]>
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
= malloc(len
);
110 if (msg
->buf
== NULL
)
112 rv
= recv_buf(fd
, msg
->buf
, len
, tmo
);
119 rv
= recv_buf(fd
, &magic
, 4, tmo
);
120 if (rv
< 0 || magic
!= end_magic
) {
128 int ack(int fd
, int tmo
)
130 struct metadata_update msg
= { .len
= 0 };
132 return send_message(fd
, &msg
, tmo
);
135 int wait_reply(int fd
, int tmo
)
137 struct metadata_update msg
;
138 int err
= receive_message(fd
, &msg
, tmo
);
140 /* mdmon sent extra data, but caller only cares that we got a
143 if (err
== 0 && msg
.len
> 0)
149 int connect_monitor(char *devname
)
154 struct sockaddr_un addr
;
158 pos
= sprintf(path
, "%s/", MDMON_DIR
);
159 if (is_subarray(devname
)) {
161 c
= strchr(devname
, '/');
164 snprintf(&path
[pos
], c
- devname
+ 1, "%s", devname
);
167 pos
+= sprintf(&path
[pos
], "%s", devname
);
168 sprintf(&path
[pos
], ".sock");
170 sfd
= socket(PF_LOCAL
, SOCK_STREAM
, 0);
174 addr
.sun_family
= PF_LOCAL
;
175 strcpy(addr
.sun_path
, path
);
176 if (connect(sfd
, &addr
, sizeof(addr
)) < 0) {
181 fl
= fcntl(sfd
, F_GETFL
, 0);
183 fcntl(sfd
, F_SETFL
, fl
);
188 int fping_monitor(int sfd
)
195 /* try to ping existing socket */
196 if (ack(sfd
, 20) != 0)
199 /* check the reply */
200 if (!err
&& wait_reply(sfd
, 20) != 0)
206 /* give the monitor a chance to update the metadata */
207 int ping_monitor(char *devname
)
209 int sfd
= connect_monitor(devname
);
210 int err
= fping_monitor(sfd
);
216 static char *ping_monitor_version(char *devname
)
218 int sfd
= connect_monitor(devname
);
219 struct metadata_update msg
;
225 if (ack(sfd
, 20) != 0)
228 if (!err
&& receive_message(sfd
, &msg
, 20) != 0)
233 if (err
|| !msg
.len
|| !msg
.buf
)
238 static int unblock_subarray(struct mdinfo
*sra
, const int unfreeze
)
244 sprintf(buf
, "external:%s\n", sra
->text_version
);
250 sysfs_set_str(sra
, NULL
, "metadata_version", buf
) ||
252 sysfs_attribute_available(sra
, NULL
, "sync_action") &&
253 sysfs_set_str(sra
, NULL
, "sync_action", "idle")))
259 * block_monitor - prevent mdmon spare assignment
260 * @container - container to block
261 * @freeze - flag to additionally freeze sync_action
263 * This is used by the reshape code to freeze the container, and the
264 * auto-rebuild implementation to atomically move spares.
265 * In both cases we need to stop mdmon from assigning spares to replace
266 * failed devices as we might have other plans for the spare.
267 * For the reshape case we also need to 'freeze' sync_action so that
268 * no recovery happens until we have fully prepared for the reshape.
270 * We tell mdmon that the array is frozen by marking the 'metadata' name
271 * with a leading '-'. The previously told mdmon "Don't make this array
272 * read/write, leave it readonly". Now it means a more general "Don't
273 * reconfigure this array at all".
274 * As older versions of mdmon (which might run from initrd) don't understand
275 * this, we first check that the running mdmon is new enough.
277 int block_monitor(char *container
, const int freeze
)
279 int devnum
= devname2devnum(container
);
280 struct mdstat_ent
*ent
, *e
, *e2
;
281 struct mdinfo
*sra
= NULL
;
282 char *version
= NULL
;
286 if (!mdmon_running(devnum
)) {
287 /* if mdmon is not active we assume that any instance that is
288 * later started will match the current mdadm version, if this
289 * assumption is violated we may inadvertantly rebuild an array
290 * that was meant for reshape, or start rebuild on a spare that
291 * was to be moved to another container
297 version
= ping_monitor_version(container
);
298 ver
= version
? mdadm_version(version
) : -1;
302 ": mdmon instance for %s cannot be disabled\n",
308 ent
= mdstat_read(0, 0);
311 ": failed to read /proc/mdstat while disabling mdmon\n");
315 /* freeze container contents */
316 for (e
= ent
; e
; e
= e
->next
) {
317 if (!is_container_member(e
, container
))
320 sra
= sysfs_read(-1, e
->devnum
, GET_VERSION
);
323 ": failed to read sysfs for subarray%s\n",
324 to_subarray(e
, container
));
327 /* can't reshape an array that we can't monitor */
328 if (sra
->text_version
[0] == '-')
331 if (freeze
&& sysfs_freeze_array(sra
) < 1)
333 /* flag this array to not be modified by mdmon (close race with
334 * takeover in reshape case and spare reassignment in the
337 sprintf(buf
, "external:%s\n", sra
->text_version
);
339 if (sysfs_set_str(sra
, NULL
, "metadata_version", buf
))
341 ping_monitor(container
);
343 /* check that we did not race with recovery */
345 !sysfs_attribute_available(sra
, NULL
, "sync_action")) ||
347 sysfs_attribute_available(sra
, NULL
, "sync_action") &&
348 sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) > 0 &&
349 strcmp(buf
, "frozen\n") == 0))
352 unblock_subarray(sra
, 0);
355 /* Double check against races - there should be no spares
359 sra
= sysfs_read(-1, e
->devnum
, GET_DEVS
| GET_STATE
);
360 if (sra
&& sra
->array
.spare_disks
> 0) {
361 unblock_subarray(sra
, freeze
);
367 fprintf(stderr
, Name
": failed to freeze subarray%s\n",
368 to_subarray(e
, container
));
370 /* thaw the partially frozen container */
371 for (e2
= ent
; e2
&& e2
!= e
; e2
= e2
->next
) {
372 if (!is_container_member(e2
, container
))
375 sra
= sysfs_read(-1, e2
->devnum
, GET_VERSION
);
376 if (unblock_subarray(sra
, freeze
))
377 fprintf(stderr
, Name
": Failed to unfreeze %s\n", e2
->dev
);
380 ping_monitor(container
); /* cleared frozen */
390 void unblock_monitor(char *container
, const int unfreeze
)
392 struct mdstat_ent
*ent
, *e
;
393 struct mdinfo
*sra
= NULL
;
396 ent
= mdstat_read(0, 0);
399 ": failed to read /proc/mdstat while unblocking container\n");
403 /* unfreeze container contents */
404 for (e
= ent
; e
; e
= e
->next
) {
405 if (!is_container_member(e
, container
))
408 sra
= sysfs_read(-1, e
->devnum
, GET_VERSION
|GET_LEVEL
);
409 if (sra
->array
.level
> 0)
411 if (unblock_subarray(sra
, unfreeze
))
412 fprintf(stderr
, Name
": Failed to unfreeze %s\n", e
->dev
);
415 ping_monitor(container
);
423 /* give the manager a chance to view the updated container state. This
424 * would naturally happen due to the manager noticing a change in
425 * /proc/mdstat; however, pinging encourages this detection to happen
426 * while an exclusive open() on the container is active
428 int ping_manager(char *devname
)
430 int sfd
= connect_monitor(devname
);
431 struct metadata_update msg
= { .len
= -1 };
437 err
= send_message(sfd
, &msg
, 20);
439 /* check the reply */
440 if (!err
&& wait_reply(sfd
, 20) != 0)