gnt-cluster: Add hv/disk state to init
[ganeti-local] / lib / cmdlib.py
index 512c228..04d26f9 100644 (file)
 
 """Module implementing the master-side code."""
 
 
 """Module implementing the master-side code."""
 
-# pylint: disable-msg=W0201,C0302
+# pylint: disable=W0201,C0302
 
 # W0201 since most LU attributes are defined in CheckPrereq or similar
 # functions
 
 
 # W0201 since most LU attributes are defined in CheckPrereq or similar
 # functions
 
-# C0302: since we have waaaay to many lines in this module
+# C0302: since we have waaaay too many lines in this module
 
 import os
 import os.path
 
 import os
 import os.path
@@ -40,6 +40,7 @@ import socket
 import tempfile
 import shutil
 import itertools
 import tempfile
 import shutil
 import itertools
+import operator
 
 from ganeti import ssh
 from ganeti import utils
 
 from ganeti import ssh
 from ganeti import utils
@@ -58,21 +59,20 @@ from ganeti import query
 from ganeti import qlang
 from ganeti import opcodes
 from ganeti import ht
 from ganeti import qlang
 from ganeti import opcodes
 from ganeti import ht
+from ganeti import rpc
 
 
-import ganeti.masterd.instance # pylint: disable-msg=W0611
+import ganeti.masterd.instance # pylint: disable=W0611
 
 
 
 
-def _SupportsOob(cfg, node):
-  """Tells if node supports OOB.
-
-  @type cfg: L{config.ConfigWriter}
-  @param cfg: The cluster configuration
-  @type node: L{objects.Node}
-  @param node: The node
-  @return: The OOB script if supported or an empty string otherwise
+#: Size of DRBD meta block device
+DRBD_META_SIZE = 128
 
 
-  """
-  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
+# States of instance
+INSTANCE_UP = [constants.ADMINST_UP]
+INSTANCE_DOWN = [constants.ADMINST_DOWN]
+INSTANCE_OFFLINE = [constants.ADMINST_OFFLINE]
+INSTANCE_ONLINE = [constants.ADMINST_DOWN, constants.ADMINST_UP]
+INSTANCE_NOT_RUNNING = [constants.ADMINST_DOWN, constants.ADMINST_OFFLINE]
 
 
 class ResultWithJobs:
 
 
 class ResultWithJobs:
@@ -120,7 +120,7 @@ class LogicalUnit(object):
   HTYPE = None
   REQ_BGL = True
 
   HTYPE = None
   REQ_BGL = True
 
-  def __init__(self, processor, op, context, rpc):
+  def __init__(self, processor, op, context, rpc_runner):
     """Constructor for LogicalUnit.
 
     This needs to be overridden in derived classes in order to check op
     """Constructor for LogicalUnit.
 
     This needs to be overridden in derived classes in order to check op
@@ -131,8 +131,10 @@ class LogicalUnit(object):
     self.op = op
     self.cfg = context.cfg
     self.glm = context.glm
     self.op = op
     self.cfg = context.cfg
     self.glm = context.glm
+    # readability alias
+    self.owned_locks = context.glm.list_owned
     self.context = context
     self.context = context
-    self.rpc = rpc
+    self.rpc = rpc_runner
     # Dicts used to declare locking needs to mcpu
     self.needed_locks = None
     self.share_locks = dict.fromkeys(locking.LEVELS, 0)
     # Dicts used to declare locking needs to mcpu
     self.needed_locks = None
     self.share_locks = dict.fromkeys(locking.LEVELS, 0)
@@ -141,10 +143,10 @@ class LogicalUnit(object):
     # Used to force good behavior when calling helper functions
     self.recalculate_locks = {}
     # logging
     # Used to force good behavior when calling helper functions
     self.recalculate_locks = {}
     # logging
-    self.Log = processor.Log # pylint: disable-msg=C0103
-    self.LogWarning = processor.LogWarning # pylint: disable-msg=C0103
-    self.LogInfo = processor.LogInfo # pylint: disable-msg=C0103
-    self.LogStep = processor.LogStep # pylint: disable-msg=C0103
+    self.Log = processor.Log # pylint: disable=C0103
+    self.LogWarning = processor.LogWarning # pylint: disable=C0103
+    self.LogInfo = processor.LogInfo # pylint: disable=C0103
+    self.LogStep = processor.LogStep # pylint: disable=C0103
     # support for dry-run
     self.dry_run_result = None
     # support for generic debug attribute
     # support for dry-run
     self.dry_run_result = None
     # support for generic debug attribute
@@ -332,7 +334,7 @@ class LogicalUnit(object):
     """
     # API must be kept, thus we ignore the unused argument and could
     # be a function warnings
     """
     # API must be kept, thus we ignore the unused argument and could
     # be a function warnings
-    # pylint: disable-msg=W0613,R0201
+    # pylint: disable=W0613,R0201
     return lu_result
 
   def _ExpandAndLockInstance(self):
     return lu_result
 
   def _ExpandAndLockInstance(self):
@@ -354,7 +356,8 @@ class LogicalUnit(object):
                                                 self.op.instance_name)
     self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
 
                                                 self.op.instance_name)
     self.needed_locks[locking.LEVEL_INSTANCE] = self.op.instance_name
 
-  def _LockInstancesNodes(self, primary_only=False):
+  def _LockInstancesNodes(self, primary_only=False,
+                          level=locking.LEVEL_NODE):
     """Helper function to declare instances' nodes for locking.
 
     This function should be called after locking one or more instances to lock
     """Helper function to declare instances' nodes for locking.
 
     This function should be called after locking one or more instances to lock
@@ -375,9 +378,10 @@ class LogicalUnit(object):
 
     @type primary_only: boolean
     @param primary_only: only lock primary nodes of locked instances
 
     @type primary_only: boolean
     @param primary_only: only lock primary nodes of locked instances
+    @param level: Which lock level to use for locking nodes
 
     """
 
     """
-    assert locking.LEVEL_NODE in self.recalculate_locks, \
+    assert level in self.recalculate_locks, \
       "_LockInstancesNodes helper function called with no nodes to recalculate"
 
     # TODO: check if we're really been called with the instance locks held
       "_LockInstancesNodes helper function called with no nodes to recalculate"
 
     # TODO: check if we're really been called with the instance locks held
@@ -386,21 +390,23 @@ class LogicalUnit(object):
     # future we might want to have different behaviors depending on the value
     # of self.recalculate_locks[locking.LEVEL_NODE]
     wanted_nodes = []
     # future we might want to have different behaviors depending on the value
     # of self.recalculate_locks[locking.LEVEL_NODE]
     wanted_nodes = []
-    for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
-      instance = self.context.cfg.GetInstanceInfo(instance_name)
+    locked_i = self.owned_locks(locking.LEVEL_INSTANCE)
+    for _, instance in self.cfg.GetMultiInstanceInfo(locked_i):
       wanted_nodes.append(instance.primary_node)
       if not primary_only:
         wanted_nodes.extend(instance.secondary_nodes)
 
       wanted_nodes.append(instance.primary_node)
       if not primary_only:
         wanted_nodes.extend(instance.secondary_nodes)
 
-    if self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_REPLACE:
-      self.needed_locks[locking.LEVEL_NODE] = wanted_nodes
-    elif self.recalculate_locks[locking.LEVEL_NODE] == constants.LOCKS_APPEND:
-      self.needed_locks[locking.LEVEL_NODE].extend(wanted_nodes)
+    if self.recalculate_locks[level] == constants.LOCKS_REPLACE:
+      self.needed_locks[level] = wanted_nodes
+    elif self.recalculate_locks[level] == constants.LOCKS_APPEND:
+      self.needed_locks[level].extend(wanted_nodes)
+    else:
+      raise errors.ProgrammerError("Unknown recalculation mode")
 
 
-    del self.recalculate_locks[locking.LEVEL_NODE]
+    del self.recalculate_locks[level]
 
 
 
 
-class NoHooksLU(LogicalUnit): # pylint: disable-msg=W0223
+class NoHooksLU(LogicalUnit): # pylint: disable=W0223
   """Simple LU which runs no hooks.
 
   This LU is intended as a parent for other LogicalUnits which will
   """Simple LU which runs no hooks.
 
   This LU is intended as a parent for other LogicalUnits which will
@@ -478,13 +484,13 @@ class _QueryBase:
   #: Attribute holding field definitions
   FIELDS = None
 
   #: Attribute holding field definitions
   FIELDS = None
 
-  def __init__(self, filter_, fields, use_locking):
+  def __init__(self, qfilter, fields, use_locking):
     """Initializes this class.
 
     """
     self.use_locking = use_locking
 
     """Initializes this class.
 
     """
     self.use_locking = use_locking
 
-    self.query = query.Query(self.FIELDS, fields, filter_=filter_,
+    self.query = query.Query(self.FIELDS, fields, qfilter=qfilter,
                              namefield="name")
     self.requested_data = self.query.RequestedData()
     self.names = self.query.RequestedNames()
                              namefield="name")
     self.requested_data = self.query.RequestedData()
     self.names = self.query.RequestedNames()
@@ -500,7 +506,7 @@ class _QueryBase:
 
     """
     if self.do_locking:
 
     """
     if self.do_locking:
-      names = lu.glm.list_owned(lock_level)
+      names = lu.owned_locks(lock_level)
     else:
       names = all_names
 
     else:
       names = all_names
 
@@ -560,6 +566,90 @@ class _QueryBase:
                                     sort_by_name=self.sort_by_name)
 
 
                                     sort_by_name=self.sort_by_name)
 
 
+def _ShareAll():
+  """Returns a dict declaring all lock levels shared.
+
+  """
+  return dict.fromkeys(locking.LEVELS, 1)
+
+
+def _MakeLegacyNodeInfo(data):
+  """Formats the data returned by L{rpc.RpcRunner.call_node_info}.
+
+  Converts the data into a single dictionary. This is fine for most use cases,
+  but some require information from more than one volume group or hypervisor.
+
+  """
+  (bootid, (vg_info, ), (hv_info, )) = data
+
+  return utils.JoinDisjointDicts(utils.JoinDisjointDicts(vg_info, hv_info), {
+    "bootid": bootid,
+    })
+
+
+def _CheckInstanceNodeGroups(cfg, instance_name, owned_groups):
+  """Checks if the owned node groups are still correct for an instance.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: The cluster configuration
+  @type instance_name: string
+  @param instance_name: Instance name
+  @type owned_groups: set or frozenset
+  @param owned_groups: List of currently owned node groups
+
+  """
+  inst_groups = cfg.GetInstanceNodeGroups(instance_name)
+
+  if not owned_groups.issuperset(inst_groups):
+    raise errors.OpPrereqError("Instance %s's node groups changed since"
+                               " locks were acquired, current groups are"
+                               " are '%s', owning groups '%s'; retry the"
+                               " operation" %
+                               (instance_name,
+                                utils.CommaJoin(inst_groups),
+                                utils.CommaJoin(owned_groups)),
+                               errors.ECODE_STATE)
+
+  return inst_groups
+
+
+def _CheckNodeGroupInstances(cfg, group_uuid, owned_instances):
+  """Checks if the instances in a node group are still correct.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: The cluster configuration
+  @type group_uuid: string
+  @param group_uuid: Node group UUID
+  @type owned_instances: set or frozenset
+  @param owned_instances: List of currently owned instances
+
+  """
+  wanted_instances = cfg.GetNodeGroupInstances(group_uuid)
+  if owned_instances != wanted_instances:
+    raise errors.OpPrereqError("Instances in node group '%s' changed since"
+                               " locks were acquired, wanted '%s', have '%s';"
+                               " retry the operation" %
+                               (group_uuid,
+                                utils.CommaJoin(wanted_instances),
+                                utils.CommaJoin(owned_instances)),
+                               errors.ECODE_STATE)
+
+  return wanted_instances
+
+
+def _SupportsOob(cfg, node):
+  """Tells if node supports OOB.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: The cluster configuration
+  @type node: L{objects.Node}
+  @param node: The node
+  @return: The OOB script if supported or an empty string otherwise
+
+  """
+  return cfg.GetNdParams(node)[constants.ND_OOB_PROGRAM]
+
+
 def _GetWantedNodes(lu, nodes):
   """Returns list of checked and expanded node names.
 
 def _GetWantedNodes(lu, nodes):
   """Returns list of checked and expanded node names.
 
@@ -631,6 +721,71 @@ def _GetUpdatedParams(old_params, update_dict,
   return params_copy
 
 
   return params_copy
 
 
+def _UpdateAndVerifySubDict(base, updates, type_check):
+  """Updates and verifies a dict with sub dicts of the same type.
+
+  @param base: The dict with the old data
+  @param updates: The dict with the new data
+  @param type_check: Dict suitable to ForceDictType to verify correct types
+  @returns: A new dict with updated and verified values
+
+  """
+  def fn(old, value):
+    new = _GetUpdatedParams(old, value)
+    utils.ForceDictType(new, type_check)
+    return new
+
+  ret = copy.deepcopy(base)
+  ret.update(dict((key, fn(base.get(key, {}), value))
+                  for key, value in updates.items()))
+  return ret
+
+
+def _MergeAndVerifyHvState(op_input, obj_input):
+  """Combines the hv state from an opcode with the one of the object
+
+  @param op_input: The input dict from the opcode
+  @param obj_input: The input dict from the objects
+  @return: The verified and updated dict
+
+  """
+  if op_input:
+    invalid_hvs = set(op_input) - constants.HYPER_TYPES
+    if invalid_hvs:
+      raise errors.OpPrereqError("Invalid hypervisor(s) in hypervisor state:"
+                                 " %s" % utils.CommaJoin(invalid_hvs),
+                                 errors.ECODE_INVAL)
+    if obj_input is None:
+      obj_input = {}
+    type_check = constants.HVSTS_PARAMETER_TYPES
+    return _UpdateAndVerifySubDict(obj_input, op_input, type_check)
+
+  return None
+
+
+def _MergeAndVerifyDiskState(op_input, obj_input):
+  """Combines the disk state from an opcode with the one of the object
+
+  @param op_input: The input dict from the opcode
+  @param obj_input: The input dict from the objects
+  @return: The verified and updated dict
+  """
+  if op_input:
+    invalid_dst = set(op_input) - constants.DS_VALID_TYPES
+    if invalid_dst:
+      raise errors.OpPrereqError("Invalid storage type(s) in disk state: %s" %
+                                 utils.CommaJoin(invalid_dst),
+                                 errors.ECODE_INVAL)
+    type_check = constants.DSS_PARAMETER_TYPES
+    if obj_input is None:
+      obj_input = {}
+    return dict((key, _UpdateAndVerifySubDict(obj_input.get(key, {}), value,
+                                              type_check))
+                for key, value in op_input.items())
+
+  return None
+
+
 def _ReleaseLocks(lu, level, names=None, keep=None):
   """Releases locks owned by an LU.
 
 def _ReleaseLocks(lu, level, names=None, keep=None):
   """Releases locks owned by an LU.
 
@@ -652,23 +807,28 @@ def _ReleaseLocks(lu, level, names=None, keep=None):
   else:
     should_release = None
 
   else:
     should_release = None
 
-  if should_release:
+  owned = lu.owned_locks(level)
+  if not owned:
+    # Not owning any lock at this level, do nothing
+    pass
+
+  elif should_release:
     retain = []
     release = []
 
     # Determine which locks to release
     retain = []
     release = []
 
     # Determine which locks to release
-    for name in lu.glm.list_owned(level):
+    for name in owned:
       if should_release(name):
         release.append(name)
       else:
         retain.append(name)
 
       if should_release(name):
         release.append(name)
       else:
         retain.append(name)
 
-    assert len(lu.glm.list_owned(level)) == (len(retain) + len(release))
+    assert len(lu.owned_locks(level)) == (len(retain) + len(release))
 
     # Release just some locks
     lu.glm.release(level, names=release)
 
 
     # Release just some locks
     lu.glm.release(level, names=release)
 
-    assert frozenset(lu.glm.list_owned(level)) == frozenset(retain)
+    assert frozenset(lu.owned_locks(level)) == frozenset(retain)
   else:
     # Release everything
     lu.glm.release(level)
   else:
     # Release everything
     lu.glm.release(level)
@@ -676,15 +836,28 @@ def _ReleaseLocks(lu, level, names=None, keep=None):
     assert not lu.glm.is_owned(level), "No locks should be owned"
 
 
     assert not lu.glm.is_owned(level), "No locks should be owned"
 
 
+def _MapInstanceDisksToNodes(instances):
+  """Creates a map from (node, volume) to instance name.
+
+  @type instances: list of L{objects.Instance}
+  @rtype: dict; tuple of (node name, volume name) as key, instance name as value
+
+  """
+  return dict(((node, vol), inst.name)
+              for inst in instances
+              for (node, vols) in inst.MapLVsByNode().items()
+              for vol in vols)
+
+
 def _RunPostHook(lu, node_name):
   """Runs the post-hook for an opcode on a single node.
 
   """
 def _RunPostHook(lu, node_name):
   """Runs the post-hook for an opcode on a single node.
 
   """
-  hm = lu.proc.hmclass(lu.rpc.call_hooks_runner, lu)
+  hm = lu.proc.BuildHooksManager(lu)
   try:
     hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
   except:
   try:
     hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
   except:
-    # pylint: disable-msg=W0702
+    # pylint: disable=W0702
     lu.LogWarning("Errors occurred running hooks on %s" % node_name)
 
 
     lu.LogWarning("Errors occurred running hooks on %s" % node_name)
 
 
@@ -816,20 +989,51 @@ def _GetClusterDomainSecret():
                                strict=True)
 
 
                                strict=True)
 
 
-def _CheckInstanceDown(lu, instance, reason):
-  """Ensure that an instance is not running."""
-  if instance.admin_up:
-    raise errors.OpPrereqError("Instance %s is marked to be up, %s" %
-                               (instance.name, reason), errors.ECODE_STATE)
+def _CheckInstanceState(lu, instance, req_states, msg=None):
+  """Ensure that an instance is in one of the required states.
+
+  @param lu: the LU on behalf of which we make the check
+  @param instance: the instance to check
+  @param msg: if passed, should be a message to replace the default one
+  @raise errors.OpPrereqError: if the instance is not in the required state
+
+  """
+  if msg is None:
+    msg = "can't use instance from outside %s states" % ", ".join(req_states)
+  if instance.admin_state not in req_states:
+    raise errors.OpPrereqError("Instance %s is marked to be %s, %s" %
+                               (instance, instance.admin_state, msg),
+                               errors.ECODE_STATE)
+
+  if constants.ADMINST_UP not in req_states:
+    pnode = instance.primary_node
+    ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
+    ins_l.Raise("Can't contact node %s for instance information" % pnode,
+                prereq=True, ecode=errors.ECODE_ENVIRON)
+
+    if instance.name in ins_l.payload:
+      raise errors.OpPrereqError("Instance %s is running, %s" %
+                                 (instance.name, msg), errors.ECODE_STATE)
+
+
+def _CheckMinMaxSpecs(name, ipolicy, value):
+  """Checks if value is in the desired range.
 
 
-  pnode = instance.primary_node
-  ins_l = lu.rpc.call_instance_list([pnode], [instance.hypervisor])[pnode]
-  ins_l.Raise("Can't contact node %s for instance information" % pnode,
-              prereq=True, ecode=errors.ECODE_ENVIRON)
+  @param name: name of the parameter for which we perform the check
+  @param ipolicy: dictionary containing min, max and std values
+  @param value: actual value that we want to use
+  @return: None or element not meeting the criteria
 
 
-  if instance.name in ins_l.payload:
-    raise errors.OpPrereqError("Instance %s is running, %s" %
-                               (instance.name, reason), errors.ECODE_STATE)
+
+  """
+  if value in [None, constants.VALUE_AUTO]:
+    return None
+  max_v = ipolicy[constants.ISPECS_MAX].get(name, value)
+  min_v = ipolicy[constants.ISPECS_MIN].get(name, value)
+  if value > max_v or min_v > value:
+    return ("%s value %s is not in range [%s, %s]" %
+            (name, value, min_v, max_v))
+  return None
 
 
 def _ExpandItemName(fn, name, kind):
 
 
 def _ExpandItemName(fn, name, kind):
@@ -860,8 +1064,8 @@ def _ExpandInstanceName(cfg, name):
 
 
 def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
 
 
 def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
-                          memory, vcpus, nics, disk_template, disks,
-                          bep, hvp, hypervisor_name):
+                          minmem, maxmem, vcpus, nics, disk_template, disks,
+                          bep, hvp, hypervisor_name, tags):
   """Builds instance related env variables for hooks
 
   This builds the hook environment from individual variables.
   """Builds instance related env variables for hooks
 
   This builds the hook environment from individual variables.
@@ -874,10 +1078,12 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   @param secondary_nodes: list of secondary nodes as strings
   @type os_type: string
   @param os_type: the name of the instance's OS
   @param secondary_nodes: list of secondary nodes as strings
   @type os_type: string
   @param os_type: the name of the instance's OS
-  @type status: boolean
-  @param status: the should_run status of the instance
-  @type memory: string
-  @param memory: the memory size of the instance
+  @type status: string
+  @param status: the desired status of the instance
+  @type minmem: string
+  @param minmem: the minimum memory size of the instance
+  @type maxmem: string
+  @param maxmem: the maximum memory size of the instance
   @type vcpus: string
   @param vcpus: the count of VCPUs the instance has
   @type nics: list
   @type vcpus: string
   @param vcpus: the count of VCPUs the instance has
   @type nics: list
@@ -893,27 +1099,27 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   @param hvp: the hypervisor parameters for the instance
   @type hypervisor_name: string
   @param hypervisor_name: the hypervisor for the instance
   @param hvp: the hypervisor parameters for the instance
   @type hypervisor_name: string
   @param hypervisor_name: the hypervisor for the instance
+  @type tags: list
+  @param tags: list of instance tags as strings
   @rtype: dict
   @return: the hook environment for this instance
 
   """
   @rtype: dict
   @return: the hook environment for this instance
 
   """
-  if status:
-    str_status = "up"
-  else:
-    str_status = "down"
   env = {
     "OP_TARGET": name,
     "INSTANCE_NAME": name,
     "INSTANCE_PRIMARY": primary_node,
     "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
     "INSTANCE_OS_TYPE": os_type,
   env = {
     "OP_TARGET": name,
     "INSTANCE_NAME": name,
     "INSTANCE_PRIMARY": primary_node,
     "INSTANCE_SECONDARIES": " ".join(secondary_nodes),
     "INSTANCE_OS_TYPE": os_type,
-    "INSTANCE_STATUS": str_status,
-    "INSTANCE_MEMORY": memory,
+    "INSTANCE_STATUS": status,
+    "INSTANCE_MINMEM": minmem,
+    "INSTANCE_MAXMEM": maxmem,
+    # TODO(2.7) remove deprecated "memory" value
+    "INSTANCE_MEMORY": maxmem,
     "INSTANCE_VCPUS": vcpus,
     "INSTANCE_DISK_TEMPLATE": disk_template,
     "INSTANCE_HYPERVISOR": hypervisor_name,
   }
     "INSTANCE_VCPUS": vcpus,
     "INSTANCE_DISK_TEMPLATE": disk_template,
     "INSTANCE_HYPERVISOR": hypervisor_name,
   }
-
   if nics:
     nic_count = len(nics)
     for idx, (ip, mac, mode, link) in enumerate(nics):
   if nics:
     nic_count = len(nics)
     for idx, (ip, mac, mode, link) in enumerate(nics):
@@ -940,6 +1146,11 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
 
   env["INSTANCE_DISK_COUNT"] = disk_count
 
 
   env["INSTANCE_DISK_COUNT"] = disk_count
 
+  if not tags:
+    tags = []
+
+  env["INSTANCE_TAGS"] = " ".join(tags)
+
   for source, kind in [(bep, "BE"), (hvp, "HV")]:
     for key, value in source.items():
       env["INSTANCE_%s_%s" % (kind, key)] = value
   for source, kind in [(bep, "BE"), (hvp, "HV")]:
     for key, value in source.items():
       env["INSTANCE_%s_%s" % (kind, key)] = value
@@ -990,23 +1201,25 @@ def _BuildInstanceHookEnvByObject(lu, instance, override=None):
   bep = cluster.FillBE(instance)
   hvp = cluster.FillHV(instance)
   args = {
   bep = cluster.FillBE(instance)
   hvp = cluster.FillHV(instance)
   args = {
-    'name': instance.name,
-    'primary_node': instance.primary_node,
-    'secondary_nodes': instance.secondary_nodes,
-    'os_type': instance.os,
-    'status': instance.admin_up,
-    'memory': bep[constants.BE_MEMORY],
-    'vcpus': bep[constants.BE_VCPUS],
-    'nics': _NICListToTuple(lu, instance.nics),
-    'disk_template': instance.disk_template,
-    'disks': [(disk.size, disk.mode) for disk in instance.disks],
-    'bep': bep,
-    'hvp': hvp,
-    'hypervisor_name': instance.hypervisor,
+    "name": instance.name,
+    "primary_node": instance.primary_node,
+    "secondary_nodes": instance.secondary_nodes,
+    "os_type": instance.os,
+    "status": instance.admin_state,
+    "maxmem": bep[constants.BE_MAXMEM],
+    "minmem": bep[constants.BE_MINMEM],
+    "vcpus": bep[constants.BE_VCPUS],
+    "nics": _NICListToTuple(lu, instance.nics),
+    "disk_template": instance.disk_template,
+    "disks": [(disk.size, disk.mode) for disk in instance.disks],
+    "bep": bep,
+    "hvp": hvp,
+    "hypervisor_name": instance.hypervisor,
+    "tags": instance.tags,
   }
   if override:
     args.update(override)
   }
   if override:
     args.update(override)
-  return _BuildInstanceHookEnv(**args) # pylint: disable-msg=W0142
+  return _BuildInstanceHookEnv(**args) # pylint: disable=W0142
 
 
 def _AdjustCandidatePool(lu, exceptions):
 
 
 def _AdjustCandidatePool(lu, exceptions):
@@ -1036,6 +1249,14 @@ def _DecideSelfPromotion(lu, exceptions=None):
   return mc_now < mc_should
 
 
   return mc_now < mc_should
 
 
+def _CalculateGroupIPolicy(cfg, group):
+  """Calculate instance policy for group.
+
+  """
+  cluster = cfg.GetClusterInfo()
+  return cluster.SimpleFillIPolicy(group.ipolicy)
+
+
 def _CheckNicsBridgesExist(lu, target_nics, target_node):
   """Check that the brigdes needed by a list of nics exist.
 
 def _CheckNicsBridgesExist(lu, target_nics, target_node):
   """Check that the brigdes needed by a list of nics exist.
 
@@ -1068,9 +1289,13 @@ def _CheckOSVariant(os_obj, name):
   @param name: OS name passed by the user, to check for validity
 
   """
   @param name: OS name passed by the user, to check for validity
 
   """
+  variant = objects.OS.GetVariant(name)
   if not os_obj.supported_variants:
   if not os_obj.supported_variants:
+    if variant:
+      raise errors.OpPrereqError("OS '%s' doesn't support variants ('%s'"
+                                 " passed)" % (os_obj.name, variant),
+                                 errors.ECODE_INVAL)
     return
     return
-  variant = objects.OS.GetVariant(name)
   if not variant:
     raise errors.OpPrereqError("OS name must include a variant",
                                errors.ECODE_INVAL)
   if not variant:
     raise errors.OpPrereqError("OS name must include a variant",
                                errors.ECODE_INVAL)
@@ -1119,13 +1344,13 @@ def _GetStorageTypeArgs(cfg, storage_type):
   return []
 
 
   return []
 
 
-def _FindFaultyInstanceDisks(cfg, rpc, instance, node_name, prereq):
+def _FindFaultyInstanceDisks(cfg, rpc_runner, instance, node_name, prereq):
   faulty = []
 
   for dev in instance.disks:
     cfg.SetDiskID(dev, node_name)
 
   faulty = []
 
   for dev in instance.disks:
     cfg.SetDiskID(dev, node_name)
 
-  result = rpc.call_blockdev_getmirrorstatus(node_name, instance.disks)
+  result = rpc_runner.call_blockdev_getmirrorstatus(node_name, instance.disks)
   result.Raise("Failed to get disk status from node %s" % node_name,
                prereq=prereq, ecode=errors.ECODE_ENVIRON)
 
   result.Raise("Failed to get disk status from node %s" % node_name,
                prereq=prereq, ecode=errors.ECODE_ENVIRON)
 
@@ -1168,6 +1393,29 @@ def _CheckIAllocatorOrNode(lu, iallocator_slot, node_slot):
                                  " iallocator")
 
 
                                  " iallocator")
 
 
+def _GetDefaultIAllocator(cfg, iallocator):
+  """Decides on which iallocator to use.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: Cluster configuration object
+  @type iallocator: string or None
+  @param iallocator: Iallocator specified in opcode
+  @rtype: string
+  @return: Iallocator name
+
+  """
+  if not iallocator:
+    # Use default iallocator
+    iallocator = cfg.GetDefaultIAllocator()
+
+  if not iallocator:
+    raise errors.OpPrereqError("No iallocator was specified, neither in the"
+                               " opcode nor as a cluster-wide default",
+                               errors.ECODE_INVAL)
+
+  return iallocator
+
+
 class LUClusterPostInit(LogicalUnit):
   """Logical unit for running hooks after cluster initialization.
 
 class LUClusterPostInit(LogicalUnit):
   """Logical unit for running hooks after cluster initialization.
 
@@ -1242,19 +1490,21 @@ class LUClusterDestroy(LogicalUnit):
     """Destroys the cluster.
 
     """
     """Destroys the cluster.
 
     """
-    master = self.cfg.GetMasterNode()
+    master_params = self.cfg.GetMasterNetworkParameters()
 
     # Run post hooks on master node before it's removed
 
     # Run post hooks on master node before it's removed
-    _RunPostHook(self, master)
+    _RunPostHook(self, master_params.name)
 
 
-    result = self.rpc.call_node_stop_master(master, False)
+    ems = self.cfg.GetUseExternalMipScript()
+    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+                                                     master_params, ems)
     result.Raise("Could not disable the master role")
 
     result.Raise("Could not disable the master role")
 
-    return master
+    return master_params.name
 
 
 def _VerifyCertificate(filename):
 
 
 def _VerifyCertificate(filename):
-  """Verifies a certificate for LUClusterVerifyConfig.
+  """Verifies a certificate for L{LUClusterVerifyConfig}.
 
   @type filename: string
   @param filename: Path to PEM file
 
   @type filename: string
   @param filename: Path to PEM file
@@ -1263,7 +1513,7 @@ def _VerifyCertificate(filename):
   try:
     cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
                                            utils.ReadFile(filename))
   try:
     cert = OpenSSL.crypto.load_certificate(OpenSSL.crypto.FILETYPE_PEM,
                                            utils.ReadFile(filename))
-  except Exception, err: # pylint: disable-msg=W0703
+  except Exception, err: # pylint: disable=W0703
     return (LUClusterVerifyConfig.ETYPE_ERROR,
             "Failed to load X509 certificate %s: %s" % (filename, err))
 
     return (LUClusterVerifyConfig.ETYPE_ERROR,
             "Failed to load X509 certificate %s: %s" % (filename, err))
 
@@ -1325,39 +1575,6 @@ class _VerifyErrors(object):
   self.op and self._feedback_fn to be available.)
 
   """
   self.op and self._feedback_fn to be available.)
 
   """
-  TCLUSTER = "cluster"
-  TNODE = "node"
-  TINSTANCE = "instance"
-
-  ECLUSTERCFG = (TCLUSTER, "ECLUSTERCFG")
-  ECLUSTERCERT = (TCLUSTER, "ECLUSTERCERT")
-  ECLUSTERFILECHECK = (TCLUSTER, "ECLUSTERFILECHECK")
-  ECLUSTERDANGLINGNODES = (TNODE, "ECLUSTERDANGLINGNODES")
-  ECLUSTERDANGLINGINST = (TNODE, "ECLUSTERDANGLINGINST")
-  EINSTANCEBADNODE = (TINSTANCE, "EINSTANCEBADNODE")
-  EINSTANCEDOWN = (TINSTANCE, "EINSTANCEDOWN")
-  EINSTANCELAYOUT = (TINSTANCE, "EINSTANCELAYOUT")
-  EINSTANCEMISSINGDISK = (TINSTANCE, "EINSTANCEMISSINGDISK")
-  EINSTANCEFAULTYDISK = (TINSTANCE, "EINSTANCEFAULTYDISK")
-  EINSTANCEWRONGNODE = (TINSTANCE, "EINSTANCEWRONGNODE")
-  EINSTANCESPLITGROUPS = (TINSTANCE, "EINSTANCESPLITGROUPS")
-  ENODEDRBD = (TNODE, "ENODEDRBD")
-  ENODEDRBDHELPER = (TNODE, "ENODEDRBDHELPER")
-  ENODEFILECHECK = (TNODE, "ENODEFILECHECK")
-  ENODEHOOKS = (TNODE, "ENODEHOOKS")
-  ENODEHV = (TNODE, "ENODEHV")
-  ENODELVM = (TNODE, "ENODELVM")
-  ENODEN1 = (TNODE, "ENODEN1")
-  ENODENET = (TNODE, "ENODENET")
-  ENODEOS = (TNODE, "ENODEOS")
-  ENODEORPHANINSTANCE = (TNODE, "ENODEORPHANINSTANCE")
-  ENODEORPHANLV = (TNODE, "ENODEORPHANLV")
-  ENODERPC = (TNODE, "ENODERPC")
-  ENODESSH = (TNODE, "ENODESSH")
-  ENODEVERSION = (TNODE, "ENODEVERSION")
-  ENODESETUP = (TNODE, "ENODESETUP")
-  ENODETIME = (TNODE, "ENODETIME")
-  ENODEOOBPATH = (TNODE, "ENODEOOBPATH")
 
   ETYPE_FIELD = "code"
   ETYPE_ERROR = "ERROR"
 
   ETYPE_FIELD = "code"
   ETYPE_ERROR = "ERROR"
@@ -1373,12 +1590,12 @@ class _VerifyErrors(object):
 
     """
     ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
 
     """
     ltype = kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR)
-    itype, etxt = ecode
+    itype, etxt, _ = ecode
     # first complete the msg
     if args:
       msg = msg % args
     # then format the whole message
     # first complete the msg
     if args:
       msg = msg % args
     # then format the whole message
-    if self.op.error_codes: # This is a mix-in. pylint: disable-msg=E1101
+    if self.op.error_codes: # This is a mix-in. pylint: disable=E1101
       msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
     else:
       if item:
       msg = "%s:%s:%s:%s:%s" % (ltype, etxt, itype, item, msg)
     else:
       if item:
@@ -1387,26 +1604,78 @@ class _VerifyErrors(object):
         item = ""
       msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
     # and finally report it via the feedback_fn
         item = ""
       msg = "%s: %s%s: %s" % (ltype, itype, item, msg)
     # and finally report it via the feedback_fn
-    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable-msg=E1101
+    self._feedback_fn("  - %s" % msg) # Mix-in. pylint: disable=E1101
 
 
-  def _ErrorIf(self, cond, *args, **kwargs):
+  def _ErrorIf(self, cond, ecode, *args, **kwargs):
     """Log an error message if the passed condition is True.
 
     """
     cond = (bool(cond)
     """Log an error message if the passed condition is True.
 
     """
     cond = (bool(cond)
-            or self.op.debug_simulate_errors) # pylint: disable-msg=E1101
+            or self.op.debug_simulate_errors) # pylint: disable=E1101
+
+    # If the error code is in the list of ignored errors, demote the error to a
+    # warning
+    (_, etxt, _) = ecode
+    if etxt in self.op.ignore_errors:     # pylint: disable=E1101
+      kwargs[self.ETYPE_FIELD] = self.ETYPE_WARNING
+
     if cond:
     if cond:
-      self._Error(*args, **kwargs)
+      self._Error(ecode, *args, **kwargs)
+
     # do not mark the operation as failed for WARN cases only
     if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
       self.bad = self.bad or cond
 
 
     # do not mark the operation as failed for WARN cases only
     if kwargs.get(self.ETYPE_FIELD, self.ETYPE_ERROR) == self.ETYPE_ERROR:
       self.bad = self.bad or cond
 
 
+class LUClusterVerify(NoHooksLU):
+  """Submits all jobs necessary to verify the cluster.
+
+  """
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.needed_locks = {}
+
+  def Exec(self, feedback_fn):
+    jobs = []
+
+    if self.op.group_name:
+      groups = [self.op.group_name]
+      depends_fn = lambda: None
+    else:
+      groups = self.cfg.GetNodeGroupList()
+
+      # Verify global configuration
+      jobs.append([
+        opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
+        ])
+
+      # Always depend on global verification
+      depends_fn = lambda: [(-len(jobs), [])]
+
+    jobs.extend([opcodes.OpClusterVerifyGroup(group_name=group,
+                                            ignore_errors=self.op.ignore_errors,
+                                            depends=depends_fn())]
+                for group in groups)
+
+    # Fix up all parameters
+    for op in itertools.chain(*jobs): # pylint: disable=W0142
+      op.debug_simulate_errors = self.op.debug_simulate_errors
+      op.verbose = self.op.verbose
+      op.error_codes = self.op.error_codes
+      try:
+        op.skip_checks = self.op.skip_checks
+      except AttributeError:
+        assert not isinstance(op, opcodes.OpClusterVerifyGroup)
+
+    return ResultWithJobs(jobs)
+
+
 class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
   """Verifies the cluster config.
 
   """
 class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
   """Verifies the cluster config.
 
   """
-  REQ_BGL = False
+  REQ_BGL = True
 
   def _VerifyHVP(self, hvp_data):
     """Verifies locally the syntax of the hypervisor parameters.
 
   def _VerifyHVP(self, hvp_data):
     """Verifies locally the syntax of the hypervisor parameters.
@@ -1420,9 +1689,12 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
         utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
         hv_class.CheckParameterSyntax(hv_params)
       except errors.GenericError, err:
         utils.ForceDictType(hv_params, constants.HVS_PARAMETER_TYPES)
         hv_class.CheckParameterSyntax(hv_params)
       except errors.GenericError, err:
-        self._ErrorIf(True, self.ECLUSTERCFG, None, msg % str(err))
+        self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg % str(err))
 
   def ExpandNames(self):
 
   def ExpandNames(self):
+    # Information can be safely retrieved as the BGL is acquired in exclusive
+    # mode
+    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
     self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
     self.all_node_info = self.cfg.GetAllNodesInfo()
     self.all_inst_info = self.cfg.GetAllInstancesInfo()
     self.all_group_info = self.cfg.GetAllNodeGroupsInfo()
     self.all_node_info = self.cfg.GetAllNodesInfo()
     self.all_inst_info = self.cfg.GetAllInstancesInfo()
@@ -1438,13 +1710,13 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
     feedback_fn("* Verifying cluster config")
 
     for msg in self.cfg.VerifyConfig():
     feedback_fn("* Verifying cluster config")
 
     for msg in self.cfg.VerifyConfig():
-      self._ErrorIf(True, self.ECLUSTERCFG, None, msg)
+      self._ErrorIf(True, constants.CV_ECLUSTERCFG, None, msg)
 
     feedback_fn("* Verifying cluster certificate files")
 
     for cert_filename in constants.ALL_CERT_FILES:
       (errcode, msg) = _VerifyCertificate(cert_filename)
 
     feedback_fn("* Verifying cluster certificate files")
 
     for cert_filename in constants.ALL_CERT_FILES:
       (errcode, msg) = _VerifyCertificate(cert_filename)
-      self._ErrorIf(errcode, self.ECLUSTERCERT, None, msg, code=errcode)
+      self._ErrorIf(errcode, constants.CV_ECLUSTERCERT, None, msg, code=errcode)
 
     feedback_fn("* Verifying hypervisor parameters")
 
 
     feedback_fn("* Verifying hypervisor parameters")
 
@@ -1454,7 +1726,7 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
     feedback_fn("* Verifying all nodes belong to an existing group")
 
     # We do this verification here because, should this bogus circumstance
     feedback_fn("* Verifying all nodes belong to an existing group")
 
     # We do this verification here because, should this bogus circumstance
-    # occur, it would never be catched by VerifyGroup, which only acts on
+    # occur, it would never be caught by VerifyGroup, which only acts on
     # nodes/instances reachable from existing node groups.
 
     dangling_nodes = set(node.name for node in self.all_node_info.values()
     # nodes/instances reachable from existing node groups.
 
     dangling_nodes = set(node.name for node in self.all_node_info.values()
@@ -1476,15 +1748,17 @@ class LUClusterVerifyConfig(NoHooksLU, _VerifyErrors):
                                                 ["no instances"])))
         for node in dangling_nodes]
 
                                                 ["no instances"])))
         for node in dangling_nodes]
 
-    self._ErrorIf(bool(dangling_nodes), self.ECLUSTERDANGLINGNODES, None,
+    self._ErrorIf(bool(dangling_nodes), constants.CV_ECLUSTERDANGLINGNODES,
+                  None,
                   "the following nodes (and their instances) belong to a non"
                   " existing group: %s", utils.CommaJoin(pretty_dangling))
 
                   "the following nodes (and their instances) belong to a non"
                   " existing group: %s", utils.CommaJoin(pretty_dangling))
 
-    self._ErrorIf(bool(no_node_instances), self.ECLUSTERDANGLINGINST, None,
+    self._ErrorIf(bool(no_node_instances), constants.CV_ECLUSTERDANGLINGINST,
+                  None,
                   "the following instances have a non-existing primary-node:"
                   " %s", utils.CommaJoin(no_node_instances))
 
                   "the following instances have a non-existing primary-node:"
                   " %s", utils.CommaJoin(no_node_instances))
 
-    return (not self.bad, [g.name for g in self.all_group_info.values()])
+    return not self.bad
 
 
 class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
 
 class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
@@ -1551,60 +1825,60 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     # This raises errors.OpPrereqError on its own:
     self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
 
     # This raises errors.OpPrereqError on its own:
     self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
 
-    all_node_info = self.cfg.GetAllNodesInfo()
-    all_inst_info = self.cfg.GetAllInstancesInfo()
+    # Get instances in node group; this is unsafe and needs verification later
+    inst_names = self.cfg.GetNodeGroupInstances(self.group_uuid)
 
 
-    node_names = set(node.name
-                     for node in all_node_info.values()
-                     if node.group == self.group_uuid)
+    self.needed_locks = {
+      locking.LEVEL_INSTANCE: inst_names,
+      locking.LEVEL_NODEGROUP: [self.group_uuid],
+      locking.LEVEL_NODE: [],
+      }
 
 
-    inst_names = [inst.name
-                  for inst in all_inst_info.values()
-                  if inst.primary_node in node_names]
+    self.share_locks = _ShareAll()
 
 
-    # In Exec(), we warn about mirrored instances that have primary and
-    # secondary living in separate node groups. To fully verify that
-    # volumes for these instances are healthy, we will need to do an
-    # extra call to their secondaries. We ensure here those nodes will
-    # be locked.
-    for inst in inst_names:
-      if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
-        node_names.update(all_inst_info[inst].secondary_nodes)
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      # Get members of node group; this is unsafe and needs verification later
+      nodes = set(self.cfg.GetNodeGroup(self.group_uuid).members)
 
 
-    self.needed_locks = {
-      locking.LEVEL_NODEGROUP: [self.group_uuid],
-      locking.LEVEL_NODE: list(node_names),
-      locking.LEVEL_INSTANCE: inst_names,
-    }
+      all_inst_info = self.cfg.GetAllInstancesInfo()
 
 
-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+      # In Exec(), we warn about mirrored instances that have primary and
+      # secondary living in separate node groups. To fully verify that
+      # volumes for these instances are healthy, we will need to do an
+      # extra call to their secondaries. We ensure here those nodes will
+      # be locked.
+      for inst in self.owned_locks(locking.LEVEL_INSTANCE):
+        # Important: access only the instances whose lock is owned
+        if all_inst_info[inst].disk_template in constants.DTS_INT_MIRROR:
+          nodes.update(all_inst_info[inst].secondary_nodes)
 
 
-  def CheckPrereq(self):
-    self.all_node_info = self.cfg.GetAllNodesInfo()
-    self.all_inst_info = self.cfg.GetAllInstancesInfo()
+      self.needed_locks[locking.LEVEL_NODE] = nodes
 
 
-    group_nodes = set(node.name
-                      for node in self.all_node_info.values()
-                      if node.group == self.group_uuid)
+  def CheckPrereq(self):
+    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+    self.group_info = self.cfg.GetNodeGroup(self.group_uuid)
 
 
-    group_instances = set(inst.name
-                          for inst in self.all_inst_info.values()
-                          if inst.primary_node in group_nodes)
+    group_nodes = set(self.group_info.members)
+    group_instances = self.cfg.GetNodeGroupInstances(self.group_uuid)
 
     unlocked_nodes = \
 
     unlocked_nodes = \
-        group_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+        group_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
 
     unlocked_instances = \
 
     unlocked_instances = \
-        group_instances.difference(self.glm.list_owned(locking.LEVEL_INSTANCE))
+        group_instances.difference(self.owned_locks(locking.LEVEL_INSTANCE))
 
     if unlocked_nodes:
 
     if unlocked_nodes:
-      raise errors.OpPrereqError("missing lock for nodes: %s" %
+      raise errors.OpPrereqError("Missing lock for nodes: %s" %
                                  utils.CommaJoin(unlocked_nodes))
 
     if unlocked_instances:
                                  utils.CommaJoin(unlocked_nodes))
 
     if unlocked_instances:
-      raise errors.OpPrereqError("missing lock for instances: %s" %
+      raise errors.OpPrereqError("Missing lock for instances: %s" %
                                  utils.CommaJoin(unlocked_instances))
 
                                  utils.CommaJoin(unlocked_instances))
 
+    self.all_node_info = self.cfg.GetAllNodesInfo()
+    self.all_inst_info = self.cfg.GetAllInstancesInfo()
+
     self.my_node_names = utils.NiceSort(group_nodes)
     self.my_inst_names = utils.NiceSort(group_instances)
 
     self.my_node_names = utils.NiceSort(group_nodes)
     self.my_inst_names = utils.NiceSort(group_instances)
 
@@ -1626,7 +1900,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
             extra_lv_nodes.add(nname)
 
     unlocked_lv_nodes = \
             extra_lv_nodes.add(nname)
 
     unlocked_lv_nodes = \
-        extra_lv_nodes.difference(self.glm.list_owned(locking.LEVEL_NODE))
+        extra_lv_nodes.difference(self.owned_locks(locking.LEVEL_NODE))
 
     if unlocked_lv_nodes:
       raise errors.OpPrereqError("these nodes could be locked: %s" %
 
     if unlocked_lv_nodes:
       raise errors.OpPrereqError("these nodes could be locked: %s" %
@@ -1649,11 +1923,11 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     # main result, nresult should be a non-empty dict
     test = not nresult or not isinstance(nresult, dict)
 
     # main result, nresult should be a non-empty dict
     test = not nresult or not isinstance(nresult, dict)
-    _ErrorIf(test, self.ENODERPC, node,
+    _ErrorIf(test, constants.CV_ENODERPC, node,
                   "unable to verify node: no data returned")
     if test:
       return False
                   "unable to verify node: no data returned")
     if test:
       return False
@@ -1664,13 +1938,13 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     test = not (remote_version and
                 isinstance(remote_version, (list, tuple)) and
                 len(remote_version) == 2)
     test = not (remote_version and
                 isinstance(remote_version, (list, tuple)) and
                 len(remote_version) == 2)
-    _ErrorIf(test, self.ENODERPC, node,
+    _ErrorIf(test, constants.CV_ENODERPC, node,
              "connection to node returned invalid data")
     if test:
       return False
 
     test = local_version != remote_version[0]
              "connection to node returned invalid data")
     if test:
       return False
 
     test = local_version != remote_version[0]
-    _ErrorIf(test, self.ENODEVERSION, node,
+    _ErrorIf(test, constants.CV_ENODEVERSION, node,
              "incompatible protocol versions: master %s,"
              " node %s", local_version, remote_version[0])
     if test:
              "incompatible protocol versions: master %s,"
              " node %s", local_version, remote_version[0])
     if test:
@@ -1680,7 +1954,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     # full package version
     self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
 
     # full package version
     self._ErrorIf(constants.RELEASE_VERSION != remote_version[1],
-                  self.ENODEVERSION, node,
+                  constants.CV_ENODEVERSION, node,
                   "software version mismatch: master %s, node %s",
                   constants.RELEASE_VERSION, remote_version[1],
                   code=self.ETYPE_WARNING)
                   "software version mismatch: master %s, node %s",
                   constants.RELEASE_VERSION, remote_version[1],
                   code=self.ETYPE_WARNING)
@@ -1689,19 +1963,19 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     if ninfo.vm_capable and isinstance(hyp_result, dict):
       for hv_name, hv_result in hyp_result.iteritems():
         test = hv_result is not None
     if ninfo.vm_capable and isinstance(hyp_result, dict):
       for hv_name, hv_result in hyp_result.iteritems():
         test = hv_result is not None
-        _ErrorIf(test, self.ENODEHV, node,
+        _ErrorIf(test, constants.CV_ENODEHV, node,
                  "hypervisor %s verify failure: '%s'", hv_name, hv_result)
 
     hvp_result = nresult.get(constants.NV_HVPARAMS, None)
     if ninfo.vm_capable and isinstance(hvp_result, list):
       for item, hv_name, hv_result in hvp_result:
                  "hypervisor %s verify failure: '%s'", hv_name, hv_result)
 
     hvp_result = nresult.get(constants.NV_HVPARAMS, None)
     if ninfo.vm_capable and isinstance(hvp_result, list):
       for item, hv_name, hv_result in hvp_result:
-        _ErrorIf(True, self.ENODEHV, node,
+        _ErrorIf(True, constants.CV_ENODEHV, node,
                  "hypervisor %s parameter verify failure (source %s): %s",
                  hv_name, item, hv_result)
 
     test = nresult.get(constants.NV_NODESETUP,
                        ["Missing NODESETUP results"])
                  "hypervisor %s parameter verify failure (source %s): %s",
                  hv_name, item, hv_result)
 
     test = nresult.get(constants.NV_NODESETUP,
                        ["Missing NODESETUP results"])
-    _ErrorIf(test, self.ENODESETUP, node, "node setup error: %s",
+    _ErrorIf(test, constants.CV_ENODESETUP, node, "node setup error: %s",
              "; ".join(test))
 
     return True
              "; ".join(test))
 
     return True
@@ -1718,13 +1992,13 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     ntime = nresult.get(constants.NV_TIME, None)
     try:
       ntime_merged = utils.MergeTime(ntime)
     except (ValueError, TypeError):
 
     ntime = nresult.get(constants.NV_TIME, None)
     try:
       ntime_merged = utils.MergeTime(ntime)
     except (ValueError, TypeError):
-      _ErrorIf(True, self.ENODETIME, node, "Node returned invalid time")
+      _ErrorIf(True, constants.CV_ENODETIME, node, "Node returned invalid time")
       return
 
     if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
       return
 
     if ntime_merged < (nvinfo_starttime - constants.NODE_MAX_CLOCK_SKEW):
@@ -1734,7 +2008,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     else:
       ntime_diff = None
 
     else:
       ntime_diff = None
 
-    _ErrorIf(ntime_diff is not None, self.ENODETIME, node,
+    _ErrorIf(ntime_diff is not None, constants.CV_ENODETIME, node,
              "Node time diverges by at least %s from master node time",
              ntime_diff)
 
              "Node time diverges by at least %s from master node time",
              ntime_diff)
 
@@ -1751,29 +2025,30 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       return
 
     node = ninfo.name
       return
 
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     # checks vg existence and size > 20G
     vglist = nresult.get(constants.NV_VGLIST, None)
     test = not vglist
 
     # checks vg existence and size > 20G
     vglist = nresult.get(constants.NV_VGLIST, None)
     test = not vglist
-    _ErrorIf(test, self.ENODELVM, node, "unable to check volume groups")
+    _ErrorIf(test, constants.CV_ENODELVM, node, "unable to check volume groups")
     if not test:
       vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
                                             constants.MIN_VG_SIZE)
     if not test:
       vgstatus = utils.CheckVolumeGroupSize(vglist, vg_name,
                                             constants.MIN_VG_SIZE)
-      _ErrorIf(vgstatus, self.ENODELVM, node, vgstatus)
+      _ErrorIf(vgstatus, constants.CV_ENODELVM, node, vgstatus)
 
     # check pv names
     pvlist = nresult.get(constants.NV_PVLIST, None)
     test = pvlist is None
 
     # check pv names
     pvlist = nresult.get(constants.NV_PVLIST, None)
     test = pvlist is None
-    _ErrorIf(test, self.ENODELVM, node, "Can't get PV list from node")
+    _ErrorIf(test, constants.CV_ENODELVM, node, "Can't get PV list from node")
     if not test:
       # check that ':' is not present in PV names, since it's a
       # special character for lvcreate (denotes the range of PEs to
       # use on the PV)
       for _, pvname, owner_vg in pvlist:
         test = ":" in pvname
     if not test:
       # check that ':' is not present in PV names, since it's a
       # special character for lvcreate (denotes the range of PEs to
       # use on the PV)
       for _, pvname, owner_vg in pvlist:
         test = ":" in pvname
-        _ErrorIf(test, self.ENODELVM, node, "Invalid character ':' in PV"
-                 " '%s' of VG '%s'", pvname, owner_vg)
+        _ErrorIf(test, constants.CV_ENODELVM, node,
+                 "Invalid character ':' in PV '%s' of VG '%s'",
+                 pvname, owner_vg)
 
   def _VerifyNodeBridges(self, ninfo, nresult, bridges):
     """Check the node bridges.
 
   def _VerifyNodeBridges(self, ninfo, nresult, bridges):
     """Check the node bridges.
@@ -1788,15 +2063,35 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       return
 
     node = ninfo.name
       return
 
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     missing = nresult.get(constants.NV_BRIDGES, None)
     test = not isinstance(missing, list)
 
     missing = nresult.get(constants.NV_BRIDGES, None)
     test = not isinstance(missing, list)
-    _ErrorIf(test, self.ENODENET, node,
+    _ErrorIf(test, constants.CV_ENODENET, node,
              "did not return valid bridge information")
     if not test:
              "did not return valid bridge information")
     if not test:
-      _ErrorIf(bool(missing), self.ENODENET, node, "missing bridges: %s" %
-               utils.CommaJoin(sorted(missing)))
+      _ErrorIf(bool(missing), constants.CV_ENODENET, node,
+               "missing bridges: %s" % utils.CommaJoin(sorted(missing)))
+
+  def _VerifyNodeUserScripts(self, ninfo, nresult):
+    """Check the results of user scripts presence and executability on the node
+
+    @type ninfo: L{objects.Node}
+    @param ninfo: the node to check
+    @param nresult: the remote results for the node
+
+    """
+    node = ninfo.name
+
+    test = not constants.NV_USERSCRIPTS in nresult
+    self._ErrorIf(test, constants.CV_ENODEUSERSCRIPTS, node,
+                  "did not return user scripts information")
+
+    broken_scripts = nresult.get(constants.NV_USERSCRIPTS, None)
+    if not test:
+      self._ErrorIf(broken_scripts, constants.CV_ENODEUSERSCRIPTS, node,
+                    "user scripts not present or not executable: %s" %
+                    utils.CommaJoin(sorted(broken_scripts)))
 
   def _VerifyNodeNetwork(self, ninfo, nresult):
     """Check the node network connectivity results.
 
   def _VerifyNodeNetwork(self, ninfo, nresult):
     """Check the node network connectivity results.
@@ -1807,30 +2102,30 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     test = constants.NV_NODELIST not in nresult
 
     test = constants.NV_NODELIST not in nresult
-    _ErrorIf(test, self.ENODESSH, node,
+    _ErrorIf(test, constants.CV_ENODESSH, node,
              "node hasn't returned node ssh connectivity data")
     if not test:
       if nresult[constants.NV_NODELIST]:
         for a_node, a_msg in nresult[constants.NV_NODELIST].items():
              "node hasn't returned node ssh connectivity data")
     if not test:
       if nresult[constants.NV_NODELIST]:
         for a_node, a_msg in nresult[constants.NV_NODELIST].items():
-          _ErrorIf(True, self.ENODESSH, node,
+          _ErrorIf(True, constants.CV_ENODESSH, node,
                    "ssh communication with node '%s': %s", a_node, a_msg)
 
     test = constants.NV_NODENETTEST not in nresult
                    "ssh communication with node '%s': %s", a_node, a_msg)
 
     test = constants.NV_NODENETTEST not in nresult
-    _ErrorIf(test, self.ENODENET, node,
+    _ErrorIf(test, constants.CV_ENODENET, node,
              "node hasn't returned node tcp connectivity data")
     if not test:
       if nresult[constants.NV_NODENETTEST]:
         nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
         for anode in nlist:
              "node hasn't returned node tcp connectivity data")
     if not test:
       if nresult[constants.NV_NODENETTEST]:
         nlist = utils.NiceSort(nresult[constants.NV_NODENETTEST].keys())
         for anode in nlist:
-          _ErrorIf(True, self.ENODENET, node,
+          _ErrorIf(True, constants.CV_ENODENET, node,
                    "tcp communication with node '%s': %s",
                    anode, nresult[constants.NV_NODENETTEST][anode])
 
     test = constants.NV_MASTERIP not in nresult
                    "tcp communication with node '%s': %s",
                    anode, nresult[constants.NV_NODENETTEST][anode])
 
     test = constants.NV_MASTERIP not in nresult
-    _ErrorIf(test, self.ENODENET, node,
+    _ErrorIf(test, constants.CV_ENODENET, node,
              "node hasn't returned node master IP reachability data")
     if not test:
       if not nresult[constants.NV_MASTERIP]:
              "node hasn't returned node master IP reachability data")
     if not test:
       if not nresult[constants.NV_MASTERIP]:
@@ -1838,7 +2133,35 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
           msg = "the master node cannot reach the master IP (not configured?)"
         else:
           msg = "cannot reach the master IP"
           msg = "the master node cannot reach the master IP (not configured?)"
         else:
           msg = "cannot reach the master IP"
-        _ErrorIf(True, self.ENODENET, node, msg)
+        _ErrorIf(True, constants.CV_ENODENET, node, msg)
+
+  def _VerifyInstancePolicy(self, instance):
+    """Verify instance specs against instance policy set on node group level.
+
+
+    """
+    cluster = self.cfg.GetClusterInfo()
+    full_beparams = cluster.FillBE(instance)
+    ipolicy = cluster.SimpleFillIPolicy(self.group_info.ipolicy)
+
+    mem_size = full_beparams.get(constants.BE_MAXMEM, None)
+    cpu_count = full_beparams.get(constants.BE_VCPUS, None)
+    disk_count = len(instance.disks)
+    disk_sizes = [disk.size for disk in instance.disks]
+    nic_count = len(instance.nics)
+
+    test_settings = [
+      (constants.ISPEC_MEM_SIZE, mem_size),
+      (constants.ISPEC_CPU_COUNT, cpu_count),
+      (constants.ISPEC_DISK_COUNT, disk_count),
+      (constants.ISPEC_NIC_COUNT, nic_count),
+      ] + map((lambda d: (constants.ISPEC_DISK_SIZE, d)), disk_sizes)
+
+    for (name, value) in test_settings:
+      test_result = _CheckMinMaxSpecs(name, ipolicy, value)
+      self._ErrorIf(test_result is not None,
+                    constants.CV_EINSTANCEPOLICY, instance.name,
+                    test_result)
 
   def _VerifyInstance(self, instance, instanceconfig, node_image,
                       diskstatus):
 
   def _VerifyInstance(self, instance, instanceconfig, node_image,
                       diskstatus):
@@ -1848,12 +2171,14 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     available on the instance's node.
 
     """
     available on the instance's node.
 
     """
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
     node_current = instanceconfig.primary_node
 
     node_vol_should = {}
     instanceconfig.MapLVsByNode(node_vol_should)
 
     node_current = instanceconfig.primary_node
 
     node_vol_should = {}
     instanceconfig.MapLVsByNode(node_vol_should)
 
+    self._VerifyInstancePolicy(instanceconfig)
+
     for node in node_vol_should:
       n_img = node_image[node]
       if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
     for node in node_vol_should:
       n_img = node_image[node]
       if n_img.offline or n_img.rpc_fail or n_img.lvm_fail:
@@ -1861,13 +2186,13 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         continue
       for volume in node_vol_should[node]:
         test = volume not in n_img.volumes
         continue
       for volume in node_vol_should[node]:
         test = volume not in n_img.volumes
-        _ErrorIf(test, self.EINSTANCEMISSINGDISK, instance,
+        _ErrorIf(test, constants.CV_EINSTANCEMISSINGDISK, instance,
                  "volume %s missing on node %s", volume, node)
 
                  "volume %s missing on node %s", volume, node)
 
-    if instanceconfig.admin_up:
+    if instanceconfig.admin_state == constants.ADMINST_UP:
       pri_img = node_image[node_current]
       test = instance not in pri_img.instances and not pri_img.offline
       pri_img = node_image[node_current]
       test = instance not in pri_img.instances and not pri_img.offline
-      _ErrorIf(test, self.EINSTANCEDOWN, instance,
+      _ErrorIf(test, constants.CV_EINSTANCEDOWN, instance,
                "instance not running on its primary node %s",
                node_current)
 
                "instance not running on its primary node %s",
                node_current)
 
@@ -1880,13 +2205,14 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       # node here
       snode = node_image[nname]
       bad_snode = snode.ghost or snode.offline
       # node here
       snode = node_image[nname]
       bad_snode = snode.ghost or snode.offline
-      _ErrorIf(instanceconfig.admin_up and not success and not bad_snode,
-               self.EINSTANCEFAULTYDISK, instance,
+      _ErrorIf(instanceconfig.admin_state == constants.ADMINST_UP and
+               not success and not bad_snode,
+               constants.CV_EINSTANCEFAULTYDISK, instance,
                "couldn't retrieve status for disk/%s on %s: %s",
                idx, nname, bdev_status)
                "couldn't retrieve status for disk/%s on %s: %s",
                idx, nname, bdev_status)
-      _ErrorIf((instanceconfig.admin_up and success and
-                bdev_status.ldisk_status == constants.LDS_FAULTY),
-               self.EINSTANCEFAULTYDISK, instance,
+      _ErrorIf((instanceconfig.admin_state == constants.ADMINST_UP and
+                success and bdev_status.ldisk_status == constants.LDS_FAULTY),
+               constants.CV_EINSTANCEFAULTYDISK, instance,
                "disk/%s on %s is faulty", idx, nname)
 
   def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
                "disk/%s on %s is faulty", idx, nname)
 
   def _VerifyOrphanVolumes(self, node_vol_should, node_image, reserved):
@@ -1907,7 +2233,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         test = ((node not in node_vol_should or
                 volume not in node_vol_should[node]) and
                 not reserved.Matches(volume))
         test = ((node not in node_vol_should or
                 volume not in node_vol_should[node]) and
                 not reserved.Matches(volume))
-        self._ErrorIf(test, self.ENODEORPHANLV, node,
+        self._ErrorIf(test, constants.CV_ENODEORPHANLV, node,
                       "volume %s is unknown", volume)
 
   def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
                       "volume %s is unknown", volume)
 
   def _VerifyNPlusOneMemory(self, node_image, instance_cfg):
@@ -1933,21 +2259,23 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         # we already list instances living on such nodes, and that's
         # enough warning
         continue
         # we already list instances living on such nodes, and that's
         # enough warning
         continue
+      #TODO(dynmem): use MINMEM for checking
+      #TODO(dynmem): also consider ballooning out other instances
       for prinode, instances in n_img.sbp.items():
         needed_mem = 0
         for instance in instances:
           bep = cluster_info.FillBE(instance_cfg[instance])
           if bep[constants.BE_AUTO_BALANCE]:
       for prinode, instances in n_img.sbp.items():
         needed_mem = 0
         for instance in instances:
           bep = cluster_info.FillBE(instance_cfg[instance])
           if bep[constants.BE_AUTO_BALANCE]:
-            needed_mem += bep[constants.BE_MEMORY]
+            needed_mem += bep[constants.BE_MAXMEM]
         test = n_img.mfree < needed_mem
         test = n_img.mfree < needed_mem
-        self._ErrorIf(test, self.ENODEN1, node,
+        self._ErrorIf(test, constants.CV_ENODEN1, node,
                       "not enough memory to accomodate instance failovers"
                       " should node %s fail (%dMiB needed, %dMiB available)",
                       prinode, needed_mem, n_img.mfree)
 
   @classmethod
   def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
                       "not enough memory to accomodate instance failovers"
                       " should node %s fail (%dMiB needed, %dMiB available)",
                       prinode, needed_mem, n_img.mfree)
 
   @classmethod
   def _VerifyFiles(cls, errorif, nodeinfo, master_node, all_nvinfo,
-                   (files_all, files_all_opt, files_mc, files_vm)):
+                   (files_all, files_opt, files_mc, files_vm)):
     """Verifies file checksums collected from all nodes.
 
     @param errorif: Callback for reporting errors
     """Verifies file checksums collected from all nodes.
 
     @param errorif: Callback for reporting errors
@@ -1956,25 +2284,35 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     @param all_nvinfo: RPC results
 
     """
     @param all_nvinfo: RPC results
 
     """
-    node_names = frozenset(node.name for node in nodeinfo)
+    # Define functions determining which nodes to consider for a file
+    files2nodefn = [
+      (files_all, None),
+      (files_mc, lambda node: (node.master_candidate or
+                               node.name == master_node)),
+      (files_vm, lambda node: node.vm_capable),
+      ]
 
 
-    assert master_node in node_names
-    assert (len(files_all | files_all_opt | files_mc | files_vm) ==
-            sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
-           "Found file listed in more than one file list"
+    # Build mapping from filename to list of nodes which should have the file
+    nodefiles = {}
+    for (files, fn) in files2nodefn:
+      if fn is None:
+        filenodes = nodeinfo
+      else:
+        filenodes = filter(fn, nodeinfo)
+      nodefiles.update((filename,
+                        frozenset(map(operator.attrgetter("name"), filenodes)))
+                       for filename in files)
 
 
-    # Define functions determining which nodes to consider for a file
-    file2nodefn = dict([(filename, fn)
-      for (files, fn) in [(files_all, None),
-                          (files_all_opt, None),
-                          (files_mc, lambda node: (node.master_candidate or
-                                                   node.name == master_node)),
-                          (files_vm, lambda node: node.vm_capable)]
-      for filename in files])
+    assert set(nodefiles) == (files_all | files_mc | files_vm)
 
 
-    fileinfo = dict((filename, {}) for filename in file2nodefn.keys())
+    fileinfo = dict((filename, {}) for filename in nodefiles)
+    ignore_nodes = set()
 
     for node in nodeinfo:
 
     for node in nodeinfo:
+      if node.offline:
+        ignore_nodes.add(node.name)
+        continue
+
       nresult = all_nvinfo[node.name]
 
       if nresult.fail_msg or not nresult.payload:
       nresult = all_nvinfo[node.name]
 
       if nresult.fail_msg or not nresult.payload:
@@ -1983,16 +2321,16 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         node_files = nresult.payload.get(constants.NV_FILELIST, None)
 
       test = not (node_files and isinstance(node_files, dict))
         node_files = nresult.payload.get(constants.NV_FILELIST, None)
 
       test = not (node_files and isinstance(node_files, dict))
-      errorif(test, cls.ENODEFILECHECK, node.name,
+      errorif(test, constants.CV_ENODEFILECHECK, node.name,
               "Node did not return file checksum data")
       if test:
               "Node did not return file checksum data")
       if test:
+        ignore_nodes.add(node.name)
         continue
 
         continue
 
+      # Build per-checksum mapping from filename to nodes having it
       for (filename, checksum) in node_files.items():
       for (filename, checksum) in node_files.items():
-        # Check if the file should be considered for a node
-        fn = file2nodefn[filename]
-        if fn is None or fn(node):
-          fileinfo[filename].setdefault(checksum, set()).add(node.name)
+        assert filename in nodefiles
+        fileinfo[filename].setdefault(checksum, set()).add(node.name)
 
     for (filename, checksums) in fileinfo.items():
       assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
 
     for (filename, checksums) in fileinfo.items():
       assert compat.all(len(i) > 10 for i in checksums), "Invalid checksum"
@@ -2000,23 +2338,32 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       # Nodes having the file
       with_file = frozenset(node_name
                             for nodes in fileinfo[filename].values()
       # Nodes having the file
       with_file = frozenset(node_name
                             for nodes in fileinfo[filename].values()
-                            for node_name in nodes)
+                            for node_name in nodes) - ignore_nodes
+
+      expected_nodes = nodefiles[filename] - ignore_nodes
 
       # Nodes missing file
 
       # Nodes missing file
-      missing_file = node_names - with_file
+      missing_file = expected_nodes - with_file
 
 
-      if filename in files_all_opt:
+      if filename in files_opt:
         # All or no nodes
         # All or no nodes
-        errorif(missing_file and missing_file != node_names,
-                cls.ECLUSTERFILECHECK, None,
-                "File %s is optional, but it must exist on all or no nodes (not"
-                " found on %s)",
+        errorif(missing_file and missing_file != expected_nodes,
+                constants.CV_ECLUSTERFILECHECK, None,
+                "File %s is optional, but it must exist on all or no"
+                " nodes (not found on %s)",
                 filename, utils.CommaJoin(utils.NiceSort(missing_file)))
       else:
                 filename, utils.CommaJoin(utils.NiceSort(missing_file)))
       else:
-        errorif(missing_file, cls.ECLUSTERFILECHECK, None,
+        errorif(missing_file, constants.CV_ECLUSTERFILECHECK, None,
                 "File %s is missing from node(s) %s", filename,
                 utils.CommaJoin(utils.NiceSort(missing_file)))
 
                 "File %s is missing from node(s) %s", filename,
                 utils.CommaJoin(utils.NiceSort(missing_file)))
 
+        # Warn if a node has a file it shouldn't
+        unexpected = with_file - expected_nodes
+        errorif(unexpected,
+                constants.CV_ECLUSTERFILECHECK, None,
+                "File %s should not exist on node(s) %s",
+                filename, utils.CommaJoin(utils.NiceSort(unexpected)))
+
       # See if there are multiple versions of the file
       test = len(checksums) > 1
       if test:
       # See if there are multiple versions of the file
       test = len(checksums) > 1
       if test:
@@ -2027,7 +2374,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       else:
         variants = []
 
       else:
         variants = []
 
-      errorif(test, cls.ECLUSTERFILECHECK, None,
+      errorif(test, constants.CV_ECLUSTERFILECHECK, None,
               "File %s found with %s different checksums (%s)",
               filename, len(checksums), "; ".join(variants))
 
               "File %s found with %s different checksums (%s)",
               filename, len(checksums), "; ".join(variants))
 
@@ -2045,27 +2392,27 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     if drbd_helper:
       helper_result = nresult.get(constants.NV_DRBDHELPER, None)
       test = (helper_result == None)
 
     if drbd_helper:
       helper_result = nresult.get(constants.NV_DRBDHELPER, None)
       test = (helper_result == None)
-      _ErrorIf(test, self.ENODEDRBDHELPER, node,
+      _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
                "no drbd usermode helper returned")
       if helper_result:
         status, payload = helper_result
         test = not status
                "no drbd usermode helper returned")
       if helper_result:
         status, payload = helper_result
         test = not status
-        _ErrorIf(test, self.ENODEDRBDHELPER, node,
+        _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
                  "drbd usermode helper check unsuccessful: %s", payload)
         test = status and (payload != drbd_helper)
                  "drbd usermode helper check unsuccessful: %s", payload)
         test = status and (payload != drbd_helper)
-        _ErrorIf(test, self.ENODEDRBDHELPER, node,
+        _ErrorIf(test, constants.CV_ENODEDRBDHELPER, node,
                  "wrong drbd usermode helper: %s", payload)
 
     # compute the DRBD minors
     node_drbd = {}
     for minor, instance in drbd_map[node].items():
       test = instance not in instanceinfo
                  "wrong drbd usermode helper: %s", payload)
 
     # compute the DRBD minors
     node_drbd = {}
     for minor, instance in drbd_map[node].items():
       test = instance not in instanceinfo
-      _ErrorIf(test, self.ECLUSTERCFG, None,
+      _ErrorIf(test, constants.CV_ECLUSTERCFG, None,
                "ghost instance '%s' in temporary DRBD map", instance)
         # ghost instance should not be running, but otherwise we
         # don't give double warnings (both ghost instance and
                "ghost instance '%s' in temporary DRBD map", instance)
         # ghost instance should not be running, but otherwise we
         # don't give double warnings (both ghost instance and
@@ -2074,12 +2421,13 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         node_drbd[minor] = (instance, False)
       else:
         instance = instanceinfo[instance]
         node_drbd[minor] = (instance, False)
       else:
         instance = instanceinfo[instance]
-        node_drbd[minor] = (instance.name, instance.admin_up)
+        node_drbd[minor] = (instance.name,
+                            instance.admin_state == constants.ADMINST_UP)
 
     # and now check them
     used_minors = nresult.get(constants.NV_DRBDLIST, [])
     test = not isinstance(used_minors, (tuple, list))
 
     # and now check them
     used_minors = nresult.get(constants.NV_DRBDLIST, [])
     test = not isinstance(used_minors, (tuple, list))
-    _ErrorIf(test, self.ENODEDRBD, node,
+    _ErrorIf(test, constants.CV_ENODEDRBD, node,
              "cannot parse drbd status file: %s", str(used_minors))
     if test:
       # we cannot check drbd status
              "cannot parse drbd status file: %s", str(used_minors))
     if test:
       # we cannot check drbd status
@@ -2087,11 +2435,11 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     for minor, (iname, must_exist) in node_drbd.items():
       test = minor not in used_minors and must_exist
 
     for minor, (iname, must_exist) in node_drbd.items():
       test = minor not in used_minors and must_exist
-      _ErrorIf(test, self.ENODEDRBD, node,
+      _ErrorIf(test, constants.CV_ENODEDRBD, node,
                "drbd minor %d of instance %s is not active", minor, iname)
     for minor in used_minors:
       test = minor not in node_drbd
                "drbd minor %d of instance %s is not active", minor, iname)
     for minor in used_minors:
       test = minor not in node_drbd
-      _ErrorIf(test, self.ENODEDRBD, node,
+      _ErrorIf(test, constants.CV_ENODEDRBD, node,
                "unallocated drbd minor %d is in use", minor)
 
   def _UpdateNodeOS(self, ninfo, nresult, nimg):
                "unallocated drbd minor %d is in use", minor)
 
   def _UpdateNodeOS(self, ninfo, nresult, nimg):
@@ -2104,14 +2452,14 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     remote_os = nresult.get(constants.NV_OSLIST, None)
     test = (not isinstance(remote_os, list) or
             not compat.all(isinstance(v, list) and len(v) == 7
                            for v in remote_os))
 
 
     remote_os = nresult.get(constants.NV_OSLIST, None)
     test = (not isinstance(remote_os, list) or
             not compat.all(isinstance(v, list) and len(v) == 7
                            for v in remote_os))
 
-    _ErrorIf(test, self.ENODEOS, node,
+    _ErrorIf(test, constants.CV_ENODEOS, node,
              "node hasn't returned valid OS data")
 
     nimg.os_fail = test
              "node hasn't returned valid OS data")
 
     nimg.os_fail = test
@@ -2145,7 +2493,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
 
 
     assert not nimg.os_fail, "Entered _VerifyNodeOS with failed OS rpc?"
 
@@ -2153,19 +2501,14 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     for os_name, os_data in nimg.oslist.items():
       assert os_data, "Empty OS status for OS %s?!" % os_name
       f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
     for os_name, os_data in nimg.oslist.items():
       assert os_data, "Empty OS status for OS %s?!" % os_name
       f_path, f_status, f_diag, f_var, f_param, f_api = os_data[0]
-      _ErrorIf(not f_status, self.ENODEOS, node,
+      _ErrorIf(not f_status, constants.CV_ENODEOS, node,
                "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
                "Invalid OS %s (located at %s): %s", os_name, f_path, f_diag)
-      _ErrorIf(len(os_data) > 1, self.ENODEOS, node,
+      _ErrorIf(len(os_data) > 1, constants.CV_ENODEOS, node,
                "OS '%s' has multiple entries (first one shadows the rest): %s",
                os_name, utils.CommaJoin([v[0] for v in os_data]))
                "OS '%s' has multiple entries (first one shadows the rest): %s",
                os_name, utils.CommaJoin([v[0] for v in os_data]))
-      # this will catched in backend too
-      _ErrorIf(compat.any(v >= constants.OS_API_V15 for v in f_api)
-               and not f_var, self.ENODEOS, node,
-               "OS %s with API at least %d does not declare any variant",
-               os_name, constants.OS_API_V15)
       # comparisons with the 'base' image
       test = os_name not in base.oslist
       # comparisons with the 'base' image
       test = os_name not in base.oslist
-      _ErrorIf(test, self.ENODEOS, node,
+      _ErrorIf(test, constants.CV_ENODEOS, node,
                "Extra OS %s not present on reference node (%s)",
                os_name, base.name)
       if test:
                "Extra OS %s not present on reference node (%s)",
                os_name, base.name)
       if test:
@@ -2179,14 +2522,14 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
                          ("variants list", f_var, b_var),
                          ("parameters", beautify_params(f_param),
                           beautify_params(b_param))]:
                          ("variants list", f_var, b_var),
                          ("parameters", beautify_params(f_param),
                           beautify_params(b_param))]:
-        _ErrorIf(a != b, self.ENODEOS, node,
+        _ErrorIf(a != b, constants.CV_ENODEOS, node,
                  "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
                  kind, os_name, base.name,
                  utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
 
     # check any missing OSes
     missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
                  "OS %s for %s differs from reference node %s: [%s] vs. [%s]",
                  kind, os_name, base.name,
                  utils.CommaJoin(sorted(a)), utils.CommaJoin(sorted(b)))
 
     # check any missing OSes
     missing = set(base.oslist.keys()).difference(nimg.oslist.keys())
-    _ErrorIf(missing, self.ENODEOS, node,
+    _ErrorIf(missing, constants.CV_ENODEOS, node,
              "OSes present on reference node %s but missing on this node: %s",
              base.name, utils.CommaJoin(missing))
 
              "OSes present on reference node %s but missing on this node: %s",
              base.name, utils.CommaJoin(missing))
 
@@ -2204,7 +2547,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     if ((ninfo.master_candidate or ninfo.master_capable) and
         constants.NV_OOB_PATHS in nresult):
       for path_result in nresult[constants.NV_OOB_PATHS]:
     if ((ninfo.master_candidate or ninfo.master_capable) and
         constants.NV_OOB_PATHS in nresult):
       for path_result in nresult[constants.NV_OOB_PATHS]:
-        self._ErrorIf(path_result, self.ENODEOOBPATH, node, path_result)
+        self._ErrorIf(path_result, constants.CV_ENODEOOBPATH, node, path_result)
 
   def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
     """Verifies and updates the node volume data.
 
   def _UpdateNodeVolumes(self, ninfo, nresult, nimg, vg_name):
     """Verifies and updates the node volume data.
@@ -2220,17 +2563,18 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     nimg.lvm_fail = True
     lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
     if vg_name is None:
       pass
     elif isinstance(lvdata, basestring):
 
     nimg.lvm_fail = True
     lvdata = nresult.get(constants.NV_LVLIST, "Missing LV data")
     if vg_name is None:
       pass
     elif isinstance(lvdata, basestring):
-      _ErrorIf(True, self.ENODELVM, node, "LVM problem on node: %s",
+      _ErrorIf(True, constants.CV_ENODELVM, node, "LVM problem on node: %s",
                utils.SafeEncode(lvdata))
     elif not isinstance(lvdata, dict):
                utils.SafeEncode(lvdata))
     elif not isinstance(lvdata, dict):
-      _ErrorIf(True, self.ENODELVM, node, "rpc call to node failed (lvlist)")
+      _ErrorIf(True, constants.CV_ENODELVM, node,
+               "rpc call to node failed (lvlist)")
     else:
       nimg.volumes = lvdata
       nimg.lvm_fail = False
     else:
       nimg.volumes = lvdata
       nimg.lvm_fail = False
@@ -2250,8 +2594,9 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     """
     idata = nresult.get(constants.NV_INSTANCELIST, None)
     test = not isinstance(idata, list)
     """
     idata = nresult.get(constants.NV_INSTANCELIST, None)
     test = not isinstance(idata, list)
-    self._ErrorIf(test, self.ENODEHV, ninfo.name, "rpc call to node failed"
-                  " (instancelist): %s", utils.SafeEncode(str(idata)))
+    self._ErrorIf(test, constants.CV_ENODEHV, ninfo.name,
+                  "rpc call to node failed (instancelist): %s",
+                  utils.SafeEncode(str(idata)))
     if test:
       nimg.hyp_fail = True
     else:
     if test:
       nimg.hyp_fail = True
     else:
@@ -2268,31 +2613,32 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     node = ninfo.name
 
     """
     node = ninfo.name
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     # try to read free memory (from the hypervisor)
     hv_info = nresult.get(constants.NV_HVINFO, None)
     test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
 
     # try to read free memory (from the hypervisor)
     hv_info = nresult.get(constants.NV_HVINFO, None)
     test = not isinstance(hv_info, dict) or "memory_free" not in hv_info
-    _ErrorIf(test, self.ENODEHV, node, "rpc call to node failed (hvinfo)")
+    _ErrorIf(test, constants.CV_ENODEHV, node,
+             "rpc call to node failed (hvinfo)")
     if not test:
       try:
         nimg.mfree = int(hv_info["memory_free"])
       except (ValueError, TypeError):
     if not test:
       try:
         nimg.mfree = int(hv_info["memory_free"])
       except (ValueError, TypeError):
-        _ErrorIf(True, self.ENODERPC, node,
+        _ErrorIf(True, constants.CV_ENODERPC, node,
                  "node returned invalid nodeinfo, check hypervisor")
 
     # FIXME: devise a free space model for file based instances as well
     if vg_name is not None:
       test = (constants.NV_VGLIST not in nresult or
               vg_name not in nresult[constants.NV_VGLIST])
                  "node returned invalid nodeinfo, check hypervisor")
 
     # FIXME: devise a free space model for file based instances as well
     if vg_name is not None:
       test = (constants.NV_VGLIST not in nresult or
               vg_name not in nresult[constants.NV_VGLIST])
-      _ErrorIf(test, self.ENODELVM, node,
+      _ErrorIf(test, constants.CV_ENODELVM, node,
                "node didn't return data for the volume group '%s'"
                " - it is either missing or broken", vg_name)
       if not test:
         try:
           nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
         except (ValueError, TypeError):
                "node didn't return data for the volume group '%s'"
                " - it is either missing or broken", vg_name)
       if not test:
         try:
           nimg.dfree = int(nresult[constants.NV_VGLIST][vg_name])
         except (ValueError, TypeError):
-          _ErrorIf(True, self.ENODERPC, node,
+          _ErrorIf(True, constants.CV_ENODERPC, node,
                    "node returned invalid LVM info, check LVM status")
 
   def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
                    "node returned invalid LVM info, check LVM status")
 
   def _CollectDiskInfo(self, nodelist, node_image, instanceinfo):
@@ -2310,7 +2656,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         list of tuples (success, payload)
 
     """
         list of tuples (success, payload)
 
     """
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
 
     node_disks = {}
     node_disks_devonly = {}
 
     node_disks = {}
     node_disks_devonly = {}
@@ -2359,7 +2705,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         data = len(disks) * [(False, "node offline")]
       else:
         msg = nres.fail_msg
         data = len(disks) * [(False, "node offline")]
       else:
         msg = nres.fail_msg
-        _ErrorIf(msg, self.ENODERPC, nname,
+        _ErrorIf(msg, constants.CV_ENODERPC, nname,
                  "while getting disk information: %s", msg)
         if msg:
           # No data from this node
                  "while getting disk information: %s", msg)
         if msg:
           # No data from this node
@@ -2392,6 +2738,40 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     return instdisk
 
 
     return instdisk
 
+  @staticmethod
+  def _SshNodeSelector(group_uuid, all_nodes):
+    """Create endless iterators for all potential SSH check hosts.
+
+    """
+    nodes = [node for node in all_nodes
+             if (node.group != group_uuid and
+                 not node.offline)]
+    keyfunc = operator.attrgetter("group")
+
+    return map(itertools.cycle,
+               [sorted(map(operator.attrgetter("name"), names))
+                for _, names in itertools.groupby(sorted(nodes, key=keyfunc),
+                                                  keyfunc)])
+
+  @classmethod
+  def _SelectSshCheckNodes(cls, group_nodes, group_uuid, all_nodes):
+    """Choose which nodes should talk to which other nodes.
+
+    We will make nodes contact all nodes in their group, and one node from
+    every other group.
+
+    @warning: This algorithm has a known issue if one node group is much
+      smaller than others (e.g. just one node). In such a case all other
+      nodes will talk to the single node.
+
+    """
+    online_nodes = sorted(node.name for node in group_nodes if not node.offline)
+    sel = cls._SshNodeSelector(group_uuid, all_nodes)
+
+    return (online_nodes,
+            dict((name, sorted([i.next() for i in sel]))
+                 for name in online_nodes))
+
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
@@ -2412,17 +2792,22 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     """Build hooks nodes.
 
     """
     """Build hooks nodes.
 
     """
-    assert self.my_node_names, ("Node list not gathered,"
-      " has CheckPrereq been executed?")
     return ([], self.my_node_names)
 
   def Exec(self, feedback_fn):
     """Verify integrity of the node group, performing various test on nodes.
 
     """
     return ([], self.my_node_names)
 
   def Exec(self, feedback_fn):
     """Verify integrity of the node group, performing various test on nodes.
 
     """
-    # This method has too many local variables. pylint: disable-msg=R0914
+    # This method has too many local variables. pylint: disable=R0914
+    feedback_fn("* Verifying group '%s'" % self.group_info.name)
+
+    if not self.my_node_names:
+      # empty node group
+      feedback_fn("* Empty node group, skipping verification")
+      return True
+
     self.bad = False
     self.bad = False
-    _ErrorIf = self._ErrorIf # pylint: disable-msg=C0103
+    _ErrorIf = self._ErrorIf # pylint: disable=C0103
     verbose = self.op.verbose
     self._feedback_fn = feedback_fn
 
     verbose = self.op.verbose
     self._feedback_fn = feedback_fn
 
@@ -2435,6 +2820,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     i_non_redundant = [] # Non redundant instances
     i_non_a_balanced = [] # Non auto-balanced instances
 
     i_non_redundant = [] # Non redundant instances
     i_non_a_balanced = [] # Non auto-balanced instances
+    i_offline = 0 # Count of offline instances
     n_offline = 0 # Count of offline nodes
     n_drained = 0 # Count of nodes being drained
     node_vol_should = {}
     n_offline = 0 # Count of offline nodes
     n_drained = 0 # Count of nodes being drained
     node_vol_should = {}
@@ -2450,25 +2836,18 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
 
 
     feedback_fn("* Gathering data (%d nodes)" % len(self.my_node_names))
 
-    # We will make nodes contact all nodes in their group, and one node from
-    # every other group.
-    # TODO: should it be a *random* node, different every time?
-    online_nodes = [node.name for node in node_data_list if not node.offline]
-    other_group_nodes = {}
-
-    for name in sorted(self.all_node_info):
-      node = self.all_node_info[name]
-      if (node.group not in other_group_nodes
-          and node.group != self.group_uuid
-          and not node.offline):
-        other_group_nodes[node.group] = node.name
+    user_scripts = []
+    if self.cfg.GetUseExternalMipScript():
+      user_scripts.append(constants.EXTERNAL_MASTER_SETUP_SCRIPT)
 
     node_verify_param = {
       constants.NV_FILELIST:
         utils.UniqueSequence(filename
                              for files in filemap
                              for filename in files),
 
     node_verify_param = {
       constants.NV_FILELIST:
         utils.UniqueSequence(filename
                              for files in filemap
                              for filename in files),
-      constants.NV_NODELIST: online_nodes + other_group_nodes.values(),
+      constants.NV_NODELIST:
+        self._SelectSshCheckNodes(node_data_list, self.group_uuid,
+                                  self.all_node_info.values()),
       constants.NV_HYPERVISOR: hypervisors,
       constants.NV_HVPARAMS:
         _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
       constants.NV_HYPERVISOR: hypervisors,
       constants.NV_HVPARAMS:
         _GetAllHypervisorParameters(cluster, self.all_inst_info.values()),
@@ -2483,6 +2862,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       constants.NV_MASTERIP: (master_node, master_ip),
       constants.NV_OSLIST: None,
       constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
       constants.NV_MASTERIP: (master_node, master_ip),
       constants.NV_OSLIST: None,
       constants.NV_VMNODES: self.cfg.GetNonVmCapableNodeList(),
+      constants.NV_USERSCRIPTS: user_scripts,
       }
 
     if vg_name is not None:
       }
 
     if vg_name is not None:
@@ -2557,6 +2937,8 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
                                            node_verify_param,
                                            self.cfg.GetClusterName())
     all_nvinfo = self.rpc.call_node_verify(self.my_node_names,
                                            node_verify_param,
                                            self.cfg.GetClusterName())
+    nvinfo_endtime = time.time()
+
     if self.extra_lv_nodes and vg_name is not None:
       extra_lv_nvinfo = \
           self.rpc.call_node_verify(self.extra_lv_nodes,
     if self.extra_lv_nodes and vg_name is not None:
       extra_lv_nvinfo = \
           self.rpc.call_node_verify(self.extra_lv_nodes,
@@ -2564,7 +2946,6 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
                                     self.cfg.GetClusterName())
     else:
       extra_lv_nvinfo = {}
                                     self.cfg.GetClusterName())
     else:
       extra_lv_nvinfo = {}
-    nvinfo_endtime = time.time()
 
     all_drbd_map = self.cfg.ComputeDRBDMap()
 
 
     all_drbd_map = self.cfg.ComputeDRBDMap()
 
@@ -2629,7 +3010,8 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         feedback_fn("* Verifying node %s (%s)" % (node, ntype))
 
       msg = all_nvinfo[node].fail_msg
         feedback_fn("* Verifying node %s (%s)" % (node, ntype))
 
       msg = all_nvinfo[node].fail_msg
-      _ErrorIf(msg, self.ENODERPC, node, "while contacting node: %s", msg)
+      _ErrorIf(msg, constants.CV_ENODERPC, node, "while contacting node: %s",
+               msg)
       if msg:
         nimg.rpc_fail = True
         continue
       if msg:
         nimg.rpc_fail = True
         continue
@@ -2639,6 +3021,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       nimg.call_ok = self._VerifyNode(node_i, nresult)
       self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
       self._VerifyNodeNetwork(node_i, nresult)
       nimg.call_ok = self._VerifyNode(node_i, nresult)
       self._VerifyNodeTime(node_i, nresult, nvinfo_starttime, nvinfo_endtime)
       self._VerifyNodeNetwork(node_i, nresult)
+      self._VerifyNodeUserScripts(node_i, nresult)
       self._VerifyOob(node_i, nresult)
 
       if nimg.vm_capable:
       self._VerifyOob(node_i, nresult)
 
       if nimg.vm_capable:
@@ -2663,10 +3046,16 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         non_primary_inst = set(nimg.instances).difference(nimg.pinst)
 
         for inst in non_primary_inst:
         non_primary_inst = set(nimg.instances).difference(nimg.pinst)
 
         for inst in non_primary_inst:
+          # FIXME: investigate best way to handle offline insts
+          if inst.admin_state == constants.ADMINST_OFFLINE:
+            if verbose:
+              feedback_fn("* Skipping offline instance %s" % inst.name)
+            i_offline += 1
+            continue
           test = inst in self.all_inst_info
           test = inst in self.all_inst_info
-          _ErrorIf(test, self.EINSTANCEWRONGNODE, inst,
+          _ErrorIf(test, constants.CV_EINSTANCEWRONGNODE, inst,
                    "instance should not run on node %s", node_i.name)
                    "instance should not run on node %s", node_i.name)
-          _ErrorIf(not test, self.ENODEORPHANINSTANCE, node_i.name,
+          _ErrorIf(not test, constants.CV_ENODEORPHANINSTANCE, node_i.name,
                    "node is running unknown instance %s", inst)
 
     for node, result in extra_lv_nvinfo.items():
                    "node is running unknown instance %s", inst)
 
     for node, result in extra_lv_nvinfo.items():
@@ -2685,11 +3074,12 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       pnode = inst_config.primary_node
       pnode_img = node_image[pnode]
       _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
       pnode = inst_config.primary_node
       pnode_img = node_image[pnode]
       _ErrorIf(pnode_img.rpc_fail and not pnode_img.offline,
-               self.ENODERPC, pnode, "instance %s, connection to"
+               constants.CV_ENODERPC, pnode, "instance %s, connection to"
                " primary node failed", instance)
 
                " primary node failed", instance)
 
-      _ErrorIf(inst_config.admin_up and pnode_img.offline,
-               self.EINSTANCEBADNODE, instance,
+      _ErrorIf(inst_config.admin_state == constants.ADMINST_UP and
+               pnode_img.offline,
+               constants.CV_EINSTANCEBADNODE, instance,
                "instance is marked as running and lives on offline node %s",
                inst_config.primary_node)
 
                "instance is marked as running and lives on offline node %s",
                inst_config.primary_node)
 
@@ -2701,7 +3091,8 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       if not inst_config.secondary_nodes:
         i_non_redundant.append(instance)
 
       if not inst_config.secondary_nodes:
         i_non_redundant.append(instance)
 
-      _ErrorIf(len(inst_config.secondary_nodes) > 1, self.EINSTANCELAYOUT,
+      _ErrorIf(len(inst_config.secondary_nodes) > 1,
+               constants.CV_EINSTANCELAYOUT,
                instance, "instance has multiple secondary nodes: %s",
                utils.CommaJoin(inst_config.secondary_nodes),
                code=self.ETYPE_WARNING)
                instance, "instance has multiple secondary nodes: %s",
                utils.CommaJoin(inst_config.secondary_nodes),
                code=self.ETYPE_WARNING)
@@ -2722,7 +3113,8 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
                                      key=lambda (_, nodes): pnode in nodes,
                                      reverse=True)]
 
                                      key=lambda (_, nodes): pnode in nodes,
                                      reverse=True)]
 
-        self._ErrorIf(len(instance_groups) > 1, self.EINSTANCESPLITGROUPS,
+        self._ErrorIf(len(instance_groups) > 1,
+                      constants.CV_EINSTANCESPLITGROUPS,
                       instance, "instance has primary and secondary nodes in"
                       " different groups: %s", utils.CommaJoin(pretty_list),
                       code=self.ETYPE_WARNING)
                       instance, "instance has primary and secondary nodes in"
                       " different groups: %s", utils.CommaJoin(pretty_list),
                       code=self.ETYPE_WARNING)
@@ -2732,21 +3124,22 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
       for snode in inst_config.secondary_nodes:
         s_img = node_image[snode]
 
       for snode in inst_config.secondary_nodes:
         s_img = node_image[snode]
-        _ErrorIf(s_img.rpc_fail and not s_img.offline, self.ENODERPC, snode,
-                 "instance %s, connection to secondary node failed", instance)
+        _ErrorIf(s_img.rpc_fail and not s_img.offline, constants.CV_ENODERPC,
+                 snode, "instance %s, connection to secondary node failed",
+                 instance)
 
         if s_img.offline:
           inst_nodes_offline.append(snode)
 
       # warn that the instance lives on offline nodes
 
         if s_img.offline:
           inst_nodes_offline.append(snode)
 
       # warn that the instance lives on offline nodes
-      _ErrorIf(inst_nodes_offline, self.EINSTANCEBADNODE, instance,
+      _ErrorIf(inst_nodes_offline, constants.CV_EINSTANCEBADNODE, instance,
                "instance has offline secondary node(s) %s",
                utils.CommaJoin(inst_nodes_offline))
       # ... or ghost/non-vm_capable nodes
       for node in inst_config.all_nodes:
                "instance has offline secondary node(s) %s",
                utils.CommaJoin(inst_nodes_offline))
       # ... or ghost/non-vm_capable nodes
       for node in inst_config.all_nodes:
-        _ErrorIf(node_image[node].ghost, self.EINSTANCEBADNODE, instance,
-                 "instance lives on ghost node %s", node)
-        _ErrorIf(not node_image[node].vm_capable, self.EINSTANCEBADNODE,
+        _ErrorIf(node_image[node].ghost, constants.CV_EINSTANCEBADNODE,
+                 instance, "instance lives on ghost node %s", node)
+        _ErrorIf(not node_image[node].vm_capable, constants.CV_EINSTANCEBADNODE,
                  instance, "instance lives on non-vm_capable node %s", node)
 
     feedback_fn("* Verifying orphan volumes")
                  instance, "instance lives on non-vm_capable node %s", node)
 
     feedback_fn("* Verifying orphan volumes")
@@ -2777,6 +3170,9 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
                   % len(i_non_a_balanced))
 
       feedback_fn("  - NOTICE: %d non-auto-balanced instance(s) found."
                   % len(i_non_a_balanced))
 
+    if i_offline:
+      feedback_fn("  - NOTICE: %d offline instance(s) found." % i_offline)
+
     if n_offline:
       feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
 
     if n_offline:
       feedback_fn("  - NOTICE: %d offline node(s) found." % n_offline)
 
@@ -2800,9 +3196,12 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         and hook results
 
     """
         and hook results
 
     """
-    # We only really run POST phase hooks, and are only interested in
-    # their results
-    if phase == constants.HOOKS_PHASE_POST:
+    # We only really run POST phase hooks, only for non-empty groups,
+    # and are only interested in their results
+    if not self.my_node_names:
+      # empty node group
+      pass
+    elif phase == constants.HOOKS_PHASE_POST:
       # Used to change hooks' output to proper indentation
       feedback_fn("* Hooks Results")
       assert hooks_results, "invalid result from hooks"
       # Used to change hooks' output to proper indentation
       feedback_fn("* Hooks Results")
       assert hooks_results, "invalid result from hooks"
@@ -2811,24 +3210,22 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
         res = hooks_results[node_name]
         msg = res.fail_msg
         test = msg and not res.offline
         res = hooks_results[node_name]
         msg = res.fail_msg
         test = msg and not res.offline
-        self._ErrorIf(test, self.ENODEHOOKS, node_name,
+        self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
                       "Communication failure in hooks execution: %s", msg)
         if res.offline or msg:
                       "Communication failure in hooks execution: %s", msg)
         if res.offline or msg:
-          # No need to investigate payload if node is offline or gave an error.
-          # override manually lu_result here as _ErrorIf only
-          # overrides self.bad
-          lu_result = 1
+          # No need to investigate payload if node is offline or gave
+          # an error.
           continue
         for script, hkr, output in res.payload:
           test = hkr == constants.HKR_FAIL
           continue
         for script, hkr, output in res.payload:
           test = hkr == constants.HKR_FAIL
-          self._ErrorIf(test, self.ENODEHOOKS, node_name,
+          self._ErrorIf(test, constants.CV_ENODEHOOKS, node_name,
                         "Script %s failed, output:", script)
           if test:
                         "Script %s failed, output:", script)
           if test:
-            output = self._HOOKS_INDENT_RE.sub('      ', output)
+            output = self._HOOKS_INDENT_RE.sub("      ", output)
             feedback_fn("%s" % output)
             feedback_fn("%s" % output)
-            lu_result = 0
+            lu_result = False
 
 
-      return lu_result
+    return lu_result
 
 
 class LUClusterVerifyDisks(NoHooksLU):
 
 
 class LUClusterVerifyDisks(NoHooksLU):
@@ -2838,11 +3235,91 @@ class LUClusterVerifyDisks(NoHooksLU):
   REQ_BGL = False
 
   def ExpandNames(self):
   REQ_BGL = False
 
   def ExpandNames(self):
+    self.share_locks = _ShareAll()
     self.needed_locks = {
     self.needed_locks = {
-      locking.LEVEL_NODE: locking.ALL_SET,
-      locking.LEVEL_INSTANCE: locking.ALL_SET,
-    }
-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+      locking.LEVEL_NODEGROUP: locking.ALL_SET,
+      }
+
+  def Exec(self, feedback_fn):
+    group_names = self.owned_locks(locking.LEVEL_NODEGROUP)
+
+    # Submit one instance of L{opcodes.OpGroupVerifyDisks} per node group
+    return ResultWithJobs([[opcodes.OpGroupVerifyDisks(group_name=group)]
+                           for group in group_names])
+
+
+class LUGroupVerifyDisks(NoHooksLU):
+  """Verifies the status of all disks in a node group.
+
+  """
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    # Raises errors.OpPrereqError on its own if group can't be found
+    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+    self.share_locks = _ShareAll()
+    self.needed_locks = {
+      locking.LEVEL_INSTANCE: [],
+      locking.LEVEL_NODEGROUP: [],
+      locking.LEVEL_NODE: [],
+      }
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_INSTANCE:
+      assert not self.needed_locks[locking.LEVEL_INSTANCE]
+
+      # Lock instances optimistically, needs verification once node and group
+      # locks have been acquired
+      self.needed_locks[locking.LEVEL_INSTANCE] = \
+        self.cfg.GetNodeGroupInstances(self.group_uuid)
+
+    elif level == locking.LEVEL_NODEGROUP:
+      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+      self.needed_locks[locking.LEVEL_NODEGROUP] = \
+        set([self.group_uuid] +
+            # Lock all groups used by instances optimistically; this requires
+            # going via the node before it's locked, requiring verification
+            # later on
+            [group_uuid
+             for instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
+             for group_uuid in self.cfg.GetInstanceNodeGroups(instance_name)])
+
+    elif level == locking.LEVEL_NODE:
+      # This will only lock the nodes in the group to be verified which contain
+      # actual instances
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+      self._LockInstancesNodes()
+
+      # Lock all nodes in group to be verified
+      assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+      member_nodes = self.cfg.GetNodeGroup(self.group_uuid).members
+      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
+
+  def CheckPrereq(self):
+    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
+
+    assert self.group_uuid in owned_groups
+
+    # Check if locked instances are still correct
+    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+    # Get instance information
+    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
+
+    # Check if node groups for locked instances are still correct
+    for (instance_name, inst) in self.instances.items():
+      assert owned_nodes.issuperset(inst.all_nodes), \
+        "Instance %s's nodes changed while we kept the lock" % instance_name
+
+      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
+                                             owned_groups)
+
+      assert self.group_uuid in inst_groups, \
+        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
 
   def Exec(self, feedback_fn):
     """Verify integrity of cluster disks.
 
   def Exec(self, feedback_fn):
     """Verify integrity of cluster disks.
@@ -2853,50 +3330,41 @@ class LUClusterVerifyDisks(NoHooksLU):
         missing volumes
 
     """
         missing volumes
 
     """
-    result = res_nodes, res_instances, res_missing = {}, [], {}
+    res_nodes = {}
+    res_instances = set()
+    res_missing = {}
 
 
-    nodes = utils.NiceSort(self.cfg.GetVmCapableNodeList())
-    instances = self.cfg.GetAllInstancesInfo().values()
+    nv_dict = _MapInstanceDisksToNodes([inst
+            for inst in self.instances.values()
+            if inst.admin_state == constants.ADMINST_UP])
 
 
-    nv_dict = {}
-    for inst in instances:
-      inst_lvs = {}
-      if not inst.admin_up:
-        continue
-      inst.MapLVsByNode(inst_lvs)
-      # transform { iname: {node: [vol,],},} to {(node, vol): iname}
-      for node, vol_list in inst_lvs.iteritems():
-        for vol in vol_list:
-          nv_dict[(node, vol)] = inst
-
-    if not nv_dict:
-      return result
-
-    node_lvs = self.rpc.call_lv_list(nodes, [])
-    for node, node_res in node_lvs.items():
-      if node_res.offline:
-        continue
-      msg = node_res.fail_msg
-      if msg:
-        logging.warning("Error enumerating LVs on node %s: %s", node, msg)
-        res_nodes[node] = msg
-        continue
+    if nv_dict:
+      nodes = utils.NiceSort(set(self.owned_locks(locking.LEVEL_NODE)) &
+                             set(self.cfg.GetVmCapableNodeList()))
 
 
-      lvs = node_res.payload
-      for lv_name, (_, _, lv_online) in lvs.items():
-        inst = nv_dict.pop((node, lv_name), None)
-        if (not lv_online and inst is not None
-            and inst.name not in res_instances):
-          res_instances.append(inst.name)
+      node_lvs = self.rpc.call_lv_list(nodes, [])
 
 
-    # any leftover items in nv_dict are missing LVs, let's arrange the
-    # data better
-    for key, inst in nv_dict.iteritems():
-      if inst.name not in res_missing:
-        res_missing[inst.name] = []
-      res_missing[inst.name].append(key)
+      for (node, node_res) in node_lvs.items():
+        if node_res.offline:
+          continue
 
 
-    return result
+        msg = node_res.fail_msg
+        if msg:
+          logging.warning("Error enumerating LVs on node %s: %s", node, msg)
+          res_nodes[node] = msg
+          continue
+
+        for lv_name, (_, _, lv_online) in node_res.payload.items():
+          inst = nv_dict.pop((node, lv_name), None)
+          if not (lv_online or inst is None):
+            res_instances.add(inst)
+
+      # any leftover items in nv_dict are missing LVs, let's arrange the data
+      # better
+      for key, inst in nv_dict.iteritems():
+        res_missing.setdefault(inst, []).append(list(key))
+
+    return (res_nodes, list(res_instances), res_missing)
 
 
 class LUClusterRepairDiskSizes(NoHooksLU):
 
 
 class LUClusterRepairDiskSizes(NoHooksLU):
@@ -2909,21 +3377,24 @@ class LUClusterRepairDiskSizes(NoHooksLU):
     if self.op.instances:
       self.wanted_names = _GetWantedInstances(self, self.op.instances)
       self.needed_locks = {
     if self.op.instances:
       self.wanted_names = _GetWantedInstances(self, self.op.instances)
       self.needed_locks = {
-        locking.LEVEL_NODE: [],
+        locking.LEVEL_NODE_RES: [],
         locking.LEVEL_INSTANCE: self.wanted_names,
         }
         locking.LEVEL_INSTANCE: self.wanted_names,
         }
-      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+      self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
     else:
       self.wanted_names = None
       self.needed_locks = {
     else:
       self.wanted_names = None
       self.needed_locks = {
-        locking.LEVEL_NODE: locking.ALL_SET,
+        locking.LEVEL_NODE_RES: locking.ALL_SET,
         locking.LEVEL_INSTANCE: locking.ALL_SET,
         }
         locking.LEVEL_INSTANCE: locking.ALL_SET,
         }
-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+    self.share_locks = {
+      locking.LEVEL_NODE_RES: 1,
+      locking.LEVEL_INSTANCE: 0,
+      }
 
   def DeclareLocks(self, level):
 
   def DeclareLocks(self, level):
-    if level == locking.LEVEL_NODE and self.wanted_names is not None:
-      self._LockInstancesNodes(primary_only=True)
+    if level == locking.LEVEL_NODE_RES and self.wanted_names is not None:
+      self._LockInstancesNodes(primary_only=True, level=level)
 
   def CheckPrereq(self):
     """Check prerequisites.
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -2932,10 +3403,10 @@ class LUClusterRepairDiskSizes(NoHooksLU):
 
     """
     if self.wanted_names is None:
 
     """
     if self.wanted_names is None:
-      self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
+      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
 
 
-    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
-                             in self.wanted_names]
+    self.wanted_instances = \
+        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
 
   def _EnsureChildSizes(self, disk):
     """Ensure children of the disk have the needed disk size.
 
   def _EnsureChildSizes(self, disk):
     """Ensure children of the disk have the needed disk size.
@@ -2974,6 +3445,11 @@ class LUClusterRepairDiskSizes(NoHooksLU):
       for idx, disk in enumerate(instance.disks):
         per_node_disks[pnode].append((instance, idx, disk))
 
       for idx, disk in enumerate(instance.disks):
         per_node_disks[pnode].append((instance, idx, disk))
 
+    assert not (frozenset(per_node_disks.keys()) -
+                self.owned_locks(locking.LEVEL_NODE_RES)), \
+      "Not owning correct locks"
+    assert not self.owned_locks(locking.LEVEL_NODE)
+
     changed = []
     for node, dskl in per_node_disks.items():
       newl = [v[2].Copy() for v in dskl]
     changed = []
     for node, dskl in per_node_disks.items():
       newl = [v[2].Copy() for v in dskl]
@@ -3063,29 +3539,33 @@ class LUClusterRename(LogicalUnit):
 
     """
     clustername = self.op.name
 
     """
     clustername = self.op.name
-    ip = self.ip
+    new_ip = self.ip
 
     # shutdown the master IP
 
     # shutdown the master IP
-    master = self.cfg.GetMasterNode()
-    result = self.rpc.call_node_stop_master(master, False)
+    master_params = self.cfg.GetMasterNetworkParameters()
+    ems = self.cfg.GetUseExternalMipScript()
+    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+                                                     master_params, ems)
     result.Raise("Could not disable the master role")
 
     try:
       cluster = self.cfg.GetClusterInfo()
       cluster.cluster_name = clustername
     result.Raise("Could not disable the master role")
 
     try:
       cluster = self.cfg.GetClusterInfo()
       cluster.cluster_name = clustername
-      cluster.master_ip = ip
+      cluster.master_ip = new_ip
       self.cfg.Update(cluster, feedback_fn)
 
       # update the known hosts file
       ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
       node_list = self.cfg.GetOnlineNodeList()
       try:
       self.cfg.Update(cluster, feedback_fn)
 
       # update the known hosts file
       ssh.WriteKnownHostsFile(self.cfg, constants.SSH_KNOWN_HOSTS_FILE)
       node_list = self.cfg.GetOnlineNodeList()
       try:
-        node_list.remove(master)
+        node_list.remove(master_params.name)
       except ValueError:
         pass
       _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
     finally:
       except ValueError:
         pass
       _UploadHelper(self, node_list, constants.SSH_KNOWN_HOSTS_FILE)
     finally:
-      result = self.rpc.call_node_start_master(master, False, False)
+      master_params.ip = new_ip
+      result = self.rpc.call_node_activate_master_ip(master_params.name,
+                                                     master_params, ems)
       msg = result.fail_msg
       if msg:
         self.LogWarning("Could not re-enable the master role on"
       msg = result.fail_msg
       if msg:
         self.LogWarning("Could not re-enable the master role on"
@@ -3094,8 +3574,29 @@ class LUClusterRename(LogicalUnit):
     return clustername
 
 
     return clustername
 
 
-class LUClusterSetParams(LogicalUnit):
-  """Change the parameters of the cluster.
+def _ValidateNetmask(cfg, netmask):
+  """Checks if a netmask is valid.
+
+  @type cfg: L{config.ConfigWriter}
+  @param cfg: The cluster configuration
+  @type netmask: int
+  @param netmask: the netmask to be verified
+  @raise errors.OpPrereqError: if the validation fails
+
+  """
+  ip_family = cfg.GetPrimaryIPFamily()
+  try:
+    ipcls = netutils.IPAddress.GetClassFromIpFamily(ip_family)
+  except errors.ProgrammerError:
+    raise errors.OpPrereqError("Invalid primary ip family: %s." %
+                               ip_family)
+  if not ipcls.ValidateNetmask(netmask):
+    raise errors.OpPrereqError("CIDR netmask (%s) not valid" %
+                                (netmask))
+
+
+class LUClusterSetParams(LogicalUnit):
+  """Change the parameters of the cluster.
 
   """
   HPATH = "cluster-modify"
 
   """
   HPATH = "cluster-modify"
@@ -3115,6 +3616,13 @@ class LUClusterSetParams(LogicalUnit):
     if self.op.remove_uids:
       uidpool.CheckUidPool(self.op.remove_uids)
 
     if self.op.remove_uids:
       uidpool.CheckUidPool(self.op.remove_uids)
 
+    if self.op.master_netmask is not None:
+      _ValidateNetmask(self.cfg, self.op.master_netmask)
+
+    if self.op.diskparams:
+      for dt_params in self.op.diskparams.values():
+        utils.ForceDictType(dt_params, constants.DISK_DT_TYPES)
+
   def ExpandNames(self):
     # FIXME: in the future maybe other cluster params won't require checking on
     # all nodes to be modified.
   def ExpandNames(self):
     # FIXME: in the future maybe other cluster params won't require checking on
     # all nodes to be modified.
@@ -3157,7 +3665,7 @@ class LUClusterSetParams(LogicalUnit):
                                    " drbd-based instances exist",
                                    errors.ECODE_INVAL)
 
                                    " drbd-based instances exist",
                                    errors.ECODE_INVAL)
 
-    node_list = self.glm.list_owned(locking.LEVEL_NODE)
+    node_list = self.owned_locks(locking.LEVEL_NODE)
 
     # if vg_name not None, checks given volume group on all nodes
     if self.op.vg_name:
 
     # if vg_name not None, checks given volume group on all nodes
     if self.op.vg_name:
@@ -3179,8 +3687,7 @@ class LUClusterSetParams(LogicalUnit):
     if self.op.drbd_helper:
       # checks given drbd helper on all nodes
       helpers = self.rpc.call_drbd_helper(node_list)
     if self.op.drbd_helper:
       # checks given drbd helper on all nodes
       helpers = self.rpc.call_drbd_helper(node_list)
-      for node in node_list:
-        ninfo = self.cfg.GetNodeInfo(node)
+      for (node, ninfo) in self.cfg.GetMultiNodeInfo(node_list):
         if ninfo.offline:
           self.LogInfo("Not checking drbd helper on offline node %s", node)
           continue
         if ninfo.offline:
           self.LogInfo("Not checking drbd helper on offline node %s", node)
           continue
@@ -3197,6 +3704,7 @@ class LUClusterSetParams(LogicalUnit):
     self.cluster = cluster = self.cfg.GetClusterInfo()
     # validate params changes
     if self.op.beparams:
     self.cluster = cluster = self.cfg.GetClusterInfo()
     # validate params changes
     if self.op.beparams:
+      objects.UpgradeBeParams(self.op.beparams)
       utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
       self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
 
       utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
       self.new_beparams = cluster.SimpleFillBE(self.op.beparams)
 
@@ -3210,6 +3718,29 @@ class LUClusterSetParams(LogicalUnit):
         self.new_ndparams["oob_program"] = \
             constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
 
         self.new_ndparams["oob_program"] = \
             constants.NDC_DEFAULTS[constants.ND_OOB_PROGRAM]
 
+    if self.op.hv_state:
+      new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
+                                            self.cluster.hv_state_static)
+      self.new_hv_state = dict((hv, cluster.SimpleFillHvState(values))
+                               for hv, values in new_hv_state.items())
+
+    if self.op.disk_state:
+      new_disk_state = _MergeAndVerifyDiskState(self.op.disk_state,
+                                                self.cluster.disk_state_static)
+      self.new_disk_state = \
+        dict((storage, dict((name, cluster.SimpleFillDiskState(values))
+                            for name, values in svalues.items()))
+             for storage, svalues in new_disk_state.items())
+
+    if self.op.ipolicy:
+      ipolicy = {}
+      for key, value in self.op.ipolicy.items():
+        utils.ForceDictType(value, constants.ISPECS_PARAMETER_TYPES)
+        ipolicy[key] = _GetUpdatedParams(cluster.ipolicy.get(key, {}),
+                                          value)
+      objects.InstancePolicy.CheckParameterSyntax(ipolicy)
+      self.new_ipolicy = ipolicy
+
     if self.op.nicparams:
       utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
       self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
     if self.op.nicparams:
       utils.ForceDictType(self.op.nicparams, constants.NICS_PARAMETER_TYPES)
       self.new_nicparams = cluster.SimpleFillNIC(self.op.nicparams)
@@ -3247,6 +3778,15 @@ class LUClusterSetParams(LogicalUnit):
         else:
           self.new_hvparams[hv_name].update(hv_dict)
 
         else:
           self.new_hvparams[hv_name].update(hv_dict)
 
+    # disk template parameters
+    self.new_diskparams = objects.FillDict(cluster.diskparams, {})
+    if self.op.diskparams:
+      for dt_name, dt_params in self.op.diskparams.items():
+        if dt_name not in self.op.diskparams:
+          self.new_diskparams[dt_name] = dt_params
+        else:
+          self.new_diskparams[dt_name].update(dt_params)
+
     # os hypervisor parameters
     self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
     if self.op.os_hvp:
     # os hypervisor parameters
     self.new_os_hvp = objects.FillDict(cluster.os_hvp, {})
     if self.op.os_hvp:
@@ -3361,10 +3901,18 @@ class LUClusterSetParams(LogicalUnit):
       self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
     if self.op.nicparams:
       self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
       self.cluster.beparams[constants.PP_DEFAULT] = self.new_beparams
     if self.op.nicparams:
       self.cluster.nicparams[constants.PP_DEFAULT] = self.new_nicparams
+    if self.op.ipolicy:
+      self.cluster.ipolicy = self.new_ipolicy
     if self.op.osparams:
       self.cluster.osparams = self.new_osp
     if self.op.ndparams:
       self.cluster.ndparams = self.new_ndparams
     if self.op.osparams:
       self.cluster.osparams = self.new_osp
     if self.op.ndparams:
       self.cluster.ndparams = self.new_ndparams
+    if self.op.diskparams:
+      self.cluster.diskparams = self.new_diskparams
+    if self.op.hv_state:
+      self.cluster.hv_state_static = self.new_hv_state
+    if self.op.disk_state:
+      self.cluster.disk_state_static = self.new_disk_state
 
     if self.op.candidate_pool_size is not None:
       self.cluster.candidate_pool_size = self.op.candidate_pool_size
 
     if self.op.candidate_pool_size is not None:
       self.cluster.candidate_pool_size = self.op.candidate_pool_size
@@ -3372,6 +3920,9 @@ class LUClusterSetParams(LogicalUnit):
       _AdjustCandidatePool(self, [])
 
     if self.op.maintain_node_health is not None:
       _AdjustCandidatePool(self, [])
 
     if self.op.maintain_node_health is not None:
+      if self.op.maintain_node_health and not constants.ENABLE_CONFD:
+        feedback_fn("Note: CONFD was disabled at build time, node health"
+                    " maintenance is not useful (still enabling it)")
       self.cluster.maintain_node_health = self.op.maintain_node_health
 
     if self.op.prealloc_wipe_disks is not None:
       self.cluster.maintain_node_health = self.op.maintain_node_health
 
     if self.op.prealloc_wipe_disks is not None:
@@ -3392,6 +3943,9 @@ class LUClusterSetParams(LogicalUnit):
     if self.op.reserved_lvs is not None:
       self.cluster.reserved_lvs = self.op.reserved_lvs
 
     if self.op.reserved_lvs is not None:
       self.cluster.reserved_lvs = self.op.reserved_lvs
 
+    if self.op.use_external_mip_script is not None:
+      self.cluster.use_external_mip_script = self.op.use_external_mip_script
+
     def helper_os(aname, mods, desc):
       desc += " OS list"
       lst = getattr(self.cluster, aname)
     def helper_os(aname, mods, desc):
       desc += " OS list"
       lst = getattr(self.cluster, aname)
@@ -3416,21 +3970,40 @@ class LUClusterSetParams(LogicalUnit):
       helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
 
     if self.op.master_netdev:
       helper_os("blacklisted_os", self.op.blacklisted_os, "blacklisted")
 
     if self.op.master_netdev:
-      master = self.cfg.GetMasterNode()
+      master_params = self.cfg.GetMasterNetworkParameters()
+      ems = self.cfg.GetUseExternalMipScript()
       feedback_fn("Shutting down master ip on the current netdev (%s)" %
                   self.cluster.master_netdev)
       feedback_fn("Shutting down master ip on the current netdev (%s)" %
                   self.cluster.master_netdev)
-      result = self.rpc.call_node_stop_master(master, False)
+      result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+                                                       master_params, ems)
       result.Raise("Could not disable the master ip")
       feedback_fn("Changing master_netdev from %s to %s" %
       result.Raise("Could not disable the master ip")
       feedback_fn("Changing master_netdev from %s to %s" %
-                  (self.cluster.master_netdev, self.op.master_netdev))
+                  (master_params.netdev, self.op.master_netdev))
       self.cluster.master_netdev = self.op.master_netdev
 
       self.cluster.master_netdev = self.op.master_netdev
 
+    if self.op.master_netmask:
+      master_params = self.cfg.GetMasterNetworkParameters()
+      feedback_fn("Changing master IP netmask to %s" % self.op.master_netmask)
+      result = self.rpc.call_node_change_master_netmask(master_params.name,
+                                                        master_params.netmask,
+                                                        self.op.master_netmask,
+                                                        master_params.ip,
+                                                        master_params.netdev)
+      if result.fail_msg:
+        msg = "Could not change the master IP netmask: %s" % result.fail_msg
+        feedback_fn(msg)
+
+      self.cluster.master_netmask = self.op.master_netmask
+
     self.cfg.Update(self.cluster, feedback_fn)
 
     if self.op.master_netdev:
     self.cfg.Update(self.cluster, feedback_fn)
 
     if self.op.master_netdev:
+      master_params = self.cfg.GetMasterNetworkParameters()
       feedback_fn("Starting the master ip on the new master netdev (%s)" %
                   self.op.master_netdev)
       feedback_fn("Starting the master ip on the new master netdev (%s)" %
                   self.op.master_netdev)
-      result = self.rpc.call_node_start_master(master, False, False)
+      ems = self.cfg.GetUseExternalMipScript()
+      result = self.rpc.call_node_activate_master_ip(master_params.name,
+                                                     master_params, ems)
       if result.fail_msg:
         self.LogWarning("Could not re-enable the master ip on"
                         " the master, please restart manually: %s",
       if result.fail_msg:
         self.LogWarning("Could not re-enable the master ip on"
                         " the master, please restart manually: %s",
@@ -3463,36 +4036,58 @@ def _ComputeAncillaryFiles(cluster, redist):
     constants.SSH_KNOWN_HOSTS_FILE,
     constants.CONFD_HMAC_KEY,
     constants.CLUSTER_DOMAIN_SECRET_FILE,
     constants.SSH_KNOWN_HOSTS_FILE,
     constants.CONFD_HMAC_KEY,
     constants.CLUSTER_DOMAIN_SECRET_FILE,
+    constants.SPICE_CERT_FILE,
+    constants.SPICE_CACERT_FILE,
+    constants.RAPI_USERS_FILE,
     ])
 
   if not redist:
     files_all.update(constants.ALL_CERT_FILES)
     files_all.update(ssconf.SimpleStore().GetFileList())
     ])
 
   if not redist:
     files_all.update(constants.ALL_CERT_FILES)
     files_all.update(ssconf.SimpleStore().GetFileList())
+  else:
+    # we need to ship at least the RAPI certificate
+    files_all.add(constants.RAPI_CERT_FILE)
 
   if cluster.modify_etc_hosts:
     files_all.add(constants.ETC_HOSTS)
 
 
   if cluster.modify_etc_hosts:
     files_all.add(constants.ETC_HOSTS)
 
-  # Files which must either exist on all nodes or on none
-  files_all_opt = set([
+  # Files which are optional, these must:
+  # - be present in one other category as well
+  # - either exist or not exist on all nodes of that category (mc, vm all)
+  files_opt = set([
     constants.RAPI_USERS_FILE,
     ])
 
   # Files which should only be on master candidates
   files_mc = set()
     constants.RAPI_USERS_FILE,
     ])
 
   # Files which should only be on master candidates
   files_mc = set()
+
   if not redist:
     files_mc.add(constants.CLUSTER_CONF_FILE)
 
   if not redist:
     files_mc.add(constants.CLUSTER_CONF_FILE)
 
+    # FIXME: this should also be replicated but Ganeti doesn't support files_mc
+    # replication
+    files_mc.add(constants.DEFAULT_MASTER_SETUP_SCRIPT)
+
   # Files which should only be on VM-capable nodes
   files_vm = set(filename
     for hv_name in cluster.enabled_hypervisors
   # Files which should only be on VM-capable nodes
   files_vm = set(filename
     for hv_name in cluster.enabled_hypervisors
-    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles())
+    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[0])
+
+  files_opt |= set(filename
+    for hv_name in cluster.enabled_hypervisors
+    for filename in hypervisor.GetHypervisor(hv_name).GetAncillaryFiles()[1])
 
 
-  # Filenames must be unique
-  assert (len(files_all | files_all_opt | files_mc | files_vm) ==
-          sum(map(len, [files_all, files_all_opt, files_mc, files_vm]))), \
+  # Filenames in each category must be unique
+  all_files_set = files_all | files_mc | files_vm
+  assert (len(all_files_set) ==
+          sum(map(len, [files_all, files_mc, files_vm]))), \
          "Found file listed in more than one file list"
 
          "Found file listed in more than one file list"
 
-  return (files_all, files_all_opt, files_mc, files_vm)
+  # Optional files must be present in one other category
+  assert all_files_set.issuperset(files_opt), \
+         "Optional file not in a different required list"
+
+  return (files_all, files_opt, files_mc, files_vm)
 
 
 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
 
 
 def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
@@ -3526,7 +4121,7 @@ def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
       nodelist.remove(master_info.name)
 
   # Gather file lists
       nodelist.remove(master_info.name)
 
   # Gather file lists
-  (files_all, files_all_opt, files_mc, files_vm) = \
+  (files_all, _, files_mc, files_vm) = \
     _ComputeAncillaryFiles(cluster, True)
 
   # Never re-distribute configuration file from here
     _ComputeAncillaryFiles(cluster, True)
 
   # Never re-distribute configuration file from here
@@ -3536,7 +4131,6 @@ def _RedistributeAncillaryFiles(lu, additional_nodes=None, additional_vm=True):
 
   filemap = [
     (online_nodes, files_all),
 
   filemap = [
     (online_nodes, files_all),
-    (online_nodes, files_all_opt),
     (vm_nodes, files_vm),
     ]
 
     (vm_nodes, files_vm),
     ]
 
@@ -3568,6 +4162,36 @@ class LUClusterRedistConf(NoHooksLU):
     _RedistributeAncillaryFiles(self)
 
 
     _RedistributeAncillaryFiles(self)
 
 
+class LUClusterActivateMasterIp(NoHooksLU):
+  """Activate the master IP on the master node.
+
+  """
+  def Exec(self, feedback_fn):
+    """Activate the master IP.
+
+    """
+    master_params = self.cfg.GetMasterNetworkParameters()
+    ems = self.cfg.GetUseExternalMipScript()
+    result = self.rpc.call_node_activate_master_ip(master_params.name,
+                                                   master_params, ems)
+    result.Raise("Could not activate the master IP")
+
+
+class LUClusterDeactivateMasterIp(NoHooksLU):
+  """Deactivate the master IP on the master node.
+
+  """
+  def Exec(self, feedback_fn):
+    """Deactivate the master IP.
+
+    """
+    master_params = self.cfg.GetMasterNetworkParameters()
+    ems = self.cfg.GetUseExternalMipScript()
+    result = self.rpc.call_node_deactivate_master_ip(master_params.name,
+                                                     master_params, ems)
+    result.Raise("Could not deactivate the master IP")
+
+
 def _WaitForSync(lu, instance, disks=None, oneshot=False):
   """Sleep and poll for an instance's disk to sync.
 
 def _WaitForSync(lu, instance, disks=None, oneshot=False):
   """Sleep and poll for an instance's disk to sync.
 
@@ -3739,9 +4363,7 @@ class LUOobCommand(NoHooksLU):
     if self.op.command in self._SKIP_MASTER:
       assert self.master_node not in self.op.node_names
 
     if self.op.command in self._SKIP_MASTER:
       assert self.master_node not in self.op.node_names
 
-    for node_name in self.op.node_names:
-      node = self.cfg.GetNodeInfo(node_name)
-
+    for (node_name, node) in self.cfg.GetMultiNodeInfo(self.op.node_names):
       if node is None:
         raise errors.OpPrereqError("Node %s not found" % node_name,
                                    errors.ECODE_NOENT)
       if node is None:
         raise errors.OpPrereqError("Node %s not found" % node_name,
                                    errors.ECODE_NOENT)
@@ -3858,6 +4480,7 @@ class LUOobCommand(NoHooksLU):
       raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
                                utils.CommaJoin(errs))
 
       raise errors.OpExecError("Check of out-of-band payload failed due to %s" %
                                utils.CommaJoin(errs))
 
+
 class _OsQuery(_QueryBase):
   FIELDS = query.OS_FIELDS
 
 class _OsQuery(_QueryBase):
   FIELDS = query.OS_FIELDS
 
@@ -4065,15 +4688,12 @@ class LUNodeRemove(LogicalUnit):
     node = self.cfg.GetNodeInfo(self.op.node_name)
     assert node is not None
 
     node = self.cfg.GetNodeInfo(self.op.node_name)
     assert node is not None
 
-    instance_list = self.cfg.GetInstanceList()
-
     masternode = self.cfg.GetMasterNode()
     if node.name == masternode:
       raise errors.OpPrereqError("Node is the master node, failover to another"
                                  " node is required", errors.ECODE_INVAL)
 
     masternode = self.cfg.GetMasterNode()
     if node.name == masternode:
       raise errors.OpPrereqError("Node is the master node, failover to another"
                                  " node is required", errors.ECODE_INVAL)
 
-    for instance_name in instance_list:
-      instance = self.cfg.GetInstanceInfo(instance_name)
+    for instance_name, instance in self.cfg.GetAllInstancesInfo().items():
       if node.name in instance.all_nodes:
         raise errors.OpPrereqError("Instance %s is still running on the node,"
                                    " please remove first" % instance_name,
       if node.name in instance.all_nodes:
         raise errors.OpPrereqError("Instance %s is still running on the node,"
                                    " please remove first" % instance_name,
@@ -4091,6 +4711,9 @@ class LUNodeRemove(LogicalUnit):
 
     modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
 
 
     modify_ssh_setup = self.cfg.GetClusterInfo().modify_ssh_setup
 
+    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+      "Not owning BGL"
+
     # Promote nodes to master candidate as needed
     _AdjustCandidatePool(self, exceptions=[node.name])
     self.context.RemoveNode(node.name)
     # Promote nodes to master candidate as needed
     _AdjustCandidatePool(self, exceptions=[node.name])
     self.context.RemoveNode(node.name)
@@ -4119,7 +4742,7 @@ class _NodeQuery(_QueryBase):
 
   def ExpandNames(self, lu):
     lu.needed_locks = {}
 
   def ExpandNames(self, lu):
     lu.needed_locks = {}
-    lu.share_locks[locking.LEVEL_NODE] = 1
+    lu.share_locks = _ShareAll()
 
     if self.names:
       self.wanted = _GetWantedNodes(lu, self.names)
 
     if self.names:
       self.wanted = _GetWantedNodes(lu, self.names)
@@ -4130,7 +4753,7 @@ class _NodeQuery(_QueryBase):
                        query.NQ_LIVE in self.requested_data)
 
     if self.do_locking:
                        query.NQ_LIVE in self.requested_data)
 
     if self.do_locking:
-      # if we don't request only static fields, we need to lock the nodes
+      # If any non-static field is requested we need to lock the nodes
       lu.needed_locks[locking.LEVEL_NODE] = self.wanted
 
   def DeclareLocks(self, lu, level):
       lu.needed_locks[locking.LEVEL_NODE] = self.wanted
 
   def DeclareLocks(self, lu, level):
@@ -4149,9 +4772,9 @@ class _NodeQuery(_QueryBase):
       # filter out non-vm_capable nodes
       toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
 
       # filter out non-vm_capable nodes
       toquery_nodes = [name for name in nodenames if all_info[name].vm_capable]
 
-      node_data = lu.rpc.call_node_info(toquery_nodes, lu.cfg.GetVGName(),
-                                        lu.cfg.GetHypervisorType())
-      live_data = dict((name, nresult.payload)
+      node_data = lu.rpc.call_node_info(toquery_nodes, [lu.cfg.GetVGName()],
+                                        [lu.cfg.GetHypervisorType()])
+      live_data = dict((name, _MakeLegacyNodeInfo(nresult.payload))
                        for (name, nresult) in node_data.items()
                        if not nresult.fail_msg and nresult.payload)
     else:
                        for (name, nresult) in node_data.items()
                        if not nresult.fail_msg and nresult.payload)
     else:
@@ -4194,7 +4817,7 @@ class LUNodeQuery(NoHooksLU):
   """Logical unit for querying nodes.
 
   """
   """Logical unit for querying nodes.
 
   """
-  # pylint: disable-msg=W0142
+  # pylint: disable=W0142
   REQ_BGL = False
 
   def CheckArguments(self):
   REQ_BGL = False
 
   def CheckArguments(self):
@@ -4204,6 +4827,9 @@ class LUNodeQuery(NoHooksLU):
   def ExpandNames(self):
     self.nq.ExpandNames(self)
 
   def ExpandNames(self):
     self.nq.ExpandNames(self)
 
+  def DeclareLocks(self, level):
+    self.nq.DeclareLocks(self, level)
+
   def Exec(self, feedback_fn):
     return self.nq.OldStyleQuery(self)
 
   def Exec(self, feedback_fn):
     return self.nq.OldStyleQuery(self)
 
@@ -4222,8 +4848,9 @@ class LUNodeQueryvols(NoHooksLU):
                        selected=self.op.output_fields)
 
   def ExpandNames(self):
                        selected=self.op.output_fields)
 
   def ExpandNames(self):
+    self.share_locks = _ShareAll()
     self.needed_locks = {}
     self.needed_locks = {}
-    self.share_locks[locking.LEVEL_NODE] = 1
+
     if not self.op.nodes:
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
     else:
     if not self.op.nodes:
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
     else:
@@ -4234,13 +4861,11 @@ class LUNodeQueryvols(NoHooksLU):
     """Computes the list of nodes and their attributes.
 
     """
     """Computes the list of nodes and their attributes.
 
     """
-    nodenames = self.glm.list_owned(locking.LEVEL_NODE)
+    nodenames = self.owned_locks(locking.LEVEL_NODE)
     volumes = self.rpc.call_node_volumes(nodenames)
 
     volumes = self.rpc.call_node_volumes(nodenames)
 
-    ilist = [self.cfg.GetInstanceInfo(iname) for iname
-             in self.cfg.GetInstanceList()]
-
-    lv_by_node = dict([(inst, inst.MapLVsByNode()) for inst in ilist])
+    ilist = self.cfg.GetAllInstancesInfo()
+    vol2inst = _MapInstanceDisksToNodes(ilist.values())
 
     output = []
     for node in nodenames:
 
     output = []
     for node in nodenames:
@@ -4252,8 +4877,8 @@ class LUNodeQueryvols(NoHooksLU):
         self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
         continue
 
         self.LogWarning("Can't compute volume data on node %s: %s", node, msg)
         continue
 
-      node_vols = nresult.payload[:]
-      node_vols.sort(key=lambda vol: vol['dev'])
+      node_vols = sorted(nresult.payload,
+                         key=operator.itemgetter("dev"))
 
       for vol in node_vols:
         node_output = []
 
       for vol in node_vols:
         node_output = []
@@ -4261,22 +4886,15 @@ class LUNodeQueryvols(NoHooksLU):
           if field == "node":
             val = node
           elif field == "phys":
           if field == "node":
             val = node
           elif field == "phys":
-            val = vol['dev']
+            val = vol["dev"]
           elif field == "vg":
           elif field == "vg":
-            val = vol['vg']
+            val = vol["vg"]
           elif field == "name":
           elif field == "name":
-            val = vol['name']
+            val = vol["name"]
           elif field == "size":
           elif field == "size":
-            val = int(float(vol['size']))
+            val = int(float(vol["size"]))
           elif field == "instance":
           elif field == "instance":
-            for inst in ilist:
-              if node not in lv_by_node[inst]:
-                continue
-              if vol['name'] in lv_by_node[inst][node]:
-                val = inst.name
-                break
-            else:
-              val = '-'
+            val = vol2inst.get((node, vol["vg"] + "/" + vol["name"]), "-")
           else:
             raise errors.ParameterError(field)
           node_output.append(str(val))
           else:
             raise errors.ParameterError(field)
           node_output.append(str(val))
@@ -4299,8 +4917,8 @@ class LUNodeQueryStorage(NoHooksLU):
                        selected=self.op.output_fields)
 
   def ExpandNames(self):
                        selected=self.op.output_fields)
 
   def ExpandNames(self):
+    self.share_locks = _ShareAll()
     self.needed_locks = {}
     self.needed_locks = {}
-    self.share_locks[locking.LEVEL_NODE] = 1
 
     if self.op.nodes:
       self.needed_locks[locking.LEVEL_NODE] = \
 
     if self.op.nodes:
       self.needed_locks[locking.LEVEL_NODE] = \
@@ -4312,7 +4930,7 @@ class LUNodeQueryStorage(NoHooksLU):
     """Computes the list of nodes and their attributes.
 
     """
     """Computes the list of nodes and their attributes.
 
     """
-    self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
+    self.nodes = self.owned_locks(locking.LEVEL_NODE)
 
     # Always get name to sort by
     if constants.SF_NAME in self.op.output_fields:
 
     # Always get name to sort by
     if constants.SF_NAME in self.op.output_fields:
@@ -4374,8 +4992,7 @@ class _InstanceQuery(_QueryBase):
 
   def ExpandNames(self, lu):
     lu.needed_locks = {}
 
   def ExpandNames(self, lu):
     lu.needed_locks = {}
-    lu.share_locks[locking.LEVEL_INSTANCE] = 1
-    lu.share_locks[locking.LEVEL_NODE] = 1
+    lu.share_locks = _ShareAll()
 
     if self.names:
       self.wanted = _GetWantedInstances(lu, self.names)
 
     if self.names:
       self.wanted = _GetWantedInstances(lu, self.names)
@@ -4386,17 +5003,43 @@ class _InstanceQuery(_QueryBase):
                        query.IQ_LIVE in self.requested_data)
     if self.do_locking:
       lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
                        query.IQ_LIVE in self.requested_data)
     if self.do_locking:
       lu.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
+      lu.needed_locks[locking.LEVEL_NODEGROUP] = []
       lu.needed_locks[locking.LEVEL_NODE] = []
       lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
       lu.needed_locks[locking.LEVEL_NODE] = []
       lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
+    self.do_grouplocks = (self.do_locking and
+                          query.IQ_NODES in self.requested_data)
+
   def DeclareLocks(self, lu, level):
   def DeclareLocks(self, lu, level):
-    if level == locking.LEVEL_NODE and self.do_locking:
-      lu._LockInstancesNodes() # pylint: disable-msg=W0212
+    if self.do_locking:
+      if level == locking.LEVEL_NODEGROUP and self.do_grouplocks:
+        assert not lu.needed_locks[locking.LEVEL_NODEGROUP]
+
+        # Lock all groups used by instances optimistically; this requires going
+        # via the node before it's locked, requiring verification later on
+        lu.needed_locks[locking.LEVEL_NODEGROUP] = \
+          set(group_uuid
+              for instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
+              for group_uuid in lu.cfg.GetInstanceNodeGroups(instance_name))
+      elif level == locking.LEVEL_NODE:
+        lu._LockInstancesNodes() # pylint: disable=W0212
+
+  @staticmethod
+  def _CheckGroupLocks(lu):
+    owned_instances = frozenset(lu.owned_locks(locking.LEVEL_INSTANCE))
+    owned_groups = frozenset(lu.owned_locks(locking.LEVEL_NODEGROUP))
+
+    # Check if node groups for locked instances are still correct
+    for instance_name in owned_instances:
+      _CheckInstanceNodeGroups(lu.cfg, instance_name, owned_groups)
 
   def _GetQueryData(self, lu):
     """Computes the list of instances and their attributes.
 
     """
 
   def _GetQueryData(self, lu):
     """Computes the list of instances and their attributes.
 
     """
+    if self.do_grouplocks:
+      self._CheckGroupLocks(lu)
+
     cluster = lu.cfg.GetClusterInfo()
     all_info = lu.cfg.GetAllInstancesInfo()
 
     cluster = lu.cfg.GetClusterInfo()
     all_info = lu.cfg.GetAllInstancesInfo()
 
@@ -4459,22 +5102,34 @@ class _InstanceQuery(_QueryBase):
     else:
       consinfo = None
 
     else:
       consinfo = None
 
+    if query.IQ_NODES in self.requested_data:
+      node_names = set(itertools.chain(*map(operator.attrgetter("all_nodes"),
+                                            instance_list)))
+      nodes = dict(lu.cfg.GetMultiNodeInfo(node_names))
+      groups = dict((uuid, lu.cfg.GetNodeGroup(uuid))
+                    for uuid in set(map(operator.attrgetter("group"),
+                                        nodes.values())))
+    else:
+      nodes = None
+      groups = None
+
     return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
                                    disk_usage, offline_nodes, bad_nodes,
     return query.InstanceQueryData(instance_list, lu.cfg.GetClusterInfo(),
                                    disk_usage, offline_nodes, bad_nodes,
-                                   live_data, wrongnode_inst, consinfo)
+                                   live_data, wrongnode_inst, consinfo,
+                                   nodes, groups)
 
 
 class LUQuery(NoHooksLU):
   """Query for resources/items of a certain kind.
 
   """
 
 
 class LUQuery(NoHooksLU):
   """Query for resources/items of a certain kind.
 
   """
-  # pylint: disable-msg=W0142
+  # pylint: disable=W0142
   REQ_BGL = False
 
   def CheckArguments(self):
     qcls = _GetQueryImplementation(self.op.what)
 
   REQ_BGL = False
 
   def CheckArguments(self):
     qcls = _GetQueryImplementation(self.op.what)
 
-    self.impl = qcls(self.op.filter, self.op.fields, False)
+    self.impl = qcls(self.op.qfilter, self.op.fields, self.op.use_locking)
 
   def ExpandNames(self):
     self.impl.ExpandNames(self)
 
   def ExpandNames(self):
     self.impl.ExpandNames(self)
@@ -4490,7 +5145,7 @@ class LUQueryFields(NoHooksLU):
   """Query for resources/items of a certain kind.
 
   """
   """Query for resources/items of a certain kind.
 
   """
-  # pylint: disable-msg=W0142
+  # pylint: disable=W0142
   REQ_BGL = False
 
   def CheckArguments(self):
   REQ_BGL = False
 
   def CheckArguments(self):
@@ -4630,9 +5285,7 @@ class LUNodeAdd(LogicalUnit):
 
     self.changed_primary_ip = False
 
 
     self.changed_primary_ip = False
 
-    for existing_node_name in node_list:
-      existing_node = cfg.GetNodeInfo(existing_node_name)
-
+    for existing_node_name, existing_node in cfg.GetMultiNodeInfo(node_list):
       if self.op.readd and node == existing_node_name:
         if existing_node.secondary_ip != secondary_ip:
           raise errors.OpPrereqError("Readded node doesn't have the same IP"
       if self.op.readd and node == existing_node_name:
         if existing_node.secondary_ip != secondary_ip:
           raise errors.OpPrereqError("Readded node doesn't have the same IP"
@@ -4731,6 +5384,9 @@ class LUNodeAdd(LogicalUnit):
     new_node = self.new_node
     node = new_node.name
 
     new_node = self.new_node
     node = new_node.name
 
+    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER), \
+      "Not owning BGL"
+
     # We adding a new node so we assume it's powered
     new_node.powered = True
 
     # We adding a new node so we assume it's powered
     new_node.powered = True
 
@@ -4739,7 +5395,7 @@ class LUNodeAdd(LogicalUnit):
     # later in the procedure; this also means that if the re-add
     # fails, we are left with a non-offlined, broken node
     if self.op.readd:
     # later in the procedure; this also means that if the re-add
     # fails, we are left with a non-offlined, broken node
     if self.op.readd:
-      new_node.drained = new_node.offline = False # pylint: disable-msg=W0201
+      new_node.drained = new_node.offline = False # pylint: disable=W0201
       self.LogInfo("Readding a node, the offline/drained flags were reset")
       # if we demote the node, we do cleanup later in the procedure
       new_node.master_candidate = self.master_candidate
       self.LogInfo("Readding a node, the offline/drained flags were reset")
       # if we demote the node, we do cleanup later in the procedure
       new_node.master_candidate = self.master_candidate
@@ -4785,7 +5441,7 @@ class LUNodeAdd(LogicalUnit):
 
     node_verify_list = [self.cfg.GetMasterNode()]
     node_verify_param = {
 
     node_verify_list = [self.cfg.GetMasterNode()]
     node_verify_param = {
-      constants.NV_NODELIST: [node],
+      constants.NV_NODELIST: ([node], {}),
       # TODO: do a node-net-test as well?
     }
 
       # TODO: do a node-net-test as well?
     }
 
@@ -4845,7 +5501,8 @@ class LUNodeSetParams(LogicalUnit):
     self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
     all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
                 self.op.master_capable, self.op.vm_capable,
     self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
     all_mods = [self.op.offline, self.op.master_candidate, self.op.drained,
                 self.op.master_capable, self.op.vm_capable,
-                self.op.secondary_ip, self.op.ndparams]
+                self.op.secondary_ip, self.op.ndparams, self.op.hv_state,
+                self.op.disk_state]
     if all_mods.count(None) == len(all_mods):
       raise errors.OpPrereqError("Please pass at least one modification",
                                  errors.ECODE_INVAL)
     if all_mods.count(None) == len(all_mods):
       raise errors.OpPrereqError("Please pass at least one modification",
                                  errors.ECODE_INVAL)
@@ -4869,35 +5526,32 @@ class LUNodeSetParams(LogicalUnit):
     self.lock_all = self.op.auto_promote and self.might_demote
     self.lock_instances = self.op.secondary_ip is not None
 
     self.lock_all = self.op.auto_promote and self.might_demote
     self.lock_instances = self.op.secondary_ip is not None
 
+  def _InstanceFilter(self, instance):
+    """Filter for getting affected instances.
+
+    """
+    return (instance.disk_template in constants.DTS_INT_MIRROR and
+            self.op.node_name in instance.all_nodes)
+
   def ExpandNames(self):
     if self.lock_all:
       self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
     else:
       self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
 
   def ExpandNames(self):
     if self.lock_all:
       self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
     else:
       self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
 
-    if self.lock_instances:
-      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+    # Since modifying a node can have severe effects on currently running
+    # operations the resource lock is at least acquired in shared mode
+    self.needed_locks[locking.LEVEL_NODE_RES] = \
+      self.needed_locks[locking.LEVEL_NODE]
 
 
-  def DeclareLocks(self, level):
-    # If we have locked all instances, before waiting to lock nodes, release
-    # all the ones living on nodes unrelated to the current operation.
-    if level == locking.LEVEL_NODE and self.lock_instances:
-      self.affected_instances = []
-      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
-        instances_keep = []
-
-        # Build list of instances to release
-        for instance_name in self.glm.list_owned(locking.LEVEL_INSTANCE):
-          instance = self.context.cfg.GetInstanceInfo(instance_name)
-          if (instance.disk_template in constants.DTS_INT_MIRROR and
-              self.op.node_name in instance.all_nodes):
-            instances_keep.append(instance_name)
-            self.affected_instances.append(instance)
-
-        _ReleaseLocks(self, locking.LEVEL_INSTANCE, keep=instances_keep)
-
-        assert (set(self.glm.list_owned(locking.LEVEL_INSTANCE)) ==
-                set(instances_keep))
+    # Get node resource and instance locks in shared mode; they are not used
+    # for anything but read-only access
+    self.share_locks[locking.LEVEL_NODE_RES] = 1
+    self.share_locks[locking.LEVEL_INSTANCE] = 1
+
+    if self.lock_instances:
+      self.needed_locks[locking.LEVEL_INSTANCE] = \
+        frozenset(self.cfg.GetInstancesInfoByFilter(self._InstanceFilter))
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -4929,6 +5583,25 @@ class LUNodeSetParams(LogicalUnit):
     """
     node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
 
     """
     node = self.node = self.cfg.GetNodeInfo(self.op.node_name)
 
+    if self.lock_instances:
+      affected_instances = \
+        self.cfg.GetInstancesInfoByFilter(self._InstanceFilter)
+
+      # Verify instance locks
+      owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
+      wanted_instances = frozenset(affected_instances.keys())
+      if wanted_instances - owned_instances:
+        raise errors.OpPrereqError("Instances affected by changing node %s's"
+                                   " secondary IP address have changed since"
+                                   " locks were acquired, wanted '%s', have"
+                                   " '%s'; retry the operation" %
+                                   (self.op.node_name,
+                                    utils.CommaJoin(wanted_instances),
+                                    utils.CommaJoin(owned_instances)),
+                                   errors.ECODE_STATE)
+    else:
+      affected_instances = None
+
     if (self.op.master_candidate is not None or
         self.op.drained is not None or
         self.op.offline is not None):
     if (self.op.master_candidate is not None or
         self.op.drained is not None or
         self.op.offline is not None):
@@ -5018,7 +5691,9 @@ class LUNodeSetParams(LogicalUnit):
 
     if old_role == self._ROLE_OFFLINE and new_role != old_role:
       # Trying to transition out of offline status
 
     if old_role == self._ROLE_OFFLINE and new_role != old_role:
       # Trying to transition out of offline status
-      result = self.rpc.call_version([node.name])[node.name]
+      # TODO: Use standard RPC runner, but make sure it works when the node is
+      # still marked offline
+      result = rpc.BootstrapRunner().call_version([node.name])[node.name]
       if result.fail_msg:
         raise errors.OpPrereqError("Node %s is being de-offlined but fails"
                                    " to report its version: %s" %
       if result.fail_msg:
         raise errors.OpPrereqError("Node %s is being de-offlined but fails"
                                    " to report its version: %s" %
@@ -5037,16 +5712,21 @@ class LUNodeSetParams(LogicalUnit):
         raise errors.OpPrereqError("Cannot change the secondary ip on a single"
                                    " homed cluster", errors.ECODE_INVAL)
 
         raise errors.OpPrereqError("Cannot change the secondary ip on a single"
                                    " homed cluster", errors.ECODE_INVAL)
 
+      assert not (frozenset(affected_instances) -
+                  self.owned_locks(locking.LEVEL_INSTANCE))
+
       if node.offline:
       if node.offline:
-        if self.affected_instances:
-          raise errors.OpPrereqError("Cannot change secondary ip: offline"
-                                     " node has instances (%s) configured"
-                                     " to use it" % self.affected_instances)
+        if affected_instances:
+          raise errors.OpPrereqError("Cannot change secondary IP address:"
+                                     " offline node has instances (%s)"
+                                     " configured to use it" %
+                                     utils.CommaJoin(affected_instances.keys()))
       else:
         # On online nodes, check that no instances are running, and that
         # the node has the new ip and we can reach it.
       else:
         # On online nodes, check that no instances are running, and that
         # the node has the new ip and we can reach it.
-        for instance in self.affected_instances:
-          _CheckInstanceDown(self, instance, "cannot change secondary ip")
+        for instance in affected_instances.values():
+          _CheckInstanceState(self, instance, INSTANCE_DOWN,
+                              msg="cannot change secondary ip")
 
         _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
         if master.name != node.name:
 
         _CheckNodeHasSecondaryIP(self, node.name, self.op.secondary_ip, True)
         if master.name != node.name:
@@ -5063,6 +5743,15 @@ class LUNodeSetParams(LogicalUnit):
       utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = new_ndparams
 
       utils.ForceDictType(new_ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = new_ndparams
 
+    if self.op.hv_state:
+      self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
+                                                 self.node.hv_state_static)
+
+    if self.op.disk_state:
+      self.new_disk_state = \
+        _MergeAndVerifyDiskState(self.op.disk_state,
+                                 self.node.disk_state_static)
+
   def Exec(self, feedback_fn):
     """Modifies a node.
 
   def Exec(self, feedback_fn):
     """Modifies a node.
 
@@ -5079,6 +5768,12 @@ class LUNodeSetParams(LogicalUnit):
     if self.op.powered is not None:
       node.powered = self.op.powered
 
     if self.op.powered is not None:
       node.powered = self.op.powered
 
+    if self.op.hv_state:
+      node.hv_state_static = self.new_hv_state
+
+    if self.op.disk_state:
+      node.disk_state_static = self.new_disk_state
+
     for attr in ["master_capable", "vm_capable"]:
       val = getattr(self.op, attr)
       if val is not None:
     for attr in ["master_capable", "vm_capable"]:
       val = getattr(self.op, attr)
       if val is not None:
@@ -5186,17 +5881,20 @@ class LUClusterQuery(NoHooksLU):
       "architecture": (platform.architecture()[0], platform.machine()),
       "name": cluster.cluster_name,
       "master": cluster.master_node,
       "architecture": (platform.architecture()[0], platform.machine()),
       "name": cluster.cluster_name,
       "master": cluster.master_node,
-      "default_hypervisor": cluster.enabled_hypervisors[0],
+      "default_hypervisor": cluster.primary_hypervisor,
       "enabled_hypervisors": cluster.enabled_hypervisors,
       "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
                         for hypervisor_name in cluster.enabled_hypervisors]),
       "os_hvp": os_hvp,
       "beparams": cluster.beparams,
       "osparams": cluster.osparams,
       "enabled_hypervisors": cluster.enabled_hypervisors,
       "hvparams": dict([(hypervisor_name, cluster.hvparams[hypervisor_name])
                         for hypervisor_name in cluster.enabled_hypervisors]),
       "os_hvp": os_hvp,
       "beparams": cluster.beparams,
       "osparams": cluster.osparams,
+      "ipolicy": cluster.ipolicy,
       "nicparams": cluster.nicparams,
       "ndparams": cluster.ndparams,
       "candidate_pool_size": cluster.candidate_pool_size,
       "master_netdev": cluster.master_netdev,
       "nicparams": cluster.nicparams,
       "ndparams": cluster.ndparams,
       "candidate_pool_size": cluster.candidate_pool_size,
       "master_netdev": cluster.master_netdev,
+      "master_netmask": cluster.master_netmask,
+      "use_external_mip_script": cluster.use_external_mip_script,
       "volume_group_name": cluster.volume_group_name,
       "drbd_usermode_helper": cluster.drbd_usermode_helper,
       "file_storage_dir": cluster.file_storage_dir,
       "volume_group_name": cluster.volume_group_name,
       "drbd_usermode_helper": cluster.drbd_usermode_helper,
       "file_storage_dir": cluster.file_storage_dir,
@@ -5442,7 +6140,7 @@ def _SafeShutdownInstanceDisks(lu, instance, disks=None):
   _ShutdownInstanceDisks.
 
   """
   _ShutdownInstanceDisks.
 
   """
-  _CheckInstanceDown(lu, instance, "cannot shutdown disks")
+  _CheckInstanceState(lu, instance, INSTANCE_DOWN, msg="cannot shutdown disks")
   _ShutdownInstanceDisks(lu, instance, disks=disks)
 
 
   _ShutdownInstanceDisks(lu, instance, disks=disks)
 
 
@@ -5512,10 +6210,12 @@ def _CheckNodeFreeMemory(lu, node, reason, requested, hypervisor_name):
       we cannot check the node
 
   """
       we cannot check the node
 
   """
-  nodeinfo = lu.rpc.call_node_info([node], None, hypervisor_name)
+  nodeinfo = lu.rpc.call_node_info([node], None, [hypervisor_name])
   nodeinfo[node].Raise("Can't get data from node %s" % node,
                        prereq=True, ecode=errors.ECODE_ENVIRON)
   nodeinfo[node].Raise("Can't get data from node %s" % node,
                        prereq=True, ecode=errors.ECODE_ENVIRON)
-  free_mem = nodeinfo[node].payload.get('memory_free', None)
+  (_, _, (hv_info, )) = nodeinfo[node].payload
+
+  free_mem = hv_info.get("memory_free", None)
   if not isinstance(free_mem, int):
     raise errors.OpPrereqError("Can't compute free memory on node %s, result"
                                " was '%s'" % (node, free_mem),
   if not isinstance(free_mem, int):
     raise errors.OpPrereqError("Can't compute free memory on node %s, result"
                                " was '%s'" % (node, free_mem),
@@ -5570,12 +6270,13 @@ def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
       or we cannot check the node
 
   """
       or we cannot check the node
 
   """
-  nodeinfo = lu.rpc.call_node_info(nodenames, vg, None)
+  nodeinfo = lu.rpc.call_node_info(nodenames, [vg], None)
   for node in nodenames:
     info = nodeinfo[node]
     info.Raise("Cannot get current information from node %s" % node,
                prereq=True, ecode=errors.ECODE_ENVIRON)
   for node in nodenames:
     info = nodeinfo[node]
     info.Raise("Cannot get current information from node %s" % node,
                prereq=True, ecode=errors.ECODE_ENVIRON)
-    vg_free = info.payload.get("vg_free", None)
+    (_, (vg_info, ), _) = info.payload
+    vg_free = vg_info.get("vg_free", None)
     if not isinstance(vg_free, int):
       raise errors.OpPrereqError("Can't compute free disk space on node"
                                  " %s for vg %s, result was '%s'" %
     if not isinstance(vg_free, int):
       raise errors.OpPrereqError("Can't compute free disk space on node"
                                  " %s for vg %s, result was '%s'" %
@@ -5587,6 +6288,41 @@ def _CheckNodesFreeDiskOnVG(lu, nodenames, vg, requested):
                                  errors.ECODE_NORES)
 
 
                                  errors.ECODE_NORES)
 
 
+def _CheckNodesPhysicalCPUs(lu, nodenames, requested, hypervisor_name):
+  """Checks if nodes have enough physical CPUs
+
+  This function checks if all given nodes have the needed number of
+  physical CPUs. In case any node has less CPUs or we cannot get the
+  information from the node, this function raises an OpPrereqError
+  exception.
+
+  @type lu: C{LogicalUnit}
+  @param lu: a logical unit from which we get configuration data
+  @type nodenames: C{list}
+  @param nodenames: the list of node names to check
+  @type requested: C{int}
+  @param requested: the minimum acceptable number of physical CPUs
+  @raise errors.OpPrereqError: if the node doesn't have enough CPUs,
+      or we cannot check the node
+
+  """
+  nodeinfo = lu.rpc.call_node_info(nodenames, None, [hypervisor_name])
+  for node in nodenames:
+    info = nodeinfo[node]
+    info.Raise("Cannot get current information from node %s" % node,
+               prereq=True, ecode=errors.ECODE_ENVIRON)
+    (_, _, (hv_info, )) = info.payload
+    num_cpus = hv_info.get("cpu_total", None)
+    if not isinstance(num_cpus, int):
+      raise errors.OpPrereqError("Can't compute the number of physical CPUs"
+                                 " on node %s, result was '%s'" %
+                                 (node, num_cpus), errors.ECODE_ENVIRON)
+    if requested > num_cpus:
+      raise errors.OpPrereqError("Node %s has %s physical CPUs, but %s are "
+                                 "required" % (node, num_cpus, requested),
+                                 errors.ECODE_NORES)
+
+
 class LUInstanceStartup(LogicalUnit):
   """Starts an instance.
 
 class LUInstanceStartup(LogicalUnit):
   """Starts an instance.
 
@@ -5599,6 +6335,7 @@ class LUInstanceStartup(LogicalUnit):
     # extra beparams
     if self.op.beparams:
       # fill the beparams dict
     # extra beparams
     if self.op.beparams:
       # fill the beparams dict
+      objects.UpgradeBeParams(self.op.beparams)
       utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
 
   def ExpandNames(self):
       utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
 
   def ExpandNames(self):
@@ -5646,6 +6383,8 @@ class LUInstanceStartup(LogicalUnit):
       hv_type.CheckParameterSyntax(filled_hvp)
       _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
 
       hv_type.CheckParameterSyntax(filled_hvp)
       _CheckHVParams(self, instance.all_nodes, instance.hypervisor, filled_hvp)
 
+    _CheckInstanceState(self, instance, INSTANCE_ONLINE)
+
     self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
 
     if self.primary_offline and self.op.ignore_offline_nodes:
     self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
 
     if self.primary_offline and self.op.ignore_offline_nodes:
@@ -5669,7 +6408,7 @@ class LUInstanceStartup(LogicalUnit):
       if not remote_info.payload: # not running already
         _CheckNodeFreeMemory(self, instance.primary_node,
                              "starting instance %s" % instance.name,
       if not remote_info.payload: # not running already
         _CheckNodeFreeMemory(self, instance.primary_node,
                              "starting instance %s" % instance.name,
-                             bep[constants.BE_MEMORY], instance.hypervisor)
+                             bep[constants.BE_MAXMEM], instance.hypervisor)
 
   def Exec(self, feedback_fn):
     """Start the instance.
 
   def Exec(self, feedback_fn):
     """Start the instance.
@@ -5689,8 +6428,11 @@ class LUInstanceStartup(LogicalUnit):
 
       _StartInstanceDisks(self, instance, force)
 
 
       _StartInstanceDisks(self, instance, force)
 
-      result = self.rpc.call_instance_start(node_current, instance,
-                                            self.op.hvparams, self.op.beparams)
+      result = \
+        self.rpc.call_instance_start(node_current,
+                                     (instance, self.op.hvparams,
+                                      self.op.beparams),
+                                     self.op.startup_paused)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
@@ -5740,7 +6482,7 @@ class LUInstanceReboot(LogicalUnit):
     self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
     assert self.instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
     self.instance = instance = self.cfg.GetInstanceInfo(self.op.instance_name)
     assert self.instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
-
+    _CheckInstanceState(self, instance, INSTANCE_ONLINE)
     _CheckNodeOnline(self, instance.primary_node)
 
     # check bridges existence
     _CheckNodeOnline(self, instance.primary_node)
 
     # check bridges existence
@@ -5780,7 +6522,8 @@ class LUInstanceReboot(LogicalUnit):
         self.LogInfo("Instance %s was already stopped, starting now",
                      instance.name)
       _StartInstanceDisks(self, instance, ignore_secondaries)
         self.LogInfo("Instance %s was already stopped, starting now",
                      instance.name)
       _StartInstanceDisks(self, instance, ignore_secondaries)
-      result = self.rpc.call_instance_start(node_current, instance, None, None)
+      result = self.rpc.call_instance_start(node_current,
+                                            (instance, None, None), False)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
@@ -5828,6 +6571,8 @@ class LUInstanceShutdown(LogicalUnit):
     assert self.instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
 
     assert self.instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
 
+    _CheckInstanceState(self, self.instance, INSTANCE_ONLINE)
+
     self.primary_offline = \
       self.cfg.GetNodeInfo(self.instance.primary_node).offline
 
     self.primary_offline = \
       self.cfg.GetNodeInfo(self.instance.primary_node).offline
 
@@ -5904,7 +6649,7 @@ class LUInstanceReinstall(LogicalUnit):
       raise errors.OpPrereqError("Instance '%s' has no disks" %
                                  self.op.instance_name,
                                  errors.ECODE_INVAL)
       raise errors.OpPrereqError("Instance '%s' has no disks" %
                                  self.op.instance_name,
                                  errors.ECODE_INVAL)
-    _CheckInstanceDown(self, instance, "cannot reinstall")
+    _CheckInstanceState(self, instance, INSTANCE_DOWN, msg="cannot reinstall")
 
     if self.op.os_type is not None:
       # OS verification
 
     if self.op.os_type is not None:
       # OS verification
@@ -5941,9 +6686,9 @@ class LUInstanceReinstall(LogicalUnit):
     try:
       feedback_fn("Running the instance OS create scripts...")
       # FIXME: pass debug option from opcode to backend
     try:
       feedback_fn("Running the instance OS create scripts...")
       # FIXME: pass debug option from opcode to backend
-      result = self.rpc.call_instance_os_add(inst.primary_node, inst, True,
-                                             self.op.debug_level,
-                                             osparams=self.os_inst)
+      result = self.rpc.call_instance_os_add(inst.primary_node,
+                                             (inst, self.os_inst), True,
+                                             self.op.debug_level)
       result.Raise("Could not install OS for instance %s on node %s" %
                    (inst.name, inst.primary_node))
     finally:
       result.Raise("Could not install OS for instance %s on node %s" %
                    (inst.name, inst.primary_node))
     finally:
@@ -5977,6 +6722,10 @@ class LUInstanceRecreateDisks(LogicalUnit):
       # otherwise we need to lock all nodes for disk re-creation
       primary_only = bool(self.op.nodes)
       self._LockInstancesNodes(primary_only=primary_only)
       # otherwise we need to lock all nodes for disk re-creation
       primary_only = bool(self.op.nodes)
       self._LockInstancesNodes(primary_only=primary_only)
+    elif level == locking.LEVEL_NODE_RES:
+      # Copy node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE][:]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -6023,10 +6772,12 @@ class LUInstanceRecreateDisks(LogicalUnit):
                                  self.op.instance_name, errors.ECODE_INVAL)
     # if we replace nodes *and* the old primary is offline, we don't
     # check
                                  self.op.instance_name, errors.ECODE_INVAL)
     # if we replace nodes *and* the old primary is offline, we don't
     # check
-    assert instance.primary_node in self.needed_locks[locking.LEVEL_NODE]
+    assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE)
+    assert instance.primary_node in self.owned_locks(locking.LEVEL_NODE_RES)
     old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
     if not (self.op.nodes and old_pnode.offline):
     old_pnode = self.cfg.GetNodeInfo(instance.primary_node)
     if not (self.op.nodes and old_pnode.offline):
-      _CheckInstanceDown(self, instance, "cannot recreate disks")
+      _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+                          msg="cannot recreate disks")
 
     if not self.op.disks:
       self.op.disks = range(len(instance.disks))
 
     if not self.op.disks:
       self.op.disks = range(len(instance.disks))
@@ -6045,31 +6796,47 @@ class LUInstanceRecreateDisks(LogicalUnit):
     """Recreate the disks.
 
     """
     """Recreate the disks.
 
     """
-    # change primary node, if needed
-    if self.op.nodes:
-      self.instance.primary_node = self.op.nodes[0]
-      self.LogWarning("Changing the instance's nodes, you will have to"
-                      " remove any disks left on the older nodes manually")
+    instance = self.instance
+
+    assert (self.owned_locks(locking.LEVEL_NODE) ==
+            self.owned_locks(locking.LEVEL_NODE_RES))
 
     to_skip = []
 
     to_skip = []
-    for idx, disk in enumerate(self.instance.disks):
+    mods = [] # keeps track of needed logical_id changes
+
+    for idx, disk in enumerate(instance.disks):
       if idx not in self.op.disks: # disk idx has not been passed in
         to_skip.append(idx)
         continue
       # update secondaries for disks, if needed
       if self.op.nodes:
         if disk.dev_type == constants.LD_DRBD8:
       if idx not in self.op.disks: # disk idx has not been passed in
         to_skip.append(idx)
         continue
       # update secondaries for disks, if needed
       if self.op.nodes:
         if disk.dev_type == constants.LD_DRBD8:
-          # need to update the nodes
+          # need to update the nodes and minors
           assert len(self.op.nodes) == 2
           assert len(self.op.nodes) == 2
-          logical_id = list(disk.logical_id)
-          logical_id[0] = self.op.nodes[0]
-          logical_id[1] = self.op.nodes[1]
-          disk.logical_id = tuple(logical_id)
+          assert len(disk.logical_id) == 6 # otherwise disk internals
+                                           # have changed
+          (_, _, old_port, _, _, old_secret) = disk.logical_id
+          new_minors = self.cfg.AllocateDRBDMinor(self.op.nodes, instance.name)
+          new_id = (self.op.nodes[0], self.op.nodes[1], old_port,
+                    new_minors[0], new_minors[1], old_secret)
+          assert len(disk.logical_id) == len(new_id)
+          mods.append((idx, new_id))
+
+    # now that we have passed all asserts above, we can apply the mods
+    # in a single run (to avoid partial changes)
+    for idx, new_id in mods:
+      instance.disks[idx].logical_id = new_id
+
+    # change primary node, if needed
+    if self.op.nodes:
+      instance.primary_node = self.op.nodes[0]
+      self.LogWarning("Changing the instance's nodes, you will have to"
+                      " remove any disks left on the older nodes manually")
 
     if self.op.nodes:
 
     if self.op.nodes:
-      self.cfg.Update(self.instance, feedback_fn)
+      self.cfg.Update(instance, feedback_fn)
 
 
-    _CreateDisks(self, self.instance, to_skip=to_skip)
+    _CreateDisks(self, instance, to_skip=to_skip)
 
 
 class LUInstanceRename(LogicalUnit):
 
 
 class LUInstanceRename(LogicalUnit):
@@ -6116,13 +6883,14 @@ class LUInstanceRename(LogicalUnit):
     instance = self.cfg.GetInstanceInfo(self.op.instance_name)
     assert instance is not None
     _CheckNodeOnline(self, instance.primary_node)
     instance = self.cfg.GetInstanceInfo(self.op.instance_name)
     assert instance is not None
     _CheckNodeOnline(self, instance.primary_node)
-    _CheckInstanceDown(self, instance, "cannot rename")
+    _CheckInstanceState(self, instance, INSTANCE_NOT_RUNNING,
+                        msg="cannot rename")
     self.instance = instance
 
     new_name = self.op.new_name
     if self.op.name_check:
       hostname = netutils.GetHostname(name=new_name)
     self.instance = instance
 
     new_name = self.op.new_name
     if self.op.name_check:
       hostname = netutils.GetHostname(name=new_name)
-      if hostname != new_name:
+      if hostname.name != new_name:
         self.LogInfo("Resolved given name '%s' to '%s'", new_name,
                      hostname.name)
       if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
         self.LogInfo("Resolved given name '%s' to '%s'", new_name,
                      hostname.name)
       if not utils.MatchNameComponent(self.op.new_name, [hostname.name]):
@@ -6150,7 +6918,7 @@ class LUInstanceRename(LogicalUnit):
     old_name = inst.name
 
     rename_file_storage = False
     old_name = inst.name
 
     rename_file_storage = False
-    if (inst.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE) and
+    if (inst.disk_template in constants.DTS_FILEBASED and
         self.op.new_name != inst.name):
       old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
       rename_file_storage = True
         self.op.new_name != inst.name):
       old_file_storage_dir = os.path.dirname(inst.disks[0].logical_id[1])
       rename_file_storage = True
@@ -6202,11 +6970,16 @@ class LUInstanceRemove(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
+    self.needed_locks[locking.LEVEL_NODE_RES] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes()
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes()
+    elif level == locking.LEVEL_NODE_RES:
+      # Copy node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE][:]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -6255,6 +7028,12 @@ class LUInstanceRemove(LogicalUnit):
                                  " node %s: %s" %
                                  (instance.name, instance.primary_node, msg))
 
                                  " node %s: %s" %
                                  (instance.name, instance.primary_node, msg))
 
+    assert (self.owned_locks(locking.LEVEL_NODE) ==
+            self.owned_locks(locking.LEVEL_NODE_RES))
+    assert not (set(instance.all_nodes) -
+                self.owned_locks(locking.LEVEL_NODE)), \
+      "Not owning correct locks"
+
     _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
 
 
     _RemoveInstance(self, feedback_fn, instance, self.op.ignore_failures)
 
 
@@ -6284,7 +7063,7 @@ class LUInstanceQuery(NoHooksLU):
   """Logical unit for querying instances.
 
   """
   """Logical unit for querying instances.
 
   """
-  # pylint: disable-msg=W0142
+  # pylint: disable=W0142
   REQ_BGL = False
 
   def CheckArguments(self):
   REQ_BGL = False
 
   def CheckArguments(self):
@@ -6468,11 +7247,16 @@ class LUInstanceMove(LogicalUnit):
     target_node = _ExpandNodeName(self.cfg, self.op.target_node)
     self.op.target_node = target_node
     self.needed_locks[locking.LEVEL_NODE] = [target_node]
     target_node = _ExpandNodeName(self.cfg, self.op.target_node)
     self.op.target_node = target_node
     self.needed_locks[locking.LEVEL_NODE] = [target_node]
+    self.needed_locks[locking.LEVEL_NODE_RES] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes(primary_only=True)
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes(primary_only=True)
+    elif level == locking.LEVEL_NODE_RES:
+      # Copy node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE][:]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -6530,10 +7314,10 @@ class LUInstanceMove(LogicalUnit):
     _CheckNodeNotDrained(self, target_node)
     _CheckNodeVmCapable(self, target_node)
 
     _CheckNodeNotDrained(self, target_node)
     _CheckNodeVmCapable(self, target_node)
 
-    if instance.admin_up:
+    if instance.admin_state == constants.ADMINST_UP:
       # check memory requirements on the secondary node
       _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
       # check memory requirements on the secondary node
       _CheckNodeFreeMemory(self, target_node, "failing over instance %s" %
-                           instance.name, bep[constants.BE_MEMORY],
+                           instance.name, bep[constants.BE_MAXMEM],
                            instance.hypervisor)
     else:
       self.LogInfo("Not checking memory on the secondary node as"
                            instance.hypervisor)
     else:
       self.LogInfo("Not checking memory on the secondary node as"
@@ -6557,6 +7341,9 @@ class LUInstanceMove(LogicalUnit):
     self.LogInfo("Shutting down instance %s on source node %s",
                  instance.name, source_node)
 
     self.LogInfo("Shutting down instance %s on source node %s",
                  instance.name, source_node)
 
+    assert (self.owned_locks(locking.LEVEL_NODE) ==
+            self.owned_locks(locking.LEVEL_NODE_RES))
+
     result = self.rpc.call_instance_shutdown(source_node, instance,
                                              self.op.shutdown_timeout)
     msg = result.fail_msg
     result = self.rpc.call_instance_shutdown(source_node, instance,
                                              self.op.shutdown_timeout)
     msg = result.fail_msg
@@ -6621,7 +7408,7 @@ class LUInstanceMove(LogicalUnit):
     _RemoveDisks(self, instance, target_node=source_node)
 
     # Only start the instance if it's marked as up
     _RemoveDisks(self, instance, target_node=source_node)
 
     # Only start the instance if it's marked as up
-    if instance.admin_up:
+    if instance.admin_state == constants.ADMINST_UP:
       self.LogInfo("Starting instance %s on node %s",
                    instance.name, target_node)
 
       self.LogInfo("Starting instance %s on node %s",
                    instance.name, target_node)
 
@@ -6631,7 +7418,8 @@ class LUInstanceMove(LogicalUnit):
         _ShutdownInstanceDisks(self, instance)
         raise errors.OpExecError("Can't activate the instance's disks")
 
         _ShutdownInstanceDisks(self, instance)
         raise errors.OpExecError("Can't activate the instance's disks")
 
-      result = self.rpc.call_instance_start(target_node, instance, None, None)
+      result = self.rpc.call_instance_start(target_node,
+                                            (instance, None, None), False)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
       msg = result.fail_msg
       if msg:
         _ShutdownInstanceDisks(self, instance)
@@ -6653,7 +7441,7 @@ class LUNodeMigrate(LogicalUnit):
   def ExpandNames(self):
     self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
 
   def ExpandNames(self):
     self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
 
-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+    self.share_locks = _ShareAll()
     self.needed_locks = {
       locking.LEVEL_NODE: [self.op.node_name],
       }
     self.needed_locks = {
       locking.LEVEL_NODE: [self.op.node_name],
       }
@@ -6694,7 +7482,7 @@ class LUNodeMigrate(LogicalUnit):
     # running the iallocator and the actual migration, a good consistency model
     # will have to be found.
 
     # running the iallocator and the actual migration, a good consistency model
     # will have to be found.
 
-    assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
             frozenset([self.op.node_name]))
 
     return ResultWithJobs(jobs)
             frozenset([self.op.node_name]))
 
     return ResultWithJobs(jobs)
@@ -6724,6 +7512,11 @@ class TLMigrateInstance(Tasklet):
   @ivar shutdown_timeout: In case of failover timeout of the shutdown
 
   """
   @ivar shutdown_timeout: In case of failover timeout of the shutdown
 
   """
+
+  # Constants
+  _MIGRATION_POLL_INTERVAL = 1      # seconds
+  _MIGRATION_FEEDBACK_INTERVAL = 10 # seconds
+
   def __init__(self, lu, instance_name, cleanup=False,
                failover=False, fallback=False,
                ignore_consistency=False,
   def __init__(self, lu, instance_name, cleanup=False,
                failover=False, fallback=False,
                ignore_consistency=False,
@@ -6753,10 +7546,11 @@ class TLMigrateInstance(Tasklet):
     assert instance is not None
     self.instance = instance
 
     assert instance is not None
     self.instance = instance
 
-    if (not self.cleanup and not instance.admin_up and not self.failover and
-        self.fallback):
-      self.lu.LogInfo("Instance is marked down, fallback allowed, switching"
-                      " to failover")
+    if (not self.cleanup and
+        not instance.admin_state == constants.ADMINST_UP and
+        not self.failover and self.fallback):
+      self.lu.LogInfo("Instance is marked down or offline, fallback allowed,"
+                      " switching to failover")
       self.failover = True
 
     if instance.disk_template not in constants.DTS_MIRRORED:
       self.failover = True
 
     if instance.disk_template not in constants.DTS_MIRRORED:
@@ -6815,14 +7609,26 @@ class TLMigrateInstance(Tasklet):
     i_be = self.cfg.GetClusterInfo().FillBE(instance)
 
     # check memory requirements on the secondary node
     i_be = self.cfg.GetClusterInfo().FillBE(instance)
 
     # check memory requirements on the secondary node
-    if not self.failover or instance.admin_up:
+    if not self.failover or instance.admin_state == constants.ADMINST_UP:
       _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
       _CheckNodeFreeMemory(self.lu, target_node, "migrating instance %s" %
-                           instance.name, i_be[constants.BE_MEMORY],
+                           instance.name, i_be[constants.BE_MAXMEM],
                            instance.hypervisor)
     else:
       self.lu.LogInfo("Not checking memory on the secondary node as"
                       " instance will not be started")
 
                            instance.hypervisor)
     else:
       self.lu.LogInfo("Not checking memory on the secondary node as"
                       " instance will not be started")
 
+    # check if failover must be forced instead of migration
+    if (not self.cleanup and not self.failover and
+        i_be[constants.BE_ALWAYS_FAILOVER]):
+      if self.fallback:
+        self.lu.LogInfo("Instance configured to always failover; fallback"
+                        " to failover")
+        self.failover = True
+      else:
+        raise errors.OpPrereqError("This instance has been configured to"
+                                   " always failover, please allow failover",
+                                   errors.ECODE_STATE)
+
     # check bridge existance
     _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
 
     # check bridge existance
     _CheckInstanceBridgesExist(self.lu, instance, node=target_node)
 
@@ -7047,12 +7853,13 @@ class TLMigrateInstance(Tasklet):
     """
     instance = self.instance
     target_node = self.target_node
     """
     instance = self.instance
     target_node = self.target_node
+    source_node = self.source_node
     migration_info = self.migration_info
 
     migration_info = self.migration_info
 
-    abort_result = self.rpc.call_finalize_migration(target_node,
-                                                    instance,
-                                                    migration_info,
-                                                    False)
+    abort_result = self.rpc.call_instance_finalize_migration_dst(target_node,
+                                                                 instance,
+                                                                 migration_info,
+                                                                 False)
     abort_msg = abort_result.fail_msg
     if abort_msg:
       logging.error("Aborting migration failed on target node %s: %s",
     abort_msg = abort_result.fail_msg
     if abort_msg:
       logging.error("Aborting migration failed on target node %s: %s",
@@ -7060,6 +7867,13 @@ class TLMigrateInstance(Tasklet):
       # Don't raise an exception here, as we stil have to try to revert the
       # disk status, even if this step failed.
 
       # Don't raise an exception here, as we stil have to try to revert the
       # disk status, even if this step failed.
 
+    abort_result = self.rpc.call_instance_finalize_migration_src(source_node,
+        instance, False, self.live)
+    abort_msg = abort_result.fail_msg
+    if abort_msg:
+      logging.error("Aborting migration failed on source node %s: %s",
+                    source_node, abort_msg)
+
   def _ExecMigration(self):
     """Migrate an instance.
 
   def _ExecMigration(self):
     """Migrate an instance.
 
@@ -7076,6 +7890,24 @@ class TLMigrateInstance(Tasklet):
     target_node = self.target_node
     source_node = self.source_node
 
     target_node = self.target_node
     source_node = self.source_node
 
+    # Check for hypervisor version mismatch and warn the user.
+    nodeinfo = self.rpc.call_node_info([source_node, target_node],
+                                       None, [self.instance.hypervisor])
+    for ninfo in nodeinfo.values():
+      ninfo.Raise("Unable to retrieve node information from node '%s'" %
+                  ninfo.node)
+    (_, _, (src_info, )) = nodeinfo[source_node].payload
+    (_, _, (dst_info, )) = nodeinfo[target_node].payload
+
+    if ((constants.HV_NODEINFO_KEY_VERSION in src_info) and
+        (constants.HV_NODEINFO_KEY_VERSION in dst_info)):
+      src_version = src_info[constants.HV_NODEINFO_KEY_VERSION]
+      dst_version = dst_info[constants.HV_NODEINFO_KEY_VERSION]
+      if src_version != dst_version:
+        self.feedback_fn("* warning: hypervisor version mismatch between"
+                         " source (%s) and target (%s) node" %
+                         (src_version, dst_version))
+
     self.feedback_fn("* checking disk consistency between source and target")
     for dev in instance.disks:
       if not _CheckDiskConsistency(self.lu, dev, target_node, False):
     self.feedback_fn("* checking disk consistency between source and target")
     for dev in instance.disks:
       if not _CheckDiskConsistency(self.lu, dev, target_node, False):
@@ -7131,18 +7963,59 @@ class TLMigrateInstance(Tasklet):
       raise errors.OpExecError("Could not migrate instance %s: %s" %
                                (instance.name, msg))
 
       raise errors.OpExecError("Could not migrate instance %s: %s" %
                                (instance.name, msg))
 
+    self.feedback_fn("* starting memory transfer")
+    last_feedback = time.time()
+    while True:
+      result = self.rpc.call_instance_get_migration_status(source_node,
+                                                           instance)
+      msg = result.fail_msg
+      ms = result.payload   # MigrationStatus instance
+      if msg or (ms.status in constants.HV_MIGRATION_FAILED_STATUSES):
+        logging.error("Instance migration failed, trying to revert"
+                      " disk status: %s", msg)
+        self.feedback_fn("Migration failed, aborting")
+        self._AbortMigration()
+        self._RevertDiskStatus()
+        raise errors.OpExecError("Could not migrate instance %s: %s" %
+                                 (instance.name, msg))
+
+      if result.payload.status != constants.HV_MIGRATION_ACTIVE:
+        self.feedback_fn("* memory transfer complete")
+        break
+
+      if (utils.TimeoutExpired(last_feedback,
+                               self._MIGRATION_FEEDBACK_INTERVAL) and
+          ms.transferred_ram is not None):
+        mem_progress = 100 * float(ms.transferred_ram) / float(ms.total_ram)
+        self.feedback_fn("* memory transfer progress: %.2f %%" % mem_progress)
+        last_feedback = time.time()
+
+      time.sleep(self._MIGRATION_POLL_INTERVAL)
+
+    result = self.rpc.call_instance_finalize_migration_src(source_node,
+                                                           instance,
+                                                           True,
+                                                           self.live)
+    msg = result.fail_msg
+    if msg:
+      logging.error("Instance migration succeeded, but finalization failed"
+                    " on the source node: %s", msg)
+      raise errors.OpExecError("Could not finalize instance migration: %s" %
+                               msg)
+
     instance.primary_node = target_node
     instance.primary_node = target_node
+
     # distribute new instance config to the other nodes
     self.cfg.Update(instance, self.feedback_fn)
 
     # distribute new instance config to the other nodes
     self.cfg.Update(instance, self.feedback_fn)
 
-    result = self.rpc.call_finalize_migration(target_node,
-                                              instance,
-                                              migration_info,
-                                              True)
+    result = self.rpc.call_instance_finalize_migration_dst(target_node,
+                                                           instance,
+                                                           migration_info,
+                                                           True)
     msg = result.fail_msg
     if msg:
     msg = result.fail_msg
     if msg:
-      logging.error("Instance migration succeeded, but finalization failed:"
-                    " %s", msg)
+      logging.error("Instance migration succeeded, but finalization failed"
+                    " on the target node: %s", msg)
       raise errors.OpExecError("Could not finalize instance migration: %s" %
                                msg)
 
       raise errors.OpExecError("Could not finalize instance migration: %s" %
                                msg)
 
@@ -7168,12 +8041,16 @@ class TLMigrateInstance(Tasklet):
     source_node = instance.primary_node
     target_node = self.target_node
 
     source_node = instance.primary_node
     target_node = self.target_node
 
-    if instance.admin_up:
+    if instance.admin_state == constants.ADMINST_UP:
       self.feedback_fn("* checking disk consistency between source and target")
       for dev in instance.disks:
         # for drbd, these are drbd over lvm
       self.feedback_fn("* checking disk consistency between source and target")
       for dev in instance.disks:
         # for drbd, these are drbd over lvm
-        if not _CheckDiskConsistency(self, dev, target_node, False):
-          if not self.ignore_consistency:
+        if not _CheckDiskConsistency(self.lu, dev, target_node, False):
+          if primary_node.offline:
+            self.feedback_fn("Node %s is offline, ignoring degraded disk %s on"
+                             " target node %s" %
+                             (primary_node.name, dev.iv_name, target_node))
+          elif not self.ignore_consistency:
             raise errors.OpExecError("Disk %s is degraded on target node,"
                                      " aborting failover" % dev.iv_name)
     else:
             raise errors.OpExecError("Disk %s is degraded on target node,"
                                      " aborting failover" % dev.iv_name)
     else:
@@ -7199,30 +8076,33 @@ class TLMigrateInstance(Tasklet):
                                  (instance.name, source_node, msg))
 
     self.feedback_fn("* deactivating the instance's disks on source node")
                                  (instance.name, source_node, msg))
 
     self.feedback_fn("* deactivating the instance's disks on source node")
-    if not _ShutdownInstanceDisks(self, instance, ignore_primary=True):
-      raise errors.OpExecError("Can't shut down the instance's disks.")
+    if not _ShutdownInstanceDisks(self.lu, instance, ignore_primary=True):
+      raise errors.OpExecError("Can't shut down the instance's disks")
 
     instance.primary_node = target_node
     # distribute new instance config to the other nodes
     self.cfg.Update(instance, self.feedback_fn)
 
     # Only start the instance if it's marked as up
 
     instance.primary_node = target_node
     # distribute new instance config to the other nodes
     self.cfg.Update(instance, self.feedback_fn)
 
     # Only start the instance if it's marked as up
-    if instance.admin_up:
-      self.feedback_fn("* activating the instance's disks on target node")
+    if instance.admin_state == constants.ADMINST_UP:
+      self.feedback_fn("* activating the instance's disks on target node %s" %
+                       target_node)
       logging.info("Starting instance %s on node %s",
                    instance.name, target_node)
 
       logging.info("Starting instance %s on node %s",
                    instance.name, target_node)
 
-      disks_ok, _ = _AssembleInstanceDisks(self, instance,
+      disks_ok, _ = _AssembleInstanceDisks(self.lu, instance,
                                            ignore_secondaries=True)
       if not disks_ok:
                                            ignore_secondaries=True)
       if not disks_ok:
-        _ShutdownInstanceDisks(self, instance)
+        _ShutdownInstanceDisks(self.lu, instance)
         raise errors.OpExecError("Can't activate the instance's disks")
 
         raise errors.OpExecError("Can't activate the instance's disks")
 
-      self.feedback_fn("* starting the instance on the target node")
-      result = self.rpc.call_instance_start(target_node, instance, None, None)
+      self.feedback_fn("* starting the instance on the target node %s" %
+                       target_node)
+      result = self.rpc.call_instance_start(target_node, (instance, None, None),
+                                            False)
       msg = result.fail_msg
       if msg:
       msg = result.fail_msg
       if msg:
-        _ShutdownInstanceDisks(self, instance)
+        _ShutdownInstanceDisks(self.lu, instance)
         raise errors.OpExecError("Could not start instance %s on node %s: %s" %
                                  (instance.name, target_node, msg))
 
         raise errors.OpExecError("Could not start instance %s on node %s: %s" %
                                  (instance.name, target_node, msg))
 
@@ -7240,10 +8120,8 @@ class TLMigrateInstance(Tasklet):
       # directly, or through an iallocator.
 
     self.all_nodes = [self.source_node, self.target_node]
       # directly, or through an iallocator.
 
     self.all_nodes = [self.source_node, self.target_node]
-    self.nodes_ip = {
-      self.source_node: self.cfg.GetNodeInfo(self.source_node).secondary_ip,
-      self.target_node: self.cfg.GetNodeInfo(self.target_node).secondary_ip,
-      }
+    self.nodes_ip = dict((name, node.secondary_ip) for (name, node)
+                         in self.cfg.GetMultiNodeInfo(self.all_nodes))
 
     if self.failover:
       feedback_fn("Failover instance %s" % self.instance.name)
 
     if self.failover:
       feedback_fn("Failover instance %s" % self.instance.name)
@@ -7342,24 +8220,104 @@ def _GenerateUniqueNames(lu, exts):
   return results
 
 
   return results
 
 
+def _ComputeLDParams(disk_template, disk_params):
+  """Computes Logical Disk parameters from Disk Template parameters.
+
+  @type disk_template: string
+  @param disk_template: disk template, one of L{constants.DISK_TEMPLATES}
+  @type disk_params: dict
+  @param disk_params: disk template parameters; dict(template_name -> parameters
+  @rtype: list(dict)
+  @return: a list of dicts, one for each node of the disk hierarchy. Each dict
+    contains the LD parameters of the node. The tree is flattened in-order.
+
+  """
+  if disk_template not in constants.DISK_TEMPLATES:
+    raise errors.ProgrammerError("Unknown disk template %s" % disk_template)
+
+  result = list()
+  dt_params = disk_params[disk_template]
+  if disk_template == constants.DT_DRBD8:
+    drbd_params = {
+      constants.LDP_RESYNC_RATE: dt_params[constants.DRBD_RESYNC_RATE],
+      constants.LDP_BARRIERS: dt_params[constants.DRBD_DISK_BARRIERS],
+      constants.LDP_NO_META_FLUSH: dt_params[constants.DRBD_META_BARRIERS],
+      constants.LDP_DEFAULT_METAVG: dt_params[constants.DRBD_DEFAULT_METAVG],
+      constants.LDP_DISK_CUSTOM: dt_params[constants.DRBD_DISK_CUSTOM],
+      constants.LDP_NET_CUSTOM: dt_params[constants.DRBD_NET_CUSTOM],
+      constants.LDP_DYNAMIC_RESYNC: dt_params[constants.DRBD_DYNAMIC_RESYNC],
+      constants.LDP_PLAN_AHEAD: dt_params[constants.DRBD_PLAN_AHEAD],
+      constants.LDP_FILL_TARGET: dt_params[constants.DRBD_FILL_TARGET],
+      constants.LDP_DELAY_TARGET: dt_params[constants.DRBD_DELAY_TARGET],
+      constants.LDP_MAX_RATE: dt_params[constants.DRBD_MAX_RATE],
+      constants.LDP_MIN_RATE: dt_params[constants.DRBD_MIN_RATE],
+      }
+
+    drbd_params = \
+      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_DRBD8],
+                       drbd_params)
+
+    result.append(drbd_params)
+
+    # data LV
+    data_params = {
+      constants.LDP_STRIPES: dt_params[constants.DRBD_DATA_STRIPES],
+      }
+    data_params = \
+      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+                       data_params)
+    result.append(data_params)
+
+    # metadata LV
+    meta_params = {
+      constants.LDP_STRIPES: dt_params[constants.DRBD_META_STRIPES],
+      }
+    meta_params = \
+      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+                       meta_params)
+    result.append(meta_params)
+
+  elif (disk_template == constants.DT_FILE or
+        disk_template == constants.DT_SHARED_FILE):
+    result.append(constants.DISK_LD_DEFAULTS[constants.LD_FILE])
+
+  elif disk_template == constants.DT_PLAIN:
+    params = {
+      constants.LDP_STRIPES: dt_params[constants.LV_STRIPES],
+      }
+    params = \
+      objects.FillDict(constants.DISK_LD_DEFAULTS[constants.LD_LV],
+                       params)
+    result.append(params)
+
+  elif disk_template == constants.DT_BLOCK:
+    result.append(constants.DISK_LD_DEFAULTS[constants.LD_BLOCKDEV])
+
+  return result
+
+
 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
 def _GenerateDRBD8Branch(lu, primary, secondary, size, vgnames, names,
-                         iv_name, p_minor, s_minor):
+                         iv_name, p_minor, s_minor, drbd_params, data_params,
+                         meta_params):
   """Generate a drbd8 device complete with its children.
 
   """
   assert len(vgnames) == len(names) == 2
   port = lu.cfg.AllocatePort()
   shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
   """Generate a drbd8 device complete with its children.
 
   """
   assert len(vgnames) == len(names) == 2
   port = lu.cfg.AllocatePort()
   shared_secret = lu.cfg.GenerateDRBDSecret(lu.proc.GetECId())
+
   dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
   dev_data = objects.Disk(dev_type=constants.LD_LV, size=size,
-                          logical_id=(vgnames[0], names[0]))
-  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
-                          logical_id=(vgnames[1], names[1]))
+                          logical_id=(vgnames[0], names[0]),
+                          params=data_params)
+  dev_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
+                          logical_id=(vgnames[1], names[1]),
+                          params=meta_params)
   drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
                           logical_id=(primary, secondary, port,
                                       p_minor, s_minor,
                                       shared_secret),
                           children=[dev_data, dev_meta],
   drbd_dev = objects.Disk(dev_type=constants.LD_DRBD8, size=size,
                           logical_id=(primary, secondary, port,
                                       p_minor, s_minor,
                                       shared_secret),
                           children=[dev_data, dev_meta],
-                          iv_name=iv_name)
+                          iv_name=iv_name, params=drbd_params)
   return drbd_dev
 
 
   return drbd_dev
 
 
@@ -7367,7 +8325,7 @@ def _GenerateDiskTemplate(lu, template_name,
                           instance_name, primary_node,
                           secondary_nodes, disk_info,
                           file_storage_dir, file_driver,
                           instance_name, primary_node,
                           secondary_nodes, disk_info,
                           file_storage_dir, file_driver,
-                          base_index, feedback_fn):
+                          base_index, feedback_fn, disk_params):
   """Generate the entire disk layout for a given template type.
 
   """
   """Generate the entire disk layout for a given template type.
 
   """
@@ -7376,6 +8334,7 @@ def _GenerateDiskTemplate(lu, template_name,
   vgname = lu.cfg.GetVGName()
   disk_count = len(disk_info)
   disks = []
   vgname = lu.cfg.GetVGName()
   disk_count = len(disk_info)
   disks = []
+  ld_params = _ComputeLDParams(template_name, disk_params)
   if template_name == constants.DT_DISKLESS:
     pass
   elif template_name == constants.DT_PLAIN:
   if template_name == constants.DT_DISKLESS:
     pass
   elif template_name == constants.DT_PLAIN:
@@ -7392,9 +8351,11 @@ def _GenerateDiskTemplate(lu, template_name,
                               size=disk[constants.IDISK_SIZE],
                               logical_id=(vg, names[idx]),
                               iv_name="disk/%d" % disk_index,
                               size=disk[constants.IDISK_SIZE],
                               logical_id=(vg, names[idx]),
                               iv_name="disk/%d" % disk_index,
-                              mode=disk[constants.IDISK_MODE])
+                              mode=disk[constants.IDISK_MODE],
+                              params=ld_params[0])
       disks.append(disk_dev)
   elif template_name == constants.DT_DRBD8:
       disks.append(disk_dev)
   elif template_name == constants.DT_DRBD8:
+    drbd_params, data_params, meta_params = ld_params
     if len(secondary_nodes) != 1:
       raise errors.ProgrammerError("Wrong template configuration")
     remote_node = secondary_nodes[0]
     if len(secondary_nodes) != 1:
       raise errors.ProgrammerError("Wrong template configuration")
     remote_node = secondary_nodes[0]
@@ -7408,14 +8369,16 @@ def _GenerateDiskTemplate(lu, template_name,
       names.append(lv_prefix + "_meta")
     for idx, disk in enumerate(disk_info):
       disk_index = idx + base_index
       names.append(lv_prefix + "_meta")
     for idx, disk in enumerate(disk_info):
       disk_index = idx + base_index
+      drbd_default_metavg = drbd_params[constants.LDP_DEFAULT_METAVG]
       data_vg = disk.get(constants.IDISK_VG, vgname)
       data_vg = disk.get(constants.IDISK_VG, vgname)
-      meta_vg = disk.get(constants.IDISK_METAVG, data_vg)
+      meta_vg = disk.get(constants.IDISK_METAVG, drbd_default_metavg)
       disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
                                       disk[constants.IDISK_SIZE],
                                       [data_vg, meta_vg],
                                       names[idx * 2:idx * 2 + 2],
                                       "disk/%d" % disk_index,
       disk_dev = _GenerateDRBD8Branch(lu, primary_node, remote_node,
                                       disk[constants.IDISK_SIZE],
                                       [data_vg, meta_vg],
                                       names[idx * 2:idx * 2 + 2],
                                       "disk/%d" % disk_index,
-                                      minors[idx * 2], minors[idx * 2 + 1])
+                                      minors[idx * 2], minors[idx * 2 + 1],
+                                      drbd_params, data_params, meta_params)
       disk_dev.mode = disk[constants.IDISK_MODE]
       disks.append(disk_dev)
   elif template_name == constants.DT_FILE:
       disk_dev.mode = disk[constants.IDISK_MODE]
       disks.append(disk_dev)
   elif template_name == constants.DT_FILE:
@@ -7432,7 +8395,8 @@ def _GenerateDiskTemplate(lu, template_name,
                               logical_id=(file_driver,
                                           "%s/disk%d" % (file_storage_dir,
                                                          disk_index)),
                               logical_id=(file_driver,
                                           "%s/disk%d" % (file_storage_dir,
                                                          disk_index)),
-                              mode=disk[constants.IDISK_MODE])
+                              mode=disk[constants.IDISK_MODE],
+                              params=ld_params[0])
       disks.append(disk_dev)
   elif template_name == constants.DT_SHARED_FILE:
     if len(secondary_nodes) != 0:
       disks.append(disk_dev)
   elif template_name == constants.DT_SHARED_FILE:
     if len(secondary_nodes) != 0:
@@ -7448,7 +8412,8 @@ def _GenerateDiskTemplate(lu, template_name,
                               logical_id=(file_driver,
                                           "%s/disk%d" % (file_storage_dir,
                                                          disk_index)),
                               logical_id=(file_driver,
                                           "%s/disk%d" % (file_storage_dir,
                                                          disk_index)),
-                              mode=disk[constants.IDISK_MODE])
+                              mode=disk[constants.IDISK_MODE],
+                              params=ld_params[0])
       disks.append(disk_dev)
   elif template_name == constants.DT_BLOCK:
     if len(secondary_nodes) != 0:
       disks.append(disk_dev)
   elif template_name == constants.DT_BLOCK:
     if len(secondary_nodes) != 0:
@@ -7461,7 +8426,8 @@ def _GenerateDiskTemplate(lu, template_name,
                               logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
                                           disk[constants.IDISK_ADOPT]),
                               iv_name="disk/%d" % disk_index,
                               logical_id=(constants.BLOCKDEV_DRIVER_MANUAL,
                                           disk[constants.IDISK_ADOPT]),
                               iv_name="disk/%d" % disk_index,
-                              mode=disk[constants.IDISK_MODE])
+                              mode=disk[constants.IDISK_MODE],
+                              params=ld_params[0])
       disks.append(disk_dev)
 
   else:
       disks.append(disk_dev)
 
   else:
@@ -7583,7 +8549,7 @@ def _CreateDisks(lu, instance, to_skip=None, target_node=None):
     pnode = target_node
     all_nodes = [pnode]
 
     pnode = target_node
     all_nodes = [pnode]
 
-  if instance.disk_template in (constants.DT_FILE, constants.DT_SHARED_FILE):
+  if instance.disk_template in constants.DTS_FILEBASED:
     file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
     result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
 
     file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
     result = lu.rpc.call_file_storage_dir_create(pnode, file_storage_dir)
 
@@ -7637,6 +8603,11 @@ def _RemoveDisks(lu, instance, target_node=None):
                       " continuing anyway: %s", device.iv_name, node, msg)
         all_result = False
 
                       " continuing anyway: %s", device.iv_name, node, msg)
         all_result = False
 
+    # if this is a DRBD disk, return its port to the pool
+    if device.dev_type in constants.LDS_DRBD:
+      tcp_port = device.logical_id[2]
+      lu.cfg.AddTcpUdpPort(tcp_port)
+
   if instance.disk_template == constants.DT_FILE:
     file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
     if target_node:
   if instance.disk_template == constants.DT_FILE:
     file_storage_dir = os.path.dirname(instance.disks[0].logical_id[1])
     if target_node:
@@ -7672,14 +8643,14 @@ def _ComputeDiskSizePerVG(disk_template, disks):
     constants.DT_DISKLESS: {},
     constants.DT_PLAIN: _compute(disks, 0),
     # 128 MB are added for drbd metadata for each disk
     constants.DT_DISKLESS: {},
     constants.DT_PLAIN: _compute(disks, 0),
     # 128 MB are added for drbd metadata for each disk
-    constants.DT_DRBD8: _compute(disks, 128),
+    constants.DT_DRBD8: _compute(disks, DRBD_META_SIZE),
     constants.DT_FILE: {},
     constants.DT_SHARED_FILE: {},
   }
 
   if disk_template not in req_size_dict:
     raise errors.ProgrammerError("Disk template '%s' size requirement"
     constants.DT_FILE: {},
     constants.DT_SHARED_FILE: {},
   }
 
   if disk_template not in req_size_dict:
     raise errors.ProgrammerError("Disk template '%s' size requirement"
-                                 " is unknown" %  disk_template)
+                                 " is unknown" % disk_template)
 
   return req_size_dict[disk_template]
 
 
   return req_size_dict[disk_template]
 
@@ -7693,7 +8664,8 @@ def _ComputeDiskSize(disk_template, disks):
     constants.DT_DISKLESS: None,
     constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
     # 128 MB are added for drbd metadata for each disk
     constants.DT_DISKLESS: None,
     constants.DT_PLAIN: sum(d[constants.IDISK_SIZE] for d in disks),
     # 128 MB are added for drbd metadata for each disk
-    constants.DT_DRBD8: sum(d[constants.IDISK_SIZE] + 128 for d in disks),
+    constants.DT_DRBD8:
+      sum(d[constants.IDISK_SIZE] + DRBD_META_SIZE for d in disks),
     constants.DT_FILE: None,
     constants.DT_SHARED_FILE: 0,
     constants.DT_BLOCK: 0,
     constants.DT_FILE: None,
     constants.DT_SHARED_FILE: 0,
     constants.DT_BLOCK: 0,
@@ -7701,7 +8673,7 @@ def _ComputeDiskSize(disk_template, disks):
 
   if disk_template not in req_size_dict:
     raise errors.ProgrammerError("Disk template '%s' size requirement"
 
   if disk_template not in req_size_dict:
     raise errors.ProgrammerError("Disk template '%s' size requirement"
-                                 " is unknown" %  disk_template)
+                                 " is unknown" % disk_template)
 
   return req_size_dict[disk_template]
 
 
   return req_size_dict[disk_template]
 
@@ -7739,9 +8711,11 @@ def _CheckHVParams(lu, nodenames, hvname, hvparams):
 
   """
   nodenames = _FilterVmNodes(lu, nodenames)
 
   """
   nodenames = _FilterVmNodes(lu, nodenames)
-  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames,
-                                                  hvname,
-                                                  hvparams)
+
+  cluster = lu.cfg.GetClusterInfo()
+  hvfull = objects.FillDict(cluster.hvparams.get(hvname, {}), hvparams)
+
+  hvinfo = lu.rpc.call_hypervisor_validate_params(nodenames, hvname, hvfull)
   for node in nodenames:
     info = hvinfo[node]
     if info.offline:
   for node in nodenames:
     info = hvinfo[node]
     if info.offline:
@@ -7767,7 +8741,7 @@ def _CheckOSParams(lu, required, nodenames, osname, osparams):
 
   """
   nodenames = _FilterVmNodes(lu, nodenames)
 
   """
   nodenames = _FilterVmNodes(lu, nodenames)
-  result = lu.rpc.call_os_validate(required, nodenames, osname,
+  result = lu.rpc.call_os_validate(nodenames, required, osname,
                                    [constants.OS_VALIDATE_PARAMETERS],
                                    osparams)
   for node, nres in result.items():
                                    [constants.OS_VALIDATE_PARAMETERS],
                                    osparams)
   for node, nres in result.items():
@@ -7856,9 +8830,10 @@ class LUInstanceCreate(LogicalUnit):
       raise errors.OpPrereqError("Invalid file driver name '%s'" %
                                  self.op.file_driver, errors.ECODE_INVAL)
 
       raise errors.OpPrereqError("Invalid file driver name '%s'" %
                                  self.op.file_driver, errors.ECODE_INVAL)
 
-    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
-      raise errors.OpPrereqError("File storage directory path not absolute",
-                                 errors.ECODE_INVAL)
+    if self.op.disk_template == constants.DT_FILE:
+      opcodes.RequireFileStorage()
+    elif self.op.disk_template == constants.DT_SHARED_FILE:
+      opcodes.RequireSharedFileStorage()
 
     ### Node/iallocator related checks
     _CheckIAllocatorOrNode(self, "iallocator", "pnode")
 
     ### Node/iallocator related checks
     _CheckIAllocatorOrNode(self, "iallocator", "pnode")
@@ -7959,7 +8934,11 @@ class LUInstanceCreate(LogicalUnit):
     self.add_locks[locking.LEVEL_INSTANCE] = instance_name
 
     if self.op.iallocator:
     self.add_locks[locking.LEVEL_INSTANCE] = instance_name
 
     if self.op.iallocator:
+      # TODO: Find a solution to not lock all nodes in the cluster, e.g. by
+      # specifying a group on instance creation and then selecting nodes from
+      # that group
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+      self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
     else:
       self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
       nodelist = [self.op.pnode]
     else:
       self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
       nodelist = [self.op.pnode]
@@ -7967,6 +8946,9 @@ class LUInstanceCreate(LogicalUnit):
         self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
         nodelist.append(self.op.snode)
       self.needed_locks[locking.LEVEL_NODE] = nodelist
         self.op.snode = _ExpandNodeName(self.cfg, self.op.snode)
         nodelist.append(self.op.snode)
       self.needed_locks[locking.LEVEL_NODE] = nodelist
+      # Lock resources of instance's primary and secondary nodes (copy to
+      # prevent accidential modification)
+      self.needed_locks[locking.LEVEL_NODE_RES] = list(nodelist)
 
     # in case of import lock the source node too
     if self.op.mode == constants.INSTANCE_IMPORT:
 
     # in case of import lock the source node too
     if self.op.mode == constants.INSTANCE_IMPORT:
@@ -7980,8 +8962,8 @@ class LUInstanceCreate(LogicalUnit):
         self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
         self.op.src_node = None
         if os.path.isabs(src_path):
         self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
         self.op.src_node = None
         if os.path.isabs(src_path):
-          raise errors.OpPrereqError("Importing an instance from an absolute"
-                                     " path requires a source node option",
+          raise errors.OpPrereqError("Importing an instance from a path"
+                                     " requires a source node option",
                                      errors.ECODE_INVAL)
       else:
         self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
                                      errors.ECODE_INVAL)
       else:
         self.op.src_node = src_node = _ExpandNodeName(self.cfg, src_node)
@@ -8003,7 +8985,7 @@ class LUInstanceCreate(LogicalUnit):
                      tags=self.op.tags,
                      os=self.op.os_type,
                      vcpus=self.be_full[constants.BE_VCPUS],
                      tags=self.op.tags,
                      os=self.op.os_type,
                      vcpus=self.be_full[constants.BE_VCPUS],
-                     memory=self.be_full[constants.BE_MEMORY],
+                     memory=self.be_full[constants.BE_MAXMEM],
                      disks=self.disks,
                      nics=nics,
                      hypervisor=self.op.hypervisor,
                      disks=self.disks,
                      nics=nics,
                      hypervisor=self.op.hypervisor,
@@ -8048,7 +9030,8 @@ class LUInstanceCreate(LogicalUnit):
       secondary_nodes=self.secondaries,
       status=self.op.start,
       os_type=self.op.os_type,
       secondary_nodes=self.secondaries,
       status=self.op.start,
       os_type=self.op.os_type,
-      memory=self.be_full[constants.BE_MEMORY],
+      minmem=self.be_full[constants.BE_MINMEM],
+      maxmem=self.be_full[constants.BE_MAXMEM],
       vcpus=self.be_full[constants.BE_VCPUS],
       nics=_NICListToTuple(self, self.nics),
       disk_template=self.op.disk_template,
       vcpus=self.be_full[constants.BE_VCPUS],
       nics=_NICListToTuple(self, self.nics),
       disk_template=self.op.disk_template,
@@ -8057,6 +9040,7 @@ class LUInstanceCreate(LogicalUnit):
       bep=self.be_full,
       hvp=self.hv_full,
       hypervisor_name=self.op.hypervisor,
       bep=self.be_full,
       hvp=self.hv_full,
       hypervisor_name=self.op.hypervisor,
+      tags=self.op.tags,
     ))
 
     return env
     ))
 
     return env
@@ -8083,7 +9067,7 @@ class LUInstanceCreate(LogicalUnit):
     src_path = self.op.src_path
 
     if src_node is None:
     src_path = self.op.src_path
 
     if src_node is None:
-      locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
+      locked_nodes = self.owned_locks(locking.LEVEL_NODE)
       exp_list = self.rpc.call_export_list(locked_nodes)
       found = False
       for node in exp_list:
       exp_list = self.rpc.call_export_list(locked_nodes)
       found = False
       for node in exp_list:
@@ -8129,38 +9113,48 @@ class LUInstanceCreate(LogicalUnit):
       if einfo.has_option(constants.INISECT_INS, "disk_template"):
         self.op.disk_template = einfo.get(constants.INISECT_INS,
                                           "disk_template")
       if einfo.has_option(constants.INISECT_INS, "disk_template"):
         self.op.disk_template = einfo.get(constants.INISECT_INS,
                                           "disk_template")
+        if self.op.disk_template not in constants.DISK_TEMPLATES:
+          raise errors.OpPrereqError("Disk template specified in configuration"
+                                     " file is not one of the allowed values:"
+                                     " %s" % " ".join(constants.DISK_TEMPLATES))
       else:
         raise errors.OpPrereqError("No disk template specified and the export"
                                    " is missing the disk_template information",
                                    errors.ECODE_INVAL)
 
     if not self.op.disks:
       else:
         raise errors.OpPrereqError("No disk template specified and the export"
                                    " is missing the disk_template information",
                                    errors.ECODE_INVAL)
 
     if not self.op.disks:
-      if einfo.has_option(constants.INISECT_INS, "disk_count"):
-        disks = []
-        # TODO: import the disk iv_name too
-        for idx in range(einfo.getint(constants.INISECT_INS, "disk_count")):
+      disks = []
+      # TODO: import the disk iv_name too
+      for idx in range(constants.MAX_DISKS):
+        if einfo.has_option(constants.INISECT_INS, "disk%d_size" % idx):
           disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
           disks.append({constants.IDISK_SIZE: disk_sz})
           disk_sz = einfo.getint(constants.INISECT_INS, "disk%d_size" % idx)
           disks.append({constants.IDISK_SIZE: disk_sz})
-        self.op.disks = disks
-      else:
+      self.op.disks = disks
+      if not disks and self.op.disk_template != constants.DT_DISKLESS:
         raise errors.OpPrereqError("No disk info specified and the export"
                                    " is missing the disk information",
                                    errors.ECODE_INVAL)
 
         raise errors.OpPrereqError("No disk info specified and the export"
                                    " is missing the disk information",
                                    errors.ECODE_INVAL)
 
-    if (not self.op.nics and
-        einfo.has_option(constants.INISECT_INS, "nic_count")):
+    if not self.op.nics:
       nics = []
       nics = []
-      for idx in range(einfo.getint(constants.INISECT_INS, "nic_count")):
-        ndict = {}
-        for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
-          v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
-          ndict[name] = v
-        nics.append(ndict)
+      for idx in range(constants.MAX_NICS):
+        if einfo.has_option(constants.INISECT_INS, "nic%d_mac" % idx):
+          ndict = {}
+          for name in list(constants.NICS_PARAMETERS) + ["ip", "mac"]:
+            v = einfo.get(constants.INISECT_INS, "nic%d_%s" % (idx, name))
+            ndict[name] = v
+          nics.append(ndict)
+        else:
+          break
       self.op.nics = nics
 
       self.op.nics = nics
 
+    if not self.op.tags and einfo.has_option(constants.INISECT_INS, "tags"):
+      self.op.tags = einfo.get(constants.INISECT_INS, "tags").split()
+
     if (self.op.hypervisor is None and
         einfo.has_option(constants.INISECT_INS, "hypervisor")):
       self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
     if (self.op.hypervisor is None and
         einfo.has_option(constants.INISECT_INS, "hypervisor")):
       self.op.hypervisor = einfo.get(constants.INISECT_INS, "hypervisor")
+
     if einfo.has_section(constants.INISECT_HYP):
       # use the export parameters but do not override the ones
       # specified by the user
     if einfo.has_section(constants.INISECT_HYP):
       # use the export parameters but do not override the ones
       # specified by the user
@@ -8173,6 +9167,12 @@ class LUInstanceCreate(LogicalUnit):
       for name, value in einfo.items(constants.INISECT_BEP):
         if name not in self.op.beparams:
           self.op.beparams[name] = value
       for name, value in einfo.items(constants.INISECT_BEP):
         if name not in self.op.beparams:
           self.op.beparams[name] = value
+        # Compatibility for the old "memory" be param
+        if name == constants.BE_MEMORY:
+          if constants.BE_MAXMEM not in self.op.beparams:
+            self.op.beparams[constants.BE_MAXMEM] = value
+          if constants.BE_MINMEM not in self.op.beparams:
+            self.op.beparams[constants.BE_MINMEM] = value
     else:
       # try to read the parameters old style, from the main section
       for name in constants.BES_PARAMETERS:
     else:
       # try to read the parameters old style, from the main section
       for name in constants.BES_PARAMETERS:
@@ -8212,10 +9212,40 @@ class LUInstanceCreate(LogicalUnit):
       if name in os_defs and os_defs[name] == self.op.osparams[name]:
         del self.op.osparams[name]
 
       if name in os_defs and os_defs[name] == self.op.osparams[name]:
         del self.op.osparams[name]
 
+  def _CalculateFileStorageDir(self):
+    """Calculate final instance file storage dir.
+
+    """
+    # file storage dir calculation/check
+    self.instance_file_storage_dir = None
+    if self.op.disk_template in constants.DTS_FILEBASED:
+      # build the full file storage dir path
+      joinargs = []
+
+      if self.op.disk_template == constants.DT_SHARED_FILE:
+        get_fsd_fn = self.cfg.GetSharedFileStorageDir
+      else:
+        get_fsd_fn = self.cfg.GetFileStorageDir
+
+      cfg_storagedir = get_fsd_fn()
+      if not cfg_storagedir:
+        raise errors.OpPrereqError("Cluster file storage dir not defined")
+      joinargs.append(cfg_storagedir)
+
+      if self.op.file_storage_dir is not None:
+        joinargs.append(self.op.file_storage_dir)
+
+      joinargs.append(self.op.instance_name)
+
+      # pylint: disable=W0142
+      self.instance_file_storage_dir = utils.PathJoin(*joinargs)
+
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
+    self._CalculateFileStorageDir()
+
     if self.op.mode == constants.INSTANCE_IMPORT:
       export_info = self._ReadExportInfo()
       self._ReadExportParams(export_info)
     if self.op.mode == constants.INSTANCE_IMPORT:
       export_info = self._ReadExportInfo()
       self._ReadExportParams(export_info)
@@ -8225,7 +9255,8 @@ class LUInstanceCreate(LogicalUnit):
       raise errors.OpPrereqError("Cluster does not support lvm-based"
                                  " instances", errors.ECODE_STATE)
 
       raise errors.OpPrereqError("Cluster does not support lvm-based"
                                  " instances", errors.ECODE_STATE)
 
-    if self.op.hypervisor is None:
+    if (self.op.hypervisor is None or
+        self.op.hypervisor == constants.VALUE_AUTO):
       self.op.hypervisor = self.cfg.GetHypervisorType()
 
     cluster = self.cfg.GetClusterInfo()
       self.op.hypervisor = self.cfg.GetHypervisorType()
 
     cluster = self.cfg.GetClusterInfo()
@@ -8251,6 +9282,11 @@ class LUInstanceCreate(LogicalUnit):
     _CheckGlobalHvParams(self.op.hvparams)
 
     # fill and remember the beparams dict
     _CheckGlobalHvParams(self.op.hvparams)
 
     # fill and remember the beparams dict
+    default_beparams = cluster.beparams[constants.PP_DEFAULT]
+    for param, value in self.op.beparams.iteritems():
+      if value == constants.VALUE_AUTO:
+        self.op.beparams[param] = default_beparams[param]
+    objects.UpgradeBeParams(self.op.beparams)
     utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
     self.be_full = cluster.SimpleFillBE(self.op.beparams)
 
     utils.ForceDictType(self.op.beparams, constants.BES_PARAMETER_TYPES)
     self.be_full = cluster.SimpleFillBE(self.op.beparams)
 
@@ -8267,7 +9303,7 @@ class LUInstanceCreate(LogicalUnit):
     for idx, nic in enumerate(self.op.nics):
       nic_mode_req = nic.get(constants.INIC_MODE, None)
       nic_mode = nic_mode_req
     for idx, nic in enumerate(self.op.nics):
       nic_mode_req = nic.get(constants.INIC_MODE, None)
       nic_mode = nic_mode_req
-      if nic_mode is None:
+      if nic_mode is None or nic_mode == constants.VALUE_AUTO:
         nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
 
       # in routed mode, for the first nic, the default ip is 'auto'
         nic_mode = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_MODE]
 
       # in routed mode, for the first nic, the default ip is 'auto'
@@ -8311,9 +9347,11 @@ class LUInstanceCreate(LogicalUnit):
 
       #  Build nic parameters
       link = nic.get(constants.INIC_LINK, None)
 
       #  Build nic parameters
       link = nic.get(constants.INIC_LINK, None)
+      if link == constants.VALUE_AUTO:
+        link = cluster.nicparams[constants.PP_DEFAULT][constants.NIC_LINK]
       nicparams = {}
       if nic_mode_req:
       nicparams = {}
       if nic_mode_req:
-        nicparams[constants.NIC_MODE] = nic_mode_req
+        nicparams[constants.NIC_MODE] = nic_mode
       if link:
         nicparams[constants.NIC_LINK] = link
 
       if link:
         nicparams[constants.NIC_LINK] = link
 
@@ -8343,26 +9381,17 @@ class LUInstanceCreate(LogicalUnit):
         constants.IDISK_SIZE: size,
         constants.IDISK_MODE: mode,
         constants.IDISK_VG: data_vg,
         constants.IDISK_SIZE: size,
         constants.IDISK_MODE: mode,
         constants.IDISK_VG: data_vg,
-        constants.IDISK_METAVG: disk.get(constants.IDISK_METAVG, data_vg),
         }
         }
+      if constants.IDISK_METAVG in disk:
+        new_disk[constants.IDISK_METAVG] = disk[constants.IDISK_METAVG]
       if constants.IDISK_ADOPT in disk:
         new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
       self.disks.append(new_disk)
 
     if self.op.mode == constants.INSTANCE_IMPORT:
       if constants.IDISK_ADOPT in disk:
         new_disk[constants.IDISK_ADOPT] = disk[constants.IDISK_ADOPT]
       self.disks.append(new_disk)
 
     if self.op.mode == constants.INSTANCE_IMPORT:
-
-      # Check that the new instance doesn't have less disks than the export
-      instance_disks = len(self.disks)
-      export_disks = export_info.getint(constants.INISECT_INS, 'disk_count')
-      if instance_disks < export_disks:
-        raise errors.OpPrereqError("Not enough disks to import."
-                                   " (instance: %d, export: %d)" %
-                                   (instance_disks, export_disks),
-                                   errors.ECODE_INVAL)
-
       disk_images = []
       disk_images = []
-      for idx in range(export_disks):
-        option = 'disk%d_dump' % idx
+      for idx in range(len(self.disks)):
+        option = "disk%d_dump" % idx
         if export_info.has_option(constants.INISECT_INS, option):
           # FIXME: are the old os-es, disk sizes, etc. useful?
           export_name = export_info.get(constants.INISECT_INS, option)
         if export_info.has_option(constants.INISECT_INS, option):
           # FIXME: are the old os-es, disk sizes, etc. useful?
           export_name = export_info.get(constants.INISECT_INS, option)
@@ -8373,17 +9402,11 @@ class LUInstanceCreate(LogicalUnit):
 
       self.src_images = disk_images
 
 
       self.src_images = disk_images
 
-      old_name = export_info.get(constants.INISECT_INS, 'name')
-      try:
-        exp_nic_count = export_info.getint(constants.INISECT_INS, 'nic_count')
-      except (TypeError, ValueError), err:
-        raise errors.OpPrereqError("Invalid export file, nic_count is not"
-                                   " an integer: %s" % str(err),
-                                   errors.ECODE_STATE)
+      old_name = export_info.get(constants.INISECT_INS, "name")
       if self.op.instance_name == old_name:
         for idx, nic in enumerate(self.nics):
       if self.op.instance_name == old_name:
         for idx, nic in enumerate(self.nics):
-          if nic.mac == constants.VALUE_AUTO and exp_nic_count >= idx:
-            nic_mac_ini = 'nic%d_mac' % idx
+          if nic.mac == constants.VALUE_AUTO:
+            nic_mac_ini = "nic%d_mac" % idx
             nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
 
     # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
             nic.mac = export_info.get(constants.INISECT_INS, nic_mac_ini)
 
     # ENDIF: self.op.mode == constants.INSTANCE_IMPORT
@@ -8412,6 +9435,11 @@ class LUInstanceCreate(LogicalUnit):
     if self.op.iallocator is not None:
       self._RunAllocator()
 
     if self.op.iallocator is not None:
       self._RunAllocator()
 
+    # Release all unneeded node locks
+    _ReleaseLocks(self, locking.LEVEL_NODE,
+                  keep=filter(None, [self.op.pnode, self.op.snode,
+                                     self.op.src_node]))
+
     #### node related checks
 
     # check primary node
     #### node related checks
 
     # check primary node
@@ -8440,8 +9468,19 @@ class LUInstanceCreate(LogicalUnit):
       _CheckNodeVmCapable(self, self.op.snode)
       self.secondaries.append(self.op.snode)
 
       _CheckNodeVmCapable(self, self.op.snode)
       self.secondaries.append(self.op.snode)
 
+      snode = self.cfg.GetNodeInfo(self.op.snode)
+      if pnode.group != snode.group:
+        self.LogWarning("The primary and secondary nodes are in two"
+                        " different node groups; the disk parameters"
+                        " from the first disk's node group will be"
+                        " used")
+
     nodenames = [pnode.name] + self.secondaries
 
     nodenames = [pnode.name] + self.secondaries
 
+    # disk parameters (not customizable at instance or node level)
+    # just use the primary node parameters, ignoring the secondary.
+    self.diskparams = self.cfg.GetNodeGroup(pnode.group).diskparams
+
     if not self.adopt_disks:
       # Check lv size requirements, if not adopting
       req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
     if not self.adopt_disks:
       # Check lv size requirements, if not adopting
       req_sizes = _ComputeDiskSizePerVG(self.op.disk_template, self.disks)
@@ -8526,10 +9565,11 @@ class LUInstanceCreate(LogicalUnit):
     _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
 
     # memory check on primary node
     _CheckNicsBridgesExist(self, self.nics, self.pnode.name)
 
     # memory check on primary node
+    #TODO(dynmem): use MINMEM for checking
     if self.op.start:
       _CheckNodeFreeMemory(self, self.pnode.name,
                            "creating instance %s" % self.op.instance_name,
     if self.op.start:
       _CheckNodeFreeMemory(self, self.pnode.name,
                            "creating instance %s" % self.op.instance_name,
-                           self.be_full[constants.BE_MEMORY],
+                           self.be_full[constants.BE_MAXMEM],
                            self.op.hypervisor)
 
     self.dry_run_result = list(nodenames)
                            self.op.hypervisor)
 
     self.dry_run_result = list(nodenames)
@@ -8541,60 +9581,50 @@ class LUInstanceCreate(LogicalUnit):
     instance = self.op.instance_name
     pnode_name = self.pnode.name
 
     instance = self.op.instance_name
     pnode_name = self.pnode.name
 
+    assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
+                self.owned_locks(locking.LEVEL_NODE)), \
+      "Node locks differ from node resource locks"
+
     ht_kind = self.op.hypervisor
     if ht_kind in constants.HTS_REQ_PORT:
       network_port = self.cfg.AllocatePort()
     else:
       network_port = None
 
     ht_kind = self.op.hypervisor
     if ht_kind in constants.HTS_REQ_PORT:
       network_port = self.cfg.AllocatePort()
     else:
       network_port = None
 
-    if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE:
-      # this is needed because os.path.join does not accept None arguments
-      if self.op.file_storage_dir is None:
-        string_file_storage_dir = ""
-      else:
-        string_file_storage_dir = self.op.file_storage_dir
-
-      # build the full file storage dir path
-      if self.op.disk_template == constants.DT_SHARED_FILE:
-        get_fsd_fn = self.cfg.GetSharedFileStorageDir
-      else:
-        get_fsd_fn = self.cfg.GetFileStorageDir
-
-      file_storage_dir = utils.PathJoin(get_fsd_fn(),
-                                        string_file_storage_dir, instance)
-    else:
-      file_storage_dir = ""
-
     disks = _GenerateDiskTemplate(self,
                                   self.op.disk_template,
                                   instance, pnode_name,
                                   self.secondaries,
                                   self.disks,
     disks = _GenerateDiskTemplate(self,
                                   self.op.disk_template,
                                   instance, pnode_name,
                                   self.secondaries,
                                   self.disks,
-                                  file_storage_dir,
+                                  self.instance_file_storage_dir,
                                   self.op.file_driver,
                                   0,
                                   self.op.file_driver,
                                   0,
-                                  feedback_fn)
+                                  feedback_fn,
+                                  self.diskparams)
 
     iobj = objects.Instance(name=instance, os=self.op.os_type,
                             primary_node=pnode_name,
                             nics=self.nics, disks=disks,
                             disk_template=self.op.disk_template,
 
     iobj = objects.Instance(name=instance, os=self.op.os_type,
                             primary_node=pnode_name,
                             nics=self.nics, disks=disks,
                             disk_template=self.op.disk_template,
-                            admin_up=False,
+                            admin_state=constants.ADMINST_DOWN,
                             network_port=network_port,
                             beparams=self.op.beparams,
                             hvparams=self.op.hvparams,
                             hypervisor=self.op.hypervisor,
                             osparams=self.op.osparams,
                             network_port=network_port,
                             beparams=self.op.beparams,
                             hvparams=self.op.hvparams,
                             hypervisor=self.op.hypervisor,
                             osparams=self.op.osparams,
-                            tags=self.op.tags,
                             )
 
                             )
 
+    if self.op.tags:
+      for tag in self.op.tags:
+        iobj.AddTag(tag)
+
     if self.adopt_disks:
       if self.op.disk_template == constants.DT_PLAIN:
         # rename LVs to the newly-generated names; we need to construct
         # 'fake' LV disks with the old data, plus the new unique_id
         tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
         rename_to = []
     if self.adopt_disks:
       if self.op.disk_template == constants.DT_PLAIN:
         # rename LVs to the newly-generated names; we need to construct
         # 'fake' LV disks with the old data, plus the new unique_id
         tmp_disks = [objects.Disk.FromDict(v.ToDict()) for v in disks]
         rename_to = []
-        for t_dsk, a_dsk in zip (tmp_disks, self.disks):
+        for t_dsk, a_dsk in zip(tmp_disks, self.disks):
           rename_to.append(t_dsk.logical_id)
           t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
           self.cfg.SetDiskID(t_dsk, pnode_name)
           rename_to.append(t_dsk.logical_id)
           t_dsk.logical_id = (t_dsk.logical_id[0], a_dsk[constants.IDISK_ADOPT])
           self.cfg.SetDiskID(t_dsk, pnode_name)
@@ -8645,7 +9675,6 @@ class LUInstanceCreate(LogicalUnit):
       disk_abort = not _WaitForSync(self, iobj)
     elif iobj.disk_template in constants.DTS_INT_MIRROR:
       # make sure the disks are not degraded (still sync-ing is ok)
       disk_abort = not _WaitForSync(self, iobj)
     elif iobj.disk_template in constants.DTS_INT_MIRROR:
       # make sure the disks are not degraded (still sync-ing is ok)
-      time.sleep(15)
       feedback_fn("* checking mirrors status")
       disk_abort = not _WaitForSync(self, iobj, oneshot=True)
     else:
       feedback_fn("* checking mirrors status")
       disk_abort = not _WaitForSync(self, iobj, oneshot=True)
     else:
@@ -8659,15 +9688,39 @@ class LUInstanceCreate(LogicalUnit):
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance")
 
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance")
 
+    # Release all node resource locks
+    _ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
     if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
       if self.op.mode == constants.INSTANCE_CREATE:
         if not self.op.no_install:
     if iobj.disk_template != constants.DT_DISKLESS and not self.adopt_disks:
       if self.op.mode == constants.INSTANCE_CREATE:
         if not self.op.no_install:
+          pause_sync = (iobj.disk_template in constants.DTS_INT_MIRROR and
+                        not self.op.wait_for_sync)
+          if pause_sync:
+            feedback_fn("* pausing disk sync to install instance OS")
+            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+                                                              iobj.disks, True)
+            for idx, success in enumerate(result.payload):
+              if not success:
+                logging.warn("pause-sync of instance %s for disk %d failed",
+                             instance, idx)
+
           feedback_fn("* running the instance OS create scripts...")
           # FIXME: pass debug option from opcode to backend
           feedback_fn("* running the instance OS create scripts...")
           # FIXME: pass debug option from opcode to backend
-          result = self.rpc.call_instance_os_add(pnode_name, iobj, False,
-                                                 self.op.debug_level)
-          result.Raise("Could not add os for instance %s"
-                       " on node %s" % (instance, pnode_name))
+          os_add_result = \
+            self.rpc.call_instance_os_add(pnode_name, (iobj, None), False,
+                                          self.op.debug_level)
+          if pause_sync:
+            feedback_fn("* resuming disk sync")
+            result = self.rpc.call_blockdev_pause_resume_sync(pnode_name,
+                                                              iobj.disks, False)
+            for idx, success in enumerate(result.payload):
+              if not success:
+                logging.warn("resume-sync of instance %s for disk %d failed",
+                             instance, idx)
+
+          os_add_result.Raise("Could not add os for instance %s"
+                              " on node %s" % (instance, pnode_name))
 
       elif self.op.mode == constants.INSTANCE_IMPORT:
         feedback_fn("* running the instance OS import scripts...")
 
       elif self.op.mode == constants.INSTANCE_IMPORT:
         feedback_fn("* running the instance OS import scripts...")
@@ -8730,12 +9783,15 @@ class LUInstanceCreate(LogicalUnit):
         raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
                                      % self.op.mode)
 
         raise errors.ProgrammerError("Unknown OS initialization mode '%s'"
                                      % self.op.mode)
 
+    assert not self.owned_locks(locking.LEVEL_NODE_RES)
+
     if self.op.start:
     if self.op.start:
-      iobj.admin_up = True
+      iobj.admin_state = constants.ADMINST_UP
       self.cfg.Update(iobj, feedback_fn)
       logging.info("Starting instance %s on node %s", instance, pnode_name)
       feedback_fn("* starting instance...")
       self.cfg.Update(iobj, feedback_fn)
       logging.info("Starting instance %s on node %s", instance, pnode_name)
       feedback_fn("* starting instance...")
-      result = self.rpc.call_instance_start(pnode_name, iobj, None, None)
+      result = self.rpc.call_instance_start(pnode_name, (iobj, None, None),
+                                            False)
       result.Raise("Could not start instance")
 
     return list(iobj.all_nodes)
       result.Raise("Could not start instance")
 
     return list(iobj.all_nodes)
@@ -8752,6 +9808,7 @@ class LUInstanceConsole(NoHooksLU):
   REQ_BGL = False
 
   def ExpandNames(self):
   REQ_BGL = False
 
   def ExpandNames(self):
+    self.share_locks = _ShareAll()
     self._ExpandAndLockInstance()
 
   def CheckPrereq(self):
     self._ExpandAndLockInstance()
 
   def CheckPrereq(self):
@@ -8777,10 +9834,12 @@ class LUInstanceConsole(NoHooksLU):
     node_insts.Raise("Can't get node information from %s" % node)
 
     if instance.name not in node_insts.payload:
     node_insts.Raise("Can't get node information from %s" % node)
 
     if instance.name not in node_insts.payload:
-      if instance.admin_up:
+      if instance.admin_state == constants.ADMINST_UP:
         state = constants.INSTST_ERRORDOWN
         state = constants.INSTST_ERRORDOWN
-      else:
+      elif instance.admin_state == constants.ADMINST_DOWN:
         state = constants.INSTST_ADMINDOWN
         state = constants.INSTST_ADMINDOWN
+      else:
+        state = constants.INSTST_ADMINOFFLINE
       raise errors.OpExecError("Instance %s is not running (state %s)" %
                                (instance.name, state))
 
       raise errors.OpExecError("Instance %s is not running (state %s)" %
                                (instance.name, state))
 
@@ -8826,6 +9885,7 @@ class LUInstanceReplaceDisks(LogicalUnit):
     self._ExpandAndLockInstance()
 
     assert locking.LEVEL_NODE not in self.needed_locks
     self._ExpandAndLockInstance()
 
     assert locking.LEVEL_NODE not in self.needed_locks
+    assert locking.LEVEL_NODE_RES not in self.needed_locks
     assert locking.LEVEL_NODEGROUP not in self.needed_locks
 
     assert self.op.iallocator is None or self.op.remote_node is None, \
     assert locking.LEVEL_NODEGROUP not in self.needed_locks
 
     assert self.op.iallocator is None or self.op.remote_node is None, \
@@ -8848,6 +9908,8 @@ class LUInstanceReplaceDisks(LogicalUnit):
         # iallocator will select a new node in the same group
         self.needed_locks[locking.LEVEL_NODEGROUP] = []
 
         # iallocator will select a new node in the same group
         self.needed_locks[locking.LEVEL_NODEGROUP] = []
 
+    self.needed_locks[locking.LEVEL_NODE_RES] = []
+
     self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
                                    self.op.iallocator, self.op.remote_node,
                                    self.op.disks, False, self.op.early_release)
     self.replacer = TLReplaceDisks(self, self.op.instance_name, self.op.mode,
                                    self.op.iallocator, self.op.remote_node,
                                    self.op.disks, False, self.op.early_release)
@@ -8861,6 +9923,8 @@ class LUInstanceReplaceDisks(LogicalUnit):
       assert not self.needed_locks[locking.LEVEL_NODEGROUP]
 
       self.share_locks[locking.LEVEL_NODEGROUP] = 1
       assert not self.needed_locks[locking.LEVEL_NODEGROUP]
 
       self.share_locks[locking.LEVEL_NODEGROUP] = 1
+      # Lock all groups used by instance optimistically; this requires going
+      # via the node before it's locked, requiring verification later on
       self.needed_locks[locking.LEVEL_NODEGROUP] = \
         self.cfg.GetInstanceNodeGroups(self.op.instance_name)
 
       self.needed_locks[locking.LEVEL_NODEGROUP] = \
         self.cfg.GetInstanceNodeGroups(self.op.instance_name)
 
@@ -8871,10 +9935,14 @@ class LUInstanceReplaceDisks(LogicalUnit):
 
         # Lock member nodes of all locked groups
         self.needed_locks[locking.LEVEL_NODE] = [node_name
 
         # Lock member nodes of all locked groups
         self.needed_locks[locking.LEVEL_NODE] = [node_name
-          for group_uuid in self.glm.list_owned(locking.LEVEL_NODEGROUP)
+          for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
           for node_name in self.cfg.GetNodeGroup(group_uuid).members]
       else:
         self._LockInstancesNodes()
           for node_name in self.cfg.GetNodeGroup(group_uuid).members]
       else:
         self._LockInstancesNodes()
+    elif level == locking.LEVEL_NODE_RES:
+      # Reuse node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -8911,16 +9979,10 @@ class LUInstanceReplaceDisks(LogicalUnit):
     assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
             self.op.iallocator is None)
 
     assert (self.glm.is_owned(locking.LEVEL_NODEGROUP) or
             self.op.iallocator is None)
 
-    owned_groups = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+    # Verify if node group locks are still correct
+    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
     if owned_groups:
     if owned_groups:
-      groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
-      if owned_groups != groups:
-        raise errors.OpExecError("Node groups used by instance '%s' changed"
-                                 " since lock was acquired, current list is %r,"
-                                 " used to be '%s'" %
-                                 (self.op.instance_name,
-                                  utils.CommaJoin(groups),
-                                  utils.CommaJoin(owned_groups)))
+      _CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
 
     return LogicalUnit.CheckPrereq(self)
 
 
     return LogicalUnit.CheckPrereq(self)
 
@@ -8985,7 +10047,7 @@ class TLReplaceDisks(Tasklet):
     ial = IAllocator(lu.cfg, lu.rpc,
                      mode=constants.IALLOCATOR_MODE_RELOC,
                      name=instance_name,
     ial = IAllocator(lu.cfg, lu.rpc,
                      mode=constants.IALLOCATOR_MODE_RELOC,
                      name=instance_name,
-                     relocate_from=relocate_from)
+                     relocate_from=list(relocate_from))
 
     ial.Run(iallocator_name)
 
 
     ial.Run(iallocator_name)
 
@@ -9009,6 +10071,9 @@ class TLReplaceDisks(Tasklet):
     return remote_node_name
 
   def _FindFaultyDisks(self, node_name):
     return remote_node_name
 
   def _FindFaultyDisks(self, node_name):
+    """Wrapper for L{_FindFaultyInstanceDisks}.
+
+    """
     return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
                                     node_name, True)
 
     return _FindFaultyInstanceDisks(self.cfg, self.rpc, self.instance,
                                     node_name, True)
 
@@ -9079,7 +10144,7 @@ class TLReplaceDisks(Tasklet):
     if remote_node is None:
       self.remote_node_info = None
     else:
     if remote_node is None:
       self.remote_node_info = None
     else:
-      assert remote_node in self.lu.glm.list_owned(locking.LEVEL_NODE), \
+      assert remote_node in self.lu.owned_locks(locking.LEVEL_NODE), \
              "Remote node '%s' is not locked" % remote_node
 
       self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
              "Remote node '%s' is not locked" % remote_node
 
       self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
@@ -9165,6 +10230,16 @@ class TLReplaceDisks(Tasklet):
       if not self.disks:
         self.disks = range(len(self.instance.disks))
 
       if not self.disks:
         self.disks = range(len(self.instance.disks))
 
+    # TODO: compute disk parameters
+    primary_node_info = self.cfg.GetNodeInfo(instance.primary_node)
+    secondary_node_info = self.cfg.GetNodeInfo(secondary_node)
+    if primary_node_info.group != secondary_node_info.group:
+      self.lu.LogInfo("The instance primary and secondary nodes are in two"
+                      " different node groups; the disk parameters of the"
+                      " primary node's group will be applied.")
+
+    self.diskparams = self.cfg.GetNodeGroup(primary_node_info.group).diskparams
+
     for node in check_nodes:
       _CheckNodeOnline(self.lu, node)
 
     for node in check_nodes:
       _CheckNodeOnline(self.lu, node)
 
@@ -9173,8 +10248,9 @@ class TLReplaceDisks(Tasklet):
                                                           self.target_node]
                               if node_name is not None)
 
                                                           self.target_node]
                               if node_name is not None)
 
-    # Release unneeded node locks
+    # Release unneeded node and node resource locks
     _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
     _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
+    _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
 
     # Release any owned node group
     if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
 
     # Release any owned node group
     if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
@@ -9185,9 +10261,8 @@ class TLReplaceDisks(Tasklet):
       instance.FindDisk(disk_idx)
 
     # Get secondary node IP addresses
       instance.FindDisk(disk_idx)
 
     # Get secondary node IP addresses
-    self.node_secondary_ip = \
-      dict((node_name, self.cfg.GetNodeInfo(node_name).secondary_ip)
-           for node_name in touched_nodes)
+    self.node_secondary_ip = dict((name, node.secondary_ip) for (name, node)
+                                  in self.cfg.GetMultiNodeInfo(touched_nodes))
 
   def Exec(self, feedback_fn):
     """Execute disk replacement.
 
   def Exec(self, feedback_fn):
     """Execute disk replacement.
@@ -9200,13 +10275,15 @@ class TLReplaceDisks(Tasklet):
 
     if __debug__:
       # Verify owned locks before starting operation
 
     if __debug__:
       # Verify owned locks before starting operation
-      owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
-      assert set(owned_locks) == set(self.node_secondary_ip), \
+      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
+      assert set(owned_nodes) == set(self.node_secondary_ip), \
           ("Incorrect node locks, owning %s, expected %s" %
           ("Incorrect node locks, owning %s, expected %s" %
-           (owned_locks, self.node_secondary_ip.keys()))
+           (owned_nodes, self.node_secondary_ip.keys()))
+      assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
+              self.lu.owned_locks(locking.LEVEL_NODE_RES))
 
 
-      owned_locks = self.lu.glm.list_owned(locking.LEVEL_INSTANCE)
-      assert list(owned_locks) == [self.instance_name], \
+      owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
+      assert list(owned_instances) == [self.instance_name], \
           "Instance '%s' not locked" % self.instance_name
 
       assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
           "Instance '%s' not locked" % self.instance_name
 
       assert not self.lu.glm.is_owned(locking.LEVEL_NODEGROUP), \
@@ -9219,7 +10296,7 @@ class TLReplaceDisks(Tasklet):
     feedback_fn("Replacing disk(s) %s for %s" %
                 (utils.CommaJoin(self.disks), self.instance.name))
 
     feedback_fn("Replacing disk(s) %s for %s" %
                 (utils.CommaJoin(self.disks), self.instance.name))
 
-    activate_disks = (not self.instance.admin_up)
+    activate_disks = (self.instance.admin_state != constants.ADMINST_UP)
 
     # Activate the instance disks if we're replacing them on a down instance
     if activate_disks:
 
     # Activate the instance disks if we're replacing them on a down instance
     if activate_disks:
@@ -9239,14 +10316,16 @@ class TLReplaceDisks(Tasklet):
       if activate_disks:
         _SafeShutdownInstanceDisks(self.lu, self.instance)
 
       if activate_disks:
         _SafeShutdownInstanceDisks(self.lu, self.instance)
 
+    assert not self.lu.owned_locks(locking.LEVEL_NODE)
+
     if __debug__:
       # Verify owned locks
     if __debug__:
       # Verify owned locks
-      owned_locks = self.lu.glm.list_owned(locking.LEVEL_NODE)
+      owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE_RES)
       nodes = frozenset(self.node_secondary_ip)
       nodes = frozenset(self.node_secondary_ip)
-      assert ((self.early_release and not owned_locks) or
-              (not self.early_release and not (set(owned_locks) - nodes))), \
+      assert ((self.early_release and not owned_nodes) or
+              (not self.early_release and not (set(owned_nodes) - nodes))), \
         ("Not owning the correct locks, early_release=%s, owned=%r,"
         ("Not owning the correct locks, early_release=%s, owned=%r,"
-         " nodes=%r" % (self.early_release, owned_locks, nodes))
+         " nodes=%r" % (self.early_release, owned_nodes, nodes))
 
     return result
 
 
     return result
 
@@ -9301,6 +10380,12 @@ class TLReplaceDisks(Tasklet):
                                  (node_name, self.instance.name))
 
   def _CreateNewStorage(self, node_name):
                                  (node_name, self.instance.name))
 
   def _CreateNewStorage(self, node_name):
+    """Create new storage on the primary or secondary node.
+
+    This is only used for same-node replaces, not for changing the
+    secondary node, hence we don't want to modify the existing disk.
+
+    """
     iv_names = {}
 
     for idx, dev in enumerate(self.instance.disks):
     iv_names = {}
 
     for idx, dev in enumerate(self.instance.disks):
@@ -9314,15 +10399,17 @@ class TLReplaceDisks(Tasklet):
       lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
       names = _GenerateUniqueNames(self.lu, lv_names)
 
       lv_names = [".disk%d_%s" % (idx, suffix) for suffix in ["data", "meta"]]
       names = _GenerateUniqueNames(self.lu, lv_names)
 
+      _, data_p, meta_p = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
+
       vg_data = dev.children[0].logical_id[0]
       lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
       vg_data = dev.children[0].logical_id[0]
       lv_data = objects.Disk(dev_type=constants.LD_LV, size=dev.size,
-                             logical_id=(vg_data, names[0]))
+                             logical_id=(vg_data, names[0]), params=data_p)
       vg_meta = dev.children[1].logical_id[0]
       vg_meta = dev.children[1].logical_id[0]
-      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=128,
-                             logical_id=(vg_meta, names[1]))
+      lv_meta = objects.Disk(dev_type=constants.LD_LV, size=DRBD_META_SIZE,
+                             logical_id=(vg_meta, names[1]), params=meta_p)
 
       new_lvs = [lv_data, lv_meta]
 
       new_lvs = [lv_data, lv_meta]
-      old_lvs = dev.children
+      old_lvs = [child.Copy() for child in dev.children]
       iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
 
       # we pass force_create=True to force the LVM creation
       iv_names[dev.iv_name] = (dev, old_lvs, new_lvs)
 
       # we pass force_create=True to force the LVM creation
@@ -9360,7 +10447,7 @@ class TLReplaceDisks(Tasklet):
           self.lu.LogWarning("Can't remove old LV: %s" % msg,
                              hint="remove unused LVs manually")
 
           self.lu.LogWarning("Can't remove old LV: %s" % msg,
                              hint="remove unused LVs manually")
 
-  def _ExecDrbd8DiskOnly(self, feedback_fn):
+  def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
     """Replace a disk on the primary or secondary for DRBD 8.
 
     The algorithm for replace is quite complicated:
     """Replace a disk on the primary or secondary for DRBD 8.
 
     The algorithm for replace is quite complicated:
@@ -9443,10 +10530,14 @@ class TLReplaceDisks(Tasklet):
                                              rename_new_to_old)
       result.Raise("Can't rename new LVs on node %s" % self.target_node)
 
                                              rename_new_to_old)
       result.Raise("Can't rename new LVs on node %s" % self.target_node)
 
+      # Intermediate steps of in memory modifications
       for old, new in zip(old_lvs, new_lvs):
         new.logical_id = old.logical_id
         self.cfg.SetDiskID(new, self.target_node)
 
       for old, new in zip(old_lvs, new_lvs):
         new.logical_id = old.logical_id
         self.cfg.SetDiskID(new, self.target_node)
 
+      # We need to modify old_lvs so that removal later removes the
+      # right LVs, not the newly added ones; note that old_lvs is a
+      # copy here
       for disk in old_lvs:
         disk.logical_id = ren_fn(disk, temp_suffix)
         self.cfg.SetDiskID(disk, self.target_node)
       for disk in old_lvs:
         disk.logical_id = ren_fn(disk, temp_suffix)
         self.cfg.SetDiskID(disk, self.target_node)
@@ -9466,25 +10557,28 @@ class TLReplaceDisks(Tasklet):
                                      "volumes"))
         raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
 
                                      "volumes"))
         raise errors.OpExecError("Can't add local storage to drbd: %s" % msg)
 
-      dev.children = new_lvs
+    cstep = itertools.count(5)
 
 
-      self.cfg.Update(self.instance, feedback_fn)
-
-    cstep = 5
     if self.early_release:
     if self.early_release:
-      self.lu.LogStep(cstep, steps_total, "Removing old storage")
-      cstep += 1
+      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
       self._RemoveOldStorage(self.target_node, iv_names)
       self._RemoveOldStorage(self.target_node, iv_names)
-      # WARNING: we release both node locks here, do not do other RPCs
-      # than WaitForSync to the primary node
-      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
-                    names=[self.target_node, self.other_node])
+      # TODO: Check if releasing locks early still makes sense
+      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+    else:
+      # Release all resource locks except those used by the instance
+      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+                    keep=self.node_secondary_ip.keys())
+
+    # Release all node locks while waiting for sync
+    _ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
+    # TODO: Can the instance lock be downgraded here? Take the optional disk
+    # shutdown in the caller into consideration.
 
     # Wait for sync
     # This can fail as the old devices are degraded and _WaitForSync
     # does a combined result over all disks, so we don't check its return value
 
     # Wait for sync
     # This can fail as the old devices are degraded and _WaitForSync
     # does a combined result over all disks, so we don't check its return value
-    self.lu.LogStep(cstep, steps_total, "Sync devices")
-    cstep += 1
+    self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
     _WaitForSync(self.lu, self.instance)
 
     # Check all devices manually
     _WaitForSync(self.lu, self.instance)
 
     # Check all devices manually
@@ -9492,8 +10586,7 @@ class TLReplaceDisks(Tasklet):
 
     # Step: remove old storage
     if not self.early_release:
 
     # Step: remove old storage
     if not self.early_release:
-      self.lu.LogStep(cstep, steps_total, "Removing old storage")
-      cstep += 1
+      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
       self._RemoveOldStorage(self.target_node, iv_names)
 
   def _ExecDrbd8Secondary(self, feedback_fn):
       self._RemoveOldStorage(self.target_node, iv_names)
 
   def _ExecDrbd8Secondary(self, feedback_fn):
@@ -9517,6 +10610,8 @@ class TLReplaceDisks(Tasklet):
     """
     steps_total = 6
 
     """
     steps_total = 6
 
+    pnode = self.instance.primary_node
+
     # Step: check device activation
     self.lu.LogStep(1, steps_total, "Check device existence")
     self._CheckDisksExistence([self.instance.primary_node])
     # Step: check device activation
     self.lu.LogStep(1, steps_total, "Check device existence")
     self._CheckDisksExistence([self.instance.primary_node])
@@ -9568,10 +10663,12 @@ class TLReplaceDisks(Tasklet):
       iv_names[idx] = (dev, dev.children, new_net_id)
       logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
                     new_net_id)
       iv_names[idx] = (dev, dev.children, new_net_id)
       logging.debug("Allocated new_minor: %s, new_logical_id: %s", new_minor,
                     new_net_id)
+      drbd_params, _, _ = _ComputeLDParams(constants.DT_DRBD8, self.diskparams)
       new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
                               logical_id=new_alone_id,
                               children=dev.children,
       new_drbd = objects.Disk(dev_type=constants.LD_DRBD8,
                               logical_id=new_alone_id,
                               children=dev.children,
-                              size=dev.size)
+                              size=dev.size,
+                              params=drbd_params)
       try:
         _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
                               _GetInstanceInfoText(self.instance), False)
       try:
         _CreateSingleBlockDev(self.lu, self.new_node, self.instance, new_drbd,
                               _GetInstanceInfoText(self.instance), False)
@@ -9591,10 +10688,8 @@ class TLReplaceDisks(Tasklet):
                                  " soon as possible"))
 
     self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
                                  " soon as possible"))
 
     self.lu.LogInfo("Detaching primary drbds from the network (=> standalone)")
-    result = self.rpc.call_drbd_disconnect_net([self.instance.primary_node],
-                                               self.node_secondary_ip,
-                                               self.instance.disks)\
-                                              [self.instance.primary_node]
+    result = self.rpc.call_drbd_disconnect_net([pnode], self.node_secondary_ip,
+                                               self.instance.disks)[pnode]
 
     msg = result.fail_msg
     if msg:
 
     msg = result.fail_msg
     if msg:
@@ -9612,6 +10707,9 @@ class TLReplaceDisks(Tasklet):
 
     self.cfg.Update(self.instance, feedback_fn)
 
 
     self.cfg.Update(self.instance, feedback_fn)
 
+    # Release all node locks (the configuration has been updated)
+    _ReleaseLocks(self.lu, locking.LEVEL_NODE)
+
     # and now perform the drbd attach
     self.lu.LogInfo("Attaching primary drbds to new secondary"
                     " (standalone => connected)")
     # and now perform the drbd attach
     self.lu.LogInfo("Attaching primary drbds to new secondary"
                     " (standalone => connected)")
@@ -9628,23 +10726,26 @@ class TLReplaceDisks(Tasklet):
                            to_node, msg,
                            hint=("please do a gnt-instance info to see the"
                                  " status of disks"))
                            to_node, msg,
                            hint=("please do a gnt-instance info to see the"
                                  " status of disks"))
-    cstep = 5
+
+    cstep = itertools.count(5)
+
     if self.early_release:
     if self.early_release:
-      self.lu.LogStep(cstep, steps_total, "Removing old storage")
-      cstep += 1
+      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
       self._RemoveOldStorage(self.target_node, iv_names)
       self._RemoveOldStorage(self.target_node, iv_names)
-      # WARNING: we release all node locks here, do not do other RPCs
-      # than WaitForSync to the primary node
-      _ReleaseLocks(self.lu, locking.LEVEL_NODE,
-                    names=[self.instance.primary_node,
-                           self.target_node,
-                           self.new_node])
+      # TODO: Check if releasing locks early still makes sense
+      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES)
+    else:
+      # Release all resource locks except those used by the instance
+      _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES,
+                    keep=self.node_secondary_ip.keys())
+
+    # TODO: Can the instance lock be downgraded here? Take the optional disk
+    # shutdown in the caller into consideration.
 
     # Wait for sync
     # This can fail as the old devices are degraded and _WaitForSync
     # does a combined result over all disks, so we don't check its return value
 
     # Wait for sync
     # This can fail as the old devices are degraded and _WaitForSync
     # does a combined result over all disks, so we don't check its return value
-    self.lu.LogStep(cstep, steps_total, "Sync devices")
-    cstep += 1
+    self.lu.LogStep(cstep.next(), steps_total, "Sync devices")
     _WaitForSync(self.lu, self.instance)
 
     # Check all devices manually
     _WaitForSync(self.lu, self.instance)
 
     # Check all devices manually
@@ -9652,7 +10753,7 @@ class TLReplaceDisks(Tasklet):
 
     # Step: remove old storage
     if not self.early_release:
 
     # Step: remove old storage
     if not self.early_release:
-      self.lu.LogStep(cstep, steps_total, "Removing old storage")
+      self.lu.LogStep(cstep.next(), steps_total, "Removing old storage")
       self._RemoveOldStorage(self.target_node, iv_names)
 
 
       self._RemoveOldStorage(self.target_node, iv_names)
 
 
@@ -9698,7 +10799,7 @@ class LURepairNodeStorage(NoHooksLU):
     """
     # Check whether any instance on this node has faulty disks
     for inst in _GetNodeInstances(self.cfg, self.op.node_name):
     """
     # Check whether any instance on this node has faulty disks
     for inst in _GetNodeInstances(self.cfg, self.op.node_name):
-      if not inst.admin_up:
+      if inst.admin_state != constants.ADMINST_UP:
         continue
       check_nodes = set(inst.all_nodes)
       check_nodes.discard(self.op.node_name)
         continue
       check_nodes = set(inst.all_nodes)
       check_nodes.discard(self.op.node_name)
@@ -9718,51 +10819,260 @@ class LURepairNodeStorage(NoHooksLU):
                  (self.op.name, self.op.node_name))
 
 
                  (self.op.name, self.op.node_name))
 
 
-class LUNodeEvacStrategy(NoHooksLU):
-  """Computes the node evacuation strategy.
+class LUNodeEvacuate(NoHooksLU):
+  """Evacuates instances off a list of nodes.
 
   """
   REQ_BGL = False
 
 
   """
   REQ_BGL = False
 
+  _MODE2IALLOCATOR = {
+    constants.NODE_EVAC_PRI: constants.IALLOCATOR_NEVAC_PRI,
+    constants.NODE_EVAC_SEC: constants.IALLOCATOR_NEVAC_SEC,
+    constants.NODE_EVAC_ALL: constants.IALLOCATOR_NEVAC_ALL,
+    }
+  assert frozenset(_MODE2IALLOCATOR.keys()) == constants.NODE_EVAC_MODES
+  assert (frozenset(_MODE2IALLOCATOR.values()) ==
+          constants.IALLOCATOR_NEVAC_MODES)
+
   def CheckArguments(self):
     _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
 
   def ExpandNames(self):
   def CheckArguments(self):
     _CheckIAllocatorOrNode(self, "iallocator", "remote_node")
 
   def ExpandNames(self):
-    self.op.nodes = _GetWantedNodes(self, self.op.nodes)
-    self.needed_locks = locks = {}
+    self.op.node_name = _ExpandNodeName(self.cfg, self.op.node_name)
+
+    if self.op.remote_node is not None:
+      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
+      assert self.op.remote_node
+
+      if self.op.remote_node == self.op.node_name:
+        raise errors.OpPrereqError("Can not use evacuated node as a new"
+                                   " secondary node", errors.ECODE_INVAL)
+
+      if self.op.mode != constants.NODE_EVAC_SEC:
+        raise errors.OpPrereqError("Without the use of an iallocator only"
+                                   " secondary instances can be evacuated",
+                                   errors.ECODE_INVAL)
+
+    # Declare locks
+    self.share_locks = _ShareAll()
+    self.needed_locks = {
+      locking.LEVEL_INSTANCE: [],
+      locking.LEVEL_NODEGROUP: [],
+      locking.LEVEL_NODE: [],
+      }
+
+    # Determine nodes (via group) optimistically, needs verification once locks
+    # have been acquired
+    self.lock_nodes = self._DetermineNodes()
+
+  def _DetermineNodes(self):
+    """Gets the list of nodes to operate on.
+
+    """
     if self.op.remote_node is None:
     if self.op.remote_node is None:
-      locks[locking.LEVEL_NODE] = locking.ALL_SET
+      # Iallocator will choose any node(s) in the same group
+      group_nodes = self.cfg.GetNodeGroupMembersByNodes([self.op.node_name])
     else:
     else:
-      self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
-      locks[locking.LEVEL_NODE] = self.op.nodes + [self.op.remote_node]
+      group_nodes = frozenset([self.op.remote_node])
 
 
-  def Exec(self, feedback_fn):
-    instances = []
-    for node in self.op.nodes:
-      instances.extend(_GetNodeSecondaryInstances(self.cfg, node))
-    if not instances:
-      return []
+    # Determine nodes to be locked
+    return set([self.op.node_name]) | group_nodes
+
+  def _DetermineInstances(self):
+    """Builds list of instances to operate on.
+
+    """
+    assert self.op.mode in constants.NODE_EVAC_MODES
+
+    if self.op.mode == constants.NODE_EVAC_PRI:
+      # Primary instances only
+      inst_fn = _GetNodePrimaryInstances
+      assert self.op.remote_node is None, \
+        "Evacuating primary instances requires iallocator"
+    elif self.op.mode == constants.NODE_EVAC_SEC:
+      # Secondary instances only
+      inst_fn = _GetNodeSecondaryInstances
+    else:
+      # All instances
+      assert self.op.mode == constants.NODE_EVAC_ALL
+      inst_fn = _GetNodeInstances
+      # TODO: In 2.6, change the iallocator interface to take an evacuation mode
+      # per instance
+      raise errors.OpPrereqError("Due to an issue with the iallocator"
+                                 " interface it is not possible to evacuate"
+                                 " all instances at once; specify explicitly"
+                                 " whether to evacuate primary or secondary"
+                                 " instances",
+                                 errors.ECODE_INVAL)
+
+    return inst_fn(self.cfg, self.op.node_name)
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_INSTANCE:
+      # Lock instances optimistically, needs verification once node and group
+      # locks have been acquired
+      self.needed_locks[locking.LEVEL_INSTANCE] = \
+        set(i.name for i in self._DetermineInstances())
+
+    elif level == locking.LEVEL_NODEGROUP:
+      # Lock node groups for all potential target nodes optimistically, needs
+      # verification once nodes have been acquired
+      self.needed_locks[locking.LEVEL_NODEGROUP] = \
+        self.cfg.GetNodeGroupsFromNodes(self.lock_nodes)
+
+    elif level == locking.LEVEL_NODE:
+      self.needed_locks[locking.LEVEL_NODE] = self.lock_nodes
+
+  def CheckPrereq(self):
+    # Verify locks
+    owned_instances = self.owned_locks(locking.LEVEL_INSTANCE)
+    owned_nodes = self.owned_locks(locking.LEVEL_NODE)
+    owned_groups = self.owned_locks(locking.LEVEL_NODEGROUP)
+
+    need_nodes = self._DetermineNodes()
+
+    if not owned_nodes.issuperset(need_nodes):
+      raise errors.OpPrereqError("Nodes in same group as '%s' changed since"
+                                 " locks were acquired, current nodes are"
+                                 " are '%s', used to be '%s'; retry the"
+                                 " operation" %
+                                 (self.op.node_name,
+                                  utils.CommaJoin(need_nodes),
+                                  utils.CommaJoin(owned_nodes)),
+                                 errors.ECODE_STATE)
+
+    wanted_groups = self.cfg.GetNodeGroupsFromNodes(owned_nodes)
+    if owned_groups != wanted_groups:
+      raise errors.OpExecError("Node groups changed since locks were acquired,"
+                               " current groups are '%s', used to be '%s';"
+                               " retry the operation" %
+                               (utils.CommaJoin(wanted_groups),
+                                utils.CommaJoin(owned_groups)))
+
+    # Determine affected instances
+    self.instances = self._DetermineInstances()
+    self.instance_names = [i.name for i in self.instances]
+
+    if set(self.instance_names) != owned_instances:
+      raise errors.OpExecError("Instances on node '%s' changed since locks"
+                               " were acquired, current instances are '%s',"
+                               " used to be '%s'; retry the operation" %
+                               (self.op.node_name,
+                                utils.CommaJoin(self.instance_names),
+                                utils.CommaJoin(owned_instances)))
+
+    if self.instance_names:
+      self.LogInfo("Evacuating instances from node '%s': %s",
+                   self.op.node_name,
+                   utils.CommaJoin(utils.NiceSort(self.instance_names)))
+    else:
+      self.LogInfo("No instances to evacuate from node '%s'",
+                   self.op.node_name)
 
     if self.op.remote_node is not None:
 
     if self.op.remote_node is not None:
-      result = []
-      for i in instances:
+      for i in self.instances:
         if i.primary_node == self.op.remote_node:
           raise errors.OpPrereqError("Node %s is the primary node of"
                                      " instance %s, cannot use it as"
                                      " secondary" %
                                      (self.op.remote_node, i.name),
                                      errors.ECODE_INVAL)
         if i.primary_node == self.op.remote_node:
           raise errors.OpPrereqError("Node %s is the primary node of"
                                      " instance %s, cannot use it as"
                                      " secondary" %
                                      (self.op.remote_node, i.name),
                                      errors.ECODE_INVAL)
-        result.append([i.name, self.op.remote_node])
-    else:
-      ial = IAllocator(self.cfg, self.rpc,
-                       mode=constants.IALLOCATOR_MODE_MEVAC,
-                       evac_nodes=self.op.nodes)
-      ial.Run(self.op.iallocator, validate=True)
+
+  def Exec(self, feedback_fn):
+    assert (self.op.iallocator is not None) ^ (self.op.remote_node is not None)
+
+    if not self.instance_names:
+      # No instances to evacuate
+      jobs = []
+
+    elif self.op.iallocator is not None:
+      # TODO: Implement relocation to other group
+      ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_NODE_EVAC,
+                       evac_mode=self._MODE2IALLOCATOR[self.op.mode],
+                       instances=list(self.instance_names))
+
+      ial.Run(self.op.iallocator)
+
       if not ial.success:
       if not ial.success:
-        raise errors.OpExecError("No valid evacuation solution: %s" % ial.info,
-                                 errors.ECODE_NORES)
-      result = ial.result
-    return result
+        raise errors.OpPrereqError("Can't compute node evacuation using"
+                                   " iallocator '%s': %s" %
+                                   (self.op.iallocator, ial.info),
+                                   errors.ECODE_NORES)
+
+      jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, True)
+
+    elif self.op.remote_node is not None:
+      assert self.op.mode == constants.NODE_EVAC_SEC
+      jobs = [
+        [opcodes.OpInstanceReplaceDisks(instance_name=instance_name,
+                                        remote_node=self.op.remote_node,
+                                        disks=[],
+                                        mode=constants.REPLACE_DISK_CHG,
+                                        early_release=self.op.early_release)]
+        for instance_name in self.instance_names
+        ]
+
+    else:
+      raise errors.ProgrammerError("No iallocator or remote node")
+
+    return ResultWithJobs(jobs)
+
+
+def _SetOpEarlyRelease(early_release, op):
+  """Sets C{early_release} flag on opcodes if available.
+
+  """
+  try:
+    op.early_release = early_release
+  except AttributeError:
+    assert not isinstance(op, opcodes.OpInstanceReplaceDisks)
+
+  return op
+
+
+def _NodeEvacDest(use_nodes, group, nodes):
+  """Returns group or nodes depending on caller's choice.
+
+  """
+  if use_nodes:
+    return utils.CommaJoin(nodes)
+  else:
+    return group
+
+
+def _LoadNodeEvacResult(lu, alloc_result, early_release, use_nodes):
+  """Unpacks the result of change-group and node-evacuate iallocator requests.
+
+  Iallocator modes L{constants.IALLOCATOR_MODE_NODE_EVAC} and
+  L{constants.IALLOCATOR_MODE_CHG_GROUP}.
+
+  @type lu: L{LogicalUnit}
+  @param lu: Logical unit instance
+  @type alloc_result: tuple/list
+  @param alloc_result: Result from iallocator
+  @type early_release: bool
+  @param early_release: Whether to release locks early if possible
+  @type use_nodes: bool
+  @param use_nodes: Whether to display node names instead of groups
+
+  """
+  (moved, failed, jobs) = alloc_result
+
+  if failed:
+    failreason = utils.CommaJoin("%s (%s)" % (name, reason)
+                                 for (name, reason) in failed)
+    lu.LogWarning("Unable to evacuate instances %s", failreason)
+    raise errors.OpExecError("Unable to evacuate instances %s" % failreason)
+
+  if moved:
+    lu.LogInfo("Instances to be moved: %s",
+               utils.CommaJoin("%s (to %s)" %
+                               (name, _NodeEvacDest(use_nodes, group, nodes))
+                               for (name, group, nodes) in moved))
+
+  return [map(compat.partial(_SetOpEarlyRelease, early_release),
+              map(opcodes.OpCode.LoadOpCode, ops))
+          for ops in jobs]
 
 
 class LUInstanceGrowDisk(LogicalUnit):
 
 
 class LUInstanceGrowDisk(LogicalUnit):
@@ -9776,11 +11086,16 @@ class LUInstanceGrowDisk(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+    self.needed_locks[locking.LEVEL_NODE_RES] = []
+    self.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes()
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
       self._LockInstancesNodes()
+    elif level == locking.LEVEL_NODE_RES:
+      # Copy node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE][:]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -9837,10 +11152,18 @@ class LUInstanceGrowDisk(LogicalUnit):
     instance = self.instance
     disk = self.disk
 
     instance = self.instance
     disk = self.disk
 
+    assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+    assert (self.owned_locks(locking.LEVEL_NODE) ==
+            self.owned_locks(locking.LEVEL_NODE_RES))
+
     disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
     if not disks_ok:
       raise errors.OpExecError("Cannot activate block device to grow")
 
     disks_ok, _ = _AssembleInstanceDisks(self, self.instance, disks=[disk])
     if not disks_ok:
       raise errors.OpExecError("Cannot activate block device to grow")
 
+    feedback_fn("Growing disk %s of instance '%s' by %s" %
+                (self.op.disk, instance.name,
+                 utils.FormatUnit(self.op.amount, "h")))
+
     # First run all grow ops in dry-run mode
     for node in instance.all_nodes:
       self.cfg.SetDiskID(disk, node)
     # First run all grow ops in dry-run mode
     for node in instance.all_nodes:
       self.cfg.SetDiskID(disk, node)
@@ -9863,18 +11186,28 @@ class LUInstanceGrowDisk(LogicalUnit):
 
     disk.RecordGrow(self.op.amount)
     self.cfg.Update(instance, feedback_fn)
 
     disk.RecordGrow(self.op.amount)
     self.cfg.Update(instance, feedback_fn)
+
+    # Changes have been recorded, release node lock
+    _ReleaseLocks(self, locking.LEVEL_NODE)
+
+    # Downgrade lock while waiting for sync
+    self.glm.downgrade(locking.LEVEL_INSTANCE)
+
     if self.op.wait_for_sync:
       disk_abort = not _WaitForSync(self, instance, disks=[disk])
       if disk_abort:
         self.proc.LogWarning("Disk sync-ing has not returned a good"
                              " status; please check the instance")
     if self.op.wait_for_sync:
       disk_abort = not _WaitForSync(self, instance, disks=[disk])
       if disk_abort:
         self.proc.LogWarning("Disk sync-ing has not returned a good"
                              " status; please check the instance")
-      if not instance.admin_up:
+      if instance.admin_state != constants.ADMINST_UP:
         _SafeShutdownInstanceDisks(self, instance, disks=[disk])
         _SafeShutdownInstanceDisks(self, instance, disks=[disk])
-    elif not instance.admin_up:
+    elif instance.admin_state != constants.ADMINST_UP:
       self.proc.LogWarning("Not shutting down the disk even if the instance is"
                            " not supposed to be running because no wait for"
                            " sync mode was requested")
 
       self.proc.LogWarning("Not shutting down the disk even if the instance is"
                            " not supposed to be running because no wait for"
                            " sync mode was requested")
 
+    assert self.owned_locks(locking.LEVEL_NODE_RES)
+    assert set([instance.name]) == self.owned_locks(locking.LEVEL_INSTANCE)
+
 
 class LUInstanceQueryData(NoHooksLU):
   """Query runtime instance data.
 
 class LUInstanceQueryData(NoHooksLU):
   """Query runtime instance data.
@@ -9898,7 +11231,7 @@ class LUInstanceQueryData(NoHooksLU):
       self.wanted_names = None
 
     if self.op.use_locking:
       self.wanted_names = None
 
     if self.op.use_locking:
-      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+      self.share_locks = _ShareAll()
 
       if self.wanted_names is None:
         self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
 
       if self.wanted_names is None:
         self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
@@ -9906,7 +11239,6 @@ class LUInstanceQueryData(NoHooksLU):
         self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
 
       self.needed_locks[locking.LEVEL_NODE] = []
         self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
 
       self.needed_locks[locking.LEVEL_NODE] = []
-      self.share_locks = dict.fromkeys(locking.LEVELS, 1)
       self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
       self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
@@ -9921,10 +11253,10 @@ class LUInstanceQueryData(NoHooksLU):
     """
     if self.wanted_names is None:
       assert self.op.use_locking, "Locking was not used"
     """
     if self.wanted_names is None:
       assert self.op.use_locking, "Locking was not used"
-      self.wanted_names = self.glm.list_owned(locking.LEVEL_INSTANCE)
+      self.wanted_names = self.owned_locks(locking.LEVEL_INSTANCE)
 
 
-    self.wanted_instances = [self.cfg.GetInstanceInfo(name)
-                             for name in self.wanted_names]
+    self.wanted_instances = \
+        map(compat.snd, self.cfg.GetMultiInstanceInfo(self.wanted_names))
 
   def _ComputeBlockdevStatus(self, node, instance_name, dev):
     """Returns the status of a block device
 
   def _ComputeBlockdevStatus(self, node, instance_name, dev):
     """Returns the status of a block device
@@ -9965,8 +11297,9 @@ class LUInstanceQueryData(NoHooksLU):
     dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
 
     if dev.children:
     dev_sstatus = self._ComputeBlockdevStatus(snode, instance.name, dev)
 
     if dev.children:
-      dev_children = [self._ComputeDiskStatus(instance, snode, child)
-                      for child in dev.children]
+      dev_children = map(compat.partial(self._ComputeDiskStatus,
+                                        instance, snode),
+                         dev.children)
     else:
       dev_children = []
 
     else:
       dev_children = []
 
@@ -9988,8 +11321,16 @@ class LUInstanceQueryData(NoHooksLU):
 
     cluster = self.cfg.GetClusterInfo()
 
 
     cluster = self.cfg.GetClusterInfo()
 
-    for instance in self.wanted_instances:
-      if not self.op.static:
+    pri_nodes = self.cfg.GetMultiNodeInfo(i.primary_node
+                                          for i in self.wanted_instances)
+    for instance, (_, pnode) in zip(self.wanted_instances, pri_nodes):
+      if self.op.static or pnode.offline:
+        remote_state = None
+        if pnode.offline:
+          self.LogWarning("Primary node %s is marked offline, returning static"
+                          " information only for instance %s" %
+                          (pnode.name, instance.name))
+      else:
         remote_info = self.rpc.call_instance_info(instance.primary_node,
                                                   instance.name,
                                                   instance.hypervisor)
         remote_info = self.rpc.call_instance_info(instance.primary_node,
                                                   instance.name,
                                                   instance.hypervisor)
@@ -9998,20 +11339,17 @@ class LUInstanceQueryData(NoHooksLU):
         if remote_info and "state" in remote_info:
           remote_state = "up"
         else:
         if remote_info and "state" in remote_info:
           remote_state = "up"
         else:
-          remote_state = "down"
-      else:
-        remote_state = None
-      if instance.admin_up:
-        config_state = "up"
-      else:
-        config_state = "down"
+          if instance.admin_state == constants.ADMINST_UP:
+            remote_state = "down"
+          else:
+            remote_state = instance.admin_state
 
 
-      disks = [self._ComputeDiskStatus(instance, None, device)
-               for device in instance.disks]
+      disks = map(compat.partial(self._ComputeDiskStatus, instance, None),
+                  instance.disks)
 
       result[instance.name] = {
         "name": instance.name,
 
       result[instance.name] = {
         "name": instance.name,
-        "config_state": config_state,
+        "config_state": instance.admin_state,
         "run_state": remote_state,
         "pnode": instance.primary_node,
         "snodes": instance.secondary_nodes,
         "run_state": remote_state,
         "pnode": instance.primary_node,
         "snodes": instance.secondary_nodes,
@@ -10047,7 +11385,8 @@ class LUInstanceSetParams(LogicalUnit):
 
   def CheckArguments(self):
     if not (self.op.nics or self.op.disks or self.op.disk_template or
 
   def CheckArguments(self):
     if not (self.op.nics or self.op.disks or self.op.disk_template or
-            self.op.hvparams or self.op.beparams or self.op.os_name):
+            self.op.hvparams or self.op.beparams or self.op.os_name or
+            self.op.online_inst or self.op.offline_inst):
       raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
 
     if self.op.hvparams:
       raise errors.OpPrereqError("No changes submitted", errors.ECODE_INVAL)
 
     if self.op.hvparams:
@@ -10132,13 +11471,13 @@ class LUInstanceSetParams(LogicalUnit):
             raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
                                        errors.ECODE_INVAL)
 
             raise errors.OpPrereqError("Invalid IP address '%s'" % nic_ip,
                                        errors.ECODE_INVAL)
 
-      nic_bridge = nic_dict.get('bridge', None)
+      nic_bridge = nic_dict.get("bridge", None)
       nic_link = nic_dict.get(constants.INIC_LINK, None)
       if nic_bridge and nic_link:
         raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
                                    " at the same time", errors.ECODE_INVAL)
       elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
       nic_link = nic_dict.get(constants.INIC_LINK, None)
       if nic_bridge and nic_link:
         raise errors.OpPrereqError("Cannot pass 'bridge' and 'link'"
                                    " at the same time", errors.ECODE_INVAL)
       elif nic_bridge and nic_bridge.lower() == constants.VALUE_NONE:
-        nic_dict['bridge'] = None
+        nic_dict["bridge"] = None
       elif nic_link and nic_link.lower() == constants.VALUE_NONE:
         nic_dict[constants.INIC_LINK] = None
 
       elif nic_link and nic_link.lower() == constants.VALUE_NONE:
         nic_dict[constants.INIC_LINK] = None
 
@@ -10163,7 +11502,10 @@ class LUInstanceSetParams(LogicalUnit):
 
   def ExpandNames(self):
     self._ExpandAndLockInstance()
 
   def ExpandNames(self):
     self._ExpandAndLockInstance()
+    # Can't even acquire node locks in shared mode as upcoming changes in
+    # Ganeti 2.6 will start to modify the node object on disk conversion
     self.needed_locks[locking.LEVEL_NODE] = []
     self.needed_locks[locking.LEVEL_NODE] = []
+    self.needed_locks[locking.LEVEL_NODE_RES] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
@@ -10172,6 +11514,10 @@ class LUInstanceSetParams(LogicalUnit):
       if self.op.disk_template and self.op.remote_node:
         self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
         self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
       if self.op.disk_template and self.op.remote_node:
         self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
         self.needed_locks[locking.LEVEL_NODE].append(self.op.remote_node)
+    elif level == locking.LEVEL_NODE_RES and self.op.disk_template:
+      # Copy node locks
+      self.needed_locks[locking.LEVEL_NODE_RES] = \
+        self.needed_locks[locking.LEVEL_NODE][:]
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -10180,14 +11526,16 @@ class LUInstanceSetParams(LogicalUnit):
 
     """
     args = dict()
 
     """
     args = dict()
-    if constants.BE_MEMORY in self.be_new:
-      args['memory'] = self.be_new[constants.BE_MEMORY]
+    if constants.BE_MINMEM in self.be_new:
+      args["minmem"] = self.be_new[constants.BE_MINMEM]
+    if constants.BE_MAXMEM in self.be_new:
+      args["maxmem"] = self.be_new[constants.BE_MAXMEM]
     if constants.BE_VCPUS in self.be_new:
     if constants.BE_VCPUS in self.be_new:
-      args['vcpus'] = self.be_new[constants.BE_VCPUS]
+      args["vcpus"] = self.be_new[constants.BE_VCPUS]
     # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
     # information at all.
     if self.op.nics:
     # TODO: export disk changes. Note: _BuildInstanceHookEnv* don't export disk
     # information at all.
     if self.op.nics:
-      args['nics'] = []
+      args["nics"] = []
       nic_override = dict(self.op.nics)
       for idx, nic in enumerate(self.instance.nics):
         if idx in nic_override:
       nic_override = dict(self.op.nics)
       for idx, nic in enumerate(self.instance.nics):
         if idx in nic_override:
@@ -10208,16 +11556,16 @@ class LUInstanceSetParams(LogicalUnit):
           nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
         mode = nicparams[constants.NIC_MODE]
         link = nicparams[constants.NIC_LINK]
           nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
         mode = nicparams[constants.NIC_MODE]
         link = nicparams[constants.NIC_LINK]
-        args['nics'].append((ip, mac, mode, link))
+        args["nics"].append((ip, mac, mode, link))
       if constants.DDM_ADD in nic_override:
         ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
         mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
         nicparams = self.nic_pnew[constants.DDM_ADD]
         mode = nicparams[constants.NIC_MODE]
         link = nicparams[constants.NIC_LINK]
       if constants.DDM_ADD in nic_override:
         ip = nic_override[constants.DDM_ADD].get(constants.INIC_IP, None)
         mac = nic_override[constants.DDM_ADD][constants.INIC_MAC]
         nicparams = self.nic_pnew[constants.DDM_ADD]
         mode = nicparams[constants.NIC_MODE]
         link = nicparams[constants.NIC_LINK]
-        args['nics'].append((ip, mac, mode, link))
+        args["nics"].append((ip, mac, mode, link))
       elif constants.DDM_REMOVE in nic_override:
       elif constants.DDM_REMOVE in nic_override:
-        del args['nics'][-1]
+        del args["nics"][-1]
 
     env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
     if self.op.disk_template:
 
     env = _BuildInstanceHookEnvByObject(self, self.instance, override=args)
     if self.op.disk_template:
@@ -10246,6 +11594,8 @@ class LUInstanceSetParams(LogicalUnit):
       "Cannot retrieve locked instance %s" % self.op.instance_name
     pnode = instance.primary_node
     nodelist = list(instance.all_nodes)
       "Cannot retrieve locked instance %s" % self.op.instance_name
     pnode = instance.primary_node
     nodelist = list(instance.all_nodes)
+    pnode_info = self.cfg.GetNodeInfo(pnode)
+    self.diskparams = self.cfg.GetNodeGroup(pnode_info.group).diskparams
 
     # OS change
     if self.op.os_name and not self.op.force:
 
     # OS change
     if self.op.os_name and not self.op.force:
@@ -10266,7 +11616,8 @@ class LUInstanceSetParams(LogicalUnit):
                                    " %s to %s" % (instance.disk_template,
                                                   self.op.disk_template),
                                    errors.ECODE_INVAL)
                                    " %s to %s" % (instance.disk_template,
                                                   self.op.disk_template),
                                    errors.ECODE_INVAL)
-      _CheckInstanceDown(self, instance, "cannot change disk template")
+      _CheckInstanceState(self, instance, INSTANCE_DOWN,
+                          msg="cannot change disk template")
       if self.op.disk_template in constants.DTS_INT_MIRROR:
         if self.op.remote_node == pnode:
           raise errors.OpPrereqError("Given new secondary node %s is the same"
       if self.op.disk_template in constants.DTS_INT_MIRROR:
         if self.op.remote_node == pnode:
           raise errors.OpPrereqError("Given new secondary node %s is the same"
@@ -10282,6 +11633,13 @@ class LUInstanceSetParams(LogicalUnit):
         required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
         _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
 
         required = _ComputeDiskSizePerVG(self.op.disk_template, disks)
         _CheckNodesFreeDiskPerVG(self, [self.op.remote_node], required)
 
+        snode_info = self.cfg.GetNodeInfo(self.op.remote_node)
+        if pnode_info.group != snode_info.group:
+          self.LogWarning("The primary and secondary nodes are in two"
+                          " different node groups; the disk parameters"
+                          " from the first disk's node group will be"
+                          " used")
+
     # hvparams processing
     if self.op.hvparams:
       hv_type = instance.hypervisor
     # hvparams processing
     if self.op.hvparams:
       hv_type = instance.hypervisor
@@ -10292,23 +11650,54 @@ class LUInstanceSetParams(LogicalUnit):
       # local check
       hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
       _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
       # local check
       hypervisor.GetHypervisor(hv_type).CheckParameterSyntax(hv_new)
       _CheckHVParams(self, nodelist, instance.hypervisor, hv_new)
-      self.hv_new = hv_new # the new actual values
+      self.hv_proposed = self.hv_new = hv_new # the new actual values
       self.hv_inst = i_hvdict # the new dict (without defaults)
     else:
       self.hv_inst = i_hvdict # the new dict (without defaults)
     else:
+      self.hv_proposed = cluster.SimpleFillHV(instance.hypervisor, instance.os,
+                                              instance.hvparams)
       self.hv_new = self.hv_inst = {}
 
     # beparams processing
     if self.op.beparams:
       i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
                                    use_none=True)
       self.hv_new = self.hv_inst = {}
 
     # beparams processing
     if self.op.beparams:
       i_bedict = _GetUpdatedParams(instance.beparams, self.op.beparams,
                                    use_none=True)
+      objects.UpgradeBeParams(i_bedict)
       utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
       be_new = cluster.SimpleFillBE(i_bedict)
       utils.ForceDictType(i_bedict, constants.BES_PARAMETER_TYPES)
       be_new = cluster.SimpleFillBE(i_bedict)
-      self.be_new = be_new # the new actual values
+      self.be_proposed = self.be_new = be_new # the new actual values
       self.be_inst = i_bedict # the new dict (without defaults)
     else:
       self.be_new = self.be_inst = {}
       self.be_inst = i_bedict # the new dict (without defaults)
     else:
       self.be_new = self.be_inst = {}
+      self.be_proposed = cluster.SimpleFillBE(instance.beparams)
     be_old = cluster.FillBE(instance)
 
     be_old = cluster.FillBE(instance)
 
+    # CPU param validation -- checking every time a paramtere is
+    # changed to cover all cases where either CPU mask or vcpus have
+    # changed
+    if (constants.BE_VCPUS in self.be_proposed and
+        constants.HV_CPU_MASK in self.hv_proposed):
+      cpu_list = \
+        utils.ParseMultiCpuMask(self.hv_proposed[constants.HV_CPU_MASK])
+      # Verify mask is consistent with number of vCPUs. Can skip this
+      # test if only 1 entry in the CPU mask, which means same mask
+      # is applied to all vCPUs.
+      if (len(cpu_list) > 1 and
+          len(cpu_list) != self.be_proposed[constants.BE_VCPUS]):
+        raise errors.OpPrereqError("Number of vCPUs [%d] does not match the"
+                                   " CPU mask [%s]" %
+                                   (self.be_proposed[constants.BE_VCPUS],
+                                    self.hv_proposed[constants.HV_CPU_MASK]),
+                                   errors.ECODE_INVAL)
+
+      # Only perform this test if a new CPU mask is given
+      if constants.HV_CPU_MASK in self.hv_new:
+        # Calculate the largest CPU number requested
+        max_requested_cpu = max(map(max, cpu_list))
+        # Check that all of the instance's nodes have enough physical CPUs to
+        # satisfy the requested CPU mask
+        _CheckNodesPhysicalCPUs(self, instance.all_nodes,
+                                max_requested_cpu + 1, instance.hypervisor)
+
     # osparams processing
     if self.op.osparams:
       i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
     # osparams processing
     if self.op.osparams:
       i_osdict = _GetUpdatedParams(instance.osparams, self.op.osparams)
@@ -10319,8 +11708,9 @@ class LUInstanceSetParams(LogicalUnit):
 
     self.warn = []
 
 
     self.warn = []
 
-    if (constants.BE_MEMORY in self.op.beparams and not self.op.force and
-        be_new[constants.BE_MEMORY] > be_old[constants.BE_MEMORY]):
+    #TODO(dynmem): do the appropriate check involving MINMEM
+    if (constants.BE_MAXMEM in self.op.beparams and not self.op.force and
+        be_new[constants.BE_MAXMEM] > be_old[constants.BE_MAXMEM]):
       mem_check_list = [pnode]
       if be_new[constants.BE_AUTO_BALANCE]:
         # either we changed auto_balance to yes or it was from before
       mem_check_list = [pnode]
       if be_new[constants.BE_AUTO_BALANCE]:
         # either we changed auto_balance to yes or it was from before
@@ -10328,34 +11718,39 @@ class LUInstanceSetParams(LogicalUnit):
       instance_info = self.rpc.call_instance_info(pnode, instance.name,
                                                   instance.hypervisor)
       nodeinfo = self.rpc.call_node_info(mem_check_list, None,
       instance_info = self.rpc.call_instance_info(pnode, instance.name,
                                                   instance.hypervisor)
       nodeinfo = self.rpc.call_node_info(mem_check_list, None,
-                                         instance.hypervisor)
+                                         [instance.hypervisor])
       pninfo = nodeinfo[pnode]
       msg = pninfo.fail_msg
       if msg:
         # Assume the primary node is unreachable and go ahead
         self.warn.append("Can't get info from primary node %s: %s" %
       pninfo = nodeinfo[pnode]
       msg = pninfo.fail_msg
       if msg:
         # Assume the primary node is unreachable and go ahead
         self.warn.append("Can't get info from primary node %s: %s" %
-                         (pnode,  msg))
-      elif not isinstance(pninfo.payload.get('memory_free', None), int):
-        self.warn.append("Node data from primary node %s doesn't contain"
-                         " free memory information" % pnode)
-      elif instance_info.fail_msg:
-        self.warn.append("Can't get instance runtime information: %s" %
-                        instance_info.fail_msg)
+                         (pnode, msg))
       else:
       else:
-        if instance_info.payload:
-          current_mem = int(instance_info.payload['memory'])
+        (_, _, (pnhvinfo, )) = pninfo.payload
+        if not isinstance(pnhvinfo.get("memory_free", None), int):
+          self.warn.append("Node data from primary node %s doesn't contain"
+                           " free memory information" % pnode)
+        elif instance_info.fail_msg:
+          self.warn.append("Can't get instance runtime information: %s" %
+                          instance_info.fail_msg)
         else:
         else:
-          # Assume instance not running
-          # (there is a slight race condition here, but it's not very probable,
-          # and we have no other way to check)
-          current_mem = 0
-        miss_mem = (be_new[constants.BE_MEMORY] - current_mem -
-                    pninfo.payload['memory_free'])
-        if miss_mem > 0:
-          raise errors.OpPrereqError("This change will prevent the instance"
-                                     " from starting, due to %d MB of memory"
-                                     " missing on its primary node" % miss_mem,
-                                     errors.ECODE_NORES)
+          if instance_info.payload:
+            current_mem = int(instance_info.payload["memory"])
+          else:
+            # Assume instance not running
+            # (there is a slight race condition here, but it's not very
+            # probable, and we have no other way to check)
+            # TODO: Describe race condition
+            current_mem = 0
+          #TODO(dynmem): do the appropriate check involving MINMEM
+          miss_mem = (be_new[constants.BE_MAXMEM] - current_mem -
+                      pnhvinfo["memory_free"])
+          if miss_mem > 0:
+            raise errors.OpPrereqError("This change will prevent the instance"
+                                       " from starting, due to %d MB of memory"
+                                       " missing on its primary node" %
+                                       miss_mem,
+                                       errors.ECODE_NORES)
 
       if be_new[constants.BE_AUTO_BALANCE]:
         for node, nres in nodeinfo.items():
 
       if be_new[constants.BE_AUTO_BALANCE]:
         for node, nres in nodeinfo.items():
@@ -10363,11 +11758,13 @@ class LUInstanceSetParams(LogicalUnit):
             continue
           nres.Raise("Can't get info from secondary node %s" % node,
                      prereq=True, ecode=errors.ECODE_STATE)
             continue
           nres.Raise("Can't get info from secondary node %s" % node,
                      prereq=True, ecode=errors.ECODE_STATE)
-          if not isinstance(nres.payload.get('memory_free', None), int):
+          (_, _, (nhvinfo, )) = nres.payload
+          if not isinstance(nhvinfo.get("memory_free", None), int):
             raise errors.OpPrereqError("Secondary node %s didn't return free"
                                        " memory information" % node,
                                        errors.ECODE_STATE)
             raise errors.OpPrereqError("Secondary node %s didn't return free"
                                        " memory information" % node,
                                        errors.ECODE_STATE)
-          elif be_new[constants.BE_MEMORY] > nres.payload['memory_free']:
+          #TODO(dynmem): do the appropriate check involving MINMEM
+          elif be_new[constants.BE_MAXMEM] > nhvinfo["memory_free"]:
             raise errors.OpPrereqError("This change will prevent the instance"
                                        " from failover to its secondary node"
                                        " %s, due to not enough memory" % node,
             raise errors.OpPrereqError("This change will prevent the instance"
                                        " from failover to its secondary node"
                                        " %s, due to not enough memory" % node,
@@ -10403,8 +11800,8 @@ class LUInstanceSetParams(LogicalUnit):
                                  for key in constants.NICS_PARAMETERS
                                  if key in nic_dict])
 
                                  for key in constants.NICS_PARAMETERS
                                  if key in nic_dict])
 
-      if 'bridge' in nic_dict:
-        update_params_dict[constants.NIC_LINK] = nic_dict['bridge']
+      if "bridge" in nic_dict:
+        update_params_dict[constants.NIC_LINK] = nic_dict["bridge"]
 
       new_nic_params = _GetUpdatedParams(old_nic_params,
                                          update_params_dict)
 
       new_nic_params = _GetUpdatedParams(old_nic_params,
                                          update_params_dict)
@@ -10430,12 +11827,12 @@ class LUInstanceSetParams(LogicalUnit):
         else:
           nic_ip = old_nic_ip
         if nic_ip is None:
         else:
           nic_ip = old_nic_ip
         if nic_ip is None:
-          raise errors.OpPrereqError('Cannot set the nic ip to None'
-                                     ' on a routed nic', errors.ECODE_INVAL)
+          raise errors.OpPrereqError("Cannot set the nic ip to None"
+                                     " on a routed nic", errors.ECODE_INVAL)
       if constants.INIC_MAC in nic_dict:
         nic_mac = nic_dict[constants.INIC_MAC]
         if nic_mac is None:
       if constants.INIC_MAC in nic_dict:
         nic_mac = nic_dict[constants.INIC_MAC]
         if nic_mac is None:
-          raise errors.OpPrereqError('Cannot set the nic mac to None',
+          raise errors.OpPrereqError("Cannot set the nic mac to None",
                                      errors.ECODE_INVAL)
         elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
           # otherwise generate the mac
                                      errors.ECODE_INVAL)
         elif nic_mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
           # otherwise generate the mac
@@ -10460,7 +11857,8 @@ class LUInstanceSetParams(LogicalUnit):
         if len(instance.disks) == 1:
           raise errors.OpPrereqError("Cannot remove the last disk of"
                                      " an instance", errors.ECODE_INVAL)
         if len(instance.disks) == 1:
           raise errors.OpPrereqError("Cannot remove the last disk of"
                                      " an instance", errors.ECODE_INVAL)
-        _CheckInstanceDown(self, instance, "cannot remove disks")
+        _CheckInstanceState(self, instance, INSTANCE_DOWN,
+                            msg="cannot remove disks")
 
       if (disk_op == constants.DDM_ADD and
           len(instance.disks) >= constants.MAX_DISKS):
 
       if (disk_op == constants.DDM_ADD and
           len(instance.disks) >= constants.MAX_DISKS):
@@ -10475,7 +11873,15 @@ class LUInstanceSetParams(LogicalUnit):
                                      (disk_op, len(instance.disks)),
                                      errors.ECODE_INVAL)
 
                                      (disk_op, len(instance.disks)),
                                      errors.ECODE_INVAL)
 
-    return
+    # disabling the instance
+    if self.op.offline_inst:
+      _CheckInstanceState(self, instance, INSTANCE_DOWN,
+                          msg="cannot change instance state to offline")
+
+    # enabling the instance
+    if self.op.online_inst:
+      _CheckInstanceState(self, instance, INSTANCE_OFFLINE,
+                          msg="cannot make instance go online")
 
   def _ConvertPlainToDrbd(self, feedback_fn):
     """Converts an instance from plain to drbd.
 
   def _ConvertPlainToDrbd(self, feedback_fn):
     """Converts an instance from plain to drbd.
@@ -10486,13 +11892,16 @@ class LUInstanceSetParams(LogicalUnit):
     pnode = instance.primary_node
     snode = self.op.remote_node
 
     pnode = instance.primary_node
     snode = self.op.remote_node
 
+    assert instance.disk_template == constants.DT_PLAIN
+
     # create a fake disk info for _GenerateDiskTemplate
     disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
                   constants.IDISK_VG: d.logical_id[0]}
                  for d in instance.disks]
     new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
                                       instance.name, pnode, [snode],
     # create a fake disk info for _GenerateDiskTemplate
     disk_info = [{constants.IDISK_SIZE: d.size, constants.IDISK_MODE: d.mode,
                   constants.IDISK_VG: d.logical_id[0]}
                  for d in instance.disks]
     new_disks = _GenerateDiskTemplate(self, self.op.disk_template,
                                       instance.name, pnode, [snode],
-                                      disk_info, None, None, 0, feedback_fn)
+                                      disk_info, None, None, 0, feedback_fn,
+                                      self.diskparams)
     info = _GetInstanceInfoText(instance)
     feedback_fn("Creating aditional volumes...")
     # first, create the missing data and meta devices
     info = _GetInstanceInfoText(instance)
     feedback_fn("Creating aditional volumes...")
     # first, create the missing data and meta devices
@@ -10522,6 +11931,9 @@ class LUInstanceSetParams(LogicalUnit):
     instance.disks = new_disks
     self.cfg.Update(instance, feedback_fn)
 
     instance.disks = new_disks
     self.cfg.Update(instance, feedback_fn)
 
+    # Release node locks while waiting for sync
+    _ReleaseLocks(self, locking.LEVEL_NODE)
+
     # disks are created, waiting for sync
     disk_abort = not _WaitForSync(self, instance,
                                   oneshot=not self.op.wait_for_sync)
     # disks are created, waiting for sync
     disk_abort = not _WaitForSync(self, instance,
                                   oneshot=not self.op.wait_for_sync)
@@ -10529,12 +11941,17 @@ class LUInstanceSetParams(LogicalUnit):
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance, please cleanup manually")
 
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance, please cleanup manually")
 
+    # Node resource locks will be released by caller
+
   def _ConvertDrbdToPlain(self, feedback_fn):
     """Converts an instance from drbd to plain.
 
     """
     instance = self.instance
   def _ConvertDrbdToPlain(self, feedback_fn):
     """Converts an instance from drbd to plain.
 
     """
     instance = self.instance
+
     assert len(instance.secondary_nodes) == 1
     assert len(instance.secondary_nodes) == 1
+    assert instance.disk_template == constants.DT_DRBD8
+
     pnode = instance.primary_node
     snode = instance.secondary_nodes[0]
     feedback_fn("Converting template to plain")
     pnode = instance.primary_node
     snode = instance.secondary_nodes[0]
     feedback_fn("Converting template to plain")
@@ -10552,6 +11969,9 @@ class LUInstanceSetParams(LogicalUnit):
     instance.disk_template = constants.DT_PLAIN
     self.cfg.Update(instance, feedback_fn)
 
     instance.disk_template = constants.DT_PLAIN
     self.cfg.Update(instance, feedback_fn)
 
+    # Release locks in case removing disks takes a while
+    _ReleaseLocks(self, locking.LEVEL_NODE)
+
     feedback_fn("Removing volumes on the secondary node...")
     for disk in old_disks:
       self.cfg.SetDiskID(disk, snode)
     feedback_fn("Removing volumes on the secondary node...")
     for disk in old_disks:
       self.cfg.SetDiskID(disk, snode)
@@ -10569,6 +11989,13 @@ class LUInstanceSetParams(LogicalUnit):
         self.LogWarning("Could not remove metadata for disk %d on node %s,"
                         " continuing anyway: %s", idx, pnode, msg)
 
         self.LogWarning("Could not remove metadata for disk %d on node %s,"
                         " continuing anyway: %s", idx, pnode, msg)
 
+    # this is a DRBD disk, return its port to the pool
+    for disk in old_disks:
+      tcp_port = disk.logical_id[2]
+      self.cfg.AddTcpUdpPort(tcp_port)
+
+    # Node resource locks will be released by caller
+
   def Exec(self, feedback_fn):
     """Modifies an instance.
 
   def Exec(self, feedback_fn):
     """Modifies an instance.
 
@@ -10580,6 +12007,10 @@ class LUInstanceSetParams(LogicalUnit):
     for warn in self.warn:
       feedback_fn("WARNING: %s" % warn)
 
     for warn in self.warn:
       feedback_fn("WARNING: %s" % warn)
 
+    assert ((self.op.disk_template is None) ^
+            bool(self.owned_locks(locking.LEVEL_NODE_RES))), \
+      "Not owning any node resource locks"
+
     result = []
     instance = self.instance
     # disk changes
     result = []
     instance = self.instance
     # disk changes
@@ -10595,6 +12026,11 @@ class LUInstanceSetParams(LogicalUnit):
             self.LogWarning("Could not remove disk/%d on node %s: %s,"
                             " continuing anyway", device_idx, node, msg)
         result.append(("disk/%d" % device_idx, "remove"))
             self.LogWarning("Could not remove disk/%d on node %s: %s,"
                             " continuing anyway", device_idx, node, msg)
         result.append(("disk/%d" % device_idx, "remove"))
+
+        # if this is a DRBD disk, return its port to the pool
+        if device.dev_type in constants.LDS_DRBD:
+          tcp_port = device.logical_id[2]
+          self.cfg.AddTcpUdpPort(tcp_port)
       elif disk_op == constants.DDM_ADD:
         # add a new disk
         if instance.disk_template in (constants.DT_FILE,
       elif disk_op == constants.DDM_ADD:
         # add a new disk
         if instance.disk_template in (constants.DT_FILE,
@@ -10611,7 +12047,9 @@ class LUInstanceSetParams(LogicalUnit):
                                          [disk_dict],
                                          file_path,
                                          file_driver,
                                          [disk_dict],
                                          file_path,
                                          file_driver,
-                                         disk_idx_base, feedback_fn)[0]
+                                         disk_idx_base,
+                                         feedback_fn,
+                                         self.diskparams)[0]
         instance.disks.append(new_disk)
         info = _GetInstanceInfoText(instance)
 
         instance.disks.append(new_disk)
         info = _GetInstanceInfoText(instance)
 
@@ -10637,6 +12075,16 @@ class LUInstanceSetParams(LogicalUnit):
                        disk_dict[constants.IDISK_MODE]))
 
     if self.op.disk_template:
                        disk_dict[constants.IDISK_MODE]))
 
     if self.op.disk_template:
+      if __debug__:
+        check_nodes = set(instance.all_nodes)
+        if self.op.remote_node:
+          check_nodes.add(self.op.remote_node)
+        for level in [locking.LEVEL_NODE, locking.LEVEL_NODE_RES]:
+          owned = self.owned_locks(level)
+          assert not (check_nodes - owned), \
+            ("Not owning the correct locks, owning %r, expected at least %r" %
+             (owned, check_nodes))
+
       r_shut = _ShutdownInstanceDisks(self, instance)
       if not r_shut:
         raise errors.OpExecError("Cannot shutdown instance disks, unable to"
       r_shut = _ShutdownInstanceDisks(self, instance)
       if not r_shut:
         raise errors.OpExecError("Cannot shutdown instance disks, unable to"
@@ -10649,6 +12097,15 @@ class LUInstanceSetParams(LogicalUnit):
         raise
       result.append(("disk_template", self.op.disk_template))
 
         raise
       result.append(("disk_template", self.op.disk_template))
 
+      assert instance.disk_template == self.op.disk_template, \
+        ("Expected disk template '%s', found '%s'" %
+         (self.op.disk_template, instance.disk_template))
+
+    # Release node and resource locks if there are any (they might already have
+    # been released during disk conversion)
+    _ReleaseLocks(self, locking.LEVEL_NODE)
+    _ReleaseLocks(self, locking.LEVEL_NODE_RES)
+
     # NIC changes
     for nic_op, nic_dict in self.op.nics:
       if nic_op == constants.DDM_REMOVE:
     # NIC changes
     for nic_op, nic_dict in self.op.nics:
       if nic_op == constants.DDM_REMOVE:
@@ -10699,8 +12156,20 @@ class LUInstanceSetParams(LogicalUnit):
       for key, val in self.op.osparams.iteritems():
         result.append(("os/%s" % key, val))
 
       for key, val in self.op.osparams.iteritems():
         result.append(("os/%s" % key, val))
 
+    # online/offline instance
+    if self.op.online_inst:
+      self.cfg.MarkInstanceDown(instance.name)
+      result.append(("admin_state", constants.ADMINST_DOWN))
+    if self.op.offline_inst:
+      self.cfg.MarkInstanceOffline(instance.name)
+      result.append(("admin_state", constants.ADMINST_OFFLINE))
+
     self.cfg.Update(instance, feedback_fn)
 
     self.cfg.Update(instance, feedback_fn)
 
+    assert not (self.owned_locks(locking.LEVEL_NODE_RES) or
+                self.owned_locks(locking.LEVEL_NODE)), \
+      "All node locks should have been released by now"
+
     return result
 
   _DISK_CONVERSIONS = {
     return result
 
   _DISK_CONVERSIONS = {
@@ -10709,6 +12178,147 @@ class LUInstanceSetParams(LogicalUnit):
     }
 
 
     }
 
 
+class LUInstanceChangeGroup(LogicalUnit):
+  HPATH = "instance-change-group"
+  HTYPE = constants.HTYPE_INSTANCE
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.share_locks = _ShareAll()
+    self.needed_locks = {
+      locking.LEVEL_NODEGROUP: [],
+      locking.LEVEL_NODE: [],
+      }
+
+    self._ExpandAndLockInstance()
+
+    if self.op.target_groups:
+      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
+                                  self.op.target_groups)
+    else:
+      self.req_target_uuids = None
+
+    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODEGROUP:
+      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+      if self.req_target_uuids:
+        lock_groups = set(self.req_target_uuids)
+
+        # Lock all groups used by instance optimistically; this requires going
+        # via the node before it's locked, requiring verification later on
+        instance_groups = self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+        lock_groups.update(instance_groups)
+      else:
+        # No target groups, need to lock all of them
+        lock_groups = locking.ALL_SET
+
+      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
+
+    elif level == locking.LEVEL_NODE:
+      if self.req_target_uuids:
+        # Lock all nodes used by instances
+        self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+        self._LockInstancesNodes()
+
+        # Lock all nodes in all potential target groups
+        lock_groups = (frozenset(self.owned_locks(locking.LEVEL_NODEGROUP)) -
+                       self.cfg.GetInstanceNodeGroups(self.op.instance_name))
+        member_nodes = [node_name
+                        for group in lock_groups
+                        for node_name in self.cfg.GetNodeGroup(group).members]
+        self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
+      else:
+        # Lock all nodes as all groups are potential targets
+        self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+  def CheckPrereq(self):
+    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
+
+    assert (self.req_target_uuids is None or
+            owned_groups.issuperset(self.req_target_uuids))
+    assert owned_instances == set([self.op.instance_name])
+
+    # Get instance information
+    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
+    # Check if node groups for locked instance are still correct
+    assert owned_nodes.issuperset(self.instance.all_nodes), \
+      ("Instance %s's nodes changed while we kept the lock" %
+       self.op.instance_name)
+
+    inst_groups = _CheckInstanceNodeGroups(self.cfg, self.op.instance_name,
+                                           owned_groups)
+
+    if self.req_target_uuids:
+      # User requested specific target groups
+      self.target_uuids = self.req_target_uuids
+    else:
+      # All groups except those used by the instance are potential targets
+      self.target_uuids = owned_groups - inst_groups
+
+    conflicting_groups = self.target_uuids & inst_groups
+    if conflicting_groups:
+      raise errors.OpPrereqError("Can't use group(s) '%s' as targets, they are"
+                                 " used by the instance '%s'" %
+                                 (utils.CommaJoin(conflicting_groups),
+                                  self.op.instance_name),
+                                 errors.ECODE_INVAL)
+
+    if not self.target_uuids:
+      raise errors.OpPrereqError("There are no possible target groups",
+                                 errors.ECODE_INVAL)
+
+  def BuildHooksEnv(self):
+    """Build hooks env.
+
+    """
+    assert self.target_uuids
+
+    env = {
+      "TARGET_GROUPS": " ".join(self.target_uuids),
+      }
+
+    env.update(_BuildInstanceHookEnvByObject(self, self.instance))
+
+    return env
+
+  def BuildHooksNodes(self):
+    """Build hooks nodes.
+
+    """
+    mn = self.cfg.GetMasterNode()
+    return ([mn], [mn])
+
+  def Exec(self, feedback_fn):
+    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
+
+    assert instances == [self.op.instance_name], "Instance not locked"
+
+    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
+                     instances=instances, target_groups=list(self.target_uuids))
+
+    ial.Run(self.op.iallocator)
+
+    if not ial.success:
+      raise errors.OpPrereqError("Can't compute solution for changing group of"
+                                 " instance '%s' using iallocator '%s': %s" %
+                                 (self.op.instance_name, self.op.iallocator,
+                                  ial.info),
+                                 errors.ECODE_NORES)
+
+    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
+
+    self.LogInfo("Iallocator returned %s job(s) for changing group of"
+                 " instance '%s'", len(jobs), self.op.instance_name)
+
+    return ResultWithJobs(jobs)
+
+
 class LUBackupQuery(NoHooksLU):
   """Query the exports list
 
 class LUBackupQuery(NoHooksLU):
   """Query the exports list
 
@@ -10733,7 +12343,7 @@ class LUBackupQuery(NoHooksLU):
         that node.
 
     """
         that node.
 
     """
-    self.nodes = self.glm.list_owned(locking.LEVEL_NODE)
+    self.nodes = self.owned_locks(locking.LEVEL_NODE)
     rpcresult = self.rpc.call_export_list(self.nodes)
     result = {}
     for node in rpcresult:
     rpcresult = self.rpc.call_export_list(self.nodes)
     result = {}
     for node in rpcresult:
@@ -10882,7 +12492,8 @@ class LUBackupExport(LogicalUnit):
           "Cannot retrieve locked instance %s" % self.op.instance_name
     _CheckNodeOnline(self, self.instance.primary_node)
 
           "Cannot retrieve locked instance %s" % self.op.instance_name
     _CheckNodeOnline(self, self.instance.primary_node)
 
-    if (self.op.remove_instance and self.instance.admin_up and
+    if (self.op.remove_instance and
+        self.instance.admin_state == constants.ADMINST_UP and
         not self.op.shutdown):
       raise errors.OpPrereqError("Can not remove instance without shutting it"
                                  " down before")
         not self.op.shutdown):
       raise errors.OpPrereqError("Can not remove instance without shutting it"
                                  " down before")
@@ -11012,7 +12623,7 @@ class LUBackupExport(LogicalUnit):
     for disk in instance.disks:
       self.cfg.SetDiskID(disk, src_node)
 
     for disk in instance.disks:
       self.cfg.SetDiskID(disk, src_node)
 
-    activate_disks = (not instance.admin_up)
+    activate_disks = (instance.admin_state != constants.ADMINST_UP)
 
     if activate_disks:
       # Activate the instance disks if we'exporting a stopped instance
 
     if activate_disks:
       # Activate the instance disks if we'exporting a stopped instance
@@ -11025,11 +12636,13 @@ class LUBackupExport(LogicalUnit):
 
       helper.CreateSnapshots()
       try:
 
       helper.CreateSnapshots()
       try:
-        if (self.op.shutdown and instance.admin_up and
+        if (self.op.shutdown and
+            instance.admin_state == constants.ADMINST_UP and
             not self.op.remove_instance):
           assert not activate_disks
           feedback_fn("Starting instance %s" % instance.name)
             not self.op.remove_instance):
           assert not activate_disks
           feedback_fn("Starting instance %s" % instance.name)
-          result = self.rpc.call_instance_start(src_node, instance, None, None)
+          result = self.rpc.call_instance_start(src_node,
+                                                (instance, None, None), False)
           msg = result.fail_msg
           if msg:
             feedback_fn("Failed to start instance: %s" % msg)
           msg = result.fail_msg
           if msg:
             feedback_fn("Failed to start instance: %s" % msg)
@@ -11115,7 +12728,7 @@ class LUBackupRemove(NoHooksLU):
       fqdn_warn = True
       instance_name = self.op.instance_name
 
       fqdn_warn = True
       instance_name = self.op.instance_name
 
-    locked_nodes = self.glm.list_owned(locking.LEVEL_NODE)
+    locked_nodes = self.owned_locks(locking.LEVEL_NODE)
     exportlist = self.rpc.call_export_list(locked_nodes)
     found = False
     for node in exportlist:
     exportlist = self.rpc.call_export_list(locked_nodes)
     found = False
     for node in exportlist:
@@ -11173,6 +12786,19 @@ class LUGroupAdd(LogicalUnit):
     if self.op.ndparams:
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
 
     if self.op.ndparams:
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
 
+    if self.op.diskparams:
+      for templ in constants.DISK_TEMPLATES:
+        if templ not in self.op.diskparams:
+          self.op.diskparams[templ] = {}
+        utils.ForceDictType(self.op.diskparams[templ], constants.DISK_DT_TYPES)
+    else:
+      self.op.diskparams = self.cfg.GetClusterInfo().diskparams
+
+    if self.op.ipolicy:
+      cluster = self.cfg.GetClusterInfo()
+      full_ipolicy = cluster.SimpleFillIPolicy(self.op.ipolicy)
+      objects.InstancePolicy.CheckParameterSyntax(full_ipolicy)
+
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
@@ -11195,7 +12821,9 @@ class LUGroupAdd(LogicalUnit):
     group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
                                   uuid=self.group_uuid,
                                   alloc_policy=self.op.alloc_policy,
     group_obj = objects.NodeGroup(name=self.op.group_name, members=[],
                                   uuid=self.group_uuid,
                                   alloc_policy=self.op.alloc_policy,
-                                  ndparams=self.op.ndparams)
+                                  ndparams=self.op.ndparams,
+                                  diskparams=self.op.diskparams,
+                                  ipolicy=self.op.ipolicy)
 
     self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
     del self.remove_locks[locking.LEVEL_NODEGROUP]
 
     self.cfg.AddNodeGroup(group_obj, self.proc.GetECId(), check_uuid=False)
     del self.remove_locks[locking.LEVEL_NODEGROUP]
@@ -11235,12 +12863,12 @@ class LUGroupAssignNodes(NoHooksLU):
 
     """
     assert self.needed_locks[locking.LEVEL_NODEGROUP]
 
     """
     assert self.needed_locks[locking.LEVEL_NODEGROUP]
-    assert (frozenset(self.glm.list_owned(locking.LEVEL_NODE)) ==
+    assert (frozenset(self.owned_locks(locking.LEVEL_NODE)) ==
             frozenset(self.op.nodes))
 
     expected_locks = (set([self.group_uuid]) |
                       self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
             frozenset(self.op.nodes))
 
     expected_locks = (set([self.group_uuid]) |
                       self.cfg.GetNodeGroupsFromNodes(self.op.nodes))
-    actual_locks = self.glm.list_owned(locking.LEVEL_NODEGROUP)
+    actual_locks = self.owned_locks(locking.LEVEL_NODEGROUP)
     if actual_locks != expected_locks:
       raise errors.OpExecError("Nodes changed groups since locks were acquired,"
                                " current groups are '%s', used to be '%s'" %
     if actual_locks != expected_locks:
       raise errors.OpExecError("Nodes changed groups since locks were acquired,"
                                " current groups are '%s', used to be '%s'" %
@@ -11280,13 +12908,9 @@ class LUGroupAssignNodes(NoHooksLU):
     """Assign nodes to a new group.
 
     """
     """Assign nodes to a new group.
 
     """
-    for node in self.op.nodes:
-      self.node_data[node].group = self.group_uuid
-
-    # FIXME: Depends on side-effects of modifying the result of
-    # C{cfg.GetAllNodesInfo}
+    mods = [(node_name, self.group_uuid) for node_name in self.op.nodes]
 
 
-    self.cfg.Update(self.group, feedback_fn) # Saves all modified nodes.
+    self.cfg.AssignGroupNodes(mods)
 
   @staticmethod
   def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
 
   @staticmethod
   def CheckAssignmentForSplitInstances(changes, node_data, instance_data):
@@ -11345,6 +12969,7 @@ class _GroupQuery(_QueryBase):
     lu.needed_locks = {}
 
     self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
     lu.needed_locks = {}
 
     self._all_groups = lu.cfg.GetAllNodeGroupsInfo()
+    self._cluster = lu.cfg.GetClusterInfo()
     name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
 
     if not self.names:
     name_to_uuid = dict((g.name, g.uuid) for g in self._all_groups.values())
 
     if not self.names:
@@ -11410,7 +13035,8 @@ class _GroupQuery(_QueryBase):
           # Do not pass on node information if it was not requested.
           group_to_nodes = None
 
           # Do not pass on node information if it was not requested.
           group_to_nodes = None
 
-    return query.GroupQueryData([self._all_groups[uuid]
+    return query.GroupQueryData(self._cluster,
+                                [self._all_groups[uuid]
                                  for uuid in self.wanted],
                                 group_to_nodes, group_to_instances)
 
                                  for uuid in self.wanted],
                                 group_to_nodes, group_to_instances)
 
@@ -11428,6 +13054,9 @@ class LUGroupQuery(NoHooksLU):
   def ExpandNames(self):
     self.gq.ExpandNames(self)
 
   def ExpandNames(self):
     self.gq.ExpandNames(self)
 
+  def DeclareLocks(self, level):
+    self.gq.DeclareLocks(self, level)
+
   def Exec(self, feedback_fn):
     return self.gq.OldStyleQuery(self)
 
   def Exec(self, feedback_fn):
     return self.gq.OldStyleQuery(self)
 
@@ -11443,7 +13072,11 @@ class LUGroupSetParams(LogicalUnit):
   def CheckArguments(self):
     all_changes = [
       self.op.ndparams,
   def CheckArguments(self):
     all_changes = [
       self.op.ndparams,
+      self.op.diskparams,
       self.op.alloc_policy,
       self.op.alloc_policy,
+      self.op.hv_state,
+      self.op.disk_state,
+      self.op.ipolicy,
       ]
 
     if all_changes.count(None) == len(all_changes):
       ]
 
     if all_changes.count(None) == len(all_changes):
@@ -11473,6 +13106,35 @@ class LUGroupSetParams(LogicalUnit):
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = new_ndparams
 
       utils.ForceDictType(self.op.ndparams, constants.NDS_PARAMETER_TYPES)
       self.new_ndparams = new_ndparams
 
+    if self.op.diskparams:
+      self.new_diskparams = dict()
+      for templ in constants.DISK_TEMPLATES:
+        if templ not in self.op.diskparams:
+          self.op.diskparams[templ] = {}
+        new_templ_params = _GetUpdatedParams(self.group.diskparams[templ],
+                                             self.op.diskparams[templ])
+        utils.ForceDictType(new_templ_params, constants.DISK_DT_TYPES)
+        self.new_diskparams[templ] = new_templ_params
+
+    if self.op.hv_state:
+      self.new_hv_state = _MergeAndVerifyHvState(self.op.hv_state,
+                                                 self.group.hv_state_static)
+
+    if self.op.disk_state:
+      self.new_disk_state = \
+        _MergeAndVerifyDiskState(self.op.disk_state,
+                                 self.group.disk_state_static)
+
+    if self.op.ipolicy:
+      g_ipolicy = {}
+      for key, value in self.op.ipolicy.iteritems():
+        g_ipolicy[key] = _GetUpdatedParams(self.group.ipolicy.get(key, {}),
+                                           value,
+                                           use_none=True)
+        utils.ForceDictType(g_ipolicy[key], constants.ISPECS_PARAMETER_TYPES)
+      self.new_ipolicy = g_ipolicy
+      objects.InstancePolicy.CheckParameterSyntax(self.new_ipolicy)
+
   def BuildHooksEnv(self):
     """Build hooks env.
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
@@ -11499,14 +13161,26 @@ class LUGroupSetParams(LogicalUnit):
       self.group.ndparams = self.new_ndparams
       result.append(("ndparams", str(self.group.ndparams)))
 
       self.group.ndparams = self.new_ndparams
       result.append(("ndparams", str(self.group.ndparams)))
 
+    if self.op.diskparams:
+      self.group.diskparams = self.new_diskparams
+      result.append(("diskparams", str(self.group.diskparams)))
+
     if self.op.alloc_policy:
       self.group.alloc_policy = self.op.alloc_policy
 
     if self.op.alloc_policy:
       self.group.alloc_policy = self.op.alloc_policy
 
+    if self.op.hv_state:
+      self.group.hv_state_static = self.new_hv_state
+
+    if self.op.disk_state:
+      self.group.disk_state_static = self.new_disk_state
+
+    if self.op.ipolicy:
+      self.group.ipolicy = self.new_ipolicy
+
     self.cfg.Update(self.group, feedback_fn)
     return result
 
 
     self.cfg.Update(self.group, feedback_fn)
     return result
 
 
-
 class LUGroupRemove(LogicalUnit):
   HPATH = "group-remove"
   HTYPE = constants.HTYPE_GROUP
 class LUGroupRemove(LogicalUnit):
   HPATH = "group-remove"
   HTYPE = constants.HTYPE_GROUP
@@ -11643,7 +13317,163 @@ class LUGroupRename(LogicalUnit):
     return self.op.new_name
 
 
     return self.op.new_name
 
 
-class TagsLU(NoHooksLU): # pylint: disable-msg=W0223
+class LUGroupEvacuate(LogicalUnit):
+  HPATH = "group-evacuate"
+  HTYPE = constants.HTYPE_GROUP
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    # This raises errors.OpPrereqError on its own:
+    self.group_uuid = self.cfg.LookupNodeGroup(self.op.group_name)
+
+    if self.op.target_groups:
+      self.req_target_uuids = map(self.cfg.LookupNodeGroup,
+                                  self.op.target_groups)
+    else:
+      self.req_target_uuids = []
+
+    if self.group_uuid in self.req_target_uuids:
+      raise errors.OpPrereqError("Group to be evacuated (%s) can not be used"
+                                 " as a target group (targets are %s)" %
+                                 (self.group_uuid,
+                                  utils.CommaJoin(self.req_target_uuids)),
+                                 errors.ECODE_INVAL)
+
+    self.op.iallocator = _GetDefaultIAllocator(self.cfg, self.op.iallocator)
+
+    self.share_locks = _ShareAll()
+    self.needed_locks = {
+      locking.LEVEL_INSTANCE: [],
+      locking.LEVEL_NODEGROUP: [],
+      locking.LEVEL_NODE: [],
+      }
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_INSTANCE:
+      assert not self.needed_locks[locking.LEVEL_INSTANCE]
+
+      # Lock instances optimistically, needs verification once node and group
+      # locks have been acquired
+      self.needed_locks[locking.LEVEL_INSTANCE] = \
+        self.cfg.GetNodeGroupInstances(self.group_uuid)
+
+    elif level == locking.LEVEL_NODEGROUP:
+      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+
+      if self.req_target_uuids:
+        lock_groups = set([self.group_uuid] + self.req_target_uuids)
+
+        # Lock all groups used by instances optimistically; this requires going
+        # via the node before it's locked, requiring verification later on
+        lock_groups.update(group_uuid
+                           for instance_name in
+                             self.owned_locks(locking.LEVEL_INSTANCE)
+                           for group_uuid in
+                             self.cfg.GetInstanceNodeGroups(instance_name))
+      else:
+        # No target groups, need to lock all of them
+        lock_groups = locking.ALL_SET
+
+      self.needed_locks[locking.LEVEL_NODEGROUP] = lock_groups
+
+    elif level == locking.LEVEL_NODE:
+      # This will only lock the nodes in the group to be evacuated which
+      # contain actual instances
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+      self._LockInstancesNodes()
+
+      # Lock all nodes in group to be evacuated and target groups
+      owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+      assert self.group_uuid in owned_groups
+      member_nodes = [node_name
+                      for group in owned_groups
+                      for node_name in self.cfg.GetNodeGroup(group).members]
+      self.needed_locks[locking.LEVEL_NODE].extend(member_nodes)
+
+  def CheckPrereq(self):
+    owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+    owned_nodes = frozenset(self.owned_locks(locking.LEVEL_NODE))
+
+    assert owned_groups.issuperset(self.req_target_uuids)
+    assert self.group_uuid in owned_groups
+
+    # Check if locked instances are still correct
+    _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+    # Get instance information
+    self.instances = dict(self.cfg.GetMultiInstanceInfo(owned_instances))
+
+    # Check if node groups for locked instances are still correct
+    for instance_name in owned_instances:
+      inst = self.instances[instance_name]
+      assert owned_nodes.issuperset(inst.all_nodes), \
+        "Instance %s's nodes changed while we kept the lock" % instance_name
+
+      inst_groups = _CheckInstanceNodeGroups(self.cfg, instance_name,
+                                             owned_groups)
+
+      assert self.group_uuid in inst_groups, \
+        "Instance %s has no node in group %s" % (instance_name, self.group_uuid)
+
+    if self.req_target_uuids:
+      # User requested specific target groups
+      self.target_uuids = self.req_target_uuids
+    else:
+      # All groups except the one to be evacuated are potential targets
+      self.target_uuids = [group_uuid for group_uuid in owned_groups
+                           if group_uuid != self.group_uuid]
+
+      if not self.target_uuids:
+        raise errors.OpPrereqError("There are no possible target groups",
+                                   errors.ECODE_INVAL)
+
+  def BuildHooksEnv(self):
+    """Build hooks env.
+
+    """
+    return {
+      "GROUP_NAME": self.op.group_name,
+      "TARGET_GROUPS": " ".join(self.target_uuids),
+      }
+
+  def BuildHooksNodes(self):
+    """Build hooks nodes.
+
+    """
+    mn = self.cfg.GetMasterNode()
+
+    assert self.group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
+
+    run_nodes = [mn] + self.cfg.GetNodeGroup(self.group_uuid).members
+
+    return (run_nodes, run_nodes)
+
+  def Exec(self, feedback_fn):
+    instances = list(self.owned_locks(locking.LEVEL_INSTANCE))
+
+    assert self.group_uuid not in self.target_uuids
+
+    ial = IAllocator(self.cfg, self.rpc, constants.IALLOCATOR_MODE_CHG_GROUP,
+                     instances=instances, target_groups=self.target_uuids)
+
+    ial.Run(self.op.iallocator)
+
+    if not ial.success:
+      raise errors.OpPrereqError("Can't compute group evacuation using"
+                                 " iallocator '%s': %s" %
+                                 (self.op.iallocator, ial.info),
+                                 errors.ECODE_NORES)
+
+    jobs = _LoadNodeEvacResult(self, ial.result, self.op.early_release, False)
+
+    self.LogInfo("Iallocator returned %s job(s) for evacuating node group %s",
+                 len(jobs), self.op.group_name)
+
+    return ResultWithJobs(jobs)
+
+
+class TagsLU(NoHooksLU): # pylint: disable=W0223
   """Generic tags LU.
 
   This is an abstract class which is the parent of all the other tags LUs.
   """Generic tags LU.
 
   This is an abstract class which is the parent of all the other tags LUs.
@@ -11691,7 +13521,7 @@ class LUTagsGet(TagsLU):
     TagsLU.ExpandNames(self)
 
     # Share locks as this is only a read operation
     TagsLU.ExpandNames(self)
 
     # Share locks as this is only a read operation
-    self.share_locks = dict.fromkeys(locking.LEVELS, 1)
+    self.share_locks = _ShareAll()
 
   def Exec(self, feedback_fn):
     """Returns the tag list.
 
   def Exec(self, feedback_fn):
     """Returns the tag list.
@@ -11903,7 +13733,7 @@ class LUTestJqueue(NoHooksLU):
     # Wait for client to close
     try:
       try:
     # Wait for client to close
     try:
       try:
-        # pylint: disable-msg=E1101
+        # pylint: disable=E1101
         # Instance of '_socketobject' has no ... member
         conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
         conn.recv(1)
         # Instance of '_socketobject' has no ... member
         conn.settimeout(cls._CLIENT_CONFIRM_TIMEOUT)
         conn.recv(1)
@@ -12000,12 +13830,12 @@ class IAllocator(object):
       easy usage
 
   """
       easy usage
 
   """
-  # pylint: disable-msg=R0902
+  # pylint: disable=R0902
   # lots of instance attributes
 
   # lots of instance attributes
 
-  def __init__(self, cfg, rpc, mode, **kwargs):
+  def __init__(self, cfg, rpc_runner, mode, **kwargs):
     self.cfg = cfg
     self.cfg = cfg
-    self.rpc = rpc
+    self.rpc = rpc_runner
     # init buffer variables
     self.in_text = self.out_text = self.in_data = self.out_data = None
     # init all input fields so that pylint is happy
     # init buffer variables
     self.in_text = self.out_text = self.in_data = self.out_data = None
     # init all input fields so that pylint is happy
@@ -12015,9 +13845,8 @@ class IAllocator(object):
     self.hypervisor = None
     self.relocate_from = None
     self.name = None
     self.hypervisor = None
     self.relocate_from = None
     self.name = None
-    self.evac_nodes = None
     self.instances = None
     self.instances = None
-    self.reloc_mode = None
+    self.evac_mode = None
     self.target_groups = []
     # computed fields
     self.required_nodes = None
     self.target_groups = []
     # computed fields
     self.required_nodes = None
@@ -12071,12 +13900,11 @@ class IAllocator(object):
       hypervisor_name = self.hypervisor
     elif self.mode == constants.IALLOCATOR_MODE_RELOC:
       hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
       hypervisor_name = self.hypervisor
     elif self.mode == constants.IALLOCATOR_MODE_RELOC:
       hypervisor_name = cfg.GetInstanceInfo(self.name).hypervisor
-    elif self.mode in (constants.IALLOCATOR_MODE_MEVAC,
-                       constants.IALLOCATOR_MODE_MRELOC):
-      hypervisor_name = cluster_info.enabled_hypervisors[0]
+    else:
+      hypervisor_name = cluster_info.primary_hypervisor
 
 
-    node_data = self.rpc.call_node_info(node_list, cfg.GetVGName(),
-                                        hypervisor_name)
+    node_data = self.rpc.call_node_info(node_list, [cfg.GetVGName()],
+                                        [hypervisor_name])
     node_iinfo = \
       self.rpc.call_all_instances_info(node_list,
                                        cluster_info.enabled_hypervisors)
     node_iinfo = \
       self.rpc.call_all_instances_info(node_list,
                                        cluster_info.enabled_hypervisors)
@@ -12138,6 +13966,7 @@ class IAllocator(object):
     @param node_results: the basic node structures as filled from the config
 
     """
     @param node_results: the basic node structures as filled from the config
 
     """
+    #TODO(dynmem): compute the right data on MAX and MIN memory
     # make a copy of the current dict
     node_results = dict(node_results)
     for nname, nresult in node_data.items():
     # make a copy of the current dict
     node_results = dict(node_results)
     for nname, nresult in node_data.items():
@@ -12148,10 +13977,10 @@ class IAllocator(object):
         nresult.Raise("Can't get data for node %s" % nname)
         node_iinfo[nname].Raise("Can't get node instance info from node %s" %
                                 nname)
         nresult.Raise("Can't get data for node %s" % nname)
         node_iinfo[nname].Raise("Can't get node instance info from node %s" %
                                 nname)
-        remote_info = nresult.payload
+        remote_info = _MakeLegacyNodeInfo(nresult.payload)
 
 
-        for attr in ['memory_total', 'memory_free', 'memory_dom0',
-                     'vg_size', 'vg_free', 'cpu_total']:
+        for attr in ["memory_total", "memory_free", "memory_dom0",
+                     "vg_size", "vg_free", "cpu_total"]:
           if attr not in remote_info:
             raise errors.OpExecError("Node '%s' didn't return attribute"
                                      " '%s'" % (nname, attr))
           if attr not in remote_info:
             raise errors.OpExecError("Node '%s' didn't return attribute"
                                      " '%s'" % (nname, attr))
@@ -12163,25 +13992,25 @@ class IAllocator(object):
         i_p_mem = i_p_up_mem = 0
         for iinfo, beinfo in i_list:
           if iinfo.primary_node == nname:
         i_p_mem = i_p_up_mem = 0
         for iinfo, beinfo in i_list:
           if iinfo.primary_node == nname:
-            i_p_mem += beinfo[constants.BE_MEMORY]
+            i_p_mem += beinfo[constants.BE_MAXMEM]
             if iinfo.name not in node_iinfo[nname].payload:
               i_used_mem = 0
             else:
             if iinfo.name not in node_iinfo[nname].payload:
               i_used_mem = 0
             else:
-              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]['memory'])
-            i_mem_diff = beinfo[constants.BE_MEMORY] - i_used_mem
-            remote_info['memory_free'] -= max(0, i_mem_diff)
+              i_used_mem = int(node_iinfo[nname].payload[iinfo.name]["memory"])
+            i_mem_diff = beinfo[constants.BE_MAXMEM] - i_used_mem
+            remote_info["memory_free"] -= max(0, i_mem_diff)
 
 
-            if iinfo.admin_up:
-              i_p_up_mem += beinfo[constants.BE_MEMORY]
+            if iinfo.admin_state == constants.ADMINST_UP:
+              i_p_up_mem += beinfo[constants.BE_MAXMEM]
 
         # compute memory used by instances
         pnr_dyn = {
 
         # compute memory used by instances
         pnr_dyn = {
-          "total_memory": remote_info['memory_total'],
-          "reserved_memory": remote_info['memory_dom0'],
-          "free_memory": remote_info['memory_free'],
-          "total_disk": remote_info['vg_size'],
-          "free_disk": remote_info['vg_free'],
-          "total_cpus": remote_info['cpu_total'],
+          "total_memory": remote_info["memory_total"],
+          "reserved_memory": remote_info["memory_dom0"],
+          "free_memory": remote_info["memory_free"],
+          "total_disk": remote_info["vg_size"],
+          "free_disk": remote_info["vg_free"],
+          "total_cpus": remote_info["cpu_total"],
           "i_pri_memory": i_p_mem,
           "i_pri_up_memory": i_p_up_mem,
           }
           "i_pri_memory": i_p_mem,
           "i_pri_up_memory": i_p_up_mem,
           }
@@ -12211,9 +14040,9 @@ class IAllocator(object):
         nic_data.append(nic_dict)
       pir = {
         "tags": list(iinfo.GetTags()),
         nic_data.append(nic_dict)
       pir = {
         "tags": list(iinfo.GetTags()),
-        "admin_up": iinfo.admin_up,
+        "admin_state": iinfo.admin_state,
         "vcpus": beinfo[constants.BE_VCPUS],
         "vcpus": beinfo[constants.BE_VCPUS],
-        "memory": beinfo[constants.BE_MEMORY],
+        "memory": beinfo[constants.BE_MAXMEM],
         "os": iinfo.os,
         "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
         "nics": nic_data,
         "os": iinfo.os,
         "nodes": [iinfo.primary_node] + list(iinfo.secondary_nodes),
         "nics": nic_data,
@@ -12298,22 +14127,21 @@ class IAllocator(object):
       }
     return request
 
       }
     return request
 
-  def _AddEvacuateNodes(self):
-    """Add evacuate nodes data to allocator structure.
+  def _AddNodeEvacuate(self):
+    """Get data for node-evacuate requests.
 
     """
 
     """
-    request = {
-      "evac_nodes": self.evac_nodes
+    return {
+      "instances": self.instances,
+      "evac_mode": self.evac_mode,
       }
       }
-    return request
 
 
-  def _AddMultiRelocate(self):
-    """Get data for multi-relocate requests.
+  def _AddChangeGroup(self):
+    """Get data for node-evacuate requests.
 
     """
     return {
       "instances": self.instances,
 
     """
     return {
       "instances": self.instances,
-      "reloc_mode": self.reloc_mode,
       "target_groups": self.target_groups,
       }
 
       "target_groups": self.target_groups,
       }
 
@@ -12339,6 +14167,28 @@ class IAllocator(object):
     self.in_text = serializer.Dump(self.in_data)
 
   _STRING_LIST = ht.TListOf(ht.TString)
     self.in_text = serializer.Dump(self.in_data)
 
   _STRING_LIST = ht.TListOf(ht.TString)
+  _JOB_LIST = ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
+     # pylint: disable=E1101
+     # Class '...' has no 'OP_ID' member
+     "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
+                          opcodes.OpInstanceMigrate.OP_ID,
+                          opcodes.OpInstanceReplaceDisks.OP_ID])
+     })))
+
+  _NEVAC_MOVED = \
+    ht.TListOf(ht.TAnd(ht.TIsLength(3),
+                       ht.TItems([ht.TNonEmptyString,
+                                  ht.TNonEmptyString,
+                                  ht.TListOf(ht.TNonEmptyString),
+                                 ])))
+  _NEVAC_FAILED = \
+    ht.TListOf(ht.TAnd(ht.TIsLength(2),
+                       ht.TItems([ht.TNonEmptyString,
+                                  ht.TMaybeString,
+                                 ])))
+  _NEVAC_RESULT = ht.TAnd(ht.TIsLength(3),
+                          ht.TItems([_NEVAC_MOVED, _NEVAC_FAILED, _JOB_LIST]))
+
   _MODE_DATA = {
     constants.IALLOCATOR_MODE_ALLOC:
       (_AddNewInstance,
   _MODE_DATA = {
     constants.IALLOCATOR_MODE_ALLOC:
       (_AddNewInstance,
@@ -12357,22 +14207,16 @@ class IAllocator(object):
       (_AddRelocateInstance,
        [("name", ht.TString), ("relocate_from", _STRING_LIST)],
        ht.TList),
       (_AddRelocateInstance,
        [("name", ht.TString), ("relocate_from", _STRING_LIST)],
        ht.TList),
-    constants.IALLOCATOR_MODE_MEVAC:
-      (_AddEvacuateNodes, [("evac_nodes", _STRING_LIST)],
-       ht.TListOf(ht.TAnd(ht.TIsLength(2), _STRING_LIST))),
-    constants.IALLOCATOR_MODE_MRELOC:
-      (_AddMultiRelocate, [
+     constants.IALLOCATOR_MODE_NODE_EVAC:
+      (_AddNodeEvacuate, [
+        ("instances", _STRING_LIST),
+        ("evac_mode", ht.TElemOf(constants.IALLOCATOR_NEVAC_MODES)),
+        ], _NEVAC_RESULT),
+     constants.IALLOCATOR_MODE_CHG_GROUP:
+      (_AddChangeGroup, [
         ("instances", _STRING_LIST),
         ("instances", _STRING_LIST),
-        ("reloc_mode", ht.TElemOf(constants.IALLOCATOR_MRELOC_MODES)),
         ("target_groups", _STRING_LIST),
         ("target_groups", _STRING_LIST),
-        ],
-       ht.TListOf(ht.TListOf(ht.TStrictDict(True, False, {
-         # pylint: disable-msg=E1101
-         # Class '...' has no 'OP_ID' member
-         "OP_ID": ht.TElemOf([opcodes.OpInstanceFailover.OP_ID,
-                              opcodes.OpInstanceMigrate.OP_ID,
-                              opcodes.OpInstanceReplaceDisks.OP_ID])
-         })))),
+        ], _NEVAC_RESULT),
     }
 
   def Run(self, name, validate=True, call_fn=None):
     }
 
   def Run(self, name, validate=True, call_fn=None):
@@ -12421,39 +14265,28 @@ class IAllocator(object):
                                (self._result_check, self.result),
                                errors.ECODE_INVAL)
 
                                (self._result_check, self.result),
                                errors.ECODE_INVAL)
 
-    if self.mode in (constants.IALLOCATOR_MODE_RELOC,
-                     constants.IALLOCATOR_MODE_MEVAC):
+    if self.mode == constants.IALLOCATOR_MODE_RELOC:
+      assert self.relocate_from is not None
+      assert self.required_nodes == 1
+
       node2group = dict((name, ndata["group"])
                         for (name, ndata) in self.in_data["nodes"].items())
 
       fn = compat.partial(self._NodesToGroups, node2group,
                           self.in_data["nodegroups"])
 
       node2group = dict((name, ndata["group"])
                         for (name, ndata) in self.in_data["nodes"].items())
 
       fn = compat.partial(self._NodesToGroups, node2group,
                           self.in_data["nodegroups"])
 
-      if self.mode == constants.IALLOCATOR_MODE_RELOC:
-        assert self.relocate_from is not None
-        assert self.required_nodes == 1
-
-        request_groups = fn(self.relocate_from)
-        result_groups = fn(rdict["result"])
-
-        if result_groups != request_groups:
-          raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
-                                   " differ from original groups (%s)" %
-                                   (utils.CommaJoin(result_groups),
-                                    utils.CommaJoin(request_groups)))
-      elif self.mode == constants.IALLOCATOR_MODE_MEVAC:
-        request_groups = fn(self.evac_nodes)
-        for (instance_name, secnode) in self.result:
-          result_groups = fn([secnode])
-          if result_groups != request_groups:
-            raise errors.OpExecError("Iallocator returned new secondary node"
-                                     " '%s' (group '%s') for instance '%s'"
-                                     " which is not in original group '%s'" %
-                                     (secnode, utils.CommaJoin(result_groups),
-                                      instance_name,
-                                      utils.CommaJoin(request_groups)))
-      else:
-        raise errors.ProgrammerError("Unhandled mode '%s'" % self.mode)
+      instance = self.cfg.GetInstanceInfo(self.name)
+      request_groups = fn(self.relocate_from + [instance.primary_node])
+      result_groups = fn(rdict["result"] + [instance.primary_node])
+
+      if self.success and not set(result_groups).issubset(request_groups):
+        raise errors.OpExecError("Groups of nodes returned by iallocator (%s)"
+                                 " differ from original groups (%s)" %
+                                 (utils.CommaJoin(result_groups),
+                                  utils.CommaJoin(request_groups)))
+
+    elif self.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
+      assert self.evac_mode in constants.IALLOCATOR_NEVAC_MODES
 
     self.out_data = rdict
 
 
     self.out_data = rdict
 
@@ -12532,17 +14365,13 @@ class LUTestAllocator(NoHooksLU):
     elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
       fname = _ExpandInstanceName(self.cfg, self.op.name)
       self.op.name = fname
     elif self.op.mode == constants.IALLOCATOR_MODE_RELOC:
       fname = _ExpandInstanceName(self.cfg, self.op.name)
       self.op.name = fname
-      self.relocate_from = self.cfg.GetInstanceInfo(fname).secondary_nodes
-    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
-      if not hasattr(self.op, "evac_nodes"):
-        raise errors.OpPrereqError("Missing attribute 'evac_nodes' on"
-                                   " opcode input", errors.ECODE_INVAL)
-    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
-      if self.op.instances:
-        self.op.instances = _GetWantedInstances(self, self.op.instances)
-      else:
-        raise errors.OpPrereqError("Missing instances to relocate",
-                                   errors.ECODE_INVAL)
+      self.relocate_from = \
+          list(self.cfg.GetInstanceInfo(fname).secondary_nodes)
+    elif self.op.mode in (constants.IALLOCATOR_MODE_CHG_GROUP,
+                          constants.IALLOCATOR_MODE_NODE_EVAC):
+      if not self.op.instances:
+        raise errors.OpPrereqError("Missing instances", errors.ECODE_INVAL)
+      self.op.instances = _GetWantedInstances(self, self.op.instances)
     else:
       raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
                                  self.op.mode, errors.ECODE_INVAL)
     else:
       raise errors.OpPrereqError("Invalid test allocator mode '%s'" %
                                  self.op.mode, errors.ECODE_INVAL)
@@ -12578,16 +14407,16 @@ class LUTestAllocator(NoHooksLU):
                        name=self.op.name,
                        relocate_from=list(self.relocate_from),
                        )
                        name=self.op.name,
                        relocate_from=list(self.relocate_from),
                        )
-    elif self.op.mode == constants.IALLOCATOR_MODE_MEVAC:
+    elif self.op.mode == constants.IALLOCATOR_MODE_CHG_GROUP:
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,
-                       evac_nodes=self.op.evac_nodes)
-    elif self.op.mode == constants.IALLOCATOR_MODE_MRELOC:
+                       instances=self.op.instances,
+                       target_groups=self.op.target_groups)
+    elif self.op.mode == constants.IALLOCATOR_MODE_NODE_EVAC:
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,
                        instances=self.op.instances,
       ial = IAllocator(self.cfg, self.rpc,
                        mode=self.op.mode,
                        instances=self.op.instances,
-                       reloc_mode=self.op.reloc_mode,
-                       target_groups=self.op.target_groups)
+                       evac_mode=self.op.evac_mode)
     else:
       raise errors.ProgrammerError("Uncatched mode %s in"
                                    " LUTestAllocator.Exec", self.op.mode)
     else:
       raise errors.ProgrammerError("Uncatched mode %s in"
                                    " LUTestAllocator.Exec", self.op.mode)