]>
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 | 119 | |
c2047875 JS |
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); | |
2a0bb19e DW |
130 | } |
131 | ||
549e9569 NB |
132 | static void free_aa(struct active_array *aa) |
133 | { | |
2a0bb19e DW |
134 | /* Note that this doesn't close fds if they are being used |
135 | * by a clone. ->container will be set for a clone | |
549e9569 | 136 | */ |
4dd2df09 | 137 | dprintf("%s: sys_name: %s\n", __func__, aa->info.sys_name); |
2a0bb19e DW |
138 | if (!aa->container) |
139 | close_aa(aa); | |
549e9569 NB |
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 | ||
6c3fb95c NB |
148 | static struct active_array *duplicate_aa(struct active_array *aa) |
149 | { | |
503975b9 | 150 | struct active_array *newa = xmalloc(sizeof(*newa)); |
6c3fb95c NB |
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 | ||
503975b9 | 165 | d = xmalloc(sizeof(*d)); |
6c3fb95c NB |
166 | *d = **dp1; |
167 | *dp2 = d; | |
168 | dp2 = & d->next; | |
169 | } | |
7e1432fb | 170 | *dp2 = NULL; |
6c3fb95c NB |
171 | |
172 | return newa; | |
173 | } | |
174 | ||
4d43913c | 175 | static void wakeup_monitor(void) |
2a0bb19e | 176 | { |
4d43913c NB |
177 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
178 | int pid = getpid(); | |
179 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
180 | } |
181 | ||
1ed3f387 NB |
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; | |
48561b01 | 190 | wakeup_monitor(); |
1ed3f387 NB |
191 | } |
192 | } | |
193 | ||
549e9569 NB |
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 | */ | |
1ed3f387 | 204 | remove_old(); |
549e9569 NB |
205 | while (pending_discard) { |
206 | while (discard_this == NULL) | |
207 | sleep(1); | |
1ed3f387 | 208 | remove_old(); |
549e9569 NB |
209 | } |
210 | pending_discard = old; | |
211 | new->replaces = old; | |
212 | new->next = container->arrays; | |
213 | container->arrays = new; | |
4d43913c | 214 | wakeup_monitor(); |
549e9569 NB |
215 | } |
216 | ||
2e735d19 NB |
217 | struct metadata_update *update_queue = NULL; |
218 | struct metadata_update *update_queue_handled = NULL; | |
219 | struct metadata_update *update_queue_pending = NULL; | |
220 | ||
071cfc42 | 221 | static void free_updates(struct metadata_update **update) |
2e735d19 | 222 | { |
071cfc42 DW |
223 | while (*update) { |
224 | struct metadata_update *this = *update; | |
cb23f1f4 | 225 | void **space_list = this->space_list; |
071cfc42 DW |
226 | |
227 | *update = this->next; | |
904c1ef7 | 228 | free(this->buf); |
071cfc42 | 229 | free(this->space); |
cb23f1f4 N |
230 | while (space_list) { |
231 | void *space = space_list; | |
232 | space_list = *space_list; | |
233 | free(space); | |
234 | } | |
2e735d19 NB |
235 | free(this); |
236 | } | |
071cfc42 DW |
237 | } |
238 | ||
239 | void check_update_queue(struct supertype *container) | |
240 | { | |
241 | free_updates(&update_queue_handled); | |
242 | ||
2e735d19 NB |
243 | if (update_queue == NULL && |
244 | update_queue_pending) { | |
245 | update_queue = update_queue_pending; | |
246 | update_queue_pending = NULL; | |
4d43913c | 247 | wakeup_monitor(); |
2e735d19 NB |
248 | } |
249 | } | |
250 | ||
6c3fb95c | 251 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
252 | { |
253 | struct metadata_update **qp; | |
254 | ||
255 | qp = &update_queue_pending; | |
256 | while (*qp) | |
257 | qp = & ((*qp)->next); | |
258 | *qp = mu; | |
259 | } | |
260 | ||
43dad3d6 DW |
261 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
262 | { | |
263 | int dfd; | |
264 | char nm[20]; | |
661dce36 | 265 | struct supertype *st2; |
43dad3d6 | 266 | struct metadata_update *update = NULL; |
661dce36 | 267 | struct mdinfo info; |
43dad3d6 DW |
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 | ||
04a8ac08 DW |
279 | sd->next = st->devs; |
280 | st->devs = sd; | |
281 | ||
43dad3d6 DW |
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 | ||
661dce36 N |
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) { | |
a5d85af7 | 292 | st2->ss->getinfo_super(st, &info, NULL); |
661dce36 N |
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 | ||
43dad3d6 | 306 | st->update_tail = &update; |
72ca9bcf | 307 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); |
43dad3d6 DW |
308 | st->ss->write_init_super(st); |
309 | queue_metadata_update(update); | |
310 | st->update_tail = NULL; | |
311 | } | |
312 | ||
1a64be56 LM |
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 | }; | |
1a64be56 LM |
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); | |
4dd968cc N |
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 | */ | |
1a64be56 LM |
337 | st->ss->write_init_super(st); |
338 | queue_metadata_update(update); | |
339 | st->update_tail = NULL; | |
340 | } | |
341 | ||
549e9569 NB |
342 | static void manage_container(struct mdstat_ent *mdstat, |
343 | struct supertype *container) | |
344 | { | |
1a64be56 | 345 | /* Of interest here are: |
1011e834 | 346 | * - if a new device has been added to the container, we |
1a64be56 LM |
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. | |
549e9569 NB |
350 | * FIXME should we look for compatible metadata and take hints |
351 | * about spare assignment.... probably not. | |
549e9569 NB |
352 | */ |
353 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
354 | struct mdinfo **cdp, *cd, *di, *mdi; |
355 | int found; | |
356 | ||
549e9569 NB |
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 | */ | |
4dd2df09 | 362 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); |
313a4a82 DW |
363 | if (!mdi) { |
364 | /* invalidate the current count so we can try again */ | |
365 | container->devcnt = -1; | |
7bc1962f | 366 | return; |
313a4a82 | 367 | } |
7bc1962f DW |
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; | |
1a64be56 | 381 | remove_disk_from_container(container, cd); |
7bc1962f DW |
382 | free(cd); |
383 | } else | |
384 | cdp = &(*cdp)->next; | |
385 | } | |
43dad3d6 DW |
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; | |
04a8ac08 | 393 | if (!cd) { |
503975b9 | 394 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
04a8ac08 | 395 | |
04a8ac08 DW |
396 | *newd = *di; |
397 | add_disk_to_container(container, newd); | |
398 | } | |
43dad3d6 | 399 | } |
7bc1962f | 400 | sysfs_free(mdi); |
549e9569 NB |
401 | container->devcnt = mdstat->devcnt; |
402 | } | |
403 | } | |
404 | ||
63b4aae3 DW |
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; | |
4dd2df09 | 412 | disk->recovery_fd = sysfs_open(aa->info.sys_name, disk->sys_name, "recovery_start"); |
3e1d79b2 JS |
413 | if (disk->recovery_fd < 0) |
414 | return -1; | |
4dd2df09 | 415 | disk->state_fd = sysfs_open(aa->info.sys_name, disk->sys_name, "state"); |
3e1d79b2 JS |
416 | if (disk->state_fd < 0) { |
417 | close(disk->recovery_fd); | |
418 | return -1; | |
419 | } | |
63b4aae3 DW |
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 | ||
549e9569 NB |
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 | * | |
0f99b4bd N |
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. | |
6c3fb95c | 439 | * |
0f99b4bd | 440 | * We also want to handle degraded arrays here by |
6c3fb95c NB |
441 | * trying to find and assign a spare. |
442 | * We do that whenever the monitor tells us too. | |
549e9569 | 443 | */ |
bc77ed53 DW |
444 | char buf[64]; |
445 | int frozen; | |
4e2c1a9a | 446 | struct supertype *container = a->container; |
4edb8530 | 447 | unsigned long long int component_size = 0; |
4e2c1a9a N |
448 | |
449 | if (container == NULL) | |
450 | /* Raced with something */ | |
451 | return; | |
bc77ed53 | 452 | |
549e9569 NB |
453 | // FIXME |
454 | a->info.array.raid_disks = mdstat->raid_disks; | |
549e9569 NB |
455 | // MORE |
456 | ||
4edb8530 PB |
457 | if (sysfs_get_ll(&a->info, NULL, "component_size", &component_size) >= 0) |
458 | a->info.component_size = component_size << 1; | |
459 | ||
bc77ed53 DW |
460 | /* honor 'frozen' */ |
461 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
462 | frozen = buf[9] == '-'; | |
463 | else | |
464 | frozen = 1; /* can't read metadata_version assume the worst */ | |
465 | ||
f54a6742 N |
466 | /* If sync_action is not 'idle' then don't try recovery now */ |
467 | if (!frozen | |
468 | && sysfs_get_str(&a->info, NULL, "sync_action", buf, sizeof(buf)) > 0 | |
469 | && strncmp(buf, "idle", 4) != 0) | |
470 | frozen = 1; | |
471 | ||
57f8c769 AK |
472 | if (mdstat->level) { |
473 | int level = map_name(pers, mdstat->level); | |
7023e0b8 | 474 | if (level == 0 || level == LEVEL_LINEAR) { |
ba714450 | 475 | a->to_remove = 1; |
84f3857f | 476 | wakeup_monitor(); |
7023e0b8 N |
477 | return; |
478 | } | |
479 | else if (a->info.array.level != level && level > 0) { | |
57f8c769 AK |
480 | struct active_array *newa = duplicate_aa(a); |
481 | if (newa) { | |
482 | newa->info.array.level = level; | |
4e2c1a9a | 483 | replace_array(container, a, newa); |
57f8c769 AK |
484 | a = newa; |
485 | } | |
486 | } | |
487 | } | |
488 | ||
50927b13 AK |
489 | /* we are after monitor kick, |
490 | * so container field can be cleared - check it again | |
491 | */ | |
492 | if (a->container == NULL) | |
493 | return; | |
494 | ||
0c4f6e37 | 495 | /* We don't check the array while any update is pending, as it |
88b496c2 | 496 | * might container a change (such as a spare assignment) which |
0c4f6e37 N |
497 | * could affect our decisions. |
498 | */ | |
88b496c2 | 499 | if (a->check_degraded && !frozen && |
0c4f6e37 | 500 | update_queue == NULL && update_queue_pending == NULL) { |
6c3fb95c | 501 | struct metadata_update *updates = NULL; |
071cfc42 | 502 | struct mdinfo *newdev = NULL; |
6c3fb95c | 503 | struct active_array *newa; |
071cfc42 | 504 | struct mdinfo *d; |
3c00ffbe | 505 | |
6c3fb95c NB |
506 | a->check_degraded = 0; |
507 | ||
508 | /* The array may not be degraded, this is just a good time | |
509 | * to check. | |
510 | */ | |
4e2c1a9a | 511 | newdev = container->ss->activate_spare(a, &updates); |
071cfc42 DW |
512 | if (!newdev) |
513 | return; | |
514 | ||
515 | newa = duplicate_aa(a); | |
516 | if (!newa) | |
517 | goto out; | |
1d446d52 DW |
518 | /* prevent the kernel from activating the disk(s) before we |
519 | * finish adding them | |
520 | */ | |
30b83120 | 521 | dprintf("%s: freezing %s\n", __func__, a->info.sys_name); |
1d446d52 | 522 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); |
071cfc42 DW |
523 | |
524 | /* Add device to array and set offset/size/slot. | |
525 | * and open files for each newdev */ | |
526 | for (d = newdev; d ; d = d->next) { | |
527 | struct mdinfo *newd; | |
528 | ||
503975b9 | 529 | newd = xmalloc(sizeof(*newd)); |
2904b26f | 530 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
531 | free(newd); |
532 | continue; | |
6c3fb95c | 533 | } |
63b4aae3 | 534 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
535 | } |
536 | queue_metadata_update(updates); | |
537 | updates = NULL; | |
6ca1e6ec MW |
538 | while (update_queue_pending || update_queue) { |
539 | check_update_queue(container); | |
540 | usleep(15*1000); | |
541 | } | |
4e2c1a9a | 542 | replace_array(container, a, newa); |
6ca1e6ec MW |
543 | if (sysfs_set_str(&a->info, NULL, "sync_action", "recover") |
544 | == 0) | |
545 | newa->prev_action = recover; | |
30b83120 MW |
546 | dprintf("%s: recovery started on %s\n", __func__, |
547 | a->info.sys_name); | |
071cfc42 DW |
548 | out: |
549 | while (newdev) { | |
550 | d = newdev->next; | |
551 | free(newdev); | |
552 | newdev = d; | |
6c3fb95c | 553 | } |
071cfc42 | 554 | free_updates(&updates); |
6c3fb95c | 555 | } |
0f99b4bd N |
556 | |
557 | if (a->check_reshape) { | |
558 | /* mdadm might have added some devices to the array. | |
559 | * We want to disk_init_and_add any such device to a | |
560 | * duplicate_aa and replace a with that. | |
561 | * mdstat doesn't have enough info so we sysfs_read | |
562 | * and look for new stuff. | |
563 | */ | |
564 | struct mdinfo *info, *d, *d2, *newd; | |
aad6f216 | 565 | unsigned long long array_size; |
0f99b4bd N |
566 | struct active_array *newa = NULL; |
567 | a->check_reshape = 0; | |
4dd2df09 | 568 | info = sysfs_read(-1, mdstat->devnm, |
0f99b4bd N |
569 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
570 | if (!info) | |
571 | goto out2; | |
572 | for (d = info->devs; d; d = d->next) { | |
573 | if (d->disk.raid_disk < 0) | |
574 | continue; | |
575 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
576 | if (d2->disk.raid_disk == | |
577 | d->disk.raid_disk) | |
578 | break; | |
579 | if (d2) | |
580 | /* already have this one */ | |
581 | continue; | |
582 | if (!newa) { | |
583 | newa = duplicate_aa(a); | |
584 | if (!newa) | |
585 | break; | |
586 | } | |
503975b9 | 587 | newd = xmalloc(sizeof(*newd)); |
0f99b4bd N |
588 | disk_init_and_add(newd, d, newa); |
589 | } | |
aad6f216 | 590 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 |
02eedb57 | 591 | && a->info.custom_array_size > array_size*2) { |
aad6f216 | 592 | sysfs_set_num(info, NULL, "array_size", |
02eedb57 | 593 | a->info.custom_array_size/2); |
aad6f216 | 594 | } |
0f99b4bd N |
595 | out2: |
596 | sysfs_free(info); | |
597 | if (newa) | |
4e2c1a9a | 598 | replace_array(container, a, newa); |
0f99b4bd | 599 | } |
549e9569 NB |
600 | } |
601 | ||
836759d5 DW |
602 | static int aa_ready(struct active_array *aa) |
603 | { | |
604 | struct mdinfo *d; | |
605 | int level = aa->info.array.level; | |
606 | ||
607 | for (d = aa->info.devs; d; d = d->next) | |
608 | if (d->state_fd < 0) | |
609 | return 0; | |
610 | ||
611 | if (aa->info.state_fd < 0) | |
612 | return 0; | |
613 | ||
614 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
615 | return 0; | |
616 | ||
617 | if (!aa->container) | |
618 | return 0; | |
619 | ||
620 | return 1; | |
621 | } | |
622 | ||
549e9569 | 623 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
624 | struct supertype *container, |
625 | struct active_array *victim) | |
549e9569 NB |
626 | { |
627 | /* A new array has appeared in this container. | |
628 | * Hopefully it is already recorded in the metadata. | |
629 | * Check, then create the new array to report it to | |
630 | * the monitor. | |
631 | */ | |
632 | ||
633 | struct active_array *new; | |
634 | struct mdinfo *mdi, *di; | |
cba0191b | 635 | char *inst; |
549e9569 | 636 | int i; |
f1d26766 | 637 | int failed = 0; |
138477db | 638 | char buf[40]; |
549e9569 | 639 | |
836759d5 | 640 | /* check if array is ready to be monitored */ |
7023e0b8 N |
641 | if (!mdstat->active || !mdstat->level) |
642 | return; | |
643 | if (strcmp(mdstat->level, "raid0") == 0 || | |
644 | strcmp(mdstat->level, "linear") == 0) | |
836759d5 DW |
645 | return; |
646 | ||
4dd2df09 | 647 | mdi = sysfs_read(-1, mdstat->devnm, |
836759d5 | 648 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| |
f1d26766 | 649 | GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
836759d5 | 650 | |
503975b9 | 651 | if (!mdi) |
836759d5 | 652 | return; |
503975b9 | 653 | new = xcalloc(1, sizeof(*new)); |
d52690ac | 654 | |
4dd2df09 | 655 | strcpy(new->info.sys_name, mdstat->devnm); |
549e9569 NB |
656 | |
657 | new->prev_state = new->curr_state = new->next_state = inactive; | |
658 | new->prev_action= new->curr_action= new->next_action= idle; | |
659 | ||
660 | new->container = container; | |
661 | ||
4dd2df09 | 662 | inst = to_subarray(mdstat, container->devnm); |
549e9569 | 663 | |
549e9569 | 664 | new->info.array = mdi->array; |
272bcc48 | 665 | new->info.component_size = mdi->component_size; |
549e9569 NB |
666 | |
667 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
503975b9 | 668 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
549e9569 NB |
669 | |
670 | for (di = mdi->devs; di; di = di->next) | |
671 | if (i == di->disk.raid_disk) | |
672 | break; | |
673 | ||
63b4aae3 | 674 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
675 | if (newd) |
676 | free(newd); | |
677 | ||
f1d26766 | 678 | failed++; |
7da80e6f DW |
679 | if (failed > new->info.array.failed_disks) { |
680 | /* we cannot properly monitor without all working disks */ | |
681 | new->container = NULL; | |
682 | break; | |
683 | } | |
549e9569 | 684 | } |
549e9569 | 685 | } |
836759d5 | 686 | |
4dd2df09 N |
687 | new->action_fd = sysfs_open(new->info.sys_name, NULL, "sync_action"); |
688 | new->info.state_fd = sysfs_open(new->info.sys_name, NULL, "array_state"); | |
689 | new->resync_start_fd = sysfs_open(new->info.sys_name, NULL, "resync_start"); | |
690 | new->metadata_fd = sysfs_open(new->info.sys_name, NULL, "metadata_version"); | |
691 | new->sync_completed_fd = sysfs_open(new->info.sys_name, NULL, "sync_completed"); | |
4e6e574a DW |
692 | dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst), |
693 | new->action_fd, new->info.state_fd); | |
549e9569 | 694 | |
138477db AK |
695 | /* reshape_position is set by mdadm in sysfs |
696 | * read this information for new arrays only (empty victim) | |
697 | */ | |
698 | if ((victim == NULL) && | |
699 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
700 | (strncmp(buf, "reshape", 7) == 0)) { | |
701 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
702 | &new->last_checkpoint) != 0) | |
703 | new->last_checkpoint = 0; | |
0d51bfa2 AK |
704 | else { |
705 | int data_disks = mdi->array.raid_disks; | |
706 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
707 | data_disks--; | |
708 | if (mdi->array.level == 6) | |
709 | data_disks -= 2; | |
710 | ||
711 | new->last_checkpoint /= data_disks; | |
712 | } | |
138477db AK |
713 | dprintf("mdmon: New monitored array is under reshape.\n" |
714 | " Last checkpoint is: %llu\n", | |
715 | new->last_checkpoint); | |
716 | } | |
717 | ||
4fa5aef9 | 718 | sysfs_free(mdi); |
836759d5 DW |
719 | |
720 | /* if everything checks out tell the metadata handler we want to | |
721 | * manage this instance | |
722 | */ | |
723 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
a88e119f | 724 | pr_err("failed to monitor %s\n", |
836759d5 | 725 | mdstat->metadata_version); |
549e9569 | 726 | new->container = NULL; |
836759d5 | 727 | free_aa(new); |
93f7caca | 728 | } else { |
2a0bb19e | 729 | replace_array(container, victim, new); |
93f7caca DW |
730 | if (failed) { |
731 | new->check_degraded = 1; | |
732 | manage_member(mdstat, new); | |
733 | } | |
734 | } | |
549e9569 NB |
735 | } |
736 | ||
5d19760d | 737 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
738 | { |
739 | /* We have just read mdstat and need to compare it with | |
740 | * the known active arrays. | |
741 | * Arrays with the wrong metadata are ignored. | |
742 | */ | |
743 | ||
744 | for ( ; mdstat ; mdstat = mdstat->next) { | |
745 | struct active_array *a; | |
4dd2df09 | 746 | if (strcmp(mdstat->devnm, container->devnm) == 0) { |
549e9569 NB |
747 | manage_container(mdstat, container); |
748 | continue; | |
749 | } | |
4dd2df09 | 750 | if (!is_container_member(mdstat, container->devnm)) |
549e9569 NB |
751 | /* Not for this array */ |
752 | continue; | |
753 | /* Looks like a member of this container */ | |
5d19760d | 754 | for (a = container->arrays; a; a = a->next) { |
4dd2df09 | 755 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { |
ba714450 | 756 | if (a->container && a->to_remove == 0) |
549e9569 NB |
757 | manage_member(mdstat, a); |
758 | break; | |
759 | } | |
760 | } | |
2a0bb19e DW |
761 | if (a == NULL || !a->container) |
762 | manage_new(mdstat, container, a); | |
549e9569 NB |
763 | } |
764 | } | |
765 | ||
edd8d13c | 766 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 767 | { |
edd8d13c NB |
768 | /* queue this metadata update through to the monitor */ |
769 | ||
770 | struct metadata_update *mu; | |
771 | ||
313a4a82 | 772 | if (msg->len <= 0) |
3c00ffbe N |
773 | while (update_queue_pending || update_queue) { |
774 | check_update_queue(container); | |
775 | usleep(15*1000); | |
776 | } | |
777 | ||
313a4a82 DW |
778 | if (msg->len == 0) { /* ping_monitor */ |
779 | int cnt; | |
1011e834 | 780 | |
3c00ffbe | 781 | cnt = monitor_loop_cnt; |
1eb252b8 N |
782 | if (cnt & 1) |
783 | cnt += 2; /* wait until next pselect */ | |
784 | else | |
785 | cnt += 3; /* wait for 2 pselects */ | |
786 | wakeup_monitor(); | |
3c00ffbe | 787 | |
1eb252b8 N |
788 | while (monitor_loop_cnt - cnt < 0) |
789 | usleep(10 * 1000); | |
313a4a82 DW |
790 | } else if (msg->len == -1) { /* ping_manager */ |
791 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
792 | ||
793 | manage(mdstat, container); | |
794 | free_mdstat(mdstat); | |
6144ed44 | 795 | } else if (!sigterm) { |
503975b9 | 796 | mu = xmalloc(sizeof(*mu)); |
edd8d13c NB |
797 | mu->len = msg->len; |
798 | mu->buf = msg->buf; | |
799 | msg->buf = NULL; | |
800 | mu->space = NULL; | |
cb23f1f4 | 801 | mu->space_list = NULL; |
edd8d13c NB |
802 | mu->next = NULL; |
803 | if (container->ss->prepare_update) | |
804 | container->ss->prepare_update(container, mu); | |
805 | queue_metadata_update(mu); | |
806 | } | |
3e70c845 DW |
807 | } |
808 | ||
809 | void read_sock(struct supertype *container) | |
549e9569 NB |
810 | { |
811 | int fd; | |
bfa44e2e | 812 | struct metadata_update msg; |
b109d928 DW |
813 | int terminate = 0; |
814 | long fl; | |
815 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 816 | |
3e70c845 | 817 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
818 | if (fd < 0) |
819 | return; | |
b109d928 DW |
820 | |
821 | fl = fcntl(fd, F_GETFL, 0); | |
822 | fl |= O_NONBLOCK; | |
823 | fcntl(fd, F_SETFL, fl); | |
824 | ||
825 | do { | |
826 | msg.buf = NULL; | |
827 | ||
828 | /* read and validate the message */ | |
829 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e | 830 | handle_message(container, &msg); |
bc77ed53 DW |
831 | if (msg.len == 0) { |
832 | /* ping reply with version */ | |
833 | msg.buf = Version; | |
834 | msg.len = strlen(Version) + 1; | |
835 | if (send_message(fd, &msg, tmo) < 0) | |
836 | terminate = 1; | |
837 | } else if (ack(fd, tmo) < 0) | |
bfa44e2e NB |
838 | terminate = 1; |
839 | } else | |
b109d928 | 840 | terminate = 1; |
b109d928 | 841 | |
b109d928 DW |
842 | } while (!terminate); |
843 | ||
549e9569 NB |
844 | close(fd); |
845 | } | |
1ed3f387 | 846 | |
e0d6609f NB |
847 | int exit_now = 0; |
848 | int manager_ready = 0; | |
549e9569 NB |
849 | void do_manager(struct supertype *container) |
850 | { | |
851 | struct mdstat_ent *mdstat; | |
4d43913c | 852 | sigset_t set; |
1ed3f387 | 853 | |
4d43913c NB |
854 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
855 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 856 | sigdelset(&set, SIGTERM); |
549e9569 NB |
857 | |
858 | do { | |
1ed3f387 | 859 | |
e0d6609f NB |
860 | if (exit_now) |
861 | exit(0); | |
862 | ||
3c00ffbe N |
863 | /* Can only 'manage' things if 'monitor' is not making |
864 | * structural changes to metadata, so need to check | |
865 | * update_queue | |
866 | */ | |
867 | if (update_queue == NULL) { | |
868 | mdstat = mdstat_read(1, 0); | |
549e9569 | 869 | |
3c00ffbe | 870 | manage(mdstat, container); |
549e9569 | 871 | |
3c00ffbe | 872 | read_sock(container); |
4fa5aef9 | 873 | |
3c00ffbe N |
874 | free_mdstat(mdstat); |
875 | } | |
1ed3f387 NB |
876 | remove_old(); |
877 | ||
2e735d19 NB |
878 | check_update_queue(container); |
879 | ||
e0d6609f | 880 | manager_ready = 1; |
4d43913c | 881 | |
6144ed44 DW |
882 | if (sigterm) |
883 | wakeup_monitor(); | |
884 | ||
5d4d1b26 | 885 | if (update_queue == NULL) |
58a4ba2a | 886 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 887 | else |
3c00ffbe N |
888 | /* If an update is happening, just wait for signal */ |
889 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
890 | } while(1); |
891 | } |