Statistics
| Branch: | Tag: | Revision:

root / lib / config.py @ 1d4a4b26

History | View | Annotate | Download (87.7 kB)

1
#
2
#
3

    
4
# Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013 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
"""Configuration management for Ganeti
23

24
This module provides the interface to the Ganeti cluster configuration.
25

26
The configuration data is stored on every node but is updated on the master
27
only. After each update, the master distributes the data to the other nodes.
28

29
Currently, the data storage format is JSON. YAML was slow and consuming too
30
much memory.
31

32
"""
33

    
34
# pylint: disable=R0904
35
# R0904: Too many public methods
36

    
37
import copy
38
import os
39
import random
40
import logging
41
import time
42
import itertools
43

    
44
from ganeti import errors
45
from ganeti import locking
46
from ganeti import utils
47
from ganeti import constants
48
from ganeti import rpc
49
from ganeti import objects
50
from ganeti import serializer
51
from ganeti import uidpool
52
from ganeti import netutils
53
from ganeti import runtime
54
from ganeti import pathutils
55
from ganeti import network
56

    
57

    
58
_config_lock = locking.SharedLock("ConfigWriter")
59

    
60
# job id used for resource management at config upgrade time
61
_UPGRADE_CONFIG_JID = "jid-cfg-upgrade"
62

    
63

    
64
def _ValidateConfig(data):
65
  """Verifies that a configuration objects looks valid.
66

67
  This only verifies the version of the configuration.
68

69
  @raise errors.ConfigurationError: if the version differs from what
70
      we expect
71

72
  """
73
  if data.version != constants.CONFIG_VERSION:
74
    raise errors.ConfigVersionMismatch(constants.CONFIG_VERSION, data.version)
75

    
76

    
77
class TemporaryReservationManager:
78
  """A temporary resource reservation manager.
79

80
  This is used to reserve resources in a job, before using them, making sure
81
  other jobs cannot get them in the meantime.
82

83
  """
84
  def __init__(self):
85
    self._ec_reserved = {}
86

    
87
  def Reserved(self, resource):
88
    for holder_reserved in self._ec_reserved.values():
89
      if resource in holder_reserved:
90
        return True
91
    return False
92

    
93
  def Reserve(self, ec_id, resource):
94
    if self.Reserved(resource):
95
      raise errors.ReservationError("Duplicate reservation for resource '%s'"
96
                                    % str(resource))
97
    if ec_id not in self._ec_reserved:
98
      self._ec_reserved[ec_id] = set([resource])
99
    else:
100
      self._ec_reserved[ec_id].add(resource)
101

    
102
  def DropECReservations(self, ec_id):
103
    if ec_id in self._ec_reserved:
104
      del self._ec_reserved[ec_id]
105

    
106
  def GetReserved(self):
107
    all_reserved = set()
108
    for holder_reserved in self._ec_reserved.values():
109
      all_reserved.update(holder_reserved)
110
    return all_reserved
111

    
112
  def GetECReserved(self, ec_id):
113
    """ Used when you want to retrieve all reservations for a specific
114
        execution context. E.g when commiting reserved IPs for a specific
115
        network.
116

117
    """
118
    ec_reserved = set()
119
    if ec_id in self._ec_reserved:
120
      ec_reserved.update(self._ec_reserved[ec_id])
121
    return ec_reserved
122

    
123
  def Generate(self, existing, generate_one_fn, ec_id):
124
    """Generate a new resource of this type
125

126
    """
127
    assert callable(generate_one_fn)
128

    
129
    all_elems = self.GetReserved()
130
    all_elems.update(existing)
131
    retries = 64
132
    while retries > 0:
133
      new_resource = generate_one_fn()
134
      if new_resource is not None and new_resource not in all_elems:
135
        break
136
    else:
137
      raise errors.ConfigurationError("Not able generate new resource"
138
                                      " (last tried: %s)" % new_resource)
139
    self.Reserve(ec_id, new_resource)
140
    return new_resource
141

    
142

    
143
def _MatchNameComponentIgnoreCase(short_name, names):
144
  """Wrapper around L{utils.text.MatchNameComponent}.
145

146
  """
147
  return utils.MatchNameComponent(short_name, names, case_sensitive=False)
148

    
149

    
150
def _CheckInstanceDiskIvNames(disks):
151
  """Checks if instance's disks' C{iv_name} attributes are in order.
152

153
  @type disks: list of L{objects.Disk}
154
  @param disks: List of disks
155
  @rtype: list of tuples; (int, string, string)
156
  @return: List of wrongly named disks, each tuple contains disk index,
157
    expected and actual name
158

159
  """
160
  result = []
161

    
162
  for (idx, disk) in enumerate(disks):
163
    exp_iv_name = "disk/%s" % idx
164
    if disk.iv_name != exp_iv_name:
165
      result.append((idx, exp_iv_name, disk.iv_name))
166

    
167
  return result
168

    
169

    
170
class ConfigWriter:
171
  """The interface to the cluster configuration.
172

173
  @ivar _temporary_lvs: reservation manager for temporary LVs
174
  @ivar _all_rms: a list of all temporary reservation managers
175

176
  """
177
  def __init__(self, cfg_file=None, offline=False, _getents=runtime.GetEnts,
178
               accept_foreign=False):
179
    self.write_count = 0
180
    self._lock = _config_lock
181
    self._config_data = None
182
    self._offline = offline
183
    if cfg_file is None:
184
      self._cfg_file = pathutils.CLUSTER_CONF_FILE
185
    else:
186
      self._cfg_file = cfg_file
187
    self._getents = _getents
188
    self._temporary_ids = TemporaryReservationManager()
189
    self._temporary_drbds = {}
190
    self._temporary_macs = TemporaryReservationManager()
191
    self._temporary_secrets = TemporaryReservationManager()
192
    self._temporary_lvs = TemporaryReservationManager()
193
    self._temporary_ips = TemporaryReservationManager()
194
    self._all_rms = [self._temporary_ids, self._temporary_macs,
195
                     self._temporary_secrets, self._temporary_lvs,
196
                     self._temporary_ips]
197
    # Note: in order to prevent errors when resolving our name in
198
    # _DistributeConfig, we compute it here once and reuse it; it's
199
    # better to raise an error before starting to modify the config
200
    # file than after it was modified
201
    self._my_hostname = netutils.Hostname.GetSysName()
202
    self._last_cluster_serial = -1
203
    self._cfg_id = None
204
    self._context = None
205
    self._OpenConfig(accept_foreign)
206

    
207
  def _GetRpc(self, address_list):
208
    """Returns RPC runner for configuration.
209

210
    """
211
    return rpc.ConfigRunner(self._context, address_list)
212

    
213
  def SetContext(self, context):
214
    """Sets Ganeti context.
215

216
    """
217
    self._context = context
218

    
219
  # this method needs to be static, so that we can call it on the class
220
  @staticmethod
221
  def IsCluster():
222
    """Check if the cluster is configured.
223

224
    """
225
    return os.path.exists(pathutils.CLUSTER_CONF_FILE)
226

    
227
  @locking.ssynchronized(_config_lock, shared=1)
228
  def GetNdParams(self, node):
229
    """Get the node params populated with cluster defaults.
230

231
    @type node: L{objects.Node}
232
    @param node: The node we want to know the params for
233
    @return: A dict with the filled in node params
234

235
    """
236
    nodegroup = self._UnlockedGetNodeGroup(node.group)
237
    return self._config_data.cluster.FillND(node, nodegroup)
238

    
239
  @locking.ssynchronized(_config_lock, shared=1)
240
  def GetInstanceDiskParams(self, instance):
241
    """Get the disk params populated with inherit chain.
242

243
    @type instance: L{objects.Instance}
244
    @param instance: The instance we want to know the params for
245
    @return: A dict with the filled in disk params
246

247
    """
248
    node = self._UnlockedGetNodeInfo(instance.primary_node)
249
    nodegroup = self._UnlockedGetNodeGroup(node.group)
250
    return self._UnlockedGetGroupDiskParams(nodegroup)
251

    
252
  @locking.ssynchronized(_config_lock, shared=1)
253
  def GetGroupDiskParams(self, group):
254
    """Get the disk params populated with inherit chain.
255

256
    @type group: L{objects.NodeGroup}
257
    @param group: The group we want to know the params for
258
    @return: A dict with the filled in disk params
259

260
    """
261
    return self._UnlockedGetGroupDiskParams(group)
262

    
263
  def _UnlockedGetGroupDiskParams(self, group):
264
    """Get the disk params populated with inherit chain down to node-group.
265

266
    @type group: L{objects.NodeGroup}
267
    @param group: The group we want to know the params for
268
    @return: A dict with the filled in disk params
269

270
    """
271
    return self._config_data.cluster.SimpleFillDP(group.diskparams)
272

    
273
  def _UnlockedGetNetworkMACPrefix(self, net_uuid):
274
    """Return the network mac prefix if it exists or the cluster level default.
275

276
    """
277
    prefix = None
278
    if net_uuid:
279
      nobj = self._UnlockedGetNetwork(net_uuid)
280
      if nobj.mac_prefix:
281
        prefix = nobj.mac_prefix
282

    
283
    return prefix
284

    
285
  def _GenerateOneMAC(self, prefix=None):
286
    """Return a function that randomly generates a MAC suffic
287
       and appends it to the given prefix. If prefix is not given get
288
       the cluster level default.
289

290
    """
291
    if not prefix:
292
      prefix = self._config_data.cluster.mac_prefix
293

    
294
    def GenMac():
295
      byte1 = random.randrange(0, 256)
296
      byte2 = random.randrange(0, 256)
297
      byte3 = random.randrange(0, 256)
298
      mac = "%s:%02x:%02x:%02x" % (prefix, byte1, byte2, byte3)
299
      return mac
300

    
301
    return GenMac
302

    
303
  @locking.ssynchronized(_config_lock, shared=1)
304
  def GenerateMAC(self, net_uuid, ec_id):
305
    """Generate a MAC for an instance.
306

307
    This should check the current instances for duplicates.
308

309
    """
310
    existing = self._AllMACs()
311
    prefix = self._UnlockedGetNetworkMACPrefix(net_uuid)
312
    gen_mac = self._GenerateOneMAC(prefix)
313
    return self._temporary_ids.Generate(existing, gen_mac, ec_id)
314

    
315
  @locking.ssynchronized(_config_lock, shared=1)
316
  def ReserveMAC(self, mac, ec_id):
317
    """Reserve a MAC for an instance.
318

319
    This only checks instances managed by this cluster, it does not
320
    check for potential collisions elsewhere.
321

322
    """
323
    all_macs = self._AllMACs()
324
    if mac in all_macs:
325
      raise errors.ReservationError("mac already in use")
326
    else:
327
      self._temporary_macs.Reserve(ec_id, mac)
328

    
329
  def _UnlockedCommitTemporaryIps(self, ec_id):
330
    """Commit all reserved IP address to their respective pools
331

332
    """
333
    for action, address, net_uuid in self._temporary_ips.GetECReserved(ec_id):
334
      self._UnlockedCommitIp(action, net_uuid, address)
335

    
336
  def _UnlockedCommitIp(self, action, net_uuid, address):
337
    """Commit a reserved IP address to an IP pool.
338

339
    The IP address is taken from the network's IP pool and marked as reserved.
340

341
    """
342
    nobj = self._UnlockedGetNetwork(net_uuid)
343
    pool = network.AddressPool(nobj)
344
    if action == constants.RESERVE_ACTION:
345
      pool.Reserve(address)
346
    elif action == constants.RELEASE_ACTION:
347
      pool.Release(address)
348

    
349
  def _UnlockedReleaseIp(self, net_uuid, address, ec_id):
350
    """Give a specific IP address back to an IP pool.
351

352
    The IP address is returned to the IP pool designated by pool_id and marked
353
    as reserved.
354

355
    """
356
    self._temporary_ips.Reserve(ec_id,
357
                                (constants.RELEASE_ACTION, address, net_uuid))
358

    
359
  @locking.ssynchronized(_config_lock, shared=1)
360
  def ReleaseIp(self, net_uuid, address, ec_id):
361
    """Give a specified IP address back to an IP pool.
362

363
    This is just a wrapper around _UnlockedReleaseIp.
364

365
    """
366
    if net_uuid:
367
      self._UnlockedReleaseIp(net_uuid, address, ec_id)
368

    
369
  @locking.ssynchronized(_config_lock, shared=1)
370
  def GenerateIp(self, net_uuid, ec_id):
371
    """Find a free IPv4 address for an instance.
372

373
    """
374
    nobj = self._UnlockedGetNetwork(net_uuid)
375
    pool = network.AddressPool(nobj)
376

    
377
    def gen_one():
378
      try:
379
        ip = pool.GenerateFree()
380
      except errors.AddressPoolError:
381
        raise errors.ReservationError("Cannot generate IP. Network is full")
382
      return (constants.RESERVE_ACTION, ip, net_uuid)
383

    
384
    _, address, _ = self._temporary_ips.Generate([], gen_one, ec_id)
385
    return address
386

    
387
  def _UnlockedReserveIp(self, net_uuid, address, ec_id):
388
    """Reserve a given IPv4 address for use by an instance.
389

390
    """
391
    nobj = self._UnlockedGetNetwork(net_uuid)
392
    pool = network.AddressPool(nobj)
393
    try:
394
      isreserved = pool.IsReserved(address)
395
    except errors.AddressPoolError:
396
      raise errors.ReservationError("IP address not in network")
397
    if isreserved:
398
      raise errors.ReservationError("IP address already in use")
399

    
400
    return self._temporary_ips.Reserve(ec_id,
401
                                       (constants.RESERVE_ACTION,
402
                                        address, net_uuid))
403

    
404
  @locking.ssynchronized(_config_lock, shared=1)
405
  def ReserveIp(self, net_uuid, address, ec_id):
406
    """Reserve a given IPv4 address for use by an instance.
407

408
    """
409
    if net_uuid:
410
      return self._UnlockedReserveIp(net_uuid, address, ec_id)
411

    
412
  @locking.ssynchronized(_config_lock, shared=1)
413
  def ReserveLV(self, lv_name, ec_id):
414
    """Reserve an VG/LV pair for an instance.
415

416
    @type lv_name: string
417
    @param lv_name: the logical volume name to reserve
418

419
    """
420
    all_lvs = self._AllLVs()
421
    if lv_name in all_lvs:
422
      raise errors.ReservationError("LV already in use")
423
    else:
424
      self._temporary_lvs.Reserve(ec_id, lv_name)
425

    
426
  @locking.ssynchronized(_config_lock, shared=1)
427
  def GenerateDRBDSecret(self, ec_id):
428
    """Generate a DRBD secret.
429

430
    This checks the current disks for duplicates.
431

432
    """
433
    return self._temporary_secrets.Generate(self._AllDRBDSecrets(),
434
                                            utils.GenerateSecret,
435
                                            ec_id)
436

    
437
  def _AllLVs(self):
438
    """Compute the list of all LVs.
439

440
    """
441
    lvnames = set()
442
    for instance in self._config_data.instances.values():
443
      node_data = instance.MapLVsByNode()
444
      for lv_list in node_data.values():
445
        lvnames.update(lv_list)
446
    return lvnames
447

    
448
  def _AllDisks(self):
449
    """Compute the list of all Disks.
450

451
    """
452
    disks = []
453
    for instance in self._config_data.instances.values():
454
      disks.extend(instance.disks)
455
    return disks
456

    
457
  def _AllNICs(self):
458
    """Compute the list of all NICs.
459

460
    """
461
    nics = []
462
    for instance in self._config_data.instances.values():
463
      nics.extend(instance.nics)
464
    return nics
465

    
466
  def _AllIDs(self, include_temporary):
467
    """Compute the list of all UUIDs and names we have.
468

469
    @type include_temporary: boolean
470
    @param include_temporary: whether to include the _temporary_ids set
471
    @rtype: set
472
    @return: a set of IDs
473

474
    """
475
    existing = set()
476
    if include_temporary:
477
      existing.update(self._temporary_ids.GetReserved())
478
    existing.update(self._AllLVs())
479
    existing.update(self._config_data.instances.keys())
480
    existing.update(self._config_data.nodes.keys())
481
    existing.update([i.uuid for i in self._AllUUIDObjects() if i.uuid])
482
    return existing
483

    
484
  def _GenerateUniqueID(self, ec_id):
485
    """Generate an unique UUID.
486

487
    This checks the current node, instances and disk names for
488
    duplicates.
489

490
    @rtype: string
491
    @return: the unique id
492

493
    """
494
    existing = self._AllIDs(include_temporary=False)
495
    return self._temporary_ids.Generate(existing, utils.NewUUID, ec_id)
496

    
497
  @locking.ssynchronized(_config_lock, shared=1)
498
  def GenerateUniqueID(self, ec_id):
499
    """Generate an unique ID.
500

501
    This is just a wrapper over the unlocked version.
502

503
    @type ec_id: string
504
    @param ec_id: unique id for the job to reserve the id to
505

506
    """
507
    return self._GenerateUniqueID(ec_id)
508

    
509
  def _AllMACs(self):
510
    """Return all MACs present in the config.
511

512
    @rtype: list
513
    @return: the list of all MACs
514

515
    """
516
    result = []
517
    for instance in self._config_data.instances.values():
518
      for nic in instance.nics:
519
        result.append(nic.mac)
520

    
521
    return result
522

    
523
  def _AllDRBDSecrets(self):
524
    """Return all DRBD secrets present in the config.
525

526
    @rtype: list
527
    @return: the list of all DRBD secrets
528

529
    """
530
    def helper(disk, result):
531
      """Recursively gather secrets from this disk."""
532
      if disk.dev_type == constants.DT_DRBD8:
533
        result.append(disk.logical_id[5])
534
      if disk.children:
535
        for child in disk.children:
536
          helper(child, result)
537

    
538
    result = []
539
    for instance in self._config_data.instances.values():
540
      for disk in instance.disks:
541
        helper(disk, result)
542

    
543
    return result
544

    
545
  def _CheckDiskIDs(self, disk, l_ids, p_ids):
546
    """Compute duplicate disk IDs
547

548
    @type disk: L{objects.Disk}
549
    @param disk: the disk at which to start searching
550
    @type l_ids: list
551
    @param l_ids: list of current logical ids
552
    @type p_ids: list
553
    @param p_ids: list of current physical ids
554
    @rtype: list
555
    @return: a list of error messages
556

557
    """
558
    result = []
559
    if disk.logical_id is not None:
560
      if disk.logical_id in l_ids:
561
        result.append("duplicate logical id %s" % str(disk.logical_id))
562
      else:
563
        l_ids.append(disk.logical_id)
564
    if disk.physical_id is not None:
565
      if disk.physical_id in p_ids:
566
        result.append("duplicate physical id %s" % str(disk.physical_id))
567
      else:
568
        p_ids.append(disk.physical_id)
569

    
570
    if disk.children:
571
      for child in disk.children:
572
        result.extend(self._CheckDiskIDs(child, l_ids, p_ids))
573
    return result
574

    
575
  def _UnlockedVerifyConfig(self):
576
    """Verify function.
577

578
    @rtype: list
579
    @return: a list of error messages; a non-empty list signifies
580
        configuration errors
581

582
    """
583
    # pylint: disable=R0914
584
    result = []
585
    seen_macs = []
586
    ports = {}
587
    data = self._config_data
588
    cluster = data.cluster
589
    seen_lids = []
590
    seen_pids = []
591

    
592
    # global cluster checks
593
    if not cluster.enabled_hypervisors:
594
      result.append("enabled hypervisors list doesn't have any entries")
595
    invalid_hvs = set(cluster.enabled_hypervisors) - constants.HYPER_TYPES
596
    if invalid_hvs:
597
      result.append("enabled hypervisors contains invalid entries: %s" %
598
                    utils.CommaJoin(invalid_hvs))
599
    missing_hvp = (set(cluster.enabled_hypervisors) -
600
                   set(cluster.hvparams.keys()))
601
    if missing_hvp:
602
      result.append("hypervisor parameters missing for the enabled"
603
                    " hypervisor(s) %s" % utils.CommaJoin(missing_hvp))
604

    
605
    if not cluster.enabled_disk_templates:
606
      result.append("enabled disk templates list doesn't have any entries")
607
    invalid_disk_templates = set(cluster.enabled_disk_templates) \
608
                               - constants.DISK_TEMPLATES
609
    if invalid_disk_templates:
610
      result.append("enabled disk templates list contains invalid entries:"
611
                    " %s" % utils.CommaJoin(invalid_disk_templates))
612

    
613
    if cluster.master_node not in data.nodes:
614
      result.append("cluster has invalid primary node '%s'" %
615
                    cluster.master_node)
616

    
617
    def _helper(owner, attr, value, template):
618
      try:
619
        utils.ForceDictType(value, template)
620
      except errors.GenericError, err:
621
        result.append("%s has invalid %s: %s" % (owner, attr, err))
622

    
623
    def _helper_nic(owner, params):
624
      try:
625
        objects.NIC.CheckParameterSyntax(params)
626
      except errors.ConfigurationError, err:
627
        result.append("%s has invalid nicparams: %s" % (owner, err))
628

    
629
    def _helper_ipolicy(owner, ipolicy, iscluster):
630
      try:
631
        objects.InstancePolicy.CheckParameterSyntax(ipolicy, iscluster)
632
      except errors.ConfigurationError, err:
633
        result.append("%s has invalid instance policy: %s" % (owner, err))
634
      for key, value in ipolicy.items():
635
        if key == constants.ISPECS_MINMAX:
636
          for k in range(len(value)):
637
            _helper_ispecs(owner, "ipolicy/%s[%s]" % (key, k), value[k])
638
        elif key == constants.ISPECS_STD:
639
          _helper(owner, "ipolicy/" + key, value,
640
                  constants.ISPECS_PARAMETER_TYPES)
641
        else:
642
          # FIXME: assuming list type
643
          if key in constants.IPOLICY_PARAMETERS:
644
            exp_type = float
645
          else:
646
            exp_type = list
647
          if not isinstance(value, exp_type):
648
            result.append("%s has invalid instance policy: for %s,"
649
                          " expecting %s, got %s" %
650
                          (owner, key, exp_type.__name__, type(value)))
651

    
652
    def _helper_ispecs(owner, parentkey, params):
653
      for (key, value) in params.items():
654
        fullkey = "/".join([parentkey, key])
655
        _helper(owner, fullkey, value, constants.ISPECS_PARAMETER_TYPES)
656

    
657
    # check cluster parameters
658
    _helper("cluster", "beparams", cluster.SimpleFillBE({}),
659
            constants.BES_PARAMETER_TYPES)
660
    _helper("cluster", "nicparams", cluster.SimpleFillNIC({}),
661
            constants.NICS_PARAMETER_TYPES)
662
    _helper_nic("cluster", cluster.SimpleFillNIC({}))
663
    _helper("cluster", "ndparams", cluster.SimpleFillND({}),
664
            constants.NDS_PARAMETER_TYPES)
665
    _helper_ipolicy("cluster", cluster.ipolicy, True)
666

    
667
    # per-instance checks
668
    for instance_name in data.instances:
669
      instance = data.instances[instance_name]
670
      if instance.name != instance_name:
671
        result.append("instance '%s' is indexed by wrong name '%s'" %
672
                      (instance.name, instance_name))
673
      if instance.primary_node not in data.nodes:
674
        result.append("instance '%s' has invalid primary node '%s'" %
675
                      (instance_name, instance.primary_node))
676
      for snode in instance.secondary_nodes:
677
        if snode not in data.nodes:
678
          result.append("instance '%s' has invalid secondary node '%s'" %
679
                        (instance_name, snode))
680
      for idx, nic in enumerate(instance.nics):
681
        if nic.mac in seen_macs:
682
          result.append("instance '%s' has NIC %d mac %s duplicate" %
683
                        (instance_name, idx, nic.mac))
684
        else:
685
          seen_macs.append(nic.mac)
686
        if nic.nicparams:
687
          filled = cluster.SimpleFillNIC(nic.nicparams)
688
          owner = "instance %s nic %d" % (instance.name, idx)
689
          _helper(owner, "nicparams",
690
                  filled, constants.NICS_PARAMETER_TYPES)
691
          _helper_nic(owner, filled)
692

    
693
      # disk template checks
694
      if not instance.disk_template in data.cluster.enabled_disk_templates:
695
        result.append("instance '%s' uses the disabled disk template '%s'." %
696
                      (instance_name, instance.disk_template))
697

    
698
      # parameter checks
699
      if instance.beparams:
700
        _helper("instance %s" % instance.name, "beparams",
701
                cluster.FillBE(instance), constants.BES_PARAMETER_TYPES)
702

    
703
      # gather the drbd ports for duplicate checks
704
      for (idx, dsk) in enumerate(instance.disks):
705
        if dsk.dev_type in constants.LDS_DRBD:
706
          tcp_port = dsk.logical_id[2]
707
          if tcp_port not in ports:
708
            ports[tcp_port] = []
709
          ports[tcp_port].append((instance.name, "drbd disk %s" % idx))
710
      # gather network port reservation
711
      net_port = getattr(instance, "network_port", None)
712
      if net_port is not None:
713
        if net_port not in ports:
714
          ports[net_port] = []
715
        ports[net_port].append((instance.name, "network port"))
716

    
717
      # instance disk verify
718
      for idx, disk in enumerate(instance.disks):
719
        result.extend(["instance '%s' disk %d error: %s" %
720
                       (instance.name, idx, msg) for msg in disk.Verify()])
721
        result.extend(self._CheckDiskIDs(disk, seen_lids, seen_pids))
722

    
723
      wrong_names = _CheckInstanceDiskIvNames(instance.disks)
724
      if wrong_names:
725
        tmp = "; ".join(("name of disk %s should be '%s', but is '%s'" %
726
                         (idx, exp_name, actual_name))
727
                        for (idx, exp_name, actual_name) in wrong_names)
728

    
729
        result.append("Instance '%s' has wrongly named disks: %s" %
730
                      (instance.name, tmp))
731

    
732
    # cluster-wide pool of free ports
733
    for free_port in cluster.tcpudp_port_pool:
734
      if free_port not in ports:
735
        ports[free_port] = []
736
      ports[free_port].append(("cluster", "port marked as free"))
737

    
738
    # compute tcp/udp duplicate ports
739
    keys = ports.keys()
740
    keys.sort()
741
    for pnum in keys:
742
      pdata = ports[pnum]
743
      if len(pdata) > 1:
744
        txt = utils.CommaJoin(["%s/%s" % val for val in pdata])
745
        result.append("tcp/udp port %s has duplicates: %s" % (pnum, txt))
746

    
747
    # highest used tcp port check
748
    if keys:
749
      if keys[-1] > cluster.highest_used_port:
750
        result.append("Highest used port mismatch, saved %s, computed %s" %
751
                      (cluster.highest_used_port, keys[-1]))
752

    
753
    if not data.nodes[cluster.master_node].master_candidate:
754
      result.append("Master node is not a master candidate")
755

    
756
    # master candidate checks
757
    mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats()
758
    if mc_now < mc_max:
759
      result.append("Not enough master candidates: actual %d, target %d" %
760
                    (mc_now, mc_max))
761

    
762
    # node checks
763
    for node_name, node in data.nodes.items():
764
      if node.name != node_name:
765
        result.append("Node '%s' is indexed by wrong name '%s'" %
766
                      (node.name, node_name))
767
      if [node.master_candidate, node.drained, node.offline].count(True) > 1:
768
        result.append("Node %s state is invalid: master_candidate=%s,"
769
                      " drain=%s, offline=%s" %
770
                      (node.name, node.master_candidate, node.drained,
771
                       node.offline))
772
      if node.group not in data.nodegroups:
773
        result.append("Node '%s' has invalid group '%s'" %
774
                      (node.name, node.group))
775
      else:
776
        _helper("node %s" % node.name, "ndparams",
777
                cluster.FillND(node, data.nodegroups[node.group]),
778
                constants.NDS_PARAMETER_TYPES)
779
      used_globals = constants.NDC_GLOBALS.intersection(node.ndparams)
780
      if used_globals:
781
        result.append("Node '%s' has some global parameters set: %s" %
782
                      (node.name, utils.CommaJoin(used_globals)))
783

    
784
    # nodegroups checks
785
    nodegroups_names = set()
786
    for nodegroup_uuid in data.nodegroups:
787
      nodegroup = data.nodegroups[nodegroup_uuid]
788
      if nodegroup.uuid != nodegroup_uuid:
789
        result.append("node group '%s' (uuid: '%s') indexed by wrong uuid '%s'"
790
                      % (nodegroup.name, nodegroup.uuid, nodegroup_uuid))
791
      if utils.UUID_RE.match(nodegroup.name.lower()):
792
        result.append("node group '%s' (uuid: '%s') has uuid-like name" %
793
                      (nodegroup.name, nodegroup.uuid))
794
      if nodegroup.name in nodegroups_names:
795
        result.append("duplicate node group name '%s'" % nodegroup.name)
796
      else:
797
        nodegroups_names.add(nodegroup.name)
798
      group_name = "group %s" % nodegroup.name
799
      _helper_ipolicy(group_name, cluster.SimpleFillIPolicy(nodegroup.ipolicy),
800
                      False)
801
      if nodegroup.ndparams:
802
        _helper(group_name, "ndparams",
803
                cluster.SimpleFillND(nodegroup.ndparams),
804
                constants.NDS_PARAMETER_TYPES)
805

    
806
    # drbd minors check
807
    _, duplicates = self._UnlockedComputeDRBDMap()
808
    for node, minor, instance_a, instance_b in duplicates:
809
      result.append("DRBD minor %d on node %s is assigned twice to instances"
810
                    " %s and %s" % (minor, node, instance_a, instance_b))
811

    
812
    # IP checks
813
    default_nicparams = cluster.nicparams[constants.PP_DEFAULT]
814
    ips = {}
815

    
816
    def _AddIpAddress(ip, name):
817
      ips.setdefault(ip, []).append(name)
818

    
819
    _AddIpAddress(cluster.master_ip, "cluster_ip")
820

    
821
    for node in data.nodes.values():
822
      _AddIpAddress(node.primary_ip, "node:%s/primary" % node.name)
823
      if node.secondary_ip != node.primary_ip:
824
        _AddIpAddress(node.secondary_ip, "node:%s/secondary" % node.name)
825

    
826
    for instance in data.instances.values():
827
      for idx, nic in enumerate(instance.nics):
828
        if nic.ip is None:
829
          continue
830

    
831
        nicparams = objects.FillDict(default_nicparams, nic.nicparams)
832
        nic_mode = nicparams[constants.NIC_MODE]
833
        nic_link = nicparams[constants.NIC_LINK]
834

    
835
        if nic_mode == constants.NIC_MODE_BRIDGED:
836
          link = "bridge:%s" % nic_link
837
        elif nic_mode == constants.NIC_MODE_ROUTED:
838
          link = "route:%s" % nic_link
839
        else:
840
          raise errors.ProgrammerError("NIC mode '%s' not handled" % nic_mode)
841

    
842
        _AddIpAddress("%s/%s/%s" % (link, nic.ip, nic.network),
843
                      "instance:%s/nic:%d" % (instance.name, idx))
844

    
845
    for ip, owners in ips.items():
846
      if len(owners) > 1:
847
        result.append("IP address %s is used by multiple owners: %s" %
848
                      (ip, utils.CommaJoin(owners)))
849

    
850
    return result
851

    
852
  @locking.ssynchronized(_config_lock, shared=1)
853
  def VerifyConfig(self):
854
    """Verify function.
855

856
    This is just a wrapper over L{_UnlockedVerifyConfig}.
857

858
    @rtype: list
859
    @return: a list of error messages; a non-empty list signifies
860
        configuration errors
861

862
    """
863
    return self._UnlockedVerifyConfig()
864

    
865
  def _UnlockedSetDiskID(self, disk, node_name):
866
    """Convert the unique ID to the ID needed on the target nodes.
867

868
    This is used only for drbd, which needs ip/port configuration.
869

870
    The routine descends down and updates its children also, because
871
    this helps when the only the top device is passed to the remote
872
    node.
873

874
    This function is for internal use, when the config lock is already held.
875

876
    """
877
    if disk.children:
878
      for child in disk.children:
879
        self._UnlockedSetDiskID(child, node_name)
880

    
881
    if disk.logical_id is None and disk.physical_id is not None:
882
      return
883
    if disk.dev_type == constants.LD_DRBD8:
884
      pnode, snode, port, pminor, sminor, secret = disk.logical_id
885
      if node_name not in (pnode, snode):
886
        raise errors.ConfigurationError("DRBD device not knowing node %s" %
887
                                        node_name)
888
      pnode_info = self._UnlockedGetNodeInfo(pnode)
889
      snode_info = self._UnlockedGetNodeInfo(snode)
890
      if pnode_info is None or snode_info is None:
891
        raise errors.ConfigurationError("Can't find primary or secondary node"
892
                                        " for %s" % str(disk))
893
      p_data = (pnode_info.secondary_ip, port)
894
      s_data = (snode_info.secondary_ip, port)
895
      if pnode == node_name:
896
        disk.physical_id = p_data + s_data + (pminor, secret)
897
      else: # it must be secondary, we tested above
898
        disk.physical_id = s_data + p_data + (sminor, secret)
899
    else:
900
      disk.physical_id = disk.logical_id
901
    return
902

    
903
  @locking.ssynchronized(_config_lock)
904
  def SetDiskID(self, disk, node_name):
905
    """Convert the unique ID to the ID needed on the target nodes.
906

907
    This is used only for drbd, which needs ip/port configuration.
908

909
    The routine descends down and updates its children also, because
910
    this helps when the only the top device is passed to the remote
911
    node.
912

913
    """
914
    return self._UnlockedSetDiskID(disk, node_name)
915

    
916
  @locking.ssynchronized(_config_lock)
917
  def AddTcpUdpPort(self, port):
918
    """Adds a new port to the available port pool.
919

920
    @warning: this method does not "flush" the configuration (via
921
        L{_WriteConfig}); callers should do that themselves once the
922
        configuration is stable
923

924
    """
925
    if not isinstance(port, int):
926
      raise errors.ProgrammerError("Invalid type passed for port")
927

    
928
    self._config_data.cluster.tcpudp_port_pool.add(port)
929

    
930
  @locking.ssynchronized(_config_lock, shared=1)
931
  def GetPortList(self):
932
    """Returns a copy of the current port list.
933

934
    """
935
    return self._config_data.cluster.tcpudp_port_pool.copy()
936

    
937
  @locking.ssynchronized(_config_lock)
938
  def AllocatePort(self):
939
    """Allocate a port.
940

941
    The port will be taken from the available port pool or from the
942
    default port range (and in this case we increase
943
    highest_used_port).
944

945
    """
946
    # If there are TCP/IP ports configured, we use them first.
947
    if self._config_data.cluster.tcpudp_port_pool:
948
      port = self._config_data.cluster.tcpudp_port_pool.pop()
949
    else:
950
      port = self._config_data.cluster.highest_used_port + 1
951
      if port >= constants.LAST_DRBD_PORT:
952
        raise errors.ConfigurationError("The highest used port is greater"
953
                                        " than %s. Aborting." %
954
                                        constants.LAST_DRBD_PORT)
955
      self._config_data.cluster.highest_used_port = port
956

    
957
    self._WriteConfig()
958
    return port
959

    
960
  def _UnlockedComputeDRBDMap(self):
961
    """Compute the used DRBD minor/nodes.
962

963
    @rtype: (dict, list)
964
    @return: dictionary of node_name: dict of minor: instance_name;
965
        the returned dict will have all the nodes in it (even if with
966
        an empty list), and a list of duplicates; if the duplicates
967
        list is not empty, the configuration is corrupted and its caller
968
        should raise an exception
969

970
    """
971
    def _AppendUsedPorts(instance_name, disk, used):
972
      duplicates = []
973
      if disk.dev_type == constants.LD_DRBD8 and len(disk.logical_id) >= 5:
974
        node_a, node_b, _, minor_a, minor_b = disk.logical_id[:5]
975
        for node, port in ((node_a, minor_a), (node_b, minor_b)):
976
          assert node in used, ("Node '%s' of instance '%s' not found"
977
                                " in node list" % (node, instance_name))
978
          if port in used[node]:
979
            duplicates.append((node, port, instance_name, used[node][port]))
980
          else:
981
            used[node][port] = instance_name
982
      if disk.children:
983
        for child in disk.children:
984
          duplicates.extend(_AppendUsedPorts(instance_name, child, used))
985
      return duplicates
986

    
987
    duplicates = []
988
    my_dict = dict((node, {}) for node in self._config_data.nodes)
989
    for instance in self._config_data.instances.itervalues():
990
      for disk in instance.disks:
991
        duplicates.extend(_AppendUsedPorts(instance.name, disk, my_dict))
992
    for (node, minor), instance in self._temporary_drbds.iteritems():
993
      if minor in my_dict[node] and my_dict[node][minor] != instance:
994
        duplicates.append((node, minor, instance, my_dict[node][minor]))
995
      else:
996
        my_dict[node][minor] = instance
997
    return my_dict, duplicates
998

    
999
  @locking.ssynchronized(_config_lock)
1000
  def ComputeDRBDMap(self):
1001
    """Compute the used DRBD minor/nodes.
1002

1003
    This is just a wrapper over L{_UnlockedComputeDRBDMap}.
1004

1005
    @return: dictionary of node_name: dict of minor: instance_name;
1006
        the returned dict will have all the nodes in it (even if with
1007
        an empty list).
1008

1009
    """
1010
    d_map, duplicates = self._UnlockedComputeDRBDMap()
1011
    if duplicates:
1012
      raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
1013
                                      str(duplicates))
1014
    return d_map
1015

    
1016
  @locking.ssynchronized(_config_lock)
1017
  def AllocateDRBDMinor(self, nodes, instance):
1018
    """Allocate a drbd minor.
1019

1020
    The free minor will be automatically computed from the existing
1021
    devices. A node can be given multiple times in order to allocate
1022
    multiple minors. The result is the list of minors, in the same
1023
    order as the passed nodes.
1024

1025
    @type instance: string
1026
    @param instance: the instance for which we allocate minors
1027

1028
    """
1029
    assert isinstance(instance, basestring), \
1030
           "Invalid argument '%s' passed to AllocateDRBDMinor" % instance
1031

    
1032
    d_map, duplicates = self._UnlockedComputeDRBDMap()
1033
    if duplicates:
1034
      raise errors.ConfigurationError("Duplicate DRBD ports detected: %s" %
1035
                                      str(duplicates))
1036
    result = []
1037
    for nname in nodes:
1038
      ndata = d_map[nname]
1039
      if not ndata:
1040
        # no minors used, we can start at 0
1041
        result.append(0)
1042
        ndata[0] = instance
1043
        self._temporary_drbds[(nname, 0)] = instance
1044
        continue
1045
      keys = ndata.keys()
1046
      keys.sort()
1047
      ffree = utils.FirstFree(keys)
1048
      if ffree is None:
1049
        # return the next minor
1050
        # TODO: implement high-limit check
1051
        minor = keys[-1] + 1
1052
      else:
1053
        minor = ffree
1054
      # double-check minor against current instances
1055
      assert minor not in d_map[nname], \
1056
             ("Attempt to reuse allocated DRBD minor %d on node %s,"
1057
              " already allocated to instance %s" %
1058
              (minor, nname, d_map[nname][minor]))
1059
      ndata[minor] = instance
1060
      # double-check minor against reservation
1061
      r_key = (nname, minor)
1062
      assert r_key not in self._temporary_drbds, \
1063
             ("Attempt to reuse reserved DRBD minor %d on node %s,"
1064
              " reserved for instance %s" %
1065
              (minor, nname, self._temporary_drbds[r_key]))
1066
      self._temporary_drbds[r_key] = instance
1067
      result.append(minor)
1068
    logging.debug("Request to allocate drbd minors, input: %s, returning %s",
1069
                  nodes, result)
1070
    return result
1071

    
1072
  def _UnlockedReleaseDRBDMinors(self, instance):
1073
    """Release temporary drbd minors allocated for a given instance.
1074

1075
    @type instance: string
1076
    @param instance: the instance for which temporary minors should be
1077
                     released
1078

1079
    """
1080
    assert isinstance(instance, basestring), \
1081
           "Invalid argument passed to ReleaseDRBDMinors"
1082
    for key, name in self._temporary_drbds.items():
1083
      if name == instance:
1084
        del self._temporary_drbds[key]
1085

    
1086
  @locking.ssynchronized(_config_lock)
1087
  def ReleaseDRBDMinors(self, instance):
1088
    """Release temporary drbd minors allocated for a given instance.
1089

1090
    This should be called on the error paths, on the success paths
1091
    it's automatically called by the ConfigWriter add and update
1092
    functions.
1093

1094
    This function is just a wrapper over L{_UnlockedReleaseDRBDMinors}.
1095

1096
    @type instance: string
1097
    @param instance: the instance for which temporary minors should be
1098
                     released
1099

1100
    """
1101
    self._UnlockedReleaseDRBDMinors(instance)
1102

    
1103
  @locking.ssynchronized(_config_lock, shared=1)
1104
  def GetConfigVersion(self):
1105
    """Get the configuration version.
1106

1107
    @return: Config version
1108

1109
    """
1110
    return self._config_data.version
1111

    
1112
  @locking.ssynchronized(_config_lock, shared=1)
1113
  def GetClusterName(self):
1114
    """Get cluster name.
1115

1116
    @return: Cluster name
1117

1118
    """
1119
    return self._config_data.cluster.cluster_name
1120

    
1121
  @locking.ssynchronized(_config_lock, shared=1)
1122
  def GetMasterNode(self):
1123
    """Get the hostname of the master node for this cluster.
1124

1125
    @return: Master hostname
1126

1127
    """
1128
    return self._config_data.cluster.master_node
1129

    
1130
  @locking.ssynchronized(_config_lock, shared=1)
1131
  def GetMasterIP(self):
1132
    """Get the IP of the master node for this cluster.
1133

1134
    @return: Master IP
1135

1136
    """
1137
    return self._config_data.cluster.master_ip
1138

    
1139
  @locking.ssynchronized(_config_lock, shared=1)
1140
  def GetMasterNetdev(self):
1141
    """Get the master network device for this cluster.
1142

1143
    """
1144
    return self._config_data.cluster.master_netdev
1145

    
1146
  @locking.ssynchronized(_config_lock, shared=1)
1147
  def GetMasterNetmask(self):
1148
    """Get the netmask of the master node for this cluster.
1149

1150
    """
1151
    return self._config_data.cluster.master_netmask
1152

    
1153
  @locking.ssynchronized(_config_lock, shared=1)
1154
  def GetUseExternalMipScript(self):
1155
    """Get flag representing whether to use the external master IP setup script.
1156

1157
    """
1158
    return self._config_data.cluster.use_external_mip_script
1159

    
1160
  @locking.ssynchronized(_config_lock, shared=1)
1161
  def GetFileStorageDir(self):
1162
    """Get the file storage dir for this cluster.
1163

1164
    """
1165
    return self._config_data.cluster.file_storage_dir
1166

    
1167
  @locking.ssynchronized(_config_lock, shared=1)
1168
  def GetSharedFileStorageDir(self):
1169
    """Get the shared file storage dir for this cluster.
1170

1171
    """
1172
    return self._config_data.cluster.shared_file_storage_dir
1173

    
1174
  @locking.ssynchronized(_config_lock, shared=1)
1175
  def GetHypervisorType(self):
1176
    """Get the hypervisor type for this cluster.
1177

1178
    """
1179
    return self._config_data.cluster.enabled_hypervisors[0]
1180

    
1181
  @locking.ssynchronized(_config_lock, shared=1)
1182
  def GetHostKey(self):
1183
    """Return the rsa hostkey from the config.
1184

1185
    @rtype: string
1186
    @return: the rsa hostkey
1187

1188
    """
1189
    return self._config_data.cluster.rsahostkeypub
1190

    
1191
  @locking.ssynchronized(_config_lock, shared=1)
1192
  def GetDefaultIAllocator(self):
1193
    """Get the default instance allocator for this cluster.
1194

1195
    """
1196
    return self._config_data.cluster.default_iallocator
1197

    
1198
  @locking.ssynchronized(_config_lock, shared=1)
1199
  def GetPrimaryIPFamily(self):
1200
    """Get cluster primary ip family.
1201

1202
    @return: primary ip family
1203

1204
    """
1205
    return self._config_data.cluster.primary_ip_family
1206

    
1207
  @locking.ssynchronized(_config_lock, shared=1)
1208
  def GetMasterNetworkParameters(self):
1209
    """Get network parameters of the master node.
1210

1211
    @rtype: L{object.MasterNetworkParameters}
1212
    @return: network parameters of the master node
1213

1214
    """
1215
    cluster = self._config_data.cluster
1216
    result = objects.MasterNetworkParameters(
1217
      name=cluster.master_node, ip=cluster.master_ip,
1218
      netmask=cluster.master_netmask, netdev=cluster.master_netdev,
1219
      ip_family=cluster.primary_ip_family)
1220

    
1221
    return result
1222

    
1223
  @locking.ssynchronized(_config_lock)
1224
  def AddNodeGroup(self, group, ec_id, check_uuid=True):
1225
    """Add a node group to the configuration.
1226

1227
    This method calls group.UpgradeConfig() to fill any missing attributes
1228
    according to their default values.
1229

1230
    @type group: L{objects.NodeGroup}
1231
    @param group: the NodeGroup object to add
1232
    @type ec_id: string
1233
    @param ec_id: unique id for the job to use when creating a missing UUID
1234
    @type check_uuid: bool
1235
    @param check_uuid: add an UUID to the group if it doesn't have one or, if
1236
                       it does, ensure that it does not exist in the
1237
                       configuration already
1238

1239
    """
1240
    self._UnlockedAddNodeGroup(group, ec_id, check_uuid)
1241
    self._WriteConfig()
1242

    
1243
  def _UnlockedAddNodeGroup(self, group, ec_id, check_uuid):
1244
    """Add a node group to the configuration.
1245

1246
    """
1247
    logging.info("Adding node group %s to configuration", group.name)
1248

    
1249
    # Some code might need to add a node group with a pre-populated UUID
1250
    # generated with ConfigWriter.GenerateUniqueID(). We allow them to bypass
1251
    # the "does this UUID" exist already check.
1252
    if check_uuid:
1253
      self._EnsureUUID(group, ec_id)
1254

    
1255
    try:
1256
      existing_uuid = self._UnlockedLookupNodeGroup(group.name)
1257
    except errors.OpPrereqError:
1258
      pass
1259
    else:
1260
      raise errors.OpPrereqError("Desired group name '%s' already exists as a"
1261
                                 " node group (UUID: %s)" %
1262
                                 (group.name, existing_uuid),
1263
                                 errors.ECODE_EXISTS)
1264

    
1265
    group.serial_no = 1
1266
    group.ctime = group.mtime = time.time()
1267
    group.UpgradeConfig()
1268

    
1269
    self._config_data.nodegroups[group.uuid] = group
1270
    self._config_data.cluster.serial_no += 1
1271

    
1272
  @locking.ssynchronized(_config_lock)
1273
  def RemoveNodeGroup(self, group_uuid):
1274
    """Remove a node group from the configuration.
1275

1276
    @type group_uuid: string
1277
    @param group_uuid: the UUID of the node group to remove
1278

1279
    """
1280
    logging.info("Removing node group %s from configuration", group_uuid)
1281

    
1282
    if group_uuid not in self._config_data.nodegroups:
1283
      raise errors.ConfigurationError("Unknown node group '%s'" % group_uuid)
1284

    
1285
    assert len(self._config_data.nodegroups) != 1, \
1286
            "Group '%s' is the only group, cannot be removed" % group_uuid
1287

    
1288
    del self._config_data.nodegroups[group_uuid]
1289
    self._config_data.cluster.serial_no += 1
1290
    self._WriteConfig()
1291

    
1292
  def _UnlockedLookupNodeGroup(self, target):
1293
    """Lookup a node group's UUID.
1294

1295
    @type target: string or None
1296
    @param target: group name or UUID or None to look for the default
1297
    @rtype: string
1298
    @return: nodegroup UUID
1299
    @raises errors.OpPrereqError: when the target group cannot be found
1300

1301
    """
1302
    if target is None:
1303
      if len(self._config_data.nodegroups) != 1:
1304
        raise errors.OpPrereqError("More than one node group exists. Target"
1305
                                   " group must be specified explicitly.")
1306
      else:
1307
        return self._config_data.nodegroups.keys()[0]
1308
    if target in self._config_data.nodegroups:
1309
      return target
1310
    for nodegroup in self._config_data.nodegroups.values():
1311
      if nodegroup.name == target:
1312
        return nodegroup.uuid
1313
    raise errors.OpPrereqError("Node group '%s' not found" % target,
1314
                               errors.ECODE_NOENT)
1315

    
1316
  @locking.ssynchronized(_config_lock, shared=1)
1317
  def LookupNodeGroup(self, target):
1318
    """Lookup a node group's UUID.
1319

1320
    This function is just a wrapper over L{_UnlockedLookupNodeGroup}.
1321

1322
    @type target: string or None
1323
    @param target: group name or UUID or None to look for the default
1324
    @rtype: string
1325
    @return: nodegroup UUID
1326

1327
    """
1328
    return self._UnlockedLookupNodeGroup(target)
1329

    
1330
  def _UnlockedGetNodeGroup(self, uuid):
1331
    """Lookup a node group.
1332

1333
    @type uuid: string
1334
    @param uuid: group UUID
1335
    @rtype: L{objects.NodeGroup} or None
1336
    @return: nodegroup object, or None if not found
1337

1338
    """
1339
    if uuid not in self._config_data.nodegroups:
1340
      return None
1341

    
1342
    return self._config_data.nodegroups[uuid]
1343

    
1344
  @locking.ssynchronized(_config_lock, shared=1)
1345
  def GetNodeGroup(self, uuid):
1346
    """Lookup a node group.
1347

1348
    @type uuid: string
1349
    @param uuid: group UUID
1350
    @rtype: L{objects.NodeGroup} or None
1351
    @return: nodegroup object, or None if not found
1352

1353
    """
1354
    return self._UnlockedGetNodeGroup(uuid)
1355

    
1356
  @locking.ssynchronized(_config_lock, shared=1)
1357
  def GetAllNodeGroupsInfo(self):
1358
    """Get the configuration of all node groups.
1359

1360
    """
1361
    return dict(self._config_data.nodegroups)
1362

    
1363
  @locking.ssynchronized(_config_lock, shared=1)
1364
  def GetNodeGroupList(self):
1365
    """Get a list of node groups.
1366

1367
    """
1368
    return self._config_data.nodegroups.keys()
1369

    
1370
  @locking.ssynchronized(_config_lock, shared=1)
1371
  def GetNodeGroupMembersByNodes(self, nodes):
1372
    """Get nodes which are member in the same nodegroups as the given nodes.
1373

1374
    """
1375
    ngfn = lambda node_name: self._UnlockedGetNodeInfo(node_name).group
1376
    return frozenset(member_name
1377
                     for node_name in nodes
1378
                     for member_name in
1379
                       self._UnlockedGetNodeGroup(ngfn(node_name)).members)
1380

    
1381
  @locking.ssynchronized(_config_lock, shared=1)
1382
  def GetMultiNodeGroupInfo(self, group_uuids):
1383
    """Get the configuration of multiple node groups.
1384

1385
    @param group_uuids: List of node group UUIDs
1386
    @rtype: list
1387
    @return: List of tuples of (group_uuid, group_info)
1388

1389
    """
1390
    return [(uuid, self._UnlockedGetNodeGroup(uuid)) for uuid in group_uuids]
1391

    
1392
  @locking.ssynchronized(_config_lock)
1393
  def AddInstance(self, instance, ec_id):
1394
    """Add an instance to the config.
1395

1396
    This should be used after creating a new instance.
1397

1398
    @type instance: L{objects.Instance}
1399
    @param instance: the instance object
1400

1401
    """
1402
    if not isinstance(instance, objects.Instance):
1403
      raise errors.ProgrammerError("Invalid type passed to AddInstance")
1404

    
1405
    if instance.disk_template != constants.DT_DISKLESS:
1406
      all_lvs = instance.MapLVsByNode()
1407
      logging.info("Instance '%s' DISK_LAYOUT: %s", instance.name, all_lvs)
1408

    
1409
    all_macs = self._AllMACs()
1410
    for nic in instance.nics:
1411
      if nic.mac in all_macs:
1412
        raise errors.ConfigurationError("Cannot add instance %s:"
1413
                                        " MAC address '%s' already in use." %
1414
                                        (instance.name, nic.mac))
1415

    
1416
    self._EnsureUUID(instance, ec_id)
1417

    
1418
    instance.serial_no = 1
1419
    instance.ctime = instance.mtime = time.time()
1420
    self._config_data.instances[instance.name] = instance
1421
    self._config_data.cluster.serial_no += 1
1422
    self._UnlockedReleaseDRBDMinors(instance.name)
1423
    self._UnlockedCommitTemporaryIps(ec_id)
1424
    self._WriteConfig()
1425

    
1426
  def _EnsureUUID(self, item, ec_id):
1427
    """Ensures a given object has a valid UUID.
1428

1429
    @param item: the instance or node to be checked
1430
    @param ec_id: the execution context id for the uuid reservation
1431

1432
    """
1433
    if not item.uuid:
1434
      item.uuid = self._GenerateUniqueID(ec_id)
1435
    elif item.uuid in self._AllIDs(include_temporary=True):
1436
      raise errors.ConfigurationError("Cannot add '%s': UUID %s already"
1437
                                      " in use" % (item.name, item.uuid))
1438

    
1439
  def _SetInstanceStatus(self, instance_name, status, disks_active):
1440
    """Set the instance's status to a given value.
1441

1442
    """
1443
    if instance_name not in self._config_data.instances:
1444
      raise errors.ConfigurationError("Unknown instance '%s'" %
1445
                                      instance_name)
1446
    instance = self._config_data.instances[instance_name]
1447

    
1448
    if status is None:
1449
      status = instance.admin_state
1450
    if disks_active is None:
1451
      disks_active = instance.disks_active
1452

    
1453
    assert status in constants.ADMINST_ALL, \
1454
           "Invalid status '%s' passed to SetInstanceStatus" % (status,)
1455

    
1456
    if instance.admin_state != status or \
1457
       instance.disks_active != disks_active:
1458
      instance.admin_state = status
1459
      instance.disks_active = disks_active
1460
      instance.serial_no += 1
1461
      instance.mtime = time.time()
1462
      self._WriteConfig()
1463

    
1464
  @locking.ssynchronized(_config_lock)
1465
  def MarkInstanceUp(self, instance_name):
1466
    """Mark the instance status to up in the config.
1467

1468
    This also sets the instance disks active flag.
1469

1470
    """
1471
    self._SetInstanceStatus(instance_name, constants.ADMINST_UP, True)
1472

    
1473
  @locking.ssynchronized(_config_lock)
1474
  def MarkInstanceOffline(self, instance_name):
1475
    """Mark the instance status to down in the config.
1476

1477
    This also clears the instance disks active flag.
1478

1479
    """
1480
    self._SetInstanceStatus(instance_name, constants.ADMINST_OFFLINE, False)
1481

    
1482
  @locking.ssynchronized(_config_lock)
1483
  def RemoveInstance(self, instance_name):
1484
    """Remove the instance from the configuration.
1485

1486
    """
1487
    if instance_name not in self._config_data.instances:
1488
      raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
1489

    
1490
    # If a network port has been allocated to the instance,
1491
    # return it to the pool of free ports.
1492
    inst = self._config_data.instances[instance_name]
1493
    network_port = getattr(inst, "network_port", None)
1494
    if network_port is not None:
1495
      self._config_data.cluster.tcpudp_port_pool.add(network_port)
1496

    
1497
    instance = self._UnlockedGetInstanceInfo(instance_name)
1498

    
1499
    for nic in instance.nics:
1500
      if nic.network and nic.ip:
1501
        # Return all IP addresses to the respective address pools
1502
        self._UnlockedCommitIp(constants.RELEASE_ACTION, nic.network, nic.ip)
1503

    
1504
    del self._config_data.instances[instance_name]
1505
    self._config_data.cluster.serial_no += 1
1506
    self._WriteConfig()
1507

    
1508
  @locking.ssynchronized(_config_lock)
1509
  def RenameInstance(self, old_name, new_name):
1510
    """Rename an instance.
1511

1512
    This needs to be done in ConfigWriter and not by RemoveInstance
1513
    combined with AddInstance as only we can guarantee an atomic
1514
    rename.
1515

1516
    """
1517
    if old_name not in self._config_data.instances:
1518
      raise errors.ConfigurationError("Unknown instance '%s'" % old_name)
1519

    
1520
    # Operate on a copy to not loose instance object in case of a failure
1521
    inst = self._config_data.instances[old_name].Copy()
1522
    inst.name = new_name
1523

    
1524
    for (idx, disk) in enumerate(inst.disks):
1525
      if disk.dev_type == constants.LD_FILE:
1526
        # rename the file paths in logical and physical id
1527
        file_storage_dir = os.path.dirname(os.path.dirname(disk.logical_id[1]))
1528
        disk.logical_id = (disk.logical_id[0],
1529
                           utils.PathJoin(file_storage_dir, inst.name,
1530
                                          "disk%s" % idx))
1531
        disk.physical_id = disk.logical_id
1532

    
1533
    # Actually replace instance object
1534
    del self._config_data.instances[old_name]
1535
    self._config_data.instances[inst.name] = inst
1536

    
1537
    # Force update of ssconf files
1538
    self._config_data.cluster.serial_no += 1
1539

    
1540
    self._WriteConfig()
1541

    
1542
  @locking.ssynchronized(_config_lock)
1543
  def MarkInstanceDown(self, instance_name):
1544
    """Mark the status of an instance to down in the configuration.
1545

1546
    This does not touch the instance disks active flag, as shut down instances
1547
    can still have active disks.
1548

1549
    """
1550
    self._SetInstanceStatus(instance_name, constants.ADMINST_DOWN, None)
1551

    
1552
  @locking.ssynchronized(_config_lock)
1553
  def MarkInstanceDisksActive(self, instance_name):
1554
    """Mark the status of instance disks active.
1555

1556
    """
1557
    self._SetInstanceStatus(instance_name, None, True)
1558

    
1559
  @locking.ssynchronized(_config_lock)
1560
  def MarkInstanceDisksInactive(self, instance_name):
1561
    """Mark the status of instance disks inactive.
1562

1563
    """
1564
    self._SetInstanceStatus(instance_name, None, False)
1565

    
1566
  def _UnlockedGetInstanceList(self):
1567
    """Get the list of instances.
1568

1569
    This function is for internal use, when the config lock is already held.
1570

1571
    """
1572
    return self._config_data.instances.keys()
1573

    
1574
  @locking.ssynchronized(_config_lock, shared=1)
1575
  def GetInstanceList(self):
1576
    """Get the list of instances.
1577

1578
    @return: array of instances, ex. ['instance2.example.com',
1579
        'instance1.example.com']
1580

1581
    """
1582
    return self._UnlockedGetInstanceList()
1583

    
1584
  def ExpandInstanceName(self, short_name):
1585
    """Attempt to expand an incomplete instance name.
1586

1587
    """
1588
    # Locking is done in L{ConfigWriter.GetInstanceList}
1589
    return _MatchNameComponentIgnoreCase(short_name, self.GetInstanceList())
1590

    
1591
  def _UnlockedGetInstanceInfo(self, instance_name):
1592
    """Returns information about an instance.
1593

1594
    This function is for internal use, when the config lock is already held.
1595

1596
    """
1597
    if instance_name not in self._config_data.instances:
1598
      return None
1599

    
1600
    return self._config_data.instances[instance_name]
1601

    
1602
  @locking.ssynchronized(_config_lock, shared=1)
1603
  def GetInstanceInfo(self, instance_name):
1604
    """Returns information about an instance.
1605

1606
    It takes the information from the configuration file. Other information of
1607
    an instance are taken from the live systems.
1608

1609
    @param instance_name: name of the instance, e.g.
1610
        I{instance1.example.com}
1611

1612
    @rtype: L{objects.Instance}
1613
    @return: the instance object
1614

1615
    """
1616
    return self._UnlockedGetInstanceInfo(instance_name)
1617

    
1618
  @locking.ssynchronized(_config_lock, shared=1)
1619
  def GetInstanceNodeGroups(self, instance_name, primary_only=False):
1620
    """Returns set of node group UUIDs for instance's nodes.
1621

1622
    @rtype: frozenset
1623

1624
    """
1625
    instance = self._UnlockedGetInstanceInfo(instance_name)
1626
    if not instance:
1627
      raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
1628

    
1629
    if primary_only:
1630
      nodes = [instance.primary_node]
1631
    else:
1632
      nodes = instance.all_nodes
1633

    
1634
    return frozenset(self._UnlockedGetNodeInfo(node_name).group
1635
                     for node_name in nodes)
1636

    
1637
  @locking.ssynchronized(_config_lock, shared=1)
1638
  def GetInstanceNetworks(self, instance_name):
1639
    """Returns set of network UUIDs for instance's nics.
1640

1641
    @rtype: frozenset
1642

1643
    """
1644
    instance = self._UnlockedGetInstanceInfo(instance_name)
1645
    if not instance:
1646
      raise errors.ConfigurationError("Unknown instance '%s'" % instance_name)
1647

    
1648
    networks = set()
1649
    for nic in instance.nics:
1650
      if nic.network:
1651
        networks.add(nic.network)
1652

    
1653
    return frozenset(networks)
1654

    
1655
  @locking.ssynchronized(_config_lock, shared=1)
1656
  def GetMultiInstanceInfo(self, instances):
1657
    """Get the configuration of multiple instances.
1658

1659
    @param instances: list of instance names
1660
    @rtype: list
1661
    @return: list of tuples (instance, instance_info), where
1662
        instance_info is what would GetInstanceInfo return for the
1663
        node, while keeping the original order
1664

1665
    """
1666
    return [(name, self._UnlockedGetInstanceInfo(name)) for name in instances]
1667

    
1668
  @locking.ssynchronized(_config_lock, shared=1)
1669
  def GetAllInstancesInfo(self):
1670
    """Get the configuration of all instances.
1671

1672
    @rtype: dict
1673
    @return: dict of (instance, instance_info), where instance_info is what
1674
              would GetInstanceInfo return for the node
1675

1676
    """
1677
    my_dict = dict([(instance, self._UnlockedGetInstanceInfo(instance))
1678
                    for instance in self._UnlockedGetInstanceList()])
1679
    return my_dict
1680

    
1681
  @locking.ssynchronized(_config_lock, shared=1)
1682
  def GetInstancesInfoByFilter(self, filter_fn):
1683
    """Get instance configuration with a filter.
1684

1685
    @type filter_fn: callable
1686
    @param filter_fn: Filter function receiving instance object as parameter,
1687
      returning boolean. Important: this function is called while the
1688
      configuration locks is held. It must not do any complex work or call
1689
      functions potentially leading to a deadlock. Ideally it doesn't call any
1690
      other functions and just compares instance attributes.
1691

1692
    """
1693
    return dict((name, inst)
1694
                for (name, inst) in self._config_data.instances.items()
1695
                if filter_fn(inst))
1696

    
1697
  @locking.ssynchronized(_config_lock)
1698
  def AddNode(self, node, ec_id):
1699
    """Add a node to the configuration.
1700

1701
    @type node: L{objects.Node}
1702
    @param node: a Node instance
1703

1704
    """
1705
    logging.info("Adding node %s to configuration", node.name)
1706

    
1707
    self._EnsureUUID(node, ec_id)
1708

    
1709
    node.serial_no = 1
1710
    node.ctime = node.mtime = time.time()
1711
    self._UnlockedAddNodeToGroup(node.name, node.group)
1712
    self._config_data.nodes[node.name] = node
1713
    self._config_data.cluster.serial_no += 1
1714
    self._WriteConfig()
1715

    
1716
  @locking.ssynchronized(_config_lock)
1717
  def RemoveNode(self, node_name):
1718
    """Remove a node from the configuration.
1719

1720
    """
1721
    logging.info("Removing node %s from configuration", node_name)
1722

    
1723
    if node_name not in self._config_data.nodes:
1724
      raise errors.ConfigurationError("Unknown node '%s'" % node_name)
1725

    
1726
    self._UnlockedRemoveNodeFromGroup(self._config_data.nodes[node_name])
1727
    del self._config_data.nodes[node_name]
1728
    self._config_data.cluster.serial_no += 1
1729
    self._WriteConfig()
1730

    
1731
  def ExpandNodeName(self, short_name):
1732
    """Attempt to expand an incomplete node name.
1733

1734
    """
1735
    # Locking is done in L{ConfigWriter.GetNodeList}
1736
    return _MatchNameComponentIgnoreCase(short_name, self.GetNodeList())
1737

    
1738
  def _UnlockedGetNodeInfo(self, node_name):
1739
    """Get the configuration of a node, as stored in the config.
1740

1741
    This function is for internal use, when the config lock is already
1742
    held.
1743

1744
    @param node_name: the node name, e.g. I{node1.example.com}
1745

1746
    @rtype: L{objects.Node}
1747
    @return: the node object
1748

1749
    """
1750
    if node_name not in self._config_data.nodes:
1751
      return None
1752

    
1753
    return self._config_data.nodes[node_name]
1754

    
1755
  @locking.ssynchronized(_config_lock, shared=1)
1756
  def GetNodeInfo(self, node_name):
1757
    """Get the configuration of a node, as stored in the config.
1758

1759
    This is just a locked wrapper over L{_UnlockedGetNodeInfo}.
1760

1761
    @param node_name: the node name, e.g. I{node1.example.com}
1762

1763
    @rtype: L{objects.Node}
1764
    @return: the node object
1765

1766
    """
1767
    return self._UnlockedGetNodeInfo(node_name)
1768

    
1769
  @locking.ssynchronized(_config_lock, shared=1)
1770
  def GetNodeInstances(self, node_name):
1771
    """Get the instances of a node, as stored in the config.
1772

1773
    @param node_name: the node name, e.g. I{node1.example.com}
1774

1775
    @rtype: (list, list)
1776
    @return: a tuple with two lists: the primary and the secondary instances
1777

1778
    """
1779
    pri = []
1780
    sec = []
1781
    for inst in self._config_data.instances.values():
1782
      if inst.primary_node == node_name:
1783
        pri.append(inst.name)
1784
      if node_name in inst.secondary_nodes:
1785
        sec.append(inst.name)
1786
    return (pri, sec)
1787

    
1788
  @locking.ssynchronized(_config_lock, shared=1)
1789
  def GetNodeGroupInstances(self, uuid, primary_only=False):
1790
    """Get the instances of a node group.
1791

1792
    @param uuid: Node group UUID
1793
    @param primary_only: Whether to only consider primary nodes
1794
    @rtype: frozenset
1795
    @return: List of instance names in node group
1796

1797
    """
1798
    if primary_only:
1799
      nodes_fn = lambda inst: [inst.primary_node]
1800
    else:
1801
      nodes_fn = lambda inst: inst.all_nodes
1802

    
1803
    return frozenset(inst.name
1804
                     for inst in self._config_data.instances.values()
1805
                     for node_name in nodes_fn(inst)
1806
                     if self._UnlockedGetNodeInfo(node_name).group == uuid)
1807

    
1808
  def _UnlockedGetNodeList(self):
1809
    """Return the list of nodes which are in the configuration.
1810

1811
    This function is for internal use, when the config lock is already
1812
    held.
1813

1814
    @rtype: list
1815

1816
    """
1817
    return self._config_data.nodes.keys()
1818

    
1819
  @locking.ssynchronized(_config_lock, shared=1)
1820
  def GetNodeList(self):
1821
    """Return the list of nodes which are in the configuration.
1822

1823
    """
1824
    return self._UnlockedGetNodeList()
1825

    
1826
  def _UnlockedGetOnlineNodeList(self):
1827
    """Return the list of nodes which are online.
1828

1829
    """
1830
    all_nodes = [self._UnlockedGetNodeInfo(node)
1831
                 for node in self._UnlockedGetNodeList()]
1832
    return [node.name for node in all_nodes if not node.offline]
1833

    
1834
  @locking.ssynchronized(_config_lock, shared=1)
1835
  def GetOnlineNodeList(self):
1836
    """Return the list of nodes which are online.
1837

1838
    """
1839
    return self._UnlockedGetOnlineNodeList()
1840

    
1841
  @locking.ssynchronized(_config_lock, shared=1)
1842
  def GetVmCapableNodeList(self):
1843
    """Return the list of nodes which are not vm capable.
1844

1845
    """
1846
    all_nodes = [self._UnlockedGetNodeInfo(node)
1847
                 for node in self._UnlockedGetNodeList()]
1848
    return [node.name for node in all_nodes if node.vm_capable]
1849

    
1850
  @locking.ssynchronized(_config_lock, shared=1)
1851
  def GetNonVmCapableNodeList(self):
1852
    """Return the list of nodes which are not vm capable.
1853

1854
    """
1855
    all_nodes = [self._UnlockedGetNodeInfo(node)
1856
                 for node in self._UnlockedGetNodeList()]
1857
    return [node.name for node in all_nodes if not node.vm_capable]
1858

    
1859
  @locking.ssynchronized(_config_lock, shared=1)
1860
  def GetMultiNodeInfo(self, nodes):
1861
    """Get the configuration of multiple nodes.
1862

1863
    @param nodes: list of node names
1864
    @rtype: list
1865
    @return: list of tuples of (node, node_info), where node_info is
1866
        what would GetNodeInfo return for the node, in the original
1867
        order
1868

1869
    """
1870
    return [(name, self._UnlockedGetNodeInfo(name)) for name in nodes]
1871

    
1872
  @locking.ssynchronized(_config_lock, shared=1)
1873
  def GetAllNodesInfo(self):
1874
    """Get the configuration of all nodes.
1875

1876
    @rtype: dict
1877
    @return: dict of (node, node_info), where node_info is what
1878
              would GetNodeInfo return for the node
1879

1880
    """
1881
    return self._UnlockedGetAllNodesInfo()
1882

    
1883
  def _UnlockedGetAllNodesInfo(self):
1884
    """Gets configuration of all nodes.
1885

1886
    @note: See L{GetAllNodesInfo}
1887

1888
    """
1889
    return dict([(node, self._UnlockedGetNodeInfo(node))
1890
                 for node in self._UnlockedGetNodeList()])
1891

    
1892
  @locking.ssynchronized(_config_lock, shared=1)
1893
  def GetNodeGroupsFromNodes(self, nodes):
1894
    """Returns groups for a list of nodes.
1895

1896
    @type nodes: list of string
1897
    @param nodes: List of node names
1898
    @rtype: frozenset
1899

1900
    """
1901
    return frozenset(self._UnlockedGetNodeInfo(name).group for name in nodes)
1902

    
1903
  def _UnlockedGetMasterCandidateStats(self, exceptions=None):
1904
    """Get the number of current and maximum desired and possible candidates.
1905

1906
    @type exceptions: list
1907
    @param exceptions: if passed, list of nodes that should be ignored
1908
    @rtype: tuple
1909
    @return: tuple of (current, desired and possible, possible)
1910

1911
    """
1912
    mc_now = mc_should = mc_max = 0
1913
    for node in self._config_data.nodes.values():
1914
      if exceptions and node.name in exceptions:
1915
        continue
1916
      if not (node.offline or node.drained) and node.master_capable:
1917
        mc_max += 1
1918
      if node.master_candidate:
1919
        mc_now += 1
1920
    mc_should = min(mc_max, self._config_data.cluster.candidate_pool_size)
1921
    return (mc_now, mc_should, mc_max)
1922

    
1923
  @locking.ssynchronized(_config_lock, shared=1)
1924
  def GetMasterCandidateStats(self, exceptions=None):
1925
    """Get the number of current and maximum possible candidates.
1926

1927
    This is just a wrapper over L{_UnlockedGetMasterCandidateStats}.
1928

1929
    @type exceptions: list
1930
    @param exceptions: if passed, list of nodes that should be ignored
1931
    @rtype: tuple
1932
    @return: tuple of (current, max)
1933

1934
    """
1935
    return self._UnlockedGetMasterCandidateStats(exceptions)
1936

    
1937
  @locking.ssynchronized(_config_lock)
1938
  def MaintainCandidatePool(self, exceptions):
1939
    """Try to grow the candidate pool to the desired size.
1940

1941
    @type exceptions: list
1942
    @param exceptions: if passed, list of nodes that should be ignored
1943
    @rtype: list
1944
    @return: list with the adjusted nodes (L{objects.Node} instances)
1945

1946
    """
1947
    mc_now, mc_max, _ = self._UnlockedGetMasterCandidateStats(exceptions)
1948
    mod_list = []
1949
    if mc_now < mc_max:
1950
      node_list = self._config_data.nodes.keys()
1951
      random.shuffle(node_list)
1952
      for name in node_list:
1953
        if mc_now >= mc_max:
1954
          break
1955
        node = self._config_data.nodes[name]
1956
        if (node.master_candidate or node.offline or node.drained or
1957
            node.name in exceptions or not node.master_capable):
1958
          continue
1959
        mod_list.append(node)
1960
        node.master_candidate = True
1961
        node.serial_no += 1
1962
        mc_now += 1
1963
      if mc_now != mc_max:
1964
        # this should not happen
1965
        logging.warning("Warning: MaintainCandidatePool didn't manage to"
1966
                        " fill the candidate pool (%d/%d)", mc_now, mc_max)
1967
      if mod_list:
1968
        self._config_data.cluster.serial_no += 1
1969
        self._WriteConfig()
1970

    
1971
    return mod_list
1972

    
1973
  def _UnlockedAddNodeToGroup(self, node_name, nodegroup_uuid):
1974
    """Add a given node to the specified group.
1975

1976
    """
1977
    if nodegroup_uuid not in self._config_data.nodegroups:
1978
      # This can happen if a node group gets deleted between its lookup and
1979
      # when we're adding the first node to it, since we don't keep a lock in
1980
      # the meantime. It's ok though, as we'll fail cleanly if the node group
1981
      # is not found anymore.
1982
      raise errors.OpExecError("Unknown node group: %s" % nodegroup_uuid)
1983
    if node_name not in self._config_data.nodegroups[nodegroup_uuid].members:
1984
      self._config_data.nodegroups[nodegroup_uuid].members.append(node_name)
1985

    
1986
  def _UnlockedRemoveNodeFromGroup(self, node):
1987
    """Remove a given node from its group.
1988

1989
    """
1990
    nodegroup = node.group
1991
    if nodegroup not in self._config_data.nodegroups:
1992
      logging.warning("Warning: node '%s' has unknown node group '%s'"
1993
                      " (while being removed from it)", node.name, nodegroup)
1994
    nodegroup_obj = self._config_data.nodegroups[nodegroup]
1995
    if node.name not in nodegroup_obj.members:
1996
      logging.warning("Warning: node '%s' not a member of its node group '%s'"
1997
                      " (while being removed from it)", node.name, nodegroup)
1998
    else:
1999
      nodegroup_obj.members.remove(node.name)
2000

    
2001
  @locking.ssynchronized(_config_lock)
2002
  def AssignGroupNodes(self, mods):
2003
    """Changes the group of a number of nodes.
2004

2005
    @type mods: list of tuples; (node name, new group UUID)
2006
    @param mods: Node membership modifications
2007

2008
    """
2009
    groups = self._config_data.nodegroups
2010
    nodes = self._config_data.nodes
2011

    
2012
    resmod = []
2013

    
2014
    # Try to resolve names/UUIDs first
2015
    for (node_name, new_group_uuid) in mods:
2016
      try:
2017
        node = nodes[node_name]
2018
      except KeyError:
2019
        raise errors.ConfigurationError("Unable to find node '%s'" % node_name)
2020

    
2021
      if node.group == new_group_uuid:
2022
        # Node is being assigned to its current group
2023
        logging.debug("Node '%s' was assigned to its current group (%s)",
2024
                      node_name, node.group)
2025
        continue
2026

    
2027
      # Try to find current group of node
2028
      try:
2029
        old_group = groups[node.group]
2030
      except KeyError:
2031
        raise errors.ConfigurationError("Unable to find old group '%s'" %
2032
                                        node.group)
2033

    
2034
      # Try to find new group for node
2035
      try:
2036
        new_group = groups[new_group_uuid]
2037
      except KeyError:
2038
        raise errors.ConfigurationError("Unable to find new group '%s'" %
2039
                                        new_group_uuid)
2040

    
2041
      assert node.name in old_group.members, \
2042
        ("Inconsistent configuration: node '%s' not listed in members for its"
2043
         " old group '%s'" % (node.name, old_group.uuid))
2044
      assert node.name not in new_group.members, \
2045
        ("Inconsistent configuration: node '%s' already listed in members for"
2046
         " its new group '%s'" % (node.name, new_group.uuid))
2047

    
2048
      resmod.append((node, old_group, new_group))
2049

    
2050
    # Apply changes
2051
    for (node, old_group, new_group) in resmod:
2052
      assert node.uuid != new_group.uuid and old_group.uuid != new_group.uuid, \
2053
        "Assigning to current group is not possible"
2054

    
2055
      node.group = new_group.uuid
2056

    
2057
      # Update members of involved groups
2058
      if node.name in old_group.members:
2059
        old_group.members.remove(node.name)
2060
      if node.name not in new_group.members:
2061
        new_group.members.append(node.name)
2062

    
2063
    # Update timestamps and serials (only once per node/group object)
2064
    now = time.time()
2065
    for obj in frozenset(itertools.chain(*resmod)): # pylint: disable=W0142
2066
      obj.serial_no += 1
2067
      obj.mtime = now
2068

    
2069
    # Force ssconf update
2070
    self._config_data.cluster.serial_no += 1
2071

    
2072
    self._WriteConfig()
2073

    
2074
  def _BumpSerialNo(self):
2075
    """Bump up the serial number of the config.
2076

2077
    """
2078
    self._config_data.serial_no += 1
2079
    self._config_data.mtime = time.time()
2080

    
2081
  def _AllUUIDObjects(self):
2082
    """Returns all objects with uuid attributes.
2083

2084
    """
2085
    return (self._config_data.instances.values() +
2086
            self._config_data.nodes.values() +
2087
            self._config_data.nodegroups.values() +
2088
            self._config_data.networks.values() +
2089
            self._AllDisks() +
2090
            self._AllNICs() +
2091
            [self._config_data.cluster])
2092

    
2093
  def _OpenConfig(self, accept_foreign):
2094
    """Read the config data from disk.
2095

2096
    """
2097
    raw_data = utils.ReadFile(self._cfg_file)
2098

    
2099
    try:
2100
      data = objects.ConfigData.FromDict(serializer.Load(raw_data))
2101
    except Exception, err:
2102
      raise errors.ConfigurationError(err)
2103

    
2104
    # Make sure the configuration has the right version
2105
    _ValidateConfig(data)
2106

    
2107
    if (not hasattr(data, "cluster") or
2108
        not hasattr(data.cluster, "rsahostkeypub")):
2109
      raise errors.ConfigurationError("Incomplete configuration"
2110
                                      " (missing cluster.rsahostkeypub)")
2111

    
2112
    if data.cluster.master_node != self._my_hostname and not accept_foreign:
2113
      msg = ("The configuration denotes node %s as master, while my"
2114
             " hostname is %s; opening a foreign configuration is only"
2115
             " possible in accept_foreign mode" %
2116
             (data.cluster.master_node, self._my_hostname))
2117
      raise errors.ConfigurationError(msg)
2118

    
2119
    self._config_data = data
2120
    # reset the last serial as -1 so that the next write will cause
2121
    # ssconf update
2122
    self._last_cluster_serial = -1
2123

    
2124
    # Upgrade configuration if needed
2125
    self._UpgradeConfig()
2126

    
2127
    self._cfg_id = utils.GetFileID(path=self._cfg_file)
2128

    
2129
  def _UpgradeConfig(self):
2130
    """Run any upgrade steps.
2131

2132
    This method performs both in-object upgrades and also update some data
2133
    elements that need uniqueness across the whole configuration or interact
2134
    with other objects.
2135

2136
    @warning: this function will call L{_WriteConfig()}, but also
2137
        L{DropECReservations} so it needs to be called only from a
2138
        "safe" place (the constructor). If one wanted to call it with
2139
        the lock held, a DropECReservationUnlocked would need to be
2140
        created first, to avoid causing deadlock.
2141

2142
    """
2143
    # Keep a copy of the persistent part of _config_data to check for changes
2144
    # Serialization doesn't guarantee order in dictionaries
2145
    oldconf = copy.deepcopy(self._config_data.ToDict())
2146

    
2147
    # In-object upgrades
2148
    self._config_data.UpgradeConfig()
2149

    
2150
    for item in self._AllUUIDObjects():
2151
      if item.uuid is None:
2152
        item.uuid = self._GenerateUniqueID(_UPGRADE_CONFIG_JID)
2153
    if not self._config_data.nodegroups:
2154
      default_nodegroup_name = constants.INITIAL_NODE_GROUP_NAME
2155
      default_nodegroup = objects.NodeGroup(name=default_nodegroup_name,
2156
                                            members=[])
2157
      self._UnlockedAddNodeGroup(default_nodegroup, _UPGRADE_CONFIG_JID, True)
2158
    for node in self._config_data.nodes.values():
2159
      if not node.group:
2160
        node.group = self.LookupNodeGroup(None)
2161
      # This is technically *not* an upgrade, but needs to be done both when
2162
      # nodegroups are being added, and upon normally loading the config,
2163
      # because the members list of a node group is discarded upon
2164
      # serializing/deserializing the object.
2165
      self._UnlockedAddNodeToGroup(node.name, node.group)
2166

    
2167
    modified = (oldconf != self._config_data.ToDict())
2168
    if modified:
2169
      self._WriteConfig()
2170
      # This is ok even if it acquires the internal lock, as _UpgradeConfig is
2171
      # only called at config init time, without the lock held
2172
      self.DropECReservations(_UPGRADE_CONFIG_JID)
2173
    else:
2174
      config_errors = self._UnlockedVerifyConfig()
2175
      if config_errors:
2176
        errmsg = ("Loaded configuration data is not consistent: %s" %
2177
                  (utils.CommaJoin(config_errors)))
2178
        logging.critical(errmsg)
2179

    
2180
  def _DistributeConfig(self, feedback_fn):
2181
    """Distribute the configuration to the other nodes.
2182

2183
    Currently, this only copies the configuration file. In the future,
2184
    it could be used to encapsulate the 2/3-phase update mechanism.
2185

2186
    """
2187
    if self._offline:
2188
      return True
2189

    
2190
    bad = False
2191

    
2192
    node_list = []
2193
    addr_list = []
2194
    myhostname = self._my_hostname
2195
    # we can skip checking whether _UnlockedGetNodeInfo returns None
2196
    # since the node list comes from _UnlocketGetNodeList, and we are
2197
    # called with the lock held, so no modifications should take place
2198
    # in between
2199
    for node_name in self._UnlockedGetNodeList():
2200
      if node_name == myhostname:
2201
        continue
2202
      node_info = self._UnlockedGetNodeInfo(node_name)
2203
      if not node_info.master_candidate:
2204
        continue
2205
      node_list.append(node_info.name)
2206
      addr_list.append(node_info.primary_ip)
2207

    
2208
    # TODO: Use dedicated resolver talking to config writer for name resolution
2209
    result = \
2210
      self._GetRpc(addr_list).call_upload_file(node_list, self._cfg_file)
2211
    for to_node, to_result in result.items():
2212
      msg = to_result.fail_msg
2213
      if msg:
2214
        msg = ("Copy of file %s to node %s failed: %s" %
2215
               (self._cfg_file, to_node, msg))
2216
        logging.error(msg)
2217

    
2218
        if feedback_fn:
2219
          feedback_fn(msg)
2220

    
2221
        bad = True
2222

    
2223
    return not bad
2224

    
2225
  def _WriteConfig(self, destination=None, feedback_fn=None):
2226
    """Write the configuration data to persistent storage.
2227

2228
    """
2229
    assert feedback_fn is None or callable(feedback_fn)
2230

    
2231
    # Warn on config errors, but don't abort the save - the
2232
    # configuration has already been modified, and we can't revert;
2233
    # the best we can do is to warn the user and save as is, leaving
2234
    # recovery to the user
2235
    config_errors = self._UnlockedVerifyConfig()
2236
    if config_errors:
2237
      errmsg = ("Configuration data is not consistent: %s" %
2238
                (utils.CommaJoin(config_errors)))
2239
      logging.critical(errmsg)
2240
      if feedback_fn:
2241
        feedback_fn(errmsg)
2242

    
2243
    if destination is None:
2244
      destination = self._cfg_file
2245
    self._BumpSerialNo()
2246
    txt = serializer.Dump(self._config_data.ToDict())
2247

    
2248
    getents = self._getents()
2249
    try:
2250
      fd = utils.SafeWriteFile(destination, self._cfg_id, data=txt,
2251
                               close=False, gid=getents.confd_gid, mode=0640)
2252
    except errors.LockError:
2253
      raise errors.ConfigurationError("The configuration file has been"
2254
                                      " modified since the last write, cannot"
2255
                                      " update")
2256
    try:
2257
      self._cfg_id = utils.GetFileID(fd=fd)
2258
    finally:
2259
      os.close(fd)
2260

    
2261
    self.write_count += 1
2262

    
2263
    # and redistribute the config file to master candidates
2264
    self._DistributeConfig(feedback_fn)
2265

    
2266
    # Write ssconf files on all nodes (including locally)
2267
    if self._last_cluster_serial < self._config_data.cluster.serial_no:
2268
      if not self._offline:
2269
        result = self._GetRpc(None).call_write_ssconf_files(
2270
          self._UnlockedGetOnlineNodeList(),
2271
          self._UnlockedGetSsconfValues())
2272

    
2273
        for nname, nresu in result.items():
2274
          msg = nresu.fail_msg
2275
          if msg:
2276
            errmsg = ("Error while uploading ssconf files to"
2277
                      " node %s: %s" % (nname, msg))
2278
            logging.warning(errmsg)
2279

    
2280
            if feedback_fn:
2281
              feedback_fn(errmsg)
2282

    
2283
      self._last_cluster_serial = self._config_data.cluster.serial_no
2284

    
2285
  def _UnlockedGetSsconfValues(self):
2286
    """Return the values needed by ssconf.
2287

2288
    @rtype: dict
2289
    @return: a dictionary with keys the ssconf names and values their
2290
        associated value
2291

2292
    """
2293
    fn = "\n".join
2294
    instance_names = utils.NiceSort(self._UnlockedGetInstanceList())
2295
    node_names = utils.NiceSort(self._UnlockedGetNodeList())
2296
    node_info = [self._UnlockedGetNodeInfo(name) for name in node_names]
2297
    node_pri_ips = ["%s %s" % (ninfo.name, ninfo.primary_ip)
2298
                    for ninfo in node_info]
2299
    node_snd_ips = ["%s %s" % (ninfo.name, ninfo.secondary_ip)
2300
                    for ninfo in node_info]
2301

    
2302
    instance_data = fn(instance_names)
2303
    off_data = fn(node.name for node in node_info if node.offline)
2304
    on_data = fn(node.name for node in node_info if not node.offline)
2305
    mc_data = fn(node.name for node in node_info if node.master_candidate)
2306
    mc_ips_data = fn(node.primary_ip for node in node_info
2307
                     if node.master_candidate)
2308
    node_data = fn(node_names)
2309
    node_pri_ips_data = fn(node_pri_ips)
2310
    node_snd_ips_data = fn(node_snd_ips)
2311

    
2312
    cluster = self._config_data.cluster
2313
    cluster_tags = fn(cluster.GetTags())
2314

    
2315
    hypervisor_list = fn(cluster.enabled_hypervisors)
2316

    
2317
    uid_pool = uidpool.FormatUidPool(cluster.uid_pool, separator="\n")
2318

    
2319
    nodegroups = ["%s %s" % (nodegroup.uuid, nodegroup.name) for nodegroup in
2320
                  self._config_data.nodegroups.values()]
2321
    nodegroups_data = fn(utils.NiceSort(nodegroups))
2322
    networks = ["%s %s" % (net.uuid, net.name) for net in
2323
                self._config_data.networks.values()]
2324
    networks_data = fn(utils.NiceSort(networks))
2325

    
2326
    ssconf_values = {
2327
      constants.SS_CLUSTER_NAME: cluster.cluster_name,
2328
      constants.SS_CLUSTER_TAGS: cluster_tags,
2329
      constants.SS_FILE_STORAGE_DIR: cluster.file_storage_dir,
2330
      constants.SS_SHARED_FILE_STORAGE_DIR: cluster.shared_file_storage_dir,
2331
      constants.SS_MASTER_CANDIDATES: mc_data,
2332
      constants.SS_MASTER_CANDIDATES_IPS: mc_ips_data,
2333
      constants.SS_MASTER_IP: cluster.master_ip,
2334
      constants.SS_MASTER_NETDEV: cluster.master_netdev,
2335
      constants.SS_MASTER_NETMASK: str(cluster.master_netmask),
2336
      constants.SS_MASTER_NODE: cluster.master_node,
2337
      constants.SS_NODE_LIST: node_data,
2338
      constants.SS_NODE_PRIMARY_IPS: node_pri_ips_data,
2339
      constants.SS_NODE_SECONDARY_IPS: node_snd_ips_data,
2340
      constants.SS_OFFLINE_NODES: off_data,
2341
      constants.SS_ONLINE_NODES: on_data,
2342
      constants.SS_PRIMARY_IP_FAMILY: str(cluster.primary_ip_family),
2343
      constants.SS_INSTANCE_LIST: instance_data,
2344
      constants.SS_RELEASE_VERSION: constants.RELEASE_VERSION,
2345
      constants.SS_HYPERVISOR_LIST: hypervisor_list,
2346
      constants.SS_MAINTAIN_NODE_HEALTH: str(cluster.maintain_node_health),
2347
      constants.SS_UID_POOL: uid_pool,
2348
      constants.SS_NODEGROUPS: nodegroups_data,
2349
      constants.SS_NETWORKS: networks_data,
2350
      }
2351
    bad_values = [(k, v) for k, v in ssconf_values.items()
2352
                  if not isinstance(v, (str, basestring))]
2353
    if bad_values:
2354
      err = utils.CommaJoin("%s=%s" % (k, v) for k, v in bad_values)
2355
      raise errors.ConfigurationError("Some ssconf key(s) have non-string"
2356
                                      " values: %s" % err)
2357
    return ssconf_values
2358

    
2359
  @locking.ssynchronized(_config_lock, shared=1)
2360
  def GetSsconfValues(self):
2361
    """Wrapper using lock around _UnlockedGetSsconf().
2362

2363
    """
2364
    return self._UnlockedGetSsconfValues()
2365

    
2366
  @locking.ssynchronized(_config_lock, shared=1)
2367
  def GetVGName(self):
2368
    """Return the volume group name.
2369

2370
    """
2371
    return self._config_data.cluster.volume_group_name
2372

    
2373
  @locking.ssynchronized(_config_lock)
2374
  def SetVGName(self, vg_name):
2375
    """Set the volume group name.
2376

2377
    """
2378
    self._config_data.cluster.volume_group_name = vg_name
2379
    self._config_data.cluster.serial_no += 1
2380
    self._WriteConfig()
2381

    
2382
  @locking.ssynchronized(_config_lock, shared=1)
2383
  def GetDRBDHelper(self):
2384
    """Return DRBD usermode helper.
2385

2386
    """
2387
    return self._config_data.cluster.drbd_usermode_helper
2388

    
2389
  @locking.ssynchronized(_config_lock)
2390
  def SetDRBDHelper(self, drbd_helper):
2391
    """Set DRBD usermode helper.
2392

2393
    """
2394
    self._config_data.cluster.drbd_usermode_helper = drbd_helper
2395
    self._config_data.cluster.serial_no += 1
2396
    self._WriteConfig()
2397

    
2398
  @locking.ssynchronized(_config_lock, shared=1)
2399
  def GetMACPrefix(self):
2400
    """Return the mac prefix.
2401

2402
    """
2403
    return self._config_data.cluster.mac_prefix
2404

    
2405
  @locking.ssynchronized(_config_lock, shared=1)
2406
  def GetClusterInfo(self):
2407
    """Returns information about the cluster
2408

2409
    @rtype: L{objects.Cluster}
2410
    @return: the cluster object
2411

2412
    """
2413
    return self._config_data.cluster
2414

    
2415
  @locking.ssynchronized(_config_lock, shared=1)
2416
  def HasAnyDiskOfType(self, dev_type):
2417
    """Check if in there is at disk of the given type in the configuration.
2418

2419
    """
2420
    return self._config_data.HasAnyDiskOfType(dev_type)
2421

    
2422
  @locking.ssynchronized(_config_lock)
2423
  def Update(self, target, feedback_fn, ec_id=None):
2424
    """Notify function to be called after updates.
2425

2426
    This function must be called when an object (as returned by
2427
    GetInstanceInfo, GetNodeInfo, GetCluster) has been updated and the
2428
    caller wants the modifications saved to the backing store. Note
2429
    that all modified objects will be saved, but the target argument
2430
    is the one the caller wants to ensure that it's saved.
2431

2432
    @param target: an instance of either L{objects.Cluster},
2433
        L{objects.Node} or L{objects.Instance} which is existing in
2434
        the cluster
2435
    @param feedback_fn: Callable feedback function
2436

2437
    """
2438
    if self._config_data is None:
2439
      raise errors.ProgrammerError("Configuration file not read,"
2440
                                   " cannot save.")
2441
    update_serial = False
2442
    if isinstance(target, objects.Cluster):
2443
      test = target == self._config_data.cluster
2444
    elif isinstance(target, objects.Node):
2445
      test = target in self._config_data.nodes.values()
2446
      update_serial = True
2447
    elif isinstance(target, objects.Instance):
2448
      test = target in self._config_data.instances.values()
2449
    elif isinstance(target, objects.NodeGroup):
2450
      test = target in self._config_data.nodegroups.values()
2451
    elif isinstance(target, objects.Network):
2452
      test = target in self._config_data.networks.values()
2453
    else:
2454
      raise errors.ProgrammerError("Invalid object type (%s) passed to"
2455
                                   " ConfigWriter.Update" % type(target))
2456
    if not test:
2457
      raise errors.ConfigurationError("Configuration updated since object"
2458
                                      " has been read or unknown object")
2459
    target.serial_no += 1
2460
    target.mtime = now = time.time()
2461

    
2462
    if update_serial:
2463
      # for node updates, we need to increase the cluster serial too
2464
      self._config_data.cluster.serial_no += 1
2465
      self._config_data.cluster.mtime = now
2466

    
2467
    if isinstance(target, objects.Instance):
2468
      self._UnlockedReleaseDRBDMinors(target.name)
2469

    
2470
    if ec_id is not None:
2471
      # Commit all ips reserved by OpInstanceSetParams and OpGroupSetParams
2472
      self._UnlockedCommitTemporaryIps(ec_id)
2473

    
2474
    self._WriteConfig(feedback_fn=feedback_fn)
2475

    
2476
  @locking.ssynchronized(_config_lock)
2477
  def DropECReservations(self, ec_id):
2478
    """Drop per-execution-context reservations
2479

2480
    """
2481
    for rm in self._all_rms:
2482
      rm.DropECReservations(ec_id)
2483

    
2484
  @locking.ssynchronized(_config_lock, shared=1)
2485
  def GetAllNetworksInfo(self):
2486
    """Get configuration info of all the networks.
2487

2488
    """
2489
    return dict(self._config_data.networks)
2490

    
2491
  def _UnlockedGetNetworkList(self):
2492
    """Get the list of networks.
2493

2494
    This function is for internal use, when the config lock is already held.
2495

2496
    """
2497
    return self._config_data.networks.keys()
2498

    
2499
  @locking.ssynchronized(_config_lock, shared=1)
2500
  def GetNetworkList(self):
2501
    """Get the list of networks.
2502

2503
    @return: array of networks, ex. ["main", "vlan100", "200]
2504

2505
    """
2506
    return self._UnlockedGetNetworkList()
2507

    
2508
  @locking.ssynchronized(_config_lock, shared=1)
2509
  def GetNetworkNames(self):
2510
    """Get a list of network names
2511

2512
    """
2513
    names = [net.name
2514
             for net in self._config_data.networks.values()]
2515
    return names
2516

    
2517
  def _UnlockedGetNetwork(self, uuid):
2518
    """Returns information about a network.
2519

2520
    This function is for internal use, when the config lock is already held.
2521

2522
    """
2523
    if uuid not in self._config_data.networks:
2524
      return None
2525

    
2526
    return self._config_data.networks[uuid]
2527

    
2528
  @locking.ssynchronized(_config_lock, shared=1)
2529
  def GetNetwork(self, uuid):
2530
    """Returns information about a network.
2531

2532
    It takes the information from the configuration file.
2533

2534
    @param uuid: UUID of the network
2535

2536
    @rtype: L{objects.Network}
2537
    @return: the network object
2538

2539
    """
2540
    return self._UnlockedGetNetwork(uuid)
2541

    
2542
  @locking.ssynchronized(_config_lock)
2543
  def AddNetwork(self, net, ec_id, check_uuid=True):
2544
    """Add a network to the configuration.
2545

2546
    @type net: L{objects.Network}
2547
    @param net: the Network object to add
2548
    @type ec_id: string
2549
    @param ec_id: unique id for the job to use when creating a missing UUID
2550

2551
    """
2552
    self._UnlockedAddNetwork(net, ec_id, check_uuid)
2553
    self._WriteConfig()
2554

    
2555
  def _UnlockedAddNetwork(self, net, ec_id, check_uuid):
2556
    """Add a network to the configuration.
2557

2558
    """
2559
    logging.info("Adding network %s to configuration", net.name)
2560

    
2561
    if check_uuid:
2562
      self._EnsureUUID(net, ec_id)
2563

    
2564
    net.serial_no = 1
2565
    self._config_data.networks[net.uuid] = net
2566
    self._config_data.cluster.serial_no += 1
2567

    
2568
  def _UnlockedLookupNetwork(self, target):
2569
    """Lookup a network's UUID.
2570

2571
    @type target: string
2572
    @param target: network name or UUID
2573
    @rtype: string
2574
    @return: network UUID
2575
    @raises errors.OpPrereqError: when the target network cannot be found
2576

2577
    """
2578
    if target is None:
2579
      return None
2580
    if target in self._config_data.networks:
2581
      return target
2582
    for net in self._config_data.networks.values():
2583
      if net.name == target:
2584
        return net.uuid
2585
    raise errors.OpPrereqError("Network '%s' not found" % target,
2586
                               errors.ECODE_NOENT)
2587

    
2588
  @locking.ssynchronized(_config_lock, shared=1)
2589
  def LookupNetwork(self, target):
2590
    """Lookup a network's UUID.
2591

2592
    This function is just a wrapper over L{_UnlockedLookupNetwork}.
2593

2594
    @type target: string
2595
    @param target: network name or UUID
2596
    @rtype: string
2597
    @return: network UUID
2598

2599
    """
2600
    return self._UnlockedLookupNetwork(target)
2601

    
2602
  @locking.ssynchronized(_config_lock)
2603
  def RemoveNetwork(self, network_uuid):
2604
    """Remove a network from the configuration.
2605

2606
    @type network_uuid: string
2607
    @param network_uuid: the UUID of the network to remove
2608

2609
    """
2610
    logging.info("Removing network %s from configuration", network_uuid)
2611

    
2612
    if network_uuid not in self._config_data.networks:
2613
      raise errors.ConfigurationError("Unknown network '%s'" % network_uuid)
2614

    
2615
    del self._config_data.networks[network_uuid]
2616
    self._config_data.cluster.serial_no += 1
2617
    self._WriteConfig()
2618

    
2619
  def _UnlockedGetGroupNetParams(self, net_uuid, node):
2620
    """Get the netparams (mode, link) of a network.
2621

2622
    Get a network's netparams for a given node.
2623

2624
    @type net_uuid: string
2625
    @param net_uuid: network uuid
2626
    @type node: string
2627
    @param node: node name
2628
    @rtype: dict or None
2629
    @return: netparams
2630

2631
    """
2632
    node_info = self._UnlockedGetNodeInfo(node)
2633
    nodegroup_info = self._UnlockedGetNodeGroup(node_info.group)
2634
    netparams = nodegroup_info.networks.get(net_uuid, None)
2635

    
2636
    return netparams
2637

    
2638
  @locking.ssynchronized(_config_lock, shared=1)
2639
  def GetGroupNetParams(self, net_uuid, node):
2640
    """Locking wrapper of _UnlockedGetGroupNetParams()
2641

2642
    """
2643
    return self._UnlockedGetGroupNetParams(net_uuid, node)
2644

    
2645
  @locking.ssynchronized(_config_lock, shared=1)
2646
  def CheckIPInNodeGroup(self, ip, node):
2647
    """Check IP uniqueness in nodegroup.
2648

2649
    Check networks that are connected in the node's node group
2650
    if ip is contained in any of them. Used when creating/adding
2651
    a NIC to ensure uniqueness among nodegroups.
2652

2653
    @type ip: string
2654
    @param ip: ip address
2655
    @type node: string
2656
    @param node: node name
2657
    @rtype: (string, dict) or (None, None)
2658
    @return: (network name, netparams)
2659

2660
    """
2661
    if ip is None:
2662
      return (None, None)
2663
    node_info = self._UnlockedGetNodeInfo(node)
2664
    nodegroup_info = self._UnlockedGetNodeGroup(node_info.group)
2665
    for net_uuid in nodegroup_info.networks.keys():
2666
      net_info = self._UnlockedGetNetwork(net_uuid)
2667
      pool = network.AddressPool(net_info)
2668
      if pool.Contains(ip):
2669
        return (net_info.name, nodegroup_info.networks[net_uuid])
2670

    
2671
    return (None, None)