root / lib / bdev.py @ 4a72cc75
History | View | Annotate | Download (45.7 kB)
1 |
#!/usr/bin/python
|
---|---|
2 |
#
|
3 |
|
4 |
# Copyright (C) 2006, 2007 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 |
|
28 |
from ganeti import utils |
29 |
from ganeti import logger |
30 |
from ganeti import errors |
31 |
|
32 |
|
33 |
class BlockDev(object): |
34 |
"""Block device abstract class.
|
35 |
|
36 |
A block device can be in the following states:
|
37 |
- not existing on the system, and by `Create()` it goes into:
|
38 |
- existing but not setup/not active, and by `Assemble()` goes into:
|
39 |
- active read-write and by `Open()` it goes into
|
40 |
- online (=used, or ready for use)
|
41 |
|
42 |
A device can also be online but read-only, however we are not using
|
43 |
the readonly state (MD and LV have it, if needed in the future)
|
44 |
and we are usually looking at this like at a stack, so it's easier
|
45 |
to conceptualise the transition from not-existing to online and back
|
46 |
like a linear one.
|
47 |
|
48 |
The many different states of the device are due to the fact that we
|
49 |
need to cover many device types:
|
50 |
- logical volumes are created, lvchange -a y $lv, and used
|
51 |
- md arrays are created or assembled and used
|
52 |
- drbd devices are attached to a local disk/remote peer and made primary
|
53 |
|
54 |
The status of the device can be examined by `GetStatus()`, which
|
55 |
returns a numerical value, depending on the position in the
|
56 |
transition stack of the device.
|
57 |
|
58 |
A block device is identified by three items:
|
59 |
- the /dev path of the device (dynamic)
|
60 |
- a unique ID of the device (static)
|
61 |
- it's major/minor pair (dynamic)
|
62 |
|
63 |
Not all devices implement both the first two as distinct items. LVM
|
64 |
logical volumes have their unique ID (the pair volume group, logical
|
65 |
volume name) in a 1-to-1 relation to the dev path. For MD devices,
|
66 |
the /dev path is dynamic and the unique ID is the UUID generated at
|
67 |
array creation plus the slave list. For DRBD devices, the /dev path
|
68 |
is again dynamic and the unique id is the pair (host1, dev1),
|
69 |
(host2, dev2).
|
70 |
|
71 |
You can get to a device in two ways:
|
72 |
- creating the (real) device, which returns you
|
73 |
an attached instance (lvcreate, mdadm --create)
|
74 |
- attaching of a python instance to an existing (real) device
|
75 |
|
76 |
The second point, the attachement to a device, is different
|
77 |
depending on whether the device is assembled or not. At init() time,
|
78 |
we search for a device with the same unique_id as us. If found,
|
79 |
good. It also means that the device is already assembled. If not,
|
80 |
after assembly we'll have our correct major/minor.
|
81 |
|
82 |
"""
|
83 |
STATUS_UNKNOWN = 0
|
84 |
STATUS_EXISTING = 1
|
85 |
STATUS_STANDBY = 2
|
86 |
STATUS_ONLINE = 3
|
87 |
|
88 |
STATUS_MAP = { |
89 |
STATUS_UNKNOWN: "unknown",
|
90 |
STATUS_EXISTING: "existing",
|
91 |
STATUS_STANDBY: "ready for use",
|
92 |
STATUS_ONLINE: "online",
|
93 |
} |
94 |
|
95 |
|
96 |
def __init__(self, unique_id, children): |
97 |
self._children = children
|
98 |
self.dev_path = None |
99 |
self.unique_id = unique_id
|
100 |
self.major = None |
101 |
self.minor = None |
102 |
|
103 |
|
104 |
def Assemble(self): |
105 |
"""Assemble the device from its components.
|
106 |
|
107 |
If this is a plain block device (e.g. LVM) than assemble does
|
108 |
nothing, as the LVM has no children and we don't put logical
|
109 |
volumes offline.
|
110 |
|
111 |
One guarantee is that after the device has been assembled, it
|
112 |
knows its major/minor numbers. This allows other devices (usually
|
113 |
parents) to probe correctly for their children.
|
114 |
|
115 |
"""
|
116 |
status = True
|
117 |
for child in self._children: |
118 |
if not isinstance(child, BlockDev): |
119 |
raise TypeError("Invalid child passed of type '%s'" % type(child)) |
120 |
if not status: |
121 |
break
|
122 |
status = status and child.Assemble()
|
123 |
if not status: |
124 |
break
|
125 |
status = status and child.Open()
|
126 |
|
127 |
if not status: |
128 |
for child in self._children: |
129 |
child.Shutdown() |
130 |
return status
|
131 |
|
132 |
|
133 |
def Attach(self): |
134 |
"""Find a device which matches our config and attach to it.
|
135 |
|
136 |
"""
|
137 |
raise NotImplementedError |
138 |
|
139 |
|
140 |
def Close(self): |
141 |
"""Notifies that the device will no longer be used for I/O.
|
142 |
|
143 |
"""
|
144 |
raise NotImplementedError |
145 |
|
146 |
|
147 |
@classmethod
|
148 |
def Create(cls, unique_id, children, size): |
149 |
"""Create the device.
|
150 |
|
151 |
If the device cannot be created, it will return None
|
152 |
instead. Error messages go to the logging system.
|
153 |
|
154 |
Note that for some devices, the unique_id is used, and for other,
|
155 |
the children. The idea is that these two, taken together, are
|
156 |
enough for both creation and assembly (later).
|
157 |
|
158 |
"""
|
159 |
raise NotImplementedError |
160 |
|
161 |
|
162 |
def Remove(self): |
163 |
"""Remove this device.
|
164 |
|
165 |
This makes sense only for some of the device types: LV and to a
|
166 |
lesser degree, md devices. Also note that if the device can't
|
167 |
attach, the removal can't be completed.
|
168 |
|
169 |
"""
|
170 |
raise NotImplementedError |
171 |
|
172 |
|
173 |
def GetStatus(self): |
174 |
"""Return the status of the device.
|
175 |
|
176 |
"""
|
177 |
raise NotImplementedError |
178 |
|
179 |
|
180 |
def Open(self, force=False): |
181 |
"""Make the device ready for use.
|
182 |
|
183 |
This makes the device ready for I/O. For now, just the DRBD
|
184 |
devices need this.
|
185 |
|
186 |
The force parameter signifies that if the device has any kind of
|
187 |
--force thing, it should be used, we know what we are doing.
|
188 |
|
189 |
"""
|
190 |
raise NotImplementedError |
191 |
|
192 |
|
193 |
def Shutdown(self): |
194 |
"""Shut down the device, freeing its children.
|
195 |
|
196 |
This undoes the `Assemble()` work, except for the child
|
197 |
assembling; as such, the children on the device are still
|
198 |
assembled after this call.
|
199 |
|
200 |
"""
|
201 |
raise NotImplementedError |
202 |
|
203 |
|
204 |
def SetSyncSpeed(self, speed): |
205 |
"""Adjust the sync speed of the mirror.
|
206 |
|
207 |
In case this is not a mirroring device, this is no-op.
|
208 |
|
209 |
"""
|
210 |
result = True
|
211 |
if self._children: |
212 |
for child in self._children: |
213 |
result = result and child.SetSyncSpeed(speed)
|
214 |
return result
|
215 |
|
216 |
|
217 |
def GetSyncStatus(self): |
218 |
"""Returns the sync status of the device.
|
219 |
|
220 |
If this device is a mirroring device, this function returns the
|
221 |
status of the mirror.
|
222 |
|
223 |
Returns:
|
224 |
(sync_percent, estimated_time, is_degraded)
|
225 |
|
226 |
If sync_percent is None, it means all is ok
|
227 |
If estimated_time is None, it means we can't estimate
|
228 |
the time needed, otherwise it's the time left in seconds
|
229 |
If is_degraded is True, it means the device is missing
|
230 |
redundancy. This is usually a sign that something went wrong in
|
231 |
the device setup, if sync_percent is None.
|
232 |
|
233 |
"""
|
234 |
return None, None, False |
235 |
|
236 |
|
237 |
def CombinedSyncStatus(self): |
238 |
"""Calculate the mirror status recursively for our children.
|
239 |
|
240 |
The return value is the same as for `GetSyncStatus()` except the
|
241 |
minimum percent and maximum time are calculated across our
|
242 |
children.
|
243 |
|
244 |
"""
|
245 |
min_percent, max_time, is_degraded = self.GetSyncStatus()
|
246 |
if self._children: |
247 |
for child in self._children: |
248 |
c_percent, c_time, c_degraded = child.GetSyncStatus() |
249 |
if min_percent is None: |
250 |
min_percent = c_percent |
251 |
elif c_percent is not None: |
252 |
min_percent = min(min_percent, c_percent)
|
253 |
if max_time is None: |
254 |
max_time = c_time |
255 |
elif c_time is not None: |
256 |
max_time = max(max_time, c_time)
|
257 |
is_degraded = is_degraded or c_degraded
|
258 |
return min_percent, max_time, is_degraded
|
259 |
|
260 |
|
261 |
def SetInfo(self, text): |
262 |
"""Update metadata with info text.
|
263 |
|
264 |
Only supported for some device types.
|
265 |
|
266 |
"""
|
267 |
for child in self._children: |
268 |
child.SetInfo(text) |
269 |
|
270 |
|
271 |
def __repr__(self): |
272 |
return ("<%s: unique_id: %s, children: %s, %s:%s, %s>" % |
273 |
(self.__class__, self.unique_id, self._children, |
274 |
self.major, self.minor, self.dev_path)) |
275 |
|
276 |
|
277 |
class LogicalVolume(BlockDev): |
278 |
"""Logical Volume block device.
|
279 |
|
280 |
"""
|
281 |
def __init__(self, unique_id, children): |
282 |
"""Attaches to a LV device.
|
283 |
|
284 |
The unique_id is a tuple (vg_name, lv_name)
|
285 |
|
286 |
"""
|
287 |
super(LogicalVolume, self).__init__(unique_id, children) |
288 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
289 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
290 |
self._vg_name, self._lv_name = unique_id |
291 |
self.dev_path = "/dev/%s/%s" % (self._vg_name, self._lv_name) |
292 |
self.Attach()
|
293 |
|
294 |
|
295 |
@classmethod
|
296 |
def Create(cls, unique_id, children, size): |
297 |
"""Create a new logical volume.
|
298 |
|
299 |
"""
|
300 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 2: |
301 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
302 |
vg_name, lv_name = unique_id |
303 |
pvs_info = cls.GetPVInfo(vg_name) |
304 |
if not pvs_info: |
305 |
raise errors.BlockDeviceError("Can't compute PV info for vg %s" % |
306 |
vg_name) |
307 |
pvs_info.sort() |
308 |
pvs_info.reverse() |
309 |
|
310 |
pvlist = [ pv[1] for pv in pvs_info ] |
311 |
free_size = sum([ pv[0] for pv in pvs_info ]) |
312 |
|
313 |
# The size constraint should have been checked from the master before
|
314 |
# calling the create function.
|
315 |
if free_size < size:
|
316 |
raise errors.BlockDeviceError("Not enough free space: required %s," |
317 |
" available %s" % (size, free_size))
|
318 |
result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-n%s" % lv_name, |
319 |
vg_name] + pvlist) |
320 |
if result.failed:
|
321 |
raise errors.BlockDeviceError(result.fail_reason)
|
322 |
return LogicalVolume(unique_id, children)
|
323 |
|
324 |
@staticmethod
|
325 |
def GetPVInfo(vg_name): |
326 |
"""Get the free space info for PVs in a volume group.
|
327 |
|
328 |
Args:
|
329 |
vg_name: the volume group name
|
330 |
|
331 |
Returns:
|
332 |
list of (free_space, name) with free_space in mebibytes
|
333 |
|
334 |
"""
|
335 |
command = ["pvs", "--noheadings", "--nosuffix", "--units=m", |
336 |
"-opv_name,vg_name,pv_free,pv_attr", "--unbuffered", |
337 |
"--separator=:"]
|
338 |
result = utils.RunCmd(command) |
339 |
if result.failed:
|
340 |
logger.Error("Can't get the PV information: %s" % result.fail_reason)
|
341 |
return None |
342 |
data = [] |
343 |
for line in result.stdout.splitlines(): |
344 |
fields = line.strip().split(':')
|
345 |
if len(fields) != 4: |
346 |
logger.Error("Can't parse pvs output: line '%s'" % line)
|
347 |
return None |
348 |
# skip over pvs from another vg or ones which are not allocatable
|
349 |
if fields[1] != vg_name or fields[3][0] != 'a': |
350 |
continue
|
351 |
data.append((float(fields[2]), fields[0])) |
352 |
|
353 |
return data
|
354 |
|
355 |
def Remove(self): |
356 |
"""Remove this logical volume.
|
357 |
|
358 |
"""
|
359 |
if not self.minor and not self.Attach(): |
360 |
# the LV does not exist
|
361 |
return True |
362 |
result = utils.RunCmd(["lvremove", "-f", "%s/%s" % |
363 |
(self._vg_name, self._lv_name)]) |
364 |
if result.failed:
|
365 |
logger.Error("Can't lvremove: %s" % result.fail_reason)
|
366 |
|
367 |
return not result.failed |
368 |
|
369 |
|
370 |
def Attach(self): |
371 |
"""Attach to an existing LV.
|
372 |
|
373 |
This method will try to see if an existing and active LV exists
|
374 |
which matches the our name. If so, its major/minor will be
|
375 |
recorded.
|
376 |
|
377 |
"""
|
378 |
result = utils.RunCmd(["lvdisplay", self.dev_path]) |
379 |
if result.failed:
|
380 |
logger.Error("Can't find LV %s: %s" %
|
381 |
(self.dev_path, result.fail_reason))
|
382 |
return False |
383 |
match = re.compile("^ *Block device *([0-9]+):([0-9]+).*$")
|
384 |
for line in result.stdout.splitlines(): |
385 |
match_result = match.match(line) |
386 |
if match_result:
|
387 |
self.major = int(match_result.group(1)) |
388 |
self.minor = int(match_result.group(2)) |
389 |
return True |
390 |
return False |
391 |
|
392 |
|
393 |
def Assemble(self): |
394 |
"""Assemble the device.
|
395 |
|
396 |
This is a no-op for the LV device type. Eventually, we could
|
397 |
lvchange -ay here if we see that the LV is not active.
|
398 |
|
399 |
"""
|
400 |
return True |
401 |
|
402 |
|
403 |
def Shutdown(self): |
404 |
"""Shutdown the device.
|
405 |
|
406 |
This is a no-op for the LV device type, as we don't deactivate the
|
407 |
volumes on shutdown.
|
408 |
|
409 |
"""
|
410 |
return True |
411 |
|
412 |
|
413 |
def GetStatus(self): |
414 |
"""Return the status of the device.
|
415 |
|
416 |
Logical volumes will can be in all four states, although we don't
|
417 |
deactivate (lvchange -an) them when shutdown, so STATUS_EXISTING
|
418 |
should not be seen for our devices.
|
419 |
|
420 |
"""
|
421 |
result = utils.RunCmd(["lvs", "--noheadings", "-olv_attr", self.dev_path]) |
422 |
if result.failed:
|
423 |
logger.Error("Can't display lv: %s" % result.fail_reason)
|
424 |
return self.STATUS_UNKNOWN |
425 |
out = result.stdout.strip() |
426 |
# format: type/permissions/alloc/fixed_minor/state/open
|
427 |
if len(out) != 6: |
428 |
return self.STATUS_UNKNOWN |
429 |
#writable = (out[1] == "w")
|
430 |
active = (out[4] == "a") |
431 |
online = (out[5] == "o") |
432 |
if online:
|
433 |
retval = self.STATUS_ONLINE
|
434 |
elif active:
|
435 |
retval = self.STATUS_STANDBY
|
436 |
else:
|
437 |
retval = self.STATUS_EXISTING
|
438 |
|
439 |
return retval
|
440 |
|
441 |
|
442 |
def Open(self, force=False): |
443 |
"""Make the device ready for I/O.
|
444 |
|
445 |
This is a no-op for the LV device type.
|
446 |
|
447 |
"""
|
448 |
return True |
449 |
|
450 |
|
451 |
def Close(self): |
452 |
"""Notifies that the device will no longer be used for I/O.
|
453 |
|
454 |
This is a no-op for the LV device type.
|
455 |
|
456 |
"""
|
457 |
return True |
458 |
|
459 |
|
460 |
def Snapshot(self, size): |
461 |
"""Create a snapshot copy of an lvm block device.
|
462 |
|
463 |
"""
|
464 |
snap_name = self._lv_name + ".snap" |
465 |
|
466 |
# remove existing snapshot if found
|
467 |
snap = LogicalVolume((self._vg_name, snap_name), None) |
468 |
snap.Remove() |
469 |
|
470 |
pvs_info = self.GetPVInfo(self._vg_name) |
471 |
if not pvs_info: |
472 |
raise errors.BlockDeviceError("Can't compute PV info for vg %s" % |
473 |
self._vg_name)
|
474 |
pvs_info.sort() |
475 |
pvs_info.reverse() |
476 |
free_size, pv_name = pvs_info[0]
|
477 |
if free_size < size:
|
478 |
raise errors.BlockDeviceError("Not enough free space: required %s," |
479 |
" available %s" % (size, free_size))
|
480 |
|
481 |
result = utils.RunCmd(["lvcreate", "-L%dm" % size, "-s", |
482 |
"-n%s" % snap_name, self.dev_path]) |
483 |
if result.failed:
|
484 |
raise errors.BlockDeviceError("command: %s error: %s" % |
485 |
(result.cmd, result.fail_reason)) |
486 |
|
487 |
return snap_name
|
488 |
|
489 |
|
490 |
def SetInfo(self, text): |
491 |
"""Update metadata with info text.
|
492 |
|
493 |
"""
|
494 |
BlockDev.SetInfo(self, text)
|
495 |
|
496 |
# Replace invalid characters
|
497 |
text = re.sub('^[^A-Za-z0-9_+.]', '_', text) |
498 |
text = re.sub('[^-A-Za-z0-9_+.]', '_', text) |
499 |
|
500 |
# Only up to 128 characters are allowed
|
501 |
text = text[:128]
|
502 |
|
503 |
result = utils.RunCmd(["lvchange", "--addtag", text, |
504 |
self.dev_path])
|
505 |
if result.failed:
|
506 |
raise errors.BlockDeviceError("Command: %s error: %s" % |
507 |
(result.cmd, result.fail_reason)) |
508 |
|
509 |
|
510 |
class MDRaid1(BlockDev): |
511 |
"""raid1 device implemented via md.
|
512 |
|
513 |
"""
|
514 |
def __init__(self, unique_id, children): |
515 |
super(MDRaid1, self).__init__(unique_id, children) |
516 |
self.major = 9 |
517 |
self.Attach()
|
518 |
|
519 |
|
520 |
def Attach(self): |
521 |
"""Find an array which matches our config and attach to it.
|
522 |
|
523 |
This tries to find a MD array which has the same UUID as our own.
|
524 |
|
525 |
"""
|
526 |
minor = self._FindMDByUUID(self.unique_id) |
527 |
if minor is not None: |
528 |
self._SetFromMinor(minor)
|
529 |
else:
|
530 |
self.minor = None |
531 |
self.dev_path = None |
532 |
|
533 |
return (minor is not None) |
534 |
|
535 |
|
536 |
@staticmethod
|
537 |
def _GetUsedDevs(): |
538 |
"""Compute the list of in-use MD devices.
|
539 |
|
540 |
It doesn't matter if the used device have other raid level, just
|
541 |
that they are in use.
|
542 |
|
543 |
"""
|
544 |
mdstat = open("/proc/mdstat", "r") |
545 |
data = mdstat.readlines() |
546 |
mdstat.close() |
547 |
|
548 |
used_md = {} |
549 |
valid_line = re.compile("^md([0-9]+) : .*$")
|
550 |
for line in data: |
551 |
match = valid_line.match(line) |
552 |
if match:
|
553 |
md_no = int(match.group(1)) |
554 |
used_md[md_no] = line |
555 |
|
556 |
return used_md
|
557 |
|
558 |
|
559 |
@staticmethod
|
560 |
def _GetDevInfo(minor): |
561 |
"""Get info about a MD device.
|
562 |
|
563 |
Currently only uuid is returned.
|
564 |
|
565 |
"""
|
566 |
result = utils.RunCmd(["mdadm", "-D", "/dev/md%d" % minor]) |
567 |
if result.failed:
|
568 |
logger.Error("Can't display md: %s" % result.fail_reason)
|
569 |
return None |
570 |
retval = {} |
571 |
for line in result.stdout.splitlines(): |
572 |
line = line.strip() |
573 |
kv = line.split(" : ", 1) |
574 |
if kv:
|
575 |
if kv[0] == "UUID": |
576 |
retval["uuid"] = kv[1].split()[0] |
577 |
elif kv[0] == "State": |
578 |
retval["state"] = kv[1].split(", ") |
579 |
return retval
|
580 |
|
581 |
|
582 |
@staticmethod
|
583 |
def _FindUnusedMinor(): |
584 |
"""Compute an unused MD minor.
|
585 |
|
586 |
This code assumes that there are 256 minors only.
|
587 |
|
588 |
"""
|
589 |
used_md = MDRaid1._GetUsedDevs() |
590 |
i = 0
|
591 |
while i < 256: |
592 |
if i not in used_md: |
593 |
break
|
594 |
i += 1
|
595 |
if i == 256: |
596 |
logger.Error("Critical: Out of md minor numbers.")
|
597 |
return None |
598 |
return i
|
599 |
|
600 |
|
601 |
@classmethod
|
602 |
def _FindMDByUUID(cls, uuid): |
603 |
"""Find the minor of an MD array with a given UUID.
|
604 |
|
605 |
"""
|
606 |
md_list = cls._GetUsedDevs() |
607 |
for minor in md_list: |
608 |
info = cls._GetDevInfo(minor) |
609 |
if info and info["uuid"] == uuid: |
610 |
return minor
|
611 |
return None |
612 |
|
613 |
|
614 |
@staticmethod
|
615 |
def _ZeroSuperblock(dev_path): |
616 |
"""Zero the possible locations for an MD superblock.
|
617 |
|
618 |
The zero-ing can't be done via ``mdadm --zero-superblock`` as that
|
619 |
fails in versions 2.x with the same error code as non-writable
|
620 |
device.
|
621 |
|
622 |
The superblocks are located at (negative values are relative to
|
623 |
the end of the block device):
|
624 |
- -128k to end for version 0.90 superblock
|
625 |
- -8k to -12k for version 1.0 superblock (included in the above)
|
626 |
- 0k to 4k for version 1.1 superblock
|
627 |
- 4k to 8k for version 1.2 superblock
|
628 |
|
629 |
To cover all situations, the zero-ing will be:
|
630 |
- 0k to 128k
|
631 |
- -128k to end
|
632 |
|
633 |
As such, the minimum device size must be 128k, otherwise we'll get
|
634 |
I/O errors.
|
635 |
|
636 |
Note that this function depends on the fact that one can open,
|
637 |
read and write block devices normally.
|
638 |
|
639 |
"""
|
640 |
overwrite_size = 128 * 1024 |
641 |
empty_buf = '\0' * overwrite_size
|
642 |
fd = open(dev_path, "r+") |
643 |
try:
|
644 |
fd.seek(0, 0) |
645 |
p1 = fd.tell() |
646 |
fd.write(empty_buf) |
647 |
p2 = fd.tell() |
648 |
logger.Debug("Zeroed %s from %d to %d" % (dev_path, p1, p2))
|
649 |
fd.seek(-overwrite_size, 2)
|
650 |
p1 = fd.tell() |
651 |
fd.write(empty_buf) |
652 |
p2 = fd.tell() |
653 |
logger.Debug("Zeroed %s from %d to %d" % (dev_path, p1, p2))
|
654 |
finally:
|
655 |
fd.close() |
656 |
|
657 |
@classmethod
|
658 |
def Create(cls, unique_id, children, size): |
659 |
"""Create a new MD raid1 array.
|
660 |
|
661 |
"""
|
662 |
if not isinstance(children, (tuple, list)): |
663 |
raise ValueError("Invalid setup data for MDRaid1 dev: %s" % |
664 |
str(children))
|
665 |
for i in children: |
666 |
if not isinstance(i, BlockDev): |
667 |
raise ValueError("Invalid member in MDRaid1 dev: %s" % type(i)) |
668 |
for i in children: |
669 |
try:
|
670 |
cls._ZeroSuperblock(i.dev_path) |
671 |
except EnvironmentError, err: |
672 |
logger.Error("Can't zero superblock for %s: %s" %
|
673 |
(i.dev_path, str(err)))
|
674 |
return None |
675 |
minor = cls._FindUnusedMinor() |
676 |
result = utils.RunCmd(["mdadm", "--create", "/dev/md%d" % minor, |
677 |
"--auto=yes", "--force", "-l1", |
678 |
"-n%d" % len(children)] + |
679 |
[dev.dev_path for dev in children]) |
680 |
|
681 |
if result.failed:
|
682 |
logger.Error("Can't create md: %s: %s" % (result.fail_reason,
|
683 |
result.output)) |
684 |
return None |
685 |
info = cls._GetDevInfo(minor) |
686 |
if not info or not "uuid" in info: |
687 |
logger.Error("Wrong information returned from mdadm -D: %s" % str(info)) |
688 |
return None |
689 |
return MDRaid1(info["uuid"], children) |
690 |
|
691 |
|
692 |
def Remove(self): |
693 |
"""Stub remove function for MD RAID 1 arrays.
|
694 |
|
695 |
We don't remove the superblock right now. Mark a to do.
|
696 |
|
697 |
"""
|
698 |
#TODO: maybe zero superblock on child devices?
|
699 |
return self.Shutdown() |
700 |
|
701 |
|
702 |
def AddChild(self, device): |
703 |
"""Add a new member to the md raid1.
|
704 |
|
705 |
"""
|
706 |
if self.minor is None and not self.Attach(): |
707 |
raise errors.BlockDeviceError("Can't attach to device") |
708 |
if device.dev_path is None: |
709 |
raise errors.BlockDeviceError("New child is not initialised") |
710 |
result = utils.RunCmd(["mdadm", "-a", self.dev_path, device.dev_path]) |
711 |
if result.failed:
|
712 |
raise errors.BlockDeviceError("Failed to add new device to array: %s" % |
713 |
result.output) |
714 |
new_len = len(self._children) + 1 |
715 |
result = utils.RunCmd(["mdadm", "--grow", self.dev_path, "-n", new_len]) |
716 |
if result.failed:
|
717 |
raise errors.BlockDeviceError("Can't grow md array: %s" % |
718 |
result.output) |
719 |
self._children.append(device)
|
720 |
|
721 |
|
722 |
def RemoveChild(self, dev_path): |
723 |
"""Remove member from the md raid1.
|
724 |
|
725 |
"""
|
726 |
if self.minor is None and not self.Attach(): |
727 |
raise errors.BlockDeviceError("Can't attach to device") |
728 |
if len(self._children) == 1: |
729 |
raise errors.BlockDeviceError("Can't reduce member when only one" |
730 |
" child left")
|
731 |
for device in self._children: |
732 |
if device.dev_path == dev_path:
|
733 |
break
|
734 |
else:
|
735 |
raise errors.BlockDeviceError("Can't find child with this path") |
736 |
new_len = len(self._children) - 1 |
737 |
result = utils.RunCmd(["mdadm", "-f", self.dev_path, dev_path]) |
738 |
if result.failed:
|
739 |
raise errors.BlockDeviceError("Failed to mark device as failed: %s" % |
740 |
result.output) |
741 |
|
742 |
# it seems here we need a short delay for MD to update its
|
743 |
# superblocks
|
744 |
time.sleep(0.5)
|
745 |
result = utils.RunCmd(["mdadm", "-r", self.dev_path, dev_path]) |
746 |
if result.failed:
|
747 |
raise errors.BlockDeviceError("Failed to remove device from array:" |
748 |
" %s" % result.output)
|
749 |
result = utils.RunCmd(["mdadm", "--grow", "--force", self.dev_path, |
750 |
"-n", new_len])
|
751 |
if result.failed:
|
752 |
raise errors.BlockDeviceError("Can't shrink md array: %s" % |
753 |
result.output) |
754 |
self._children.remove(device)
|
755 |
|
756 |
|
757 |
def GetStatus(self): |
758 |
"""Return the status of the device.
|
759 |
|
760 |
"""
|
761 |
self.Attach()
|
762 |
if self.minor is None: |
763 |
retval = self.STATUS_UNKNOWN
|
764 |
else:
|
765 |
retval = self.STATUS_ONLINE
|
766 |
return retval
|
767 |
|
768 |
|
769 |
def _SetFromMinor(self, minor): |
770 |
"""Set our parameters based on the given minor.
|
771 |
|
772 |
This sets our minor variable and our dev_path.
|
773 |
|
774 |
"""
|
775 |
self.minor = minor
|
776 |
self.dev_path = "/dev/md%d" % minor |
777 |
|
778 |
|
779 |
def Assemble(self): |
780 |
"""Assemble the MD device.
|
781 |
|
782 |
At this point we should have:
|
783 |
- list of children devices
|
784 |
- uuid
|
785 |
|
786 |
"""
|
787 |
result = super(MDRaid1, self).Assemble() |
788 |
if not result: |
789 |
return result
|
790 |
md_list = self._GetUsedDevs()
|
791 |
for minor in md_list: |
792 |
info = self._GetDevInfo(minor)
|
793 |
if info and info["uuid"] == self.unique_id: |
794 |
self._SetFromMinor(minor)
|
795 |
logger.Info("MD array %s already started" % str(self)) |
796 |
return True |
797 |
free_minor = self._FindUnusedMinor()
|
798 |
result = utils.RunCmd(["mdadm", "-A", "--auto=yes", "--uuid", |
799 |
self.unique_id, "/dev/md%d" % free_minor] + |
800 |
[bdev.dev_path for bdev in self._children]) |
801 |
if result.failed:
|
802 |
logger.Error("Can't assemble MD array: %s: %s" %
|
803 |
(result.fail_reason, result.output)) |
804 |
self.minor = None |
805 |
else:
|
806 |
self.minor = free_minor
|
807 |
return not result.failed |
808 |
|
809 |
|
810 |
def Shutdown(self): |
811 |
"""Tear down the MD array.
|
812 |
|
813 |
This does a 'mdadm --stop' so after this command, the array is no
|
814 |
longer available.
|
815 |
|
816 |
"""
|
817 |
if self.minor is None and not self.Attach(): |
818 |
logger.Info("MD object not attached to a device")
|
819 |
return True |
820 |
|
821 |
result = utils.RunCmd(["mdadm", "--stop", "/dev/md%d" % self.minor]) |
822 |
if result.failed:
|
823 |
logger.Error("Can't stop MD array: %s" % result.fail_reason)
|
824 |
return False |
825 |
self.minor = None |
826 |
self.dev_path = None |
827 |
return True |
828 |
|
829 |
|
830 |
def SetSyncSpeed(self, kbytes): |
831 |
"""Set the maximum sync speed for the MD array.
|
832 |
|
833 |
"""
|
834 |
result = super(MDRaid1, self).SetSyncSpeed(kbytes) |
835 |
if self.minor is None: |
836 |
logger.Error("MD array not attached to a device")
|
837 |
return False |
838 |
f = open("/sys/block/md%d/md/sync_speed_max" % self.minor, "w") |
839 |
try:
|
840 |
f.write("%d" % kbytes)
|
841 |
finally:
|
842 |
f.close() |
843 |
f = open("/sys/block/md%d/md/sync_speed_min" % self.minor, "w") |
844 |
try:
|
845 |
f.write("%d" % (kbytes/2)) |
846 |
finally:
|
847 |
f.close() |
848 |
return result
|
849 |
|
850 |
|
851 |
def GetSyncStatus(self): |
852 |
"""Returns the sync status of the device.
|
853 |
|
854 |
Returns:
|
855 |
(sync_percent, estimated_time)
|
856 |
|
857 |
If sync_percent is None, it means all is ok
|
858 |
If estimated_time is None, it means we can't esimate
|
859 |
the time needed, otherwise it's the time left in seconds
|
860 |
|
861 |
"""
|
862 |
if self.minor is None and not self.Attach(): |
863 |
raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") |
864 |
dev_info = self._GetDevInfo(self.minor) |
865 |
is_clean = ("state" in dev_info and |
866 |
len(dev_info["state"]) == 1 and |
867 |
dev_info["state"][0] in ("clean", "active")) |
868 |
sys_path = "/sys/block/md%s/md/" % self.minor |
869 |
f = file(sys_path + "sync_action") |
870 |
sync_status = f.readline().strip() |
871 |
f.close() |
872 |
if sync_status == "idle": |
873 |
return None, None, not is_clean |
874 |
f = file(sys_path + "sync_completed") |
875 |
sync_completed = f.readline().strip().split(" / ")
|
876 |
f.close() |
877 |
if len(sync_completed) != 2: |
878 |
return 0, None, not is_clean |
879 |
sync_done, sync_total = [float(i) for i in sync_completed] |
880 |
sync_percent = 100.0*sync_done/sync_total
|
881 |
f = file(sys_path + "sync_speed") |
882 |
sync_speed_k = int(f.readline().strip())
|
883 |
if sync_speed_k == 0: |
884 |
time_est = None
|
885 |
else:
|
886 |
time_est = (sync_total - sync_done) / 2 / sync_speed_k
|
887 |
return sync_percent, time_est, not is_clean |
888 |
|
889 |
|
890 |
def Open(self, force=False): |
891 |
"""Make the device ready for I/O.
|
892 |
|
893 |
This is a no-op for the MDRaid1 device type, although we could use
|
894 |
the 2.6.18's new array_state thing.
|
895 |
|
896 |
"""
|
897 |
return True |
898 |
|
899 |
|
900 |
def Close(self): |
901 |
"""Notifies that the device will no longer be used for I/O.
|
902 |
|
903 |
This is a no-op for the MDRaid1 device type, but see comment for
|
904 |
`Open()`.
|
905 |
|
906 |
"""
|
907 |
return True |
908 |
|
909 |
|
910 |
|
911 |
class DRBDev(BlockDev): |
912 |
"""DRBD block device.
|
913 |
|
914 |
This implements the local host part of the DRBD device, i.e. it
|
915 |
doesn't do anything to the supposed peer. If you need a fully
|
916 |
connected DRBD pair, you need to use this class on both hosts.
|
917 |
|
918 |
The unique_id for the drbd device is the (local_ip, local_port,
|
919 |
remote_ip, remote_port) tuple, and it must have two children: the
|
920 |
data device and the meta_device. The meta device is checked for
|
921 |
valid size and is zeroed on create.
|
922 |
|
923 |
"""
|
924 |
_DRBD_MAJOR = 147
|
925 |
_ST_UNCONFIGURED = "Unconfigured"
|
926 |
_ST_WFCONNECTION = "WFConnection"
|
927 |
_ST_CONNECTED = "Connected"
|
928 |
|
929 |
def __init__(self, unique_id, children): |
930 |
super(DRBDev, self).__init__(unique_id, children) |
931 |
self.major = self._DRBD_MAJOR |
932 |
if len(children) != 2: |
933 |
raise ValueError("Invalid configuration data %s" % str(children)) |
934 |
if not isinstance(unique_id, (tuple, list)) or len(unique_id) != 4: |
935 |
raise ValueError("Invalid configuration data %s" % str(unique_id)) |
936 |
self._lhost, self._lport, self._rhost, self._rport = unique_id |
937 |
self.Attach()
|
938 |
|
939 |
@staticmethod
|
940 |
def _DevPath(minor): |
941 |
"""Return the path to a drbd device for a given minor.
|
942 |
|
943 |
"""
|
944 |
return "/dev/drbd%d" % minor |
945 |
|
946 |
@staticmethod
|
947 |
def _GetProcData(): |
948 |
"""Return data from /proc/drbd.
|
949 |
|
950 |
"""
|
951 |
stat = open("/proc/drbd", "r") |
952 |
data = stat.read().splitlines() |
953 |
stat.close() |
954 |
return data
|
955 |
|
956 |
|
957 |
@classmethod
|
958 |
def _GetUsedDevs(cls): |
959 |
"""Compute the list of used DRBD devices.
|
960 |
|
961 |
"""
|
962 |
data = cls._GetProcData() |
963 |
|
964 |
used_devs = {} |
965 |
valid_line = re.compile("^ *([0-9]+): cs:([^ ]+).*$")
|
966 |
for line in data: |
967 |
match = valid_line.match(line) |
968 |
if not match: |
969 |
continue
|
970 |
minor = int(match.group(1)) |
971 |
state = match.group(2)
|
972 |
if state == cls._ST_UNCONFIGURED:
|
973 |
continue
|
974 |
used_devs[minor] = state, line |
975 |
|
976 |
return used_devs
|
977 |
|
978 |
|
979 |
@classmethod
|
980 |
def _FindUnusedMinor(cls): |
981 |
"""Find an unused DRBD device.
|
982 |
|
983 |
"""
|
984 |
data = cls._GetProcData() |
985 |
|
986 |
valid_line = re.compile("^ *([0-9]+): cs:Unconfigured$")
|
987 |
for line in data: |
988 |
match = valid_line.match(line) |
989 |
if match:
|
990 |
return int(match.group(1)) |
991 |
logger.Error("Error: no free drbd minors!")
|
992 |
return None |
993 |
|
994 |
|
995 |
@classmethod
|
996 |
def _GetDevInfo(cls, minor): |
997 |
"""Get details about a given DRBD minor.
|
998 |
|
999 |
This return, if available, the local backing device in (major,
|
1000 |
minor) formant and the local and remote (ip, port) information.
|
1001 |
|
1002 |
"""
|
1003 |
data = {} |
1004 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "show"]) |
1005 |
if result.failed:
|
1006 |
logger.Error("Can't display the drbd config: %s" % result.fail_reason)
|
1007 |
return data
|
1008 |
out = result.stdout |
1009 |
if out == "Not configured\n": |
1010 |
return data
|
1011 |
for line in out.splitlines(): |
1012 |
if "local_dev" not in data: |
1013 |
match = re.match("^Lower device: ([0-9]+):([0-9]+) .*$", line)
|
1014 |
if match:
|
1015 |
data["local_dev"] = (int(match.group(1)), int(match.group(2))) |
1016 |
continue
|
1017 |
if "meta_dev" not in data: |
1018 |
match = re.match("^Meta device: (([0-9]+):([0-9]+)|internal).*$", line)
|
1019 |
if match:
|
1020 |
if match.group(2) is not None and match.group(3) is not None: |
1021 |
# matched on the major/minor
|
1022 |
data["meta_dev"] = (int(match.group(2)), int(match.group(3))) |
1023 |
else:
|
1024 |
# matched on the "internal" string
|
1025 |
data["meta_dev"] = match.group(1) |
1026 |
# in this case, no meta_index is in the output
|
1027 |
data["meta_index"] = -1 |
1028 |
continue
|
1029 |
if "meta_index" not in data: |
1030 |
match = re.match("^Meta index: ([0-9]+).*$", line)
|
1031 |
if match:
|
1032 |
data["meta_index"] = int(match.group(1)) |
1033 |
continue
|
1034 |
if "local_addr" not in data: |
1035 |
match = re.match("^Local address: ([0-9.]+):([0-9]+)$", line)
|
1036 |
if match:
|
1037 |
data["local_addr"] = (match.group(1), int(match.group(2))) |
1038 |
continue
|
1039 |
if "remote_addr" not in data: |
1040 |
match = re.match("^Remote address: ([0-9.]+):([0-9]+)$", line)
|
1041 |
if match:
|
1042 |
data["remote_addr"] = (match.group(1), int(match.group(2))) |
1043 |
continue
|
1044 |
return data
|
1045 |
|
1046 |
|
1047 |
def _MatchesLocal(self, info): |
1048 |
"""Test if our local config matches with an existing device.
|
1049 |
|
1050 |
The parameter should be as returned from `_GetDevInfo()`. This
|
1051 |
method tests if our local backing device is the same as the one in
|
1052 |
the info parameter, in effect testing if we look like the given
|
1053 |
device.
|
1054 |
|
1055 |
"""
|
1056 |
if not ("local_dev" in info and "meta_dev" in info and |
1057 |
"meta_index" in info): |
1058 |
return False |
1059 |
|
1060 |
backend = self._children[0] |
1061 |
if backend is not None: |
1062 |
retval = (info["local_dev"] == (backend.major, backend.minor))
|
1063 |
else:
|
1064 |
retval = (info["local_dev"] == (0, 0)) |
1065 |
meta = self._children[1] |
1066 |
if meta is not None: |
1067 |
retval = retval and (info["meta_dev"] == (meta.major, meta.minor)) |
1068 |
retval = retval and (info["meta_index"] == 0) |
1069 |
else:
|
1070 |
retval = retval and (info["meta_dev"] == "internal" and |
1071 |
info["meta_index"] == -1) |
1072 |
return retval
|
1073 |
|
1074 |
|
1075 |
def _MatchesNet(self, info): |
1076 |
"""Test if our network config matches with an existing device.
|
1077 |
|
1078 |
The parameter should be as returned from `_GetDevInfo()`. This
|
1079 |
method tests if our network configuration is the same as the one
|
1080 |
in the info parameter, in effect testing if we look like the given
|
1081 |
device.
|
1082 |
|
1083 |
"""
|
1084 |
if (((self._lhost is None and not ("local_addr" in info)) and |
1085 |
(self._rhost is None and not ("remote_addr" in info)))): |
1086 |
return True |
1087 |
|
1088 |
if self._lhost is None: |
1089 |
return False |
1090 |
|
1091 |
if not ("local_addr" in info and |
1092 |
"remote_addr" in info): |
1093 |
return False |
1094 |
|
1095 |
retval = (info["local_addr"] == (self._lhost, self._lport)) |
1096 |
retval = (retval and
|
1097 |
info["remote_addr"] == (self._rhost, self._rport)) |
1098 |
return retval
|
1099 |
|
1100 |
|
1101 |
@staticmethod
|
1102 |
def _IsValidMeta(meta_device): |
1103 |
"""Check if the given meta device looks like a valid one.
|
1104 |
|
1105 |
This currently only check the size, which must be around
|
1106 |
128MiB.
|
1107 |
|
1108 |
"""
|
1109 |
result = utils.RunCmd(["blockdev", "--getsize", meta_device]) |
1110 |
if result.failed:
|
1111 |
logger.Error("Failed to get device size: %s" % result.fail_reason)
|
1112 |
return False |
1113 |
try:
|
1114 |
sectors = int(result.stdout)
|
1115 |
except ValueError: |
1116 |
logger.Error("Invalid output from blockdev: '%s'" % result.stdout)
|
1117 |
return False |
1118 |
bytes = sectors * 512
|
1119 |
if bytes < 128*1024*1024: # less than 128MiB |
1120 |
logger.Error("Meta device too small (%.2fMib)" % (bytes/1024/1024)) |
1121 |
return False |
1122 |
if bytes > (128+32)*1024*1024: # account for an extra (big) PE on LVM |
1123 |
logger.Error("Meta device too big (%.2fMiB)" % (bytes/1024/1024)) |
1124 |
return False |
1125 |
return True |
1126 |
|
1127 |
|
1128 |
@classmethod
|
1129 |
def _AssembleLocal(cls, minor, backend, meta): |
1130 |
"""Configure the local part of a DRBD device.
|
1131 |
|
1132 |
This is the first thing that must be done on an unconfigured DRBD
|
1133 |
device. And it must be done only once.
|
1134 |
|
1135 |
"""
|
1136 |
if not cls._IsValidMeta(meta): |
1137 |
return False |
1138 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disk", |
1139 |
backend, meta, "0", "-e", "detach"]) |
1140 |
if result.failed:
|
1141 |
logger.Error("Can't attach local disk: %s" % result.output)
|
1142 |
return not result.failed |
1143 |
|
1144 |
|
1145 |
@classmethod
|
1146 |
def _ShutdownLocal(cls, minor): |
1147 |
"""Detach from the local device.
|
1148 |
|
1149 |
I/Os will continue to be served from the remote device. If we
|
1150 |
don't have a remote device, this operation will fail.
|
1151 |
|
1152 |
"""
|
1153 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "detach"]) |
1154 |
if result.failed:
|
1155 |
logger.Error("Can't detach local device: %s" % result.output)
|
1156 |
return not result.failed |
1157 |
|
1158 |
|
1159 |
@staticmethod
|
1160 |
def _ShutdownAll(minor): |
1161 |
"""Deactivate the device.
|
1162 |
|
1163 |
This will, of course, fail if the device is in use.
|
1164 |
|
1165 |
"""
|
1166 |
result = utils.RunCmd(["drbdsetup", DRBDev._DevPath(minor), "down"]) |
1167 |
if result.failed:
|
1168 |
logger.Error("Can't shutdown drbd device: %s" % result.output)
|
1169 |
return not result.failed |
1170 |
|
1171 |
|
1172 |
@classmethod
|
1173 |
def _AssembleNet(cls, minor, net_info, protocol): |
1174 |
"""Configure the network part of the device.
|
1175 |
|
1176 |
This operation can be, in theory, done multiple times, but there
|
1177 |
have been cases (in lab testing) in which the network part of the
|
1178 |
device had become stuck and couldn't be shut down because activity
|
1179 |
from the new peer (also stuck) triggered a timer re-init and
|
1180 |
needed remote peer interface shutdown in order to clear. So please
|
1181 |
don't change online the net config.
|
1182 |
|
1183 |
"""
|
1184 |
lhost, lport, rhost, rport = net_info |
1185 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "net", |
1186 |
"%s:%s" % (lhost, lport), "%s:%s" % (rhost, rport), |
1187 |
protocol]) |
1188 |
if result.failed:
|
1189 |
logger.Error("Can't setup network for dbrd device: %s" %
|
1190 |
result.fail_reason) |
1191 |
return False |
1192 |
|
1193 |
timeout = time.time() + 10
|
1194 |
ok = False
|
1195 |
while time.time() < timeout:
|
1196 |
info = cls._GetDevInfo(minor) |
1197 |
if not "local_addr" in info or not "remote_addr" in info: |
1198 |
time.sleep(1)
|
1199 |
continue
|
1200 |
if (info["local_addr"] != (lhost, lport) or |
1201 |
info["remote_addr"] != (rhost, rport)):
|
1202 |
time.sleep(1)
|
1203 |
continue
|
1204 |
ok = True
|
1205 |
break
|
1206 |
if not ok: |
1207 |
logger.Error("Timeout while configuring network")
|
1208 |
return False |
1209 |
return True |
1210 |
|
1211 |
|
1212 |
@classmethod
|
1213 |
def _ShutdownNet(cls, minor): |
1214 |
"""Disconnect from the remote peer.
|
1215 |
|
1216 |
This fails if we don't have a local device.
|
1217 |
|
1218 |
"""
|
1219 |
result = utils.RunCmd(["drbdsetup", cls._DevPath(minor), "disconnect"]) |
1220 |
logger.Error("Can't shutdown network: %s" % result.output)
|
1221 |
return not result.failed |
1222 |
|
1223 |
|
1224 |
def _SetFromMinor(self, minor): |
1225 |
"""Set our parameters based on the given minor.
|
1226 |
|
1227 |
This sets our minor variable and our dev_path.
|
1228 |
|
1229 |
"""
|
1230 |
if minor is None: |
1231 |
self.minor = self.dev_path = None |
1232 |
else:
|
1233 |
self.minor = minor
|
1234 |
self.dev_path = self._DevPath(minor) |
1235 |
|
1236 |
|
1237 |
def Assemble(self): |
1238 |
"""Assemble the drbd.
|
1239 |
|
1240 |
Method:
|
1241 |
- if we have a local backing device, we bind to it by:
|
1242 |
- checking the list of used drbd devices
|
1243 |
- check if the local minor use of any of them is our own device
|
1244 |
- if yes, abort?
|
1245 |
- if not, bind
|
1246 |
- if we have a local/remote net info:
|
1247 |
- redo the local backing device step for the remote device
|
1248 |
- check if any drbd device is using the local port,
|
1249 |
if yes abort
|
1250 |
- check if any remote drbd device is using the remote
|
1251 |
port, if yes abort (for now)
|
1252 |
- bind our net port
|
1253 |
- bind the remote net port
|
1254 |
|
1255 |
"""
|
1256 |
self.Attach()
|
1257 |
if self.minor is not None: |
1258 |
logger.Info("Already assembled")
|
1259 |
return True |
1260 |
|
1261 |
result = super(DRBDev, self).Assemble() |
1262 |
if not result: |
1263 |
return result
|
1264 |
|
1265 |
minor = self._FindUnusedMinor()
|
1266 |
if minor is None: |
1267 |
raise errors.BlockDeviceError("Not enough free minors for DRBD!") |
1268 |
need_localdev_teardown = False
|
1269 |
if self._children[0]: |
1270 |
result = self._AssembleLocal(minor, self._children[0].dev_path, |
1271 |
self._children[1].dev_path) |
1272 |
if not result: |
1273 |
return False |
1274 |
need_localdev_teardown = True
|
1275 |
if self._lhost and self._lport and self._rhost and self._rport: |
1276 |
result = self._AssembleNet(minor,
|
1277 |
(self._lhost, self._lport, |
1278 |
self._rhost, self._rport), |
1279 |
"C")
|
1280 |
if not result: |
1281 |
if need_localdev_teardown:
|
1282 |
# we will ignore failures from this
|
1283 |
logger.Error("net setup failed, tearing down local device")
|
1284 |
self._ShutdownAll(minor)
|
1285 |
return False |
1286 |
self._SetFromMinor(minor)
|
1287 |
return True |
1288 |
|
1289 |
|
1290 |
def Shutdown(self): |
1291 |
"""Shutdown the DRBD device.
|
1292 |
|
1293 |
"""
|
1294 |
if self.minor is None and not self.Attach(): |
1295 |
logger.Info("DRBD device not attached to a device during Shutdown")
|
1296 |
return True |
1297 |
if not self._ShutdownAll(self.minor): |
1298 |
return False |
1299 |
self.minor = None |
1300 |
self.dev_path = None |
1301 |
return True |
1302 |
|
1303 |
|
1304 |
def Attach(self): |
1305 |
"""Find a DRBD device which matches our config and attach to it.
|
1306 |
|
1307 |
In case of partially attached (local device matches but no network
|
1308 |
setup), we perform the network attach. If successful, we re-test
|
1309 |
the attach if can return success.
|
1310 |
|
1311 |
"""
|
1312 |
for minor in self._GetUsedDevs(): |
1313 |
info = self._GetDevInfo(minor)
|
1314 |
match_l = self._MatchesLocal(info)
|
1315 |
match_r = self._MatchesNet(info)
|
1316 |
if match_l and match_r: |
1317 |
break
|
1318 |
if match_l and not match_r and "local_addr" not in info: |
1319 |
res_r = self._AssembleNet(minor,
|
1320 |
(self._lhost, self._lport, |
1321 |
self._rhost, self._rport), |
1322 |
"C")
|
1323 |
if res_r and self._MatchesNet(self._GetDevInfo(minor)): |
1324 |
break
|
1325 |
else:
|
1326 |
minor = None
|
1327 |
|
1328 |
self._SetFromMinor(minor)
|
1329 |
return minor is not None |
1330 |
|
1331 |
|
1332 |
def Open(self, force=False): |
1333 |
"""Make the local state primary.
|
1334 |
|
1335 |
If the 'force' parameter is given, the '--do-what-I-say' parameter
|
1336 |
is given. Since this is a pottentialy dangerous operation, the
|
1337 |
force flag should be only given after creation, when it actually
|
1338 |
has to be given.
|
1339 |
|
1340 |
"""
|
1341 |
if self.minor is None and not self.Attach(): |
1342 |
logger.Error("DRBD cannot attach to a device during open")
|
1343 |
return False |
1344 |
cmd = ["drbdsetup", self.dev_path, "primary"] |
1345 |
if force:
|
1346 |
cmd.append("--do-what-I-say")
|
1347 |
result = utils.RunCmd(cmd) |
1348 |
if result.failed:
|
1349 |
logger.Error("Can't make drbd device primary: %s" % result.output)
|
1350 |
return False |
1351 |
return True |
1352 |
|
1353 |
|
1354 |
def Close(self): |
1355 |
"""Make the local state secondary.
|
1356 |
|
1357 |
This will, of course, fail if the device is in use.
|
1358 |
|
1359 |
"""
|
1360 |
if self.minor is None and not self.Attach(): |
1361 |
logger.Info("Instance not attached to a device")
|
1362 |
raise errors.BlockDeviceError("Can't find device") |
1363 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "secondary"]) |
1364 |
if result.failed:
|
1365 |
logger.Error("Can't switch drbd device to secondary: %s" % result.output)
|
1366 |
raise errors.BlockDeviceError("Can't switch drbd device to secondary") |
1367 |
|
1368 |
|
1369 |
def SetSyncSpeed(self, kbytes): |
1370 |
"""Set the speed of the DRBD syncer.
|
1371 |
|
1372 |
"""
|
1373 |
children_result = super(DRBDev, self).SetSyncSpeed(kbytes) |
1374 |
if self.minor is None: |
1375 |
logger.Info("Instance not attached to a device")
|
1376 |
return False |
1377 |
result = utils.RunCmd(["drbdsetup", self.dev_path, "syncer", "-r", "%d" % |
1378 |
kbytes]) |
1379 |
if result.failed:
|
1380 |
logger.Error("Can't change syncer rate: %s " % result.fail_reason)
|
1381 |
return not result.failed and children_result |
1382 |
|
1383 |
|
1384 |
def GetSyncStatus(self): |
1385 |
"""Returns the sync status of the device.
|
1386 |
|
1387 |
Returns:
|
1388 |
(sync_percent, estimated_time)
|
1389 |
|
1390 |
If sync_percent is None, it means all is ok
|
1391 |
If estimated_time is None, it means we can't esimate
|
1392 |
the time needed, otherwise it's the time left in seconds
|
1393 |
|
1394 |
"""
|
1395 |
if self.minor is None and not self.Attach(): |
1396 |
raise errors.BlockDeviceError("Can't attach to device in GetSyncStatus") |
1397 |
proc_info = self._MassageProcData(self._GetProcData()) |
1398 |
if self.minor not in proc_info: |
1399 |
raise errors.BlockDeviceError("Can't find myself in /proc (minor %d)" % |
1400 |
self.minor)
|
1401 |
line = proc_info[self.minor]
|
1402 |
match = re.match("^.*sync'ed: *([0-9.]+)%.*"
|
1403 |
" finish: ([0-9]+):([0-9]+):([0-9]+) .*$", line)
|
1404 |
if match:
|
1405 |
sync_percent = float(match.group(1)) |
1406 |
hours = int(match.group(2)) |
1407 |
minutes = int(match.group(3)) |
1408 |
seconds = int(match.group(4)) |
1409 |
est_time = hours * 3600 + minutes * 60 + seconds |
1410 |
else:
|
1411 |
sync_percent = None
|
1412 |
est_time = None
|
1413 |
match = re.match("^ *[0-9]+: cs:([^ ]+).*$", line)
|
1414 |
if not match: |
1415 |
raise errors.BlockDeviceError("Can't find my data in /proc (minor %d)" % |
1416 |
self.minor)
|
1417 |
client_state = match.group(1)
|
1418 |
is_degraded = client_state != "Connected"
|
1419 |
return sync_percent, est_time, is_degraded
|
1420 |
|
1421 |
|
1422 |
@staticmethod
|
1423 |
def _MassageProcData(data): |
1424 |
"""Transform the output of _GetProdData into a nicer form.
|
1425 |
|
1426 |
Returns:
|
1427 |
a dictionary of minor: joined lines from /proc/drbd for that minor
|
1428 |
|
1429 |
"""
|
1430 |
lmatch = re.compile("^ *([0-9]+):.*$")
|
1431 |
results = {} |
1432 |
old_minor = old_line = None
|
1433 |
for line in data: |
1434 |
lresult = lmatch.match(line) |
1435 |
if lresult is not None: |
1436 |
if old_minor is not None: |
1437 |
results[old_minor] = old_line |
1438 |
old_minor = int(lresult.group(1)) |
1439 |
old_line = line |
1440 |
else:
|
1441 |
if old_minor is not None: |
1442 |
old_line += " " + line.strip()
|
1443 |
# add last line
|
1444 |
if old_minor is not None: |
1445 |
results[old_minor] = old_line |
1446 |
return results
|
1447 |
|
1448 |
|
1449 |
def GetStatus(self): |
1450 |
"""Compute the status of the DRBD device
|
1451 |
|
1452 |
Note that DRBD devices don't have the STATUS_EXISTING state.
|
1453 |
|
1454 |
"""
|
1455 |
if self.minor is None and not self.Attach(): |
1456 |
return self.STATUS_UNKNOWN |
1457 |
|
1458 |
data = self._GetProcData()
|
1459 |
match = re.compile("^ *%d: cs:[^ ]+ st:(Primary|Secondary)/.*$" %
|
1460 |
self.minor)
|
1461 |
for line in data: |
1462 |
mresult = match.match(line) |
1463 |
if mresult:
|
1464 |
break
|
1465 |
else:
|
1466 |
logger.Error("Can't find myself!")
|
1467 |
return self.STATUS_UNKNOWN |
1468 |
|
1469 |
state = mresult.group(2)
|
1470 |
if state == "Primary": |
1471 |
result = self.STATUS_ONLINE
|
1472 |
else:
|
1473 |
result = self.STATUS_STANDBY
|
1474 |
|
1475 |
return result
|
1476 |
|
1477 |
|
1478 |
@staticmethod
|
1479 |
def _ZeroDevice(device): |
1480 |
"""Zero a device.
|
1481 |
|
1482 |
This writes until we get ENOSPC.
|
1483 |
|
1484 |
"""
|
1485 |
f = open(device, "w") |
1486 |
buf = "\0" * 1048576 |
1487 |
try:
|
1488 |
while True: |
1489 |
f.write(buf) |
1490 |
except IOError, err: |
1491 |
if err.errno != errno.ENOSPC:
|
1492 |
raise
|
1493 |
|
1494 |
|
1495 |
@classmethod
|
1496 |
def Create(cls, unique_id, children, size): |
1497 |
"""Create a new DRBD device.
|
1498 |
|
1499 |
Since DRBD devices are not created per se, just assembled, this
|
1500 |
function just zeroes the meta device.
|
1501 |
|
1502 |
"""
|
1503 |
if len(children) != 2: |
1504 |
raise errors.ProgrammerError("Invalid setup for the drbd device") |
1505 |
meta = children[1]
|
1506 |
meta.Assemble() |
1507 |
if not meta.Attach(): |
1508 |
raise errors.BlockDeviceError("Can't attach to meta device") |
1509 |
if not cls._IsValidMeta(meta.dev_path): |
1510 |
raise errors.BlockDeviceError("Invalid meta device") |
1511 |
logger.Info("Started zeroing device %s" % meta.dev_path)
|
1512 |
cls._ZeroDevice(meta.dev_path) |
1513 |
logger.Info("Done zeroing device %s" % meta.dev_path)
|
1514 |
return cls(unique_id, children)
|
1515 |
|
1516 |
|
1517 |
def Remove(self): |
1518 |
"""Stub remove for DRBD devices.
|
1519 |
|
1520 |
"""
|
1521 |
return self.Shutdown() |
1522 |
|
1523 |
|
1524 |
DEV_MAP = { |
1525 |
"lvm": LogicalVolume,
|
1526 |
"md_raid1": MDRaid1,
|
1527 |
"drbd": DRBDev,
|
1528 |
} |
1529 |
|
1530 |
|
1531 |
def FindDevice(dev_type, unique_id, children): |
1532 |
"""Search for an existing, assembled device.
|
1533 |
|
1534 |
This will succeed only if the device exists and is assembled, but it
|
1535 |
does not do any actions in order to activate the device.
|
1536 |
|
1537 |
"""
|
1538 |
if dev_type not in DEV_MAP: |
1539 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
1540 |
device = DEV_MAP[dev_type](unique_id, children) |
1541 |
if not device.Attach(): |
1542 |
return None |
1543 |
return device
|
1544 |
|
1545 |
|
1546 |
def AttachOrAssemble(dev_type, unique_id, children): |
1547 |
"""Try to attach or assemble an existing device.
|
1548 |
|
1549 |
This will attach to an existing assembled device or will assemble
|
1550 |
the device, as needed, to bring it fully up.
|
1551 |
|
1552 |
"""
|
1553 |
if dev_type not in DEV_MAP: |
1554 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
1555 |
device = DEV_MAP[dev_type](unique_id, children) |
1556 |
if not device.Attach(): |
1557 |
device.Assemble() |
1558 |
if not device.Attach(): |
1559 |
raise errors.BlockDeviceError("Can't find a valid block device for" |
1560 |
" %s/%s/%s" %
|
1561 |
(dev_type, unique_id, children)) |
1562 |
return device
|
1563 |
|
1564 |
|
1565 |
def Create(dev_type, unique_id, children, size): |
1566 |
"""Create a device.
|
1567 |
|
1568 |
"""
|
1569 |
if dev_type not in DEV_MAP: |
1570 |
raise errors.ProgrammerError("Invalid block device type '%s'" % dev_type) |
1571 |
device = DEV_MAP[dev_type].Create(unique_id, children, size) |
1572 |
return device
|