Statistics
| Branch: | Tag: | Revision:

root / lib / cmdlib.py @ c1c31426

History | View | Annotate | Download (333.1 kB)

1
#
2
#
3

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

    
21

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

    
24
# pylint: disable-msg=W0201
25

    
26
# W0201 since most LU attributes are defined in CheckPrereq or similar
27
# functions
28

    
29
import os
30
import os.path
31
import time
32
import re
33
import platform
34
import logging
35
import copy
36
import OpenSSL
37

    
38
from ganeti import ssh
39
from ganeti import utils
40
from ganeti import errors
41
from ganeti import hypervisor
42
from ganeti import locking
43
from ganeti import constants
44
from ganeti import objects
45
from ganeti import serializer
46
from ganeti import ssconf
47

    
48

    
49
class LogicalUnit(object):
50
  """Logical Unit base class.
51

52
  Subclasses must follow these rules:
53
    - implement ExpandNames
54
    - implement CheckPrereq (except when tasklets are used)
55
    - implement Exec (except when tasklets are used)
56
    - implement BuildHooksEnv
57
    - redefine HPATH and HTYPE
58
    - optionally redefine their run requirements:
59
        REQ_BGL: the LU needs to hold the Big Ganeti Lock exclusively
60

61
  Note that all commands require root permissions.
62

63
  @ivar dry_run_result: the value (if any) that will be returned to the caller
64
      in dry-run mode (signalled by opcode dry_run parameter)
65

66
  """
67
  HPATH = None
68
  HTYPE = None
69
  _OP_REQP = []
70
  REQ_BGL = True
71

    
72
  def __init__(self, processor, op, context, rpc):
73
    """Constructor for LogicalUnit.
74

75
    This needs to be overridden in derived classes in order to check op
76
    validity.
77

78
    """
79
    self.proc = processor
80
    self.op = op
81
    self.cfg = context.cfg
82
    self.context = context
83
    self.rpc = rpc
84
    # Dicts used to declare locking needs to mcpu
85
    self.needed_locks = None
86
    self.acquired_locks = {}
87
    self.share_locks = dict.fromkeys(locking.LEVELS, 0)
88
    self.add_locks = {}
89
    self.remove_locks = {}
90
    # Used to force good behavior when calling helper functions
91
    self.recalculate_locks = {}
92
    self.__ssh = None
93
    # logging
94
    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
95
    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
96
    self.LogStep = processor.LogStep # pylint: disable-msg=C0103
97
    # support for dry-run
98
    self.dry_run_result = None
99
    # support for generic debug attribute
100
    if (not hasattr(self.op, "debug_level") or
101
        not isinstance(self.op.debug_level, int)):
102
      self.op.debug_level = 0
103

    
104
    # Tasklets
105
    self.tasklets = None
106

    
107
    for attr_name in self._OP_REQP:
108
      attr_val = getattr(op, attr_name, None)
109
      if attr_val is None:
110
        raise errors.OpPrereqError("Required parameter '%s' missing" %
111
                                   attr_name, errors.ECODE_INVAL)
112

    
113
    self.CheckArguments()
114

    
115
  def __GetSSH(self):
116
    """Returns the SshRunner object
117

118
    """
119
    if not self.__ssh:
120
      self.__ssh = ssh.SshRunner(self.cfg.GetClusterName())
121
    return self.__ssh
122

    
123
  ssh = property(fget=__GetSSH)
124

    
125
  def CheckArguments(self):
126
    """Check syntactic validity for the opcode arguments.
127

128
    This method is for doing a simple syntactic check and ensure
129
    validity of opcode parameters, without any cluster-related
130
    checks. While the same can be accomplished in ExpandNames and/or
131
    CheckPrereq, doing these separate is better because:
132

133
      - ExpandNames is left as as purely a lock-related function
134
      - CheckPrereq is run after we have acquired locks (and possible
135
        waited for them)
136

137
    The function is allowed to change the self.op attribute so that
138
    later methods can no longer worry about missing parameters.
139

140
    """
141
    pass
142

    
143
  def ExpandNames(self):
144
    """Expand names for this LU.
145

146
    This method is called before starting to execute the opcode, and it should
147
    update all the parameters of the opcode to their canonical form (e.g. a
148
    short node name must be fully expanded after this method has successfully
149
    completed). This way locking, hooks, logging, ecc. can work correctly.
150

151
    LUs which implement this method must also populate the self.needed_locks
152
    member, as a dict with lock levels as keys, and a list of needed lock names
153
    as values. Rules:
154

155
      - use an empty dict if you don't need any lock
156
      - if you don't need any lock at a particular level omit that level
157
      - don't put anything for the BGL level
158
      - if you want all locks at a level use locking.ALL_SET as a value
159

160
    If you need to share locks (rather than acquire them exclusively) at one
161
    level you can modify self.share_locks, setting a true value (usually 1) for
162
    that level. By default locks are not shared.
163

164
    This function can also define a list of tasklets, which then will be
165
    executed in order instead of the usual LU-level CheckPrereq and Exec
166
    functions, if those are not defined by the LU.
167

168
    Examples::
169

170
      # Acquire all nodes and one instance
171
      self.needed_locks = {
172
        locking.LEVEL_NODE: locking.ALL_SET,
173
        locking.LEVEL_INSTANCE: ['instance1.example.tld'],
174
      }
175
      # Acquire just two nodes
176
      self.needed_locks = {
177
        locking.LEVEL_NODE: ['node1.example.tld', 'node2.example.tld'],
178
      }
179
      # Acquire no locks
180
      self.needed_locks = {} # No, you can't leave it to the default value None
181

182
    """
183
    # The implementation of this method is mandatory only if the new LU is
184
    # concurrent, so that old LUs don't need to be changed all at the same
185
    # time.
186
    if self.REQ_BGL:
187
      self.needed_locks = {} # Exclusive LUs don't need locks.
188
    else:
189
      raise NotImplementedError
190

    
191
  def DeclareLocks(self, level):
192
    """Declare LU locking needs for a level
193

194
    While most LUs can just declare their locking needs at ExpandNames time,
195
    sometimes there's the need to calculate some locks after having acquired
196
    the ones before. This function is called just before acquiring locks at a
197
    particular level, but after acquiring the ones at lower levels, and permits
198
    such calculations. It can be used to modify self.needed_locks, and by
199
    default it does nothing.
200

201
    This function is only called if you have something already set in
202
    self.needed_locks for the level.
203

204
    @param level: Locking level which is going to be locked
205
    @type level: member of ganeti.locking.LEVELS
206

207
    """
208

    
209
  def CheckPrereq(self):
210
    """Check prerequisites for this LU.
211

212
    This method should check that the prerequisites for the execution
213
    of this LU are fulfilled. It can do internode communication, but
214
    it should be idempotent - no cluster or system changes are
215
    allowed.
216

217
    The method should raise errors.OpPrereqError in case something is
218
    not fulfilled. Its return value is ignored.
219

220
    This method should also update all the parameters of the opcode to
221
    their canonical form if it hasn't been done by ExpandNames before.
222

223
    """
224
    if self.tasklets is not None:
225
      for (idx, tl) in enumerate(self.tasklets):
226
        logging.debug("Checking prerequisites for tasklet %s/%s",
227
                      idx + 1, len(self.tasklets))
228
        tl.CheckPrereq()
229
    else:
230
      raise NotImplementedError
231

    
232
  def Exec(self, feedback_fn):
233
    """Execute the LU.
234

235
    This method should implement the actual work. It should raise
236
    errors.OpExecError for failures that are somewhat dealt with in
237
    code, or expected.
238

239
    """
240
    if self.tasklets is not None:
241
      for (idx, tl) in enumerate(self.tasklets):
242
        logging.debug("Executing tasklet %s/%s", idx + 1, len(self.tasklets))
243
        tl.Exec(feedback_fn)
244
    else:
245
      raise NotImplementedError
246

    
247
  def BuildHooksEnv(self):
248
    """Build hooks environment for this LU.
249

250
    This method should return a three-node tuple consisting of: a dict
251
    containing the environment that will be used for running the
252
    specific hook for this LU, a list of node names on which the hook
253
    should run before the execution, and a list of node names on which
254
    the hook should run after the execution.
255

256
    The keys of the dict must not have 'GANETI_' prefixed as this will
257
    be handled in the hooks runner. Also note additional keys will be
258
    added by the hooks runner. If the LU doesn't define any
259
    environment, an empty dict (and not None) should be returned.
260

261
    No nodes should be returned as an empty list (and not None).
262

263
    Note that if the HPATH for a LU class is None, this function will
264
    not be called.
265

266
    """
267
    raise NotImplementedError
268

    
269
  def HooksCallBack(self, phase, hook_results, feedback_fn, lu_result):
270
    """Notify the LU about the results of its hooks.
271

272
    This method is called every time a hooks phase is executed, and notifies
273
    the Logical Unit about the hooks' result. The LU can then use it to alter
274
    its result based on the hooks.  By default the method does nothing and the
275
    previous result is passed back unchanged but any LU can define it if it
276
    wants to use the local cluster hook-scripts somehow.
277

278
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
279
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
280
    @param hook_results: the results of the multi-node hooks rpc call
281
    @param feedback_fn: function used send feedback back to the caller
282
    @param lu_result: the previous Exec result this LU had, or None
283
        in the PRE phase
284
    @return: the new Exec result, based on the previous result
285
        and hook results
286

287
    """
288
    # API must be kept, thus we ignore the unused argument and could
289
    # be a function warnings
290
    # pylint: disable-msg=W0613,R0201
291
    return lu_result
292

    
293
  def _ExpandAndLockInstance(self):
294
    """Helper function to expand and lock an instance.
295

296
    Many LUs that work on an instance take its name in self.op.instance_name
297
    and need to expand it and then declare the expanded name for locking. This
298
    function does it, and then updates self.op.instance_name to the expanded
299
    name. It also initializes needed_locks as a dict, if this hasn't been done
300
    before.
301

302
    """
303
    if self.needed_locks is None:
304
      self.needed_locks = {}
305
    else:
306
      assert locking.LEVEL_INSTANCE not in self.needed_locks, \
307
        "_ExpandAndLockInstance called with instance-level locks set"
308
    self.op.instance_name = _ExpandInstanceName(self.cfg,
309
                                                self.op.instance_name)
310
    self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
311

    
312
  def _LockInstancesNodes(self, primary_only=False):
313
    """Helper function to declare instances' nodes for locking.
314

315
    This function should be called after locking one or more instances to lock
316
    their nodes. Its effect is populating self.needed_locks[locking.LEVEL_NODE]
317
    with all primary or secondary nodes for instances already locked and
318
    present in self.needed_locks[locking.LEVEL_INSTANCE].
319

320
    It should be called from DeclareLocks, and for safety only works if
321
    self.recalculate_locks[locking.LEVEL_NODE] is set.
322

323
    In the future it may grow parameters to just lock some instance's nodes, or
324
    to just lock primaries or secondary nodes, if needed.
325

326
    If should be called in DeclareLocks in a way similar to::
327

328
      if level == locking.LEVEL_NODE:
329
        self._LockInstancesNodes()
330

331
    @type primary_only: boolean
332
    @param primary_only: only lock primary nodes of locked instances
333

334
    """
335
    assert locking.LEVEL_NODE in self.recalculate_locks, \
336
      "_LockInstancesNodes helper function called with no nodes to recalculate"
337

    
338
    # TODO: check if we're really been called with the instance locks held
339

    
340
    # For now we'll replace self.needed_locks[locking.LEVEL_NODE], but in the
341
    # future we might want to have different behaviors depending on the value
342
    # of self.recalculate_locks[locking.LEVEL_NODE]
343
    wanted_nodes = []
344
    for instance_name in self.acquired_locks[locking.LEVEL_INSTANCE]:
345
      instance = self.context.cfg.GetInstanceInfo(instance_name)
346
      wanted_nodes.append(instance.primary_node)
347
      if not primary_only:
348
        wanted_nodes.extend(instance.secondary_nodes)
349

    
350
    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
351
      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
352
    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
353
      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
354

    
355
    del self.recalculate_locks[locking.LEVEL_NODE]
356

    
357

    
358
class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
359
  """Simple LU which runs no hooks.
360

361
  This LU is intended as a parent for other LogicalUnits which will
362
  run no hooks, in order to reduce duplicate code.
363

364
  """
365
  HPATH = None
366
  HTYPE = None
367

    
368
  def BuildHooksEnv(self):
369
    """Empty BuildHooksEnv for NoHooksLu.
370

371
    This just raises an error.
372

373
    """
374
    assert False, "BuildHooksEnv called for NoHooksLUs"
375

    
376

    
377
class Tasklet:
378
  """Tasklet base class.
379

380
  Tasklets are subcomponents for LUs. LUs can consist entirely of tasklets or
381
  they can mix legacy code with tasklets. Locking needs to be done in the LU,
382
  tasklets know nothing about locks.
383

384
  Subclasses must follow these rules:
385
    - Implement CheckPrereq
386
    - Implement Exec
387

388
  """
389
  def __init__(self, lu):
390
    self.lu = lu
391

    
392
    # Shortcuts
393
    self.cfg = lu.cfg
394
    self.rpc = lu.rpc
395

    
396
  def CheckPrereq(self):
397
    """Check prerequisites for this tasklets.
398

399
    This method should check whether the prerequisites for the execution of
400
    this tasklet are fulfilled. It can do internode communication, but it
401
    should be idempotent - no cluster or system changes are allowed.
402

403
    The method should raise errors.OpPrereqError in case something is not
404
    fulfilled. Its return value is ignored.
405

406
    This method should also update all parameters to their canonical form if it
407
    hasn't been done before.
408

409
    """
410
    raise NotImplementedError
411

    
412
  def Exec(self, feedback_fn):
413
    """Execute the tasklet.
414

415
    This method should implement the actual work. It should raise
416
    errors.OpExecError for failures that are somewhat dealt with in code, or
417
    expected.
418

419
    """
420
    raise NotImplementedError
421

    
422

    
423
def _GetWantedNodes(lu, nodes):
424
  """Returns list of checked and expanded node names.
425

426
  @type lu: L{LogicalUnit}
427
  @param lu: the logical unit on whose behalf we execute
428
  @type nodes: list
429
  @param nodes: list of node names or None for all nodes
430
  @rtype: list
431
  @return: the list of nodes, sorted
432
  @raise errors.ProgrammerError: if the nodes parameter is wrong type
433

434
  """
435
  if not isinstance(nodes, list):
436
    raise errors.OpPrereqError("Invalid argument type 'nodes'",
437
                               errors.ECODE_INVAL)
438

    
439
  if not nodes:
440
    raise errors.ProgrammerError("_GetWantedNodes should only be called with a"
441
      " non-empty list of nodes whose name is to be expanded.")
442

    
443
  wanted = [_ExpandNodeName(lu.cfg, name) for name in nodes]
444
  return utils.NiceSort(wanted)
445

    
446

    
447
def _GetWantedInstances(lu, instances):
448
  """Returns list of checked and expanded instance names.
449

450
  @type lu: L{LogicalUnit}
451
  @param lu: the logical unit on whose behalf we execute
452
  @type instances: list
453
  @param instances: list of instance names or None for all instances
454
  @rtype: list
455
  @return: the list of instances, sorted
456
  @raise errors.OpPrereqError: if the instances parameter is wrong type
457
  @raise errors.OpPrereqError: if any of the passed instances is not found
458

459
  """
460
  if not isinstance(instances, list):
461
    raise errors.OpPrereqError("Invalid argument type 'instances'",
462
                               errors.ECODE_INVAL)
463

    
464
  if instances:
465
    wanted = [_ExpandInstanceName(lu.cfg, name) for name in instances]
466
  else:
467
    wanted = utils.NiceSort(lu.cfg.GetInstanceList())
468
  return wanted
469

    
470

    
471
def _CheckOutputFields(static, dynamic, selected):
472
  """Checks whether all selected fields are valid.
473

474
  @type static: L{utils.FieldSet}
475
  @param static: static fields set
476
  @type dynamic: L{utils.FieldSet}
477
  @param dynamic: dynamic fields set
478

479
  """
480
  f = utils.FieldSet()
481
  f.Extend(static)
482
  f.Extend(dynamic)
483

    
484
  delta = f.NonMatching(selected)
485
  if delta:
486
    raise errors.OpPrereqError("Unknown output fields selected: %s"
487
                               % ",".join(delta), errors.ECODE_INVAL)
488

    
489

    
490
def _CheckBooleanOpField(op, name):
491
  """Validates boolean opcode parameters.
492

493
  This will ensure that an opcode parameter is either a boolean value,
494
  or None (but that it always exists).
495

496
  """
497
  val = getattr(op, name, None)
498
  if not (val is None or isinstance(val, bool)):
499
    raise errors.OpPrereqError("Invalid boolean parameter '%s' (%s)" %
500
                               (name, str(val)), errors.ECODE_INVAL)
501
  setattr(op, name, val)
502

    
503

    
504
def _CheckGlobalHvParams(params):
505
  """Validates that given hypervisor params are not global ones.
506

507
  This will ensure that instances don't get customised versions of
508
  global params.
509

510
  """
511
  used_globals = constants.HVC_GLOBALS.intersection(params)
512
  if used_globals:
513
    msg = ("The following hypervisor parameters are global and cannot"
514
           " be customized at instance level, please modify them at"
515
           " cluster level: %s" % utils.CommaJoin(used_globals))
516
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
517

    
518

    
519
def _CheckNodeOnline(lu, node):
520
  """Ensure that a given node is online.
521

522
  @param lu: the LU on behalf of which we make the check
523
  @param node: the node to check
524
  @raise errors.OpPrereqError: if the node is offline
525

526
  """
527
  if lu.cfg.GetNodeInfo(node).offline:
528
    raise errors.OpPrereqError("Can't use offline node %s" % node,
529
                               errors.ECODE_INVAL)
530

    
531

    
532
def _CheckNodeNotDrained(lu, node):
533
  """Ensure that a given node is not drained.
534

535
  @param lu: the LU on behalf of which we make the check
536
  @param node: the node to check
537
  @raise errors.OpPrereqError: if the node is drained
538

539
  """
540
  if lu.cfg.GetNodeInfo(node).drained:
541
    raise errors.OpPrereqError("Can't use drained node %s" % node,
542
                               errors.ECODE_INVAL)
543

    
544

    
545
def _CheckNodeHasOS(lu, node, os_name, force_variant):
546
  """Ensure that a node supports a given OS.
547

548
  @param lu: the LU on behalf of which we make the check
549
  @param node: the node to check
550
  @param os_name: the OS to query about
551
  @param force_variant: whether to ignore variant errors
552
  @raise errors.OpPrereqError: if the node is not supporting the OS
553

554
  """
555
  result = lu.rpc.call_os_get(node, os_name)
556
  result.Raise("OS '%s' not in supported OS list for node %s" %
557
               (os_name, node),
558
               prereq=True, ecode=errors.ECODE_INVAL)
559
  if not force_variant:
560
    _CheckOSVariant(result.payload, os_name)
561

    
562

    
563
def _CheckDiskTemplate(template):
564
  """Ensure a given disk template is valid.
565

566
  """
567
  if template not in constants.DISK_TEMPLATES:
568
    msg = ("Invalid disk template name '%s', valid templates are: %s" %
569
           (template, utils.CommaJoin(constants.DISK_TEMPLATES)))
570
    raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
571

    
572

    
573
def _CheckInstanceDown(lu, instance, reason):
574
  """Ensure that an instance is not running."""
575
  if instance.admin_up:
576
    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
577
                               (instance.name, reason), errors.ECODE_STATE)
578

    
579
  pnode = instance.primary_node
580
  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
581
  ins_l.Raise("Can't contact node %s for instance information" % pnode,
582
              prereq=True, ecode=errors.ECODE_ENVIRON)
583

    
584
  if instance.name in ins_l.payload:
585
    raise errors.OpPrereqError("Instance %s is running, %s" %
586
                               (instance.name, reason), errors.ECODE_STATE)
587

    
588

    
589
def _ExpandItemName(fn, name, kind):
590
  """Expand an item name.
591

592
  @param fn: the function to use for expansion
593
  @param name: requested item name
594
  @param kind: text description ('Node' or 'Instance')
595
  @return: the resolved (full) name
596
  @raise errors.OpPrereqError: if the item is not found
597

598
  """
599
  full_name = fn(name)
600
  if full_name is None:
601
    raise errors.OpPrereqError("%s '%s' not known" % (kind, name),
602
                               errors.ECODE_NOENT)
603
  return full_name
604

    
605

    
606
def _ExpandNodeName(cfg, name):
607
  """Wrapper over L{_ExpandItemName} for nodes."""
608
  return _ExpandItemName(cfg.ExpandNodeName, name, "Node")
609

    
610

    
611
def _ExpandInstanceName(cfg, name):
612
  """Wrapper over L{_ExpandItemName} for instance."""
613
  return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
614

    
615

    
616
def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
617
                          memory, vcpus, nics, disk_template, disks,
618
                          bep, hvp, hypervisor_name):
619
  """Builds instance related env variables for hooks
620

621
  This builds the hook environment from individual variables.
622

623
  @type name: string
624
  @param name: the name of the instance
625
  @type primary_node: string
626
  @param primary_node: the name of the instance's primary node
627
  @type secondary_nodes: list
628
  @param secondary_nodes: list of secondary nodes as strings
629
  @type os_type: string
630
  @param os_type: the name of the instance's OS
631
  @type status: boolean
632
  @param status: the should_run status of the instance
633
  @type memory: string
634
  @param memory: the memory size of the instance
635
  @type vcpus: string
636
  @param vcpus: the count of VCPUs the instance has
637
  @type nics: list
638
  @param nics: list of tuples (ip, mac, mode, link) representing
639
      the NICs the instance has
640
  @type disk_template: string
641
  @param disk_template: the disk template of the instance
642
  @type disks: list
643
  @param disks: the list of (size, mode) pairs
644
  @type bep: dict
645
  @param bep: the backend parameters for the instance
646
  @type hvp: dict
647
  @param hvp: the hypervisor parameters for the instance
648
  @type hypervisor_name: string
649
  @param hypervisor_name: the hypervisor for the instance
650
  @rtype: dict
651
  @return: the hook environment for this instance
652

653
  """
654
  if status:
655
    str_status = "up"
656
  else:
657
    str_status = "down"
658
  env = {
659
    "OP_TARGET": name,
660
    "INSTANCE_NAME": name,
661
    "INSTANCE_PRIMARY": primary_node,
662
    "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
663
    "INSTANCE_OS_TYPE": os_type,
664
    "INSTANCE_STATUS": str_status,
665
    "INSTANCE_MEMORY": memory,
666
    "INSTANCE_VCPUS": vcpus,
667
    "INSTANCE_DISK_TEMPLATE": disk_template,
668
    "INSTANCE_HYPERVISOR": hypervisor_name,
669
  }
670

    
671
  if nics:
672
    nic_count = len(nics)
673
    for idx, (ip, mac, mode, link) in enumerate(nics):
674
      if ip is None:
675
        ip = ""
676
      env["INSTANCE_NIC%d_IP" % idx] = ip
677
      env["INSTANCE_NIC%d_MAC" % idx] = mac
678
      env["INSTANCE_NIC%d_MODE" % idx] = mode
679
      env["INSTANCE_NIC%d_LINK" % idx] = link
680
      if mode == constants.NIC_MODE_BRIDGED:
681
        env["INSTANCE_NIC%d_BRIDGE" % idx] = link
682
  else:
683
    nic_count = 0
684

    
685
  env["INSTANCE_NIC_COUNT"] = nic_count
686

    
687
  if disks:
688
    disk_count = len(disks)
689
    for idx, (size, mode) in enumerate(disks):
690
      env["INSTANCE_DISK%d_SIZE" % idx] = size
691
      env["INSTANCE_DISK%d_MODE" % idx] = mode
692
  else:
693
    disk_count = 0
694

    
695
  env["INSTANCE_DISK_COUNT"] = disk_count
696

    
697
  for source, kind in [(bep, "BE"), (hvp, "HV")]:
698
    for key, value in source.items():
699
      env["INSTANCE_%s_%s" % (kind, key)] = value
700

    
701
  return env
702

    
703

    
704
def _NICListToTuple(lu, nics):
705
  """Build a list of nic information tuples.
706

707
  This list is suitable to be passed to _BuildInstanceHookEnv or as a return
708
  value in LUQueryInstanceData.
709

710
  @type lu:  L{LogicalUnit}
711
  @param lu: the logical unit on whose behalf we execute
712
  @type nics: list of L{objects.NIC}
713
  @param nics: list of nics to convert to hooks tuples
714

715
  """
716
  hooks_nics = []
717
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[constants.PP_DEFAULT]
718
  for nic in nics:
719
    ip = nic.ip
720
    mac = nic.mac
721
    filled_params = objects.FillDict(c_nicparams, nic.nicparams)
722
    mode = filled_params[constants.NIC_MODE]
723
    link = filled_params[constants.NIC_LINK]
724
    hooks_nics.append((ip, mac, mode, link))
725
  return hooks_nics
726

    
727

    
728
def _BuildInstanceHookEnvByObject(lu, instance, override=None):
729
  """Builds instance related env variables for hooks from an object.
730

731
  @type lu: L{LogicalUnit}
732
  @param lu: the logical unit on whose behalf we execute
733
  @type instance: L{objects.Instance}
734
  @param instance: the instance for which we should build the
735
      environment
736
  @type override: dict
737
  @param override: dictionary with key/values that will override
738
      our values
739
  @rtype: dict
740
  @return: the hook environment dictionary
741

742
  """
743
  cluster = lu.cfg.GetClusterInfo()
744
  bep = cluster.FillBE(instance)
745
  hvp = cluster.FillHV(instance)
746
  args = {
747
    'name': instance.name,
748
    'primary_node': instance.primary_node,
749
    'secondary_nodes': instance.secondary_nodes,
750
    'os_type': instance.os,
751
    'status': instance.admin_up,
752
    'memory': bep[constants.BE_MEMORY],
753
    'vcpus': bep[constants.BE_VCPUS],
754
    'nics': _NICListToTuple(lu, instance.nics),
755
    'disk_template': instance.disk_template,
756
    'disks': [(disk.size, disk.mode) for disk in instance.disks],
757
    'bep': bep,
758
    'hvp': hvp,
759
    'hypervisor_name': instance.hypervisor,
760
  }
761
  if override:
762
    args.update(override)
763
  return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
764

    
765

    
766
def _AdjustCandidatePool(lu, exceptions):
767
  """Adjust the candidate pool after node operations.
768

769
  """
770
  mod_list = lu.cfg.MaintainCandidatePool(exceptions)
771
  if mod_list:
772
    lu.LogInfo("Promoted nodes to master candidate role: %s",
773
               utils.CommaJoin(node.name for node in mod_list))
774
    for name in mod_list:
775
      lu.context.ReaddNode(name)
776
  mc_now, mc_max, _ = lu.cfg.GetMasterCandidateStats(exceptions)
777
  if mc_now > mc_max:
778
    lu.LogInfo("Note: more nodes are candidates (%d) than desired (%d)" %
779
               (mc_now, mc_max))
780

    
781

    
782
def _DecideSelfPromotion(lu, exceptions=None):
783
  """Decide whether I should promote myself as a master candidate.
784

785
  """
786
  cp_size = lu.cfg.GetClusterInfo().candidate_pool_size
787
  mc_now, mc_should, _ = lu.cfg.GetMasterCandidateStats(exceptions)
788
  # the new node will increase mc_max with one, so:
789
  mc_should = min(mc_should + 1, cp_size)
790
  return mc_now < mc_should
791

    
792

    
793
def _CheckNicsBridgesExist(lu, target_nics, target_node,
794
                               profile=constants.PP_DEFAULT):
795
  """Check that the brigdes needed by a list of nics exist.
796

797
  """
798
  c_nicparams = lu.cfg.GetClusterInfo().nicparams[profile]
799
  paramslist = [objects.FillDict(c_nicparams, nic.nicparams)
800
                for nic in target_nics]
801
  brlist = [params[constants.NIC_LINK] for params in paramslist
802
            if params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED]
803
  if brlist:
804
    result = lu.rpc.call_bridges_exist(target_node, brlist)
805
    result.Raise("Error checking bridges on destination node '%s'" %
806
                 target_node, prereq=True, ecode=errors.ECODE_ENVIRON)
807

    
808

    
809
def _CheckInstanceBridgesExist(lu, instance, node=None):
810
  """Check that the brigdes needed by an instance exist.
811

812
  """
813
  if node is None:
814
    node = instance.primary_node
815
  _CheckNicsBridgesExist(lu, instance.nics, node)
816

    
817

    
818
def _CheckOSVariant(os_obj, name):
819
  """Check whether an OS name conforms to the os variants specification.
820

821
  @type os_obj: L{objects.OS}
822
  @param os_obj: OS object to check
823
  @type name: string
824
  @param name: OS name passed by the user, to check for validity
825

826
  """
827
  if not os_obj.supported_variants:
828
    return
829
  try:
830
    variant = name.split("+", 1)[1]
831
  except IndexError:
832
    raise errors.OpPrereqError("OS name must include a variant",
833
                               errors.ECODE_INVAL)
834

    
835
  if variant not in os_obj.supported_variants:
836
    raise errors.OpPrereqError("Unsupported OS variant", errors.ECODE_INVAL)
837

    
838

    
839
def _GetNodeInstancesInner(cfg, fn):
840
  return [i for i in cfg.GetAllInstancesInfo().values() if fn(i)]
841

    
842

    
843
def _GetNodeInstances(cfg, node_name):
844
  """Returns a list of all primary and secondary instances on a node.
845

846
  """
847

    
848
  return _GetNodeInstancesInner(cfg, lambda inst: node_name in inst.all_nodes)
849

    
850

    
851
def _GetNodePrimaryInstances(cfg, node_name):
852
  """Returns primary instances on a node.
853

854
  """
855
  return _GetNodeInstancesInner(cfg,
856
                                lambda inst: node_name == inst.primary_node)
857

    
858

    
859
def _GetNodeSecondaryInstances(cfg, node_name):
860
  """Returns secondary instances on a node.
861

862
  """
863
  return _GetNodeInstancesInner(cfg,
864
                                lambda inst: node_name in inst.secondary_nodes)
865

    
866

    
867
def _GetStorageTypeArgs(cfg, storage_type):
868
  """Returns the arguments for a storage type.
869

870
  """
871
  # Special case for file storage
872
  if storage_type == constants.ST_FILE:
873
    # storage.FileStorage wants a list of storage directories
874
    return [[cfg.GetFileStorageDir()]]
875

    
876
  return []
877

    
878

    
879
def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
880
  faulty = []
881

    
882
  for dev in instance.disks:
883
    cfg.SetDiskID(dev, node_name)
884

    
885
  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
886
  result.Raise("Failed to get disk status from node %s" % node_name,
887
               prereq=prereq, ecode=errors.ECODE_ENVIRON)
888

    
889
  for idx, bdev_status in enumerate(result.payload):
890
    if bdev_status and bdev_status.ldisk_status == constants.LDS_FAULTY:
891
      faulty.append(idx)
892

    
893
  return faulty
894

    
895

    
896
def _FormatTimestamp(secs):
897
  """Formats a Unix timestamp with the local timezone.
898

899
  """
900
  return time.strftime("%F %T %Z", time.gmtime(secs))
901

    
902

    
903
class LUPostInitCluster(LogicalUnit):
904
  """Logical unit for running hooks after cluster initialization.
905

906
  """
907
  HPATH = "cluster-init"
908
  HTYPE = constants.HTYPE_CLUSTER
909
  _OP_REQP = []
910

    
911
  def BuildHooksEnv(self):
912
    """Build hooks env.
913

914
    """
915
    env = {"OP_TARGET": self.cfg.GetClusterName()}
916
    mn = self.cfg.GetMasterNode()
917
    return env, [], [mn]
918

    
919
  def CheckPrereq(self):
920
    """No prerequisites to check.
921

922
    """
923
    return True
924

    
925
  def Exec(self, feedback_fn):
926
    """Nothing to do.
927

928
    """
929
    return True
930

    
931

    
932
class LUDestroyCluster(LogicalUnit):
933
  """Logical unit for destroying the cluster.
934

935
  """
936
  HPATH = "cluster-destroy"
937
  HTYPE = constants.HTYPE_CLUSTER
938
  _OP_REQP = []
939

    
940
  def BuildHooksEnv(self):
941
    """Build hooks env.
942

943
    """
944
    env = {"OP_TARGET": self.cfg.GetClusterName()}
945
    return env, [], []
946

    
947
  def CheckPrereq(self):
948
    """Check prerequisites.
949

950
    This checks whether the cluster is empty.
951

952
    Any errors are signaled by raising errors.OpPrereqError.
953

954
    """
955
    master = self.cfg.GetMasterNode()
956

    
957
    nodelist = self.cfg.GetNodeList()
958
    if len(nodelist) != 1 or nodelist[0] != master:
959
      raise errors.OpPrereqError("There are still %d node(s) in"
960
                                 " this cluster." % (len(nodelist) - 1),
961
                                 errors.ECODE_INVAL)
962
    instancelist = self.cfg.GetInstanceList()
963
    if instancelist:
964
      raise errors.OpPrereqError("There are still %d instance(s) in"
965
                                 " this cluster." % len(instancelist),
966
                                 errors.ECODE_INVAL)
967

    
968
  def Exec(self, feedback_fn):
969
    """Destroys the cluster.
970

971
    """
972
    master = self.cfg.GetMasterNode()
973
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
974

    
975
    # Run post hooks on master node before it's removed
976
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
977
    try:
978
      hm.RunPhase(constants.HOOKS_PHASE_POST, [master])
979
    except:
980
      # pylint: disable-msg=W0702
981
      self.LogWarning("Errors occurred running hooks on %s" % master)
982

    
983
    result = self.rpc.call_node_stop_master(master, False)
984
    result.Raise("Could not disable the master role")
985

    
986
    if modify_ssh_setup:
987
      priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
988
      utils.CreateBackup(priv_key)
989
      utils.CreateBackup(pub_key)
990

    
991
    return master
992

    
993

    
994
def _VerifyCertificateInner(filename, expired, not_before, not_after, now,
995
                            warn_days=constants.SSL_CERT_EXPIRATION_WARN,
996
                            error_days=constants.SSL_CERT_EXPIRATION_ERROR):
997
  """Verifies certificate details for LUVerifyCluster.
998

999
  """
1000
  if expired:
1001
    msg = "Certificate %s is expired" % filename
1002

    
1003
    if not_before is not None and not_after is not None:
1004
      msg += (" (valid from %s to %s)" %
1005
              (_FormatTimestamp(not_before),
1006
               _FormatTimestamp(not_after)))
1007
    elif not_before is not None:
1008
      msg += " (valid from %s)" % _FormatTimestamp(not_before)
1009
    elif not_after is not None:
1010
      msg += " (valid until %s)" % _FormatTimestamp(not_after)
1011

    
1012
    return (LUVerifyCluster.ETYPE_ERROR, msg)
1013

    
1014
  elif not_before is not None and not_before > now:
1015
    return (LUVerifyCluster.ETYPE_WARNING,
1016
            "Certificate %s not yet valid (valid from %s)" %
1017
            (filename, _FormatTimestamp(not_before)))
1018

    
1019
  elif not_after is not None:
1020
    remaining_days = int((not_after - now) / (24 * 3600))
1021

    
1022
    msg = ("Certificate %s expires in %d days" % (filename, remaining_days))
1023

    
1024
    if remaining_days <= error_days:
1025
      return (LUVerifyCluster.ETYPE_ERROR, msg)
1026

    
1027
    if remaining_days <= warn_days:
1028
      return (LUVerifyCluster.ETYPE_WARNING, msg)
1029

    
1030
  return (None, None)
1031

    
1032

    
1033
def _VerifyCertificate(filename):
1034
  """Verifies a certificate for LUVerifyCluster.
1035

1036
  @type filename: string
1037
  @param filename: Path to PEM file
1038

1039
  """
1040
  try:
1041
    cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
1042
                                           utils.ReadFile(filename))
1043
  except Exception, err: # pylint: disable-msg=W0703
1044
    return (LUVerifyCluster.ETYPE_ERROR,
1045
            "Failed to load X509 certificate %s: %s" % (filename, err))
1046

    
1047
  # Depending on the pyOpenSSL version, this can just return (None, None)
1048
  (not_before, not_after) = utils.GetX509CertValidity(cert)
1049

    
1050
  return _VerifyCertificateInner(filename, cert.has_expired(),
1051
                                 not_before, not_after, time.time())
1052

    
1053

    
1054
class LUVerifyCluster(LogicalUnit):
1055
  """Verifies the cluster status.
1056

1057
  """
1058
  HPATH = "cluster-verify"
1059
  HTYPE = constants.HTYPE_CLUSTER
1060
  _OP_REQP = ["skip_checks", "verbose", "error_codes", "debug_simulate_errors"]
1061
  REQ_BGL = False
1062

    
1063
  TCLUSTER = "cluster"
1064
  TNODE = "node"
1065
  TINSTANCE = "instance"
1066

    
1067
  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
1068
  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
1069
  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
1070
  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
1071
  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
1072
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1073
  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
1074
  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
1075
  ENODEDRBD = (TNODE, "ENODEDRBD")
1076
  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
1077
  ENODEHOOKS = (TNODE, "ENODEHOOKS")
1078
  ENODEHV = (TNODE, "ENODEHV")
1079
  ENODELVM = (TNODE, "ENODELVM")
1080
  ENODEN1 = (TNODE, "ENODEN1")
1081
  ENODENET = (TNODE, "ENODENET")
1082
  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
1083
  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
1084
  ENODERPC = (TNODE, "ENODERPC")
1085
  ENODESSH = (TNODE, "ENODESSH")
1086
  ENODEVERSION = (TNODE, "ENODEVERSION")
1087
  ENODESETUP = (TNODE, "ENODESETUP")
1088
  ENODETIME = (TNODE, "ENODETIME")
1089

    
1090
  ETYPE_FIELD = "code"
1091
  ETYPE_ERROR = "ERROR"
1092
  ETYPE_WARNING = "WARNING"
1093

    
1094
  class NodeImage(object):
1095
    """A class representing the logical and physical status of a node.
1096

1097
    @ivar volumes: a structure as returned from
1098
        L{ganeti.backend.GetVolumeList} (runtime)
1099
    @ivar instances: a list of running instances (runtime)
1100
    @ivar pinst: list of configured primary instances (config)
1101
    @ivar sinst: list of configured secondary instances (config)
1102
    @ivar sbp: diction of {secondary-node: list of instances} of all peers
1103
        of this node (config)
1104
    @ivar mfree: free memory, as reported by hypervisor (runtime)
1105
    @ivar dfree: free disk, as reported by the node (runtime)
1106
    @ivar offline: the offline status (config)
1107
    @type rpc_fail: boolean
1108
    @ivar rpc_fail: whether the RPC verify call was successfull (overall,
1109
        not whether the individual keys were correct) (runtime)
1110
    @type lvm_fail: boolean
1111
    @ivar lvm_fail: whether the RPC call didn't return valid LVM data
1112
    @type hyp_fail: boolean
1113
    @ivar hyp_fail: whether the RPC call didn't return the instance list
1114
    @type ghost: boolean
1115
    @ivar ghost: whether this is a known node or not (config)
1116

1117
    """
1118
    def __init__(self, offline=False):
1119
      self.volumes = {}
1120
      self.instances = []
1121
      self.pinst = []
1122
      self.sinst = []
1123
      self.sbp = {}
1124
      self.mfree = 0
1125
      self.dfree = 0
1126
      self.offline = offline
1127
      self.rpc_fail = False
1128
      self.lvm_fail = False
1129
      self.hyp_fail = False
1130
      self.ghost = False
1131

    
1132
  def ExpandNames(self):
1133
    self.needed_locks = {
1134
      locking.LEVEL_NODE: locking.ALL_SET,
1135
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1136
    }
1137
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
1138

    
1139
  def _Error(self, ecode, item, msg, *args, **kwargs):
1140
    """Format an error message.
1141

1142
    Based on the opcode's error_codes parameter, either format a
1143
    parseable error code, or a simpler error string.
1144

1145
    This must be called only from Exec and functions called from Exec.
1146

1147
    """
1148
    ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
1149
    itype, etxt = ecode
1150
    # first complete the msg
1151
    if args:
1152
      msg = msg % args
1153
    # then format the whole message
1154
    if self.op.error_codes:
1155
      msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
1156
    else:
1157
      if item:
1158
        item = " " + item
1159
      else:
1160
        item = ""
1161
      msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
1162
    # and finally report it via the feedback_fn
1163
    self._feedback_fn("  - %s" % msg)
1164

    
1165
  def _ErrorIf(self, cond, *args, **kwargs):
1166
    """Log an error message if the passed condition is True.
1167

1168
    """
1169
    cond = bool(cond) or self.op.debug_simulate_errors
1170
    if cond:
1171
      self._Error(*args, **kwargs)
1172
    # do not mark the operation as failed for WARN cases only
1173
    if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
1174
      self.bad = self.bad or cond
1175

    
1176
  def _VerifyNode(self, ninfo, nresult):
1177
    """Run multiple tests against a node.
1178

1179
    Test list:
1180

1181
      - compares ganeti version
1182
      - checks vg existence and size > 20G
1183
      - checks config file checksum
1184
      - checks ssh to other nodes
1185

1186
    @type ninfo: L{objects.Node}
1187
    @param ninfo: the node to check
1188
    @param nresult: the results from the node
1189
    @rtype: boolean
1190
    @return: whether overall this call was successful (and we can expect
1191
         reasonable values in the respose)
1192

1193
    """
1194
    node = ninfo.name
1195
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1196

    
1197
    # main result, nresult should be a non-empty dict
1198
    test = not nresult or not isinstance(nresult, dict)
1199
    _ErrorIf(test, self.ENODERPC, node,
1200
                  "unable to verify node: no data returned")
1201
    if test:
1202
      return False
1203

    
1204
    # compares ganeti version
1205
    local_version = constants.PROTOCOL_VERSION
1206
    remote_version = nresult.get("version", None)
1207
    test = not (remote_version and
1208
                isinstance(remote_version, (list, tuple)) and
1209
                len(remote_version) == 2)
1210
    _ErrorIf(test, self.ENODERPC, node,
1211
             "connection to node returned invalid data")
1212
    if test:
1213
      return False
1214

    
1215
    test = local_version != remote_version[0]
1216
    _ErrorIf(test, self.ENODEVERSION, node,
1217
             "incompatible protocol versions: master %s,"
1218
             " node %s", local_version, remote_version[0])
1219
    if test:
1220
      return False
1221

    
1222
    # node seems compatible, we can actually try to look into its results
1223

    
1224
    # full package version
1225
    self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
1226
                  self.ENODEVERSION, node,
1227
                  "software version mismatch: master %s, node %s",
1228
                  constants.RELEASE_VERSION, remote_version[1],
1229
                  code=self.ETYPE_WARNING)
1230

    
1231
    hyp_result = nresult.get(constants.NV_HYPERVISOR, None)
1232
    if isinstance(hyp_result, dict):
1233
      for hv_name, hv_result in hyp_result.iteritems():
1234
        test = hv_result is not None
1235
        _ErrorIf(test, self.ENODEHV, node,
1236
                 "hypervisor %s verify failure: '%s'", hv_name, hv_result)
1237

    
1238

    
1239
    test = nresult.get(constants.NV_NODESETUP,
1240
                           ["Missing NODESETUP results"])
1241
    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
1242
             "; ".join(test))
1243

    
1244
    return True
1245

    
1246
  def _VerifyNodeTime(self, ninfo, nresult,
1247
                      nvinfo_starttime, nvinfo_endtime):
1248
    """Check the node time.
1249

1250
    @type ninfo: L{objects.Node}
1251
    @param ninfo: the node to check
1252
    @param nresult: the remote results for the node
1253
    @param nvinfo_starttime: the start time of the RPC call
1254
    @param nvinfo_endtime: the end time of the RPC call
1255

1256
    """
1257
    node = ninfo.name
1258
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1259

    
1260
    ntime = nresult.get(constants.NV_TIME, None)
1261
    try:
1262
      ntime_merged = utils.MergeTime(ntime)
1263
    except (ValueError, TypeError):
1264
      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
1265
      return
1266

    
1267
    if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
1268
      ntime_diff = "%.01fs" % abs(nvinfo_starttime - ntime_merged)
1269
    elif ntime_merged > (nvinfo_endtime + constants.NODE_MAX_CLOCK_SKEW):
1270
      ntime_diff = "%.01fs" % abs(ntime_merged - nvinfo_endtime)
1271
    else:
1272
      ntime_diff = None
1273

    
1274
    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
1275
             "Node time diverges by at least %s from master node time",
1276
             ntime_diff)
1277

    
1278
  def _VerifyNodeLVM(self, ninfo, nresult, vg_name):
1279
    """Check the node time.
1280

1281
    @type ninfo: L{objects.Node}
1282
    @param ninfo: the node to check
1283
    @param nresult: the remote results for the node
1284
    @param vg_name: the configured VG name
1285

1286
    """
1287
    if vg_name is None:
1288
      return
1289

    
1290
    node = ninfo.name
1291
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1292

    
1293
    # checks vg existence and size > 20G
1294
    vglist = nresult.get(constants.NV_VGLIST, None)
1295
    test = not vglist
1296
    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
1297
    if not test:
1298
      vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
1299
                                            constants.MIN_VG_SIZE)
1300
      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
1301

    
1302
    # check pv names
1303
    pvlist = nresult.get(constants.NV_PVLIST, None)
1304
    test = pvlist is None
1305
    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
1306
    if not test:
1307
      # check that ':' is not present in PV names, since it's a
1308
      # special character for lvcreate (denotes the range of PEs to
1309
      # use on the PV)
1310
      for _, pvname, owner_vg in pvlist:
1311
        test = ":" in pvname
1312
        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
1313
                 " '%s' of VG '%s'", pvname, owner_vg)
1314

    
1315
  def _VerifyNodeNetwork(self, ninfo, nresult):
1316
    """Check the node time.
1317

1318
    @type ninfo: L{objects.Node}
1319
    @param ninfo: the node to check
1320
    @param nresult: the remote results for the node
1321

1322
    """
1323
    node = ninfo.name
1324
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1325

    
1326
    test = constants.NV_NODELIST not in nresult
1327
    _ErrorIf(test, self.ENODESSH, node,
1328
             "node hasn't returned node ssh connectivity data")
1329
    if not test:
1330
      if nresult[constants.NV_NODELIST]:
1331
        for a_node, a_msg in nresult[constants.NV_NODELIST].items():
1332
          _ErrorIf(True, self.ENODESSH, node,
1333
                   "ssh communication with node '%s': %s", a_node, a_msg)
1334

    
1335
    test = constants.NV_NODENETTEST not in nresult
1336
    _ErrorIf(test, self.ENODENET, node,
1337
             "node hasn't returned node tcp connectivity data")
1338
    if not test:
1339
      if nresult[constants.NV_NODENETTEST]:
1340
        nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
1341
        for anode in nlist:
1342
          _ErrorIf(True, self.ENODENET, node,
1343
                   "tcp communication with node '%s': %s",
1344
                   anode, nresult[constants.NV_NODENETTEST][anode])
1345

    
1346
  def _VerifyInstance(self, instance, instanceconfig, node_image):
1347
    """Verify an instance.
1348

1349
    This function checks to see if the required block devices are
1350
    available on the instance's node.
1351

1352
    """
1353
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1354
    node_current = instanceconfig.primary_node
1355

    
1356
    node_vol_should = {}
1357
    instanceconfig.MapLVsByNode(node_vol_should)
1358

    
1359
    for node in node_vol_should:
1360
      n_img = node_image[node]
1361
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1362
        # ignore missing volumes on offline or broken nodes
1363
        continue
1364
      for volume in node_vol_should[node]:
1365
        test = volume not in n_img.volumes
1366
        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
1367
                 "volume %s missing on node %s", volume, node)
1368

    
1369
    if instanceconfig.admin_up:
1370
      pri_img = node_image[node_current]
1371
      test = instance not in pri_img.instances and not pri_img.offline
1372
      _ErrorIf(test, self.EINSTANCEDOWN, instance,
1373
               "instance not running on its primary node %s",
1374
               node_current)
1375

    
1376
    for node, n_img in node_image.items():
1377
      if (not node == node_current):
1378
        test = instance in n_img.instances
1379
        _ErrorIf(test, self.EINSTANCEWRONGNODE, instance,
1380
                 "instance should not run on node %s", node)
1381

    
1382
  def _VerifyOrphanVolumes(self, node_vol_should, node_image):
1383
    """Verify if there are any unknown volumes in the cluster.
1384

1385
    The .os, .swap and backup volumes are ignored. All other volumes are
1386
    reported as unknown.
1387

1388
    """
1389
    for node, n_img in node_image.items():
1390
      if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
1391
        # skip non-healthy nodes
1392
        continue
1393
      for volume in n_img.volumes:
1394
        test = (node not in node_vol_should or
1395
                volume not in node_vol_should[node])
1396
        self._ErrorIf(test, self.ENODEORPHANLV, node,
1397
                      "volume %s is unknown", volume)
1398

    
1399
  def _VerifyOrphanInstances(self, instancelist, node_image):
1400
    """Verify the list of running instances.
1401

1402
    This checks what instances are running but unknown to the cluster.
1403

1404
    """
1405
    for node, n_img in node_image.items():
1406
      for o_inst in n_img.instances:
1407
        test = o_inst not in instancelist
1408
        self._ErrorIf(test, self.ENODEORPHANINSTANCE, node,
1409
                      "instance %s on node %s should not exist", o_inst, node)
1410

    
1411
  def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
1412
    """Verify N+1 Memory Resilience.
1413

1414
    Check that if one single node dies we can still start all the
1415
    instances it was primary for.
1416

1417
    """
1418
    for node, n_img in node_image.items():
1419
      # This code checks that every node which is now listed as
1420
      # secondary has enough memory to host all instances it is
1421
      # supposed to should a single other node in the cluster fail.
1422
      # FIXME: not ready for failover to an arbitrary node
1423
      # FIXME: does not support file-backed instances
1424
      # WARNING: we currently take into account down instances as well
1425
      # as up ones, considering that even if they're down someone
1426
      # might want to start them even in the event of a node failure.
1427
      for prinode, instances in n_img.sbp.items():
1428
        needed_mem = 0
1429
        for instance in instances:
1430
          bep = self.cfg.GetClusterInfo().FillBE(instance_cfg[instance])
1431
          if bep[constants.BE_AUTO_BALANCE]:
1432
            needed_mem += bep[constants.BE_MEMORY]
1433
        test = n_img.mfree < needed_mem
1434
        self._ErrorIf(test, self.ENODEN1, node,
1435
                      "not enough memory on to accommodate"
1436
                      " failovers should peer node %s fail", prinode)
1437

    
1438
  def _VerifyNodeFiles(self, ninfo, nresult, file_list, local_cksum,
1439
                       master_files):
1440
    """Verifies and computes the node required file checksums.
1441

1442
    @type ninfo: L{objects.Node}
1443
    @param ninfo: the node to check
1444
    @param nresult: the remote results for the node
1445
    @param file_list: required list of files
1446
    @param local_cksum: dictionary of local files and their checksums
1447
    @param master_files: list of files that only masters should have
1448

1449
    """
1450
    node = ninfo.name
1451
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1452

    
1453
    remote_cksum = nresult.get(constants.NV_FILELIST, None)
1454
    test = not isinstance(remote_cksum, dict)
1455
    _ErrorIf(test, self.ENODEFILECHECK, node,
1456
             "node hasn't returned file checksum data")
1457
    if test:
1458
      return
1459

    
1460
    for file_name in file_list:
1461
      node_is_mc = ninfo.master_candidate
1462
      must_have = (file_name not in master_files) or node_is_mc
1463
      # missing
1464
      test1 = file_name not in remote_cksum
1465
      # invalid checksum
1466
      test2 = not test1 and remote_cksum[file_name] != local_cksum[file_name]
1467
      # existing and good
1468
      test3 = not test1 and remote_cksum[file_name] == local_cksum[file_name]
1469
      _ErrorIf(test1 and must_have, self.ENODEFILECHECK, node,
1470
               "file '%s' missing", file_name)
1471
      _ErrorIf(test2 and must_have, self.ENODEFILECHECK, node,
1472
               "file '%s' has wrong checksum", file_name)
1473
      # not candidate and this is not a must-have file
1474
      _ErrorIf(test2 and not must_have, self.ENODEFILECHECK, node,
1475
               "file '%s' should not exist on non master"
1476
               " candidates (and the file is outdated)", file_name)
1477
      # all good, except non-master/non-must have combination
1478
      _ErrorIf(test3 and not must_have, self.ENODEFILECHECK, node,
1479
               "file '%s' should not exist"
1480
               " on non master candidates", file_name)
1481

    
1482
  def _VerifyNodeDrbd(self, ninfo, nresult, instanceinfo, drbd_map):
1483
    """Verifies and the node DRBD status.
1484

1485
    @type ninfo: L{objects.Node}
1486
    @param ninfo: the node to check
1487
    @param nresult: the remote results for the node
1488
    @param instanceinfo: the dict of instances
1489
    @param drbd_map: the DRBD map as returned by
1490
        L{ganeti.config.ConfigWriter.ComputeDRBDMap}
1491

1492
    """
1493
    node = ninfo.name
1494
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1495

    
1496
    # compute the DRBD minors
1497
    node_drbd = {}
1498
    for minor, instance in drbd_map[node].items():
1499
      test = instance not in instanceinfo
1500
      _ErrorIf(test, self.ECLUSTERCFG, None,
1501
               "ghost instance '%s' in temporary DRBD map", instance)
1502
        # ghost instance should not be running, but otherwise we
1503
        # don't give double warnings (both ghost instance and
1504
        # unallocated minor in use)
1505
      if test:
1506
        node_drbd[minor] = (instance, False)
1507
      else:
1508
        instance = instanceinfo[instance]
1509
        node_drbd[minor] = (instance.name, instance.admin_up)
1510

    
1511
    # and now check them
1512
    used_minors = nresult.get(constants.NV_DRBDLIST, [])
1513
    test = not isinstance(used_minors, (tuple, list))
1514
    _ErrorIf(test, self.ENODEDRBD, node,
1515
             "cannot parse drbd status file: %s", str(used_minors))
1516
    if test:
1517
      # we cannot check drbd status
1518
      return
1519

    
1520
    for minor, (iname, must_exist) in node_drbd.items():
1521
      test = minor not in used_minors and must_exist
1522
      _ErrorIf(test, self.ENODEDRBD, node,
1523
               "drbd minor %d of instance %s is not active", minor, iname)
1524
    for minor in used_minors:
1525
      test = minor not in node_drbd
1526
      _ErrorIf(test, self.ENODEDRBD, node,
1527
               "unallocated drbd minor %d is in use", minor)
1528

    
1529
  def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
1530
    """Verifies and updates the node volume data.
1531

1532
    This function will update a L{NodeImage}'s internal structures
1533
    with data from the remote call.
1534

1535
    @type ninfo: L{objects.Node}
1536
    @param ninfo: the node to check
1537
    @param nresult: the remote results for the node
1538
    @param nimg: the node image object
1539
    @param vg_name: the configured VG name
1540

1541
    """
1542
    node = ninfo.name
1543
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1544

    
1545
    nimg.lvm_fail = True
1546
    lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
1547
    if vg_name is None:
1548
      pass
1549
    elif isinstance(lvdata, basestring):
1550
      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
1551
               utils.SafeEncode(lvdata))
1552
    elif not isinstance(lvdata, dict):
1553
      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
1554
    else:
1555
      nimg.volumes = lvdata
1556
      nimg.lvm_fail = False
1557

    
1558
  def _UpdateNodeInstances(self, ninfo, nresult, nimg):
1559
    """Verifies and updates the node instance list.
1560

1561
    If the listing was successful, then updates this node's instance
1562
    list. Otherwise, it marks the RPC call as failed for the instance
1563
    list key.
1564

1565
    @type ninfo: L{objects.Node}
1566
    @param ninfo: the node to check
1567
    @param nresult: the remote results for the node
1568
    @param nimg: the node image object
1569

1570
    """
1571
    idata = nresult.get(constants.NV_INSTANCELIST, None)
1572
    test = not isinstance(idata, list)
1573
    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
1574
                  " (instancelist): %s", utils.SafeEncode(str(idata)))
1575
    if test:
1576
      nimg.hyp_fail = True
1577
    else:
1578
      nimg.instances = idata
1579

    
1580
  def _UpdateNodeInfo(self, ninfo, nresult, nimg, vg_name):
1581
    """Verifies and computes a node information map
1582

1583
    @type ninfo: L{objects.Node}
1584
    @param ninfo: the node to check
1585
    @param nresult: the remote results for the node
1586
    @param nimg: the node image object
1587
    @param vg_name: the configured VG name
1588

1589
    """
1590
    node = ninfo.name
1591
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1592

    
1593
    # try to read free memory (from the hypervisor)
1594
    hv_info = nresult.get(constants.NV_HVINFO, None)
1595
    test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
1596
    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
1597
    if not test:
1598
      try:
1599
        nimg.mfree = int(hv_info["memory_free"])
1600
      except (ValueError, TypeError):
1601
        _ErrorIf(True, self.ENODERPC, node,
1602
                 "node returned invalid nodeinfo, check hypervisor")
1603

    
1604
    # FIXME: devise a free space model for file based instances as well
1605
    if vg_name is not None:
1606
      test = (constants.NV_VGLIST not in nresult or
1607
              vg_name not in nresult[constants.NV_VGLIST])
1608
      _ErrorIf(test, self.ENODELVM, node,
1609
               "node didn't return data for the volume group '%s'"
1610
               " - it is either missing or broken", vg_name)
1611
      if not test:
1612
        try:
1613
          nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
1614
        except (ValueError, TypeError):
1615
          _ErrorIf(True, self.ENODERPC, node,
1616
                   "node returned invalid LVM info, check LVM status")
1617

    
1618
  def CheckPrereq(self):
1619
    """Check prerequisites.
1620

1621
    Transform the list of checks we're going to skip into a set and check that
1622
    all its members are valid.
1623

1624
    """
1625
    self.skip_set = frozenset(self.op.skip_checks)
1626
    if not constants.VERIFY_OPTIONAL_CHECKS.issuperset(self.skip_set):
1627
      raise errors.OpPrereqError("Invalid checks to be skipped specified",
1628
                                 errors.ECODE_INVAL)
1629

    
1630
  def BuildHooksEnv(self):
1631
    """Build hooks env.
1632

1633
    Cluster-Verify hooks just ran in the post phase and their failure makes
1634
    the output be logged in the verify output and the verification to fail.
1635

1636
    """
1637
    all_nodes = self.cfg.GetNodeList()
1638
    env = {
1639
      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
1640
      }
1641
    for node in self.cfg.GetAllNodesInfo().values():
1642
      env["NODE_TAGS_%s" % node.name] = " ".join(node.GetTags())
1643

    
1644
    return env, [], all_nodes
1645

    
1646
  def Exec(self, feedback_fn):
1647
    """Verify integrity of cluster, performing various test on nodes.
1648

1649
    """
1650
    self.bad = False
1651
    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
1652
    verbose = self.op.verbose
1653
    self._feedback_fn = feedback_fn
1654
    feedback_fn("* Verifying global settings")
1655
    for msg in self.cfg.VerifyConfig():
1656
      _ErrorIf(True, self.ECLUSTERCFG, None, msg)
1657

    
1658
    # Check the cluster certificates
1659
    for cert_filename in constants.ALL_CERT_FILES:
1660
      (errcode, msg) = _VerifyCertificate(cert_filename)
1661
      _ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
1662

    
1663
    vg_name = self.cfg.GetVGName()
1664
    hypervisors = self.cfg.GetClusterInfo().enabled_hypervisors
1665
    nodelist = utils.NiceSort(self.cfg.GetNodeList())
1666
    nodeinfo = [self.cfg.GetNodeInfo(nname) for nname in nodelist]
1667
    instancelist = utils.NiceSort(self.cfg.GetInstanceList())
1668
    instanceinfo = dict((iname, self.cfg.GetInstanceInfo(iname))
1669
                        for iname in instancelist)
1670
    i_non_redundant = [] # Non redundant instances
1671
    i_non_a_balanced = [] # Non auto-balanced instances
1672
    n_offline = 0 # Count of offline nodes
1673
    n_drained = 0 # Count of nodes being drained
1674
    node_vol_should = {}
1675

    
1676
    # FIXME: verify OS list
1677
    # do local checksums
1678
    master_files = [constants.CLUSTER_CONF_FILE]
1679

    
1680
    file_names = ssconf.SimpleStore().GetFileList()
1681
    file_names.extend(constants.ALL_CERT_FILES)
1682
    file_names.extend(master_files)
1683

    
1684
    local_checksums = utils.FingerprintFiles(file_names)
1685

    
1686
    feedback_fn("* Gathering data (%d nodes)" % len(nodelist))
1687
    node_verify_param = {
1688
      constants.NV_FILELIST: file_names,
1689
      constants.NV_NODELIST: [node.name for node in nodeinfo
1690
                              if not node.offline],
1691
      constants.NV_HYPERVISOR: hypervisors,
1692
      constants.NV_NODENETTEST: [(node.name, node.primary_ip,
1693
                                  node.secondary_ip) for node in nodeinfo
1694
                                 if not node.offline],
1695
      constants.NV_INSTANCELIST: hypervisors,
1696
      constants.NV_VERSION: None,
1697
      constants.NV_HVINFO: self.cfg.GetHypervisorType(),
1698
      constants.NV_NODESETUP: None,
1699
      constants.NV_TIME: None,
1700
      }
1701

    
1702
    if vg_name is not None:
1703
      node_verify_param[constants.NV_VGLIST] = None
1704
      node_verify_param[constants.NV_LVLIST] = vg_name
1705
      node_verify_param[constants.NV_PVLIST] = [vg_name]
1706
      node_verify_param[constants.NV_DRBDLIST] = None
1707

    
1708
    # Build our expected cluster state
1709
    node_image = dict((node.name, self.NodeImage(offline=node.offline))
1710
                      for node in nodeinfo)
1711

    
1712
    for instance in instancelist:
1713
      inst_config = instanceinfo[instance]
1714

    
1715
      for nname in inst_config.all_nodes:
1716
        if nname not in node_image:
1717
          # ghost node
1718
          gnode = self.NodeImage()
1719
          gnode.ghost = True
1720
          node_image[nname] = gnode
1721

    
1722
      inst_config.MapLVsByNode(node_vol_should)
1723

    
1724
      pnode = inst_config.primary_node
1725
      node_image[pnode].pinst.append(instance)
1726

    
1727
      for snode in inst_config.secondary_nodes:
1728
        nimg = node_image[snode]
1729
        nimg.sinst.append(instance)
1730
        if pnode not in nimg.sbp:
1731
          nimg.sbp[pnode] = []
1732
        nimg.sbp[pnode].append(instance)
1733

    
1734
    # At this point, we have the in-memory data structures complete,
1735
    # except for the runtime information, which we'll gather next
1736

    
1737
    # Due to the way our RPC system works, exact response times cannot be
1738
    # guaranteed (e.g. a broken node could run into a timeout). By keeping the
1739
    # time before and after executing the request, we can at least have a time
1740
    # window.
1741
    nvinfo_starttime = time.time()
1742
    all_nvinfo = self.rpc.call_node_verify(nodelist, node_verify_param,
1743
                                           self.cfg.GetClusterName())
1744
    nvinfo_endtime = time.time()
1745

    
1746
    cluster = self.cfg.GetClusterInfo()
1747
    master_node = self.cfg.GetMasterNode()
1748
    all_drbd_map = self.cfg.ComputeDRBDMap()
1749

    
1750
    feedback_fn("* Verifying node status")
1751
    for node_i in nodeinfo:
1752
      node = node_i.name
1753
      nimg = node_image[node]
1754

    
1755
      if node_i.offline:
1756
        if verbose:
1757
          feedback_fn("* Skipping offline node %s" % (node,))
1758
        n_offline += 1
1759
        continue
1760

    
1761
      if node == master_node:
1762
        ntype = "master"
1763
      elif node_i.master_candidate:
1764
        ntype = "master candidate"
1765
      elif node_i.drained:
1766
        ntype = "drained"
1767
        n_drained += 1
1768
      else:
1769
        ntype = "regular"
1770
      if verbose:
1771
        feedback_fn("* Verifying node %s (%s)" % (node, ntype))
1772

    
1773
      msg = all_nvinfo[node].fail_msg
1774
      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
1775
      if msg:
1776
        nimg.rpc_fail = True
1777
        continue
1778

    
1779
      nresult = all_nvinfo[node].payload
1780

    
1781
      nimg.call_ok = self._VerifyNode(node_i, nresult)
1782
      self._VerifyNodeNetwork(node_i, nresult)
1783
      self._VerifyNodeLVM(node_i, nresult, vg_name)
1784
      self._VerifyNodeFiles(node_i, nresult, file_names, local_checksums,
1785
                            master_files)
1786
      self._VerifyNodeDrbd(node_i, nresult, instanceinfo, all_drbd_map)
1787
      self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
1788

    
1789
      self._UpdateNodeVolumes(node_i, nresult, nimg, vg_name)
1790
      self._UpdateNodeInstances(node_i, nresult, nimg)
1791
      self._UpdateNodeInfo(node_i, nresult, nimg, vg_name)
1792

    
1793
    feedback_fn("* Verifying instance status")
1794
    for instance in instancelist:
1795
      if verbose:
1796
        feedback_fn("* Verifying instance %s" % instance)
1797
      inst_config = instanceinfo[instance]
1798
      self._VerifyInstance(instance, inst_config, node_image)
1799
      inst_nodes_offline = []
1800

    
1801
      pnode = inst_config.primary_node
1802
      pnode_img = node_image[pnode]
1803
      _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
1804
               self.ENODERPC, pnode, "instance %s, connection to"
1805
               " primary node failed", instance)
1806

    
1807
      if pnode_img.offline:
1808
        inst_nodes_offline.append(pnode)
1809

    
1810
      # If the instance is non-redundant we cannot survive losing its primary
1811
      # node, so we are not N+1 compliant. On the other hand we have no disk
1812
      # templates with more than one secondary so that situation is not well
1813
      # supported either.
1814
      # FIXME: does not support file-backed instances
1815
      if not inst_config.secondary_nodes:
1816
        i_non_redundant.append(instance)
1817
      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
1818
               instance, "instance has multiple secondary nodes: %s",
1819
               utils.CommaJoin(inst_config.secondary_nodes),
1820
               code=self.ETYPE_WARNING)
1821

    
1822
      if not cluster.FillBE(inst_config)[constants.BE_AUTO_BALANCE]:
1823
        i_non_a_balanced.append(instance)
1824

    
1825
      for snode in inst_config.secondary_nodes:
1826
        s_img = node_image[snode]
1827
        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
1828
                 "instance %s, connection to secondary node failed", instance)
1829

    
1830
        if s_img.offline:
1831
          inst_nodes_offline.append(snode)
1832

    
1833
      # warn that the instance lives on offline nodes
1834
      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
1835
               "instance lives on offline node(s) %s",
1836
               utils.CommaJoin(inst_nodes_offline))
1837
      # ... or ghost nodes
1838
      for node in inst_config.all_nodes:
1839
        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
1840
                 "instance lives on ghost node %s", node)
1841

    
1842
    feedback_fn("* Verifying orphan volumes")
1843
    self._VerifyOrphanVolumes(node_vol_should, node_image)
1844

    
1845
    feedback_fn("* Verifying oprhan instances")
1846
    self._VerifyOrphanInstances(instancelist, node_image)
1847

    
1848
    if constants.VERIFY_NPLUSONE_MEM not in self.skip_set:
1849
      feedback_fn("* Verifying N+1 Memory redundancy")
1850
      self._VerifyNPlusOneMemory(node_image, instanceinfo)
1851

    
1852
    feedback_fn("* Other Notes")
1853
    if i_non_redundant:
1854
      feedback_fn("  - NOTICE: %d non-redundant instance(s) found."
1855
                  % len(i_non_redundant))
1856

    
1857
    if i_non_a_balanced:
1858
      feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
1859
                  % len(i_non_a_balanced))
1860

    
1861
    if n_offline:
1862
      feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
1863

    
1864
    if n_drained:
1865
      feedback_fn("  - NOTICE: %d drained node(s) found." % n_drained)
1866

    
1867
    return not self.bad
1868

    
1869
  def HooksCallBack(self, phase, hooks_results, feedback_fn, lu_result):
1870
    """Analyze the post-hooks' result
1871

1872
    This method analyses the hook result, handles it, and sends some
1873
    nicely-formatted feedback back to the user.
1874

1875
    @param phase: one of L{constants.HOOKS_PHASE_POST} or
1876
        L{constants.HOOKS_PHASE_PRE}; it denotes the hooks phase
1877
    @param hooks_results: the results of the multi-node hooks rpc call
1878
    @param feedback_fn: function used send feedback back to the caller
1879
    @param lu_result: previous Exec result
1880
    @return: the new Exec result, based on the previous result
1881
        and hook results
1882

1883
    """
1884
    # We only really run POST phase hooks, and are only interested in
1885
    # their results
1886
    if phase == constants.HOOKS_PHASE_POST:
1887
      # Used to change hooks' output to proper indentation
1888
      indent_re = re.compile('^', re.M)
1889
      feedback_fn("* Hooks Results")
1890
      assert hooks_results, "invalid result from hooks"
1891

    
1892
      for node_name in hooks_results:
1893
        res = hooks_results[node_name]
1894
        msg = res.fail_msg
1895
        test = msg and not res.offline
1896
        self._ErrorIf(test, self.ENODEHOOKS, node_name,
1897
                      "Communication failure in hooks execution: %s", msg)
1898
        if res.offline or msg:
1899
          # No need to investigate payload if node is offline or gave an error.
1900
          # override manually lu_result here as _ErrorIf only
1901
          # overrides self.bad
1902
          lu_result = 1
1903
          continue
1904
        for script, hkr, output in res.payload:
1905
          test = hkr == constants.HKR_FAIL
1906
          self._ErrorIf(test, self.ENODEHOOKS, node_name,
1907
                        "Script %s failed, output:", script)
1908
          if test:
1909
            output = indent_re.sub('      ', output)
1910
            feedback_fn("%s" % output)
1911
            lu_result = 0
1912

    
1913
      return lu_result
1914

    
1915

    
1916
class LUVerifyDisks(NoHooksLU):
1917
  """Verifies the cluster disks status.
1918

1919
  """
1920
  _OP_REQP = []
1921
  REQ_BGL = False
1922

    
1923
  def ExpandNames(self):
1924
    self.needed_locks = {
1925
      locking.LEVEL_NODE: locking.ALL_SET,
1926
      locking.LEVEL_INSTANCE: locking.ALL_SET,
1927
    }
1928
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
1929

    
1930
  def CheckPrereq(self):
1931
    """Check prerequisites.
1932

1933
    This has no prerequisites.
1934

1935
    """
1936
    pass
1937

    
1938
  def Exec(self, feedback_fn):
1939
    """Verify integrity of cluster disks.
1940

1941
    @rtype: tuple of three items
1942
    @return: a tuple of (dict of node-to-node_error, list of instances
1943
        which need activate-disks, dict of instance: (node, volume) for
1944
        missing volumes
1945

1946
    """
1947
    result = res_nodes, res_instances, res_missing = {}, [], {}
1948

    
1949
    vg_name = self.cfg.GetVGName()
1950
    nodes = utils.NiceSort(self.cfg.GetNodeList())
1951
    instances = [self.cfg.GetInstanceInfo(name)
1952
                 for name in self.cfg.GetInstanceList()]
1953

    
1954
    nv_dict = {}
1955
    for inst in instances:
1956
      inst_lvs = {}
1957
      if (not inst.admin_up or
1958
          inst.disk_template not in constants.DTS_NET_MIRROR):
1959
        continue
1960
      inst.MapLVsByNode(inst_lvs)
1961
      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
1962
      for node, vol_list in inst_lvs.iteritems():
1963
        for vol in vol_list:
1964
          nv_dict[(node, vol)] = inst
1965

    
1966
    if not nv_dict:
1967
      return result
1968

    
1969
    node_lvs = self.rpc.call_lv_list(nodes, vg_name)
1970

    
1971
    for node in nodes:
1972
      # node_volume
1973
      node_res = node_lvs[node]
1974
      if node_res.offline:
1975
        continue
1976
      msg = node_res.fail_msg
1977
      if msg:
1978
        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
1979
        res_nodes[node] = msg
1980
        continue
1981

    
1982
      lvs = node_res.payload
1983
      for lv_name, (_, _, lv_online) in lvs.items():
1984
        inst = nv_dict.pop((node, lv_name), None)
1985
        if (not lv_online and inst is not None
1986
            and inst.name not in res_instances):
1987
          res_instances.append(inst.name)
1988

    
1989
    # any leftover items in nv_dict are missing LVs, let's arrange the
1990
    # data better
1991
    for key, inst in nv_dict.iteritems():
1992
      if inst.name not in res_missing:
1993
        res_missing[inst.name] = []
1994
      res_missing[inst.name].append(key)
1995

    
1996
    return result
1997

    
1998

    
1999
class LURepairDiskSizes(NoHooksLU):
2000
  """Verifies the cluster disks sizes.
2001

2002
  """
2003
  _OP_REQP = ["instances"]
2004
  REQ_BGL = False
2005

    
2006
  def ExpandNames(self):
2007
    if not isinstance(self.op.instances, list):
2008
      raise errors.OpPrereqError("Invalid argument type 'instances'",
2009
                                 errors.ECODE_INVAL)
2010

    
2011
    if self.op.instances:
2012
      self.wanted_names = []
2013
      for name in self.op.instances:
2014
        full_name = _ExpandInstanceName(self.cfg, name)
2015
        self.wanted_names.append(full_name)
2016
      self.needed_locks = {
2017
        locking.LEVEL_NODE: [],
2018
        locking.LEVEL_INSTANCE: self.wanted_names,
2019
        }
2020
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
2021
    else:
2022
      self.wanted_names = None
2023
      self.needed_locks = {
2024
        locking.LEVEL_NODE: locking.ALL_SET,
2025
        locking.LEVEL_INSTANCE: locking.ALL_SET,
2026
        }
2027
    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
2028

    
2029
  def DeclareLocks(self, level):
2030
    if level == locking.LEVEL_NODE and self.wanted_names is not None:
2031
      self._LockInstancesNodes(primary_only=True)
2032

    
2033
  def CheckPrereq(self):
2034
    """Check prerequisites.
2035

2036
    This only checks the optional instance list against the existing names.
2037

2038
    """
2039
    if self.wanted_names is None:
2040
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
2041

    
2042
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
2043
                             in self.wanted_names]
2044

    
2045
  def _EnsureChildSizes(self, disk):
2046
    """Ensure children of the disk have the needed disk size.
2047

2048
    This is valid mainly for DRBD8 and fixes an issue where the
2049
    children have smaller disk size.
2050

2051
    @param disk: an L{ganeti.objects.Disk} object
2052

2053
    """
2054
    if disk.dev_type == constants.LD_DRBD8:
2055
      assert disk.children, "Empty children for DRBD8?"
2056
      fchild = disk.children[0]
2057
      mismatch = fchild.size < disk.size
2058
      if mismatch:
2059
        self.LogInfo("Child disk has size %d, parent %d, fixing",
2060
                     fchild.size, disk.size)
2061
        fchild.size = disk.size
2062

    
2063
      # and we recurse on this child only, not on the metadev
2064
      return self._EnsureChildSizes(fchild) or mismatch
2065
    else:
2066
      return False
2067

    
2068
  def Exec(self, feedback_fn):
2069
    """Verify the size of cluster disks.
2070

2071
    """
2072
    # TODO: check child disks too
2073
    # TODO: check differences in size between primary/secondary nodes
2074
    per_node_disks = {}
2075
    for instance in self.wanted_instances:
2076
      pnode = instance.primary_node
2077
      if pnode not in per_node_disks:
2078
        per_node_disks[pnode] = []
2079
      for idx, disk in enumerate(instance.disks):
2080
        per_node_disks[pnode].append((instance, idx, disk))
2081

    
2082
    changed = []
2083
    for node, dskl in per_node_disks.items():
2084
      newl = [v[2].Copy() for v in dskl]
2085
      for dsk in newl:
2086
        self.cfg.SetDiskID(dsk, node)
2087
      result = self.rpc.call_blockdev_getsizes(node, newl)
2088
      if result.fail_msg:
2089
        self.LogWarning("Failure in blockdev_getsizes call to node"
2090
                        " %s, ignoring", node)
2091
        continue
2092
      if len(result.data) != len(dskl):
2093
        self.LogWarning("Invalid result from node %s, ignoring node results",
2094
                        node)
2095
        continue
2096
      for ((instance, idx, disk), size) in zip(dskl, result.data):
2097
        if size is None:
2098
          self.LogWarning("Disk %d of instance %s did not return size"
2099
                          " information, ignoring", idx, instance.name)
2100
          continue
2101
        if not isinstance(size, (int, long)):
2102
          self.LogWarning("Disk %d of instance %s did not return valid"
2103
                          " size information, ignoring", idx, instance.name)
2104
          continue
2105
        size = size >> 20
2106
        if size != disk.size:
2107
          self.LogInfo("Disk %d of instance %s has mismatched size,"
2108
                       " correcting: recorded %d, actual %d", idx,
2109
                       instance.name, disk.size, size)
2110
          disk.size = size
2111
          self.cfg.Update(instance, feedback_fn)
2112
          changed.append((instance.name, idx, size))
2113
        if self._EnsureChildSizes(disk):
2114
          self.cfg.Update(instance, feedback_fn)
2115
          changed.append((instance.name, idx, disk.size))
2116
    return changed
2117

    
2118

    
2119
class LURenameCluster(LogicalUnit):
2120
  """Rename the cluster.
2121

2122
  """
2123
  HPATH = "cluster-rename"
2124
  HTYPE = constants.HTYPE_CLUSTER
2125
  _OP_REQP = ["name"]
2126

    
2127
  def BuildHooksEnv(self):
2128
    """Build hooks env.
2129

2130
    """
2131
    env = {
2132
      "OP_TARGET": self.cfg.GetClusterName(),
2133
      "NEW_NAME": self.op.name,
2134
      }
2135
    mn = self.cfg.GetMasterNode()
2136
    all_nodes = self.cfg.GetNodeList()
2137
    return env, [mn], all_nodes
2138

    
2139
  def CheckPrereq(self):
2140
    """Verify that the passed name is a valid one.
2141

2142
    """
2143
    hostname = utils.GetHostInfo(self.op.name)
2144

    
2145
    new_name = hostname.name
2146
    self.ip = new_ip = hostname.ip
2147
    old_name = self.cfg.GetClusterName()
2148
    old_ip = self.cfg.GetMasterIP()
2149
    if new_name == old_name and new_ip == old_ip:
2150
      raise errors.OpPrereqError("Neither the name nor the IP address of the"
2151
                                 " cluster has changed",
2152
                                 errors.ECODE_INVAL)
2153
    if new_ip != old_ip:
2154
      if utils.TcpPing(new_ip, constants.DEFAULT_NODED_PORT):
2155
        raise errors.OpPrereqError("The given cluster IP address (%s) is"
2156
                                   " reachable on the network. Aborting." %
2157
                                   new_ip, errors.ECODE_NOTUNIQUE)
2158

    
2159
    self.op.name = new_name
2160

    
2161
  def Exec(self, feedback_fn):
2162
    """Rename the cluster.
2163

2164
    """
2165
    clustername = self.op.name
2166
    ip = self.ip
2167

    
2168
    # shutdown the master IP
2169
    master = self.cfg.GetMasterNode()
2170
    result = self.rpc.call_node_stop_master(master, False)
2171
    result.Raise("Could not disable the master role")
2172

    
2173
    try:
2174
      cluster = self.cfg.GetClusterInfo()
2175
      cluster.cluster_name = clustername
2176
      cluster.master_ip = ip
2177
      self.cfg.Update(cluster, feedback_fn)
2178

    
2179
      # update the known hosts file
2180
      ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
2181
      node_list = self.cfg.GetNodeList()
2182
      try:
2183
        node_list.remove(master)
2184
      except ValueError:
2185
        pass
2186
      result = self.rpc.call_upload_file(node_list,
2187
                                         constants.SSH_KNOWN_HOSTS_FILE)
2188
      for to_node, to_result in result.iteritems():
2189
        msg = to_result.fail_msg
2190
        if msg:
2191
          msg = ("Copy of file %s to node %s failed: %s" %
2192
                 (constants.SSH_KNOWN_HOSTS_FILE, to_node, msg))
2193
          self.proc.LogWarning(msg)
2194

    
2195
    finally:
2196
      result = self.rpc.call_node_start_master(master, False, False)
2197
      msg = result.fail_msg
2198
      if msg:
2199
        self.LogWarning("Could not re-enable the master role on"
2200
                        " the master, please restart manually: %s", msg)
2201

    
2202

    
2203
def _RecursiveCheckIfLVMBased(disk):
2204
  """Check if the given disk or its children are lvm-based.
2205

2206
  @type disk: L{objects.Disk}
2207
  @param disk: the disk to check
2208
  @rtype: boolean
2209
  @return: boolean indicating whether a LD_LV dev_type was found or not
2210

2211
  """
2212
  if disk.children:
2213
    for chdisk in disk.children:
2214
      if _RecursiveCheckIfLVMBased(chdisk):
2215
        return True
2216
  return disk.dev_type == constants.LD_LV
2217

    
2218

    
2219
class LUSetClusterParams(LogicalUnit):
2220
  """Change the parameters of the cluster.
2221

2222
  """
2223
  HPATH = "cluster-modify"
2224
  HTYPE = constants.HTYPE_CLUSTER
2225
  _OP_REQP = []
2226
  REQ_BGL = False
2227

    
2228
  def CheckArguments(self):
2229
    """Check parameters
2230

2231
    """
2232
    if not hasattr(self.op, "candidate_pool_size"):
2233
      self.op.candidate_pool_size = None
2234
    if self.op.candidate_pool_size is not None:
2235
      try:
2236
        self.op.candidate_pool_size = int(self.op.candidate_pool_size)
2237
      except (ValueError, TypeError), err:
2238
        raise errors.OpPrereqError("Invalid candidate_pool_size value: %s" %
2239
                                   str(err), errors.ECODE_INVAL)
2240
      if self.op.candidate_pool_size < 1:
2241
        raise errors.OpPrereqError("At least one master candidate needed",
2242
                                   errors.ECODE_INVAL)
2243
    _CheckBooleanOpField(self.op, "maintain_node_health")
2244

    
2245
  def ExpandNames(self):
2246
    # FIXME: in the future maybe other cluster params won't require checking on
2247
    # all nodes to be modified.
2248
    self.needed_locks = {
2249
      locking.LEVEL_NODE: locking.ALL_SET,
2250
    }
2251
    self.share_locks[locking.LEVEL_NODE] = 1
2252

    
2253
  def BuildHooksEnv(self):
2254
    """Build hooks env.
2255

2256
    """
2257
    env = {
2258
      "OP_TARGET": self.cfg.GetClusterName(),
2259
      "NEW_VG_NAME": self.op.vg_name,
2260
      }
2261
    mn = self.cfg.GetMasterNode()
2262
    return env, [mn], [mn]
2263

    
2264
  def CheckPrereq(self):
2265
    """Check prerequisites.
2266

2267
    This checks whether the given params don't conflict and
2268
    if the given volume group is valid.
2269

2270
    """
2271
    if self.op.vg_name is not None and not self.op.vg_name:
2272
      instances = self.cfg.GetAllInstancesInfo().values()
2273
      for inst in instances:
2274
        for disk in inst.disks:
2275
          if _RecursiveCheckIfLVMBased(disk):
2276
            raise errors.OpPrereqError("Cannot disable lvm storage while"
2277
                                       " lvm-based instances exist",
2278
                                       errors.ECODE_INVAL)
2279

    
2280
    node_list = self.acquired_locks[locking.LEVEL_NODE]
2281

    
2282
    # if vg_name not None, checks given volume group on all nodes
2283
    if self.op.vg_name:
2284
      vglist = self.rpc.call_vg_list(node_list)
2285
      for node in node_list:
2286
        msg = vglist[node].fail_msg
2287
        if msg:
2288
          # ignoring down node
2289
          self.LogWarning("Error while gathering data on node %s"
2290
                          " (ignoring node): %s", node, msg)
2291
          continue
2292
        vgstatus = utils.CheckVolumeGroupSize(vglist[node].payload,
2293
                                              self.op.vg_name,
2294
                                              constants.MIN_VG_SIZE)
2295
        if vgstatus:
2296
          raise errors.OpPrereqError("Error on node '%s': %s" %
2297
                                     (node, vgstatus), errors.ECODE_ENVIRON)
2298

    
2299
    self.cluster = cluster = self.cfg.GetClusterInfo()
2300
    # validate params changes
2301
    if self.op.beparams:
2302
      utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
2303
      self.new_beparams = objects.FillDict(
2304
        cluster.beparams[constants.PP_DEFAULT], self.op.beparams)
2305

    
2306
    if self.op.nicparams:
2307
      utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
2308
      self.new_nicparams = objects.FillDict(
2309
        cluster.nicparams[constants.PP_DEFAULT], self.op.nicparams)
2310
      objects.NIC.CheckParameterSyntax(self.new_nicparams)
2311
      nic_errors = []
2312

    
2313
      # check all instances for consistency
2314
      for instance in self.cfg.GetAllInstancesInfo().values():
2315
        for nic_idx, nic in enumerate(instance.nics):
2316
          params_copy = copy.deepcopy(nic.nicparams)
2317
          params_filled = objects.FillDict(self.new_nicparams, params_copy)
2318

    
2319
          # check parameter syntax
2320
          try:
2321
            objects.NIC.CheckParameterSyntax(params_filled)
2322
          except errors.ConfigurationError, err:
2323
            nic_errors.append("Instance %s, nic/%d: %s" %
2324
                              (instance.name, nic_idx, err))
2325

    
2326
          # if we're moving instances to routed, check that they have an ip
2327
          target_mode = params_filled[constants.NIC_MODE]
2328
          if target_mode == constants.NIC_MODE_ROUTED and not nic.ip:
2329
            nic_errors.append("Instance %s, nic/%d: routed nick with no ip" %
2330
                              (instance.name, nic_idx))
2331
      if nic_errors:
2332
        raise errors.OpPrereqError("Cannot apply the change, errors:\n%s" %
2333
                                   "\n".join(nic_errors))
2334

    
2335
    # hypervisor list/parameters
2336
    self.new_hvparams = objects.FillDict(cluster.hvparams, {})
2337
    if self.op.hvparams:
2338
      if not isinstance(self.op.hvparams, dict):
2339
        raise errors.OpPrereqError("Invalid 'hvparams' parameter on input",
2340
                                   errors.ECODE_INVAL)
2341
      for hv_name, hv_dict in self.op.hvparams.items():
2342
        if hv_name not in self.new_hvparams:
2343
          self.new_hvparams[hv_name] = hv_dict
2344
        else:
2345
          self.new_hvparams[hv_name].update(hv_dict)
2346

    
2347
    # os hypervisor parameters
2348
    self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
2349
    if self.op.os_hvp:
2350
      if not isinstance(self.op.os_hvp, dict):
2351
        raise errors.OpPrereqError("Invalid 'os_hvp' parameter on input",
2352
                                   errors.ECODE_INVAL)
2353
      for os_name, hvs in self.op.os_hvp.items():
2354
        if not isinstance(hvs, dict):
2355
          raise errors.OpPrereqError(("Invalid 'os_hvp' parameter on"
2356
                                      " input"), errors.ECODE_INVAL)
2357
        if os_name not in self.new_os_hvp:
2358
          self.new_os_hvp[os_name] = hvs
2359
        else:
2360
          for hv_name, hv_dict in hvs.items():
2361
            if hv_name not in self.new_os_hvp[os_name]:
2362
              self.new_os_hvp[os_name][hv_name] = hv_dict
2363
            else:
2364
              self.new_os_hvp[os_name][hv_name].update(hv_dict)
2365

    
2366
    if self.op.enabled_hypervisors is not None:
2367
      self.hv_list = self.op.enabled_hypervisors
2368
      if not self.hv_list:
2369
        raise errors.OpPrereqError("Enabled hypervisors list must contain at"
2370
                                   " least one member",
2371
                                   errors.ECODE_INVAL)
2372
      invalid_hvs = set(self.hv_list) - constants.HYPER_TYPES
2373
      if invalid_hvs:
2374
        raise errors.OpPrereqError("Enabled hypervisors contains invalid"
2375
                                   " entries: %s" %
2376
                                   utils.CommaJoin(invalid_hvs),
2377
                                   errors.ECODE_INVAL)
2378
    else:
2379
      self.hv_list = cluster.enabled_hypervisors
2380

    
2381
    if self.op.hvparams or self.op.enabled_hypervisors is not None:
2382
      # either the enabled list has changed, or the parameters have, validate
2383
      for hv_name, hv_params in self.new_hvparams.items():
2384
        if ((self.op.hvparams and hv_name in self.op.hvparams) or
2385
            (self.op.enabled_hypervisors and
2386
             hv_name in self.op.enabled_hypervisors)):
2387
          # either this is a new hypervisor, or its parameters have changed
2388
          hv_class = hypervisor.GetHypervisor(hv_name)
2389
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2390
          hv_class.CheckParameterSyntax(hv_params)
2391
          _CheckHVParams(self, node_list, hv_name, hv_params)
2392

    
2393
    if self.op.os_hvp:
2394
      # no need to check any newly-enabled hypervisors, since the
2395
      # defaults have already been checked in the above code-block
2396
      for os_name, os_hvp in self.new_os_hvp.items():
2397
        for hv_name, hv_params in os_hvp.items():
2398
          utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
2399
          # we need to fill in the new os_hvp on top of the actual hv_p
2400
          cluster_defaults = self.new_hvparams.get(hv_name, {})
2401
          new_osp = objects.FillDict(cluster_defaults, hv_params)
2402
          hv_class = hypervisor.GetHypervisor(hv_name)
2403
          hv_class.CheckParameterSyntax(new_osp)
2404
          _CheckHVParams(self, node_list, hv_name, new_osp)
2405

    
2406

    
2407
  def Exec(self, feedback_fn):
2408
    """Change the parameters of the cluster.
2409

2410
    """
2411
    if self.op.vg_name is not None:
2412
      new_volume = self.op.vg_name
2413
      if not new_volume:
2414
        new_volume = None
2415
      if new_volume != self.cfg.GetVGName():
2416
        self.cfg.SetVGName(new_volume)
2417
      else:
2418
        feedback_fn("Cluster LVM configuration already in desired"
2419
                    " state, not changing")
2420
    if self.op.hvparams:
2421
      self.cluster.hvparams = self.new_hvparams
2422
    if self.op.os_hvp:
2423
      self.cluster.os_hvp = self.new_os_hvp
2424
    if self.op.enabled_hypervisors is not None:
2425
      self.cluster.enabled_hypervisors = self.op.enabled_hypervisors
2426
    if self.op.beparams:
2427
      self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
2428
    if self.op.nicparams:
2429
      self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
2430

    
2431
    if self.op.candidate_pool_size is not None:
2432
      self.cluster.candidate_pool_size = self.op.candidate_pool_size
2433
      # we need to update the pool size here, otherwise the save will fail
2434
      _AdjustCandidatePool(self, [])
2435

    
2436
    if self.op.maintain_node_health is not None:
2437
      self.cluster.maintain_node_health = self.op.maintain_node_health
2438

    
2439
    self.cfg.Update(self.cluster, feedback_fn)
2440

    
2441

    
2442
def _RedistributeAncillaryFiles(lu, additional_nodes=None):
2443
  """Distribute additional files which are part of the cluster configuration.
2444

2445
  ConfigWriter takes care of distributing the config and ssconf files, but
2446
  there are more files which should be distributed to all nodes. This function
2447
  makes sure those are copied.
2448

2449
  @param lu: calling logical unit
2450
  @param additional_nodes: list of nodes not in the config to distribute to
2451

2452
  """
2453
  # 1. Gather target nodes
2454
  myself = lu.cfg.GetNodeInfo(lu.cfg.GetMasterNode())
2455
  dist_nodes = lu.cfg.GetOnlineNodeList()
2456
  if additional_nodes is not None:
2457
    dist_nodes.extend(additional_nodes)
2458
  if myself.name in dist_nodes:
2459
    dist_nodes.remove(myself.name)
2460

    
2461
  # 2. Gather files to distribute
2462
  dist_files = set([constants.ETC_HOSTS,
2463
                    constants.SSH_KNOWN_HOSTS_FILE,
2464
                    constants.RAPI_CERT_FILE,
2465
                    constants.RAPI_USERS_FILE,
2466
                    constants.CONFD_HMAC_KEY,
2467
                   ])
2468

    
2469
  enabled_hypervisors = lu.cfg.GetClusterInfo().enabled_hypervisors
2470
  for hv_name in enabled_hypervisors:
2471
    hv_class = hypervisor.GetHypervisor(hv_name)
2472
    dist_files.update(hv_class.GetAncillaryFiles())
2473

    
2474
  # 3. Perform the files upload
2475
  for fname in dist_files:
2476
    if os.path.exists(fname):
2477
      result = lu.rpc.call_upload_file(dist_nodes, fname)
2478
      for to_node, to_result in result.items():
2479
        msg = to_result.fail_msg
2480
        if msg:
2481
          msg = ("Copy of file %s to node %s failed: %s" %
2482
                 (fname, to_node, msg))
2483
          lu.proc.LogWarning(msg)
2484

    
2485

    
2486
class LURedistributeConfig(NoHooksLU):
2487
  """Force the redistribution of cluster configuration.
2488

2489
  This is a very simple LU.
2490

2491
  """
2492
  _OP_REQP = []
2493
  REQ_BGL = False
2494

    
2495
  def ExpandNames(self):
2496
    self.needed_locks = {
2497
      locking.LEVEL_NODE: locking.ALL_SET,
2498
    }
2499
    self.share_locks[locking.LEVEL_NODE] = 1
2500

    
2501
  def CheckPrereq(self):
2502
    """Check prerequisites.
2503

2504
    """
2505

    
2506
  def Exec(self, feedback_fn):
2507
    """Redistribute the configuration.
2508

2509
    """
2510
    self.cfg.Update(self.cfg.GetClusterInfo(), feedback_fn)
2511
    _RedistributeAncillaryFiles(self)
2512

    
2513

    
2514
def _WaitForSync(lu, instance, oneshot=False):
2515
  """Sleep and poll for an instance's disk to sync.
2516

2517
  """
2518
  if not instance.disks:
2519
    return True
2520

    
2521
  if not oneshot:
2522
    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
2523

    
2524
  node = instance.primary_node
2525

    
2526
  for dev in instance.disks:
2527
    lu.cfg.SetDiskID(dev, node)
2528

    
2529
  # TODO: Convert to utils.Retry
2530

    
2531
  retries = 0
2532
  degr_retries = 10 # in seconds, as we sleep 1 second each time
2533
  while True:
2534
    max_time = 0
2535
    done = True
2536
    cumul_degraded = False
2537
    rstats = lu.rpc.call_blockdev_getmirrorstatus(node, instance.disks)
2538
    msg = rstats.fail_msg
2539
    if msg:
2540
      lu.LogWarning("Can't get any data from node %s: %s", node, msg)
2541
      retries += 1
2542
      if retries >= 10:
2543
        raise errors.RemoteError("Can't contact node %s for mirror data,"
2544
                                 " aborting." % node)
2545
      time.sleep(6)
2546
      continue
2547
    rstats = rstats.payload
2548
    retries = 0
2549
    for i, mstat in enumerate(rstats):
2550
      if mstat is None:
2551
        lu.LogWarning("Can't compute data for node %s/%s",
2552
                           node, instance.disks[i].iv_name)
2553
        continue
2554

    
2555
      cumul_degraded = (cumul_degraded or
2556
                        (mstat.is_degraded and mstat.sync_percent is None))
2557
      if mstat.sync_percent is not None:
2558
        done = False
2559
        if mstat.estimated_time is not None:
2560
          rem_time = "%d estimated seconds remaining" % mstat.estimated_time
2561
          max_time = mstat.estimated_time
2562
        else:
2563
          rem_time = "no time estimate"
2564
        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
2565
                        (instance.disks[i].iv_name, mstat.sync_percent,
2566
                         rem_time))
2567

    
2568
    # if we're done but degraded, let's do a few small retries, to
2569
    # make sure we see a stable and not transient situation; therefore
2570
    # we force restart of the loop
2571
    if (done or oneshot) and cumul_degraded and degr_retries > 0:
2572
      logging.info("Degraded disks found, %d retries left", degr_retries)
2573
      degr_retries -= 1
2574
      time.sleep(1)
2575
      continue
2576

    
2577
    if done or oneshot:
2578
      break
2579

    
2580
    time.sleep(min(60, max_time))
2581

    
2582
  if done:
2583
    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
2584
  return not cumul_degraded
2585

    
2586

    
2587
def _CheckDiskConsistency(lu, dev, node, on_primary, ldisk=False):
2588
  """Check that mirrors are not degraded.
2589

2590
  The ldisk parameter, if True, will change the test from the
2591
  is_degraded attribute (which represents overall non-ok status for
2592
  the device(s)) to the ldisk (representing the local storage status).
2593

2594
  """
2595
  lu.cfg.SetDiskID(dev, node)
2596

    
2597
  result = True
2598

    
2599
  if on_primary or dev.AssembleOnSecondary():
2600
    rstats = lu.rpc.call_blockdev_find(node, dev)
2601
    msg = rstats.fail_msg
2602
    if msg:
2603
      lu.LogWarning("Can't find disk on node %s: %s", node, msg)
2604
      result = False
2605
    elif not rstats.payload:
2606
      lu.LogWarning("Can't find disk on node %s", node)
2607
      result = False
2608
    else:
2609
      if ldisk:
2610
        result = result and rstats.payload.ldisk_status == constants.LDS_OKAY
2611
      else:
2612
        result = result and not rstats.payload.is_degraded
2613

    
2614
  if dev.children:
2615
    for child in dev.children:
2616
      result = result and _CheckDiskConsistency(lu, child, node, on_primary)
2617

    
2618
  return result
2619

    
2620

    
2621
class LUDiagnoseOS(NoHooksLU):
2622
  """Logical unit for OS diagnose/query.
2623

2624
  """
2625
  _OP_REQP = ["output_fields", "names"]
2626
  REQ_BGL = False
2627
  _FIELDS_STATIC = utils.FieldSet()
2628
  _FIELDS_DYNAMIC = utils.FieldSet("name", "valid", "node_status", "variants")
2629
  # Fields that need calculation of global os validity
2630
  _FIELDS_NEEDVALID = frozenset(["valid", "variants"])
2631

    
2632
  def ExpandNames(self):
2633
    if self.op.names:
2634
      raise errors.OpPrereqError("Selective OS query not supported",
2635
                                 errors.ECODE_INVAL)
2636

    
2637
    _CheckOutputFields(static=self._FIELDS_STATIC,
2638
                       dynamic=self._FIELDS_DYNAMIC,
2639
                       selected=self.op.output_fields)
2640

    
2641
    # Lock all nodes, in shared mode
2642
    # Temporary removal of locks, should be reverted later
2643
    # TODO: reintroduce locks when they are lighter-weight
2644
    self.needed_locks = {}
2645
    #self.share_locks[locking.LEVEL_NODE] = 1
2646
    #self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
2647

    
2648
  def CheckPrereq(self):
2649
    """Check prerequisites.
2650

2651
    """
2652

    
2653
  @staticmethod
2654
  def _DiagnoseByOS(rlist):
2655
    """Remaps a per-node return list into an a per-os per-node dictionary
2656

2657
    @param rlist: a map with node names as keys and OS objects as values
2658

2659
    @rtype: dict
2660
    @return: a dictionary with osnames as keys and as value another map, with
2661
        nodes as keys and tuples of (path, status, diagnose) as values, eg::
2662

2663
          {"debian-etch": {"node1": [(/usr/lib/..., True, ""),
2664
                                     (/srv/..., False, "invalid api")],
2665
                           "node2": [(/srv/..., True, "")]}
2666
          }
2667

2668
    """
2669
    all_os = {}
2670
    # we build here the list of nodes that didn't fail the RPC (at RPC
2671
    # level), so that nodes with a non-responding node daemon don't
2672
    # make all OSes invalid
2673
    good_nodes = [node_name for node_name in rlist
2674
                  if not rlist[node_name].fail_msg]
2675
    for node_name, nr in rlist.items():
2676
      if nr.fail_msg or not nr.payload:
2677
        continue
2678
      for name, path, status, diagnose, variants in nr.payload:
2679
        if name not in all_os:
2680
          # build a list of nodes for this os containing empty lists
2681
          # for each node in node_list
2682
          all_os[name] = {}
2683
          for nname in good_nodes:
2684
            all_os[name][nname] = []
2685
        all_os[name][node_name].append((path, status, diagnose, variants))
2686
    return all_os
2687

    
2688
  def Exec(self, feedback_fn):
2689
    """Compute the list of OSes.
2690

2691
    """
2692
    valid_nodes = [node for node in self.cfg.GetOnlineNodeList()]
2693
    node_data = self.rpc.call_os_diagnose(valid_nodes)
2694
    pol = self._DiagnoseByOS(node_data)
2695
    output = []
2696
    calc_valid = self._FIELDS_NEEDVALID.intersection(self.op.output_fields)
2697
    calc_variants = "variants" in self.op.output_fields
2698

    
2699
    for os_name, os_data in pol.items():
2700
      row = []
2701
      if calc_valid:
2702
        valid = True
2703
        variants = None
2704
        for osl in os_data.values():
2705
          valid = valid and osl and osl[0][1]
2706
          if not valid:
2707
            variants = None
2708
            break
2709
          if calc_variants:
2710
            node_variants = osl[0][3]
2711
            if variants is None:
2712
              variants = node_variants
2713
            else:
2714
              variants = [v for v in variants if v in node_variants]
2715

    
2716
      for field in self.op.output_fields:
2717
        if field == "name":
2718
          val = os_name
2719
        elif field == "valid":
2720
          val = valid
2721
        elif field == "node_status":
2722
          # this is just a copy of the dict
2723
          val = {}
2724
          for node_name, nos_list in os_data.items():
2725
            val[node_name] = nos_list
2726
        elif field == "variants":
2727
          val =  variants
2728
        else:
2729
          raise errors.ParameterError(field)
2730
        row.append(val)
2731
      output.append(row)
2732

    
2733
    return output
2734

    
2735

    
2736
class LURemoveNode(LogicalUnit):
2737
  """Logical unit for removing a node.
2738

2739
  """
2740
  HPATH = "node-remove"
2741
  HTYPE = constants.HTYPE_NODE
2742
  _OP_REQP = ["node_name"]
2743

    
2744
  def BuildHooksEnv(self):
2745
    """Build hooks env.
2746

2747
    This doesn't run on the target node in the pre phase as a failed
2748
    node would then be impossible to remove.
2749

2750
    """
2751
    env = {
2752
      "OP_TARGET": self.op.node_name,
2753
      "NODE_NAME": self.op.node_name,
2754
      }
2755
    all_nodes = self.cfg.GetNodeList()
2756
    try:
2757
      all_nodes.remove(self.op.node_name)
2758
    except ValueError:
2759
      logging.warning("Node %s which is about to be removed not found"
2760
                      " in the all nodes list", self.op.node_name)
2761
    return env, all_nodes, all_nodes
2762

    
2763
  def CheckPrereq(self):
2764
    """Check prerequisites.
2765

2766
    This checks:
2767
     - the node exists in the configuration
2768
     - it does not have primary or secondary instances
2769
     - it's not the master
2770

2771
    Any errors are signaled by raising errors.OpPrereqError.
2772

2773
    """
2774
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
2775
    node = self.cfg.GetNodeInfo(self.op.node_name)
2776
    assert node is not None
2777

    
2778
    instance_list = self.cfg.GetInstanceList()
2779

    
2780
    masternode = self.cfg.GetMasterNode()
2781
    if node.name == masternode:
2782
      raise errors.OpPrereqError("Node is the master node,"
2783
                                 " you need to failover first.",
2784
                                 errors.ECODE_INVAL)
2785

    
2786
    for instance_name in instance_list:
2787
      instance = self.cfg.GetInstanceInfo(instance_name)
2788
      if node.name in instance.all_nodes:
2789
        raise errors.OpPrereqError("Instance %s is still running on the node,"
2790
                                   " please remove first." % instance_name,
2791
                                   errors.ECODE_INVAL)
2792
    self.op.node_name = node.name
2793
    self.node = node
2794

    
2795
  def Exec(self, feedback_fn):
2796
    """Removes the node from the cluster.
2797

2798
    """
2799
    node = self.node
2800
    logging.info("Stopping the node daemon and removing configs from node %s",
2801
                 node.name)
2802

    
2803
    modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
2804

    
2805
    # Promote nodes to master candidate as needed
2806
    _AdjustCandidatePool(self, exceptions=[node.name])
2807
    self.context.RemoveNode(node.name)
2808

    
2809
    # Run post hooks on the node before it's removed
2810
    hm = self.proc.hmclass(self.rpc.call_hooks_runner, self)
2811
    try:
2812
      hm.RunPhase(constants.HOOKS_PHASE_POST, [node.name])
2813
    except:
2814
      # pylint: disable-msg=W0702
2815
      self.LogWarning("Errors occurred running hooks on %s" % node.name)
2816

    
2817
    result = self.rpc.call_node_leave_cluster(node.name, modify_ssh_setup)
2818
    msg = result.fail_msg
2819
    if msg:
2820
      self.LogWarning("Errors encountered on the remote node while leaving"
2821
                      " the cluster: %s", msg)
2822

    
2823

    
2824
class LUQueryNodes(NoHooksLU):
2825
  """Logical unit for querying nodes.
2826

2827
  """
2828
  # pylint: disable-msg=W0142
2829
  _OP_REQP = ["output_fields", "names", "use_locking"]
2830
  REQ_BGL = False
2831

    
2832
  _SIMPLE_FIELDS = ["name", "serial_no", "ctime", "mtime", "uuid",
2833
                    "master_candidate", "offline", "drained"]
2834

    
2835
  _FIELDS_DYNAMIC = utils.FieldSet(
2836
    "dtotal", "dfree",
2837
    "mtotal", "mnode", "mfree",
2838
    "bootid",
2839
    "ctotal", "cnodes", "csockets",
2840
    )
2841

    
2842
  _FIELDS_STATIC = utils.FieldSet(*[
2843
    "pinst_cnt", "sinst_cnt",
2844
    "pinst_list", "sinst_list",
2845
    "pip", "sip", "tags",
2846
    "master",
2847
    "role"] + _SIMPLE_FIELDS
2848
    )
2849

    
2850
  def ExpandNames(self):
2851
    _CheckOutputFields(static=self._FIELDS_STATIC,
2852
                       dynamic=self._FIELDS_DYNAMIC,
2853
                       selected=self.op.output_fields)
2854

    
2855
    self.needed_locks = {}
2856
    self.share_locks[locking.LEVEL_NODE] = 1
2857

    
2858
    if self.op.names:
2859
      self.wanted = _GetWantedNodes(self, self.op.names)
2860
    else:
2861
      self.wanted = locking.ALL_SET
2862

    
2863
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
2864
    self.do_locking = self.do_node_query and self.op.use_locking
2865
    if self.do_locking:
2866
      # if we don't request only static fields, we need to lock the nodes
2867
      self.needed_locks[locking.LEVEL_NODE] = self.wanted
2868

    
2869
  def CheckPrereq(self):
2870
    """Check prerequisites.
2871

2872
    """
2873
    # The validation of the node list is done in the _GetWantedNodes,
2874
    # if non empty, and if empty, there's no validation to do
2875
    pass
2876

    
2877
  def Exec(self, feedback_fn):
2878
    """Computes the list of nodes and their attributes.
2879

2880
    """
2881
    all_info = self.cfg.GetAllNodesInfo()
2882
    if self.do_locking:
2883
      nodenames = self.acquired_locks[locking.LEVEL_NODE]
2884
    elif self.wanted != locking.ALL_SET:
2885
      nodenames = self.wanted
2886
      missing = set(nodenames).difference(all_info.keys())
2887
      if missing:
2888
        raise errors.OpExecError(
2889
          "Some nodes were removed before retrieving their data: %s" % missing)
2890
    else:
2891
      nodenames = all_info.keys()
2892

    
2893
    nodenames = utils.NiceSort(nodenames)
2894
    nodelist = [all_info[name] for name in nodenames]
2895

    
2896
    # begin data gathering
2897

    
2898
    if self.do_node_query:
2899
      live_data = {}
2900
      node_data = self.rpc.call_node_info(nodenames, self.cfg.GetVGName(),
2901
                                          self.cfg.GetHypervisorType())
2902
      for name in nodenames:
2903
        nodeinfo = node_data[name]
2904
        if not nodeinfo.fail_msg and nodeinfo.payload:
2905
          nodeinfo = nodeinfo.payload
2906
          fn = utils.TryConvert
2907
          live_data[name] = {
2908
            "mtotal": fn(int, nodeinfo.get('memory_total', None)),
2909
            "mnode": fn(int, nodeinfo.get('memory_dom0', None)),
2910
            "mfree": fn(int, nodeinfo.get('memory_free', None)),
2911
            "dtotal": fn(int, nodeinfo.get('vg_size', None)),
2912
            "dfree": fn(int, nodeinfo.get('vg_free', None)),
2913
            "ctotal": fn(int, nodeinfo.get('cpu_total', None)),
2914
            "bootid": nodeinfo.get('bootid', None),
2915
            "cnodes": fn(int, nodeinfo.get('cpu_nodes', None)),
2916
            "csockets": fn(int, nodeinfo.get('cpu_sockets', None)),
2917
            }
2918
        else:
2919
          live_data[name] = {}
2920
    else:
2921
      live_data = dict.fromkeys(nodenames, {})
2922

    
2923
    node_to_primary = dict([(name, set()) for name in nodenames])
2924
    node_to_secondary = dict([(name, set()) for name in nodenames])
2925

    
2926
    inst_fields = frozenset(("pinst_cnt", "pinst_list",
2927
                             "sinst_cnt", "sinst_list"))
2928
    if inst_fields & frozenset(self.op.output_fields):
2929
      inst_data = self.cfg.GetAllInstancesInfo()
2930

    
2931
      for inst in inst_data.values():
2932
        if inst.primary_node in node_to_primary:
2933
          node_to_primary[inst.primary_node].add(inst.name)
2934
        for secnode in inst.secondary_nodes:
2935
          if secnode in node_to_secondary:
2936
            node_to_secondary[secnode].add(inst.name)
2937

    
2938
    master_node = self.cfg.GetMasterNode()
2939

    
2940
    # end data gathering
2941

    
2942
    output = []
2943
    for node in nodelist:
2944
      node_output = []
2945
      for field in self.op.output_fields:
2946
        if field in self._SIMPLE_FIELDS:
2947
          val = getattr(node, field)
2948
        elif field == "pinst_list":
2949
          val = list(node_to_primary[node.name])
2950
        elif field == "sinst_list":
2951
          val = list(node_to_secondary[node.name])
2952
        elif field == "pinst_cnt":
2953
          val = len(node_to_primary[node.name])
2954
        elif field == "sinst_cnt":
2955
          val = len(node_to_secondary[node.name])
2956
        elif field == "pip":
2957
          val = node.primary_ip
2958
        elif field == "sip":
2959
          val = node.secondary_ip
2960
        elif field == "tags":
2961
          val = list(node.GetTags())
2962
        elif field == "master":
2963
          val = node.name == master_node
2964
        elif self._FIELDS_DYNAMIC.Matches(field):
2965
          val = live_data[node.name].get(field, None)
2966
        elif field == "role":
2967
          if node.name == master_node:
2968
            val = "M"
2969
          elif node.master_candidate:
2970
            val = "C"
2971
          elif node.drained:
2972
            val = "D"
2973
          elif node.offline:
2974
            val = "O"
2975
          else:
2976
            val = "R"
2977
        else:
2978
          raise errors.ParameterError(field)
2979
        node_output.append(val)
2980
      output.append(node_output)
2981

    
2982
    return output
2983

    
2984

    
2985
class LUQueryNodeVolumes(NoHooksLU):
2986
  """Logical unit for getting volumes on node(s).
2987

2988
  """
2989
  _OP_REQP = ["nodes", "output_fields"]
2990
  REQ_BGL = False
2991
  _FIELDS_DYNAMIC = utils.FieldSet("phys", "vg", "name", "size", "instance")
2992
  _FIELDS_STATIC = utils.FieldSet("node")
2993

    
2994
  def ExpandNames(self):
2995
    _CheckOutputFields(static=self._FIELDS_STATIC,
2996
                       dynamic=self._FIELDS_DYNAMIC,
2997
                       selected=self.op.output_fields)
2998

    
2999
    self.needed_locks = {}
3000
    self.share_locks[locking.LEVEL_NODE] = 1
3001
    if not self.op.nodes:
3002
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3003
    else:
3004
      self.needed_locks[locking.LEVEL_NODE] = \
3005
        _GetWantedNodes(self, self.op.nodes)
3006

    
3007
  def CheckPrereq(self):
3008
    """Check prerequisites.
3009

3010
    This checks that the fields required are valid output fields.
3011

3012
    """
3013
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
3014

    
3015
  def Exec(self, feedback_fn):
3016
    """Computes the list of nodes and their attributes.
3017

3018
    """
3019
    nodenames = self.nodes
3020
    volumes = self.rpc.call_node_volumes(nodenames)
3021

    
3022
    ilist = [self.cfg.GetInstanceInfo(iname) for iname
3023
             in self.cfg.GetInstanceList()]
3024

    
3025
    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
3026

    
3027
    output = []
3028
    for node in nodenames:
3029
      nresult = volumes[node]
3030
      if nresult.offline:
3031
        continue
3032
      msg = nresult.fail_msg
3033
      if msg:
3034
        self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
3035
        continue
3036

    
3037
      node_vols = nresult.payload[:]
3038
      node_vols.sort(key=lambda vol: vol['dev'])
3039

    
3040
      for vol in node_vols:
3041
        node_output = []
3042
        for field in self.op.output_fields:
3043
          if field == "node":
3044
            val = node
3045
          elif field == "phys":
3046
            val = vol['dev']
3047
          elif field == "vg":
3048
            val = vol['vg']
3049
          elif field == "name":
3050
            val = vol['name']
3051
          elif field == "size":
3052
            val = int(float(vol['size']))
3053
          elif field == "instance":
3054
            for inst in ilist:
3055
              if node not in lv_by_node[inst]:
3056
                continue
3057
              if vol['name'] in lv_by_node[inst][node]:
3058
                val = inst.name
3059
                break
3060
            else:
3061
              val = '-'
3062
          else:
3063
            raise errors.ParameterError(field)
3064
          node_output.append(str(val))
3065

    
3066
        output.append(node_output)
3067

    
3068
    return output
3069

    
3070

    
3071
class LUQueryNodeStorage(NoHooksLU):
3072
  """Logical unit for getting information on storage units on node(s).
3073

3074
  """
3075
  _OP_REQP = ["nodes", "storage_type", "output_fields"]
3076
  REQ_BGL = False
3077
  _FIELDS_STATIC = utils.FieldSet(constants.SF_NODE)
3078

    
3079
  def ExpandNames(self):
3080
    storage_type = self.op.storage_type
3081

    
3082
    if storage_type not in constants.VALID_STORAGE_TYPES:
3083
      raise errors.OpPrereqError("Unknown storage type: %s" % storage_type,
3084
                                 errors.ECODE_INVAL)
3085

    
3086
    _CheckOutputFields(static=self._FIELDS_STATIC,
3087
                       dynamic=utils.FieldSet(*constants.VALID_STORAGE_FIELDS),
3088
                       selected=self.op.output_fields)
3089

    
3090
    self.needed_locks = {}
3091
    self.share_locks[locking.LEVEL_NODE] = 1
3092

    
3093
    if self.op.nodes:
3094
      self.needed_locks[locking.LEVEL_NODE] = \
3095
        _GetWantedNodes(self, self.op.nodes)
3096
    else:
3097
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
3098

    
3099
  def CheckPrereq(self):
3100
    """Check prerequisites.
3101

3102
    This checks that the fields required are valid output fields.
3103

3104
    """
3105
    self.op.name = getattr(self.op, "name", None)
3106

    
3107
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
3108

    
3109
  def Exec(self, feedback_fn):
3110
    """Computes the list of nodes and their attributes.
3111

3112
    """
3113
    # Always get name to sort by
3114
    if constants.SF_NAME in self.op.output_fields:
3115
      fields = self.op.output_fields[:]
3116
    else:
3117
      fields = [constants.SF_NAME] + self.op.output_fields
3118

    
3119
    # Never ask for node or type as it's only known to the LU
3120
    for extra in [constants.SF_NODE, constants.SF_TYPE]:
3121
      while extra in fields:
3122
        fields.remove(extra)
3123

    
3124
    field_idx = dict([(name, idx) for (idx, name) in enumerate(fields)])
3125
    name_idx = field_idx[constants.SF_NAME]
3126

    
3127
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
3128
    data = self.rpc.call_storage_list(self.nodes,
3129
                                      self.op.storage_type, st_args,
3130
                                      self.op.name, fields)
3131

    
3132
    result = []
3133

    
3134
    for node in utils.NiceSort(self.nodes):
3135
      nresult = data[node]
3136
      if nresult.offline:
3137
        continue
3138

    
3139
      msg = nresult.fail_msg
3140
      if msg:
3141
        self.LogWarning("Can't get storage data from node %s: %s", node, msg)
3142
        continue
3143

    
3144
      rows = dict([(row[name_idx], row) for row in nresult.payload])
3145

    
3146
      for name in utils.NiceSort(rows.keys()):
3147
        row = rows[name]
3148

    
3149
        out = []
3150

    
3151
        for field in self.op.output_fields:
3152
          if field == constants.SF_NODE:
3153
            val = node
3154
          elif field == constants.SF_TYPE:
3155
            val = self.op.storage_type
3156
          elif field in field_idx:
3157
            val = row[field_idx[field]]
3158
          else:
3159
            raise errors.ParameterError(field)
3160

    
3161
          out.append(val)
3162

    
3163
        result.append(out)
3164

    
3165
    return result
3166

    
3167

    
3168
class LUModifyNodeStorage(NoHooksLU):
3169
  """Logical unit for modifying a storage volume on a node.
3170

3171
  """
3172
  _OP_REQP = ["node_name", "storage_type", "name", "changes"]
3173
  REQ_BGL = False
3174

    
3175
  def CheckArguments(self):
3176
    self.opnode_name = _ExpandNodeName(self.cfg, self.op.node_name)
3177

    
3178
    storage_type = self.op.storage_type
3179
    if storage_type not in constants.VALID_STORAGE_TYPES:
3180
      raise errors.OpPrereqError("Unknown storage type: %s" % storage_type,
3181
                                 errors.ECODE_INVAL)
3182

    
3183
  def ExpandNames(self):
3184
    self.needed_locks = {
3185
      locking.LEVEL_NODE: self.op.node_name,
3186
      }
3187

    
3188
  def CheckPrereq(self):
3189
    """Check prerequisites.
3190

3191
    """
3192
    storage_type = self.op.storage_type
3193

    
3194
    try:
3195
      modifiable = constants.MODIFIABLE_STORAGE_FIELDS[storage_type]
3196
    except KeyError:
3197
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
3198
                                 " modified" % storage_type,
3199
                                 errors.ECODE_INVAL)
3200

    
3201
    diff = set(self.op.changes.keys()) - modifiable
3202
    if diff:
3203
      raise errors.OpPrereqError("The following fields can not be modified for"
3204
                                 " storage units of type '%s': %r" %
3205
                                 (storage_type, list(diff)),
3206
                                 errors.ECODE_INVAL)
3207

    
3208
  def Exec(self, feedback_fn):
3209
    """Computes the list of nodes and their attributes.
3210

3211
    """
3212
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
3213
    result = self.rpc.call_storage_modify(self.op.node_name,
3214
                                          self.op.storage_type, st_args,
3215
                                          self.op.name, self.op.changes)
3216
    result.Raise("Failed to modify storage unit '%s' on %s" %
3217
                 (self.op.name, self.op.node_name))
3218

    
3219

    
3220
class LUAddNode(LogicalUnit):
3221
  """Logical unit for adding node to the cluster.
3222

3223
  """
3224
  HPATH = "node-add"
3225
  HTYPE = constants.HTYPE_NODE
3226
  _OP_REQP = ["node_name"]
3227

    
3228
  def CheckArguments(self):
3229
    # validate/normalize the node name
3230
    self.op.node_name = utils.HostInfo.NormalizeName(self.op.node_name)
3231

    
3232
  def BuildHooksEnv(self):
3233
    """Build hooks env.
3234

3235
    This will run on all nodes before, and on all nodes + the new node after.
3236

3237
    """
3238
    env = {
3239
      "OP_TARGET": self.op.node_name,
3240
      "NODE_NAME": self.op.node_name,
3241
      "NODE_PIP": self.op.primary_ip,
3242
      "NODE_SIP": self.op.secondary_ip,
3243
      }
3244
    nodes_0 = self.cfg.GetNodeList()
3245
    nodes_1 = nodes_0 + [self.op.node_name, ]
3246
    return env, nodes_0, nodes_1
3247

    
3248
  def CheckPrereq(self):
3249
    """Check prerequisites.
3250

3251
    This checks:
3252
     - the new node is not already in the config
3253
     - it is resolvable
3254
     - its parameters (single/dual homed) matches the cluster
3255

3256
    Any errors are signaled by raising errors.OpPrereqError.
3257

3258
    """
3259
    node_name = self.op.node_name
3260
    cfg = self.cfg
3261

    
3262
    dns_data = utils.GetHostInfo(node_name)
3263

    
3264
    node = dns_data.name
3265
    primary_ip = self.op.primary_ip = dns_data.ip
3266
    secondary_ip = getattr(self.op, "secondary_ip", None)
3267
    if secondary_ip is None:
3268
      secondary_ip = primary_ip
3269
    if not utils.IsValidIP(secondary_ip):
3270
      raise errors.OpPrereqError("Invalid secondary IP given",
3271
                                 errors.ECODE_INVAL)
3272
    self.op.secondary_ip = secondary_ip
3273

    
3274
    node_list = cfg.GetNodeList()
3275
    if not self.op.readd and node in node_list:
3276
      raise errors.OpPrereqError("Node %s is already in the configuration" %
3277
                                 node, errors.ECODE_EXISTS)
3278
    elif self.op.readd and node not in node_list:
3279
      raise errors.OpPrereqError("Node %s is not in the configuration" % node,
3280
                                 errors.ECODE_NOENT)
3281

    
3282
    for existing_node_name in node_list:
3283
      existing_node = cfg.GetNodeInfo(existing_node_name)
3284

    
3285
      if self.op.readd and node == existing_node_name:
3286
        if (existing_node.primary_ip != primary_ip or
3287
            existing_node.secondary_ip != secondary_ip):
3288
          raise errors.OpPrereqError("Readded node doesn't have the same IP"
3289
                                     " address configuration as before",
3290
                                     errors.ECODE_INVAL)
3291
        continue
3292

    
3293
      if (existing_node.primary_ip == primary_ip or
3294
          existing_node.secondary_ip == primary_ip or
3295
          existing_node.primary_ip == secondary_ip or
3296
          existing_node.secondary_ip == secondary_ip):
3297
        raise errors.OpPrereqError("New node ip address(es) conflict with"
3298
                                   " existing node %s" % existing_node.name,
3299
                                   errors.ECODE_NOTUNIQUE)
3300

    
3301
    # check that the type of the node (single versus dual homed) is the
3302
    # same as for the master
3303
    myself = cfg.GetNodeInfo(self.cfg.GetMasterNode())
3304
    master_singlehomed = myself.secondary_ip == myself.primary_ip
3305
    newbie_singlehomed = secondary_ip == primary_ip
3306
    if master_singlehomed != newbie_singlehomed:
3307
      if master_singlehomed:
3308
        raise errors.OpPrereqError("The master has no private ip but the"
3309
                                   " new node has one",
3310
                                   errors.ECODE_INVAL)
3311
      else:
3312
        raise errors.OpPrereqError("The master has a private ip but the"
3313
                                   " new node doesn't have one",
3314
                                   errors.ECODE_INVAL)
3315

    
3316
    # checks reachability
3317
    if not utils.TcpPing(primary_ip, constants.DEFAULT_NODED_PORT):
3318
      raise errors.OpPrereqError("Node not reachable by ping",
3319
                                 errors.ECODE_ENVIRON)
3320

    
3321
    if not newbie_singlehomed:
3322
      # check reachability from my secondary ip to newbie's secondary ip
3323
      if not utils.TcpPing(secondary_ip, constants.DEFAULT_NODED_PORT,
3324
                           source=myself.secondary_ip):
3325
        raise errors.OpPrereqError("Node secondary ip not reachable by TCP"
3326
                                   " based ping to noded port",
3327
                                   errors.ECODE_ENVIRON)
3328

    
3329
    if self.op.readd:
3330
      exceptions = [node]
3331
    else:
3332
      exceptions = []
3333

    
3334
    self.master_candidate = _DecideSelfPromotion(self, exceptions=exceptions)
3335

    
3336
    if self.op.readd:
3337
      self.new_node = self.cfg.GetNodeInfo(node)
3338
      assert self.new_node is not None, "Can't retrieve locked node %s" % node
3339
    else:
3340
      self.new_node = objects.Node(name=node,
3341
                                   primary_ip=primary_ip,
3342
                                   secondary_ip=secondary_ip,
3343
                                   master_candidate=self.master_candidate,
3344
                                   offline=False, drained=False)
3345

    
3346
  def Exec(self, feedback_fn):
3347
    """Adds the new node to the cluster.
3348

3349
    """
3350
    new_node = self.new_node
3351
    node = new_node.name
3352

    
3353
    # for re-adds, reset the offline/drained/master-candidate flags;
3354
    # we need to reset here, otherwise offline would prevent RPC calls
3355
    # later in the procedure; this also means that if the re-add
3356
    # fails, we are left with a non-offlined, broken node
3357
    if self.op.readd:
3358
      new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
3359
      self.LogInfo("Readding a node, the offline/drained flags were reset")
3360
      # if we demote the node, we do cleanup later in the procedure
3361
      new_node.master_candidate = self.master_candidate
3362

    
3363
    # notify the user about any possible mc promotion
3364
    if new_node.master_candidate:
3365
      self.LogInfo("Node will be a master candidate")
3366

    
3367
    # check connectivity
3368
    result = self.rpc.call_version([node])[node]
3369
    result.Raise("Can't get version information from node %s" % node)
3370
    if constants.PROTOCOL_VERSION == result.payload:
3371
      logging.info("Communication to node %s fine, sw version %s match",
3372
                   node, result.payload)
3373
    else:
3374
      raise errors.OpExecError("Version mismatch master version %s,"
3375
                               " node version %s" %
3376
                               (constants.PROTOCOL_VERSION, result.payload))
3377

    
3378
    # setup ssh on node
3379
    if self.cfg.GetClusterInfo().modify_ssh_setup:
3380
      logging.info("Copy ssh key to node %s", node)
3381
      priv_key, pub_key, _ = ssh.GetUserFiles(constants.GANETI_RUNAS)
3382
      keyarray = []
3383
      keyfiles = [constants.SSH_HOST_DSA_PRIV, constants.SSH_HOST_DSA_PUB,
3384
                  constants.SSH_HOST_RSA_PRIV, constants.SSH_HOST_RSA_PUB,
3385
                  priv_key, pub_key]
3386

    
3387
      for i in keyfiles:
3388
        keyarray.append(utils.ReadFile(i))
3389

    
3390
      result = self.rpc.call_node_add(node, keyarray[0], keyarray[1],
3391
                                      keyarray[2], keyarray[3], keyarray[4],
3392
                                      keyarray[5])
3393
      result.Raise("Cannot transfer ssh keys to the new node")
3394

    
3395
    # Add node to our /etc/hosts, and add key to known_hosts
3396
    if self.cfg.GetClusterInfo().modify_etc_hosts:
3397
      utils.AddHostToEtcHosts(new_node.name)
3398

    
3399
    if new_node.secondary_ip != new_node.primary_ip:
3400
      result = self.rpc.call_node_has_ip_address(new_node.name,
3401
                                                 new_node.secondary_ip)
3402
      result.Raise("Failure checking secondary ip on node %s" % new_node.name,
3403
                   prereq=True, ecode=errors.ECODE_ENVIRON)
3404
      if not result.payload:
3405
        raise errors.OpExecError("Node claims it doesn't have the secondary ip"
3406
                                 " you gave (%s). Please fix and re-run this"
3407
                                 " command." % new_node.secondary_ip)
3408

    
3409
    node_verify_list = [self.cfg.GetMasterNode()]
3410
    node_verify_param = {
3411
      constants.NV_NODELIST: [node],
3412
      # TODO: do a node-net-test as well?
3413
    }
3414

    
3415
    result = self.rpc.call_node_verify(node_verify_list, node_verify_param,
3416
                                       self.cfg.GetClusterName())
3417
    for verifier in node_verify_list:
3418
      result[verifier].Raise("Cannot communicate with node %s" % verifier)
3419
      nl_payload = result[verifier].payload[constants.NV_NODELIST]
3420
      if nl_payload:
3421
        for failed in nl_payload:
3422
          feedback_fn("ssh/hostname verification failed"
3423
                      " (checking from %s): %s" %
3424
                      (verifier, nl_payload[failed]))
3425
        raise errors.OpExecError("ssh/hostname verification failed.")
3426

    
3427
    if self.op.readd:
3428
      _RedistributeAncillaryFiles(self)
3429
      self.context.ReaddNode(new_node)
3430
      # make sure we redistribute the config
3431
      self.cfg.Update(new_node, feedback_fn)
3432
      # and make sure the new node will not have old files around
3433
      if not new_node.master_candidate:
3434
        result = self.rpc.call_node_demote_from_mc(new_node.name)
3435
        msg = result.fail_msg
3436
        if msg:
3437
          self.LogWarning("Node failed to demote itself from master"
3438
                          " candidate status: %s" % msg)
3439
    else:
3440
      _RedistributeAncillaryFiles(self, additional_nodes=[node])
3441
      self.context.AddNode(new_node, self.proc.GetECId())
3442

    
3443

    
3444
class LUSetNodeParams(LogicalUnit):
3445
  """Modifies the parameters of a node.
3446

3447
  """
3448
  HPATH = "node-modify"
3449
  HTYPE = constants.HTYPE_NODE
3450
  _OP_REQP = ["node_name"]
3451
  REQ_BGL = False
3452

    
3453
  def CheckArguments(self):
3454
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
3455
    _CheckBooleanOpField(self.op, 'master_candidate')
3456
    _CheckBooleanOpField(self.op, 'offline')
3457
    _CheckBooleanOpField(self.op, 'drained')
3458
    _CheckBooleanOpField(self.op, 'auto_promote')
3459
    all_mods = [self.op.offline, self.op.master_candidate, self.op.drained]
3460
    if all_mods.count(None) == 3:
3461
      raise errors.OpPrereqError("Please pass at least one modification",
3462
                                 errors.ECODE_INVAL)
3463
    if all_mods.count(True) > 1:
3464
      raise errors.OpPrereqError("Can't set the node into more than one"
3465
                                 " state at the same time",
3466
                                 errors.ECODE_INVAL)
3467

    
3468
    # Boolean value that tells us whether we're offlining or draining the node
3469
    self.offline_or_drain = (self.op.offline == True or
3470
                             self.op.drained == True)
3471
    self.deoffline_or_drain = (self.op.offline == False or
3472
                               self.op.drained == False)
3473
    self.might_demote = (self.op.master_candidate == False or
3474
                         self.offline_or_drain)
3475

    
3476
    self.lock_all = self.op.auto_promote and self.might_demote
3477

    
3478

    
3479
  def ExpandNames(self):
3480
    if self.lock_all:
3481
      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
3482
    else:
3483
      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
3484

    
3485
  def BuildHooksEnv(self):
3486
    """Build hooks env.
3487

3488
    This runs on the master node.
3489

3490
    """
3491
    env = {
3492
      "OP_TARGET": self.op.node_name,
3493
      "MASTER_CANDIDATE": str(self.op.master_candidate),
3494
      "OFFLINE": str(self.op.offline),
3495
      "DRAINED": str(self.op.drained),
3496
      }
3497
    nl = [self.cfg.GetMasterNode(),
3498
          self.op.node_name]
3499
    return env, nl, nl
3500

    
3501
  def CheckPrereq(self):
3502
    """Check prerequisites.
3503

3504
    This only checks the instance list against the existing names.
3505

3506
    """
3507
    node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
3508

    
3509
    if (self.op.master_candidate is not None or
3510
        self.op.drained is not None or
3511
        self.op.offline is not None):
3512
      # we can't change the master's node flags
3513
      if self.op.node_name == self.cfg.GetMasterNode():
3514
        raise errors.OpPrereqError("The master role can be changed"
3515
                                   " only via masterfailover",
3516
                                   errors.ECODE_INVAL)
3517

    
3518

    
3519
    if node.master_candidate and self.might_demote and not self.lock_all:
3520
      assert not self.op.auto_promote, "auto-promote set but lock_all not"
3521
      # check if after removing the current node, we're missing master
3522
      # candidates
3523
      (mc_remaining, mc_should, _) = \
3524
          self.cfg.GetMasterCandidateStats(exceptions=[node.name])
3525
      if mc_remaining < mc_should:
3526
        raise errors.OpPrereqError("Not enough master candidates, please"
3527
                                   " pass auto_promote to allow promotion",
3528
                                   errors.ECODE_INVAL)
3529

    
3530
    if (self.op.master_candidate == True and
3531
        ((node.offline and not self.op.offline == False) or
3532
         (node.drained and not self.op.drained == False))):
3533
      raise errors.OpPrereqError("Node '%s' is offline or drained, can't set"
3534
                                 " to master_candidate" % node.name,
3535
                                 errors.ECODE_INVAL)
3536

    
3537
    # If we're being deofflined/drained, we'll MC ourself if needed
3538
    if (self.deoffline_or_drain and not self.offline_or_drain and not
3539
        self.op.master_candidate == True and not node.master_candidate):
3540
      self.op.master_candidate = _DecideSelfPromotion(self)
3541
      if self.op.master_candidate:
3542
        self.LogInfo("Autopromoting node to master candidate")
3543

    
3544
    return
3545

    
3546
  def Exec(self, feedback_fn):
3547
    """Modifies a node.
3548

3549
    """
3550
    node = self.node
3551

    
3552
    result = []
3553
    changed_mc = False
3554

    
3555
    if self.op.offline is not None:
3556
      node.offline = self.op.offline
3557
      result.append(("offline", str(self.op.offline)))
3558
      if self.op.offline == True:
3559
        if node.master_candidate:
3560
          node.master_candidate = False
3561
          changed_mc = True
3562
          result.append(("master_candidate", "auto-demotion due to offline"))
3563
        if node.drained:
3564
          node.drained = False
3565
          result.append(("drained", "clear drained status due to offline"))
3566

    
3567
    if self.op.master_candidate is not None:
3568
      node.master_candidate = self.op.master_candidate
3569
      changed_mc = True
3570
      result.append(("master_candidate", str(self.op.master_candidate)))
3571
      if self.op.master_candidate == False:
3572
        rrc = self.rpc.call_node_demote_from_mc(node.name)
3573
        msg = rrc.fail_msg
3574
        if msg:
3575
          self.LogWarning("Node failed to demote itself: %s" % msg)
3576

    
3577
    if self.op.drained is not None:
3578
      node.drained = self.op.drained
3579
      result.append(("drained", str(self.op.drained)))
3580
      if self.op.drained == True:
3581
        if node.master_candidate:
3582
          node.master_candidate = False
3583
          changed_mc = True
3584
          result.append(("master_candidate", "auto-demotion due to drain"))
3585
          rrc = self.rpc.call_node_demote_from_mc(node.name)
3586
          msg = rrc.fail_msg
3587
          if msg:
3588
            self.LogWarning("Node failed to demote itself: %s" % msg)
3589
        if node.offline:
3590
          node.offline = False
3591
          result.append(("offline", "clear offline status due to drain"))
3592

    
3593
    # we locked all nodes, we adjust the CP before updating this node
3594
    if self.lock_all:
3595
      _AdjustCandidatePool(self, [node.name])
3596

    
3597
    # this will trigger configuration file update, if needed
3598
    self.cfg.Update(node, feedback_fn)
3599

    
3600
    # this will trigger job queue propagation or cleanup
3601
    if changed_mc:
3602
      self.context.ReaddNode(node)
3603

    
3604
    return result
3605

    
3606

    
3607
class LUPowercycleNode(NoHooksLU):
3608
  """Powercycles a node.
3609

3610
  """
3611
  _OP_REQP = ["node_name", "force"]
3612
  REQ_BGL = False
3613

    
3614
  def CheckArguments(self):
3615
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
3616
    if self.op.node_name == self.cfg.GetMasterNode() and not self.op.force:
3617
      raise errors.OpPrereqError("The node is the master and the force"
3618
                                 " parameter was not set",
3619
                                 errors.ECODE_INVAL)
3620

    
3621
  def ExpandNames(self):
3622
    """Locking for PowercycleNode.
3623

3624
    This is a last-resort option and shouldn't block on other
3625
    jobs. Therefore, we grab no locks.
3626

3627
    """
3628
    self.needed_locks = {}
3629

    
3630
  def CheckPrereq(self):
3631
    """Check prerequisites.
3632

3633
    This LU has no prereqs.
3634

3635
    """
3636
    pass
3637

    
3638
  def Exec(self, feedback_fn):
3639
    """Reboots a node.
3640

3641
    """
3642
    result = self.rpc.call_node_powercycle(self.op.node_name,
3643
                                           self.cfg.GetHypervisorType())
3644
    result.Raise("Failed to schedule the reboot")
3645
    return result.payload
3646

    
3647

    
3648
class LUQueryClusterInfo(NoHooksLU):
3649
  """Query cluster configuration.
3650

3651
  """
3652
  _OP_REQP = []
3653
  REQ_BGL = False
3654

    
3655
  def ExpandNames(self):
3656
    self.needed_locks = {}
3657

    
3658
  def CheckPrereq(self):
3659
    """No prerequsites needed for this LU.
3660

3661
    """
3662
    pass
3663

    
3664
  def Exec(self, feedback_fn):
3665
    """Return cluster config.
3666

3667
    """
3668
    cluster = self.cfg.GetClusterInfo()
3669
    os_hvp = {}
3670

    
3671
    # Filter just for enabled hypervisors
3672
    for os_name, hv_dict in cluster.os_hvp.items():
3673
      os_hvp[os_name] = {}
3674
      for hv_name, hv_params in hv_dict.items():
3675
        if hv_name in cluster.enabled_hypervisors:
3676
          os_hvp[os_name][hv_name] = hv_params
3677

    
3678
    result = {
3679
      "software_version": constants.RELEASE_VERSION,
3680
      "protocol_version": constants.PROTOCOL_VERSION,
3681
      "config_version": constants.CONFIG_VERSION,
3682
      "os_api_version": max(constants.OS_API_VERSIONS),
3683
      "export_version": constants.EXPORT_VERSION,
3684
      "architecture": (platform.architecture()[0], platform.machine()),
3685
      "name": cluster.cluster_name,
3686
      "master": cluster.master_node,
3687
      "default_hypervisor": cluster.enabled_hypervisors[0],
3688
      "enabled_hypervisors": cluster.enabled_hypervisors,
3689
      "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
3690
                        for hypervisor_name in cluster.enabled_hypervisors]),
3691
      "os_hvp": os_hvp,
3692
      "beparams": cluster.beparams,
3693
      "nicparams": cluster.nicparams,
3694
      "candidate_pool_size": cluster.candidate_pool_size,
3695
      "master_netdev": cluster.master_netdev,
3696
      "volume_group_name": cluster.volume_group_name,
3697
      "file_storage_dir": cluster.file_storage_dir,
3698
      "maintain_node_health": cluster.maintain_node_health,
3699
      "ctime": cluster.ctime,
3700
      "mtime": cluster.mtime,
3701
      "uuid": cluster.uuid,
3702
      "tags": list(cluster.GetTags()),
3703
      }
3704

    
3705
    return result
3706

    
3707

    
3708
class LUQueryConfigValues(NoHooksLU):
3709
  """Return configuration values.
3710

3711
  """
3712
  _OP_REQP = []
3713
  REQ_BGL = False
3714
  _FIELDS_DYNAMIC = utils.FieldSet()
3715
  _FIELDS_STATIC = utils.FieldSet("cluster_name", "master_node", "drain_flag",
3716
                                  "watcher_pause")
3717

    
3718
  def ExpandNames(self):
3719
    self.needed_locks = {}
3720

    
3721
    _CheckOutputFields(static=self._FIELDS_STATIC,
3722
                       dynamic=self._FIELDS_DYNAMIC,
3723
                       selected=self.op.output_fields)
3724

    
3725
  def CheckPrereq(self):
3726
    """No prerequisites.
3727

3728
    """
3729
    pass
3730

    
3731
  def Exec(self, feedback_fn):
3732
    """Dump a representation of the cluster config to the standard output.
3733

3734
    """
3735
    values = []
3736
    for field in self.op.output_fields:
3737
      if field == "cluster_name":
3738
        entry = self.cfg.GetClusterName()
3739
      elif field == "master_node":
3740
        entry = self.cfg.GetMasterNode()
3741
      elif field == "drain_flag":
3742
        entry = os.path.exists(constants.JOB_QUEUE_DRAIN_FILE)
3743
      elif field == "watcher_pause":
3744
        entry = utils.ReadWatcherPauseFile(constants.WATCHER_PAUSEFILE)
3745
      else:
3746
        raise errors.ParameterError(field)
3747
      values.append(entry)
3748
    return values
3749

    
3750

    
3751
class LUActivateInstanceDisks(NoHooksLU):
3752
  """Bring up an instance's disks.
3753

3754
  """
3755
  _OP_REQP = ["instance_name"]
3756
  REQ_BGL = False
3757

    
3758
  def ExpandNames(self):
3759
    self._ExpandAndLockInstance()
3760
    self.needed_locks[locking.LEVEL_NODE] = []
3761
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3762

    
3763
  def DeclareLocks(self, level):
3764
    if level == locking.LEVEL_NODE:
3765
      self._LockInstancesNodes()
3766

    
3767
  def CheckPrereq(self):
3768
    """Check prerequisites.
3769

3770
    This checks that the instance is in the cluster.
3771

3772
    """
3773
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3774
    assert self.instance is not None, \
3775
      "Cannot retrieve locked instance %s" % self.op.instance_name
3776
    _CheckNodeOnline(self, self.instance.primary_node)
3777
    if not hasattr(self.op, "ignore_size"):
3778
      self.op.ignore_size = False
3779

    
3780
  def Exec(self, feedback_fn):
3781
    """Activate the disks.
3782

3783
    """
3784
    disks_ok, disks_info = \
3785
              _AssembleInstanceDisks(self, self.instance,
3786
                                     ignore_size=self.op.ignore_size)
3787
    if not disks_ok:
3788
      raise errors.OpExecError("Cannot activate block devices")
3789

    
3790
    return disks_info
3791

    
3792

    
3793
def _AssembleInstanceDisks(lu, instance, ignore_secondaries=False,
3794
                           ignore_size=False):
3795
  """Prepare the block devices for an instance.
3796

3797
  This sets up the block devices on all nodes.
3798

3799
  @type lu: L{LogicalUnit}
3800
  @param lu: the logical unit on whose behalf we execute
3801
  @type instance: L{objects.Instance}
3802
  @param instance: the instance for whose disks we assemble
3803
  @type ignore_secondaries: boolean
3804
  @param ignore_secondaries: if true, errors on secondary nodes
3805
      won't result in an error return from the function
3806
  @type ignore_size: boolean
3807
  @param ignore_size: if true, the current known size of the disk
3808
      will not be used during the disk activation, useful for cases
3809
      when the size is wrong
3810
  @return: False if the operation failed, otherwise a list of
3811
      (host, instance_visible_name, node_visible_name)
3812
      with the mapping from node devices to instance devices
3813

3814
  """
3815
  device_info = []
3816
  disks_ok = True
3817
  iname = instance.name
3818
  # With the two passes mechanism we try to reduce the window of
3819
  # opportunity for the race condition of switching DRBD to primary
3820
  # before handshaking occured, but we do not eliminate it
3821

    
3822
  # The proper fix would be to wait (with some limits) until the
3823
  # connection has been made and drbd transitions from WFConnection
3824
  # into any other network-connected state (Connected, SyncTarget,
3825
  # SyncSource, etc.)
3826

    
3827
  # 1st pass, assemble on all nodes in secondary mode
3828
  for inst_disk in instance.disks:
3829
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
3830
      if ignore_size:
3831
        node_disk = node_disk.Copy()
3832
        node_disk.UnsetSize()
3833
      lu.cfg.SetDiskID(node_disk, node)
3834
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, False)
3835
      msg = result.fail_msg
3836
      if msg:
3837
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
3838
                           " (is_primary=False, pass=1): %s",
3839
                           inst_disk.iv_name, node, msg)
3840
        if not ignore_secondaries:
3841
          disks_ok = False
3842

    
3843
  # FIXME: race condition on drbd migration to primary
3844

    
3845
  # 2nd pass, do only the primary node
3846
  for inst_disk in instance.disks:
3847
    dev_path = None
3848

    
3849
    for node, node_disk in inst_disk.ComputeNodeTree(instance.primary_node):
3850
      if node != instance.primary_node:
3851
        continue
3852
      if ignore_size:
3853
        node_disk = node_disk.Copy()
3854
        node_disk.UnsetSize()
3855
      lu.cfg.SetDiskID(node_disk, node)
3856
      result = lu.rpc.call_blockdev_assemble(node, node_disk, iname, True)
3857
      msg = result.fail_msg
3858
      if msg:
3859
        lu.proc.LogWarning("Could not prepare block device %s on node %s"
3860
                           " (is_primary=True, pass=2): %s",
3861
                           inst_disk.iv_name, node, msg)
3862
        disks_ok = False
3863
      else:
3864
        dev_path = result.payload
3865

    
3866
    device_info.append((instance.primary_node, inst_disk.iv_name, dev_path))
3867

    
3868
  # leave the disks configured for the primary node
3869
  # this is a workaround that would be fixed better by
3870
  # improving the logical/physical id handling
3871
  for disk in instance.disks:
3872
    lu.cfg.SetDiskID(disk, instance.primary_node)
3873

    
3874
  return disks_ok, device_info
3875

    
3876

    
3877
def _StartInstanceDisks(lu, instance, force):
3878
  """Start the disks of an instance.
3879

3880
  """
3881
  disks_ok, _ = _AssembleInstanceDisks(lu, instance,
3882
                                           ignore_secondaries=force)
3883
  if not disks_ok:
3884
    _ShutdownInstanceDisks(lu, instance)
3885
    if force is not None and not force:
3886
      lu.proc.LogWarning("", hint="If the message above refers to a"
3887
                         " secondary node,"
3888
                         " you can retry the operation using '--force'.")
3889
    raise errors.OpExecError("Disk consistency error")
3890

    
3891

    
3892
class LUDeactivateInstanceDisks(NoHooksLU):
3893
  """Shutdown an instance's disks.
3894

3895
  """
3896
  _OP_REQP = ["instance_name"]
3897
  REQ_BGL = False
3898

    
3899
  def ExpandNames(self):
3900
    self._ExpandAndLockInstance()
3901
    self.needed_locks[locking.LEVEL_NODE] = []
3902
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
3903

    
3904
  def DeclareLocks(self, level):
3905
    if level == locking.LEVEL_NODE:
3906
      self._LockInstancesNodes()
3907

    
3908
  def CheckPrereq(self):
3909
    """Check prerequisites.
3910

3911
    This checks that the instance is in the cluster.
3912

3913
    """
3914
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
3915
    assert self.instance is not None, \
3916
      "Cannot retrieve locked instance %s" % self.op.instance_name
3917

    
3918
  def Exec(self, feedback_fn):
3919
    """Deactivate the disks
3920

3921
    """
3922
    instance = self.instance
3923
    _SafeShutdownInstanceDisks(self, instance)
3924

    
3925

    
3926
def _SafeShutdownInstanceDisks(lu, instance):
3927
  """Shutdown block devices of an instance.
3928

3929
  This function checks if an instance is running, before calling
3930
  _ShutdownInstanceDisks.
3931

3932
  """
3933
  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
3934
  _ShutdownInstanceDisks(lu, instance)
3935

    
3936

    
3937
def _ShutdownInstanceDisks(lu, instance, ignore_primary=False):
3938
  """Shutdown block devices of an instance.
3939

3940
  This does the shutdown on all nodes of the instance.
3941

3942
  If the ignore_primary is false, errors on the primary node are
3943
  ignored.
3944

3945
  """
3946
  all_result = True
3947
  for disk in instance.disks:
3948
    for node, top_disk in disk.ComputeNodeTree(instance.primary_node):
3949
      lu.cfg.SetDiskID(top_disk, node)
3950
      result = lu.rpc.call_blockdev_shutdown(node, top_disk)
3951
      msg = result.fail_msg
3952
      if msg:
3953
        lu.LogWarning("Could not shutdown block device %s on node %s: %s",
3954
                      disk.iv_name, node, msg)
3955
        if not ignore_primary or node != instance.primary_node:
3956
          all_result = False
3957
  return all_result
3958

    
3959

    
3960
def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
3961
  """Checks if a node has enough free memory.
3962

3963
  This function check if a given node has the needed amount of free
3964
  memory. In case the node has less memory or we cannot get the
3965
  information from the node, this function raise an OpPrereqError
3966
  exception.
3967

3968
  @type lu: C{LogicalUnit}
3969
  @param lu: a logical unit from which we get configuration data
3970
  @type node: C{str}
3971
  @param node: the node to check
3972
  @type reason: C{str}
3973
  @param reason: string to use in the error message
3974
  @type requested: C{int}
3975
  @param requested: the amount of memory in MiB to check for
3976
  @type hypervisor_name: C{str}
3977
  @param hypervisor_name: the hypervisor to ask for memory stats
3978
  @raise errors.OpPrereqError: if the node doesn't have enough memory, or
3979
      we cannot check the node
3980

3981
  """
3982
  nodeinfo = lu.rpc.call_node_info([node], lu.cfg.GetVGName(), hypervisor_name)
3983
  nodeinfo[node].Raise("Can't get data from node %s" % node,
3984
                       prereq=True, ecode=errors.ECODE_ENVIRON)
3985
  free_mem = nodeinfo[node].payload.get('memory_free', None)
3986
  if not isinstance(free_mem, int):
3987
    raise errors.OpPrereqError("Can't compute free memory on node %s, result"
3988
                               " was '%s'" % (node, free_mem),
3989
                               errors.ECODE_ENVIRON)
3990
  if requested > free_mem:
3991
    raise errors.OpPrereqError("Not enough memory on node %s for %s:"
3992
                               " needed %s MiB, available %s MiB" %
3993
                               (node, reason, requested, free_mem),
3994
                               errors.ECODE_NORES)
3995

    
3996

    
3997
def _CheckNodesFreeDisk(lu, nodenames, requested):
3998
  """Checks if nodes have enough free disk space in the default VG.
3999

4000
  This function check if all given nodes have the needed amount of
4001
  free disk. In case any node has less disk or we cannot get the
4002
  information from the node, this function raise an OpPrereqError
4003
  exception.
4004

4005
  @type lu: C{LogicalUnit}
4006
  @param lu: a logical unit from which we get configuration data
4007
  @type nodenames: C{list}
4008
  @param nodenames: the list of node names to check
4009
  @type requested: C{int}
4010
  @param requested: the amount of disk in MiB to check for
4011
  @raise errors.OpPrereqError: if the node doesn't have enough disk, or
4012
      we cannot check the node
4013

4014
  """
4015
  nodeinfo = lu.rpc.call_node_info(nodenames, lu.cfg.GetVGName(),
4016
                                   lu.cfg.GetHypervisorType())
4017
  for node in nodenames:
4018
    info = nodeinfo[node]
4019
    info.Raise("Cannot get current information from node %s" % node,
4020
               prereq=True, ecode=errors.ECODE_ENVIRON)
4021
    vg_free = info.payload.get("vg_free", None)
4022
    if not isinstance(vg_free, int):
4023
      raise errors.OpPrereqError("Can't compute free disk space on node %s,"
4024
                                 " result was '%s'" % (node, vg_free),
4025
                                 errors.ECODE_ENVIRON)
4026
    if requested > vg_free:
4027
      raise errors.OpPrereqError("Not enough disk space on target node %s:"
4028
                                 " required %d MiB, available %d MiB" %
4029
                                 (node, requested, vg_free),
4030
                                 errors.ECODE_NORES)
4031

    
4032

    
4033
class LUStartupInstance(LogicalUnit):
4034
  """Starts an instance.
4035

4036
  """
4037
  HPATH = "instance-start"
4038
  HTYPE = constants.HTYPE_INSTANCE
4039
  _OP_REQP = ["instance_name", "force"]
4040
  REQ_BGL = False
4041

    
4042
  def ExpandNames(self):
4043
    self._ExpandAndLockInstance()
4044

    
4045
  def BuildHooksEnv(self):
4046
    """Build hooks env.
4047

4048
    This runs on master, primary and secondary nodes of the instance.
4049

4050
    """
4051
    env = {
4052
      "FORCE": self.op.force,
4053
      }
4054
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4055
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4056
    return env, nl, nl
4057

    
4058
  def CheckPrereq(self):
4059
    """Check prerequisites.
4060

4061
    This checks that the instance is in the cluster.
4062

4063
    """
4064
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4065
    assert self.instance is not None, \
4066
      "Cannot retrieve locked instance %s" % self.op.instance_name
4067

    
4068
    # extra beparams
4069
    self.beparams = getattr(self.op, "beparams", {})
4070
    if self.beparams:
4071
      if not isinstance(self.beparams, dict):
4072
        raise errors.OpPrereqError("Invalid beparams passed: %s, expected"
4073
                                   " dict" % (type(self.beparams), ),
4074
                                   errors.ECODE_INVAL)
4075
      # fill the beparams dict
4076
      utils.ForceDictType(self.beparams, constants.BES_PARAMETER_TYPES)
4077
      self.op.beparams = self.beparams
4078

    
4079
    # extra hvparams
4080
    self.hvparams = getattr(self.op, "hvparams", {})
4081
    if self.hvparams:
4082
      if not isinstance(self.hvparams, dict):
4083
        raise errors.OpPrereqError("Invalid hvparams passed: %s, expected"
4084
                                   " dict" % (type(self.hvparams), ),
4085
                                   errors.ECODE_INVAL)
4086

    
4087
      # check hypervisor parameter syntax (locally)
4088
      cluster = self.cfg.GetClusterInfo()
4089
      utils.ForceDictType(self.hvparams, constants.HVS_PARAMETER_TYPES)
4090
      filled_hvp = objects.FillDict(cluster.hvparams[instance.hypervisor],
4091
                                    instance.hvparams)
4092
      filled_hvp.update(self.hvparams)
4093
      hv_type = hypervisor.GetHypervisor(instance.hypervisor)
4094
      hv_type.CheckParameterSyntax(filled_hvp)
4095
      _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
4096
      self.op.hvparams = self.hvparams
4097

    
4098
    _CheckNodeOnline(self, instance.primary_node)
4099

    
4100
    bep = self.cfg.GetClusterInfo().FillBE(instance)
4101
    # check bridges existence
4102
    _CheckInstanceBridgesExist(self, instance)
4103

    
4104
    remote_info = self.rpc.call_instance_info(instance.primary_node,
4105
                                              instance.name,
4106
                                              instance.hypervisor)
4107
    remote_info.Raise("Error checking node %s" % instance.primary_node,
4108
                      prereq=True, ecode=errors.ECODE_ENVIRON)
4109
    if not remote_info.payload: # not running already
4110
      _CheckNodeFreeMemory(self, instance.primary_node,
4111
                           "starting instance %s" % instance.name,
4112
                           bep[constants.BE_MEMORY], instance.hypervisor)
4113

    
4114
  def Exec(self, feedback_fn):
4115
    """Start the instance.
4116

4117
    """
4118
    instance = self.instance
4119
    force = self.op.force
4120

    
4121
    self.cfg.MarkInstanceUp(instance.name)
4122

    
4123
    node_current = instance.primary_node
4124

    
4125
    _StartInstanceDisks(self, instance, force)
4126

    
4127
    result = self.rpc.call_instance_start(node_current, instance,
4128
                                          self.hvparams, self.beparams)
4129
    msg = result.fail_msg
4130
    if msg:
4131
      _ShutdownInstanceDisks(self, instance)
4132
      raise errors.OpExecError("Could not start instance: %s" % msg)
4133

    
4134

    
4135
class LURebootInstance(LogicalUnit):
4136
  """Reboot an instance.
4137

4138
  """
4139
  HPATH = "instance-reboot"
4140
  HTYPE = constants.HTYPE_INSTANCE
4141
  _OP_REQP = ["instance_name", "ignore_secondaries", "reboot_type"]
4142
  REQ_BGL = False
4143

    
4144
  def CheckArguments(self):
4145
    """Check the arguments.
4146

4147
    """
4148
    self.shutdown_timeout = getattr(self.op, "shutdown_timeout",
4149
                                    constants.DEFAULT_SHUTDOWN_TIMEOUT)
4150

    
4151
  def ExpandNames(self):
4152
    if self.op.reboot_type not in [constants.INSTANCE_REBOOT_SOFT,
4153
                                   constants.INSTANCE_REBOOT_HARD,
4154
                                   constants.INSTANCE_REBOOT_FULL]:
4155
      raise errors.ParameterError("reboot type not in [%s, %s, %s]" %
4156
                                  (constants.INSTANCE_REBOOT_SOFT,
4157
                                   constants.INSTANCE_REBOOT_HARD,
4158
                                   constants.INSTANCE_REBOOT_FULL))
4159
    self._ExpandAndLockInstance()
4160

    
4161
  def BuildHooksEnv(self):
4162
    """Build hooks env.
4163

4164
    This runs on master, primary and secondary nodes of the instance.
4165

4166
    """
4167
    env = {
4168
      "IGNORE_SECONDARIES": self.op.ignore_secondaries,
4169
      "REBOOT_TYPE": self.op.reboot_type,
4170
      "SHUTDOWN_TIMEOUT": self.shutdown_timeout,
4171
      }
4172
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
4173
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4174
    return env, nl, nl
4175

    
4176
  def CheckPrereq(self):
4177
    """Check prerequisites.
4178

4179
    This checks that the instance is in the cluster.
4180

4181
    """
4182
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4183
    assert self.instance is not None, \
4184
      "Cannot retrieve locked instance %s" % self.op.instance_name
4185

    
4186
    _CheckNodeOnline(self, instance.primary_node)
4187

    
4188
    # check bridges existence
4189
    _CheckInstanceBridgesExist(self, instance)
4190

    
4191
  def Exec(self, feedback_fn):
4192
    """Reboot the instance.
4193

4194
    """
4195
    instance = self.instance
4196
    ignore_secondaries = self.op.ignore_secondaries
4197
    reboot_type = self.op.reboot_type
4198

    
4199
    node_current = instance.primary_node
4200

    
4201
    if reboot_type in [constants.INSTANCE_REBOOT_SOFT,
4202
                       constants.INSTANCE_REBOOT_HARD]:
4203
      for disk in instance.disks:
4204
        self.cfg.SetDiskID(disk, node_current)
4205
      result = self.rpc.call_instance_reboot(node_current, instance,
4206
                                             reboot_type,
4207
                                             self.shutdown_timeout)
4208
      result.Raise("Could not reboot instance")
4209
    else:
4210
      result = self.rpc.call_instance_shutdown(node_current, instance,
4211
                                               self.shutdown_timeout)
4212
      result.Raise("Could not shutdown instance for full reboot")
4213
      _ShutdownInstanceDisks(self, instance)
4214
      _StartInstanceDisks(self, instance, ignore_secondaries)
4215
      result = self.rpc.call_instance_start(node_current, instance, None, None)
4216
      msg = result.fail_msg
4217
      if msg:
4218
        _ShutdownInstanceDisks(self, instance)
4219
        raise errors.OpExecError("Could not start instance for"
4220
                                 " full reboot: %s" % msg)
4221

    
4222
    self.cfg.MarkInstanceUp(instance.name)
4223

    
4224

    
4225
class LUShutdownInstance(LogicalUnit):
4226
  """Shutdown an instance.
4227

4228
  """
4229
  HPATH = "instance-stop"
4230
  HTYPE = constants.HTYPE_INSTANCE
4231
  _OP_REQP = ["instance_name"]
4232
  REQ_BGL = False
4233

    
4234
  def CheckArguments(self):
4235
    """Check the arguments.
4236

4237
    """
4238
    self.timeout = getattr(self.op, "timeout",
4239
                           constants.DEFAULT_SHUTDOWN_TIMEOUT)
4240

    
4241
  def ExpandNames(self):
4242
    self._ExpandAndLockInstance()
4243

    
4244
  def BuildHooksEnv(self):
4245
    """Build hooks env.
4246

4247
    This runs on master, primary and secondary nodes of the instance.
4248

4249
    """
4250
    env = _BuildInstanceHookEnvByObject(self, self.instance)
4251
    env["TIMEOUT"] = self.timeout
4252
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4253
    return env, nl, nl
4254

    
4255
  def CheckPrereq(self):
4256
    """Check prerequisites.
4257

4258
    This checks that the instance is in the cluster.
4259

4260
    """
4261
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4262
    assert self.instance is not None, \
4263
      "Cannot retrieve locked instance %s" % self.op.instance_name
4264
    _CheckNodeOnline(self, self.instance.primary_node)
4265

    
4266
  def Exec(self, feedback_fn):
4267
    """Shutdown the instance.
4268

4269
    """
4270
    instance = self.instance
4271
    node_current = instance.primary_node
4272
    timeout = self.timeout
4273
    self.cfg.MarkInstanceDown(instance.name)
4274
    result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
4275
    msg = result.fail_msg
4276
    if msg:
4277
      self.proc.LogWarning("Could not shutdown instance: %s" % msg)
4278

    
4279
    _ShutdownInstanceDisks(self, instance)
4280

    
4281

    
4282
class LUReinstallInstance(LogicalUnit):
4283
  """Reinstall an instance.
4284

4285
  """
4286
  HPATH = "instance-reinstall"
4287
  HTYPE = constants.HTYPE_INSTANCE
4288
  _OP_REQP = ["instance_name"]
4289
  REQ_BGL = False
4290

    
4291
  def ExpandNames(self):
4292
    self._ExpandAndLockInstance()
4293

    
4294
  def BuildHooksEnv(self):
4295
    """Build hooks env.
4296

4297
    This runs on master, primary and secondary nodes of the instance.
4298

4299
    """
4300
    env = _BuildInstanceHookEnvByObject(self, self.instance)
4301
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4302
    return env, nl, nl
4303

    
4304
  def CheckPrereq(self):
4305
    """Check prerequisites.
4306

4307
    This checks that the instance is in the cluster and is not running.
4308

4309
    """
4310
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4311
    assert instance is not None, \
4312
      "Cannot retrieve locked instance %s" % self.op.instance_name
4313
    _CheckNodeOnline(self, instance.primary_node)
4314

    
4315
    if instance.disk_template == constants.DT_DISKLESS:
4316
      raise errors.OpPrereqError("Instance '%s' has no disks" %
4317
                                 self.op.instance_name,
4318
                                 errors.ECODE_INVAL)
4319
    _CheckInstanceDown(self, instance, "cannot reinstall")
4320

    
4321
    self.op.os_type = getattr(self.op, "os_type", None)
4322
    self.op.force_variant = getattr(self.op, "force_variant", False)
4323
    if self.op.os_type is not None:
4324
      # OS verification
4325
      pnode = _ExpandNodeName(self.cfg, instance.primary_node)
4326
      _CheckNodeHasOS(self, pnode, self.op.os_type, self.op.force_variant)
4327

    
4328
    self.instance = instance
4329

    
4330
  def Exec(self, feedback_fn):
4331
    """Reinstall the instance.
4332

4333
    """
4334
    inst = self.instance
4335

    
4336
    if self.op.os_type is not None:
4337
      feedback_fn("Changing OS to '%s'..." % self.op.os_type)
4338
      inst.os = self.op.os_type
4339
      self.cfg.Update(inst, feedback_fn)
4340

    
4341
    _StartInstanceDisks(self, inst, None)
4342
    try:
4343
      feedback_fn("Running the instance OS create scripts...")
4344
      # FIXME: pass debug option from opcode to backend
4345
      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
4346
                                             self.op.debug_level)
4347
      result.Raise("Could not install OS for instance %s on node %s" %
4348
                   (inst.name, inst.primary_node))
4349
    finally:
4350
      _ShutdownInstanceDisks(self, inst)
4351

    
4352

    
4353
class LURecreateInstanceDisks(LogicalUnit):
4354
  """Recreate an instance's missing disks.
4355

4356
  """
4357
  HPATH = "instance-recreate-disks"
4358
  HTYPE = constants.HTYPE_INSTANCE
4359
  _OP_REQP = ["instance_name", "disks"]
4360
  REQ_BGL = False
4361

    
4362
  def CheckArguments(self):
4363
    """Check the arguments.
4364

4365
    """
4366
    if not isinstance(self.op.disks, list):
4367
      raise errors.OpPrereqError("Invalid disks parameter", errors.ECODE_INVAL)
4368
    for item in self.op.disks:
4369
      if (not isinstance(item, int) or
4370
          item < 0):
4371
        raise errors.OpPrereqError("Invalid disk specification '%s'" %
4372
                                   str(item), errors.ECODE_INVAL)
4373

    
4374
  def ExpandNames(self):
4375
    self._ExpandAndLockInstance()
4376

    
4377
  def BuildHooksEnv(self):
4378
    """Build hooks env.
4379

4380
    This runs on master, primary and secondary nodes of the instance.
4381

4382
    """
4383
    env = _BuildInstanceHookEnvByObject(self, self.instance)
4384
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4385
    return env, nl, nl
4386

    
4387
  def CheckPrereq(self):
4388
    """Check prerequisites.
4389

4390
    This checks that the instance is in the cluster and is not running.
4391

4392
    """
4393
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4394
    assert instance is not None, \
4395
      "Cannot retrieve locked instance %s" % self.op.instance_name
4396
    _CheckNodeOnline(self, instance.primary_node)
4397

    
4398
    if instance.disk_template == constants.DT_DISKLESS:
4399
      raise errors.OpPrereqError("Instance '%s' has no disks" %
4400
                                 self.op.instance_name, errors.ECODE_INVAL)
4401
    _CheckInstanceDown(self, instance, "cannot recreate disks")
4402

    
4403
    if not self.op.disks:
4404
      self.op.disks = range(len(instance.disks))
4405
    else:
4406
      for idx in self.op.disks:
4407
        if idx >= len(instance.disks):
4408
          raise errors.OpPrereqError("Invalid disk index passed '%s'" % idx,
4409
                                     errors.ECODE_INVAL)
4410

    
4411
    self.instance = instance
4412

    
4413
  def Exec(self, feedback_fn):
4414
    """Recreate the disks.
4415

4416
    """
4417
    to_skip = []
4418
    for idx, _ in enumerate(self.instance.disks):
4419
      if idx not in self.op.disks: # disk idx has not been passed in
4420
        to_skip.append(idx)
4421
        continue
4422

    
4423
    _CreateDisks(self, self.instance, to_skip=to_skip)
4424

    
4425

    
4426
class LURenameInstance(LogicalUnit):
4427
  """Rename an instance.
4428

4429
  """
4430
  HPATH = "instance-rename"
4431
  HTYPE = constants.HTYPE_INSTANCE
4432
  _OP_REQP = ["instance_name", "new_name"]
4433

    
4434
  def BuildHooksEnv(self):
4435
    """Build hooks env.
4436

4437
    This runs on master, primary and secondary nodes of the instance.
4438

4439
    """
4440
    env = _BuildInstanceHookEnvByObject(self, self.instance)
4441
    env["INSTANCE_NEW_NAME"] = self.op.new_name
4442
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
4443
    return env, nl, nl
4444

    
4445
  def CheckPrereq(self):
4446
    """Check prerequisites.
4447

4448
    This checks that the instance is in the cluster and is not running.
4449

4450
    """
4451
    self.op.instance_name = _ExpandInstanceName(self.cfg,
4452
                                                self.op.instance_name)
4453
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4454
    assert instance is not None
4455
    _CheckNodeOnline(self, instance.primary_node)
4456
    _CheckInstanceDown(self, instance, "cannot rename")
4457
    self.instance = instance
4458

    
4459
    # new name verification
4460
    name_info = utils.GetHostInfo(self.op.new_name)
4461

    
4462
    self.op.new_name = new_name = name_info.name
4463
    instance_list = self.cfg.GetInstanceList()
4464
    if new_name in instance_list:
4465
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
4466
                                 new_name, errors.ECODE_EXISTS)
4467

    
4468
    if not getattr(self.op, "ignore_ip", False):
4469
      if utils.TcpPing(name_info.ip, constants.DEFAULT_NODED_PORT):
4470
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
4471
                                   (name_info.ip, new_name),
4472
                                   errors.ECODE_NOTUNIQUE)
4473

    
4474

    
4475
  def Exec(self, feedback_fn):
4476
    """Reinstall the instance.
4477

4478
    """
4479
    inst = self.instance
4480
    old_name = inst.name
4481

    
4482
    if inst.disk_template == constants.DT_FILE:
4483
      old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
4484

    
4485
    self.cfg.RenameInstance(inst.name, self.op.new_name)
4486
    # Change the instance lock. This is definitely safe while we hold the BGL
4487
    self.context.glm.remove(locking.LEVEL_INSTANCE, old_name)
4488
    self.context.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
4489

    
4490
    # re-read the instance from the configuration after rename
4491
    inst = self.cfg.GetInstanceInfo(self.op.new_name)
4492

    
4493
    if inst.disk_template == constants.DT_FILE:
4494
      new_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
4495
      result = self.rpc.call_file_storage_dir_rename(inst.primary_node,
4496
                                                     old_file_storage_dir,
4497
                                                     new_file_storage_dir)
4498
      result.Raise("Could not rename on node %s directory '%s' to '%s'"
4499
                   " (but the instance has been renamed in Ganeti)" %
4500
                   (inst.primary_node, old_file_storage_dir,
4501
                    new_file_storage_dir))
4502

    
4503
    _StartInstanceDisks(self, inst, None)
4504
    try:
4505
      result = self.rpc.call_instance_run_rename(inst.primary_node, inst,
4506
                                                 old_name, self.op.debug_level)
4507
      msg = result.fail_msg
4508
      if msg:
4509
        msg = ("Could not run OS rename script for instance %s on node %s"
4510
               " (but the instance has been renamed in Ganeti): %s" %
4511
               (inst.name, inst.primary_node, msg))
4512
        self.proc.LogWarning(msg)
4513
    finally:
4514
      _ShutdownInstanceDisks(self, inst)
4515

    
4516

    
4517
class LURemoveInstance(LogicalUnit):
4518
  """Remove an instance.
4519

4520
  """
4521
  HPATH = "instance-remove"
4522
  HTYPE = constants.HTYPE_INSTANCE
4523
  _OP_REQP = ["instance_name", "ignore_failures"]
4524
  REQ_BGL = False
4525

    
4526
  def CheckArguments(self):
4527
    """Check the arguments.
4528

4529
    """
4530
    self.shutdown_timeout = getattr(self.op, "shutdown_timeout",
4531
                                    constants.DEFAULT_SHUTDOWN_TIMEOUT)
4532

    
4533
  def ExpandNames(self):
4534
    self._ExpandAndLockInstance()
4535
    self.needed_locks[locking.LEVEL_NODE] = []
4536
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4537

    
4538
  def DeclareLocks(self, level):
4539
    if level == locking.LEVEL_NODE:
4540
      self._LockInstancesNodes()
4541

    
4542
  def BuildHooksEnv(self):
4543
    """Build hooks env.
4544

4545
    This runs on master, primary and secondary nodes of the instance.
4546

4547
    """
4548
    env = _BuildInstanceHookEnvByObject(self, self.instance)
4549
    env["SHUTDOWN_TIMEOUT"] = self.shutdown_timeout
4550
    nl = [self.cfg.GetMasterNode()]
4551
    nl_post = list(self.instance.all_nodes) + nl
4552
    return env, nl, nl_post
4553

    
4554
  def CheckPrereq(self):
4555
    """Check prerequisites.
4556

4557
    This checks that the instance is in the cluster.
4558

4559
    """
4560
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4561
    assert self.instance is not None, \
4562
      "Cannot retrieve locked instance %s" % self.op.instance_name
4563

    
4564
  def Exec(self, feedback_fn):
4565
    """Remove the instance.
4566

4567
    """
4568
    instance = self.instance
4569
    logging.info("Shutting down instance %s on node %s",
4570
                 instance.name, instance.primary_node)
4571

    
4572
    result = self.rpc.call_instance_shutdown(instance.primary_node, instance,
4573
                                             self.shutdown_timeout)
4574
    msg = result.fail_msg
4575
    if msg:
4576
      if self.op.ignore_failures:
4577
        feedback_fn("Warning: can't shutdown instance: %s" % msg)
4578
      else:
4579
        raise errors.OpExecError("Could not shutdown instance %s on"
4580
                                 " node %s: %s" %
4581
                                 (instance.name, instance.primary_node, msg))
4582

    
4583
    logging.info("Removing block devices for instance %s", instance.name)
4584

    
4585
    if not _RemoveDisks(self, instance):
4586
      if self.op.ignore_failures:
4587
        feedback_fn("Warning: can't remove instance's disks")
4588
      else:
4589
        raise errors.OpExecError("Can't remove instance's disks")
4590

    
4591
    logging.info("Removing instance %s out of cluster config", instance.name)
4592

    
4593
    self.cfg.RemoveInstance(instance.name)
4594
    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
4595

    
4596

    
4597
class LUQueryInstances(NoHooksLU):
4598
  """Logical unit for querying instances.
4599

4600
  """
4601
  # pylint: disable-msg=W0142
4602
  _OP_REQP = ["output_fields", "names", "use_locking"]
4603
  REQ_BGL = False
4604
  _SIMPLE_FIELDS = ["name", "os", "network_port", "hypervisor",
4605
                    "serial_no", "ctime", "mtime", "uuid"]
4606
  _FIELDS_STATIC = utils.FieldSet(*["name", "os", "pnode", "snodes",
4607
                                    "admin_state",
4608
                                    "disk_template", "ip", "mac", "bridge",
4609
                                    "nic_mode", "nic_link",
4610
                                    "sda_size", "sdb_size", "vcpus", "tags",
4611
                                    "network_port", "beparams",
4612
                                    r"(disk)\.(size)/([0-9]+)",
4613
                                    r"(disk)\.(sizes)", "disk_usage",
4614
                                    r"(nic)\.(mac|ip|mode|link)/([0-9]+)",
4615
                                    r"(nic)\.(bridge)/([0-9]+)",
4616
                                    r"(nic)\.(macs|ips|modes|links|bridges)",
4617
                                    r"(disk|nic)\.(count)",
4618
                                    "hvparams",
4619
                                    ] + _SIMPLE_FIELDS +
4620
                                  ["hv/%s" % name
4621
                                   for name in constants.HVS_PARAMETERS
4622
                                   if name not in constants.HVC_GLOBALS] +
4623
                                  ["be/%s" % name
4624
                                   for name in constants.BES_PARAMETERS])
4625
  _FIELDS_DYNAMIC = utils.FieldSet("oper_state", "oper_ram", "status")
4626

    
4627

    
4628
  def ExpandNames(self):
4629
    _CheckOutputFields(static=self._FIELDS_STATIC,
4630
                       dynamic=self._FIELDS_DYNAMIC,
4631
                       selected=self.op.output_fields)
4632

    
4633
    self.needed_locks = {}
4634
    self.share_locks[locking.LEVEL_INSTANCE] = 1
4635
    self.share_locks[locking.LEVEL_NODE] = 1
4636

    
4637
    if self.op.names:
4638
      self.wanted = _GetWantedInstances(self, self.op.names)
4639
    else:
4640
      self.wanted = locking.ALL_SET
4641

    
4642
    self.do_node_query = self._FIELDS_STATIC.NonMatching(self.op.output_fields)
4643
    self.do_locking = self.do_node_query and self.op.use_locking
4644
    if self.do_locking:
4645
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
4646
      self.needed_locks[locking.LEVEL_NODE] = []
4647
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4648

    
4649
  def DeclareLocks(self, level):
4650
    if level == locking.LEVEL_NODE and self.do_locking:
4651
      self._LockInstancesNodes()
4652

    
4653
  def CheckPrereq(self):
4654
    """Check prerequisites.
4655

4656
    """
4657
    pass
4658

    
4659
  def Exec(self, feedback_fn):
4660
    """Computes the list of nodes and their attributes.
4661

4662
    """
4663
    # pylint: disable-msg=R0912
4664
    # way too many branches here
4665
    all_info = self.cfg.GetAllInstancesInfo()
4666
    if self.wanted == locking.ALL_SET:
4667
      # caller didn't specify instance names, so ordering is not important
4668
      if self.do_locking:
4669
        instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
4670
      else:
4671
        instance_names = all_info.keys()
4672
      instance_names = utils.NiceSort(instance_names)
4673
    else:
4674
      # caller did specify names, so we must keep the ordering
4675
      if self.do_locking:
4676
        tgt_set = self.acquired_locks[locking.LEVEL_INSTANCE]
4677
      else:
4678
        tgt_set = all_info.keys()
4679
      missing = set(self.wanted).difference(tgt_set)
4680
      if missing:
4681
        raise errors.OpExecError("Some instances were removed before"
4682
                                 " retrieving their data: %s" % missing)
4683
      instance_names = self.wanted
4684

    
4685
    instance_list = [all_info[iname] for iname in instance_names]
4686

    
4687
    # begin data gathering
4688

    
4689
    nodes = frozenset([inst.primary_node for inst in instance_list])
4690
    hv_list = list(set([inst.hypervisor for inst in instance_list]))
4691

    
4692
    bad_nodes = []
4693
    off_nodes = []
4694
    if self.do_node_query:
4695
      live_data = {}
4696
      node_data = self.rpc.call_all_instances_info(nodes, hv_list)
4697
      for name in nodes:
4698
        result = node_data[name]
4699
        if result.offline:
4700
          # offline nodes will be in both lists
4701
          off_nodes.append(name)
4702
        if result.fail_msg:
4703
          bad_nodes.append(name)
4704
        else:
4705
          if result.payload:
4706
            live_data.update(result.payload)
4707
          # else no instance is alive
4708
    else:
4709
      live_data = dict([(name, {}) for name in instance_names])
4710

    
4711
    # end data gathering
4712

    
4713
    HVPREFIX = "hv/"
4714
    BEPREFIX = "be/"
4715
    output = []
4716
    cluster = self.cfg.GetClusterInfo()
4717
    for instance in instance_list:
4718
      iout = []
4719
      i_hv = cluster.FillHV(instance, skip_globals=True)
4720
      i_be = cluster.FillBE(instance)
4721
      i_nicp = [objects.FillDict(cluster.nicparams[constants.PP_DEFAULT],
4722
                                 nic.nicparams) for nic in instance.nics]
4723
      for field in self.op.output_fields:
4724
        st_match = self._FIELDS_STATIC.Matches(field)
4725
        if field in self._SIMPLE_FIELDS:
4726
          val = getattr(instance, field)
4727
        elif field == "pnode":
4728
          val = instance.primary_node
4729
        elif field == "snodes":
4730
          val = list(instance.secondary_nodes)
4731
        elif field == "admin_state":
4732
          val = instance.admin_up
4733
        elif field == "oper_state":
4734
          if instance.primary_node in bad_nodes:
4735
            val = None
4736
          else:
4737
            val = bool(live_data.get(instance.name))
4738
        elif field == "status":
4739
          if instance.primary_node in off_nodes:
4740
            val = "ERROR_nodeoffline"
4741
          elif instance.primary_node in bad_nodes:
4742
            val = "ERROR_nodedown"
4743
          else:
4744
            running = bool(live_data.get(instance.name))
4745
            if running:
4746
              if instance.admin_up:
4747
                val = "running"
4748
              else:
4749
                val = "ERROR_up"
4750
            else:
4751
              if instance.admin_up:
4752
                val = "ERROR_down"
4753
              else:
4754
                val = "ADMIN_down"
4755
        elif field == "oper_ram":
4756
          if instance.primary_node in bad_nodes:
4757
            val = None
4758
          elif instance.name in live_data:
4759
            val = live_data[instance.name].get("memory", "?")
4760
          else:
4761
            val = "-"
4762
        elif field == "vcpus":
4763
          val = i_be[constants.BE_VCPUS]
4764
        elif field == "disk_template":
4765
          val = instance.disk_template
4766
        elif field == "ip":
4767
          if instance.nics:
4768
            val = instance.nics[0].ip
4769
          else:
4770
            val = None
4771
        elif field == "nic_mode":
4772
          if instance.nics:
4773
            val = i_nicp[0][constants.NIC_MODE]
4774
          else:
4775
            val = None
4776
        elif field == "nic_link":
4777
          if instance.nics:
4778
            val = i_nicp[0][constants.NIC_LINK]
4779
          else:
4780
            val = None
4781
        elif field == "bridge":
4782
          if (instance.nics and
4783
              i_nicp[0][constants.NIC_MODE] == constants.NIC_MODE_BRIDGED):
4784
            val = i_nicp[0][constants.NIC_LINK]
4785
          else:
4786
            val = None
4787
        elif field == "mac":
4788
          if instance.nics:
4789
            val = instance.nics[0].mac
4790
          else:
4791
            val = None
4792
        elif field == "sda_size" or field == "sdb_size":
4793
          idx = ord(field[2]) - ord('a')
4794
          try:
4795
            val = instance.FindDisk(idx).size
4796
          except errors.OpPrereqError:
4797
            val = None
4798
        elif field == "disk_usage": # total disk usage per node
4799
          disk_sizes = [{'size': disk.size} for disk in instance.disks]
4800
          val = _ComputeDiskSize(instance.disk_template, disk_sizes)
4801
        elif field == "tags":
4802
          val = list(instance.GetTags())
4803
        elif field == "hvparams":
4804
          val = i_hv
4805
        elif (field.startswith(HVPREFIX) and
4806
              field[len(HVPREFIX):] in constants.HVS_PARAMETERS and
4807
              field[len(HVPREFIX):] not in constants.HVC_GLOBALS):
4808
          val = i_hv.get(field[len(HVPREFIX):], None)
4809
        elif field == "beparams":
4810
          val = i_be
4811
        elif (field.startswith(BEPREFIX) and
4812
              field[len(BEPREFIX):] in constants.BES_PARAMETERS):
4813
          val = i_be.get(field[len(BEPREFIX):], None)
4814
        elif st_match and st_match.groups():
4815
          # matches a variable list
4816
          st_groups = st_match.groups()
4817
          if st_groups and st_groups[0] == "disk":
4818
            if st_groups[1] == "count":
4819
              val = len(instance.disks)
4820
            elif st_groups[1] == "sizes":
4821
              val = [disk.size for disk in instance.disks]
4822
            elif st_groups[1] == "size":
4823
              try:
4824
                val = instance.FindDisk(st_groups[2]).size
4825
              except errors.OpPrereqError:
4826
                val = None
4827
            else:
4828
              assert False, "Unhandled disk parameter"
4829
          elif st_groups[0] == "nic":
4830
            if st_groups[1] == "count":
4831
              val = len(instance.nics)
4832
            elif st_groups[1] == "macs":
4833
              val = [nic.mac for nic in instance.nics]
4834
            elif st_groups[1] == "ips":
4835
              val = [nic.ip for nic in instance.nics]
4836
            elif st_groups[1] == "modes":
4837
              val = [nicp[constants.NIC_MODE] for nicp in i_nicp]
4838
            elif st_groups[1] == "links":
4839
              val = [nicp[constants.NIC_LINK] for nicp in i_nicp]
4840
            elif st_groups[1] == "bridges":
4841
              val = []
4842
              for nicp in i_nicp:
4843
                if nicp[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
4844
                  val.append(nicp[constants.NIC_LINK])
4845
                else:
4846
                  val.append(None)
4847
            else:
4848
              # index-based item
4849
              nic_idx = int(st_groups[2])
4850
              if nic_idx >= len(instance.nics):
4851
                val = None
4852
              else:
4853
                if st_groups[1] == "mac":
4854
                  val = instance.nics[nic_idx].mac
4855
                elif st_groups[1] == "ip":
4856
                  val = instance.nics[nic_idx].ip
4857
                elif st_groups[1] == "mode":
4858
                  val = i_nicp[nic_idx][constants.NIC_MODE]
4859
                elif st_groups[1] == "link":
4860
                  val = i_nicp[nic_idx][constants.NIC_LINK]
4861
                elif st_groups[1] == "bridge":
4862
                  nic_mode = i_nicp[nic_idx][constants.NIC_MODE]
4863
                  if nic_mode == constants.NIC_MODE_BRIDGED:
4864
                    val = i_nicp[nic_idx][constants.NIC_LINK]
4865
                  else:
4866
                    val = None
4867
                else:
4868
                  assert False, "Unhandled NIC parameter"
4869
          else:
4870
            assert False, ("Declared but unhandled variable parameter '%s'" %
4871
                           field)
4872
        else:
4873
          assert False, "Declared but unhandled parameter '%s'" % field
4874
        iout.append(val)
4875
      output.append(iout)
4876

    
4877
    return output
4878

    
4879

    
4880
class LUFailoverInstance(LogicalUnit):
4881
  """Failover an instance.
4882

4883
  """
4884
  HPATH = "instance-failover"
4885
  HTYPE = constants.HTYPE_INSTANCE
4886
  _OP_REQP = ["instance_name", "ignore_consistency"]
4887
  REQ_BGL = False
4888

    
4889
  def CheckArguments(self):
4890
    """Check the arguments.
4891

4892
    """
4893
    self.shutdown_timeout = getattr(self.op, "shutdown_timeout",
4894
                                    constants.DEFAULT_SHUTDOWN_TIMEOUT)
4895

    
4896
  def ExpandNames(self):
4897
    self._ExpandAndLockInstance()
4898
    self.needed_locks[locking.LEVEL_NODE] = []
4899
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
4900

    
4901
  def DeclareLocks(self, level):
4902
    if level == locking.LEVEL_NODE:
4903
      self._LockInstancesNodes()
4904

    
4905
  def BuildHooksEnv(self):
4906
    """Build hooks env.
4907

4908
    This runs on master, primary and secondary nodes of the instance.
4909

4910
    """
4911
    instance = self.instance
4912
    source_node = instance.primary_node
4913
    target_node = instance.secondary_nodes[0]
4914
    env = {
4915
      "IGNORE_CONSISTENCY": self.op.ignore_consistency,
4916
      "SHUTDOWN_TIMEOUT": self.shutdown_timeout,
4917
      "OLD_PRIMARY": source_node,
4918
      "OLD_SECONDARY": target_node,
4919
      "NEW_PRIMARY": target_node,
4920
      "NEW_SECONDARY": source_node,
4921
      }
4922
    env.update(_BuildInstanceHookEnvByObject(self, instance))
4923
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
4924
    nl_post = list(nl)
4925
    nl_post.append(source_node)
4926
    return env, nl, nl_post
4927

    
4928
  def CheckPrereq(self):
4929
    """Check prerequisites.
4930

4931
    This checks that the instance is in the cluster.
4932

4933
    """
4934
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
4935
    assert self.instance is not None, \
4936
      "Cannot retrieve locked instance %s" % self.op.instance_name
4937

    
4938
    bep = self.cfg.GetClusterInfo().FillBE(instance)
4939
    if instance.disk_template not in constants.DTS_NET_MIRROR:
4940
      raise errors.OpPrereqError("Instance's disk layout is not"
4941
                                 " network mirrored, cannot failover.",
4942
                                 errors.ECODE_STATE)
4943

    
4944
    secondary_nodes = instance.secondary_nodes
4945
    if not secondary_nodes:
4946
      raise errors.ProgrammerError("no secondary node but using "
4947
                                   "a mirrored disk template")
4948

    
4949
    target_node = secondary_nodes[0]
4950
    _CheckNodeOnline(self, target_node)
4951
    _CheckNodeNotDrained(self, target_node)
4952
    if instance.admin_up:
4953
      # check memory requirements on the secondary node
4954
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
4955
                           instance.name, bep[constants.BE_MEMORY],
4956
                           instance.hypervisor)
4957
    else:
4958
      self.LogInfo("Not checking memory on the secondary node as"
4959
                   " instance will not be started")
4960

    
4961
    # check bridge existance
4962
    _CheckInstanceBridgesExist(self, instance, node=target_node)
4963

    
4964
  def Exec(self, feedback_fn):
4965
    """Failover an instance.
4966

4967
    The failover is done by shutting it down on its present node and
4968
    starting it on the secondary.
4969

4970
    """
4971
    instance = self.instance
4972

    
4973
    source_node = instance.primary_node
4974
    target_node = instance.secondary_nodes[0]
4975

    
4976
    if instance.admin_up:
4977
      feedback_fn("* checking disk consistency between source and target")
4978
      for dev in instance.disks:
4979
        # for drbd, these are drbd over lvm
4980
        if not _CheckDiskConsistency(self, dev, target_node, False):
4981
          if not self.op.ignore_consistency:
4982
            raise errors.OpExecError("Disk %s is degraded on target node,"
4983
                                     " aborting failover." % dev.iv_name)
4984
    else:
4985
      feedback_fn("* not checking disk consistency as instance is not running")
4986

    
4987
    feedback_fn("* shutting down instance on source node")
4988
    logging.info("Shutting down instance %s on node %s",
4989
                 instance.name, source_node)
4990

    
4991
    result = self.rpc.call_instance_shutdown(source_node, instance,
4992
                                             self.shutdown_timeout)
4993
    msg = result.fail_msg
4994
    if msg:
4995
      if self.op.ignore_consistency:
4996
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
4997
                             " Proceeding anyway. Please make sure node"
4998
                             " %s is down. Error details: %s",
4999
                             instance.name, source_node, source_node, msg)
5000
      else:
5001
        raise errors.OpExecError("Could not shutdown instance %s on"
5002
                                 " node %s: %s" %
5003
                                 (instance.name, source_node, msg))
5004

    
5005
    feedback_fn("* deactivating the instance's disks on source node")
5006
    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
5007
      raise errors.OpExecError("Can't shut down the instance's disks.")
5008

    
5009
    instance.primary_node = target_node
5010
    # distribute new instance config to the other nodes
5011
    self.cfg.Update(instance, feedback_fn)
5012

    
5013
    # Only start the instance if it's marked as up
5014
    if instance.admin_up:
5015
      feedback_fn("* activating the instance's disks on target node")
5016
      logging.info("Starting instance %s on node %s",
5017
                   instance.name, target_node)
5018

    
5019
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
5020
                                               ignore_secondaries=True)
5021
      if not disks_ok:
5022
        _ShutdownInstanceDisks(self, instance)
5023
        raise errors.OpExecError("Can't activate the instance's disks")
5024

    
5025
      feedback_fn("* starting the instance on the target node")
5026
      result = self.rpc.call_instance_start(target_node, instance, None, None)
5027
      msg = result.fail_msg
5028
      if msg:
5029
        _ShutdownInstanceDisks(self, instance)
5030
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
5031
                                 (instance.name, target_node, msg))
5032

    
5033

    
5034
class LUMigrateInstance(LogicalUnit):
5035
  """Migrate an instance.
5036

5037
  This is migration without shutting down, compared to the failover,
5038
  which is done with shutdown.
5039

5040
  """
5041
  HPATH = "instance-migrate"
5042
  HTYPE = constants.HTYPE_INSTANCE
5043
  _OP_REQP = ["instance_name", "live", "cleanup"]
5044

    
5045
  REQ_BGL = False
5046

    
5047
  def ExpandNames(self):
5048
    self._ExpandAndLockInstance()
5049

    
5050
    self.needed_locks[locking.LEVEL_NODE] = []
5051
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
5052

    
5053
    self._migrater = TLMigrateInstance(self, self.op.instance_name,
5054
                                       self.op.live, self.op.cleanup)
5055
    self.tasklets = [self._migrater]
5056

    
5057
  def DeclareLocks(self, level):
5058
    if level == locking.LEVEL_NODE:
5059
      self._LockInstancesNodes()
5060

    
5061
  def BuildHooksEnv(self):
5062
    """Build hooks env.
5063

5064
    This runs on master, primary and secondary nodes of the instance.
5065

5066
    """
5067
    instance = self._migrater.instance
5068
    source_node = instance.primary_node
5069
    target_node = instance.secondary_nodes[0]
5070
    env = _BuildInstanceHookEnvByObject(self, instance)
5071
    env["MIGRATE_LIVE"] = self.op.live
5072
    env["MIGRATE_CLEANUP"] = self.op.cleanup
5073
    env.update({
5074
        "OLD_PRIMARY": source_node,
5075
        "OLD_SECONDARY": target_node,
5076
        "NEW_PRIMARY": target_node,
5077
        "NEW_SECONDARY": source_node,
5078
        })
5079
    nl = [self.cfg.GetMasterNode()] + list(instance.secondary_nodes)
5080
    nl_post = list(nl)
5081
    nl_post.append(source_node)
5082
    return env, nl, nl_post
5083

    
5084

    
5085
class LUMoveInstance(LogicalUnit):
5086
  """Move an instance by data-copying.
5087

5088
  """
5089
  HPATH = "instance-move"
5090
  HTYPE = constants.HTYPE_INSTANCE
5091
  _OP_REQP = ["instance_name", "target_node"]
5092
  REQ_BGL = False
5093

    
5094
  def CheckArguments(self):
5095
    """Check the arguments.
5096

5097
    """
5098
    self.shutdown_timeout = getattr(self.op, "shutdown_timeout",
5099
                                    constants.DEFAULT_SHUTDOWN_TIMEOUT)
5100

    
5101
  def ExpandNames(self):
5102
    self._ExpandAndLockInstance()
5103
    target_node = _ExpandNodeName(self.cfg, self.op.target_node)
5104
    self.op.target_node = target_node
5105
    self.needed_locks[locking.LEVEL_NODE] = [target_node]
5106
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5107

    
5108
  def DeclareLocks(self, level):
5109
    if level == locking.LEVEL_NODE:
5110
      self._LockInstancesNodes(primary_only=True)
5111

    
5112
  def BuildHooksEnv(self):
5113
    """Build hooks env.
5114

5115
    This runs on master, primary and secondary nodes of the instance.
5116

5117
    """
5118
    env = {
5119
      "TARGET_NODE": self.op.target_node,
5120
      "SHUTDOWN_TIMEOUT": self.shutdown_timeout,
5121
      }
5122
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
5123
    nl = [self.cfg.GetMasterNode()] + [self.instance.primary_node,
5124
                                       self.op.target_node]
5125
    return env, nl, nl
5126

    
5127
  def CheckPrereq(self):
5128
    """Check prerequisites.
5129

5130
    This checks that the instance is in the cluster.
5131

5132
    """
5133
    self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
5134
    assert self.instance is not None, \
5135
      "Cannot retrieve locked instance %s" % self.op.instance_name
5136

    
5137
    node = self.cfg.GetNodeInfo(self.op.target_node)
5138
    assert node is not None, \
5139
      "Cannot retrieve locked node %s" % self.op.target_node
5140

    
5141
    self.target_node = target_node = node.name
5142

    
5143
    if target_node == instance.primary_node:
5144
      raise errors.OpPrereqError("Instance %s is already on the node %s" %
5145
                                 (instance.name, target_node),
5146
                                 errors.ECODE_STATE)
5147

    
5148
    bep = self.cfg.GetClusterInfo().FillBE(instance)
5149

    
5150
    for idx, dsk in enumerate(instance.disks):
5151
      if dsk.dev_type not in (constants.LD_LV, constants.LD_FILE):
5152
        raise errors.OpPrereqError("Instance disk %d has a complex layout,"
5153
                                   " cannot copy" % idx, errors.ECODE_STATE)
5154

    
5155
    _CheckNodeOnline(self, target_node)
5156
    _CheckNodeNotDrained(self, target_node)
5157

    
5158
    if instance.admin_up:
5159
      # check memory requirements on the secondary node
5160
      _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
5161
                           instance.name, bep[constants.BE_MEMORY],
5162
                           instance.hypervisor)
5163
    else:
5164
      self.LogInfo("Not checking memory on the secondary node as"
5165
                   " instance will not be started")
5166

    
5167
    # check bridge existance
5168
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5169

    
5170
  def Exec(self, feedback_fn):
5171
    """Move an instance.
5172

5173
    The move is done by shutting it down on its present node, copying
5174
    the data over (slow) and starting it on the new node.
5175

5176
    """
5177
    instance = self.instance
5178

    
5179
    source_node = instance.primary_node
5180
    target_node = self.target_node
5181

    
5182
    self.LogInfo("Shutting down instance %s on source node %s",
5183
                 instance.name, source_node)
5184

    
5185
    result = self.rpc.call_instance_shutdown(source_node, instance,
5186
                                             self.shutdown_timeout)
5187
    msg = result.fail_msg
5188
    if msg:
5189
      if self.op.ignore_consistency:
5190
        self.proc.LogWarning("Could not shutdown instance %s on node %s."
5191
                             " Proceeding anyway. Please make sure node"
5192
                             " %s is down. Error details: %s",
5193
                             instance.name, source_node, source_node, msg)
5194
      else:
5195
        raise errors.OpExecError("Could not shutdown instance %s on"
5196
                                 " node %s: %s" %
5197
                                 (instance.name, source_node, msg))
5198

    
5199
    # create the target disks
5200
    try:
5201
      _CreateDisks(self, instance, target_node=target_node)
5202
    except errors.OpExecError:
5203
      self.LogWarning("Device creation failed, reverting...")
5204
      try:
5205
        _RemoveDisks(self, instance, target_node=target_node)
5206
      finally:
5207
        self.cfg.ReleaseDRBDMinors(instance.name)
5208
        raise
5209

    
5210
    cluster_name = self.cfg.GetClusterInfo().cluster_name
5211

    
5212
    errs = []
5213
    # activate, get path, copy the data over
5214
    for idx, disk in enumerate(instance.disks):
5215
      self.LogInfo("Copying data for disk %d", idx)
5216
      result = self.rpc.call_blockdev_assemble(target_node, disk,
5217
                                               instance.name, True)
5218
      if result.fail_msg:
5219
        self.LogWarning("Can't assemble newly created disk %d: %s",
5220
                        idx, result.fail_msg)
5221
        errs.append(result.fail_msg)
5222
        break
5223
      dev_path = result.payload
5224
      result = self.rpc.call_blockdev_export(source_node, disk,
5225
                                             target_node, dev_path,
5226
                                             cluster_name)
5227
      if result.fail_msg:
5228
        self.LogWarning("Can't copy data over for disk %d: %s",
5229
                        idx, result.fail_msg)
5230
        errs.append(result.fail_msg)
5231
        break
5232

    
5233
    if errs:
5234
      self.LogWarning("Some disks failed to copy, aborting")
5235
      try:
5236
        _RemoveDisks(self, instance, target_node=target_node)
5237
      finally:
5238
        self.cfg.ReleaseDRBDMinors(instance.name)
5239
        raise errors.OpExecError("Errors during disk copy: %s" %
5240
                                 (",".join(errs),))
5241

    
5242
    instance.primary_node = target_node
5243
    self.cfg.Update(instance, feedback_fn)
5244

    
5245
    self.LogInfo("Removing the disks on the original node")
5246
    _RemoveDisks(self, instance, target_node=source_node)
5247

    
5248
    # Only start the instance if it's marked as up
5249
    if instance.admin_up:
5250
      self.LogInfo("Starting instance %s on node %s",
5251
                   instance.name, target_node)
5252

    
5253
      disks_ok, _ = _AssembleInstanceDisks(self, instance,
5254
                                           ignore_secondaries=True)
5255
      if not disks_ok:
5256
        _ShutdownInstanceDisks(self, instance)
5257
        raise errors.OpExecError("Can't activate the instance's disks")
5258

    
5259
      result = self.rpc.call_instance_start(target_node, instance, None, None)
5260
      msg = result.fail_msg
5261
      if msg:
5262
        _ShutdownInstanceDisks(self, instance)
5263
        raise errors.OpExecError("Could not start instance %s on node %s: %s" %
5264
                                 (instance.name, target_node, msg))
5265

    
5266

    
5267
class LUMigrateNode(LogicalUnit):
5268
  """Migrate all instances from a node.
5269

5270
  """
5271
  HPATH = "node-migrate"
5272
  HTYPE = constants.HTYPE_NODE
5273
  _OP_REQP = ["node_name", "live"]
5274
  REQ_BGL = False
5275

    
5276
  def ExpandNames(self):
5277
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
5278

    
5279
    self.needed_locks = {
5280
      locking.LEVEL_NODE: [self.op.node_name],
5281
      }
5282

    
5283
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
5284

    
5285
    # Create tasklets for migrating instances for all instances on this node
5286
    names = []
5287
    tasklets = []
5288

    
5289
    for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name):
5290
      logging.debug("Migrating instance %s", inst.name)
5291
      names.append(inst.name)
5292

    
5293
      tasklets.append(TLMigrateInstance(self, inst.name, self.op.live, False))
5294

    
5295
    self.tasklets = tasklets
5296

    
5297
    # Declare instance locks
5298
    self.needed_locks[locking.LEVEL_INSTANCE] = names
5299

    
5300
  def DeclareLocks(self, level):
5301
    if level == locking.LEVEL_NODE:
5302
      self._LockInstancesNodes()
5303

    
5304
  def BuildHooksEnv(self):
5305
    """Build hooks env.
5306

5307
    This runs on the master, the primary and all the secondaries.
5308

5309
    """
5310
    env = {
5311
      "NODE_NAME": self.op.node_name,
5312
      }
5313

    
5314
    nl = [self.cfg.GetMasterNode()]
5315

    
5316
    return (env, nl, nl)
5317

    
5318

    
5319
class TLMigrateInstance(Tasklet):
5320
  def __init__(self, lu, instance_name, live, cleanup):
5321
    """Initializes this class.
5322

5323
    """
5324
    Tasklet.__init__(self, lu)
5325

    
5326
    # Parameters
5327
    self.instance_name = instance_name
5328
    self.live = live
5329
    self.cleanup = cleanup
5330

    
5331
  def CheckPrereq(self):
5332
    """Check prerequisites.
5333

5334
    This checks that the instance is in the cluster.
5335

5336
    """
5337
    instance_name = _ExpandInstanceName(self.lu.cfg, self.instance_name)
5338
    instance = self.cfg.GetInstanceInfo(instance_name)
5339
    assert instance is not None
5340

    
5341
    if instance.disk_template != constants.DT_DRBD8:
5342
      raise errors.OpPrereqError("Instance's disk layout is not"
5343
                                 " drbd8, cannot migrate.", errors.ECODE_STATE)
5344

    
5345
    secondary_nodes = instance.secondary_nodes
5346
    if not secondary_nodes:
5347
      raise errors.ConfigurationError("No secondary node but using"
5348
                                      " drbd8 disk template")
5349

    
5350
    i_be = self.cfg.GetClusterInfo().FillBE(instance)
5351

    
5352
    target_node = secondary_nodes[0]
5353
    # check memory requirements on the secondary node
5354
    _CheckNodeFreeMemory(self, target_node, "migrating instance %s" %
5355
                         instance.name, i_be[constants.BE_MEMORY],
5356
                         instance.hypervisor)
5357

    
5358
    # check bridge existance
5359
    _CheckInstanceBridgesExist(self, instance, node=target_node)
5360

    
5361
    if not self.cleanup:
5362
      _CheckNodeNotDrained(self, target_node)
5363
      result = self.rpc.call_instance_migratable(instance.primary_node,
5364
                                                 instance)
5365
      result.Raise("Can't migrate, please use failover",
5366
                   prereq=True, ecode=errors.ECODE_STATE)
5367

    
5368
    self.instance = instance
5369

    
5370
  def _WaitUntilSync(self):
5371
    """Poll with custom rpc for disk sync.
5372

5373
    This uses our own step-based rpc call.
5374

5375
    """
5376
    self.feedback_fn("* wait until resync is done")
5377
    all_done = False
5378
    while not all_done:
5379
      all_done = True
5380
      result = self.rpc.call_drbd_wait_sync(self.all_nodes,
5381
                                            self.nodes_ip,
5382
                                            self.instance.disks)
5383
      min_percent = 100
5384
      for node, nres in result.items():
5385
        nres.Raise("Cannot resync disks on node %s" % node)
5386
        node_done, node_percent = nres.payload
5387
        all_done = all_done and node_done
5388
        if node_percent is not None:
5389
          min_percent = min(min_percent, node_percent)
5390
      if not all_done:
5391
        if min_percent < 100:
5392
          self.feedback_fn("   - progress: %.1f%%" % min_percent)
5393
        time.sleep(2)
5394

    
5395
  def _EnsureSecondary(self, node):
5396
    """Demote a node to secondary.
5397

5398
    """
5399
    self.feedback_fn("* switching node %s to secondary mode" % node)
5400

    
5401
    for dev in self.instance.disks:
5402
      self.cfg.SetDiskID(dev, node)
5403

    
5404
    result = self.rpc.call_blockdev_close(node, self.instance.name,
5405
                                          self.instance.disks)
5406
    result.Raise("Cannot change disk to secondary on node %s" % node)
5407

    
5408
  def _GoStandalone(self):
5409
    """Disconnect from the network.
5410

5411
    """
5412
    self.feedback_fn("* changing into standalone mode")
5413
    result = self.rpc.call_drbd_disconnect_net(self.all_nodes, self.nodes_ip,
5414
                                               self.instance.disks)
5415
    for node, nres in result.items():
5416
      nres.Raise("Cannot disconnect disks node %s" % node)
5417

    
5418
  def _GoReconnect(self, multimaster):
5419
    """Reconnect to the network.
5420

5421
    """
5422
    if multimaster:
5423
      msg = "dual-master"
5424
    else:
5425
      msg = "single-master"
5426
    self.feedback_fn("* changing disks into %s mode" % msg)
5427
    result = self.rpc.call_drbd_attach_net(self.all_nodes, self.nodes_ip,
5428
                                           self.instance.disks,
5429
                                           self.instance.name, multimaster)
5430
    for node, nres in result.items():
5431
      nres.Raise("Cannot change disks config on node %s" % node)
5432

    
5433
  def _ExecCleanup(self):
5434
    """Try to cleanup after a failed migration.
5435

5436
    The cleanup is done by:
5437
      - check that the instance is running only on one node
5438
        (and update the config if needed)
5439
      - change disks on its secondary node to secondary
5440
      - wait until disks are fully synchronized
5441
      - disconnect from the network
5442
      - change disks into single-master mode
5443
      - wait again until disks are fully synchronized
5444

5445
    """
5446
    instance = self.instance
5447
    target_node = self.target_node
5448
    source_node = self.source_node
5449

    
5450
    # check running on only one node
5451
    self.feedback_fn("* checking where the instance actually runs"
5452
                     " (if this hangs, the hypervisor might be in"
5453
                     " a bad state)")
5454
    ins_l = self.rpc.call_instance_list(self.all_nodes, [instance.hypervisor])
5455
    for node, result in ins_l.items():
5456
      result.Raise("Can't contact node %s" % node)
5457

    
5458
    runningon_source = instance.name in ins_l[source_node].payload
5459
    runningon_target = instance.name in ins_l[target_node].payload
5460

    
5461
    if runningon_source and runningon_target:
5462
      raise errors.OpExecError("Instance seems to be running on two nodes,"
5463
                               " or the hypervisor is confused. You will have"
5464
                               " to ensure manually that it runs only on one"
5465
                               " and restart this operation.")
5466

    
5467
    if not (runningon_source or runningon_target):
5468
      raise errors.OpExecError("Instance does not seem to be running at all."
5469
                               " In this case, it's safer to repair by"
5470
                               " running 'gnt-instance stop' to ensure disk"
5471
                               " shutdown, and then restarting it.")
5472

    
5473
    if runningon_target:
5474
      # the migration has actually succeeded, we need to update the config
5475
      self.feedback_fn("* instance running on secondary node (%s),"
5476
                       " updating config" % target_node)
5477
      instance.primary_node = target_node
5478
      self.cfg.Update(instance, self.feedback_fn)
5479
      demoted_node = source_node
5480
    else:
5481
      self.feedback_fn("* instance confirmed to be running on its"
5482
                       " primary node (%s)" % source_node)
5483
      demoted_node = target_node
5484

    
5485
    self._EnsureSecondary(demoted_node)
5486
    try:
5487
      self._WaitUntilSync()
5488
    except errors.OpExecError:
5489
      # we ignore here errors, since if the device is standalone, it
5490
      # won't be able to sync
5491
      pass
5492
    self._GoStandalone()
5493
    self._GoReconnect(False)
5494
    self._WaitUntilSync()
5495

    
5496
    self.feedback_fn("* done")
5497

    
5498
  def _RevertDiskStatus(self):
5499
    """Try to revert the disk status after a failed migration.
5500

5501
    """
5502
    target_node = self.target_node
5503
    try:
5504
      self._EnsureSecondary(target_node)
5505
      self._GoStandalone()
5506
      self._GoReconnect(False)
5507
      self._WaitUntilSync()
5508
    except errors.OpExecError, err:
5509
      self.lu.LogWarning("Migration failed and I can't reconnect the"
5510
                         " drives: error '%s'\n"
5511
                         "Please look and recover the instance status" %
5512
                         str(err))
5513

    
5514
  def _AbortMigration(self):
5515
    """Call the hypervisor code to abort a started migration.
5516

5517
    """
5518
    instance = self.instance
5519
    target_node = self.target_node
5520
    migration_info = self.migration_info
5521

    
5522
    abort_result = self.rpc.call_finalize_migration(target_node,
5523
                                                    instance,
5524
                                                    migration_info,
5525
                                                    False)
5526
    abort_msg = abort_result.fail_msg
5527
    if abort_msg:
5528
      logging.error("Aborting migration failed on target node %s: %s",
5529
                    target_node, abort_msg)
5530
      # Don't raise an exception here, as we stil have to try to revert the
5531
      # disk status, even if this step failed.
5532

    
5533
  def _ExecMigration(self):
5534
    """Migrate an instance.
5535

5536
    The migrate is done by:
5537
      - change the disks into dual-master mode
5538
      - wait until disks are fully synchronized again
5539
      - migrate the instance
5540
      - change disks on the new secondary node (the old primary) to secondary
5541
      - wait until disks are fully synchronized
5542
      - change disks into single-master mode
5543

5544
    """
5545
    instance = self.instance
5546
    target_node = self.target_node
5547
    source_node = self.source_node
5548

    
5549
    self.feedback_fn("* checking disk consistency between source and target")
5550
    for dev in instance.disks:
5551
      if not _CheckDiskConsistency(self, dev, target_node, False):
5552
        raise errors.OpExecError("Disk %s is degraded or not fully"
5553
                                 " synchronized on target node,"
5554
                                 " aborting migrate." % dev.iv_name)
5555

    
5556
    # First get the migration information from the remote node
5557
    result = self.rpc.call_migration_info(source_node, instance)
5558
    msg = result.fail_msg
5559
    if msg:
5560
      log_err = ("Failed fetching source migration information from %s: %s" %
5561
                 (source_node, msg))
5562
      logging.error(log_err)
5563
      raise errors.OpExecError(log_err)
5564

    
5565
    self.migration_info = migration_info = result.payload
5566

    
5567
    # Then switch the disks to master/master mode
5568
    self._EnsureSecondary(target_node)
5569
    self._GoStandalone()
5570
    self._GoReconnect(True)
5571
    self._WaitUntilSync()
5572

    
5573
    self.feedback_fn("* preparing %s to accept the instance" % target_node)
5574
    result = self.rpc.call_accept_instance(target_node,
5575
                                           instance,
5576
                                           migration_info,
5577
                                           self.nodes_ip[target_node])
5578

    
5579
    msg = result.fail_msg
5580
    if msg:
5581
      logging.error("Instance pre-migration failed, trying to revert"
5582
                    " disk status: %s", msg)
5583
      self.feedback_fn("Pre-migration failed, aborting")
5584
      self._AbortMigration()
5585
      self._RevertDiskStatus()
5586
      raise errors.OpExecError("Could not pre-migrate instance %s: %s" %
5587
                               (instance.name, msg))
5588

    
5589
    self.feedback_fn("* migrating instance to %s" % target_node)
5590
    time.sleep(10)
5591
    result = self.rpc.call_instance_migrate(source_node, instance,
5592
                                            self.nodes_ip[target_node],
5593
                                            self.live)
5594
    msg = result.fail_msg
5595
    if msg:
5596
      logging.error("Instance migration failed, trying to revert"
5597
                    " disk status: %s", msg)
5598
      self.feedback_fn("Migration failed, aborting")
5599
      self._AbortMigration()
5600
      self._RevertDiskStatus()
5601
      raise errors.OpExecError("Could not migrate instance %s: %s" %
5602
                               (instance.name, msg))
5603
    time.sleep(10)
5604

    
5605
    instance.primary_node = target_node
5606
    # distribute new instance config to the other nodes
5607
    self.cfg.Update(instance, self.feedback_fn)
5608

    
5609
    result = self.rpc.call_finalize_migration(target_node,
5610
                                              instance,
5611
                                              migration_info,
5612
                                              True)
5613
    msg = result.fail_msg
5614
    if msg:
5615
      logging.error("Instance migration succeeded, but finalization failed:"
5616
                    " %s", msg)
5617
      raise errors.OpExecError("Could not finalize instance migration: %s" %
5618
                               msg)
5619

    
5620
    self._EnsureSecondary(source_node)
5621
    self._WaitUntilSync()
5622
    self._GoStandalone()
5623
    self._GoReconnect(False)
5624
    self._WaitUntilSync()
5625

    
5626
    self.feedback_fn("* done")
5627

    
5628
  def Exec(self, feedback_fn):
5629
    """Perform the migration.
5630

5631
    """
5632
    feedback_fn("Migrating instance %s" % self.instance.name)
5633

    
5634
    self.feedback_fn = feedback_fn
5635

    
5636
    self.source_node = self.instance.primary_node
5637
    self.target_node = self.instance.secondary_nodes[0]
5638
    self.all_nodes = [self.source_node, self.target_node]
5639
    self.nodes_ip = {
5640
      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
5641
      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
5642
      }
5643

    
5644
    if self.cleanup:
5645
      return self._ExecCleanup()
5646
    else:
5647
      return self._ExecMigration()
5648

    
5649

    
5650
def _CreateBlockDev(lu, node, instance, device, force_create,
5651
                    info, force_open):
5652
  """Create a tree of block devices on a given node.
5653

5654
  If this device type has to be created on secondaries, create it and
5655
  all its children.
5656

5657
  If not, just recurse to children keeping the same 'force' value.
5658

5659
  @param lu: the lu on whose behalf we execute
5660
  @param node: the node on which to create the device
5661
  @type instance: L{objects.Instance}
5662
  @param instance: the instance which owns the device
5663
  @type device: L{objects.Disk}
5664
  @param device: the device to create
5665
  @type force_create: boolean
5666
  @param force_create: whether to force creation of this device; this
5667
      will be change to True whenever we find a device which has
5668
      CreateOnSecondary() attribute
5669
  @param info: the extra 'metadata' we should attach to the device
5670
      (this will be represented as a LVM tag)
5671
  @type force_open: boolean
5672
  @param force_open: this parameter will be passes to the
5673
      L{backend.BlockdevCreate} function where it specifies
5674
      whether we run on primary or not, and it affects both
5675
      the child assembly and the device own Open() execution
5676

5677
  """
5678
  if device.CreateOnSecondary():
5679
    force_create = True
5680

    
5681
  if device.children:
5682
    for child in device.children:
5683
      _CreateBlockDev(lu, node, instance, child, force_create,
5684
                      info, force_open)
5685

    
5686
  if not force_create:
5687
    return
5688

    
5689
  _CreateSingleBlockDev(lu, node, instance, device, info, force_open)
5690

    
5691

    
5692
def _CreateSingleBlockDev(lu, node, instance, device, info, force_open):
5693
  """Create a single block device on a given node.
5694

5695
  This will not recurse over children of the device, so they must be
5696
  created in advance.
5697

5698
  @param lu: the lu on whose behalf we execute
5699
  @param node: the node on which to create the device
5700
  @type instance: L{objects.Instance}
5701
  @param instance: the instance which owns the device
5702
  @type device: L{objects.Disk}
5703
  @param device: the device to create
5704
  @param info: the extra 'metadata' we should attach to the device
5705
      (this will be represented as a LVM tag)
5706
  @type force_open: boolean
5707
  @param force_open: this parameter will be passes to the
5708
      L{backend.BlockdevCreate} function where it specifies
5709
      whether we run on primary or not, and it affects both
5710
      the child assembly and the device own Open() execution
5711

5712
  """
5713
  lu.cfg.SetDiskID(device, node)
5714
  result = lu.rpc.call_blockdev_create(node, device, device.size,
5715
                                       instance.name, force_open, info)
5716
  result.Raise("Can't create block device %s on"
5717
               " node %s for instance %s" % (device, node, instance.name))
5718
  if device.physical_id is None:
5719
    device.physical_id = result.payload
5720

    
5721

    
5722
def _GenerateUniqueNames(lu, exts):
5723
  """Generate a suitable LV name.
5724

5725
  This will generate a logical volume name for the given instance.
5726

5727
  """
5728
  results = []
5729
  for val in exts:
5730
    new_id = lu.cfg.GenerateUniqueID(lu.proc.GetECId())
5731
    results.append("%s%s" % (new_id, val))
5732
  return results
5733

    
5734

    
5735
def _GenerateDRBD8Branch(lu, primary, secondary, size, names, iv_name,
5736
                         p_minor, s_minor):
5737
  """Generate a drbd8 device complete with its children.
5738

5739
  """
5740
  port = lu.cfg.AllocatePort()
5741
  vgname = lu.cfg.GetVGName()
5742
  shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
5743
  dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
5744
                          logical_id=(vgname, names[0]))
5745
  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
5746
                          logical_id=(vgname, names[1]))
5747
  drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
5748
                          logical_id=(primary, secondary, port,
5749
                                      p_minor, s_minor,
5750
                                      shared_secret),
5751
                          children=[dev_data, dev_meta],
5752
                          iv_name=iv_name)
5753
  return drbd_dev
5754

    
5755

    
5756
def _GenerateDiskTemplate(lu, template_name,
5757
                          instance_name, primary_node,
5758
                          secondary_nodes, disk_info,
5759
                          file_storage_dir, file_driver,
5760
                          base_index):
5761
  """Generate the entire disk layout for a given template type.
5762

5763
  """
5764
  #TODO: compute space requirements
5765

    
5766
  vgname = lu.cfg.GetVGName()
5767
  disk_count = len(disk_info)
5768
  disks = []
5769
  if template_name == constants.DT_DISKLESS:
5770
    pass
5771
  elif template_name == constants.DT_PLAIN:
5772
    if len(secondary_nodes) != 0:
5773
      raise errors.ProgrammerError("Wrong template configuration")
5774

    
5775
    names = _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
5776
                                      for i in range(disk_count)])
5777
    for idx, disk in enumerate(disk_info):
5778
      disk_index = idx + base_index
5779
      disk_dev = objects.Disk(dev_type=constants.LD_LV, size=disk["size"],
5780
                              logical_id=(vgname, names[idx]),
5781
                              iv_name="disk/%d" % disk_index,
5782
                              mode=disk["mode"])
5783
      disks.append(disk_dev)
5784
  elif template_name == constants.DT_DRBD8:
5785
    if len(secondary_nodes) != 1:
5786
      raise errors.ProgrammerError("Wrong template configuration")
5787
    remote_node = secondary_nodes[0]
5788
    minors = lu.cfg.AllocateDRBDMinor(
5789
      [primary_node, remote_node] * len(disk_info), instance_name)
5790

    
5791
    names = []
5792
    for lv_prefix in _GenerateUniqueNames(lu, [".disk%d" % (base_index + i)
5793
                                               for i in range(disk_count)]):
5794
      names.append(lv_prefix + "_data")
5795
      names.append(lv_prefix + "_meta")
5796
    for idx, disk in enumerate(disk_info):
5797
      disk_index = idx + base_index
5798
      disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
5799
                                      disk["size"], names[idx*2:idx*2+2],
5800
                                      "disk/%d" % disk_index,
5801
                                      minors[idx*2], minors[idx*2+1])
5802
      disk_dev.mode = disk["mode"]
5803
      disks.append(disk_dev)
5804
  elif template_name == constants.DT_FILE:
5805
    if len(secondary_nodes) != 0:
5806
      raise errors.ProgrammerError("Wrong template configuration")
5807

    
5808
    for idx, disk in enumerate(disk_info):
5809
      disk_index = idx + base_index
5810
      disk_dev = objects.Disk(dev_type=constants.LD_FILE, size=disk["size"],
5811
                              iv_name="disk/%d" % disk_index,
5812
                              logical_id=(file_driver,
5813
                                          "%s/disk%d" % (file_storage_dir,
5814
                                                         disk_index)),
5815
                              mode=disk["mode"])
5816
      disks.append(disk_dev)
5817
  else:
5818
    raise errors.ProgrammerError("Invalid disk template '%s'" % template_name)
5819
  return disks
5820

    
5821

    
5822
def _GetInstanceInfoText(instance):
5823
  """Compute that text that should be added to the disk's metadata.
5824

5825
  """
5826
  return "originstname+%s" % instance.name
5827

    
5828

    
5829
def _CreateDisks(lu, instance, to_skip=None, target_node=None):
5830
  """Create all disks for an instance.
5831

5832
  This abstracts away some work from AddInstance.
5833

5834
  @type lu: L{LogicalUnit}
5835
  @param lu: the logical unit on whose behalf we execute
5836
  @type instance: L{objects.Instance}
5837
  @param instance: the instance whose disks we should create
5838
  @type to_skip: list
5839
  @param to_skip: list of indices to skip
5840
  @type target_node: string
5841
  @param target_node: if passed, overrides the target node for creation
5842
  @rtype: boolean
5843
  @return: the success of the creation
5844

5845
  """
5846
  info = _GetInstanceInfoText(instance)
5847
  if target_node is None:
5848
    pnode = instance.primary_node
5849
    all_nodes = instance.all_nodes
5850
  else:
5851
    pnode = target_node
5852
    all_nodes = [pnode]
5853

    
5854
  if instance.disk_template == constants.DT_FILE:
5855
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
5856
    result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
5857

    
5858
    result.Raise("Failed to create directory '%s' on"
5859
                 " node %s" % (file_storage_dir, pnode))
5860

    
5861
  # Note: this needs to be kept in sync with adding of disks in
5862
  # LUSetInstanceParams
5863
  for idx, device in enumerate(instance.disks):
5864
    if to_skip and idx in to_skip:
5865
      continue
5866
    logging.info("Creating volume %s for instance %s",
5867
                 device.iv_name, instance.name)
5868
    #HARDCODE
5869
    for node in all_nodes:
5870
      f_create = node == pnode
5871
      _CreateBlockDev(lu, node, instance, device, f_create, info, f_create)
5872

    
5873

    
5874
def _RemoveDisks(lu, instance, target_node=None):
5875
  """Remove all disks for an instance.
5876

5877
  This abstracts away some work from `AddInstance()` and
5878
  `RemoveInstance()`. Note that in case some of the devices couldn't
5879
  be removed, the removal will continue with the other ones (compare
5880
  with `_CreateDisks()`).
5881

5882
  @type lu: L{LogicalUnit}
5883
  @param lu: the logical unit on whose behalf we execute
5884
  @type instance: L{objects.Instance}
5885
  @param instance: the instance whose disks we should remove
5886
  @type target_node: string
5887
  @param target_node: used to override the node on which to remove the disks
5888
  @rtype: boolean
5889
  @return: the success of the removal
5890

5891
  """
5892
  logging.info("Removing block devices for instance %s", instance.name)
5893

    
5894
  all_result = True
5895
  for device in instance.disks:
5896
    if target_node:
5897
      edata = [(target_node, device)]
5898
    else:
5899
      edata = device.ComputeNodeTree(instance.primary_node)
5900
    for node, disk in edata:
5901
      lu.cfg.SetDiskID(disk, node)
5902
      msg = lu.rpc.call_blockdev_remove(node, disk).fail_msg
5903
      if msg:
5904
        lu.LogWarning("Could not remove block device %s on node %s,"
5905
                      " continuing anyway: %s", device.iv_name, node, msg)
5906
        all_result = False
5907

    
5908
  if instance.disk_template == constants.DT_FILE:
5909
    file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
5910
    if target_node:
5911
      tgt = target_node
5912
    else:
5913
      tgt = instance.primary_node
5914
    result = lu.rpc.call_file_storage_dir_remove(tgt, file_storage_dir)
5915
    if result.fail_msg:
5916
      lu.LogWarning("Could not remove directory '%s' on node %s: %s",
5917
                    file_storage_dir, instance.primary_node, result.fail_msg)
5918
      all_result = False
5919

    
5920
  return all_result
5921

    
5922

    
5923
def _ComputeDiskSize(disk_template, disks):
5924
  """Compute disk size requirements in the volume group
5925

5926
  """
5927
  # Required free disk space as a function of disk and swap space
5928
  req_size_dict = {
5929
    constants.DT_DISKLESS: None,
5930
    constants.DT_PLAIN: sum(d["size"] for d in disks),
5931
    # 128 MB are added for drbd metadata for each disk
5932
    constants.DT_DRBD8: sum(d["size"] + 128 for d in disks),
5933
    constants.DT_FILE: None,
5934
  }
5935

    
5936
  if disk_template not in req_size_dict:
5937
    raise errors.ProgrammerError("Disk template '%s' size requirement"
5938
                                 " is unknown" %  disk_template)
5939

    
5940
  return req_size_dict[disk_template]
5941

    
5942

    
5943
def _CheckHVParams(lu, nodenames, hvname, hvparams):
5944
  """Hypervisor parameter validation.
5945

5946
  This function abstract the hypervisor parameter validation to be
5947
  used in both instance create and instance modify.
5948

5949
  @type lu: L{LogicalUnit}
5950
  @param lu: the logical unit for which we check
5951
  @type nodenames: list
5952
  @param nodenames: the list of nodes on which we should check
5953
  @type hvname: string
5954
  @param hvname: the name of the hypervisor we should use
5955
  @type hvparams: dict
5956
  @param hvparams: the parameters which we need to check
5957
  @raise errors.OpPrereqError: if the parameters are not valid
5958

5959
  """
5960
  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
5961
                                                  hvname,
5962
                                                  hvparams)
5963
  for node in nodenames:
5964
    info = hvinfo[node]
5965
    if info.offline:
5966
      continue
5967
    info.Raise("Hypervisor parameter validation failed on node %s" % node)
5968

    
5969

    
5970
class LUCreateInstance(LogicalUnit):
5971
  """Create an instance.
5972

5973
  """
5974
  HPATH = "instance-add"
5975
  HTYPE = constants.HTYPE_INSTANCE
5976
  _OP_REQP = ["instance_name", "disks", "disk_template",
5977
              "mode", "start",
5978
              "wait_for_sync", "ip_check", "nics",
5979
              "hvparams", "beparams"]
5980
  REQ_BGL = False
5981

    
5982
  def CheckArguments(self):
5983
    """Check arguments.
5984

5985
    """
5986
    # set optional parameters to none if they don't exist
5987
    for attr in ["pnode", "snode", "iallocator", "hypervisor"]:
5988
      if not hasattr(self.op, attr):
5989
        setattr(self.op, attr, None)
5990

    
5991
    # do not require name_check to ease forward/backward compatibility
5992
    # for tools
5993
    if not hasattr(self.op, "name_check"):
5994
      self.op.name_check = True
5995
    if not hasattr(self.op, "no_install"):
5996
      self.op.no_install = False
5997
    if self.op.no_install and self.op.start:
5998
      self.LogInfo("No-installation mode selected, disabling startup")
5999
      self.op.start = False
6000
    # validate/normalize the instance name
6001
    self.op.instance_name = utils.HostInfo.NormalizeName(self.op.instance_name)
6002
    if self.op.ip_check and not self.op.name_check:
6003
      # TODO: make the ip check more flexible and not depend on the name check
6004
      raise errors.OpPrereqError("Cannot do ip checks without a name check",
6005
                                 errors.ECODE_INVAL)
6006
    if (self.op.disk_template == constants.DT_FILE and
6007
        not constants.ENABLE_FILE_STORAGE):
6008
      raise errors.OpPrereqError("File storage disabled at configure time",
6009
                                 errors.ECODE_INVAL)
6010
    # check disk information: either all adopt, or no adopt
6011
    has_adopt = has_no_adopt = False
6012
    for disk in self.op.disks:
6013
      if "adopt" in disk:
6014
        has_adopt = True
6015
      else:
6016
        has_no_adopt = True
6017
    if has_adopt and has_no_adopt:
6018
      raise errors.OpPrereqError("Either all disks are adopted or none is",
6019
                                 errors.ECODE_INVAL)
6020
    if has_adopt:
6021
      if self.op.disk_template != constants.DT_PLAIN:
6022
        raise errors.OpPrereqError("Disk adoption is only supported for the"
6023
                                   " 'plain' disk template",
6024
                                   errors.ECODE_INVAL)
6025
      if self.op.iallocator is not None:
6026
        raise errors.OpPrereqError("Disk adoption not allowed with an"
6027
                                   " iallocator script", errors.ECODE_INVAL)
6028
      if self.op.mode == constants.INSTANCE_IMPORT:
6029
        raise errors.OpPrereqError("Disk adoption not allowed for"
6030
                                   " instance import", errors.ECODE_INVAL)
6031

    
6032
    self.adopt_disks = has_adopt
6033

    
6034
    # verify creation mode
6035
    if self.op.mode not in (constants.INSTANCE_CREATE,
6036
                            constants.INSTANCE_IMPORT):
6037
      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
6038
                                 self.op.mode, errors.ECODE_INVAL)
6039

    
6040
    # disk template
6041
    _CheckDiskTemplate(self.op.disk_template)
6042

    
6043
    # instance name verification
6044
    if self.op.name_check:
6045
      self.hostname1 = utils.GetHostInfo(self.op.instance_name)
6046
      self.op.instance_name = self.hostname1.name
6047
      # used in CheckPrereq for ip ping check
6048
      self.check_ip = self.hostname1.ip
6049
    else:
6050
      self.check_ip = None
6051

    
6052
    # file storage checks
6053
    if (self.op.file_driver and
6054
        not self.op.file_driver in constants.FILE_DRIVER):
6055
      raise errors.OpPrereqError("Invalid file driver name '%s'" %
6056
                                 self.op.file_driver, errors.ECODE_INVAL)
6057

    
6058
    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
6059
      raise errors.OpPrereqError("File storage directory path not absolute",
6060
                                 errors.ECODE_INVAL)
6061

    
6062
    ### Node/iallocator related checks
6063
    if [self.op.iallocator, self.op.pnode].count(None) != 1:
6064
      raise errors.OpPrereqError("One and only one of iallocator and primary"
6065
                                 " node must be given",
6066
                                 errors.ECODE_INVAL)
6067

    
6068
    if self.op.mode == constants.INSTANCE_IMPORT:
6069
      # On import force_variant must be True, because if we forced it at
6070
      # initial install, our only chance when importing it back is that it
6071
      # works again!
6072
      self.op.force_variant = True
6073

    
6074
      if self.op.no_install:
6075
        self.LogInfo("No-installation mode has no effect during import")
6076

    
6077
    else: # INSTANCE_CREATE
6078
      if getattr(self.op, "os_type", None) is None:
6079
        raise errors.OpPrereqError("No guest OS specified",
6080
                                   errors.ECODE_INVAL)
6081
      self.op.force_variant = getattr(self.op, "force_variant", False)
6082

    
6083
  def ExpandNames(self):
6084
    """ExpandNames for CreateInstance.
6085

6086
    Figure out the right locks for instance creation.
6087

6088
    """
6089
    self.needed_locks = {}
6090

    
6091
    instance_name = self.op.instance_name
6092
    # this is just a preventive check, but someone might still add this
6093
    # instance in the meantime, and creation will fail at lock-add time
6094
    if instance_name in self.cfg.GetInstanceList():
6095
      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
6096
                                 instance_name, errors.ECODE_EXISTS)
6097

    
6098
    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
6099

    
6100
    if self.op.iallocator:
6101
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6102
    else:
6103
      self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
6104
      nodelist = [self.op.pnode]
6105
      if self.op.snode is not None:
6106
        self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
6107
        nodelist.append(self.op.snode)
6108
      self.needed_locks[locking.LEVEL_NODE] = nodelist
6109

    
6110
    # in case of import lock the source node too
6111
    if self.op.mode == constants.INSTANCE_IMPORT:
6112
      src_node = getattr(self.op, "src_node", None)
6113
      src_path = getattr(self.op, "src_path", None)
6114

    
6115
      if src_path is None:
6116
        self.op.src_path = src_path = self.op.instance_name
6117

    
6118
      if src_node is None:
6119
        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6120
        self.op.src_node = None
6121
        if os.path.isabs(src_path):
6122
          raise errors.OpPrereqError("Importing an instance from an absolute"
6123
                                     " path requires a source node option.",
6124
                                     errors.ECODE_INVAL)
6125
      else:
6126
        self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
6127
        if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
6128
          self.needed_locks[locking.LEVEL_NODE].append(src_node)
6129
        if not os.path.isabs(src_path):
6130
          self.op.src_path = src_path = \
6131
            utils.PathJoin(constants.EXPORT_DIR, src_path)
6132

    
6133
  def _RunAllocator(self):
6134
    """Run the allocator based on input opcode.
6135

6136
    """
6137
    nics = [n.ToDict() for n in self.nics]
6138
    ial = IAllocator(self.cfg, self.rpc,
6139
                     mode=constants.IALLOCATOR_MODE_ALLOC,
6140
                     name=self.op.instance_name,
6141
                     disk_template=self.op.disk_template,
6142
                     tags=[],
6143
                     os=self.op.os_type,
6144
                     vcpus=self.be_full[constants.BE_VCPUS],
6145
                     mem_size=self.be_full[constants.BE_MEMORY],
6146
                     disks=self.disks,
6147
                     nics=nics,
6148
                     hypervisor=self.op.hypervisor,
6149
                     )
6150

    
6151
    ial.Run(self.op.iallocator)
6152

    
6153
    if not ial.success:
6154
      raise errors.OpPrereqError("Can't compute nodes using"
6155
                                 " iallocator '%s': %s" %
6156
                                 (self.op.iallocator, ial.info),
6157
                                 errors.ECODE_NORES)
6158
    if len(ial.result) != ial.required_nodes:
6159
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6160
                                 " of nodes (%s), required %s" %
6161
                                 (self.op.iallocator, len(ial.result),
6162
                                  ial.required_nodes), errors.ECODE_FAULT)
6163
    self.op.pnode = ial.result[0]
6164
    self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
6165
                 self.op.instance_name, self.op.iallocator,
6166
                 utils.CommaJoin(ial.result))
6167
    if ial.required_nodes == 2:
6168
      self.op.snode = ial.result[1]
6169

    
6170
  def BuildHooksEnv(self):
6171
    """Build hooks env.
6172

6173
    This runs on master, primary and secondary nodes of the instance.
6174

6175
    """
6176
    env = {
6177
      "ADD_MODE": self.op.mode,
6178
      }
6179
    if self.op.mode == constants.INSTANCE_IMPORT:
6180
      env["SRC_NODE"] = self.op.src_node
6181
      env["SRC_PATH"] = self.op.src_path
6182
      env["SRC_IMAGES"] = self.src_images
6183

    
6184
    env.update(_BuildInstanceHookEnv(
6185
      name=self.op.instance_name,
6186
      primary_node=self.op.pnode,
6187
      secondary_nodes=self.secondaries,
6188
      status=self.op.start,
6189
      os_type=self.op.os_type,
6190
      memory=self.be_full[constants.BE_MEMORY],
6191
      vcpus=self.be_full[constants.BE_VCPUS],
6192
      nics=_NICListToTuple(self, self.nics),
6193
      disk_template=self.op.disk_template,
6194
      disks=[(d["size"], d["mode"]) for d in self.disks],
6195
      bep=self.be_full,
6196
      hvp=self.hv_full,
6197
      hypervisor_name=self.op.hypervisor,
6198
    ))
6199

    
6200
    nl = ([self.cfg.GetMasterNode(), self.op.pnode] +
6201
          self.secondaries)
6202
    return env, nl, nl
6203

    
6204
  def _ReadExportInfo(self):
6205
    """Reads the export information from disk.
6206

6207
    It will override the opcode source node and path with the actual
6208
    information, if these two were not specified before.
6209

6210
    @return: the export information
6211

6212
    """
6213
    assert self.op.mode == constants.INSTANCE_IMPORT
6214

    
6215
    src_node = self.op.src_node
6216
    src_path = self.op.src_path
6217

    
6218
    if src_node is None:
6219
      locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
6220
      exp_list = self.rpc.call_export_list(locked_nodes)
6221
      found = False
6222
      for node in exp_list:
6223
        if exp_list[node].fail_msg:
6224
          continue
6225
        if src_path in exp_list[node].payload:
6226
          found = True
6227
          self.op.src_node = src_node = node
6228
          self.op.src_path = src_path = utils.PathJoin(constants.EXPORT_DIR,
6229
                                                       src_path)
6230
          break
6231
      if not found:
6232
        raise errors.OpPrereqError("No export found for relative path %s" %
6233
                                    src_path, errors.ECODE_INVAL)
6234

    
6235
    _CheckNodeOnline(self, src_node)
6236
    result = self.rpc.call_export_info(src_node, src_path)
6237
    result.Raise("No export or invalid export found in dir %s" % src_path)
6238

    
6239
    export_info = objects.SerializableConfigParser.Loads(str(result.payload))
6240
    if not export_info.has_section(constants.INISECT_EXP):
6241
      raise errors.ProgrammerError("Corrupted export config",
6242
                                   errors.ECODE_ENVIRON)
6243

    
6244
    ei_version = export_info.get(constants.INISECT_EXP, "version")
6245
    if (int(ei_version) != constants.EXPORT_VERSION):
6246
      raise errors.OpPrereqError("Wrong export version %s (wanted %d)" %
6247
                                 (ei_version, constants.EXPORT_VERSION),
6248
                                 errors.ECODE_ENVIRON)
6249
    return export_info
6250

    
6251
  def CheckPrereq(self):
6252
    """Check prerequisites.
6253

6254
    """
6255
    if self.op.mode == constants.INSTANCE_IMPORT:
6256
      export_info = self._ReadExportInfo()
6257

    
6258
    if (not self.cfg.GetVGName() and
6259
        self.op.disk_template not in constants.DTS_NOT_LVM):
6260
      raise errors.OpPrereqError("Cluster does not support lvm-based"
6261
                                 " instances", errors.ECODE_STATE)
6262

    
6263
    if self.op.hypervisor is None:
6264
      self.op.hypervisor = self.cfg.GetHypervisorType()
6265

    
6266
    cluster = self.cfg.GetClusterInfo()
6267
    enabled_hvs = cluster.enabled_hypervisors
6268
    if self.op.hypervisor not in enabled_hvs:
6269
      raise errors.OpPrereqError("Selected hypervisor (%s) not enabled in the"
6270
                                 " cluster (%s)" % (self.op.hypervisor,
6271
                                  ",".join(enabled_hvs)),
6272
                                 errors.ECODE_STATE)
6273

    
6274
    # check hypervisor parameter syntax (locally)
6275
    utils.ForceDictType(self.op.hvparams, constants.HVS_PARAMETER_TYPES)
6276
    filled_hvp = objects.FillDict(cluster.hvparams[self.op.hypervisor],
6277
                                  self.op.hvparams)
6278
    hv_type = hypervisor.GetHypervisor(self.op.hypervisor)
6279
    hv_type.CheckParameterSyntax(filled_hvp)
6280
    self.hv_full = filled_hvp
6281
    # check that we don't specify global parameters on an instance
6282
    _CheckGlobalHvParams(self.op.hvparams)
6283

    
6284
    # fill and remember the beparams dict
6285
    utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
6286
    self.be_full = objects.FillDict(cluster.beparams[constants.PP_DEFAULT],
6287
                                    self.op.beparams)
6288

    
6289
    # NIC buildup
6290
    self.nics = []
6291
    for idx, nic in enumerate(self.op.nics):
6292
      nic_mode_req = nic.get("mode", None)
6293
      nic_mode = nic_mode_req
6294
      if nic_mode is None:
6295
        nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
6296

    
6297
      # in routed mode, for the first nic, the default ip is 'auto'
6298
      if nic_mode == constants.NIC_MODE_ROUTED and idx == 0:
6299
        default_ip_mode = constants.VALUE_AUTO
6300
      else:
6301
        default_ip_mode = constants.VALUE_NONE
6302

    
6303
      # ip validity checks
6304
      ip = nic.get("ip", default_ip_mode)
6305
      if ip is None or ip.lower() == constants.VALUE_NONE:
6306
        nic_ip = None
6307
      elif ip.lower() == constants.VALUE_AUTO:
6308
        if not self.op.name_check:
6309
          raise errors.OpPrereqError("IP address set to auto but name checks"
6310
                                     " have been skipped. Aborting.",
6311
                                     errors.ECODE_INVAL)
6312
        nic_ip = self.hostname1.ip
6313
      else:
6314
        if not utils.IsValidIP(ip):
6315
          raise errors.OpPrereqError("Given IP address '%s' doesn't look"
6316
                                     " like a valid IP" % ip,
6317
                                     errors.ECODE_INVAL)
6318
        nic_ip = ip
6319

    
6320
      # TODO: check the ip address for uniqueness
6321
      if nic_mode == constants.NIC_MODE_ROUTED and not nic_ip:
6322
        raise errors.OpPrereqError("Routed nic mode requires an ip address",
6323
                                   errors.ECODE_INVAL)
6324

    
6325
      # MAC address verification
6326
      mac = nic.get("mac", constants.VALUE_AUTO)
6327
      if mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6328
        mac = utils.NormalizeAndValidateMac(mac)
6329

    
6330
        try:
6331
          self.cfg.ReserveMAC(mac, self.proc.GetECId())
6332
        except errors.ReservationError:
6333
          raise errors.OpPrereqError("MAC address %s already in use"
6334
                                     " in cluster" % mac,
6335
                                     errors.ECODE_NOTUNIQUE)
6336

    
6337
      # bridge verification
6338
      bridge = nic.get("bridge", None)
6339
      link = nic.get("link", None)
6340
      if bridge and link:
6341
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
6342
                                   " at the same time", errors.ECODE_INVAL)
6343
      elif bridge and nic_mode == constants.NIC_MODE_ROUTED:
6344
        raise errors.OpPrereqError("Cannot pass 'bridge' on a routed nic",
6345
                                   errors.ECODE_INVAL)
6346
      elif bridge:
6347
        link = bridge
6348

    
6349
      nicparams = {}
6350
      if nic_mode_req:
6351
        nicparams[constants.NIC_MODE] = nic_mode_req
6352
      if link:
6353
        nicparams[constants.NIC_LINK] = link
6354

    
6355
      check_params = objects.FillDict(cluster.nicparams[constants.PP_DEFAULT],
6356
                                      nicparams)
6357
      objects.NIC.CheckParameterSyntax(check_params)
6358
      self.nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
6359

    
6360
    # disk checks/pre-build
6361
    self.disks = []
6362
    for disk in self.op.disks:
6363
      mode = disk.get("mode", constants.DISK_RDWR)
6364
      if mode not in constants.DISK_ACCESS_SET:
6365
        raise errors.OpPrereqError("Invalid disk access mode '%s'" %
6366
                                   mode, errors.ECODE_INVAL)
6367
      size = disk.get("size", None)
6368
      if size is None:
6369
        raise errors.OpPrereqError("Missing disk size", errors.ECODE_INVAL)
6370
      try:
6371
        size = int(size)
6372
      except (TypeError, ValueError):
6373
        raise errors.OpPrereqError("Invalid disk size '%s'" % size,
6374
                                   errors.ECODE_INVAL)
6375
      new_disk = {"size": size, "mode": mode}
6376
      if "adopt" in disk:
6377
        new_disk["adopt"] = disk["adopt"]
6378
      self.disks.append(new_disk)
6379

    
6380
    if self.op.mode == constants.INSTANCE_IMPORT:
6381

    
6382
      # Check that the new instance doesn't have less disks than the export
6383
      instance_disks = len(self.disks)
6384
      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
6385
      if instance_disks < export_disks:
6386
        raise errors.OpPrereqError("Not enough disks to import."
6387
                                   " (instance: %d, export: %d)" %
6388
                                   (instance_disks, export_disks),
6389
                                   errors.ECODE_INVAL)
6390

    
6391
      self.op.os_type = export_info.get(constants.INISECT_EXP, 'os')
6392
      disk_images = []
6393
      for idx in range(export_disks):
6394
        option = 'disk%d_dump' % idx
6395
        if export_info.has_option(constants.INISECT_INS, option):
6396
          # FIXME: are the old os-es, disk sizes, etc. useful?
6397
          export_name = export_info.get(constants.INISECT_INS, option)
6398
          image = utils.PathJoin(self.op.src_path, export_name)
6399
          disk_images.append(image)
6400
        else:
6401
          disk_images.append(False)
6402

    
6403
      self.src_images = disk_images
6404

    
6405
      old_name = export_info.get(constants.INISECT_INS, 'name')
6406
      try:
6407
        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
6408
      except (TypeError, ValueError), err:
6409
        raise errors.OpPrereqError("Invalid export file, nic_count is not"
6410
                                   " an integer: %s" % str(err),
6411
                                   errors.ECODE_STATE)
6412
      if self.op.instance_name == old_name:
6413
        for idx, nic in enumerate(self.nics):
6414
          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
6415
            nic_mac_ini = 'nic%d_mac' % idx
6416
            nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
6417

    
6418
    # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
6419

    
6420
    # ip ping checks (we use the same ip that was resolved in ExpandNames)
6421
    if self.op.ip_check:
6422
      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
6423
        raise errors.OpPrereqError("IP %s of instance %s already in use" %
6424
                                   (self.check_ip, self.op.instance_name),
6425
                                   errors.ECODE_NOTUNIQUE)
6426

    
6427
    #### mac address generation
6428
    # By generating here the mac address both the allocator and the hooks get
6429
    # the real final mac address rather than the 'auto' or 'generate' value.
6430
    # There is a race condition between the generation and the instance object
6431
    # creation, which means that we know the mac is valid now, but we're not
6432
    # sure it will be when we actually add the instance. If things go bad
6433
    # adding the instance will abort because of a duplicate mac, and the
6434
    # creation job will fail.
6435
    for nic in self.nics:
6436
      if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
6437
        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
6438

    
6439
    #### allocator run
6440

    
6441
    if self.op.iallocator is not None:
6442
      self._RunAllocator()
6443

    
6444
    #### node related checks
6445

    
6446
    # check primary node
6447
    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
6448
    assert self.pnode is not None, \
6449
      "Cannot retrieve locked node %s" % self.op.pnode
6450
    if pnode.offline:
6451
      raise errors.OpPrereqError("Cannot use offline primary node '%s'" %
6452
                                 pnode.name, errors.ECODE_STATE)
6453
    if pnode.drained:
6454
      raise errors.OpPrereqError("Cannot use drained primary node '%s'" %
6455
                                 pnode.name, errors.ECODE_STATE)
6456

    
6457
    self.secondaries = []
6458

    
6459
    # mirror node verification
6460
    if self.op.disk_template in constants.DTS_NET_MIRROR:
6461
      if self.op.snode is None:
6462
        raise errors.OpPrereqError("The networked disk templates need"
6463
                                   " a mirror node", errors.ECODE_INVAL)
6464
      if self.op.snode == pnode.name:
6465
        raise errors.OpPrereqError("The secondary node cannot be the"
6466
                                   " primary node.", errors.ECODE_INVAL)
6467
      _CheckNodeOnline(self, self.op.snode)
6468
      _CheckNodeNotDrained(self, self.op.snode)
6469
      self.secondaries.append(self.op.snode)
6470

    
6471
    nodenames = [pnode.name] + self.secondaries
6472

    
6473
    req_size = _ComputeDiskSize(self.op.disk_template,
6474
                                self.disks)
6475

    
6476
    # Check lv size requirements, if not adopting
6477
    if req_size is not None and not self.adopt_disks:
6478
      _CheckNodesFreeDisk(self, nodenames, req_size)
6479

    
6480
    if self.adopt_disks: # instead, we must check the adoption data
6481
      all_lvs = set([i["adopt"] for i in self.disks])
6482
      if len(all_lvs) != len(self.disks):
6483
        raise errors.OpPrereqError("Duplicate volume names given for adoption",
6484
                                   errors.ECODE_INVAL)
6485
      for lv_name in all_lvs:
6486
        try:
6487
          self.cfg.ReserveLV(lv_name, self.proc.GetECId())
6488
        except errors.ReservationError:
6489
          raise errors.OpPrereqError("LV named %s used by another instance" %
6490
                                     lv_name, errors.ECODE_NOTUNIQUE)
6491

    
6492
      node_lvs = self.rpc.call_lv_list([pnode.name],
6493
                                       self.cfg.GetVGName())[pnode.name]
6494
      node_lvs.Raise("Cannot get LV information from node %s" % pnode.name)
6495
      node_lvs = node_lvs.payload
6496
      delta = all_lvs.difference(node_lvs.keys())
6497
      if delta:
6498
        raise errors.OpPrereqError("Missing logical volume(s): %s" %
6499
                                   utils.CommaJoin(delta),
6500
                                   errors.ECODE_INVAL)
6501
      online_lvs = [lv for lv in all_lvs if node_lvs[lv][2]]
6502
      if online_lvs:
6503
        raise errors.OpPrereqError("Online logical volumes found, cannot"
6504
                                   " adopt: %s" % utils.CommaJoin(online_lvs),
6505
                                   errors.ECODE_STATE)
6506
      # update the size of disk based on what is found
6507
      for dsk in self.disks:
6508
        dsk["size"] = int(float(node_lvs[dsk["adopt"]][0]))
6509

    
6510
    _CheckHVParams(self, nodenames, self.op.hypervisor, self.op.hvparams)
6511

    
6512
    _CheckNodeHasOS(self, pnode.name, self.op.os_type, self.op.force_variant)
6513

    
6514
    _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
6515

    
6516
    # memory check on primary node
6517
    if self.op.start:
6518
      _CheckNodeFreeMemory(self, self.pnode.name,
6519
                           "creating instance %s" % self.op.instance_name,
6520
                           self.be_full[constants.BE_MEMORY],
6521
                           self.op.hypervisor)
6522

    
6523
    self.dry_run_result = list(nodenames)
6524

    
6525
  def Exec(self, feedback_fn):
6526
    """Create and add the instance to the cluster.
6527

6528
    """
6529
    instance = self.op.instance_name
6530
    pnode_name = self.pnode.name
6531

    
6532
    ht_kind = self.op.hypervisor
6533
    if ht_kind in constants.HTS_REQ_PORT:
6534
      network_port = self.cfg.AllocatePort()
6535
    else:
6536
      network_port = None
6537

    
6538
    ##if self.op.vnc_bind_address is None:
6539
    ##  self.op.vnc_bind_address = constants.VNC_DEFAULT_BIND_ADDRESS
6540

    
6541
    # this is needed because os.path.join does not accept None arguments
6542
    if self.op.file_storage_dir is None:
6543
      string_file_storage_dir = ""
6544
    else:
6545
      string_file_storage_dir = self.op.file_storage_dir
6546

    
6547
    # build the full file storage dir path
6548
    file_storage_dir = utils.PathJoin(self.cfg.GetFileStorageDir(),
6549
                                      string_file_storage_dir, instance)
6550

    
6551

    
6552
    disks = _GenerateDiskTemplate(self,
6553
                                  self.op.disk_template,
6554
                                  instance, pnode_name,
6555
                                  self.secondaries,
6556
                                  self.disks,
6557
                                  file_storage_dir,
6558
                                  self.op.file_driver,
6559
                                  0)
6560

    
6561
    iobj = objects.Instance(name=instance, os=self.op.os_type,
6562
                            primary_node=pnode_name,
6563
                            nics=self.nics, disks=disks,
6564
                            disk_template=self.op.disk_template,
6565
                            admin_up=False,
6566
                            network_port=network_port,
6567
                            beparams=self.op.beparams,
6568
                            hvparams=self.op.hvparams,
6569
                            hypervisor=self.op.hypervisor,
6570
                            )
6571

    
6572
    if self.adopt_disks:
6573
      # rename LVs to the newly-generated names; we need to construct
6574
      # 'fake' LV disks with the old data, plus the new unique_id
6575
      tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
6576
      rename_to = []
6577
      for t_dsk, a_dsk in zip (tmp_disks, self.disks):
6578
        rename_to.append(t_dsk.logical_id)
6579
        t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk["adopt"])
6580
        self.cfg.SetDiskID(t_dsk, pnode_name)
6581
      result = self.rpc.call_blockdev_rename(pnode_name,
6582
                                             zip(tmp_disks, rename_to))
6583
      result.Raise("Failed to rename adoped LVs")
6584
    else:
6585
      feedback_fn("* creating instance disks...")
6586
      try:
6587
        _CreateDisks(self, iobj)
6588
      except errors.OpExecError:
6589
        self.LogWarning("Device creation failed, reverting...")
6590
        try:
6591
          _RemoveDisks(self, iobj)
6592
        finally:
6593
          self.cfg.ReleaseDRBDMinors(instance)
6594
          raise
6595

    
6596
    feedback_fn("adding instance %s to cluster config" % instance)
6597

    
6598
    self.cfg.AddInstance(iobj, self.proc.GetECId())
6599

    
6600
    # Declare that we don't want to remove the instance lock anymore, as we've
6601
    # added the instance to the config
6602
    del self.remove_locks[locking.LEVEL_INSTANCE]
6603
    # Unlock all the nodes
6604
    if self.op.mode == constants.INSTANCE_IMPORT:
6605
      nodes_keep = [self.op.src_node]
6606
      nodes_release = [node for node in self.acquired_locks[locking.LEVEL_NODE]
6607
                       if node != self.op.src_node]
6608
      self.context.glm.release(locking.LEVEL_NODE, nodes_release)
6609
      self.acquired_locks[locking.LEVEL_NODE] = nodes_keep
6610
    else:
6611
      self.context.glm.release(locking.LEVEL_NODE)
6612
      del self.acquired_locks[locking.LEVEL_NODE]
6613

    
6614
    if self.op.wait_for_sync:
6615
      disk_abort = not _WaitForSync(self, iobj)
6616
    elif iobj.disk_template in constants.DTS_NET_MIRROR:
6617
      # make sure the disks are not degraded (still sync-ing is ok)
6618
      time.sleep(15)
6619
      feedback_fn("* checking mirrors status")
6620
      disk_abort = not _WaitForSync(self, iobj, oneshot=True)
6621
    else:
6622
      disk_abort = False
6623

    
6624
    if disk_abort:
6625
      _RemoveDisks(self, iobj)
6626
      self.cfg.RemoveInstance(iobj.name)
6627
      # Make sure the instance lock gets removed
6628
      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
6629
      raise errors.OpExecError("There are some degraded disks for"
6630
                               " this instance")
6631

    
6632
    if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
6633
      if self.op.mode == constants.INSTANCE_CREATE:
6634
        if not self.op.no_install:
6635
          feedback_fn("* running the instance OS create scripts...")
6636
          # FIXME: pass debug option from opcode to backend
6637
          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
6638
                                                 self.op.debug_level)
6639
          result.Raise("Could not add os for instance %s"
6640
                       " on node %s" % (instance, pnode_name))
6641

    
6642
      elif self.op.mode == constants.INSTANCE_IMPORT:
6643
        feedback_fn("* running the instance OS import scripts...")
6644
        src_node = self.op.src_node
6645
        src_images = self.src_images
6646
        cluster_name = self.cfg.GetClusterName()
6647
        # FIXME: pass debug option from opcode to backend
6648
        import_result = self.rpc.call_instance_os_import(pnode_name, iobj,
6649
                                                         src_node, src_images,
6650
                                                         cluster_name,
6651
                                                         self.op.debug_level)
6652
        msg = import_result.fail_msg
6653
        if msg:
6654
          self.LogWarning("Error while importing the disk images for instance"
6655
                          " %s on node %s: %s" % (instance, pnode_name, msg))
6656
      else:
6657
        # also checked in the prereq part
6658
        raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
6659
                                     % self.op.mode)
6660

    
6661
    if self.op.start:
6662
      iobj.admin_up = True
6663
      self.cfg.Update(iobj, feedback_fn)
6664
      logging.info("Starting instance %s on node %s", instance, pnode_name)
6665
      feedback_fn("* starting instance...")
6666
      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
6667
      result.Raise("Could not start instance")
6668

    
6669
    return list(iobj.all_nodes)
6670

    
6671

    
6672
class LUConnectConsole(NoHooksLU):
6673
  """Connect to an instance's console.
6674

6675
  This is somewhat special in that it returns the command line that
6676
  you need to run on the master node in order to connect to the
6677
  console.
6678

6679
  """
6680
  _OP_REQP = ["instance_name"]
6681
  REQ_BGL = False
6682

    
6683
  def ExpandNames(self):
6684
    self._ExpandAndLockInstance()
6685

    
6686
  def CheckPrereq(self):
6687
    """Check prerequisites.
6688

6689
    This checks that the instance is in the cluster.
6690

6691
    """
6692
    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
6693
    assert self.instance is not None, \
6694
      "Cannot retrieve locked instance %s" % self.op.instance_name
6695
    _CheckNodeOnline(self, self.instance.primary_node)
6696

    
6697
  def Exec(self, feedback_fn):
6698
    """Connect to the console of an instance
6699

6700
    """
6701
    instance = self.instance
6702
    node = instance.primary_node
6703

    
6704
    node_insts = self.rpc.call_instance_list([node],
6705
                                             [instance.hypervisor])[node]
6706
    node_insts.Raise("Can't get node information from %s" % node)
6707

    
6708
    if instance.name not in node_insts.payload:
6709
      raise errors.OpExecError("Instance %s is not running." % instance.name)
6710

    
6711
    logging.debug("Connecting to console of %s on %s", instance.name, node)
6712

    
6713
    hyper = hypervisor.GetHypervisor(instance.hypervisor)
6714
    cluster = self.cfg.GetClusterInfo()
6715
    # beparams and hvparams are passed separately, to avoid editing the
6716
    # instance and then saving the defaults in the instance itself.
6717
    hvparams = cluster.FillHV(instance)
6718
    beparams = cluster.FillBE(instance)
6719
    console_cmd = hyper.GetShellCommandForConsole(instance, hvparams, beparams)
6720

    
6721
    # build ssh cmdline
6722
    return self.ssh.BuildCmd(node, "root", console_cmd, batch=True, tty=True)
6723

    
6724

    
6725
class LUReplaceDisks(LogicalUnit):
6726
  """Replace the disks of an instance.
6727

6728
  """
6729
  HPATH = "mirrors-replace"
6730
  HTYPE = constants.HTYPE_INSTANCE
6731
  _OP_REQP = ["instance_name", "mode", "disks"]
6732
  REQ_BGL = False
6733

    
6734
  def CheckArguments(self):
6735
    if not hasattr(self.op, "remote_node"):
6736
      self.op.remote_node = None
6737
    if not hasattr(self.op, "iallocator"):
6738
      self.op.iallocator = None
6739
    if not hasattr(self.op, "early_release"):
6740
      self.op.early_release = False
6741

    
6742
    TLReplaceDisks.CheckArguments(self.op.mode, self.op.remote_node,
6743
                                  self.op.iallocator)
6744

    
6745
  def ExpandNames(self):
6746
    self._ExpandAndLockInstance()
6747

    
6748
    if self.op.iallocator is not None:
6749
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6750

    
6751
    elif self.op.remote_node is not None:
6752
      remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
6753
      self.op.remote_node = remote_node
6754

    
6755
      # Warning: do not remove the locking of the new secondary here
6756
      # unless DRBD8.AddChildren is changed to work in parallel;
6757
      # currently it doesn't since parallel invocations of
6758
      # FindUnusedMinor will conflict
6759
      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
6760
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6761

    
6762
    else:
6763
      self.needed_locks[locking.LEVEL_NODE] = []
6764
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
6765

    
6766
    self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
6767
                                   self.op.iallocator, self.op.remote_node,
6768
                                   self.op.disks, False, self.op.early_release)
6769

    
6770
    self.tasklets = [self.replacer]
6771

    
6772
  def DeclareLocks(self, level):
6773
    # If we're not already locking all nodes in the set we have to declare the
6774
    # instance's primary/secondary nodes.
6775
    if (level == locking.LEVEL_NODE and
6776
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
6777
      self._LockInstancesNodes()
6778

    
6779
  def BuildHooksEnv(self):
6780
    """Build hooks env.
6781

6782
    This runs on the master, the primary and all the secondaries.
6783

6784
    """
6785
    instance = self.replacer.instance
6786
    env = {
6787
      "MODE": self.op.mode,
6788
      "NEW_SECONDARY": self.op.remote_node,
6789
      "OLD_SECONDARY": instance.secondary_nodes[0],
6790
      }
6791
    env.update(_BuildInstanceHookEnvByObject(self, instance))
6792
    nl = [
6793
      self.cfg.GetMasterNode(),
6794
      instance.primary_node,
6795
      ]
6796
    if self.op.remote_node is not None:
6797
      nl.append(self.op.remote_node)
6798
    return env, nl, nl
6799

    
6800

    
6801
class LUEvacuateNode(LogicalUnit):
6802
  """Relocate the secondary instances from a node.
6803

6804
  """
6805
  HPATH = "node-evacuate"
6806
  HTYPE = constants.HTYPE_NODE
6807
  _OP_REQP = ["node_name"]
6808
  REQ_BGL = False
6809

    
6810
  def CheckArguments(self):
6811
    if not hasattr(self.op, "remote_node"):
6812
      self.op.remote_node = None
6813
    if not hasattr(self.op, "iallocator"):
6814
      self.op.iallocator = None
6815
    if not hasattr(self.op, "early_release"):
6816
      self.op.early_release = False
6817

    
6818
    TLReplaceDisks.CheckArguments(constants.REPLACE_DISK_CHG,
6819
                                  self.op.remote_node,
6820
                                  self.op.iallocator)
6821

    
6822
  def ExpandNames(self):
6823
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
6824

    
6825
    self.needed_locks = {}
6826

    
6827
    # Declare node locks
6828
    if self.op.iallocator is not None:
6829
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
6830

    
6831
    elif self.op.remote_node is not None:
6832
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
6833

    
6834
      # Warning: do not remove the locking of the new secondary here
6835
      # unless DRBD8.AddChildren is changed to work in parallel;
6836
      # currently it doesn't since parallel invocations of
6837
      # FindUnusedMinor will conflict
6838
      self.needed_locks[locking.LEVEL_NODE] = [self.op.remote_node]
6839
      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
6840

    
6841
    else:
6842
      raise errors.OpPrereqError("Invalid parameters", errors.ECODE_INVAL)
6843

    
6844
    # Create tasklets for replacing disks for all secondary instances on this
6845
    # node
6846
    names = []
6847
    tasklets = []
6848

    
6849
    for inst in _GetNodeSecondaryInstances(self.cfg, self.op.node_name):
6850
      logging.debug("Replacing disks for instance %s", inst.name)
6851
      names.append(inst.name)
6852

    
6853
      replacer = TLReplaceDisks(self, inst.name, constants.REPLACE_DISK_CHG,
6854
                                self.op.iallocator, self.op.remote_node, [],
6855
                                True, self.op.early_release)
6856
      tasklets.append(replacer)
6857

    
6858
    self.tasklets = tasklets
6859
    self.instance_names = names
6860

    
6861
    # Declare instance locks
6862
    self.needed_locks[locking.LEVEL_INSTANCE] = self.instance_names
6863

    
6864
  def DeclareLocks(self, level):
6865
    # If we're not already locking all nodes in the set we have to declare the
6866
    # instance's primary/secondary nodes.
6867
    if (level == locking.LEVEL_NODE and
6868
        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
6869
      self._LockInstancesNodes()
6870

    
6871
  def BuildHooksEnv(self):
6872
    """Build hooks env.
6873

6874
    This runs on the master, the primary and all the secondaries.
6875

6876
    """
6877
    env = {
6878
      "NODE_NAME": self.op.node_name,
6879
      }
6880

    
6881
    nl = [self.cfg.GetMasterNode()]
6882

    
6883
    if self.op.remote_node is not None:
6884
      env["NEW_SECONDARY"] = self.op.remote_node
6885
      nl.append(self.op.remote_node)
6886

    
6887
    return (env, nl, nl)
6888

    
6889

    
6890
class TLReplaceDisks(Tasklet):
6891
  """Replaces disks for an instance.
6892

6893
  Note: Locking is not within the scope of this class.
6894

6895
  """
6896
  def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
6897
               disks, delay_iallocator, early_release):
6898
    """Initializes this class.
6899

6900
    """
6901
    Tasklet.__init__(self, lu)
6902

    
6903
    # Parameters
6904
    self.instance_name = instance_name
6905
    self.mode = mode
6906
    self.iallocator_name = iallocator_name
6907
    self.remote_node = remote_node
6908
    self.disks = disks
6909
    self.delay_iallocator = delay_iallocator
6910
    self.early_release = early_release
6911

    
6912
    # Runtime data
6913
    self.instance = None
6914
    self.new_node = None
6915
    self.target_node = None
6916
    self.other_node = None
6917
    self.remote_node_info = None
6918
    self.node_secondary_ip = None
6919

    
6920
  @staticmethod
6921
  def CheckArguments(mode, remote_node, iallocator):
6922
    """Helper function for users of this class.
6923

6924
    """
6925
    # check for valid parameter combination
6926
    if mode == constants.REPLACE_DISK_CHG:
6927
      if remote_node is None and iallocator is None:
6928
        raise errors.OpPrereqError("When changing the secondary either an"
6929
                                   " iallocator script must be used or the"
6930
                                   " new node given", errors.ECODE_INVAL)
6931

    
6932
      if remote_node is not None and iallocator is not None:
6933
        raise errors.OpPrereqError("Give either the iallocator or the new"
6934
                                   " secondary, not both", errors.ECODE_INVAL)
6935

    
6936
    elif remote_node is not None or iallocator is not None:
6937
      # Not replacing the secondary
6938
      raise errors.OpPrereqError("The iallocator and new node options can"
6939
                                 " only be used when changing the"
6940
                                 " secondary node", errors.ECODE_INVAL)
6941

    
6942
  @staticmethod
6943
  def _RunAllocator(lu, iallocator_name, instance_name, relocate_from):
6944
    """Compute a new secondary node using an IAllocator.
6945

6946
    """
6947
    ial = IAllocator(lu.cfg, lu.rpc,
6948
                     mode=constants.IALLOCATOR_MODE_RELOC,
6949
                     name=instance_name,
6950
                     relocate_from=relocate_from)
6951

    
6952
    ial.Run(iallocator_name)
6953

    
6954
    if not ial.success:
6955
      raise errors.OpPrereqError("Can't compute nodes using iallocator '%s':"
6956
                                 " %s" % (iallocator_name, ial.info),
6957
                                 errors.ECODE_NORES)
6958

    
6959
    if len(ial.result) != ial.required_nodes:
6960
      raise errors.OpPrereqError("iallocator '%s' returned invalid number"
6961
                                 " of nodes (%s), required %s" %
6962
                                 (iallocator_name,
6963
                                  len(ial.result), ial.required_nodes),
6964
                                 errors.ECODE_FAULT)
6965

    
6966
    remote_node_name = ial.result[0]
6967

    
6968
    lu.LogInfo("Selected new secondary for instance '%s': %s",
6969
               instance_name, remote_node_name)
6970

    
6971
    return remote_node_name
6972

    
6973
  def _FindFaultyDisks(self, node_name):
6974
    return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
6975
                                    node_name, True)
6976

    
6977
  def CheckPrereq(self):
6978
    """Check prerequisites.
6979

6980
    This checks that the instance is in the cluster.
6981

6982
    """
6983
    self.instance = instance = self.cfg.GetInstanceInfo(self.instance_name)
6984
    assert instance is not None, \
6985
      "Cannot retrieve locked instance %s" % self.instance_name
6986

    
6987
    if instance.disk_template != constants.DT_DRBD8:
6988
      raise errors.OpPrereqError("Can only run replace disks for DRBD8-based"
6989
                                 " instances", errors.ECODE_INVAL)
6990

    
6991
    if len(instance.secondary_nodes) != 1:
6992
      raise errors.OpPrereqError("The instance has a strange layout,"
6993
                                 " expected one secondary but found %d" %
6994
                                 len(instance.secondary_nodes),
6995
                                 errors.ECODE_FAULT)
6996

    
6997
    if not self.delay_iallocator:
6998
      self._CheckPrereq2()
6999

    
7000
  def _CheckPrereq2(self):
7001
    """Check prerequisites, second part.
7002

7003
    This function should always be part of CheckPrereq. It was separated and is
7004
    now called from Exec because during node evacuation iallocator was only
7005
    called with an unmodified cluster model, not taking planned changes into
7006
    account.
7007

7008
    """
7009
    instance = self.instance
7010
    secondary_node = instance.secondary_nodes[0]
7011

    
7012
    if self.iallocator_name is None:
7013
      remote_node = self.remote_node
7014
    else:
7015
      remote_node = self._RunAllocator(self.lu, self.iallocator_name,
7016
                                       instance.name, instance.secondary_nodes)
7017

    
7018
    if remote_node is not None:
7019
      self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
7020
      assert self.remote_node_info is not None, \
7021
        "Cannot retrieve locked node %s" % remote_node
7022
    else:
7023
      self.remote_node_info = None
7024

    
7025
    if remote_node == self.instance.primary_node:
7026
      raise errors.OpPrereqError("The specified node is the primary node of"
7027
                                 " the instance.", errors.ECODE_INVAL)
7028

    
7029
    if remote_node == secondary_node:
7030
      raise errors.OpPrereqError("The specified node is already the"
7031
                                 " secondary node of the instance.",
7032
                                 errors.ECODE_INVAL)
7033

    
7034
    if self.disks and self.mode in (constants.REPLACE_DISK_AUTO,
7035
                                    constants.REPLACE_DISK_CHG):
7036
      raise errors.OpPrereqError("Cannot specify disks to be replaced",
7037
                                 errors.ECODE_INVAL)
7038

    
7039
    if self.mode == constants.REPLACE_DISK_AUTO:
7040
      faulty_primary = self._FindFaultyDisks(instance.primary_node)
7041
      faulty_secondary = self._FindFaultyDisks(secondary_node)
7042

    
7043
      if faulty_primary and faulty_secondary:
7044
        raise errors.OpPrereqError("Instance %s has faulty disks on more than"
7045
                                   " one node and can not be repaired"
7046
                                   " automatically" % self.instance_name,
7047
                                   errors.ECODE_STATE)
7048

    
7049
      if faulty_primary:
7050
        self.disks = faulty_primary
7051
        self.target_node = instance.primary_node
7052
        self.other_node = secondary_node
7053
        check_nodes = [self.target_node, self.other_node]
7054
      elif faulty_secondary:
7055
        self.disks = faulty_secondary
7056
        self.target_node = secondary_node
7057
        self.other_node = instance.primary_node
7058
        check_nodes = [self.target_node, self.other_node]
7059
      else:
7060
        self.disks = []
7061
        check_nodes = []
7062

    
7063
    else:
7064
      # Non-automatic modes
7065
      if self.mode == constants.REPLACE_DISK_PRI:
7066
        self.target_node = instance.primary_node
7067
        self.other_node = secondary_node
7068
        check_nodes = [self.target_node, self.other_node]
7069

    
7070
      elif self.mode == constants.REPLACE_DISK_SEC:
7071
        self.target_node = secondary_node
7072
        self.other_node = instance.primary_node
7073
        check_nodes = [self.target_node, self.other_node]
7074

    
7075
      elif self.mode == constants.REPLACE_DISK_CHG:
7076
        self.new_node = remote_node
7077
        self.other_node = instance.primary_node
7078
        self.target_node = secondary_node
7079
        check_nodes = [self.new_node, self.other_node]
7080

    
7081
        _CheckNodeNotDrained(self.lu, remote_node)
7082

    
7083
        old_node_info = self.cfg.GetNodeInfo(secondary_node)
7084
        assert old_node_info is not None
7085
        if old_node_info.offline and not self.early_release:
7086
          # doesn't make sense to delay the release
7087
          self.early_release = True
7088
          self.lu.LogInfo("Old secondary %s is offline, automatically enabling"
7089
                          " early-release mode", secondary_node)
7090

    
7091
      else:
7092
        raise errors.ProgrammerError("Unhandled disk replace mode (%s)" %
7093
                                     self.mode)
7094

    
7095
      # If not specified all disks should be replaced
7096
      if not self.disks:
7097
        self.disks = range(len(self.instance.disks))
7098

    
7099
    for node in check_nodes:
7100
      _CheckNodeOnline(self.lu, node)
7101

    
7102
    # Check whether disks are valid
7103
    for disk_idx in self.disks:
7104
      instance.FindDisk(disk_idx)
7105

    
7106
    # Get secondary node IP addresses
7107
    node_2nd_ip = {}
7108

    
7109
    for node_name in [self.target_node, self.other_node, self.new_node]:
7110
      if node_name is not None:
7111
        node_2nd_ip[node_name] = self.cfg.GetNodeInfo(node_name).secondary_ip
7112

    
7113
    self.node_secondary_ip = node_2nd_ip
7114

    
7115
  def Exec(self, feedback_fn):
7116
    """Execute disk replacement.
7117

7118
    This dispatches the disk replacement to the appropriate handler.
7119

7120
    """
7121
    if self.delay_iallocator:
7122
      self._CheckPrereq2()
7123

    
7124
    if not self.disks:
7125
      feedback_fn("No disks need replacement")
7126
      return
7127

    
7128
    feedback_fn("Replacing disk(s) %s for %s" %
7129
                (utils.CommaJoin(self.disks), self.instance.name))
7130

    
7131
    activate_disks = (not self.instance.admin_up)
7132

    
7133
    # Activate the instance disks if we're replacing them on a down instance
7134
    if activate_disks:
7135
      _StartInstanceDisks(self.lu, self.instance, True)
7136

    
7137
    try:
7138
      # Should we replace the secondary node?
7139
      if self.new_node is not None:
7140
        fn = self._ExecDrbd8Secondary
7141
      else:
7142
        fn = self._ExecDrbd8DiskOnly
7143

    
7144
      return fn(feedback_fn)
7145

    
7146
    finally:
7147
      # Deactivate the instance disks if we're replacing them on a
7148
      # down instance
7149
      if activate_disks:
7150
        _SafeShutdownInstanceDisks(self.lu, self.instance)
7151

    
7152
  def _CheckVolumeGroup(self, nodes):
7153
    self.lu.LogInfo("Checking volume groups")
7154

    
7155
    vgname = self.cfg.GetVGName()
7156

    
7157
    # Make sure volume group exists on all involved nodes
7158
    results = self.rpc.call_vg_list(nodes)
7159
    if not results:
7160
      raise errors.OpExecError("Can't list volume groups on the nodes")
7161

    
7162
    for node in nodes:
7163
      res = results[node]
7164
      res.Raise("Error checking node %s" % node)
7165
      if vgname not in res.payload:
7166
        raise errors.OpExecError("Volume group '%s' not found on node %s" %
7167
                                 (vgname, node))
7168

    
7169
  def _CheckDisksExistence(self, nodes):
7170
    # Check disk existence
7171
    for idx, dev in enumerate(self.instance.disks):
7172
      if idx not in self.disks:
7173
        continue
7174

    
7175
      for node in nodes:
7176
        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
7177
        self.cfg.SetDiskID(dev, node)
7178

    
7179
        result = self.rpc.call_blockdev_find(node, dev)
7180

    
7181
        msg = result.fail_msg
7182
        if msg or not result.payload:
7183
          if not msg:
7184
            msg = "disk not found"
7185
          raise errors.OpExecError("Can't find disk/%d on node %s: %s" %
7186
                                   (idx, node, msg))
7187

    
7188
  def _CheckDisksConsistency(self, node_name, on_primary, ldisk):
7189
    for idx, dev in enumerate(self.instance.disks):
7190
      if idx not in self.disks:
7191
        continue
7192

    
7193
      self.lu.LogInfo("Checking disk/%d consistency on node %s" %
7194
                      (idx, node_name))
7195

    
7196
      if not _CheckDiskConsistency(self.lu, dev, node_name, on_primary,
7197
                                   ldisk=ldisk):
7198
        raise errors.OpExecError("Node %s has degraded storage, unsafe to"
7199
                                 " replace disks for instance %s" %
7200
                                 (node_name, self.instance.name))
7201

    
7202
  def _CreateNewStorage(self, node_name):
7203
    vgname = self.cfg.GetVGName()
7204
    iv_names = {}
7205

    
7206
    for idx, dev in enumerate(self.instance.disks):
7207
      if idx not in self.disks:
7208
        continue
7209

    
7210
      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
7211

    
7212
      self.cfg.SetDiskID(dev, node_name)
7213

    
7214
      lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
7215
      names = _GenerateUniqueNames(self.lu, lv_names)
7216

    
7217
      lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
7218
                             logical_id=(vgname, names[0]))
7219
      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
7220
                             logical_id=(vgname, names[1]))
7221

    
7222
      new_lvs = [lv_data, lv_meta]
7223
      old_lvs = dev.children
7224
      iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
7225

    
7226
      # we pass force_create=True to force the LVM creation
7227
      for new_lv in new_lvs:
7228
        _CreateBlockDev(self.lu, node_name, self.instance, new_lv, True,
7229
                        _GetInstanceInfoText(self.instance), False)
7230

    
7231
    return iv_names
7232

    
7233
  def _CheckDevices(self, node_name, iv_names):
7234
    for name, (dev, _, _) in iv_names.iteritems():
7235
      self.cfg.SetDiskID(dev, node_name)
7236

    
7237
      result = self.rpc.call_blockdev_find(node_name, dev)
7238

    
7239
      msg = result.fail_msg
7240
      if msg or not result.payload:
7241
        if not msg:
7242
          msg = "disk not found"
7243
        raise errors.OpExecError("Can't find DRBD device %s: %s" %
7244
                                 (name, msg))
7245

    
7246
      if result.payload.is_degraded:
7247
        raise errors.OpExecError("DRBD device %s is degraded!" % name)
7248

    
7249
  def _RemoveOldStorage(self, node_name, iv_names):
7250
    for name, (_, old_lvs, _) in iv_names.iteritems():
7251
      self.lu.LogInfo("Remove logical volumes for %s" % name)
7252

    
7253
      for lv in old_lvs:
7254
        self.cfg.SetDiskID(lv, node_name)
7255

    
7256
        msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
7257
        if msg:
7258
          self.lu.LogWarning("Can't remove old LV: %s" % msg,
7259
                             hint="remove unused LVs manually")
7260

    
7261
  def _ReleaseNodeLock(self, node_name):
7262
    """Releases the lock for a given node."""
7263
    self.lu.context.glm.release(locking.LEVEL_NODE, node_name)
7264

    
7265
  def _ExecDrbd8DiskOnly(self, feedback_fn):
7266
    """Replace a disk on the primary or secondary for DRBD 8.
7267

7268
    The algorithm for replace is quite complicated:
7269

7270
      1. for each disk to be replaced:
7271

7272
        1. create new LVs on the target node with unique names
7273
        1. detach old LVs from the drbd device
7274
        1. rename old LVs to name_replaced.<time_t>
7275
        1. rename new LVs to old LVs
7276
        1. attach the new LVs (with the old names now) to the drbd device
7277

7278
      1. wait for sync across all devices
7279

7280
      1. for each modified disk:
7281

7282
        1. remove old LVs (which have the name name_replaces.<time_t>)
7283

7284
    Failures are not very well handled.
7285

7286
    """
7287
    steps_total = 6
7288

    
7289
    # Step: check device activation
7290
    self.lu.LogStep(1, steps_total, "Check device existence")
7291
    self._CheckDisksExistence([self.other_node, self.target_node])
7292
    self._CheckVolumeGroup([self.target_node, self.other_node])
7293

    
7294
    # Step: check other node consistency
7295
    self.lu.LogStep(2, steps_total, "Check peer consistency")
7296
    self._CheckDisksConsistency(self.other_node,
7297
                                self.other_node == self.instance.primary_node,
7298
                                False)
7299

    
7300
    # Step: create new storage
7301
    self.lu.LogStep(3, steps_total, "Allocate new storage")
7302
    iv_names = self._CreateNewStorage(self.target_node)
7303

    
7304
    # Step: for each lv, detach+rename*2+attach
7305
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
7306
    for dev, old_lvs, new_lvs in iv_names.itervalues():
7307
      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
7308

    
7309
      result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
7310
                                                     old_lvs)
7311
      result.Raise("Can't detach drbd from local storage on node"
7312
                   " %s for device %s" % (self.target_node, dev.iv_name))
7313
      #dev.children = []
7314
      #cfg.Update(instance)
7315

    
7316
      # ok, we created the new LVs, so now we know we have the needed
7317
      # storage; as such, we proceed on the target node to rename
7318
      # old_lv to _old, and new_lv to old_lv; note that we rename LVs
7319
      # using the assumption that logical_id == physical_id (which in
7320
      # turn is the unique_id on that node)
7321

    
7322
      # FIXME(iustin): use a better name for the replaced LVs
7323
      temp_suffix = int(time.time())
7324
      ren_fn = lambda d, suff: (d.physical_id[0],
7325
                                d.physical_id[1] + "_replaced-%s" % suff)
7326

    
7327
      # Build the rename list based on what LVs exist on the node
7328
      rename_old_to_new = []
7329
      for to_ren in old_lvs:
7330
        result = self.rpc.call_blockdev_find(self.target_node, to_ren)
7331
        if not result.fail_msg and result.payload:
7332
          # device exists
7333
          rename_old_to_new.append((to_ren, ren_fn(to_ren, temp_suffix)))
7334

    
7335
      self.lu.LogInfo("Renaming the old LVs on the target node")
7336
      result = self.rpc.call_blockdev_rename(self.target_node,
7337
                                             rename_old_to_new)
7338
      result.Raise("Can't rename old LVs on node %s" % self.target_node)
7339

    
7340
      # Now we rename the new LVs to the old LVs
7341
      self.lu.LogInfo("Renaming the new LVs on the target node")
7342
      rename_new_to_old = [(new, old.physical_id)
7343
                           for old, new in zip(old_lvs, new_lvs)]
7344
      result = self.rpc.call_blockdev_rename(self.target_node,
7345
                                             rename_new_to_old)
7346
      result.Raise("Can't rename new LVs on node %s" % self.target_node)
7347

    
7348
      for old, new in zip(old_lvs, new_lvs):
7349
        new.logical_id = old.logical_id
7350
        self.cfg.SetDiskID(new, self.target_node)
7351

    
7352
      for disk in old_lvs:
7353
        disk.logical_id = ren_fn(disk, temp_suffix)
7354
        self.cfg.SetDiskID(disk, self.target_node)
7355

    
7356
      # Now that the new lvs have the old name, we can add them to the device
7357
      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
7358
      result = self.rpc.call_blockdev_addchildren(self.target_node, dev,
7359
                                                  new_lvs)
7360
      msg = result.fail_msg
7361
      if msg:
7362
        for new_lv in new_lvs:
7363
          msg2 = self.rpc.call_blockdev_remove(self.target_node,
7364
                                               new_lv).fail_msg
7365
          if msg2:
7366
            self.lu.LogWarning("Can't rollback device %s: %s", dev, msg2,
7367
                               hint=("cleanup manually the unused logical"
7368
                                     "volumes"))
7369
        raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
7370

    
7371
      dev.children = new_lvs
7372

    
7373
      self.cfg.Update(self.instance, feedback_fn)
7374

    
7375
    cstep = 5
7376
    if self.early_release:
7377
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
7378
      cstep += 1
7379
      self._RemoveOldStorage(self.target_node, iv_names)
7380
      # WARNING: we release both node locks here, do not do other RPCs
7381
      # than WaitForSync to the primary node
7382
      self._ReleaseNodeLock([self.target_node, self.other_node])
7383

    
7384
    # Wait for sync
7385
    # This can fail as the old devices are degraded and _WaitForSync
7386
    # does a combined result over all disks, so we don't check its return value
7387
    self.lu.LogStep(cstep, steps_total, "Sync devices")
7388
    cstep += 1
7389
    _WaitForSync(self.lu, self.instance)
7390

    
7391
    # Check all devices manually
7392
    self._CheckDevices(self.instance.primary_node, iv_names)
7393

    
7394
    # Step: remove old storage
7395
    if not self.early_release:
7396
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
7397
      cstep += 1
7398
      self._RemoveOldStorage(self.target_node, iv_names)
7399

    
7400
  def _ExecDrbd8Secondary(self, feedback_fn):
7401
    """Replace the secondary node for DRBD 8.
7402

7403
    The algorithm for replace is quite complicated:
7404
      - for all disks of the instance:
7405
        - create new LVs on the new node with same names
7406
        - shutdown the drbd device on the old secondary
7407
        - disconnect the drbd network on the primary
7408
        - create the drbd device on the new secondary
7409
        - network attach the drbd on the primary, using an artifice:
7410
          the drbd code for Attach() will connect to the network if it
7411
          finds a device which is connected to the good local disks but
7412
          not network enabled
7413
      - wait for sync across all devices
7414
      - remove all disks from the old secondary
7415

7416
    Failures are not very well handled.
7417

7418
    """
7419
    steps_total = 6
7420

    
7421
    # Step: check device activation
7422
    self.lu.LogStep(1, steps_total, "Check device existence")
7423
    self._CheckDisksExistence([self.instance.primary_node])
7424
    self._CheckVolumeGroup([self.instance.primary_node])
7425

    
7426
    # Step: check other node consistency
7427
    self.lu.LogStep(2, steps_total, "Check peer consistency")
7428
    self._CheckDisksConsistency(self.instance.primary_node, True, True)
7429

    
7430
    # Step: create new storage
7431
    self.lu.LogStep(3, steps_total, "Allocate new storage")
7432
    for idx, dev in enumerate(self.instance.disks):
7433
      self.lu.LogInfo("Adding new local storage on %s for disk/%d" %
7434
                      (self.new_node, idx))
7435
      # we pass force_create=True to force LVM creation
7436
      for new_lv in dev.children:
7437
        _CreateBlockDev(self.lu, self.new_node, self.instance, new_lv, True,
7438
                        _GetInstanceInfoText(self.instance), False)
7439

    
7440
    # Step 4: dbrd minors and drbd setups changes
7441
    # after this, we must manually remove the drbd minors on both the
7442
    # error and the success paths
7443
    self.lu.LogStep(4, steps_total, "Changing drbd configuration")
7444
    minors = self.cfg.AllocateDRBDMinor([self.new_node
7445
                                         for dev in self.instance.disks],
7446
                                        self.instance.name)
7447
    logging.debug("Allocated minors %r", minors)
7448

    
7449
    iv_names = {}
7450
    for idx, (dev, new_minor) in enumerate(zip(self.instance.disks, minors)):
7451
      self.lu.LogInfo("activating a new drbd on %s for disk/%d" %
7452
                      (self.new_node, idx))
7453
      # create new devices on new_node; note that we create two IDs:
7454
      # one without port, so the drbd will be activated without
7455
      # networking information on the new node at this stage, and one
7456
      # with network, for the latter activation in step 4
7457
      (o_node1, o_node2, o_port, o_minor1, o_minor2, o_secret) = dev.logical_id
7458
      if self.instance.primary_node == o_node1:
7459
        p_minor = o_minor1
7460
      else:
7461
        assert self.instance.primary_node == o_node2, "Three-node instance?"
7462
        p_minor = o_minor2
7463

    
7464
      new_alone_id = (self.instance.primary_node, self.new_node, None,
7465
                      p_minor, new_minor, o_secret)
7466
      new_net_id = (self.instance.primary_node, self.new_node, o_port,
7467
                    p_minor, new_minor, o_secret)
7468

    
7469
      iv_names[idx] = (dev, dev.children, new_net_id)
7470
      logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
7471
                    new_net_id)
7472
      new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
7473
                              logical_id=new_alone_id,
7474
                              children=dev.children,
7475
                              size=dev.size)
7476
      try:
7477
        _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
7478
                              _GetInstanceInfoText(self.instance), False)
7479
      except errors.GenericError:
7480
        self.cfg.ReleaseDRBDMinors(self.instance.name)
7481
        raise
7482

    
7483
    # We have new devices, shutdown the drbd on the old secondary
7484
    for idx, dev in enumerate(self.instance.disks):
7485
      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
7486
      self.cfg.SetDiskID(dev, self.target_node)
7487
      msg = self.rpc.call_blockdev_shutdown(self.target_node, dev).fail_msg
7488
      if msg:
7489
        self.lu.LogWarning("Failed to shutdown drbd for disk/%d on old"
7490
                           "node: %s" % (idx, msg),
7491
                           hint=("Please cleanup this device manually as"
7492
                                 " soon as possible"))
7493

    
7494
    self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
7495
    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
7496
                                               self.node_secondary_ip,
7497
                                               self.instance.disks)\
7498
                                              [self.instance.primary_node]
7499

    
7500
    msg = result.fail_msg
7501
    if msg:
7502
      # detaches didn't succeed (unlikely)
7503
      self.cfg.ReleaseDRBDMinors(self.instance.name)
7504
      raise errors.OpExecError("Can't detach the disks from the network on"
7505
                               " old node: %s" % (msg,))
7506

    
7507
    # if we managed to detach at least one, we update all the disks of
7508
    # the instance to point to the new secondary
7509
    self.lu.LogInfo("Updating instance configuration")
7510
    for dev, _, new_logical_id in iv_names.itervalues():
7511
      dev.logical_id = new_logical_id
7512
      self.cfg.SetDiskID(dev, self.instance.primary_node)
7513

    
7514
    self.cfg.Update(self.instance, feedback_fn)
7515

    
7516
    # and now perform the drbd attach
7517
    self.lu.LogInfo("Attaching primary drbds to new secondary"
7518
                    " (standalone => connected)")
7519
    result = self.rpc.call_drbd_attach_net([self.instance.primary_node,
7520
                                            self.new_node],
7521
                                           self.node_secondary_ip,
7522
                                           self.instance.disks,
7523
                                           self.instance.name,
7524
                                           False)
7525
    for to_node, to_result in result.items():
7526
      msg = to_result.fail_msg
7527
      if msg:
7528
        self.lu.LogWarning("Can't attach drbd disks on node %s: %s",
7529
                           to_node, msg,
7530
                           hint=("please do a gnt-instance info to see the"
7531
                                 " status of disks"))
7532
    cstep = 5
7533
    if self.early_release:
7534
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
7535
      cstep += 1
7536
      self._RemoveOldStorage(self.target_node, iv_names)
7537
      # WARNING: we release all node locks here, do not do other RPCs
7538
      # than WaitForSync to the primary node
7539
      self._ReleaseNodeLock([self.instance.primary_node,
7540
                             self.target_node,
7541
                             self.new_node])
7542

    
7543
    # Wait for sync
7544
    # This can fail as the old devices are degraded and _WaitForSync
7545
    # does a combined result over all disks, so we don't check its return value
7546
    self.lu.LogStep(cstep, steps_total, "Sync devices")
7547
    cstep += 1
7548
    _WaitForSync(self.lu, self.instance)
7549

    
7550
    # Check all devices manually
7551
    self._CheckDevices(self.instance.primary_node, iv_names)
7552

    
7553
    # Step: remove old storage
7554
    if not self.early_release:
7555
      self.lu.LogStep(cstep, steps_total, "Removing old storage")
7556
      self._RemoveOldStorage(self.target_node, iv_names)
7557

    
7558

    
7559
class LURepairNodeStorage(NoHooksLU):
7560
  """Repairs the volume group on a node.
7561

7562
  """
7563
  _OP_REQP = ["node_name"]
7564
  REQ_BGL = False
7565

    
7566
  def CheckArguments(self):
7567
    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
7568

    
7569
  def ExpandNames(self):
7570
    self.needed_locks = {
7571
      locking.LEVEL_NODE: [self.op.node_name],
7572
      }
7573

    
7574
  def _CheckFaultyDisks(self, instance, node_name):
7575
    """Ensure faulty disks abort the opcode or at least warn."""
7576
    try:
7577
      if _FindFaultyInstanceDisks(self.cfg, self.rpc, instance,
7578
                                  node_name, True):
7579
        raise errors.OpPrereqError("Instance '%s' has faulty disks on"
7580
                                   " node '%s'" % (instance.name, node_name),
7581
                                   errors.ECODE_STATE)
7582
    except errors.OpPrereqError, err:
7583
      if self.op.ignore_consistency:
7584
        self.proc.LogWarning(str(err.args[0]))
7585
      else:
7586
        raise
7587

    
7588
  def CheckPrereq(self):
7589
    """Check prerequisites.
7590

7591
    """
7592
    storage_type = self.op.storage_type
7593

    
7594
    if (constants.SO_FIX_CONSISTENCY not in
7595
        constants.VALID_STORAGE_OPERATIONS.get(storage_type, [])):
7596
      raise errors.OpPrereqError("Storage units of type '%s' can not be"
7597
                                 " repaired" % storage_type,
7598
                                 errors.ECODE_INVAL)
7599

    
7600
    # Check whether any instance on this node has faulty disks
7601
    for inst in _GetNodeInstances(self.cfg, self.op.node_name):
7602
      if not inst.admin_up:
7603
        continue
7604
      check_nodes = set(inst.all_nodes)
7605
      check_nodes.discard(self.op.node_name)
7606
      for inst_node_name in check_nodes:
7607
        self._CheckFaultyDisks(inst, inst_node_name)
7608

    
7609
  def Exec(self, feedback_fn):
7610
    feedback_fn("Repairing storage unit '%s' on %s ..." %
7611
                (self.op.name, self.op.node_name))
7612

    
7613
    st_args = _GetStorageTypeArgs(self.cfg, self.op.storage_type)
7614
    result = self.rpc.call_storage_execute(self.op.node_name,
7615
                                           self.op.storage_type, st_args,
7616
                                           self.op.name,
7617
                                           constants.SO_FIX_CONSISTENCY)
7618
    result.Raise("Failed to repair storage unit '%s' on %s" %
7619
                 (self.op.name, self.op.node_name))
7620

    
7621

    
7622
class LUNodeEvacuationStrategy(NoHooksLU):
7623
  """Computes the node evacuation strategy.
7624

7625
  """
7626
  _OP_REQP = ["nodes"]
7627
  REQ_BGL = False
7628

    
7629
  def CheckArguments(self):
7630
    if not hasattr(self.op, "remote_node"):
7631
      self.op.remote_node = None
7632
    if not hasattr(self.op, "iallocator"):
7633
      self.op.iallocator = None
7634
    if self.op.remote_node is not None and self.op.iallocator is not None:
7635
      raise errors.OpPrereqError("Give either the iallocator or the new"
7636
                                 " secondary, not both", errors.ECODE_INVAL)
7637

    
7638
  def ExpandNames(self):
7639
    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
7640
    self.needed_locks = locks = {}
7641
    if self.op.remote_node is None:
7642
      locks[locking.LEVEL_NODE] = locking.ALL_SET
7643
    else:
7644
      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
7645
      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
7646

    
7647
  def CheckPrereq(self):
7648
    pass
7649

    
7650
  def Exec(self, feedback_fn):
7651
    if self.op.remote_node is not None:
7652
      instances = []
7653
      for node in self.op.nodes:
7654
        instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
7655
      result = []
7656
      for i in instances:
7657
        if i.primary_node == self.op.remote_node:
7658
          raise errors.OpPrereqError("Node %s is the primary node of"
7659
                                     " instance %s, cannot use it as"
7660
                                     " secondary" %
7661
                                     (self.op.remote_node, i.name),
7662
                                     errors.ECODE_INVAL)
7663
        result.append([i.name, self.op.remote_node])
7664
    else:
7665
      ial = IAllocator(self.cfg, self.rpc,
7666
                       mode=constants.IALLOCATOR_MODE_MEVAC,
7667
                       evac_nodes=self.op.nodes)
7668
      ial.Run(self.op.iallocator, validate=True)
7669
      if not ial.success:
7670
        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
7671
                                 errors.ECODE_NORES)
7672
      result = ial.result
7673
    return result
7674

    
7675

    
7676
class LUGrowDisk(LogicalUnit):
7677
  """Grow a disk of an instance.
7678

7679
  """
7680
  HPATH = "disk-grow"
7681
  HTYPE = constants.HTYPE_INSTANCE
7682
  _OP_REQP = ["instance_name", "disk", "amount", "wait_for_sync"]
7683
  REQ_BGL = False
7684

    
7685
  def ExpandNames(self):
7686
    self._ExpandAndLockInstance()
7687
    self.needed_locks[locking.LEVEL_NODE] = []
7688
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7689

    
7690
  def DeclareLocks(self, level):
7691
    if level == locking.LEVEL_NODE:
7692
      self._LockInstancesNodes()
7693

    
7694
  def BuildHooksEnv(self):
7695
    """Build hooks env.
7696

7697
    This runs on the master, the primary and all the secondaries.
7698

7699
    """
7700
    env = {
7701
      "DISK": self.op.disk,
7702
      "AMOUNT": self.op.amount,
7703
      }
7704
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
7705
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
7706
    return env, nl, nl
7707

    
7708
  def CheckPrereq(self):
7709
    """Check prerequisites.
7710

7711
    This checks that the instance is in the cluster.
7712

7713
    """
7714
    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
7715
    assert instance is not None, \
7716
      "Cannot retrieve locked instance %s" % self.op.instance_name
7717
    nodenames = list(instance.all_nodes)
7718
    for node in nodenames:
7719
      _CheckNodeOnline(self, node)
7720

    
7721

    
7722
    self.instance = instance
7723

    
7724
    if instance.disk_template not in constants.DTS_GROWABLE:
7725
      raise errors.OpPrereqError("Instance's disk layout does not support"
7726
                                 " growing.", errors.ECODE_INVAL)
7727

    
7728
    self.disk = instance.FindDisk(self.op.disk)
7729

    
7730
    if instance.disk_template != constants.DT_FILE:
7731
      # TODO: check the free disk space for file, when that feature will be
7732
      # supported
7733
      _CheckNodesFreeDisk(self, nodenames, self.op.amount)
7734

    
7735
  def Exec(self, feedback_fn):
7736
    """Execute disk grow.
7737

7738
    """
7739
    instance = self.instance
7740
    disk = self.disk
7741
    for node in instance.all_nodes:
7742
      self.cfg.SetDiskID(disk, node)
7743
      result = self.rpc.call_blockdev_grow(node, disk, self.op.amount)
7744
      result.Raise("Grow request failed to node %s" % node)
7745

    
7746
      # TODO: Rewrite code to work properly
7747
      # DRBD goes into sync mode for a short amount of time after executing the
7748
      # "resize" command. DRBD 8.x below version 8.0.13 contains a bug whereby
7749
      # calling "resize" in sync mode fails. Sleeping for a short amount of
7750
      # time is a work-around.
7751
      time.sleep(5)
7752

    
7753
    disk.RecordGrow(self.op.amount)
7754
    self.cfg.Update(instance, feedback_fn)
7755
    if self.op.wait_for_sync:
7756
      disk_abort = not _WaitForSync(self, instance)
7757
      if disk_abort:
7758
        self.proc.LogWarning("Warning: disk sync-ing has not returned a good"
7759
                             " status.\nPlease check the instance.")
7760

    
7761

    
7762
class LUQueryInstanceData(NoHooksLU):
7763
  """Query runtime instance data.
7764

7765
  """
7766
  _OP_REQP = ["instances", "static"]
7767
  REQ_BGL = False
7768

    
7769
  def ExpandNames(self):
7770
    self.needed_locks = {}
7771
    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
7772

    
7773
    if not isinstance(self.op.instances, list):
7774
      raise errors.OpPrereqError("Invalid argument type 'instances'",
7775
                                 errors.ECODE_INVAL)
7776

    
7777
    if self.op.instances:
7778
      self.wanted_names = []
7779
      for name in self.op.instances:
7780
        full_name = _ExpandInstanceName(self.cfg, name)
7781
        self.wanted_names.append(full_name)
7782
      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
7783
    else:
7784
      self.wanted_names = None
7785
      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
7786

    
7787
    self.needed_locks[locking.LEVEL_NODE] = []
7788
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
7789

    
7790
  def DeclareLocks(self, level):
7791
    if level == locking.LEVEL_NODE:
7792
      self._LockInstancesNodes()
7793

    
7794
  def CheckPrereq(self):
7795
    """Check prerequisites.
7796

7797
    This only checks the optional instance list against the existing names.
7798

7799
    """
7800
    if self.wanted_names is None:
7801
      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
7802

    
7803
    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
7804
                             in self.wanted_names]
7805
    return
7806

    
7807
  def _ComputeBlockdevStatus(self, node, instance_name, dev):
7808
    """Returns the status of a block device
7809

7810
    """
7811
    if self.op.static or not node:
7812
      return None
7813

    
7814
    self.cfg.SetDiskID(dev, node)
7815

    
7816
    result = self.rpc.call_blockdev_find(node, dev)
7817
    if result.offline:
7818
      return None
7819

    
7820
    result.Raise("Can't compute disk status for %s" % instance_name)
7821

    
7822
    status = result.payload
7823
    if status is None:
7824
      return None
7825

    
7826
    return (status.dev_path, status.major, status.minor,
7827
            status.sync_percent, status.estimated_time,
7828
            status.is_degraded, status.ldisk_status)
7829

    
7830
  def _ComputeDiskStatus(self, instance, snode, dev):
7831
    """Compute block device status.
7832

7833
    """
7834
    if dev.dev_type in constants.LDS_DRBD:
7835
      # we change the snode then (otherwise we use the one passed in)
7836
      if dev.logical_id[0] == instance.primary_node:
7837
        snode = dev.logical_id[1]
7838
      else:
7839
        snode = dev.logical_id[0]
7840

    
7841
    dev_pstatus = self._ComputeBlockdevStatus(instance.primary_node,
7842
                                              instance.name, dev)
7843
    dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
7844

    
7845
    if dev.children:
7846
      dev_children = [self._ComputeDiskStatus(instance, snode, child)
7847
                      for child in dev.children]
7848
    else:
7849
      dev_children = []
7850

    
7851
    data = {
7852
      "iv_name": dev.iv_name,
7853
      "dev_type": dev.dev_type,
7854
      "logical_id": dev.logical_id,
7855
      "physical_id": dev.physical_id,
7856
      "pstatus": dev_pstatus,
7857
      "sstatus": dev_sstatus,
7858
      "children": dev_children,
7859
      "mode": dev.mode,
7860
      "size": dev.size,
7861
      }
7862

    
7863
    return data
7864

    
7865
  def Exec(self, feedback_fn):
7866
    """Gather and return data"""
7867
    result = {}
7868

    
7869
    cluster = self.cfg.GetClusterInfo()
7870

    
7871
    for instance in self.wanted_instances:
7872
      if not self.op.static:
7873
        remote_info = self.rpc.call_instance_info(instance.primary_node,
7874
                                                  instance.name,
7875
                                                  instance.hypervisor)
7876
        remote_info.Raise("Error checking node %s" % instance.primary_node)
7877
        remote_info = remote_info.payload
7878
        if remote_info and "state" in remote_info:
7879
          remote_state = "up"
7880
        else:
7881
          remote_state = "down"
7882
      else:
7883
        remote_state = None
7884
      if instance.admin_up:
7885
        config_state = "up"
7886
      else:
7887
        config_state = "down"
7888

    
7889
      disks = [self._ComputeDiskStatus(instance, None, device)
7890
               for device in instance.disks]
7891

    
7892
      idict = {
7893
        "name": instance.name,
7894
        "config_state": config_state,
7895
        "run_state": remote_state,
7896
        "pnode": instance.primary_node,
7897
        "snodes": instance.secondary_nodes,
7898
        "os": instance.os,
7899
        # this happens to be the same format used for hooks
7900
        "nics": _NICListToTuple(self, instance.nics),
7901
        "disks": disks,
7902
        "hypervisor": instance.hypervisor,
7903
        "network_port": instance.network_port,
7904
        "hv_instance": instance.hvparams,
7905
        "hv_actual": cluster.FillHV(instance, skip_globals=True),
7906
        "be_instance": instance.beparams,
7907
        "be_actual": cluster.FillBE(instance),
7908
        "serial_no": instance.serial_no,
7909
        "mtime": instance.mtime,
7910
        "ctime": instance.ctime,
7911
        "uuid": instance.uuid,
7912
        }
7913

    
7914
      result[instance.name] = idict
7915

    
7916
    return result
7917

    
7918

    
7919
class LUSetInstanceParams(LogicalUnit):
7920
  """Modifies an instances's parameters.
7921

7922
  """
7923
  HPATH = "instance-modify"
7924
  HTYPE = constants.HTYPE_INSTANCE
7925
  _OP_REQP = ["instance_name"]
7926
  REQ_BGL = False
7927

    
7928
  def CheckArguments(self):
7929
    if not hasattr(self.op, 'nics'):
7930
      self.op.nics = []
7931
    if not hasattr(self.op, 'disks'):
7932
      self.op.disks = []
7933
    if not hasattr(self.op, 'beparams'):
7934
      self.op.beparams = {}
7935
    if not hasattr(self.op, 'hvparams'):
7936
      self.op.hvparams = {}
7937
    if not hasattr(self.op, "disk_template"):
7938
      self.op.disk_template = None
7939
    if not hasattr(self.op, "remote_node"):
7940
      self.op.remote_node = None
7941
    if not hasattr(self.op, "os_name"):
7942
      self.op.os_name = None
7943
    if not hasattr(self.op, "force_variant"):
7944
      self.op.force_variant = False
7945
    self.op.force = getattr(self.op, "force", False)
7946
    if not (self.op.nics or self.op.disks or self.op.disk_template or
7947
            self.op.hvparams or self.op.beparams or self.op.os_name):
7948
      raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
7949

    
7950
    if self.op.hvparams:
7951
      _CheckGlobalHvParams(self.op.hvparams)
7952

    
7953
    # Disk validation
7954
    disk_addremove = 0
7955
    for disk_op, disk_dict in self.op.disks:
7956
      if disk_op == constants.DDM_REMOVE:
7957
        disk_addremove += 1
7958
        continue
7959
      elif disk_op == constants.DDM_ADD:
7960
        disk_addremove += 1
7961
      else:
7962
        if not isinstance(disk_op, int):
7963
          raise errors.OpPrereqError("Invalid disk index", errors.ECODE_INVAL)
7964
        if not isinstance(disk_dict, dict):
7965
          msg = "Invalid disk value: expected dict, got '%s'" % disk_dict
7966
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
7967

    
7968
      if disk_op == constants.DDM_ADD:
7969
        mode = disk_dict.setdefault('mode', constants.DISK_RDWR)
7970
        if mode not in constants.DISK_ACCESS_SET:
7971
          raise errors.OpPrereqError("Invalid disk access mode '%s'" % mode,
7972
                                     errors.ECODE_INVAL)
7973
        size = disk_dict.get('size', None)
7974
        if size is None:
7975
          raise errors.OpPrereqError("Required disk parameter size missing",
7976
                                     errors.ECODE_INVAL)
7977
        try:
7978
          size = int(size)
7979
        except (TypeError, ValueError), err:
7980
          raise errors.OpPrereqError("Invalid disk size parameter: %s" %
7981
                                     str(err), errors.ECODE_INVAL)
7982
        disk_dict['size'] = size
7983
      else:
7984
        # modification of disk
7985
        if 'size' in disk_dict:
7986
          raise errors.OpPrereqError("Disk size change not possible, use"
7987
                                     " grow-disk", errors.ECODE_INVAL)
7988

    
7989
    if disk_addremove > 1:
7990
      raise errors.OpPrereqError("Only one disk add or remove operation"
7991
                                 " supported at a time", errors.ECODE_INVAL)
7992

    
7993
    if self.op.disks and self.op.disk_template is not None:
7994
      raise errors.OpPrereqError("Disk template conversion and other disk"
7995
                                 " changes not supported at the same time",
7996
                                 errors.ECODE_INVAL)
7997

    
7998
    if self.op.disk_template:
7999
      _CheckDiskTemplate(self.op.disk_template)
8000
      if (self.op.disk_template in constants.DTS_NET_MIRROR and
8001
          self.op.remote_node is None):
8002
        raise errors.OpPrereqError("Changing the disk template to a mirrored"
8003
                                   " one requires specifying a secondary node",
8004
                                   errors.ECODE_INVAL)
8005

    
8006
    # NIC validation
8007
    nic_addremove = 0
8008
    for nic_op, nic_dict in self.op.nics:
8009
      if nic_op == constants.DDM_REMOVE:
8010
        nic_addremove += 1
8011
        continue
8012
      elif nic_op == constants.DDM_ADD:
8013
        nic_addremove += 1
8014
      else:
8015
        if not isinstance(nic_op, int):
8016
          raise errors.OpPrereqError("Invalid nic index", errors.ECODE_INVAL)
8017
        if not isinstance(nic_dict, dict):
8018
          msg = "Invalid nic value: expected dict, got '%s'" % nic_dict
8019
          raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
8020

    
8021
      # nic_dict should be a dict
8022
      nic_ip = nic_dict.get('ip', None)
8023
      if nic_ip is not None:
8024
        if nic_ip.lower() == constants.VALUE_NONE:
8025
          nic_dict['ip'] = None
8026
        else:
8027
          if not utils.IsValidIP(nic_ip):
8028
            raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
8029
                                       errors.ECODE_INVAL)
8030

    
8031
      nic_bridge = nic_dict.get('bridge', None)
8032
      nic_link = nic_dict.get('link', None)
8033
      if nic_bridge and nic_link:
8034
        raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
8035
                                   " at the same time", errors.ECODE_INVAL)
8036
      elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
8037
        nic_dict['bridge'] = None
8038
      elif nic_link and nic_link.lower() == constants.VALUE_NONE:
8039
        nic_dict['link'] = None
8040

    
8041
      if nic_op == constants.DDM_ADD:
8042
        nic_mac = nic_dict.get('mac', None)
8043
        if nic_mac is None:
8044
          nic_dict['mac'] = constants.VALUE_AUTO
8045

    
8046
      if 'mac' in nic_dict:
8047
        nic_mac = nic_dict['mac']
8048
        if nic_mac not in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8049
          nic_mac = utils.NormalizeAndValidateMac(nic_mac)
8050

    
8051
        if nic_op != constants.DDM_ADD and nic_mac == constants.VALUE_AUTO:
8052
          raise errors.OpPrereqError("'auto' is not a valid MAC address when"
8053
                                     " modifying an existing nic",
8054
                                     errors.ECODE_INVAL)
8055

    
8056
    if nic_addremove > 1:
8057
      raise errors.OpPrereqError("Only one NIC add or remove operation"
8058
                                 " supported at a time", errors.ECODE_INVAL)
8059

    
8060
  def ExpandNames(self):
8061
    self._ExpandAndLockInstance()
8062
    self.needed_locks[locking.LEVEL_NODE] = []
8063
    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
8064

    
8065
  def DeclareLocks(self, level):
8066
    if level == locking.LEVEL_NODE:
8067
      self._LockInstancesNodes()
8068
      if self.op.disk_template and self.op.remote_node:
8069
        self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
8070
        self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
8071

    
8072
  def BuildHooksEnv(self):
8073
    """Build hooks env.
8074

8075
    This runs on the master, primary and secondaries.
8076

8077
    """
8078
    args = dict()
8079
    if constants.BE_MEMORY in self.be_new:
8080
      args['memory'] = self.be_new[constants.BE_MEMORY]
8081
    if constants.BE_VCPUS in self.be_new:
8082
      args['vcpus'] = self.be_new[constants.BE_VCPUS]
8083
    # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
8084
    # information at all.
8085
    if self.op.nics:
8086
      args['nics'] = []
8087
      nic_override = dict(self.op.nics)
8088
      c_nicparams = self.cluster.nicparams[constants.PP_DEFAULT]
8089
      for idx, nic in enumerate(self.instance.nics):
8090
        if idx in nic_override:
8091
          this_nic_override = nic_override[idx]
8092
        else:
8093
          this_nic_override = {}
8094
        if 'ip' in this_nic_override:
8095
          ip = this_nic_override['ip']
8096
        else:
8097
          ip = nic.ip
8098
        if 'mac' in this_nic_override:
8099
          mac = this_nic_override['mac']
8100
        else:
8101
          mac = nic.mac
8102
        if idx in self.nic_pnew:
8103
          nicparams = self.nic_pnew[idx]
8104
        else:
8105
          nicparams = objects.FillDict(c_nicparams, nic.nicparams)
8106
        mode = nicparams[constants.NIC_MODE]
8107
        link = nicparams[constants.NIC_LINK]
8108
        args['nics'].append((ip, mac, mode, link))
8109
      if constants.DDM_ADD in nic_override:
8110
        ip = nic_override[constants.DDM_ADD].get('ip', None)
8111
        mac = nic_override[constants.DDM_ADD]['mac']
8112
        nicparams = self.nic_pnew[constants.DDM_ADD]
8113
        mode = nicparams[constants.NIC_MODE]
8114
        link = nicparams[constants.NIC_LINK]
8115
        args['nics'].append((ip, mac, mode, link))
8116
      elif constants.DDM_REMOVE in nic_override:
8117
        del args['nics'][-1]
8118

    
8119
    env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
8120
    if self.op.disk_template:
8121
      env["NEW_DISK_TEMPLATE"] = self.op.disk_template
8122
    nl = [self.cfg.GetMasterNode()] + list(self.instance.all_nodes)
8123
    return env, nl, nl
8124

    
8125
  @staticmethod
8126
  def _GetUpdatedParams(old_params, update_dict,
8127
                        default_values, parameter_types):
8128
    """Return the new params dict for the given params.
8129

8130
    @type old_params: dict
8131
    @param old_params: old parameters
8132
    @type update_dict: dict
8133
    @param update_dict: dict containing new parameter values,
8134
                        or constants.VALUE_DEFAULT to reset the
8135
                        parameter to its default value
8136
    @type default_values: dict
8137
    @param default_values: default values for the filled parameters
8138
    @type parameter_types: dict
8139
    @param parameter_types: dict mapping target dict keys to types
8140
                            in constants.ENFORCEABLE_TYPES
8141
    @rtype: (dict, dict)
8142
    @return: (new_parameters, filled_parameters)
8143

8144
    """
8145
    params_copy = copy.deepcopy(old_params)
8146
    for key, val in update_dict.iteritems():
8147
      if val == constants.VALUE_DEFAULT:
8148
        try:
8149
          del params_copy[key]
8150
        except KeyError:
8151
          pass
8152
      else:
8153
        params_copy[key] = val
8154
    utils.ForceDictType(params_copy, parameter_types)
8155
    params_filled = objects.FillDict(default_values, params_copy)
8156
    return (params_copy, params_filled)
8157

    
8158
  def CheckPrereq(self):
8159
    """Check prerequisites.
8160

8161
    This only checks the instance list against the existing names.
8162

8163
    """
8164
    self.force = self.op.force
8165

    
8166
    # checking the new params on the primary/secondary nodes
8167

    
8168
    instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
8169
    cluster = self.cluster = self.cfg.GetClusterInfo()
8170
    assert self.instance is not None, \
8171
      "Cannot retrieve locked instance %s" % self.op.instance_name
8172
    pnode = instance.primary_node
8173
    nodelist = list(instance.all_nodes)
8174

    
8175
    if self.op.disk_template:
8176
      if instance.disk_template == self.op.disk_template:
8177
        raise errors.OpPrereqError("Instance already has disk template %s" %
8178
                                   instance.disk_template, errors.ECODE_INVAL)
8179

    
8180
      if (instance.disk_template,
8181
          self.op.disk_template) not in self._DISK_CONVERSIONS:
8182
        raise errors.OpPrereqError("Unsupported disk template conversion from"
8183
                                   " %s to %s" % (instance.disk_template,
8184
                                                  self.op.disk_template),
8185
                                   errors.ECODE_INVAL)
8186
      if self.op.disk_template in constants.DTS_NET_MIRROR:
8187
        _CheckNodeOnline(self, self.op.remote_node)
8188
        _CheckNodeNotDrained(self, self.op.remote_node)
8189
        disks = [{"size": d.size} for d in instance.disks]
8190
        required = _ComputeDiskSize(self.op.disk_template, disks)
8191
        _CheckNodesFreeDisk(self, [self.op.remote_node], required)
8192
        _CheckInstanceDown(self, instance, "cannot change disk template")
8193

    
8194
    # hvparams processing
8195
    if self.op.hvparams:
8196
      i_hvdict, hv_new = self._GetUpdatedParams(
8197
                             instance.hvparams, self.op.hvparams,
8198
                             cluster.hvparams[instance.hypervisor],
8199
                             constants.HVS_PARAMETER_TYPES)
8200
      # local check
8201
      hypervisor.GetHypervisor(
8202
        instance.hypervisor).CheckParameterSyntax(hv_new)
8203
      _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
8204
      self.hv_new = hv_new # the new actual values
8205
      self.hv_inst = i_hvdict # the new dict (without defaults)
8206
    else:
8207
      self.hv_new = self.hv_inst = {}
8208

    
8209
    # beparams processing
8210
    if self.op.beparams:
8211
      i_bedict, be_new = self._GetUpdatedParams(
8212
                             instance.beparams, self.op.beparams,
8213
                             cluster.beparams[constants.PP_DEFAULT],
8214
                             constants.BES_PARAMETER_TYPES)
8215
      self.be_new = be_new # the new actual values
8216
      self.be_inst = i_bedict # the new dict (without defaults)
8217
    else:
8218
      self.be_new = self.be_inst = {}
8219

    
8220
    self.warn = []
8221

    
8222
    if constants.BE_MEMORY in self.op.beparams and not self.force:
8223
      mem_check_list = [pnode]
8224
      if be_new[constants.BE_AUTO_BALANCE]:
8225
        # either we changed auto_balance to yes or it was from before
8226
        mem_check_list.extend(instance.secondary_nodes)
8227
      instance_info = self.rpc.call_instance_info(pnode, instance.name,
8228
                                                  instance.hypervisor)
8229
      nodeinfo = self.rpc.call_node_info(mem_check_list, self.cfg.GetVGName(),
8230
                                         instance.hypervisor)
8231
      pninfo = nodeinfo[pnode]
8232
      msg = pninfo.fail_msg
8233
      if msg:
8234
        # Assume the primary node is unreachable and go ahead
8235
        self.warn.append("Can't get info from primary node %s: %s" %
8236
                         (pnode,  msg))
8237
      elif not isinstance(pninfo.payload.get('memory_free', None), int):
8238
        self.warn.append("Node data from primary node %s doesn't contain"
8239
                         " free memory information" % pnode)
8240
      elif instance_info.fail_msg:
8241
        self.warn.append("Can't get instance runtime information: %s" %
8242
                        instance_info.fail_msg)
8243
      else:
8244
        if instance_info.payload:
8245
          current_mem = int(instance_info.payload['memory'])
8246
        else:
8247
          # Assume instance not running
8248
          # (there is a slight race condition here, but it's not very probable,
8249
          # and we have no other way to check)
8250
          current_mem = 0
8251
        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
8252
                    pninfo.payload['memory_free'])
8253
        if miss_mem > 0:
8254
          raise errors.OpPrereqError("This change will prevent the instance"
8255
                                     " from starting, due to %d MB of memory"
8256
                                     " missing on its primary node" % miss_mem,
8257
                                     errors.ECODE_NORES)
8258

    
8259
      if be_new[constants.BE_AUTO_BALANCE]:
8260
        for node, nres in nodeinfo.items():
8261
          if node not in instance.secondary_nodes:
8262
            continue
8263
          msg = nres.fail_msg
8264
          if msg:
8265
            self.warn.append("Can't get info from secondary node %s: %s" %
8266
                             (node, msg))
8267
          elif not isinstance(nres.payload.get('memory_free', None), int):
8268
            self.warn.append("Secondary node %s didn't return free"
8269
                             " memory information" % node)
8270
          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
8271
            self.warn.append("Not enough memory to failover instance to"
8272
                             " secondary node %s" % node)
8273

    
8274
    # NIC processing
8275
    self.nic_pnew = {}
8276
    self.nic_pinst = {}
8277
    for nic_op, nic_dict in self.op.nics:
8278
      if nic_op == constants.DDM_REMOVE:
8279
        if not instance.nics:
8280
          raise errors.OpPrereqError("Instance has no NICs, cannot remove",
8281
                                     errors.ECODE_INVAL)
8282
        continue
8283
      if nic_op != constants.DDM_ADD:
8284
        # an existing nic
8285
        if not instance.nics:
8286
          raise errors.OpPrereqError("Invalid NIC index %s, instance has"
8287
                                     " no NICs" % nic_op,
8288
                                     errors.ECODE_INVAL)
8289
        if nic_op < 0 or nic_op >= len(instance.nics):
8290
          raise errors.OpPrereqError("Invalid NIC index %s, valid values"
8291
                                     " are 0 to %d" %
8292
                                     (nic_op, len(instance.nics) - 1),
8293
                                     errors.ECODE_INVAL)
8294
        old_nic_params = instance.nics[nic_op].nicparams
8295
        old_nic_ip = instance.nics[nic_op].ip
8296
      else:
8297
        old_nic_params = {}
8298
        old_nic_ip = None
8299

    
8300
      update_params_dict = dict([(key, nic_dict[key])
8301
                                 for key in constants.NICS_PARAMETERS
8302
                                 if key in nic_dict])
8303

    
8304
      if 'bridge' in nic_dict:
8305
        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
8306

    
8307
      new_nic_params, new_filled_nic_params = \
8308
          self._GetUpdatedParams(old_nic_params, update_params_dict,
8309
                                 cluster.nicparams[constants.PP_DEFAULT],
8310
                                 constants.NICS_PARAMETER_TYPES)
8311
      objects.NIC.CheckParameterSyntax(new_filled_nic_params)
8312
      self.nic_pinst[nic_op] = new_nic_params
8313
      self.nic_pnew[nic_op] = new_filled_nic_params
8314
      new_nic_mode = new_filled_nic_params[constants.NIC_MODE]
8315

    
8316
      if new_nic_mode == constants.NIC_MODE_BRIDGED:
8317
        nic_bridge = new_filled_nic_params[constants.NIC_LINK]
8318
        msg = self.rpc.call_bridges_exist(pnode, [nic_bridge]).fail_msg
8319
        if msg:
8320
          msg = "Error checking bridges on node %s: %s" % (pnode, msg)
8321
          if self.force:
8322
            self.warn.append(msg)
8323
          else:
8324
            raise errors.OpPrereqError(msg, errors.ECODE_ENVIRON)
8325
      if new_nic_mode == constants.NIC_MODE_ROUTED:
8326
        if 'ip' in nic_dict:
8327
          nic_ip = nic_dict['ip']
8328
        else:
8329
          nic_ip = old_nic_ip
8330
        if nic_ip is None:
8331
          raise errors.OpPrereqError('Cannot set the nic ip to None'
8332
                                     ' on a routed nic', errors.ECODE_INVAL)
8333
      if 'mac' in nic_dict:
8334
        nic_mac = nic_dict['mac']
8335
        if nic_mac is None:
8336
          raise errors.OpPrereqError('Cannot set the nic mac to None',
8337
                                     errors.ECODE_INVAL)
8338
        elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
8339
          # otherwise generate the mac
8340
          nic_dict['mac'] = self.cfg.GenerateMAC(self.proc.GetECId())
8341
        else:
8342
          # or validate/reserve the current one
8343
          try:
8344
            self.cfg.ReserveMAC(nic_mac, self.proc.GetECId())
8345
          except errors.ReservationError:
8346
            raise errors.OpPrereqError("MAC address %s already in use"
8347
                                       " in cluster" % nic_mac,
8348
                                       errors.ECODE_NOTUNIQUE)
8349

    
8350
    # DISK processing
8351
    if self.op.disks and instance.disk_template == constants.DT_DISKLESS:
8352
      raise errors.OpPrereqError("Disk operations not supported for"
8353
                                 " diskless instances",
8354
                                 errors.ECODE_INVAL)
8355
    for disk_op, _ in self.op.disks:
8356
      if disk_op == constants.DDM_REMOVE:
8357
        if len(instance.disks) == 1:
8358
          raise errors.OpPrereqError("Cannot remove the last disk of"
8359
                                     " an instance", errors.ECODE_INVAL)
8360
        _CheckInstanceDown(self, instance, "cannot remove disks")
8361

    
8362
      if (disk_op == constants.DDM_ADD and
8363
          len(instance.nics) >= constants.MAX_DISKS):
8364
        raise errors.OpPrereqError("Instance has too many disks (%d), cannot"
8365
                                   " add more" % constants.MAX_DISKS,
8366
                                   errors.ECODE_STATE)
8367
      if disk_op not in (constants.DDM_ADD, constants.DDM_REMOVE):
8368
        # an existing disk
8369
        if disk_op < 0 or disk_op >= len(instance.disks):
8370
          raise errors.OpPrereqError("Invalid disk index %s, valid values"
8371
                                     " are 0 to %d" %
8372
                                     (disk_op, len(instance.disks)),
8373
                                     errors.ECODE_INVAL)
8374

    
8375
    # OS change
8376
    if self.op.os_name and not self.op.force:
8377
      _CheckNodeHasOS(self, instance.primary_node, self.op.os_name,
8378
                      self.op.force_variant)
8379

    
8380
    return
8381

    
8382
  def _ConvertPlainToDrbd(self, feedback_fn):
8383
    """Converts an instance from plain to drbd.
8384

8385
    """
8386
    feedback_fn("Converting template to drbd")
8387
    instance = self.instance
8388
    pnode = instance.primary_node
8389
    snode = self.op.remote_node
8390

    
8391
    # create a fake disk info for _GenerateDiskTemplate
8392
    disk_info = [{"size": d.size, "mode": d.mode} for d in instance.disks]
8393
    new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
8394
                                      instance.name, pnode, [snode],
8395
                                      disk_info, None, None, 0)
8396
    info = _GetInstanceInfoText(instance)
8397
    feedback_fn("Creating aditional volumes...")
8398
    # first, create the missing data and meta devices
8399
    for disk in new_disks:
8400
      # unfortunately this is... not too nice
8401
      _CreateSingleBlockDev(self, pnode, instance, disk.children[1],
8402
                            info, True)
8403
      for child in disk.children:
8404
        _CreateSingleBlockDev(self, snode, instance, child, info, True)
8405
    # at this stage, all new LVs have been created, we can rename the
8406
    # old ones
8407
    feedback_fn("Renaming original volumes...")
8408
    rename_list = [(o, n.children[0].logical_id)
8409
                   for (o, n) in zip(instance.disks, new_disks)]
8410
    result = self.rpc.call_blockdev_rename(pnode, rename_list)
8411
    result.Raise("Failed to rename original LVs")
8412

    
8413
    feedback_fn("Initializing DRBD devices...")
8414
    # all child devices are in place, we can now create the DRBD devices
8415
    for disk in new_disks:
8416
      for node in [pnode, snode]:
8417
        f_create = node == pnode
8418
        _CreateSingleBlockDev(self, node, instance, disk, info, f_create)
8419

    
8420
    # at this point, the instance has been modified
8421
    instance.disk_template = constants.DT_DRBD8
8422
    instance.disks = new_disks
8423
    self.cfg.Update(instance, feedback_fn)
8424

    
8425
    # disks are created, waiting for sync
8426
    disk_abort = not _WaitForSync(self, instance)
8427
    if disk_abort:
8428
      raise errors.OpExecError("There are some degraded disks for"
8429
                               " this instance, please cleanup manually")
8430

    
8431
  def _ConvertDrbdToPlain(self, feedback_fn):
8432
    """Converts an instance from drbd to plain.
8433

8434
    """
8435
    instance = self.instance
8436
    assert len(instance.secondary_nodes) == 1
8437
    pnode = instance.primary_node
8438
    snode = instance.secondary_nodes[0]
8439
    feedback_fn("Converting template to plain")
8440

    
8441
    old_disks = instance.disks
8442
    new_disks = [d.children[0] for d in old_disks]
8443

    
8444
    # copy over size and mode
8445
    for parent, child in zip(old_disks, new_disks):
8446
      child.size = parent.size
8447
      child.mode = parent.mode
8448

    
8449
    # update instance structure
8450
    instance.disks = new_disks
8451
    instance.disk_template = constants.DT_PLAIN
8452
    self.cfg.Update(instance, feedback_fn)
8453

    
8454
    feedback_fn("Removing volumes on the secondary node...")
8455
    for disk in old_disks:
8456
      self.cfg.SetDiskID(disk, snode)
8457
      msg = self.rpc.call_blockdev_remove(snode, disk).fail_msg
8458
      if msg:
8459
        self.LogWarning("Could not remove block device %s on node %s,"
8460
                        " continuing anyway: %s", disk.iv_name, snode, msg)
8461

    
8462
    feedback_fn("Removing unneeded volumes on the primary node...")
8463
    for idx, disk in enumerate(old_disks):
8464
      meta = disk.children[1]
8465
      self.cfg.SetDiskID(meta, pnode)
8466
      msg = self.rpc.call_blockdev_remove(pnode, meta).fail_msg
8467
      if msg:
8468
        self.LogWarning("Could not remove metadata for disk %d on node %s,"
8469
                        " continuing anyway: %s", idx, pnode, msg)
8470

    
8471

    
8472
  def Exec(self, feedback_fn):
8473
    """Modifies an instance.
8474

8475
    All parameters take effect only at the next restart of the instance.
8476

8477
    """
8478
    # Process here the warnings from CheckPrereq, as we don't have a
8479
    # feedback_fn there.
8480
    for warn in self.warn:
8481
      feedback_fn("WARNING: %s" % warn)
8482

    
8483
    result = []
8484
    instance = self.instance
8485
    # disk changes
8486
    for disk_op, disk_dict in self.op.disks:
8487
      if disk_op == constants.DDM_REMOVE:
8488
        # remove the last disk
8489
        device = instance.disks.pop()
8490
        device_idx = len(instance.disks)
8491
        for node, disk in device.ComputeNodeTree(instance.primary_node):
8492
          self.cfg.SetDiskID(disk, node)
8493
          msg = self.rpc.call_blockdev_remove(node, disk).fail_msg
8494
          if msg:
8495
            self.LogWarning("Could not remove disk/%d on node %s: %s,"
8496
                            " continuing anyway", device_idx, node, msg)
8497
        result.append(("disk/%d" % device_idx, "remove"))
8498
      elif disk_op == constants.DDM_ADD:
8499
        # add a new disk
8500
        if instance.disk_template == constants.DT_FILE:
8501
          file_driver, file_path = instance.disks[0].logical_id
8502
          file_path = os.path.dirname(file_path)
8503
        else:
8504
          file_driver = file_path = None
8505
        disk_idx_base = len(instance.disks)
8506
        new_disk = _GenerateDiskTemplate(self,
8507
                                         instance.disk_template,
8508
                                         instance.name, instance.primary_node,
8509
                                         instance.secondary_nodes,
8510
                                         [disk_dict],
8511
                                         file_path,
8512
                                         file_driver,
8513
                                         disk_idx_base)[0]
8514
        instance.disks.append(new_disk)
8515
        info = _GetInstanceInfoText(instance)
8516

    
8517
        logging.info("Creating volume %s for instance %s",
8518
                     new_disk.iv_name, instance.name)
8519
        # Note: this needs to be kept in sync with _CreateDisks
8520
        #HARDCODE
8521
        for node in instance.all_nodes:
8522
          f_create = node == instance.primary_node
8523
          try:
8524
            _CreateBlockDev(self, node, instance, new_disk,
8525
                            f_create, info, f_create)
8526
          except errors.OpExecError, err:
8527
            self.LogWarning("Failed to create volume %s (%s) on"
8528
                            " node %s: %s",
8529
                            new_disk.iv_name, new_disk, node, err)
8530
        result.append(("disk/%d" % disk_idx_base, "add:size=%s,mode=%s" %
8531
                       (new_disk.size, new_disk.mode)))
8532
      else:
8533
        # change a given disk
8534
        instance.disks[disk_op].mode = disk_dict['mode']
8535
        result.append(("disk.mode/%d" % disk_op, disk_dict['mode']))
8536

    
8537
    if self.op.disk_template:
8538
      r_shut = _ShutdownInstanceDisks(self, instance)
8539
      if not r_shut:
8540
        raise errors.OpExecError("Cannot shutdow instance disks, unable to"
8541
                                 " proceed with disk template conversion")
8542
      mode = (instance.disk_template, self.op.disk_template)
8543
      try:
8544
        self._DISK_CONVERSIONS[mode](self, feedback_fn)
8545
      except:
8546
        self.cfg.ReleaseDRBDMinors(instance.name)
8547
        raise
8548
      result.append(("disk_template", self.op.disk_template))
8549

    
8550
    # NIC changes
8551
    for nic_op, nic_dict in self.op.nics:
8552
      if nic_op == constants.DDM_REMOVE:
8553
        # remove the last nic
8554
        del instance.nics[-1]
8555
        result.append(("nic.%d" % len(instance.nics), "remove"))
8556
      elif nic_op == constants.DDM_ADD:
8557
        # mac and bridge should be set, by now
8558
        mac = nic_dict['mac']
8559
        ip = nic_dict.get('ip', None)
8560
        nicparams = self.nic_pinst[constants.DDM_ADD]
8561
        new_nic = objects.NIC(mac=mac, ip=ip, nicparams=nicparams)
8562
        instance.nics.append(new_nic)
8563
        result.append(("nic.%d" % (len(instance.nics) - 1),
8564
                       "add:mac=%s,ip=%s,mode=%s,link=%s" %
8565
                       (new_nic.mac, new_nic.ip,
8566
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_MODE],
8567
                        self.nic_pnew[constants.DDM_ADD][constants.NIC_LINK]
8568
                       )))
8569
      else:
8570
        for key in 'mac', 'ip':
8571
          if key in nic_dict:
8572
            setattr(instance.nics[nic_op], key, nic_dict[key])
8573
        if nic_op in self.nic_pinst:
8574
          instance.nics[nic_op].nicparams = self.nic_pinst[nic_op]
8575
        for key, val in nic_dict.iteritems():
8576
          result.append(("nic.%s/%d" % (key, nic_op), val))
8577

    
8578
    # hvparams changes
8579
    if self.op.hvparams:
8580
      instance.hvparams = self.hv_inst
8581
      for key, val in self.op.hvparams.iteritems():
8582
        result.append(("hv/%s" % key, val))
8583

    
8584
    # beparams changes
8585
    if self.op.beparams:
8586
      instance.beparams = self.be_inst
8587
      for key, val in self.op.beparams.iteritems():
8588
        result.append(("be/%s" % key, val))
8589

    
8590
    # OS change
8591
    if self.op.os_name:
8592
      instance.os = self.op.os_name
8593

    
8594
    self.cfg.Update(instance, feedback_fn)
8595

    
8596
    return result
8597

    
8598
  _DISK_CONVERSIONS = {
8599
    (constants.DT_PLAIN, constants.DT_DRBD8): _ConvertPlainToDrbd,
8600
    (constants.DT_DRBD8, constants.DT_PLAIN): _ConvertDrbdToPlain,
8601
    }
8602

    
8603
class LUQueryExports(NoHooksLU):
8604
  """Query the exports list
8605

8606
  """
8607
  _OP_REQP = ['nodes']
8608
  REQ_BGL = False
8609

    
8610
  def ExpandNames(self):
8611
    self.needed_locks = {}
8612
    self.share_locks[locking.LEVEL_NODE] = 1
8613
    if not self.op.nodes:
8614
      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8615
    else:
8616
      self.needed_locks[locking.LEVEL_NODE] = \
8617
        _GetWantedNodes(self, self.op.nodes)
8618

    
8619
  def CheckPrereq(self):
8620
    """Check prerequisites.
8621

8622
    """
8623
    self.nodes = self.acquired_locks[locking.LEVEL_NODE]
8624

    
8625
  def Exec(self, feedback_fn):
8626
    """Compute the list of all the exported system images.
8627

8628
    @rtype: dict
8629
    @return: a dictionary with the structure node->(export-list)
8630
        where export-list is a list of the instances exported on
8631
        that node.
8632

8633
    """
8634
    rpcresult = self.rpc.call_export_list(self.nodes)
8635
    result = {}
8636
    for node in rpcresult:
8637
      if rpcresult[node].fail_msg:
8638
        result[node] = False
8639
      else:
8640
        result[node] = rpcresult[node].payload
8641

    
8642
    return result
8643

    
8644

    
8645
class LUExportInstance(LogicalUnit):
8646
  """Export an instance to an image in the cluster.
8647

8648
  """
8649
  HPATH = "instance-export"
8650
  HTYPE = constants.HTYPE_INSTANCE
8651
  _OP_REQP = ["instance_name", "target_node", "shutdown"]
8652
  REQ_BGL = False
8653

    
8654
  def CheckArguments(self):
8655
    """Check the arguments.
8656

8657
    """
8658
    self.shutdown_timeout = getattr(self.op, "shutdown_timeout",
8659
                                    constants.DEFAULT_SHUTDOWN_TIMEOUT)
8660

    
8661
  def ExpandNames(self):
8662
    self._ExpandAndLockInstance()
8663
    # FIXME: lock only instance primary and destination node
8664
    #
8665
    # Sad but true, for now we have do lock all nodes, as we don't know where
8666
    # the previous export might be, and and in this LU we search for it and
8667
    # remove it from its current node. In the future we could fix this by:
8668
    #  - making a tasklet to search (share-lock all), then create the new one,
8669
    #    then one to remove, after
8670
    #  - removing the removal operation altogether
8671
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8672

    
8673
  def DeclareLocks(self, level):
8674
    """Last minute lock declaration."""
8675
    # All nodes are locked anyway, so nothing to do here.
8676

    
8677
  def BuildHooksEnv(self):
8678
    """Build hooks env.
8679

8680
    This will run on the master, primary node and target node.
8681

8682
    """
8683
    env = {
8684
      "EXPORT_NODE": self.op.target_node,
8685
      "EXPORT_DO_SHUTDOWN": self.op.shutdown,
8686
      "SHUTDOWN_TIMEOUT": self.shutdown_timeout,
8687
      }
8688
    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
8689
    nl = [self.cfg.GetMasterNode(), self.instance.primary_node,
8690
          self.op.target_node]
8691
    return env, nl, nl
8692

    
8693
  def CheckPrereq(self):
8694
    """Check prerequisites.
8695

8696
    This checks that the instance and node names are valid.
8697

8698
    """
8699
    instance_name = self.op.instance_name
8700
    self.instance = self.cfg.GetInstanceInfo(instance_name)
8701
    assert self.instance is not None, \
8702
          "Cannot retrieve locked instance %s" % self.op.instance_name
8703
    _CheckNodeOnline(self, self.instance.primary_node)
8704

    
8705
    self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
8706
    self.dst_node = self.cfg.GetNodeInfo(self.op.target_node)
8707
    assert self.dst_node is not None
8708

    
8709
    _CheckNodeOnline(self, self.dst_node.name)
8710
    _CheckNodeNotDrained(self, self.dst_node.name)
8711

    
8712
    # instance disk type verification
8713
    for disk in self.instance.disks:
8714
      if disk.dev_type == constants.LD_FILE:
8715
        raise errors.OpPrereqError("Export not supported for instances with"
8716
                                   " file-based disks", errors.ECODE_INVAL)
8717

    
8718
  def Exec(self, feedback_fn):
8719
    """Export an instance to an image in the cluster.
8720

8721
    """
8722
    instance = self.instance
8723
    dst_node = self.dst_node
8724
    src_node = instance.primary_node
8725

    
8726
    if self.op.shutdown:
8727
      # shutdown the instance, but not the disks
8728
      feedback_fn("Shutting down instance %s" % instance.name)
8729
      result = self.rpc.call_instance_shutdown(src_node, instance,
8730
                                               self.shutdown_timeout)
8731
      result.Raise("Could not shutdown instance %s on"
8732
                   " node %s" % (instance.name, src_node))
8733

    
8734
    vgname = self.cfg.GetVGName()
8735

    
8736
    snap_disks = []
8737

    
8738
    # set the disks ID correctly since call_instance_start needs the
8739
    # correct drbd minor to create the symlinks
8740
    for disk in instance.disks:
8741
      self.cfg.SetDiskID(disk, src_node)
8742

    
8743
    activate_disks = (not instance.admin_up)
8744

    
8745
    if activate_disks:
8746
      # Activate the instance disks if we'exporting a stopped instance
8747
      feedback_fn("Activating disks for %s" % instance.name)
8748
      _StartInstanceDisks(self, instance, None)
8749

    
8750
    try:
8751
      # per-disk results
8752
      dresults = []
8753
      try:
8754
        for idx, disk in enumerate(instance.disks):
8755
          feedback_fn("Creating a snapshot of disk/%s on node %s" %
8756
                      (idx, src_node))
8757

    
8758
          # result.payload will be a snapshot of an lvm leaf of the one we
8759
          # passed
8760
          result = self.rpc.call_blockdev_snapshot(src_node, disk)
8761
          msg = result.fail_msg
8762
          if msg:
8763
            self.LogWarning("Could not snapshot disk/%s on node %s: %s",
8764
                            idx, src_node, msg)
8765
            snap_disks.append(False)
8766
          else:
8767
            disk_id = (vgname, result.payload)
8768
            new_dev = objects.Disk(dev_type=constants.LD_LV, size=disk.size,
8769
                                   logical_id=disk_id, physical_id=disk_id,
8770
                                   iv_name=disk.iv_name)
8771
            snap_disks.append(new_dev)
8772

    
8773
      finally:
8774
        if self.op.shutdown and instance.admin_up:
8775
          feedback_fn("Starting instance %s" % instance.name)
8776
          result = self.rpc.call_instance_start(src_node, instance, None, None)
8777
          msg = result.fail_msg
8778
          if msg:
8779
            _ShutdownInstanceDisks(self, instance)
8780
            raise errors.OpExecError("Could not start instance: %s" % msg)
8781

    
8782
      # TODO: check for size
8783

    
8784
      cluster_name = self.cfg.GetClusterName()
8785
      for idx, dev in enumerate(snap_disks):
8786
        feedback_fn("Exporting snapshot %s from %s to %s" %
8787
                    (idx, src_node, dst_node.name))
8788
        if dev:
8789
          # FIXME: pass debug from opcode to backend
8790
          result = self.rpc.call_snapshot_export(src_node, dev, dst_node.name,
8791
                                                 instance, cluster_name,
8792
                                                 idx, self.op.debug_level)
8793
          msg = result.fail_msg
8794
          if msg:
8795
            self.LogWarning("Could not export disk/%s from node %s to"
8796
                            " node %s: %s", idx, src_node, dst_node.name, msg)
8797
            dresults.append(False)
8798
          else:
8799
            dresults.append(True)
8800
          msg = self.rpc.call_blockdev_remove(src_node, dev).fail_msg
8801
          if msg:
8802
            self.LogWarning("Could not remove snapshot for disk/%d from node"
8803
                            " %s: %s", idx, src_node, msg)
8804
        else:
8805
          dresults.append(False)
8806

    
8807
      feedback_fn("Finalizing export on %s" % dst_node.name)
8808
      result = self.rpc.call_finalize_export(dst_node.name, instance,
8809
                                             snap_disks)
8810
      fin_resu = True
8811
      msg = result.fail_msg
8812
      if msg:
8813
        self.LogWarning("Could not finalize export for instance %s"
8814
                        " on node %s: %s", instance.name, dst_node.name, msg)
8815
        fin_resu = False
8816

    
8817
    finally:
8818
      if activate_disks:
8819
        feedback_fn("Deactivating disks for %s" % instance.name)
8820
        _ShutdownInstanceDisks(self, instance)
8821

    
8822
    nodelist = self.cfg.GetNodeList()
8823
    nodelist.remove(dst_node.name)
8824

    
8825
    # on one-node clusters nodelist will be empty after the removal
8826
    # if we proceed the backup would be removed because OpQueryExports
8827
    # substitutes an empty list with the full cluster node list.
8828
    iname = instance.name
8829
    if nodelist:
8830
      feedback_fn("Removing old exports for instance %s" % iname)
8831
      exportlist = self.rpc.call_export_list(nodelist)
8832
      for node in exportlist:
8833
        if exportlist[node].fail_msg:
8834
          continue
8835
        if iname in exportlist[node].payload:
8836
          msg = self.rpc.call_export_remove(node, iname).fail_msg
8837
          if msg:
8838
            self.LogWarning("Could not remove older export for instance %s"
8839
                            " on node %s: %s", iname, node, msg)
8840
    return fin_resu, dresults
8841

    
8842

    
8843
class LURemoveExport(NoHooksLU):
8844
  """Remove exports related to the named instance.
8845

8846
  """
8847
  _OP_REQP = ["instance_name"]
8848
  REQ_BGL = False
8849

    
8850
  def ExpandNames(self):
8851
    self.needed_locks = {}
8852
    # We need all nodes to be locked in order for RemoveExport to work, but we
8853
    # don't need to lock the instance itself, as nothing will happen to it (and
8854
    # we can remove exports also for a removed instance)
8855
    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
8856

    
8857
  def CheckPrereq(self):
8858
    """Check prerequisites.
8859
    """
8860
    pass
8861

    
8862
  def Exec(self, feedback_fn):
8863
    """Remove any export.
8864

8865
    """
8866
    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
8867
    # If the instance was not found we'll try with the name that was passed in.
8868
    # This will only work if it was an FQDN, though.
8869
    fqdn_warn = False
8870
    if not instance_name:
8871
      fqdn_warn = True
8872
      instance_name = self.op.instance_name
8873

    
8874
    locked_nodes = self.acquired_locks[locking.LEVEL_NODE]
8875
    exportlist = self.rpc.call_export_list(locked_nodes)
8876
    found = False
8877
    for node in exportlist:
8878
      msg = exportlist[node].fail_msg
8879
      if msg:
8880
        self.LogWarning("Failed to query node %s (continuing): %s", node, msg)
8881
        continue
8882
      if instance_name in exportlist[node].payload:
8883
        found = True
8884
        result = self.rpc.call_export_remove(node, instance_name)
8885
        msg = result.fail_msg
8886
        if msg:
8887
          logging.error("Could not remove export for instance %s"
8888
                        " on node %s: %s", instance_name, node, msg)
8889

    
8890
    if fqdn_warn and not found:
8891
      feedback_fn("Export not found. If trying to remove an export belonging"
8892
                  " to a deleted instance please use its Fully Qualified"
8893
                  " Domain Name.")
8894

    
8895

    
8896
class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
8897
  """Generic tags LU.
8898

8899
  This is an abstract class which is the parent of all the other tags LUs.
8900

8901
  """
8902

    
8903
  def ExpandNames(self):
8904
    self.needed_locks = {}
8905
    if self.op.kind == constants.TAG_NODE:
8906
      self.op.name = _ExpandNodeName(self.cfg, self.op.name)
8907
      self.needed_locks[locking.LEVEL_NODE] = self.op.name
8908
    elif self.op.kind == constants.TAG_INSTANCE:
8909
      self.op.name = _ExpandInstanceName(self.cfg, self.op.name)
8910
      self.needed_locks[locking.LEVEL_INSTANCE] = self.op.name
8911

    
8912
  def CheckPrereq(self):
8913
    """Check prerequisites.
8914

8915
    """
8916
    if self.op.kind == constants.TAG_CLUSTER:
8917
      self.target = self.cfg.GetClusterInfo()
8918
    elif self.op.kind == constants.TAG_NODE:
8919
      self.target = self.cfg.GetNodeInfo(self.op.name)
8920
    elif self.op.kind == constants.TAG_INSTANCE:
8921
      self.target = self.cfg.GetInstanceInfo(self.op.name)
8922
    else:
8923
      raise errors.OpPrereqError("Wrong tag type requested (%s)" %
8924
                                 str(self.op.kind), errors.ECODE_INVAL)
8925

    
8926

    
8927
class LUGetTags(TagsLU):
8928
  """Returns the tags of a given object.
8929

8930
  """
8931
  _OP_REQP = ["kind", "name"]
8932
  REQ_BGL = False
8933

    
8934
  def Exec(self, feedback_fn):
8935
    """Returns the tag list.
8936

8937
    """
8938
    return list(self.target.GetTags())
8939

    
8940

    
8941
class LUSearchTags(NoHooksLU):
8942
  """Searches the tags for a given pattern.
8943

8944
  """
8945
  _OP_REQP = ["pattern"]
8946
  REQ_BGL = False
8947

    
8948
  def ExpandNames(self):
8949
    self.needed_locks = {}
8950

    
8951
  def CheckPrereq(self):
8952
    """Check prerequisites.
8953

8954
    This checks the pattern passed for validity by compiling it.
8955

8956
    """
8957
    try:
8958
      self.re = re.compile(self.op.pattern)
8959
    except re.error, err:
8960
      raise errors.OpPrereqError("Invalid search pattern '%s': %s" %
8961
                                 (self.op.pattern, err), errors.ECODE_INVAL)
8962

    
8963
  def Exec(self, feedback_fn):
8964
    """Returns the tag list.
8965

8966
    """
8967
    cfg = self.cfg
8968
    tgts = [("/cluster", cfg.GetClusterInfo())]
8969
    ilist = cfg.GetAllInstancesInfo().values()
8970
    tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
8971
    nlist = cfg.GetAllNodesInfo().values()
8972
    tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
8973
    results = []
8974
    for path, target in tgts:
8975
      for tag in target.GetTags():
8976
        if self.re.search(tag):
8977
          results.append((path, tag))
8978
    return results
8979

    
8980

    
8981
class LUAddTags(TagsLU):
8982
  """Sets a tag on a given object.
8983

8984
  """
8985
  _OP_REQP = ["kind", "name", "tags"]
8986
  REQ_BGL = False
8987

    
8988
  def CheckPrereq(self):
8989
    """Check prerequisites.
8990

8991
    This checks the type and length of the tag name and value.
8992

8993
    """
8994
    TagsLU.CheckPrereq(self)
8995
    for tag in self.op.tags:
8996
      objects.TaggableObject.ValidateTag(tag)
8997

    
8998
  def Exec(self, feedback_fn):
8999
    """Sets the tag.
9000

9001
    """
9002
    try:
9003
      for tag in self.op.tags:
9004
        self.target.AddTag(tag)
9005
    except errors.TagError, err:
9006
      raise errors.OpExecError("Error while setting tag: %s" % str(err))
9007
    self.cfg.Update(self.target, feedback_fn)
9008

    
9009

    
9010
class LUDelTags(TagsLU):
9011
  """Delete a list of tags from a given object.
9012

9013
  """
9014
  _OP_REQP = ["kind", "name", "tags"]
9015
  REQ_BGL = False
9016

    
9017
  def CheckPrereq(self):
9018
    """Check prerequisites.
9019

9020
    This checks that we have the given tag.
9021

9022
    """
9023
    TagsLU.CheckPrereq(self)
9024
    for tag in self.op.tags:
9025
      objects.TaggableObject.ValidateTag(tag)
9026
    del_tags = frozenset(self.op.tags)
9027
    cur_tags = self.target.GetTags()
9028
    if not del_tags <= cur_tags:
9029
      diff_tags = del_tags - cur_tags
9030
      diff_names = ["'%s'" % tag for tag in diff_tags]
9031
      diff_names.sort()
9032
      raise errors.OpPrereqError("Tag(s) %s not found" %
9033
                                 (",".join(diff_names)), errors.ECODE_NOENT)
9034

    
9035
  def Exec(self, feedback_fn):
9036
    """Remove the tag from the object.
9037

9038
    """
9039
    for tag in self.op.tags:
9040
      self.target.RemoveTag(tag)
9041
    self.cfg.Update(self.target, feedback_fn)
9042

    
9043

    
9044
class LUTestDelay(NoHooksLU):
9045
  """Sleep for a specified amount of time.
9046

9047
  This LU sleeps on the master and/or nodes for a specified amount of
9048
  time.
9049

9050
  """
9051
  _OP_REQP = ["duration", "on_master", "on_nodes"]
9052
  REQ_BGL = False
9053

    
9054
  def ExpandNames(self):
9055
    """Expand names and set required locks.
9056

9057
    This expands the node list, if any.
9058

9059
    """
9060
    self.needed_locks = {}
9061
    if self.op.on_nodes:
9062
      # _GetWantedNodes can be used here, but is not always appropriate to use
9063
      # this way in ExpandNames. Check LogicalUnit.ExpandNames docstring for
9064
      # more information.
9065
      self.op.on_nodes = _GetWantedNodes(self, self.op.on_nodes)
9066
      self.needed_locks[locking.LEVEL_NODE] = self.op.on_nodes
9067

    
9068
  def CheckPrereq(self):
9069
    """Check prerequisites.
9070

9071
    """
9072

    
9073
  def Exec(self, feedback_fn):
9074
    """Do the actual sleep.
9075

9076
    """
9077
    if self.op.on_master:
9078
      if not utils.TestDelay(self.op.duration):
9079
        raise errors.OpExecError("Error during master delay test")
9080
    if self.op.on_nodes:
9081
      result = self.rpc.call_test_delay(self.op.on_nodes, self.op.duration)
9082
      for node, node_result in result.items():
9083
        node_result.Raise("Failure during rpc call to node %s" % node)
9084

    
9085

    
9086
class IAllocator(object):
9087
  """IAllocator framework.
9088

9089
  An IAllocator instance has three sets of attributes:
9090
    - cfg that is needed to query the cluster
9091
    - input data (all members of the _KEYS class attribute are required)
9092
    - four buffer attributes (in|out_data|text), that represent the
9093
      input (to the external script) in text and data structure format,
9094
      and the output from it, again in two formats
9095
    - the result variables from the script (success, info, nodes) for
9096
      easy usage
9097

9098
  """
9099
  # pylint: disable-msg=R0902
9100
  # lots of instance attributes
9101
  _ALLO_KEYS = [
9102
    "name", "mem_size", "disks", "disk_template",
9103
    "os", "tags", "nics", "vcpus", "hypervisor",
9104
    ]
9105
  _RELO_KEYS = [
9106
    "name", "relocate_from",
9107
    ]
9108
  _EVAC_KEYS = [
9109
    "evac_nodes",
9110
    ]
9111

    
9112
  def __init__(self, cfg, rpc, mode, **kwargs):
9113
    self.cfg = cfg
9114
    self.rpc = rpc
9115
    # init buffer variables
9116
    self.in_text = self.out_text = self.in_data = self.out_data = None
9117
    # init all input fields so that pylint is happy
9118
    self.mode = mode
9119
    self.mem_size = self.disks = self.disk_template = None
9120
    self.os = self.tags = self.nics = self.vcpus = None
9121
    self.hypervisor = None
9122
    self.relocate_from = None
9123
    self.name = None
9124
    self.evac_nodes = None
9125
    # computed fields
9126
    self.required_nodes = None
9127
    # init result fields
9128
    self.success = self.info = self.result = None
9129
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
9130
      keyset = self._ALLO_KEYS
9131
      fn = self._AddNewInstance
9132
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
9133
      keyset = self._RELO_KEYS
9134
      fn = self._AddRelocateInstance
9135
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
9136
      keyset = self._EVAC_KEYS
9137
      fn = self._AddEvacuateNodes
9138
    else:
9139
      raise errors.ProgrammerError("Unknown mode '%s' passed to the"
9140
                                   " IAllocator" % self.mode)
9141
    for key in kwargs:
9142
      if key not in keyset:
9143
        raise errors.ProgrammerError("Invalid input parameter '%s' to"
9144
                                     " IAllocator" % key)
9145
      setattr(self, key, kwargs[key])
9146

    
9147
    for key in keyset:
9148
      if key not in kwargs:
9149
        raise errors.ProgrammerError("Missing input parameter '%s' to"
9150
                                     " IAllocator" % key)
9151
    self._BuildInputData(fn)
9152

    
9153
  def _ComputeClusterData(self):
9154
    """Compute the generic allocator input data.
9155

9156
    This is the data that is independent of the actual operation.
9157

9158
    """
9159
    cfg = self.cfg
9160
    cluster_info = cfg.GetClusterInfo()
9161
    # cluster data
9162
    data = {
9163
      "version": constants.IALLOCATOR_VERSION,
9164
      "cluster_name": cfg.GetClusterName(),
9165
      "cluster_tags": list(cluster_info.GetTags()),
9166
      "enabled_hypervisors": list(cluster_info.enabled_hypervisors),
9167
      # we don't have job IDs
9168
      }
9169
    iinfo = cfg.GetAllInstancesInfo().values()
9170
    i_list = [(inst, cluster_info.FillBE(inst)) for inst in iinfo]
9171

    
9172
    # node data
9173
    node_results = {}
9174
    node_list = cfg.GetNodeList()
9175

    
9176
    if self.mode == constants.IALLOCATOR_MODE_ALLOC:
9177
      hypervisor_name = self.hypervisor
9178
    elif self.mode == constants.IALLOCATOR_MODE_RELOC:
9179
      hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
9180
    elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
9181
      hypervisor_name = cluster_info.enabled_hypervisors[0]
9182

    
9183
    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
9184
                                        hypervisor_name)
9185
    node_iinfo = \
9186
      self.rpc.call_all_instances_info(node_list,
9187
                                       cluster_info.enabled_hypervisors)
9188
    for nname, nresult in node_data.items():
9189
      # first fill in static (config-based) values
9190
      ninfo = cfg.GetNodeInfo(nname)
9191
      pnr = {
9192
        "tags": list(ninfo.GetTags()),
9193
        "primary_ip": ninfo.primary_ip,
9194
        "secondary_ip": ninfo.secondary_ip,
9195
        "offline": ninfo.offline,
9196
        "drained": ninfo.drained,
9197
        "master_candidate": ninfo.master_candidate,
9198
        }
9199

    
9200
      if not (ninfo.offline or ninfo.drained):
9201
        nresult.Raise("Can't get data for node %s" % nname)
9202
        node_iinfo[nname].Raise("Can't get node instance info from node %s" %
9203
                                nname)
9204
        remote_info = nresult.payload
9205

    
9206
        for attr in ['memory_total', 'memory_free', 'memory_dom0',
9207
                     'vg_size', 'vg_free', 'cpu_total']:
9208
          if attr not in remote_info:
9209
            raise errors.OpExecError("Node '%s' didn't return attribute"
9210
                                     " '%s'" % (nname, attr))
9211
          if not isinstance(remote_info[attr], int):
9212
            raise errors.OpExecError("Node '%s' returned invalid value"
9213
                                     " for '%s': %s" %
9214
                                     (nname, attr, remote_info[attr]))
9215
        # compute memory used by primary instances
9216
        i_p_mem = i_p_up_mem = 0
9217
        for iinfo, beinfo in i_list:
9218
          if iinfo.primary_node == nname:
9219
            i_p_mem += beinfo[constants.BE_MEMORY]
9220
            if iinfo.name not in node_iinfo[nname].payload:
9221
              i_used_mem = 0
9222
            else:
9223
              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
9224
            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
9225
            remote_info['memory_free'] -= max(0, i_mem_diff)
9226

    
9227
            if iinfo.admin_up:
9228
              i_p_up_mem += beinfo[constants.BE_MEMORY]
9229

    
9230
        # compute memory used by instances
9231
        pnr_dyn = {
9232
          "total_memory": remote_info['memory_total'],
9233
          "reserved_memory": remote_info['memory_dom0'],
9234
          "free_memory": remote_info['memory_free'],
9235
          "total_disk": remote_info['vg_size'],
9236
          "free_disk": remote_info['vg_free'],
9237
          "total_cpus": remote_info['cpu_total'],
9238
          "i_pri_memory": i_p_mem,
9239
          "i_pri_up_memory": i_p_up_mem,
9240
          }
9241
        pnr.update(pnr_dyn)
9242

    
9243
      node_results[nname] = pnr
9244
    data["nodes"] = node_results
9245

    
9246
    # instance data
9247
    instance_data = {}
9248
    for iinfo, beinfo in i_list:
9249
      nic_data = []
9250
      for nic in iinfo.nics:
9251
        filled_params = objects.FillDict(
9252
            cluster_info.nicparams[constants.PP_DEFAULT],
9253
            nic.nicparams)
9254
        nic_dict = {"mac": nic.mac,
9255
                    "ip": nic.ip,
9256
                    "mode": filled_params[constants.NIC_MODE],
9257
                    "link": filled_params[constants.NIC_LINK],
9258
                   }
9259
        if filled_params[constants.NIC_MODE] == constants.NIC_MODE_BRIDGED:
9260
          nic_dict["bridge"] = filled_params[constants.NIC_LINK]
9261
        nic_data.append(nic_dict)
9262
      pir = {
9263
        "tags": list(iinfo.GetTags()),
9264
        "admin_up": iinfo.admin_up,
9265
        "vcpus": beinfo[constants.BE_VCPUS],
9266
        "memory": beinfo[constants.BE_MEMORY],
9267
        "os": iinfo.os,
9268
        "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
9269
        "nics": nic_data,
9270
        "disks": [{"size": dsk.size, "mode": dsk.mode} for dsk in iinfo.disks],
9271
        "disk_template": iinfo.disk_template,
9272
        "hypervisor": iinfo.hypervisor,
9273
        }
9274
      pir["disk_space_total"] = _ComputeDiskSize(iinfo.disk_template,
9275
                                                 pir["disks"])
9276
      instance_data[iinfo.name] = pir
9277

    
9278
    data["instances"] = instance_data
9279

    
9280
    self.in_data = data
9281

    
9282
  def _AddNewInstance(self):
9283
    """Add new instance data to allocator structure.
9284

9285
    This in combination with _AllocatorGetClusterData will create the
9286
    correct structure needed as input for the allocator.
9287

9288
    The checks for the completeness of the opcode must have already been
9289
    done.
9290

9291
    """
9292
    disk_space = _ComputeDiskSize(self.disk_template, self.disks)
9293

    
9294
    if self.disk_template in constants.DTS_NET_MIRROR:
9295
      self.required_nodes = 2
9296
    else:
9297
      self.required_nodes = 1
9298
    request = {
9299
      "name": self.name,
9300
      "disk_template": self.disk_template,
9301
      "tags": self.tags,
9302
      "os": self.os,
9303
      "vcpus": self.vcpus,
9304
      "memory": self.mem_size,
9305
      "disks": self.disks,
9306
      "disk_space_total": disk_space,
9307
      "nics": self.nics,
9308
      "required_nodes": self.required_nodes,
9309
      }
9310
    return request
9311

    
9312
  def _AddRelocateInstance(self):
9313
    """Add relocate instance data to allocator structure.
9314

9315
    This in combination with _IAllocatorGetClusterData will create the
9316
    correct structure needed as input for the allocator.
9317

9318
    The checks for the completeness of the opcode must have already been
9319
    done.
9320

9321
    """
9322
    instance = self.cfg.GetInstanceInfo(self.name)
9323
    if instance is None:
9324
      raise errors.ProgrammerError("Unknown instance '%s' passed to"
9325
                                   " IAllocator" % self.name)
9326

    
9327
    if instance.disk_template not in constants.DTS_NET_MIRROR:
9328
      raise errors.OpPrereqError("Can't relocate non-mirrored instances",
9329
                                 errors.ECODE_INVAL)
9330

    
9331
    if len(instance.secondary_nodes) != 1:
9332
      raise errors.OpPrereqError("Instance has not exactly one secondary node",
9333
                                 errors.ECODE_STATE)
9334

    
9335
    self.required_nodes = 1
9336
    disk_sizes = [{'size': disk.size} for disk in instance.disks]
9337
    disk_space = _ComputeDiskSize(instance.disk_template, disk_sizes)
9338

    
9339
    request = {
9340
      "name": self.name,
9341
      "disk_space_total": disk_space,
9342
      "required_nodes": self.required_nodes,
9343
      "relocate_from": self.relocate_from,
9344
      }
9345
    return request
9346

    
9347
  def _AddEvacuateNodes(self):
9348
    """Add evacuate nodes data to allocator structure.
9349

9350
    """
9351
    request = {
9352
      "evac_nodes": self.evac_nodes
9353
      }
9354
    return request
9355

    
9356
  def _BuildInputData(self, fn):
9357
    """Build input data structures.
9358

9359
    """
9360
    self._ComputeClusterData()
9361

    
9362
    request = fn()
9363
    request["type"] = self.mode
9364
    self.in_data["request"] = request
9365

    
9366
    self.in_text = serializer.Dump(self.in_data)
9367

    
9368
  def Run(self, name, validate=True, call_fn=None):
9369
    """Run an instance allocator and return the results.
9370

9371
    """
9372
    if call_fn is None:
9373
      call_fn = self.rpc.call_iallocator_runner
9374

    
9375
    result = call_fn(self.cfg.GetMasterNode(), name, self.in_text)
9376
    result.Raise("Failure while running the iallocator script")
9377

    
9378
    self.out_text = result.payload
9379
    if validate:
9380
      self._ValidateResult()
9381

    
9382
  def _ValidateResult(self):
9383
    """Process the allocator results.
9384

9385
    This will process and if successful save the result in
9386
    self.out_data and the other parameters.
9387

9388
    """
9389
    try:
9390
      rdict = serializer.Load(self.out_text)
9391
    except Exception, err:
9392
      raise errors.OpExecError("Can't parse iallocator results: %s" % str(err))
9393

    
9394
    if not isinstance(rdict, dict):
9395
      raise errors.OpExecError("Can't parse iallocator results: not a dict")
9396

    
9397
    # TODO: remove backwards compatiblity in later versions
9398
    if "nodes" in rdict and "result" not in rdict:
9399
      rdict["result"] = rdict["nodes"]
9400
      del rdict["nodes"]
9401

    
9402
    for key in "success", "info", "result":
9403
      if key not in rdict:
9404
        raise errors.OpExecError("Can't parse iallocator results:"
9405
                                 " missing key '%s'" % key)
9406
      setattr(self, key, rdict[key])
9407

    
9408
    if not isinstance(rdict["result"], list):
9409
      raise errors.OpExecError("Can't parse iallocator results: 'result' key"
9410
                               " is not a list")
9411
    self.out_data = rdict
9412

    
9413

    
9414
class LUTestAllocator(NoHooksLU):
9415
  """Run allocator tests.
9416

9417
  This LU runs the allocator tests
9418

9419
  """
9420
  _OP_REQP = ["direction", "mode", "name"]
9421

    
9422
  def CheckPrereq(self):
9423
    """Check prerequisites.
9424

9425
    This checks the opcode parameters depending on the director and mode test.
9426

9427
    """
9428
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
9429
      for attr in ["name", "mem_size", "disks", "disk_template",
9430
                   "os", "tags", "nics", "vcpus"]:
9431
        if not hasattr(self.op, attr):
9432
          raise errors.OpPrereqError("Missing attribute '%s' on opcode input" %
9433
                                     attr, errors.ECODE_INVAL)
9434
      iname = self.cfg.ExpandInstanceName(self.op.name)
9435
      if iname is not None:
9436
        raise errors.OpPrereqError("Instance '%s' already in the cluster" %
9437
                                   iname, errors.ECODE_EXISTS)
9438
      if not isinstance(self.op.nics, list):
9439
        raise errors.OpPrereqError("Invalid parameter 'nics'",
9440
                                   errors.ECODE_INVAL)
9441
      for row in self.op.nics:
9442
        if (not isinstance(row, dict) or
9443
            "mac" not in row or
9444
            "ip" not in row or
9445
            "bridge" not in row):
9446
          raise errors.OpPrereqError("Invalid contents of the 'nics'"
9447
                                     " parameter", errors.ECODE_INVAL)
9448
      if not isinstance(self.op.disks, list):
9449
        raise errors.OpPrereqError("Invalid parameter 'disks'",
9450
                                   errors.ECODE_INVAL)
9451
      for row in self.op.disks:
9452
        if (not isinstance(row, dict) or
9453
            "size" not in row or
9454
            not isinstance(row["size"], int) or
9455
            "mode" not in row or
9456
            row["mode"] not in ['r', 'w']):
9457
          raise errors.OpPrereqError("Invalid contents of the 'disks'"
9458
                                     " parameter", errors.ECODE_INVAL)
9459
      if not hasattr(self.op, "hypervisor") or self.op.hypervisor is None:
9460
        self.op.hypervisor = self.cfg.GetHypervisorType()
9461
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
9462
      if not hasattr(self.op, "name"):
9463
        raise errors.OpPrereqError("Missing attribute 'name' on opcode input",
9464
                                   errors.ECODE_INVAL)
9465
      fname = _ExpandInstanceName(self.cfg, self.op.name)
9466
      self.op.name = fname
9467
      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
9468
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
9469
      if not hasattr(self.op, "evac_nodes"):
9470
        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
9471
                                   " opcode input", errors.ECODE_INVAL)
9472
    else:
9473
      raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
9474
                                 self.op.mode, errors.ECODE_INVAL)
9475

    
9476
    if self.op.direction == constants.IALLOCATOR_DIR_OUT:
9477
      if not hasattr(self.op, "allocator") or self.op.allocator is None:
9478
        raise errors.OpPrereqError("Missing allocator name",
9479
                                   errors.ECODE_INVAL)
9480
    elif self.op.direction != constants.IALLOCATOR_DIR_IN:
9481
      raise errors.OpPrereqError("Wrong allocator test '%s'" %
9482
                                 self.op.direction, errors.ECODE_INVAL)
9483

    
9484
  def Exec(self, feedback_fn):
9485
    """Run the allocator test.
9486

9487
    """
9488
    if self.op.mode == constants.IALLOCATOR_MODE_ALLOC:
9489
      ial = IAllocator(self.cfg, self.rpc,
9490
                       mode=self.op.mode,
9491
                       name=self.op.name,
9492
                       mem_size=self.op.mem_size,
9493
                       disks=self.op.disks,
9494
                       disk_template=self.op.disk_template,
9495
                       os=self.op.os,
9496
                       tags=self.op.tags,
9497
                       nics=self.op.nics,
9498
                       vcpus=self.op.vcpus,
9499
                       hypervisor=self.op.hypervisor,
9500
                       )
9501
    elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
9502
      ial = IAllocator(self.cfg, self.rpc,
9503
                       mode=self.op.mode,
9504
                       name=self.op.name,
9505
                       relocate_from=list(self.relocate_from),
9506
                       )
9507
    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
9508
      ial = IAllocator(self.cfg, self.rpc,
9509
                       mode=self.op.mode,
9510
                       evac_nodes=self.op.evac_nodes)
9511
    else:
9512
      raise errors.ProgrammerError("Uncatched mode %s in"
9513
                                   " LUTestAllocator.Exec", self.op.mode)
9514

    
9515
    if self.op.direction == constants.IALLOCATOR_DIR_IN:
9516
      result = ial.in_text
9517
    else:
9518
      ial.Run(self.op.allocator, validate=False)
9519
      result = ial.out_text
9520
    return result