Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ f1926756

History | View | Annotate | Download (242 kB)

1
#
2
#
3

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

    
21

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

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

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

    
36
from ganeti import ssh
37
from ganeti import utils
38
from ganeti import errors
39
from ganeti import hypervisor
40
from ganeti import locking
41
from ganeti import constants
42
from ganeti import objects
43
from ganeti import opcodes
44
from ganeti import serializer
45
from ganeti import ssconf
46

    
47

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

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

60
  Note that all commands require root permissions.
61

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

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

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

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

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

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

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

    
108
  ssh = property(fget=__GetSSH)
109

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

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

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

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

125
    """
126
    pass
127

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

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

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

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

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

149
    Examples::
150

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

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

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

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

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

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

188
    """
189

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

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

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

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

204
    """
205
    raise NotImplementedError
206

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

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

214
    """
215
    raise NotImplementedError
216

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

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

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

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

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

236
    """
237
    raise NotImplementedError
238

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

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

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

257
    """
258
    return lu_result
259

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
325
    del self.recalculate_locks[locking.LEVEL_NODE]
326

    
327

    
328
class NoHooksLU(LogicalUnit):
329
  """Simple LU which runs no hooks.
330

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

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

    
338

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

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

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

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

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

    
365
  return utils.NiceSort(wanted)
366

    
367

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

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

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

    
384
  if instances:
385
    wanted = []
386

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

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

    
397

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

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

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

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

    
416

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

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

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

    
430

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

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

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

    
442

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

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

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

    
454

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

459
  This builds the hook environment from individual variables.
460

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

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

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

    
513
  env["INSTANCE_NIC_COUNT"] = nic_count
514

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

    
523
  env["INSTANCE_DISK_COUNT"] = disk_count
524

    
525
  return env
526

    
527

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

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

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

    
560

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

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

    
576

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

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

    
590

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

594
  """
595
  _OP_REQP = []
596

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

600
    This checks whether the cluster is empty.
601

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

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

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

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

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

    
630

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

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

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

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

652
    Test list:
653

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

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

671
    """
672
    node = nodeinfo.name
673

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

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

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

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

    
694
    bad = False
695

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

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

    
716
    # checks config file checksum
717

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

    
745
    # checks ssh to any
746

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

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

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

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

    
793
    return bad
794

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

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

802
    """
803
    bad = False
804

    
805
    node_current = instanceconfig.primary_node
806

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

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

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

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

    
835
    return bad
836

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

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

843
    """
844
    bad = False
845

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

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

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

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

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

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

875
    """
876
    bad = False
877

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

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

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

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

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

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

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

    
924
    return env, [], all_nodes
925

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

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

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

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

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

    
960
    local_checksums = utils.FingerprintFiles(file_names)
961

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

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

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

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

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

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

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

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

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

    
1051
      node_instance[node] = idata
1052

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

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

    
1089
    node_vol_should = {}
1090

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

    
1099
      inst_config.MapLVsByNode(node_vol_should)
1100

    
1101
      instance_cfg[instance] = inst_config
1102

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

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

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

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

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

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

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

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

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

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

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

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

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

    
1177
    return not bad
1178

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

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

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

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

    
1226
      return lu_result
1227

    
1228

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

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

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

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

1246
    This has no prerequisites.
1247

1248
    """
1249
    pass
1250

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

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

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

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

    
1274
    if not nv_dict:
1275
      return result
1276

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

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

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

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

    
1312
    return result
1313

    
1314

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

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

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

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

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

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

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

    
1353
    self.op.name = new_name
1354

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

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

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

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

    
1374
      # update the known hosts file
1375
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
1376
      node_list = self.cfg.GetNodeList()
1377
      try:
1378
        node_list.remove(master)
1379
      except ValueError:
1380
        pass
1381
      result = self.rpc.call_upload_file(node_list,
1382
                                         constants.SSH_KNOWN_HOSTS_FILE)
1383
      for to_node, to_result in result.iteritems():
1384
        if to_result.failed or not to_result.data:
1385
          logging.error("Copy of file %s to node %s failed",
1386
                        constants.SSH_KNOWN_HOSTS_FILE, to_node)
1387

    
1388
    finally:
1389
      result = self.rpc.call_node_start_master(master, False)
1390
      if result.failed or not result.data:
1391
        self.LogWarning("Could not re-enable the master role on"
1392
                        " the master, please restart manually.")
1393

    
1394

    
1395
def _RecursiveCheckIfLVMBased(disk):
1396
  """Check if the given disk or its children are lvm-based.
1397

1398
  @type disk: L{objects.Disk}
1399
  @param disk: the disk to check
1400
  @rtype: booleean
1401
  @return: boolean indicating whether a LD_LV dev_type was found or not
1402

1403
  """
1404
  if disk.children:
1405
    for chdisk in disk.children:
1406
      if _RecursiveCheckIfLVMBased(chdisk):
1407
        return True
1408
  return disk.dev_type == constants.LD_LV
1409

    
1410

    
1411
class LUSetClusterParams(LogicalUnit):
1412
  """Change the parameters of the cluster.
1413

1414
  """
1415
  HPATH = "cluster-modify"
1416
  HTYPE = constants.HTYPE_CLUSTER
1417
  _OP_REQP = []
1418
  REQ_BGL = False
1419

    
1420
  def CheckParameters(self):
1421
    """Check parameters
1422

1423
    """
1424
    if not hasattr(self.op, "candidate_pool_size"):
1425
      self.op.candidate_pool_size = None
1426
    if self.op.candidate_pool_size is not None:
1427
      try:
1428
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1429
      except ValueError, err:
1430
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1431
                                   str(err))
1432
      if self.op.candidate_pool_size < 1:
1433
        raise errors.OpPrereqError("At least one master candidate needed")
1434

    
1435
  def ExpandNames(self):
1436
    # FIXME: in the future maybe other cluster params won't require checking on
1437
    # all nodes to be modified.
1438
    self.needed_locks = {
1439
      locking.LEVEL_NODE: locking.ALL_SET,
1440
    }
1441
    self.share_locks[locking.LEVEL_NODE] = 1
1442

    
1443
  def BuildHooksEnv(self):
1444
    """Build hooks env.
1445

1446
    """
1447
    env = {
1448
      "OP_TARGET": self.cfg.GetClusterName(),
1449
      "NEW_VG_NAME": self.op.vg_name,
1450
      }
1451
    mn = self.cfg.GetMasterNode()
1452
    return env, [mn], [mn]
1453

    
1454
  def CheckPrereq(self):
1455
    """Check prerequisites.
1456

1457
    This checks whether the given params don't conflict and
1458
    if the given volume group is valid.
1459

1460
    """
1461
    if self.op.vg_name is not None and not self.op.vg_name:
1462
      instances = self.cfg.GetAllInstancesInfo().values()
1463
      for inst in instances:
1464
        for disk in inst.disks:
1465
          if _RecursiveCheckIfLVMBased(disk):
1466
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1467
                                       " lvm-based instances exist")
1468

    
1469
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1470

    
1471
    # if vg_name not None, checks given volume group on all nodes
1472
    if self.op.vg_name:
1473
      vglist = self.rpc.call_vg_list(node_list)
1474
      for node in node_list:
1475
        if vglist[node].failed:
1476
          # ignoring down node
1477
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1478
          continue
1479
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1480
                                              self.op.vg_name,
1481
                                              constants.MIN_VG_SIZE)
1482
        if vgstatus:
1483
          raise errors.OpPrereqError("Error on node '%s': %s" %
1484
                                     (node, vgstatus))
1485

    
1486
    self.cluster = cluster = self.cfg.GetClusterInfo()
1487
    # validate beparams changes
1488
    if self.op.beparams:
1489
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1490
      self.new_beparams = cluster.FillDict(
1491
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1492

    
1493
    # hypervisor list/parameters
1494
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1495
    if self.op.hvparams:
1496
      if not isinstance(self.op.hvparams, dict):
1497
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1498
      for hv_name, hv_dict in self.op.hvparams.items():
1499
        if hv_name not in self.new_hvparams:
1500
          self.new_hvparams[hv_name] = hv_dict
1501
        else:
1502
          self.new_hvparams[hv_name].update(hv_dict)
1503

    
1504
    if self.op.enabled_hypervisors is not None:
1505
      self.hv_list = self.op.enabled_hypervisors
1506
    else:
1507
      self.hv_list = cluster.enabled_hypervisors
1508

    
1509
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1510
      # either the enabled list has changed, or the parameters have, validate
1511
      for hv_name, hv_params in self.new_hvparams.items():
1512
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1513
            (self.op.enabled_hypervisors and
1514
             hv_name in self.op.enabled_hypervisors)):
1515
          # either this is a new hypervisor, or its parameters have changed
1516
          hv_class = hypervisor.GetHypervisor(hv_name)
1517
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1518
          hv_class.CheckParameterSyntax(hv_params)
1519
          _CheckHVParams(self, node_list, hv_name, hv_params)
1520

    
1521
  def Exec(self, feedback_fn):
1522
    """Change the parameters of the cluster.
1523

1524
    """
1525
    if self.op.vg_name is not None:
1526
      new_volume = self.op.vg_name
1527
      if not new_volume:
1528
        new_volume = None
1529
      if new_volume != self.cfg.GetVGName():
1530
        self.cfg.SetVGName(new_volume)
1531
      else:
1532
        feedback_fn("Cluster LVM configuration already in desired"
1533
                    " state, not changing")
1534
    if self.op.hvparams:
1535
      self.cluster.hvparams = self.new_hvparams
1536
    if self.op.enabled_hypervisors is not None:
1537
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1538
    if self.op.beparams:
1539
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1540
    if self.op.candidate_pool_size is not None:
1541
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1542

    
1543
    self.cfg.Update(self.cluster)
1544

    
1545
    # we want to update nodes after the cluster so that if any errors
1546
    # happen, we have recorded and saved the cluster info
1547
    if self.op.candidate_pool_size is not None:
1548
      _AdjustCandidatePool(self)
1549

    
1550

    
1551
class LURedistributeConfig(NoHooksLU):
1552
  """Force the redistribution of cluster configuration.
1553

1554
  This is a very simple LU.
1555

1556
  """
1557
  _OP_REQP = []
1558
  REQ_BGL = False
1559

    
1560
  def ExpandNames(self):
1561
    self.needed_locks = {
1562
      locking.LEVEL_NODE: locking.ALL_SET,
1563
    }
1564
    self.share_locks[locking.LEVEL_NODE] = 1
1565

    
1566
  def CheckPrereq(self):
1567
    """Check prerequisites.
1568

1569
    """
1570

    
1571
  def Exec(self, feedback_fn):
1572
    """Redistribute the configuration.
1573

1574
    """
1575
    self.cfg.Update(self.cfg.GetClusterInfo())
1576

    
1577

    
1578
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1579
  """Sleep and poll for an instance's disk to sync.
1580

1581
  """
1582
  if not instance.disks:
1583
    return True
1584

    
1585
  if not oneshot:
1586
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1587

    
1588
  node = instance.primary_node
1589

    
1590
  for dev in instance.disks:
1591
    lu.cfg.SetDiskID(dev, node)
1592

    
1593
  retries = 0
1594
  while True:
1595
    max_time = 0
1596
    done = True
1597
    cumul_degraded = False
1598
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1599
    if rstats.failed or not rstats.data:
1600
      lu.LogWarning("Can't get any data from node %s", node)
1601
      retries += 1
1602
      if retries >= 10:
1603
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1604
                                 " aborting." % node)
1605
      time.sleep(6)
1606
      continue
1607
    rstats = rstats.data
1608
    retries = 0
1609
    for i, mstat in enumerate(rstats):
1610
      if mstat is None:
1611
        lu.LogWarning("Can't compute data for node %s/%s",
1612
                           node, instance.disks[i].iv_name)
1613
        continue
1614
      # we ignore the ldisk parameter
1615
      perc_done, est_time, is_degraded, _ = mstat
1616
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1617
      if perc_done is not None:
1618
        done = False
1619
        if est_time is not None:
1620
          rem_time = "%d estimated seconds remaining" % est_time
1621
          max_time = est_time
1622
        else:
1623
          rem_time = "no time estimate"
1624
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1625
                        (instance.disks[i].iv_name, perc_done, rem_time))
1626
    if done or oneshot:
1627
      break
1628

    
1629
    time.sleep(min(60, max_time))
1630

    
1631
  if done:
1632
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1633
  return not cumul_degraded
1634

    
1635

    
1636
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1637
  """Check that mirrors are not degraded.
1638

1639
  The ldisk parameter, if True, will change the test from the
1640
  is_degraded attribute (which represents overall non-ok status for
1641
  the device(s)) to the ldisk (representing the local storage status).
1642

1643
  """
1644
  lu.cfg.SetDiskID(dev, node)
1645
  if ldisk:
1646
    idx = 6
1647
  else:
1648
    idx = 5
1649

    
1650
  result = True
1651
  if on_primary or dev.AssembleOnSecondary():
1652
    rstats = lu.rpc.call_blockdev_find(node, dev)
1653
    msg = rstats.RemoteFailMsg()
1654
    if msg:
1655
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1656
      result = False
1657
    elif not rstats.payload:
1658
      lu.LogWarning("Can't find disk on node %s", node)
1659
      result = False
1660
    else:
1661
      result = result and (not rstats.payload[idx])
1662
  if dev.children:
1663
    for child in dev.children:
1664
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1665

    
1666
  return result
1667

    
1668

    
1669
class LUDiagnoseOS(NoHooksLU):
1670
  """Logical unit for OS diagnose/query.
1671

1672
  """
1673
  _OP_REQP = ["output_fields", "names"]
1674
  REQ_BGL = False
1675
  _FIELDS_STATIC = utils.FieldSet()
1676
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1677

    
1678
  def ExpandNames(self):
1679
    if self.op.names:
1680
      raise errors.OpPrereqError("Selective OS query not supported")
1681

    
1682
    _CheckOutputFields(static=self._FIELDS_STATIC,
1683
                       dynamic=self._FIELDS_DYNAMIC,
1684
                       selected=self.op.output_fields)
1685

    
1686
    # Lock all nodes, in shared mode
1687
    # Temporary removal of locks, should be reverted later
1688
    # TODO: reintroduce locks when they are lighter-weight
1689
    self.needed_locks = {}
1690
    #self.share_locks[locking.LEVEL_NODE] = 1
1691
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1692

    
1693
  def CheckPrereq(self):
1694
    """Check prerequisites.
1695

1696
    """
1697

    
1698
  @staticmethod
1699
  def _DiagnoseByOS(node_list, rlist):
1700
    """Remaps a per-node return list into an a per-os per-node dictionary
1701

1702
    @param node_list: a list with the names of all nodes
1703
    @param rlist: a map with node names as keys and OS objects as values
1704

1705
    @rtype: dict
1706
    @return: a dictionary with osnames as keys and as value another map, with
1707
        nodes as keys and list of OS objects as values, eg::
1708

1709
          {"debian-etch": {"node1": [<object>,...],
1710
                           "node2": [<object>,]}
1711
          }
1712

1713
    """
1714
    all_os = {}
1715
    # we build here the list of nodes that didn't fail the RPC (at RPC
1716
    # level), so that nodes with a non-responding node daemon don't
1717
    # make all OSes invalid
1718
    good_nodes = [node_name for node_name in rlist
1719
                  if not rlist[node_name].failed]
1720
    for node_name, nr in rlist.iteritems():
1721
      if nr.failed or not nr.data:
1722
        continue
1723
      for os_obj in nr.data:
1724
        if os_obj.name not in all_os:
1725
          # build a list of nodes for this os containing empty lists
1726
          # for each node in node_list
1727
          all_os[os_obj.name] = {}
1728
          for nname in good_nodes:
1729
            all_os[os_obj.name][nname] = []
1730
        all_os[os_obj.name][node_name].append(os_obj)
1731
    return all_os
1732

    
1733
  def Exec(self, feedback_fn):
1734
    """Compute the list of OSes.
1735

1736
    """
1737
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1738
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1739
    if node_data == False:
1740
      raise errors.OpExecError("Can't gather the list of OSes")
1741
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1742
    output = []
1743
    for os_name, os_data in pol.iteritems():
1744
      row = []
1745
      for field in self.op.output_fields:
1746
        if field == "name":
1747
          val = os_name
1748
        elif field == "valid":
1749
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1750
        elif field == "node_status":
1751
          val = {}
1752
          for node_name, nos_list in os_data.iteritems():
1753
            val[node_name] = [(v.status, v.path) for v in nos_list]
1754
        else:
1755
          raise errors.ParameterError(field)
1756
        row.append(val)
1757
      output.append(row)
1758

    
1759
    return output
1760

    
1761

    
1762
class LURemoveNode(LogicalUnit):
1763
  """Logical unit for removing a node.
1764

1765
  """
1766
  HPATH = "node-remove"
1767
  HTYPE = constants.HTYPE_NODE
1768
  _OP_REQP = ["node_name"]
1769

    
1770
  def BuildHooksEnv(self):
1771
    """Build hooks env.
1772

1773
    This doesn't run on the target node in the pre phase as a failed
1774
    node would then be impossible to remove.
1775

1776
    """
1777
    env = {
1778
      "OP_TARGET": self.op.node_name,
1779
      "NODE_NAME": self.op.node_name,
1780
      }
1781
    all_nodes = self.cfg.GetNodeList()
1782
    all_nodes.remove(self.op.node_name)
1783
    return env, all_nodes, all_nodes
1784

    
1785
  def CheckPrereq(self):
1786
    """Check prerequisites.
1787

1788
    This checks:
1789
     - the node exists in the configuration
1790
     - it does not have primary or secondary instances
1791
     - it's not the master
1792

1793
    Any errors are signalled by raising errors.OpPrereqError.
1794

1795
    """
1796
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1797
    if node is None:
1798
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1799

    
1800
    instance_list = self.cfg.GetInstanceList()
1801

    
1802
    masternode = self.cfg.GetMasterNode()
1803
    if node.name == masternode:
1804
      raise errors.OpPrereqError("Node is the master node,"
1805
                                 " you need to failover first.")
1806

    
1807
    for instance_name in instance_list:
1808
      instance = self.cfg.GetInstanceInfo(instance_name)
1809
      if node.name in instance.all_nodes:
1810
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1811
                                   " please remove first." % instance_name)
1812
    self.op.node_name = node.name
1813
    self.node = node
1814

    
1815
  def Exec(self, feedback_fn):
1816
    """Removes the node from the cluster.
1817

1818
    """
1819
    node = self.node
1820
    logging.info("Stopping the node daemon and removing configs from node %s",
1821
                 node.name)
1822

    
1823
    self.context.RemoveNode(node.name)
1824

    
1825
    self.rpc.call_node_leave_cluster(node.name)
1826

    
1827
    # Promote nodes to master candidate as needed
1828
    _AdjustCandidatePool(self)
1829

    
1830

    
1831
class LUQueryNodes(NoHooksLU):
1832
  """Logical unit for querying nodes.
1833

1834
  """
1835
  _OP_REQP = ["output_fields", "names", "use_locking"]
1836
  REQ_BGL = False
1837
  _FIELDS_DYNAMIC = utils.FieldSet(
1838
    "dtotal", "dfree",
1839
    "mtotal", "mnode", "mfree",
1840
    "bootid",
1841
    "ctotal", "cnodes", "csockets",
1842
    )
1843

    
1844
  _FIELDS_STATIC = utils.FieldSet(
1845
    "name", "pinst_cnt", "sinst_cnt",
1846
    "pinst_list", "sinst_list",
1847
    "pip", "sip", "tags",
1848
    "serial_no",
1849
    "master_candidate",
1850
    "master",
1851
    "offline",
1852
    "drained",
1853
    )
1854

    
1855
  def ExpandNames(self):
1856
    _CheckOutputFields(static=self._FIELDS_STATIC,
1857
                       dynamic=self._FIELDS_DYNAMIC,
1858
                       selected=self.op.output_fields)
1859

    
1860
    self.needed_locks = {}
1861
    self.share_locks[locking.LEVEL_NODE] = 1
1862

    
1863
    if self.op.names:
1864
      self.wanted = _GetWantedNodes(self, self.op.names)
1865
    else:
1866
      self.wanted = locking.ALL_SET
1867

    
1868
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1869
    self.do_locking = self.do_node_query and self.op.use_locking
1870
    if self.do_locking:
1871
      # if we don't request only static fields, we need to lock the nodes
1872
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1873

    
1874

    
1875
  def CheckPrereq(self):
1876
    """Check prerequisites.
1877

1878
    """
1879
    # The validation of the node list is done in the _GetWantedNodes,
1880
    # if non empty, and if empty, there's no validation to do
1881
    pass
1882

    
1883
  def Exec(self, feedback_fn):
1884
    """Computes the list of nodes and their attributes.
1885

1886
    """
1887
    all_info = self.cfg.GetAllNodesInfo()
1888
    if self.do_locking:
1889
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1890
    elif self.wanted != locking.ALL_SET:
1891
      nodenames = self.wanted
1892
      missing = set(nodenames).difference(all_info.keys())
1893
      if missing:
1894
        raise errors.OpExecError(
1895
          "Some nodes were removed before retrieving their data: %s" % missing)
1896
    else:
1897
      nodenames = all_info.keys()
1898

    
1899
    nodenames = utils.NiceSort(nodenames)
1900
    nodelist = [all_info[name] for name in nodenames]
1901

    
1902
    # begin data gathering
1903

    
1904
    if self.do_node_query:
1905
      live_data = {}
1906
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1907
                                          self.cfg.GetHypervisorType())
1908
      for name in nodenames:
1909
        nodeinfo = node_data[name]
1910
        if not nodeinfo.failed and nodeinfo.data:
1911
          nodeinfo = nodeinfo.data
1912
          fn = utils.TryConvert
1913
          live_data[name] = {
1914
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1915
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1916
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1917
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1918
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1919
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1920
            "bootid": nodeinfo.get('bootid', None),
1921
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
1922
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
1923
            }
1924
        else:
1925
          live_data[name] = {}
1926
    else:
1927
      live_data = dict.fromkeys(nodenames, {})
1928

    
1929
    node_to_primary = dict([(name, set()) for name in nodenames])
1930
    node_to_secondary = dict([(name, set()) for name in nodenames])
1931

    
1932
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1933
                             "sinst_cnt", "sinst_list"))
1934
    if inst_fields & frozenset(self.op.output_fields):
1935
      instancelist = self.cfg.GetInstanceList()
1936

    
1937
      for instance_name in instancelist:
1938
        inst = self.cfg.GetInstanceInfo(instance_name)
1939
        if inst.primary_node in node_to_primary:
1940
          node_to_primary[inst.primary_node].add(inst.name)
1941
        for secnode in inst.secondary_nodes:
1942
          if secnode in node_to_secondary:
1943
            node_to_secondary[secnode].add(inst.name)
1944

    
1945
    master_node = self.cfg.GetMasterNode()
1946

    
1947
    # end data gathering
1948

    
1949
    output = []
1950
    for node in nodelist:
1951
      node_output = []
1952
      for field in self.op.output_fields:
1953
        if field == "name":
1954
          val = node.name
1955
        elif field == "pinst_list":
1956
          val = list(node_to_primary[node.name])
1957
        elif field == "sinst_list":
1958
          val = list(node_to_secondary[node.name])
1959
        elif field == "pinst_cnt":
1960
          val = len(node_to_primary[node.name])
1961
        elif field == "sinst_cnt":
1962
          val = len(node_to_secondary[node.name])
1963
        elif field == "pip":
1964
          val = node.primary_ip
1965
        elif field == "sip":
1966
          val = node.secondary_ip
1967
        elif field == "tags":
1968
          val = list(node.GetTags())
1969
        elif field == "serial_no":
1970
          val = node.serial_no
1971
        elif field == "master_candidate":
1972
          val = node.master_candidate
1973
        elif field == "master":
1974
          val = node.name == master_node
1975
        elif field == "offline":
1976
          val = node.offline
1977
        elif field == "drained":
1978
          val = node.drained
1979
        elif self._FIELDS_DYNAMIC.Matches(field):
1980
          val = live_data[node.name].get(field, None)
1981
        else:
1982
          raise errors.ParameterError(field)
1983
        node_output.append(val)
1984
      output.append(node_output)
1985

    
1986
    return output
1987

    
1988

    
1989
class LUQueryNodeVolumes(NoHooksLU):
1990
  """Logical unit for getting volumes on node(s).
1991

1992
  """
1993
  _OP_REQP = ["nodes", "output_fields"]
1994
  REQ_BGL = False
1995
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
1996
  _FIELDS_STATIC = utils.FieldSet("node")
1997

    
1998
  def ExpandNames(self):
1999
    _CheckOutputFields(static=self._FIELDS_STATIC,
2000
                       dynamic=self._FIELDS_DYNAMIC,
2001
                       selected=self.op.output_fields)
2002

    
2003
    self.needed_locks = {}
2004
    self.share_locks[locking.LEVEL_NODE] = 1
2005
    if not self.op.nodes:
2006
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2007
    else:
2008
      self.needed_locks[locking.LEVEL_NODE] = \
2009
        _GetWantedNodes(self, self.op.nodes)
2010

    
2011
  def CheckPrereq(self):
2012
    """Check prerequisites.
2013

2014
    This checks that the fields required are valid output fields.
2015

2016
    """
2017
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2018

    
2019
  def Exec(self, feedback_fn):
2020
    """Computes the list of nodes and their attributes.
2021

2022
    """
2023
    nodenames = self.nodes
2024
    volumes = self.rpc.call_node_volumes(nodenames)
2025

    
2026
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2027
             in self.cfg.GetInstanceList()]
2028

    
2029
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2030

    
2031
    output = []
2032
    for node in nodenames:
2033
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2034
        continue
2035

    
2036
      node_vols = volumes[node].data[:]
2037
      node_vols.sort(key=lambda vol: vol['dev'])
2038

    
2039
      for vol in node_vols:
2040
        node_output = []
2041
        for field in self.op.output_fields:
2042
          if field == "node":
2043
            val = node
2044
          elif field == "phys":
2045
            val = vol['dev']
2046
          elif field == "vg":
2047
            val = vol['vg']
2048
          elif field == "name":
2049
            val = vol['name']
2050
          elif field == "size":
2051
            val = int(float(vol['size']))
2052
          elif field == "instance":
2053
            for inst in ilist:
2054
              if node not in lv_by_node[inst]:
2055
                continue
2056
              if vol['name'] in lv_by_node[inst][node]:
2057
                val = inst.name
2058
                break
2059
            else:
2060
              val = '-'
2061
          else:
2062
            raise errors.ParameterError(field)
2063
          node_output.append(str(val))
2064

    
2065
        output.append(node_output)
2066

    
2067
    return output
2068

    
2069

    
2070
class LUAddNode(LogicalUnit):
2071
  """Logical unit for adding node to the cluster.
2072

2073
  """
2074
  HPATH = "node-add"
2075
  HTYPE = constants.HTYPE_NODE
2076
  _OP_REQP = ["node_name"]
2077

    
2078
  def BuildHooksEnv(self):
2079
    """Build hooks env.
2080

2081
    This will run on all nodes before, and on all nodes + the new node after.
2082

2083
    """
2084
    env = {
2085
      "OP_TARGET": self.op.node_name,
2086
      "NODE_NAME": self.op.node_name,
2087
      "NODE_PIP": self.op.primary_ip,
2088
      "NODE_SIP": self.op.secondary_ip,
2089
      }
2090
    nodes_0 = self.cfg.GetNodeList()
2091
    nodes_1 = nodes_0 + [self.op.node_name, ]
2092
    return env, nodes_0, nodes_1
2093

    
2094
  def CheckPrereq(self):
2095
    """Check prerequisites.
2096

2097
    This checks:
2098
     - the new node is not already in the config
2099
     - it is resolvable
2100
     - its parameters (single/dual homed) matches the cluster
2101

2102
    Any errors are signalled by raising errors.OpPrereqError.
2103

2104
    """
2105
    node_name = self.op.node_name
2106
    cfg = self.cfg
2107

    
2108
    dns_data = utils.HostInfo(node_name)
2109

    
2110
    node = dns_data.name
2111
    primary_ip = self.op.primary_ip = dns_data.ip
2112
    secondary_ip = getattr(self.op, "secondary_ip", None)
2113
    if secondary_ip is None:
2114
      secondary_ip = primary_ip
2115
    if not utils.IsValidIP(secondary_ip):
2116
      raise errors.OpPrereqError("Invalid secondary IP given")
2117
    self.op.secondary_ip = secondary_ip
2118

    
2119
    node_list = cfg.GetNodeList()
2120
    if not self.op.readd and node in node_list:
2121
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2122
                                 node)
2123
    elif self.op.readd and node not in node_list:
2124
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2125

    
2126
    for existing_node_name in node_list:
2127
      existing_node = cfg.GetNodeInfo(existing_node_name)
2128

    
2129
      if self.op.readd and node == existing_node_name:
2130
        if (existing_node.primary_ip != primary_ip or
2131
            existing_node.secondary_ip != secondary_ip):
2132
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2133
                                     " address configuration as before")
2134
        continue
2135

    
2136
      if (existing_node.primary_ip == primary_ip or
2137
          existing_node.secondary_ip == primary_ip or
2138
          existing_node.primary_ip == secondary_ip or
2139
          existing_node.secondary_ip == secondary_ip):
2140
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2141
                                   " existing node %s" % existing_node.name)
2142

    
2143
    # check that the type of the node (single versus dual homed) is the
2144
    # same as for the master
2145
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2146
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2147
    newbie_singlehomed = secondary_ip == primary_ip
2148
    if master_singlehomed != newbie_singlehomed:
2149
      if master_singlehomed:
2150
        raise errors.OpPrereqError("The master has no private ip but the"
2151
                                   " new node has one")
2152
      else:
2153
        raise errors.OpPrereqError("The master has a private ip but the"
2154
                                   " new node doesn't have one")
2155

    
2156
    # checks reachablity
2157
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2158
      raise errors.OpPrereqError("Node not reachable by ping")
2159

    
2160
    if not newbie_singlehomed:
2161
      # check reachability from my secondary ip to newbie's secondary ip
2162
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2163
                           source=myself.secondary_ip):
2164
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2165
                                   " based ping to noded port")
2166

    
2167
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2168
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2169
    master_candidate = mc_now < cp_size
2170

    
2171
    self.new_node = objects.Node(name=node,
2172
                                 primary_ip=primary_ip,
2173
                                 secondary_ip=secondary_ip,
2174
                                 master_candidate=master_candidate,
2175
                                 offline=False, drained=False)
2176

    
2177
  def Exec(self, feedback_fn):
2178
    """Adds the new node to the cluster.
2179

2180
    """
2181
    new_node = self.new_node
2182
    node = new_node.name
2183

    
2184
    # check connectivity
2185
    result = self.rpc.call_version([node])[node]
2186
    result.Raise()
2187
    if result.data:
2188
      if constants.PROTOCOL_VERSION == result.data:
2189
        logging.info("Communication to node %s fine, sw version %s match",
2190
                     node, result.data)
2191
      else:
2192
        raise errors.OpExecError("Version mismatch master version %s,"
2193
                                 " node version %s" %
2194
                                 (constants.PROTOCOL_VERSION, result.data))
2195
    else:
2196
      raise errors.OpExecError("Cannot get version from the new node")
2197

    
2198
    # setup ssh on node
2199
    logging.info("Copy ssh key to node %s", node)
2200
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2201
    keyarray = []
2202
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2203
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2204
                priv_key, pub_key]
2205

    
2206
    for i in keyfiles:
2207
      f = open(i, 'r')
2208
      try:
2209
        keyarray.append(f.read())
2210
      finally:
2211
        f.close()
2212

    
2213
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2214
                                    keyarray[2],
2215
                                    keyarray[3], keyarray[4], keyarray[5])
2216

    
2217
    msg = result.RemoteFailMsg()
2218
    if msg:
2219
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2220
                               " new node: %s" % msg)
2221

    
2222
    # Add node to our /etc/hosts, and add key to known_hosts
2223
    utils.AddHostToEtcHosts(new_node.name)
2224

    
2225
    if new_node.secondary_ip != new_node.primary_ip:
2226
      result = self.rpc.call_node_has_ip_address(new_node.name,
2227
                                                 new_node.secondary_ip)
2228
      if result.failed or not result.data:
2229
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2230
                                 " you gave (%s). Please fix and re-run this"
2231
                                 " command." % new_node.secondary_ip)
2232

    
2233
    node_verify_list = [self.cfg.GetMasterNode()]
2234
    node_verify_param = {
2235
      'nodelist': [node],
2236
      # TODO: do a node-net-test as well?
2237
    }
2238

    
2239
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2240
                                       self.cfg.GetClusterName())
2241
    for verifier in node_verify_list:
2242
      if result[verifier].failed or not result[verifier].data:
2243
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2244
                                 " for remote verification" % verifier)
2245
      if result[verifier].data['nodelist']:
2246
        for failed in result[verifier].data['nodelist']:
2247
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2248
                      (verifier, result[verifier].data['nodelist'][failed]))
2249
        raise errors.OpExecError("ssh/hostname verification failed.")
2250

    
2251
    # Distribute updated /etc/hosts and known_hosts to all nodes,
2252
    # including the node just added
2253
    myself = self.cfg.GetNodeInfo(self.cfg.GetMasterNode())
2254
    dist_nodes = self.cfg.GetNodeList()
2255
    if not self.op.readd:
2256
      dist_nodes.append(node)
2257
    if myself.name in dist_nodes:
2258
      dist_nodes.remove(myself.name)
2259

    
2260
    logging.debug("Copying hosts and known_hosts to all nodes")
2261
    for fname in (constants.ETC_HOSTS, constants.SSH_KNOWN_HOSTS_FILE):
2262
      result = self.rpc.call_upload_file(dist_nodes, fname)
2263
      for to_node, to_result in result.iteritems():
2264
        if to_result.failed or not to_result.data:
2265
          logging.error("Copy of file %s to node %s failed", fname, to_node)
2266

    
2267
    to_copy = []
2268
    enabled_hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
2269
    if constants.HTS_COPY_VNC_PASSWORD.intersection(enabled_hypervisors):
2270
      to_copy.append(constants.VNC_PASSWORD_FILE)
2271

    
2272
    for fname in to_copy:
2273
      result = self.rpc.call_upload_file([node], fname)
2274
      if result[node].failed or not result[node]:
2275
        logging.error("Could not copy file %s to node %s", fname, node)
2276

    
2277
    if self.op.readd:
2278
      self.context.ReaddNode(new_node)
2279
    else:
2280
      self.context.AddNode(new_node)
2281

    
2282

    
2283
class LUSetNodeParams(LogicalUnit):
2284
  """Modifies the parameters of a node.
2285

2286
  """
2287
  HPATH = "node-modify"
2288
  HTYPE = constants.HTYPE_NODE
2289
  _OP_REQP = ["node_name"]
2290
  REQ_BGL = False
2291

    
2292
  def CheckArguments(self):
2293
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2294
    if node_name is None:
2295
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2296
    self.op.node_name = node_name
2297
    _CheckBooleanOpField(self.op, 'master_candidate')
2298
    _CheckBooleanOpField(self.op, 'offline')
2299
    _CheckBooleanOpField(self.op, 'drained')
2300
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2301
    if all_mods.count(None) == 3:
2302
      raise errors.OpPrereqError("Please pass at least one modification")
2303
    if all_mods.count(True) > 1:
2304
      raise errors.OpPrereqError("Can't set the node into more than one"
2305
                                 " state at the same time")
2306

    
2307
  def ExpandNames(self):
2308
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2309

    
2310
  def BuildHooksEnv(self):
2311
    """Build hooks env.
2312

2313
    This runs on the master node.
2314

2315
    """
2316
    env = {
2317
      "OP_TARGET": self.op.node_name,
2318
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2319
      "OFFLINE": str(self.op.offline),
2320
      "DRAINED": str(self.op.drained),
2321
      }
2322
    nl = [self.cfg.GetMasterNode(),
2323
          self.op.node_name]
2324
    return env, nl, nl
2325

    
2326
  def CheckPrereq(self):
2327
    """Check prerequisites.
2328

2329
    This only checks the instance list against the existing names.
2330

2331
    """
2332
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2333

    
2334
    if ((self.op.master_candidate == False or self.op.offline == True or
2335
         self.op.drained == True) and node.master_candidate):
2336
      # we will demote the node from master_candidate
2337
      if self.op.node_name == self.cfg.GetMasterNode():
2338
        raise errors.OpPrereqError("The master node has to be a"
2339
                                   " master candidate, online and not drained")
2340
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2341
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2342
      if num_candidates <= cp_size:
2343
        msg = ("Not enough master candidates (desired"
2344
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2345
        if self.op.force:
2346
          self.LogWarning(msg)
2347
        else:
2348
          raise errors.OpPrereqError(msg)
2349

    
2350
    if (self.op.master_candidate == True and
2351
        ((node.offline and not self.op.offline == False) or
2352
         (node.drained and not self.op.drained == False))):
2353
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2354
                                 " to master_candidate" % node.name)
2355

    
2356
    return
2357

    
2358
  def Exec(self, feedback_fn):
2359
    """Modifies a node.
2360

2361
    """
2362
    node = self.node
2363

    
2364
    result = []
2365
    changed_mc = False
2366

    
2367
    if self.op.offline is not None:
2368
      node.offline = self.op.offline
2369
      result.append(("offline", str(self.op.offline)))
2370
      if self.op.offline == True:
2371
        if node.master_candidate:
2372
          node.master_candidate = False
2373
          changed_mc = True
2374
          result.append(("master_candidate", "auto-demotion due to offline"))
2375
        if node.drained:
2376
          node.drained = False
2377
          result.append(("drained", "clear drained status due to offline"))
2378

    
2379
    if self.op.master_candidate is not None:
2380
      node.master_candidate = self.op.master_candidate
2381
      changed_mc = True
2382
      result.append(("master_candidate", str(self.op.master_candidate)))
2383
      if self.op.master_candidate == False:
2384
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2385
        msg = rrc.RemoteFailMsg()
2386
        if msg:
2387
          self.LogWarning("Node failed to demote itself: %s" % msg)
2388

    
2389
    if self.op.drained is not None:
2390
      node.drained = self.op.drained
2391
      result.append(("drained", str(self.op.drained)))
2392
      if self.op.drained == True:
2393
        if node.master_candidate:
2394
          node.master_candidate = False
2395
          changed_mc = True
2396
          result.append(("master_candidate", "auto-demotion due to drain"))
2397
        if node.offline:
2398
          node.offline = False
2399
          result.append(("offline", "clear offline status due to drain"))
2400

    
2401
    # this will trigger configuration file update, if needed
2402
    self.cfg.Update(node)
2403
    # this will trigger job queue propagation or cleanup
2404
    if changed_mc:
2405
      self.context.ReaddNode(node)
2406

    
2407
    return result
2408

    
2409

    
2410
class LUQueryClusterInfo(NoHooksLU):
2411
  """Query cluster configuration.
2412

2413
  """
2414
  _OP_REQP = []
2415
  REQ_BGL = False
2416

    
2417
  def ExpandNames(self):
2418
    self.needed_locks = {}
2419

    
2420
  def CheckPrereq(self):
2421
    """No prerequsites needed for this LU.
2422

2423
    """
2424
    pass
2425

    
2426
  def Exec(self, feedback_fn):
2427
    """Return cluster config.
2428

2429
    """
2430
    cluster = self.cfg.GetClusterInfo()
2431
    result = {
2432
      "software_version": constants.RELEASE_VERSION,
2433
      "protocol_version": constants.PROTOCOL_VERSION,
2434
      "config_version": constants.CONFIG_VERSION,
2435
      "os_api_version": constants.OS_API_VERSION,
2436
      "export_version": constants.EXPORT_VERSION,
2437
      "architecture": (platform.architecture()[0], platform.machine()),
2438
      "name": cluster.cluster_name,
2439
      "master": cluster.master_node,
2440
      "default_hypervisor": cluster.default_hypervisor,
2441
      "enabled_hypervisors": cluster.enabled_hypervisors,
2442
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2443
                        for hypervisor in cluster.enabled_hypervisors]),
2444
      "beparams": cluster.beparams,
2445
      "candidate_pool_size": cluster.candidate_pool_size,
2446
      "default_bridge": cluster.default_bridge,
2447
      "master_netdev": cluster.master_netdev,
2448
      "volume_group_name": cluster.volume_group_name,
2449
      "file_storage_dir": cluster.file_storage_dir,
2450
      }
2451

    
2452
    return result
2453

    
2454

    
2455
class LUQueryConfigValues(NoHooksLU):
2456
  """Return configuration values.
2457

2458
  """
2459
  _OP_REQP = []
2460
  REQ_BGL = False
2461
  _FIELDS_DYNAMIC = utils.FieldSet()
2462
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2463

    
2464
  def ExpandNames(self):
2465
    self.needed_locks = {}
2466

    
2467
    _CheckOutputFields(static=self._FIELDS_STATIC,
2468
                       dynamic=self._FIELDS_DYNAMIC,
2469
                       selected=self.op.output_fields)
2470

    
2471
  def CheckPrereq(self):
2472
    """No prerequisites.
2473

2474
    """
2475
    pass
2476

    
2477
  def Exec(self, feedback_fn):
2478
    """Dump a representation of the cluster config to the standard output.
2479

2480
    """
2481
    values = []
2482
    for field in self.op.output_fields:
2483
      if field == "cluster_name":
2484
        entry = self.cfg.GetClusterName()
2485
      elif field == "master_node":
2486
        entry = self.cfg.GetMasterNode()
2487
      elif field == "drain_flag":
2488
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2489
      else:
2490
        raise errors.ParameterError(field)
2491
      values.append(entry)
2492
    return values
2493

    
2494

    
2495
class LUActivateInstanceDisks(NoHooksLU):
2496
  """Bring up an instance's disks.
2497

2498
  """
2499
  _OP_REQP = ["instance_name"]
2500
  REQ_BGL = False
2501

    
2502
  def ExpandNames(self):
2503
    self._ExpandAndLockInstance()
2504
    self.needed_locks[locking.LEVEL_NODE] = []
2505
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2506

    
2507
  def DeclareLocks(self, level):
2508
    if level == locking.LEVEL_NODE:
2509
      self._LockInstancesNodes()
2510

    
2511
  def CheckPrereq(self):
2512
    """Check prerequisites.
2513

2514
    This checks that the instance is in the cluster.
2515

2516
    """
2517
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2518
    assert self.instance is not None, \
2519
      "Cannot retrieve locked instance %s" % self.op.instance_name
2520
    _CheckNodeOnline(self, self.instance.primary_node)
2521

    
2522
  def Exec(self, feedback_fn):
2523
    """Activate the disks.
2524

2525
    """
2526
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2527
    if not disks_ok:
2528
      raise errors.OpExecError("Cannot activate block devices")
2529

    
2530
    return disks_info
2531

    
2532

    
2533
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2534
  """Prepare the block devices for an instance.
2535

2536
  This sets up the block devices on all nodes.
2537

2538
  @type lu: L{LogicalUnit}
2539
  @param lu: the logical unit on whose behalf we execute
2540
  @type instance: L{objects.Instance}
2541
  @param instance: the instance for whose disks we assemble
2542
  @type ignore_secondaries: boolean
2543
  @param ignore_secondaries: if true, errors on secondary nodes
2544
      won't result in an error return from the function
2545
  @return: False if the operation failed, otherwise a list of
2546
      (host, instance_visible_name, node_visible_name)
2547
      with the mapping from node devices to instance devices
2548

2549
  """
2550
  device_info = []
2551
  disks_ok = True
2552
  iname = instance.name
2553
  # With the two passes mechanism we try to reduce the window of
2554
  # opportunity for the race condition of switching DRBD to primary
2555
  # before handshaking occured, but we do not eliminate it
2556

    
2557
  # The proper fix would be to wait (with some limits) until the
2558
  # connection has been made and drbd transitions from WFConnection
2559
  # into any other network-connected state (Connected, SyncTarget,
2560
  # SyncSource, etc.)
2561

    
2562
  # 1st pass, assemble on all nodes in secondary mode
2563
  for inst_disk in instance.disks:
2564
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2565
      lu.cfg.SetDiskID(node_disk, node)
2566
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2567
      msg = result.RemoteFailMsg()
2568
      if msg:
2569
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2570
                           " (is_primary=False, pass=1): %s",
2571
                           inst_disk.iv_name, node, msg)
2572
        if not ignore_secondaries:
2573
          disks_ok = False
2574

    
2575
  # FIXME: race condition on drbd migration to primary
2576

    
2577
  # 2nd pass, do only the primary node
2578
  for inst_disk in instance.disks:
2579
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2580
      if node != instance.primary_node:
2581
        continue
2582
      lu.cfg.SetDiskID(node_disk, node)
2583
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2584
      msg = result.RemoteFailMsg()
2585
      if msg:
2586
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2587
                           " (is_primary=True, pass=2): %s",
2588
                           inst_disk.iv_name, node, msg)
2589
        disks_ok = False
2590
    device_info.append((instance.primary_node, inst_disk.iv_name,
2591
                        result.payload))
2592

    
2593
  # leave the disks configured for the primary node
2594
  # this is a workaround that would be fixed better by
2595
  # improving the logical/physical id handling
2596
  for disk in instance.disks:
2597
    lu.cfg.SetDiskID(disk, instance.primary_node)
2598

    
2599
  return disks_ok, device_info
2600

    
2601

    
2602
def _StartInstanceDisks(lu, instance, force):
2603
  """Start the disks of an instance.
2604

2605
  """
2606
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2607
                                           ignore_secondaries=force)
2608
  if not disks_ok:
2609
    _ShutdownInstanceDisks(lu, instance)
2610
    if force is not None and not force:
2611
      lu.proc.LogWarning("", hint="If the message above refers to a"
2612
                         " secondary node,"
2613
                         " you can retry the operation using '--force'.")
2614
    raise errors.OpExecError("Disk consistency error")
2615

    
2616

    
2617
class LUDeactivateInstanceDisks(NoHooksLU):
2618
  """Shutdown an instance's disks.
2619

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

    
2624
  def ExpandNames(self):
2625
    self._ExpandAndLockInstance()
2626
    self.needed_locks[locking.LEVEL_NODE] = []
2627
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2628

    
2629
  def DeclareLocks(self, level):
2630
    if level == locking.LEVEL_NODE:
2631
      self._LockInstancesNodes()
2632

    
2633
  def CheckPrereq(self):
2634
    """Check prerequisites.
2635

2636
    This checks that the instance is in the cluster.
2637

2638
    """
2639
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2640
    assert self.instance is not None, \
2641
      "Cannot retrieve locked instance %s" % self.op.instance_name
2642

    
2643
  def Exec(self, feedback_fn):
2644
    """Deactivate the disks
2645

2646
    """
2647
    instance = self.instance
2648
    _SafeShutdownInstanceDisks(self, instance)
2649

    
2650

    
2651
def _SafeShutdownInstanceDisks(lu, instance):
2652
  """Shutdown block devices of an instance.
2653

2654
  This function checks if an instance is running, before calling
2655
  _ShutdownInstanceDisks.
2656

2657
  """
2658
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2659
                                      [instance.hypervisor])
2660
  ins_l = ins_l[instance.primary_node]
2661
  if ins_l.failed or not isinstance(ins_l.data, list):
2662
    raise errors.OpExecError("Can't contact node '%s'" %
2663
                             instance.primary_node)
2664

    
2665
  if instance.name in ins_l.data:
2666
    raise errors.OpExecError("Instance is running, can't shutdown"
2667
                             " block devices.")
2668

    
2669
  _ShutdownInstanceDisks(lu, instance)
2670

    
2671

    
2672
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2673
  """Shutdown block devices of an instance.
2674

2675
  This does the shutdown on all nodes of the instance.
2676

2677
  If the ignore_primary is false, errors on the primary node are
2678
  ignored.
2679

2680
  """
2681
  all_result = True
2682
  for disk in instance.disks:
2683
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2684
      lu.cfg.SetDiskID(top_disk, node)
2685
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2686
      msg = result.RemoteFailMsg()
2687
      if msg:
2688
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2689
                      disk.iv_name, node, msg)
2690
        if not ignore_primary or node != instance.primary_node:
2691
          all_result = False
2692
  return all_result
2693

    
2694

    
2695
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2696
  """Checks if a node has enough free memory.
2697

2698
  This function check if a given node has the needed amount of free
2699
  memory. In case the node has less memory or we cannot get the
2700
  information from the node, this function raise an OpPrereqError
2701
  exception.
2702

2703
  @type lu: C{LogicalUnit}
2704
  @param lu: a logical unit from which we get configuration data
2705
  @type node: C{str}
2706
  @param node: the node to check
2707
  @type reason: C{str}
2708
  @param reason: string to use in the error message
2709
  @type requested: C{int}
2710
  @param requested: the amount of memory in MiB to check for
2711
  @type hypervisor_name: C{str}
2712
  @param hypervisor_name: the hypervisor to ask for memory stats
2713
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2714
      we cannot check the node
2715

2716
  """
2717
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2718
  nodeinfo[node].Raise()
2719
  free_mem = nodeinfo[node].data.get('memory_free')
2720
  if not isinstance(free_mem, int):
2721
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2722
                             " was '%s'" % (node, free_mem))
2723
  if requested > free_mem:
2724
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2725
                             " needed %s MiB, available %s MiB" %
2726
                             (node, reason, requested, free_mem))
2727

    
2728

    
2729
class LUStartupInstance(LogicalUnit):
2730
  """Starts an instance.
2731

2732
  """
2733
  HPATH = "instance-start"
2734
  HTYPE = constants.HTYPE_INSTANCE
2735
  _OP_REQP = ["instance_name", "force"]
2736
  REQ_BGL = False
2737

    
2738
  def ExpandNames(self):
2739
    self._ExpandAndLockInstance()
2740

    
2741
  def BuildHooksEnv(self):
2742
    """Build hooks env.
2743

2744
    This runs on master, primary and secondary nodes of the instance.
2745

2746
    """
2747
    env = {
2748
      "FORCE": self.op.force,
2749
      }
2750
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2751
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2752
    return env, nl, nl
2753

    
2754
  def CheckPrereq(self):
2755
    """Check prerequisites.
2756

2757
    This checks that the instance is in the cluster.
2758

2759
    """
2760
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2761
    assert self.instance is not None, \
2762
      "Cannot retrieve locked instance %s" % self.op.instance_name
2763

    
2764
    _CheckNodeOnline(self, instance.primary_node)
2765

    
2766
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2767
    # check bridges existance
2768
    _CheckInstanceBridgesExist(self, instance)
2769

    
2770
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2771
                                              instance.name,
2772
                                              instance.hypervisor)
2773
    remote_info.Raise()
2774
    if not remote_info.data:
2775
      _CheckNodeFreeMemory(self, instance.primary_node,
2776
                           "starting instance %s" % instance.name,
2777
                           bep[constants.BE_MEMORY], instance.hypervisor)
2778

    
2779
  def Exec(self, feedback_fn):
2780
    """Start the instance.
2781

2782
    """
2783
    instance = self.instance
2784
    force = self.op.force
2785

    
2786
    self.cfg.MarkInstanceUp(instance.name)
2787

    
2788
    node_current = instance.primary_node
2789

    
2790
    _StartInstanceDisks(self, instance, force)
2791

    
2792
    result = self.rpc.call_instance_start(node_current, instance)
2793
    msg = result.RemoteFailMsg()
2794
    if msg:
2795
      _ShutdownInstanceDisks(self, instance)
2796
      raise errors.OpExecError("Could not start instance: %s" % msg)
2797

    
2798

    
2799
class LURebootInstance(LogicalUnit):
2800
  """Reboot an instance.
2801

2802
  """
2803
  HPATH = "instance-reboot"
2804
  HTYPE = constants.HTYPE_INSTANCE
2805
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2806
  REQ_BGL = False
2807

    
2808
  def ExpandNames(self):
2809
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2810
                                   constants.INSTANCE_REBOOT_HARD,
2811
                                   constants.INSTANCE_REBOOT_FULL]:
2812
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2813
                                  (constants.INSTANCE_REBOOT_SOFT,
2814
                                   constants.INSTANCE_REBOOT_HARD,
2815
                                   constants.INSTANCE_REBOOT_FULL))
2816
    self._ExpandAndLockInstance()
2817

    
2818
  def BuildHooksEnv(self):
2819
    """Build hooks env.
2820

2821
    This runs on master, primary and secondary nodes of the instance.
2822

2823
    """
2824
    env = {
2825
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2826
      "REBOOT_TYPE": self.op.reboot_type,
2827
      }
2828
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2829
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2830
    return env, nl, nl
2831

    
2832
  def CheckPrereq(self):
2833
    """Check prerequisites.
2834

2835
    This checks that the instance is in the cluster.
2836

2837
    """
2838
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2839
    assert self.instance is not None, \
2840
      "Cannot retrieve locked instance %s" % self.op.instance_name
2841

    
2842
    _CheckNodeOnline(self, instance.primary_node)
2843

    
2844
    # check bridges existance
2845
    _CheckInstanceBridgesExist(self, instance)
2846

    
2847
  def Exec(self, feedback_fn):
2848
    """Reboot the instance.
2849

2850
    """
2851
    instance = self.instance
2852
    ignore_secondaries = self.op.ignore_secondaries
2853
    reboot_type = self.op.reboot_type
2854

    
2855
    node_current = instance.primary_node
2856

    
2857
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2858
                       constants.INSTANCE_REBOOT_HARD]:
2859
      for disk in instance.disks:
2860
        self.cfg.SetDiskID(disk, node_current)
2861
      result = self.rpc.call_instance_reboot(node_current, instance,
2862
                                             reboot_type)
2863
      msg = result.RemoteFailMsg()
2864
      if msg:
2865
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2866
    else:
2867
      result = self.rpc.call_instance_shutdown(node_current, instance)
2868
      msg = result.RemoteFailMsg()
2869
      if msg:
2870
        raise errors.OpExecError("Could not shutdown instance for"
2871
                                 " full reboot: %s" % msg)
2872
      _ShutdownInstanceDisks(self, instance)
2873
      _StartInstanceDisks(self, instance, ignore_secondaries)
2874
      result = self.rpc.call_instance_start(node_current, instance)
2875
      msg = result.RemoteFailMsg()
2876
      if msg:
2877
        _ShutdownInstanceDisks(self, instance)
2878
        raise errors.OpExecError("Could not start instance for"
2879
                                 " full reboot: %s" % msg)
2880

    
2881
    self.cfg.MarkInstanceUp(instance.name)
2882

    
2883

    
2884
class LUShutdownInstance(LogicalUnit):
2885
  """Shutdown an instance.
2886

2887
  """
2888
  HPATH = "instance-stop"
2889
  HTYPE = constants.HTYPE_INSTANCE
2890
  _OP_REQP = ["instance_name"]
2891
  REQ_BGL = False
2892

    
2893
  def ExpandNames(self):
2894
    self._ExpandAndLockInstance()
2895

    
2896
  def BuildHooksEnv(self):
2897
    """Build hooks env.
2898

2899
    This runs on master, primary and secondary nodes of the instance.
2900

2901
    """
2902
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2903
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2904
    return env, nl, nl
2905

    
2906
  def CheckPrereq(self):
2907
    """Check prerequisites.
2908

2909
    This checks that the instance is in the cluster.
2910

2911
    """
2912
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2913
    assert self.instance is not None, \
2914
      "Cannot retrieve locked instance %s" % self.op.instance_name
2915
    _CheckNodeOnline(self, self.instance.primary_node)
2916

    
2917
  def Exec(self, feedback_fn):
2918
    """Shutdown the instance.
2919

2920
    """
2921
    instance = self.instance
2922
    node_current = instance.primary_node
2923
    self.cfg.MarkInstanceDown(instance.name)
2924
    result = self.rpc.call_instance_shutdown(node_current, instance)
2925
    msg = result.RemoteFailMsg()
2926
    if msg:
2927
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
2928

    
2929
    _ShutdownInstanceDisks(self, instance)
2930

    
2931

    
2932
class LUReinstallInstance(LogicalUnit):
2933
  """Reinstall an instance.
2934

2935
  """
2936
  HPATH = "instance-reinstall"
2937
  HTYPE = constants.HTYPE_INSTANCE
2938
  _OP_REQP = ["instance_name"]
2939
  REQ_BGL = False
2940

    
2941
  def ExpandNames(self):
2942
    self._ExpandAndLockInstance()
2943

    
2944
  def BuildHooksEnv(self):
2945
    """Build hooks env.
2946

2947
    This runs on master, primary and secondary nodes of the instance.
2948

2949
    """
2950
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2951
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2952
    return env, nl, nl
2953

    
2954
  def CheckPrereq(self):
2955
    """Check prerequisites.
2956

2957
    This checks that the instance is in the cluster and is not running.
2958

2959
    """
2960
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2961
    assert instance is not None, \
2962
      "Cannot retrieve locked instance %s" % self.op.instance_name
2963
    _CheckNodeOnline(self, instance.primary_node)
2964

    
2965
    if instance.disk_template == constants.DT_DISKLESS:
2966
      raise errors.OpPrereqError("Instance '%s' has no disks" %
2967
                                 self.op.instance_name)
2968
    if instance.admin_up:
2969
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
2970
                                 self.op.instance_name)
2971
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2972
                                              instance.name,
2973
                                              instance.hypervisor)
2974
    if remote_info.failed or remote_info.data:
2975
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
2976
                                 (self.op.instance_name,
2977
                                  instance.primary_node))
2978

    
2979
    self.op.os_type = getattr(self.op, "os_type", None)
2980
    if self.op.os_type is not None:
2981
      # OS verification
2982
      pnode = self.cfg.GetNodeInfo(
2983
        self.cfg.ExpandNodeName(instance.primary_node))
2984
      if pnode is None:
2985
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
2986
                                   self.op.pnode)
2987
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
2988
      result.Raise()
2989
      if not isinstance(result.data, objects.OS):
2990
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
2991
                                   " primary node"  % self.op.os_type)
2992

    
2993
    self.instance = instance
2994

    
2995
  def Exec(self, feedback_fn):
2996
    """Reinstall the instance.
2997

2998
    """
2999
    inst = self.instance
3000

    
3001
    if self.op.os_type is not None:
3002
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3003
      inst.os = self.op.os_type
3004
      self.cfg.Update(inst)
3005

    
3006
    _StartInstanceDisks(self, inst, None)
3007
    try:
3008
      feedback_fn("Running the instance OS create scripts...")
3009
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3010
      msg = result.RemoteFailMsg()
3011
      if msg:
3012
        raise errors.OpExecError("Could not install OS for instance %s"
3013
                                 " on node %s: %s" %
3014
                                 (inst.name, inst.primary_node, msg))
3015
    finally:
3016
      _ShutdownInstanceDisks(self, inst)
3017

    
3018

    
3019
class LURenameInstance(LogicalUnit):
3020
  """Rename an instance.
3021

3022
  """
3023
  HPATH = "instance-rename"
3024
  HTYPE = constants.HTYPE_INSTANCE
3025
  _OP_REQP = ["instance_name", "new_name"]
3026

    
3027
  def BuildHooksEnv(self):
3028
    """Build hooks env.
3029

3030
    This runs on master, primary and secondary nodes of the instance.
3031

3032
    """
3033
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3034
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3035
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3036
    return env, nl, nl
3037

    
3038
  def CheckPrereq(self):
3039
    """Check prerequisites.
3040

3041
    This checks that the instance is in the cluster and is not running.
3042

3043
    """
3044
    instance = self.cfg.GetInstanceInfo(
3045
      self.cfg.ExpandInstanceName(self.op.instance_name))
3046
    if instance is None:
3047
      raise errors.OpPrereqError("Instance '%s' not known" %
3048
                                 self.op.instance_name)
3049
    _CheckNodeOnline(self, instance.primary_node)
3050

    
3051
    if instance.admin_up:
3052
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3053
                                 self.op.instance_name)
3054
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3055
                                              instance.name,
3056
                                              instance.hypervisor)
3057
    remote_info.Raise()
3058
    if remote_info.data:
3059
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3060
                                 (self.op.instance_name,
3061
                                  instance.primary_node))
3062
    self.instance = instance
3063

    
3064
    # new name verification
3065
    name_info = utils.HostInfo(self.op.new_name)
3066

    
3067
    self.op.new_name = new_name = name_info.name
3068
    instance_list = self.cfg.GetInstanceList()
3069
    if new_name in instance_list:
3070
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3071
                                 new_name)
3072

    
3073
    if not getattr(self.op, "ignore_ip", False):
3074
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3075
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3076
                                   (name_info.ip, new_name))
3077

    
3078

    
3079
  def Exec(self, feedback_fn):
3080
    """Reinstall the instance.
3081

3082
    """
3083
    inst = self.instance
3084
    old_name = inst.name
3085

    
3086
    if inst.disk_template == constants.DT_FILE:
3087
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3088

    
3089
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3090
    # Change the instance lock. This is definitely safe while we hold the BGL
3091
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3092
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3093

    
3094
    # re-read the instance from the configuration after rename
3095
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3096

    
3097
    if inst.disk_template == constants.DT_FILE:
3098
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3099
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3100
                                                     old_file_storage_dir,
3101
                                                     new_file_storage_dir)
3102
      result.Raise()
3103
      if not result.data:
3104
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3105
                                 " directory '%s' to '%s' (but the instance"
3106
                                 " has been renamed in Ganeti)" % (
3107
                                 inst.primary_node, old_file_storage_dir,
3108
                                 new_file_storage_dir))
3109

    
3110
      if not result.data[0]:
3111
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3112
                                 " (but the instance has been renamed in"
3113
                                 " Ganeti)" % (old_file_storage_dir,
3114
                                               new_file_storage_dir))
3115

    
3116
    _StartInstanceDisks(self, inst, None)
3117
    try:
3118
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3119
                                                 old_name)
3120
      msg = result.RemoteFailMsg()
3121
      if msg:
3122
        msg = ("Could not run OS rename script for instance %s on node %s"
3123
               " (but the instance has been renamed in Ganeti): %s" %
3124
               (inst.name, inst.primary_node, msg))
3125
        self.proc.LogWarning(msg)
3126
    finally:
3127
      _ShutdownInstanceDisks(self, inst)
3128

    
3129

    
3130
class LURemoveInstance(LogicalUnit):
3131
  """Remove an instance.
3132

3133
  """
3134
  HPATH = "instance-remove"
3135
  HTYPE = constants.HTYPE_INSTANCE
3136
  _OP_REQP = ["instance_name", "ignore_failures"]
3137
  REQ_BGL = False
3138

    
3139
  def ExpandNames(self):
3140
    self._ExpandAndLockInstance()
3141
    self.needed_locks[locking.LEVEL_NODE] = []
3142
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3143

    
3144
  def DeclareLocks(self, level):
3145
    if level == locking.LEVEL_NODE:
3146
      self._LockInstancesNodes()
3147

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

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

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

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

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

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

    
3168
  def Exec(self, feedback_fn):
3169
    """Remove the instance.
3170

3171
    """
3172
    instance = self.instance
3173
    logging.info("Shutting down instance %s on node %s",
3174
                 instance.name, instance.primary_node)
3175

    
3176
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3177
    msg = result.RemoteFailMsg()
3178
    if msg:
3179
      if self.op.ignore_failures:
3180
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3181
      else:
3182
        raise errors.OpExecError("Could not shutdown instance %s on"
3183
                                 " node %s: %s" %
3184
                                 (instance.name, instance.primary_node, msg))
3185

    
3186
    logging.info("Removing block devices for instance %s", instance.name)
3187

    
3188
    if not _RemoveDisks(self, instance):
3189
      if self.op.ignore_failures:
3190
        feedback_fn("Warning: can't remove instance's disks")
3191
      else:
3192
        raise errors.OpExecError("Can't remove instance's disks")
3193

    
3194
    logging.info("Removing instance %s out of cluster config", instance.name)
3195

    
3196
    self.cfg.RemoveInstance(instance.name)
3197
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3198

    
3199

    
3200
class LUQueryInstances(NoHooksLU):
3201
  """Logical unit for querying instances.
3202

3203
  """
3204
  _OP_REQP = ["output_fields", "names", "use_locking"]
3205
  REQ_BGL = False
3206
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3207
                                    "admin_state",
3208
                                    "disk_template", "ip", "mac", "bridge",
3209
                                    "sda_size", "sdb_size", "vcpus", "tags",
3210
                                    "network_port", "beparams",
3211
                                    r"(disk)\.(size)/([0-9]+)",
3212
                                    r"(disk)\.(sizes)", "disk_usage",
3213
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3214
                                    r"(nic)\.(macs|ips|bridges)",
3215
                                    r"(disk|nic)\.(count)",
3216
                                    "serial_no", "hypervisor", "hvparams",] +
3217
                                  ["hv/%s" % name
3218
                                   for name in constants.HVS_PARAMETERS] +
3219
                                  ["be/%s" % name
3220
                                   for name in constants.BES_PARAMETERS])
3221
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3222

    
3223

    
3224
  def ExpandNames(self):
3225
    _CheckOutputFields(static=self._FIELDS_STATIC,
3226
                       dynamic=self._FIELDS_DYNAMIC,
3227
                       selected=self.op.output_fields)
3228

    
3229
    self.needed_locks = {}
3230
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3231
    self.share_locks[locking.LEVEL_NODE] = 1
3232

    
3233
    if self.op.names:
3234
      self.wanted = _GetWantedInstances(self, self.op.names)
3235
    else:
3236
      self.wanted = locking.ALL_SET
3237

    
3238
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3239
    self.do_locking = self.do_node_query and self.op.use_locking
3240
    if self.do_locking:
3241
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3242
      self.needed_locks[locking.LEVEL_NODE] = []
3243
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3244

    
3245
  def DeclareLocks(self, level):
3246
    if level == locking.LEVEL_NODE and self.do_locking:
3247
      self._LockInstancesNodes()
3248

    
3249
  def CheckPrereq(self):
3250
    """Check prerequisites.
3251

3252
    """
3253
    pass
3254

    
3255
  def Exec(self, feedback_fn):
3256
    """Computes the list of nodes and their attributes.
3257

3258
    """
3259
    all_info = self.cfg.GetAllInstancesInfo()
3260
    if self.wanted == locking.ALL_SET:
3261
      # caller didn't specify instance names, so ordering is not important
3262
      if self.do_locking:
3263
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3264
      else:
3265
        instance_names = all_info.keys()
3266
      instance_names = utils.NiceSort(instance_names)
3267
    else:
3268
      # caller did specify names, so we must keep the ordering
3269
      if self.do_locking:
3270
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3271
      else:
3272
        tgt_set = all_info.keys()
3273
      missing = set(self.wanted).difference(tgt_set)
3274
      if missing:
3275
        raise errors.OpExecError("Some instances were removed before"
3276
                                 " retrieving their data: %s" % missing)
3277
      instance_names = self.wanted
3278

    
3279
    instance_list = [all_info[iname] for iname in instance_names]
3280

    
3281
    # begin data gathering
3282

    
3283
    nodes = frozenset([inst.primary_node for inst in instance_list])
3284
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3285

    
3286
    bad_nodes = []
3287
    off_nodes = []
3288
    if self.do_node_query:
3289
      live_data = {}
3290
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3291
      for name in nodes:
3292
        result = node_data[name]
3293
        if result.offline:
3294
          # offline nodes will be in both lists
3295
          off_nodes.append(name)
3296
        if result.failed:
3297
          bad_nodes.append(name)
3298
        else:
3299
          if result.data:
3300
            live_data.update(result.data)
3301
            # else no instance is alive
3302
    else:
3303
      live_data = dict([(name, {}) for name in instance_names])
3304

    
3305
    # end data gathering
3306

    
3307
    HVPREFIX = "hv/"
3308
    BEPREFIX = "be/"
3309
    output = []
3310
    for instance in instance_list:
3311
      iout = []
3312
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3313
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3314
      for field in self.op.output_fields:
3315
        st_match = self._FIELDS_STATIC.Matches(field)
3316
        if field == "name":
3317
          val = instance.name
3318
        elif field == "os":
3319
          val = instance.os
3320
        elif field == "pnode":
3321
          val = instance.primary_node
3322
        elif field == "snodes":
3323
          val = list(instance.secondary_nodes)
3324
        elif field == "admin_state":
3325
          val = instance.admin_up
3326
        elif field == "oper_state":
3327
          if instance.primary_node in bad_nodes:
3328
            val = None
3329
          else:
3330
            val = bool(live_data.get(instance.name))
3331
        elif field == "status":
3332
          if instance.primary_node in off_nodes:
3333
            val = "ERROR_nodeoffline"
3334
          elif instance.primary_node in bad_nodes:
3335
            val = "ERROR_nodedown"
3336
          else:
3337
            running = bool(live_data.get(instance.name))
3338
            if running:
3339
              if instance.admin_up:
3340
                val = "running"
3341
              else:
3342
                val = "ERROR_up"
3343
            else:
3344
              if instance.admin_up:
3345
                val = "ERROR_down"
3346
              else:
3347
                val = "ADMIN_down"
3348
        elif field == "oper_ram":
3349
          if instance.primary_node in bad_nodes:
3350
            val = None
3351
          elif instance.name in live_data:
3352
            val = live_data[instance.name].get("memory", "?")
3353
          else:
3354
            val = "-"
3355
        elif field == "disk_template":
3356
          val = instance.disk_template
3357
        elif field == "ip":
3358
          val = instance.nics[0].ip
3359
        elif field == "bridge":
3360
          val = instance.nics[0].bridge
3361
        elif field == "mac":
3362
          val = instance.nics[0].mac
3363
        elif field == "sda_size" or field == "sdb_size":
3364
          idx = ord(field[2]) - ord('a')
3365
          try:
3366
            val = instance.FindDisk(idx).size
3367
          except errors.OpPrereqError:
3368
            val = None
3369
        elif field == "disk_usage": # total disk usage per node
3370
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3371
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3372
        elif field == "tags":
3373
          val = list(instance.GetTags())
3374
        elif field == "serial_no":
3375
          val = instance.serial_no
3376
        elif field == "network_port":
3377
          val = instance.network_port
3378
        elif field == "hypervisor":
3379
          val = instance.hypervisor
3380
        elif field == "hvparams":
3381
          val = i_hv
3382
        elif (field.startswith(HVPREFIX) and
3383
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3384
          val = i_hv.get(field[len(HVPREFIX):], None)
3385
        elif field == "beparams":
3386
          val = i_be
3387
        elif (field.startswith(BEPREFIX) and
3388
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3389
          val = i_be.get(field[len(BEPREFIX):], None)
3390
        elif st_match and st_match.groups():
3391
          # matches a variable list
3392
          st_groups = st_match.groups()
3393
          if st_groups and st_groups[0] == "disk":
3394
            if st_groups[1] == "count":
3395
              val = len(instance.disks)
3396
            elif st_groups[1] == "sizes":
3397
              val = [disk.size for disk in instance.disks]
3398
            elif st_groups[1] == "size":
3399
              try:
3400
                val = instance.FindDisk(st_groups[2]).size
3401
              except errors.OpPrereqError:
3402
                val = None
3403
            else:
3404
              assert False, "Unhandled disk parameter"
3405
          elif st_groups[0] == "nic":
3406
            if st_groups[1] == "count":
3407
              val = len(instance.nics)
3408
            elif st_groups[1] == "macs":
3409
              val = [nic.mac for nic in instance.nics]
3410
            elif st_groups[1] == "ips":
3411
              val = [nic.ip for nic in instance.nics]
3412
            elif st_groups[1] == "bridges":
3413
              val = [nic.bridge for nic in instance.nics]
3414
            else:
3415
              # index-based item
3416
              nic_idx = int(st_groups[2])
3417
              if nic_idx >= len(instance.nics):
3418
                val = None
3419
              else:
3420
                if st_groups[1] == "mac":
3421
                  val = instance.nics[nic_idx].mac
3422
                elif st_groups[1] == "ip":
3423
                  val = instance.nics[nic_idx].ip
3424
                elif st_groups[1] == "bridge":
3425
                  val = instance.nics[nic_idx].bridge
3426
                else:
3427
                  assert False, "Unhandled NIC parameter"
3428
          else:
3429
            assert False, "Unhandled variable parameter"
3430
        else:
3431
          raise errors.ParameterError(field)
3432
        iout.append(val)
3433
      output.append(iout)
3434

    
3435
    return output
3436

    
3437

    
3438
class LUFailoverInstance(LogicalUnit):
3439
  """Failover an instance.
3440

3441
  """
3442
  HPATH = "instance-failover"
3443
  HTYPE = constants.HTYPE_INSTANCE
3444
  _OP_REQP = ["instance_name", "ignore_consistency"]
3445
  REQ_BGL = False
3446

    
3447
  def ExpandNames(self):
3448
    self._ExpandAndLockInstance()
3449
    self.needed_locks[locking.LEVEL_NODE] = []
3450
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3451

    
3452
  def DeclareLocks(self, level):
3453
    if level == locking.LEVEL_NODE:
3454
      self._LockInstancesNodes()
3455

    
3456
  def BuildHooksEnv(self):
3457
    """Build hooks env.
3458

3459
    This runs on master, primary and secondary nodes of the instance.
3460

3461
    """
3462
    env = {
3463
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3464
      }
3465
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3466
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3467
    return env, nl, nl
3468

    
3469
  def CheckPrereq(self):
3470
    """Check prerequisites.
3471

3472
    This checks that the instance is in the cluster.
3473

3474
    """
3475
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3476
    assert self.instance is not None, \
3477
      "Cannot retrieve locked instance %s" % self.op.instance_name
3478

    
3479
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3480
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3481
      raise errors.OpPrereqError("Instance's disk layout is not"
3482
                                 " network mirrored, cannot failover.")
3483

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

    
3489
    target_node = secondary_nodes[0]
3490
    _CheckNodeOnline(self, target_node)
3491
    _CheckNodeNotDrained(self, target_node)
3492
    # check memory requirements on the secondary node
3493
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3494
                         instance.name, bep[constants.BE_MEMORY],
3495
                         instance.hypervisor)
3496

    
3497
    # check bridge existance
3498
    brlist = [nic.bridge for nic in instance.nics]
3499
    result = self.rpc.call_bridges_exist(target_node, brlist)
3500
    result.Raise()
3501
    if not result.data:
3502
      raise errors.OpPrereqError("One or more target bridges %s does not"
3503
                                 " exist on destination node '%s'" %
3504
                                 (brlist, target_node))
3505

    
3506
  def Exec(self, feedback_fn):
3507
    """Failover an instance.
3508

3509
    The failover is done by shutting it down on its present node and
3510
    starting it on the secondary.
3511

3512
    """
3513
    instance = self.instance
3514

    
3515
    source_node = instance.primary_node
3516
    target_node = instance.secondary_nodes[0]
3517

    
3518
    feedback_fn("* checking disk consistency between source and target")
3519
    for dev in instance.disks:
3520
      # for drbd, these are drbd over lvm
3521
      if not _CheckDiskConsistency(self, dev, target_node, False):
3522
        if instance.admin_up and not self.op.ignore_consistency:
3523
          raise errors.OpExecError("Disk %s is degraded on target node,"
3524
                                   " aborting failover." % dev.iv_name)
3525

    
3526
    feedback_fn("* shutting down instance on source node")
3527
    logging.info("Shutting down instance %s on node %s",
3528
                 instance.name, source_node)
3529

    
3530
    result = self.rpc.call_instance_shutdown(source_node, instance)
3531
    msg = result.RemoteFailMsg()
3532
    if msg:
3533
      if self.op.ignore_consistency:
3534
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3535
                             " Proceeding anyway. Please make sure node"
3536
                             " %s is down. Error details: %s",
3537
                             instance.name, source_node, source_node, msg)
3538
      else:
3539
        raise errors.OpExecError("Could not shutdown instance %s on"
3540
                                 " node %s: %s" %
3541
                                 (instance.name, source_node, msg))
3542

    
3543
    feedback_fn("* deactivating the instance's disks on source node")
3544
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3545
      raise errors.OpExecError("Can't shut down the instance's disks.")
3546

    
3547
    instance.primary_node = target_node
3548
    # distribute new instance config to the other nodes
3549
    self.cfg.Update(instance)
3550

    
3551
    # Only start the instance if it's marked as up
3552
    if instance.admin_up:
3553
      feedback_fn("* activating the instance's disks on target node")
3554
      logging.info("Starting instance %s on node %s",
3555
                   instance.name, target_node)
3556

    
3557
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3558
                                               ignore_secondaries=True)
3559
      if not disks_ok:
3560
        _ShutdownInstanceDisks(self, instance)
3561
        raise errors.OpExecError("Can't activate the instance's disks")
3562

    
3563
      feedback_fn("* starting the instance on the target node")
3564
      result = self.rpc.call_instance_start(target_node, instance)
3565
      msg = result.RemoteFailMsg()
3566
      if msg:
3567
        _ShutdownInstanceDisks(self, instance)
3568
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3569
                                 (instance.name, target_node, msg))
3570

    
3571

    
3572
class LUMigrateInstance(LogicalUnit):
3573
  """Migrate an instance.
3574

3575
  This is migration without shutting down, compared to the failover,
3576
  which is done with shutdown.
3577

3578
  """
3579
  HPATH = "instance-migrate"
3580
  HTYPE = constants.HTYPE_INSTANCE
3581
  _OP_REQP = ["instance_name", "live", "cleanup"]
3582

    
3583
  REQ_BGL = False
3584

    
3585
  def ExpandNames(self):
3586
    self._ExpandAndLockInstance()
3587
    self.needed_locks[locking.LEVEL_NODE] = []
3588
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3589

    
3590
  def DeclareLocks(self, level):
3591
    if level == locking.LEVEL_NODE:
3592
      self._LockInstancesNodes()
3593

    
3594
  def BuildHooksEnv(self):
3595
    """Build hooks env.
3596

3597
    This runs on master, primary and secondary nodes of the instance.
3598

3599
    """
3600
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3601
    env["MIGRATE_LIVE"] = self.op.live
3602
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3603
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3604
    return env, nl, nl
3605

    
3606
  def CheckPrereq(self):
3607
    """Check prerequisites.
3608

3609
    This checks that the instance is in the cluster.
3610

3611
    """
3612
    instance = self.cfg.GetInstanceInfo(
3613
      self.cfg.ExpandInstanceName(self.op.instance_name))
3614
    if instance is None:
3615
      raise errors.OpPrereqError("Instance '%s' not known" %
3616
                                 self.op.instance_name)
3617

    
3618
    if instance.disk_template != constants.DT_DRBD8:
3619
      raise errors.OpPrereqError("Instance's disk layout is not"
3620
                                 " drbd8, cannot migrate.")
3621

    
3622
    secondary_nodes = instance.secondary_nodes
3623
    if not secondary_nodes:
3624
      raise errors.ConfigurationError("No secondary node but using"
3625
                                      " drbd8 disk template")
3626

    
3627
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3628

    
3629
    target_node = secondary_nodes[0]
3630
    # check memory requirements on the secondary node
3631
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3632
                         instance.name, i_be[constants.BE_MEMORY],
3633
                         instance.hypervisor)
3634

    
3635
    # check bridge existance
3636
    brlist = [nic.bridge for nic in instance.nics]
3637
    result = self.rpc.call_bridges_exist(target_node, brlist)
3638
    if result.failed or not result.data:
3639
      raise errors.OpPrereqError("One or more target bridges %s does not"
3640
                                 " exist on destination node '%s'" %
3641
                                 (brlist, target_node))
3642

    
3643
    if not self.op.cleanup:
3644
      _CheckNodeNotDrained(self, target_node)
3645
      result = self.rpc.call_instance_migratable(instance.primary_node,
3646
                                                 instance)
3647
      msg = result.RemoteFailMsg()
3648
      if msg:
3649
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3650
                                   msg)
3651

    
3652
    self.instance = instance
3653

    
3654
  def _WaitUntilSync(self):
3655
    """Poll with custom rpc for disk sync.
3656

3657
    This uses our own step-based rpc call.
3658

3659
    """
3660
    self.feedback_fn("* wait until resync is done")
3661
    all_done = False
3662
    while not all_done:
3663
      all_done = True
3664
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3665
                                            self.nodes_ip,
3666
                                            self.instance.disks)
3667
      min_percent = 100
3668
      for node, nres in result.items():
3669
        msg = nres.RemoteFailMsg()
3670
        if msg:
3671
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3672
                                   (node, msg))
3673
        node_done, node_percent = nres.payload
3674
        all_done = all_done and node_done
3675
        if node_percent is not None:
3676
          min_percent = min(min_percent, node_percent)
3677
      if not all_done:
3678
        if min_percent < 100:
3679
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3680
        time.sleep(2)
3681

    
3682
  def _EnsureSecondary(self, node):
3683
    """Demote a node to secondary.
3684

3685
    """
3686
    self.feedback_fn("* switching node %s to secondary mode" % node)
3687

    
3688
    for dev in self.instance.disks:
3689
      self.cfg.SetDiskID(dev, node)
3690

    
3691
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3692
                                          self.instance.disks)
3693
    msg = result.RemoteFailMsg()
3694
    if msg:
3695
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3696
                               " error %s" % (node, msg))
3697

    
3698
  def _GoStandalone(self):
3699
    """Disconnect from the network.
3700

3701
    """
3702
    self.feedback_fn("* changing into standalone mode")
3703
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3704
                                               self.instance.disks)
3705
    for node, nres in result.items():
3706
      msg = nres.RemoteFailMsg()
3707
      if msg:
3708
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3709
                                 " error %s" % (node, msg))
3710

    
3711
  def _GoReconnect(self, multimaster):
3712
    """Reconnect to the network.
3713

3714
    """
3715
    if multimaster:
3716
      msg = "dual-master"
3717
    else:
3718
      msg = "single-master"
3719
    self.feedback_fn("* changing disks into %s mode" % msg)
3720
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3721
                                           self.instance.disks,
3722
                                           self.instance.name, multimaster)
3723
    for node, nres in result.items():
3724
      msg = nres.RemoteFailMsg()
3725
      if msg:
3726
        raise errors.OpExecError("Cannot change disks config on node %s,"
3727
                                 " error: %s" % (node, msg))
3728

    
3729
  def _ExecCleanup(self):
3730
    """Try to cleanup after a failed migration.
3731

3732
    The cleanup is done by:
3733
      - check that the instance is running only on one node
3734
        (and update the config if needed)
3735
      - change disks on its secondary node to secondary
3736
      - wait until disks are fully synchronized
3737
      - disconnect from the network
3738
      - change disks into single-master mode
3739
      - wait again until disks are fully synchronized
3740

3741
    """
3742
    instance = self.instance
3743
    target_node = self.target_node
3744
    source_node = self.source_node
3745

    
3746
    # check running on only one node
3747
    self.feedback_fn("* checking where the instance actually runs"
3748
                     " (if this hangs, the hypervisor might be in"
3749
                     " a bad state)")
3750
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3751
    for node, result in ins_l.items():
3752
      result.Raise()
3753
      if not isinstance(result.data, list):
3754
        raise errors.OpExecError("Can't contact node '%s'" % node)
3755

    
3756
    runningon_source = instance.name in ins_l[source_node].data
3757
    runningon_target = instance.name in ins_l[target_node].data
3758

    
3759
    if runningon_source and runningon_target:
3760
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3761
                               " or the hypervisor is confused. You will have"
3762
                               " to ensure manually that it runs only on one"
3763
                               " and restart this operation.")
3764

    
3765
    if not (runningon_source or runningon_target):
3766
      raise errors.OpExecError("Instance does not seem to be running at all."
3767
                               " In this case, it's safer to repair by"
3768
                               " running 'gnt-instance stop' to ensure disk"
3769
                               " shutdown, and then restarting it.")
3770

    
3771
    if runningon_target:
3772
      # the migration has actually succeeded, we need to update the config
3773
      self.feedback_fn("* instance running on secondary node (%s),"
3774
                       " updating config" % target_node)
3775
      instance.primary_node = target_node
3776
      self.cfg.Update(instance)
3777
      demoted_node = source_node
3778
    else:
3779
      self.feedback_fn("* instance confirmed to be running on its"
3780
                       " primary node (%s)" % source_node)
3781
      demoted_node = target_node
3782

    
3783
    self._EnsureSecondary(demoted_node)
3784
    try:
3785
      self._WaitUntilSync()
3786
    except errors.OpExecError:
3787
      # we ignore here errors, since if the device is standalone, it
3788
      # won't be able to sync
3789
      pass
3790
    self._GoStandalone()
3791
    self._GoReconnect(False)
3792
    self._WaitUntilSync()
3793

    
3794
    self.feedback_fn("* done")
3795

    
3796
  def _RevertDiskStatus(self):
3797
    """Try to revert the disk status after a failed migration.
3798

3799
    """
3800
    target_node = self.target_node
3801
    try:
3802
      self._EnsureSecondary(target_node)
3803
      self._GoStandalone()
3804
      self._GoReconnect(False)
3805
      self._WaitUntilSync()
3806
    except errors.OpExecError, err:
3807
      self.LogWarning("Migration failed and I can't reconnect the"
3808
                      " drives: error '%s'\n"
3809
                      "Please look and recover the instance status" %
3810
                      str(err))
3811

    
3812
  def _AbortMigration(self):
3813
    """Call the hypervisor code to abort a started migration.
3814

3815
    """
3816
    instance = self.instance
3817
    target_node = self.target_node
3818
    migration_info = self.migration_info
3819

    
3820
    abort_result = self.rpc.call_finalize_migration(target_node,
3821
                                                    instance,
3822
                                                    migration_info,
3823
                                                    False)
3824
    abort_msg = abort_result.RemoteFailMsg()
3825
    if abort_msg:
3826
      logging.error("Aborting migration failed on target node %s: %s" %
3827
                    (target_node, abort_msg))
3828
      # Don't raise an exception here, as we stil have to try to revert the
3829
      # disk status, even if this step failed.
3830

    
3831
  def _ExecMigration(self):
3832
    """Migrate an instance.
3833

3834
    The migrate is done by:
3835
      - change the disks into dual-master mode
3836
      - wait until disks are fully synchronized again
3837
      - migrate the instance
3838
      - change disks on the new secondary node (the old primary) to secondary
3839
      - wait until disks are fully synchronized
3840
      - change disks into single-master mode
3841

3842
    """
3843
    instance = self.instance
3844
    target_node = self.target_node
3845
    source_node = self.source_node
3846

    
3847
    self.feedback_fn("* checking disk consistency between source and target")
3848
    for dev in instance.disks:
3849
      if not _CheckDiskConsistency(self, dev, target_node, False):
3850
        raise errors.OpExecError("Disk %s is degraded or not fully"
3851
                                 " synchronized on target node,"
3852
                                 " aborting migrate." % dev.iv_name)
3853

    
3854
    # First get the migration information from the remote node
3855
    result = self.rpc.call_migration_info(source_node, instance)
3856
    msg = result.RemoteFailMsg()
3857
    if msg:
3858
      log_err = ("Failed fetching source migration information from %s: %s" %
3859
                 (source_node, msg))
3860
      logging.error(log_err)
3861
      raise errors.OpExecError(log_err)
3862

    
3863
    self.migration_info = migration_info = result.payload
3864

    
3865
    # Then switch the disks to master/master mode
3866
    self._EnsureSecondary(target_node)
3867
    self._GoStandalone()
3868
    self._GoReconnect(True)
3869
    self._WaitUntilSync()
3870

    
3871
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3872
    result = self.rpc.call_accept_instance(target_node,
3873
                                           instance,
3874
                                           migration_info,
3875
                                           self.nodes_ip[target_node])
3876

    
3877
    msg = result.RemoteFailMsg()
3878
    if msg:
3879
      logging.error("Instance pre-migration failed, trying to revert"
3880
                    " disk status: %s", msg)
3881
      self._AbortMigration()
3882
      self._RevertDiskStatus()
3883
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3884
                               (instance.name, msg))
3885

    
3886
    self.feedback_fn("* migrating instance to %s" % target_node)
3887
    time.sleep(10)
3888
    result = self.rpc.call_instance_migrate(source_node, instance,
3889
                                            self.nodes_ip[target_node],
3890
                                            self.op.live)
3891
    msg = result.RemoteFailMsg()
3892
    if msg:
3893
      logging.error("Instance migration failed, trying to revert"
3894
                    " disk status: %s", msg)
3895
      self._AbortMigration()
3896
      self._RevertDiskStatus()
3897
      raise errors.OpExecError("Could not migrate instance %s: %s" %
3898
                               (instance.name, msg))
3899
    time.sleep(10)
3900

    
3901
    instance.primary_node = target_node
3902
    # distribute new instance config to the other nodes
3903
    self.cfg.Update(instance)
3904

    
3905
    result = self.rpc.call_finalize_migration(target_node,
3906
                                              instance,
3907
                                              migration_info,
3908
                                              True)
3909
    msg = result.RemoteFailMsg()
3910
    if msg:
3911
      logging.error("Instance migration succeeded, but finalization failed:"
3912
                    " %s" % msg)
3913
      raise errors.OpExecError("Could not finalize instance migration: %s" %
3914
                               msg)
3915

    
3916
    self._EnsureSecondary(source_node)
3917
    self._WaitUntilSync()
3918
    self._GoStandalone()
3919
    self._GoReconnect(False)
3920
    self._WaitUntilSync()
3921

    
3922
    self.feedback_fn("* done")
3923

    
3924
  def Exec(self, feedback_fn):
3925
    """Perform the migration.
3926

3927
    """
3928
    self.feedback_fn = feedback_fn
3929

    
3930
    self.source_node = self.instance.primary_node
3931
    self.target_node = self.instance.secondary_nodes[0]
3932
    self.all_nodes = [self.source_node, self.target_node]
3933
    self.nodes_ip = {
3934
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
3935
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
3936
      }
3937
    if self.op.cleanup:
3938
      return self._ExecCleanup()
3939
    else:
3940
      return self._ExecMigration()
3941

    
3942

    
3943
def _CreateBlockDev(lu, node, instance, device, force_create,
3944
                    info, force_open):
3945
  """Create a tree of block devices on a given node.
3946

3947
  If this device type has to be created on secondaries, create it and
3948
  all its children.
3949

3950
  If not, just recurse to children keeping the same 'force' value.
3951

3952
  @param lu: the lu on whose behalf we execute
3953
  @param node: the node on which to create the device
3954
  @type instance: L{objects.Instance}
3955
  @param instance: the instance which owns the device
3956
  @type device: L{objects.Disk}
3957
  @param device: the device to create
3958
  @type force_create: boolean
3959
  @param force_create: whether to force creation of this device; this
3960
      will be change to True whenever we find a device which has
3961
      CreateOnSecondary() attribute
3962
  @param info: the extra 'metadata' we should attach to the device
3963
      (this will be represented as a LVM tag)
3964
  @type force_open: boolean
3965
  @param force_open: this parameter will be passes to the
3966
      L{backend.BlockdevCreate} function where it specifies
3967
      whether we run on primary or not, and it affects both
3968
      the child assembly and the device own Open() execution
3969

3970
  """
3971
  if device.CreateOnSecondary():
3972
    force_create = True
3973

    
3974
  if device.children:
3975
    for child in device.children:
3976
      _CreateBlockDev(lu, node, instance, child, force_create,
3977
                      info, force_open)
3978

    
3979
  if not force_create:
3980
    return
3981

    
3982
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
3983

    
3984

    
3985
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
3986
  """Create a single block device on a given node.
3987

3988
  This will not recurse over children of the device, so they must be
3989
  created in advance.
3990

3991
  @param lu: the lu on whose behalf we execute
3992
  @param node: the node on which to create the device
3993
  @type instance: L{objects.Instance}
3994
  @param instance: the instance which owns the device
3995
  @type device: L{objects.Disk}
3996
  @param device: the device to create
3997
  @param info: the extra 'metadata' we should attach to the device
3998
      (this will be represented as a LVM tag)
3999
  @type force_open: boolean
4000
  @param force_open: this parameter will be passes to the
4001
      L{backend.BlockdevCreate} function where it specifies
4002
      whether we run on primary or not, and it affects both
4003
      the child assembly and the device own Open() execution
4004

4005
  """
4006
  lu.cfg.SetDiskID(device, node)
4007
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4008
                                       instance.name, force_open, info)
4009
  msg = result.RemoteFailMsg()
4010
  if msg:
4011
    raise errors.OpExecError("Can't create block device %s on"
4012
                             " node %s for instance %s: %s" %
4013
                             (device, node, instance.name, msg))
4014
  if device.physical_id is None:
4015
    device.physical_id = result.payload
4016

    
4017

    
4018
def _GenerateUniqueNames(lu, exts):
4019
  """Generate a suitable LV name.
4020

4021
  This will generate a logical volume name for the given instance.
4022

4023
  """
4024
  results = []
4025
  for val in exts:
4026
    new_id = lu.cfg.GenerateUniqueID()
4027
    results.append("%s%s" % (new_id, val))
4028
  return results
4029

    
4030

    
4031
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4032
                         p_minor, s_minor):
4033
  """Generate a drbd8 device complete with its children.
4034

4035
  """
4036
  port = lu.cfg.AllocatePort()
4037
  vgname = lu.cfg.GetVGName()
4038
  shared_secret = lu.cfg.GenerateDRBDSecret()
4039
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4040
                          logical_id=(vgname, names[0]))
4041
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4042
                          logical_id=(vgname, names[1]))
4043
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4044
                          logical_id=(primary, secondary, port,
4045
                                      p_minor, s_minor,
4046
                                      shared_secret),
4047
                          children=[dev_data, dev_meta],
4048
                          iv_name=iv_name)
4049
  return drbd_dev
4050

    
4051

    
4052
def _GenerateDiskTemplate(lu, template_name,
4053
                          instance_name, primary_node,
4054
                          secondary_nodes, disk_info,
4055
                          file_storage_dir, file_driver,
4056
                          base_index):
4057
  """Generate the entire disk layout for a given template type.
4058

4059
  """
4060
  #TODO: compute space requirements
4061

    
4062
  vgname = lu.cfg.GetVGName()
4063
  disk_count = len(disk_info)
4064
  disks = []
4065
  if template_name == constants.DT_DISKLESS:
4066
    pass
4067
  elif template_name == constants.DT_PLAIN:
4068
    if len(secondary_nodes) != 0:
4069
      raise errors.ProgrammerError("Wrong template configuration")
4070

    
4071
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4072
                                      for i in range(disk_count)])
4073
    for idx, disk in enumerate(disk_info):
4074
      disk_index = idx + base_index
4075
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4076
                              logical_id=(vgname, names[idx]),
4077
                              iv_name="disk/%d" % disk_index,
4078
                              mode=disk["mode"])
4079
      disks.append(disk_dev)
4080
  elif template_name == constants.DT_DRBD8:
4081
    if len(secondary_nodes) != 1:
4082
      raise errors.ProgrammerError("Wrong template configuration")
4083
    remote_node = secondary_nodes[0]
4084
    minors = lu.cfg.AllocateDRBDMinor(
4085
      [primary_node, remote_node] * len(disk_info), instance_name)
4086

    
4087
    names = []
4088
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4089
                                               for i in range(disk_count)]):
4090
      names.append(lv_prefix + "_data")
4091
      names.append(lv_prefix + "_meta")
4092
    for idx, disk in enumerate(disk_info):
4093
      disk_index = idx + base_index
4094
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4095
                                      disk["size"], names[idx*2:idx*2+2],
4096
                                      "disk/%d" % disk_index,
4097
                                      minors[idx*2], minors[idx*2+1])
4098
      disk_dev.mode = disk["mode"]
4099
      disks.append(disk_dev)
4100
  elif template_name == constants.DT_FILE:
4101
    if len(secondary_nodes) != 0:
4102
      raise errors.ProgrammerError("Wrong template configuration")
4103

    
4104
    for idx, disk in enumerate(disk_info):
4105
      disk_index = idx + base_index
4106
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4107
                              iv_name="disk/%d" % disk_index,
4108
                              logical_id=(file_driver,
4109
                                          "%s/disk%d" % (file_storage_dir,
4110
                                                         disk_index)),
4111
                              mode=disk["mode"])
4112
      disks.append(disk_dev)
4113
  else:
4114
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4115
  return disks
4116

    
4117

    
4118
def _GetInstanceInfoText(instance):
4119
  """Compute that text that should be added to the disk's metadata.
4120

4121
  """
4122
  return "originstname+%s" % instance.name
4123

    
4124

    
4125
def _CreateDisks(lu, instance):
4126
  """Create all disks for an instance.
4127

4128
  This abstracts away some work from AddInstance.
4129

4130
  @type lu: L{LogicalUnit}
4131
  @param lu: the logical unit on whose behalf we execute
4132
  @type instance: L{objects.Instance}
4133
  @param instance: the instance whose disks we should create
4134
  @rtype: boolean
4135
  @return: the success of the creation
4136

4137
  """
4138
  info = _GetInstanceInfoText(instance)
4139
  pnode = instance.primary_node
4140

    
4141
  if instance.disk_template == constants.DT_FILE:
4142
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4143
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4144

    
4145
    if result.failed or not result.data:
4146
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4147

    
4148
    if not result.data[0]:
4149
      raise errors.OpExecError("Failed to create directory '%s'" %
4150
                               file_storage_dir)
4151

    
4152
  # Note: this needs to be kept in sync with adding of disks in
4153
  # LUSetInstanceParams
4154
  for device in instance.disks:
4155
    logging.info("Creating volume %s for instance %s",
4156
                 device.iv_name, instance.name)
4157
    #HARDCODE
4158
    for node in instance.all_nodes:
4159
      f_create = node == pnode
4160
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4161

    
4162

    
4163
def _RemoveDisks(lu, instance):
4164
  """Remove all disks for an instance.
4165

4166
  This abstracts away some work from `AddInstance()` and
4167
  `RemoveInstance()`. Note that in case some of the devices couldn't
4168
  be removed, the removal will continue with the other ones (compare
4169
  with `_CreateDisks()`).
4170

4171
  @type lu: L{LogicalUnit}
4172
  @param lu: the logical unit on whose behalf we execute
4173
  @type instance: L{objects.Instance}
4174
  @param instance: the instance whose disks we should remove
4175
  @rtype: boolean
4176
  @return: the success of the removal
4177

4178
  """
4179
  logging.info("Removing block devices for instance %s", instance.name)
4180

    
4181
  all_result = True
4182
  for device in instance.disks:
4183
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4184
      lu.cfg.SetDiskID(disk, node)
4185
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4186
      if msg:
4187
        lu.LogWarning("Could not remove block device %s on node %s,"
4188
                      " continuing anyway: %s", device.iv_name, node, msg)
4189
        all_result = False
4190

    
4191
  if instance.disk_template == constants.DT_FILE:
4192
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4193
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4194
                                                 file_storage_dir)
4195
    if result.failed or not result.data:
4196
      logging.error("Could not remove directory '%s'", file_storage_dir)
4197
      all_result = False
4198

    
4199
  return all_result
4200

    
4201

    
4202
def _ComputeDiskSize(disk_template, disks):
4203
  """Compute disk size requirements in the volume group
4204

4205
  """
4206
  # Required free disk space as a function of disk and swap space
4207
  req_size_dict = {
4208
    constants.DT_DISKLESS: None,
4209
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4210
    # 128 MB are added for drbd metadata for each disk
4211
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4212
    constants.DT_FILE: None,
4213
  }
4214

    
4215
  if disk_template not in req_size_dict:
4216
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4217
                                 " is unknown" %  disk_template)
4218

    
4219
  return req_size_dict[disk_template]
4220

    
4221

    
4222
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4223
  """Hypervisor parameter validation.
4224

4225
  This function abstract the hypervisor parameter validation to be
4226
  used in both instance create and instance modify.
4227

4228
  @type lu: L{LogicalUnit}
4229
  @param lu: the logical unit for which we check
4230
  @type nodenames: list
4231
  @param nodenames: the list of nodes on which we should check
4232
  @type hvname: string
4233
  @param hvname: the name of the hypervisor we should use
4234
  @type hvparams: dict
4235
  @param hvparams: the parameters which we need to check
4236
  @raise errors.OpPrereqError: if the parameters are not valid
4237

4238
  """
4239
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4240
                                                  hvname,
4241
                                                  hvparams)
4242
  for node in nodenames:
4243
    info = hvinfo[node]
4244
    if info.offline:
4245
      continue
4246
    msg = info.RemoteFailMsg()
4247
    if msg:
4248
      raise errors.OpPrereqError("Hypervisor parameter validation"
4249
                                 " failed on node %s: %s" % (node, msg))
4250

    
4251

    
4252
class LUCreateInstance(LogicalUnit):
4253
  """Create an instance.
4254

4255
  """
4256
  HPATH = "instance-add"
4257
  HTYPE = constants.HTYPE_INSTANCE
4258
  _OP_REQP = ["instance_name", "disks", "disk_template",
4259
              "mode", "start",
4260
              "wait_for_sync", "ip_check", "nics",
4261
              "hvparams", "beparams"]
4262
  REQ_BGL = False
4263

    
4264
  def _ExpandNode(self, node):
4265
    """Expands and checks one node name.
4266

4267
    """
4268
    node_full = self.cfg.ExpandNodeName(node)
4269
    if node_full is None:
4270
      raise errors.OpPrereqError("Unknown node %s" % node)
4271
    return node_full
4272

    
4273
  def ExpandNames(self):
4274
    """ExpandNames for CreateInstance.
4275

4276
    Figure out the right locks for instance creation.
4277

4278
    """
4279
    self.needed_locks = {}
4280

    
4281
    # set optional parameters to none if they don't exist
4282
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4283
      if not hasattr(self.op, attr):
4284
        setattr(self.op, attr, None)
4285

    
4286
    # cheap checks, mostly valid constants given
4287

    
4288
    # verify creation mode
4289
    if self.op.mode not in (constants.INSTANCE_CREATE,
4290
                            constants.INSTANCE_IMPORT):
4291
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4292
                                 self.op.mode)
4293

    
4294
    # disk template and mirror node verification
4295
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4296
      raise errors.OpPrereqError("Invalid disk template name")
4297

    
4298
    if self.op.hypervisor is None:
4299
      self.op.hypervisor = self.cfg.GetHypervisorType()
4300

    
4301
    cluster = self.cfg.GetClusterInfo()
4302
    enabled_hvs = cluster.enabled_hypervisors
4303
    if self.op.hypervisor not in enabled_hvs:
4304
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4305
                                 " cluster (%s)" % (self.op.hypervisor,
4306
                                  ",".join(enabled_hvs)))
4307

    
4308
    # check hypervisor parameter syntax (locally)
4309
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4310
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4311
                                  self.op.hvparams)
4312
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4313
    hv_type.CheckParameterSyntax(filled_hvp)
4314

    
4315
    # fill and remember the beparams dict
4316
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4317
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4318
                                    self.op.beparams)
4319

    
4320
    #### instance parameters check
4321

    
4322
    # instance name verification
4323
    hostname1 = utils.HostInfo(self.op.instance_name)
4324
    self.op.instance_name = instance_name = hostname1.name
4325

    
4326
    # this is just a preventive check, but someone might still add this
4327
    # instance in the meantime, and creation will fail at lock-add time
4328
    if instance_name in self.cfg.GetInstanceList():
4329
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4330
                                 instance_name)
4331

    
4332
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4333

    
4334
    # NIC buildup
4335
    self.nics = []
4336
    for nic in self.op.nics:
4337
      # ip validity checks
4338
      ip = nic.get("ip", None)
4339
      if ip is None or ip.lower() == "none":
4340
        nic_ip = None
4341
      elif ip.lower() == constants.VALUE_AUTO:
4342
        nic_ip = hostname1.ip
4343
      else:
4344
        if not utils.IsValidIP(ip):
4345
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4346
                                     " like a valid IP" % ip)
4347
        nic_ip = ip
4348

    
4349
      # MAC address verification
4350
      mac = nic.get("mac", constants.VALUE_AUTO)
4351
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4352
        if not utils.IsValidMac(mac.lower()):
4353
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4354
                                     mac)
4355
      # bridge verification
4356
      bridge = nic.get("bridge", None)
4357
      if bridge is None:
4358
        bridge = self.cfg.GetDefBridge()
4359
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4360

    
4361
    # disk checks/pre-build
4362
    self.disks = []
4363
    for disk in self.op.disks:
4364
      mode = disk.get("mode", constants.DISK_RDWR)
4365
      if mode not in constants.DISK_ACCESS_SET:
4366
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4367
                                   mode)
4368
      size = disk.get("size", None)
4369
      if size is None:
4370
        raise errors.OpPrereqError("Missing disk size")
4371
      try:
4372
        size = int(size)
4373
      except ValueError:
4374
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4375
      self.disks.append({"size": size, "mode": mode})
4376

    
4377
    # used in CheckPrereq for ip ping check
4378
    self.check_ip = hostname1.ip
4379

    
4380
    # file storage checks
4381
    if (self.op.file_driver and
4382
        not self.op.file_driver in constants.FILE_DRIVER):
4383
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4384
                                 self.op.file_driver)
4385

    
4386
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4387
      raise errors.OpPrereqError("File storage directory path not absolute")
4388

    
4389
    ### Node/iallocator related checks
4390
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4391
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4392
                                 " node must be given")
4393

    
4394
    if self.op.iallocator:
4395
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4396
    else:
4397
      self.op.pnode = self._ExpandNode(self.op.pnode)
4398
      nodelist = [self.op.pnode]
4399
      if self.op.snode is not None:
4400
        self.op.snode = self._ExpandNode(self.op.snode)
4401
        nodelist.append(self.op.snode)
4402
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4403

    
4404
    # in case of import lock the source node too
4405
    if self.op.mode == constants.INSTANCE_IMPORT:
4406
      src_node = getattr(self.op, "src_node", None)
4407
      src_path = getattr(self.op, "src_path", None)
4408

    
4409
      if src_path is None:
4410
        self.op.src_path = src_path = self.op.instance_name
4411

    
4412
      if src_node is None:
4413
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4414
        self.op.src_node = None
4415
        if os.path.isabs(src_path):
4416
          raise errors.OpPrereqError("Importing an instance from an absolute"
4417
                                     " path requires a source node option.")
4418
      else:
4419
        self.op.src_node = src_node = self._ExpandNode(src_node)
4420
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4421
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4422
        if not os.path.isabs(src_path):
4423
          self.op.src_path = src_path = \
4424
            os.path.join(constants.EXPORT_DIR, src_path)
4425

    
4426
    else: # INSTANCE_CREATE
4427
      if getattr(self.op, "os_type", None) is None:
4428
        raise errors.OpPrereqError("No guest OS specified")
4429

    
4430
  def _RunAllocator(self):
4431
    """Run the allocator based on input opcode.
4432

4433
    """
4434
    nics = [n.ToDict() for n in self.nics]
4435
    ial = IAllocator(self,
4436
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4437
                     name=self.op.instance_name,
4438
                     disk_template=self.op.disk_template,
4439
                     tags=[],
4440
                     os=self.op.os_type,
4441
                     vcpus=self.be_full[constants.BE_VCPUS],
4442
                     mem_size=self.be_full[constants.BE_MEMORY],
4443
                     disks=self.disks,
4444
                     nics=nics,
4445
                     hypervisor=self.op.hypervisor,
4446
                     )
4447

    
4448
    ial.Run(self.op.iallocator)
4449

    
4450
    if not ial.success:
4451
      raise errors.OpPrereqError("Can't compute nodes using"
4452
                                 " iallocator '%s': %s" % (self.op.iallocator,
4453
                                                           ial.info))
4454
    if len(ial.nodes) != ial.required_nodes:
4455
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4456
                                 " of nodes (%s), required %s" %
4457
                                 (self.op.iallocator, len(ial.nodes),
4458
                                  ial.required_nodes))
4459
    self.op.pnode = ial.nodes[0]
4460
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4461
                 self.op.instance_name, self.op.iallocator,
4462
                 ", ".join(ial.nodes))
4463
    if ial.required_nodes == 2:
4464
      self.op.snode = ial.nodes[1]
4465

    
4466
  def BuildHooksEnv(self):
4467
    """Build hooks env.
4468

4469
    This runs on master, primary and secondary nodes of the instance.
4470

4471
    """
4472
    env = {
4473
      "ADD_MODE": self.op.mode,
4474
      }
4475
    if self.op.mode == constants.INSTANCE_IMPORT:
4476
      env["SRC_NODE"] = self.op.src_node
4477
      env["SRC_PATH"] = self.op.src_path
4478
      env["SRC_IMAGES"] = self.src_images
4479

    
4480
    env.update(_BuildInstanceHookEnv(
4481
      name=self.op.instance_name,
4482
      primary_node=self.op.pnode,
4483
      secondary_nodes=self.secondaries,
4484
      status=self.op.start,
4485
      os_type=self.op.os_type,
4486
      memory=self.be_full[constants.BE_MEMORY],
4487
      vcpus=self.be_full[constants.BE_VCPUS],
4488
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4489
      disk_template=self.op.disk_template,
4490
      disks=[(d["size"], d["mode"]) for d in self.disks],
4491
    ))
4492

    
4493
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4494
          self.secondaries)
4495
    return env, nl, nl
4496

    
4497

    
4498
  def CheckPrereq(self):
4499
    """Check prerequisites.
4500

4501
    """
4502
    if (not self.cfg.GetVGName() and
4503
        self.op.disk_template not in constants.DTS_NOT_LVM):
4504
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4505
                                 " instances")
4506

    
4507
    if self.op.mode == constants.INSTANCE_IMPORT:
4508
      src_node = self.op.src_node
4509
      src_path = self.op.src_path
4510

    
4511
      if src_node is None:
4512
        exp_list = self.rpc.call_export_list(
4513
          self.acquired_locks[locking.LEVEL_NODE])
4514
        found = False
4515
        for node in exp_list:
4516
          if not exp_list[node].failed and src_path in exp_list[node].data:
4517
            found = True
4518
            self.op.src_node = src_node = node
4519
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4520
                                                       src_path)
4521
            break
4522
        if not found:
4523
          raise errors.OpPrereqError("No export found for relative path %s" %
4524
                                      src_path)
4525

    
4526
      _CheckNodeOnline(self, src_node)
4527
      result = self.rpc.call_export_info(src_node, src_path)
4528
      result.Raise()
4529
      if not result.data:
4530
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4531

    
4532
      export_info = result.data
4533
      if not export_info.has_section(constants.INISECT_EXP):
4534
        raise errors.ProgrammerError("Corrupted export config")
4535

    
4536
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4537
      if (int(ei_version) != constants.EXPORT_VERSION):
4538
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4539
                                   (ei_version, constants.EXPORT_VERSION))
4540

    
4541
      # Check that the new instance doesn't have less disks than the export
4542
      instance_disks = len(self.disks)
4543
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4544
      if instance_disks < export_disks:
4545
        raise errors.OpPrereqError("Not enough disks to import."
4546
                                   " (instance: %d, export: %d)" %
4547
                                   (instance_disks, export_disks))
4548

    
4549
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4550
      disk_images = []
4551
      for idx in range(export_disks):
4552
        option = 'disk%d_dump' % idx
4553
        if export_info.has_option(constants.INISECT_INS, option):
4554
          # FIXME: are the old os-es, disk sizes, etc. useful?
4555
          export_name = export_info.get(constants.INISECT_INS, option)
4556
          image = os.path.join(src_path, export_name)
4557
          disk_images.append(image)
4558
        else:
4559
          disk_images.append(False)
4560

    
4561
      self.src_images = disk_images
4562

    
4563
      old_name = export_info.get(constants.INISECT_INS, 'name')
4564
      # FIXME: int() here could throw a ValueError on broken exports
4565
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4566
      if self.op.instance_name == old_name:
4567
        for idx, nic in enumerate(self.nics):
4568
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4569
            nic_mac_ini = 'nic%d_mac' % idx
4570
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4571

    
4572
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4573
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4574
    if self.op.start and not self.op.ip_check:
4575
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4576
                                 " adding an instance in start mode")
4577

    
4578
    if self.op.ip_check:
4579
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4580
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4581
                                   (self.check_ip, self.op.instance_name))
4582

    
4583
    #### mac address generation
4584
    # By generating here the mac address both the allocator and the hooks get
4585
    # the real final mac address rather than the 'auto' or 'generate' value.
4586
    # There is a race condition between the generation and the instance object
4587
    # creation, which means that we know the mac is valid now, but we're not
4588
    # sure it will be when we actually add the instance. If things go bad
4589
    # adding the instance will abort because of a duplicate mac, and the
4590
    # creation job will fail.
4591
    for nic in self.nics:
4592
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4593
        nic.mac = self.cfg.GenerateMAC()
4594

    
4595
    #### allocator run
4596

    
4597
    if self.op.iallocator is not None:
4598
      self._RunAllocator()
4599

    
4600
    #### node related checks
4601

    
4602
    # check primary node
4603
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4604
    assert self.pnode is not None, \
4605
      "Cannot retrieve locked node %s" % self.op.pnode
4606
    if pnode.offline:
4607
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4608
                                 pnode.name)
4609
    if pnode.drained:
4610
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4611
                                 pnode.name)
4612

    
4613
    self.secondaries = []
4614

    
4615
    # mirror node verification
4616
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4617
      if self.op.snode is None:
4618
        raise errors.OpPrereqError("The networked disk templates need"
4619
                                   " a mirror node")
4620
      if self.op.snode == pnode.name:
4621
        raise errors.OpPrereqError("The secondary node cannot be"
4622
                                   " the primary node.")
4623
      _CheckNodeOnline(self, self.op.snode)
4624
      _CheckNodeNotDrained(self, self.op.snode)
4625
      self.secondaries.append(self.op.snode)
4626

    
4627
    nodenames = [pnode.name] + self.secondaries
4628

    
4629
    req_size = _ComputeDiskSize(self.op.disk_template,
4630
                                self.disks)
4631

    
4632
    # Check lv size requirements
4633
    if req_size is not None:
4634
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4635
                                         self.op.hypervisor)
4636
      for node in nodenames:
4637
        info = nodeinfo[node]
4638
        info.Raise()
4639
        info = info.data
4640
        if not info:
4641
          raise errors.OpPrereqError("Cannot get current information"
4642
                                     " from node '%s'" % node)
4643
        vg_free = info.get('vg_free', None)
4644
        if not isinstance(vg_free, int):
4645
          raise errors.OpPrereqError("Can't compute free disk space on"
4646
                                     " node %s" % node)
4647
        if req_size > info['vg_free']:
4648
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4649
                                     " %d MB available, %d MB required" %
4650
                                     (node, info['vg_free'], req_size))
4651

    
4652
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4653

    
4654
    # os verification
4655
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4656
    result.Raise()
4657
    if not isinstance(result.data, objects.OS):
4658
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4659
                                 " primary node"  % self.op.os_type)
4660

    
4661
    # bridge check on primary node
4662
    bridges = [n.bridge for n in self.nics]
4663
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4664
    result.Raise()
4665
    if not result.data:
4666
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4667
                                 " exist on destination node '%s'" %
4668
                                 (",".join(bridges), pnode.name))
4669

    
4670
    # memory check on primary node
4671
    if self.op.start:
4672
      _CheckNodeFreeMemory(self, self.pnode.name,
4673
                           "creating instance %s" % self.op.instance_name,
4674
                           self.be_full[constants.BE_MEMORY],
4675
                           self.op.hypervisor)
4676

    
4677
  def Exec(self, feedback_fn):
4678
    """Create and add the instance to the cluster.
4679

4680
    """
4681
    instance = self.op.instance_name
4682
    pnode_name = self.pnode.name
4683

    
4684
    ht_kind = self.op.hypervisor
4685
    if ht_kind in constants.HTS_REQ_PORT:
4686
      network_port = self.cfg.AllocatePort()
4687
    else:
4688
      network_port = None
4689

    
4690
    ##if self.op.vnc_bind_address is None:
4691
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4692

    
4693
    # this is needed because os.path.join does not accept None arguments
4694
    if self.op.file_storage_dir is None:
4695
      string_file_storage_dir = ""
4696
    else:
4697
      string_file_storage_dir = self.op.file_storage_dir
4698

    
4699
    # build the full file storage dir path
4700
    file_storage_dir = os.path.normpath(os.path.join(
4701
                                        self.cfg.GetFileStorageDir(),
4702
                                        string_file_storage_dir, instance))
4703

    
4704

    
4705
    disks = _GenerateDiskTemplate(self,
4706
                                  self.op.disk_template,
4707
                                  instance, pnode_name,
4708
                                  self.secondaries,
4709
                                  self.disks,
4710
                                  file_storage_dir,
4711
                                  self.op.file_driver,
4712
                                  0)
4713

    
4714
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4715
                            primary_node=pnode_name,
4716
                            nics=self.nics, disks=disks,
4717
                            disk_template=self.op.disk_template,
4718
                            admin_up=False,
4719
                            network_port=network_port,
4720
                            beparams=self.op.beparams,
4721
                            hvparams=self.op.hvparams,
4722
                            hypervisor=self.op.hypervisor,
4723
                            )
4724

    
4725
    feedback_fn("* creating instance disks...")
4726
    try:
4727
      _CreateDisks(self, iobj)
4728
    except errors.OpExecError:
4729
      self.LogWarning("Device creation failed, reverting...")
4730
      try:
4731
        _RemoveDisks(self, iobj)
4732
      finally:
4733
        self.cfg.ReleaseDRBDMinors(instance)
4734
        raise
4735

    
4736
    feedback_fn("adding instance %s to cluster config" % instance)
4737

    
4738
    self.cfg.AddInstance(iobj)
4739
    # Declare that we don't want to remove the instance lock anymore, as we've
4740
    # added the instance to the config
4741
    del self.remove_locks[locking.LEVEL_INSTANCE]
4742
    # Unlock all the nodes
4743
    if self.op.mode == constants.INSTANCE_IMPORT:
4744
      nodes_keep = [self.op.src_node]
4745
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4746
                       if node != self.op.src_node]
4747
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4748
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4749
    else:
4750
      self.context.glm.release(locking.LEVEL_NODE)
4751
      del self.acquired_locks[locking.LEVEL_NODE]
4752

    
4753
    if self.op.wait_for_sync:
4754
      disk_abort = not _WaitForSync(self, iobj)
4755
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4756
      # make sure the disks are not degraded (still sync-ing is ok)
4757
      time.sleep(15)
4758
      feedback_fn("* checking mirrors status")
4759
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4760
    else:
4761
      disk_abort = False
4762

    
4763
    if disk_abort:
4764
      _RemoveDisks(self, iobj)
4765
      self.cfg.RemoveInstance(iobj.name)
4766
      # Make sure the instance lock gets removed
4767
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4768
      raise errors.OpExecError("There are some degraded disks for"
4769
                               " this instance")
4770

    
4771
    feedback_fn("creating os for instance %s on node %s" %
4772
                (instance, pnode_name))
4773

    
4774
    if iobj.disk_template != constants.DT_DISKLESS:
4775
      if self.op.mode == constants.INSTANCE_CREATE:
4776
        feedback_fn("* running the instance OS create scripts...")
4777
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4778
        msg = result.RemoteFailMsg()
4779
        if msg:
4780
          raise errors.OpExecError("Could not add os for instance %s"
4781
                                   " on node %s: %s" %
4782
                                   (instance, pnode_name, msg))
4783

    
4784
      elif self.op.mode == constants.INSTANCE_IMPORT:
4785
        feedback_fn("* running the instance OS import scripts...")
4786
        src_node = self.op.src_node
4787
        src_images = self.src_images
4788
        cluster_name = self.cfg.GetClusterName()
4789
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4790
                                                         src_node, src_images,
4791
                                                         cluster_name)
4792
        import_result.Raise()
4793
        for idx, result in enumerate(import_result.data):
4794
          if not result:
4795
            self.LogWarning("Could not import the image %s for instance"
4796
                            " %s, disk %d, on node %s" %
4797
                            (src_images[idx], instance, idx, pnode_name))
4798
      else:
4799
        # also checked in the prereq part
4800
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4801
                                     % self.op.mode)
4802

    
4803
    if self.op.start:
4804
      iobj.admin_up = True
4805
      self.cfg.Update(iobj)
4806
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4807
      feedback_fn("* starting instance...")
4808
      result = self.rpc.call_instance_start(pnode_name, iobj)
4809
      msg = result.RemoteFailMsg()
4810
      if msg:
4811
        raise errors.OpExecError("Could not start instance: %s" % msg)
4812

    
4813

    
4814
class LUConnectConsole(NoHooksLU):
4815
  """Connect to an instance's console.
4816

4817
  This is somewhat special in that it returns the command line that
4818
  you need to run on the master node in order to connect to the
4819
  console.
4820

4821
  """
4822
  _OP_REQP = ["instance_name"]
4823
  REQ_BGL = False
4824

    
4825
  def ExpandNames(self):
4826
    self._ExpandAndLockInstance()
4827

    
4828
  def CheckPrereq(self):
4829
    """Check prerequisites.
4830

4831
    This checks that the instance is in the cluster.
4832

4833
    """
4834
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4835
    assert self.instance is not None, \
4836
      "Cannot retrieve locked instance %s" % self.op.instance_name
4837
    _CheckNodeOnline(self, self.instance.primary_node)
4838

    
4839
  def Exec(self, feedback_fn):
4840
    """Connect to the console of an instance
4841

4842
    """
4843
    instance = self.instance
4844
    node = instance.primary_node
4845

    
4846
    node_insts = self.rpc.call_instance_list([node],
4847
                                             [instance.hypervisor])[node]
4848
    node_insts.Raise()
4849

    
4850
    if instance.name not in node_insts.data:
4851
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4852

    
4853
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4854

    
4855
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4856
    cluster = self.cfg.GetClusterInfo()
4857
    # beparams and hvparams are passed separately, to avoid editing the
4858
    # instance and then saving the defaults in the instance itself.
4859
    hvparams = cluster.FillHV(instance)
4860
    beparams = cluster.FillBE(instance)
4861
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4862

    
4863
    # build ssh cmdline
4864
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
4865

    
4866

    
4867
class LUReplaceDisks(LogicalUnit):
4868
  """Replace the disks of an instance.
4869

4870
  """
4871
  HPATH = "mirrors-replace"
4872
  HTYPE = constants.HTYPE_INSTANCE
4873
  _OP_REQP = ["instance_name", "mode", "disks"]
4874
  REQ_BGL = False
4875

    
4876
  def CheckArguments(self):
4877
    if not hasattr(self.op, "remote_node"):
4878
      self.op.remote_node = None
4879
    if not hasattr(self.op, "iallocator"):
4880
      self.op.iallocator = None
4881

    
4882
    # check for valid parameter combination
4883
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
4884
    if self.op.mode == constants.REPLACE_DISK_CHG:
4885
      if cnt == 2:
4886
        raise errors.OpPrereqError("When changing the secondary either an"
4887
                                   " iallocator script must be used or the"
4888
                                   " new node given")
4889
      elif cnt == 0:
4890
        raise errors.OpPrereqError("Give either the iallocator or the new"
4891
                                   " secondary, not both")
4892
    else: # not replacing the secondary
4893
      if cnt != 2:
4894
        raise errors.OpPrereqError("The iallocator and new node options can"
4895
                                   " be used only when changing the"
4896
                                   " secondary node")
4897

    
4898
  def ExpandNames(self):
4899
    self._ExpandAndLockInstance()
4900

    
4901
    if self.op.iallocator is not None:
4902
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4903
    elif self.op.remote_node is not None:
4904
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
4905
      if remote_node is None:
4906
        raise errors.OpPrereqError("Node '%s' not known" %
4907
                                   self.op.remote_node)
4908
      self.op.remote_node = remote_node
4909
      # Warning: do not remove the locking of the new secondary here
4910
      # unless DRBD8.AddChildren is changed to work in parallel;
4911
      # currently it doesn't since parallel invocations of
4912
      # FindUnusedMinor will conflict
4913
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
4914
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
4915
    else:
4916
      self.needed_locks[locking.LEVEL_NODE] = []
4917
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4918

    
4919
  def DeclareLocks(self, level):
4920
    # If we're not already locking all nodes in the set we have to declare the
4921
    # instance's primary/secondary nodes.
4922
    if (level == locking.LEVEL_NODE and
4923
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
4924
      self._LockInstancesNodes()
4925

    
4926
  def _RunAllocator(self):
4927
    """Compute a new secondary node using an IAllocator.
4928

4929
    """
4930
    ial = IAllocator(self,
4931
                     mode=constants.IALLOCATOR_MODE_RELOC,
4932
                     name=self.op.instance_name,
4933
                     relocate_from=[self.sec_node])
4934

    
4935
    ial.Run(self.op.iallocator)
4936

    
4937
    if not ial.success:
4938
      raise errors.OpPrereqError("Can't compute nodes using"
4939
                                 " iallocator '%s': %s" % (self.op.iallocator,
4940
                                                           ial.info))
4941
    if len(ial.nodes) != ial.required_nodes:
4942
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4943
                                 " of nodes (%s), required %s" %
4944
                                 (len(ial.nodes), ial.required_nodes))
4945
    self.op.remote_node = ial.nodes[0]
4946
    self.LogInfo("Selected new secondary for the instance: %s",
4947
                 self.op.remote_node)
4948

    
4949
  def BuildHooksEnv(self):
4950
    """Build hooks env.
4951

4952
    This runs on the master, the primary and all the secondaries.
4953

4954
    """
4955
    env = {
4956
      "MODE": self.op.mode,
4957
      "NEW_SECONDARY": self.op.remote_node,
4958
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
4959
      }
4960
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4961
    nl = [
4962
      self.cfg.GetMasterNode(),
4963
      self.instance.primary_node,
4964
      ]
4965
    if self.op.remote_node is not None:
4966
      nl.append(self.op.remote_node)
4967
    return env, nl, nl
4968

    
4969
  def CheckPrereq(self):
4970
    """Check prerequisites.
4971

4972
    This checks that the instance is in the cluster.
4973

4974
    """
4975
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4976
    assert instance is not None, \
4977
      "Cannot retrieve locked instance %s" % self.op.instance_name
4978
    self.instance = instance
4979

    
4980
    if instance.disk_template != constants.DT_DRBD8:
4981
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
4982
                                 " instances")
4983

    
4984
    if len(instance.secondary_nodes) != 1:
4985
      raise errors.OpPrereqError("The instance has a strange layout,"
4986
                                 " expected one secondary but found %d" %
4987
                                 len(instance.secondary_nodes))
4988

    
4989
    self.sec_node = instance.secondary_nodes[0]
4990

    
4991
    if self.op.iallocator is not None:
4992
      self._RunAllocator()
4993

    
4994
    remote_node = self.op.remote_node
4995
    if remote_node is not None:
4996
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
4997
      assert self.remote_node_info is not None, \
4998
        "Cannot retrieve locked node %s" % remote_node
4999
    else:
5000
      self.remote_node_info = None
5001
    if remote_node == instance.primary_node:
5002
      raise errors.OpPrereqError("The specified node is the primary node of"
5003
                                 " the instance.")
5004
    elif remote_node == self.sec_node:
5005
      raise errors.OpPrereqError("The specified node is already the"
5006
                                 " secondary node of the instance.")
5007

    
5008
    if self.op.mode == constants.REPLACE_DISK_PRI:
5009
      n1 = self.tgt_node = instance.primary_node
5010
      n2 = self.oth_node = self.sec_node
5011
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5012
      n1 = self.tgt_node = self.sec_node
5013
      n2 = self.oth_node = instance.primary_node
5014
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5015
      n1 = self.new_node = remote_node
5016
      n2 = self.oth_node = instance.primary_node
5017
      self.tgt_node = self.sec_node
5018
      _CheckNodeNotDrained(self, remote_node)
5019
    else:
5020
      raise errors.ProgrammerError("Unhandled disk replace mode")
5021

    
5022
    _CheckNodeOnline(self, n1)
5023
    _CheckNodeOnline(self, n2)
5024

    
5025
    if not self.op.disks:
5026
      self.op.disks = range(len(instance.disks))
5027

    
5028
    for disk_idx in self.op.disks:
5029
      instance.FindDisk(disk_idx)
5030

    
5031
  def _ExecD8DiskOnly(self, feedback_fn):
5032
    """Replace a disk on the primary or secondary for dbrd8.
5033

5034
    The algorithm for replace is quite complicated:
5035

5036
      1. for each disk to be replaced:
5037

5038
        1. create new LVs on the target node with unique names
5039
        1. detach old LVs from the drbd device
5040
        1. rename old LVs to name_replaced.<time_t>
5041
        1. rename new LVs to old LVs
5042
        1. attach the new LVs (with the old names now) to the drbd device
5043

5044
      1. wait for sync across all devices
5045

5046
      1. for each modified disk:
5047

5048
        1. remove old LVs (which have the name name_replaces.<time_t>)
5049

5050
    Failures are not very well handled.
5051

5052
    """
5053
    steps_total = 6
5054
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5055
    instance = self.instance
5056
    iv_names = {}
5057
    vgname = self.cfg.GetVGName()
5058
    # start of work
5059
    cfg = self.cfg
5060
    tgt_node = self.tgt_node
5061
    oth_node = self.oth_node
5062

    
5063
    # Step: check device activation
5064
    self.proc.LogStep(1, steps_total, "check device existence")
5065
    info("checking volume groups")
5066
    my_vg = cfg.GetVGName()
5067
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5068
    if not results:
5069
      raise errors.OpExecError("Can't list volume groups on the nodes")
5070
    for node in oth_node, tgt_node:
5071
      res = results[node]
5072
      if res.failed or not res.data or my_vg not in res.data:
5073
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5074
                                 (my_vg, node))
5075
    for idx, dev in enumerate(instance.disks):
5076
      if idx not in self.op.disks:
5077
        continue
5078
      for node in tgt_node, oth_node:
5079
        info("checking disk/%d on %s" % (idx, node))
5080
        cfg.SetDiskID(dev, node)
5081
        result = self.rpc.call_blockdev_find(node, dev)
5082
        msg = result.RemoteFailMsg()
5083
        if not msg and not result.payload:
5084
          msg = "disk not found"
5085
        if msg:
5086
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5087
                                   (idx, node, msg))
5088

    
5089
    # Step: check other node consistency
5090
    self.proc.LogStep(2, steps_total, "check peer consistency")
5091
    for idx, dev in enumerate(instance.disks):
5092
      if idx not in self.op.disks:
5093
        continue
5094
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5095
      if not _CheckDiskConsistency(self, dev, oth_node,
5096
                                   oth_node==instance.primary_node):
5097
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5098
                                 " to replace disks on this node (%s)" %
5099
                                 (oth_node, tgt_node))
5100

    
5101
    # Step: create new storage
5102
    self.proc.LogStep(3, steps_total, "allocate new storage")
5103
    for idx, dev in enumerate(instance.disks):
5104
      if idx not in self.op.disks:
5105
        continue
5106
      size = dev.size
5107
      cfg.SetDiskID(dev, tgt_node)
5108
      lv_names = [".disk%d_%s" % (idx, suf)
5109
                  for suf in ["data", "meta"]]
5110
      names = _GenerateUniqueNames(self, lv_names)
5111
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5112
                             logical_id=(vgname, names[0]))
5113
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5114
                             logical_id=(vgname, names[1]))
5115
      new_lvs = [lv_data, lv_meta]
5116
      old_lvs = dev.children
5117
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5118
      info("creating new local storage on %s for %s" %
5119
           (tgt_node, dev.iv_name))
5120
      # we pass force_create=True to force the LVM creation
5121
      for new_lv in new_lvs:
5122
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5123
                        _GetInstanceInfoText(instance), False)
5124

    
5125
    # Step: for each lv, detach+rename*2+attach
5126
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5127
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5128
      info("detaching %s drbd from local storage" % dev.iv_name)
5129
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5130
      result.Raise()
5131
      if not result.data:
5132
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5133
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5134
      #dev.children = []
5135
      #cfg.Update(instance)
5136

    
5137
      # ok, we created the new LVs, so now we know we have the needed
5138
      # storage; as such, we proceed on the target node to rename
5139
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5140
      # using the assumption that logical_id == physical_id (which in
5141
      # turn is the unique_id on that node)
5142

    
5143
      # FIXME(iustin): use a better name for the replaced LVs
5144
      temp_suffix = int(time.time())
5145
      ren_fn = lambda d, suff: (d.physical_id[0],
5146
                                d.physical_id[1] + "_replaced-%s" % suff)
5147
      # build the rename list based on what LVs exist on the node
5148
      rlist = []
5149
      for to_ren in old_lvs:
5150
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5151
        if not result.RemoteFailMsg() and result.payload:
5152
          # device exists
5153
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5154

    
5155
      info("renaming the old LVs on the target node")
5156
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5157
      result.Raise()
5158
      if not result.data:
5159
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5160
      # now we rename the new LVs to the old LVs
5161
      info("renaming the new LVs on the target node")
5162
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5163
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5164
      result.Raise()
5165
      if not result.data:
5166
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5167

    
5168
      for old, new in zip(old_lvs, new_lvs):
5169
        new.logical_id = old.logical_id
5170
        cfg.SetDiskID(new, tgt_node)
5171

    
5172
      for disk in old_lvs:
5173
        disk.logical_id = ren_fn(disk, temp_suffix)
5174
        cfg.SetDiskID(disk, tgt_node)
5175

    
5176
      # now that the new lvs have the old name, we can add them to the device
5177
      info("adding new mirror component on %s" % tgt_node)
5178
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5179
      if result.failed or not result.data:
5180
        for new_lv in new_lvs:
5181
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5182
          if msg:
5183
            warning("Can't rollback device %s: %s", dev, msg,
5184
                    hint="cleanup manually the unused logical volumes")
5185
        raise errors.OpExecError("Can't add local storage to drbd")
5186

    
5187
      dev.children = new_lvs
5188
      cfg.Update(instance)
5189

    
5190
    # Step: wait for sync
5191

    
5192
    # this can fail as the old devices are degraded and _WaitForSync
5193
    # does a combined result over all disks, so we don't check its
5194
    # return value
5195
    self.proc.LogStep(5, steps_total, "sync devices")
5196
    _WaitForSync(self, instance, unlock=True)
5197

    
5198
    # so check manually all the devices
5199
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5200
      cfg.SetDiskID(dev, instance.primary_node)
5201
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5202
      msg = result.RemoteFailMsg()
5203
      if not msg and not result.payload:
5204
        msg = "disk not found"
5205
      if msg:
5206
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5207
                                 (name, msg))
5208
      if result.payload[5]:
5209
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5210

    
5211
    # Step: remove old storage
5212
    self.proc.LogStep(6, steps_total, "removing old storage")
5213
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5214
      info("remove logical volumes for %s" % name)
5215
      for lv in old_lvs:
5216
        cfg.SetDiskID(lv, tgt_node)
5217
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5218
        if msg:
5219
          warning("Can't remove old LV: %s" % msg,
5220
                  hint="manually remove unused LVs")
5221
          continue
5222

    
5223
  def _ExecD8Secondary(self, feedback_fn):
5224
    """Replace the secondary node for drbd8.
5225

5226
    The algorithm for replace is quite complicated:
5227
      - for all disks of the instance:
5228
        - create new LVs on the new node with same names
5229
        - shutdown the drbd device on the old secondary
5230
        - disconnect the drbd network on the primary
5231
        - create the drbd device on the new secondary
5232
        - network attach the drbd on the primary, using an artifice:
5233
          the drbd code for Attach() will connect to the network if it
5234
          finds a device which is connected to the good local disks but
5235
          not network enabled
5236
      - wait for sync across all devices
5237
      - remove all disks from the old secondary
5238

5239
    Failures are not very well handled.
5240

5241
    """
5242
    steps_total = 6
5243
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5244
    instance = self.instance
5245
    iv_names = {}
5246
    # start of work
5247
    cfg = self.cfg
5248
    old_node = self.tgt_node
5249
    new_node = self.new_node
5250
    pri_node = instance.primary_node
5251
    nodes_ip = {
5252
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5253
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5254
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5255
      }
5256

    
5257
    # Step: check device activation
5258
    self.proc.LogStep(1, steps_total, "check device existence")
5259
    info("checking volume groups")
5260
    my_vg = cfg.GetVGName()
5261
    results = self.rpc.call_vg_list([pri_node, new_node])
5262
    for node in pri_node, new_node:
5263
      res = results[node]
5264
      if res.failed or not res.data or my_vg not in res.data:
5265
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5266
                                 (my_vg, node))
5267
    for idx, dev in enumerate(instance.disks):
5268
      if idx not in self.op.disks:
5269
        continue
5270
      info("checking disk/%d on %s" % (idx, pri_node))
5271
      cfg.SetDiskID(dev, pri_node)
5272
      result = self.rpc.call_blockdev_find(pri_node, dev)
5273
      msg = result.RemoteFailMsg()
5274
      if not msg and not result.payload:
5275
        msg = "disk not found"
5276
      if msg:
5277
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5278
                                 (idx, pri_node, msg))
5279

    
5280
    # Step: check other node consistency
5281
    self.proc.LogStep(2, steps_total, "check peer consistency")
5282
    for idx, dev in enumerate(instance.disks):
5283
      if idx not in self.op.disks:
5284
        continue
5285
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5286
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5287
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5288
                                 " unsafe to replace the secondary" %
5289
                                 pri_node)
5290

    
5291
    # Step: create new storage
5292
    self.proc.LogStep(3, steps_total, "allocate new storage")
5293
    for idx, dev in enumerate(instance.disks):
5294
      info("adding new local storage on %s for disk/%d" %
5295
           (new_node, idx))
5296
      # we pass force_create=True to force LVM creation
5297
      for new_lv in dev.children:
5298
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5299
                        _GetInstanceInfoText(instance), False)
5300

    
5301
    # Step 4: dbrd minors and drbd setups changes
5302
    # after this, we must manually remove the drbd minors on both the
5303
    # error and the success paths
5304
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5305
                                   instance.name)
5306
    logging.debug("Allocated minors %s" % (minors,))
5307
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5308
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5309
      size = dev.size
5310
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5311
      # create new devices on new_node; note that we create two IDs:
5312
      # one without port, so the drbd will be activated without
5313
      # networking information on the new node at this stage, and one
5314
      # with network, for the latter activation in step 4
5315
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5316
      if pri_node == o_node1:
5317
        p_minor = o_minor1
5318
      else:
5319
        p_minor = o_minor2
5320

    
5321
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5322
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5323

    
5324
      iv_names[idx] = (dev, dev.children, new_net_id)
5325
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5326
                    new_net_id)
5327
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5328
                              logical_id=new_alone_id,
5329
                              children=dev.children)
5330
      try:
5331
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5332
                              _GetInstanceInfoText(instance), False)
5333
      except errors.GenericError:
5334
        self.cfg.ReleaseDRBDMinors(instance.name)
5335
        raise
5336

    
5337
    for idx, dev in enumerate(instance.disks):
5338
      # we have new devices, shutdown the drbd on the old secondary
5339
      info("shutting down drbd for disk/%d on old node" % idx)
5340
      cfg.SetDiskID(dev, old_node)
5341
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5342
      if msg:
5343
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5344
                (idx, msg),
5345
                hint="Please cleanup this device manually as soon as possible")
5346

    
5347
    info("detaching primary drbds from the network (=> standalone)")
5348
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5349
                                               instance.disks)[pri_node]
5350

    
5351
    msg = result.RemoteFailMsg()
5352
    if msg:
5353
      # detaches didn't succeed (unlikely)
5354
      self.cfg.ReleaseDRBDMinors(instance.name)
5355
      raise errors.OpExecError("Can't detach the disks from the network on"
5356
                               " old node: %s" % (msg,))
5357

    
5358
    # if we managed to detach at least one, we update all the disks of
5359
    # the instance to point to the new secondary
5360
    info("updating instance configuration")
5361
    for dev, _, new_logical_id in iv_names.itervalues():
5362
      dev.logical_id = new_logical_id
5363
      cfg.SetDiskID(dev, pri_node)
5364
    cfg.Update(instance)
5365

    
5366
    # and now perform the drbd attach
5367
    info("attaching primary drbds to new secondary (standalone => connected)")
5368
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5369
                                           instance.disks, instance.name,
5370
                                           False)
5371
    for to_node, to_result in result.items():
5372
      msg = to_result.RemoteFailMsg()
5373
      if msg:
5374
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5375
                hint="please do a gnt-instance info to see the"
5376
                " status of disks")
5377

    
5378
    # this can fail as the old devices are degraded and _WaitForSync
5379
    # does a combined result over all disks, so we don't check its
5380
    # return value
5381
    self.proc.LogStep(5, steps_total, "sync devices")
5382
    _WaitForSync(self, instance, unlock=True)
5383

    
5384
    # so check manually all the devices
5385
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5386
      cfg.SetDiskID(dev, pri_node)
5387
      result = self.rpc.call_blockdev_find(pri_node, dev)
5388
      msg = result.RemoteFailMsg()
5389
      if not msg and not result.payload:
5390
        msg = "disk not found"
5391
      if msg:
5392
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5393
                                 (idx, msg))
5394
      if result.payload[5]:
5395
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5396

    
5397
    self.proc.LogStep(6, steps_total, "removing old storage")
5398
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5399
      info("remove logical volumes for disk/%d" % idx)
5400
      for lv in old_lvs:
5401
        cfg.SetDiskID(lv, old_node)
5402
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5403
        if msg:
5404
          warning("Can't remove LV on old secondary: %s", msg,
5405
                  hint="Cleanup stale volumes by hand")
5406

    
5407
  def Exec(self, feedback_fn):
5408
    """Execute disk replacement.
5409

5410
    This dispatches the disk replacement to the appropriate handler.
5411

5412
    """
5413
    instance = self.instance
5414

    
5415
    # Activate the instance disks if we're replacing them on a down instance
5416
    if not instance.admin_up:
5417
      _StartInstanceDisks(self, instance, True)
5418

    
5419
    if self.op.mode == constants.REPLACE_DISK_CHG:
5420
      fn = self._ExecD8Secondary
5421
    else:
5422
      fn = self._ExecD8DiskOnly
5423

    
5424
    ret = fn(feedback_fn)
5425

    
5426
    # Deactivate the instance disks if we're replacing them on a down instance
5427
    if not instance.admin_up:
5428
      _SafeShutdownInstanceDisks(self, instance)
5429

    
5430
    return ret
5431

    
5432

    
5433
class LUGrowDisk(LogicalUnit):
5434
  """Grow a disk of an instance.
5435

5436
  """
5437
  HPATH = "disk-grow"
5438
  HTYPE = constants.HTYPE_INSTANCE
5439
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5440
  REQ_BGL = False
5441

    
5442
  def ExpandNames(self):
5443
    self._ExpandAndLockInstance()
5444
    self.needed_locks[locking.LEVEL_NODE] = []
5445
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5446

    
5447
  def DeclareLocks(self, level):
5448
    if level == locking.LEVEL_NODE:
5449
      self._LockInstancesNodes()
5450

    
5451
  def BuildHooksEnv(self):
5452
    """Build hooks env.
5453

5454
    This runs on the master, the primary and all the secondaries.
5455

5456
    """
5457
    env = {
5458
      "DISK": self.op.disk,
5459
      "AMOUNT": self.op.amount,
5460
      }
5461
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5462
    nl = [
5463
      self.cfg.GetMasterNode(),
5464
      self.instance.primary_node,
5465
      ]
5466
    return env, nl, nl
5467

    
5468
  def CheckPrereq(self):
5469
    """Check prerequisites.
5470

5471
    This checks that the instance is in the cluster.
5472

5473
    """
5474
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5475
    assert instance is not None, \
5476
      "Cannot retrieve locked instance %s" % self.op.instance_name
5477
    nodenames = list(instance.all_nodes)
5478
    for node in nodenames:
5479
      _CheckNodeOnline(self, node)
5480

    
5481

    
5482
    self.instance = instance
5483

    
5484
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5485
      raise errors.OpPrereqError("Instance's disk layout does not support"
5486
                                 " growing.")
5487

    
5488
    self.disk = instance.FindDisk(self.op.disk)
5489

    
5490
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5491
                                       instance.hypervisor)
5492
    for node in nodenames:
5493
      info = nodeinfo[node]
5494
      if info.failed or not info.data:
5495
        raise errors.OpPrereqError("Cannot get current information"
5496
                                   " from node '%s'" % node)
5497
      vg_free = info.data.get('vg_free', None)
5498
      if not isinstance(vg_free, int):
5499
        raise errors.OpPrereqError("Can't compute free disk space on"
5500
                                   " node %s" % node)
5501
      if self.op.amount > vg_free:
5502
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5503
                                   " %d MiB available, %d MiB required" %
5504
                                   (node, vg_free, self.op.amount))
5505

    
5506
  def Exec(self, feedback_fn):
5507
    """Execute disk grow.
5508

5509
    """
5510
    instance = self.instance
5511
    disk = self.disk
5512
    for node in instance.all_nodes:
5513
      self.cfg.SetDiskID(disk, node)
5514
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5515
      msg = result.RemoteFailMsg()
5516
      if msg:
5517
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5518
                                 (node, msg))
5519
    disk.RecordGrow(self.op.amount)
5520
    self.cfg.Update(instance)
5521
    if self.op.wait_for_sync:
5522
      disk_abort = not _WaitForSync(self, instance)
5523
      if disk_abort:
5524
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5525
                             " status.\nPlease check the instance.")
5526

    
5527

    
5528
class LUQueryInstanceData(NoHooksLU):
5529
  """Query runtime instance data.
5530

5531
  """
5532
  _OP_REQP = ["instances", "static"]
5533
  REQ_BGL = False
5534

    
5535
  def ExpandNames(self):
5536
    self.needed_locks = {}
5537
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5538

    
5539
    if not isinstance(self.op.instances, list):
5540
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5541

    
5542
    if self.op.instances:
5543
      self.wanted_names = []
5544
      for name in self.op.instances:
5545
        full_name = self.cfg.ExpandInstanceName(name)
5546
        if full_name is None:
5547
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5548
        self.wanted_names.append(full_name)
5549
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5550
    else:
5551
      self.wanted_names = None
5552
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5553

    
5554
    self.needed_locks[locking.LEVEL_NODE] = []
5555
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5556

    
5557
  def DeclareLocks(self, level):
5558
    if level == locking.LEVEL_NODE:
5559
      self._LockInstancesNodes()
5560

    
5561
  def CheckPrereq(self):
5562
    """Check prerequisites.
5563

5564
    This only checks the optional instance list against the existing names.
5565

5566
    """
5567
    if self.wanted_names is None:
5568
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5569

    
5570
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5571
                             in self.wanted_names]
5572
    return
5573

    
5574
  def _ComputeDiskStatus(self, instance, snode, dev):
5575
    """Compute block device status.
5576

5577
    """
5578
    static = self.op.static
5579
    if not static:
5580
      self.cfg.SetDiskID(dev, instance.primary_node)
5581
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5582
      if dev_pstatus.offline:
5583
        dev_pstatus = None
5584
      else:
5585
        msg = dev_pstatus.RemoteFailMsg()
5586
        if msg:
5587
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5588
                                   (instance.name, msg))
5589
        dev_pstatus = dev_pstatus.payload
5590
    else:
5591
      dev_pstatus = None
5592

    
5593
    if dev.dev_type in constants.LDS_DRBD:
5594
      # we change the snode then (otherwise we use the one passed in)
5595
      if dev.logical_id[0] == instance.primary_node:
5596
        snode = dev.logical_id[1]
5597
      else:
5598
        snode = dev.logical_id[0]
5599

    
5600
    if snode and not static:
5601
      self.cfg.SetDiskID(dev, snode)
5602
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5603
      if dev_sstatus.offline:
5604
        dev_sstatus = None
5605
      else:
5606
        msg = dev_sstatus.RemoteFailMsg()
5607
        if msg:
5608
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5609
                                   (instance.name, msg))
5610
        dev_sstatus = dev_sstatus.payload
5611
    else:
5612
      dev_sstatus = None
5613

    
5614
    if dev.children:
5615
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5616
                      for child in dev.children]
5617
    else:
5618
      dev_children = []
5619

    
5620
    data = {
5621
      "iv_name": dev.iv_name,
5622
      "dev_type": dev.dev_type,
5623
      "logical_id": dev.logical_id,
5624
      "physical_id": dev.physical_id,
5625
      "pstatus": dev_pstatus,
5626
      "sstatus": dev_sstatus,
5627
      "children": dev_children,
5628
      "mode": dev.mode,
5629
      }
5630

    
5631
    return data
5632

    
5633
  def Exec(self, feedback_fn):
5634
    """Gather and return data"""
5635
    result = {}
5636

    
5637
    cluster = self.cfg.GetClusterInfo()
5638

    
5639
    for instance in self.wanted_instances:
5640
      if not self.op.static:
5641
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5642
                                                  instance.name,
5643
                                                  instance.hypervisor)
5644
        remote_info.Raise()
5645
        remote_info = remote_info.data
5646
        if remote_info and "state" in remote_info:
5647
          remote_state = "up"
5648
        else:
5649
          remote_state = "down"
5650
      else:
5651
        remote_state = None
5652
      if instance.admin_up:
5653
        config_state = "up"
5654
      else:
5655
        config_state = "down"
5656

    
5657
      disks = [self._ComputeDiskStatus(instance, None, device)
5658
               for device in instance.disks]
5659

    
5660
      idict = {
5661
        "name": instance.name,
5662
        "config_state": config_state,
5663
        "run_state": remote_state,
5664
        "pnode": instance.primary_node,
5665
        "snodes": instance.secondary_nodes,
5666
        "os": instance.os,
5667
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5668
        "disks": disks,
5669
        "hypervisor": instance.hypervisor,
5670
        "network_port": instance.network_port,
5671
        "hv_instance": instance.hvparams,
5672
        "hv_actual": cluster.FillHV(instance),
5673
        "be_instance": instance.beparams,
5674
        "be_actual": cluster.FillBE(instance),
5675
        }
5676

    
5677
      result[instance.name] = idict
5678

    
5679
    return result
5680

    
5681

    
5682
class LUSetInstanceParams(LogicalUnit):
5683
  """Modifies an instances's parameters.
5684

5685
  """
5686
  HPATH = "instance-modify"
5687
  HTYPE = constants.HTYPE_INSTANCE
5688
  _OP_REQP = ["instance_name"]
5689
  REQ_BGL = False
5690

    
5691
  def CheckArguments(self):
5692
    if not hasattr(self.op, 'nics'):
5693
      self.op.nics = []
5694
    if not hasattr(self.op, 'disks'):
5695
      self.op.disks = []
5696
    if not hasattr(self.op, 'beparams'):
5697
      self.op.beparams = {}
5698
    if not hasattr(self.op, 'hvparams'):
5699
      self.op.hvparams = {}
5700
    self.op.force = getattr(self.op, "force", False)
5701
    if not (self.op.nics or self.op.disks or
5702
            self.op.hvparams or self.op.beparams):
5703
      raise errors.OpPrereqError("No changes submitted")
5704

    
5705
    # Disk validation
5706
    disk_addremove = 0
5707
    for disk_op, disk_dict in self.op.disks:
5708
      if disk_op == constants.DDM_REMOVE:
5709
        disk_addremove += 1
5710
        continue
5711
      elif disk_op == constants.DDM_ADD:
5712
        disk_addremove += 1
5713
      else:
5714
        if not isinstance(disk_op, int):
5715
          raise errors.OpPrereqError("Invalid disk index")
5716
      if disk_op == constants.DDM_ADD:
5717
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5718
        if mode not in constants.DISK_ACCESS_SET:
5719
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5720
        size = disk_dict.get('size', None)
5721
        if size is None:
5722
          raise errors.OpPrereqError("Required disk parameter size missing")
5723
        try:
5724
          size = int(size)
5725
        except ValueError, err:
5726
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5727
                                     str(err))
5728
        disk_dict['size'] = size
5729
      else:
5730
        # modification of disk
5731
        if 'size' in disk_dict:
5732
          raise errors.OpPrereqError("Disk size change not possible, use"
5733
                                     " grow-disk")
5734

    
5735
    if disk_addremove > 1:
5736
      raise errors.OpPrereqError("Only one disk add or remove operation"
5737
                                 " supported at a time")
5738

    
5739
    # NIC validation
5740
    nic_addremove = 0
5741
    for nic_op, nic_dict in self.op.nics:
5742
      if nic_op == constants.DDM_REMOVE:
5743
        nic_addremove += 1
5744
        continue
5745
      elif nic_op == constants.DDM_ADD:
5746
        nic_addremove += 1
5747
      else:
5748
        if not isinstance(nic_op, int):
5749
          raise errors.OpPrereqError("Invalid nic index")
5750

    
5751
      # nic_dict should be a dict
5752
      nic_ip = nic_dict.get('ip', None)
5753
      if nic_ip is not None:
5754
        if nic_ip.lower() == constants.VALUE_NONE:
5755
          nic_dict['ip'] = None
5756
        else:
5757
          if not utils.IsValidIP(nic_ip):
5758
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5759

    
5760
      if nic_op == constants.DDM_ADD:
5761
        nic_bridge = nic_dict.get('bridge', None)
5762
        if nic_bridge is None:
5763
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5764
        nic_mac = nic_dict.get('mac', None)
5765
        if nic_mac is None:
5766
          nic_dict['mac'] = constants.VALUE_AUTO
5767

    
5768
      if 'mac' in nic_dict:
5769
        nic_mac = nic_dict['mac']
5770
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5771
          if not utils.IsValidMac(nic_mac):
5772
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5773
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5774
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5775
                                     " modifying an existing nic")
5776

    
5777
    if nic_addremove > 1:
5778
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5779
                                 " supported at a time")
5780

    
5781
  def ExpandNames(self):
5782
    self._ExpandAndLockInstance()
5783
    self.needed_locks[locking.LEVEL_NODE] = []
5784
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5785

    
5786
  def DeclareLocks(self, level):
5787
    if level == locking.LEVEL_NODE:
5788
      self._LockInstancesNodes()
5789

    
5790
  def BuildHooksEnv(self):
5791
    """Build hooks env.
5792

5793
    This runs on the master, primary and secondaries.
5794

5795
    """
5796
    args = dict()
5797
    if constants.BE_MEMORY in self.be_new:
5798
      args['memory'] = self.be_new[constants.BE_MEMORY]
5799
    if constants.BE_VCPUS in self.be_new:
5800
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5801
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5802
    # information at all.
5803
    if self.op.nics:
5804
      args['nics'] = []
5805
      nic_override = dict(self.op.nics)
5806
      for idx, nic in enumerate(self.instance.nics):
5807
        if idx in nic_override:
5808
          this_nic_override = nic_override[idx]
5809
        else:
5810
          this_nic_override = {}
5811
        if 'ip' in this_nic_override:
5812
          ip = this_nic_override['ip']
5813
        else:
5814
          ip = nic.ip
5815
        if 'bridge' in this_nic_override:
5816
          bridge = this_nic_override['bridge']
5817
        else:
5818
          bridge = nic.bridge
5819
        if 'mac' in this_nic_override:
5820
          mac = this_nic_override['mac']
5821
        else:
5822
          mac = nic.mac
5823
        args['nics'].append((ip, bridge, mac))
5824
      if constants.DDM_ADD in nic_override:
5825
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5826
        bridge = nic_override[constants.DDM_ADD]['bridge']
5827
        mac = nic_override[constants.DDM_ADD]['mac']
5828
        args['nics'].append((ip, bridge, mac))
5829
      elif constants.DDM_REMOVE in nic_override:
5830
        del args['nics'][-1]
5831

    
5832
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5833
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5834
    return env, nl, nl
5835

    
5836
  def CheckPrereq(self):
5837
    """Check prerequisites.
5838

5839
    This only checks the instance list against the existing names.
5840

5841
    """
5842
    force = self.force = self.op.force
5843

    
5844
    # checking the new params on the primary/secondary nodes
5845

    
5846
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5847
    assert self.instance is not None, \
5848
      "Cannot retrieve locked instance %s" % self.op.instance_name
5849
    pnode = instance.primary_node
5850
    nodelist = list(instance.all_nodes)
5851

    
5852
    # hvparams processing
5853
    if self.op.hvparams:
5854
      i_hvdict = copy.deepcopy(instance.hvparams)
5855
      for key, val in self.op.hvparams.iteritems():
5856
        if val == constants.VALUE_DEFAULT:
5857
          try:
5858
            del i_hvdict[key]
5859
          except KeyError:
5860
            pass
5861
        else:
5862
          i_hvdict[key] = val
5863
      cluster = self.cfg.GetClusterInfo()
5864
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
5865
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
5866
                                i_hvdict)
5867
      # local check
5868
      hypervisor.GetHypervisor(
5869
        instance.hypervisor).CheckParameterSyntax(hv_new)
5870
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
5871
      self.hv_new = hv_new # the new actual values
5872
      self.hv_inst = i_hvdict # the new dict (without defaults)
5873
    else:
5874
      self.hv_new = self.hv_inst = {}
5875

    
5876
    # beparams processing
5877
    if self.op.beparams:
5878
      i_bedict = copy.deepcopy(instance.beparams)
5879
      for key, val in self.op.beparams.iteritems():
5880
        if val == constants.VALUE_DEFAULT:
5881
          try:
5882
            del i_bedict[key]
5883
          except KeyError:
5884
            pass
5885
        else:
5886
          i_bedict[key] = val
5887
      cluster = self.cfg.GetClusterInfo()
5888
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
5889
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
5890
                                i_bedict)
5891
      self.be_new = be_new # the new actual values
5892
      self.be_inst = i_bedict # the new dict (without defaults)
5893
    else:
5894
      self.be_new = self.be_inst = {}
5895

    
5896
    self.warn = []
5897

    
5898
    if constants.BE_MEMORY in self.op.beparams and not self.force:
5899
      mem_check_list = [pnode]
5900
      if be_new[constants.BE_AUTO_BALANCE]:
5901
        # either we changed auto_balance to yes or it was from before
5902
        mem_check_list.extend(instance.secondary_nodes)
5903
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
5904
                                                  instance.hypervisor)
5905
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
5906
                                         instance.hypervisor)
5907
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
5908
        # Assume the primary node is unreachable and go ahead
5909
        self.warn.append("Can't get info from primary node %s" % pnode)
5910
      else:
5911
        if not instance_info.failed and instance_info.data:
5912
          current_mem = instance_info.data['memory']
5913
        else:
5914
          # Assume instance not running
5915
          # (there is a slight race condition here, but it's not very probable,
5916
          # and we have no other way to check)
5917
          current_mem = 0
5918
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
5919
                    nodeinfo[pnode].data['memory_free'])
5920
        if miss_mem > 0:
5921
          raise errors.OpPrereqError("This change will prevent the instance"
5922
                                     " from starting, due to %d MB of memory"
5923
                                     " missing on its primary node" % miss_mem)
5924

    
5925
      if be_new[constants.BE_AUTO_BALANCE]:
5926
        for node, nres in nodeinfo.iteritems():
5927
          if node not in instance.secondary_nodes:
5928
            continue
5929
          if nres.failed or not isinstance(nres.data, dict):
5930
            self.warn.append("Can't get info from secondary node %s" % node)
5931
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
5932
            self.warn.append("Not enough memory to failover instance to"
5933
                             " secondary node %s" % node)
5934

    
5935
    # NIC processing
5936
    for nic_op, nic_dict in self.op.nics:
5937
      if nic_op == constants.DDM_REMOVE:
5938
        if not instance.nics:
5939
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
5940
        continue
5941
      if nic_op != constants.DDM_ADD:
5942
        # an existing nic
5943
        if nic_op < 0 or nic_op >= len(instance.nics):
5944
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
5945
                                     " are 0 to %d" %
5946
                                     (nic_op, len(instance.nics)))
5947
      if 'bridge' in nic_dict:
5948
        nic_bridge = nic_dict['bridge']
5949
        if nic_bridge is None:
5950
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
5951
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
5952
          msg = ("Bridge '%s' doesn't exist on one of"
5953
                 " the instance nodes" % nic_bridge)
5954
          if self.force:
5955
            self.warn.append(msg)
5956
          else:
5957
            raise errors.OpPrereqError(msg)
5958
      if 'mac' in nic_dict:
5959
        nic_mac = nic_dict['mac']
5960
        if nic_mac is None:
5961
          raise errors.OpPrereqError('Cannot set the nic mac to None')
5962
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5963
          # otherwise generate the mac
5964
          nic_dict['mac'] = self.cfg.GenerateMAC()
5965
        else:
5966
          # or validate/reserve the current one
5967
          if self.cfg.IsMacInUse(nic_mac):
5968
            raise errors.OpPrereqError("MAC address %s already in use"
5969
                                       " in cluster" % nic_mac)
5970

    
5971
    # DISK processing
5972
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
5973
      raise errors.OpPrereqError("Disk operations not supported for"
5974
                                 " diskless instances")
5975
    for disk_op, disk_dict in self.op.disks:
5976
      if disk_op == constants.DDM_REMOVE:
5977
        if len(instance.disks) == 1:
5978
          raise errors.OpPrereqError("Cannot remove the last disk of"
5979
                                     " an instance")
5980
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
5981
        ins_l = ins_l[pnode]
5982
        if ins_l.failed or not isinstance(ins_l.data, list):
5983
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
5984
        if instance.name in ins_l.data:
5985
          raise errors.OpPrereqError("Instance is running, can't remove"
5986
                                     " disks.")
5987

    
5988
      if (disk_op == constants.DDM_ADD and
5989
          len(instance.nics) >= constants.MAX_DISKS):
5990
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
5991
                                   " add more" % constants.MAX_DISKS)
5992
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
5993
        # an existing disk
5994
        if disk_op < 0 or disk_op >= len(instance.disks):
5995
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
5996
                                     " are 0 to %d" %
5997
                                     (disk_op, len(instance.disks)))
5998

    
5999
    return
6000

    
6001
  def Exec(self, feedback_fn):
6002
    """Modifies an instance.
6003

6004
    All parameters take effect only at the next restart of the instance.
6005

6006
    """
6007
    # Process here the warnings from CheckPrereq, as we don't have a
6008
    # feedback_fn there.
6009
    for warn in self.warn:
6010
      feedback_fn("WARNING: %s" % warn)
6011

    
6012
    result = []
6013
    instance = self.instance
6014
    # disk changes
6015
    for disk_op, disk_dict in self.op.disks:
6016
      if disk_op == constants.DDM_REMOVE:
6017
        # remove the last disk
6018
        device = instance.disks.pop()
6019
        device_idx = len(instance.disks)
6020
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6021
          self.cfg.SetDiskID(disk, node)
6022
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6023
          if msg:
6024
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6025
                            " continuing anyway", device_idx, node, msg)
6026
        result.append(("disk/%d" % device_idx, "remove"))
6027
      elif disk_op == constants.DDM_ADD:
6028
        # add a new disk
6029
        if instance.disk_template == constants.DT_FILE:
6030
          file_driver, file_path = instance.disks[0].logical_id
6031
          file_path = os.path.dirname(file_path)
6032
        else:
6033
          file_driver = file_path = None
6034
        disk_idx_base = len(instance.disks)
6035
        new_disk = _GenerateDiskTemplate(self,
6036
                                         instance.disk_template,
6037
                                         instance.name, instance.primary_node,
6038
                                         instance.secondary_nodes,
6039
                                         [disk_dict],
6040
                                         file_path,
6041
                                         file_driver,
6042
                                         disk_idx_base)[0]
6043
        instance.disks.append(new_disk)
6044
        info = _GetInstanceInfoText(instance)
6045

    
6046
        logging.info("Creating volume %s for instance %s",
6047
                     new_disk.iv_name, instance.name)
6048
        # Note: this needs to be kept in sync with _CreateDisks
6049
        #HARDCODE
6050
        for node in instance.all_nodes:
6051
          f_create = node == instance.primary_node
6052
          try:
6053
            _CreateBlockDev(self, node, instance, new_disk,
6054
                            f_create, info, f_create)
6055
          except errors.OpExecError, err:
6056
            self.LogWarning("Failed to create volume %s (%s) on"
6057
                            " node %s: %s",
6058
                            new_disk.iv_name, new_disk, node, err)
6059
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6060
                       (new_disk.size, new_disk.mode)))
6061
      else:
6062
        # change a given disk
6063
        instance.disks[disk_op].mode = disk_dict['mode']
6064
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6065
    # NIC changes
6066
    for nic_op, nic_dict in self.op.nics:
6067
      if nic_op == constants.DDM_REMOVE:
6068
        # remove the last nic
6069
        del instance.nics[-1]
6070
        result.append(("nic.%d" % len(instance.nics), "remove"))
6071
      elif nic_op == constants.DDM_ADD:
6072
        # mac and bridge should be set, by now
6073
        mac = nic_dict['mac']
6074
        bridge = nic_dict['bridge']
6075
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6076
                              bridge=bridge)
6077
        instance.nics.append(new_nic)
6078
        result.append(("nic.%d" % (len(instance.nics) - 1),
6079
                       "add:mac=%s,ip=%s,bridge=%s" %
6080
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6081
      else:
6082
        # change a given nic
6083
        for key in 'mac', 'ip', 'bridge':
6084
          if key in nic_dict:
6085
            setattr(instance.nics[nic_op], key, nic_dict[key])
6086
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6087

    
6088
    # hvparams changes
6089
    if self.op.hvparams:
6090
      instance.hvparams = self.hv_inst
6091
      for key, val in self.op.hvparams.iteritems():
6092
        result.append(("hv/%s" % key, val))
6093

    
6094
    # beparams changes
6095
    if self.op.beparams:
6096
      instance.beparams = self.be_inst
6097
      for key, val in self.op.beparams.iteritems():
6098
        result.append(("be/%s" % key, val))
6099

    
6100
    self.cfg.Update(instance)
6101

    
6102
    return result
6103

    
6104

    
6105
class LUQueryExports(NoHooksLU):
6106
  """Query the exports list
6107

6108
  """
6109
  _OP_REQP = ['nodes']
6110
  REQ_BGL = False
6111

    
6112
  def ExpandNames(self):
6113
    self.needed_locks = {}
6114
    self.share_locks[locking.LEVEL_NODE] = 1
6115
    if not self.op.nodes:
6116
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6117
    else:
6118
      self.needed_locks[locking.LEVEL_NODE] = \
6119
        _GetWantedNodes(self, self.op.nodes)
6120

    
6121
  def CheckPrereq(self):
6122
    """Check prerequisites.
6123

6124
    """
6125
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6126

    
6127
  def Exec(self, feedback_fn):
6128
    """Compute the list of all the exported system images.
6129

6130
    @rtype: dict
6131
    @return: a dictionary with the structure node->(export-list)
6132
        where export-list is a list of the instances exported on
6133
        that node.
6134

6135
    """
6136
    rpcresult = self.rpc.call_export_list(self.nodes)
6137
    result = {}
6138
    for node in rpcresult:
6139
      if rpcresult[node].failed:
6140
        result[node] = False
6141
      else:
6142
        result[node] = rpcresult[node].data
6143

    
6144
    return result
6145

    
6146

    
6147
class LUExportInstance(LogicalUnit):
6148
  """Export an instance to an image in the cluster.
6149

6150
  """
6151
  HPATH = "instance-export"
6152
  HTYPE = constants.HTYPE_INSTANCE
6153
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6154
  REQ_BGL = False
6155

    
6156
  def ExpandNames(self):
6157
    self._ExpandAndLockInstance()
6158
    # FIXME: lock only instance primary and destination node
6159
    #
6160
    # Sad but true, for now we have do lock all nodes, as we don't know where
6161
    # the previous export might be, and and in this LU we search for it and
6162
    # remove it from its current node. In the future we could fix this by:
6163
    #  - making a tasklet to search (share-lock all), then create the new one,
6164
    #    then one to remove, after
6165
    #  - removing the removal operation altoghether
6166
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6167

    
6168
  def DeclareLocks(self, level):
6169
    """Last minute lock declaration."""
6170
    # All nodes are locked anyway, so nothing to do here.
6171

    
6172
  def BuildHooksEnv(self):
6173
    """Build hooks env.
6174

6175
    This will run on the master, primary node and target node.
6176

6177
    """
6178
    env = {
6179
      "EXPORT_NODE": self.op.target_node,
6180
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6181
      }
6182
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6183
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6184
          self.op.target_node]
6185
    return env, nl, nl
6186

    
6187
  def CheckPrereq(self):
6188
    """Check prerequisites.
6189

6190
    This checks that the instance and node names are valid.
6191

6192
    """
6193
    instance_name = self.op.instance_name
6194
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6195
    assert self.instance is not None, \
6196
          "Cannot retrieve locked instance %s" % self.op.instance_name
6197
    _CheckNodeOnline(self, self.instance.primary_node)
6198

    
6199
    self.dst_node = self.cfg.GetNodeInfo(
6200
      self.cfg.ExpandNodeName(self.op.target_node))
6201

    
6202
    if self.dst_node is None:
6203
      # This is wrong node name, not a non-locked node
6204
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6205
    _CheckNodeOnline(self, self.dst_node.name)
6206
    _CheckNodeNotDrained(self, self.dst_node.name)
6207

    
6208
    # instance disk type verification
6209
    for disk in self.instance.disks:
6210
      if disk.dev_type == constants.LD_FILE:
6211
        raise errors.OpPrereqError("Export not supported for instances with"
6212
                                   " file-based disks")
6213

    
6214
  def Exec(self, feedback_fn):
6215
    """Export an instance to an image in the cluster.
6216

6217
    """
6218
    instance = self.instance
6219
    dst_node = self.dst_node
6220
    src_node = instance.primary_node
6221
    if self.op.shutdown:
6222
      # shutdown the instance, but not the disks
6223
      result = self.rpc.call_instance_shutdown(src_node, instance)
6224
      msg = result.RemoteFailMsg()
6225
      if msg:
6226
        raise errors.OpExecError("Could not shutdown instance %s on"
6227
                                 " node %s: %s" %
6228
                                 (instance.name, src_node, msg))
6229

    
6230
    vgname = self.cfg.GetVGName()
6231

    
6232
    snap_disks = []
6233

    
6234
    # set the disks ID correctly since call_instance_start needs the
6235
    # correct drbd minor to create the symlinks
6236
    for disk in instance.disks:
6237
      self.cfg.SetDiskID(disk, src_node)
6238

    
6239
    try:
6240
      for disk in instance.disks:
6241
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6242
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6243
        if new_dev_name.failed or not new_dev_name.data:
6244
          self.LogWarning("Could not snapshot block device %s on node %s",
6245
                          disk.logical_id[1], src_node)
6246
          snap_disks.append(False)
6247
        else:
6248
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6249
                                 logical_id=(vgname, new_dev_name.data),
6250
                                 physical_id=(vgname, new_dev_name.data),
6251
                                 iv_name=disk.iv_name)
6252
          snap_disks.append(new_dev)
6253

    
6254
    finally:
6255
      if self.op.shutdown and instance.admin_up:
6256
        result = self.rpc.call_instance_start(src_node, instance)
6257
        msg = result.RemoteFailMsg()
6258
        if msg:
6259
          _ShutdownInstanceDisks(self, instance)
6260
          raise errors.OpExecError("Could not start instance: %s" % msg)
6261

    
6262
    # TODO: check for size
6263

    
6264
    cluster_name = self.cfg.GetClusterName()
6265
    for idx, dev in enumerate(snap_disks):
6266
      if dev:
6267
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6268
                                               instance, cluster_name, idx)
6269
        if result.failed or not result.data:
6270
          self.LogWarning("Could not export block device %s from node %s to"
6271
                          " node %s", dev.logical_id[1], src_node,
6272
                          dst_node.name)
6273
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6274
        if msg:
6275
          self.LogWarning("Could not remove snapshot block device %s from node"
6276
                          " %s: %s", dev.logical_id[1], src_node, msg)
6277

    
6278
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6279
    if result.failed or not result.data:
6280
      self.LogWarning("Could not finalize export for instance %s on node %s",
6281
                      instance.name, dst_node.name)
6282

    
6283
    nodelist = self.cfg.GetNodeList()
6284
    nodelist.remove(dst_node.name)
6285

    
6286
    # on one-node clusters nodelist will be empty after the removal
6287
    # if we proceed the backup would be removed because OpQueryExports
6288
    # substitutes an empty list with the full cluster node list.
6289
    if nodelist:
6290
      exportlist = self.rpc.call_export_list(nodelist)
6291
      for node in exportlist:
6292
        if exportlist[node].failed:
6293
          continue
6294
        if instance.name in exportlist[node].data:
6295
          if not self.rpc.call_export_remove(node, instance.name):
6296
            self.LogWarning("Could not remove older export for instance %s"
6297
                            " on node %s", instance.name, node)
6298

    
6299

    
6300
class LURemoveExport(NoHooksLU):
6301
  """Remove exports related to the named instance.
6302

6303
  """
6304
  _OP_REQP = ["instance_name"]
6305
  REQ_BGL = False
6306

    
6307
  def ExpandNames(self):
6308
    self.needed_locks = {}
6309
    # We need all nodes to be locked in order for RemoveExport to work, but we
6310
    # don't need to lock the instance itself, as nothing will happen to it (and
6311
    # we can remove exports also for a removed instance)
6312
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6313

    
6314
  def CheckPrereq(self):
6315
    """Check prerequisites.
6316
    """
6317
    pass
6318

    
6319
  def Exec(self, feedback_fn):
6320
    """Remove any export.
6321

6322
    """
6323
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6324
    # If the instance was not found we'll try with the name that was passed in.
6325
    # This will only work if it was an FQDN, though.
6326
    fqdn_warn = False
6327
    if not instance_name:
6328
      fqdn_warn = True
6329
      instance_name = self.op.instance_name
6330

    
6331
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6332
      locking.LEVEL_NODE])
6333
    found = False
6334
    for node in exportlist:
6335
      if exportlist[node].failed:
6336
        self.LogWarning("Failed to query node %s, continuing" % node)
6337
        continue
6338
      if instance_name in exportlist[node].data:
6339
        found = True
6340
        result = self.rpc.call_export_remove(node, instance_name)
6341
        if result.failed or not result.data:
6342
          logging.error("Could not remove export for instance %s"
6343
                        " on node %s", instance_name, node)
6344

    
6345
    if fqdn_warn and not found:
6346
      feedback_fn("Export not found. If trying to remove an export belonging"
6347
                  " to a deleted instance please use its Fully Qualified"
6348
                  " Domain Name.")
6349

    
6350

    
6351
class TagsLU(NoHooksLU):
6352
  """Generic tags LU.
6353

6354
  This is an abstract class which is the parent of all the other tags LUs.
6355

6356
  """
6357

    
6358
  def ExpandNames(self):
6359
    self.needed_locks = {}
6360
    if self.op.kind == constants.TAG_NODE:
6361
      name = self.cfg.ExpandNodeName(self.op.name)
6362
      if name is None:
6363
        raise errors.OpPrereqError("Invalid node name (%s)" %
6364
                                   (self.op.name,))
6365
      self.op.name = name
6366
      self.needed_locks[locking.LEVEL_NODE] = name
6367
    elif self.op.kind == constants.TAG_INSTANCE:
6368
      name = self.cfg.ExpandInstanceName(self.op.name)
6369
      if name is None:
6370
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6371
                                   (self.op.name,))
6372
      self.op.name = name
6373
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6374

    
6375
  def CheckPrereq(self):
6376
    """Check prerequisites.
6377

6378
    """
6379
    if self.op.kind == constants.TAG_CLUSTER:
6380
      self.target = self.cfg.GetClusterInfo()
6381
    elif self.op.kind == constants.TAG_NODE:
6382
      self.target = self.cfg.GetNodeInfo(self.op.name)
6383
    elif self.op.kind == constants.TAG_INSTANCE:
6384
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6385
    else:
6386
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6387
                                 str(self.op.kind))
6388

    
6389

    
6390
class LUGetTags(TagsLU):
6391
  """Returns the tags of a given object.
6392

6393
  """
6394
  _OP_REQP = ["kind", "name"]
6395
  REQ_BGL = False
6396

    
6397
  def Exec(self, feedback_fn):
6398
    """Returns the tag list.
6399

6400
    """
6401
    return list(self.target.GetTags())
6402

    
6403

    
6404
class LUSearchTags(NoHooksLU):
6405
  """Searches the tags for a given pattern.
6406

6407
  """
6408
  _OP_REQP = ["pattern"]
6409
  REQ_BGL = False
6410

    
6411
  def ExpandNames(self):
6412
    self.needed_locks = {}
6413

    
6414
  def CheckPrereq(self):
6415
    """Check prerequisites.
6416

6417
    This checks the pattern passed for validity by compiling it.
6418

6419
    """
6420
    try:
6421
      self.re = re.compile(self.op.pattern)
6422
    except re.error, err:
6423
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6424
                                 (self.op.pattern, err))
6425

    
6426
  def Exec(self, feedback_fn):
6427
    """Returns the tag list.
6428

6429
    """
6430
    cfg = self.cfg
6431
    tgts = [("/cluster", cfg.GetClusterInfo())]
6432
    ilist = cfg.GetAllInstancesInfo().values()
6433
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6434
    nlist = cfg.GetAllNodesInfo().values()
6435
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6436
    results = []
6437
    for path, target in tgts:
6438
      for tag in target.GetTags():
6439
        if self.re.search(tag):
6440
          results.append((path, tag))
6441
    return results
6442

    
6443

    
6444
class LUAddTags(TagsLU):
6445
  """Sets a tag on a given object.
6446

6447
  """
6448
  _OP_REQP = ["kind", "name", "tags"]
6449
  REQ_BGL = False
6450

    
6451
  def CheckPrereq(self):
6452
    """Check prerequisites.
6453

6454
    This checks the type and length of the tag name and value.
6455

6456
    """
6457
    TagsLU.CheckPrereq(self)
6458
    for tag in self.op.tags:
6459
      objects.TaggableObject.ValidateTag(tag)
6460

    
6461
  def Exec(self, feedback_fn):
6462
    """Sets the tag.
6463

6464
    """
6465
    try:
6466
      for tag in self.op.tags:
6467
        self.target.AddTag(tag)
6468
    except errors.TagError, err:
6469
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6470
    try:
6471
      self.cfg.Update(self.target)
6472
    except errors.ConfigurationError:
6473
      raise errors.OpRetryError("There has been a modification to the"
6474
                                " config file and the operation has been"
6475
                                " aborted. Please retry.")
6476

    
6477

    
6478
class LUDelTags(TagsLU):
6479
  """Delete a list of tags from a given object.
6480

6481
  """
6482
  _OP_REQP = ["kind", "name", "tags"]
6483
  REQ_BGL = False
6484

    
6485
  def CheckPrereq(self):
6486
    """Check prerequisites.
6487

6488
    This checks that we have the given tag.
6489

6490
    """
6491
    TagsLU.CheckPrereq(self)
6492
    for tag in self.op.tags:
6493
      objects.TaggableObject.ValidateTag(tag)
6494
    del_tags = frozenset(self.op.tags)
6495
    cur_tags = self.target.GetTags()
6496
    if not del_tags <= cur_tags:
6497
      diff_tags = del_tags - cur_tags
6498
      diff_names = ["'%s'" % tag for tag in diff_tags]
6499
      diff_names.sort()
6500
      raise errors.OpPrereqError("Tag(s) %s not found" %
6501
                                 (",".join(diff_names)))
6502

    
6503
  def Exec(self, feedback_fn):
6504
    """Remove the tag from the object.
6505

6506
    """
6507
    for tag in self.op.tags:
6508
      self.target.RemoveTag(tag)
6509
    try:
6510
      self.cfg.Update(self.target)
6511
    except errors.ConfigurationError:
6512
      raise errors.OpRetryError("There has been a modification to the"
6513
                                " config file and the operation has been"
6514
                                " aborted. Please retry.")
6515

    
6516

    
6517
class LUTestDelay(NoHooksLU):
6518
  """Sleep for a specified amount of time.
6519

6520
  This LU sleeps on the master and/or nodes for a specified amount of
6521
  time.
6522

6523
  """
6524
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6525
  REQ_BGL = False
6526

    
6527
  def ExpandNames(self):
6528
    """Expand names and set required locks.
6529

6530
    This expands the node list, if any.
6531

6532
    """
6533
    self.needed_locks = {}
6534
    if self.op.on_nodes:
6535
      # _GetWantedNodes can be used here, but is not always appropriate to use
6536
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6537
      # more information.
6538
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6539
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6540

    
6541
  def CheckPrereq(self):
6542
    """Check prerequisites.
6543

6544
    """
6545

    
6546
  def Exec(self, feedback_fn):
6547
    """Do the actual sleep.
6548

6549
    """
6550
    if self.op.on_master:
6551
      if not utils.TestDelay(self.op.duration):
6552
        raise errors.OpExecError("Error during master delay test")
6553
    if self.op.on_nodes:
6554
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6555
      if not result:
6556
        raise errors.OpExecError("Complete failure from rpc call")
6557
      for node, node_result in result.items():
6558
        node_result.Raise()
6559
        if not node_result.data:
6560
          raise errors.OpExecError("Failure during rpc call to node %s,"
6561
                                   " result: %s" % (node, node_result.data))
6562

    
6563

    
6564
class IAllocator(object):
6565
  """IAllocator framework.
6566

6567
  An IAllocator instance has three sets of attributes:
6568
    - cfg that is needed to query the cluster
6569
    - input data (all members of the _KEYS class attribute are required)
6570
    - four buffer attributes (in|out_data|text), that represent the
6571
      input (to the external script) in text and data structure format,
6572
      and the output from it, again in two formats
6573
    - the result variables from the script (success, info, nodes) for
6574
      easy usage
6575

6576
  """
6577
  _ALLO_KEYS = [
6578
    "mem_size", "disks", "disk_template",
6579
    "os", "tags", "nics", "vcpus", "hypervisor",
6580
    ]
6581
  _RELO_KEYS = [
6582
    "relocate_from",
6583
    ]
6584

    
6585
  def __init__(self, lu, mode, name, **kwargs):
6586
    self.lu = lu
6587
    # init buffer variables
6588
    self.in_text = self.out_text = self.in_data = self.out_data = None
6589
    # init all input fields so that pylint is happy
6590
    self.mode = mode
6591
    self.name = name
6592
    self.mem_size = self.disks = self.disk_template = None
6593
    self.os = self.tags = self.nics = self.vcpus = None
6594
    self.hypervisor = None
6595
    self.relocate_from = None
6596
    # computed fields
6597
    self.required_nodes = None
6598
    # init result fields
6599
    self.success = self.info = self.nodes = None
6600
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6601
      keyset = self._ALLO_KEYS
6602
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6603
      keyset = self._RELO_KEYS
6604
    else:
6605
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6606
                                   " IAllocator" % self.mode)
6607
    for key in kwargs:
6608
      if key not in keyset:
6609
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6610
                                     " IAllocator" % key)
6611
      setattr(self, key, kwargs[key])
6612
    for key in keyset:
6613
      if key not in kwargs:
6614
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6615
                                     " IAllocator" % key)
6616
    self._BuildInputData()
6617

    
6618
  def _ComputeClusterData(self):
6619
    """Compute the generic allocator input data.
6620

6621
    This is the data that is independent of the actual operation.
6622

6623
    """
6624
    cfg = self.lu.cfg
6625
    cluster_info = cfg.GetClusterInfo()
6626
    # cluster data
6627
    data = {
6628
      "version": constants.IALLOCATOR_VERSION,
6629
      "cluster_name": cfg.GetClusterName(),
6630
      "cluster_tags": list(cluster_info.GetTags()),
6631
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6632
      # we don't have job IDs
6633
      }
6634
    iinfo = cfg.GetAllInstancesInfo().values()
6635
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6636

    
6637
    # node data
6638
    node_results = {}
6639
    node_list = cfg.GetNodeList()
6640

    
6641
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6642
      hypervisor_name = self.hypervisor
6643
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6644
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6645

    
6646
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6647
                                           hypervisor_name)
6648
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6649
                       cluster_info.enabled_hypervisors)
6650
    for nname, nresult in node_data.items():
6651
      # first fill in static (config-based) values
6652
      ninfo = cfg.GetNodeInfo(nname)
6653
      pnr = {
6654
        "tags": list(ninfo.GetTags()),
6655
        "primary_ip": ninfo.primary_ip,
6656
        "secondary_ip": ninfo.secondary_ip,
6657
        "offline": ninfo.offline,
6658
        "drained": ninfo.drained,
6659
        "master_candidate": ninfo.master_candidate,
6660
        }
6661

    
6662
      if not ninfo.offline:
6663
        nresult.Raise()
6664
        if not isinstance(nresult.data, dict):
6665
          raise errors.OpExecError("Can't get data for node %s" % nname)
6666
        remote_info = nresult.data
6667
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6668
                     'vg_size', 'vg_free', 'cpu_total']:
6669
          if attr not in remote_info:
6670
            raise errors.OpExecError("Node '%s' didn't return attribute"
6671
                                     " '%s'" % (nname, attr))
6672
          try:
6673
            remote_info[attr] = int(remote_info[attr])
6674
          except ValueError, err:
6675
            raise errors.OpExecError("Node '%s' returned invalid value"
6676
                                     " for '%s': %s" % (nname, attr, err))
6677
        # compute memory used by primary instances
6678
        i_p_mem = i_p_up_mem = 0
6679
        for iinfo, beinfo in i_list:
6680
          if iinfo.primary_node == nname:
6681
            i_p_mem += beinfo[constants.BE_MEMORY]
6682
            if iinfo.name not in node_iinfo[nname].data:
6683
              i_used_mem = 0
6684
            else:
6685
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6686
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6687
            remote_info['memory_free'] -= max(0, i_mem_diff)
6688

    
6689
            if iinfo.admin_up:
6690
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6691

    
6692
        # compute memory used by instances
6693
        pnr_dyn = {
6694
          "total_memory": remote_info['memory_total'],
6695
          "reserved_memory": remote_info['memory_dom0'],
6696
          "free_memory": remote_info['memory_free'],
6697
          "total_disk": remote_info['vg_size'],
6698
          "free_disk": remote_info['vg_free'],
6699
          "total_cpus": remote_info['cpu_total'],
6700
          "i_pri_memory": i_p_mem,
6701
          "i_pri_up_memory": i_p_up_mem,
6702
          }
6703
        pnr.update(pnr_dyn)
6704

    
6705
      node_results[nname] = pnr
6706
    data["nodes"] = node_results
6707

    
6708
    # instance data
6709
    instance_data = {}
6710
    for iinfo, beinfo in i_list:
6711
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6712
                  for n in iinfo.nics]
6713
      pir = {
6714
        "tags": list(iinfo.GetTags()),
6715
        "admin_up": iinfo.admin_up,
6716
        "vcpus": beinfo[constants.BE_VCPUS],
6717
        "memory": beinfo[constants.BE_MEMORY],
6718
        "os": iinfo.os,
6719
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6720
        "nics": nic_data,
6721
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6722
        "disk_template": iinfo.disk_template,
6723
        "hypervisor": iinfo.hypervisor,
6724
        }
6725
      instance_data[iinfo.name] = pir
6726

    
6727
    data["instances"] = instance_data
6728

    
6729
    self.in_data = data
6730

    
6731
  def _AddNewInstance(self):
6732
    """Add new instance data to allocator structure.
6733

6734
    This in combination with _AllocatorGetClusterData will create the
6735
    correct structure needed as input for the allocator.
6736

6737
    The checks for the completeness of the opcode must have already been
6738
    done.
6739

6740
    """
6741
    data = self.in_data
6742

    
6743
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6744

    
6745
    if self.disk_template in constants.DTS_NET_MIRROR:
6746
      self.required_nodes = 2
6747
    else:
6748
      self.required_nodes = 1
6749
    request = {
6750
      "type": "allocate",
6751
      "name": self.name,
6752
      "disk_template": self.disk_template,
6753
      "tags": self.tags,
6754
      "os": self.os,
6755
      "vcpus": self.vcpus,
6756
      "memory": self.mem_size,
6757
      "disks": self.disks,
6758
      "disk_space_total": disk_space,
6759
      "nics": self.nics,
6760
      "required_nodes": self.required_nodes,
6761
      }
6762
    data["request"] = request
6763

    
6764
  def _AddRelocateInstance(self):
6765
    """Add relocate instance data to allocator structure.
6766

6767
    This in combination with _IAllocatorGetClusterData will create the
6768
    correct structure needed as input for the allocator.
6769

6770
    The checks for the completeness of the opcode must have already been
6771
    done.
6772

6773
    """
6774
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6775
    if instance is None:
6776
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6777
                                   " IAllocator" % self.name)
6778

    
6779
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6780
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6781

    
6782
    if len(instance.secondary_nodes) != 1:
6783
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6784

    
6785
    self.required_nodes = 1
6786
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6787
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6788

    
6789
    request = {
6790
      "type": "relocate",
6791
      "name": self.name,
6792
      "disk_space_total": disk_space,
6793
      "required_nodes": self.required_nodes,
6794
      "relocate_from": self.relocate_from,
6795
      }
6796
    self.in_data["request"] = request
6797

    
6798
  def _BuildInputData(self):
6799
    """Build input data structures.
6800

6801
    """
6802
    self._ComputeClusterData()
6803

    
6804
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6805
      self._AddNewInstance()
6806
    else:
6807
      self._AddRelocateInstance()
6808

    
6809
    self.in_text = serializer.Dump(self.in_data)
6810

    
6811
  def Run(self, name, validate=True, call_fn=None):
6812
    """Run an instance allocator and return the results.
6813

6814
    """
6815
    if call_fn is None:
6816
      call_fn = self.lu.rpc.call_iallocator_runner
6817
    data = self.in_text
6818

    
6819
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6820
    result.Raise()
6821

    
6822
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6823
      raise errors.OpExecError("Invalid result from master iallocator runner")
6824

    
6825
    rcode, stdout, stderr, fail = result.data
6826

    
6827
    if rcode == constants.IARUN_NOTFOUND:
6828
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6829
    elif rcode == constants.IARUN_FAILURE:
6830
      raise errors.OpExecError("Instance allocator call failed: %s,"
6831
                               " output: %s" % (fail, stdout+stderr))
6832
    self.out_text = stdout
6833
    if validate:
6834
      self._ValidateResult()
6835

    
6836
  def _ValidateResult(self):
6837
    """Process the allocator results.
6838

6839
    This will process and if successful save the result in
6840
    self.out_data and the other parameters.
6841

6842
    """
6843
    try:
6844
      rdict = serializer.Load(self.out_text)
6845
    except Exception, err:
6846
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6847

    
6848
    if not isinstance(rdict, dict):
6849
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6850

    
6851
    for key in "success", "info", "nodes":
6852
      if key not in rdict:
6853
        raise errors.OpExecError("Can't parse iallocator results:"
6854
                                 " missing key '%s'" % key)
6855
      setattr(self, key, rdict[key])
6856

    
6857
    if not isinstance(rdict["nodes"], list):
6858
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6859
                               " is not a list")
6860
    self.out_data = rdict
6861

    
6862

    
6863
class LUTestAllocator(NoHooksLU):
6864
  """Run allocator tests.
6865

6866
  This LU runs the allocator tests
6867

6868
  """
6869
  _OP_REQP = ["direction", "mode", "name"]
6870

    
6871
  def CheckPrereq(self):
6872
    """Check prerequisites.
6873

6874
    This checks the opcode parameters depending on the director and mode test.
6875

6876
    """
6877
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6878
      for attr in ["name", "mem_size", "disks", "disk_template",
6879
                   "os", "tags", "nics", "vcpus"]:
6880
        if not hasattr(self.op, attr):
6881
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
6882
                                     attr)
6883
      iname = self.cfg.ExpandInstanceName(self.op.name)
6884
      if iname is not None:
6885
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
6886
                                   iname)
6887
      if not isinstance(self.op.nics, list):
6888
        raise errors.OpPrereqError("Invalid parameter 'nics'")
6889
      for row in self.op.nics:
6890
        if (not isinstance(row, dict) or
6891
            "mac" not in row or
6892
            "ip" not in row or
6893
            "bridge" not in row):
6894
          raise errors.OpPrereqError("Invalid contents of the"
6895
                                     " 'nics' parameter")
6896
      if not isinstance(self.op.disks, list):
6897
        raise errors.OpPrereqError("Invalid parameter 'disks'")
6898
      for row in self.op.disks:
6899
        if (not isinstance(row, dict) or
6900
            "size" not in row or
6901
            not isinstance(row["size"], int) or
6902
            "mode" not in row or
6903
            row["mode"] not in ['r', 'w']):
6904
          raise errors.OpPrereqError("Invalid contents of the"
6905
                                     " 'disks' parameter")
6906
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
6907
        self.op.hypervisor = self.cfg.GetHypervisorType()
6908
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
6909
      if not hasattr(self.op, "name"):
6910
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
6911
      fname = self.cfg.ExpandInstanceName(self.op.name)
6912
      if fname is None:
6913
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
6914
                                   self.op.name)
6915
      self.op.name = fname
6916
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
6917
    else:
6918
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
6919
                                 self.op.mode)
6920

    
6921
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
6922
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
6923
        raise errors.OpPrereqError("Missing allocator name")
6924
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
6925
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
6926
                                 self.op.direction)
6927

    
6928
  def Exec(self, feedback_fn):
6929
    """Run the allocator test.
6930

6931
    """
6932
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6933
      ial = IAllocator(self,
6934
                       mode=self.op.mode,
6935
                       name=self.op.name,
6936
                       mem_size=self.op.mem_size,
6937
                       disks=self.op.disks,
6938
                       disk_template=self.op.disk_template,
6939
                       os=self.op.os,
6940
                       tags=self.op.tags,
6941
                       nics=self.op.nics,
6942
                       vcpus=self.op.vcpus,
6943
                       hypervisor=self.op.hypervisor,
6944
                       )
6945
    else:
6946
      ial = IAllocator(self,
6947
                       mode=self.op.mode,
6948
                       name=self.op.name,
6949
                       relocate_from=list(self.relocate_from),
6950
                       )
6951

    
6952
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
6953
      result = ial.in_text
6954
    else:
6955
      ial.Run(self.op.allocator, validate=False)
6956
      result = ial.out_text
6957
    return result