X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/ffa1c0dc146c9929b7baf7f2d3b18f9e0144fcbf..71910715d49828acf8cc89d59ed2303c01b44b30:/lib/bdev.py diff --git a/lib/bdev.py b/lib/bdev.py index c3f4a7c..7904eea 100644 --- a/lib/bdev.py +++ b/lib/bdev.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2006, 2007 Google Inc. +# Copyright (C) 2006, 2007, 2010 Google Inc. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by @@ -31,6 +31,59 @@ import logging from ganeti import utils from ganeti import errors from ganeti import constants +from ganeti import objects +from ganeti import compat +from ganeti import netutils + + +# 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): @@ -76,47 +129,28 @@ class BlockDev(object): after assembly we'll have our correct major/minor. """ - def __init__(self, unique_id, children): + def __init__(self, unique_id, children, size): self._children = children self.dev_path = None self.unique_id = unique_id self.major = None self.minor = None self.attached = False + self.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 - - try: - child.Open() - except errors.BlockDeviceError: - for child in self._children: - child.Shutdown() - raise - - 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. @@ -148,7 +182,7 @@ class BlockDev(object): """Remove this device. This makes sense only for some of the device types: LV and file - storeage. Also note that if the device can't attach, the removal + storage. Also note that if the device can't attach, the removal can't be completed. """ @@ -196,15 +230,26 @@ class BlockDev(object): result = result and child.SetSyncSpeed(speed) return result + def PauseResumeSync(self, pause): + """Pause/Resume the sync of the mirror. + + In case this is not a mirroring device, this is no-op. + + @param pause: Wheater to pause or resume + + """ + result = True + if self._children: + for child in self._children: + result = result and child.PauseResumeSync(pause) + return result + def GetSyncStatus(self): """Returns the sync status of the device. If this device is a mirroring device, this function returns the status of the mirror. - 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 @@ -218,9 +263,16 @@ 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. @@ -229,22 +281,44 @@ 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): @@ -259,14 +333,28 @@ class BlockDev(object): def Grow(self, amount): """Grow the block device. - Arguments: - amount: the amount (in mebibytes) to grow with - - Returns: None + @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>" % (self.__class__, self.unique_id, self._children, @@ -277,19 +365,25 @@ 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 = None + self.major = self.minor = self.pe_size = self.stripe_count = None self.Attach() @classmethod @@ -298,76 +392,164 @@ 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(":" in v for v in pvlist): + _ThrowError("Some of your PVs have the invalid character ':' in their" + " name, this is not supported - please filter them out" + " in lvm.conf using either 'filter' or 'preferred_names'") 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("%s - %s" % (result.fail_reason, - result.output)) - 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 _GetVolumeInfo(lvm_cmd, fields): + """Returns LVM Volumen infos using lvm_cmd + + @param lvm_cmd: Should be one of "pvs", "vgs" or "lvs" + @param fields: Fields to return + @return: A list of dicts each with the parsed fields + + """ + if not fields: + raise errors.ProgrammerError("No fields specified") + + sep = "|" + cmd = [lvm_cmd, "--noheadings", "--nosuffix", "--units=m", "--unbuffered", + "--separator=%s" % sep, "-o%s" % ",".join(fields)] + + result = utils.RunCmd(cmd) + if result.failed: + raise errors.CommandError("Can't get the volume information: %s - %s" % + (result.fail_reason, result.output)) + + data = [] + for line in result.stdout.splitlines(): + splitted_fields = line.strip().split(sep) + + if len(fields) != len(splitted_fields): + raise errors.CommandError("Can't parse %s output: line '%s'" % + (lvm_cmd, line)) + + data.append(splitted_fields) + + return data + + @classmethod + def GetPVInfo(cls, vg_names, filter_allocatable=True): """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 """ - command = ["pvs", "--noheadings", "--nosuffix", "--units=m", - "-opv_name,vg_name,pv_free,pv_attr", "--unbuffered", - "--separator=:"] - result = utils.RunCmd(command) - if result.failed: - logging.error("Can't get the PV information: %s - %s", - result.fail_reason, result.output) + try: + info = cls._GetVolumeInfo("pvs", ["pv_name", "vg_name", "pv_free", + "pv_attr"]) + except errors.GenericError, err: + logging.error("Can't get PV information: %s", err) return None + data = [] - for line in result.stdout.splitlines(): - fields = line.strip().split(':') - if len(fields) != 4: - logging.error("Can't parse pvs output: line '%s'", line) - return None - # skip over pvs from another vg or ones which are not allocatable - if fields[1] != vg_name or fields[3][0] != 'a': + for pv_name, vg_name, pv_free, pv_attr in info: + # (possibly) skip over pvs which are not allocatable + if filter_allocatable and pv_attr[0] != "a": + continue + # (possibly) skip over pvs which are not in the right volume group(s) + if vg_names and vg_name not in vg_names: + continue + data.append((float(pv_free), pv_name, vg_name)) + + return data + + @classmethod + def GetVGInfo(cls, vg_names, filter_readonly=True): + """Get the free space info for specific VGs. + + @param vg_names: list of volume group names, if empty all will be returned + @param filter_readonly: whether to skip over readonly VGs + + @rtype: list + @return: list of tuples (free_space, total_size, name) with free_space in + MiB + + """ + try: + info = cls._GetVolumeInfo("vgs", ["vg_name", "vg_free", "vg_attr", + "vg_size"]) + except errors.GenericError, err: + logging.error("Can't get VG information: %s", err) + return None + + data = [] + for vg_name, vg_free, vg_attr, vg_size in info: + # (possibly) skip over vgs which are not writable + if filter_readonly and vg_attr[0] == "r": continue - data.append((float(fields[2]), fields[0])) + # (possibly) skip over vgs which are not in the right volume group(s) + if vg_names and vg_name not in vg_names: + continue + data.append((float(vg_free), float(vg_size), vg_name)) return data + @classmethod + def _ValidateName(cls, name): + """Validates that a given name is valid as VG or LV name. + + The list of valid characters and restricted names is taken out of + the lvm(8) manpage, with the simplification that we enforce both + VG and LV restrictions on the names. + + """ + if (not cls._VALID_NAME_RE.match(name) or + name in cls._INVALID_NAMES or + compat.any(substring in name for substring in cls._INVALID_SUBSTRINGS)): + _ThrowError("Invalid LVM name '%s'", name) + def Remove(self): """Remove this logical volume. """ if not self.minor and not self.Attach(): # the LV does not exist - return True + return result = utils.RunCmd(["lvremove", "-f", "%s/%s" % (self._vg_name, self._lv_name)]) if result.failed: - logging.error("Can't lvremove: %s - %s", - result.fail_reason, result.output) - - return not result.failed + _ThrowError("Can't lvremove: %s - %s", result.fail_reason, result.output) def Rename(self, new_id): """Rename this logical volume. @@ -382,10 +564,9 @@ 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. @@ -397,19 +578,30 @@ class LogicalVolume(BlockDev): """ self.attached = False result = utils.RunCmd(["lvs", "--noheadings", "--separator=,", - "-olv_attr,lv_kernel_major,lv_kernel_minor", - self.dev_path]) + "--units=m", "--nosuffix", + "-olv_attr,lv_kernel_major,lv_kernel_minor," + "vg_extent_size,stripes", self.dev_path]) if result.failed: logging.error("Can't find LV %s: %s, %s", self.dev_path, result.fail_reason, result.output) return False - out = result.stdout.strip().rstrip(',') + # the output can (and will) have multiple lines for multi-segment + # LVs, as the 'stripes' parameter is a segment one, so we take + # only the last entry, which is the one we're interested in; note + # that with LVM2 anyway the 'stripes' value must be constant + # across segments, so this is a no-op actually + out = result.stdout.splitlines() + if not out: # totally empty result? splitlines() returns at least + # one line for any non-empty string + logging.error("Can't parse LVS output, no lines? Got '%s'", str(out)) + return False + out = out[-1].strip().rstrip(',') out = out.split(",") - if len(out) != 3: - logging.error("Can't parse LVS output, len(%s) != 3", str(out)) + if len(out) != 5: + logging.error("Can't parse LVS output, len(%s) != 5", str(out)) return False - status, major, minor = out[:3] + status, major, minor, pe_size, stripes = out if len(status) != 6: logging.error("lvs lv_attr is not 6 characters (%s)", status) return False @@ -417,11 +609,25 @@ class LogicalVolume(BlockDev): try: major = int(major) minor = int(minor) - except ValueError, err: + except (TypeError, ValueError), err: logging.error("lvs major/minor cannot be parsed: %s", str(err)) + try: + pe_size = int(float(pe_size)) + except (TypeError, ValueError), err: + logging.error("Can't parse vg extent size: %s", err) + return False + + try: + stripes = int(stripes) + except (TypeError, ValueError), err: + logging.error("Can't parse the number of stripes: %s", err) + return False + self.major = major self.minor = minor + self.pe_size = pe_size + self.stripe_count = stripes self._degraded = status[0] == 'v' # virtual volume, i.e. doesn't backing # storage self.attached = True @@ -430,16 +636,14 @@ class LogicalVolume(BlockDev): def Assemble(self): """Assemble the device. - We alway run `lvchange -ay` on the LV to ensure it's active before + We always run `lvchange -ay` on the LV to ensure it's active before use, as there were cases when xenvg was not active after boot (also possibly after disk issues). """ result = utils.RunCmd(["lvchange", "-ay", self.dev_path]) if result.failed: - logging.error("Can't activate lv %s: %s", self.dev_path, result.output) - return False - return self.Attach() + _ThrowError("Can't activate lv %s: %s", self.dev_path, result.output) def Shutdown(self): """Shutdown the device. @@ -448,7 +652,7 @@ class LogicalVolume(BlockDev): volumes on shutdown. """ - return True + pass def GetSyncStatus(self): """Returns the sync status of the device. @@ -456,9 +660,6 @@ 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 @@ -472,8 +673,21 @@ class LogicalVolume(BlockDev): The status was already read in Attach, so we just return it. + @rtype: objects.BlockDevStatus + """ - return None, None, self._degraded, self._degraded + if self._degraded: + ldisk_status = constants.LDS_FAULTY + else: + ldisk_status = constants.LDS_OKAY + + return objects.BlockDevStatus(dev_path=self.dev_path, + major=self.major, + minor=self.minor, + sync_percent=None, + estimated_time=None, + is_degraded=self._degraded, + ldisk_status=ldisk_status) def Open(self, force=False): """Make the device ready for I/O. @@ -494,32 +708,30 @@ class LogicalVolume(BlockDev): def Snapshot(self, size): """Create a snapshot copy of an lvm block device. + @returns: tuple (vg, lv) + """ snap_name = self._lv_name + ".snap" # remove existing snapshot if found - snap = LogicalVolume((self._vg_name, snap_name), None) - snap.Remove() + snap = LogicalVolume((self._vg_name, snap_name), None, size) + _IgnoreError(snap.Remove) - pvs_info = self.GetPVInfo(self._vg_name) - if not pvs_info: - raise errors.BlockDeviceError("Can't compute PV info for vg %s" % - self._vg_name) - pvs_info.sort() - pvs_info.reverse() - free_size, pv_name = pvs_info[0] + vg_info = self.GetVGInfo([self._vg_name]) + if not vg_info: + _ThrowError("Can't compute VG info for vg %s", self._vg_name) + free_size, _, _ = vg_info[0] if free_size < size: - raise errors.BlockDeviceError("Not enough free space: required %s," - " available %s" % (size, free_size)) + _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 - %s" % - (result.cmd, result.fail_reason, - result.output)) + _ThrowError("command: %s error: %s - %s", + result.cmd, result.fail_reason, result.output) - return snap_name + return (self._vg_name, snap_name) def SetInfo(self, text): """Update metadata with info text. @@ -537,13 +749,20 @@ class LogicalVolume(BlockDev): result = utils.RunCmd(["lvchange", "--addtag", text, self.dev_path]) if result.failed: - raise errors.BlockDeviceError("Command: %s error: %s - %s" % - (result.cmd, result.fail_reason, - result.output)) + _ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason, + result.output) + def Grow(self, amount): """Grow the logical volume. """ + if self.pe_size is None or self.stripe_count is None: + if not self.Attach(): + _ThrowError("Can't attach to LV during Grow()") + full_stripe_size = self.pe_size * self.stripe_count + rest = amount % full_stripe_size + if rest != 0: + amount += full_stripe_size - rest # we try multiple algorithms since the 'best' ones might not have # space available in the right place, but later ones might (since # they have less constraints); also note that only recent LVM @@ -553,8 +772,7 @@ class LogicalVolume(BlockDev): "-L", "+%dm" % amount, self.dev_path]) if not result.failed: return - raise errors.BlockDeviceError("Can't grow LV %s: %s" % - (self.dev_path, result.output)) + _ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) class DRBD8Status(object): @@ -563,33 +781,85 @@ class DRBD8Status(object): Note that this doesn't support unconfigured devices (cs:Unconfigured). """ - LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+st:([^/]+)/(\S+)" + 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): - 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) - - self.is_standalone = self.cstatus == "StandAlone" - self.is_wfconn = self.cstatus == "WFConnection" - self.is_connected = self.cstatus == "Connected" - self.is_primary = self.lrole == "Primary" - self.is_secondary = self.lrole == "Secondary" - self.peer_primary = self.rrole == "Primary" - self.peer_secondary = self.rrole == "Secondary" + 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 == "Diskless" - self.is_disk_uptodate = self.ldisk == "UpToDate" + self.is_diskless = self.ldisk == self.DS_DISKLESS + self.is_disk_uptodate = self.ldisk == self.DS_UPTODATE + + self.is_in_resync = self.cstatus in self.CSET_SYNC + self.is_in_use = self.cstatus != self.CS_UNCONFIGURED m = self.SYNC_RE.match(procline) if m: @@ -599,23 +869,28 @@ class DRBD8Status(object): seconds = int(m.group(4)) self.est_time = hours * 3600 + minutes * 60 + seconds else: - self.sync_percent = None + # we have (in this if branch) no percent information, but if + # we're resyncing we need to 'fake' a sync percent information, + # as this is how cmdlib determines if it makes sense to wait for + # resyncing or not + if self.is_in_resync: + self.sync_percent = 0 + else: + self.sync_percent = None self.est_time = None - self.is_sync_target = self.peer_sync_source = self.cstatus == "SyncTarget" - self.peer_sync_target = self.is_sync_source = self.cstatus == "SyncSource" - self.is_resync = self.is_sync_target or self.is_sync_source - -class BaseDRBD(BlockDev): +class BaseDRBD(BlockDev): # pylint: disable-msg=W0223 """Base DRBD class. This class contains a few bits of common functionality between the 0.7 and 8.x versions of DRBD. """ - _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)" + _VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)(?:\.\d+)?" r" \(api:(\d+)/proto:(\d+)(?:-(\d+))?\)") + _VALID_LINE_RE = re.compile("^ *([0-9]+): cs:([^ ]+).*$") + _UNUSED_LINE_RE = re.compile("^ *([0-9]+): cs:Unconfigured$") _DRBD_MAJOR = 147 _ST_UNCONFIGURED = "Unconfigured" @@ -623,34 +898,39 @@ class BaseDRBD(BlockDev): _ST_CONNECTED = "Connected" _STATUS_FILE = "/proc/drbd" + _USERMODE_HELPER_FILE = "/sys/module/drbd/parameters/usermode_helper" @staticmethod def _GetProcData(filename=_STATUS_FILE): """Return data from /proc/drbd. """ - stat = open(filename, "r") try: - data = stat.read().splitlines() - finally: - stat.close() + data = utils.ReadFile(filename).splitlines() + except EnvironmentError, err: + if err.errno == errno.ENOENT: + _ThrowError("The file %s cannot be opened, check if the module" + " is loaded (%s)", filename, str(err)) + else: + _ThrowError("Can't read the DRBD proc file %s: %s", filename, str(err)) if not data: - raise errors.BlockDeviceError("Can't read any data from %s" % filename) + _ThrowError("Can't read any data from %s", filename) return data - @staticmethod - def _MassageProcData(data): + @classmethod + def _MassageProcData(cls, data): """Transform the output of _GetProdData into a nicer form. - Returns: - a dictionary of minor: joined lines from /proc/drbd for that minor + @return: a dictionary of minor: joined lines from /proc/drbd + for that minor """ - lmatch = re.compile("^ *([0-9]+):.*$") results = {} old_minor = old_line = None for line in data: - lresult = lmatch.match(line) + if not line: # completely empty lines, as can be returned by drbd8.0+ + continue + lresult = cls._VALID_LINE_RE.match(line) if lresult is not None: if old_minor is not None: results[old_minor] = old_line @@ -665,19 +945,18 @@ class BaseDRBD(BlockDev): return results @classmethod - def _GetVersion(cls): + def _GetVersion(cls, proc_data): """Return the DRBD version. This will return a dict with keys: - k_major, - k_minor, - k_point, - api, - proto, - proto2 (only on drbd > 8.2.X) + - k_major + - k_minor + - k_point + - api + - proto + - proto2 (only on drbd > 8.2.X) """ - proc_data = cls._GetProcData() first_line = proc_data[0].strip() version = cls._VERSION_RE.match(first_line) if not version: @@ -697,6 +976,23 @@ class BaseDRBD(BlockDev): return retval @staticmethod + def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): + """Returns DRBD usermode_helper currently set. + + """ + try: + helper = utils.ReadFile(filename).splitlines()[0] + except EnvironmentError, err: + if err.errno == errno.ENOENT: + _ThrowError("The file %s cannot be opened, check if the module" + " is loaded (%s)", filename, str(err)) + else: + _ThrowError("Can't read DRBD helper file %s: %s", filename, str(err)) + if not helper: + _ThrowError("Can't read any data from %s", filename) + return helper + + @staticmethod def _DevPath(minor): """Return the path to a drbd device for a given minor. @@ -704,16 +1000,15 @@ class BaseDRBD(BlockDev): return "/dev/drbd%d" % minor @classmethod - def _GetUsedDevs(cls): + def GetUsedDevs(cls): """Compute the list of used DRBD devices. """ data = cls._GetProcData() used_devs = {} - valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$") 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)) @@ -748,22 +1043,23 @@ class BaseDRBD(BlockDev): """ result = utils.RunCmd(["blockdev", "--getsize", meta_device]) if result.failed: - logging.error("Failed to get device size: %s - %s", - result.fail_reason, result.output) - return False + _ThrowError("Failed to get device size: %s - %s", + result.fail_reason, result.output) try: sectors = int(result.stdout) - except ValueError: - logging.error("Invalid output from blockdev: '%s'", result.stdout) - return False - bytes = sectors * 512 - if bytes < 128 * 1024 * 1024: # less than 128MiB - logging.error("Meta device too small (%.2fMib)", (bytes / 1024 / 1024)) - return False - if bytes > (128 + 32) * 1024 * 1024: # account for an extra (big) PE on LVM - logging.error("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024)) - return False - return True + except (TypeError, ValueError): + _ThrowError("Invalid output from blockdev: '%s'", result.stdout) + num_bytes = sectors * 512 + if num_bytes < 128 * 1024 * 1024: # less than 128MiB + _ThrowError("Meta device too small (%.2fMib)", (num_bytes / 1024 / 1024)) + # the maximum *valid* size of the meta device when living on top + # of LVM is hard to compute: it depends on the number of stripes + # and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB + # (normal size), but an eight-stripe 128MB PE will result in a 1GB + # size meta device; as such, we restrict it to 1GB (a little bit + # too generous, but making assumptions about PE size is hard) + if num_bytes > 1024 * 1024 * 1024: + _ThrowError("Meta device too big (%.2fMiB)", (num_bytes / 1024 / 1024)) def Rename(self, new_id): """Rename a device. @@ -790,25 +1086,31 @@ 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 - version = self._GetVersion() - if version['k_major'] != 8 : - raise errors.BlockDeviceError("Mismatch in DRBD kernel version and" - " requested ganeti usage: kernel is" - " %s.%s, ganeti wants 8.x" % - (version['k_major'], version['k_minor'])) - if len(children) not in (0, 2): raise ValueError("Invalid configuration data %s" % str(children)) - if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 5: + 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, - self._aminor) = unique_id + 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(self._GetProcData()) + if version['k_major'] != 8 : + _ThrowError("Mismatch in DRBD kernel version and requested ganeti" + " usage: kernel is %s.%s, ganeti wants 8.x", + version['k_major'], version['k_minor']) + if (self._lhost is not None and self._lhost == self._rhost and self._lport == self._rport): raise ValueError("Invalid configuration data, same local/remote %s" % @@ -825,8 +1127,7 @@ 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): @@ -838,14 +1139,12 @@ 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) @@ -857,21 +1156,6 @@ class DRBD8(BaseDRBD): 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: - logging.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. @@ -885,7 +1169,10 @@ class DRBD8(BaseDRBD): # pyparsing setup lbrace = pyp.Literal("{").suppress() rbrace = pyp.Literal("}").suppress() + lbracket = pyp.Literal("[").suppress() + rbracket = pyp.Literal("]").suppress() semi = pyp.Literal(";").suppress() + colon = pyp.Literal(":").suppress() # this also converts the value to an int number = pyp.Word(pyp.nums).setParseAction(lambda s, l, t: int(t[0])) @@ -898,15 +1185,20 @@ 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) + ipv4_addr = (pyp.Optional(pyp.Literal("ipv4")).suppress() + + pyp.Word(pyp.nums + ".") + colon + number) + ipv6_addr = (pyp.Optional(pyp.Literal("ipv6")).suppress() + + pyp.Optional(lbracket) + pyp.Word(pyp.hexnums + ":") + + pyp.Optional(rbracket) + colon + number) # meta device, extended syntax - meta_value = ((value ^ quoted) + pyp.Literal('[').suppress() + - number + pyp.Word(']').suppress()) + meta_value = ((value ^ quoted) + lbracket + number + rbracket) + # device name, extended syntax + device_value = pyp.Literal("minor").suppress() + number # a statement stmt = (~rbrace + keyword + ~lbrace + - pyp.Optional(addr_port ^ value ^ quoted ^ meta_value) + + pyp.Optional(ipv4_addr ^ ipv6_addr ^ value ^ quoted ^ meta_value ^ + device_value) + pyp.Optional(defa) + semi + pyp.Optional(pyp.restOfLine).suppress()) @@ -953,8 +1245,7 @@ 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: @@ -1029,21 +1320,37 @@ 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 args = ["drbdsetup", cls._DevPath(minor), "disk", - backend, meta, "0", "-e", "detach", "--create-device"] + backend, meta, "0", + "-e", "detach", + "--create-device"] + if size: + args.extend(["-d", "%sm" % size]) + if not constants.DRBD_BARRIERS: # disable barriers, if configured so + version = cls._GetVersion(cls._GetProcData()) + # 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: - logging.error("Can't attach local disk: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output) @classmethod def _AssembleNet(cls, minor, net_info, protocol, @@ -1055,10 +1362,33 @@ 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) + + if netutils.IP6Address.IsValid(lhost): + if not netutils.IP6Address.IsValid(rhost): + _ThrowError("drbd%d: can't connect ip %s to ip %s" % + (minor, lhost, rhost)) + family = "ipv6" + elif netutils.IP4Address.IsValid(lhost): + if not netutils.IP4Address.IsValid(rhost): + _ThrowError("drbd%d: can't connect ip %s to ip %s" % + (minor, lhost, rhost)) + family = "ipv4" + else: + _ThrowError("drbd%d: Invalid ip %s" % (minor, lhost)) args = ["drbdsetup", cls._DevPath(minor), "net", - "%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), protocol, + "%s:%s:%s" % (family, lhost, lport), + "%s:%s:%s" % (family, rhost, rport), protocol, "-A", "discard-zero-changes", "-B", "consensus", "--create-device", @@ -1069,52 +1399,44 @@ class DRBD8(BaseDRBD): args.extend(["-a", hmac, "-x", secret]) result = utils.RunCmd(args) if result.failed: - logging.error("Can't setup network for dbrd device: %s - %s", - result.fail_reason, result.output) - return False + _ThrowError("drbd%d: can't setup network: %s - %s", + minor, result.fail_reason, result.output) - timeout = time.time() + 10 - ok = False - while time.time() < timeout: + def _CheckNetworkConfig(): info = cls._GetDevInfo(cls._GetShowData(minor)) if not "local_addr" in info or not "remote_addr" in info: - 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: - logging.error("Timeout while configuring network") - return False - return True + raise utils.RetryAgain() + + try: + utils.Retry(_CheckNetworkConfig, 1.0, 10.0) + except utils.RetryTimeout: + _ThrowError("drbd%d: timeout while configuring network", minor) def AddChildren(self, devices): """Add a disk to the DRBD device. """ 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): @@ -1122,42 +1444,85 @@ 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 :) - logging.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 + """ + if self.minor is None: + logging.info("Not attached during SetSyncSpeed") + return False children_result = super(DRBD8, self).SetSyncSpeed(kbytes) + return self._SetMinorSyncSpeed(self.minor, kbytes) and children_result + + def PauseResumeSync(self, pause): + """Pauses or resumes the sync of a DRBD device. + + @param pause: Wether to pause or resume + @return: the success of the operation + + """ if self.minor is None: - logging.info("Instance not attached to a device") + logging.info("Not attached during PauseSync") return False - result = utils.RunCmd(["drbdsetup", self.dev_path, "syncer", "-r", "%d" % - kbytes]) + + children_result = super(DRBD8, self).PauseResumeSync(pause) + + if pause: + cmd = "pause-sync" + else: + cmd = "resume-sync" + + result = utils.RunCmd(["drbdsetup", self.dev_path, cmd]) if result.failed: - logging.error("Can't change syncer rate: %s - %s", + logging.error("Can't %s: %s - %s", cmd, result.fail_reason, result.output) return not result.failed and children_result @@ -1166,37 +1531,50 @@ class DRBD8(BaseDRBD): """ if self.minor is None: - raise errors.BlockDeviceError("GetStats() called while not attached") + _ThrowError("drbd%d: GetStats() called while not attached", self._aminor) 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) + _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. + @rtype: objects.BlockDevStatus + """ if self.minor is None and not self.Attach(): - raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") + _ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) + stats = self.GetProcStatus() - ldisk = not stats.is_disk_uptodate - is_degraded = not stats.is_connected - return stats.sync_percent, stats.est_time, is_degraded or ldisk, ldisk + is_degraded = not stats.is_connected or not stats.is_disk_uptodate + + if stats.is_disk_uptodate: + ldisk_status = constants.LDS_OKAY + elif stats.is_diskless: + ldisk_status = constants.LDS_FAULTY + else: + ldisk_status = constants.LDS_UNKNOWN + + return objects.BlockDevStatus(dev_path=self.dev_path, + major=self.major, + minor=self.minor, + sync_percent=stats.sync_percent, + estimated_time=stats.est_time, + is_degraded=is_degraded, + ldisk_status=ldisk_status) def Open(self, force=False): """Make the local state primary. @@ -1215,9 +1593,8 @@ class DRBD8(BaseDRBD): cmd.append("-o") result = utils.RunCmd(cmd) if result.failed: - msg = ("Can't make drbd device primary: %s" % result.output) - logging.error(msg) - raise errors.BlockDeviceError(msg) + _ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, + result.output) def Close(self): """Make the local state secondary. @@ -1226,41 +1603,178 @@ class DRBD8(BaseDRBD): """ if self.minor is None and not self.Attach(): - logging.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: - msg = ("Can't switch drbd device to" - " secondary: %s" % result.output) - logging.error(msg) - raise errors.BlockDeviceError(msg) + _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 + # 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 @@ -1272,74 +1786,47 @@ class DRBD8(BaseDRBD): # else, even though its local storage is ours; as we own the # drbd space, we try to disconnect from the remote peer and # reconnect to our correct one - 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: - logging.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 - logging.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): @@ -1351,8 +1838,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) if result.failed: - logging.error("Can't detach local device: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't detach local disk: %s", minor, result.output) @classmethod def _ShutdownNet(cls, minor): @@ -1363,8 +1849,7 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) if result.failed: - logging.error("Can't shutdown network: %s", result.output) - return not result.failed + _ThrowError("drbd%d: can't shutdown network: %s", minor, result.output) @classmethod def _ShutdownAll(cls, minor): @@ -1375,27 +1860,26 @@ class DRBD8(BaseDRBD): """ result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) if result.failed: - logging.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(): - logging.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): @@ -1407,31 +1891,38 @@ 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: - raise errors.ProgrammerError("drbd8: Grow called while not attached") + _ThrowError("drbd%d: Grow called while not attached", self._aminor) if len(self._children) != 2 or None in self._children: - raise errors.BlockDeviceError("Cannot grow diskless DRBD8 device") + _ThrowError("drbd%d: cannot grow diskless device", self.minor) self._children[0].Grow(amount) - result = utils.RunCmd(["drbdsetup", self.dev_path, "resize"]) + result = utils.RunCmd(["drbdsetup", self.dev_path, "resize", "-s", + "%dm" % (self.size + amount)]) if result.failed: - raise errors.BlockDeviceError("resize failed for %s: %s" % - (self.dev_path, result.output)) - return + _ThrowError("drbd%d: resize failed: %s", self.minor, result.output) class FileStorage(BlockDev): @@ -1442,17 +1933,18 @@ class FileStorage(BlockDev): The unique_id for the file device is a (file_driver, file_path) tuple. """ - def __init__(self, unique_id, children): + def __init__(self, unique_id, children, size): """Initalizes a file device backend. """ if children: raise errors.BlockDeviceError("Invalid setup for file device") - super(FileStorage, self).__init__(unique_id, children) + super(FileStorage, self).__init__(unique_id, children, size) if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) self.driver = unique_id[0] self.dev_path = unique_id[1] + self.Attach() def Assemble(self): """Assemble the device. @@ -1461,18 +1953,16 @@ class FileStorage(BlockDev): """ if not os.path.exists(self.dev_path): - raise errors.BlockDeviceError("File device '%s' does not exist." % - self.dev_path) - return True + _ThrowError("File device '%s' does not exist" % self.dev_path) def Shutdown(self): """Shutdown the device. - This is a no-op for the file type, as we don't deacivate + This is a no-op for the file type, as we don't deactivate the file on shutdown. """ - return True + pass def Open(self, force=False): """Make the device ready for I/O. @@ -1493,66 +1983,102 @@ class FileStorage(BlockDev): def Remove(self): """Remove the file backing the block device. - Returns: - boolean indicating wheter removal of file was successful or not. + @rtype: boolean + @return: True if the removal was successful """ - if not os.path.exists(self.dev_path): - return True try: os.remove(self.dev_path) - return True except OSError, err: - logging.error("Can't remove file '%s': %s", self.dev_path, err) - return False + 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. - Returns: - boolean indicating if file exists or not. + @rtype: boolean + @return: True if file exists """ - if os.path.exists(self.dev_path): - return True - return False + 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. - Args: - children: - size: integer size of file in MiB + @param size: the size of file in MiB - Returns: - A ganeti.bdev.FileStorage object. + @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: - f = open(dev_path, 'w') - except IOError, err: - raise errors.BlockDeviceError("Could not create '%'" % err) - else: + 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) + return FileStorage(unique_id, children, size) DEV_MAP = { constants.LD_LV: LogicalVolume, constants.LD_DRBD8: DRBD8, - constants.LD_FILE: FileStorage, } +if constants.ENABLE_FILE_STORAGE: + DEV_MAP[constants.LD_FILE] = FileStorage + -def FindDevice(dev_type, unique_id, children): +def FindDevice(dev_type, unique_id, children, size): """Search for an existing, assembled device. This will succeed only if the device exists and is assembled, but it @@ -1561,28 +2087,23 @@ 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) + 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.attached: - device.Assemble() - if not device.attached: - 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