]>
Commit | Line | Data |
---|---|---|
a54d5262 DW |
1 | /* |
2 | * mdmon - monitor external metadata arrays | |
3 | * | |
e736b623 N |
4 | * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de> |
5 | * Copyright (C) 2007-2009 Intel Corporation | |
a54d5262 DW |
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 | */ | |
549e9569 NB |
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 | * | |
6c3fb95c NB |
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.... | |
549e9569 NB |
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" | |
4d43913c | 107 | #include <sys/syscall.h> |
549e9569 | 108 | #include <sys/socket.h> |
1ed3f387 | 109 | #include <signal.h> |
549e9569 | 110 | |
2a0bb19e DW |
111 | static void close_aa(struct active_array *aa) |
112 | { | |
113 | struct mdinfo *d; | |
114 | ||
e1516be1 DW |
115 | for (d = aa->info.devs; d; d = d->next) { |
116 | close(d->recovery_fd); | |
2a0bb19e | 117 | close(d->state_fd); |
e1516be1 | 118 | } |
2a0bb19e DW |
119 | |
120 | close(aa->action_fd); | |
121 | close(aa->info.state_fd); | |
122 | close(aa->resync_start_fd); | |
a9d868c3 AK |
123 | close(aa->metadata_fd); |
124 | close(aa->sync_completed_fd); | |
2a0bb19e DW |
125 | } |
126 | ||
549e9569 NB |
127 | static void free_aa(struct active_array *aa) |
128 | { | |
2a0bb19e DW |
129 | /* Note that this doesn't close fds if they are being used |
130 | * by a clone. ->container will be set for a clone | |
549e9569 | 131 | */ |
4e6e574a | 132 | dprintf("%s: devnum: %d\n", __func__, aa->devnum); |
2a0bb19e DW |
133 | if (!aa->container) |
134 | close_aa(aa); | |
549e9569 NB |
135 | while (aa->info.devs) { |
136 | struct mdinfo *d = aa->info.devs; | |
137 | aa->info.devs = d->next; | |
138 | free(d); | |
139 | } | |
140 | free(aa); | |
141 | } | |
142 | ||
6c3fb95c NB |
143 | static struct active_array *duplicate_aa(struct active_array *aa) |
144 | { | |
145 | struct active_array *newa = malloc(sizeof(*newa)); | |
146 | struct mdinfo **dp1, **dp2; | |
147 | ||
148 | *newa = *aa; | |
149 | newa->next = NULL; | |
150 | newa->replaces = NULL; | |
151 | newa->info.next = NULL; | |
152 | ||
153 | dp2 = &newa->info.devs; | |
154 | ||
155 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
156 | struct mdinfo *d; | |
157 | if ((*dp1)->state_fd < 0) | |
158 | continue; | |
159 | ||
160 | d = malloc(sizeof(*d)); | |
161 | *d = **dp1; | |
162 | *dp2 = d; | |
163 | dp2 = & d->next; | |
164 | } | |
7e1432fb | 165 | *dp2 = NULL; |
6c3fb95c NB |
166 | |
167 | return newa; | |
168 | } | |
169 | ||
4d43913c | 170 | static void wakeup_monitor(void) |
2a0bb19e | 171 | { |
4d43913c NB |
172 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
173 | int pid = getpid(); | |
174 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
175 | } |
176 | ||
1ed3f387 NB |
177 | static void remove_old(void) |
178 | { | |
179 | if (discard_this) { | |
180 | discard_this->next = NULL; | |
181 | free_aa(discard_this); | |
182 | if (pending_discard == discard_this) | |
183 | pending_discard = NULL; | |
184 | discard_this = NULL; | |
48561b01 | 185 | wakeup_monitor(); |
1ed3f387 NB |
186 | } |
187 | } | |
188 | ||
549e9569 NB |
189 | static void replace_array(struct supertype *container, |
190 | struct active_array *old, | |
191 | struct active_array *new) | |
192 | { | |
193 | /* To replace an array, we add it to the top of the list | |
194 | * marked with ->replaces to point to the original. | |
195 | * 'monitor' will take the original out of the list | |
196 | * and put it on 'discard_this'. We take it from there | |
197 | * and discard it. | |
198 | */ | |
1ed3f387 | 199 | remove_old(); |
549e9569 NB |
200 | while (pending_discard) { |
201 | while (discard_this == NULL) | |
202 | sleep(1); | |
1ed3f387 | 203 | remove_old(); |
549e9569 NB |
204 | } |
205 | pending_discard = old; | |
206 | new->replaces = old; | |
207 | new->next = container->arrays; | |
208 | container->arrays = new; | |
4d43913c | 209 | wakeup_monitor(); |
549e9569 NB |
210 | } |
211 | ||
2e735d19 NB |
212 | struct metadata_update *update_queue = NULL; |
213 | struct metadata_update *update_queue_handled = NULL; | |
214 | struct metadata_update *update_queue_pending = NULL; | |
215 | ||
071cfc42 | 216 | static void free_updates(struct metadata_update **update) |
2e735d19 | 217 | { |
071cfc42 DW |
218 | while (*update) { |
219 | struct metadata_update *this = *update; | |
220 | ||
221 | *update = this->next; | |
904c1ef7 | 222 | free(this->buf); |
071cfc42 | 223 | free(this->space); |
2e735d19 NB |
224 | free(this); |
225 | } | |
071cfc42 DW |
226 | } |
227 | ||
228 | void check_update_queue(struct supertype *container) | |
229 | { | |
230 | free_updates(&update_queue_handled); | |
231 | ||
2e735d19 NB |
232 | if (update_queue == NULL && |
233 | update_queue_pending) { | |
234 | update_queue = update_queue_pending; | |
235 | update_queue_pending = NULL; | |
4d43913c | 236 | wakeup_monitor(); |
2e735d19 NB |
237 | } |
238 | } | |
239 | ||
6c3fb95c | 240 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
241 | { |
242 | struct metadata_update **qp; | |
243 | ||
244 | qp = &update_queue_pending; | |
245 | while (*qp) | |
246 | qp = & ((*qp)->next); | |
247 | *qp = mu; | |
248 | } | |
249 | ||
43dad3d6 DW |
250 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
251 | { | |
252 | int dfd; | |
253 | char nm[20]; | |
661dce36 | 254 | struct supertype *st2; |
43dad3d6 | 255 | struct metadata_update *update = NULL; |
661dce36 | 256 | struct mdinfo info; |
43dad3d6 DW |
257 | mdu_disk_info_t dk = { |
258 | .number = -1, | |
259 | .major = sd->disk.major, | |
260 | .minor = sd->disk.minor, | |
261 | .raid_disk = -1, | |
262 | .state = 0, | |
263 | }; | |
264 | ||
265 | dprintf("%s: add %d:%d to container\n", | |
266 | __func__, sd->disk.major, sd->disk.minor); | |
267 | ||
04a8ac08 DW |
268 | sd->next = st->devs; |
269 | st->devs = sd; | |
270 | ||
43dad3d6 DW |
271 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); |
272 | dfd = dev_open(nm, O_RDWR); | |
273 | if (dfd < 0) | |
274 | return; | |
275 | ||
661dce36 N |
276 | /* Check the metadata and see if it is already part of this |
277 | * array | |
278 | */ | |
279 | st2 = dup_super(st); | |
280 | if (st2->ss->load_super(st2, dfd, NULL) == 0) { | |
a5d85af7 | 281 | st2->ss->getinfo_super(st, &info, NULL); |
661dce36 N |
282 | if (st->ss->compare_super(st, st2) == 0 && |
283 | info.disk.raid_disk >= 0) { | |
284 | /* Looks like a good member of array. | |
285 | * Just accept it. | |
286 | * mdadm will incorporate any parts into | |
287 | * active arrays. | |
288 | */ | |
289 | st2->ss->free_super(st2); | |
290 | return; | |
291 | } | |
292 | } | |
293 | st2->ss->free_super(st2); | |
294 | ||
43dad3d6 DW |
295 | st->update_tail = &update; |
296 | st->ss->add_to_super(st, &dk, dfd, NULL); | |
297 | st->ss->write_init_super(st); | |
298 | queue_metadata_update(update); | |
299 | st->update_tail = NULL; | |
300 | } | |
301 | ||
1a64be56 LM |
302 | /* |
303 | * Create and queue update structure about the removed disks. | |
304 | * The update is prepared by super type handler and passed to the monitor | |
305 | * thread. | |
306 | */ | |
307 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
308 | { | |
309 | struct metadata_update *update = NULL; | |
310 | mdu_disk_info_t dk = { | |
311 | .number = -1, | |
312 | .major = sd->disk.major, | |
313 | .minor = sd->disk.minor, | |
314 | .raid_disk = -1, | |
315 | .state = 0, | |
316 | }; | |
317 | /* nothing to do if super type handler does not support | |
318 | * remove disk primitive | |
319 | */ | |
320 | if (!st->ss->remove_from_super) | |
321 | return; | |
322 | dprintf("%s: remove %d:%d from container\n", | |
323 | __func__, sd->disk.major, sd->disk.minor); | |
324 | ||
325 | st->update_tail = &update; | |
326 | st->ss->remove_from_super(st, &dk); | |
327 | st->ss->write_init_super(st); | |
328 | queue_metadata_update(update); | |
329 | st->update_tail = NULL; | |
330 | } | |
331 | ||
549e9569 NB |
332 | static void manage_container(struct mdstat_ent *mdstat, |
333 | struct supertype *container) | |
334 | { | |
1a64be56 LM |
335 | /* Of interest here are: |
336 | * - if a new device has been added to the container, we | |
337 | * add it to the array ignoring any metadata on it. | |
338 | * - if a device has been removed from the container, we | |
339 | * remove it from the device list and update the metadata. | |
549e9569 NB |
340 | * FIXME should we look for compatible metadata and take hints |
341 | * about spare assignment.... probably not. | |
549e9569 NB |
342 | */ |
343 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
344 | struct mdinfo **cdp, *cd, *di, *mdi; |
345 | int found; | |
346 | ||
549e9569 NB |
347 | /* read /sys/block/NAME/md/dev-??/block/dev to find out |
348 | * what is there, and compare with container->info.devs | |
349 | * To see what is removed and what is added. | |
350 | * These need to be remove from, or added to, the array | |
351 | */ | |
b526e52d | 352 | mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS); |
313a4a82 DW |
353 | if (!mdi) { |
354 | /* invalidate the current count so we can try again */ | |
355 | container->devcnt = -1; | |
7bc1962f | 356 | return; |
313a4a82 | 357 | } |
7bc1962f DW |
358 | |
359 | /* check for removals */ | |
360 | for (cdp = &container->devs; *cdp; ) { | |
361 | found = 0; | |
362 | for (di = mdi->devs; di; di = di->next) | |
363 | if (di->disk.major == (*cdp)->disk.major && | |
364 | di->disk.minor == (*cdp)->disk.minor) { | |
365 | found = 1; | |
366 | break; | |
367 | } | |
368 | if (!found) { | |
369 | cd = *cdp; | |
370 | *cdp = (*cdp)->next; | |
1a64be56 | 371 | remove_disk_from_container(container, cd); |
7bc1962f DW |
372 | free(cd); |
373 | } else | |
374 | cdp = &(*cdp)->next; | |
375 | } | |
43dad3d6 DW |
376 | |
377 | /* check for additions */ | |
378 | for (di = mdi->devs; di; di = di->next) { | |
379 | for (cd = container->devs; cd; cd = cd->next) | |
380 | if (di->disk.major == cd->disk.major && | |
381 | di->disk.minor == cd->disk.minor) | |
382 | break; | |
04a8ac08 DW |
383 | if (!cd) { |
384 | struct mdinfo *newd = malloc(sizeof(*newd)); | |
385 | ||
386 | if (!newd) { | |
387 | container->devcnt = -1; | |
388 | continue; | |
389 | } | |
390 | *newd = *di; | |
391 | add_disk_to_container(container, newd); | |
392 | } | |
43dad3d6 | 393 | } |
7bc1962f | 394 | sysfs_free(mdi); |
549e9569 NB |
395 | container->devcnt = mdstat->devcnt; |
396 | } | |
397 | } | |
398 | ||
63b4aae3 DW |
399 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, |
400 | struct active_array *aa) | |
401 | { | |
402 | if (!disk || !clone) | |
403 | return -1; | |
404 | ||
405 | *disk = *clone; | |
406 | disk->recovery_fd = sysfs_open(aa->devnum, disk->sys_name, "recovery_start"); | |
407 | disk->state_fd = sysfs_open(aa->devnum, disk->sys_name, "state"); | |
408 | disk->prev_state = read_dev_state(disk->state_fd); | |
409 | disk->curr_state = disk->prev_state; | |
410 | disk->next = aa->info.devs; | |
411 | aa->info.devs = disk; | |
412 | ||
413 | return 0; | |
414 | } | |
415 | ||
549e9569 NB |
416 | static void manage_member(struct mdstat_ent *mdstat, |
417 | struct active_array *a) | |
418 | { | |
419 | /* Compare mdstat info with known state of member array. | |
420 | * We do not need to look for device state changes here, that | |
421 | * is dealt with by the monitor. | |
422 | * | |
423 | * We just look for changes which suggest that a reshape is | |
424 | * being requested. | |
425 | * Unfortunately decreases in raid_disks don't show up in | |
426 | * mdstat until the reshape completes FIXME. | |
6c3fb95c NB |
427 | * |
428 | * Actually, we also want to handle degraded arrays here by | |
429 | * trying to find and assign a spare. | |
430 | * We do that whenever the monitor tells us too. | |
549e9569 | 431 | */ |
bc77ed53 DW |
432 | char buf[64]; |
433 | int frozen; | |
434 | ||
549e9569 NB |
435 | // FIXME |
436 | a->info.array.raid_disks = mdstat->raid_disks; | |
549e9569 NB |
437 | // MORE |
438 | ||
bc77ed53 DW |
439 | /* honor 'frozen' */ |
440 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
441 | frozen = buf[9] == '-'; | |
442 | else | |
443 | frozen = 1; /* can't read metadata_version assume the worst */ | |
444 | ||
445 | if (a->check_degraded && !frozen) { | |
6c3fb95c | 446 | struct metadata_update *updates = NULL; |
071cfc42 | 447 | struct mdinfo *newdev = NULL; |
6c3fb95c | 448 | struct active_array *newa; |
071cfc42 | 449 | struct mdinfo *d; |
3c00ffbe | 450 | |
6c3fb95c NB |
451 | a->check_degraded = 0; |
452 | ||
453 | /* The array may not be degraded, this is just a good time | |
454 | * to check. | |
455 | */ | |
456 | newdev = a->container->ss->activate_spare(a, &updates); | |
071cfc42 DW |
457 | if (!newdev) |
458 | return; | |
459 | ||
460 | newa = duplicate_aa(a); | |
461 | if (!newa) | |
462 | goto out; | |
463 | /* Cool, we can add a device or several. */ | |
464 | ||
465 | /* Add device to array and set offset/size/slot. | |
466 | * and open files for each newdev */ | |
467 | for (d = newdev; d ; d = d->next) { | |
468 | struct mdinfo *newd; | |
469 | ||
470 | newd = malloc(sizeof(*newd)); | |
471 | if (!newd) | |
472 | continue; | |
2904b26f | 473 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
474 | free(newd); |
475 | continue; | |
6c3fb95c | 476 | } |
63b4aae3 | 477 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
478 | } |
479 | queue_metadata_update(updates); | |
480 | updates = NULL; | |
481 | replace_array(a->container, a, newa); | |
482 | sysfs_set_str(&a->info, NULL, "sync_action", "recover"); | |
483 | out: | |
484 | while (newdev) { | |
485 | d = newdev->next; | |
486 | free(newdev); | |
487 | newdev = d; | |
6c3fb95c | 488 | } |
071cfc42 | 489 | free_updates(&updates); |
6c3fb95c | 490 | } |
549e9569 NB |
491 | } |
492 | ||
836759d5 DW |
493 | static int aa_ready(struct active_array *aa) |
494 | { | |
495 | struct mdinfo *d; | |
496 | int level = aa->info.array.level; | |
497 | ||
498 | for (d = aa->info.devs; d; d = d->next) | |
499 | if (d->state_fd < 0) | |
500 | return 0; | |
501 | ||
502 | if (aa->info.state_fd < 0) | |
503 | return 0; | |
504 | ||
505 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
506 | return 0; | |
507 | ||
508 | if (!aa->container) | |
509 | return 0; | |
510 | ||
511 | return 1; | |
512 | } | |
513 | ||
549e9569 | 514 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
515 | struct supertype *container, |
516 | struct active_array *victim) | |
549e9569 NB |
517 | { |
518 | /* A new array has appeared in this container. | |
519 | * Hopefully it is already recorded in the metadata. | |
520 | * Check, then create the new array to report it to | |
521 | * the monitor. | |
522 | */ | |
523 | ||
524 | struct active_array *new; | |
525 | struct mdinfo *mdi, *di; | |
cba0191b | 526 | char *inst; |
549e9569 | 527 | int i; |
f1d26766 | 528 | int failed = 0; |
549e9569 | 529 | |
836759d5 DW |
530 | /* check if array is ready to be monitored */ |
531 | if (!mdstat->active) | |
532 | return; | |
533 | ||
534 | mdi = sysfs_read(-1, mdstat->devnum, | |
535 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| | |
f1d26766 | 536 | GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
836759d5 | 537 | |
549e9569 NB |
538 | new = malloc(sizeof(*new)); |
539 | ||
836759d5 DW |
540 | if (!new || !mdi) { |
541 | if (mdi) | |
542 | sysfs_free(mdi); | |
543 | if (new) | |
544 | free(new); | |
545 | return; | |
546 | } | |
d52690ac NB |
547 | memset(new, 0, sizeof(*new)); |
548 | ||
549e9569 | 549 | new->devnum = mdstat->devnum; |
7e1432fb | 550 | strcpy(new->info.sys_name, devnum2devname(new->devnum)); |
549e9569 NB |
551 | |
552 | new->prev_state = new->curr_state = new->next_state = inactive; | |
553 | new->prev_action= new->curr_action= new->next_action= idle; | |
554 | ||
555 | new->container = container; | |
556 | ||
e5408a32 | 557 | inst = to_subarray(mdstat, container->devname); |
549e9569 | 558 | |
549e9569 | 559 | new->info.array = mdi->array; |
272bcc48 | 560 | new->info.component_size = mdi->component_size; |
549e9569 NB |
561 | |
562 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
563 | struct mdinfo *newd = malloc(sizeof(*newd)); | |
564 | ||
565 | for (di = mdi->devs; di; di = di->next) | |
566 | if (i == di->disk.raid_disk) | |
567 | break; | |
568 | ||
63b4aae3 | 569 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
570 | if (newd) |
571 | free(newd); | |
572 | ||
f1d26766 | 573 | failed++; |
7da80e6f DW |
574 | if (failed > new->info.array.failed_disks) { |
575 | /* we cannot properly monitor without all working disks */ | |
576 | new->container = NULL; | |
577 | break; | |
578 | } | |
549e9569 | 579 | } |
549e9569 | 580 | } |
836759d5 | 581 | |
549e9569 NB |
582 | new->action_fd = sysfs_open(new->devnum, NULL, "sync_action"); |
583 | new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state"); | |
c052ba30 | 584 | new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start"); |
e9dd1598 | 585 | new->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version"); |
484240d8 | 586 | new->sync_completed_fd = sysfs_open(new->devnum, NULL, "sync_completed"); |
4e6e574a DW |
587 | dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst), |
588 | new->action_fd, new->info.state_fd); | |
549e9569 | 589 | |
4fa5aef9 | 590 | sysfs_free(mdi); |
836759d5 DW |
591 | |
592 | /* if everything checks out tell the metadata handler we want to | |
593 | * manage this instance | |
594 | */ | |
595 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
596 | fprintf(stderr, "mdmon: failed to monitor %s\n", | |
597 | mdstat->metadata_version); | |
549e9569 | 598 | new->container = NULL; |
836759d5 | 599 | free_aa(new); |
93f7caca | 600 | } else { |
2a0bb19e | 601 | replace_array(container, victim, new); |
93f7caca DW |
602 | if (failed) { |
603 | new->check_degraded = 1; | |
604 | manage_member(mdstat, new); | |
605 | } | |
606 | } | |
549e9569 NB |
607 | } |
608 | ||
5d19760d | 609 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
610 | { |
611 | /* We have just read mdstat and need to compare it with | |
612 | * the known active arrays. | |
613 | * Arrays with the wrong metadata are ignored. | |
614 | */ | |
615 | ||
616 | for ( ; mdstat ; mdstat = mdstat->next) { | |
617 | struct active_array *a; | |
618 | if (mdstat->devnum == container->devnum) { | |
619 | manage_container(mdstat, container); | |
620 | continue; | |
621 | } | |
883a6142 | 622 | if (!is_container_member(mdstat, container->devname)) |
549e9569 NB |
623 | /* Not for this array */ |
624 | continue; | |
625 | /* Looks like a member of this container */ | |
5d19760d | 626 | for (a = container->arrays; a; a = a->next) { |
549e9569 NB |
627 | if (mdstat->devnum == a->devnum) { |
628 | if (a->container) | |
629 | manage_member(mdstat, a); | |
630 | break; | |
631 | } | |
632 | } | |
2a0bb19e DW |
633 | if (a == NULL || !a->container) |
634 | manage_new(mdstat, container, a); | |
549e9569 NB |
635 | } |
636 | } | |
637 | ||
edd8d13c | 638 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 639 | { |
edd8d13c NB |
640 | /* queue this metadata update through to the monitor */ |
641 | ||
642 | struct metadata_update *mu; | |
643 | ||
313a4a82 | 644 | if (msg->len <= 0) |
3c00ffbe N |
645 | while (update_queue_pending || update_queue) { |
646 | check_update_queue(container); | |
647 | usleep(15*1000); | |
648 | } | |
649 | ||
313a4a82 DW |
650 | if (msg->len == 0) { /* ping_monitor */ |
651 | int cnt; | |
652 | ||
3c00ffbe | 653 | cnt = monitor_loop_cnt; |
1eb252b8 N |
654 | if (cnt & 1) |
655 | cnt += 2; /* wait until next pselect */ | |
656 | else | |
657 | cnt += 3; /* wait for 2 pselects */ | |
658 | wakeup_monitor(); | |
3c00ffbe | 659 | |
1eb252b8 N |
660 | while (monitor_loop_cnt - cnt < 0) |
661 | usleep(10 * 1000); | |
313a4a82 DW |
662 | } else if (msg->len == -1) { /* ping_manager */ |
663 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
664 | ||
665 | manage(mdstat, container); | |
666 | free_mdstat(mdstat); | |
6144ed44 | 667 | } else if (!sigterm) { |
edd8d13c NB |
668 | mu = malloc(sizeof(*mu)); |
669 | mu->len = msg->len; | |
670 | mu->buf = msg->buf; | |
671 | msg->buf = NULL; | |
672 | mu->space = NULL; | |
673 | mu->next = NULL; | |
674 | if (container->ss->prepare_update) | |
675 | container->ss->prepare_update(container, mu); | |
676 | queue_metadata_update(mu); | |
677 | } | |
3e70c845 DW |
678 | } |
679 | ||
680 | void read_sock(struct supertype *container) | |
549e9569 NB |
681 | { |
682 | int fd; | |
bfa44e2e | 683 | struct metadata_update msg; |
b109d928 DW |
684 | int terminate = 0; |
685 | long fl; | |
686 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 687 | |
3e70c845 | 688 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
689 | if (fd < 0) |
690 | return; | |
b109d928 DW |
691 | |
692 | fl = fcntl(fd, F_GETFL, 0); | |
693 | fl |= O_NONBLOCK; | |
694 | fcntl(fd, F_SETFL, fl); | |
695 | ||
696 | do { | |
697 | msg.buf = NULL; | |
698 | ||
699 | /* read and validate the message */ | |
700 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e | 701 | handle_message(container, &msg); |
bc77ed53 DW |
702 | if (msg.len == 0) { |
703 | /* ping reply with version */ | |
704 | msg.buf = Version; | |
705 | msg.len = strlen(Version) + 1; | |
706 | if (send_message(fd, &msg, tmo) < 0) | |
707 | terminate = 1; | |
708 | } else if (ack(fd, tmo) < 0) | |
bfa44e2e NB |
709 | terminate = 1; |
710 | } else | |
b109d928 | 711 | terminate = 1; |
b109d928 | 712 | |
b109d928 DW |
713 | } while (!terminate); |
714 | ||
549e9569 NB |
715 | close(fd); |
716 | } | |
1ed3f387 | 717 | |
e0d6609f NB |
718 | int exit_now = 0; |
719 | int manager_ready = 0; | |
549e9569 NB |
720 | void do_manager(struct supertype *container) |
721 | { | |
722 | struct mdstat_ent *mdstat; | |
4d43913c | 723 | sigset_t set; |
1ed3f387 | 724 | |
4d43913c NB |
725 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
726 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 727 | sigdelset(&set, SIGTERM); |
549e9569 NB |
728 | |
729 | do { | |
1ed3f387 | 730 | |
e0d6609f NB |
731 | if (exit_now) |
732 | exit(0); | |
733 | ||
3c00ffbe N |
734 | /* Can only 'manage' things if 'monitor' is not making |
735 | * structural changes to metadata, so need to check | |
736 | * update_queue | |
737 | */ | |
738 | if (update_queue == NULL) { | |
739 | mdstat = mdstat_read(1, 0); | |
549e9569 | 740 | |
3c00ffbe | 741 | manage(mdstat, container); |
549e9569 | 742 | |
3c00ffbe | 743 | read_sock(container); |
4fa5aef9 | 744 | |
3c00ffbe N |
745 | free_mdstat(mdstat); |
746 | } | |
1ed3f387 NB |
747 | remove_old(); |
748 | ||
2e735d19 NB |
749 | check_update_queue(container); |
750 | ||
e0d6609f | 751 | manager_ready = 1; |
4d43913c | 752 | |
6144ed44 DW |
753 | if (sigterm) |
754 | wakeup_monitor(); | |
755 | ||
5d4d1b26 | 756 | if (update_queue == NULL) |
58a4ba2a | 757 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 758 | else |
3c00ffbe N |
759 | /* If an update is happening, just wait for signal */ |
760 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
761 | } while(1); |
762 | } |