Synnefo » snf-ganeti

#

#

# Copyright (C) 2006, 2007, 2010, 2011, 2012 Google Inc.

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful, but

# WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

# General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

# 02110-1301, USA.

"""Block device abstraction"""

import re

import time

import errno

import shlex

import stat

import pyparsing as pyp

import os

import logging

import math

from ganeti import utils

from ganeti import errors

from ganeti import constants

from ganeti import objects

from ganeti import compat

from ganeti import netutils

from ganeti import pathutils

# Size of reads in _CanReadDevice

_DEVICE_READ_SIZE = 128 * 1024

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

  """Executes the given function, ignoring BlockDeviceErrors.

  This is used in order to simplify the execution of cleanup or

  rollback functions.

  @rtype: boolean

  @return: True when fn didn't raise an exception, False otherwise

  """

  try:

    fn(*args, **kwargs)

    return True

  except errors.BlockDeviceError, err:

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

    return False

def _ThrowError(msg, *args):

  """Log an error to the node daemon and the raise an exception.

  @type msg: string

  @param msg: the text of the exception

  @raise errors.BlockDeviceError

  """

  if args:

    msg = msg % args

  logging.error(msg)

  raise errors.BlockDeviceError(msg)

def _CheckResult(result):

  """Throws an error if the given result is a failed one.

  @param result: result from RunCmd

  """

  if result.failed:

    _ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason,

                result.output)

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

def _GetForbiddenFileStoragePaths():

  """Builds a list of path prefixes which shouldn't be used for file storage.

  @rtype: frozenset

  """

  paths = set([

    "/boot",

    "/dev",

    "/etc",

    "/home",

    "/proc",

    "/root",

    "/sys",

    ])

  for prefix in ["", "/usr", "/usr/local"]:

    paths.update(map(lambda s: "%s/%s" % (prefix, s),

                     ["bin", "lib", "lib32", "lib64", "sbin"]))

  return compat.UniqueFrozenset(map(os.path.normpath, paths))

def _ComputeWrongFileStoragePaths(paths,

                                  _forbidden=_GetForbiddenFileStoragePaths()):

  """Cross-checks a list of paths for prefixes considered bad.

  Some paths, e.g. "/bin", should not be used for file storage.

  @type paths: list

  @param paths: List of paths to be checked

  @rtype: list

  @return: Sorted list of paths for which the user should be warned

  """

  def _Check(path):

    return (not os.path.isabs(path) or

            path in _forbidden or

            filter(lambda p: utils.IsBelowDir(p, path), _forbidden))

  return utils.NiceSort(filter(_Check, map(os.path.normpath, paths)))

def ComputeWrongFileStoragePaths(_filename=pathutils.FILE_STORAGE_PATHS_FILE):

  """Returns a list of file storage paths whose prefix is considered bad.

  See L{_ComputeWrongFileStoragePaths}.

  """

  return _ComputeWrongFileStoragePaths(_LoadAllowedFileStoragePaths(_filename))

def _CheckFileStoragePath(path, allowed):

  """Checks if a path is in a list of allowed paths for file storage.

  @type path: string

  @param path: Path to check

  @type allowed: list

  @param allowed: List of allowed paths

  @raise errors.FileStoragePathError: If the path is not allowed

  """

  if not os.path.isabs(path):

    raise errors.FileStoragePathError("File storage path must be absolute,"

                                      " got '%s'" % path)

  for i in allowed:

    if not os.path.isabs(i):

      logging.info("Ignoring relative path '%s' for file storage", i)

      continue

    if utils.IsBelowDir(i, path):

      break

  else:

    raise errors.FileStoragePathError("Path '%s' is not acceptable for file"

                                      " storage" % path)

def _LoadAllowedFileStoragePaths(filename):

  """Loads file containing allowed file storage paths.

  @rtype: list

  @return: List of allowed paths (can be an empty list)

  """

  try:

    contents = utils.ReadFile(filename)

  except EnvironmentError:

    return []

  else:

    return utils.FilterEmptyLinesAndComments(contents)

def CheckFileStoragePath(path, _filename=pathutils.FILE_STORAGE_PATHS_FILE):

  """Checks if a path is allowed for file storage.

  @type path: string

  @param path: Path to check

  @raise errors.FileStoragePathError: If the path is not allowed

  """

  allowed = _LoadAllowedFileStoragePaths(_filename)

  if _ComputeWrongFileStoragePaths([path]):

    raise errors.FileStoragePathError("Path '%s' uses a forbidden prefix" %

                                      path)

  _CheckFileStoragePath(path, allowed)

class BlockDev(object):

  """Block device abstract class.

  A block device can be in the following states:

    - not existing on the system, and by `Create()` it goes into:

    - existing but not setup/not active, and by `Assemble()` goes into:

    - active read-write and by `Open()` it goes into

    - online (=used, or ready for use)

  A device can also be online but read-only, however we are not using

  the readonly state (LV has it, if needed in the future) and we are

  usually looking at this like at a stack, so it's easier to

  conceptualise the transition from not-existing to online and back

  like a linear one.

  The many different states of the device are due to the fact that we

  need to cover many device types:

    - logical volumes are created, lvchange -a y $lv, and used

    - drbd devices are attached to a local disk/remote peer and made primary

  A block device is identified by three items:

    - the /dev path of the device (dynamic)

    - a unique ID of the device (static)

    - it's major/minor pair (dynamic)

  Not all devices implement both the first two as distinct items. LVM

  logical volumes have their unique ID (the pair volume group, logical

  volume name) in a 1-to-1 relation to the dev path. For DRBD devices,

  the /dev path is again dynamic and the unique id is the pair (host1,

  dev1), (host2, dev2).

  You can get to a device in two ways:

    - creating the (real) device, which returns you

      an attached instance (lvcreate)

    - attaching of a python instance to an existing (real) device

  The second point, the attachement to a device, is different

  depending on whether the device is assembled or not. At init() time,

  we search for a device with the same unique_id as us. If found,

  good. It also means that the device is already assembled. If not,

  after assembly we'll have our correct major/minor.

  """

  def __init__(self, unique_id, children, size, params):

    self._children = children

    self.dev_path = None

    self.unique_id = unique_id

    self.major = None

    self.minor = None

    self.attached = False

    self.size = size

    self.params = params

  def Assemble(self):

    """Assemble the device from its components.

    Implementations of this method by child classes must ensure that:

      - after the device has been assembled, it knows its major/minor

        numbers; this allows other devices (usually parents) to probe

        correctly for their children

      - calling this method on an existing, in-use device is safe

      - if the device is already configured (and in an OK state),

        this method is idempotent

    """

    pass

  def Attach(self):

    """Find a device which matches our config and attach to it.

    """

    raise NotImplementedError

  def Close(self):

    """Notifies that the device will no longer be used for I/O.

    """

    raise NotImplementedError

  @classmethod

  def Create(cls, unique_id, children, size, params, excl_stor):

    """Create the device.

    If the device cannot be created, it will return None

    instead. Error messages go to the logging system.

    Note that for some devices, the unique_id is used, and for other,

    the children. The idea is that these two, taken together, are

    enough for both creation and assembly (later).

    """

    raise NotImplementedError

  def Remove(self):

    """Remove this device.

    This makes sense only for some of the device types: LV and file

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

    can't be completed.

    """

    raise NotImplementedError

  def Rename(self, new_id):

    """Rename this device.

    This may or may not make sense for a given device type.

    """

    raise NotImplementedError

  def Open(self, force=False):

    """Make the device ready for use.

    This makes the device ready for I/O. For now, just the DRBD

    devices need this.

    The force parameter signifies that if the device has any kind of

    --force thing, it should be used, we know what we are doing.

    """

    raise NotImplementedError

  def Shutdown(self):

    """Shut down the device, freeing its children.

    This undoes the `Assemble()` work, except for the child

    assembling; as such, the children on the device are still

    assembled after this call.

    """

    raise NotImplementedError

  def SetSyncParams(self, params):

    """Adjust the synchronization parameters of the mirror.

    In case this is not a mirroring device, this is no-op.

    @param params: dictionary of LD level disk parameters related to the

    synchronization.

    @rtype: list

    @return: a list of error messages, emitted both by the current node and by

    children. An empty list means no errors.

    """

    result = []

    if self._children:

      for child in self._children:

        result.extend(child.SetSyncParams(params))

    return result

  def PauseResumeSync(self, pause):

    """Pause/Resume the sync of the mirror.

    In case this is not a mirroring device, this is no-op.

    @param pause: Whether to pause or resume

    """

    result = True

    if self._children:

      for child in self._children:

        result = result and child.PauseResumeSync(pause)

    return result

  def GetSyncStatus(self):

    """Returns the sync status of the device.

    If this device is a mirroring device, this function returns the

    status of the mirror.

    If sync_percent is None, it means the device is not syncing.

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

    the time needed, otherwise it's the time left in seconds.

    If is_degraded is True, it means the device is missing

    redundancy. This is usually a sign that something went wrong in

    the device setup, if sync_percent is None.

    The ldisk parameter represents the degradation of the local

    data. This is only valid for some devices, the rest will always

    return False (not degraded).

    @rtype: objects.BlockDevStatus

    """

    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.

    The return value is the same as for `GetSyncStatus()` except the

    minimum percent and maximum time are calculated across our

    children.

    @rtype: objects.BlockDevStatus

    """

    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:

        child_status = child.GetSyncStatus()

        if min_percent is None:

          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:

          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):

    """Update metadata with info text.

    Only supported for some device types.

    """

    for child in self._children:

      child.SetInfo(text)

  def Grow(self, amount, dryrun, backingstore):

    """Grow the block device.

    @type amount: integer

    @param amount: the amount (in mebibytes) to grow with

    @type dryrun: boolean

    @param dryrun: whether to execute the operation in simulation mode

        only, without actually increasing the size

    @param backingstore: whether to execute the operation on backing storage

        only, or on "logical" storage only; e.g. DRBD is logical storage,

        whereas LVM, file, RBD are backing storage

    """

    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,

             self.major, self.minor, self.dev_path))

class LogicalVolume(BlockDev):

  """Logical Volume block device.

  """

  _VALID_NAME_RE = re.compile("^[a-zA-Z0-9+_.-]*$")

  _INVALID_NAMES = compat.UniqueFrozenset([".", "..", "snapshot", "pvmove"])

  _INVALID_SUBSTRINGS = compat.UniqueFrozenset(["_mlog", "_mimage"])

  def __init__(self, unique_id, children, size, params):

    """Attaches to a LV device.

    The unique_id is a tuple (vg_name, lv_name)

    """

    super(LogicalVolume, self).__init__(unique_id, children, size, params)

    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._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.major = self.minor = self.pe_size = self.stripe_count = None

    self.Attach()

  @staticmethod

  def _GetStdPvSize(pvs_info):

    """Return the the standard PV size (used with exclusive storage).

    @param pvs_info: list of objects.LvmPvInfo, cannot be empty

    @rtype: float

    @return: size in MiB

    """

    assert len(pvs_info) > 0

    smallest = min([pv.size for pv in pvs_info])

    return smallest / (1 + constants.PART_MARGIN + constants.PART_RESERVED)

  @staticmethod

  def _ComputeNumPvs(size, pvs_info):

    """Compute the number of PVs needed for an LV (with exclusive storage).

    @type size: float

    @param size: LV size in MiB

    @param pvs_info: list of objects.LvmPvInfo, cannot be empty

    @rtype: integer

    @return: number of PVs needed

    """

    assert len(pvs_info) > 0

    pv_size = float(LogicalVolume._GetStdPvSize(pvs_info))

    return int(math.ceil(float(size) / pv_size))

  @staticmethod

  def _GetEmptyPvNames(pvs_info, max_pvs=None):

    """Return a list of empty PVs, by name.

    """

    empty_pvs = filter(objects.LvmPvInfo.IsEmpty, pvs_info)

    if max_pvs is not None:

      empty_pvs = empty_pvs[:max_pvs]

    return map((lambda pv: pv.name), empty_pvs)

  @classmethod

  def Create(cls, unique_id, children, size, params, excl_stor):

    """Create a new logical volume.

    """

    if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2:

      raise errors.ProgrammerError("Invalid configuration data %s" %

                                   str(unique_id))

    vg_name, lv_name = unique_id

    cls._ValidateName(vg_name)

    cls._ValidateName(lv_name)

    pvs_info = cls.GetPVInfo([vg_name])

    if not pvs_info:

      if excl_stor:

        msg = "No (empty) PVs found"

      else:

        msg = "Can't compute PV info for vg %s" % vg_name

      _ThrowError(msg)

    pvs_info.sort(key=(lambda pv: pv.free), reverse=True)

    pvlist = [pv.name for pv in pvs_info]

    if compat.any(":" in v for v in pvlist):

      _ThrowError("Some of your PVs have the invalid character ':' in their"

                  " name, this is not supported - please filter them out"

                  " in lvm.conf using either 'filter' or 'preferred_names'")

    current_pvs = len(pvlist)

    desired_stripes = params[constants.LDP_STRIPES]

    stripes = min(current_pvs, desired_stripes)

    if excl_stor:

      (err_msgs, _) = utils.LvmExclusiveCheckNodePvs(pvs_info)

      if err_msgs:

        for m in err_msgs:

          logging.warning(m)

      req_pvs = cls._ComputeNumPvs(size, pvs_info)

      pvlist = cls._GetEmptyPvNames(pvs_info, req_pvs)

      current_pvs = len(pvlist)

      if current_pvs < req_pvs:

        _ThrowError("Not enough empty PVs to create a disk of %d MB:"

                    " %d available, %d needed", size, current_pvs, req_pvs)

      assert current_pvs == len(pvlist)

      if stripes > current_pvs:

        # No warning issued for this, as it's no surprise

        stripes = current_pvs

    else:

      if stripes < desired_stripes:

        logging.warning("Could not use %d stripes for VG %s, as only %d PVs are"

                        " available.", desired_stripes, vg_name, current_pvs)

      free_size = sum([pv.free for pv in pvs_info])

      # 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)

    # 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

    cmd = ["lvcreate", "-L%dm" % size, "-n%s" % lv_name]

    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)

    return LogicalVolume(unique_id, children, size, params)

  @staticmethod

  def _GetVolumeInfo(lvm_cmd, fields):

    """Returns LVM Volumen infos using lvm_cmd

    @param lvm_cmd: Should be one of "pvs", "vgs" or "lvs"

    @param fields: Fields to return

    @return: A list of dicts each with the parsed fields

    """

    if not fields:

      raise errors.ProgrammerError("No fields specified")

    sep = "|"

    cmd = [lvm_cmd, "--noheadings", "--nosuffix", "--units=m", "--unbuffered",

           "--separator=%s" % sep, "-o%s" % ",".join(fields)]

    result = utils.RunCmd(cmd)

    if result.failed:

      raise errors.CommandError("Can't get the volume information: %s - %s" %

                                (result.fail_reason, result.output))

    data = []

    for line in result.stdout.splitlines():

      splitted_fields = line.strip().split(sep)

      if len(fields) != len(splitted_fields):

        raise errors.CommandError("Can't parse %s output: line '%s'" %

                                  (lvm_cmd, line))

      data.append(splitted_fields)

    return data

  @classmethod

  def GetPVInfo(cls, vg_names, filter_allocatable=True, include_lvs=False):

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

    @param vg_names: list of volume group names, if empty all will be returned

    @param filter_allocatable: whether to skip over unallocatable PVs

    @param include_lvs: whether to include a list of LVs hosted on each PV

    @rtype: list

    @return: list of objects.LvmPvInfo objects

    """

    # We request "lv_name" field only if we care about LVs, so we don't get

    # a long list of entries with many duplicates unless we really have to.

    # The duplicate "pv_name" field will be ignored.

    if include_lvs:

      lvfield = "lv_name"

    else:

      lvfield = "pv_name"

    try:

      info = cls._GetVolumeInfo("pvs", ["pv_name", "vg_name", "pv_free",

                                        "pv_attr", "pv_size", lvfield])

    except errors.GenericError, err:

      logging.error("Can't get PV information: %s", err)

      return None

    # When asked for LVs, "pvs" may return multiple entries for the same PV-LV

    # pair. We sort entries by PV name and then LV name, so it's easy to weed

    # out duplicates.

    if include_lvs:

      info.sort(key=(lambda i: (i[0], i[5])))

    data = []

    lastpvi = None

    for (pv_name, vg_name, pv_free, pv_attr, pv_size, lv_name) in info:

      # (possibly) skip over pvs which are not allocatable

      if filter_allocatable and pv_attr[0] != "a":

        continue

      # (possibly) skip over pvs which are not in the right volume group(s)

      if vg_names and vg_name not in vg_names:

        continue

      # Beware of duplicates (check before inserting)

      if lastpvi and lastpvi.name == pv_name:

        if include_lvs and lv_name:

          if not lastpvi.lv_list or lastpvi.lv_list[-1] != lv_name:

            lastpvi.lv_list.append(lv_name)

      else:

        if include_lvs and lv_name:

          lvl = [lv_name]

        else:

          lvl = []

        lastpvi = objects.LvmPvInfo(name=pv_name, vg_name=vg_name,

                                    size=float(pv_size), free=float(pv_free),

                                    attributes=pv_attr, lv_list=lvl)

        data.append(lastpvi)

    return data

  @classmethod

  def _GetExclusiveStorageVgFree(cls, vg_name):

    """Return the free disk space in the given VG, in exclusive storage mode.

    @type vg_name: string