]>
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); | |
2a0bb19e DW |
132 | } |
133 | ||
549e9569 NB |
134 | static void free_aa(struct active_array *aa) |
135 | { | |
2a0bb19e DW |
136 | /* Note that this doesn't close fds if they are being used |
137 | * by a clone. ->container will be set for a clone | |
549e9569 | 138 | */ |
1ade5cc1 | 139 | dprintf("sys_name: %s\n", aa->info.sys_name); |
2a0bb19e DW |
140 | if (!aa->container) |
141 | close_aa(aa); | |
549e9569 NB |
142 | while (aa->info.devs) { |
143 | struct mdinfo *d = aa->info.devs; | |
144 | aa->info.devs = d->next; | |
145 | free(d); | |
146 | } | |
147 | free(aa); | |
148 | } | |
149 | ||
6c3fb95c NB |
150 | static struct active_array *duplicate_aa(struct active_array *aa) |
151 | { | |
503975b9 | 152 | struct active_array *newa = xmalloc(sizeof(*newa)); |
6c3fb95c NB |
153 | struct mdinfo **dp1, **dp2; |
154 | ||
155 | *newa = *aa; | |
156 | newa->next = NULL; | |
157 | newa->replaces = NULL; | |
158 | newa->info.next = NULL; | |
159 | ||
160 | dp2 = &newa->info.devs; | |
161 | ||
162 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
163 | struct mdinfo *d; | |
164 | if ((*dp1)->state_fd < 0) | |
165 | continue; | |
166 | ||
503975b9 | 167 | d = xmalloc(sizeof(*d)); |
6c3fb95c NB |
168 | *d = **dp1; |
169 | *dp2 = d; | |
170 | dp2 = & d->next; | |
171 | } | |
7e1432fb | 172 | *dp2 = NULL; |
6c3fb95c NB |
173 | |
174 | return newa; | |
175 | } | |
176 | ||
4d43913c | 177 | static void wakeup_monitor(void) |
2a0bb19e | 178 | { |
4d43913c NB |
179 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
180 | int pid = getpid(); | |
181 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
182 | } |
183 | ||
1ed3f387 NB |
184 | static void remove_old(void) |
185 | { | |
186 | if (discard_this) { | |
187 | discard_this->next = NULL; | |
188 | free_aa(discard_this); | |
189 | if (pending_discard == discard_this) | |
190 | pending_discard = NULL; | |
191 | discard_this = NULL; | |
48561b01 | 192 | wakeup_monitor(); |
1ed3f387 NB |
193 | } |
194 | } | |
195 | ||
549e9569 NB |
196 | static void replace_array(struct supertype *container, |
197 | struct active_array *old, | |
198 | struct active_array *new) | |
199 | { | |
200 | /* To replace an array, we add it to the top of the list | |
201 | * marked with ->replaces to point to the original. | |
202 | * 'monitor' will take the original out of the list | |
203 | * and put it on 'discard_this'. We take it from there | |
204 | * and discard it. | |
205 | */ | |
1ed3f387 | 206 | remove_old(); |
549e9569 NB |
207 | while (pending_discard) { |
208 | while (discard_this == NULL) | |
209 | sleep(1); | |
1ed3f387 | 210 | remove_old(); |
549e9569 NB |
211 | } |
212 | pending_discard = old; | |
213 | new->replaces = old; | |
214 | new->next = container->arrays; | |
215 | container->arrays = new; | |
4d43913c | 216 | wakeup_monitor(); |
549e9569 NB |
217 | } |
218 | ||
2e735d19 NB |
219 | struct metadata_update *update_queue = NULL; |
220 | struct metadata_update *update_queue_handled = NULL; | |
221 | struct metadata_update *update_queue_pending = NULL; | |
222 | ||
071cfc42 | 223 | static void free_updates(struct metadata_update **update) |
2e735d19 | 224 | { |
071cfc42 DW |
225 | while (*update) { |
226 | struct metadata_update *this = *update; | |
cb23f1f4 | 227 | void **space_list = this->space_list; |
071cfc42 DW |
228 | |
229 | *update = this->next; | |
904c1ef7 | 230 | free(this->buf); |
071cfc42 | 231 | free(this->space); |
cb23f1f4 N |
232 | while (space_list) { |
233 | void *space = space_list; | |
234 | space_list = *space_list; | |
235 | free(space); | |
236 | } | |
2e735d19 NB |
237 | free(this); |
238 | } | |
071cfc42 DW |
239 | } |
240 | ||
241 | void check_update_queue(struct supertype *container) | |
242 | { | |
243 | free_updates(&update_queue_handled); | |
244 | ||
2e735d19 NB |
245 | if (update_queue == NULL && |
246 | update_queue_pending) { | |
247 | update_queue = update_queue_pending; | |
248 | update_queue_pending = NULL; | |
4d43913c | 249 | wakeup_monitor(); |
2e735d19 NB |
250 | } |
251 | } | |
252 | ||
6c3fb95c | 253 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
254 | { |
255 | struct metadata_update **qp; | |
256 | ||
257 | qp = &update_queue_pending; | |
258 | while (*qp) | |
259 | qp = & ((*qp)->next); | |
260 | *qp = mu; | |
261 | } | |
262 | ||
43dad3d6 DW |
263 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
264 | { | |
265 | int dfd; | |
266 | char nm[20]; | |
661dce36 | 267 | struct supertype *st2; |
43dad3d6 | 268 | struct metadata_update *update = NULL; |
661dce36 | 269 | struct mdinfo info; |
43dad3d6 DW |
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 | ||
661dce36 N |
288 | /* Check the metadata and see if it is already part of this |
289 | * array | |
290 | */ | |
291 | st2 = dup_super(st); | |
292 | if (st2->ss->load_super(st2, dfd, NULL) == 0) { | |
4389b648 | 293 | st2->ss->getinfo_super(st2, &info, NULL); |
661dce36 N |
294 | if (st->ss->compare_super(st, st2) == 0 && |
295 | info.disk.raid_disk >= 0) { | |
296 | /* Looks like a good member of array. | |
297 | * Just accept it. | |
298 | * mdadm will incorporate any parts into | |
299 | * active arrays. | |
300 | */ | |
301 | st2->ss->free_super(st2); | |
302 | return; | |
303 | } | |
304 | } | |
305 | st2->ss->free_super(st2); | |
306 | ||
43dad3d6 | 307 | st->update_tail = &update; |
72ca9bcf | 308 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); |
43dad3d6 DW |
309 | st->ss->write_init_super(st); |
310 | queue_metadata_update(update); | |
311 | st->update_tail = NULL; | |
312 | } | |
313 | ||
1a64be56 LM |
314 | /* |
315 | * Create and queue update structure about the removed disks. | |
316 | * The update is prepared by super type handler and passed to the monitor | |
317 | * thread. | |
318 | */ | |
319 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
320 | { | |
321 | struct metadata_update *update = NULL; | |
322 | mdu_disk_info_t dk = { | |
323 | .number = -1, | |
324 | .major = sd->disk.major, | |
325 | .minor = sd->disk.minor, | |
326 | .raid_disk = -1, | |
327 | .state = 0, | |
328 | }; | |
1ade5cc1 N |
329 | dprintf("remove %d:%d from container\n", |
330 | sd->disk.major, sd->disk.minor); | |
1a64be56 LM |
331 | |
332 | st->update_tail = &update; | |
333 | st->ss->remove_from_super(st, &dk); | |
4dd968cc N |
334 | /* FIXME this write_init_super shouldn't be here. |
335 | * We have it after add_to_super to write to new device, | |
336 | * but with 'remove' we don't ant to write to that device! | |
337 | */ | |
1a64be56 LM |
338 | st->ss->write_init_super(st); |
339 | queue_metadata_update(update); | |
340 | st->update_tail = NULL; | |
341 | } | |
342 | ||
549e9569 NB |
343 | static void manage_container(struct mdstat_ent *mdstat, |
344 | struct supertype *container) | |
345 | { | |
1a64be56 | 346 | /* Of interest here are: |
1011e834 | 347 | * - if a new device has been added to the container, we |
1a64be56 LM |
348 | * add it to the array ignoring any metadata on it. |
349 | * - if a device has been removed from the container, we | |
350 | * remove it from the device list and update the metadata. | |
549e9569 NB |
351 | * FIXME should we look for compatible metadata and take hints |
352 | * about spare assignment.... probably not. | |
549e9569 NB |
353 | */ |
354 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
355 | struct mdinfo **cdp, *cd, *di, *mdi; |
356 | int found; | |
357 | ||
549e9569 NB |
358 | /* read /sys/block/NAME/md/dev-??/block/dev to find out |
359 | * what is there, and compare with container->info.devs | |
360 | * To see what is removed and what is added. | |
361 | * These need to be remove from, or added to, the array | |
362 | */ | |
4dd2df09 | 363 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); |
313a4a82 DW |
364 | if (!mdi) { |
365 | /* invalidate the current count so we can try again */ | |
366 | container->devcnt = -1; | |
7bc1962f | 367 | return; |
313a4a82 | 368 | } |
7bc1962f DW |
369 | |
370 | /* check for removals */ | |
371 | for (cdp = &container->devs; *cdp; ) { | |
372 | found = 0; | |
373 | for (di = mdi->devs; di; di = di->next) | |
374 | if (di->disk.major == (*cdp)->disk.major && | |
375 | di->disk.minor == (*cdp)->disk.minor) { | |
376 | found = 1; | |
377 | break; | |
378 | } | |
379 | if (!found) { | |
380 | cd = *cdp; | |
381 | *cdp = (*cdp)->next; | |
1a64be56 | 382 | remove_disk_from_container(container, cd); |
7bc1962f DW |
383 | free(cd); |
384 | } else | |
385 | cdp = &(*cdp)->next; | |
386 | } | |
43dad3d6 DW |
387 | |
388 | /* check for additions */ | |
389 | for (di = mdi->devs; di; di = di->next) { | |
390 | for (cd = container->devs; cd; cd = cd->next) | |
391 | if (di->disk.major == cd->disk.major && | |
392 | di->disk.minor == cd->disk.minor) | |
393 | break; | |
04a8ac08 | 394 | if (!cd) { |
503975b9 | 395 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
04a8ac08 | 396 | |
04a8ac08 DW |
397 | *newd = *di; |
398 | add_disk_to_container(container, newd); | |
399 | } | |
43dad3d6 | 400 | } |
7bc1962f | 401 | sysfs_free(mdi); |
549e9569 NB |
402 | container->devcnt = mdstat->devcnt; |
403 | } | |
404 | } | |
405 | ||
da5a36fa N |
406 | static int sysfs_open2(char *devnum, char *name, char *attr) |
407 | { | |
408 | int fd = sysfs_open(devnum, name, attr); | |
409 | if (fd >= 0) { | |
410 | /* seq_file in the kernel allocates buffer space | |
411 | * on the first read. Do that now so 'monitor' | |
412 | * never needs too. | |
413 | */ | |
414 | char buf[200]; | |
27aefbdb N |
415 | if (read(fd, buf, sizeof(buf)) < 0) |
416 | /* pretend not to ignore return value */ | |
417 | return fd; | |
da5a36fa N |
418 | } |
419 | return fd; | |
420 | } | |
421 | ||
63b4aae3 DW |
422 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, |
423 | struct active_array *aa) | |
424 | { | |
425 | if (!disk || !clone) | |
426 | return -1; | |
427 | ||
428 | *disk = *clone; | |
da5a36fa N |
429 | disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
430 | "recovery_start"); | |
3e1d79b2 JS |
431 | if (disk->recovery_fd < 0) |
432 | return -1; | |
da5a36fa | 433 | disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state"); |
3e1d79b2 JS |
434 | if (disk->state_fd < 0) { |
435 | close(disk->recovery_fd); | |
436 | return -1; | |
437 | } | |
6dc1785f TM |
438 | disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
439 | "bad_blocks"); | |
440 | if (disk->bb_fd < 0) { | |
441 | close(disk->recovery_fd); | |
442 | close(disk->state_fd); | |
443 | return -1; | |
444 | } | |
445 | disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
446 | "unacknowledged_bad_blocks"); | |
447 | if (disk->ubb_fd < 0) { | |
448 | close(disk->recovery_fd); | |
449 | close(disk->state_fd); | |
450 | close(disk->bb_fd); | |
451 | return -1; | |
452 | } | |
63b4aae3 DW |
453 | disk->prev_state = read_dev_state(disk->state_fd); |
454 | disk->curr_state = disk->prev_state; | |
455 | disk->next = aa->info.devs; | |
456 | aa->info.devs = disk; | |
457 | ||
458 | return 0; | |
459 | } | |
460 | ||
549e9569 NB |
461 | static void manage_member(struct mdstat_ent *mdstat, |
462 | struct active_array *a) | |
463 | { | |
464 | /* Compare mdstat info with known state of member array. | |
465 | * We do not need to look for device state changes here, that | |
466 | * is dealt with by the monitor. | |
467 | * | |
0f99b4bd N |
468 | * If a reshape is being requested, monitor will have noticed |
469 | * that sync_action changed and will have set check_reshape. | |
470 | * We just need to see if new devices have appeared. All metadata | |
471 | * updates will already have been processed. | |
6c3fb95c | 472 | * |
0f99b4bd | 473 | * We also want to handle degraded arrays here by |
6c3fb95c NB |
474 | * trying to find and assign a spare. |
475 | * We do that whenever the monitor tells us too. | |
549e9569 | 476 | */ |
bc77ed53 DW |
477 | char buf[64]; |
478 | int frozen; | |
4e2c1a9a | 479 | struct supertype *container = a->container; |
4edb8530 | 480 | unsigned long long int component_size = 0; |
4e2c1a9a N |
481 | |
482 | if (container == NULL) | |
483 | /* Raced with something */ | |
484 | return; | |
bc77ed53 | 485 | |
e49a8a80 N |
486 | if (mdstat->active) { |
487 | // FIXME | |
488 | a->info.array.raid_disks = mdstat->raid_disks; | |
489 | // MORE | |
490 | } | |
549e9569 | 491 | |
4edb8530 PB |
492 | if (sysfs_get_ll(&a->info, NULL, "component_size", &component_size) >= 0) |
493 | a->info.component_size = component_size << 1; | |
494 | ||
bc77ed53 DW |
495 | /* honor 'frozen' */ |
496 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
497 | frozen = buf[9] == '-'; | |
498 | else | |
499 | frozen = 1; /* can't read metadata_version assume the worst */ | |
500 | ||
f54a6742 | 501 | /* If sync_action is not 'idle' then don't try recovery now */ |
fc54fe7a JS |
502 | if (!frozen && |
503 | sysfs_get_str(&a->info, NULL, "sync_action", | |
504 | buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0) | |
f54a6742 N |
505 | frozen = 1; |
506 | ||
57f8c769 AK |
507 | if (mdstat->level) { |
508 | int level = map_name(pers, mdstat->level); | |
7023e0b8 | 509 | if (level == 0 || level == LEVEL_LINEAR) { |
ba714450 | 510 | a->to_remove = 1; |
84f3857f | 511 | wakeup_monitor(); |
7023e0b8 N |
512 | return; |
513 | } | |
514 | else if (a->info.array.level != level && level > 0) { | |
57f8c769 AK |
515 | struct active_array *newa = duplicate_aa(a); |
516 | if (newa) { | |
517 | newa->info.array.level = level; | |
4e2c1a9a | 518 | replace_array(container, a, newa); |
57f8c769 AK |
519 | a = newa; |
520 | } | |
521 | } | |
522 | } | |
523 | ||
50927b13 AK |
524 | /* we are after monitor kick, |
525 | * so container field can be cleared - check it again | |
526 | */ | |
527 | if (a->container == NULL) | |
528 | return; | |
529 | ||
4e5e54cf | 530 | if (sigterm && a->info.safe_mode_delay != 1) { |
2ef21963 N |
531 | sysfs_set_safemode(&a->info, 1); |
532 | a->info.safe_mode_delay = 1; | |
533 | } | |
534 | ||
0c4f6e37 | 535 | /* We don't check the array while any update is pending, as it |
88b496c2 | 536 | * might container a change (such as a spare assignment) which |
0c4f6e37 N |
537 | * could affect our decisions. |
538 | */ | |
88b496c2 | 539 | if (a->check_degraded && !frozen && |
0c4f6e37 | 540 | update_queue == NULL && update_queue_pending == NULL) { |
6c3fb95c | 541 | struct metadata_update *updates = NULL; |
071cfc42 | 542 | struct mdinfo *newdev = NULL; |
6c3fb95c | 543 | struct active_array *newa; |
071cfc42 | 544 | struct mdinfo *d; |
3c00ffbe | 545 | |
6c3fb95c NB |
546 | a->check_degraded = 0; |
547 | ||
548 | /* The array may not be degraded, this is just a good time | |
549 | * to check. | |
550 | */ | |
4e2c1a9a | 551 | newdev = container->ss->activate_spare(a, &updates); |
071cfc42 DW |
552 | if (!newdev) |
553 | return; | |
554 | ||
555 | newa = duplicate_aa(a); | |
556 | if (!newa) | |
557 | goto out; | |
1d446d52 DW |
558 | /* prevent the kernel from activating the disk(s) before we |
559 | * finish adding them | |
560 | */ | |
1ade5cc1 | 561 | dprintf("freezing %s\n", a->info.sys_name); |
1d446d52 | 562 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); |
071cfc42 DW |
563 | |
564 | /* Add device to array and set offset/size/slot. | |
565 | * and open files for each newdev */ | |
566 | for (d = newdev; d ; d = d->next) { | |
567 | struct mdinfo *newd; | |
568 | ||
503975b9 | 569 | newd = xmalloc(sizeof(*newd)); |
2904b26f | 570 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
571 | free(newd); |
572 | continue; | |
6c3fb95c | 573 | } |
63b4aae3 | 574 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
575 | } |
576 | queue_metadata_update(updates); | |
577 | updates = NULL; | |
6ca1e6ec MW |
578 | while (update_queue_pending || update_queue) { |
579 | check_update_queue(container); | |
580 | usleep(15*1000); | |
581 | } | |
4e2c1a9a | 582 | replace_array(container, a, newa); |
b831b299 JS |
583 | if (sysfs_set_str(&a->info, NULL, |
584 | "sync_action", "recover") == 0) | |
6ca1e6ec | 585 | newa->prev_action = recover; |
1ade5cc1 | 586 | dprintf("recovery started on %s\n", a->info.sys_name); |
071cfc42 DW |
587 | out: |
588 | while (newdev) { | |
589 | d = newdev->next; | |
590 | free(newdev); | |
591 | newdev = d; | |
6c3fb95c | 592 | } |
071cfc42 | 593 | free_updates(&updates); |
6c3fb95c | 594 | } |
0f99b4bd N |
595 | |
596 | if (a->check_reshape) { | |
597 | /* mdadm might have added some devices to the array. | |
598 | * We want to disk_init_and_add any such device to a | |
599 | * duplicate_aa and replace a with that. | |
600 | * mdstat doesn't have enough info so we sysfs_read | |
601 | * and look for new stuff. | |
602 | */ | |
603 | struct mdinfo *info, *d, *d2, *newd; | |
aad6f216 | 604 | unsigned long long array_size; |
0f99b4bd N |
605 | struct active_array *newa = NULL; |
606 | a->check_reshape = 0; | |
4dd2df09 | 607 | info = sysfs_read(-1, mdstat->devnm, |
0f99b4bd N |
608 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
609 | if (!info) | |
610 | goto out2; | |
611 | for (d = info->devs; d; d = d->next) { | |
612 | if (d->disk.raid_disk < 0) | |
613 | continue; | |
614 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
615 | if (d2->disk.raid_disk == | |
616 | d->disk.raid_disk) | |
617 | break; | |
618 | if (d2) | |
619 | /* already have this one */ | |
620 | continue; | |
621 | if (!newa) { | |
622 | newa = duplicate_aa(a); | |
623 | if (!newa) | |
624 | break; | |
625 | } | |
503975b9 | 626 | newd = xmalloc(sizeof(*newd)); |
0f99b4bd N |
627 | disk_init_and_add(newd, d, newa); |
628 | } | |
fc54fe7a JS |
629 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 && |
630 | a->info.custom_array_size > array_size*2) { | |
aad6f216 | 631 | sysfs_set_num(info, NULL, "array_size", |
02eedb57 | 632 | a->info.custom_array_size/2); |
aad6f216 | 633 | } |
0f99b4bd N |
634 | out2: |
635 | sysfs_free(info); | |
636 | if (newa) | |
4e2c1a9a | 637 | replace_array(container, a, newa); |
0f99b4bd | 638 | } |
549e9569 NB |
639 | } |
640 | ||
836759d5 DW |
641 | static int aa_ready(struct active_array *aa) |
642 | { | |
643 | struct mdinfo *d; | |
644 | int level = aa->info.array.level; | |
645 | ||
646 | for (d = aa->info.devs; d; d = d->next) | |
647 | if (d->state_fd < 0) | |
648 | return 0; | |
649 | ||
650 | if (aa->info.state_fd < 0) | |
651 | return 0; | |
652 | ||
653 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
654 | return 0; | |
655 | ||
656 | if (!aa->container) | |
657 | return 0; | |
658 | ||
659 | return 1; | |
660 | } | |
661 | ||
549e9569 | 662 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
663 | struct supertype *container, |
664 | struct active_array *victim) | |
549e9569 NB |
665 | { |
666 | /* A new array has appeared in this container. | |
667 | * Hopefully it is already recorded in the metadata. | |
668 | * Check, then create the new array to report it to | |
669 | * the monitor. | |
670 | */ | |
671 | ||
672 | struct active_array *new; | |
673 | struct mdinfo *mdi, *di; | |
cba0191b | 674 | char *inst; |
549e9569 | 675 | int i; |
f1d26766 | 676 | int failed = 0; |
138477db | 677 | char buf[40]; |
549e9569 | 678 | |
836759d5 | 679 | /* check if array is ready to be monitored */ |
7023e0b8 N |
680 | if (!mdstat->active || !mdstat->level) |
681 | return; | |
682 | if (strcmp(mdstat->level, "raid0") == 0 || | |
683 | strcmp(mdstat->level, "linear") == 0) | |
836759d5 DW |
684 | return; |
685 | ||
4dd2df09 | 686 | mdi = sysfs_read(-1, mdstat->devnm, |
836759d5 | 687 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| |
b13b52c8 TM |
688 | GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| |
689 | GET_LAYOUT); | |
836759d5 | 690 | |
503975b9 | 691 | if (!mdi) |
836759d5 | 692 | return; |
503975b9 | 693 | new = xcalloc(1, sizeof(*new)); |
d52690ac | 694 | |
4dd2df09 | 695 | strcpy(new->info.sys_name, mdstat->devnm); |
549e9569 NB |
696 | |
697 | new->prev_state = new->curr_state = new->next_state = inactive; | |
698 | new->prev_action= new->curr_action= new->next_action= idle; | |
699 | ||
700 | new->container = container; | |
701 | ||
4dd2df09 | 702 | inst = to_subarray(mdstat, container->devnm); |
549e9569 | 703 | |
549e9569 | 704 | new->info.array = mdi->array; |
272bcc48 | 705 | new->info.component_size = mdi->component_size; |
549e9569 NB |
706 | |
707 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
503975b9 | 708 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
549e9569 NB |
709 | |
710 | for (di = mdi->devs; di; di = di->next) | |
711 | if (i == di->disk.raid_disk) | |
712 | break; | |
713 | ||
63b4aae3 | 714 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
715 | if (newd) |
716 | free(newd); | |
717 | ||
f1d26766 | 718 | failed++; |
7da80e6f DW |
719 | if (failed > new->info.array.failed_disks) { |
720 | /* we cannot properly monitor without all working disks */ | |
721 | new->container = NULL; | |
722 | break; | |
723 | } | |
549e9569 | 724 | } |
549e9569 | 725 | } |
836759d5 | 726 | |
da5a36fa N |
727 | new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action"); |
728 | new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state"); | |
729 | new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start"); | |
730 | new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version"); | |
731 | new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed"); | |
732 | ||
1ade5cc1 | 733 | dprintf("inst: %s action: %d state: %d\n", inst, |
4e6e574a | 734 | new->action_fd, new->info.state_fd); |
549e9569 | 735 | |
2ef21963 N |
736 | if (sigterm) |
737 | new->info.safe_mode_delay = 1; | |
738 | else if (mdi->safe_mode_delay >= 50) | |
739 | /* Normal start, mdadm set this. */ | |
740 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
741 | else | |
742 | /* Restart, just pick a number */ | |
743 | new->info.safe_mode_delay = 5000; | |
744 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
745 | ||
138477db AK |
746 | /* reshape_position is set by mdadm in sysfs |
747 | * read this information for new arrays only (empty victim) | |
748 | */ | |
749 | if ((victim == NULL) && | |
750 | (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) && | |
751 | (strncmp(buf, "reshape", 7) == 0)) { | |
752 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
753 | &new->last_checkpoint) != 0) | |
754 | new->last_checkpoint = 0; | |
0d51bfa2 AK |
755 | else { |
756 | int data_disks = mdi->array.raid_disks; | |
757 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
758 | data_disks--; | |
759 | if (mdi->array.level == 6) | |
760 | data_disks -= 2; | |
761 | ||
762 | new->last_checkpoint /= data_disks; | |
763 | } | |
138477db AK |
764 | dprintf("mdmon: New monitored array is under reshape.\n" |
765 | " Last checkpoint is: %llu\n", | |
766 | new->last_checkpoint); | |
767 | } | |
768 | ||
4fa5aef9 | 769 | sysfs_free(mdi); |
836759d5 DW |
770 | |
771 | /* if everything checks out tell the metadata handler we want to | |
772 | * manage this instance | |
773 | */ | |
774 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
a88e119f | 775 | pr_err("failed to monitor %s\n", |
836759d5 | 776 | mdstat->metadata_version); |
549e9569 | 777 | new->container = NULL; |
836759d5 | 778 | free_aa(new); |
93f7caca | 779 | } else { |
2a0bb19e | 780 | replace_array(container, victim, new); |
93f7caca DW |
781 | if (failed) { |
782 | new->check_degraded = 1; | |
783 | manage_member(mdstat, new); | |
784 | } | |
785 | } | |
549e9569 NB |
786 | } |
787 | ||
5d19760d | 788 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
789 | { |
790 | /* We have just read mdstat and need to compare it with | |
791 | * the known active arrays. | |
792 | * Arrays with the wrong metadata are ignored. | |
793 | */ | |
794 | ||
795 | for ( ; mdstat ; mdstat = mdstat->next) { | |
796 | struct active_array *a; | |
4dd2df09 | 797 | if (strcmp(mdstat->devnm, container->devnm) == 0) { |
549e9569 NB |
798 | manage_container(mdstat, container); |
799 | continue; | |
800 | } | |
4dd2df09 | 801 | if (!is_container_member(mdstat, container->devnm)) |
549e9569 NB |
802 | /* Not for this array */ |
803 | continue; | |
804 | /* Looks like a member of this container */ | |
5d19760d | 805 | for (a = container->arrays; a; a = a->next) { |
4dd2df09 | 806 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { |
ba714450 | 807 | if (a->container && a->to_remove == 0) |
549e9569 NB |
808 | manage_member(mdstat, a); |
809 | break; | |
810 | } | |
811 | } | |
2a0bb19e DW |
812 | if (a == NULL || !a->container) |
813 | manage_new(mdstat, container, a); | |
549e9569 NB |
814 | } |
815 | } | |
816 | ||
edd8d13c | 817 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 818 | { |
edd8d13c NB |
819 | /* queue this metadata update through to the monitor */ |
820 | ||
821 | struct metadata_update *mu; | |
822 | ||
313a4a82 | 823 | if (msg->len <= 0) |
3c00ffbe N |
824 | while (update_queue_pending || update_queue) { |
825 | check_update_queue(container); | |
826 | usleep(15*1000); | |
827 | } | |
828 | ||
313a4a82 DW |
829 | if (msg->len == 0) { /* ping_monitor */ |
830 | int cnt; | |
1011e834 | 831 | |
3c00ffbe | 832 | cnt = monitor_loop_cnt; |
1eb252b8 N |
833 | if (cnt & 1) |
834 | cnt += 2; /* wait until next pselect */ | |
835 | else | |
836 | cnt += 3; /* wait for 2 pselects */ | |
837 | wakeup_monitor(); | |
3c00ffbe | 838 | |
1eb252b8 N |
839 | while (monitor_loop_cnt - cnt < 0) |
840 | usleep(10 * 1000); | |
313a4a82 DW |
841 | } else if (msg->len == -1) { /* ping_manager */ |
842 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
843 | ||
844 | manage(mdstat, container); | |
845 | free_mdstat(mdstat); | |
6144ed44 | 846 | } else if (!sigterm) { |
503975b9 | 847 | mu = xmalloc(sizeof(*mu)); |
edd8d13c NB |
848 | mu->len = msg->len; |
849 | mu->buf = msg->buf; | |
850 | msg->buf = NULL; | |
851 | mu->space = NULL; | |
cb23f1f4 | 852 | mu->space_list = NULL; |
edd8d13c NB |
853 | mu->next = NULL; |
854 | if (container->ss->prepare_update) | |
5fe6f031 N |
855 | if (!container->ss->prepare_update(container, mu)) |
856 | free_updates(&mu); | |
edd8d13c NB |
857 | queue_metadata_update(mu); |
858 | } | |
3e70c845 DW |
859 | } |
860 | ||
861 | void read_sock(struct supertype *container) | |
549e9569 NB |
862 | { |
863 | int fd; | |
bfa44e2e | 864 | struct metadata_update msg; |
b109d928 DW |
865 | int terminate = 0; |
866 | long fl; | |
867 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 868 | |
3e70c845 | 869 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
870 | if (fd < 0) |
871 | return; | |
b109d928 DW |
872 | |
873 | fl = fcntl(fd, F_GETFL, 0); | |
874 | fl |= O_NONBLOCK; | |
875 | fcntl(fd, F_SETFL, fl); | |
876 | ||
877 | do { | |
878 | msg.buf = NULL; | |
879 | ||
880 | /* read and validate the message */ | |
881 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e | 882 | handle_message(container, &msg); |
bc77ed53 DW |
883 | if (msg.len == 0) { |
884 | /* ping reply with version */ | |
885 | msg.buf = Version; | |
886 | msg.len = strlen(Version) + 1; | |
887 | if (send_message(fd, &msg, tmo) < 0) | |
888 | terminate = 1; | |
889 | } else if (ack(fd, tmo) < 0) | |
bfa44e2e NB |
890 | terminate = 1; |
891 | } else | |
b109d928 | 892 | terminate = 1; |
b109d928 | 893 | |
b109d928 DW |
894 | } while (!terminate); |
895 | ||
549e9569 NB |
896 | close(fd); |
897 | } | |
1ed3f387 | 898 | |
e0d6609f NB |
899 | int exit_now = 0; |
900 | int manager_ready = 0; | |
549e9569 NB |
901 | void do_manager(struct supertype *container) |
902 | { | |
903 | struct mdstat_ent *mdstat; | |
4d43913c | 904 | sigset_t set; |
1ed3f387 | 905 | |
4d43913c NB |
906 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
907 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 908 | sigdelset(&set, SIGTERM); |
549e9569 NB |
909 | |
910 | do { | |
1ed3f387 | 911 | |
e0d6609f NB |
912 | if (exit_now) |
913 | exit(0); | |
914 | ||
3c00ffbe N |
915 | /* Can only 'manage' things if 'monitor' is not making |
916 | * structural changes to metadata, so need to check | |
917 | * update_queue | |
918 | */ | |
919 | if (update_queue == NULL) { | |
920 | mdstat = mdstat_read(1, 0); | |
549e9569 | 921 | |
3c00ffbe | 922 | manage(mdstat, container); |
549e9569 | 923 | |
3c00ffbe | 924 | read_sock(container); |
4fa5aef9 | 925 | |
3c00ffbe N |
926 | free_mdstat(mdstat); |
927 | } | |
1ed3f387 NB |
928 | remove_old(); |
929 | ||
2e735d19 NB |
930 | check_update_queue(container); |
931 | ||
e0d6609f | 932 | manager_ready = 1; |
4d43913c | 933 | |
6144ed44 DW |
934 | if (sigterm) |
935 | wakeup_monitor(); | |
936 | ||
5d4d1b26 | 937 | if (update_queue == NULL) |
58a4ba2a | 938 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 939 | else |
3c00ffbe N |
940 | /* If an update is happening, just wait for signal */ |
941 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
942 | } while(1); |
943 | } |