]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - msg.c
2 * Copyright (C) 2008 Intel Corporation
4 * mdmon socket / message handling
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
28 #include <sys/types.h>
29 #include <sys/socket.h>
34 static const __u32 start_magic
= 0x5a5aa5a5;
35 static const __u32 end_magic
= 0xa5a55a5a;
37 static int send_buf(int fd
, const void* buf
, int len
, int tmo
)
41 struct timeval timeout
= {tmo
, 0};
42 struct timeval
*ptmo
= tmo
? &timeout
: NULL
;
47 rv
= select(fd
+1, NULL
, &set
, NULL
, ptmo
);
50 rv
= write(fd
, buf
, len
);
59 static int recv_buf(int fd
, void* buf
, int len
, int tmo
)
63 struct timeval timeout
= {tmo
, 0};
64 struct timeval
*ptmo
= tmo
? &timeout
: NULL
;
69 rv
= select(fd
+1, &set
, NULL
, NULL
, ptmo
);
72 rv
= read(fd
, buf
, len
);
82 int send_message(int fd
, struct metadata_update
*msg
, int tmo
)
87 rv
= send_buf(fd
, &start_magic
, 4, tmo
);
88 rv
= rv
?: send_buf(fd
, &len
, 4, tmo
);
90 rv
= rv
?: send_buf(fd
, msg
->buf
, msg
->len
, tmo
);
91 rv
= send_buf(fd
, &end_magic
, 4, tmo
);
96 int receive_message(int fd
, struct metadata_update
*msg
, int tmo
)
102 rv
= recv_buf(fd
, &magic
, 4, tmo
);
103 if (rv
< 0 || magic
!= start_magic
)
105 rv
= recv_buf(fd
, &len
, 4, tmo
);
106 if (rv
< 0 || len
> MSG_MAX_LEN
)
109 msg
->buf
= xmalloc(len
);
110 rv
= recv_buf(fd
, msg
->buf
, len
, tmo
);
117 rv
= recv_buf(fd
, &magic
, 4, tmo
);
118 if (rv
< 0 || magic
!= end_magic
) {
126 int ack(int fd
, int tmo
)
128 struct metadata_update msg
= { .len
= 0 };
130 return send_message(fd
, &msg
, tmo
);
133 int wait_reply(int fd
, int tmo
)
135 struct metadata_update msg
;
136 int err
= receive_message(fd
, &msg
, tmo
);
138 /* mdmon sent extra data, but caller only cares that we got a
141 if (err
== 0 && msg
.len
> 0)
147 int connect_monitor(char *devname
)
152 struct sockaddr_un addr
;
156 pos
= sprintf(path
, "%s/", MDMON_DIR
);
157 if (is_subarray(devname
)) {
159 c
= strchr(devname
, '/');
162 snprintf(&path
[pos
], c
- devname
+ 1, "%s", devname
);
165 pos
+= sprintf(&path
[pos
], "%s", devname
);
166 sprintf(&path
[pos
], ".sock");
168 sfd
= socket(PF_LOCAL
, SOCK_STREAM
, 0);
172 addr
.sun_family
= PF_LOCAL
;
173 strcpy(addr
.sun_path
, path
);
174 if (connect(sfd
, &addr
, sizeof(addr
)) < 0) {
179 fl
= fcntl(sfd
, F_GETFL
, 0);
181 fcntl(sfd
, F_SETFL
, fl
);
186 int fping_monitor(int sfd
)
193 /* try to ping existing socket */
194 if (ack(sfd
, 20) != 0)
197 /* check the reply */
198 if (!err
&& wait_reply(sfd
, 20) != 0)
204 /* give the monitor a chance to update the metadata */
205 int ping_monitor(char *devname
)
207 int sfd
= connect_monitor(devname
);
211 err
= fping_monitor(sfd
);
219 /* ping monitor using device number */
220 int ping_monitor_by_id(int devnum
)
223 char *container
= devnum2devname(devnum
);
226 err
= ping_monitor(container
);
233 static char *ping_monitor_version(char *devname
)
235 int sfd
= connect_monitor(devname
);
236 struct metadata_update msg
;
242 if (ack(sfd
, 20) != 0)
245 if (!err
&& receive_message(sfd
, &msg
, 20) != 0)
250 if (err
|| !msg
.len
|| !msg
.buf
)
255 int unblock_subarray(struct mdinfo
*sra
, const int unfreeze
)
261 sprintf(buf
, "external:%s\n", sra
->text_version
);
267 sysfs_set_str(sra
, NULL
, "metadata_version", buf
) ||
269 sysfs_attribute_available(sra
, NULL
, "sync_action") &&
270 sysfs_set_str(sra
, NULL
, "sync_action", "idle")))
275 int block_subarray(struct mdinfo
*sra
)
280 sprintf(buf
, "external:%s\n", sra
->text_version
);
282 if (sysfs_set_str(sra
, NULL
, "metadata_version", buf
))
288 /* check mdmon version if it supports
289 * array blocking mechanism
291 int check_mdmon_version(char *container
)
293 char *version
= NULL
;
294 int devnum
= devname2devnum(container
);
296 if (!mdmon_running(devnum
)) {
297 /* if mdmon is not active we assume that any instance that is
298 * later started will match the current mdadm version, if this
299 * assumption is violated we may inadvertantly rebuild an array
300 * that was meant for reshape, or start rebuild on a spare that
301 * was to be moved to another container
307 version
= ping_monitor_version(container
);
308 ver
= version
? mdadm_version(version
) : -1;
311 pr_err("mdmon instance for %s cannot be disabled\n",
321 * block_monitor - prevent mdmon spare assignment
322 * @container - container to block
323 * @freeze - flag to additionally freeze sync_action
325 * This is used by the reshape code to freeze the container, and the
326 * auto-rebuild implementation to atomically move spares.
327 * In both cases we need to stop mdmon from assigning spares to replace
328 * failed devices as we might have other plans for the spare.
329 * For the reshape case we also need to 'freeze' sync_action so that
330 * no recovery happens until we have fully prepared for the reshape.
332 * We tell mdmon that the array is frozen by marking the 'metadata' name
333 * with a leading '-'. The previously told mdmon "Don't make this array
334 * read/write, leave it readonly". Now it means a more general "Don't
335 * reconfigure this array at all".
336 * As older versions of mdmon (which might run from initrd) don't understand
337 * this, we first check that the running mdmon is new enough.
339 int block_monitor(char *container
, const int freeze
)
341 struct mdstat_ent
*ent
, *e
, *e2
;
342 struct mdinfo
*sra
= NULL
;
346 if (check_mdmon_version(container
))
349 ent
= mdstat_read(0, 0);
351 pr_err("failed to read /proc/mdstat while disabling mdmon\n");
355 /* freeze container contents */
356 for (e
= ent
; e
; e
= e
->next
) {
357 if (!is_container_member(e
, container
))
360 sra
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
362 pr_err("failed to read sysfs for subarray%s\n",
363 to_subarray(e
, container
));
366 /* can't reshape an array that we can't monitor */
367 if (sra
->text_version
[0] == '-')
370 if (freeze
&& sysfs_freeze_array(sra
) < 1)
372 /* flag this array to not be modified by mdmon (close race with
373 * takeover in reshape case and spare reassignment in the
376 if (block_subarray(sra
))
378 ping_monitor(container
);
380 /* check that we did not race with recovery */
382 !sysfs_attribute_available(sra
, NULL
, "sync_action")) ||
384 sysfs_attribute_available(sra
, NULL
, "sync_action") &&
385 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
386 strcmp(buf
, "frozen\n") == 0))
389 unblock_subarray(sra
, 0);
392 /* Double check against races - there should be no spares
396 sra
= sysfs_read(-1, e
->devnum
, GET_DEVS
| GET_STATE
);
397 if (sra
&& sra
->array
.spare_disks
> 0) {
398 unblock_subarray(sra
, freeze
);
404 pr_err("failed to freeze subarray%s\n",
405 to_subarray(e
, container
));
407 /* thaw the partially frozen container */
408 for (e2
= ent
; e2
&& e2
!= e
; e2
= e2
->next
) {
409 if (!is_container_member(e2
, container
))
412 sra
= sysfs_read(-1, e2
->devnum
, GET_VERSION
);
413 if (unblock_subarray(sra
, freeze
))
414 pr_err("Failed to unfreeze %s\n", e2
->dev
);
417 ping_monitor(container
); /* cleared frozen */
427 void unblock_monitor(char *container
, const int unfreeze
)
429 struct mdstat_ent
*ent
, *e
;
430 struct mdinfo
*sra
= NULL
;
433 ent
= mdstat_read(0, 0);
435 pr_err("failed to read /proc/mdstat while unblocking container\n");
439 /* unfreeze container contents */
440 for (e
= ent
; e
; e
= e
->next
) {
441 if (!is_container_member(e
, container
))
444 sra
= sysfs_read(-1, e
->devnum
, GET_VERSION
|GET_LEVEL
);
447 if (sra
->array
.level
> 0)
449 if (unblock_subarray(sra
, unfreeze
))
450 pr_err("Failed to unfreeze %s\n", e
->dev
);
453 ping_monitor(container
);
461 /* give the manager a chance to view the updated container state. This
462 * would naturally happen due to the manager noticing a change in
463 * /proc/mdstat; however, pinging encourages this detection to happen
464 * while an exclusive open() on the container is active
466 int ping_manager(char *devname
)
468 int sfd
= connect_monitor(devname
);
469 struct metadata_update msg
= { .len
= -1 };
475 err
= send_message(sfd
, &msg
, 20);
477 /* check the reply */
478 if (!err
&& wait_reply(sfd
, 20) != 0)
485 /* using takeover operation for grow purposes, mdadm has to be sure
486 * that mdmon processes all updates, and if necessary it will be closed
487 * at takeover to raid0 operation
489 void flush_mdmon(char *container
)
491 ping_manager(container
);
492 ping_monitor(container
);