]>
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
) {
233 if (mdi
->state_fd
>= 0) {
234 mdi
->recovery_start
= read_resync_start(mdi
->recovery_fd
);
235 mdi
->curr_state
= read_dev_state(mdi
->state_fd
);
239 if (a
->curr_state
> inactive
&&
240 a
->prev_state
== inactive
) {
241 /* array has been started
242 * possible that container operation has to be completed
244 a
->container
->ss
->set_array_state(a
, 0);
246 if (a
->curr_state
<= inactive
&&
247 a
->prev_state
> inactive
) {
248 /* array has been stopped */
249 a
->container
->ss
->set_array_state(a
, 1);
250 a
->next_state
= clear
;
253 if (a
->curr_state
== write_pending
) {
254 a
->container
->ss
->set_array_state(a
, 0);
255 a
->next_state
= active
;
258 if (a
->curr_state
== active_idle
) {
259 /* Set array to 'clean' FIRST, then mark clean
262 a
->next_state
= clean
;
265 if (a
->curr_state
== clean
) {
266 a
->container
->ss
->set_array_state(a
, 1);
268 if (a
->curr_state
== active
||
269 a
->curr_state
== suspended
||
270 a
->curr_state
== bad_word
)
272 if (a
->curr_state
== readonly
) {
273 /* Well, I'm ready to handle things. If readonly
274 * wasn't requested, transition to read-auto.
277 read_attr(buf
, sizeof(buf
), a
->metadata_fd
);
278 if (strncmp(buf
, "external:-", 10) == 0) {
279 /* explicit request for readonly array. Leave it alone */
282 if (a
->container
->ss
->set_array_state(a
, 2))
283 a
->next_state
= read_auto
; /* array is clean */
285 a
->next_state
= active
; /* Now active for recovery etc */
292 a
->curr_action
== idle
&&
293 a
->prev_action
== resync
) {
294 /* A resync has finished. The endpoint is recorded in
295 * 'sync_start'. We don't update the metadata
296 * until the array goes inactive or readonly though.
297 * Just check if we need to fiddle spares.
299 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
304 a
->curr_action
== idle
&&
305 a
->prev_action
== recover
) {
306 /* A recovery has finished. Some disks may be in sync now,
307 * and the array may no longer be degraded
309 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
310 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
312 if (! (mdi
->curr_state
& DS_INSYNC
))
318 a
->curr_action
== reshape
&&
319 a
->prev_action
!= reshape
)
320 /* reshape was requested by mdadm. Need to see if
321 * new devices have been added. Manager does that
322 * when it sees check_reshape
326 /* Check for failures and if found:
327 * 1/ Record the failure in the metadata and unblock the device.
328 * FIXME update the kernel to stop notifying on failed drives when
329 * the array is readonly and we have cleared 'blocked'
330 * 2/ Try to remove the device if the array is writable, or can be
333 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
334 if (mdi
->curr_state
& DS_FAULTY
) {
335 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
338 mdi
->next_state
|= DS_UNBLOCK
;
339 if (a
->curr_state
== read_auto
) {
340 a
->container
->ss
->set_array_state(a
, 0);
341 a
->next_state
= active
;
343 if (a
->curr_state
> readonly
)
344 mdi
->next_state
|= DS_REMOVE
;
348 /* Check for recovery checkpoint notifications. We need to be a
349 * minimum distance away from the last checkpoint to prevent
350 * over checkpointing. Note reshape checkpointing is handled
351 * in the second branch.
353 if (sync_completed
> a
->last_checkpoint
&&
354 sync_completed
- a
->last_checkpoint
> a
->info
.component_size
>> 4 &&
355 a
->curr_action
> reshape
) {
356 /* A (non-reshape) sync_action has reached a checkpoint.
357 * Record the updated position in the metadata
359 a
->last_checkpoint
= sync_completed
;
360 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
361 } else if ((a
->curr_action
== idle
&& a
->prev_action
== reshape
) ||
362 (a
->curr_action
== reshape
363 && sync_completed
> a
->last_checkpoint
) ) {
364 /* Reshape has progressed or completed so we need to
365 * update the array state - and possibly the array size
367 if (sync_completed
!= 0)
368 a
->last_checkpoint
= sync_completed
;
369 /* We might need to update last_checkpoint depending on
370 * the reason that reshape finished.
371 * if array reshape is really finished:
372 * set check point to the end, this allows
373 * set_array_state() to finalize reshape in metadata
374 * if reshape if broken: do not set checkpoint to the end
375 * this allows for reshape restart from checkpoint
377 if ((a
->curr_action
!= reshape
) &&
378 (a
->prev_action
== reshape
)) {
380 if ((sysfs_get_str(&a
->info
, NULL
,
383 sizeof(buf
)) >= 0) &&
384 strncmp(buf
, "none", 4) == 0)
385 a
->last_checkpoint
= a
->info
.component_size
;
387 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
388 a
->last_checkpoint
= sync_completed
;
391 if (sync_completed
> a
->last_checkpoint
)
392 a
->last_checkpoint
= sync_completed
;
394 a
->container
->ss
->sync_metadata(a
->container
);
395 dprintf("%s(%d): state:%s action:%s next(", __func__
, a
->info
.container_member
,
396 array_states
[a
->curr_state
], sync_actions
[a
->curr_action
]);
398 /* Effect state changes in the array */
399 if (a
->next_state
!= bad_word
) {
400 dprintf(" state:%s", array_states
[a
->next_state
]);
401 write_attr(array_states
[a
->next_state
], a
->info
.state_fd
);
403 if (a
->next_action
!= bad_action
) {
404 write_attr(sync_actions
[a
->next_action
], a
->action_fd
);
405 dprintf(" action:%s", sync_actions
[a
->next_action
]);
407 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
408 if (mdi
->next_state
& DS_UNBLOCK
) {
409 dprintf(" %d:-blocked", mdi
->disk
.raid_disk
);
410 write_attr("-blocked", mdi
->state_fd
);
413 if ((mdi
->next_state
& DS_REMOVE
) && mdi
->state_fd
>= 0) {
416 /* the kernel may not be able to immediately remove the
417 * disk, we can simply wait until the next event to try
420 remove_result
= write_attr("remove", mdi
->state_fd
);
421 if (remove_result
> 0) {
422 dprintf(" %d:removed", mdi
->disk
.raid_disk
);
423 close(mdi
->state_fd
);
427 if (mdi
->next_state
& DS_INSYNC
) {
428 write_attr("+in_sync", mdi
->state_fd
);
429 dprintf(" %d:+in_sync", mdi
->disk
.raid_disk
);
434 /* move curr_ to prev_ */
435 a
->prev_state
= a
->curr_state
;
437 a
->prev_action
= a
->curr_action
;
439 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
440 mdi
->prev_state
= mdi
->curr_state
;
444 if (check_degraded
|| check_reshape
) {
445 /* manager will do the actual check */
447 a
->check_degraded
= 1;
449 a
->check_reshape
= 1;
459 static struct mdinfo
*
460 find_device(struct active_array
*a
, int major
, int minor
)
464 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
465 if (mdi
->disk
.major
== major
&& mdi
->disk
.minor
== minor
)
471 static void reconcile_failed(struct active_array
*aa
, struct mdinfo
*failed
)
473 struct active_array
*a
;
474 struct mdinfo
*victim
;
476 for (a
= aa
; a
; a
= a
->next
) {
479 victim
= find_device(a
, failed
->disk
.major
, failed
->disk
.minor
);
483 if (!(victim
->curr_state
& DS_FAULTY
))
484 write_attr("faulty", victim
->state_fd
);
489 static void dprint_wake_reasons(fd_set
*fds
)
497 fprintf(stderr
, "monitor: wake ( ");
498 for (i
= 0; i
< FD_SETSIZE
; i
++) {
499 if (FD_ISSET(i
, fds
)) {
500 sprintf(proc_path
, "/proc/%d/fd/%d",
503 rv
= readlink(proc_path
, link
, sizeof(link
) - 1);
505 fprintf(stderr
, "%d:unknown ", i
);
509 basename
= strrchr(link
, '/');
510 fprintf(stderr
, "%d:%s ",
511 i
, basename
? ++basename
: link
);
514 fprintf(stderr
, ")\n");
518 int monitor_loop_cnt
;
520 static int wait_and_act(struct supertype
*container
, int nowait
)
524 struct active_array
**aap
= &container
->arrays
;
525 struct active_array
*a
, **ap
;
528 static unsigned int dirty_arrays
= ~0; /* start at some non-zero value */
532 for (ap
= aap
; *ap
;) {
534 /* once an array has been deactivated we want to
535 * ask the manager to discard it.
537 if (!a
->container
|| (a
->info
.array
.level
== 0)) {
549 add_fd(&rfds
, &maxfd
, a
->info
.state_fd
);
550 add_fd(&rfds
, &maxfd
, a
->action_fd
);
551 add_fd(&rfds
, &maxfd
, a
->sync_completed_fd
);
552 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
553 add_fd(&rfds
, &maxfd
, mdi
->state_fd
);
558 if (manager_ready
&& (*aap
== NULL
|| (sigterm
&& !dirty_arrays
))) {
559 /* No interesting arrays, or we have been told to
560 * terminate and everything is clean. Lets see about
561 * exiting. Note that blocking at this point is not a
562 * problem as there are no active arrays, there is
563 * nothing that we need to be ready to do.
565 int fd
= open_dev_excl(container
->devnum
);
566 if (fd
>= 0 || errno
!= EBUSY
) {
567 /* OK, we are safe to leave */
568 if (sigterm
&& !dirty_arrays
)
569 dprintf("caught sigterm, all clean... exiting\n");
571 dprintf("no arrays to monitor... exiting\n");
573 /* On SIGTERM, someone (the take-over mdmon) will
576 remove_pidfile(container
->devname
);
586 sigprocmask(SIG_UNBLOCK
, NULL
, &set
);
587 sigdelset(&set
, SIGUSR1
);
588 monitor_loop_cnt
|= 1;
589 rv
= pselect(maxfd
+1, NULL
, NULL
, &rfds
, NULL
, &set
);
590 monitor_loop_cnt
+= 1;
591 if (rv
== -1 && errno
== EINTR
)
594 dprint_wake_reasons(&rfds
);
600 struct metadata_update
*this;
602 for (this = update_queue
; this ; this = this->next
)
603 container
->ss
->process_update(container
, this);
605 update_queue_handled
= update_queue
;
608 container
->ss
->sync_metadata(container
);
613 for (a
= *aap
; a
; a
= a
->next
) {
616 if (a
->replaces
&& !discard_this
) {
617 struct active_array
**ap
;
618 for (ap
= &a
->next
; *ap
&& *ap
!= a
->replaces
;
623 discard_this
= a
->replaces
;
625 /* FIXME check if device->state_fd need to be cleared?*/
629 is_dirty
= read_and_act(a
);
631 dirty_arrays
+= is_dirty
;
632 /* when terminating stop manipulating the array after it
633 * is clean, but make sure read_and_act() is given a
634 * chance to handle 'active_idle'
636 if (sigterm
&& !is_dirty
)
637 a
->container
= NULL
; /* stop touching this array */
641 /* propagate failures across container members */
642 for (a
= *aap
; a
; a
= a
->next
) {
645 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
646 if (mdi
->curr_state
& DS_FAULTY
)
647 reconcile_failed(*aap
, mdi
);
653 void do_monitor(struct supertype
*container
)
658 rv
= wait_and_act(container
, first
);