Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ b2482333

History | View | Annotate | Download (241.8 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
    _CheckNodeFreeMemory(self, instance.primary_node,
2771
                         "starting instance %s" % instance.name,
2772
                         bep[constants.BE_MEMORY], instance.hypervisor)
2773

    
2774
  def Exec(self, feedback_fn):
2775
    """Start the instance.
2776

2777
    """
2778
    instance = self.instance
2779
    force = self.op.force
2780

    
2781
    self.cfg.MarkInstanceUp(instance.name)
2782

    
2783
    node_current = instance.primary_node
2784

    
2785
    _StartInstanceDisks(self, instance, force)
2786

    
2787
    result = self.rpc.call_instance_start(node_current, instance)
2788
    msg = result.RemoteFailMsg()
2789
    if msg:
2790
      _ShutdownInstanceDisks(self, instance)
2791
      raise errors.OpExecError("Could not start instance: %s" % msg)
2792

    
2793

    
2794
class LURebootInstance(LogicalUnit):
2795
  """Reboot an instance.
2796

2797
  """
2798
  HPATH = "instance-reboot"
2799
  HTYPE = constants.HTYPE_INSTANCE
2800
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2801
  REQ_BGL = False
2802

    
2803
  def ExpandNames(self):
2804
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2805
                                   constants.INSTANCE_REBOOT_HARD,
2806
                                   constants.INSTANCE_REBOOT_FULL]:
2807
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2808
                                  (constants.INSTANCE_REBOOT_SOFT,
2809
                                   constants.INSTANCE_REBOOT_HARD,
2810
                                   constants.INSTANCE_REBOOT_FULL))
2811
    self._ExpandAndLockInstance()
2812

    
2813
  def BuildHooksEnv(self):
2814
    """Build hooks env.
2815

2816
    This runs on master, primary and secondary nodes of the instance.
2817

2818
    """
2819
    env = {
2820
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2821
      "REBOOT_TYPE": self.op.reboot_type,
2822
      }
2823
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2824
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2825
    return env, nl, nl
2826

    
2827
  def CheckPrereq(self):
2828
    """Check prerequisites.
2829

2830
    This checks that the instance is in the cluster.
2831

2832
    """
2833
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2834
    assert self.instance is not None, \
2835
      "Cannot retrieve locked instance %s" % self.op.instance_name
2836

    
2837
    _CheckNodeOnline(self, instance.primary_node)
2838

    
2839
    # check bridges existance
2840
    _CheckInstanceBridgesExist(self, instance)
2841

    
2842
  def Exec(self, feedback_fn):
2843
    """Reboot the instance.
2844

2845
    """
2846
    instance = self.instance
2847
    ignore_secondaries = self.op.ignore_secondaries
2848
    reboot_type = self.op.reboot_type
2849

    
2850
    node_current = instance.primary_node
2851

    
2852
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2853
                       constants.INSTANCE_REBOOT_HARD]:
2854
      for disk in instance.disks:
2855
        self.cfg.SetDiskID(disk, node_current)
2856
      result = self.rpc.call_instance_reboot(node_current, instance,
2857
                                             reboot_type)
2858
      msg = result.RemoteFailMsg()
2859
      if msg:
2860
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2861
    else:
2862
      result = self.rpc.call_instance_shutdown(node_current, instance)
2863
      msg = result.RemoteFailMsg()
2864
      if msg:
2865
        raise errors.OpExecError("Could not shutdown instance for"
2866
                                 " full reboot: %s" % msg)
2867
      _ShutdownInstanceDisks(self, instance)
2868
      _StartInstanceDisks(self, instance, ignore_secondaries)
2869
      result = self.rpc.call_instance_start(node_current, instance)
2870
      msg = result.RemoteFailMsg()
2871
      if msg:
2872
        _ShutdownInstanceDisks(self, instance)
2873
        raise errors.OpExecError("Could not start instance for"
2874
                                 " full reboot: %s" % msg)
2875

    
2876
    self.cfg.MarkInstanceUp(instance.name)
2877

    
2878

    
2879
class LUShutdownInstance(LogicalUnit):
2880
  """Shutdown an instance.
2881

2882
  """
2883
  HPATH = "instance-stop"
2884
  HTYPE = constants.HTYPE_INSTANCE
2885
  _OP_REQP = ["instance_name"]
2886
  REQ_BGL = False
2887

    
2888
  def ExpandNames(self):
2889
    self._ExpandAndLockInstance()
2890

    
2891
  def BuildHooksEnv(self):
2892
    """Build hooks env.
2893

2894
    This runs on master, primary and secondary nodes of the instance.
2895

2896
    """
2897
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2898
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2899
    return env, nl, nl
2900

    
2901
  def CheckPrereq(self):
2902
    """Check prerequisites.
2903

2904
    This checks that the instance is in the cluster.
2905

2906
    """
2907
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2908
    assert self.instance is not None, \
2909
      "Cannot retrieve locked instance %s" % self.op.instance_name
2910
    _CheckNodeOnline(self, self.instance.primary_node)
2911

    
2912
  def Exec(self, feedback_fn):
2913
    """Shutdown the instance.
2914

2915
    """
2916
    instance = self.instance
2917
    node_current = instance.primary_node
2918
    self.cfg.MarkInstanceDown(instance.name)
2919
    result = self.rpc.call_instance_shutdown(node_current, instance)
2920
    msg = result.RemoteFailMsg()
2921
    if msg:
2922
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
2923

    
2924
    _ShutdownInstanceDisks(self, instance)
2925

    
2926

    
2927
class LUReinstallInstance(LogicalUnit):
2928
  """Reinstall an instance.
2929

2930
  """
2931
  HPATH = "instance-reinstall"
2932
  HTYPE = constants.HTYPE_INSTANCE
2933
  _OP_REQP = ["instance_name"]
2934
  REQ_BGL = False
2935

    
2936
  def ExpandNames(self):
2937
    self._ExpandAndLockInstance()
2938

    
2939
  def BuildHooksEnv(self):
2940
    """Build hooks env.
2941

2942
    This runs on master, primary and secondary nodes of the instance.
2943

2944
    """
2945
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2946
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2947
    return env, nl, nl
2948

    
2949
  def CheckPrereq(self):
2950
    """Check prerequisites.
2951

2952
    This checks that the instance is in the cluster and is not running.
2953

2954
    """
2955
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2956
    assert instance is not None, \
2957
      "Cannot retrieve locked instance %s" % self.op.instance_name
2958
    _CheckNodeOnline(self, instance.primary_node)
2959

    
2960
    if instance.disk_template == constants.DT_DISKLESS:
2961
      raise errors.OpPrereqError("Instance '%s' has no disks" %
2962
                                 self.op.instance_name)
2963
    if instance.admin_up:
2964
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
2965
                                 self.op.instance_name)
2966
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2967
                                              instance.name,
2968
                                              instance.hypervisor)
2969
    if remote_info.failed or remote_info.data:
2970
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
2971
                                 (self.op.instance_name,
2972
                                  instance.primary_node))
2973

    
2974
    self.op.os_type = getattr(self.op, "os_type", None)
2975
    if self.op.os_type is not None:
2976
      # OS verification
2977
      pnode = self.cfg.GetNodeInfo(
2978
        self.cfg.ExpandNodeName(instance.primary_node))
2979
      if pnode is None:
2980
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
2981
                                   self.op.pnode)
2982
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
2983
      result.Raise()
2984
      if not isinstance(result.data, objects.OS):
2985
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
2986
                                   " primary node"  % self.op.os_type)
2987

    
2988
    self.instance = instance
2989

    
2990
  def Exec(self, feedback_fn):
2991
    """Reinstall the instance.
2992

2993
    """
2994
    inst = self.instance
2995

    
2996
    if self.op.os_type is not None:
2997
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
2998
      inst.os = self.op.os_type
2999
      self.cfg.Update(inst)
3000

    
3001
    _StartInstanceDisks(self, inst, None)
3002
    try:
3003
      feedback_fn("Running the instance OS create scripts...")
3004
      result = self.rpc.call_instance_os_add(inst.primary_node, inst)
3005
      msg = result.RemoteFailMsg()
3006
      if msg:
3007
        raise errors.OpExecError("Could not install OS for instance %s"
3008
                                 " on node %s: %s" %
3009
                                 (inst.name, inst.primary_node, msg))
3010
    finally:
3011
      _ShutdownInstanceDisks(self, inst)
3012

    
3013

    
3014
class LURenameInstance(LogicalUnit):
3015
  """Rename an instance.
3016

3017
  """
3018
  HPATH = "instance-rename"
3019
  HTYPE = constants.HTYPE_INSTANCE
3020
  _OP_REQP = ["instance_name", "new_name"]
3021

    
3022
  def BuildHooksEnv(self):
3023
    """Build hooks env.
3024

3025
    This runs on master, primary and secondary nodes of the instance.
3026

3027
    """
3028
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3029
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3030
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3031
    return env, nl, nl
3032

    
3033
  def CheckPrereq(self):
3034
    """Check prerequisites.
3035

3036
    This checks that the instance is in the cluster and is not running.
3037

3038
    """
3039
    instance = self.cfg.GetInstanceInfo(
3040
      self.cfg.ExpandInstanceName(self.op.instance_name))
3041
    if instance is None:
3042
      raise errors.OpPrereqError("Instance '%s' not known" %
3043
                                 self.op.instance_name)
3044
    _CheckNodeOnline(self, instance.primary_node)
3045

    
3046
    if instance.admin_up:
3047
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3048
                                 self.op.instance_name)
3049
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3050
                                              instance.name,
3051
                                              instance.hypervisor)
3052
    remote_info.Raise()
3053
    if remote_info.data:
3054
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3055
                                 (self.op.instance_name,
3056
                                  instance.primary_node))
3057
    self.instance = instance
3058

    
3059
    # new name verification
3060
    name_info = utils.HostInfo(self.op.new_name)
3061

    
3062
    self.op.new_name = new_name = name_info.name
3063
    instance_list = self.cfg.GetInstanceList()
3064
    if new_name in instance_list:
3065
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3066
                                 new_name)
3067

    
3068
    if not getattr(self.op, "ignore_ip", False):
3069
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3070
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3071
                                   (name_info.ip, new_name))
3072

    
3073

    
3074
  def Exec(self, feedback_fn):
3075
    """Reinstall the instance.
3076

3077
    """
3078
    inst = self.instance
3079
    old_name = inst.name
3080

    
3081
    if inst.disk_template == constants.DT_FILE:
3082
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3083

    
3084
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3085
    # Change the instance lock. This is definitely safe while we hold the BGL
3086
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3087
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3088

    
3089
    # re-read the instance from the configuration after rename
3090
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3091

    
3092
    if inst.disk_template == constants.DT_FILE:
3093
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3094
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3095
                                                     old_file_storage_dir,
3096
                                                     new_file_storage_dir)
3097
      result.Raise()
3098
      if not result.data:
3099
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3100
                                 " directory '%s' to '%s' (but the instance"
3101
                                 " has been renamed in Ganeti)" % (
3102
                                 inst.primary_node, old_file_storage_dir,
3103
                                 new_file_storage_dir))
3104

    
3105
      if not result.data[0]:
3106
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3107
                                 " (but the instance has been renamed in"
3108
                                 " Ganeti)" % (old_file_storage_dir,
3109
                                               new_file_storage_dir))
3110

    
3111
    _StartInstanceDisks(self, inst, None)
3112
    try:
3113
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3114
                                                 old_name)
3115
      msg = result.RemoteFailMsg()
3116
      if msg:
3117
        msg = ("Could not run OS rename script for instance %s on node %s"
3118
               " (but the instance has been renamed in Ganeti): %s" %
3119
               (inst.name, inst.primary_node, msg))
3120
        self.proc.LogWarning(msg)
3121
    finally:
3122
      _ShutdownInstanceDisks(self, inst)
3123

    
3124

    
3125
class LURemoveInstance(LogicalUnit):
3126
  """Remove an instance.
3127

3128
  """
3129
  HPATH = "instance-remove"
3130
  HTYPE = constants.HTYPE_INSTANCE
3131
  _OP_REQP = ["instance_name", "ignore_failures"]
3132
  REQ_BGL = False
3133

    
3134
  def ExpandNames(self):
3135
    self._ExpandAndLockInstance()
3136
    self.needed_locks[locking.LEVEL_NODE] = []
3137
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3138

    
3139
  def DeclareLocks(self, level):
3140
    if level == locking.LEVEL_NODE:
3141
      self._LockInstancesNodes()
3142

    
3143
  def BuildHooksEnv(self):
3144
    """Build hooks env.
3145

3146
    This runs on master, primary and secondary nodes of the instance.
3147

3148
    """
3149
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3150
    nl = [self.cfg.GetMasterNode()]
3151
    return env, nl, nl
3152

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

3156
    This checks that the instance is in the cluster.
3157

3158
    """
3159
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3160
    assert self.instance is not None, \
3161
      "Cannot retrieve locked instance %s" % self.op.instance_name
3162

    
3163
  def Exec(self, feedback_fn):
3164
    """Remove the instance.
3165

3166
    """
3167
    instance = self.instance
3168
    logging.info("Shutting down instance %s on node %s",
3169
                 instance.name, instance.primary_node)
3170

    
3171
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3172
    msg = result.RemoteFailMsg()
3173
    if msg:
3174
      if self.op.ignore_failures:
3175
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3176
      else:
3177
        raise errors.OpExecError("Could not shutdown instance %s on"
3178
                                 " node %s: %s" %
3179
                                 (instance.name, instance.primary_node, msg))
3180

    
3181
    logging.info("Removing block devices for instance %s", instance.name)
3182

    
3183
    if not _RemoveDisks(self, instance):
3184
      if self.op.ignore_failures:
3185
        feedback_fn("Warning: can't remove instance's disks")
3186
      else:
3187
        raise errors.OpExecError("Can't remove instance's disks")
3188

    
3189
    logging.info("Removing instance %s out of cluster config", instance.name)
3190

    
3191
    self.cfg.RemoveInstance(instance.name)
3192
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3193

    
3194

    
3195
class LUQueryInstances(NoHooksLU):
3196
  """Logical unit for querying instances.
3197

3198
  """
3199
  _OP_REQP = ["output_fields", "names", "use_locking"]
3200
  REQ_BGL = False
3201
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3202
                                    "admin_state",
3203
                                    "disk_template", "ip", "mac", "bridge",
3204
                                    "sda_size", "sdb_size", "vcpus", "tags",
3205
                                    "network_port", "beparams",
3206
                                    r"(disk)\.(size)/([0-9]+)",
3207
                                    r"(disk)\.(sizes)", "disk_usage",
3208
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3209
                                    r"(nic)\.(macs|ips|bridges)",
3210
                                    r"(disk|nic)\.(count)",
3211
                                    "serial_no", "hypervisor", "hvparams",] +
3212
                                  ["hv/%s" % name
3213
                                   for name in constants.HVS_PARAMETERS] +
3214
                                  ["be/%s" % name
3215
                                   for name in constants.BES_PARAMETERS])
3216
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3217

    
3218

    
3219
  def ExpandNames(self):
3220
    _CheckOutputFields(static=self._FIELDS_STATIC,
3221
                       dynamic=self._FIELDS_DYNAMIC,
3222
                       selected=self.op.output_fields)
3223

    
3224
    self.needed_locks = {}
3225
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3226
    self.share_locks[locking.LEVEL_NODE] = 1
3227

    
3228
    if self.op.names:
3229
      self.wanted = _GetWantedInstances(self, self.op.names)
3230
    else:
3231
      self.wanted = locking.ALL_SET
3232

    
3233
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3234
    self.do_locking = self.do_node_query and self.op.use_locking
3235
    if self.do_locking:
3236
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3237
      self.needed_locks[locking.LEVEL_NODE] = []
3238
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3239

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

    
3244
  def CheckPrereq(self):
3245
    """Check prerequisites.
3246

3247
    """
3248
    pass
3249

    
3250
  def Exec(self, feedback_fn):
3251
    """Computes the list of nodes and their attributes.
3252

3253
    """
3254
    all_info = self.cfg.GetAllInstancesInfo()
3255
    if self.wanted == locking.ALL_SET:
3256
      # caller didn't specify instance names, so ordering is not important
3257
      if self.do_locking:
3258
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3259
      else:
3260
        instance_names = all_info.keys()
3261
      instance_names = utils.NiceSort(instance_names)
3262
    else:
3263
      # caller did specify names, so we must keep the ordering
3264
      if self.do_locking:
3265
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3266
      else:
3267
        tgt_set = all_info.keys()
3268
      missing = set(self.wanted).difference(tgt_set)
3269
      if missing:
3270
        raise errors.OpExecError("Some instances were removed before"
3271
                                 " retrieving their data: %s" % missing)
3272
      instance_names = self.wanted
3273

    
3274
    instance_list = [all_info[iname] for iname in instance_names]
3275

    
3276
    # begin data gathering
3277

    
3278
    nodes = frozenset([inst.primary_node for inst in instance_list])
3279
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3280

    
3281
    bad_nodes = []
3282
    off_nodes = []
3283
    if self.do_node_query:
3284
      live_data = {}
3285
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3286
      for name in nodes:
3287
        result = node_data[name]
3288
        if result.offline:
3289
          # offline nodes will be in both lists
3290
          off_nodes.append(name)
3291
        if result.failed:
3292
          bad_nodes.append(name)
3293
        else:
3294
          if result.data:
3295
            live_data.update(result.data)
3296
            # else no instance is alive
3297
    else:
3298
      live_data = dict([(name, {}) for name in instance_names])
3299

    
3300
    # end data gathering
3301

    
3302
    HVPREFIX = "hv/"
3303
    BEPREFIX = "be/"
3304
    output = []
3305
    for instance in instance_list:
3306
      iout = []
3307
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3308
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3309
      for field in self.op.output_fields:
3310
        st_match = self._FIELDS_STATIC.Matches(field)
3311
        if field == "name":
3312
          val = instance.name
3313
        elif field == "os":
3314
          val = instance.os
3315
        elif field == "pnode":
3316
          val = instance.primary_node
3317
        elif field == "snodes":
3318
          val = list(instance.secondary_nodes)
3319
        elif field == "admin_state":
3320
          val = instance.admin_up
3321
        elif field == "oper_state":
3322
          if instance.primary_node in bad_nodes:
3323
            val = None
3324
          else:
3325
            val = bool(live_data.get(instance.name))
3326
        elif field == "status":
3327
          if instance.primary_node in off_nodes:
3328
            val = "ERROR_nodeoffline"
3329
          elif instance.primary_node in bad_nodes:
3330
            val = "ERROR_nodedown"
3331
          else:
3332
            running = bool(live_data.get(instance.name))
3333
            if running:
3334
              if instance.admin_up:
3335
                val = "running"
3336
              else:
3337
                val = "ERROR_up"
3338
            else:
3339
              if instance.admin_up:
3340
                val = "ERROR_down"
3341
              else:
3342
                val = "ADMIN_down"
3343
        elif field == "oper_ram":
3344
          if instance.primary_node in bad_nodes:
3345
            val = None
3346
          elif instance.name in live_data:
3347
            val = live_data[instance.name].get("memory", "?")
3348
          else:
3349
            val = "-"
3350
        elif field == "disk_template":
3351
          val = instance.disk_template
3352
        elif field == "ip":
3353
          val = instance.nics[0].ip
3354
        elif field == "bridge":
3355
          val = instance.nics[0].bridge
3356
        elif field == "mac":
3357
          val = instance.nics[0].mac
3358
        elif field == "sda_size" or field == "sdb_size":
3359
          idx = ord(field[2]) - ord('a')
3360
          try:
3361
            val = instance.FindDisk(idx).size
3362
          except errors.OpPrereqError:
3363
            val = None
3364
        elif field == "disk_usage": # total disk usage per node
3365
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3366
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3367
        elif field == "tags":
3368
          val = list(instance.GetTags())
3369
        elif field == "serial_no":
3370
          val = instance.serial_no
3371
        elif field == "network_port":
3372
          val = instance.network_port
3373
        elif field == "hypervisor":
3374
          val = instance.hypervisor
3375
        elif field == "hvparams":
3376
          val = i_hv
3377
        elif (field.startswith(HVPREFIX) and
3378
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3379
          val = i_hv.get(field[len(HVPREFIX):], None)
3380
        elif field == "beparams":
3381
          val = i_be
3382
        elif (field.startswith(BEPREFIX) and
3383
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3384
          val = i_be.get(field[len(BEPREFIX):], None)
3385
        elif st_match and st_match.groups():
3386
          # matches a variable list
3387
          st_groups = st_match.groups()
3388
          if st_groups and st_groups[0] == "disk":
3389
            if st_groups[1] == "count":
3390
              val = len(instance.disks)
3391
            elif st_groups[1] == "sizes":
3392
              val = [disk.size for disk in instance.disks]
3393
            elif st_groups[1] == "size":
3394
              try:
3395
                val = instance.FindDisk(st_groups[2]).size
3396
              except errors.OpPrereqError:
3397
                val = None
3398
            else:
3399
              assert False, "Unhandled disk parameter"
3400
          elif st_groups[0] == "nic":
3401
            if st_groups[1] == "count":
3402
              val = len(instance.nics)
3403
            elif st_groups[1] == "macs":
3404
              val = [nic.mac for nic in instance.nics]
3405
            elif st_groups[1] == "ips":
3406
              val = [nic.ip for nic in instance.nics]
3407
            elif st_groups[1] == "bridges":
3408
              val = [nic.bridge for nic in instance.nics]
3409
            else:
3410
              # index-based item
3411
              nic_idx = int(st_groups[2])
3412
              if nic_idx >= len(instance.nics):
3413
                val = None
3414
              else:
3415
                if st_groups[1] == "mac":
3416
                  val = instance.nics[nic_idx].mac
3417
                elif st_groups[1] == "ip":
3418
                  val = instance.nics[nic_idx].ip
3419
                elif st_groups[1] == "bridge":
3420
                  val = instance.nics[nic_idx].bridge
3421
                else:
3422
                  assert False, "Unhandled NIC parameter"
3423
          else:
3424
            assert False, "Unhandled variable parameter"
3425
        else:
3426
          raise errors.ParameterError(field)
3427
        iout.append(val)
3428
      output.append(iout)
3429

    
3430
    return output
3431

    
3432

    
3433
class LUFailoverInstance(LogicalUnit):
3434
  """Failover an instance.
3435

3436
  """
3437
  HPATH = "instance-failover"
3438
  HTYPE = constants.HTYPE_INSTANCE
3439
  _OP_REQP = ["instance_name", "ignore_consistency"]
3440
  REQ_BGL = False
3441

    
3442
  def ExpandNames(self):
3443
    self._ExpandAndLockInstance()
3444
    self.needed_locks[locking.LEVEL_NODE] = []
3445
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3446

    
3447
  def DeclareLocks(self, level):
3448
    if level == locking.LEVEL_NODE:
3449
      self._LockInstancesNodes()
3450

    
3451
  def BuildHooksEnv(self):
3452
    """Build hooks env.
3453

3454
    This runs on master, primary and secondary nodes of the instance.
3455

3456
    """
3457
    env = {
3458
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3459
      }
3460
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3461
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3462
    return env, nl, nl
3463

    
3464
  def CheckPrereq(self):
3465
    """Check prerequisites.
3466

3467
    This checks that the instance is in the cluster.
3468

3469
    """
3470
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3471
    assert self.instance is not None, \
3472
      "Cannot retrieve locked instance %s" % self.op.instance_name
3473

    
3474
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3475
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3476
      raise errors.OpPrereqError("Instance's disk layout is not"
3477
                                 " network mirrored, cannot failover.")
3478

    
3479
    secondary_nodes = instance.secondary_nodes
3480
    if not secondary_nodes:
3481
      raise errors.ProgrammerError("no secondary node but using "
3482
                                   "a mirrored disk template")
3483

    
3484
    target_node = secondary_nodes[0]
3485
    _CheckNodeOnline(self, target_node)
3486
    _CheckNodeNotDrained(self, target_node)
3487
    # check memory requirements on the secondary node
3488
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3489
                         instance.name, bep[constants.BE_MEMORY],
3490
                         instance.hypervisor)
3491

    
3492
    # check bridge existance
3493
    brlist = [nic.bridge for nic in instance.nics]
3494
    result = self.rpc.call_bridges_exist(target_node, brlist)
3495
    result.Raise()
3496
    if not result.data:
3497
      raise errors.OpPrereqError("One or more target bridges %s does not"
3498
                                 " exist on destination node '%s'" %
3499
                                 (brlist, target_node))
3500

    
3501
  def Exec(self, feedback_fn):
3502
    """Failover an instance.
3503

3504
    The failover is done by shutting it down on its present node and
3505
    starting it on the secondary.
3506

3507
    """
3508
    instance = self.instance
3509

    
3510
    source_node = instance.primary_node
3511
    target_node = instance.secondary_nodes[0]
3512

    
3513
    feedback_fn("* checking disk consistency between source and target")
3514
    for dev in instance.disks:
3515
      # for drbd, these are drbd over lvm
3516
      if not _CheckDiskConsistency(self, dev, target_node, False):
3517
        if instance.admin_up and not self.op.ignore_consistency:
3518
          raise errors.OpExecError("Disk %s is degraded on target node,"
3519
                                   " aborting failover." % dev.iv_name)
3520

    
3521
    feedback_fn("* shutting down instance on source node")
3522
    logging.info("Shutting down instance %s on node %s",
3523
                 instance.name, source_node)
3524

    
3525
    result = self.rpc.call_instance_shutdown(source_node, instance)
3526
    msg = result.RemoteFailMsg()
3527
    if msg:
3528
      if self.op.ignore_consistency:
3529
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3530
                             " Proceeding anyway. Please make sure node"
3531
                             " %s is down. Error details: %s",
3532
                             instance.name, source_node, source_node, msg)
3533
      else:
3534
        raise errors.OpExecError("Could not shutdown instance %s on"
3535
                                 " node %s: %s" %
3536
                                 (instance.name, source_node, msg))
3537

    
3538
    feedback_fn("* deactivating the instance's disks on source node")
3539
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3540
      raise errors.OpExecError("Can't shut down the instance's disks.")
3541

    
3542
    instance.primary_node = target_node
3543
    # distribute new instance config to the other nodes
3544
    self.cfg.Update(instance)
3545

    
3546
    # Only start the instance if it's marked as up
3547
    if instance.admin_up:
3548
      feedback_fn("* activating the instance's disks on target node")
3549
      logging.info("Starting instance %s on node %s",
3550
                   instance.name, target_node)
3551

    
3552
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3553
                                               ignore_secondaries=True)
3554
      if not disks_ok:
3555
        _ShutdownInstanceDisks(self, instance)
3556
        raise errors.OpExecError("Can't activate the instance's disks")
3557

    
3558
      feedback_fn("* starting the instance on the target node")
3559
      result = self.rpc.call_instance_start(target_node, instance)
3560
      msg = result.RemoteFailMsg()
3561
      if msg:
3562
        _ShutdownInstanceDisks(self, instance)
3563
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3564
                                 (instance.name, target_node, msg))
3565

    
3566

    
3567
class LUMigrateInstance(LogicalUnit):
3568
  """Migrate an instance.
3569

3570
  This is migration without shutting down, compared to the failover,
3571
  which is done with shutdown.
3572

3573
  """
3574
  HPATH = "instance-migrate"
3575
  HTYPE = constants.HTYPE_INSTANCE
3576
  _OP_REQP = ["instance_name", "live", "cleanup"]
3577

    
3578
  REQ_BGL = False
3579

    
3580
  def ExpandNames(self):
3581
    self._ExpandAndLockInstance()
3582
    self.needed_locks[locking.LEVEL_NODE] = []
3583
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3584

    
3585
  def DeclareLocks(self, level):
3586
    if level == locking.LEVEL_NODE:
3587
      self._LockInstancesNodes()
3588

    
3589
  def BuildHooksEnv(self):
3590
    """Build hooks env.
3591

3592
    This runs on master, primary and secondary nodes of the instance.
3593

3594
    """
3595
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3596
    env["MIGRATE_LIVE"] = self.op.live
3597
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3598
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3599
    return env, nl, nl
3600

    
3601
  def CheckPrereq(self):
3602
    """Check prerequisites.
3603

3604
    This checks that the instance is in the cluster.
3605

3606
    """
3607
    instance = self.cfg.GetInstanceInfo(
3608
      self.cfg.ExpandInstanceName(self.op.instance_name))
3609
    if instance is None:
3610
      raise errors.OpPrereqError("Instance '%s' not known" %
3611
                                 self.op.instance_name)
3612

    
3613
    if instance.disk_template != constants.DT_DRBD8:
3614
      raise errors.OpPrereqError("Instance's disk layout is not"
3615
                                 " drbd8, cannot migrate.")
3616

    
3617
    secondary_nodes = instance.secondary_nodes
3618
    if not secondary_nodes:
3619
      raise errors.ConfigurationError("No secondary node but using"
3620
                                      " drbd8 disk template")
3621

    
3622
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3623

    
3624
    target_node = secondary_nodes[0]
3625
    # check memory requirements on the secondary node
3626
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3627
                         instance.name, i_be[constants.BE_MEMORY],
3628
                         instance.hypervisor)
3629

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

    
3638
    if not self.op.cleanup:
3639
      _CheckNodeNotDrained(self, target_node)
3640
      result = self.rpc.call_instance_migratable(instance.primary_node,
3641
                                                 instance)
3642
      msg = result.RemoteFailMsg()
3643
      if msg:
3644
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3645
                                   msg)
3646

    
3647
    self.instance = instance
3648

    
3649
  def _WaitUntilSync(self):
3650
    """Poll with custom rpc for disk sync.
3651

3652
    This uses our own step-based rpc call.
3653

3654
    """
3655
    self.feedback_fn("* wait until resync is done")
3656
    all_done = False
3657
    while not all_done:
3658
      all_done = True
3659
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3660
                                            self.nodes_ip,
3661
                                            self.instance.disks)
3662
      min_percent = 100
3663
      for node, nres in result.items():
3664
        msg = nres.RemoteFailMsg()
3665
        if msg:
3666
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3667
                                   (node, msg))
3668
        node_done, node_percent = nres.payload
3669
        all_done = all_done and node_done
3670
        if node_percent is not None:
3671
          min_percent = min(min_percent, node_percent)
3672
      if not all_done:
3673
        if min_percent < 100:
3674
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3675
        time.sleep(2)
3676

    
3677
  def _EnsureSecondary(self, node):
3678
    """Demote a node to secondary.
3679

3680
    """
3681
    self.feedback_fn("* switching node %s to secondary mode" % node)
3682

    
3683
    for dev in self.instance.disks:
3684
      self.cfg.SetDiskID(dev, node)
3685

    
3686
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3687
                                          self.instance.disks)
3688
    msg = result.RemoteFailMsg()
3689
    if msg:
3690
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3691
                               " error %s" % (node, msg))
3692

    
3693
  def _GoStandalone(self):
3694
    """Disconnect from the network.
3695

3696
    """
3697
    self.feedback_fn("* changing into standalone mode")
3698
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3699
                                               self.instance.disks)
3700
    for node, nres in result.items():
3701
      msg = nres.RemoteFailMsg()
3702
      if msg:
3703
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3704
                                 " error %s" % (node, msg))
3705

    
3706
  def _GoReconnect(self, multimaster):
3707
    """Reconnect to the network.
3708

3709
    """
3710
    if multimaster:
3711
      msg = "dual-master"
3712
    else:
3713
      msg = "single-master"
3714
    self.feedback_fn("* changing disks into %s mode" % msg)
3715
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3716
                                           self.instance.disks,
3717
                                           self.instance.name, multimaster)
3718
    for node, nres in result.items():
3719
      msg = nres.RemoteFailMsg()
3720
      if msg:
3721
        raise errors.OpExecError("Cannot change disks config on node %s,"
3722
                                 " error: %s" % (node, msg))
3723

    
3724
  def _ExecCleanup(self):
3725
    """Try to cleanup after a failed migration.
3726

3727
    The cleanup is done by:
3728
      - check that the instance is running only on one node
3729
        (and update the config if needed)
3730
      - change disks on its secondary node to secondary
3731
      - wait until disks are fully synchronized
3732
      - disconnect from the network
3733
      - change disks into single-master mode
3734
      - wait again until disks are fully synchronized
3735

3736
    """
3737
    instance = self.instance
3738
    target_node = self.target_node
3739
    source_node = self.source_node
3740

    
3741
    # check running on only one node
3742
    self.feedback_fn("* checking where the instance actually runs"
3743
                     " (if this hangs, the hypervisor might be in"
3744
                     " a bad state)")
3745
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3746
    for node, result in ins_l.items():
3747
      result.Raise()
3748
      if not isinstance(result.data, list):
3749
        raise errors.OpExecError("Can't contact node '%s'" % node)
3750

    
3751
    runningon_source = instance.name in ins_l[source_node].data
3752
    runningon_target = instance.name in ins_l[target_node].data
3753

    
3754
    if runningon_source and runningon_target:
3755
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3756
                               " or the hypervisor is confused. You will have"
3757
                               " to ensure manually that it runs only on one"
3758
                               " and restart this operation.")
3759

    
3760
    if not (runningon_source or runningon_target):
3761
      raise errors.OpExecError("Instance does not seem to be running at all."
3762
                               " In this case, it's safer to repair by"
3763
                               " running 'gnt-instance stop' to ensure disk"
3764
                               " shutdown, and then restarting it.")
3765

    
3766
    if runningon_target:
3767
      # the migration has actually succeeded, we need to update the config
3768
      self.feedback_fn("* instance running on secondary node (%s),"
3769
                       " updating config" % target_node)
3770
      instance.primary_node = target_node
3771
      self.cfg.Update(instance)
3772
      demoted_node = source_node
3773
    else:
3774
      self.feedback_fn("* instance confirmed to be running on its"
3775
                       " primary node (%s)" % source_node)
3776
      demoted_node = target_node
3777

    
3778
    self._EnsureSecondary(demoted_node)
3779
    try:
3780
      self._WaitUntilSync()
3781
    except errors.OpExecError:
3782
      # we ignore here errors, since if the device is standalone, it
3783
      # won't be able to sync
3784
      pass
3785
    self._GoStandalone()
3786
    self._GoReconnect(False)
3787
    self._WaitUntilSync()
3788

    
3789
    self.feedback_fn("* done")
3790

    
3791
  def _RevertDiskStatus(self):
3792
    """Try to revert the disk status after a failed migration.
3793

3794
    """
3795
    target_node = self.target_node
3796
    try:
3797
      self._EnsureSecondary(target_node)
3798
      self._GoStandalone()
3799
      self._GoReconnect(False)
3800
      self._WaitUntilSync()
3801
    except errors.OpExecError, err:
3802
      self.LogWarning("Migration failed and I can't reconnect the"
3803
                      " drives: error '%s'\n"
3804
                      "Please look and recover the instance status" %
3805
                      str(err))
3806

    
3807
  def _AbortMigration(self):
3808
    """Call the hypervisor code to abort a started migration.
3809

3810
    """
3811
    instance = self.instance
3812
    target_node = self.target_node
3813
    migration_info = self.migration_info
3814

    
3815
    abort_result = self.rpc.call_finalize_migration(target_node,
3816
                                                    instance,
3817
                                                    migration_info,
3818
                                                    False)
3819
    abort_msg = abort_result.RemoteFailMsg()
3820
    if abort_msg:
3821
      logging.error("Aborting migration failed on target node %s: %s" %
3822
                    (target_node, abort_msg))
3823
      # Don't raise an exception here, as we stil have to try to revert the
3824
      # disk status, even if this step failed.
3825

    
3826
  def _ExecMigration(self):
3827
    """Migrate an instance.
3828

3829
    The migrate is done by:
3830
      - change the disks into dual-master mode
3831
      - wait until disks are fully synchronized again
3832
      - migrate the instance
3833
      - change disks on the new secondary node (the old primary) to secondary
3834
      - wait until disks are fully synchronized
3835
      - change disks into single-master mode
3836

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

    
3842
    self.feedback_fn("* checking disk consistency between source and target")
3843
    for dev in instance.disks:
3844
      if not _CheckDiskConsistency(self, dev, target_node, False):
3845
        raise errors.OpExecError("Disk %s is degraded or not fully"
3846
                                 " synchronized on target node,"
3847
                                 " aborting migrate." % dev.iv_name)
3848

    
3849
    # First get the migration information from the remote node
3850
    result = self.rpc.call_migration_info(source_node, instance)
3851
    msg = result.RemoteFailMsg()
3852
    if msg:
3853
      log_err = ("Failed fetching source migration information from %s: %s" %
3854
                 (source_node, msg))
3855
      logging.error(log_err)
3856
      raise errors.OpExecError(log_err)
3857

    
3858
    self.migration_info = migration_info = result.payload
3859

    
3860
    # Then switch the disks to master/master mode
3861
    self._EnsureSecondary(target_node)
3862
    self._GoStandalone()
3863
    self._GoReconnect(True)
3864
    self._WaitUntilSync()
3865

    
3866
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3867
    result = self.rpc.call_accept_instance(target_node,
3868
                                           instance,
3869
                                           migration_info,
3870
                                           self.nodes_ip[target_node])
3871

    
3872
    msg = result.RemoteFailMsg()
3873
    if msg:
3874
      logging.error("Instance pre-migration failed, trying to revert"
3875
                    " disk status: %s", msg)
3876
      self._AbortMigration()
3877
      self._RevertDiskStatus()
3878
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3879
                               (instance.name, msg))
3880

    
3881
    self.feedback_fn("* migrating instance to %s" % target_node)
3882
    time.sleep(10)
3883
    result = self.rpc.call_instance_migrate(source_node, instance,
3884
                                            self.nodes_ip[target_node],
3885
                                            self.op.live)
3886
    msg = result.RemoteFailMsg()
3887
    if msg:
3888
      logging.error("Instance migration failed, trying to revert"
3889
                    " disk status: %s", msg)
3890
      self._AbortMigration()
3891
      self._RevertDiskStatus()
3892
      raise errors.OpExecError("Could not migrate instance %s: %s" %
3893
                               (instance.name, msg))
3894
    time.sleep(10)
3895

    
3896
    instance.primary_node = target_node
3897
    # distribute new instance config to the other nodes
3898
    self.cfg.Update(instance)
3899

    
3900
    result = self.rpc.call_finalize_migration(target_node,
3901
                                              instance,
3902
                                              migration_info,
3903
                                              True)
3904
    msg = result.RemoteFailMsg()
3905
    if msg:
3906
      logging.error("Instance migration succeeded, but finalization failed:"
3907
                    " %s" % msg)
3908
      raise errors.OpExecError("Could not finalize instance migration: %s" %
3909
                               msg)
3910

    
3911
    self._EnsureSecondary(source_node)
3912
    self._WaitUntilSync()
3913
    self._GoStandalone()
3914
    self._GoReconnect(False)
3915
    self._WaitUntilSync()
3916

    
3917
    self.feedback_fn("* done")
3918

    
3919
  def Exec(self, feedback_fn):
3920
    """Perform the migration.
3921

3922
    """
3923
    self.feedback_fn = feedback_fn
3924

    
3925
    self.source_node = self.instance.primary_node
3926
    self.target_node = self.instance.secondary_nodes[0]
3927
    self.all_nodes = [self.source_node, self.target_node]
3928
    self.nodes_ip = {
3929
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
3930
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
3931
      }
3932
    if self.op.cleanup:
3933
      return self._ExecCleanup()
3934
    else:
3935
      return self._ExecMigration()
3936

    
3937

    
3938
def _CreateBlockDev(lu, node, instance, device, force_create,
3939
                    info, force_open):
3940
  """Create a tree of block devices on a given node.
3941

3942
  If this device type has to be created on secondaries, create it and
3943
  all its children.
3944

3945
  If not, just recurse to children keeping the same 'force' value.
3946

3947
  @param lu: the lu on whose behalf we execute
3948
  @param node: the node on which to create the device
3949
  @type instance: L{objects.Instance}
3950
  @param instance: the instance which owns the device
3951
  @type device: L{objects.Disk}
3952
  @param device: the device to create
3953
  @type force_create: boolean
3954
  @param force_create: whether to force creation of this device; this
3955
      will be change to True whenever we find a device which has
3956
      CreateOnSecondary() attribute
3957
  @param info: the extra 'metadata' we should attach to the device
3958
      (this will be represented as a LVM tag)
3959
  @type force_open: boolean
3960
  @param force_open: this parameter will be passes to the
3961
      L{backend.BlockdevCreate} function where it specifies
3962
      whether we run on primary or not, and it affects both
3963
      the child assembly and the device own Open() execution
3964

3965
  """
3966
  if device.CreateOnSecondary():
3967
    force_create = True
3968

    
3969
  if device.children:
3970
    for child in device.children:
3971
      _CreateBlockDev(lu, node, instance, child, force_create,
3972
                      info, force_open)
3973

    
3974
  if not force_create:
3975
    return
3976

    
3977
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
3978

    
3979

    
3980
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
3981
  """Create a single block device on a given node.
3982

3983
  This will not recurse over children of the device, so they must be
3984
  created in advance.
3985

3986
  @param lu: the lu on whose behalf we execute
3987
  @param node: the node on which to create the device
3988
  @type instance: L{objects.Instance}
3989
  @param instance: the instance which owns the device
3990
  @type device: L{objects.Disk}
3991
  @param device: the device to create
3992
  @param info: the extra 'metadata' we should attach to the device
3993
      (this will be represented as a LVM tag)
3994
  @type force_open: boolean
3995
  @param force_open: this parameter will be passes to the
3996
      L{backend.BlockdevCreate} function where it specifies
3997
      whether we run on primary or not, and it affects both
3998
      the child assembly and the device own Open() execution
3999

4000
  """
4001
  lu.cfg.SetDiskID(device, node)
4002
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4003
                                       instance.name, force_open, info)
4004
  msg = result.RemoteFailMsg()
4005
  if msg:
4006
    raise errors.OpExecError("Can't create block device %s on"
4007
                             " node %s for instance %s: %s" %
4008
                             (device, node, instance.name, msg))
4009
  if device.physical_id is None:
4010
    device.physical_id = result.payload
4011

    
4012

    
4013
def _GenerateUniqueNames(lu, exts):
4014
  """Generate a suitable LV name.
4015

4016
  This will generate a logical volume name for the given instance.
4017

4018
  """
4019
  results = []
4020
  for val in exts:
4021
    new_id = lu.cfg.GenerateUniqueID()
4022
    results.append("%s%s" % (new_id, val))
4023
  return results
4024

    
4025

    
4026
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4027
                         p_minor, s_minor):
4028
  """Generate a drbd8 device complete with its children.
4029

4030
  """
4031
  port = lu.cfg.AllocatePort()
4032
  vgname = lu.cfg.GetVGName()
4033
  shared_secret = lu.cfg.GenerateDRBDSecret()
4034
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4035
                          logical_id=(vgname, names[0]))
4036
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4037
                          logical_id=(vgname, names[1]))
4038
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4039
                          logical_id=(primary, secondary, port,
4040
                                      p_minor, s_minor,
4041
                                      shared_secret),
4042
                          children=[dev_data, dev_meta],
4043
                          iv_name=iv_name)
4044
  return drbd_dev
4045

    
4046

    
4047
def _GenerateDiskTemplate(lu, template_name,
4048
                          instance_name, primary_node,
4049
                          secondary_nodes, disk_info,
4050
                          file_storage_dir, file_driver,
4051
                          base_index):
4052
  """Generate the entire disk layout for a given template type.
4053

4054
  """
4055
  #TODO: compute space requirements
4056

    
4057
  vgname = lu.cfg.GetVGName()
4058
  disk_count = len(disk_info)
4059
  disks = []
4060
  if template_name == constants.DT_DISKLESS:
4061
    pass
4062
  elif template_name == constants.DT_PLAIN:
4063
    if len(secondary_nodes) != 0:
4064
      raise errors.ProgrammerError("Wrong template configuration")
4065

    
4066
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4067
                                      for i in range(disk_count)])
4068
    for idx, disk in enumerate(disk_info):
4069
      disk_index = idx + base_index
4070
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4071
                              logical_id=(vgname, names[idx]),
4072
                              iv_name="disk/%d" % disk_index,
4073
                              mode=disk["mode"])
4074
      disks.append(disk_dev)
4075
  elif template_name == constants.DT_DRBD8:
4076
    if len(secondary_nodes) != 1:
4077
      raise errors.ProgrammerError("Wrong template configuration")
4078
    remote_node = secondary_nodes[0]
4079
    minors = lu.cfg.AllocateDRBDMinor(
4080
      [primary_node, remote_node] * len(disk_info), instance_name)
4081

    
4082
    names = []
4083
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4084
                                               for i in range(disk_count)]):
4085
      names.append(lv_prefix + "_data")
4086
      names.append(lv_prefix + "_meta")
4087
    for idx, disk in enumerate(disk_info):
4088
      disk_index = idx + base_index
4089
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4090
                                      disk["size"], names[idx*2:idx*2+2],
4091
                                      "disk/%d" % disk_index,
4092
                                      minors[idx*2], minors[idx*2+1])
4093
      disk_dev.mode = disk["mode"]
4094
      disks.append(disk_dev)
4095
  elif template_name == constants.DT_FILE:
4096
    if len(secondary_nodes) != 0:
4097
      raise errors.ProgrammerError("Wrong template configuration")
4098

    
4099
    for idx, disk in enumerate(disk_info):
4100
      disk_index = idx + base_index
4101
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4102
                              iv_name="disk/%d" % disk_index,
4103
                              logical_id=(file_driver,
4104
                                          "%s/disk%d" % (file_storage_dir,
4105
                                                         disk_index)),
4106
                              mode=disk["mode"])
4107
      disks.append(disk_dev)
4108
  else:
4109
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4110
  return disks
4111

    
4112

    
4113
def _GetInstanceInfoText(instance):
4114
  """Compute that text that should be added to the disk's metadata.
4115

4116
  """
4117
  return "originstname+%s" % instance.name
4118

    
4119

    
4120
def _CreateDisks(lu, instance):
4121
  """Create all disks for an instance.
4122

4123
  This abstracts away some work from AddInstance.
4124

4125
  @type lu: L{LogicalUnit}
4126
  @param lu: the logical unit on whose behalf we execute
4127
  @type instance: L{objects.Instance}
4128
  @param instance: the instance whose disks we should create
4129
  @rtype: boolean
4130
  @return: the success of the creation
4131

4132
  """
4133
  info = _GetInstanceInfoText(instance)
4134
  pnode = instance.primary_node
4135

    
4136
  if instance.disk_template == constants.DT_FILE:
4137
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4138
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4139

    
4140
    if result.failed or not result.data:
4141
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4142

    
4143
    if not result.data[0]:
4144
      raise errors.OpExecError("Failed to create directory '%s'" %
4145
                               file_storage_dir)
4146

    
4147
  # Note: this needs to be kept in sync with adding of disks in
4148
  # LUSetInstanceParams
4149
  for device in instance.disks:
4150
    logging.info("Creating volume %s for instance %s",
4151
                 device.iv_name, instance.name)
4152
    #HARDCODE
4153
    for node in instance.all_nodes:
4154
      f_create = node == pnode
4155
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4156

    
4157

    
4158
def _RemoveDisks(lu, instance):
4159
  """Remove all disks for an instance.
4160

4161
  This abstracts away some work from `AddInstance()` and
4162
  `RemoveInstance()`. Note that in case some of the devices couldn't
4163
  be removed, the removal will continue with the other ones (compare
4164
  with `_CreateDisks()`).
4165

4166
  @type lu: L{LogicalUnit}
4167
  @param lu: the logical unit on whose behalf we execute
4168
  @type instance: L{objects.Instance}
4169
  @param instance: the instance whose disks we should remove
4170
  @rtype: boolean
4171
  @return: the success of the removal
4172

4173
  """
4174
  logging.info("Removing block devices for instance %s", instance.name)
4175

    
4176
  all_result = True
4177
  for device in instance.disks:
4178
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4179
      lu.cfg.SetDiskID(disk, node)
4180
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4181
      if msg:
4182
        lu.LogWarning("Could not remove block device %s on node %s,"
4183
                      " continuing anyway: %s", device.iv_name, node, msg)
4184
        all_result = False
4185

    
4186
  if instance.disk_template == constants.DT_FILE:
4187
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4188
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4189
                                                 file_storage_dir)
4190
    if result.failed or not result.data:
4191
      logging.error("Could not remove directory '%s'", file_storage_dir)
4192
      all_result = False
4193

    
4194
  return all_result
4195

    
4196

    
4197
def _ComputeDiskSize(disk_template, disks):
4198
  """Compute disk size requirements in the volume group
4199

4200
  """
4201
  # Required free disk space as a function of disk and swap space
4202
  req_size_dict = {
4203
    constants.DT_DISKLESS: None,
4204
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4205
    # 128 MB are added for drbd metadata for each disk
4206
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4207
    constants.DT_FILE: None,
4208
  }
4209

    
4210
  if disk_template not in req_size_dict:
4211
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4212
                                 " is unknown" %  disk_template)
4213

    
4214
  return req_size_dict[disk_template]
4215

    
4216

    
4217
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4218
  """Hypervisor parameter validation.
4219

4220
  This function abstract the hypervisor parameter validation to be
4221
  used in both instance create and instance modify.
4222

4223
  @type lu: L{LogicalUnit}
4224
  @param lu: the logical unit for which we check
4225
  @type nodenames: list
4226
  @param nodenames: the list of nodes on which we should check
4227
  @type hvname: string
4228
  @param hvname: the name of the hypervisor we should use
4229
  @type hvparams: dict
4230
  @param hvparams: the parameters which we need to check
4231
  @raise errors.OpPrereqError: if the parameters are not valid
4232

4233
  """
4234
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4235
                                                  hvname,
4236
                                                  hvparams)
4237
  for node in nodenames:
4238
    info = hvinfo[node]
4239
    if info.offline:
4240
      continue
4241
    msg = info.RemoteFailMsg()
4242
    if msg:
4243
      raise errors.OpPrereqError("Hypervisor parameter validation"
4244
                                 " failed on node %s: %s" % (node, msg))
4245

    
4246

    
4247
class LUCreateInstance(LogicalUnit):
4248
  """Create an instance.
4249

4250
  """
4251
  HPATH = "instance-add"
4252
  HTYPE = constants.HTYPE_INSTANCE
4253
  _OP_REQP = ["instance_name", "disks", "disk_template",
4254
              "mode", "start",
4255
              "wait_for_sync", "ip_check", "nics",
4256
              "hvparams", "beparams"]
4257
  REQ_BGL = False
4258

    
4259
  def _ExpandNode(self, node):
4260
    """Expands and checks one node name.
4261

4262
    """
4263
    node_full = self.cfg.ExpandNodeName(node)
4264
    if node_full is None:
4265
      raise errors.OpPrereqError("Unknown node %s" % node)
4266
    return node_full
4267

    
4268
  def ExpandNames(self):
4269
    """ExpandNames for CreateInstance.
4270

4271
    Figure out the right locks for instance creation.
4272

4273
    """
4274
    self.needed_locks = {}
4275

    
4276
    # set optional parameters to none if they don't exist
4277
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4278
      if not hasattr(self.op, attr):
4279
        setattr(self.op, attr, None)
4280

    
4281
    # cheap checks, mostly valid constants given
4282

    
4283
    # verify creation mode
4284
    if self.op.mode not in (constants.INSTANCE_CREATE,
4285
                            constants.INSTANCE_IMPORT):
4286
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4287
                                 self.op.mode)
4288

    
4289
    # disk template and mirror node verification
4290
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4291
      raise errors.OpPrereqError("Invalid disk template name")
4292

    
4293
    if self.op.hypervisor is None:
4294
      self.op.hypervisor = self.cfg.GetHypervisorType()
4295

    
4296
    cluster = self.cfg.GetClusterInfo()
4297
    enabled_hvs = cluster.enabled_hypervisors
4298
    if self.op.hypervisor not in enabled_hvs:
4299
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4300
                                 " cluster (%s)" % (self.op.hypervisor,
4301
                                  ",".join(enabled_hvs)))
4302

    
4303
    # check hypervisor parameter syntax (locally)
4304
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4305
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4306
                                  self.op.hvparams)
4307
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4308
    hv_type.CheckParameterSyntax(filled_hvp)
4309

    
4310
    # fill and remember the beparams dict
4311
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4312
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4313
                                    self.op.beparams)
4314

    
4315
    #### instance parameters check
4316

    
4317
    # instance name verification
4318
    hostname1 = utils.HostInfo(self.op.instance_name)
4319
    self.op.instance_name = instance_name = hostname1.name
4320

    
4321
    # this is just a preventive check, but someone might still add this
4322
    # instance in the meantime, and creation will fail at lock-add time
4323
    if instance_name in self.cfg.GetInstanceList():
4324
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4325
                                 instance_name)
4326

    
4327
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4328

    
4329
    # NIC buildup
4330
    self.nics = []
4331
    for nic in self.op.nics:
4332
      # ip validity checks
4333
      ip = nic.get("ip", None)
4334
      if ip is None or ip.lower() == "none":
4335
        nic_ip = None
4336
      elif ip.lower() == constants.VALUE_AUTO:
4337
        nic_ip = hostname1.ip
4338
      else:
4339
        if not utils.IsValidIP(ip):
4340
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4341
                                     " like a valid IP" % ip)
4342
        nic_ip = ip
4343

    
4344
      # MAC address verification
4345
      mac = nic.get("mac", constants.VALUE_AUTO)
4346
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4347
        if not utils.IsValidMac(mac.lower()):
4348
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4349
                                     mac)
4350
      # bridge verification
4351
      bridge = nic.get("bridge", None)
4352
      if bridge is None:
4353
        bridge = self.cfg.GetDefBridge()
4354
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4355

    
4356
    # disk checks/pre-build
4357
    self.disks = []
4358
    for disk in self.op.disks:
4359
      mode = disk.get("mode", constants.DISK_RDWR)
4360
      if mode not in constants.DISK_ACCESS_SET:
4361
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4362
                                   mode)
4363
      size = disk.get("size", None)
4364
      if size is None:
4365
        raise errors.OpPrereqError("Missing disk size")
4366
      try:
4367
        size = int(size)
4368
      except ValueError:
4369
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4370
      self.disks.append({"size": size, "mode": mode})
4371

    
4372
    # used in CheckPrereq for ip ping check
4373
    self.check_ip = hostname1.ip
4374

    
4375
    # file storage checks
4376
    if (self.op.file_driver and
4377
        not self.op.file_driver in constants.FILE_DRIVER):
4378
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4379
                                 self.op.file_driver)
4380

    
4381
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4382
      raise errors.OpPrereqError("File storage directory path not absolute")
4383

    
4384
    ### Node/iallocator related checks
4385
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4386
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4387
                                 " node must be given")
4388

    
4389
    if self.op.iallocator:
4390
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4391
    else:
4392
      self.op.pnode = self._ExpandNode(self.op.pnode)
4393
      nodelist = [self.op.pnode]
4394
      if self.op.snode is not None:
4395
        self.op.snode = self._ExpandNode(self.op.snode)
4396
        nodelist.append(self.op.snode)
4397
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4398

    
4399
    # in case of import lock the source node too
4400
    if self.op.mode == constants.INSTANCE_IMPORT:
4401
      src_node = getattr(self.op, "src_node", None)
4402
      src_path = getattr(self.op, "src_path", None)
4403

    
4404
      if src_path is None:
4405
        self.op.src_path = src_path = self.op.instance_name
4406

    
4407
      if src_node is None:
4408
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4409
        self.op.src_node = None
4410
        if os.path.isabs(src_path):
4411
          raise errors.OpPrereqError("Importing an instance from an absolute"
4412
                                     " path requires a source node option.")
4413
      else:
4414
        self.op.src_node = src_node = self._ExpandNode(src_node)
4415
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4416
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4417
        if not os.path.isabs(src_path):
4418
          self.op.src_path = src_path = \
4419
            os.path.join(constants.EXPORT_DIR, src_path)
4420

    
4421
    else: # INSTANCE_CREATE
4422
      if getattr(self.op, "os_type", None) is None:
4423
        raise errors.OpPrereqError("No guest OS specified")
4424

    
4425
  def _RunAllocator(self):
4426
    """Run the allocator based on input opcode.
4427

4428
    """
4429
    nics = [n.ToDict() for n in self.nics]
4430
    ial = IAllocator(self,
4431
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4432
                     name=self.op.instance_name,
4433
                     disk_template=self.op.disk_template,
4434
                     tags=[],
4435
                     os=self.op.os_type,
4436
                     vcpus=self.be_full[constants.BE_VCPUS],
4437
                     mem_size=self.be_full[constants.BE_MEMORY],
4438
                     disks=self.disks,
4439
                     nics=nics,
4440
                     hypervisor=self.op.hypervisor,
4441
                     )
4442

    
4443
    ial.Run(self.op.iallocator)
4444

    
4445
    if not ial.success:
4446
      raise errors.OpPrereqError("Can't compute nodes using"
4447
                                 " iallocator '%s': %s" % (self.op.iallocator,
4448
                                                           ial.info))
4449
    if len(ial.nodes) != ial.required_nodes:
4450
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4451
                                 " of nodes (%s), required %s" %
4452
                                 (self.op.iallocator, len(ial.nodes),
4453
                                  ial.required_nodes))
4454
    self.op.pnode = ial.nodes[0]
4455
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4456
                 self.op.instance_name, self.op.iallocator,
4457
                 ", ".join(ial.nodes))
4458
    if ial.required_nodes == 2:
4459
      self.op.snode = ial.nodes[1]
4460

    
4461
  def BuildHooksEnv(self):
4462
    """Build hooks env.
4463

4464
    This runs on master, primary and secondary nodes of the instance.
4465

4466
    """
4467
    env = {
4468
      "ADD_MODE": self.op.mode,
4469
      }
4470
    if self.op.mode == constants.INSTANCE_IMPORT:
4471
      env["SRC_NODE"] = self.op.src_node
4472
      env["SRC_PATH"] = self.op.src_path
4473
      env["SRC_IMAGES"] = self.src_images
4474

    
4475
    env.update(_BuildInstanceHookEnv(
4476
      name=self.op.instance_name,
4477
      primary_node=self.op.pnode,
4478
      secondary_nodes=self.secondaries,
4479
      status=self.op.start,
4480
      os_type=self.op.os_type,
4481
      memory=self.be_full[constants.BE_MEMORY],
4482
      vcpus=self.be_full[constants.BE_VCPUS],
4483
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4484
      disk_template=self.op.disk_template,
4485
      disks=[(d["size"], d["mode"]) for d in self.disks],
4486
    ))
4487

    
4488
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4489
          self.secondaries)
4490
    return env, nl, nl
4491

    
4492

    
4493
  def CheckPrereq(self):
4494
    """Check prerequisites.
4495

4496
    """
4497
    if (not self.cfg.GetVGName() and
4498
        self.op.disk_template not in constants.DTS_NOT_LVM):
4499
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4500
                                 " instances")
4501

    
4502
    if self.op.mode == constants.INSTANCE_IMPORT:
4503
      src_node = self.op.src_node
4504
      src_path = self.op.src_path
4505

    
4506
      if src_node is None:
4507
        exp_list = self.rpc.call_export_list(
4508
          self.acquired_locks[locking.LEVEL_NODE])
4509
        found = False
4510
        for node in exp_list:
4511
          if not exp_list[node].failed and src_path in exp_list[node].data:
4512
            found = True
4513
            self.op.src_node = src_node = node
4514
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4515
                                                       src_path)
4516
            break
4517
        if not found:
4518
          raise errors.OpPrereqError("No export found for relative path %s" %
4519
                                      src_path)
4520

    
4521
      _CheckNodeOnline(self, src_node)
4522
      result = self.rpc.call_export_info(src_node, src_path)
4523
      result.Raise()
4524
      if not result.data:
4525
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4526

    
4527
      export_info = result.data
4528
      if not export_info.has_section(constants.INISECT_EXP):
4529
        raise errors.ProgrammerError("Corrupted export config")
4530

    
4531
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4532
      if (int(ei_version) != constants.EXPORT_VERSION):
4533
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4534
                                   (ei_version, constants.EXPORT_VERSION))
4535

    
4536
      # Check that the new instance doesn't have less disks than the export
4537
      instance_disks = len(self.disks)
4538
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4539
      if instance_disks < export_disks:
4540
        raise errors.OpPrereqError("Not enough disks to import."
4541
                                   " (instance: %d, export: %d)" %
4542
                                   (instance_disks, export_disks))
4543

    
4544
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4545
      disk_images = []
4546
      for idx in range(export_disks):
4547
        option = 'disk%d_dump' % idx
4548
        if export_info.has_option(constants.INISECT_INS, option):
4549
          # FIXME: are the old os-es, disk sizes, etc. useful?
4550
          export_name = export_info.get(constants.INISECT_INS, option)
4551
          image = os.path.join(src_path, export_name)
4552
          disk_images.append(image)
4553
        else:
4554
          disk_images.append(False)
4555

    
4556
      self.src_images = disk_images
4557

    
4558
      old_name = export_info.get(constants.INISECT_INS, 'name')
4559
      # FIXME: int() here could throw a ValueError on broken exports
4560
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4561
      if self.op.instance_name == old_name:
4562
        for idx, nic in enumerate(self.nics):
4563
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4564
            nic_mac_ini = 'nic%d_mac' % idx
4565
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4566

    
4567
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4568
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4569
    if self.op.start and not self.op.ip_check:
4570
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4571
                                 " adding an instance in start mode")
4572

    
4573
    if self.op.ip_check:
4574
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4575
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4576
                                   (self.check_ip, self.op.instance_name))
4577

    
4578
    #### mac address generation
4579
    # By generating here the mac address both the allocator and the hooks get
4580
    # the real final mac address rather than the 'auto' or 'generate' value.
4581
    # There is a race condition between the generation and the instance object
4582
    # creation, which means that we know the mac is valid now, but we're not
4583
    # sure it will be when we actually add the instance. If things go bad
4584
    # adding the instance will abort because of a duplicate mac, and the
4585
    # creation job will fail.
4586
    for nic in self.nics:
4587
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4588
        nic.mac = self.cfg.GenerateMAC()
4589

    
4590
    #### allocator run
4591

    
4592
    if self.op.iallocator is not None:
4593
      self._RunAllocator()
4594

    
4595
    #### node related checks
4596

    
4597
    # check primary node
4598
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4599
    assert self.pnode is not None, \
4600
      "Cannot retrieve locked node %s" % self.op.pnode
4601
    if pnode.offline:
4602
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4603
                                 pnode.name)
4604
    if pnode.drained:
4605
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4606
                                 pnode.name)
4607

    
4608
    self.secondaries = []
4609

    
4610
    # mirror node verification
4611
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4612
      if self.op.snode is None:
4613
        raise errors.OpPrereqError("The networked disk templates need"
4614
                                   " a mirror node")
4615
      if self.op.snode == pnode.name:
4616
        raise errors.OpPrereqError("The secondary node cannot be"
4617
                                   " the primary node.")
4618
      _CheckNodeOnline(self, self.op.snode)
4619
      _CheckNodeNotDrained(self, self.op.snode)
4620
      self.secondaries.append(self.op.snode)
4621

    
4622
    nodenames = [pnode.name] + self.secondaries
4623

    
4624
    req_size = _ComputeDiskSize(self.op.disk_template,
4625
                                self.disks)
4626

    
4627
    # Check lv size requirements
4628
    if req_size is not None:
4629
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4630
                                         self.op.hypervisor)
4631
      for node in nodenames:
4632
        info = nodeinfo[node]
4633
        info.Raise()
4634
        info = info.data
4635
        if not info:
4636
          raise errors.OpPrereqError("Cannot get current information"
4637
                                     " from node '%s'" % node)
4638
        vg_free = info.get('vg_free', None)
4639
        if not isinstance(vg_free, int):
4640
          raise errors.OpPrereqError("Can't compute free disk space on"
4641
                                     " node %s" % node)
4642
        if req_size > info['vg_free']:
4643
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4644
                                     " %d MB available, %d MB required" %
4645
                                     (node, info['vg_free'], req_size))
4646

    
4647
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4648

    
4649
    # os verification
4650
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4651
    result.Raise()
4652
    if not isinstance(result.data, objects.OS):
4653
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4654
                                 " primary node"  % self.op.os_type)
4655

    
4656
    # bridge check on primary node
4657
    bridges = [n.bridge for n in self.nics]
4658
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4659
    result.Raise()
4660
    if not result.data:
4661
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4662
                                 " exist on destination node '%s'" %
4663
                                 (",".join(bridges), pnode.name))
4664

    
4665
    # memory check on primary node
4666
    if self.op.start:
4667
      _CheckNodeFreeMemory(self, self.pnode.name,
4668
                           "creating instance %s" % self.op.instance_name,
4669
                           self.be_full[constants.BE_MEMORY],
4670
                           self.op.hypervisor)
4671

    
4672
  def Exec(self, feedback_fn):
4673
    """Create and add the instance to the cluster.
4674

4675
    """
4676
    instance = self.op.instance_name
4677
    pnode_name = self.pnode.name
4678

    
4679
    ht_kind = self.op.hypervisor
4680
    if ht_kind in constants.HTS_REQ_PORT:
4681
      network_port = self.cfg.AllocatePort()
4682
    else:
4683
      network_port = None
4684

    
4685
    ##if self.op.vnc_bind_address is None:
4686
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4687

    
4688
    # this is needed because os.path.join does not accept None arguments
4689
    if self.op.file_storage_dir is None:
4690
      string_file_storage_dir = ""
4691
    else:
4692
      string_file_storage_dir = self.op.file_storage_dir
4693

    
4694
    # build the full file storage dir path
4695
    file_storage_dir = os.path.normpath(os.path.join(
4696
                                        self.cfg.GetFileStorageDir(),
4697
                                        string_file_storage_dir, instance))
4698

    
4699

    
4700
    disks = _GenerateDiskTemplate(self,
4701
                                  self.op.disk_template,
4702
                                  instance, pnode_name,
4703
                                  self.secondaries,
4704
                                  self.disks,
4705
                                  file_storage_dir,
4706
                                  self.op.file_driver,
4707
                                  0)
4708

    
4709
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4710
                            primary_node=pnode_name,
4711
                            nics=self.nics, disks=disks,
4712
                            disk_template=self.op.disk_template,
4713
                            admin_up=False,
4714
                            network_port=network_port,
4715
                            beparams=self.op.beparams,
4716
                            hvparams=self.op.hvparams,
4717
                            hypervisor=self.op.hypervisor,
4718
                            )
4719

    
4720
    feedback_fn("* creating instance disks...")
4721
    try:
4722
      _CreateDisks(self, iobj)
4723
    except errors.OpExecError:
4724
      self.LogWarning("Device creation failed, reverting...")
4725
      try:
4726
        _RemoveDisks(self, iobj)
4727
      finally:
4728
        self.cfg.ReleaseDRBDMinors(instance)
4729
        raise
4730

    
4731
    feedback_fn("adding instance %s to cluster config" % instance)
4732

    
4733
    self.cfg.AddInstance(iobj)
4734
    # Declare that we don't want to remove the instance lock anymore, as we've
4735
    # added the instance to the config
4736
    del self.remove_locks[locking.LEVEL_INSTANCE]
4737
    # Unlock all the nodes
4738
    if self.op.mode == constants.INSTANCE_IMPORT:
4739
      nodes_keep = [self.op.src_node]
4740
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4741
                       if node != self.op.src_node]
4742
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4743
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4744
    else:
4745
      self.context.glm.release(locking.LEVEL_NODE)
4746
      del self.acquired_locks[locking.LEVEL_NODE]
4747

    
4748
    if self.op.wait_for_sync:
4749
      disk_abort = not _WaitForSync(self, iobj)
4750
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4751
      # make sure the disks are not degraded (still sync-ing is ok)
4752
      time.sleep(15)
4753
      feedback_fn("* checking mirrors status")
4754
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4755
    else:
4756
      disk_abort = False
4757

    
4758
    if disk_abort:
4759
      _RemoveDisks(self, iobj)
4760
      self.cfg.RemoveInstance(iobj.name)
4761
      # Make sure the instance lock gets removed
4762
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4763
      raise errors.OpExecError("There are some degraded disks for"
4764
                               " this instance")
4765

    
4766
    feedback_fn("creating os for instance %s on node %s" %
4767
                (instance, pnode_name))
4768

    
4769
    if iobj.disk_template != constants.DT_DISKLESS:
4770
      if self.op.mode == constants.INSTANCE_CREATE:
4771
        feedback_fn("* running the instance OS create scripts...")
4772
        result = self.rpc.call_instance_os_add(pnode_name, iobj)
4773
        msg = result.RemoteFailMsg()
4774
        if msg:
4775
          raise errors.OpExecError("Could not add os for instance %s"
4776
                                   " on node %s: %s" %
4777
                                   (instance, pnode_name, msg))
4778

    
4779
      elif self.op.mode == constants.INSTANCE_IMPORT:
4780
        feedback_fn("* running the instance OS import scripts...")
4781
        src_node = self.op.src_node
4782
        src_images = self.src_images
4783
        cluster_name = self.cfg.GetClusterName()
4784
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4785
                                                         src_node, src_images,
4786
                                                         cluster_name)
4787
        import_result.Raise()
4788
        for idx, result in enumerate(import_result.data):
4789
          if not result:
4790
            self.LogWarning("Could not import the image %s for instance"
4791
                            " %s, disk %d, on node %s" %
4792
                            (src_images[idx], instance, idx, pnode_name))
4793
      else:
4794
        # also checked in the prereq part
4795
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4796
                                     % self.op.mode)
4797

    
4798
    if self.op.start:
4799
      iobj.admin_up = True
4800
      self.cfg.Update(iobj)
4801
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4802
      feedback_fn("* starting instance...")
4803
      result = self.rpc.call_instance_start(pnode_name, iobj)
4804
      msg = result.RemoteFailMsg()
4805
      if msg:
4806
        raise errors.OpExecError("Could not start instance: %s" % msg)
4807

    
4808

    
4809
class LUConnectConsole(NoHooksLU):
4810
  """Connect to an instance's console.
4811

4812
  This is somewhat special in that it returns the command line that
4813
  you need to run on the master node in order to connect to the
4814
  console.
4815

4816
  """
4817
  _OP_REQP = ["instance_name"]
4818
  REQ_BGL = False
4819

    
4820
  def ExpandNames(self):
4821
    self._ExpandAndLockInstance()
4822

    
4823
  def CheckPrereq(self):
4824
    """Check prerequisites.
4825

4826
    This checks that the instance is in the cluster.
4827

4828
    """
4829
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4830
    assert self.instance is not None, \
4831
      "Cannot retrieve locked instance %s" % self.op.instance_name
4832
    _CheckNodeOnline(self, self.instance.primary_node)
4833

    
4834
  def Exec(self, feedback_fn):
4835
    """Connect to the console of an instance
4836

4837
    """
4838
    instance = self.instance
4839
    node = instance.primary_node
4840

    
4841
    node_insts = self.rpc.call_instance_list([node],
4842
                                             [instance.hypervisor])[node]
4843
    node_insts.Raise()
4844

    
4845
    if instance.name not in node_insts.data:
4846
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4847

    
4848
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4849

    
4850
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4851
    cluster = self.cfg.GetClusterInfo()
4852
    # beparams and hvparams are passed separately, to avoid editing the
4853
    # instance and then saving the defaults in the instance itself.
4854
    hvparams = cluster.FillHV(instance)
4855
    beparams = cluster.FillBE(instance)
4856
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4857

    
4858
    # build ssh cmdline
4859
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
4860

    
4861

    
4862
class LUReplaceDisks(LogicalUnit):
4863
  """Replace the disks of an instance.
4864

4865
  """
4866
  HPATH = "mirrors-replace"
4867
  HTYPE = constants.HTYPE_INSTANCE
4868
  _OP_REQP = ["instance_name", "mode", "disks"]
4869
  REQ_BGL = False
4870

    
4871
  def CheckArguments(self):
4872
    if not hasattr(self.op, "remote_node"):
4873
      self.op.remote_node = None
4874
    if not hasattr(self.op, "iallocator"):
4875
      self.op.iallocator = None
4876

    
4877
    # check for valid parameter combination
4878
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
4879
    if self.op.mode == constants.REPLACE_DISK_CHG:
4880
      if cnt == 2:
4881
        raise errors.OpPrereqError("When changing the secondary either an"
4882
                                   " iallocator script must be used or the"
4883
                                   " new node given")
4884
      elif cnt == 0:
4885
        raise errors.OpPrereqError("Give either the iallocator or the new"
4886
                                   " secondary, not both")
4887
    else: # not replacing the secondary
4888
      if cnt != 2:
4889
        raise errors.OpPrereqError("The iallocator and new node options can"
4890
                                   " be used only when changing the"
4891
                                   " secondary node")
4892

    
4893
  def ExpandNames(self):
4894
    self._ExpandAndLockInstance()
4895

    
4896
    if self.op.iallocator is not None:
4897
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4898
    elif self.op.remote_node is not None:
4899
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
4900
      if remote_node is None:
4901
        raise errors.OpPrereqError("Node '%s' not known" %
4902
                                   self.op.remote_node)
4903
      self.op.remote_node = remote_node
4904
      # Warning: do not remove the locking of the new secondary here
4905
      # unless DRBD8.AddChildren is changed to work in parallel;
4906
      # currently it doesn't since parallel invocations of
4907
      # FindUnusedMinor will conflict
4908
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
4909
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
4910
    else:
4911
      self.needed_locks[locking.LEVEL_NODE] = []
4912
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4913

    
4914
  def DeclareLocks(self, level):
4915
    # If we're not already locking all nodes in the set we have to declare the
4916
    # instance's primary/secondary nodes.
4917
    if (level == locking.LEVEL_NODE and
4918
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
4919
      self._LockInstancesNodes()
4920

    
4921
  def _RunAllocator(self):
4922
    """Compute a new secondary node using an IAllocator.
4923

4924
    """
4925
    ial = IAllocator(self,
4926
                     mode=constants.IALLOCATOR_MODE_RELOC,
4927
                     name=self.op.instance_name,
4928
                     relocate_from=[self.sec_node])
4929

    
4930
    ial.Run(self.op.iallocator)
4931

    
4932
    if not ial.success:
4933
      raise errors.OpPrereqError("Can't compute nodes using"
4934
                                 " iallocator '%s': %s" % (self.op.iallocator,
4935
                                                           ial.info))
4936
    if len(ial.nodes) != ial.required_nodes:
4937
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4938
                                 " of nodes (%s), required %s" %
4939
                                 (len(ial.nodes), ial.required_nodes))
4940
    self.op.remote_node = ial.nodes[0]
4941
    self.LogInfo("Selected new secondary for the instance: %s",
4942
                 self.op.remote_node)
4943

    
4944
  def BuildHooksEnv(self):
4945
    """Build hooks env.
4946

4947
    This runs on the master, the primary and all the secondaries.
4948

4949
    """
4950
    env = {
4951
      "MODE": self.op.mode,
4952
      "NEW_SECONDARY": self.op.remote_node,
4953
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
4954
      }
4955
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4956
    nl = [
4957
      self.cfg.GetMasterNode(),
4958
      self.instance.primary_node,
4959
      ]
4960
    if self.op.remote_node is not None:
4961
      nl.append(self.op.remote_node)
4962
    return env, nl, nl
4963

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

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

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

    
4975
    if instance.disk_template != constants.DT_DRBD8:
4976
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
4977
                                 " instances")
4978

    
4979
    if len(instance.secondary_nodes) != 1:
4980
      raise errors.OpPrereqError("The instance has a strange layout,"
4981
                                 " expected one secondary but found %d" %
4982
                                 len(instance.secondary_nodes))
4983

    
4984
    self.sec_node = instance.secondary_nodes[0]
4985

    
4986
    if self.op.iallocator is not None:
4987
      self._RunAllocator()
4988

    
4989
    remote_node = self.op.remote_node
4990
    if remote_node is not None:
4991
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
4992
      assert self.remote_node_info is not None, \
4993
        "Cannot retrieve locked node %s" % remote_node
4994
    else:
4995
      self.remote_node_info = None
4996
    if remote_node == instance.primary_node:
4997
      raise errors.OpPrereqError("The specified node is the primary node of"
4998
                                 " the instance.")
4999
    elif remote_node == self.sec_node:
5000
      raise errors.OpPrereqError("The specified node is already the"
5001
                                 " secondary node of the instance.")
5002

    
5003
    if self.op.mode == constants.REPLACE_DISK_PRI:
5004
      n1 = self.tgt_node = instance.primary_node
5005
      n2 = self.oth_node = self.sec_node
5006
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5007
      n1 = self.tgt_node = self.sec_node
5008
      n2 = self.oth_node = instance.primary_node
5009
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5010
      n1 = self.new_node = remote_node
5011
      n2 = self.oth_node = instance.primary_node
5012
      self.tgt_node = self.sec_node
5013
      _CheckNodeNotDrained(self, remote_node)
5014
    else:
5015
      raise errors.ProgrammerError("Unhandled disk replace mode")
5016

    
5017
    _CheckNodeOnline(self, n1)
5018
    _CheckNodeOnline(self, n2)
5019

    
5020
    if not self.op.disks:
5021
      self.op.disks = range(len(instance.disks))
5022

    
5023
    for disk_idx in self.op.disks:
5024
      instance.FindDisk(disk_idx)
5025

    
5026
  def _ExecD8DiskOnly(self, feedback_fn):
5027
    """Replace a disk on the primary or secondary for dbrd8.
5028

5029
    The algorithm for replace is quite complicated:
5030

5031
      1. for each disk to be replaced:
5032

5033
        1. create new LVs on the target node with unique names
5034
        1. detach old LVs from the drbd device
5035
        1. rename old LVs to name_replaced.<time_t>
5036
        1. rename new LVs to old LVs
5037
        1. attach the new LVs (with the old names now) to the drbd device
5038

5039
      1. wait for sync across all devices
5040

5041
      1. for each modified disk:
5042

5043
        1. remove old LVs (which have the name name_replaces.<time_t>)
5044

5045
    Failures are not very well handled.
5046

5047
    """
5048
    steps_total = 6
5049
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5050
    instance = self.instance
5051
    iv_names = {}
5052
    vgname = self.cfg.GetVGName()
5053
    # start of work
5054
    cfg = self.cfg
5055
    tgt_node = self.tgt_node
5056
    oth_node = self.oth_node
5057

    
5058
    # Step: check device activation
5059
    self.proc.LogStep(1, steps_total, "check device existence")
5060
    info("checking volume groups")
5061
    my_vg = cfg.GetVGName()
5062
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5063
    if not results:
5064
      raise errors.OpExecError("Can't list volume groups on the nodes")
5065
    for node in oth_node, tgt_node:
5066
      res = results[node]
5067
      if res.failed or not res.data or my_vg not in res.data:
5068
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5069
                                 (my_vg, node))
5070
    for idx, dev in enumerate(instance.disks):
5071
      if idx not in self.op.disks:
5072
        continue
5073
      for node in tgt_node, oth_node:
5074
        info("checking disk/%d on %s" % (idx, node))
5075
        cfg.SetDiskID(dev, node)
5076
        result = self.rpc.call_blockdev_find(node, dev)
5077
        msg = result.RemoteFailMsg()
5078
        if not msg and not result.payload:
5079
          msg = "disk not found"
5080
        if msg:
5081
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5082
                                   (idx, node, msg))
5083

    
5084
    # Step: check other node consistency
5085
    self.proc.LogStep(2, steps_total, "check peer consistency")
5086
    for idx, dev in enumerate(instance.disks):
5087
      if idx not in self.op.disks:
5088
        continue
5089
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5090
      if not _CheckDiskConsistency(self, dev, oth_node,
5091
                                   oth_node==instance.primary_node):
5092
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5093
                                 " to replace disks on this node (%s)" %
5094
                                 (oth_node, tgt_node))
5095

    
5096
    # Step: create new storage
5097
    self.proc.LogStep(3, steps_total, "allocate new storage")
5098
    for idx, dev in enumerate(instance.disks):
5099
      if idx not in self.op.disks:
5100
        continue
5101
      size = dev.size
5102
      cfg.SetDiskID(dev, tgt_node)
5103
      lv_names = [".disk%d_%s" % (idx, suf)
5104
                  for suf in ["data", "meta"]]
5105
      names = _GenerateUniqueNames(self, lv_names)
5106
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5107
                             logical_id=(vgname, names[0]))
5108
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5109
                             logical_id=(vgname, names[1]))
5110
      new_lvs = [lv_data, lv_meta]
5111
      old_lvs = dev.children
5112
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5113
      info("creating new local storage on %s for %s" %
5114
           (tgt_node, dev.iv_name))
5115
      # we pass force_create=True to force the LVM creation
5116
      for new_lv in new_lvs:
5117
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5118
                        _GetInstanceInfoText(instance), False)
5119

    
5120
    # Step: for each lv, detach+rename*2+attach
5121
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5122
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5123
      info("detaching %s drbd from local storage" % dev.iv_name)
5124
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5125
      result.Raise()
5126
      if not result.data:
5127
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5128
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5129
      #dev.children = []
5130
      #cfg.Update(instance)
5131

    
5132
      # ok, we created the new LVs, so now we know we have the needed
5133
      # storage; as such, we proceed on the target node to rename
5134
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5135
      # using the assumption that logical_id == physical_id (which in
5136
      # turn is the unique_id on that node)
5137

    
5138
      # FIXME(iustin): use a better name for the replaced LVs
5139
      temp_suffix = int(time.time())
5140
      ren_fn = lambda d, suff: (d.physical_id[0],
5141
                                d.physical_id[1] + "_replaced-%s" % suff)
5142
      # build the rename list based on what LVs exist on the node
5143
      rlist = []
5144
      for to_ren in old_lvs:
5145
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5146
        if not result.RemoteFailMsg() and result.payload:
5147
          # device exists
5148
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5149

    
5150
      info("renaming the old LVs on the target node")
5151
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5152
      result.Raise()
5153
      if not result.data:
5154
        raise errors.OpExecError("Can't rename old LVs on node %s" % tgt_node)
5155
      # now we rename the new LVs to the old LVs
5156
      info("renaming the new LVs on the target node")
5157
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5158
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5159
      result.Raise()
5160
      if not result.data:
5161
        raise errors.OpExecError("Can't rename new LVs on node %s" % tgt_node)
5162

    
5163
      for old, new in zip(old_lvs, new_lvs):
5164
        new.logical_id = old.logical_id
5165
        cfg.SetDiskID(new, tgt_node)
5166

    
5167
      for disk in old_lvs:
5168
        disk.logical_id = ren_fn(disk, temp_suffix)
5169
        cfg.SetDiskID(disk, tgt_node)
5170

    
5171
      # now that the new lvs have the old name, we can add them to the device
5172
      info("adding new mirror component on %s" % tgt_node)
5173
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5174
      if result.failed or not result.data:
5175
        for new_lv in new_lvs:
5176
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5177
          if msg:
5178
            warning("Can't rollback device %s: %s", dev, msg,
5179
                    hint="cleanup manually the unused logical volumes")
5180
        raise errors.OpExecError("Can't add local storage to drbd")
5181

    
5182
      dev.children = new_lvs
5183
      cfg.Update(instance)
5184

    
5185
    # Step: wait for sync
5186

    
5187
    # this can fail as the old devices are degraded and _WaitForSync
5188
    # does a combined result over all disks, so we don't check its
5189
    # return value
5190
    self.proc.LogStep(5, steps_total, "sync devices")
5191
    _WaitForSync(self, instance, unlock=True)
5192

    
5193
    # so check manually all the devices
5194
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5195
      cfg.SetDiskID(dev, instance.primary_node)
5196
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5197
      msg = result.RemoteFailMsg()
5198
      if not msg and not result.payload:
5199
        msg = "disk not found"
5200
      if msg:
5201
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5202
                                 (name, msg))
5203
      if result.payload[5]:
5204
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5205

    
5206
    # Step: remove old storage
5207
    self.proc.LogStep(6, steps_total, "removing old storage")
5208
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5209
      info("remove logical volumes for %s" % name)
5210
      for lv in old_lvs:
5211
        cfg.SetDiskID(lv, tgt_node)
5212
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5213
        if msg:
5214
          warning("Can't remove old LV: %s" % msg,
5215
                  hint="manually remove unused LVs")
5216
          continue
5217

    
5218
  def _ExecD8Secondary(self, feedback_fn):
5219
    """Replace the secondary node for drbd8.
5220

5221
    The algorithm for replace is quite complicated:
5222
      - for all disks of the instance:
5223
        - create new LVs on the new node with same names
5224
        - shutdown the drbd device on the old secondary
5225
        - disconnect the drbd network on the primary
5226
        - create the drbd device on the new secondary
5227
        - network attach the drbd on the primary, using an artifice:
5228
          the drbd code for Attach() will connect to the network if it
5229
          finds a device which is connected to the good local disks but
5230
          not network enabled
5231
      - wait for sync across all devices
5232
      - remove all disks from the old secondary
5233

5234
    Failures are not very well handled.
5235

5236
    """
5237
    steps_total = 6
5238
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5239
    instance = self.instance
5240
    iv_names = {}
5241
    # start of work
5242
    cfg = self.cfg
5243
    old_node = self.tgt_node
5244
    new_node = self.new_node
5245
    pri_node = instance.primary_node
5246
    nodes_ip = {
5247
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5248
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5249
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5250
      }
5251

    
5252
    # Step: check device activation
5253
    self.proc.LogStep(1, steps_total, "check device existence")
5254
    info("checking volume groups")
5255
    my_vg = cfg.GetVGName()
5256
    results = self.rpc.call_vg_list([pri_node, new_node])
5257
    for node in pri_node, new_node:
5258
      res = results[node]
5259
      if res.failed or not res.data or my_vg not in res.data:
5260
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5261
                                 (my_vg, node))
5262
    for idx, dev in enumerate(instance.disks):
5263
      if idx not in self.op.disks:
5264
        continue
5265
      info("checking disk/%d on %s" % (idx, pri_node))
5266
      cfg.SetDiskID(dev, pri_node)
5267
      result = self.rpc.call_blockdev_find(pri_node, dev)
5268
      msg = result.RemoteFailMsg()
5269
      if not msg and not result.payload:
5270
        msg = "disk not found"
5271
      if msg:
5272
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5273
                                 (idx, pri_node, msg))
5274

    
5275
    # Step: check other node consistency
5276
    self.proc.LogStep(2, steps_total, "check peer consistency")
5277
    for idx, dev in enumerate(instance.disks):
5278
      if idx not in self.op.disks:
5279
        continue
5280
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5281
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5282
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5283
                                 " unsafe to replace the secondary" %
5284
                                 pri_node)
5285

    
5286
    # Step: create new storage
5287
    self.proc.LogStep(3, steps_total, "allocate new storage")
5288
    for idx, dev in enumerate(instance.disks):
5289
      info("adding new local storage on %s for disk/%d" %
5290
           (new_node, idx))
5291
      # we pass force_create=True to force LVM creation
5292
      for new_lv in dev.children:
5293
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5294
                        _GetInstanceInfoText(instance), False)
5295

    
5296
    # Step 4: dbrd minors and drbd setups changes
5297
    # after this, we must manually remove the drbd minors on both the
5298
    # error and the success paths
5299
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5300
                                   instance.name)
5301
    logging.debug("Allocated minors %s" % (minors,))
5302
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5303
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5304
      size = dev.size
5305
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5306
      # create new devices on new_node; note that we create two IDs:
5307
      # one without port, so the drbd will be activated without
5308
      # networking information on the new node at this stage, and one
5309
      # with network, for the latter activation in step 4
5310
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5311
      if pri_node == o_node1:
5312
        p_minor = o_minor1
5313
      else:
5314
        p_minor = o_minor2
5315

    
5316
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5317
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5318

    
5319
      iv_names[idx] = (dev, dev.children, new_net_id)
5320
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5321
                    new_net_id)
5322
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5323
                              logical_id=new_alone_id,
5324
                              children=dev.children)
5325
      try:
5326
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5327
                              _GetInstanceInfoText(instance), False)
5328
      except errors.GenericError:
5329
        self.cfg.ReleaseDRBDMinors(instance.name)
5330
        raise
5331

    
5332
    for idx, dev in enumerate(instance.disks):
5333
      # we have new devices, shutdown the drbd on the old secondary
5334
      info("shutting down drbd for disk/%d on old node" % idx)
5335
      cfg.SetDiskID(dev, old_node)
5336
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5337
      if msg:
5338
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5339
                (idx, msg),
5340
                hint="Please cleanup this device manually as soon as possible")
5341

    
5342
    info("detaching primary drbds from the network (=> standalone)")
5343
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5344
                                               instance.disks)[pri_node]
5345

    
5346
    msg = result.RemoteFailMsg()
5347
    if msg:
5348
      # detaches didn't succeed (unlikely)
5349
      self.cfg.ReleaseDRBDMinors(instance.name)
5350
      raise errors.OpExecError("Can't detach the disks from the network on"
5351
                               " old node: %s" % (msg,))
5352

    
5353
    # if we managed to detach at least one, we update all the disks of
5354
    # the instance to point to the new secondary
5355
    info("updating instance configuration")
5356
    for dev, _, new_logical_id in iv_names.itervalues():
5357
      dev.logical_id = new_logical_id
5358
      cfg.SetDiskID(dev, pri_node)
5359
    cfg.Update(instance)
5360

    
5361
    # and now perform the drbd attach
5362
    info("attaching primary drbds to new secondary (standalone => connected)")
5363
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5364
                                           instance.disks, instance.name,
5365
                                           False)
5366
    for to_node, to_result in result.items():
5367
      msg = to_result.RemoteFailMsg()
5368
      if msg:
5369
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5370
                hint="please do a gnt-instance info to see the"
5371
                " status of disks")
5372

    
5373
    # this can fail as the old devices are degraded and _WaitForSync
5374
    # does a combined result over all disks, so we don't check its
5375
    # return value
5376
    self.proc.LogStep(5, steps_total, "sync devices")
5377
    _WaitForSync(self, instance, unlock=True)
5378

    
5379
    # so check manually all the devices
5380
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5381
      cfg.SetDiskID(dev, pri_node)
5382
      result = self.rpc.call_blockdev_find(pri_node, dev)
5383
      msg = result.RemoteFailMsg()
5384
      if not msg and not result.payload:
5385
        msg = "disk not found"
5386
      if msg:
5387
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5388
                                 (idx, msg))
5389
      if result.payload[5]:
5390
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5391

    
5392
    self.proc.LogStep(6, steps_total, "removing old storage")
5393
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5394
      info("remove logical volumes for disk/%d" % idx)
5395
      for lv in old_lvs:
5396
        cfg.SetDiskID(lv, old_node)
5397
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5398
        if msg:
5399
          warning("Can't remove LV on old secondary: %s", msg,
5400
                  hint="Cleanup stale volumes by hand")
5401

    
5402
  def Exec(self, feedback_fn):
5403
    """Execute disk replacement.
5404

5405
    This dispatches the disk replacement to the appropriate handler.
5406

5407
    """
5408
    instance = self.instance
5409

    
5410
    # Activate the instance disks if we're replacing them on a down instance
5411
    if not instance.admin_up:
5412
      _StartInstanceDisks(self, instance, True)
5413

    
5414
    if self.op.mode == constants.REPLACE_DISK_CHG:
5415
      fn = self._ExecD8Secondary
5416
    else:
5417
      fn = self._ExecD8DiskOnly
5418

    
5419
    ret = fn(feedback_fn)
5420

    
5421
    # Deactivate the instance disks if we're replacing them on a down instance
5422
    if not instance.admin_up:
5423
      _SafeShutdownInstanceDisks(self, instance)
5424

    
5425
    return ret
5426

    
5427

    
5428
class LUGrowDisk(LogicalUnit):
5429
  """Grow a disk of an instance.
5430

5431
  """
5432
  HPATH = "disk-grow"
5433
  HTYPE = constants.HTYPE_INSTANCE
5434
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5435
  REQ_BGL = False
5436

    
5437
  def ExpandNames(self):
5438
    self._ExpandAndLockInstance()
5439
    self.needed_locks[locking.LEVEL_NODE] = []
5440
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5441

    
5442
  def DeclareLocks(self, level):
5443
    if level == locking.LEVEL_NODE:
5444
      self._LockInstancesNodes()
5445

    
5446
  def BuildHooksEnv(self):
5447
    """Build hooks env.
5448

5449
    This runs on the master, the primary and all the secondaries.
5450

5451
    """
5452
    env = {
5453
      "DISK": self.op.disk,
5454
      "AMOUNT": self.op.amount,
5455
      }
5456
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5457
    nl = [
5458
      self.cfg.GetMasterNode(),
5459
      self.instance.primary_node,
5460
      ]
5461
    return env, nl, nl
5462

    
5463
  def CheckPrereq(self):
5464
    """Check prerequisites.
5465

5466
    This checks that the instance is in the cluster.
5467

5468
    """
5469
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5470
    assert instance is not None, \
5471
      "Cannot retrieve locked instance %s" % self.op.instance_name
5472
    nodenames = list(instance.all_nodes)
5473
    for node in nodenames:
5474
      _CheckNodeOnline(self, node)
5475

    
5476

    
5477
    self.instance = instance
5478

    
5479
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5480
      raise errors.OpPrereqError("Instance's disk layout does not support"
5481
                                 " growing.")
5482

    
5483
    self.disk = instance.FindDisk(self.op.disk)
5484

    
5485
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5486
                                       instance.hypervisor)
5487
    for node in nodenames:
5488
      info = nodeinfo[node]
5489
      if info.failed or not info.data:
5490
        raise errors.OpPrereqError("Cannot get current information"
5491
                                   " from node '%s'" % node)
5492
      vg_free = info.data.get('vg_free', None)
5493
      if not isinstance(vg_free, int):
5494
        raise errors.OpPrereqError("Can't compute free disk space on"
5495
                                   " node %s" % node)
5496
      if self.op.amount > vg_free:
5497
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5498
                                   " %d MiB available, %d MiB required" %
5499
                                   (node, vg_free, self.op.amount))
5500

    
5501
  def Exec(self, feedback_fn):
5502
    """Execute disk grow.
5503

5504
    """
5505
    instance = self.instance
5506
    disk = self.disk
5507
    for node in instance.all_nodes:
5508
      self.cfg.SetDiskID(disk, node)
5509
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5510
      msg = result.RemoteFailMsg()
5511
      if msg:
5512
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5513
                                 (node, msg))
5514
    disk.RecordGrow(self.op.amount)
5515
    self.cfg.Update(instance)
5516
    if self.op.wait_for_sync:
5517
      disk_abort = not _WaitForSync(self, instance)
5518
      if disk_abort:
5519
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5520
                             " status.\nPlease check the instance.")
5521

    
5522

    
5523
class LUQueryInstanceData(NoHooksLU):
5524
  """Query runtime instance data.
5525

5526
  """
5527
  _OP_REQP = ["instances", "static"]
5528
  REQ_BGL = False
5529

    
5530
  def ExpandNames(self):
5531
    self.needed_locks = {}
5532
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5533

    
5534
    if not isinstance(self.op.instances, list):
5535
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5536

    
5537
    if self.op.instances:
5538
      self.wanted_names = []
5539
      for name in self.op.instances:
5540
        full_name = self.cfg.ExpandInstanceName(name)
5541
        if full_name is None:
5542
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5543
        self.wanted_names.append(full_name)
5544
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5545
    else:
5546
      self.wanted_names = None
5547
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5548

    
5549
    self.needed_locks[locking.LEVEL_NODE] = []
5550
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5551

    
5552
  def DeclareLocks(self, level):
5553
    if level == locking.LEVEL_NODE:
5554
      self._LockInstancesNodes()
5555

    
5556
  def CheckPrereq(self):
5557
    """Check prerequisites.
5558

5559
    This only checks the optional instance list against the existing names.
5560

5561
    """
5562
    if self.wanted_names is None:
5563
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5564

    
5565
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5566
                             in self.wanted_names]
5567
    return
5568

    
5569
  def _ComputeDiskStatus(self, instance, snode, dev):
5570
    """Compute block device status.
5571

5572
    """
5573
    static = self.op.static
5574
    if not static:
5575
      self.cfg.SetDiskID(dev, instance.primary_node)
5576
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5577
      if dev_pstatus.offline:
5578
        dev_pstatus = None
5579
      else:
5580
        msg = dev_pstatus.RemoteFailMsg()
5581
        if msg:
5582
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5583
                                   (instance.name, msg))
5584
        dev_pstatus = dev_pstatus.payload
5585
    else:
5586
      dev_pstatus = None
5587

    
5588
    if dev.dev_type in constants.LDS_DRBD:
5589
      # we change the snode then (otherwise we use the one passed in)
5590
      if dev.logical_id[0] == instance.primary_node:
5591
        snode = dev.logical_id[1]
5592
      else:
5593
        snode = dev.logical_id[0]
5594

    
5595
    if snode and not static:
5596
      self.cfg.SetDiskID(dev, snode)
5597
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5598
      if dev_sstatus.offline:
5599
        dev_sstatus = None
5600
      else:
5601
        msg = dev_sstatus.RemoteFailMsg()
5602
        if msg:
5603
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5604
                                   (instance.name, msg))
5605
        dev_sstatus = dev_sstatus.payload
5606
    else:
5607
      dev_sstatus = None
5608

    
5609
    if dev.children:
5610
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5611
                      for child in dev.children]
5612
    else:
5613
      dev_children = []
5614

    
5615
    data = {
5616
      "iv_name": dev.iv_name,
5617
      "dev_type": dev.dev_type,
5618
      "logical_id": dev.logical_id,
5619
      "physical_id": dev.physical_id,
5620
      "pstatus": dev_pstatus,
5621
      "sstatus": dev_sstatus,
5622
      "children": dev_children,
5623
      "mode": dev.mode,
5624
      }
5625

    
5626
    return data
5627

    
5628
  def Exec(self, feedback_fn):
5629
    """Gather and return data"""
5630
    result = {}
5631

    
5632
    cluster = self.cfg.GetClusterInfo()
5633

    
5634
    for instance in self.wanted_instances:
5635
      if not self.op.static:
5636
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5637
                                                  instance.name,
5638
                                                  instance.hypervisor)
5639
        remote_info.Raise()
5640
        remote_info = remote_info.data
5641
        if remote_info and "state" in remote_info:
5642
          remote_state = "up"
5643
        else:
5644
          remote_state = "down"
5645
      else:
5646
        remote_state = None
5647
      if instance.admin_up:
5648
        config_state = "up"
5649
      else:
5650
        config_state = "down"
5651

    
5652
      disks = [self._ComputeDiskStatus(instance, None, device)
5653
               for device in instance.disks]
5654

    
5655
      idict = {
5656
        "name": instance.name,
5657
        "config_state": config_state,
5658
        "run_state": remote_state,
5659
        "pnode": instance.primary_node,
5660
        "snodes": instance.secondary_nodes,
5661
        "os": instance.os,
5662
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5663
        "disks": disks,
5664
        "hypervisor": instance.hypervisor,
5665
        "network_port": instance.network_port,
5666
        "hv_instance": instance.hvparams,
5667
        "hv_actual": cluster.FillHV(instance),
5668
        "be_instance": instance.beparams,
5669
        "be_actual": cluster.FillBE(instance),
5670
        }
5671

    
5672
      result[instance.name] = idict
5673

    
5674
    return result
5675

    
5676

    
5677
class LUSetInstanceParams(LogicalUnit):
5678
  """Modifies an instances's parameters.
5679

5680
  """
5681
  HPATH = "instance-modify"
5682
  HTYPE = constants.HTYPE_INSTANCE
5683
  _OP_REQP = ["instance_name"]
5684
  REQ_BGL = False
5685

    
5686
  def CheckArguments(self):
5687
    if not hasattr(self.op, 'nics'):
5688
      self.op.nics = []
5689
    if not hasattr(self.op, 'disks'):
5690
      self.op.disks = []
5691
    if not hasattr(self.op, 'beparams'):
5692
      self.op.beparams = {}
5693
    if not hasattr(self.op, 'hvparams'):
5694
      self.op.hvparams = {}
5695
    self.op.force = getattr(self.op, "force", False)
5696
    if not (self.op.nics or self.op.disks or
5697
            self.op.hvparams or self.op.beparams):
5698
      raise errors.OpPrereqError("No changes submitted")
5699

    
5700
    # Disk validation
5701
    disk_addremove = 0
5702
    for disk_op, disk_dict in self.op.disks:
5703
      if disk_op == constants.DDM_REMOVE:
5704
        disk_addremove += 1
5705
        continue
5706
      elif disk_op == constants.DDM_ADD:
5707
        disk_addremove += 1
5708
      else:
5709
        if not isinstance(disk_op, int):
5710
          raise errors.OpPrereqError("Invalid disk index")
5711
      if disk_op == constants.DDM_ADD:
5712
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5713
        if mode not in constants.DISK_ACCESS_SET:
5714
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5715
        size = disk_dict.get('size', None)
5716
        if size is None:
5717
          raise errors.OpPrereqError("Required disk parameter size missing")
5718
        try:
5719
          size = int(size)
5720
        except ValueError, err:
5721
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5722
                                     str(err))
5723
        disk_dict['size'] = size
5724
      else:
5725
        # modification of disk
5726
        if 'size' in disk_dict:
5727
          raise errors.OpPrereqError("Disk size change not possible, use"
5728
                                     " grow-disk")
5729

    
5730
    if disk_addremove > 1:
5731
      raise errors.OpPrereqError("Only one disk add or remove operation"
5732
                                 " supported at a time")
5733

    
5734
    # NIC validation
5735
    nic_addremove = 0
5736
    for nic_op, nic_dict in self.op.nics:
5737
      if nic_op == constants.DDM_REMOVE:
5738
        nic_addremove += 1
5739
        continue
5740
      elif nic_op == constants.DDM_ADD:
5741
        nic_addremove += 1
5742
      else:
5743
        if not isinstance(nic_op, int):
5744
          raise errors.OpPrereqError("Invalid nic index")
5745

    
5746
      # nic_dict should be a dict
5747
      nic_ip = nic_dict.get('ip', None)
5748
      if nic_ip is not None:
5749
        if nic_ip.lower() == constants.VALUE_NONE:
5750
          nic_dict['ip'] = None
5751
        else:
5752
          if not utils.IsValidIP(nic_ip):
5753
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5754

    
5755
      if nic_op == constants.DDM_ADD:
5756
        nic_bridge = nic_dict.get('bridge', None)
5757
        if nic_bridge is None:
5758
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5759
        nic_mac = nic_dict.get('mac', None)
5760
        if nic_mac is None:
5761
          nic_dict['mac'] = constants.VALUE_AUTO
5762

    
5763
      if 'mac' in nic_dict:
5764
        nic_mac = nic_dict['mac']
5765
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5766
          if not utils.IsValidMac(nic_mac):
5767
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5768
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5769
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5770
                                     " modifying an existing nic")
5771

    
5772
    if nic_addremove > 1:
5773
      raise errors.OpPrereqError("Only one NIC add or remove operation"
5774
                                 " supported at a time")
5775

    
5776
  def ExpandNames(self):
5777
    self._ExpandAndLockInstance()
5778
    self.needed_locks[locking.LEVEL_NODE] = []
5779
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5780

    
5781
  def DeclareLocks(self, level):
5782
    if level == locking.LEVEL_NODE:
5783
      self._LockInstancesNodes()
5784

    
5785
  def BuildHooksEnv(self):
5786
    """Build hooks env.
5787

5788
    This runs on the master, primary and secondaries.
5789

5790
    """
5791
    args = dict()
5792
    if constants.BE_MEMORY in self.be_new:
5793
      args['memory'] = self.be_new[constants.BE_MEMORY]
5794
    if constants.BE_VCPUS in self.be_new:
5795
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5796
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5797
    # information at all.
5798
    if self.op.nics:
5799
      args['nics'] = []
5800
      nic_override = dict(self.op.nics)
5801
      for idx, nic in enumerate(self.instance.nics):
5802
        if idx in nic_override:
5803
          this_nic_override = nic_override[idx]
5804
        else:
5805
          this_nic_override = {}
5806
        if 'ip' in this_nic_override:
5807
          ip = this_nic_override['ip']
5808
        else:
5809
          ip = nic.ip
5810
        if 'bridge' in this_nic_override:
5811
          bridge = this_nic_override['bridge']
5812
        else:
5813
          bridge = nic.bridge
5814
        if 'mac' in this_nic_override:
5815
          mac = this_nic_override['mac']
5816
        else:
5817
          mac = nic.mac
5818
        args['nics'].append((ip, bridge, mac))
5819
      if constants.DDM_ADD in nic_override:
5820
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5821
        bridge = nic_override[constants.DDM_ADD]['bridge']
5822
        mac = nic_override[constants.DDM_ADD]['mac']
5823
        args['nics'].append((ip, bridge, mac))
5824
      elif constants.DDM_REMOVE in nic_override:
5825
        del args['nics'][-1]
5826

    
5827
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5828
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5829
    return env, nl, nl
5830

    
5831
  def CheckPrereq(self):
5832
    """Check prerequisites.
5833

5834
    This only checks the instance list against the existing names.
5835

5836
    """
5837
    force = self.force = self.op.force
5838

    
5839
    # checking the new params on the primary/secondary nodes
5840

    
5841
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5842
    assert self.instance is not None, \
5843
      "Cannot retrieve locked instance %s" % self.op.instance_name
5844
    pnode = instance.primary_node
5845
    nodelist = list(instance.all_nodes)
5846

    
5847
    # hvparams processing
5848
    if self.op.hvparams:
5849
      i_hvdict = copy.deepcopy(instance.hvparams)
5850
      for key, val in self.op.hvparams.iteritems():
5851
        if val == constants.VALUE_DEFAULT:
5852
          try:
5853
            del i_hvdict[key]
5854
          except KeyError:
5855
            pass
5856
        else:
5857
          i_hvdict[key] = val
5858
      cluster = self.cfg.GetClusterInfo()
5859
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
5860
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
5861
                                i_hvdict)
5862
      # local check
5863
      hypervisor.GetHypervisor(
5864
        instance.hypervisor).CheckParameterSyntax(hv_new)
5865
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
5866
      self.hv_new = hv_new # the new actual values
5867
      self.hv_inst = i_hvdict # the new dict (without defaults)
5868
    else:
5869
      self.hv_new = self.hv_inst = {}
5870

    
5871
    # beparams processing
5872
    if self.op.beparams:
5873
      i_bedict = copy.deepcopy(instance.beparams)
5874
      for key, val in self.op.beparams.iteritems():
5875
        if val == constants.VALUE_DEFAULT:
5876
          try:
5877
            del i_bedict[key]
5878
          except KeyError:
5879
            pass
5880
        else:
5881
          i_bedict[key] = val
5882
      cluster = self.cfg.GetClusterInfo()
5883
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
5884
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
5885
                                i_bedict)
5886
      self.be_new = be_new # the new actual values
5887
      self.be_inst = i_bedict # the new dict (without defaults)
5888
    else:
5889
      self.be_new = self.be_inst = {}
5890

    
5891
    self.warn = []
5892

    
5893
    if constants.BE_MEMORY in self.op.beparams and not self.force:
5894
      mem_check_list = [pnode]
5895
      if be_new[constants.BE_AUTO_BALANCE]:
5896
        # either we changed auto_balance to yes or it was from before
5897
        mem_check_list.extend(instance.secondary_nodes)
5898
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
5899
                                                  instance.hypervisor)
5900
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
5901
                                         instance.hypervisor)
5902
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
5903
        # Assume the primary node is unreachable and go ahead
5904
        self.warn.append("Can't get info from primary node %s" % pnode)
5905
      else:
5906
        if not instance_info.failed and instance_info.data:
5907
          current_mem = instance_info.data['memory']
5908
        else:
5909
          # Assume instance not running
5910
          # (there is a slight race condition here, but it's not very probable,
5911
          # and we have no other way to check)
5912
          current_mem = 0
5913
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
5914
                    nodeinfo[pnode].data['memory_free'])
5915
        if miss_mem > 0:
5916
          raise errors.OpPrereqError("This change will prevent the instance"
5917
                                     " from starting, due to %d MB of memory"
5918
                                     " missing on its primary node" % miss_mem)
5919

    
5920
      if be_new[constants.BE_AUTO_BALANCE]:
5921
        for node, nres in nodeinfo.iteritems():
5922
          if node not in instance.secondary_nodes:
5923
            continue
5924
          if nres.failed or not isinstance(nres.data, dict):
5925
            self.warn.append("Can't get info from secondary node %s" % node)
5926
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
5927
            self.warn.append("Not enough memory to failover instance to"
5928
                             " secondary node %s" % node)
5929

    
5930
    # NIC processing
5931
    for nic_op, nic_dict in self.op.nics:
5932
      if nic_op == constants.DDM_REMOVE:
5933
        if not instance.nics:
5934
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
5935
        continue
5936
      if nic_op != constants.DDM_ADD:
5937
        # an existing nic
5938
        if nic_op < 0 or nic_op >= len(instance.nics):
5939
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
5940
                                     " are 0 to %d" %
5941
                                     (nic_op, len(instance.nics)))
5942
      if 'bridge' in nic_dict:
5943
        nic_bridge = nic_dict['bridge']
5944
        if nic_bridge is None:
5945
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
5946
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
5947
          msg = ("Bridge '%s' doesn't exist on one of"
5948
                 " the instance nodes" % nic_bridge)
5949
          if self.force:
5950
            self.warn.append(msg)
5951
          else:
5952
            raise errors.OpPrereqError(msg)
5953
      if 'mac' in nic_dict:
5954
        nic_mac = nic_dict['mac']
5955
        if nic_mac is None:
5956
          raise errors.OpPrereqError('Cannot set the nic mac to None')
5957
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5958
          # otherwise generate the mac
5959
          nic_dict['mac'] = self.cfg.GenerateMAC()
5960
        else:
5961
          # or validate/reserve the current one
5962
          if self.cfg.IsMacInUse(nic_mac):
5963
            raise errors.OpPrereqError("MAC address %s already in use"
5964
                                       " in cluster" % nic_mac)
5965

    
5966
    # DISK processing
5967
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
5968
      raise errors.OpPrereqError("Disk operations not supported for"
5969
                                 " diskless instances")
5970
    for disk_op, disk_dict in self.op.disks:
5971
      if disk_op == constants.DDM_REMOVE:
5972
        if len(instance.disks) == 1:
5973
          raise errors.OpPrereqError("Cannot remove the last disk of"
5974
                                     " an instance")
5975
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
5976
        ins_l = ins_l[pnode]
5977
        if ins_l.failed or not isinstance(ins_l.data, list):
5978
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
5979
        if instance.name in ins_l.data:
5980
          raise errors.OpPrereqError("Instance is running, can't remove"
5981
                                     " disks.")
5982

    
5983
      if (disk_op == constants.DDM_ADD and
5984
          len(instance.nics) >= constants.MAX_DISKS):
5985
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
5986
                                   " add more" % constants.MAX_DISKS)
5987
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
5988
        # an existing disk
5989
        if disk_op < 0 or disk_op >= len(instance.disks):
5990
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
5991
                                     " are 0 to %d" %
5992
                                     (disk_op, len(instance.disks)))
5993

    
5994
    return
5995

    
5996
  def Exec(self, feedback_fn):
5997
    """Modifies an instance.
5998

5999
    All parameters take effect only at the next restart of the instance.
6000

6001
    """
6002
    # Process here the warnings from CheckPrereq, as we don't have a
6003
    # feedback_fn there.
6004
    for warn in self.warn:
6005
      feedback_fn("WARNING: %s" % warn)
6006

    
6007
    result = []
6008
    instance = self.instance
6009
    # disk changes
6010
    for disk_op, disk_dict in self.op.disks:
6011
      if disk_op == constants.DDM_REMOVE:
6012
        # remove the last disk
6013
        device = instance.disks.pop()
6014
        device_idx = len(instance.disks)
6015
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6016
          self.cfg.SetDiskID(disk, node)
6017
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6018
          if msg:
6019
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6020
                            " continuing anyway", device_idx, node, msg)
6021
        result.append(("disk/%d" % device_idx, "remove"))
6022
      elif disk_op == constants.DDM_ADD:
6023
        # add a new disk
6024
        if instance.disk_template == constants.DT_FILE:
6025
          file_driver, file_path = instance.disks[0].logical_id
6026
          file_path = os.path.dirname(file_path)
6027
        else:
6028
          file_driver = file_path = None
6029
        disk_idx_base = len(instance.disks)
6030
        new_disk = _GenerateDiskTemplate(self,
6031
                                         instance.disk_template,
6032
                                         instance.name, instance.primary_node,
6033
                                         instance.secondary_nodes,
6034
                                         [disk_dict],
6035
                                         file_path,
6036
                                         file_driver,
6037
                                         disk_idx_base)[0]
6038
        instance.disks.append(new_disk)
6039
        info = _GetInstanceInfoText(instance)
6040

    
6041
        logging.info("Creating volume %s for instance %s",
6042
                     new_disk.iv_name, instance.name)
6043
        # Note: this needs to be kept in sync with _CreateDisks
6044
        #HARDCODE
6045
        for node in instance.all_nodes:
6046
          f_create = node == instance.primary_node
6047
          try:
6048
            _CreateBlockDev(self, node, instance, new_disk,
6049
                            f_create, info, f_create)
6050
          except errors.OpExecError, err:
6051
            self.LogWarning("Failed to create volume %s (%s) on"
6052
                            " node %s: %s",
6053
                            new_disk.iv_name, new_disk, node, err)
6054
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6055
                       (new_disk.size, new_disk.mode)))
6056
      else:
6057
        # change a given disk
6058
        instance.disks[disk_op].mode = disk_dict['mode']
6059
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6060
    # NIC changes
6061
    for nic_op, nic_dict in self.op.nics:
6062
      if nic_op == constants.DDM_REMOVE:
6063
        # remove the last nic
6064
        del instance.nics[-1]
6065
        result.append(("nic.%d" % len(instance.nics), "remove"))
6066
      elif nic_op == constants.DDM_ADD:
6067
        # mac and bridge should be set, by now
6068
        mac = nic_dict['mac']
6069
        bridge = nic_dict['bridge']
6070
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6071
                              bridge=bridge)
6072
        instance.nics.append(new_nic)
6073
        result.append(("nic.%d" % (len(instance.nics) - 1),
6074
                       "add:mac=%s,ip=%s,bridge=%s" %
6075
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6076
      else:
6077
        # change a given nic
6078
        for key in 'mac', 'ip', 'bridge':
6079
          if key in nic_dict:
6080
            setattr(instance.nics[nic_op], key, nic_dict[key])
6081
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6082

    
6083
    # hvparams changes
6084
    if self.op.hvparams:
6085
      instance.hvparams = self.hv_inst
6086
      for key, val in self.op.hvparams.iteritems():
6087
        result.append(("hv/%s" % key, val))
6088

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

    
6095
    self.cfg.Update(instance)
6096

    
6097
    return result
6098

    
6099

    
6100
class LUQueryExports(NoHooksLU):
6101
  """Query the exports list
6102

6103
  """
6104
  _OP_REQP = ['nodes']
6105
  REQ_BGL = False
6106

    
6107
  def ExpandNames(self):
6108
    self.needed_locks = {}
6109
    self.share_locks[locking.LEVEL_NODE] = 1
6110
    if not self.op.nodes:
6111
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6112
    else:
6113
      self.needed_locks[locking.LEVEL_NODE] = \
6114
        _GetWantedNodes(self, self.op.nodes)
6115

    
6116
  def CheckPrereq(self):
6117
    """Check prerequisites.
6118

6119
    """
6120
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6121

    
6122
  def Exec(self, feedback_fn):
6123
    """Compute the list of all the exported system images.
6124

6125
    @rtype: dict
6126
    @return: a dictionary with the structure node->(export-list)
6127
        where export-list is a list of the instances exported on
6128
        that node.
6129

6130
    """
6131
    rpcresult = self.rpc.call_export_list(self.nodes)
6132
    result = {}
6133
    for node in rpcresult:
6134
      if rpcresult[node].failed:
6135
        result[node] = False
6136
      else:
6137
        result[node] = rpcresult[node].data
6138

    
6139
    return result
6140

    
6141

    
6142
class LUExportInstance(LogicalUnit):
6143
  """Export an instance to an image in the cluster.
6144

6145
  """
6146
  HPATH = "instance-export"
6147
  HTYPE = constants.HTYPE_INSTANCE
6148
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6149
  REQ_BGL = False
6150

    
6151
  def ExpandNames(self):
6152
    self._ExpandAndLockInstance()
6153
    # FIXME: lock only instance primary and destination node
6154
    #
6155
    # Sad but true, for now we have do lock all nodes, as we don't know where
6156
    # the previous export might be, and and in this LU we search for it and
6157
    # remove it from its current node. In the future we could fix this by:
6158
    #  - making a tasklet to search (share-lock all), then create the new one,
6159
    #    then one to remove, after
6160
    #  - removing the removal operation altoghether
6161
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6162

    
6163
  def DeclareLocks(self, level):
6164
    """Last minute lock declaration."""
6165
    # All nodes are locked anyway, so nothing to do here.
6166

    
6167
  def BuildHooksEnv(self):
6168
    """Build hooks env.
6169

6170
    This will run on the master, primary node and target node.
6171

6172
    """
6173
    env = {
6174
      "EXPORT_NODE": self.op.target_node,
6175
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6176
      }
6177
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6178
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6179
          self.op.target_node]
6180
    return env, nl, nl
6181

    
6182
  def CheckPrereq(self):
6183
    """Check prerequisites.
6184

6185
    This checks that the instance and node names are valid.
6186

6187
    """
6188
    instance_name = self.op.instance_name
6189
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6190
    assert self.instance is not None, \
6191
          "Cannot retrieve locked instance %s" % self.op.instance_name
6192
    _CheckNodeOnline(self, self.instance.primary_node)
6193

    
6194
    self.dst_node = self.cfg.GetNodeInfo(
6195
      self.cfg.ExpandNodeName(self.op.target_node))
6196

    
6197
    if self.dst_node is None:
6198
      # This is wrong node name, not a non-locked node
6199
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6200
    _CheckNodeOnline(self, self.dst_node.name)
6201
    _CheckNodeNotDrained(self, self.dst_node.name)
6202

    
6203
    # instance disk type verification
6204
    for disk in self.instance.disks:
6205
      if disk.dev_type == constants.LD_FILE:
6206
        raise errors.OpPrereqError("Export not supported for instances with"
6207
                                   " file-based disks")
6208

    
6209
  def Exec(self, feedback_fn):
6210
    """Export an instance to an image in the cluster.
6211

6212
    """
6213
    instance = self.instance
6214
    dst_node = self.dst_node
6215
    src_node = instance.primary_node
6216
    if self.op.shutdown:
6217
      # shutdown the instance, but not the disks
6218
      result = self.rpc.call_instance_shutdown(src_node, instance)
6219
      msg = result.RemoteFailMsg()
6220
      if msg:
6221
        raise errors.OpExecError("Could not shutdown instance %s on"
6222
                                 " node %s: %s" %
6223
                                 (instance.name, src_node, msg))
6224

    
6225
    vgname = self.cfg.GetVGName()
6226

    
6227
    snap_disks = []
6228

    
6229
    # set the disks ID correctly since call_instance_start needs the
6230
    # correct drbd minor to create the symlinks
6231
    for disk in instance.disks:
6232
      self.cfg.SetDiskID(disk, src_node)
6233

    
6234
    try:
6235
      for disk in instance.disks:
6236
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6237
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6238
        if new_dev_name.failed or not new_dev_name.data:
6239
          self.LogWarning("Could not snapshot block device %s on node %s",
6240
                          disk.logical_id[1], src_node)
6241
          snap_disks.append(False)
6242
        else:
6243
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6244
                                 logical_id=(vgname, new_dev_name.data),
6245
                                 physical_id=(vgname, new_dev_name.data),
6246
                                 iv_name=disk.iv_name)
6247
          snap_disks.append(new_dev)
6248

    
6249
    finally:
6250
      if self.op.shutdown and instance.admin_up:
6251
        result = self.rpc.call_instance_start(src_node, instance)
6252
        msg = result.RemoteFailMsg()
6253
        if msg:
6254
          _ShutdownInstanceDisks(self, instance)
6255
          raise errors.OpExecError("Could not start instance: %s" % msg)
6256

    
6257
    # TODO: check for size
6258

    
6259
    cluster_name = self.cfg.GetClusterName()
6260
    for idx, dev in enumerate(snap_disks):
6261
      if dev:
6262
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6263
                                               instance, cluster_name, idx)
6264
        if result.failed or not result.data:
6265
          self.LogWarning("Could not export block device %s from node %s to"
6266
                          " node %s", dev.logical_id[1], src_node,
6267
                          dst_node.name)
6268
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6269
        if msg:
6270
          self.LogWarning("Could not remove snapshot block device %s from node"
6271
                          " %s: %s", dev.logical_id[1], src_node, msg)
6272

    
6273
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6274
    if result.failed or not result.data:
6275
      self.LogWarning("Could not finalize export for instance %s on node %s",
6276
                      instance.name, dst_node.name)
6277

    
6278
    nodelist = self.cfg.GetNodeList()
6279
    nodelist.remove(dst_node.name)
6280

    
6281
    # on one-node clusters nodelist will be empty after the removal
6282
    # if we proceed the backup would be removed because OpQueryExports
6283
    # substitutes an empty list with the full cluster node list.
6284
    if nodelist:
6285
      exportlist = self.rpc.call_export_list(nodelist)
6286
      for node in exportlist:
6287
        if exportlist[node].failed:
6288
          continue
6289
        if instance.name in exportlist[node].data:
6290
          if not self.rpc.call_export_remove(node, instance.name):
6291
            self.LogWarning("Could not remove older export for instance %s"
6292
                            " on node %s", instance.name, node)
6293

    
6294

    
6295
class LURemoveExport(NoHooksLU):
6296
  """Remove exports related to the named instance.
6297

6298
  """
6299
  _OP_REQP = ["instance_name"]
6300
  REQ_BGL = False
6301

    
6302
  def ExpandNames(self):
6303
    self.needed_locks = {}
6304
    # We need all nodes to be locked in order for RemoveExport to work, but we
6305
    # don't need to lock the instance itself, as nothing will happen to it (and
6306
    # we can remove exports also for a removed instance)
6307
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6308

    
6309
  def CheckPrereq(self):
6310
    """Check prerequisites.
6311
    """
6312
    pass
6313

    
6314
  def Exec(self, feedback_fn):
6315
    """Remove any export.
6316

6317
    """
6318
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6319
    # If the instance was not found we'll try with the name that was passed in.
6320
    # This will only work if it was an FQDN, though.
6321
    fqdn_warn = False
6322
    if not instance_name:
6323
      fqdn_warn = True
6324
      instance_name = self.op.instance_name
6325

    
6326
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6327
      locking.LEVEL_NODE])
6328
    found = False
6329
    for node in exportlist:
6330
      if exportlist[node].failed:
6331
        self.LogWarning("Failed to query node %s, continuing" % node)
6332
        continue
6333
      if instance_name in exportlist[node].data:
6334
        found = True
6335
        result = self.rpc.call_export_remove(node, instance_name)
6336
        if result.failed or not result.data:
6337
          logging.error("Could not remove export for instance %s"
6338
                        " on node %s", instance_name, node)
6339

    
6340
    if fqdn_warn and not found:
6341
      feedback_fn("Export not found. If trying to remove an export belonging"
6342
                  " to a deleted instance please use its Fully Qualified"
6343
                  " Domain Name.")
6344

    
6345

    
6346
class TagsLU(NoHooksLU):
6347
  """Generic tags LU.
6348

6349
  This is an abstract class which is the parent of all the other tags LUs.
6350

6351
  """
6352

    
6353
  def ExpandNames(self):
6354
    self.needed_locks = {}
6355
    if self.op.kind == constants.TAG_NODE:
6356
      name = self.cfg.ExpandNodeName(self.op.name)
6357
      if name is None:
6358
        raise errors.OpPrereqError("Invalid node name (%s)" %
6359
                                   (self.op.name,))
6360
      self.op.name = name
6361
      self.needed_locks[locking.LEVEL_NODE] = name
6362
    elif self.op.kind == constants.TAG_INSTANCE:
6363
      name = self.cfg.ExpandInstanceName(self.op.name)
6364
      if name is None:
6365
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6366
                                   (self.op.name,))
6367
      self.op.name = name
6368
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6369

    
6370
  def CheckPrereq(self):
6371
    """Check prerequisites.
6372

6373
    """
6374
    if self.op.kind == constants.TAG_CLUSTER:
6375
      self.target = self.cfg.GetClusterInfo()
6376
    elif self.op.kind == constants.TAG_NODE:
6377
      self.target = self.cfg.GetNodeInfo(self.op.name)
6378
    elif self.op.kind == constants.TAG_INSTANCE:
6379
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6380
    else:
6381
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6382
                                 str(self.op.kind))
6383

    
6384

    
6385
class LUGetTags(TagsLU):
6386
  """Returns the tags of a given object.
6387

6388
  """
6389
  _OP_REQP = ["kind", "name"]
6390
  REQ_BGL = False
6391

    
6392
  def Exec(self, feedback_fn):
6393
    """Returns the tag list.
6394

6395
    """
6396
    return list(self.target.GetTags())
6397

    
6398

    
6399
class LUSearchTags(NoHooksLU):
6400
  """Searches the tags for a given pattern.
6401

6402
  """
6403
  _OP_REQP = ["pattern"]
6404
  REQ_BGL = False
6405

    
6406
  def ExpandNames(self):
6407
    self.needed_locks = {}
6408

    
6409
  def CheckPrereq(self):
6410
    """Check prerequisites.
6411

6412
    This checks the pattern passed for validity by compiling it.
6413

6414
    """
6415
    try:
6416
      self.re = re.compile(self.op.pattern)
6417
    except re.error, err:
6418
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6419
                                 (self.op.pattern, err))
6420

    
6421
  def Exec(self, feedback_fn):
6422
    """Returns the tag list.
6423

6424
    """
6425
    cfg = self.cfg
6426
    tgts = [("/cluster", cfg.GetClusterInfo())]
6427
    ilist = cfg.GetAllInstancesInfo().values()
6428
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6429
    nlist = cfg.GetAllNodesInfo().values()
6430
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6431
    results = []
6432
    for path, target in tgts:
6433
      for tag in target.GetTags():
6434
        if self.re.search(tag):
6435
          results.append((path, tag))
6436
    return results
6437

    
6438

    
6439
class LUAddTags(TagsLU):
6440
  """Sets a tag on a given object.
6441

6442
  """
6443
  _OP_REQP = ["kind", "name", "tags"]
6444
  REQ_BGL = False
6445

    
6446
  def CheckPrereq(self):
6447
    """Check prerequisites.
6448

6449
    This checks the type and length of the tag name and value.
6450

6451
    """
6452
    TagsLU.CheckPrereq(self)
6453
    for tag in self.op.tags:
6454
      objects.TaggableObject.ValidateTag(tag)
6455

    
6456
  def Exec(self, feedback_fn):
6457
    """Sets the tag.
6458

6459
    """
6460
    try:
6461
      for tag in self.op.tags:
6462
        self.target.AddTag(tag)
6463
    except errors.TagError, err:
6464
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6465
    try:
6466
      self.cfg.Update(self.target)
6467
    except errors.ConfigurationError:
6468
      raise errors.OpRetryError("There has been a modification to the"
6469
                                " config file and the operation has been"
6470
                                " aborted. Please retry.")
6471

    
6472

    
6473
class LUDelTags(TagsLU):
6474
  """Delete a list of tags from a given object.
6475

6476
  """
6477
  _OP_REQP = ["kind", "name", "tags"]
6478
  REQ_BGL = False
6479

    
6480
  def CheckPrereq(self):
6481
    """Check prerequisites.
6482

6483
    This checks that we have the given tag.
6484

6485
    """
6486
    TagsLU.CheckPrereq(self)
6487
    for tag in self.op.tags:
6488
      objects.TaggableObject.ValidateTag(tag)
6489
    del_tags = frozenset(self.op.tags)
6490
    cur_tags = self.target.GetTags()
6491
    if not del_tags <= cur_tags:
6492
      diff_tags = del_tags - cur_tags
6493
      diff_names = ["'%s'" % tag for tag in diff_tags]
6494
      diff_names.sort()
6495
      raise errors.OpPrereqError("Tag(s) %s not found" %
6496
                                 (",".join(diff_names)))
6497

    
6498
  def Exec(self, feedback_fn):
6499
    """Remove the tag from the object.
6500

6501
    """
6502
    for tag in self.op.tags:
6503
      self.target.RemoveTag(tag)
6504
    try:
6505
      self.cfg.Update(self.target)
6506
    except errors.ConfigurationError:
6507
      raise errors.OpRetryError("There has been a modification to the"
6508
                                " config file and the operation has been"
6509
                                " aborted. Please retry.")
6510

    
6511

    
6512
class LUTestDelay(NoHooksLU):
6513
  """Sleep for a specified amount of time.
6514

6515
  This LU sleeps on the master and/or nodes for a specified amount of
6516
  time.
6517

6518
  """
6519
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6520
  REQ_BGL = False
6521

    
6522
  def ExpandNames(self):
6523
    """Expand names and set required locks.
6524

6525
    This expands the node list, if any.
6526

6527
    """
6528
    self.needed_locks = {}
6529
    if self.op.on_nodes:
6530
      # _GetWantedNodes can be used here, but is not always appropriate to use
6531
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6532
      # more information.
6533
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6534
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6535

    
6536
  def CheckPrereq(self):
6537
    """Check prerequisites.
6538

6539
    """
6540

    
6541
  def Exec(self, feedback_fn):
6542
    """Do the actual sleep.
6543

6544
    """
6545
    if self.op.on_master:
6546
      if not utils.TestDelay(self.op.duration):
6547
        raise errors.OpExecError("Error during master delay test")
6548
    if self.op.on_nodes:
6549
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6550
      if not result:
6551
        raise errors.OpExecError("Complete failure from rpc call")
6552
      for node, node_result in result.items():
6553
        node_result.Raise()
6554
        if not node_result.data:
6555
          raise errors.OpExecError("Failure during rpc call to node %s,"
6556
                                   " result: %s" % (node, node_result.data))
6557

    
6558

    
6559
class IAllocator(object):
6560
  """IAllocator framework.
6561

6562
  An IAllocator instance has three sets of attributes:
6563
    - cfg that is needed to query the cluster
6564
    - input data (all members of the _KEYS class attribute are required)
6565
    - four buffer attributes (in|out_data|text), that represent the
6566
      input (to the external script) in text and data structure format,
6567
      and the output from it, again in two formats
6568
    - the result variables from the script (success, info, nodes) for
6569
      easy usage
6570

6571
  """
6572
  _ALLO_KEYS = [
6573
    "mem_size", "disks", "disk_template",
6574
    "os", "tags", "nics", "vcpus", "hypervisor",
6575
    ]
6576
  _RELO_KEYS = [
6577
    "relocate_from",
6578
    ]
6579

    
6580
  def __init__(self, lu, mode, name, **kwargs):
6581
    self.lu = lu
6582
    # init buffer variables
6583
    self.in_text = self.out_text = self.in_data = self.out_data = None
6584
    # init all input fields so that pylint is happy
6585
    self.mode = mode
6586
    self.name = name
6587
    self.mem_size = self.disks = self.disk_template = None
6588
    self.os = self.tags = self.nics = self.vcpus = None
6589
    self.hypervisor = None
6590
    self.relocate_from = None
6591
    # computed fields
6592
    self.required_nodes = None
6593
    # init result fields
6594
    self.success = self.info = self.nodes = None
6595
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6596
      keyset = self._ALLO_KEYS
6597
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6598
      keyset = self._RELO_KEYS
6599
    else:
6600
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6601
                                   " IAllocator" % self.mode)
6602
    for key in kwargs:
6603
      if key not in keyset:
6604
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6605
                                     " IAllocator" % key)
6606
      setattr(self, key, kwargs[key])
6607
    for key in keyset:
6608
      if key not in kwargs:
6609
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6610
                                     " IAllocator" % key)
6611
    self._BuildInputData()
6612

    
6613
  def _ComputeClusterData(self):
6614
    """Compute the generic allocator input data.
6615

6616
    This is the data that is independent of the actual operation.
6617

6618
    """
6619
    cfg = self.lu.cfg
6620
    cluster_info = cfg.GetClusterInfo()
6621
    # cluster data
6622
    data = {
6623
      "version": constants.IALLOCATOR_VERSION,
6624
      "cluster_name": cfg.GetClusterName(),
6625
      "cluster_tags": list(cluster_info.GetTags()),
6626
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6627
      # we don't have job IDs
6628
      }
6629
    iinfo = cfg.GetAllInstancesInfo().values()
6630
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6631

    
6632
    # node data
6633
    node_results = {}
6634
    node_list = cfg.GetNodeList()
6635

    
6636
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6637
      hypervisor_name = self.hypervisor
6638
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6639
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6640

    
6641
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6642
                                           hypervisor_name)
6643
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6644
                       cluster_info.enabled_hypervisors)
6645
    for nname, nresult in node_data.items():
6646
      # first fill in static (config-based) values
6647
      ninfo = cfg.GetNodeInfo(nname)
6648
      pnr = {
6649
        "tags": list(ninfo.GetTags()),
6650
        "primary_ip": ninfo.primary_ip,
6651
        "secondary_ip": ninfo.secondary_ip,
6652
        "offline": ninfo.offline,
6653
        "drained": ninfo.drained,
6654
        "master_candidate": ninfo.master_candidate,
6655
        }
6656

    
6657
      if not ninfo.offline:
6658
        nresult.Raise()
6659
        if not isinstance(nresult.data, dict):
6660
          raise errors.OpExecError("Can't get data for node %s" % nname)
6661
        remote_info = nresult.data
6662
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6663
                     'vg_size', 'vg_free', 'cpu_total']:
6664
          if attr not in remote_info:
6665
            raise errors.OpExecError("Node '%s' didn't return attribute"
6666
                                     " '%s'" % (nname, attr))
6667
          try:
6668
            remote_info[attr] = int(remote_info[attr])
6669
          except ValueError, err:
6670
            raise errors.OpExecError("Node '%s' returned invalid value"
6671
                                     " for '%s': %s" % (nname, attr, err))
6672
        # compute memory used by primary instances
6673
        i_p_mem = i_p_up_mem = 0
6674
        for iinfo, beinfo in i_list:
6675
          if iinfo.primary_node == nname:
6676
            i_p_mem += beinfo[constants.BE_MEMORY]
6677
            if iinfo.name not in node_iinfo[nname].data:
6678
              i_used_mem = 0
6679
            else:
6680
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6681
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6682
            remote_info['memory_free'] -= max(0, i_mem_diff)
6683

    
6684
            if iinfo.admin_up:
6685
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6686

    
6687
        # compute memory used by instances
6688
        pnr_dyn = {
6689
          "total_memory": remote_info['memory_total'],
6690
          "reserved_memory": remote_info['memory_dom0'],
6691
          "free_memory": remote_info['memory_free'],
6692
          "total_disk": remote_info['vg_size'],
6693
          "free_disk": remote_info['vg_free'],
6694
          "total_cpus": remote_info['cpu_total'],
6695
          "i_pri_memory": i_p_mem,
6696
          "i_pri_up_memory": i_p_up_mem,
6697
          }
6698
        pnr.update(pnr_dyn)
6699

    
6700
      node_results[nname] = pnr
6701
    data["nodes"] = node_results
6702

    
6703
    # instance data
6704
    instance_data = {}
6705
    for iinfo, beinfo in i_list:
6706
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6707
                  for n in iinfo.nics]
6708
      pir = {
6709
        "tags": list(iinfo.GetTags()),
6710
        "admin_up": iinfo.admin_up,
6711
        "vcpus": beinfo[constants.BE_VCPUS],
6712
        "memory": beinfo[constants.BE_MEMORY],
6713
        "os": iinfo.os,
6714
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6715
        "nics": nic_data,
6716
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6717
        "disk_template": iinfo.disk_template,
6718
        "hypervisor": iinfo.hypervisor,
6719
        }
6720
      instance_data[iinfo.name] = pir
6721

    
6722
    data["instances"] = instance_data
6723

    
6724
    self.in_data = data
6725

    
6726
  def _AddNewInstance(self):
6727
    """Add new instance data to allocator structure.
6728

6729
    This in combination with _AllocatorGetClusterData will create the
6730
    correct structure needed as input for the allocator.
6731

6732
    The checks for the completeness of the opcode must have already been
6733
    done.
6734

6735
    """
6736
    data = self.in_data
6737

    
6738
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6739

    
6740
    if self.disk_template in constants.DTS_NET_MIRROR:
6741
      self.required_nodes = 2
6742
    else:
6743
      self.required_nodes = 1
6744
    request = {
6745
      "type": "allocate",
6746
      "name": self.name,
6747
      "disk_template": self.disk_template,
6748
      "tags": self.tags,
6749
      "os": self.os,
6750
      "vcpus": self.vcpus,
6751
      "memory": self.mem_size,
6752
      "disks": self.disks,
6753
      "disk_space_total": disk_space,
6754
      "nics": self.nics,
6755
      "required_nodes": self.required_nodes,
6756
      }
6757
    data["request"] = request
6758

    
6759
  def _AddRelocateInstance(self):
6760
    """Add relocate instance data to allocator structure.
6761

6762
    This in combination with _IAllocatorGetClusterData will create the
6763
    correct structure needed as input for the allocator.
6764

6765
    The checks for the completeness of the opcode must have already been
6766
    done.
6767

6768
    """
6769
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6770
    if instance is None:
6771
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6772
                                   " IAllocator" % self.name)
6773

    
6774
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6775
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6776

    
6777
    if len(instance.secondary_nodes) != 1:
6778
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6779

    
6780
    self.required_nodes = 1
6781
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6782
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6783

    
6784
    request = {
6785
      "type": "relocate",
6786
      "name": self.name,
6787
      "disk_space_total": disk_space,
6788
      "required_nodes": self.required_nodes,
6789
      "relocate_from": self.relocate_from,
6790
      }
6791
    self.in_data["request"] = request
6792

    
6793
  def _BuildInputData(self):
6794
    """Build input data structures.
6795

6796
    """
6797
    self._ComputeClusterData()
6798

    
6799
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6800
      self._AddNewInstance()
6801
    else:
6802
      self._AddRelocateInstance()
6803

    
6804
    self.in_text = serializer.Dump(self.in_data)
6805

    
6806
  def Run(self, name, validate=True, call_fn=None):
6807
    """Run an instance allocator and return the results.
6808

6809
    """
6810
    if call_fn is None:
6811
      call_fn = self.lu.rpc.call_iallocator_runner
6812
    data = self.in_text
6813

    
6814
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6815
    result.Raise()
6816

    
6817
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6818
      raise errors.OpExecError("Invalid result from master iallocator runner")
6819

    
6820
    rcode, stdout, stderr, fail = result.data
6821

    
6822
    if rcode == constants.IARUN_NOTFOUND:
6823
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6824
    elif rcode == constants.IARUN_FAILURE:
6825
      raise errors.OpExecError("Instance allocator call failed: %s,"
6826
                               " output: %s" % (fail, stdout+stderr))
6827
    self.out_text = stdout
6828
    if validate:
6829
      self._ValidateResult()
6830

    
6831
  def _ValidateResult(self):
6832
    """Process the allocator results.
6833

6834
    This will process and if successful save the result in
6835
    self.out_data and the other parameters.
6836

6837
    """
6838
    try:
6839
      rdict = serializer.Load(self.out_text)
6840
    except Exception, err:
6841
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6842

    
6843
    if not isinstance(rdict, dict):
6844
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6845

    
6846
    for key in "success", "info", "nodes":
6847
      if key not in rdict:
6848
        raise errors.OpExecError("Can't parse iallocator results:"
6849
                                 " missing key '%s'" % key)
6850
      setattr(self, key, rdict[key])
6851

    
6852
    if not isinstance(rdict["nodes"], list):
6853
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6854
                               " is not a list")
6855
    self.out_data = rdict
6856

    
6857

    
6858
class LUTestAllocator(NoHooksLU):
6859
  """Run allocator tests.
6860

6861
  This LU runs the allocator tests
6862

6863
  """
6864
  _OP_REQP = ["direction", "mode", "name"]
6865

    
6866
  def CheckPrereq(self):
6867
    """Check prerequisites.
6868

6869
    This checks the opcode parameters depending on the director and mode test.
6870

6871
    """
6872
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6873
      for attr in ["name", "mem_size", "disks", "disk_template",
6874
                   "os", "tags", "nics", "vcpus"]:
6875
        if not hasattr(self.op, attr):
6876
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
6877
                                     attr)
6878
      iname = self.cfg.ExpandInstanceName(self.op.name)
6879
      if iname is not None:
6880
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
6881
                                   iname)
6882
      if not isinstance(self.op.nics, list):
6883
        raise errors.OpPrereqError("Invalid parameter 'nics'")
6884
      for row in self.op.nics:
6885
        if (not isinstance(row, dict) or
6886
            "mac" not in row or
6887
            "ip" not in row or
6888
            "bridge" not in row):
6889
          raise errors.OpPrereqError("Invalid contents of the"
6890
                                     " 'nics' parameter")
6891
      if not isinstance(self.op.disks, list):
6892
        raise errors.OpPrereqError("Invalid parameter 'disks'")
6893
      for row in self.op.disks:
6894
        if (not isinstance(row, dict) or
6895
            "size" not in row or
6896
            not isinstance(row["size"], int) or
6897
            "mode" not in row or
6898
            row["mode"] not in ['r', 'w']):
6899
          raise errors.OpPrereqError("Invalid contents of the"
6900
                                     " 'disks' parameter")
6901
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
6902
        self.op.hypervisor = self.cfg.GetHypervisorType()
6903
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
6904
      if not hasattr(self.op, "name"):
6905
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
6906
      fname = self.cfg.ExpandInstanceName(self.op.name)
6907
      if fname is None:
6908
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
6909
                                   self.op.name)
6910
      self.op.name = fname
6911
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
6912
    else:
6913
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
6914
                                 self.op.mode)
6915

    
6916
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
6917
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
6918
        raise errors.OpPrereqError("Missing allocator name")
6919
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
6920
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
6921
                                 self.op.direction)
6922

    
6923
  def Exec(self, feedback_fn):
6924
    """Run the allocator test.
6925

6926
    """
6927
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6928
      ial = IAllocator(self,
6929
                       mode=self.op.mode,
6930
                       name=self.op.name,
6931
                       mem_size=self.op.mem_size,
6932
                       disks=self.op.disks,
6933
                       disk_template=self.op.disk_template,
6934
                       os=self.op.os,
6935
                       tags=self.op.tags,
6936
                       nics=self.op.nics,
6937
                       vcpus=self.op.vcpus,
6938
                       hypervisor=self.op.hypervisor,
6939
                       )
6940
    else:
6941
      ial = IAllocator(self,
6942
                       mode=self.op.mode,
6943
                       name=self.op.name,
6944
                       relocate_from=list(self.relocate_from),
6945
                       )
6946

    
6947
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
6948
      result = ial.in_text
6949
    else:
6950
      ial.Run(self.op.allocator, validate=False)
6951
      result = ial.out_text
6952
    return result