Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ 2cc1da8b

History | View | Annotate | Download (245.5 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
         msg = to_result.RemoteFailMsg()
1385
         if msg:
1386
           msg = ("Copy of file %s to node %s failed: %s" %
1387
                   (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
1388
           self.proc.LogWarning(msg)
1389

    
1390
    finally:
1391
      result = self.rpc.call_node_start_master(master, False)
1392
      if result.failed or not result.data:
1393
        self.LogWarning("Could not re-enable the master role on"
1394
                        " the master, please restart manually.")
1395

    
1396

    
1397
def _RecursiveCheckIfLVMBased(disk):
1398
  """Check if the given disk or its children are lvm-based.
1399

1400
  @type disk: L{objects.Disk}
1401
  @param disk: the disk to check
1402
  @rtype: booleean
1403
  @return: boolean indicating whether a LD_LV dev_type was found or not
1404

1405
  """
1406
  if disk.children:
1407
    for chdisk in disk.children:
1408
      if _RecursiveCheckIfLVMBased(chdisk):
1409
        return True
1410
  return disk.dev_type == constants.LD_LV
1411

    
1412

    
1413
class LUSetClusterParams(LogicalUnit):
1414
  """Change the parameters of the cluster.
1415

1416
  """
1417
  HPATH = "cluster-modify"
1418
  HTYPE = constants.HTYPE_CLUSTER
1419
  _OP_REQP = []
1420
  REQ_BGL = False
1421

    
1422
  def CheckArguments(self):
1423
    """Check parameters
1424

1425
    """
1426
    if not hasattr(self.op, "candidate_pool_size"):
1427
      self.op.candidate_pool_size = None
1428
    if self.op.candidate_pool_size is not None:
1429
      try:
1430
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
1431
      except (ValueError, TypeError), err:
1432
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
1433
                                   str(err))
1434
      if self.op.candidate_pool_size < 1:
1435
        raise errors.OpPrereqError("At least one master candidate needed")
1436

    
1437
  def ExpandNames(self):
1438
    # FIXME: in the future maybe other cluster params won't require checking on
1439
    # all nodes to be modified.
1440
    self.needed_locks = {
1441
      locking.LEVEL_NODE: locking.ALL_SET,
1442
    }
1443
    self.share_locks[locking.LEVEL_NODE] = 1
1444

    
1445
  def BuildHooksEnv(self):
1446
    """Build hooks env.
1447

1448
    """
1449
    env = {
1450
      "OP_TARGET": self.cfg.GetClusterName(),
1451
      "NEW_VG_NAME": self.op.vg_name,
1452
      }
1453
    mn = self.cfg.GetMasterNode()
1454
    return env, [mn], [mn]
1455

    
1456
  def CheckPrereq(self):
1457
    """Check prerequisites.
1458

1459
    This checks whether the given params don't conflict and
1460
    if the given volume group is valid.
1461

1462
    """
1463
    if self.op.vg_name is not None and not self.op.vg_name:
1464
      instances = self.cfg.GetAllInstancesInfo().values()
1465
      for inst in instances:
1466
        for disk in inst.disks:
1467
          if _RecursiveCheckIfLVMBased(disk):
1468
            raise errors.OpPrereqError("Cannot disable lvm storage while"
1469
                                       " lvm-based instances exist")
1470

    
1471
    node_list = self.acquired_locks[locking.LEVEL_NODE]
1472

    
1473
    # if vg_name not None, checks given volume group on all nodes
1474
    if self.op.vg_name:
1475
      vglist = self.rpc.call_vg_list(node_list)
1476
      for node in node_list:
1477
        if vglist[node].failed:
1478
          # ignoring down node
1479
          self.LogWarning("Node %s unreachable/error, ignoring" % node)
1480
          continue
1481
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].data,
1482
                                              self.op.vg_name,
1483
                                              constants.MIN_VG_SIZE)
1484
        if vgstatus:
1485
          raise errors.OpPrereqError("Error on node '%s': %s" %
1486
                                     (node, vgstatus))
1487

    
1488
    self.cluster = cluster = self.cfg.GetClusterInfo()
1489
    # validate beparams changes
1490
    if self.op.beparams:
1491
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
1492
      self.new_beparams = cluster.FillDict(
1493
        cluster.beparams[constants.BEGR_DEFAULT], self.op.beparams)
1494

    
1495
    # hypervisor list/parameters
1496
    self.new_hvparams = cluster.FillDict(cluster.hvparams, {})
1497
    if self.op.hvparams:
1498
      if not isinstance(self.op.hvparams, dict):
1499
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input")
1500
      for hv_name, hv_dict in self.op.hvparams.items():
1501
        if hv_name not in self.new_hvparams:
1502
          self.new_hvparams[hv_name] = hv_dict
1503
        else:
1504
          self.new_hvparams[hv_name].update(hv_dict)
1505

    
1506
    if self.op.enabled_hypervisors is not None:
1507
      self.hv_list = self.op.enabled_hypervisors
1508
    else:
1509
      self.hv_list = cluster.enabled_hypervisors
1510

    
1511
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
1512
      # either the enabled list has changed, or the parameters have, validate
1513
      for hv_name, hv_params in self.new_hvparams.items():
1514
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
1515
            (self.op.enabled_hypervisors and
1516
             hv_name in self.op.enabled_hypervisors)):
1517
          # either this is a new hypervisor, or its parameters have changed
1518
          hv_class = hypervisor.GetHypervisor(hv_name)
1519
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
1520
          hv_class.CheckParameterSyntax(hv_params)
1521
          _CheckHVParams(self, node_list, hv_name, hv_params)
1522

    
1523
  def Exec(self, feedback_fn):
1524
    """Change the parameters of the cluster.
1525

1526
    """
1527
    if self.op.vg_name is not None:
1528
      new_volume = self.op.vg_name
1529
      if not new_volume:
1530
        new_volume = None
1531
      if new_volume != self.cfg.GetVGName():
1532
        self.cfg.SetVGName(new_volume)
1533
      else:
1534
        feedback_fn("Cluster LVM configuration already in desired"
1535
                    " state, not changing")
1536
    if self.op.hvparams:
1537
      self.cluster.hvparams = self.new_hvparams
1538
    if self.op.enabled_hypervisors is not None:
1539
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
1540
    if self.op.beparams:
1541
      self.cluster.beparams[constants.BEGR_DEFAULT] = self.new_beparams
1542
    if self.op.candidate_pool_size is not None:
1543
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
1544

    
1545
    self.cfg.Update(self.cluster)
1546

    
1547
    # we want to update nodes after the cluster so that if any errors
1548
    # happen, we have recorded and saved the cluster info
1549
    if self.op.candidate_pool_size is not None:
1550
      _AdjustCandidatePool(self)
1551

    
1552

    
1553
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
1554
  """Distribute additional files which are part of the cluster configuration.
1555

1556
  ConfigWriter takes care of distributing the config and ssconf files, but
1557
  there are more files which should be distributed to all nodes. This function
1558
  makes sure those are copied.
1559

1560
  @param lu: calling logical unit
1561
  @param additional_nodes: list of nodes not in the config to distribute to
1562

1563
  """
1564
  # 1. Gather target nodes
1565
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
1566
  dist_nodes = lu.cfg.GetNodeList()
1567
  if additional_nodes is not None:
1568
    dist_nodes.extend(additional_nodes)
1569
  if myself.name in dist_nodes:
1570
    dist_nodes.remove(myself.name)
1571
  # 2. Gather files to distribute
1572
  dist_files = set([constants.ETC_HOSTS,
1573
                    constants.SSH_KNOWN_HOSTS_FILE,
1574
                    constants.RAPI_CERT_FILE,
1575
                    constants.RAPI_USERS_FILE,
1576
                   ])
1577

    
1578
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
1579
  for hv_name in enabled_hypervisors:
1580
    hv_class = hypervisor.GetHypervisor(hv_name)
1581
    dist_files.update(hv_class.GetAncillaryFiles())
1582

    
1583
  # 3. Perform the files upload
1584
  for fname in dist_files:
1585
    if os.path.exists(fname):
1586
      result = lu.rpc.call_upload_file(dist_nodes, fname)
1587
      for to_node, to_result in result.items():
1588
         msg = to_result.RemoteFailMsg()
1589
         if msg:
1590
           msg = ("Copy of file %s to node %s failed: %s" %
1591
                   (fname, to_node, msg))
1592
           lu.proc.LogWarning(msg)
1593

    
1594

    
1595
class LURedistributeConfig(NoHooksLU):
1596
  """Force the redistribution of cluster configuration.
1597

1598
  This is a very simple LU.
1599

1600
  """
1601
  _OP_REQP = []
1602
  REQ_BGL = False
1603

    
1604
  def ExpandNames(self):
1605
    self.needed_locks = {
1606
      locking.LEVEL_NODE: locking.ALL_SET,
1607
    }
1608
    self.share_locks[locking.LEVEL_NODE] = 1
1609

    
1610
  def CheckPrereq(self):
1611
    """Check prerequisites.
1612

1613
    """
1614

    
1615
  def Exec(self, feedback_fn):
1616
    """Redistribute the configuration.
1617

1618
    """
1619
    self.cfg.Update(self.cfg.GetClusterInfo())
1620
    _RedistributeAncillaryFiles(self)
1621

    
1622

    
1623
def _WaitForSync(lu, instance, oneshot=False, unlock=False):
1624
  """Sleep and poll for an instance's disk to sync.
1625

1626
  """
1627
  if not instance.disks:
1628
    return True
1629

    
1630
  if not oneshot:
1631
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
1632

    
1633
  node = instance.primary_node
1634

    
1635
  for dev in instance.disks:
1636
    lu.cfg.SetDiskID(dev, node)
1637

    
1638
  retries = 0
1639
  while True:
1640
    max_time = 0
1641
    done = True
1642
    cumul_degraded = False
1643
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
1644
    if rstats.failed or not rstats.data:
1645
      lu.LogWarning("Can't get any data from node %s", node)
1646
      retries += 1
1647
      if retries >= 10:
1648
        raise errors.RemoteError("Can't contact node %s for mirror data,"
1649
                                 " aborting." % node)
1650
      time.sleep(6)
1651
      continue
1652
    rstats = rstats.data
1653
    retries = 0
1654
    for i, mstat in enumerate(rstats):
1655
      if mstat is None:
1656
        lu.LogWarning("Can't compute data for node %s/%s",
1657
                           node, instance.disks[i].iv_name)
1658
        continue
1659
      # we ignore the ldisk parameter
1660
      perc_done, est_time, is_degraded, _ = mstat
1661
      cumul_degraded = cumul_degraded or (is_degraded and perc_done is None)
1662
      if perc_done is not None:
1663
        done = False
1664
        if est_time is not None:
1665
          rem_time = "%d estimated seconds remaining" % est_time
1666
          max_time = est_time
1667
        else:
1668
          rem_time = "no time estimate"
1669
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
1670
                        (instance.disks[i].iv_name, perc_done, rem_time))
1671
    if done or oneshot:
1672
      break
1673

    
1674
    time.sleep(min(60, max_time))
1675

    
1676
  if done:
1677
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
1678
  return not cumul_degraded
1679

    
1680

    
1681
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
1682
  """Check that mirrors are not degraded.
1683

1684
  The ldisk parameter, if True, will change the test from the
1685
  is_degraded attribute (which represents overall non-ok status for
1686
  the device(s)) to the ldisk (representing the local storage status).
1687

1688
  """
1689
  lu.cfg.SetDiskID(dev, node)
1690
  if ldisk:
1691
    idx = 6
1692
  else:
1693
    idx = 5
1694

    
1695
  result = True
1696
  if on_primary or dev.AssembleOnSecondary():
1697
    rstats = lu.rpc.call_blockdev_find(node, dev)
1698
    msg = rstats.RemoteFailMsg()
1699
    if msg:
1700
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
1701
      result = False
1702
    elif not rstats.payload:
1703
      lu.LogWarning("Can't find disk on node %s", node)
1704
      result = False
1705
    else:
1706
      result = result and (not rstats.payload[idx])
1707
  if dev.children:
1708
    for child in dev.children:
1709
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
1710

    
1711
  return result
1712

    
1713

    
1714
class LUDiagnoseOS(NoHooksLU):
1715
  """Logical unit for OS diagnose/query.
1716

1717
  """
1718
  _OP_REQP = ["output_fields", "names"]
1719
  REQ_BGL = False
1720
  _FIELDS_STATIC = utils.FieldSet()
1721
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status")
1722

    
1723
  def ExpandNames(self):
1724
    if self.op.names:
1725
      raise errors.OpPrereqError("Selective OS query not supported")
1726

    
1727
    _CheckOutputFields(static=self._FIELDS_STATIC,
1728
                       dynamic=self._FIELDS_DYNAMIC,
1729
                       selected=self.op.output_fields)
1730

    
1731
    # Lock all nodes, in shared mode
1732
    # Temporary removal of locks, should be reverted later
1733
    # TODO: reintroduce locks when they are lighter-weight
1734
    self.needed_locks = {}
1735
    #self.share_locks[locking.LEVEL_NODE] = 1
1736
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
1737

    
1738
  def CheckPrereq(self):
1739
    """Check prerequisites.
1740

1741
    """
1742

    
1743
  @staticmethod
1744
  def _DiagnoseByOS(node_list, rlist):
1745
    """Remaps a per-node return list into an a per-os per-node dictionary
1746

1747
    @param node_list: a list with the names of all nodes
1748
    @param rlist: a map with node names as keys and OS objects as values
1749

1750
    @rtype: dict
1751
    @return: a dictionary with osnames as keys and as value another map, with
1752
        nodes as keys and list of OS objects as values, eg::
1753

1754
          {"debian-etch": {"node1": [<object>,...],
1755
                           "node2": [<object>,]}
1756
          }
1757

1758
    """
1759
    all_os = {}
1760
    # we build here the list of nodes that didn't fail the RPC (at RPC
1761
    # level), so that nodes with a non-responding node daemon don't
1762
    # make all OSes invalid
1763
    good_nodes = [node_name for node_name in rlist
1764
                  if not rlist[node_name].failed]
1765
    for node_name, nr in rlist.iteritems():
1766
      if nr.failed or not nr.data:
1767
        continue
1768
      for os_obj in nr.data:
1769
        if os_obj.name not in all_os:
1770
          # build a list of nodes for this os containing empty lists
1771
          # for each node in node_list
1772
          all_os[os_obj.name] = {}
1773
          for nname in good_nodes:
1774
            all_os[os_obj.name][nname] = []
1775
        all_os[os_obj.name][node_name].append(os_obj)
1776
    return all_os
1777

    
1778
  def Exec(self, feedback_fn):
1779
    """Compute the list of OSes.
1780

1781
    """
1782
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
1783
    node_data = self.rpc.call_os_diagnose(valid_nodes)
1784
    if node_data == False:
1785
      raise errors.OpExecError("Can't gather the list of OSes")
1786
    pol = self._DiagnoseByOS(valid_nodes, node_data)
1787
    output = []
1788
    for os_name, os_data in pol.iteritems():
1789
      row = []
1790
      for field in self.op.output_fields:
1791
        if field == "name":
1792
          val = os_name
1793
        elif field == "valid":
1794
          val = utils.all([osl and osl[0] for osl in os_data.values()])
1795
        elif field == "node_status":
1796
          val = {}
1797
          for node_name, nos_list in os_data.iteritems():
1798
            val[node_name] = [(v.status, v.path) for v in nos_list]
1799
        else:
1800
          raise errors.ParameterError(field)
1801
        row.append(val)
1802
      output.append(row)
1803

    
1804
    return output
1805

    
1806

    
1807
class LURemoveNode(LogicalUnit):
1808
  """Logical unit for removing a node.
1809

1810
  """
1811
  HPATH = "node-remove"
1812
  HTYPE = constants.HTYPE_NODE
1813
  _OP_REQP = ["node_name"]
1814

    
1815
  def BuildHooksEnv(self):
1816
    """Build hooks env.
1817

1818
    This doesn't run on the target node in the pre phase as a failed
1819
    node would then be impossible to remove.
1820

1821
    """
1822
    env = {
1823
      "OP_TARGET": self.op.node_name,
1824
      "NODE_NAME": self.op.node_name,
1825
      }
1826
    all_nodes = self.cfg.GetNodeList()
1827
    all_nodes.remove(self.op.node_name)
1828
    return env, all_nodes, all_nodes
1829

    
1830
  def CheckPrereq(self):
1831
    """Check prerequisites.
1832

1833
    This checks:
1834
     - the node exists in the configuration
1835
     - it does not have primary or secondary instances
1836
     - it's not the master
1837

1838
    Any errors are signalled by raising errors.OpPrereqError.
1839

1840
    """
1841
    node = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.node_name))
1842
    if node is None:
1843
      raise errors.OpPrereqError, ("Node '%s' is unknown." % self.op.node_name)
1844

    
1845
    instance_list = self.cfg.GetInstanceList()
1846

    
1847
    masternode = self.cfg.GetMasterNode()
1848
    if node.name == masternode:
1849
      raise errors.OpPrereqError("Node is the master node,"
1850
                                 " you need to failover first.")
1851

    
1852
    for instance_name in instance_list:
1853
      instance = self.cfg.GetInstanceInfo(instance_name)
1854
      if node.name in instance.all_nodes:
1855
        raise errors.OpPrereqError("Instance %s is still running on the node,"
1856
                                   " please remove first." % instance_name)
1857
    self.op.node_name = node.name
1858
    self.node = node
1859

    
1860
  def Exec(self, feedback_fn):
1861
    """Removes the node from the cluster.
1862

1863
    """
1864
    node = self.node
1865
    logging.info("Stopping the node daemon and removing configs from node %s",
1866
                 node.name)
1867

    
1868
    self.context.RemoveNode(node.name)
1869

    
1870
    self.rpc.call_node_leave_cluster(node.name)
1871

    
1872
    # Promote nodes to master candidate as needed
1873
    _AdjustCandidatePool(self)
1874

    
1875

    
1876
class LUQueryNodes(NoHooksLU):
1877
  """Logical unit for querying nodes.
1878

1879
  """
1880
  _OP_REQP = ["output_fields", "names", "use_locking"]
1881
  REQ_BGL = False
1882
  _FIELDS_DYNAMIC = utils.FieldSet(
1883
    "dtotal", "dfree",
1884
    "mtotal", "mnode", "mfree",
1885
    "bootid",
1886
    "ctotal", "cnodes", "csockets",
1887
    )
1888

    
1889
  _FIELDS_STATIC = utils.FieldSet(
1890
    "name", "pinst_cnt", "sinst_cnt",
1891
    "pinst_list", "sinst_list",
1892
    "pip", "sip", "tags",
1893
    "serial_no",
1894
    "master_candidate",
1895
    "master",
1896
    "offline",
1897
    "drained",
1898
    )
1899

    
1900
  def ExpandNames(self):
1901
    _CheckOutputFields(static=self._FIELDS_STATIC,
1902
                       dynamic=self._FIELDS_DYNAMIC,
1903
                       selected=self.op.output_fields)
1904

    
1905
    self.needed_locks = {}
1906
    self.share_locks[locking.LEVEL_NODE] = 1
1907

    
1908
    if self.op.names:
1909
      self.wanted = _GetWantedNodes(self, self.op.names)
1910
    else:
1911
      self.wanted = locking.ALL_SET
1912

    
1913
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
1914
    self.do_locking = self.do_node_query and self.op.use_locking
1915
    if self.do_locking:
1916
      # if we don't request only static fields, we need to lock the nodes
1917
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
1918

    
1919

    
1920
  def CheckPrereq(self):
1921
    """Check prerequisites.
1922

1923
    """
1924
    # The validation of the node list is done in the _GetWantedNodes,
1925
    # if non empty, and if empty, there's no validation to do
1926
    pass
1927

    
1928
  def Exec(self, feedback_fn):
1929
    """Computes the list of nodes and their attributes.
1930

1931
    """
1932
    all_info = self.cfg.GetAllNodesInfo()
1933
    if self.do_locking:
1934
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
1935
    elif self.wanted != locking.ALL_SET:
1936
      nodenames = self.wanted
1937
      missing = set(nodenames).difference(all_info.keys())
1938
      if missing:
1939
        raise errors.OpExecError(
1940
          "Some nodes were removed before retrieving their data: %s" % missing)
1941
    else:
1942
      nodenames = all_info.keys()
1943

    
1944
    nodenames = utils.NiceSort(nodenames)
1945
    nodelist = [all_info[name] for name in nodenames]
1946

    
1947
    # begin data gathering
1948

    
1949
    if self.do_node_query:
1950
      live_data = {}
1951
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
1952
                                          self.cfg.GetHypervisorType())
1953
      for name in nodenames:
1954
        nodeinfo = node_data[name]
1955
        if not nodeinfo.failed and nodeinfo.data:
1956
          nodeinfo = nodeinfo.data
1957
          fn = utils.TryConvert
1958
          live_data[name] = {
1959
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
1960
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
1961
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
1962
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
1963
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
1964
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
1965
            "bootid": nodeinfo.get('bootid', None),
1966
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
1967
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
1968
            }
1969
        else:
1970
          live_data[name] = {}
1971
    else:
1972
      live_data = dict.fromkeys(nodenames, {})
1973

    
1974
    node_to_primary = dict([(name, set()) for name in nodenames])
1975
    node_to_secondary = dict([(name, set()) for name in nodenames])
1976

    
1977
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
1978
                             "sinst_cnt", "sinst_list"))
1979
    if inst_fields & frozenset(self.op.output_fields):
1980
      instancelist = self.cfg.GetInstanceList()
1981

    
1982
      for instance_name in instancelist:
1983
        inst = self.cfg.GetInstanceInfo(instance_name)
1984
        if inst.primary_node in node_to_primary:
1985
          node_to_primary[inst.primary_node].add(inst.name)
1986
        for secnode in inst.secondary_nodes:
1987
          if secnode in node_to_secondary:
1988
            node_to_secondary[secnode].add(inst.name)
1989

    
1990
    master_node = self.cfg.GetMasterNode()
1991

    
1992
    # end data gathering
1993

    
1994
    output = []
1995
    for node in nodelist:
1996
      node_output = []
1997
      for field in self.op.output_fields:
1998
        if field == "name":
1999
          val = node.name
2000
        elif field == "pinst_list":
2001
          val = list(node_to_primary[node.name])
2002
        elif field == "sinst_list":
2003
          val = list(node_to_secondary[node.name])
2004
        elif field == "pinst_cnt":
2005
          val = len(node_to_primary[node.name])
2006
        elif field == "sinst_cnt":
2007
          val = len(node_to_secondary[node.name])
2008
        elif field == "pip":
2009
          val = node.primary_ip
2010
        elif field == "sip":
2011
          val = node.secondary_ip
2012
        elif field == "tags":
2013
          val = list(node.GetTags())
2014
        elif field == "serial_no":
2015
          val = node.serial_no
2016
        elif field == "master_candidate":
2017
          val = node.master_candidate
2018
        elif field == "master":
2019
          val = node.name == master_node
2020
        elif field == "offline":
2021
          val = node.offline
2022
        elif field == "drained":
2023
          val = node.drained
2024
        elif self._FIELDS_DYNAMIC.Matches(field):
2025
          val = live_data[node.name].get(field, None)
2026
        else:
2027
          raise errors.ParameterError(field)
2028
        node_output.append(val)
2029
      output.append(node_output)
2030

    
2031
    return output
2032

    
2033

    
2034
class LUQueryNodeVolumes(NoHooksLU):
2035
  """Logical unit for getting volumes on node(s).
2036

2037
  """
2038
  _OP_REQP = ["nodes", "output_fields"]
2039
  REQ_BGL = False
2040
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2041
  _FIELDS_STATIC = utils.FieldSet("node")
2042

    
2043
  def ExpandNames(self):
2044
    _CheckOutputFields(static=self._FIELDS_STATIC,
2045
                       dynamic=self._FIELDS_DYNAMIC,
2046
                       selected=self.op.output_fields)
2047

    
2048
    self.needed_locks = {}
2049
    self.share_locks[locking.LEVEL_NODE] = 1
2050
    if not self.op.nodes:
2051
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2052
    else:
2053
      self.needed_locks[locking.LEVEL_NODE] = \
2054
        _GetWantedNodes(self, self.op.nodes)
2055

    
2056
  def CheckPrereq(self):
2057
    """Check prerequisites.
2058

2059
    This checks that the fields required are valid output fields.
2060

2061
    """
2062
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
2063

    
2064
  def Exec(self, feedback_fn):
2065
    """Computes the list of nodes and their attributes.
2066

2067
    """
2068
    nodenames = self.nodes
2069
    volumes = self.rpc.call_node_volumes(nodenames)
2070

    
2071
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
2072
             in self.cfg.GetInstanceList()]
2073

    
2074
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
2075

    
2076
    output = []
2077
    for node in nodenames:
2078
      if node not in volumes or volumes[node].failed or not volumes[node].data:
2079
        continue
2080

    
2081
      node_vols = volumes[node].data[:]
2082
      node_vols.sort(key=lambda vol: vol['dev'])
2083

    
2084
      for vol in node_vols:
2085
        node_output = []
2086
        for field in self.op.output_fields:
2087
          if field == "node":
2088
            val = node
2089
          elif field == "phys":
2090
            val = vol['dev']
2091
          elif field == "vg":
2092
            val = vol['vg']
2093
          elif field == "name":
2094
            val = vol['name']
2095
          elif field == "size":
2096
            val = int(float(vol['size']))
2097
          elif field == "instance":
2098
            for inst in ilist:
2099
              if node not in lv_by_node[inst]:
2100
                continue
2101
              if vol['name'] in lv_by_node[inst][node]:
2102
                val = inst.name
2103
                break
2104
            else:
2105
              val = '-'
2106
          else:
2107
            raise errors.ParameterError(field)
2108
          node_output.append(str(val))
2109

    
2110
        output.append(node_output)
2111

    
2112
    return output
2113

    
2114

    
2115
class LUAddNode(LogicalUnit):
2116
  """Logical unit for adding node to the cluster.
2117

2118
  """
2119
  HPATH = "node-add"
2120
  HTYPE = constants.HTYPE_NODE
2121
  _OP_REQP = ["node_name"]
2122

    
2123
  def BuildHooksEnv(self):
2124
    """Build hooks env.
2125

2126
    This will run on all nodes before, and on all nodes + the new node after.
2127

2128
    """
2129
    env = {
2130
      "OP_TARGET": self.op.node_name,
2131
      "NODE_NAME": self.op.node_name,
2132
      "NODE_PIP": self.op.primary_ip,
2133
      "NODE_SIP": self.op.secondary_ip,
2134
      }
2135
    nodes_0 = self.cfg.GetNodeList()
2136
    nodes_1 = nodes_0 + [self.op.node_name, ]
2137
    return env, nodes_0, nodes_1
2138

    
2139
  def CheckPrereq(self):
2140
    """Check prerequisites.
2141

2142
    This checks:
2143
     - the new node is not already in the config
2144
     - it is resolvable
2145
     - its parameters (single/dual homed) matches the cluster
2146

2147
    Any errors are signalled by raising errors.OpPrereqError.
2148

2149
    """
2150
    node_name = self.op.node_name
2151
    cfg = self.cfg
2152

    
2153
    dns_data = utils.HostInfo(node_name)
2154

    
2155
    node = dns_data.name
2156
    primary_ip = self.op.primary_ip = dns_data.ip
2157
    secondary_ip = getattr(self.op, "secondary_ip", None)
2158
    if secondary_ip is None:
2159
      secondary_ip = primary_ip
2160
    if not utils.IsValidIP(secondary_ip):
2161
      raise errors.OpPrereqError("Invalid secondary IP given")
2162
    self.op.secondary_ip = secondary_ip
2163

    
2164
    node_list = cfg.GetNodeList()
2165
    if not self.op.readd and node in node_list:
2166
      raise errors.OpPrereqError("Node %s is already in the configuration" %
2167
                                 node)
2168
    elif self.op.readd and node not in node_list:
2169
      raise errors.OpPrereqError("Node %s is not in the configuration" % node)
2170

    
2171
    for existing_node_name in node_list:
2172
      existing_node = cfg.GetNodeInfo(existing_node_name)
2173

    
2174
      if self.op.readd and node == existing_node_name:
2175
        if (existing_node.primary_ip != primary_ip or
2176
            existing_node.secondary_ip != secondary_ip):
2177
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
2178
                                     " address configuration as before")
2179
        continue
2180

    
2181
      if (existing_node.primary_ip == primary_ip or
2182
          existing_node.secondary_ip == primary_ip or
2183
          existing_node.primary_ip == secondary_ip or
2184
          existing_node.secondary_ip == secondary_ip):
2185
        raise errors.OpPrereqError("New node ip address(es) conflict with"
2186
                                   " existing node %s" % existing_node.name)
2187

    
2188
    # check that the type of the node (single versus dual homed) is the
2189
    # same as for the master
2190
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
2191
    master_singlehomed = myself.secondary_ip == myself.primary_ip
2192
    newbie_singlehomed = secondary_ip == primary_ip
2193
    if master_singlehomed != newbie_singlehomed:
2194
      if master_singlehomed:
2195
        raise errors.OpPrereqError("The master has no private ip but the"
2196
                                   " new node has one")
2197
      else:
2198
        raise errors.OpPrereqError("The master has a private ip but the"
2199
                                   " new node doesn't have one")
2200

    
2201
    # checks reachablity
2202
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
2203
      raise errors.OpPrereqError("Node not reachable by ping")
2204

    
2205
    if not newbie_singlehomed:
2206
      # check reachability from my secondary ip to newbie's secondary ip
2207
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
2208
                           source=myself.secondary_ip):
2209
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
2210
                                   " based ping to noded port")
2211

    
2212
    cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2213
    mc_now, _ = self.cfg.GetMasterCandidateStats()
2214
    master_candidate = mc_now < cp_size
2215

    
2216
    self.new_node = objects.Node(name=node,
2217
                                 primary_ip=primary_ip,
2218
                                 secondary_ip=secondary_ip,
2219
                                 master_candidate=master_candidate,
2220
                                 offline=False, drained=False)
2221

    
2222
  def Exec(self, feedback_fn):
2223
    """Adds the new node to the cluster.
2224

2225
    """
2226
    new_node = self.new_node
2227
    node = new_node.name
2228

    
2229
    # check connectivity
2230
    result = self.rpc.call_version([node])[node]
2231
    result.Raise()
2232
    if result.data:
2233
      if constants.PROTOCOL_VERSION == result.data:
2234
        logging.info("Communication to node %s fine, sw version %s match",
2235
                     node, result.data)
2236
      else:
2237
        raise errors.OpExecError("Version mismatch master version %s,"
2238
                                 " node version %s" %
2239
                                 (constants.PROTOCOL_VERSION, result.data))
2240
    else:
2241
      raise errors.OpExecError("Cannot get version from the new node")
2242

    
2243
    # setup ssh on node
2244
    logging.info("Copy ssh key to node %s", node)
2245
    priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
2246
    keyarray = []
2247
    keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
2248
                constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
2249
                priv_key, pub_key]
2250

    
2251
    for i in keyfiles:
2252
      f = open(i, 'r')
2253
      try:
2254
        keyarray.append(f.read())
2255
      finally:
2256
        f.close()
2257

    
2258
    result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
2259
                                    keyarray[2],
2260
                                    keyarray[3], keyarray[4], keyarray[5])
2261

    
2262
    msg = result.RemoteFailMsg()
2263
    if msg:
2264
      raise errors.OpExecError("Cannot transfer ssh keys to the"
2265
                               " new node: %s" % msg)
2266

    
2267
    # Add node to our /etc/hosts, and add key to known_hosts
2268
    if self.cfg.GetClusterInfo().modify_etc_hosts:
2269
      utils.AddHostToEtcHosts(new_node.name)
2270

    
2271
    if new_node.secondary_ip != new_node.primary_ip:
2272
      result = self.rpc.call_node_has_ip_address(new_node.name,
2273
                                                 new_node.secondary_ip)
2274
      if result.failed or not result.data:
2275
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
2276
                                 " you gave (%s). Please fix and re-run this"
2277
                                 " command." % new_node.secondary_ip)
2278

    
2279
    node_verify_list = [self.cfg.GetMasterNode()]
2280
    node_verify_param = {
2281
      'nodelist': [node],
2282
      # TODO: do a node-net-test as well?
2283
    }
2284

    
2285
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
2286
                                       self.cfg.GetClusterName())
2287
    for verifier in node_verify_list:
2288
      if result[verifier].failed or not result[verifier].data:
2289
        raise errors.OpExecError("Cannot communicate with %s's node daemon"
2290
                                 " for remote verification" % verifier)
2291
      if result[verifier].data['nodelist']:
2292
        for failed in result[verifier].data['nodelist']:
2293
          feedback_fn("ssh/hostname verification failed %s -> %s" %
2294
                      (verifier, result[verifier].data['nodelist'][failed]))
2295
        raise errors.OpExecError("ssh/hostname verification failed.")
2296

    
2297
    if self.op.readd:
2298
      _RedistributeAncillaryFiles(self)
2299
      self.context.ReaddNode(new_node)
2300
    else:
2301
      _RedistributeAncillaryFiles(self, additional_nodes=node)
2302
      self.context.AddNode(new_node)
2303

    
2304

    
2305
class LUSetNodeParams(LogicalUnit):
2306
  """Modifies the parameters of a node.
2307

2308
  """
2309
  HPATH = "node-modify"
2310
  HTYPE = constants.HTYPE_NODE
2311
  _OP_REQP = ["node_name"]
2312
  REQ_BGL = False
2313

    
2314
  def CheckArguments(self):
2315
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2316
    if node_name is None:
2317
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2318
    self.op.node_name = node_name
2319
    _CheckBooleanOpField(self.op, 'master_candidate')
2320
    _CheckBooleanOpField(self.op, 'offline')
2321
    _CheckBooleanOpField(self.op, 'drained')
2322
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
2323
    if all_mods.count(None) == 3:
2324
      raise errors.OpPrereqError("Please pass at least one modification")
2325
    if all_mods.count(True) > 1:
2326
      raise errors.OpPrereqError("Can't set the node into more than one"
2327
                                 " state at the same time")
2328

    
2329
  def ExpandNames(self):
2330
    self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
2331

    
2332
  def BuildHooksEnv(self):
2333
    """Build hooks env.
2334

2335
    This runs on the master node.
2336

2337
    """
2338
    env = {
2339
      "OP_TARGET": self.op.node_name,
2340
      "MASTER_CANDIDATE": str(self.op.master_candidate),
2341
      "OFFLINE": str(self.op.offline),
2342
      "DRAINED": str(self.op.drained),
2343
      }
2344
    nl = [self.cfg.GetMasterNode(),
2345
          self.op.node_name]
2346
    return env, nl, nl
2347

    
2348
  def CheckPrereq(self):
2349
    """Check prerequisites.
2350

2351
    This only checks the instance list against the existing names.
2352

2353
    """
2354
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
2355

    
2356
    if ((self.op.master_candidate == False or self.op.offline == True or
2357
         self.op.drained == True) and node.master_candidate):
2358
      # we will demote the node from master_candidate
2359
      if self.op.node_name == self.cfg.GetMasterNode():
2360
        raise errors.OpPrereqError("The master node has to be a"
2361
                                   " master candidate, online and not drained")
2362
      cp_size = self.cfg.GetClusterInfo().candidate_pool_size
2363
      num_candidates, _ = self.cfg.GetMasterCandidateStats()
2364
      if num_candidates <= cp_size:
2365
        msg = ("Not enough master candidates (desired"
2366
               " %d, new value will be %d)" % (cp_size, num_candidates-1))
2367
        if self.op.force:
2368
          self.LogWarning(msg)
2369
        else:
2370
          raise errors.OpPrereqError(msg)
2371

    
2372
    if (self.op.master_candidate == True and
2373
        ((node.offline and not self.op.offline == False) or
2374
         (node.drained and not self.op.drained == False))):
2375
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
2376
                                 " to master_candidate" % node.name)
2377

    
2378
    return
2379

    
2380
  def Exec(self, feedback_fn):
2381
    """Modifies a node.
2382

2383
    """
2384
    node = self.node
2385

    
2386
    result = []
2387
    changed_mc = False
2388

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

    
2401
    if self.op.master_candidate is not None:
2402
      node.master_candidate = self.op.master_candidate
2403
      changed_mc = True
2404
      result.append(("master_candidate", str(self.op.master_candidate)))
2405
      if self.op.master_candidate == False:
2406
        rrc = self.rpc.call_node_demote_from_mc(node.name)
2407
        msg = rrc.RemoteFailMsg()
2408
        if msg:
2409
          self.LogWarning("Node failed to demote itself: %s" % msg)
2410

    
2411
    if self.op.drained is not None:
2412
      node.drained = self.op.drained
2413
      result.append(("drained", str(self.op.drained)))
2414
      if self.op.drained == True:
2415
        if node.master_candidate:
2416
          node.master_candidate = False
2417
          changed_mc = True
2418
          result.append(("master_candidate", "auto-demotion due to drain"))
2419
        if node.offline:
2420
          node.offline = False
2421
          result.append(("offline", "clear offline status due to drain"))
2422

    
2423
    # this will trigger configuration file update, if needed
2424
    self.cfg.Update(node)
2425
    # this will trigger job queue propagation or cleanup
2426
    if changed_mc:
2427
      self.context.ReaddNode(node)
2428

    
2429
    return result
2430

    
2431

    
2432
class LUPowercycleNode(NoHooksLU):
2433
  """Powercycles a node.
2434

2435
  """
2436
  _OP_REQP = ["node_name", "force"]
2437
  REQ_BGL = False
2438

    
2439
  def CheckArguments(self):
2440
    node_name = self.cfg.ExpandNodeName(self.op.node_name)
2441
    if node_name is None:
2442
      raise errors.OpPrereqError("Invalid node name '%s'" % self.op.node_name)
2443
    self.op.node_name = node_name
2444
    if node_name == self.cfg.GetMasterNode() and not self.op.force:
2445
      raise errors.OpPrereqError("The node is the master and the force"
2446
                                 " parameter was not set")
2447

    
2448
  def ExpandNames(self):
2449
    """Locking for PowercycleNode.
2450

2451
    This is a last-resource option and shouldn't block on other
2452
    jobs. Therefore, we grab no locks.
2453

2454
    """
2455
    self.needed_locks = {}
2456

    
2457
  def CheckPrereq(self):
2458
    """Check prerequisites.
2459

2460
    This LU has no prereqs.
2461

2462
    """
2463
    pass
2464

    
2465
  def Exec(self, feedback_fn):
2466
    """Reboots a node.
2467

2468
    """
2469
    result = self.rpc.call_node_powercycle(self.op.node_name,
2470
                                           self.cfg.GetHypervisorType())
2471
    msg = result.RemoteFailMsg()
2472
    if msg:
2473
      raise errors.OpExecError("Failed to schedule the reboot: %s" % msg)
2474
    return result.payload
2475

    
2476

    
2477
class LUQueryClusterInfo(NoHooksLU):
2478
  """Query cluster configuration.
2479

2480
  """
2481
  _OP_REQP = []
2482
  REQ_BGL = False
2483

    
2484
  def ExpandNames(self):
2485
    self.needed_locks = {}
2486

    
2487
  def CheckPrereq(self):
2488
    """No prerequsites needed for this LU.
2489

2490
    """
2491
    pass
2492

    
2493
  def Exec(self, feedback_fn):
2494
    """Return cluster config.
2495

2496
    """
2497
    cluster = self.cfg.GetClusterInfo()
2498
    result = {
2499
      "software_version": constants.RELEASE_VERSION,
2500
      "protocol_version": constants.PROTOCOL_VERSION,
2501
      "config_version": constants.CONFIG_VERSION,
2502
      "os_api_version": constants.OS_API_VERSION,
2503
      "export_version": constants.EXPORT_VERSION,
2504
      "architecture": (platform.architecture()[0], platform.machine()),
2505
      "name": cluster.cluster_name,
2506
      "master": cluster.master_node,
2507
      "default_hypervisor": cluster.default_hypervisor,
2508
      "enabled_hypervisors": cluster.enabled_hypervisors,
2509
      "hvparams": dict([(hypervisor, cluster.hvparams[hypervisor])
2510
                        for hypervisor in cluster.enabled_hypervisors]),
2511
      "beparams": cluster.beparams,
2512
      "candidate_pool_size": cluster.candidate_pool_size,
2513
      "default_bridge": cluster.default_bridge,
2514
      "master_netdev": cluster.master_netdev,
2515
      "volume_group_name": cluster.volume_group_name,
2516
      "file_storage_dir": cluster.file_storage_dir,
2517
      }
2518

    
2519
    return result
2520

    
2521

    
2522
class LUQueryConfigValues(NoHooksLU):
2523
  """Return configuration values.
2524

2525
  """
2526
  _OP_REQP = []
2527
  REQ_BGL = False
2528
  _FIELDS_DYNAMIC = utils.FieldSet()
2529
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag")
2530

    
2531
  def ExpandNames(self):
2532
    self.needed_locks = {}
2533

    
2534
    _CheckOutputFields(static=self._FIELDS_STATIC,
2535
                       dynamic=self._FIELDS_DYNAMIC,
2536
                       selected=self.op.output_fields)
2537

    
2538
  def CheckPrereq(self):
2539
    """No prerequisites.
2540

2541
    """
2542
    pass
2543

    
2544
  def Exec(self, feedback_fn):
2545
    """Dump a representation of the cluster config to the standard output.
2546

2547
    """
2548
    values = []
2549
    for field in self.op.output_fields:
2550
      if field == "cluster_name":
2551
        entry = self.cfg.GetClusterName()
2552
      elif field == "master_node":
2553
        entry = self.cfg.GetMasterNode()
2554
      elif field == "drain_flag":
2555
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
2556
      else:
2557
        raise errors.ParameterError(field)
2558
      values.append(entry)
2559
    return values
2560

    
2561

    
2562
class LUActivateInstanceDisks(NoHooksLU):
2563
  """Bring up an instance's disks.
2564

2565
  """
2566
  _OP_REQP = ["instance_name"]
2567
  REQ_BGL = False
2568

    
2569
  def ExpandNames(self):
2570
    self._ExpandAndLockInstance()
2571
    self.needed_locks[locking.LEVEL_NODE] = []
2572
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2573

    
2574
  def DeclareLocks(self, level):
2575
    if level == locking.LEVEL_NODE:
2576
      self._LockInstancesNodes()
2577

    
2578
  def CheckPrereq(self):
2579
    """Check prerequisites.
2580

2581
    This checks that the instance is in the cluster.
2582

2583
    """
2584
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2585
    assert self.instance is not None, \
2586
      "Cannot retrieve locked instance %s" % self.op.instance_name
2587
    _CheckNodeOnline(self, self.instance.primary_node)
2588

    
2589
  def Exec(self, feedback_fn):
2590
    """Activate the disks.
2591

2592
    """
2593
    disks_ok, disks_info = _AssembleInstanceDisks(self, self.instance)
2594
    if not disks_ok:
2595
      raise errors.OpExecError("Cannot activate block devices")
2596

    
2597
    return disks_info
2598

    
2599

    
2600
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False):
2601
  """Prepare the block devices for an instance.
2602

2603
  This sets up the block devices on all nodes.
2604

2605
  @type lu: L{LogicalUnit}
2606
  @param lu: the logical unit on whose behalf we execute
2607
  @type instance: L{objects.Instance}
2608
  @param instance: the instance for whose disks we assemble
2609
  @type ignore_secondaries: boolean
2610
  @param ignore_secondaries: if true, errors on secondary nodes
2611
      won't result in an error return from the function
2612
  @return: False if the operation failed, otherwise a list of
2613
      (host, instance_visible_name, node_visible_name)
2614
      with the mapping from node devices to instance devices
2615

2616
  """
2617
  device_info = []
2618
  disks_ok = True
2619
  iname = instance.name
2620
  # With the two passes mechanism we try to reduce the window of
2621
  # opportunity for the race condition of switching DRBD to primary
2622
  # before handshaking occured, but we do not eliminate it
2623

    
2624
  # The proper fix would be to wait (with some limits) until the
2625
  # connection has been made and drbd transitions from WFConnection
2626
  # into any other network-connected state (Connected, SyncTarget,
2627
  # SyncSource, etc.)
2628

    
2629
  # 1st pass, assemble on all nodes in secondary mode
2630
  for inst_disk in instance.disks:
2631
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2632
      lu.cfg.SetDiskID(node_disk, node)
2633
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
2634
      msg = result.RemoteFailMsg()
2635
      if msg:
2636
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2637
                           " (is_primary=False, pass=1): %s",
2638
                           inst_disk.iv_name, node, msg)
2639
        if not ignore_secondaries:
2640
          disks_ok = False
2641

    
2642
  # FIXME: race condition on drbd migration to primary
2643

    
2644
  # 2nd pass, do only the primary node
2645
  for inst_disk in instance.disks:
2646
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
2647
      if node != instance.primary_node:
2648
        continue
2649
      lu.cfg.SetDiskID(node_disk, node)
2650
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
2651
      msg = result.RemoteFailMsg()
2652
      if msg:
2653
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
2654
                           " (is_primary=True, pass=2): %s",
2655
                           inst_disk.iv_name, node, msg)
2656
        disks_ok = False
2657
    device_info.append((instance.primary_node, inst_disk.iv_name,
2658
                        result.payload))
2659

    
2660
  # leave the disks configured for the primary node
2661
  # this is a workaround that would be fixed better by
2662
  # improving the logical/physical id handling
2663
  for disk in instance.disks:
2664
    lu.cfg.SetDiskID(disk, instance.primary_node)
2665

    
2666
  return disks_ok, device_info
2667

    
2668

    
2669
def _StartInstanceDisks(lu, instance, force):
2670
  """Start the disks of an instance.
2671

2672
  """
2673
  disks_ok, dummy = _AssembleInstanceDisks(lu, instance,
2674
                                           ignore_secondaries=force)
2675
  if not disks_ok:
2676
    _ShutdownInstanceDisks(lu, instance)
2677
    if force is not None and not force:
2678
      lu.proc.LogWarning("", hint="If the message above refers to a"
2679
                         " secondary node,"
2680
                         " you can retry the operation using '--force'.")
2681
    raise errors.OpExecError("Disk consistency error")
2682

    
2683

    
2684
class LUDeactivateInstanceDisks(NoHooksLU):
2685
  """Shutdown an instance's disks.
2686

2687
  """
2688
  _OP_REQP = ["instance_name"]
2689
  REQ_BGL = False
2690

    
2691
  def ExpandNames(self):
2692
    self._ExpandAndLockInstance()
2693
    self.needed_locks[locking.LEVEL_NODE] = []
2694
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2695

    
2696
  def DeclareLocks(self, level):
2697
    if level == locking.LEVEL_NODE:
2698
      self._LockInstancesNodes()
2699

    
2700
  def CheckPrereq(self):
2701
    """Check prerequisites.
2702

2703
    This checks that the instance is in the cluster.
2704

2705
    """
2706
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2707
    assert self.instance is not None, \
2708
      "Cannot retrieve locked instance %s" % self.op.instance_name
2709

    
2710
  def Exec(self, feedback_fn):
2711
    """Deactivate the disks
2712

2713
    """
2714
    instance = self.instance
2715
    _SafeShutdownInstanceDisks(self, instance)
2716

    
2717

    
2718
def _SafeShutdownInstanceDisks(lu, instance):
2719
  """Shutdown block devices of an instance.
2720

2721
  This function checks if an instance is running, before calling
2722
  _ShutdownInstanceDisks.
2723

2724
  """
2725
  ins_l = lu.rpc.call_instance_list([instance.primary_node],
2726
                                      [instance.hypervisor])
2727
  ins_l = ins_l[instance.primary_node]
2728
  if ins_l.failed or not isinstance(ins_l.data, list):
2729
    raise errors.OpExecError("Can't contact node '%s'" %
2730
                             instance.primary_node)
2731

    
2732
  if instance.name in ins_l.data:
2733
    raise errors.OpExecError("Instance is running, can't shutdown"
2734
                             " block devices.")
2735

    
2736
  _ShutdownInstanceDisks(lu, instance)
2737

    
2738

    
2739
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
2740
  """Shutdown block devices of an instance.
2741

2742
  This does the shutdown on all nodes of the instance.
2743

2744
  If the ignore_primary is false, errors on the primary node are
2745
  ignored.
2746

2747
  """
2748
  all_result = True
2749
  for disk in instance.disks:
2750
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
2751
      lu.cfg.SetDiskID(top_disk, node)
2752
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
2753
      msg = result.RemoteFailMsg()
2754
      if msg:
2755
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
2756
                      disk.iv_name, node, msg)
2757
        if not ignore_primary or node != instance.primary_node:
2758
          all_result = False
2759
  return all_result
2760

    
2761

    
2762
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
2763
  """Checks if a node has enough free memory.
2764

2765
  This function check if a given node has the needed amount of free
2766
  memory. In case the node has less memory or we cannot get the
2767
  information from the node, this function raise an OpPrereqError
2768
  exception.
2769

2770
  @type lu: C{LogicalUnit}
2771
  @param lu: a logical unit from which we get configuration data
2772
  @type node: C{str}
2773
  @param node: the node to check
2774
  @type reason: C{str}
2775
  @param reason: string to use in the error message
2776
  @type requested: C{int}
2777
  @param requested: the amount of memory in MiB to check for
2778
  @type hypervisor_name: C{str}
2779
  @param hypervisor_name: the hypervisor to ask for memory stats
2780
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
2781
      we cannot check the node
2782

2783
  """
2784
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
2785
  nodeinfo[node].Raise()
2786
  free_mem = nodeinfo[node].data.get('memory_free')
2787
  if not isinstance(free_mem, int):
2788
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
2789
                             " was '%s'" % (node, free_mem))
2790
  if requested > free_mem:
2791
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
2792
                             " needed %s MiB, available %s MiB" %
2793
                             (node, reason, requested, free_mem))
2794

    
2795

    
2796
class LUStartupInstance(LogicalUnit):
2797
  """Starts an instance.
2798

2799
  """
2800
  HPATH = "instance-start"
2801
  HTYPE = constants.HTYPE_INSTANCE
2802
  _OP_REQP = ["instance_name", "force"]
2803
  REQ_BGL = False
2804

    
2805
  def ExpandNames(self):
2806
    self._ExpandAndLockInstance()
2807

    
2808
  def BuildHooksEnv(self):
2809
    """Build hooks env.
2810

2811
    This runs on master, primary and secondary nodes of the instance.
2812

2813
    """
2814
    env = {
2815
      "FORCE": self.op.force,
2816
      }
2817
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2818
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2819
    return env, nl, nl
2820

    
2821
  def CheckPrereq(self):
2822
    """Check prerequisites.
2823

2824
    This checks that the instance is in the cluster.
2825

2826
    """
2827
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2828
    assert self.instance is not None, \
2829
      "Cannot retrieve locked instance %s" % self.op.instance_name
2830

    
2831
    # extra beparams
2832
    self.beparams = getattr(self.op, "beparams", {})
2833
    if self.beparams:
2834
      if not isinstance(self.beparams, dict):
2835
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
2836
                                   " dict" % (type(self.beparams), ))
2837
      # fill the beparams dict
2838
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
2839
      self.op.beparams = self.beparams
2840

    
2841
    # extra hvparams
2842
    self.hvparams = getattr(self.op, "hvparams", {})
2843
    if self.hvparams:
2844
      if not isinstance(self.hvparams, dict):
2845
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
2846
                                   " dict" % (type(self.hvparams), ))
2847

    
2848
      # check hypervisor parameter syntax (locally)
2849
      cluster = self.cfg.GetClusterInfo()
2850
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
2851
      filled_hvp = cluster.FillDict(cluster.hvparams[instance.hypervisor],
2852
                                    instance.hvparams)
2853
      filled_hvp.update(self.hvparams)
2854
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
2855
      hv_type.CheckParameterSyntax(filled_hvp)
2856
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
2857
      self.op.hvparams = self.hvparams
2858

    
2859
    _CheckNodeOnline(self, instance.primary_node)
2860

    
2861
    bep = self.cfg.GetClusterInfo().FillBE(instance)
2862
    # check bridges existance
2863
    _CheckInstanceBridgesExist(self, instance)
2864

    
2865
    remote_info = self.rpc.call_instance_info(instance.primary_node,
2866
                                              instance.name,
2867
                                              instance.hypervisor)
2868
    remote_info.Raise()
2869
    if not remote_info.data:
2870
      _CheckNodeFreeMemory(self, instance.primary_node,
2871
                           "starting instance %s" % instance.name,
2872
                           bep[constants.BE_MEMORY], instance.hypervisor)
2873

    
2874
  def Exec(self, feedback_fn):
2875
    """Start the instance.
2876

2877
    """
2878
    instance = self.instance
2879
    force = self.op.force
2880

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

    
2883
    node_current = instance.primary_node
2884

    
2885
    _StartInstanceDisks(self, instance, force)
2886

    
2887
    result = self.rpc.call_instance_start(node_current, instance,
2888
                                          self.hvparams, self.beparams)
2889
    msg = result.RemoteFailMsg()
2890
    if msg:
2891
      _ShutdownInstanceDisks(self, instance)
2892
      raise errors.OpExecError("Could not start instance: %s" % msg)
2893

    
2894

    
2895
class LURebootInstance(LogicalUnit):
2896
  """Reboot an instance.
2897

2898
  """
2899
  HPATH = "instance-reboot"
2900
  HTYPE = constants.HTYPE_INSTANCE
2901
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
2902
  REQ_BGL = False
2903

    
2904
  def ExpandNames(self):
2905
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
2906
                                   constants.INSTANCE_REBOOT_HARD,
2907
                                   constants.INSTANCE_REBOOT_FULL]:
2908
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
2909
                                  (constants.INSTANCE_REBOOT_SOFT,
2910
                                   constants.INSTANCE_REBOOT_HARD,
2911
                                   constants.INSTANCE_REBOOT_FULL))
2912
    self._ExpandAndLockInstance()
2913

    
2914
  def BuildHooksEnv(self):
2915
    """Build hooks env.
2916

2917
    This runs on master, primary and secondary nodes of the instance.
2918

2919
    """
2920
    env = {
2921
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
2922
      "REBOOT_TYPE": self.op.reboot_type,
2923
      }
2924
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
2925
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
2926
    return env, nl, nl
2927

    
2928
  def CheckPrereq(self):
2929
    """Check prerequisites.
2930

2931
    This checks that the instance is in the cluster.
2932

2933
    """
2934
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
2935
    assert self.instance is not None, \
2936
      "Cannot retrieve locked instance %s" % self.op.instance_name
2937

    
2938
    _CheckNodeOnline(self, instance.primary_node)
2939

    
2940
    # check bridges existance
2941
    _CheckInstanceBridgesExist(self, instance)
2942

    
2943
  def Exec(self, feedback_fn):
2944
    """Reboot the instance.
2945

2946
    """
2947
    instance = self.instance
2948
    ignore_secondaries = self.op.ignore_secondaries
2949
    reboot_type = self.op.reboot_type
2950

    
2951
    node_current = instance.primary_node
2952

    
2953
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
2954
                       constants.INSTANCE_REBOOT_HARD]:
2955
      for disk in instance.disks:
2956
        self.cfg.SetDiskID(disk, node_current)
2957
      result = self.rpc.call_instance_reboot(node_current, instance,
2958
                                             reboot_type)
2959
      msg = result.RemoteFailMsg()
2960
      if msg:
2961
        raise errors.OpExecError("Could not reboot instance: %s" % msg)
2962
    else:
2963
      result = self.rpc.call_instance_shutdown(node_current, instance)
2964
      msg = result.RemoteFailMsg()
2965
      if msg:
2966
        raise errors.OpExecError("Could not shutdown instance for"
2967
                                 " full reboot: %s" % msg)
2968
      _ShutdownInstanceDisks(self, instance)
2969
      _StartInstanceDisks(self, instance, ignore_secondaries)
2970
      result = self.rpc.call_instance_start(node_current, instance, None, None)
2971
      msg = result.RemoteFailMsg()
2972
      if msg:
2973
        _ShutdownInstanceDisks(self, instance)
2974
        raise errors.OpExecError("Could not start instance for"
2975
                                 " full reboot: %s" % msg)
2976

    
2977
    self.cfg.MarkInstanceUp(instance.name)
2978

    
2979

    
2980
class LUShutdownInstance(LogicalUnit):
2981
  """Shutdown an instance.
2982

2983
  """
2984
  HPATH = "instance-stop"
2985
  HTYPE = constants.HTYPE_INSTANCE
2986
  _OP_REQP = ["instance_name"]
2987
  REQ_BGL = False
2988

    
2989
  def ExpandNames(self):
2990
    self._ExpandAndLockInstance()
2991

    
2992
  def BuildHooksEnv(self):
2993
    """Build hooks env.
2994

2995
    This runs on master, primary and secondary nodes of the instance.
2996

2997
    """
2998
    env = _BuildInstanceHookEnvByObject(self, self.instance)
2999
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3000
    return env, nl, nl
3001

    
3002
  def CheckPrereq(self):
3003
    """Check prerequisites.
3004

3005
    This checks that the instance is in the cluster.
3006

3007
    """
3008
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3009
    assert self.instance is not None, \
3010
      "Cannot retrieve locked instance %s" % self.op.instance_name
3011
    _CheckNodeOnline(self, self.instance.primary_node)
3012

    
3013
  def Exec(self, feedback_fn):
3014
    """Shutdown the instance.
3015

3016
    """
3017
    instance = self.instance
3018
    node_current = instance.primary_node
3019
    self.cfg.MarkInstanceDown(instance.name)
3020
    result = self.rpc.call_instance_shutdown(node_current, instance)
3021
    msg = result.RemoteFailMsg()
3022
    if msg:
3023
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
3024

    
3025
    _ShutdownInstanceDisks(self, instance)
3026

    
3027

    
3028
class LUReinstallInstance(LogicalUnit):
3029
  """Reinstall an instance.
3030

3031
  """
3032
  HPATH = "instance-reinstall"
3033
  HTYPE = constants.HTYPE_INSTANCE
3034
  _OP_REQP = ["instance_name"]
3035
  REQ_BGL = False
3036

    
3037
  def ExpandNames(self):
3038
    self._ExpandAndLockInstance()
3039

    
3040
  def BuildHooksEnv(self):
3041
    """Build hooks env.
3042

3043
    This runs on master, primary and secondary nodes of the instance.
3044

3045
    """
3046
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3047
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3048
    return env, nl, nl
3049

    
3050
  def CheckPrereq(self):
3051
    """Check prerequisites.
3052

3053
    This checks that the instance is in the cluster and is not running.
3054

3055
    """
3056
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3057
    assert instance is not None, \
3058
      "Cannot retrieve locked instance %s" % self.op.instance_name
3059
    _CheckNodeOnline(self, instance.primary_node)
3060

    
3061
    if instance.disk_template == constants.DT_DISKLESS:
3062
      raise errors.OpPrereqError("Instance '%s' has no disks" %
3063
                                 self.op.instance_name)
3064
    if instance.admin_up:
3065
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3066
                                 self.op.instance_name)
3067
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3068
                                              instance.name,
3069
                                              instance.hypervisor)
3070
    remote_info.Raise()
3071
    if remote_info.data:
3072
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3073
                                 (self.op.instance_name,
3074
                                  instance.primary_node))
3075

    
3076
    self.op.os_type = getattr(self.op, "os_type", None)
3077
    if self.op.os_type is not None:
3078
      # OS verification
3079
      pnode = self.cfg.GetNodeInfo(
3080
        self.cfg.ExpandNodeName(instance.primary_node))
3081
      if pnode is None:
3082
        raise errors.OpPrereqError("Primary node '%s' is unknown" %
3083
                                   self.op.pnode)
3084
      result = self.rpc.call_os_get(pnode.name, self.op.os_type)
3085
      result.Raise()
3086
      if not isinstance(result.data, objects.OS):
3087
        raise errors.OpPrereqError("OS '%s' not in supported OS list for"
3088
                                   " primary node"  % self.op.os_type)
3089

    
3090
    self.instance = instance
3091

    
3092
  def Exec(self, feedback_fn):
3093
    """Reinstall the instance.
3094

3095
    """
3096
    inst = self.instance
3097

    
3098
    if self.op.os_type is not None:
3099
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
3100
      inst.os = self.op.os_type
3101
      self.cfg.Update(inst)
3102

    
3103
    _StartInstanceDisks(self, inst, None)
3104
    try:
3105
      feedback_fn("Running the instance OS create scripts...")
3106
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True)
3107
      msg = result.RemoteFailMsg()
3108
      if msg:
3109
        raise errors.OpExecError("Could not install OS for instance %s"
3110
                                 " on node %s: %s" %
3111
                                 (inst.name, inst.primary_node, msg))
3112
    finally:
3113
      _ShutdownInstanceDisks(self, inst)
3114

    
3115

    
3116
class LURenameInstance(LogicalUnit):
3117
  """Rename an instance.
3118

3119
  """
3120
  HPATH = "instance-rename"
3121
  HTYPE = constants.HTYPE_INSTANCE
3122
  _OP_REQP = ["instance_name", "new_name"]
3123

    
3124
  def BuildHooksEnv(self):
3125
    """Build hooks env.
3126

3127
    This runs on master, primary and secondary nodes of the instance.
3128

3129
    """
3130
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3131
    env["INSTANCE_NEW_NAME"] = self.op.new_name
3132
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
3133
    return env, nl, nl
3134

    
3135
  def CheckPrereq(self):
3136
    """Check prerequisites.
3137

3138
    This checks that the instance is in the cluster and is not running.
3139

3140
    """
3141
    instance = self.cfg.GetInstanceInfo(
3142
      self.cfg.ExpandInstanceName(self.op.instance_name))
3143
    if instance is None:
3144
      raise errors.OpPrereqError("Instance '%s' not known" %
3145
                                 self.op.instance_name)
3146
    _CheckNodeOnline(self, instance.primary_node)
3147

    
3148
    if instance.admin_up:
3149
      raise errors.OpPrereqError("Instance '%s' is marked to be up" %
3150
                                 self.op.instance_name)
3151
    remote_info = self.rpc.call_instance_info(instance.primary_node,
3152
                                              instance.name,
3153
                                              instance.hypervisor)
3154
    remote_info.Raise()
3155
    if remote_info.data:
3156
      raise errors.OpPrereqError("Instance '%s' is running on the node %s" %
3157
                                 (self.op.instance_name,
3158
                                  instance.primary_node))
3159
    self.instance = instance
3160

    
3161
    # new name verification
3162
    name_info = utils.HostInfo(self.op.new_name)
3163

    
3164
    self.op.new_name = new_name = name_info.name
3165
    instance_list = self.cfg.GetInstanceList()
3166
    if new_name in instance_list:
3167
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
3168
                                 new_name)
3169

    
3170
    if not getattr(self.op, "ignore_ip", False):
3171
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
3172
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
3173
                                   (name_info.ip, new_name))
3174

    
3175

    
3176
  def Exec(self, feedback_fn):
3177
    """Reinstall the instance.
3178

3179
    """
3180
    inst = self.instance
3181
    old_name = inst.name
3182

    
3183
    if inst.disk_template == constants.DT_FILE:
3184
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3185

    
3186
    self.cfg.RenameInstance(inst.name, self.op.new_name)
3187
    # Change the instance lock. This is definitely safe while we hold the BGL
3188
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
3189
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
3190

    
3191
    # re-read the instance from the configuration after rename
3192
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
3193

    
3194
    if inst.disk_template == constants.DT_FILE:
3195
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
3196
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
3197
                                                     old_file_storage_dir,
3198
                                                     new_file_storage_dir)
3199
      result.Raise()
3200
      if not result.data:
3201
        raise errors.OpExecError("Could not connect to node '%s' to rename"
3202
                                 " directory '%s' to '%s' (but the instance"
3203
                                 " has been renamed in Ganeti)" % (
3204
                                 inst.primary_node, old_file_storage_dir,
3205
                                 new_file_storage_dir))
3206

    
3207
      if not result.data[0]:
3208
        raise errors.OpExecError("Could not rename directory '%s' to '%s'"
3209
                                 " (but the instance has been renamed in"
3210
                                 " Ganeti)" % (old_file_storage_dir,
3211
                                               new_file_storage_dir))
3212

    
3213
    _StartInstanceDisks(self, inst, None)
3214
    try:
3215
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
3216
                                                 old_name)
3217
      msg = result.RemoteFailMsg()
3218
      if msg:
3219
        msg = ("Could not run OS rename script for instance %s on node %s"
3220
               " (but the instance has been renamed in Ganeti): %s" %
3221
               (inst.name, inst.primary_node, msg))
3222
        self.proc.LogWarning(msg)
3223
    finally:
3224
      _ShutdownInstanceDisks(self, inst)
3225

    
3226

    
3227
class LURemoveInstance(LogicalUnit):
3228
  """Remove an instance.
3229

3230
  """
3231
  HPATH = "instance-remove"
3232
  HTYPE = constants.HTYPE_INSTANCE
3233
  _OP_REQP = ["instance_name", "ignore_failures"]
3234
  REQ_BGL = False
3235

    
3236
  def ExpandNames(self):
3237
    self._ExpandAndLockInstance()
3238
    self.needed_locks[locking.LEVEL_NODE] = []
3239
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3240

    
3241
  def DeclareLocks(self, level):
3242
    if level == locking.LEVEL_NODE:
3243
      self._LockInstancesNodes()
3244

    
3245
  def BuildHooksEnv(self):
3246
    """Build hooks env.
3247

3248
    This runs on master, primary and secondary nodes of the instance.
3249

3250
    """
3251
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3252
    nl = [self.cfg.GetMasterNode()]
3253
    return env, nl, nl
3254

    
3255
  def CheckPrereq(self):
3256
    """Check prerequisites.
3257

3258
    This checks that the instance is in the cluster.
3259

3260
    """
3261
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3262
    assert self.instance is not None, \
3263
      "Cannot retrieve locked instance %s" % self.op.instance_name
3264

    
3265
  def Exec(self, feedback_fn):
3266
    """Remove the instance.
3267

3268
    """
3269
    instance = self.instance
3270
    logging.info("Shutting down instance %s on node %s",
3271
                 instance.name, instance.primary_node)
3272

    
3273
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance)
3274
    msg = result.RemoteFailMsg()
3275
    if msg:
3276
      if self.op.ignore_failures:
3277
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
3278
      else:
3279
        raise errors.OpExecError("Could not shutdown instance %s on"
3280
                                 " node %s: %s" %
3281
                                 (instance.name, instance.primary_node, msg))
3282

    
3283
    logging.info("Removing block devices for instance %s", instance.name)
3284

    
3285
    if not _RemoveDisks(self, instance):
3286
      if self.op.ignore_failures:
3287
        feedback_fn("Warning: can't remove instance's disks")
3288
      else:
3289
        raise errors.OpExecError("Can't remove instance's disks")
3290

    
3291
    logging.info("Removing instance %s out of cluster config", instance.name)
3292

    
3293
    self.cfg.RemoveInstance(instance.name)
3294
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
3295

    
3296

    
3297
class LUQueryInstances(NoHooksLU):
3298
  """Logical unit for querying instances.
3299

3300
  """
3301
  _OP_REQP = ["output_fields", "names", "use_locking"]
3302
  REQ_BGL = False
3303
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
3304
                                    "admin_state",
3305
                                    "disk_template", "ip", "mac", "bridge",
3306
                                    "sda_size", "sdb_size", "vcpus", "tags",
3307
                                    "network_port", "beparams",
3308
                                    r"(disk)\.(size)/([0-9]+)",
3309
                                    r"(disk)\.(sizes)", "disk_usage",
3310
                                    r"(nic)\.(mac|ip|bridge)/([0-9]+)",
3311
                                    r"(nic)\.(macs|ips|bridges)",
3312
                                    r"(disk|nic)\.(count)",
3313
                                    "serial_no", "hypervisor", "hvparams",] +
3314
                                  ["hv/%s" % name
3315
                                   for name in constants.HVS_PARAMETERS] +
3316
                                  ["be/%s" % name
3317
                                   for name in constants.BES_PARAMETERS])
3318
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
3319

    
3320

    
3321
  def ExpandNames(self):
3322
    _CheckOutputFields(static=self._FIELDS_STATIC,
3323
                       dynamic=self._FIELDS_DYNAMIC,
3324
                       selected=self.op.output_fields)
3325

    
3326
    self.needed_locks = {}
3327
    self.share_locks[locking.LEVEL_INSTANCE] = 1
3328
    self.share_locks[locking.LEVEL_NODE] = 1
3329

    
3330
    if self.op.names:
3331
      self.wanted = _GetWantedInstances(self, self.op.names)
3332
    else:
3333
      self.wanted = locking.ALL_SET
3334

    
3335
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
3336
    self.do_locking = self.do_node_query and self.op.use_locking
3337
    if self.do_locking:
3338
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
3339
      self.needed_locks[locking.LEVEL_NODE] = []
3340
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3341

    
3342
  def DeclareLocks(self, level):
3343
    if level == locking.LEVEL_NODE and self.do_locking:
3344
      self._LockInstancesNodes()
3345

    
3346
  def CheckPrereq(self):
3347
    """Check prerequisites.
3348

3349
    """
3350
    pass
3351

    
3352
  def Exec(self, feedback_fn):
3353
    """Computes the list of nodes and their attributes.
3354

3355
    """
3356
    all_info = self.cfg.GetAllInstancesInfo()
3357
    if self.wanted == locking.ALL_SET:
3358
      # caller didn't specify instance names, so ordering is not important
3359
      if self.do_locking:
3360
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
3361
      else:
3362
        instance_names = all_info.keys()
3363
      instance_names = utils.NiceSort(instance_names)
3364
    else:
3365
      # caller did specify names, so we must keep the ordering
3366
      if self.do_locking:
3367
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
3368
      else:
3369
        tgt_set = all_info.keys()
3370
      missing = set(self.wanted).difference(tgt_set)
3371
      if missing:
3372
        raise errors.OpExecError("Some instances were removed before"
3373
                                 " retrieving their data: %s" % missing)
3374
      instance_names = self.wanted
3375

    
3376
    instance_list = [all_info[iname] for iname in instance_names]
3377

    
3378
    # begin data gathering
3379

    
3380
    nodes = frozenset([inst.primary_node for inst in instance_list])
3381
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
3382

    
3383
    bad_nodes = []
3384
    off_nodes = []
3385
    if self.do_node_query:
3386
      live_data = {}
3387
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
3388
      for name in nodes:
3389
        result = node_data[name]
3390
        if result.offline:
3391
          # offline nodes will be in both lists
3392
          off_nodes.append(name)
3393
        if result.failed:
3394
          bad_nodes.append(name)
3395
        else:
3396
          if result.data:
3397
            live_data.update(result.data)
3398
            # else no instance is alive
3399
    else:
3400
      live_data = dict([(name, {}) for name in instance_names])
3401

    
3402
    # end data gathering
3403

    
3404
    HVPREFIX = "hv/"
3405
    BEPREFIX = "be/"
3406
    output = []
3407
    for instance in instance_list:
3408
      iout = []
3409
      i_hv = self.cfg.GetClusterInfo().FillHV(instance)
3410
      i_be = self.cfg.GetClusterInfo().FillBE(instance)
3411
      for field in self.op.output_fields:
3412
        st_match = self._FIELDS_STATIC.Matches(field)
3413
        if field == "name":
3414
          val = instance.name
3415
        elif field == "os":
3416
          val = instance.os
3417
        elif field == "pnode":
3418
          val = instance.primary_node
3419
        elif field == "snodes":
3420
          val = list(instance.secondary_nodes)
3421
        elif field == "admin_state":
3422
          val = instance.admin_up
3423
        elif field == "oper_state":
3424
          if instance.primary_node in bad_nodes:
3425
            val = None
3426
          else:
3427
            val = bool(live_data.get(instance.name))
3428
        elif field == "status":
3429
          if instance.primary_node in off_nodes:
3430
            val = "ERROR_nodeoffline"
3431
          elif instance.primary_node in bad_nodes:
3432
            val = "ERROR_nodedown"
3433
          else:
3434
            running = bool(live_data.get(instance.name))
3435
            if running:
3436
              if instance.admin_up:
3437
                val = "running"
3438
              else:
3439
                val = "ERROR_up"
3440
            else:
3441
              if instance.admin_up:
3442
                val = "ERROR_down"
3443
              else:
3444
                val = "ADMIN_down"
3445
        elif field == "oper_ram":
3446
          if instance.primary_node in bad_nodes:
3447
            val = None
3448
          elif instance.name in live_data:
3449
            val = live_data[instance.name].get("memory", "?")
3450
          else:
3451
            val = "-"
3452
        elif field == "disk_template":
3453
          val = instance.disk_template
3454
        elif field == "ip":
3455
          val = instance.nics[0].ip
3456
        elif field == "bridge":
3457
          val = instance.nics[0].bridge
3458
        elif field == "mac":
3459
          val = instance.nics[0].mac
3460
        elif field == "sda_size" or field == "sdb_size":
3461
          idx = ord(field[2]) - ord('a')
3462
          try:
3463
            val = instance.FindDisk(idx).size
3464
          except errors.OpPrereqError:
3465
            val = None
3466
        elif field == "disk_usage": # total disk usage per node
3467
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
3468
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
3469
        elif field == "tags":
3470
          val = list(instance.GetTags())
3471
        elif field == "serial_no":
3472
          val = instance.serial_no
3473
        elif field == "network_port":
3474
          val = instance.network_port
3475
        elif field == "hypervisor":
3476
          val = instance.hypervisor
3477
        elif field == "hvparams":
3478
          val = i_hv
3479
        elif (field.startswith(HVPREFIX) and
3480
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS):
3481
          val = i_hv.get(field[len(HVPREFIX):], None)
3482
        elif field == "beparams":
3483
          val = i_be
3484
        elif (field.startswith(BEPREFIX) and
3485
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
3486
          val = i_be.get(field[len(BEPREFIX):], None)
3487
        elif st_match and st_match.groups():
3488
          # matches a variable list
3489
          st_groups = st_match.groups()
3490
          if st_groups and st_groups[0] == "disk":
3491
            if st_groups[1] == "count":
3492
              val = len(instance.disks)
3493
            elif st_groups[1] == "sizes":
3494
              val = [disk.size for disk in instance.disks]
3495
            elif st_groups[1] == "size":
3496
              try:
3497
                val = instance.FindDisk(st_groups[2]).size
3498
              except errors.OpPrereqError:
3499
                val = None
3500
            else:
3501
              assert False, "Unhandled disk parameter"
3502
          elif st_groups[0] == "nic":
3503
            if st_groups[1] == "count":
3504
              val = len(instance.nics)
3505
            elif st_groups[1] == "macs":
3506
              val = [nic.mac for nic in instance.nics]
3507
            elif st_groups[1] == "ips":
3508
              val = [nic.ip for nic in instance.nics]
3509
            elif st_groups[1] == "bridges":
3510
              val = [nic.bridge for nic in instance.nics]
3511
            else:
3512
              # index-based item
3513
              nic_idx = int(st_groups[2])
3514
              if nic_idx >= len(instance.nics):
3515
                val = None
3516
              else:
3517
                if st_groups[1] == "mac":
3518
                  val = instance.nics[nic_idx].mac
3519
                elif st_groups[1] == "ip":
3520
                  val = instance.nics[nic_idx].ip
3521
                elif st_groups[1] == "bridge":
3522
                  val = instance.nics[nic_idx].bridge
3523
                else:
3524
                  assert False, "Unhandled NIC parameter"
3525
          else:
3526
            assert False, "Unhandled variable parameter"
3527
        else:
3528
          raise errors.ParameterError(field)
3529
        iout.append(val)
3530
      output.append(iout)
3531

    
3532
    return output
3533

    
3534

    
3535
class LUFailoverInstance(LogicalUnit):
3536
  """Failover an instance.
3537

3538
  """
3539
  HPATH = "instance-failover"
3540
  HTYPE = constants.HTYPE_INSTANCE
3541
  _OP_REQP = ["instance_name", "ignore_consistency"]
3542
  REQ_BGL = False
3543

    
3544
  def ExpandNames(self):
3545
    self._ExpandAndLockInstance()
3546
    self.needed_locks[locking.LEVEL_NODE] = []
3547
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3548

    
3549
  def DeclareLocks(self, level):
3550
    if level == locking.LEVEL_NODE:
3551
      self._LockInstancesNodes()
3552

    
3553
  def BuildHooksEnv(self):
3554
    """Build hooks env.
3555

3556
    This runs on master, primary and secondary nodes of the instance.
3557

3558
    """
3559
    env = {
3560
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
3561
      }
3562
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
3563
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3564
    return env, nl, nl
3565

    
3566
  def CheckPrereq(self):
3567
    """Check prerequisites.
3568

3569
    This checks that the instance is in the cluster.
3570

3571
    """
3572
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3573
    assert self.instance is not None, \
3574
      "Cannot retrieve locked instance %s" % self.op.instance_name
3575

    
3576
    bep = self.cfg.GetClusterInfo().FillBE(instance)
3577
    if instance.disk_template not in constants.DTS_NET_MIRROR:
3578
      raise errors.OpPrereqError("Instance's disk layout is not"
3579
                                 " network mirrored, cannot failover.")
3580

    
3581
    secondary_nodes = instance.secondary_nodes
3582
    if not secondary_nodes:
3583
      raise errors.ProgrammerError("no secondary node but using "
3584
                                   "a mirrored disk template")
3585

    
3586
    target_node = secondary_nodes[0]
3587
    _CheckNodeOnline(self, target_node)
3588
    _CheckNodeNotDrained(self, target_node)
3589
    # check memory requirements on the secondary node
3590
    _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
3591
                         instance.name, bep[constants.BE_MEMORY],
3592
                         instance.hypervisor)
3593

    
3594
    # check bridge existance
3595
    brlist = [nic.bridge for nic in instance.nics]
3596
    result = self.rpc.call_bridges_exist(target_node, brlist)
3597
    result.Raise()
3598
    if not result.data:
3599
      raise errors.OpPrereqError("One or more target bridges %s does not"
3600
                                 " exist on destination node '%s'" %
3601
                                 (brlist, target_node))
3602

    
3603
  def Exec(self, feedback_fn):
3604
    """Failover an instance.
3605

3606
    The failover is done by shutting it down on its present node and
3607
    starting it on the secondary.
3608

3609
    """
3610
    instance = self.instance
3611

    
3612
    source_node = instance.primary_node
3613
    target_node = instance.secondary_nodes[0]
3614

    
3615
    feedback_fn("* checking disk consistency between source and target")
3616
    for dev in instance.disks:
3617
      # for drbd, these are drbd over lvm
3618
      if not _CheckDiskConsistency(self, dev, target_node, False):
3619
        if instance.admin_up and not self.op.ignore_consistency:
3620
          raise errors.OpExecError("Disk %s is degraded on target node,"
3621
                                   " aborting failover." % dev.iv_name)
3622

    
3623
    feedback_fn("* shutting down instance on source node")
3624
    logging.info("Shutting down instance %s on node %s",
3625
                 instance.name, source_node)
3626

    
3627
    result = self.rpc.call_instance_shutdown(source_node, instance)
3628
    msg = result.RemoteFailMsg()
3629
    if msg:
3630
      if self.op.ignore_consistency:
3631
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
3632
                             " Proceeding anyway. Please make sure node"
3633
                             " %s is down. Error details: %s",
3634
                             instance.name, source_node, source_node, msg)
3635
      else:
3636
        raise errors.OpExecError("Could not shutdown instance %s on"
3637
                                 " node %s: %s" %
3638
                                 (instance.name, source_node, msg))
3639

    
3640
    feedback_fn("* deactivating the instance's disks on source node")
3641
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
3642
      raise errors.OpExecError("Can't shut down the instance's disks.")
3643

    
3644
    instance.primary_node = target_node
3645
    # distribute new instance config to the other nodes
3646
    self.cfg.Update(instance)
3647

    
3648
    # Only start the instance if it's marked as up
3649
    if instance.admin_up:
3650
      feedback_fn("* activating the instance's disks on target node")
3651
      logging.info("Starting instance %s on node %s",
3652
                   instance.name, target_node)
3653

    
3654
      disks_ok, dummy = _AssembleInstanceDisks(self, instance,
3655
                                               ignore_secondaries=True)
3656
      if not disks_ok:
3657
        _ShutdownInstanceDisks(self, instance)
3658
        raise errors.OpExecError("Can't activate the instance's disks")
3659

    
3660
      feedback_fn("* starting the instance on the target node")
3661
      result = self.rpc.call_instance_start(target_node, instance, None, None)
3662
      msg = result.RemoteFailMsg()
3663
      if msg:
3664
        _ShutdownInstanceDisks(self, instance)
3665
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
3666
                                 (instance.name, target_node, msg))
3667

    
3668

    
3669
class LUMigrateInstance(LogicalUnit):
3670
  """Migrate an instance.
3671

3672
  This is migration without shutting down, compared to the failover,
3673
  which is done with shutdown.
3674

3675
  """
3676
  HPATH = "instance-migrate"
3677
  HTYPE = constants.HTYPE_INSTANCE
3678
  _OP_REQP = ["instance_name", "live", "cleanup"]
3679

    
3680
  REQ_BGL = False
3681

    
3682
  def ExpandNames(self):
3683
    self._ExpandAndLockInstance()
3684
    self.needed_locks[locking.LEVEL_NODE] = []
3685
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3686

    
3687
  def DeclareLocks(self, level):
3688
    if level == locking.LEVEL_NODE:
3689
      self._LockInstancesNodes()
3690

    
3691
  def BuildHooksEnv(self):
3692
    """Build hooks env.
3693

3694
    This runs on master, primary and secondary nodes of the instance.
3695

3696
    """
3697
    env = _BuildInstanceHookEnvByObject(self, self.instance)
3698
    env["MIGRATE_LIVE"] = self.op.live
3699
    env["MIGRATE_CLEANUP"] = self.op.cleanup
3700
    nl = [self.cfg.GetMasterNode()] + list(self.instance.secondary_nodes)
3701
    return env, nl, nl
3702

    
3703
  def CheckPrereq(self):
3704
    """Check prerequisites.
3705

3706
    This checks that the instance is in the cluster.
3707

3708
    """
3709
    instance = self.cfg.GetInstanceInfo(
3710
      self.cfg.ExpandInstanceName(self.op.instance_name))
3711
    if instance is None:
3712
      raise errors.OpPrereqError("Instance '%s' not known" %
3713
                                 self.op.instance_name)
3714

    
3715
    if instance.disk_template != constants.DT_DRBD8:
3716
      raise errors.OpPrereqError("Instance's disk layout is not"
3717
                                 " drbd8, cannot migrate.")
3718

    
3719
    secondary_nodes = instance.secondary_nodes
3720
    if not secondary_nodes:
3721
      raise errors.ConfigurationError("No secondary node but using"
3722
                                      " drbd8 disk template")
3723

    
3724
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
3725

    
3726
    target_node = secondary_nodes[0]
3727
    # check memory requirements on the secondary node
3728
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
3729
                         instance.name, i_be[constants.BE_MEMORY],
3730
                         instance.hypervisor)
3731

    
3732
    # check bridge existance
3733
    brlist = [nic.bridge for nic in instance.nics]
3734
    result = self.rpc.call_bridges_exist(target_node, brlist)
3735
    if result.failed or not result.data:
3736
      raise errors.OpPrereqError("One or more target bridges %s does not"
3737
                                 " exist on destination node '%s'" %
3738
                                 (brlist, target_node))
3739

    
3740
    if not self.op.cleanup:
3741
      _CheckNodeNotDrained(self, target_node)
3742
      result = self.rpc.call_instance_migratable(instance.primary_node,
3743
                                                 instance)
3744
      msg = result.RemoteFailMsg()
3745
      if msg:
3746
        raise errors.OpPrereqError("Can't migrate: %s - please use failover" %
3747
                                   msg)
3748

    
3749
    self.instance = instance
3750

    
3751
  def _WaitUntilSync(self):
3752
    """Poll with custom rpc for disk sync.
3753

3754
    This uses our own step-based rpc call.
3755

3756
    """
3757
    self.feedback_fn("* wait until resync is done")
3758
    all_done = False
3759
    while not all_done:
3760
      all_done = True
3761
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
3762
                                            self.nodes_ip,
3763
                                            self.instance.disks)
3764
      min_percent = 100
3765
      for node, nres in result.items():
3766
        msg = nres.RemoteFailMsg()
3767
        if msg:
3768
          raise errors.OpExecError("Cannot resync disks on node %s: %s" %
3769
                                   (node, msg))
3770
        node_done, node_percent = nres.payload
3771
        all_done = all_done and node_done
3772
        if node_percent is not None:
3773
          min_percent = min(min_percent, node_percent)
3774
      if not all_done:
3775
        if min_percent < 100:
3776
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
3777
        time.sleep(2)
3778

    
3779
  def _EnsureSecondary(self, node):
3780
    """Demote a node to secondary.
3781

3782
    """
3783
    self.feedback_fn("* switching node %s to secondary mode" % node)
3784

    
3785
    for dev in self.instance.disks:
3786
      self.cfg.SetDiskID(dev, node)
3787

    
3788
    result = self.rpc.call_blockdev_close(node, self.instance.name,
3789
                                          self.instance.disks)
3790
    msg = result.RemoteFailMsg()
3791
    if msg:
3792
      raise errors.OpExecError("Cannot change disk to secondary on node %s,"
3793
                               " error %s" % (node, msg))
3794

    
3795
  def _GoStandalone(self):
3796
    """Disconnect from the network.
3797

3798
    """
3799
    self.feedback_fn("* changing into standalone mode")
3800
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
3801
                                               self.instance.disks)
3802
    for node, nres in result.items():
3803
      msg = nres.RemoteFailMsg()
3804
      if msg:
3805
        raise errors.OpExecError("Cannot disconnect disks node %s,"
3806
                                 " error %s" % (node, msg))
3807

    
3808
  def _GoReconnect(self, multimaster):
3809
    """Reconnect to the network.
3810

3811
    """
3812
    if multimaster:
3813
      msg = "dual-master"
3814
    else:
3815
      msg = "single-master"
3816
    self.feedback_fn("* changing disks into %s mode" % msg)
3817
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
3818
                                           self.instance.disks,
3819
                                           self.instance.name, multimaster)
3820
    for node, nres in result.items():
3821
      msg = nres.RemoteFailMsg()
3822
      if msg:
3823
        raise errors.OpExecError("Cannot change disks config on node %s,"
3824
                                 " error: %s" % (node, msg))
3825

    
3826
  def _ExecCleanup(self):
3827
    """Try to cleanup after a failed migration.
3828

3829
    The cleanup is done by:
3830
      - check that the instance is running only on one node
3831
        (and update the config if needed)
3832
      - change disks on its secondary node to secondary
3833
      - wait until disks are fully synchronized
3834
      - disconnect from the network
3835
      - change disks into single-master mode
3836
      - wait again until disks are fully synchronized
3837

3838
    """
3839
    instance = self.instance
3840
    target_node = self.target_node
3841
    source_node = self.source_node
3842

    
3843
    # check running on only one node
3844
    self.feedback_fn("* checking where the instance actually runs"
3845
                     " (if this hangs, the hypervisor might be in"
3846
                     " a bad state)")
3847
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
3848
    for node, result in ins_l.items():
3849
      result.Raise()
3850
      if not isinstance(result.data, list):
3851
        raise errors.OpExecError("Can't contact node '%s'" % node)
3852

    
3853
    runningon_source = instance.name in ins_l[source_node].data
3854
    runningon_target = instance.name in ins_l[target_node].data
3855

    
3856
    if runningon_source and runningon_target:
3857
      raise errors.OpExecError("Instance seems to be running on two nodes,"
3858
                               " or the hypervisor is confused. You will have"
3859
                               " to ensure manually that it runs only on one"
3860
                               " and restart this operation.")
3861

    
3862
    if not (runningon_source or runningon_target):
3863
      raise errors.OpExecError("Instance does not seem to be running at all."
3864
                               " In this case, it's safer to repair by"
3865
                               " running 'gnt-instance stop' to ensure disk"
3866
                               " shutdown, and then restarting it.")
3867

    
3868
    if runningon_target:
3869
      # the migration has actually succeeded, we need to update the config
3870
      self.feedback_fn("* instance running on secondary node (%s),"
3871
                       " updating config" % target_node)
3872
      instance.primary_node = target_node
3873
      self.cfg.Update(instance)
3874
      demoted_node = source_node
3875
    else:
3876
      self.feedback_fn("* instance confirmed to be running on its"
3877
                       " primary node (%s)" % source_node)
3878
      demoted_node = target_node
3879

    
3880
    self._EnsureSecondary(demoted_node)
3881
    try:
3882
      self._WaitUntilSync()
3883
    except errors.OpExecError:
3884
      # we ignore here errors, since if the device is standalone, it
3885
      # won't be able to sync
3886
      pass
3887
    self._GoStandalone()
3888
    self._GoReconnect(False)
3889
    self._WaitUntilSync()
3890

    
3891
    self.feedback_fn("* done")
3892

    
3893
  def _RevertDiskStatus(self):
3894
    """Try to revert the disk status after a failed migration.
3895

3896
    """
3897
    target_node = self.target_node
3898
    try:
3899
      self._EnsureSecondary(target_node)
3900
      self._GoStandalone()
3901
      self._GoReconnect(False)
3902
      self._WaitUntilSync()
3903
    except errors.OpExecError, err:
3904
      self.LogWarning("Migration failed and I can't reconnect the"
3905
                      " drives: error '%s'\n"
3906
                      "Please look and recover the instance status" %
3907
                      str(err))
3908

    
3909
  def _AbortMigration(self):
3910
    """Call the hypervisor code to abort a started migration.
3911

3912
    """
3913
    instance = self.instance
3914
    target_node = self.target_node
3915
    migration_info = self.migration_info
3916

    
3917
    abort_result = self.rpc.call_finalize_migration(target_node,
3918
                                                    instance,
3919
                                                    migration_info,
3920
                                                    False)
3921
    abort_msg = abort_result.RemoteFailMsg()
3922
    if abort_msg:
3923
      logging.error("Aborting migration failed on target node %s: %s" %
3924
                    (target_node, abort_msg))
3925
      # Don't raise an exception here, as we stil have to try to revert the
3926
      # disk status, even if this step failed.
3927

    
3928
  def _ExecMigration(self):
3929
    """Migrate an instance.
3930

3931
    The migrate is done by:
3932
      - change the disks into dual-master mode
3933
      - wait until disks are fully synchronized again
3934
      - migrate the instance
3935
      - change disks on the new secondary node (the old primary) to secondary
3936
      - wait until disks are fully synchronized
3937
      - change disks into single-master mode
3938

3939
    """
3940
    instance = self.instance
3941
    target_node = self.target_node
3942
    source_node = self.source_node
3943

    
3944
    self.feedback_fn("* checking disk consistency between source and target")
3945
    for dev in instance.disks:
3946
      if not _CheckDiskConsistency(self, dev, target_node, False):
3947
        raise errors.OpExecError("Disk %s is degraded or not fully"
3948
                                 " synchronized on target node,"
3949
                                 " aborting migrate." % dev.iv_name)
3950

    
3951
    # First get the migration information from the remote node
3952
    result = self.rpc.call_migration_info(source_node, instance)
3953
    msg = result.RemoteFailMsg()
3954
    if msg:
3955
      log_err = ("Failed fetching source migration information from %s: %s" %
3956
                 (source_node, msg))
3957
      logging.error(log_err)
3958
      raise errors.OpExecError(log_err)
3959

    
3960
    self.migration_info = migration_info = result.payload
3961

    
3962
    # Then switch the disks to master/master mode
3963
    self._EnsureSecondary(target_node)
3964
    self._GoStandalone()
3965
    self._GoReconnect(True)
3966
    self._WaitUntilSync()
3967

    
3968
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
3969
    result = self.rpc.call_accept_instance(target_node,
3970
                                           instance,
3971
                                           migration_info,
3972
                                           self.nodes_ip[target_node])
3973

    
3974
    msg = result.RemoteFailMsg()
3975
    if msg:
3976
      logging.error("Instance pre-migration failed, trying to revert"
3977
                    " disk status: %s", msg)
3978
      self._AbortMigration()
3979
      self._RevertDiskStatus()
3980
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
3981
                               (instance.name, msg))
3982

    
3983
    self.feedback_fn("* migrating instance to %s" % target_node)
3984
    time.sleep(10)
3985
    result = self.rpc.call_instance_migrate(source_node, instance,
3986
                                            self.nodes_ip[target_node],
3987
                                            self.op.live)
3988
    msg = result.RemoteFailMsg()
3989
    if msg:
3990
      logging.error("Instance migration failed, trying to revert"
3991
                    " disk status: %s", msg)
3992
      self._AbortMigration()
3993
      self._RevertDiskStatus()
3994
      raise errors.OpExecError("Could not migrate instance %s: %s" %
3995
                               (instance.name, msg))
3996
    time.sleep(10)
3997

    
3998
    instance.primary_node = target_node
3999
    # distribute new instance config to the other nodes
4000
    self.cfg.Update(instance)
4001

    
4002
    result = self.rpc.call_finalize_migration(target_node,
4003
                                              instance,
4004
                                              migration_info,
4005
                                              True)
4006
    msg = result.RemoteFailMsg()
4007
    if msg:
4008
      logging.error("Instance migration succeeded, but finalization failed:"
4009
                    " %s" % msg)
4010
      raise errors.OpExecError("Could not finalize instance migration: %s" %
4011
                               msg)
4012

    
4013
    self._EnsureSecondary(source_node)
4014
    self._WaitUntilSync()
4015
    self._GoStandalone()
4016
    self._GoReconnect(False)
4017
    self._WaitUntilSync()
4018

    
4019
    self.feedback_fn("* done")
4020

    
4021
  def Exec(self, feedback_fn):
4022
    """Perform the migration.
4023

4024
    """
4025
    self.feedback_fn = feedback_fn
4026

    
4027
    self.source_node = self.instance.primary_node
4028
    self.target_node = self.instance.secondary_nodes[0]
4029
    self.all_nodes = [self.source_node, self.target_node]
4030
    self.nodes_ip = {
4031
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
4032
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
4033
      }
4034
    if self.op.cleanup:
4035
      return self._ExecCleanup()
4036
    else:
4037
      return self._ExecMigration()
4038

    
4039

    
4040
def _CreateBlockDev(lu, node, instance, device, force_create,
4041
                    info, force_open):
4042
  """Create a tree of block devices on a given node.
4043

4044
  If this device type has to be created on secondaries, create it and
4045
  all its children.
4046

4047
  If not, just recurse to children keeping the same 'force' value.
4048

4049
  @param lu: the lu on whose behalf we execute
4050
  @param node: the node on which to create the device
4051
  @type instance: L{objects.Instance}
4052
  @param instance: the instance which owns the device
4053
  @type device: L{objects.Disk}
4054
  @param device: the device to create
4055
  @type force_create: boolean
4056
  @param force_create: whether to force creation of this device; this
4057
      will be change to True whenever we find a device which has
4058
      CreateOnSecondary() attribute
4059
  @param info: the extra 'metadata' we should attach to the device
4060
      (this will be represented as a LVM tag)
4061
  @type force_open: boolean
4062
  @param force_open: this parameter will be passes to the
4063
      L{backend.BlockdevCreate} function where it specifies
4064
      whether we run on primary or not, and it affects both
4065
      the child assembly and the device own Open() execution
4066

4067
  """
4068
  if device.CreateOnSecondary():
4069
    force_create = True
4070

    
4071
  if device.children:
4072
    for child in device.children:
4073
      _CreateBlockDev(lu, node, instance, child, force_create,
4074
                      info, force_open)
4075

    
4076
  if not force_create:
4077
    return
4078

    
4079
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
4080

    
4081

    
4082
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
4083
  """Create a single block device on a given node.
4084

4085
  This will not recurse over children of the device, so they must be
4086
  created in advance.
4087

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

4102
  """
4103
  lu.cfg.SetDiskID(device, node)
4104
  result = lu.rpc.call_blockdev_create(node, device, device.size,
4105
                                       instance.name, force_open, info)
4106
  msg = result.RemoteFailMsg()
4107
  if msg:
4108
    raise errors.OpExecError("Can't create block device %s on"
4109
                             " node %s for instance %s: %s" %
4110
                             (device, node, instance.name, msg))
4111
  if device.physical_id is None:
4112
    device.physical_id = result.payload
4113

    
4114

    
4115
def _GenerateUniqueNames(lu, exts):
4116
  """Generate a suitable LV name.
4117

4118
  This will generate a logical volume name for the given instance.
4119

4120
  """
4121
  results = []
4122
  for val in exts:
4123
    new_id = lu.cfg.GenerateUniqueID()
4124
    results.append("%s%s" % (new_id, val))
4125
  return results
4126

    
4127

    
4128
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
4129
                         p_minor, s_minor):
4130
  """Generate a drbd8 device complete with its children.
4131

4132
  """
4133
  port = lu.cfg.AllocatePort()
4134
  vgname = lu.cfg.GetVGName()
4135
  shared_secret = lu.cfg.GenerateDRBDSecret()
4136
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
4137
                          logical_id=(vgname, names[0]))
4138
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
4139
                          logical_id=(vgname, names[1]))
4140
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
4141
                          logical_id=(primary, secondary, port,
4142
                                      p_minor, s_minor,
4143
                                      shared_secret),
4144
                          children=[dev_data, dev_meta],
4145
                          iv_name=iv_name)
4146
  return drbd_dev
4147

    
4148

    
4149
def _GenerateDiskTemplate(lu, template_name,
4150
                          instance_name, primary_node,
4151
                          secondary_nodes, disk_info,
4152
                          file_storage_dir, file_driver,
4153
                          base_index):
4154
  """Generate the entire disk layout for a given template type.
4155

4156
  """
4157
  #TODO: compute space requirements
4158

    
4159
  vgname = lu.cfg.GetVGName()
4160
  disk_count = len(disk_info)
4161
  disks = []
4162
  if template_name == constants.DT_DISKLESS:
4163
    pass
4164
  elif template_name == constants.DT_PLAIN:
4165
    if len(secondary_nodes) != 0:
4166
      raise errors.ProgrammerError("Wrong template configuration")
4167

    
4168
    names = _GenerateUniqueNames(lu, [".disk%d" % i
4169
                                      for i in range(disk_count)])
4170
    for idx, disk in enumerate(disk_info):
4171
      disk_index = idx + base_index
4172
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
4173
                              logical_id=(vgname, names[idx]),
4174
                              iv_name="disk/%d" % disk_index,
4175
                              mode=disk["mode"])
4176
      disks.append(disk_dev)
4177
  elif template_name == constants.DT_DRBD8:
4178
    if len(secondary_nodes) != 1:
4179
      raise errors.ProgrammerError("Wrong template configuration")
4180
    remote_node = secondary_nodes[0]
4181
    minors = lu.cfg.AllocateDRBDMinor(
4182
      [primary_node, remote_node] * len(disk_info), instance_name)
4183

    
4184
    names = []
4185
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % i
4186
                                               for i in range(disk_count)]):
4187
      names.append(lv_prefix + "_data")
4188
      names.append(lv_prefix + "_meta")
4189
    for idx, disk in enumerate(disk_info):
4190
      disk_index = idx + base_index
4191
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
4192
                                      disk["size"], names[idx*2:idx*2+2],
4193
                                      "disk/%d" % disk_index,
4194
                                      minors[idx*2], minors[idx*2+1])
4195
      disk_dev.mode = disk["mode"]
4196
      disks.append(disk_dev)
4197
  elif template_name == constants.DT_FILE:
4198
    if len(secondary_nodes) != 0:
4199
      raise errors.ProgrammerError("Wrong template configuration")
4200

    
4201
    for idx, disk in enumerate(disk_info):
4202
      disk_index = idx + base_index
4203
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
4204
                              iv_name="disk/%d" % disk_index,
4205
                              logical_id=(file_driver,
4206
                                          "%s/disk%d" % (file_storage_dir,
4207
                                                         disk_index)),
4208
                              mode=disk["mode"])
4209
      disks.append(disk_dev)
4210
  else:
4211
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
4212
  return disks
4213

    
4214

    
4215
def _GetInstanceInfoText(instance):
4216
  """Compute that text that should be added to the disk's metadata.
4217

4218
  """
4219
  return "originstname+%s" % instance.name
4220

    
4221

    
4222
def _CreateDisks(lu, instance):
4223
  """Create all disks for an instance.
4224

4225
  This abstracts away some work from AddInstance.
4226

4227
  @type lu: L{LogicalUnit}
4228
  @param lu: the logical unit on whose behalf we execute
4229
  @type instance: L{objects.Instance}
4230
  @param instance: the instance whose disks we should create
4231
  @rtype: boolean
4232
  @return: the success of the creation
4233

4234
  """
4235
  info = _GetInstanceInfoText(instance)
4236
  pnode = instance.primary_node
4237

    
4238
  if instance.disk_template == constants.DT_FILE:
4239
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4240
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
4241

    
4242
    if result.failed or not result.data:
4243
      raise errors.OpExecError("Could not connect to node '%s'" % pnode)
4244

    
4245
    if not result.data[0]:
4246
      raise errors.OpExecError("Failed to create directory '%s'" %
4247
                               file_storage_dir)
4248

    
4249
  # Note: this needs to be kept in sync with adding of disks in
4250
  # LUSetInstanceParams
4251
  for device in instance.disks:
4252
    logging.info("Creating volume %s for instance %s",
4253
                 device.iv_name, instance.name)
4254
    #HARDCODE
4255
    for node in instance.all_nodes:
4256
      f_create = node == pnode
4257
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
4258

    
4259

    
4260
def _RemoveDisks(lu, instance):
4261
  """Remove all disks for an instance.
4262

4263
  This abstracts away some work from `AddInstance()` and
4264
  `RemoveInstance()`. Note that in case some of the devices couldn't
4265
  be removed, the removal will continue with the other ones (compare
4266
  with `_CreateDisks()`).
4267

4268
  @type lu: L{LogicalUnit}
4269
  @param lu: the logical unit on whose behalf we execute
4270
  @type instance: L{objects.Instance}
4271
  @param instance: the instance whose disks we should remove
4272
  @rtype: boolean
4273
  @return: the success of the removal
4274

4275
  """
4276
  logging.info("Removing block devices for instance %s", instance.name)
4277

    
4278
  all_result = True
4279
  for device in instance.disks:
4280
    for node, disk in device.ComputeNodeTree(instance.primary_node):
4281
      lu.cfg.SetDiskID(disk, node)
4282
      msg = lu.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
4283
      if msg:
4284
        lu.LogWarning("Could not remove block device %s on node %s,"
4285
                      " continuing anyway: %s", device.iv_name, node, msg)
4286
        all_result = False
4287

    
4288
  if instance.disk_template == constants.DT_FILE:
4289
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
4290
    result = lu.rpc.call_file_storage_dir_remove(instance.primary_node,
4291
                                                 file_storage_dir)
4292
    if result.failed or not result.data:
4293
      logging.error("Could not remove directory '%s'", file_storage_dir)
4294
      all_result = False
4295

    
4296
  return all_result
4297

    
4298

    
4299
def _ComputeDiskSize(disk_template, disks):
4300
  """Compute disk size requirements in the volume group
4301

4302
  """
4303
  # Required free disk space as a function of disk and swap space
4304
  req_size_dict = {
4305
    constants.DT_DISKLESS: None,
4306
    constants.DT_PLAIN: sum(d["size"] for d in disks),
4307
    # 128 MB are added for drbd metadata for each disk
4308
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
4309
    constants.DT_FILE: None,
4310
  }
4311

    
4312
  if disk_template not in req_size_dict:
4313
    raise errors.ProgrammerError("Disk template '%s' size requirement"
4314
                                 " is unknown" %  disk_template)
4315

    
4316
  return req_size_dict[disk_template]
4317

    
4318

    
4319
def _CheckHVParams(lu, nodenames, hvname, hvparams):
4320
  """Hypervisor parameter validation.
4321

4322
  This function abstract the hypervisor parameter validation to be
4323
  used in both instance create and instance modify.
4324

4325
  @type lu: L{LogicalUnit}
4326
  @param lu: the logical unit for which we check
4327
  @type nodenames: list
4328
  @param nodenames: the list of nodes on which we should check
4329
  @type hvname: string
4330
  @param hvname: the name of the hypervisor we should use
4331
  @type hvparams: dict
4332
  @param hvparams: the parameters which we need to check
4333
  @raise errors.OpPrereqError: if the parameters are not valid
4334

4335
  """
4336
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
4337
                                                  hvname,
4338
                                                  hvparams)
4339
  for node in nodenames:
4340
    info = hvinfo[node]
4341
    if info.offline:
4342
      continue
4343
    msg = info.RemoteFailMsg()
4344
    if msg:
4345
      raise errors.OpPrereqError("Hypervisor parameter validation"
4346
                                 " failed on node %s: %s" % (node, msg))
4347

    
4348

    
4349
class LUCreateInstance(LogicalUnit):
4350
  """Create an instance.
4351

4352
  """
4353
  HPATH = "instance-add"
4354
  HTYPE = constants.HTYPE_INSTANCE
4355
  _OP_REQP = ["instance_name", "disks", "disk_template",
4356
              "mode", "start",
4357
              "wait_for_sync", "ip_check", "nics",
4358
              "hvparams", "beparams"]
4359
  REQ_BGL = False
4360

    
4361
  def _ExpandNode(self, node):
4362
    """Expands and checks one node name.
4363

4364
    """
4365
    node_full = self.cfg.ExpandNodeName(node)
4366
    if node_full is None:
4367
      raise errors.OpPrereqError("Unknown node %s" % node)
4368
    return node_full
4369

    
4370
  def ExpandNames(self):
4371
    """ExpandNames for CreateInstance.
4372

4373
    Figure out the right locks for instance creation.
4374

4375
    """
4376
    self.needed_locks = {}
4377

    
4378
    # set optional parameters to none if they don't exist
4379
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
4380
      if not hasattr(self.op, attr):
4381
        setattr(self.op, attr, None)
4382

    
4383
    # cheap checks, mostly valid constants given
4384

    
4385
    # verify creation mode
4386
    if self.op.mode not in (constants.INSTANCE_CREATE,
4387
                            constants.INSTANCE_IMPORT):
4388
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
4389
                                 self.op.mode)
4390

    
4391
    # disk template and mirror node verification
4392
    if self.op.disk_template not in constants.DISK_TEMPLATES:
4393
      raise errors.OpPrereqError("Invalid disk template name")
4394

    
4395
    if self.op.hypervisor is None:
4396
      self.op.hypervisor = self.cfg.GetHypervisorType()
4397

    
4398
    cluster = self.cfg.GetClusterInfo()
4399
    enabled_hvs = cluster.enabled_hypervisors
4400
    if self.op.hypervisor not in enabled_hvs:
4401
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
4402
                                 " cluster (%s)" % (self.op.hypervisor,
4403
                                  ",".join(enabled_hvs)))
4404

    
4405
    # check hypervisor parameter syntax (locally)
4406
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
4407
    filled_hvp = cluster.FillDict(cluster.hvparams[self.op.hypervisor],
4408
                                  self.op.hvparams)
4409
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
4410
    hv_type.CheckParameterSyntax(filled_hvp)
4411

    
4412
    # fill and remember the beparams dict
4413
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
4414
    self.be_full = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
4415
                                    self.op.beparams)
4416

    
4417
    #### instance parameters check
4418

    
4419
    # instance name verification
4420
    hostname1 = utils.HostInfo(self.op.instance_name)
4421
    self.op.instance_name = instance_name = hostname1.name
4422

    
4423
    # this is just a preventive check, but someone might still add this
4424
    # instance in the meantime, and creation will fail at lock-add time
4425
    if instance_name in self.cfg.GetInstanceList():
4426
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4427
                                 instance_name)
4428

    
4429
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
4430

    
4431
    # NIC buildup
4432
    self.nics = []
4433
    for nic in self.op.nics:
4434
      # ip validity checks
4435
      ip = nic.get("ip", None)
4436
      if ip is None or ip.lower() == "none":
4437
        nic_ip = None
4438
      elif ip.lower() == constants.VALUE_AUTO:
4439
        nic_ip = hostname1.ip
4440
      else:
4441
        if not utils.IsValidIP(ip):
4442
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
4443
                                     " like a valid IP" % ip)
4444
        nic_ip = ip
4445

    
4446
      # MAC address verification
4447
      mac = nic.get("mac", constants.VALUE_AUTO)
4448
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4449
        if not utils.IsValidMac(mac.lower()):
4450
          raise errors.OpPrereqError("Invalid MAC address specified: %s" %
4451
                                     mac)
4452
      # bridge verification
4453
      bridge = nic.get("bridge", None)
4454
      if bridge is None:
4455
        bridge = self.cfg.GetDefBridge()
4456
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, bridge=bridge))
4457

    
4458
    # disk checks/pre-build
4459
    self.disks = []
4460
    for disk in self.op.disks:
4461
      mode = disk.get("mode", constants.DISK_RDWR)
4462
      if mode not in constants.DISK_ACCESS_SET:
4463
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
4464
                                   mode)
4465
      size = disk.get("size", None)
4466
      if size is None:
4467
        raise errors.OpPrereqError("Missing disk size")
4468
      try:
4469
        size = int(size)
4470
      except ValueError:
4471
        raise errors.OpPrereqError("Invalid disk size '%s'" % size)
4472
      self.disks.append({"size": size, "mode": mode})
4473

    
4474
    # used in CheckPrereq for ip ping check
4475
    self.check_ip = hostname1.ip
4476

    
4477
    # file storage checks
4478
    if (self.op.file_driver and
4479
        not self.op.file_driver in constants.FILE_DRIVER):
4480
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
4481
                                 self.op.file_driver)
4482

    
4483
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
4484
      raise errors.OpPrereqError("File storage directory path not absolute")
4485

    
4486
    ### Node/iallocator related checks
4487
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
4488
      raise errors.OpPrereqError("One and only one of iallocator and primary"
4489
                                 " node must be given")
4490

    
4491
    if self.op.iallocator:
4492
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4493
    else:
4494
      self.op.pnode = self._ExpandNode(self.op.pnode)
4495
      nodelist = [self.op.pnode]
4496
      if self.op.snode is not None:
4497
        self.op.snode = self._ExpandNode(self.op.snode)
4498
        nodelist.append(self.op.snode)
4499
      self.needed_locks[locking.LEVEL_NODE] = nodelist
4500

    
4501
    # in case of import lock the source node too
4502
    if self.op.mode == constants.INSTANCE_IMPORT:
4503
      src_node = getattr(self.op, "src_node", None)
4504
      src_path = getattr(self.op, "src_path", None)
4505

    
4506
      if src_path is None:
4507
        self.op.src_path = src_path = self.op.instance_name
4508

    
4509
      if src_node is None:
4510
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
4511
        self.op.src_node = None
4512
        if os.path.isabs(src_path):
4513
          raise errors.OpPrereqError("Importing an instance from an absolute"
4514
                                     " path requires a source node option.")
4515
      else:
4516
        self.op.src_node = src_node = self._ExpandNode(src_node)
4517
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
4518
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
4519
        if not os.path.isabs(src_path):
4520
          self.op.src_path = src_path = \
4521
            os.path.join(constants.EXPORT_DIR, src_path)
4522

    
4523
    else: # INSTANCE_CREATE
4524
      if getattr(self.op, "os_type", None) is None:
4525
        raise errors.OpPrereqError("No guest OS specified")
4526

    
4527
  def _RunAllocator(self):
4528
    """Run the allocator based on input opcode.
4529

4530
    """
4531
    nics = [n.ToDict() for n in self.nics]
4532
    ial = IAllocator(self,
4533
                     mode=constants.IALLOCATOR_MODE_ALLOC,
4534
                     name=self.op.instance_name,
4535
                     disk_template=self.op.disk_template,
4536
                     tags=[],
4537
                     os=self.op.os_type,
4538
                     vcpus=self.be_full[constants.BE_VCPUS],
4539
                     mem_size=self.be_full[constants.BE_MEMORY],
4540
                     disks=self.disks,
4541
                     nics=nics,
4542
                     hypervisor=self.op.hypervisor,
4543
                     )
4544

    
4545
    ial.Run(self.op.iallocator)
4546

    
4547
    if not ial.success:
4548
      raise errors.OpPrereqError("Can't compute nodes using"
4549
                                 " iallocator '%s': %s" % (self.op.iallocator,
4550
                                                           ial.info))
4551
    if len(ial.nodes) != ial.required_nodes:
4552
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
4553
                                 " of nodes (%s), required %s" %
4554
                                 (self.op.iallocator, len(ial.nodes),
4555
                                  ial.required_nodes))
4556
    self.op.pnode = ial.nodes[0]
4557
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
4558
                 self.op.instance_name, self.op.iallocator,
4559
                 ", ".join(ial.nodes))
4560
    if ial.required_nodes == 2:
4561
      self.op.snode = ial.nodes[1]
4562

    
4563
  def BuildHooksEnv(self):
4564
    """Build hooks env.
4565

4566
    This runs on master, primary and secondary nodes of the instance.
4567

4568
    """
4569
    env = {
4570
      "ADD_MODE": self.op.mode,
4571
      }
4572
    if self.op.mode == constants.INSTANCE_IMPORT:
4573
      env["SRC_NODE"] = self.op.src_node
4574
      env["SRC_PATH"] = self.op.src_path
4575
      env["SRC_IMAGES"] = self.src_images
4576

    
4577
    env.update(_BuildInstanceHookEnv(
4578
      name=self.op.instance_name,
4579
      primary_node=self.op.pnode,
4580
      secondary_nodes=self.secondaries,
4581
      status=self.op.start,
4582
      os_type=self.op.os_type,
4583
      memory=self.be_full[constants.BE_MEMORY],
4584
      vcpus=self.be_full[constants.BE_VCPUS],
4585
      nics=[(n.ip, n.bridge, n.mac) for n in self.nics],
4586
      disk_template=self.op.disk_template,
4587
      disks=[(d["size"], d["mode"]) for d in self.disks],
4588
    ))
4589

    
4590
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
4591
          self.secondaries)
4592
    return env, nl, nl
4593

    
4594

    
4595
  def CheckPrereq(self):
4596
    """Check prerequisites.
4597

4598
    """
4599
    if (not self.cfg.GetVGName() and
4600
        self.op.disk_template not in constants.DTS_NOT_LVM):
4601
      raise errors.OpPrereqError("Cluster does not support lvm-based"
4602
                                 " instances")
4603

    
4604
    if self.op.mode == constants.INSTANCE_IMPORT:
4605
      src_node = self.op.src_node
4606
      src_path = self.op.src_path
4607

    
4608
      if src_node is None:
4609
        exp_list = self.rpc.call_export_list(
4610
          self.acquired_locks[locking.LEVEL_NODE])
4611
        found = False
4612
        for node in exp_list:
4613
          if not exp_list[node].failed and src_path in exp_list[node].data:
4614
            found = True
4615
            self.op.src_node = src_node = node
4616
            self.op.src_path = src_path = os.path.join(constants.EXPORT_DIR,
4617
                                                       src_path)
4618
            break
4619
        if not found:
4620
          raise errors.OpPrereqError("No export found for relative path %s" %
4621
                                      src_path)
4622

    
4623
      _CheckNodeOnline(self, src_node)
4624
      result = self.rpc.call_export_info(src_node, src_path)
4625
      result.Raise()
4626
      if not result.data:
4627
        raise errors.OpPrereqError("No export found in dir %s" % src_path)
4628

    
4629
      export_info = result.data
4630
      if not export_info.has_section(constants.INISECT_EXP):
4631
        raise errors.ProgrammerError("Corrupted export config")
4632

    
4633
      ei_version = export_info.get(constants.INISECT_EXP, 'version')
4634
      if (int(ei_version) != constants.EXPORT_VERSION):
4635
        raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
4636
                                   (ei_version, constants.EXPORT_VERSION))
4637

    
4638
      # Check that the new instance doesn't have less disks than the export
4639
      instance_disks = len(self.disks)
4640
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
4641
      if instance_disks < export_disks:
4642
        raise errors.OpPrereqError("Not enough disks to import."
4643
                                   " (instance: %d, export: %d)" %
4644
                                   (instance_disks, export_disks))
4645

    
4646
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
4647
      disk_images = []
4648
      for idx in range(export_disks):
4649
        option = 'disk%d_dump' % idx
4650
        if export_info.has_option(constants.INISECT_INS, option):
4651
          # FIXME: are the old os-es, disk sizes, etc. useful?
4652
          export_name = export_info.get(constants.INISECT_INS, option)
4653
          image = os.path.join(src_path, export_name)
4654
          disk_images.append(image)
4655
        else:
4656
          disk_images.append(False)
4657

    
4658
      self.src_images = disk_images
4659

    
4660
      old_name = export_info.get(constants.INISECT_INS, 'name')
4661
      # FIXME: int() here could throw a ValueError on broken exports
4662
      exp_nic_count = int(export_info.get(constants.INISECT_INS, 'nic_count'))
4663
      if self.op.instance_name == old_name:
4664
        for idx, nic in enumerate(self.nics):
4665
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
4666
            nic_mac_ini = 'nic%d_mac' % idx
4667
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
4668

    
4669
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
4670
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
4671
    if self.op.start and not self.op.ip_check:
4672
      raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
4673
                                 " adding an instance in start mode")
4674

    
4675
    if self.op.ip_check:
4676
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
4677
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4678
                                   (self.check_ip, self.op.instance_name))
4679

    
4680
    #### mac address generation
4681
    # By generating here the mac address both the allocator and the hooks get
4682
    # the real final mac address rather than the 'auto' or 'generate' value.
4683
    # There is a race condition between the generation and the instance object
4684
    # creation, which means that we know the mac is valid now, but we're not
4685
    # sure it will be when we actually add the instance. If things go bad
4686
    # adding the instance will abort because of a duplicate mac, and the
4687
    # creation job will fail.
4688
    for nic in self.nics:
4689
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
4690
        nic.mac = self.cfg.GenerateMAC()
4691

    
4692
    #### allocator run
4693

    
4694
    if self.op.iallocator is not None:
4695
      self._RunAllocator()
4696

    
4697
    #### node related checks
4698

    
4699
    # check primary node
4700
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
4701
    assert self.pnode is not None, \
4702
      "Cannot retrieve locked node %s" % self.op.pnode
4703
    if pnode.offline:
4704
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
4705
                                 pnode.name)
4706
    if pnode.drained:
4707
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
4708
                                 pnode.name)
4709

    
4710
    self.secondaries = []
4711

    
4712
    # mirror node verification
4713
    if self.op.disk_template in constants.DTS_NET_MIRROR:
4714
      if self.op.snode is None:
4715
        raise errors.OpPrereqError("The networked disk templates need"
4716
                                   " a mirror node")
4717
      if self.op.snode == pnode.name:
4718
        raise errors.OpPrereqError("The secondary node cannot be"
4719
                                   " the primary node.")
4720
      _CheckNodeOnline(self, self.op.snode)
4721
      _CheckNodeNotDrained(self, self.op.snode)
4722
      self.secondaries.append(self.op.snode)
4723

    
4724
    nodenames = [pnode.name] + self.secondaries
4725

    
4726
    req_size = _ComputeDiskSize(self.op.disk_template,
4727
                                self.disks)
4728

    
4729
    # Check lv size requirements
4730
    if req_size is not None:
4731
      nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
4732
                                         self.op.hypervisor)
4733
      for node in nodenames:
4734
        info = nodeinfo[node]
4735
        info.Raise()
4736
        info = info.data
4737
        if not info:
4738
          raise errors.OpPrereqError("Cannot get current information"
4739
                                     " from node '%s'" % node)
4740
        vg_free = info.get('vg_free', None)
4741
        if not isinstance(vg_free, int):
4742
          raise errors.OpPrereqError("Can't compute free disk space on"
4743
                                     " node %s" % node)
4744
        if req_size > info['vg_free']:
4745
          raise errors.OpPrereqError("Not enough disk space on target node %s."
4746
                                     " %d MB available, %d MB required" %
4747
                                     (node, info['vg_free'], req_size))
4748

    
4749
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
4750

    
4751
    # os verification
4752
    result = self.rpc.call_os_get(pnode.name, self.op.os_type)
4753
    result.Raise()
4754
    if not isinstance(result.data, objects.OS):
4755
      raise errors.OpPrereqError("OS '%s' not in supported os list for"
4756
                                 " primary node"  % self.op.os_type)
4757

    
4758
    # bridge check on primary node
4759
    bridges = [n.bridge for n in self.nics]
4760
    result = self.rpc.call_bridges_exist(self.pnode.name, bridges)
4761
    result.Raise()
4762
    if not result.data:
4763
      raise errors.OpPrereqError("One of the target bridges '%s' does not"
4764
                                 " exist on destination node '%s'" %
4765
                                 (",".join(bridges), pnode.name))
4766

    
4767
    # memory check on primary node
4768
    if self.op.start:
4769
      _CheckNodeFreeMemory(self, self.pnode.name,
4770
                           "creating instance %s" % self.op.instance_name,
4771
                           self.be_full[constants.BE_MEMORY],
4772
                           self.op.hypervisor)
4773

    
4774
  def Exec(self, feedback_fn):
4775
    """Create and add the instance to the cluster.
4776

4777
    """
4778
    instance = self.op.instance_name
4779
    pnode_name = self.pnode.name
4780

    
4781
    ht_kind = self.op.hypervisor
4782
    if ht_kind in constants.HTS_REQ_PORT:
4783
      network_port = self.cfg.AllocatePort()
4784
    else:
4785
      network_port = None
4786

    
4787
    ##if self.op.vnc_bind_address is None:
4788
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
4789

    
4790
    # this is needed because os.path.join does not accept None arguments
4791
    if self.op.file_storage_dir is None:
4792
      string_file_storage_dir = ""
4793
    else:
4794
      string_file_storage_dir = self.op.file_storage_dir
4795

    
4796
    # build the full file storage dir path
4797
    file_storage_dir = os.path.normpath(os.path.join(
4798
                                        self.cfg.GetFileStorageDir(),
4799
                                        string_file_storage_dir, instance))
4800

    
4801

    
4802
    disks = _GenerateDiskTemplate(self,
4803
                                  self.op.disk_template,
4804
                                  instance, pnode_name,
4805
                                  self.secondaries,
4806
                                  self.disks,
4807
                                  file_storage_dir,
4808
                                  self.op.file_driver,
4809
                                  0)
4810

    
4811
    iobj = objects.Instance(name=instance, os=self.op.os_type,
4812
                            primary_node=pnode_name,
4813
                            nics=self.nics, disks=disks,
4814
                            disk_template=self.op.disk_template,
4815
                            admin_up=False,
4816
                            network_port=network_port,
4817
                            beparams=self.op.beparams,
4818
                            hvparams=self.op.hvparams,
4819
                            hypervisor=self.op.hypervisor,
4820
                            )
4821

    
4822
    feedback_fn("* creating instance disks...")
4823
    try:
4824
      _CreateDisks(self, iobj)
4825
    except errors.OpExecError:
4826
      self.LogWarning("Device creation failed, reverting...")
4827
      try:
4828
        _RemoveDisks(self, iobj)
4829
      finally:
4830
        self.cfg.ReleaseDRBDMinors(instance)
4831
        raise
4832

    
4833
    feedback_fn("adding instance %s to cluster config" % instance)
4834

    
4835
    self.cfg.AddInstance(iobj)
4836
    # Declare that we don't want to remove the instance lock anymore, as we've
4837
    # added the instance to the config
4838
    del self.remove_locks[locking.LEVEL_INSTANCE]
4839
    # Unlock all the nodes
4840
    if self.op.mode == constants.INSTANCE_IMPORT:
4841
      nodes_keep = [self.op.src_node]
4842
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
4843
                       if node != self.op.src_node]
4844
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
4845
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
4846
    else:
4847
      self.context.glm.release(locking.LEVEL_NODE)
4848
      del self.acquired_locks[locking.LEVEL_NODE]
4849

    
4850
    if self.op.wait_for_sync:
4851
      disk_abort = not _WaitForSync(self, iobj)
4852
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
4853
      # make sure the disks are not degraded (still sync-ing is ok)
4854
      time.sleep(15)
4855
      feedback_fn("* checking mirrors status")
4856
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
4857
    else:
4858
      disk_abort = False
4859

    
4860
    if disk_abort:
4861
      _RemoveDisks(self, iobj)
4862
      self.cfg.RemoveInstance(iobj.name)
4863
      # Make sure the instance lock gets removed
4864
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
4865
      raise errors.OpExecError("There are some degraded disks for"
4866
                               " this instance")
4867

    
4868
    feedback_fn("creating os for instance %s on node %s" %
4869
                (instance, pnode_name))
4870

    
4871
    if iobj.disk_template != constants.DT_DISKLESS:
4872
      if self.op.mode == constants.INSTANCE_CREATE:
4873
        feedback_fn("* running the instance OS create scripts...")
4874
        result = self.rpc.call_instance_os_add(pnode_name, iobj, False)
4875
        msg = result.RemoteFailMsg()
4876
        if msg:
4877
          raise errors.OpExecError("Could not add os for instance %s"
4878
                                   " on node %s: %s" %
4879
                                   (instance, pnode_name, msg))
4880

    
4881
      elif self.op.mode == constants.INSTANCE_IMPORT:
4882
        feedback_fn("* running the instance OS import scripts...")
4883
        src_node = self.op.src_node
4884
        src_images = self.src_images
4885
        cluster_name = self.cfg.GetClusterName()
4886
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
4887
                                                         src_node, src_images,
4888
                                                         cluster_name)
4889
        import_result.Raise()
4890
        for idx, result in enumerate(import_result.data):
4891
          if not result:
4892
            self.LogWarning("Could not import the image %s for instance"
4893
                            " %s, disk %d, on node %s" %
4894
                            (src_images[idx], instance, idx, pnode_name))
4895
      else:
4896
        # also checked in the prereq part
4897
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
4898
                                     % self.op.mode)
4899

    
4900
    if self.op.start:
4901
      iobj.admin_up = True
4902
      self.cfg.Update(iobj)
4903
      logging.info("Starting instance %s on node %s", instance, pnode_name)
4904
      feedback_fn("* starting instance...")
4905
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
4906
      msg = result.RemoteFailMsg()
4907
      if msg:
4908
        raise errors.OpExecError("Could not start instance: %s" % msg)
4909

    
4910

    
4911
class LUConnectConsole(NoHooksLU):
4912
  """Connect to an instance's console.
4913

4914
  This is somewhat special in that it returns the command line that
4915
  you need to run on the master node in order to connect to the
4916
  console.
4917

4918
  """
4919
  _OP_REQP = ["instance_name"]
4920
  REQ_BGL = False
4921

    
4922
  def ExpandNames(self):
4923
    self._ExpandAndLockInstance()
4924

    
4925
  def CheckPrereq(self):
4926
    """Check prerequisites.
4927

4928
    This checks that the instance is in the cluster.
4929

4930
    """
4931
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4932
    assert self.instance is not None, \
4933
      "Cannot retrieve locked instance %s" % self.op.instance_name
4934
    _CheckNodeOnline(self, self.instance.primary_node)
4935

    
4936
  def Exec(self, feedback_fn):
4937
    """Connect to the console of an instance
4938

4939
    """
4940
    instance = self.instance
4941
    node = instance.primary_node
4942

    
4943
    node_insts = self.rpc.call_instance_list([node],
4944
                                             [instance.hypervisor])[node]
4945
    node_insts.Raise()
4946

    
4947
    if instance.name not in node_insts.data:
4948
      raise errors.OpExecError("Instance %s is not running." % instance.name)
4949

    
4950
    logging.debug("Connecting to console of %s on %s", instance.name, node)
4951

    
4952
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
4953
    cluster = self.cfg.GetClusterInfo()
4954
    # beparams and hvparams are passed separately, to avoid editing the
4955
    # instance and then saving the defaults in the instance itself.
4956
    hvparams = cluster.FillHV(instance)
4957
    beparams = cluster.FillBE(instance)
4958
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
4959

    
4960
    # build ssh cmdline
4961
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
4962

    
4963

    
4964
class LUReplaceDisks(LogicalUnit):
4965
  """Replace the disks of an instance.
4966

4967
  """
4968
  HPATH = "mirrors-replace"
4969
  HTYPE = constants.HTYPE_INSTANCE
4970
  _OP_REQP = ["instance_name", "mode", "disks"]
4971
  REQ_BGL = False
4972

    
4973
  def CheckArguments(self):
4974
    if not hasattr(self.op, "remote_node"):
4975
      self.op.remote_node = None
4976
    if not hasattr(self.op, "iallocator"):
4977
      self.op.iallocator = None
4978

    
4979
    # check for valid parameter combination
4980
    cnt = [self.op.remote_node, self.op.iallocator].count(None)
4981
    if self.op.mode == constants.REPLACE_DISK_CHG:
4982
      if cnt == 2:
4983
        raise errors.OpPrereqError("When changing the secondary either an"
4984
                                   " iallocator script must be used or the"
4985
                                   " new node given")
4986
      elif cnt == 0:
4987
        raise errors.OpPrereqError("Give either the iallocator or the new"
4988
                                   " secondary, not both")
4989
    else: # not replacing the secondary
4990
      if cnt != 2:
4991
        raise errors.OpPrereqError("The iallocator and new node options can"
4992
                                   " be used only when changing the"
4993
                                   " secondary node")
4994

    
4995
  def ExpandNames(self):
4996
    self._ExpandAndLockInstance()
4997

    
4998
    if self.op.iallocator is not None:
4999
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
5000
    elif self.op.remote_node is not None:
5001
      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
5002
      if remote_node is None:
5003
        raise errors.OpPrereqError("Node '%s' not known" %
5004
                                   self.op.remote_node)
5005
      self.op.remote_node = remote_node
5006
      # Warning: do not remove the locking of the new secondary here
5007
      # unless DRBD8.AddChildren is changed to work in parallel;
5008
      # currently it doesn't since parallel invocations of
5009
      # FindUnusedMinor will conflict
5010
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
5011
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5012
    else:
5013
      self.needed_locks[locking.LEVEL_NODE] = []
5014
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5015

    
5016
  def DeclareLocks(self, level):
5017
    # If we're not already locking all nodes in the set we have to declare the
5018
    # instance's primary/secondary nodes.
5019
    if (level == locking.LEVEL_NODE and
5020
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
5021
      self._LockInstancesNodes()
5022

    
5023
  def _RunAllocator(self):
5024
    """Compute a new secondary node using an IAllocator.
5025

5026
    """
5027
    ial = IAllocator(self,
5028
                     mode=constants.IALLOCATOR_MODE_RELOC,
5029
                     name=self.op.instance_name,
5030
                     relocate_from=[self.sec_node])
5031

    
5032
    ial.Run(self.op.iallocator)
5033

    
5034
    if not ial.success:
5035
      raise errors.OpPrereqError("Can't compute nodes using"
5036
                                 " iallocator '%s': %s" % (self.op.iallocator,
5037
                                                           ial.info))
5038
    if len(ial.nodes) != ial.required_nodes:
5039
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
5040
                                 " of nodes (%s), required %s" %
5041
                                 (len(ial.nodes), ial.required_nodes))
5042
    self.op.remote_node = ial.nodes[0]
5043
    self.LogInfo("Selected new secondary for the instance: %s",
5044
                 self.op.remote_node)
5045

    
5046
  def BuildHooksEnv(self):
5047
    """Build hooks env.
5048

5049
    This runs on the master, the primary and all the secondaries.
5050

5051
    """
5052
    env = {
5053
      "MODE": self.op.mode,
5054
      "NEW_SECONDARY": self.op.remote_node,
5055
      "OLD_SECONDARY": self.instance.secondary_nodes[0],
5056
      }
5057
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5058
    nl = [
5059
      self.cfg.GetMasterNode(),
5060
      self.instance.primary_node,
5061
      ]
5062
    if self.op.remote_node is not None:
5063
      nl.append(self.op.remote_node)
5064
    return env, nl, nl
5065

    
5066
  def CheckPrereq(self):
5067
    """Check prerequisites.
5068

5069
    This checks that the instance is in the cluster.
5070

5071
    """
5072
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5073
    assert instance is not None, \
5074
      "Cannot retrieve locked instance %s" % self.op.instance_name
5075
    self.instance = instance
5076

    
5077
    if instance.disk_template != constants.DT_DRBD8:
5078
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
5079
                                 " instances")
5080

    
5081
    if len(instance.secondary_nodes) != 1:
5082
      raise errors.OpPrereqError("The instance has a strange layout,"
5083
                                 " expected one secondary but found %d" %
5084
                                 len(instance.secondary_nodes))
5085

    
5086
    self.sec_node = instance.secondary_nodes[0]
5087

    
5088
    if self.op.iallocator is not None:
5089
      self._RunAllocator()
5090

    
5091
    remote_node = self.op.remote_node
5092
    if remote_node is not None:
5093
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
5094
      assert self.remote_node_info is not None, \
5095
        "Cannot retrieve locked node %s" % remote_node
5096
    else:
5097
      self.remote_node_info = None
5098
    if remote_node == instance.primary_node:
5099
      raise errors.OpPrereqError("The specified node is the primary node of"
5100
                                 " the instance.")
5101
    elif remote_node == self.sec_node:
5102
      raise errors.OpPrereqError("The specified node is already the"
5103
                                 " secondary node of the instance.")
5104

    
5105
    if self.op.mode == constants.REPLACE_DISK_PRI:
5106
      n1 = self.tgt_node = instance.primary_node
5107
      n2 = self.oth_node = self.sec_node
5108
    elif self.op.mode == constants.REPLACE_DISK_SEC:
5109
      n1 = self.tgt_node = self.sec_node
5110
      n2 = self.oth_node = instance.primary_node
5111
    elif self.op.mode == constants.REPLACE_DISK_CHG:
5112
      n1 = self.new_node = remote_node
5113
      n2 = self.oth_node = instance.primary_node
5114
      self.tgt_node = self.sec_node
5115
      _CheckNodeNotDrained(self, remote_node)
5116
    else:
5117
      raise errors.ProgrammerError("Unhandled disk replace mode")
5118

    
5119
    _CheckNodeOnline(self, n1)
5120
    _CheckNodeOnline(self, n2)
5121

    
5122
    if not self.op.disks:
5123
      self.op.disks = range(len(instance.disks))
5124

    
5125
    for disk_idx in self.op.disks:
5126
      instance.FindDisk(disk_idx)
5127

    
5128
  def _ExecD8DiskOnly(self, feedback_fn):
5129
    """Replace a disk on the primary or secondary for dbrd8.
5130

5131
    The algorithm for replace is quite complicated:
5132

5133
      1. for each disk to be replaced:
5134

5135
        1. create new LVs on the target node with unique names
5136
        1. detach old LVs from the drbd device
5137
        1. rename old LVs to name_replaced.<time_t>
5138
        1. rename new LVs to old LVs
5139
        1. attach the new LVs (with the old names now) to the drbd device
5140

5141
      1. wait for sync across all devices
5142

5143
      1. for each modified disk:
5144

5145
        1. remove old LVs (which have the name name_replaces.<time_t>)
5146

5147
    Failures are not very well handled.
5148

5149
    """
5150
    steps_total = 6
5151
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5152
    instance = self.instance
5153
    iv_names = {}
5154
    vgname = self.cfg.GetVGName()
5155
    # start of work
5156
    cfg = self.cfg
5157
    tgt_node = self.tgt_node
5158
    oth_node = self.oth_node
5159

    
5160
    # Step: check device activation
5161
    self.proc.LogStep(1, steps_total, "check device existence")
5162
    info("checking volume groups")
5163
    my_vg = cfg.GetVGName()
5164
    results = self.rpc.call_vg_list([oth_node, tgt_node])
5165
    if not results:
5166
      raise errors.OpExecError("Can't list volume groups on the nodes")
5167
    for node in oth_node, tgt_node:
5168
      res = results[node]
5169
      if res.failed or not res.data or my_vg not in res.data:
5170
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5171
                                 (my_vg, node))
5172
    for idx, dev in enumerate(instance.disks):
5173
      if idx not in self.op.disks:
5174
        continue
5175
      for node in tgt_node, oth_node:
5176
        info("checking disk/%d on %s" % (idx, node))
5177
        cfg.SetDiskID(dev, node)
5178
        result = self.rpc.call_blockdev_find(node, dev)
5179
        msg = result.RemoteFailMsg()
5180
        if not msg and not result.payload:
5181
          msg = "disk not found"
5182
        if msg:
5183
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5184
                                   (idx, node, msg))
5185

    
5186
    # Step: check other node consistency
5187
    self.proc.LogStep(2, steps_total, "check peer consistency")
5188
    for idx, dev in enumerate(instance.disks):
5189
      if idx not in self.op.disks:
5190
        continue
5191
      info("checking disk/%d consistency on %s" % (idx, oth_node))
5192
      if not _CheckDiskConsistency(self, dev, oth_node,
5193
                                   oth_node==instance.primary_node):
5194
        raise errors.OpExecError("Peer node (%s) has degraded storage, unsafe"
5195
                                 " to replace disks on this node (%s)" %
5196
                                 (oth_node, tgt_node))
5197

    
5198
    # Step: create new storage
5199
    self.proc.LogStep(3, steps_total, "allocate new storage")
5200
    for idx, dev in enumerate(instance.disks):
5201
      if idx not in self.op.disks:
5202
        continue
5203
      size = dev.size
5204
      cfg.SetDiskID(dev, tgt_node)
5205
      lv_names = [".disk%d_%s" % (idx, suf)
5206
                  for suf in ["data", "meta"]]
5207
      names = _GenerateUniqueNames(self, lv_names)
5208
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5209
                             logical_id=(vgname, names[0]))
5210
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5211
                             logical_id=(vgname, names[1]))
5212
      new_lvs = [lv_data, lv_meta]
5213
      old_lvs = dev.children
5214
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
5215
      info("creating new local storage on %s for %s" %
5216
           (tgt_node, dev.iv_name))
5217
      # we pass force_create=True to force the LVM creation
5218
      for new_lv in new_lvs:
5219
        _CreateBlockDev(self, tgt_node, instance, new_lv, True,
5220
                        _GetInstanceInfoText(instance), False)
5221

    
5222
    # Step: for each lv, detach+rename*2+attach
5223
    self.proc.LogStep(4, steps_total, "change drbd configuration")
5224
    for dev, old_lvs, new_lvs in iv_names.itervalues():
5225
      info("detaching %s drbd from local storage" % dev.iv_name)
5226
      result = self.rpc.call_blockdev_removechildren(tgt_node, dev, old_lvs)
5227
      result.Raise()
5228
      if not result.data:
5229
        raise errors.OpExecError("Can't detach drbd from local storage on node"
5230
                                 " %s for device %s" % (tgt_node, dev.iv_name))
5231
      #dev.children = []
5232
      #cfg.Update(instance)
5233

    
5234
      # ok, we created the new LVs, so now we know we have the needed
5235
      # storage; as such, we proceed on the target node to rename
5236
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
5237
      # using the assumption that logical_id == physical_id (which in
5238
      # turn is the unique_id on that node)
5239

    
5240
      # FIXME(iustin): use a better name for the replaced LVs
5241
      temp_suffix = int(time.time())
5242
      ren_fn = lambda d, suff: (d.physical_id[0],
5243
                                d.physical_id[1] + "_replaced-%s" % suff)
5244
      # build the rename list based on what LVs exist on the node
5245
      rlist = []
5246
      for to_ren in old_lvs:
5247
        result = self.rpc.call_blockdev_find(tgt_node, to_ren)
5248
        if not result.RemoteFailMsg() and result.payload:
5249
          # device exists
5250
          rlist.append((to_ren, ren_fn(to_ren, temp_suffix)))
5251

    
5252
      info("renaming the old LVs on the target node")
5253
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5254
      msg = result.RemoteFailMsg()
5255
      if msg:
5256
        raise errors.OpExecError("Can't rename old LVs on node %s: %s" %
5257
                                 (tgt_node, msg))
5258
      # now we rename the new LVs to the old LVs
5259
      info("renaming the new LVs on the target node")
5260
      rlist = [(new, old.physical_id) for old, new in zip(old_lvs, new_lvs)]
5261
      result = self.rpc.call_blockdev_rename(tgt_node, rlist)
5262
      msg = result.RemoteFailMsg()
5263
      if msg:
5264
        raise errors.OpExecError("Can't rename new LVs on node %s: %s" %
5265
                                 (tgt_node, msg))
5266

    
5267
      for old, new in zip(old_lvs, new_lvs):
5268
        new.logical_id = old.logical_id
5269
        cfg.SetDiskID(new, tgt_node)
5270

    
5271
      for disk in old_lvs:
5272
        disk.logical_id = ren_fn(disk, temp_suffix)
5273
        cfg.SetDiskID(disk, tgt_node)
5274

    
5275
      # now that the new lvs have the old name, we can add them to the device
5276
      info("adding new mirror component on %s" % tgt_node)
5277
      result = self.rpc.call_blockdev_addchildren(tgt_node, dev, new_lvs)
5278
      msg = result.RemoteFailMsg()
5279
      if msg:
5280
        for new_lv in new_lvs:
5281
          msg = self.rpc.call_blockdev_remove(tgt_node, new_lv).RemoteFailMsg()
5282
          if msg:
5283
            warning("Can't rollback device %s: %s", dev, msg,
5284
                    hint="cleanup manually the unused logical volumes")
5285
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
5286

    
5287
      dev.children = new_lvs
5288
      cfg.Update(instance)
5289

    
5290
    # Step: wait for sync
5291

    
5292
    # this can fail as the old devices are degraded and _WaitForSync
5293
    # does a combined result over all disks, so we don't check its
5294
    # return value
5295
    self.proc.LogStep(5, steps_total, "sync devices")
5296
    _WaitForSync(self, instance, unlock=True)
5297

    
5298
    # so check manually all the devices
5299
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5300
      cfg.SetDiskID(dev, instance.primary_node)
5301
      result = self.rpc.call_blockdev_find(instance.primary_node, dev)
5302
      msg = result.RemoteFailMsg()
5303
      if not msg and not result.payload:
5304
        msg = "disk not found"
5305
      if msg:
5306
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
5307
                                 (name, msg))
5308
      if result.payload[5]:
5309
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
5310

    
5311
    # Step: remove old storage
5312
    self.proc.LogStep(6, steps_total, "removing old storage")
5313
    for name, (dev, old_lvs, new_lvs) in iv_names.iteritems():
5314
      info("remove logical volumes for %s" % name)
5315
      for lv in old_lvs:
5316
        cfg.SetDiskID(lv, tgt_node)
5317
        msg = self.rpc.call_blockdev_remove(tgt_node, lv).RemoteFailMsg()
5318
        if msg:
5319
          warning("Can't remove old LV: %s" % msg,
5320
                  hint="manually remove unused LVs")
5321
          continue
5322

    
5323
  def _ExecD8Secondary(self, feedback_fn):
5324
    """Replace the secondary node for drbd8.
5325

5326
    The algorithm for replace is quite complicated:
5327
      - for all disks of the instance:
5328
        - create new LVs on the new node with same names
5329
        - shutdown the drbd device on the old secondary
5330
        - disconnect the drbd network on the primary
5331
        - create the drbd device on the new secondary
5332
        - network attach the drbd on the primary, using an artifice:
5333
          the drbd code for Attach() will connect to the network if it
5334
          finds a device which is connected to the good local disks but
5335
          not network enabled
5336
      - wait for sync across all devices
5337
      - remove all disks from the old secondary
5338

5339
    Failures are not very well handled.
5340

5341
    """
5342
    steps_total = 6
5343
    warning, info = (self.proc.LogWarning, self.proc.LogInfo)
5344
    instance = self.instance
5345
    iv_names = {}
5346
    # start of work
5347
    cfg = self.cfg
5348
    old_node = self.tgt_node
5349
    new_node = self.new_node
5350
    pri_node = instance.primary_node
5351
    nodes_ip = {
5352
      old_node: self.cfg.GetNodeInfo(old_node).secondary_ip,
5353
      new_node: self.cfg.GetNodeInfo(new_node).secondary_ip,
5354
      pri_node: self.cfg.GetNodeInfo(pri_node).secondary_ip,
5355
      }
5356

    
5357
    # Step: check device activation
5358
    self.proc.LogStep(1, steps_total, "check device existence")
5359
    info("checking volume groups")
5360
    my_vg = cfg.GetVGName()
5361
    results = self.rpc.call_vg_list([pri_node, new_node])
5362
    for node in pri_node, new_node:
5363
      res = results[node]
5364
      if res.failed or not res.data or my_vg not in res.data:
5365
        raise errors.OpExecError("Volume group '%s' not found on %s" %
5366
                                 (my_vg, node))
5367
    for idx, dev in enumerate(instance.disks):
5368
      if idx not in self.op.disks:
5369
        continue
5370
      info("checking disk/%d on %s" % (idx, pri_node))
5371
      cfg.SetDiskID(dev, pri_node)
5372
      result = self.rpc.call_blockdev_find(pri_node, dev)
5373
      msg = result.RemoteFailMsg()
5374
      if not msg and not result.payload:
5375
        msg = "disk not found"
5376
      if msg:
5377
        raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
5378
                                 (idx, pri_node, msg))
5379

    
5380
    # Step: check other node consistency
5381
    self.proc.LogStep(2, steps_total, "check peer consistency")
5382
    for idx, dev in enumerate(instance.disks):
5383
      if idx not in self.op.disks:
5384
        continue
5385
      info("checking disk/%d consistency on %s" % (idx, pri_node))
5386
      if not _CheckDiskConsistency(self, dev, pri_node, True, ldisk=True):
5387
        raise errors.OpExecError("Primary node (%s) has degraded storage,"
5388
                                 " unsafe to replace the secondary" %
5389
                                 pri_node)
5390

    
5391
    # Step: create new storage
5392
    self.proc.LogStep(3, steps_total, "allocate new storage")
5393
    for idx, dev in enumerate(instance.disks):
5394
      info("adding new local storage on %s for disk/%d" %
5395
           (new_node, idx))
5396
      # we pass force_create=True to force LVM creation
5397
      for new_lv in dev.children:
5398
        _CreateBlockDev(self, new_node, instance, new_lv, True,
5399
                        _GetInstanceInfoText(instance), False)
5400

    
5401
    # Step 4: dbrd minors and drbd setups changes
5402
    # after this, we must manually remove the drbd minors on both the
5403
    # error and the success paths
5404
    minors = cfg.AllocateDRBDMinor([new_node for dev in instance.disks],
5405
                                   instance.name)
5406
    logging.debug("Allocated minors %s" % (minors,))
5407
    self.proc.LogStep(4, steps_total, "changing drbd configuration")
5408
    for idx, (dev, new_minor) in enumerate(zip(instance.disks, minors)):
5409
      size = dev.size
5410
      info("activating a new drbd on %s for disk/%d" % (new_node, idx))
5411
      # create new devices on new_node; note that we create two IDs:
5412
      # one without port, so the drbd will be activated without
5413
      # networking information on the new node at this stage, and one
5414
      # with network, for the latter activation in step 4
5415
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
5416
      if pri_node == o_node1:
5417
        p_minor = o_minor1
5418
      else:
5419
        p_minor = o_minor2
5420

    
5421
      new_alone_id = (pri_node, new_node, None, p_minor, new_minor, o_secret)
5422
      new_net_id = (pri_node, new_node, o_port, p_minor, new_minor, o_secret)
5423

    
5424
      iv_names[idx] = (dev, dev.children, new_net_id)
5425
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
5426
                    new_net_id)
5427
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
5428
                              logical_id=new_alone_id,
5429
                              children=dev.children)
5430
      try:
5431
        _CreateSingleBlockDev(self, new_node, instance, new_drbd,
5432
                              _GetInstanceInfoText(instance), False)
5433
      except errors.GenericError:
5434
        self.cfg.ReleaseDRBDMinors(instance.name)
5435
        raise
5436

    
5437
    for idx, dev in enumerate(instance.disks):
5438
      # we have new devices, shutdown the drbd on the old secondary
5439
      info("shutting down drbd for disk/%d on old node" % idx)
5440
      cfg.SetDiskID(dev, old_node)
5441
      msg = self.rpc.call_blockdev_shutdown(old_node, dev).RemoteFailMsg()
5442
      if msg:
5443
        warning("Failed to shutdown drbd for disk/%d on old node: %s" %
5444
                (idx, msg),
5445
                hint="Please cleanup this device manually as soon as possible")
5446

    
5447
    info("detaching primary drbds from the network (=> standalone)")
5448
    result = self.rpc.call_drbd_disconnect_net([pri_node], nodes_ip,
5449
                                               instance.disks)[pri_node]
5450

    
5451
    msg = result.RemoteFailMsg()
5452
    if msg:
5453
      # detaches didn't succeed (unlikely)
5454
      self.cfg.ReleaseDRBDMinors(instance.name)
5455
      raise errors.OpExecError("Can't detach the disks from the network on"
5456
                               " old node: %s" % (msg,))
5457

    
5458
    # if we managed to detach at least one, we update all the disks of
5459
    # the instance to point to the new secondary
5460
    info("updating instance configuration")
5461
    for dev, _, new_logical_id in iv_names.itervalues():
5462
      dev.logical_id = new_logical_id
5463
      cfg.SetDiskID(dev, pri_node)
5464
    cfg.Update(instance)
5465

    
5466
    # and now perform the drbd attach
5467
    info("attaching primary drbds to new secondary (standalone => connected)")
5468
    result = self.rpc.call_drbd_attach_net([pri_node, new_node], nodes_ip,
5469
                                           instance.disks, instance.name,
5470
                                           False)
5471
    for to_node, to_result in result.items():
5472
      msg = to_result.RemoteFailMsg()
5473
      if msg:
5474
        warning("can't attach drbd disks on node %s: %s", to_node, msg,
5475
                hint="please do a gnt-instance info to see the"
5476
                " status of disks")
5477

    
5478
    # this can fail as the old devices are degraded and _WaitForSync
5479
    # does a combined result over all disks, so we don't check its
5480
    # return value
5481
    self.proc.LogStep(5, steps_total, "sync devices")
5482
    _WaitForSync(self, instance, unlock=True)
5483

    
5484
    # so check manually all the devices
5485
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5486
      cfg.SetDiskID(dev, pri_node)
5487
      result = self.rpc.call_blockdev_find(pri_node, dev)
5488
      msg = result.RemoteFailMsg()
5489
      if not msg and not result.payload:
5490
        msg = "disk not found"
5491
      if msg:
5492
        raise errors.OpExecError("Can't find DRBD device disk/%d: %s" %
5493
                                 (idx, msg))
5494
      if result.payload[5]:
5495
        raise errors.OpExecError("DRBD device disk/%d is degraded!" % idx)
5496

    
5497
    self.proc.LogStep(6, steps_total, "removing old storage")
5498
    for idx, (dev, old_lvs, _) in iv_names.iteritems():
5499
      info("remove logical volumes for disk/%d" % idx)
5500
      for lv in old_lvs:
5501
        cfg.SetDiskID(lv, old_node)
5502
        msg = self.rpc.call_blockdev_remove(old_node, lv).RemoteFailMsg()
5503
        if msg:
5504
          warning("Can't remove LV on old secondary: %s", msg,
5505
                  hint="Cleanup stale volumes by hand")
5506

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

5510
    This dispatches the disk replacement to the appropriate handler.
5511

5512
    """
5513
    instance = self.instance
5514

    
5515
    # Activate the instance disks if we're replacing them on a down instance
5516
    if not instance.admin_up:
5517
      _StartInstanceDisks(self, instance, True)
5518

    
5519
    if self.op.mode == constants.REPLACE_DISK_CHG:
5520
      fn = self._ExecD8Secondary
5521
    else:
5522
      fn = self._ExecD8DiskOnly
5523

    
5524
    ret = fn(feedback_fn)
5525

    
5526
    # Deactivate the instance disks if we're replacing them on a down instance
5527
    if not instance.admin_up:
5528
      _SafeShutdownInstanceDisks(self, instance)
5529

    
5530
    return ret
5531

    
5532

    
5533
class LUGrowDisk(LogicalUnit):
5534
  """Grow a disk of an instance.
5535

5536
  """
5537
  HPATH = "disk-grow"
5538
  HTYPE = constants.HTYPE_INSTANCE
5539
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
5540
  REQ_BGL = False
5541

    
5542
  def ExpandNames(self):
5543
    self._ExpandAndLockInstance()
5544
    self.needed_locks[locking.LEVEL_NODE] = []
5545
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5546

    
5547
  def DeclareLocks(self, level):
5548
    if level == locking.LEVEL_NODE:
5549
      self._LockInstancesNodes()
5550

    
5551
  def BuildHooksEnv(self):
5552
    """Build hooks env.
5553

5554
    This runs on the master, the primary and all the secondaries.
5555

5556
    """
5557
    env = {
5558
      "DISK": self.op.disk,
5559
      "AMOUNT": self.op.amount,
5560
      }
5561
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5562
    nl = [
5563
      self.cfg.GetMasterNode(),
5564
      self.instance.primary_node,
5565
      ]
5566
    return env, nl, nl
5567

    
5568
  def CheckPrereq(self):
5569
    """Check prerequisites.
5570

5571
    This checks that the instance is in the cluster.
5572

5573
    """
5574
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5575
    assert instance is not None, \
5576
      "Cannot retrieve locked instance %s" % self.op.instance_name
5577
    nodenames = list(instance.all_nodes)
5578
    for node in nodenames:
5579
      _CheckNodeOnline(self, node)
5580

    
5581

    
5582
    self.instance = instance
5583

    
5584
    if instance.disk_template not in (constants.DT_PLAIN, constants.DT_DRBD8):
5585
      raise errors.OpPrereqError("Instance's disk layout does not support"
5586
                                 " growing.")
5587

    
5588
    self.disk = instance.FindDisk(self.op.disk)
5589

    
5590
    nodeinfo = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
5591
                                       instance.hypervisor)
5592
    for node in nodenames:
5593
      info = nodeinfo[node]
5594
      if info.failed or not info.data:
5595
        raise errors.OpPrereqError("Cannot get current information"
5596
                                   " from node '%s'" % node)
5597
      vg_free = info.data.get('vg_free', None)
5598
      if not isinstance(vg_free, int):
5599
        raise errors.OpPrereqError("Can't compute free disk space on"
5600
                                   " node %s" % node)
5601
      if self.op.amount > vg_free:
5602
        raise errors.OpPrereqError("Not enough disk space on target node %s:"
5603
                                   " %d MiB available, %d MiB required" %
5604
                                   (node, vg_free, self.op.amount))
5605

    
5606
  def Exec(self, feedback_fn):
5607
    """Execute disk grow.
5608

5609
    """
5610
    instance = self.instance
5611
    disk = self.disk
5612
    for node in instance.all_nodes:
5613
      self.cfg.SetDiskID(disk, node)
5614
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
5615
      msg = result.RemoteFailMsg()
5616
      if msg:
5617
        raise errors.OpExecError("Grow request failed to node %s: %s" %
5618
                                 (node, msg))
5619
    disk.RecordGrow(self.op.amount)
5620
    self.cfg.Update(instance)
5621
    if self.op.wait_for_sync:
5622
      disk_abort = not _WaitForSync(self, instance)
5623
      if disk_abort:
5624
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
5625
                             " status.\nPlease check the instance.")
5626

    
5627

    
5628
class LUQueryInstanceData(NoHooksLU):
5629
  """Query runtime instance data.
5630

5631
  """
5632
  _OP_REQP = ["instances", "static"]
5633
  REQ_BGL = False
5634

    
5635
  def ExpandNames(self):
5636
    self.needed_locks = {}
5637
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
5638

    
5639
    if not isinstance(self.op.instances, list):
5640
      raise errors.OpPrereqError("Invalid argument type 'instances'")
5641

    
5642
    if self.op.instances:
5643
      self.wanted_names = []
5644
      for name in self.op.instances:
5645
        full_name = self.cfg.ExpandInstanceName(name)
5646
        if full_name is None:
5647
          raise errors.OpPrereqError("Instance '%s' not known" % name)
5648
        self.wanted_names.append(full_name)
5649
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
5650
    else:
5651
      self.wanted_names = None
5652
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
5653

    
5654
    self.needed_locks[locking.LEVEL_NODE] = []
5655
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5656

    
5657
  def DeclareLocks(self, level):
5658
    if level == locking.LEVEL_NODE:
5659
      self._LockInstancesNodes()
5660

    
5661
  def CheckPrereq(self):
5662
    """Check prerequisites.
5663

5664
    This only checks the optional instance list against the existing names.
5665

5666
    """
5667
    if self.wanted_names is None:
5668
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
5669

    
5670
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
5671
                             in self.wanted_names]
5672
    return
5673

    
5674
  def _ComputeDiskStatus(self, instance, snode, dev):
5675
    """Compute block device status.
5676

5677
    """
5678
    static = self.op.static
5679
    if not static:
5680
      self.cfg.SetDiskID(dev, instance.primary_node)
5681
      dev_pstatus = self.rpc.call_blockdev_find(instance.primary_node, dev)
5682
      if dev_pstatus.offline:
5683
        dev_pstatus = None
5684
      else:
5685
        msg = dev_pstatus.RemoteFailMsg()
5686
        if msg:
5687
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5688
                                   (instance.name, msg))
5689
        dev_pstatus = dev_pstatus.payload
5690
    else:
5691
      dev_pstatus = None
5692

    
5693
    if dev.dev_type in constants.LDS_DRBD:
5694
      # we change the snode then (otherwise we use the one passed in)
5695
      if dev.logical_id[0] == instance.primary_node:
5696
        snode = dev.logical_id[1]
5697
      else:
5698
        snode = dev.logical_id[0]
5699

    
5700
    if snode and not static:
5701
      self.cfg.SetDiskID(dev, snode)
5702
      dev_sstatus = self.rpc.call_blockdev_find(snode, dev)
5703
      if dev_sstatus.offline:
5704
        dev_sstatus = None
5705
      else:
5706
        msg = dev_sstatus.RemoteFailMsg()
5707
        if msg:
5708
          raise errors.OpExecError("Can't compute disk status for %s: %s" %
5709
                                   (instance.name, msg))
5710
        dev_sstatus = dev_sstatus.payload
5711
    else:
5712
      dev_sstatus = None
5713

    
5714
    if dev.children:
5715
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
5716
                      for child in dev.children]
5717
    else:
5718
      dev_children = []
5719

    
5720
    data = {
5721
      "iv_name": dev.iv_name,
5722
      "dev_type": dev.dev_type,
5723
      "logical_id": dev.logical_id,
5724
      "physical_id": dev.physical_id,
5725
      "pstatus": dev_pstatus,
5726
      "sstatus": dev_sstatus,
5727
      "children": dev_children,
5728
      "mode": dev.mode,
5729
      }
5730

    
5731
    return data
5732

    
5733
  def Exec(self, feedback_fn):
5734
    """Gather and return data"""
5735
    result = {}
5736

    
5737
    cluster = self.cfg.GetClusterInfo()
5738

    
5739
    for instance in self.wanted_instances:
5740
      if not self.op.static:
5741
        remote_info = self.rpc.call_instance_info(instance.primary_node,
5742
                                                  instance.name,
5743
                                                  instance.hypervisor)
5744
        remote_info.Raise()
5745
        remote_info = remote_info.data
5746
        if remote_info and "state" in remote_info:
5747
          remote_state = "up"
5748
        else:
5749
          remote_state = "down"
5750
      else:
5751
        remote_state = None
5752
      if instance.admin_up:
5753
        config_state = "up"
5754
      else:
5755
        config_state = "down"
5756

    
5757
      disks = [self._ComputeDiskStatus(instance, None, device)
5758
               for device in instance.disks]
5759

    
5760
      idict = {
5761
        "name": instance.name,
5762
        "config_state": config_state,
5763
        "run_state": remote_state,
5764
        "pnode": instance.primary_node,
5765
        "snodes": instance.secondary_nodes,
5766
        "os": instance.os,
5767
        "nics": [(nic.mac, nic.ip, nic.bridge) for nic in instance.nics],
5768
        "disks": disks,
5769
        "hypervisor": instance.hypervisor,
5770
        "network_port": instance.network_port,
5771
        "hv_instance": instance.hvparams,
5772
        "hv_actual": cluster.FillHV(instance),
5773
        "be_instance": instance.beparams,
5774
        "be_actual": cluster.FillBE(instance),
5775
        }
5776

    
5777
      result[instance.name] = idict
5778

    
5779
    return result
5780

    
5781

    
5782
class LUSetInstanceParams(LogicalUnit):
5783
  """Modifies an instances's parameters.
5784

5785
  """
5786
  HPATH = "instance-modify"
5787
  HTYPE = constants.HTYPE_INSTANCE
5788
  _OP_REQP = ["instance_name"]
5789
  REQ_BGL = False
5790

    
5791
  def CheckArguments(self):
5792
    if not hasattr(self.op, 'nics'):
5793
      self.op.nics = []
5794
    if not hasattr(self.op, 'disks'):
5795
      self.op.disks = []
5796
    if not hasattr(self.op, 'beparams'):
5797
      self.op.beparams = {}
5798
    if not hasattr(self.op, 'hvparams'):
5799
      self.op.hvparams = {}
5800
    self.op.force = getattr(self.op, "force", False)
5801
    if not (self.op.nics or self.op.disks or
5802
            self.op.hvparams or self.op.beparams):
5803
      raise errors.OpPrereqError("No changes submitted")
5804

    
5805
    # Disk validation
5806
    disk_addremove = 0
5807
    for disk_op, disk_dict in self.op.disks:
5808
      if disk_op == constants.DDM_REMOVE:
5809
        disk_addremove += 1
5810
        continue
5811
      elif disk_op == constants.DDM_ADD:
5812
        disk_addremove += 1
5813
      else:
5814
        if not isinstance(disk_op, int):
5815
          raise errors.OpPrereqError("Invalid disk index")
5816
      if disk_op == constants.DDM_ADD:
5817
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
5818
        if mode not in constants.DISK_ACCESS_SET:
5819
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode)
5820
        size = disk_dict.get('size', None)
5821
        if size is None:
5822
          raise errors.OpPrereqError("Required disk parameter size missing")
5823
        try:
5824
          size = int(size)
5825
        except ValueError, err:
5826
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
5827
                                     str(err))
5828
        disk_dict['size'] = size
5829
      else:
5830
        # modification of disk
5831
        if 'size' in disk_dict:
5832
          raise errors.OpPrereqError("Disk size change not possible, use"
5833
                                     " grow-disk")
5834

    
5835
    if disk_addremove > 1:
5836
      raise errors.OpPrereqError("Only one disk add or remove operation"
5837
                                 " supported at a time")
5838

    
5839
    # NIC validation
5840
    nic_addremove = 0
5841
    for nic_op, nic_dict in self.op.nics:
5842
      if nic_op == constants.DDM_REMOVE:
5843
        nic_addremove += 1
5844
        continue
5845
      elif nic_op == constants.DDM_ADD:
5846
        nic_addremove += 1
5847
      else:
5848
        if not isinstance(nic_op, int):
5849
          raise errors.OpPrereqError("Invalid nic index")
5850

    
5851
      # nic_dict should be a dict
5852
      nic_ip = nic_dict.get('ip', None)
5853
      if nic_ip is not None:
5854
        if nic_ip.lower() == constants.VALUE_NONE:
5855
          nic_dict['ip'] = None
5856
        else:
5857
          if not utils.IsValidIP(nic_ip):
5858
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip)
5859

    
5860
      if nic_op == constants.DDM_ADD:
5861
        nic_bridge = nic_dict.get('bridge', None)
5862
        if nic_bridge is None:
5863
          nic_dict['bridge'] = self.cfg.GetDefBridge()
5864
        nic_mac = nic_dict.get('mac', None)
5865
        if nic_mac is None:
5866
          nic_dict['mac'] = constants.VALUE_AUTO
5867

    
5868
      if 'mac' in nic_dict:
5869
        nic_mac = nic_dict['mac']
5870
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
5871
          if not utils.IsValidMac(nic_mac):
5872
            raise errors.OpPrereqError("Invalid MAC address %s" % nic_mac)
5873
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
5874
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
5875
                                     " modifying an existing nic")
5876

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

    
5881
  def ExpandNames(self):
5882
    self._ExpandAndLockInstance()
5883
    self.needed_locks[locking.LEVEL_NODE] = []
5884
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5885

    
5886
  def DeclareLocks(self, level):
5887
    if level == locking.LEVEL_NODE:
5888
      self._LockInstancesNodes()
5889

    
5890
  def BuildHooksEnv(self):
5891
    """Build hooks env.
5892

5893
    This runs on the master, primary and secondaries.
5894

5895
    """
5896
    args = dict()
5897
    if constants.BE_MEMORY in self.be_new:
5898
      args['memory'] = self.be_new[constants.BE_MEMORY]
5899
    if constants.BE_VCPUS in self.be_new:
5900
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
5901
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
5902
    # information at all.
5903
    if self.op.nics:
5904
      args['nics'] = []
5905
      nic_override = dict(self.op.nics)
5906
      for idx, nic in enumerate(self.instance.nics):
5907
        if idx in nic_override:
5908
          this_nic_override = nic_override[idx]
5909
        else:
5910
          this_nic_override = {}
5911
        if 'ip' in this_nic_override:
5912
          ip = this_nic_override['ip']
5913
        else:
5914
          ip = nic.ip
5915
        if 'bridge' in this_nic_override:
5916
          bridge = this_nic_override['bridge']
5917
        else:
5918
          bridge = nic.bridge
5919
        if 'mac' in this_nic_override:
5920
          mac = this_nic_override['mac']
5921
        else:
5922
          mac = nic.mac
5923
        args['nics'].append((ip, bridge, mac))
5924
      if constants.DDM_ADD in nic_override:
5925
        ip = nic_override[constants.DDM_ADD].get('ip', None)
5926
        bridge = nic_override[constants.DDM_ADD]['bridge']
5927
        mac = nic_override[constants.DDM_ADD]['mac']
5928
        args['nics'].append((ip, bridge, mac))
5929
      elif constants.DDM_REMOVE in nic_override:
5930
        del args['nics'][-1]
5931

    
5932
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
5933
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
5934
    return env, nl, nl
5935

    
5936
  def CheckPrereq(self):
5937
    """Check prerequisites.
5938

5939
    This only checks the instance list against the existing names.
5940

5941
    """
5942
    force = self.force = self.op.force
5943

    
5944
    # checking the new params on the primary/secondary nodes
5945

    
5946
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5947
    assert self.instance is not None, \
5948
      "Cannot retrieve locked instance %s" % self.op.instance_name
5949
    pnode = instance.primary_node
5950
    nodelist = list(instance.all_nodes)
5951

    
5952
    # hvparams processing
5953
    if self.op.hvparams:
5954
      i_hvdict = copy.deepcopy(instance.hvparams)
5955
      for key, val in self.op.hvparams.iteritems():
5956
        if val == constants.VALUE_DEFAULT:
5957
          try:
5958
            del i_hvdict[key]
5959
          except KeyError:
5960
            pass
5961
        else:
5962
          i_hvdict[key] = val
5963
      cluster = self.cfg.GetClusterInfo()
5964
      utils.ForceDictType(i_hvdict, constants.HVS_PARAMETER_TYPES)
5965
      hv_new = cluster.FillDict(cluster.hvparams[instance.hypervisor],
5966
                                i_hvdict)
5967
      # local check
5968
      hypervisor.GetHypervisor(
5969
        instance.hypervisor).CheckParameterSyntax(hv_new)
5970
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
5971
      self.hv_new = hv_new # the new actual values
5972
      self.hv_inst = i_hvdict # the new dict (without defaults)
5973
    else:
5974
      self.hv_new = self.hv_inst = {}
5975

    
5976
    # beparams processing
5977
    if self.op.beparams:
5978
      i_bedict = copy.deepcopy(instance.beparams)
5979
      for key, val in self.op.beparams.iteritems():
5980
        if val == constants.VALUE_DEFAULT:
5981
          try:
5982
            del i_bedict[key]
5983
          except KeyError:
5984
            pass
5985
        else:
5986
          i_bedict[key] = val
5987
      cluster = self.cfg.GetClusterInfo()
5988
      utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
5989
      be_new = cluster.FillDict(cluster.beparams[constants.BEGR_DEFAULT],
5990
                                i_bedict)
5991
      self.be_new = be_new # the new actual values
5992
      self.be_inst = i_bedict # the new dict (without defaults)
5993
    else:
5994
      self.be_new = self.be_inst = {}
5995

    
5996
    self.warn = []
5997

    
5998
    if constants.BE_MEMORY in self.op.beparams and not self.force:
5999
      mem_check_list = [pnode]
6000
      if be_new[constants.BE_AUTO_BALANCE]:
6001
        # either we changed auto_balance to yes or it was from before
6002
        mem_check_list.extend(instance.secondary_nodes)
6003
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
6004
                                                  instance.hypervisor)
6005
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
6006
                                         instance.hypervisor)
6007
      if nodeinfo[pnode].failed or not isinstance(nodeinfo[pnode].data, dict):
6008
        # Assume the primary node is unreachable and go ahead
6009
        self.warn.append("Can't get info from primary node %s" % pnode)
6010
      else:
6011
        if not instance_info.failed and instance_info.data:
6012
          current_mem = int(instance_info.data['memory'])
6013
        else:
6014
          # Assume instance not running
6015
          # (there is a slight race condition here, but it's not very probable,
6016
          # and we have no other way to check)
6017
          current_mem = 0
6018
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
6019
                    nodeinfo[pnode].data['memory_free'])
6020
        if miss_mem > 0:
6021
          raise errors.OpPrereqError("This change will prevent the instance"
6022
                                     " from starting, due to %d MB of memory"
6023
                                     " missing on its primary node" % miss_mem)
6024

    
6025
      if be_new[constants.BE_AUTO_BALANCE]:
6026
        for node, nres in nodeinfo.iteritems():
6027
          if node not in instance.secondary_nodes:
6028
            continue
6029
          if nres.failed or not isinstance(nres.data, dict):
6030
            self.warn.append("Can't get info from secondary node %s" % node)
6031
          elif be_new[constants.BE_MEMORY] > nres.data['memory_free']:
6032
            self.warn.append("Not enough memory to failover instance to"
6033
                             " secondary node %s" % node)
6034

    
6035
    # NIC processing
6036
    for nic_op, nic_dict in self.op.nics:
6037
      if nic_op == constants.DDM_REMOVE:
6038
        if not instance.nics:
6039
          raise errors.OpPrereqError("Instance has no NICs, cannot remove")
6040
        continue
6041
      if nic_op != constants.DDM_ADD:
6042
        # an existing nic
6043
        if nic_op < 0 or nic_op >= len(instance.nics):
6044
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
6045
                                     " are 0 to %d" %
6046
                                     (nic_op, len(instance.nics)))
6047
      if 'bridge' in nic_dict:
6048
        nic_bridge = nic_dict['bridge']
6049
        if nic_bridge is None:
6050
          raise errors.OpPrereqError('Cannot set the nic bridge to None')
6051
        if not self.rpc.call_bridges_exist(pnode, [nic_bridge]):
6052
          msg = ("Bridge '%s' doesn't exist on one of"
6053
                 " the instance nodes" % nic_bridge)
6054
          if self.force:
6055
            self.warn.append(msg)
6056
          else:
6057
            raise errors.OpPrereqError(msg)
6058
      if 'mac' in nic_dict:
6059
        nic_mac = nic_dict['mac']
6060
        if nic_mac is None:
6061
          raise errors.OpPrereqError('Cannot set the nic mac to None')
6062
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6063
          # otherwise generate the mac
6064
          nic_dict['mac'] = self.cfg.GenerateMAC()
6065
        else:
6066
          # or validate/reserve the current one
6067
          if self.cfg.IsMacInUse(nic_mac):
6068
            raise errors.OpPrereqError("MAC address %s already in use"
6069
                                       " in cluster" % nic_mac)
6070

    
6071
    # DISK processing
6072
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
6073
      raise errors.OpPrereqError("Disk operations not supported for"
6074
                                 " diskless instances")
6075
    for disk_op, disk_dict in self.op.disks:
6076
      if disk_op == constants.DDM_REMOVE:
6077
        if len(instance.disks) == 1:
6078
          raise errors.OpPrereqError("Cannot remove the last disk of"
6079
                                     " an instance")
6080
        ins_l = self.rpc.call_instance_list([pnode], [instance.hypervisor])
6081
        ins_l = ins_l[pnode]
6082
        if ins_l.failed or not isinstance(ins_l.data, list):
6083
          raise errors.OpPrereqError("Can't contact node '%s'" % pnode)
6084
        if instance.name in ins_l.data:
6085
          raise errors.OpPrereqError("Instance is running, can't remove"
6086
                                     " disks.")
6087

    
6088
      if (disk_op == constants.DDM_ADD and
6089
          len(instance.nics) >= constants.MAX_DISKS):
6090
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
6091
                                   " add more" % constants.MAX_DISKS)
6092
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
6093
        # an existing disk
6094
        if disk_op < 0 or disk_op >= len(instance.disks):
6095
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
6096
                                     " are 0 to %d" %
6097
                                     (disk_op, len(instance.disks)))
6098

    
6099
    return
6100

    
6101
  def Exec(self, feedback_fn):
6102
    """Modifies an instance.
6103

6104
    All parameters take effect only at the next restart of the instance.
6105

6106
    """
6107
    # Process here the warnings from CheckPrereq, as we don't have a
6108
    # feedback_fn there.
6109
    for warn in self.warn:
6110
      feedback_fn("WARNING: %s" % warn)
6111

    
6112
    result = []
6113
    instance = self.instance
6114
    # disk changes
6115
    for disk_op, disk_dict in self.op.disks:
6116
      if disk_op == constants.DDM_REMOVE:
6117
        # remove the last disk
6118
        device = instance.disks.pop()
6119
        device_idx = len(instance.disks)
6120
        for node, disk in device.ComputeNodeTree(instance.primary_node):
6121
          self.cfg.SetDiskID(disk, node)
6122
          msg = self.rpc.call_blockdev_remove(node, disk).RemoteFailMsg()
6123
          if msg:
6124
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
6125
                            " continuing anyway", device_idx, node, msg)
6126
        result.append(("disk/%d" % device_idx, "remove"))
6127
      elif disk_op == constants.DDM_ADD:
6128
        # add a new disk
6129
        if instance.disk_template == constants.DT_FILE:
6130
          file_driver, file_path = instance.disks[0].logical_id
6131
          file_path = os.path.dirname(file_path)
6132
        else:
6133
          file_driver = file_path = None
6134
        disk_idx_base = len(instance.disks)
6135
        new_disk = _GenerateDiskTemplate(self,
6136
                                         instance.disk_template,
6137
                                         instance.name, instance.primary_node,
6138
                                         instance.secondary_nodes,
6139
                                         [disk_dict],
6140
                                         file_path,
6141
                                         file_driver,
6142
                                         disk_idx_base)[0]
6143
        instance.disks.append(new_disk)
6144
        info = _GetInstanceInfoText(instance)
6145

    
6146
        logging.info("Creating volume %s for instance %s",
6147
                     new_disk.iv_name, instance.name)
6148
        # Note: this needs to be kept in sync with _CreateDisks
6149
        #HARDCODE
6150
        for node in instance.all_nodes:
6151
          f_create = node == instance.primary_node
6152
          try:
6153
            _CreateBlockDev(self, node, instance, new_disk,
6154
                            f_create, info, f_create)
6155
          except errors.OpExecError, err:
6156
            self.LogWarning("Failed to create volume %s (%s) on"
6157
                            " node %s: %s",
6158
                            new_disk.iv_name, new_disk, node, err)
6159
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
6160
                       (new_disk.size, new_disk.mode)))
6161
      else:
6162
        # change a given disk
6163
        instance.disks[disk_op].mode = disk_dict['mode']
6164
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
6165
    # NIC changes
6166
    for nic_op, nic_dict in self.op.nics:
6167
      if nic_op == constants.DDM_REMOVE:
6168
        # remove the last nic
6169
        del instance.nics[-1]
6170
        result.append(("nic.%d" % len(instance.nics), "remove"))
6171
      elif nic_op == constants.DDM_ADD:
6172
        # mac and bridge should be set, by now
6173
        mac = nic_dict['mac']
6174
        bridge = nic_dict['bridge']
6175
        new_nic = objects.NIC(mac=mac, ip=nic_dict.get('ip', None),
6176
                              bridge=bridge)
6177
        instance.nics.append(new_nic)
6178
        result.append(("nic.%d" % (len(instance.nics) - 1),
6179
                       "add:mac=%s,ip=%s,bridge=%s" %
6180
                       (new_nic.mac, new_nic.ip, new_nic.bridge)))
6181
      else:
6182
        # change a given nic
6183
        for key in 'mac', 'ip', 'bridge':
6184
          if key in nic_dict:
6185
            setattr(instance.nics[nic_op], key, nic_dict[key])
6186
            result.append(("nic.%s/%d" % (key, nic_op), nic_dict[key]))
6187

    
6188
    # hvparams changes
6189
    if self.op.hvparams:
6190
      instance.hvparams = self.hv_inst
6191
      for key, val in self.op.hvparams.iteritems():
6192
        result.append(("hv/%s" % key, val))
6193

    
6194
    # beparams changes
6195
    if self.op.beparams:
6196
      instance.beparams = self.be_inst
6197
      for key, val in self.op.beparams.iteritems():
6198
        result.append(("be/%s" % key, val))
6199

    
6200
    self.cfg.Update(instance)
6201

    
6202
    return result
6203

    
6204

    
6205
class LUQueryExports(NoHooksLU):
6206
  """Query the exports list
6207

6208
  """
6209
  _OP_REQP = ['nodes']
6210
  REQ_BGL = False
6211

    
6212
  def ExpandNames(self):
6213
    self.needed_locks = {}
6214
    self.share_locks[locking.LEVEL_NODE] = 1
6215
    if not self.op.nodes:
6216
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6217
    else:
6218
      self.needed_locks[locking.LEVEL_NODE] = \
6219
        _GetWantedNodes(self, self.op.nodes)
6220

    
6221
  def CheckPrereq(self):
6222
    """Check prerequisites.
6223

6224
    """
6225
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
6226

    
6227
  def Exec(self, feedback_fn):
6228
    """Compute the list of all the exported system images.
6229

6230
    @rtype: dict
6231
    @return: a dictionary with the structure node->(export-list)
6232
        where export-list is a list of the instances exported on
6233
        that node.
6234

6235
    """
6236
    rpcresult = self.rpc.call_export_list(self.nodes)
6237
    result = {}
6238
    for node in rpcresult:
6239
      if rpcresult[node].failed:
6240
        result[node] = False
6241
      else:
6242
        result[node] = rpcresult[node].data
6243

    
6244
    return result
6245

    
6246

    
6247
class LUExportInstance(LogicalUnit):
6248
  """Export an instance to an image in the cluster.
6249

6250
  """
6251
  HPATH = "instance-export"
6252
  HTYPE = constants.HTYPE_INSTANCE
6253
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
6254
  REQ_BGL = False
6255

    
6256
  def ExpandNames(self):
6257
    self._ExpandAndLockInstance()
6258
    # FIXME: lock only instance primary and destination node
6259
    #
6260
    # Sad but true, for now we have do lock all nodes, as we don't know where
6261
    # the previous export might be, and and in this LU we search for it and
6262
    # remove it from its current node. In the future we could fix this by:
6263
    #  - making a tasklet to search (share-lock all), then create the new one,
6264
    #    then one to remove, after
6265
    #  - removing the removal operation altoghether
6266
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6267

    
6268
  def DeclareLocks(self, level):
6269
    """Last minute lock declaration."""
6270
    # All nodes are locked anyway, so nothing to do here.
6271

    
6272
  def BuildHooksEnv(self):
6273
    """Build hooks env.
6274

6275
    This will run on the master, primary node and target node.
6276

6277
    """
6278
    env = {
6279
      "EXPORT_NODE": self.op.target_node,
6280
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
6281
      }
6282
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
6283
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
6284
          self.op.target_node]
6285
    return env, nl, nl
6286

    
6287
  def CheckPrereq(self):
6288
    """Check prerequisites.
6289

6290
    This checks that the instance and node names are valid.
6291

6292
    """
6293
    instance_name = self.op.instance_name
6294
    self.instance = self.cfg.GetInstanceInfo(instance_name)
6295
    assert self.instance is not None, \
6296
          "Cannot retrieve locked instance %s" % self.op.instance_name
6297
    _CheckNodeOnline(self, self.instance.primary_node)
6298

    
6299
    self.dst_node = self.cfg.GetNodeInfo(
6300
      self.cfg.ExpandNodeName(self.op.target_node))
6301

    
6302
    if self.dst_node is None:
6303
      # This is wrong node name, not a non-locked node
6304
      raise errors.OpPrereqError("Wrong node name %s" % self.op.target_node)
6305
    _CheckNodeOnline(self, self.dst_node.name)
6306
    _CheckNodeNotDrained(self, self.dst_node.name)
6307

    
6308
    # instance disk type verification
6309
    for disk in self.instance.disks:
6310
      if disk.dev_type == constants.LD_FILE:
6311
        raise errors.OpPrereqError("Export not supported for instances with"
6312
                                   " file-based disks")
6313

    
6314
  def Exec(self, feedback_fn):
6315
    """Export an instance to an image in the cluster.
6316

6317
    """
6318
    instance = self.instance
6319
    dst_node = self.dst_node
6320
    src_node = instance.primary_node
6321
    if self.op.shutdown:
6322
      # shutdown the instance, but not the disks
6323
      result = self.rpc.call_instance_shutdown(src_node, instance)
6324
      msg = result.RemoteFailMsg()
6325
      if msg:
6326
        raise errors.OpExecError("Could not shutdown instance %s on"
6327
                                 " node %s: %s" %
6328
                                 (instance.name, src_node, msg))
6329

    
6330
    vgname = self.cfg.GetVGName()
6331

    
6332
    snap_disks = []
6333

    
6334
    # set the disks ID correctly since call_instance_start needs the
6335
    # correct drbd minor to create the symlinks
6336
    for disk in instance.disks:
6337
      self.cfg.SetDiskID(disk, src_node)
6338

    
6339
    try:
6340
      for disk in instance.disks:
6341
        # new_dev_name will be a snapshot of an lvm leaf of the one we passed
6342
        new_dev_name = self.rpc.call_blockdev_snapshot(src_node, disk)
6343
        if new_dev_name.failed or not new_dev_name.data:
6344
          self.LogWarning("Could not snapshot block device %s on node %s",
6345
                          disk.logical_id[1], src_node)
6346
          snap_disks.append(False)
6347
        else:
6348
          new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
6349
                                 logical_id=(vgname, new_dev_name.data),
6350
                                 physical_id=(vgname, new_dev_name.data),
6351
                                 iv_name=disk.iv_name)
6352
          snap_disks.append(new_dev)
6353

    
6354
    finally:
6355
      if self.op.shutdown and instance.admin_up:
6356
        result = self.rpc.call_instance_start(src_node, instance, None, None)
6357
        msg = result.RemoteFailMsg()
6358
        if msg:
6359
          _ShutdownInstanceDisks(self, instance)
6360
          raise errors.OpExecError("Could not start instance: %s" % msg)
6361

    
6362
    # TODO: check for size
6363

    
6364
    cluster_name = self.cfg.GetClusterName()
6365
    for idx, dev in enumerate(snap_disks):
6366
      if dev:
6367
        result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
6368
                                               instance, cluster_name, idx)
6369
        if result.failed or not result.data:
6370
          self.LogWarning("Could not export block device %s from node %s to"
6371
                          " node %s", dev.logical_id[1], src_node,
6372
                          dst_node.name)
6373
        msg = self.rpc.call_blockdev_remove(src_node, dev).RemoteFailMsg()
6374
        if msg:
6375
          self.LogWarning("Could not remove snapshot block device %s from node"
6376
                          " %s: %s", dev.logical_id[1], src_node, msg)
6377

    
6378
    result = self.rpc.call_finalize_export(dst_node.name, instance, snap_disks)
6379
    if result.failed or not result.data:
6380
      self.LogWarning("Could not finalize export for instance %s on node %s",
6381
                      instance.name, dst_node.name)
6382

    
6383
    nodelist = self.cfg.GetNodeList()
6384
    nodelist.remove(dst_node.name)
6385

    
6386
    # on one-node clusters nodelist will be empty after the removal
6387
    # if we proceed the backup would be removed because OpQueryExports
6388
    # substitutes an empty list with the full cluster node list.
6389
    if nodelist:
6390
      exportlist = self.rpc.call_export_list(nodelist)
6391
      for node in exportlist:
6392
        if exportlist[node].failed:
6393
          continue
6394
        if instance.name in exportlist[node].data:
6395
          if not self.rpc.call_export_remove(node, instance.name):
6396
            self.LogWarning("Could not remove older export for instance %s"
6397
                            " on node %s", instance.name, node)
6398

    
6399

    
6400
class LURemoveExport(NoHooksLU):
6401
  """Remove exports related to the named instance.
6402

6403
  """
6404
  _OP_REQP = ["instance_name"]
6405
  REQ_BGL = False
6406

    
6407
  def ExpandNames(self):
6408
    self.needed_locks = {}
6409
    # We need all nodes to be locked in order for RemoveExport to work, but we
6410
    # don't need to lock the instance itself, as nothing will happen to it (and
6411
    # we can remove exports also for a removed instance)
6412
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6413

    
6414
  def CheckPrereq(self):
6415
    """Check prerequisites.
6416
    """
6417
    pass
6418

    
6419
  def Exec(self, feedback_fn):
6420
    """Remove any export.
6421

6422
    """
6423
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
6424
    # If the instance was not found we'll try with the name that was passed in.
6425
    # This will only work if it was an FQDN, though.
6426
    fqdn_warn = False
6427
    if not instance_name:
6428
      fqdn_warn = True
6429
      instance_name = self.op.instance_name
6430

    
6431
    exportlist = self.rpc.call_export_list(self.acquired_locks[
6432
      locking.LEVEL_NODE])
6433
    found = False
6434
    for node in exportlist:
6435
      if exportlist[node].failed:
6436
        self.LogWarning("Failed to query node %s, continuing" % node)
6437
        continue
6438
      if instance_name in exportlist[node].data:
6439
        found = True
6440
        result = self.rpc.call_export_remove(node, instance_name)
6441
        if result.failed or not result.data:
6442
          logging.error("Could not remove export for instance %s"
6443
                        " on node %s", instance_name, node)
6444

    
6445
    if fqdn_warn and not found:
6446
      feedback_fn("Export not found. If trying to remove an export belonging"
6447
                  " to a deleted instance please use its Fully Qualified"
6448
                  " Domain Name.")
6449

    
6450

    
6451
class TagsLU(NoHooksLU):
6452
  """Generic tags LU.
6453

6454
  This is an abstract class which is the parent of all the other tags LUs.
6455

6456
  """
6457

    
6458
  def ExpandNames(self):
6459
    self.needed_locks = {}
6460
    if self.op.kind == constants.TAG_NODE:
6461
      name = self.cfg.ExpandNodeName(self.op.name)
6462
      if name is None:
6463
        raise errors.OpPrereqError("Invalid node name (%s)" %
6464
                                   (self.op.name,))
6465
      self.op.name = name
6466
      self.needed_locks[locking.LEVEL_NODE] = name
6467
    elif self.op.kind == constants.TAG_INSTANCE:
6468
      name = self.cfg.ExpandInstanceName(self.op.name)
6469
      if name is None:
6470
        raise errors.OpPrereqError("Invalid instance name (%s)" %
6471
                                   (self.op.name,))
6472
      self.op.name = name
6473
      self.needed_locks[locking.LEVEL_INSTANCE] = name
6474

    
6475
  def CheckPrereq(self):
6476
    """Check prerequisites.
6477

6478
    """
6479
    if self.op.kind == constants.TAG_CLUSTER:
6480
      self.target = self.cfg.GetClusterInfo()
6481
    elif self.op.kind == constants.TAG_NODE:
6482
      self.target = self.cfg.GetNodeInfo(self.op.name)
6483
    elif self.op.kind == constants.TAG_INSTANCE:
6484
      self.target = self.cfg.GetInstanceInfo(self.op.name)
6485
    else:
6486
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
6487
                                 str(self.op.kind))
6488

    
6489

    
6490
class LUGetTags(TagsLU):
6491
  """Returns the tags of a given object.
6492

6493
  """
6494
  _OP_REQP = ["kind", "name"]
6495
  REQ_BGL = False
6496

    
6497
  def Exec(self, feedback_fn):
6498
    """Returns the tag list.
6499

6500
    """
6501
    return list(self.target.GetTags())
6502

    
6503

    
6504
class LUSearchTags(NoHooksLU):
6505
  """Searches the tags for a given pattern.
6506

6507
  """
6508
  _OP_REQP = ["pattern"]
6509
  REQ_BGL = False
6510

    
6511
  def ExpandNames(self):
6512
    self.needed_locks = {}
6513

    
6514
  def CheckPrereq(self):
6515
    """Check prerequisites.
6516

6517
    This checks the pattern passed for validity by compiling it.
6518

6519
    """
6520
    try:
6521
      self.re = re.compile(self.op.pattern)
6522
    except re.error, err:
6523
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
6524
                                 (self.op.pattern, err))
6525

    
6526
  def Exec(self, feedback_fn):
6527
    """Returns the tag list.
6528

6529
    """
6530
    cfg = self.cfg
6531
    tgts = [("/cluster", cfg.GetClusterInfo())]
6532
    ilist = cfg.GetAllInstancesInfo().values()
6533
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
6534
    nlist = cfg.GetAllNodesInfo().values()
6535
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
6536
    results = []
6537
    for path, target in tgts:
6538
      for tag in target.GetTags():
6539
        if self.re.search(tag):
6540
          results.append((path, tag))
6541
    return results
6542

    
6543

    
6544
class LUAddTags(TagsLU):
6545
  """Sets a tag on a given object.
6546

6547
  """
6548
  _OP_REQP = ["kind", "name", "tags"]
6549
  REQ_BGL = False
6550

    
6551
  def CheckPrereq(self):
6552
    """Check prerequisites.
6553

6554
    This checks the type and length of the tag name and value.
6555

6556
    """
6557
    TagsLU.CheckPrereq(self)
6558
    for tag in self.op.tags:
6559
      objects.TaggableObject.ValidateTag(tag)
6560

    
6561
  def Exec(self, feedback_fn):
6562
    """Sets the tag.
6563

6564
    """
6565
    try:
6566
      for tag in self.op.tags:
6567
        self.target.AddTag(tag)
6568
    except errors.TagError, err:
6569
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
6570
    try:
6571
      self.cfg.Update(self.target)
6572
    except errors.ConfigurationError:
6573
      raise errors.OpRetryError("There has been a modification to the"
6574
                                " config file and the operation has been"
6575
                                " aborted. Please retry.")
6576

    
6577

    
6578
class LUDelTags(TagsLU):
6579
  """Delete a list of tags from a given object.
6580

6581
  """
6582
  _OP_REQP = ["kind", "name", "tags"]
6583
  REQ_BGL = False
6584

    
6585
  def CheckPrereq(self):
6586
    """Check prerequisites.
6587

6588
    This checks that we have the given tag.
6589

6590
    """
6591
    TagsLU.CheckPrereq(self)
6592
    for tag in self.op.tags:
6593
      objects.TaggableObject.ValidateTag(tag)
6594
    del_tags = frozenset(self.op.tags)
6595
    cur_tags = self.target.GetTags()
6596
    if not del_tags <= cur_tags:
6597
      diff_tags = del_tags - cur_tags
6598
      diff_names = ["'%s'" % tag for tag in diff_tags]
6599
      diff_names.sort()
6600
      raise errors.OpPrereqError("Tag(s) %s not found" %
6601
                                 (",".join(diff_names)))
6602

    
6603
  def Exec(self, feedback_fn):
6604
    """Remove the tag from the object.
6605

6606
    """
6607
    for tag in self.op.tags:
6608
      self.target.RemoveTag(tag)
6609
    try:
6610
      self.cfg.Update(self.target)
6611
    except errors.ConfigurationError:
6612
      raise errors.OpRetryError("There has been a modification to the"
6613
                                " config file and the operation has been"
6614
                                " aborted. Please retry.")
6615

    
6616

    
6617
class LUTestDelay(NoHooksLU):
6618
  """Sleep for a specified amount of time.
6619

6620
  This LU sleeps on the master and/or nodes for a specified amount of
6621
  time.
6622

6623
  """
6624
  _OP_REQP = ["duration", "on_master", "on_nodes"]
6625
  REQ_BGL = False
6626

    
6627
  def ExpandNames(self):
6628
    """Expand names and set required locks.
6629

6630
    This expands the node list, if any.
6631

6632
    """
6633
    self.needed_locks = {}
6634
    if self.op.on_nodes:
6635
      # _GetWantedNodes can be used here, but is not always appropriate to use
6636
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
6637
      # more information.
6638
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
6639
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
6640

    
6641
  def CheckPrereq(self):
6642
    """Check prerequisites.
6643

6644
    """
6645

    
6646
  def Exec(self, feedback_fn):
6647
    """Do the actual sleep.
6648

6649
    """
6650
    if self.op.on_master:
6651
      if not utils.TestDelay(self.op.duration):
6652
        raise errors.OpExecError("Error during master delay test")
6653
    if self.op.on_nodes:
6654
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
6655
      if not result:
6656
        raise errors.OpExecError("Complete failure from rpc call")
6657
      for node, node_result in result.items():
6658
        node_result.Raise()
6659
        if not node_result.data:
6660
          raise errors.OpExecError("Failure during rpc call to node %s,"
6661
                                   " result: %s" % (node, node_result.data))
6662

    
6663

    
6664
class IAllocator(object):
6665
  """IAllocator framework.
6666

6667
  An IAllocator instance has three sets of attributes:
6668
    - cfg that is needed to query the cluster
6669
    - input data (all members of the _KEYS class attribute are required)
6670
    - four buffer attributes (in|out_data|text), that represent the
6671
      input (to the external script) in text and data structure format,
6672
      and the output from it, again in two formats
6673
    - the result variables from the script (success, info, nodes) for
6674
      easy usage
6675

6676
  """
6677
  _ALLO_KEYS = [
6678
    "mem_size", "disks", "disk_template",
6679
    "os", "tags", "nics", "vcpus", "hypervisor",
6680
    ]
6681
  _RELO_KEYS = [
6682
    "relocate_from",
6683
    ]
6684

    
6685
  def __init__(self, lu, mode, name, **kwargs):
6686
    self.lu = lu
6687
    # init buffer variables
6688
    self.in_text = self.out_text = self.in_data = self.out_data = None
6689
    # init all input fields so that pylint is happy
6690
    self.mode = mode
6691
    self.name = name
6692
    self.mem_size = self.disks = self.disk_template = None
6693
    self.os = self.tags = self.nics = self.vcpus = None
6694
    self.hypervisor = None
6695
    self.relocate_from = None
6696
    # computed fields
6697
    self.required_nodes = None
6698
    # init result fields
6699
    self.success = self.info = self.nodes = None
6700
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6701
      keyset = self._ALLO_KEYS
6702
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6703
      keyset = self._RELO_KEYS
6704
    else:
6705
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
6706
                                   " IAllocator" % self.mode)
6707
    for key in kwargs:
6708
      if key not in keyset:
6709
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
6710
                                     " IAllocator" % key)
6711
      setattr(self, key, kwargs[key])
6712
    for key in keyset:
6713
      if key not in kwargs:
6714
        raise errors.ProgrammerError("Missing input parameter '%s' to"
6715
                                     " IAllocator" % key)
6716
    self._BuildInputData()
6717

    
6718
  def _ComputeClusterData(self):
6719
    """Compute the generic allocator input data.
6720

6721
    This is the data that is independent of the actual operation.
6722

6723
    """
6724
    cfg = self.lu.cfg
6725
    cluster_info = cfg.GetClusterInfo()
6726
    # cluster data
6727
    data = {
6728
      "version": constants.IALLOCATOR_VERSION,
6729
      "cluster_name": cfg.GetClusterName(),
6730
      "cluster_tags": list(cluster_info.GetTags()),
6731
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
6732
      # we don't have job IDs
6733
      }
6734
    iinfo = cfg.GetAllInstancesInfo().values()
6735
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
6736

    
6737
    # node data
6738
    node_results = {}
6739
    node_list = cfg.GetNodeList()
6740

    
6741
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6742
      hypervisor_name = self.hypervisor
6743
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
6744
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
6745

    
6746
    node_data = self.lu.rpc.call_node_info(node_list, cfg.GetVGName(),
6747
                                           hypervisor_name)
6748
    node_iinfo = self.lu.rpc.call_all_instances_info(node_list,
6749
                       cluster_info.enabled_hypervisors)
6750
    for nname, nresult in node_data.items():
6751
      # first fill in static (config-based) values
6752
      ninfo = cfg.GetNodeInfo(nname)
6753
      pnr = {
6754
        "tags": list(ninfo.GetTags()),
6755
        "primary_ip": ninfo.primary_ip,
6756
        "secondary_ip": ninfo.secondary_ip,
6757
        "offline": ninfo.offline,
6758
        "drained": ninfo.drained,
6759
        "master_candidate": ninfo.master_candidate,
6760
        }
6761

    
6762
      if not ninfo.offline:
6763
        nresult.Raise()
6764
        if not isinstance(nresult.data, dict):
6765
          raise errors.OpExecError("Can't get data for node %s" % nname)
6766
        remote_info = nresult.data
6767
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
6768
                     'vg_size', 'vg_free', 'cpu_total']:
6769
          if attr not in remote_info:
6770
            raise errors.OpExecError("Node '%s' didn't return attribute"
6771
                                     " '%s'" % (nname, attr))
6772
          try:
6773
            remote_info[attr] = int(remote_info[attr])
6774
          except ValueError, err:
6775
            raise errors.OpExecError("Node '%s' returned invalid value"
6776
                                     " for '%s': %s" % (nname, attr, err))
6777
        # compute memory used by primary instances
6778
        i_p_mem = i_p_up_mem = 0
6779
        for iinfo, beinfo in i_list:
6780
          if iinfo.primary_node == nname:
6781
            i_p_mem += beinfo[constants.BE_MEMORY]
6782
            if iinfo.name not in node_iinfo[nname].data:
6783
              i_used_mem = 0
6784
            else:
6785
              i_used_mem = int(node_iinfo[nname].data[iinfo.name]['memory'])
6786
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
6787
            remote_info['memory_free'] -= max(0, i_mem_diff)
6788

    
6789
            if iinfo.admin_up:
6790
              i_p_up_mem += beinfo[constants.BE_MEMORY]
6791

    
6792
        # compute memory used by instances
6793
        pnr_dyn = {
6794
          "total_memory": remote_info['memory_total'],
6795
          "reserved_memory": remote_info['memory_dom0'],
6796
          "free_memory": remote_info['memory_free'],
6797
          "total_disk": remote_info['vg_size'],
6798
          "free_disk": remote_info['vg_free'],
6799
          "total_cpus": remote_info['cpu_total'],
6800
          "i_pri_memory": i_p_mem,
6801
          "i_pri_up_memory": i_p_up_mem,
6802
          }
6803
        pnr.update(pnr_dyn)
6804

    
6805
      node_results[nname] = pnr
6806
    data["nodes"] = node_results
6807

    
6808
    # instance data
6809
    instance_data = {}
6810
    for iinfo, beinfo in i_list:
6811
      nic_data = [{"mac": n.mac, "ip": n.ip, "bridge": n.bridge}
6812
                  for n in iinfo.nics]
6813
      pir = {
6814
        "tags": list(iinfo.GetTags()),
6815
        "admin_up": iinfo.admin_up,
6816
        "vcpus": beinfo[constants.BE_VCPUS],
6817
        "memory": beinfo[constants.BE_MEMORY],
6818
        "os": iinfo.os,
6819
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
6820
        "nics": nic_data,
6821
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
6822
        "disk_template": iinfo.disk_template,
6823
        "hypervisor": iinfo.hypervisor,
6824
        }
6825
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
6826
                                                 pir["disks"])
6827
      instance_data[iinfo.name] = pir
6828

    
6829
    data["instances"] = instance_data
6830

    
6831
    self.in_data = data
6832

    
6833
  def _AddNewInstance(self):
6834
    """Add new instance data to allocator structure.
6835

6836
    This in combination with _AllocatorGetClusterData will create the
6837
    correct structure needed as input for the allocator.
6838

6839
    The checks for the completeness of the opcode must have already been
6840
    done.
6841

6842
    """
6843
    data = self.in_data
6844

    
6845
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
6846

    
6847
    if self.disk_template in constants.DTS_NET_MIRROR:
6848
      self.required_nodes = 2
6849
    else:
6850
      self.required_nodes = 1
6851
    request = {
6852
      "type": "allocate",
6853
      "name": self.name,
6854
      "disk_template": self.disk_template,
6855
      "tags": self.tags,
6856
      "os": self.os,
6857
      "vcpus": self.vcpus,
6858
      "memory": self.mem_size,
6859
      "disks": self.disks,
6860
      "disk_space_total": disk_space,
6861
      "nics": self.nics,
6862
      "required_nodes": self.required_nodes,
6863
      }
6864
    data["request"] = request
6865

    
6866
  def _AddRelocateInstance(self):
6867
    """Add relocate instance data to allocator structure.
6868

6869
    This in combination with _IAllocatorGetClusterData will create the
6870
    correct structure needed as input for the allocator.
6871

6872
    The checks for the completeness of the opcode must have already been
6873
    done.
6874

6875
    """
6876
    instance = self.lu.cfg.GetInstanceInfo(self.name)
6877
    if instance is None:
6878
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
6879
                                   " IAllocator" % self.name)
6880

    
6881
    if instance.disk_template not in constants.DTS_NET_MIRROR:
6882
      raise errors.OpPrereqError("Can't relocate non-mirrored instances")
6883

    
6884
    if len(instance.secondary_nodes) != 1:
6885
      raise errors.OpPrereqError("Instance has not exactly one secondary node")
6886

    
6887
    self.required_nodes = 1
6888
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
6889
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
6890

    
6891
    request = {
6892
      "type": "relocate",
6893
      "name": self.name,
6894
      "disk_space_total": disk_space,
6895
      "required_nodes": self.required_nodes,
6896
      "relocate_from": self.relocate_from,
6897
      }
6898
    self.in_data["request"] = request
6899

    
6900
  def _BuildInputData(self):
6901
    """Build input data structures.
6902

6903
    """
6904
    self._ComputeClusterData()
6905

    
6906
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
6907
      self._AddNewInstance()
6908
    else:
6909
      self._AddRelocateInstance()
6910

    
6911
    self.in_text = serializer.Dump(self.in_data)
6912

    
6913
  def Run(self, name, validate=True, call_fn=None):
6914
    """Run an instance allocator and return the results.
6915

6916
    """
6917
    if call_fn is None:
6918
      call_fn = self.lu.rpc.call_iallocator_runner
6919
    data = self.in_text
6920

    
6921
    result = call_fn(self.lu.cfg.GetMasterNode(), name, self.in_text)
6922
    result.Raise()
6923

    
6924
    if not isinstance(result.data, (list, tuple)) or len(result.data) != 4:
6925
      raise errors.OpExecError("Invalid result from master iallocator runner")
6926

    
6927
    rcode, stdout, stderr, fail = result.data
6928

    
6929
    if rcode == constants.IARUN_NOTFOUND:
6930
      raise errors.OpExecError("Can't find allocator '%s'" % name)
6931
    elif rcode == constants.IARUN_FAILURE:
6932
      raise errors.OpExecError("Instance allocator call failed: %s,"
6933
                               " output: %s" % (fail, stdout+stderr))
6934
    self.out_text = stdout
6935
    if validate:
6936
      self._ValidateResult()
6937

    
6938
  def _ValidateResult(self):
6939
    """Process the allocator results.
6940

6941
    This will process and if successful save the result in
6942
    self.out_data and the other parameters.
6943

6944
    """
6945
    try:
6946
      rdict = serializer.Load(self.out_text)
6947
    except Exception, err:
6948
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
6949

    
6950
    if not isinstance(rdict, dict):
6951
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
6952

    
6953
    for key in "success", "info", "nodes":
6954
      if key not in rdict:
6955
        raise errors.OpExecError("Can't parse iallocator results:"
6956
                                 " missing key '%s'" % key)
6957
      setattr(self, key, rdict[key])
6958

    
6959
    if not isinstance(rdict["nodes"], list):
6960
      raise errors.OpExecError("Can't parse iallocator results: 'nodes' key"
6961
                               " is not a list")
6962
    self.out_data = rdict
6963

    
6964

    
6965
class LUTestAllocator(NoHooksLU):
6966
  """Run allocator tests.
6967

6968
  This LU runs the allocator tests
6969

6970
  """
6971
  _OP_REQP = ["direction", "mode", "name"]
6972

    
6973
  def CheckPrereq(self):
6974
    """Check prerequisites.
6975

6976
    This checks the opcode parameters depending on the director and mode test.
6977

6978
    """
6979
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
6980
      for attr in ["name", "mem_size", "disks", "disk_template",
6981
                   "os", "tags", "nics", "vcpus"]:
6982
        if not hasattr(self.op, attr):
6983
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
6984
                                     attr)
6985
      iname = self.cfg.ExpandInstanceName(self.op.name)
6986
      if iname is not None:
6987
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
6988
                                   iname)
6989
      if not isinstance(self.op.nics, list):
6990
        raise errors.OpPrereqError("Invalid parameter 'nics'")
6991
      for row in self.op.nics:
6992
        if (not isinstance(row, dict) or
6993
            "mac" not in row or
6994
            "ip" not in row or
6995
            "bridge" not in row):
6996
          raise errors.OpPrereqError("Invalid contents of the"
6997
                                     " 'nics' parameter")
6998
      if not isinstance(self.op.disks, list):
6999
        raise errors.OpPrereqError("Invalid parameter 'disks'")
7000
      for row in self.op.disks:
7001
        if (not isinstance(row, dict) or
7002
            "size" not in row or
7003
            not isinstance(row["size"], int) or
7004
            "mode" not in row or
7005
            row["mode"] not in ['r', 'w']):
7006
          raise errors.OpPrereqError("Invalid contents of the"
7007
                                     " 'disks' parameter")
7008
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
7009
        self.op.hypervisor = self.cfg.GetHypervisorType()
7010
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
7011
      if not hasattr(self.op, "name"):
7012
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input")
7013
      fname = self.cfg.ExpandInstanceName(self.op.name)
7014
      if fname is None:
7015
        raise errors.OpPrereqError("Instance '%s' not found for relocation" %
7016
                                   self.op.name)
7017
      self.op.name = fname
7018
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
7019
    else:
7020
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
7021
                                 self.op.mode)
7022

    
7023
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
7024
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
7025
        raise errors.OpPrereqError("Missing allocator name")
7026
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
7027
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
7028
                                 self.op.direction)
7029

    
7030
  def Exec(self, feedback_fn):
7031
    """Run the allocator test.
7032

7033
    """
7034
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
7035
      ial = IAllocator(self,
7036
                       mode=self.op.mode,
7037
                       name=self.op.name,
7038
                       mem_size=self.op.mem_size,
7039
                       disks=self.op.disks,
7040
                       disk_template=self.op.disk_template,
7041
                       os=self.op.os,
7042
                       tags=self.op.tags,
7043
                       nics=self.op.nics,
7044
                       vcpus=self.op.vcpus,
7045
                       hypervisor=self.op.hypervisor,
7046
                       )
7047
    else:
7048
      ial = IAllocator(self,
7049
                       mode=self.op.mode,
7050
                       name=self.op.name,
7051
                       relocate_from=list(self.relocate_from),
7052
                       )
7053

    
7054
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
7055
      result = ial.in_text
7056
    else:
7057
      ial.Run(self.op.allocator, validate=False)
7058
      result = ial.out_text
7059
    return result