root / lib / storage / drbd.py @ 65fc2388
History | View | Annotate | Download (34.7 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 time |
27 |
|
28 |
from ganeti import constants |
29 |
from ganeti import utils |
30 |
from ganeti import errors |
31 |
from ganeti import netutils |
32 |
from ganeti import objects |
33 |
from ganeti.storage import base |
34 |
from ganeti.storage.drbd_info import DRBD8Info |
35 |
from ganeti.storage import drbd_info |
36 |
from ganeti.storage import drbd_cmdgen |
37 |
|
38 |
|
39 |
# Size of reads in _CanReadDevice
|
40 |
|
41 |
_DEVICE_READ_SIZE = 128 * 1024 |
42 |
|
43 |
|
44 |
class DRBD8(object): |
45 |
"""Various methods to deals with the DRBD system as a whole.
|
46 |
|
47 |
This class provides a set of methods to deal with the DRBD installation on
|
48 |
the node or with uninitialized devices as opposed to a DRBD device.
|
49 |
|
50 |
"""
|
51 |
_USERMODE_HELPER_FILE = "/sys/module/drbd/parameters/usermode_helper"
|
52 |
|
53 |
_MAX_MINORS = 255
|
54 |
|
55 |
@staticmethod
|
56 |
def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): |
57 |
"""Returns DRBD usermode_helper currently set.
|
58 |
|
59 |
@type filename: string
|
60 |
@param filename: the filename to read the usermode helper from
|
61 |
@rtype: string
|
62 |
@return: the currently configured DRBD usermode helper
|
63 |
|
64 |
"""
|
65 |
try:
|
66 |
helper = utils.ReadFile(filename).splitlines()[0]
|
67 |
except EnvironmentError, err: |
68 |
if err.errno == errno.ENOENT:
|
69 |
base.ThrowError("The file %s cannot be opened, check if the module"
|
70 |
" is loaded (%s)", filename, str(err)) |
71 |
else:
|
72 |
base.ThrowError("Can't read DRBD helper file %s: %s",
|
73 |
filename, str(err))
|
74 |
if not helper: |
75 |
base.ThrowError("Can't read any data from %s", filename)
|
76 |
return helper
|
77 |
|
78 |
@staticmethod
|
79 |
def GetProcInfo(): |
80 |
"""Reads and parses information from /proc/drbd.
|
81 |
|
82 |
@rtype: DRBD8Info
|
83 |
@return: a L{DRBD8Info} instance containing the current /proc/drbd info
|
84 |
|
85 |
"""
|
86 |
return DRBD8Info.CreateFromFile()
|
87 |
|
88 |
@staticmethod
|
89 |
def GetUsedDevs(): |
90 |
"""Compute the list of used DRBD minors.
|
91 |
|
92 |
@rtype: list of ints
|
93 |
|
94 |
"""
|
95 |
info = DRBD8.GetProcInfo() |
96 |
return filter(lambda m: not info.GetMinorStatus(m).is_unconfigured, |
97 |
info.GetMinors()) |
98 |
|
99 |
@staticmethod
|
100 |
def FindUnusedMinor(): |
101 |
"""Find an unused DRBD device.
|
102 |
|
103 |
This is specific to 8.x as the minors are allocated dynamically,
|
104 |
so non-existing numbers up to a max minor count are actually free.
|
105 |
|
106 |
@rtype: int
|
107 |
|
108 |
"""
|
109 |
highest = None
|
110 |
info = DRBD8.GetProcInfo() |
111 |
for minor in info.GetMinors(): |
112 |
status = info.GetMinorStatus(minor) |
113 |
if not status.is_in_use: |
114 |
return minor
|
115 |
highest = max(highest, minor)
|
116 |
|
117 |
if highest is None: # there are no minors in use at all |
118 |
return 0 |
119 |
if highest >= DRBD8._MAX_MINORS:
|
120 |
logging.error("Error: no free drbd minors!")
|
121 |
raise errors.BlockDeviceError("Can't find a free DRBD minor") |
122 |
|
123 |
return highest + 1 |
124 |
|
125 |
@staticmethod
|
126 |
def GetCmdGenerator(info): |
127 |
"""Creates a suitable L{BaseDRBDCmdGenerator} based on the given info.
|
128 |
|
129 |
@type info: DRBD8Info
|
130 |
@rtype: BaseDRBDCmdGenerator
|
131 |
|
132 |
"""
|
133 |
version = info.GetVersion() |
134 |
if version["k_minor"] <= 3: |
135 |
return drbd_cmdgen.DRBD83CmdGenerator(version)
|
136 |
else:
|
137 |
return drbd_cmdgen.DRBD84CmdGenerator(version)
|
138 |
|
139 |
@staticmethod
|
140 |
def ShutdownAll(minor): |
141 |
"""Deactivate the device.
|
142 |
|
143 |
This will, of course, fail if the device is in use.
|
144 |
|
145 |
@type minor: int
|
146 |
@param minor: the minor to shut down
|
147 |
|
148 |
"""
|
149 |
info = DRBD8.GetProcInfo() |
150 |
cmd_gen = DRBD8.GetCmdGenerator(info) |
151 |
|
152 |
cmd = cmd_gen.GenDownCmd(minor) |
153 |
result = utils.RunCmd(cmd) |
154 |
if result.failed:
|
155 |
base.ThrowError("drbd%d: can't shutdown drbd device: %s",
|
156 |
minor, result.output) |
157 |
|
158 |
|
159 |
class DRBD8Dev(base.BlockDev): |
160 |
"""DRBD v8.x block device.
|
161 |
|
162 |
This implements the local host part of the DRBD device, i.e. it
|
163 |
doesn't do anything to the supposed peer. If you need a fully
|
164 |
connected DRBD pair, you need to use this class on both hosts.
|
165 |
|
166 |
The unique_id for the drbd device is a (local_ip, local_port,
|
167 |
remote_ip, remote_port, local_minor, secret) tuple, and it must have
|
168 |
two children: the data device and the meta_device. The meta device
|
169 |
is checked for valid size and is zeroed on create.
|
170 |
|
171 |
"""
|
172 |
_DRBD_MAJOR = 147
|
173 |
|
174 |
# timeout constants
|
175 |
_NET_RECONFIG_TIMEOUT = 60
|
176 |
|
177 |
def __init__(self, unique_id, children, size, params): |
178 |
if children and children.count(None) > 0: |
179 |
children = [] |
180 |
if len(children) not in (0, 2): |
181 |
raise ValueError("Invalid configuration data %s" % str(children)) |
182 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6: |
183 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
184 |
(self._lhost, self._lport, |
185 |
self._rhost, self._rport, |
186 |
self._aminor, self._secret) = unique_id |
187 |
if children:
|
188 |
if not _CanReadDevice(children[1].dev_path): |
189 |
logging.info("drbd%s: Ignoring unreadable meta device", self._aminor) |
190 |
children = [] |
191 |
super(DRBD8Dev, self).__init__(unique_id, children, size, params) |
192 |
self.major = self._DRBD_MAJOR |
193 |
|
194 |
info = DRBD8.GetProcInfo() |
195 |
version = info.GetVersion() |
196 |
if version["k_major"] != 8: |
197 |
base.ThrowError("Mismatch in DRBD kernel version and requested ganeti"
|
198 |
" usage: kernel is %s.%s, ganeti wants 8.x",
|
199 |
version["k_major"], version["k_minor"]) |
200 |
|
201 |
if version["k_minor"] <= 3: |
202 |
self._show_info_cls = drbd_info.DRBD83ShowInfo
|
203 |
else:
|
204 |
self._show_info_cls = drbd_info.DRBD84ShowInfo
|
205 |
|
206 |
self._cmd_gen = DRBD8.GetCmdGenerator(info)
|
207 |
|
208 |
if (self._lhost is not None and self._lhost == self._rhost and |
209 |
self._lport == self._rport): |
210 |
raise ValueError("Invalid configuration data, same local/remote %s" % |
211 |
(unique_id,)) |
212 |
self.Attach()
|
213 |
|
214 |
@staticmethod
|
215 |
def _DevPath(minor): |
216 |
"""Return the path to a drbd device for a given minor.
|
217 |
|
218 |
@type minor: int
|
219 |
@rtype: string
|
220 |
|
221 |
"""
|
222 |
return "/dev/drbd%d" % minor |
223 |
|
224 |
def _SetFromMinor(self, minor): |
225 |
"""Set our parameters based on the given minor.
|
226 |
|
227 |
This sets our minor variable and our dev_path.
|
228 |
|
229 |
@type minor: int
|
230 |
|
231 |
"""
|
232 |
if minor is None: |
233 |
self.minor = self.dev_path = None |
234 |
self.attached = False |
235 |
else:
|
236 |
self.minor = minor
|
237 |
self.dev_path = self._DevPath(minor) |
238 |
self.attached = True |
239 |
|
240 |
@staticmethod
|
241 |
def _CheckMetaSize(meta_device): |
242 |
"""Check if the given meta device looks like a valid one.
|
243 |
|
244 |
This currently only checks the size, which must be around
|
245 |
128MiB.
|
246 |
|
247 |
@type meta_device: string
|
248 |
@param meta_device: the path to the device to check
|
249 |
|
250 |
"""
|
251 |
result = utils.RunCmd(["blockdev", "--getsize", meta_device]) |
252 |
if result.failed:
|
253 |
base.ThrowError("Failed to get device size: %s - %s",
|
254 |
result.fail_reason, result.output) |
255 |
try:
|
256 |
sectors = int(result.stdout)
|
257 |
except (TypeError, ValueError): |
258 |
base.ThrowError("Invalid output from blockdev: '%s'", result.stdout)
|
259 |
num_bytes = sectors * 512
|
260 |
if num_bytes < 128 * 1024 * 1024: # less than 128MiB |
261 |
base.ThrowError("Meta device too small (%.2fMib)",
|
262 |
(num_bytes / 1024 / 1024)) |
263 |
# the maximum *valid* size of the meta device when living on top
|
264 |
# of LVM is hard to compute: it depends on the number of stripes
|
265 |
# and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB
|
266 |
# (normal size), but an eight-stripe 128MB PE will result in a 1GB
|
267 |
# size meta device; as such, we restrict it to 1GB (a little bit
|
268 |
# too generous, but making assumptions about PE size is hard)
|
269 |
if num_bytes > 1024 * 1024 * 1024: |
270 |
base.ThrowError("Meta device too big (%.2fMiB)",
|
271 |
(num_bytes / 1024 / 1024)) |
272 |
|
273 |
def _GetShowData(self, minor): |
274 |
"""Return the `drbdsetup show` data.
|
275 |
|
276 |
@type minor: int
|
277 |
@param minor: the minor to collect show output for
|
278 |
@rtype: string
|
279 |
|
280 |
"""
|
281 |
result = utils.RunCmd(self._cmd_gen.GenShowCmd(minor))
|
282 |
if result.failed:
|
283 |
logging.error("Can't display the drbd config: %s - %s",
|
284 |
result.fail_reason, result.output) |
285 |
return None |
286 |
return result.stdout
|
287 |
|
288 |
def _GetShowInfo(self, minor): |
289 |
"""Return parsed information from `drbdsetup show`.
|
290 |
|
291 |
@type minor: int
|
292 |
@param minor: the minor to return information for
|
293 |
@rtype: dict as described in L{drbd_info.BaseShowInfo.GetDevInfo}
|
294 |
|
295 |
"""
|
296 |
return self._show_info_cls.GetDevInfo(self._GetShowData(minor)) |
297 |
|
298 |
def _MatchesLocal(self, info): |
299 |
"""Test if our local config matches with an existing device.
|
300 |
|
301 |
The parameter should be as returned from `_GetShowInfo()`. This
|
302 |
method tests if our local backing device is the same as the one in
|
303 |
the info parameter, in effect testing if we look like the given
|
304 |
device.
|
305 |
|
306 |
@type info: dict as described in L{drbd_info.BaseShowInfo.GetDevInfo}
|
307 |
@rtype: boolean
|
308 |
|
309 |
"""
|
310 |
if self._children: |
311 |
backend, meta = self._children
|
312 |
else:
|
313 |
backend = meta = None
|
314 |
|
315 |
if backend is not None: |
316 |
retval = ("local_dev" in info and info["local_dev"] == backend.dev_path) |
317 |
else:
|
318 |
retval = ("local_dev" not in info) |
319 |
|
320 |
if meta is not None: |
321 |
retval = retval and ("meta_dev" in info and |
322 |
info["meta_dev"] == meta.dev_path)
|
323 |
if "meta_index" in info: |
324 |
retval = retval and info["meta_index"] == 0 |
325 |
else:
|
326 |
retval = retval and ("meta_dev" not in info and |
327 |
"meta_index" not in info) |
328 |
return retval
|
329 |
|
330 |
def _MatchesNet(self, info): |
331 |
"""Test if our network config matches with an existing device.
|
332 |
|
333 |
The parameter should be as returned from `_GetShowInfo()`. This
|
334 |
method tests if our network configuration is the same as the one
|
335 |
in the info parameter, in effect testing if we look like the given
|
336 |
device.
|
337 |
|
338 |
@type info: dict as described in L{drbd_info.BaseShowInfo.GetDevInfo}
|
339 |
@rtype: boolean
|
340 |
|
341 |
"""
|
342 |
if (((self._lhost is None and not ("local_addr" in info)) and |
343 |
(self._rhost is None and not ("remote_addr" in info)))): |
344 |
return True |
345 |
|
346 |
if self._lhost is None: |
347 |
return False |
348 |
|
349 |
if not ("local_addr" in info and |
350 |
"remote_addr" in info): |
351 |
return False |
352 |
|
353 |
retval = (info["local_addr"] == (self._lhost, self._lport)) |
354 |
retval = (retval and
|
355 |
info["remote_addr"] == (self._rhost, self._rport)) |
356 |
return retval
|
357 |
|
358 |
def _AssembleLocal(self, minor, backend, meta, size): |
359 |
"""Configure the local part of a DRBD device.
|
360 |
|
361 |
@type minor: int
|
362 |
@param minor: the minor to assemble locally
|
363 |
@type backend: string
|
364 |
@param backend: path to the data device to use
|
365 |
@type meta: string
|
366 |
@param meta: path to the meta device to use
|
367 |
@type size: int
|
368 |
@param size: size in MiB
|
369 |
|
370 |
"""
|
371 |
cmds = self._cmd_gen.GenLocalInitCmds(minor, backend, meta,
|
372 |
size, self.params)
|
373 |
|
374 |
for cmd in cmds: |
375 |
result = utils.RunCmd(cmd) |
376 |
if result.failed:
|
377 |
base.ThrowError("drbd%d: can't attach local disk: %s",
|
378 |
minor, result.output) |
379 |
|
380 |
def _AssembleNet(self, minor, net_info, dual_pri=False, hmac=None, |
381 |
secret=None):
|
382 |
"""Configure the network part of the device.
|
383 |
|
384 |
@type minor: int
|
385 |
@param minor: the minor to assemble the network for
|
386 |
@type net_info: (string, int, string, int)
|
387 |
@param net_info: tuple containing the local address, local port, remote
|
388 |
address and remote port
|
389 |
@type dual_pri: boolean
|
390 |
@param dual_pri: whether two primaries should be allowed or not
|
391 |
@type hmac: string
|
392 |
@param hmac: the HMAC algorithm to use
|
393 |
@type secret: string
|
394 |
@param secret: the shared secret to use
|
395 |
|
396 |
"""
|
397 |
lhost, lport, rhost, rport = net_info |
398 |
if None in net_info: |
399 |
# we don't want network connection and actually want to make
|
400 |
# sure its shutdown
|
401 |
self._ShutdownNet(minor)
|
402 |
return
|
403 |
|
404 |
if dual_pri:
|
405 |
protocol = constants.DRBD_MIGRATION_NET_PROTOCOL |
406 |
else:
|
407 |
protocol = self.params[constants.LDP_PROTOCOL]
|
408 |
|
409 |
# Workaround for a race condition. When DRBD is doing its dance to
|
410 |
# establish a connection with its peer, it also sends the
|
411 |
# synchronization speed over the wire. In some cases setting the
|
412 |
# sync speed only after setting up both sides can race with DRBD
|
413 |
# connecting, hence we set it here before telling DRBD anything
|
414 |
# about its peer.
|
415 |
sync_errors = self._SetMinorSyncParams(minor, self.params) |
416 |
if sync_errors:
|
417 |
base.ThrowError("drbd%d: can't set the synchronization parameters: %s" %
|
418 |
(minor, utils.CommaJoin(sync_errors))) |
419 |
|
420 |
family = self._GetNetFamily(minor, lhost, rhost)
|
421 |
|
422 |
cmd = self._cmd_gen.GenNetInitCmd(minor, family, lhost, lport,
|
423 |
rhost, rport, protocol, |
424 |
dual_pri, hmac, secret, self.params)
|
425 |
|
426 |
result = utils.RunCmd(cmd) |
427 |
if result.failed:
|
428 |
base.ThrowError("drbd%d: can't setup network: %s - %s",
|
429 |
minor, result.fail_reason, result.output) |
430 |
|
431 |
def _CheckNetworkConfig(): |
432 |
info = self._GetShowInfo(minor)
|
433 |
if not "local_addr" in info or not "remote_addr" in info: |
434 |
raise utils.RetryAgain()
|
435 |
|
436 |
if (info["local_addr"] != (lhost, lport) or |
437 |
info["remote_addr"] != (rhost, rport)):
|
438 |
raise utils.RetryAgain()
|
439 |
|
440 |
try:
|
441 |
utils.Retry(_CheckNetworkConfig, 1.0, 10.0) |
442 |
except utils.RetryTimeout:
|
443 |
base.ThrowError("drbd%d: timeout while configuring network", minor)
|
444 |
|
445 |
@staticmethod
|
446 |
def _GetNetFamily(minor, lhost, rhost): |
447 |
if netutils.IP6Address.IsValid(lhost):
|
448 |
if not netutils.IP6Address.IsValid(rhost): |
449 |
base.ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
450 |
(minor, lhost, rhost)) |
451 |
return "ipv6" |
452 |
elif netutils.IP4Address.IsValid(lhost):
|
453 |
if not netutils.IP4Address.IsValid(rhost): |
454 |
base.ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
455 |
(minor, lhost, rhost)) |
456 |
return "ipv4" |
457 |
else:
|
458 |
base.ThrowError("drbd%d: Invalid ip %s" % (minor, lhost))
|
459 |
|
460 |
def AddChildren(self, devices): |
461 |
"""Add a disk to the DRBD device.
|
462 |
|
463 |
@type devices: list of L{BlockDev}
|
464 |
@param devices: a list of exactly two L{BlockDev} objects; the first
|
465 |
denotes the data device, the second the meta device for this DRBD device
|
466 |
|
467 |
"""
|
468 |
if self.minor is None: |
469 |
base.ThrowError("drbd%d: can't attach to dbrd8 during AddChildren",
|
470 |
self._aminor)
|
471 |
if len(devices) != 2: |
472 |
base.ThrowError("drbd%d: need two devices for AddChildren", self.minor) |
473 |
info = self._GetShowInfo(self.minor) |
474 |
if "local_dev" in info: |
475 |
base.ThrowError("drbd%d: already attached to a local disk", self.minor) |
476 |
backend, meta = devices |
477 |
if backend.dev_path is None or meta.dev_path is None: |
478 |
base.ThrowError("drbd%d: children not ready during AddChildren",
|
479 |
self.minor)
|
480 |
backend.Open() |
481 |
meta.Open() |
482 |
self._CheckMetaSize(meta.dev_path)
|
483 |
self._InitMeta(DRBD8.FindUnusedMinor(), meta.dev_path)
|
484 |
|
485 |
self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size) |
486 |
self._children = devices
|
487 |
|
488 |
def RemoveChildren(self, devices): |
489 |
"""Detach the drbd device from local storage.
|
490 |
|
491 |
@type devices: list of L{BlockDev}
|
492 |
@param devices: a list of exactly two L{BlockDev} objects; the first
|
493 |
denotes the data device, the second the meta device for this DRBD device
|
494 |
|
495 |
"""
|
496 |
if self.minor is None: |
497 |
base.ThrowError("drbd%d: can't attach to drbd8 during RemoveChildren",
|
498 |
self._aminor)
|
499 |
# early return if we don't actually have backing storage
|
500 |
info = self._GetShowInfo(self.minor) |
501 |
if "local_dev" not in info: |
502 |
return
|
503 |
if len(self._children) != 2: |
504 |
base.ThrowError("drbd%d: we don't have two children: %s", self.minor, |
505 |
self._children)
|
506 |
if self._children.count(None) == 2: # we don't actually have children :) |
507 |
logging.warning("drbd%d: requested detach while detached", self.minor) |
508 |
return
|
509 |
if len(devices) != 2: |
510 |
base.ThrowError("drbd%d: we need two children in RemoveChildren",
|
511 |
self.minor)
|
512 |
for child, dev in zip(self._children, devices): |
513 |
if dev != child.dev_path:
|
514 |
base.ThrowError("drbd%d: mismatch in local storage (%s != %s) in"
|
515 |
" RemoveChildren", self.minor, dev, child.dev_path) |
516 |
|
517 |
self._ShutdownLocal(self.minor) |
518 |
self._children = []
|
519 |
|
520 |
def _SetMinorSyncParams(self, minor, params): |
521 |
"""Set the parameters of the DRBD syncer.
|
522 |
|
523 |
This is the low-level implementation.
|
524 |
|
525 |
@type minor: int
|
526 |
@param minor: the drbd minor whose settings we change
|
527 |
@type params: dict
|
528 |
@param params: LD level disk parameters related to the synchronization
|
529 |
@rtype: list
|
530 |
@return: a list of error messages
|
531 |
|
532 |
"""
|
533 |
cmd = self._cmd_gen.GenSyncParamsCmd(minor, params)
|
534 |
result = utils.RunCmd(cmd) |
535 |
if result.failed:
|
536 |
msg = ("Can't change syncer rate: %s - %s" %
|
537 |
(result.fail_reason, result.output)) |
538 |
logging.error(msg) |
539 |
return [msg]
|
540 |
|
541 |
return []
|
542 |
|
543 |
def SetSyncParams(self, params): |
544 |
"""Set the synchronization parameters of the DRBD syncer.
|
545 |
|
546 |
See L{BlockDev.SetSyncParams} for parameter description.
|
547 |
|
548 |
"""
|
549 |
if self.minor is None: |
550 |
err = "Not attached during SetSyncParams"
|
551 |
logging.info(err) |
552 |
return [err]
|
553 |
|
554 |
children_result = super(DRBD8Dev, self).SetSyncParams(params) |
555 |
children_result.extend(self._SetMinorSyncParams(self.minor, params)) |
556 |
return children_result
|
557 |
|
558 |
def PauseResumeSync(self, pause): |
559 |
"""Pauses or resumes the sync of a DRBD device.
|
560 |
|
561 |
See L{BlockDev.PauseResumeSync} for parameter description.
|
562 |
|
563 |
"""
|
564 |
if self.minor is None: |
565 |
logging.info("Not attached during PauseSync")
|
566 |
return False |
567 |
|
568 |
children_result = super(DRBD8Dev, self).PauseResumeSync(pause) |
569 |
|
570 |
if pause:
|
571 |
cmd = self._cmd_gen.GenPauseSyncCmd(self.minor) |
572 |
else:
|
573 |
cmd = self._cmd_gen.GenResumeSyncCmd(self.minor) |
574 |
|
575 |
result = utils.RunCmd(cmd) |
576 |
if result.failed:
|
577 |
logging.error("Can't %s: %s - %s", cmd,
|
578 |
result.fail_reason, result.output) |
579 |
return not result.failed and children_result |
580 |
|
581 |
def GetProcStatus(self): |
582 |
"""Return the current status data from /proc/drbd for this device.
|
583 |
|
584 |
@rtype: DRBD8Status
|
585 |
|
586 |
"""
|
587 |
if self.minor is None: |
588 |
base.ThrowError("drbd%d: GetStats() called while not attached",
|
589 |
self._aminor)
|
590 |
info = DRBD8.GetProcInfo() |
591 |
if not info.HasMinorStatus(self.minor): |
592 |
base.ThrowError("drbd%d: can't find myself in /proc", self.minor) |
593 |
return info.GetMinorStatus(self.minor) |
594 |
|
595 |
def GetSyncStatus(self): |
596 |
"""Returns the sync status of the device.
|
597 |
|
598 |
If sync_percent is None, it means all is ok
|
599 |
If estimated_time is None, it means we can't estimate
|
600 |
the time needed, otherwise it's the time left in seconds.
|
601 |
|
602 |
We set the is_degraded parameter to True on two conditions:
|
603 |
network not connected or local disk missing.
|
604 |
|
605 |
We compute the ldisk parameter based on whether we have a local
|
606 |
disk or not.
|
607 |
|
608 |
@rtype: objects.BlockDevStatus
|
609 |
|
610 |
"""
|
611 |
if self.minor is None and not self.Attach(): |
612 |
base.ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) |
613 |
|
614 |
stats = self.GetProcStatus()
|
615 |
is_degraded = not stats.is_connected or not stats.is_disk_uptodate |
616 |
|
617 |
if stats.is_disk_uptodate:
|
618 |
ldisk_status = constants.LDS_OKAY |
619 |
elif stats.is_diskless:
|
620 |
ldisk_status = constants.LDS_FAULTY |
621 |
else:
|
622 |
ldisk_status = constants.LDS_UNKNOWN |
623 |
|
624 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
625 |
major=self.major,
|
626 |
minor=self.minor,
|
627 |
sync_percent=stats.sync_percent, |
628 |
estimated_time=stats.est_time, |
629 |
is_degraded=is_degraded, |
630 |
ldisk_status=ldisk_status) |
631 |
|
632 |
def Open(self, force=False): |
633 |
"""Make the local state primary.
|
634 |
|
635 |
If the 'force' parameter is given, DRBD is instructed to switch the device
|
636 |
into primary mode. Since this is a potentially dangerous operation, the
|
637 |
force flag should be only given after creation, when it actually is
|
638 |
mandatory.
|
639 |
|
640 |
"""
|
641 |
if self.minor is None and not self.Attach(): |
642 |
logging.error("DRBD cannot attach to a device during open")
|
643 |
return False |
644 |
|
645 |
cmd = self._cmd_gen.GenPrimaryCmd(self.minor, force) |
646 |
|
647 |
result = utils.RunCmd(cmd) |
648 |
if result.failed:
|
649 |
base.ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, |
650 |
result.output) |
651 |
|
652 |
def Close(self): |
653 |
"""Make the local state secondary.
|
654 |
|
655 |
This will, of course, fail if the device is in use.
|
656 |
|
657 |
"""
|
658 |
if self.minor is None and not self.Attach(): |
659 |
base.ThrowError("drbd%d: can't Attach() in Close()", self._aminor) |
660 |
cmd = self._cmd_gen.GenSecondaryCmd(self.minor) |
661 |
result = utils.RunCmd(cmd) |
662 |
if result.failed:
|
663 |
base.ThrowError("drbd%d: can't switch drbd device to secondary: %s",
|
664 |
self.minor, result.output)
|
665 |
|
666 |
def DisconnectNet(self): |
667 |
"""Removes network configuration.
|
668 |
|
669 |
This method shutdowns the network side of the device.
|
670 |
|
671 |
The method will wait up to a hardcoded timeout for the device to
|
672 |
go into standalone after the 'disconnect' command before
|
673 |
re-configuring it, as sometimes it takes a while for the
|
674 |
disconnect to actually propagate and thus we might issue a 'net'
|
675 |
command while the device is still connected. If the device will
|
676 |
still be attached to the network and we time out, we raise an
|
677 |
exception.
|
678 |
|
679 |
"""
|
680 |
if self.minor is None: |
681 |
base.ThrowError("drbd%d: disk not attached in re-attach net",
|
682 |
self._aminor)
|
683 |
|
684 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
685 |
base.ThrowError("drbd%d: DRBD disk missing network info in"
|
686 |
" DisconnectNet()", self.minor) |
687 |
|
688 |
class _DisconnectStatus: |
689 |
def __init__(self, ever_disconnected): |
690 |
self.ever_disconnected = ever_disconnected
|
691 |
|
692 |
dstatus = _DisconnectStatus(base.IgnoreError(self._ShutdownNet, self.minor)) |
693 |
|
694 |
def _WaitForDisconnect(): |
695 |
if self.GetProcStatus().is_standalone: |
696 |
return
|
697 |
|
698 |
# retry the disconnect, it seems possible that due to a well-time
|
699 |
# disconnect on the peer, my disconnect command might be ignored and
|
700 |
# forgotten
|
701 |
dstatus.ever_disconnected = \ |
702 |
base.IgnoreError(self._ShutdownNet, self.minor) or \ |
703 |
dstatus.ever_disconnected |
704 |
|
705 |
raise utils.RetryAgain()
|
706 |
|
707 |
# Keep start time
|
708 |
start_time = time.time() |
709 |
|
710 |
try:
|
711 |
# Start delay at 100 milliseconds and grow up to 2 seconds
|
712 |
utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0), |
713 |
self._NET_RECONFIG_TIMEOUT)
|
714 |
except utils.RetryTimeout:
|
715 |
if dstatus.ever_disconnected:
|
716 |
msg = ("drbd%d: device did not react to the"
|
717 |
" 'disconnect' command in a timely manner")
|
718 |
else:
|
719 |
msg = "drbd%d: can't shutdown network, even after multiple retries"
|
720 |
|
721 |
base.ThrowError(msg, self.minor)
|
722 |
|
723 |
reconfig_time = time.time() - start_time |
724 |
if reconfig_time > (self._NET_RECONFIG_TIMEOUT * 0.25): |
725 |
logging.info("drbd%d: DisconnectNet: detach took %.3f seconds",
|
726 |
self.minor, reconfig_time)
|
727 |
|
728 |
def AttachNet(self, multimaster): |
729 |
"""Reconnects the network.
|
730 |
|
731 |
This method connects the network side of the device with a
|
732 |
specified multi-master flag. The device needs to be 'Standalone'
|
733 |
but have valid network configuration data.
|
734 |
|
735 |
@type multimaster: boolean
|
736 |
@param multimaster: init the network in dual-primary mode
|
737 |
|
738 |
"""
|
739 |
if self.minor is None: |
740 |
base.ThrowError("drbd%d: device not attached in AttachNet", self._aminor) |
741 |
|
742 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
743 |
base.ThrowError("drbd%d: missing network info in AttachNet()", self.minor) |
744 |
|
745 |
status = self.GetProcStatus()
|
746 |
|
747 |
if not status.is_standalone: |
748 |
base.ThrowError("drbd%d: device is not standalone in AttachNet",
|
749 |
self.minor)
|
750 |
|
751 |
self._AssembleNet(self.minor, |
752 |
(self._lhost, self._lport, self._rhost, self._rport), |
753 |
dual_pri=multimaster, hmac=constants.DRBD_HMAC_ALG, |
754 |
secret=self._secret)
|
755 |
|
756 |
def Attach(self): |
757 |
"""Check if our minor is configured.
|
758 |
|
759 |
This doesn't do any device configurations - it only checks if the
|
760 |
minor is in a state different from Unconfigured.
|
761 |
|
762 |
Note that this function will not change the state of the system in
|
763 |
any way (except in case of side-effects caused by reading from
|
764 |
/proc).
|
765 |
|
766 |
"""
|
767 |
used_devs = DRBD8.GetUsedDevs() |
768 |
if self._aminor in used_devs: |
769 |
minor = self._aminor
|
770 |
else:
|
771 |
minor = None
|
772 |
|
773 |
self._SetFromMinor(minor)
|
774 |
return minor is not None |
775 |
|
776 |
def Assemble(self): |
777 |
"""Assemble the drbd.
|
778 |
|
779 |
Method:
|
780 |
- if we have a configured device, we try to ensure that it matches
|
781 |
our config
|
782 |
- if not, we create it from zero
|
783 |
- anyway, set the device parameters
|
784 |
|
785 |
"""
|
786 |
super(DRBD8Dev, self).Assemble() |
787 |
|
788 |
self.Attach()
|
789 |
if self.minor is None: |
790 |
# local device completely unconfigured
|
791 |
self._FastAssemble()
|
792 |
else:
|
793 |
# we have to recheck the local and network status and try to fix
|
794 |
# the device
|
795 |
self._SlowAssemble()
|
796 |
|
797 |
sync_errors = self.SetSyncParams(self.params) |
798 |
if sync_errors:
|
799 |
base.ThrowError("drbd%d: can't set the synchronization parameters: %s" %
|
800 |
(self.minor, utils.CommaJoin(sync_errors)))
|
801 |
|
802 |
def _SlowAssemble(self): |
803 |
"""Assembles the DRBD device from a (partially) configured device.
|
804 |
|
805 |
In case of partially attached (local device matches but no network
|
806 |
setup), we perform the network attach. If successful, we re-test
|
807 |
the attach if can return success.
|
808 |
|
809 |
"""
|
810 |
# TODO: Rewrite to not use a for loop just because there is 'break'
|
811 |
# pylint: disable=W0631
|
812 |
net_data = (self._lhost, self._lport, self._rhost, self._rport) |
813 |
for minor in (self._aminor,): |
814 |
info = self._GetShowInfo(minor)
|
815 |
match_l = self._MatchesLocal(info)
|
816 |
match_r = self._MatchesNet(info)
|
817 |
|
818 |
if match_l and match_r: |
819 |
# everything matches
|
820 |
break
|
821 |
|
822 |
if match_l and not match_r and "local_addr" not in info: |
823 |
# disk matches, but not attached to network, attach and recheck
|
824 |
self._AssembleNet(minor, net_data, hmac=constants.DRBD_HMAC_ALG,
|
825 |
secret=self._secret)
|
826 |
if self._MatchesNet(self._GetShowInfo(minor)): |
827 |
break
|
828 |
else:
|
829 |
base.ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
830 |
" show' disagrees", minor)
|
831 |
|
832 |
if match_r and "local_dev" not in info: |
833 |
# no local disk, but network attached and it matches
|
834 |
self._AssembleLocal(minor, self._children[0].dev_path, |
835 |
self._children[1].dev_path, self.size) |
836 |
if self._MatchesLocal(self._GetShowInfo(minor)): |
837 |
break
|
838 |
else:
|
839 |
base.ThrowError("drbd%d: disk attach successful, but 'drbdsetup"
|
840 |
" show' disagrees", minor)
|
841 |
|
842 |
# this case must be considered only if we actually have local
|
843 |
# storage, i.e. not in diskless mode, because all diskless
|
844 |
# devices are equal from the point of view of local
|
845 |
# configuration
|
846 |
if (match_l and "local_dev" in info and |
847 |
not match_r and "local_addr" in info): |
848 |
# strange case - the device network part points to somewhere
|
849 |
# else, even though its local storage is ours; as we own the
|
850 |
# drbd space, we try to disconnect from the remote peer and
|
851 |
# reconnect to our correct one
|
852 |
try:
|
853 |
self._ShutdownNet(minor)
|
854 |
except errors.BlockDeviceError, err:
|
855 |
base.ThrowError("drbd%d: device has correct local storage, wrong"
|
856 |
" remote peer and is unable to disconnect in order"
|
857 |
" to attach to the correct peer: %s", minor, str(err)) |
858 |
# note: _AssembleNet also handles the case when we don't want
|
859 |
# local storage (i.e. one or more of the _[lr](host|port) is
|
860 |
# None)
|
861 |
self._AssembleNet(minor, net_data, hmac=constants.DRBD_HMAC_ALG,
|
862 |
secret=self._secret)
|
863 |
if self._MatchesNet(self._GetShowInfo(minor)): |
864 |
break
|
865 |
else:
|
866 |
base.ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
867 |
" show' disagrees", minor)
|
868 |
|
869 |
else:
|
870 |
minor = None
|
871 |
|
872 |
self._SetFromMinor(minor)
|
873 |
if minor is None: |
874 |
base.ThrowError("drbd%d: cannot activate, unknown or unhandled reason",
|
875 |
self._aminor)
|
876 |
|
877 |
def _FastAssemble(self): |
878 |
"""Assemble the drbd device from zero.
|
879 |
|
880 |
This is run when in Assemble we detect our minor is unused.
|
881 |
|
882 |
"""
|
883 |
minor = self._aminor
|
884 |
if self._children and self._children[0] and self._children[1]: |
885 |
self._AssembleLocal(minor, self._children[0].dev_path, |
886 |
self._children[1].dev_path, self.size) |
887 |
if self._lhost and self._lport and self._rhost and self._rport: |
888 |
self._AssembleNet(minor,
|
889 |
(self._lhost, self._lport, self._rhost, self._rport), |
890 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
891 |
self._SetFromMinor(minor)
|
892 |
|
893 |
def _ShutdownLocal(self, minor): |
894 |
"""Detach from the local device.
|
895 |
|
896 |
I/Os will continue to be served from the remote device. If we
|
897 |
don't have a remote device, this operation will fail.
|
898 |
|
899 |
@type minor: int
|
900 |
@param minor: the device to detach from the local device
|
901 |
|
902 |
"""
|
903 |
cmd = self._cmd_gen.GenDetachCmd(minor)
|
904 |
result = utils.RunCmd(cmd) |
905 |
if result.failed:
|
906 |
base.ThrowError("drbd%d: can't detach local disk: %s",
|
907 |
minor, result.output) |
908 |
|
909 |
def _ShutdownNet(self, minor): |
910 |
"""Disconnect from the remote peer.
|
911 |
|
912 |
This fails if we don't have a local device.
|
913 |
|
914 |
@type minor: boolean
|
915 |
@param minor: the device to disconnect from the remote peer
|
916 |
|
917 |
"""
|
918 |
family = self._GetNetFamily(minor, self._lhost, self._rhost) |
919 |
cmd = self._cmd_gen.GenDisconnectCmd(minor, family,
|
920 |
self._lhost, self._lport, |
921 |
self._rhost, self._rport) |
922 |
result = utils.RunCmd(cmd) |
923 |
if result.failed:
|
924 |
base.ThrowError("drbd%d: can't shutdown network: %s",
|
925 |
minor, result.output) |
926 |
|
927 |
def Shutdown(self): |
928 |
"""Shutdown the DRBD device.
|
929 |
|
930 |
"""
|
931 |
if self.minor is None and not self.Attach(): |
932 |
logging.info("drbd%d: not attached during Shutdown()", self._aminor) |
933 |
return
|
934 |
|
935 |
try:
|
936 |
DRBD8.ShutdownAll(self.minor)
|
937 |
finally:
|
938 |
self.minor = None |
939 |
self.dev_path = None |
940 |
|
941 |
def Remove(self): |
942 |
"""Stub remove for DRBD devices.
|
943 |
|
944 |
"""
|
945 |
self.Shutdown()
|
946 |
|
947 |
def Rename(self, new_id): |
948 |
"""Rename a device.
|
949 |
|
950 |
This is not supported for drbd devices.
|
951 |
|
952 |
"""
|
953 |
raise errors.ProgrammerError("Can't rename a drbd device") |
954 |
|
955 |
def Grow(self, amount, dryrun, backingstore, excl_stor): |
956 |
"""Resize the DRBD device and its backing storage.
|
957 |
|
958 |
See L{BlockDev.Grow} for parameter description.
|
959 |
|
960 |
"""
|
961 |
if self.minor is None: |
962 |
base.ThrowError("drbd%d: Grow called while not attached", self._aminor) |
963 |
if len(self._children) != 2 or None in self._children: |
964 |
base.ThrowError("drbd%d: cannot grow diskless device", self.minor) |
965 |
self._children[0].Grow(amount, dryrun, backingstore, excl_stor) |
966 |
if dryrun or backingstore: |
967 |
# DRBD does not support dry-run mode and is not backing storage,
|
968 |
# so we'll return here
|
969 |
return
|
970 |
cmd = self._cmd_gen.GenResizeCmd(self.minor, self.size + amount) |
971 |
result = utils.RunCmd(cmd) |
972 |
if result.failed:
|
973 |
base.ThrowError("drbd%d: resize failed: %s", self.minor, result.output) |
974 |
|
975 |
@classmethod
|
976 |
def _InitMeta(cls, minor, dev_path): |
977 |
"""Initialize a meta device.
|
978 |
|
979 |
This will not work if the given minor is in use.
|
980 |
|
981 |
@type minor: int
|
982 |
@param minor: the DRBD minor whose (future) meta device should be
|
983 |
initialized
|
984 |
@type dev_path: string
|
985 |
@param dev_path: path to the meta device to initialize
|
986 |
|
987 |
"""
|
988 |
# Zero the metadata first, in order to make sure drbdmeta doesn't
|
989 |
# try to auto-detect existing filesystems or similar (see
|
990 |
# http://code.google.com/p/ganeti/issues/detail?id=182); we only
|
991 |
# care about the first 128MB of data in the device, even though it
|
992 |
# can be bigger
|
993 |
result = utils.RunCmd([constants.DD_CMD, |
994 |
"if=/dev/zero", "of=%s" % dev_path, |
995 |
"bs=1048576", "count=128", "oflag=direct"]) |
996 |
if result.failed:
|
997 |
base.ThrowError("Can't wipe the meta device: %s", result.output)
|
998 |
|
999 |
info = DRBD8.GetProcInfo() |
1000 |
cmd_gen = DRBD8.GetCmdGenerator(info) |
1001 |
cmd = cmd_gen.GenInitMetaCmd(minor, dev_path) |
1002 |
|
1003 |
result = utils.RunCmd(cmd) |
1004 |
if result.failed:
|
1005 |
base.ThrowError("Can't initialize meta device: %s", result.output)
|
1006 |
|
1007 |
@classmethod
|
1008 |
def Create(cls, unique_id, children, size, spindles, params, excl_stor): |
1009 |
"""Create a new DRBD8 device.
|
1010 |
|
1011 |
Since DRBD devices are not created per se, just assembled, this
|
1012 |
function only initializes the metadata.
|
1013 |
|
1014 |
"""
|
1015 |
if len(children) != 2: |
1016 |
raise errors.ProgrammerError("Invalid setup for the drbd device") |
1017 |
if excl_stor:
|
1018 |
raise errors.ProgrammerError("DRBD device requested with" |
1019 |
" exclusive_storage")
|
1020 |
# check that the minor is unused
|
1021 |
aminor = unique_id[4]
|
1022 |
|
1023 |
info = DRBD8.GetProcInfo() |
1024 |
if info.HasMinorStatus(aminor):
|
1025 |
status = info.GetMinorStatus(aminor) |
1026 |
in_use = status.is_in_use |
1027 |
else:
|
1028 |
in_use = False
|
1029 |
if in_use:
|
1030 |
base.ThrowError("drbd%d: minor is already in use at Create() time",
|
1031 |
aminor) |
1032 |
meta = children[1]
|
1033 |
meta.Assemble() |
1034 |
if not meta.Attach(): |
1035 |
base.ThrowError("drbd%d: can't attach to meta device '%s'",
|
1036 |
aminor, meta) |
1037 |
cls._CheckMetaSize(meta.dev_path) |
1038 |
cls._InitMeta(aminor, meta.dev_path) |
1039 |
return cls(unique_id, children, size, params)
|
1040 |
|
1041 |
|
1042 |
def _CanReadDevice(path): |
1043 |
"""Check if we can read from the given device.
|
1044 |
|
1045 |
This tries to read the first 128k of the device.
|
1046 |
|
1047 |
@type path: string
|
1048 |
|
1049 |
"""
|
1050 |
try:
|
1051 |
utils.ReadFile(path, size=_DEVICE_READ_SIZE) |
1052 |
return True |
1053 |
except EnvironmentError: |
1054 |
logging.warning("Can't read from device %s", path, exc_info=True) |
1055 |
return False |