X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/3840729daa88e268c09cf64665b6769d426a40c8..cea881e554d3bb44d8ea83db6275ceb124915c34:/lib/bdev.py diff --git a/lib/bdev.py b/lib/bdev.py index 8b27971..5469283 100644 --- a/lib/bdev.py +++ b/lib/bdev.py @@ -25,11 +25,64 @@ import re import time import errno import pyparsing as pyp +import os +import logging from ganeti import utils -from ganeti import logger from ganeti import errors from ganeti import constants +from ganeti import objects +from ganeti import compat + + +# Size of reads in _CanReadDevice +_DEVICE_READ_SIZE = 128 * 1024 + + +def _IgnoreError(fn, *args, **kwargs): + """Executes the given function, ignoring BlockDeviceErrors. + + This is used in order to simplify the execution of cleanup or + rollback functions. + + @rtype: boolean + @return: True when fn didn't raise an exception, False otherwise + + """ + try: + fn(*args, **kwargs) + return True + except errors.BlockDeviceError, err: + logging.warning("Caught BlockDeviceError but ignoring: %s", str(err)) + return False + + +def _ThrowError(msg, *args): + """Log an error to the node daemon and the raise an exception. + + @type msg: string + @param msg: the text of the exception + @raise errors.BlockDeviceError + + """ + if args: + msg = msg % args + logging.error(msg) + raise errors.BlockDeviceError(msg) + + +def _CanReadDevice(path): + """Check if we can read from the given device. + + This tries to read the first 128k of the device. + + """ + try: + utils.ReadFile(path, size=_DEVICE_READ_SIZE) + return True + except EnvironmentError: + logging.warning("Can't read from device %s", path, exc_info=True) + return False class BlockDev(object): @@ -42,21 +95,16 @@ 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) @@ -64,15 +112,13 @@ 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 @@ -82,52 +128,28 @@ class BlockDev(object): 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): self._children = children self.dev_path = None self.unique_id = unique_id self.major = None self.minor = None + self.attached = False + self.size = size def Assemble(self): """Assemble the device from its components. - 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. @@ -158,9 +180,9 @@ class BlockDev(object): 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 @@ -173,12 +195,6 @@ class BlockDev(object): """ raise NotImplementedError - def GetStatus(self): - """Return the status of the device. - - """ - raise NotImplementedError - def Open(self, force=False): """Make the device ready for use. @@ -219,9 +235,6 @@ class BlockDev(object): If this device is a mirroring device, this function returns the status of the mirror. - Returns: - (sync_percent, estimated_time, is_degraded, ldisk) - If sync_percent is None, it means the device is not syncing. If estimated_time is None, it means we can't estimate @@ -235,9 +248,16 @@ class BlockDev(object): data. This is only valid for some devices, the rest will always return False (not degraded). - """ - return None, None, False, False + @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. @@ -246,22 +266,44 @@ class BlockDev(object): minimum percent and maximum time are calculated across our children. + @rtype: objects.BlockDevStatus + """ - min_percent, max_time, is_degraded, ldisk = self.GetSyncStatus() + status = self.GetSyncStatus() + + min_percent = status.sync_percent + max_time = status.estimated_time + is_degraded = status.is_degraded + ldisk_status = status.ldisk_status + if self._children: for child in self._children: - c_percent, c_time, c_degraded, c_ldisk = 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 - ldisk = ldisk or c_ldisk - return min_percent, max_time, is_degraded, ldisk + max_time = child_status.estimated_time + elif child_status.estimated_time is not None: + max_time = max(max_time, child_status.estimated_time) + + is_degraded = is_degraded or child_status.is_degraded + + if ldisk_status is None: + ldisk_status = child_status.ldisk_status + elif child_status.ldisk_status is not None: + ldisk_status = max(ldisk_status, child_status.ldisk_status) + + return objects.BlockDevStatus(dev_path=self.dev_path, + major=self.major, + minor=self.minor, + sync_percent=min_percent, + estimated_time=max_time, + is_degraded=is_degraded, + ldisk_status=ldisk_status) def SetInfo(self, text): @@ -273,6 +315,30 @@ class BlockDev(object): for child in self._children: child.SetInfo(text) + def Grow(self, amount): + """Grow the block device. + + @param amount: the amount (in mebibytes) to grow with + + """ + 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>" % @@ -284,17 +350,25 @@ class LogicalVolume(BlockDev): """Logical Volume block device. """ - def __init__(self, unique_id, children): + _VALID_NAME_RE = re.compile("^[a-zA-Z0-9+_.-]*$") + _INVALID_NAMES = frozenset([".", "..", "snapshot", "pvmove"]) + _INVALID_SUBSTRINGS = frozenset(["_mlog", "_mimage"]) + + def __init__(self, unique_id, children, size): """Attaches to a LV device. The unique_id is a tuple (vg_name, lv_name) """ - super(LogicalVolume, self).__init__(unique_id, children) + super(LogicalVolume, self).__init__(unique_id, children, size) if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) self._vg_name, self._lv_name = unique_id - 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() @classmethod @@ -303,73 +377,106 @@ class LogicalVolume(BlockDev): """ 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) + _ThrowError("Can't compute PV info for vg %s", vg_name) pvs_info.sort() pvs_info.reverse() pvlist = [ pv[1] for pv in pvs_info ] + if compat.any(pvlist, lambda v: ":" in v): + _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'") free_size = sum([ pv[0] for pv in pvs_info ]) + current_pvs = len(pvlist) + stripes = min(current_pvs, constants.LVM_STRIPECOUNT) # The size constraint should have been checked from the master before # calling the create function. if free_size < size: - 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) + _ThrowError("Not enough free space: required %s," + " available %s", size, free_size) + cmd = ["lvcreate", "-L%dm" % size, "-n%s" % lv_name] + # If the free space is not well distributed, we won't be able to + # create an optimally-striped volume; in that case, we want to try + # with N, N-1, ..., 2, and finally 1 (non-stripped) number of + # stripes + for stripes_arg in range(stripes, 0, -1): + result = utils.RunCmd(cmd + ["-i%d" % stripes_arg] + [vg_name] + pvlist) + if not result.failed: + break if result.failed: - 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) @staticmethod - def GetPVInfo(vg_name): + def GetPVInfo(vg_names, filter_allocatable=True): """Get the free space info for PVs in a volume group. - Args: - vg_name: the volume group name + @param vg_names: list of volume group names, if empty all will be returned + @param filter_allocatable: whether to skip over unallocatable PVs - Returns: - list of (free_space, name) with free_space in mebibytes + @rtype: list + @return: list of tuples (free_space, name) with free_space in mebibytes """ + sep = "|" command = ["pvs", "--noheadings", "--nosuffix", "--units=m", "-opv_name,vg_name,pv_free,pv_attr", "--unbuffered", - "--separator=:"] + "--separator=%s" % sep ] result = utils.RunCmd(command) if result.failed: - logger.Error("Can't get the PV information: %s" % result.fail_reason) + logging.error("Can't get the PV information: %s - %s", + result.fail_reason, result.output) return None data = [] for line in result.stdout.splitlines(): - fields = line.strip().split(':') + fields = line.strip().split(sep) if len(fields) != 4: - logger.Error("Can't parse pvs output: line '%s'" % line) + logging.error("Can't parse pvs output: line '%s'", line) return None - # skip over pvs from another vg or ones which are not allocatable - if fields[1] != vg_name or fields[3][0] != 'a': + # (possibly) skip over pvs which are not allocatable + if filter_allocatable and fields[3][0] != 'a': continue - data.append((float(fields[2]), fields[0])) + # (possibly) skip over pvs which are not in the right volume group(s) + if vg_names and fields[1] not in vg_names: + continue + data.append((float(fields[2]), fields[0], fields[1])) 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(cls._INVALID_SUBSTRINGS, lambda x: x in name)): + _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) - - return not result.failed + _ThrowError("Can't lvremove: %s - %s", result.fail_reason, result.output) def Rename(self, new_id): """Rename this logical volume. @@ -384,42 +491,86 @@ class LogicalVolume(BlockDev): (self._vg_name, new_vg)) result = utils.RunCmd(["lvrename", new_vg, self._lv_name, new_name]) if result.failed: - raise errors.BlockDeviceError("Failed to rename the logical volume: %s" % - result.output) + _ThrowError("Failed to rename the logical volume: %s", result.output) self._lv_name = new_name - self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name) - + self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name) def Attach(self): """Attach to an existing LV. 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=m", "--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, %s" % - (self.dev_path, result.fail_reason, result.output)) + 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 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 - 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 + + try: + stripes = int(stripes) + except (TypeError, ValueError), err: + logging.error("Can't parse the number of stripes: %s", err) + return False + + self.major = major + self.minor = minor + self.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. @@ -428,35 +579,7 @@ class LogicalVolume(BlockDev): volumes on shutdown. """ - return True - - def GetStatus(self): - """Return the status of the device. - - 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. - - """ - 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 - else: - retval = self.STATUS_EXISTING - - return retval + pass def GetSyncStatus(self): """Returns the sync status of the device. @@ -464,9 +587,6 @@ class LogicalVolume(BlockDev): If this device is a mirroring device, this function returns the status of the mirror. - Returns: - (sync_percent, estimated_time, is_degraded, ldisk) - 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 @@ -478,19 +598,23 @@ class LogicalVolume(BlockDev): physical disk failure and subsequent 'vgreduce --removemissing' on the volume group. + 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 None, None, True, True - out = result.stdout.strip() - # format: type/permissions/alloc/fixed_minor/state/open - if len(out) != 6: - logger.Debug("Error in lvs output: attrs=%s, len != 6" % out) - return None, None, True, True - ldisk = out[0] == 'v' # virtual volume, i.e. doesn't have - # backing storage - return None, None, ldisk, ldisk + if self._degraded: + ldisk_status = constants.LDS_FAULTY + else: + ldisk_status = constants.LDS_OKAY + + return objects.BlockDevStatus(dev_path=self.dev_path, + major=self.major, + minor=self.minor, + sync_percent=None, + estimated_time=None, + is_degraded=self._degraded, + ldisk_status=ldisk_status) def Open(self, force=False): """Make the device ready for I/O. @@ -498,7 +622,7 @@ 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. @@ -506,7 +630,7 @@ 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. @@ -515,25 +639,24 @@ class LogicalVolume(BlockDev): 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) + _IgnoreError(snap.Remove) - pvs_info = self.GetPVInfo(self._vg_name) + 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) + _ThrowError("Can't compute PV info for vg %s", self._vg_name) pvs_info.sort() pvs_info.reverse() - free_size, pv_name = pvs_info[0] + free_size, _, _ = pvs_info[0] if free_size < size: - raise errors.BlockDeviceError("Not enough free space: required %s," - " available %s" % (size, free_size)) + _ThrowError("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)) + _ThrowError("command: %s error: %s - %s", + result.cmd, result.fail_reason, result.output) return snap_name @@ -553,502 +676,249 @@ class LogicalVolume(BlockDev): result = utils.RunCmd(["lvchange", "--addtag", text, self.dev_path]) if result.failed: - raise errors.BlockDeviceError("Command: %s error: %s" % - (result.cmd, result.fail_reason)) - - -class MDRaid1(BlockDev): - """raid1 device implemented via md. - - """ - def __init__(self, unique_id, children): - super(MDRaid1, self).__init__(unique_id, children) - self.major = 9 - self.Attach() - - def Attach(self): - """Find an array which matches our config and attach to it. + _ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason, + result.output) - This tries to find a MD array which has the same UUID as our own. + def Grow(self, amount): + """Grow the logical volume. """ - 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) - - @staticmethod - def _GetUsedDevs(): - """Compute the list of in-use MD devices. + if self.pe_size is None or self.stripe_count is None: + if not self.Attach(): + _ThrowError("Can't attach to LV during Grow()") + full_stripe_size = self.pe_size * self.stripe_count + rest = amount % full_stripe_size + if rest != 0: + amount += full_stripe_size - rest + # we try multiple algorithms since the 'best' ones might not have + # space available in the right place, but later ones might (since + # they have less constraints); also note that only recent LVM + # supports 'cling' + for alloc_policy in "contiguous", "cling", "normal": + result = utils.RunCmd(["lvextend", "--alloc", alloc_policy, + "-L", "+%dm" % amount, self.dev_path]) + if not result.failed: + return + _ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) - It doesn't matter if the used device have other raid level, just - that they are in use. - - """ - mdstat = open("/proc/mdstat", "r") - data = mdstat.readlines() - mdstat.close() - - 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 - return used_md +class DRBD8Status(object): + """A DRBD status representation class. - @staticmethod - def _GetDevInfo(minor): - """Get info about a MD device. + Note that this doesn't support unconfigured devices (cs:Unconfigured). - Currently only uuid is returned. + """ + 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.]+)%.*" + "\sfinish: ([0-9]+):([0-9]+):([0-9]+)\s.*$") + + CS_UNCONFIGURED = "Unconfigured" + CS_STANDALONE = "StandAlone" + CS_WFCONNECTION = "WFConnection" + CS_WFREPORTPARAMS = "WFReportParams" + CS_CONNECTED = "Connected" + CS_STARTINGSYNCS = "StartingSyncS" + CS_STARTINGSYNCT = "StartingSyncT" + CS_WFBITMAPS = "WFBitMapS" + CS_WFBITMAPT = "WFBitMapT" + CS_WFSYNCUUID = "WFSyncUUID" + CS_SYNCSOURCE = "SyncSource" + CS_SYNCTARGET = "SyncTarget" + CS_PAUSEDSYNCS = "PausedSyncS" + CS_PAUSEDSYNCT = "PausedSyncT" + CSET_SYNC = frozenset([ + CS_WFREPORTPARAMS, + CS_STARTINGSYNCS, + CS_STARTINGSYNCT, + CS_WFBITMAPS, + CS_WFBITMAPT, + CS_WFSYNCUUID, + CS_SYNCSOURCE, + CS_SYNCTARGET, + CS_PAUSEDSYNCS, + CS_PAUSEDSYNCT, + ]) + + DS_DISKLESS = "Diskless" + DS_ATTACHING = "Attaching" # transient state + DS_FAILED = "Failed" # transient state, next: diskless + DS_NEGOTIATING = "Negotiating" # transient state + DS_INCONSISTENT = "Inconsistent" # while syncing or after creation + DS_OUTDATED = "Outdated" + DS_DUNKNOWN = "DUnknown" # shown for peer disk when not connected + DS_CONSISTENT = "Consistent" + DS_UPTODATE = "UpToDate" # normal state + + RO_PRIMARY = "Primary" + RO_SECONDARY = "Secondary" + RO_UNKNOWN = "Unknown" + + def __init__(self, procline): + u = self.UNCONF_RE.match(procline) + if u: + 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 - """ - 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].split()[0] - elif kv[0] == "State": - retval["state"] = kv[1].split(", ") - return retval - @staticmethod - def _FindUnusedMinor(): - """Compute an unused MD minor. +class BaseDRBD(BlockDev): # pylint: disable-msg=W0223 + """Base DRBD class. - This code assumes that there are 256 minors only. + This class contains a few bits of common functionality between the + 0.7 and 8.x versions of DRBD. - """ - 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.") - raise errors.BlockDeviceError("Can't find a free MD minor") - return i + """ + _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)" + r" \(api:(\d+)/proto:(\d+)(?:-(\d+))?\)") + _VALID_LINE_RE = re.compile("^ *([0-9]+): cs:([^ ]+).*$") + _UNUSED_LINE_RE = re.compile("^ *([0-9]+): cs:Unconfigured$") - @classmethod - def _FindMDByUUID(cls, uuid): - """Find the minor of an MD array with a given UUID. + _DRBD_MAJOR = 147 + _ST_UNCONFIGURED = "Unconfigured" + _ST_WFCONNECTION = "WFConnection" + _ST_CONNECTED = "Connected" - """ - md_list = cls._GetUsedDevs() - for minor in md_list: - info = cls._GetDevInfo(minor) - if info and info["uuid"] == uuid: - return minor - return None + _STATUS_FILE = "/proc/drbd" @staticmethod - def _ZeroSuperblock(dev_path): - """Zero the possible locations for an MD superblock. - - The zero-ing can't be done via ``mdadm --zero-superblock`` as that - fails in versions 2.x with the same error code as non-writable - device. - - The superblocks are located at (negative values are relative to - the end of the block device): - - -128k to end for version 0.90 superblock - - -8k to -12k for version 1.0 superblock (included in the above) - - 0k to 4k for version 1.1 superblock - - 4k to 8k for version 1.2 superblock - - To cover all situations, the zero-ing will be: - - 0k to 128k - - -128k to end - - As such, the minimum device size must be 128k, otherwise we'll get - I/O errors. - - Note that this function depends on the fact that one can open, - read and write block devices normally. + def _GetProcData(filename=_STATUS_FILE): + """Return data from /proc/drbd. """ - overwrite_size = 128 * 1024 - empty_buf = '\0' * overwrite_size - fd = open(dev_path, "r+") try: - fd.seek(0, 0) - p1 = fd.tell() - fd.write(empty_buf) - p2 = fd.tell() - logger.Debug("Zeroed %s from %d to %d" % (dev_path, p1, p2)) - fd.seek(-overwrite_size, 2) - p1 = fd.tell() - fd.write(empty_buf) - p2 = fd.tell() - logger.Debug("Zeroed %s from %d to %d" % (dev_path, p1, p2)) - finally: - fd.close() + data = utils.ReadFile(filename).splitlines() + except EnvironmentError, err: + if err.errno == errno.ENOENT: + _ThrowError("The file %s cannot be opened, check if the module" + " is loaded (%s)", filename, str(err)) + else: + _ThrowError("Can't read the DRBD proc file %s: %s", filename, str(err)) + if not data: + _ThrowError("Can't read any data from %s", filename) + return data @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: - try: - cls._ZeroSuperblock(i.dev_path) - except EnvironmentError, err: - logger.Error("Can't zero superblock for %s: %s" % - (i.dev_path, str(err))) - 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]) - - if result.failed: - logger.Error("Can't create md: %s: %s" % (result.fail_reason, - result.output)) - 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) - - def Remove(self): - """Stub remove function for MD RAID 1 arrays. - - We don't remove the superblock right now. Mark a to do. - - """ - #TODO: maybe zero superblock on child devices? - return self.Shutdown() - - def Rename(self, new_id): - """Rename a device. - - This is not supported for md raid1 devices. - - """ - raise errors.ProgrammerError("Can't rename a md raid1 device") - - def AddChildren(self, devices): - """Add new member(s) to the md raid1. - - """ - if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError("Can't attach to device") - - args = ["mdadm", "-a", self.dev_path] - for dev in devices: - if dev.dev_path is None: - raise errors.BlockDeviceError("Child '%s' is not initialised" % dev) - dev.Open() - args.append(dev.dev_path) - result = utils.RunCmd(args) - if result.failed: - raise errors.BlockDeviceError("Failed to add new device to array: %s" % - result.output) - new_len = len(self._children) + len(devices) - 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.extend(devices) + def _MassageProcData(cls, data): + """Transform the output of _GetProdData into a nicer form. - def RemoveChildren(self, devices): - """Remove member(s) from 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") - new_len = len(self._children) - len(devices) - if new_len < 1: - raise errors.BlockDeviceError("Can't reduce to less than one child") - args = ["mdadm", "-f", self.dev_path] - orig_devs = [] - for dev in devices: - args.append(dev) - for c in self._children: - if c.dev_path == dev: - orig_devs.append(c) - break + 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: - raise errors.BlockDeviceError("Can't find device '%s' for removal" % - dev) - result = utils.RunCmd(args) - if result.failed: - raise errors.BlockDeviceError("Failed to mark device(s) as failed: %s" % - result.output) - - # it seems here we need a short delay for MD to update its - # superblocks - time.sleep(0.5) - args[1] = "-r" - result = utils.RunCmd(args) - if result.failed: - raise errors.BlockDeviceError("Failed to remove device(s) 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) - for dev in orig_devs: - self._children.remove(dev) - - def GetStatus(self): - """Return the status of the device. - - """ - self.Attach() - if self.minor is None: - retval = self.STATUS_UNKNOWN - else: - retval = self.STATUS_ONLINE - return retval + 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 - def _SetFromMinor(self, minor): - """Set our parameters based on the given minor. + @classmethod + def _GetVersion(cls): + """Return the DRBD version. - This sets our minor variable and our dev_path. + This will return a dict with keys: + - k_major + - k_minor + - k_point + - api + - proto + - proto2 (only on drbd > 8.2.X) """ - self.minor = minor - self.dev_path = "/dev/md%d" % minor + proc_data = cls._GetProcData() + 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) - 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: %s" % - (result.fail_reason, result.output)) - self.minor = None - else: - self.minor = free_minor - return not result.failed + 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] - def Shutdown(self): - """Tear down the MD array. + return retval - This does a 'mdadm --stop' so after this command, the array is no - longer available. + @staticmethod + def _DevPath(minor): + """Return the path to a drbd device for a given minor. """ - 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]) - 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 + return "/dev/drbd%d" % minor - def SetSyncSpeed(self, kbytes): - """Set the maximum sync speed for the MD array. + @classmethod + def GetUsedDevs(cls): + """Compute the list of used DRBD devices. """ - 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") - 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, is_degraded, ldisk) - - 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. - - The ldisk parameter is always true for MD devices. - - """ - 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, False - 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, False - 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, False - - 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 - - def Close(self): - """Notifies that the device will no longer be used for I/O. - - This is a no-op for the MDRaid1 device type, but see comment for - `Open()`. - - """ - return True - - -class BaseDRBD(BlockDev): - """Base DRBD class. - - This class contains a few bits of common functionality between the - 0.7 and 8.x versions of DRBD. - - """ - _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)" - r" \(api:(\d+)/proto:(\d+)\)") - _DRBD_MAJOR = 147 - _ST_UNCONFIGURED = "Unconfigured" - _ST_WFCONNECTION = "WFConnection" - _ST_CONNECTED = "Connected" - - @staticmethod - def _GetProcData(): - """Return data from /proc/drbd. - - """ - stat = open("/proc/drbd", "r") - try: - data = stat.read().splitlines() - finally: - stat.close() - if not data: - raise errors.BlockDeviceError("Can't read any data from /proc/drbd") - return data - - @staticmethod - def _MassageProcData(data): - """Transform the output of _GetProdData into a nicer form. - - Returns: - a dictionary of minor: joined lines from /proc/drbd for that minor - - """ - lmatch = re.compile("^ *([0-9]+):.*$") - results = {} - old_minor = old_line = None - for line in data: - lresult = lmatch.match(line) - if lresult is not None: - if old_minor is not None: - results[old_minor] = old_line - old_minor = int(lresult.group(1)) - old_line = line - else: - if old_minor is not None: - old_line += " " + line.strip() - # add last line - if old_minor is not None: - results[old_minor] = old_line - return results - - @classmethod - def _GetVersion(cls): - """Return the DRBD version. - - This will return a list [k_major, k_minor, k_point, api, proto]. - - """ - proc_data = cls._GetProcData() - first_line = proc_data[0].strip() - version = cls._VERSION_RE.match(first_line) - if not version: - raise errors.BlockDeviceError("Can't parse DRBD version from '%s'" % - first_line) - return [int(val) for val in version.groups()] - - @staticmethod - def _DevPath(minor): - """Return the path to a drbd device for a given minor. - - """ - return "/dev/drbd%d" % minor - - @classmethod - def _GetUsedDevs(cls): - """Compute the list of used DRBD devices. - - """ - data = cls._GetProcData() + data = cls._GetProcData() used_devs = {} - valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$") for line in data: - match = valid_line.match(line) + match = cls._VALID_LINE_RE.match(line) if not match: continue minor = int(match.group(1)) @@ -1057,543 +927,57 @@ class BaseDRBD(BlockDev): continue used_devs[minor] = state, line - return used_devs - - def _SetFromMinor(self, minor): - """Set our parameters based on the given minor. - - This sets our minor variable and our dev_path. - - """ - if minor is None: - self.minor = self.dev_path = None - else: - self.minor = minor - self.dev_path = self._DevPath(minor) - - @staticmethod - def _CheckMetaSize(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 - - def Rename(self, new_id): - """Rename a device. - - This is not supported for drbd devices. - - """ - raise errors.ProgrammerError("Can't rename a drbd device") - - -class DRBDev(BaseDRBD): - """DRBD 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. - - """ - def __init__(self, unique_id, children): - super(DRBDev, self).__init__(unique_id, children) - self.major = self._DRBD_MAJOR - [kmaj, kmin, kfix, api, proto] = self._GetVersion() - if kmaj != 0 and kmin != 7: - raise errors.BlockDeviceError("Mismatch in DRBD kernel version and" - " requested ganeti usage: kernel is" - " %s.%s, ganeti wants 0.7" % (kmaj, kmin)) - - if len(children) != 2: - raise ValueError("Invalid configuration data %s" % str(children)) - if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 4: - raise ValueError("Invalid configuration data %s" % str(unique_id)) - self._lhost, self._lport, self._rhost, self._rport = unique_id - self.Attach() - - @classmethod - def _FindUnusedMinor(cls): - """Find an unused DRBD device. - - """ - data = cls._GetProcData() - - valid_line = re.compile("^ *([0-9]+): cs:Unconfigured$") - for line in data: - match = valid_line.match(line) - if match: - return int(match.group(1)) - logger.Error("Error: no free drbd minors!") - raise errors.BlockDeviceError("Can't find a free DRBD minor") - - @classmethod - def _GetDevInfo(cls, minor): - """Get details about a given DRBD minor. - - This return, if available, the local backing device in (major, - minor) formant and the local and remote (ip, port) information. - - """ - data = {} - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) - if result.failed: - logger.Error("Can't display the drbd config: %s" % result.fail_reason) - return data - out = result.stdout - if out == "Not configured\n": - return data - for line in out.splitlines(): - if "local_dev" not in data: - match = re.match("^Lower device: ([0-9]+):([0-9]+) .*$", 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 - - def _MatchesLocal(self, info): - """Test if our local config matches with an existing device. - - The parameter should be as returned from `_GetDevInfo()`. This - method tests if our local backing device is the same as the one in - the info parameter, in effect testing if we look like the given - device. - - """ - if not ("local_dev" in info and "meta_dev" in info and - "meta_index" in info): - return False - - backend = self._children[0] - if backend is not None: - retval = (info["local_dev"] == (backend.major, backend.minor)) - else: - retval = (info["local_dev"] == (0, 0)) - meta = self._children[1] - if meta is not None: - retval = retval and (info["meta_dev"] == (meta.major, meta.minor)) - retval = retval and (info["meta_index"] == 0) - else: - retval = retval and (info["meta_dev"] == "internal" and - info["meta_index"] == -1) - return retval - - def _MatchesNet(self, info): - """Test if our network config matches with an existing device. - - The parameter should be as returned from `_GetDevInfo()`. This - method tests if our network configuration is the same as the one - in the info parameter, in effect testing if we look like the given - device. - - """ - if (((self._lhost is None and not ("local_addr" in info)) and - (self._rhost is None and not ("remote_addr" in info)))): - return True - - if self._lhost is None: - return False - - if not ("local_addr" in info and - "remote_addr" in info): - return False - - retval = (info["local_addr"] == (self._lhost, self._lport)) - retval = (retval and - info["remote_addr"] == (self._rhost, self._rport)) - return retval - - @classmethod - def _AssembleLocal(cls, minor, backend, meta): - """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._CheckMetaSize(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 - - @classmethod - def _ShutdownLocal(cls, minor): - """Detach from the local device. - - I/Os will continue to be served from the remote device. If we - don't have a remote device, this operation will fail. - - """ - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) - if result.failed: - logger.Error("Can't detach local device: %s" % result.output) - return not result.failed - - @staticmethod - def _ShutdownAll(minor): - """Deactivate the device. - - This will, of course, fail if the device is in use. - - """ - 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 - - @classmethod - def _AssembleNet(cls, minor, net_info, protocol): - """Configure the network part of the device. - - 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. - - """ - lhost, lport, rhost, rport = net_info - result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "net", - "%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), - protocol]) - if result.failed: - logger.Error("Can't setup network for dbrd device: %s" % - result.fail_reason) - return False - - timeout = time.time() + 10 - ok = False - while time.time() < timeout: - info = cls._GetDevInfo(minor) - if not "local_addr" in info or not "remote_addr" in info: - time.sleep(1) - continue - 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 - - @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: - logger.Error("Can't shutdown network: %s" % result.output) - return not result.failed - - def Assemble(self): - """Assemble the drbd. - - 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 - - """ - self.Attach() - if self.minor is not None: - logger.Info("Already assembled") - return True - - result = super(DRBDev, self).Assemble() - if not result: - return result - - minor = self._FindUnusedMinor() - 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 - - def Shutdown(self): - """Shutdown the DRBD device. - - """ - 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 - - 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. - - """ - 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 - - self._SetFromMinor(minor) - return minor is not None - - 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 - force flag should be only given after creation, when it actually - has to be given. - - """ - if self.minor is None and not self.Attach(): - logger.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") - result = utils.RunCmd(cmd) - if result.failed: - logger.Error("Can't make drbd device primary: %s" % result.output) - return False - return True - - def Close(self): - """Make the local state secondary. - - This will, of course, fail if the device is in use. - - """ - if self.minor is None and not self.Attach(): - logger.Info("Instance not attached to a device") - raise errors.BlockDeviceError("Can't find device") - 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") - - def SetSyncSpeed(self, kbytes): - """Set the speed of the DRBD syncer. - - """ - 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 - - def GetSyncStatus(self): - """Returns the sync status of the device. - - Returns: - (sync_percent, estimated_time, is_degraded, ldisk) - - 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. - - The ldisk parameter will be returned as True, since the DRBD7 - devices have not been converted. - - """ - 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]+) .*$", line) - if match: - sync_percent = float(match.group(1)) - hours = int(match.group(2)) - minutes = int(match.group(3)) - seconds = int(match.group(4)) - est_time = hours * 3600 + minutes * 60 + seconds - else: - sync_percent = None - est_time = None - match = re.match("^ *[0-9]+: cs:([^ ]+).*$", 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, False - - def GetStatus(self): - """Compute the status of the DRBD device - - Note that DRBD devices don't have the STATUS_EXISTING state. + return used_devs - """ - if self.minor is None and not self.Attach(): - return self.STATUS_UNKNOWN + def _SetFromMinor(self, minor): + """Set our parameters based on the given minor. - data = self._GetProcData() - match = re.compile("^ *%d: cs:[^ ]+ st:(Primary|Secondary)/.*$" % - self.minor) - for line in data: - mresult = match.match(line) - if mresult: - break - else: - logger.Error("Can't find myself!") - return self.STATUS_UNKNOWN + This sets our minor variable and our dev_path. - state = mresult.group(2) - if state == "Primary": - result = self.STATUS_ONLINE + """ + if minor is None: + self.minor = self.dev_path = None + self.attached = False else: - result = self.STATUS_STANDBY - - return result + self.minor = minor + self.dev_path = self._DevPath(minor) + self.attached = True @staticmethod - def _ZeroDevice(device): - """Zero a device. + def _CheckMetaSize(meta_device): + """Check if the given meta device looks like a valid one. - This writes until we get ENOSPC. + This currently only check the size, which must be around + 128MiB. """ - f = open(device, "w") - buf = "\0" * 1048576 + result = utils.RunCmd(["blockdev", "--getsize", meta_device]) + if result.failed: + _ThrowError("Failed to get device size: %s - %s", + result.fail_reason, result.output) try: - while True: - f.write(buf) - except IOError, err: - if err.errno != errno.ENOSPC: - raise - - @classmethod - def Create(cls, unique_id, children, size): - """Create a new DRBD device. - - Since DRBD devices are not created per se, just assembled, this - function just zeroes the meta device. + sectors = int(result.stdout) + except (TypeError, ValueError): + _ThrowError("Invalid output from blockdev: '%s'", result.stdout) + bytes = sectors * 512 + if bytes < 128 * 1024 * 1024: # less than 128MiB + _ThrowError("Meta device too small (%.2fMib)", (bytes / 1024 / 1024)) + # the maximum *valid* size of the meta device when living on top + # of LVM is hard to compute: it depends on the number of stripes + # and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB + # (normal size), but an eight-stripe 128MB PE will result in a 1GB + # size meta device; as such, we restrict it to 1GB (a little bit + # too generous, but making assumptions about PE size is hard) + if bytes > 1024 * 1024 * 1024: + _ThrowError("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024)) - """ - if len(children) != 2: - raise errors.ProgrammerError("Invalid setup for the drbd device") - meta = children[1] - meta.Assemble() - if not meta.Attach(): - raise errors.BlockDeviceError("Can't attach to meta device") - if not cls._CheckMetaSize(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device") - 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) + def Rename(self, new_id): + """Rename a device. - def Remove(self): - """Stub remove for DRBD devices. + This is not supported for drbd devices. """ - return self.Shutdown() + raise errors.ProgrammerError("Can't rename a drbd device") class DRBD8(BaseDRBD): @@ -1612,22 +996,35 @@ class DRBD8(BaseDRBD): _MAX_MINORS = 255 _PARSE_SHOW = None - def __init__(self, unique_id, children): + # timeout constants + _NET_RECONFIG_TIMEOUT = 60 + + def __init__(self, unique_id, children, size): if children and children.count(None) > 0: children = [] - super(DRBD8, self).__init__(unique_id, children) - self.major = self._DRBD_MAJOR - [kmaj, kmin, kfix, api, proto] = self._GetVersion() - if kmaj != 8: - raise errors.BlockDeviceError("Mismatch in DRBD kernel version and" - " requested ganeti usage: kernel is" - " %s.%s, ganeti wants 8.x" % (kmaj, kmin)) - 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) + self.major = self._DRBD_MAJOR + version = self._GetVersion() + if version['k_major'] != 8 : + _ThrowError("Mismatch in DRBD kernel version and requested ganeti" + " usage: kernel is %s.%s, ganeti wants 8.x", + version['k_major'], version['k_minor']) + + if (self._lhost is not None and self._lhost == self._rhost and + self._lport == self._rport): + raise ValueError("Invalid configuration data, same local/remote %s" % + (unique_id,)) self.Attach() @classmethod @@ -1640,8 +1037,7 @@ class DRBD8(BaseDRBD): result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), "v08", dev_path, "0", "create-md"]) if result.failed: - raise errors.BlockDeviceError("Can't initialize meta device: %s" % - result.output) + _ThrowError("Can't initialize meta device: %s", result.output) @classmethod def _FindUnusedMinor(cls): @@ -1653,40 +1049,23 @@ class DRBD8(BaseDRBD): """ data = cls._GetProcData() - unused_line = re.compile("^ *([0-9]+): cs:Unconfigured$") - used_line = re.compile("^ *([0-9]+): cs:") highest = None for line in data: - match = unused_line.match(line) + match = cls._UNUSED_LINE_RE.match(line) if match: return int(match.group(1)) - match = used_line.match(line) + 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: - logger.Error("Error: no free drbd minors!") + logging.error("Error: no free drbd minors!") raise errors.BlockDeviceError("Can't find a free DRBD minor") return highest + 1 @classmethod - def _IsValidMeta(cls, meta_device): - """Check if the given meta device looks like a valid one. - - """ - minor = cls._FindUnusedMinor() - minor_path = cls._DevPath(minor) - result = utils.RunCmd(["drbdmeta", minor_path, - "v08", meta_device, "0", - "dstate"]) - if result.failed: - logger.Error("Invalid meta device %s: %s" % (meta_device, result.output)) - return False - return True - - @classmethod def _GetShowParser(cls): """Return a parser for `drbd show` output. @@ -1713,15 +1092,20 @@ class DRBD8(BaseDRBD): # value types value = pyp.Word(pyp.alphanums + '_-/.:') quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote - addr_port = (pyp.Word(pyp.nums + '.') + pyp.Literal(':').suppress() + - number) + addr_type = (pyp.Optional(pyp.Literal("ipv4")).suppress() + + pyp.Optional(pyp.Literal("ipv6")).suppress()) + addr_port = (addr_type + pyp.Word(pyp.nums + '.') + + pyp.Literal(':').suppress() + number) # meta device, extended syntax meta_value = ((value ^ quoted) + pyp.Literal('[').suppress() + number + pyp.Word(']').suppress()) + # device name, extended syntax + device_value = pyp.Literal("minor").suppress() + number # a statement stmt = (~rbrace + keyword + ~lbrace + - (addr_port ^ value ^ quoted ^ meta_value) + + pyp.Optional(addr_port ^ value ^ quoted ^ meta_value ^ + device_value) + pyp.Optional(defa) + semi + pyp.Optional(pyp.restOfLine).suppress()) @@ -1743,7 +1127,8 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) if result.failed: - logger.Error("Can't display the drbd config: %s" % result.fail_reason) + logging.error("Can't display the drbd config: %s - %s", + result.fail_reason, result.output) return None return result.stdout @@ -1767,8 +1152,7 @@ class DRBD8(BaseDRBD): try: results = bnf.parseString(out) except pyp.ParseException, err: - raise errors.BlockDeviceError("Can't parse drbdsetup show output: %s" % - str(err)) + _ThrowError("Can't parse drbdsetup show output: %s", str(err)) # and massage the results into our desired format for section in results: @@ -1843,21 +1227,37 @@ class DRBD8(BaseDRBD): return retval @classmethod - def _AssembleLocal(cls, minor, backend, meta): + def _AssembleLocal(cls, 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", - "--create-device"]) + args = ["drbdsetup", cls._DevPath(minor), "disk", + backend, meta, "0", + "-e", "detach", + "--create-device"] + if size: + args.extend(["-d", "%sm" % size]) + if not constants.DRBD_BARRIERS: # disable barriers, if configured so + version = cls._GetVersion() + # various DRBD versions support different disk barrier options; + # what we aim here is to revert back to the 'drain' method of + # disk flushes and to disable metadata barriers, in effect going + # back to pre-8.0.7 behaviour + vmaj = version['k_major'] + vmin = version['k_minor'] + vrel = version['k_point'] + assert vmaj == 8 + if vmin == 0: # 8.0.x + if vrel >= 12: + args.extend(['-i', '-m']) + elif vmin == 2: # 8.2.x + if vrel >= 7: + args.extend(['-i', '-m']) + elif vmaj >= 3: # 8.3.x or newer + args.extend(['-i', '-a', 'm']) + result = utils.RunCmd(args) if result.failed: - logger.Error("Can't attach local disk: %s" % result.output) - return not result.failed + _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output) @classmethod def _AssembleNet(cls, minor, net_info, protocol, @@ -1869,12 +1269,22 @@ class DRBD8(BaseDRBD): if None in net_info: # we don't want network connection and actually want to make # sure its shutdown - return cls._ShutdownNet(minor) + cls._ShutdownNet(minor) + return + + # Workaround for a race condition. When DRBD is doing its dance to + # establish a connection with its peer, it also sends the + # 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. + cls._SetMinorSyncSpeed(minor, constants.SYNC_SPEED) args = ["drbdsetup", cls._DevPath(minor), "net", "%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), protocol, "-A", "discard-zero-changes", "-B", "consensus", + "--create-device", ] if dual_pri: args.append("-m") @@ -1882,52 +1292,44 @@ class DRBD8(BaseDRBD): args.extend(["-a", hmac, "-x", secret]) 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: + def _CheckNetworkConfig(): info = cls._GetDevInfo(cls._GetShowData(minor)) if not "local_addr" in info or not "remote_addr" in info: - time.sleep(1) - continue + raise utils.RetryAgain() + if (info["local_addr"] != (lhost, lport) or info["remote_addr"] != (rhost, rport)): - 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) def AddChildren(self, devices): """Add a disk to the DRBD device. """ if self.minor is None: - raise errors.BlockDeviceError("Can't attach to dbrd8 during AddChildren") + _ThrowError("drbd%d: can't attach to dbrd8 during AddChildren", + self._aminor) if len(devices) != 2: - raise errors.BlockDeviceError("Need two devices for AddChildren") + _ThrowError("drbd%d: need two devices for AddChildren", self.minor) info = self._GetDevInfo(self._GetShowData(self.minor)) if "local_dev" in info: - raise errors.BlockDeviceError("DRBD8 already attached to a local disk") + _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: - raise errors.BlockDeviceError("Children not ready during AddChildren") + _ThrowError("drbd%d: children not ready during AddChildren", self.minor) backend.Open() meta.Open() - if not self._CheckMetaSize(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device size") + self._CheckMetaSize(meta.dev_path) self._InitMeta(self._FindUnusedMinor(), meta.dev_path) - if not self._IsValidMeta(meta.dev_path): - raise errors.BlockDeviceError("Cannot initalize meta device") - if not self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path): - raise errors.BlockDeviceError("Can't attach to local storage") + self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size) self._children = devices def RemoveChildren(self, devices): @@ -1935,138 +1337,133 @@ class DRBD8(BaseDRBD): """ if self.minor is None: - raise errors.BlockDeviceError("Can't attach to drbd8 during" - " RemoveChildren") + _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: - raise errors.BlockDeviceError("We don't have two children: %s" % - self._children) + _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 :) - logger.Error("Requested detach while detached") + logging.warning("drbd%d: requested detach while detached", self.minor) return if len(devices) != 2: - raise errors.BlockDeviceError("We need two children in RemoveChildren") + _ThrowError("drbd%d: we need two children in RemoveChildren", self.minor) for child, dev in zip(self._children, devices): if dev != child.dev_path: - raise errors.BlockDeviceError("Mismatch in local storage" - " (%s != %s) in RemoveChildren" % - (dev, child.dev_path)) + _ThrowError("drbd%d: mismatch in local storage (%s != %s) in" + " RemoveChildren", self.minor, dev, child.dev_path) - if not self._ShutdownLocal(self.minor): - raise errors.BlockDeviceError("Can't detach from local storage") + self._ShutdownLocal(self.minor) self._children = [] + @classmethod + def _SetMinorSyncSpeed(cls, minor, kbytes): + """Set the speed of the DRBD syncer. + + This is the low-level implementation. + + @type minor: int + @param minor: the drbd minor whose settings we change + @type kbytes: int + @param kbytes: the speed in kbytes/second + @rtype: boolean + @return: the success of the operation + + """ + result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "syncer", + "-r", "%d" % kbytes, "--create-device"]) + if result.failed: + logging.error("Can't change syncer rate: %s - %s", + result.fail_reason, result.output) + return not result.failed + def SetSyncSpeed(self, kbytes): """Set the speed of the DRBD syncer. + @type kbytes: int + @param kbytes: the speed in kbytes/second + @rtype: boolean + @return: the success of the operation + """ - children_result = super(DRBD8, self).SetSyncSpeed(kbytes) if self.minor is None: - logger.Info("Instance not attached to a device") + logging.info("Not attached during SetSyncSpeed") 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 + children_result = super(DRBD8, self).SetSyncSpeed(kbytes) + return self._SetMinorSyncSpeed(self.minor, kbytes) and children_result + + def GetProcStatus(self): + """Return device data from /proc. + + """ + 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. - Returns: - (sync_percent, estimated_time, is_degraded) If sync_percent is None, it means all is ok - If estimated_time is None, it means we can't esimate + If estimated_time is None, it means we can't estimate the time needed, otherwise it's the time left in seconds. We set the is_degraded parameter to True on two conditions: network not connected or local disk missing. - We compute the ldisk parameter based on wheter we have a local + We compute the ldisk parameter based on whether we have a local disk or not. - """ - 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]+) .*$", line) - if match: - sync_percent = float(match.group(1)) - hours = int(match.group(2)) - minutes = int(match.group(3)) - seconds = int(match.group(4)) - est_time = hours * 3600 + minutes * 60 + seconds - else: - sync_percent = None - est_time = None - match = re.match("^ *\d+: cs:(\w+).*ds:(\w+)/(\w+).*$", line) - if not match: - raise errors.BlockDeviceError("Can't find my data in /proc (minor %d)" % - self.minor) - client_state = match.group(1) - local_disk_state = match.group(2) - ldisk = local_disk_state != "UpToDate" - is_degraded = client_state != "Connected" - return sync_percent, est_time, is_degraded or ldisk, ldisk - - def GetStatus(self): - """Compute the status of the DRBD device - - Note that DRBD devices don't have the STATUS_EXISTING state. + @rtype: objects.BlockDevStatus """ if self.minor is None and not self.Attach(): - return self.STATUS_UNKNOWN + _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) - data = self._GetProcData() - match = re.compile("^ *%d: cs:[^ ]+ st:(Primary|Secondary)/.*$" % - self.minor) - for line in data: - mresult = match.match(line) - if mresult: - break - else: - logger.Error("Can't find myself!") - return self.STATUS_UNKNOWN + stats = self.GetProcStatus() + is_degraded = not stats.is_connected or not stats.is_disk_uptodate - state = mresult.group(2) - if state == "Primary": - result = self.STATUS_ONLINE + if stats.is_disk_uptodate: + ldisk_status = constants.LDS_OKAY + elif stats.is_diskless: + ldisk_status = constants.LDS_FAULTY else: - result = self.STATUS_STANDBY + ldisk_status = constants.LDS_UNKNOWN - return result + 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("-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. @@ -2075,112 +1472,230 @@ class DRBD8(BaseDRBD): """ 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. + + 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. + + """ + if self.minor is None: + _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) + + class _DisconnectStatus: + def __init__(self, ever_disconnected): + self.ever_disconnected = ever_disconnected + + dstatus = _DisconnectStatus(_IgnoreError(self._ShutdownNet, self.minor)) + + def _WaitForDisconnect(): + if self.GetProcStatus().is_standalone: + return + + # retry the disconnect, it seems possible that due to a well-time + # disconnect on the peer, my disconnect command might be ignored and + # forgotten + dstatus.ever_disconnected = \ + _IgnoreError(self._ShutdownNet, self.minor) or dstatus.ever_disconnected + + raise utils.RetryAgain() + + # Keep start time + start_time = time.time() + + try: + # Start delay at 100 milliseconds and grow up to 2 seconds + utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0), + self._NET_RECONFIG_TIMEOUT) + except utils.RetryTimeout: + if dstatus.ever_disconnected: + msg = ("drbd%d: device did not react to the" + " 'disconnect' command in a timely manner") + else: + msg = "drbd%d: can't shutdown network, even after multiple retries" + + _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: + _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): - """Find a DRBD device which matches our config and attach to it. + """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: + minor = None + + self._SetFromMinor(minor) + return minor is not None + + def Assemble(self): + """Assemble the drbd. + + Method: + - if we have a configured device, we try to ensure that it matches + our config + - if not, we create it from zero + + """ + 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() + + def _SlowAssemble(self): + """Assembles the DRBD device from a (partially) configured device. 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. """ - for minor in self._GetUsedDevs(): + # TODO: Rewrite to not use a for loop just because there is 'break' + # pylint: disable-msg=W0631 + net_data = (self._lhost, self._lport, self._rhost, self._rport) + for minor in (self._aminor,): info = self._GetDevInfo(self._GetShowData(minor)) match_l = self._MatchesLocal(info) match_r = self._MatchesNet(info) + if match_l and match_r: + # everything matches 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: - if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): - break - # the weakest case: we find something that is only net attached - # even though we were passed some children at init time + # disk matches, but not attached to network, attach and recheck + self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) + if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): + break + else: + _ThrowError("drbd%d: network attach successful, but 'drbdsetup" + " show' disagrees", minor) + if match_r and "local_dev" not in info: - break - if match_l and not match_r and "local_addr" 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 - if not self._ShutdownNet(minor): - raise errors.BlockDeviceError("Device has correct local storage," - " wrong remote peer and is unable to" - " disconnect in order to attach to" - " the correct peer") + try: + self._ShutdownNet(minor) + except errors.BlockDeviceError, err: + _ThrowError("drbd%d: device has correct local storage, wrong" + " remote peer and is unable to disconnect in order" + " to attach to the correct peer: %s", minor, str(err)) # note: _AssembleNet also handles the case when we don't want # local storage (i.e. one or more of the _[lr](host|port) is # None) - if (self._AssembleNet(minor, (self._lhost, self._lport, - self._rhost, self._rport), "C") and - self._MatchesNet(self._GetDevInfo(self._GetShowData(minor)))): + self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL, + hmac=constants.DRBD_HMAC_ALG, secret=self._secret) + if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): break + else: + _ThrowError("drbd%d: network attach successful, but 'drbdsetup" + " show' disagrees", minor) else: minor = None self._SetFromMinor(minor) - return minor is not None + if minor is None: + _ThrowError("drbd%d: cannot activate, unknown or unhandled reason", + self._aminor) - def Assemble(self): - """Assemble the drbd. + def _FastAssemble(self): + """Assemble the drbd device from zero. - 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 + This is run when in Assemble we detect our minor is unused. """ - self.Attach() - if self.minor is not None: - logger.Info("Already assembled") - return True - - result = super(DRBD8, self).Assemble() - if not result: - return result - - minor = self._FindUnusedMinor() - need_localdev_teardown = False + minor = self._aminor if self._children and self._children[0] and self._children[1]: - result = self._AssembleLocal(minor, self._children[0].dev_path, - self._children[1].dev_path) - if not result: - return False - need_localdev_teardown = True + 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: - 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._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) - return True @classmethod def _ShutdownLocal(cls, minor): @@ -2192,8 +1707,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) if result.failed: - logger.Error("Can't detach local device: %s" % result.output) - return not result.failed + _ThrowError("drbd%d: can't detach local disk: %s", minor, result.output) @classmethod def _ShutdownNet(cls, minor): @@ -2204,8 +1718,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) if result.failed: - logger.Error("Can't shutdown network: %s" % result.output) - return not result.failed + _ThrowError("drbd%d: can't shutdown network: %s", minor, result.output) @classmethod def _ShutdownAll(cls, minor): @@ -2216,27 +1729,26 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) if result.failed: - logger.Error("Can't shutdown drbd device: %s" % result.output) - return not 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(): - logger.Info("DRBD device not attached to a device during Shutdown") - return True - if not self._ShutdownAll(self.minor): - return False + logging.info("drbd%d: not attached during Shutdown()", self._aminor) + return + minor = self.minor self.minor = None self.dev_path = None - return True + self._ShutdownAll(minor) def Remove(self): """Stub remove for DRBD devices. """ - return self.Shutdown() + self.Shutdown() @classmethod def Create(cls, unique_id, children, size): @@ -2248,27 +1760,194 @@ class DRBD8(BaseDRBD): """ if len(children) != 2: raise errors.ProgrammerError("Invalid setup for the drbd device") + # 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._CheckMetaSize(meta.dev_path): - raise errors.BlockDeviceError("Invalid meta device size") - cls._InitMeta(cls._FindUnusedMinor(), meta.dev_path) - if not cls._IsValidMeta(meta.dev_path): - raise errors.BlockDeviceError("Cannot initalize meta device") - 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) + + def Grow(self, amount): + """Resize the DRBD device and its backing storage. + + """ + 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) + 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): + """Initalizes a file device backend. + + """ + if children: + raise errors.BlockDeviceError("Invalid setup for file device") + super(FileStorage, self).__init__(unique_id, children, size) + if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: + raise ValueError("Invalid configuration data %s" % str(unique_id)) + self.driver = unique_id[0] + self.dev_path = unique_id[1] + self.Attach() + + def Assemble(self): + """Assemble the device. + + Checks whether the file device exists, raises BlockDeviceError otherwise. + + """ + if not os.path.exists(self.dev_path): + _ThrowError("File device '%s' does not exist" % self.dev_path) + + def Shutdown(self): + """Shutdown the device. + + This is a no-op for the file type, as we don't deactivate + the file on shutdown. + + """ + pass + + def Open(self, force=False): + """Make the device ready for I/O. + + This is a no-op for the file type. + + """ + 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. + + """ + pass + + 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): + """Grow the file + + @param amount: the amount (in mebibytes) to grow with + + """ + # 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" + 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): + """Create a new file. + + @param size: the size of file in MiB + + @rtype: L{bdev.FileStorage} + @return: an instance of FileStorage + + """ + 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] + 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) DEV_MAP = { constants.LD_LV: LogicalVolume, - constants.LD_MD_R1: MDRaid1, - constants.LD_DRBD7: DRBDev, constants.LD_DRBD8: DRBD8, } +if constants.ENABLE_FILE_STORAGE: + DEV_MAP[constants.LD_FILE] = FileStorage -def FindDevice(dev_type, unique_id, children): + +def FindDevice(dev_type, unique_id, children, size): """Search for an existing, assembled device. This will succeed only if the device exists and is assembled, but it @@ -2277,28 +1956,23 @@ def FindDevice(dev_type, unique_id, children): """ 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 = DEV_MAP[dev_type](unique_id, children, size) + if not device.attached: return None - return device + return device -def AttachOrAssemble(dev_type, unique_id, children): +def Assemble(dev_type, unique_id, children, size): """Try to attach or assemble an existing device. - This will attach to an existing assembled device or will assemble - the device, as needed, to bring it fully up. + This will attach to assemble the device, as needed, to bring it + fully up. It must be safe to run on already-assembled devices. """ 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)) + device = DEV_MAP[dev_type](unique_id, children, size) + device.Assemble() return device