Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ e3443b36

History | View | Annotate | Download (247.4 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
                          bep, hvp, hypervisor):
458
  """Builds instance related env variables for hooks
459

460
  This builds the hook environment from individual variables.
461

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

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

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

    
521
  env["INSTANCE_NIC_COUNT"] = nic_count
522

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

    
531
  env["INSTANCE_DISK_COUNT"] = disk_count
532

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

    
537
  return env
538

    
539

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

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

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

    
577

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

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

    
593

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

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

    
607

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

611
  """
612
  _OP_REQP = []
613

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

617
    This checks whether the cluster is empty.
618

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

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

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

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

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

    
647

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

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

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

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

669
    Test list:
670

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

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

688
    """
689
    node = nodeinfo.name
690

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

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

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

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

    
711
    bad = False
712

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

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

    
733
    # checks config file checksum
734

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

    
762
    # checks ssh to any
763

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

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

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

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

    
810
    return bad
811

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

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

819
    """
820
    bad = False
821

    
822
    node_current = instanceconfig.primary_node
823

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

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

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

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

    
852
    return bad
853

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

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

860
    """
861
    bad = False
862

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

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

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

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

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

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

892
    """
893
    bad = False
894

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

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

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

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

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

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

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

    
941
    return env, [], all_nodes
942

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

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

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

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

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

    
977
    local_checksums = utils.FingerprintFiles(file_names)
978

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

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

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

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

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

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

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

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

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

    
1068
      node_instance[node] = idata
1069

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

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

    
1106
    node_vol_should = {}
1107

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

    
1116
      inst_config.MapLVsByNode(node_vol_should)
1117

    
1118
      instance_cfg[instance] = inst_config
1119

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1194
    return not bad
1195

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

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

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

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

    
1243
      return lu_result
1244

    
1245

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

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

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

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

1263
    This has no prerequisites.
1264

1265
    """
1266
    pass
1267

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

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

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

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

    
1291
    if not nv_dict:
1292
      return result
1293

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

    
1296
    to_act = set()
1297
    for node in nodes:
1298
      # node_volume
1299
      lvs = node_lvs[node]
1300
      if lvs.failed:
1301
        if not lvs.offline:
1302
          self.LogWarning("Connection to node %s failed: %s" %
1303
                          (node, lvs.data))
1304
        continue
1305
      lvs = lvs.data
1306
      if isinstance(lvs, basestring):
1307
        logging.warning("Error enumerating LVs on node %s: %s", node, lvs)
1308
        res_nlvm[node] = lvs
1309
        continue
1310
      elif not isinstance(lvs, dict):
1311
        logging.warning("Connection to node %s failed or invalid data"
1312
                        " returned", node)
1313
        res_nodes.append(node)
1314
        continue
1315

    
1316
      for lv_name, (_, lv_inactive, lv_online) in lvs.iteritems():
1317
        inst = nv_dict.pop((node, lv_name), None)
1318
        if (not lv_online and inst is not None
1319
            and inst.name not in res_instances):
1320
          res_instances.append(inst.name)
1321

    
1322
    # any leftover items in nv_dict are missing LVs, let's arrange the
1323
    # data better
1324
    for key, inst in nv_dict.iteritems():
1325
      if inst.name not in res_missing:
1326
        res_missing[inst.name] = []
1327
      res_missing[inst.name].append(key)
1328

    
1329
    return result
1330

    
1331

    
1332
class LURenameCluster(LogicalUnit):
1333
  """Rename the cluster.
1334

1335
  """
1336
  HPATH = "cluster-rename"
1337
  HTYPE = constants.HTYPE_CLUSTER
1338
  _OP_REQP = ["name"]
1339

    
1340
  def BuildHooksEnv(self):
1341
    """Build hooks env.
1342

1343
    """
1344
    env = {
1345
      "OP_TARGET": self.cfg.GetClusterName(),
1346
      "NEW_NAME": self.op.name,
1347
      }
1348
    mn = self.cfg.GetMasterNode()
1349
    return env, [mn], [mn]
1350

    
1351
  def CheckPrereq(self):
1352
    """Verify that the passed name is a valid one.
1353

1354
    """
1355
    hostname = utils.HostInfo(self.op.name)
1356

    
1357
    new_name = hostname.name
1358
    self.ip = new_ip = hostname.ip
1359
    old_name = self.cfg.GetClusterName()
1360
    old_ip = self.cfg.GetMasterIP()
1361
    if new_name == old_name and new_ip == old_ip:
1362
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
1363
                                 " cluster has changed")
1364
    if new_ip != old_ip:
1365
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
1366
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
1367
                                   " reachable on the network. Aborting." %
1368
                                   new_ip)
1369

    
1370
    self.op.name = new_name
1371

    
1372
  def Exec(self, feedback_fn):
1373
    """Rename the cluster.
1374

1375
    """
1376
    clustername = self.op.name
1377
    ip = self.ip
1378

    
1379
    # shutdown the master IP
1380
    master = self.cfg.GetMasterNode()
1381
    result = self.rpc.call_node_stop_master(master, False)
1382
    if result.failed or not result.data:
1383
      raise errors.OpExecError("Could not disable the master role")
1384

    
1385
    try:
1386
      cluster = self.cfg.GetClusterInfo()
1387
      cluster.cluster_name = clustername
1388
      cluster.master_ip = ip
1389
      self.cfg.Update(cluster)
1390

    
1391
      # update the known hosts file
1392
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1393
      node_list = self.cfg.GetNodeList()
1394
      try:
1395
        node_list.remove(master)
1396
      except ValueError:
1397
        pass
1398
      result = self.rpc.call_upload_file(node_list,
1399
                                         constants.SSH_KNOWN_HOSTS_FILE)
1400
      for to_node, to_result in result.iteritems():
1401
        if to_result.failed or not to_result.data:
1402
          logging.error("Copy of file %s to node %s failed",
1403
                        constants.SSH_KNOWN_HOSTS_FILE, to_node)
1404

    
1405
    finally:
1406
      result = self.rpc.call_node_start_master(master, False, False)
1407
      if result.failed or not result.data:
1408
        self.LogWarning("Could not re-enable the master role on"
1409
                        " the master, please restart manually.")
1410

    
1411

    
1412
def _RecursiveCheckIfLVMBased(disk):
1413
  """Check if the given disk or its children are lvm-based.
1414

1415
  @type disk: L{objects.Disk}
1416
  @param disk: the disk to check
1417
  @rtype: booleean
1418
  @return: boolean indicating whether a LD_LV dev_type was found or not
1419

1420
  """
1421
  if disk.children:
1422
    for chdisk in disk.children:
1423
      if _RecursiveCheckIfLVMBased(chdisk):
1424
        return True
1425
  return disk.dev_type == constants.LD_LV
1426

    
1427

    
1428
class LUSetClusterParams(LogicalUnit):
1429
  """Change the parameters of the cluster.
1430

1431
  """
1432
  HPATH = "cluster-modify"
1433
  HTYPE = constants.HTYPE_CLUSTER
1434
  _OP_REQP = []
1435
  REQ_BGL = False
1436

    
1437
  def CheckArguments(self):
1438
    """Check parameters
1439

1440
    """
1441
    if not hasattr(self.op, "candidate_pool_size"):
1442
      self.op.candidate_pool_size = None
1443
    if self.op.candidate_pool_size is not None:
1444
      try:
1445
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1446
      except (ValueError, TypeError), err:
1447
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1448
                                   str(err))
1449
      if self.op.candidate_pool_size < 1:
1450
        raise errors.OpPrereqError("At least one master candidate needed")
1451

    
1452
  def ExpandNames(self):
1453
    # FIXME: in the future maybe other cluster params won't require checking on
1454
    # all nodes to be modified.
1455
    self.needed_locks = {
1456
      locking.LEVEL_NODE: locking.ALL_SET,
1457
    }
1458
    self.share_locks[locking.LEVEL_NODE] = 1
1459

    
1460
  def BuildHooksEnv(self):
1461
    """Build hooks env.
1462

1463
    """
1464
    env = {
1465
      "OP_TARGET": self.cfg.GetClusterName(),
1466
      "NEW_VG_NAME": self.op.vg_name,
1467
      }
1468
    mn = self.cfg.GetMasterNode()
1469
    return env, [mn], [mn]
1470

    
1471
  def CheckPrereq(self):
1472
    """Check prerequisites.
1473

1474
    This checks whether the given params don't conflict and
1475
    if the given volume group is valid.
1476

1477
    """
1478
    if self.op.vg_name is not None and not self.op.vg_name:
1479
      instances = self.cfg.GetAllInstancesInfo().values()
1480
      for inst in instances:
1481
        for disk in inst.disks:
1482
          if _RecursiveCheckIfLVMBased(disk):
1483
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1484
                                       " lvm-based instances exist")
1485

    
1486
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1487

    
1488
    # if vg_name not None, checks given volume group on all nodes
1489
    if self.op.vg_name:
1490
      vglist = self.rpc.call_vg_list(node_list)
1491
      for node in node_list:
1492
        if vglist[node].failed:
1493
          # ignoring down node
1494
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1495
          continue
1496
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1497
                                              self.op.vg_name,
1498
                                              constants.MIN_VG_SIZE)
1499
        if vgstatus:
1500
          raise errors.OpPrereqError("Error on node '%s': %s" %
1501
                                     (node, vgstatus))
1502

    
1503
    self.cluster = cluster = self.cfg.GetClusterInfo()
1504
    # validate beparams changes
1505
    if self.op.beparams:
1506
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1507
      self.new_beparams = cluster.FillDict(
1508
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1509

    
1510
    # hypervisor list/parameters
1511
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1512
    if self.op.hvparams:
1513
      if not isinstance(self.op.hvparams, dict):
1514
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1515
      for hv_name, hv_dict in self.op.hvparams.items():
1516
        if hv_name not in self.new_hvparams:
1517
          self.new_hvparams[hv_name] = hv_dict
1518
        else:
1519
          self.new_hvparams[hv_name].update(hv_dict)
1520

    
1521
    if self.op.enabled_hypervisors is not None:
1522
      self.hv_list = self.op.enabled_hypervisors
1523
    else:
1524
      self.hv_list = cluster.enabled_hypervisors
1525

    
1526
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1527
      # either the enabled list has changed, or the parameters have, validate
1528
      for hv_name, hv_params in self.new_hvparams.items():
1529
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1530
            (self.op.enabled_hypervisors and
1531
             hv_name in self.op.enabled_hypervisors)):
1532
          # either this is a new hypervisor, or its parameters have changed
1533
          hv_class = hypervisor.GetHypervisor(hv_name)
1534
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1535
          hv_class.CheckParameterSyntax(hv_params)
1536
          _CheckHVParams(self, node_list, hv_name, hv_params)
1537

    
1538
  def Exec(self, feedback_fn):
1539
    """Change the parameters of the cluster.
1540

1541
    """
1542
    if self.op.vg_name is not None:
1543
      new_volume = self.op.vg_name
1544
      if not new_volume:
1545
        new_volume = None
1546
      if new_volume != self.cfg.GetVGName():
1547
        self.cfg.SetVGName(new_volume)
1548
      else:
1549
        feedback_fn("Cluster LVM configuration already in desired"
1550
                    " state, not changing")
1551
    if self.op.hvparams:
1552
      self.cluster.hvparams = self.new_hvparams
1553
    if self.op.enabled_hypervisors is not None:
1554
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1555
    if self.op.beparams:
1556
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1557
    if self.op.candidate_pool_size is not None:
1558
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1559
      # we need to update the pool size here, otherwise the save will fail
1560
      _AdjustCandidatePool(self)
1561

    
1562
    self.cfg.Update(self.cluster)
1563

    
1564

    
1565
class LURedistributeConfig(NoHooksLU):
1566
  """Force the redistribution of cluster configuration.
1567

1568
  This is a very simple LU.
1569

1570
  """
1571
  _OP_REQP = []
1572
  REQ_BGL = False
1573

    
1574
  def ExpandNames(self):
1575
    self.needed_locks = {
1576
      locking.LEVEL_NODE: locking.ALL_SET,
1577
    }
1578
    self.share_locks[locking.LEVEL_NODE] = 1
1579

    
1580
  def CheckPrereq(self):
1581
    """Check prerequisites.
1582

1583
    """
1584

    
1585
  def Exec(self, feedback_fn):
1586
    """Redistribute the configuration.
1587

1588
    """
1589
    self.cfg.Update(self.cfg.GetClusterInfo())
1590

    
1591

    
1592
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1593
  """Sleep and poll for an instance's disk to sync.
1594

1595
  """
1596
  if not instance.disks:
1597
    return True
1598

    
1599
  if not oneshot:
1600
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1601

    
1602
  node = instance.primary_node
1603

    
1604
  for dev in instance.disks:
1605
    lu.cfg.SetDiskID(dev, node)
1606

    
1607
  retries = 0
1608
  degr_retries = 10 # in seconds, as we sleep 1 second each time
1609
  while True:
1610
    max_time = 0
1611
    done = True
1612
    cumul_degraded = False
1613
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1614
    if rstats.failed or not rstats.data:
1615
      lu.LogWarning("Can't get any data from node %s", node)
1616
      retries += 1
1617
      if retries >= 10:
1618
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1619
                                 " aborting." % node)
1620
      time.sleep(6)
1621
      continue
1622
    rstats = rstats.data
1623
    retries = 0
1624
    for i, mstat in enumerate(rstats):
1625
      if mstat is None:
1626
        lu.LogWarning("Can't compute data for node %s/%s",
1627
                           node, instance.disks[i].iv_name)
1628
        continue
1629
      # we ignore the ldisk parameter
1630
      perc_done, est_time, is_degraded, _ = mstat
1631
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1632
      if perc_done is not None:
1633
        done = False
1634
        if est_time is not None:
1635
          rem_time = "%d estimated seconds remaining" % est_time
1636
          max_time = est_time
1637
        else:
1638
          rem_time = "no time estimate"
1639
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1640
                        (instance.disks[i].iv_name, perc_done, rem_time))
1641

    
1642
    # if we're done but degraded, let's do a few small retries, to
1643
    # make sure we see a stable and not transient situation; therefore
1644
    # we force restart of the loop
1645
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
1646
      logging.info("Degraded disks found, %d retries left", degr_retries)
1647
      degr_retries -= 1
1648
      time.sleep(1)
1649
      continue
1650

    
1651
    if done or oneshot:
1652
      break
1653

    
1654
    time.sleep(min(60, max_time))
1655

    
1656
  if done:
1657
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1658
  return not cumul_degraded
1659

    
1660

    
1661
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1662
  """Check that mirrors are not degraded.
1663

1664
  The ldisk parameter, if True, will change the test from the
1665
  is_degraded attribute (which represents overall non-ok status for
1666
  the device(s)) to the ldisk (representing the local storage status).
1667

1668
  """
1669
  lu.cfg.SetDiskID(dev, node)
1670
  if ldisk:
1671
    idx = 6
1672
  else:
1673
    idx = 5
1674

    
1675
  result = True
1676
  if on_primary or dev.AssembleOnSecondary():
1677
    rstats = lu.rpc.call_blockdev_find(node, dev)
1678
    msg = rstats.RemoteFailMsg()
1679
    if msg:
1680
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1681
      result = False
1682
    elif not rstats.payload:
1683
      lu.LogWarning("Can't find disk on node %s", node)
1684
      result = False
1685
    else:
1686
      result = result and (not rstats.payload[idx])
1687
  if dev.children:
1688
    for child in dev.children:
1689
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1690

    
1691
  return result
1692

    
1693

    
1694
class LUDiagnoseOS(NoHooksLU):
1695
  """Logical unit for OS diagnose/query.
1696

1697
  """
1698
  _OP_REQP = ["output_fields", "names"]
1699
  REQ_BGL = False
1700
  _FIELDS_STATIC = utils.FieldSet()
1701
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1702

    
1703
  def ExpandNames(self):
1704
    if self.op.names:
1705
      raise errors.OpPrereqError("Selective OS query not supported")
1706

    
1707
    _CheckOutputFields(static=self._FIELDS_STATIC,
1708
                       dynamic=self._FIELDS_DYNAMIC,
1709
                       selected=self.op.output_fields)
1710

    
1711
    # Lock all nodes, in shared mode
1712
    # Temporary removal of locks, should be reverted later
1713
    # TODO: reintroduce locks when they are lighter-weight
1714
    self.needed_locks = {}
1715
    #self.share_locks[locking.LEVEL_NODE] = 1
1716
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1717

    
1718
  def CheckPrereq(self):
1719
    """Check prerequisites.
1720

1721
    """
1722

    
1723
  @staticmethod
1724
  def _DiagnoseByOS(node_list, rlist):
1725
    """Remaps a per-node return list into an a per-os per-node dictionary
1726

1727
    @param node_list: a list with the names of all nodes
1728
    @param rlist: a map with node names as keys and OS objects as values
1729

1730
    @rtype: dict
1731
    @return: a dictionary with osnames as keys and as value another map, with
1732
        nodes as keys and list of OS objects as values, eg::
1733

1734
          {"debian-etch": {"node1": [<object>,...],
1735
                           "node2": [<object>,]}
1736
          }
1737

1738
    """
1739
    all_os = {}
1740
    # we build here the list of nodes that didn't fail the RPC (at RPC
1741
    # level), so that nodes with a non-responding node daemon don't
1742
    # make all OSes invalid
1743
    good_nodes = [node_name for node_name in rlist
1744
                  if not rlist[node_name].failed]
1745
    for node_name, nr in rlist.iteritems():
1746
      if nr.failed or not nr.data:
1747
        continue
1748
      for os_obj in nr.data:
1749
        if os_obj.name not in all_os:
1750
          # build a list of nodes for this os containing empty lists
1751
          # for each node in node_list
1752
          all_os[os_obj.name] = {}
1753
          for nname in good_nodes:
1754
            all_os[os_obj.name][nname] = []
1755
        all_os[os_obj.name][node_name].append(os_obj)
1756
    return all_os
1757

    
1758
  def Exec(self, feedback_fn):
1759
    """Compute the list of OSes.
1760

1761
    """
1762
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1763
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1764
    if node_data == False:
1765
      raise errors.OpExecError("Can't gather the list of OSes")
1766
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1767
    output = []
1768
    for os_name, os_data in pol.iteritems():
1769
      row = []
1770
      for field in self.op.output_fields:
1771
        if field == "name":
1772
          val = os_name
1773
        elif field == "valid":
1774
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1775
        elif field == "node_status":
1776
          val = {}
1777
          for node_name, nos_list in os_data.iteritems():
1778
            val[node_name] = [(v.status, v.path) for v in nos_list]
1779
        else:
1780
          raise errors.ParameterError(field)
1781
        row.append(val)
1782
      output.append(row)
1783

    
1784
    return output
1785

    
1786

    
1787
class LURemoveNode(LogicalUnit):
1788
  """Logical unit for removing a node.
1789

1790
  """
1791
  HPATH = "node-remove"
1792
  HTYPE = constants.HTYPE_NODE
1793
  _OP_REQP = ["node_name"]
1794

    
1795
  def BuildHooksEnv(self):
1796
    """Build hooks env.
1797

1798
    This doesn't run on the target node in the pre phase as a failed
1799
    node would then be impossible to remove.
1800

1801
    """
1802
    env = {
1803
      "OP_TARGET": self.op.node_name,
1804
      "NODE_NAME": self.op.node_name,
1805
      }
1806
    all_nodes = self.cfg.GetNodeList()
1807
    all_nodes.remove(self.op.node_name)
1808
    return env, all_nodes, all_nodes
1809

    
1810
  def CheckPrereq(self):
1811
    """Check prerequisites.
1812

1813
    This checks:
1814
     - the node exists in the configuration
1815
     - it does not have primary or secondary instances
1816
     - it's not the master
1817

1818
    Any errors are signalled by raising errors.OpPrereqError.
1819

1820
    """
1821
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1822
    if node is None:
1823
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1824

    
1825
    instance_list = self.cfg.GetInstanceList()
1826

    
1827
    masternode = self.cfg.GetMasterNode()
1828
    if node.name == masternode:
1829
      raise errors.OpPrereqError("Node is the master node,"
1830
                                 " you need to failover first.")
1831

    
1832
    for instance_name in instance_list:
1833
      instance = self.cfg.GetInstanceInfo(instance_name)
1834
      if node.name in instance.all_nodes:
1835
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1836
                                   " please remove first." % instance_name)
1837
    self.op.node_name = node.name
1838
    self.node = node
1839

    
1840
  def Exec(self, feedback_fn):
1841
    """Removes the node from the cluster.
1842

1843
    """
1844
    node = self.node
1845
    logging.info("Stopping the node daemon and removing configs from node %s",
1846
                 node.name)
1847

    
1848
    self.context.RemoveNode(node.name)
1849

    
1850
    self.rpc.call_node_leave_cluster(node.name)
1851

    
1852
    # Promote nodes to master candidate as needed
1853
    _AdjustCandidatePool(self)
1854

    
1855

    
1856
class LUQueryNodes(NoHooksLU):
1857
  """Logical unit for querying nodes.
1858

1859
  """
1860
  _OP_REQP = ["output_fields", "names", "use_locking"]
1861
  REQ_BGL = False
1862
  _FIELDS_DYNAMIC = utils.FieldSet(
1863
    "dtotal", "dfree",
1864
    "mtotal", "mnode", "mfree",
1865
    "bootid",
1866
    "ctotal", "cnodes", "csockets",
1867
    )
1868

    
1869
  _FIELDS_STATIC = utils.FieldSet(
1870
    "name", "pinst_cnt", "sinst_cnt",
1871
    "pinst_list", "sinst_list",
1872
    "pip", "sip", "tags",
1873
    "serial_no",
1874
    "master_candidate",
1875
    "master",
1876
    "offline",
1877
    "drained",
1878
    "role",
1879
    )
1880

    
1881
  def ExpandNames(self):
1882
    _CheckOutputFields(static=self._FIELDS_STATIC,
1883
                       dynamic=self._FIELDS_DYNAMIC,
1884
                       selected=self.op.output_fields)
1885

    
1886
    self.needed_locks = {}
1887
    self.share_locks[locking.LEVEL_NODE] = 1
1888

    
1889
    if self.op.names:
1890
      self.wanted = _GetWantedNodes(self, self.op.names)
1891
    else:
1892
      self.wanted = locking.ALL_SET
1893

    
1894
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1895
    self.do_locking = self.do_node_query and self.op.use_locking
1896
    if self.do_locking:
1897
      # if we don't request only static fields, we need to lock the nodes
1898
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1899

    
1900

    
1901
  def CheckPrereq(self):
1902
    """Check prerequisites.
1903

1904
    """
1905
    # The validation of the node list is done in the _GetWantedNodes,
1906
    # if non empty, and if empty, there's no validation to do
1907
    pass
1908

    
1909
  def Exec(self, feedback_fn):
1910
    """Computes the list of nodes and their attributes.
1911

1912
    """
1913
    all_info = self.cfg.GetAllNodesInfo()
1914
    if self.do_locking:
1915
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1916
    elif self.wanted != locking.ALL_SET:
1917
      nodenames = self.wanted
1918
      missing = set(nodenames).difference(all_info.keys())
1919
      if missing:
1920
        raise errors.OpExecError(
1921
          "Some nodes were removed before retrieving their data: %s" % missing)
1922
    else:
1923
      nodenames = all_info.keys()
1924

    
1925
    nodenames = utils.NiceSort(nodenames)
1926
    nodelist = [all_info[name] for name in nodenames]
1927

    
1928
    # begin data gathering
1929

    
1930
    if self.do_node_query:
1931
      live_data = {}
1932
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1933
                                          self.cfg.GetHypervisorType())
1934
      for name in nodenames:
1935
        nodeinfo = node_data[name]
1936
        if not nodeinfo.failed and nodeinfo.data:
1937
          nodeinfo = nodeinfo.data
1938
          fn = utils.TryConvert
1939
          live_data[name] = {
1940
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1941
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1942
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1943
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1944
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1945
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1946
            "bootid": nodeinfo.get('bootid', None),
1947
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
1948
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
1949
            }
1950
        else:
1951
          live_data[name] = {}
1952
    else:
1953
      live_data = dict.fromkeys(nodenames, {})
1954

    
1955
    node_to_primary = dict([(name, set()) for name in nodenames])
1956
    node_to_secondary = dict([(name, set()) for name in nodenames])
1957

    
1958
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1959
                             "sinst_cnt", "sinst_list"))
1960
    if inst_fields & frozenset(self.op.output_fields):
1961
      instancelist = self.cfg.GetInstanceList()
1962

    
1963
      for instance_name in instancelist:
1964
        inst = self.cfg.GetInstanceInfo(instance_name)
1965
        if inst.primary_node in node_to_primary:
1966
          node_to_primary[inst.primary_node].add(inst.name)
1967
        for secnode in inst.secondary_nodes:
1968
          if secnode in node_to_secondary:
1969
            node_to_secondary[secnode].add(inst.name)
1970

    
1971
    master_node = self.cfg.GetMasterNode()
1972

    
1973
    # end data gathering
1974

    
1975
    output = []
1976
    for node in nodelist:
1977
      node_output = []
1978
      for field in self.op.output_fields:
1979
        if field == "name":
1980
          val = node.name
1981
        elif field == "pinst_list":
1982
          val = list(node_to_primary[node.name])
1983
        elif field == "sinst_list":
1984
          val = list(node_to_secondary[node.name])
1985
        elif field == "pinst_cnt":
1986
          val = len(node_to_primary[node.name])
1987
        elif field == "sinst_cnt":
1988
          val = len(node_to_secondary[node.name])
1989
        elif field == "pip":
1990
          val = node.primary_ip
1991
        elif field == "sip":
1992
          val = node.secondary_ip
1993
        elif field == "tags":
1994
          val = list(node.GetTags())
1995
        elif field == "serial_no":
1996
          val = node.serial_no
1997
        elif field == "master_candidate":
1998
          val = node.master_candidate
1999
        elif field == "master":
2000
          val = node.name == master_node
2001
        elif field == "offline":
2002
          val = node.offline
2003
        elif field == "drained":
2004
          val = node.drained
2005
        elif self._FIELDS_DYNAMIC.Matches(field):
2006
          val = live_data[node.name].get(field, None)
2007
        elif field == "role":
2008
          if node.name == master_node:
2009
            val = "M"
2010
          elif node.master_candidate:
2011
            val = "C"
2012
          elif node.drained:
2013
            val = "D"
2014
          elif node.offline:
2015
            val = "O"
2016
          else:
2017
            val = "R"
2018
        else:
2019
          raise errors.ParameterError(field)
2020
        node_output.append(val)
2021
      output.append(node_output)
2022

    
2023
    return output
2024

    
2025

    
2026
class LUQueryNodeVolumes(NoHooksLU):
2027
  """Logical unit for getting volumes on node(s).
2028

2029
  """
2030
  _OP_REQP = ["nodes", "output_fields"]
2031
  REQ_BGL = False
2032
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2033
  _FIELDS_STATIC = utils.FieldSet("node")
2034

    
2035
  def ExpandNames(self):
2036
    _CheckOutputFields(static=self._FIELDS_STATIC,
2037
                       dynamic=self._FIELDS_DYNAMIC,
2038
                       selected=self.op.output_fields)
2039

    
2040
    self.needed_locks = {}
2041
    self.share_locks[locking.LEVEL_NODE] = 1
2042
    if not self.op.nodes:
2043
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2044
    else:
2045
      self.needed_locks[locking.LEVEL_NODE] = \
2046
        _GetWantedNodes(self, self.op.nodes)
2047

    
2048
  def CheckPrereq(self):
2049
    """Check prerequisites.
2050

2051
    This checks that the fields required are valid output fields.
2052

2053
    """
2054
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2055

    
2056
  def Exec(self, feedback_fn):
2057
    """Computes the list of nodes and their attributes.
2058

2059
    """
2060
    nodenames = self.nodes
2061
    volumes = self.rpc.call_node_volumes(nodenames)
2062

    
2063
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2064
             in self.cfg.GetInstanceList()]
2065

    
2066
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2067

    
2068
    output = []
2069
    for node in nodenames:
2070
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2071
        continue
2072

    
2073
      node_vols = volumes[node].data[:]
2074
      node_vols.sort(key=lambda vol: vol['dev'])
2075

    
2076
      for vol in node_vols:
2077
        node_output = []
2078
        for field in self.op.output_fields:
2079
          if field == "node":
2080
            val = node
2081
          elif field == "phys":
2082
            val = vol['dev']
2083
          elif field == "vg":
2084
            val = vol['vg']
2085
          elif field == "name":
2086
            val = vol['name']
2087
          elif field == "size":
2088
            val = int(float(vol['size']))
2089
          elif field == "instance":
2090
            for inst in ilist:
2091
              if node not in lv_by_node[inst]:
2092
                continue
2093
              if vol['name'] in lv_by_node[inst][node]:
2094
                val = inst.name
2095
                break
2096
            else:
2097
              val = '-'
2098
          else:
2099
            raise errors.ParameterError(field)
2100
          node_output.append(str(val))
2101

    
2102
        output.append(node_output)
2103

    
2104
    return output
2105

    
2106

    
2107
class LUAddNode(LogicalUnit):
2108
  """Logical unit for adding node to the cluster.
2109

2110
  """
2111
  HPATH = "node-add"
2112
  HTYPE = constants.HTYPE_NODE
2113
  _OP_REQP = ["node_name"]
2114

    
2115
  def BuildHooksEnv(self):
2116
    """Build hooks env.
2117

2118
    This will run on all nodes before, and on all nodes + the new node after.
2119

2120
    """
2121
    env = {
2122
      "OP_TARGET": self.op.node_name,
2123
      "NODE_NAME": self.op.node_name,
2124
      "NODE_PIP": self.op.primary_ip,
2125
      "NODE_SIP": self.op.secondary_ip,
2126
      }
2127
    nodes_0 = self.cfg.GetNodeList()
2128
    nodes_1 = nodes_0 + [self.op.node_name, ]
2129
    return env, nodes_0, nodes_1
2130

    
2131
  def CheckPrereq(self):
2132
    """Check prerequisites.
2133

2134
    This checks:
2135
     - the new node is not already in the config
2136
     - it is resolvable
2137
     - its parameters (single/dual homed) matches the cluster
2138

2139
    Any errors are signalled by raising errors.OpPrereqError.
2140

2141
    """
2142
    node_name = self.op.node_name
2143
    cfg = self.cfg
2144

    
2145
    dns_data = utils.HostInfo(node_name)
2146

    
2147
    node = dns_data.name
2148
    primary_ip = self.op.primary_ip = dns_data.ip
2149
    secondary_ip = getattr(self.op, "secondary_ip", None)
2150
    if secondary_ip is None:
2151
      secondary_ip = primary_ip
2152
    if not utils.IsValidIP(secondary_ip):
2153
      raise errors.OpPrereqError("Invalid secondary IP given")
2154
    self.op.secondary_ip = secondary_ip
2155

    
2156
    node_list = cfg.GetNodeList()
2157
    if not self.op.readd and node in node_list:
2158
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2159
                                 node)
2160
    elif self.op.readd and node not in node_list:
2161
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2162

    
2163
    for existing_node_name in node_list:
2164
      existing_node = cfg.GetNodeInfo(existing_node_name)
2165

    
2166
      if self.op.readd and node == existing_node_name:
2167
        if (existing_node.primary_ip != primary_ip or
2168
            existing_node.secondary_ip != secondary_ip):
2169
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2170
                                     " address configuration as before")
2171
        continue
2172

    
2173
      if (existing_node.primary_ip == primary_ip or
2174
          existing_node.secondary_ip == primary_ip or
2175
          existing_node.primary_ip == secondary_ip or
2176
          existing_node.secondary_ip == secondary_ip):
2177
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2178
                                   " existing node %s" % existing_node.name)
2179

    
2180
    # check that the type of the node (single versus dual homed) is the
2181
    # same as for the master
2182
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2183
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2184
    newbie_singlehomed = secondary_ip == primary_ip
2185
    if master_singlehomed != newbie_singlehomed:
2186
      if master_singlehomed:
2187
        raise errors.OpPrereqError("The master has no private ip but the"
2188
                                   " new node has one")
2189
      else:
2190
        raise errors.OpPrereqError("The master has a private ip but the"
2191
                                   " new node doesn't have one")
2192

    
2193
    # checks reachablity
2194
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2195
      raise errors.OpPrereqError("Node not reachable by ping")
2196

    
2197
    if not newbie_singlehomed:
2198
      # check reachability from my secondary ip to newbie's secondary ip
2199
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2200
                           source=myself.secondary_ip):
2201
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2202
                                   " based ping to noded port")
2203

    
2204
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2205
    if self.op.readd:
2206
      exceptions = [node]
2207
    else:
2208
      exceptions = []
2209
    mc_now, mc_max = self.cfg.GetMasterCandidateStats(exceptions)
2210
    # the new node will increase mc_max with one, so:
2211
    mc_max = min(mc_max + 1, cp_size)
2212
    self.master_candidate = mc_now < mc_max
2213

    
2214
    if self.op.readd:
2215
      self.new_node = self.cfg.GetNodeInfo(node)
2216
      assert self.new_node is not None, "Can't retrieve locked node %s" % node
2217
    else:
2218
      self.new_node = objects.Node(name=node,
2219
                                   primary_ip=primary_ip,
2220
                                   secondary_ip=secondary_ip,
2221
                                   master_candidate=self.master_candidate,
2222
                                   offline=False, drained=False)
2223

    
2224
  def Exec(self, feedback_fn):
2225
    """Adds the new node to the cluster.
2226

2227
    """
2228
    new_node = self.new_node
2229
    node = new_node.name
2230

    
2231
    # for re-adds, reset the offline/drained/master-candidate flags;
2232
    # we need to reset here, otherwise offline would prevent RPC calls
2233
    # later in the procedure; this also means that if the re-add
2234
    # fails, we are left with a non-offlined, broken node
2235
    if self.op.readd:
2236
      new_node.drained = new_node.offline = False
2237
      self.LogInfo("Readding a node, the offline/drained flags were reset")
2238
      # if we demote the node, we do cleanup later in the procedure
2239
      new_node.master_candidate = self.master_candidate
2240

    
2241
    # notify the user about any possible mc promotion
2242
    if new_node.master_candidate:
2243
      self.LogInfo("Node will be a master candidate")
2244

    
2245
    # check connectivity
2246
    result = self.rpc.call_version([node])[node]
2247
    result.Raise()
2248
    if result.data:
2249
      if constants.PROTOCOL_VERSION == result.data:
2250
        logging.info("Communication to node %s fine, sw version %s match",
2251
                     node, result.data)
2252
      else:
2253
        raise errors.OpExecError("Version mismatch master version %s,"
2254
                                 " node version %s" %
2255
                                 (constants.PROTOCOL_VERSION, result.data))
2256
    else:
2257
      raise errors.OpExecError("Cannot get version from the new node")
2258

    
2259
    # setup ssh on node
2260
    logging.info("Copy ssh key to node %s", node)
2261
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2262
    keyarray = []
2263
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2264
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2265
                priv_key, pub_key]
2266

    
2267
    for i in keyfiles:
2268
      f = open(i, 'r')
2269
      try:
2270
        keyarray.append(f.read())
2271
      finally:
2272
        f.close()
2273

    
2274
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2275
                                    keyarray[2],
2276
                                    keyarray[3], keyarray[4], keyarray[5])
2277

    
2278
    msg = result.RemoteFailMsg()
2279
    if msg:
2280
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2281
                               " new node: %s" % msg)
2282

    
2283
    # Add node to our /etc/hosts, and add key to known_hosts
2284
    utils.AddHostToEtcHosts(new_node.name)
2285

    
2286
    if new_node.secondary_ip != new_node.primary_ip:
2287
      result = self.rpc.call_node_has_ip_address(new_node.name,
2288
                                                 new_node.secondary_ip)
2289
      if result.failed or not result.data:
2290
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2291
                                 " you gave (%s). Please fix and re-run this"
2292
                                 " command." % new_node.secondary_ip)
2293

    
2294
    node_verify_list = [self.cfg.GetMasterNode()]
2295
    node_verify_param = {
2296
      'nodelist': [node],
2297
      # TODO: do a node-net-test as well?
2298
    }
2299

    
2300
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2301
                                       self.cfg.GetClusterName())
2302
    for verifier in node_verify_list:
2303
      if result[verifier].failed or not result[verifier].data:
2304
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2305
                                 " for remote verification" % verifier)
2306
      if result[verifier].data['nodelist']:
2307
        for failed in result[verifier].data['nodelist']:
2308
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2309
                      (verifier, result[verifier].data['nodelist'][failed]))
2310
        raise errors.OpExecError("ssh/hostname verification failed.")
2311

    
2312
    # Distribute updated /etc/hosts and known_hosts to all nodes,
2313
    # including the node just added
2314
    myself = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
2315
    dist_nodes = self.cfg.GetNodeList()
2316
    if not self.op.readd:
2317
      dist_nodes.append(node)
2318
    if myself.name in dist_nodes:
2319
      dist_nodes.remove(myself.name)
2320

    
2321
    logging.debug("Copying hosts and known_hosts to all nodes")
2322
    for fname in (constants.ETC_HOSTS, constants.SSH_KNOWN_HOSTS_FILE):
2323
      result = self.rpc.call_upload_file(dist_nodes, fname)
2324
      for to_node, to_result in result.iteritems():
2325
        if to_result.failed or not to_result.data:
2326
          logging.error("Copy of file %s to node %s failed", fname, to_node)
2327

    
2328
    to_copy = []
2329
    enabled_hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2330
    if constants.HTS_COPY_VNC_PASSWORD.intersection(enabled_hypervisors):
2331
      to_copy.append(constants.VNC_PASSWORD_FILE)
2332

    
2333
    for fname in to_copy:
2334
      result = self.rpc.call_upload_file([node], fname)
2335
      if result[node].failed or not result[node]:
2336
        logging.error("Could not copy file %s to node %s", fname, node)
2337

    
2338
    if self.op.readd:
2339
      self.context.ReaddNode(new_node)
2340
      # make sure we redistribute the config
2341
      self.cfg.Update(new_node)
2342
      # and make sure the new node will not have old files around
2343
      if not new_node.master_candidate:
2344
        result = self.rpc.call_node_demote_from_mc(new_node.name)
2345
        msg = result.RemoteFailMsg()
2346
        if msg:
2347
          self.LogWarning("Node failed to demote itself from master"
2348
                          " candidate status: %s" % msg)
2349
    else:
2350
      self.context.AddNode(new_node)
2351

    
2352

    
2353
class LUSetNodeParams(LogicalUnit):
2354
  """Modifies the parameters of a node.
2355

2356
  """
2357
  HPATH = "node-modify"
2358
  HTYPE = constants.HTYPE_NODE
2359
  _OP_REQP = ["node_name"]
2360
  REQ_BGL = False
2361

    
2362
  def CheckArguments(self):
2363
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2364
    if node_name is None:
2365
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2366
    self.op.node_name = node_name
2367
    _CheckBooleanOpField(self.op, 'master_candidate')
2368
    _CheckBooleanOpField(self.op, 'offline')
2369
    _CheckBooleanOpField(self.op, 'drained')
2370
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2371
    if all_mods.count(None) == 3:
2372
      raise errors.OpPrereqError("Please pass at least one modification")
2373
    if all_mods.count(True) > 1:
2374
      raise errors.OpPrereqError("Can't set the node into more than one"
2375
                                 " state at the same time")
2376

    
2377
  def ExpandNames(self):
2378
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2379

    
2380
  def BuildHooksEnv(self):
2381
    """Build hooks env.
2382

2383
    This runs on the master node.
2384

2385
    """
2386
    env = {
2387
      "OP_TARGET": self.op.node_name,
2388
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2389
      "OFFLINE": str(self.op.offline),
2390
      "DRAINED": str(self.op.drained),
2391
      }
2392
    nl = [self.cfg.GetMasterNode(),
2393
          self.op.node_name]
2394
    return env, nl, nl
2395

    
2396
  def CheckPrereq(self):
2397
    """Check prerequisites.
2398

2399
    This only checks the instance list against the existing names.
2400

2401
    """
2402
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2403

    
2404
    if ((self.op.master_candidate == False or self.op.offline == True or
2405
         self.op.drained == True) and node.master_candidate):
2406
      # we will demote the node from master_candidate
2407
      if self.op.node_name == self.cfg.GetMasterNode():
2408
        raise errors.OpPrereqError("The master node has to be a"
2409
                                   " master candidate, online and not drained")
2410
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2411
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2412
      if num_candidates <= cp_size:
2413
        msg = ("Not enough master candidates (desired"
2414
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2415
        if self.op.force:
2416
          self.LogWarning(msg)
2417
        else:
2418
          raise errors.OpPrereqError(msg)
2419

    
2420
    if (self.op.master_candidate == True and
2421
        ((node.offline and not self.op.offline == False) or
2422
         (node.drained and not self.op.drained == False))):
2423
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2424
                                 " to master_candidate" % node.name)
2425

    
2426
    return
2427

    
2428
  def Exec(self, feedback_fn):
2429
    """Modifies a node.
2430

2431
    """
2432
    node = self.node
2433

    
2434
    result = []
2435
    changed_mc = False
2436

    
2437
    if self.op.offline is not None:
2438
      node.offline = self.op.offline
2439
      result.append(("offline", str(self.op.offline)))
2440
      if self.op.offline == True:
2441
        if node.master_candidate:
2442
          node.master_candidate = False
2443
          changed_mc = True
2444
          result.append(("master_candidate", "auto-demotion due to offline"))
2445
        if node.drained:
2446
          node.drained = False
2447
          result.append(("drained", "clear drained status due to offline"))
2448

    
2449
    if self.op.master_candidate is not None:
2450
      node.master_candidate = self.op.master_candidate
2451
      changed_mc = True
2452
      result.append(("master_candidate", str(self.op.master_candidate)))
2453
      if self.op.master_candidate == False:
2454
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2455
        msg = rrc.RemoteFailMsg()
2456
        if msg:
2457
          self.LogWarning("Node failed to demote itself: %s" % msg)
2458

    
2459
    if self.op.drained is not None:
2460
      node.drained = self.op.drained
2461
      result.append(("drained", str(self.op.drained)))
2462
      if self.op.drained == True:
2463
        if node.master_candidate:
2464
          node.master_candidate = False
2465
          changed_mc = True
2466
          result.append(("master_candidate", "auto-demotion due to drain"))
2467
          rrc = self.rpc.call_node_demote_from_mc(node.name)
2468
          msg = rrc.RemoteFailMsg()
2469
          if msg:
2470
            self.LogWarning("Node failed to demote itself: %s" % msg)
2471
        if node.offline:
2472
          node.offline = False
2473
          result.append(("offline", "clear offline status due to drain"))
2474

    
2475
    # this will trigger configuration file update, if needed
2476
    self.cfg.Update(node)
2477
    # this will trigger job queue propagation or cleanup
2478
    if changed_mc:
2479
      self.context.ReaddNode(node)
2480

    
2481
    return result
2482

    
2483

    
2484
class LUQueryClusterInfo(NoHooksLU):
2485
  """Query cluster configuration.
2486

2487
  """
2488
  _OP_REQP = []
2489
  REQ_BGL = False
2490

    
2491
  def ExpandNames(self):
2492
    self.needed_locks = {}
2493

    
2494
  def CheckPrereq(self):
2495
    """No prerequsites needed for this LU.
2496

2497
    """
2498
    pass
2499

    
2500
  def Exec(self, feedback_fn):
2501
    """Return cluster config.
2502

2503
    """
2504
    cluster = self.cfg.GetClusterInfo()
2505
    result = {
2506
      "software_version": constants.RELEASE_VERSION,
2507
      "protocol_version": constants.PROTOCOL_VERSION,
2508
      "config_version": constants.CONFIG_VERSION,
2509
      "os_api_version": constants.OS_API_VERSION,
2510
      "export_version": constants.EXPORT_VERSION,
2511
      "architecture": (platform.architecture()[0], platform.machine()),
2512
      "name": cluster.cluster_name,
2513
      "master": cluster.master_node,
2514
      "default_hypervisor": cluster.default_hypervisor,
2515
      "enabled_hypervisors": cluster.enabled_hypervisors,
2516
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2517
                        for hypervisor in cluster.enabled_hypervisors]),
2518
      "beparams": cluster.beparams,
2519
      "candidate_pool_size": cluster.candidate_pool_size,
2520
      "default_bridge": cluster.default_bridge,
2521
      "master_netdev": cluster.master_netdev,
2522
      "volume_group_name": cluster.volume_group_name,
2523
      "file_storage_dir": cluster.file_storage_dir,
2524
      }
2525

    
2526
    return result
2527

    
2528

    
2529
class LUQueryConfigValues(NoHooksLU):
2530
  """Return configuration values.
2531

2532
  """
2533
  _OP_REQP = []
2534
  REQ_BGL = False
2535
  _FIELDS_DYNAMIC = utils.FieldSet()
2536
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2537

    
2538
  def ExpandNames(self):
2539
    self.needed_locks = {}
2540

    
2541
    _CheckOutputFields(static=self._FIELDS_STATIC,
2542
                       dynamic=self._FIELDS_DYNAMIC,
2543
                       selected=self.op.output_fields)
2544

    
2545
  def CheckPrereq(self):
2546
    """No prerequisites.
2547

2548
    """
2549
    pass
2550

    
2551
  def Exec(self, feedback_fn):
2552
    """Dump a representation of the cluster config to the standard output.
2553

2554
    """
2555
    values = []
2556
    for field in self.op.output_fields:
2557
      if field == "cluster_name":
2558
        entry = self.cfg.GetClusterName()
2559
      elif field == "master_node":
2560
        entry = self.cfg.GetMasterNode()
2561
      elif field == "drain_flag":
2562
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2563
      else:
2564
        raise errors.ParameterError(field)
2565
      values.append(entry)
2566
    return values
2567

    
2568

    
2569
class LUActivateInstanceDisks(NoHooksLU):
2570
  """Bring up an instance's disks.
2571

2572
  """
2573
  _OP_REQP = ["instance_name"]
2574
  REQ_BGL = False
2575

    
2576
  def ExpandNames(self):
2577
    self._ExpandAndLockInstance()
2578
    self.needed_locks[locking.LEVEL_NODE] = []
2579
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2580

    
2581
  def DeclareLocks(self, level):
2582
    if level == locking.LEVEL_NODE:
2583
      self._LockInstancesNodes()
2584

    
2585
  def CheckPrereq(self):
2586
    """Check prerequisites.
2587

2588
    This checks that the instance is in the cluster.
2589

2590
    """
2591
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2592
    assert self.instance is not None, \
2593
      "Cannot retrieve locked instance %s" % self.op.instance_name
2594
    _CheckNodeOnline(self, self.instance.primary_node)
2595

    
2596
  def Exec(self, feedback_fn):
2597
    """Activate the disks.
2598

2599
    """
2600
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2601
    if not disks_ok:
2602
      raise errors.OpExecError("Cannot activate block devices")
2603

    
2604
    return disks_info
2605

    
2606

    
2607
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False,
2608
                           ignore_size=False):
2609
  """Prepare the block devices for an instance.
2610

2611
  This sets up the block devices on all nodes.
2612

2613
  @type lu: L{LogicalUnit}
2614
  @param lu: the logical unit on whose behalf we execute
2615
  @type instance: L{objects.Instance}
2616
  @param instance: the instance for whose disks we assemble
2617
  @type ignore_secondaries: boolean
2618
  @param ignore_secondaries: if true, errors on secondary nodes
2619
      won't result in an error return from the function
2620
  @type ignore_size: boolean
2621
  @param ignore_size: if true, the current known size of the disk
2622
      will not be used during the disk activation, useful for cases
2623
      when the size is wrong
2624
  @return: False if the operation failed, otherwise a list of
2625
      (host, instance_visible_name, node_visible_name)
2626
      with the mapping from node devices to instance devices
2627

2628
  """
2629
  device_info = []
2630
  disks_ok = True
2631
  iname = instance.name
2632
  # With the two passes mechanism we try to reduce the window of
2633
  # opportunity for the race condition of switching DRBD to primary
2634
  # before handshaking occured, but we do not eliminate it
2635

    
2636
  # The proper fix would be to wait (with some limits) until the
2637
  # connection has been made and drbd transitions from WFConnection
2638
  # into any other network-connected state (Connected, SyncTarget,
2639
  # SyncSource, etc.)
2640

    
2641
  # 1st pass, assemble on all nodes in secondary mode
2642
  for inst_disk in instance.disks:
2643
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2644
      if ignore_size:
2645
        node_disk = node_disk.Copy()
2646
        node_disk.UnsetSize()
2647
      lu.cfg.SetDiskID(node_disk, node)
2648
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2649
      msg = result.RemoteFailMsg()
2650
      if msg:
2651
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2652
                           " (is_primary=False, pass=1): %s",
2653
                           inst_disk.iv_name, node, msg)
2654
        if not ignore_secondaries:
2655
          disks_ok = False
2656

    
2657
  # FIXME: race condition on drbd migration to primary
2658

    
2659
  # 2nd pass, do only the primary node
2660
  for inst_disk in instance.disks:
2661
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2662
      if node != instance.primary_node:
2663
        continue
2664
      if ignore_size:
2665
        node_disk = node_disk.Copy()
2666
        node_disk.UnsetSize()
2667
      lu.cfg.SetDiskID(node_disk, node)
2668
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2669
      msg = result.RemoteFailMsg()
2670
      if msg:
2671
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2672
                           " (is_primary=True, pass=2): %s",
2673
                           inst_disk.iv_name, node, msg)
2674
        disks_ok = False
2675
    device_info.append((instance.primary_node, inst_disk.iv_name,
2676
                        result.payload))
2677

    
2678
  # leave the disks configured for the primary node
2679
  # this is a workaround that would be fixed better by
2680
  # improving the logical/physical id handling
2681
  for disk in instance.disks:
2682
    lu.cfg.SetDiskID(disk, instance.primary_node)
2683

    
2684
  return disks_ok, device_info
2685

    
2686

    
2687
def _StartInstanceDisks(lu, instance, force):
2688
  """Start the disks of an instance.
2689

2690
  """
2691
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2692
                                           ignore_secondaries=force)
2693
  if not disks_ok:
2694
    _ShutdownInstanceDisks(lu, instance)
2695
    if force is not None and not force:
2696
      lu.proc.LogWarning("", hint="If the message above refers to a"
2697
                         " secondary node,"
2698
                         " you can retry the operation using '--force'.")
2699
    raise errors.OpExecError("Disk consistency error")
2700

    
2701

    
2702
class LUDeactivateInstanceDisks(NoHooksLU):
2703
  """Shutdown an instance's disks.
2704

2705
  """
2706
  _OP_REQP = ["instance_name"]
2707
  REQ_BGL = False
2708

    
2709
  def ExpandNames(self):
2710
    self._ExpandAndLockInstance()
2711
    self.needed_locks[locking.LEVEL_NODE] = []
2712
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2713

    
2714
  def DeclareLocks(self, level):
2715
    if level == locking.LEVEL_NODE:
2716
      self._LockInstancesNodes()
2717

    
2718
  def CheckPrereq(self):
2719
    """Check prerequisites.
2720

2721
    This checks that the instance is in the cluster.
2722

2723
    """
2724
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2725
    assert self.instance is not None, \
2726
      "Cannot retrieve locked instance %s" % self.op.instance_name
2727

    
2728
  def Exec(self, feedback_fn):
2729
    """Deactivate the disks
2730

2731
    """
2732
    instance = self.instance
2733
    _SafeShutdownInstanceDisks(self, instance)
2734

    
2735

    
2736
def _SafeShutdownInstanceDisks(lu, instance):
2737
  """Shutdown block devices of an instance.
2738

2739
  This function checks if an instance is running, before calling
2740
  _ShutdownInstanceDisks.
2741

2742
  """
2743
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2744
                                      [instance.hypervisor])
2745
  ins_l = ins_l[instance.primary_node]
2746
  if ins_l.failed or not isinstance(ins_l.data, list):
2747
    raise errors.OpExecError("Can't contact node '%s'" %
2748
                             instance.primary_node)
2749

    
2750
  if instance.name in ins_l.data:
2751
    raise errors.OpExecError("Instance is running, can't shutdown"
2752
                             " block devices.")
2753

    
2754
  _ShutdownInstanceDisks(lu, instance)
2755

    
2756

    
2757
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2758
  """Shutdown block devices of an instance.
2759

2760
  This does the shutdown on all nodes of the instance.
2761

2762
  If the ignore_primary is false, errors on the primary node are
2763
  ignored.
2764

2765
  """
2766
  all_result = True
2767
  for disk in instance.disks:
2768
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2769
      lu.cfg.SetDiskID(top_disk, node)
2770
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2771
      msg = result.RemoteFailMsg()
2772
      if msg:
2773
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2774
                      disk.iv_name, node, msg)
2775
        if not ignore_primary or node != instance.primary_node:
2776
          all_result = False
2777
  return all_result
2778

    
2779

    
2780
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2781
  """Checks if a node has enough free memory.
2782

2783
  This function check if a given node has the needed amount of free
2784
  memory. In case the node has less memory or we cannot get the
2785
  information from the node, this function raise an OpPrereqError
2786
  exception.
2787

2788
  @type lu: C{LogicalUnit}
2789
  @param lu: a logical unit from which we get configuration data
2790
  @type node: C{str}
2791
  @param node: the node to check
2792
  @type reason: C{str}
2793
  @param reason: string to use in the error message
2794
  @type requested: C{int}
2795
  @param requested: the amount of memory in MiB to check for
2796
  @type hypervisor_name: C{str}
2797
  @param hypervisor_name: the hypervisor to ask for memory stats
2798
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2799
      we cannot check the node
2800

2801
  """
2802
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2803
  nodeinfo[node].Raise()
2804
  free_mem = nodeinfo[node].data.get('memory_free')
2805
  if not isinstance(free_mem, int):
2806
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2807
                             " was '%s'" % (node, free_mem))
2808
  if requested > free_mem:
2809
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2810
                             " needed %s MiB, available %s MiB" %
2811
                             (node, reason, requested, free_mem))
2812

    
2813

    
2814
class LUStartupInstance(LogicalUnit):
2815
  """Starts an instance.
2816

2817
  """
2818
  HPATH = "instance-start"
2819
  HTYPE = constants.HTYPE_INSTANCE
2820
  _OP_REQP = ["instance_name", "force"]
2821
  REQ_BGL = False
2822

    
2823
  def ExpandNames(self):
2824
    self._ExpandAndLockInstance()
2825

    
2826
  def BuildHooksEnv(self):
2827
    """Build hooks env.
2828

2829
    This runs on master, primary and secondary nodes of the instance.
2830

2831
    """
2832
    env = {
2833
      "FORCE": self.op.force,
2834
      }
2835
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2836
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2837
    return env, nl, nl
2838

    
2839
  def CheckPrereq(self):
2840
    """Check prerequisites.
2841

2842
    This checks that the instance is in the cluster.
2843

2844
    """
2845
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2846
    assert self.instance is not None, \
2847
      "Cannot retrieve locked instance %s" % self.op.instance_name
2848

    
2849
    # extra beparams
2850
    self.beparams = getattr(self.op, "beparams", {})
2851
    if self.beparams:
2852
      if not isinstance(self.beparams, dict):
2853
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2854
                                   " dict" % (type(self.beparams), ))
2855
      # fill the beparams dict
2856
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2857
      self.op.beparams = self.beparams
2858

    
2859
    # extra hvparams
2860
    self.hvparams = getattr(self.op, "hvparams", {})
2861
    if self.hvparams:
2862
      if not isinstance(self.hvparams, dict):
2863
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2864
                                   " dict" % (type(self.hvparams), ))
2865

    
2866
      # check hypervisor parameter syntax (locally)
2867
      cluster = self.cfg.GetClusterInfo()
2868
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2869
      filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
2870
                                    instance.hvparams)
2871
      filled_hvp.update(self.hvparams)
2872
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2873
      hv_type.CheckParameterSyntax(filled_hvp)
2874
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2875
      self.op.hvparams = self.hvparams
2876

    
2877
    _CheckNodeOnline(self, instance.primary_node)
2878

    
2879
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2880
    # check bridges existance
2881
    _CheckInstanceBridgesExist(self, instance)
2882

    
2883
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2884
                                              instance.name,
2885
                                              instance.hypervisor)
2886
    remote_info.Raise()
2887
    if not remote_info.data:
2888
      _CheckNodeFreeMemory(self, instance.primary_node,
2889
                           "starting instance %s" % instance.name,
2890
                           bep[constants.BE_MEMORY], instance.hypervisor)
2891

    
2892
  def Exec(self, feedback_fn):
2893
    """Start the instance.
2894

2895
    """
2896
    instance = self.instance
2897
    force = self.op.force
2898

    
2899
    self.cfg.MarkInstanceUp(instance.name)
2900

    
2901
    node_current = instance.primary_node
2902

    
2903
    _StartInstanceDisks(self, instance, force)
2904

    
2905
    result = self.rpc.call_instance_start(node_current, instance,
2906
                                          self.hvparams, self.beparams)
2907
    msg = result.RemoteFailMsg()
2908
    if msg:
2909
      _ShutdownInstanceDisks(self, instance)
2910
      raise errors.OpExecError("Could not start instance: %s" % msg)
2911

    
2912

    
2913
class LURebootInstance(LogicalUnit):
2914
  """Reboot an instance.
2915

2916
  """
2917
  HPATH = "instance-reboot"
2918
  HTYPE = constants.HTYPE_INSTANCE
2919
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2920
  REQ_BGL = False
2921

    
2922
  def ExpandNames(self):
2923
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2924
                                   constants.INSTANCE_REBOOT_HARD,
2925
                                   constants.INSTANCE_REBOOT_FULL]:
2926
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2927
                                  (constants.INSTANCE_REBOOT_SOFT,
2928
                                   constants.INSTANCE_REBOOT_HARD,
2929
                                   constants.INSTANCE_REBOOT_FULL))
2930
    self._ExpandAndLockInstance()
2931

    
2932
  def BuildHooksEnv(self):
2933
    """Build hooks env.
2934

2935
    This runs on master, primary and secondary nodes of the instance.
2936

2937
    """
2938
    env = {
2939
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2940
      "REBOOT_TYPE": self.op.reboot_type,
2941
      }
2942
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2943
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2944
    return env, nl, nl
2945

    
2946
  def CheckPrereq(self):
2947
    """Check prerequisites.
2948

2949
    This checks that the instance is in the cluster.
2950

2951
    """
2952
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2953
    assert self.instance is not None, \
2954
      "Cannot retrieve locked instance %s" % self.op.instance_name
2955

    
2956
    _CheckNodeOnline(self, instance.primary_node)
2957

    
2958
    # check bridges existance
2959
    _CheckInstanceBridgesExist(self, instance)
2960

    
2961
  def Exec(self, feedback_fn):
2962
    """Reboot the instance.
2963

2964
    """
2965
    instance = self.instance
2966
    ignore_secondaries = self.op.ignore_secondaries
2967
    reboot_type = self.op.reboot_type
2968

    
2969
    node_current = instance.primary_node
2970

    
2971
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2972
                       constants.INSTANCE_REBOOT_HARD]:
2973
      for disk in instance.disks:
2974
        self.cfg.SetDiskID(disk, node_current)
2975
      result = self.rpc.call_instance_reboot(node_current, instance,
2976
                                             reboot_type)
2977
      msg = result.RemoteFailMsg()
2978
      if msg:
2979
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2980
    else:
2981
      result = self.rpc.call_instance_shutdown(node_current, instance)
2982
      msg = result.RemoteFailMsg()
2983
      if msg:
2984
        raise errors.OpExecError("Could not shutdown instance for"
2985
                                 " full reboot: %s" % msg)
2986
      _ShutdownInstanceDisks(self, instance)
2987
      _StartInstanceDisks(self, instance, ignore_secondaries)
2988
      result = self.rpc.call_instance_start(node_current, instance, None, None)
2989
      msg = result.RemoteFailMsg()
2990
      if msg:
2991
        _ShutdownInstanceDisks(self, instance)
2992
        raise errors.OpExecError("Could not start instance for"
2993
                                 " full reboot: %s" % msg)
2994

    
2995
    self.cfg.MarkInstanceUp(instance.name)
2996

    
2997

    
2998
class LUShutdownInstance(LogicalUnit):
2999
  """Shutdown an instance.
3000

3001
  """
3002
  HPATH = "instance-stop"
3003
  HTYPE = constants.HTYPE_INSTANCE
3004
  _OP_REQP = ["instance_name"]
3005
  REQ_BGL = False
3006

    
3007
  def ExpandNames(self):
3008
    self._ExpandAndLockInstance()
3009

    
3010
  def BuildHooksEnv(self):
3011
    """Build hooks env.
3012

3013
    This runs on master, primary and secondary nodes of the instance.
3014

3015
    """
3016
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3017
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3018
    return env, nl, nl
3019

    
3020
  def CheckPrereq(self):
3021
    """Check prerequisites.
3022

3023
    This checks that the instance is in the cluster.
3024

3025
    """
3026
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3027
    assert self.instance is not None, \
3028
      "Cannot retrieve locked instance %s" % self.op.instance_name
3029
    _CheckNodeOnline(self, self.instance.primary_node)
3030

    
3031
  def Exec(self, feedback_fn):
3032
    """Shutdown the instance.
3033

3034
    """
3035
    instance = self.instance
3036
    node_current = instance.primary_node
3037
    self.cfg.MarkInstanceDown(instance.name)
3038
    result = self.rpc.call_instance_shutdown(node_current, instance)
3039
    msg = result.RemoteFailMsg()
3040
    if msg:
3041
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3042

    
3043
    _ShutdownInstanceDisks(self, instance)
3044

    
3045

    
3046
class LUReinstallInstance(LogicalUnit):
3047
  """Reinstall an instance.
3048

3049
  """
3050
  HPATH = "instance-reinstall"
3051
  HTYPE = constants.HTYPE_INSTANCE
3052
  _OP_REQP = ["instance_name"]
3053
  REQ_BGL = False
3054

    
3055
  def ExpandNames(self):
3056
    self._ExpandAndLockInstance()
3057

    
3058
  def BuildHooksEnv(self):
3059
    """Build hooks env.
3060

3061
    This runs on master, primary and secondary nodes of the instance.
3062

3063
    """
3064
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3065
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3066
    return env, nl, nl
3067

    
3068
  def CheckPrereq(self):
3069
    """Check prerequisites.
3070

3071
    This checks that the instance is in the cluster and is not running.
3072

3073
    """
3074
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3075
    assert instance is not None, \
3076
      "Cannot retrieve locked instance %s" % self.op.instance_name
3077
    _CheckNodeOnline(self, instance.primary_node)
3078

    
3079
    if instance.disk_template == constants.DT_DISKLESS:
3080
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3081
                                 self.op.instance_name)
3082
    if instance.admin_up:
3083
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3084
                                 self.op.instance_name)
3085
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3086
                                              instance.name,
3087
                                              instance.hypervisor)
3088
    remote_info.Raise()
3089
    if remote_info.data:
3090
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3091
                                 (self.op.instance_name,
3092
                                  instance.primary_node))
3093

    
3094
    self.op.os_type = getattr(self.op, "os_type", None)
3095
    if self.op.os_type is not None:
3096
      # OS verification
3097
      pnode = self.cfg.GetNodeInfo(
3098
        self.cfg.ExpandNodeName(instance.primary_node))
3099
      if pnode is None:
3100
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3101
                                   self.op.pnode)
3102
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3103
      result.Raise()
3104
      if not isinstance(result.data, objects.OS):
3105
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3106
                                   " primary node"  % self.op.os_type)
3107

    
3108
    self.instance = instance
3109

    
3110
  def Exec(self, feedback_fn):
3111
    """Reinstall the instance.
3112

3113
    """
3114
    inst = self.instance
3115

    
3116
    if self.op.os_type is not None:
3117
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3118
      inst.os = self.op.os_type
3119
      self.cfg.Update(inst)
3120

    
3121
    _StartInstanceDisks(self, inst, None)
3122
    try:
3123
      feedback_fn("Running the instance OS create scripts...")
3124
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3125
      msg = result.RemoteFailMsg()
3126
      if msg:
3127
        raise errors.OpExecError("Could not install OS for instance %s"
3128
                                 " on node %s: %s" %
3129
                                 (inst.name, inst.primary_node, msg))
3130
    finally:
3131
      _ShutdownInstanceDisks(self, inst)
3132

    
3133

    
3134
class LURenameInstance(LogicalUnit):
3135
  """Rename an instance.
3136

3137
  """
3138
  HPATH = "instance-rename"
3139
  HTYPE = constants.HTYPE_INSTANCE
3140
  _OP_REQP = ["instance_name", "new_name"]
3141

    
3142
  def BuildHooksEnv(self):
3143
    """Build hooks env.
3144

3145
    This runs on master, primary and secondary nodes of the instance.
3146

3147
    """
3148
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3149
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3150
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3151
    return env, nl, nl
3152

    
3153
  def CheckPrereq(self):
3154
    """Check prerequisites.
3155

3156
    This checks that the instance is in the cluster and is not running.
3157

3158
    """
3159
    instance = self.cfg.GetInstanceInfo(
3160
      self.cfg.ExpandInstanceName(self.op.instance_name))
3161
    if instance is None:
3162
      raise errors.OpPrereqError("Instance '%s' not known" %
3163
                                 self.op.instance_name)
3164
    _CheckNodeOnline(self, instance.primary_node)
3165

    
3166
    if instance.admin_up:
3167
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3168
                                 self.op.instance_name)
3169
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3170
                                              instance.name,
3171
                                              instance.hypervisor)
3172
    remote_info.Raise()
3173
    if remote_info.data:
3174
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3175
                                 (self.op.instance_name,
3176
                                  instance.primary_node))
3177
    self.instance = instance
3178

    
3179
    # new name verification
3180
    name_info = utils.HostInfo(self.op.new_name)
3181

    
3182
    self.op.new_name = new_name = name_info.name
3183
    instance_list = self.cfg.GetInstanceList()
3184
    if new_name in instance_list:
3185
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3186
                                 new_name)
3187

    
3188
    if not getattr(self.op, "ignore_ip", False):
3189
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3190
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3191
                                   (name_info.ip, new_name))
3192

    
3193

    
3194
  def Exec(self, feedback_fn):
3195
    """Reinstall the instance.
3196

3197
    """
3198
    inst = self.instance
3199
    old_name = inst.name
3200

    
3201
    if inst.disk_template == constants.DT_FILE:
3202
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3203

    
3204
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3205
    # Change the instance lock. This is definitely safe while we hold the BGL
3206
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3207
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3208

    
3209
    # re-read the instance from the configuration after rename
3210
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3211

    
3212
    if inst.disk_template == constants.DT_FILE:
3213
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3214
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3215
                                                     old_file_storage_dir,
3216
                                                     new_file_storage_dir)
3217
      result.Raise()
3218
      if not result.data:
3219
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3220
                                 " directory '%s' to '%s' (but the instance"
3221
                                 " has been renamed in Ganeti)" % (
3222
                                 inst.primary_node, old_file_storage_dir,
3223
                                 new_file_storage_dir))
3224

    
3225
      if not result.data[0]:
3226
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3227
                                 " (but the instance has been renamed in"
3228
                                 " Ganeti)" % (old_file_storage_dir,
3229
                                               new_file_storage_dir))
3230

    
3231
    _StartInstanceDisks(self, inst, None)
3232
    try:
3233
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3234
                                                 old_name)
3235
      msg = result.RemoteFailMsg()
3236
      if msg:
3237
        msg = ("Could not run OS rename script for instance %s on node %s"
3238
               " (but the instance has been renamed in Ganeti): %s" %
3239
               (inst.name, inst.primary_node, msg))
3240
        self.proc.LogWarning(msg)
3241
    finally:
3242
      _ShutdownInstanceDisks(self, inst)
3243

    
3244

    
3245
class LURemoveInstance(LogicalUnit):
3246
  """Remove an instance.
3247

3248
  """
3249
  HPATH = "instance-remove"
3250
  HTYPE = constants.HTYPE_INSTANCE
3251
  _OP_REQP = ["instance_name", "ignore_failures"]
3252
  REQ_BGL = False
3253

    
3254
  def ExpandNames(self):
3255
    self._ExpandAndLockInstance()
3256
    self.needed_locks[locking.LEVEL_NODE] = []
3257
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3258

    
3259
  def DeclareLocks(self, level):
3260
    if level == locking.LEVEL_NODE:
3261
      self._LockInstancesNodes()
3262

    
3263
  def BuildHooksEnv(self):
3264
    """Build hooks env.
3265

3266
    This runs on master, primary and secondary nodes of the instance.
3267

3268
    """
3269
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3270
    nl = [self.cfg.GetMasterNode()]
3271
    return env, nl, nl
3272

    
3273
  def CheckPrereq(self):
3274
    """Check prerequisites.
3275

3276
    This checks that the instance is in the cluster.
3277

3278
    """
3279
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3280
    assert self.instance is not None, \
3281
      "Cannot retrieve locked instance %s" % self.op.instance_name
3282

    
3283
  def Exec(self, feedback_fn):
3284
    """Remove the instance.
3285

3286
    """
3287
    instance = self.instance
3288
    logging.info("Shutting down instance %s on node %s",
3289
                 instance.name, instance.primary_node)
3290

    
3291
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3292
    msg = result.RemoteFailMsg()
3293
    if msg:
3294
      if self.op.ignore_failures:
3295
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3296
      else:
3297
        raise errors.OpExecError("Could not shutdown instance %s on"
3298
                                 " node %s: %s" %
3299
                                 (instance.name, instance.primary_node, msg))
3300

    
3301
    logging.info("Removing block devices for instance %s", instance.name)
3302

    
3303
    if not _RemoveDisks(self, instance):
3304
      if self.op.ignore_failures:
3305
        feedback_fn("Warning: can't remove instance's disks")
3306
      else:
3307
        raise errors.OpExecError("Can't remove instance's disks")
3308

    
3309
    logging.info("Removing instance %s out of cluster config", instance.name)
3310

    
3311
    self.cfg.RemoveInstance(instance.name)
3312
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3313

    
3314

    
3315
class LUQueryInstances(NoHooksLU):
3316
  """Logical unit for querying instances.
3317

3318
  """
3319
  _OP_REQP = ["output_fields", "names", "use_locking"]
3320
  REQ_BGL = False
3321
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3322
                                    "admin_state",
3323
                                    "disk_template", "ip", "mac", "bridge",
3324
                                    "sda_size", "sdb_size", "vcpus", "tags",
3325
                                    "network_port", "beparams",
3326
                                    r"(disk)\.(size)/([0-9]+)",
3327
                                    r"(disk)\.(sizes)", "disk_usage",
3328
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3329
                                    r"(nic)\.(macs|ips|bridges)",
3330
                                    r"(disk|nic)\.(count)",
3331
                                    "serial_no", "hypervisor", "hvparams",] +
3332
                                  ["hv/%s" % name
3333
                                   for name in constants.HVS_PARAMETERS] +
3334
                                  ["be/%s" % name
3335
                                   for name in constants.BES_PARAMETERS])
3336
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3337

    
3338

    
3339
  def ExpandNames(self):
3340
    _CheckOutputFields(static=self._FIELDS_STATIC,
3341
                       dynamic=self._FIELDS_DYNAMIC,
3342
                       selected=self.op.output_fields)
3343

    
3344
    self.needed_locks = {}
3345
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3346
    self.share_locks[locking.LEVEL_NODE] = 1
3347

    
3348
    if self.op.names:
3349
      self.wanted = _GetWantedInstances(self, self.op.names)
3350
    else:
3351
      self.wanted = locking.ALL_SET
3352

    
3353
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3354
    self.do_locking = self.do_node_query and self.op.use_locking
3355
    if self.do_locking:
3356
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3357
      self.needed_locks[locking.LEVEL_NODE] = []
3358
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3359

    
3360
  def DeclareLocks(self, level):
3361
    if level == locking.LEVEL_NODE and self.do_locking:
3362
      self._LockInstancesNodes()
3363

    
3364
  def CheckPrereq(self):
3365
    """Check prerequisites.
3366

3367
    """
3368
    pass
3369

    
3370
  def Exec(self, feedback_fn):
3371
    """Computes the list of nodes and their attributes.
3372

3373
    """
3374
    all_info = self.cfg.GetAllInstancesInfo()
3375
    if self.wanted == locking.ALL_SET:
3376
      # caller didn't specify instance names, so ordering is not important
3377
      if self.do_locking:
3378
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3379
      else:
3380
        instance_names = all_info.keys()
3381
      instance_names = utils.NiceSort(instance_names)
3382
    else:
3383
      # caller did specify names, so we must keep the ordering
3384
      if self.do_locking:
3385
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3386
      else:
3387
        tgt_set = all_info.keys()
3388
      missing = set(self.wanted).difference(tgt_set)
3389
      if missing:
3390
        raise errors.OpExecError("Some instances were removed before"
3391
                                 " retrieving their data: %s" % missing)
3392
      instance_names = self.wanted
3393

    
3394
    instance_list = [all_info[iname] for iname in instance_names]
3395

    
3396
    # begin data gathering
3397

    
3398
    nodes = frozenset([inst.primary_node for inst in instance_list])
3399
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3400

    
3401
    bad_nodes = []
3402
    off_nodes = []
3403
    if self.do_node_query:
3404
      live_data = {}
3405
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3406
      for name in nodes:
3407
        result = node_data[name]
3408
        if result.offline:
3409
          # offline nodes will be in both lists
3410
          off_nodes.append(name)
3411
        if result.failed:
3412
          bad_nodes.append(name)
3413
        else:
3414
          if result.data:
3415
            live_data.update(result.data)
3416
            # else no instance is alive
3417
    else:
3418
      live_data = dict([(name, {}) for name in instance_names])
3419

    
3420
    # end data gathering
3421

    
3422
    HVPREFIX = "hv/"
3423
    BEPREFIX = "be/"
3424
    output = []
3425
    for instance in instance_list:
3426
      iout = []
3427
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3428
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3429
      for field in self.op.output_fields:
3430
        st_match = self._FIELDS_STATIC.Matches(field)
3431
        if field == "name":
3432
          val = instance.name
3433
        elif field == "os":
3434
          val = instance.os
3435
        elif field == "pnode":
3436
          val = instance.primary_node
3437
        elif field == "snodes":
3438
          val = list(instance.secondary_nodes)
3439
        elif field == "admin_state":
3440
          val = instance.admin_up
3441
        elif field == "oper_state":
3442
          if instance.primary_node in bad_nodes:
3443
            val = None
3444
          else:
3445
            val = bool(live_data.get(instance.name))
3446
        elif field == "status":
3447
          if instance.primary_node in off_nodes:
3448
            val = "ERROR_nodeoffline"
3449
          elif instance.primary_node in bad_nodes:
3450
            val = "ERROR_nodedown"
3451
          else:
3452
            running = bool(live_data.get(instance.name))
3453
            if running:
3454
              if instance.admin_up:
3455
                val = "running"
3456
              else:
3457
                val = "ERROR_up"
3458
            else:
3459
              if instance.admin_up:
3460
                val = "ERROR_down"
3461
              else:
3462
                val = "ADMIN_down"
3463
        elif field == "oper_ram":
3464
          if instance.primary_node in bad_nodes:
3465
            val = None
3466
          elif instance.name in live_data:
3467
            val = live_data[instance.name].get("memory", "?")
3468
          else:
3469
            val = "-"
3470
        elif field == "vcpus":
3471
          val = i_be[constants.BE_VCPUS]
3472
        elif field == "disk_template":
3473
          val = instance.disk_template
3474
        elif field == "ip":
3475
          if instance.nics:
3476
            val = instance.nics[0].ip
3477
          else:
3478
            val = None
3479
        elif field == "bridge":
3480
          if instance.nics:
3481
            val = instance.nics[0].bridge
3482
          else:
3483
            val = None
3484
        elif field == "mac":
3485
          if instance.nics:
3486
            val = instance.nics[0].mac
3487
          else:
3488
            val = None
3489
        elif field == "sda_size" or field == "sdb_size":
3490
          idx = ord(field[2]) - ord('a')
3491
          try:
3492
            val = instance.FindDisk(idx).size
3493
          except errors.OpPrereqError:
3494
            val = None
3495
        elif field == "disk_usage": # total disk usage per node
3496
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3497
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3498
        elif field == "tags":
3499
          val = list(instance.GetTags())
3500
        elif field == "serial_no":
3501
          val = instance.serial_no
3502
        elif field == "network_port":
3503
          val = instance.network_port
3504
        elif field == "hypervisor":
3505
          val = instance.hypervisor
3506
        elif field == "hvparams":
3507
          val = i_hv
3508
        elif (field.startswith(HVPREFIX) and
3509
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3510
          val = i_hv.get(field[len(HVPREFIX):], None)
3511
        elif field == "beparams":
3512
          val = i_be
3513
        elif (field.startswith(BEPREFIX) and
3514
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3515
          val = i_be.get(field[len(BEPREFIX):], None)
3516
        elif st_match and st_match.groups():
3517
          # matches a variable list
3518
          st_groups = st_match.groups()
3519
          if st_groups and st_groups[0] == "disk":
3520
            if st_groups[1] == "count":
3521
              val = len(instance.disks)
3522
            elif st_groups[1] == "sizes":
3523
              val = [disk.size for disk in instance.disks]
3524
            elif st_groups[1] == "size":
3525
              try:
3526
                val = instance.FindDisk(st_groups[2]).size
3527
              except errors.OpPrereqError:
3528
                val = None
3529
            else:
3530
              assert False, "Unhandled disk parameter"
3531
          elif st_groups[0] == "nic":
3532
            if st_groups[1] == "count":
3533
              val = len(instance.nics)
3534
            elif st_groups[1] == "macs":
3535
              val = [nic.mac for nic in instance.nics]
3536
            elif st_groups[1] == "ips":
3537
              val = [nic.ip for nic in instance.nics]
3538
            elif st_groups[1] == "bridges":
3539
              val = [nic.bridge for nic in instance.nics]
3540
            else:
3541
              # index-based item
3542
              nic_idx = int(st_groups[2])
3543
              if nic_idx >= len(instance.nics):
3544
                val = None
3545
              else:
3546
                if st_groups[1] == "mac":
3547
                  val = instance.nics[nic_idx].mac
3548
                elif st_groups[1] == "ip":
3549
                  val = instance.nics[nic_idx].ip
3550
                elif st_groups[1] == "bridge":
3551
                  val = instance.nics[nic_idx].bridge
3552
                else:
3553
                  assert False, "Unhandled NIC parameter"
3554
          else:
3555
            assert False, ("Declared but unhandled variable parameter '%s'" %
3556
                           field)
3557
        else:
3558
          assert False, "Declared but unhandled parameter '%s'" % field
3559
        iout.append(val)
3560
      output.append(iout)
3561

    
3562
    return output
3563

    
3564

    
3565
class LUFailoverInstance(LogicalUnit):
3566
  """Failover an instance.
3567

3568
  """
3569
  HPATH = "instance-failover"
3570
  HTYPE = constants.HTYPE_INSTANCE
3571
  _OP_REQP = ["instance_name", "ignore_consistency"]
3572
  REQ_BGL = False
3573

    
3574
  def ExpandNames(self):
3575
    self._ExpandAndLockInstance()
3576
    self.needed_locks[locking.LEVEL_NODE] = []
3577
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3578

    
3579
  def DeclareLocks(self, level):
3580
    if level == locking.LEVEL_NODE:
3581
      self._LockInstancesNodes()
3582

    
3583
  def BuildHooksEnv(self):
3584
    """Build hooks env.
3585

3586
    This runs on master, primary and secondary nodes of the instance.
3587

3588
    """
3589
    env = {
3590
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3591
      }
3592
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3593
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3594
    return env, nl, nl
3595

    
3596
  def CheckPrereq(self):
3597
    """Check prerequisites.
3598

3599
    This checks that the instance is in the cluster.
3600

3601
    """
3602
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3603
    assert self.instance is not None, \
3604
      "Cannot retrieve locked instance %s" % self.op.instance_name
3605

    
3606
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3607
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3608
      raise errors.OpPrereqError("Instance's disk layout is not"
3609
                                 " network mirrored, cannot failover.")
3610

    
3611
    secondary_nodes = instance.secondary_nodes
3612
    if not secondary_nodes:
3613
      raise errors.ProgrammerError("no secondary node but using "
3614
                                   "a mirrored disk template")
3615

    
3616
    target_node = secondary_nodes[0]
3617
    _CheckNodeOnline(self, target_node)
3618
    _CheckNodeNotDrained(self, target_node)
3619

    
3620
    if instance.admin_up:
3621
      # check memory requirements on the secondary node
3622
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3623
                           instance.name, bep[constants.BE_MEMORY],
3624
                           instance.hypervisor)
3625
    else:
3626
      self.LogInfo("Not checking memory on the secondary node as"
3627
                   " instance will not be started")
3628

    
3629
    # check bridge existance
3630
    brlist = [nic.bridge for nic in instance.nics]
3631
    result = self.rpc.call_bridges_exist(target_node, brlist)
3632
    result.Raise()
3633
    if not result.data:
3634
      raise errors.OpPrereqError("One or more target bridges %s does not"
3635
                                 " exist on destination node '%s'" %
3636
                                 (brlist, target_node))
3637

    
3638
  def Exec(self, feedback_fn):
3639
    """Failover an instance.
3640

3641
    The failover is done by shutting it down on its present node and
3642
    starting it on the secondary.
3643

3644
    """
3645
    instance = self.instance
3646

    
3647
    source_node = instance.primary_node
3648
    target_node = instance.secondary_nodes[0]
3649

    
3650
    feedback_fn("* checking disk consistency between source and target")
3651
    for dev in instance.disks:
3652
      # for drbd, these are drbd over lvm
3653
      if not _CheckDiskConsistency(self, dev, target_node, False):
3654
        if instance.admin_up and not self.op.ignore_consistency:
3655
          raise errors.OpExecError("Disk %s is degraded on target node,"
3656
                                   " aborting failover." % dev.iv_name)
3657

    
3658
    feedback_fn("* shutting down instance on source node")
3659
    logging.info("Shutting down instance %s on node %s",
3660
                 instance.name, source_node)
3661

    
3662
    result = self.rpc.call_instance_shutdown(source_node, instance)
3663
    msg = result.RemoteFailMsg()
3664
    if msg:
3665
      if self.op.ignore_consistency:
3666
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3667
                             " Proceeding anyway. Please make sure node"
3668
                             " %s is down. Error details: %s",
3669
                             instance.name, source_node, source_node, msg)
3670
      else:
3671
        raise errors.OpExecError("Could not shutdown instance %s on"
3672
                                 " node %s: %s" %
3673
                                 (instance.name, source_node, msg))
3674

    
3675
    feedback_fn("* deactivating the instance's disks on source node")
3676
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3677
      raise errors.OpExecError("Can't shut down the instance's disks.")
3678

    
3679
    instance.primary_node = target_node
3680
    # distribute new instance config to the other nodes
3681
    self.cfg.Update(instance)
3682

    
3683
    # Only start the instance if it's marked as up
3684
    if instance.admin_up:
3685
      feedback_fn("* activating the instance's disks on target node")
3686
      logging.info("Starting instance %s on node %s",
3687
                   instance.name, target_node)
3688

    
3689
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3690
                                               ignore_secondaries=True)
3691
      if not disks_ok:
3692
        _ShutdownInstanceDisks(self, instance)
3693
        raise errors.OpExecError("Can't activate the instance's disks")
3694

    
3695
      feedback_fn("* starting the instance on the target node")
3696
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3697
      msg = result.RemoteFailMsg()
3698
      if msg:
3699
        _ShutdownInstanceDisks(self, instance)
3700
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3701
                                 (instance.name, target_node, msg))
3702

    
3703

    
3704
class LUMigrateInstance(LogicalUnit):
3705
  """Migrate an instance.
3706

3707
  This is migration without shutting down, compared to the failover,
3708
  which is done with shutdown.
3709

3710
  """
3711
  HPATH = "instance-migrate"
3712
  HTYPE = constants.HTYPE_INSTANCE
3713
  _OP_REQP = ["instance_name", "live", "cleanup"]
3714

    
3715
  REQ_BGL = False
3716

    
3717
  def ExpandNames(self):
3718
    self._ExpandAndLockInstance()
3719
    self.needed_locks[locking.LEVEL_NODE] = []
3720
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3721

    
3722
  def DeclareLocks(self, level):
3723
    if level == locking.LEVEL_NODE:
3724
      self._LockInstancesNodes()
3725

    
3726
  def BuildHooksEnv(self):
3727
    """Build hooks env.
3728

3729
    This runs on master, primary and secondary nodes of the instance.
3730

3731
    """
3732
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3733
    env["MIGRATE_LIVE"] = self.op.live
3734
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3735
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3736
    return env, nl, nl
3737

    
3738
  def CheckPrereq(self):
3739
    """Check prerequisites.
3740

3741
    This checks that the instance is in the cluster.
3742

3743
    """
3744
    instance = self.cfg.GetInstanceInfo(
3745
      self.cfg.ExpandInstanceName(self.op.instance_name))
3746
    if instance is None:
3747
      raise errors.OpPrereqError("Instance '%s' not known" %
3748
                                 self.op.instance_name)
3749

    
3750
    if instance.disk_template != constants.DT_DRBD8:
3751
      raise errors.OpPrereqError("Instance's disk layout is not"
3752
                                 " drbd8, cannot migrate.")
3753

    
3754
    secondary_nodes = instance.secondary_nodes
3755
    if not secondary_nodes:
3756
      raise errors.ConfigurationError("No secondary node but using"
3757
                                      " drbd8 disk template")
3758

    
3759
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3760

    
3761
    target_node = secondary_nodes[0]
3762
    # check memory requirements on the secondary node
3763
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3764
                         instance.name, i_be[constants.BE_MEMORY],
3765
                         instance.hypervisor)
3766

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

    
3775
    if not self.op.cleanup:
3776
      _CheckNodeNotDrained(self, target_node)
3777
      result = self.rpc.call_instance_migratable(instance.primary_node,
3778
                                                 instance)
3779
      msg = result.RemoteFailMsg()
3780
      if msg:
3781
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3782
                                   msg)
3783

    
3784
    self.instance = instance
3785

    
3786
  def _WaitUntilSync(self):
3787
    """Poll with custom rpc for disk sync.
3788

3789
    This uses our own step-based rpc call.
3790

3791
    """
3792
    self.feedback_fn("* wait until resync is done")
3793
    all_done = False
3794
    while not all_done:
3795
      all_done = True
3796
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3797
                                            self.nodes_ip,
3798
                                            self.instance.disks)
3799
      min_percent = 100
3800
      for node, nres in result.items():
3801
        msg = nres.RemoteFailMsg()
3802
        if msg:
3803
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3804
                                   (node, msg))
3805
        node_done, node_percent = nres.payload
3806
        all_done = all_done and node_done
3807
        if node_percent is not None:
3808
          min_percent = min(min_percent, node_percent)
3809
      if not all_done:
3810
        if min_percent < 100:
3811
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3812
        time.sleep(2)
3813

    
3814
  def _EnsureSecondary(self, node):
3815
    """Demote a node to secondary.
3816

3817
    """
3818
    self.feedback_fn("* switching node %s to secondary mode" % node)
3819

    
3820
    for dev in self.instance.disks:
3821
      self.cfg.SetDiskID(dev, node)
3822

    
3823
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3824
                                          self.instance.disks)
3825
    msg = result.RemoteFailMsg()
3826
    if msg:
3827
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3828
                               " error %s" % (node, msg))
3829

    
3830
  def _GoStandalone(self):
3831
    """Disconnect from the network.
3832

3833
    """
3834
    self.feedback_fn("* changing into standalone mode")
3835
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3836
                                               self.instance.disks)
3837
    for node, nres in result.items():
3838
      msg = nres.RemoteFailMsg()
3839
      if msg:
3840
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3841
                                 " error %s" % (node, msg))
3842

    
3843
  def _GoReconnect(self, multimaster):
3844
    """Reconnect to the network.
3845

3846
    """
3847
    if multimaster:
3848
      msg = "dual-master"
3849
    else:
3850
      msg = "single-master"
3851
    self.feedback_fn("* changing disks into %s mode" % msg)
3852
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3853
                                           self.instance.disks,
3854
                                           self.instance.name, multimaster)
3855
    for node, nres in result.items():
3856
      msg = nres.RemoteFailMsg()
3857
      if msg:
3858
        raise errors.OpExecError("Cannot change disks config on node %s,"
3859
                                 " error: %s" % (node, msg))
3860

    
3861
  def _ExecCleanup(self):
3862
    """Try to cleanup after a failed migration.
3863

3864
    The cleanup is done by:
3865
      - check that the instance is running only on one node
3866
        (and update the config if needed)
3867
      - change disks on its secondary node to secondary
3868
      - wait until disks are fully synchronized
3869
      - disconnect from the network
3870
      - change disks into single-master mode
3871
      - wait again until disks are fully synchronized
3872

3873
    """
3874
    instance = self.instance
3875
    target_node = self.target_node
3876
    source_node = self.source_node
3877

    
3878
    # check running on only one node
3879
    self.feedback_fn("* checking where the instance actually runs"
3880
                     " (if this hangs, the hypervisor might be in"
3881
                     " a bad state)")
3882
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3883
    for node, result in ins_l.items():
3884
      result.Raise()
3885
      if not isinstance(result.data, list):
3886
        raise errors.OpExecError("Can't contact node '%s'" % node)
3887

    
3888
    runningon_source = instance.name in ins_l[source_node].data
3889
    runningon_target = instance.name in ins_l[target_node].data
3890

    
3891
    if runningon_source and runningon_target:
3892
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3893
                               " or the hypervisor is confused. You will have"
3894
                               " to ensure manually that it runs only on one"
3895
                               " and restart this operation.")
3896

    
3897
    if not (runningon_source or runningon_target):
3898
      raise errors.OpExecError("Instance does not seem to be running at all."
3899
                               " In this case, it's safer to repair by"
3900
                               " running 'gnt-instance stop' to ensure disk"
3901
                               " shutdown, and then restarting it.")
3902

    
3903
    if runningon_target:
3904
      # the migration has actually succeeded, we need to update the config
3905
      self.feedback_fn("* instance running on secondary node (%s),"
3906
                       " updating config" % target_node)
3907
      instance.primary_node = target_node
3908
      self.cfg.Update(instance)
3909
      demoted_node = source_node
3910
    else:
3911
      self.feedback_fn("* instance confirmed to be running on its"
3912
                       " primary node (%s)" % source_node)
3913
      demoted_node = target_node
3914

    
3915
    self._EnsureSecondary(demoted_node)
3916
    try:
3917
      self._WaitUntilSync()
3918
    except errors.OpExecError:
3919
      # we ignore here errors, since if the device is standalone, it
3920
      # won't be able to sync
3921
      pass
3922
    self._GoStandalone()
3923
    self._GoReconnect(False)
3924
    self._WaitUntilSync()
3925

    
3926
    self.feedback_fn("* done")
3927

    
3928
  def _RevertDiskStatus(self):
3929
    """Try to revert the disk status after a failed migration.
3930

3931
    """
3932
    target_node = self.target_node
3933
    try:
3934
      self._EnsureSecondary(target_node)
3935
      self._GoStandalone()
3936
      self._GoReconnect(False)
3937
      self._WaitUntilSync()
3938
    except errors.OpExecError, err:
3939
      self.LogWarning("Migration failed and I can't reconnect the"
3940
                      " drives: error '%s'\n"
3941
                      "Please look and recover the instance status" %
3942
                      str(err))
3943

    
3944
  def _AbortMigration(self):
3945
    """Call the hypervisor code to abort a started migration.
3946

3947
    """
3948
    instance = self.instance
3949
    target_node = self.target_node
3950
    migration_info = self.migration_info
3951

    
3952
    abort_result = self.rpc.call_finalize_migration(target_node,
3953
                                                    instance,
3954
                                                    migration_info,
3955
                                                    False)
3956
    abort_msg = abort_result.RemoteFailMsg()
3957
    if abort_msg:
3958
      logging.error("Aborting migration failed on target node %s: %s" %
3959
                    (target_node, abort_msg))
3960
      # Don't raise an exception here, as we stil have to try to revert the
3961
      # disk status, even if this step failed.
3962

    
3963
  def _ExecMigration(self):
3964
    """Migrate an instance.
3965

3966
    The migrate is done by:
3967
      - change the disks into dual-master mode
3968
      - wait until disks are fully synchronized again
3969
      - migrate the instance
3970
      - change disks on the new secondary node (the old primary) to secondary
3971
      - wait until disks are fully synchronized
3972
      - change disks into single-master mode
3973

3974
    """
3975
    instance = self.instance
3976
    target_node = self.target_node
3977
    source_node = self.source_node
3978

    
3979
    self.feedback_fn("* checking disk consistency between source and target")
3980
    for dev in instance.disks:
3981
      if not _CheckDiskConsistency(self, dev, target_node, False):
3982
        raise errors.OpExecError("Disk %s is degraded or not fully"
3983
                                 " synchronized on target node,"
3984
                                 " aborting migrate." % dev.iv_name)
3985

    
3986
    # First get the migration information from the remote node
3987
    result = self.rpc.call_migration_info(source_node, instance)
3988
    msg = result.RemoteFailMsg()
3989
    if msg:
3990
      log_err = ("Failed fetching source migration information from %s: %s" %
3991
                 (source_node, msg))
3992
      logging.error(log_err)
3993
      raise errors.OpExecError(log_err)
3994

    
3995
    self.migration_info = migration_info = result.payload
3996

    
3997
    # Then switch the disks to master/master mode
3998
    self._EnsureSecondary(target_node)
3999
    self._GoStandalone()
4000
    self._GoReconnect(True)
4001
    self._WaitUntilSync()
4002

    
4003
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
4004
    result = self.rpc.call_accept_instance(target_node,
4005
                                           instance,
4006
                                           migration_info,
4007
                                           self.nodes_ip[target_node])
4008

    
4009
    msg = result.RemoteFailMsg()
4010
    if msg:
4011
      logging.error("Instance pre-migration failed, trying to revert"
4012
                    " disk status: %s", msg)
4013
      self._AbortMigration()
4014
      self._RevertDiskStatus()
4015
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
4016
                               (instance.name, msg))
4017

    
4018
    self.feedback_fn("* migrating instance to %s" % target_node)
4019
    time.sleep(10)
4020
    result = self.rpc.call_instance_migrate(source_node, instance,
4021
                                            self.nodes_ip[target_node],
4022
                                            self.op.live)
4023
    msg = result.RemoteFailMsg()
4024
    if msg:
4025
      logging.error("Instance migration failed, trying to revert"
4026
                    " disk status: %s", msg)
4027
      self._AbortMigration()
4028
      self._RevertDiskStatus()
4029
      raise errors.OpExecError("Could not migrate instance %s: %s" %
4030
                               (instance.name, msg))
4031
    time.sleep(10)
4032

    
4033
    instance.primary_node = target_node
4034
    # distribute new instance config to the other nodes
4035
    self.cfg.Update(instance)
4036

    
4037
    result = self.rpc.call_finalize_migration(target_node,
4038
                                              instance,
4039
                                              migration_info,
4040
                                              True)
4041
    msg = result.RemoteFailMsg()
4042
    if msg:
4043
      logging.error("Instance migration succeeded, but finalization failed:"
4044
                    " %s" % msg)
4045
      raise errors.OpExecError("Could not finalize instance migration: %s" %
4046
                               msg)
4047

    
4048
    self._EnsureSecondary(source_node)
4049
    self._WaitUntilSync()
4050
    self._GoStandalone()
4051
    self._GoReconnect(False)
4052
    self._WaitUntilSync()
4053

    
4054
    self.feedback_fn("* done")
4055

    
4056
  def Exec(self, feedback_fn):
4057
    """Perform the migration.
4058

4059
    """
4060
    self.feedback_fn = feedback_fn
4061

    
4062
    self.source_node = self.instance.primary_node
4063
    self.target_node = self.instance.secondary_nodes[0]
4064
    self.all_nodes = [self.source_node, self.target_node]
4065
    self.nodes_ip = {
4066
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
4067
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
4068
      }
4069
    if self.op.cleanup:
4070
      return self._ExecCleanup()
4071
    else:
4072
      return self._ExecMigration()
4073

    
4074

    
4075
def _CreateBlockDev(lu, node, instance, device, force_create,
4076
                    info, force_open):
4077
  """Create a tree of block devices on a given node.
4078

4079
  If this device type has to be created on secondaries, create it and
4080
  all its children.
4081

4082
  If not, just recurse to children keeping the same 'force' value.
4083

4084
  @param lu: the lu on whose behalf we execute
4085
  @param node: the node on which to create the device
4086
  @type instance: L{objects.Instance}
4087
  @param instance: the instance which owns the device
4088
  @type device: L{objects.Disk}
4089
  @param device: the device to create
4090
  @type force_create: boolean
4091
  @param force_create: whether to force creation of this device; this
4092
      will be change to True whenever we find a device which has
4093
      CreateOnSecondary() attribute
4094
  @param info: the extra 'metadata' we should attach to the device
4095
      (this will be represented as a LVM tag)
4096
  @type force_open: boolean
4097
  @param force_open: this parameter will be passes to the
4098
      L{backend.BlockdevCreate} function where it specifies
4099
      whether we run on primary or not, and it affects both
4100
      the child assembly and the device own Open() execution
4101

4102
  """
4103
  if device.CreateOnSecondary():
4104
    force_create = True
4105

    
4106
  if device.children:
4107
    for child in device.children:
4108
      _CreateBlockDev(lu, node, instance, child, force_create,
4109
                      info, force_open)
4110

    
4111
  if not force_create:
4112
    return
4113

    
4114
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
4115

    
4116

    
4117
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
4118
  """Create a single block device on a given node.
4119

4120
  This will not recurse over children of the device, so they must be
4121
  created in advance.
4122

4123
  @param lu: the lu on whose behalf we execute
4124
  @param node: the node on which to create the device
4125
  @type instance: L{objects.Instance}
4126
  @param instance: the instance which owns the device
4127
  @type device: L{objects.Disk}
4128
  @param device: the device to create
4129
  @param info: the extra 'metadata' we should attach to the device
4130
      (this will be represented as a LVM tag)
4131
  @type force_open: boolean
4132
  @param force_open: this parameter will be passes to the
4133
      L{backend.BlockdevCreate} function where it specifies
4134
      whether we run on primary or not, and it affects both
4135
      the child assembly and the device own Open() execution
4136

4137
  """
4138
  lu.cfg.SetDiskID(device, node)
4139
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4140
                                       instance.name, force_open, info)
4141
  msg = result.RemoteFailMsg()
4142
  if msg:
4143
    raise errors.OpExecError("Can't create block device %s on"
4144
                             " node %s for instance %s: %s" %
4145
                             (device, node, instance.name, msg))
4146
  if device.physical_id is None:
4147
    device.physical_id = result.payload
4148

    
4149

    
4150
def _GenerateUniqueNames(lu, exts):
4151
  """Generate a suitable LV name.
4152

4153
  This will generate a logical volume name for the given instance.
4154

4155
  """
4156
  results = []
4157
  for val in exts:
4158
    new_id = lu.cfg.GenerateUniqueID()
4159
    results.append("%s%s" % (new_id, val))
4160
  return results
4161

    
4162

    
4163
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4164
                         p_minor, s_minor):
4165
  """Generate a drbd8 device complete with its children.
4166

4167
  """
4168
  port = lu.cfg.AllocatePort()
4169
  vgname = lu.cfg.GetVGName()
4170
  shared_secret = lu.cfg.GenerateDRBDSecret()
4171
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4172
                          logical_id=(vgname, names[0]))
4173
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4174
                          logical_id=(vgname, names[1]))
4175
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4176
                          logical_id=(primary, secondary, port,
4177
                                      p_minor, s_minor,
4178
                                      shared_secret),
4179
                          children=[dev_data, dev_meta],
4180
                          iv_name=iv_name)
4181
  return drbd_dev
4182

    
4183

    
4184
def _GenerateDiskTemplate(lu, template_name,
4185
                          instance_name, primary_node,
4186
                          secondary_nodes, disk_info,
4187
                          file_storage_dir, file_driver,
4188
                          base_index):
4189
  """Generate the entire disk layout for a given template type.
4190

4191
  """
4192
  #TODO: compute space requirements
4193

    
4194
  vgname = lu.cfg.GetVGName()
4195
  disk_count = len(disk_info)
4196
  disks = []
4197
  if template_name == constants.DT_DISKLESS:
4198
    pass
4199
  elif template_name == constants.DT_PLAIN:
4200
    if len(secondary_nodes) != 0:
4201
      raise errors.ProgrammerError("Wrong template configuration")
4202

    
4203
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4204
                                      for i in range(disk_count)])
4205
    for idx, disk in enumerate(disk_info):
4206
      disk_index = idx + base_index
4207
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4208
                              logical_id=(vgname, names[idx]),
4209
                              iv_name="disk/%d" % disk_index,
4210
                              mode=disk["mode"])
4211
      disks.append(disk_dev)
4212
  elif template_name == constants.DT_DRBD8:
4213
    if len(secondary_nodes) != 1:
4214
      raise errors.ProgrammerError("Wrong template configuration")
4215
    remote_node = secondary_nodes[0]
4216
    minors = lu.cfg.AllocateDRBDMinor(
4217
      [primary_node, remote_node] * len(disk_info), instance_name)
4218

    
4219
    names = []
4220
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4221
                                               for i in range(disk_count)]):
4222
      names.append(lv_prefix + "_data")
4223
      names.append(lv_prefix + "_meta")
4224
    for idx, disk in enumerate(disk_info):
4225
      disk_index = idx + base_index
4226
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4227
                                      disk["size"], names[idx*2:idx*2+2],
4228
                                      "disk/%d" % disk_index,
4229
                                      minors[idx*2], minors[idx*2+1])
4230
      disk_dev.mode = disk["mode"]
4231
      disks.append(disk_dev)
4232
  elif template_name == constants.DT_FILE:
4233
    if len(secondary_nodes) != 0:
4234
      raise errors.ProgrammerError("Wrong template configuration")
4235

    
4236
    for idx, disk in enumerate(disk_info):
4237
      disk_index = idx + base_index
4238
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4239
                              iv_name="disk/%d" % disk_index,
4240
                              logical_id=(file_driver,
4241
                                          "%s/disk%d" % (file_storage_dir,
4242
                                                         disk_index)),
4243
                              mode=disk["mode"])
4244
      disks.append(disk_dev)
4245
  else:
4246
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4247
  return disks
4248

    
4249

    
4250
def _GetInstanceInfoText(instance):
4251
  """Compute that text that should be added to the disk's metadata.
4252

4253
  """
4254
  return "originstname+%s" % instance.name
4255

    
4256

    
4257
def _CreateDisks(lu, instance):
4258
  """Create all disks for an instance.
4259

4260
  This abstracts away some work from AddInstance.
4261

4262
  @type lu: L{LogicalUnit}
4263
  @param lu: the logical unit on whose behalf we execute
4264
  @type instance: L{objects.Instance}
4265
  @param instance: the instance whose disks we should create
4266
  @rtype: boolean
4267
  @return: the success of the creation
4268

4269
  """
4270
  info = _GetInstanceInfoText(instance)
4271
  pnode = instance.primary_node
4272

    
4273
  if instance.disk_template == constants.DT_FILE:
4274
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4275
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4276

    
4277
    if result.failed or not result.data:
4278
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4279

    
4280
    if not result.data[0]:
4281
      raise errors.OpExecError("Failed to create directory '%s'" %
4282
                               file_storage_dir)
4283

    
4284
  # Note: this needs to be kept in sync with adding of disks in
4285
  # LUSetInstanceParams
4286
  for device in instance.disks:
4287
    logging.info("Creating volume %s for instance %s",
4288
                 device.iv_name, instance.name)
4289
    #HARDCODE
4290
    for node in instance.all_nodes:
4291
      f_create = node == pnode
4292
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4293

    
4294

    
4295
def _RemoveDisks(lu, instance):
4296
  """Remove all disks for an instance.
4297

4298
  This abstracts away some work from `AddInstance()` and
4299
  `RemoveInstance()`. Note that in case some of the devices couldn't
4300
  be removed, the removal will continue with the other ones (compare
4301
  with `_CreateDisks()`).
4302

4303
  @type lu: L{LogicalUnit}
4304
  @param lu: the logical unit on whose behalf we execute
4305
  @type instance: L{objects.Instance}
4306
  @param instance: the instance whose disks we should remove
4307
  @rtype: boolean
4308
  @return: the success of the removal
4309

4310
  """
4311
  logging.info("Removing block devices for instance %s", instance.name)
4312

    
4313
  all_result = True
4314
  for device in instance.disks:
4315
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4316
      lu.cfg.SetDiskID(disk, node)
4317
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4318
      if msg:
4319
        lu.LogWarning("Could not remove block device %s on node %s,"
4320
                      " continuing anyway: %s", device.iv_name, node, msg)
4321
        all_result = False
4322

    
4323
  if instance.disk_template == constants.DT_FILE:
4324
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4325
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4326
                                                 file_storage_dir)
4327
    if result.failed or not result.data:
4328
      logging.error("Could not remove directory '%s'", file_storage_dir)
4329
      all_result = False
4330

    
4331
  return all_result
4332

    
4333

    
4334
def _ComputeDiskSize(disk_template, disks):
4335
  """Compute disk size requirements in the volume group
4336

4337
  """
4338
  # Required free disk space as a function of disk and swap space
4339
  req_size_dict = {
4340
    constants.DT_DISKLESS: None,
4341
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4342
    # 128 MB are added for drbd metadata for each disk
4343
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4344
    constants.DT_FILE: None,
4345
  }
4346

    
4347
  if disk_template not in req_size_dict:
4348
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4349
                                 " is unknown" %  disk_template)
4350

    
4351
  return req_size_dict[disk_template]
4352

    
4353

    
4354
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4355
  """Hypervisor parameter validation.
4356

4357
  This function abstract the hypervisor parameter validation to be
4358
  used in both instance create and instance modify.
4359

4360
  @type lu: L{LogicalUnit}
4361
  @param lu: the logical unit for which we check
4362
  @type nodenames: list
4363
  @param nodenames: the list of nodes on which we should check
4364
  @type hvname: string
4365
  @param hvname: the name of the hypervisor we should use
4366
  @type hvparams: dict
4367
  @param hvparams: the parameters which we need to check
4368
  @raise errors.OpPrereqError: if the parameters are not valid
4369

4370
  """
4371
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4372
                                                  hvname,
4373
                                                  hvparams)
4374
  for node in nodenames:
4375
    info = hvinfo[node]
4376
    if info.offline:
4377
      continue
4378
    msg = info.RemoteFailMsg()
4379
    if msg:
4380
      raise errors.OpPrereqError("Hypervisor parameter validation"
4381
                                 " failed on node %s: %s" % (node, msg))
4382

    
4383

    
4384
class LUCreateInstance(LogicalUnit):
4385
  """Create an instance.
4386

4387
  """
4388
  HPATH = "instance-add"
4389
  HTYPE = constants.HTYPE_INSTANCE
4390
  _OP_REQP = ["instance_name", "disks", "disk_template",
4391
              "mode", "start",
4392
              "wait_for_sync", "ip_check", "nics",
4393
              "hvparams", "beparams"]
4394
  REQ_BGL = False
4395

    
4396
  def _ExpandNode(self, node):
4397
    """Expands and checks one node name.
4398

4399
    """
4400
    node_full = self.cfg.ExpandNodeName(node)
4401
    if node_full is None:
4402
      raise errors.OpPrereqError("Unknown node %s" % node)
4403
    return node_full
4404

    
4405
  def ExpandNames(self):
4406
    """ExpandNames for CreateInstance.
4407

4408
    Figure out the right locks for instance creation.
4409

4410
    """
4411
    self.needed_locks = {}
4412

    
4413
    # set optional parameters to none if they don't exist
4414
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4415
      if not hasattr(self.op, attr):
4416
        setattr(self.op, attr, None)
4417

    
4418
    # cheap checks, mostly valid constants given
4419

    
4420
    # verify creation mode
4421
    if self.op.mode not in (constants.INSTANCE_CREATE,
4422
                            constants.INSTANCE_IMPORT):
4423
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4424
                                 self.op.mode)
4425

    
4426
    # disk template and mirror node verification
4427
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4428
      raise errors.OpPrereqError("Invalid disk template name")
4429

    
4430
    if self.op.hypervisor is None:
4431
      self.op.hypervisor = self.cfg.GetHypervisorType()
4432

    
4433
    cluster = self.cfg.GetClusterInfo()
4434
    enabled_hvs = cluster.enabled_hypervisors
4435
    if self.op.hypervisor not in enabled_hvs:
4436
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4437
                                 " cluster (%s)" % (self.op.hypervisor,
4438
                                  ",".join(enabled_hvs)))
4439

    
4440
    # check hypervisor parameter syntax (locally)
4441
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4442
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4443
                                  self.op.hvparams)
4444
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4445
    hv_type.CheckParameterSyntax(filled_hvp)
4446
    self.hv_full = filled_hvp
4447

    
4448
    # fill and remember the beparams dict
4449
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4450
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4451
                                    self.op.beparams)
4452

    
4453
    #### instance parameters check
4454

    
4455
    # instance name verification
4456
    hostname1 = utils.HostInfo(self.op.instance_name)
4457
    self.op.instance_name = instance_name = hostname1.name
4458

    
4459
    # this is just a preventive check, but someone might still add this
4460
    # instance in the meantime, and creation will fail at lock-add time
4461
    if instance_name in self.cfg.GetInstanceList():
4462
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4463
                                 instance_name)
4464

    
4465
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4466

    
4467
    # NIC buildup
4468
    self.nics = []
4469
    for nic in self.op.nics:
4470
      # ip validity checks
4471
      ip = nic.get("ip", None)
4472
      if ip is None or ip.lower() == "none":
4473
        nic_ip = None
4474
      elif ip.lower() == constants.VALUE_AUTO:
4475
        nic_ip = hostname1.ip
4476
      else:
4477
        if not utils.IsValidIP(ip):
4478
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4479
                                     " like a valid IP" % ip)
4480
        nic_ip = ip
4481

    
4482
      # MAC address verification
4483
      mac = nic.get("mac", constants.VALUE_AUTO)
4484
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4485
        if not utils.IsValidMac(mac.lower()):
4486
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4487
                                     mac)
4488
      # bridge verification
4489
      bridge = nic.get("bridge", None)
4490
      if bridge is None:
4491
        bridge = self.cfg.GetDefBridge()
4492
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4493

    
4494
    # disk checks/pre-build
4495
    self.disks = []
4496
    for disk in self.op.disks:
4497
      mode = disk.get("mode", constants.DISK_RDWR)
4498
      if mode not in constants.DISK_ACCESS_SET:
4499
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4500
                                   mode)
4501
      size = disk.get("size", None)
4502
      if size is None:
4503
        raise errors.OpPrereqError("Missing disk size")
4504
      try:
4505
        size = int(size)
4506
      except ValueError:
4507
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4508
      self.disks.append({"size": size, "mode": mode})
4509

    
4510
    # used in CheckPrereq for ip ping check
4511
    self.check_ip = hostname1.ip
4512

    
4513
    # file storage checks
4514
    if (self.op.file_driver and
4515
        not self.op.file_driver in constants.FILE_DRIVER):
4516
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4517
                                 self.op.file_driver)
4518

    
4519
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4520
      raise errors.OpPrereqError("File storage directory path not absolute")
4521

    
4522
    ### Node/iallocator related checks
4523
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4524
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4525
                                 " node must be given")
4526

    
4527
    if self.op.iallocator:
4528
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4529
    else:
4530
      self.op.pnode = self._ExpandNode(self.op.pnode)
4531
      nodelist = [self.op.pnode]
4532
      if self.op.snode is not None:
4533
        self.op.snode = self._ExpandNode(self.op.snode)
4534
        nodelist.append(self.op.snode)
4535
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4536

    
4537
    # in case of import lock the source node too
4538
    if self.op.mode == constants.INSTANCE_IMPORT:
4539
      src_node = getattr(self.op, "src_node", None)
4540
      src_path = getattr(self.op, "src_path", None)
4541

    
4542
      if src_path is None:
4543
        self.op.src_path = src_path = self.op.instance_name
4544

    
4545
      if src_node is None:
4546
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4547
        self.op.src_node = None
4548
        if os.path.isabs(src_path):
4549
          raise errors.OpPrereqError("Importing an instance from an absolute"
4550
                                     " path requires a source node option.")
4551
      else:
4552
        self.op.src_node = src_node = self._ExpandNode(src_node)
4553
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4554
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4555
        if not os.path.isabs(src_path):
4556
          self.op.src_path = src_path = \
4557
            os.path.join(constants.EXPORT_DIR, src_path)
4558

    
4559
    else: # INSTANCE_CREATE
4560
      if getattr(self.op, "os_type", None) is None:
4561
        raise errors.OpPrereqError("No guest OS specified")
4562

    
4563
  def _RunAllocator(self):
4564
    """Run the allocator based on input opcode.
4565

4566
    """
4567
    nics = [n.ToDict() for n in self.nics]
4568
    ial = IAllocator(self,
4569
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4570
                     name=self.op.instance_name,
4571
                     disk_template=self.op.disk_template,
4572
                     tags=[],
4573
                     os=self.op.os_type,
4574
                     vcpus=self.be_full[constants.BE_VCPUS],
4575
                     mem_size=self.be_full[constants.BE_MEMORY],
4576
                     disks=self.disks,
4577
                     nics=nics,
4578
                     hypervisor=self.op.hypervisor,
4579
                     )
4580

    
4581
    ial.Run(self.op.iallocator)
4582

    
4583
    if not ial.success:
4584
      raise errors.OpPrereqError("Can't compute nodes using"
4585
                                 " iallocator '%s': %s" % (self.op.iallocator,
4586
                                                           ial.info))
4587
    if len(ial.nodes) != ial.required_nodes:
4588
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4589
                                 " of nodes (%s), required %s" %
4590
                                 (self.op.iallocator, len(ial.nodes),
4591
                                  ial.required_nodes))
4592
    self.op.pnode = ial.nodes[0]
4593
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4594
                 self.op.instance_name, self.op.iallocator,
4595
                 ", ".join(ial.nodes))
4596
    if ial.required_nodes == 2:
4597
      self.op.snode = ial.nodes[1]
4598

    
4599
  def BuildHooksEnv(self):
4600
    """Build hooks env.
4601

4602
    This runs on master, primary and secondary nodes of the instance.
4603

4604
    """
4605
    env = {
4606
      "ADD_MODE": self.op.mode,
4607
      }
4608
    if self.op.mode == constants.INSTANCE_IMPORT:
4609
      env["SRC_NODE"] = self.op.src_node
4610
      env["SRC_PATH"] = self.op.src_path
4611
      env["SRC_IMAGES"] = self.src_images
4612

    
4613
    env.update(_BuildInstanceHookEnv(
4614
      name=self.op.instance_name,
4615
      primary_node=self.op.pnode,
4616
      secondary_nodes=self.secondaries,
4617
      status=self.op.start,
4618
      os_type=self.op.os_type,
4619
      memory=self.be_full[constants.BE_MEMORY],
4620
      vcpus=self.be_full[constants.BE_VCPUS],
4621
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4622
      disk_template=self.op.disk_template,
4623
      disks=[(d["size"], d["mode"]) for d in self.disks],
4624
      bep=self.be_full,
4625
      hvp=self.hv_full,
4626
      hypervisor=self.op.hypervisor,
4627
    ))
4628

    
4629
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4630
          self.secondaries)
4631
    return env, nl, nl
4632

    
4633

    
4634
  def CheckPrereq(self):
4635
    """Check prerequisites.
4636

4637
    """
4638
    if (not self.cfg.GetVGName() and
4639
        self.op.disk_template not in constants.DTS_NOT_LVM):
4640
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4641
                                 " instances")
4642

    
4643
    if self.op.mode == constants.INSTANCE_IMPORT:
4644
      src_node = self.op.src_node
4645
      src_path = self.op.src_path
4646

    
4647
      if src_node is None:
4648
        exp_list = self.rpc.call_export_list(
4649
          self.acquired_locks[locking.LEVEL_NODE])
4650
        found = False
4651
        for node in exp_list:
4652
          if not exp_list[node].failed and src_path in exp_list[node].data:
4653
            found = True
4654
            self.op.src_node = src_node = node
4655
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4656
                                                       src_path)
4657
            break
4658
        if not found:
4659
          raise errors.OpPrereqError("No export found for relative path %s" %
4660
                                      src_path)
4661

    
4662
      _CheckNodeOnline(self, src_node)
4663
      result = self.rpc.call_export_info(src_node, src_path)
4664
      result.Raise()
4665
      if not result.data:
4666
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4667

    
4668
      export_info = result.data
4669
      if not export_info.has_section(constants.INISECT_EXP):
4670
        raise errors.ProgrammerError("Corrupted export config")
4671

    
4672
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4673
      if (int(ei_version) != constants.EXPORT_VERSION):
4674
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4675
                                   (ei_version, constants.EXPORT_VERSION))
4676

    
4677
      # Check that the new instance doesn't have less disks than the export
4678
      instance_disks = len(self.disks)
4679
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4680
      if instance_disks < export_disks:
4681
        raise errors.OpPrereqError("Not enough disks to import."
4682
                                   " (instance: %d, export: %d)" %
4683
                                   (instance_disks, export_disks))
4684

    
4685
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4686
      disk_images = []
4687
      for idx in range(export_disks):
4688
        option = 'disk%d_dump' % idx
4689
        if export_info.has_option(constants.INISECT_INS, option):
4690
          # FIXME: are the old os-es, disk sizes, etc. useful?
4691
          export_name = export_info.get(constants.INISECT_INS, option)
4692
          image = os.path.join(src_path, export_name)
4693
          disk_images.append(image)
4694
        else:
4695
          disk_images.append(False)
4696

    
4697
      self.src_images = disk_images
4698

    
4699
      old_name = export_info.get(constants.INISECT_INS, 'name')
4700
      # FIXME: int() here could throw a ValueError on broken exports
4701
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4702
      if self.op.instance_name == old_name:
4703
        for idx, nic in enumerate(self.nics):
4704
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4705
            nic_mac_ini = 'nic%d_mac' % idx
4706
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4707

    
4708
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4709
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4710
    if self.op.start and not self.op.ip_check:
4711
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4712
                                 " adding an instance in start mode")
4713

    
4714
    if self.op.ip_check:
4715
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4716
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4717
                                   (self.check_ip, self.op.instance_name))
4718

    
4719
    #### mac address generation
4720
    # By generating here the mac address both the allocator and the hooks get
4721
    # the real final mac address rather than the 'auto' or 'generate' value.
4722
    # There is a race condition between the generation and the instance object
4723
    # creation, which means that we know the mac is valid now, but we're not
4724
    # sure it will be when we actually add the instance. If things go bad
4725
    # adding the instance will abort because of a duplicate mac, and the
4726
    # creation job will fail.
4727
    for nic in self.nics:
4728
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4729
        nic.mac = self.cfg.GenerateMAC()
4730

    
4731
    #### allocator run
4732

    
4733
    if self.op.iallocator is not None:
4734
      self._RunAllocator()
4735

    
4736
    #### node related checks
4737

    
4738
    # check primary node
4739
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4740
    assert self.pnode is not None, \
4741
      "Cannot retrieve locked node %s" % self.op.pnode
4742
    if pnode.offline:
4743
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4744
                                 pnode.name)
4745
    if pnode.drained:
4746
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4747
                                 pnode.name)
4748

    
4749
    self.secondaries = []
4750

    
4751
    # mirror node verification
4752
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4753
      if self.op.snode is None:
4754
        raise errors.OpPrereqError("The networked disk templates need"
4755
                                   " a mirror node")
4756
      if self.op.snode == pnode.name:
4757
        raise errors.OpPrereqError("The secondary node cannot be"
4758
                                   " the primary node.")
4759
      _CheckNodeOnline(self, self.op.snode)
4760
      _CheckNodeNotDrained(self, self.op.snode)
4761
      self.secondaries.append(self.op.snode)
4762

    
4763
    nodenames = [pnode.name] + self.secondaries
4764

    
4765
    req_size = _ComputeDiskSize(self.op.disk_template,
4766
                                self.disks)
4767

    
4768
    # Check lv size requirements
4769
    if req_size is not None:
4770
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4771
                                         self.op.hypervisor)
4772
      for node in nodenames:
4773
        info = nodeinfo[node]
4774
        info.Raise()
4775
        info = info.data
4776
        if not info:
4777
          raise errors.OpPrereqError("Cannot get current information"
4778
                                     " from node '%s'" % node)
4779
        vg_free = info.get('vg_free', None)
4780
        if not isinstance(vg_free, int):
4781
          raise errors.OpPrereqError("Can't compute free disk space on"
4782
                                     " node %s" % node)
4783
        if req_size > info['vg_free']:
4784
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4785
                                     " %d MB available, %d MB required" %
4786
                                     (node, info['vg_free'], req_size))
4787

    
4788
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4789

    
4790
    # os verification
4791
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4792
    result.Raise()
4793
    if not isinstance(result.data, objects.OS) or not result.data:
4794
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4795
                                 " primary node"  % self.op.os_type)
4796

    
4797
    # bridge check on primary node
4798
    bridges = [n.bridge for n in self.nics]
4799
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4800
    result.Raise()
4801
    if not result.data:
4802
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4803
                                 " exist on destination node '%s'" %
4804
                                 (",".join(bridges), pnode.name))
4805

    
4806
    # memory check on primary node
4807
    if self.op.start:
4808
      _CheckNodeFreeMemory(self, self.pnode.name,
4809
                           "creating instance %s" % self.op.instance_name,
4810
                           self.be_full[constants.BE_MEMORY],
4811
                           self.op.hypervisor)
4812

    
4813
  def Exec(self, feedback_fn):
4814
    """Create and add the instance to the cluster.
4815

4816
    """
4817
    instance = self.op.instance_name
4818
    pnode_name = self.pnode.name
4819

    
4820
    ht_kind = self.op.hypervisor
4821
    if ht_kind in constants.HTS_REQ_PORT:
4822
      network_port = self.cfg.AllocatePort()
4823
    else:
4824
      network_port = None
4825

    
4826
    ##if self.op.vnc_bind_address is None:
4827
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4828

    
4829
    # this is needed because os.path.join does not accept None arguments
4830
    if self.op.file_storage_dir is None:
4831
      string_file_storage_dir = ""
4832
    else:
4833
      string_file_storage_dir = self.op.file_storage_dir
4834

    
4835
    # build the full file storage dir path
4836
    file_storage_dir = os.path.normpath(os.path.join(
4837
                                        self.cfg.GetFileStorageDir(),
4838
                                        string_file_storage_dir, instance))
4839

    
4840

    
4841
    disks = _GenerateDiskTemplate(self,
4842
                                  self.op.disk_template,
4843
                                  instance, pnode_name,
4844
                                  self.secondaries,
4845
                                  self.disks,
4846
                                  file_storage_dir,
4847
                                  self.op.file_driver,
4848
                                  0)
4849

    
4850
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4851
                            primary_node=pnode_name,
4852
                            nics=self.nics, disks=disks,
4853
                            disk_template=self.op.disk_template,
4854
                            admin_up=False,
4855
                            network_port=network_port,
4856
                            beparams=self.op.beparams,
4857
                            hvparams=self.op.hvparams,
4858
                            hypervisor=self.op.hypervisor,
4859
                            )
4860

    
4861
    feedback_fn("* creating instance disks...")
4862
    try:
4863
      _CreateDisks(self, iobj)
4864
    except errors.OpExecError:
4865
      self.LogWarning("Device creation failed, reverting...")
4866
      try:
4867
        _RemoveDisks(self, iobj)
4868
      finally:
4869
        self.cfg.ReleaseDRBDMinors(instance)
4870
        raise
4871

    
4872
    feedback_fn("adding instance %s to cluster config" % instance)
4873

    
4874
    self.cfg.AddInstance(iobj)
4875
    # Declare that we don't want to remove the instance lock anymore, as we've
4876
    # added the instance to the config
4877
    del self.remove_locks[locking.LEVEL_INSTANCE]
4878
    # Unlock all the nodes
4879
    if self.op.mode == constants.INSTANCE_IMPORT:
4880
      nodes_keep = [self.op.src_node]
4881
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4882
                       if node != self.op.src_node]
4883
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4884
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4885
    else:
4886
      self.context.glm.release(locking.LEVEL_NODE)
4887
      del self.acquired_locks[locking.LEVEL_NODE]
4888

    
4889
    if self.op.wait_for_sync:
4890
      disk_abort = not _WaitForSync(self, iobj)
4891
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4892
      # make sure the disks are not degraded (still sync-ing is ok)
4893
      time.sleep(15)
4894
      feedback_fn("* checking mirrors status")
4895
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4896
    else:
4897
      disk_abort = False
4898

    
4899
    if disk_abort:
4900
      _RemoveDisks(self, iobj)
4901
      self.cfg.RemoveInstance(iobj.name)
4902
      # Make sure the instance lock gets removed
4903
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4904
      raise errors.OpExecError("There are some degraded disks for"
4905
                               " this instance")
4906

    
4907
    feedback_fn("creating os for instance %s on node %s" %
4908
                (instance, pnode_name))
4909

    
4910
    if iobj.disk_template != constants.DT_DISKLESS:
4911
      if self.op.mode == constants.INSTANCE_CREATE:
4912
        feedback_fn("* running the instance OS create scripts...")
4913
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4914
        msg = result.RemoteFailMsg()
4915
        if msg:
4916
          raise errors.OpExecError("Could not add os for instance %s"
4917
                                   " on node %s: %s" %
4918
                                   (instance, pnode_name, msg))
4919

    
4920
      elif self.op.mode == constants.INSTANCE_IMPORT:
4921
        feedback_fn("* running the instance OS import scripts...")
4922
        src_node = self.op.src_node
4923
        src_images = self.src_images
4924
        cluster_name = self.cfg.GetClusterName()
4925
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4926
                                                         src_node, src_images,
4927
                                                         cluster_name)
4928
        import_result.Raise()
4929
        for idx, result in enumerate(import_result.data):
4930
          if not result:
4931
            self.LogWarning("Could not import the image %s for instance"
4932
                            " %s, disk %d, on node %s" %
4933
                            (src_images[idx], instance, idx, pnode_name))
4934
      else:
4935
        # also checked in the prereq part
4936
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4937
                                     % self.op.mode)
4938

    
4939
    if self.op.start:
4940
      iobj.admin_up = True
4941
      self.cfg.Update(iobj)
4942
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4943
      feedback_fn("* starting instance...")
4944
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
4945
      msg = result.RemoteFailMsg()
4946
      if msg:
4947
        raise errors.OpExecError("Could not start instance: %s" % msg)
4948

    
4949

    
4950
class LUConnectConsole(NoHooksLU):
4951
  """Connect to an instance's console.
4952

4953
  This is somewhat special in that it returns the command line that
4954
  you need to run on the master node in order to connect to the
4955
  console.
4956

4957
  """
4958
  _OP_REQP = ["instance_name"]
4959
  REQ_BGL = False
4960

    
4961
  def ExpandNames(self):
4962
    self._ExpandAndLockInstance()
4963

    
4964
  def CheckPrereq(self):
4965
    """Check prerequisites.
4966

4967
    This checks that the instance is in the cluster.
4968

4969
    """
4970
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4971
    assert self.instance is not None, \
4972
      "Cannot retrieve locked instance %s" % self.op.instance_name
4973
    _CheckNodeOnline(self, self.instance.primary_node)
4974

    
4975
  def Exec(self, feedback_fn):
4976
    """Connect to the console of an instance
4977

4978
    """
4979
    instance = self.instance
4980
    node = instance.primary_node
4981

    
4982
    node_insts = self.rpc.call_instance_list([node],
4983
                                             [instance.hypervisor])[node]
4984
    node_insts.Raise()
4985

    
4986
    if instance.name not in node_insts.data:
4987
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4988

    
4989
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4990

    
4991
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4992
    cluster = self.cfg.GetClusterInfo()
4993
    # beparams and hvparams are passed separately, to avoid editing the
4994
    # instance and then saving the defaults in the instance itself.
4995
    hvparams = cluster.FillHV(instance)
4996
    beparams = cluster.FillBE(instance)
4997
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4998

    
4999
    # build ssh cmdline
5000
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
5001

    
5002

    
5003
class LUReplaceDisks(LogicalUnit):
5004
  """Replace the disks of an instance.
5005

5006
  """
5007
  HPATH = "mirrors-replace"
5008
  HTYPE = constants.HTYPE_INSTANCE
5009
  _OP_REQP = ["instance_name", "mode", "disks"]
5010
  REQ_BGL = False
5011

    
5012
  def CheckArguments(self):
5013
    if not hasattr(self.op, "remote_node"):
5014
      self.op.remote_node = None
5015
    if not hasattr(self.op, "iallocator"):
5016
      self.op.iallocator = None
5017

    
5018
    # check for valid parameter combination
5019
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
5020
    if self.op.mode == constants.REPLACE_DISK_CHG:
5021
      if cnt == 2:
5022
        raise errors.OpPrereqError("When changing the secondary either an"
5023
                                   " iallocator script must be used or the"
5024
                                   " new node given")
5025
      elif cnt == 0:
5026
        raise errors.OpPrereqError("Give either the iallocator or the new"
5027
                                   " secondary, not both")
5028
    else: # not replacing the secondary
5029
      if cnt != 2:
5030
        raise errors.OpPrereqError("The iallocator and new node options can"
5031
                                   " be used only when changing the"
5032
                                   " secondary node")
5033

    
5034
  def ExpandNames(self):
5035
    self._ExpandAndLockInstance()
5036

    
5037
    if self.op.iallocator is not None:
5038
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5039
    elif self.op.remote_node is not None:
5040
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
5041
      if remote_node is None:
5042
        raise errors.OpPrereqError("Node '%s' not known" %
5043
                                   self.op.remote_node)
5044
      self.op.remote_node = remote_node
5045
      # Warning: do not remove the locking of the new secondary here
5046
      # unless DRBD8.AddChildren is changed to work in parallel;
5047
      # currently it doesn't since parallel invocations of
5048
      # FindUnusedMinor will conflict
5049
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
5050
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5051
    else:
5052
      self.needed_locks[locking.LEVEL_NODE] = []
5053
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5054

    
5055
  def DeclareLocks(self, level):
5056
    # If we're not already locking all nodes in the set we have to declare the
5057
    # instance's primary/secondary nodes.
5058
    if (level == locking.LEVEL_NODE and
5059
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
5060
      self._LockInstancesNodes()
5061

    
5062
  def _RunAllocator(self):
5063
    """Compute a new secondary node using an IAllocator.
5064

5065
    """
5066
    ial = IAllocator(self,
5067
                     mode=constants.IALLOCATOR_MODE_RELOC,
5068
                     name=self.op.instance_name,
5069
                     relocate_from=[self.sec_node])
5070

    
5071
    ial.Run(self.op.iallocator)
5072

    
5073
    if not ial.success:
5074
      raise errors.OpPrereqError("Can't compute nodes using"
5075
                                 " iallocator '%s': %s" % (self.op.iallocator,
5076
                                                           ial.info))
5077
    if len(ial.nodes) != ial.required_nodes:
5078
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
5079
                                 " of nodes (%s), required %s" %
5080
                                 (len(ial.nodes), ial.required_nodes))
5081
    self.op.remote_node = ial.nodes[0]
5082
    self.LogInfo("Selected new secondary for the instance: %s",
5083
                 self.op.remote_node)
5084

    
5085
  def BuildHooksEnv(self):
5086
    """Build hooks env.
5087

5088
    This runs on the master, the primary and all the secondaries.
5089

5090
    """
5091
    env = {
5092
      "MODE": self.op.mode,
5093
      "NEW_SECONDARY": self.op.remote_node,
5094
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
5095
      }
5096
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5097
    nl = [
5098
      self.cfg.GetMasterNode(),
5099
      self.instance.primary_node,
5100
      ]
5101
    if self.op.remote_node is not None:
5102
      nl.append(self.op.remote_node)
5103
    return env, nl, nl
5104

    
5105
  def CheckPrereq(self):
5106
    """Check prerequisites.
5107

5108
    This checks that the instance is in the cluster.
5109

5110
    """
5111
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5112
    assert instance is not None, \
5113
      "Cannot retrieve locked instance %s" % self.op.instance_name
5114
    self.instance = instance
5115

    
5116
    if instance.disk_template != constants.DT_DRBD8:
5117
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
5118
                                 " instances")
5119

    
5120
    if len(instance.secondary_nodes) != 1:
5121
      raise errors.OpPrereqError("The instance has a strange layout,"
5122
                                 " expected one secondary but found %d" %
5123
                                 len(instance.secondary_nodes))
5124

    
5125
    self.sec_node = instance.secondary_nodes[0]
5126

    
5127
    if self.op.iallocator is not None:
5128
      self._RunAllocator()
5129

    
5130
    remote_node = self.op.remote_node
5131
    if remote_node is not None:
5132
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
5133
      assert self.remote_node_info is not None, \
5134
        "Cannot retrieve locked node %s" % remote_node
5135
    else:
5136
      self.remote_node_info = None
5137
    if remote_node == instance.primary_node:
5138
      raise errors.OpPrereqError("The specified node is the primary node of"
5139
                                 " the instance.")
5140
    elif remote_node == self.sec_node:
5141
      raise errors.OpPrereqError("The specified node is already the"
5142
                                 " secondary node of the instance.")
5143

    
5144
    if self.op.mode == constants.REPLACE_DISK_PRI:
5145
      n1 = self.tgt_node = instance.primary_node
5146
      n2 = self.oth_node = self.sec_node
5147
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5148
      n1 = self.tgt_node = self.sec_node
5149
      n2 = self.oth_node = instance.primary_node
5150
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5151
      n1 = self.new_node = remote_node
5152
      n2 = self.oth_node = instance.primary_node
5153
      self.tgt_node = self.sec_node
5154
      _CheckNodeNotDrained(self, remote_node)
5155
    else:
5156
      raise errors.ProgrammerError("Unhandled disk replace mode")
5157

    
5158
    _CheckNodeOnline(self, n1)
5159
    _CheckNodeOnline(self, n2)
5160

    
5161
    if not self.op.disks:
5162
      self.op.disks = range(len(instance.disks))
5163

    
5164
    for disk_idx in self.op.disks:
5165
      instance.FindDisk(disk_idx)
5166

    
5167
  def _ExecD8DiskOnly(self, feedback_fn):
5168
    """Replace a disk on the primary or secondary for dbrd8.
5169

5170
    The algorithm for replace is quite complicated:
5171

5172
      1. for each disk to be replaced:
5173

5174
        1. create new LVs on the target node with unique names
5175
        1. detach old LVs from the drbd device
5176
        1. rename old LVs to name_replaced.<time_t>
5177
        1. rename new LVs to old LVs
5178
        1. attach the new LVs (with the old names now) to the drbd device
5179

5180
      1. wait for sync across all devices
5181

5182
      1. for each modified disk:
5183

5184
        1. remove old LVs (which have the name name_replaces.<time_t>)
5185

5186
    Failures are not very well handled.
5187

5188
    """
5189
    steps_total = 6
5190
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5191
    instance = self.instance
5192
    iv_names = {}
5193
    vgname = self.cfg.GetVGName()
5194
    # start of work
5195
    cfg = self.cfg
5196
    tgt_node = self.tgt_node
5197
    oth_node = self.oth_node
5198

    
5199
    # Step: check device activation
5200
    self.proc.LogStep(1, steps_total, "check device existence")
5201
    info("checking volume groups")
5202
    my_vg = cfg.GetVGName()
5203
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5204
    if not results:
5205
      raise errors.OpExecError("Can't list volume groups on the nodes")
5206
    for node in oth_node, tgt_node:
5207
      res = results[node]
5208
      if res.failed or not res.data or my_vg not in res.data:
5209
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5210
                                 (my_vg, node))
5211
    for idx, dev in enumerate(instance.disks):
5212
      if idx not in self.op.disks:
5213
        continue
5214
      for node in tgt_node, oth_node:
5215
        info("checking disk/%d on %s" % (idx, node))
5216
        cfg.SetDiskID(dev, node)
5217
        result = self.rpc.call_blockdev_find(node, dev)
5218
        msg = result.RemoteFailMsg()
5219
        if not msg and not result.payload:
5220
          msg = "disk not found"
5221
        if msg:
5222
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5223
                                   (idx, node, msg))
5224

    
5225
    # Step: check other node consistency
5226
    self.proc.LogStep(2, steps_total, "check peer consistency")
5227
    for idx, dev in enumerate(instance.disks):
5228
      if idx not in self.op.disks:
5229
        continue
5230
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5231
      if not _CheckDiskConsistency(self, dev, oth_node,
5232
                                   oth_node==instance.primary_node):
5233
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5234
                                 " to replace disks on this node (%s)" %
5235
                                 (oth_node, tgt_node))
5236

    
5237
    # Step: create new storage
5238
    self.proc.LogStep(3, steps_total, "allocate new storage")
5239
    for idx, dev in enumerate(instance.disks):
5240
      if idx not in self.op.disks:
5241
        continue
5242
      size = dev.size
5243
      cfg.SetDiskID(dev, tgt_node)
5244
      lv_names = [".disk%d_%s" % (idx, suf)
5245
                  for suf in ["data", "meta"]]
5246
      names = _GenerateUniqueNames(self, lv_names)
5247
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5248
                             logical_id=(vgname, names[0]))
5249
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5250
                             logical_id=(vgname, names[1]))
5251
      new_lvs = [lv_data, lv_meta]
5252
      old_lvs = dev.children
5253
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5254
      info("creating new local storage on %s for %s" %
5255
           (tgt_node, dev.iv_name))
5256
      # we pass force_create=True to force the LVM creation
5257
      for new_lv in new_lvs:
5258
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5259
                        _GetInstanceInfoText(instance), False)
5260

    
5261
    # Step: for each lv, detach+rename*2+attach
5262
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5263
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5264
      info("detaching %s drbd from local storage" % dev.iv_name)
5265
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5266
      result.Raise()
5267
      if not result.data:
5268
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5269
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5270
      #dev.children = []
5271
      #cfg.Update(instance)
5272

    
5273
      # ok, we created the new LVs, so now we know we have the needed
5274
      # storage; as such, we proceed on the target node to rename
5275
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5276
      # using the assumption that logical_id == physical_id (which in
5277
      # turn is the unique_id on that node)
5278

    
5279
      # FIXME(iustin): use a better name for the replaced LVs
5280
      temp_suffix = int(time.time())
5281
      ren_fn = lambda d, suff: (d.physical_id[0],
5282
                                d.physical_id[1] + "_replaced-%s" % suff)
5283
      # build the rename list based on what LVs exist on the node
5284
      rlist = []
5285
      for to_ren in old_lvs:
5286
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5287
        if not result.RemoteFailMsg() and result.payload:
5288
          # device exists
5289
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5290

    
5291
      info("renaming the old LVs on the target node")
5292
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5293
      result.Raise()
5294
      if not result.data:
5295
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5296
      # now we rename the new LVs to the old LVs
5297
      info("renaming the new LVs on the target node")
5298
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5299
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5300
      result.Raise()
5301
      if not result.data:
5302
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5303

    
5304
      for old, new in zip(old_lvs, new_lvs):
5305
        new.logical_id = old.logical_id
5306
        cfg.SetDiskID(new, tgt_node)
5307

    
5308
      for disk in old_lvs:
5309
        disk.logical_id = ren_fn(disk, temp_suffix)
5310
        cfg.SetDiskID(disk, tgt_node)
5311

    
5312
      # now that the new lvs have the old name, we can add them to the device
5313
      info("adding new mirror component on %s" % tgt_node)
5314
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5315
      if result.failed or not result.data:
5316
        for new_lv in new_lvs:
5317
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5318
          if msg:
5319
            warning("Can't rollback device %s: %s", dev, msg,
5320
                    hint="cleanup manually the unused logical volumes")
5321
        raise errors.OpExecError("Can't add local storage to drbd")
5322

    
5323
      dev.children = new_lvs
5324
      cfg.Update(instance)
5325

    
5326
    # Step: wait for sync
5327

    
5328
    # this can fail as the old devices are degraded and _WaitForSync
5329
    # does a combined result over all disks, so we don't check its
5330
    # return value
5331
    self.proc.LogStep(5, steps_total, "sync devices")
5332
    _WaitForSync(self, instance, unlock=True)
5333

    
5334
    # so check manually all the devices
5335
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5336
      cfg.SetDiskID(dev, instance.primary_node)
5337
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5338
      msg = result.RemoteFailMsg()
5339
      if not msg and not result.payload:
5340
        msg = "disk not found"
5341
      if msg:
5342
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5343
                                 (name, msg))
5344
      if result.payload[5]:
5345
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5346

    
5347
    # Step: remove old storage
5348
    self.proc.LogStep(6, steps_total, "removing old storage")
5349
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5350
      info("remove logical volumes for %s" % name)
5351
      for lv in old_lvs:
5352
        cfg.SetDiskID(lv, tgt_node)
5353
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5354
        if msg:
5355
          warning("Can't remove old LV: %s" % msg,
5356
                  hint="manually remove unused LVs")
5357
          continue
5358

    
5359
  def _ExecD8Secondary(self, feedback_fn):
5360
    """Replace the secondary node for drbd8.
5361

5362
    The algorithm for replace is quite complicated:
5363
      - for all disks of the instance:
5364
        - create new LVs on the new node with same names
5365
        - shutdown the drbd device on the old secondary
5366
        - disconnect the drbd network on the primary
5367
        - create the drbd device on the new secondary
5368
        - network attach the drbd on the primary, using an artifice:
5369
          the drbd code for Attach() will connect to the network if it
5370
          finds a device which is connected to the good local disks but
5371
          not network enabled
5372
      - wait for sync across all devices
5373
      - remove all disks from the old secondary
5374

5375
    Failures are not very well handled.
5376

5377
    """
5378
    steps_total = 6
5379
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5380
    instance = self.instance
5381
    iv_names = {}
5382
    # start of work
5383
    cfg = self.cfg
5384
    old_node = self.tgt_node
5385
    new_node = self.new_node
5386
    pri_node = instance.primary_node
5387
    nodes_ip = {
5388
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5389
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5390
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5391
      }
5392

    
5393
    # Step: check device activation
5394
    self.proc.LogStep(1, steps_total, "check device existence")
5395
    info("checking volume groups")
5396
    my_vg = cfg.GetVGName()
5397
    results = self.rpc.call_vg_list([pri_node, new_node])
5398
    for node in pri_node, new_node:
5399
      res = results[node]
5400
      if res.failed or not res.data or my_vg not in res.data:
5401
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5402
                                 (my_vg, node))
5403
    for idx, dev in enumerate(instance.disks):
5404
      if idx not in self.op.disks:
5405
        continue
5406
      info("checking disk/%d on %s" % (idx, pri_node))
5407
      cfg.SetDiskID(dev, pri_node)
5408
      result = self.rpc.call_blockdev_find(pri_node, dev)
5409
      msg = result.RemoteFailMsg()
5410
      if not msg and not result.payload:
5411
        msg = "disk not found"
5412
      if msg:
5413
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5414
                                 (idx, pri_node, msg))
5415

    
5416
    # Step: check other node consistency
5417
    self.proc.LogStep(2, steps_total, "check peer consistency")
5418
    for idx, dev in enumerate(instance.disks):
5419
      if idx not in self.op.disks:
5420
        continue
5421
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5422
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5423
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5424
                                 " unsafe to replace the secondary" %
5425
                                 pri_node)
5426

    
5427
    # Step: create new storage
5428
    self.proc.LogStep(3, steps_total, "allocate new storage")
5429
    for idx, dev in enumerate(instance.disks):
5430
      info("adding new local storage on %s for disk/%d" %
5431
           (new_node, idx))
5432
      # we pass force_create=True to force LVM creation
5433
      for new_lv in dev.children:
5434
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5435
                        _GetInstanceInfoText(instance), False)
5436

    
5437
    # Step 4: dbrd minors and drbd setups changes
5438
    # after this, we must manually remove the drbd minors on both the
5439
    # error and the success paths
5440
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5441
                                   instance.name)
5442
    logging.debug("Allocated minors %s" % (minors,))
5443
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5444
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5445
      size = dev.size
5446
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5447
      # create new devices on new_node; note that we create two IDs:
5448
      # one without port, so the drbd will be activated without
5449
      # networking information on the new node at this stage, and one
5450
      # with network, for the latter activation in step 4
5451
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5452
      if pri_node == o_node1:
5453
        p_minor = o_minor1
5454
      else:
5455
        p_minor = o_minor2
5456

    
5457
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5458
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5459

    
5460
      iv_names[idx] = (dev, dev.children, new_net_id)
5461
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5462
                    new_net_id)
5463
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5464
                              logical_id=new_alone_id,
5465
                              children=dev.children,
5466
                              size=dev.size)
5467
      try:
5468
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5469
                              _GetInstanceInfoText(instance), False)
5470
      except errors.GenericError:
5471
        self.cfg.ReleaseDRBDMinors(instance.name)
5472
        raise
5473

    
5474
    for idx, dev in enumerate(instance.disks):
5475
      # we have new devices, shutdown the drbd on the old secondary
5476
      info("shutting down drbd for disk/%d on old node" % idx)
5477
      cfg.SetDiskID(dev, old_node)
5478
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5479
      if msg:
5480
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5481
                (idx, msg),
5482
                hint="Please cleanup this device manually as soon as possible")
5483

    
5484
    info("detaching primary drbds from the network (=> standalone)")
5485
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5486
                                               instance.disks)[pri_node]
5487

    
5488
    msg = result.RemoteFailMsg()
5489
    if msg:
5490
      # detaches didn't succeed (unlikely)
5491
      self.cfg.ReleaseDRBDMinors(instance.name)
5492
      raise errors.OpExecError("Can't detach the disks from the network on"
5493
                               " old node: %s" % (msg,))
5494

    
5495
    # if we managed to detach at least one, we update all the disks of
5496
    # the instance to point to the new secondary
5497
    info("updating instance configuration")
5498
    for dev, _, new_logical_id in iv_names.itervalues():
5499
      dev.logical_id = new_logical_id
5500
      cfg.SetDiskID(dev, pri_node)
5501
    cfg.Update(instance)
5502

    
5503
    # and now perform the drbd attach
5504
    info("attaching primary drbds to new secondary (standalone => connected)")
5505
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5506
                                           instance.disks, instance.name,
5507
                                           False)
5508
    for to_node, to_result in result.items():
5509
      msg = to_result.RemoteFailMsg()
5510
      if msg:
5511
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5512
                hint="please do a gnt-instance info to see the"
5513
                " status of disks")
5514

    
5515
    # this can fail as the old devices are degraded and _WaitForSync
5516
    # does a combined result over all disks, so we don't check its
5517
    # return value
5518
    self.proc.LogStep(5, steps_total, "sync devices")
5519
    _WaitForSync(self, instance, unlock=True)
5520

    
5521
    # so check manually all the devices
5522
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5523
      cfg.SetDiskID(dev, pri_node)
5524
      result = self.rpc.call_blockdev_find(pri_node, dev)
5525
      msg = result.RemoteFailMsg()
5526
      if not msg and not result.payload:
5527
        msg = "disk not found"
5528
      if msg:
5529
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5530
                                 (idx, msg))
5531
      if result.payload[5]:
5532
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5533

    
5534
    self.proc.LogStep(6, steps_total, "removing old storage")
5535
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5536
      info("remove logical volumes for disk/%d" % idx)
5537
      for lv in old_lvs:
5538
        cfg.SetDiskID(lv, old_node)
5539
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5540
        if msg:
5541
          warning("Can't remove LV on old secondary: %s", msg,
5542
                  hint="Cleanup stale volumes by hand")
5543

    
5544
  def Exec(self, feedback_fn):
5545
    """Execute disk replacement.
5546

5547
    This dispatches the disk replacement to the appropriate handler.
5548

5549
    """
5550
    instance = self.instance
5551

    
5552
    # Activate the instance disks if we're replacing them on a down instance
5553
    if not instance.admin_up:
5554
      _StartInstanceDisks(self, instance, True)
5555

    
5556
    if self.op.mode == constants.REPLACE_DISK_CHG:
5557
      fn = self._ExecD8Secondary
5558
    else:
5559
      fn = self._ExecD8DiskOnly
5560

    
5561
    ret = fn(feedback_fn)
5562

    
5563
    # Deactivate the instance disks if we're replacing them on a down instance
5564
    if not instance.admin_up:
5565
      _SafeShutdownInstanceDisks(self, instance)
5566

    
5567
    return ret
5568

    
5569

    
5570
class LUGrowDisk(LogicalUnit):
5571
  """Grow a disk of an instance.
5572

5573
  """
5574
  HPATH = "disk-grow"
5575
  HTYPE = constants.HTYPE_INSTANCE
5576
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5577
  REQ_BGL = False
5578

    
5579
  def ExpandNames(self):
5580
    self._ExpandAndLockInstance()
5581
    self.needed_locks[locking.LEVEL_NODE] = []
5582
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5583

    
5584
  def DeclareLocks(self, level):
5585
    if level == locking.LEVEL_NODE:
5586
      self._LockInstancesNodes()
5587

    
5588
  def BuildHooksEnv(self):
5589
    """Build hooks env.
5590

5591
    This runs on the master, the primary and all the secondaries.
5592

5593
    """
5594
    env = {
5595
      "DISK": self.op.disk,
5596
      "AMOUNT": self.op.amount,
5597
      }
5598
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5599
    nl = [
5600
      self.cfg.GetMasterNode(),
5601
      self.instance.primary_node,
5602
      ]
5603
    return env, nl, nl
5604

    
5605
  def CheckPrereq(self):
5606
    """Check prerequisites.
5607

5608
    This checks that the instance is in the cluster.
5609

5610
    """
5611
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5612
    assert instance is not None, \
5613
      "Cannot retrieve locked instance %s" % self.op.instance_name
5614
    nodenames = list(instance.all_nodes)
5615
    for node in nodenames:
5616
      _CheckNodeOnline(self, node)
5617

    
5618

    
5619
    self.instance = instance
5620

    
5621
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5622
      raise errors.OpPrereqError("Instance's disk layout does not support"
5623
                                 " growing.")
5624

    
5625
    self.disk = instance.FindDisk(self.op.disk)
5626

    
5627
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5628
                                       instance.hypervisor)
5629
    for node in nodenames:
5630
      info = nodeinfo[node]
5631
      if info.failed or not info.data:
5632
        raise errors.OpPrereqError("Cannot get current information"
5633
                                   " from node '%s'" % node)
5634
      vg_free = info.data.get('vg_free', None)
5635
      if not isinstance(vg_free, int):
5636
        raise errors.OpPrereqError("Can't compute free disk space on"
5637
                                   " node %s" % node)
5638
      if self.op.amount > vg_free:
5639
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5640
                                   " %d MiB available, %d MiB required" %
5641
                                   (node, vg_free, self.op.amount))
5642

    
5643
  def Exec(self, feedback_fn):
5644
    """Execute disk grow.
5645

5646
    """
5647
    instance = self.instance
5648
    disk = self.disk
5649
    for node in instance.all_nodes:
5650
      self.cfg.SetDiskID(disk, node)
5651
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5652
      msg = result.RemoteFailMsg()
5653
      if msg:
5654
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5655
                                 (node, msg))
5656
    disk.RecordGrow(self.op.amount)
5657
    self.cfg.Update(instance)
5658
    if self.op.wait_for_sync:
5659
      disk_abort = not _WaitForSync(self, instance)
5660
      if disk_abort:
5661
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5662
                             " status.\nPlease check the instance.")
5663

    
5664

    
5665
class LUQueryInstanceData(NoHooksLU):
5666
  """Query runtime instance data.
5667

5668
  """
5669
  _OP_REQP = ["instances", "static"]
5670
  REQ_BGL = False
5671

    
5672
  def ExpandNames(self):
5673
    self.needed_locks = {}
5674
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5675

    
5676
    if not isinstance(self.op.instances, list):
5677
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5678

    
5679
    if self.op.instances:
5680
      self.wanted_names = []
5681
      for name in self.op.instances:
5682
        full_name = self.cfg.ExpandInstanceName(name)
5683
        if full_name is None:
5684
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5685
        self.wanted_names.append(full_name)
5686
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5687
    else:
5688
      self.wanted_names = None
5689
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5690

    
5691
    self.needed_locks[locking.LEVEL_NODE] = []
5692
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5693

    
5694
  def DeclareLocks(self, level):
5695
    if level == locking.LEVEL_NODE:
5696
      self._LockInstancesNodes()
5697

    
5698
  def CheckPrereq(self):
5699
    """Check prerequisites.
5700

5701
    This only checks the optional instance list against the existing names.
5702

5703
    """
5704
    if self.wanted_names is None:
5705
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5706

    
5707
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5708
                             in self.wanted_names]
5709
    return
5710

    
5711
  def _ComputeDiskStatus(self, instance, snode, dev):
5712
    """Compute block device status.
5713

5714
    """
5715
    static = self.op.static
5716
    if not static:
5717
      self.cfg.SetDiskID(dev, instance.primary_node)
5718
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5719
      if dev_pstatus.offline:
5720
        dev_pstatus = None
5721
      else:
5722
        msg = dev_pstatus.RemoteFailMsg()
5723
        if msg:
5724
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5725
                                   (instance.name, msg))
5726
        dev_pstatus = dev_pstatus.payload
5727
    else:
5728
      dev_pstatus = None
5729

    
5730
    if dev.dev_type in constants.LDS_DRBD:
5731
      # we change the snode then (otherwise we use the one passed in)
5732
      if dev.logical_id[0] == instance.primary_node:
5733
        snode = dev.logical_id[1]
5734
      else:
5735
        snode = dev.logical_id[0]
5736

    
5737
    if snode and not static:
5738
      self.cfg.SetDiskID(dev, snode)
5739
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5740
      if dev_sstatus.offline:
5741
        dev_sstatus = None
5742
      else:
5743
        msg = dev_sstatus.RemoteFailMsg()
5744
        if msg:
5745
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5746
                                   (instance.name, msg))
5747
        dev_sstatus = dev_sstatus.payload
5748
    else:
5749
      dev_sstatus = None
5750

    
5751
    if dev.children:
5752
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5753
                      for child in dev.children]
5754
    else:
5755
      dev_children = []
5756

    
5757
    data = {
5758
      "iv_name": dev.iv_name,
5759
      "dev_type": dev.dev_type,
5760
      "logical_id": dev.logical_id,
5761
      "physical_id": dev.physical_id,
5762
      "pstatus": dev_pstatus,
5763
      "sstatus": dev_sstatus,
5764
      "children": dev_children,
5765
      "mode": dev.mode,
5766
      "size": dev.size,
5767
      }
5768

    
5769
    return data
5770

    
5771
  def Exec(self, feedback_fn):
5772
    """Gather and return data"""
5773
    result = {}
5774

    
5775
    cluster = self.cfg.GetClusterInfo()
5776

    
5777
    for instance in self.wanted_instances:
5778
      if not self.op.static:
5779
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5780
                                                  instance.name,
5781
                                                  instance.hypervisor)
5782
        remote_info.Raise()
5783
        remote_info = remote_info.data
5784
        if remote_info and "state" in remote_info:
5785
          remote_state = "up"
5786
        else:
5787
          remote_state = "down"
5788
      else:
5789
        remote_state = None
5790
      if instance.admin_up:
5791
        config_state = "up"
5792
      else:
5793
        config_state = "down"
5794

    
5795
      disks = [self._ComputeDiskStatus(instance, None, device)
5796
               for device in instance.disks]
5797

    
5798
      idict = {
5799
        "name": instance.name,
5800
        "config_state": config_state,
5801
        "run_state": remote_state,
5802
        "pnode": instance.primary_node,
5803
        "snodes": instance.secondary_nodes,
5804
        "os": instance.os,
5805
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5806
        "disks": disks,
5807
        "hypervisor": instance.hypervisor,
5808
        "network_port": instance.network_port,
5809
        "hv_instance": instance.hvparams,
5810
        "hv_actual": cluster.FillHV(instance),
5811
        "be_instance": instance.beparams,
5812
        "be_actual": cluster.FillBE(instance),
5813
        }
5814

    
5815
      result[instance.name] = idict
5816

    
5817
    return result
5818

    
5819

    
5820
class LUSetInstanceParams(LogicalUnit):
5821
  """Modifies an instances's parameters.
5822

5823
  """
5824
  HPATH = "instance-modify"
5825
  HTYPE = constants.HTYPE_INSTANCE
5826
  _OP_REQP = ["instance_name"]
5827
  REQ_BGL = False
5828

    
5829
  def CheckArguments(self):
5830
    if not hasattr(self.op, 'nics'):
5831
      self.op.nics = []
5832
    if not hasattr(self.op, 'disks'):
5833
      self.op.disks = []
5834
    if not hasattr(self.op, 'beparams'):
5835
      self.op.beparams = {}
5836
    if not hasattr(self.op, 'hvparams'):
5837
      self.op.hvparams = {}
5838
    self.op.force = getattr(self.op, "force", False)
5839
    if not (self.op.nics or self.op.disks or
5840
            self.op.hvparams or self.op.beparams):
5841
      raise errors.OpPrereqError("No changes submitted")
5842

    
5843
    # Disk validation
5844
    disk_addremove = 0
5845
    for disk_op, disk_dict in self.op.disks:
5846
      if disk_op == constants.DDM_REMOVE:
5847
        disk_addremove += 1
5848
        continue
5849
      elif disk_op == constants.DDM_ADD:
5850
        disk_addremove += 1
5851
      else:
5852
        if not isinstance(disk_op, int):
5853
          raise errors.OpPrereqError("Invalid disk index")
5854
      if disk_op == constants.DDM_ADD:
5855
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5856
        if mode not in constants.DISK_ACCESS_SET:
5857
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5858
        size = disk_dict.get('size', None)
5859
        if size is None:
5860
          raise errors.OpPrereqError("Required disk parameter size missing")
5861
        try:
5862
          size = int(size)
5863
        except ValueError, err:
5864
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5865
                                     str(err))
5866
        disk_dict['size'] = size
5867
      else:
5868
        # modification of disk
5869
        if 'size' in disk_dict:
5870
          raise errors.OpPrereqError("Disk size change not possible, use"
5871
                                     " grow-disk")
5872

    
5873
    if disk_addremove > 1:
5874
      raise errors.OpPrereqError("Only one disk add or remove operation"
5875
                                 " supported at a time")
5876

    
5877
    # NIC validation
5878
    nic_addremove = 0
5879
    for nic_op, nic_dict in self.op.nics:
5880
      if nic_op == constants.DDM_REMOVE:
5881
        nic_addremove += 1
5882
        continue
5883
      elif nic_op == constants.DDM_ADD:
5884
        nic_addremove += 1
5885
      else:
5886
        if not isinstance(nic_op, int):
5887
          raise errors.OpPrereqError("Invalid nic index")
5888

    
5889
      # nic_dict should be a dict
5890
      nic_ip = nic_dict.get('ip', None)
5891
      if nic_ip is not None:
5892
        if nic_ip.lower() == constants.VALUE_NONE:
5893
          nic_dict['ip'] = None
5894
        else:
5895
          if not utils.IsValidIP(nic_ip):
5896
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5897

    
5898
      if nic_op == constants.DDM_ADD:
5899
        nic_bridge = nic_dict.get('bridge', None)
5900
        if nic_bridge is None:
5901
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5902
        nic_mac = nic_dict.get('mac', None)
5903
        if nic_mac is None:
5904
          nic_dict['mac'] = constants.VALUE_AUTO
5905

    
5906
      if 'mac' in nic_dict:
5907
        nic_mac = nic_dict['mac']
5908
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5909
          if not utils.IsValidMac(nic_mac):
5910
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5911
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5912
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5913
                                     " modifying an existing nic")
5914

    
5915
    if nic_addremove > 1:
5916
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5917
                                 " supported at a time")
5918

    
5919
  def ExpandNames(self):
5920
    self._ExpandAndLockInstance()
5921
    self.needed_locks[locking.LEVEL_NODE] = []
5922
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5923

    
5924
  def DeclareLocks(self, level):
5925
    if level == locking.LEVEL_NODE:
5926
      self._LockInstancesNodes()
5927

    
5928
  def BuildHooksEnv(self):
5929
    """Build hooks env.
5930

5931
    This runs on the master, primary and secondaries.
5932

5933
    """
5934
    args = dict()
5935
    if constants.BE_MEMORY in self.be_new:
5936
      args['memory'] = self.be_new[constants.BE_MEMORY]
5937
    if constants.BE_VCPUS in self.be_new:
5938
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5939
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5940
    # information at all.
5941
    if self.op.nics:
5942
      args['nics'] = []
5943
      nic_override = dict(self.op.nics)
5944
      for idx, nic in enumerate(self.instance.nics):
5945
        if idx in nic_override:
5946
          this_nic_override = nic_override[idx]
5947
        else:
5948
          this_nic_override = {}
5949
        if 'ip' in this_nic_override:
5950
          ip = this_nic_override['ip']
5951
        else:
5952
          ip = nic.ip
5953
        if 'bridge' in this_nic_override:
5954
          bridge = this_nic_override['bridge']
5955
        else:
5956
          bridge = nic.bridge
5957
        if 'mac' in this_nic_override:
5958
          mac = this_nic_override['mac']
5959
        else:
5960
          mac = nic.mac
5961
        args['nics'].append((ip, bridge, mac))
5962
      if constants.DDM_ADD in nic_override:
5963
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5964
        bridge = nic_override[constants.DDM_ADD]['bridge']
5965
        mac = nic_override[constants.DDM_ADD]['mac']
5966
        args['nics'].append((ip, bridge, mac))
5967
      elif constants.DDM_REMOVE in nic_override:
5968
        del args['nics'][-1]
5969

    
5970
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5971
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5972
    return env, nl, nl
5973

    
5974
  def CheckPrereq(self):
5975
    """Check prerequisites.
5976

5977
    This only checks the instance list against the existing names.
5978

5979
    """
5980
    force = self.force = self.op.force
5981

    
5982
    # checking the new params on the primary/secondary nodes
5983

    
5984
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5985
    assert self.instance is not None, \
5986
      "Cannot retrieve locked instance %s" % self.op.instance_name
5987
    pnode = instance.primary_node
5988
    nodelist = list(instance.all_nodes)
5989

    
5990
    # hvparams processing
5991
    if self.op.hvparams:
5992
      i_hvdict = copy.deepcopy(instance.hvparams)
5993
      for key, val in self.op.hvparams.iteritems():
5994
        if val == constants.VALUE_DEFAULT:
5995
          try:
5996
            del i_hvdict[key]
5997
          except KeyError:
5998
            pass
5999
        else:
6000
          i_hvdict[key] = val
6001
      cluster = self.cfg.GetClusterInfo()
6002
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
6003
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
6004
                                i_hvdict)
6005
      # local check
6006
      hypervisor.GetHypervisor(
6007
        instance.hypervisor).CheckParameterSyntax(hv_new)
6008
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
6009
      self.hv_new = hv_new # the new actual values
6010
      self.hv_inst = i_hvdict # the new dict (without defaults)
6011
    else:
6012
      self.hv_new = self.hv_inst = {}
6013

    
6014
    # beparams processing
6015
    if self.op.beparams:
6016
      i_bedict = copy.deepcopy(instance.beparams)
6017
      for key, val in self.op.beparams.iteritems():
6018
        if val == constants.VALUE_DEFAULT:
6019
          try:
6020
            del i_bedict[key]
6021
          except KeyError:
6022
            pass
6023
        else:
6024
          i_bedict[key] = val
6025
      cluster = self.cfg.GetClusterInfo()
6026
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
6027
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
6028
                                i_bedict)
6029
      self.be_new = be_new # the new actual values
6030
      self.be_inst = i_bedict # the new dict (without defaults)
6031
    else:
6032
      self.be_new = self.be_inst = {}
6033

    
6034
    self.warn = []
6035

    
6036
    if constants.BE_MEMORY in self.op.beparams and not self.force:
6037
      mem_check_list = [pnode]
6038
      if be_new[constants.BE_AUTO_BALANCE]:
6039
        # either we changed auto_balance to yes or it was from before
6040
        mem_check_list.extend(instance.secondary_nodes)
6041
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
6042
                                                  instance.hypervisor)
6043
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
6044
                                         instance.hypervisor)
6045
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
6046
        # Assume the primary node is unreachable and go ahead
6047
        self.warn.append("Can't get info from primary node %s" % pnode)
6048
      else:
6049
        if not instance_info.failed and instance_info.data:
6050
          current_mem = int(instance_info.data['memory'])
6051
        else:
6052
          # Assume instance not running
6053
          # (there is a slight race condition here, but it's not very probable,
6054
          # and we have no other way to check)
6055
          current_mem = 0
6056
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
6057
                    nodeinfo[pnode].data['memory_free'])
6058
        if miss_mem > 0:
6059
          raise errors.OpPrereqError("This change will prevent the instance"
6060
                                     " from starting, due to %d MB of memory"
6061
                                     " missing on its primary node" % miss_mem)
6062

    
6063
      if be_new[constants.BE_AUTO_BALANCE]:
6064
        for node, nres in nodeinfo.iteritems():
6065
          if node not in instance.secondary_nodes:
6066
            continue
6067
          if nres.failed or not isinstance(nres.data, dict):
6068
            self.warn.append("Can't get info from secondary node %s" % node)
6069
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
6070
            self.warn.append("Not enough memory to failover instance to"
6071
                             " secondary node %s" % node)
6072

    
6073
    # NIC processing
6074
    for nic_op, nic_dict in self.op.nics:
6075
      if nic_op == constants.DDM_REMOVE:
6076
        if not instance.nics:
6077
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
6078
        continue
6079
      if nic_op != constants.DDM_ADD:
6080
        # an existing nic
6081
        if nic_op < 0 or nic_op >= len(instance.nics):
6082
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
6083
                                     " are 0 to %d" %
6084
                                     (nic_op, len(instance.nics)))
6085
      if 'bridge' in nic_dict:
6086
        nic_bridge = nic_dict['bridge']
6087
        if nic_bridge is None:
6088
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
6089
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
6090
          msg = ("Bridge '%s' doesn't exist on one of"
6091
                 " the instance nodes" % nic_bridge)
6092
          if self.force:
6093
            self.warn.append(msg)
6094
          else:
6095
            raise errors.OpPrereqError(msg)
6096
      if 'mac' in nic_dict:
6097
        nic_mac = nic_dict['mac']
6098
        if nic_mac is None:
6099
          raise errors.OpPrereqError('Cannot set the nic mac to None')
6100
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6101
          # otherwise generate the mac
6102
          nic_dict['mac'] = self.cfg.GenerateMAC()
6103
        else:
6104
          # or validate/reserve the current one
6105
          if self.cfg.IsMacInUse(nic_mac):
6106
            raise errors.OpPrereqError("MAC address %s already in use"
6107
                                       " in cluster" % nic_mac)
6108

    
6109
    # DISK processing
6110
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
6111
      raise errors.OpPrereqError("Disk operations not supported for"
6112
                                 " diskless instances")
6113
    for disk_op, disk_dict in self.op.disks:
6114
      if disk_op == constants.DDM_REMOVE:
6115
        if len(instance.disks) == 1:
6116
          raise errors.OpPrereqError("Cannot remove the last disk of"
6117
                                     " an instance")
6118
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
6119
        ins_l = ins_l[pnode]
6120
        if ins_l.failed or not isinstance(ins_l.data, list):
6121
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
6122
        if instance.name in ins_l.data:
6123
          raise errors.OpPrereqError("Instance is running, can't remove"
6124
                                     " disks.")
6125

    
6126
      if (disk_op == constants.DDM_ADD and
6127
          len(instance.nics) >= constants.MAX_DISKS):
6128
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
6129
                                   " add more" % constants.MAX_DISKS)
6130
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
6131
        # an existing disk
6132
        if disk_op < 0 or disk_op >= len(instance.disks):
6133
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
6134
                                     " are 0 to %d" %
6135
                                     (disk_op, len(instance.disks)))
6136

    
6137
    return
6138

    
6139
  def Exec(self, feedback_fn):
6140
    """Modifies an instance.
6141

6142
    All parameters take effect only at the next restart of the instance.
6143

6144
    """
6145
    # Process here the warnings from CheckPrereq, as we don't have a
6146
    # feedback_fn there.
6147
    for warn in self.warn:
6148
      feedback_fn("WARNING: %s" % warn)
6149

    
6150
    result = []
6151
    instance = self.instance
6152
    # disk changes
6153
    for disk_op, disk_dict in self.op.disks:
6154
      if disk_op == constants.DDM_REMOVE:
6155
        # remove the last disk
6156
        device = instance.disks.pop()
6157
        device_idx = len(instance.disks)
6158
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6159
          self.cfg.SetDiskID(disk, node)
6160
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6161
          if msg:
6162
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6163
                            " continuing anyway", device_idx, node, msg)
6164
        result.append(("disk/%d" % device_idx, "remove"))
6165
      elif disk_op == constants.DDM_ADD:
6166
        # add a new disk
6167
        if instance.disk_template == constants.DT_FILE:
6168
          file_driver, file_path = instance.disks[0].logical_id
6169
          file_path = os.path.dirname(file_path)
6170
        else:
6171
          file_driver = file_path = None
6172
        disk_idx_base = len(instance.disks)
6173
        new_disk = _GenerateDiskTemplate(self,
6174
                                         instance.disk_template,
6175
                                         instance.name, instance.primary_node,
6176
                                         instance.secondary_nodes,
6177
                                         [disk_dict],
6178
                                         file_path,
6179
                                         file_driver,
6180
                                         disk_idx_base)[0]
6181
        instance.disks.append(new_disk)
6182
        info = _GetInstanceInfoText(instance)
6183

    
6184
        logging.info("Creating volume %s for instance %s",
6185
                     new_disk.iv_name, instance.name)
6186
        # Note: this needs to be kept in sync with _CreateDisks
6187
        #HARDCODE
6188
        for node in instance.all_nodes:
6189
          f_create = node == instance.primary_node
6190
          try:
6191
            _CreateBlockDev(self, node, instance, new_disk,
6192
                            f_create, info, f_create)
6193
          except errors.OpExecError, err:
6194
            self.LogWarning("Failed to create volume %s (%s) on"
6195
                            " node %s: %s",
6196
                            new_disk.iv_name, new_disk, node, err)
6197
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6198
                       (new_disk.size, new_disk.mode)))
6199
      else:
6200
        # change a given disk
6201
        instance.disks[disk_op].mode = disk_dict['mode']
6202
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6203
    # NIC changes
6204
    for nic_op, nic_dict in self.op.nics:
6205
      if nic_op == constants.DDM_REMOVE:
6206
        # remove the last nic
6207
        del instance.nics[-1]
6208
        result.append(("nic.%d" % len(instance.nics), "remove"))
6209
      elif nic_op == constants.DDM_ADD:
6210
        # mac and bridge should be set, by now
6211
        mac = nic_dict['mac']
6212
        bridge = nic_dict['bridge']
6213
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6214
                              bridge=bridge)
6215
        instance.nics.append(new_nic)
6216
        result.append(("nic.%d" % (len(instance.nics) - 1),
6217
                       "add:mac=%s,ip=%s,bridge=%s" %
6218
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6219
      else:
6220
        # change a given nic
6221
        for key in 'mac', 'ip', 'bridge':
6222
          if key in nic_dict:
6223
            setattr(instance.nics[nic_op], key, nic_dict[key])
6224
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6225

    
6226
    # hvparams changes
6227
    if self.op.hvparams:
6228
      instance.hvparams = self.hv_inst
6229
      for key, val in self.op.hvparams.iteritems():
6230
        result.append(("hv/%s" % key, val))
6231

    
6232
    # beparams changes
6233
    if self.op.beparams:
6234
      instance.beparams = self.be_inst
6235
      for key, val in self.op.beparams.iteritems():
6236
        result.append(("be/%s" % key, val))
6237

    
6238
    self.cfg.Update(instance)
6239

    
6240
    return result
6241

    
6242

    
6243
class LUQueryExports(NoHooksLU):
6244
  """Query the exports list
6245

6246
  """
6247
  _OP_REQP = ['nodes']
6248
  REQ_BGL = False
6249

    
6250
  def ExpandNames(self):
6251
    self.needed_locks = {}
6252
    self.share_locks[locking.LEVEL_NODE] = 1
6253
    if not self.op.nodes:
6254
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6255
    else:
6256
      self.needed_locks[locking.LEVEL_NODE] = \
6257
        _GetWantedNodes(self, self.op.nodes)
6258

    
6259
  def CheckPrereq(self):
6260
    """Check prerequisites.
6261

6262
    """
6263
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6264

    
6265
  def Exec(self, feedback_fn):
6266
    """Compute the list of all the exported system images.
6267

6268
    @rtype: dict
6269
    @return: a dictionary with the structure node->(export-list)
6270
        where export-list is a list of the instances exported on
6271
        that node.
6272

6273
    """
6274
    rpcresult = self.rpc.call_export_list(self.nodes)
6275
    result = {}
6276
    for node in rpcresult:
6277
      if rpcresult[node].failed:
6278
        result[node] = False
6279
      else:
6280
        result[node] = rpcresult[node].data
6281

    
6282
    return result
6283

    
6284

    
6285
class LUExportInstance(LogicalUnit):
6286
  """Export an instance to an image in the cluster.
6287

6288
  """
6289
  HPATH = "instance-export"
6290
  HTYPE = constants.HTYPE_INSTANCE
6291
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6292
  REQ_BGL = False
6293

    
6294
  def ExpandNames(self):
6295
    self._ExpandAndLockInstance()
6296
    # FIXME: lock only instance primary and destination node
6297
    #
6298
    # Sad but true, for now we have do lock all nodes, as we don't know where
6299
    # the previous export might be, and and in this LU we search for it and
6300
    # remove it from its current node. In the future we could fix this by:
6301
    #  - making a tasklet to search (share-lock all), then create the new one,
6302
    #    then one to remove, after
6303
    #  - removing the removal operation altoghether
6304
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6305

    
6306
  def DeclareLocks(self, level):
6307
    """Last minute lock declaration."""
6308
    # All nodes are locked anyway, so nothing to do here.
6309

    
6310
  def BuildHooksEnv(self):
6311
    """Build hooks env.
6312

6313
    This will run on the master, primary node and target node.
6314

6315
    """
6316
    env = {
6317
      "EXPORT_NODE": self.op.target_node,
6318
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6319
      }
6320
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6321
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6322
          self.op.target_node]
6323
    return env, nl, nl
6324

    
6325
  def CheckPrereq(self):
6326
    """Check prerequisites.
6327

6328
    This checks that the instance and node names are valid.
6329

6330
    """
6331
    instance_name = self.op.instance_name
6332
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6333
    assert self.instance is not None, \
6334
          "Cannot retrieve locked instance %s" % self.op.instance_name
6335
    _CheckNodeOnline(self, self.instance.primary_node)
6336

    
6337
    self.dst_node = self.cfg.GetNodeInfo(
6338
      self.cfg.ExpandNodeName(self.op.target_node))
6339

    
6340
    if self.dst_node is None:
6341
      # This is wrong node name, not a non-locked node
6342
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6343
    _CheckNodeOnline(self, self.dst_node.name)
6344
    _CheckNodeNotDrained(self, self.dst_node.name)
6345

    
6346
    # instance disk type verification
6347
    for disk in self.instance.disks:
6348
      if disk.dev_type == constants.LD_FILE:
6349
        raise errors.OpPrereqError("Export not supported for instances with"
6350
                                   " file-based disks")
6351

    
6352
  def Exec(self, feedback_fn):
6353
    """Export an instance to an image in the cluster.
6354

6355
    """
6356
    instance = self.instance
6357
    dst_node = self.dst_node
6358
    src_node = instance.primary_node
6359
    if self.op.shutdown:
6360
      # shutdown the instance, but not the disks
6361
      result = self.rpc.call_instance_shutdown(src_node, instance)
6362
      msg = result.RemoteFailMsg()
6363
      if msg:
6364
        raise errors.OpExecError("Could not shutdown instance %s on"
6365
                                 " node %s: %s" %
6366
                                 (instance.name, src_node, msg))
6367

    
6368
    vgname = self.cfg.GetVGName()
6369

    
6370
    snap_disks = []
6371

    
6372
    # set the disks ID correctly since call_instance_start needs the
6373
    # correct drbd minor to create the symlinks
6374
    for disk in instance.disks:
6375
      self.cfg.SetDiskID(disk, src_node)
6376

    
6377
    try:
6378
      for idx, disk in enumerate(instance.disks):
6379
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6380
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6381
        if new_dev_name.failed or not new_dev_name.data:
6382
          self.LogWarning("Could not snapshot disk/%d on node %s",
6383
                          idx, src_node)
6384
          snap_disks.append(False)
6385
        else:
6386
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6387
                                 logical_id=(vgname, new_dev_name.data),
6388
                                 physical_id=(vgname, new_dev_name.data),
6389
                                 iv_name=disk.iv_name)
6390
          snap_disks.append(new_dev)
6391

    
6392
    finally:
6393
      if self.op.shutdown and instance.admin_up:
6394
        result = self.rpc.call_instance_start(src_node, instance, None, None)
6395
        msg = result.RemoteFailMsg()
6396
        if msg:
6397
          _ShutdownInstanceDisks(self, instance)
6398
          raise errors.OpExecError("Could not start instance: %s" % msg)
6399

    
6400
    # TODO: check for size
6401

    
6402
    cluster_name = self.cfg.GetClusterName()
6403
    for idx, dev in enumerate(snap_disks):
6404
      if dev:
6405
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6406
                                               instance, cluster_name, idx)
6407
        if result.failed or not result.data:
6408
          self.LogWarning("Could not export disk/%d from node %s to"
6409
                          " node %s", idx, src_node, dst_node.name)
6410
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6411
        if msg:
6412
          self.LogWarning("Could not remove snapshot for disk/%d from node"
6413
                          " %s: %s", idx, src_node, msg)
6414

    
6415
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6416
    if result.failed or not result.data:
6417
      self.LogWarning("Could not finalize export for instance %s on node %s",
6418
                      instance.name, dst_node.name)
6419

    
6420
    nodelist = self.cfg.GetNodeList()
6421
    nodelist.remove(dst_node.name)
6422

    
6423
    # on one-node clusters nodelist will be empty after the removal
6424
    # if we proceed the backup would be removed because OpQueryExports
6425
    # substitutes an empty list with the full cluster node list.
6426
    if nodelist:
6427
      exportlist = self.rpc.call_export_list(nodelist)
6428
      for node in exportlist:
6429
        if exportlist[node].failed:
6430
          continue
6431
        if instance.name in exportlist[node].data:
6432
          if not self.rpc.call_export_remove(node, instance.name):
6433
            self.LogWarning("Could not remove older export for instance %s"
6434
                            " on node %s", instance.name, node)
6435

    
6436

    
6437
class LURemoveExport(NoHooksLU):
6438
  """Remove exports related to the named instance.
6439

6440
  """
6441
  _OP_REQP = ["instance_name"]
6442
  REQ_BGL = False
6443

    
6444
  def ExpandNames(self):
6445
    self.needed_locks = {}
6446
    # We need all nodes to be locked in order for RemoveExport to work, but we
6447
    # don't need to lock the instance itself, as nothing will happen to it (and
6448
    # we can remove exports also for a removed instance)
6449
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6450

    
6451
  def CheckPrereq(self):
6452
    """Check prerequisites.
6453
    """
6454
    pass
6455

    
6456
  def Exec(self, feedback_fn):
6457
    """Remove any export.
6458

6459
    """
6460
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6461
    # If the instance was not found we'll try with the name that was passed in.
6462
    # This will only work if it was an FQDN, though.
6463
    fqdn_warn = False
6464
    if not instance_name:
6465
      fqdn_warn = True
6466
      instance_name = self.op.instance_name
6467

    
6468
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6469
      locking.LEVEL_NODE])
6470
    found = False
6471
    for node in exportlist:
6472
      if exportlist[node].failed:
6473
        self.LogWarning("Failed to query node %s, continuing" % node)
6474
        continue
6475
      if instance_name in exportlist[node].data:
6476
        found = True
6477
        result = self.rpc.call_export_remove(node, instance_name)
6478
        if result.failed or not result.data:
6479
          logging.error("Could not remove export for instance %s"
6480
                        " on node %s", instance_name, node)
6481

    
6482
    if fqdn_warn and not found:
6483
      feedback_fn("Export not found. If trying to remove an export belonging"
6484
                  " to a deleted instance please use its Fully Qualified"
6485
                  " Domain Name.")
6486

    
6487

    
6488
class TagsLU(NoHooksLU):
6489
  """Generic tags LU.
6490

6491
  This is an abstract class which is the parent of all the other tags LUs.
6492

6493
  """
6494

    
6495
  def ExpandNames(self):
6496
    self.needed_locks = {}
6497
    if self.op.kind == constants.TAG_NODE:
6498
      name = self.cfg.ExpandNodeName(self.op.name)
6499
      if name is None:
6500
        raise errors.OpPrereqError("Invalid node name (%s)" %
6501
                                   (self.op.name,))
6502
      self.op.name = name
6503
      self.needed_locks[locking.LEVEL_NODE] = name
6504
    elif self.op.kind == constants.TAG_INSTANCE:
6505
      name = self.cfg.ExpandInstanceName(self.op.name)
6506
      if name is None:
6507
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6508
                                   (self.op.name,))
6509
      self.op.name = name
6510
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6511

    
6512
  def CheckPrereq(self):
6513
    """Check prerequisites.
6514

6515
    """
6516
    if self.op.kind == constants.TAG_CLUSTER:
6517
      self.target = self.cfg.GetClusterInfo()
6518
    elif self.op.kind == constants.TAG_NODE:
6519
      self.target = self.cfg.GetNodeInfo(self.op.name)
6520
    elif self.op.kind == constants.TAG_INSTANCE:
6521
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6522
    else:
6523
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6524
                                 str(self.op.kind))
6525

    
6526

    
6527
class LUGetTags(TagsLU):
6528
  """Returns the tags of a given object.
6529

6530
  """
6531
  _OP_REQP = ["kind", "name"]
6532
  REQ_BGL = False
6533

    
6534
  def Exec(self, feedback_fn):
6535
    """Returns the tag list.
6536

6537
    """
6538
    return list(self.target.GetTags())
6539

    
6540

    
6541
class LUSearchTags(NoHooksLU):
6542
  """Searches the tags for a given pattern.
6543

6544
  """
6545
  _OP_REQP = ["pattern"]
6546
  REQ_BGL = False
6547

    
6548
  def ExpandNames(self):
6549
    self.needed_locks = {}
6550

    
6551
  def CheckPrereq(self):
6552
    """Check prerequisites.
6553

6554
    This checks the pattern passed for validity by compiling it.
6555

6556
    """
6557
    try:
6558
      self.re = re.compile(self.op.pattern)
6559
    except re.error, err:
6560
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6561
                                 (self.op.pattern, err))
6562

    
6563
  def Exec(self, feedback_fn):
6564
    """Returns the tag list.
6565

6566
    """
6567
    cfg = self.cfg
6568
    tgts = [("/cluster", cfg.GetClusterInfo())]
6569
    ilist = cfg.GetAllInstancesInfo().values()
6570
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6571
    nlist = cfg.GetAllNodesInfo().values()
6572
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6573
    results = []
6574
    for path, target in tgts:
6575
      for tag in target.GetTags():
6576
        if self.re.search(tag):
6577
          results.append((path, tag))
6578
    return results
6579

    
6580

    
6581
class LUAddTags(TagsLU):
6582
  """Sets a tag on a given object.
6583

6584
  """
6585
  _OP_REQP = ["kind", "name", "tags"]
6586
  REQ_BGL = False
6587

    
6588
  def CheckPrereq(self):
6589
    """Check prerequisites.
6590

6591
    This checks the type and length of the tag name and value.
6592

6593
    """
6594
    TagsLU.CheckPrereq(self)
6595
    for tag in self.op.tags:
6596
      objects.TaggableObject.ValidateTag(tag)
6597

    
6598
  def Exec(self, feedback_fn):
6599
    """Sets the tag.
6600

6601
    """
6602
    try:
6603
      for tag in self.op.tags:
6604
        self.target.AddTag(tag)
6605
    except errors.TagError, err:
6606
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6607
    try:
6608
      self.cfg.Update(self.target)
6609
    except errors.ConfigurationError:
6610
      raise errors.OpRetryError("There has been a modification to the"
6611
                                " config file and the operation has been"
6612
                                " aborted. Please retry.")
6613

    
6614

    
6615
class LUDelTags(TagsLU):
6616
  """Delete a list of tags from a given object.
6617

6618
  """
6619
  _OP_REQP = ["kind", "name", "tags"]
6620
  REQ_BGL = False
6621

    
6622
  def CheckPrereq(self):
6623
    """Check prerequisites.
6624

6625
    This checks that we have the given tag.
6626

6627
    """
6628
    TagsLU.CheckPrereq(self)
6629
    for tag in self.op.tags:
6630
      objects.TaggableObject.ValidateTag(tag)
6631
    del_tags = frozenset(self.op.tags)
6632
    cur_tags = self.target.GetTags()
6633
    if not del_tags <= cur_tags:
6634
      diff_tags = del_tags - cur_tags
6635
      diff_names = ["'%s'" % tag for tag in diff_tags]
6636
      diff_names.sort()
6637
      raise errors.OpPrereqError("Tag(s) %s not found" %
6638
                                 (",".join(diff_names)))
6639

    
6640
  def Exec(self, feedback_fn):
6641
    """Remove the tag from the object.
6642

6643
    """
6644
    for tag in self.op.tags:
6645
      self.target.RemoveTag(tag)
6646
    try:
6647
      self.cfg.Update(self.target)
6648
    except errors.ConfigurationError:
6649
      raise errors.OpRetryError("There has been a modification to the"
6650
                                " config file and the operation has been"
6651
                                " aborted. Please retry.")
6652

    
6653

    
6654
class LUTestDelay(NoHooksLU):
6655
  """Sleep for a specified amount of time.
6656

6657
  This LU sleeps on the master and/or nodes for a specified amount of
6658
  time.
6659

6660
  """
6661
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6662
  REQ_BGL = False
6663

    
6664
  def ExpandNames(self):
6665
    """Expand names and set required locks.
6666

6667
    This expands the node list, if any.
6668

6669
    """
6670
    self.needed_locks = {}
6671
    if self.op.on_nodes:
6672
      # _GetWantedNodes can be used here, but is not always appropriate to use
6673
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6674
      # more information.
6675
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6676
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6677

    
6678
  def CheckPrereq(self):
6679
    """Check prerequisites.
6680

6681
    """
6682

    
6683
  def Exec(self, feedback_fn):
6684
    """Do the actual sleep.
6685

6686
    """
6687
    if self.op.on_master:
6688
      if not utils.TestDelay(self.op.duration):
6689
        raise errors.OpExecError("Error during master delay test")
6690
    if self.op.on_nodes:
6691
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6692
      if not result:
6693
        raise errors.OpExecError("Complete failure from rpc call")
6694
      for node, node_result in result.items():
6695
        node_result.Raise()
6696
        if not node_result.data:
6697
          raise errors.OpExecError("Failure during rpc call to node %s,"
6698
                                   " result: %s" % (node, node_result.data))
6699

    
6700

    
6701
class IAllocator(object):
6702
  """IAllocator framework.
6703

6704
  An IAllocator instance has three sets of attributes:
6705
    - cfg that is needed to query the cluster
6706
    - input data (all members of the _KEYS class attribute are required)
6707
    - four buffer attributes (in|out_data|text), that represent the
6708
      input (to the external script) in text and data structure format,
6709
      and the output from it, again in two formats
6710
    - the result variables from the script (success, info, nodes) for
6711
      easy usage
6712

6713
  """
6714
  _ALLO_KEYS = [
6715
    "mem_size", "disks", "disk_template",
6716
    "os", "tags", "nics", "vcpus", "hypervisor",
6717
    ]
6718
  _RELO_KEYS = [
6719
    "relocate_from",
6720
    ]
6721

    
6722
  def __init__(self, lu, mode, name, **kwargs):
6723
    self.lu = lu
6724
    # init buffer variables
6725
    self.in_text = self.out_text = self.in_data = self.out_data = None
6726
    # init all input fields so that pylint is happy
6727
    self.mode = mode
6728
    self.name = name
6729
    self.mem_size = self.disks = self.disk_template = None
6730
    self.os = self.tags = self.nics = self.vcpus = None
6731
    self.hypervisor = None
6732
    self.relocate_from = None
6733
    # computed fields
6734
    self.required_nodes = None
6735
    # init result fields
6736
    self.success = self.info = self.nodes = None
6737
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6738
      keyset = self._ALLO_KEYS
6739
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6740
      keyset = self._RELO_KEYS
6741
    else:
6742
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6743
                                   " IAllocator" % self.mode)
6744
    for key in kwargs:
6745
      if key not in keyset:
6746
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6747
                                     " IAllocator" % key)
6748
      setattr(self, key, kwargs[key])
6749
    for key in keyset:
6750
      if key not in kwargs:
6751
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6752
                                     " IAllocator" % key)
6753
    self._BuildInputData()
6754

    
6755
  def _ComputeClusterData(self):
6756
    """Compute the generic allocator input data.
6757

6758
    This is the data that is independent of the actual operation.
6759

6760
    """
6761
    cfg = self.lu.cfg
6762
    cluster_info = cfg.GetClusterInfo()
6763
    # cluster data
6764
    data = {
6765
      "version": constants.IALLOCATOR_VERSION,
6766
      "cluster_name": cfg.GetClusterName(),
6767
      "cluster_tags": list(cluster_info.GetTags()),
6768
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6769
      # we don't have job IDs
6770
      }
6771
    iinfo = cfg.GetAllInstancesInfo().values()
6772
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6773

    
6774
    # node data
6775
    node_results = {}
6776
    node_list = cfg.GetNodeList()
6777

    
6778
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6779
      hypervisor_name = self.hypervisor
6780
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6781
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6782

    
6783
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6784
                                           hypervisor_name)
6785
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6786
                       cluster_info.enabled_hypervisors)
6787
    for nname, nresult in node_data.items():
6788
      # first fill in static (config-based) values
6789
      ninfo = cfg.GetNodeInfo(nname)
6790
      pnr = {
6791
        "tags": list(ninfo.GetTags()),
6792
        "primary_ip": ninfo.primary_ip,
6793
        "secondary_ip": ninfo.secondary_ip,
6794
        "offline": ninfo.offline,
6795
        "drained": ninfo.drained,
6796
        "master_candidate": ninfo.master_candidate,
6797
        }
6798

    
6799
      if not ninfo.offline:
6800
        nresult.Raise()
6801
        if not isinstance(nresult.data, dict):
6802
          raise errors.OpExecError("Can't get data for node %s" % nname)
6803
        remote_info = nresult.data
6804
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6805
                     'vg_size', 'vg_free', 'cpu_total']:
6806
          if attr not in remote_info:
6807
            raise errors.OpExecError("Node '%s' didn't return attribute"
6808
                                     " '%s'" % (nname, attr))
6809
          try:
6810
            remote_info[attr] = int(remote_info[attr])
6811
          except ValueError, err:
6812
            raise errors.OpExecError("Node '%s' returned invalid value"
6813
                                     " for '%s': %s" % (nname, attr, err))
6814
        # compute memory used by primary instances
6815
        i_p_mem = i_p_up_mem = 0
6816
        for iinfo, beinfo in i_list:
6817
          if iinfo.primary_node == nname:
6818
            i_p_mem += beinfo[constants.BE_MEMORY]
6819
            if iinfo.name not in node_iinfo[nname].data:
6820
              i_used_mem = 0
6821
            else:
6822
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6823
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6824
            remote_info['memory_free'] -= max(0, i_mem_diff)
6825

    
6826
            if iinfo.admin_up:
6827
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6828

    
6829
        # compute memory used by instances
6830
        pnr_dyn = {
6831
          "total_memory": remote_info['memory_total'],
6832
          "reserved_memory": remote_info['memory_dom0'],
6833
          "free_memory": remote_info['memory_free'],
6834
          "total_disk": remote_info['vg_size'],
6835
          "free_disk": remote_info['vg_free'],
6836
          "total_cpus": remote_info['cpu_total'],
6837
          "i_pri_memory": i_p_mem,
6838
          "i_pri_up_memory": i_p_up_mem,
6839
          }
6840
        pnr.update(pnr_dyn)
6841

    
6842
      node_results[nname] = pnr
6843
    data["nodes"] = node_results
6844

    
6845
    # instance data
6846
    instance_data = {}
6847
    for iinfo, beinfo in i_list:
6848
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6849
                  for n in iinfo.nics]
6850
      pir = {
6851
        "tags": list(iinfo.GetTags()),
6852
        "admin_up": iinfo.admin_up,
6853
        "vcpus": beinfo[constants.BE_VCPUS],
6854
        "memory": beinfo[constants.BE_MEMORY],
6855
        "os": iinfo.os,
6856
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6857
        "nics": nic_data,
6858
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6859
        "disk_template": iinfo.disk_template,
6860
        "hypervisor": iinfo.hypervisor,
6861
        }
6862
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
6863
                                                 pir["disks"])
6864
      instance_data[iinfo.name] = pir
6865

    
6866
    data["instances"] = instance_data
6867

    
6868
    self.in_data = data
6869

    
6870
  def _AddNewInstance(self):
6871
    """Add new instance data to allocator structure.
6872

6873
    This in combination with _AllocatorGetClusterData will create the
6874
    correct structure needed as input for the allocator.
6875

6876
    The checks for the completeness of the opcode must have already been
6877
    done.
6878

6879
    """
6880
    data = self.in_data
6881

    
6882
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6883

    
6884
    if self.disk_template in constants.DTS_NET_MIRROR:
6885
      self.required_nodes = 2
6886
    else:
6887
      self.required_nodes = 1
6888
    request = {
6889
      "type": "allocate",
6890
      "name": self.name,
6891
      "disk_template": self.disk_template,
6892
      "tags": self.tags,
6893
      "os": self.os,
6894
      "vcpus": self.vcpus,
6895
      "memory": self.mem_size,
6896
      "disks": self.disks,
6897
      "disk_space_total": disk_space,
6898
      "nics": self.nics,
6899
      "required_nodes": self.required_nodes,
6900
      }
6901
    data["request"] = request
6902

    
6903
  def _AddRelocateInstance(self):
6904
    """Add relocate instance data to allocator structure.
6905

6906
    This in combination with _IAllocatorGetClusterData will create the
6907
    correct structure needed as input for the allocator.
6908

6909
    The checks for the completeness of the opcode must have already been
6910
    done.
6911

6912
    """
6913
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6914
    if instance is None:
6915
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6916
                                   " IAllocator" % self.name)
6917

    
6918
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6919
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6920

    
6921
    if len(instance.secondary_nodes) != 1:
6922
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6923

    
6924
    self.required_nodes = 1
6925
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6926
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6927

    
6928
    request = {
6929
      "type": "relocate",
6930
      "name": self.name,
6931
      "disk_space_total": disk_space,
6932
      "required_nodes": self.required_nodes,
6933
      "relocate_from": self.relocate_from,
6934
      }
6935
    self.in_data["request"] = request
6936

    
6937
  def _BuildInputData(self):
6938
    """Build input data structures.
6939

6940
    """
6941
    self._ComputeClusterData()
6942

    
6943
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6944
      self._AddNewInstance()
6945
    else:
6946
      self._AddRelocateInstance()
6947

    
6948
    self.in_text = serializer.Dump(self.in_data)
6949

    
6950
  def Run(self, name, validate=True, call_fn=None):
6951
    """Run an instance allocator and return the results.
6952

6953
    """
6954
    if call_fn is None:
6955
      call_fn = self.lu.rpc.call_iallocator_runner
6956
    data = self.in_text
6957

    
6958
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6959
    result.Raise()
6960

    
6961
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6962
      raise errors.OpExecError("Invalid result from master iallocator runner")
6963

    
6964
    rcode, stdout, stderr, fail = result.data
6965

    
6966
    if rcode == constants.IARUN_NOTFOUND:
6967
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6968
    elif rcode == constants.IARUN_FAILURE:
6969
      raise errors.OpExecError("Instance allocator call failed: %s,"
6970
                               " output: %s" % (fail, stdout+stderr))
6971
    self.out_text = stdout
6972
    if validate:
6973
      self._ValidateResult()
6974

    
6975
  def _ValidateResult(self):
6976
    """Process the allocator results.
6977

6978
    This will process and if successful save the result in
6979
    self.out_data and the other parameters.
6980

6981
    """
6982
    try:
6983
      rdict = serializer.Load(self.out_text)
6984
    except Exception, err:
6985
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6986

    
6987
    if not isinstance(rdict, dict):
6988
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6989

    
6990
    for key in "success", "info", "nodes":
6991
      if key not in rdict:
6992
        raise errors.OpExecError("Can't parse iallocator results:"
6993
                                 " missing key '%s'" % key)
6994
      setattr(self, key, rdict[key])
6995

    
6996
    if not isinstance(rdict["nodes"], list):
6997
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6998
                               " is not a list")
6999
    self.out_data = rdict
7000

    
7001

    
7002
class LUTestAllocator(NoHooksLU):
7003
  """Run allocator tests.
7004

7005
  This LU runs the allocator tests
7006

7007
  """
7008
  _OP_REQP = ["direction", "mode", "name"]
7009

    
7010
  def CheckPrereq(self):
7011
    """Check prerequisites.
7012

7013
    This checks the opcode parameters depending on the director and mode test.
7014

7015
    """
7016
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7017
      for attr in ["name", "mem_size", "disks", "disk_template",
7018
                   "os", "tags", "nics", "vcpus"]:
7019
        if not hasattr(self.op, attr):
7020
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
7021
                                     attr)
7022
      iname = self.cfg.ExpandInstanceName(self.op.name)
7023
      if iname is not None:
7024
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
7025
                                   iname)
7026
      if not isinstance(self.op.nics, list):
7027
        raise errors.OpPrereqError("Invalid parameter 'nics'")
7028
      for row in self.op.nics:
7029
        if (not isinstance(row, dict) or
7030
            "mac" not in row or
7031
            "ip" not in row or
7032
            "bridge" not in row):
7033
          raise errors.OpPrereqError("Invalid contents of the"
7034
                                     " 'nics' parameter")
7035
      if not isinstance(self.op.disks, list):
7036
        raise errors.OpPrereqError("Invalid parameter 'disks'")
7037
      for row in self.op.disks:
7038
        if (not isinstance(row, dict) or
7039
            "size" not in row or
7040
            not isinstance(row["size"], int) or
7041
            "mode" not in row or
7042
            row["mode"] not in ['r', 'w']):
7043
          raise errors.OpPrereqError("Invalid contents of the"
7044
                                     " 'disks' parameter")
7045
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
7046
        self.op.hypervisor = self.cfg.GetHypervisorType()
7047
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
7048
      if not hasattr(self.op, "name"):
7049
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
7050
      fname = self.cfg.ExpandInstanceName(self.op.name)
7051
      if fname is None:
7052
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
7053
                                   self.op.name)
7054
      self.op.name = fname
7055
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
7056
    else:
7057
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
7058
                                 self.op.mode)
7059

    
7060
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
7061
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
7062
        raise errors.OpPrereqError("Missing allocator name")
7063
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
7064
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
7065
                                 self.op.direction)
7066

    
7067
  def Exec(self, feedback_fn):
7068
    """Run the allocator test.
7069

7070
    """
7071
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7072
      ial = IAllocator(self,
7073
                       mode=self.op.mode,
7074
                       name=self.op.name,
7075
                       mem_size=self.op.mem_size,
7076
                       disks=self.op.disks,
7077
                       disk_template=self.op.disk_template,
7078
                       os=self.op.os,
7079
                       tags=self.op.tags,
7080
                       nics=self.op.nics,
7081
                       vcpus=self.op.vcpus,
7082
                       hypervisor=self.op.hypervisor,
7083
                       )
7084
    else:
7085
      ial = IAllocator(self,
7086
                       mode=self.op.mode,
7087
                       name=self.op.name,
7088
                       relocate_from=list(self.relocate_from),
7089
                       )
7090

    
7091
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
7092
      result = ial.in_text
7093
    else:
7094
      ial.Run(self.op.allocator, validate=False)
7095
      result = ial.out_text
7096
    return result