root / lib / block / drbd.py @ 27c7d9c3
History | View | Annotate | Download (36.2 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2010, 2011, 2012, 2013 Google Inc.
|
5 |
#
|
6 |
# This program is free software; you can redistribute it and/or modify
|
7 |
# it under the terms of the GNU General Public License as published by
|
8 |
# the Free Software Foundation; either version 2 of the License, or
|
9 |
# (at your option) any later version.
|
10 |
#
|
11 |
# This program is distributed in the hope that it will be useful, but
|
12 |
# WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 |
# General Public License for more details.
|
15 |
#
|
16 |
# You should have received a copy of the GNU General Public License
|
17 |
# along with this program; if not, write to the Free Software
|
18 |
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
|
19 |
# 02110-1301, USA.
|
20 |
|
21 |
|
22 |
"""DRBD block device related functionality"""
|
23 |
|
24 |
import errno |
25 |
import logging |
26 |
import shlex |
27 |
import time |
28 |
|
29 |
from ganeti import constants |
30 |
from ganeti import utils |
31 |
from ganeti import errors |
32 |
from ganeti import netutils |
33 |
from ganeti import objects |
34 |
from ganeti.block import base |
35 |
from ganeti.block.drbd_info import DRBD8Info |
36 |
from ganeti.block.drbd_info import DRBD83ShowInfo |
37 |
from ganeti.block.drbd_info import DRBD84ShowInfo |
38 |
|
39 |
|
40 |
# Size of reads in _CanReadDevice
|
41 |
|
42 |
_DEVICE_READ_SIZE = 128 * 1024 |
43 |
|
44 |
|
45 |
class DRBD8(base.BlockDev): |
46 |
"""DRBD v8.x block device.
|
47 |
|
48 |
This implements the local host part of the DRBD device, i.e. it
|
49 |
doesn't do anything to the supposed peer. If you need a fully
|
50 |
connected DRBD pair, you need to use this class on both hosts.
|
51 |
|
52 |
The unique_id for the drbd device is a (local_ip, local_port,
|
53 |
remote_ip, remote_port, local_minor, secret) tuple, and it must have
|
54 |
two children: the data device and the meta_device. The meta device
|
55 |
is checked for valid size and is zeroed on create.
|
56 |
|
57 |
"""
|
58 |
_DRBD_MAJOR = 147
|
59 |
|
60 |
_USERMODE_HELPER_FILE = "/sys/module/drbd/parameters/usermode_helper"
|
61 |
|
62 |
_MAX_MINORS = 255
|
63 |
|
64 |
# timeout constants
|
65 |
_NET_RECONFIG_TIMEOUT = 60
|
66 |
|
67 |
# command line options for barriers
|
68 |
_DISABLE_DISK_OPTION = "--no-disk-barrier" # -a |
69 |
_DISABLE_DRAIN_OPTION = "--no-disk-drain" # -D |
70 |
_DISABLE_FLUSH_OPTION = "--no-disk-flushes" # -i |
71 |
_DISABLE_META_FLUSH_OPTION = "--no-md-flushes" # -m |
72 |
|
73 |
def __init__(self, unique_id, children, size, params): |
74 |
if children and children.count(None) > 0: |
75 |
children = [] |
76 |
if len(children) not in (0, 2): |
77 |
raise ValueError("Invalid configuration data %s" % str(children)) |
78 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6: |
79 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
80 |
(self._lhost, self._lport, |
81 |
self._rhost, self._rport, |
82 |
self._aminor, self._secret) = unique_id |
83 |
if children:
|
84 |
if not _CanReadDevice(children[1].dev_path): |
85 |
logging.info("drbd%s: Ignoring unreadable meta device", self._aminor) |
86 |
children = [] |
87 |
super(DRBD8, self).__init__(unique_id, children, size, params) |
88 |
self.major = self._DRBD_MAJOR |
89 |
|
90 |
drbd_info = DRBD8Info.CreateFromFile() |
91 |
version = drbd_info.GetVersion() |
92 |
if version["k_major"] != 8: |
93 |
base.ThrowError("Mismatch in DRBD kernel version and requested ganeti"
|
94 |
" usage: kernel is %s.%s, ganeti wants 8.x",
|
95 |
version["k_major"], version["k_minor"]) |
96 |
|
97 |
if version["k_minor"] <= 3: |
98 |
self._show_info_cls = DRBD83ShowInfo
|
99 |
else:
|
100 |
self._show_info_cls = DRBD84ShowInfo
|
101 |
|
102 |
if (self._lhost is not None and self._lhost == self._rhost and |
103 |
self._lport == self._rport): |
104 |
raise ValueError("Invalid configuration data, same local/remote %s" % |
105 |
(unique_id,)) |
106 |
self.Attach()
|
107 |
|
108 |
@staticmethod
|
109 |
def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): |
110 |
"""Returns DRBD usermode_helper currently set.
|
111 |
|
112 |
"""
|
113 |
try:
|
114 |
helper = utils.ReadFile(filename).splitlines()[0]
|
115 |
except EnvironmentError, err: |
116 |
if err.errno == errno.ENOENT:
|
117 |
base.ThrowError("The file %s cannot be opened, check if the module"
|
118 |
" is loaded (%s)", filename, str(err)) |
119 |
else:
|
120 |
base.ThrowError("Can't read DRBD helper file %s: %s",
|
121 |
filename, str(err))
|
122 |
if not helper: |
123 |
base.ThrowError("Can't read any data from %s", filename)
|
124 |
return helper
|
125 |
|
126 |
@staticmethod
|
127 |
def _DevPath(minor): |
128 |
"""Return the path to a drbd device for a given minor.
|
129 |
|
130 |
"""
|
131 |
return "/dev/drbd%d" % minor |
132 |
|
133 |
@classmethod
|
134 |
def GetUsedDevs(cls): |
135 |
"""Compute the list of used DRBD devices.
|
136 |
|
137 |
"""
|
138 |
drbd_info = DRBD8Info.CreateFromFile() |
139 |
return filter(lambda m: not drbd_info.GetMinorStatus(m).is_unconfigured, |
140 |
drbd_info.GetMinors()) |
141 |
|
142 |
def _SetFromMinor(self, minor): |
143 |
"""Set our parameters based on the given minor.
|
144 |
|
145 |
This sets our minor variable and our dev_path.
|
146 |
|
147 |
"""
|
148 |
if minor is None: |
149 |
self.minor = self.dev_path = None |
150 |
self.attached = False |
151 |
else:
|
152 |
self.minor = minor
|
153 |
self.dev_path = self._DevPath(minor) |
154 |
self.attached = True |
155 |
|
156 |
@staticmethod
|
157 |
def _CheckMetaSize(meta_device): |
158 |
"""Check if the given meta device looks like a valid one.
|
159 |
|
160 |
This currently only checks the size, which must be around
|
161 |
128MiB.
|
162 |
|
163 |
"""
|
164 |
result = utils.RunCmd(["blockdev", "--getsize", meta_device]) |
165 |
if result.failed:
|
166 |
base.ThrowError("Failed to get device size: %s - %s",
|
167 |
result.fail_reason, result.output) |
168 |
try:
|
169 |
sectors = int(result.stdout)
|
170 |
except (TypeError, ValueError): |
171 |
base.ThrowError("Invalid output from blockdev: '%s'", result.stdout)
|
172 |
num_bytes = sectors * 512
|
173 |
if num_bytes < 128 * 1024 * 1024: # less than 128MiB |
174 |
base.ThrowError("Meta device too small (%.2fMib)",
|
175 |
(num_bytes / 1024 / 1024)) |
176 |
# the maximum *valid* size of the meta device when living on top
|
177 |
# of LVM is hard to compute: it depends on the number of stripes
|
178 |
# and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB
|
179 |
# (normal size), but an eight-stripe 128MB PE will result in a 1GB
|
180 |
# size meta device; as such, we restrict it to 1GB (a little bit
|
181 |
# too generous, but making assumptions about PE size is hard)
|
182 |
if num_bytes > 1024 * 1024 * 1024: |
183 |
base.ThrowError("Meta device too big (%.2fMiB)",
|
184 |
(num_bytes / 1024 / 1024)) |
185 |
|
186 |
@classmethod
|
187 |
def _InitMeta(cls, minor, dev_path): |
188 |
"""Initialize a meta device.
|
189 |
|
190 |
This will not work if the given minor is in use.
|
191 |
|
192 |
"""
|
193 |
# Zero the metadata first, in order to make sure drbdmeta doesn't
|
194 |
# try to auto-detect existing filesystems or similar (see
|
195 |
# http://code.google.com/p/ganeti/issues/detail?id=182); we only
|
196 |
# care about the first 128MB of data in the device, even though it
|
197 |
# can be bigger
|
198 |
result = utils.RunCmd([constants.DD_CMD, |
199 |
"if=/dev/zero", "of=%s" % dev_path, |
200 |
"bs=1048576", "count=128", "oflag=direct"]) |
201 |
if result.failed:
|
202 |
base.ThrowError("Can't wipe the meta device: %s", result.output)
|
203 |
|
204 |
result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), |
205 |
"v08", dev_path, "0", "create-md"]) |
206 |
if result.failed:
|
207 |
base.ThrowError("Can't initialize meta device: %s", result.output)
|
208 |
|
209 |
def _FindUnusedMinor(self): |
210 |
"""Find an unused DRBD device.
|
211 |
|
212 |
This is specific to 8.x as the minors are allocated dynamically,
|
213 |
so non-existing numbers up to a max minor count are actually free.
|
214 |
|
215 |
"""
|
216 |
|
217 |
highest = None
|
218 |
drbd_info = DRBD8Info.CreateFromFile() |
219 |
for minor in drbd_info.GetMinors(): |
220 |
status = drbd_info.GetMinorStatus(minor) |
221 |
if not status.is_in_use: |
222 |
return minor
|
223 |
highest = max(highest, minor)
|
224 |
|
225 |
if highest is None: # there are no minors in use at all |
226 |
return 0 |
227 |
if highest >= self._MAX_MINORS: |
228 |
logging.error("Error: no free drbd minors!")
|
229 |
raise errors.BlockDeviceError("Can't find a free DRBD minor") |
230 |
|
231 |
return highest + 1 |
232 |
|
233 |
@classmethod
|
234 |
def _GetShowData(cls, minor): |
235 |
"""Return the `drbdsetup show` data for a minor.
|
236 |
|
237 |
"""
|
238 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) |
239 |
if result.failed:
|
240 |
logging.error("Can't display the drbd config: %s - %s",
|
241 |
result.fail_reason, result.output) |
242 |
return None |
243 |
return result.stdout
|
244 |
|
245 |
def _GetShowInfo(self, minor): |
246 |
return self._show_info_cls.GetDevInfo(self._GetShowData(minor)) |
247 |
|
248 |
def _MatchesLocal(self, info): |
249 |
"""Test if our local config matches with an existing device.
|
250 |
|
251 |
The parameter should be as returned from `_GetDevInfo()`. This
|
252 |
method tests if our local backing device is the same as the one in
|
253 |
the info parameter, in effect testing if we look like the given
|
254 |
device.
|
255 |
|
256 |
"""
|
257 |
if self._children: |
258 |
backend, meta = self._children
|
259 |
else:
|
260 |
backend = meta = None
|
261 |
|
262 |
if backend is not None: |
263 |
retval = ("local_dev" in info and info["local_dev"] == backend.dev_path) |
264 |
else:
|
265 |
retval = ("local_dev" not in info) |
266 |
|
267 |
if meta is not None: |
268 |
retval = retval and ("meta_dev" in info and |
269 |
info["meta_dev"] == meta.dev_path)
|
270 |
retval = retval and ("meta_index" in info and |
271 |
info["meta_index"] == 0) |
272 |
else:
|
273 |
retval = retval and ("meta_dev" not in info and |
274 |
"meta_index" not in info) |
275 |
return retval
|
276 |
|
277 |
def _MatchesNet(self, info): |
278 |
"""Test if our network config matches with an existing device.
|
279 |
|
280 |
The parameter should be as returned from `_GetDevInfo()`. This
|
281 |
method tests if our network configuration is the same as the one
|
282 |
in the info parameter, in effect testing if we look like the given
|
283 |
device.
|
284 |
|
285 |
"""
|
286 |
if (((self._lhost is None and not ("local_addr" in info)) and |
287 |
(self._rhost is None and not ("remote_addr" in info)))): |
288 |
return True |
289 |
|
290 |
if self._lhost is None: |
291 |
return False |
292 |
|
293 |
if not ("local_addr" in info and |
294 |
"remote_addr" in info): |
295 |
return False |
296 |
|
297 |
retval = (info["local_addr"] == (self._lhost, self._lport)) |
298 |
retval = (retval and
|
299 |
info["remote_addr"] == (self._rhost, self._rport)) |
300 |
return retval
|
301 |
|
302 |
def _AssembleLocal(self, minor, backend, meta, size): |
303 |
"""Configure the local part of a DRBD device.
|
304 |
|
305 |
"""
|
306 |
args = ["drbdsetup", self._DevPath(minor), "disk", |
307 |
backend, meta, "0",
|
308 |
"-e", "detach", |
309 |
"--create-device"]
|
310 |
if size:
|
311 |
args.extend(["-d", "%sm" % size]) |
312 |
|
313 |
drbd_info = DRBD8Info.CreateFromFile() |
314 |
version = drbd_info.GetVersion() |
315 |
vmaj = version["k_major"]
|
316 |
vmin = version["k_minor"]
|
317 |
vrel = version["k_point"]
|
318 |
|
319 |
barrier_args = \ |
320 |
self._ComputeDiskBarrierArgs(vmaj, vmin, vrel,
|
321 |
self.params[constants.LDP_BARRIERS],
|
322 |
self.params[constants.LDP_NO_META_FLUSH])
|
323 |
args.extend(barrier_args) |
324 |
|
325 |
if self.params[constants.LDP_DISK_CUSTOM]: |
326 |
args.extend(shlex.split(self.params[constants.LDP_DISK_CUSTOM]))
|
327 |
|
328 |
result = utils.RunCmd(args) |
329 |
if result.failed:
|
330 |
base.ThrowError("drbd%d: can't attach local disk: %s",
|
331 |
minor, result.output) |
332 |
|
333 |
@classmethod
|
334 |
def _ComputeDiskBarrierArgs(cls, vmaj, vmin, vrel, disabled_barriers, |
335 |
disable_meta_flush): |
336 |
"""Compute the DRBD command line parameters for disk barriers
|
337 |
|
338 |
Returns a list of the disk barrier parameters as requested via the
|
339 |
disabled_barriers and disable_meta_flush arguments, and according to the
|
340 |
supported ones in the DRBD version vmaj.vmin.vrel
|
341 |
|
342 |
If the desired option is unsupported, raises errors.BlockDeviceError.
|
343 |
|
344 |
"""
|
345 |
disabled_barriers_set = frozenset(disabled_barriers)
|
346 |
if not disabled_barriers_set in constants.DRBD_VALID_BARRIER_OPT: |
347 |
raise errors.BlockDeviceError("%s is not a valid option set for DRBD" |
348 |
" barriers" % disabled_barriers)
|
349 |
|
350 |
args = [] |
351 |
|
352 |
# The following code assumes DRBD 8.x, with x < 4 and x != 1 (DRBD 8.1.x
|
353 |
# does not exist)
|
354 |
if not vmaj == 8 and vmin in (0, 2, 3): |
355 |
raise errors.BlockDeviceError("Unsupported DRBD version: %d.%d.%d" % |
356 |
(vmaj, vmin, vrel)) |
357 |
|
358 |
def _AppendOrRaise(option, min_version): |
359 |
"""Helper for DRBD options"""
|
360 |
if min_version is not None and vrel >= min_version: |
361 |
args.append(option) |
362 |
else:
|
363 |
raise errors.BlockDeviceError("Could not use the option %s as the" |
364 |
" DRBD version %d.%d.%d does not support"
|
365 |
" it." % (option, vmaj, vmin, vrel))
|
366 |
|
367 |
# the minimum version for each feature is encoded via pairs of (minor
|
368 |
# version -> x) where x is version in which support for the option was
|
369 |
# introduced.
|
370 |
meta_flush_supported = disk_flush_supported = { |
371 |
0: 12, |
372 |
2: 7, |
373 |
3: 0, |
374 |
} |
375 |
|
376 |
disk_drain_supported = { |
377 |
2: 7, |
378 |
3: 0, |
379 |
} |
380 |
|
381 |
disk_barriers_supported = { |
382 |
3: 0, |
383 |
} |
384 |
|
385 |
# meta flushes
|
386 |
if disable_meta_flush:
|
387 |
_AppendOrRaise(cls._DISABLE_META_FLUSH_OPTION, |
388 |
meta_flush_supported.get(vmin, None))
|
389 |
|
390 |
# disk flushes
|
391 |
if constants.DRBD_B_DISK_FLUSH in disabled_barriers_set: |
392 |
_AppendOrRaise(cls._DISABLE_FLUSH_OPTION, |
393 |
disk_flush_supported.get(vmin, None))
|
394 |
|
395 |
# disk drain
|
396 |
if constants.DRBD_B_DISK_DRAIN in disabled_barriers_set: |
397 |
_AppendOrRaise(cls._DISABLE_DRAIN_OPTION, |
398 |
disk_drain_supported.get(vmin, None))
|
399 |
|
400 |
# disk barriers
|
401 |
if constants.DRBD_B_DISK_BARRIERS in disabled_barriers_set: |
402 |
_AppendOrRaise(cls._DISABLE_DISK_OPTION, |
403 |
disk_barriers_supported.get(vmin, None))
|
404 |
|
405 |
return args
|
406 |
|
407 |
def _AssembleNet(self, minor, net_info, protocol, |
408 |
dual_pri=False, hmac=None, secret=None): |
409 |
"""Configure the network part of the device.
|
410 |
|
411 |
"""
|
412 |
lhost, lport, rhost, rport = net_info |
413 |
if None in net_info: |
414 |
# we don't want network connection and actually want to make
|
415 |
# sure its shutdown
|
416 |
self._ShutdownNet(minor)
|
417 |
return
|
418 |
|
419 |
# Workaround for a race condition. When DRBD is doing its dance to
|
420 |
# establish a connection with its peer, it also sends the
|
421 |
# synchronization speed over the wire. In some cases setting the
|
422 |
# sync speed only after setting up both sides can race with DRBD
|
423 |
# connecting, hence we set it here before telling DRBD anything
|
424 |
# about its peer.
|
425 |
sync_errors = self._SetMinorSyncParams(minor, self.params) |
426 |
if sync_errors:
|
427 |
base.ThrowError("drbd%d: can't set the synchronization parameters: %s" %
|
428 |
(minor, utils.CommaJoin(sync_errors))) |
429 |
|
430 |
if netutils.IP6Address.IsValid(lhost):
|
431 |
if not netutils.IP6Address.IsValid(rhost): |
432 |
base.ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
433 |
(minor, lhost, rhost)) |
434 |
family = "ipv6"
|
435 |
elif netutils.IP4Address.IsValid(lhost):
|
436 |
if not netutils.IP4Address.IsValid(rhost): |
437 |
base.ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
438 |
(minor, lhost, rhost)) |
439 |
family = "ipv4"
|
440 |
else:
|
441 |
base.ThrowError("drbd%d: Invalid ip %s" % (minor, lhost))
|
442 |
|
443 |
args = ["drbdsetup", self._DevPath(minor), "net", |
444 |
"%s:%s:%s" % (family, lhost, lport),
|
445 |
"%s:%s:%s" % (family, rhost, rport), protocol,
|
446 |
"-A", "discard-zero-changes", |
447 |
"-B", "consensus", |
448 |
"--create-device",
|
449 |
] |
450 |
if dual_pri:
|
451 |
args.append("-m")
|
452 |
if hmac and secret: |
453 |
args.extend(["-a", hmac, "-x", secret]) |
454 |
|
455 |
if self.params[constants.LDP_NET_CUSTOM]: |
456 |
args.extend(shlex.split(self.params[constants.LDP_NET_CUSTOM]))
|
457 |
|
458 |
result = utils.RunCmd(args) |
459 |
if result.failed:
|
460 |
base.ThrowError("drbd%d: can't setup network: %s - %s",
|
461 |
minor, result.fail_reason, result.output) |
462 |
|
463 |
def _CheckNetworkConfig(): |
464 |
info = self._GetShowInfo(minor)
|
465 |
if not "local_addr" in info or not "remote_addr" in info: |
466 |
raise utils.RetryAgain()
|
467 |
|
468 |
if (info["local_addr"] != (lhost, lport) or |
469 |
info["remote_addr"] != (rhost, rport)):
|
470 |
raise utils.RetryAgain()
|
471 |
|
472 |
try:
|
473 |
utils.Retry(_CheckNetworkConfig, 1.0, 10.0) |
474 |
except utils.RetryTimeout:
|
475 |
base.ThrowError("drbd%d: timeout while configuring network", minor)
|
476 |
|
477 |
def AddChildren(self, devices): |
478 |
"""Add a disk to the DRBD device.
|
479 |
|
480 |
"""
|
481 |
if self.minor is None: |
482 |
base.ThrowError("drbd%d: can't attach to dbrd8 during AddChildren",
|
483 |
self._aminor)
|
484 |
if len(devices) != 2: |
485 |
base.ThrowError("drbd%d: need two devices for AddChildren", self.minor) |
486 |
info = self._GetShowInfo(self.minor) |
487 |
if "local_dev" in info: |
488 |
base.ThrowError("drbd%d: already attached to a local disk", self.minor) |
489 |
backend, meta = devices |
490 |
if backend.dev_path is None or meta.dev_path is None: |
491 |
base.ThrowError("drbd%d: children not ready during AddChildren",
|
492 |
self.minor)
|
493 |
backend.Open() |
494 |
meta.Open() |
495 |
self._CheckMetaSize(meta.dev_path)
|
496 |
self._InitMeta(self._FindUnusedMinor(), meta.dev_path) |
497 |
|
498 |
self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size) |
499 |
self._children = devices
|
500 |
|
501 |
def RemoveChildren(self, devices): |
502 |
"""Detach the drbd device from local storage.
|
503 |
|
504 |
"""
|
505 |
if self.minor is None: |
506 |
base.ThrowError("drbd%d: can't attach to drbd8 during RemoveChildren",
|
507 |
self._aminor)
|
508 |
# early return if we don't actually have backing storage
|
509 |
info = self._GetShowInfo(self.minor) |
510 |
if "local_dev" not in info: |
511 |
return
|
512 |
if len(self._children) != 2: |
513 |
base.ThrowError("drbd%d: we don't have two children: %s", self.minor, |
514 |
self._children)
|
515 |
if self._children.count(None) == 2: # we don't actually have children :) |
516 |
logging.warning("drbd%d: requested detach while detached", self.minor) |
517 |
return
|
518 |
if len(devices) != 2: |
519 |
base.ThrowError("drbd%d: we need two children in RemoveChildren",
|
520 |
self.minor)
|
521 |
for child, dev in zip(self._children, devices): |
522 |
if dev != child.dev_path:
|
523 |
base.ThrowError("drbd%d: mismatch in local storage (%s != %s) in"
|
524 |
" RemoveChildren", self.minor, dev, child.dev_path) |
525 |
|
526 |
self._ShutdownLocal(self.minor) |
527 |
self._children = []
|
528 |
|
529 |
def _SetMinorSyncParams(self, minor, params): |
530 |
"""Set the parameters of the DRBD syncer.
|
531 |
|
532 |
This is the low-level implementation.
|
533 |
|
534 |
@type minor: int
|
535 |
@param minor: the drbd minor whose settings we change
|
536 |
@type params: dict
|
537 |
@param params: LD level disk parameters related to the synchronization
|
538 |
@rtype: list
|
539 |
@return: a list of error messages
|
540 |
|
541 |
"""
|
542 |
|
543 |
args = ["drbdsetup", self._DevPath(minor), "syncer"] |
544 |
if params[constants.LDP_DYNAMIC_RESYNC]:
|
545 |
drbd_info = DRBD8Info.CreateFromFile() |
546 |
version = drbd_info.GetVersion() |
547 |
vmin = version["k_minor"]
|
548 |
vrel = version["k_point"]
|
549 |
|
550 |
# By definition we are using 8.x, so just check the rest of the version
|
551 |
# number
|
552 |
if vmin != 3 or vrel < 9: |
553 |
msg = ("The current DRBD version (8.%d.%d) does not support the "
|
554 |
"dynamic resync speed controller" % (vmin, vrel))
|
555 |
logging.error(msg) |
556 |
return [msg]
|
557 |
|
558 |
if params[constants.LDP_PLAN_AHEAD] == 0: |
559 |
msg = ("A value of 0 for c-plan-ahead disables the dynamic sync speed"
|
560 |
" controller at DRBD level. If you want to disable it, please"
|
561 |
" set the dynamic-resync disk parameter to False.")
|
562 |
logging.error(msg) |
563 |
return [msg]
|
564 |
|
565 |
# add the c-* parameters to args
|
566 |
args.extend(["--c-plan-ahead", params[constants.LDP_PLAN_AHEAD],
|
567 |
"--c-fill-target", params[constants.LDP_FILL_TARGET],
|
568 |
"--c-delay-target", params[constants.LDP_DELAY_TARGET],
|
569 |
"--c-max-rate", params[constants.LDP_MAX_RATE],
|
570 |
"--c-min-rate", params[constants.LDP_MIN_RATE],
|
571 |
]) |
572 |
|
573 |
else:
|
574 |
args.extend(["-r", "%d" % params[constants.LDP_RESYNC_RATE]]) |
575 |
|
576 |
args.append("--create-device")
|
577 |
result = utils.RunCmd(args) |
578 |
if result.failed:
|
579 |
msg = ("Can't change syncer rate: %s - %s" %
|
580 |
(result.fail_reason, result.output)) |
581 |
logging.error(msg) |
582 |
return [msg]
|
583 |
|
584 |
return []
|
585 |
|
586 |
def SetSyncParams(self, params): |
587 |
"""Set the synchronization parameters of the DRBD syncer.
|
588 |
|
589 |
@type params: dict
|
590 |
@param params: LD level disk parameters related to the synchronization
|
591 |
@rtype: list
|
592 |
@return: a list of error messages, emitted both by the current node and by
|
593 |
children. An empty list means no errors
|
594 |
|
595 |
"""
|
596 |
if self.minor is None: |
597 |
err = "Not attached during SetSyncParams"
|
598 |
logging.info(err) |
599 |
return [err]
|
600 |
|
601 |
children_result = super(DRBD8, self).SetSyncParams(params) |
602 |
children_result.extend(self._SetMinorSyncParams(self.minor, params)) |
603 |
return children_result
|
604 |
|
605 |
def PauseResumeSync(self, pause): |
606 |
"""Pauses or resumes the sync of a DRBD device.
|
607 |
|
608 |
@param pause: Wether to pause or resume
|
609 |
@return: the success of the operation
|
610 |
|
611 |
"""
|
612 |
if self.minor is None: |
613 |
logging.info("Not attached during PauseSync")
|
614 |
return False |
615 |
|
616 |
children_result = super(DRBD8, self).PauseResumeSync(pause) |
617 |
|
618 |
if pause:
|
619 |
cmd = "pause-sync"
|
620 |
else:
|
621 |
cmd = "resume-sync"
|
622 |
|
623 |
result = utils.RunCmd(["drbdsetup", self.dev_path, cmd]) |
624 |
if result.failed:
|
625 |
logging.error("Can't %s: %s - %s", cmd,
|
626 |
result.fail_reason, result.output) |
627 |
return not result.failed and children_result |
628 |
|
629 |
def GetProcStatus(self): |
630 |
"""Return device data from /proc.
|
631 |
|
632 |
"""
|
633 |
if self.minor is None: |
634 |
base.ThrowError("drbd%d: GetStats() called while not attached",
|
635 |
self._aminor)
|
636 |
drbd_info = DRBD8Info.CreateFromFile() |
637 |
if not drbd_info.HasMinorStatus(self.minor): |
638 |
base.ThrowError("drbd%d: can't find myself in /proc", self.minor) |
639 |
return drbd_info.GetMinorStatus(self.minor) |
640 |
|
641 |
def GetSyncStatus(self): |
642 |
"""Returns the sync status of the device.
|
643 |
|
644 |
|
645 |
If sync_percent is None, it means all is ok
|
646 |
If estimated_time is None, it means we can't estimate
|
647 |
the time needed, otherwise it's the time left in seconds.
|
648 |
|
649 |
|
650 |
We set the is_degraded parameter to True on two conditions:
|
651 |
network not connected or local disk missing.
|
652 |
|
653 |
We compute the ldisk parameter based on whether we have a local
|
654 |
disk or not.
|
655 |
|
656 |
@rtype: objects.BlockDevStatus
|
657 |
|
658 |
"""
|
659 |
if self.minor is None and not self.Attach(): |
660 |
base.ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) |
661 |
|
662 |
stats = self.GetProcStatus()
|
663 |
is_degraded = not stats.is_connected or not stats.is_disk_uptodate |
664 |
|
665 |
if stats.is_disk_uptodate:
|
666 |
ldisk_status = constants.LDS_OKAY |
667 |
elif stats.is_diskless:
|
668 |
ldisk_status = constants.LDS_FAULTY |
669 |
else:
|
670 |
ldisk_status = constants.LDS_UNKNOWN |
671 |
|
672 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
673 |
major=self.major,
|
674 |
minor=self.minor,
|
675 |
sync_percent=stats.sync_percent, |
676 |
estimated_time=stats.est_time, |
677 |
is_degraded=is_degraded, |
678 |
ldisk_status=ldisk_status) |
679 |
|
680 |
def Open(self, force=False): |
681 |
"""Make the local state primary.
|
682 |
|
683 |
If the 'force' parameter is given, the '-o' option is passed to
|
684 |
drbdsetup. Since this is a potentially dangerous operation, the
|
685 |
force flag should be only given after creation, when it actually
|
686 |
is mandatory.
|
687 |
|
688 |
"""
|
689 |
if self.minor is None and not self.Attach(): |
690 |
logging.error("DRBD cannot attach to a device during open")
|
691 |
return False |
692 |
cmd = ["drbdsetup", self.dev_path, "primary"] |
693 |
if force:
|
694 |
cmd.append("-o")
|
695 |
result = utils.RunCmd(cmd) |
696 |
if result.failed:
|
697 |
base.ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, |
698 |
result.output) |
699 |
|
700 |
def Close(self): |
701 |
"""Make the local state secondary.
|
702 |
|
703 |
This will, of course, fail if the device is in use.
|
704 |
|
705 |
"""
|
706 |
if self.minor is None and not self.Attach(): |
707 |
base.ThrowError("drbd%d: can't Attach() in Close()", self._aminor) |
708 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "secondary"]) |
709 |
if result.failed:
|
710 |
base.ThrowError("drbd%d: can't switch drbd device to secondary: %s",
|
711 |
self.minor, result.output)
|
712 |
|
713 |
def DisconnectNet(self): |
714 |
"""Removes network configuration.
|
715 |
|
716 |
This method shutdowns the network side of the device.
|
717 |
|
718 |
The method will wait up to a hardcoded timeout for the device to
|
719 |
go into standalone after the 'disconnect' command before
|
720 |
re-configuring it, as sometimes it takes a while for the
|
721 |
disconnect to actually propagate and thus we might issue a 'net'
|
722 |
command while the device is still connected. If the device will
|
723 |
still be attached to the network and we time out, we raise an
|
724 |
exception.
|
725 |
|
726 |
"""
|
727 |
if self.minor is None: |
728 |
base.ThrowError("drbd%d: disk not attached in re-attach net",
|
729 |
self._aminor)
|
730 |
|
731 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
732 |
base.ThrowError("drbd%d: DRBD disk missing network info in"
|
733 |
" DisconnectNet()", self.minor) |
734 |
|
735 |
class _DisconnectStatus: |
736 |
def __init__(self, ever_disconnected): |
737 |
self.ever_disconnected = ever_disconnected
|
738 |
|
739 |
dstatus = _DisconnectStatus(base.IgnoreError(self._ShutdownNet, self.minor)) |
740 |
|
741 |
def _WaitForDisconnect(): |
742 |
if self.GetProcStatus().is_standalone: |
743 |
return
|
744 |
|
745 |
# retry the disconnect, it seems possible that due to a well-time
|
746 |
# disconnect on the peer, my disconnect command might be ignored and
|
747 |
# forgotten
|
748 |
dstatus.ever_disconnected = \ |
749 |
base.IgnoreError(self._ShutdownNet, self.minor) or \ |
750 |
dstatus.ever_disconnected |
751 |
|
752 |
raise utils.RetryAgain()
|
753 |
|
754 |
# Keep start time
|
755 |
start_time = time.time() |
756 |
|
757 |
try:
|
758 |
# Start delay at 100 milliseconds and grow up to 2 seconds
|
759 |
utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0), |
760 |
self._NET_RECONFIG_TIMEOUT)
|
761 |
except utils.RetryTimeout:
|
762 |
if dstatus.ever_disconnected:
|
763 |
msg = ("drbd%d: device did not react to the"
|
764 |
" 'disconnect' command in a timely manner")
|
765 |
else:
|
766 |
msg = "drbd%d: can't shutdown network, even after multiple retries"
|
767 |
|
768 |
base.ThrowError(msg, self.minor)
|
769 |
|
770 |
reconfig_time = time.time() - start_time |
771 |
if reconfig_time > (self._NET_RECONFIG_TIMEOUT * 0.25): |
772 |
logging.info("drbd%d: DisconnectNet: detach took %.3f seconds",
|
773 |
self.minor, reconfig_time)
|
774 |
|
775 |
def AttachNet(self, multimaster): |
776 |
"""Reconnects the network.
|
777 |
|
778 |
This method connects the network side of the device with a
|
779 |
specified multi-master flag. The device needs to be 'Standalone'
|
780 |
but have valid network configuration data.
|
781 |
|
782 |
Args:
|
783 |
- multimaster: init the network in dual-primary mode
|
784 |
|
785 |
"""
|
786 |
if self.minor is None: |
787 |
base.ThrowError("drbd%d: device not attached in AttachNet", self._aminor) |
788 |
|
789 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
790 |
base.ThrowError("drbd%d: missing network info in AttachNet()", self.minor) |
791 |
|
792 |
status = self.GetProcStatus()
|
793 |
|
794 |
if not status.is_standalone: |
795 |
base.ThrowError("drbd%d: device is not standalone in AttachNet",
|
796 |
self.minor)
|
797 |
|
798 |
self._AssembleNet(self.minor, |
799 |
(self._lhost, self._lport, self._rhost, self._rport), |
800 |
constants.DRBD_NET_PROTOCOL, dual_pri=multimaster, |
801 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
802 |
|
803 |
def Attach(self): |
804 |
"""Check if our minor is configured.
|
805 |
|
806 |
This doesn't do any device configurations - it only checks if the
|
807 |
minor is in a state different from Unconfigured.
|
808 |
|
809 |
Note that this function will not change the state of the system in
|
810 |
any way (except in case of side-effects caused by reading from
|
811 |
/proc).
|
812 |
|
813 |
"""
|
814 |
used_devs = self.GetUsedDevs()
|
815 |
if self._aminor in used_devs: |
816 |
minor = self._aminor
|
817 |
else:
|
818 |
minor = None
|
819 |
|
820 |
self._SetFromMinor(minor)
|
821 |
return minor is not None |
822 |
|
823 |
def Assemble(self): |
824 |
"""Assemble the drbd.
|
825 |
|
826 |
Method:
|
827 |
- if we have a configured device, we try to ensure that it matches
|
828 |
our config
|
829 |
- if not, we create it from zero
|
830 |
- anyway, set the device parameters
|
831 |
|
832 |
"""
|
833 |
super(DRBD8, self).Assemble() |
834 |
|
835 |
self.Attach()
|
836 |
if self.minor is None: |
837 |
# local device completely unconfigured
|
838 |
self._FastAssemble()
|
839 |
else:
|
840 |
# we have to recheck the local and network status and try to fix
|
841 |
# the device
|
842 |
self._SlowAssemble()
|
843 |
|
844 |
sync_errors = self.SetSyncParams(self.params) |
845 |
if sync_errors:
|
846 |
base.ThrowError("drbd%d: can't set the synchronization parameters: %s" %
|
847 |
(self.minor, utils.CommaJoin(sync_errors)))
|
848 |
|
849 |
def _SlowAssemble(self): |
850 |
"""Assembles the DRBD device from a (partially) configured device.
|
851 |
|
852 |
In case of partially attached (local device matches but no network
|
853 |
setup), we perform the network attach. If successful, we re-test
|
854 |
the attach if can return success.
|
855 |
|
856 |
"""
|
857 |
# TODO: Rewrite to not use a for loop just because there is 'break'
|
858 |
# pylint: disable=W0631
|
859 |
net_data = (self._lhost, self._lport, self._rhost, self._rport) |
860 |
for minor in (self._aminor,): |
861 |
info = self._GetShowInfo(minor)
|
862 |
match_l = self._MatchesLocal(info)
|
863 |
match_r = self._MatchesNet(info)
|
864 |
|
865 |
if match_l and match_r: |
866 |
# everything matches
|
867 |
break
|
868 |
|
869 |
if match_l and not match_r and "local_addr" not in info: |
870 |
# disk matches, but not attached to network, attach and recheck
|
871 |
self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
|
872 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
873 |
if self._MatchesNet(self._GetShowInfo(minor)): |
874 |
break
|
875 |
else:
|
876 |
base.ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
877 |
" show' disagrees", minor)
|
878 |
|
879 |
if match_r and "local_dev" not in info: |
880 |
# no local disk, but network attached and it matches
|
881 |
self._AssembleLocal(minor, self._children[0].dev_path, |
882 |
self._children[1].dev_path, self.size) |
883 |
if self._MatchesNet(self._GetShowInfo(minor)): |
884 |
break
|
885 |
else:
|
886 |
base.ThrowError("drbd%d: disk attach successful, but 'drbdsetup"
|
887 |
" show' disagrees", minor)
|
888 |
|
889 |
# this case must be considered only if we actually have local
|
890 |
# storage, i.e. not in diskless mode, because all diskless
|
891 |
# devices are equal from the point of view of local
|
892 |
# configuration
|
893 |
if (match_l and "local_dev" in info and |
894 |
not match_r and "local_addr" in info): |
895 |
# strange case - the device network part points to somewhere
|
896 |
# else, even though its local storage is ours; as we own the
|
897 |
# drbd space, we try to disconnect from the remote peer and
|
898 |
# reconnect to our correct one
|
899 |
try:
|
900 |
self._ShutdownNet(minor)
|
901 |
except errors.BlockDeviceError, err:
|
902 |
base.ThrowError("drbd%d: device has correct local storage, wrong"
|
903 |
" remote peer and is unable to disconnect in order"
|
904 |
" to attach to the correct peer: %s", minor, str(err)) |
905 |
# note: _AssembleNet also handles the case when we don't want
|
906 |
# local storage (i.e. one or more of the _[lr](host|port) is
|
907 |
# None)
|
908 |
self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
|
909 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
910 |
if self._MatchesNet(self._GetShowInfo(minor)): |
911 |
break
|
912 |
else:
|
913 |
base.ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
914 |
" show' disagrees", minor)
|
915 |
|
916 |
else:
|
917 |
minor = None
|
918 |
|
919 |
self._SetFromMinor(minor)
|
920 |
if minor is None: |
921 |
base.ThrowError("drbd%d: cannot activate, unknown or unhandled reason",
|
922 |
self._aminor)
|
923 |
|
924 |
def _FastAssemble(self): |
925 |
"""Assemble the drbd device from zero.
|
926 |
|
927 |
This is run when in Assemble we detect our minor is unused.
|
928 |
|
929 |
"""
|
930 |
minor = self._aminor
|
931 |
if self._children and self._children[0] and self._children[1]: |
932 |
self._AssembleLocal(minor, self._children[0].dev_path, |
933 |
self._children[1].dev_path, self.size) |
934 |
if self._lhost and self._lport and self._rhost and self._rport: |
935 |
self._AssembleNet(minor,
|
936 |
(self._lhost, self._lport, self._rhost, self._rport), |
937 |
constants.DRBD_NET_PROTOCOL, |
938 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
939 |
self._SetFromMinor(minor)
|
940 |
|
941 |
@classmethod
|
942 |
def _ShutdownLocal(cls, minor): |
943 |
"""Detach from the local device.
|
944 |
|
945 |
I/Os will continue to be served from the remote device. If we
|
946 |
don't have a remote device, this operation will fail.
|
947 |
|
948 |
"""
|
949 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) |
950 |
if result.failed:
|
951 |
base.ThrowError("drbd%d: can't detach local disk: %s",
|
952 |
minor, result.output) |
953 |
|
954 |
@classmethod
|
955 |
def _ShutdownNet(cls, minor): |
956 |
"""Disconnect from the remote peer.
|
957 |
|
958 |
This fails if we don't have a local device.
|
959 |
|
960 |
"""
|
961 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) |
962 |
if result.failed:
|
963 |
base.ThrowError("drbd%d: can't shutdown network: %s",
|
964 |
minor, result.output) |
965 |
|
966 |
@classmethod
|
967 |
def _ShutdownAll(cls, minor): |
968 |
"""Deactivate the device.
|
969 |
|
970 |
This will, of course, fail if the device is in use.
|
971 |
|
972 |
"""
|
973 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) |
974 |
if result.failed:
|
975 |
base.ThrowError("drbd%d: can't shutdown drbd device: %s",
|
976 |
minor, result.output) |
977 |
|
978 |
def Shutdown(self): |
979 |
"""Shutdown the DRBD device.
|
980 |
|
981 |
"""
|
982 |
if self.minor is None and not self.Attach(): |
983 |
logging.info("drbd%d: not attached during Shutdown()", self._aminor) |
984 |
return
|
985 |
minor = self.minor
|
986 |
self.minor = None |
987 |
self.dev_path = None |
988 |
self._ShutdownAll(minor)
|
989 |
|
990 |
def Remove(self): |
991 |
"""Stub remove for DRBD devices.
|
992 |
|
993 |
"""
|
994 |
self.Shutdown()
|
995 |
|
996 |
def Rename(self, new_id): |
997 |
"""Rename a device.
|
998 |
|
999 |
This is not supported for drbd devices.
|
1000 |
|
1001 |
"""
|
1002 |
raise errors.ProgrammerError("Can't rename a drbd device") |
1003 |
|
1004 |
@classmethod
|
1005 |
def Create(cls, unique_id, children, size, params, excl_stor): |
1006 |
"""Create a new DRBD8 device.
|
1007 |
|
1008 |
Since DRBD devices are not created per se, just assembled, this
|
1009 |
function only initializes the metadata.
|
1010 |
|
1011 |
"""
|
1012 |
if len(children) != 2: |
1013 |
raise errors.ProgrammerError("Invalid setup for the drbd device") |
1014 |
if excl_stor:
|
1015 |
raise errors.ProgrammerError("DRBD device requested with" |
1016 |
" exclusive_storage")
|
1017 |
# check that the minor is unused
|
1018 |
aminor = unique_id[4]
|
1019 |
|
1020 |
drbd_info = DRBD8Info.CreateFromFile() |
1021 |
if drbd_info.HasMinorStatus(aminor):
|
1022 |
status = drbd_info.GetMinorStatus(aminor) |
1023 |
in_use = status.is_in_use |
1024 |
else:
|
1025 |
in_use = False
|
1026 |
if in_use:
|
1027 |
base.ThrowError("drbd%d: minor is already in use at Create() time",
|
1028 |
aminor) |
1029 |
meta = children[1]
|
1030 |
meta.Assemble() |
1031 |
if not meta.Attach(): |
1032 |
base.ThrowError("drbd%d: can't attach to meta device '%s'",
|
1033 |
aminor, meta) |
1034 |
cls._CheckMetaSize(meta.dev_path) |
1035 |
cls._InitMeta(aminor, meta.dev_path) |
1036 |
return cls(unique_id, children, size, params)
|
1037 |
|
1038 |
def Grow(self, amount, dryrun, backingstore): |
1039 |
"""Resize the DRBD device and its backing storage.
|
1040 |
|
1041 |
"""
|
1042 |
if self.minor is None: |
1043 |
base.ThrowError("drbd%d: Grow called while not attached", self._aminor) |
1044 |
if len(self._children) != 2 or None in self._children: |
1045 |
base.ThrowError("drbd%d: cannot grow diskless device", self.minor) |
1046 |
self._children[0].Grow(amount, dryrun, backingstore) |
1047 |
if dryrun or backingstore: |
1048 |
# DRBD does not support dry-run mode and is not backing storage,
|
1049 |
# so we'll return here
|
1050 |
return
|
1051 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "resize", "-s", |
1052 |
"%dm" % (self.size + amount)]) |
1053 |
if result.failed:
|
1054 |
base.ThrowError("drbd%d: resize failed: %s", self.minor, result.output) |
1055 |
|
1056 |
|
1057 |
def _CanReadDevice(path): |
1058 |
"""Check if we can read from the given device.
|
1059 |
|
1060 |
This tries to read the first 128k of the device.
|
1061 |
|
1062 |
"""
|
1063 |
try:
|
1064 |
utils.ReadFile(path, size=_DEVICE_READ_SIZE) |
1065 |
return True |
1066 |
except EnvironmentError: |
1067 |
logging.warning("Can't read from device %s", path, exc_info=True) |
1068 |
return False |