X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/f208978a68d2e4145254f2047eff977e97d262f5..5959f75d30de7352520ec7a611614c1fd8b3d95e:/lib/bdev.py diff --git a/lib/bdev.py b/lib/bdev.py index 534b835..3f9441d 100644 --- a/lib/bdev.py +++ b/lib/bdev.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2006, 2007 Google Inc. +# Copyright (C) 2006, 2007, 2010, 2011 Google Inc. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by @@ -24,6 +24,8 @@ import re import time import errno +import shlex +import stat import pyparsing as pyp import os import logging @@ -32,6 +34,12 @@ 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): @@ -48,7 +56,7 @@ def _IgnoreError(fn, *args, **kwargs): fn(*args, **kwargs) return True except errors.BlockDeviceError, err: - logging.warning("Caught BlockDeviceError but ignoring: %s" % str(err)) + logging.warning("Caught BlockDeviceError but ignoring: %s", str(err)) return False @@ -66,6 +74,20 @@ def _ThrowError(msg, *args): raise errors.BlockDeviceError(msg) +def _CanReadDevice(path): + """Check if we can read from the given device. + + This tries to read the first 128k of the device. + + """ + try: + utils.ReadFile(path, size=_DEVICE_READ_SIZE) + return True + except EnvironmentError: + logging.warning("Can't read from device %s", path, exc_info=True) + return False + + class BlockDev(object): """Block device abstract class. @@ -109,7 +131,7 @@ class BlockDev(object): after assembly we'll have our correct major/minor. """ - def __init__(self, unique_id, children, size): + def __init__(self, unique_id, children, size, params): self._children = children self.dev_path = None self.unique_id = unique_id @@ -117,6 +139,7 @@ class BlockDev(object): self.minor = None self.attached = False self.size = size + self.params = params def Assemble(self): """Assemble the device from its components. @@ -145,7 +168,7 @@ class BlockDev(object): raise NotImplementedError @classmethod - def Create(cls, unique_id, children, size): + def Create(cls, unique_id, children, size, params): """Create the device. If the device cannot be created, it will return None @@ -210,6 +233,20 @@ 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. @@ -286,7 +323,6 @@ class BlockDev(object): is_degraded=is_degraded, ldisk_status=ldisk_status) - def SetInfo(self, text): """Update metadata with info text. @@ -296,14 +332,35 @@ class BlockDev(object): for child in self._children: child.SetInfo(text) - def Grow(self, amount): + def Grow(self, amount, dryrun): """Grow the block device. + @type amount: integer @param amount: the amount (in mebibytes) to grow with + @type dryrun: boolean + @param dryrun: whether to execute the operation in simulation mode + only, without actually increasing the size """ 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, @@ -314,23 +371,29 @@ class LogicalVolume(BlockDev): """Logical Volume block device. """ - def __init__(self, unique_id, children, size): + _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, params): """Attaches to a LV device. The unique_id is a tuple (vg_name, lv_name) """ - super(LogicalVolume, self).__init__(unique_id, children, size) + super(LogicalVolume, self).__init__(unique_id, children, size, params) if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) self._vg_name, self._lv_name = unique_id - self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name) + self._ValidateName(self._vg_name) + self._ValidateName(self._lv_name) + self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name) self._degraded = True self.major = self.minor = self.pe_size = self.stripe_count = None self.Attach() @classmethod - def Create(cls, unique_id, children, size): + def Create(cls, unique_id, children, size, params): """Create a new logical volume. """ @@ -338,16 +401,26 @@ class LogicalVolume(BlockDev): raise errors.ProgrammerError("Invalid configuration data %s" % str(unique_id)) vg_name, lv_name = unique_id - pvs_info = cls.GetPVInfo(vg_name) + cls._ValidateName(vg_name) + cls._ValidateName(lv_name) + pvs_info = cls.GetPVInfo([vg_name]) if not pvs_info: _ThrowError("Can't compute PV info for vg %s", vg_name) pvs_info.sort() pvs_info.reverse() - pvlist = [ pv[1] for pv in pvs_info ] - free_size = sum([ pv[0] for pv in pvs_info ]) + 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) + desired_stripes = params[constants.LDP_STRIPES] + stripes = min(current_pvs, desired_stripes) + if stripes < desired_stripes: + logging.warning("Could not use %d stripes for VG %s, as only %d PVs are" + " available.", desired_stripes, vg_name, current_pvs) # The size constraint should have been checked from the master before # calling the create function. @@ -366,39 +439,116 @@ class LogicalVolume(BlockDev): if result.failed: _ThrowError("LV create failed (%s): %s", result.fail_reason, result.output) - return LogicalVolume(unique_id, children, size) + return LogicalVolume(unique_id, children, size, params) @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. - @param vg_name: the volume group name + @param vg_names: list of volume group names, if empty all will be returned + @param filter_allocatable: whether to skip over unallocatable PVs @rtype: list @return: list of tuples (free_space, name) with free_space in mebibytes """ - 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 + # (possibly) skip over vgs which are not in the right volume group(s) + if vg_names and vg_name not in vg_names: continue - data.append((float(fields[2]), fields[0])) + 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. @@ -426,7 +576,7 @@ class LogicalVolume(BlockDev): if result.failed: _ThrowError("Failed to rename the logical volume: %s", result.output) self._lv_name = new_name - self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name) + self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name) def Attach(self): """Attach to an existing LV. @@ -455,7 +605,7 @@ class LogicalVolume(BlockDev): # 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[-1].strip().rstrip(",") out = out.split(",") if len(out) != 5: logging.error("Can't parse LVS output, len(%s) != 5", str(out)) @@ -469,7 +619,7 @@ 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: @@ -488,7 +638,7 @@ class LogicalVolume(BlockDev): self.minor = minor self.pe_size = pe_size self.stripe_count = stripes - self._degraded = status[0] == 'v' # virtual volume, i.e. doesn't backing + self._degraded = status[0] == "v" # virtual volume, i.e. doesn't backing # storage self.attached = True return True @@ -568,19 +718,19 @@ 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, size) + snap = LogicalVolume((self._vg_name, snap_name), None, size, self.params) _IgnoreError(snap.Remove) - pvs_info = self.GetPVInfo(self._vg_name) - if not pvs_info: - _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] + 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: _ThrowError("Not enough free space: required %s," " available %s", size, free_size) @@ -591,7 +741,7 @@ class LogicalVolume(BlockDev): _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. @@ -600,8 +750,8 @@ class LogicalVolume(BlockDev): BlockDev.SetInfo(self, text) # Replace invalid characters - text = re.sub('^[^A-Za-z0-9_+.]', '_', text) - text = re.sub('[^-A-Za-z0-9_+.]', '_', text) + text = re.sub("^[^A-Za-z0-9_+.]", "_", text) + text = re.sub("[^-A-Za-z0-9_+.]", "_", text) # Only up to 128 characters are allowed text = text[:128] @@ -612,7 +762,7 @@ class LogicalVolume(BlockDev): _ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason, result.output) - def Grow(self, amount): + def Grow(self, amount, dryrun): """Grow the logical volume. """ @@ -623,13 +773,15 @@ class LogicalVolume(BlockDev): rest = amount % full_stripe_size if rest != 0: amount += full_stripe_size - rest + cmd = ["lvextend", "-L", "+%dm" % amount] + if dryrun: + cmd.append("--test") # we try multiple algorithms since the 'best' ones might not have # space available in the right place, but later ones might (since # they have less constraints); also note that only recent LVM # supports 'cling' for alloc_policy in "contiguous", "cling", "normal": - result = utils.RunCmd(["lvextend", "--alloc", alloc_policy, - "-L", "+%dm" % amount, self.dev_path]) + result = utils.RunCmd(cmd + ["--alloc", alloc_policy, self.dev_path]) if not result.failed: return _ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) @@ -645,7 +797,10 @@ class DRBD8Status(object): 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.*$") + # Due to a bug in drbd in the kernel, introduced in + # commit 4b0715f096 (still unfixed as of 2011-08-22) + "(?:\s|M)" + "finish: ([0-9]+):([0-9]+):([0-9]+)\s.*$") CS_UNCONFIGURED = "Unconfigured" CS_STANDALONE = "StandAlone" @@ -740,15 +895,17 @@ class DRBD8Status(object): self.est_time = None -class BaseDRBD(BlockDev): +class BaseDRBD(BlockDev): # pylint: disable=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" @@ -756,6 +913,7 @@ 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): @@ -763,11 +921,7 @@ class BaseDRBD(BlockDev): """ try: - 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" @@ -778,19 +932,20 @@ class BaseDRBD(BlockDev): _ThrowError("Can't read any data from %s", filename) return data - @staticmethod - def _MassageProcData(data): + @classmethod + def _MassageProcData(cls, data): """Transform the output of _GetProdData into a nicer form. @return: a dictionary of minor: joined lines from /proc/drbd for that minor """ - lmatch = re.compile("^ *([0-9]+):.*$") results = {} old_minor = old_line = None for line in data: - lresult = lmatch.match(line) + if not line: # completely empty lines, as can be returned by drbd8.0+ + continue + lresult = cls._VALID_LINE_RE.match(line) if lresult is not None: if old_minor is not None: results[old_minor] = old_line @@ -805,7 +960,7 @@ class BaseDRBD(BlockDev): return results @classmethod - def _GetVersion(cls): + def _GetVersion(cls, proc_data): """Return the DRBD version. This will return a dict with keys: @@ -817,7 +972,6 @@ class BaseDRBD(BlockDev): - proto2 (only on drbd > 8.2.X) """ - proc_data = cls._GetProcData() first_line = proc_data[0].strip() version = cls._VERSION_RE.match(first_line) if not version: @@ -825,18 +979,35 @@ class BaseDRBD(BlockDev): first_line) values = version.groups() - retval = {'k_major': int(values[0]), - 'k_minor': int(values[1]), - 'k_point': int(values[2]), - 'api': int(values[3]), - 'proto': int(values[4]), + retval = {"k_major": int(values[0]), + "k_minor": int(values[1]), + "k_point": int(values[2]), + "api": int(values[3]), + "proto": int(values[4]), } if values[5] is not None: - retval['proto2'] = values[5] + retval["proto2"] = values[5] return retval @staticmethod + def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): + """Returns DRBD usermode_helper currently set. + + """ + try: + helper = utils.ReadFile(filename).splitlines()[0] + except EnvironmentError, err: + if err.errno == errno.ENOENT: + _ThrowError("The file %s cannot be opened, check if the module" + " is loaded (%s)", filename, str(err)) + else: + _ThrowError("Can't read DRBD helper file %s: %s", filename, str(err)) + if not helper: + _ThrowError("Can't read any data from %s", filename) + return helper + + @staticmethod def _DevPath(minor): """Return the path to a drbd device for a given minor. @@ -851,9 +1022,8 @@ class BaseDRBD(BlockDev): data = cls._GetProcData() used_devs = {} - valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$") 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)) @@ -892,19 +1062,19 @@ class BaseDRBD(BlockDev): result.fail_reason, result.output) try: sectors = int(result.stdout) - except ValueError: + 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)) + 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 bytes > 1024 * 1024 * 1024: - _ThrowError("Meta device too big (%.2fMiB)", (bytes / 1024 / 1024)) + if num_bytes > 1024 * 1024 * 1024: + _ThrowError("Meta device too big (%.2fMiB)", (num_bytes / 1024 / 1024)) def Rename(self, new_id): """Rename a device. @@ -934,17 +1104,15 @@ class DRBD8(BaseDRBD): # timeout constants _NET_RECONFIG_TIMEOUT = 60 - def __init__(self, unique_id, children, size): + # command line options for barriers + _DISABLE_DISK_OPTION = "--no-disk-barrier" # -a + _DISABLE_DRAIN_OPTION = "--no-disk-drain" # -D + _DISABLE_FLUSH_OPTION = "--no-disk-flushes" # -i + _DISABLE_META_FLUSH_OPTION = "--no-md-flushes" # -m + + def __init__(self, unique_id, children, size, params): if children and children.count(None) > 0: children = [] - 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 len(children) not in (0, 2): raise ValueError("Invalid configuration data %s" % str(children)) if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6: @@ -952,6 +1120,18 @@ class DRBD8(BaseDRBD): (self._lhost, self._lport, self._rhost, self._rport, self._aminor, self._secret) = unique_id + if children: + if not _CanReadDevice(children[1].dev_path): + logging.info("drbd%s: Ignoring unreadable meta device", self._aminor) + children = [] + super(DRBD8, self).__init__(unique_id, children, size, params) + self.major = self._DRBD_MAJOR + version = self._GetVersion(self._GetProcData()) + if version["k_major"] != 8: + _ThrowError("Mismatch in DRBD kernel version and requested ganeti" + " usage: kernel is %s.%s, ganeti wants 8.x", + version["k_major"], version["k_minor"]) + if (self._lhost is not None and self._lhost == self._rhost and self._lport == self._rport): raise ValueError("Invalid configuration data, same local/remote %s" % @@ -965,6 +1145,17 @@ class DRBD8(BaseDRBD): This will not work if the given minor is in use. """ + # Zero the metadata first, in order to make sure drbdmeta doesn't + # try to auto-detect existing filesystems or similar (see + # http://code.google.com/p/ganeti/issues/detail?id=182); we only + # care about the first 128MB of data in the device, even though it + # can be bigger + result = utils.RunCmd([constants.DD_CMD, + "if=/dev/zero", "of=%s" % dev_path, + "bs=1048576", "count=128", "oflag=direct"]) + if result.failed: + _ThrowError("Can't wipe the meta device: %s", result.output) + result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), "v08", dev_path, "0", "create-md"]) if result.failed: @@ -980,14 +1171,12 @@ 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) @@ -1012,11 +1201,14 @@ 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])) - comment = pyp.Literal ("#") + pyp.Optional(pyp.restOfLine) + comment = pyp.Literal("#") + pyp.Optional(pyp.restOfLine) defa = pyp.Literal("_is_default").suppress() dbl_quote = pyp.Literal('"').suppress() @@ -1025,25 +1217,25 @@ class DRBD8(BaseDRBD): # value types value = pyp.Word(pyp.alphanums + '_-/.:') quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote - 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) + 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()) # an entire section - section_name = pyp.Word(pyp.alphas + '_') + section_name = pyp.Word(pyp.alphas + "_") section = section_name + lbrace + pyp.ZeroOrMore(pyp.Group(stmt)) + rbrace bnf = pyp.ZeroOrMore(pyp.Group(section ^ stmt)) @@ -1159,22 +1351,110 @@ class DRBD8(BaseDRBD): info["remote_addr"] == (self._rhost, self._rport)) return retval - @classmethod - def _AssembleLocal(cls, minor, backend, meta, size): + def _AssembleLocal(self, minor, backend, meta, size): """Configure the local part of a DRBD device. """ - args = ["drbdsetup", cls._DevPath(minor), "disk", + args = ["drbdsetup", self._DevPath(minor), "disk", backend, meta, "0", - "-d", "%sm" % size, "-e", "detach", "--create-device"] + if size: + args.extend(["-d", "%sm" % size]) + + version = self._GetVersion(self._GetProcData()) + vmaj = version["k_major"] + vmin = version["k_minor"] + vrel = version["k_point"] + + barrier_args = \ + self._ComputeDiskBarrierArgs(vmaj, vmin, vrel, + self.params[constants.LDP_BARRIERS], + self.params[constants.LDP_NO_META_FLUSH]) + args.extend(barrier_args) + + if self.params[constants.LDP_DISK_CUSTOM]: + args.extend(shlex.split(self.params[constants.LDP_DISK_CUSTOM])) + result = utils.RunCmd(args) if result.failed: _ThrowError("drbd%d: can't attach local disk: %s", minor, result.output) @classmethod - def _AssembleNet(cls, minor, net_info, protocol, + def _ComputeDiskBarrierArgs(cls, vmaj, vmin, vrel, disabled_barriers, + disable_meta_flush): + """Compute the DRBD command line parameters for disk barriers + + Returns a list of the disk barrier parameters as requested via the + disabled_barriers and disable_meta_flush arguments, and according to the + supported ones in the DRBD version vmaj.vmin.vrel + + If the desired option is unsupported, raises errors.BlockDeviceError. + + """ + disabled_barriers_set = frozenset(disabled_barriers) + if not disabled_barriers_set in constants.DRBD_VALID_BARRIER_OPT: + raise errors.BlockDeviceError("%s is not a valid option set for DRBD" + " barriers" % disabled_barriers) + + args = [] + + # The following code assumes DRBD 8.x, with x < 4 and x != 1 (DRBD 8.1.x + # does not exist) + if not vmaj == 8 and vmin in (0, 2, 3): + raise errors.BlockDeviceError("Unsupported DRBD version: %d.%d.%d" % + (vmaj, vmin, vrel)) + + def _AppendOrRaise(option, min_version): + """Helper for DRBD options""" + if min_version is not None and vrel >= min_version: + args.append(option) + else: + raise errors.BlockDeviceError("Could not use the option %s as the" + " DRBD version %d.%d.%d does not support" + " it." % (option, vmaj, vmin, vrel)) + + # the minimum version for each feature is encoded via pairs of (minor + # version -> x) where x is version in which support for the option was + # introduced. + meta_flush_supported = disk_flush_supported = { + 0: 12, + 2: 7, + 3: 0, + } + + disk_drain_supported = { + 2: 7, + 3: 0, + } + + disk_barriers_supported = { + 3: 0, + } + + # meta flushes + if disable_meta_flush: + _AppendOrRaise(cls._DISABLE_META_FLUSH_OPTION, + meta_flush_supported.get(vmin, None)) + + # disk flushes + if constants.DRBD_B_DISK_FLUSH in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_FLUSH_OPTION, + disk_flush_supported.get(vmin, None)) + + # disk drain + if constants.DRBD_B_DISK_DRAIN in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_DRAIN_OPTION, + disk_drain_supported.get(vmin, None)) + + # disk barriers + if constants.DRBD_B_DISK_BARRIERS in disabled_barriers_set: + _AppendOrRaise(cls._DISABLE_DISK_OPTION, + disk_barriers_supported.get(vmin, None)) + + return args + + def _AssembleNet(self, minor, net_info, protocol, dual_pri=False, hmac=None, secret=None): """Configure the network part of the device. @@ -1183,7 +1463,7 @@ class DRBD8(BaseDRBD): if None in net_info: # we don't want network connection and actually want to make # sure its shutdown - cls._ShutdownNet(minor) + self._ShutdownNet(minor) return # Workaround for a race condition. When DRBD is doing its dance to @@ -1192,10 +1472,25 @@ class DRBD8(BaseDRBD): # 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) + sync_speed = self.params.get(constants.LDP_RESYNC_RATE) + self._SetMinorSyncSpeed(minor, 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, + args = ["drbdsetup", self._DevPath(minor), "net", + "%s:%s:%s" % (family, lhost, lport), + "%s:%s:%s" % (family, rhost, rport), protocol, "-A", "discard-zero-changes", "-B", "consensus", "--create-device", @@ -1204,25 +1499,27 @@ class DRBD8(BaseDRBD): args.append("-m") if hmac and secret: args.extend(["-a", hmac, "-x", secret]) + + if self.params[constants.LDP_NET_CUSTOM]: + args.extend(shlex.split(self.params[constants.LDP_NET_CUSTOM])) + result = utils.RunCmd(args) if result.failed: _ThrowError("drbd%d: can't setup network: %s - %s", minor, result.fail_reason, result.output) - timeout = time.time() + 10 - ok = False - while time.time() < timeout: - info = cls._GetDevInfo(cls._GetShowData(minor)) + def _CheckNetworkConfig(): + info = self._GetDevInfo(self._GetShowData(minor)) if not "local_addr" in info or not "remote_addr" in info: - time.sleep(1) - continue + raise utils.RetryAgain() + if (info["local_addr"] != (lhost, lport) or info["remote_addr"] != (rhost, rport)): - time.sleep(1) - continue - ok = True - break - if not ok: + 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): @@ -1311,6 +1608,30 @@ class DRBD8(BaseDRBD): 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("Not attached during PauseSync") + return False + + 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 %s: %s - %s", cmd, + result.fail_reason, result.output) + return not result.failed and children_result + def GetProcStatus(self): """Return device data from /proc. @@ -1415,31 +1736,42 @@ class DRBD8(BaseDRBD): _ThrowError("drbd%d: DRBD disk missing network info in" " DisconnectNet()", self.minor) - ever_disconnected = _IgnoreError(self._ShutdownNet, self.minor) - timeout_limit = time.time() + self._NET_RECONFIG_TIMEOUT - sleep_time = 0.100 # we start the retry time at 100 milliseconds - while time.time() < timeout_limit: - status = self.GetProcStatus() - if status.is_standalone: - break - # retry the disconnect, it seems possible that due to a - # well-time disconnect on the peer, my disconnect command might - # be ignored and forgotten - ever_disconnected = _IgnoreError(self._ShutdownNet, self.minor) or \ - ever_disconnected - time.sleep(sleep_time) - sleep_time = min(2, sleep_time * 1.5) + class _DisconnectStatus: + def __init__(self, ever_disconnected): + self.ever_disconnected = ever_disconnected - if not status.is_standalone: - if ever_disconnected: + dstatus = _DisconnectStatus(_IgnoreError(self._ShutdownNet, self.minor)) + + def _WaitForDisconnect(): + if self.GetProcStatus().is_standalone: + return + + # retry the disconnect, it seems possible that due to a well-time + # disconnect on the peer, my disconnect command might be ignored and + # forgotten + dstatus.ever_disconnected = \ + _IgnoreError(self._ShutdownNet, self.minor) or dstatus.ever_disconnected + + raise utils.RetryAgain() + + # Keep start time + start_time = time.time() + + try: + # Start delay at 100 milliseconds and grow up to 2 seconds + utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0), + self._NET_RECONFIG_TIMEOUT) + except utils.RetryTimeout: + if dstatus.ever_disconnected: msg = ("drbd%d: device did not react to the" " 'disconnect' command in a timely manner") else: msg = "drbd%d: can't shutdown network, even after multiple retries" + _ThrowError(msg, self.minor) - reconfig_time = time.time() - timeout_limit + self._NET_RECONFIG_TIMEOUT - if reconfig_time > 15: # hardcoded alert limit + reconfig_time = time.time() - start_time + if reconfig_time > (self._NET_RECONFIG_TIMEOUT * 0.25): logging.info("drbd%d: DisconnectNet: detach took %.3f seconds", self.minor, reconfig_time) @@ -1497,6 +1829,7 @@ class DRBD8(BaseDRBD): - if we have a configured device, we try to ensure that it matches our config - if not, we create it from zero + - anyway, set the device parameters """ super(DRBD8, self).Assemble() @@ -1510,6 +1843,9 @@ class DRBD8(BaseDRBD): # the device self._SlowAssemble() + sync_speed = self.params.get(constants.LDP_RESYNC_RATE) + self.SetSyncSpeed(sync_speed) + def _SlowAssemble(self): """Assembles the DRBD device from a (partially) configured device. @@ -1518,6 +1854,8 @@ class DRBD8(BaseDRBD): the attach if can return success. """ + # TODO: Rewrite to not use a for loop just because there is 'break' + # pylint: disable=W0631 net_data = (self._lhost, self._lport, self._rhost, self._rport) for minor in (self._aminor,): info = self._GetDevInfo(self._GetShowData(minor)) @@ -1654,7 +1992,7 @@ class DRBD8(BaseDRBD): self.Shutdown() @classmethod - def Create(cls, unique_id, children, size): + def Create(cls, unique_id, children, size, params): """Create a new DRBD8 device. Since DRBD devices are not created per se, just assembled, this @@ -1680,9 +2018,9 @@ class DRBD8(BaseDRBD): aminor, meta) cls._CheckMetaSize(meta.dev_path) cls._InitMeta(aminor, meta.dev_path) - return cls(unique_id, children, size) + return cls(unique_id, children, size, params) - def Grow(self, amount): + def Grow(self, amount, dryrun): """Resize the DRBD device and its backing storage. """ @@ -1690,7 +2028,10 @@ class DRBD8(BaseDRBD): _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) + self._children[0].Grow(amount, dryrun) + if dryrun: + # DRBD does not support dry-run mode, so we'll return here + return result = utils.RunCmd(["drbdsetup", self.dev_path, "resize", "-s", "%dm" % (self.size + amount)]) if result.failed: @@ -1705,13 +2046,13 @@ class FileStorage(BlockDev): The unique_id for the file device is a (file_driver, file_path) tuple. """ - def __init__(self, unique_id, children, size): + def __init__(self, unique_id, children, size, params): """Initalizes a file device backend. """ if children: raise errors.BlockDeviceError("Invalid setup for file device") - super(FileStorage, self).__init__(unique_id, children, size) + super(FileStorage, self).__init__(unique_id, children, size, params) if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) self.driver = unique_id[0] @@ -1765,6 +2106,34 @@ class FileStorage(BlockDev): if err.errno != errno.ENOENT: _ThrowError("Can't remove file '%s': %s", self.dev_path, err) + def Rename(self, new_id): + """Renames the file. + + """ + # TODO: implement rename for file-based storage + _ThrowError("Rename is not supported for file-based storage") + + def Grow(self, amount, dryrun): + """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" + # We can't really simulate the growth + if dryrun: + return + try: + f = open(self.dev_path, "a+") + f.truncate(new_size) + f.close() + except EnvironmentError, err: + _ThrowError("Error in file growth: %", str(err)) + def Attach(self): """Attach to an existing file. @@ -1777,8 +2146,21 @@ class FileStorage(BlockDev): 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): + def Create(cls, unique_id, children, size, params): """Create a new file. @param size: the size of file in MiB @@ -1790,59 +2172,203 @@ class FileStorage(BlockDev): if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: raise ValueError("Invalid configuration data %s" % str(unique_id)) dev_path = unique_id[1] - if os.path.exists(dev_path): - _ThrowError("File already existing: %s", dev_path) try: - f = open(dev_path, 'w') + 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 IOError, err: + 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) + return FileStorage(unique_id, children, size, params) + + +class PersistentBlockDevice(BlockDev): + """A block device with persistent node + + May be either directly attached, or exposed through DM (e.g. dm-multipath). + udev helpers are probably required to give persistent, human-friendly + names. + + For the time being, pathnames are required to lie under /dev. + + """ + def __init__(self, unique_id, children, size, params): + """Attaches to a static block device. + + The unique_id is a path under /dev. + + """ + super(PersistentBlockDevice, self).__init__(unique_id, children, size, + params) + if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: + raise ValueError("Invalid configuration data %s" % str(unique_id)) + self.dev_path = unique_id[1] + if not os.path.realpath(self.dev_path).startswith("/dev/"): + raise ValueError("Full path '%s' lies outside /dev" % + os.path.realpath(self.dev_path)) + # TODO: this is just a safety guard checking that we only deal with devices + # we know how to handle. In the future this will be integrated with + # external storage backends and possible values will probably be collected + # from the cluster configuration. + if unique_id[0] != constants.BLOCKDEV_DRIVER_MANUAL: + raise ValueError("Got persistent block device of invalid type: %s" % + unique_id[0]) + + self.major = self.minor = None + self.Attach() + + @classmethod + def Create(cls, unique_id, children, size, params): + """Create a new device + + This is a noop, we only return a PersistentBlockDevice instance + + """ + return PersistentBlockDevice(unique_id, children, 0, params) + + def Remove(self): + """Remove a device + + This is a noop + + """ + pass + + def Rename(self, new_id): + """Rename this device. + + """ + _ThrowError("Rename is not supported for PersistentBlockDev storage") + + def Attach(self): + """Attach to an existing block device. + + + """ + self.attached = False + try: + st = os.stat(self.dev_path) + except OSError, err: + logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) + return False + + if not stat.S_ISBLK(st.st_mode): + logging.error("%s is not a block device", self.dev_path) + return False + + self.major = os.major(st.st_rdev) + self.minor = os.minor(st.st_rdev) + self.attached = True + + return True + + def Assemble(self): + """Assemble the device. + + """ + pass + + def Shutdown(self): + """Shutdown the device. + + """ + pass + + def Open(self, force=False): + """Make the device ready for I/O. + + """ + pass + + def Close(self): + """Notifies that the device will no longer be used for I/O. + + """ + pass + + def Grow(self, amount, dryrun): + """Grow the logical volume. + + """ + _ThrowError("Grow is not supported for PersistentBlockDev storage") DEV_MAP = { constants.LD_LV: LogicalVolume, constants.LD_DRBD8: DRBD8, - constants.LD_FILE: FileStorage, + constants.LD_BLOCKDEV: PersistentBlockDevice, } +if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE: + DEV_MAP[constants.LD_FILE] = FileStorage + + +def _VerifyDiskType(dev_type): + if dev_type not in DEV_MAP: + raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) + -def FindDevice(dev_type, unique_id, children, size): +def FindDevice(disk, children): """Search for an existing, assembled device. This will succeed only if the device exists and is assembled, but it does not do any actions in order to activate the device. + @type disk: L{objects.Disk} + @param disk: the disk object to find + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + """ - 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, size) + _VerifyDiskType(disk.dev_type) + dev_params = objects.FillDict(constants.DISK_LD_DEFAULTS[disk.dev_type], + disk.params) + device = DEV_MAP[disk.dev_type](disk.physical_id, children, disk.size, + dev_params) if not device.attached: return None return device -def Assemble(dev_type, unique_id, children, size): +def Assemble(disk, children): """Try to attach or assemble an existing device. This will attach to assemble the device, as needed, to bring it fully up. It must be safe to run on already-assembled devices. + @type disk: L{objects.Disk} + @param disk: the disk object to assemble + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + """ - 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, size) + _VerifyDiskType(disk.dev_type) + dev_params = objects.FillDict(constants.DISK_LD_DEFAULTS[disk.dev_type], + disk.params) + device = DEV_MAP[disk.dev_type](disk.physical_id, children, disk.size, + dev_params) device.Assemble() return device -def Create(dev_type, unique_id, children, size): +def Create(disk, children): """Create a device. + @type disk: L{objects.Disk} + @param disk: the disk object to create + @type children: list of L{bdev.BlockDev} + @param children: the list of block devices that are children of the device + represented by the disk parameter + """ - if dev_type not in DEV_MAP: - raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) - device = DEV_MAP[dev_type].Create(unique_id, children, size) + _VerifyDiskType(disk.dev_type) + dev_params = objects.FillDict(constants.DISK_LD_DEFAULTS[disk.dev_type], + disk.params) + device = DEV_MAP[disk.dev_type].Create(disk.physical_id, children, disk.size, + dev_params) return device