X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/5b7b5d49a689c7023aa37b989c60c4483dbc230a..95990bc55368300f63d9289b1b362713eb05976f:/lib/bdev.py diff --git a/lib/bdev.py b/lib/bdev.py index 7409493..0f9339e 100644 --- a/lib/bdev.py +++ b/lib/bdev.py @@ -1,7 +1,7 @@ -#!/usr/bin/python +# # -# Copyright (C) 2006, 2007 Google Inc. +# Copyright (C) 2006, 2007, 2010, 2011, 2012, 2013 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 @@ -24,10 +24,199 @@ 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 logger from ganeti import errors +from ganeti import constants +from ganeti import objects +from ganeti import compat +from ganeti import netutils +from ganeti import pathutils +from ganeti import serializer + + +# Size of reads in _CanReadDevice +_DEVICE_READ_SIZE = 128 * 1024 + + +class RbdShowmappedJsonError(Exception): + """`rbd showmmapped' JSON formatting error Exception class. + + """ + pass + + +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): @@ -40,21 +229,16 @@ class BlockDev(object): - online (=used, or ready for use) A device can also be online but read-only, however we are not using - the readonly state (MD and LV have 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 + 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 - - md arrays are created or assembled and used - drbd devices are attached to a local disk/remote peer and made primary - The status of the device can be examined by `GetStatus()`, which - returns a numerical value, depending on the position in the - transition stack of the device. - A block device is identified by three items: - the /dev path of the device (dynamic) - a unique ID of the device (static) @@ -62,74 +246,45 @@ class BlockDev(object): 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 MD devices, - the /dev path is dynamic and the unique ID is the UUID generated at - array creation plus the slave list. For DRBD devices, the /dev path - is again dynamic and the unique id is the pair (host1, dev1), - (host2, dev2). + 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, mdadm --create) + an attached instance (lvcreate) - attaching of a python instance to an existing (real) device - The second point, the attachement to a device, is different + The second point, the attachment 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. """ - - STATUS_UNKNOWN = 0 - STATUS_EXISTING = 1 - STATUS_STANDBY = 2 - STATUS_ONLINE = 3 - - STATUS_MAP = { - STATUS_UNKNOWN: "unknown", - STATUS_EXISTING: "existing", - STATUS_STANDBY: "ready for use", - STATUS_ONLINE: "online", - } - - - def __init__(self, unique_id, children): + 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. - If this is a plain block device (e.g. LVM) than assemble does - nothing, as the LVM has no children and we don't put logical - volumes offline. - - One guarantee is 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. + 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 """ - status = True - for child in self._children: - if not isinstance(child, BlockDev): - raise TypeError("Invalid child passed of type '%s'" % type(child)) - if not status: - break - status = status and child.Assemble() - if not status: - break - status = status and child.Open() - - if not status: - for child in self._children: - child.Shutdown() - return status - + pass def Attach(self): """Find a device which matches our config and attach to it. @@ -137,16 +292,14 @@ class BlockDev(object): """ 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): + def Create(cls, unique_id, children, size, params, excl_stor): """Create the device. If the device cannot be created, it will return None @@ -159,25 +312,24 @@ class BlockDev(object): """ raise NotImplementedError - def Remove(self): """Remove this device. - This makes sense only for some of the device types: LV and to a - lesser degree, md devices. Also note that if the device can't - attach, the removal can't be completed. + 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. - def GetStatus(self): - """Return the status of the 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. @@ -190,7 +342,6 @@ class BlockDev(object): """ raise NotImplementedError - def Shutdown(self): """Shut down the device, freeing its children. @@ -201,39 +352,67 @@ class BlockDev(object): """ 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 SetSyncSpeed(self, speed): - """Adjust the sync speed of the mirror. + 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.SetSyncSpeed(speed) + 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. - Returns: - (sync_percent, estimated_time, is_degraded) + If sync_percent is None, it means the device is not syncing. - If sync_percent is None, it means all is ok If estimated_time is None, it means we can't estimate - the time needed, otherwise it's the time left in seconds + 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. - """ - return None, None, False + 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. @@ -242,22 +421,85 @@ class BlockDev(object): minimum percent and maximum time are calculated across our children. + @rtype: objects.BlockDevStatus + """ - min_percent, max_time, is_degraded = 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: - c_percent, c_time, c_degraded = child.GetSyncStatus() + child_status = child.GetSyncStatus() + 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: - max_time = c_time - elif c_time is not None: - max_time = max(max_time, c_time) - is_degraded = is_degraded or c_degraded - return min_percent, max_time, is_degraded + 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>" % @@ -269,126 +511,397 @@ class LogicalVolume(BlockDev): """Logical Volume block device. """ - def __init__(self, unique_id, children): + _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) + 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.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.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): + 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 ValueError("Invalid configuration data %s" % str(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: - raise errors.BlockDeviceError, ("Can't compute PV info for vg %s" % - vg_name) - pvs_info.sort() - pvs_info.reverse() - - pvlist = [ pv[1] for pv in pvs_info ] - free_size = sum([ pv[0] for pv in 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 - # The size constraint should have been checked from the master before - # calling the create function. - if free_size < size: - raise errors.BlockDeviceError, ("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) + 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: - raise errors.BlockDeviceError(result.fail_reason) - return LogicalVolume(unique_id, children) + _ThrowError("LV create failed (%s): %s", + result.fail_reason, result.output) + return LogicalVolume(unique_id, children, size, params) @staticmethod - def GetPVInfo(vg_name): - """Get the free space info for PVs in a volume group. + def _GetVolumeInfo(lvm_cmd, fields): + """Returns LVM Volume infos using lvm_cmd - Args: - vg_name: the volume group name + @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 - Returns: - list of (free_space, name) with free_space in mebibytes """ - command = ["pvs", "--noheadings", "--nosuffix", "--units=m", - "-opv_name,vg_name,pv_free,pv_attr", "--unbuffered", - "--separator=:"] - result = utils.RunCmd(command) + 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: - logger.Error("Can't get the PV information: %s" % result.fail_reason) - return None + raise errors.CommandError("Can't get the volume information: %s - %s" % + (result.fail_reason, result.output)) + data = [] for line in result.stdout.splitlines(): - fields = line.strip().split(':') - if len(fields) != 4: - logger.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': + 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 + @param vg_name: VG name + @rtype: float + @return: free space in MiB + """ + pvs_info = cls.GetPVInfo([vg_name]) + if not pvs_info: + return 0.0 + pv_size = cls._GetStdPvSize(pvs_info) + num_pvs = len(cls._GetEmptyPvNames(pvs_info)) + return pv_size * num_pvs + + @classmethod + def GetVGInfo(cls, vg_names, excl_stor, filter_readonly=True): + """Get the free space info for specific VGs. + + @param vg_names: list of volume group names, if empty all will be returned + @param excl_stor: whether exclusive_storage is enabled + @param filter_readonly: whether to skip over readonly VGs + + @rtype: list + @return: list of tuples (free_space, total_size, name) with free_space in + MiB + + """ + try: + info = cls._GetVolumeInfo("vgs", ["vg_name", "vg_free", "vg_attr", + "vg_size"]) + except errors.GenericError, err: + logging.error("Can't get VG information: %s", err) + return None + + data = [] + for vg_name, vg_free, vg_attr, vg_size in info: + # (possibly) skip over vgs which are not writable + if filter_readonly and vg_attr[0] == "r": + continue + # (possibly) skip over vgs which are not in the right volume group(s) + if vg_names and vg_name not in vg_names: continue - data.append((float(fields[2]), fields[0])) + # Exclusive storage needs a different concept of free space + if excl_stor: + es_free = cls._GetExclusiveStorageVgFree(vg_name) + assert es_free <= vg_free + vg_free = es_free + data.append((float(vg_free), float(vg_size), vg_name)) 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 + compat.any(substring in name for substring in cls._INVALID_SUBSTRINGS)): + _ThrowError("Invalid LVM name '%s'", name) + def Remove(self): """Remove this logical volume. """ if not self.minor and not self.Attach(): # the LV does not exist - return True + return result = utils.RunCmd(["lvremove", "-f", "%s/%s" % (self._vg_name, self._lv_name)]) if result.failed: - logger.Error("Can't lvremove: %s" % result.fail_reason) + _ThrowError("Can't lvremove: %s - %s", result.fail_reason, result.output) - return not result.failed + def Rename(self, new_id): + """Rename this logical volume. + """ + if not isinstance(new_id, (tuple, list)) or len(new_id) != 2: + raise errors.ProgrammerError("Invalid new logical id '%s'" % new_id) + new_vg, new_name = new_id + if new_vg != self._vg_name: + raise errors.ProgrammerError("Can't move a logical volume across" + " volume groups (from %s to to %s)" % + (self._vg_name, new_vg)) + result = utils.RunCmd(["lvrename", new_vg, 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 = utils.PathJoin("/dev", self._vg_name, self._lv_name) def Attach(self): """Attach to an existing LV. This method will try to see if an existing and active LV exists - which matches the our name. If so, its major/minor will be + which matches our name. If so, its major/minor will be recorded. """ - result = utils.RunCmd(["lvdisplay", self.dev_path]) + self.attached = False + result = utils.RunCmd(["lvs", "--noheadings", "--separator=,", + "--units=k", "--nosuffix", + "-olv_attr,lv_kernel_major,lv_kernel_minor," + "vg_extent_size,stripes", self.dev_path]) if result.failed: - logger.Error("Can't find LV %s: %s" % - (self.dev_path, result.fail_reason)) + logging.error("Can't find LV %s: %s, %s", + self.dev_path, result.fail_reason, result.output) + return False + # 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(",") + if len(out) != 5: + logging.error("Can't parse LVS output, len(%s) != 5", str(out)) + return False + + status, major, minor, pe_size, stripes = out + if len(status) < 6: + logging.error("lvs lv_attr is not at least 6 characters (%s)", status) + return False + + try: + major = int(major) + minor = int(minor) + except (TypeError, ValueError), 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 - match = re.compile("^ *Block device *([0-9]+):([0-9]+).*$") - for line in result.stdout.splitlines(): - match_result = match.match(line) - if match_result: - self.major = int(match_result.group(1)) - self.minor = int(match_result.group(2)) - return True - return False + 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 + return True def Assemble(self): """Assemble the device. - This is a no-op for the LV device type. Eventually, we could - lvchange -ay here if we see that the LV is not active. + 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). """ - return True - + result = utils.RunCmd(["lvchange", "-ay", self.dev_path]) + if result.failed: + _ThrowError("Can't activate lv %s: %s", self.dev_path, result.output) def Shutdown(self): """Shutdown the device. @@ -397,37 +910,42 @@ class LogicalVolume(BlockDev): volumes on shutdown. """ - return True + pass + + 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. + For logical volumes, sync_percent and estimated_time are always + None (no recovery in progress, as we don't handle the mirrored LV + case). The is_degraded parameter is the inverse of the ldisk + parameter. - def GetStatus(self): - """Return the status of the device. + For the ldisk parameter, we check if the logical volume has the + 'virtual' type, which means it's not backed by existing storage + anymore (read from it return I/O error). This happens after a + physical disk failure and subsequent 'vgreduce --removemissing' on + the volume group. - Logical volumes will can be in all four states, although we don't - deactivate (lvchange -an) them when shutdown, so STATUS_EXISTING - should not be seen for our devices. + The status was already read in Attach, so we just return it. + + @rtype: objects.BlockDevStatus """ - result = utils.RunCmd(["lvs", "--noheadings", "-olv_attr", self.dev_path]) - if result.failed: - logger.Error("Can't display lv: %s" % result.fail_reason) - return self.STATUS_UNKNOWN - out = result.stdout.strip() - # format: type/permissions/alloc/fixed_minor/state/open - if len(out) != 6: - return self.STATUS_UNKNOWN - #writable = (out[1] == "w") - active = (out[4] == "a") - online = (out[5] == "o") - if online: - retval = self.STATUS_ONLINE - elif active: - retval = self.STATUS_STANDBY + if self._degraded: + ldisk_status = constants.LDS_FAULTY else: - retval = self.STATUS_EXISTING - - return retval + 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. @@ -435,8 +953,7 @@ class LogicalVolume(BlockDev): This is a no-op for the LV device type. """ - return True - + pass def Close(self): """Notifies that the device will no longer be used for I/O. @@ -444,253 +961,340 @@ class LogicalVolume(BlockDev): This is a no-op for the LV device type. """ - return True - + pass def Snapshot(self, size): """Create a snapshot copy of an lvm block device. - """ + @returns: tuple (vg, lv) + """ snap_name = self._lv_name + ".snap" # remove existing snapshot if found - snap = LogicalVolume((self._vg_name, snap_name), None) - snap.Remove() + snap = LogicalVolume((self._vg_name, snap_name), None, size, self.params) + _IgnoreError(snap.Remove) - pvs_info = self.GetPVInfo(self._vg_name) - if not pvs_info: - raise errors.BlockDeviceError, ("Can't compute PV info for vg %s" % - self._vg_name) - pvs_info.sort() - pvs_info.reverse() - free_size, pv_name = pvs_info[0] + vg_info = self.GetVGInfo([self._vg_name], False) + if not vg_info: + _ThrowError("Can't compute VG info for vg %s", self._vg_name) + free_size, _, _ = vg_info[0] if free_size < size: - raise errors.BlockDeviceError, ("Not enough free space: required %s," - " available %s" % (size, free_size)) - - result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-s", - "-n%s" % snap_name, self.dev_path]) - if result.failed: - raise errors.BlockDeviceError, ("command: %s error: %s" % - (result.cmd, result.fail_reason)) - - return snap_name - - -class MDRaid1(BlockDev): - """raid1 device implemented via md. + _ThrowError("Not enough free space: required %s," + " available %s", size, free_size) - """ - def __init__(self, unique_id, children): - super(MDRaid1, self).__init__(unique_id, children) - self.major = 9 - self.Attach() + _CheckResult(utils.RunCmd(["lvcreate", "-L%dm" % size, "-s", + "-n%s" % snap_name, self.dev_path])) + return (self._vg_name, snap_name) - def Attach(self): - """Find an array which matches our config and attach to it. - - This tries to find a MD array which has the same UUID as our own. + def _RemoveOldInfo(self): + """Try to remove old tags from the lv. """ - minor = self._FindMDByUUID(self.unique_id) - if minor is not None: - self._SetFromMinor(minor) - else: - self.minor = None - self.dev_path = None - - return (minor is not None) + result = utils.RunCmd(["lvs", "-o", "tags", "--noheadings", "--nosuffix", + self.dev_path]) + _CheckResult(result) + raw_tags = result.stdout.strip() + if raw_tags: + for tag in raw_tags.split(","): + _CheckResult(utils.RunCmd(["lvchange", "--deltag", + tag.strip(), self.dev_path])) - @staticmethod - def _GetUsedDevs(): - """Compute the list of in-use MD devices. - - It doesn't matter if the used device have other raid level, just - that they are in use. + def SetInfo(self, text): + """Update metadata with info text. """ - mdstat = open("/proc/mdstat", "r") - data = mdstat.readlines() - mdstat.close() + BlockDev.SetInfo(self, text) - used_md = {} - valid_line = re.compile("^md([0-9]+) : .*$") - for line in data: - match = valid_line.match(line) - if match: - md_no = int(match.group(1)) - used_md[md_no] = line + self._RemoveOldInfo() - return used_md + # Replace invalid characters + text = re.sub("^[^A-Za-z0-9_+.]", "_", text) + text = re.sub("[^-A-Za-z0-9_+.]", "_", text) + # Only up to 128 characters are allowed + text = text[:128] - @staticmethod - def _GetDevInfo(minor): - """Get info about a MD device. + _CheckResult(utils.RunCmd(["lvchange", "--addtag", text, self.dev_path])) - Currently only uuid is returned. + def Grow(self, amount, dryrun, backingstore): + """Grow the logical volume. """ - result = utils.RunCmd(["mdadm", "-D", "/dev/md%d" % minor]) - if result.failed: - logger.Error("Can't display md: %s" % result.fail_reason) - return None - retval = {} - for line in result.stdout.splitlines(): - line = line.strip() - kv = line.split(" : ", 1) - if kv: - if kv[0] == "UUID": - retval["uuid"] = kv[1] - elif kv[0] == "State": - retval["state"] = kv[1].split(", ") - return retval + if not backingstore: + return + 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 + # pe_size is in KB + amount *= 1024 + rest = amount % full_stripe_size + if rest != 0: + amount += full_stripe_size - rest + cmd = ["lvextend", "-L", "+%dk" % amount] + if dryrun: + cmd.append("--test") + # 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 + # supports 'cling' + for alloc_policy in "contiguous", "cling", "normal": + result = utils.RunCmd(cmd + ["--alloc", alloc_policy, self.dev_path]) + if not result.failed: + return + _ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) - @staticmethod - def _FindUnusedMinor(): - """Compute an unused MD minor. +class DRBD8Status(object): + """A DRBD status representation class. - This code assumes that there are 256 minors only. + Note that this doesn't support unconfigured devices (cs:Unconfigured). - """ - used_md = MDRaid1._GetUsedDevs() - i = 0 - while i < 256: - if i not in used_md: - break - i += 1 - if i == 256: - logger.Error("Critical: Out of md minor numbers.") - return None - return i - - - @classmethod - def _FindMDByUUID(cls, uuid): - """Find the minor of an MD array with a given UUID. + """ + UNCONF_RE = re.compile(r"\s*[0-9]+:\s*cs:Unconfigured$") + LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+(?:st|ro):([^/]+)/(\S+)" + "\s+ds:([^/]+)/(\S+)\s+.*$") + SYNC_RE = re.compile(r"^.*\ssync'ed:\s*([0-9.]+)%.*" + # Due to a bug in drbd in the kernel, introduced in + # commit 4b0715f096 (still unfixed as of 2011-08-22) + "(?:\s|M)" + "finish: ([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 = compat.UniqueFrozenset([ + 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: + self.cstatus = self.CS_UNCONFIGURED + self.lrole = self.rrole = self.ldisk = self.rdisk = None + else: + m = self.LINE_RE.match(procline) + if not m: + raise errors.BlockDeviceError("Can't parse input data '%s'" % procline) + self.cstatus = m.group(1) + self.lrole = m.group(2) + self.rrole = m.group(3) + self.ldisk = m.group(4) + self.rdisk = m.group(5) + + # end reading of data from the LINE_RE or UNCONF_RE + + 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.is_diskless = self.ldisk == self.DS_DISKLESS + self.is_disk_uptodate = self.ldisk == self.DS_UPTODATE + + 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: + self.sync_percent = float(m.group(1)) + hours = int(m.group(2)) + minutes = int(m.group(3)) + seconds = int(m.group(4)) + self.est_time = hours * 3600 + minutes * 60 + seconds + else: + # 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 - """ - md_list = cls._GetUsedDevs() - for minor in md_list: - info = cls._GetDevInfo(minor) - if info and info["uuid"] == uuid: - return minor - return None +class BaseDRBD(BlockDev): # pylint: disable=W0223 + """Base DRBD class. - @classmethod - def Create(cls, unique_id, children, size): - """Create a new MD raid1 array. - - """ - if not isinstance(children, (tuple, list)): - raise ValueError("Invalid setup data for MDRaid1 dev: %s" % - str(children)) - for i in children: - if not isinstance(i, BlockDev): - raise ValueError("Invalid member in MDRaid1 dev: %s" % type(i)) - for i in children: - result = utils.RunCmd(["mdadm", "--zero-superblock", "--force", - i.dev_path]) - if result.failed: - logger.Error("Can't zero superblock: %s" % result.fail_reason) - return None - minor = cls._FindUnusedMinor() - result = utils.RunCmd(["mdadm", "--create", "/dev/md%d" % minor, - "--auto=yes", "--force", "-l1", - "-n%d" % len(children)] + - [dev.dev_path for dev in children]) + This class contains a few bits of common functionality between the + 0.7 and 8.x versions of DRBD. - if result.failed: - logger.Error("Can't create md: %s" % result.fail_reason) - return None - info = cls._GetDevInfo(minor) - if not info or not "uuid" in info: - logger.Error("Wrong information returned from mdadm -D: %s" % str(info)) - return None - return MDRaid1(info["uuid"], children) + """ + _VERSION_RE = re.compile(r"^version: (\d+)\.(\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" + _ST_WFCONNECTION = "WFConnection" + _ST_CONNECTED = "Connected" - def Remove(self): - """Stub remove function for MD RAID 1 arrays. + _STATUS_FILE = constants.DRBD_STATUS_FILE + _USERMODE_HELPER_FILE = "/sys/module/drbd/parameters/usermode_helper" - We don't remove the superblock right now. Mark a to do. + @staticmethod + def _GetProcData(filename=_STATUS_FILE): + """Return data from /proc/drbd. """ - #TODO: maybe zero superblock on child devices? - return self.Shutdown() + try: + 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 + @classmethod + def _MassageProcData(cls, data): + """Transform the output of _GetProdData into a nicer form. - def AddChild(self, device): - """Add a new member to the md raid1. + @return: a dictionary of minor: joined lines from /proc/drbd + for that minor """ - if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError, "Can't attach to device" - if device.dev_path is None: - raise errors.BlockDeviceError, "New child is not initialised" - result = utils.RunCmd(["mdadm", "-a", self.dev_path, device.dev_path]) - if result.failed: - raise errors.BlockDeviceError, ("Failed to add new device to array: %s" % - result.output) - new_len = len(self._children) + 1 - result = utils.RunCmd(["mdadm", "--grow", self.dev_path, "-n", new_len]) - if result.failed: - raise errors.BlockDeviceError, ("Can't grow md array: %s" % - result.output) - self._children.append(device) + results = {} + old_minor = old_line = None + for line in data: + 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 + old_minor = int(lresult.group(1)) + old_line = line + else: + if old_minor is not None: + old_line += " " + line.strip() + # add last line + if old_minor is not None: + results[old_minor] = old_line + return results + + @classmethod + def _GetVersion(cls, proc_data): + """Return the DRBD version. + + This will return a dict with keys: + - k_major + - k_minor + - k_point + - api + - proto + - proto2 (only on drbd > 8.2.X) + + """ + first_line = proc_data[0].strip() + version = cls._VERSION_RE.match(first_line) + if not version: + raise errors.BlockDeviceError("Can't parse DRBD version from '%s'" % + first_line) + + values = version.groups() + retval = { + "k_major": int(values[0]), + "k_minor": int(values[1]), + "k_point": int(values[2]), + "api": int(values[3]), + "proto": int(values[4]), + } + if values[5] is not None: + retval["proto2"] = values[5] + return retval - def RemoveChild(self, dev_path): - """Remove member from the md raid1. + @staticmethod + def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): + """Returns DRBD usermode_helper currently set. """ - if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError, "Can't attach to device" - if len(self._children) == 1: - raise errors.BlockDeviceError, ("Can't reduce member when only one" - " child left") - for device in self._children: - if device.dev_path == dev_path: - break - else: - raise errors.BlockDeviceError, "Can't find child with this path" - new_len = len(self._children) - 1 - result = utils.RunCmd(["mdadm", "-f", self.dev_path, dev_path]) - if result.failed: - raise errors.BlockDeviceError, ("Failed to mark device as failed: %s" % - result.output) + try: + helper = utils.ReadFile(filename).splitlines()[0] + 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 DRBD helper file %s: %s", filename, str(err)) + if not helper: + _ThrowError("Can't read any data from %s", filename) + return helper - # it seems here we need a short delay for MD to update its - # superblocks - time.sleep(0.5) - result = utils.RunCmd(["mdadm", "-r", self.dev_path, dev_path]) - if result.failed: - raise errors.BlockDeviceError, ("Failed to remove device from array:" - " %s" % result.output) - result = utils.RunCmd(["mdadm", "--grow", "--force", self.dev_path, - "-n", new_len]) - if result.failed: - raise errors.BlockDeviceError, ("Can't shrink md array: %s" % - result.output) - self._children.remove(device) + @staticmethod + def _DevPath(minor): + """Return the path to a drbd device for a given minor. + """ + return "/dev/drbd%d" % minor - def GetStatus(self): - """Return the status of the device. + @classmethod + def GetUsedDevs(cls): + """Compute the list of used DRBD devices. """ - self.Attach() - if self.minor is None: - retval = self.STATUS_UNKNOWN - else: - retval = self.STATUS_ONLINE - return retval + data = cls._GetProcData() + + used_devs = {} + for line in data: + match = cls._VALID_LINE_RE.match(line) + if not match: + continue + minor = int(match.group(1)) + state = match.group(2) + if state == cls._ST_UNCONFIGURED: + continue + used_devs[minor] = state, line + return used_devs def _SetFromMinor(self, minor): """Set our parameters based on the given minor. @@ -698,276 +1302,262 @@ class MDRaid1(BlockDev): This sets our minor variable and our dev_path. """ - self.minor = minor - self.dev_path = "/dev/md%d" % minor - - - def Assemble(self): - """Assemble the MD device. - - At this point we should have: - - list of children devices - - uuid - - """ - result = super(MDRaid1, self).Assemble() - if not result: - return result - md_list = self._GetUsedDevs() - for minor in md_list: - info = self._GetDevInfo(minor) - if info and info["uuid"] == self.unique_id: - self._SetFromMinor(minor) - logger.Info("MD array %s already started" % str(self)) - return True - free_minor = self._FindUnusedMinor() - result = utils.RunCmd(["mdadm", "-A", "--auto=yes", "--uuid", - self.unique_id, "/dev/md%d" % free_minor] + - [bdev.dev_path for bdev in self._children]) - if result.failed: - logger.Error("Can't assemble MD array: %s" % result.fail_reason) - self.minor = None + if minor is None: + self.minor = self.dev_path = None + self.attached = False else: - self.minor = free_minor - return not result.failed - + self.minor = minor + self.dev_path = self._DevPath(minor) + self.attached = True - def Shutdown(self): - """Tear down the MD array. + @staticmethod + def _CheckMetaSize(meta_device): + """Check if the given meta device looks like a valid one. - This does a 'mdadm --stop' so after this command, the array is no - longer available. + This currently only checks the size, which must be around + 128MiB. """ - if self.minor is None and not self.Attach(): - logger.Info("MD object not attached to a device") - return True - - result = utils.RunCmd(["mdadm", "--stop", "/dev/md%d" % self.minor]) + result = utils.RunCmd(["blockdev", "--getsize", meta_device]) if result.failed: - logger.Error("Can't stop MD array: %s" % result.fail_reason) - return False - self.minor = None - self.dev_path = None - return True - - - def SetSyncSpeed(self, kbytes): - """Set the maximum sync speed for the MD array. - - """ - result = super(MDRaid1, self).SetSyncSpeed(kbytes) - if self.minor is None: - logger.Error("MD array not attached to a device") - return False - f = open("/sys/block/md%d/md/sync_speed_max" % self.minor, "w") + _ThrowError("Failed to get device size: %s - %s", + result.fail_reason, result.output) try: - f.write("%d" % kbytes) - finally: - f.close() - f = open("/sys/block/md%d/md/sync_speed_min" % self.minor, "w") - try: - f.write("%d" % (kbytes/2)) - finally: - f.close() - return result - - - def GetSyncStatus(self): - """Returns the sync status of the device. - - Returns: - (sync_percent, estimated_time) - - If sync_percent is None, it means all is ok - If estimated_time is None, it means we can't esimate - the time needed, otherwise it's the time left in seconds - - """ - if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") - dev_info = self._GetDevInfo(self.minor) - is_clean = ("state" in dev_info and - len(dev_info["state"]) == 1 and - dev_info["state"][0] in ("clean", "active")) - sys_path = "/sys/block/md%s/md/" % self.minor - f = file(sys_path + "sync_action") - sync_status = f.readline().strip() - f.close() - if sync_status == "idle": - return None, None, not is_clean - f = file(sys_path + "sync_completed") - sync_completed = f.readline().strip().split(" / ") - f.close() - if len(sync_completed) != 2: - return 0, None, not is_clean - sync_done, sync_total = [float(i) for i in sync_completed] - sync_percent = 100.0*sync_done/sync_total - f = file(sys_path + "sync_speed") - sync_speed_k = int(f.readline().strip()) - if sync_speed_k == 0: - time_est = None - else: - time_est = (sync_total - sync_done) / 2 / sync_speed_k - return sync_percent, time_est, not is_clean - - - def Open(self, force=False): - """Make the device ready for I/O. - - This is a no-op for the MDRaid1 device type, although we could use - the 2.6.18's new array_state thing. - - """ - return True - + sectors = int(result.stdout) + except (TypeError, ValueError): + _ThrowError("Invalid output from blockdev: '%s'", result.stdout) + num_bytes = sectors * 512 + if num_bytes < 128 * 1024 * 1024: # less than 128MiB + _ThrowError("Meta device too small (%.2fMib)", (num_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 num_bytes > 1024 * 1024 * 1024: + _ThrowError("Meta device too big (%.2fMiB)", (num_bytes / 1024 / 1024)) - def Close(self): - """Notifies that the device will no longer be used for I/O. + def Rename(self, new_id): + """Rename a device. - This is a no-op for the MDRaid1 device type, but see comment for - `Open()`. + This is not supported for drbd devices. """ - return True + raise errors.ProgrammerError("Can't rename a drbd device") - -class DRBDev(BlockDev): - """DRBD block device. +class DRBD8(BaseDRBD): + """DRBD v8.x block device. This implements the local host part of the DRBD device, i.e. it doesn't do anything to the supposed peer. If you need a fully connected DRBD pair, you need to use this class on both hosts. - The unique_id for the drbd device is the (local_ip, local_port, - remote_ip, remote_port) tuple, and it must have two children: the - data device and the meta_device. The meta device is checked for - valid size and is zeroed on create. + The unique_id for the drbd device is a (local_ip, local_port, + remote_ip, remote_port, local_minor, secret) tuple, and it must have + two children: the data device and the meta_device. The meta device + is checked for valid size and is zeroed on create. """ - _DRBD_MAJOR = 147 - _ST_UNCONFIGURED = "Unconfigured" - _ST_WFCONNECTION = "WFConnection" - _ST_CONNECTED = "Connected" - - def __init__(self, unique_id, children): - super(DRBDev, self).__init__(unique_id, children) - self.major = self._DRBD_MAJOR - if len(children) != 2: + _MAX_MINORS = 255 + _PARSE_SHOW = None + + # timeout constants + _NET_RECONFIG_TIMEOUT = 60 + + # command line options for barriers + _DISABLE_DISK_OPTION = "--no-disk-barrier" # -a + _DISABLE_DRAIN_OPTION = "--no-disk-drain" # -D + _DISABLE_FLUSH_OPTION = "--no-disk-flushes" # -i + _DISABLE_META_FLUSH_OPTION = "--no-md-flushes" # -m + + def __init__(self, unique_id, children, size, params): + if children and children.count(None) > 0: + children = [] + 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) != 4: + if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6: raise ValueError("Invalid configuration data %s" % str(unique_id)) - self._lhost, self._lport, self._rhost, self._rport = 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, params) + self.major = self._DRBD_MAJOR + version = self._GetVersion(self._GetProcData()) + 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" % + (unique_id,)) self.Attach() - @staticmethod - def _DevPath(minor): - """Return the path to a drbd device for a given minor. - - """ - return "/dev/drbd%d" % minor - - @staticmethod - def _GetProcData(): - """Return data from /proc/drbd. - - """ - stat = open("/proc/drbd", "r") - data = stat.read().splitlines() - stat.close() - return data - - @classmethod - def _GetUsedDevs(cls): - """Compute the list of used DRBD devices. + def _InitMeta(cls, minor, dev_path): + """Initialize a meta device. - """ - data = cls._GetProcData() - - used_devs = {} - valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$") - for line in data: - match = valid_line.match(line) - if not match: - continue - minor = int(match.group(1)) - state = match.group(2) - if state == cls._ST_UNCONFIGURED: - continue - used_devs[minor] = state, line + This will not work if the given minor is in use. - return used_devs + """ + # Zero the metadata first, in order to make sure drbdmeta doesn't + # try to auto-detect existing filesystems or similar (see + # http://code.google.com/p/ganeti/issues/detail?id=182); we only + # care about the first 128MB of data in the device, even though it + # can be bigger + result = utils.RunCmd([constants.DD_CMD, + "if=/dev/zero", "of=%s" % dev_path, + "bs=1048576", "count=128", "oflag=direct"]) + if result.failed: + _ThrowError("Can't wipe the meta device: %s", result.output) + result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), + "v08", dev_path, "0", "create-md"]) + if result.failed: + _ThrowError("Can't initialize meta device: %s", result.output) @classmethod def _FindUnusedMinor(cls): """Find an unused DRBD device. + This is specific to 8.x as the minors are allocated dynamically, + so non-existing numbers up to a max minor count are actually free. + """ data = cls._GetProcData() - valid_line = re.compile("^ *([0-9]+): cs:Unconfigured$") + highest = None for line in data: - match = valid_line.match(line) + match = cls._UNUSED_LINE_RE.match(line) if match: return int(match.group(1)) - logger.Error("Error: no free drbd minors!") - return None - + match = cls._VALID_LINE_RE.match(line) + if match: + minor = int(match.group(1)) + highest = max(highest, minor) + if highest is None: # there are no minors in use at all + return 0 + if highest >= cls._MAX_MINORS: + logging.error("Error: no free drbd minors!") + raise errors.BlockDeviceError("Can't find a free DRBD minor") + return highest + 1 @classmethod - def _GetDevInfo(cls, minor): - """Get details about a given DRBD minor. + def _GetShowParser(cls): + """Return a parser for `drbd show` output. + + This will either create or return an already-created parser for the + output of the command `drbd show`. + + """ + if cls._PARSE_SHOW is not None: + return cls._PARSE_SHOW + + # pyparsing setup + lbrace = pyp.Literal("{").suppress() + rbrace = pyp.Literal("}").suppress() + lbracket = pyp.Literal("[").suppress() + rbracket = pyp.Literal("]").suppress() + semi = pyp.Literal(";").suppress() + colon = pyp.Literal(":").suppress() + # this also converts the value to an int + number = pyp.Word(pyp.nums).setParseAction(lambda s, l, t: int(t[0])) + + comment = pyp.Literal("#") + pyp.Optional(pyp.restOfLine) + defa = pyp.Literal("_is_default").suppress() + dbl_quote = pyp.Literal('"').suppress() + + keyword = pyp.Word(pyp.alphanums + "-") + + # value types + value = pyp.Word(pyp.alphanums + "_-/.:") + quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote + ipv4_addr = (pyp.Optional(pyp.Literal("ipv4")).suppress() + + pyp.Word(pyp.nums + ".") + colon + number) + ipv6_addr = (pyp.Optional(pyp.Literal("ipv6")).suppress() + + pyp.Optional(lbracket) + pyp.Word(pyp.hexnums + ":") + + pyp.Optional(rbracket) + colon + number) + # meta device, extended syntax + meta_value = ((value ^ quoted) + lbracket + number + rbracket) + # device name, extended syntax + device_value = pyp.Literal("minor").suppress() + number + + # a statement + stmt = (~rbrace + keyword + ~lbrace + + pyp.Optional(ipv4_addr ^ ipv6_addr ^ value ^ quoted ^ meta_value ^ + device_value) + + pyp.Optional(defa) + semi + + pyp.Optional(pyp.restOfLine).suppress()) + + # an entire section + section_name = pyp.Word(pyp.alphas + "_") + section = section_name + lbrace + pyp.ZeroOrMore(pyp.Group(stmt)) + rbrace + + bnf = pyp.ZeroOrMore(pyp.Group(section ^ stmt)) + bnf.ignore(comment) + + cls._PARSE_SHOW = bnf + + return bnf - This return, if available, the local backing device in (major, - minor) formant and the local and remote (ip, port) information. + @classmethod + def _GetShowData(cls, minor): + """Return the `drbdsetup show` data for a minor. """ - data = {} result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) if result.failed: - logger.Error("Can't display the drbd config: %s" % result.fail_reason) - return data - out = result.stdout - if out == "Not configured\n": + logging.error("Can't display the drbd config: %s - %s", + result.fail_reason, result.output) + return None + return result.stdout + + @classmethod + def _GetDevInfo(cls, out): + """Parse details about a given DRBD minor. + + This return, if available, the local backing device (as a path) + and the local and remote (ip, port) information from a string + containing the output of the `drbdsetup show` command as returned + by _GetShowData. + + """ + data = {} + if not out: return data - for line in out.splitlines(): - if "local_dev" not in data: - match = re.match("^Lower device: ([0-9]+):([0-9]+) .*$", line) - if match: - data["local_dev"] = (int(match.group(1)), int(match.group(2))) - continue - if "meta_dev" not in data: - match = re.match("^Meta device: (([0-9]+):([0-9]+)|internal).*$", line) - if match: - if match.group(2) is not None and match.group(3) is not None: - # matched on the major/minor - data["meta_dev"] = (int(match.group(2)), int(match.group(3))) - else: - # matched on the "internal" string - data["meta_dev"] = match.group(1) - # in this case, no meta_index is in the output - data["meta_index"] = -1 - continue - if "meta_index" not in data: - match = re.match("^Meta index: ([0-9]+).*$", line) - if match: - data["meta_index"] = int(match.group(1)) - continue - if "local_addr" not in data: - match = re.match("^Local address: ([0-9.]+):([0-9]+)$", line) - if match: - data["local_addr"] = (match.group(1), int(match.group(2))) - continue - if "remote_addr" not in data: - match = re.match("^Remote address: ([0-9.]+):([0-9]+)$", line) - if match: - data["remote_addr"] = (match.group(1), int(match.group(2))) - continue - return data + bnf = cls._GetShowParser() + # run pyparse + + try: + results = bnf.parseString(out) + except pyp.ParseException, err: + _ThrowError("Can't parse drbdsetup show output: %s", str(err)) + + # and massage the results into our desired format + for section in results: + sname = section[0] + if sname == "_this_host": + for lst in section[1:]: + if lst[0] == "disk": + data["local_dev"] = lst[1] + elif lst[0] == "meta-disk": + data["meta_dev"] = lst[1] + data["meta_index"] = lst[2] + elif lst[0] == "address": + data["local_addr"] = tuple(lst[1:]) + elif sname == "_remote_host": + for lst in section[1:]: + if lst[0] == "address": + data["remote_addr"] = tuple(lst[1:]) + return data def _MatchesLocal(self, info): """Test if our local config matches with an existing device. @@ -978,25 +1568,26 @@ class DRBDev(BlockDev): device. """ - if not ("local_dev" in info and "meta_dev" in info and - "meta_index" in info): - return False + if self._children: + backend, meta = self._children + else: + backend = meta = None - backend = self._children[0] if backend is not None: - retval = (info["local_dev"] == (backend.major, backend.minor)) + retval = ("local_dev" in info and info["local_dev"] == backend.dev_path) else: - retval = (info["local_dev"] == (0, 0)) - meta = self._children[1] + retval = ("local_dev" not in info) + if meta is not None: - retval = retval and (info["meta_dev"] == (meta.major, meta.minor)) - retval = retval and (info["meta_index"] == 0) + retval = retval and ("meta_dev" in info and + info["meta_dev"] == meta.dev_path) + retval = retval and ("meta_index" in info and + info["meta_index"] == 0) else: - retval = retval and (info["meta_dev"] == "internal" and - info["meta_index"] == -1) + retval = retval and ("meta_dev" not in info and + "meta_index" not in info) return retval - def _MatchesNet(self, info): """Test if our network config matches with an existing device. @@ -1022,259 +1613,398 @@ class DRBDev(BlockDev): info["remote_addr"] == (self._rhost, self._rport)) return retval - - @staticmethod - def _IsValidMeta(meta_device): - """Check if the given meta device looks like a valid one. - - This currently only check the size, which must be around - 128MiB. - - """ - result = utils.RunCmd(["blockdev", "--getsize", meta_device]) - if result.failed: - logger.Error("Failed to get device size: %s" % result.fail_reason) - return False - try: - sectors = int(result.stdout) - except ValueError: - logger.Error("Invalid output from blockdev: '%s'" % result.stdout) - return False - bytes = sectors * 512 - if bytes < 128*1024*1024: # less than 128MiB - logger.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 - logger.Error("Meta device too big (%.2fMiB)" % (bytes/1024/1024)) - return False - return True - - - @classmethod - def _AssembleLocal(cls, minor, backend, meta): + def _AssembleLocal(self, minor, backend, meta, size): """Configure the local part of a DRBD device. - This is the first thing that must be done on an unconfigured DRBD - device. And it must be done only once. - """ - if not cls._IsValidMeta(meta): - return False - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disk", - backend, meta, "0", "-e", "detach"]) - if result.failed: - logger.Error("Can't attach local disk: %s" % result.output) - return not result.failed + args = ["drbdsetup", self._DevPath(minor), "disk", + backend, meta, "0", + "-e", "detach", + "--create-device"] + if size: + args.extend(["-d", "%sm" % size]) + version = self._GetVersion(self._GetProcData()) + vmaj = version["k_major"] + vmin = version["k_minor"] + vrel = version["k_point"] - @classmethod - def _ShutdownLocal(cls, minor): - """Detach from the local device. + barrier_args = \ + self._ComputeDiskBarrierArgs(vmaj, vmin, vrel, + self.params[constants.LDP_BARRIERS], + self.params[constants.LDP_NO_META_FLUSH]) + args.extend(barrier_args) - I/Os will continue to be served from the remote device. If we - don't have a remote device, this operation will fail. + if self.params[constants.LDP_DISK_CUSTOM]: + args.extend(shlex.split(self.params[constants.LDP_DISK_CUSTOM])) - """ - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) + result = utils.RunCmd(args) if result.failed: - logger.Error("Can't detach local device: %s" % result.output) - return not result.failed + _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output) + @classmethod + def _ComputeDiskBarrierArgs(cls, vmaj, vmin, vrel, disabled_barriers, + disable_meta_flush): + """Compute the DRBD command line parameters for disk barriers - @staticmethod - def _ShutdownAll(minor): - """Deactivate the device. + Returns a list of the disk barrier parameters as requested via the + disabled_barriers and disable_meta_flush arguments, and according to the + supported ones in the DRBD version vmaj.vmin.vrel - This will, of course, fail if the device is in use. + If the desired option is unsupported, raises errors.BlockDeviceError. """ - result = utils.RunCmd(["drbdsetup", DRBDev._DevPath(minor), "down"]) - if result.failed: - logger.Error("Can't shutdown drbd device: %s" % result.output) - return not result.failed + disabled_barriers_set = frozenset(disabled_barriers) + if not disabled_barriers_set in constants.DRBD_VALID_BARRIER_OPT: + raise errors.BlockDeviceError("%s is not a valid option set for DRBD" + " barriers" % disabled_barriers) + args = [] - @classmethod - def _AssembleNet(cls, minor, net_info, protocol): - """Configure the network part of the device. + # The following code assumes DRBD 8.x, with x < 4 and x != 1 (DRBD 8.1.x + # does not exist) + if not vmaj == 8 and vmin in (0, 2, 3): + raise errors.BlockDeviceError("Unsupported DRBD version: %d.%d.%d" % + (vmaj, vmin, vrel)) - This operation can be, in theory, done multiple times, but there - have been cases (in lab testing) in which the network part of the - device had become stuck and couldn't be shut down because activity - from the new peer (also stuck) triggered a timer re-init and - needed remote peer interface shutdown in order to clear. So please - don't change online the net config. + def _AppendOrRaise(option, min_version): + """Helper for DRBD options""" + if min_version is not None and vrel >= min_version: + args.append(option) + else: + raise errors.BlockDeviceError("Could not use the option %s as the" + " DRBD version %d.%d.%d does not support" + " it." % (option, vmaj, vmin, vrel)) + + # the minimum version for each feature is encoded via pairs of (minor + # version -> x) where x is version in which support for the option was + # introduced. + meta_flush_supported = disk_flush_supported = { + 0: 12, + 2: 7, + 3: 0, + } + + disk_drain_supported = { + 2: 7, + 3: 0, + } + + disk_barriers_supported = { + 3: 0, + } + + # meta flushes + if disable_meta_flush: + _AppendOrRaise(cls._DISABLE_META_FLUSH_OPTION, + meta_flush_supported.get(vmin, None)) + + # disk flushes + if constants.DRBD_B_DISK_FLUSH in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_FLUSH_OPTION, + disk_flush_supported.get(vmin, None)) + + # disk drain + if constants.DRBD_B_DISK_DRAIN in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_DRAIN_OPTION, + disk_drain_supported.get(vmin, None)) + + # disk barriers + if constants.DRBD_B_DISK_BARRIERS in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_DISK_OPTION, + disk_barriers_supported.get(vmin, None)) + + return args + + def _AssembleNet(self, minor, net_info, protocol, + dual_pri=False, hmac=None, secret=None): + """Configure the network part of the device. """ lhost, lport, rhost, rport = net_info - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "net", - "%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), - protocol]) + if None in net_info: + # we don't want network connection and actually want to make + # sure its shutdown + self._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 + # synchronization speed over the wire. In some cases setting the + # sync speed only after setting up both sides can race with DRBD + # connecting, hence we set it here before telling DRBD anything + # about its peer. + sync_errors = self._SetMinorSyncParams(minor, self.params) + if sync_errors: + _ThrowError("drbd%d: can't set the synchronization parameters: %s" % + (minor, utils.CommaJoin(sync_errors))) + + if netutils.IP6Address.IsValid(lhost): + if not netutils.IP6Address.IsValid(rhost): + _ThrowError("drbd%d: can't connect ip %s to ip %s" % + (minor, lhost, rhost)) + family = "ipv6" + elif netutils.IP4Address.IsValid(lhost): + if not netutils.IP4Address.IsValid(rhost): + _ThrowError("drbd%d: can't connect ip %s to ip %s" % + (minor, lhost, rhost)) + family = "ipv4" + else: + _ThrowError("drbd%d: Invalid ip %s" % (minor, lhost)) + + args = ["drbdsetup", self._DevPath(minor), "net", + "%s:%s:%s" % (family, lhost, lport), + "%s:%s:%s" % (family, rhost, rport), protocol, + "-A", "discard-zero-changes", + "-B", "consensus", + "--create-device", + ] + if dual_pri: + args.append("-m") + if hmac and secret: + args.extend(["-a", hmac, "-x", secret]) + + if self.params[constants.LDP_NET_CUSTOM]: + args.extend(shlex.split(self.params[constants.LDP_NET_CUSTOM])) + + result = utils.RunCmd(args) if result.failed: - logger.Error("Can't setup network for dbrd device: %s" % - result.fail_reason) - 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: - info = cls._GetDevInfo(minor) + def _CheckNetworkConfig(): + info = self._GetDevInfo(self._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)): - time.sleep(1) - continue - ok = True - break - if not ok: - logger.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) - @classmethod - def _ShutdownNet(cls, minor): - """Disconnect from the remote peer. + def AddChildren(self, devices): + """Add a disk to the DRBD device. - This fails if we don't have a local device. + """ + if self.minor is None: + _ThrowError("drbd%d: can't attach to dbrd8 during AddChildren", + self._aminor) + if len(devices) != 2: + _ThrowError("drbd%d: need two devices for AddChildren", self.minor) + info = self._GetDevInfo(self._GetShowData(self.minor)) + if "local_dev" in info: + _ThrowError("drbd%d: already attached to a local disk", self.minor) + backend, meta = devices + if backend.dev_path is None or meta.dev_path is None: + _ThrowError("drbd%d: children not ready during AddChildren", self.minor) + backend.Open() + meta.Open() + self._CheckMetaSize(meta.dev_path) + self._InitMeta(self._FindUnusedMinor(), meta.dev_path) + + self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size) + self._children = devices + + def RemoveChildren(self, devices): + """Detach the drbd device from local storage. """ - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) - logger.Error("Can't shutdown network: %s" % result.output) - return not result.failed + if self.minor is None: + _ThrowError("drbd%d: can't attach to drbd8 during RemoveChildren", + self._aminor) + # early return if we don't actually have backing storage + info = self._GetDevInfo(self._GetShowData(self.minor)) + if "local_dev" not in info: + return + if len(self._children) != 2: + _ThrowError("drbd%d: we don't have two children: %s", self.minor, + self._children) + if self._children.count(None) == 2: # we don't actually have children :) + logging.warning("drbd%d: requested detach while detached", self.minor) + return + if len(devices) != 2: + _ThrowError("drbd%d: we need two children in RemoveChildren", self.minor) + for child, dev in zip(self._children, devices): + if dev != child.dev_path: + _ThrowError("drbd%d: mismatch in local storage (%s != %s) in" + " RemoveChildren", self.minor, dev, child.dev_path) + + self._ShutdownLocal(self.minor) + self._children = [] + @classmethod + def _SetMinorSyncParams(cls, minor, params): + """Set the parameters of the DRBD syncer. + + This is the low-level implementation. + + @type minor: int + @param minor: the drbd minor whose settings we change + @type params: dict + @param params: LD level disk parameters related to the synchronization + @rtype: list + @return: a list of error messages + + """ + + args = ["drbdsetup", cls._DevPath(minor), "syncer"] + if params[constants.LDP_DYNAMIC_RESYNC]: + version = cls._GetVersion(cls._GetProcData()) + vmin = version["k_minor"] + vrel = version["k_point"] + + # By definition we are using 8.x, so just check the rest of the version + # number + if vmin != 3 or vrel < 9: + msg = ("The current DRBD version (8.%d.%d) does not support the " + "dynamic resync speed controller" % (vmin, vrel)) + logging.error(msg) + return [msg] + + if params[constants.LDP_PLAN_AHEAD] == 0: + msg = ("A value of 0 for c-plan-ahead disables the dynamic sync speed" + " controller at DRBD level. If you want to disable it, please" + " set the dynamic-resync disk parameter to False.") + logging.error(msg) + return [msg] + + # add the c-* parameters to args + args.extend(["--c-plan-ahead", params[constants.LDP_PLAN_AHEAD], + "--c-fill-target", params[constants.LDP_FILL_TARGET], + "--c-delay-target", params[constants.LDP_DELAY_TARGET], + "--c-max-rate", params[constants.LDP_MAX_RATE], + "--c-min-rate", params[constants.LDP_MIN_RATE], + ]) - def _SetFromMinor(self, minor): - """Set our parameters based on the given minor. + else: + args.extend(["-r", "%d" % params[constants.LDP_RESYNC_RATE]]) - This sets our minor variable and our dev_path. + args.append("--create-device") + result = utils.RunCmd(args) + if result.failed: + msg = ("Can't change syncer rate: %s - %s" % + (result.fail_reason, result.output)) + logging.error(msg) + return [msg] + + return [] + + def SetSyncParams(self, params): + """Set the synchronization parameters of the DRBD syncer. + + @type params: dict + @param params: 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 """ - if minor is None: - self.minor = self.dev_path = None - else: - self.minor = minor - self.dev_path = self._DevPath(minor) + if self.minor is None: + err = "Not attached during SetSyncParams" + logging.info(err) + return [err] + children_result = super(DRBD8, self).SetSyncParams(params) + children_result.extend(self._SetMinorSyncParams(self.minor, params)) + return children_result - def Assemble(self): - """Assemble the drbd. + def PauseResumeSync(self, pause): + """Pauses or resumes the sync of a DRBD device. - Method: - - if we have a local backing device, we bind to it by: - - checking the list of used drbd devices - - check if the local minor use of any of them is our own device - - if yes, abort? - - if not, bind - - if we have a local/remote net info: - - redo the local backing device step for the remote device - - check if any drbd device is using the local port, - if yes abort - - check if any remote drbd device is using the remote - port, if yes abort (for now) - - bind our net port - - bind the remote net port + @param pause: Wether to pause or resume + @return: the success of the operation """ - self.Attach() - if self.minor is not None: - logger.Info("Already assembled") - return True + if self.minor is None: + logging.info("Not attached during PauseSync") + return False - result = super(DRBDev, self).Assemble() - if not result: - return result + children_result = super(DRBD8, self).PauseResumeSync(pause) - minor = self._FindUnusedMinor() - if minor is None: - raise errors.BlockDeviceError, "Not enough free minors for DRBD!" - need_localdev_teardown = False - if self._children[0]: - result = self._AssembleLocal(minor, self._children[0].dev_path, - self._children[1].dev_path) - if not result: - return False - need_localdev_teardown = True - if self._lhost and self._lport and self._rhost and self._rport: - result = self._AssembleNet(minor, - (self._lhost, self._lport, - self._rhost, self._rport), - "C") - if not result: - if need_localdev_teardown: - # we will ignore failures from this - logger.Error("net setup failed, tearing down local device") - self._ShutdownAll(minor) - return False - self._SetFromMinor(minor) - return True + if pause: + cmd = "pause-sync" + else: + cmd = "resume-sync" + result = utils.RunCmd(["drbdsetup", self.dev_path, cmd]) + if result.failed: + logging.error("Can't %s: %s - %s", cmd, + result.fail_reason, result.output) + return not result.failed and children_result - def Shutdown(self): - """Shutdown the DRBD device. + def GetProcStatus(self): + """Return device data from /proc. """ - if self.minor is None and not self.Attach(): - logger.Info("DRBD device not attached to a device during Shutdown") - return True - if not self._ShutdownAll(self.minor): - return False - self.minor = None - self.dev_path = None - return True + if self.minor is None: + _ThrowError("drbd%d: GetStats() called while not attached", self._aminor) + proc_info = self._MassageProcData(self._GetProcData()) + if self.minor not in proc_info: + _ThrowError("drbd%d: can't find myself in /proc", self.minor) + return DRBD8Status(proc_info[self.minor]) + def GetSyncStatus(self): + """Returns the sync status of the device. - def Attach(self): - """Find a DRBD device which matches our config and attach to it. - In case of partially attached (local device matches but no network - setup), we perform the network attach. If successful, we re-test - the attach if can return success. + If sync_percent is None, it means all is ok + 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. + + We compute the ldisk parameter based on whether we have a local + disk or not. + + @rtype: objects.BlockDevStatus """ - for minor in self._GetUsedDevs(): - info = self._GetDevInfo(minor) - match_l = self._MatchesLocal(info) - match_r = self._MatchesNet(info) - if match_l and match_r: - break - 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), - "C") - if res_r and self._MatchesNet(self._GetDevInfo(minor)): - break - else: - minor = None + if self.minor is None and not self.Attach(): + _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) - self._SetFromMinor(minor) - return minor is not None + stats = self.GetProcStatus() + 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. - If the 'force' parameter is given, the '--do-what-I-say' parameter - is given. Since this is a pottentialy dangerous operation, the + If the 'force' parameter is given, the '-o' option is passed to + drbdsetup. Since this is a potentially dangerous operation, the force flag should be only given after creation, when it actually - has to be given. + is mandatory. """ if self.minor is None and not self.Attach(): - logger.Error("DRBD cannot attach to a device during open") + logging.error("DRBD cannot attach to a device during open") return False cmd = ["drbdsetup", self.dev_path, "primary"] if force: - cmd.append("--do-what-I-say") + cmd.append("-o") result = utils.RunCmd(cmd) if result.failed: - logger.Error("Can't make drbd device primary: %s" % result.output) - return False - return True - + _ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, + result.output) def Close(self): """Make the local state secondary. @@ -1283,215 +2013,1432 @@ class DRBDev(BlockDev): """ if self.minor is None and not self.Attach(): - logger.Info("Instance not attached to a device") - raise errors.BlockDeviceError("Can't find device") + _ThrowError("drbd%d: can't Attach() in Close()", self._aminor) result = utils.RunCmd(["drbdsetup", self.dev_path, "secondary"]) if result.failed: - logger.Error("Can't switch drbd device to secondary: %s" % result.output) - raise errors.BlockDeviceError("Can't switch drbd device to secondary") + _ThrowError("drbd%d: can't switch drbd device to secondary: %s", + self.minor, result.output) + def DisconnectNet(self): + """Removes network configuration. - def SetSyncSpeed(self, kbytes): - """Set the speed of the DRBD syncer. + This method shutdowns the network side of the device. + + The method will wait up to a hardcoded timeout for the device to + go into standalone after the 'disconnect' command before + re-configuring it, as sometimes it takes a while for the + disconnect to actually propagate and thus we might issue a 'net' + command while the device is still connected. If the device will + still be attached to the network and we time out, we raise an + exception. """ - children_result = super(DRBDev, self).SetSyncSpeed(kbytes) if self.minor is None: - logger.Info("Instance not attached to a device") - return False - result = utils.RunCmd(["drbdsetup", self.dev_path, "syncer", "-r", "%d" % - kbytes]) - if result.failed: - logger.Error("Can't change syncer rate: %s " % result.fail_reason) - return not result.failed and children_result + _ThrowError("drbd%d: disk not attached in re-attach net", self._aminor) + if None in (self._lhost, self._lport, self._rhost, self._rport): + _ThrowError("drbd%d: DRBD disk missing network info in" + " DisconnectNet()", self.minor) - def GetSyncStatus(self): - """Returns the sync status of the device. + class _DisconnectStatus: + def __init__(self, ever_disconnected): + self.ever_disconnected = ever_disconnected - Returns: - (sync_percent, estimated_time) + dstatus = _DisconnectStatus(_IgnoreError(self._ShutdownNet, self.minor)) - If sync_percent is None, it means all is ok - If estimated_time is None, it means we can't esimate - the time needed, otherwise it's the time left in seconds + 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" + + _ThrowError(msg, self.minor) + + 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) + + def AttachNet(self, multimaster): + """Reconnects the network. + + This method connects the network side of the device with a + specified multi-master flag. The device needs to be 'Standalone' + but have valid network configuration data. + + Args: + - multimaster: init the network in dual-primary mode """ - if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") - proc_info = self._MassageProcData(self._GetProcData()) - if self.minor not in proc_info: - raise errors.BlockDeviceError("Can't find myself in /proc (minor %d)" % - self.minor) - line = proc_info[self.minor] - match = re.match("^.*sync'ed: *([0-9.]+)%.*" - " finish: ([0-9]+):([0-9]+):([0-9]+) .*$", line) - if match: - sync_percent = float(match.group(1)) - hours = int(match.group(2)) - minutes = int(match.group(3)) - seconds = int(match.group(4)) - est_time = hours * 3600 + minutes * 60 + seconds + if self.minor is None: + _ThrowError("drbd%d: device not attached in AttachNet", self._aminor) + + if None in (self._lhost, self._lport, self._rhost, self._rport): + _ThrowError("drbd%d: missing network info in AttachNet()", self.minor) + + status = self.GetProcStatus() + + if not status.is_standalone: + _ThrowError("drbd%d: device is not standalone in AttachNet", self.minor) + + 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. + + This doesn't do any device configurations - it only checks if the + minor is in a state different from Unconfigured. + + Note that this function will not change the state of the system in + any way (except in case of side-effects caused by reading from + /proc). + + """ + used_devs = self.GetUsedDevs() + if self._aminor in used_devs: + minor = self._aminor else: - sync_percent = None - est_time = None - match = re.match("^ *[0-9]+: cs:([^ ]+).*$", line) - if not match: - raise errors.BlockDeviceError("Can't find my data in /proc (minor %d)" % - self.minor) - client_state = match.group(1) - is_degraded = client_state != "Connected" - return sync_percent, est_time, is_degraded + minor = None + self._SetFromMinor(minor) + return minor is not None - @staticmethod - def _MassageProcData(data): - """Transform the output of _GetProdData into a nicer form. + def Assemble(self): + """Assemble the drbd. - Returns: - a dictionary of minor: joined lines from /proc/drbd for that minor + Method: + - if we have a configured device, we try to ensure that it matches + our config + - if not, we create it from zero + - anyway, set the device parameters """ - lmatch = re.compile("^ *([0-9]+):.*$") - results = {} - old_minor = old_line = None - for line in data: - lresult = lmatch.match(line) - if lresult is not None: - if old_minor is not None: - results[old_minor] = old_line - old_minor = int(lresult.group(1)) - old_line = line - else: - if old_minor is not None: - old_line += " " + line.strip() - # add last line - if old_minor is not None: - results[old_minor] = old_line - return results + super(DRBD8, self).Assemble() + + self.Attach() + if self.minor is None: + # local device completely unconfigured + self._FastAssemble() + else: + # we have to recheck the local and network status and try to fix + # the device + self._SlowAssemble() + sync_errors = self.SetSyncParams(self.params) + if sync_errors: + _ThrowError("drbd%d: can't set the synchronization parameters: %s" % + (self.minor, utils.CommaJoin(sync_errors))) - def GetStatus(self): - """Compute the status of the DRBD device + def _SlowAssemble(self): + """Assembles the DRBD device from a (partially) configured device. - Note that DRBD devices don't have the STATUS_EXISTING state. + In case of partially attached (local device matches but no network + setup), we perform the network attach. If successful, we re-test + the attach if can return success. """ - if self.minor is None and not self.Attach(): - return self.STATUS_UNKNOWN + # TODO: Rewrite to not use a for loop just because there is 'break' + # pylint: disable=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) - data = self._GetProcData() - match = re.compile("^ *%d: cs:[^ ]+ st:(Primary|Secondary)/.*$" % - self.minor) - for line in data: - mresult = match.match(line) - if mresult: + if match_l and match_r: + # everything matches break - else: - logger.Error("Can't find myself!") - return self.STATUS_UNKNOWN - state = mresult.group(2) - if state == "Primary": - result = self.STATUS_ONLINE + if match_l and not match_r and "local_addr" not in info: + # 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: + # 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 + # devices are equal from the point of view of local + # configuration + if (match_l and "local_dev" in info and + not match_r and "local_addr" in info): + # strange case - the device network part points to somewhere + # 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 + 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) + 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) + else: - result = self.STATUS_STANDBY + minor = None - return result + self._SetFromMinor(minor) + if minor is None: + _ThrowError("drbd%d: cannot activate, unknown or unhandled reason", + self._aminor) + def _FastAssemble(self): + """Assemble the drbd device from zero. - @staticmethod - def _ZeroDevice(device): - """Zero a device. + This is run when in Assemble we detect our minor is unused. + + """ + minor = self._aminor + if self._children and self._children[0] and self._children[1]: + 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: + 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) + + @classmethod + def _ShutdownLocal(cls, minor): + """Detach from the local device. - This writes until we get ENOSPC. + I/Os will continue to be served from the remote device. If we + don't have a remote device, this operation will fail. """ - f = open(device, "w") - buf = "\0" * 1048576 - try: - while True: - f.write(buf) - except IOError, err: - if err.errno != errno.ENOSPC: - raise + result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) + if result.failed: + _ThrowError("drbd%d: can't detach local disk: %s", minor, result.output) + + @classmethod + def _ShutdownNet(cls, minor): + """Disconnect from the remote peer. + + This fails if we don't have a local device. + + """ + result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) + if result.failed: + _ThrowError("drbd%d: can't shutdown network: %s", minor, result.output) + + @classmethod + def _ShutdownAll(cls, minor): + """Deactivate the device. + + This will, of course, fail if the device is in use. + + """ + result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) + if result.failed: + _ThrowError("drbd%d: can't shutdown drbd device: %s", + minor, result.output) + + def Shutdown(self): + """Shutdown the DRBD device. + + """ + if self.minor is None and not self.Attach(): + logging.info("drbd%d: not attached during Shutdown()", self._aminor) + return + minor = self.minor + self.minor = None + self.dev_path = None + self._ShutdownAll(minor) + def Remove(self): + """Stub remove for DRBD devices. + + """ + self.Shutdown() @classmethod - def Create(cls, unique_id, children, size): - """Create a new DRBD device. + def Create(cls, unique_id, children, size, params, excl_stor): + """Create a new DRBD8 device. Since DRBD devices are not created per se, just assembled, this - function just zeroes the meta device. + function only initializes the metadata. """ if len(children) != 2: raise errors.ProgrammerError("Invalid setup for the drbd device") + if excl_stor: + raise errors.ProgrammerError("DRBD device requested with" + " exclusive_storage") + # check that the minor is unused + aminor = unique_id[4] + proc_info = cls._MassageProcData(cls._GetProcData()) + if aminor in proc_info: + status = DRBD8Status(proc_info[aminor]) + in_use = status.is_in_use + else: + in_use = False + if in_use: + _ThrowError("drbd%d: minor is already in use at Create() time", aminor) meta = children[1] meta.Assemble() if not meta.Attach(): - raise errors.BlockDeviceError("Can't attach to meta device") - if not cls._IsValidMeta(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device") - logger.Info("Started zeroing device %s" % meta.dev_path) - cls._ZeroDevice(meta.dev_path) - logger.Info("Done zeroing device %s" % meta.dev_path) - return cls(unique_id, children) + _ThrowError("drbd%d: can't attach to meta device '%s'", + aminor, meta) + cls._CheckMetaSize(meta.dev_path) + cls._InitMeta(aminor, meta.dev_path) + return cls(unique_id, children, size, params) + def Grow(self, amount, dryrun, backingstore): + """Resize the DRBD device and its backing storage. - def Remove(self): - """Stub remove for DRBD devices. + """ + if self.minor is None: + _ThrowError("drbd%d: Grow called while not attached", self._aminor) + if len(self._children) != 2 or None in self._children: + _ThrowError("drbd%d: cannot grow diskless device", self.minor) + self._children[0].Grow(amount, dryrun, backingstore) + if dryrun or backingstore: + # DRBD does not support dry-run mode and is not backing storage, + # so we'll return here + return + 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) + + +class FileStorage(BlockDev): + """File device. + + This class represents the a file storage backend device. + + The unique_id for the file device is a (file_driver, file_path) tuple. + + """ + def __init__(self, unique_id, children, size, params): + """Initalizes a file device backend. """ - return self.Shutdown() + if children: + raise errors.BlockDeviceError("Invalid setup for file device") + super(FileStorage, 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.driver = unique_id[0] + self.dev_path = unique_id[1] + CheckFileStoragePath(self.dev_path) -DEV_MAP = { - "lvm": LogicalVolume, - "md_raid1": MDRaid1, - "drbd": DRBDev, - } + self.Attach() + def Assemble(self): + """Assemble the device. -def FindDevice(dev_type, unique_id, children): - """Search for an existing, assembled device. + Checks whether the file device exists, raises BlockDeviceError otherwise. - This will succeed only if the device exists and is assembled, but it - does not do any actions in order to activate the device. + """ + if not os.path.exists(self.dev_path): + _ThrowError("File device '%s' does not exist" % self.dev_path) - """ - 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.Attach(): - return None - return device + def Shutdown(self): + """Shutdown the device. + This is a no-op for the file type, as we don't deactivate + the file on shutdown. -def AttachOrAssemble(dev_type, unique_id, children): - """Try to attach or assemble an existing device. + """ + pass + + def Open(self, force=False): + """Make the device ready for I/O. - This will attach to an existing assembled device or will assemble - the device, as needed, to bring it fully up. + This is a no-op for the file 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.Attach(): - device.Assemble() - if not device.Attach(): - raise errors.BlockDeviceError("Can't find a valid block device for" - " %s/%s/%s" % - (dev_type, unique_id, children)) - return device + """ + pass + + def Close(self): + """Notifies that the device will no longer be used for I/O. + This is a no-op for the file type. -def Create(dev_type, unique_id, children, size): - """Create a device. + """ + pass - """ - if dev_type not in DEV_MAP: - raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) - device = DEV_MAP[dev_type].Create(unique_id, children, size) + def Remove(self): + """Remove the file backing the block device. + + @rtype: boolean + @return: True if the removal was successful + + """ + try: + os.remove(self.dev_path) + except OSError, 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, dryrun, backingstore): + """Grow the file + + @param amount: the amount (in mebibytes) to grow with + + """ + if not backingstore: + return + # Check that the file exists + self.Assemble() + current_size = self.GetActualSize() + new_size = current_size + amount * 1024 * 1024 + assert new_size > current_size, "Cannot Grow with a negative amount" + # We can't really simulate the growth + if dryrun: + return + try: + f = open(self.dev_path, "a+") + f.truncate(new_size) + f.close() + except EnvironmentError, err: + _ThrowError("Error in file growth: %", str(err)) + + def Attach(self): + """Attach to an existing file. + + Check if this file already exists. + + @rtype: boolean + @return: True if file exists + + """ + 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, params, excl_stor): + """Create a new file. + + @param size: the size of file in MiB + + @rtype: L{bdev.FileStorage} + @return: an instance of FileStorage + + """ + if excl_stor: + raise errors.ProgrammerError("FileStorage device requested with" + " exclusive_storage") + 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] + + CheckFileStoragePath(dev_path) + + try: + fd = os.open(dev_path, os.O_RDWR | os.O_CREAT | os.O_EXCL) + f = os.fdopen(fd, "w") + f.truncate(size * 1024 * 1024) + f.close() + except EnvironmentError, err: + if err.errno == errno.EEXIST: + _ThrowError("File already existing: %s", dev_path) + _ThrowError("Error in file creation: %", str(err)) + + return FileStorage(unique_id, children, size, params) + + +class PersistentBlockDevice(BlockDev): + """A block device with persistent node + + May be either directly attached, or exposed through DM (e.g. dm-multipath). + udev helpers are probably required to give persistent, human-friendly + names. + + For the time being, pathnames are required to lie under /dev. + + """ + def __init__(self, unique_id, children, size, params): + """Attaches to a static block device. + + The unique_id is a path under /dev. + + """ + super(PersistentBlockDevice, 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.dev_path = unique_id[1] + if not os.path.realpath(self.dev_path).startswith("/dev/"): + raise ValueError("Full path '%s' lies outside /dev" % + os.path.realpath(self.dev_path)) + # TODO: this is just a safety guard checking that we only deal with devices + # we know how to handle. In the future this will be integrated with + # external storage backends and possible values will probably be collected + # from the cluster configuration. + if unique_id[0] != constants.BLOCKDEV_DRIVER_MANUAL: + raise ValueError("Got persistent block device of invalid type: %s" % + unique_id[0]) + + self.major = self.minor = None + self.Attach() + + @classmethod + def Create(cls, unique_id, children, size, params, excl_stor): + """Create a new device + + This is a noop, we only return a PersistentBlockDevice instance + + """ + if excl_stor: + raise errors.ProgrammerError("Persistent block device requested with" + " exclusive_storage") + return PersistentBlockDevice(unique_id, children, 0, params) + + def Remove(self): + """Remove a device + + This is a noop + + """ + pass + + def Rename(self, new_id): + """Rename this device. + + """ + _ThrowError("Rename is not supported for PersistentBlockDev storage") + + def Attach(self): + """Attach to an existing block device. + + + """ + self.attached = False + try: + st = os.stat(self.dev_path) + except OSError, err: + logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) + return False + + if not stat.S_ISBLK(st.st_mode): + logging.error("%s is not a block device", self.dev_path) + return False + + self.major = os.major(st.st_rdev) + self.minor = os.minor(st.st_rdev) + self.attached = True + + return True + + def Assemble(self): + """Assemble the device. + + """ + pass + + def Shutdown(self): + """Shutdown the device. + + """ + pass + + def Open(self, force=False): + """Make the device ready for I/O. + + """ + pass + + def Close(self): + """Notifies that the device will no longer be used for I/O. + + """ + pass + + def Grow(self, amount, dryrun, backingstore): + """Grow the logical volume. + + """ + _ThrowError("Grow is not supported for PersistentBlockDev storage") + + +class RADOSBlockDevice(BlockDev): + """A RADOS Block Device (rbd). + + This class implements the RADOS Block Device for the backend. You need + the rbd kernel driver, the RADOS Tools and a working RADOS cluster for + this to be functional. + + """ + def __init__(self, unique_id, children, size, params): + """Attaches to an rbd device. + + """ + super(RADOSBlockDevice, 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.driver, self.rbd_name = unique_id + + self.major = self.minor = None + self.Attach() + + @classmethod + def Create(cls, unique_id, children, size, params, excl_stor): + """Create a new rbd device. + + Provision a new rbd volume inside a RADOS pool. + + """ + if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: + raise errors.ProgrammerError("Invalid configuration data %s" % + str(unique_id)) + if excl_stor: + raise errors.ProgrammerError("RBD device requested with" + " exclusive_storage") + rbd_pool = params[constants.LDP_POOL] + rbd_name = unique_id[1] + + # Provision a new rbd volume (Image) inside the RADOS cluster. + cmd = [constants.RBD_CMD, "create", "-p", rbd_pool, + rbd_name, "--size", "%s" % size] + result = utils.RunCmd(cmd) + if result.failed: + _ThrowError("rbd creation failed (%s): %s", + result.fail_reason, result.output) + + return RADOSBlockDevice(unique_id, children, size, params) + + def Remove(self): + """Remove the rbd device. + + """ + rbd_pool = self.params[constants.LDP_POOL] + rbd_name = self.unique_id[1] + + if not self.minor and not self.Attach(): + # The rbd device doesn't exist. + return + + # First shutdown the device (remove mappings). + self.Shutdown() + + # Remove the actual Volume (Image) from the RADOS cluster. + cmd = [constants.RBD_CMD, "rm", "-p", rbd_pool, rbd_name] + result = utils.RunCmd(cmd) + if result.failed: + _ThrowError("Can't remove Volume from cluster with rbd rm: %s - %s", + result.fail_reason, result.output) + + def Rename(self, new_id): + """Rename this device. + + """ + pass + + def Attach(self): + """Attach to an existing rbd device. + + This method maps the rbd volume that matches our name with + an rbd device and then attaches to this device. + + """ + self.attached = False + + # Map the rbd volume to a block device under /dev + self.dev_path = self._MapVolumeToBlockdev(self.unique_id) + + try: + st = os.stat(self.dev_path) + except OSError, err: + logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) + return False + + if not stat.S_ISBLK(st.st_mode): + logging.error("%s is not a block device", self.dev_path) + return False + + self.major = os.major(st.st_rdev) + self.minor = os.minor(st.st_rdev) + self.attached = True + + return True + + def _MapVolumeToBlockdev(self, unique_id): + """Maps existing rbd volumes to block devices. + + This method should be idempotent if the mapping already exists. + + @rtype: string + @return: the block device path that corresponds to the volume + + """ + pool = self.params[constants.LDP_POOL] + name = unique_id[1] + + # Check if the mapping already exists. + rbd_dev = self._VolumeToBlockdev(pool, name) + if rbd_dev: + # The mapping exists. Return it. + return rbd_dev + + # The mapping doesn't exist. Create it. + map_cmd = [constants.RBD_CMD, "map", "-p", pool, name] + result = utils.RunCmd(map_cmd) + if result.failed: + _ThrowError("rbd map failed (%s): %s", + result.fail_reason, result.output) + + # Find the corresponding rbd device. + rbd_dev = self._VolumeToBlockdev(pool, name) + if not rbd_dev: + _ThrowError("rbd map succeeded, but could not find the rbd block" + " device in output of showmapped, for volume: %s", name) + + # The device was successfully mapped. Return it. + return rbd_dev + + @classmethod + def _VolumeToBlockdev(cls, pool, volume_name): + """Do the 'volume name'-to-'rbd block device' resolving. + + @type pool: string + @param pool: RADOS pool to use + @type volume_name: string + @param volume_name: the name of the volume whose device we search for + @rtype: string or None + @return: block device path if the volume is mapped, else None + + """ + try: + # Newer versions of the rbd tool support json output formatting. Use it + # if available. + showmap_cmd = [ + constants.RBD_CMD, + "showmapped", + "-p", + pool, + "--format", + "json" + ] + result = utils.RunCmd(showmap_cmd) + if result.failed: + logging.error("rbd JSON output formatting returned error (%s): %s," + "falling back to plain output parsing", + result.fail_reason, result.output) + raise RbdShowmappedJsonError + + return cls._ParseRbdShowmappedJson(result.output, volume_name) + except RbdShowmappedJsonError: + # For older versions of rbd, we have to parse the plain / text output + # manually. + showmap_cmd = [constants.RBD_CMD, "showmapped", "-p", pool] + result = utils.RunCmd(showmap_cmd) + if result.failed: + _ThrowError("rbd showmapped failed (%s): %s", + result.fail_reason, result.output) + + return cls._ParseRbdShowmappedPlain(result.output, volume_name) + + @staticmethod + def _ParseRbdShowmappedJson(output, volume_name): + """Parse the json output of `rbd showmapped'. + + This method parses the json output of `rbd showmapped' and returns the rbd + block device path (e.g. /dev/rbd0) that matches the given rbd volume. + + @type output: string + @param output: the json output of `rbd showmapped' + @type volume_name: string + @param volume_name: the name of the volume whose device we search for + @rtype: string or None + @return: block device path if the volume is mapped, else None + + """ + try: + devices = serializer.LoadJson(output) + except ValueError, err: + _ThrowError("Unable to parse JSON data: %s" % err) + + rbd_dev = None + for d in devices.values(): # pylint: disable=E1103 + try: + name = d["name"] + except KeyError: + _ThrowError("'name' key missing from json object %s", devices) + + if name == volume_name: + if rbd_dev is not None: + _ThrowError("rbd volume %s is mapped more than once", volume_name) + + rbd_dev = d["device"] + + return rbd_dev + + @staticmethod + def _ParseRbdShowmappedPlain(output, volume_name): + """Parse the (plain / text) output of `rbd showmapped'. + + This method parses the output of `rbd showmapped' and returns + the rbd block device path (e.g. /dev/rbd0) that matches the + given rbd volume. + + @type output: string + @param output: the plain text output of `rbd showmapped' + @type volume_name: string + @param volume_name: the name of the volume whose device we search for + @rtype: string or None + @return: block device path if the volume is mapped, else None + + """ + allfields = 5 + volumefield = 2 + devicefield = 4 + + lines = output.splitlines() + + # Try parsing the new output format (ceph >= 0.55). + splitted_lines = map(lambda l: l.split(), lines) + + # Check for empty output. + if not splitted_lines: + return None + + # Check showmapped output, to determine number of fields. + field_cnt = len(splitted_lines[0]) + if field_cnt != allfields: + # Parsing the new format failed. Fallback to parsing the old output + # format (< 0.55). + splitted_lines = map(lambda l: l.split("\t"), lines) + if field_cnt != allfields: + _ThrowError("Cannot parse rbd showmapped output expected %s fields," + " found %s", allfields, field_cnt) + + matched_lines = \ + filter(lambda l: len(l) == allfields and l[volumefield] == volume_name, + splitted_lines) + + if len(matched_lines) > 1: + _ThrowError("rbd volume %s mapped more than once", volume_name) + + if matched_lines: + # rbd block device found. Return it. + rbd_dev = matched_lines[0][devicefield] + return rbd_dev + + # The given volume is not mapped. + return None + + def Assemble(self): + """Assemble the device. + + """ + pass + + def Shutdown(self): + """Shutdown the device. + + """ + if not self.minor and not self.Attach(): + # The rbd device doesn't exist. + return + + # Unmap the block device from the Volume. + self._UnmapVolumeFromBlockdev(self.unique_id) + + self.minor = None + self.dev_path = None + + def _UnmapVolumeFromBlockdev(self, unique_id): + """Unmaps the rbd device from the Volume it is mapped. + + Unmaps the rbd device from the Volume it was previously mapped to. + This method should be idempotent if the Volume isn't mapped. + + """ + pool = self.params[constants.LDP_POOL] + name = unique_id[1] + + # Check if the mapping already exists. + rbd_dev = self._VolumeToBlockdev(pool, name) + + if rbd_dev: + # The mapping exists. Unmap the rbd device. + unmap_cmd = [constants.RBD_CMD, "unmap", "%s" % rbd_dev] + result = utils.RunCmd(unmap_cmd) + if result.failed: + _ThrowError("rbd unmap failed (%s): %s", + result.fail_reason, result.output) + + def Open(self, force=False): + """Make the device ready for I/O. + + """ + pass + + def Close(self): + """Notifies that the device will no longer be used for I/O. + + """ + pass + + def Grow(self, amount, dryrun, backingstore): + """Grow the Volume. + + @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 + + """ + if not backingstore: + return + if not self.Attach(): + _ThrowError("Can't attach to rbd device during Grow()") + + if dryrun: + # the rbd tool does not support dry runs of resize operations. + # Since rbd volumes are thinly provisioned, we assume + # there is always enough free space for the operation. + return + + rbd_pool = self.params[constants.LDP_POOL] + rbd_name = self.unique_id[1] + new_size = self.size + amount + + # Resize the rbd volume (Image) inside the RADOS cluster. + cmd = [constants.RBD_CMD, "resize", "-p", rbd_pool, + rbd_name, "--size", "%s" % new_size] + result = utils.RunCmd(cmd) + if result.failed: + _ThrowError("rbd resize failed (%s): %s", + result.fail_reason, result.output) + + +class ExtStorageDevice(BlockDev): + """A block device provided by an ExtStorage Provider. + + This class implements the External Storage Interface, which means + handling of the externally provided block devices. + + """ + def __init__(self, unique_id, children, size, params): + """Attaches to an extstorage block device. + + """ + super(ExtStorageDevice, 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.driver, self.vol_name = unique_id + self.ext_params = params + + self.major = self.minor = None + self.Attach() + + @classmethod + def Create(cls, unique_id, children, size, params, excl_stor): + """Create a new extstorage device. + + Provision a new volume using an extstorage provider, which will + then be mapped to a block device. + + """ + if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: + raise errors.ProgrammerError("Invalid configuration data %s" % + str(unique_id)) + if excl_stor: + raise errors.ProgrammerError("extstorage device requested with" + " exclusive_storage") + + # Call the External Storage's create script, + # to provision a new Volume inside the External Storage + _ExtStorageAction(constants.ES_ACTION_CREATE, unique_id, + params, str(size)) + + return ExtStorageDevice(unique_id, children, size, params) + + def Remove(self): + """Remove the extstorage device. + + """ + if not self.minor and not self.Attach(): + # The extstorage device doesn't exist. + return + + # First shutdown the device (remove mappings). + self.Shutdown() + + # Call the External Storage's remove script, + # to remove the Volume from the External Storage + _ExtStorageAction(constants.ES_ACTION_REMOVE, self.unique_id, + self.ext_params) + + def Rename(self, new_id): + """Rename this device. + + """ + pass + + def Attach(self): + """Attach to an existing extstorage device. + + This method maps the extstorage volume that matches our name with + a corresponding block device and then attaches to this device. + + """ + self.attached = False + + # Call the External Storage's attach script, + # to attach an existing Volume to a block device under /dev + self.dev_path = _ExtStorageAction(constants.ES_ACTION_ATTACH, + self.unique_id, self.ext_params) + + try: + st = os.stat(self.dev_path) + except OSError, err: + logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) + return False + + if not stat.S_ISBLK(st.st_mode): + logging.error("%s is not a block device", self.dev_path) + return False + + self.major = os.major(st.st_rdev) + self.minor = os.minor(st.st_rdev) + self.attached = True + + return True + + def Assemble(self): + """Assemble the device. + + """ + pass + + def Shutdown(self): + """Shutdown the device. + + """ + if not self.minor and not self.Attach(): + # The extstorage device doesn't exist. + return + + # Call the External Storage's detach script, + # to detach an existing Volume from it's block device under /dev + _ExtStorageAction(constants.ES_ACTION_DETACH, self.unique_id, + self.ext_params) + + self.minor = None + self.dev_path = None + + def Open(self, force=False): + """Make the device ready for I/O. + + """ + pass + + def Close(self): + """Notifies that the device will no longer be used for I/O. + + """ + pass + + def Grow(self, amount, dryrun, backingstore): + """Grow the Volume. + + @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 + + """ + if not backingstore: + return + if not self.Attach(): + _ThrowError("Can't attach to extstorage device during Grow()") + + if dryrun: + # we do not support dry runs of resize operations for now. + return + + new_size = self.size + amount + + # Call the External Storage's grow script, + # to grow an existing Volume inside the External Storage + _ExtStorageAction(constants.ES_ACTION_GROW, self.unique_id, + self.ext_params, str(self.size), grow=str(new_size)) + + def SetInfo(self, text): + """Update metadata with info text. + + """ + # Replace invalid characters + text = re.sub("^[^A-Za-z0-9_+.]", "_", text) + text = re.sub("[^-A-Za-z0-9_+.]", "_", text) + + # Only up to 128 characters are allowed + text = text[:128] + + # Call the External Storage's setinfo script, + # to set metadata for an existing Volume inside the External Storage + _ExtStorageAction(constants.ES_ACTION_SETINFO, self.unique_id, + self.ext_params, metadata=text) + + +def _ExtStorageAction(action, unique_id, ext_params, + size=None, grow=None, metadata=None): + """Take an External Storage action. + + Take an External Storage action concerning or affecting + a specific Volume inside the External Storage. + + @type action: string + @param action: which action to perform. One of: + create / remove / grow / attach / detach + @type unique_id: tuple (driver, vol_name) + @param unique_id: a tuple containing the type of ExtStorage (driver) + and the Volume name + @type ext_params: dict + @param ext_params: ExtStorage parameters + @type size: integer + @param size: the size of the Volume in mebibytes + @type grow: integer + @param grow: the new size in mebibytes (after grow) + @type metadata: string + @param metadata: metadata info of the Volume, for use by the provider + @rtype: None or a block device path (during attach) + + """ + driver, vol_name = unique_id + + # Create an External Storage instance of type `driver' + status, inst_es = ExtStorageFromDisk(driver) + if not status: + _ThrowError("%s" % inst_es) + + # Create the basic environment for the driver's scripts + create_env = _ExtStorageEnvironment(unique_id, ext_params, size, + grow, metadata) + + # Do not use log file for action `attach' as we need + # to get the output from RunResult + # TODO: find a way to have a log file for attach too + logfile = None + if action is not constants.ES_ACTION_ATTACH: + logfile = _VolumeLogName(action, driver, vol_name) + + # Make sure the given action results in a valid script + if action not in constants.ES_SCRIPTS: + _ThrowError("Action '%s' doesn't result in a valid ExtStorage script" % + action) + + # Find out which external script to run according the given action + script_name = action + "_script" + script = getattr(inst_es, script_name) + + # Run the external script + result = utils.RunCmd([script], env=create_env, + cwd=inst_es.path, output=logfile,) + if result.failed: + logging.error("External storage's %s command '%s' returned" + " error: %s, logfile: %s, output: %s", + action, result.cmd, result.fail_reason, + logfile, result.output) + + # If logfile is 'None' (during attach), it breaks TailFile + # TODO: have a log file for attach too + if action is not constants.ES_ACTION_ATTACH: + lines = [utils.SafeEncode(val) + for val in utils.TailFile(logfile, lines=20)] + else: + lines = result.output[-20:] + + _ThrowError("External storage's %s script failed (%s), last" + " lines of output:\n%s", + action, result.fail_reason, "\n".join(lines)) + + if action == constants.ES_ACTION_ATTACH: + return result.stdout + + +def ExtStorageFromDisk(name, base_dir=None): + """Create an ExtStorage instance from disk. + + This function will return an ExtStorage instance + if the given name is a valid ExtStorage name. + + @type base_dir: string + @keyword base_dir: Base directory containing ExtStorage installations. + Defaults to a search in all the ES_SEARCH_PATH dirs. + @rtype: tuple + @return: True and the ExtStorage instance if we find a valid one, or + False and the diagnose message on error + + """ + if base_dir is None: + es_base_dir = pathutils.ES_SEARCH_PATH + else: + es_base_dir = [base_dir] + + es_dir = utils.FindFile(name, es_base_dir, os.path.isdir) + + if es_dir is None: + return False, ("Directory for External Storage Provider %s not" + " found in search path" % name) + + # ES Files dictionary, we will populate it with the absolute path + # names; if the value is True, then it is a required file, otherwise + # an optional one + es_files = dict.fromkeys(constants.ES_SCRIPTS, True) + + es_files[constants.ES_PARAMETERS_FILE] = True + + for (filename, _) in es_files.items(): + es_files[filename] = utils.PathJoin(es_dir, filename) + + try: + st = os.stat(es_files[filename]) + except EnvironmentError, err: + return False, ("File '%s' under path '%s' is missing (%s)" % + (filename, es_dir, utils.ErrnoOrStr(err))) + + if not stat.S_ISREG(stat.S_IFMT(st.st_mode)): + return False, ("File '%s' under path '%s' is not a regular file" % + (filename, es_dir)) + + if filename in constants.ES_SCRIPTS: + if stat.S_IMODE(st.st_mode) & stat.S_IXUSR != stat.S_IXUSR: + return False, ("File '%s' under path '%s' is not executable" % + (filename, es_dir)) + + parameters = [] + if constants.ES_PARAMETERS_FILE in es_files: + parameters_file = es_files[constants.ES_PARAMETERS_FILE] + try: + parameters = utils.ReadFile(parameters_file).splitlines() + except EnvironmentError, err: + return False, ("Error while reading the EXT parameters file at %s: %s" % + (parameters_file, utils.ErrnoOrStr(err))) + parameters = [v.split(None, 1) for v in parameters] + + es_obj = \ + objects.ExtStorage(name=name, path=es_dir, + create_script=es_files[constants.ES_SCRIPT_CREATE], + remove_script=es_files[constants.ES_SCRIPT_REMOVE], + grow_script=es_files[constants.ES_SCRIPT_GROW], + attach_script=es_files[constants.ES_SCRIPT_ATTACH], + detach_script=es_files[constants.ES_SCRIPT_DETACH], + setinfo_script=es_files[constants.ES_SCRIPT_SETINFO], + verify_script=es_files[constants.ES_SCRIPT_VERIFY], + supported_parameters=parameters) + return True, es_obj + + +def _ExtStorageEnvironment(unique_id, ext_params, + size=None, grow=None, metadata=None): + """Calculate the environment for an External Storage script. + + @type unique_id: tuple (driver, vol_name) + @param unique_id: ExtStorage pool and name of the Volume + @type ext_params: dict + @param ext_params: the EXT parameters + @type size: string + @param size: size of the Volume (in mebibytes) + @type grow: string + @param grow: new size of Volume after grow (in mebibytes) + @type metadata: string + @param metadata: metadata info of the Volume + @rtype: dict + @return: dict of environment variables + + """ + vol_name = unique_id[1] + + result = {} + result["VOL_NAME"] = vol_name + + # EXT params + for pname, pvalue in ext_params.items(): + result["EXTP_%s" % pname.upper()] = str(pvalue) + + if size is not None: + result["VOL_SIZE"] = size + + if grow is not None: + result["VOL_NEW_SIZE"] = grow + + if metadata is not None: + result["VOL_METADATA"] = metadata + + return result + + +def _VolumeLogName(kind, es_name, volume): + """Compute the ExtStorage log filename for a given Volume and operation. + + @type kind: string + @param kind: the operation type (e.g. create, remove etc.) + @type es_name: string + @param es_name: the ExtStorage name + @type volume: string + @param volume: the name of the Volume inside the External Storage + + """ + # Check if the extstorage log dir is a valid dir + if not os.path.isdir(pathutils.LOG_ES_DIR): + _ThrowError("Cannot find log directory: %s", pathutils.LOG_ES_DIR) + + # TODO: Use tempfile.mkstemp to create unique filename + base = ("%s-%s-%s-%s.log" % + (kind, es_name, volume, utils.TimestampForFilename())) + return utils.PathJoin(pathutils.LOG_ES_DIR, base) + + +DEV_MAP = { + constants.LD_LV: LogicalVolume, + constants.LD_DRBD8: DRBD8, + constants.LD_BLOCKDEV: PersistentBlockDevice, + constants.LD_RBD: RADOSBlockDevice, + constants.LD_EXT: ExtStorageDevice, + } + +if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE: + DEV_MAP[constants.LD_FILE] = FileStorage + + +def _VerifyDiskType(dev_type): + if dev_type not in DEV_MAP: + raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) + + +def _VerifyDiskParams(disk): + """Verifies if all disk parameters are set. + + """ + missing = set(constants.DISK_LD_DEFAULTS[disk.dev_type]) - set(disk.params) + if missing: + raise errors.ProgrammerError("Block device is missing disk parameters: %s" % + missing) + + +def FindDevice(disk, children): + """Search for an existing, assembled device. + + This will succeed only if the device exists and is assembled, but it + does not do any actions in order to activate the device. + + @type disk: L{objects.Disk} + @param disk: the disk object to find + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + + """ + _VerifyDiskType(disk.dev_type) + device = DEV_MAP[disk.dev_type](disk.physical_id, children, disk.size, + disk.params) + if not device.attached: + return None + return device + + +def Assemble(disk, children): + """Try to attach or assemble an existing device. + + This will attach to assemble the device, as needed, to bring it + fully up. It must be safe to run on already-assembled devices. + + @type disk: L{objects.Disk} + @param disk: the disk object to assemble + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + + """ + _VerifyDiskType(disk.dev_type) + _VerifyDiskParams(disk) + device = DEV_MAP[disk.dev_type](disk.physical_id, children, disk.size, + disk.params) + device.Assemble() + return device + + +def Create(disk, children, excl_stor): + """Create a device. + + @type disk: L{objects.Disk} + @param disk: the disk object to create + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + @type excl_stor: boolean + @param excl_stor: Whether exclusive_storage is active + + """ + _VerifyDiskType(disk.dev_type) + _VerifyDiskParams(disk) + device = DEV_MAP[disk.dev_type].Create(disk.physical_id, children, disk.size, + disk.params, excl_stor) return device