Add error code for temporary lack of resources
[ganeti-local] / lib / cmdlib.py
index 690053a..d29631a 100644 (file)
@@ -40,7 +40,6 @@ import tempfile
 import shutil
 import itertools
 import operator
-import ipaddr
 
 from ganeti import ssh
 from ganeti import utils
@@ -139,13 +138,18 @@ class LogicalUnit(object):
     self.owned_locks = context.glm.list_owned
     self.context = context
     self.rpc = rpc_runner
-    # Dicts used to declare locking needs to mcpu
+
+    # Dictionaries used to declare locking needs to mcpu
     self.needed_locks = None
     self.share_locks = dict.fromkeys(locking.LEVELS, 0)
+    self.opportunistic_locks = dict.fromkeys(locking.LEVELS, False)
+
     self.add_locks = {}
     self.remove_locks = {}
+
     # Used to force good behavior when calling helper functions
     self.recalculate_locks = {}
+
     # logging
     self.Log = processor.Log # pylint: disable=C0103
     self.LogWarning = processor.LogWarning # pylint: disable=C0103
@@ -960,7 +964,8 @@ def _RunPostHook(lu, node_name):
   try:
     hm.RunPhase(constants.HOOKS_PHASE_POST, nodes=[node_name])
   except Exception, err: # pylint: disable=W0703
-    lu.LogWarning("Errors occurred running hooks on %s: %s" % (node_name, err))
+    lu.LogWarning("Errors occurred running hooks on %s: %s",
+                  node_name, err)
 
 
 def _CheckOutputFields(static, dynamic, selected):
@@ -1101,7 +1106,8 @@ def _CheckInstanceState(lu, instance, req_states, msg=None):
 
   """
   if msg is None:
-    msg = "can't use instance from outside %s states" % ", ".join(req_states)
+    msg = ("can't use instance from outside %s states" %
+           utils.CommaJoin(req_states))
   if instance.admin_state not in req_states:
     raise errors.OpPrereqError("Instance '%s' is marked to be %s, %s" %
                                (instance.name, instance.admin_state, msg),
@@ -1316,6 +1322,72 @@ def _ExpandInstanceName(cfg, name):
   return _ExpandItemName(cfg.ExpandInstanceName, name, "Instance")
 
 
+def _BuildNetworkHookEnv(name, subnet, gateway, network6, gateway6,
+                         network_type, mac_prefix, tags):
+  """Builds network related env variables for hooks
+
+  This builds the hook environment from individual variables.
+
+  @type name: string
+  @param name: the name of the network
+  @type subnet: string
+  @param subnet: the ipv4 subnet
+  @type gateway: string
+  @param gateway: the ipv4 gateway
+  @type network6: string
+  @param network6: the ipv6 subnet
+  @type gateway6: string
+  @param gateway6: the ipv6 gateway
+  @type network_type: string
+  @param network_type: the type of the network
+  @type mac_prefix: string
+  @param mac_prefix: the mac_prefix
+  @type tags: list
+  @param tags: the tags of the network
+
+  """
+  env = {}
+  if name:
+    env["NETWORK_NAME"] = name
+  if subnet:
+    env["NETWORK_SUBNET"] = subnet
+  if gateway:
+    env["NETWORK_GATEWAY"] = gateway
+  if network6:
+    env["NETWORK_SUBNET6"] = network6
+  if gateway6:
+    env["NETWORK_GATEWAY6"] = gateway6
+  if mac_prefix:
+    env["NETWORK_MAC_PREFIX"] = mac_prefix
+  if network_type:
+    env["NETWORK_TYPE"] = network_type
+  if tags:
+    env["NETWORK_TAGS"] = " ".join(tags)
+
+  return env
+
+
+def _BuildNetworkHookEnvByObject(net):
+  """Builds network related env varliables for hooks
+
+  @type net: L{objects.Network}
+  @param net: the network object
+
+  """
+  args = {
+    "name": net.name,
+    "subnet": net.network,
+    "gateway": net.gateway,
+    "network6": net.network6,
+    "gateway6": net.gateway6,
+    "network_type": net.network_type,
+    "mac_prefix": net.mac_prefix,
+    "tags": net.tags,
+  }
+
+  return _BuildNetworkHookEnv(**args) # pylint: disable=W0142
+
+
 def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
                           minmem, maxmem, vcpus, nics, disk_template, disks,
                           bep, hvp, hypervisor_name, tags):
@@ -1375,14 +1447,31 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   }
   if nics:
     nic_count = len(nics)
-    for idx, (ip, mac, mode, link, network) in enumerate(nics):
+    for idx, (ip, mac, mode, link, net, netinfo) in enumerate(nics):
       if ip is None:
         ip = ""
       env["INSTANCE_NIC%d_IP" % idx] = ip
       env["INSTANCE_NIC%d_MAC" % idx] = mac
       env["INSTANCE_NIC%d_MODE" % idx] = mode
       env["INSTANCE_NIC%d_LINK" % idx] = link
-      env["INSTANCE_NIC%d_NETWORK" % idx] = network
+      if network:
+        env["INSTANCE_NIC%d_NETWORK" % idx] = net
+        if netinfo:
+          nobj = objects.Network.FromDict(netinfo)
+          if nobj.network:
+            env["INSTANCE_NIC%d_NETWORK_SUBNET" % idx] = nobj.network
+          if nobj.gateway:
+            env["INSTANCE_NIC%d_NETWORK_GATEWAY" % idx] = nobj.gateway
+          if nobj.network6:
+            env["INSTANCE_NIC%d_NETWORK_SUBNET6" % idx] = nobj.network6
+          if nobj.gateway6:
+            env["INSTANCE_NIC%d_NETWORK_GATEWAY6" % idx] = nobj.gateway6
+          if nobj.mac_prefix:
+            env["INSTANCE_NIC%d_NETWORK_MAC_PREFIX" % idx] = nobj.mac_prefix
+          if nobj.network_type:
+            env["INSTANCE_NIC%d_NETWORK_TYPE" % idx] = nobj.network_type
+          if nobj.tags:
+            env["INSTANCE_NIC%d_NETWORK_TAGS" % idx] = " ".join(nobj.tags)
       if mode == constants.NIC_MODE_BRIDGED:
         env["INSTANCE_NIC%d_BRIDGE" % idx] = link
   else:
@@ -1412,6 +1501,31 @@ def _BuildInstanceHookEnv(name, primary_node, secondary_nodes, os_type, status,
   return env
 
 
+def _NICToTuple(lu, nic):
+  """Build a tupple of nic information.
+
+  @type lu:  L{LogicalUnit}
+  @param lu: the logical unit on whose behalf we execute
+  @type nic: L{objects.NIC}
+  @param nic: nic to convert to hooks tuple
+
+  """
+  ip = nic.ip
+  mac = nic.mac
+  cluster = lu.cfg.GetClusterInfo()
+  filled_params = cluster.SimpleFillNIC(nic.nicparams)
+  mode = filled_params[constants.NIC_MODE]
+  link = filled_params[constants.NIC_LINK]
+  net = nic.network
+  netinfo = None
+  if net:
+    net_uuid = lu.cfg.LookupNetwork(net)
+    if net_uuid:
+      nobj = lu.cfg.GetNetwork(net_uuid)
+      netinfo = objects.Network.ToDict(nobj)
+  return (ip, mac, mode, link, net, netinfo)
+
+
 def _NICListToTuple(lu, nics):
   """Build a list of nic information tuples.
 
@@ -1425,15 +1539,8 @@ def _NICListToTuple(lu, nics):
 
   """
   hooks_nics = []
-  cluster = lu.cfg.GetClusterInfo()
   for nic in nics:
-    ip = nic.ip
-    mac = nic.mac
-    filled_params = cluster.SimpleFillNIC(nic.nicparams)
-    mode = filled_params[constants.NIC_MODE]
-    link = filled_params[constants.NIC_LINK]
-    network = nic.network
-    hooks_nics.append((ip, mac, mode, link, network))
+    hooks_nics.append(_NICToTuple(lu, nic))
   return hooks_nics
 
 
@@ -1935,7 +2042,7 @@ class LUClusterVerify(NoHooksLU):
 
       # Verify global configuration
       jobs.append([
-        opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors)
+        opcodes.OpClusterVerifyConfig(ignore_errors=self.op.ignore_errors),
         ])
 
       # Always depend on global verification
@@ -2126,6 +2233,11 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
       locking.LEVEL_INSTANCE: inst_names,
       locking.LEVEL_NODEGROUP: [self.group_uuid],
       locking.LEVEL_NODE: [],
+
+      # This opcode is run by watcher every five minutes and acquires all nodes
+      # for a group. It doesn't run for a long time, so it's better to acquire
+      # the node allocation lock as well.
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
       }
 
     self.share_locks = _ShareAll()
@@ -2450,7 +2562,8 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
     ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
                                                             self.group_info)
     err = _ComputeIPolicyInstanceViolation(ipolicy, instanceconfig)
-    _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err))
+    _ErrorIf(err, constants.CV_EINSTANCEPOLICY, instance, utils.CommaJoin(err),
+             code=self.ETYPE_WARNING)
 
     for node in node_vol_should:
       n_img = node_image[node]
@@ -3087,7 +3200,7 @@ class LUClusterVerifyGroup(LogicalUnit, _VerifyErrors):
 
     """
     env = {
-      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags())
+      "CLUSTER_TAGS": " ".join(self.cfg.GetClusterInfo().GetTags()),
       }
 
     env.update(("NODE_TAGS_%s" % node.name, " ".join(node.GetTags()))
@@ -3572,6 +3685,12 @@ class LUGroupVerifyDisks(NoHooksLU):
       locking.LEVEL_INSTANCE: [],
       locking.LEVEL_NODEGROUP: [],
       locking.LEVEL_NODE: [],
+
+      # This opcode is acquires all node locks in a group. LUClusterVerifyDisks
+      # starts one instance of this opcode for every group, which means all
+      # nodes will be locked for a short amount of time, so it's better to
+      # acquire the node allocation lock as well.
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
       }
 
   def DeclareLocks(self, level):
@@ -3678,6 +3797,8 @@ class LUClusterRepairDiskSizes(NoHooksLU):
   def ExpandNames(self):
     if self.op.instances:
       self.wanted_names = _GetWantedInstances(self, self.op.instances)
+      # Not getting the node allocation lock as only a specific set of
+      # instances (and their nodes) is going to be acquired
       self.needed_locks = {
         locking.LEVEL_NODE_RES: [],
         locking.LEVEL_INSTANCE: self.wanted_names,
@@ -3688,10 +3809,15 @@ class LUClusterRepairDiskSizes(NoHooksLU):
       self.needed_locks = {
         locking.LEVEL_NODE_RES: locking.ALL_SET,
         locking.LEVEL_INSTANCE: locking.ALL_SET,
+
+        # This opcode is acquires the node locks for all instances
+        locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
         }
+
     self.share_locks = {
       locking.LEVEL_NODE_RES: 1,
       locking.LEVEL_INSTANCE: 0,
+      locking.LEVEL_NODE_ALLOC: 1,
       }
 
   def DeclareLocks(self, level):
@@ -3933,16 +4059,15 @@ class LUClusterSetParams(LogicalUnit):
   def ExpandNames(self):
     # FIXME: in the future maybe other cluster params won't require checking on
     # all nodes to be modified.
+    # FIXME: This opcode changes cluster-wide settings. Is acquiring all
+    # resource locks the right thing, shouldn't it be the BGL instead?
     self.needed_locks = {
       locking.LEVEL_NODE: locking.ALL_SET,
       locking.LEVEL_INSTANCE: locking.ALL_SET,
       locking.LEVEL_NODEGROUP: locking.ALL_SET,
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
     }
-    self.share_locks = {
-        locking.LEVEL_NODE: 1,
-        locking.LEVEL_INSTANCE: 1,
-        locking.LEVEL_NODEGROUP: 1,
-    }
+    self.share_locks = _ShareAll()
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -4347,7 +4472,7 @@ def _UploadHelper(lu, nodes, fname):
       if msg:
         msg = ("Copy of file %s to node %s failed: %s" %
                (fname, to_node, msg))
-        lu.proc.LogWarning(msg)
+        lu.LogWarning(msg)
 
 
 def _ComputeAncillaryFiles(cluster, redist):
@@ -4489,8 +4614,9 @@ class LUClusterRedistConf(NoHooksLU):
   def ExpandNames(self):
     self.needed_locks = {
       locking.LEVEL_NODE: locking.ALL_SET,
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
     }
-    self.share_locks[locking.LEVEL_NODE] = 1
+    self.share_locks = _ShareAll()
 
   def Exec(self, feedback_fn):
     """Redistribute the configuration.
@@ -4540,7 +4666,7 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False):
   disks = _ExpandCheckDisks(instance, disks)
 
   if not oneshot:
-    lu.proc.LogInfo("Waiting for instance %s to sync disks." % instance.name)
+    lu.LogInfo("Waiting for instance %s to sync disks", instance.name)
 
   node = instance.primary_node
 
@@ -4583,8 +4709,8 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False):
           max_time = mstat.estimated_time
         else:
           rem_time = "no time estimate"
-        lu.proc.LogInfo("- device %s: %5.2f%% done, %s" %
-                        (disks[i].iv_name, mstat.sync_percent, rem_time))
+        lu.LogInfo("- device %s: %5.2f%% done, %s",
+                   disks[i].iv_name, mstat.sync_percent, rem_time)
 
     # if we're done but degraded, let's do a few small retries, to
     # make sure we see a stable and not transient situation; therefore
@@ -4601,7 +4727,8 @@ def _WaitForSync(lu, instance, disks=None, oneshot=False):
     time.sleep(min(60, max_time))
 
   if done:
-    lu.proc.LogInfo("Instance %s's disks are in sync." % instance.name)
+    lu.LogInfo("Instance %s's disks are in sync", instance.name)
+
   return not cumul_degraded
 
 
@@ -4687,6 +4814,11 @@ class LUOobCommand(NoHooksLU):
       locking.LEVEL_NODE: lock_names,
       }
 
+    if not self.op.node_names:
+      # Acquire node allocation lock only if all nodes are affected
+      self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+      self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
+
   def CheckPrereq(self):
     """Check prerequisites.
 
@@ -5119,6 +5251,7 @@ class _NodeQuery(_QueryBase):
     if self.do_locking:
       # If any non-static field is requested we need to lock the nodes
       lu.needed_locks[locking.LEVEL_NODE] = self.wanted
+      lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
 
   def DeclareLocks(self, lu, level):
     pass
@@ -5213,13 +5346,16 @@ class LUNodeQueryvols(NoHooksLU):
 
   def ExpandNames(self):
     self.share_locks = _ShareAll()
-    self.needed_locks = {}
 
-    if not self.op.nodes:
-      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+    if self.op.nodes:
+      self.needed_locks = {
+        locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes),
+        }
     else:
-      self.needed_locks[locking.LEVEL_NODE] = \
-        _GetWantedNodes(self, self.op.nodes)
+      self.needed_locks = {
+        locking.LEVEL_NODE: locking.ALL_SET,
+        locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+        }
 
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
@@ -5282,13 +5418,16 @@ class LUNodeQueryStorage(NoHooksLU):
 
   def ExpandNames(self):
     self.share_locks = _ShareAll()
-    self.needed_locks = {}
 
     if self.op.nodes:
-      self.needed_locks[locking.LEVEL_NODE] = \
-        _GetWantedNodes(self, self.op.nodes)
+      self.needed_locks = {
+        locking.LEVEL_NODE: _GetWantedNodes(self, self.op.nodes),
+        }
     else:
-      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+      self.needed_locks = {
+        locking.LEVEL_NODE: locking.ALL_SET,
+        locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+        }
 
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
@@ -5914,19 +6053,28 @@ class LUNodeSetParams(LogicalUnit):
 
   def ExpandNames(self):
     if self.lock_all:
-      self.needed_locks = {locking.LEVEL_NODE: locking.ALL_SET}
+      self.needed_locks = {
+        locking.LEVEL_NODE: locking.ALL_SET,
+
+        # Block allocations when all nodes are locked
+        locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+        }
     else:
-      self.needed_locks = {locking.LEVEL_NODE: self.op.node_name}
+      self.needed_locks = {
+        locking.LEVEL_NODE: self.op.node_name,
+        }
 
     # 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]
 
-    # 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
+    # Get all locks except nodes in shared mode; they are not used for anything
+    # but read-only access
+    self.share_locks = _ShareAll()
+    self.share_locks[locking.LEVEL_NODE] = 0
+    self.share_locks[locking.LEVEL_NODE_RES] = 0
+    self.share_locks[locking.LEVEL_NODE_ALLOC] = 0
 
     if self.lock_instances:
       self.needed_locks[locking.LEVEL_INSTANCE] = \
@@ -6093,7 +6241,8 @@ class LUNodeSetParams(LogicalUnit):
       if master_singlehomed and self.op.secondary_ip != node.primary_ip:
         if self.op.force and node.name == master.name:
           self.LogWarning("Transitioning from single-homed to multi-homed"
-                          " cluster. All nodes will require a secondary ip.")
+                          " cluster; all nodes will require a secondary IP"
+                          " address")
         else:
           raise errors.OpPrereqError("Changing the secondary ip on a"
                                      " single-homed cluster requires the"
@@ -6103,7 +6252,8 @@ class LUNodeSetParams(LogicalUnit):
       elif not master_singlehomed and self.op.secondary_ip == node.primary_ip:
         if self.op.force and node.name == master.name:
           self.LogWarning("Transitioning from multi-homed to single-homed"
-                          " cluster. Secondary IPs will have to be removed.")
+                          " cluster; secondary IP addresses will have to be"
+                          " removed")
         else:
           raise errors.OpPrereqError("Cannot set the secondary IP to be the"
                                      " same as the primary IP on a multi-homed"
@@ -6481,9 +6631,9 @@ def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
       if msg:
         is_offline_secondary = (node in instance.secondary_nodes and
                                 result.offline)
-        lu.proc.LogWarning("Could not prepare block device %s on node %s"
-                           " (is_primary=False, pass=1): %s",
-                           inst_disk.iv_name, node, msg)
+        lu.LogWarning("Could not prepare block device %s on node %s"
+                      " (is_primary=False, pass=1): %s",
+                      inst_disk.iv_name, node, msg)
         if not (ignore_secondaries or is_offline_secondary):
           disks_ok = False
 
@@ -6504,9 +6654,9 @@ def _AssembleInstanceDisks(lu, instance, disks=None, ignore_secondaries=False,
                                              True, idx)
       msg = result.fail_msg
       if msg:
-        lu.proc.LogWarning("Could not prepare block device %s on node %s"
-                           " (is_primary=True, pass=2): %s",
-                           inst_disk.iv_name, node, msg)
+        lu.LogWarning("Could not prepare block device %s on node %s"
+                      " (is_primary=True, pass=2): %s",
+                      inst_disk.iv_name, node, msg)
         disks_ok = False
       else:
         dev_path = result.payload
@@ -6531,9 +6681,9 @@ def _StartInstanceDisks(lu, instance, force):
   if not disks_ok:
     _ShutdownInstanceDisks(lu, instance)
     if force is not None and not force:
-      lu.proc.LogWarning("", hint="If the message above refers to a"
-                         " secondary node,"
-                         " you can retry the operation using '--force'.")
+      lu.LogWarning("",
+                    hint=("If the message above refers to a secondary node,"
+                          " you can retry the operation using '--force'"))
     raise errors.OpExecError("Disk consistency error")
 
 
@@ -6836,10 +6986,10 @@ class LUInstanceStartup(LogicalUnit):
     self.primary_offline = self.cfg.GetNodeInfo(instance.primary_node).offline
 
     if self.primary_offline and self.op.ignore_offline_nodes:
-      self.proc.LogWarning("Ignoring offline primary node")
+      self.LogWarning("Ignoring offline primary node")
 
       if self.op.hvparams or self.op.beparams:
-        self.proc.LogWarning("Overridden parameters are ignored")
+        self.LogWarning("Overridden parameters are ignored")
     else:
       _CheckNodeOnline(self, instance.primary_node)
 
@@ -6871,7 +7021,7 @@ class LUInstanceStartup(LogicalUnit):
 
     if self.primary_offline:
       assert self.op.ignore_offline_nodes
-      self.proc.LogInfo("Primary node offline, marked instance as started")
+      self.LogInfo("Primary node offline, marked instance as started")
     else:
       node_current = instance.primary_node
 
@@ -7026,7 +7176,7 @@ class LUInstanceShutdown(LogicalUnit):
       self.cfg.GetNodeInfo(self.instance.primary_node).offline
 
     if self.primary_offline and self.op.ignore_offline_nodes:
-      self.proc.LogWarning("Ignoring offline primary node")
+      self.LogWarning("Ignoring offline primary node")
     else:
       _CheckNodeOnline(self, self.instance.primary_node)
 
@@ -7043,12 +7193,12 @@ class LUInstanceShutdown(LogicalUnit):
 
     if self.primary_offline:
       assert self.op.ignore_offline_nodes
-      self.proc.LogInfo("Primary node offline, marked instance as stopped")
+      self.LogInfo("Primary node offline, marked instance as stopped")
     else:
       result = self.rpc.call_instance_shutdown(node_current, instance, timeout)
       msg = result.fail_msg
       if msg:
-        self.proc.LogWarning("Could not shutdown instance: %s" % msg)
+        self.LogWarning("Could not shutdown instance: %s", msg)
 
       _ShutdownInstanceDisks(self, instance)
 
@@ -7244,6 +7394,7 @@ class LUInstanceRecreateDisks(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+
     if self.op.nodes:
       self.op.nodes = [_ExpandNodeName(self.cfg, n) for n in self.op.nodes]
       self.needed_locks[locking.LEVEL_NODE] = list(self.op.nodes)
@@ -7252,6 +7403,8 @@ class LUInstanceRecreateDisks(LogicalUnit):
       if self.op.iallocator:
         # iallocator will select a new node in the same group
         self.needed_locks[locking.LEVEL_NODEGROUP] = []
+        self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
     self.needed_locks[locking.LEVEL_NODE_RES] = []
 
   def DeclareLocks(self, level):
@@ -7281,6 +7434,8 @@ class LUInstanceRecreateDisks(LogicalUnit):
         for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP):
           self.needed_locks[locking.LEVEL_NODE].extend(
             self.cfg.GetNodeGroup(group_uuid).members)
+
+        assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
       elif not self.op.nodes:
         self._LockInstancesNodes(primary_only=False)
     elif level == locking.LEVEL_NODE_RES:
@@ -7371,6 +7526,9 @@ class LUInstanceRecreateDisks(LogicalUnit):
       # Release unneeded node and node resource locks
       _ReleaseLocks(self, locking.LEVEL_NODE, keep=self.op.nodes)
       _ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=self.op.nodes)
+      _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+    assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
 
   def Exec(self, feedback_fn):
     """Recreate the disks.
@@ -7514,6 +7672,7 @@ class LUInstanceRename(LogicalUnit):
     # Change the instance lock. This is definitely safe while we hold the BGL.
     # Otherwise the new lock would have to be added in acquired mode.
     assert self.REQ_BGL
+    assert locking.BGL in self.owned_locks(locking.LEVEL_CLUSTER)
     self.glm.remove(locking.LEVEL_INSTANCE, old_name)
     self.glm.add(locking.LEVEL_INSTANCE, self.op.new_name)
 
@@ -7548,7 +7707,7 @@ class LUInstanceRename(LogicalUnit):
         msg = ("Could not run OS rename script for instance %s on node %s"
                " (but the instance has been renamed in Ganeti): %s" %
                (inst.name, inst.primary_node, msg))
-        self.proc.LogWarning(msg)
+        self.LogWarning(msg)
     finally:
       _ShutdownInstanceDisks(self, inst)
 
@@ -7676,6 +7835,59 @@ class LUInstanceQuery(NoHooksLU):
     return self.iq.OldStyleQuery(self)
 
 
+def _ExpandNamesForMigration(lu):
+  """Expands names for use with L{TLMigrateInstance}.
+
+  @type lu: L{LogicalUnit}
+
+  """
+  if lu.op.target_node is not None:
+    lu.op.target_node = _ExpandNodeName(lu.cfg, lu.op.target_node)
+
+  lu.needed_locks[locking.LEVEL_NODE] = []
+  lu.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  lu.needed_locks[locking.LEVEL_NODE_RES] = []
+  lu.recalculate_locks[locking.LEVEL_NODE_RES] = constants.LOCKS_REPLACE
+
+  # The node allocation lock is actually only needed for replicated instances
+  # (e.g. DRBD8) and if an iallocator is used.
+  lu.needed_locks[locking.LEVEL_NODE_ALLOC] = []
+
+
+def _DeclareLocksForMigration(lu, level):
+  """Declares locks for L{TLMigrateInstance}.
+
+  @type lu: L{LogicalUnit}
+  @param level: Lock level
+
+  """
+  if level == locking.LEVEL_NODE_ALLOC:
+    assert lu.op.instance_name in lu.owned_locks(locking.LEVEL_INSTANCE)
+
+    instance = lu.cfg.GetInstanceInfo(lu.op.instance_name)
+
+    if instance.disk_template in constants.DTS_EXT_MIRROR:
+      if lu.op.target_node is None:
+        lu.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+        lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+      else:
+        lu.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
+                                               lu.op.target_node]
+      del lu.recalculate_locks[locking.LEVEL_NODE]
+    else:
+      lu._LockInstancesNodes() # pylint: disable=W0212
+
+  elif level == locking.LEVEL_NODE:
+    # Node locks are declared together with the node allocation lock
+    assert lu.needed_locks[locking.LEVEL_NODE]
+
+  elif level == locking.LEVEL_NODE_RES:
+    # Copy node locks
+    lu.needed_locks[locking.LEVEL_NODE_RES] = \
+      _CopyLockList(lu.needed_locks[locking.LEVEL_NODE])
+
+
 class LUInstanceFailover(LogicalUnit):
   """Failover an instance.
 
@@ -7693,42 +7905,17 @@ class LUInstanceFailover(LogicalUnit):
 
   def ExpandNames(self):
     self._ExpandAndLockInstance()
+    _ExpandNamesForMigration(self)
 
-    if self.op.target_node is not None:
-      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
-
-    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
+    self._migrater = \
+      TLMigrateInstance(self, self.op.instance_name, False, True, False,
+                        self.op.ignore_consistency, True,
+                        self.op.shutdown_timeout, self.op.ignore_ipolicy)
 
-    ignore_consistency = self.op.ignore_consistency
-    shutdown_timeout = self.op.shutdown_timeout
-    self._migrater = TLMigrateInstance(self, self.op.instance_name,
-                                       cleanup=False,
-                                       failover=True,
-                                       ignore_consistency=ignore_consistency,
-                                       shutdown_timeout=shutdown_timeout,
-                                       ignore_ipolicy=self.op.ignore_ipolicy)
     self.tasklets = [self._migrater]
 
   def DeclareLocks(self, level):
-    if level == locking.LEVEL_NODE:
-      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
-      if instance.disk_template in constants.DTS_EXT_MIRROR:
-        if self.op.target_node is None:
-          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
-        else:
-          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
-                                                   self.op.target_node]
-        del self.recalculate_locks[locking.LEVEL_NODE]
-      else:
-        self._LockInstancesNodes()
-    elif level == locking.LEVEL_NODE_RES:
-      # Copy node locks
-      self.needed_locks[locking.LEVEL_NODE_RES] = \
-        _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+    _DeclareLocksForMigration(self, level)
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -7778,41 +7965,19 @@ class LUInstanceMigrate(LogicalUnit):
 
   def ExpandNames(self):
     self._ExpandAndLockInstance()
-
-    if self.op.target_node is not None:
-      self.op.target_node = _ExpandNodeName(self.cfg, self.op.target_node)
-
-    self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
-
-    self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+    _ExpandNamesForMigration(self)
 
     self._migrater = \
-      TLMigrateInstance(self, self.op.instance_name,
-                        cleanup=self.op.cleanup,
-                        failover=False,
-                        fallback=self.op.allow_failover,
-                        allow_runtime_changes=self.op.allow_runtime_changes,
-                        ignore_ipolicy=self.op.ignore_ipolicy)
+      TLMigrateInstance(self, self.op.instance_name, self.op.cleanup,
+                        False, self.op.allow_failover, False,
+                        self.op.allow_runtime_changes,
+                        constants.DEFAULT_SHUTDOWN_TIMEOUT,
+                        self.op.ignore_ipolicy)
+
     self.tasklets = [self._migrater]
 
   def DeclareLocks(self, level):
-    if level == locking.LEVEL_NODE:
-      instance = self.context.cfg.GetInstanceInfo(self.op.instance_name)
-      if instance.disk_template in constants.DTS_EXT_MIRROR:
-        if self.op.target_node is None:
-          self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
-        else:
-          self.needed_locks[locking.LEVEL_NODE] = [instance.primary_node,
-                                                   self.op.target_node]
-        del self.recalculate_locks[locking.LEVEL_NODE]
-      else:
-        self._LockInstancesNodes()
-    elif level == locking.LEVEL_NODE_RES:
-      # Copy node locks
-      self.needed_locks[locking.LEVEL_NODE_RES] = \
-        _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+    _DeclareLocksForMigration(self, level)
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -7969,10 +8134,10 @@ class LUInstanceMove(LogicalUnit):
     msg = result.fail_msg
     if msg:
       if self.op.ignore_consistency:
-        self.proc.LogWarning("Could not shutdown instance %s on node %s."
-                             " Proceeding anyway. Please make sure node"
-                             " %s is down. Error details: %s",
-                             instance.name, source_node, source_node, msg)
+        self.LogWarning("Could not shutdown instance %s on node %s."
+                        " Proceeding anyway. Please make sure node"
+                        " %s is down. Error details: %s",
+                        instance.name, source_node, source_node, msg)
       else:
         raise errors.OpExecError("Could not shutdown instance %s on"
                                  " node %s: %s" %
@@ -8098,8 +8263,7 @@ class LUNodeMigrate(LogicalUnit):
                                  target_node=self.op.target_node,
                                  allow_runtime_changes=allow_runtime_changes,
                                  ignore_ipolicy=self.op.ignore_ipolicy)]
-      for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)
-      ]
+      for inst in _GetNodePrimaryInstances(self.cfg, self.op.node_name)]
 
     # TODO: Run iallocator in this opcode and pass correct placement options to
     # OpInstanceMigrate. Since other jobs can modify the cluster between
@@ -8143,12 +8307,9 @@ class TLMigrateInstance(Tasklet):
   _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,
-               allow_runtime_changes=True,
-               shutdown_timeout=constants.DEFAULT_SHUTDOWN_TIMEOUT,
-               ignore_ipolicy=False):
+  def __init__(self, lu, instance_name, cleanup, failover, fallback,
+               ignore_consistency, allow_runtime_changes, shutdown_timeout,
+               ignore_ipolicy):
     """Initializes this class.
 
     """
@@ -8194,6 +8355,8 @@ class TLMigrateInstance(Tasklet):
                                  errors.ECODE_STATE)
 
     if instance.disk_template in constants.DTS_EXT_MIRROR:
+      assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC)
+
       _CheckIAllocatorOrNode(self.lu, "iallocator", "target_node")
 
       if self.lu.op.iallocator:
@@ -8225,8 +8388,11 @@ class TLMigrateInstance(Tasklet):
         # in the LU
         _ReleaseLocks(self.lu, locking.LEVEL_NODE,
                       keep=[instance.primary_node, self.target_node])
+        _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
 
     else:
+      assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
       secondary_nodes = instance.secondary_nodes
       if not secondary_nodes:
         raise errors.ConfigurationError("No secondary node but using"
@@ -8328,6 +8494,8 @@ class TLMigrateInstance(Tasklet):
     """Run the allocator based on input opcode.
 
     """
+    assert locking.NAL in self.lu.owned_locks(locking.LEVEL_NODE_ALLOC)
+
     # FIXME: add a self.ignore_ipolicy option
     req = iallocator.IAReqRelocate(name=self.instance_name,
                                    relocate_from=[self.instance.primary_node])
@@ -9013,9 +9181,11 @@ def _GenerateDiskTemplate(
                                         for i in range(disk_count)])
 
     if template_name == constants.DT_PLAIN:
+
       def logical_id_fn(idx, _, disk):
         vg = disk.get(constants.IDISK_VG, vgname)
         return (vg, names[idx])
+
     elif template_name in (constants.DT_FILE, constants.DT_SHARED_FILE):
       logical_id_fn = \
         lambda _, disk_index, disk: (file_driver,
@@ -9370,13 +9540,15 @@ def _CheckOSParams(lu, required, nodenames, osname, osparams):
                  osname, node)
 
 
-def _CreateInstanceAllocRequest(op, disks, nics, beparams):
+def _CreateInstanceAllocRequest(op, disks, nics, beparams, node_whitelist):
   """Wrapper around IAReqInstanceAlloc.
 
   @param op: The instance opcode
   @param disks: The computed disks
   @param nics: The computed nics
   @param beparams: The full filled beparams
+  @param node_whitelist: List of nodes which should appear as online to the
+    allocator (unless the node is already marked offline)
 
   @returns: A filled L{iallocator.IAReqInstanceAlloc}
 
@@ -9391,23 +9563,24 @@ def _CreateInstanceAllocRequest(op, disks, nics, beparams):
                                        spindle_use=spindle_use,
                                        disks=disks,
                                        nics=[n.ToDict() for n in nics],
-                                       hypervisor=op.hypervisor)
+                                       hypervisor=op.hypervisor,
+                                       node_whitelist=node_whitelist)
 
 
-def _ComputeNics(op, cluster, default_ip, cfg, proc):
+def _ComputeNics(op, cluster, default_ip, cfg, ec_id):
   """Computes the nics.
 
   @param op: The instance opcode
   @param cluster: Cluster configuration object
   @param default_ip: The default ip to assign
   @param cfg: An instance of the configuration object
-  @param proc: The executer instance
+  @param ec_id: Execution context ID
 
   @returns: The build up nics
 
   """
   nics = []
-  for idx, nic in enumerate(op.nics):
+  for nic in op.nics:
     nic_mode_req = nic.get(constants.INIC_MODE, None)
     nic_mode = nic_mode_req
     if nic_mode is None or nic_mode == constants.VALUE_AUTO:
@@ -9461,7 +9634,7 @@ def _ComputeNics(op, cluster, default_ip, cfg, proc):
 
       try:
         # TODO: We need to factor this out
-        cfg.ReserveMAC(mac, proc.GetECId())
+        cfg.ReserveMAC(mac, ec_id)
       except errors.ReservationError:
         raise errors.OpPrereqError("MAC address %s already in use"
                                    " in cluster" % mac,
@@ -9476,7 +9649,8 @@ def _ComputeNics(op, cluster, default_ip, cfg, proc):
 
     check_params = cluster.SimpleFillNIC(nicparams)
     objects.NIC.CheckParameterSyntax(check_params)
-    nics.append(objects.NIC(mac=mac, ip=nic_ip, nicparams=nicparams))
+    nics.append(objects.NIC(mac=mac, ip=nic_ip,
+                            network=net, nicparams=nicparams))
 
   return nics
 
@@ -9723,7 +9897,11 @@ class LUInstanceCreate(LogicalUnit):
       # 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_RES] = locking.ALL_SET
+      self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
+      if self.op.opportunistic_locking:
+        self.opportunistic_locks[locking.LEVEL_NODE] = True
+        self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
     else:
       self.op.pnode = _ExpandNodeName(self.cfg, self.op.pnode)
       nodelist = [self.op.pnode]
@@ -9731,9 +9909,6 @@ class LUInstanceCreate(LogicalUnit):
         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:
@@ -9745,6 +9920,7 @@ class LUInstanceCreate(LogicalUnit):
 
       if src_node is None:
         self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+        self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
         self.op.src_node = None
         if os.path.isabs(src_path):
           raise errors.OpPrereqError("Importing an instance from a path"
@@ -9758,23 +9934,40 @@ class LUInstanceCreate(LogicalUnit):
           self.op.src_path = src_path = \
             utils.PathJoin(pathutils.EXPORT_DIR, src_path)
 
+    self.needed_locks[locking.LEVEL_NODE_RES] = \
+      _CopyLockList(self.needed_locks[locking.LEVEL_NODE])
+
   def _RunAllocator(self):
     """Run the allocator based on input opcode.
 
     """
+    if self.op.opportunistic_locking:
+      # Only consider nodes for which a lock is held
+      node_whitelist = self.owned_locks(locking.LEVEL_NODE)
+    else:
+      node_whitelist = None
+
     #TODO Export network to iallocator so that it chooses a pnode
     #     in a nodegroup that has the desired network connected to
     req = _CreateInstanceAllocRequest(self.op, self.disks,
-                                      self.nics, self.be_full)
+                                      self.nics, self.be_full,
+                                      node_whitelist)
     ial = iallocator.IAllocator(self.cfg, self.rpc, req)
 
     ial.Run(self.op.iallocator)
 
     if not ial.success:
+      # When opportunistic locks are used only a temporary failure is generated
+      if self.op.opportunistic_locking:
+        ecode = errors.ECODE_TEMP_NORES
+      else:
+        ecode = errors.ECODE_NORES
+
       raise errors.OpPrereqError("Can't compute nodes using"
                                  " iallocator '%s': %s" %
                                  (self.op.iallocator, ial.info),
-                                 errors.ECODE_NORES)
+                                 ecode)
+
     self.op.pnode = ial.result[0]
     self.LogInfo("Selected nodes for instance %s via iallocator %s: %s",
                  self.op.instance_name, self.op.iallocator,
@@ -10075,7 +10268,7 @@ class LUInstanceCreate(LogicalUnit):
 
     # NIC buildup
     self.nics = _ComputeNics(self.op, cluster, self.hostname1.ip, self.cfg,
-                             self.proc)
+                             self.proc.GetECId())
 
     # disk checks/pre-build
     default_vg = self.cfg.GetVGName()
@@ -10120,7 +10313,7 @@ class LUInstanceCreate(LogicalUnit):
     # creation job will fail.
     for nic in self.nics:
       if nic.mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
-        nic.mac = self.cfg.GenerateMAC(self.proc.GetECId())
+        nic.mac = self.cfg.GenerateMAC(nic.network, self.proc.GetECId())
 
     #### allocator run
 
@@ -10128,12 +10321,14 @@ class LUInstanceCreate(LogicalUnit):
       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]))
-    _ReleaseLocks(self, locking.LEVEL_NODE_RES,
-                  keep=filter(None, [self.op.pnode, self.op.snode,
-                                     self.op.src_node]))
+    keep_locks = filter(None, [self.op.pnode, self.op.snode, self.op.src_node])
+    _ReleaseLocks(self, locking.LEVEL_NODE, keep=keep_locks)
+    _ReleaseLocks(self, locking.LEVEL_NODE_RES, keep=keep_locks)
+    _ReleaseLocks(self, locking.LEVEL_NODE_ALLOC)
+
+    assert (self.owned_locks(locking.LEVEL_NODE) ==
+            self.owned_locks(locking.LEVEL_NODE_RES)), \
+      "Node locks differ from node resource locks"
 
     #### node related checks
 
@@ -10169,7 +10364,6 @@ class LUInstanceCreate(LogicalUnit):
                      (idx, netparams.values()))
         nic.nicparams = dict(netparams)
         if nic.ip is not None:
-          filled_params = cluster.SimpleFillNIC(nic.nicparams)
           if nic.ip.lower() == constants.NIC_IP_POOL:
             try:
               nic.ip = self.cfg.GenerateIp(net, self.proc.GetECId())
@@ -10191,7 +10385,6 @@ class LUInstanceCreate(LogicalUnit):
         if self.op.conflicts_check:
           _CheckForConflictingIp(self, nic.ip, self.pnode.name)
 
-
     # mirror node verification
     if self.op.disk_template in constants.DTS_INT_MIRROR:
       if self.op.snode == pnode.name:
@@ -10293,7 +10486,7 @@ class LUInstanceCreate(LogicalUnit):
       if baddisks:
         raise errors.OpPrereqError("Device node(s) %s lie outside %s and"
                                    " cannot be adopted" %
-                                   (", ".join(baddisks),
+                                   (utils.CommaJoin(baddisks),
                                     constants.ADOPTABLE_BLOCKDEV_ROOT),
                                    errors.ECODE_INVAL)
 
@@ -10360,6 +10553,7 @@ class LUInstanceCreate(LogicalUnit):
     assert not (self.owned_locks(locking.LEVEL_NODE_RES) -
                 self.owned_locks(locking.LEVEL_NODE)), \
       "Node locks differ from node resource locks"
+    assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
 
     ht_kind = self.op.hypervisor
     if ht_kind in constants.HTS_REQ_PORT:
@@ -10638,11 +10832,19 @@ class LUInstanceMultiAlloc(NoHooksLU):
 
     """
     self.share_locks = _ShareAll()
-    self.needed_locks = {}
+    self.needed_locks = {
+      # iallocator will select nodes and even if no iallocator is used,
+      # collisions with LUInstanceCreate should be avoided
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+      }
 
     if self.op.iallocator:
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
       self.needed_locks[locking.LEVEL_NODE_RES] = locking.ALL_SET
+
+      if self.op.opportunistic_locking:
+        self.opportunistic_locks[locking.LEVEL_NODE] = True
+        self.opportunistic_locks[locking.LEVEL_NODE_RES] = True
     else:
       nodeslist = []
       for inst in self.op.instances:
@@ -10663,11 +10865,21 @@ class LUInstanceMultiAlloc(NoHooksLU):
     """
     cluster = self.cfg.GetClusterInfo()
     default_vg = self.cfg.GetVGName()
+    ec_id = self.proc.GetECId()
+
+    if self.op.opportunistic_locking:
+      # Only consider nodes for which a lock is held
+      node_whitelist = self.owned_locks(locking.LEVEL_NODE)
+    else:
+      node_whitelist = None
+
     insts = [_CreateInstanceAllocRequest(op, _ComputeDisks(op, default_vg),
                                          _ComputeNics(op, cluster, None,
-                                                      self.cfg, self.proc),
-                                         _ComputeFullBeParams(op, cluster))
+                                                      self.cfg, ec_id),
+                                         _ComputeFullBeParams(op, cluster),
+                                         node_whitelist)
              for op in self.op.instances]
+
     req = iallocator.IAReqMultiInstanceAlloc(instances=insts)
     ial = iallocator.IAllocator(self.cfg, self.rpc, req)
 
@@ -10856,12 +11068,13 @@ class LUInstanceReplaceDisks(LogicalUnit):
       if self.op.iallocator is not None:
         # iallocator will select a new node in the same group
         self.needed_locks[locking.LEVEL_NODEGROUP] = []
+        self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
 
     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.op.disks, self.op.early_release,
                                    self.op.ignore_ipolicy)
 
     self.tasklets = [self.replacer]
@@ -10882,6 +11095,7 @@ class LUInstanceReplaceDisks(LogicalUnit):
       if self.op.iallocator is not None:
         assert self.op.remote_node is None
         assert not self.needed_locks[locking.LEVEL_NODE]
+        assert locking.NAL in self.owned_locks(locking.LEVEL_NODE_ALLOC)
 
         # Lock member nodes of all locked groups
         self.needed_locks[locking.LEVEL_NODE] = \
@@ -10889,7 +11103,10 @@ class LUInstanceReplaceDisks(LogicalUnit):
              for group_uuid in self.owned_locks(locking.LEVEL_NODEGROUP)
              for node_name in self.cfg.GetNodeGroup(group_uuid).members]
       else:
+        assert not self.glm.is_owned(locking.LEVEL_NODE_ALLOC)
+
         self._LockInstancesNodes()
+
     elif level == locking.LEVEL_NODE_RES:
       # Reuse node locks
       self.needed_locks[locking.LEVEL_NODE_RES] = \
@@ -10945,7 +11162,7 @@ class TLReplaceDisks(Tasklet):
 
   """
   def __init__(self, lu, instance_name, mode, iallocator_name, remote_node,
-               disks, delay_iallocator, early_release, ignore_ipolicy):
+               disks, early_release, ignore_ipolicy):
     """Initializes this class.
 
     """
@@ -10957,7 +11174,6 @@ class TLReplaceDisks(Tasklet):
     self.iallocator_name = iallocator_name
     self.remote_node = remote_node
     self.disks = disks
-    self.delay_iallocator = delay_iallocator
     self.early_release = early_release
     self.ignore_ipolicy = ignore_ipolicy
 
@@ -11042,18 +11258,6 @@ class TLReplaceDisks(Tasklet):
                                  len(instance.secondary_nodes),
                                  errors.ECODE_FAULT)
 
-    if not self.delay_iallocator:
-      self._CheckPrereq2()
-
-  def _CheckPrereq2(self):
-    """Check prerequisites, second part.
-
-    This function should always be part of CheckPrereq. It was separated and is
-    now called from Exec because during node evacuation iallocator was only
-    called with an unmodified cluster model, not taking planned changes into
-    account.
-
-    """
     instance = self.instance
     secondary_node = instance.secondary_nodes[0]
 
@@ -11174,10 +11378,10 @@ class TLReplaceDisks(Tasklet):
     # Release unneeded node and node resource locks
     _ReleaseLocks(self.lu, locking.LEVEL_NODE, keep=touched_nodes)
     _ReleaseLocks(self.lu, locking.LEVEL_NODE_RES, keep=touched_nodes)
+    _ReleaseLocks(self.lu, locking.LEVEL_NODE_ALLOC)
 
     # Release any owned node group
-    if self.lu.glm.is_owned(locking.LEVEL_NODEGROUP):
-      _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
+    _ReleaseLocks(self.lu, locking.LEVEL_NODEGROUP)
 
     # Check whether disks are valid
     for disk_idx in self.disks:
@@ -11193,9 +11397,6 @@ class TLReplaceDisks(Tasklet):
     This dispatches the disk replacement to the appropriate handler.
 
     """
-    if self.delay_iallocator:
-      self._CheckPrereq2()
-
     if __debug__:
       # Verify owned locks before starting operation
       owned_nodes = self.lu.owned_locks(locking.LEVEL_NODE)
@@ -11204,6 +11405,7 @@ class TLReplaceDisks(Tasklet):
            (owned_nodes, self.node_secondary_ip.keys()))
       assert (self.lu.owned_locks(locking.LEVEL_NODE) ==
               self.lu.owned_locks(locking.LEVEL_NODE_RES))
+      assert not self.lu.glm.is_owned(locking.LEVEL_NODE_ALLOC)
 
       owned_instances = self.lu.owned_locks(locking.LEVEL_INSTANCE)
       assert list(owned_instances) == [self.instance_name], \
@@ -11280,7 +11482,7 @@ class TLReplaceDisks(Tasklet):
         continue
 
       for node in nodes:
-        self.lu.LogInfo("Checking disk/%d on %s" % (idx, node))
+        self.lu.LogInfo("Checking disk/%d on %s", idx, node)
         self.cfg.SetDiskID(dev, node)
 
         result = _BlockdevFind(self, node, dev, self.instance)
@@ -11320,7 +11522,7 @@ class TLReplaceDisks(Tasklet):
       if idx not in self.disks:
         continue
 
-      self.lu.LogInfo("Adding storage on %s for disk/%d" % (node_name, idx))
+      self.lu.LogInfo("Adding storage on %s for disk/%d", node_name, idx)
 
       self.cfg.SetDiskID(dev, node_name)
 
@@ -11367,14 +11569,14 @@ class TLReplaceDisks(Tasklet):
 
   def _RemoveOldStorage(self, node_name, iv_names):
     for name, (_, old_lvs, _) in iv_names.iteritems():
-      self.lu.LogInfo("Remove logical volumes for %s" % name)
+      self.lu.LogInfo("Remove logical volumes for %s", name)
 
       for lv in old_lvs:
         self.cfg.SetDiskID(lv, node_name)
 
         msg = self.rpc.call_blockdev_remove(node_name, lv).fail_msg
         if msg:
-          self.lu.LogWarning("Can't remove old LV: %s" % msg,
+          self.lu.LogWarning("Can't remove old LV: %s", msg,
                              hint="remove unused LVs manually")
 
   def _ExecDrbd8DiskOnly(self, feedback_fn): # pylint: disable=W0613
@@ -11419,7 +11621,7 @@ class TLReplaceDisks(Tasklet):
     # Step: for each lv, detach+rename*2+attach
     self.lu.LogStep(4, steps_total, "Changing drbd configuration")
     for dev, old_lvs, new_lvs in iv_names.itervalues():
-      self.lu.LogInfo("Detaching %s drbd from local storage" % dev.iv_name)
+      self.lu.LogInfo("Detaching %s drbd from local storage", dev.iv_name)
 
       result = self.rpc.call_blockdev_removechildren(self.target_node, dev,
                                                      old_lvs)
@@ -11473,7 +11675,7 @@ class TLReplaceDisks(Tasklet):
         self.cfg.SetDiskID(disk, self.target_node)
 
       # Now that the new lvs have the old name, we can add them to the device
-      self.lu.LogInfo("Adding new mirror component on %s" % self.target_node)
+      self.lu.LogInfo("Adding new mirror component on %s", self.target_node)
       result = self.rpc.call_blockdev_addchildren(self.target_node,
                                                   (dev, self.instance), new_lvs)
       msg = result.fail_msg
@@ -11611,7 +11813,7 @@ class TLReplaceDisks(Tasklet):
 
     # We have new devices, shutdown the drbd on the old secondary
     for idx, dev in enumerate(self.instance.disks):
-      self.lu.LogInfo("Shutting down drbd for disk/%d on old node" % idx)
+      self.lu.LogInfo("Shutting down drbd for disk/%d on old node", idx)
       self.cfg.SetDiskID(dev, self.target_node)
       msg = self.rpc.call_blockdev_shutdown(self.target_node,
                                             (dev, self.instance)).fail_msg
@@ -11723,7 +11925,7 @@ class LURepairNodeStorage(NoHooksLU):
                                    errors.ECODE_STATE)
     except errors.OpPrereqError, err:
       if self.op.ignore_consistency:
-        self.proc.LogWarning(str(err.args[0]))
+        self.LogWarning(str(err.args[0]))
       else:
         raise
 
@@ -11944,8 +12146,7 @@ class LUNodeEvacuate(NoHooksLU):
                                         disks=[],
                                         mode=constants.REPLACE_DISK_CHG,
                                         early_release=self.op.early_release)]
-        for instance_name in self.instance_names
-        ]
+        for instance_name in self.instance_names]
 
     else:
       raise errors.ProgrammerError("No iallocator or remote node")
@@ -12202,14 +12403,14 @@ class LUInstanceGrowDisk(LogicalUnit):
     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")
+        self.LogWarning("Disk syncing has not returned a good status; check"
+                        " the instance")
       if instance.admin_state != constants.ADMINST_UP:
         _SafeShutdownInstanceDisks(self, instance, disks=[disk])
     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.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)
@@ -12735,15 +12936,23 @@ class LUInstanceSetParams(LogicalUnit):
 
   def ExpandNames(self):
     self._ExpandAndLockInstance()
+    self.needed_locks[locking.LEVEL_NODEGROUP] = []
     # 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_RES] = []
     self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+    # Look node group to look up the ipolicy
+    self.share_locks[locking.LEVEL_NODEGROUP] = 1
 
   def DeclareLocks(self, level):
-    # TODO: Acquire group lock in shared mode (disk parameters)
-    if level == locking.LEVEL_NODE:
+    if level == locking.LEVEL_NODEGROUP:
+      assert not self.needed_locks[locking.LEVEL_NODEGROUP]
+      # Acquire locks for the instance's nodegroups optimistically. Needs
+      # to be verified in CheckPrereq
+      self.needed_locks[locking.LEVEL_NODEGROUP] = \
+        self.cfg.GetInstanceNodeGroups(self.op.instance_name)
+    elif level == locking.LEVEL_NODE:
       self._LockInstancesNodes()
       if self.op.disk_template and self.op.remote_node:
         self.op.remote_node = _ExpandNodeName(self.cfg, self.op.remote_node)
@@ -12759,7 +12968,7 @@ class LUInstanceSetParams(LogicalUnit):
     This runs on the master, primary and secondaries.
 
     """
-    args = dict()
+    args = {}
     if constants.BE_MINMEM in self.be_new:
       args["minmem"] = self.be_new[constants.BE_MINMEM]
     if constants.BE_MAXMEM in self.be_new:
@@ -12773,10 +12982,10 @@ class LUInstanceSetParams(LogicalUnit):
       nics = []
 
       for nic in self._new_nics:
-        nicparams = self.cluster.SimpleFillNIC(nic.nicparams)
-        mode = nicparams[constants.NIC_MODE]
-        link = nicparams[constants.NIC_LINK]
-        nics.append((nic.ip, nic.mac, mode, link, nic.network))
+        n = copy.deepcopy(nic)
+        nicparams = self.cluster.SimpleFillNIC(n.nicparams)
+        n.nicparams = nicparams
+        nics.append(_NICToTuple(self, n))
 
       args["nics"] = nics
 
@@ -12810,7 +13019,7 @@ class LUInstanceSetParams(LogicalUnit):
       netparams = self.cfg.GetGroupNetParams(new_net, pnode)
       if netparams is None:
         raise errors.OpPrereqError("No netparams found for the network"
-                                   " %s, propably not connected." % new_net,
+                                   " %s, probably not connected" % new_net,
                                    errors.ECODE_INVAL)
       new_params = dict(netparams)
     else:
@@ -12846,7 +13055,7 @@ class LUInstanceSetParams(LogicalUnit):
       elif mac in (constants.VALUE_AUTO, constants.VALUE_GENERATE):
         # otherwise generate the MAC address
         params[constants.INIC_MAC] = \
-          self.cfg.GenerateMAC(self.proc.GetECId())
+          self.cfg.GenerateMAC(new_net, self.proc.GetECId())
       else:
         # or validate/reserve the current one
         try:
@@ -12856,6 +13065,7 @@ class LUInstanceSetParams(LogicalUnit):
                                      " in cluster" % mac,
                                      errors.ECODE_NOTUNIQUE)
     elif new_net != old_net:
+
       def get_net_prefix(net):
         if net:
           uuid = self.cfg.LookupNetwork(net)
@@ -12863,11 +13073,12 @@ class LUInstanceSetParams(LogicalUnit):
             nobj = self.cfg.GetNetwork(uuid)
             return nobj.mac_prefix
         return None
+
       new_prefix = get_net_prefix(new_net)
       old_prefix = get_net_prefix(old_net)
       if old_prefix != new_prefix:
         params[constants.INIC_MAC] = \
-          self.cfg.GenerateMAC(self.proc.GetECId())
+          self.cfg.GenerateMAC(new_net, self.proc.GetECId())
 
     #if there is a change in nic-network configuration
     new_ip = params.get(constants.INIC_IP, old_ip)
@@ -12893,7 +13104,7 @@ class LUInstanceSetParams(LogicalUnit):
                                          errors.ECODE_NOTUNIQUE)
         elif new_ip.lower() == constants.NIC_IP_POOL:
           raise errors.OpPrereqError("ip=pool, but no network found",
-                                     ECODEE_INVAL)
+                                     errors.ECODE_INVAL)
         else:
           # new net is None
           if self.op.conflicts_check:
@@ -12911,7 +13122,7 @@ class LUInstanceSetParams(LogicalUnit):
     elif (old_net is not None and
           (req_link is not None or req_mode is not None)):
       raise errors.OpPrereqError("Not allowed to change link or mode of"
-                                 " a NIC that is connected to a network.",
+                                 " a NIC that is connected to a network",
                                  errors.ECODE_INVAL)
 
     private.params = new_params
@@ -12923,17 +13134,26 @@ class LUInstanceSetParams(LogicalUnit):
     This only checks the instance list against the existing names.
 
     """
-    # checking the new params on the primary/secondary nodes
-
+    assert self.op.instance_name in self.owned_locks(locking.LEVEL_INSTANCE)
     instance = self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+
     cluster = self.cluster = self.cfg.GetClusterInfo()
     assert self.instance is not None, \
       "Cannot retrieve locked instance %s" % self.op.instance_name
+
     pnode = instance.primary_node
+    assert pnode in self.owned_locks(locking.LEVEL_NODE)
     nodelist = list(instance.all_nodes)
     pnode_info = self.cfg.GetNodeInfo(pnode)
     self.diskparams = self.cfg.GetInstanceDiskParams(instance)
 
+    #_CheckInstanceNodeGroups(self.cfg, self.op.instance_name, owned_groups)
+    assert pnode_info.group in self.owned_locks(locking.LEVEL_NODEGROUP)
+    group_info = self.cfg.GetNodeGroup(pnode_info.group)
+
+    # dictionary with instance information after the modification
+    ispec = {}
+
     # Prepare disk/NIC modifications
     self.diskmod = PrepareContainerMods(self.op.disks, None)
     self.nicmod = PrepareContainerMods(self.op.nics, _InstNicModPrivate)
@@ -13159,7 +13379,7 @@ class LUInstanceSetParams(LogicalUnit):
                                    nic.nicparams, cluster, pnode)
       return None
 
-    def _PrepareNicRemove(_, params, private):
+    def _PrepareNicRemove(_, params, __):
       ip = params.ip
       net = params.network
       if net is not None and ip is not None:
@@ -13181,6 +13401,11 @@ class LUInstanceSetParams(LogicalUnit):
       raise errors.OpPrereqError("Instance has too many disks (%d), cannot add"
                                  " more" % constants.MAX_DISKS,
                                  errors.ECODE_STATE)
+    disk_sizes = [disk.size for disk in instance.disks]
+    disk_sizes.extend(params["size"] for (op, idx, params, private) in
+                      self.diskmod if op == constants.DDM_ADD)
+    ispec[constants.ISPEC_DISK_COUNT] = len(disk_sizes)
+    ispec[constants.ISPEC_DISK_SIZE] = disk_sizes
 
     if self.op.offline is not None:
       if self.op.offline:
@@ -13197,8 +13422,38 @@ class LUInstanceSetParams(LogicalUnit):
       ApplyContainerMods("NIC", nics, self._nic_chgdesc, self.nicmod,
                          self._CreateNewNic, self._ApplyNicMods, None)
       self._new_nics = nics
+      ispec[constants.ISPEC_NIC_COUNT] = len(self._new_nics)
     else:
       self._new_nics = None
+      ispec[constants.ISPEC_NIC_COUNT] = len(instance.nics)
+
+    if not self.op.ignore_ipolicy:
+      ipolicy = ganeti.masterd.instance.CalculateGroupIPolicy(cluster,
+                                                              group_info)
+
+      # Fill ispec with backend parameters
+      ispec[constants.ISPEC_SPINDLE_USE] = \
+        self.be_new.get(constants.BE_SPINDLE_USE, None)
+      ispec[constants.ISPEC_CPU_COUNT] = self.be_new.get(constants.BE_VCPUS,
+                                                         None)
+
+      # Copy ispec to verify parameters with min/max values separately
+      ispec_max = ispec.copy()
+      ispec_max[constants.ISPEC_MEM_SIZE] = \
+        self.be_new.get(constants.BE_MAXMEM, None)
+      res_max = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_max)
+      ispec_min = ispec.copy()
+      ispec_min[constants.ISPEC_MEM_SIZE] = \
+        self.be_new.get(constants.BE_MINMEM, None)
+      res_min = _ComputeIPolicyInstanceSpecViolation(ipolicy, ispec_min)
+
+      if (res_max or res_min):
+        # FIXME: Improve error message by including information about whether
+        # the upper or lower limit of the parameter fails the ipolicy.
+        msg = ("Instance allocation to group %s (%s) violates policy: %s" %
+               (group_info, group_info.name,
+                utils.CommaJoin(set(res_max + res_min))))
+        raise errors.OpPrereqError(msg, errors.ECODE_INVAL)
 
   def _ConvertPlainToDrbd(self, feedback_fn):
     """Converts an instance from plain to drbd.
@@ -13385,16 +13640,16 @@ class LUInstanceSetParams(LogicalUnit):
     """
     mac = params[constants.INIC_MAC]
     ip = params.get(constants.INIC_IP, None)
-    network = params.get(constants.INIC_NETWORK, None)
+    net = params.get(constants.INIC_NETWORK, None)
     #TODO: not private.filled?? can a nic have no nicparams??
     nicparams = private.filled
 
-    return (objects.NIC(mac=mac, ip=ip, network=network, nicparams=nicparams), [
+    return (objects.NIC(mac=mac, ip=ip, network=net, nicparams=nicparams), [
       ("nic.%d" % idx,
        "add:mac=%s,ip=%s,mode=%s,link=%s,network=%s" %
        (mac, ip, private.filled[constants.NIC_MODE],
        private.filled[constants.NIC_LINK],
-       network)),
+       net)),
       ])
 
   @staticmethod
@@ -13541,9 +13796,11 @@ class LUInstanceChangeGroup(LogicalUnit):
 
   def ExpandNames(self):
     self.share_locks = _ShareAll()
+
     self.needed_locks = {
       locking.LEVEL_NODEGROUP: [],
       locking.LEVEL_NODE: [],
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
       }
 
     self._ExpandAndLockInstance()
@@ -13726,6 +13983,9 @@ class _ExportQuery(_QueryBase):
         locking.LEVEL_NODE: self.wanted,
         }
 
+      if not self.names:
+        lu.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
   def DeclareLocks(self, lu, level):
     pass
 
@@ -13843,6 +14103,11 @@ class LUBackupExport(LogicalUnit):
       #  - removing the removal operation altogether
       self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
 
+      # Allocations should be stopped while this LU runs with node locks, but
+      # it doesn't have to be exclusive
+      self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
+      self.needed_locks[locking.LEVEL_NODE_ALLOC] = locking.ALL_SET
+
   def DeclareLocks(self, level):
     """Last minute lock declaration."""
     # All nodes are locked anyway, so nothing to do here.
@@ -14109,11 +14374,19 @@ class LUBackupRemove(NoHooksLU):
   REQ_BGL = False
 
   def ExpandNames(self):
-    self.needed_locks = {}
-    # We need all nodes to be locked in order for RemoveExport to work, but we
-    # don't need to lock the instance itself, as nothing will happen to it (and
-    # we can remove exports also for a removed instance)
-    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+    self.needed_locks = {
+      # We need all nodes to be locked in order for RemoveExport to work, but
+      # we don't need to lock the instance itself, as nothing will happen to it
+      # (and we can remove exports also for a removed instance)
+      locking.LEVEL_NODE: locking.ALL_SET,
+
+      # Removing backups is quick, so blocking allocations is justified
+      locking.LEVEL_NODE_ALLOC: locking.ALL_SET,
+      }
+
+    # Allocations should be stopped while this LU runs with node locks, but it
+    # doesn't have to be exclusive
+    self.share_locks[locking.LEVEL_NODE_ALLOC] = 1
 
   def Exec(self, feedback_fn):
     """Remove any export.
@@ -14953,6 +15226,10 @@ class TagsLU(NoHooksLU): # pylint: disable=W0223
       self.group_uuid = self.cfg.LookupNodeGroup(self.op.name)
       lock_level = locking.LEVEL_NODEGROUP
       lock_name = self.group_uuid
+    elif self.op.kind == constants.TAG_NETWORK:
+      self.network_uuid = self.cfg.LookupNetwork(self.op.name)
+      lock_level = locking.LEVEL_NETWORK
+      lock_name = self.network_uuid
     else:
       lock_level = None
       lock_name = None
@@ -14975,6 +15252,8 @@ class TagsLU(NoHooksLU): # pylint: disable=W0223
       self.target = self.cfg.GetInstanceInfo(self.op.name)
     elif self.op.kind == constants.TAG_NODEGROUP:
       self.target = self.cfg.GetNodeGroup(self.group_uuid)
+    elif self.op.kind == constants.TAG_NETWORK:
+      self.target = self.cfg.GetNetwork(self.network_uuid)
     else:
       raise errors.OpPrereqError("Wrong tag type requested (%s)" %
                                  str(self.op.kind), errors.ECODE_INVAL)
@@ -15146,7 +15425,7 @@ class LUTestDelay(NoHooksLU):
     else:
       top_value = self.op.repeat - 1
       for i in range(self.op.repeat):
-        self.LogInfo("Test delay iteration %d/%d" % (i, top_value))
+        self.LogInfo("Test delay iteration %d/%d", i, top_value)
         self._TestDelay()
 
 
@@ -15387,7 +15666,7 @@ class LUTestAllocator(NoHooksLU):
                                  self.op.mode, errors.ECODE_INVAL)
 
     if self.op.direction == constants.IALLOCATOR_DIR_OUT:
-      if self.op.allocator is None:
+      if self.op.iallocator is None:
         raise errors.OpPrereqError("Missing allocator name",
                                    errors.ECODE_INVAL)
     elif self.op.direction != constants.IALLOCATOR_DIR_IN:
@@ -15440,11 +15719,11 @@ class LUTestAllocator(NoHooksLU):
     if self.op.direction == constants.IALLOCATOR_DIR_IN:
       result = ial.in_text
     else:
-      ial.Run(self.op.allocator, validate=False)
+      ial.Run(self.op.iallocator, validate=False)
       result = ial.out_text
     return result
 
-# Network LUs
+
 class LUNetworkAdd(LogicalUnit):
   """Logical unit for creating networks.
 
@@ -15460,18 +15739,25 @@ class LUNetworkAdd(LogicalUnit):
     mn = self.cfg.GetMasterNode()
     return ([mn], [mn])
 
+  def CheckArguments(self):
+    if self.op.mac_prefix:
+      self.op.mac_prefix = \
+        utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix)
+
   def ExpandNames(self):
     self.network_uuid = self.cfg.GenerateUniqueID(self.proc.GetECId())
-    self.needed_locks = {}
-    self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid
 
-  def CheckPrereq(self):
-    """Check prerequisites.
+    if self.op.conflicts_check:
+      self.share_locks[locking.LEVEL_NODE] = 1
+      self.needed_locks = {
+        locking.LEVEL_NODE: locking.ALL_SET,
+        }
+    else:
+      self.needed_locks = {}
 
-    This checks that the given group name is not an existing node group
-    already.
+    self.add_locks[locking.LEVEL_NETWORK] = self.network_uuid
 
-    """
+  def CheckPrereq(self):
     if self.op.network is None:
       raise errors.OpPrereqError("Network must be given",
                                  errors.ECODE_INVAL)
@@ -15482,21 +15768,25 @@ class LUNetworkAdd(LogicalUnit):
       raise errors.OpPrereqError("Network '%s' already defined" %
                                  self.op.network, errors.ECODE_EXISTS)
 
+    # Check tag validity
+    for tag in self.op.tags:
+      objects.TaggableObject.ValidateTag(tag)
 
   def BuildHooksEnv(self):
     """Build hooks env.
 
     """
-    env = {
-      "NETWORK_NAME": self.op.network_name,
-      "NETWORK_SUBNET": self.op.network,
-      "NETWORK_GATEWAY": self.op.gateway,
-      "NETWORK_SUBNET6": self.op.network6,
-      "NETWORK_GATEWAY6": self.op.gateway6,
-      "NETWORK_MAC_PREFIX": self.op.mac_prefix,
-      "NETWORK_TYPE": self.op.network_type,
+    args = {
+      "name": self.op.network_name,
+      "subnet": self.op.network,
+      "gateway": self.op.gateway,
+      "network6": self.op.network6,
+      "gateway6": self.op.gateway6,
+      "mac_prefix": self.op.mac_prefix,
+      "network_type": self.op.network_type,
+      "tags": self.op.tags,
       }
-    return env
+    return _BuildNetworkHookEnv(**args) # pylint: disable=W0142
 
   def Exec(self, feedback_fn):
     """Add the ip pool to the cluster.
@@ -15510,7 +15800,7 @@ class LUNetworkAdd(LogicalUnit):
                            mac_prefix=self.op.mac_prefix,
                            network_type=self.op.network_type,
                            uuid=self.network_uuid,
-                           family=4)
+                           family=constants.IP4_VERSION)
     # Initialize the associated address pool
     try:
       pool = network.AddressPool.InitializeNetwork(nobj)
@@ -15520,21 +15810,26 @@ class LUNetworkAdd(LogicalUnit):
     # Check if we need to reserve the nodes and the cluster master IP
     # These may not be allocated to any instances in routed mode, as
     # they wouldn't function anyway.
-    for node in self.cfg.GetAllNodesInfo().values():
-      for ip in [node.primary_ip, node.secondary_ip]:
-        try:
-          pool.Reserve(ip)
-          self.LogInfo("Reserved node %s's IP (%s)", node.name, ip)
-
-        except errors.AddressPoolError:
-          pass
+    if self.op.conflicts_check:
+      for node in self.cfg.GetAllNodesInfo().values():
+        for ip in [node.primary_ip, node.secondary_ip]:
+          try:
+            if pool.Contains(ip):
+              pool.Reserve(ip)
+              self.LogInfo("Reserved IP address of node '%s' (%s)",
+                           node.name, ip)
+          except errors.AddressPoolError:
+            self.LogWarning("Cannot reserve IP address of node '%s' (%s)",
+                            node.name, ip)
 
-    master_ip = self.cfg.GetClusterInfo().master_ip
-    try:
-      pool.Reserve(master_ip)
-      self.LogInfo("Reserved cluster master IP (%s)", master_ip)
-    except errors.AddressPoolError:
-      pass
+      master_ip = self.cfg.GetClusterInfo().master_ip
+      try:
+        if pool.Contains(master_ip):
+          pool.Reserve(master_ip)
+          self.LogInfo("Reserved cluster master IP address (%s)", master_ip)
+      except errors.AddressPoolError:
+        self.LogWarning("Cannot reserve cluster master IP address (%s)",
+                        master_ip)
 
     if self.op.add_reserved_ips:
       for ip in self.op.add_reserved_ips:
@@ -15543,6 +15838,10 @@ class LUNetworkAdd(LogicalUnit):
         except errors.AddressPoolError, e:
           raise errors.OpExecError("Cannot reserve IP %s. %s " % (ip, e))
 
+    if self.op.tags:
+      for tag in self.op.tags:
+        nobj.AddTag(tag)
+
     self.cfg.AddNetwork(nobj, self.proc.GetECId(), check_uuid=False)
     del self.remove_locks[locking.LEVEL_NETWORK]
 
@@ -15555,11 +15854,17 @@ class LUNetworkRemove(LogicalUnit):
   def ExpandNames(self):
     self.network_uuid = self.cfg.LookupNetwork(self.op.network_name)
 
+    if not self.network_uuid:
+      raise errors.OpPrereqError(("Network '%s' not found" %
+                                  self.op.network_name),
+                                 errors.ECODE_INVAL)
+
+    self.share_locks[locking.LEVEL_NODEGROUP] = 1
     self.needed_locks = {
       locking.LEVEL_NETWORK: [self.network_uuid],
+      locking.LEVEL_NODEGROUP: locking.ALL_SET,
       }
 
-
   def CheckPrereq(self):
     """Check prerequisites.
 
@@ -15568,22 +15873,17 @@ class LUNetworkRemove(LogicalUnit):
     cluster.
 
     """
-    if not self.network_uuid:
-      raise errors.OpPrereqError("Network %s not found" % self.op.network_name,
-                                 errors.ECODE_INVAL)
-
     # Verify that the network is not conncted.
     node_groups = [group.name
                    for group in self.cfg.GetAllNodeGroupsInfo().values()
-                   for network in group.networks.keys()
-                   if network == self.network_uuid]
+                   if self.network_uuid in group.networks]
 
     if node_groups:
-      self.LogWarning("Nework '%s' is connected to the following"
-                      " node groups: %s" % (self.op.network_name,
-                      utils.CommaJoin(utils.NiceSort(node_groups))))
-      raise errors.OpPrereqError("Network still connected",
-                                 errors.ECODE_STATE)
+      self.LogWarning("Network '%s' is connected to the following"
+                      " node groups: %s" %
+                      (self.op.network_name,
+                       utils.CommaJoin(utils.NiceSort(node_groups))))
+      raise errors.OpPrereqError("Network still connected", errors.ECODE_STATE)
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -15625,29 +15925,28 @@ class LUNetworkSetParams(LogicalUnit):
       raise errors.OpPrereqError("Cannot modify gateway and reserved ips"
                                  " at once", errors.ECODE_INVAL)
 
-
   def ExpandNames(self):
     self.network_uuid = self.cfg.LookupNetwork(self.op.network_name)
-    self.network = self.cfg.GetNetwork(self.network_uuid)
+    if self.network_uuid is None:
+      raise errors.OpPrereqError(("Network '%s' not found" %
+                                  self.op.network_name),
+                                 errors.ECODE_INVAL)
+
     self.needed_locks = {
       locking.LEVEL_NETWORK: [self.network_uuid],
       }
 
-
-    if self.network is None:
-      raise errors.OpPrereqError("Could not retrieve network '%s' (UUID: %s)" %
-                                 (self.op.network_name, self.network_uuid),
-                                 errors.ECODE_INVAL)
-
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
+    self.network = self.cfg.GetNetwork(self.network_uuid)
     self.gateway = self.network.gateway
     self.network_type = self.network.network_type
     self.mac_prefix = self.network.mac_prefix
     self.network6 = self.network.network6
     self.gateway6 = self.network.gateway6
+    self.tags = self.network.tags
 
     self.pool = network.AddressPool(self.network)
 
@@ -15670,7 +15969,8 @@ class LUNetworkSetParams(LogicalUnit):
       if self.op.mac_prefix == constants.VALUE_NONE:
         self.mac_prefix = None
       else:
-        self.mac_prefix = self.op.mac_prefix
+        self.mac_prefix = \
+          utils.NormalizeAndValidateThreeOctetMacPrefix(self.op.mac_prefix)
 
     if self.op.gateway6:
       if self.op.gateway6 == constants.VALUE_NONE:
@@ -15684,22 +15984,21 @@ class LUNetworkSetParams(LogicalUnit):
       else:
         self.network6 = self.op.network6
 
-
-
   def BuildHooksEnv(self):
     """Build hooks env.
 
     """
-    env = {
-      "NETWORK_NAME": self.op.network_name,
-      "NETWORK_SUBNET": self.network.network,
-      "NETWORK_GATEWAY": self.gateway,
-      "NETWORK_SUBNET6": self.network6,
-      "NETWORK_GATEWAY6": self.gateway6,
-      "NETWORK_MAC_PREFIX": self.mac_prefix,
-      "NETWORK_TYPE": self.network_type,
+    args = {
+      "name": self.op.network_name,
+      "subnet": self.network.network,
+      "gateway": self.gateway,
+      "network6": self.network6,
+      "gateway6": self.gateway6,
+      "mac_prefix": self.mac_prefix,
+      "network_type": self.network_type,
+      "tags": self.tags,
       }
-    return env
+    return _BuildNetworkHookEnv(**args) # pylint: disable=W0142
 
   def BuildHooksNodes(self):
     """Build hooks nodes.
@@ -15716,7 +16015,7 @@ class LUNetworkSetParams(LogicalUnit):
     #      extend cfg.ReserveIp/ReleaseIp with the external flag
     if self.op.gateway:
       if self.gateway == self.network.gateway:
-        self.LogWarning("Gateway is already %s" % self.gateway)
+        self.LogWarning("Gateway is already %s", self.gateway)
       else:
         if self.gateway:
           self.pool.Reserve(self.gateway, external=True)
@@ -15728,11 +16027,11 @@ class LUNetworkSetParams(LogicalUnit):
       for ip in self.op.add_reserved_ips:
         try:
           if self.pool.IsReserved(ip):
-            self.LogWarning("IP %s is already reserved" % ip)
+            self.LogWarning("IP address %s is already reserved", ip)
           else:
             self.pool.Reserve(ip, external=True)
-        except errors.AddressPoolError, e:
-          self.LogWarning("Cannot reserve ip %s. %s" % (ip, e))
+        except errors.AddressPoolError, err:
+          self.LogWarning("Cannot reserve IP address %s: %s", ip, err)
 
     if self.op.remove_reserved_ips:
       for ip in self.op.remove_reserved_ips:
@@ -15741,11 +16040,11 @@ class LUNetworkSetParams(LogicalUnit):
           continue
         try:
           if not self.pool.IsReserved(ip):
-            self.LogWarning("IP %s is already unreserved" % ip)
+            self.LogWarning("IP address %s is already unreserved", ip)
           else:
             self.pool.Release(ip, external=True)
-        except errors.AddressPoolError, e:
-          self.LogWarning("Cannot release ip %s. %s" % (ip, e))
+        except errors.AddressPoolError, err:
+          self.LogWarning("Cannot release IP address %s: %s", ip, err)
 
     if self.op.mac_prefix:
       self.network.mac_prefix = self.mac_prefix
@@ -15804,7 +16103,6 @@ class _NetworkQuery(_QueryBase):
     do_instances = query.NETQ_INST in self.requested_data
     do_groups = do_instances or (query.NETQ_GROUP in self.requested_data)
     do_stats = query.NETQ_STATS in self.requested_data
-    cluster = lu.cfg.GetClusterInfo()
 
     network_to_groups = None
     network_to_instances = None
@@ -15814,14 +16112,12 @@ class _NetworkQuery(_QueryBase):
     if do_groups:
       all_groups = lu.cfg.GetAllNodeGroupsInfo()
       network_to_groups = dict((uuid, []) for uuid in self.wanted)
-      default_nicpp = cluster.nicparams[constants.PP_DEFAULT]
 
       if do_instances:
         all_instances = lu.cfg.GetAllInstancesInfo()
         all_nodes = lu.cfg.GetAllNodesInfo()
         network_to_instances = dict((uuid, []) for uuid in self.wanted)
 
-
       for group in all_groups.values():
         if do_instances:
           group_nodes = [node.name for node in all_nodes.values() if
@@ -15834,7 +16130,7 @@ class _NetworkQuery(_QueryBase):
             netparams = group.networks[net_uuid]
             mode = netparams[constants.NIC_MODE]
             link = netparams[constants.NIC_LINK]
-            info = group.name + '(' + mode + ', ' + link + ')'
+            info = group.name + "(" + mode + ", " + link + ")"
             network_to_groups[net_uuid].append(info)
 
             if do_instances:
@@ -15853,7 +16149,8 @@ class _NetworkQuery(_QueryBase):
             "free_count": pool.GetFreeCount(),
             "reserved_count": pool.GetReservedCount(),
             "map": pool.GetMap(),
-            "external_reservations": ", ".join(pool.GetExternalReservations()),
+            "external_reservations":
+              utils.CommaJoin(pool.GetExternalReservations()),
             }
 
     return query.NetworkQueryData([self._all_networks[uuid]
@@ -15880,7 +16177,6 @@ class LUNetworkQuery(NoHooksLU):
     return self.nq.OldStyleQuery(self)
 
 
-
 class LUNetworkConnect(LogicalUnit):
   """Connect a network to a nodegroup
 
@@ -15896,9 +16192,14 @@ class LUNetworkConnect(LogicalUnit):
     self.network_link = self.op.network_link
 
     self.network_uuid = self.cfg.LookupNetwork(self.network_name)
-    self.network = self.cfg.GetNetwork(self.network_uuid)
+    if self.network_uuid is None:
+      raise errors.OpPrereqError("Network %s does not exist" %
+                                 self.network_name, errors.ECODE_INVAL)
+
     self.group_uuid = self.cfg.LookupNodeGroup(self.group_name)
-    self.group = self.cfg.GetNodeGroup(self.group_uuid)
+    if self.group_uuid is None:
+      raise errors.OpPrereqError("Group %s does not exist" %
+                                 self.group_name, errors.ECODE_INVAL)
 
     self.needed_locks = {
       locking.LEVEL_INSTANCE: [],
@@ -15906,41 +16207,47 @@ class LUNetworkConnect(LogicalUnit):
       }
     self.share_locks[locking.LEVEL_INSTANCE] = 1
 
+    if self.op.conflicts_check:
+      self.needed_locks[locking.LEVEL_NETWORK] = [self.network_uuid]
+      self.share_locks[locking.LEVEL_NETWORK] = 1
+
   def DeclareLocks(self, level):
     if level == locking.LEVEL_INSTANCE:
       assert not self.needed_locks[locking.LEVEL_INSTANCE]
 
       # Lock instances optimistically, needs verification once group lock has
       # been acquired
-      self.needed_locks[locking.LEVEL_INSTANCE] = \
-          self.cfg.GetNodeGroupInstances(self.group_uuid)
+      if self.op.conflicts_check:
+        self.needed_locks[locking.LEVEL_INSTANCE] = \
+            self.cfg.GetNodeGroupInstances(self.group_uuid)
 
   def BuildHooksEnv(self):
-    ret = dict()
-    ret["GROUP_NAME"] = self.group_name
-    ret["GROUP_NETWORK_NAME"] = self.network_name
-    ret["GROUP_NETWORK_MODE"] = self.network_mode
-    ret["GROUP_NETWORK_LINK"] = self.network_link
+    ret = {
+      "GROUP_NAME": self.group_name,
+      "GROUP_NETWORK_MODE": self.network_mode,
+      "GROUP_NETWORK_LINK": self.network_link,
+      }
     return ret
 
   def BuildHooksNodes(self):
     nodes = self.cfg.GetNodeGroup(self.group_uuid).members
     return (nodes, nodes)
 
-
   def CheckPrereq(self):
-    l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2])
-                                   for i in value)
+    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
 
-    if self.network is None:
-      raise errors.OpPrereqError("Network %s does not exist" %
-                                 self.network_name, errors.ECODE_INVAL)
+    assert self.group_uuid in owned_groups
+
+    l = lambda value: utils.CommaJoin("%s: %s/%s" % (i[0], i[1], i[2])
+                                      for i in value)
 
-    self.netparams = dict()
-    self.netparams[constants.NIC_MODE] = self.network_mode
-    self.netparams[constants.NIC_LINK] = self.network_link
+    self.netparams = {
+      constants.NIC_MODE: self.network_mode,
+      constants.NIC_LINK: self.network_link,
+      }
     objects.NIC.CheckParameterSyntax(self.netparams)
 
+    self.group = self.cfg.GetNodeGroup(self.group_uuid)
     #if self.network_mode == constants.NIC_MODE_BRIDGED:
     #  _CheckNodeGroupBridgesExist(self, self.network_link, self.group_uuid)
     self.connected = False
@@ -15950,20 +16257,25 @@ class LUNetworkConnect(LogicalUnit):
       self.connected = True
       return
 
-    pool = network.AddressPool(self.network)
     if self.op.conflicts_check:
-      groupinstances = []
-      for n in self.cfg.GetNodeGroupInstances(self.group_uuid):
-        groupinstances.append(self.cfg.GetInstanceInfo(n))
-      instances = [(instance.name, idx, nic.ip)
-                   for instance in groupinstances
-                   for idx, nic in enumerate(instance.nics)
-                   if (not nic.network and pool._Contains(nic.ip))]
-      if instances:
+      # Check if locked instances are still correct
+      owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+      _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+      nobj = self.cfg.GetNetwork(self.network_uuid)
+      pool = network.AddressPool(nobj)
+      conflicting_instances = []
+
+      for (_, instance) in self.cfg.GetMultiInstanceInfo(owned_instances):
+        for idx, nic in enumerate(instance.nics):
+          if pool.Contains(nic.ip):
+            conflicting_instances.append((instance.name, idx, nic.ip))
+
+      if conflicting_instances:
         self.LogWarning("Following occurences use IPs from network %s"
                         " that is about to connect to nodegroup %s: %s" %
                         (self.network_name, self.group.name,
-                        l(instances)))
+                        l(conflicting_instances)))
         raise errors.OpPrereqError("Conflicting IPs found."
                                    " Please remove/modify"
                                    " corresponding NICs",
@@ -15990,9 +16302,14 @@ class LUNetworkDisconnect(LogicalUnit):
     self.group_name = self.op.group_name
 
     self.network_uuid = self.cfg.LookupNetwork(self.network_name)
-    self.network = self.cfg.GetNetwork(self.network_uuid)
+    if self.network_uuid is None:
+      raise errors.OpPrereqError("Network %s does not exist" %
+                                 self.network_name, errors.ECODE_INVAL)
+
     self.group_uuid = self.cfg.LookupNodeGroup(self.group_name)
-    self.group = self.cfg.GetNodeGroup(self.group_uuid)
+    if self.group_uuid is None:
+      raise errors.OpPrereqError("Group %s does not exist" %
+                                 self.group_name, errors.ECODE_INVAL)
 
     self.needed_locks = {
       locking.LEVEL_INSTANCE: [],
@@ -16006,46 +16323,54 @@ class LUNetworkDisconnect(LogicalUnit):
 
       # Lock instances optimistically, needs verification once group lock has
       # been acquired
-      self.needed_locks[locking.LEVEL_INSTANCE] = \
+      if self.op.conflicts_check:
+        self.needed_locks[locking.LEVEL_INSTANCE] = \
           self.cfg.GetNodeGroupInstances(self.group_uuid)
 
   def BuildHooksEnv(self):
-    ret = dict()
-    ret["GROUP_NAME"] = self.group_name
-    ret["GROUP_NETWORK_NAME"] = self.network_name
+    ret = {
+      "GROUP_NAME": self.group_name,
+      }
     return ret
 
   def BuildHooksNodes(self):
     nodes = self.cfg.GetNodeGroup(self.group_uuid).members
     return (nodes, nodes)
 
-
   def CheckPrereq(self):
-    l = lambda value: ", ".join("%s: %s/%s" % (i[0], i[1], i[2])
-                                   for i in value)
+    owned_groups = frozenset(self.owned_locks(locking.LEVEL_NODEGROUP))
+
+    assert self.group_uuid in owned_groups
 
+    l = lambda value: utils.CommaJoin("%s: %s/%s" % (i[0], i[1], i[2])
+                                      for i in value)
+
+    self.group = self.cfg.GetNodeGroup(self.group_uuid)
     self.connected = True
     if self.network_uuid not in self.group.networks:
-      self.LogWarning("Network '%s' is"
-                         " not mapped to group '%s'" %
-                         (self.network_name, self.group.name))
+      self.LogWarning("Network '%s' is not mapped to group '%s'",
+                      self.network_name, self.group.name)
       self.connected = False
       return
 
     if self.op.conflicts_check:
-      groupinstances = []
-      for n in self.cfg.GetNodeGroupInstances(self.group_uuid):
-        groupinstances.append(self.cfg.GetInstanceInfo(n))
-      instances = [(instance.name, idx, nic.ip)
-                   for instance in groupinstances
-                   for idx, nic in enumerate(instance.nics)
-                   if nic.network == self.network_name]
-      if instances:
+      # Check if locked instances are still correct
+      owned_instances = frozenset(self.owned_locks(locking.LEVEL_INSTANCE))
+      _CheckNodeGroupInstances(self.cfg, self.group_uuid, owned_instances)
+
+      conflicting_instances = []
+
+      for (_, instance) in self.cfg.GetMultiInstanceInfo(owned_instances):
+        for idx, nic in enumerate(instance.nics):
+          if nic.network == self.network_name:
+            conflicting_instances.append((instance.name, idx, nic.ip))
+
+      if conflicting_instances:
         self.LogWarning("Following occurences use IPs from network %s"
                            " that is about to disconnected from the nodegroup"
                            " %s: %s" %
                            (self.network_name, self.group.name,
-                            l(instances)))
+                            l(conflicting_instances)))
         raise errors.OpPrereqError("Conflicting IPs."
                                    " Please remove/modify"
                                    " corresponding NICS",
@@ -16085,6 +16410,7 @@ def _GetQueryImplementation(name):
     raise errors.OpPrereqError("Unknown query resource '%s'" % name,
                                errors.ECODE_INVAL)
 
+
 def _CheckForConflictingIp(lu, ip, node):
   """In case of conflicting ip raise error.
 
@@ -16094,7 +16420,7 @@ def _CheckForConflictingIp(lu, ip, node):
   @param node: node name
 
   """
-  (conf_net, conf_netparams) = lu.cfg.CheckIPInNodeGroup(ip, node)
+  (conf_net, _) = lu.cfg.CheckIPInNodeGroup(ip, node)
   if conf_net is not None:
     raise errors.OpPrereqError("Conflicting IP found:"
                                " %s <> %s." % (ip, conf_net),