]>
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 | } | |
119 | ||
120 | if (aa->action_fd >= 0) | |
121 | close(aa->action_fd); | |
122 | if (aa->info.state_fd >= 0) | |
123 | close(aa->info.state_fd); | |
124 | if (aa->resync_start_fd >= 0) | |
125 | close(aa->resync_start_fd); | |
126 | if (aa->metadata_fd >= 0) | |
127 | close(aa->metadata_fd); | |
128 | if (aa->sync_completed_fd >= 0) | |
129 | close(aa->sync_completed_fd); | |
130 | } | |
131 | ||
132 | static void free_aa(struct active_array *aa) | |
133 | { | |
134 | /* Note that this doesn't close fds if they are being used | |
135 | * by a clone. ->container will be set for a clone | |
136 | */ | |
137 | dprintf("%s: sys_name: %s\n", __func__, aa->info.sys_name); | |
138 | if (!aa->container) | |
139 | close_aa(aa); | |
140 | while (aa->info.devs) { | |
141 | struct mdinfo *d = aa->info.devs; | |
142 | aa->info.devs = d->next; | |
143 | free(d); | |
144 | } | |
145 | free(aa); | |
146 | } | |
147 | ||
148 | static struct active_array *duplicate_aa(struct active_array *aa) | |
149 | { | |
150 | struct active_array *newa = xmalloc(sizeof(*newa)); | |
151 | struct mdinfo **dp1, **dp2; | |
152 | ||
153 | *newa = *aa; | |
154 | newa->next = NULL; | |
155 | newa->replaces = NULL; | |
156 | newa->info.next = NULL; | |
157 | ||
158 | dp2 = &newa->info.devs; | |
159 | ||
160 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
161 | struct mdinfo *d; | |
162 | if ((*dp1)->state_fd < 0) | |
163 | continue; | |
164 | ||
165 | d = xmalloc(sizeof(*d)); | |
166 | *d = **dp1; | |
167 | *dp2 = d; | |
168 | dp2 = & d->next; | |
169 | } | |
170 | *dp2 = NULL; | |
171 | ||
172 | return newa; | |
173 | } | |
174 | ||
175 | static void wakeup_monitor(void) | |
176 | { | |
177 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ | |
178 | int pid = getpid(); | |
179 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
180 | } | |
181 | ||
182 | static void remove_old(void) | |
183 | { | |
184 | if (discard_this) { | |
185 | discard_this->next = NULL; | |
186 | free_aa(discard_this); | |
187 | if (pending_discard == discard_this) | |
188 | pending_discard = NULL; | |
189 | discard_this = NULL; | |
190 | wakeup_monitor(); | |
191 | } | |
192 | } | |
193 | ||
194 | static void replace_array(struct supertype *container, | |
195 | struct active_array *old, | |
196 | struct active_array *new) | |
197 | { | |
198 | /* To replace an array, we add it to the top of the list | |
199 | * marked with ->replaces to point to the original. | |
200 | * 'monitor' will take the original out of the list | |
201 | * and put it on 'discard_this'. We take it from there | |
202 | * and discard it. | |
203 | */ | |
204 | remove_old(); | |
205 | while (pending_discard) { | |
206 | while (discard_this == NULL) | |
207 | sleep(1); | |
208 | remove_old(); | |
209 | } | |
210 | pending_discard = old; | |
211 | new->replaces = old; | |
212 | new->next = container->arrays; | |
213 | container->arrays = new; | |
214 | wakeup_monitor(); | |
215 | } | |
216 | ||
217 | struct metadata_update *update_queue = NULL; | |
218 | struct metadata_update *update_queue_handled = NULL; | |
219 | struct metadata_update *update_queue_pending = NULL; | |
220 | ||
221 | static void free_updates(struct metadata_update **update) | |
222 | { | |
223 | while (*update) { | |
224 | struct metadata_update *this = *update; | |
225 | void **space_list = this->space_list; | |
226 | ||
227 | *update = this->next; | |
228 | free(this->buf); | |
229 | free(this->space); | |
230 | while (space_list) { | |
231 | void *space = space_list; | |
232 | space_list = *space_list; | |
233 | free(space); | |
234 | } | |
235 | free(this); | |
236 | } | |
237 | } | |
238 | ||
239 | void check_update_queue(struct supertype *container) | |
240 | { | |
241 | free_updates(&update_queue_handled); | |
242 | ||
243 | if (update_queue == NULL && | |
244 | update_queue_pending) { | |
245 | update_queue = update_queue_pending; | |
246 | update_queue_pending = NULL; | |
247 | wakeup_monitor(); | |
248 | } | |
249 | } | |
250 | ||
251 | static void queue_metadata_update(struct metadata_update *mu) | |
252 | { | |
253 | struct metadata_update **qp; | |
254 | ||
255 | qp = &update_queue_pending; | |
256 | while (*qp) | |
257 | qp = & ((*qp)->next); | |
258 | *qp = mu; | |
259 | } | |
260 | ||
261 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) | |
262 | { | |
263 | int dfd; | |
264 | char nm[20]; | |
265 | struct supertype *st2; | |
266 | struct metadata_update *update = NULL; | |
267 | struct mdinfo info; | |
268 | mdu_disk_info_t dk = { | |
269 | .number = -1, | |
270 | .major = sd->disk.major, | |
271 | .minor = sd->disk.minor, | |
272 | .raid_disk = -1, | |
273 | .state = 0, | |
274 | }; | |
275 | ||
276 | dprintf("%s: add %d:%d to container\n", | |
277 | __func__, sd->disk.major, sd->disk.minor); | |
278 | ||
279 | sd->next = st->devs; | |
280 | st->devs = sd; | |
281 | ||
282 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); | |
283 | dfd = dev_open(nm, O_RDWR); | |
284 | if (dfd < 0) | |
285 | return; | |
286 | ||
287 | /* Check the metadata and see if it is already part of this | |
288 | * array | |
289 | */ | |
290 | st2 = dup_super(st); | |
291 | if (st2->ss->load_super(st2, dfd, NULL) == 0) { | |
292 | st2->ss->getinfo_super(st2, &info, NULL); | |
293 | if (st->ss->compare_super(st, st2) == 0 && | |
294 | info.disk.raid_disk >= 0) { | |
295 | /* Looks like a good member of array. | |
296 | * Just accept it. | |
297 | * mdadm will incorporate any parts into | |
298 | * active arrays. | |
299 | */ | |
300 | st2->ss->free_super(st2); | |
301 | return; | |
302 | } | |
303 | } | |
304 | st2->ss->free_super(st2); | |
305 | ||
306 | st->update_tail = &update; | |
307 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); | |
308 | st->ss->write_init_super(st); | |
309 | queue_metadata_update(update); | |
310 | st->update_tail = NULL; | |
311 | } | |
312 | ||
313 | /* | |
314 | * Create and queue update structure about the removed disks. | |
315 | * The update is prepared by super type handler and passed to the monitor | |
316 | * thread. | |
317 | */ | |
318 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
319 | { | |
320 | struct metadata_update *update = NULL; | |
321 | mdu_disk_info_t dk = { | |
322 | .number = -1, | |
323 | .major = sd->disk.major, | |
324 | .minor = sd->disk.minor, | |
325 | .raid_disk = -1, | |
326 | .state = 0, | |
327 | }; | |
328 | dprintf("%s: remove %d:%d from container\n", | |
329 | __func__, sd->disk.major, sd->disk.minor); | |
330 | ||
331 | st->update_tail = &update; | |
332 | st->ss->remove_from_super(st, &dk); | |
333 | /* FIXME this write_init_super shouldn't be here. | |
334 | * We have it after add_to_super to write to new device, | |
335 | * but with 'remove' we don't ant to write to that device! | |
336 | */ | |
337 | st->ss->write_init_super(st); | |
338 | queue_metadata_update(update); | |
339 | st->update_tail = NULL; | |
340 | } | |
341 | ||
342 | static void manage_container(struct mdstat_ent *mdstat, | |
343 | struct supertype *container) | |
344 | { | |
345 | /* Of interest here are: | |
346 | * - if a new device has been added to the container, we | |
347 | * add it to the array ignoring any metadata on it. | |
348 | * - if a device has been removed from the container, we | |
349 | * remove it from the device list and update the metadata. | |
350 | * FIXME should we look for compatible metadata and take hints | |
351 | * about spare assignment.... probably not. | |
352 | */ | |
353 | if (mdstat->devcnt != container->devcnt) { | |
354 | struct mdinfo **cdp, *cd, *di, *mdi; | |
355 | int found; | |
356 | ||
357 | /* read /sys/block/NAME/md/dev-??/block/dev to find out | |
358 | * what is there, and compare with container->info.devs | |
359 | * To see what is removed and what is added. | |
360 | * These need to be remove from, or added to, the array | |
361 | */ | |
362 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); | |
363 | if (!mdi) { | |
364 | /* invalidate the current count so we can try again */ | |
365 | container->devcnt = -1; | |
366 | return; | |
367 | } | |
368 | ||
369 | /* check for removals */ | |
370 | for (cdp = &container->devs; *cdp; ) { | |
371 | found = 0; | |
372 | for (di = mdi->devs; di; di = di->next) | |
373 | if (di->disk.major == (*cdp)->disk.major && | |
374 | di->disk.minor == (*cdp)->disk.minor) { | |
375 | found = 1; | |
376 | break; | |
377 | } | |
378 | if (!found) { | |
379 | cd = *cdp; | |
380 | *cdp = (*cdp)->next; | |
381 | remove_disk_from_container(container, cd); | |
382 | free(cd); | |
383 | } else | |
384 | cdp = &(*cdp)->next; | |
385 | } | |
386 | ||
387 | /* check for additions */ | |
388 | for (di = mdi->devs; di; di = di->next) { | |
389 | for (cd = container->devs; cd; cd = cd->next) | |
390 | if (di->disk.major == cd->disk.major && | |
391 | di->disk.minor == cd->disk.minor) | |
392 | break; | |
393 | if (!cd) { | |
394 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
395 | ||
396 | *newd = *di; | |
397 | add_disk_to_container(container, newd); | |
398 | } | |
399 | } | |
400 | sysfs_free(mdi); | |
401 | container->devcnt = mdstat->devcnt; | |
402 | } | |
403 | } | |
404 | ||
405 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, | |
406 | struct active_array *aa) | |
407 | { | |
408 | if (!disk || !clone) | |
409 | return -1; | |
410 | ||
411 | *disk = *clone; | |
412 | disk->recovery_fd = sysfs_open(aa->info.sys_name, disk->sys_name, "recovery_start"); | |
413 | if (disk->recovery_fd < 0) | |
414 | return -1; | |
415 | disk->state_fd = sysfs_open(aa->info.sys_name, disk->sys_name, "state"); | |
416 | if (disk->state_fd < 0) { | |
417 | close(disk->recovery_fd); | |
418 | return -1; | |
419 | } | |
420 | disk->prev_state = read_dev_state(disk->state_fd); | |
421 | disk->curr_state = disk->prev_state; | |
422 | disk->next = aa->info.devs; | |
423 | aa->info.devs = disk; | |
424 | ||
425 | return 0; | |
426 | } | |
427 | ||
428 | static void manage_member(struct mdstat_ent *mdstat, | |
429 | struct active_array *a) | |
430 | { | |
431 | /* Compare mdstat info with known state of member array. | |
432 | * We do not need to look for device state changes here, that | |
433 | * is dealt with by the monitor. | |
434 | * | |
435 | * If a reshape is being requested, monitor will have noticed | |
436 | * that sync_action changed and will have set check_reshape. | |
437 | * We just need to see if new devices have appeared. All metadata | |
438 | * updates will already have been processed. | |
439 | * | |
440 | * We also want to handle degraded arrays here by | |
441 | * trying to find and assign a spare. | |
442 | * We do that whenever the monitor tells us too. | |
443 | */ | |
444 | char buf[64]; | |
445 | int frozen; | |
446 | struct supertype *container = a->container; | |
447 | unsigned long long int component_size = 0; | |
448 | ||
449 | if (container == NULL) | |
450 | /* Raced with something */ | |
451 | return; | |
452 | ||
453 | if (mdstat->active) { | |
454 | // FIXME | |
455 | a->info.array.raid_disks = mdstat->raid_disks; | |
456 | // MORE | |
457 | } | |
458 | ||
459 | if (sysfs_get_ll(&a->info, NULL, "component_size", &component_size) >= 0) | |
460 | a->info.component_size = component_size << 1; | |
461 | ||
462 | /* honor 'frozen' */ | |
463 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
464 | frozen = buf[9] == '-'; | |
465 | else | |
466 | frozen = 1; /* can't read metadata_version assume the worst */ | |
467 | ||
468 | /* If sync_action is not 'idle' then don't try recovery now */ | |
469 | if (!frozen | |
470 | && sysfs_get_str(&a->info, NULL, "sync_action", buf, sizeof(buf)) > 0 | |
471 | && strncmp(buf, "idle", 4) != 0) | |
472 | frozen = 1; | |
473 | ||
474 | if (mdstat->level) { | |
475 | int level = map_name(pers, mdstat->level); | |
476 | if (level == 0 || level == LEVEL_LINEAR) { | |
477 | a->to_remove = 1; | |
478 | wakeup_monitor(); | |
479 | return; | |
480 | } | |
481 | else if (a->info.array.level != level && level > 0) { | |
482 | struct active_array *newa = duplicate_aa(a); | |
483 | if (newa) { | |
484 | newa->info.array.level = level; | |
485 | replace_array(container, a, newa); | |
486 | a = newa; | |
487 | } | |
488 | } | |
489 | } | |
490 | ||
491 | /* we are after monitor kick, | |
492 | * so container field can be cleared - check it again | |
493 | */ | |
494 | if (a->container == NULL) | |
495 | return; | |
496 | ||
497 | if (sigterm && a->info.safe_mode_delay != 1) { | |
498 | sysfs_set_safemode(&a->info, 1); | |
499 | a->info.safe_mode_delay = 1; | |
500 | } | |
501 | ||
502 | /* We don't check the array while any update is pending, as it | |
503 | * might container a change (such as a spare assignment) which | |
504 | * could affect our decisions. | |
505 | */ | |
506 | if (a->check_degraded && !frozen && | |
507 | update_queue == NULL && update_queue_pending == NULL) { | |
508 | struct metadata_update *updates = NULL; | |
509 | struct mdinfo *newdev = NULL; | |
510 | struct active_array *newa; | |
511 | struct mdinfo *d; | |
512 | ||
513 | a->check_degraded = 0; | |
514 | ||
515 | /* The array may not be degraded, this is just a good time | |
516 | * to check. | |
517 | */ | |
518 | newdev = container->ss->activate_spare(a, &updates); | |
519 | if (!newdev) | |
520 | return; | |
521 | ||
522 | newa = duplicate_aa(a); | |
523 | if (!newa) | |
524 | goto out; | |
525 | /* prevent the kernel from activating the disk(s) before we | |
526 | * finish adding them | |
527 | */ | |
528 | dprintf("%s: freezing %s\n", __func__, a->info.sys_name); | |
529 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); | |
530 | ||
531 | /* Add device to array and set offset/size/slot. | |
532 | * and open files for each newdev */ | |
533 | for (d = newdev; d ; d = d->next) { | |
534 | struct mdinfo *newd; | |
535 | ||
536 | newd = xmalloc(sizeof(*newd)); | |
537 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { | |
538 | free(newd); | |
539 | continue; | |
540 | } | |
541 | disk_init_and_add(newd, d, newa); | |
542 | } | |
543 | queue_metadata_update(updates); | |
544 | updates = NULL; | |
545 | while (update_queue_pending || update_queue) { | |
546 | check_update_queue(container); | |
547 | usleep(15*1000); | |
548 | } | |
549 | replace_array(container, a, newa); | |
550 | if (sysfs_set_str(&a->info, NULL, "sync_action", "recover") | |
551 | == 0) | |
552 | newa->prev_action = recover; | |
553 | dprintf("%s: recovery started on %s\n", __func__, | |
554 | a->info.sys_name); | |
555 | out: | |
556 | while (newdev) { | |
557 | d = newdev->next; | |
558 | free(newdev); | |
559 | newdev = d; | |
560 | } | |
561 | free_updates(&updates); | |
562 | } | |
563 | ||
564 | if (a->check_reshape) { | |
565 | /* mdadm might have added some devices to the array. | |
566 | * We want to disk_init_and_add any such device to a | |
567 | * duplicate_aa and replace a with that. | |
568 | * mdstat doesn't have enough info so we sysfs_read | |
569 | * and look for new stuff. | |
570 | */ | |
571 | struct mdinfo *info, *d, *d2, *newd; | |
572 | unsigned long long array_size; | |
573 | struct active_array *newa = NULL; | |
574 | a->check_reshape = 0; | |
575 | info = sysfs_read(-1, mdstat->devnm, | |
576 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); | |
577 | if (!info) | |
578 | goto out2; | |
579 | for (d = info->devs; d; d = d->next) { | |
580 | if (d->disk.raid_disk < 0) | |
581 | continue; | |
582 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
583 | if (d2->disk.raid_disk == | |
584 | d->disk.raid_disk) | |
585 | break; | |
586 | if (d2) | |
587 | /* already have this one */ | |
588 | continue; | |
589 | if (!newa) { | |
590 | newa = duplicate_aa(a); | |
591 | if (!newa) | |
592 | break; | |
593 | } | |
594 | newd = xmalloc(sizeof(*newd)); | |
595 | disk_init_and_add(newd, d, newa); | |
596 | } | |
597 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 | |
598 | && a->info.custom_array_size > array_size*2) { | |
599 | sysfs_set_num(info, NULL, "array_size", | |
600 | a->info.custom_array_size/2); | |
601 | } | |
602 | out2: | |
603 | sysfs_free(info); | |
604 | if (newa) | |
605 | replace_array(container, a, newa); | |
606 | } | |
607 | } | |
608 | ||
609 | static int aa_ready(struct active_array *aa) | |
610 | { | |
611 | struct mdinfo *d; | |
612 | int level = aa->info.array.level; | |
613 | ||
614 | for (d = aa->info.devs; d; d = d->next) | |
615 | if (d->state_fd < 0) | |
616 | return 0; | |
617 | ||
618 | if (aa->info.state_fd < 0) | |
619 | return 0; | |
620 | ||
621 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
622 | return 0; | |
623 | ||
624 | if (!aa->container) | |
625 | return 0; | |
626 | ||
627 | return 1; | |
628 | } | |
629 | ||
630 | static void manage_new(struct mdstat_ent *mdstat, | |
631 | struct supertype *container, | |
632 | struct active_array *victim) | |
633 | { | |
634 | /* A new array has appeared in this container. | |
635 | * Hopefully it is already recorded in the metadata. | |
636 | * Check, then create the new array to report it to | |
637 | * the monitor. | |
638 | */ | |
639 | ||
640 | struct active_array *new; | |
641 | struct mdinfo *mdi, *di; | |
642 | char *inst; | |
643 | int i; | |
644 | int failed = 0; | |
645 | char buf[40]; | |
646 | ||
647 | /* check if array is ready to be monitored */ | |
648 | if (!mdstat->active || !mdstat->level) | |
649 | return; | |
650 | if (strcmp(mdstat->level, "raid0") == 0 || | |
651 | strcmp(mdstat->level, "linear") == 0) | |
652 | return; | |
653 | ||
654 | mdi = sysfs_read(-1, mdstat->devnm, | |
655 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| | |
656 | GET_DEGRADED|GET_SAFEMODE| | |
657 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|GET_LAYOUT); | |
658 | ||
659 | if (!mdi) | |
660 | return; | |
661 | new = xcalloc(1, sizeof(*new)); | |
662 | ||
663 | strcpy(new->info.sys_name, mdstat->devnm); | |
664 | ||
665 | new->prev_state = new->curr_state = new->next_state = inactive; | |
666 | new->prev_action= new->curr_action= new->next_action= idle; | |
667 | ||
668 | new->container = container; | |
669 | ||
670 | inst = to_subarray(mdstat, container->devnm); | |
671 | ||
672 | new->info.array = mdi->array; | |
673 | new->info.component_size = mdi->component_size; | |
674 | ||
675 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
676 | struct mdinfo *newd = xmalloc(sizeof(*newd)); | |
677 | ||
678 | for (di = mdi->devs; di; di = di->next) | |
679 | if (i == di->disk.raid_disk) | |
680 | break; | |
681 | ||
682 | if (disk_init_and_add(newd, di, new) != 0) { | |
683 | if (newd) | |
684 | free(newd); | |
685 | ||
686 | failed++; | |
687 | if (failed > new->info.array.failed_disks) { | |
688 | /* we cannot properly monitor without all working disks */ | |
689 | new->container = NULL; | |
690 | break; | |
691 | } | |
692 | } | |
693 | } | |
694 | ||
695 | new->action_fd = sysfs_open(new->info.sys_name, NULL, "sync_action"); | |
696 | new->info.state_fd = sysfs_open(new->info.sys_name, NULL, "array_state"); | |
697 | new->resync_start_fd = sysfs_open(new->info.sys_name, NULL, "resync_start"); | |
698 | new->metadata_fd = sysfs_open(new->info.sys_name, NULL, "metadata_version"); | |
699 | new->sync_completed_fd = sysfs_open(new->info.sys_name, NULL, "sync_completed"); | |
700 | dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst), | |
701 | new->action_fd, new->info.state_fd); | |
702 | ||
703 | if (sigterm) | |
704 | new->info.safe_mode_delay = 1; | |
705 | else if (mdi->safe_mode_delay >= 50) | |
706 | /* Normal start, mdadm set this. */ | |
707 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
708 | else | |
709 | /* Restart, just pick a number */ | |
710 | new->info.safe_mode_delay = 5000; | |
711 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
712 | ||
713 | /* reshape_position is set by mdadm in sysfs | |
714 | * read this information for new arrays only (empty victim) | |
715 | */ | |
716 | if ((victim == NULL) && | |
717 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
718 | (strncmp(buf, "reshape", 7) == 0)) { | |
719 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
720 | &new->last_checkpoint) != 0) | |
721 | new->last_checkpoint = 0; | |
722 | else { | |
723 | int data_disks = mdi->array.raid_disks; | |
724 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
725 | data_disks--; | |
726 | if (mdi->array.level == 6) | |
727 | data_disks -= 2; | |
728 | ||
729 | new->last_checkpoint /= data_disks; | |
730 | } | |
731 | dprintf("mdmon: New monitored array is under reshape.\n" | |
732 | " Last checkpoint is: %llu\n", | |
733 | new->last_checkpoint); | |
734 | } | |
735 | ||
736 | sysfs_free(mdi); | |
737 | ||
738 | /* if everything checks out tell the metadata handler we want to | |
739 | * manage this instance | |
740 | */ | |
741 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
742 | pr_err("failed to monitor %s\n", | |
743 | mdstat->metadata_version); | |
744 | new->container = NULL; | |
745 | free_aa(new); | |
746 | } else { | |
747 | replace_array(container, victim, new); | |
748 | if (failed) { | |
749 | new->check_degraded = 1; | |
750 | manage_member(mdstat, new); | |
751 | } | |
752 | } | |
753 | } | |
754 | ||
755 | void manage(struct mdstat_ent *mdstat, struct supertype *container) | |
756 | { | |
757 | /* We have just read mdstat and need to compare it with | |
758 | * the known active arrays. | |
759 | * Arrays with the wrong metadata are ignored. | |
760 | */ | |
761 | ||
762 | for ( ; mdstat ; mdstat = mdstat->next) { | |
763 | struct active_array *a; | |
764 | if (strcmp(mdstat->devnm, container->devnm) == 0) { | |
765 | manage_container(mdstat, container); | |
766 | continue; | |
767 | } | |
768 | if (!is_container_member(mdstat, container->devnm)) | |
769 | /* Not for this array */ | |
770 | continue; | |
771 | /* Looks like a member of this container */ | |
772 | for (a = container->arrays; a; a = a->next) { | |
773 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { | |
774 | if (a->container && a->to_remove == 0) | |
775 | manage_member(mdstat, a); | |
776 | break; | |
777 | } | |
778 | } | |
779 | if (a == NULL || !a->container) | |
780 | manage_new(mdstat, container, a); | |
781 | } | |
782 | } | |
783 | ||
784 | static void handle_message(struct supertype *container, struct metadata_update *msg) | |
785 | { | |
786 | /* queue this metadata update through to the monitor */ | |
787 | ||
788 | struct metadata_update *mu; | |
789 | ||
790 | if (msg->len <= 0) | |
791 | while (update_queue_pending || update_queue) { | |
792 | check_update_queue(container); | |
793 | usleep(15*1000); | |
794 | } | |
795 | ||
796 | if (msg->len == 0) { /* ping_monitor */ | |
797 | int cnt; | |
798 | ||
799 | cnt = monitor_loop_cnt; | |
800 | if (cnt & 1) | |
801 | cnt += 2; /* wait until next pselect */ | |
802 | else | |
803 | cnt += 3; /* wait for 2 pselects */ | |
804 | wakeup_monitor(); | |
805 | ||
806 | while (monitor_loop_cnt - cnt < 0) | |
807 | usleep(10 * 1000); | |
808 | } else if (msg->len == -1) { /* ping_manager */ | |
809 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
810 | ||
811 | manage(mdstat, container); | |
812 | free_mdstat(mdstat); | |
813 | } else if (!sigterm) { | |
814 | mu = xmalloc(sizeof(*mu)); | |
815 | mu->len = msg->len; | |
816 | mu->buf = msg->buf; | |
817 | msg->buf = NULL; | |
818 | mu->space = NULL; | |
819 | mu->space_list = NULL; | |
820 | mu->next = NULL; | |
821 | if (container->ss->prepare_update) | |
822 | container->ss->prepare_update(container, mu); | |
823 | queue_metadata_update(mu); | |
824 | } | |
825 | } | |
826 | ||
827 | void read_sock(struct supertype *container) | |
828 | { | |
829 | int fd; | |
830 | struct metadata_update msg; | |
831 | int terminate = 0; | |
832 | long fl; | |
833 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
834 | ||
835 | fd = accept(container->sock, NULL, NULL); | |
836 | if (fd < 0) | |
837 | return; | |
838 | ||
839 | fl = fcntl(fd, F_GETFL, 0); | |
840 | fl |= O_NONBLOCK; | |
841 | fcntl(fd, F_SETFL, fl); | |
842 | ||
843 | do { | |
844 | msg.buf = NULL; | |
845 | ||
846 | /* read and validate the message */ | |
847 | if (receive_message(fd, &msg, tmo) == 0) { | |
848 | handle_message(container, &msg); | |
849 | if (msg.len == 0) { | |
850 | /* ping reply with version */ | |
851 | msg.buf = Version; | |
852 | msg.len = strlen(Version) + 1; | |
853 | if (send_message(fd, &msg, tmo) < 0) | |
854 | terminate = 1; | |
855 | } else if (ack(fd, tmo) < 0) | |
856 | terminate = 1; | |
857 | } else | |
858 | terminate = 1; | |
859 | ||
860 | } while (!terminate); | |
861 | ||
862 | close(fd); | |
863 | } | |
864 | ||
865 | int exit_now = 0; | |
866 | int manager_ready = 0; | |
867 | void do_manager(struct supertype *container) | |
868 | { | |
869 | struct mdstat_ent *mdstat; | |
870 | sigset_t set; | |
871 | ||
872 | sigprocmask(SIG_UNBLOCK, NULL, &set); | |
873 | sigdelset(&set, SIGUSR1); | |
874 | sigdelset(&set, SIGTERM); | |
875 | ||
876 | do { | |
877 | ||
878 | if (exit_now) | |
879 | exit(0); | |
880 | ||
881 | /* Can only 'manage' things if 'monitor' is not making | |
882 | * structural changes to metadata, so need to check | |
883 | * update_queue | |
884 | */ | |
885 | if (update_queue == NULL) { | |
886 | mdstat = mdstat_read(1, 0); | |
887 | ||
888 | manage(mdstat, container); | |
889 | ||
890 | read_sock(container); | |
891 | ||
892 | free_mdstat(mdstat); | |
893 | } | |
894 | remove_old(); | |
895 | ||
896 | check_update_queue(container); | |
897 | ||
898 | manager_ready = 1; | |
899 | ||
900 | if (sigterm) | |
901 | wakeup_monitor(); | |
902 | ||
903 | if (update_queue == NULL) | |
904 | mdstat_wait_fd(container->sock, &set); | |
905 | else | |
906 | /* If an update is happening, just wait for signal */ | |
907 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
908 | } while(1); | |
909 | } |