root / lib / bdev.py @ fffe93e7
History | View | Annotate | Download (78.1 kB)
1 |
#
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007, 2010, 2011 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 |
"""Block device abstraction"""
|
23 |
|
24 |
import re |
25 |
import time |
26 |
import errno |
27 |
import stat |
28 |
import pyparsing as pyp |
29 |
import os |
30 |
import logging |
31 |
|
32 |
from ganeti import utils |
33 |
from ganeti import errors |
34 |
from ganeti import constants |
35 |
from ganeti import objects |
36 |
from ganeti import compat |
37 |
from ganeti import netutils |
38 |
|
39 |
|
40 |
# Size of reads in _CanReadDevice
|
41 |
_DEVICE_READ_SIZE = 128 * 1024 |
42 |
|
43 |
|
44 |
def _IgnoreError(fn, *args, **kwargs): |
45 |
"""Executes the given function, ignoring BlockDeviceErrors.
|
46 |
|
47 |
This is used in order to simplify the execution of cleanup or
|
48 |
rollback functions.
|
49 |
|
50 |
@rtype: boolean
|
51 |
@return: True when fn didn't raise an exception, False otherwise
|
52 |
|
53 |
"""
|
54 |
try:
|
55 |
fn(*args, **kwargs) |
56 |
return True |
57 |
except errors.BlockDeviceError, err:
|
58 |
logging.warning("Caught BlockDeviceError but ignoring: %s", str(err)) |
59 |
return False |
60 |
|
61 |
|
62 |
def _ThrowError(msg, *args): |
63 |
"""Log an error to the node daemon and the raise an exception.
|
64 |
|
65 |
@type msg: string
|
66 |
@param msg: the text of the exception
|
67 |
@raise errors.BlockDeviceError
|
68 |
|
69 |
"""
|
70 |
if args:
|
71 |
msg = msg % args |
72 |
logging.error(msg) |
73 |
raise errors.BlockDeviceError(msg)
|
74 |
|
75 |
|
76 |
def _CanReadDevice(path): |
77 |
"""Check if we can read from the given device.
|
78 |
|
79 |
This tries to read the first 128k of the device.
|
80 |
|
81 |
"""
|
82 |
try:
|
83 |
utils.ReadFile(path, size=_DEVICE_READ_SIZE) |
84 |
return True |
85 |
except EnvironmentError: |
86 |
logging.warning("Can't read from device %s", path, exc_info=True) |
87 |
return False |
88 |
|
89 |
|
90 |
class BlockDev(object): |
91 |
"""Block device abstract class.
|
92 |
|
93 |
A block device can be in the following states:
|
94 |
- not existing on the system, and by `Create()` it goes into:
|
95 |
- existing but not setup/not active, and by `Assemble()` goes into:
|
96 |
- active read-write and by `Open()` it goes into
|
97 |
- online (=used, or ready for use)
|
98 |
|
99 |
A device can also be online but read-only, however we are not using
|
100 |
the readonly state (LV has it, if needed in the future) and we are
|
101 |
usually looking at this like at a stack, so it's easier to
|
102 |
conceptualise the transition from not-existing to online and back
|
103 |
like a linear one.
|
104 |
|
105 |
The many different states of the device are due to the fact that we
|
106 |
need to cover many device types:
|
107 |
- logical volumes are created, lvchange -a y $lv, and used
|
108 |
- drbd devices are attached to a local disk/remote peer and made primary
|
109 |
|
110 |
A block device is identified by three items:
|
111 |
- the /dev path of the device (dynamic)
|
112 |
- a unique ID of the device (static)
|
113 |
- it's major/minor pair (dynamic)
|
114 |
|
115 |
Not all devices implement both the first two as distinct items. LVM
|
116 |
logical volumes have their unique ID (the pair volume group, logical
|
117 |
volume name) in a 1-to-1 relation to the dev path. For DRBD devices,
|
118 |
the /dev path is again dynamic and the unique id is the pair (host1,
|
119 |
dev1), (host2, dev2).
|
120 |
|
121 |
You can get to a device in two ways:
|
122 |
- creating the (real) device, which returns you
|
123 |
an attached instance (lvcreate)
|
124 |
- attaching of a python instance to an existing (real) device
|
125 |
|
126 |
The second point, the attachement to a device, is different
|
127 |
depending on whether the device is assembled or not. At init() time,
|
128 |
we search for a device with the same unique_id as us. If found,
|
129 |
good. It also means that the device is already assembled. If not,
|
130 |
after assembly we'll have our correct major/minor.
|
131 |
|
132 |
"""
|
133 |
def __init__(self, unique_id, children, size): |
134 |
self._children = children
|
135 |
self.dev_path = None |
136 |
self.unique_id = unique_id
|
137 |
self.major = None |
138 |
self.minor = None |
139 |
self.attached = False |
140 |
self.size = size
|
141 |
|
142 |
def Assemble(self): |
143 |
"""Assemble the device from its components.
|
144 |
|
145 |
Implementations of this method by child classes must ensure that:
|
146 |
- after the device has been assembled, it knows its major/minor
|
147 |
numbers; this allows other devices (usually parents) to probe
|
148 |
correctly for their children
|
149 |
- calling this method on an existing, in-use device is safe
|
150 |
- if the device is already configured (and in an OK state),
|
151 |
this method is idempotent
|
152 |
|
153 |
"""
|
154 |
pass
|
155 |
|
156 |
def Attach(self): |
157 |
"""Find a device which matches our config and attach to it.
|
158 |
|
159 |
"""
|
160 |
raise NotImplementedError |
161 |
|
162 |
def Close(self): |
163 |
"""Notifies that the device will no longer be used for I/O.
|
164 |
|
165 |
"""
|
166 |
raise NotImplementedError |
167 |
|
168 |
@classmethod
|
169 |
def Create(cls, unique_id, children, size): |
170 |
"""Create the device.
|
171 |
|
172 |
If the device cannot be created, it will return None
|
173 |
instead. Error messages go to the logging system.
|
174 |
|
175 |
Note that for some devices, the unique_id is used, and for other,
|
176 |
the children. The idea is that these two, taken together, are
|
177 |
enough for both creation and assembly (later).
|
178 |
|
179 |
"""
|
180 |
raise NotImplementedError |
181 |
|
182 |
def Remove(self): |
183 |
"""Remove this device.
|
184 |
|
185 |
This makes sense only for some of the device types: LV and file
|
186 |
storage. Also note that if the device can't attach, the removal
|
187 |
can't be completed.
|
188 |
|
189 |
"""
|
190 |
raise NotImplementedError |
191 |
|
192 |
def Rename(self, new_id): |
193 |
"""Rename this device.
|
194 |
|
195 |
This may or may not make sense for a given device type.
|
196 |
|
197 |
"""
|
198 |
raise NotImplementedError |
199 |
|
200 |
def Open(self, force=False): |
201 |
"""Make the device ready for use.
|
202 |
|
203 |
This makes the device ready for I/O. For now, just the DRBD
|
204 |
devices need this.
|
205 |
|
206 |
The force parameter signifies that if the device has any kind of
|
207 |
--force thing, it should be used, we know what we are doing.
|
208 |
|
209 |
"""
|
210 |
raise NotImplementedError |
211 |
|
212 |
def Shutdown(self): |
213 |
"""Shut down the device, freeing its children.
|
214 |
|
215 |
This undoes the `Assemble()` work, except for the child
|
216 |
assembling; as such, the children on the device are still
|
217 |
assembled after this call.
|
218 |
|
219 |
"""
|
220 |
raise NotImplementedError |
221 |
|
222 |
def SetSyncSpeed(self, speed): |
223 |
"""Adjust the sync speed of the mirror.
|
224 |
|
225 |
In case this is not a mirroring device, this is no-op.
|
226 |
|
227 |
"""
|
228 |
result = True
|
229 |
if self._children: |
230 |
for child in self._children: |
231 |
result = result and child.SetSyncSpeed(speed)
|
232 |
return result
|
233 |
|
234 |
def PauseResumeSync(self, pause): |
235 |
"""Pause/Resume the sync of the mirror.
|
236 |
|
237 |
In case this is not a mirroring device, this is no-op.
|
238 |
|
239 |
@param pause: Wheater to pause or resume
|
240 |
|
241 |
"""
|
242 |
result = True
|
243 |
if self._children: |
244 |
for child in self._children: |
245 |
result = result and child.PauseResumeSync(pause)
|
246 |
return result
|
247 |
|
248 |
def GetSyncStatus(self): |
249 |
"""Returns the sync status of the device.
|
250 |
|
251 |
If this device is a mirroring device, this function returns the
|
252 |
status of the mirror.
|
253 |
|
254 |
If sync_percent is None, it means the device is not syncing.
|
255 |
|
256 |
If estimated_time is None, it means we can't estimate
|
257 |
the time needed, otherwise it's the time left in seconds.
|
258 |
|
259 |
If is_degraded is True, it means the device is missing
|
260 |
redundancy. This is usually a sign that something went wrong in
|
261 |
the device setup, if sync_percent is None.
|
262 |
|
263 |
The ldisk parameter represents the degradation of the local
|
264 |
data. This is only valid for some devices, the rest will always
|
265 |
return False (not degraded).
|
266 |
|
267 |
@rtype: objects.BlockDevStatus
|
268 |
|
269 |
"""
|
270 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
271 |
major=self.major,
|
272 |
minor=self.minor,
|
273 |
sync_percent=None,
|
274 |
estimated_time=None,
|
275 |
is_degraded=False,
|
276 |
ldisk_status=constants.LDS_OKAY) |
277 |
|
278 |
def CombinedSyncStatus(self): |
279 |
"""Calculate the mirror status recursively for our children.
|
280 |
|
281 |
The return value is the same as for `GetSyncStatus()` except the
|
282 |
minimum percent and maximum time are calculated across our
|
283 |
children.
|
284 |
|
285 |
@rtype: objects.BlockDevStatus
|
286 |
|
287 |
"""
|
288 |
status = self.GetSyncStatus()
|
289 |
|
290 |
min_percent = status.sync_percent |
291 |
max_time = status.estimated_time |
292 |
is_degraded = status.is_degraded |
293 |
ldisk_status = status.ldisk_status |
294 |
|
295 |
if self._children: |
296 |
for child in self._children: |
297 |
child_status = child.GetSyncStatus() |
298 |
|
299 |
if min_percent is None: |
300 |
min_percent = child_status.sync_percent |
301 |
elif child_status.sync_percent is not None: |
302 |
min_percent = min(min_percent, child_status.sync_percent)
|
303 |
|
304 |
if max_time is None: |
305 |
max_time = child_status.estimated_time |
306 |
elif child_status.estimated_time is not None: |
307 |
max_time = max(max_time, child_status.estimated_time)
|
308 |
|
309 |
is_degraded = is_degraded or child_status.is_degraded
|
310 |
|
311 |
if ldisk_status is None: |
312 |
ldisk_status = child_status.ldisk_status |
313 |
elif child_status.ldisk_status is not None: |
314 |
ldisk_status = max(ldisk_status, child_status.ldisk_status)
|
315 |
|
316 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
317 |
major=self.major,
|
318 |
minor=self.minor,
|
319 |
sync_percent=min_percent, |
320 |
estimated_time=max_time, |
321 |
is_degraded=is_degraded, |
322 |
ldisk_status=ldisk_status) |
323 |
|
324 |
|
325 |
def SetInfo(self, text): |
326 |
"""Update metadata with info text.
|
327 |
|
328 |
Only supported for some device types.
|
329 |
|
330 |
"""
|
331 |
for child in self._children: |
332 |
child.SetInfo(text) |
333 |
|
334 |
def Grow(self, amount): |
335 |
"""Grow the block device.
|
336 |
|
337 |
@param amount: the amount (in mebibytes) to grow with
|
338 |
|
339 |
"""
|
340 |
raise NotImplementedError |
341 |
|
342 |
def GetActualSize(self): |
343 |
"""Return the actual disk size.
|
344 |
|
345 |
@note: the device needs to be active when this is called
|
346 |
|
347 |
"""
|
348 |
assert self.attached, "BlockDevice not attached in GetActualSize()" |
349 |
result = utils.RunCmd(["blockdev", "--getsize64", self.dev_path]) |
350 |
if result.failed:
|
351 |
_ThrowError("blockdev failed (%s): %s",
|
352 |
result.fail_reason, result.output) |
353 |
try:
|
354 |
sz = int(result.output.strip())
|
355 |
except (ValueError, TypeError), err: |
356 |
_ThrowError("Failed to parse blockdev output: %s", str(err)) |
357 |
return sz
|
358 |
|
359 |
def __repr__(self): |
360 |
return ("<%s: unique_id: %s, children: %s, %s:%s, %s>" % |
361 |
(self.__class__, self.unique_id, self._children, |
362 |
self.major, self.minor, self.dev_path)) |
363 |
|
364 |
|
365 |
class LogicalVolume(BlockDev): |
366 |
"""Logical Volume block device.
|
367 |
|
368 |
"""
|
369 |
_VALID_NAME_RE = re.compile("^[a-zA-Z0-9+_.-]*$")
|
370 |
_INVALID_NAMES = frozenset([".", "..", "snapshot", "pvmove"]) |
371 |
_INVALID_SUBSTRINGS = frozenset(["_mlog", "_mimage"]) |
372 |
|
373 |
def __init__(self, unique_id, children, size): |
374 |
"""Attaches to a LV device.
|
375 |
|
376 |
The unique_id is a tuple (vg_name, lv_name)
|
377 |
|
378 |
"""
|
379 |
super(LogicalVolume, self).__init__(unique_id, children, size) |
380 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
381 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
382 |
self._vg_name, self._lv_name = unique_id |
383 |
self._ValidateName(self._vg_name) |
384 |
self._ValidateName(self._lv_name) |
385 |
self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name) |
386 |
self._degraded = True |
387 |
self.major = self.minor = self.pe_size = self.stripe_count = None |
388 |
self.Attach()
|
389 |
|
390 |
@classmethod
|
391 |
def Create(cls, unique_id, children, size): |
392 |
"""Create a new logical volume.
|
393 |
|
394 |
"""
|
395 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
396 |
raise errors.ProgrammerError("Invalid configuration data %s" % |
397 |
str(unique_id))
|
398 |
vg_name, lv_name = unique_id |
399 |
cls._ValidateName(vg_name) |
400 |
cls._ValidateName(lv_name) |
401 |
pvs_info = cls.GetPVInfo([vg_name]) |
402 |
if not pvs_info: |
403 |
_ThrowError("Can't compute PV info for vg %s", vg_name)
|
404 |
pvs_info.sort() |
405 |
pvs_info.reverse() |
406 |
|
407 |
pvlist = [ pv[1] for pv in pvs_info ] |
408 |
if compat.any(":" in v for v in pvlist): |
409 |
_ThrowError("Some of your PVs have the invalid character ':' in their"
|
410 |
" name, this is not supported - please filter them out"
|
411 |
" in lvm.conf using either 'filter' or 'preferred_names'")
|
412 |
free_size = sum([ pv[0] for pv in pvs_info ]) |
413 |
current_pvs = len(pvlist)
|
414 |
stripes = min(current_pvs, constants.LVM_STRIPECOUNT)
|
415 |
|
416 |
# The size constraint should have been checked from the master before
|
417 |
# calling the create function.
|
418 |
if free_size < size:
|
419 |
_ThrowError("Not enough free space: required %s,"
|
420 |
" available %s", size, free_size)
|
421 |
cmd = ["lvcreate", "-L%dm" % size, "-n%s" % lv_name] |
422 |
# If the free space is not well distributed, we won't be able to
|
423 |
# create an optimally-striped volume; in that case, we want to try
|
424 |
# with N, N-1, ..., 2, and finally 1 (non-stripped) number of
|
425 |
# stripes
|
426 |
for stripes_arg in range(stripes, 0, -1): |
427 |
result = utils.RunCmd(cmd + ["-i%d" % stripes_arg] + [vg_name] + pvlist)
|
428 |
if not result.failed: |
429 |
break
|
430 |
if result.failed:
|
431 |
_ThrowError("LV create failed (%s): %s",
|
432 |
result.fail_reason, result.output) |
433 |
return LogicalVolume(unique_id, children, size)
|
434 |
|
435 |
@staticmethod
|
436 |
def _GetVolumeInfo(lvm_cmd, fields): |
437 |
"""Returns LVM Volumen infos using lvm_cmd
|
438 |
|
439 |
@param lvm_cmd: Should be one of "pvs", "vgs" or "lvs"
|
440 |
@param fields: Fields to return
|
441 |
@return: A list of dicts each with the parsed fields
|
442 |
|
443 |
"""
|
444 |
if not fields: |
445 |
raise errors.ProgrammerError("No fields specified") |
446 |
|
447 |
sep = "|"
|
448 |
cmd = [lvm_cmd, "--noheadings", "--nosuffix", "--units=m", "--unbuffered", |
449 |
"--separator=%s" % sep, "-o%s" % ",".join(fields)] |
450 |
|
451 |
result = utils.RunCmd(cmd) |
452 |
if result.failed:
|
453 |
raise errors.CommandError("Can't get the volume information: %s - %s" % |
454 |
(result.fail_reason, result.output)) |
455 |
|
456 |
data = [] |
457 |
for line in result.stdout.splitlines(): |
458 |
splitted_fields = line.strip().split(sep) |
459 |
|
460 |
if len(fields) != len(splitted_fields): |
461 |
raise errors.CommandError("Can't parse %s output: line '%s'" % |
462 |
(lvm_cmd, line)) |
463 |
|
464 |
data.append(splitted_fields) |
465 |
|
466 |
return data
|
467 |
|
468 |
@classmethod
|
469 |
def GetPVInfo(cls, vg_names, filter_allocatable=True): |
470 |
"""Get the free space info for PVs in a volume group.
|
471 |
|
472 |
@param vg_names: list of volume group names, if empty all will be returned
|
473 |
@param filter_allocatable: whether to skip over unallocatable PVs
|
474 |
|
475 |
@rtype: list
|
476 |
@return: list of tuples (free_space, name) with free_space in mebibytes
|
477 |
|
478 |
"""
|
479 |
try:
|
480 |
info = cls._GetVolumeInfo("pvs", ["pv_name", "vg_name", "pv_free", |
481 |
"pv_attr"])
|
482 |
except errors.GenericError, err:
|
483 |
logging.error("Can't get PV information: %s", err)
|
484 |
return None |
485 |
|
486 |
data = [] |
487 |
for pv_name, vg_name, pv_free, pv_attr in info: |
488 |
# (possibly) skip over pvs which are not allocatable
|
489 |
if filter_allocatable and pv_attr[0] != "a": |
490 |
continue
|
491 |
# (possibly) skip over pvs which are not in the right volume group(s)
|
492 |
if vg_names and vg_name not in vg_names: |
493 |
continue
|
494 |
data.append((float(pv_free), pv_name, vg_name))
|
495 |
|
496 |
return data
|
497 |
|
498 |
@classmethod
|
499 |
def GetVGInfo(cls, vg_names, filter_readonly=True): |
500 |
"""Get the free space info for specific VGs.
|
501 |
|
502 |
@param vg_names: list of volume group names, if empty all will be returned
|
503 |
@param filter_readonly: whether to skip over readonly VGs
|
504 |
|
505 |
@rtype: list
|
506 |
@return: list of tuples (free_space, total_size, name) with free_space in
|
507 |
MiB
|
508 |
|
509 |
"""
|
510 |
try:
|
511 |
info = cls._GetVolumeInfo("vgs", ["vg_name", "vg_free", "vg_attr", |
512 |
"vg_size"])
|
513 |
except errors.GenericError, err:
|
514 |
logging.error("Can't get VG information: %s", err)
|
515 |
return None |
516 |
|
517 |
data = [] |
518 |
for vg_name, vg_free, vg_attr, vg_size in info: |
519 |
# (possibly) skip over vgs which are not writable
|
520 |
if filter_readonly and vg_attr[0] == "r": |
521 |
continue
|
522 |
# (possibly) skip over vgs which are not in the right volume group(s)
|
523 |
if vg_names and vg_name not in vg_names: |
524 |
continue
|
525 |
data.append((float(vg_free), float(vg_size), vg_name)) |
526 |
|
527 |
return data
|
528 |
|
529 |
@classmethod
|
530 |
def _ValidateName(cls, name): |
531 |
"""Validates that a given name is valid as VG or LV name.
|
532 |
|
533 |
The list of valid characters and restricted names is taken out of
|
534 |
the lvm(8) manpage, with the simplification that we enforce both
|
535 |
VG and LV restrictions on the names.
|
536 |
|
537 |
"""
|
538 |
if (not cls._VALID_NAME_RE.match(name) or |
539 |
name in cls._INVALID_NAMES or |
540 |
compat.any(substring in name for substring in cls._INVALID_SUBSTRINGS)): |
541 |
_ThrowError("Invalid LVM name '%s'", name)
|
542 |
|
543 |
def Remove(self): |
544 |
"""Remove this logical volume.
|
545 |
|
546 |
"""
|
547 |
if not self.minor and not self.Attach(): |
548 |
# the LV does not exist
|
549 |
return
|
550 |
result = utils.RunCmd(["lvremove", "-f", "%s/%s" % |
551 |
(self._vg_name, self._lv_name)]) |
552 |
if result.failed:
|
553 |
_ThrowError("Can't lvremove: %s - %s", result.fail_reason, result.output)
|
554 |
|
555 |
def Rename(self, new_id): |
556 |
"""Rename this logical volume.
|
557 |
|
558 |
"""
|
559 |
if not isinstance(new_id, (tuple, list)) or len(new_id) != 2: |
560 |
raise errors.ProgrammerError("Invalid new logical id '%s'" % new_id) |
561 |
new_vg, new_name = new_id |
562 |
if new_vg != self._vg_name: |
563 |
raise errors.ProgrammerError("Can't move a logical volume across" |
564 |
" volume groups (from %s to to %s)" %
|
565 |
(self._vg_name, new_vg))
|
566 |
result = utils.RunCmd(["lvrename", new_vg, self._lv_name, new_name]) |
567 |
if result.failed:
|
568 |
_ThrowError("Failed to rename the logical volume: %s", result.output)
|
569 |
self._lv_name = new_name
|
570 |
self.dev_path = utils.PathJoin("/dev", self._vg_name, self._lv_name) |
571 |
|
572 |
def Attach(self): |
573 |
"""Attach to an existing LV.
|
574 |
|
575 |
This method will try to see if an existing and active LV exists
|
576 |
which matches our name. If so, its major/minor will be
|
577 |
recorded.
|
578 |
|
579 |
"""
|
580 |
self.attached = False |
581 |
result = utils.RunCmd(["lvs", "--noheadings", "--separator=,", |
582 |
"--units=m", "--nosuffix", |
583 |
"-olv_attr,lv_kernel_major,lv_kernel_minor,"
|
584 |
"vg_extent_size,stripes", self.dev_path]) |
585 |
if result.failed:
|
586 |
logging.error("Can't find LV %s: %s, %s",
|
587 |
self.dev_path, result.fail_reason, result.output)
|
588 |
return False |
589 |
# the output can (and will) have multiple lines for multi-segment
|
590 |
# LVs, as the 'stripes' parameter is a segment one, so we take
|
591 |
# only the last entry, which is the one we're interested in; note
|
592 |
# that with LVM2 anyway the 'stripes' value must be constant
|
593 |
# across segments, so this is a no-op actually
|
594 |
out = result.stdout.splitlines() |
595 |
if not out: # totally empty result? splitlines() returns at least |
596 |
# one line for any non-empty string
|
597 |
logging.error("Can't parse LVS output, no lines? Got '%s'", str(out)) |
598 |
return False |
599 |
out = out[-1].strip().rstrip(',') |
600 |
out = out.split(",")
|
601 |
if len(out) != 5: |
602 |
logging.error("Can't parse LVS output, len(%s) != 5", str(out)) |
603 |
return False |
604 |
|
605 |
status, major, minor, pe_size, stripes = out |
606 |
if len(status) != 6: |
607 |
logging.error("lvs lv_attr is not 6 characters (%s)", status)
|
608 |
return False |
609 |
|
610 |
try:
|
611 |
major = int(major)
|
612 |
minor = int(minor)
|
613 |
except (TypeError, ValueError), err: |
614 |
logging.error("lvs major/minor cannot be parsed: %s", str(err)) |
615 |
|
616 |
try:
|
617 |
pe_size = int(float(pe_size)) |
618 |
except (TypeError, ValueError), err: |
619 |
logging.error("Can't parse vg extent size: %s", err)
|
620 |
return False |
621 |
|
622 |
try:
|
623 |
stripes = int(stripes)
|
624 |
except (TypeError, ValueError), err: |
625 |
logging.error("Can't parse the number of stripes: %s", err)
|
626 |
return False |
627 |
|
628 |
self.major = major
|
629 |
self.minor = minor
|
630 |
self.pe_size = pe_size
|
631 |
self.stripe_count = stripes
|
632 |
self._degraded = status[0] == 'v' # virtual volume, i.e. doesn't backing |
633 |
# storage
|
634 |
self.attached = True |
635 |
return True |
636 |
|
637 |
def Assemble(self): |
638 |
"""Assemble the device.
|
639 |
|
640 |
We always run `lvchange -ay` on the LV to ensure it's active before
|
641 |
use, as there were cases when xenvg was not active after boot
|
642 |
(also possibly after disk issues).
|
643 |
|
644 |
"""
|
645 |
result = utils.RunCmd(["lvchange", "-ay", self.dev_path]) |
646 |
if result.failed:
|
647 |
_ThrowError("Can't activate lv %s: %s", self.dev_path, result.output) |
648 |
|
649 |
def Shutdown(self): |
650 |
"""Shutdown the device.
|
651 |
|
652 |
This is a no-op for the LV device type, as we don't deactivate the
|
653 |
volumes on shutdown.
|
654 |
|
655 |
"""
|
656 |
pass
|
657 |
|
658 |
def GetSyncStatus(self): |
659 |
"""Returns the sync status of the device.
|
660 |
|
661 |
If this device is a mirroring device, this function returns the
|
662 |
status of the mirror.
|
663 |
|
664 |
For logical volumes, sync_percent and estimated_time are always
|
665 |
None (no recovery in progress, as we don't handle the mirrored LV
|
666 |
case). The is_degraded parameter is the inverse of the ldisk
|
667 |
parameter.
|
668 |
|
669 |
For the ldisk parameter, we check if the logical volume has the
|
670 |
'virtual' type, which means it's not backed by existing storage
|
671 |
anymore (read from it return I/O error). This happens after a
|
672 |
physical disk failure and subsequent 'vgreduce --removemissing' on
|
673 |
the volume group.
|
674 |
|
675 |
The status was already read in Attach, so we just return it.
|
676 |
|
677 |
@rtype: objects.BlockDevStatus
|
678 |
|
679 |
"""
|
680 |
if self._degraded: |
681 |
ldisk_status = constants.LDS_FAULTY |
682 |
else:
|
683 |
ldisk_status = constants.LDS_OKAY |
684 |
|
685 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
686 |
major=self.major,
|
687 |
minor=self.minor,
|
688 |
sync_percent=None,
|
689 |
estimated_time=None,
|
690 |
is_degraded=self._degraded,
|
691 |
ldisk_status=ldisk_status) |
692 |
|
693 |
def Open(self, force=False): |
694 |
"""Make the device ready for I/O.
|
695 |
|
696 |
This is a no-op for the LV device type.
|
697 |
|
698 |
"""
|
699 |
pass
|
700 |
|
701 |
def Close(self): |
702 |
"""Notifies that the device will no longer be used for I/O.
|
703 |
|
704 |
This is a no-op for the LV device type.
|
705 |
|
706 |
"""
|
707 |
pass
|
708 |
|
709 |
def Snapshot(self, size): |
710 |
"""Create a snapshot copy of an lvm block device.
|
711 |
|
712 |
@returns: tuple (vg, lv)
|
713 |
|
714 |
"""
|
715 |
snap_name = self._lv_name + ".snap" |
716 |
|
717 |
# remove existing snapshot if found
|
718 |
snap = LogicalVolume((self._vg_name, snap_name), None, size) |
719 |
_IgnoreError(snap.Remove) |
720 |
|
721 |
vg_info = self.GetVGInfo([self._vg_name]) |
722 |
if not vg_info: |
723 |
_ThrowError("Can't compute VG info for vg %s", self._vg_name) |
724 |
free_size, _, _ = vg_info[0]
|
725 |
if free_size < size:
|
726 |
_ThrowError("Not enough free space: required %s,"
|
727 |
" available %s", size, free_size)
|
728 |
|
729 |
result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-s", |
730 |
"-n%s" % snap_name, self.dev_path]) |
731 |
if result.failed:
|
732 |
_ThrowError("command: %s error: %s - %s",
|
733 |
result.cmd, result.fail_reason, result.output) |
734 |
|
735 |
return (self._vg_name, snap_name) |
736 |
|
737 |
def SetInfo(self, text): |
738 |
"""Update metadata with info text.
|
739 |
|
740 |
"""
|
741 |
BlockDev.SetInfo(self, text)
|
742 |
|
743 |
# Replace invalid characters
|
744 |
text = re.sub('^[^A-Za-z0-9_+.]', '_', text) |
745 |
text = re.sub('[^-A-Za-z0-9_+.]', '_', text) |
746 |
|
747 |
# Only up to 128 characters are allowed
|
748 |
text = text[:128]
|
749 |
|
750 |
result = utils.RunCmd(["lvchange", "--addtag", text, |
751 |
self.dev_path])
|
752 |
if result.failed:
|
753 |
_ThrowError("Command: %s error: %s - %s", result.cmd, result.fail_reason,
|
754 |
result.output) |
755 |
|
756 |
def Grow(self, amount): |
757 |
"""Grow the logical volume.
|
758 |
|
759 |
"""
|
760 |
if self.pe_size is None or self.stripe_count is None: |
761 |
if not self.Attach(): |
762 |
_ThrowError("Can't attach to LV during Grow()")
|
763 |
full_stripe_size = self.pe_size * self.stripe_count |
764 |
rest = amount % full_stripe_size |
765 |
if rest != 0: |
766 |
amount += full_stripe_size - rest |
767 |
# we try multiple algorithms since the 'best' ones might not have
|
768 |
# space available in the right place, but later ones might (since
|
769 |
# they have less constraints); also note that only recent LVM
|
770 |
# supports 'cling'
|
771 |
for alloc_policy in "contiguous", "cling", "normal": |
772 |
result = utils.RunCmd(["lvextend", "--alloc", alloc_policy, |
773 |
"-L", "+%dm" % amount, self.dev_path]) |
774 |
if not result.failed: |
775 |
return
|
776 |
_ThrowError("Can't grow LV %s: %s", self.dev_path, result.output) |
777 |
|
778 |
|
779 |
class DRBD8Status(object): |
780 |
"""A DRBD status representation class.
|
781 |
|
782 |
Note that this doesn't support unconfigured devices (cs:Unconfigured).
|
783 |
|
784 |
"""
|
785 |
UNCONF_RE = re.compile(r"\s*[0-9]+:\s*cs:Unconfigured$")
|
786 |
LINE_RE = re.compile(r"\s*[0-9]+:\s*cs:(\S+)\s+(?:st|ro):([^/]+)/(\S+)"
|
787 |
"\s+ds:([^/]+)/(\S+)\s+.*$")
|
788 |
SYNC_RE = re.compile(r"^.*\ssync'ed:\s*([0-9.]+)%.*"
|
789 |
# Due to a bug in drbd in the kernel, introduced in
|
790 |
# commit 4b0715f096 (still unfixed as of 2011-08-22)
|
791 |
"(?:\s|M)"
|
792 |
"finish: ([0-9]+):([0-9]+):([0-9]+)\s.*$")
|
793 |
|
794 |
CS_UNCONFIGURED = "Unconfigured"
|
795 |
CS_STANDALONE = "StandAlone"
|
796 |
CS_WFCONNECTION = "WFConnection"
|
797 |
CS_WFREPORTPARAMS = "WFReportParams"
|
798 |
CS_CONNECTED = "Connected"
|
799 |
CS_STARTINGSYNCS = "StartingSyncS"
|
800 |
CS_STARTINGSYNCT = "StartingSyncT"
|
801 |
CS_WFBITMAPS = "WFBitMapS"
|
802 |
CS_WFBITMAPT = "WFBitMapT"
|
803 |
CS_WFSYNCUUID = "WFSyncUUID"
|
804 |
CS_SYNCSOURCE = "SyncSource"
|
805 |
CS_SYNCTARGET = "SyncTarget"
|
806 |
CS_PAUSEDSYNCS = "PausedSyncS"
|
807 |
CS_PAUSEDSYNCT = "PausedSyncT"
|
808 |
CSET_SYNC = frozenset([
|
809 |
CS_WFREPORTPARAMS, |
810 |
CS_STARTINGSYNCS, |
811 |
CS_STARTINGSYNCT, |
812 |
CS_WFBITMAPS, |
813 |
CS_WFBITMAPT, |
814 |
CS_WFSYNCUUID, |
815 |
CS_SYNCSOURCE, |
816 |
CS_SYNCTARGET, |
817 |
CS_PAUSEDSYNCS, |
818 |
CS_PAUSEDSYNCT, |
819 |
]) |
820 |
|
821 |
DS_DISKLESS = "Diskless"
|
822 |
DS_ATTACHING = "Attaching" # transient state |
823 |
DS_FAILED = "Failed" # transient state, next: diskless |
824 |
DS_NEGOTIATING = "Negotiating" # transient state |
825 |
DS_INCONSISTENT = "Inconsistent" # while syncing or after creation |
826 |
DS_OUTDATED = "Outdated"
|
827 |
DS_DUNKNOWN = "DUnknown" # shown for peer disk when not connected |
828 |
DS_CONSISTENT = "Consistent"
|
829 |
DS_UPTODATE = "UpToDate" # normal state |
830 |
|
831 |
RO_PRIMARY = "Primary"
|
832 |
RO_SECONDARY = "Secondary"
|
833 |
RO_UNKNOWN = "Unknown"
|
834 |
|
835 |
def __init__(self, procline): |
836 |
u = self.UNCONF_RE.match(procline)
|
837 |
if u:
|
838 |
self.cstatus = self.CS_UNCONFIGURED |
839 |
self.lrole = self.rrole = self.ldisk = self.rdisk = None |
840 |
else:
|
841 |
m = self.LINE_RE.match(procline)
|
842 |
if not m: |
843 |
raise errors.BlockDeviceError("Can't parse input data '%s'" % procline) |
844 |
self.cstatus = m.group(1) |
845 |
self.lrole = m.group(2) |
846 |
self.rrole = m.group(3) |
847 |
self.ldisk = m.group(4) |
848 |
self.rdisk = m.group(5) |
849 |
|
850 |
# end reading of data from the LINE_RE or UNCONF_RE
|
851 |
|
852 |
self.is_standalone = self.cstatus == self.CS_STANDALONE |
853 |
self.is_wfconn = self.cstatus == self.CS_WFCONNECTION |
854 |
self.is_connected = self.cstatus == self.CS_CONNECTED |
855 |
self.is_primary = self.lrole == self.RO_PRIMARY |
856 |
self.is_secondary = self.lrole == self.RO_SECONDARY |
857 |
self.peer_primary = self.rrole == self.RO_PRIMARY |
858 |
self.peer_secondary = self.rrole == self.RO_SECONDARY |
859 |
self.both_primary = self.is_primary and self.peer_primary |
860 |
self.both_secondary = self.is_secondary and self.peer_secondary |
861 |
|
862 |
self.is_diskless = self.ldisk == self.DS_DISKLESS |
863 |
self.is_disk_uptodate = self.ldisk == self.DS_UPTODATE |
864 |
|
865 |
self.is_in_resync = self.cstatus in self.CSET_SYNC |
866 |
self.is_in_use = self.cstatus != self.CS_UNCONFIGURED |
867 |
|
868 |
m = self.SYNC_RE.match(procline)
|
869 |
if m:
|
870 |
self.sync_percent = float(m.group(1)) |
871 |
hours = int(m.group(2)) |
872 |
minutes = int(m.group(3)) |
873 |
seconds = int(m.group(4)) |
874 |
self.est_time = hours * 3600 + minutes * 60 + seconds |
875 |
else:
|
876 |
# we have (in this if branch) no percent information, but if
|
877 |
# we're resyncing we need to 'fake' a sync percent information,
|
878 |
# as this is how cmdlib determines if it makes sense to wait for
|
879 |
# resyncing or not
|
880 |
if self.is_in_resync: |
881 |
self.sync_percent = 0 |
882 |
else:
|
883 |
self.sync_percent = None |
884 |
self.est_time = None |
885 |
|
886 |
|
887 |
class BaseDRBD(BlockDev): # pylint: disable-msg=W0223 |
888 |
"""Base DRBD class.
|
889 |
|
890 |
This class contains a few bits of common functionality between the
|
891 |
0.7 and 8.x versions of DRBD.
|
892 |
|
893 |
"""
|
894 |
_VERSION_RE = re.compile(r"^version: (\d+)\.(\d+)\.(\d+)(?:\.\d+)?"
|
895 |
r" \(api:(\d+)/proto:(\d+)(?:-(\d+))?\)")
|
896 |
_VALID_LINE_RE = re.compile("^ *([0-9]+): cs:([^ ]+).*$")
|
897 |
_UNUSED_LINE_RE = re.compile("^ *([0-9]+): cs:Unconfigured$")
|
898 |
|
899 |
_DRBD_MAJOR = 147
|
900 |
_ST_UNCONFIGURED = "Unconfigured"
|
901 |
_ST_WFCONNECTION = "WFConnection"
|
902 |
_ST_CONNECTED = "Connected"
|
903 |
|
904 |
_STATUS_FILE = "/proc/drbd"
|
905 |
_USERMODE_HELPER_FILE = "/sys/module/drbd/parameters/usermode_helper"
|
906 |
|
907 |
@staticmethod
|
908 |
def _GetProcData(filename=_STATUS_FILE): |
909 |
"""Return data from /proc/drbd.
|
910 |
|
911 |
"""
|
912 |
try:
|
913 |
data = utils.ReadFile(filename).splitlines() |
914 |
except EnvironmentError, err: |
915 |
if err.errno == errno.ENOENT:
|
916 |
_ThrowError("The file %s cannot be opened, check if the module"
|
917 |
" is loaded (%s)", filename, str(err)) |
918 |
else:
|
919 |
_ThrowError("Can't read the DRBD proc file %s: %s", filename, str(err)) |
920 |
if not data: |
921 |
_ThrowError("Can't read any data from %s", filename)
|
922 |
return data
|
923 |
|
924 |
@classmethod
|
925 |
def _MassageProcData(cls, data): |
926 |
"""Transform the output of _GetProdData into a nicer form.
|
927 |
|
928 |
@return: a dictionary of minor: joined lines from /proc/drbd
|
929 |
for that minor
|
930 |
|
931 |
"""
|
932 |
results = {} |
933 |
old_minor = old_line = None
|
934 |
for line in data: |
935 |
if not line: # completely empty lines, as can be returned by drbd8.0+ |
936 |
continue
|
937 |
lresult = cls._VALID_LINE_RE.match(line) |
938 |
if lresult is not None: |
939 |
if old_minor is not None: |
940 |
results[old_minor] = old_line |
941 |
old_minor = int(lresult.group(1)) |
942 |
old_line = line |
943 |
else:
|
944 |
if old_minor is not None: |
945 |
old_line += " " + line.strip()
|
946 |
# add last line
|
947 |
if old_minor is not None: |
948 |
results[old_minor] = old_line |
949 |
return results
|
950 |
|
951 |
@classmethod
|
952 |
def _GetVersion(cls, proc_data): |
953 |
"""Return the DRBD version.
|
954 |
|
955 |
This will return a dict with keys:
|
956 |
- k_major
|
957 |
- k_minor
|
958 |
- k_point
|
959 |
- api
|
960 |
- proto
|
961 |
- proto2 (only on drbd > 8.2.X)
|
962 |
|
963 |
"""
|
964 |
first_line = proc_data[0].strip()
|
965 |
version = cls._VERSION_RE.match(first_line) |
966 |
if not version: |
967 |
raise errors.BlockDeviceError("Can't parse DRBD version from '%s'" % |
968 |
first_line) |
969 |
|
970 |
values = version.groups() |
971 |
retval = {'k_major': int(values[0]), |
972 |
'k_minor': int(values[1]), |
973 |
'k_point': int(values[2]), |
974 |
'api': int(values[3]), |
975 |
'proto': int(values[4]), |
976 |
} |
977 |
if values[5] is not None: |
978 |
retval['proto2'] = values[5] |
979 |
|
980 |
return retval
|
981 |
|
982 |
@staticmethod
|
983 |
def GetUsermodeHelper(filename=_USERMODE_HELPER_FILE): |
984 |
"""Returns DRBD usermode_helper currently set.
|
985 |
|
986 |
"""
|
987 |
try:
|
988 |
helper = utils.ReadFile(filename).splitlines()[0]
|
989 |
except EnvironmentError, err: |
990 |
if err.errno == errno.ENOENT:
|
991 |
_ThrowError("The file %s cannot be opened, check if the module"
|
992 |
" is loaded (%s)", filename, str(err)) |
993 |
else:
|
994 |
_ThrowError("Can't read DRBD helper file %s: %s", filename, str(err)) |
995 |
if not helper: |
996 |
_ThrowError("Can't read any data from %s", filename)
|
997 |
return helper
|
998 |
|
999 |
@staticmethod
|
1000 |
def _DevPath(minor): |
1001 |
"""Return the path to a drbd device for a given minor.
|
1002 |
|
1003 |
"""
|
1004 |
return "/dev/drbd%d" % minor |
1005 |
|
1006 |
@classmethod
|
1007 |
def GetUsedDevs(cls): |
1008 |
"""Compute the list of used DRBD devices.
|
1009 |
|
1010 |
"""
|
1011 |
data = cls._GetProcData() |
1012 |
|
1013 |
used_devs = {} |
1014 |
for line in data: |
1015 |
match = cls._VALID_LINE_RE.match(line) |
1016 |
if not match: |
1017 |
continue
|
1018 |
minor = int(match.group(1)) |
1019 |
state = match.group(2)
|
1020 |
if state == cls._ST_UNCONFIGURED:
|
1021 |
continue
|
1022 |
used_devs[minor] = state, line |
1023 |
|
1024 |
return used_devs
|
1025 |
|
1026 |
def _SetFromMinor(self, minor): |
1027 |
"""Set our parameters based on the given minor.
|
1028 |
|
1029 |
This sets our minor variable and our dev_path.
|
1030 |
|
1031 |
"""
|
1032 |
if minor is None: |
1033 |
self.minor = self.dev_path = None |
1034 |
self.attached = False |
1035 |
else:
|
1036 |
self.minor = minor
|
1037 |
self.dev_path = self._DevPath(minor) |
1038 |
self.attached = True |
1039 |
|
1040 |
@staticmethod
|
1041 |
def _CheckMetaSize(meta_device): |
1042 |
"""Check if the given meta device looks like a valid one.
|
1043 |
|
1044 |
This currently only check the size, which must be around
|
1045 |
128MiB.
|
1046 |
|
1047 |
"""
|
1048 |
result = utils.RunCmd(["blockdev", "--getsize", meta_device]) |
1049 |
if result.failed:
|
1050 |
_ThrowError("Failed to get device size: %s - %s",
|
1051 |
result.fail_reason, result.output) |
1052 |
try:
|
1053 |
sectors = int(result.stdout)
|
1054 |
except (TypeError, ValueError): |
1055 |
_ThrowError("Invalid output from blockdev: '%s'", result.stdout)
|
1056 |
num_bytes = sectors * 512
|
1057 |
if num_bytes < 128 * 1024 * 1024: # less than 128MiB |
1058 |
_ThrowError("Meta device too small (%.2fMib)", (num_bytes / 1024 / 1024)) |
1059 |
# the maximum *valid* size of the meta device when living on top
|
1060 |
# of LVM is hard to compute: it depends on the number of stripes
|
1061 |
# and the PE size; e.g. a 2-stripe, 64MB PE will result in a 128MB
|
1062 |
# (normal size), but an eight-stripe 128MB PE will result in a 1GB
|
1063 |
# size meta device; as such, we restrict it to 1GB (a little bit
|
1064 |
# too generous, but making assumptions about PE size is hard)
|
1065 |
if num_bytes > 1024 * 1024 * 1024: |
1066 |
_ThrowError("Meta device too big (%.2fMiB)", (num_bytes / 1024 / 1024)) |
1067 |
|
1068 |
def Rename(self, new_id): |
1069 |
"""Rename a device.
|
1070 |
|
1071 |
This is not supported for drbd devices.
|
1072 |
|
1073 |
"""
|
1074 |
raise errors.ProgrammerError("Can't rename a drbd device") |
1075 |
|
1076 |
|
1077 |
class DRBD8(BaseDRBD): |
1078 |
"""DRBD v8.x block device.
|
1079 |
|
1080 |
This implements the local host part of the DRBD device, i.e. it
|
1081 |
doesn't do anything to the supposed peer. If you need a fully
|
1082 |
connected DRBD pair, you need to use this class on both hosts.
|
1083 |
|
1084 |
The unique_id for the drbd device is the (local_ip, local_port,
|
1085 |
remote_ip, remote_port) tuple, and it must have two children: the
|
1086 |
data device and the meta_device. The meta device is checked for
|
1087 |
valid size and is zeroed on create.
|
1088 |
|
1089 |
"""
|
1090 |
_MAX_MINORS = 255
|
1091 |
_PARSE_SHOW = None
|
1092 |
|
1093 |
# timeout constants
|
1094 |
_NET_RECONFIG_TIMEOUT = 60
|
1095 |
|
1096 |
def __init__(self, unique_id, children, size): |
1097 |
if children and children.count(None) > 0: |
1098 |
children = [] |
1099 |
if len(children) not in (0, 2): |
1100 |
raise ValueError("Invalid configuration data %s" % str(children)) |
1101 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 6: |
1102 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
1103 |
(self._lhost, self._lport, |
1104 |
self._rhost, self._rport, |
1105 |
self._aminor, self._secret) = unique_id |
1106 |
if children:
|
1107 |
if not _CanReadDevice(children[1].dev_path): |
1108 |
logging.info("drbd%s: Ignoring unreadable meta device", self._aminor) |
1109 |
children = [] |
1110 |
super(DRBD8, self).__init__(unique_id, children, size) |
1111 |
self.major = self._DRBD_MAJOR |
1112 |
version = self._GetVersion(self._GetProcData()) |
1113 |
if version['k_major'] != 8 : |
1114 |
_ThrowError("Mismatch in DRBD kernel version and requested ganeti"
|
1115 |
" usage: kernel is %s.%s, ganeti wants 8.x",
|
1116 |
version['k_major'], version['k_minor']) |
1117 |
|
1118 |
if (self._lhost is not None and self._lhost == self._rhost and |
1119 |
self._lport == self._rport): |
1120 |
raise ValueError("Invalid configuration data, same local/remote %s" % |
1121 |
(unique_id,)) |
1122 |
self.Attach()
|
1123 |
|
1124 |
@classmethod
|
1125 |
def _InitMeta(cls, minor, dev_path): |
1126 |
"""Initialize a meta device.
|
1127 |
|
1128 |
This will not work if the given minor is in use.
|
1129 |
|
1130 |
"""
|
1131 |
result = utils.RunCmd(["drbdmeta", "--force", cls._DevPath(minor), |
1132 |
"v08", dev_path, "0", "create-md"]) |
1133 |
if result.failed:
|
1134 |
_ThrowError("Can't initialize meta device: %s", result.output)
|
1135 |
|
1136 |
@classmethod
|
1137 |
def _FindUnusedMinor(cls): |
1138 |
"""Find an unused DRBD device.
|
1139 |
|
1140 |
This is specific to 8.x as the minors are allocated dynamically,
|
1141 |
so non-existing numbers up to a max minor count are actually free.
|
1142 |
|
1143 |
"""
|
1144 |
data = cls._GetProcData() |
1145 |
|
1146 |
highest = None
|
1147 |
for line in data: |
1148 |
match = cls._UNUSED_LINE_RE.match(line) |
1149 |
if match:
|
1150 |
return int(match.group(1)) |
1151 |
match = cls._VALID_LINE_RE.match(line) |
1152 |
if match:
|
1153 |
minor = int(match.group(1)) |
1154 |
highest = max(highest, minor)
|
1155 |
if highest is None: # there are no minors in use at all |
1156 |
return 0 |
1157 |
if highest >= cls._MAX_MINORS:
|
1158 |
logging.error("Error: no free drbd minors!")
|
1159 |
raise errors.BlockDeviceError("Can't find a free DRBD minor") |
1160 |
return highest + 1 |
1161 |
|
1162 |
@classmethod
|
1163 |
def _GetShowParser(cls): |
1164 |
"""Return a parser for `drbd show` output.
|
1165 |
|
1166 |
This will either create or return an already-create parser for the
|
1167 |
output of the command `drbd show`.
|
1168 |
|
1169 |
"""
|
1170 |
if cls._PARSE_SHOW is not None: |
1171 |
return cls._PARSE_SHOW
|
1172 |
|
1173 |
# pyparsing setup
|
1174 |
lbrace = pyp.Literal("{").suppress()
|
1175 |
rbrace = pyp.Literal("}").suppress()
|
1176 |
lbracket = pyp.Literal("[").suppress()
|
1177 |
rbracket = pyp.Literal("]").suppress()
|
1178 |
semi = pyp.Literal(";").suppress()
|
1179 |
colon = pyp.Literal(":").suppress()
|
1180 |
# this also converts the value to an int
|
1181 |
number = pyp.Word(pyp.nums).setParseAction(lambda s, l, t: int(t[0])) |
1182 |
|
1183 |
comment = pyp.Literal ("#") + pyp.Optional(pyp.restOfLine)
|
1184 |
defa = pyp.Literal("_is_default").suppress()
|
1185 |
dbl_quote = pyp.Literal('"').suppress()
|
1186 |
|
1187 |
keyword = pyp.Word(pyp.alphanums + '-')
|
1188 |
|
1189 |
# value types
|
1190 |
value = pyp.Word(pyp.alphanums + '_-/.:')
|
1191 |
quoted = dbl_quote + pyp.CharsNotIn('"') + dbl_quote
|
1192 |
ipv4_addr = (pyp.Optional(pyp.Literal("ipv4")).suppress() +
|
1193 |
pyp.Word(pyp.nums + ".") + colon + number)
|
1194 |
ipv6_addr = (pyp.Optional(pyp.Literal("ipv6")).suppress() +
|
1195 |
pyp.Optional(lbracket) + pyp.Word(pyp.hexnums + ":") +
|
1196 |
pyp.Optional(rbracket) + colon + number) |
1197 |
# meta device, extended syntax
|
1198 |
meta_value = ((value ^ quoted) + lbracket + number + rbracket) |
1199 |
# device name, extended syntax
|
1200 |
device_value = pyp.Literal("minor").suppress() + number
|
1201 |
|
1202 |
# a statement
|
1203 |
stmt = (~rbrace + keyword + ~lbrace + |
1204 |
pyp.Optional(ipv4_addr ^ ipv6_addr ^ value ^ quoted ^ meta_value ^ |
1205 |
device_value) + |
1206 |
pyp.Optional(defa) + semi + |
1207 |
pyp.Optional(pyp.restOfLine).suppress()) |
1208 |
|
1209 |
# an entire section
|
1210 |
section_name = pyp.Word(pyp.alphas + '_')
|
1211 |
section = section_name + lbrace + pyp.ZeroOrMore(pyp.Group(stmt)) + rbrace |
1212 |
|
1213 |
bnf = pyp.ZeroOrMore(pyp.Group(section ^ stmt)) |
1214 |
bnf.ignore(comment) |
1215 |
|
1216 |
cls._PARSE_SHOW = bnf |
1217 |
|
1218 |
return bnf
|
1219 |
|
1220 |
@classmethod
|
1221 |
def _GetShowData(cls, minor): |
1222 |
"""Return the `drbdsetup show` data for a minor.
|
1223 |
|
1224 |
"""
|
1225 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) |
1226 |
if result.failed:
|
1227 |
logging.error("Can't display the drbd config: %s - %s",
|
1228 |
result.fail_reason, result.output) |
1229 |
return None |
1230 |
return result.stdout
|
1231 |
|
1232 |
@classmethod
|
1233 |
def _GetDevInfo(cls, out): |
1234 |
"""Parse details about a given DRBD minor.
|
1235 |
|
1236 |
This return, if available, the local backing device (as a path)
|
1237 |
and the local and remote (ip, port) information from a string
|
1238 |
containing the output of the `drbdsetup show` command as returned
|
1239 |
by _GetShowData.
|
1240 |
|
1241 |
"""
|
1242 |
data = {} |
1243 |
if not out: |
1244 |
return data
|
1245 |
|
1246 |
bnf = cls._GetShowParser() |
1247 |
# run pyparse
|
1248 |
|
1249 |
try:
|
1250 |
results = bnf.parseString(out) |
1251 |
except pyp.ParseException, err:
|
1252 |
_ThrowError("Can't parse drbdsetup show output: %s", str(err)) |
1253 |
|
1254 |
# and massage the results into our desired format
|
1255 |
for section in results: |
1256 |
sname = section[0]
|
1257 |
if sname == "_this_host": |
1258 |
for lst in section[1:]: |
1259 |
if lst[0] == "disk": |
1260 |
data["local_dev"] = lst[1] |
1261 |
elif lst[0] == "meta-disk": |
1262 |
data["meta_dev"] = lst[1] |
1263 |
data["meta_index"] = lst[2] |
1264 |
elif lst[0] == "address": |
1265 |
data["local_addr"] = tuple(lst[1:]) |
1266 |
elif sname == "_remote_host": |
1267 |
for lst in section[1:]: |
1268 |
if lst[0] == "address": |
1269 |
data["remote_addr"] = tuple(lst[1:]) |
1270 |
return data
|
1271 |
|
1272 |
def _MatchesLocal(self, info): |
1273 |
"""Test if our local config matches with an existing device.
|
1274 |
|
1275 |
The parameter should be as returned from `_GetDevInfo()`. This
|
1276 |
method tests if our local backing device is the same as the one in
|
1277 |
the info parameter, in effect testing if we look like the given
|
1278 |
device.
|
1279 |
|
1280 |
"""
|
1281 |
if self._children: |
1282 |
backend, meta = self._children
|
1283 |
else:
|
1284 |
backend = meta = None
|
1285 |
|
1286 |
if backend is not None: |
1287 |
retval = ("local_dev" in info and info["local_dev"] == backend.dev_path) |
1288 |
else:
|
1289 |
retval = ("local_dev" not in info) |
1290 |
|
1291 |
if meta is not None: |
1292 |
retval = retval and ("meta_dev" in info and |
1293 |
info["meta_dev"] == meta.dev_path)
|
1294 |
retval = retval and ("meta_index" in info and |
1295 |
info["meta_index"] == 0) |
1296 |
else:
|
1297 |
retval = retval and ("meta_dev" not in info and |
1298 |
"meta_index" not in info) |
1299 |
return retval
|
1300 |
|
1301 |
def _MatchesNet(self, info): |
1302 |
"""Test if our network config matches with an existing device.
|
1303 |
|
1304 |
The parameter should be as returned from `_GetDevInfo()`. This
|
1305 |
method tests if our network configuration is the same as the one
|
1306 |
in the info parameter, in effect testing if we look like the given
|
1307 |
device.
|
1308 |
|
1309 |
"""
|
1310 |
if (((self._lhost is None and not ("local_addr" in info)) and |
1311 |
(self._rhost is None and not ("remote_addr" in info)))): |
1312 |
return True |
1313 |
|
1314 |
if self._lhost is None: |
1315 |
return False |
1316 |
|
1317 |
if not ("local_addr" in info and |
1318 |
"remote_addr" in info): |
1319 |
return False |
1320 |
|
1321 |
retval = (info["local_addr"] == (self._lhost, self._lport)) |
1322 |
retval = (retval and
|
1323 |
info["remote_addr"] == (self._rhost, self._rport)) |
1324 |
return retval
|
1325 |
|
1326 |
@classmethod
|
1327 |
def _AssembleLocal(cls, minor, backend, meta, size): |
1328 |
"""Configure the local part of a DRBD device.
|
1329 |
|
1330 |
"""
|
1331 |
args = ["drbdsetup", cls._DevPath(minor), "disk", |
1332 |
backend, meta, "0",
|
1333 |
"-e", "detach", |
1334 |
"--create-device"]
|
1335 |
if size:
|
1336 |
args.extend(["-d", "%sm" % size]) |
1337 |
if not constants.DRBD_BARRIERS: # disable barriers, if configured so |
1338 |
version = cls._GetVersion(cls._GetProcData()) |
1339 |
# various DRBD versions support different disk barrier options;
|
1340 |
# what we aim here is to revert back to the 'drain' method of
|
1341 |
# disk flushes and to disable metadata barriers, in effect going
|
1342 |
# back to pre-8.0.7 behaviour
|
1343 |
vmaj = version['k_major']
|
1344 |
vmin = version['k_minor']
|
1345 |
vrel = version['k_point']
|
1346 |
assert vmaj == 8 |
1347 |
if vmin == 0: # 8.0.x |
1348 |
if vrel >= 12: |
1349 |
args.extend(['-i', '-m']) |
1350 |
elif vmin == 2: # 8.2.x |
1351 |
if vrel >= 7: |
1352 |
args.extend(['-i', '-m']) |
1353 |
elif vmaj >= 3: # 8.3.x or newer |
1354 |
args.extend(['-i', '-a', 'm']) |
1355 |
result = utils.RunCmd(args) |
1356 |
if result.failed:
|
1357 |
_ThrowError("drbd%d: can't attach local disk: %s", minor, result.output)
|
1358 |
|
1359 |
@classmethod
|
1360 |
def _AssembleNet(cls, minor, net_info, protocol, |
1361 |
dual_pri=False, hmac=None, secret=None): |
1362 |
"""Configure the network part of the device.
|
1363 |
|
1364 |
"""
|
1365 |
lhost, lport, rhost, rport = net_info |
1366 |
if None in net_info: |
1367 |
# we don't want network connection and actually want to make
|
1368 |
# sure its shutdown
|
1369 |
cls._ShutdownNet(minor) |
1370 |
return
|
1371 |
|
1372 |
# Workaround for a race condition. When DRBD is doing its dance to
|
1373 |
# establish a connection with its peer, it also sends the
|
1374 |
# synchronization speed over the wire. In some cases setting the
|
1375 |
# sync speed only after setting up both sides can race with DRBD
|
1376 |
# connecting, hence we set it here before telling DRBD anything
|
1377 |
# about its peer.
|
1378 |
cls._SetMinorSyncSpeed(minor, constants.SYNC_SPEED) |
1379 |
|
1380 |
if netutils.IP6Address.IsValid(lhost):
|
1381 |
if not netutils.IP6Address.IsValid(rhost): |
1382 |
_ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
1383 |
(minor, lhost, rhost)) |
1384 |
family = "ipv6"
|
1385 |
elif netutils.IP4Address.IsValid(lhost):
|
1386 |
if not netutils.IP4Address.IsValid(rhost): |
1387 |
_ThrowError("drbd%d: can't connect ip %s to ip %s" %
|
1388 |
(minor, lhost, rhost)) |
1389 |
family = "ipv4"
|
1390 |
else:
|
1391 |
_ThrowError("drbd%d: Invalid ip %s" % (minor, lhost))
|
1392 |
|
1393 |
args = ["drbdsetup", cls._DevPath(minor), "net", |
1394 |
"%s:%s:%s" % (family, lhost, lport),
|
1395 |
"%s:%s:%s" % (family, rhost, rport), protocol,
|
1396 |
"-A", "discard-zero-changes", |
1397 |
"-B", "consensus", |
1398 |
"--create-device",
|
1399 |
] |
1400 |
if dual_pri:
|
1401 |
args.append("-m")
|
1402 |
if hmac and secret: |
1403 |
args.extend(["-a", hmac, "-x", secret]) |
1404 |
result = utils.RunCmd(args) |
1405 |
if result.failed:
|
1406 |
_ThrowError("drbd%d: can't setup network: %s - %s",
|
1407 |
minor, result.fail_reason, result.output) |
1408 |
|
1409 |
def _CheckNetworkConfig(): |
1410 |
info = cls._GetDevInfo(cls._GetShowData(minor)) |
1411 |
if not "local_addr" in info or not "remote_addr" in info: |
1412 |
raise utils.RetryAgain()
|
1413 |
|
1414 |
if (info["local_addr"] != (lhost, lport) or |
1415 |
info["remote_addr"] != (rhost, rport)):
|
1416 |
raise utils.RetryAgain()
|
1417 |
|
1418 |
try:
|
1419 |
utils.Retry(_CheckNetworkConfig, 1.0, 10.0) |
1420 |
except utils.RetryTimeout:
|
1421 |
_ThrowError("drbd%d: timeout while configuring network", minor)
|
1422 |
|
1423 |
def AddChildren(self, devices): |
1424 |
"""Add a disk to the DRBD device.
|
1425 |
|
1426 |
"""
|
1427 |
if self.minor is None: |
1428 |
_ThrowError("drbd%d: can't attach to dbrd8 during AddChildren",
|
1429 |
self._aminor)
|
1430 |
if len(devices) != 2: |
1431 |
_ThrowError("drbd%d: need two devices for AddChildren", self.minor) |
1432 |
info = self._GetDevInfo(self._GetShowData(self.minor)) |
1433 |
if "local_dev" in info: |
1434 |
_ThrowError("drbd%d: already attached to a local disk", self.minor) |
1435 |
backend, meta = devices |
1436 |
if backend.dev_path is None or meta.dev_path is None: |
1437 |
_ThrowError("drbd%d: children not ready during AddChildren", self.minor) |
1438 |
backend.Open() |
1439 |
meta.Open() |
1440 |
self._CheckMetaSize(meta.dev_path)
|
1441 |
self._InitMeta(self._FindUnusedMinor(), meta.dev_path) |
1442 |
|
1443 |
self._AssembleLocal(self.minor, backend.dev_path, meta.dev_path, self.size) |
1444 |
self._children = devices
|
1445 |
|
1446 |
def RemoveChildren(self, devices): |
1447 |
"""Detach the drbd device from local storage.
|
1448 |
|
1449 |
"""
|
1450 |
if self.minor is None: |
1451 |
_ThrowError("drbd%d: can't attach to drbd8 during RemoveChildren",
|
1452 |
self._aminor)
|
1453 |
# early return if we don't actually have backing storage
|
1454 |
info = self._GetDevInfo(self._GetShowData(self.minor)) |
1455 |
if "local_dev" not in info: |
1456 |
return
|
1457 |
if len(self._children) != 2: |
1458 |
_ThrowError("drbd%d: we don't have two children: %s", self.minor, |
1459 |
self._children)
|
1460 |
if self._children.count(None) == 2: # we don't actually have children :) |
1461 |
logging.warning("drbd%d: requested detach while detached", self.minor) |
1462 |
return
|
1463 |
if len(devices) != 2: |
1464 |
_ThrowError("drbd%d: we need two children in RemoveChildren", self.minor) |
1465 |
for child, dev in zip(self._children, devices): |
1466 |
if dev != child.dev_path:
|
1467 |
_ThrowError("drbd%d: mismatch in local storage (%s != %s) in"
|
1468 |
" RemoveChildren", self.minor, dev, child.dev_path) |
1469 |
|
1470 |
self._ShutdownLocal(self.minor) |
1471 |
self._children = []
|
1472 |
|
1473 |
@classmethod
|
1474 |
def _SetMinorSyncSpeed(cls, minor, kbytes): |
1475 |
"""Set the speed of the DRBD syncer.
|
1476 |
|
1477 |
This is the low-level implementation.
|
1478 |
|
1479 |
@type minor: int
|
1480 |
@param minor: the drbd minor whose settings we change
|
1481 |
@type kbytes: int
|
1482 |
@param kbytes: the speed in kbytes/second
|
1483 |
@rtype: boolean
|
1484 |
@return: the success of the operation
|
1485 |
|
1486 |
"""
|
1487 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "syncer", |
1488 |
"-r", "%d" % kbytes, "--create-device"]) |
1489 |
if result.failed:
|
1490 |
logging.error("Can't change syncer rate: %s - %s",
|
1491 |
result.fail_reason, result.output) |
1492 |
return not result.failed |
1493 |
|
1494 |
def SetSyncSpeed(self, kbytes): |
1495 |
"""Set the speed of the DRBD syncer.
|
1496 |
|
1497 |
@type kbytes: int
|
1498 |
@param kbytes: the speed in kbytes/second
|
1499 |
@rtype: boolean
|
1500 |
@return: the success of the operation
|
1501 |
|
1502 |
"""
|
1503 |
if self.minor is None: |
1504 |
logging.info("Not attached during SetSyncSpeed")
|
1505 |
return False |
1506 |
children_result = super(DRBD8, self).SetSyncSpeed(kbytes) |
1507 |
return self._SetMinorSyncSpeed(self.minor, kbytes) and children_result |
1508 |
|
1509 |
def PauseResumeSync(self, pause): |
1510 |
"""Pauses or resumes the sync of a DRBD device.
|
1511 |
|
1512 |
@param pause: Wether to pause or resume
|
1513 |
@return: the success of the operation
|
1514 |
|
1515 |
"""
|
1516 |
if self.minor is None: |
1517 |
logging.info("Not attached during PauseSync")
|
1518 |
return False |
1519 |
|
1520 |
children_result = super(DRBD8, self).PauseResumeSync(pause) |
1521 |
|
1522 |
if pause:
|
1523 |
cmd = "pause-sync"
|
1524 |
else:
|
1525 |
cmd = "resume-sync"
|
1526 |
|
1527 |
result = utils.RunCmd(["drbdsetup", self.dev_path, cmd]) |
1528 |
if result.failed:
|
1529 |
logging.error("Can't %s: %s - %s", cmd,
|
1530 |
result.fail_reason, result.output) |
1531 |
return not result.failed and children_result |
1532 |
|
1533 |
def GetProcStatus(self): |
1534 |
"""Return device data from /proc.
|
1535 |
|
1536 |
"""
|
1537 |
if self.minor is None: |
1538 |
_ThrowError("drbd%d: GetStats() called while not attached", self._aminor) |
1539 |
proc_info = self._MassageProcData(self._GetProcData()) |
1540 |
if self.minor not in proc_info: |
1541 |
_ThrowError("drbd%d: can't find myself in /proc", self.minor) |
1542 |
return DRBD8Status(proc_info[self.minor]) |
1543 |
|
1544 |
def GetSyncStatus(self): |
1545 |
"""Returns the sync status of the device.
|
1546 |
|
1547 |
|
1548 |
If sync_percent is None, it means all is ok
|
1549 |
If estimated_time is None, it means we can't estimate
|
1550 |
the time needed, otherwise it's the time left in seconds.
|
1551 |
|
1552 |
|
1553 |
We set the is_degraded parameter to True on two conditions:
|
1554 |
network not connected or local disk missing.
|
1555 |
|
1556 |
We compute the ldisk parameter based on whether we have a local
|
1557 |
disk or not.
|
1558 |
|
1559 |
@rtype: objects.BlockDevStatus
|
1560 |
|
1561 |
"""
|
1562 |
if self.minor is None and not self.Attach(): |
1563 |
_ThrowError("drbd%d: can't Attach() in GetSyncStatus", self._aminor) |
1564 |
|
1565 |
stats = self.GetProcStatus()
|
1566 |
is_degraded = not stats.is_connected or not stats.is_disk_uptodate |
1567 |
|
1568 |
if stats.is_disk_uptodate:
|
1569 |
ldisk_status = constants.LDS_OKAY |
1570 |
elif stats.is_diskless:
|
1571 |
ldisk_status = constants.LDS_FAULTY |
1572 |
else:
|
1573 |
ldisk_status = constants.LDS_UNKNOWN |
1574 |
|
1575 |
return objects.BlockDevStatus(dev_path=self.dev_path, |
1576 |
major=self.major,
|
1577 |
minor=self.minor,
|
1578 |
sync_percent=stats.sync_percent, |
1579 |
estimated_time=stats.est_time, |
1580 |
is_degraded=is_degraded, |
1581 |
ldisk_status=ldisk_status) |
1582 |
|
1583 |
def Open(self, force=False): |
1584 |
"""Make the local state primary.
|
1585 |
|
1586 |
If the 'force' parameter is given, the '-o' option is passed to
|
1587 |
drbdsetup. Since this is a potentially dangerous operation, the
|
1588 |
force flag should be only given after creation, when it actually
|
1589 |
is mandatory.
|
1590 |
|
1591 |
"""
|
1592 |
if self.minor is None and not self.Attach(): |
1593 |
logging.error("DRBD cannot attach to a device during open")
|
1594 |
return False |
1595 |
cmd = ["drbdsetup", self.dev_path, "primary"] |
1596 |
if force:
|
1597 |
cmd.append("-o")
|
1598 |
result = utils.RunCmd(cmd) |
1599 |
if result.failed:
|
1600 |
_ThrowError("drbd%d: can't make drbd device primary: %s", self.minor, |
1601 |
result.output) |
1602 |
|
1603 |
def Close(self): |
1604 |
"""Make the local state secondary.
|
1605 |
|
1606 |
This will, of course, fail if the device is in use.
|
1607 |
|
1608 |
"""
|
1609 |
if self.minor is None and not self.Attach(): |
1610 |
_ThrowError("drbd%d: can't Attach() in Close()", self._aminor) |
1611 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "secondary"]) |
1612 |
if result.failed:
|
1613 |
_ThrowError("drbd%d: can't switch drbd device to secondary: %s",
|
1614 |
self.minor, result.output)
|
1615 |
|
1616 |
def DisconnectNet(self): |
1617 |
"""Removes network configuration.
|
1618 |
|
1619 |
This method shutdowns the network side of the device.
|
1620 |
|
1621 |
The method will wait up to a hardcoded timeout for the device to
|
1622 |
go into standalone after the 'disconnect' command before
|
1623 |
re-configuring it, as sometimes it takes a while for the
|
1624 |
disconnect to actually propagate and thus we might issue a 'net'
|
1625 |
command while the device is still connected. If the device will
|
1626 |
still be attached to the network and we time out, we raise an
|
1627 |
exception.
|
1628 |
|
1629 |
"""
|
1630 |
if self.minor is None: |
1631 |
_ThrowError("drbd%d: disk not attached in re-attach net", self._aminor) |
1632 |
|
1633 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
1634 |
_ThrowError("drbd%d: DRBD disk missing network info in"
|
1635 |
" DisconnectNet()", self.minor) |
1636 |
|
1637 |
class _DisconnectStatus: |
1638 |
def __init__(self, ever_disconnected): |
1639 |
self.ever_disconnected = ever_disconnected
|
1640 |
|
1641 |
dstatus = _DisconnectStatus(_IgnoreError(self._ShutdownNet, self.minor)) |
1642 |
|
1643 |
def _WaitForDisconnect(): |
1644 |
if self.GetProcStatus().is_standalone: |
1645 |
return
|
1646 |
|
1647 |
# retry the disconnect, it seems possible that due to a well-time
|
1648 |
# disconnect on the peer, my disconnect command might be ignored and
|
1649 |
# forgotten
|
1650 |
dstatus.ever_disconnected = \ |
1651 |
_IgnoreError(self._ShutdownNet, self.minor) or dstatus.ever_disconnected |
1652 |
|
1653 |
raise utils.RetryAgain()
|
1654 |
|
1655 |
# Keep start time
|
1656 |
start_time = time.time() |
1657 |
|
1658 |
try:
|
1659 |
# Start delay at 100 milliseconds and grow up to 2 seconds
|
1660 |
utils.Retry(_WaitForDisconnect, (0.1, 1.5, 2.0), |
1661 |
self._NET_RECONFIG_TIMEOUT)
|
1662 |
except utils.RetryTimeout:
|
1663 |
if dstatus.ever_disconnected:
|
1664 |
msg = ("drbd%d: device did not react to the"
|
1665 |
" 'disconnect' command in a timely manner")
|
1666 |
else:
|
1667 |
msg = "drbd%d: can't shutdown network, even after multiple retries"
|
1668 |
|
1669 |
_ThrowError(msg, self.minor)
|
1670 |
|
1671 |
reconfig_time = time.time() - start_time |
1672 |
if reconfig_time > (self._NET_RECONFIG_TIMEOUT * 0.25): |
1673 |
logging.info("drbd%d: DisconnectNet: detach took %.3f seconds",
|
1674 |
self.minor, reconfig_time)
|
1675 |
|
1676 |
def AttachNet(self, multimaster): |
1677 |
"""Reconnects the network.
|
1678 |
|
1679 |
This method connects the network side of the device with a
|
1680 |
specified multi-master flag. The device needs to be 'Standalone'
|
1681 |
but have valid network configuration data.
|
1682 |
|
1683 |
Args:
|
1684 |
- multimaster: init the network in dual-primary mode
|
1685 |
|
1686 |
"""
|
1687 |
if self.minor is None: |
1688 |
_ThrowError("drbd%d: device not attached in AttachNet", self._aminor) |
1689 |
|
1690 |
if None in (self._lhost, self._lport, self._rhost, self._rport): |
1691 |
_ThrowError("drbd%d: missing network info in AttachNet()", self.minor) |
1692 |
|
1693 |
status = self.GetProcStatus()
|
1694 |
|
1695 |
if not status.is_standalone: |
1696 |
_ThrowError("drbd%d: device is not standalone in AttachNet", self.minor) |
1697 |
|
1698 |
self._AssembleNet(self.minor, |
1699 |
(self._lhost, self._lport, self._rhost, self._rport), |
1700 |
constants.DRBD_NET_PROTOCOL, dual_pri=multimaster, |
1701 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
1702 |
|
1703 |
def Attach(self): |
1704 |
"""Check if our minor is configured.
|
1705 |
|
1706 |
This doesn't do any device configurations - it only checks if the
|
1707 |
minor is in a state different from Unconfigured.
|
1708 |
|
1709 |
Note that this function will not change the state of the system in
|
1710 |
any way (except in case of side-effects caused by reading from
|
1711 |
/proc).
|
1712 |
|
1713 |
"""
|
1714 |
used_devs = self.GetUsedDevs()
|
1715 |
if self._aminor in used_devs: |
1716 |
minor = self._aminor
|
1717 |
else:
|
1718 |
minor = None
|
1719 |
|
1720 |
self._SetFromMinor(minor)
|
1721 |
return minor is not None |
1722 |
|
1723 |
def Assemble(self): |
1724 |
"""Assemble the drbd.
|
1725 |
|
1726 |
Method:
|
1727 |
- if we have a configured device, we try to ensure that it matches
|
1728 |
our config
|
1729 |
- if not, we create it from zero
|
1730 |
|
1731 |
"""
|
1732 |
super(DRBD8, self).Assemble() |
1733 |
|
1734 |
self.Attach()
|
1735 |
if self.minor is None: |
1736 |
# local device completely unconfigured
|
1737 |
self._FastAssemble()
|
1738 |
else:
|
1739 |
# we have to recheck the local and network status and try to fix
|
1740 |
# the device
|
1741 |
self._SlowAssemble()
|
1742 |
|
1743 |
def _SlowAssemble(self): |
1744 |
"""Assembles the DRBD device from a (partially) configured device.
|
1745 |
|
1746 |
In case of partially attached (local device matches but no network
|
1747 |
setup), we perform the network attach. If successful, we re-test
|
1748 |
the attach if can return success.
|
1749 |
|
1750 |
"""
|
1751 |
# TODO: Rewrite to not use a for loop just because there is 'break'
|
1752 |
# pylint: disable-msg=W0631
|
1753 |
net_data = (self._lhost, self._lport, self._rhost, self._rport) |
1754 |
for minor in (self._aminor,): |
1755 |
info = self._GetDevInfo(self._GetShowData(minor)) |
1756 |
match_l = self._MatchesLocal(info)
|
1757 |
match_r = self._MatchesNet(info)
|
1758 |
|
1759 |
if match_l and match_r: |
1760 |
# everything matches
|
1761 |
break
|
1762 |
|
1763 |
if match_l and not match_r and "local_addr" not in info: |
1764 |
# disk matches, but not attached to network, attach and recheck
|
1765 |
self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
|
1766 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
1767 |
if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): |
1768 |
break
|
1769 |
else:
|
1770 |
_ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
1771 |
" show' disagrees", minor)
|
1772 |
|
1773 |
if match_r and "local_dev" not in info: |
1774 |
# no local disk, but network attached and it matches
|
1775 |
self._AssembleLocal(minor, self._children[0].dev_path, |
1776 |
self._children[1].dev_path, self.size) |
1777 |
if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): |
1778 |
break
|
1779 |
else:
|
1780 |
_ThrowError("drbd%d: disk attach successful, but 'drbdsetup"
|
1781 |
" show' disagrees", minor)
|
1782 |
|
1783 |
# this case must be considered only if we actually have local
|
1784 |
# storage, i.e. not in diskless mode, because all diskless
|
1785 |
# devices are equal from the point of view of local
|
1786 |
# configuration
|
1787 |
if (match_l and "local_dev" in info and |
1788 |
not match_r and "local_addr" in info): |
1789 |
# strange case - the device network part points to somewhere
|
1790 |
# else, even though its local storage is ours; as we own the
|
1791 |
# drbd space, we try to disconnect from the remote peer and
|
1792 |
# reconnect to our correct one
|
1793 |
try:
|
1794 |
self._ShutdownNet(minor)
|
1795 |
except errors.BlockDeviceError, err:
|
1796 |
_ThrowError("drbd%d: device has correct local storage, wrong"
|
1797 |
" remote peer and is unable to disconnect in order"
|
1798 |
" to attach to the correct peer: %s", minor, str(err)) |
1799 |
# note: _AssembleNet also handles the case when we don't want
|
1800 |
# local storage (i.e. one or more of the _[lr](host|port) is
|
1801 |
# None)
|
1802 |
self._AssembleNet(minor, net_data, constants.DRBD_NET_PROTOCOL,
|
1803 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
1804 |
if self._MatchesNet(self._GetDevInfo(self._GetShowData(minor))): |
1805 |
break
|
1806 |
else:
|
1807 |
_ThrowError("drbd%d: network attach successful, but 'drbdsetup"
|
1808 |
" show' disagrees", minor)
|
1809 |
|
1810 |
else:
|
1811 |
minor = None
|
1812 |
|
1813 |
self._SetFromMinor(minor)
|
1814 |
if minor is None: |
1815 |
_ThrowError("drbd%d: cannot activate, unknown or unhandled reason",
|
1816 |
self._aminor)
|
1817 |
|
1818 |
def _FastAssemble(self): |
1819 |
"""Assemble the drbd device from zero.
|
1820 |
|
1821 |
This is run when in Assemble we detect our minor is unused.
|
1822 |
|
1823 |
"""
|
1824 |
minor = self._aminor
|
1825 |
if self._children and self._children[0] and self._children[1]: |
1826 |
self._AssembleLocal(minor, self._children[0].dev_path, |
1827 |
self._children[1].dev_path, self.size) |
1828 |
if self._lhost and self._lport and self._rhost and self._rport: |
1829 |
self._AssembleNet(minor,
|
1830 |
(self._lhost, self._lport, self._rhost, self._rport), |
1831 |
constants.DRBD_NET_PROTOCOL, |
1832 |
hmac=constants.DRBD_HMAC_ALG, secret=self._secret)
|
1833 |
self._SetFromMinor(minor)
|
1834 |
|
1835 |
@classmethod
|
1836 |
def _ShutdownLocal(cls, minor): |
1837 |
"""Detach from the local device.
|
1838 |
|
1839 |
I/Os will continue to be served from the remote device. If we
|
1840 |
don't have a remote device, this operation will fail.
|
1841 |
|
1842 |
"""
|
1843 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) |
1844 |
if result.failed:
|
1845 |
_ThrowError("drbd%d: can't detach local disk: %s", minor, result.output)
|
1846 |
|
1847 |
@classmethod
|
1848 |
def _ShutdownNet(cls, minor): |
1849 |
"""Disconnect from the remote peer.
|
1850 |
|
1851 |
This fails if we don't have a local device.
|
1852 |
|
1853 |
"""
|
1854 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) |
1855 |
if result.failed:
|
1856 |
_ThrowError("drbd%d: can't shutdown network: %s", minor, result.output)
|
1857 |
|
1858 |
@classmethod
|
1859 |
def _ShutdownAll(cls, minor): |
1860 |
"""Deactivate the device.
|
1861 |
|
1862 |
This will, of course, fail if the device is in use.
|
1863 |
|
1864 |
"""
|
1865 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "down"]) |
1866 |
if result.failed:
|
1867 |
_ThrowError("drbd%d: can't shutdown drbd device: %s",
|
1868 |
minor, result.output) |
1869 |
|
1870 |
def Shutdown(self): |
1871 |
"""Shutdown the DRBD device.
|
1872 |
|
1873 |
"""
|
1874 |
if self.minor is None and not self.Attach(): |
1875 |
logging.info("drbd%d: not attached during Shutdown()", self._aminor) |
1876 |
return
|
1877 |
minor = self.minor
|
1878 |
self.minor = None |
1879 |
self.dev_path = None |
1880 |
self._ShutdownAll(minor)
|
1881 |
|
1882 |
def Remove(self): |
1883 |
"""Stub remove for DRBD devices.
|
1884 |
|
1885 |
"""
|
1886 |
self.Shutdown()
|
1887 |
|
1888 |
@classmethod
|
1889 |
def Create(cls, unique_id, children, size): |
1890 |
"""Create a new DRBD8 device.
|
1891 |
|
1892 |
Since DRBD devices are not created per se, just assembled, this
|
1893 |
function only initializes the metadata.
|
1894 |
|
1895 |
"""
|
1896 |
if len(children) != 2: |
1897 |
raise errors.ProgrammerError("Invalid setup for the drbd device") |
1898 |
# check that the minor is unused
|
1899 |
aminor = unique_id[4]
|
1900 |
proc_info = cls._MassageProcData(cls._GetProcData()) |
1901 |
if aminor in proc_info: |
1902 |
status = DRBD8Status(proc_info[aminor]) |
1903 |
in_use = status.is_in_use |
1904 |
else:
|
1905 |
in_use = False
|
1906 |
if in_use:
|
1907 |
_ThrowError("drbd%d: minor is already in use at Create() time", aminor)
|
1908 |
meta = children[1]
|
1909 |
meta.Assemble() |
1910 |
if not meta.Attach(): |
1911 |
_ThrowError("drbd%d: can't attach to meta device '%s'",
|
1912 |
aminor, meta) |
1913 |
cls._CheckMetaSize(meta.dev_path) |
1914 |
cls._InitMeta(aminor, meta.dev_path) |
1915 |
return cls(unique_id, children, size)
|
1916 |
|
1917 |
def Grow(self, amount): |
1918 |
"""Resize the DRBD device and its backing storage.
|
1919 |
|
1920 |
"""
|
1921 |
if self.minor is None: |
1922 |
_ThrowError("drbd%d: Grow called while not attached", self._aminor) |
1923 |
if len(self._children) != 2 or None in self._children: |
1924 |
_ThrowError("drbd%d: cannot grow diskless device", self.minor) |
1925 |
self._children[0].Grow(amount) |
1926 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "resize", "-s", |
1927 |
"%dm" % (self.size + amount)]) |
1928 |
if result.failed:
|
1929 |
_ThrowError("drbd%d: resize failed: %s", self.minor, result.output) |
1930 |
|
1931 |
|
1932 |
class FileStorage(BlockDev): |
1933 |
"""File device.
|
1934 |
|
1935 |
This class represents the a file storage backend device.
|
1936 |
|
1937 |
The unique_id for the file device is a (file_driver, file_path) tuple.
|
1938 |
|
1939 |
"""
|
1940 |
def __init__(self, unique_id, children, size): |
1941 |
"""Initalizes a file device backend.
|
1942 |
|
1943 |
"""
|
1944 |
if children:
|
1945 |
raise errors.BlockDeviceError("Invalid setup for file device") |
1946 |
super(FileStorage, self).__init__(unique_id, children, size) |
1947 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
1948 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
1949 |
self.driver = unique_id[0] |
1950 |
self.dev_path = unique_id[1] |
1951 |
self.Attach()
|
1952 |
|
1953 |
def Assemble(self): |
1954 |
"""Assemble the device.
|
1955 |
|
1956 |
Checks whether the file device exists, raises BlockDeviceError otherwise.
|
1957 |
|
1958 |
"""
|
1959 |
if not os.path.exists(self.dev_path): |
1960 |
_ThrowError("File device '%s' does not exist" % self.dev_path) |
1961 |
|
1962 |
def Shutdown(self): |
1963 |
"""Shutdown the device.
|
1964 |
|
1965 |
This is a no-op for the file type, as we don't deactivate
|
1966 |
the file on shutdown.
|
1967 |
|
1968 |
"""
|
1969 |
pass
|
1970 |
|
1971 |
def Open(self, force=False): |
1972 |
"""Make the device ready for I/O.
|
1973 |
|
1974 |
This is a no-op for the file type.
|
1975 |
|
1976 |
"""
|
1977 |
pass
|
1978 |
|
1979 |
def Close(self): |
1980 |
"""Notifies that the device will no longer be used for I/O.
|
1981 |
|
1982 |
This is a no-op for the file type.
|
1983 |
|
1984 |
"""
|
1985 |
pass
|
1986 |
|
1987 |
def Remove(self): |
1988 |
"""Remove the file backing the block device.
|
1989 |
|
1990 |
@rtype: boolean
|
1991 |
@return: True if the removal was successful
|
1992 |
|
1993 |
"""
|
1994 |
try:
|
1995 |
os.remove(self.dev_path)
|
1996 |
except OSError, err: |
1997 |
if err.errno != errno.ENOENT:
|
1998 |
_ThrowError("Can't remove file '%s': %s", self.dev_path, err) |
1999 |
|
2000 |
def Rename(self, new_id): |
2001 |
"""Renames the file.
|
2002 |
|
2003 |
"""
|
2004 |
# TODO: implement rename for file-based storage
|
2005 |
_ThrowError("Rename is not supported for file-based storage")
|
2006 |
|
2007 |
def Grow(self, amount): |
2008 |
"""Grow the file
|
2009 |
|
2010 |
@param amount: the amount (in mebibytes) to grow with
|
2011 |
|
2012 |
"""
|
2013 |
# Check that the file exists
|
2014 |
self.Assemble()
|
2015 |
current_size = self.GetActualSize()
|
2016 |
new_size = current_size + amount * 1024 * 1024 |
2017 |
assert new_size > current_size, "Cannot Grow with a negative amount" |
2018 |
try:
|
2019 |
f = open(self.dev_path, "a+") |
2020 |
f.truncate(new_size) |
2021 |
f.close() |
2022 |
except EnvironmentError, err: |
2023 |
_ThrowError("Error in file growth: %", str(err)) |
2024 |
|
2025 |
def Attach(self): |
2026 |
"""Attach to an existing file.
|
2027 |
|
2028 |
Check if this file already exists.
|
2029 |
|
2030 |
@rtype: boolean
|
2031 |
@return: True if file exists
|
2032 |
|
2033 |
"""
|
2034 |
self.attached = os.path.exists(self.dev_path) |
2035 |
return self.attached |
2036 |
|
2037 |
def GetActualSize(self): |
2038 |
"""Return the actual disk size.
|
2039 |
|
2040 |
@note: the device needs to be active when this is called
|
2041 |
|
2042 |
"""
|
2043 |
assert self.attached, "BlockDevice not attached in GetActualSize()" |
2044 |
try:
|
2045 |
st = os.stat(self.dev_path)
|
2046 |
return st.st_size
|
2047 |
except OSError, err: |
2048 |
_ThrowError("Can't stat %s: %s", self.dev_path, err) |
2049 |
|
2050 |
@classmethod
|
2051 |
def Create(cls, unique_id, children, size): |
2052 |
"""Create a new file.
|
2053 |
|
2054 |
@param size: the size of file in MiB
|
2055 |
|
2056 |
@rtype: L{bdev.FileStorage}
|
2057 |
@return: an instance of FileStorage
|
2058 |
|
2059 |
"""
|
2060 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
2061 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
2062 |
dev_path = unique_id[1]
|
2063 |
try:
|
2064 |
fd = os.open(dev_path, os.O_RDWR | os.O_CREAT | os.O_EXCL) |
2065 |
f = os.fdopen(fd, "w")
|
2066 |
f.truncate(size * 1024 * 1024) |
2067 |
f.close() |
2068 |
except EnvironmentError, err: |
2069 |
if err.errno == errno.EEXIST:
|
2070 |
_ThrowError("File already existing: %s", dev_path)
|
2071 |
_ThrowError("Error in file creation: %", str(err)) |
2072 |
|
2073 |
return FileStorage(unique_id, children, size)
|
2074 |
|
2075 |
|
2076 |
class PersistentBlockDevice(BlockDev): |
2077 |
"""A block device with persistent node
|
2078 |
|
2079 |
May be either directly attached, or exposed through DM (e.g. dm-multipath).
|
2080 |
udev helpers are probably required to give persistent, human-friendly
|
2081 |
names.
|
2082 |
|
2083 |
For the time being, pathnames are required to lie under /dev.
|
2084 |
|
2085 |
"""
|
2086 |
def __init__(self, unique_id, children, size): |
2087 |
"""Attaches to a static block device.
|
2088 |
|
2089 |
The unique_id is a path under /dev.
|
2090 |
|
2091 |
"""
|
2092 |
super(PersistentBlockDevice, self).__init__(unique_id, children, size) |
2093 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
2094 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
2095 |
self.dev_path = unique_id[1] |
2096 |
if not os.path.realpath(self.dev_path).startswith('/dev/'): |
2097 |
raise ValueError("Full path '%s' lies outside /dev" % |
2098 |
os.path.realpath(self.dev_path))
|
2099 |
# TODO: this is just a safety guard checking that we only deal with devices
|
2100 |
# we know how to handle. In the future this will be integrated with
|
2101 |
# external storage backends and possible values will probably be collected
|
2102 |
# from the cluster configuration.
|
2103 |
if unique_id[0] != constants.BLOCKDEV_DRIVER_MANUAL: |
2104 |
raise ValueError("Got persistent block device of invalid type: %s" % |
2105 |
unique_id[0])
|
2106 |
|
2107 |
self.major = self.minor = None |
2108 |
self.Attach()
|
2109 |
|
2110 |
@classmethod
|
2111 |
def Create(cls, unique_id, children, size): |
2112 |
"""Create a new device
|
2113 |
|
2114 |
This is a noop, we only return a PersistentBlockDevice instance
|
2115 |
|
2116 |
"""
|
2117 |
return PersistentBlockDevice(unique_id, children, 0) |
2118 |
|
2119 |
def Remove(self): |
2120 |
"""Remove a device
|
2121 |
|
2122 |
This is a noop
|
2123 |
|
2124 |
"""
|
2125 |
pass
|
2126 |
|
2127 |
def Rename(self, new_id): |
2128 |
"""Rename this device.
|
2129 |
|
2130 |
"""
|
2131 |
_ThrowError("Rename is not supported for PersistentBlockDev storage")
|
2132 |
|
2133 |
def Attach(self): |
2134 |
"""Attach to an existing block device.
|
2135 |
|
2136 |
|
2137 |
"""
|
2138 |
self.attached = False |
2139 |
try:
|
2140 |
st = os.stat(self.dev_path)
|
2141 |
except OSError, err: |
2142 |
logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) |
2143 |
return False |
2144 |
|
2145 |
if not stat.S_ISBLK(st.st_mode): |
2146 |
logging.error("%s is not a block device", self.dev_path) |
2147 |
return False |
2148 |
|
2149 |
self.major = os.major(st.st_rdev)
|
2150 |
self.minor = os.minor(st.st_rdev)
|
2151 |
self.attached = True |
2152 |
|
2153 |
return True |
2154 |
|
2155 |
def Assemble(self): |
2156 |
"""Assemble the device.
|
2157 |
|
2158 |
"""
|
2159 |
pass
|
2160 |
|
2161 |
def Shutdown(self): |
2162 |
"""Shutdown the device.
|
2163 |
|
2164 |
"""
|
2165 |
pass
|
2166 |
|
2167 |
def Open(self, force=False): |
2168 |
"""Make the device ready for I/O.
|
2169 |
|
2170 |
"""
|
2171 |
pass
|
2172 |
|
2173 |
def Close(self): |
2174 |
"""Notifies that the device will no longer be used for I/O.
|
2175 |
|
2176 |
"""
|
2177 |
pass
|
2178 |
|
2179 |
def Grow(self, amount): |
2180 |
"""Grow the logical volume.
|
2181 |
|
2182 |
"""
|
2183 |
_ThrowError("Grow is not supported for PersistentBlockDev storage")
|
2184 |
|
2185 |
|
2186 |
class RADOSBlockDevice(BlockDev): |
2187 |
"""A RADOS Block Device (rbd)
|
2188 |
|
2189 |
This class implements the RADOS Block Device for the backend. You need
|
2190 |
the rbd kernel driver, the RADOS Tools and a RADOS working cluster for
|
2191 |
this to be functional.
|
2192 |
|
2193 |
"""
|
2194 |
def __init__(self, unique_id, children, size): |
2195 |
"""Attaches to an rbd device.
|
2196 |
|
2197 |
"""
|
2198 |
super(RADOSBlockDevice, self).__init__(unique_id, children, size) |
2199 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
2200 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
2201 |
|
2202 |
self.unique_id = unique_id
|
2203 |
|
2204 |
self.major = self.minor = None |
2205 |
self.Attach()
|
2206 |
|
2207 |
@classmethod
|
2208 |
def Create(cls, unique_id, children, size): |
2209 |
"""Create a new rbd device
|
2210 |
|
2211 |
Provision a new rbd image file inside a RADOS pool
|
2212 |
|
2213 |
"""
|
2214 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
2215 |
raise errors.ProgrammerError("Invalid configuration data %s" % |
2216 |
str(unique_id))
|
2217 |
rbd_pool, rbd_name = unique_id |
2218 |
|
2219 |
# Provision a new rbd Volume (Image) inside the RADOS cluster
|
2220 |
cmd = "rbd create -p %s %s --size %s" % (rbd_pool, rbd_name, size)
|
2221 |
result = utils.RunCmd(cmd) |
2222 |
if result.failed:
|
2223 |
_ThrowError("rbd creation failed (%s): %s",
|
2224 |
result.fail_reason, result.output) |
2225 |
|
2226 |
return RADOSBlockDevice(unique_id, children, size)
|
2227 |
|
2228 |
def Remove(self): |
2229 |
"""Remove the rbd device
|
2230 |
|
2231 |
"""
|
2232 |
rbd_pool, rbd_name = self.unique_id
|
2233 |
|
2234 |
if not self.minor and not self.Attach(): |
2235 |
# the rbd device doesn't exist
|
2236 |
return
|
2237 |
|
2238 |
# First shutdown the device (remove mappings).
|
2239 |
self.Shutdown()
|
2240 |
|
2241 |
# Remove the actual Volume (Image) from the RADOS cluster
|
2242 |
cmd = "rbd rm -p %s %s" % (rbd_pool, rbd_name)
|
2243 |
result = utils.RunCmd(cmd) |
2244 |
if result.failed:
|
2245 |
_ThrowError("Can't remove Volume from cluster with rbd rm: %s - %s",
|
2246 |
result.fail_reason, result.output) |
2247 |
|
2248 |
def Rename(self, new_id): |
2249 |
"""Rename this device.
|
2250 |
|
2251 |
"""
|
2252 |
pass
|
2253 |
|
2254 |
def Attach(self): |
2255 |
"""Attach to an existing rbd device.
|
2256 |
|
2257 |
This method maps the rbd volume that matches our name with
|
2258 |
an rbd device and then attaches to this device
|
2259 |
|
2260 |
"""
|
2261 |
self.attached = False |
2262 |
|
2263 |
# Map the rbd Volume to a block device under /dev
|
2264 |
self.dev_path = self._MapVolumeToBlockdev(self.unique_id) |
2265 |
|
2266 |
try:
|
2267 |
st = os.stat(self.dev_path)
|
2268 |
except OSError, err: |
2269 |
logging.error("Error stat()'ing %s: %s", self.dev_path, str(err)) |
2270 |
return False |
2271 |
|
2272 |
if not stat.S_ISBLK(st.st_mode): |
2273 |
logging.error("%s is not a block device", self.dev_path) |
2274 |
return False |
2275 |
|
2276 |
self.major = os.major(st.st_rdev)
|
2277 |
self.minor = os.minor(st.st_rdev)
|
2278 |
self.attached = True |
2279 |
|
2280 |
return True |
2281 |
|
2282 |
def _MapVolumeToBlockdev(self, unique_id): |
2283 |
"""Maps existing rbd Volumes (Images) to block devices
|
2284 |
|
2285 |
This method should be idempotent if the mapping already exists.
|
2286 |
If the mapping exists it returns the device path.
|
2287 |
|
2288 |
If the mapping doesn't exist:
|
2289 |
Maps the existing Volume (Image) named `name' (inside the RADOS
|
2290 |
pool named `pool'), to a block device e.g. /dev/rbd{X}. Then
|
2291 |
returns it's device path.
|
2292 |
|
2293 |
"""
|
2294 |
pool, name = unique_id |
2295 |
|
2296 |
# Check if the mapping already exists
|
2297 |
cmd1 = "rbd showmapped"
|
2298 |
result = utils.RunCmd(cmd1) |
2299 |
if result.failed:
|
2300 |
_ThrowError("rbd showmapped failed (%s): %s",
|
2301 |
result.fail_reason, result.output) |
2302 |
else:
|
2303 |
cmd2 = "echo '%s' | grep %s | grep %s" % (result.output, pool, name)
|
2304 |
result = utils.RunCmd(cmd2) |
2305 |
if not result.failed: |
2306 |
# The mapping already exists.
|
2307 |
# Parse the result and return the rbd device
|
2308 |
try:
|
2309 |
rbd_dev = re.search("(/dev/rbd\d+)", result.output).group(1) |
2310 |
except:
|
2311 |
# maybe we can add an assert here
|
2312 |
_ThrowError("You shouldn't get here")
|
2313 |
|
2314 |
return rbd_dev
|
2315 |
|
2316 |
else:
|
2317 |
# The mapping doesn't exist. Create it
|
2318 |
cmd = "rbd map -p %s %s" % (pool, name)
|
2319 |
result = utils.RunCmd(cmd) |
2320 |
if result.failed:
|
2321 |
_ThrowError("rbd map failed (%s): %s",
|
2322 |
result.fail_reason, result.output) |
2323 |
# Use rbd showmapped again to find the rbd device
|
2324 |
# the image was mapped to
|
2325 |
cmd3 = "rbd showmapped | grep %s | grep %s" % (pool, name)
|
2326 |
result = utils.RunCmd(cmd3) |
2327 |
if result.failed:
|
2328 |
_ThrowError("Can't find mapped device. "
|
2329 |
"rbd showmapped failed (%s): %s",
|
2330 |
result.fail_reason, result.output) |
2331 |
try:
|
2332 |
rbd_dev = re.search("(/dev/rbd\d+)", result.output).group(1) |
2333 |
except:
|
2334 |
# maybe we can add an assert here
|
2335 |
_ThrowError("You shouldn't get here")
|
2336 |
|
2337 |
# The device was successfully mapped. Return it
|
2338 |
return rbd_dev
|
2339 |
|
2340 |
def Assemble(self): |
2341 |
"""Assemble the device.
|
2342 |
|
2343 |
"""
|
2344 |
pass
|
2345 |
|
2346 |
def Shutdown(self): |
2347 |
"""Shutdown the device.
|
2348 |
|
2349 |
"""
|
2350 |
# Unmap the rbd device
|
2351 |
# !doesn't unmap during migration because the
|
2352 |
# !shutdown method is never called if the machine was up
|
2353 |
# TODO: Fix cmdlib.py to shutdown the machine in the source node
|
2354 |
# once the migration ends successfully
|
2355 |
if not self.minor and not self.Attach(): |
2356 |
# the rbd device doesn't exist
|
2357 |
return
|
2358 |
|
2359 |
# Unmap the block device from the Volume
|
2360 |
self._UnmapVolumeFromBlockdev(self.unique_id) |
2361 |
|
2362 |
self.minor = None |
2363 |
self.dev_path = None |
2364 |
|
2365 |
def _UnmapVolumeFromBlockdev(self, unique_id): |
2366 |
"""Unmaps the rbd device from the Volume it is mapped
|
2367 |
|
2368 |
Unmaps the rbd device from the Volume (Image) it was
|
2369 |
previously mapped to. This method should be idempotent if
|
2370 |
the Volume isn't mapped.
|
2371 |
|
2372 |
"""
|
2373 |
pool, name = unique_id |
2374 |
|
2375 |
# Check if the mapping already exists
|
2376 |
cmd1 = "rbd showmapped"
|
2377 |
result = utils.RunCmd(cmd1) |
2378 |
if result.failed:
|
2379 |
_ThrowError("rbd showmapped failed [during unmap](%s): %s",
|
2380 |
result.fail_reason, result.output) |
2381 |
else:
|
2382 |
cmd2 = "echo '%s' | grep %s | grep %s" % (result.output, pool, name)
|
2383 |
result = utils.RunCmd(cmd2) |
2384 |
if not result.failed: |
2385 |
# The mapping already exists.
|
2386 |
# Parse the result to find the rbd device
|
2387 |
try:
|
2388 |
rbd_dev = re.search("(/dev/rbd\d+)", result.output).group(1) |
2389 |
except:
|
2390 |
# maybe we can add an assert here
|
2391 |
_ThrowError("You shouldn't get here")
|
2392 |
|
2393 |
# Unmap the rbd device
|
2394 |
cmd = "rbd unmap %s" % rbd_dev
|
2395 |
result = utils.RunCmd(cmd) |
2396 |
if result.failed:
|
2397 |
_ThrowError("rbd unmap failed (%s): %s",
|
2398 |
result.fail_reason, result.output) |
2399 |
|
2400 |
else:
|
2401 |
# The mapping doesn't exist. Do nothing
|
2402 |
pass
|
2403 |
|
2404 |
def Open(self, force=False): |
2405 |
"""Make the device ready for I/O.
|
2406 |
|
2407 |
"""
|
2408 |
pass
|
2409 |
|
2410 |
def Close(self): |
2411 |
"""Notifies that the device will no longer be used for I/O.
|
2412 |
|
2413 |
"""
|
2414 |
pass
|
2415 |
|
2416 |
def Grow(self, amount): |
2417 |
"""Grow the volume.
|
2418 |
|
2419 |
Grow the rbd Volume by `amount' megabytes.
|
2420 |
Total block size will be: size + amount
|
2421 |
|
2422 |
"""
|
2423 |
if not self.Attach(): |
2424 |
_ThrowError("Can't attach to rbd device during Grow()")
|
2425 |
|
2426 |
rbd_pool, rbd_name = self.unique_id
|
2427 |
new_size = self.size + amount
|
2428 |
|
2429 |
# Resize the rbd Volume (Image) inside the RADOS cluster
|
2430 |
cmd = "rbd resize -p %s %s --size %s" % (rbd_pool, rbd_name, new_size)
|
2431 |
result = utils.RunCmd(cmd) |
2432 |
if result.failed:
|
2433 |
_ThrowError("rbd resize failed (%s): %s",
|
2434 |
result.fail_reason, result.output) |
2435 |
|
2436 |
|
2437 |
DEV_MAP = { |
2438 |
constants.LD_LV: LogicalVolume, |
2439 |
constants.LD_DRBD8: DRBD8, |
2440 |
constants.LD_BLOCKDEV: PersistentBlockDevice, |
2441 |
constants.LD_RBD: RADOSBlockDevice, |
2442 |
} |
2443 |
|
2444 |
if constants.ENABLE_FILE_STORAGE or constants.ENABLE_SHARED_FILE_STORAGE: |
2445 |
DEV_MAP[constants.LD_FILE] = FileStorage |
2446 |
|
2447 |
|
2448 |
def FindDevice(dev_type, unique_id, children, size): |
2449 |
"""Search for an existing, assembled device.
|
2450 |
|
2451 |
This will succeed only if the device exists and is assembled, but it
|
2452 |
does not do any actions in order to activate the device.
|
2453 |
|
2454 |
"""
|
2455 |
if dev_type not in DEV_MAP: |
2456 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
2457 |
device = DEV_MAP[dev_type](unique_id, children, size) |
2458 |
if not device.attached: |
2459 |
return None |
2460 |
return device
|
2461 |
|
2462 |
|
2463 |
def Assemble(dev_type, unique_id, children, size): |
2464 |
"""Try to attach or assemble an existing device.
|
2465 |
|
2466 |
This will attach to assemble the device, as needed, to bring it
|
2467 |
fully up. It must be safe to run on already-assembled devices.
|
2468 |
|
2469 |
"""
|
2470 |
if dev_type not in DEV_MAP: |
2471 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
2472 |
device = DEV_MAP[dev_type](unique_id, children, size) |
2473 |
device.Assemble() |
2474 |
return device
|
2475 |
|
2476 |
|
2477 |
def Create(dev_type, unique_id, children, size): |
2478 |
"""Create a device.
|
2479 |
|
2480 |
"""
|
2481 |
if dev_type not in DEV_MAP: |
2482 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
2483 |
device = DEV_MAP[dev_type].Create(unique_id, children, size) |
2484 |
return device
|