]>
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 NB |
108 | #include <sys/socket.h> |
109 | ||
2a0bb19e DW |
110 | static void close_aa(struct active_array *aa) |
111 | { | |
112 | struct mdinfo *d; | |
113 | ||
e1516be1 DW |
114 | for (d = aa->info.devs; d; d = d->next) { |
115 | close(d->recovery_fd); | |
2a0bb19e | 116 | close(d->state_fd); |
6dc1785f TM |
117 | close(d->bb_fd); |
118 | close(d->ubb_fd); | |
e1516be1 | 119 | } |
2a0bb19e | 120 | |
c2047875 JS |
121 | if (aa->action_fd >= 0) |
122 | close(aa->action_fd); | |
123 | if (aa->info.state_fd >= 0) | |
124 | close(aa->info.state_fd); | |
125 | if (aa->resync_start_fd >= 0) | |
126 | close(aa->resync_start_fd); | |
127 | if (aa->metadata_fd >= 0) | |
128 | close(aa->metadata_fd); | |
129 | if (aa->sync_completed_fd >= 0) | |
130 | close(aa->sync_completed_fd); | |
c76242c5 TM |
131 | if (aa->safe_mode_delay_fd >= 0) |
132 | close(aa->safe_mode_delay_fd); | |
2a0bb19e DW |
133 | } |
134 | ||
549e9569 NB |
135 | static void free_aa(struct active_array *aa) |
136 | { | |
2a0bb19e DW |
137 | /* Note that this doesn't close fds if they are being used |
138 | * by a clone. ->container will be set for a clone | |
549e9569 | 139 | */ |
1ade5cc1 | 140 | dprintf("sys_name: %s\n", aa->info.sys_name); |
2a0bb19e DW |
141 | if (!aa->container) |
142 | close_aa(aa); | |
549e9569 NB |
143 | while (aa->info.devs) { |
144 | struct mdinfo *d = aa->info.devs; | |
145 | aa->info.devs = d->next; | |
146 | free(d); | |
147 | } | |
148 | free(aa); | |
149 | } | |
150 | ||
6c3fb95c NB |
151 | static struct active_array *duplicate_aa(struct active_array *aa) |
152 | { | |
503975b9 | 153 | struct active_array *newa = xmalloc(sizeof(*newa)); |
6c3fb95c NB |
154 | struct mdinfo **dp1, **dp2; |
155 | ||
156 | *newa = *aa; | |
157 | newa->next = NULL; | |
158 | newa->replaces = NULL; | |
159 | newa->info.next = NULL; | |
160 | ||
161 | dp2 = &newa->info.devs; | |
162 | ||
163 | for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) { | |
164 | struct mdinfo *d; | |
165 | if ((*dp1)->state_fd < 0) | |
166 | continue; | |
167 | ||
503975b9 | 168 | d = xmalloc(sizeof(*d)); |
6c3fb95c NB |
169 | *d = **dp1; |
170 | *dp2 = d; | |
171 | dp2 = & d->next; | |
172 | } | |
7e1432fb | 173 | *dp2 = NULL; |
6c3fb95c NB |
174 | |
175 | return newa; | |
176 | } | |
177 | ||
4d43913c | 178 | static void wakeup_monitor(void) |
2a0bb19e | 179 | { |
4d43913c NB |
180 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
181 | int pid = getpid(); | |
182 | syscall(SYS_tgkill, pid, mon_tid, SIGUSR1); | |
2a0bb19e DW |
183 | } |
184 | ||
1ed3f387 NB |
185 | static void remove_old(void) |
186 | { | |
187 | if (discard_this) { | |
188 | discard_this->next = NULL; | |
189 | free_aa(discard_this); | |
190 | if (pending_discard == discard_this) | |
191 | pending_discard = NULL; | |
192 | discard_this = NULL; | |
48561b01 | 193 | wakeup_monitor(); |
1ed3f387 NB |
194 | } |
195 | } | |
196 | ||
549e9569 NB |
197 | static void replace_array(struct supertype *container, |
198 | struct active_array *old, | |
199 | struct active_array *new) | |
200 | { | |
201 | /* To replace an array, we add it to the top of the list | |
202 | * marked with ->replaces to point to the original. | |
203 | * 'monitor' will take the original out of the list | |
204 | * and put it on 'discard_this'. We take it from there | |
205 | * and discard it. | |
206 | */ | |
1ed3f387 | 207 | remove_old(); |
549e9569 NB |
208 | while (pending_discard) { |
209 | while (discard_this == NULL) | |
239b3cc0 | 210 | sleep_for(1, 0, true); |
1ed3f387 | 211 | remove_old(); |
549e9569 NB |
212 | } |
213 | pending_discard = old; | |
214 | new->replaces = old; | |
215 | new->next = container->arrays; | |
216 | container->arrays = new; | |
4d43913c | 217 | wakeup_monitor(); |
549e9569 NB |
218 | } |
219 | ||
2e735d19 NB |
220 | struct metadata_update *update_queue = NULL; |
221 | struct metadata_update *update_queue_handled = NULL; | |
222 | struct metadata_update *update_queue_pending = NULL; | |
223 | ||
071cfc42 | 224 | static void free_updates(struct metadata_update **update) |
2e735d19 | 225 | { |
071cfc42 DW |
226 | while (*update) { |
227 | struct metadata_update *this = *update; | |
cb23f1f4 | 228 | void **space_list = this->space_list; |
071cfc42 DW |
229 | |
230 | *update = this->next; | |
904c1ef7 | 231 | free(this->buf); |
071cfc42 | 232 | free(this->space); |
cb23f1f4 N |
233 | while (space_list) { |
234 | void *space = space_list; | |
235 | space_list = *space_list; | |
236 | free(space); | |
237 | } | |
2e735d19 NB |
238 | free(this); |
239 | } | |
071cfc42 DW |
240 | } |
241 | ||
242 | void check_update_queue(struct supertype *container) | |
243 | { | |
244 | free_updates(&update_queue_handled); | |
245 | ||
2e735d19 NB |
246 | if (update_queue == NULL && |
247 | update_queue_pending) { | |
248 | update_queue = update_queue_pending; | |
249 | update_queue_pending = NULL; | |
4d43913c | 250 | wakeup_monitor(); |
2e735d19 NB |
251 | } |
252 | } | |
253 | ||
6c3fb95c | 254 | static void queue_metadata_update(struct metadata_update *mu) |
2e735d19 NB |
255 | { |
256 | struct metadata_update **qp; | |
257 | ||
258 | qp = &update_queue_pending; | |
259 | while (*qp) | |
260 | qp = & ((*qp)->next); | |
261 | *qp = mu; | |
262 | } | |
263 | ||
43dad3d6 DW |
264 | static void add_disk_to_container(struct supertype *st, struct mdinfo *sd) |
265 | { | |
266 | int dfd; | |
267 | char nm[20]; | |
268 | struct metadata_update *update = NULL; | |
269 | mdu_disk_info_t dk = { | |
270 | .number = -1, | |
271 | .major = sd->disk.major, | |
272 | .minor = sd->disk.minor, | |
273 | .raid_disk = -1, | |
274 | .state = 0, | |
275 | }; | |
276 | ||
1ade5cc1 | 277 | dprintf("add %d:%d to container\n", sd->disk.major, sd->disk.minor); |
43dad3d6 | 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 | ||
287 | st->update_tail = &update; | |
72ca9bcf | 288 | st->ss->add_to_super(st, &dk, dfd, NULL, INVALID_SECTORS); |
43dad3d6 DW |
289 | st->ss->write_init_super(st); |
290 | queue_metadata_update(update); | |
291 | st->update_tail = NULL; | |
292 | } | |
293 | ||
1a64be56 LM |
294 | /* |
295 | * Create and queue update structure about the removed disks. | |
296 | * The update is prepared by super type handler and passed to the monitor | |
297 | * thread. | |
298 | */ | |
299 | static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd) | |
300 | { | |
301 | struct metadata_update *update = NULL; | |
302 | mdu_disk_info_t dk = { | |
303 | .number = -1, | |
304 | .major = sd->disk.major, | |
305 | .minor = sd->disk.minor, | |
306 | .raid_disk = -1, | |
307 | .state = 0, | |
308 | }; | |
1ade5cc1 N |
309 | dprintf("remove %d:%d from container\n", |
310 | sd->disk.major, sd->disk.minor); | |
1a64be56 LM |
311 | |
312 | st->update_tail = &update; | |
313 | st->ss->remove_from_super(st, &dk); | |
4dd968cc N |
314 | /* FIXME this write_init_super shouldn't be here. |
315 | * We have it after add_to_super to write to new device, | |
316 | * but with 'remove' we don't ant to write to that device! | |
317 | */ | |
1a64be56 LM |
318 | st->ss->write_init_super(st); |
319 | queue_metadata_update(update); | |
320 | st->update_tail = NULL; | |
321 | } | |
322 | ||
549e9569 NB |
323 | static void manage_container(struct mdstat_ent *mdstat, |
324 | struct supertype *container) | |
325 | { | |
1a64be56 | 326 | /* Of interest here are: |
1011e834 | 327 | * - if a new device has been added to the container, we |
1a64be56 LM |
328 | * add it to the array ignoring any metadata on it. |
329 | * - if a device has been removed from the container, we | |
330 | * remove it from the device list and update the metadata. | |
549e9569 NB |
331 | * FIXME should we look for compatible metadata and take hints |
332 | * about spare assignment.... probably not. | |
549e9569 NB |
333 | */ |
334 | if (mdstat->devcnt != container->devcnt) { | |
7bc1962f DW |
335 | struct mdinfo **cdp, *cd, *di, *mdi; |
336 | int found; | |
337 | ||
549e9569 NB |
338 | /* read /sys/block/NAME/md/dev-??/block/dev to find out |
339 | * what is there, and compare with container->info.devs | |
340 | * To see what is removed and what is added. | |
341 | * These need to be remove from, or added to, the array | |
342 | */ | |
4dd2df09 | 343 | mdi = sysfs_read(-1, mdstat->devnm, GET_DEVS); |
313a4a82 DW |
344 | if (!mdi) { |
345 | /* invalidate the current count so we can try again */ | |
346 | container->devcnt = -1; | |
7bc1962f | 347 | return; |
313a4a82 | 348 | } |
7bc1962f DW |
349 | |
350 | /* check for removals */ | |
351 | for (cdp = &container->devs; *cdp; ) { | |
352 | found = 0; | |
353 | for (di = mdi->devs; di; di = di->next) | |
354 | if (di->disk.major == (*cdp)->disk.major && | |
355 | di->disk.minor == (*cdp)->disk.minor) { | |
356 | found = 1; | |
357 | break; | |
358 | } | |
359 | if (!found) { | |
360 | cd = *cdp; | |
361 | *cdp = (*cdp)->next; | |
1a64be56 | 362 | remove_disk_from_container(container, cd); |
7bc1962f DW |
363 | free(cd); |
364 | } else | |
365 | cdp = &(*cdp)->next; | |
366 | } | |
43dad3d6 DW |
367 | |
368 | /* check for additions */ | |
369 | for (di = mdi->devs; di; di = di->next) { | |
370 | for (cd = container->devs; cd; cd = cd->next) | |
371 | if (di->disk.major == cd->disk.major && | |
372 | di->disk.minor == cd->disk.minor) | |
373 | break; | |
04a8ac08 | 374 | if (!cd) { |
503975b9 | 375 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
04a8ac08 | 376 | |
04a8ac08 DW |
377 | *newd = *di; |
378 | add_disk_to_container(container, newd); | |
379 | } | |
43dad3d6 | 380 | } |
7bc1962f | 381 | sysfs_free(mdi); |
549e9569 NB |
382 | container->devcnt = mdstat->devcnt; |
383 | } | |
384 | } | |
385 | ||
da5a36fa N |
386 | static int sysfs_open2(char *devnum, char *name, char *attr) |
387 | { | |
388 | int fd = sysfs_open(devnum, name, attr); | |
389 | if (fd >= 0) { | |
390 | /* seq_file in the kernel allocates buffer space | |
391 | * on the first read. Do that now so 'monitor' | |
392 | * never needs too. | |
393 | */ | |
394 | char buf[200]; | |
27aefbdb N |
395 | if (read(fd, buf, sizeof(buf)) < 0) |
396 | /* pretend not to ignore return value */ | |
397 | return fd; | |
da5a36fa N |
398 | } |
399 | return fd; | |
400 | } | |
401 | ||
63b4aae3 DW |
402 | static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone, |
403 | struct active_array *aa) | |
404 | { | |
405 | if (!disk || !clone) | |
406 | return -1; | |
407 | ||
408 | *disk = *clone; | |
da5a36fa N |
409 | disk->recovery_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
410 | "recovery_start"); | |
3e1d79b2 JS |
411 | if (disk->recovery_fd < 0) |
412 | return -1; | |
da5a36fa | 413 | disk->state_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, "state"); |
3e1d79b2 JS |
414 | if (disk->state_fd < 0) { |
415 | close(disk->recovery_fd); | |
416 | return -1; | |
417 | } | |
6dc1785f TM |
418 | disk->bb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, |
419 | "bad_blocks"); | |
420 | if (disk->bb_fd < 0) { | |
421 | close(disk->recovery_fd); | |
422 | close(disk->state_fd); | |
423 | return -1; | |
424 | } | |
425 | disk->ubb_fd = sysfs_open2(aa->info.sys_name, disk->sys_name, | |
426 | "unacknowledged_bad_blocks"); | |
427 | if (disk->ubb_fd < 0) { | |
428 | close(disk->recovery_fd); | |
429 | close(disk->state_fd); | |
430 | close(disk->bb_fd); | |
431 | return -1; | |
432 | } | |
63b4aae3 DW |
433 | disk->prev_state = read_dev_state(disk->state_fd); |
434 | disk->curr_state = disk->prev_state; | |
435 | disk->next = aa->info.devs; | |
436 | aa->info.devs = disk; | |
437 | ||
438 | return 0; | |
439 | } | |
440 | ||
549e9569 NB |
441 | static void manage_member(struct mdstat_ent *mdstat, |
442 | struct active_array *a) | |
443 | { | |
444 | /* Compare mdstat info with known state of member array. | |
445 | * We do not need to look for device state changes here, that | |
446 | * is dealt with by the monitor. | |
447 | * | |
0f99b4bd N |
448 | * If a reshape is being requested, monitor will have noticed |
449 | * that sync_action changed and will have set check_reshape. | |
450 | * We just need to see if new devices have appeared. All metadata | |
451 | * updates will already have been processed. | |
6c3fb95c | 452 | * |
0f99b4bd | 453 | * We also want to handle degraded arrays here by |
6c3fb95c NB |
454 | * trying to find and assign a spare. |
455 | * We do that whenever the monitor tells us too. | |
549e9569 | 456 | */ |
90fd7001 | 457 | char buf[SYSFS_MAX_BUF_SIZE]; |
bc77ed53 | 458 | int frozen; |
4e2c1a9a | 459 | struct supertype *container = a->container; |
2c8890e9 | 460 | struct mdinfo *mdi; |
4e2c1a9a N |
461 | |
462 | if (container == NULL) | |
463 | /* Raced with something */ | |
464 | return; | |
bc77ed53 | 465 | |
e49a8a80 N |
466 | if (mdstat->active) { |
467 | // FIXME | |
468 | a->info.array.raid_disks = mdstat->raid_disks; | |
469 | // MORE | |
470 | } | |
549e9569 | 471 | |
2c8890e9 AP |
472 | mdi = sysfs_read(-1, mdstat->devnm, |
473 | GET_COMPONENT|GET_CONSISTENCY_POLICY); | |
474 | if (mdi) { | |
475 | a->info.component_size = mdi->component_size; | |
476 | a->info.consistency_policy = mdi->consistency_policy; | |
477 | sysfs_free(mdi); | |
478 | } | |
4edb8530 | 479 | |
bc77ed53 DW |
480 | /* honor 'frozen' */ |
481 | if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0) | |
482 | frozen = buf[9] == '-'; | |
483 | else | |
484 | frozen = 1; /* can't read metadata_version assume the worst */ | |
485 | ||
f54a6742 | 486 | /* If sync_action is not 'idle' then don't try recovery now */ |
fc54fe7a JS |
487 | if (!frozen && |
488 | sysfs_get_str(&a->info, NULL, "sync_action", | |
489 | buf, sizeof(buf)) > 0 && strncmp(buf, "idle", 4) != 0) | |
f54a6742 N |
490 | frozen = 1; |
491 | ||
57f8c769 AK |
492 | if (mdstat->level) { |
493 | int level = map_name(pers, mdstat->level); | |
7023e0b8 | 494 | if (level == 0 || level == LEVEL_LINEAR) { |
ba714450 | 495 | a->to_remove = 1; |
84f3857f | 496 | wakeup_monitor(); |
7023e0b8 N |
497 | return; |
498 | } | |
499 | else if (a->info.array.level != level && level > 0) { | |
57f8c769 AK |
500 | struct active_array *newa = duplicate_aa(a); |
501 | if (newa) { | |
502 | newa->info.array.level = level; | |
4e2c1a9a | 503 | replace_array(container, a, newa); |
57f8c769 AK |
504 | a = newa; |
505 | } | |
506 | } | |
507 | } | |
508 | ||
50927b13 AK |
509 | /* we are after monitor kick, |
510 | * so container field can be cleared - check it again | |
511 | */ | |
512 | if (a->container == NULL) | |
513 | return; | |
514 | ||
c76242c5 TM |
515 | if (sigterm && a->info.safe_mode_delay != 1 && |
516 | a->safe_mode_delay_fd >= 0) { | |
517 | long int new_delay = 1; | |
518 | char delay[10]; | |
519 | ssize_t len; | |
520 | ||
521 | len = snprintf(delay, sizeof(delay), "0.%03ld\n", new_delay); | |
522 | if (write(a->safe_mode_delay_fd, delay, len) == len) | |
523 | a->info.safe_mode_delay = new_delay; | |
2ef21963 N |
524 | } |
525 | ||
0c4f6e37 | 526 | /* We don't check the array while any update is pending, as it |
88b496c2 | 527 | * might container a change (such as a spare assignment) which |
0c4f6e37 N |
528 | * could affect our decisions. |
529 | */ | |
88b496c2 | 530 | if (a->check_degraded && !frozen && |
0c4f6e37 | 531 | update_queue == NULL && update_queue_pending == NULL) { |
6c3fb95c | 532 | struct metadata_update *updates = NULL; |
071cfc42 | 533 | struct mdinfo *newdev = NULL; |
6c3fb95c | 534 | struct active_array *newa; |
071cfc42 | 535 | struct mdinfo *d; |
3c00ffbe | 536 | |
6c3fb95c NB |
537 | a->check_degraded = 0; |
538 | ||
539 | /* The array may not be degraded, this is just a good time | |
540 | * to check. | |
541 | */ | |
4e2c1a9a | 542 | newdev = container->ss->activate_spare(a, &updates); |
071cfc42 DW |
543 | if (!newdev) |
544 | return; | |
545 | ||
546 | newa = duplicate_aa(a); | |
547 | if (!newa) | |
548 | goto out; | |
1d446d52 DW |
549 | /* prevent the kernel from activating the disk(s) before we |
550 | * finish adding them | |
551 | */ | |
1ade5cc1 | 552 | dprintf("freezing %s\n", a->info.sys_name); |
1d446d52 | 553 | sysfs_set_str(&a->info, NULL, "sync_action", "frozen"); |
071cfc42 DW |
554 | |
555 | /* Add device to array and set offset/size/slot. | |
556 | * and open files for each newdev */ | |
557 | for (d = newdev; d ; d = d->next) { | |
558 | struct mdinfo *newd; | |
559 | ||
503975b9 | 560 | newd = xmalloc(sizeof(*newd)); |
2904b26f | 561 | if (sysfs_add_disk(&newa->info, d, 0) < 0) { |
071cfc42 DW |
562 | free(newd); |
563 | continue; | |
6c3fb95c | 564 | } |
63b4aae3 | 565 | disk_init_and_add(newd, d, newa); |
071cfc42 DW |
566 | } |
567 | queue_metadata_update(updates); | |
568 | updates = NULL; | |
6ca1e6ec MW |
569 | while (update_queue_pending || update_queue) { |
570 | check_update_queue(container); | |
239b3cc0 | 571 | sleep_for(0, MSEC_TO_NSEC(15), true); |
6ca1e6ec | 572 | } |
4e2c1a9a | 573 | replace_array(container, a, newa); |
b831b299 JS |
574 | if (sysfs_set_str(&a->info, NULL, |
575 | "sync_action", "recover") == 0) | |
6ca1e6ec | 576 | newa->prev_action = recover; |
1ade5cc1 | 577 | dprintf("recovery started on %s\n", a->info.sys_name); |
071cfc42 DW |
578 | out: |
579 | while (newdev) { | |
580 | d = newdev->next; | |
581 | free(newdev); | |
582 | newdev = d; | |
6c3fb95c | 583 | } |
071cfc42 | 584 | free_updates(&updates); |
6c3fb95c | 585 | } |
0f99b4bd N |
586 | |
587 | if (a->check_reshape) { | |
588 | /* mdadm might have added some devices to the array. | |
589 | * We want to disk_init_and_add any such device to a | |
590 | * duplicate_aa and replace a with that. | |
591 | * mdstat doesn't have enough info so we sysfs_read | |
592 | * and look for new stuff. | |
593 | */ | |
594 | struct mdinfo *info, *d, *d2, *newd; | |
aad6f216 | 595 | unsigned long long array_size; |
0f99b4bd N |
596 | struct active_array *newa = NULL; |
597 | a->check_reshape = 0; | |
4dd2df09 | 598 | info = sysfs_read(-1, mdstat->devnm, |
0f99b4bd N |
599 | GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE); |
600 | if (!info) | |
601 | goto out2; | |
602 | for (d = info->devs; d; d = d->next) { | |
603 | if (d->disk.raid_disk < 0) | |
604 | continue; | |
605 | for (d2 = a->info.devs; d2; d2 = d2->next) | |
606 | if (d2->disk.raid_disk == | |
607 | d->disk.raid_disk) | |
608 | break; | |
609 | if (d2) | |
610 | /* already have this one */ | |
611 | continue; | |
612 | if (!newa) { | |
613 | newa = duplicate_aa(a); | |
614 | if (!newa) | |
615 | break; | |
616 | } | |
503975b9 | 617 | newd = xmalloc(sizeof(*newd)); |
0f99b4bd N |
618 | disk_init_and_add(newd, d, newa); |
619 | } | |
fc54fe7a JS |
620 | if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0 && |
621 | a->info.custom_array_size > array_size*2) { | |
aad6f216 | 622 | sysfs_set_num(info, NULL, "array_size", |
02eedb57 | 623 | a->info.custom_array_size/2); |
aad6f216 | 624 | } |
0f99b4bd N |
625 | out2: |
626 | sysfs_free(info); | |
627 | if (newa) | |
4e2c1a9a | 628 | replace_array(container, a, newa); |
0f99b4bd | 629 | } |
549e9569 NB |
630 | } |
631 | ||
836759d5 DW |
632 | static int aa_ready(struct active_array *aa) |
633 | { | |
634 | struct mdinfo *d; | |
635 | int level = aa->info.array.level; | |
636 | ||
637 | for (d = aa->info.devs; d; d = d->next) | |
638 | if (d->state_fd < 0) | |
639 | return 0; | |
640 | ||
641 | if (aa->info.state_fd < 0) | |
642 | return 0; | |
643 | ||
644 | if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0)) | |
645 | return 0; | |
646 | ||
647 | if (!aa->container) | |
648 | return 0; | |
649 | ||
650 | return 1; | |
651 | } | |
652 | ||
549e9569 | 653 | static void manage_new(struct mdstat_ent *mdstat, |
2a0bb19e DW |
654 | struct supertype *container, |
655 | struct active_array *victim) | |
549e9569 NB |
656 | { |
657 | /* A new array has appeared in this container. | |
658 | * Hopefully it is already recorded in the metadata. | |
659 | * Check, then create the new array to report it to | |
660 | * the monitor. | |
661 | */ | |
662 | ||
60815698 MG |
663 | struct active_array *new = NULL; |
664 | struct mdinfo *mdi = NULL, *di; | |
665 | int i, inst; | |
f1d26766 | 666 | int failed = 0; |
90fd7001 | 667 | char buf[SYSFS_MAX_BUF_SIZE]; |
549e9569 | 668 | |
836759d5 | 669 | /* check if array is ready to be monitored */ |
7023e0b8 N |
670 | if (!mdstat->active || !mdstat->level) |
671 | return; | |
60815698 MG |
672 | if (strncmp(mdstat->level, "raid0", strlen("raid0")) == 0 || |
673 | strncmp(mdstat->level, "linear", strlen("linear")) == 0) | |
836759d5 DW |
674 | return; |
675 | ||
4dd2df09 | 676 | mdi = sysfs_read(-1, mdstat->devnm, |
836759d5 | 677 | GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT| |
b13b52c8 | 678 | GET_SAFEMODE|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE| |
ae7d61e3 | 679 | GET_LAYOUT|GET_DEVS_ALL); |
836759d5 | 680 | |
503975b9 | 681 | if (!mdi) |
836759d5 | 682 | return; |
503975b9 | 683 | new = xcalloc(1, sizeof(*new)); |
d52690ac | 684 | |
4dd2df09 | 685 | strcpy(new->info.sys_name, mdstat->devnm); |
549e9569 NB |
686 | |
687 | new->prev_state = new->curr_state = new->next_state = inactive; | |
688 | new->prev_action= new->curr_action= new->next_action= idle; | |
689 | ||
690 | new->container = container; | |
691 | ||
60815698 MG |
692 | if (parse_num(&inst, to_subarray(mdstat, container->devnm)) != 0) |
693 | goto error; | |
549e9569 | 694 | |
549e9569 | 695 | new->info.array = mdi->array; |
272bcc48 | 696 | new->info.component_size = mdi->component_size; |
549e9569 NB |
697 | |
698 | for (i = 0; i < new->info.array.raid_disks; i++) { | |
503975b9 | 699 | struct mdinfo *newd = xmalloc(sizeof(*newd)); |
549e9569 NB |
700 | |
701 | for (di = mdi->devs; di; di = di->next) | |
702 | if (i == di->disk.raid_disk) | |
703 | break; | |
704 | ||
63b4aae3 | 705 | if (disk_init_and_add(newd, di, new) != 0) { |
7da80e6f DW |
706 | if (newd) |
707 | free(newd); | |
708 | ||
f1d26766 | 709 | failed++; |
7da80e6f DW |
710 | if (failed > new->info.array.failed_disks) { |
711 | /* we cannot properly monitor without all working disks */ | |
712 | new->container = NULL; | |
713 | break; | |
714 | } | |
549e9569 | 715 | } |
549e9569 | 716 | } |
836759d5 | 717 | |
da5a36fa N |
718 | new->action_fd = sysfs_open2(new->info.sys_name, NULL, "sync_action"); |
719 | new->info.state_fd = sysfs_open2(new->info.sys_name, NULL, "array_state"); | |
720 | new->resync_start_fd = sysfs_open2(new->info.sys_name, NULL, "resync_start"); | |
721 | new->metadata_fd = sysfs_open2(new->info.sys_name, NULL, "metadata_version"); | |
722 | new->sync_completed_fd = sysfs_open2(new->info.sys_name, NULL, "sync_completed"); | |
c76242c5 TM |
723 | new->safe_mode_delay_fd = sysfs_open2(new->info.sys_name, NULL, |
724 | "safe_mode_delay"); | |
da5a36fa | 725 | |
60815698 | 726 | dprintf("inst: %d action: %d state: %d\n", inst, |
4e6e574a | 727 | new->action_fd, new->info.state_fd); |
549e9569 | 728 | |
69d08478 | 729 | if (mdi->safe_mode_delay >= 50) |
2ef21963 N |
730 | /* Normal start, mdadm set this. */ |
731 | new->info.safe_mode_delay = mdi->safe_mode_delay; | |
732 | else | |
733 | /* Restart, just pick a number */ | |
734 | new->info.safe_mode_delay = 5000; | |
735 | sysfs_set_safemode(&new->info, new->info.safe_mode_delay); | |
736 | ||
138477db AK |
737 | /* reshape_position is set by mdadm in sysfs |
738 | * read this information for new arrays only (empty victim) | |
739 | */ | |
740 | if ((victim == NULL) && | |
90fd7001 | 741 | (sysfs_get_str(mdi, NULL, "sync_action", buf, sizeof(buf)) > 0) && |
138477db AK |
742 | (strncmp(buf, "reshape", 7) == 0)) { |
743 | if (sysfs_get_ll(mdi, NULL, "reshape_position", | |
744 | &new->last_checkpoint) != 0) | |
745 | new->last_checkpoint = 0; | |
0d51bfa2 AK |
746 | else { |
747 | int data_disks = mdi->array.raid_disks; | |
748 | if (mdi->array.level == 4 || mdi->array.level == 5) | |
749 | data_disks--; | |
750 | if (mdi->array.level == 6) | |
751 | data_disks -= 2; | |
752 | ||
753 | new->last_checkpoint /= data_disks; | |
754 | } | |
138477db AK |
755 | dprintf("mdmon: New monitored array is under reshape.\n" |
756 | " Last checkpoint is: %llu\n", | |
757 | new->last_checkpoint); | |
758 | } | |
759 | ||
4fa5aef9 | 760 | sysfs_free(mdi); |
60815698 | 761 | mdi = NULL; |
836759d5 DW |
762 | |
763 | /* if everything checks out tell the metadata handler we want to | |
764 | * manage this instance | |
765 | */ | |
766 | if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) { | |
60815698 | 767 | goto error; |
93f7caca | 768 | } else { |
2a0bb19e | 769 | replace_array(container, victim, new); |
93f7caca DW |
770 | if (failed) { |
771 | new->check_degraded = 1; | |
772 | manage_member(mdstat, new); | |
773 | } | |
774 | } | |
60815698 MG |
775 | return; |
776 | ||
777 | error: | |
778 | pr_err("failed to monitor %s\n", mdstat->metadata_version); | |
779 | if (new) { | |
780 | new->container = NULL; | |
781 | free_aa(new); | |
782 | } | |
783 | if (mdi) | |
784 | sysfs_free(mdi); | |
549e9569 NB |
785 | } |
786 | ||
5d19760d | 787 | void manage(struct mdstat_ent *mdstat, struct supertype *container) |
549e9569 NB |
788 | { |
789 | /* We have just read mdstat and need to compare it with | |
790 | * the known active arrays. | |
791 | * Arrays with the wrong metadata are ignored. | |
792 | */ | |
793 | ||
794 | for ( ; mdstat ; mdstat = mdstat->next) { | |
795 | struct active_array *a; | |
4dd2df09 | 796 | if (strcmp(mdstat->devnm, container->devnm) == 0) { |
549e9569 NB |
797 | manage_container(mdstat, container); |
798 | continue; | |
799 | } | |
4dd2df09 | 800 | if (!is_container_member(mdstat, container->devnm)) |
549e9569 NB |
801 | /* Not for this array */ |
802 | continue; | |
803 | /* Looks like a member of this container */ | |
5d19760d | 804 | for (a = container->arrays; a; a = a->next) { |
4dd2df09 | 805 | if (strcmp(mdstat->devnm, a->info.sys_name) == 0) { |
ba714450 | 806 | if (a->container && a->to_remove == 0) |
549e9569 NB |
807 | manage_member(mdstat, a); |
808 | break; | |
809 | } | |
810 | } | |
69d08478 | 811 | if ((a == NULL || !a->container) && !sigterm) |
2a0bb19e | 812 | manage_new(mdstat, container, a); |
549e9569 NB |
813 | } |
814 | } | |
815 | ||
edd8d13c | 816 | static void handle_message(struct supertype *container, struct metadata_update *msg) |
3e70c845 | 817 | { |
edd8d13c NB |
818 | /* queue this metadata update through to the monitor */ |
819 | ||
820 | struct metadata_update *mu; | |
821 | ||
313a4a82 | 822 | if (msg->len <= 0) |
3c00ffbe N |
823 | while (update_queue_pending || update_queue) { |
824 | check_update_queue(container); | |
239b3cc0 | 825 | sleep_for(0, MSEC_TO_NSEC(15), true); |
3c00ffbe N |
826 | } |
827 | ||
313a4a82 DW |
828 | if (msg->len == 0) { /* ping_monitor */ |
829 | int cnt; | |
1011e834 | 830 | |
3c00ffbe | 831 | cnt = monitor_loop_cnt; |
1eb252b8 N |
832 | if (cnt & 1) |
833 | cnt += 2; /* wait until next pselect */ | |
834 | else | |
835 | cnt += 3; /* wait for 2 pselects */ | |
836 | wakeup_monitor(); | |
3c00ffbe | 837 | |
1eb252b8 | 838 | while (monitor_loop_cnt - cnt < 0) |
239b3cc0 | 839 | sleep_for(0, MSEC_TO_NSEC(10), true); |
313a4a82 DW |
840 | } else if (msg->len == -1) { /* ping_manager */ |
841 | struct mdstat_ent *mdstat = mdstat_read(1, 0); | |
842 | ||
843 | manage(mdstat, container); | |
844 | free_mdstat(mdstat); | |
6144ed44 | 845 | } else if (!sigterm) { |
503975b9 | 846 | mu = xmalloc(sizeof(*mu)); |
edd8d13c NB |
847 | mu->len = msg->len; |
848 | mu->buf = msg->buf; | |
849 | msg->buf = NULL; | |
850 | mu->space = NULL; | |
cb23f1f4 | 851 | mu->space_list = NULL; |
edd8d13c NB |
852 | mu->next = NULL; |
853 | if (container->ss->prepare_update) | |
5fe6f031 N |
854 | if (!container->ss->prepare_update(container, mu)) |
855 | free_updates(&mu); | |
edd8d13c NB |
856 | queue_metadata_update(mu); |
857 | } | |
3e70c845 DW |
858 | } |
859 | ||
860 | void read_sock(struct supertype *container) | |
549e9569 NB |
861 | { |
862 | int fd; | |
bfa44e2e | 863 | struct metadata_update msg; |
b109d928 DW |
864 | int terminate = 0; |
865 | long fl; | |
866 | int tmo = 3; /* 3 second timeout before hanging up the socket */ | |
549e9569 | 867 | |
3e70c845 | 868 | fd = accept(container->sock, NULL, NULL); |
549e9569 NB |
869 | if (fd < 0) |
870 | return; | |
b109d928 DW |
871 | |
872 | fl = fcntl(fd, F_GETFL, 0); | |
873 | fl |= O_NONBLOCK; | |
874 | fcntl(fd, F_SETFL, fl); | |
875 | ||
876 | do { | |
877 | msg.buf = NULL; | |
878 | ||
879 | /* read and validate the message */ | |
880 | if (receive_message(fd, &msg, tmo) == 0) { | |
bfa44e2e | 881 | handle_message(container, &msg); |
bc77ed53 DW |
882 | if (msg.len == 0) { |
883 | /* ping reply with version */ | |
884 | msg.buf = Version; | |
885 | msg.len = strlen(Version) + 1; | |
886 | if (send_message(fd, &msg, tmo) < 0) | |
887 | terminate = 1; | |
888 | } else if (ack(fd, tmo) < 0) | |
bfa44e2e NB |
889 | terminate = 1; |
890 | } else | |
b109d928 | 891 | terminate = 1; |
b109d928 | 892 | |
b109d928 DW |
893 | } while (!terminate); |
894 | ||
549e9569 NB |
895 | close(fd); |
896 | } | |
1ed3f387 | 897 | |
e0d6609f NB |
898 | int exit_now = 0; |
899 | int manager_ready = 0; | |
549e9569 NB |
900 | void do_manager(struct supertype *container) |
901 | { | |
902 | struct mdstat_ent *mdstat; | |
4d43913c | 903 | sigset_t set; |
1ed3f387 | 904 | |
4d43913c NB |
905 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
906 | sigdelset(&set, SIGUSR1); | |
6144ed44 | 907 | sigdelset(&set, SIGTERM); |
549e9569 NB |
908 | |
909 | do { | |
1ed3f387 | 910 | |
e0d6609f NB |
911 | if (exit_now) |
912 | exit(0); | |
913 | ||
3c00ffbe N |
914 | /* Can only 'manage' things if 'monitor' is not making |
915 | * structural changes to metadata, so need to check | |
916 | * update_queue | |
917 | */ | |
918 | if (update_queue == NULL) { | |
919 | mdstat = mdstat_read(1, 0); | |
549e9569 | 920 | |
3c00ffbe | 921 | manage(mdstat, container); |
549e9569 | 922 | |
3c00ffbe | 923 | read_sock(container); |
4fa5aef9 | 924 | |
3c00ffbe N |
925 | free_mdstat(mdstat); |
926 | } | |
1ed3f387 NB |
927 | remove_old(); |
928 | ||
2e735d19 NB |
929 | check_update_queue(container); |
930 | ||
e0d6609f | 931 | manager_ready = 1; |
4d43913c | 932 | |
6144ed44 DW |
933 | if (sigterm) |
934 | wakeup_monitor(); | |
935 | ||
5d4d1b26 | 936 | if (update_queue == NULL) |
58a4ba2a | 937 | mdstat_wait_fd(container->sock, &set); |
5d4d1b26 | 938 | else |
3c00ffbe N |
939 | /* If an update is happening, just wait for signal */ |
940 | pselect(0, NULL, NULL, NULL, NULL, &set); | |
549e9569 NB |
941 | } while(1); |
942 | } |