]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - managemon.c
2 * mdmon - monitor external metadata arrays
4 * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2009 Intel Corporation
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms and conditions of the GNU General Public License,
9 * version 2, as published by the Free Software Foundation.
11 * This program is distributed in the hope it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
22 * The management thread for monitoring active md arrays.
23 * This thread does things which might block such as memory
27 * - Find out about new arrays in this container.
28 * Allocate the data structures and open the files.
30 * For this we watch /proc/mdstat and find new arrays with
31 * metadata type that confirms sharing. e.g. "md4"
32 * When we find a new array we slip it into the list of
33 * arrays and signal 'monitor' by writing to a pipe.
35 * - Respond to reshape requests by allocating new data structures
36 * and opening new files.
38 * These come as a change to raid_disks. We allocate a new
39 * version of the data structures and slip it into the list.
40 * 'monitor' will notice and release the old version.
41 * Changes to level, chunksize, layout.. do not need re-allocation.
42 * Reductions in raid_disks don't really either, but we handle
43 * them the same way for consistency.
45 * - When a device is added to the container, we add it to the metadata
48 * - Deal with degraded array
49 * We only do this when first noticing the array is degraded.
50 * This can be when we first see the array, when sync completes or
51 * when recovery completes.
53 * Check if number of failed devices suggests recovery is needed, and
55 * Ask metadata to allocate a spare device
56 * Add device as not in_sync and give a role
58 * Open sysfs files and pass to monitor.
59 * Make sure that monitor Starts recovery....
61 * - Pass on metadata updates from external programs such as
62 * mdadm creating a new array.
65 * It might involve adding a new array or changing the status of
66 * a spare, or any reconfig that the kernel doesn't get involved in.
68 * The required updates are received via a named pipe. There will
69 * be one named pipe for each container. Each message contains a
70 * sync marker: 0x5a5aa5a5, A byte count, and the message. This is
71 * passed to the metadata handler which will interpret and process it.
72 * For 'DDF' messages are internal data blocks with the leading
73 * 'magic number' signifying what sort of data it is.
78 * We select on /proc/mdstat and the named pipe.
79 * We create new arrays or updated version of arrays and slip
80 * them into the head of the list, then signal 'monitor' via a pipe write.
81 * 'monitor' will notice and place the old array on a return list.
82 * Metadata updates are placed on a queue just like they arrive
83 * from the named pipe.
85 * When new arrays are found based on correct metadata string, we
86 * need to identify them with an entry in the metadata. Maybe we require
87 * the metadata to be mdX/NN when NN is the index into an appropriate table.
93 * - Watch for spares to be added to the container, and write updated
95 * - Watch for new arrays using this container, confirm they match metadata
96 * and if so, start monitoring them
97 * - Watch for spares being added to monitored arrays. This shouldn't
98 * happen, as we should do all the adding. Just remove them.
99 * - Watch for change in raid-disks, chunk-size, etc. Update metadata and
107 #include <sys/syscall.h>
108 #include <sys/socket.h>
111 static void close_aa(struct active_array
*aa
)
115 for (d
= aa
->info
.devs
; d
; d
= d
->next
) {
116 close(d
->recovery_fd
);
120 close(aa
->action_fd
);
121 close(aa
->info
.state_fd
);
122 close(aa
->resync_start_fd
);
125 static void free_aa(struct active_array
*aa
)
127 /* Note that this doesn't close fds if they are being used
128 * by a clone. ->container will be set for a clone
130 dprintf("%s: devnum: %d\n", __func__
, aa
->devnum
);
133 while (aa
->info
.devs
) {
134 struct mdinfo
*d
= aa
->info
.devs
;
135 aa
->info
.devs
= d
->next
;
141 static struct active_array
*duplicate_aa(struct active_array
*aa
)
143 struct active_array
*newa
= malloc(sizeof(*newa
));
144 struct mdinfo
**dp1
, **dp2
;
148 newa
->replaces
= NULL
;
149 newa
->info
.next
= NULL
;
151 dp2
= &newa
->info
.devs
;
153 for (dp1
= &aa
->info
.devs
; *dp1
; dp1
= &(*dp1
)->next
) {
155 if ((*dp1
)->state_fd
< 0)
158 d
= malloc(sizeof(*d
));
168 static void wakeup_monitor(void)
170 /* tgkill(getpid(), mon_tid, SIGUSR1); */
172 syscall(SYS_tgkill
, pid
, mon_tid
, SIGUSR1
);
175 static void remove_old(void)
178 discard_this
->next
= NULL
;
179 free_aa(discard_this
);
180 if (pending_discard
== discard_this
)
181 pending_discard
= NULL
;
187 static void replace_array(struct supertype
*container
,
188 struct active_array
*old
,
189 struct active_array
*new)
191 /* To replace an array, we add it to the top of the list
192 * marked with ->replaces to point to the original.
193 * 'monitor' will take the original out of the list
194 * and put it on 'discard_this'. We take it from there
198 while (pending_discard
) {
199 while (discard_this
== NULL
)
203 pending_discard
= old
;
205 new->next
= container
->arrays
;
206 container
->arrays
= new;
210 struct metadata_update
*update_queue
= NULL
;
211 struct metadata_update
*update_queue_handled
= NULL
;
212 struct metadata_update
*update_queue_pending
= NULL
;
214 static void free_updates(struct metadata_update
**update
)
217 struct metadata_update
*this = *update
;
219 *update
= this->next
;
226 void check_update_queue(struct supertype
*container
)
228 free_updates(&update_queue_handled
);
230 if (update_queue
== NULL
&&
231 update_queue_pending
) {
232 update_queue
= update_queue_pending
;
233 update_queue_pending
= NULL
;
238 static void queue_metadata_update(struct metadata_update
*mu
)
240 struct metadata_update
**qp
;
242 qp
= &update_queue_pending
;
244 qp
= & ((*qp
)->next
);
248 static void add_disk_to_container(struct supertype
*st
, struct mdinfo
*sd
)
252 struct supertype
*st2
;
253 struct metadata_update
*update
= NULL
;
255 mdu_disk_info_t dk
= {
257 .major
= sd
->disk
.major
,
258 .minor
= sd
->disk
.minor
,
263 dprintf("%s: add %d:%d to container\n",
264 __func__
, sd
->disk
.major
, sd
->disk
.minor
);
269 sprintf(nm
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
270 dfd
= dev_open(nm
, O_RDWR
);
274 /* Check the metadata and see if it is already part of this
278 if (st2
->ss
->load_super(st2
, dfd
, NULL
) == 0) {
279 st2
->ss
->getinfo_super(st
, &info
, NULL
);
280 if (st
->ss
->compare_super(st
, st2
) == 0 &&
281 info
.disk
.raid_disk
>= 0) {
282 /* Looks like a good member of array.
284 * mdadm will incorporate any parts into
287 st2
->ss
->free_super(st2
);
291 st2
->ss
->free_super(st2
);
293 st
->update_tail
= &update
;
294 st
->ss
->add_to_super(st
, &dk
, dfd
, NULL
);
295 st
->ss
->write_init_super(st
);
296 queue_metadata_update(update
);
297 st
->update_tail
= NULL
;
300 static void manage_container(struct mdstat_ent
*mdstat
,
301 struct supertype
*container
)
303 /* The only thing of interest here is if a new device
304 * has been added to the container. We add it to the
305 * array ignoring any metadata on it.
306 * FIXME should we look for compatible metadata and take hints
307 * about spare assignment.... probably not.
309 if (mdstat
->devcnt
!= container
->devcnt
) {
310 struct mdinfo
**cdp
, *cd
, *di
, *mdi
;
313 /* read /sys/block/NAME/md/dev-??/block/dev to find out
314 * what is there, and compare with container->info.devs
315 * To see what is removed and what is added.
316 * These need to be remove from, or added to, the array
318 mdi
= sysfs_read(-1, mdstat
->devnum
, GET_DEVS
);
320 /* invalidate the current count so we can try again */
321 container
->devcnt
= -1;
325 /* check for removals */
326 for (cdp
= &container
->devs
; *cdp
; ) {
328 for (di
= mdi
->devs
; di
; di
= di
->next
)
329 if (di
->disk
.major
== (*cdp
)->disk
.major
&&
330 di
->disk
.minor
== (*cdp
)->disk
.minor
) {
342 /* check for additions */
343 for (di
= mdi
->devs
; di
; di
= di
->next
) {
344 for (cd
= container
->devs
; cd
; cd
= cd
->next
)
345 if (di
->disk
.major
== cd
->disk
.major
&&
346 di
->disk
.minor
== cd
->disk
.minor
)
349 struct mdinfo
*newd
= malloc(sizeof(*newd
));
352 container
->devcnt
= -1;
356 add_disk_to_container(container
, newd
);
360 container
->devcnt
= mdstat
->devcnt
;
364 static int disk_init_and_add(struct mdinfo
*disk
, struct mdinfo
*clone
,
365 struct active_array
*aa
)
371 disk
->recovery_fd
= sysfs_open(aa
->devnum
, disk
->sys_name
, "recovery_start");
372 disk
->state_fd
= sysfs_open(aa
->devnum
, disk
->sys_name
, "state");
373 disk
->prev_state
= read_dev_state(disk
->state_fd
);
374 disk
->curr_state
= disk
->prev_state
;
375 disk
->next
= aa
->info
.devs
;
376 aa
->info
.devs
= disk
;
381 static void manage_member(struct mdstat_ent
*mdstat
,
382 struct active_array
*a
)
384 /* Compare mdstat info with known state of member array.
385 * We do not need to look for device state changes here, that
386 * is dealt with by the monitor.
388 * We just look for changes which suggest that a reshape is
390 * Unfortunately decreases in raid_disks don't show up in
391 * mdstat until the reshape completes FIXME.
393 * Actually, we also want to handle degraded arrays here by
394 * trying to find and assign a spare.
395 * We do that whenever the monitor tells us too.
401 a
->info
.array
.raid_disks
= mdstat
->raid_disks
;
402 a
->info
.array
.chunk_size
= mdstat
->chunk_size
;
406 if (sysfs_get_str(&a
->info
, NULL
, "metadata_version", buf
, sizeof(buf
)) > 0)
407 frozen
= buf
[9] == '-';
409 frozen
= 1; /* can't read metadata_version assume the worst */
411 if (a
->check_degraded
&& !frozen
) {
412 struct metadata_update
*updates
= NULL
;
413 struct mdinfo
*newdev
= NULL
;
414 struct active_array
*newa
;
417 a
->check_degraded
= 0;
419 /* The array may not be degraded, this is just a good time
422 newdev
= a
->container
->ss
->activate_spare(a
, &updates
);
426 newa
= duplicate_aa(a
);
429 /* Cool, we can add a device or several. */
431 /* Add device to array and set offset/size/slot.
432 * and open files for each newdev */
433 for (d
= newdev
; d
; d
= d
->next
) {
436 newd
= malloc(sizeof(*newd
));
439 if (sysfs_add_disk(&newa
->info
, d
, 0) < 0) {
443 disk_init_and_add(newd
, d
, newa
);
445 queue_metadata_update(updates
);
447 replace_array(a
->container
, a
, newa
);
448 sysfs_set_str(&a
->info
, NULL
, "sync_action", "recover");
455 free_updates(&updates
);
459 static int aa_ready(struct active_array
*aa
)
462 int level
= aa
->info
.array
.level
;
464 for (d
= aa
->info
.devs
; d
; d
= d
->next
)
468 if (aa
->info
.state_fd
< 0)
471 if (level
> 0 && (aa
->action_fd
< 0 || aa
->resync_start_fd
< 0))
480 static void manage_new(struct mdstat_ent
*mdstat
,
481 struct supertype
*container
,
482 struct active_array
*victim
)
484 /* A new array has appeared in this container.
485 * Hopefully it is already recorded in the metadata.
486 * Check, then create the new array to report it to
490 struct active_array
*new;
491 struct mdinfo
*mdi
, *di
;
496 /* check if array is ready to be monitored */
500 mdi
= sysfs_read(-1, mdstat
->devnum
,
501 GET_LEVEL
|GET_CHUNK
|GET_DISKS
|GET_COMPONENT
|
502 GET_DEGRADED
|GET_DEVS
|GET_OFFSET
|GET_SIZE
|GET_STATE
);
504 new = malloc(sizeof(*new));
513 memset(new, 0, sizeof(*new));
515 new->devnum
= mdstat
->devnum
;
516 strcpy(new->info
.sys_name
, devnum2devname(new->devnum
));
518 new->prev_state
= new->curr_state
= new->next_state
= inactive
;
519 new->prev_action
= new->curr_action
= new->next_action
= idle
;
521 new->container
= container
;
523 inst
= to_subarray(mdstat
, container
->devname
);
525 new->info
.array
= mdi
->array
;
526 new->info
.component_size
= mdi
->component_size
;
528 for (i
= 0; i
< new->info
.array
.raid_disks
; i
++) {
529 struct mdinfo
*newd
= malloc(sizeof(*newd
));
531 for (di
= mdi
->devs
; di
; di
= di
->next
)
532 if (i
== di
->disk
.raid_disk
)
535 if (disk_init_and_add(newd
, di
, new) != 0) {
540 if (failed
> new->info
.array
.failed_disks
) {
541 /* we cannot properly monitor without all working disks */
542 new->container
= NULL
;
548 new->action_fd
= sysfs_open(new->devnum
, NULL
, "sync_action");
549 new->info
.state_fd
= sysfs_open(new->devnum
, NULL
, "array_state");
550 new->resync_start_fd
= sysfs_open(new->devnum
, NULL
, "resync_start");
551 new->metadata_fd
= sysfs_open(new->devnum
, NULL
, "metadata_version");
552 new->sync_completed_fd
= sysfs_open(new->devnum
, NULL
, "sync_completed");
553 dprintf("%s: inst: %d action: %d state: %d\n", __func__
, atoi(inst
),
554 new->action_fd
, new->info
.state_fd
);
558 /* if everything checks out tell the metadata handler we want to
559 * manage this instance
561 if (!aa_ready(new) || container
->ss
->open_new(container
, new, inst
) < 0) {
562 fprintf(stderr
, "mdmon: failed to monitor %s\n",
563 mdstat
->metadata_version
);
564 new->container
= NULL
;
567 replace_array(container
, victim
, new);
569 new->check_degraded
= 1;
570 manage_member(mdstat
, new);
575 void manage(struct mdstat_ent
*mdstat
, struct supertype
*container
)
577 /* We have just read mdstat and need to compare it with
578 * the known active arrays.
579 * Arrays with the wrong metadata are ignored.
582 for ( ; mdstat
; mdstat
= mdstat
->next
) {
583 struct active_array
*a
;
584 if (mdstat
->devnum
== container
->devnum
) {
585 manage_container(mdstat
, container
);
588 if (!is_container_member(mdstat
, container
->devname
))
589 /* Not for this array */
591 /* Looks like a member of this container */
592 for (a
= container
->arrays
; a
; a
= a
->next
) {
593 if (mdstat
->devnum
== a
->devnum
) {
595 manage_member(mdstat
, a
);
599 if (a
== NULL
|| !a
->container
)
600 manage_new(mdstat
, container
, a
);
604 static void handle_message(struct supertype
*container
, struct metadata_update
*msg
)
606 /* queue this metadata update through to the monitor */
608 struct metadata_update
*mu
;
611 while (update_queue_pending
|| update_queue
) {
612 check_update_queue(container
);
616 if (msg
->len
== 0) { /* ping_monitor */
619 cnt
= monitor_loop_cnt
;
621 cnt
+= 2; /* wait until next pselect */
623 cnt
+= 3; /* wait for 2 pselects */
626 while (monitor_loop_cnt
- cnt
< 0)
628 } else if (msg
->len
== -1) { /* ping_manager */
629 struct mdstat_ent
*mdstat
= mdstat_read(1, 0);
631 manage(mdstat
, container
);
633 } else if (!sigterm
) {
634 mu
= malloc(sizeof(*mu
));
640 if (container
->ss
->prepare_update
)
641 container
->ss
->prepare_update(container
, mu
);
642 queue_metadata_update(mu
);
646 void read_sock(struct supertype
*container
)
649 struct metadata_update msg
;
652 int tmo
= 3; /* 3 second timeout before hanging up the socket */
654 fd
= accept(container
->sock
, NULL
, NULL
);
658 fl
= fcntl(fd
, F_GETFL
, 0);
660 fcntl(fd
, F_SETFL
, fl
);
665 /* read and validate the message */
666 if (receive_message(fd
, &msg
, tmo
) == 0) {
667 handle_message(container
, &msg
);
669 /* ping reply with version */
671 msg
.len
= strlen(Version
) + 1;
672 if (send_message(fd
, &msg
, tmo
) < 0)
674 } else if (ack(fd
, tmo
) < 0)
679 } while (!terminate
);
685 int manager_ready
= 0;
686 void do_manager(struct supertype
*container
)
688 struct mdstat_ent
*mdstat
;
691 sigprocmask(SIG_UNBLOCK
, NULL
, &set
);
692 sigdelset(&set
, SIGUSR1
);
693 sigdelset(&set
, SIGTERM
);
700 /* Can only 'manage' things if 'monitor' is not making
701 * structural changes to metadata, so need to check
704 if (update_queue
== NULL
) {
705 mdstat
= mdstat_read(1, 0);
707 manage(mdstat
, container
);
709 read_sock(container
);
715 check_update_queue(container
);
722 if (update_queue
== NULL
)
723 mdstat_wait_fd(container
->sock
, &set
);
725 /* If an update is happening, just wait for signal */
726 pselect(0, NULL
, NULL
, NULL
, NULL
, &set
);