]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdmon - monitor external metadata arrays | |
3 | * | |
4 | * Copyright (C) 2007-2008 Neil Brown <neilb@suse.de> | |
5 | * Copyright (C) 2007-2008 Intel Corporation | |
6 | * | |
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. | |
10 | * | |
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 | |
14 | * more details. | |
15 | * | |
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. | |
19 | */ | |
20 | ||
21 | #include "mdadm.h" | |
22 | #include "mdmon.h" | |
23 | #include <sys/syscall.h> | |
24 | #include <sys/select.h> | |
25 | #include <signal.h> | |
26 | ||
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 | |
32 | }; | |
33 | ||
34 | static int write_attr(char *attr, int fd) | |
35 | { | |
36 | return write(fd, attr, strlen(attr)); | |
37 | } | |
38 | ||
39 | static void add_fd(fd_set *fds, int *maxfd, int fd) | |
40 | { | |
41 | if (fd < 0) | |
42 | return; | |
43 | if (fd > *maxfd) | |
44 | *maxfd = fd; | |
45 | FD_SET(fd, fds); | |
46 | } | |
47 | ||
48 | static int read_attr(char *buf, int len, int fd) | |
49 | { | |
50 | int n; | |
51 | ||
52 | if (fd < 0) { | |
53 | buf[0] = 0; | |
54 | return 0; | |
55 | } | |
56 | lseek(fd, 0, 0); | |
57 | n = read(fd, buf, len - 1); | |
58 | ||
59 | if (n <= 0) { | |
60 | buf[0] = 0; | |
61 | return 0; | |
62 | } | |
63 | buf[n] = 0; | |
64 | if (buf[n-1] == '\n') | |
65 | buf[n-1] = 0; | |
66 | return n; | |
67 | } | |
68 | ||
69 | int get_resync_start(struct active_array *a) | |
70 | { | |
71 | char buf[30]; | |
72 | int n; | |
73 | ||
74 | n = read_attr(buf, 30, a->resync_start_fd); | |
75 | if (n <= 0) | |
76 | return n; | |
77 | if (strncmp(buf, "none", 4) == 0) | |
78 | a->resync_start = ~0ULL; | |
79 | else | |
80 | a->resync_start = strtoull(buf, NULL, 10); | |
81 | ||
82 | return 1; | |
83 | } | |
84 | ||
85 | ||
86 | static enum array_state read_state(int fd) | |
87 | { | |
88 | char buf[20]; | |
89 | int n = read_attr(buf, 20, fd); | |
90 | ||
91 | if (n <= 0) | |
92 | return bad_word; | |
93 | return (enum array_state) sysfs_match_word(buf, array_states); | |
94 | } | |
95 | ||
96 | static enum sync_action read_action( int fd) | |
97 | { | |
98 | char buf[20]; | |
99 | int n = read_attr(buf, 20, fd); | |
100 | ||
101 | if (n <= 0) | |
102 | return bad_action; | |
103 | return (enum sync_action) sysfs_match_word(buf, sync_actions); | |
104 | } | |
105 | ||
106 | int read_dev_state(int fd) | |
107 | { | |
108 | char buf[60]; | |
109 | int n = read_attr(buf, 60, fd); | |
110 | char *cp; | |
111 | int rv = 0; | |
112 | ||
113 | if (n <= 0) | |
114 | return 0; | |
115 | ||
116 | cp = buf; | |
117 | while (cp) { | |
118 | if (sysfs_attr_match(cp, "faulty")) | |
119 | rv |= DS_FAULTY; | |
120 | if (sysfs_attr_match(cp, "in_sync")) | |
121 | rv |= DS_INSYNC; | |
122 | if (sysfs_attr_match(cp, "write_mostly")) | |
123 | rv |= DS_WRITE_MOSTLY; | |
124 | if (sysfs_attr_match(cp, "spare")) | |
125 | rv |= DS_SPARE; | |
126 | if (sysfs_attr_match(cp, "blocked")) | |
127 | rv |= DS_BLOCKED; | |
128 | cp = strchr(cp, ','); | |
129 | if (cp) | |
130 | cp++; | |
131 | } | |
132 | return rv; | |
133 | } | |
134 | ||
135 | static void signal_manager(void) | |
136 | { | |
137 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ | |
138 | int pid = getpid(); | |
139 | syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1); | |
140 | } | |
141 | ||
142 | /* Monitor a set of active md arrays - all of which share the | |
143 | * same metadata - and respond to events that require | |
144 | * metadata update. | |
145 | * | |
146 | * New arrays are detected by another thread which allocates | |
147 | * required memory and attaches the data structure to our list. | |
148 | * | |
149 | * Events: | |
150 | * Array stops. | |
151 | * This is detected by array_state going to 'clear' or 'inactive'. | |
152 | * while we thought it was active. | |
153 | * Response is to mark metadata as clean and 'clear' the array(??) | |
154 | * write-pending | |
155 | * array_state if 'write-pending' | |
156 | * We mark metadata as 'dirty' then set array to 'active'. | |
157 | * active_idle | |
158 | * Either ignore, or mark clean, then mark metadata as clean. | |
159 | * | |
160 | * device fails | |
161 | * detected by rd-N/state reporting "faulty" | |
162 | * mark device as 'failed' in metadata, let the kernel release the | |
163 | * device by writing '-blocked' to rd/state, and finally write 'remove' to | |
164 | * rd/state. Before a disk can be replaced it must be failed and removed | |
165 | * from all container members, this will be preemptive for the other | |
166 | * arrays... safe? | |
167 | * | |
168 | * sync completes | |
169 | * sync_action was 'resync' and becomes 'idle' and resync_start becomes | |
170 | * MaxSector | |
171 | * Notify metadata that sync is complete. | |
172 | * | |
173 | * recovery completes | |
174 | * sync_action changes from 'recover' to 'idle' | |
175 | * Check each device state and mark metadata if 'faulty' or 'in_sync'. | |
176 | * | |
177 | * deal with resync | |
178 | * This only happens on finding a new array... mdadm will have set | |
179 | * 'resync_start' to the correct value. If 'resync_start' indicates that an | |
180 | * resync needs to occur set the array to the 'active' state rather than the | |
181 | * initial read-auto state. | |
182 | * | |
183 | * | |
184 | * | |
185 | * We wait for a change (poll/select) on array_state, sync_action, and | |
186 | * each rd-X/state file. | |
187 | * When we get any change, we check everything. So read each state file, | |
188 | * then decide what to do. | |
189 | * | |
190 | * The core action is to write new metadata to all devices in the array. | |
191 | * This is done at most once on any wakeup. | |
192 | * After that we might: | |
193 | * - update the array_state | |
194 | * - set the role of some devices. | |
195 | * - request a sync_action | |
196 | * | |
197 | */ | |
198 | ||
199 | static int read_and_act(struct active_array *a) | |
200 | { | |
201 | int check_degraded = 0; | |
202 | int deactivate = 0; | |
203 | struct mdinfo *mdi; | |
204 | int dirty = 0; | |
205 | ||
206 | a->next_state = bad_word; | |
207 | a->next_action = bad_action; | |
208 | ||
209 | a->curr_state = read_state(a->info.state_fd); | |
210 | a->curr_action = read_action(a->action_fd); | |
211 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { | |
212 | mdi->next_state = 0; | |
213 | if (mdi->state_fd >= 0) | |
214 | mdi->curr_state = read_dev_state(mdi->state_fd); | |
215 | } | |
216 | ||
217 | if (a->curr_state <= inactive && | |
218 | a->prev_state > inactive) { | |
219 | /* array has been stopped */ | |
220 | get_resync_start(a); | |
221 | a->container->ss->set_array_state(a, 1); | |
222 | a->next_state = clear; | |
223 | deactivate = 1; | |
224 | } | |
225 | if (a->curr_state == write_pending) { | |
226 | get_resync_start(a); | |
227 | a->container->ss->set_array_state(a, 0); | |
228 | a->next_state = active; | |
229 | dirty = 1; | |
230 | } | |
231 | if (a->curr_state == active_idle) { | |
232 | /* Set array to 'clean' FIRST, then mark clean | |
233 | * in the metadata | |
234 | */ | |
235 | a->next_state = clean; | |
236 | dirty = 1; | |
237 | } | |
238 | if (a->curr_state == clean) { | |
239 | get_resync_start(a); | |
240 | a->container->ss->set_array_state(a, 1); | |
241 | } | |
242 | if (a->curr_state == active || | |
243 | a->curr_state == suspended || | |
244 | a->curr_state == bad_word) | |
245 | dirty = 1; | |
246 | if (a->curr_state == readonly) { | |
247 | /* Well, I'm ready to handle things. If readonly | |
248 | * wasn't requested, transition to read-auto. | |
249 | */ | |
250 | char buf[64]; | |
251 | read_attr(buf, sizeof(buf), a->metadata_fd); | |
252 | if (strncmp(buf, "external:-", 10) == 0) { | |
253 | /* explicit request for readonly array. Leave it alone */ | |
254 | ; | |
255 | } else { | |
256 | get_resync_start(a); | |
257 | if (a->container->ss->set_array_state(a, 2)) | |
258 | a->next_state = read_auto; /* array is clean */ | |
259 | else { | |
260 | a->next_state = active; /* Now active for recovery etc */ | |
261 | dirty = 1; | |
262 | } | |
263 | } | |
264 | } | |
265 | ||
266 | if (!deactivate && | |
267 | a->curr_action == idle && | |
268 | a->prev_action == resync) { | |
269 | /* A resync has finished. The endpoint is recorded in | |
270 | * 'sync_start'. We don't update the metadata | |
271 | * until the array goes inactive or readonly though. | |
272 | * Just check if we need to fiddle spares. | |
273 | */ | |
274 | get_resync_start(a); | |
275 | a->container->ss->set_array_state(a, a->curr_state <= clean); | |
276 | check_degraded = 1; | |
277 | } | |
278 | ||
279 | if (!deactivate && | |
280 | a->curr_action == idle && | |
281 | a->prev_action == recover) { | |
282 | /* A recovery has finished. Some disks may be in sync now, | |
283 | * and the array may no longer be degraded | |
284 | */ | |
285 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { | |
286 | a->container->ss->set_disk(a, mdi->disk.raid_disk, | |
287 | mdi->curr_state); | |
288 | if (! (mdi->curr_state & DS_INSYNC)) | |
289 | check_degraded = 1; | |
290 | } | |
291 | } | |
292 | ||
293 | /* Check for failures and if found: | |
294 | * 1/ Record the failure in the metadata and unblock the device. | |
295 | * FIXME update the kernel to stop notifying on failed drives when | |
296 | * the array is readonly and we have cleared 'blocked' | |
297 | * 2/ Try to remove the device if the array is writable, or can be | |
298 | * made writable. | |
299 | */ | |
300 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { | |
301 | if (mdi->curr_state & DS_FAULTY) { | |
302 | a->container->ss->set_disk(a, mdi->disk.raid_disk, | |
303 | mdi->curr_state); | |
304 | check_degraded = 1; | |
305 | mdi->next_state |= DS_UNBLOCK; | |
306 | if (a->curr_state == read_auto) { | |
307 | a->container->ss->set_array_state(a, 0); | |
308 | a->next_state = active; | |
309 | } | |
310 | if (a->curr_state > readonly) | |
311 | mdi->next_state |= DS_REMOVE; | |
312 | } | |
313 | } | |
314 | ||
315 | a->container->ss->sync_metadata(a->container); | |
316 | dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member, | |
317 | array_states[a->curr_state], sync_actions[a->curr_action]); | |
318 | ||
319 | /* Effect state changes in the array */ | |
320 | if (a->next_state != bad_word) { | |
321 | dprintf(" state:%s", array_states[a->next_state]); | |
322 | write_attr(array_states[a->next_state], a->info.state_fd); | |
323 | } | |
324 | if (a->next_action != bad_action) { | |
325 | write_attr(sync_actions[a->next_action], a->action_fd); | |
326 | dprintf(" action:%s", sync_actions[a->next_action]); | |
327 | } | |
328 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { | |
329 | if (mdi->next_state & DS_UNBLOCK) { | |
330 | dprintf(" %d:-blocked", mdi->disk.raid_disk); | |
331 | write_attr("-blocked", mdi->state_fd); | |
332 | } | |
333 | ||
334 | if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) { | |
335 | int remove_result; | |
336 | ||
337 | /* the kernel may not be able to immediately remove the | |
338 | * disk, we can simply wait until the next event to try | |
339 | * again. | |
340 | */ | |
341 | remove_result = write_attr("remove", mdi->state_fd); | |
342 | if (remove_result > 0) { | |
343 | dprintf(" %d:removed", mdi->disk.raid_disk); | |
344 | close(mdi->state_fd); | |
345 | mdi->state_fd = -1; | |
346 | } | |
347 | } | |
348 | if (mdi->next_state & DS_INSYNC) { | |
349 | write_attr("+in_sync", mdi->state_fd); | |
350 | dprintf(" %d:+in_sync", mdi->disk.raid_disk); | |
351 | } | |
352 | } | |
353 | dprintf(" )\n"); | |
354 | ||
355 | /* move curr_ to prev_ */ | |
356 | a->prev_state = a->curr_state; | |
357 | ||
358 | a->prev_action = a->curr_action; | |
359 | ||
360 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { | |
361 | mdi->prev_state = mdi->curr_state; | |
362 | mdi->next_state = 0; | |
363 | } | |
364 | ||
365 | if (check_degraded) { | |
366 | /* manager will do the actual check */ | |
367 | a->check_degraded = 1; | |
368 | signal_manager(); | |
369 | } | |
370 | ||
371 | if (deactivate) | |
372 | a->container = NULL; | |
373 | ||
374 | return dirty; | |
375 | } | |
376 | ||
377 | static struct mdinfo * | |
378 | find_device(struct active_array *a, int major, int minor) | |
379 | { | |
380 | struct mdinfo *mdi; | |
381 | ||
382 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
383 | if (mdi->disk.major == major && mdi->disk.minor == minor) | |
384 | return mdi; | |
385 | ||
386 | return NULL; | |
387 | } | |
388 | ||
389 | static void reconcile_failed(struct active_array *aa, struct mdinfo *failed) | |
390 | { | |
391 | struct active_array *a; | |
392 | struct mdinfo *victim; | |
393 | ||
394 | for (a = aa; a; a = a->next) { | |
395 | if (!a->container) | |
396 | continue; | |
397 | victim = find_device(a, failed->disk.major, failed->disk.minor); | |
398 | if (!victim) | |
399 | continue; | |
400 | ||
401 | if (!(victim->curr_state & DS_FAULTY)) | |
402 | write_attr("faulty", victim->state_fd); | |
403 | } | |
404 | } | |
405 | ||
406 | #ifdef DEBUG | |
407 | static void dprint_wake_reasons(fd_set *fds) | |
408 | { | |
409 | int i; | |
410 | char proc_path[256]; | |
411 | char link[256]; | |
412 | char *basename; | |
413 | int rv; | |
414 | ||
415 | fprintf(stderr, "monitor: wake ( "); | |
416 | for (i = 0; i < FD_SETSIZE; i++) { | |
417 | if (FD_ISSET(i, fds)) { | |
418 | sprintf(proc_path, "/proc/%d/fd/%d", | |
419 | (int) getpid(), i); | |
420 | ||
421 | rv = readlink(proc_path, link, sizeof(link) - 1); | |
422 | if (rv < 0) { | |
423 | fprintf(stderr, "%d:unknown ", i); | |
424 | continue; | |
425 | } | |
426 | link[rv] = '\0'; | |
427 | basename = strrchr(link, '/'); | |
428 | fprintf(stderr, "%d:%s ", | |
429 | i, basename ? ++basename : link); | |
430 | } | |
431 | } | |
432 | fprintf(stderr, ")\n"); | |
433 | } | |
434 | #endif | |
435 | ||
436 | int monitor_loop_cnt; | |
437 | ||
438 | static int wait_and_act(struct supertype *container, int nowait) | |
439 | { | |
440 | fd_set rfds; | |
441 | int maxfd = 0; | |
442 | struct active_array **aap = &container->arrays; | |
443 | struct active_array *a, **ap; | |
444 | int rv; | |
445 | struct mdinfo *mdi; | |
446 | static unsigned int dirty_arrays = ~0; /* start at some non-zero value */ | |
447 | ||
448 | FD_ZERO(&rfds); | |
449 | ||
450 | for (ap = aap ; *ap ;) { | |
451 | a = *ap; | |
452 | /* once an array has been deactivated we want to | |
453 | * ask the manager to discard it. | |
454 | */ | |
455 | if (!a->container) { | |
456 | if (discard_this) { | |
457 | ap = &(*ap)->next; | |
458 | continue; | |
459 | } | |
460 | *ap = a->next; | |
461 | a->next = NULL; | |
462 | discard_this = a; | |
463 | signal_manager(); | |
464 | continue; | |
465 | } | |
466 | ||
467 | add_fd(&rfds, &maxfd, a->info.state_fd); | |
468 | add_fd(&rfds, &maxfd, a->action_fd); | |
469 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
470 | add_fd(&rfds, &maxfd, mdi->state_fd); | |
471 | ||
472 | ap = &(*ap)->next; | |
473 | } | |
474 | ||
475 | if (manager_ready && (*aap == NULL || (sigterm && !dirty_arrays))) { | |
476 | /* No interesting arrays, or we have been told to | |
477 | * terminate and everything is clean. Lets see about | |
478 | * exiting. Note that blocking at this point is not a | |
479 | * problem as there are no active arrays, there is | |
480 | * nothing that we need to be ready to do. | |
481 | */ | |
482 | int fd = open_dev_excl(container->devnum); | |
483 | if (fd >= 0 || errno != EBUSY) { | |
484 | /* OK, we are safe to leave */ | |
485 | if (sigterm && !dirty_arrays) | |
486 | dprintf("caught sigterm, all clean... exiting\n"); | |
487 | else | |
488 | dprintf("no arrays to monitor... exiting\n"); | |
489 | remove_pidfile(container->devname); | |
490 | exit_now = 1; | |
491 | signal_manager(); | |
492 | exit(0); | |
493 | } | |
494 | } | |
495 | ||
496 | if (!nowait) { | |
497 | sigset_t set; | |
498 | sigprocmask(SIG_UNBLOCK, NULL, &set); | |
499 | sigdelset(&set, SIGUSR1); | |
500 | monitor_loop_cnt |= 1; | |
501 | rv = pselect(maxfd+1, NULL, NULL, &rfds, NULL, &set); | |
502 | monitor_loop_cnt += 1; | |
503 | if (rv == -1 && errno == EINTR) | |
504 | rv = 0; | |
505 | #ifdef DEBUG | |
506 | dprint_wake_reasons(&rfds); | |
507 | #endif | |
508 | ||
509 | } | |
510 | ||
511 | if (update_queue) { | |
512 | struct metadata_update *this; | |
513 | ||
514 | for (this = update_queue; this ; this = this->next) | |
515 | container->ss->process_update(container, this); | |
516 | ||
517 | update_queue_handled = update_queue; | |
518 | update_queue = NULL; | |
519 | signal_manager(); | |
520 | container->ss->sync_metadata(container); | |
521 | } | |
522 | ||
523 | rv = 0; | |
524 | dirty_arrays = 0; | |
525 | for (a = *aap; a ; a = a->next) { | |
526 | int is_dirty; | |
527 | ||
528 | if (a->replaces && !discard_this) { | |
529 | struct active_array **ap; | |
530 | for (ap = &a->next; *ap && *ap != a->replaces; | |
531 | ap = & (*ap)->next) | |
532 | ; | |
533 | if (*ap) | |
534 | *ap = (*ap)->next; | |
535 | discard_this = a->replaces; | |
536 | a->replaces = NULL; | |
537 | /* FIXME check if device->state_fd need to be cleared?*/ | |
538 | signal_manager(); | |
539 | } | |
540 | if (a->container) { | |
541 | is_dirty = read_and_act(a); | |
542 | rv |= 1; | |
543 | dirty_arrays += is_dirty; | |
544 | /* when terminating stop manipulating the array after it | |
545 | * is clean, but make sure read_and_act() is given a | |
546 | * chance to handle 'active_idle' | |
547 | */ | |
548 | if (sigterm && !is_dirty) | |
549 | a->container = NULL; /* stop touching this array */ | |
550 | } | |
551 | } | |
552 | ||
553 | /* propagate failures across container members */ | |
554 | for (a = *aap; a ; a = a->next) { | |
555 | if (!a->container) | |
556 | continue; | |
557 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
558 | if (mdi->curr_state & DS_FAULTY) | |
559 | reconcile_failed(*aap, mdi); | |
560 | } | |
561 | ||
562 | return rv; | |
563 | } | |
564 | ||
565 | void do_monitor(struct supertype *container) | |
566 | { | |
567 | int rv; | |
568 | int first = 1; | |
569 | do { | |
570 | rv = wait_and_act(container, first); | |
571 | first = 0; | |
572 | } while (rv >= 0); | |
573 | } |