]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - monitor.c
2 * mdmon - monitor external metadata arrays
4 * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2008 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 int get_resync_start(struct active_array
*a
)
74 n
= read_attr(buf
, 30, a
->resync_start_fd
);
78 a
->resync_start
= strtoull(buf
, NULL
, 10);
84 static enum array_state
read_state(int fd
)
87 int n
= read_attr(buf
, 20, fd
);
91 return (enum array_state
) sysfs_match_word(buf
, array_states
);
94 static enum sync_action
read_action( int fd
)
97 int n
= read_attr(buf
, 20, fd
);
101 return (enum sync_action
) sysfs_match_word(buf
, sync_actions
);
104 int read_dev_state(int fd
)
107 int n
= read_attr(buf
, 60, fd
);
116 if (sysfs_attr_match(cp
, "faulty"))
118 if (sysfs_attr_match(cp
, "in_sync"))
120 if (sysfs_attr_match(cp
, "write_mostly"))
121 rv
|= DS_WRITE_MOSTLY
;
122 if (sysfs_attr_match(cp
, "spare"))
124 if (sysfs_attr_match(cp
, "blocked"))
126 cp
= strchr(cp
, ',');
133 static void signal_manager(void)
135 /* tgkill(getpid(), mon_tid, SIGUSR1); */
137 syscall(SYS_tgkill
, pid
, mgr_tid
, SIGUSR1
);
140 /* Monitor a set of active md arrays - all of which share the
141 * same metadata - and respond to events that require
144 * New arrays are detected by another thread which allocates
145 * required memory and attaches the data structure to our list.
149 * This is detected by array_state going to 'clear' or 'inactive'.
150 * while we thought it was active.
151 * Response is to mark metadata as clean and 'clear' the array(??)
153 * array_state if 'write-pending'
154 * We mark metadata as 'dirty' then set array to 'active'.
156 * Either ignore, or mark clean, then mark metadata as clean.
159 * detected by rd-N/state reporting "faulty"
160 * mark device as 'failed' in metadata, let the kernel release the
161 * device by writing '-blocked' to rd/state, and finally write 'remove' to
162 * rd/state. Before a disk can be replaced it must be failed and removed
163 * from all container members, this will be preemptive for the other
167 * sync_action was 'resync' and becomes 'idle' and resync_start becomes
169 * Notify metadata that sync is complete.
172 * sync_action changes from 'recover' to 'idle'
173 * Check each device state and mark metadata if 'faulty' or 'in_sync'.
176 * This only happens on finding a new array... mdadm will have set
177 * 'resync_start' to the correct value. If 'resync_start' indicates that an
178 * resync needs to occur set the array to the 'active' state rather than the
179 * initial read-auto state.
183 * We wait for a change (poll/select) on array_state, sync_action, and
184 * each rd-X/state file.
185 * When we get any change, we check everything. So read each state file,
186 * then decide what to do.
188 * The core action is to write new metadata to all devices in the array.
189 * This is done at most once on any wakeup.
190 * After that we might:
191 * - update the array_state
192 * - set the role of some devices.
193 * - request a sync_action
197 static int read_and_act(struct active_array
*a
)
199 int check_degraded
= 0;
204 a
->next_state
= bad_word
;
205 a
->next_action
= bad_action
;
207 a
->curr_state
= read_state(a
->info
.state_fd
);
208 a
->curr_action
= read_action(a
->action_fd
);
209 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
211 if (mdi
->state_fd
>= 0)
212 mdi
->curr_state
= read_dev_state(mdi
->state_fd
);
215 if (a
->curr_state
<= inactive
&&
216 a
->prev_state
> inactive
) {
217 /* array has been stopped */
219 a
->container
->ss
->set_array_state(a
, 1);
220 a
->next_state
= clear
;
223 if (a
->curr_state
== write_pending
) {
225 a
->container
->ss
->set_array_state(a
, 0);
226 a
->next_state
= active
;
229 if (a
->curr_state
== active_idle
) {
230 /* Set array to 'clean' FIRST, then mark clean
233 a
->next_state
= clean
;
236 if (a
->curr_state
== clean
) {
238 a
->container
->ss
->set_array_state(a
, 1);
240 if (a
->curr_state
== active
||
241 a
->curr_state
== suspended
||
242 a
->curr_state
== bad_word
)
244 if (a
->curr_state
== readonly
) {
245 /* Well, I'm ready to handle things. If readonly
246 * wasn't requested, transition to read-auto.
249 read_attr(buf
, sizeof(buf
), a
->metadata_fd
);
250 if (strncmp(buf
, "external:-", 10) == 0) {
251 /* explicit request for readonly array. Leave it alone */
255 if (a
->container
->ss
->set_array_state(a
, 2))
256 a
->next_state
= read_auto
; /* array is clean */
258 a
->next_state
= active
; /* Now active for recovery etc */
265 a
->curr_action
== idle
&&
266 a
->prev_action
== resync
) {
267 /* A resync has finished. The endpoint is recorded in
268 * 'sync_start'. We don't update the metadata
269 * until the array goes inactive or readonly though.
270 * Just check if we need to fiddle spares.
273 a
->container
->ss
->set_array_state(a
, a
->curr_state
<= clean
);
278 a
->curr_action
== idle
&&
279 a
->prev_action
== recover
) {
280 /* A recovery has finished. Some disks may be in sync now,
281 * and the array may no longer be degraded
283 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
284 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
286 if (! (mdi
->curr_state
& DS_INSYNC
))
291 /* Check for failures and if found:
292 * 1/ Record the failure in the metadata and unblock the device.
293 * FIXME update the kernel to stop notifying on failed drives when
294 * the array is readonly and we have cleared 'blocked'
295 * 2/ Try to remove the device if the array is writable, or can be
298 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
299 if (mdi
->curr_state
& DS_FAULTY
) {
300 a
->container
->ss
->set_disk(a
, mdi
->disk
.raid_disk
,
303 mdi
->next_state
|= DS_UNBLOCK
;
304 if (a
->curr_state
== read_auto
) {
305 a
->container
->ss
->set_array_state(a
, 0);
306 a
->next_state
= active
;
308 if (a
->curr_state
> readonly
)
309 mdi
->next_state
|= DS_REMOVE
;
313 a
->container
->ss
->sync_metadata(a
->container
);
314 dprintf("%s(%d): state:%s action:%s next(", __func__
, a
->info
.container_member
,
315 array_states
[a
->curr_state
], sync_actions
[a
->curr_action
]);
317 /* Effect state changes in the array */
318 if (a
->next_state
!= bad_word
) {
319 dprintf(" state:%s", array_states
[a
->next_state
]);
320 write_attr(array_states
[a
->next_state
], a
->info
.state_fd
);
322 if (a
->next_action
!= bad_action
) {
323 write_attr(sync_actions
[a
->next_action
], a
->action_fd
);
324 dprintf(" action:%s", sync_actions
[a
->next_action
]);
326 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
327 if (mdi
->next_state
& DS_UNBLOCK
) {
328 dprintf(" %d:-blocked", mdi
->disk
.raid_disk
);
329 write_attr("-blocked", mdi
->state_fd
);
332 if ((mdi
->next_state
& DS_REMOVE
) && mdi
->state_fd
>= 0) {
335 /* the kernel may not be able to immediately remove the
336 * disk, we can simply wait until the next event to try
339 remove_result
= write_attr("remove", mdi
->state_fd
);
340 if (remove_result
> 0) {
341 dprintf(" %d:removed", mdi
->disk
.raid_disk
);
342 close(mdi
->state_fd
);
346 if (mdi
->next_state
& DS_INSYNC
) {
347 write_attr("+in_sync", mdi
->state_fd
);
348 dprintf(" %d:+in_sync", mdi
->disk
.raid_disk
);
353 /* move curr_ to prev_ */
354 a
->prev_state
= a
->curr_state
;
356 a
->prev_action
= a
->curr_action
;
358 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
) {
359 mdi
->prev_state
= mdi
->curr_state
;
363 if (check_degraded
) {
364 /* manager will do the actual check */
365 a
->check_degraded
= 1;
375 static struct mdinfo
*
376 find_device(struct active_array
*a
, int major
, int minor
)
380 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
381 if (mdi
->disk
.major
== major
&& mdi
->disk
.minor
== minor
)
387 static void reconcile_failed(struct active_array
*aa
, struct mdinfo
*failed
)
389 struct active_array
*a
;
390 struct mdinfo
*victim
;
392 for (a
= aa
; a
; a
= a
->next
) {
395 victim
= find_device(a
, failed
->disk
.major
, failed
->disk
.minor
);
399 if (!(victim
->curr_state
& DS_FAULTY
))
400 write_attr("faulty", victim
->state_fd
);
405 static void dprint_wake_reasons(fd_set
*fds
)
413 fprintf(stderr
, "monitor: wake ( ");
414 for (i
= 0; i
< FD_SETSIZE
; i
++) {
415 if (FD_ISSET(i
, fds
)) {
416 sprintf(proc_path
, "/proc/%d/fd/%d",
419 rv
= readlink(proc_path
, link
, sizeof(link
) - 1);
421 fprintf(stderr
, "%d:unknown ", i
);
425 basename
= strrchr(link
, '/');
426 fprintf(stderr
, "%d:%s ",
427 i
, basename
? ++basename
: link
);
430 fprintf(stderr
, ")\n");
434 int monitor_loop_cnt
;
436 static int wait_and_act(struct supertype
*container
, int nowait
)
440 struct active_array
**aap
= &container
->arrays
;
441 struct active_array
*a
, **ap
;
444 static unsigned int dirty_arrays
= ~0; /* start at some non-zero value */
448 for (ap
= aap
; *ap
;) {
450 /* once an array has been deactivated we want to
451 * ask the manager to discard it.
465 add_fd(&rfds
, &maxfd
, a
->info
.state_fd
);
466 add_fd(&rfds
, &maxfd
, a
->action_fd
);
467 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
468 add_fd(&rfds
, &maxfd
, mdi
->state_fd
);
473 if (manager_ready
&& (*aap
== NULL
|| (sigterm
&& !dirty_arrays
))) {
474 /* No interesting arrays, or we have been told to
475 * terminate and everything is clean. Lets see about
476 * exiting. Note that blocking at this point is not a
477 * problem as there are no active arrays, there is
478 * nothing that we need to be ready to do.
480 int fd
= open_dev_excl(container
->devnum
);
481 if (fd
>= 0 || errno
!= EBUSY
) {
482 /* OK, we are safe to leave */
483 if (sigterm
&& !dirty_arrays
)
484 dprintf("caught sigterm, all clean... exiting\n");
486 dprintf("no arrays to monitor... exiting\n");
487 remove_pidfile(container
->devname
);
496 sigprocmask(SIG_UNBLOCK
, NULL
, &set
);
497 sigdelset(&set
, SIGUSR1
);
498 monitor_loop_cnt
|= 1;
499 rv
= pselect(maxfd
+1, &rfds
, NULL
, NULL
, NULL
, &set
);
500 monitor_loop_cnt
+= 1;
501 if (rv
== -1 && errno
== EINTR
)
504 dprint_wake_reasons(&rfds
);
510 struct metadata_update
*this;
512 for (this = update_queue
; this ; this = this->next
)
513 container
->ss
->process_update(container
, this);
515 update_queue_handled
= update_queue
;
518 container
->ss
->sync_metadata(container
);
523 for (a
= *aap
; a
; a
= a
->next
) {
526 if (a
->replaces
&& !discard_this
) {
527 struct active_array
**ap
;
528 for (ap
= &a
->next
; *ap
&& *ap
!= a
->replaces
;
533 discard_this
= a
->replaces
;
535 /* FIXME check if device->state_fd need to be cleared?*/
539 is_dirty
= read_and_act(a
);
541 dirty_arrays
+= is_dirty
;
542 /* when terminating stop manipulating the array after it
543 * is clean, but make sure read_and_act() is given a
544 * chance to handle 'active_idle'
546 if (sigterm
&& !is_dirty
)
547 a
->container
= NULL
; /* stop touching this array */
551 /* propagate failures across container members */
552 for (a
= *aap
; a
; a
= a
->next
) {
555 for (mdi
= a
->info
.devs
; mdi
; mdi
= mdi
->next
)
556 if (mdi
->curr_state
& DS_FAULTY
)
557 reconcile_failed(*aap
, mdi
);
563 void do_monitor(struct supertype
*container
)
568 rv
= wait_and_act(container
, first
);