]>
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); |
6dc1785f TM |
118 | close(d->bb_fd); |
119 | close(d->ubb_fd); | |
e1516be1 | 120 | } |
2a0bb19e | 121 | |
c2047875 JS |
122 | if (aa->action_fd >= 0) |
123 | close(aa->action_fd); | |
124 | if (aa->info.state_fd >= 0) | |
125 | close(aa->info.state_fd); | |
126 | if (aa->resync_start_fd >= 0) | |
127 | close(aa->resync_start_fd); | |
128 | if (aa->metadata_fd >= 0) | |
129 | close(aa->metadata_fd); | |
130 | if (aa->sync_completed_fd >= 0) | |
131 | close(aa->sync_completed_fd); | |
c76242c5 TM |
132 | if (aa->safe_mode_delay_fd >= 0) |
133 | close(aa->safe_mode_delay_fd); | |
2a0bb19e DW |
134 | } |
135 | ||
549e9569 NB |
136 | static void free_aa(struct active_array *aa) |
137 | { | |
2a0bb19e DW |
138 | /* Note that this doesn't close fds if they are being used |
139 | * by a clone. ->container will be set for a clone | |
549e9569 | 140 | */ |
1ade5cc1 | 141 | dprintf("sys_name: %s\n", aa->info.sys_name); |
2a0bb19e DW |
142 | if (!aa->container) |
143 | close_aa(aa); | |
549e9569 NB |
144 | while (aa->info.devs) { |
145 | struct mdinfo *d = aa->info.devs; | |
146 | aa->info.devs = d->next; | |
147 | free(d); | |
148 | } | |
149 | free(aa); | |
150 | } | |
151 | ||
6c3fb95c NB |
152 | static struct active_array *duplicate_aa(struct active_array *aa) |
153 | { | |
503975b9 | 154 | struct active_array *newa = xmalloc(sizeof(*newa)); |
6c3fb95c NB |
155 | struct mdinfo **dp1, **dp2; |
156 | ||
157 | *newa = *aa; | |
158 | newa->next = NULL; | |
159 | newa->replaces = NULL; | |
160 | newa->info.next = NULL; | |
161 | ||
162 | dp2 = &newa->info.devs; | |
163 | ||
164 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
165 | struct mdinfo *d; | |
166 | if ((*dp1)->state_fd < 0) | |
167 | continue; | |
168 | ||
503975b9 | 169 | d = xmalloc(sizeof(*d)); |
6c3fb95c NB |
170 | *d = **dp1; |
171 | *dp2 = d; | |
172 | dp2 = & d->next; | |
173 | } | |
7e1432fb | 174 | *dp2 = NULL; |
6c3fb95c NB |
175 | |
176 | return newa; | |
177 | } | |
178 | ||
4d43913c | 179 | static void wakeup_monitor(void) |
2a0bb19e | 180 | { |
4d43913c NB |
181 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
182 | int pid = getpid(); | |
183 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
184 | } |
185 | ||
1ed3f387 NB |
186 | static void remove_old(void) |
187 | { | |
188 | if (discard_this) { | |
189 | discard_this->next = NULL; | |
190 | free_aa(discard_this); | |
191 | if (pending_discard == discard_this) | |
192 | pending_discard = NULL; | |
193 | discard_this = NULL; | |
48561b01 | 194 | wakeup_monitor(); |
1ed3f387 NB |
195 | } |
196 | } | |
197 | ||
549e9569 NB |
198 | static void replace_array(struct supertype *container, |
199 | struct active_array *old, | |
200 | struct active_array *new) | |
201 | { | |
202 | /* To replace an array, we add it to the top of the list | |
203 | * marked with ->replaces to point to the original. | |
204 | * 'monitor' will take the original out of the list | |
205 | * and put it on 'discard_this'. We take it from there | |
206 | * and discard it. | |
207 | */ | |
1ed3f387 | 208 | remove_old(); |
549e9569 NB |
209 | while (pending_discard) { |
210 | while (discard_this == NULL) | |
211 | sleep(1); | |
1ed3f387 | 212 | remove_old(); |
549e9569 NB |
213 | } |
214 | pending_discard = old; | |
215 | new->replaces = old; | |
216 | new->next = container->arrays; | |
217 | container->arrays = new; | |
4d43913c | 218 | wakeup_monitor(); |
549e9569 NB |
219 | } |
220 | ||
2e735d19 NB |
221 | struct metadata_update *update_queue = NULL; |
222 | struct metadata_update *update_queue_handled = NULL; | |
223 | struct metadata_update *update_queue_pending = NULL; | |
224 | ||
071cfc42 | 225 | static void free_updates(struct metadata_update **update) |
2e735d19 | 226 | { |
071cfc42 DW |
227 | while (*update) { |
228 | struct metadata_update *this = *update; | |
cb23f1f4 | 229 | void **space_list = this->space_list; |
071cfc42 DW |
230 | |
231 | *update = this->next; | |
904c1ef7 | 232 | free(this->buf); |
071cfc42 | 233 | free(this->space); |
cb23f1f4 N |
234 | while (space_list) { |
235 | void *space = space_list; | |
236 | space_list = *space_list; | |
237 | free(space); | |
238 | } | |
2e735d19 NB |
239 | free(this); |
240 | } | |
071cfc42 DW |
241 | } |
242 | ||
243 | void check_update_queue(struct supertype *container) | |
244 | { | |
245 | free_updates(&update_queue_handled); | |
246 | ||
2e735d19 NB |
247 | if (update_queue == NULL && |
248 | update_queue_pending) { | |
249 | update_queue = update_queue_pending; | |
250 | update_queue_pending = NULL; | |
4d43913c | 251 | wakeup_monitor(); |
2e735d19 NB |
252 | } |
253 | } | |
254 | ||
6c3fb95c | 255 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
256 | { |
257 | struct metadata_update **qp; | |
258 | ||
259 | qp = &update_queue_pending; | |
260 | while (*qp) | |
261 | qp = & ((*qp)->next); | |
262 | *qp = mu; | |
263 | } | |
264 | ||
43dad3d6 DW |
265 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
266 | { | |
267 | int dfd; | |
268 | char nm[20]; | |
269 | struct metadata_update *update = NULL; | |
270 | mdu_disk_info_t dk = { | |
271 | .number = -1, | |
272 | .major = sd->disk.major, | |
273 | .minor = sd->disk.minor, | |
274 | .raid_disk = -1, | |
275 | .state = 0, | |
276 | }; | |
277 | ||
1ade5cc1 | 278 | dprintf("add %d:%d to container\n", sd->disk.major, sd->disk.minor); |
43dad3d6 | 279 | |
04a8ac08 DW |
280 | sd->next = st->devs; |
281 | st->devs = sd; | |
282 | ||
43dad3d6 DW |
283 | sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor); |
284 | dfd = dev_open(nm, O_RDWR); | |
285 | if (dfd < 0) | |
286 | return; | |
287 | ||
288 | st->update_tail = &update; | |
72ca9bcf | 289 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); |
43dad3d6 DW |
290 | st->ss->write_init_super(st); |
291 | queue_metadata_update(update); | |
292 | st->update_tail = NULL; | |
293 | } | |
294 | ||
1a64be56 LM |
295 | /* |
296 | * Create and queue update structure about the removed disks. | |
297 | * The update is prepared by super type handler and passed to the monitor | |
298 | * thread. | |
299 | */ | |
300 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
301 | { | |
302 | struct metadata_update *update = NULL; | |
303 | mdu_disk_info_t dk = { | |
304 | .number = -1, | |
305 | .major = sd->disk.major, | |
306 | .minor = sd->disk.minor, | |
307 | .raid_disk = -1, | |
308 | .state = 0, | |
309 | }; | |
1ade5cc1 N |
310 | dprintf("remove %d:%d from container\n", |
311 | sd->disk.major, sd->disk.minor); | |
1a64be56 LM |
312 | |
313 | st->update_tail = &update; | |
314 | st->ss->remove_from_super(st, &dk); | |
4dd968cc N |
315 | /* FIXME this write_init_super shouldn't be here. |
316 | * We have it after add_to_super to write to new device, | |
317 | * but with 'remove' we don't ant to write to that device! | |
318 | */ | |
1a64be56 LM |
319 | st->ss->write_init_super(st); |
320 | queue_metadata_update(update); | |
321 | st->update_tail = NULL; | |
322 | } | |
323 | ||
549e9569 NB |
324 | static void manage_container(struct mdstat_ent *mdstat, |
325 | struct supertype *container) | |
326 | { | |
1a64be56 | 327 | /* Of interest here are: |
1011e834 | 328 | * - if a new device has been added to the container, we |
1a64be56 LM |
329 | * add it to the array ignoring any metadata on it. |
330 | * - if a device has been removed from the container, we | |
331 | * remove it from the device list and update the metadata. | |
549e9569 NB |
332 | * FIXME should we look for compatible metadata and take hints |
333 | * about spare assignment.... probably not. | |
549e9569 NB |
334 | */ |
335 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
336 | struct mdinfo **cdp, *cd, *di, *mdi; |
337 | int found; | |
338 | ||
549e9569 NB |
339 | /* read /sys/block/NAME/md/dev-??/block/dev to find out |
340 | * what is there, and compare with container->info.devs | |
341 | * To see what is removed and what is added. | |
342 | * These need to be remove from, or added to, the array | |
343 | */ | |
4dd2df09 | 344 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); |
313a4a82 DW |
345 | if (!mdi) { |
346 | /* invalidate the current count so we can try again */ | |
347 | container->devcnt = -1; | |
7bc1962f | 348 | return; |
313a4a82 | 349 | } |
7bc1962f DW |
350 | |
351 | /* check for removals */ | |
352 | for (cdp = &container->devs; *cdp; ) { | |
353 | found = 0; | |
354 | for (di = mdi->devs; di; di = di->next) | |
355 | if (di->disk.major == (*cdp)->disk.major && | |
356 | di->disk.minor == (*cdp)->disk.minor) { | |
357 | found = 1; | |
358 | break; | |
359 | } | |
360 | if (!found) { | |
361 | cd = *cdp; | |
362 | *cdp = (*cdp)->next; | |
1a64be56 | 363 | remove_disk_from_container(container, cd); |
7bc1962f DW |
364 | free(cd); |
365 | } else | |
366 | cdp = &(*cdp)->next; | |
367 | } | |
43dad3d6 DW |
368 | |
369 | /* check for additions */ | |
370 | for (di = mdi->devs; di; di = di->next) { | |
371 | for (cd = container->devs; cd; cd = cd->next) | |
372 | if (di->disk.major == cd->disk.major && | |
373 | di->disk.minor == cd->disk.minor) | |
374 | break; | |
04a8ac08 | 375 | if (!cd) { |
503975b9 | 376 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
04a8ac08 | 377 | |
04a8ac08 DW |
378 | *newd = *di; |
379 | add_disk_to_container(container, newd); | |
380 | } | |
43dad3d6 | 381 | } |
7bc1962f | 382 | sysfs_free(mdi); |
549e9569 NB |
383 | container->devcnt = mdstat->devcnt; |
384 | } | |
385 | } | |
386 | ||
da5a36fa N |
387 | static int sysfs_open2(char *devnum, char *name, char *attr) |
388 | { | |
389 | int fd = sysfs_open(devnum, name, attr); | |
390 | if (fd >= 0) { | |
391 | /* seq_file in the kernel allocates buffer space | |
392 | * on the first read. Do that now so 'monitor' | |
393 | * never needs too. | |
394 | */ | |
395 | char buf[200]; | |
27aefbdb N |
396 | if (read(fd, buf, sizeof(buf)) < 0) |
397 | /* pretend not to ignore return value */ | |
398 | return fd; | |
da5a36fa N |
399 | } |
400 | return fd; | |
401 | } | |
402 | ||
63b4aae3 DW |
403 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, |
404 | struct active_array *aa) | |
405 | { | |
406 | if (!disk || !clone) | |
407 | return -1; | |
408 | ||
409 | *disk = *clone; | |
da5a36fa N |
410 | disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
411 | "recovery_start"); | |
3e1d79b2 JS |
412 | if (disk->recovery_fd < 0) |
413 | return -1; | |
da5a36fa | 414 | disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state"); |
3e1d79b2 JS |
415 | if (disk->state_fd < 0) { |
416 | close(disk->recovery_fd); | |
417 | return -1; | |
418 | } | |
6dc1785f TM |
419 | disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
420 | "bad_blocks"); | |
421 | if (disk->bb_fd < 0) { | |
422 | close(disk->recovery_fd); | |
423 | close(disk->state_fd); | |
424 | return -1; | |
425 | } | |
426 | disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
427 | "unacknowledged_bad_blocks"); | |
428 | if (disk->ubb_fd < 0) { | |
429 | close(disk->recovery_fd); | |
430 | close(disk->state_fd); | |
431 | close(disk->bb_fd); | |
432 | return -1; | |
433 | } | |
63b4aae3 DW |
434 | disk->prev_state = read_dev_state(disk->state_fd); |
435 | disk->curr_state = disk->prev_state; | |
436 | disk->next = aa->info.devs; | |
437 | aa->info.devs = disk; | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
549e9569 NB |
442 | static void manage_member(struct mdstat_ent *mdstat, |
443 | struct active_array *a) | |
444 | { | |
445 | /* Compare mdstat info with known state of member array. | |
446 | * We do not need to look for device state changes here, that | |
447 | * is dealt with by the monitor. | |
448 | * | |
0f99b4bd N |
449 | * If a reshape is being requested, monitor will have noticed |
450 | * that sync_action changed and will have set check_reshape. | |
451 | * We just need to see if new devices have appeared. All metadata | |
452 | * updates will already have been processed. | |
6c3fb95c | 453 | * |
0f99b4bd | 454 | * We also want to handle degraded arrays here by |
6c3fb95c NB |
455 | * trying to find and assign a spare. |
456 | * We do that whenever the monitor tells us too. | |
549e9569 | 457 | */ |
bc77ed53 DW |
458 | char buf[64]; |
459 | int frozen; | |
4e2c1a9a | 460 | struct supertype *container = a->container; |
2c8890e9 | 461 | struct mdinfo *mdi; |
4e2c1a9a N |
462 | |
463 | if (container == NULL) | |
464 | /* Raced with something */ | |
465 | return; | |
bc77ed53 | 466 | |
e49a8a80 N |
467 | if (mdstat->active) { |
468 | // FIXME | |
469 | a->info.array.raid_disks = mdstat->raid_disks; | |
470 | // MORE | |
471 | } | |
549e9569 | 472 | |
2c8890e9 AP |
473 | mdi = sysfs_read(-1, mdstat->devnm, |
474 | GET_COMPONENT|GET_CONSISTENCY_POLICY); | |
475 | if (mdi) { | |
476 | a->info.component_size = mdi->component_size; | |
477 | a->info.consistency_policy = mdi->consistency_policy; | |
478 | sysfs_free(mdi); | |
479 | } | |
4edb8530 | 480 | |
bc77ed53 DW |
481 | /* honor 'frozen' */ |
482 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
483 | frozen = buf[9] == '-'; | |
484 | else | |
485 | frozen = 1; /* can't read metadata_version assume the worst */ | |
486 | ||
f54a6742 | 487 | /* If sync_action is not 'idle' then don't try recovery now */ |
fc54fe7a JS |
488 | if (!frozen && |
489 | sysfs_get_str(&a->info, NULL, "sync_action", | |
490 | buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0) | |
f54a6742 N |
491 | frozen = 1; |
492 | ||
57f8c769 AK |
493 | if (mdstat->level) { |
494 | int level = map_name(pers, mdstat->level); | |
7023e0b8 | 495 | if (level == 0 || level == LEVEL_LINEAR) { |
ba714450 | 496 | a->to_remove = 1; |
84f3857f | 497 | wakeup_monitor(); |
7023e0b8 N |
498 | return; |
499 | } | |
500 | else if (a->info.array.level != level && level > 0) { | |
57f8c769 AK |
501 | struct active_array *newa = duplicate_aa(a); |
502 | if (newa) { | |
503 | newa->info.array.level = level; | |
4e2c1a9a | 504 | replace_array(container, a, newa); |
57f8c769 AK |
505 | a = newa; |
506 | } | |
507 | } | |
508 | } | |
509 | ||
50927b13 AK |
510 | /* we are after monitor kick, |
511 | * so container field can be cleared - check it again | |
512 | */ | |
513 | if (a->container == NULL) | |
514 | return; | |
515 | ||
c76242c5 TM |
516 | if (sigterm && a->info.safe_mode_delay != 1 && |
517 | a->safe_mode_delay_fd >= 0) { | |
518 | long int new_delay = 1; | |
519 | char delay[10]; | |
520 | ssize_t len; | |
521 | ||
522 | len = snprintf(delay, sizeof(delay), "0.%03ld\n", new_delay); | |
523 | if (write(a->safe_mode_delay_fd, delay, len) == len) | |
524 | a->info.safe_mode_delay = new_delay; | |
2ef21963 N |
525 | } |
526 | ||
0c4f6e37 | 527 | /* We don't check the array while any update is pending, as it |
88b496c2 | 528 | * might container a change (such as a spare assignment) which |
0c4f6e37 N |
529 | * could affect our decisions. |
530 | */ | |
88b496c2 | 531 | if (a->check_degraded && !frozen && |
0c4f6e37 | 532 | update_queue == NULL && update_queue_pending == NULL) { |
6c3fb95c | 533 | struct metadata_update *updates = NULL; |
071cfc42 | 534 | struct mdinfo *newdev = NULL; |
6c3fb95c | 535 | struct active_array *newa; |
071cfc42 | 536 | struct mdinfo *d; |
3c00ffbe | 537 | |
6c3fb95c NB |
538 | a->check_degraded = 0; |
539 | ||
540 | /* The array may not be degraded, this is just a good time | |
541 | * to check. | |
542 | */ | |
4e2c1a9a | 543 | newdev = container->ss->activate_spare(a, &updates); |
071cfc42 DW |
544 | if (!newdev) |
545 | return; | |
546 | ||
547 | newa = duplicate_aa(a); | |
548 | if (!newa) | |
549 | goto out; | |
1d446d52 DW |
550 | /* prevent the kernel from activating the disk(s) before we |
551 | * finish adding them | |
552 | */ | |
1ade5cc1 | 553 | dprintf("freezing %s\n", a->info.sys_name); |
1d446d52 | 554 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); |
071cfc42 DW |
555 | |
556 | /* Add device to array and set offset/size/slot. | |
557 | * and open files for each newdev */ | |
558 | for (d = newdev; d ; d = d->next) { | |
559 | struct mdinfo *newd; | |
560 | ||
503975b9 | 561 | newd = xmalloc(sizeof(*newd)); |
2904b26f | 562 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
563 | free(newd); |
564 | continue; | |
6c3fb95c | 565 | } |
63b4aae3 | 566 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
567 | } |
568 | queue_metadata_update(updates); | |
569 | updates = NULL; | |
6ca1e6ec MW |
570 | while (update_queue_pending || update_queue) { |
571 | check_update_queue(container); | |
572 | usleep(15*1000); | |
573 | } | |
4e2c1a9a | 574 | replace_array(container, a, newa); |
b831b299 JS |
575 | if (sysfs_set_str(&a->info, NULL, |
576 | "sync_action", "recover") == 0) | |
6ca1e6ec | 577 | newa->prev_action = recover; |
1ade5cc1 | 578 | dprintf("recovery started on %s\n", a->info.sys_name); |
071cfc42 DW |
579 | out: |
580 | while (newdev) { | |
581 | d = newdev->next; | |
582 | free(newdev); | |
583 | newdev = d; | |
6c3fb95c | 584 | } |
071cfc42 | 585 | free_updates(&updates); |
6c3fb95c | 586 | } |
0f99b4bd N |
587 | |
588 | if (a->check_reshape) { | |
589 | /* mdadm might have added some devices to the array. | |
590 | * We want to disk_init_and_add any such device to a | |
591 | * duplicate_aa and replace a with that. | |
592 | * mdstat doesn't have enough info so we sysfs_read | |
593 | * and look for new stuff. | |
594 | */ | |
595 | struct mdinfo *info, *d, *d2, *newd; | |
aad6f216 | 596 | unsigned long long array_size; |
0f99b4bd N |
597 | struct active_array *newa = NULL; |
598 | a->check_reshape = 0; | |
4dd2df09 | 599 | info = sysfs_read(-1, mdstat->devnm, |
0f99b4bd N |
600 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
601 | if (!info) | |
602 | goto out2; | |
603 | for (d = info->devs; d; d = d->next) { | |
604 | if (d->disk.raid_disk < 0) | |
605 | continue; | |
606 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
607 | if (d2->disk.raid_disk == | |
608 | d->disk.raid_disk) | |
609 | break; | |
610 | if (d2) | |
611 | /* already have this one */ | |
612 | continue; | |
613 | if (!newa) { | |
614 | newa = duplicate_aa(a); | |
615 | if (!newa) | |
616 | break; | |
617 | } | |
503975b9 | 618 | newd = xmalloc(sizeof(*newd)); |
0f99b4bd N |
619 | disk_init_and_add(newd, d, newa); |
620 | } | |
fc54fe7a JS |
621 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 && |
622 | a->info.custom_array_size > array_size*2) { | |
aad6f216 | 623 | sysfs_set_num(info, NULL, "array_size", |
02eedb57 | 624 | a->info.custom_array_size/2); |
aad6f216 | 625 | } |
0f99b4bd N |
626 | out2: |
627 | sysfs_free(info); | |
628 | if (newa) | |
4e2c1a9a | 629 | replace_array(container, a, newa); |
0f99b4bd | 630 | } |
549e9569 NB |
631 | } |
632 | ||
836759d5 DW |
633 | static int aa_ready(struct active_array *aa) |
634 | { | |
635 | struct mdinfo *d; | |
636 | int level = aa->info.array.level; | |
637 | ||
638 | for (d = aa->info.devs; d; d = d->next) | |
639 | if (d->state_fd < 0) | |
640 | return 0; | |
641 | ||
642 | if (aa->info.state_fd < 0) | |
643 | return 0; | |
644 | ||
645 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
646 | return 0; | |
647 | ||
648 | if (!aa->container) | |
649 | return 0; | |
650 | ||
651 | return 1; | |
652 | } | |
653 | ||
549e9569 | 654 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
655 | struct supertype *container, |
656 | struct active_array *victim) | |
549e9569 NB |
657 | { |
658 | /* A new array has appeared in this container. | |
659 | * Hopefully it is already recorded in the metadata. | |
660 | * Check, then create the new array to report it to | |
661 | * the monitor. | |
662 | */ | |
663 | ||
664 | struct active_array *new; | |
665 | struct mdinfo *mdi, *di; | |
cba0191b | 666 | char *inst; |
549e9569 | 667 | int i; |
f1d26766 | 668 | int failed = 0; |
138477db | 669 | char buf[40]; |
549e9569 | 670 | |
836759d5 | 671 | /* check if array is ready to be monitored */ |
7023e0b8 N |
672 | if (!mdstat->active || !mdstat->level) |
673 | return; | |
674 | if (strcmp(mdstat->level, "raid0") == 0 || | |
675 | strcmp(mdstat->level, "linear") == 0) | |
836759d5 DW |
676 | return; |
677 | ||
4dd2df09 | 678 | mdi = sysfs_read(-1, mdstat->devnm, |
836759d5 | 679 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| |
b13b52c8 TM |
680 | GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| |
681 | GET_LAYOUT); | |
836759d5 | 682 | |
503975b9 | 683 | if (!mdi) |
836759d5 | 684 | return; |
503975b9 | 685 | new = xcalloc(1, sizeof(*new)); |
d52690ac | 686 | |
4dd2df09 | 687 | strcpy(new->info.sys_name, mdstat->devnm); |
549e9569 NB |
688 | |
689 | new->prev_state = new->curr_state = new->next_state = inactive; | |
690 | new->prev_action= new->curr_action= new->next_action= idle; | |
691 | ||
692 | new->container = container; | |
693 | ||
4dd2df09 | 694 | inst = to_subarray(mdstat, container->devnm); |
549e9569 | 695 | |
549e9569 | 696 | new->info.array = mdi->array; |
272bcc48 | 697 | new->info.component_size = mdi->component_size; |
549e9569 NB |
698 | |
699 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
503975b9 | 700 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
549e9569 NB |
701 | |
702 | for (di = mdi->devs; di; di = di->next) | |
703 | if (i == di->disk.raid_disk) | |
704 | break; | |
705 | ||
63b4aae3 | 706 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
707 | if (newd) |
708 | free(newd); | |
709 | ||
f1d26766 | 710 | failed++; |
7da80e6f DW |
711 | if (failed > new->info.array.failed_disks) { |
712 | /* we cannot properly monitor without all working disks */ | |
713 | new->container = NULL; | |
714 | break; | |
715 | } | |
549e9569 | 716 | } |
549e9569 | 717 | } |
836759d5 | 718 | |
da5a36fa N |
719 | new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action"); |
720 | new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state"); | |
721 | new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start"); | |
722 | new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version"); | |
723 | new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed"); | |
c76242c5 TM |
724 | new->safe_mode_delay_fd = sysfs_open2(new->info.sys_name, NULL, |
725 | "safe_mode_delay"); | |
da5a36fa | 726 | |
1ade5cc1 | 727 | dprintf("inst: %s action: %d state: %d\n", inst, |
4e6e574a | 728 | new->action_fd, new->info.state_fd); |
549e9569 | 729 | |
2ef21963 N |
730 | if (sigterm) |
731 | new->info.safe_mode_delay = 1; | |
732 | else if (mdi->safe_mode_delay >= 50) | |
733 | /* Normal start, mdadm set this. */ | |
734 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
735 | else | |
736 | /* Restart, just pick a number */ | |
737 | new->info.safe_mode_delay = 5000; | |
738 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
739 | ||
138477db AK |
740 | /* reshape_position is set by mdadm in sysfs |
741 | * read this information for new arrays only (empty victim) | |
742 | */ | |
743 | if ((victim == NULL) && | |
744 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
745 | (strncmp(buf, "reshape", 7) == 0)) { | |
746 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
747 | &new->last_checkpoint) != 0) | |
748 | new->last_checkpoint = 0; | |
0d51bfa2 AK |
749 | else { |
750 | int data_disks = mdi->array.raid_disks; | |
751 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
752 | data_disks--; | |
753 | if (mdi->array.level == 6) | |
754 | data_disks -= 2; | |
755 | ||
756 | new->last_checkpoint /= data_disks; | |
757 | } | |
138477db AK |
758 | dprintf("mdmon: New monitored array is under reshape.\n" |
759 | " Last checkpoint is: %llu\n", | |
760 | new->last_checkpoint); | |
761 | } | |
762 | ||
4fa5aef9 | 763 | sysfs_free(mdi); |
836759d5 DW |
764 | |
765 | /* if everything checks out tell the metadata handler we want to | |
766 | * manage this instance | |
767 | */ | |
768 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
a88e119f | 769 | pr_err("failed to monitor %s\n", |
836759d5 | 770 | mdstat->metadata_version); |
549e9569 | 771 | new->container = NULL; |
836759d5 | 772 | free_aa(new); |
93f7caca | 773 | } else { |
2a0bb19e | 774 | replace_array(container, victim, new); |
93f7caca DW |
775 | if (failed) { |
776 | new->check_degraded = 1; | |
777 | manage_member(mdstat, new); | |
778 | } | |
779 | } | |
549e9569 NB |
780 | } |
781 | ||
5d19760d | 782 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
783 | { |
784 | /* We have just read mdstat and need to compare it with | |
785 | * the known active arrays. | |
786 | * Arrays with the wrong metadata are ignored. | |
787 | */ | |
788 | ||
789 | for ( ; mdstat ; mdstat = mdstat->next) { | |
790 | struct active_array *a; | |
4dd2df09 | 791 | if (strcmp(mdstat->devnm, container->devnm) == 0) { |
549e9569 NB |
792 | manage_container(mdstat, container); |
793 | continue; | |
794 | } | |
4dd2df09 | 795 | if (!is_container_member(mdstat, container->devnm)) |
549e9569 NB |
796 | /* Not for this array */ |
797 | continue; | |
798 | /* Looks like a member of this container */ | |
5d19760d | 799 | for (a = container->arrays; a; a = a->next) { |
4dd2df09 | 800 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { |
ba714450 | 801 | if (a->container && a->to_remove == 0) |
549e9569 NB |
802 | manage_member(mdstat, a); |
803 | break; | |
804 | } | |
805 | } | |
2a0bb19e DW |
806 | if (a == NULL || !a->container) |
807 | manage_new(mdstat, container, a); | |
549e9569 NB |
808 | } |
809 | } | |
810 | ||
edd8d13c | 811 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 812 | { |
edd8d13c NB |
813 | /* queue this metadata update through to the monitor */ |
814 | ||
815 | struct metadata_update *mu; | |
816 | ||
313a4a82 | 817 | if (msg->len <= 0) |
3c00ffbe N |
818 | while (update_queue_pending || update_queue) { |
819 | check_update_queue(container); | |
820 | usleep(15*1000); | |
821 | } | |
822 | ||
313a4a82 DW |
823 | if (msg->len == 0) { /* ping_monitor */ |
824 | int cnt; | |
1011e834 | 825 | |
3c00ffbe | 826 | cnt = monitor_loop_cnt; |
1eb252b8 N |
827 | if (cnt & 1) |
828 | cnt += 2; /* wait until next pselect */ | |
829 | else | |
830 | cnt += 3; /* wait for 2 pselects */ | |
831 | wakeup_monitor(); | |
3c00ffbe | 832 | |
1eb252b8 N |
833 | while (monitor_loop_cnt - cnt < 0) |
834 | usleep(10 * 1000); | |
313a4a82 DW |
835 | } else if (msg->len == -1) { /* ping_manager */ |
836 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
837 | ||
838 | manage(mdstat, container); | |
839 | free_mdstat(mdstat); | |
6144ed44 | 840 | } else if (!sigterm) { |
503975b9 | 841 | mu = xmalloc(sizeof(*mu)); |
edd8d13c NB |
842 | mu->len = msg->len; |
843 | mu->buf = msg->buf; | |
844 | msg->buf = NULL; | |
845 | mu->space = NULL; | |
cb23f1f4 | 846 | mu->space_list = NULL; |
edd8d13c NB |
847 | mu->next = NULL; |
848 | if (container->ss->prepare_update) | |
5fe6f031 N |
849 | if (!container->ss->prepare_update(container, mu)) |
850 | free_updates(&mu); | |
edd8d13c NB |
851 | queue_metadata_update(mu); |
852 | } | |
3e70c845 DW |
853 | } |
854 | ||
855 | void read_sock(struct supertype *container) | |
549e9569 NB |
856 | { |
857 | int fd; | |
bfa44e2e | 858 | struct metadata_update msg; |
b109d928 DW |
859 | int terminate = 0; |
860 | long fl; | |
861 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 862 | |
3e70c845 | 863 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
864 | if (fd < 0) |
865 | return; | |
b109d928 DW |
866 | |
867 | fl = fcntl(fd, F_GETFL, 0); | |
868 | fl |= O_NONBLOCK; | |
869 | fcntl(fd, F_SETFL, fl); | |
870 | ||
871 | do { | |
872 | msg.buf = NULL; | |
873 | ||
874 | /* read and validate the message */ | |
875 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e | 876 | handle_message(container, &msg); |
bc77ed53 DW |
877 | if (msg.len == 0) { |
878 | /* ping reply with version */ | |
879 | msg.buf = Version; | |
880 | msg.len = strlen(Version) + 1; | |
881 | if (send_message(fd, &msg, tmo) < 0) | |
882 | terminate = 1; | |
883 | } else if (ack(fd, tmo) < 0) | |
bfa44e2e NB |
884 | terminate = 1; |
885 | } else | |
b109d928 | 886 | terminate = 1; |
b109d928 | 887 | |
b109d928 DW |
888 | } while (!terminate); |
889 | ||
549e9569 NB |
890 | close(fd); |
891 | } | |
1ed3f387 | 892 | |
e0d6609f NB |
893 | int exit_now = 0; |
894 | int manager_ready = 0; | |
549e9569 NB |
895 | void do_manager(struct supertype *container) |
896 | { | |
897 | struct mdstat_ent *mdstat; | |
4d43913c | 898 | sigset_t set; |
1ed3f387 | 899 | |
4d43913c NB |
900 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
901 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 902 | sigdelset(&set, SIGTERM); |
549e9569 NB |
903 | |
904 | do { | |
1ed3f387 | 905 | |
e0d6609f NB |
906 | if (exit_now) |
907 | exit(0); | |
908 | ||
3c00ffbe N |
909 | /* Can only 'manage' things if 'monitor' is not making |
910 | * structural changes to metadata, so need to check | |
911 | * update_queue | |
912 | */ | |
913 | if (update_queue == NULL) { | |
914 | mdstat = mdstat_read(1, 0); | |
549e9569 | 915 | |
3c00ffbe | 916 | manage(mdstat, container); |
549e9569 | 917 | |
3c00ffbe | 918 | read_sock(container); |
4fa5aef9 | 919 | |
3c00ffbe N |
920 | free_mdstat(mdstat); |
921 | } | |
1ed3f387 NB |
922 | remove_old(); |
923 | ||
2e735d19 NB |
924 | check_update_queue(container); |
925 | ||
e0d6609f | 926 | manager_ready = 1; |
4d43913c | 927 | |
6144ed44 DW |
928 | if (sigterm) |
929 | wakeup_monitor(); | |
930 | ||
5d4d1b26 | 931 | if (update_queue == NULL) |
58a4ba2a | 932 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 933 | else |
3c00ffbe N |
934 | /* If an update is happening, just wait for signal */ |
935 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
936 | } while(1); |
937 | } |