]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - monitor.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.
23 #include <sys/syscall.h>
24 #include <sys/select.h>
27 static char *array_states
[] = {
28 "clear", "inactive", "suspended", "readonly", "read-auto",
29 "clean", "active", "write-pending", "active-idle", NULL
};
30 static char *sync_actions
[] = {
31 "idle", "reshape", "resync", "recover", "check", "repair", NULL
34 static int write_attr(char *attr
, int fd
)
36 return write(fd
, attr
, strlen(attr
));
39 static void add_fd(fd_set
*fds
, int *maxfd
, int fd
)
48 static int read_attr(char *buf
, int len
, int fd
)
57 n
= read(fd
, buf
, len
- 1);
69 static unsigned long long read_resync_start(int fd
)
74 n
= read_attr(buf
, 30, fd
);
77 if (strncmp(buf
, "none", 4) == 0)
80 return strtoull(buf
, NULL
, 10);
83 static unsigned long long read_sync_completed(int fd
)
85 unsigned long long val
;
90 n
= read_attr(buf
, 50, fd
);
95 val
= strtoull(buf
, &ep
, 0);
96 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
101 static enum array_state
read_state(int fd
)
104 int n
= read_attr(buf
, 20, fd
);
108 return (enum array_state
) sysfs_match_word(buf
, array_states
);
111 static enum sync_action
read_action( int fd
)
114 int n
= read_attr(buf
, 20, fd
);
118 return (enum sync_action
) sysfs_match_word(buf
, sync_actions
);
121 int read_dev_state(int fd
)
124 int n
= read_attr(buf
, 60, fd
);
133 if (sysfs_attr_match(cp
, "faulty"))
135 if (sysfs_attr_match(cp
, "in_sync"))
137 if (sysfs_attr_match(cp
, "write_mostly"))
138 rv
|= DS_WRITE_MOSTLY
;
139 if (sysfs_attr_match(cp
, "spare"))
141 if (sysfs_attr_match(cp
, "blocked"))
143 cp
= strchr(cp
, ',');
150 static void signal_manager(void)
152 /* tgkill(getpid(), mon_tid, SIGUSR1); */
154 syscall(SYS_tgkill
, pid
, mgr_tid
, SIGUSR1
);
157 /* Monitor a set of active md arrays - all of which share the
158 * same metadata - and respond to events that require
161 * New arrays are detected by another thread which allocates
162 * required memory and attaches the data structure to our list.
166 * This is detected by array_state going to 'clear' or 'inactive'.
167 * while we thought it was active.
168 * Response is to mark metadata as clean and 'clear' the array(??)
170 * array_state if 'write-pending'
171 * We mark metadata as 'dirty' then set array to 'active'.
173 * Either ignore, or mark clean, then mark metadata as clean.
176 * detected by rd-N/state reporting "faulty"
177 * mark device as 'failed' in metadata, let the kernel release the
178 * device by writing '-blocked' to rd/state, and finally write 'remove' to
179 * rd/state. Before a disk can be replaced it must be failed and removed
180 * from all container members, this will be preemptive for the other
184 * sync_action was 'resync' and becomes 'idle' and resync_start becomes
186 * Notify metadata that sync is complete.
189 * sync_action changes from 'recover' to 'idle'
190 * Check each device state and mark metadata if 'faulty' or 'in_sync'.
193 * This only happens on finding a new array... mdadm will have set
194 * 'resync_start' to the correct value. If 'resync_start' indicates that an
195 * resync needs to occur set the array to the 'active' state rather than the
196 * initial read-auto state.
200 * We wait for a change (poll/select) on array_state, sync_action, and
201 * each rd-X/state file.
202 * When we get any change, we check everything. So read each state file,
203 * then decide what to do.
205 * The core action is to write new metadata to all devices in the array.
206 * This is done at most once on any wakeup.
207 * After that we might:
208 * - update the array_state
209 * - set the role of some devices.
210 * - request a sync_action
214 static int read_and_act(struct active_array
*a
)
216 unsigned long long sync_completed
;
217 int check_degraded
= 0;
218 int check_reshape
= 0;
223 a
->next_state
= bad_word
;
224 a
->next_action
= bad_action
;
226 a
->curr_state
= read_state(a
->info
.state_fd
);
227 a
->curr_action
= read_action(a
->action_fd
);
228 a
->info
.resync_start
= read_resync_start(a
->resync_start_fd
);
229 sync_completed
= read_sync_completed(a
->sync_completed_fd
);
230 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
232 if (mdi
->state_fd
>= 0) {
233 mdi
->recovery_start
= read_resync_start(mdi
->recovery_fd
);
234 mdi
->curr_state
= read_dev_state(mdi
->state_fd
);
238 if (a
->curr_state
> inactive
&&
239 a
->prev_state
== inactive
) {
240 /* array has been started
241 * possible that container operation has to be completed
243 a
->container
->ss
->set_array_state(a
, 0);
245 if (a
->curr_state
<= inactive
&&
246 a
->prev_state
> inactive
) {
247 /* array has been stopped */
248 a
->container
->ss
->set_array_state(a
, 1);
249 a
->next_state
= clear
;
252 if (a
->curr_state
== write_pending
) {
253 a
->container
->ss
->set_array_state(a
, 0);
254 a
->next_state
= active
;
257 if (a
->curr_state
== active_idle
) {
258 /* Set array to 'clean' FIRST, then mark clean
261 a
->next_state
= clean
;
264 if (a
->curr_state
== clean
) {
265 a
->container
->ss
->set_array_state(a
, 1);
267 if (a
->curr_state
== active
||
268 a
->curr_state
== suspended
||
269 a
->curr_state
== bad_word
)
271 if (a
->curr_state
== readonly
) {
272 /* Well, I'm ready to handle things. If readonly
273 * wasn't requested, transition to read-auto.
276 read_attr(buf
, sizeof(buf
), a
->metadata_fd
);
277 if (strncmp(buf
, "external:-", 10) == 0) {
278 /* explicit request for readonly array. Leave it alone */
281 if (a
->container
->ss
->set_array_state(a
, 2))
282 a
->next_state
= read_auto
; /* array is clean */
284 a
->next_state
= active
; /* Now active for recovery etc */
291 a
->curr_action
== idle
&&
292 a
->prev_action
== resync
) {
293 /* A resync has finished. The endpoint is recorded in
294 * 'sync_start'. We don't update the metadata
295 * until the array goes inactive or readonly though.
296 * Just check if we need to fiddle spares.
298 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
303 a
->curr_action
== idle
&&
304 a
->prev_action
== recover
) {
305 /* A recovery has finished. Some disks may be in sync now,
306 * and the array may no longer be degraded
308 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
309 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
311 if (! (mdi
->curr_state
& DS_INSYNC
))
317 a
->curr_action
== reshape
&&
318 a
->prev_action
!= reshape
)
319 /* reshape was requested by mdadm. Need to see if
320 * new devices have been added. Manager does that
321 * when it sees check_reshape
325 /* Check for failures and if found:
326 * 1/ Record the failure in the metadata and unblock the device.
327 * FIXME update the kernel to stop notifying on failed drives when
328 * the array is readonly and we have cleared 'blocked'
329 * 2/ Try to remove the device if the array is writable, or can be
332 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
333 if (mdi
->curr_state
& DS_FAULTY
) {
334 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
337 mdi
->next_state
|= DS_UNBLOCK
;
338 if (a
->curr_state
== read_auto
) {
339 a
->container
->ss
->set_array_state(a
, 0);
340 a
->next_state
= active
;
342 if (a
->curr_state
> readonly
)
343 mdi
->next_state
|= DS_REMOVE
;
347 /* Check for recovery checkpoint notifications. We need to be a
348 * minimum distance away from the last checkpoint to prevent
349 * over checkpointing. Note reshape checkpointing is handled
350 * in the second branch.
352 if (sync_completed
> a
->last_checkpoint
&&
353 sync_completed
- a
->last_checkpoint
> a
->info
.component_size
>> 4 &&
354 a
->curr_action
> reshape
) {
355 /* A (non-reshape) sync_action has reached a checkpoint.
356 * Record the updated position in the metadata
358 a
->last_checkpoint
= sync_completed
;
359 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
360 } else if ((a
->curr_action
== idle
&& a
->prev_action
== reshape
) ||
361 (a
->curr_action
== reshape
362 && sync_completed
> a
->last_checkpoint
) ) {
363 /* Reshape has progressed or completed so we need to
364 * update the array state - and possibly the array size
366 if (sync_completed
!= 0)
367 a
->last_checkpoint
= sync_completed
;
368 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
369 a
->last_checkpoint
= sync_completed
;
372 if (sync_completed
> a
->last_checkpoint
)
373 a
->last_checkpoint
= sync_completed
;
375 a
->container
->ss
->sync_metadata(a
->container
);
376 dprintf("%s(%d): state:%s action:%s next(", __func__
, a
->info
.container_member
,
377 array_states
[a
->curr_state
], sync_actions
[a
->curr_action
]);
379 /* Effect state changes in the array */
380 if (a
->next_state
!= bad_word
) {
381 dprintf(" state:%s", array_states
[a
->next_state
]);
382 write_attr(array_states
[a
->next_state
], a
->info
.state_fd
);
384 if (a
->next_action
!= bad_action
) {
385 write_attr(sync_actions
[a
->next_action
], a
->action_fd
);
386 dprintf(" action:%s", sync_actions
[a
->next_action
]);
388 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
389 if (mdi
->next_state
& DS_UNBLOCK
) {
390 dprintf(" %d:-blocked", mdi
->disk
.raid_disk
);
391 write_attr("-blocked", mdi
->state_fd
);
394 if ((mdi
->next_state
& DS_REMOVE
) && mdi
->state_fd
>= 0) {
397 /* the kernel may not be able to immediately remove the
398 * disk, we can simply wait until the next event to try
401 remove_result
= write_attr("remove", mdi
->state_fd
);
402 if (remove_result
> 0) {
403 dprintf(" %d:removed", mdi
->disk
.raid_disk
);
404 close(mdi
->state_fd
);
408 if (mdi
->next_state
& DS_INSYNC
) {
409 write_attr("+in_sync", mdi
->state_fd
);
410 dprintf(" %d:+in_sync", mdi
->disk
.raid_disk
);
415 /* move curr_ to prev_ */
416 a
->prev_state
= a
->curr_state
;
418 a
->prev_action
= a
->curr_action
;
420 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
421 mdi
->prev_state
= mdi
->curr_state
;
425 if (check_degraded
|| check_reshape
) {
426 /* manager will do the actual check */
428 a
->check_degraded
= 1;
430 a
->check_reshape
= 1;
440 static struct mdinfo
*
441 find_device(struct active_array
*a
, int major
, int minor
)
445 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
446 if (mdi
->disk
.major
== major
&& mdi
->disk
.minor
== minor
)
452 static void reconcile_failed(struct active_array
*aa
, struct mdinfo
*failed
)
454 struct active_array
*a
;
455 struct mdinfo
*victim
;
457 for (a
= aa
; a
; a
= a
->next
) {
460 victim
= find_device(a
, failed
->disk
.major
, failed
->disk
.minor
);
464 if (!(victim
->curr_state
& DS_FAULTY
))
465 write_attr("faulty", victim
->state_fd
);
470 static void dprint_wake_reasons(fd_set
*fds
)
478 fprintf(stderr
, "monitor: wake ( ");
479 for (i
= 0; i
< FD_SETSIZE
; i
++) {
480 if (FD_ISSET(i
, fds
)) {
481 sprintf(proc_path
, "/proc/%d/fd/%d",
484 rv
= readlink(proc_path
, link
, sizeof(link
) - 1);
486 fprintf(stderr
, "%d:unknown ", i
);
490 basename
= strrchr(link
, '/');
491 fprintf(stderr
, "%d:%s ",
492 i
, basename
? ++basename
: link
);
495 fprintf(stderr
, ")\n");
499 int monitor_loop_cnt
;
501 static int wait_and_act(struct supertype
*container
, int nowait
)
505 struct active_array
**aap
= &container
->arrays
;
506 struct active_array
*a
, **ap
;
509 static unsigned int dirty_arrays
= ~0; /* start at some non-zero value */
513 for (ap
= aap
; *ap
;) {
515 /* once an array has been deactivated we want to
516 * ask the manager to discard it.
518 if (!a
->container
|| (a
->info
.array
.level
== 0)) {
530 add_fd(&rfds
, &maxfd
, a
->info
.state_fd
);
531 add_fd(&rfds
, &maxfd
, a
->action_fd
);
532 add_fd(&rfds
, &maxfd
, a
->sync_completed_fd
);
533 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
534 add_fd(&rfds
, &maxfd
, mdi
->state_fd
);
539 if (manager_ready
&& (*aap
== NULL
|| (sigterm
&& !dirty_arrays
))) {
540 /* No interesting arrays, or we have been told to
541 * terminate and everything is clean. Lets see about
542 * exiting. Note that blocking at this point is not a
543 * problem as there are no active arrays, there is
544 * nothing that we need to be ready to do.
546 int fd
= open_dev_excl(container
->devnum
);
547 if (fd
>= 0 || errno
!= EBUSY
) {
548 /* OK, we are safe to leave */
549 if (sigterm
&& !dirty_arrays
)
550 dprintf("caught sigterm, all clean... exiting\n");
552 dprintf("no arrays to monitor... exiting\n");
554 /* On SIGTERM, someone (the take-over mdmon) will
557 remove_pidfile(container
->devname
);
567 sigprocmask(SIG_UNBLOCK
, NULL
, &set
);
568 sigdelset(&set
, SIGUSR1
);
569 monitor_loop_cnt
|= 1;
570 rv
= pselect(maxfd
+1, NULL
, NULL
, &rfds
, NULL
, &set
);
571 monitor_loop_cnt
+= 1;
572 if (rv
== -1 && errno
== EINTR
)
575 dprint_wake_reasons(&rfds
);
581 struct metadata_update
*this;
583 for (this = update_queue
; this ; this = this->next
)
584 container
->ss
->process_update(container
, this);
586 update_queue_handled
= update_queue
;
589 container
->ss
->sync_metadata(container
);
594 for (a
= *aap
; a
; a
= a
->next
) {
597 if (a
->replaces
&& !discard_this
) {
598 struct active_array
**ap
;
599 for (ap
= &a
->next
; *ap
&& *ap
!= a
->replaces
;
604 discard_this
= a
->replaces
;
606 /* FIXME check if device->state_fd need to be cleared?*/
610 is_dirty
= read_and_act(a
);
612 dirty_arrays
+= is_dirty
;
613 /* when terminating stop manipulating the array after it
614 * is clean, but make sure read_and_act() is given a
615 * chance to handle 'active_idle'
617 if (sigterm
&& !is_dirty
)
618 a
->container
= NULL
; /* stop touching this array */
622 /* propagate failures across container members */
623 for (a
= *aap
; a
; a
= a
->next
) {
626 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
627 if (mdi
->curr_state
& DS_FAULTY
)
628 reconcile_failed(*aap
, mdi
);
634 void do_monitor(struct supertype
*container
)
639 rv
= wait_and_act(container
, first
);