]>
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); | |
2a0bb19e DW |
123 | } |
124 | ||
549e9569 NB |
125 | static void free_aa(struct active_array *aa) |
126 | { | |
2a0bb19e DW |
127 | /* Note that this doesn't close fds if they are being used |
128 | * by a clone. ->container will be set for a clone | |
549e9569 | 129 | */ |
4e6e574a | 130 | dprintf("%s: devnum: %d\n", __func__, aa->devnum); |
2a0bb19e DW |
131 | if (!aa->container) |
132 | close_aa(aa); | |
549e9569 NB |
133 | while (aa->info.devs) { |
134 | struct mdinfo *d = aa->info.devs; | |
135 | aa->info.devs = d->next; | |
136 | free(d); | |
137 | } | |
138 | free(aa); | |
139 | } | |
140 | ||
6c3fb95c NB |
141 | static struct active_array *duplicate_aa(struct active_array *aa) |
142 | { | |
143 | struct active_array *newa = malloc(sizeof(*newa)); | |
144 | struct mdinfo **dp1, **dp2; | |
145 | ||
146 | *newa = *aa; | |
147 | newa->next = NULL; | |
148 | newa->replaces = NULL; | |
149 | newa->info.next = NULL; | |
150 | ||
151 | dp2 = &newa->info.devs; | |
152 | ||
153 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
154 | struct mdinfo *d; | |
155 | if ((*dp1)->state_fd < 0) | |
156 | continue; | |
157 | ||
158 | d = malloc(sizeof(*d)); | |
159 | *d = **dp1; | |
160 | *dp2 = d; | |
161 | dp2 = & d->next; | |
162 | } | |
7e1432fb | 163 | *dp2 = NULL; |
6c3fb95c NB |
164 | |
165 | return newa; | |
166 | } | |
167 | ||
4d43913c | 168 | static void wakeup_monitor(void) |
2a0bb19e | 169 | { |
4d43913c NB |
170 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
171 | int pid = getpid(); | |
172 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
173 | } |
174 | ||
1ed3f387 NB |
175 | static void remove_old(void) |
176 | { | |
177 | if (discard_this) { | |
178 | discard_this->next = NULL; | |
179 | free_aa(discard_this); | |
180 | if (pending_discard == discard_this) | |
181 | pending_discard = NULL; | |
182 | discard_this = NULL; | |
48561b01 | 183 | wakeup_monitor(); |
1ed3f387 NB |
184 | } |
185 | } | |
186 | ||
549e9569 NB |
187 | static void replace_array(struct supertype *container, |
188 | struct active_array *old, | |
189 | struct active_array *new) | |
190 | { | |
191 | /* To replace an array, we add it to the top of the list | |
192 | * marked with ->replaces to point to the original. | |
193 | * 'monitor' will take the original out of the list | |
194 | * and put it on 'discard_this'. We take it from there | |
195 | * and discard it. | |
196 | */ | |
1ed3f387 | 197 | remove_old(); |
549e9569 NB |
198 | while (pending_discard) { |
199 | while (discard_this == NULL) | |
200 | sleep(1); | |
1ed3f387 | 201 | remove_old(); |
549e9569 NB |
202 | } |
203 | pending_discard = old; | |
204 | new->replaces = old; | |
205 | new->next = container->arrays; | |
206 | container->arrays = new; | |
4d43913c | 207 | wakeup_monitor(); |
549e9569 NB |
208 | } |
209 | ||
2e735d19 NB |
210 | struct metadata_update *update_queue = NULL; |
211 | struct metadata_update *update_queue_handled = NULL; | |
212 | struct metadata_update *update_queue_pending = NULL; | |
213 | ||
071cfc42 | 214 | static void free_updates(struct metadata_update **update) |
2e735d19 | 215 | { |
071cfc42 DW |
216 | while (*update) { |
217 | struct metadata_update *this = *update; | |
218 | ||
219 | *update = this->next; | |
904c1ef7 | 220 | free(this->buf); |
071cfc42 | 221 | free(this->space); |
2e735d19 NB |
222 | free(this); |
223 | } | |
071cfc42 DW |
224 | } |
225 | ||
226 | void check_update_queue(struct supertype *container) | |
227 | { | |
228 | free_updates(&update_queue_handled); | |
229 | ||
2e735d19 NB |
230 | if (update_queue == NULL && |
231 | update_queue_pending) { | |
232 | update_queue = update_queue_pending; | |
233 | update_queue_pending = NULL; | |
4d43913c | 234 | wakeup_monitor(); |
2e735d19 NB |
235 | } |
236 | } | |
237 | ||
6c3fb95c | 238 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
239 | { |
240 | struct metadata_update **qp; | |
241 | ||
242 | qp = &update_queue_pending; | |
243 | while (*qp) | |
244 | qp = & ((*qp)->next); | |
245 | *qp = mu; | |
246 | } | |
247 | ||
43dad3d6 DW |
248 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
249 | { | |
250 | int dfd; | |
251 | char nm[20]; | |
661dce36 | 252 | struct supertype *st2; |
43dad3d6 | 253 | struct metadata_update *update = NULL; |
661dce36 | 254 | struct mdinfo info; |
43dad3d6 DW |
255 | mdu_disk_info_t dk = { |
256 | .number = -1, | |
257 | .major = sd->disk.major, | |
258 | .minor = sd->disk.minor, | |
259 | .raid_disk = -1, | |
260 | .state = 0, | |
261 | }; | |
262 | ||
263 | dprintf("%s: add %d:%d to container\n", | |
264 | __func__, sd->disk.major, sd->disk.minor); | |
265 | ||
04a8ac08 DW |
266 | sd->next = st->devs; |
267 | st->devs = sd; | |
268 | ||
43dad3d6 DW |
269 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); |
270 | dfd = dev_open(nm, O_RDWR); | |
271 | if (dfd < 0) | |
272 | return; | |
273 | ||
661dce36 N |
274 | /* Check the metadata and see if it is already part of this |
275 | * array | |
276 | */ | |
277 | st2 = dup_super(st); | |
278 | if (st2->ss->load_super(st2, dfd, NULL) == 0) { | |
279 | st2->ss->getinfo_super(st, &info); | |
280 | if (st->ss->compare_super(st, st2) == 0 && | |
281 | info.disk.raid_disk >= 0) { | |
282 | /* Looks like a good member of array. | |
283 | * Just accept it. | |
284 | * mdadm will incorporate any parts into | |
285 | * active arrays. | |
286 | */ | |
287 | st2->ss->free_super(st2); | |
288 | return; | |
289 | } | |
290 | } | |
291 | st2->ss->free_super(st2); | |
292 | ||
43dad3d6 DW |
293 | st->update_tail = &update; |
294 | st->ss->add_to_super(st, &dk, dfd, NULL); | |
295 | st->ss->write_init_super(st); | |
296 | queue_metadata_update(update); | |
297 | st->update_tail = NULL; | |
298 | } | |
299 | ||
258c3e92 LM |
300 | /* |
301 | * Create and queue update structure about the removed disks. | |
302 | * The update is prepared by super type handler and passed to the monitor | |
303 | * thread. | |
304 | */ | |
305 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
306 | { | |
307 | struct metadata_update *update = NULL; | |
308 | mdu_disk_info_t dk = { | |
309 | .number = -1, | |
310 | .major = sd->disk.major, | |
311 | .minor = sd->disk.minor, | |
312 | .raid_disk = -1, | |
313 | .state = 0, | |
314 | }; | |
258c3e92 LM |
315 | dprintf("%s: remove %d:%d from container\n", |
316 | __func__, sd->disk.major, sd->disk.minor); | |
317 | ||
318 | st->update_tail = &update; | |
319 | st->ss->remove_from_super(st, &dk); | |
b0edee6e N |
320 | /* FIXME this write_init_super shouldn't be here. |
321 | * We have it after add_to_super to write to new device, | |
322 | * but with 'remove' we don't ant to write to that device! | |
323 | */ | |
258c3e92 LM |
324 | st->ss->write_init_super(st); |
325 | queue_metadata_update(update); | |
326 | st->update_tail = NULL; | |
327 | } | |
328 | ||
549e9569 NB |
329 | static void manage_container(struct mdstat_ent *mdstat, |
330 | struct supertype *container) | |
331 | { | |
258c3e92 LM |
332 | /* Of interest here are: |
333 | * - if a new device has been added to the container, we | |
334 | * add it to the array ignoring any metadata on it. | |
335 | * - if a device has been removed from the container, we | |
336 | * remove it from the device list and update the metadata. | |
549e9569 NB |
337 | * FIXME should we look for compatible metadata and take hints |
338 | * about spare assignment.... probably not. | |
549e9569 NB |
339 | */ |
340 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
341 | struct mdinfo **cdp, *cd, *di, *mdi; |
342 | int found; | |
343 | ||
549e9569 NB |
344 | /* read /sys/block/NAME/md/dev-??/block/dev to find out |
345 | * what is there, and compare with container->info.devs | |
346 | * To see what is removed and what is added. | |
347 | * These need to be remove from, or added to, the array | |
348 | */ | |
b526e52d | 349 | mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS); |
313a4a82 DW |
350 | if (!mdi) { |
351 | /* invalidate the current count so we can try again */ | |
352 | container->devcnt = -1; | |
7bc1962f | 353 | return; |
313a4a82 | 354 | } |
7bc1962f DW |
355 | |
356 | /* check for removals */ | |
357 | for (cdp = &container->devs; *cdp; ) { | |
358 | found = 0; | |
359 | for (di = mdi->devs; di; di = di->next) | |
360 | if (di->disk.major == (*cdp)->disk.major && | |
361 | di->disk.minor == (*cdp)->disk.minor) { | |
362 | found = 1; | |
363 | break; | |
364 | } | |
365 | if (!found) { | |
366 | cd = *cdp; | |
367 | *cdp = (*cdp)->next; | |
258c3e92 | 368 | remove_disk_from_container(container, cd); |
7bc1962f DW |
369 | free(cd); |
370 | } else | |
371 | cdp = &(*cdp)->next; | |
372 | } | |
43dad3d6 DW |
373 | |
374 | /* check for additions */ | |
375 | for (di = mdi->devs; di; di = di->next) { | |
376 | for (cd = container->devs; cd; cd = cd->next) | |
377 | if (di->disk.major == cd->disk.major && | |
378 | di->disk.minor == cd->disk.minor) | |
379 | break; | |
04a8ac08 DW |
380 | if (!cd) { |
381 | struct mdinfo *newd = malloc(sizeof(*newd)); | |
382 | ||
383 | if (!newd) { | |
384 | container->devcnt = -1; | |
385 | continue; | |
386 | } | |
387 | *newd = *di; | |
388 | add_disk_to_container(container, newd); | |
389 | } | |
43dad3d6 | 390 | } |
7bc1962f | 391 | sysfs_free(mdi); |
549e9569 NB |
392 | container->devcnt = mdstat->devcnt; |
393 | } | |
394 | } | |
395 | ||
63b4aae3 DW |
396 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, |
397 | struct active_array *aa) | |
398 | { | |
399 | if (!disk || !clone) | |
400 | return -1; | |
401 | ||
402 | *disk = *clone; | |
403 | disk->recovery_fd = sysfs_open(aa->devnum, disk->sys_name, "recovery_start"); | |
404 | disk->state_fd = sysfs_open(aa->devnum, disk->sys_name, "state"); | |
405 | disk->prev_state = read_dev_state(disk->state_fd); | |
406 | disk->curr_state = disk->prev_state; | |
407 | disk->next = aa->info.devs; | |
408 | aa->info.devs = disk; | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
549e9569 NB |
413 | static void manage_member(struct mdstat_ent *mdstat, |
414 | struct active_array *a) | |
415 | { | |
416 | /* Compare mdstat info with known state of member array. | |
417 | * We do not need to look for device state changes here, that | |
418 | * is dealt with by the monitor. | |
419 | * | |
420 | * We just look for changes which suggest that a reshape is | |
421 | * being requested. | |
422 | * Unfortunately decreases in raid_disks don't show up in | |
423 | * mdstat until the reshape completes FIXME. | |
6c3fb95c NB |
424 | * |
425 | * Actually, we also want to handle degraded arrays here by | |
426 | * trying to find and assign a spare. | |
427 | * We do that whenever the monitor tells us too. | |
549e9569 NB |
428 | */ |
429 | // FIXME | |
430 | a->info.array.raid_disks = mdstat->raid_disks; | |
549e9569 NB |
431 | // MORE |
432 | ||
0c4f6e37 N |
433 | /* We don't check the array while any update is pending, as it |
434 | * might container a change (such as a spare assignment) which | |
435 | * could affect our decisions. | |
436 | */ | |
437 | if (a->check_degraded && | |
438 | update_queue == NULL && update_queue_pending == NULL) { | |
6c3fb95c | 439 | struct metadata_update *updates = NULL; |
071cfc42 | 440 | struct mdinfo *newdev = NULL; |
6c3fb95c | 441 | struct active_array *newa; |
071cfc42 | 442 | struct mdinfo *d; |
3c00ffbe | 443 | |
6c3fb95c NB |
444 | a->check_degraded = 0; |
445 | ||
446 | /* The array may not be degraded, this is just a good time | |
447 | * to check. | |
448 | */ | |
449 | newdev = a->container->ss->activate_spare(a, &updates); | |
071cfc42 DW |
450 | if (!newdev) |
451 | return; | |
452 | ||
453 | newa = duplicate_aa(a); | |
454 | if (!newa) | |
455 | goto out; | |
456 | /* Cool, we can add a device or several. */ | |
457 | ||
458 | /* Add device to array and set offset/size/slot. | |
459 | * and open files for each newdev */ | |
460 | for (d = newdev; d ; d = d->next) { | |
461 | struct mdinfo *newd; | |
462 | ||
463 | newd = malloc(sizeof(*newd)); | |
464 | if (!newd) | |
465 | continue; | |
2904b26f | 466 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
467 | free(newd); |
468 | continue; | |
6c3fb95c | 469 | } |
63b4aae3 | 470 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
471 | } |
472 | queue_metadata_update(updates); | |
473 | updates = NULL; | |
474 | replace_array(a->container, a, newa); | |
475 | sysfs_set_str(&a->info, NULL, "sync_action", "recover"); | |
476 | out: | |
477 | while (newdev) { | |
478 | d = newdev->next; | |
479 | free(newdev); | |
480 | newdev = d; | |
6c3fb95c | 481 | } |
071cfc42 | 482 | free_updates(&updates); |
6c3fb95c | 483 | } |
549e9569 NB |
484 | } |
485 | ||
836759d5 DW |
486 | static int aa_ready(struct active_array *aa) |
487 | { | |
488 | struct mdinfo *d; | |
489 | int level = aa->info.array.level; | |
490 | ||
491 | for (d = aa->info.devs; d; d = d->next) | |
492 | if (d->state_fd < 0) | |
493 | return 0; | |
494 | ||
495 | if (aa->info.state_fd < 0) | |
496 | return 0; | |
497 | ||
498 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
499 | return 0; | |
500 | ||
501 | if (!aa->container) | |
502 | return 0; | |
503 | ||
504 | return 1; | |
505 | } | |
506 | ||
549e9569 | 507 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
508 | struct supertype *container, |
509 | struct active_array *victim) | |
549e9569 NB |
510 | { |
511 | /* A new array has appeared in this container. | |
512 | * Hopefully it is already recorded in the metadata. | |
513 | * Check, then create the new array to report it to | |
514 | * the monitor. | |
515 | */ | |
516 | ||
517 | struct active_array *new; | |
518 | struct mdinfo *mdi, *di; | |
cba0191b | 519 | char *inst; |
549e9569 | 520 | int i; |
f1d26766 | 521 | int failed = 0; |
549e9569 | 522 | |
836759d5 DW |
523 | /* check if array is ready to be monitored */ |
524 | if (!mdstat->active) | |
525 | return; | |
526 | ||
527 | mdi = sysfs_read(-1, mdstat->devnum, | |
528 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| | |
f1d26766 | 529 | GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
836759d5 | 530 | |
549e9569 NB |
531 | new = malloc(sizeof(*new)); |
532 | ||
836759d5 DW |
533 | if (!new || !mdi) { |
534 | if (mdi) | |
535 | sysfs_free(mdi); | |
536 | if (new) | |
537 | free(new); | |
538 | return; | |
539 | } | |
d52690ac NB |
540 | memset(new, 0, sizeof(*new)); |
541 | ||
549e9569 | 542 | new->devnum = mdstat->devnum; |
7e1432fb | 543 | strcpy(new->info.sys_name, devnum2devname(new->devnum)); |
549e9569 NB |
544 | |
545 | new->prev_state = new->curr_state = new->next_state = inactive; | |
546 | new->prev_action= new->curr_action= new->next_action= idle; | |
547 | ||
548 | new->container = container; | |
549 | ||
cba0191b | 550 | inst = &mdstat->metadata_version[10+strlen(container->devname)+1]; |
549e9569 | 551 | |
549e9569 | 552 | new->info.array = mdi->array; |
272bcc48 | 553 | new->info.component_size = mdi->component_size; |
549e9569 NB |
554 | |
555 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
556 | struct mdinfo *newd = malloc(sizeof(*newd)); | |
557 | ||
558 | for (di = mdi->devs; di; di = di->next) | |
559 | if (i == di->disk.raid_disk) | |
560 | break; | |
561 | ||
63b4aae3 | 562 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
563 | if (newd) |
564 | free(newd); | |
565 | ||
f1d26766 | 566 | failed++; |
7da80e6f DW |
567 | if (failed > new->info.array.failed_disks) { |
568 | /* we cannot properly monitor without all working disks */ | |
569 | new->container = NULL; | |
570 | break; | |
571 | } | |
549e9569 | 572 | } |
549e9569 | 573 | } |
836759d5 | 574 | |
549e9569 NB |
575 | new->action_fd = sysfs_open(new->devnum, NULL, "sync_action"); |
576 | new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state"); | |
c052ba30 | 577 | new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start"); |
e9dd1598 | 578 | new->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version"); |
484240d8 | 579 | new->sync_completed_fd = sysfs_open(new->devnum, NULL, "sync_completed"); |
4e6e574a DW |
580 | dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst), |
581 | new->action_fd, new->info.state_fd); | |
549e9569 | 582 | |
4fa5aef9 | 583 | sysfs_free(mdi); |
836759d5 DW |
584 | |
585 | /* if everything checks out tell the metadata handler we want to | |
586 | * manage this instance | |
587 | */ | |
588 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
589 | fprintf(stderr, "mdmon: failed to monitor %s\n", | |
590 | mdstat->metadata_version); | |
549e9569 | 591 | new->container = NULL; |
836759d5 | 592 | free_aa(new); |
93f7caca | 593 | } else { |
2a0bb19e | 594 | replace_array(container, victim, new); |
93f7caca DW |
595 | if (failed) { |
596 | new->check_degraded = 1; | |
597 | manage_member(mdstat, new); | |
598 | } | |
599 | } | |
549e9569 NB |
600 | } |
601 | ||
5d19760d | 602 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
603 | { |
604 | /* We have just read mdstat and need to compare it with | |
605 | * the known active arrays. | |
606 | * Arrays with the wrong metadata are ignored. | |
607 | */ | |
608 | ||
609 | for ( ; mdstat ; mdstat = mdstat->next) { | |
610 | struct active_array *a; | |
611 | if (mdstat->devnum == container->devnum) { | |
612 | manage_container(mdstat, container); | |
613 | continue; | |
614 | } | |
883a6142 | 615 | if (!is_container_member(mdstat, container->devname)) |
549e9569 NB |
616 | /* Not for this array */ |
617 | continue; | |
618 | /* Looks like a member of this container */ | |
5d19760d | 619 | for (a = container->arrays; a; a = a->next) { |
549e9569 NB |
620 | if (mdstat->devnum == a->devnum) { |
621 | if (a->container) | |
622 | manage_member(mdstat, a); | |
623 | break; | |
624 | } | |
625 | } | |
2a0bb19e DW |
626 | if (a == NULL || !a->container) |
627 | manage_new(mdstat, container, a); | |
549e9569 NB |
628 | } |
629 | } | |
630 | ||
edd8d13c | 631 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 632 | { |
edd8d13c NB |
633 | /* queue this metadata update through to the monitor */ |
634 | ||
635 | struct metadata_update *mu; | |
636 | ||
313a4a82 | 637 | if (msg->len <= 0) |
3c00ffbe N |
638 | while (update_queue_pending || update_queue) { |
639 | check_update_queue(container); | |
640 | usleep(15*1000); | |
641 | } | |
642 | ||
313a4a82 DW |
643 | if (msg->len == 0) { /* ping_monitor */ |
644 | int cnt; | |
645 | ||
3c00ffbe | 646 | cnt = monitor_loop_cnt; |
1eb252b8 N |
647 | if (cnt & 1) |
648 | cnt += 2; /* wait until next pselect */ | |
649 | else | |
650 | cnt += 3; /* wait for 2 pselects */ | |
651 | wakeup_monitor(); | |
3c00ffbe | 652 | |
1eb252b8 N |
653 | while (monitor_loop_cnt - cnt < 0) |
654 | usleep(10 * 1000); | |
313a4a82 DW |
655 | } else if (msg->len == -1) { /* ping_manager */ |
656 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
657 | ||
658 | manage(mdstat, container); | |
659 | free_mdstat(mdstat); | |
6144ed44 | 660 | } else if (!sigterm) { |
edd8d13c NB |
661 | mu = malloc(sizeof(*mu)); |
662 | mu->len = msg->len; | |
663 | mu->buf = msg->buf; | |
664 | msg->buf = NULL; | |
665 | mu->space = NULL; | |
666 | mu->next = NULL; | |
667 | if (container->ss->prepare_update) | |
668 | container->ss->prepare_update(container, mu); | |
669 | queue_metadata_update(mu); | |
670 | } | |
3e70c845 DW |
671 | } |
672 | ||
673 | void read_sock(struct supertype *container) | |
549e9569 NB |
674 | { |
675 | int fd; | |
bfa44e2e | 676 | struct metadata_update msg; |
b109d928 DW |
677 | int terminate = 0; |
678 | long fl; | |
679 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 680 | |
3e70c845 | 681 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
682 | if (fd < 0) |
683 | return; | |
b109d928 DW |
684 | |
685 | fl = fcntl(fd, F_GETFL, 0); | |
686 | fl |= O_NONBLOCK; | |
687 | fcntl(fd, F_SETFL, fl); | |
688 | ||
689 | do { | |
690 | msg.buf = NULL; | |
691 | ||
692 | /* read and validate the message */ | |
693 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e NB |
694 | handle_message(container, &msg); |
695 | if (ack(fd, tmo) < 0) | |
696 | terminate = 1; | |
697 | } else | |
b109d928 | 698 | terminate = 1; |
b109d928 | 699 | |
b109d928 DW |
700 | } while (!terminate); |
701 | ||
549e9569 NB |
702 | close(fd); |
703 | } | |
1ed3f387 | 704 | |
e0d6609f NB |
705 | int exit_now = 0; |
706 | int manager_ready = 0; | |
549e9569 NB |
707 | void do_manager(struct supertype *container) |
708 | { | |
709 | struct mdstat_ent *mdstat; | |
4d43913c | 710 | sigset_t set; |
1ed3f387 | 711 | |
4d43913c NB |
712 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
713 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 714 | sigdelset(&set, SIGTERM); |
549e9569 NB |
715 | |
716 | do { | |
1ed3f387 | 717 | |
e0d6609f NB |
718 | if (exit_now) |
719 | exit(0); | |
720 | ||
3c00ffbe N |
721 | /* Can only 'manage' things if 'monitor' is not making |
722 | * structural changes to metadata, so need to check | |
723 | * update_queue | |
724 | */ | |
725 | if (update_queue == NULL) { | |
726 | mdstat = mdstat_read(1, 0); | |
549e9569 | 727 | |
3c00ffbe | 728 | manage(mdstat, container); |
549e9569 | 729 | |
3c00ffbe | 730 | read_sock(container); |
4fa5aef9 | 731 | |
3c00ffbe N |
732 | free_mdstat(mdstat); |
733 | } | |
1ed3f387 NB |
734 | remove_old(); |
735 | ||
2e735d19 NB |
736 | check_update_queue(container); |
737 | ||
e0d6609f | 738 | manager_ready = 1; |
4d43913c | 739 | |
6144ed44 DW |
740 | if (sigterm) |
741 | wakeup_monitor(); | |
742 | ||
5d4d1b26 | 743 | if (update_queue == NULL) |
58a4ba2a | 744 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 745 | else |
3c00ffbe N |
746 | /* If an update is happening, just wait for signal */ |
747 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
748 | } while(1); |
749 | } |