Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ b4ec07f8

History | View | Annotate | Download (247.6 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
    if not hasattr(self.op, "ignore_size"):
2596
      self.op.ignore_size = False
2597

    
2598
  def Exec(self, feedback_fn):
2599
    """Activate the disks.
2600

2601
    """
2602
    disks_ok, disks_info = \
2603
              _AssembleInstanceDisks(self, self.instance,
2604
                                     ignore_size=self.op.ignore_size)
2605
    if not disks_ok:
2606
      raise errors.OpExecError("Cannot activate block devices")
2607

    
2608
    return disks_info
2609

    
2610

    
2611
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False,
2612
                           ignore_size=False):
2613
  """Prepare the block devices for an instance.
2614

2615
  This sets up the block devices on all nodes.
2616

2617
  @type lu: L{LogicalUnit}
2618
  @param lu: the logical unit on whose behalf we execute
2619
  @type instance: L{objects.Instance}
2620
  @param instance: the instance for whose disks we assemble
2621
  @type ignore_secondaries: boolean
2622
  @param ignore_secondaries: if true, errors on secondary nodes
2623
      won't result in an error return from the function
2624
  @type ignore_size: boolean
2625
  @param ignore_size: if true, the current known size of the disk
2626
      will not be used during the disk activation, useful for cases
2627
      when the size is wrong
2628
  @return: False if the operation failed, otherwise a list of
2629
      (host, instance_visible_name, node_visible_name)
2630
      with the mapping from node devices to instance devices
2631

2632
  """
2633
  device_info = []
2634
  disks_ok = True
2635
  iname = instance.name
2636
  # With the two passes mechanism we try to reduce the window of
2637
  # opportunity for the race condition of switching DRBD to primary
2638
  # before handshaking occured, but we do not eliminate it
2639

    
2640
  # The proper fix would be to wait (with some limits) until the
2641
  # connection has been made and drbd transitions from WFConnection
2642
  # into any other network-connected state (Connected, SyncTarget,
2643
  # SyncSource, etc.)
2644

    
2645
  # 1st pass, assemble on all nodes in secondary mode
2646
  for inst_disk in instance.disks:
2647
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2648
      if ignore_size:
2649
        node_disk = node_disk.Copy()
2650
        node_disk.UnsetSize()
2651
      lu.cfg.SetDiskID(node_disk, node)
2652
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2653
      msg = result.RemoteFailMsg()
2654
      if msg:
2655
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2656
                           " (is_primary=False, pass=1): %s",
2657
                           inst_disk.iv_name, node, msg)
2658
        if not ignore_secondaries:
2659
          disks_ok = False
2660

    
2661
  # FIXME: race condition on drbd migration to primary
2662

    
2663
  # 2nd pass, do only the primary node
2664
  for inst_disk in instance.disks:
2665
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2666
      if node != instance.primary_node:
2667
        continue
2668
      if ignore_size:
2669
        node_disk = node_disk.Copy()
2670
        node_disk.UnsetSize()
2671
      lu.cfg.SetDiskID(node_disk, node)
2672
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2673
      msg = result.RemoteFailMsg()
2674
      if msg:
2675
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2676
                           " (is_primary=True, pass=2): %s",
2677
                           inst_disk.iv_name, node, msg)
2678
        disks_ok = False
2679
    device_info.append((instance.primary_node, inst_disk.iv_name,
2680
                        result.payload))
2681

    
2682
  # leave the disks configured for the primary node
2683
  # this is a workaround that would be fixed better by
2684
  # improving the logical/physical id handling
2685
  for disk in instance.disks:
2686
    lu.cfg.SetDiskID(disk, instance.primary_node)
2687

    
2688
  return disks_ok, device_info
2689

    
2690

    
2691
def _StartInstanceDisks(lu, instance, force):
2692
  """Start the disks of an instance.
2693

2694
  """
2695
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2696
                                           ignore_secondaries=force)
2697
  if not disks_ok:
2698
    _ShutdownInstanceDisks(lu, instance)
2699
    if force is not None and not force:
2700
      lu.proc.LogWarning("", hint="If the message above refers to a"
2701
                         " secondary node,"
2702
                         " you can retry the operation using '--force'.")
2703
    raise errors.OpExecError("Disk consistency error")
2704

    
2705

    
2706
class LUDeactivateInstanceDisks(NoHooksLU):
2707
  """Shutdown an instance's disks.
2708

2709
  """
2710
  _OP_REQP = ["instance_name"]
2711
  REQ_BGL = False
2712

    
2713
  def ExpandNames(self):
2714
    self._ExpandAndLockInstance()
2715
    self.needed_locks[locking.LEVEL_NODE] = []
2716
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2717

    
2718
  def DeclareLocks(self, level):
2719
    if level == locking.LEVEL_NODE:
2720
      self._LockInstancesNodes()
2721

    
2722
  def CheckPrereq(self):
2723
    """Check prerequisites.
2724

2725
    This checks that the instance is in the cluster.
2726

2727
    """
2728
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2729
    assert self.instance is not None, \
2730
      "Cannot retrieve locked instance %s" % self.op.instance_name
2731

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

2735
    """
2736
    instance = self.instance
2737
    _SafeShutdownInstanceDisks(self, instance)
2738

    
2739

    
2740
def _SafeShutdownInstanceDisks(lu, instance):
2741
  """Shutdown block devices of an instance.
2742

2743
  This function checks if an instance is running, before calling
2744
  _ShutdownInstanceDisks.
2745

2746
  """
2747
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2748
                                      [instance.hypervisor])
2749
  ins_l = ins_l[instance.primary_node]
2750
  if ins_l.failed or not isinstance(ins_l.data, list):
2751
    raise errors.OpExecError("Can't contact node '%s'" %
2752
                             instance.primary_node)
2753

    
2754
  if instance.name in ins_l.data:
2755
    raise errors.OpExecError("Instance is running, can't shutdown"
2756
                             " block devices.")
2757

    
2758
  _ShutdownInstanceDisks(lu, instance)
2759

    
2760

    
2761
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2762
  """Shutdown block devices of an instance.
2763

2764
  This does the shutdown on all nodes of the instance.
2765

2766
  If the ignore_primary is false, errors on the primary node are
2767
  ignored.
2768

2769
  """
2770
  all_result = True
2771
  for disk in instance.disks:
2772
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2773
      lu.cfg.SetDiskID(top_disk, node)
2774
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2775
      msg = result.RemoteFailMsg()
2776
      if msg:
2777
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2778
                      disk.iv_name, node, msg)
2779
        if not ignore_primary or node != instance.primary_node:
2780
          all_result = False
2781
  return all_result
2782

    
2783

    
2784
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2785
  """Checks if a node has enough free memory.
2786

2787
  This function check if a given node has the needed amount of free
2788
  memory. In case the node has less memory or we cannot get the
2789
  information from the node, this function raise an OpPrereqError
2790
  exception.
2791

2792
  @type lu: C{LogicalUnit}
2793
  @param lu: a logical unit from which we get configuration data
2794
  @type node: C{str}
2795
  @param node: the node to check
2796
  @type reason: C{str}
2797
  @param reason: string to use in the error message
2798
  @type requested: C{int}
2799
  @param requested: the amount of memory in MiB to check for
2800
  @type hypervisor_name: C{str}
2801
  @param hypervisor_name: the hypervisor to ask for memory stats
2802
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2803
      we cannot check the node
2804

2805
  """
2806
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2807
  nodeinfo[node].Raise()
2808
  free_mem = nodeinfo[node].data.get('memory_free')
2809
  if not isinstance(free_mem, int):
2810
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2811
                             " was '%s'" % (node, free_mem))
2812
  if requested > free_mem:
2813
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2814
                             " needed %s MiB, available %s MiB" %
2815
                             (node, reason, requested, free_mem))
2816

    
2817

    
2818
class LUStartupInstance(LogicalUnit):
2819
  """Starts an instance.
2820

2821
  """
2822
  HPATH = "instance-start"
2823
  HTYPE = constants.HTYPE_INSTANCE
2824
  _OP_REQP = ["instance_name", "force"]
2825
  REQ_BGL = False
2826

    
2827
  def ExpandNames(self):
2828
    self._ExpandAndLockInstance()
2829

    
2830
  def BuildHooksEnv(self):
2831
    """Build hooks env.
2832

2833
    This runs on master, primary and secondary nodes of the instance.
2834

2835
    """
2836
    env = {
2837
      "FORCE": self.op.force,
2838
      }
2839
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2840
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2841
    return env, nl, nl
2842

    
2843
  def CheckPrereq(self):
2844
    """Check prerequisites.
2845

2846
    This checks that the instance is in the cluster.
2847

2848
    """
2849
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2850
    assert self.instance is not None, \
2851
      "Cannot retrieve locked instance %s" % self.op.instance_name
2852

    
2853
    # extra beparams
2854
    self.beparams = getattr(self.op, "beparams", {})
2855
    if self.beparams:
2856
      if not isinstance(self.beparams, dict):
2857
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2858
                                   " dict" % (type(self.beparams), ))
2859
      # fill the beparams dict
2860
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2861
      self.op.beparams = self.beparams
2862

    
2863
    # extra hvparams
2864
    self.hvparams = getattr(self.op, "hvparams", {})
2865
    if self.hvparams:
2866
      if not isinstance(self.hvparams, dict):
2867
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2868
                                   " dict" % (type(self.hvparams), ))
2869

    
2870
      # check hypervisor parameter syntax (locally)
2871
      cluster = self.cfg.GetClusterInfo()
2872
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2873
      filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
2874
                                    instance.hvparams)
2875
      filled_hvp.update(self.hvparams)
2876
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2877
      hv_type.CheckParameterSyntax(filled_hvp)
2878
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2879
      self.op.hvparams = self.hvparams
2880

    
2881
    _CheckNodeOnline(self, instance.primary_node)
2882

    
2883
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2884
    # check bridges existance
2885
    _CheckInstanceBridgesExist(self, instance)
2886

    
2887
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2888
                                              instance.name,
2889
                                              instance.hypervisor)
2890
    remote_info.Raise()
2891
    if not remote_info.data:
2892
      _CheckNodeFreeMemory(self, instance.primary_node,
2893
                           "starting instance %s" % instance.name,
2894
                           bep[constants.BE_MEMORY], instance.hypervisor)
2895

    
2896
  def Exec(self, feedback_fn):
2897
    """Start the instance.
2898

2899
    """
2900
    instance = self.instance
2901
    force = self.op.force
2902

    
2903
    self.cfg.MarkInstanceUp(instance.name)
2904

    
2905
    node_current = instance.primary_node
2906

    
2907
    _StartInstanceDisks(self, instance, force)
2908

    
2909
    result = self.rpc.call_instance_start(node_current, instance,
2910
                                          self.hvparams, self.beparams)
2911
    msg = result.RemoteFailMsg()
2912
    if msg:
2913
      _ShutdownInstanceDisks(self, instance)
2914
      raise errors.OpExecError("Could not start instance: %s" % msg)
2915

    
2916

    
2917
class LURebootInstance(LogicalUnit):
2918
  """Reboot an instance.
2919

2920
  """
2921
  HPATH = "instance-reboot"
2922
  HTYPE = constants.HTYPE_INSTANCE
2923
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2924
  REQ_BGL = False
2925

    
2926
  def ExpandNames(self):
2927
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2928
                                   constants.INSTANCE_REBOOT_HARD,
2929
                                   constants.INSTANCE_REBOOT_FULL]:
2930
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2931
                                  (constants.INSTANCE_REBOOT_SOFT,
2932
                                   constants.INSTANCE_REBOOT_HARD,
2933
                                   constants.INSTANCE_REBOOT_FULL))
2934
    self._ExpandAndLockInstance()
2935

    
2936
  def BuildHooksEnv(self):
2937
    """Build hooks env.
2938

2939
    This runs on master, primary and secondary nodes of the instance.
2940

2941
    """
2942
    env = {
2943
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2944
      "REBOOT_TYPE": self.op.reboot_type,
2945
      }
2946
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2947
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2948
    return env, nl, nl
2949

    
2950
  def CheckPrereq(self):
2951
    """Check prerequisites.
2952

2953
    This checks that the instance is in the cluster.
2954

2955
    """
2956
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2957
    assert self.instance is not None, \
2958
      "Cannot retrieve locked instance %s" % self.op.instance_name
2959

    
2960
    _CheckNodeOnline(self, instance.primary_node)
2961

    
2962
    # check bridges existance
2963
    _CheckInstanceBridgesExist(self, instance)
2964

    
2965
  def Exec(self, feedback_fn):
2966
    """Reboot the instance.
2967

2968
    """
2969
    instance = self.instance
2970
    ignore_secondaries = self.op.ignore_secondaries
2971
    reboot_type = self.op.reboot_type
2972

    
2973
    node_current = instance.primary_node
2974

    
2975
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2976
                       constants.INSTANCE_REBOOT_HARD]:
2977
      for disk in instance.disks:
2978
        self.cfg.SetDiskID(disk, node_current)
2979
      result = self.rpc.call_instance_reboot(node_current, instance,
2980
                                             reboot_type)
2981
      msg = result.RemoteFailMsg()
2982
      if msg:
2983
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2984
    else:
2985
      result = self.rpc.call_instance_shutdown(node_current, instance)
2986
      msg = result.RemoteFailMsg()
2987
      if msg:
2988
        raise errors.OpExecError("Could not shutdown instance for"
2989
                                 " full reboot: %s" % msg)
2990
      _ShutdownInstanceDisks(self, instance)
2991
      _StartInstanceDisks(self, instance, ignore_secondaries)
2992
      result = self.rpc.call_instance_start(node_current, instance, None, None)
2993
      msg = result.RemoteFailMsg()
2994
      if msg:
2995
        _ShutdownInstanceDisks(self, instance)
2996
        raise errors.OpExecError("Could not start instance for"
2997
                                 " full reboot: %s" % msg)
2998

    
2999
    self.cfg.MarkInstanceUp(instance.name)
3000

    
3001

    
3002
class LUShutdownInstance(LogicalUnit):
3003
  """Shutdown an instance.
3004

3005
  """
3006
  HPATH = "instance-stop"
3007
  HTYPE = constants.HTYPE_INSTANCE
3008
  _OP_REQP = ["instance_name"]
3009
  REQ_BGL = False
3010

    
3011
  def ExpandNames(self):
3012
    self._ExpandAndLockInstance()
3013

    
3014
  def BuildHooksEnv(self):
3015
    """Build hooks env.
3016

3017
    This runs on master, primary and secondary nodes of the instance.
3018

3019
    """
3020
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3021
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3022
    return env, nl, nl
3023

    
3024
  def CheckPrereq(self):
3025
    """Check prerequisites.
3026

3027
    This checks that the instance is in the cluster.
3028

3029
    """
3030
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3031
    assert self.instance is not None, \
3032
      "Cannot retrieve locked instance %s" % self.op.instance_name
3033
    _CheckNodeOnline(self, self.instance.primary_node)
3034

    
3035
  def Exec(self, feedback_fn):
3036
    """Shutdown the instance.
3037

3038
    """
3039
    instance = self.instance
3040
    node_current = instance.primary_node
3041
    self.cfg.MarkInstanceDown(instance.name)
3042
    result = self.rpc.call_instance_shutdown(node_current, instance)
3043
    msg = result.RemoteFailMsg()
3044
    if msg:
3045
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3046

    
3047
    _ShutdownInstanceDisks(self, instance)
3048

    
3049

    
3050
class LUReinstallInstance(LogicalUnit):
3051
  """Reinstall an instance.
3052

3053
  """
3054
  HPATH = "instance-reinstall"
3055
  HTYPE = constants.HTYPE_INSTANCE
3056
  _OP_REQP = ["instance_name"]
3057
  REQ_BGL = False
3058

    
3059
  def ExpandNames(self):
3060
    self._ExpandAndLockInstance()
3061

    
3062
  def BuildHooksEnv(self):
3063
    """Build hooks env.
3064

3065
    This runs on master, primary and secondary nodes of the instance.
3066

3067
    """
3068
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3069
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3070
    return env, nl, nl
3071

    
3072
  def CheckPrereq(self):
3073
    """Check prerequisites.
3074

3075
    This checks that the instance is in the cluster and is not running.
3076

3077
    """
3078
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3079
    assert instance is not None, \
3080
      "Cannot retrieve locked instance %s" % self.op.instance_name
3081
    _CheckNodeOnline(self, instance.primary_node)
3082

    
3083
    if instance.disk_template == constants.DT_DISKLESS:
3084
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3085
                                 self.op.instance_name)
3086
    if instance.admin_up:
3087
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3088
                                 self.op.instance_name)
3089
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3090
                                              instance.name,
3091
                                              instance.hypervisor)
3092
    remote_info.Raise()
3093
    if remote_info.data:
3094
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3095
                                 (self.op.instance_name,
3096
                                  instance.primary_node))
3097

    
3098
    self.op.os_type = getattr(self.op, "os_type", None)
3099
    if self.op.os_type is not None:
3100
      # OS verification
3101
      pnode = self.cfg.GetNodeInfo(
3102
        self.cfg.ExpandNodeName(instance.primary_node))
3103
      if pnode is None:
3104
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3105
                                   self.op.pnode)
3106
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3107
      result.Raise()
3108
      if not isinstance(result.data, objects.OS):
3109
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3110
                                   " primary node"  % self.op.os_type)
3111

    
3112
    self.instance = instance
3113

    
3114
  def Exec(self, feedback_fn):
3115
    """Reinstall the instance.
3116

3117
    """
3118
    inst = self.instance
3119

    
3120
    if self.op.os_type is not None:
3121
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3122
      inst.os = self.op.os_type
3123
      self.cfg.Update(inst)
3124

    
3125
    _StartInstanceDisks(self, inst, None)
3126
    try:
3127
      feedback_fn("Running the instance OS create scripts...")
3128
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3129
      msg = result.RemoteFailMsg()
3130
      if msg:
3131
        raise errors.OpExecError("Could not install OS for instance %s"
3132
                                 " on node %s: %s" %
3133
                                 (inst.name, inst.primary_node, msg))
3134
    finally:
3135
      _ShutdownInstanceDisks(self, inst)
3136

    
3137

    
3138
class LURenameInstance(LogicalUnit):
3139
  """Rename an instance.
3140

3141
  """
3142
  HPATH = "instance-rename"
3143
  HTYPE = constants.HTYPE_INSTANCE
3144
  _OP_REQP = ["instance_name", "new_name"]
3145

    
3146
  def BuildHooksEnv(self):
3147
    """Build hooks env.
3148

3149
    This runs on master, primary and secondary nodes of the instance.
3150

3151
    """
3152
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3153
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3154
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3155
    return env, nl, nl
3156

    
3157
  def CheckPrereq(self):
3158
    """Check prerequisites.
3159

3160
    This checks that the instance is in the cluster and is not running.
3161

3162
    """
3163
    instance = self.cfg.GetInstanceInfo(
3164
      self.cfg.ExpandInstanceName(self.op.instance_name))
3165
    if instance is None:
3166
      raise errors.OpPrereqError("Instance '%s' not known" %
3167
                                 self.op.instance_name)
3168
    _CheckNodeOnline(self, instance.primary_node)
3169

    
3170
    if instance.admin_up:
3171
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3172
                                 self.op.instance_name)
3173
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3174
                                              instance.name,
3175
                                              instance.hypervisor)
3176
    remote_info.Raise()
3177
    if remote_info.data:
3178
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3179
                                 (self.op.instance_name,
3180
                                  instance.primary_node))
3181
    self.instance = instance
3182

    
3183
    # new name verification
3184
    name_info = utils.HostInfo(self.op.new_name)
3185

    
3186
    self.op.new_name = new_name = name_info.name
3187
    instance_list = self.cfg.GetInstanceList()
3188
    if new_name in instance_list:
3189
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3190
                                 new_name)
3191

    
3192
    if not getattr(self.op, "ignore_ip", False):
3193
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3194
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3195
                                   (name_info.ip, new_name))
3196

    
3197

    
3198
  def Exec(self, feedback_fn):
3199
    """Reinstall the instance.
3200

3201
    """
3202
    inst = self.instance
3203
    old_name = inst.name
3204

    
3205
    if inst.disk_template == constants.DT_FILE:
3206
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3207

    
3208
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3209
    # Change the instance lock. This is definitely safe while we hold the BGL
3210
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3211
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3212

    
3213
    # re-read the instance from the configuration after rename
3214
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3215

    
3216
    if inst.disk_template == constants.DT_FILE:
3217
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3218
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3219
                                                     old_file_storage_dir,
3220
                                                     new_file_storage_dir)
3221
      result.Raise()
3222
      if not result.data:
3223
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3224
                                 " directory '%s' to '%s' (but the instance"
3225
                                 " has been renamed in Ganeti)" % (
3226
                                 inst.primary_node, old_file_storage_dir,
3227
                                 new_file_storage_dir))
3228

    
3229
      if not result.data[0]:
3230
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3231
                                 " (but the instance has been renamed in"
3232
                                 " Ganeti)" % (old_file_storage_dir,
3233
                                               new_file_storage_dir))
3234

    
3235
    _StartInstanceDisks(self, inst, None)
3236
    try:
3237
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3238
                                                 old_name)
3239
      msg = result.RemoteFailMsg()
3240
      if msg:
3241
        msg = ("Could not run OS rename script for instance %s on node %s"
3242
               " (but the instance has been renamed in Ganeti): %s" %
3243
               (inst.name, inst.primary_node, msg))
3244
        self.proc.LogWarning(msg)
3245
    finally:
3246
      _ShutdownInstanceDisks(self, inst)
3247

    
3248

    
3249
class LURemoveInstance(LogicalUnit):
3250
  """Remove an instance.
3251

3252
  """
3253
  HPATH = "instance-remove"
3254
  HTYPE = constants.HTYPE_INSTANCE
3255
  _OP_REQP = ["instance_name", "ignore_failures"]
3256
  REQ_BGL = False
3257

    
3258
  def ExpandNames(self):
3259
    self._ExpandAndLockInstance()
3260
    self.needed_locks[locking.LEVEL_NODE] = []
3261
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3262

    
3263
  def DeclareLocks(self, level):
3264
    if level == locking.LEVEL_NODE:
3265
      self._LockInstancesNodes()
3266

    
3267
  def BuildHooksEnv(self):
3268
    """Build hooks env.
3269

3270
    This runs on master, primary and secondary nodes of the instance.
3271

3272
    """
3273
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3274
    nl = [self.cfg.GetMasterNode()]
3275
    return env, nl, nl
3276

    
3277
  def CheckPrereq(self):
3278
    """Check prerequisites.
3279

3280
    This checks that the instance is in the cluster.
3281

3282
    """
3283
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3284
    assert self.instance is not None, \
3285
      "Cannot retrieve locked instance %s" % self.op.instance_name
3286

    
3287
  def Exec(self, feedback_fn):
3288
    """Remove the instance.
3289

3290
    """
3291
    instance = self.instance
3292
    logging.info("Shutting down instance %s on node %s",
3293
                 instance.name, instance.primary_node)
3294

    
3295
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3296
    msg = result.RemoteFailMsg()
3297
    if msg:
3298
      if self.op.ignore_failures:
3299
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3300
      else:
3301
        raise errors.OpExecError("Could not shutdown instance %s on"
3302
                                 " node %s: %s" %
3303
                                 (instance.name, instance.primary_node, msg))
3304

    
3305
    logging.info("Removing block devices for instance %s", instance.name)
3306

    
3307
    if not _RemoveDisks(self, instance):
3308
      if self.op.ignore_failures:
3309
        feedback_fn("Warning: can't remove instance's disks")
3310
      else:
3311
        raise errors.OpExecError("Can't remove instance's disks")
3312

    
3313
    logging.info("Removing instance %s out of cluster config", instance.name)
3314

    
3315
    self.cfg.RemoveInstance(instance.name)
3316
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3317

    
3318

    
3319
class LUQueryInstances(NoHooksLU):
3320
  """Logical unit for querying instances.
3321

3322
  """
3323
  _OP_REQP = ["output_fields", "names", "use_locking"]
3324
  REQ_BGL = False
3325
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3326
                                    "admin_state",
3327
                                    "disk_template", "ip", "mac", "bridge",
3328
                                    "sda_size", "sdb_size", "vcpus", "tags",
3329
                                    "network_port", "beparams",
3330
                                    r"(disk)\.(size)/([0-9]+)",
3331
                                    r"(disk)\.(sizes)", "disk_usage",
3332
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3333
                                    r"(nic)\.(macs|ips|bridges)",
3334
                                    r"(disk|nic)\.(count)",
3335
                                    "serial_no", "hypervisor", "hvparams",] +
3336
                                  ["hv/%s" % name
3337
                                   for name in constants.HVS_PARAMETERS] +
3338
                                  ["be/%s" % name
3339
                                   for name in constants.BES_PARAMETERS])
3340
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3341

    
3342

    
3343
  def ExpandNames(self):
3344
    _CheckOutputFields(static=self._FIELDS_STATIC,
3345
                       dynamic=self._FIELDS_DYNAMIC,
3346
                       selected=self.op.output_fields)
3347

    
3348
    self.needed_locks = {}
3349
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3350
    self.share_locks[locking.LEVEL_NODE] = 1
3351

    
3352
    if self.op.names:
3353
      self.wanted = _GetWantedInstances(self, self.op.names)
3354
    else:
3355
      self.wanted = locking.ALL_SET
3356

    
3357
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3358
    self.do_locking = self.do_node_query and self.op.use_locking
3359
    if self.do_locking:
3360
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3361
      self.needed_locks[locking.LEVEL_NODE] = []
3362
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3363

    
3364
  def DeclareLocks(self, level):
3365
    if level == locking.LEVEL_NODE and self.do_locking:
3366
      self._LockInstancesNodes()
3367

    
3368
  def CheckPrereq(self):
3369
    """Check prerequisites.
3370

3371
    """
3372
    pass
3373

    
3374
  def Exec(self, feedback_fn):
3375
    """Computes the list of nodes and their attributes.
3376

3377
    """
3378
    all_info = self.cfg.GetAllInstancesInfo()
3379
    if self.wanted == locking.ALL_SET:
3380
      # caller didn't specify instance names, so ordering is not important
3381
      if self.do_locking:
3382
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3383
      else:
3384
        instance_names = all_info.keys()
3385
      instance_names = utils.NiceSort(instance_names)
3386
    else:
3387
      # caller did specify names, so we must keep the ordering
3388
      if self.do_locking:
3389
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3390
      else:
3391
        tgt_set = all_info.keys()
3392
      missing = set(self.wanted).difference(tgt_set)
3393
      if missing:
3394
        raise errors.OpExecError("Some instances were removed before"
3395
                                 " retrieving their data: %s" % missing)
3396
      instance_names = self.wanted
3397

    
3398
    instance_list = [all_info[iname] for iname in instance_names]
3399

    
3400
    # begin data gathering
3401

    
3402
    nodes = frozenset([inst.primary_node for inst in instance_list])
3403
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3404

    
3405
    bad_nodes = []
3406
    off_nodes = []
3407
    if self.do_node_query:
3408
      live_data = {}
3409
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3410
      for name in nodes:
3411
        result = node_data[name]
3412
        if result.offline:
3413
          # offline nodes will be in both lists
3414
          off_nodes.append(name)
3415
        if result.failed:
3416
          bad_nodes.append(name)
3417
        else:
3418
          if result.data:
3419
            live_data.update(result.data)
3420
            # else no instance is alive
3421
    else:
3422
      live_data = dict([(name, {}) for name in instance_names])
3423

    
3424
    # end data gathering
3425

    
3426
    HVPREFIX = "hv/"
3427
    BEPREFIX = "be/"
3428
    output = []
3429
    for instance in instance_list:
3430
      iout = []
3431
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3432
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3433
      for field in self.op.output_fields:
3434
        st_match = self._FIELDS_STATIC.Matches(field)
3435
        if field == "name":
3436
          val = instance.name
3437
        elif field == "os":
3438
          val = instance.os
3439
        elif field == "pnode":
3440
          val = instance.primary_node
3441
        elif field == "snodes":
3442
          val = list(instance.secondary_nodes)
3443
        elif field == "admin_state":
3444
          val = instance.admin_up
3445
        elif field == "oper_state":
3446
          if instance.primary_node in bad_nodes:
3447
            val = None
3448
          else:
3449
            val = bool(live_data.get(instance.name))
3450
        elif field == "status":
3451
          if instance.primary_node in off_nodes:
3452
            val = "ERROR_nodeoffline"
3453
          elif instance.primary_node in bad_nodes:
3454
            val = "ERROR_nodedown"
3455
          else:
3456
            running = bool(live_data.get(instance.name))
3457
            if running:
3458
              if instance.admin_up:
3459
                val = "running"
3460
              else:
3461
                val = "ERROR_up"
3462
            else:
3463
              if instance.admin_up:
3464
                val = "ERROR_down"
3465
              else:
3466
                val = "ADMIN_down"
3467
        elif field == "oper_ram":
3468
          if instance.primary_node in bad_nodes:
3469
            val = None
3470
          elif instance.name in live_data:
3471
            val = live_data[instance.name].get("memory", "?")
3472
          else:
3473
            val = "-"
3474
        elif field == "vcpus":
3475
          val = i_be[constants.BE_VCPUS]
3476
        elif field == "disk_template":
3477
          val = instance.disk_template
3478
        elif field == "ip":
3479
          if instance.nics:
3480
            val = instance.nics[0].ip
3481
          else:
3482
            val = None
3483
        elif field == "bridge":
3484
          if instance.nics:
3485
            val = instance.nics[0].bridge
3486
          else:
3487
            val = None
3488
        elif field == "mac":
3489
          if instance.nics:
3490
            val = instance.nics[0].mac
3491
          else:
3492
            val = None
3493
        elif field == "sda_size" or field == "sdb_size":
3494
          idx = ord(field[2]) - ord('a')
3495
          try:
3496
            val = instance.FindDisk(idx).size
3497
          except errors.OpPrereqError:
3498
            val = None
3499
        elif field == "disk_usage": # total disk usage per node
3500
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3501
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3502
        elif field == "tags":
3503
          val = list(instance.GetTags())
3504
        elif field == "serial_no":
3505
          val = instance.serial_no
3506
        elif field == "network_port":
3507
          val = instance.network_port
3508
        elif field == "hypervisor":
3509
          val = instance.hypervisor
3510
        elif field == "hvparams":
3511
          val = i_hv
3512
        elif (field.startswith(HVPREFIX) and
3513
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3514
          val = i_hv.get(field[len(HVPREFIX):], None)
3515
        elif field == "beparams":
3516
          val = i_be
3517
        elif (field.startswith(BEPREFIX) and
3518
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3519
          val = i_be.get(field[len(BEPREFIX):], None)
3520
        elif st_match and st_match.groups():
3521
          # matches a variable list
3522
          st_groups = st_match.groups()
3523
          if st_groups and st_groups[0] == "disk":
3524
            if st_groups[1] == "count":
3525
              val = len(instance.disks)
3526
            elif st_groups[1] == "sizes":
3527
              val = [disk.size for disk in instance.disks]
3528
            elif st_groups[1] == "size":
3529
              try:
3530
                val = instance.FindDisk(st_groups[2]).size
3531
              except errors.OpPrereqError:
3532
                val = None
3533
            else:
3534
              assert False, "Unhandled disk parameter"
3535
          elif st_groups[0] == "nic":
3536
            if st_groups[1] == "count":
3537
              val = len(instance.nics)
3538
            elif st_groups[1] == "macs":
3539
              val = [nic.mac for nic in instance.nics]
3540
            elif st_groups[1] == "ips":
3541
              val = [nic.ip for nic in instance.nics]
3542
            elif st_groups[1] == "bridges":
3543
              val = [nic.bridge for nic in instance.nics]
3544
            else:
3545
              # index-based item
3546
              nic_idx = int(st_groups[2])
3547
              if nic_idx >= len(instance.nics):
3548
                val = None
3549
              else:
3550
                if st_groups[1] == "mac":
3551
                  val = instance.nics[nic_idx].mac
3552
                elif st_groups[1] == "ip":
3553
                  val = instance.nics[nic_idx].ip
3554
                elif st_groups[1] == "bridge":
3555
                  val = instance.nics[nic_idx].bridge
3556
                else:
3557
                  assert False, "Unhandled NIC parameter"
3558
          else:
3559
            assert False, ("Declared but unhandled variable parameter '%s'" %
3560
                           field)
3561
        else:
3562
          assert False, "Declared but unhandled parameter '%s'" % field
3563
        iout.append(val)
3564
      output.append(iout)
3565

    
3566
    return output
3567

    
3568

    
3569
class LUFailoverInstance(LogicalUnit):
3570
  """Failover an instance.
3571

3572
  """
3573
  HPATH = "instance-failover"
3574
  HTYPE = constants.HTYPE_INSTANCE
3575
  _OP_REQP = ["instance_name", "ignore_consistency"]
3576
  REQ_BGL = False
3577

    
3578
  def ExpandNames(self):
3579
    self._ExpandAndLockInstance()
3580
    self.needed_locks[locking.LEVEL_NODE] = []
3581
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3582

    
3583
  def DeclareLocks(self, level):
3584
    if level == locking.LEVEL_NODE:
3585
      self._LockInstancesNodes()
3586

    
3587
  def BuildHooksEnv(self):
3588
    """Build hooks env.
3589

3590
    This runs on master, primary and secondary nodes of the instance.
3591

3592
    """
3593
    env = {
3594
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3595
      }
3596
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3597
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3598
    return env, nl, nl
3599

    
3600
  def CheckPrereq(self):
3601
    """Check prerequisites.
3602

3603
    This checks that the instance is in the cluster.
3604

3605
    """
3606
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3607
    assert self.instance is not None, \
3608
      "Cannot retrieve locked instance %s" % self.op.instance_name
3609

    
3610
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3611
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3612
      raise errors.OpPrereqError("Instance's disk layout is not"
3613
                                 " network mirrored, cannot failover.")
3614

    
3615
    secondary_nodes = instance.secondary_nodes
3616
    if not secondary_nodes:
3617
      raise errors.ProgrammerError("no secondary node but using "
3618
                                   "a mirrored disk template")
3619

    
3620
    target_node = secondary_nodes[0]
3621
    _CheckNodeOnline(self, target_node)
3622
    _CheckNodeNotDrained(self, target_node)
3623

    
3624
    if instance.admin_up:
3625
      # check memory requirements on the secondary node
3626
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3627
                           instance.name, bep[constants.BE_MEMORY],
3628
                           instance.hypervisor)
3629
    else:
3630
      self.LogInfo("Not checking memory on the secondary node as"
3631
                   " instance will not be started")
3632

    
3633
    # check bridge existance
3634
    brlist = [nic.bridge for nic in instance.nics]
3635
    result = self.rpc.call_bridges_exist(target_node, brlist)
3636
    result.Raise()
3637
    if not result.data:
3638
      raise errors.OpPrereqError("One or more target bridges %s does not"
3639
                                 " exist on destination node '%s'" %
3640
                                 (brlist, target_node))
3641

    
3642
  def Exec(self, feedback_fn):
3643
    """Failover an instance.
3644

3645
    The failover is done by shutting it down on its present node and
3646
    starting it on the secondary.
3647

3648
    """
3649
    instance = self.instance
3650

    
3651
    source_node = instance.primary_node
3652
    target_node = instance.secondary_nodes[0]
3653

    
3654
    feedback_fn("* checking disk consistency between source and target")
3655
    for dev in instance.disks:
3656
      # for drbd, these are drbd over lvm
3657
      if not _CheckDiskConsistency(self, dev, target_node, False):
3658
        if instance.admin_up and not self.op.ignore_consistency:
3659
          raise errors.OpExecError("Disk %s is degraded on target node,"
3660
                                   " aborting failover." % dev.iv_name)
3661

    
3662
    feedback_fn("* shutting down instance on source node")
3663
    logging.info("Shutting down instance %s on node %s",
3664
                 instance.name, source_node)
3665

    
3666
    result = self.rpc.call_instance_shutdown(source_node, instance)
3667
    msg = result.RemoteFailMsg()
3668
    if msg:
3669
      if self.op.ignore_consistency:
3670
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3671
                             " Proceeding anyway. Please make sure node"
3672
                             " %s is down. Error details: %s",
3673
                             instance.name, source_node, source_node, msg)
3674
      else:
3675
        raise errors.OpExecError("Could not shutdown instance %s on"
3676
                                 " node %s: %s" %
3677
                                 (instance.name, source_node, msg))
3678

    
3679
    feedback_fn("* deactivating the instance's disks on source node")
3680
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3681
      raise errors.OpExecError("Can't shut down the instance's disks.")
3682

    
3683
    instance.primary_node = target_node
3684
    # distribute new instance config to the other nodes
3685
    self.cfg.Update(instance)
3686

    
3687
    # Only start the instance if it's marked as up
3688
    if instance.admin_up:
3689
      feedback_fn("* activating the instance's disks on target node")
3690
      logging.info("Starting instance %s on node %s",
3691
                   instance.name, target_node)
3692

    
3693
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3694
                                               ignore_secondaries=True)
3695
      if not disks_ok:
3696
        _ShutdownInstanceDisks(self, instance)
3697
        raise errors.OpExecError("Can't activate the instance's disks")
3698

    
3699
      feedback_fn("* starting the instance on the target node")
3700
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3701
      msg = result.RemoteFailMsg()
3702
      if msg:
3703
        _ShutdownInstanceDisks(self, instance)
3704
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3705
                                 (instance.name, target_node, msg))
3706

    
3707

    
3708
class LUMigrateInstance(LogicalUnit):
3709
  """Migrate an instance.
3710

3711
  This is migration without shutting down, compared to the failover,
3712
  which is done with shutdown.
3713

3714
  """
3715
  HPATH = "instance-migrate"
3716
  HTYPE = constants.HTYPE_INSTANCE
3717
  _OP_REQP = ["instance_name", "live", "cleanup"]
3718

    
3719
  REQ_BGL = False
3720

    
3721
  def ExpandNames(self):
3722
    self._ExpandAndLockInstance()
3723
    self.needed_locks[locking.LEVEL_NODE] = []
3724
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3725

    
3726
  def DeclareLocks(self, level):
3727
    if level == locking.LEVEL_NODE:
3728
      self._LockInstancesNodes()
3729

    
3730
  def BuildHooksEnv(self):
3731
    """Build hooks env.
3732

3733
    This runs on master, primary and secondary nodes of the instance.
3734

3735
    """
3736
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3737
    env["MIGRATE_LIVE"] = self.op.live
3738
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3739
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3740
    return env, nl, nl
3741

    
3742
  def CheckPrereq(self):
3743
    """Check prerequisites.
3744

3745
    This checks that the instance is in the cluster.
3746

3747
    """
3748
    instance = self.cfg.GetInstanceInfo(
3749
      self.cfg.ExpandInstanceName(self.op.instance_name))
3750
    if instance is None:
3751
      raise errors.OpPrereqError("Instance '%s' not known" %
3752
                                 self.op.instance_name)
3753

    
3754
    if instance.disk_template != constants.DT_DRBD8:
3755
      raise errors.OpPrereqError("Instance's disk layout is not"
3756
                                 " drbd8, cannot migrate.")
3757

    
3758
    secondary_nodes = instance.secondary_nodes
3759
    if not secondary_nodes:
3760
      raise errors.ConfigurationError("No secondary node but using"
3761
                                      " drbd8 disk template")
3762

    
3763
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3764

    
3765
    target_node = secondary_nodes[0]
3766
    # check memory requirements on the secondary node
3767
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3768
                         instance.name, i_be[constants.BE_MEMORY],
3769
                         instance.hypervisor)
3770

    
3771
    # check bridge existance
3772
    brlist = [nic.bridge for nic in instance.nics]
3773
    result = self.rpc.call_bridges_exist(target_node, brlist)
3774
    if result.failed or not result.data:
3775
      raise errors.OpPrereqError("One or more target bridges %s does not"
3776
                                 " exist on destination node '%s'" %
3777
                                 (brlist, target_node))
3778

    
3779
    if not self.op.cleanup:
3780
      _CheckNodeNotDrained(self, target_node)
3781
      result = self.rpc.call_instance_migratable(instance.primary_node,
3782
                                                 instance)
3783
      msg = result.RemoteFailMsg()
3784
      if msg:
3785
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3786
                                   msg)
3787

    
3788
    self.instance = instance
3789

    
3790
  def _WaitUntilSync(self):
3791
    """Poll with custom rpc for disk sync.
3792

3793
    This uses our own step-based rpc call.
3794

3795
    """
3796
    self.feedback_fn("* wait until resync is done")
3797
    all_done = False
3798
    while not all_done:
3799
      all_done = True
3800
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3801
                                            self.nodes_ip,
3802
                                            self.instance.disks)
3803
      min_percent = 100
3804
      for node, nres in result.items():
3805
        msg = nres.RemoteFailMsg()
3806
        if msg:
3807
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3808
                                   (node, msg))
3809
        node_done, node_percent = nres.payload
3810
        all_done = all_done and node_done
3811
        if node_percent is not None:
3812
          min_percent = min(min_percent, node_percent)
3813
      if not all_done:
3814
        if min_percent < 100:
3815
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3816
        time.sleep(2)
3817

    
3818
  def _EnsureSecondary(self, node):
3819
    """Demote a node to secondary.
3820

3821
    """
3822
    self.feedback_fn("* switching node %s to secondary mode" % node)
3823

    
3824
    for dev in self.instance.disks:
3825
      self.cfg.SetDiskID(dev, node)
3826

    
3827
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3828
                                          self.instance.disks)
3829
    msg = result.RemoteFailMsg()
3830
    if msg:
3831
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3832
                               " error %s" % (node, msg))
3833

    
3834
  def _GoStandalone(self):
3835
    """Disconnect from the network.
3836

3837
    """
3838
    self.feedback_fn("* changing into standalone mode")
3839
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3840
                                               self.instance.disks)
3841
    for node, nres in result.items():
3842
      msg = nres.RemoteFailMsg()
3843
      if msg:
3844
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3845
                                 " error %s" % (node, msg))
3846

    
3847
  def _GoReconnect(self, multimaster):
3848
    """Reconnect to the network.
3849

3850
    """
3851
    if multimaster:
3852
      msg = "dual-master"
3853
    else:
3854
      msg = "single-master"
3855
    self.feedback_fn("* changing disks into %s mode" % msg)
3856
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3857
                                           self.instance.disks,
3858
                                           self.instance.name, multimaster)
3859
    for node, nres in result.items():
3860
      msg = nres.RemoteFailMsg()
3861
      if msg:
3862
        raise errors.OpExecError("Cannot change disks config on node %s,"
3863
                                 " error: %s" % (node, msg))
3864

    
3865
  def _ExecCleanup(self):
3866
    """Try to cleanup after a failed migration.
3867

3868
    The cleanup is done by:
3869
      - check that the instance is running only on one node
3870
        (and update the config if needed)
3871
      - change disks on its secondary node to secondary
3872
      - wait until disks are fully synchronized
3873
      - disconnect from the network
3874
      - change disks into single-master mode
3875
      - wait again until disks are fully synchronized
3876

3877
    """
3878
    instance = self.instance
3879
    target_node = self.target_node
3880
    source_node = self.source_node
3881

    
3882
    # check running on only one node
3883
    self.feedback_fn("* checking where the instance actually runs"
3884
                     " (if this hangs, the hypervisor might be in"
3885
                     " a bad state)")
3886
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3887
    for node, result in ins_l.items():
3888
      result.Raise()
3889
      if not isinstance(result.data, list):
3890
        raise errors.OpExecError("Can't contact node '%s'" % node)
3891

    
3892
    runningon_source = instance.name in ins_l[source_node].data
3893
    runningon_target = instance.name in ins_l[target_node].data
3894

    
3895
    if runningon_source and runningon_target:
3896
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3897
                               " or the hypervisor is confused. You will have"
3898
                               " to ensure manually that it runs only on one"
3899
                               " and restart this operation.")
3900

    
3901
    if not (runningon_source or runningon_target):
3902
      raise errors.OpExecError("Instance does not seem to be running at all."
3903
                               " In this case, it's safer to repair by"
3904
                               " running 'gnt-instance stop' to ensure disk"
3905
                               " shutdown, and then restarting it.")
3906

    
3907
    if runningon_target:
3908
      # the migration has actually succeeded, we need to update the config
3909
      self.feedback_fn("* instance running on secondary node (%s),"
3910
                       " updating config" % target_node)
3911
      instance.primary_node = target_node
3912
      self.cfg.Update(instance)
3913
      demoted_node = source_node
3914
    else:
3915
      self.feedback_fn("* instance confirmed to be running on its"
3916
                       " primary node (%s)" % source_node)
3917
      demoted_node = target_node
3918

    
3919
    self._EnsureSecondary(demoted_node)
3920
    try:
3921
      self._WaitUntilSync()
3922
    except errors.OpExecError:
3923
      # we ignore here errors, since if the device is standalone, it
3924
      # won't be able to sync
3925
      pass
3926
    self._GoStandalone()
3927
    self._GoReconnect(False)
3928
    self._WaitUntilSync()
3929

    
3930
    self.feedback_fn("* done")
3931

    
3932
  def _RevertDiskStatus(self):
3933
    """Try to revert the disk status after a failed migration.
3934

3935
    """
3936
    target_node = self.target_node
3937
    try:
3938
      self._EnsureSecondary(target_node)
3939
      self._GoStandalone()
3940
      self._GoReconnect(False)
3941
      self._WaitUntilSync()
3942
    except errors.OpExecError, err:
3943
      self.LogWarning("Migration failed and I can't reconnect the"
3944
                      " drives: error '%s'\n"
3945
                      "Please look and recover the instance status" %
3946
                      str(err))
3947

    
3948
  def _AbortMigration(self):
3949
    """Call the hypervisor code to abort a started migration.
3950

3951
    """
3952
    instance = self.instance
3953
    target_node = self.target_node
3954
    migration_info = self.migration_info
3955

    
3956
    abort_result = self.rpc.call_finalize_migration(target_node,
3957
                                                    instance,
3958
                                                    migration_info,
3959
                                                    False)
3960
    abort_msg = abort_result.RemoteFailMsg()
3961
    if abort_msg:
3962
      logging.error("Aborting migration failed on target node %s: %s" %
3963
                    (target_node, abort_msg))
3964
      # Don't raise an exception here, as we stil have to try to revert the
3965
      # disk status, even if this step failed.
3966

    
3967
  def _ExecMigration(self):
3968
    """Migrate an instance.
3969

3970
    The migrate is done by:
3971
      - change the disks into dual-master mode
3972
      - wait until disks are fully synchronized again
3973
      - migrate the instance
3974
      - change disks on the new secondary node (the old primary) to secondary
3975
      - wait until disks are fully synchronized
3976
      - change disks into single-master mode
3977

3978
    """
3979
    instance = self.instance
3980
    target_node = self.target_node
3981
    source_node = self.source_node
3982

    
3983
    self.feedback_fn("* checking disk consistency between source and target")
3984
    for dev in instance.disks:
3985
      if not _CheckDiskConsistency(self, dev, target_node, False):
3986
        raise errors.OpExecError("Disk %s is degraded or not fully"
3987
                                 " synchronized on target node,"
3988
                                 " aborting migrate." % dev.iv_name)
3989

    
3990
    # First get the migration information from the remote node
3991
    result = self.rpc.call_migration_info(source_node, instance)
3992
    msg = result.RemoteFailMsg()
3993
    if msg:
3994
      log_err = ("Failed fetching source migration information from %s: %s" %
3995
                 (source_node, msg))
3996
      logging.error(log_err)
3997
      raise errors.OpExecError(log_err)
3998

    
3999
    self.migration_info = migration_info = result.payload
4000

    
4001
    # Then switch the disks to master/master mode
4002
    self._EnsureSecondary(target_node)
4003
    self._GoStandalone()
4004
    self._GoReconnect(True)
4005
    self._WaitUntilSync()
4006

    
4007
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
4008
    result = self.rpc.call_accept_instance(target_node,
4009
                                           instance,
4010
                                           migration_info,
4011
                                           self.nodes_ip[target_node])
4012

    
4013
    msg = result.RemoteFailMsg()
4014
    if msg:
4015
      logging.error("Instance pre-migration failed, trying to revert"
4016
                    " disk status: %s", msg)
4017
      self._AbortMigration()
4018
      self._RevertDiskStatus()
4019
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
4020
                               (instance.name, msg))
4021

    
4022
    self.feedback_fn("* migrating instance to %s" % target_node)
4023
    time.sleep(10)
4024
    result = self.rpc.call_instance_migrate(source_node, instance,
4025
                                            self.nodes_ip[target_node],
4026
                                            self.op.live)
4027
    msg = result.RemoteFailMsg()
4028
    if msg:
4029
      logging.error("Instance migration failed, trying to revert"
4030
                    " disk status: %s", msg)
4031
      self._AbortMigration()
4032
      self._RevertDiskStatus()
4033
      raise errors.OpExecError("Could not migrate instance %s: %s" %
4034
                               (instance.name, msg))
4035
    time.sleep(10)
4036

    
4037
    instance.primary_node = target_node
4038
    # distribute new instance config to the other nodes
4039
    self.cfg.Update(instance)
4040

    
4041
    result = self.rpc.call_finalize_migration(target_node,
4042
                                              instance,
4043
                                              migration_info,
4044
                                              True)
4045
    msg = result.RemoteFailMsg()
4046
    if msg:
4047
      logging.error("Instance migration succeeded, but finalization failed:"
4048
                    " %s" % msg)
4049
      raise errors.OpExecError("Could not finalize instance migration: %s" %
4050
                               msg)
4051

    
4052
    self._EnsureSecondary(source_node)
4053
    self._WaitUntilSync()
4054
    self._GoStandalone()
4055
    self._GoReconnect(False)
4056
    self._WaitUntilSync()
4057

    
4058
    self.feedback_fn("* done")
4059

    
4060
  def Exec(self, feedback_fn):
4061
    """Perform the migration.
4062

4063
    """
4064
    self.feedback_fn = feedback_fn
4065

    
4066
    self.source_node = self.instance.primary_node
4067
    self.target_node = self.instance.secondary_nodes[0]
4068
    self.all_nodes = [self.source_node, self.target_node]
4069
    self.nodes_ip = {
4070
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
4071
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
4072
      }
4073
    if self.op.cleanup:
4074
      return self._ExecCleanup()
4075
    else:
4076
      return self._ExecMigration()
4077

    
4078

    
4079
def _CreateBlockDev(lu, node, instance, device, force_create,
4080
                    info, force_open):
4081
  """Create a tree of block devices on a given node.
4082

4083
  If this device type has to be created on secondaries, create it and
4084
  all its children.
4085

4086
  If not, just recurse to children keeping the same 'force' value.
4087

4088
  @param lu: the lu on whose behalf we execute
4089
  @param node: the node on which to create the device
4090
  @type instance: L{objects.Instance}
4091
  @param instance: the instance which owns the device
4092
  @type device: L{objects.Disk}
4093
  @param device: the device to create
4094
  @type force_create: boolean
4095
  @param force_create: whether to force creation of this device; this
4096
      will be change to True whenever we find a device which has
4097
      CreateOnSecondary() attribute
4098
  @param info: the extra 'metadata' we should attach to the device
4099
      (this will be represented as a LVM tag)
4100
  @type force_open: boolean
4101
  @param force_open: this parameter will be passes to the
4102
      L{backend.BlockdevCreate} function where it specifies
4103
      whether we run on primary or not, and it affects both
4104
      the child assembly and the device own Open() execution
4105

4106
  """
4107
  if device.CreateOnSecondary():
4108
    force_create = True
4109

    
4110
  if device.children:
4111
    for child in device.children:
4112
      _CreateBlockDev(lu, node, instance, child, force_create,
4113
                      info, force_open)
4114

    
4115
  if not force_create:
4116
    return
4117

    
4118
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
4119

    
4120

    
4121
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
4122
  """Create a single block device on a given node.
4123

4124
  This will not recurse over children of the device, so they must be
4125
  created in advance.
4126

4127
  @param lu: the lu on whose behalf we execute
4128
  @param node: the node on which to create the device
4129
  @type instance: L{objects.Instance}
4130
  @param instance: the instance which owns the device
4131
  @type device: L{objects.Disk}
4132
  @param device: the device to create
4133
  @param info: the extra 'metadata' we should attach to the device
4134
      (this will be represented as a LVM tag)
4135
  @type force_open: boolean
4136
  @param force_open: this parameter will be passes to the
4137
      L{backend.BlockdevCreate} function where it specifies
4138
      whether we run on primary or not, and it affects both
4139
      the child assembly and the device own Open() execution
4140

4141
  """
4142
  lu.cfg.SetDiskID(device, node)
4143
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4144
                                       instance.name, force_open, info)
4145
  msg = result.RemoteFailMsg()
4146
  if msg:
4147
    raise errors.OpExecError("Can't create block device %s on"
4148
                             " node %s for instance %s: %s" %
4149
                             (device, node, instance.name, msg))
4150
  if device.physical_id is None:
4151
    device.physical_id = result.payload
4152

    
4153

    
4154
def _GenerateUniqueNames(lu, exts):
4155
  """Generate a suitable LV name.
4156

4157
  This will generate a logical volume name for the given instance.
4158

4159
  """
4160
  results = []
4161
  for val in exts:
4162
    new_id = lu.cfg.GenerateUniqueID()
4163
    results.append("%s%s" % (new_id, val))
4164
  return results
4165

    
4166

    
4167
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4168
                         p_minor, s_minor):
4169
  """Generate a drbd8 device complete with its children.
4170

4171
  """
4172
  port = lu.cfg.AllocatePort()
4173
  vgname = lu.cfg.GetVGName()
4174
  shared_secret = lu.cfg.GenerateDRBDSecret()
4175
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4176
                          logical_id=(vgname, names[0]))
4177
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4178
                          logical_id=(vgname, names[1]))
4179
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4180
                          logical_id=(primary, secondary, port,
4181
                                      p_minor, s_minor,
4182
                                      shared_secret),
4183
                          children=[dev_data, dev_meta],
4184
                          iv_name=iv_name)
4185
  return drbd_dev
4186

    
4187

    
4188
def _GenerateDiskTemplate(lu, template_name,
4189
                          instance_name, primary_node,
4190
                          secondary_nodes, disk_info,
4191
                          file_storage_dir, file_driver,
4192
                          base_index):
4193
  """Generate the entire disk layout for a given template type.
4194

4195
  """
4196
  #TODO: compute space requirements
4197

    
4198
  vgname = lu.cfg.GetVGName()
4199
  disk_count = len(disk_info)
4200
  disks = []
4201
  if template_name == constants.DT_DISKLESS:
4202
    pass
4203
  elif template_name == constants.DT_PLAIN:
4204
    if len(secondary_nodes) != 0:
4205
      raise errors.ProgrammerError("Wrong template configuration")
4206

    
4207
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4208
                                      for i in range(disk_count)])
4209
    for idx, disk in enumerate(disk_info):
4210
      disk_index = idx + base_index
4211
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4212
                              logical_id=(vgname, names[idx]),
4213
                              iv_name="disk/%d" % disk_index,
4214
                              mode=disk["mode"])
4215
      disks.append(disk_dev)
4216
  elif template_name == constants.DT_DRBD8:
4217
    if len(secondary_nodes) != 1:
4218
      raise errors.ProgrammerError("Wrong template configuration")
4219
    remote_node = secondary_nodes[0]
4220
    minors = lu.cfg.AllocateDRBDMinor(
4221
      [primary_node, remote_node] * len(disk_info), instance_name)
4222

    
4223
    names = []
4224
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4225
                                               for i in range(disk_count)]):
4226
      names.append(lv_prefix + "_data")
4227
      names.append(lv_prefix + "_meta")
4228
    for idx, disk in enumerate(disk_info):
4229
      disk_index = idx + base_index
4230
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4231
                                      disk["size"], names[idx*2:idx*2+2],
4232
                                      "disk/%d" % disk_index,
4233
                                      minors[idx*2], minors[idx*2+1])
4234
      disk_dev.mode = disk["mode"]
4235
      disks.append(disk_dev)
4236
  elif template_name == constants.DT_FILE:
4237
    if len(secondary_nodes) != 0:
4238
      raise errors.ProgrammerError("Wrong template configuration")
4239

    
4240
    for idx, disk in enumerate(disk_info):
4241
      disk_index = idx + base_index
4242
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4243
                              iv_name="disk/%d" % disk_index,
4244
                              logical_id=(file_driver,
4245
                                          "%s/disk%d" % (file_storage_dir,
4246
                                                         disk_index)),
4247
                              mode=disk["mode"])
4248
      disks.append(disk_dev)
4249
  else:
4250
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4251
  return disks
4252

    
4253

    
4254
def _GetInstanceInfoText(instance):
4255
  """Compute that text that should be added to the disk's metadata.
4256

4257
  """
4258
  return "originstname+%s" % instance.name
4259

    
4260

    
4261
def _CreateDisks(lu, instance):
4262
  """Create all disks for an instance.
4263

4264
  This abstracts away some work from AddInstance.
4265

4266
  @type lu: L{LogicalUnit}
4267
  @param lu: the logical unit on whose behalf we execute
4268
  @type instance: L{objects.Instance}
4269
  @param instance: the instance whose disks we should create
4270
  @rtype: boolean
4271
  @return: the success of the creation
4272

4273
  """
4274
  info = _GetInstanceInfoText(instance)
4275
  pnode = instance.primary_node
4276

    
4277
  if instance.disk_template == constants.DT_FILE:
4278
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4279
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4280

    
4281
    if result.failed or not result.data:
4282
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4283

    
4284
    if not result.data[0]:
4285
      raise errors.OpExecError("Failed to create directory '%s'" %
4286
                               file_storage_dir)
4287

    
4288
  # Note: this needs to be kept in sync with adding of disks in
4289
  # LUSetInstanceParams
4290
  for device in instance.disks:
4291
    logging.info("Creating volume %s for instance %s",
4292
                 device.iv_name, instance.name)
4293
    #HARDCODE
4294
    for node in instance.all_nodes:
4295
      f_create = node == pnode
4296
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4297

    
4298

    
4299
def _RemoveDisks(lu, instance):
4300
  """Remove all disks for an instance.
4301

4302
  This abstracts away some work from `AddInstance()` and
4303
  `RemoveInstance()`. Note that in case some of the devices couldn't
4304
  be removed, the removal will continue with the other ones (compare
4305
  with `_CreateDisks()`).
4306

4307
  @type lu: L{LogicalUnit}
4308
  @param lu: the logical unit on whose behalf we execute
4309
  @type instance: L{objects.Instance}
4310
  @param instance: the instance whose disks we should remove
4311
  @rtype: boolean
4312
  @return: the success of the removal
4313

4314
  """
4315
  logging.info("Removing block devices for instance %s", instance.name)
4316

    
4317
  all_result = True
4318
  for device in instance.disks:
4319
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4320
      lu.cfg.SetDiskID(disk, node)
4321
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4322
      if msg:
4323
        lu.LogWarning("Could not remove block device %s on node %s,"
4324
                      " continuing anyway: %s", device.iv_name, node, msg)
4325
        all_result = False
4326

    
4327
  if instance.disk_template == constants.DT_FILE:
4328
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4329
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4330
                                                 file_storage_dir)
4331
    if result.failed or not result.data:
4332
      logging.error("Could not remove directory '%s'", file_storage_dir)
4333
      all_result = False
4334

    
4335
  return all_result
4336

    
4337

    
4338
def _ComputeDiskSize(disk_template, disks):
4339
  """Compute disk size requirements in the volume group
4340

4341
  """
4342
  # Required free disk space as a function of disk and swap space
4343
  req_size_dict = {
4344
    constants.DT_DISKLESS: None,
4345
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4346
    # 128 MB are added for drbd metadata for each disk
4347
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4348
    constants.DT_FILE: None,
4349
  }
4350

    
4351
  if disk_template not in req_size_dict:
4352
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4353
                                 " is unknown" %  disk_template)
4354

    
4355
  return req_size_dict[disk_template]
4356

    
4357

    
4358
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4359
  """Hypervisor parameter validation.
4360

4361
  This function abstract the hypervisor parameter validation to be
4362
  used in both instance create and instance modify.
4363

4364
  @type lu: L{LogicalUnit}
4365
  @param lu: the logical unit for which we check
4366
  @type nodenames: list
4367
  @param nodenames: the list of nodes on which we should check
4368
  @type hvname: string
4369
  @param hvname: the name of the hypervisor we should use
4370
  @type hvparams: dict
4371
  @param hvparams: the parameters which we need to check
4372
  @raise errors.OpPrereqError: if the parameters are not valid
4373

4374
  """
4375
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4376
                                                  hvname,
4377
                                                  hvparams)
4378
  for node in nodenames:
4379
    info = hvinfo[node]
4380
    if info.offline:
4381
      continue
4382
    msg = info.RemoteFailMsg()
4383
    if msg:
4384
      raise errors.OpPrereqError("Hypervisor parameter validation"
4385
                                 " failed on node %s: %s" % (node, msg))
4386

    
4387

    
4388
class LUCreateInstance(LogicalUnit):
4389
  """Create an instance.
4390

4391
  """
4392
  HPATH = "instance-add"
4393
  HTYPE = constants.HTYPE_INSTANCE
4394
  _OP_REQP = ["instance_name", "disks", "disk_template",
4395
              "mode", "start",
4396
              "wait_for_sync", "ip_check", "nics",
4397
              "hvparams", "beparams"]
4398
  REQ_BGL = False
4399

    
4400
  def _ExpandNode(self, node):
4401
    """Expands and checks one node name.
4402

4403
    """
4404
    node_full = self.cfg.ExpandNodeName(node)
4405
    if node_full is None:
4406
      raise errors.OpPrereqError("Unknown node %s" % node)
4407
    return node_full
4408

    
4409
  def ExpandNames(self):
4410
    """ExpandNames for CreateInstance.
4411

4412
    Figure out the right locks for instance creation.
4413

4414
    """
4415
    self.needed_locks = {}
4416

    
4417
    # set optional parameters to none if they don't exist
4418
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4419
      if not hasattr(self.op, attr):
4420
        setattr(self.op, attr, None)
4421

    
4422
    # cheap checks, mostly valid constants given
4423

    
4424
    # verify creation mode
4425
    if self.op.mode not in (constants.INSTANCE_CREATE,
4426
                            constants.INSTANCE_IMPORT):
4427
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4428
                                 self.op.mode)
4429

    
4430
    # disk template and mirror node verification
4431
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4432
      raise errors.OpPrereqError("Invalid disk template name")
4433

    
4434
    if self.op.hypervisor is None:
4435
      self.op.hypervisor = self.cfg.GetHypervisorType()
4436

    
4437
    cluster = self.cfg.GetClusterInfo()
4438
    enabled_hvs = cluster.enabled_hypervisors
4439
    if self.op.hypervisor not in enabled_hvs:
4440
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4441
                                 " cluster (%s)" % (self.op.hypervisor,
4442
                                  ",".join(enabled_hvs)))
4443

    
4444
    # check hypervisor parameter syntax (locally)
4445
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4446
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4447
                                  self.op.hvparams)
4448
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4449
    hv_type.CheckParameterSyntax(filled_hvp)
4450
    self.hv_full = filled_hvp
4451

    
4452
    # fill and remember the beparams dict
4453
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4454
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4455
                                    self.op.beparams)
4456

    
4457
    #### instance parameters check
4458

    
4459
    # instance name verification
4460
    hostname1 = utils.HostInfo(self.op.instance_name)
4461
    self.op.instance_name = instance_name = hostname1.name
4462

    
4463
    # this is just a preventive check, but someone might still add this
4464
    # instance in the meantime, and creation will fail at lock-add time
4465
    if instance_name in self.cfg.GetInstanceList():
4466
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4467
                                 instance_name)
4468

    
4469
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4470

    
4471
    # NIC buildup
4472
    self.nics = []
4473
    for nic in self.op.nics:
4474
      # ip validity checks
4475
      ip = nic.get("ip", None)
4476
      if ip is None or ip.lower() == "none":
4477
        nic_ip = None
4478
      elif ip.lower() == constants.VALUE_AUTO:
4479
        nic_ip = hostname1.ip
4480
      else:
4481
        if not utils.IsValidIP(ip):
4482
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4483
                                     " like a valid IP" % ip)
4484
        nic_ip = ip
4485

    
4486
      # MAC address verification
4487
      mac = nic.get("mac", constants.VALUE_AUTO)
4488
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4489
        if not utils.IsValidMac(mac.lower()):
4490
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4491
                                     mac)
4492
      # bridge verification
4493
      bridge = nic.get("bridge", None)
4494
      if bridge is None:
4495
        bridge = self.cfg.GetDefBridge()
4496
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4497

    
4498
    # disk checks/pre-build
4499
    self.disks = []
4500
    for disk in self.op.disks:
4501
      mode = disk.get("mode", constants.DISK_RDWR)
4502
      if mode not in constants.DISK_ACCESS_SET:
4503
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4504
                                   mode)
4505
      size = disk.get("size", None)
4506
      if size is None:
4507
        raise errors.OpPrereqError("Missing disk size")
4508
      try:
4509
        size = int(size)
4510
      except ValueError:
4511
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4512
      self.disks.append({"size": size, "mode": mode})
4513

    
4514
    # used in CheckPrereq for ip ping check
4515
    self.check_ip = hostname1.ip
4516

    
4517
    # file storage checks
4518
    if (self.op.file_driver and
4519
        not self.op.file_driver in constants.FILE_DRIVER):
4520
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4521
                                 self.op.file_driver)
4522

    
4523
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4524
      raise errors.OpPrereqError("File storage directory path not absolute")
4525

    
4526
    ### Node/iallocator related checks
4527
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4528
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4529
                                 " node must be given")
4530

    
4531
    if self.op.iallocator:
4532
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4533
    else:
4534
      self.op.pnode = self._ExpandNode(self.op.pnode)
4535
      nodelist = [self.op.pnode]
4536
      if self.op.snode is not None:
4537
        self.op.snode = self._ExpandNode(self.op.snode)
4538
        nodelist.append(self.op.snode)
4539
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4540

    
4541
    # in case of import lock the source node too
4542
    if self.op.mode == constants.INSTANCE_IMPORT:
4543
      src_node = getattr(self.op, "src_node", None)
4544
      src_path = getattr(self.op, "src_path", None)
4545

    
4546
      if src_path is None:
4547
        self.op.src_path = src_path = self.op.instance_name
4548

    
4549
      if src_node is None:
4550
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4551
        self.op.src_node = None
4552
        if os.path.isabs(src_path):
4553
          raise errors.OpPrereqError("Importing an instance from an absolute"
4554
                                     " path requires a source node option.")
4555
      else:
4556
        self.op.src_node = src_node = self._ExpandNode(src_node)
4557
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4558
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4559
        if not os.path.isabs(src_path):
4560
          self.op.src_path = src_path = \
4561
            os.path.join(constants.EXPORT_DIR, src_path)
4562

    
4563
    else: # INSTANCE_CREATE
4564
      if getattr(self.op, "os_type", None) is None:
4565
        raise errors.OpPrereqError("No guest OS specified")
4566

    
4567
  def _RunAllocator(self):
4568
    """Run the allocator based on input opcode.
4569

4570
    """
4571
    nics = [n.ToDict() for n in self.nics]
4572
    ial = IAllocator(self,
4573
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4574
                     name=self.op.instance_name,
4575
                     disk_template=self.op.disk_template,
4576
                     tags=[],
4577
                     os=self.op.os_type,
4578
                     vcpus=self.be_full[constants.BE_VCPUS],
4579
                     mem_size=self.be_full[constants.BE_MEMORY],
4580
                     disks=self.disks,
4581
                     nics=nics,
4582
                     hypervisor=self.op.hypervisor,
4583
                     )
4584

    
4585
    ial.Run(self.op.iallocator)
4586

    
4587
    if not ial.success:
4588
      raise errors.OpPrereqError("Can't compute nodes using"
4589
                                 " iallocator '%s': %s" % (self.op.iallocator,
4590
                                                           ial.info))
4591
    if len(ial.nodes) != ial.required_nodes:
4592
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4593
                                 " of nodes (%s), required %s" %
4594
                                 (self.op.iallocator, len(ial.nodes),
4595
                                  ial.required_nodes))
4596
    self.op.pnode = ial.nodes[0]
4597
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4598
                 self.op.instance_name, self.op.iallocator,
4599
                 ", ".join(ial.nodes))
4600
    if ial.required_nodes == 2:
4601
      self.op.snode = ial.nodes[1]
4602

    
4603
  def BuildHooksEnv(self):
4604
    """Build hooks env.
4605

4606
    This runs on master, primary and secondary nodes of the instance.
4607

4608
    """
4609
    env = {
4610
      "ADD_MODE": self.op.mode,
4611
      }
4612
    if self.op.mode == constants.INSTANCE_IMPORT:
4613
      env["SRC_NODE"] = self.op.src_node
4614
      env["SRC_PATH"] = self.op.src_path
4615
      env["SRC_IMAGES"] = self.src_images
4616

    
4617
    env.update(_BuildInstanceHookEnv(
4618
      name=self.op.instance_name,
4619
      primary_node=self.op.pnode,
4620
      secondary_nodes=self.secondaries,
4621
      status=self.op.start,
4622
      os_type=self.op.os_type,
4623
      memory=self.be_full[constants.BE_MEMORY],
4624
      vcpus=self.be_full[constants.BE_VCPUS],
4625
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4626
      disk_template=self.op.disk_template,
4627
      disks=[(d["size"], d["mode"]) for d in self.disks],
4628
      bep=self.be_full,
4629
      hvp=self.hv_full,
4630
      hypervisor=self.op.hypervisor,
4631
    ))
4632

    
4633
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4634
          self.secondaries)
4635
    return env, nl, nl
4636

    
4637

    
4638
  def CheckPrereq(self):
4639
    """Check prerequisites.
4640

4641
    """
4642
    if (not self.cfg.GetVGName() and
4643
        self.op.disk_template not in constants.DTS_NOT_LVM):
4644
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4645
                                 " instances")
4646

    
4647
    if self.op.mode == constants.INSTANCE_IMPORT:
4648
      src_node = self.op.src_node
4649
      src_path = self.op.src_path
4650

    
4651
      if src_node is None:
4652
        exp_list = self.rpc.call_export_list(
4653
          self.acquired_locks[locking.LEVEL_NODE])
4654
        found = False
4655
        for node in exp_list:
4656
          if not exp_list[node].failed and src_path in exp_list[node].data:
4657
            found = True
4658
            self.op.src_node = src_node = node
4659
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4660
                                                       src_path)
4661
            break
4662
        if not found:
4663
          raise errors.OpPrereqError("No export found for relative path %s" %
4664
                                      src_path)
4665

    
4666
      _CheckNodeOnline(self, src_node)
4667
      result = self.rpc.call_export_info(src_node, src_path)
4668
      result.Raise()
4669
      if not result.data:
4670
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4671

    
4672
      export_info = result.data
4673
      if not export_info.has_section(constants.INISECT_EXP):
4674
        raise errors.ProgrammerError("Corrupted export config")
4675

    
4676
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4677
      if (int(ei_version) != constants.EXPORT_VERSION):
4678
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4679
                                   (ei_version, constants.EXPORT_VERSION))
4680

    
4681
      # Check that the new instance doesn't have less disks than the export
4682
      instance_disks = len(self.disks)
4683
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4684
      if instance_disks < export_disks:
4685
        raise errors.OpPrereqError("Not enough disks to import."
4686
                                   " (instance: %d, export: %d)" %
4687
                                   (instance_disks, export_disks))
4688

    
4689
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4690
      disk_images = []
4691
      for idx in range(export_disks):
4692
        option = 'disk%d_dump' % idx
4693
        if export_info.has_option(constants.INISECT_INS, option):
4694
          # FIXME: are the old os-es, disk sizes, etc. useful?
4695
          export_name = export_info.get(constants.INISECT_INS, option)
4696
          image = os.path.join(src_path, export_name)
4697
          disk_images.append(image)
4698
        else:
4699
          disk_images.append(False)
4700

    
4701
      self.src_images = disk_images
4702

    
4703
      old_name = export_info.get(constants.INISECT_INS, 'name')
4704
      # FIXME: int() here could throw a ValueError on broken exports
4705
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4706
      if self.op.instance_name == old_name:
4707
        for idx, nic in enumerate(self.nics):
4708
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4709
            nic_mac_ini = 'nic%d_mac' % idx
4710
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4711

    
4712
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4713
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4714
    if self.op.start and not self.op.ip_check:
4715
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4716
                                 " adding an instance in start mode")
4717

    
4718
    if self.op.ip_check:
4719
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4720
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4721
                                   (self.check_ip, self.op.instance_name))
4722

    
4723
    #### mac address generation
4724
    # By generating here the mac address both the allocator and the hooks get
4725
    # the real final mac address rather than the 'auto' or 'generate' value.
4726
    # There is a race condition between the generation and the instance object
4727
    # creation, which means that we know the mac is valid now, but we're not
4728
    # sure it will be when we actually add the instance. If things go bad
4729
    # adding the instance will abort because of a duplicate mac, and the
4730
    # creation job will fail.
4731
    for nic in self.nics:
4732
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4733
        nic.mac = self.cfg.GenerateMAC()
4734

    
4735
    #### allocator run
4736

    
4737
    if self.op.iallocator is not None:
4738
      self._RunAllocator()
4739

    
4740
    #### node related checks
4741

    
4742
    # check primary node
4743
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4744
    assert self.pnode is not None, \
4745
      "Cannot retrieve locked node %s" % self.op.pnode
4746
    if pnode.offline:
4747
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4748
                                 pnode.name)
4749
    if pnode.drained:
4750
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4751
                                 pnode.name)
4752

    
4753
    self.secondaries = []
4754

    
4755
    # mirror node verification
4756
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4757
      if self.op.snode is None:
4758
        raise errors.OpPrereqError("The networked disk templates need"
4759
                                   " a mirror node")
4760
      if self.op.snode == pnode.name:
4761
        raise errors.OpPrereqError("The secondary node cannot be"
4762
                                   " the primary node.")
4763
      _CheckNodeOnline(self, self.op.snode)
4764
      _CheckNodeNotDrained(self, self.op.snode)
4765
      self.secondaries.append(self.op.snode)
4766

    
4767
    nodenames = [pnode.name] + self.secondaries
4768

    
4769
    req_size = _ComputeDiskSize(self.op.disk_template,
4770
                                self.disks)
4771

    
4772
    # Check lv size requirements
4773
    if req_size is not None:
4774
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4775
                                         self.op.hypervisor)
4776
      for node in nodenames:
4777
        info = nodeinfo[node]
4778
        info.Raise()
4779
        info = info.data
4780
        if not info:
4781
          raise errors.OpPrereqError("Cannot get current information"
4782
                                     " from node '%s'" % node)
4783
        vg_free = info.get('vg_free', None)
4784
        if not isinstance(vg_free, int):
4785
          raise errors.OpPrereqError("Can't compute free disk space on"
4786
                                     " node %s" % node)
4787
        if req_size > info['vg_free']:
4788
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4789
                                     " %d MB available, %d MB required" %
4790
                                     (node, info['vg_free'], req_size))
4791

    
4792
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4793

    
4794
    # os verification
4795
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4796
    result.Raise()
4797
    if not isinstance(result.data, objects.OS) or not result.data:
4798
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4799
                                 " primary node"  % self.op.os_type)
4800

    
4801
    # bridge check on primary node
4802
    bridges = [n.bridge for n in self.nics]
4803
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4804
    result.Raise()
4805
    if not result.data:
4806
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4807
                                 " exist on destination node '%s'" %
4808
                                 (",".join(bridges), pnode.name))
4809

    
4810
    # memory check on primary node
4811
    if self.op.start:
4812
      _CheckNodeFreeMemory(self, self.pnode.name,
4813
                           "creating instance %s" % self.op.instance_name,
4814
                           self.be_full[constants.BE_MEMORY],
4815
                           self.op.hypervisor)
4816

    
4817
  def Exec(self, feedback_fn):
4818
    """Create and add the instance to the cluster.
4819

4820
    """
4821
    instance = self.op.instance_name
4822
    pnode_name = self.pnode.name
4823

    
4824
    ht_kind = self.op.hypervisor
4825
    if ht_kind in constants.HTS_REQ_PORT:
4826
      network_port = self.cfg.AllocatePort()
4827
    else:
4828
      network_port = None
4829

    
4830
    ##if self.op.vnc_bind_address is None:
4831
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4832

    
4833
    # this is needed because os.path.join does not accept None arguments
4834
    if self.op.file_storage_dir is None:
4835
      string_file_storage_dir = ""
4836
    else:
4837
      string_file_storage_dir = self.op.file_storage_dir
4838

    
4839
    # build the full file storage dir path
4840
    file_storage_dir = os.path.normpath(os.path.join(
4841
                                        self.cfg.GetFileStorageDir(),
4842
                                        string_file_storage_dir, instance))
4843

    
4844

    
4845
    disks = _GenerateDiskTemplate(self,
4846
                                  self.op.disk_template,
4847
                                  instance, pnode_name,
4848
                                  self.secondaries,
4849
                                  self.disks,
4850
                                  file_storage_dir,
4851
                                  self.op.file_driver,
4852
                                  0)
4853

    
4854
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4855
                            primary_node=pnode_name,
4856
                            nics=self.nics, disks=disks,
4857
                            disk_template=self.op.disk_template,
4858
                            admin_up=False,
4859
                            network_port=network_port,
4860
                            beparams=self.op.beparams,
4861
                            hvparams=self.op.hvparams,
4862
                            hypervisor=self.op.hypervisor,
4863
                            )
4864

    
4865
    feedback_fn("* creating instance disks...")
4866
    try:
4867
      _CreateDisks(self, iobj)
4868
    except errors.OpExecError:
4869
      self.LogWarning("Device creation failed, reverting...")
4870
      try:
4871
        _RemoveDisks(self, iobj)
4872
      finally:
4873
        self.cfg.ReleaseDRBDMinors(instance)
4874
        raise
4875

    
4876
    feedback_fn("adding instance %s to cluster config" % instance)
4877

    
4878
    self.cfg.AddInstance(iobj)
4879
    # Declare that we don't want to remove the instance lock anymore, as we've
4880
    # added the instance to the config
4881
    del self.remove_locks[locking.LEVEL_INSTANCE]
4882
    # Unlock all the nodes
4883
    if self.op.mode == constants.INSTANCE_IMPORT:
4884
      nodes_keep = [self.op.src_node]
4885
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4886
                       if node != self.op.src_node]
4887
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4888
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4889
    else:
4890
      self.context.glm.release(locking.LEVEL_NODE)
4891
      del self.acquired_locks[locking.LEVEL_NODE]
4892

    
4893
    if self.op.wait_for_sync:
4894
      disk_abort = not _WaitForSync(self, iobj)
4895
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4896
      # make sure the disks are not degraded (still sync-ing is ok)
4897
      time.sleep(15)
4898
      feedback_fn("* checking mirrors status")
4899
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4900
    else:
4901
      disk_abort = False
4902

    
4903
    if disk_abort:
4904
      _RemoveDisks(self, iobj)
4905
      self.cfg.RemoveInstance(iobj.name)
4906
      # Make sure the instance lock gets removed
4907
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4908
      raise errors.OpExecError("There are some degraded disks for"
4909
                               " this instance")
4910

    
4911
    feedback_fn("creating os for instance %s on node %s" %
4912
                (instance, pnode_name))
4913

    
4914
    if iobj.disk_template != constants.DT_DISKLESS:
4915
      if self.op.mode == constants.INSTANCE_CREATE:
4916
        feedback_fn("* running the instance OS create scripts...")
4917
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4918
        msg = result.RemoteFailMsg()
4919
        if msg:
4920
          raise errors.OpExecError("Could not add os for instance %s"
4921
                                   " on node %s: %s" %
4922
                                   (instance, pnode_name, msg))
4923

    
4924
      elif self.op.mode == constants.INSTANCE_IMPORT:
4925
        feedback_fn("* running the instance OS import scripts...")
4926
        src_node = self.op.src_node
4927
        src_images = self.src_images
4928
        cluster_name = self.cfg.GetClusterName()
4929
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4930
                                                         src_node, src_images,
4931
                                                         cluster_name)
4932
        import_result.Raise()
4933
        for idx, result in enumerate(import_result.data):
4934
          if not result:
4935
            self.LogWarning("Could not import the image %s for instance"
4936
                            " %s, disk %d, on node %s" %
4937
                            (src_images[idx], instance, idx, pnode_name))
4938
      else:
4939
        # also checked in the prereq part
4940
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4941
                                     % self.op.mode)
4942

    
4943
    if self.op.start:
4944
      iobj.admin_up = True
4945
      self.cfg.Update(iobj)
4946
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4947
      feedback_fn("* starting instance...")
4948
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
4949
      msg = result.RemoteFailMsg()
4950
      if msg:
4951
        raise errors.OpExecError("Could not start instance: %s" % msg)
4952

    
4953

    
4954
class LUConnectConsole(NoHooksLU):
4955
  """Connect to an instance's console.
4956

4957
  This is somewhat special in that it returns the command line that
4958
  you need to run on the master node in order to connect to the
4959
  console.
4960

4961
  """
4962
  _OP_REQP = ["instance_name"]
4963
  REQ_BGL = False
4964

    
4965
  def ExpandNames(self):
4966
    self._ExpandAndLockInstance()
4967

    
4968
  def CheckPrereq(self):
4969
    """Check prerequisites.
4970

4971
    This checks that the instance is in the cluster.
4972

4973
    """
4974
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4975
    assert self.instance is not None, \
4976
      "Cannot retrieve locked instance %s" % self.op.instance_name
4977
    _CheckNodeOnline(self, self.instance.primary_node)
4978

    
4979
  def Exec(self, feedback_fn):
4980
    """Connect to the console of an instance
4981

4982
    """
4983
    instance = self.instance
4984
    node = instance.primary_node
4985

    
4986
    node_insts = self.rpc.call_instance_list([node],
4987
                                             [instance.hypervisor])[node]
4988
    node_insts.Raise()
4989

    
4990
    if instance.name not in node_insts.data:
4991
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4992

    
4993
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4994

    
4995
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4996
    cluster = self.cfg.GetClusterInfo()
4997
    # beparams and hvparams are passed separately, to avoid editing the
4998
    # instance and then saving the defaults in the instance itself.
4999
    hvparams = cluster.FillHV(instance)
5000
    beparams = cluster.FillBE(instance)
5001
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
5002

    
5003
    # build ssh cmdline
5004
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
5005

    
5006

    
5007
class LUReplaceDisks(LogicalUnit):
5008
  """Replace the disks of an instance.
5009

5010
  """
5011
  HPATH = "mirrors-replace"
5012
  HTYPE = constants.HTYPE_INSTANCE
5013
  _OP_REQP = ["instance_name", "mode", "disks"]
5014
  REQ_BGL = False
5015

    
5016
  def CheckArguments(self):
5017
    if not hasattr(self.op, "remote_node"):
5018
      self.op.remote_node = None
5019
    if not hasattr(self.op, "iallocator"):
5020
      self.op.iallocator = None
5021

    
5022
    # check for valid parameter combination
5023
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
5024
    if self.op.mode == constants.REPLACE_DISK_CHG:
5025
      if cnt == 2:
5026
        raise errors.OpPrereqError("When changing the secondary either an"
5027
                                   " iallocator script must be used or the"
5028
                                   " new node given")
5029
      elif cnt == 0:
5030
        raise errors.OpPrereqError("Give either the iallocator or the new"
5031
                                   " secondary, not both")
5032
    else: # not replacing the secondary
5033
      if cnt != 2:
5034
        raise errors.OpPrereqError("The iallocator and new node options can"
5035
                                   " be used only when changing the"
5036
                                   " secondary node")
5037

    
5038
  def ExpandNames(self):
5039
    self._ExpandAndLockInstance()
5040

    
5041
    if self.op.iallocator is not None:
5042
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5043
    elif self.op.remote_node is not None:
5044
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
5045
      if remote_node is None:
5046
        raise errors.OpPrereqError("Node '%s' not known" %
5047
                                   self.op.remote_node)
5048
      self.op.remote_node = remote_node
5049
      # Warning: do not remove the locking of the new secondary here
5050
      # unless DRBD8.AddChildren is changed to work in parallel;
5051
      # currently it doesn't since parallel invocations of
5052
      # FindUnusedMinor will conflict
5053
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
5054
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5055
    else:
5056
      self.needed_locks[locking.LEVEL_NODE] = []
5057
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5058

    
5059
  def DeclareLocks(self, level):
5060
    # If we're not already locking all nodes in the set we have to declare the
5061
    # instance's primary/secondary nodes.
5062
    if (level == locking.LEVEL_NODE and
5063
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
5064
      self._LockInstancesNodes()
5065

    
5066
  def _RunAllocator(self):
5067
    """Compute a new secondary node using an IAllocator.
5068

5069
    """
5070
    ial = IAllocator(self,
5071
                     mode=constants.IALLOCATOR_MODE_RELOC,
5072
                     name=self.op.instance_name,
5073
                     relocate_from=[self.sec_node])
5074

    
5075
    ial.Run(self.op.iallocator)
5076

    
5077
    if not ial.success:
5078
      raise errors.OpPrereqError("Can't compute nodes using"
5079
                                 " iallocator '%s': %s" % (self.op.iallocator,
5080
                                                           ial.info))
5081
    if len(ial.nodes) != ial.required_nodes:
5082
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
5083
                                 " of nodes (%s), required %s" %
5084
                                 (len(ial.nodes), ial.required_nodes))
5085
    self.op.remote_node = ial.nodes[0]
5086
    self.LogInfo("Selected new secondary for the instance: %s",
5087
                 self.op.remote_node)
5088

    
5089
  def BuildHooksEnv(self):
5090
    """Build hooks env.
5091

5092
    This runs on the master, the primary and all the secondaries.
5093

5094
    """
5095
    env = {
5096
      "MODE": self.op.mode,
5097
      "NEW_SECONDARY": self.op.remote_node,
5098
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
5099
      }
5100
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5101
    nl = [
5102
      self.cfg.GetMasterNode(),
5103
      self.instance.primary_node,
5104
      ]
5105
    if self.op.remote_node is not None:
5106
      nl.append(self.op.remote_node)
5107
    return env, nl, nl
5108

    
5109
  def CheckPrereq(self):
5110
    """Check prerequisites.
5111

5112
    This checks that the instance is in the cluster.
5113

5114
    """
5115
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5116
    assert instance is not None, \
5117
      "Cannot retrieve locked instance %s" % self.op.instance_name
5118
    self.instance = instance
5119

    
5120
    if instance.disk_template != constants.DT_DRBD8:
5121
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
5122
                                 " instances")
5123

    
5124
    if len(instance.secondary_nodes) != 1:
5125
      raise errors.OpPrereqError("The instance has a strange layout,"
5126
                                 " expected one secondary but found %d" %
5127
                                 len(instance.secondary_nodes))
5128

    
5129
    self.sec_node = instance.secondary_nodes[0]
5130

    
5131
    if self.op.iallocator is not None:
5132
      self._RunAllocator()
5133

    
5134
    remote_node = self.op.remote_node
5135
    if remote_node is not None:
5136
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
5137
      assert self.remote_node_info is not None, \
5138
        "Cannot retrieve locked node %s" % remote_node
5139
    else:
5140
      self.remote_node_info = None
5141
    if remote_node == instance.primary_node:
5142
      raise errors.OpPrereqError("The specified node is the primary node of"
5143
                                 " the instance.")
5144
    elif remote_node == self.sec_node:
5145
      raise errors.OpPrereqError("The specified node is already the"
5146
                                 " secondary node of the instance.")
5147

    
5148
    if self.op.mode == constants.REPLACE_DISK_PRI:
5149
      n1 = self.tgt_node = instance.primary_node
5150
      n2 = self.oth_node = self.sec_node
5151
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5152
      n1 = self.tgt_node = self.sec_node
5153
      n2 = self.oth_node = instance.primary_node
5154
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5155
      n1 = self.new_node = remote_node
5156
      n2 = self.oth_node = instance.primary_node
5157
      self.tgt_node = self.sec_node
5158
      _CheckNodeNotDrained(self, remote_node)
5159
    else:
5160
      raise errors.ProgrammerError("Unhandled disk replace mode")
5161

    
5162
    _CheckNodeOnline(self, n1)
5163
    _CheckNodeOnline(self, n2)
5164

    
5165
    if not self.op.disks:
5166
      self.op.disks = range(len(instance.disks))
5167

    
5168
    for disk_idx in self.op.disks:
5169
      instance.FindDisk(disk_idx)
5170

    
5171
  def _ExecD8DiskOnly(self, feedback_fn):
5172
    """Replace a disk on the primary or secondary for dbrd8.
5173

5174
    The algorithm for replace is quite complicated:
5175

5176
      1. for each disk to be replaced:
5177

5178
        1. create new LVs on the target node with unique names
5179
        1. detach old LVs from the drbd device
5180
        1. rename old LVs to name_replaced.<time_t>
5181
        1. rename new LVs to old LVs
5182
        1. attach the new LVs (with the old names now) to the drbd device
5183

5184
      1. wait for sync across all devices
5185

5186
      1. for each modified disk:
5187

5188
        1. remove old LVs (which have the name name_replaces.<time_t>)
5189

5190
    Failures are not very well handled.
5191

5192
    """
5193
    steps_total = 6
5194
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5195
    instance = self.instance
5196
    iv_names = {}
5197
    vgname = self.cfg.GetVGName()
5198
    # start of work
5199
    cfg = self.cfg
5200
    tgt_node = self.tgt_node
5201
    oth_node = self.oth_node
5202

    
5203
    # Step: check device activation
5204
    self.proc.LogStep(1, steps_total, "check device existence")
5205
    info("checking volume groups")
5206
    my_vg = cfg.GetVGName()
5207
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5208
    if not results:
5209
      raise errors.OpExecError("Can't list volume groups on the nodes")
5210
    for node in oth_node, tgt_node:
5211
      res = results[node]
5212
      if res.failed or not res.data or my_vg not in res.data:
5213
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5214
                                 (my_vg, node))
5215
    for idx, dev in enumerate(instance.disks):
5216
      if idx not in self.op.disks:
5217
        continue
5218
      for node in tgt_node, oth_node:
5219
        info("checking disk/%d on %s" % (idx, node))
5220
        cfg.SetDiskID(dev, node)
5221
        result = self.rpc.call_blockdev_find(node, dev)
5222
        msg = result.RemoteFailMsg()
5223
        if not msg and not result.payload:
5224
          msg = "disk not found"
5225
        if msg:
5226
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5227
                                   (idx, node, msg))
5228

    
5229
    # Step: check other node consistency
5230
    self.proc.LogStep(2, steps_total, "check peer consistency")
5231
    for idx, dev in enumerate(instance.disks):
5232
      if idx not in self.op.disks:
5233
        continue
5234
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5235
      if not _CheckDiskConsistency(self, dev, oth_node,
5236
                                   oth_node==instance.primary_node):
5237
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5238
                                 " to replace disks on this node (%s)" %
5239
                                 (oth_node, tgt_node))
5240

    
5241
    # Step: create new storage
5242
    self.proc.LogStep(3, steps_total, "allocate new storage")
5243
    for idx, dev in enumerate(instance.disks):
5244
      if idx not in self.op.disks:
5245
        continue
5246
      size = dev.size
5247
      cfg.SetDiskID(dev, tgt_node)
5248
      lv_names = [".disk%d_%s" % (idx, suf)
5249
                  for suf in ["data", "meta"]]
5250
      names = _GenerateUniqueNames(self, lv_names)
5251
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5252
                             logical_id=(vgname, names[0]))
5253
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5254
                             logical_id=(vgname, names[1]))
5255
      new_lvs = [lv_data, lv_meta]
5256
      old_lvs = dev.children
5257
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5258
      info("creating new local storage on %s for %s" %
5259
           (tgt_node, dev.iv_name))
5260
      # we pass force_create=True to force the LVM creation
5261
      for new_lv in new_lvs:
5262
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5263
                        _GetInstanceInfoText(instance), False)
5264

    
5265
    # Step: for each lv, detach+rename*2+attach
5266
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5267
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5268
      info("detaching %s drbd from local storage" % dev.iv_name)
5269
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5270
      result.Raise()
5271
      if not result.data:
5272
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5273
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5274
      #dev.children = []
5275
      #cfg.Update(instance)
5276

    
5277
      # ok, we created the new LVs, so now we know we have the needed
5278
      # storage; as such, we proceed on the target node to rename
5279
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5280
      # using the assumption that logical_id == physical_id (which in
5281
      # turn is the unique_id on that node)
5282

    
5283
      # FIXME(iustin): use a better name for the replaced LVs
5284
      temp_suffix = int(time.time())
5285
      ren_fn = lambda d, suff: (d.physical_id[0],
5286
                                d.physical_id[1] + "_replaced-%s" % suff)
5287
      # build the rename list based on what LVs exist on the node
5288
      rlist = []
5289
      for to_ren in old_lvs:
5290
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5291
        if not result.RemoteFailMsg() and result.payload:
5292
          # device exists
5293
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5294

    
5295
      info("renaming the old LVs on the target node")
5296
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5297
      result.Raise()
5298
      if not result.data:
5299
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5300
      # now we rename the new LVs to the old LVs
5301
      info("renaming the new LVs on the target node")
5302
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5303
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5304
      result.Raise()
5305
      if not result.data:
5306
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5307

    
5308
      for old, new in zip(old_lvs, new_lvs):
5309
        new.logical_id = old.logical_id
5310
        cfg.SetDiskID(new, tgt_node)
5311

    
5312
      for disk in old_lvs:
5313
        disk.logical_id = ren_fn(disk, temp_suffix)
5314
        cfg.SetDiskID(disk, tgt_node)
5315

    
5316
      # now that the new lvs have the old name, we can add them to the device
5317
      info("adding new mirror component on %s" % tgt_node)
5318
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5319
      if result.failed or not result.data:
5320
        for new_lv in new_lvs:
5321
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5322
          if msg:
5323
            warning("Can't rollback device %s: %s", dev, msg,
5324
                    hint="cleanup manually the unused logical volumes")
5325
        raise errors.OpExecError("Can't add local storage to drbd")
5326

    
5327
      dev.children = new_lvs
5328
      cfg.Update(instance)
5329

    
5330
    # Step: wait for sync
5331

    
5332
    # this can fail as the old devices are degraded and _WaitForSync
5333
    # does a combined result over all disks, so we don't check its
5334
    # return value
5335
    self.proc.LogStep(5, steps_total, "sync devices")
5336
    _WaitForSync(self, instance, unlock=True)
5337

    
5338
    # so check manually all the devices
5339
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5340
      cfg.SetDiskID(dev, instance.primary_node)
5341
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5342
      msg = result.RemoteFailMsg()
5343
      if not msg and not result.payload:
5344
        msg = "disk not found"
5345
      if msg:
5346
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5347
                                 (name, msg))
5348
      if result.payload[5]:
5349
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5350

    
5351
    # Step: remove old storage
5352
    self.proc.LogStep(6, steps_total, "removing old storage")
5353
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5354
      info("remove logical volumes for %s" % name)
5355
      for lv in old_lvs:
5356
        cfg.SetDiskID(lv, tgt_node)
5357
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5358
        if msg:
5359
          warning("Can't remove old LV: %s" % msg,
5360
                  hint="manually remove unused LVs")
5361
          continue
5362

    
5363
  def _ExecD8Secondary(self, feedback_fn):
5364
    """Replace the secondary node for drbd8.
5365

5366
    The algorithm for replace is quite complicated:
5367
      - for all disks of the instance:
5368
        - create new LVs on the new node with same names
5369
        - shutdown the drbd device on the old secondary
5370
        - disconnect the drbd network on the primary
5371
        - create the drbd device on the new secondary
5372
        - network attach the drbd on the primary, using an artifice:
5373
          the drbd code for Attach() will connect to the network if it
5374
          finds a device which is connected to the good local disks but
5375
          not network enabled
5376
      - wait for sync across all devices
5377
      - remove all disks from the old secondary
5378

5379
    Failures are not very well handled.
5380

5381
    """
5382
    steps_total = 6
5383
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5384
    instance = self.instance
5385
    iv_names = {}
5386
    # start of work
5387
    cfg = self.cfg
5388
    old_node = self.tgt_node
5389
    new_node = self.new_node
5390
    pri_node = instance.primary_node
5391
    nodes_ip = {
5392
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5393
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5394
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5395
      }
5396

    
5397
    # Step: check device activation
5398
    self.proc.LogStep(1, steps_total, "check device existence")
5399
    info("checking volume groups")
5400
    my_vg = cfg.GetVGName()
5401
    results = self.rpc.call_vg_list([pri_node, new_node])
5402
    for node in pri_node, new_node:
5403
      res = results[node]
5404
      if res.failed or not res.data or my_vg not in res.data:
5405
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5406
                                 (my_vg, node))
5407
    for idx, dev in enumerate(instance.disks):
5408
      if idx not in self.op.disks:
5409
        continue
5410
      info("checking disk/%d on %s" % (idx, pri_node))
5411
      cfg.SetDiskID(dev, pri_node)
5412
      result = self.rpc.call_blockdev_find(pri_node, dev)
5413
      msg = result.RemoteFailMsg()
5414
      if not msg and not result.payload:
5415
        msg = "disk not found"
5416
      if msg:
5417
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5418
                                 (idx, pri_node, msg))
5419

    
5420
    # Step: check other node consistency
5421
    self.proc.LogStep(2, steps_total, "check peer consistency")
5422
    for idx, dev in enumerate(instance.disks):
5423
      if idx not in self.op.disks:
5424
        continue
5425
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5426
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5427
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5428
                                 " unsafe to replace the secondary" %
5429
                                 pri_node)
5430

    
5431
    # Step: create new storage
5432
    self.proc.LogStep(3, steps_total, "allocate new storage")
5433
    for idx, dev in enumerate(instance.disks):
5434
      info("adding new local storage on %s for disk/%d" %
5435
           (new_node, idx))
5436
      # we pass force_create=True to force LVM creation
5437
      for new_lv in dev.children:
5438
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5439
                        _GetInstanceInfoText(instance), False)
5440

    
5441
    # Step 4: dbrd minors and drbd setups changes
5442
    # after this, we must manually remove the drbd minors on both the
5443
    # error and the success paths
5444
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5445
                                   instance.name)
5446
    logging.debug("Allocated minors %s" % (minors,))
5447
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5448
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5449
      size = dev.size
5450
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5451
      # create new devices on new_node; note that we create two IDs:
5452
      # one without port, so the drbd will be activated without
5453
      # networking information on the new node at this stage, and one
5454
      # with network, for the latter activation in step 4
5455
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5456
      if pri_node == o_node1:
5457
        p_minor = o_minor1
5458
      else:
5459
        p_minor = o_minor2
5460

    
5461
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5462
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5463

    
5464
      iv_names[idx] = (dev, dev.children, new_net_id)
5465
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5466
                    new_net_id)
5467
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5468
                              logical_id=new_alone_id,
5469
                              children=dev.children,
5470
                              size=dev.size)
5471
      try:
5472
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5473
                              _GetInstanceInfoText(instance), False)
5474
      except errors.GenericError:
5475
        self.cfg.ReleaseDRBDMinors(instance.name)
5476
        raise
5477

    
5478
    for idx, dev in enumerate(instance.disks):
5479
      # we have new devices, shutdown the drbd on the old secondary
5480
      info("shutting down drbd for disk/%d on old node" % idx)
5481
      cfg.SetDiskID(dev, old_node)
5482
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5483
      if msg:
5484
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5485
                (idx, msg),
5486
                hint="Please cleanup this device manually as soon as possible")
5487

    
5488
    info("detaching primary drbds from the network (=> standalone)")
5489
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5490
                                               instance.disks)[pri_node]
5491

    
5492
    msg = result.RemoteFailMsg()
5493
    if msg:
5494
      # detaches didn't succeed (unlikely)
5495
      self.cfg.ReleaseDRBDMinors(instance.name)
5496
      raise errors.OpExecError("Can't detach the disks from the network on"
5497
                               " old node: %s" % (msg,))
5498

    
5499
    # if we managed to detach at least one, we update all the disks of
5500
    # the instance to point to the new secondary
5501
    info("updating instance configuration")
5502
    for dev, _, new_logical_id in iv_names.itervalues():
5503
      dev.logical_id = new_logical_id
5504
      cfg.SetDiskID(dev, pri_node)
5505
    cfg.Update(instance)
5506

    
5507
    # and now perform the drbd attach
5508
    info("attaching primary drbds to new secondary (standalone => connected)")
5509
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5510
                                           instance.disks, instance.name,
5511
                                           False)
5512
    for to_node, to_result in result.items():
5513
      msg = to_result.RemoteFailMsg()
5514
      if msg:
5515
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5516
                hint="please do a gnt-instance info to see the"
5517
                " status of disks")
5518

    
5519
    # this can fail as the old devices are degraded and _WaitForSync
5520
    # does a combined result over all disks, so we don't check its
5521
    # return value
5522
    self.proc.LogStep(5, steps_total, "sync devices")
5523
    _WaitForSync(self, instance, unlock=True)
5524

    
5525
    # so check manually all the devices
5526
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5527
      cfg.SetDiskID(dev, pri_node)
5528
      result = self.rpc.call_blockdev_find(pri_node, dev)
5529
      msg = result.RemoteFailMsg()
5530
      if not msg and not result.payload:
5531
        msg = "disk not found"
5532
      if msg:
5533
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5534
                                 (idx, msg))
5535
      if result.payload[5]:
5536
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5537

    
5538
    self.proc.LogStep(6, steps_total, "removing old storage")
5539
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5540
      info("remove logical volumes for disk/%d" % idx)
5541
      for lv in old_lvs:
5542
        cfg.SetDiskID(lv, old_node)
5543
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5544
        if msg:
5545
          warning("Can't remove LV on old secondary: %s", msg,
5546
                  hint="Cleanup stale volumes by hand")
5547

    
5548
  def Exec(self, feedback_fn):
5549
    """Execute disk replacement.
5550

5551
    This dispatches the disk replacement to the appropriate handler.
5552

5553
    """
5554
    instance = self.instance
5555

    
5556
    # Activate the instance disks if we're replacing them on a down instance
5557
    if not instance.admin_up:
5558
      _StartInstanceDisks(self, instance, True)
5559

    
5560
    if self.op.mode == constants.REPLACE_DISK_CHG:
5561
      fn = self._ExecD8Secondary
5562
    else:
5563
      fn = self._ExecD8DiskOnly
5564

    
5565
    ret = fn(feedback_fn)
5566

    
5567
    # Deactivate the instance disks if we're replacing them on a down instance
5568
    if not instance.admin_up:
5569
      _SafeShutdownInstanceDisks(self, instance)
5570

    
5571
    return ret
5572

    
5573

    
5574
class LUGrowDisk(LogicalUnit):
5575
  """Grow a disk of an instance.
5576

5577
  """
5578
  HPATH = "disk-grow"
5579
  HTYPE = constants.HTYPE_INSTANCE
5580
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5581
  REQ_BGL = False
5582

    
5583
  def ExpandNames(self):
5584
    self._ExpandAndLockInstance()
5585
    self.needed_locks[locking.LEVEL_NODE] = []
5586
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5587

    
5588
  def DeclareLocks(self, level):
5589
    if level == locking.LEVEL_NODE:
5590
      self._LockInstancesNodes()
5591

    
5592
  def BuildHooksEnv(self):
5593
    """Build hooks env.
5594

5595
    This runs on the master, the primary and all the secondaries.
5596

5597
    """
5598
    env = {
5599
      "DISK": self.op.disk,
5600
      "AMOUNT": self.op.amount,
5601
      }
5602
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5603
    nl = [
5604
      self.cfg.GetMasterNode(),
5605
      self.instance.primary_node,
5606
      ]
5607
    return env, nl, nl
5608

    
5609
  def CheckPrereq(self):
5610
    """Check prerequisites.
5611

5612
    This checks that the instance is in the cluster.
5613

5614
    """
5615
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5616
    assert instance is not None, \
5617
      "Cannot retrieve locked instance %s" % self.op.instance_name
5618
    nodenames = list(instance.all_nodes)
5619
    for node in nodenames:
5620
      _CheckNodeOnline(self, node)
5621

    
5622

    
5623
    self.instance = instance
5624

    
5625
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5626
      raise errors.OpPrereqError("Instance's disk layout does not support"
5627
                                 " growing.")
5628

    
5629
    self.disk = instance.FindDisk(self.op.disk)
5630

    
5631
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5632
                                       instance.hypervisor)
5633
    for node in nodenames:
5634
      info = nodeinfo[node]
5635
      if info.failed or not info.data:
5636
        raise errors.OpPrereqError("Cannot get current information"
5637
                                   " from node '%s'" % node)
5638
      vg_free = info.data.get('vg_free', None)
5639
      if not isinstance(vg_free, int):
5640
        raise errors.OpPrereqError("Can't compute free disk space on"
5641
                                   " node %s" % node)
5642
      if self.op.amount > vg_free:
5643
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5644
                                   " %d MiB available, %d MiB required" %
5645
                                   (node, vg_free, self.op.amount))
5646

    
5647
  def Exec(self, feedback_fn):
5648
    """Execute disk grow.
5649

5650
    """
5651
    instance = self.instance
5652
    disk = self.disk
5653
    for node in instance.all_nodes:
5654
      self.cfg.SetDiskID(disk, node)
5655
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5656
      msg = result.RemoteFailMsg()
5657
      if msg:
5658
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5659
                                 (node, msg))
5660
    disk.RecordGrow(self.op.amount)
5661
    self.cfg.Update(instance)
5662
    if self.op.wait_for_sync:
5663
      disk_abort = not _WaitForSync(self, instance)
5664
      if disk_abort:
5665
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5666
                             " status.\nPlease check the instance.")
5667

    
5668

    
5669
class LUQueryInstanceData(NoHooksLU):
5670
  """Query runtime instance data.
5671

5672
  """
5673
  _OP_REQP = ["instances", "static"]
5674
  REQ_BGL = False
5675

    
5676
  def ExpandNames(self):
5677
    self.needed_locks = {}
5678
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5679

    
5680
    if not isinstance(self.op.instances, list):
5681
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5682

    
5683
    if self.op.instances:
5684
      self.wanted_names = []
5685
      for name in self.op.instances:
5686
        full_name = self.cfg.ExpandInstanceName(name)
5687
        if full_name is None:
5688
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5689
        self.wanted_names.append(full_name)
5690
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5691
    else:
5692
      self.wanted_names = None
5693
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5694

    
5695
    self.needed_locks[locking.LEVEL_NODE] = []
5696
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5697

    
5698
  def DeclareLocks(self, level):
5699
    if level == locking.LEVEL_NODE:
5700
      self._LockInstancesNodes()
5701

    
5702
  def CheckPrereq(self):
5703
    """Check prerequisites.
5704

5705
    This only checks the optional instance list against the existing names.
5706

5707
    """
5708
    if self.wanted_names is None:
5709
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5710

    
5711
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5712
                             in self.wanted_names]
5713
    return
5714

    
5715
  def _ComputeDiskStatus(self, instance, snode, dev):
5716
    """Compute block device status.
5717

5718
    """
5719
    static = self.op.static
5720
    if not static:
5721
      self.cfg.SetDiskID(dev, instance.primary_node)
5722
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5723
      if dev_pstatus.offline:
5724
        dev_pstatus = None
5725
      else:
5726
        msg = dev_pstatus.RemoteFailMsg()
5727
        if msg:
5728
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5729
                                   (instance.name, msg))
5730
        dev_pstatus = dev_pstatus.payload
5731
    else:
5732
      dev_pstatus = None
5733

    
5734
    if dev.dev_type in constants.LDS_DRBD:
5735
      # we change the snode then (otherwise we use the one passed in)
5736
      if dev.logical_id[0] == instance.primary_node:
5737
        snode = dev.logical_id[1]
5738
      else:
5739
        snode = dev.logical_id[0]
5740

    
5741
    if snode and not static:
5742
      self.cfg.SetDiskID(dev, snode)
5743
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5744
      if dev_sstatus.offline:
5745
        dev_sstatus = None
5746
      else:
5747
        msg = dev_sstatus.RemoteFailMsg()
5748
        if msg:
5749
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5750
                                   (instance.name, msg))
5751
        dev_sstatus = dev_sstatus.payload
5752
    else:
5753
      dev_sstatus = None
5754

    
5755
    if dev.children:
5756
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5757
                      for child in dev.children]
5758
    else:
5759
      dev_children = []
5760

    
5761
    data = {
5762
      "iv_name": dev.iv_name,
5763
      "dev_type": dev.dev_type,
5764
      "logical_id": dev.logical_id,
5765
      "physical_id": dev.physical_id,
5766
      "pstatus": dev_pstatus,
5767
      "sstatus": dev_sstatus,
5768
      "children": dev_children,
5769
      "mode": dev.mode,
5770
      "size": dev.size,
5771
      }
5772

    
5773
    return data
5774

    
5775
  def Exec(self, feedback_fn):
5776
    """Gather and return data"""
5777
    result = {}
5778

    
5779
    cluster = self.cfg.GetClusterInfo()
5780

    
5781
    for instance in self.wanted_instances:
5782
      if not self.op.static:
5783
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5784
                                                  instance.name,
5785
                                                  instance.hypervisor)
5786
        remote_info.Raise()
5787
        remote_info = remote_info.data
5788
        if remote_info and "state" in remote_info:
5789
          remote_state = "up"
5790
        else:
5791
          remote_state = "down"
5792
      else:
5793
        remote_state = None
5794
      if instance.admin_up:
5795
        config_state = "up"
5796
      else:
5797
        config_state = "down"
5798

    
5799
      disks = [self._ComputeDiskStatus(instance, None, device)
5800
               for device in instance.disks]
5801

    
5802
      idict = {
5803
        "name": instance.name,
5804
        "config_state": config_state,
5805
        "run_state": remote_state,
5806
        "pnode": instance.primary_node,
5807
        "snodes": instance.secondary_nodes,
5808
        "os": instance.os,
5809
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5810
        "disks": disks,
5811
        "hypervisor": instance.hypervisor,
5812
        "network_port": instance.network_port,
5813
        "hv_instance": instance.hvparams,
5814
        "hv_actual": cluster.FillHV(instance),
5815
        "be_instance": instance.beparams,
5816
        "be_actual": cluster.FillBE(instance),
5817
        }
5818

    
5819
      result[instance.name] = idict
5820

    
5821
    return result
5822

    
5823

    
5824
class LUSetInstanceParams(LogicalUnit):
5825
  """Modifies an instances's parameters.
5826

5827
  """
5828
  HPATH = "instance-modify"
5829
  HTYPE = constants.HTYPE_INSTANCE
5830
  _OP_REQP = ["instance_name"]
5831
  REQ_BGL = False
5832

    
5833
  def CheckArguments(self):
5834
    if not hasattr(self.op, 'nics'):
5835
      self.op.nics = []
5836
    if not hasattr(self.op, 'disks'):
5837
      self.op.disks = []
5838
    if not hasattr(self.op, 'beparams'):
5839
      self.op.beparams = {}
5840
    if not hasattr(self.op, 'hvparams'):
5841
      self.op.hvparams = {}
5842
    self.op.force = getattr(self.op, "force", False)
5843
    if not (self.op.nics or self.op.disks or
5844
            self.op.hvparams or self.op.beparams):
5845
      raise errors.OpPrereqError("No changes submitted")
5846

    
5847
    # Disk validation
5848
    disk_addremove = 0
5849
    for disk_op, disk_dict in self.op.disks:
5850
      if disk_op == constants.DDM_REMOVE:
5851
        disk_addremove += 1
5852
        continue
5853
      elif disk_op == constants.DDM_ADD:
5854
        disk_addremove += 1
5855
      else:
5856
        if not isinstance(disk_op, int):
5857
          raise errors.OpPrereqError("Invalid disk index")
5858
      if disk_op == constants.DDM_ADD:
5859
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5860
        if mode not in constants.DISK_ACCESS_SET:
5861
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5862
        size = disk_dict.get('size', None)
5863
        if size is None:
5864
          raise errors.OpPrereqError("Required disk parameter size missing")
5865
        try:
5866
          size = int(size)
5867
        except ValueError, err:
5868
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5869
                                     str(err))
5870
        disk_dict['size'] = size
5871
      else:
5872
        # modification of disk
5873
        if 'size' in disk_dict:
5874
          raise errors.OpPrereqError("Disk size change not possible, use"
5875
                                     " grow-disk")
5876

    
5877
    if disk_addremove > 1:
5878
      raise errors.OpPrereqError("Only one disk add or remove operation"
5879
                                 " supported at a time")
5880

    
5881
    # NIC validation
5882
    nic_addremove = 0
5883
    for nic_op, nic_dict in self.op.nics:
5884
      if nic_op == constants.DDM_REMOVE:
5885
        nic_addremove += 1
5886
        continue
5887
      elif nic_op == constants.DDM_ADD:
5888
        nic_addremove += 1
5889
      else:
5890
        if not isinstance(nic_op, int):
5891
          raise errors.OpPrereqError("Invalid nic index")
5892

    
5893
      # nic_dict should be a dict
5894
      nic_ip = nic_dict.get('ip', None)
5895
      if nic_ip is not None:
5896
        if nic_ip.lower() == constants.VALUE_NONE:
5897
          nic_dict['ip'] = None
5898
        else:
5899
          if not utils.IsValidIP(nic_ip):
5900
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5901

    
5902
      if nic_op == constants.DDM_ADD:
5903
        nic_bridge = nic_dict.get('bridge', None)
5904
        if nic_bridge is None:
5905
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5906
        nic_mac = nic_dict.get('mac', None)
5907
        if nic_mac is None:
5908
          nic_dict['mac'] = constants.VALUE_AUTO
5909

    
5910
      if 'mac' in nic_dict:
5911
        nic_mac = nic_dict['mac']
5912
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5913
          if not utils.IsValidMac(nic_mac):
5914
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5915
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5916
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5917
                                     " modifying an existing nic")
5918

    
5919
    if nic_addremove > 1:
5920
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5921
                                 " supported at a time")
5922

    
5923
  def ExpandNames(self):
5924
    self._ExpandAndLockInstance()
5925
    self.needed_locks[locking.LEVEL_NODE] = []
5926
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5927

    
5928
  def DeclareLocks(self, level):
5929
    if level == locking.LEVEL_NODE:
5930
      self._LockInstancesNodes()
5931

    
5932
  def BuildHooksEnv(self):
5933
    """Build hooks env.
5934

5935
    This runs on the master, primary and secondaries.
5936

5937
    """
5938
    args = dict()
5939
    if constants.BE_MEMORY in self.be_new:
5940
      args['memory'] = self.be_new[constants.BE_MEMORY]
5941
    if constants.BE_VCPUS in self.be_new:
5942
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5943
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5944
    # information at all.
5945
    if self.op.nics:
5946
      args['nics'] = []
5947
      nic_override = dict(self.op.nics)
5948
      for idx, nic in enumerate(self.instance.nics):
5949
        if idx in nic_override:
5950
          this_nic_override = nic_override[idx]
5951
        else:
5952
          this_nic_override = {}
5953
        if 'ip' in this_nic_override:
5954
          ip = this_nic_override['ip']
5955
        else:
5956
          ip = nic.ip
5957
        if 'bridge' in this_nic_override:
5958
          bridge = this_nic_override['bridge']
5959
        else:
5960
          bridge = nic.bridge
5961
        if 'mac' in this_nic_override:
5962
          mac = this_nic_override['mac']
5963
        else:
5964
          mac = nic.mac
5965
        args['nics'].append((ip, bridge, mac))
5966
      if constants.DDM_ADD in nic_override:
5967
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5968
        bridge = nic_override[constants.DDM_ADD]['bridge']
5969
        mac = nic_override[constants.DDM_ADD]['mac']
5970
        args['nics'].append((ip, bridge, mac))
5971
      elif constants.DDM_REMOVE in nic_override:
5972
        del args['nics'][-1]
5973

    
5974
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5975
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5976
    return env, nl, nl
5977

    
5978
  def CheckPrereq(self):
5979
    """Check prerequisites.
5980

5981
    This only checks the instance list against the existing names.
5982

5983
    """
5984
    force = self.force = self.op.force
5985

    
5986
    # checking the new params on the primary/secondary nodes
5987

    
5988
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5989
    assert self.instance is not None, \
5990
      "Cannot retrieve locked instance %s" % self.op.instance_name
5991
    pnode = instance.primary_node
5992
    nodelist = list(instance.all_nodes)
5993

    
5994
    # hvparams processing
5995
    if self.op.hvparams:
5996
      i_hvdict = copy.deepcopy(instance.hvparams)
5997
      for key, val in self.op.hvparams.iteritems():
5998
        if val == constants.VALUE_DEFAULT:
5999
          try:
6000
            del i_hvdict[key]
6001
          except KeyError:
6002
            pass
6003
        else:
6004
          i_hvdict[key] = val
6005
      cluster = self.cfg.GetClusterInfo()
6006
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
6007
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
6008
                                i_hvdict)
6009
      # local check
6010
      hypervisor.GetHypervisor(
6011
        instance.hypervisor).CheckParameterSyntax(hv_new)
6012
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
6013
      self.hv_new = hv_new # the new actual values
6014
      self.hv_inst = i_hvdict # the new dict (without defaults)
6015
    else:
6016
      self.hv_new = self.hv_inst = {}
6017

    
6018
    # beparams processing
6019
    if self.op.beparams:
6020
      i_bedict = copy.deepcopy(instance.beparams)
6021
      for key, val in self.op.beparams.iteritems():
6022
        if val == constants.VALUE_DEFAULT:
6023
          try:
6024
            del i_bedict[key]
6025
          except KeyError:
6026
            pass
6027
        else:
6028
          i_bedict[key] = val
6029
      cluster = self.cfg.GetClusterInfo()
6030
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
6031
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
6032
                                i_bedict)
6033
      self.be_new = be_new # the new actual values
6034
      self.be_inst = i_bedict # the new dict (without defaults)
6035
    else:
6036
      self.be_new = self.be_inst = {}
6037

    
6038
    self.warn = []
6039

    
6040
    if constants.BE_MEMORY in self.op.beparams and not self.force:
6041
      mem_check_list = [pnode]
6042
      if be_new[constants.BE_AUTO_BALANCE]:
6043
        # either we changed auto_balance to yes or it was from before
6044
        mem_check_list.extend(instance.secondary_nodes)
6045
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
6046
                                                  instance.hypervisor)
6047
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
6048
                                         instance.hypervisor)
6049
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
6050
        # Assume the primary node is unreachable and go ahead
6051
        self.warn.append("Can't get info from primary node %s" % pnode)
6052
      else:
6053
        if not instance_info.failed and instance_info.data:
6054
          current_mem = int(instance_info.data['memory'])
6055
        else:
6056
          # Assume instance not running
6057
          # (there is a slight race condition here, but it's not very probable,
6058
          # and we have no other way to check)
6059
          current_mem = 0
6060
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
6061
                    nodeinfo[pnode].data['memory_free'])
6062
        if miss_mem > 0:
6063
          raise errors.OpPrereqError("This change will prevent the instance"
6064
                                     " from starting, due to %d MB of memory"
6065
                                     " missing on its primary node" % miss_mem)
6066

    
6067
      if be_new[constants.BE_AUTO_BALANCE]:
6068
        for node, nres in nodeinfo.iteritems():
6069
          if node not in instance.secondary_nodes:
6070
            continue
6071
          if nres.failed or not isinstance(nres.data, dict):
6072
            self.warn.append("Can't get info from secondary node %s" % node)
6073
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
6074
            self.warn.append("Not enough memory to failover instance to"
6075
                             " secondary node %s" % node)
6076

    
6077
    # NIC processing
6078
    for nic_op, nic_dict in self.op.nics:
6079
      if nic_op == constants.DDM_REMOVE:
6080
        if not instance.nics:
6081
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
6082
        continue
6083
      if nic_op != constants.DDM_ADD:
6084
        # an existing nic
6085
        if nic_op < 0 or nic_op >= len(instance.nics):
6086
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
6087
                                     " are 0 to %d" %
6088
                                     (nic_op, len(instance.nics)))
6089
      if 'bridge' in nic_dict:
6090
        nic_bridge = nic_dict['bridge']
6091
        if nic_bridge is None:
6092
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
6093
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
6094
          msg = ("Bridge '%s' doesn't exist on one of"
6095
                 " the instance nodes" % nic_bridge)
6096
          if self.force:
6097
            self.warn.append(msg)
6098
          else:
6099
            raise errors.OpPrereqError(msg)
6100
      if 'mac' in nic_dict:
6101
        nic_mac = nic_dict['mac']
6102
        if nic_mac is None:
6103
          raise errors.OpPrereqError('Cannot set the nic mac to None')
6104
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6105
          # otherwise generate the mac
6106
          nic_dict['mac'] = self.cfg.GenerateMAC()
6107
        else:
6108
          # or validate/reserve the current one
6109
          if self.cfg.IsMacInUse(nic_mac):
6110
            raise errors.OpPrereqError("MAC address %s already in use"
6111
                                       " in cluster" % nic_mac)
6112

    
6113
    # DISK processing
6114
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
6115
      raise errors.OpPrereqError("Disk operations not supported for"
6116
                                 " diskless instances")
6117
    for disk_op, disk_dict in self.op.disks:
6118
      if disk_op == constants.DDM_REMOVE:
6119
        if len(instance.disks) == 1:
6120
          raise errors.OpPrereqError("Cannot remove the last disk of"
6121
                                     " an instance")
6122
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
6123
        ins_l = ins_l[pnode]
6124
        if ins_l.failed or not isinstance(ins_l.data, list):
6125
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
6126
        if instance.name in ins_l.data:
6127
          raise errors.OpPrereqError("Instance is running, can't remove"
6128
                                     " disks.")
6129

    
6130
      if (disk_op == constants.DDM_ADD and
6131
          len(instance.nics) >= constants.MAX_DISKS):
6132
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
6133
                                   " add more" % constants.MAX_DISKS)
6134
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
6135
        # an existing disk
6136
        if disk_op < 0 or disk_op >= len(instance.disks):
6137
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
6138
                                     " are 0 to %d" %
6139
                                     (disk_op, len(instance.disks)))
6140

    
6141
    return
6142

    
6143
  def Exec(self, feedback_fn):
6144
    """Modifies an instance.
6145

6146
    All parameters take effect only at the next restart of the instance.
6147

6148
    """
6149
    # Process here the warnings from CheckPrereq, as we don't have a
6150
    # feedback_fn there.
6151
    for warn in self.warn:
6152
      feedback_fn("WARNING: %s" % warn)
6153

    
6154
    result = []
6155
    instance = self.instance
6156
    # disk changes
6157
    for disk_op, disk_dict in self.op.disks:
6158
      if disk_op == constants.DDM_REMOVE:
6159
        # remove the last disk
6160
        device = instance.disks.pop()
6161
        device_idx = len(instance.disks)
6162
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6163
          self.cfg.SetDiskID(disk, node)
6164
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6165
          if msg:
6166
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6167
                            " continuing anyway", device_idx, node, msg)
6168
        result.append(("disk/%d" % device_idx, "remove"))
6169
      elif disk_op == constants.DDM_ADD:
6170
        # add a new disk
6171
        if instance.disk_template == constants.DT_FILE:
6172
          file_driver, file_path = instance.disks[0].logical_id
6173
          file_path = os.path.dirname(file_path)
6174
        else:
6175
          file_driver = file_path = None
6176
        disk_idx_base = len(instance.disks)
6177
        new_disk = _GenerateDiskTemplate(self,
6178
                                         instance.disk_template,
6179
                                         instance.name, instance.primary_node,
6180
                                         instance.secondary_nodes,
6181
                                         [disk_dict],
6182
                                         file_path,
6183
                                         file_driver,
6184
                                         disk_idx_base)[0]
6185
        instance.disks.append(new_disk)
6186
        info = _GetInstanceInfoText(instance)
6187

    
6188
        logging.info("Creating volume %s for instance %s",
6189
                     new_disk.iv_name, instance.name)
6190
        # Note: this needs to be kept in sync with _CreateDisks
6191
        #HARDCODE
6192
        for node in instance.all_nodes:
6193
          f_create = node == instance.primary_node
6194
          try:
6195
            _CreateBlockDev(self, node, instance, new_disk,
6196
                            f_create, info, f_create)
6197
          except errors.OpExecError, err:
6198
            self.LogWarning("Failed to create volume %s (%s) on"
6199
                            " node %s: %s",
6200
                            new_disk.iv_name, new_disk, node, err)
6201
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6202
                       (new_disk.size, new_disk.mode)))
6203
      else:
6204
        # change a given disk
6205
        instance.disks[disk_op].mode = disk_dict['mode']
6206
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6207
    # NIC changes
6208
    for nic_op, nic_dict in self.op.nics:
6209
      if nic_op == constants.DDM_REMOVE:
6210
        # remove the last nic
6211
        del instance.nics[-1]
6212
        result.append(("nic.%d" % len(instance.nics), "remove"))
6213
      elif nic_op == constants.DDM_ADD:
6214
        # mac and bridge should be set, by now
6215
        mac = nic_dict['mac']
6216
        bridge = nic_dict['bridge']
6217
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6218
                              bridge=bridge)
6219
        instance.nics.append(new_nic)
6220
        result.append(("nic.%d" % (len(instance.nics) - 1),
6221
                       "add:mac=%s,ip=%s,bridge=%s" %
6222
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6223
      else:
6224
        # change a given nic
6225
        for key in 'mac', 'ip', 'bridge':
6226
          if key in nic_dict:
6227
            setattr(instance.nics[nic_op], key, nic_dict[key])
6228
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6229

    
6230
    # hvparams changes
6231
    if self.op.hvparams:
6232
      instance.hvparams = self.hv_inst
6233
      for key, val in self.op.hvparams.iteritems():
6234
        result.append(("hv/%s" % key, val))
6235

    
6236
    # beparams changes
6237
    if self.op.beparams:
6238
      instance.beparams = self.be_inst
6239
      for key, val in self.op.beparams.iteritems():
6240
        result.append(("be/%s" % key, val))
6241

    
6242
    self.cfg.Update(instance)
6243

    
6244
    return result
6245

    
6246

    
6247
class LUQueryExports(NoHooksLU):
6248
  """Query the exports list
6249

6250
  """
6251
  _OP_REQP = ['nodes']
6252
  REQ_BGL = False
6253

    
6254
  def ExpandNames(self):
6255
    self.needed_locks = {}
6256
    self.share_locks[locking.LEVEL_NODE] = 1
6257
    if not self.op.nodes:
6258
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6259
    else:
6260
      self.needed_locks[locking.LEVEL_NODE] = \
6261
        _GetWantedNodes(self, self.op.nodes)
6262

    
6263
  def CheckPrereq(self):
6264
    """Check prerequisites.
6265

6266
    """
6267
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6268

    
6269
  def Exec(self, feedback_fn):
6270
    """Compute the list of all the exported system images.
6271

6272
    @rtype: dict
6273
    @return: a dictionary with the structure node->(export-list)
6274
        where export-list is a list of the instances exported on
6275
        that node.
6276

6277
    """
6278
    rpcresult = self.rpc.call_export_list(self.nodes)
6279
    result = {}
6280
    for node in rpcresult:
6281
      if rpcresult[node].failed:
6282
        result[node] = False
6283
      else:
6284
        result[node] = rpcresult[node].data
6285

    
6286
    return result
6287

    
6288

    
6289
class LUExportInstance(LogicalUnit):
6290
  """Export an instance to an image in the cluster.
6291

6292
  """
6293
  HPATH = "instance-export"
6294
  HTYPE = constants.HTYPE_INSTANCE
6295
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6296
  REQ_BGL = False
6297

    
6298
  def ExpandNames(self):
6299
    self._ExpandAndLockInstance()
6300
    # FIXME: lock only instance primary and destination node
6301
    #
6302
    # Sad but true, for now we have do lock all nodes, as we don't know where
6303
    # the previous export might be, and and in this LU we search for it and
6304
    # remove it from its current node. In the future we could fix this by:
6305
    #  - making a tasklet to search (share-lock all), then create the new one,
6306
    #    then one to remove, after
6307
    #  - removing the removal operation altoghether
6308
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6309

    
6310
  def DeclareLocks(self, level):
6311
    """Last minute lock declaration."""
6312
    # All nodes are locked anyway, so nothing to do here.
6313

    
6314
  def BuildHooksEnv(self):
6315
    """Build hooks env.
6316

6317
    This will run on the master, primary node and target node.
6318

6319
    """
6320
    env = {
6321
      "EXPORT_NODE": self.op.target_node,
6322
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6323
      }
6324
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6325
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6326
          self.op.target_node]
6327
    return env, nl, nl
6328

    
6329
  def CheckPrereq(self):
6330
    """Check prerequisites.
6331

6332
    This checks that the instance and node names are valid.
6333

6334
    """
6335
    instance_name = self.op.instance_name
6336
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6337
    assert self.instance is not None, \
6338
          "Cannot retrieve locked instance %s" % self.op.instance_name
6339
    _CheckNodeOnline(self, self.instance.primary_node)
6340

    
6341
    self.dst_node = self.cfg.GetNodeInfo(
6342
      self.cfg.ExpandNodeName(self.op.target_node))
6343

    
6344
    if self.dst_node is None:
6345
      # This is wrong node name, not a non-locked node
6346
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6347
    _CheckNodeOnline(self, self.dst_node.name)
6348
    _CheckNodeNotDrained(self, self.dst_node.name)
6349

    
6350
    # instance disk type verification
6351
    for disk in self.instance.disks:
6352
      if disk.dev_type == constants.LD_FILE:
6353
        raise errors.OpPrereqError("Export not supported for instances with"
6354
                                   " file-based disks")
6355

    
6356
  def Exec(self, feedback_fn):
6357
    """Export an instance to an image in the cluster.
6358

6359
    """
6360
    instance = self.instance
6361
    dst_node = self.dst_node
6362
    src_node = instance.primary_node
6363
    if self.op.shutdown:
6364
      # shutdown the instance, but not the disks
6365
      result = self.rpc.call_instance_shutdown(src_node, instance)
6366
      msg = result.RemoteFailMsg()
6367
      if msg:
6368
        raise errors.OpExecError("Could not shutdown instance %s on"
6369
                                 " node %s: %s" %
6370
                                 (instance.name, src_node, msg))
6371

    
6372
    vgname = self.cfg.GetVGName()
6373

    
6374
    snap_disks = []
6375

    
6376
    # set the disks ID correctly since call_instance_start needs the
6377
    # correct drbd minor to create the symlinks
6378
    for disk in instance.disks:
6379
      self.cfg.SetDiskID(disk, src_node)
6380

    
6381
    try:
6382
      for idx, disk in enumerate(instance.disks):
6383
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6384
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6385
        if new_dev_name.failed or not new_dev_name.data:
6386
          self.LogWarning("Could not snapshot disk/%d on node %s",
6387
                          idx, src_node)
6388
          snap_disks.append(False)
6389
        else:
6390
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6391
                                 logical_id=(vgname, new_dev_name.data),
6392
                                 physical_id=(vgname, new_dev_name.data),
6393
                                 iv_name=disk.iv_name)
6394
          snap_disks.append(new_dev)
6395

    
6396
    finally:
6397
      if self.op.shutdown and instance.admin_up:
6398
        result = self.rpc.call_instance_start(src_node, instance, None, None)
6399
        msg = result.RemoteFailMsg()
6400
        if msg:
6401
          _ShutdownInstanceDisks(self, instance)
6402
          raise errors.OpExecError("Could not start instance: %s" % msg)
6403

    
6404
    # TODO: check for size
6405

    
6406
    cluster_name = self.cfg.GetClusterName()
6407
    for idx, dev in enumerate(snap_disks):
6408
      if dev:
6409
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6410
                                               instance, cluster_name, idx)
6411
        if result.failed or not result.data:
6412
          self.LogWarning("Could not export disk/%d from node %s to"
6413
                          " node %s", idx, src_node, dst_node.name)
6414
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6415
        if msg:
6416
          self.LogWarning("Could not remove snapshot for disk/%d from node"
6417
                          " %s: %s", idx, src_node, msg)
6418

    
6419
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6420
    if result.failed or not result.data:
6421
      self.LogWarning("Could not finalize export for instance %s on node %s",
6422
                      instance.name, dst_node.name)
6423

    
6424
    nodelist = self.cfg.GetNodeList()
6425
    nodelist.remove(dst_node.name)
6426

    
6427
    # on one-node clusters nodelist will be empty after the removal
6428
    # if we proceed the backup would be removed because OpQueryExports
6429
    # substitutes an empty list with the full cluster node list.
6430
    if nodelist:
6431
      exportlist = self.rpc.call_export_list(nodelist)
6432
      for node in exportlist:
6433
        if exportlist[node].failed:
6434
          continue
6435
        if instance.name in exportlist[node].data:
6436
          if not self.rpc.call_export_remove(node, instance.name):
6437
            self.LogWarning("Could not remove older export for instance %s"
6438
                            " on node %s", instance.name, node)
6439

    
6440

    
6441
class LURemoveExport(NoHooksLU):
6442
  """Remove exports related to the named instance.
6443

6444
  """
6445
  _OP_REQP = ["instance_name"]
6446
  REQ_BGL = False
6447

    
6448
  def ExpandNames(self):
6449
    self.needed_locks = {}
6450
    # We need all nodes to be locked in order for RemoveExport to work, but we
6451
    # don't need to lock the instance itself, as nothing will happen to it (and
6452
    # we can remove exports also for a removed instance)
6453
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6454

    
6455
  def CheckPrereq(self):
6456
    """Check prerequisites.
6457
    """
6458
    pass
6459

    
6460
  def Exec(self, feedback_fn):
6461
    """Remove any export.
6462

6463
    """
6464
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6465
    # If the instance was not found we'll try with the name that was passed in.
6466
    # This will only work if it was an FQDN, though.
6467
    fqdn_warn = False
6468
    if not instance_name:
6469
      fqdn_warn = True
6470
      instance_name = self.op.instance_name
6471

    
6472
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6473
      locking.LEVEL_NODE])
6474
    found = False
6475
    for node in exportlist:
6476
      if exportlist[node].failed:
6477
        self.LogWarning("Failed to query node %s, continuing" % node)
6478
        continue
6479
      if instance_name in exportlist[node].data:
6480
        found = True
6481
        result = self.rpc.call_export_remove(node, instance_name)
6482
        if result.failed or not result.data:
6483
          logging.error("Could not remove export for instance %s"
6484
                        " on node %s", instance_name, node)
6485

    
6486
    if fqdn_warn and not found:
6487
      feedback_fn("Export not found. If trying to remove an export belonging"
6488
                  " to a deleted instance please use its Fully Qualified"
6489
                  " Domain Name.")
6490

    
6491

    
6492
class TagsLU(NoHooksLU):
6493
  """Generic tags LU.
6494

6495
  This is an abstract class which is the parent of all the other tags LUs.
6496

6497
  """
6498

    
6499
  def ExpandNames(self):
6500
    self.needed_locks = {}
6501
    if self.op.kind == constants.TAG_NODE:
6502
      name = self.cfg.ExpandNodeName(self.op.name)
6503
      if name is None:
6504
        raise errors.OpPrereqError("Invalid node name (%s)" %
6505
                                   (self.op.name,))
6506
      self.op.name = name
6507
      self.needed_locks[locking.LEVEL_NODE] = name
6508
    elif self.op.kind == constants.TAG_INSTANCE:
6509
      name = self.cfg.ExpandInstanceName(self.op.name)
6510
      if name is None:
6511
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6512
                                   (self.op.name,))
6513
      self.op.name = name
6514
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6515

    
6516
  def CheckPrereq(self):
6517
    """Check prerequisites.
6518

6519
    """
6520
    if self.op.kind == constants.TAG_CLUSTER:
6521
      self.target = self.cfg.GetClusterInfo()
6522
    elif self.op.kind == constants.TAG_NODE:
6523
      self.target = self.cfg.GetNodeInfo(self.op.name)
6524
    elif self.op.kind == constants.TAG_INSTANCE:
6525
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6526
    else:
6527
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6528
                                 str(self.op.kind))
6529

    
6530

    
6531
class LUGetTags(TagsLU):
6532
  """Returns the tags of a given object.
6533

6534
  """
6535
  _OP_REQP = ["kind", "name"]
6536
  REQ_BGL = False
6537

    
6538
  def Exec(self, feedback_fn):
6539
    """Returns the tag list.
6540

6541
    """
6542
    return list(self.target.GetTags())
6543

    
6544

    
6545
class LUSearchTags(NoHooksLU):
6546
  """Searches the tags for a given pattern.
6547

6548
  """
6549
  _OP_REQP = ["pattern"]
6550
  REQ_BGL = False
6551

    
6552
  def ExpandNames(self):
6553
    self.needed_locks = {}
6554

    
6555
  def CheckPrereq(self):
6556
    """Check prerequisites.
6557

6558
    This checks the pattern passed for validity by compiling it.
6559

6560
    """
6561
    try:
6562
      self.re = re.compile(self.op.pattern)
6563
    except re.error, err:
6564
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6565
                                 (self.op.pattern, err))
6566

    
6567
  def Exec(self, feedback_fn):
6568
    """Returns the tag list.
6569

6570
    """
6571
    cfg = self.cfg
6572
    tgts = [("/cluster", cfg.GetClusterInfo())]
6573
    ilist = cfg.GetAllInstancesInfo().values()
6574
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6575
    nlist = cfg.GetAllNodesInfo().values()
6576
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6577
    results = []
6578
    for path, target in tgts:
6579
      for tag in target.GetTags():
6580
        if self.re.search(tag):
6581
          results.append((path, tag))
6582
    return results
6583

    
6584

    
6585
class LUAddTags(TagsLU):
6586
  """Sets a tag on a given object.
6587

6588
  """
6589
  _OP_REQP = ["kind", "name", "tags"]
6590
  REQ_BGL = False
6591

    
6592
  def CheckPrereq(self):
6593
    """Check prerequisites.
6594

6595
    This checks the type and length of the tag name and value.
6596

6597
    """
6598
    TagsLU.CheckPrereq(self)
6599
    for tag in self.op.tags:
6600
      objects.TaggableObject.ValidateTag(tag)
6601

    
6602
  def Exec(self, feedback_fn):
6603
    """Sets the tag.
6604

6605
    """
6606
    try:
6607
      for tag in self.op.tags:
6608
        self.target.AddTag(tag)
6609
    except errors.TagError, err:
6610
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6611
    try:
6612
      self.cfg.Update(self.target)
6613
    except errors.ConfigurationError:
6614
      raise errors.OpRetryError("There has been a modification to the"
6615
                                " config file and the operation has been"
6616
                                " aborted. Please retry.")
6617

    
6618

    
6619
class LUDelTags(TagsLU):
6620
  """Delete a list of tags from a given object.
6621

6622
  """
6623
  _OP_REQP = ["kind", "name", "tags"]
6624
  REQ_BGL = False
6625

    
6626
  def CheckPrereq(self):
6627
    """Check prerequisites.
6628

6629
    This checks that we have the given tag.
6630

6631
    """
6632
    TagsLU.CheckPrereq(self)
6633
    for tag in self.op.tags:
6634
      objects.TaggableObject.ValidateTag(tag)
6635
    del_tags = frozenset(self.op.tags)
6636
    cur_tags = self.target.GetTags()
6637
    if not del_tags <= cur_tags:
6638
      diff_tags = del_tags - cur_tags
6639
      diff_names = ["'%s'" % tag for tag in diff_tags]
6640
      diff_names.sort()
6641
      raise errors.OpPrereqError("Tag(s) %s not found" %
6642
                                 (",".join(diff_names)))
6643

    
6644
  def Exec(self, feedback_fn):
6645
    """Remove the tag from the object.
6646

6647
    """
6648
    for tag in self.op.tags:
6649
      self.target.RemoveTag(tag)
6650
    try:
6651
      self.cfg.Update(self.target)
6652
    except errors.ConfigurationError:
6653
      raise errors.OpRetryError("There has been a modification to the"
6654
                                " config file and the operation has been"
6655
                                " aborted. Please retry.")
6656

    
6657

    
6658
class LUTestDelay(NoHooksLU):
6659
  """Sleep for a specified amount of time.
6660

6661
  This LU sleeps on the master and/or nodes for a specified amount of
6662
  time.
6663

6664
  """
6665
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6666
  REQ_BGL = False
6667

    
6668
  def ExpandNames(self):
6669
    """Expand names and set required locks.
6670

6671
    This expands the node list, if any.
6672

6673
    """
6674
    self.needed_locks = {}
6675
    if self.op.on_nodes:
6676
      # _GetWantedNodes can be used here, but is not always appropriate to use
6677
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6678
      # more information.
6679
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6680
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6681

    
6682
  def CheckPrereq(self):
6683
    """Check prerequisites.
6684

6685
    """
6686

    
6687
  def Exec(self, feedback_fn):
6688
    """Do the actual sleep.
6689

6690
    """
6691
    if self.op.on_master:
6692
      if not utils.TestDelay(self.op.duration):
6693
        raise errors.OpExecError("Error during master delay test")
6694
    if self.op.on_nodes:
6695
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6696
      if not result:
6697
        raise errors.OpExecError("Complete failure from rpc call")
6698
      for node, node_result in result.items():
6699
        node_result.Raise()
6700
        if not node_result.data:
6701
          raise errors.OpExecError("Failure during rpc call to node %s,"
6702
                                   " result: %s" % (node, node_result.data))
6703

    
6704

    
6705
class IAllocator(object):
6706
  """IAllocator framework.
6707

6708
  An IAllocator instance has three sets of attributes:
6709
    - cfg that is needed to query the cluster
6710
    - input data (all members of the _KEYS class attribute are required)
6711
    - four buffer attributes (in|out_data|text), that represent the
6712
      input (to the external script) in text and data structure format,
6713
      and the output from it, again in two formats
6714
    - the result variables from the script (success, info, nodes) for
6715
      easy usage
6716

6717
  """
6718
  _ALLO_KEYS = [
6719
    "mem_size", "disks", "disk_template",
6720
    "os", "tags", "nics", "vcpus", "hypervisor",
6721
    ]
6722
  _RELO_KEYS = [
6723
    "relocate_from",
6724
    ]
6725

    
6726
  def __init__(self, lu, mode, name, **kwargs):
6727
    self.lu = lu
6728
    # init buffer variables
6729
    self.in_text = self.out_text = self.in_data = self.out_data = None
6730
    # init all input fields so that pylint is happy
6731
    self.mode = mode
6732
    self.name = name
6733
    self.mem_size = self.disks = self.disk_template = None
6734
    self.os = self.tags = self.nics = self.vcpus = None
6735
    self.hypervisor = None
6736
    self.relocate_from = None
6737
    # computed fields
6738
    self.required_nodes = None
6739
    # init result fields
6740
    self.success = self.info = self.nodes = None
6741
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6742
      keyset = self._ALLO_KEYS
6743
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6744
      keyset = self._RELO_KEYS
6745
    else:
6746
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6747
                                   " IAllocator" % self.mode)
6748
    for key in kwargs:
6749
      if key not in keyset:
6750
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6751
                                     " IAllocator" % key)
6752
      setattr(self, key, kwargs[key])
6753
    for key in keyset:
6754
      if key not in kwargs:
6755
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6756
                                     " IAllocator" % key)
6757
    self._BuildInputData()
6758

    
6759
  def _ComputeClusterData(self):
6760
    """Compute the generic allocator input data.
6761

6762
    This is the data that is independent of the actual operation.
6763

6764
    """
6765
    cfg = self.lu.cfg
6766
    cluster_info = cfg.GetClusterInfo()
6767
    # cluster data
6768
    data = {
6769
      "version": constants.IALLOCATOR_VERSION,
6770
      "cluster_name": cfg.GetClusterName(),
6771
      "cluster_tags": list(cluster_info.GetTags()),
6772
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6773
      # we don't have job IDs
6774
      }
6775
    iinfo = cfg.GetAllInstancesInfo().values()
6776
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6777

    
6778
    # node data
6779
    node_results = {}
6780
    node_list = cfg.GetNodeList()
6781

    
6782
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6783
      hypervisor_name = self.hypervisor
6784
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6785
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6786

    
6787
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6788
                                           hypervisor_name)
6789
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6790
                       cluster_info.enabled_hypervisors)
6791
    for nname, nresult in node_data.items():
6792
      # first fill in static (config-based) values
6793
      ninfo = cfg.GetNodeInfo(nname)
6794
      pnr = {
6795
        "tags": list(ninfo.GetTags()),
6796
        "primary_ip": ninfo.primary_ip,
6797
        "secondary_ip": ninfo.secondary_ip,
6798
        "offline": ninfo.offline,
6799
        "drained": ninfo.drained,
6800
        "master_candidate": ninfo.master_candidate,
6801
        }
6802

    
6803
      if not ninfo.offline:
6804
        nresult.Raise()
6805
        if not isinstance(nresult.data, dict):
6806
          raise errors.OpExecError("Can't get data for node %s" % nname)
6807
        remote_info = nresult.data
6808
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6809
                     'vg_size', 'vg_free', 'cpu_total']:
6810
          if attr not in remote_info:
6811
            raise errors.OpExecError("Node '%s' didn't return attribute"
6812
                                     " '%s'" % (nname, attr))
6813
          try:
6814
            remote_info[attr] = int(remote_info[attr])
6815
          except ValueError, err:
6816
            raise errors.OpExecError("Node '%s' returned invalid value"
6817
                                     " for '%s': %s" % (nname, attr, err))
6818
        # compute memory used by primary instances
6819
        i_p_mem = i_p_up_mem = 0
6820
        for iinfo, beinfo in i_list:
6821
          if iinfo.primary_node == nname:
6822
            i_p_mem += beinfo[constants.BE_MEMORY]
6823
            if iinfo.name not in node_iinfo[nname].data:
6824
              i_used_mem = 0
6825
            else:
6826
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6827
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6828
            remote_info['memory_free'] -= max(0, i_mem_diff)
6829

    
6830
            if iinfo.admin_up:
6831
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6832

    
6833
        # compute memory used by instances
6834
        pnr_dyn = {
6835
          "total_memory": remote_info['memory_total'],
6836
          "reserved_memory": remote_info['memory_dom0'],
6837
          "free_memory": remote_info['memory_free'],
6838
          "total_disk": remote_info['vg_size'],
6839
          "free_disk": remote_info['vg_free'],
6840
          "total_cpus": remote_info['cpu_total'],
6841
          "i_pri_memory": i_p_mem,
6842
          "i_pri_up_memory": i_p_up_mem,
6843
          }
6844
        pnr.update(pnr_dyn)
6845

    
6846
      node_results[nname] = pnr
6847
    data["nodes"] = node_results
6848

    
6849
    # instance data
6850
    instance_data = {}
6851
    for iinfo, beinfo in i_list:
6852
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6853
                  for n in iinfo.nics]
6854
      pir = {
6855
        "tags": list(iinfo.GetTags()),
6856
        "admin_up": iinfo.admin_up,
6857
        "vcpus": beinfo[constants.BE_VCPUS],
6858
        "memory": beinfo[constants.BE_MEMORY],
6859
        "os": iinfo.os,
6860
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6861
        "nics": nic_data,
6862
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6863
        "disk_template": iinfo.disk_template,
6864
        "hypervisor": iinfo.hypervisor,
6865
        }
6866
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
6867
                                                 pir["disks"])
6868
      instance_data[iinfo.name] = pir
6869

    
6870
    data["instances"] = instance_data
6871

    
6872
    self.in_data = data
6873

    
6874
  def _AddNewInstance(self):
6875
    """Add new instance data to allocator structure.
6876

6877
    This in combination with _AllocatorGetClusterData will create the
6878
    correct structure needed as input for the allocator.
6879

6880
    The checks for the completeness of the opcode must have already been
6881
    done.
6882

6883
    """
6884
    data = self.in_data
6885

    
6886
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6887

    
6888
    if self.disk_template in constants.DTS_NET_MIRROR:
6889
      self.required_nodes = 2
6890
    else:
6891
      self.required_nodes = 1
6892
    request = {
6893
      "type": "allocate",
6894
      "name": self.name,
6895
      "disk_template": self.disk_template,
6896
      "tags": self.tags,
6897
      "os": self.os,
6898
      "vcpus": self.vcpus,
6899
      "memory": self.mem_size,
6900
      "disks": self.disks,
6901
      "disk_space_total": disk_space,
6902
      "nics": self.nics,
6903
      "required_nodes": self.required_nodes,
6904
      }
6905
    data["request"] = request
6906

    
6907
  def _AddRelocateInstance(self):
6908
    """Add relocate instance data to allocator structure.
6909

6910
    This in combination with _IAllocatorGetClusterData will create the
6911
    correct structure needed as input for the allocator.
6912

6913
    The checks for the completeness of the opcode must have already been
6914
    done.
6915

6916
    """
6917
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6918
    if instance is None:
6919
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6920
                                   " IAllocator" % self.name)
6921

    
6922
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6923
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6924

    
6925
    if len(instance.secondary_nodes) != 1:
6926
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6927

    
6928
    self.required_nodes = 1
6929
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6930
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6931

    
6932
    request = {
6933
      "type": "relocate",
6934
      "name": self.name,
6935
      "disk_space_total": disk_space,
6936
      "required_nodes": self.required_nodes,
6937
      "relocate_from": self.relocate_from,
6938
      }
6939
    self.in_data["request"] = request
6940

    
6941
  def _BuildInputData(self):
6942
    """Build input data structures.
6943

6944
    """
6945
    self._ComputeClusterData()
6946

    
6947
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6948
      self._AddNewInstance()
6949
    else:
6950
      self._AddRelocateInstance()
6951

    
6952
    self.in_text = serializer.Dump(self.in_data)
6953

    
6954
  def Run(self, name, validate=True, call_fn=None):
6955
    """Run an instance allocator and return the results.
6956

6957
    """
6958
    if call_fn is None:
6959
      call_fn = self.lu.rpc.call_iallocator_runner
6960
    data = self.in_text
6961

    
6962
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6963
    result.Raise()
6964

    
6965
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6966
      raise errors.OpExecError("Invalid result from master iallocator runner")
6967

    
6968
    rcode, stdout, stderr, fail = result.data
6969

    
6970
    if rcode == constants.IARUN_NOTFOUND:
6971
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6972
    elif rcode == constants.IARUN_FAILURE:
6973
      raise errors.OpExecError("Instance allocator call failed: %s,"
6974
                               " output: %s" % (fail, stdout+stderr))
6975
    self.out_text = stdout
6976
    if validate:
6977
      self._ValidateResult()
6978

    
6979
  def _ValidateResult(self):
6980
    """Process the allocator results.
6981

6982
    This will process and if successful save the result in
6983
    self.out_data and the other parameters.
6984

6985
    """
6986
    try:
6987
      rdict = serializer.Load(self.out_text)
6988
    except Exception, err:
6989
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6990

    
6991
    if not isinstance(rdict, dict):
6992
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6993

    
6994
    for key in "success", "info", "nodes":
6995
      if key not in rdict:
6996
        raise errors.OpExecError("Can't parse iallocator results:"
6997
                                 " missing key '%s'" % key)
6998
      setattr(self, key, rdict[key])
6999

    
7000
    if not isinstance(rdict["nodes"], list):
7001
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
7002
                               " is not a list")
7003
    self.out_data = rdict
7004

    
7005

    
7006
class LUTestAllocator(NoHooksLU):
7007
  """Run allocator tests.
7008

7009
  This LU runs the allocator tests
7010

7011
  """
7012
  _OP_REQP = ["direction", "mode", "name"]
7013

    
7014
  def CheckPrereq(self):
7015
    """Check prerequisites.
7016

7017
    This checks the opcode parameters depending on the director and mode test.
7018

7019
    """
7020
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7021
      for attr in ["name", "mem_size", "disks", "disk_template",
7022
                   "os", "tags", "nics", "vcpus"]:
7023
        if not hasattr(self.op, attr):
7024
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
7025
                                     attr)
7026
      iname = self.cfg.ExpandInstanceName(self.op.name)
7027
      if iname is not None:
7028
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
7029
                                   iname)
7030
      if not isinstance(self.op.nics, list):
7031
        raise errors.OpPrereqError("Invalid parameter 'nics'")
7032
      for row in self.op.nics:
7033
        if (not isinstance(row, dict) or
7034
            "mac" not in row or
7035
            "ip" not in row or
7036
            "bridge" not in row):
7037
          raise errors.OpPrereqError("Invalid contents of the"
7038
                                     " 'nics' parameter")
7039
      if not isinstance(self.op.disks, list):
7040
        raise errors.OpPrereqError("Invalid parameter 'disks'")
7041
      for row in self.op.disks:
7042
        if (not isinstance(row, dict) or
7043
            "size" not in row or
7044
            not isinstance(row["size"], int) or
7045
            "mode" not in row or
7046
            row["mode"] not in ['r', 'w']):
7047
          raise errors.OpPrereqError("Invalid contents of the"
7048
                                     " 'disks' parameter")
7049
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
7050
        self.op.hypervisor = self.cfg.GetHypervisorType()
7051
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
7052
      if not hasattr(self.op, "name"):
7053
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
7054
      fname = self.cfg.ExpandInstanceName(self.op.name)
7055
      if fname is None:
7056
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
7057
                                   self.op.name)
7058
      self.op.name = fname
7059
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
7060
    else:
7061
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
7062
                                 self.op.mode)
7063

    
7064
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
7065
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
7066
        raise errors.OpPrereqError("Missing allocator name")
7067
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
7068
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
7069
                                 self.op.direction)
7070

    
7071
  def Exec(self, feedback_fn):
7072
    """Run the allocator test.
7073

7074
    """
7075
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7076
      ial = IAllocator(self,
7077
                       mode=self.op.mode,
7078
                       name=self.op.name,
7079
                       mem_size=self.op.mem_size,
7080
                       disks=self.op.disks,
7081
                       disk_template=self.op.disk_template,
7082
                       os=self.op.os,
7083
                       tags=self.op.tags,
7084
                       nics=self.op.nics,
7085
                       vcpus=self.op.vcpus,
7086
                       hypervisor=self.op.hypervisor,
7087
                       )
7088
    else:
7089
      ial = IAllocator(self,
7090
                       mode=self.op.mode,
7091
                       name=self.op.name,
7092
                       relocate_from=list(self.relocate_from),
7093
                       )
7094

    
7095
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
7096
      result = ial.in_text
7097
    else:
7098
      ial.Run(self.op.allocator, validate=False)
7099
      result = ial.out_text
7100
    return result