Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 0eca8e0c

History | View | Annotate | Download (242.1 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, None, None)
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, None, None)
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
    remote_info.Raise()
2975
    if remote_info.data:
2976
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
2977
                                 (self.op.instance_name,
2978
                                  instance.primary_node))
2979

    
2980
    self.op.os_type = getattr(self.op, "os_type", None)
2981
    if self.op.os_type is not None:
2982
      # OS verification
2983
      pnode = self.cfg.GetNodeInfo(
2984
        self.cfg.ExpandNodeName(instance.primary_node))
2985
      if pnode is None:
2986
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
2987
                                   self.op.pnode)
2988
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
2989
      result.Raise()
2990
      if not isinstance(result.data, objects.OS):
2991
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
2992
                                   " primary node"  % self.op.os_type)
2993

    
2994
    self.instance = instance
2995

    
2996
  def Exec(self, feedback_fn):
2997
    """Reinstall the instance.
2998

2999
    """
3000
    inst = self.instance
3001

    
3002
    if self.op.os_type is not None:
3003
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3004
      inst.os = self.op.os_type
3005
      self.cfg.Update(inst)
3006

    
3007
    _StartInstanceDisks(self, inst, None)
3008
    try:
3009
      feedback_fn("Running the instance OS create scripts...")
3010
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3011
      msg = result.RemoteFailMsg()
3012
      if msg:
3013
        raise errors.OpExecError("Could not install OS for instance %s"
3014
                                 " on node %s: %s" %
3015
                                 (inst.name, inst.primary_node, msg))
3016
    finally:
3017
      _ShutdownInstanceDisks(self, inst)
3018

    
3019

    
3020
class LURenameInstance(LogicalUnit):
3021
  """Rename an instance.
3022

3023
  """
3024
  HPATH = "instance-rename"
3025
  HTYPE = constants.HTYPE_INSTANCE
3026
  _OP_REQP = ["instance_name", "new_name"]
3027

    
3028
  def BuildHooksEnv(self):
3029
    """Build hooks env.
3030

3031
    This runs on master, primary and secondary nodes of the instance.
3032

3033
    """
3034
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3035
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3036
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3037
    return env, nl, nl
3038

    
3039
  def CheckPrereq(self):
3040
    """Check prerequisites.
3041

3042
    This checks that the instance is in the cluster and is not running.
3043

3044
    """
3045
    instance = self.cfg.GetInstanceInfo(
3046
      self.cfg.ExpandInstanceName(self.op.instance_name))
3047
    if instance is None:
3048
      raise errors.OpPrereqError("Instance '%s' not known" %
3049
                                 self.op.instance_name)
3050
    _CheckNodeOnline(self, instance.primary_node)
3051

    
3052
    if instance.admin_up:
3053
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3054
                                 self.op.instance_name)
3055
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3056
                                              instance.name,
3057
                                              instance.hypervisor)
3058
    remote_info.Raise()
3059
    if remote_info.data:
3060
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3061
                                 (self.op.instance_name,
3062
                                  instance.primary_node))
3063
    self.instance = instance
3064

    
3065
    # new name verification
3066
    name_info = utils.HostInfo(self.op.new_name)
3067

    
3068
    self.op.new_name = new_name = name_info.name
3069
    instance_list = self.cfg.GetInstanceList()
3070
    if new_name in instance_list:
3071
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3072
                                 new_name)
3073

    
3074
    if not getattr(self.op, "ignore_ip", False):
3075
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3076
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3077
                                   (name_info.ip, new_name))
3078

    
3079

    
3080
  def Exec(self, feedback_fn):
3081
    """Reinstall the instance.
3082

3083
    """
3084
    inst = self.instance
3085
    old_name = inst.name
3086

    
3087
    if inst.disk_template == constants.DT_FILE:
3088
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3089

    
3090
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3091
    # Change the instance lock. This is definitely safe while we hold the BGL
3092
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3093
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3094

    
3095
    # re-read the instance from the configuration after rename
3096
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3097

    
3098
    if inst.disk_template == constants.DT_FILE:
3099
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3100
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3101
                                                     old_file_storage_dir,
3102
                                                     new_file_storage_dir)
3103
      result.Raise()
3104
      if not result.data:
3105
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3106
                                 " directory '%s' to '%s' (but the instance"
3107
                                 " has been renamed in Ganeti)" % (
3108
                                 inst.primary_node, old_file_storage_dir,
3109
                                 new_file_storage_dir))
3110

    
3111
      if not result.data[0]:
3112
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3113
                                 " (but the instance has been renamed in"
3114
                                 " Ganeti)" % (old_file_storage_dir,
3115
                                               new_file_storage_dir))
3116

    
3117
    _StartInstanceDisks(self, inst, None)
3118
    try:
3119
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3120
                                                 old_name)
3121
      msg = result.RemoteFailMsg()
3122
      if msg:
3123
        msg = ("Could not run OS rename script for instance %s on node %s"
3124
               " (but the instance has been renamed in Ganeti): %s" %
3125
               (inst.name, inst.primary_node, msg))
3126
        self.proc.LogWarning(msg)
3127
    finally:
3128
      _ShutdownInstanceDisks(self, inst)
3129

    
3130

    
3131
class LURemoveInstance(LogicalUnit):
3132
  """Remove an instance.
3133

3134
  """
3135
  HPATH = "instance-remove"
3136
  HTYPE = constants.HTYPE_INSTANCE
3137
  _OP_REQP = ["instance_name", "ignore_failures"]
3138
  REQ_BGL = False
3139

    
3140
  def ExpandNames(self):
3141
    self._ExpandAndLockInstance()
3142
    self.needed_locks[locking.LEVEL_NODE] = []
3143
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3144

    
3145
  def DeclareLocks(self, level):
3146
    if level == locking.LEVEL_NODE:
3147
      self._LockInstancesNodes()
3148

    
3149
  def BuildHooksEnv(self):
3150
    """Build hooks env.
3151

3152
    This runs on master, primary and secondary nodes of the instance.
3153

3154
    """
3155
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3156
    nl = [self.cfg.GetMasterNode()]
3157
    return env, nl, nl
3158

    
3159
  def CheckPrereq(self):
3160
    """Check prerequisites.
3161

3162
    This checks that the instance is in the cluster.
3163

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

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

3172
    """
3173
    instance = self.instance
3174
    logging.info("Shutting down instance %s on node %s",
3175
                 instance.name, instance.primary_node)
3176

    
3177
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3178
    msg = result.RemoteFailMsg()
3179
    if msg:
3180
      if self.op.ignore_failures:
3181
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3182
      else:
3183
        raise errors.OpExecError("Could not shutdown instance %s on"
3184
                                 " node %s: %s" %
3185
                                 (instance.name, instance.primary_node, msg))
3186

    
3187
    logging.info("Removing block devices for instance %s", instance.name)
3188

    
3189
    if not _RemoveDisks(self, instance):
3190
      if self.op.ignore_failures:
3191
        feedback_fn("Warning: can't remove instance's disks")
3192
      else:
3193
        raise errors.OpExecError("Can't remove instance's disks")
3194

    
3195
    logging.info("Removing instance %s out of cluster config", instance.name)
3196

    
3197
    self.cfg.RemoveInstance(instance.name)
3198
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3199

    
3200

    
3201
class LUQueryInstances(NoHooksLU):
3202
  """Logical unit for querying instances.
3203

3204
  """
3205
  _OP_REQP = ["output_fields", "names", "use_locking"]
3206
  REQ_BGL = False
3207
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3208
                                    "admin_state",
3209
                                    "disk_template", "ip", "mac", "bridge",
3210
                                    "sda_size", "sdb_size", "vcpus", "tags",
3211
                                    "network_port", "beparams",
3212
                                    r"(disk)\.(size)/([0-9]+)",
3213
                                    r"(disk)\.(sizes)", "disk_usage",
3214
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3215
                                    r"(nic)\.(macs|ips|bridges)",
3216
                                    r"(disk|nic)\.(count)",
3217
                                    "serial_no", "hypervisor", "hvparams",] +
3218
                                  ["hv/%s" % name
3219
                                   for name in constants.HVS_PARAMETERS] +
3220
                                  ["be/%s" % name
3221
                                   for name in constants.BES_PARAMETERS])
3222
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3223

    
3224

    
3225
  def ExpandNames(self):
3226
    _CheckOutputFields(static=self._FIELDS_STATIC,
3227
                       dynamic=self._FIELDS_DYNAMIC,
3228
                       selected=self.op.output_fields)
3229

    
3230
    self.needed_locks = {}
3231
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3232
    self.share_locks[locking.LEVEL_NODE] = 1
3233

    
3234
    if self.op.names:
3235
      self.wanted = _GetWantedInstances(self, self.op.names)
3236
    else:
3237
      self.wanted = locking.ALL_SET
3238

    
3239
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3240
    self.do_locking = self.do_node_query and self.op.use_locking
3241
    if self.do_locking:
3242
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3243
      self.needed_locks[locking.LEVEL_NODE] = []
3244
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3245

    
3246
  def DeclareLocks(self, level):
3247
    if level == locking.LEVEL_NODE and self.do_locking:
3248
      self._LockInstancesNodes()
3249

    
3250
  def CheckPrereq(self):
3251
    """Check prerequisites.
3252

3253
    """
3254
    pass
3255

    
3256
  def Exec(self, feedback_fn):
3257
    """Computes the list of nodes and their attributes.
3258

3259
    """
3260
    all_info = self.cfg.GetAllInstancesInfo()
3261
    if self.wanted == locking.ALL_SET:
3262
      # caller didn't specify instance names, so ordering is not important
3263
      if self.do_locking:
3264
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3265
      else:
3266
        instance_names = all_info.keys()
3267
      instance_names = utils.NiceSort(instance_names)
3268
    else:
3269
      # caller did specify names, so we must keep the ordering
3270
      if self.do_locking:
3271
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3272
      else:
3273
        tgt_set = all_info.keys()
3274
      missing = set(self.wanted).difference(tgt_set)
3275
      if missing:
3276
        raise errors.OpExecError("Some instances were removed before"
3277
                                 " retrieving their data: %s" % missing)
3278
      instance_names = self.wanted
3279

    
3280
    instance_list = [all_info[iname] for iname in instance_names]
3281

    
3282
    # begin data gathering
3283

    
3284
    nodes = frozenset([inst.primary_node for inst in instance_list])
3285
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3286

    
3287
    bad_nodes = []
3288
    off_nodes = []
3289
    if self.do_node_query:
3290
      live_data = {}
3291
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3292
      for name in nodes:
3293
        result = node_data[name]
3294
        if result.offline:
3295
          # offline nodes will be in both lists
3296
          off_nodes.append(name)
3297
        if result.failed:
3298
          bad_nodes.append(name)
3299
        else:
3300
          if result.data:
3301
            live_data.update(result.data)
3302
            # else no instance is alive
3303
    else:
3304
      live_data = dict([(name, {}) for name in instance_names])
3305

    
3306
    # end data gathering
3307

    
3308
    HVPREFIX = "hv/"
3309
    BEPREFIX = "be/"
3310
    output = []
3311
    for instance in instance_list:
3312
      iout = []
3313
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3314
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3315
      for field in self.op.output_fields:
3316
        st_match = self._FIELDS_STATIC.Matches(field)
3317
        if field == "name":
3318
          val = instance.name
3319
        elif field == "os":
3320
          val = instance.os
3321
        elif field == "pnode":
3322
          val = instance.primary_node
3323
        elif field == "snodes":
3324
          val = list(instance.secondary_nodes)
3325
        elif field == "admin_state":
3326
          val = instance.admin_up
3327
        elif field == "oper_state":
3328
          if instance.primary_node in bad_nodes:
3329
            val = None
3330
          else:
3331
            val = bool(live_data.get(instance.name))
3332
        elif field == "status":
3333
          if instance.primary_node in off_nodes:
3334
            val = "ERROR_nodeoffline"
3335
          elif instance.primary_node in bad_nodes:
3336
            val = "ERROR_nodedown"
3337
          else:
3338
            running = bool(live_data.get(instance.name))
3339
            if running:
3340
              if instance.admin_up:
3341
                val = "running"
3342
              else:
3343
                val = "ERROR_up"
3344
            else:
3345
              if instance.admin_up:
3346
                val = "ERROR_down"
3347
              else:
3348
                val = "ADMIN_down"
3349
        elif field == "oper_ram":
3350
          if instance.primary_node in bad_nodes:
3351
            val = None
3352
          elif instance.name in live_data:
3353
            val = live_data[instance.name].get("memory", "?")
3354
          else:
3355
            val = "-"
3356
        elif field == "disk_template":
3357
          val = instance.disk_template
3358
        elif field == "ip":
3359
          val = instance.nics[0].ip
3360
        elif field == "bridge":
3361
          val = instance.nics[0].bridge
3362
        elif field == "mac":
3363
          val = instance.nics[0].mac
3364
        elif field == "sda_size" or field == "sdb_size":
3365
          idx = ord(field[2]) - ord('a')
3366
          try:
3367
            val = instance.FindDisk(idx).size
3368
          except errors.OpPrereqError:
3369
            val = None
3370
        elif field == "disk_usage": # total disk usage per node
3371
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3372
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3373
        elif field == "tags":
3374
          val = list(instance.GetTags())
3375
        elif field == "serial_no":
3376
          val = instance.serial_no
3377
        elif field == "network_port":
3378
          val = instance.network_port
3379
        elif field == "hypervisor":
3380
          val = instance.hypervisor
3381
        elif field == "hvparams":
3382
          val = i_hv
3383
        elif (field.startswith(HVPREFIX) and
3384
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3385
          val = i_hv.get(field[len(HVPREFIX):], None)
3386
        elif field == "beparams":
3387
          val = i_be
3388
        elif (field.startswith(BEPREFIX) and
3389
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3390
          val = i_be.get(field[len(BEPREFIX):], None)
3391
        elif st_match and st_match.groups():
3392
          # matches a variable list
3393
          st_groups = st_match.groups()
3394
          if st_groups and st_groups[0] == "disk":
3395
            if st_groups[1] == "count":
3396
              val = len(instance.disks)
3397
            elif st_groups[1] == "sizes":
3398
              val = [disk.size for disk in instance.disks]
3399
            elif st_groups[1] == "size":
3400
              try:
3401
                val = instance.FindDisk(st_groups[2]).size
3402
              except errors.OpPrereqError:
3403
                val = None
3404
            else:
3405
              assert False, "Unhandled disk parameter"
3406
          elif st_groups[0] == "nic":
3407
            if st_groups[1] == "count":
3408
              val = len(instance.nics)
3409
            elif st_groups[1] == "macs":
3410
              val = [nic.mac for nic in instance.nics]
3411
            elif st_groups[1] == "ips":
3412
              val = [nic.ip for nic in instance.nics]
3413
            elif st_groups[1] == "bridges":
3414
              val = [nic.bridge for nic in instance.nics]
3415
            else:
3416
              # index-based item
3417
              nic_idx = int(st_groups[2])
3418
              if nic_idx >= len(instance.nics):
3419
                val = None
3420
              else:
3421
                if st_groups[1] == "mac":
3422
                  val = instance.nics[nic_idx].mac
3423
                elif st_groups[1] == "ip":
3424
                  val = instance.nics[nic_idx].ip
3425
                elif st_groups[1] == "bridge":
3426
                  val = instance.nics[nic_idx].bridge
3427
                else:
3428
                  assert False, "Unhandled NIC parameter"
3429
          else:
3430
            assert False, "Unhandled variable parameter"
3431
        else:
3432
          raise errors.ParameterError(field)
3433
        iout.append(val)
3434
      output.append(iout)
3435

    
3436
    return output
3437

    
3438

    
3439
class LUFailoverInstance(LogicalUnit):
3440
  """Failover an instance.
3441

3442
  """
3443
  HPATH = "instance-failover"
3444
  HTYPE = constants.HTYPE_INSTANCE
3445
  _OP_REQP = ["instance_name", "ignore_consistency"]
3446
  REQ_BGL = False
3447

    
3448
  def ExpandNames(self):
3449
    self._ExpandAndLockInstance()
3450
    self.needed_locks[locking.LEVEL_NODE] = []
3451
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3452

    
3453
  def DeclareLocks(self, level):
3454
    if level == locking.LEVEL_NODE:
3455
      self._LockInstancesNodes()
3456

    
3457
  def BuildHooksEnv(self):
3458
    """Build hooks env.
3459

3460
    This runs on master, primary and secondary nodes of the instance.
3461

3462
    """
3463
    env = {
3464
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3465
      }
3466
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3467
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3468
    return env, nl, nl
3469

    
3470
  def CheckPrereq(self):
3471
    """Check prerequisites.
3472

3473
    This checks that the instance is in the cluster.
3474

3475
    """
3476
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3477
    assert self.instance is not None, \
3478
      "Cannot retrieve locked instance %s" % self.op.instance_name
3479

    
3480
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3481
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3482
      raise errors.OpPrereqError("Instance's disk layout is not"
3483
                                 " network mirrored, cannot failover.")
3484

    
3485
    secondary_nodes = instance.secondary_nodes
3486
    if not secondary_nodes:
3487
      raise errors.ProgrammerError("no secondary node but using "
3488
                                   "a mirrored disk template")
3489

    
3490
    target_node = secondary_nodes[0]
3491
    _CheckNodeOnline(self, target_node)
3492
    _CheckNodeNotDrained(self, target_node)
3493
    # check memory requirements on the secondary node
3494
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3495
                         instance.name, bep[constants.BE_MEMORY],
3496
                         instance.hypervisor)
3497

    
3498
    # check bridge existance
3499
    brlist = [nic.bridge for nic in instance.nics]
3500
    result = self.rpc.call_bridges_exist(target_node, brlist)
3501
    result.Raise()
3502
    if not result.data:
3503
      raise errors.OpPrereqError("One or more target bridges %s does not"
3504
                                 " exist on destination node '%s'" %
3505
                                 (brlist, target_node))
3506

    
3507
  def Exec(self, feedback_fn):
3508
    """Failover an instance.
3509

3510
    The failover is done by shutting it down on its present node and
3511
    starting it on the secondary.
3512

3513
    """
3514
    instance = self.instance
3515

    
3516
    source_node = instance.primary_node
3517
    target_node = instance.secondary_nodes[0]
3518

    
3519
    feedback_fn("* checking disk consistency between source and target")
3520
    for dev in instance.disks:
3521
      # for drbd, these are drbd over lvm
3522
      if not _CheckDiskConsistency(self, dev, target_node, False):
3523
        if instance.admin_up and not self.op.ignore_consistency:
3524
          raise errors.OpExecError("Disk %s is degraded on target node,"
3525
                                   " aborting failover." % dev.iv_name)
3526

    
3527
    feedback_fn("* shutting down instance on source node")
3528
    logging.info("Shutting down instance %s on node %s",
3529
                 instance.name, source_node)
3530

    
3531
    result = self.rpc.call_instance_shutdown(source_node, instance)
3532
    msg = result.RemoteFailMsg()
3533
    if msg:
3534
      if self.op.ignore_consistency:
3535
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3536
                             " Proceeding anyway. Please make sure node"
3537
                             " %s is down. Error details: %s",
3538
                             instance.name, source_node, source_node, msg)
3539
      else:
3540
        raise errors.OpExecError("Could not shutdown instance %s on"
3541
                                 " node %s: %s" %
3542
                                 (instance.name, source_node, msg))
3543

    
3544
    feedback_fn("* deactivating the instance's disks on source node")
3545
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3546
      raise errors.OpExecError("Can't shut down the instance's disks.")
3547

    
3548
    instance.primary_node = target_node
3549
    # distribute new instance config to the other nodes
3550
    self.cfg.Update(instance)
3551

    
3552
    # Only start the instance if it's marked as up
3553
    if instance.admin_up:
3554
      feedback_fn("* activating the instance's disks on target node")
3555
      logging.info("Starting instance %s on node %s",
3556
                   instance.name, target_node)
3557

    
3558
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3559
                                               ignore_secondaries=True)
3560
      if not disks_ok:
3561
        _ShutdownInstanceDisks(self, instance)
3562
        raise errors.OpExecError("Can't activate the instance's disks")
3563

    
3564
      feedback_fn("* starting the instance on the target node")
3565
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3566
      msg = result.RemoteFailMsg()
3567
      if msg:
3568
        _ShutdownInstanceDisks(self, instance)
3569
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3570
                                 (instance.name, target_node, msg))
3571

    
3572

    
3573
class LUMigrateInstance(LogicalUnit):
3574
  """Migrate an instance.
3575

3576
  This is migration without shutting down, compared to the failover,
3577
  which is done with shutdown.
3578

3579
  """
3580
  HPATH = "instance-migrate"
3581
  HTYPE = constants.HTYPE_INSTANCE
3582
  _OP_REQP = ["instance_name", "live", "cleanup"]
3583

    
3584
  REQ_BGL = False
3585

    
3586
  def ExpandNames(self):
3587
    self._ExpandAndLockInstance()
3588
    self.needed_locks[locking.LEVEL_NODE] = []
3589
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3590

    
3591
  def DeclareLocks(self, level):
3592
    if level == locking.LEVEL_NODE:
3593
      self._LockInstancesNodes()
3594

    
3595
  def BuildHooksEnv(self):
3596
    """Build hooks env.
3597

3598
    This runs on master, primary and secondary nodes of the instance.
3599

3600
    """
3601
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3602
    env["MIGRATE_LIVE"] = self.op.live
3603
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3604
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3605
    return env, nl, nl
3606

    
3607
  def CheckPrereq(self):
3608
    """Check prerequisites.
3609

3610
    This checks that the instance is in the cluster.
3611

3612
    """
3613
    instance = self.cfg.GetInstanceInfo(
3614
      self.cfg.ExpandInstanceName(self.op.instance_name))
3615
    if instance is None:
3616
      raise errors.OpPrereqError("Instance '%s' not known" %
3617
                                 self.op.instance_name)
3618

    
3619
    if instance.disk_template != constants.DT_DRBD8:
3620
      raise errors.OpPrereqError("Instance's disk layout is not"
3621
                                 " drbd8, cannot migrate.")
3622

    
3623
    secondary_nodes = instance.secondary_nodes
3624
    if not secondary_nodes:
3625
      raise errors.ConfigurationError("No secondary node but using"
3626
                                      " drbd8 disk template")
3627

    
3628
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3629

    
3630
    target_node = secondary_nodes[0]
3631
    # check memory requirements on the secondary node
3632
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3633
                         instance.name, i_be[constants.BE_MEMORY],
3634
                         instance.hypervisor)
3635

    
3636
    # check bridge existance
3637
    brlist = [nic.bridge for nic in instance.nics]
3638
    result = self.rpc.call_bridges_exist(target_node, brlist)
3639
    if result.failed or not result.data:
3640
      raise errors.OpPrereqError("One or more target bridges %s does not"
3641
                                 " exist on destination node '%s'" %
3642
                                 (brlist, target_node))
3643

    
3644
    if not self.op.cleanup:
3645
      _CheckNodeNotDrained(self, target_node)
3646
      result = self.rpc.call_instance_migratable(instance.primary_node,
3647
                                                 instance)
3648
      msg = result.RemoteFailMsg()
3649
      if msg:
3650
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3651
                                   msg)
3652

    
3653
    self.instance = instance
3654

    
3655
  def _WaitUntilSync(self):
3656
    """Poll with custom rpc for disk sync.
3657

3658
    This uses our own step-based rpc call.
3659

3660
    """
3661
    self.feedback_fn("* wait until resync is done")
3662
    all_done = False
3663
    while not all_done:
3664
      all_done = True
3665
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3666
                                            self.nodes_ip,
3667
                                            self.instance.disks)
3668
      min_percent = 100
3669
      for node, nres in result.items():
3670
        msg = nres.RemoteFailMsg()
3671
        if msg:
3672
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3673
                                   (node, msg))
3674
        node_done, node_percent = nres.payload
3675
        all_done = all_done and node_done
3676
        if node_percent is not None:
3677
          min_percent = min(min_percent, node_percent)
3678
      if not all_done:
3679
        if min_percent < 100:
3680
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3681
        time.sleep(2)
3682

    
3683
  def _EnsureSecondary(self, node):
3684
    """Demote a node to secondary.
3685

3686
    """
3687
    self.feedback_fn("* switching node %s to secondary mode" % node)
3688

    
3689
    for dev in self.instance.disks:
3690
      self.cfg.SetDiskID(dev, node)
3691

    
3692
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3693
                                          self.instance.disks)
3694
    msg = result.RemoteFailMsg()
3695
    if msg:
3696
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3697
                               " error %s" % (node, msg))
3698

    
3699
  def _GoStandalone(self):
3700
    """Disconnect from the network.
3701

3702
    """
3703
    self.feedback_fn("* changing into standalone mode")
3704
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3705
                                               self.instance.disks)
3706
    for node, nres in result.items():
3707
      msg = nres.RemoteFailMsg()
3708
      if msg:
3709
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3710
                                 " error %s" % (node, msg))
3711

    
3712
  def _GoReconnect(self, multimaster):
3713
    """Reconnect to the network.
3714

3715
    """
3716
    if multimaster:
3717
      msg = "dual-master"
3718
    else:
3719
      msg = "single-master"
3720
    self.feedback_fn("* changing disks into %s mode" % msg)
3721
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3722
                                           self.instance.disks,
3723
                                           self.instance.name, multimaster)
3724
    for node, nres in result.items():
3725
      msg = nres.RemoteFailMsg()
3726
      if msg:
3727
        raise errors.OpExecError("Cannot change disks config on node %s,"
3728
                                 " error: %s" % (node, msg))
3729

    
3730
  def _ExecCleanup(self):
3731
    """Try to cleanup after a failed migration.
3732

3733
    The cleanup is done by:
3734
      - check that the instance is running only on one node
3735
        (and update the config if needed)
3736
      - change disks on its secondary node to secondary
3737
      - wait until disks are fully synchronized
3738
      - disconnect from the network
3739
      - change disks into single-master mode
3740
      - wait again until disks are fully synchronized
3741

3742
    """
3743
    instance = self.instance
3744
    target_node = self.target_node
3745
    source_node = self.source_node
3746

    
3747
    # check running on only one node
3748
    self.feedback_fn("* checking where the instance actually runs"
3749
                     " (if this hangs, the hypervisor might be in"
3750
                     " a bad state)")
3751
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3752
    for node, result in ins_l.items():
3753
      result.Raise()
3754
      if not isinstance(result.data, list):
3755
        raise errors.OpExecError("Can't contact node '%s'" % node)
3756

    
3757
    runningon_source = instance.name in ins_l[source_node].data
3758
    runningon_target = instance.name in ins_l[target_node].data
3759

    
3760
    if runningon_source and runningon_target:
3761
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3762
                               " or the hypervisor is confused. You will have"
3763
                               " to ensure manually that it runs only on one"
3764
                               " and restart this operation.")
3765

    
3766
    if not (runningon_source or runningon_target):
3767
      raise errors.OpExecError("Instance does not seem to be running at all."
3768
                               " In this case, it's safer to repair by"
3769
                               " running 'gnt-instance stop' to ensure disk"
3770
                               " shutdown, and then restarting it.")
3771

    
3772
    if runningon_target:
3773
      # the migration has actually succeeded, we need to update the config
3774
      self.feedback_fn("* instance running on secondary node (%s),"
3775
                       " updating config" % target_node)
3776
      instance.primary_node = target_node
3777
      self.cfg.Update(instance)
3778
      demoted_node = source_node
3779
    else:
3780
      self.feedback_fn("* instance confirmed to be running on its"
3781
                       " primary node (%s)" % source_node)
3782
      demoted_node = target_node
3783

    
3784
    self._EnsureSecondary(demoted_node)
3785
    try:
3786
      self._WaitUntilSync()
3787
    except errors.OpExecError:
3788
      # we ignore here errors, since if the device is standalone, it
3789
      # won't be able to sync
3790
      pass
3791
    self._GoStandalone()
3792
    self._GoReconnect(False)
3793
    self._WaitUntilSync()
3794

    
3795
    self.feedback_fn("* done")
3796

    
3797
  def _RevertDiskStatus(self):
3798
    """Try to revert the disk status after a failed migration.
3799

3800
    """
3801
    target_node = self.target_node
3802
    try:
3803
      self._EnsureSecondary(target_node)
3804
      self._GoStandalone()
3805
      self._GoReconnect(False)
3806
      self._WaitUntilSync()
3807
    except errors.OpExecError, err:
3808
      self.LogWarning("Migration failed and I can't reconnect the"
3809
                      " drives: error '%s'\n"
3810
                      "Please look and recover the instance status" %
3811
                      str(err))
3812

    
3813
  def _AbortMigration(self):
3814
    """Call the hypervisor code to abort a started migration.
3815

3816
    """
3817
    instance = self.instance
3818
    target_node = self.target_node
3819
    migration_info = self.migration_info
3820

    
3821
    abort_result = self.rpc.call_finalize_migration(target_node,
3822
                                                    instance,
3823
                                                    migration_info,
3824
                                                    False)
3825
    abort_msg = abort_result.RemoteFailMsg()
3826
    if abort_msg:
3827
      logging.error("Aborting migration failed on target node %s: %s" %
3828
                    (target_node, abort_msg))
3829
      # Don't raise an exception here, as we stil have to try to revert the
3830
      # disk status, even if this step failed.
3831

    
3832
  def _ExecMigration(self):
3833
    """Migrate an instance.
3834

3835
    The migrate is done by:
3836
      - change the disks into dual-master mode
3837
      - wait until disks are fully synchronized again
3838
      - migrate the instance
3839
      - change disks on the new secondary node (the old primary) to secondary
3840
      - wait until disks are fully synchronized
3841
      - change disks into single-master mode
3842

3843
    """
3844
    instance = self.instance
3845
    target_node = self.target_node
3846
    source_node = self.source_node
3847

    
3848
    self.feedback_fn("* checking disk consistency between source and target")
3849
    for dev in instance.disks:
3850
      if not _CheckDiskConsistency(self, dev, target_node, False):
3851
        raise errors.OpExecError("Disk %s is degraded or not fully"
3852
                                 " synchronized on target node,"
3853
                                 " aborting migrate." % dev.iv_name)
3854

    
3855
    # First get the migration information from the remote node
3856
    result = self.rpc.call_migration_info(source_node, instance)
3857
    msg = result.RemoteFailMsg()
3858
    if msg:
3859
      log_err = ("Failed fetching source migration information from %s: %s" %
3860
                 (source_node, msg))
3861
      logging.error(log_err)
3862
      raise errors.OpExecError(log_err)
3863

    
3864
    self.migration_info = migration_info = result.payload
3865

    
3866
    # Then switch the disks to master/master mode
3867
    self._EnsureSecondary(target_node)
3868
    self._GoStandalone()
3869
    self._GoReconnect(True)
3870
    self._WaitUntilSync()
3871

    
3872
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3873
    result = self.rpc.call_accept_instance(target_node,
3874
                                           instance,
3875
                                           migration_info,
3876
                                           self.nodes_ip[target_node])
3877

    
3878
    msg = result.RemoteFailMsg()
3879
    if msg:
3880
      logging.error("Instance pre-migration failed, trying to revert"
3881
                    " disk status: %s", msg)
3882
      self._AbortMigration()
3883
      self._RevertDiskStatus()
3884
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3885
                               (instance.name, msg))
3886

    
3887
    self.feedback_fn("* migrating instance to %s" % target_node)
3888
    time.sleep(10)
3889
    result = self.rpc.call_instance_migrate(source_node, instance,
3890
                                            self.nodes_ip[target_node],
3891
                                            self.op.live)
3892
    msg = result.RemoteFailMsg()
3893
    if msg:
3894
      logging.error("Instance migration failed, trying to revert"
3895
                    " disk status: %s", msg)
3896
      self._AbortMigration()
3897
      self._RevertDiskStatus()
3898
      raise errors.OpExecError("Could not migrate instance %s: %s" %
3899
                               (instance.name, msg))
3900
    time.sleep(10)
3901

    
3902
    instance.primary_node = target_node
3903
    # distribute new instance config to the other nodes
3904
    self.cfg.Update(instance)
3905

    
3906
    result = self.rpc.call_finalize_migration(target_node,
3907
                                              instance,
3908
                                              migration_info,
3909
                                              True)
3910
    msg = result.RemoteFailMsg()
3911
    if msg:
3912
      logging.error("Instance migration succeeded, but finalization failed:"
3913
                    " %s" % msg)
3914
      raise errors.OpExecError("Could not finalize instance migration: %s" %
3915
                               msg)
3916

    
3917
    self._EnsureSecondary(source_node)
3918
    self._WaitUntilSync()
3919
    self._GoStandalone()
3920
    self._GoReconnect(False)
3921
    self._WaitUntilSync()
3922

    
3923
    self.feedback_fn("* done")
3924

    
3925
  def Exec(self, feedback_fn):
3926
    """Perform the migration.
3927

3928
    """
3929
    self.feedback_fn = feedback_fn
3930

    
3931
    self.source_node = self.instance.primary_node
3932
    self.target_node = self.instance.secondary_nodes[0]
3933
    self.all_nodes = [self.source_node, self.target_node]
3934
    self.nodes_ip = {
3935
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
3936
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
3937
      }
3938
    if self.op.cleanup:
3939
      return self._ExecCleanup()
3940
    else:
3941
      return self._ExecMigration()
3942

    
3943

    
3944
def _CreateBlockDev(lu, node, instance, device, force_create,
3945
                    info, force_open):
3946
  """Create a tree of block devices on a given node.
3947

3948
  If this device type has to be created on secondaries, create it and
3949
  all its children.
3950

3951
  If not, just recurse to children keeping the same 'force' value.
3952

3953
  @param lu: the lu on whose behalf we execute
3954
  @param node: the node on which to create the device
3955
  @type instance: L{objects.Instance}
3956
  @param instance: the instance which owns the device
3957
  @type device: L{objects.Disk}
3958
  @param device: the device to create
3959
  @type force_create: boolean
3960
  @param force_create: whether to force creation of this device; this
3961
      will be change to True whenever we find a device which has
3962
      CreateOnSecondary() attribute
3963
  @param info: the extra 'metadata' we should attach to the device
3964
      (this will be represented as a LVM tag)
3965
  @type force_open: boolean
3966
  @param force_open: this parameter will be passes to the
3967
      L{backend.BlockdevCreate} function where it specifies
3968
      whether we run on primary or not, and it affects both
3969
      the child assembly and the device own Open() execution
3970

3971
  """
3972
  if device.CreateOnSecondary():
3973
    force_create = True
3974

    
3975
  if device.children:
3976
    for child in device.children:
3977
      _CreateBlockDev(lu, node, instance, child, force_create,
3978
                      info, force_open)
3979

    
3980
  if not force_create:
3981
    return
3982

    
3983
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
3984

    
3985

    
3986
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
3987
  """Create a single block device on a given node.
3988

3989
  This will not recurse over children of the device, so they must be
3990
  created in advance.
3991

3992
  @param lu: the lu on whose behalf we execute
3993
  @param node: the node on which to create the device
3994
  @type instance: L{objects.Instance}
3995
  @param instance: the instance which owns the device
3996
  @type device: L{objects.Disk}
3997
  @param device: the device to create
3998
  @param info: the extra 'metadata' we should attach to the device
3999
      (this will be represented as a LVM tag)
4000
  @type force_open: boolean
4001
  @param force_open: this parameter will be passes to the
4002
      L{backend.BlockdevCreate} function where it specifies
4003
      whether we run on primary or not, and it affects both
4004
      the child assembly and the device own Open() execution
4005

4006
  """
4007
  lu.cfg.SetDiskID(device, node)
4008
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4009
                                       instance.name, force_open, info)
4010
  msg = result.RemoteFailMsg()
4011
  if msg:
4012
    raise errors.OpExecError("Can't create block device %s on"
4013
                             " node %s for instance %s: %s" %
4014
                             (device, node, instance.name, msg))
4015
  if device.physical_id is None:
4016
    device.physical_id = result.payload
4017

    
4018

    
4019
def _GenerateUniqueNames(lu, exts):
4020
  """Generate a suitable LV name.
4021

4022
  This will generate a logical volume name for the given instance.
4023

4024
  """
4025
  results = []
4026
  for val in exts:
4027
    new_id = lu.cfg.GenerateUniqueID()
4028
    results.append("%s%s" % (new_id, val))
4029
  return results
4030

    
4031

    
4032
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4033
                         p_minor, s_minor):
4034
  """Generate a drbd8 device complete with its children.
4035

4036
  """
4037
  port = lu.cfg.AllocatePort()
4038
  vgname = lu.cfg.GetVGName()
4039
  shared_secret = lu.cfg.GenerateDRBDSecret()
4040
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4041
                          logical_id=(vgname, names[0]))
4042
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4043
                          logical_id=(vgname, names[1]))
4044
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4045
                          logical_id=(primary, secondary, port,
4046
                                      p_minor, s_minor,
4047
                                      shared_secret),
4048
                          children=[dev_data, dev_meta],
4049
                          iv_name=iv_name)
4050
  return drbd_dev
4051

    
4052

    
4053
def _GenerateDiskTemplate(lu, template_name,
4054
                          instance_name, primary_node,
4055
                          secondary_nodes, disk_info,
4056
                          file_storage_dir, file_driver,
4057
                          base_index):
4058
  """Generate the entire disk layout for a given template type.
4059

4060
  """
4061
  #TODO: compute space requirements
4062

    
4063
  vgname = lu.cfg.GetVGName()
4064
  disk_count = len(disk_info)
4065
  disks = []
4066
  if template_name == constants.DT_DISKLESS:
4067
    pass
4068
  elif template_name == constants.DT_PLAIN:
4069
    if len(secondary_nodes) != 0:
4070
      raise errors.ProgrammerError("Wrong template configuration")
4071

    
4072
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4073
                                      for i in range(disk_count)])
4074
    for idx, disk in enumerate(disk_info):
4075
      disk_index = idx + base_index
4076
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4077
                              logical_id=(vgname, names[idx]),
4078
                              iv_name="disk/%d" % disk_index,
4079
                              mode=disk["mode"])
4080
      disks.append(disk_dev)
4081
  elif template_name == constants.DT_DRBD8:
4082
    if len(secondary_nodes) != 1:
4083
      raise errors.ProgrammerError("Wrong template configuration")
4084
    remote_node = secondary_nodes[0]
4085
    minors = lu.cfg.AllocateDRBDMinor(
4086
      [primary_node, remote_node] * len(disk_info), instance_name)
4087

    
4088
    names = []
4089
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4090
                                               for i in range(disk_count)]):
4091
      names.append(lv_prefix + "_data")
4092
      names.append(lv_prefix + "_meta")
4093
    for idx, disk in enumerate(disk_info):
4094
      disk_index = idx + base_index
4095
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4096
                                      disk["size"], names[idx*2:idx*2+2],
4097
                                      "disk/%d" % disk_index,
4098
                                      minors[idx*2], minors[idx*2+1])
4099
      disk_dev.mode = disk["mode"]
4100
      disks.append(disk_dev)
4101
  elif template_name == constants.DT_FILE:
4102
    if len(secondary_nodes) != 0:
4103
      raise errors.ProgrammerError("Wrong template configuration")
4104

    
4105
    for idx, disk in enumerate(disk_info):
4106
      disk_index = idx + base_index
4107
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4108
                              iv_name="disk/%d" % disk_index,
4109
                              logical_id=(file_driver,
4110
                                          "%s/disk%d" % (file_storage_dir,
4111
                                                         disk_index)),
4112
                              mode=disk["mode"])
4113
      disks.append(disk_dev)
4114
  else:
4115
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4116
  return disks
4117

    
4118

    
4119
def _GetInstanceInfoText(instance):
4120
  """Compute that text that should be added to the disk's metadata.
4121

4122
  """
4123
  return "originstname+%s" % instance.name
4124

    
4125

    
4126
def _CreateDisks(lu, instance):
4127
  """Create all disks for an instance.
4128

4129
  This abstracts away some work from AddInstance.
4130

4131
  @type lu: L{LogicalUnit}
4132
  @param lu: the logical unit on whose behalf we execute
4133
  @type instance: L{objects.Instance}
4134
  @param instance: the instance whose disks we should create
4135
  @rtype: boolean
4136
  @return: the success of the creation
4137

4138
  """
4139
  info = _GetInstanceInfoText(instance)
4140
  pnode = instance.primary_node
4141

    
4142
  if instance.disk_template == constants.DT_FILE:
4143
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4144
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4145

    
4146
    if result.failed or not result.data:
4147
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4148

    
4149
    if not result.data[0]:
4150
      raise errors.OpExecError("Failed to create directory '%s'" %
4151
                               file_storage_dir)
4152

    
4153
  # Note: this needs to be kept in sync with adding of disks in
4154
  # LUSetInstanceParams
4155
  for device in instance.disks:
4156
    logging.info("Creating volume %s for instance %s",
4157
                 device.iv_name, instance.name)
4158
    #HARDCODE
4159
    for node in instance.all_nodes:
4160
      f_create = node == pnode
4161
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4162

    
4163

    
4164
def _RemoveDisks(lu, instance):
4165
  """Remove all disks for an instance.
4166

4167
  This abstracts away some work from `AddInstance()` and
4168
  `RemoveInstance()`. Note that in case some of the devices couldn't
4169
  be removed, the removal will continue with the other ones (compare
4170
  with `_CreateDisks()`).
4171

4172
  @type lu: L{LogicalUnit}
4173
  @param lu: the logical unit on whose behalf we execute
4174
  @type instance: L{objects.Instance}
4175
  @param instance: the instance whose disks we should remove
4176
  @rtype: boolean
4177
  @return: the success of the removal
4178

4179
  """
4180
  logging.info("Removing block devices for instance %s", instance.name)
4181

    
4182
  all_result = True
4183
  for device in instance.disks:
4184
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4185
      lu.cfg.SetDiskID(disk, node)
4186
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4187
      if msg:
4188
        lu.LogWarning("Could not remove block device %s on node %s,"
4189
                      " continuing anyway: %s", device.iv_name, node, msg)
4190
        all_result = False
4191

    
4192
  if instance.disk_template == constants.DT_FILE:
4193
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4194
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4195
                                                 file_storage_dir)
4196
    if result.failed or not result.data:
4197
      logging.error("Could not remove directory '%s'", file_storage_dir)
4198
      all_result = False
4199

    
4200
  return all_result
4201

    
4202

    
4203
def _ComputeDiskSize(disk_template, disks):
4204
  """Compute disk size requirements in the volume group
4205

4206
  """
4207
  # Required free disk space as a function of disk and swap space
4208
  req_size_dict = {
4209
    constants.DT_DISKLESS: None,
4210
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4211
    # 128 MB are added for drbd metadata for each disk
4212
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4213
    constants.DT_FILE: None,
4214
  }
4215

    
4216
  if disk_template not in req_size_dict:
4217
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4218
                                 " is unknown" %  disk_template)
4219

    
4220
  return req_size_dict[disk_template]
4221

    
4222

    
4223
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4224
  """Hypervisor parameter validation.
4225

4226
  This function abstract the hypervisor parameter validation to be
4227
  used in both instance create and instance modify.
4228

4229
  @type lu: L{LogicalUnit}
4230
  @param lu: the logical unit for which we check
4231
  @type nodenames: list
4232
  @param nodenames: the list of nodes on which we should check
4233
  @type hvname: string
4234
  @param hvname: the name of the hypervisor we should use
4235
  @type hvparams: dict
4236
  @param hvparams: the parameters which we need to check
4237
  @raise errors.OpPrereqError: if the parameters are not valid
4238

4239
  """
4240
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4241
                                                  hvname,
4242
                                                  hvparams)
4243
  for node in nodenames:
4244
    info = hvinfo[node]
4245
    if info.offline:
4246
      continue
4247
    msg = info.RemoteFailMsg()
4248
    if msg:
4249
      raise errors.OpPrereqError("Hypervisor parameter validation"
4250
                                 " failed on node %s: %s" % (node, msg))
4251

    
4252

    
4253
class LUCreateInstance(LogicalUnit):
4254
  """Create an instance.
4255

4256
  """
4257
  HPATH = "instance-add"
4258
  HTYPE = constants.HTYPE_INSTANCE
4259
  _OP_REQP = ["instance_name", "disks", "disk_template",
4260
              "mode", "start",
4261
              "wait_for_sync", "ip_check", "nics",
4262
              "hvparams", "beparams"]
4263
  REQ_BGL = False
4264

    
4265
  def _ExpandNode(self, node):
4266
    """Expands and checks one node name.
4267

4268
    """
4269
    node_full = self.cfg.ExpandNodeName(node)
4270
    if node_full is None:
4271
      raise errors.OpPrereqError("Unknown node %s" % node)
4272
    return node_full
4273

    
4274
  def ExpandNames(self):
4275
    """ExpandNames for CreateInstance.
4276

4277
    Figure out the right locks for instance creation.
4278

4279
    """
4280
    self.needed_locks = {}
4281

    
4282
    # set optional parameters to none if they don't exist
4283
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4284
      if not hasattr(self.op, attr):
4285
        setattr(self.op, attr, None)
4286

    
4287
    # cheap checks, mostly valid constants given
4288

    
4289
    # verify creation mode
4290
    if self.op.mode not in (constants.INSTANCE_CREATE,
4291
                            constants.INSTANCE_IMPORT):
4292
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4293
                                 self.op.mode)
4294

    
4295
    # disk template and mirror node verification
4296
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4297
      raise errors.OpPrereqError("Invalid disk template name")
4298

    
4299
    if self.op.hypervisor is None:
4300
      self.op.hypervisor = self.cfg.GetHypervisorType()
4301

    
4302
    cluster = self.cfg.GetClusterInfo()
4303
    enabled_hvs = cluster.enabled_hypervisors
4304
    if self.op.hypervisor not in enabled_hvs:
4305
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4306
                                 " cluster (%s)" % (self.op.hypervisor,
4307
                                  ",".join(enabled_hvs)))
4308

    
4309
    # check hypervisor parameter syntax (locally)
4310
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4311
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4312
                                  self.op.hvparams)
4313
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4314
    hv_type.CheckParameterSyntax(filled_hvp)
4315

    
4316
    # fill and remember the beparams dict
4317
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4318
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4319
                                    self.op.beparams)
4320

    
4321
    #### instance parameters check
4322

    
4323
    # instance name verification
4324
    hostname1 = utils.HostInfo(self.op.instance_name)
4325
    self.op.instance_name = instance_name = hostname1.name
4326

    
4327
    # this is just a preventive check, but someone might still add this
4328
    # instance in the meantime, and creation will fail at lock-add time
4329
    if instance_name in self.cfg.GetInstanceList():
4330
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4331
                                 instance_name)
4332

    
4333
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4334

    
4335
    # NIC buildup
4336
    self.nics = []
4337
    for nic in self.op.nics:
4338
      # ip validity checks
4339
      ip = nic.get("ip", None)
4340
      if ip is None or ip.lower() == "none":
4341
        nic_ip = None
4342
      elif ip.lower() == constants.VALUE_AUTO:
4343
        nic_ip = hostname1.ip
4344
      else:
4345
        if not utils.IsValidIP(ip):
4346
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4347
                                     " like a valid IP" % ip)
4348
        nic_ip = ip
4349

    
4350
      # MAC address verification
4351
      mac = nic.get("mac", constants.VALUE_AUTO)
4352
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4353
        if not utils.IsValidMac(mac.lower()):
4354
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4355
                                     mac)
4356
      # bridge verification
4357
      bridge = nic.get("bridge", None)
4358
      if bridge is None:
4359
        bridge = self.cfg.GetDefBridge()
4360
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4361

    
4362
    # disk checks/pre-build
4363
    self.disks = []
4364
    for disk in self.op.disks:
4365
      mode = disk.get("mode", constants.DISK_RDWR)
4366
      if mode not in constants.DISK_ACCESS_SET:
4367
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4368
                                   mode)
4369
      size = disk.get("size", None)
4370
      if size is None:
4371
        raise errors.OpPrereqError("Missing disk size")
4372
      try:
4373
        size = int(size)
4374
      except ValueError:
4375
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4376
      self.disks.append({"size": size, "mode": mode})
4377

    
4378
    # used in CheckPrereq for ip ping check
4379
    self.check_ip = hostname1.ip
4380

    
4381
    # file storage checks
4382
    if (self.op.file_driver and
4383
        not self.op.file_driver in constants.FILE_DRIVER):
4384
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4385
                                 self.op.file_driver)
4386

    
4387
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4388
      raise errors.OpPrereqError("File storage directory path not absolute")
4389

    
4390
    ### Node/iallocator related checks
4391
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4392
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4393
                                 " node must be given")
4394

    
4395
    if self.op.iallocator:
4396
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4397
    else:
4398
      self.op.pnode = self._ExpandNode(self.op.pnode)
4399
      nodelist = [self.op.pnode]
4400
      if self.op.snode is not None:
4401
        self.op.snode = self._ExpandNode(self.op.snode)
4402
        nodelist.append(self.op.snode)
4403
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4404

    
4405
    # in case of import lock the source node too
4406
    if self.op.mode == constants.INSTANCE_IMPORT:
4407
      src_node = getattr(self.op, "src_node", None)
4408
      src_path = getattr(self.op, "src_path", None)
4409

    
4410
      if src_path is None:
4411
        self.op.src_path = src_path = self.op.instance_name
4412

    
4413
      if src_node is None:
4414
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4415
        self.op.src_node = None
4416
        if os.path.isabs(src_path):
4417
          raise errors.OpPrereqError("Importing an instance from an absolute"
4418
                                     " path requires a source node option.")
4419
      else:
4420
        self.op.src_node = src_node = self._ExpandNode(src_node)
4421
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4422
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4423
        if not os.path.isabs(src_path):
4424
          self.op.src_path = src_path = \
4425
            os.path.join(constants.EXPORT_DIR, src_path)
4426

    
4427
    else: # INSTANCE_CREATE
4428
      if getattr(self.op, "os_type", None) is None:
4429
        raise errors.OpPrereqError("No guest OS specified")
4430

    
4431
  def _RunAllocator(self):
4432
    """Run the allocator based on input opcode.
4433

4434
    """
4435
    nics = [n.ToDict() for n in self.nics]
4436
    ial = IAllocator(self,
4437
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4438
                     name=self.op.instance_name,
4439
                     disk_template=self.op.disk_template,
4440
                     tags=[],
4441
                     os=self.op.os_type,
4442
                     vcpus=self.be_full[constants.BE_VCPUS],
4443
                     mem_size=self.be_full[constants.BE_MEMORY],
4444
                     disks=self.disks,
4445
                     nics=nics,
4446
                     hypervisor=self.op.hypervisor,
4447
                     )
4448

    
4449
    ial.Run(self.op.iallocator)
4450

    
4451
    if not ial.success:
4452
      raise errors.OpPrereqError("Can't compute nodes using"
4453
                                 " iallocator '%s': %s" % (self.op.iallocator,
4454
                                                           ial.info))
4455
    if len(ial.nodes) != ial.required_nodes:
4456
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4457
                                 " of nodes (%s), required %s" %
4458
                                 (self.op.iallocator, len(ial.nodes),
4459
                                  ial.required_nodes))
4460
    self.op.pnode = ial.nodes[0]
4461
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4462
                 self.op.instance_name, self.op.iallocator,
4463
                 ", ".join(ial.nodes))
4464
    if ial.required_nodes == 2:
4465
      self.op.snode = ial.nodes[1]
4466

    
4467
  def BuildHooksEnv(self):
4468
    """Build hooks env.
4469

4470
    This runs on master, primary and secondary nodes of the instance.
4471

4472
    """
4473
    env = {
4474
      "ADD_MODE": self.op.mode,
4475
      }
4476
    if self.op.mode == constants.INSTANCE_IMPORT:
4477
      env["SRC_NODE"] = self.op.src_node
4478
      env["SRC_PATH"] = self.op.src_path
4479
      env["SRC_IMAGES"] = self.src_images
4480

    
4481
    env.update(_BuildInstanceHookEnv(
4482
      name=self.op.instance_name,
4483
      primary_node=self.op.pnode,
4484
      secondary_nodes=self.secondaries,
4485
      status=self.op.start,
4486
      os_type=self.op.os_type,
4487
      memory=self.be_full[constants.BE_MEMORY],
4488
      vcpus=self.be_full[constants.BE_VCPUS],
4489
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4490
      disk_template=self.op.disk_template,
4491
      disks=[(d["size"], d["mode"]) for d in self.disks],
4492
    ))
4493

    
4494
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4495
          self.secondaries)
4496
    return env, nl, nl
4497

    
4498

    
4499
  def CheckPrereq(self):
4500
    """Check prerequisites.
4501

4502
    """
4503
    if (not self.cfg.GetVGName() and
4504
        self.op.disk_template not in constants.DTS_NOT_LVM):
4505
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4506
                                 " instances")
4507

    
4508
    if self.op.mode == constants.INSTANCE_IMPORT:
4509
      src_node = self.op.src_node
4510
      src_path = self.op.src_path
4511

    
4512
      if src_node is None:
4513
        exp_list = self.rpc.call_export_list(
4514
          self.acquired_locks[locking.LEVEL_NODE])
4515
        found = False
4516
        for node in exp_list:
4517
          if not exp_list[node].failed and src_path in exp_list[node].data:
4518
            found = True
4519
            self.op.src_node = src_node = node
4520
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4521
                                                       src_path)
4522
            break
4523
        if not found:
4524
          raise errors.OpPrereqError("No export found for relative path %s" %
4525
                                      src_path)
4526

    
4527
      _CheckNodeOnline(self, src_node)
4528
      result = self.rpc.call_export_info(src_node, src_path)
4529
      result.Raise()
4530
      if not result.data:
4531
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4532

    
4533
      export_info = result.data
4534
      if not export_info.has_section(constants.INISECT_EXP):
4535
        raise errors.ProgrammerError("Corrupted export config")
4536

    
4537
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4538
      if (int(ei_version) != constants.EXPORT_VERSION):
4539
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4540
                                   (ei_version, constants.EXPORT_VERSION))
4541

    
4542
      # Check that the new instance doesn't have less disks than the export
4543
      instance_disks = len(self.disks)
4544
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4545
      if instance_disks < export_disks:
4546
        raise errors.OpPrereqError("Not enough disks to import."
4547
                                   " (instance: %d, export: %d)" %
4548
                                   (instance_disks, export_disks))
4549

    
4550
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4551
      disk_images = []
4552
      for idx in range(export_disks):
4553
        option = 'disk%d_dump' % idx
4554
        if export_info.has_option(constants.INISECT_INS, option):
4555
          # FIXME: are the old os-es, disk sizes, etc. useful?
4556
          export_name = export_info.get(constants.INISECT_INS, option)
4557
          image = os.path.join(src_path, export_name)
4558
          disk_images.append(image)
4559
        else:
4560
          disk_images.append(False)
4561

    
4562
      self.src_images = disk_images
4563

    
4564
      old_name = export_info.get(constants.INISECT_INS, 'name')
4565
      # FIXME: int() here could throw a ValueError on broken exports
4566
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4567
      if self.op.instance_name == old_name:
4568
        for idx, nic in enumerate(self.nics):
4569
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4570
            nic_mac_ini = 'nic%d_mac' % idx
4571
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4572

    
4573
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4574
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4575
    if self.op.start and not self.op.ip_check:
4576
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4577
                                 " adding an instance in start mode")
4578

    
4579
    if self.op.ip_check:
4580
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4581
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4582
                                   (self.check_ip, self.op.instance_name))
4583

    
4584
    #### mac address generation
4585
    # By generating here the mac address both the allocator and the hooks get
4586
    # the real final mac address rather than the 'auto' or 'generate' value.
4587
    # There is a race condition between the generation and the instance object
4588
    # creation, which means that we know the mac is valid now, but we're not
4589
    # sure it will be when we actually add the instance. If things go bad
4590
    # adding the instance will abort because of a duplicate mac, and the
4591
    # creation job will fail.
4592
    for nic in self.nics:
4593
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4594
        nic.mac = self.cfg.GenerateMAC()
4595

    
4596
    #### allocator run
4597

    
4598
    if self.op.iallocator is not None:
4599
      self._RunAllocator()
4600

    
4601
    #### node related checks
4602

    
4603
    # check primary node
4604
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4605
    assert self.pnode is not None, \
4606
      "Cannot retrieve locked node %s" % self.op.pnode
4607
    if pnode.offline:
4608
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4609
                                 pnode.name)
4610
    if pnode.drained:
4611
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4612
                                 pnode.name)
4613

    
4614
    self.secondaries = []
4615

    
4616
    # mirror node verification
4617
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4618
      if self.op.snode is None:
4619
        raise errors.OpPrereqError("The networked disk templates need"
4620
                                   " a mirror node")
4621
      if self.op.snode == pnode.name:
4622
        raise errors.OpPrereqError("The secondary node cannot be"
4623
                                   " the primary node.")
4624
      _CheckNodeOnline(self, self.op.snode)
4625
      _CheckNodeNotDrained(self, self.op.snode)
4626
      self.secondaries.append(self.op.snode)
4627

    
4628
    nodenames = [pnode.name] + self.secondaries
4629

    
4630
    req_size = _ComputeDiskSize(self.op.disk_template,
4631
                                self.disks)
4632

    
4633
    # Check lv size requirements
4634
    if req_size is not None:
4635
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4636
                                         self.op.hypervisor)
4637
      for node in nodenames:
4638
        info = nodeinfo[node]
4639
        info.Raise()
4640
        info = info.data
4641
        if not info:
4642
          raise errors.OpPrereqError("Cannot get current information"
4643
                                     " from node '%s'" % node)
4644
        vg_free = info.get('vg_free', None)
4645
        if not isinstance(vg_free, int):
4646
          raise errors.OpPrereqError("Can't compute free disk space on"
4647
                                     " node %s" % node)
4648
        if req_size > info['vg_free']:
4649
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4650
                                     " %d MB available, %d MB required" %
4651
                                     (node, info['vg_free'], req_size))
4652

    
4653
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4654

    
4655
    # os verification
4656
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4657
    result.Raise()
4658
    if not isinstance(result.data, objects.OS):
4659
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4660
                                 " primary node"  % self.op.os_type)
4661

    
4662
    # bridge check on primary node
4663
    bridges = [n.bridge for n in self.nics]
4664
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4665
    result.Raise()
4666
    if not result.data:
4667
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4668
                                 " exist on destination node '%s'" %
4669
                                 (",".join(bridges), pnode.name))
4670

    
4671
    # memory check on primary node
4672
    if self.op.start:
4673
      _CheckNodeFreeMemory(self, self.pnode.name,
4674
                           "creating instance %s" % self.op.instance_name,
4675
                           self.be_full[constants.BE_MEMORY],
4676
                           self.op.hypervisor)
4677

    
4678
  def Exec(self, feedback_fn):
4679
    """Create and add the instance to the cluster.
4680

4681
    """
4682
    instance = self.op.instance_name
4683
    pnode_name = self.pnode.name
4684

    
4685
    ht_kind = self.op.hypervisor
4686
    if ht_kind in constants.HTS_REQ_PORT:
4687
      network_port = self.cfg.AllocatePort()
4688
    else:
4689
      network_port = None
4690

    
4691
    ##if self.op.vnc_bind_address is None:
4692
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4693

    
4694
    # this is needed because os.path.join does not accept None arguments
4695
    if self.op.file_storage_dir is None:
4696
      string_file_storage_dir = ""
4697
    else:
4698
      string_file_storage_dir = self.op.file_storage_dir
4699

    
4700
    # build the full file storage dir path
4701
    file_storage_dir = os.path.normpath(os.path.join(
4702
                                        self.cfg.GetFileStorageDir(),
4703
                                        string_file_storage_dir, instance))
4704

    
4705

    
4706
    disks = _GenerateDiskTemplate(self,
4707
                                  self.op.disk_template,
4708
                                  instance, pnode_name,
4709
                                  self.secondaries,
4710
                                  self.disks,
4711
                                  file_storage_dir,
4712
                                  self.op.file_driver,
4713
                                  0)
4714

    
4715
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4716
                            primary_node=pnode_name,
4717
                            nics=self.nics, disks=disks,
4718
                            disk_template=self.op.disk_template,
4719
                            admin_up=False,
4720
                            network_port=network_port,
4721
                            beparams=self.op.beparams,
4722
                            hvparams=self.op.hvparams,
4723
                            hypervisor=self.op.hypervisor,
4724
                            )
4725

    
4726
    feedback_fn("* creating instance disks...")
4727
    try:
4728
      _CreateDisks(self, iobj)
4729
    except errors.OpExecError:
4730
      self.LogWarning("Device creation failed, reverting...")
4731
      try:
4732
        _RemoveDisks(self, iobj)
4733
      finally:
4734
        self.cfg.ReleaseDRBDMinors(instance)
4735
        raise
4736

    
4737
    feedback_fn("adding instance %s to cluster config" % instance)
4738

    
4739
    self.cfg.AddInstance(iobj)
4740
    # Declare that we don't want to remove the instance lock anymore, as we've
4741
    # added the instance to the config
4742
    del self.remove_locks[locking.LEVEL_INSTANCE]
4743
    # Unlock all the nodes
4744
    if self.op.mode == constants.INSTANCE_IMPORT:
4745
      nodes_keep = [self.op.src_node]
4746
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4747
                       if node != self.op.src_node]
4748
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4749
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4750
    else:
4751
      self.context.glm.release(locking.LEVEL_NODE)
4752
      del self.acquired_locks[locking.LEVEL_NODE]
4753

    
4754
    if self.op.wait_for_sync:
4755
      disk_abort = not _WaitForSync(self, iobj)
4756
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4757
      # make sure the disks are not degraded (still sync-ing is ok)
4758
      time.sleep(15)
4759
      feedback_fn("* checking mirrors status")
4760
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4761
    else:
4762
      disk_abort = False
4763

    
4764
    if disk_abort:
4765
      _RemoveDisks(self, iobj)
4766
      self.cfg.RemoveInstance(iobj.name)
4767
      # Make sure the instance lock gets removed
4768
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4769
      raise errors.OpExecError("There are some degraded disks for"
4770
                               " this instance")
4771

    
4772
    feedback_fn("creating os for instance %s on node %s" %
4773
                (instance, pnode_name))
4774

    
4775
    if iobj.disk_template != constants.DT_DISKLESS:
4776
      if self.op.mode == constants.INSTANCE_CREATE:
4777
        feedback_fn("* running the instance OS create scripts...")
4778
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4779
        msg = result.RemoteFailMsg()
4780
        if msg:
4781
          raise errors.OpExecError("Could not add os for instance %s"
4782
                                   " on node %s: %s" %
4783
                                   (instance, pnode_name, msg))
4784

    
4785
      elif self.op.mode == constants.INSTANCE_IMPORT:
4786
        feedback_fn("* running the instance OS import scripts...")
4787
        src_node = self.op.src_node
4788
        src_images = self.src_images
4789
        cluster_name = self.cfg.GetClusterName()
4790
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4791
                                                         src_node, src_images,
4792
                                                         cluster_name)
4793
        import_result.Raise()
4794
        for idx, result in enumerate(import_result.data):
4795
          if not result:
4796
            self.LogWarning("Could not import the image %s for instance"
4797
                            " %s, disk %d, on node %s" %
4798
                            (src_images[idx], instance, idx, pnode_name))
4799
      else:
4800
        # also checked in the prereq part
4801
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4802
                                     % self.op.mode)
4803

    
4804
    if self.op.start:
4805
      iobj.admin_up = True
4806
      self.cfg.Update(iobj)
4807
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4808
      feedback_fn("* starting instance...")
4809
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
4810
      msg = result.RemoteFailMsg()
4811
      if msg:
4812
        raise errors.OpExecError("Could not start instance: %s" % msg)
4813

    
4814

    
4815
class LUConnectConsole(NoHooksLU):
4816
  """Connect to an instance's console.
4817

4818
  This is somewhat special in that it returns the command line that
4819
  you need to run on the master node in order to connect to the
4820
  console.
4821

4822
  """
4823
  _OP_REQP = ["instance_name"]
4824
  REQ_BGL = False
4825

    
4826
  def ExpandNames(self):
4827
    self._ExpandAndLockInstance()
4828

    
4829
  def CheckPrereq(self):
4830
    """Check prerequisites.
4831

4832
    This checks that the instance is in the cluster.
4833

4834
    """
4835
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4836
    assert self.instance is not None, \
4837
      "Cannot retrieve locked instance %s" % self.op.instance_name
4838
    _CheckNodeOnline(self, self.instance.primary_node)
4839

    
4840
  def Exec(self, feedback_fn):
4841
    """Connect to the console of an instance
4842

4843
    """
4844
    instance = self.instance
4845
    node = instance.primary_node
4846

    
4847
    node_insts = self.rpc.call_instance_list([node],
4848
                                             [instance.hypervisor])[node]
4849
    node_insts.Raise()
4850

    
4851
    if instance.name not in node_insts.data:
4852
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4853

    
4854
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4855

    
4856
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4857
    cluster = self.cfg.GetClusterInfo()
4858
    # beparams and hvparams are passed separately, to avoid editing the
4859
    # instance and then saving the defaults in the instance itself.
4860
    hvparams = cluster.FillHV(instance)
4861
    beparams = cluster.FillBE(instance)
4862
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4863

    
4864
    # build ssh cmdline
4865
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
4866

    
4867

    
4868
class LUReplaceDisks(LogicalUnit):
4869
  """Replace the disks of an instance.
4870

4871
  """
4872
  HPATH = "mirrors-replace"
4873
  HTYPE = constants.HTYPE_INSTANCE
4874
  _OP_REQP = ["instance_name", "mode", "disks"]
4875
  REQ_BGL = False
4876

    
4877
  def CheckArguments(self):
4878
    if not hasattr(self.op, "remote_node"):
4879
      self.op.remote_node = None
4880
    if not hasattr(self.op, "iallocator"):
4881
      self.op.iallocator = None
4882

    
4883
    # check for valid parameter combination
4884
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
4885
    if self.op.mode == constants.REPLACE_DISK_CHG:
4886
      if cnt == 2:
4887
        raise errors.OpPrereqError("When changing the secondary either an"
4888
                                   " iallocator script must be used or the"
4889
                                   " new node given")
4890
      elif cnt == 0:
4891
        raise errors.OpPrereqError("Give either the iallocator or the new"
4892
                                   " secondary, not both")
4893
    else: # not replacing the secondary
4894
      if cnt != 2:
4895
        raise errors.OpPrereqError("The iallocator and new node options can"
4896
                                   " be used only when changing the"
4897
                                   " secondary node")
4898

    
4899
  def ExpandNames(self):
4900
    self._ExpandAndLockInstance()
4901

    
4902
    if self.op.iallocator is not None:
4903
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4904
    elif self.op.remote_node is not None:
4905
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
4906
      if remote_node is None:
4907
        raise errors.OpPrereqError("Node '%s' not known" %
4908
                                   self.op.remote_node)
4909
      self.op.remote_node = remote_node
4910
      # Warning: do not remove the locking of the new secondary here
4911
      # unless DRBD8.AddChildren is changed to work in parallel;
4912
      # currently it doesn't since parallel invocations of
4913
      # FindUnusedMinor will conflict
4914
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
4915
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
4916
    else:
4917
      self.needed_locks[locking.LEVEL_NODE] = []
4918
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4919

    
4920
  def DeclareLocks(self, level):
4921
    # If we're not already locking all nodes in the set we have to declare the
4922
    # instance's primary/secondary nodes.
4923
    if (level == locking.LEVEL_NODE and
4924
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
4925
      self._LockInstancesNodes()
4926

    
4927
  def _RunAllocator(self):
4928
    """Compute a new secondary node using an IAllocator.
4929

4930
    """
4931
    ial = IAllocator(self,
4932
                     mode=constants.IALLOCATOR_MODE_RELOC,
4933
                     name=self.op.instance_name,
4934
                     relocate_from=[self.sec_node])
4935

    
4936
    ial.Run(self.op.iallocator)
4937

    
4938
    if not ial.success:
4939
      raise errors.OpPrereqError("Can't compute nodes using"
4940
                                 " iallocator '%s': %s" % (self.op.iallocator,
4941
                                                           ial.info))
4942
    if len(ial.nodes) != ial.required_nodes:
4943
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4944
                                 " of nodes (%s), required %s" %
4945
                                 (len(ial.nodes), ial.required_nodes))
4946
    self.op.remote_node = ial.nodes[0]
4947
    self.LogInfo("Selected new secondary for the instance: %s",
4948
                 self.op.remote_node)
4949

    
4950
  def BuildHooksEnv(self):
4951
    """Build hooks env.
4952

4953
    This runs on the master, the primary and all the secondaries.
4954

4955
    """
4956
    env = {
4957
      "MODE": self.op.mode,
4958
      "NEW_SECONDARY": self.op.remote_node,
4959
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
4960
      }
4961
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4962
    nl = [
4963
      self.cfg.GetMasterNode(),
4964
      self.instance.primary_node,
4965
      ]
4966
    if self.op.remote_node is not None:
4967
      nl.append(self.op.remote_node)
4968
    return env, nl, nl
4969

    
4970
  def CheckPrereq(self):
4971
    """Check prerequisites.
4972

4973
    This checks that the instance is in the cluster.
4974

4975
    """
4976
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4977
    assert instance is not None, \
4978
      "Cannot retrieve locked instance %s" % self.op.instance_name
4979
    self.instance = instance
4980

    
4981
    if instance.disk_template != constants.DT_DRBD8:
4982
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
4983
                                 " instances")
4984

    
4985
    if len(instance.secondary_nodes) != 1:
4986
      raise errors.OpPrereqError("The instance has a strange layout,"
4987
                                 " expected one secondary but found %d" %
4988
                                 len(instance.secondary_nodes))
4989

    
4990
    self.sec_node = instance.secondary_nodes[0]
4991

    
4992
    if self.op.iallocator is not None:
4993
      self._RunAllocator()
4994

    
4995
    remote_node = self.op.remote_node
4996
    if remote_node is not None:
4997
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
4998
      assert self.remote_node_info is not None, \
4999
        "Cannot retrieve locked node %s" % remote_node
5000
    else:
5001
      self.remote_node_info = None
5002
    if remote_node == instance.primary_node:
5003
      raise errors.OpPrereqError("The specified node is the primary node of"
5004
                                 " the instance.")
5005
    elif remote_node == self.sec_node:
5006
      raise errors.OpPrereqError("The specified node is already the"
5007
                                 " secondary node of the instance.")
5008

    
5009
    if self.op.mode == constants.REPLACE_DISK_PRI:
5010
      n1 = self.tgt_node = instance.primary_node
5011
      n2 = self.oth_node = self.sec_node
5012
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5013
      n1 = self.tgt_node = self.sec_node
5014
      n2 = self.oth_node = instance.primary_node
5015
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5016
      n1 = self.new_node = remote_node
5017
      n2 = self.oth_node = instance.primary_node
5018
      self.tgt_node = self.sec_node
5019
      _CheckNodeNotDrained(self, remote_node)
5020
    else:
5021
      raise errors.ProgrammerError("Unhandled disk replace mode")
5022

    
5023
    _CheckNodeOnline(self, n1)
5024
    _CheckNodeOnline(self, n2)
5025

    
5026
    if not self.op.disks:
5027
      self.op.disks = range(len(instance.disks))
5028

    
5029
    for disk_idx in self.op.disks:
5030
      instance.FindDisk(disk_idx)
5031

    
5032
  def _ExecD8DiskOnly(self, feedback_fn):
5033
    """Replace a disk on the primary or secondary for dbrd8.
5034

5035
    The algorithm for replace is quite complicated:
5036

5037
      1. for each disk to be replaced:
5038

5039
        1. create new LVs on the target node with unique names
5040
        1. detach old LVs from the drbd device
5041
        1. rename old LVs to name_replaced.<time_t>
5042
        1. rename new LVs to old LVs
5043
        1. attach the new LVs (with the old names now) to the drbd device
5044

5045
      1. wait for sync across all devices
5046

5047
      1. for each modified disk:
5048

5049
        1. remove old LVs (which have the name name_replaces.<time_t>)
5050

5051
    Failures are not very well handled.
5052

5053
    """
5054
    steps_total = 6
5055
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5056
    instance = self.instance
5057
    iv_names = {}
5058
    vgname = self.cfg.GetVGName()
5059
    # start of work
5060
    cfg = self.cfg
5061
    tgt_node = self.tgt_node
5062
    oth_node = self.oth_node
5063

    
5064
    # Step: check device activation
5065
    self.proc.LogStep(1, steps_total, "check device existence")
5066
    info("checking volume groups")
5067
    my_vg = cfg.GetVGName()
5068
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5069
    if not results:
5070
      raise errors.OpExecError("Can't list volume groups on the nodes")
5071
    for node in oth_node, tgt_node:
5072
      res = results[node]
5073
      if res.failed or not res.data or my_vg not in res.data:
5074
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5075
                                 (my_vg, node))
5076
    for idx, dev in enumerate(instance.disks):
5077
      if idx not in self.op.disks:
5078
        continue
5079
      for node in tgt_node, oth_node:
5080
        info("checking disk/%d on %s" % (idx, node))
5081
        cfg.SetDiskID(dev, node)
5082
        result = self.rpc.call_blockdev_find(node, dev)
5083
        msg = result.RemoteFailMsg()
5084
        if not msg and not result.payload:
5085
          msg = "disk not found"
5086
        if msg:
5087
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5088
                                   (idx, node, msg))
5089

    
5090
    # Step: check other node consistency
5091
    self.proc.LogStep(2, steps_total, "check peer consistency")
5092
    for idx, dev in enumerate(instance.disks):
5093
      if idx not in self.op.disks:
5094
        continue
5095
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5096
      if not _CheckDiskConsistency(self, dev, oth_node,
5097
                                   oth_node==instance.primary_node):
5098
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5099
                                 " to replace disks on this node (%s)" %
5100
                                 (oth_node, tgt_node))
5101

    
5102
    # Step: create new storage
5103
    self.proc.LogStep(3, steps_total, "allocate new storage")
5104
    for idx, dev in enumerate(instance.disks):
5105
      if idx not in self.op.disks:
5106
        continue
5107
      size = dev.size
5108
      cfg.SetDiskID(dev, tgt_node)
5109
      lv_names = [".disk%d_%s" % (idx, suf)
5110
                  for suf in ["data", "meta"]]
5111
      names = _GenerateUniqueNames(self, lv_names)
5112
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5113
                             logical_id=(vgname, names[0]))
5114
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5115
                             logical_id=(vgname, names[1]))
5116
      new_lvs = [lv_data, lv_meta]
5117
      old_lvs = dev.children
5118
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5119
      info("creating new local storage on %s for %s" %
5120
           (tgt_node, dev.iv_name))
5121
      # we pass force_create=True to force the LVM creation
5122
      for new_lv in new_lvs:
5123
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5124
                        _GetInstanceInfoText(instance), False)
5125

    
5126
    # Step: for each lv, detach+rename*2+attach
5127
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5128
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5129
      info("detaching %s drbd from local storage" % dev.iv_name)
5130
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5131
      result.Raise()
5132
      if not result.data:
5133
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5134
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5135
      #dev.children = []
5136
      #cfg.Update(instance)
5137

    
5138
      # ok, we created the new LVs, so now we know we have the needed
5139
      # storage; as such, we proceed on the target node to rename
5140
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5141
      # using the assumption that logical_id == physical_id (which in
5142
      # turn is the unique_id on that node)
5143

    
5144
      # FIXME(iustin): use a better name for the replaced LVs
5145
      temp_suffix = int(time.time())
5146
      ren_fn = lambda d, suff: (d.physical_id[0],
5147
                                d.physical_id[1] + "_replaced-%s" % suff)
5148
      # build the rename list based on what LVs exist on the node
5149
      rlist = []
5150
      for to_ren in old_lvs:
5151
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5152
        if not result.RemoteFailMsg() and result.payload:
5153
          # device exists
5154
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5155

    
5156
      info("renaming the old LVs on the target node")
5157
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5158
      result.Raise()
5159
      if not result.data:
5160
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5161
      # now we rename the new LVs to the old LVs
5162
      info("renaming the new LVs on the target node")
5163
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5164
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5165
      result.Raise()
5166
      if not result.data:
5167
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5168

    
5169
      for old, new in zip(old_lvs, new_lvs):
5170
        new.logical_id = old.logical_id
5171
        cfg.SetDiskID(new, tgt_node)
5172

    
5173
      for disk in old_lvs:
5174
        disk.logical_id = ren_fn(disk, temp_suffix)
5175
        cfg.SetDiskID(disk, tgt_node)
5176

    
5177
      # now that the new lvs have the old name, we can add them to the device
5178
      info("adding new mirror component on %s" % tgt_node)
5179
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5180
      if result.failed or not result.data:
5181
        for new_lv in new_lvs:
5182
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5183
          if msg:
5184
            warning("Can't rollback device %s: %s", dev, msg,
5185
                    hint="cleanup manually the unused logical volumes")
5186
        raise errors.OpExecError("Can't add local storage to drbd")
5187

    
5188
      dev.children = new_lvs
5189
      cfg.Update(instance)
5190

    
5191
    # Step: wait for sync
5192

    
5193
    # this can fail as the old devices are degraded and _WaitForSync
5194
    # does a combined result over all disks, so we don't check its
5195
    # return value
5196
    self.proc.LogStep(5, steps_total, "sync devices")
5197
    _WaitForSync(self, instance, unlock=True)
5198

    
5199
    # so check manually all the devices
5200
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5201
      cfg.SetDiskID(dev, instance.primary_node)
5202
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5203
      msg = result.RemoteFailMsg()
5204
      if not msg and not result.payload:
5205
        msg = "disk not found"
5206
      if msg:
5207
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5208
                                 (name, msg))
5209
      if result.payload[5]:
5210
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5211

    
5212
    # Step: remove old storage
5213
    self.proc.LogStep(6, steps_total, "removing old storage")
5214
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5215
      info("remove logical volumes for %s" % name)
5216
      for lv in old_lvs:
5217
        cfg.SetDiskID(lv, tgt_node)
5218
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5219
        if msg:
5220
          warning("Can't remove old LV: %s" % msg,
5221
                  hint="manually remove unused LVs")
5222
          continue
5223

    
5224
  def _ExecD8Secondary(self, feedback_fn):
5225
    """Replace the secondary node for drbd8.
5226

5227
    The algorithm for replace is quite complicated:
5228
      - for all disks of the instance:
5229
        - create new LVs on the new node with same names
5230
        - shutdown the drbd device on the old secondary
5231
        - disconnect the drbd network on the primary
5232
        - create the drbd device on the new secondary
5233
        - network attach the drbd on the primary, using an artifice:
5234
          the drbd code for Attach() will connect to the network if it
5235
          finds a device which is connected to the good local disks but
5236
          not network enabled
5237
      - wait for sync across all devices
5238
      - remove all disks from the old secondary
5239

5240
    Failures are not very well handled.
5241

5242
    """
5243
    steps_total = 6
5244
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5245
    instance = self.instance
5246
    iv_names = {}
5247
    # start of work
5248
    cfg = self.cfg
5249
    old_node = self.tgt_node
5250
    new_node = self.new_node
5251
    pri_node = instance.primary_node
5252
    nodes_ip = {
5253
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5254
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5255
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5256
      }
5257

    
5258
    # Step: check device activation
5259
    self.proc.LogStep(1, steps_total, "check device existence")
5260
    info("checking volume groups")
5261
    my_vg = cfg.GetVGName()
5262
    results = self.rpc.call_vg_list([pri_node, new_node])
5263
    for node in pri_node, new_node:
5264
      res = results[node]
5265
      if res.failed or not res.data or my_vg not in res.data:
5266
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5267
                                 (my_vg, node))
5268
    for idx, dev in enumerate(instance.disks):
5269
      if idx not in self.op.disks:
5270
        continue
5271
      info("checking disk/%d on %s" % (idx, pri_node))
5272
      cfg.SetDiskID(dev, pri_node)
5273
      result = self.rpc.call_blockdev_find(pri_node, dev)
5274
      msg = result.RemoteFailMsg()
5275
      if not msg and not result.payload:
5276
        msg = "disk not found"
5277
      if msg:
5278
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5279
                                 (idx, pri_node, msg))
5280

    
5281
    # Step: check other node consistency
5282
    self.proc.LogStep(2, steps_total, "check peer consistency")
5283
    for idx, dev in enumerate(instance.disks):
5284
      if idx not in self.op.disks:
5285
        continue
5286
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5287
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5288
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5289
                                 " unsafe to replace the secondary" %
5290
                                 pri_node)
5291

    
5292
    # Step: create new storage
5293
    self.proc.LogStep(3, steps_total, "allocate new storage")
5294
    for idx, dev in enumerate(instance.disks):
5295
      info("adding new local storage on %s for disk/%d" %
5296
           (new_node, idx))
5297
      # we pass force_create=True to force LVM creation
5298
      for new_lv in dev.children:
5299
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5300
                        _GetInstanceInfoText(instance), False)
5301

    
5302
    # Step 4: dbrd minors and drbd setups changes
5303
    # after this, we must manually remove the drbd minors on both the
5304
    # error and the success paths
5305
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5306
                                   instance.name)
5307
    logging.debug("Allocated minors %s" % (minors,))
5308
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5309
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5310
      size = dev.size
5311
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5312
      # create new devices on new_node; note that we create two IDs:
5313
      # one without port, so the drbd will be activated without
5314
      # networking information on the new node at this stage, and one
5315
      # with network, for the latter activation in step 4
5316
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5317
      if pri_node == o_node1:
5318
        p_minor = o_minor1
5319
      else:
5320
        p_minor = o_minor2
5321

    
5322
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5323
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5324

    
5325
      iv_names[idx] = (dev, dev.children, new_net_id)
5326
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5327
                    new_net_id)
5328
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5329
                              logical_id=new_alone_id,
5330
                              children=dev.children)
5331
      try:
5332
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5333
                              _GetInstanceInfoText(instance), False)
5334
      except errors.GenericError:
5335
        self.cfg.ReleaseDRBDMinors(instance.name)
5336
        raise
5337

    
5338
    for idx, dev in enumerate(instance.disks):
5339
      # we have new devices, shutdown the drbd on the old secondary
5340
      info("shutting down drbd for disk/%d on old node" % idx)
5341
      cfg.SetDiskID(dev, old_node)
5342
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5343
      if msg:
5344
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5345
                (idx, msg),
5346
                hint="Please cleanup this device manually as soon as possible")
5347

    
5348
    info("detaching primary drbds from the network (=> standalone)")
5349
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5350
                                               instance.disks)[pri_node]
5351

    
5352
    msg = result.RemoteFailMsg()
5353
    if msg:
5354
      # detaches didn't succeed (unlikely)
5355
      self.cfg.ReleaseDRBDMinors(instance.name)
5356
      raise errors.OpExecError("Can't detach the disks from the network on"
5357
                               " old node: %s" % (msg,))
5358

    
5359
    # if we managed to detach at least one, we update all the disks of
5360
    # the instance to point to the new secondary
5361
    info("updating instance configuration")
5362
    for dev, _, new_logical_id in iv_names.itervalues():
5363
      dev.logical_id = new_logical_id
5364
      cfg.SetDiskID(dev, pri_node)
5365
    cfg.Update(instance)
5366

    
5367
    # and now perform the drbd attach
5368
    info("attaching primary drbds to new secondary (standalone => connected)")
5369
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5370
                                           instance.disks, instance.name,
5371
                                           False)
5372
    for to_node, to_result in result.items():
5373
      msg = to_result.RemoteFailMsg()
5374
      if msg:
5375
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5376
                hint="please do a gnt-instance info to see the"
5377
                " status of disks")
5378

    
5379
    # this can fail as the old devices are degraded and _WaitForSync
5380
    # does a combined result over all disks, so we don't check its
5381
    # return value
5382
    self.proc.LogStep(5, steps_total, "sync devices")
5383
    _WaitForSync(self, instance, unlock=True)
5384

    
5385
    # so check manually all the devices
5386
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5387
      cfg.SetDiskID(dev, pri_node)
5388
      result = self.rpc.call_blockdev_find(pri_node, dev)
5389
      msg = result.RemoteFailMsg()
5390
      if not msg and not result.payload:
5391
        msg = "disk not found"
5392
      if msg:
5393
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5394
                                 (idx, msg))
5395
      if result.payload[5]:
5396
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5397

    
5398
    self.proc.LogStep(6, steps_total, "removing old storage")
5399
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5400
      info("remove logical volumes for disk/%d" % idx)
5401
      for lv in old_lvs:
5402
        cfg.SetDiskID(lv, old_node)
5403
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5404
        if msg:
5405
          warning("Can't remove LV on old secondary: %s", msg,
5406
                  hint="Cleanup stale volumes by hand")
5407

    
5408
  def Exec(self, feedback_fn):
5409
    """Execute disk replacement.
5410

5411
    This dispatches the disk replacement to the appropriate handler.
5412

5413
    """
5414
    instance = self.instance
5415

    
5416
    # Activate the instance disks if we're replacing them on a down instance
5417
    if not instance.admin_up:
5418
      _StartInstanceDisks(self, instance, True)
5419

    
5420
    if self.op.mode == constants.REPLACE_DISK_CHG:
5421
      fn = self._ExecD8Secondary
5422
    else:
5423
      fn = self._ExecD8DiskOnly
5424

    
5425
    ret = fn(feedback_fn)
5426

    
5427
    # Deactivate the instance disks if we're replacing them on a down instance
5428
    if not instance.admin_up:
5429
      _SafeShutdownInstanceDisks(self, instance)
5430

    
5431
    return ret
5432

    
5433

    
5434
class LUGrowDisk(LogicalUnit):
5435
  """Grow a disk of an instance.
5436

5437
  """
5438
  HPATH = "disk-grow"
5439
  HTYPE = constants.HTYPE_INSTANCE
5440
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5441
  REQ_BGL = False
5442

    
5443
  def ExpandNames(self):
5444
    self._ExpandAndLockInstance()
5445
    self.needed_locks[locking.LEVEL_NODE] = []
5446
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5447

    
5448
  def DeclareLocks(self, level):
5449
    if level == locking.LEVEL_NODE:
5450
      self._LockInstancesNodes()
5451

    
5452
  def BuildHooksEnv(self):
5453
    """Build hooks env.
5454

5455
    This runs on the master, the primary and all the secondaries.
5456

5457
    """
5458
    env = {
5459
      "DISK": self.op.disk,
5460
      "AMOUNT": self.op.amount,
5461
      }
5462
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5463
    nl = [
5464
      self.cfg.GetMasterNode(),
5465
      self.instance.primary_node,
5466
      ]
5467
    return env, nl, nl
5468

    
5469
  def CheckPrereq(self):
5470
    """Check prerequisites.
5471

5472
    This checks that the instance is in the cluster.
5473

5474
    """
5475
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5476
    assert instance is not None, \
5477
      "Cannot retrieve locked instance %s" % self.op.instance_name
5478
    nodenames = list(instance.all_nodes)
5479
    for node in nodenames:
5480
      _CheckNodeOnline(self, node)
5481

    
5482

    
5483
    self.instance = instance
5484

    
5485
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5486
      raise errors.OpPrereqError("Instance's disk layout does not support"
5487
                                 " growing.")
5488

    
5489
    self.disk = instance.FindDisk(self.op.disk)
5490

    
5491
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5492
                                       instance.hypervisor)
5493
    for node in nodenames:
5494
      info = nodeinfo[node]
5495
      if info.failed or not info.data:
5496
        raise errors.OpPrereqError("Cannot get current information"
5497
                                   " from node '%s'" % node)
5498
      vg_free = info.data.get('vg_free', None)
5499
      if not isinstance(vg_free, int):
5500
        raise errors.OpPrereqError("Can't compute free disk space on"
5501
                                   " node %s" % node)
5502
      if self.op.amount > vg_free:
5503
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5504
                                   " %d MiB available, %d MiB required" %
5505
                                   (node, vg_free, self.op.amount))
5506

    
5507
  def Exec(self, feedback_fn):
5508
    """Execute disk grow.
5509

5510
    """
5511
    instance = self.instance
5512
    disk = self.disk
5513
    for node in instance.all_nodes:
5514
      self.cfg.SetDiskID(disk, node)
5515
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5516
      msg = result.RemoteFailMsg()
5517
      if msg:
5518
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5519
                                 (node, msg))
5520
    disk.RecordGrow(self.op.amount)
5521
    self.cfg.Update(instance)
5522
    if self.op.wait_for_sync:
5523
      disk_abort = not _WaitForSync(self, instance)
5524
      if disk_abort:
5525
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5526
                             " status.\nPlease check the instance.")
5527

    
5528

    
5529
class LUQueryInstanceData(NoHooksLU):
5530
  """Query runtime instance data.
5531

5532
  """
5533
  _OP_REQP = ["instances", "static"]
5534
  REQ_BGL = False
5535

    
5536
  def ExpandNames(self):
5537
    self.needed_locks = {}
5538
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5539

    
5540
    if not isinstance(self.op.instances, list):
5541
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5542

    
5543
    if self.op.instances:
5544
      self.wanted_names = []
5545
      for name in self.op.instances:
5546
        full_name = self.cfg.ExpandInstanceName(name)
5547
        if full_name is None:
5548
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5549
        self.wanted_names.append(full_name)
5550
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5551
    else:
5552
      self.wanted_names = None
5553
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5554

    
5555
    self.needed_locks[locking.LEVEL_NODE] = []
5556
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5557

    
5558
  def DeclareLocks(self, level):
5559
    if level == locking.LEVEL_NODE:
5560
      self._LockInstancesNodes()
5561

    
5562
  def CheckPrereq(self):
5563
    """Check prerequisites.
5564

5565
    This only checks the optional instance list against the existing names.
5566

5567
    """
5568
    if self.wanted_names is None:
5569
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5570

    
5571
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5572
                             in self.wanted_names]
5573
    return
5574

    
5575
  def _ComputeDiskStatus(self, instance, snode, dev):
5576
    """Compute block device status.
5577

5578
    """
5579
    static = self.op.static
5580
    if not static:
5581
      self.cfg.SetDiskID(dev, instance.primary_node)
5582
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5583
      if dev_pstatus.offline:
5584
        dev_pstatus = None
5585
      else:
5586
        msg = dev_pstatus.RemoteFailMsg()
5587
        if msg:
5588
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5589
                                   (instance.name, msg))
5590
        dev_pstatus = dev_pstatus.payload
5591
    else:
5592
      dev_pstatus = None
5593

    
5594
    if dev.dev_type in constants.LDS_DRBD:
5595
      # we change the snode then (otherwise we use the one passed in)
5596
      if dev.logical_id[0] == instance.primary_node:
5597
        snode = dev.logical_id[1]
5598
      else:
5599
        snode = dev.logical_id[0]
5600

    
5601
    if snode and not static:
5602
      self.cfg.SetDiskID(dev, snode)
5603
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5604
      if dev_sstatus.offline:
5605
        dev_sstatus = None
5606
      else:
5607
        msg = dev_sstatus.RemoteFailMsg()
5608
        if msg:
5609
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5610
                                   (instance.name, msg))
5611
        dev_sstatus = dev_sstatus.payload
5612
    else:
5613
      dev_sstatus = None
5614

    
5615
    if dev.children:
5616
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5617
                      for child in dev.children]
5618
    else:
5619
      dev_children = []
5620

    
5621
    data = {
5622
      "iv_name": dev.iv_name,
5623
      "dev_type": dev.dev_type,
5624
      "logical_id": dev.logical_id,
5625
      "physical_id": dev.physical_id,
5626
      "pstatus": dev_pstatus,
5627
      "sstatus": dev_sstatus,
5628
      "children": dev_children,
5629
      "mode": dev.mode,
5630
      }
5631

    
5632
    return data
5633

    
5634
  def Exec(self, feedback_fn):
5635
    """Gather and return data"""
5636
    result = {}
5637

    
5638
    cluster = self.cfg.GetClusterInfo()
5639

    
5640
    for instance in self.wanted_instances:
5641
      if not self.op.static:
5642
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5643
                                                  instance.name,
5644
                                                  instance.hypervisor)
5645
        remote_info.Raise()
5646
        remote_info = remote_info.data
5647
        if remote_info and "state" in remote_info:
5648
          remote_state = "up"
5649
        else:
5650
          remote_state = "down"
5651
      else:
5652
        remote_state = None
5653
      if instance.admin_up:
5654
        config_state = "up"
5655
      else:
5656
        config_state = "down"
5657

    
5658
      disks = [self._ComputeDiskStatus(instance, None, device)
5659
               for device in instance.disks]
5660

    
5661
      idict = {
5662
        "name": instance.name,
5663
        "config_state": config_state,
5664
        "run_state": remote_state,
5665
        "pnode": instance.primary_node,
5666
        "snodes": instance.secondary_nodes,
5667
        "os": instance.os,
5668
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5669
        "disks": disks,
5670
        "hypervisor": instance.hypervisor,
5671
        "network_port": instance.network_port,
5672
        "hv_instance": instance.hvparams,
5673
        "hv_actual": cluster.FillHV(instance),
5674
        "be_instance": instance.beparams,
5675
        "be_actual": cluster.FillBE(instance),
5676
        }
5677

    
5678
      result[instance.name] = idict
5679

    
5680
    return result
5681

    
5682

    
5683
class LUSetInstanceParams(LogicalUnit):
5684
  """Modifies an instances's parameters.
5685

5686
  """
5687
  HPATH = "instance-modify"
5688
  HTYPE = constants.HTYPE_INSTANCE
5689
  _OP_REQP = ["instance_name"]
5690
  REQ_BGL = False
5691

    
5692
  def CheckArguments(self):
5693
    if not hasattr(self.op, 'nics'):
5694
      self.op.nics = []
5695
    if not hasattr(self.op, 'disks'):
5696
      self.op.disks = []
5697
    if not hasattr(self.op, 'beparams'):
5698
      self.op.beparams = {}
5699
    if not hasattr(self.op, 'hvparams'):
5700
      self.op.hvparams = {}
5701
    self.op.force = getattr(self.op, "force", False)
5702
    if not (self.op.nics or self.op.disks or
5703
            self.op.hvparams or self.op.beparams):
5704
      raise errors.OpPrereqError("No changes submitted")
5705

    
5706
    # Disk validation
5707
    disk_addremove = 0
5708
    for disk_op, disk_dict in self.op.disks:
5709
      if disk_op == constants.DDM_REMOVE:
5710
        disk_addremove += 1
5711
        continue
5712
      elif disk_op == constants.DDM_ADD:
5713
        disk_addremove += 1
5714
      else:
5715
        if not isinstance(disk_op, int):
5716
          raise errors.OpPrereqError("Invalid disk index")
5717
      if disk_op == constants.DDM_ADD:
5718
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5719
        if mode not in constants.DISK_ACCESS_SET:
5720
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5721
        size = disk_dict.get('size', None)
5722
        if size is None:
5723
          raise errors.OpPrereqError("Required disk parameter size missing")
5724
        try:
5725
          size = int(size)
5726
        except ValueError, err:
5727
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5728
                                     str(err))
5729
        disk_dict['size'] = size
5730
      else:
5731
        # modification of disk
5732
        if 'size' in disk_dict:
5733
          raise errors.OpPrereqError("Disk size change not possible, use"
5734
                                     " grow-disk")
5735

    
5736
    if disk_addremove > 1:
5737
      raise errors.OpPrereqError("Only one disk add or remove operation"
5738
                                 " supported at a time")
5739

    
5740
    # NIC validation
5741
    nic_addremove = 0
5742
    for nic_op, nic_dict in self.op.nics:
5743
      if nic_op == constants.DDM_REMOVE:
5744
        nic_addremove += 1
5745
        continue
5746
      elif nic_op == constants.DDM_ADD:
5747
        nic_addremove += 1
5748
      else:
5749
        if not isinstance(nic_op, int):
5750
          raise errors.OpPrereqError("Invalid nic index")
5751

    
5752
      # nic_dict should be a dict
5753
      nic_ip = nic_dict.get('ip', None)
5754
      if nic_ip is not None:
5755
        if nic_ip.lower() == constants.VALUE_NONE:
5756
          nic_dict['ip'] = None
5757
        else:
5758
          if not utils.IsValidIP(nic_ip):
5759
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5760

    
5761
      if nic_op == constants.DDM_ADD:
5762
        nic_bridge = nic_dict.get('bridge', None)
5763
        if nic_bridge is None:
5764
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5765
        nic_mac = nic_dict.get('mac', None)
5766
        if nic_mac is None:
5767
          nic_dict['mac'] = constants.VALUE_AUTO
5768

    
5769
      if 'mac' in nic_dict:
5770
        nic_mac = nic_dict['mac']
5771
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5772
          if not utils.IsValidMac(nic_mac):
5773
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5774
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5775
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5776
                                     " modifying an existing nic")
5777

    
5778
    if nic_addremove > 1:
5779
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5780
                                 " supported at a time")
5781

    
5782
  def ExpandNames(self):
5783
    self._ExpandAndLockInstance()
5784
    self.needed_locks[locking.LEVEL_NODE] = []
5785
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5786

    
5787
  def DeclareLocks(self, level):
5788
    if level == locking.LEVEL_NODE:
5789
      self._LockInstancesNodes()
5790

    
5791
  def BuildHooksEnv(self):
5792
    """Build hooks env.
5793

5794
    This runs on the master, primary and secondaries.
5795

5796
    """
5797
    args = dict()
5798
    if constants.BE_MEMORY in self.be_new:
5799
      args['memory'] = self.be_new[constants.BE_MEMORY]
5800
    if constants.BE_VCPUS in self.be_new:
5801
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5802
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5803
    # information at all.
5804
    if self.op.nics:
5805
      args['nics'] = []
5806
      nic_override = dict(self.op.nics)
5807
      for idx, nic in enumerate(self.instance.nics):
5808
        if idx in nic_override:
5809
          this_nic_override = nic_override[idx]
5810
        else:
5811
          this_nic_override = {}
5812
        if 'ip' in this_nic_override:
5813
          ip = this_nic_override['ip']
5814
        else:
5815
          ip = nic.ip
5816
        if 'bridge' in this_nic_override:
5817
          bridge = this_nic_override['bridge']
5818
        else:
5819
          bridge = nic.bridge
5820
        if 'mac' in this_nic_override:
5821
          mac = this_nic_override['mac']
5822
        else:
5823
          mac = nic.mac
5824
        args['nics'].append((ip, bridge, mac))
5825
      if constants.DDM_ADD in nic_override:
5826
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5827
        bridge = nic_override[constants.DDM_ADD]['bridge']
5828
        mac = nic_override[constants.DDM_ADD]['mac']
5829
        args['nics'].append((ip, bridge, mac))
5830
      elif constants.DDM_REMOVE in nic_override:
5831
        del args['nics'][-1]
5832

    
5833
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5834
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5835
    return env, nl, nl
5836

    
5837
  def CheckPrereq(self):
5838
    """Check prerequisites.
5839

5840
    This only checks the instance list against the existing names.
5841

5842
    """
5843
    force = self.force = self.op.force
5844

    
5845
    # checking the new params on the primary/secondary nodes
5846

    
5847
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5848
    assert self.instance is not None, \
5849
      "Cannot retrieve locked instance %s" % self.op.instance_name
5850
    pnode = instance.primary_node
5851
    nodelist = list(instance.all_nodes)
5852

    
5853
    # hvparams processing
5854
    if self.op.hvparams:
5855
      i_hvdict = copy.deepcopy(instance.hvparams)
5856
      for key, val in self.op.hvparams.iteritems():
5857
        if val == constants.VALUE_DEFAULT:
5858
          try:
5859
            del i_hvdict[key]
5860
          except KeyError:
5861
            pass
5862
        else:
5863
          i_hvdict[key] = val
5864
      cluster = self.cfg.GetClusterInfo()
5865
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
5866
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
5867
                                i_hvdict)
5868
      # local check
5869
      hypervisor.GetHypervisor(
5870
        instance.hypervisor).CheckParameterSyntax(hv_new)
5871
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
5872
      self.hv_new = hv_new # the new actual values
5873
      self.hv_inst = i_hvdict # the new dict (without defaults)
5874
    else:
5875
      self.hv_new = self.hv_inst = {}
5876

    
5877
    # beparams processing
5878
    if self.op.beparams:
5879
      i_bedict = copy.deepcopy(instance.beparams)
5880
      for key, val in self.op.beparams.iteritems():
5881
        if val == constants.VALUE_DEFAULT:
5882
          try:
5883
            del i_bedict[key]
5884
          except KeyError:
5885
            pass
5886
        else:
5887
          i_bedict[key] = val
5888
      cluster = self.cfg.GetClusterInfo()
5889
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
5890
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
5891
                                i_bedict)
5892
      self.be_new = be_new # the new actual values
5893
      self.be_inst = i_bedict # the new dict (without defaults)
5894
    else:
5895
      self.be_new = self.be_inst = {}
5896

    
5897
    self.warn = []
5898

    
5899
    if constants.BE_MEMORY in self.op.beparams and not self.force:
5900
      mem_check_list = [pnode]
5901
      if be_new[constants.BE_AUTO_BALANCE]:
5902
        # either we changed auto_balance to yes or it was from before
5903
        mem_check_list.extend(instance.secondary_nodes)
5904
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
5905
                                                  instance.hypervisor)
5906
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
5907
                                         instance.hypervisor)
5908
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
5909
        # Assume the primary node is unreachable and go ahead
5910
        self.warn.append("Can't get info from primary node %s" % pnode)
5911
      else:
5912
        if not instance_info.failed and instance_info.data:
5913
          current_mem = instance_info.data['memory']
5914
        else:
5915
          # Assume instance not running
5916
          # (there is a slight race condition here, but it's not very probable,
5917
          # and we have no other way to check)
5918
          current_mem = 0
5919
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
5920
                    nodeinfo[pnode].data['memory_free'])
5921
        if miss_mem > 0:
5922
          raise errors.OpPrereqError("This change will prevent the instance"
5923
                                     " from starting, due to %d MB of memory"
5924
                                     " missing on its primary node" % miss_mem)
5925

    
5926
      if be_new[constants.BE_AUTO_BALANCE]:
5927
        for node, nres in nodeinfo.iteritems():
5928
          if node not in instance.secondary_nodes:
5929
            continue
5930
          if nres.failed or not isinstance(nres.data, dict):
5931
            self.warn.append("Can't get info from secondary node %s" % node)
5932
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
5933
            self.warn.append("Not enough memory to failover instance to"
5934
                             " secondary node %s" % node)
5935

    
5936
    # NIC processing
5937
    for nic_op, nic_dict in self.op.nics:
5938
      if nic_op == constants.DDM_REMOVE:
5939
        if not instance.nics:
5940
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
5941
        continue
5942
      if nic_op != constants.DDM_ADD:
5943
        # an existing nic
5944
        if nic_op < 0 or nic_op >= len(instance.nics):
5945
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
5946
                                     " are 0 to %d" %
5947
                                     (nic_op, len(instance.nics)))
5948
      if 'bridge' in nic_dict:
5949
        nic_bridge = nic_dict['bridge']
5950
        if nic_bridge is None:
5951
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
5952
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
5953
          msg = ("Bridge '%s' doesn't exist on one of"
5954
                 " the instance nodes" % nic_bridge)
5955
          if self.force:
5956
            self.warn.append(msg)
5957
          else:
5958
            raise errors.OpPrereqError(msg)
5959
      if 'mac' in nic_dict:
5960
        nic_mac = nic_dict['mac']
5961
        if nic_mac is None:
5962
          raise errors.OpPrereqError('Cannot set the nic mac to None')
5963
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5964
          # otherwise generate the mac
5965
          nic_dict['mac'] = self.cfg.GenerateMAC()
5966
        else:
5967
          # or validate/reserve the current one
5968
          if self.cfg.IsMacInUse(nic_mac):
5969
            raise errors.OpPrereqError("MAC address %s already in use"
5970
                                       " in cluster" % nic_mac)
5971

    
5972
    # DISK processing
5973
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
5974
      raise errors.OpPrereqError("Disk operations not supported for"
5975
                                 " diskless instances")
5976
    for disk_op, disk_dict in self.op.disks:
5977
      if disk_op == constants.DDM_REMOVE:
5978
        if len(instance.disks) == 1:
5979
          raise errors.OpPrereqError("Cannot remove the last disk of"
5980
                                     " an instance")
5981
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
5982
        ins_l = ins_l[pnode]
5983
        if ins_l.failed or not isinstance(ins_l.data, list):
5984
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
5985
        if instance.name in ins_l.data:
5986
          raise errors.OpPrereqError("Instance is running, can't remove"
5987
                                     " disks.")
5988

    
5989
      if (disk_op == constants.DDM_ADD and
5990
          len(instance.nics) >= constants.MAX_DISKS):
5991
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
5992
                                   " add more" % constants.MAX_DISKS)
5993
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
5994
        # an existing disk
5995
        if disk_op < 0 or disk_op >= len(instance.disks):
5996
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
5997
                                     " are 0 to %d" %
5998
                                     (disk_op, len(instance.disks)))
5999

    
6000
    return
6001

    
6002
  def Exec(self, feedback_fn):
6003
    """Modifies an instance.
6004

6005
    All parameters take effect only at the next restart of the instance.
6006

6007
    """
6008
    # Process here the warnings from CheckPrereq, as we don't have a
6009
    # feedback_fn there.
6010
    for warn in self.warn:
6011
      feedback_fn("WARNING: %s" % warn)
6012

    
6013
    result = []
6014
    instance = self.instance
6015
    # disk changes
6016
    for disk_op, disk_dict in self.op.disks:
6017
      if disk_op == constants.DDM_REMOVE:
6018
        # remove the last disk
6019
        device = instance.disks.pop()
6020
        device_idx = len(instance.disks)
6021
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6022
          self.cfg.SetDiskID(disk, node)
6023
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6024
          if msg:
6025
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6026
                            " continuing anyway", device_idx, node, msg)
6027
        result.append(("disk/%d" % device_idx, "remove"))
6028
      elif disk_op == constants.DDM_ADD:
6029
        # add a new disk
6030
        if instance.disk_template == constants.DT_FILE:
6031
          file_driver, file_path = instance.disks[0].logical_id
6032
          file_path = os.path.dirname(file_path)
6033
        else:
6034
          file_driver = file_path = None
6035
        disk_idx_base = len(instance.disks)
6036
        new_disk = _GenerateDiskTemplate(self,
6037
                                         instance.disk_template,
6038
                                         instance.name, instance.primary_node,
6039
                                         instance.secondary_nodes,
6040
                                         [disk_dict],
6041
                                         file_path,
6042
                                         file_driver,
6043
                                         disk_idx_base)[0]
6044
        instance.disks.append(new_disk)
6045
        info = _GetInstanceInfoText(instance)
6046

    
6047
        logging.info("Creating volume %s for instance %s",
6048
                     new_disk.iv_name, instance.name)
6049
        # Note: this needs to be kept in sync with _CreateDisks
6050
        #HARDCODE
6051
        for node in instance.all_nodes:
6052
          f_create = node == instance.primary_node
6053
          try:
6054
            _CreateBlockDev(self, node, instance, new_disk,
6055
                            f_create, info, f_create)
6056
          except errors.OpExecError, err:
6057
            self.LogWarning("Failed to create volume %s (%s) on"
6058
                            " node %s: %s",
6059
                            new_disk.iv_name, new_disk, node, err)
6060
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6061
                       (new_disk.size, new_disk.mode)))
6062
      else:
6063
        # change a given disk
6064
        instance.disks[disk_op].mode = disk_dict['mode']
6065
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6066
    # NIC changes
6067
    for nic_op, nic_dict in self.op.nics:
6068
      if nic_op == constants.DDM_REMOVE:
6069
        # remove the last nic
6070
        del instance.nics[-1]
6071
        result.append(("nic.%d" % len(instance.nics), "remove"))
6072
      elif nic_op == constants.DDM_ADD:
6073
        # mac and bridge should be set, by now
6074
        mac = nic_dict['mac']
6075
        bridge = nic_dict['bridge']
6076
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6077
                              bridge=bridge)
6078
        instance.nics.append(new_nic)
6079
        result.append(("nic.%d" % (len(instance.nics) - 1),
6080
                       "add:mac=%s,ip=%s,bridge=%s" %
6081
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6082
      else:
6083
        # change a given nic
6084
        for key in 'mac', 'ip', 'bridge':
6085
          if key in nic_dict:
6086
            setattr(instance.nics[nic_op], key, nic_dict[key])
6087
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6088

    
6089
    # hvparams changes
6090
    if self.op.hvparams:
6091
      instance.hvparams = self.hv_inst
6092
      for key, val in self.op.hvparams.iteritems():
6093
        result.append(("hv/%s" % key, val))
6094

    
6095
    # beparams changes
6096
    if self.op.beparams:
6097
      instance.beparams = self.be_inst
6098
      for key, val in self.op.beparams.iteritems():
6099
        result.append(("be/%s" % key, val))
6100

    
6101
    self.cfg.Update(instance)
6102

    
6103
    return result
6104

    
6105

    
6106
class LUQueryExports(NoHooksLU):
6107
  """Query the exports list
6108

6109
  """
6110
  _OP_REQP = ['nodes']
6111
  REQ_BGL = False
6112

    
6113
  def ExpandNames(self):
6114
    self.needed_locks = {}
6115
    self.share_locks[locking.LEVEL_NODE] = 1
6116
    if not self.op.nodes:
6117
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6118
    else:
6119
      self.needed_locks[locking.LEVEL_NODE] = \
6120
        _GetWantedNodes(self, self.op.nodes)
6121

    
6122
  def CheckPrereq(self):
6123
    """Check prerequisites.
6124

6125
    """
6126
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6127

    
6128
  def Exec(self, feedback_fn):
6129
    """Compute the list of all the exported system images.
6130

6131
    @rtype: dict
6132
    @return: a dictionary with the structure node->(export-list)
6133
        where export-list is a list of the instances exported on
6134
        that node.
6135

6136
    """
6137
    rpcresult = self.rpc.call_export_list(self.nodes)
6138
    result = {}
6139
    for node in rpcresult:
6140
      if rpcresult[node].failed:
6141
        result[node] = False
6142
      else:
6143
        result[node] = rpcresult[node].data
6144

    
6145
    return result
6146

    
6147

    
6148
class LUExportInstance(LogicalUnit):
6149
  """Export an instance to an image in the cluster.
6150

6151
  """
6152
  HPATH = "instance-export"
6153
  HTYPE = constants.HTYPE_INSTANCE
6154
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6155
  REQ_BGL = False
6156

    
6157
  def ExpandNames(self):
6158
    self._ExpandAndLockInstance()
6159
    # FIXME: lock only instance primary and destination node
6160
    #
6161
    # Sad but true, for now we have do lock all nodes, as we don't know where
6162
    # the previous export might be, and and in this LU we search for it and
6163
    # remove it from its current node. In the future we could fix this by:
6164
    #  - making a tasklet to search (share-lock all), then create the new one,
6165
    #    then one to remove, after
6166
    #  - removing the removal operation altoghether
6167
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6168

    
6169
  def DeclareLocks(self, level):
6170
    """Last minute lock declaration."""
6171
    # All nodes are locked anyway, so nothing to do here.
6172

    
6173
  def BuildHooksEnv(self):
6174
    """Build hooks env.
6175

6176
    This will run on the master, primary node and target node.
6177

6178
    """
6179
    env = {
6180
      "EXPORT_NODE": self.op.target_node,
6181
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6182
      }
6183
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6184
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6185
          self.op.target_node]
6186
    return env, nl, nl
6187

    
6188
  def CheckPrereq(self):
6189
    """Check prerequisites.
6190

6191
    This checks that the instance and node names are valid.
6192

6193
    """
6194
    instance_name = self.op.instance_name
6195
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6196
    assert self.instance is not None, \
6197
          "Cannot retrieve locked instance %s" % self.op.instance_name
6198
    _CheckNodeOnline(self, self.instance.primary_node)
6199

    
6200
    self.dst_node = self.cfg.GetNodeInfo(
6201
      self.cfg.ExpandNodeName(self.op.target_node))
6202

    
6203
    if self.dst_node is None:
6204
      # This is wrong node name, not a non-locked node
6205
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6206
    _CheckNodeOnline(self, self.dst_node.name)
6207
    _CheckNodeNotDrained(self, self.dst_node.name)
6208

    
6209
    # instance disk type verification
6210
    for disk in self.instance.disks:
6211
      if disk.dev_type == constants.LD_FILE:
6212
        raise errors.OpPrereqError("Export not supported for instances with"
6213
                                   " file-based disks")
6214

    
6215
  def Exec(self, feedback_fn):
6216
    """Export an instance to an image in the cluster.
6217

6218
    """
6219
    instance = self.instance
6220
    dst_node = self.dst_node
6221
    src_node = instance.primary_node
6222
    if self.op.shutdown:
6223
      # shutdown the instance, but not the disks
6224
      result = self.rpc.call_instance_shutdown(src_node, instance)
6225
      msg = result.RemoteFailMsg()
6226
      if msg:
6227
        raise errors.OpExecError("Could not shutdown instance %s on"
6228
                                 " node %s: %s" %
6229
                                 (instance.name, src_node, msg))
6230

    
6231
    vgname = self.cfg.GetVGName()
6232

    
6233
    snap_disks = []
6234

    
6235
    # set the disks ID correctly since call_instance_start needs the
6236
    # correct drbd minor to create the symlinks
6237
    for disk in instance.disks:
6238
      self.cfg.SetDiskID(disk, src_node)
6239

    
6240
    try:
6241
      for disk in instance.disks:
6242
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6243
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6244
        if new_dev_name.failed or not new_dev_name.data:
6245
          self.LogWarning("Could not snapshot block device %s on node %s",
6246
                          disk.logical_id[1], src_node)
6247
          snap_disks.append(False)
6248
        else:
6249
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6250
                                 logical_id=(vgname, new_dev_name.data),
6251
                                 physical_id=(vgname, new_dev_name.data),
6252
                                 iv_name=disk.iv_name)
6253
          snap_disks.append(new_dev)
6254

    
6255
    finally:
6256
      if self.op.shutdown and instance.admin_up:
6257
        result = self.rpc.call_instance_start(src_node, instance, None, None)
6258
        msg = result.RemoteFailMsg()
6259
        if msg:
6260
          _ShutdownInstanceDisks(self, instance)
6261
          raise errors.OpExecError("Could not start instance: %s" % msg)
6262

    
6263
    # TODO: check for size
6264

    
6265
    cluster_name = self.cfg.GetClusterName()
6266
    for idx, dev in enumerate(snap_disks):
6267
      if dev:
6268
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6269
                                               instance, cluster_name, idx)
6270
        if result.failed or not result.data:
6271
          self.LogWarning("Could not export block device %s from node %s to"
6272
                          " node %s", dev.logical_id[1], src_node,
6273
                          dst_node.name)
6274
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6275
        if msg:
6276
          self.LogWarning("Could not remove snapshot block device %s from node"
6277
                          " %s: %s", dev.logical_id[1], src_node, msg)
6278

    
6279
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6280
    if result.failed or not result.data:
6281
      self.LogWarning("Could not finalize export for instance %s on node %s",
6282
                      instance.name, dst_node.name)
6283

    
6284
    nodelist = self.cfg.GetNodeList()
6285
    nodelist.remove(dst_node.name)
6286

    
6287
    # on one-node clusters nodelist will be empty after the removal
6288
    # if we proceed the backup would be removed because OpQueryExports
6289
    # substitutes an empty list with the full cluster node list.
6290
    if nodelist:
6291
      exportlist = self.rpc.call_export_list(nodelist)
6292
      for node in exportlist:
6293
        if exportlist[node].failed:
6294
          continue
6295
        if instance.name in exportlist[node].data:
6296
          if not self.rpc.call_export_remove(node, instance.name):
6297
            self.LogWarning("Could not remove older export for instance %s"
6298
                            " on node %s", instance.name, node)
6299

    
6300

    
6301
class LURemoveExport(NoHooksLU):
6302
  """Remove exports related to the named instance.
6303

6304
  """
6305
  _OP_REQP = ["instance_name"]
6306
  REQ_BGL = False
6307

    
6308
  def ExpandNames(self):
6309
    self.needed_locks = {}
6310
    # We need all nodes to be locked in order for RemoveExport to work, but we
6311
    # don't need to lock the instance itself, as nothing will happen to it (and
6312
    # we can remove exports also for a removed instance)
6313
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6314

    
6315
  def CheckPrereq(self):
6316
    """Check prerequisites.
6317
    """
6318
    pass
6319

    
6320
  def Exec(self, feedback_fn):
6321
    """Remove any export.
6322

6323
    """
6324
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6325
    # If the instance was not found we'll try with the name that was passed in.
6326
    # This will only work if it was an FQDN, though.
6327
    fqdn_warn = False
6328
    if not instance_name:
6329
      fqdn_warn = True
6330
      instance_name = self.op.instance_name
6331

    
6332
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6333
      locking.LEVEL_NODE])
6334
    found = False
6335
    for node in exportlist:
6336
      if exportlist[node].failed:
6337
        self.LogWarning("Failed to query node %s, continuing" % node)
6338
        continue
6339
      if instance_name in exportlist[node].data:
6340
        found = True
6341
        result = self.rpc.call_export_remove(node, instance_name)
6342
        if result.failed or not result.data:
6343
          logging.error("Could not remove export for instance %s"
6344
                        " on node %s", instance_name, node)
6345

    
6346
    if fqdn_warn and not found:
6347
      feedback_fn("Export not found. If trying to remove an export belonging"
6348
                  " to a deleted instance please use its Fully Qualified"
6349
                  " Domain Name.")
6350

    
6351

    
6352
class TagsLU(NoHooksLU):
6353
  """Generic tags LU.
6354

6355
  This is an abstract class which is the parent of all the other tags LUs.
6356

6357
  """
6358

    
6359
  def ExpandNames(self):
6360
    self.needed_locks = {}
6361
    if self.op.kind == constants.TAG_NODE:
6362
      name = self.cfg.ExpandNodeName(self.op.name)
6363
      if name is None:
6364
        raise errors.OpPrereqError("Invalid node name (%s)" %
6365
                                   (self.op.name,))
6366
      self.op.name = name
6367
      self.needed_locks[locking.LEVEL_NODE] = name
6368
    elif self.op.kind == constants.TAG_INSTANCE:
6369
      name = self.cfg.ExpandInstanceName(self.op.name)
6370
      if name is None:
6371
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6372
                                   (self.op.name,))
6373
      self.op.name = name
6374
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6375

    
6376
  def CheckPrereq(self):
6377
    """Check prerequisites.
6378

6379
    """
6380
    if self.op.kind == constants.TAG_CLUSTER:
6381
      self.target = self.cfg.GetClusterInfo()
6382
    elif self.op.kind == constants.TAG_NODE:
6383
      self.target = self.cfg.GetNodeInfo(self.op.name)
6384
    elif self.op.kind == constants.TAG_INSTANCE:
6385
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6386
    else:
6387
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6388
                                 str(self.op.kind))
6389

    
6390

    
6391
class LUGetTags(TagsLU):
6392
  """Returns the tags of a given object.
6393

6394
  """
6395
  _OP_REQP = ["kind", "name"]
6396
  REQ_BGL = False
6397

    
6398
  def Exec(self, feedback_fn):
6399
    """Returns the tag list.
6400

6401
    """
6402
    return list(self.target.GetTags())
6403

    
6404

    
6405
class LUSearchTags(NoHooksLU):
6406
  """Searches the tags for a given pattern.
6407

6408
  """
6409
  _OP_REQP = ["pattern"]
6410
  REQ_BGL = False
6411

    
6412
  def ExpandNames(self):
6413
    self.needed_locks = {}
6414

    
6415
  def CheckPrereq(self):
6416
    """Check prerequisites.
6417

6418
    This checks the pattern passed for validity by compiling it.
6419

6420
    """
6421
    try:
6422
      self.re = re.compile(self.op.pattern)
6423
    except re.error, err:
6424
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6425
                                 (self.op.pattern, err))
6426

    
6427
  def Exec(self, feedback_fn):
6428
    """Returns the tag list.
6429

6430
    """
6431
    cfg = self.cfg
6432
    tgts = [("/cluster", cfg.GetClusterInfo())]
6433
    ilist = cfg.GetAllInstancesInfo().values()
6434
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6435
    nlist = cfg.GetAllNodesInfo().values()
6436
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6437
    results = []
6438
    for path, target in tgts:
6439
      for tag in target.GetTags():
6440
        if self.re.search(tag):
6441
          results.append((path, tag))
6442
    return results
6443

    
6444

    
6445
class LUAddTags(TagsLU):
6446
  """Sets a tag on a given object.
6447

6448
  """
6449
  _OP_REQP = ["kind", "name", "tags"]
6450
  REQ_BGL = False
6451

    
6452
  def CheckPrereq(self):
6453
    """Check prerequisites.
6454

6455
    This checks the type and length of the tag name and value.
6456

6457
    """
6458
    TagsLU.CheckPrereq(self)
6459
    for tag in self.op.tags:
6460
      objects.TaggableObject.ValidateTag(tag)
6461

    
6462
  def Exec(self, feedback_fn):
6463
    """Sets the tag.
6464

6465
    """
6466
    try:
6467
      for tag in self.op.tags:
6468
        self.target.AddTag(tag)
6469
    except errors.TagError, err:
6470
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6471
    try:
6472
      self.cfg.Update(self.target)
6473
    except errors.ConfigurationError:
6474
      raise errors.OpRetryError("There has been a modification to the"
6475
                                " config file and the operation has been"
6476
                                " aborted. Please retry.")
6477

    
6478

    
6479
class LUDelTags(TagsLU):
6480
  """Delete a list of tags from a given object.
6481

6482
  """
6483
  _OP_REQP = ["kind", "name", "tags"]
6484
  REQ_BGL = False
6485

    
6486
  def CheckPrereq(self):
6487
    """Check prerequisites.
6488

6489
    This checks that we have the given tag.
6490

6491
    """
6492
    TagsLU.CheckPrereq(self)
6493
    for tag in self.op.tags:
6494
      objects.TaggableObject.ValidateTag(tag)
6495
    del_tags = frozenset(self.op.tags)
6496
    cur_tags = self.target.GetTags()
6497
    if not del_tags <= cur_tags:
6498
      diff_tags = del_tags - cur_tags
6499
      diff_names = ["'%s'" % tag for tag in diff_tags]
6500
      diff_names.sort()
6501
      raise errors.OpPrereqError("Tag(s) %s not found" %
6502
                                 (",".join(diff_names)))
6503

    
6504
  def Exec(self, feedback_fn):
6505
    """Remove the tag from the object.
6506

6507
    """
6508
    for tag in self.op.tags:
6509
      self.target.RemoveTag(tag)
6510
    try:
6511
      self.cfg.Update(self.target)
6512
    except errors.ConfigurationError:
6513
      raise errors.OpRetryError("There has been a modification to the"
6514
                                " config file and the operation has been"
6515
                                " aborted. Please retry.")
6516

    
6517

    
6518
class LUTestDelay(NoHooksLU):
6519
  """Sleep for a specified amount of time.
6520

6521
  This LU sleeps on the master and/or nodes for a specified amount of
6522
  time.
6523

6524
  """
6525
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6526
  REQ_BGL = False
6527

    
6528
  def ExpandNames(self):
6529
    """Expand names and set required locks.
6530

6531
    This expands the node list, if any.
6532

6533
    """
6534
    self.needed_locks = {}
6535
    if self.op.on_nodes:
6536
      # _GetWantedNodes can be used here, but is not always appropriate to use
6537
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6538
      # more information.
6539
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6540
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6541

    
6542
  def CheckPrereq(self):
6543
    """Check prerequisites.
6544

6545
    """
6546

    
6547
  def Exec(self, feedback_fn):
6548
    """Do the actual sleep.
6549

6550
    """
6551
    if self.op.on_master:
6552
      if not utils.TestDelay(self.op.duration):
6553
        raise errors.OpExecError("Error during master delay test")
6554
    if self.op.on_nodes:
6555
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6556
      if not result:
6557
        raise errors.OpExecError("Complete failure from rpc call")
6558
      for node, node_result in result.items():
6559
        node_result.Raise()
6560
        if not node_result.data:
6561
          raise errors.OpExecError("Failure during rpc call to node %s,"
6562
                                   " result: %s" % (node, node_result.data))
6563

    
6564

    
6565
class IAllocator(object):
6566
  """IAllocator framework.
6567

6568
  An IAllocator instance has three sets of attributes:
6569
    - cfg that is needed to query the cluster
6570
    - input data (all members of the _KEYS class attribute are required)
6571
    - four buffer attributes (in|out_data|text), that represent the
6572
      input (to the external script) in text and data structure format,
6573
      and the output from it, again in two formats
6574
    - the result variables from the script (success, info, nodes) for
6575
      easy usage
6576

6577
  """
6578
  _ALLO_KEYS = [
6579
    "mem_size", "disks", "disk_template",
6580
    "os", "tags", "nics", "vcpus", "hypervisor",
6581
    ]
6582
  _RELO_KEYS = [
6583
    "relocate_from",
6584
    ]
6585

    
6586
  def __init__(self, lu, mode, name, **kwargs):
6587
    self.lu = lu
6588
    # init buffer variables
6589
    self.in_text = self.out_text = self.in_data = self.out_data = None
6590
    # init all input fields so that pylint is happy
6591
    self.mode = mode
6592
    self.name = name
6593
    self.mem_size = self.disks = self.disk_template = None
6594
    self.os = self.tags = self.nics = self.vcpus = None
6595
    self.hypervisor = None
6596
    self.relocate_from = None
6597
    # computed fields
6598
    self.required_nodes = None
6599
    # init result fields
6600
    self.success = self.info = self.nodes = None
6601
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6602
      keyset = self._ALLO_KEYS
6603
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6604
      keyset = self._RELO_KEYS
6605
    else:
6606
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6607
                                   " IAllocator" % self.mode)
6608
    for key in kwargs:
6609
      if key not in keyset:
6610
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6611
                                     " IAllocator" % key)
6612
      setattr(self, key, kwargs[key])
6613
    for key in keyset:
6614
      if key not in kwargs:
6615
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6616
                                     " IAllocator" % key)
6617
    self._BuildInputData()
6618

    
6619
  def _ComputeClusterData(self):
6620
    """Compute the generic allocator input data.
6621

6622
    This is the data that is independent of the actual operation.
6623

6624
    """
6625
    cfg = self.lu.cfg
6626
    cluster_info = cfg.GetClusterInfo()
6627
    # cluster data
6628
    data = {
6629
      "version": constants.IALLOCATOR_VERSION,
6630
      "cluster_name": cfg.GetClusterName(),
6631
      "cluster_tags": list(cluster_info.GetTags()),
6632
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6633
      # we don't have job IDs
6634
      }
6635
    iinfo = cfg.GetAllInstancesInfo().values()
6636
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6637

    
6638
    # node data
6639
    node_results = {}
6640
    node_list = cfg.GetNodeList()
6641

    
6642
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6643
      hypervisor_name = self.hypervisor
6644
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6645
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6646

    
6647
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6648
                                           hypervisor_name)
6649
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6650
                       cluster_info.enabled_hypervisors)
6651
    for nname, nresult in node_data.items():
6652
      # first fill in static (config-based) values
6653
      ninfo = cfg.GetNodeInfo(nname)
6654
      pnr = {
6655
        "tags": list(ninfo.GetTags()),
6656
        "primary_ip": ninfo.primary_ip,
6657
        "secondary_ip": ninfo.secondary_ip,
6658
        "offline": ninfo.offline,
6659
        "drained": ninfo.drained,
6660
        "master_candidate": ninfo.master_candidate,
6661
        }
6662

    
6663
      if not ninfo.offline:
6664
        nresult.Raise()
6665
        if not isinstance(nresult.data, dict):
6666
          raise errors.OpExecError("Can't get data for node %s" % nname)
6667
        remote_info = nresult.data
6668
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6669
                     'vg_size', 'vg_free', 'cpu_total']:
6670
          if attr not in remote_info:
6671
            raise errors.OpExecError("Node '%s' didn't return attribute"
6672
                                     " '%s'" % (nname, attr))
6673
          try:
6674
            remote_info[attr] = int(remote_info[attr])
6675
          except ValueError, err:
6676
            raise errors.OpExecError("Node '%s' returned invalid value"
6677
                                     " for '%s': %s" % (nname, attr, err))
6678
        # compute memory used by primary instances
6679
        i_p_mem = i_p_up_mem = 0
6680
        for iinfo, beinfo in i_list:
6681
          if iinfo.primary_node == nname:
6682
            i_p_mem += beinfo[constants.BE_MEMORY]
6683
            if iinfo.name not in node_iinfo[nname].data:
6684
              i_used_mem = 0
6685
            else:
6686
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6687
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6688
            remote_info['memory_free'] -= max(0, i_mem_diff)
6689

    
6690
            if iinfo.admin_up:
6691
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6692

    
6693
        # compute memory used by instances
6694
        pnr_dyn = {
6695
          "total_memory": remote_info['memory_total'],
6696
          "reserved_memory": remote_info['memory_dom0'],
6697
          "free_memory": remote_info['memory_free'],
6698
          "total_disk": remote_info['vg_size'],
6699
          "free_disk": remote_info['vg_free'],
6700
          "total_cpus": remote_info['cpu_total'],
6701
          "i_pri_memory": i_p_mem,
6702
          "i_pri_up_memory": i_p_up_mem,
6703
          }
6704
        pnr.update(pnr_dyn)
6705

    
6706
      node_results[nname] = pnr
6707
    data["nodes"] = node_results
6708

    
6709
    # instance data
6710
    instance_data = {}
6711
    for iinfo, beinfo in i_list:
6712
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6713
                  for n in iinfo.nics]
6714
      pir = {
6715
        "tags": list(iinfo.GetTags()),
6716
        "admin_up": iinfo.admin_up,
6717
        "vcpus": beinfo[constants.BE_VCPUS],
6718
        "memory": beinfo[constants.BE_MEMORY],
6719
        "os": iinfo.os,
6720
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6721
        "nics": nic_data,
6722
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6723
        "disk_template": iinfo.disk_template,
6724
        "hypervisor": iinfo.hypervisor,
6725
        }
6726
      instance_data[iinfo.name] = pir
6727

    
6728
    data["instances"] = instance_data
6729

    
6730
    self.in_data = data
6731

    
6732
  def _AddNewInstance(self):
6733
    """Add new instance data to allocator structure.
6734

6735
    This in combination with _AllocatorGetClusterData will create the
6736
    correct structure needed as input for the allocator.
6737

6738
    The checks for the completeness of the opcode must have already been
6739
    done.
6740

6741
    """
6742
    data = self.in_data
6743

    
6744
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6745

    
6746
    if self.disk_template in constants.DTS_NET_MIRROR:
6747
      self.required_nodes = 2
6748
    else:
6749
      self.required_nodes = 1
6750
    request = {
6751
      "type": "allocate",
6752
      "name": self.name,
6753
      "disk_template": self.disk_template,
6754
      "tags": self.tags,
6755
      "os": self.os,
6756
      "vcpus": self.vcpus,
6757
      "memory": self.mem_size,
6758
      "disks": self.disks,
6759
      "disk_space_total": disk_space,
6760
      "nics": self.nics,
6761
      "required_nodes": self.required_nodes,
6762
      }
6763
    data["request"] = request
6764

    
6765
  def _AddRelocateInstance(self):
6766
    """Add relocate instance data to allocator structure.
6767

6768
    This in combination with _IAllocatorGetClusterData will create the
6769
    correct structure needed as input for the allocator.
6770

6771
    The checks for the completeness of the opcode must have already been
6772
    done.
6773

6774
    """
6775
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6776
    if instance is None:
6777
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6778
                                   " IAllocator" % self.name)
6779

    
6780
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6781
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6782

    
6783
    if len(instance.secondary_nodes) != 1:
6784
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6785

    
6786
    self.required_nodes = 1
6787
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6788
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6789

    
6790
    request = {
6791
      "type": "relocate",
6792
      "name": self.name,
6793
      "disk_space_total": disk_space,
6794
      "required_nodes": self.required_nodes,
6795
      "relocate_from": self.relocate_from,
6796
      }
6797
    self.in_data["request"] = request
6798

    
6799
  def _BuildInputData(self):
6800
    """Build input data structures.
6801

6802
    """
6803
    self._ComputeClusterData()
6804

    
6805
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6806
      self._AddNewInstance()
6807
    else:
6808
      self._AddRelocateInstance()
6809

    
6810
    self.in_text = serializer.Dump(self.in_data)
6811

    
6812
  def Run(self, name, validate=True, call_fn=None):
6813
    """Run an instance allocator and return the results.
6814

6815
    """
6816
    if call_fn is None:
6817
      call_fn = self.lu.rpc.call_iallocator_runner
6818
    data = self.in_text
6819

    
6820
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6821
    result.Raise()
6822

    
6823
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6824
      raise errors.OpExecError("Invalid result from master iallocator runner")
6825

    
6826
    rcode, stdout, stderr, fail = result.data
6827

    
6828
    if rcode == constants.IARUN_NOTFOUND:
6829
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6830
    elif rcode == constants.IARUN_FAILURE:
6831
      raise errors.OpExecError("Instance allocator call failed: %s,"
6832
                               " output: %s" % (fail, stdout+stderr))
6833
    self.out_text = stdout
6834
    if validate:
6835
      self._ValidateResult()
6836

    
6837
  def _ValidateResult(self):
6838
    """Process the allocator results.
6839

6840
    This will process and if successful save the result in
6841
    self.out_data and the other parameters.
6842

6843
    """
6844
    try:
6845
      rdict = serializer.Load(self.out_text)
6846
    except Exception, err:
6847
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6848

    
6849
    if not isinstance(rdict, dict):
6850
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6851

    
6852
    for key in "success", "info", "nodes":
6853
      if key not in rdict:
6854
        raise errors.OpExecError("Can't parse iallocator results:"
6855
                                 " missing key '%s'" % key)
6856
      setattr(self, key, rdict[key])
6857

    
6858
    if not isinstance(rdict["nodes"], list):
6859
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6860
                               " is not a list")
6861
    self.out_data = rdict
6862

    
6863

    
6864
class LUTestAllocator(NoHooksLU):
6865
  """Run allocator tests.
6866

6867
  This LU runs the allocator tests
6868

6869
  """
6870
  _OP_REQP = ["direction", "mode", "name"]
6871

    
6872
  def CheckPrereq(self):
6873
    """Check prerequisites.
6874

6875
    This checks the opcode parameters depending on the director and mode test.
6876

6877
    """
6878
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6879
      for attr in ["name", "mem_size", "disks", "disk_template",
6880
                   "os", "tags", "nics", "vcpus"]:
6881
        if not hasattr(self.op, attr):
6882
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
6883
                                     attr)
6884
      iname = self.cfg.ExpandInstanceName(self.op.name)
6885
      if iname is not None:
6886
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
6887
                                   iname)
6888
      if not isinstance(self.op.nics, list):
6889
        raise errors.OpPrereqError("Invalid parameter 'nics'")
6890
      for row in self.op.nics:
6891
        if (not isinstance(row, dict) or
6892
            "mac" not in row or
6893
            "ip" not in row or
6894
            "bridge" not in row):
6895
          raise errors.OpPrereqError("Invalid contents of the"
6896
                                     " 'nics' parameter")
6897
      if not isinstance(self.op.disks, list):
6898
        raise errors.OpPrereqError("Invalid parameter 'disks'")
6899
      for row in self.op.disks:
6900
        if (not isinstance(row, dict) or
6901
            "size" not in row or
6902
            not isinstance(row["size"], int) or
6903
            "mode" not in row or
6904
            row["mode"] not in ['r', 'w']):
6905
          raise errors.OpPrereqError("Invalid contents of the"
6906
                                     " 'disks' parameter")
6907
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
6908
        self.op.hypervisor = self.cfg.GetHypervisorType()
6909
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
6910
      if not hasattr(self.op, "name"):
6911
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
6912
      fname = self.cfg.ExpandInstanceName(self.op.name)
6913
      if fname is None:
6914
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
6915
                                   self.op.name)
6916
      self.op.name = fname
6917
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
6918
    else:
6919
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
6920
                                 self.op.mode)
6921

    
6922
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
6923
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
6924
        raise errors.OpPrereqError("Missing allocator name")
6925
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
6926
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
6927
                                 self.op.direction)
6928

    
6929
  def Exec(self, feedback_fn):
6930
    """Run the allocator test.
6931

6932
    """
6933
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6934
      ial = IAllocator(self,
6935
                       mode=self.op.mode,
6936
                       name=self.op.name,
6937
                       mem_size=self.op.mem_size,
6938
                       disks=self.op.disks,
6939
                       disk_template=self.op.disk_template,
6940
                       os=self.op.os,
6941
                       tags=self.op.tags,
6942
                       nics=self.op.nics,
6943
                       vcpus=self.op.vcpus,
6944
                       hypervisor=self.op.hypervisor,
6945
                       )
6946
    else:
6947
      ial = IAllocator(self,
6948
                       mode=self.op.mode,
6949
                       name=self.op.name,
6950
                       relocate_from=list(self.relocate_from),
6951
                       )
6952

    
6953
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
6954
      result = ial.in_text
6955
    else:
6956
      ial.Run(self.op.allocator, validate=False)
6957
      result = ial.out_text
6958
    return result