]>
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 | } | |
133 | ||
134 | static void free_aa(struct active_array *aa) | |
135 | { | |
136 | /* Note that this doesn't close fds if they are being used | |
137 | * by a clone. ->container will be set for a clone | |
138 | */ | |
139 | dprintf("sys_name: %s\n", aa->info.sys_name); | |
140 | if (!aa->container) | |
141 | close_aa(aa); | |
142 | while (aa->info.devs) { | |
143 | struct mdinfo *d = aa->info.devs; | |
144 | aa->info.devs = d->next; | |
145 | free(d); | |
146 | } | |
147 | free(aa); | |
148 | } | |
149 | ||
150 | static struct active_array *duplicate_aa(struct active_array *aa) | |
151 | { | |
152 | struct active_array *newa = xmalloc(sizeof(*newa)); | |
153 | struct mdinfo **dp1, **dp2; | |
154 | ||
155 | *newa = *aa; | |
156 | newa->next = NULL; | |
157 | newa->replaces = NULL; | |
158 | newa->info.next = NULL; | |
159 | ||
160 | dp2 = &newa->info.devs; | |
161 | ||
162 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
163 | struct mdinfo *d; | |
164 | if ((*dp1)->state_fd < 0) | |
165 | continue; | |
166 | ||
167 | d = xmalloc(sizeof(*d)); | |
168 | *d = **dp1; | |
169 | *dp2 = d; | |
170 | dp2 = & d->next; | |
171 | } | |
172 | *dp2 = NULL; | |
173 | ||
174 | return newa; | |
175 | } | |
176 | ||
177 | static void wakeup_monitor(void) | |
178 | { | |
179 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ | |
180 | int pid = getpid(); | |
181 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
182 | } | |
183 | ||
184 | static void remove_old(void) | |
185 | { | |
186 | if (discard_this) { | |
187 | discard_this->next = NULL; | |
188 | free_aa(discard_this); | |
189 | if (pending_discard == discard_this) | |
190 | pending_discard = NULL; | |
191 | discard_this = NULL; | |
192 | wakeup_monitor(); | |
193 | } | |
194 | } | |
195 | ||
196 | static void replace_array(struct supertype *container, | |
197 | struct active_array *old, | |
198 | struct active_array *new) | |
199 | { | |
200 | /* To replace an array, we add it to the top of the list | |
201 | * marked with ->replaces to point to the original. | |
202 | * 'monitor' will take the original out of the list | |
203 | * and put it on 'discard_this'. We take it from there | |
204 | * and discard it. | |
205 | */ | |
206 | remove_old(); | |
207 | while (pending_discard) { | |
208 | while (discard_this == NULL) | |
209 | sleep(1); | |
210 | remove_old(); | |
211 | } | |
212 | pending_discard = old; | |
213 | new->replaces = old; | |
214 | new->next = container->arrays; | |
215 | container->arrays = new; | |
216 | wakeup_monitor(); | |
217 | } | |
218 | ||
219 | struct metadata_update *update_queue = NULL; | |
220 | struct metadata_update *update_queue_handled = NULL; | |
221 | struct metadata_update *update_queue_pending = NULL; | |
222 | ||
223 | static void free_updates(struct metadata_update **update) | |
224 | { | |
225 | while (*update) { | |
226 | struct metadata_update *this = *update; | |
227 | void **space_list = this->space_list; | |
228 | ||
229 | *update = this->next; | |
230 | free(this->buf); | |
231 | free(this->space); | |
232 | while (space_list) { | |
233 | void *space = space_list; | |
234 | space_list = *space_list; | |
235 | free(space); | |
236 | } | |
237 | free(this); | |
238 | } | |
239 | } | |
240 | ||
241 | void check_update_queue(struct supertype *container) | |
242 | { | |
243 | free_updates(&update_queue_handled); | |
244 | ||
245 | if (update_queue == NULL && | |
246 | update_queue_pending) { | |
247 | update_queue = update_queue_pending; | |
248 | update_queue_pending = NULL; | |
249 | wakeup_monitor(); | |
250 | } | |
251 | } | |
252 | ||
253 | static void queue_metadata_update(struct metadata_update *mu) | |
254 | { | |
255 | struct metadata_update **qp; | |
256 | ||
257 | qp = &update_queue_pending; | |
258 | while (*qp) | |
259 | qp = & ((*qp)->next); | |
260 | *qp = mu; | |
261 | } | |
262 | ||
263 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) | |
264 | { | |
265 | int dfd; | |
266 | char nm[20]; | |
267 | struct supertype *st2; | |
268 | struct metadata_update *update = NULL; | |
269 | struct mdinfo info; | |
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 | /* Check the metadata and see if it is already part of this | |
289 | * array | |
290 | */ | |
291 | st2 = dup_super(st); | |
292 | if (st2->ss->load_super(st2, dfd, NULL) == 0) { | |
293 | st2->ss->getinfo_super(st2, &info, NULL); | |
294 | if (st->ss->compare_super(st, st2) == 0 && | |
295 | info.disk.raid_disk >= 0) { | |
296 | /* Looks like a good member of array. | |
297 | * Just accept it. | |
298 | * mdadm will incorporate any parts into | |
299 | * active arrays. | |
300 | */ | |
301 | st2->ss->free_super(st2); | |
302 | return; | |
303 | } | |
304 | } | |
305 | st2->ss->free_super(st2); | |
306 | ||
307 | st->update_tail = &update; | |
308 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); | |
309 | st->ss->write_init_super(st); | |
310 | queue_metadata_update(update); | |
311 | st->update_tail = NULL; | |
312 | } | |
313 | ||
314 | /* | |
315 | * Create and queue update structure about the removed disks. | |
316 | * The update is prepared by super type handler and passed to the monitor | |
317 | * thread. | |
318 | */ | |
319 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
320 | { | |
321 | struct metadata_update *update = NULL; | |
322 | mdu_disk_info_t dk = { | |
323 | .number = -1, | |
324 | .major = sd->disk.major, | |
325 | .minor = sd->disk.minor, | |
326 | .raid_disk = -1, | |
327 | .state = 0, | |
328 | }; | |
329 | dprintf("remove %d:%d from container\n", | |
330 | sd->disk.major, sd->disk.minor); | |
331 | ||
332 | st->update_tail = &update; | |
333 | st->ss->remove_from_super(st, &dk); | |
334 | /* FIXME this write_init_super shouldn't be here. | |
335 | * We have it after add_to_super to write to new device, | |
336 | * but with 'remove' we don't ant to write to that device! | |
337 | */ | |
338 | st->ss->write_init_super(st); | |
339 | queue_metadata_update(update); | |
340 | st->update_tail = NULL; | |
341 | } | |
342 | ||
343 | static void manage_container(struct mdstat_ent *mdstat, | |
344 | struct supertype *container) | |
345 | { | |
346 | /* Of interest here are: | |
347 | * - if a new device has been added to the container, we | |
348 | * add it to the array ignoring any metadata on it. | |
349 | * - if a device has been removed from the container, we | |
350 | * remove it from the device list and update the metadata. | |
351 | * FIXME should we look for compatible metadata and take hints | |
352 | * about spare assignment.... probably not. | |
353 | */ | |
354 | if (mdstat->devcnt != container->devcnt) { | |
355 | struct mdinfo **cdp, *cd, *di, *mdi; | |
356 | int found; | |
357 | ||
358 | /* read /sys/block/NAME/md/dev-??/block/dev to find out | |
359 | * what is there, and compare with container->info.devs | |
360 | * To see what is removed and what is added. | |
361 | * These need to be remove from, or added to, the array | |
362 | */ | |
363 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); | |
364 | if (!mdi) { | |
365 | /* invalidate the current count so we can try again */ | |
366 | container->devcnt = -1; | |
367 | return; | |
368 | } | |
369 | ||
370 | /* check for removals */ | |
371 | for (cdp = &container->devs; *cdp; ) { | |
372 | found = 0; | |
373 | for (di = mdi->devs; di; di = di->next) | |
374 | if (di->disk.major == (*cdp)->disk.major && | |
375 | di->disk.minor == (*cdp)->disk.minor) { | |
376 | found = 1; | |
377 | break; | |
378 | } | |
379 | if (!found) { | |
380 | cd = *cdp; | |
381 | *cdp = (*cdp)->next; | |
382 | remove_disk_from_container(container, cd); | |
383 | free(cd); | |
384 | } else | |
385 | cdp = &(*cdp)->next; | |
386 | } | |
387 | ||
388 | /* check for additions */ | |
389 | for (di = mdi->devs; di; di = di->next) { | |
390 | for (cd = container->devs; cd; cd = cd->next) | |
391 | if (di->disk.major == cd->disk.major && | |
392 | di->disk.minor == cd->disk.minor) | |
393 | break; | |
394 | if (!cd) { | |
395 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
396 | ||
397 | *newd = *di; | |
398 | add_disk_to_container(container, newd); | |
399 | } | |
400 | } | |
401 | sysfs_free(mdi); | |
402 | container->devcnt = mdstat->devcnt; | |
403 | } | |
404 | } | |
405 | ||
406 | static int sysfs_open2(char *devnum, char *name, char *attr) | |
407 | { | |
408 | int fd = sysfs_open(devnum, name, attr); | |
409 | if (fd >= 0) { | |
410 | /* seq_file in the kernel allocates buffer space | |
411 | * on the first read. Do that now so 'monitor' | |
412 | * never needs too. | |
413 | */ | |
414 | char buf[200]; | |
415 | if (read(fd, buf, sizeof(buf)) < 0) | |
416 | /* pretend not to ignore return value */ | |
417 | return fd; | |
418 | } | |
419 | return fd; | |
420 | } | |
421 | ||
422 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, | |
423 | struct active_array *aa) | |
424 | { | |
425 | if (!disk || !clone) | |
426 | return -1; | |
427 | ||
428 | *disk = *clone; | |
429 | disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
430 | "recovery_start"); | |
431 | if (disk->recovery_fd < 0) | |
432 | return -1; | |
433 | disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state"); | |
434 | if (disk->state_fd < 0) { | |
435 | close(disk->recovery_fd); | |
436 | return -1; | |
437 | } | |
438 | disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
439 | "bad_blocks"); | |
440 | if (disk->bb_fd < 0) { | |
441 | close(disk->recovery_fd); | |
442 | close(disk->state_fd); | |
443 | return -1; | |
444 | } | |
445 | disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
446 | "unacknowledged_bad_blocks"); | |
447 | if (disk->ubb_fd < 0) { | |
448 | close(disk->recovery_fd); | |
449 | close(disk->state_fd); | |
450 | close(disk->bb_fd); | |
451 | return -1; | |
452 | } | |
453 | disk->prev_state = read_dev_state(disk->state_fd); | |
454 | disk->curr_state = disk->prev_state; | |
455 | disk->next = aa->info.devs; | |
456 | aa->info.devs = disk; | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
461 | static void manage_member(struct mdstat_ent *mdstat, | |
462 | struct active_array *a) | |
463 | { | |
464 | /* Compare mdstat info with known state of member array. | |
465 | * We do not need to look for device state changes here, that | |
466 | * is dealt with by the monitor. | |
467 | * | |
468 | * If a reshape is being requested, monitor will have noticed | |
469 | * that sync_action changed and will have set check_reshape. | |
470 | * We just need to see if new devices have appeared. All metadata | |
471 | * updates will already have been processed. | |
472 | * | |
473 | * We also want to handle degraded arrays here by | |
474 | * trying to find and assign a spare. | |
475 | * We do that whenever the monitor tells us too. | |
476 | */ | |
477 | char buf[64]; | |
478 | int frozen; | |
479 | struct supertype *container = a->container; | |
480 | struct mdinfo *mdi; | |
481 | ||
482 | if (container == NULL) | |
483 | /* Raced with something */ | |
484 | return; | |
485 | ||
486 | if (mdstat->active) { | |
487 | // FIXME | |
488 | a->info.array.raid_disks = mdstat->raid_disks; | |
489 | // MORE | |
490 | } | |
491 | ||
492 | mdi = sysfs_read(-1, mdstat->devnm, | |
493 | GET_COMPONENT|GET_CONSISTENCY_POLICY); | |
494 | if (mdi) { | |
495 | a->info.component_size = mdi->component_size; | |
496 | a->info.consistency_policy = mdi->consistency_policy; | |
497 | sysfs_free(mdi); | |
498 | } | |
499 | ||
500 | /* honor 'frozen' */ | |
501 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
502 | frozen = buf[9] == '-'; | |
503 | else | |
504 | frozen = 1; /* can't read metadata_version assume the worst */ | |
505 | ||
506 | /* If sync_action is not 'idle' then don't try recovery now */ | |
507 | if (!frozen && | |
508 | sysfs_get_str(&a->info, NULL, "sync_action", | |
509 | buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0) | |
510 | frozen = 1; | |
511 | ||
512 | if (mdstat->level) { | |
513 | int level = map_name(pers, mdstat->level); | |
514 | if (level == 0 || level == LEVEL_LINEAR) { | |
515 | a->to_remove = 1; | |
516 | wakeup_monitor(); | |
517 | return; | |
518 | } | |
519 | else if (a->info.array.level != level && level > 0) { | |
520 | struct active_array *newa = duplicate_aa(a); | |
521 | if (newa) { | |
522 | newa->info.array.level = level; | |
523 | replace_array(container, a, newa); | |
524 | a = newa; | |
525 | } | |
526 | } | |
527 | } | |
528 | ||
529 | /* we are after monitor kick, | |
530 | * so container field can be cleared - check it again | |
531 | */ | |
532 | if (a->container == NULL) | |
533 | return; | |
534 | ||
535 | if (sigterm && a->info.safe_mode_delay != 1) { | |
536 | sysfs_set_safemode(&a->info, 1); | |
537 | a->info.safe_mode_delay = 1; | |
538 | } | |
539 | ||
540 | /* We don't check the array while any update is pending, as it | |
541 | * might container a change (such as a spare assignment) which | |
542 | * could affect our decisions. | |
543 | */ | |
544 | if (a->check_degraded && !frozen && | |
545 | update_queue == NULL && update_queue_pending == NULL) { | |
546 | struct metadata_update *updates = NULL; | |
547 | struct mdinfo *newdev = NULL; | |
548 | struct active_array *newa; | |
549 | struct mdinfo *d; | |
550 | ||
551 | a->check_degraded = 0; | |
552 | ||
553 | /* The array may not be degraded, this is just a good time | |
554 | * to check. | |
555 | */ | |
556 | newdev = container->ss->activate_spare(a, &updates); | |
557 | if (!newdev) | |
558 | return; | |
559 | ||
560 | newa = duplicate_aa(a); | |
561 | if (!newa) | |
562 | goto out; | |
563 | /* prevent the kernel from activating the disk(s) before we | |
564 | * finish adding them | |
565 | */ | |
566 | dprintf("freezing %s\n", a->info.sys_name); | |
567 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); | |
568 | ||
569 | /* Add device to array and set offset/size/slot. | |
570 | * and open files for each newdev */ | |
571 | for (d = newdev; d ; d = d->next) { | |
572 | struct mdinfo *newd; | |
573 | ||
574 | newd = xmalloc(sizeof(*newd)); | |
575 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { | |
576 | free(newd); | |
577 | continue; | |
578 | } | |
579 | disk_init_and_add(newd, d, newa); | |
580 | } | |
581 | queue_metadata_update(updates); | |
582 | updates = NULL; | |
583 | while (update_queue_pending || update_queue) { | |
584 | check_update_queue(container); | |
585 | usleep(15*1000); | |
586 | } | |
587 | replace_array(container, a, newa); | |
588 | if (sysfs_set_str(&a->info, NULL, | |
589 | "sync_action", "recover") == 0) | |
590 | newa->prev_action = recover; | |
591 | dprintf("recovery started on %s\n", a->info.sys_name); | |
592 | out: | |
593 | while (newdev) { | |
594 | d = newdev->next; | |
595 | free(newdev); | |
596 | newdev = d; | |
597 | } | |
598 | free_updates(&updates); | |
599 | } | |
600 | ||
601 | if (a->check_reshape) { | |
602 | /* mdadm might have added some devices to the array. | |
603 | * We want to disk_init_and_add any such device to a | |
604 | * duplicate_aa and replace a with that. | |
605 | * mdstat doesn't have enough info so we sysfs_read | |
606 | * and look for new stuff. | |
607 | */ | |
608 | struct mdinfo *info, *d, *d2, *newd; | |
609 | unsigned long long array_size; | |
610 | struct active_array *newa = NULL; | |
611 | a->check_reshape = 0; | |
612 | info = sysfs_read(-1, mdstat->devnm, | |
613 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); | |
614 | if (!info) | |
615 | goto out2; | |
616 | for (d = info->devs; d; d = d->next) { | |
617 | if (d->disk.raid_disk < 0) | |
618 | continue; | |
619 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
620 | if (d2->disk.raid_disk == | |
621 | d->disk.raid_disk) | |
622 | break; | |
623 | if (d2) | |
624 | /* already have this one */ | |
625 | continue; | |
626 | if (!newa) { | |
627 | newa = duplicate_aa(a); | |
628 | if (!newa) | |
629 | break; | |
630 | } | |
631 | newd = xmalloc(sizeof(*newd)); | |
632 | disk_init_and_add(newd, d, newa); | |
633 | } | |
634 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 && | |
635 | a->info.custom_array_size > array_size*2) { | |
636 | sysfs_set_num(info, NULL, "array_size", | |
637 | a->info.custom_array_size/2); | |
638 | } | |
639 | out2: | |
640 | sysfs_free(info); | |
641 | if (newa) | |
642 | replace_array(container, a, newa); | |
643 | } | |
644 | } | |
645 | ||
646 | static int aa_ready(struct active_array *aa) | |
647 | { | |
648 | struct mdinfo *d; | |
649 | int level = aa->info.array.level; | |
650 | ||
651 | for (d = aa->info.devs; d; d = d->next) | |
652 | if (d->state_fd < 0) | |
653 | return 0; | |
654 | ||
655 | if (aa->info.state_fd < 0) | |
656 | return 0; | |
657 | ||
658 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
659 | return 0; | |
660 | ||
661 | if (!aa->container) | |
662 | return 0; | |
663 | ||
664 | return 1; | |
665 | } | |
666 | ||
667 | static void manage_new(struct mdstat_ent *mdstat, | |
668 | struct supertype *container, | |
669 | struct active_array *victim) | |
670 | { | |
671 | /* A new array has appeared in this container. | |
672 | * Hopefully it is already recorded in the metadata. | |
673 | * Check, then create the new array to report it to | |
674 | * the monitor. | |
675 | */ | |
676 | ||
677 | struct active_array *new; | |
678 | struct mdinfo *mdi, *di; | |
679 | char *inst; | |
680 | int i; | |
681 | int failed = 0; | |
682 | char buf[40]; | |
683 | ||
684 | /* check if array is ready to be monitored */ | |
685 | if (!mdstat->active || !mdstat->level) | |
686 | return; | |
687 | if (strcmp(mdstat->level, "raid0") == 0 || | |
688 | strcmp(mdstat->level, "linear") == 0) | |
689 | return; | |
690 | ||
691 | mdi = sysfs_read(-1, mdstat->devnm, | |
692 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| | |
693 | GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| | |
694 | GET_LAYOUT); | |
695 | ||
696 | if (!mdi) | |
697 | return; | |
698 | new = xcalloc(1, sizeof(*new)); | |
699 | ||
700 | strcpy(new->info.sys_name, mdstat->devnm); | |
701 | ||
702 | new->prev_state = new->curr_state = new->next_state = inactive; | |
703 | new->prev_action= new->curr_action= new->next_action= idle; | |
704 | ||
705 | new->container = container; | |
706 | ||
707 | inst = to_subarray(mdstat, container->devnm); | |
708 | ||
709 | new->info.array = mdi->array; | |
710 | new->info.component_size = mdi->component_size; | |
711 | ||
712 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
713 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
714 | ||
715 | for (di = mdi->devs; di; di = di->next) | |
716 | if (i == di->disk.raid_disk) | |
717 | break; | |
718 | ||
719 | if (disk_init_and_add(newd, di, new) != 0) { | |
720 | if (newd) | |
721 | free(newd); | |
722 | ||
723 | failed++; | |
724 | if (failed > new->info.array.failed_disks) { | |
725 | /* we cannot properly monitor without all working disks */ | |
726 | new->container = NULL; | |
727 | break; | |
728 | } | |
729 | } | |
730 | } | |
731 | ||
732 | new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action"); | |
733 | new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state"); | |
734 | new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start"); | |
735 | new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version"); | |
736 | new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed"); | |
737 | ||
738 | dprintf("inst: %s action: %d state: %d\n", inst, | |
739 | new->action_fd, new->info.state_fd); | |
740 | ||
741 | if (sigterm) | |
742 | new->info.safe_mode_delay = 1; | |
743 | else if (mdi->safe_mode_delay >= 50) | |
744 | /* Normal start, mdadm set this. */ | |
745 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
746 | else | |
747 | /* Restart, just pick a number */ | |
748 | new->info.safe_mode_delay = 5000; | |
749 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
750 | ||
751 | /* reshape_position is set by mdadm in sysfs | |
752 | * read this information for new arrays only (empty victim) | |
753 | */ | |
754 | if ((victim == NULL) && | |
755 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
756 | (strncmp(buf, "reshape", 7) == 0)) { | |
757 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
758 | &new->last_checkpoint) != 0) | |
759 | new->last_checkpoint = 0; | |
760 | else { | |
761 | int data_disks = mdi->array.raid_disks; | |
762 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
763 | data_disks--; | |
764 | if (mdi->array.level == 6) | |
765 | data_disks -= 2; | |
766 | ||
767 | new->last_checkpoint /= data_disks; | |
768 | } | |
769 | dprintf("mdmon: New monitored array is under reshape.\n" | |
770 | " Last checkpoint is: %llu\n", | |
771 | new->last_checkpoint); | |
772 | } | |
773 | ||
774 | sysfs_free(mdi); | |
775 | ||
776 | /* if everything checks out tell the metadata handler we want to | |
777 | * manage this instance | |
778 | */ | |
779 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
780 | pr_err("failed to monitor %s\n", | |
781 | mdstat->metadata_version); | |
782 | new->container = NULL; | |
783 | free_aa(new); | |
784 | } else { | |
785 | replace_array(container, victim, new); | |
786 | if (failed) { | |
787 | new->check_degraded = 1; | |
788 | manage_member(mdstat, new); | |
789 | } | |
790 | } | |
791 | } | |
792 | ||
793 | void manage(struct mdstat_ent *mdstat, struct supertype *container) | |
794 | { | |
795 | /* We have just read mdstat and need to compare it with | |
796 | * the known active arrays. | |
797 | * Arrays with the wrong metadata are ignored. | |
798 | */ | |
799 | ||
800 | for ( ; mdstat ; mdstat = mdstat->next) { | |
801 | struct active_array *a; | |
802 | if (strcmp(mdstat->devnm, container->devnm) == 0) { | |
803 | manage_container(mdstat, container); | |
804 | continue; | |
805 | } | |
806 | if (!is_container_member(mdstat, container->devnm)) | |
807 | /* Not for this array */ | |
808 | continue; | |
809 | /* Looks like a member of this container */ | |
810 | for (a = container->arrays; a; a = a->next) { | |
811 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { | |
812 | if (a->container && a->to_remove == 0) | |
813 | manage_member(mdstat, a); | |
814 | break; | |
815 | } | |
816 | } | |
817 | if (a == NULL || !a->container) | |
818 | manage_new(mdstat, container, a); | |
819 | } | |
820 | } | |
821 | ||
822 | static void handle_message(struct supertype *container, struct metadata_update *msg) | |
823 | { | |
824 | /* queue this metadata update through to the monitor */ | |
825 | ||
826 | struct metadata_update *mu; | |
827 | ||
828 | if (msg->len <= 0) | |
829 | while (update_queue_pending || update_queue) { | |
830 | check_update_queue(container); | |
831 | usleep(15*1000); | |
832 | } | |
833 | ||
834 | if (msg->len == 0) { /* ping_monitor */ | |
835 | int cnt; | |
836 | ||
837 | cnt = monitor_loop_cnt; | |
838 | if (cnt & 1) | |
839 | cnt += 2; /* wait until next pselect */ | |
840 | else | |
841 | cnt += 3; /* wait for 2 pselects */ | |
842 | wakeup_monitor(); | |
843 | ||
844 | while (monitor_loop_cnt - cnt < 0) | |
845 | usleep(10 * 1000); | |
846 | } else if (msg->len == -1) { /* ping_manager */ | |
847 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
848 | ||
849 | manage(mdstat, container); | |
850 | free_mdstat(mdstat); | |
851 | } else if (!sigterm) { | |
852 | mu = xmalloc(sizeof(*mu)); | |
853 | mu->len = msg->len; | |
854 | mu->buf = msg->buf; | |
855 | msg->buf = NULL; | |
856 | mu->space = NULL; | |
857 | mu->space_list = NULL; | |
858 | mu->next = NULL; | |
859 | if (container->ss->prepare_update) | |
860 | if (!container->ss->prepare_update(container, mu)) | |
861 | free_updates(&mu); | |
862 | queue_metadata_update(mu); | |
863 | } | |
864 | } | |
865 | ||
866 | void read_sock(struct supertype *container) | |
867 | { | |
868 | int fd; | |
869 | struct metadata_update msg; | |
870 | int terminate = 0; | |
871 | long fl; | |
872 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
873 | ||
874 | fd = accept(container->sock, NULL, NULL); | |
875 | if (fd < 0) | |
876 | return; | |
877 | ||
878 | fl = fcntl(fd, F_GETFL, 0); | |
879 | fl |= O_NONBLOCK; | |
880 | fcntl(fd, F_SETFL, fl); | |
881 | ||
882 | do { | |
883 | msg.buf = NULL; | |
884 | ||
885 | /* read and validate the message */ | |
886 | if (receive_message(fd, &msg, tmo) == 0) { | |
887 | handle_message(container, &msg); | |
888 | if (msg.len == 0) { | |
889 | /* ping reply with version */ | |
890 | msg.buf = Version; | |
891 | msg.len = strlen(Version) + 1; | |
892 | if (send_message(fd, &msg, tmo) < 0) | |
893 | terminate = 1; | |
894 | } else if (ack(fd, tmo) < 0) | |
895 | terminate = 1; | |
896 | } else | |
897 | terminate = 1; | |
898 | ||
899 | } while (!terminate); | |
900 | ||
901 | close(fd); | |
902 | } | |
903 | ||
904 | int exit_now = 0; | |
905 | int manager_ready = 0; | |
906 | void do_manager(struct supertype *container) | |
907 | { | |
908 | struct mdstat_ent *mdstat; | |
909 | sigset_t set; | |
910 | ||
911 | sigprocmask(SIG_UNBLOCK, NULL, &set); | |
912 | sigdelset(&set, SIGUSR1); | |
913 | sigdelset(&set, SIGTERM); | |
914 | ||
915 | do { | |
916 | ||
917 | if (exit_now) | |
918 | exit(0); | |
919 | ||
920 | /* Can only 'manage' things if 'monitor' is not making | |
921 | * structural changes to metadata, so need to check | |
922 | * update_queue | |
923 | */ | |
924 | if (update_queue == NULL) { | |
925 | mdstat = mdstat_read(1, 0); | |
926 | ||
927 | manage(mdstat, container); | |
928 | ||
929 | read_sock(container); | |
930 | ||
931 | free_mdstat(mdstat); | |
932 | } | |
933 | remove_old(); | |
934 | ||
935 | check_update_queue(container); | |
936 | ||
937 | manager_ready = 1; | |
938 | ||
939 | if (sigterm) | |
940 | wakeup_monitor(); | |
941 | ||
942 | if (update_queue == NULL) | |
943 | mdstat_wait_fd(container->sock, &set); | |
944 | else | |
945 | /* If an update is happening, just wait for signal */ | |
946 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
947 | } while(1); | |
948 | } |