hv_chroot: remove hard-coded path constructs

[ganeti-local] / lib / bdev.py

diff --git a/lib/bdev.py b/lib/bdev.py
index 8fa718d..bdfbec7 100644 (file)
--- a/lib/bdev.py
+++ b/lib/bdev.py
@@ -31,6 +31,11 @@ import logging
 from ganeti import utils
 from ganeti import errors
 from ganeti import constants
 from ganeti import utils
 from ganeti import errors
 from ganeti import constants

+from ganeti import objects
+
+
+# Size of reads in _CanReadDevice
+_DEVICE_READ_SIZE = 128 * 1024

 
 
 def _IgnoreError(fn, *args, **kwargs):
 
 
 def _IgnoreError(fn, *args, **kwargs):

@@ -47,7 +52,7 @@ def _IgnoreError(fn, *args, **kwargs):
     fn(*args, **kwargs)
     return True
   except errors.BlockDeviceError, err:
     fn(*args, **kwargs)
     return True
   except errors.BlockDeviceError, err:

-    logging.warning("Caught BlockDeviceError but ignoring: %s" % str(err))
+    logging.warning("Caught BlockDeviceError but ignoring: %s", str(err))

     return False
 
 
     return False
 
 

@@ -65,6 +70,20 @@ def _ThrowError(msg, *args):
   raise errors.BlockDeviceError(msg)
 
 
   raise errors.BlockDeviceError(msg)
 
 

+def _CanReadDevice(path):
+  """Check if we can read from the given device.
+
+  This tries to read the first 128k of the device.
+
+  """
+  try:
+    utils.ReadFile(path, size=_DEVICE_READ_SIZE)
+    return True
+  except EnvironmentError:
+    logging.warning("Can't read from device %s", path, exc_info=True)
+    return False
+
+

 class BlockDev(object):
   """Block device abstract class.
 
 class BlockDev(object):
   """Block device abstract class.
 

@@ -108,13 +127,14 @@ class BlockDev(object):
   after assembly we'll have our correct major/minor.
 
   """
   after assembly we'll have our correct major/minor.
 
   """

-  def __init__(self, unique_id, children):
+  def __init__(self, unique_id, children, size):

     self._children = children
     self.dev_path = None
     self.unique_id = unique_id
     self.major = None
     self.minor = None
     self.attached = False
     self._children = children
     self.dev_path = None
     self.unique_id = unique_id
     self.major = None
     self.minor = None
     self.attached = False

+    self.size = size

 
   def Assemble(self):
     """Assemble the device from its components.
 
   def Assemble(self):
     """Assemble the device from its components.

@@ -128,7 +148,7 @@ class BlockDev(object):
         this method is idempotent
 
     """
         this method is idempotent
 
     """

-    return True
+    pass

 
   def Attach(self):
     """Find a device which matches our config and attach to it.
 
   def Attach(self):
     """Find a device which matches our config and attach to it.

@@ -160,7 +180,7 @@ class BlockDev(object):
     """Remove this device.
 
     This makes sense only for some of the device types: LV and file
     """Remove this device.
 
     This makes sense only for some of the device types: LV and file

-    storeage. Also note that if the device can't attach, the removal
+    storage. Also note that if the device can't attach, the removal

     can't be completed.
 
     """
     can't be completed.
 
     """

@@ -227,12 +247,16 @@ class BlockDev(object):
     data. This is only valid for some devices, the rest will always
     return False (not degraded).
 
     data. This is only valid for some devices, the rest will always
     return False (not degraded).
 

-    @rtype: tuple
-    @return: (sync_percent, estimated_time, is_degraded, ldisk)
+    @rtype: objects.BlockDevStatus

 
     """
 
     """

-    return None, None, False, False
-
+    return objects.BlockDevStatus(dev_path=self.dev_path,
+                                  major=self.major,
+                                  minor=self.minor,
+                                  sync_percent=None,
+                                  estimated_time=None,
+                                  is_degraded=False,
+                                  ldisk_status=constants.LDS_OKAY)

 
   def CombinedSyncStatus(self):
     """Calculate the mirror status recursively for our children.
 
   def CombinedSyncStatus(self):
     """Calculate the mirror status recursively for our children.

@@ -241,22 +265,44 @@ class BlockDev(object):
     minimum percent and maximum time are calculated across our
     children.
 
     minimum percent and maximum time are calculated across our
     children.
 

+    @rtype: objects.BlockDevStatus
+

     """
     """

-    min_percent, max_time, is_degraded, ldisk = self.GetSyncStatus()
+    status = self.GetSyncStatus()
+
+    min_percent = status.sync_percent
+    max_time = status.estimated_time
+    is_degraded = status.is_degraded
+    ldisk_status = status.ldisk_status
+

     if self._children:
       for child in self._children:
     if self._children:
       for child in self._children:

-        c_percent, c_time, c_degraded, c_ldisk = child.GetSyncStatus()
+        child_status = child.GetSyncStatus()
+

         if min_percent is None:
         if min_percent is None:

-          min_percent = c_percent
-        elif c_percent is not None:
-          min_percent = min(min_percent, c_percent)
+          min_percent = child_status.sync_percent
+        elif child_status.sync_percent is not None:
+          min_percent = min(min_percent, child_status.sync_percent)
+

         if max_time is None:
         if max_time is None:

-          max_time = c_time
-        elif c_time is not None:
-          max_time = max(max_time, c_time)
-        is_degraded = is_degraded or c_degraded
-        ldisk = ldisk or c_ldisk
-    return min_percent, max_time, is_degraded, ldisk
+          max_time = child_status.estimated_time
+        elif child_status.estimated_time is not None:
+          max_time = max(max_time, child_status.estimated_time)
+
+        is_degraded = is_degraded or child_status.is_degraded
+
+        if ldisk_status is None:
+          ldisk_status = child_status.ldisk_status
+        elif child_status.ldisk_status is not None:
+          ldisk_status = max(ldisk_status, child_status.ldisk_status)
+
+    return objects.BlockDevStatus(dev_path=self.dev_path,
+                                  major=self.major,
+                                  minor=self.minor,
+                                  sync_percent=min_percent,
+                                  estimated_time=max_time,
+                                  is_degraded=is_degraded,
+                                  ldisk_status=ldisk_status)

 
 
   def SetInfo(self, text):
 
 
   def SetInfo(self, text):

@@ -276,6 +322,23 @@ class BlockDev(object):
     """
     raise NotImplementedError
 
     """
     raise NotImplementedError
 

+  def GetActualSize(self):
+    """Return the actual disk size.
+
+    @note: the device needs to be active when this is called
+
+    """
+    assert self.attached, "BlockDevice not attached in GetActualSize()"
+    result = utils.RunCmd(["blockdev", "--getsize64", self.dev_path])
+    if result.failed:
+      _ThrowError("blockdev failed (%s): %s",
+                  result.fail_reason, result.output)
+    try:
+      sz = int(result.output.strip())
+    except (ValueError, TypeError), err:
+      _ThrowError("Failed to parse blockdev output: %s", str(err))
+    return sz
+

   def __repr__(self):
     return ("<%s: unique_id: %s, children: %s, %s:%s, %s>" %
             (self.__class__, self.unique_id, self._children,
   def __repr__(self):
     return ("<%s: unique_id: %s, children: %s, %s:%s, %s>" %
             (self.__class__, self.unique_id, self._children,

@@ -286,19 +349,25 @@ class LogicalVolume(BlockDev):
   """Logical Volume block device.
 
   """
   """Logical Volume block device.
 
   """

-  def __init__(self, unique_id, children):
+  _VALID_NAME_RE = re.compile("^[a-zA-Z0-9+_.-]*$")
+  _INVALID_NAMES = frozenset([".", "..", "snapshot", "pvmove"])
+  _INVALID_SUBSTRINGS = frozenset(["_mlog", "_mimage"])
+
+  def __init__(self, unique_id, children, size):

     """Attaches to a LV device.
 
     The unique_id is a tuple (vg_name, lv_name)
 
     """
     """Attaches to a LV device.
 
     The unique_id is a tuple (vg_name, lv_name)
 
     """

-    super(LogicalVolume, self).__init__(unique_id, children)
+    super(LogicalVolume, self).__init__(unique_id, children, size)

     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     self._vg_name, self._lv_name = unique_id
     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     self._vg_name, self._lv_name = unique_id

-    self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name)
+    self._ValidateName(self._vg_name)
+    self._ValidateName(self._lv_name)
+    self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name)

     self._degraded = True
     self._degraded = True

-    self.major = self.minor = None
+    self.major = self.minor = self.pe_size = self.stripe_count = None

     self.Attach()
 
   @classmethod
     self.Attach()
 
   @classmethod

@@ -310,40 +379,56 @@ class LogicalVolume(BlockDev):
       raise errors.ProgrammerError("Invalid configuration data %s" %
                                    str(unique_id))
     vg_name, lv_name = unique_id
       raise errors.ProgrammerError("Invalid configuration data %s" %
                                    str(unique_id))
     vg_name, lv_name = unique_id

-    pvs_info = cls.GetPVInfo(vg_name)
+    cls._ValidateName(vg_name)
+    cls._ValidateName(lv_name)
+    pvs_info = cls.GetPVInfo([vg_name])

     if not pvs_info:
       _ThrowError("Can't compute PV info for vg %s", vg_name)
     pvs_info.sort()
     pvs_info.reverse()
 
     pvlist = [ pv[1] for pv in pvs_info ]
     if not pvs_info:
       _ThrowError("Can't compute PV info for vg %s", vg_name)
     pvs_info.sort()
     pvs_info.reverse()
 
     pvlist = [ pv[1] for pv in pvs_info ]

+    if utils.any(pvlist, lambda v: ":" in v):
+      _ThrowError("Some of your PVs have invalid character ':'"
+                  " in their name")

     free_size = sum([ pv[0] for pv in pvs_info ])
     free_size = sum([ pv[0] for pv in pvs_info ])

+    current_pvs = len(pvlist)
+    stripes = min(current_pvs, constants.LVM_STRIPECOUNT)

 
     # The size constraint should have been checked from the master before
     # calling the create function.
     if free_size < size:
       _ThrowError("Not enough free space: required %s,"
                   " available %s", size, free_size)
 
     # The size constraint should have been checked from the master before
     # calling the create function.
     if free_size < size:
       _ThrowError("Not enough free space: required %s,"
                   " available %s", size, free_size)

-    result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-n%s" % lv_name,
-                           vg_name] + pvlist)
+    cmd = ["lvcreate", "-L%dm" % size, "-n%s" % lv_name]
+    # If the free space is not well distributed, we won't be able to
+    # create an optimally-striped volume; in that case, we want to try
+    # with N, N-1, ..., 2, and finally 1 (non-stripped) number of
+    # stripes
+    for stripes_arg in range(stripes, 0, -1):
+      result = utils.RunCmd(cmd + ["-i%d" % stripes_arg] + [vg_name] + pvlist)
+      if not result.failed:
+        break

     if result.failed:
       _ThrowError("LV create failed (%s): %s",
                   result.fail_reason, result.output)
     if result.failed:
       _ThrowError("LV create failed (%s): %s",
                   result.fail_reason, result.output)

-    return LogicalVolume(unique_id, children)
+    return LogicalVolume(unique_id, children, size)

 
   @staticmethod
 
   @staticmethod

-  def GetPVInfo(vg_name):
+  def GetPVInfo(vg_names, filter_allocatable=True):

     """Get the free space info for PVs in a volume group.
 
     """Get the free space info for PVs in a volume group.
 

-    @param vg_name: the volume group name
+    @param vg_names: list of volume group names, if empty all will be returned
+    @param filter_allocatable: whether to skip over unallocatable PVs

 
     @rtype: list
     @return: list of tuples (free_space, name) with free_space in mebibytes
 
     """
 
     @rtype: list
     @return: list of tuples (free_space, name) with free_space in mebibytes
 
     """

+    sep = "|"

     command = ["pvs", "--noheadings", "--nosuffix", "--units=m",
                "-opv_name,vg_name,pv_free,pv_attr", "--unbuffered",
     command = ["pvs", "--noheadings", "--nosuffix", "--units=m",
                "-opv_name,vg_name,pv_free,pv_attr", "--unbuffered",

-               "--separator=:"]
+               "--separator=%s" % sep ]

     result = utils.RunCmd(command)
     if result.failed:
       logging.error("Can't get the PV information: %s - %s",
     result = utils.RunCmd(command)
     if result.failed:
       logging.error("Can't get the PV information: %s - %s",

@@ -351,17 +436,34 @@ class LogicalVolume(BlockDev):
       return None
     data = []
     for line in result.stdout.splitlines():
       return None
     data = []
     for line in result.stdout.splitlines():

-      fields = line.strip().split(':')
+      fields = line.strip().split(sep)

       if len(fields) != 4:
         logging.error("Can't parse pvs output: line '%s'", line)
         return None
       if len(fields) != 4:
         logging.error("Can't parse pvs output: line '%s'", line)
         return None

-      # skip over pvs from another vg or ones which are not allocatable
-      if fields[1] != vg_name or fields[3][0] != 'a':
+      # (possibly) skip over pvs which are not allocatable
+      if filter_allocatable and fields[3][0] != 'a':
+        continue
+      # (possibly) skip over pvs which are not in the right volume group(s)
+      if vg_names and fields[1] not in vg_names:

         continue
         continue

-      data.append((float(fields[2]), fields[0]))
+      data.append((float(fields[2]), fields[0], fields[1]))

 
     return data
 
 
     return data
 

+  @classmethod
+  def _ValidateName(cls, name):
+    """Validates that a given name is valid as VG or LV name.
+
+    The list of valid characters and restricted names is taken out of
+    the lvm(8) manpage, with the simplification that we enforce both
+    VG and LV restrictions on the names.
+
+    """
+    if (not cls._VALID_NAME_RE.match(name) or
+        name in cls._INVALID_NAMES or
+        utils.any(cls._INVALID_SUBSTRINGS, lambda x: x in name)):
+      _ThrowError("Invalid LVM name '%s'", name)
+

   def Remove(self):
     """Remove this logical volume.
 
   def Remove(self):
     """Remove this logical volume.
 

@@ -389,7 +491,7 @@ class LogicalVolume(BlockDev):
     if result.failed:
       _ThrowError("Failed to rename the logical volume: %s", result.output)
     self._lv_name = new_name
     if result.failed:
       _ThrowError("Failed to rename the logical volume: %s", result.output)
     self._lv_name = new_name

-    self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name)
+    self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name)

 
   def Attach(self):
     """Attach to an existing LV.
 
   def Attach(self):
     """Attach to an existing LV.

@@ -401,19 +503,30 @@ class LogicalVolume(BlockDev):
     """
     self.attached = False
     result = utils.RunCmd(["lvs", "--noheadings", "--separator=,",
     """
     self.attached = False
     result = utils.RunCmd(["lvs", "--noheadings", "--separator=,",

-                           "-olv_attr,lv_kernel_major,lv_kernel_minor",
-                           self.dev_path])
+                           "--units=m", "--nosuffix",
+                           "-olv_attr,lv_kernel_major,lv_kernel_minor,"
+                           "vg_extent_size,stripes", self.dev_path])

     if result.failed:
       logging.error("Can't find LV %s: %s, %s",
                     self.dev_path, result.fail_reason, result.output)
       return False
     if result.failed:
       logging.error("Can't find LV %s: %s, %s",
                     self.dev_path, result.fail_reason, result.output)
       return False

-    out = result.stdout.strip().rstrip(',')
+    # the output can (and will) have multiple lines for multi-segment
+    # LVs, as the 'stripes' parameter is a segment one, so we take
+    # only the last entry, which is the one we're interested in; note
+    # that with LVM2 anyway the 'stripes' value must be constant
+    # across segments, so this is a no-op actually
+    out = result.stdout.splitlines()
+    if not out: # totally empty result? splitlines() returns at least
+                # one line for any non-empty string
+      logging.error("Can't parse LVS output, no lines? Got '%s'", str(out))
+      return False
+    out = out[-1].strip().rstrip(',')

     out = out.split(",")
     out = out.split(",")

-    if len(out) != 3:
-      logging.error("Can't parse LVS output, len(%s) != 3", str(out))
+    if len(out) != 5:
+      logging.error("Can't parse LVS output, len(%s) != 5", str(out))

       return False
 
       return False
 

-    status, major, minor = out[:3]
+    status, major, minor, pe_size, stripes = out

     if len(status) != 6:
       logging.error("lvs lv_attr is not 6 characters (%s)", status)
       return False
     if len(status) != 6:
       logging.error("lvs lv_attr is not 6 characters (%s)", status)
       return False

@@ -421,11 +534,25 @@ class LogicalVolume(BlockDev):
     try:
       major = int(major)
       minor = int(minor)
     try:
       major = int(major)
       minor = int(minor)

-    except ValueError, err:
+    except (TypeError, ValueError), err:

       logging.error("lvs major/minor cannot be parsed: %s", str(err))
 
       logging.error("lvs major/minor cannot be parsed: %s", str(err))
 

+    try:
+      pe_size = int(float(pe_size))
+    except (TypeError, ValueError), err:
+      logging.error("Can't parse vg extent size: %s", err)
+      return False
+
+    try:
+      stripes = int(stripes)
+    except (TypeError, ValueError), err:
+      logging.error("Can't parse the number of stripes: %s", err)
+      return False
+

     self.major = major
     self.minor = minor
     self.major = major
     self.minor = minor

+    self.pe_size = pe_size
+    self.stripe_count = stripes

     self._degraded = status[0] == 'v' # virtual volume, i.e. doesn't backing
                                       # storage
     self.attached = True
     self._degraded = status[0] == 'v' # virtual volume, i.e. doesn't backing
                                       # storage
     self.attached = True

@@ -434,16 +561,14 @@ class LogicalVolume(BlockDev):
   def Assemble(self):
     """Assemble the device.
 
   def Assemble(self):
     """Assemble the device.
 

-    We alway run `lvchange -ay` on the LV to ensure it's active before
+    We always run `lvchange -ay` on the LV to ensure it's active before

     use, as there were cases when xenvg was not active after boot
     (also possibly after disk issues).
 
     """
     result = utils.RunCmd(["lvchange", "-ay", self.dev_path])
     if result.failed:
     use, as there were cases when xenvg was not active after boot
     (also possibly after disk issues).
 
     """
     result = utils.RunCmd(["lvchange", "-ay", self.dev_path])
     if result.failed:

-      logging.error("Can't activate lv %s: %s", self.dev_path, result.output)
-      return False
-    return self.Attach()
+      _ThrowError("Can't activate lv %s: %s", self.dev_path, result.output)

 
   def Shutdown(self):
     """Shutdown the device.
 
   def Shutdown(self):
     """Shutdown the device.

@@ -473,11 +598,21 @@ class LogicalVolume(BlockDev):
 
     The status was already read in Attach, so we just return it.
 
 
     The status was already read in Attach, so we just return it.
 

-    @rtype: tuple
-    @return: (sync_percent, estimated_time, is_degraded, ldisk)
+    @rtype: objects.BlockDevStatus

 
     """
 
     """

-    return None, None, self._degraded, self._degraded
+    if self._degraded:
+      ldisk_status = constants.LDS_FAULTY
+    else:
+      ldisk_status = constants.LDS_OKAY
+
+    return objects.BlockDevStatus(dev_path=self.dev_path,
+                                  major=self.major,
+                                  minor=self.minor,
+                                  sync_percent=None,
+                                  estimated_time=None,
+                                  is_degraded=self._degraded,
+                                  ldisk_status=ldisk_status)

 
   def Open(self, force=False):
     """Make the device ready for I/O.
 
   def Open(self, force=False):
     """Make the device ready for I/O.

@@ -502,15 +637,15 @@ class LogicalVolume(BlockDev):
     snap_name = self._lv_name + ".snap"
 
     # remove existing snapshot if found
     snap_name = self._lv_name + ".snap"
 
     # remove existing snapshot if found

-    snap = LogicalVolume((self._vg_name, snap_name), None)
+    snap = LogicalVolume((self._vg_name, snap_name), None, size)

     _IgnoreError(snap.Remove)
 
     _IgnoreError(snap.Remove)
 

-    pvs_info = self.GetPVInfo(self._vg_name)
+    pvs_info = self.GetPVInfo([self._vg_name])

     if not pvs_info:
       _ThrowError("Can't compute PV info for vg %s", self._vg_name)
     pvs_info.sort()
     pvs_info.reverse()
     if not pvs_info:
       _ThrowError("Can't compute PV info for vg %s", self._vg_name)
     pvs_info.sort()
     pvs_info.reverse()

-    free_size, pv_name = pvs_info[0]
+    free_size, _, _ = pvs_info[0]

     if free_size < size:
       _ThrowError("Not enough free space: required %s,"
                   " available %s", size, free_size)
     if free_size < size:
       _ThrowError("Not enough free space: required %s,"
                   " available %s", size, free_size)

@@ -546,6 +681,13 @@ class LogicalVolume(BlockDev):
     """Grow the logical volume.
 
     """
     """Grow the logical volume.
 
     """

+    if self.pe_size is None or self.stripe_count is None:
+      if not self.Attach():
+        _ThrowError("Can't attach to LV during Grow()")
+    full_stripe_size = self.pe_size * self.stripe_count
+    rest = amount % full_stripe_size
+    if rest != 0:
+      amount += full_stripe_size - rest

     # we try multiple algorithms since the 'best' ones might not have
     # space available in the right place, but later ones might (since
     # they have less constraints); also note that only recent LVM
     # we try multiple algorithms since the 'best' ones might not have
     # space available in the right place, but later ones might (since
     # they have less constraints); also note that only recent LVM

@@ -565,15 +707,56 @@ class DRBD8Status(object):
 
   """
   UNCONF_RE = re.compile(r"\s*[0-9]+:\s*cs:Unconfigured$")
 
   """
   UNCONF_RE = re.compile(r"\s*[0-9]+:\s*cs:Unconfigured$")

-  LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+st:([^/]+)/(\S+)"
+  LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+(?:st|ro):([^/]+)/(\S+)"

                        "\s+ds:([^/]+)/(\S+)\s+.*$")
   SYNC_RE = re.compile(r"^.*\ssync'ed:\s*([0-9.]+)%.*"
                        "\sfinish: ([0-9]+):([0-9]+):([0-9]+)\s.*$")
 
                        "\s+ds:([^/]+)/(\S+)\s+.*$")
   SYNC_RE = re.compile(r"^.*\ssync'ed:\s*([0-9.]+)%.*"
                        "\sfinish: ([0-9]+):([0-9]+):([0-9]+)\s.*$")
 

+  CS_UNCONFIGURED = "Unconfigured"
+  CS_STANDALONE = "StandAlone"
+  CS_WFCONNECTION = "WFConnection"
+  CS_WFREPORTPARAMS = "WFReportParams"
+  CS_CONNECTED = "Connected"
+  CS_STARTINGSYNCS = "StartingSyncS"
+  CS_STARTINGSYNCT = "StartingSyncT"
+  CS_WFBITMAPS = "WFBitMapS"
+  CS_WFBITMAPT = "WFBitMapT"
+  CS_WFSYNCUUID = "WFSyncUUID"
+  CS_SYNCSOURCE = "SyncSource"
+  CS_SYNCTARGET = "SyncTarget"
+  CS_PAUSEDSYNCS = "PausedSyncS"
+  CS_PAUSEDSYNCT = "PausedSyncT"
+  CSET_SYNC = frozenset([
+    CS_WFREPORTPARAMS,
+    CS_STARTINGSYNCS,
+    CS_STARTINGSYNCT,
+    CS_WFBITMAPS,
+    CS_WFBITMAPT,
+    CS_WFSYNCUUID,
+    CS_SYNCSOURCE,
+    CS_SYNCTARGET,
+    CS_PAUSEDSYNCS,
+    CS_PAUSEDSYNCT,
+    ])
+
+  DS_DISKLESS = "Diskless"
+  DS_ATTACHING = "Attaching" # transient state
+  DS_FAILED = "Failed" # transient state, next: diskless
+  DS_NEGOTIATING = "Negotiating" # transient state
+  DS_INCONSISTENT = "Inconsistent" # while syncing or after creation
+  DS_OUTDATED = "Outdated"
+  DS_DUNKNOWN = "DUnknown" # shown for peer disk when not connected
+  DS_CONSISTENT = "Consistent"
+  DS_UPTODATE = "UpToDate" # normal state
+
+  RO_PRIMARY = "Primary"
+  RO_SECONDARY = "Secondary"
+  RO_UNKNOWN = "Unknown"
+

   def __init__(self, procline):
     u = self.UNCONF_RE.match(procline)
     if u:
   def __init__(self, procline):
     u = self.UNCONF_RE.match(procline)
     if u:

-      self.cstatus = "Unconfigured"
+      self.cstatus = self.CS_UNCONFIGURED

       self.lrole = self.rrole = self.ldisk = self.rdisk = None
     else:
       m = self.LINE_RE.match(procline)
       self.lrole = self.rrole = self.ldisk = self.rdisk = None
     else:
       m = self.LINE_RE.match(procline)

@@ -587,21 +770,21 @@ class DRBD8Status(object):
 
     # end reading of data from the LINE_RE or UNCONF_RE
 
 
     # end reading of data from the LINE_RE or UNCONF_RE
 

-    self.is_standalone = self.cstatus == "StandAlone"
-    self.is_wfconn = self.cstatus == "WFConnection"
-    self.is_connected = self.cstatus == "Connected"
-    self.is_primary = self.lrole == "Primary"
-    self.is_secondary = self.lrole == "Secondary"
-    self.peer_primary = self.rrole == "Primary"
-    self.peer_secondary = self.rrole == "Secondary"
+    self.is_standalone = self.cstatus == self.CS_STANDALONE
+    self.is_wfconn = self.cstatus == self.CS_WFCONNECTION
+    self.is_connected = self.cstatus == self.CS_CONNECTED
+    self.is_primary = self.lrole == self.RO_PRIMARY
+    self.is_secondary = self.lrole == self.RO_SECONDARY
+    self.peer_primary = self.rrole == self.RO_PRIMARY
+    self.peer_secondary = self.rrole == self.RO_SECONDARY

     self.both_primary = self.is_primary and self.peer_primary
     self.both_secondary = self.is_secondary and self.peer_secondary
 
     self.both_primary = self.is_primary and self.peer_primary
     self.both_secondary = self.is_secondary and self.peer_secondary
 

-    self.is_diskless = self.ldisk == "Diskless"
-    self.is_disk_uptodate = self.ldisk == "UpToDate"
+    self.is_diskless = self.ldisk == self.DS_DISKLESS
+    self.is_disk_uptodate = self.ldisk == self.DS_UPTODATE

 
 

-    self.is_in_resync = self.cstatus in ("SyncSource", "SyncTarget")
-    self.is_in_use = self.cstatus != "Unconfigured"
+    self.is_in_resync = self.cstatus in self.CSET_SYNC
+    self.is_in_use = self.cstatus != self.CS_UNCONFIGURED

 
     m = self.SYNC_RE.match(procline)
     if m:
 
     m = self.SYNC_RE.match(procline)
     if m:

@@ -611,15 +794,18 @@ class DRBD8Status(object):
       seconds = int(m.group(4))
       self.est_time = hours * 3600 + minutes * 60 + seconds
     else:
       seconds = int(m.group(4))
       self.est_time = hours * 3600 + minutes * 60 + seconds
     else:

-      self.sync_percent = None
+      # we have (in this if branch) no percent information, but if
+      # we're resyncing we need to 'fake' a sync percent information,
+      # as this is how cmdlib determines if it makes sense to wait for
+      # resyncing or not
+      if self.is_in_resync:
+        self.sync_percent = 0
+      else:
+        self.sync_percent = None

       self.est_time = None
 
       self.est_time = None
 

-    self.is_sync_target = self.peer_sync_source = self.cstatus == "SyncTarget"
-    self.peer_sync_target = self.is_sync_source = self.cstatus == "SyncSource"
-    self.is_resync = self.is_sync_target or self.is_sync_source
-

 
 

-class BaseDRBD(BlockDev):
+class BaseDRBD(BlockDev): # pylint: disable-msg=W0223

   """Base DRBD class.
 
   This class contains a few bits of common functionality between the
   """Base DRBD class.
 
   This class contains a few bits of common functionality between the

@@ -628,6 +814,8 @@ class BaseDRBD(BlockDev):
   """
   _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)"
                            r" \(api:(\d+)/proto:(\d+)(?:-(\d+))?\)")
   """
   _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)"
                            r" \(api:(\d+)/proto:(\d+)(?:-(\d+))?\)")

+  _VALID_LINE_RE = re.compile("^ *([0-9]+): cs:([^ ]+).*$")
+  _UNUSED_LINE_RE = re.compile("^ *([0-9]+): cs:Unconfigured$")

 
   _DRBD_MAJOR = 147
   _ST_UNCONFIGURED = "Unconfigured"
 
   _DRBD_MAJOR = 147
   _ST_UNCONFIGURED = "Unconfigured"

@@ -641,28 +829,32 @@ class BaseDRBD(BlockDev):
     """Return data from /proc/drbd.
 
     """
     """Return data from /proc/drbd.
 
     """

-    stat = open(filename, "r")

     try:
     try:

-      data = stat.read().splitlines()
-    finally:
-      stat.close()
+      data = utils.ReadFile(filename).splitlines()
+    except EnvironmentError, err:
+      if err.errno == errno.ENOENT:
+        _ThrowError("The file %s cannot be opened, check if the module"
+                    " is loaded (%s)", filename, str(err))
+      else:
+        _ThrowError("Can't read the DRBD proc file %s: %s", filename, str(err))

     if not data:
       _ThrowError("Can't read any data from %s", filename)
     return data
 
     if not data:
       _ThrowError("Can't read any data from %s", filename)
     return data
 

-  @staticmethod
-  def _MassageProcData(data):
+  @classmethod
+  def _MassageProcData(cls, data):

     """Transform the output of _GetProdData into a nicer form.
 
     @return: a dictionary of minor: joined lines from /proc/drbd
         for that minor
 
     """
     """Transform the output of _GetProdData into a nicer form.
 
     @return: a dictionary of minor: joined lines from /proc/drbd
         for that minor
 
     """

-    lmatch = re.compile("^ *([0-9]+):.*$")

     results = {}
     old_minor = old_line = None
     for line in data:
     results = {}
     old_minor = old_line = None
     for line in data:

-      lresult = lmatch.match(line)
+      if not line: # completely empty lines, as can be returned by drbd8.0+
+        continue
+      lresult = cls._VALID_LINE_RE.match(line)

       if lresult is not None:
         if old_minor is not None:
           results[old_minor] = old_line
       if lresult is not None:
         if old_minor is not None:
           results[old_minor] = old_line

@@ -723,9 +915,8 @@ class BaseDRBD(BlockDev):
     data = cls._GetProcData()
 
     used_devs = {}
     data = cls._GetProcData()
 
     used_devs = {}

-    valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$")

     for line in data:
     for line in data:

-      match = valid_line.match(line)
+      match = cls._VALID_LINE_RE.match(line)

       if not match:
         continue
       minor = int(match.group(1))
       if not match:
         continue
       minor = int(match.group(1))

@@ -760,22 +951,23 @@ class BaseDRBD(BlockDev):
     """
     result = utils.RunCmd(["blockdev", "--getsize", meta_device])
     if result.failed:
     """
     result = utils.RunCmd(["blockdev", "--getsize", meta_device])
     if result.failed:

-      logging.error("Failed to get device size: %s - %s",
-                    result.fail_reason, result.output)
-      return False
+      _ThrowError("Failed to get device size: %s - %s",
+                  result.fail_reason, result.output)

     try:
       sectors = int(result.stdout)
     try:
       sectors = int(result.stdout)

-    except ValueError:
-      logging.error("Invalid output from blockdev: '%s'", result.stdout)
-      return False
+    except (TypeError, ValueError):
+      _ThrowError("Invalid output from blockdev: '%s'", result.stdout)

     bytes = sectors * 512
     if bytes < 128 * 1024 * 1024: # less than 128MiB
     bytes = sectors * 512
     if bytes < 128 * 1024 * 1024: # less than 128MiB

-      logging.error("Meta device too small (%.2fMib)", (bytes / 1024 / 1024))
-      return False
-    if bytes > (128 + 32) * 1024 * 1024: # account for an extra (big) PE on LVM
-      logging.error("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024))
-      return False
-    return True
+      _ThrowError("Meta device too small (%.2fMib)", (bytes / 1024 / 1024))
+    # the maximum *valid* size of the meta device when living on top
+    # of LVM is hard to compute: it depends on the number of stripes
+    # and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB
+    # (normal size), but an eight-stripe 128MB PE will result in a 1GB
+    # size meta device; as such, we restrict it to 1GB (a little bit
+    # too generous, but making assumptions about PE size is hard)
+    if bytes > 1024 * 1024 * 1024:
+      _ThrowError("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024))

 
   def Rename(self, new_id):
     """Rename a device.
 
   def Rename(self, new_id):
     """Rename a device.

@@ -805,17 +997,9 @@ class DRBD8(BaseDRBD):
   # timeout constants
   _NET_RECONFIG_TIMEOUT = 60
 
   # timeout constants
   _NET_RECONFIG_TIMEOUT = 60
 

-  def __init__(self, unique_id, children):
+  def __init__(self, unique_id, children, size):

     if children and children.count(None) > 0:
       children = []
     if children and children.count(None) > 0:
       children = []

-    super(DRBD8, self).__init__(unique_id, children)
-    self.major = self._DRBD_MAJOR
-    version = self._GetVersion()
-    if version['k_major'] != 8 :
-      _ThrowError("Mismatch in DRBD kernel version and requested ganeti"
-                  " usage: kernel is %s.%s, ganeti wants 8.x",
-                  version['k_major'], version['k_minor'])
-

     if len(children) not in (0, 2):
       raise ValueError("Invalid configuration data %s" % str(children))
     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6:
     if len(children) not in (0, 2):
       raise ValueError("Invalid configuration data %s" % str(children))
     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6:

@@ -823,6 +1007,18 @@ class DRBD8(BaseDRBD):
     (self._lhost, self._lport,
      self._rhost, self._rport,
      self._aminor, self._secret) = unique_id
     (self._lhost, self._lport,
      self._rhost, self._rport,
      self._aminor, self._secret) = unique_id

+    if children:
+      if not _CanReadDevice(children[1].dev_path):
+        logging.info("drbd%s: Ignoring unreadable meta device", self._aminor)
+        children = []
+    super(DRBD8, self).__init__(unique_id, children, size)
+    self.major = self._DRBD_MAJOR
+    version = self._GetVersion()
+    if version['k_major'] != 8 :
+      _ThrowError("Mismatch in DRBD kernel version and requested ganeti"
+                  " usage: kernel is %s.%s, ganeti wants 8.x",
+                  version['k_major'], version['k_minor'])
+

     if (self._lhost is not None and self._lhost == self._rhost and
         self._lport == self._rport):
       raise ValueError("Invalid configuration data, same local/remote %s" %
     if (self._lhost is not None and self._lhost == self._rhost and
         self._lport == self._rport):
       raise ValueError("Invalid configuration data, same local/remote %s" %

@@ -851,14 +1047,12 @@ class DRBD8(BaseDRBD):
     """
     data = cls._GetProcData()
 
     """
     data = cls._GetProcData()
 

-    unused_line = re.compile("^ *([0-9]+): cs:Unconfigured$")
-    used_line = re.compile("^ *([0-9]+): cs:")

     highest = None
     for line in data:
     highest = None
     for line in data:

-      match = unused_line.match(line)
+      match = cls._UNUSED_LINE_RE.match(line)

       if match:
         return int(match.group(1))
       if match:
         return int(match.group(1))

-      match = used_line.match(line)
+      match = cls._VALID_LINE_RE.match(line)

       if match:
         minor = int(match.group(1))
         highest = max(highest, minor)
       if match:
         minor = int(match.group(1))
         highest = max(highest, minor)

@@ -896,15 +1090,20 @@ class DRBD8(BaseDRBD):
     # value types
     value = pyp.Word(pyp.alphanums + '_-/.:')
     quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote
     # value types
     value = pyp.Word(pyp.alphanums + '_-/.:')
     quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote

-    addr_port = (pyp.Word(pyp.nums + '.') + pyp.Literal(':').suppress() +
-                 number)
+    addr_type = (pyp.Optional(pyp.Literal("ipv4")).suppress() +
+                 pyp.Optional(pyp.Literal("ipv6")).suppress())
+    addr_port = (addr_type + pyp.Word(pyp.nums + '.') +
+                 pyp.Literal(':').suppress() + number)

     # meta device, extended syntax
     meta_value = ((value ^ quoted) + pyp.Literal('[').suppress() +
                   number + pyp.Word(']').suppress())
     # meta device, extended syntax
     meta_value = ((value ^ quoted) + pyp.Literal('[').suppress() +
                   number + pyp.Word(']').suppress())

+    # device name, extended syntax
+    device_value = pyp.Literal("minor").suppress() + number

 
     # a statement
     stmt = (~rbrace + keyword + ~lbrace +
 
     # a statement
     stmt = (~rbrace + keyword + ~lbrace +

-            pyp.Optional(addr_port ^ value ^ quoted ^ meta_value) +
+            pyp.Optional(addr_port ^ value ^ quoted ^ meta_value ^
+                         device_value) +

             pyp.Optional(defa) + semi +
             pyp.Optional(pyp.restOfLine).suppress())
 
             pyp.Optional(defa) + semi +
             pyp.Optional(pyp.restOfLine).suppress())
 

@@ -1026,16 +1225,37 @@ class DRBD8(BaseDRBD):
     return retval
 
   @classmethod
     return retval
 
   @classmethod

-  def _AssembleLocal(cls, minor, backend, meta):
+  def _AssembleLocal(cls, minor, backend, meta, size):

     """Configure the local part of a DRBD device.
 
     """
     args = ["drbdsetup", cls._DevPath(minor), "disk",
     """Configure the local part of a DRBD device.
 
     """
     args = ["drbdsetup", cls._DevPath(minor), "disk",

-            backend, meta, "0", "-e", "detach", "--create-device"]
+            backend, meta, "0",
+            "-e", "detach",
+            "--create-device"]
+    if size:
+      args.extend(["-d", "%sm" % size])
+    if not constants.DRBD_BARRIERS: # disable barriers, if configured so
+      version = cls._GetVersion()
+      # various DRBD versions support different disk barrier options;
+      # what we aim here is to revert back to the 'drain' method of
+      # disk flushes and to disable metadata barriers, in effect going
+      # back to pre-8.0.7 behaviour
+      vmaj = version['k_major']
+      vmin = version['k_minor']
+      vrel = version['k_point']
+      assert vmaj == 8
+      if vmin == 0: # 8.0.x
+        if vrel >= 12:
+          args.extend(['-i', '-m'])
+      elif vmin == 2: # 8.2.x
+        if vrel >= 7:
+          args.extend(['-i', '-m'])
+      elif vmaj >= 3: # 8.3.x or newer
+        args.extend(['-i', '-a', 'm'])

     result = utils.RunCmd(args)
     if result.failed:
     result = utils.RunCmd(args)
     if result.failed:

-      logging.error("Can't attach local disk: %s", result.output)
-    return not result.failed
+      _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output)

 
   @classmethod
   def _AssembleNet(cls, minor, net_info, protocol,
 
   @classmethod
   def _AssembleNet(cls, minor, net_info, protocol,

@@ -1047,7 +1267,8 @@ class DRBD8(BaseDRBD):
     if None in net_info:
       # we don't want network connection and actually want to make
       # sure its shutdown
     if None in net_info:
       # we don't want network connection and actually want to make
       # sure its shutdown

-      return cls._ShutdownNet(minor)
+      cls._ShutdownNet(minor)
+      return

 
     # Workaround for a race condition. When DRBD is doing its dance to
     # establish a connection with its peer, it also sends the
 
     # Workaround for a race condition. When DRBD is doing its dance to
     # establish a connection with its peer, it also sends the

@@ -1069,27 +1290,22 @@ class DRBD8(BaseDRBD):
       args.extend(["-a", hmac, "-x", secret])
     result = utils.RunCmd(args)
     if result.failed:
       args.extend(["-a", hmac, "-x", secret])
     result = utils.RunCmd(args)
     if result.failed:

-      logging.error("Can't setup network for dbrd device: %s - %s",
-                    result.fail_reason, result.output)
-      return False
+      _ThrowError("drbd%d: can't setup network: %s - %s",
+                  minor, result.fail_reason, result.output)

 
 

-    timeout = time.time() + 10
-    ok = False
-    while time.time() < timeout:
+    def _CheckNetworkConfig():

       info = cls._GetDevInfo(cls._GetShowData(minor))
       if not "local_addr" in info or not "remote_addr" in info:
       info = cls._GetDevInfo(cls._GetShowData(minor))
       if not "local_addr" in info or not "remote_addr" in info:

-        time.sleep(1)
-        continue
+        raise utils.RetryAgain()
+

       if (info["local_addr"] != (lhost, lport) or
           info["remote_addr"] != (rhost, rport)):
       if (info["local_addr"] != (lhost, lport) or
           info["remote_addr"] != (rhost, rport)):

-        time.sleep(1)
-        continue
-      ok = True
-      break
-    if not ok:
-      logging.error("Timeout while configuring network")
-      return False
-    return True
+        raise utils.RetryAgain()
+
+    try:
+      utils.Retry(_CheckNetworkConfig, 1.0, 10.0)
+    except utils.RetryTimeout:
+      _ThrowError("drbd%d: timeout while configuring network", minor)

 
   def AddChildren(self, devices):
     """Add a disk to the DRBD device.
 
   def AddChildren(self, devices):
     """Add a disk to the DRBD device.

@@ -1108,12 +1324,10 @@ class DRBD8(BaseDRBD):
       _ThrowError("drbd%d: children not ready during AddChildren", self.minor)
     backend.Open()
     meta.Open()
       _ThrowError("drbd%d: children not ready during AddChildren", self.minor)
     backend.Open()
     meta.Open()

-    if not self._CheckMetaSize(meta.dev_path):
-      raise errors.BlockDeviceError("Invalid meta device size")
+    self._CheckMetaSize(meta.dev_path)

     self._InitMeta(self._FindUnusedMinor(), meta.dev_path)
 
     self._InitMeta(self._FindUnusedMinor(), meta.dev_path)
 

-    if not self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path):
-      raise errors.BlockDeviceError("Can't attach to local storage")
+    self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size)

     self._children = devices
 
   def RemoveChildren(self, devices):
     self._children = devices
 
   def RemoveChildren(self, devices):

@@ -1140,8 +1354,7 @@ class DRBD8(BaseDRBD):
         _ThrowError("drbd%d: mismatch in local storage (%s != %s) in"
                     " RemoveChildren", self.minor, dev, child.dev_path)
 
         _ThrowError("drbd%d: mismatch in local storage (%s != %s) in"
                     " RemoveChildren", self.minor, dev, child.dev_path)
 

-    if not self._ShutdownLocal(self.minor):
-      raise errors.BlockDeviceError("Can't detach from local storage")
+    self._ShutdownLocal(self.minor)

     self._children = []
 
   @classmethod
     self._children = []
 
   @classmethod

@@ -1196,26 +1409,39 @@ class DRBD8(BaseDRBD):
 
 
     If sync_percent is None, it means all is ok
 
 
     If sync_percent is None, it means all is ok

-    If estimated_time is None, it means we can't esimate
+    If estimated_time is None, it means we can't estimate

     the time needed, otherwise it's the time left in seconds.
 
 
     We set the is_degraded parameter to True on two conditions:
     network not connected or local disk missing.
 
     the time needed, otherwise it's the time left in seconds.
 
 
     We set the is_degraded parameter to True on two conditions:
     network not connected or local disk missing.
 

-    We compute the ldisk parameter based on wheter we have a local
+    We compute the ldisk parameter based on whether we have a local

     disk or not.
 
     disk or not.
 

-    @rtype: tuple
-    @return: (sync_percent, estimated_time, is_degraded, ldisk)
+    @rtype: objects.BlockDevStatus

 
     """
     if self.minor is None and not self.Attach():
       _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor)
 
     """
     if self.minor is None and not self.Attach():
       _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor)

+

     stats = self.GetProcStatus()
     stats = self.GetProcStatus()

-    ldisk = not stats.is_disk_uptodate
-    is_degraded = not stats.is_connected
-    return stats.sync_percent, stats.est_time, is_degraded or ldisk, ldisk
+    is_degraded = not stats.is_connected or not stats.is_disk_uptodate
+
+    if stats.is_disk_uptodate:
+      ldisk_status = constants.LDS_OKAY
+    elif stats.is_diskless:
+      ldisk_status = constants.LDS_FAULTY
+    else:
+      ldisk_status = constants.LDS_UNKNOWN
+
+    return objects.BlockDevStatus(dev_path=self.dev_path,
+                                  major=self.major,
+                                  minor=self.minor,
+                                  sync_percent=stats.sync_percent,
+                                  estimated_time=stats.est_time,
+                                  is_degraded=is_degraded,
+                                  ldisk_status=ldisk_status)

 
   def Open(self, force=False):
     """Make the local state primary.
 
   def Open(self, force=False):
     """Make the local state primary.

@@ -1271,30 +1497,42 @@ class DRBD8(BaseDRBD):
       _ThrowError("drbd%d: DRBD disk missing network info in"
                   " DisconnectNet()", self.minor)
 
       _ThrowError("drbd%d: DRBD disk missing network info in"
                   " DisconnectNet()", self.minor)
 

-    ever_disconnected = self._ShutdownNet(self.minor)
-    timeout_limit = time.time() + self._NET_RECONFIG_TIMEOUT
-    sleep_time = 0.100 # we start the retry time at 100 miliseconds
-    while time.time() < timeout_limit:
-      status = self.GetProcStatus()
-      if status.is_standalone:
-        break
-      # retry the disconnect, it seems possible that due to a
-      # well-time disconnect on the peer, my disconnect command might
-      # be ingored and forgotten
-      ever_disconnected = self._ShutdownNet(self.minor) or ever_disconnected
-      time.sleep(sleep_time)
-      sleep_time = min(2, sleep_time * 1.5)
+    class _DisconnectStatus:
+      def __init__(self, ever_disconnected):
+        self.ever_disconnected = ever_disconnected

 
 

-    if not status.is_standalone:
-      if ever_disconnected:
+    dstatus = _DisconnectStatus(_IgnoreError(self._ShutdownNet, self.minor))
+
+    def _WaitForDisconnect():
+      if self.GetProcStatus().is_standalone:
+        return
+
+      # retry the disconnect, it seems possible that due to a well-time
+      # disconnect on the peer, my disconnect command might be ignored and
+      # forgotten
+      dstatus.ever_disconnected = \
+        _IgnoreError(self._ShutdownNet, self.minor) or dstatus.ever_disconnected
+
+      raise utils.RetryAgain()
+
+    # Keep start time
+    start_time = time.time()
+
+    try:
+      # Start delay at 100 milliseconds and grow up to 2 seconds
+      utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0),
+                  self._NET_RECONFIG_TIMEOUT)
+    except utils.RetryTimeout:
+      if dstatus.ever_disconnected:

         msg = ("drbd%d: device did not react to the"
                " 'disconnect' command in a timely manner")
       else:
         msg = "drbd%d: can't shutdown network, even after multiple retries"
         msg = ("drbd%d: device did not react to the"
                " 'disconnect' command in a timely manner")
       else:
         msg = "drbd%d: can't shutdown network, even after multiple retries"

+

       _ThrowError(msg, self.minor)
 
       _ThrowError(msg, self.minor)
 

-    reconfig_time = time.time() - timeout_limit + self._NET_RECONFIG_TIMEOUT
-    if reconfig_time > 15: # hardcoded alert limit
+    reconfig_time = time.time() - start_time
+    if reconfig_time > (self._NET_RECONFIG_TIMEOUT * 0.25):

       logging.info("drbd%d: DisconnectNet: detach took %.3f seconds",
                    self.minor, reconfig_time)
 
       logging.info("drbd%d: DisconnectNet: detach took %.3f seconds",
                    self.minor, reconfig_time)
 

@@ -1320,10 +1558,10 @@ class DRBD8(BaseDRBD):
     if not status.is_standalone:
       _ThrowError("drbd%d: device is not standalone in AttachNet", self.minor)
 
     if not status.is_standalone:
       _ThrowError("drbd%d: device is not standalone in AttachNet", self.minor)
 

-    return self._AssembleNet(self.minor,
-                             (self._lhost, self._lport,
-                              self._rhost, self._rport),
-                             "C", dual_pri=multimaster)
+    self._AssembleNet(self.minor,
+                      (self._lhost, self._lport, self._rhost, self._rport),
+                      constants.DRBD_NET_PROTOCOL, dual_pri=multimaster,
+                      hmac=constants.DRBD_HMAC_ALG, secret=self._secret)

 
   def Attach(self):
     """Check if our minor is configured.
 
   def Attach(self):
     """Check if our minor is configured.

@@ -1354,18 +1592,16 @@ class DRBD8(BaseDRBD):
       - if not, we create it from zero
 
     """
       - if not, we create it from zero
 
     """

-    result = super(DRBD8, self).Assemble()
-    if not result:
-      return result
+    super(DRBD8, self).Assemble()

 
     self.Attach()
     if self.minor is None:
       # local device completely unconfigured
 
     self.Attach()
     if self.minor is None:
       # local device completely unconfigured

-      return self._FastAssemble()
+      self._FastAssemble()

     else:
       # we have to recheck the local and network status and try to fix
       # the device
     else:
       # we have to recheck the local and network status and try to fix
       # the device

-      return self._SlowAssemble()
+      self._SlowAssemble()

 
   def _SlowAssemble(self):
     """Assembles the DRBD device from a (partially) configured device.
 
   def _SlowAssemble(self):
     """Assembles the DRBD device from a (partially) configured device.

@@ -1375,27 +1611,37 @@ class DRBD8(BaseDRBD):
     the attach if can return success.
 
     """
     the attach if can return success.
 
     """

+    # TODO: Rewrite to not use a for loop just because there is 'break'
+    # pylint: disable-msg=W0631
+    net_data = (self._lhost, self._lport, self._rhost, self._rport)

     for minor in (self._aminor,):
       info = self._GetDevInfo(self._GetShowData(minor))
       match_l = self._MatchesLocal(info)
       match_r = self._MatchesNet(info)
     for minor in (self._aminor,):
       info = self._GetDevInfo(self._GetShowData(minor))
       match_l = self._MatchesLocal(info)
       match_r = self._MatchesNet(info)

+

       if match_l and match_r:
       if match_l and match_r:

+        # everything matches

         break
         break

+

       if match_l and not match_r and "local_addr" not in info:
       if match_l and not match_r and "local_addr" not in info:

-        res_r = self._AssembleNet(minor,
-                                  (self._lhost, self._lport,
-                                   self._rhost, self._rport),
-                                  constants.DRBD_NET_PROTOCOL,
-                                  hmac=constants.DRBD_HMAC_ALG,
-                                  secret=self._secret
-                                  )
-        if res_r:
-          if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))):
-            break
-      # the weakest case: we find something that is only net attached
-      # even though we were passed some children at init time
+        # disk matches, but not attached to network, attach and recheck
+        self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
+                          hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
+        if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))):
+          break
+        else:
+          _ThrowError("drbd%d: network attach successful, but 'drbdsetup"
+                      " show' disagrees", minor)
+

       if match_r and "local_dev" not in info:
       if match_r and "local_dev" not in info:

-        break
+        # no local disk, but network attached and it matches
+        self._AssembleLocal(minor, self._children[0].dev_path,
+                            self._children[1].dev_path, self.size)
+        if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))):
+          break
+        else:
+          _ThrowError("drbd%d: disk attach successful, but 'drbdsetup"
+                      " show' disagrees", minor)

 
       # this case must be considered only if we actually have local
       # storage, i.e. not in diskless mode, because all diskless
 
       # this case must be considered only if we actually have local
       # storage, i.e. not in diskless mode, because all diskless

@@ -1407,27 +1653,30 @@ class DRBD8(BaseDRBD):
         # else, even though its local storage is ours; as we own the
         # drbd space, we try to disconnect from the remote peer and
         # reconnect to our correct one
         # else, even though its local storage is ours; as we own the
         # drbd space, we try to disconnect from the remote peer and
         # reconnect to our correct one

-        if not self._ShutdownNet(minor):
-          raise errors.BlockDeviceError("Device has correct local storage,"
-                                        " wrong remote peer and is unable to"
-                                        " disconnect in order to attach to"
-                                        " the correct peer")
+        try:
+          self._ShutdownNet(minor)
+        except errors.BlockDeviceError, err:
+          _ThrowError("drbd%d: device has correct local storage, wrong"
+                      " remote peer and is unable to disconnect in order"
+                      " to attach to the correct peer: %s", minor, str(err))

         # note: _AssembleNet also handles the case when we don't want
         # local storage (i.e. one or more of the _[lr](host|port) is
         # None)
         # note: _AssembleNet also handles the case when we don't want
         # local storage (i.e. one or more of the _[lr](host|port) is
         # None)

-        if (self._AssembleNet(minor, (self._lhost, self._lport,
-                                      self._rhost, self._rport),
-                              constants.DRBD_NET_PROTOCOL,
-                              hmac=constants.DRBD_HMAC_ALG,
-                              secret=self._secret) and
-            self._MatchesNet(self._GetDevInfo(self._GetShowData(minor)))):
+        self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
+                          hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
+        if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))):

           break
           break

+        else:
+          _ThrowError("drbd%d: network attach successful, but 'drbdsetup"
+                      " show' disagrees", minor)

 
     else:
       minor = None
 
     self._SetFromMinor(minor)
 
     else:
       minor = None
 
     self._SetFromMinor(minor)

-    return minor is not None
+    if minor is None:
+      _ThrowError("drbd%d: cannot activate, unknown or unhandled reason",
+                  self._aminor)

 
   def _FastAssemble(self):
     """Assemble the drbd device from zero.
 
   def _FastAssemble(self):
     """Assemble the drbd device from zero.

@@ -1435,26 +1684,16 @@ class DRBD8(BaseDRBD):
     This is run when in Assemble we detect our minor is unused.
 
     """
     This is run when in Assemble we detect our minor is unused.
 
     """

-    # TODO: maybe completely tear-down the minor (drbdsetup ... down)
-    # before attaching our own?

     minor = self._aminor
     minor = self._aminor

-    need_localdev_teardown = False

     if self._children and self._children[0] and self._children[1]:
     if self._children and self._children[0] and self._children[1]:

-      result = self._AssembleLocal(minor, self._children[0].dev_path,
-                                   self._children[1].dev_path)
-      if not result:
-        return False
+      self._AssembleLocal(minor, self._children[0].dev_path,
+                          self._children[1].dev_path, self.size)

     if self._lhost and self._lport and self._rhost and self._rport:
     if self._lhost and self._lport and self._rhost and self._rport:

-      result = self._AssembleNet(minor,
-                                 (self._lhost, self._lport,
-                                  self._rhost, self._rport),
-                                 constants.DRBD_NET_PROTOCOL,
-                                 hmac=constants.DRBD_HMAC_ALG,
-                                 secret=self._secret)
-      if not result:
-        return False
+      self._AssembleNet(minor,
+                        (self._lhost, self._lport, self._rhost, self._rport),
+                        constants.DRBD_NET_PROTOCOL,
+                        hmac=constants.DRBD_HMAC_ALG, secret=self._secret)

     self._SetFromMinor(minor)
     self._SetFromMinor(minor)

-    return True

 
   @classmethod
   def _ShutdownLocal(cls, minor):
 
   @classmethod
   def _ShutdownLocal(cls, minor):

@@ -1466,8 +1705,7 @@ class DRBD8(BaseDRBD):
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"])
     if result.failed:
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"])
     if result.failed:

-      logging.error("Can't detach local device: %s", result.output)
-    return not result.failed
+      _ThrowError("drbd%d: can't detach local disk: %s", minor, result.output)

 
   @classmethod
   def _ShutdownNet(cls, minor):
 
   @classmethod
   def _ShutdownNet(cls, minor):

@@ -1478,8 +1716,7 @@ class DRBD8(BaseDRBD):
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"])
     if result.failed:
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"])
     if result.failed:

-      logging.error("Can't shutdown network: %s", result.output)
-    return not result.failed
+      _ThrowError("drbd%d: can't shutdown network: %s", minor, result.output)

 
   @classmethod
   def _ShutdownAll(cls, minor):
 
   @classmethod
   def _ShutdownAll(cls, minor):

@@ -1490,7 +1727,8 @@ class DRBD8(BaseDRBD):
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"])
     if result.failed:
     """
     result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"])
     if result.failed:

-      _ThrowError("Can't shutdown drbd device: %s", result.output)
+      _ThrowError("drbd%d: can't shutdown drbd device: %s",
+                  minor, result.output)

 
   def Shutdown(self):
     """Shutdown the DRBD device.
 
   def Shutdown(self):
     """Shutdown the DRBD device.

@@ -1529,15 +1767,15 @@ class DRBD8(BaseDRBD):
     else:
       in_use = False
     if in_use:
     else:
       in_use = False
     if in_use:

-      _ThrowError("DRBD minor %d already in use at Create() time", aminor)
+      _ThrowError("drbd%d: minor is already in use at Create() time", aminor)

     meta = children[1]
     meta.Assemble()
     if not meta.Attach():
     meta = children[1]
     meta.Assemble()
     if not meta.Attach():

-      raise errors.BlockDeviceError("Can't attach to meta device")
-    if not cls._CheckMetaSize(meta.dev_path):
-      raise errors.BlockDeviceError("Invalid meta device size")
+      _ThrowError("drbd%d: can't attach to meta device '%s'",
+                  aminor, meta)
+    cls._CheckMetaSize(meta.dev_path)

     cls._InitMeta(aminor, meta.dev_path)
     cls._InitMeta(aminor, meta.dev_path)

-    return cls(unique_id, children)
+    return cls(unique_id, children, size)

 
   def Grow(self, amount):
     """Resize the DRBD device and its backing storage.
 
   def Grow(self, amount):
     """Resize the DRBD device and its backing storage.

@@ -1548,7 +1786,8 @@ class DRBD8(BaseDRBD):
     if len(self._children) != 2 or None in self._children:
       _ThrowError("drbd%d: cannot grow diskless device", self.minor)
     self._children[0].Grow(amount)
     if len(self._children) != 2 or None in self._children:
       _ThrowError("drbd%d: cannot grow diskless device", self.minor)
     self._children[0].Grow(amount)

-    result = utils.RunCmd(["drbdsetup", self.dev_path, "resize"])
+    result = utils.RunCmd(["drbdsetup", self.dev_path, "resize", "-s",
+                           "%dm" % (self.size + amount)])

     if result.failed:
       _ThrowError("drbd%d: resize failed: %s", self.minor, result.output)
 
     if result.failed:
       _ThrowError("drbd%d: resize failed: %s", self.minor, result.output)
 

@@ -1561,13 +1800,13 @@ class FileStorage(BlockDev):
   The unique_id for the file device is a (file_driver, file_path) tuple.
 
   """
   The unique_id for the file device is a (file_driver, file_path) tuple.
 
   """

-  def __init__(self, unique_id, children):
+  def __init__(self, unique_id, children, size):

     """Initalizes a file device backend.
 
     """
     if children:
       raise errors.BlockDeviceError("Invalid setup for file device")
     """Initalizes a file device backend.
 
     """
     if children:
       raise errors.BlockDeviceError("Invalid setup for file device")

-    super(FileStorage, self).__init__(unique_id, children)
+    super(FileStorage, self).__init__(unique_id, children, size)

     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     self.driver = unique_id[0]
     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     self.driver = unique_id[0]

@@ -1581,14 +1820,12 @@ class FileStorage(BlockDev):
 
     """
     if not os.path.exists(self.dev_path):
 
     """
     if not os.path.exists(self.dev_path):

-      raise errors.BlockDeviceError("File device '%s' does not exist." %
-                                    self.dev_path)
-    return True
+      _ThrowError("File device '%s' does not exist" % self.dev_path)

 
   def Shutdown(self):
     """Shutdown the device.
 
 
   def Shutdown(self):
     """Shutdown the device.
 

-    This is a no-op for the file type, as we don't deacivate
+    This is a no-op for the file type, as we don't deactivate

     the file on shutdown.
 
     """
     the file on shutdown.
 
     """

@@ -1623,6 +1860,22 @@ class FileStorage(BlockDev):
       if err.errno != errno.ENOENT:
         _ThrowError("Can't remove file '%s': %s", self.dev_path, err)
 
       if err.errno != errno.ENOENT:
         _ThrowError("Can't remove file '%s': %s", self.dev_path, err)
 

+  def Rename(self, new_id):
+    """Renames the file.
+
+    """
+    # TODO: implement rename for file-based storage
+    _ThrowError("Rename is not supported for file-based storage")
+
+  def Grow(self, amount):
+    """Grow the file
+
+    @param amount: the amount (in mebibytes) to grow with
+
+    """
+    # TODO: implement grow for file-based storage
+    _ThrowError("Grow not supported for file-based storage")
+

   def Attach(self):
     """Attach to an existing file.
 
   def Attach(self):
     """Attach to an existing file.
 

@@ -1635,6 +1888,19 @@ class FileStorage(BlockDev):
     self.attached = os.path.exists(self.dev_path)
     return self.attached
 
     self.attached = os.path.exists(self.dev_path)
     return self.attached
 

+  def GetActualSize(self):
+    """Return the actual disk size.
+
+    @note: the device needs to be active when this is called
+
+    """
+    assert self.attached, "BlockDevice not attached in GetActualSize()"
+    try:
+      st = os.stat(self.dev_path)
+      return st.st_size
+    except OSError, err:
+      _ThrowError("Can't stat %s: %s", self.dev_path, err)
+

   @classmethod
   def Create(cls, unique_id, children, size):
     """Create a new file.
   @classmethod
   def Create(cls, unique_id, children, size):
     """Create a new file.

@@ -1645,11 +1911,11 @@ class FileStorage(BlockDev):
     @return: an instance of FileStorage
 
     """
     @return: an instance of FileStorage
 
     """

-    # TODO: decide whether we should check for existing files and
-    # abort or not

     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     dev_path = unique_id[1]
     if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:
       raise ValueError("Invalid configuration data %s" % str(unique_id))
     dev_path = unique_id[1]

+    if os.path.exists(dev_path):
+      _ThrowError("File already existing: %s", dev_path)

     try:
       f = open(dev_path, 'w')
       f.truncate(size * 1024 * 1024)
     try:
       f = open(dev_path, 'w')
       f.truncate(size * 1024 * 1024)

@@ -1657,17 +1923,19 @@ class FileStorage(BlockDev):
     except IOError, err:
       _ThrowError("Error in file creation: %", str(err))
 
     except IOError, err:
       _ThrowError("Error in file creation: %", str(err))
 

-    return FileStorage(unique_id, children)
+    return FileStorage(unique_id, children, size)

 
 
 DEV_MAP = {
   constants.LD_LV: LogicalVolume,
   constants.LD_DRBD8: DRBD8,
 
 
 DEV_MAP = {
   constants.LD_LV: LogicalVolume,
   constants.LD_DRBD8: DRBD8,

-  constants.LD_FILE: FileStorage,

   }
 
   }
 

+if constants.ENABLE_FILE_STORAGE:
+  DEV_MAP[constants.LD_FILE] = FileStorage
+

 
 

-def FindDevice(dev_type, unique_id, children):
+def FindDevice(dev_type, unique_id, children, size):

   """Search for an existing, assembled device.
 
   This will succeed only if the device exists and is assembled, but it
   """Search for an existing, assembled device.
 
   This will succeed only if the device exists and is assembled, but it

@@ -1676,13 +1944,13 @@ def FindDevice(dev_type, unique_id, children):
   """
   if dev_type not in DEV_MAP:
     raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type)
   """
   if dev_type not in DEV_MAP:
     raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type)

-  device = DEV_MAP[dev_type](unique_id, children)
+  device = DEV_MAP[dev_type](unique_id, children, size)

   if not device.attached:
     return None
   return device
 
 
   if not device.attached:
     return None
   return device
 
 

-def Assemble(dev_type, unique_id, children):
+def Assemble(dev_type, unique_id, children, size):

   """Try to attach or assemble an existing device.
 
   This will attach to assemble the device, as needed, to bring it
   """Try to attach or assemble an existing device.
 
   This will attach to assemble the device, as needed, to bring it

@@ -1691,11 +1959,8 @@ def Assemble(dev_type, unique_id, children):
   """
   if dev_type not in DEV_MAP:
     raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type)
   """
   if dev_type not in DEV_MAP:
     raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type)

-  device = DEV_MAP[dev_type](unique_id, children)
-  if not device.Assemble():
-    raise errors.BlockDeviceError("Can't find a valid block device for"
-                                  " %s/%s/%s" %
-                                  (dev_type, unique_id, children))
+  device = DEV_MAP[dev_type](unique_id, children, size)
+  device.Assemble()

   return device
 
 
   return device