Parallelize Tag operations
[ganeti-local] / lib / cmdlib.py
index b79acb6..6dbaa92 100644 (file)
@@ -80,9 +80,12 @@ class LogicalUnit(object):
     self.cfg = context.cfg
     self.sstore = sstore
     self.context = context
+    # Dicts used to declare locking needs to mcpu
     self.needed_locks = None
     self.acquired_locks = {}
     self.share_locks = dict(((i, 0) for i in locking.LEVELS))
+    self.add_locks = {}
+    self.remove_locks = {}
     # Used to force good behavior when calling helper functions
     self.recalculate_locks = {}
     self.__ssh = None
@@ -136,7 +139,7 @@ class LogicalUnit(object):
     # Acquire all nodes and one instance
     self.needed_locks = {
       locking.LEVEL_NODE: locking.ALL_SET,
-      locking.LEVEL_INSTANCES: ['instance1.example.tld'],
+      locking.LEVEL_INSTANCE: ['instance1.example.tld'],
     }
     # Acquire just two nodes
     self.needed_locks = {
@@ -298,7 +301,11 @@ class LogicalUnit(object):
       wanted_nodes.append(instance.primary_node)
       if not primary_only:
         wanted_nodes.extend(instance.secondary_nodes)
-    self.needed_locks[locking.LEVEL_NODE] = wanted_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)
 
     del self.recalculate_locks[locking.LEVEL_NODE]
 
@@ -494,6 +501,14 @@ class LUVerifyCluster(LogicalUnit):
   HPATH = "cluster-verify"
   HTYPE = constants.HTYPE_CLUSTER
   _OP_REQP = ["skip_checks"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.needed_locks = {
+      locking.LEVEL_NODE: locking.ALL_SET,
+      locking.LEVEL_INSTANCE: locking.ALL_SET,
+    }
+    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
 
   def _VerifyNode(self, node, file_list, local_cksum, vglist, node_result,
                   remote_version, feedback_fn):
@@ -908,6 +923,14 @@ class LUVerifyDisks(NoHooksLU):
 
   """
   _OP_REQP = []
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.needed_locks = {
+      locking.LEVEL_NODE: locking.ALL_SET,
+      locking.LEVEL_INSTANCE: locking.ALL_SET,
+    }
+    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -1078,6 +1101,15 @@ class LUSetClusterParams(LogicalUnit):
   HPATH = "cluster-modify"
   HTYPE = constants.HTYPE_CLUSTER
   _OP_REQP = []
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    # FIXME: in the future maybe other cluster params won't require checking on
+    # all nodes to be modified.
+    self.needed_locks = {
+      locking.LEVEL_NODE: locking.ALL_SET,
+    }
+    self.share_locks[locking.LEVEL_NODE] = 1
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -1097,9 +1129,10 @@ class LUSetClusterParams(LogicalUnit):
     if the given volume group is valid.
 
     """
+    # FIXME: This only works because there is only one parameter that can be
+    # changed or removed.
     if not self.op.vg_name:
-      instances = [self.cfg.GetInstanceInfo(name)
-                   for name in self.cfg.GetInstanceList()]
+      instances = self.cfg.GetAllInstancesInfo().values()
       for inst in instances:
         for disk in inst.disks:
           if _RecursiveCheckIfLVMBased(disk):
@@ -1108,7 +1141,7 @@ class LUSetClusterParams(LogicalUnit):
 
     # if vg_name not None, checks given volume group on all nodes
     if self.op.vg_name:
-      node_list = self.cfg.GetNodeList()
+      node_list = self.acquired_locks[locking.LEVEL_NODE]
       vglist = rpc.call_vg_list(node_list)
       for node in node_list:
         vgstatus = utils.CheckVolumeGroupSize(vglist[node], self.op.vg_name,
@@ -1385,36 +1418,48 @@ class LUQueryNodes(NoHooksLU):
       "ctotal",
       ])
 
-    _CheckOutputFields(static=["name", "pinst_cnt", "sinst_cnt",
-                               "pinst_list", "sinst_list",
-                               "pip", "sip", "tags"],
+    self.static_fields = frozenset([
+      "name", "pinst_cnt", "sinst_cnt",
+      "pinst_list", "sinst_list",
+      "pip", "sip", "tags",
+      ])
+
+    _CheckOutputFields(static=self.static_fields,
                        dynamic=self.dynamic_fields,
                        selected=self.op.output_fields)
 
     self.needed_locks = {}
     self.share_locks[locking.LEVEL_NODE] = 1
-    # TODO: we could lock nodes only if the user asked for dynamic fields. For
-    # that we need atomic ways to get info for a group of nodes from the
-    # config, though.
-    if not self.op.names:
-      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+    if self.op.names:
+      self.wanted = _GetWantedNodes(self, self.op.names)
     else:
-      self.needed_locks[locking.LEVEL_NODE] = \
-        _GetWantedNodes(self, self.op.names)
+      self.wanted = locking.ALL_SET
+
+    self.do_locking = not self.static_fields.issuperset(self.op.output_fields)
+    if self.do_locking:
+      # if we don't request only static fields, we need to lock the nodes
+      self.needed_locks[locking.LEVEL_NODE] = self.wanted
+
 
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
-    # This of course is valid only if we locked the nodes
-    self.wanted = self.acquired_locks[locking.LEVEL_NODE]
+    # The validation of the node list is done in the _GetWantedNodes,
+    # if non empty, and if empty, there's no validation to do
+    pass
 
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
 
     """
-    nodenames = self.wanted
-    nodelist = [self.cfg.GetNodeInfo(name) for name in nodenames]
+    all_info = self.cfg.GetAllNodesInfo()
+    if self.do_locking:
+      nodenames = self.acquired_locks[locking.LEVEL_NODE]
+    else:
+      nodenames = all_info.keys()
+    nodelist = [all_info[name] for name in nodenames]
 
     # begin data gathering
 
@@ -1832,6 +1877,16 @@ class LUActivateInstanceDisks(NoHooksLU):
 
   """
   _OP_REQP = ["instance_name"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self._ExpandAndLockInstance()
+    self.needed_locks[locking.LEVEL_NODE] = []
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      self._LockInstancesNodes()
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -1839,13 +1894,9 @@ class LUActivateInstanceDisks(NoHooksLU):
     This checks that the instance is in the cluster.
 
     """
-    instance = self.cfg.GetInstanceInfo(
-      self.cfg.ExpandInstanceName(self.op.instance_name))
-    if instance is None:
-      raise errors.OpPrereqError("Instance '%s' not known" %
-                                 self.op.instance_name)
-    self.instance = instance
-
+    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+    assert self.instance is not None, \
+      "Cannot retrieve locked instance %s" % self.op.instance_name
 
   def Exec(self, feedback_fn):
     """Activate the disks.
@@ -1939,6 +1990,16 @@ class LUDeactivateInstanceDisks(NoHooksLU):
 
   """
   _OP_REQP = ["instance_name"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self._ExpandAndLockInstance()
+    self.needed_locks[locking.LEVEL_NODE] = []
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      self._LockInstancesNodes()
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -1946,29 +2007,36 @@ class LUDeactivateInstanceDisks(NoHooksLU):
     This checks that the instance is in the cluster.
 
     """
-    instance = self.cfg.GetInstanceInfo(
-      self.cfg.ExpandInstanceName(self.op.instance_name))
-    if instance is None:
-      raise errors.OpPrereqError("Instance '%s' not known" %
-                                 self.op.instance_name)
-    self.instance = instance
+    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+    assert self.instance is not None, \
+      "Cannot retrieve locked instance %s" % self.op.instance_name
 
   def Exec(self, feedback_fn):
     """Deactivate the disks
 
     """
     instance = self.instance
-    ins_l = rpc.call_instance_list([instance.primary_node])
-    ins_l = ins_l[instance.primary_node]
-    if not type(ins_l) is list:
-      raise errors.OpExecError("Can't contact node '%s'" %
-                               instance.primary_node)
+    _SafeShutdownInstanceDisks(instance, self.cfg)
 
-    if self.instance.name in ins_l:
-      raise errors.OpExecError("Instance is running, can't shutdown"
-                               " block devices.")
 
-    _ShutdownInstanceDisks(instance, self.cfg)
+def _SafeShutdownInstanceDisks(instance, cfg):
+  """Shutdown block devices of an instance.
+
+  This function checks if an instance is running, before calling
+  _ShutdownInstanceDisks.
+
+  """
+  ins_l = rpc.call_instance_list([instance.primary_node])
+  ins_l = ins_l[instance.primary_node]
+  if not type(ins_l) is list:
+    raise errors.OpExecError("Can't contact node '%s'" %
+                             instance.primary_node)
+
+  if instance.name in ins_l:
+    raise errors.OpExecError("Instance is running, can't shutdown"
+                             " block devices.")
+
+  _ShutdownInstanceDisks(instance, cfg)
 
 
 def _ShutdownInstanceDisks(instance, cfg, ignore_primary=False):
@@ -2034,7 +2102,7 @@ class LUStartupInstance(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -2109,7 +2177,7 @@ class LURebootInstance(LogicalUnit):
                                    constants.INSTANCE_REBOOT_FULL))
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -2183,7 +2251,7 @@ class LUShutdownInstance(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -2235,7 +2303,7 @@ class LUReinstallInstance(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -2423,6 +2491,16 @@ class LURemoveInstance(LogicalUnit):
   HPATH = "instance-remove"
   HTYPE = constants.HTYPE_INSTANCE
   _OP_REQP = ["instance_name", "ignore_failures"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self._ExpandAndLockInstance()
+    self.needed_locks[locking.LEVEL_NODE] = []
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      self._LockInstancesNodes()
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -2440,12 +2518,9 @@ class LURemoveInstance(LogicalUnit):
     This checks that the instance is in the cluster.
 
     """
-    instance = self.cfg.GetInstanceInfo(
-      self.cfg.ExpandInstanceName(self.op.instance_name))
-    if instance is None:
-      raise errors.OpPrereqError("Instance '%s' not known" %
-                                 self.op.instance_name)
-    self.instance = instance
+    self.instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+    assert self.instance is not None, \
+      "Cannot retrieve locked instance %s" % self.op.instance_name
 
   def Exec(self, feedback_fn):
     """Remove the instance.
@@ -2473,8 +2548,7 @@ class LURemoveInstance(LogicalUnit):
     logger.Info("removing instance %s out of cluster config" % instance.name)
 
     self.cfg.RemoveInstance(instance.name)
-    # Remove the new instance from the Ganeti Lock Manager
-    self.context.glm.remove(locking.LEVEL_INSTANCE, instance.name)
+    self.remove_locks[locking.LEVEL_INSTANCE] = instance.name
 
 
 class LUQueryInstances(NoHooksLU):
@@ -2486,15 +2560,18 @@ class LUQueryInstances(NoHooksLU):
 
   def ExpandNames(self):
     self.dynamic_fields = frozenset(["oper_state", "oper_ram", "status"])
-    _CheckOutputFields(static=["name", "os", "pnode", "snodes",
-                               "admin_state", "admin_ram",
-                               "disk_template", "ip", "mac", "bridge",
-                               "sda_size", "sdb_size", "vcpus", "tags",
-                               "auto_balance",
-                               "network_port", "kernel_path", "initrd_path",
-                               "hvm_boot_order", "hvm_acpi", "hvm_pae",
-                               "hvm_cdrom_image_path", "hvm_nic_type",
-                               "hvm_disk_type", "vnc_bind_address"],
+    self.static_fields = frozenset([
+      "name", "os", "pnode", "snodes",
+      "admin_state", "admin_ram",
+      "disk_template", "ip", "mac", "bridge",
+      "sda_size", "sdb_size", "vcpus", "tags",
+      "auto_balance",
+      "network_port", "kernel_path", "initrd_path",
+      "hvm_boot_order", "hvm_acpi", "hvm_pae",
+      "hvm_cdrom_image_path", "hvm_nic_type",
+      "hvm_disk_type", "vnc_bind_address",
+      ])
+    _CheckOutputFields(static=self.static_fields,
                        dynamic=self.dynamic_fields,
                        selected=self.op.output_fields)
 
@@ -2502,37 +2579,37 @@ class LUQueryInstances(NoHooksLU):
     self.share_locks[locking.LEVEL_INSTANCE] = 1
     self.share_locks[locking.LEVEL_NODE] = 1
 
-    # TODO: we could lock instances (and nodes) only if the user asked for
-    # dynamic fields. For that we need atomic ways to get info for a group of
-    # instances from the config, though.
-    if not self.op.names:
-      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+    if self.op.names:
+      self.wanted = _GetWantedInstances(self, self.op.names)
     else:
-      self.needed_locks[locking.LEVEL_INSTANCE] = \
-        _GetWantedInstances(self, self.op.names)
+      self.wanted = locking.ALL_SET
 
-    self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.do_locking = not self.static_fields.issuperset(self.op.output_fields)
+    if self.do_locking:
+      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted
+      self.needed_locks[locking.LEVEL_NODE] = []
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
-    # TODO: locking of nodes could be avoided when not querying them
-    if level == locking.LEVEL_NODE:
+    if level == locking.LEVEL_NODE and self.do_locking:
       self._LockInstancesNodes()
 
   def CheckPrereq(self):
     """Check prerequisites.
 
     """
-    # This of course is valid only if we locked the instances
-    self.wanted = self.acquired_locks[locking.LEVEL_INSTANCE]
+    pass
 
   def Exec(self, feedback_fn):
     """Computes the list of nodes and their attributes.
 
     """
-    instance_names = self.wanted
-    instance_list = [self.cfg.GetInstanceInfo(iname) for iname
-                     in instance_names]
+    all_info = self.cfg.GetAllInstancesInfo()
+    if self.do_locking:
+      instance_names = self.acquired_locks[locking.LEVEL_INSTANCE]
+    else:
+      instance_names = all_info.keys()
+    instance_list = [all_info[iname] for iname in instance_names]
 
     # begin data gathering
 
@@ -2647,7 +2724,7 @@ class LUFailoverInstance(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -3010,6 +3087,125 @@ class LUCreateInstance(LogicalUnit):
   _OP_REQP = ["instance_name", "mem_size", "disk_size",
               "disk_template", "swap_size", "mode", "start", "vcpus",
               "wait_for_sync", "ip_check", "mac"]
+  REQ_BGL = False
+
+  def _ExpandNode(self, node):
+    """Expands and checks one node name.
+
+    """
+    node_full = self.cfg.ExpandNodeName(node)
+    if node_full is None:
+      raise errors.OpPrereqError("Unknown node %s" % node)
+    return node_full
+
+  def ExpandNames(self):
+    """ExpandNames for CreateInstance.
+
+    Figure out the right locks for instance creation.
+
+    """
+    self.needed_locks = {}
+
+    # set optional parameters to none if they don't exist
+    for attr in ["kernel_path", "initrd_path", "pnode", "snode",
+                 "iallocator", "hvm_boot_order", "hvm_acpi", "hvm_pae",
+                 "hvm_cdrom_image_path", "hvm_nic_type", "hvm_disk_type",
+                 "vnc_bind_address"]:
+      if not hasattr(self.op, attr):
+        setattr(self.op, attr, None)
+
+    # verify creation mode
+    if self.op.mode not in (constants.INSTANCE_CREATE,
+                            constants.INSTANCE_IMPORT):
+      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
+                                 self.op.mode)
+    # disk template and mirror node verification
+    if self.op.disk_template not in constants.DISK_TEMPLATES:
+      raise errors.OpPrereqError("Invalid disk template name")
+
+    #### instance parameters check
+
+    # instance name verification
+    hostname1 = utils.HostInfo(self.op.instance_name)
+    self.op.instance_name = instance_name = hostname1.name
+
+    # this is just a preventive check, but someone might still add this
+    # instance in the meantime, and creation will fail at lock-add time
+    if instance_name in self.cfg.GetInstanceList():
+      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
+                                 instance_name)
+
+    self.add_locks[locking.LEVEL_INSTANCE] = instance_name
+
+    # ip validity checks
+    ip = getattr(self.op, "ip", None)
+    if ip is None or ip.lower() == "none":
+      inst_ip = None
+    elif ip.lower() == "auto":
+      inst_ip = hostname1.ip
+    else:
+      if not utils.IsValidIP(ip):
+        raise errors.OpPrereqError("given IP address '%s' doesn't look"
+                                   " like a valid IP" % ip)
+      inst_ip = ip
+    self.inst_ip = self.op.ip = inst_ip
+    # used in CheckPrereq for ip ping check
+    self.check_ip = hostname1.ip
+
+    # MAC address verification
+    if self.op.mac != "auto":
+      if not utils.IsValidMac(self.op.mac.lower()):
+        raise errors.OpPrereqError("invalid MAC address specified: %s" %
+                                   self.op.mac)
+
+    # boot order verification
+    if self.op.hvm_boot_order is not None:
+      if len(self.op.hvm_boot_order.strip("acdn")) != 0:
+        raise errors.OpPrereqError("invalid boot order specified,"
+                                   " must be one or more of [acdn]")
+    # file storage checks
+    if (self.op.file_driver and
+        not self.op.file_driver in constants.FILE_DRIVER):
+      raise errors.OpPrereqError("Invalid file driver name '%s'" %
+                                 self.op.file_driver)
+
+    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
+      raise errors.OpPrereqError("File storage directory path not absolute")
+
+    ### Node/iallocator related checks
+    if [self.op.iallocator, self.op.pnode].count(None) != 1:
+      raise errors.OpPrereqError("One and only one of iallocator and primary"
+                                 " node must be given")
+
+    if self.op.iallocator:
+      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+    else:
+      self.op.pnode = self._ExpandNode(self.op.pnode)
+      nodelist = [self.op.pnode]
+      if self.op.snode is not None:
+        self.op.snode = self._ExpandNode(self.op.snode)
+        nodelist.append(self.op.snode)
+      self.needed_locks[locking.LEVEL_NODE] = nodelist
+
+    # in case of import lock the source node too
+    if self.op.mode == constants.INSTANCE_IMPORT:
+      src_node = getattr(self.op, "src_node", None)
+      src_path = getattr(self.op, "src_path", None)
+
+      if src_node is None or src_path is None:
+        raise errors.OpPrereqError("Importing an instance requires source"
+                                   " node and path options")
+
+      if not os.path.isabs(src_path):
+        raise errors.OpPrereqError("The source path must be absolute")
+
+      self.op.src_node = src_node = self._ExpandNode(src_node)
+      if self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET:
+        self.needed_locks[locking.LEVEL_NODE].append(src_node)
+
+    else: # INSTANCE_CREATE
+      if getattr(self.op, "os_type", None) is None:
+        raise errors.OpPrereqError("No guest OS specified")
 
   def _RunAllocator(self):
     """Run the allocator based on input opcode.
@@ -3085,36 +3281,14 @@ class LUCreateInstance(LogicalUnit):
     """Check prerequisites.
 
     """
-    # set optional parameters to none if they don't exist
-    for attr in ["kernel_path", "initrd_path", "hvm_boot_order", "pnode",
-                 "iallocator", "hvm_acpi", "hvm_pae", "hvm_cdrom_image_path",
-                 "hvm_nic_type", "hvm_disk_type", "vnc_bind_address"]:
-      if not hasattr(self.op, attr):
-        setattr(self.op, attr, None)
-
-    if self.op.mode not in (constants.INSTANCE_CREATE,
-                            constants.INSTANCE_IMPORT):
-      raise errors.OpPrereqError("Invalid instance creation mode '%s'" %
-                                 self.op.mode)
-
     if (not self.cfg.GetVGName() and
         self.op.disk_template not in constants.DTS_NOT_LVM):
       raise errors.OpPrereqError("Cluster does not support lvm-based"
                                  " instances")
 
     if self.op.mode == constants.INSTANCE_IMPORT:
-      src_node = getattr(self.op, "src_node", None)
-      src_path = getattr(self.op, "src_path", None)
-      if src_node is None or src_path is None:
-        raise errors.OpPrereqError("Importing an instance requires source"
-                                   " node and path options")
-      src_node_full = self.cfg.ExpandNodeName(src_node)
-      if src_node_full is None:
-        raise errors.OpPrereqError("Unknown source node '%s'" % src_node)
-      self.op.src_node = src_node = src_node_full
-
-      if not os.path.isabs(src_path):
-        raise errors.OpPrereqError("The source path must be absolute")
+      src_node = self.op.src_node
+      src_path = self.op.src_path
 
       export_info = rpc.call_export_info(src_node, src_path)
 
@@ -3138,52 +3312,17 @@ class LUCreateInstance(LogicalUnit):
       diskimage = os.path.join(src_path, export_info.get(constants.INISECT_INS,
                                                          'disk0_dump'))
       self.src_image = diskimage
-    else: # INSTANCE_CREATE
-      if getattr(self.op, "os_type", None) is None:
-        raise errors.OpPrereqError("No guest OS specified")
-
-    #### instance parameters check
-
-    # disk template and mirror node verification
-    if self.op.disk_template not in constants.DISK_TEMPLATES:
-      raise errors.OpPrereqError("Invalid disk template name")
-
-    # instance name verification
-    hostname1 = utils.HostInfo(self.op.instance_name)
 
-    self.op.instance_name = instance_name = hostname1.name
-    instance_list = self.cfg.GetInstanceList()
-    if instance_name in instance_list:
-      raise errors.OpPrereqError("Instance '%s' is already in the cluster" %
-                                 instance_name)
-
-    # ip validity checks
-    ip = getattr(self.op, "ip", None)
-    if ip is None or ip.lower() == "none":
-      inst_ip = None
-    elif ip.lower() == "auto":
-      inst_ip = hostname1.ip
-    else:
-      if not utils.IsValidIP(ip):
-        raise errors.OpPrereqError("given IP address '%s' doesn't look"
-                                   " like a valid IP" % ip)
-      inst_ip = ip
-    self.inst_ip = self.op.ip = inst_ip
+    # ip ping checks (we use the same ip that was resolved in ExpandNames)
 
     if self.op.start and not self.op.ip_check:
       raise errors.OpPrereqError("Cannot ignore IP address conflicts when"
                                  " adding an instance in start mode")
 
     if self.op.ip_check:
-      if utils.TcpPing(hostname1.ip, constants.DEFAULT_NODED_PORT):
+      if utils.TcpPing(self.check_ip, constants.DEFAULT_NODED_PORT):
         raise errors.OpPrereqError("IP %s of instance %s already in use" %
-                                   (hostname1.ip, instance_name))
-
-    # MAC address verification
-    if self.op.mac != "auto":
-      if not utils.IsValidMac(self.op.mac.lower()):
-        raise errors.OpPrereqError("invalid MAC address specified: %s" %
-                                   self.op.mac)
+                                   (self.check_ip, instance_name))
 
     # bridge verification
     bridge = getattr(self.op, "bridge", None)
@@ -3192,54 +3331,28 @@ class LUCreateInstance(LogicalUnit):
     else:
       self.op.bridge = bridge
 
-    # boot order verification
-    if self.op.hvm_boot_order is not None:
-      if len(self.op.hvm_boot_order.strip("acdn")) != 0:
-        raise errors.OpPrereqError("invalid boot order specified,"
-                                   " must be one or more of [acdn]")
-    # file storage checks
-    if (self.op.file_driver and
-        not self.op.file_driver in constants.FILE_DRIVER):
-      raise errors.OpPrereqError("Invalid file driver name '%s'" %
-                                 self.op.file_driver)
-
-    if self.op.file_storage_dir and os.path.isabs(self.op.file_storage_dir):
-      raise errors.OpPrereqError("File storage directory not a relative"
-                                 " path")
     #### allocator run
 
-    if [self.op.iallocator, self.op.pnode].count(None) != 1:
-      raise errors.OpPrereqError("One and only one of iallocator and primary"
-                                 " node must be given")
-
     if self.op.iallocator is not None:
       self._RunAllocator()
 
     #### node related checks
 
     # check primary node
-    pnode = self.cfg.GetNodeInfo(self.cfg.ExpandNodeName(self.op.pnode))
-    if pnode is None:
-      raise errors.OpPrereqError("Primary node '%s' is unknown" %
-                                 self.op.pnode)
-    self.op.pnode = pnode.name
-    self.pnode = pnode
+    self.pnode = pnode = self.cfg.GetNodeInfo(self.op.pnode)
+    assert self.pnode is not None, \
+      "Cannot retrieve locked node %s" % self.op.pnode
     self.secondaries = []
 
     # mirror node verification
     if self.op.disk_template in constants.DTS_NET_MIRROR:
-      if getattr(self.op, "snode", None) is None:
+      if self.op.snode is None:
         raise errors.OpPrereqError("The networked disk templates need"
                                    " a mirror node")
-
-      snode_name = self.cfg.ExpandNodeName(self.op.snode)
-      if snode_name is None:
-        raise errors.OpPrereqError("Unknown secondary node '%s'" %
-                                   self.op.snode)
-      elif snode_name == pnode.name:
+      if self.op.snode == pnode.name:
         raise errors.OpPrereqError("The secondary node cannot be"
                                    " the primary node.")
-      self.secondaries.append(snode_name)
+      self.secondaries.append(self.op.snode)
 
     req_size = _ComputeDiskSize(self.op.disk_template,
                                 self.op.disk_size, self.op.swap_size)
@@ -3271,7 +3384,6 @@ class LUCreateInstance(LogicalUnit):
     if self.op.kernel_path == constants.VALUE_NONE:
       raise errors.OpPrereqError("Can't set instance kernel to none")
 
-
     # bridge check on primary node
     if not rpc.call_bridges_exist(self.pnode.name, [self.op.bridge]):
       raise errors.OpPrereqError("target bridge '%s' does not exist on"
@@ -3286,6 +3398,7 @@ class LUCreateInstance(LogicalUnit):
 
     # hvm_cdrom_image_path verification
     if self.op.hvm_cdrom_image_path is not None:
+      # FIXME (als): shouldn't these checks happen on the destination node?
       if not os.path.isabs(self.op.hvm_cdrom_image_path):
         raise errors.OpPrereqError("The path to the HVM CDROM image must"
                                    " be an absolute path or None, not %s" %
@@ -3389,8 +3502,9 @@ class LUCreateInstance(LogicalUnit):
     feedback_fn("adding instance %s to cluster config" % instance)
 
     self.cfg.AddInstance(iobj)
-    # Add the new instance to the Ganeti Lock Manager
-    self.context.glm.add(locking.LEVEL_INSTANCE, instance)
+    # Declare that we don't want to remove the instance lock anymore, as we've
+    # added the instance to the config
+    del self.remove_locks[locking.LEVEL_INSTANCE]
 
     if self.op.wait_for_sync:
       disk_abort = not _WaitForSync(self.cfg, iobj, self.proc)
@@ -3405,8 +3519,8 @@ class LUCreateInstance(LogicalUnit):
     if disk_abort:
       _RemoveDisks(iobj, self.cfg)
       self.cfg.RemoveInstance(iobj.name)
-      # Remove the new instance from the Ganeti Lock Manager
-      self.context.glm.remove(locking.LEVEL_INSTANCE, iobj.name)
+      # Make sure the instance lock gets removed
+      self.remove_locks[locking.LEVEL_INSTANCE] = iobj.name
       raise errors.OpExecError("There are some degraded disks for"
                                " this instance")
 
@@ -3496,6 +3610,38 @@ class LUReplaceDisks(LogicalUnit):
   HPATH = "mirrors-replace"
   HTYPE = constants.HTYPE_INSTANCE
   _OP_REQP = ["instance_name", "mode", "disks"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self._ExpandAndLockInstance()
+
+    if not hasattr(self.op, "remote_node"):
+      self.op.remote_node = None
+
+    ia_name = getattr(self.op, "iallocator", None)
+    if ia_name is not None:
+      if self.op.remote_node is not None:
+        raise errors.OpPrereqError("Give either the iallocator or the new"
+                                   " secondary, not both")
+      self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+    elif self.op.remote_node is not None:
+      remote_node = self.cfg.ExpandNodeName(self.op.remote_node)
+      if remote_node is None:
+        raise errors.OpPrereqError("Node '%s' not known" %
+                                   self.op.remote_node)
+      self.op.remote_node = remote_node
+      self.needed_locks[locking.LEVEL_NODE] = [remote_node]
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_APPEND
+    else:
+      self.needed_locks[locking.LEVEL_NODE] = []
+      self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  def DeclareLocks(self, level):
+    # If we're not already locking all nodes in the set we have to declare the
+    # instance's primary/secondary nodes.
+    if (level == locking.LEVEL_NODE and
+        self.needed_locks[locking.LEVEL_NODE] is not locking.ALL_SET):
+      self._LockInstancesNodes()
 
   def _RunAllocator(self):
     """Compute a new secondary node using an IAllocator.
@@ -3546,16 +3692,10 @@ class LUReplaceDisks(LogicalUnit):
     This checks that the instance is in the cluster.
 
     """
-    if not hasattr(self.op, "remote_node"):
-      self.op.remote_node = None
-
-    instance = self.cfg.GetInstanceInfo(
-      self.cfg.ExpandInstanceName(self.op.instance_name))
-    if instance is None:
-      raise errors.OpPrereqError("Instance '%s' not known" %
-                                 self.op.instance_name)
+    instance = self.cfg.GetInstanceInfo(self.op.instance_name)
+    assert instance is not None, \
+      "Cannot retrieve locked instance %s" % self.op.instance_name
     self.instance = instance
-    self.op.instance_name = instance.name
 
     if instance.disk_template not in constants.DTS_NET_MIRROR:
       raise errors.OpPrereqError("Instance's disk layout is not"
@@ -3570,18 +3710,13 @@ class LUReplaceDisks(LogicalUnit):
 
     ia_name = getattr(self.op, "iallocator", None)
     if ia_name is not None:
-      if self.op.remote_node is not None:
-        raise errors.OpPrereqError("Give either the iallocator or the new"
-                                   " secondary, not both")
-      self.op.remote_node = self._RunAllocator()
+      self._RunAllocator()
 
     remote_node = self.op.remote_node
     if remote_node is not None:
-      remote_node = self.cfg.ExpandNodeName(remote_node)
-      if remote_node is None:
-        raise errors.OpPrereqError("Node '%s' not known" %
-                                   self.op.remote_node)
       self.remote_node_info = self.cfg.GetNodeInfo(remote_node)
+      assert self.remote_node_info is not None, \
+        "Cannot retrieve locked node %s" % remote_node
     else:
       self.remote_node_info = None
     if remote_node == instance.primary_node:
@@ -3622,7 +3757,6 @@ class LUReplaceDisks(LogicalUnit):
       if instance.FindDisk(name) is None:
         raise errors.OpPrereqError("Disk '%s' not found for instance '%s'" %
                                    (name, instance.name))
-    self.op.remote_node = remote_node
 
   def _ExecD8DiskOnly(self, feedback_fn):
     """Replace a disk on the primary or secondary for dbrd8.
@@ -3968,8 +4102,7 @@ class LUReplaceDisks(LogicalUnit):
 
     # Activate the instance disks if we're replacing them on a down instance
     if instance.status == "down":
-      op = opcodes.OpActivateInstanceDisks(instance_name=instance.name)
-      self.proc.ChainOpCode(op)
+      _StartInstanceDisks(self.cfg, instance, True)
 
     if instance.disk_template == constants.DT_DRBD8:
       if self.op.remote_node is None:
@@ -3983,8 +4116,7 @@ class LUReplaceDisks(LogicalUnit):
 
     # Deactivate the instance disks if we're replacing them on a down instance
     if instance.status == "down":
-      op = opcodes.OpDeactivateInstanceDisks(instance_name=instance.name)
-      self.proc.ChainOpCode(op)
+      _SafeShutdownInstanceDisks(instance, self.cfg)
 
     return ret
 
@@ -4001,7 +4133,7 @@ class LUGrowDisk(LogicalUnit):
   def ExpandNames(self):
     self._ExpandAndLockInstance()
     self.needed_locks[locking.LEVEL_NODE] = []
-    self.recalculate_locks[locking.LEVEL_NODE] = 'replace'
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
 
   def DeclareLocks(self, level):
     if level == locking.LEVEL_NODE:
@@ -4069,7 +4201,7 @@ class LUGrowDisk(LogicalUnit):
     for node in (instance.secondary_nodes + (instance.primary_node,)):
       self.cfg.SetDiskID(disk, node)
       result = rpc.call_blockdev_grow(node, disk, self.op.amount)
-      if not result or not isinstance(result, tuple) or len(result) != 2:
+      if not result or not isinstance(result, (list, tuple)) or len(result) != 2:
         raise errors.OpExecError("grow request failed to node %s" % node)
       elif not result[0]:
         raise errors.OpExecError("grow request failed to node %s: %s" %
@@ -4084,6 +4216,33 @@ class LUQueryInstanceData(NoHooksLU):
 
   """
   _OP_REQP = ["instances"]
+  REQ_BGL = False
+  def ExpandNames(self):
+    self.needed_locks = {}
+    self.share_locks = dict(((i, 1) for i in locking.LEVELS))
+
+    if not isinstance(self.op.instances, list):
+      raise errors.OpPrereqError("Invalid argument type 'instances'")
+
+    if self.op.instances:
+      self.wanted_names = []
+      for name in self.op.instances:
+        full_name = self.cfg.ExpandInstanceName(name)
+        if full_name is None:
+          raise errors.OpPrereqError("Instance '%s' not known" %
+                                     self.op.instance_name)
+        self.wanted_names.append(full_name)
+      self.needed_locks[locking.LEVEL_INSTANCE] = self.wanted_names
+    else:
+      self.wanted_names = None
+      self.needed_locks[locking.LEVEL_INSTANCE] = locking.ALL_SET
+
+    self.needed_locks[locking.LEVEL_NODE] = []
+    self.recalculate_locks[locking.LEVEL_NODE] = constants.LOCKS_REPLACE
+
+  def DeclareLocks(self, level):
+    if level == locking.LEVEL_NODE:
+      self._LockInstancesNodes()
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -4091,21 +4250,12 @@ class LUQueryInstanceData(NoHooksLU):
     This only checks the optional instance list against the existing names.
 
     """
-    if not isinstance(self.op.instances, list):
-      raise errors.OpPrereqError("Invalid argument type 'instances'")
-    if self.op.instances:
-      self.wanted_instances = []
-      names = self.op.instances
-      for name in names:
-        instance = self.cfg.GetInstanceInfo(self.cfg.ExpandInstanceName(name))
-        if instance is None:
-          raise errors.OpPrereqError("No such instance name '%s'" % name)
-        self.wanted_instances.append(instance)
-    else:
-      self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
-                               in self.cfg.GetInstanceList()]
-    return
+    if self.wanted_names is None:
+      self.wanted_names = self.acquired_locks[locking.LEVEL_INSTANCE]
 
+    self.wanted_instances = [self.cfg.GetInstanceInfo(name) for name
+                             in self.wanted_names]
+    return
 
   def _ComputeDiskStatus(self, instance, snode, dev):
     """Compute block device status.
@@ -4519,6 +4669,23 @@ class LUExportInstance(LogicalUnit):
   HPATH = "instance-export"
   HTYPE = constants.HTYPE_INSTANCE
   _OP_REQP = ["instance_name", "target_node", "shutdown"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self._ExpandAndLockInstance()
+    # FIXME: lock only instance primary and destination node
+    #
+    # Sad but true, for now we have do lock all nodes, as we don't know where
+    # the previous export might be, and and in this LU we search for it and
+    # remove it from its current node. In the future we could fix this by:
+    #  - making a tasklet to search (share-lock all), then create the new one,
+    #    then one to remove, after
+    #  - removing the removal operation altoghether
+    self.needed_locks[locking.LEVEL_NODE] = locking.ALL_SET
+
+  def DeclareLocks(self, level):
+    """Last minute lock declaration."""
+    # All nodes are locked anyway, so nothing to do here.
 
   def BuildHooksEnv(self):
     """Build hooks env.
@@ -4541,20 +4708,16 @@ class LUExportInstance(LogicalUnit):
     This checks that the instance and node names are valid.
 
     """
-    instance_name = self.cfg.ExpandInstanceName(self.op.instance_name)
+    instance_name = self.op.instance_name
     self.instance = self.cfg.GetInstanceInfo(instance_name)
-    if self.instance is None:
-      raise errors.OpPrereqError("Instance '%s' not found" %
-                                 self.op.instance_name)
+    assert self.instance is not None, \
+          "Cannot retrieve locked instance %s" % self.op.instance_name
 
-    # node verification
-    dst_node_short = self.cfg.ExpandNodeName(self.op.target_node)
-    self.dst_node = self.cfg.GetNodeInfo(dst_node_short)
+    self.dst_node = self.cfg.GetNodeInfo(
+      self.cfg.ExpandNodeName(self.op.target_node))
 
-    if self.dst_node is None:
-      raise errors.OpPrereqError("Destination node '%s' is unknown." %
-                                 self.op.target_node)
-    self.op.target_node = self.dst_node.name
+    assert self.dst_node is not None, \
+          "Cannot retrieve locked node %s" % self.op.target_node
 
     # instance disk type verification
     for disk in self.instance.disks:
@@ -4635,6 +4798,14 @@ class LURemoveExport(NoHooksLU):
 
   """
   _OP_REQP = ["instance_name"]
+  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
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -4653,7 +4824,7 @@ class LURemoveExport(NoHooksLU):
       fqdn_warn = True
       instance_name = self.op.instance_name
 
-    exportlist = rpc.call_export_list(self.cfg.GetNodeList())
+    exportlist = rpc.call_export_list(self.acquired_locks[locking.LEVEL_NODE])
     found = False
     for node in exportlist:
       if instance_name in exportlist[node]:
@@ -4674,26 +4845,34 @@ class TagsLU(NoHooksLU):
   This is an abstract class which is the parent of all the other tags LUs.
 
   """
-  def CheckPrereq(self):
-    """Check prerequisites.
 
-    """
-    if self.op.kind == constants.TAG_CLUSTER:
-      self.target = self.cfg.GetClusterInfo()
-    elif self.op.kind == constants.TAG_NODE:
+  def ExpandNames(self):
+    self.needed_locks = {}
+    if self.op.kind == constants.TAG_NODE:
       name = self.cfg.ExpandNodeName(self.op.name)
       if name is None:
         raise errors.OpPrereqError("Invalid node name (%s)" %
                                    (self.op.name,))
       self.op.name = name
-      self.target = self.cfg.GetNodeInfo(name)
+      self.needed_locks[locking.LEVEL_NODE] = name
     elif self.op.kind == constants.TAG_INSTANCE:
       name = self.cfg.ExpandInstanceName(self.op.name)
       if name is None:
         raise errors.OpPrereqError("Invalid instance name (%s)" %
                                    (self.op.name,))
       self.op.name = name
-      self.target = self.cfg.GetInstanceInfo(name)
+      self.needed_locks[locking.LEVEL_INSTANCE] = name
+
+  def CheckPrereq(self):
+    """Check prerequisites.
+
+    """
+    if self.op.kind == constants.TAG_CLUSTER:
+      self.target = self.cfg.GetClusterInfo()
+    elif self.op.kind == constants.TAG_NODE:
+      self.target = self.cfg.GetNodeInfo(self.op.name)
+    elif self.op.kind == constants.TAG_INSTANCE:
+      self.target = self.cfg.GetInstanceInfo(self.op.name)
     else:
       raise errors.OpPrereqError("Wrong tag type requested (%s)" %
                                  str(self.op.kind))
@@ -4704,6 +4883,7 @@ class LUGetTags(TagsLU):
 
   """
   _OP_REQP = ["kind", "name"]
+  REQ_BGL = False
 
   def Exec(self, feedback_fn):
     """Returns the tag list.
@@ -4717,6 +4897,10 @@ class LUSearchTags(NoHooksLU):
 
   """
   _OP_REQP = ["pattern"]
+  REQ_BGL = False
+
+  def ExpandNames(self):
+    self.needed_locks = {}
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -4736,9 +4920,9 @@ class LUSearchTags(NoHooksLU):
     """
     cfg = self.cfg
     tgts = [("/cluster", cfg.GetClusterInfo())]
-    ilist = [cfg.GetInstanceInfo(name) for name in cfg.GetInstanceList()]
+    ilist = cfg.GetAllInstancesInfo().values()
     tgts.extend([("/instances/%s" % i.name, i) for i in ilist])
-    nlist = [cfg.GetNodeInfo(name) for name in cfg.GetNodeList()]
+    nlist = cfg.GetAllNodesInfo().values()
     tgts.extend([("/nodes/%s" % n.name, n) for n in nlist])
     results = []
     for path, target in tgts:
@@ -4753,6 +4937,7 @@ class LUAddTags(TagsLU):
 
   """
   _OP_REQP = ["kind", "name", "tags"]
+  REQ_BGL = False
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -4786,6 +4971,7 @@ class LUDelTags(TagsLU):
 
   """
   _OP_REQP = ["kind", "name", "tags"]
+  REQ_BGL = False
 
   def CheckPrereq(self):
     """Check prerequisites.
@@ -5096,7 +5282,7 @@ class IAllocator(object):
 
     result = call_fn(self.sstore.GetMasterNode(), name, self.in_text)
 
-    if not isinstance(result, tuple) or len(result) != 4:
+    if not isinstance(result, (list, tuple)) or len(result) != 4:
       raise errors.OpExecError("Invalid result from master iallocator runner")
 
     rcode, stdout, stderr, fail = result