]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mdmon - monitor external metadata arrays | |
3 | * | |
4 | * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de> | |
5 | * Copyright (C) 2007-2009 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 | /* | |
22 | * The management thread for monitoring active md arrays. | |
23 | * This thread does things which might block such as memory | |
24 | * allocation. | |
25 | * In particular: | |
26 | * | |
27 | * - Find out about new arrays in this container. | |
28 | * Allocate the data structures and open the files. | |
29 | * | |
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. | |
34 | * | |
35 | * - Respond to reshape requests by allocating new data structures | |
36 | * and opening new files. | |
37 | * | |
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. | |
44 | * | |
45 | * - When a device is added to the container, we add it to the metadata | |
46 | * as a spare. | |
47 | * | |
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. | |
52 | * | |
53 | * Check if number of failed devices suggests recovery is needed, and | |
54 | * skip if not. | |
55 | * Ask metadata to allocate a spare device | |
56 | * Add device as not in_sync and give a role | |
57 | * Update metadata. | |
58 | * Open sysfs files and pass to monitor. | |
59 | * Make sure that monitor Starts recovery.... | |
60 | * | |
61 | * - Pass on metadata updates from external programs such as | |
62 | * mdadm creating a new array. | |
63 | * | |
64 | * This is most-messy. | |
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. | |
67 | * | |
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. | |
74 | * | |
75 | */ | |
76 | ||
77 | /* | |
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. | |
84 | * | |
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. | |
88 | * | |
89 | */ | |
90 | ||
91 | /* | |
92 | * List of tasks: | |
93 | * - Watch for spares to be added to the container, and write updated | |
94 | * metadata to them. | |
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 | |
100 | * start a reshape. | |
101 | */ | |
102 | #ifndef _GNU_SOURCE | |
103 | #define _GNU_SOURCE | |
104 | #endif | |
105 | #include "mdadm.h" | |
106 | #include "mdmon.h" | |
107 | #include <sys/syscall.h> | |
108 | #include <sys/socket.h> | |
109 | #include <signal.h> | |
110 | ||
111 | static void close_aa(struct active_array *aa) | |
112 | { | |
113 | struct mdinfo *d; | |
114 | ||
115 | for (d = aa->info.devs; d; d = d->next) { | |
116 | close(d->recovery_fd); | |
117 | close(d->state_fd); | |
118 | close(d->bb_fd); | |
119 | close(d->ubb_fd); | |
120 | } | |
121 | ||
122 | if (aa->action_fd >= 0) | |
123 | close(aa->action_fd); | |
124 | if (aa->info.state_fd >= 0) | |
125 | close(aa->info.state_fd); | |
126 | if (aa->resync_start_fd >= 0) | |
127 | close(aa->resync_start_fd); | |
128 | if (aa->metadata_fd >= 0) | |
129 | close(aa->metadata_fd); | |
130 | if (aa->sync_completed_fd >= 0) | |
131 | close(aa->sync_completed_fd); | |
132 | if (aa->safe_mode_delay_fd >= 0) | |
133 | close(aa->safe_mode_delay_fd); | |
134 | } | |
135 | ||
136 | static void free_aa(struct active_array *aa) | |
137 | { | |
138 | /* Note that this doesn't close fds if they are being used | |
139 | * by a clone. ->container will be set for a clone | |
140 | */ | |
141 | dprintf("sys_name: %s\n", aa->info.sys_name); | |
142 | if (!aa->container) | |
143 | close_aa(aa); | |
144 | while (aa->info.devs) { | |
145 | struct mdinfo *d = aa->info.devs; | |
146 | aa->info.devs = d->next; | |
147 | free(d); | |
148 | } | |
149 | free(aa); | |
150 | } | |
151 | ||
152 | static struct active_array *duplicate_aa(struct active_array *aa) | |
153 | { | |
154 | struct active_array *newa = xmalloc(sizeof(*newa)); | |
155 | struct mdinfo **dp1, **dp2; | |
156 | ||
157 | *newa = *aa; | |
158 | newa->next = NULL; | |
159 | newa->replaces = NULL; | |
160 | newa->info.next = NULL; | |
161 | ||
162 | dp2 = &newa->info.devs; | |
163 | ||
164 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
165 | struct mdinfo *d; | |
166 | if ((*dp1)->state_fd < 0) | |
167 | continue; | |
168 | ||
169 | d = xmalloc(sizeof(*d)); | |
170 | *d = **dp1; | |
171 | *dp2 = d; | |
172 | dp2 = & d->next; | |
173 | } | |
174 | *dp2 = NULL; | |
175 | ||
176 | return newa; | |
177 | } | |
178 | ||
179 | static void wakeup_monitor(void) | |
180 | { | |
181 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ | |
182 | int pid = getpid(); | |
183 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
184 | } | |
185 | ||
186 | static void remove_old(void) | |
187 | { | |
188 | if (discard_this) { | |
189 | discard_this->next = NULL; | |
190 | free_aa(discard_this); | |
191 | if (pending_discard == discard_this) | |
192 | pending_discard = NULL; | |
193 | discard_this = NULL; | |
194 | wakeup_monitor(); | |
195 | } | |
196 | } | |
197 | ||
198 | static void replace_array(struct supertype *container, | |
199 | struct active_array *old, | |
200 | struct active_array *new) | |
201 | { | |
202 | /* To replace an array, we add it to the top of the list | |
203 | * marked with ->replaces to point to the original. | |
204 | * 'monitor' will take the original out of the list | |
205 | * and put it on 'discard_this'. We take it from there | |
206 | * and discard it. | |
207 | */ | |
208 | remove_old(); | |
209 | while (pending_discard) { | |
210 | while (discard_this == NULL) | |
211 | sleep(1); | |
212 | remove_old(); | |
213 | } | |
214 | pending_discard = old; | |
215 | new->replaces = old; | |
216 | new->next = container->arrays; | |
217 | container->arrays = new; | |
218 | wakeup_monitor(); | |
219 | } | |
220 | ||
221 | struct metadata_update *update_queue = NULL; | |
222 | struct metadata_update *update_queue_handled = NULL; | |
223 | struct metadata_update *update_queue_pending = NULL; | |
224 | ||
225 | static void free_updates(struct metadata_update **update) | |
226 | { | |
227 | while (*update) { | |
228 | struct metadata_update *this = *update; | |
229 | void **space_list = this->space_list; | |
230 | ||
231 | *update = this->next; | |
232 | free(this->buf); | |
233 | free(this->space); | |
234 | while (space_list) { | |
235 | void *space = space_list; | |
236 | space_list = *space_list; | |
237 | free(space); | |
238 | } | |
239 | free(this); | |
240 | } | |
241 | } | |
242 | ||
243 | void check_update_queue(struct supertype *container) | |
244 | { | |
245 | free_updates(&update_queue_handled); | |
246 | ||
247 | if (update_queue == NULL && | |
248 | update_queue_pending) { | |
249 | update_queue = update_queue_pending; | |
250 | update_queue_pending = NULL; | |
251 | wakeup_monitor(); | |
252 | } | |
253 | } | |
254 | ||
255 | static void queue_metadata_update(struct metadata_update *mu) | |
256 | { | |
257 | struct metadata_update **qp; | |
258 | ||
259 | qp = &update_queue_pending; | |
260 | while (*qp) | |
261 | qp = & ((*qp)->next); | |
262 | *qp = mu; | |
263 | } | |
264 | ||
265 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) | |
266 | { | |
267 | int dfd; | |
268 | char nm[20]; | |
269 | struct metadata_update *update = NULL; | |
270 | mdu_disk_info_t dk = { | |
271 | .number = -1, | |
272 | .major = sd->disk.major, | |
273 | .minor = sd->disk.minor, | |
274 | .raid_disk = -1, | |
275 | .state = 0, | |
276 | }; | |
277 | ||
278 | dprintf("add %d:%d to container\n", sd->disk.major, sd->disk.minor); | |
279 | ||
280 | sd->next = st->devs; | |
281 | st->devs = sd; | |
282 | ||
283 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
284 | dfd = dev_open(nm, O_RDWR); | |
285 | if (dfd < 0) | |
286 | return; | |
287 | ||
288 | st->update_tail = &update; | |
289 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); | |
290 | st->ss->write_init_super(st); | |
291 | queue_metadata_update(update); | |
292 | st->update_tail = NULL; | |
293 | } | |
294 | ||
295 | /* | |
296 | * Create and queue update structure about the removed disks. | |
297 | * The update is prepared by super type handler and passed to the monitor | |
298 | * thread. | |
299 | */ | |
300 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
301 | { | |
302 | struct metadata_update *update = NULL; | |
303 | mdu_disk_info_t dk = { | |
304 | .number = -1, | |
305 | .major = sd->disk.major, | |
306 | .minor = sd->disk.minor, | |
307 | .raid_disk = -1, | |
308 | .state = 0, | |
309 | }; | |
310 | dprintf("remove %d:%d from container\n", | |
311 | sd->disk.major, sd->disk.minor); | |
312 | ||
313 | st->update_tail = &update; | |
314 | st->ss->remove_from_super(st, &dk); | |
315 | /* FIXME this write_init_super shouldn't be here. | |
316 | * We have it after add_to_super to write to new device, | |
317 | * but with 'remove' we don't ant to write to that device! | |
318 | */ | |
319 | st->ss->write_init_super(st); | |
320 | queue_metadata_update(update); | |
321 | st->update_tail = NULL; | |
322 | } | |
323 | ||
324 | static void manage_container(struct mdstat_ent *mdstat, | |
325 | struct supertype *container) | |
326 | { | |
327 | /* Of interest here are: | |
328 | * - if a new device has been added to the container, we | |
329 | * add it to the array ignoring any metadata on it. | |
330 | * - if a device has been removed from the container, we | |
331 | * remove it from the device list and update the metadata. | |
332 | * FIXME should we look for compatible metadata and take hints | |
333 | * about spare assignment.... probably not. | |
334 | */ | |
335 | if (mdstat->devcnt != container->devcnt) { | |
336 | struct mdinfo **cdp, *cd, *di, *mdi; | |
337 | int found; | |
338 | ||
339 | /* read /sys/block/NAME/md/dev-??/block/dev to find out | |
340 | * what is there, and compare with container->info.devs | |
341 | * To see what is removed and what is added. | |
342 | * These need to be remove from, or added to, the array | |
343 | */ | |
344 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); | |
345 | if (!mdi) { | |
346 | /* invalidate the current count so we can try again */ | |
347 | container->devcnt = -1; | |
348 | return; | |
349 | } | |
350 | ||
351 | /* check for removals */ | |
352 | for (cdp = &container->devs; *cdp; ) { | |
353 | found = 0; | |
354 | for (di = mdi->devs; di; di = di->next) | |
355 | if (di->disk.major == (*cdp)->disk.major && | |
356 | di->disk.minor == (*cdp)->disk.minor) { | |
357 | found = 1; | |
358 | break; | |
359 | } | |
360 | if (!found) { | |
361 | cd = *cdp; | |
362 | *cdp = (*cdp)->next; | |
363 | remove_disk_from_container(container, cd); | |
364 | free(cd); | |
365 | } else | |
366 | cdp = &(*cdp)->next; | |
367 | } | |
368 | ||
369 | /* check for additions */ | |
370 | for (di = mdi->devs; di; di = di->next) { | |
371 | for (cd = container->devs; cd; cd = cd->next) | |
372 | if (di->disk.major == cd->disk.major && | |
373 | di->disk.minor == cd->disk.minor) | |
374 | break; | |
375 | if (!cd) { | |
376 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
377 | ||
378 | *newd = *di; | |
379 | add_disk_to_container(container, newd); | |
380 | } | |
381 | } | |
382 | sysfs_free(mdi); | |
383 | container->devcnt = mdstat->devcnt; | |
384 | } | |
385 | } | |
386 | ||
387 | static int sysfs_open2(char *devnum, char *name, char *attr) | |
388 | { | |
389 | int fd = sysfs_open(devnum, name, attr); | |
390 | if (fd >= 0) { | |
391 | /* seq_file in the kernel allocates buffer space | |
392 | * on the first read. Do that now so 'monitor' | |
393 | * never needs too. | |
394 | */ | |
395 | char buf[200]; | |
396 | if (read(fd, buf, sizeof(buf)) < 0) | |
397 | /* pretend not to ignore return value */ | |
398 | return fd; | |
399 | } | |
400 | return fd; | |
401 | } | |
402 | ||
403 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, | |
404 | struct active_array *aa) | |
405 | { | |
406 | if (!disk || !clone) | |
407 | return -1; | |
408 | ||
409 | *disk = *clone; | |
410 | disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
411 | "recovery_start"); | |
412 | if (disk->recovery_fd < 0) | |
413 | return -1; | |
414 | disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state"); | |
415 | if (disk->state_fd < 0) { | |
416 | close(disk->recovery_fd); | |
417 | return -1; | |
418 | } | |
419 | disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
420 | "bad_blocks"); | |
421 | if (disk->bb_fd < 0) { | |
422 | close(disk->recovery_fd); | |
423 | close(disk->state_fd); | |
424 | return -1; | |
425 | } | |
426 | disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
427 | "unacknowledged_bad_blocks"); | |
428 | if (disk->ubb_fd < 0) { | |
429 | close(disk->recovery_fd); | |
430 | close(disk->state_fd); | |
431 | close(disk->bb_fd); | |
432 | return -1; | |
433 | } | |
434 | disk->prev_state = read_dev_state(disk->state_fd); | |
435 | disk->curr_state = disk->prev_state; | |
436 | disk->next = aa->info.devs; | |
437 | aa->info.devs = disk; | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
442 | static void manage_member(struct mdstat_ent *mdstat, | |
443 | struct active_array *a) | |
444 | { | |
445 | /* Compare mdstat info with known state of member array. | |
446 | * We do not need to look for device state changes here, that | |
447 | * is dealt with by the monitor. | |
448 | * | |
449 | * If a reshape is being requested, monitor will have noticed | |
450 | * that sync_action changed and will have set check_reshape. | |
451 | * We just need to see if new devices have appeared. All metadata | |
452 | * updates will already have been processed. | |
453 | * | |
454 | * We also want to handle degraded arrays here by | |
455 | * trying to find and assign a spare. | |
456 | * We do that whenever the monitor tells us too. | |
457 | */ | |
458 | char buf[64]; | |
459 | int frozen; | |
460 | struct supertype *container = a->container; | |
461 | struct mdinfo *mdi; | |
462 | ||
463 | if (container == NULL) | |
464 | /* Raced with something */ | |
465 | return; | |
466 | ||
467 | if (mdstat->active) { | |
468 | // FIXME | |
469 | a->info.array.raid_disks = mdstat->raid_disks; | |
470 | // MORE | |
471 | } | |
472 | ||
473 | mdi = sysfs_read(-1, mdstat->devnm, | |
474 | GET_COMPONENT|GET_CONSISTENCY_POLICY); | |
475 | if (mdi) { | |
476 | a->info.component_size = mdi->component_size; | |
477 | a->info.consistency_policy = mdi->consistency_policy; | |
478 | sysfs_free(mdi); | |
479 | } | |
480 | ||
481 | /* honor 'frozen' */ | |
482 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
483 | frozen = buf[9] == '-'; | |
484 | else | |
485 | frozen = 1; /* can't read metadata_version assume the worst */ | |
486 | ||
487 | /* If sync_action is not 'idle' then don't try recovery now */ | |
488 | if (!frozen && | |
489 | sysfs_get_str(&a->info, NULL, "sync_action", | |
490 | buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0) | |
491 | frozen = 1; | |
492 | ||
493 | if (mdstat->level) { | |
494 | int level = map_name(pers, mdstat->level); | |
495 | if (level == 0 || level == LEVEL_LINEAR) { | |
496 | a->to_remove = 1; | |
497 | wakeup_monitor(); | |
498 | return; | |
499 | } | |
500 | else if (a->info.array.level != level && level > 0) { | |
501 | struct active_array *newa = duplicate_aa(a); | |
502 | if (newa) { | |
503 | newa->info.array.level = level; | |
504 | replace_array(container, a, newa); | |
505 | a = newa; | |
506 | } | |
507 | } | |
508 | } | |
509 | ||
510 | /* we are after monitor kick, | |
511 | * so container field can be cleared - check it again | |
512 | */ | |
513 | if (a->container == NULL) | |
514 | return; | |
515 | ||
516 | if (sigterm && a->info.safe_mode_delay != 1 && | |
517 | a->safe_mode_delay_fd >= 0) { | |
518 | long int new_delay = 1; | |
519 | char delay[10]; | |
520 | ssize_t len; | |
521 | ||
522 | len = snprintf(delay, sizeof(delay), "0.%03ld\n", new_delay); | |
523 | if (write(a->safe_mode_delay_fd, delay, len) == len) | |
524 | a->info.safe_mode_delay = new_delay; | |
525 | } | |
526 | ||
527 | /* We don't check the array while any update is pending, as it | |
528 | * might container a change (such as a spare assignment) which | |
529 | * could affect our decisions. | |
530 | */ | |
531 | if (a->check_degraded && !frozen && | |
532 | update_queue == NULL && update_queue_pending == NULL) { | |
533 | struct metadata_update *updates = NULL; | |
534 | struct mdinfo *newdev = NULL; | |
535 | struct active_array *newa; | |
536 | struct mdinfo *d; | |
537 | ||
538 | a->check_degraded = 0; | |
539 | ||
540 | /* The array may not be degraded, this is just a good time | |
541 | * to check. | |
542 | */ | |
543 | newdev = container->ss->activate_spare(a, &updates); | |
544 | if (!newdev) | |
545 | return; | |
546 | ||
547 | newa = duplicate_aa(a); | |
548 | if (!newa) | |
549 | goto out; | |
550 | /* prevent the kernel from activating the disk(s) before we | |
551 | * finish adding them | |
552 | */ | |
553 | dprintf("freezing %s\n", a->info.sys_name); | |
554 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); | |
555 | ||
556 | /* Add device to array and set offset/size/slot. | |
557 | * and open files for each newdev */ | |
558 | for (d = newdev; d ; d = d->next) { | |
559 | struct mdinfo *newd; | |
560 | ||
561 | newd = xmalloc(sizeof(*newd)); | |
562 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { | |
563 | free(newd); | |
564 | continue; | |
565 | } | |
566 | disk_init_and_add(newd, d, newa); | |
567 | } | |
568 | queue_metadata_update(updates); | |
569 | updates = NULL; | |
570 | while (update_queue_pending || update_queue) { | |
571 | check_update_queue(container); | |
572 | usleep(15*1000); | |
573 | } | |
574 | replace_array(container, a, newa); | |
575 | if (sysfs_set_str(&a->info, NULL, | |
576 | "sync_action", "recover") == 0) | |
577 | newa->prev_action = recover; | |
578 | dprintf("recovery started on %s\n", a->info.sys_name); | |
579 | out: | |
580 | while (newdev) { | |
581 | d = newdev->next; | |
582 | free(newdev); | |
583 | newdev = d; | |
584 | } | |
585 | free_updates(&updates); | |
586 | } | |
587 | ||
588 | if (a->check_reshape) { | |
589 | /* mdadm might have added some devices to the array. | |
590 | * We want to disk_init_and_add any such device to a | |
591 | * duplicate_aa and replace a with that. | |
592 | * mdstat doesn't have enough info so we sysfs_read | |
593 | * and look for new stuff. | |
594 | */ | |
595 | struct mdinfo *info, *d, *d2, *newd; | |
596 | unsigned long long array_size; | |
597 | struct active_array *newa = NULL; | |
598 | a->check_reshape = 0; | |
599 | info = sysfs_read(-1, mdstat->devnm, | |
600 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); | |
601 | if (!info) | |
602 | goto out2; | |
603 | for (d = info->devs; d; d = d->next) { | |
604 | if (d->disk.raid_disk < 0) | |
605 | continue; | |
606 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
607 | if (d2->disk.raid_disk == | |
608 | d->disk.raid_disk) | |
609 | break; | |
610 | if (d2) | |
611 | /* already have this one */ | |
612 | continue; | |
613 | if (!newa) { | |
614 | newa = duplicate_aa(a); | |
615 | if (!newa) | |
616 | break; | |
617 | } | |
618 | newd = xmalloc(sizeof(*newd)); | |
619 | disk_init_and_add(newd, d, newa); | |
620 | } | |
621 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 && | |
622 | a->info.custom_array_size > array_size*2) { | |
623 | sysfs_set_num(info, NULL, "array_size", | |
624 | a->info.custom_array_size/2); | |
625 | } | |
626 | out2: | |
627 | sysfs_free(info); | |
628 | if (newa) | |
629 | replace_array(container, a, newa); | |
630 | } | |
631 | } | |
632 | ||
633 | static int aa_ready(struct active_array *aa) | |
634 | { | |
635 | struct mdinfo *d; | |
636 | int level = aa->info.array.level; | |
637 | ||
638 | for (d = aa->info.devs; d; d = d->next) | |
639 | if (d->state_fd < 0) | |
640 | return 0; | |
641 | ||
642 | if (aa->info.state_fd < 0) | |
643 | return 0; | |
644 | ||
645 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
646 | return 0; | |
647 | ||
648 | if (!aa->container) | |
649 | return 0; | |
650 | ||
651 | return 1; | |
652 | } | |
653 | ||
654 | static void manage_new(struct mdstat_ent *mdstat, | |
655 | struct supertype *container, | |
656 | struct active_array *victim) | |
657 | { | |
658 | /* A new array has appeared in this container. | |
659 | * Hopefully it is already recorded in the metadata. | |
660 | * Check, then create the new array to report it to | |
661 | * the monitor. | |
662 | */ | |
663 | ||
664 | struct active_array *new; | |
665 | struct mdinfo *mdi, *di; | |
666 | char *inst; | |
667 | int i; | |
668 | int failed = 0; | |
669 | char buf[40]; | |
670 | ||
671 | /* check if array is ready to be monitored */ | |
672 | if (!mdstat->active || !mdstat->level) | |
673 | return; | |
674 | if (strcmp(mdstat->level, "raid0") == 0 || | |
675 | strcmp(mdstat->level, "linear") == 0) | |
676 | return; | |
677 | ||
678 | mdi = sysfs_read(-1, mdstat->devnm, | |
679 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| | |
680 | GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| | |
681 | GET_LAYOUT|GET_DEVS_ALL); | |
682 | ||
683 | if (!mdi) | |
684 | return; | |
685 | new = xcalloc(1, sizeof(*new)); | |
686 | ||
687 | strcpy(new->info.sys_name, mdstat->devnm); | |
688 | ||
689 | new->prev_state = new->curr_state = new->next_state = inactive; | |
690 | new->prev_action= new->curr_action= new->next_action= idle; | |
691 | ||
692 | new->container = container; | |
693 | ||
694 | inst = to_subarray(mdstat, container->devnm); | |
695 | ||
696 | new->info.array = mdi->array; | |
697 | new->info.component_size = mdi->component_size; | |
698 | ||
699 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
700 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
701 | ||
702 | for (di = mdi->devs; di; di = di->next) | |
703 | if (i == di->disk.raid_disk) | |
704 | break; | |
705 | ||
706 | if (disk_init_and_add(newd, di, new) != 0) { | |
707 | if (newd) | |
708 | free(newd); | |
709 | ||
710 | failed++; | |
711 | if (failed > new->info.array.failed_disks) { | |
712 | /* we cannot properly monitor without all working disks */ | |
713 | new->container = NULL; | |
714 | break; | |
715 | } | |
716 | } | |
717 | } | |
718 | ||
719 | new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action"); | |
720 | new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state"); | |
721 | new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start"); | |
722 | new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version"); | |
723 | new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed"); | |
724 | new->safe_mode_delay_fd = sysfs_open2(new->info.sys_name, NULL, | |
725 | "safe_mode_delay"); | |
726 | ||
727 | dprintf("inst: %s action: %d state: %d\n", inst, | |
728 | new->action_fd, new->info.state_fd); | |
729 | ||
730 | if (mdi->safe_mode_delay >= 50) | |
731 | /* Normal start, mdadm set this. */ | |
732 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
733 | else | |
734 | /* Restart, just pick a number */ | |
735 | new->info.safe_mode_delay = 5000; | |
736 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
737 | ||
738 | /* reshape_position is set by mdadm in sysfs | |
739 | * read this information for new arrays only (empty victim) | |
740 | */ | |
741 | if ((victim == NULL) && | |
742 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
743 | (strncmp(buf, "reshape", 7) == 0)) { | |
744 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
745 | &new->last_checkpoint) != 0) | |
746 | new->last_checkpoint = 0; | |
747 | else { | |
748 | int data_disks = mdi->array.raid_disks; | |
749 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
750 | data_disks--; | |
751 | if (mdi->array.level == 6) | |
752 | data_disks -= 2; | |
753 | ||
754 | new->last_checkpoint /= data_disks; | |
755 | } | |
756 | dprintf("mdmon: New monitored array is under reshape.\n" | |
757 | " Last checkpoint is: %llu\n", | |
758 | new->last_checkpoint); | |
759 | } | |
760 | ||
761 | sysfs_free(mdi); | |
762 | ||
763 | /* if everything checks out tell the metadata handler we want to | |
764 | * manage this instance | |
765 | */ | |
766 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
767 | pr_err("failed to monitor %s\n", | |
768 | mdstat->metadata_version); | |
769 | new->container = NULL; | |
770 | free_aa(new); | |
771 | } else { | |
772 | replace_array(container, victim, new); | |
773 | if (failed) { | |
774 | new->check_degraded = 1; | |
775 | manage_member(mdstat, new); | |
776 | } | |
777 | } | |
778 | } | |
779 | ||
780 | void manage(struct mdstat_ent *mdstat, struct supertype *container) | |
781 | { | |
782 | /* We have just read mdstat and need to compare it with | |
783 | * the known active arrays. | |
784 | * Arrays with the wrong metadata are ignored. | |
785 | */ | |
786 | ||
787 | for ( ; mdstat ; mdstat = mdstat->next) { | |
788 | struct active_array *a; | |
789 | if (strcmp(mdstat->devnm, container->devnm) == 0) { | |
790 | manage_container(mdstat, container); | |
791 | continue; | |
792 | } | |
793 | if (!is_container_member(mdstat, container->devnm)) | |
794 | /* Not for this array */ | |
795 | continue; | |
796 | /* Looks like a member of this container */ | |
797 | for (a = container->arrays; a; a = a->next) { | |
798 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { | |
799 | if (a->container && a->to_remove == 0) | |
800 | manage_member(mdstat, a); | |
801 | break; | |
802 | } | |
803 | } | |
804 | if ((a == NULL || !a->container) && !sigterm) | |
805 | manage_new(mdstat, container, a); | |
806 | } | |
807 | } | |
808 | ||
809 | static void handle_message(struct supertype *container, struct metadata_update *msg) | |
810 | { | |
811 | /* queue this metadata update through to the monitor */ | |
812 | ||
813 | struct metadata_update *mu; | |
814 | ||
815 | if (msg->len <= 0) | |
816 | while (update_queue_pending || update_queue) { | |
817 | check_update_queue(container); | |
818 | usleep(15*1000); | |
819 | } | |
820 | ||
821 | if (msg->len == 0) { /* ping_monitor */ | |
822 | int cnt; | |
823 | ||
824 | cnt = monitor_loop_cnt; | |
825 | if (cnt & 1) | |
826 | cnt += 2; /* wait until next pselect */ | |
827 | else | |
828 | cnt += 3; /* wait for 2 pselects */ | |
829 | wakeup_monitor(); | |
830 | ||
831 | while (monitor_loop_cnt - cnt < 0) | |
832 | usleep(10 * 1000); | |
833 | } else if (msg->len == -1) { /* ping_manager */ | |
834 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
835 | ||
836 | manage(mdstat, container); | |
837 | free_mdstat(mdstat); | |
838 | } else if (!sigterm) { | |
839 | mu = xmalloc(sizeof(*mu)); | |
840 | mu->len = msg->len; | |
841 | mu->buf = msg->buf; | |
842 | msg->buf = NULL; | |
843 | mu->space = NULL; | |
844 | mu->space_list = NULL; | |
845 | mu->next = NULL; | |
846 | if (container->ss->prepare_update) | |
847 | if (!container->ss->prepare_update(container, mu)) | |
848 | free_updates(&mu); | |
849 | queue_metadata_update(mu); | |
850 | } | |
851 | } | |
852 | ||
853 | void read_sock(struct supertype *container) | |
854 | { | |
855 | int fd; | |
856 | struct metadata_update msg; | |
857 | int terminate = 0; | |
858 | long fl; | |
859 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
860 | ||
861 | fd = accept(container->sock, NULL, NULL); | |
862 | if (fd < 0) | |
863 | return; | |
864 | ||
865 | fl = fcntl(fd, F_GETFL, 0); | |
866 | fl |= O_NONBLOCK; | |
867 | fcntl(fd, F_SETFL, fl); | |
868 | ||
869 | do { | |
870 | msg.buf = NULL; | |
871 | ||
872 | /* read and validate the message */ | |
873 | if (receive_message(fd, &msg, tmo) == 0) { | |
874 | handle_message(container, &msg); | |
875 | if (msg.len == 0) { | |
876 | /* ping reply with version */ | |
877 | msg.buf = Version; | |
878 | msg.len = strlen(Version) + 1; | |
879 | if (send_message(fd, &msg, tmo) < 0) | |
880 | terminate = 1; | |
881 | } else if (ack(fd, tmo) < 0) | |
882 | terminate = 1; | |
883 | } else | |
884 | terminate = 1; | |
885 | ||
886 | } while (!terminate); | |
887 | ||
888 | close(fd); | |
889 | } | |
890 | ||
891 | int exit_now = 0; | |
892 | int manager_ready = 0; | |
893 | void do_manager(struct supertype *container) | |
894 | { | |
895 | struct mdstat_ent *mdstat; | |
896 | sigset_t set; | |
897 | ||
898 | sigprocmask(SIG_UNBLOCK, NULL, &set); | |
899 | sigdelset(&set, SIGUSR1); | |
900 | sigdelset(&set, SIGTERM); | |
901 | ||
902 | do { | |
903 | ||
904 | if (exit_now) | |
905 | exit(0); | |
906 | ||
907 | /* Can only 'manage' things if 'monitor' is not making | |
908 | * structural changes to metadata, so need to check | |
909 | * update_queue | |
910 | */ | |
911 | if (update_queue == NULL) { | |
912 | mdstat = mdstat_read(1, 0); | |
913 | ||
914 | manage(mdstat, container); | |
915 | ||
916 | read_sock(container); | |
917 | ||
918 | free_mdstat(mdstat); | |
919 | } | |
920 | remove_old(); | |
921 | ||
922 | check_update_queue(container); | |
923 | ||
924 | manager_ready = 1; | |
925 | ||
926 | if (sigterm) | |
927 | wakeup_monitor(); | |
928 | ||
929 | if (update_queue == NULL) | |
930 | mdstat_wait_fd(container->sock, &set); | |
931 | else | |
932 | /* If an update is happening, just wait for signal */ | |
933 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
934 | } while(1); | |
935 | } |