]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Incremental.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Add data_offset arg to ->avail_size
[thirdparty/mdadm.git] / Incremental.c
1 /*
2 * Incremental.c - support --incremental. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
4 *
5 * Copyright (C) 2006-2012 Neil Brown <neilb@suse.de>
6 *
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 *
22 * Author: Neil Brown
23 * Email: <neilb@suse.de>
24 * Paper: Neil Brown
25 * Novell Inc
26 * GPO Box Q1283
27 * QVB Post Office, NSW 1230
28 * Australia
29 */
30
31 #include "mdadm.h"
32 #include <dirent.h>
33 #include <ctype.h>
34
35 static int count_active(struct supertype *st, struct mdinfo *sra,
36 int mdfd, char **availp,
37 struct mdinfo *info);
38 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
39 int number, __u64 events, int verbose,
40 char *array_name);
41 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
42 struct map_ent *target,
43 struct supertype *st, int verbose);
44
45 static int Incremental_container(struct supertype *st, char *devname,
46 struct context *c);
47
48 int Incremental(char *devname, struct context *c,
49 struct supertype *st)
50 {
51 /* Add this device to an array, creating the array if necessary
52 * and starting the array if sensible or - if runstop>0 - if possible.
53 *
54 * This has several steps:
55 *
56 * 1/ Check if device is permitted by mdadm.conf, reject if not.
57 * 2/ Find metadata, reject if none appropriate (check
58 * version/name from args)
59 * 3/ Check if there is a match in mdadm.conf
60 * 3a/ if not, check for homehost match. If no match, assemble as
61 * a 'foreign' array.
62 * 4/ Determine device number.
63 * - If in mdadm.conf with std name, use that
64 * - UUID in /var/run/mdadm.map use that
65 * - If name is suggestive, use that. unless in use with different uuid.
66 * - Choose a free, high number.
67 * - Use a partitioned device unless strong suggestion not to.
68 * e.g. auto=md
69 * Don't choose partitioned for containers.
70 * 5/ Find out if array already exists
71 * 5a/ if it does not
72 * - choose a name, from mdadm.conf or 'name' field in array.
73 * - create the array
74 * - add the device
75 * 5b/ if it does
76 * - check one drive in array to make sure metadata is a reasonably
77 * close match. Reject if not (e.g. different type)
78 * - add the device
79 * 6/ Make sure /var/run/mdadm.map contains this array.
80 * 7/ Is there enough devices to possibly start the array?
81 * For a container, this means running Incremental_container.
82 * 7a/ if not, finish with success.
83 * 7b/ if yes,
84 * - read all metadata and arrange devices like -A does
85 * - if number of OK devices match expected, or -R and there are enough,
86 * start the array (auto-readonly).
87 */
88 struct stat stb;
89 struct mdinfo info, dinfo;
90 struct mdinfo *sra = NULL, *d;
91 struct mddev_ident *match;
92 char chosen_name[1024];
93 int rv = 1;
94 struct map_ent *mp, *map = NULL;
95 int dfd = -1, mdfd = -1;
96 char *avail = NULL;
97 int active_disks;
98 int trustworthy;
99 char *name_to_use;
100 mdu_array_info_t ainf;
101 struct dev_policy *policy = NULL;
102 struct map_ent target_array;
103 int have_target;
104
105 struct createinfo *ci = conf_get_create_info();
106
107 if (stat(devname, &stb) < 0) {
108 if (c->verbose >= 0)
109 pr_err("stat failed for %s: %s.\n",
110 devname, strerror(errno));
111 return rv;
112 }
113 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
114 if (c->verbose >= 0)
115 pr_err("%s is not a block device.\n",
116 devname);
117 return rv;
118 }
119 dfd = dev_open(devname, O_RDONLY|O_EXCL);
120 if (dfd < 0) {
121 if (c->verbose >= 0)
122 pr_err("cannot open %s: %s.\n",
123 devname, strerror(errno));
124 return rv;
125 }
126 /* If the device is a container, we do something very different */
127 if (must_be_container(dfd)) {
128 if (!st)
129 st = super_by_fd(dfd, NULL);
130 if (st && st->ss->load_container)
131 rv = st->ss->load_container(st, dfd, NULL);
132
133 close(dfd);
134 if (!rv && st->ss->container_content) {
135 if (map_lock(&map))
136 pr_err("failed to get "
137 "exclusive lock on mapfile\n");
138 rv = Incremental_container(st, devname, c);
139 map_unlock(&map);
140 return rv;
141 }
142
143 pr_err("%s is not part of an md array.\n",
144 devname);
145 return rv;
146 }
147
148 /* 1/ Check if device is permitted by mdadm.conf */
149
150 if (!conf_test_dev(devname)) {
151 if (c->verbose >= 0)
152 pr_err("%s not permitted by mdadm.conf.\n",
153 devname);
154 goto out;
155 }
156
157 /* 2/ Find metadata, reject if none appropriate (check
158 * version/name from args) */
159
160 if (fstat(dfd, &stb) < 0) {
161 if (c->verbose >= 0)
162 pr_err("fstat failed for %s: %s.\n",
163 devname, strerror(errno));
164 goto out;
165 }
166 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
167 if (c->verbose >= 0)
168 pr_err("%s is not a block device.\n",
169 devname);
170 goto out;
171 }
172
173 dinfo.disk.major = major(stb.st_rdev);
174 dinfo.disk.minor = minor(stb.st_rdev);
175
176 policy = disk_policy(&dinfo);
177 have_target = policy_check_path(&dinfo, &target_array);
178
179 if (st == NULL && (st = guess_super(dfd)) == NULL) {
180 if (c->verbose >= 0)
181 pr_err("no recognisable superblock on %s.\n",
182 devname);
183 rv = try_spare(devname, &dfd, policy,
184 have_target ? &target_array : NULL,
185 st, c->verbose);
186 goto out;
187 }
188 if (st->ss->compare_super == NULL ||
189 st->ss->load_super(st, dfd, NULL)) {
190 if (c->verbose >= 0)
191 pr_err("no RAID superblock on %s.\n",
192 devname);
193 rv = try_spare(devname, &dfd, policy,
194 have_target ? &target_array : NULL,
195 st, c->verbose);
196 free(st);
197 goto out;
198 }
199 close (dfd); dfd = -1;
200
201 st->ss->getinfo_super(st, &info, NULL);
202
203 /* 3/ Check if there is a match in mdadm.conf */
204 match = conf_match(st, &info, devname, c->verbose, &rv);
205 if (!match && rv == 2)
206 goto out;
207
208 if (match && match->devname
209 && strcasecmp(match->devname, "<ignore>") == 0) {
210 if (c->verbose >= 0)
211 pr_err("array containing %s is explicitly"
212 " ignored by mdadm.conf\n",
213 devname);
214 goto out;
215 }
216
217 /* 3a/ if not, check for homehost match. If no match, continue
218 * but don't trust the 'name' in the array. Thus a 'random' minor
219 * number will be assigned, and the device name will be based
220 * on that. */
221 if (match)
222 trustworthy = LOCAL;
223 else if (st->ss->match_home(st, c->homehost) == 1)
224 trustworthy = LOCAL;
225 else if (st->ss->match_home(st, "any") == 1)
226 trustworthy = LOCAL_ANY;
227 else
228 trustworthy = FOREIGN;
229
230 if (!match && !conf_test_metadata(st->ss->name, policy,
231 (trustworthy == LOCAL))) {
232 if (c->verbose >= 1)
233 pr_err("%s has metadata type %s for which "
234 "auto-assembly is disabled\n",
235 devname, st->ss->name);
236 goto out;
237 }
238 if (trustworthy == LOCAL_ANY)
239 trustworthy = LOCAL;
240
241 /* There are three possible sources for 'autof': command line,
242 * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
243 * ARRAY takes precedence, then command line, then
244 * CREATE.
245 */
246 if (match && match->autof)
247 c->autof = match->autof;
248 if (c->autof == 0)
249 c->autof = ci->autof;
250
251 name_to_use = info.name;
252 if (name_to_use[0] == 0 &&
253 info.array.level == LEVEL_CONTAINER) {
254 name_to_use = info.text_version;
255 trustworthy = METADATA;
256 }
257 if (name_to_use[0] && trustworthy != LOCAL &&
258 ! c->require_homehost &&
259 conf_name_is_free(name_to_use))
260 trustworthy = LOCAL;
261
262 /* strip "hostname:" prefix from name if we have decided
263 * to treat it as LOCAL
264 */
265 if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
266 name_to_use = strchr(name_to_use, ':')+1;
267
268 /* 4/ Check if array exists.
269 */
270 if (map_lock(&map))
271 pr_err("failed to get exclusive lock on "
272 "mapfile\n");
273 mp = map_by_uuid(&map, info.uuid);
274 if (mp)
275 mdfd = open_dev(mp->devnum);
276 else
277 mdfd = -1;
278
279 if (mdfd < 0) {
280
281 /* Couldn't find an existing array, maybe make a new one */
282 mdfd = create_mddev(match ? match->devname : NULL,
283 name_to_use, c->autof, trustworthy, chosen_name);
284
285 if (mdfd < 0)
286 goto out_unlock;
287
288 sysfs_init(&info, mdfd, 0);
289
290 if (set_array_info(mdfd, st, &info) != 0) {
291 pr_err("failed to set array info for %s: %s\n",
292 chosen_name, strerror(errno));
293 rv = 2;
294 goto out_unlock;
295 }
296
297 dinfo = info;
298 dinfo.disk.major = major(stb.st_rdev);
299 dinfo.disk.minor = minor(stb.st_rdev);
300 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
301 pr_err("failed to add %s to %s: %s.\n",
302 devname, chosen_name, strerror(errno));
303 ioctl(mdfd, STOP_ARRAY, 0);
304 rv = 2;
305 goto out_unlock;
306 }
307 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
308 GET_OFFSET | GET_SIZE));
309
310 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
311 /* It really should be 'none' - must be old buggy
312 * kernel, and mdadm -I may not be able to complete.
313 * So reject it.
314 */
315 ioctl(mdfd, STOP_ARRAY, NULL);
316 pr_err("You have an old buggy kernel which cannot support\n"
317 " --incremental reliably. Aborting.\n");
318 rv = 2;
319 goto out_unlock;
320 }
321 info.array.working_disks = 1;
322 /* 6/ Make sure /var/run/mdadm.map contains this array. */
323 map_update(&map, fd2devnum(mdfd),
324 info.text_version,
325 info.uuid, chosen_name);
326 } else {
327 /* 5b/ if it does */
328 /* - check one drive in array to make sure metadata is a reasonably */
329 /* close match. Reject if not (e.g. different type) */
330 /* - add the device */
331 char dn[20];
332 int dfd2;
333 int err;
334 struct supertype *st2;
335 struct mdinfo info2, *d;
336
337 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
338 GET_OFFSET | GET_SIZE));
339
340 if (mp->path)
341 strcpy(chosen_name, mp->path);
342 else
343 strcpy(chosen_name, devnum2devname(mp->devnum));
344
345 /* It is generally not OK to add non-spare drives to a
346 * running array as they are probably missing because
347 * they failed. However if runstop is 1, then the
348 * array was possibly started early and our best bet is
349 * to add this anyway.
350 * Also if action policy is re-add or better we allow
351 * re-add.
352 * This doesn't apply to containers as the 'non-spare'
353 * flag has a different meaning. The test has to happen
354 * at the device level there
355 */
356 if (!st->ss->external
357 && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
358 && ! policy_action_allows(policy, st->ss->name,
359 act_re_add)
360 && c->runstop < 1) {
361 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
362 pr_err("not adding %s to active array (without --run) %s\n",
363 devname, chosen_name);
364 rv = 2;
365 goto out_unlock;
366 }
367 }
368 if (!sra) {
369 rv = 2;
370 goto out_unlock;
371 }
372 if (sra->devs) {
373 sprintf(dn, "%d:%d", sra->devs->disk.major,
374 sra->devs->disk.minor);
375 dfd2 = dev_open(dn, O_RDONLY);
376 if (dfd2 < 0) {
377 pr_err("unable to open %s\n", devname);
378 rv = 2;
379 goto out_unlock;
380 }
381 st2 = dup_super(st);
382 if (st2->ss->load_super(st2, dfd2, NULL) ||
383 st->ss->compare_super(st, st2) != 0) {
384 pr_err("metadata mismatch between %s and "
385 "chosen array %s\n",
386 devname, chosen_name);
387 close(dfd2);
388 rv = 2;
389 goto out_unlock;
390 }
391 close(dfd2);
392 st2->ss->getinfo_super(st2, &info2, NULL);
393 st2->ss->free_super(st2);
394 if (info.array.level != info2.array.level ||
395 memcmp(info.uuid, info2.uuid, 16) != 0 ||
396 info.array.raid_disks != info2.array.raid_disks) {
397 pr_err("unexpected difference between %s and %s.\n",
398 chosen_name, devname);
399 rv = 2;
400 goto out_unlock;
401 }
402 }
403 info.disk.major = major(stb.st_rdev);
404 info.disk.minor = minor(stb.st_rdev);
405 /* add disk needs to know about containers */
406 if (st->ss->external)
407 sra->array.level = LEVEL_CONTAINER;
408 err = add_disk(mdfd, st, sra, &info);
409 if (err < 0 && errno == EBUSY) {
410 /* could be another device present with the same
411 * disk.number. Find and reject any such
412 */
413 find_reject(mdfd, st, sra, info.disk.number,
414 info.events, c->verbose, chosen_name);
415 err = add_disk(mdfd, st, sra, &info);
416 }
417 if (err < 0) {
418 pr_err("failed to add %s to %s: %s.\n",
419 devname, chosen_name, strerror(errno));
420 rv = 2;
421 goto out_unlock;
422 }
423 info.array.working_disks = 0;
424 for (d = sra->devs; d; d=d->next)
425 info.array.working_disks ++;
426
427 }
428
429 /* 7/ Is there enough devices to possibly start the array? */
430 /* 7a/ if not, finish with success. */
431 if (info.array.level == LEVEL_CONTAINER) {
432 int devnum = devnum; /* defined and used iff ->external */
433 /* Try to assemble within the container */
434 sysfs_uevent(sra, "change");
435 if (c->verbose >= 0)
436 pr_err("container %s now has %d device%s\n",
437 chosen_name, info.array.working_disks,
438 info.array.working_disks == 1?"":"s");
439 wait_for(chosen_name, mdfd);
440 if (st->ss->external)
441 devnum = fd2devnum(mdfd);
442 if (st->ss->load_container)
443 rv = st->ss->load_container(st, mdfd, NULL);
444 close(mdfd);
445 sysfs_free(sra);
446 if (!rv)
447 rv = Incremental_container(st, chosen_name, c);
448 map_unlock(&map);
449 if (rv == 1)
450 /* Don't fail the whole -I if a subarray didn't
451 * have enough devices to start yet
452 */
453 rv = 0;
454 /* after spare is added, ping monitor for external metadata
455 * so that it can eg. try to rebuild degraded array */
456 if (st->ss->external)
457 ping_monitor_by_id(devnum);
458 return rv;
459 }
460
461 /* We have added something to the array, so need to re-read the
462 * state. Eventually this state should be kept up-to-date as
463 * things change.
464 */
465 sysfs_free(sra);
466 sra = sysfs_read(mdfd, -1, (GET_DEVS | GET_STATE |
467 GET_OFFSET | GET_SIZE));
468 active_disks = count_active(st, sra, mdfd, &avail, &info);
469 if (enough(info.array.level, info.array.raid_disks,
470 info.array.layout, info.array.state & 1,
471 avail) == 0) {
472 if (c->verbose >= 0)
473 pr_err("%s attached to %s, not enough to start (%d).\n",
474 devname, chosen_name, active_disks);
475 rv = 0;
476 goto out_unlock;
477 }
478
479 /* 7b/ if yes, */
480 /* - if number of OK devices match expected, or -R and there */
481 /* are enough, */
482 /* + add any bitmap file */
483 /* + start the array (auto-readonly). */
484
485 if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
486 if (c->verbose >= 0)
487 pr_err("%s attached to %s which is already active.\n",
488 devname, chosen_name);
489 rv = 0;
490 goto out_unlock;
491 }
492
493 map_unlock(&map);
494 if (c->runstop > 0 || active_disks >= info.array.working_disks) {
495 struct mdinfo *dsk;
496 /* Let's try to start it */
497 if (match && match->bitmap_file) {
498 int bmfd = open(match->bitmap_file, O_RDWR);
499 if (bmfd < 0) {
500 pr_err("Could not open bitmap file %s.\n",
501 match->bitmap_file);
502 goto out;
503 }
504 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
505 close(bmfd);
506 pr_err("Failed to set bitmapfile for %s.\n",
507 chosen_name);
508 goto out;
509 }
510 close(bmfd);
511 }
512 /* Need to remove from the array any devices which
513 * 'count_active' discerned were too old or inappropriate
514 */
515 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
516 if (d->disk.state & (1<<MD_DISK_REMOVED))
517 remove_disk(mdfd, st, sra, d);
518
519 if ((sra == NULL || active_disks >= info.array.working_disks)
520 && trustworthy != FOREIGN)
521 rv = ioctl(mdfd, RUN_ARRAY, NULL);
522 else
523 rv = sysfs_set_str(sra, NULL,
524 "array_state", "read-auto");
525 if (rv == 0) {
526 if (c->verbose >= 0)
527 pr_err("%s attached to %s, which has been started.\n",
528 devname, chosen_name);
529 rv = 0;
530 wait_for(chosen_name, mdfd);
531 /* We just started the array, so some devices
532 * might have been evicted from the array
533 * because their event counts were too old.
534 * If the action=re-add policy is in-force for
535 * those devices we should re-add them now.
536 */
537 for (dsk = sra->devs; dsk ; dsk = dsk->next) {
538 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
539 add_disk(mdfd, st, sra, dsk) == 0)
540 pr_err("%s re-added to %s\n",
541 dsk->sys_name, chosen_name);
542 }
543 } else {
544 pr_err("%s attached to %s, but failed to start: %s.\n",
545 devname, chosen_name, strerror(errno));
546 rv = 1;
547 }
548 } else {
549 if (c->verbose >= 0)
550 pr_err("%s attached to %s, not enough to start safely.\n",
551 devname, chosen_name);
552 rv = 0;
553 }
554 out:
555 free(avail);
556 if (dfd >= 0)
557 close(dfd);
558 if (mdfd >= 0)
559 close(mdfd);
560 if (policy)
561 dev_policy_free(policy);
562 if (sra)
563 sysfs_free(sra);
564 return rv;
565 out_unlock:
566 map_unlock(&map);
567 goto out;
568 }
569
570 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
571 int number, __u64 events, int verbose,
572 char *array_name)
573 {
574 /* Find a device attached to this array with a disk.number of number
575 * and events less than the passed events, and remove the device.
576 */
577 struct mdinfo *d;
578 mdu_array_info_t ra;
579
580 if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
581 return; /* not safe to remove from active arrays
582 * without thinking more */
583
584 for (d = sra->devs; d ; d = d->next) {
585 char dn[10];
586 int dfd;
587 struct mdinfo info;
588 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
589 dfd = dev_open(dn, O_RDONLY);
590 if (dfd < 0)
591 continue;
592 if (st->ss->load_super(st, dfd, NULL)) {
593 close(dfd);
594 continue;
595 }
596 st->ss->getinfo_super(st, &info, NULL);
597 st->ss->free_super(st);
598 close(dfd);
599
600 if (info.disk.number != number ||
601 info.events >= events)
602 continue;
603
604 if (d->disk.raid_disk > -1)
605 sysfs_set_str(sra, d, "slot", "none");
606 if (sysfs_set_str(sra, d, "state", "remove") == 0)
607 if (verbose >= 0)
608 pr_err("removing old device %s from %s\n",
609 d->sys_name+4, array_name);
610 }
611 }
612
613 static int count_active(struct supertype *st, struct mdinfo *sra,
614 int mdfd, char **availp,
615 struct mdinfo *bestinfo)
616 {
617 /* count how many devices in sra think they are active */
618 struct mdinfo *d;
619 int cnt = 0;
620 __u64 max_events = 0;
621 char *avail = NULL;
622 int *best = NULL;
623 char *devmap = NULL;
624 int numdevs = 0;
625 int devnum;
626 int b, i;
627 int raid_disks = 0;
628
629 if (!sra)
630 return 0;
631
632 for (d = sra->devs ; d ; d = d->next)
633 numdevs++;
634 for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
635 char dn[30];
636 int dfd;
637 int ok;
638 struct mdinfo info;
639
640 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
641 dfd = dev_open(dn, O_RDONLY);
642 if (dfd < 0)
643 continue;
644 ok = st->ss->load_super(st, dfd, NULL);
645 close(dfd);
646 if (ok != 0)
647 continue;
648 info.array.raid_disks = raid_disks;
649 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
650 if (!avail) {
651 raid_disks = info.array.raid_disks;
652 avail = xcalloc(raid_disks, 1);
653 *availp = avail;
654
655 best = xcalloc(raid_disks, sizeof(int));
656 devmap = xcalloc(raid_disks, numdevs);
657
658 st->ss->getinfo_super(st, &info, devmap);
659 }
660
661 if (info.disk.state & (1<<MD_DISK_SYNC))
662 {
663 if (cnt == 0) {
664 cnt++;
665 max_events = info.events;
666 avail[info.disk.raid_disk] = 2;
667 best[info.disk.raid_disk] = devnum;
668 st->ss->getinfo_super(st, bestinfo, NULL);
669 } else if (info.events == max_events) {
670 avail[info.disk.raid_disk] = 2;
671 best[info.disk.raid_disk] = devnum;
672 } else if (info.events == max_events-1) {
673 if (avail[info.disk.raid_disk] == 0) {
674 avail[info.disk.raid_disk] = 1;
675 best[info.disk.raid_disk] = devnum;
676 }
677 } else if (info.events < max_events - 1)
678 ;
679 else if (info.events == max_events+1) {
680 int i;
681 max_events = info.events;
682 for (i = 0; i < raid_disks; i++)
683 if (avail[i])
684 avail[i]--;
685 avail[info.disk.raid_disk] = 2;
686 best[info.disk.raid_disk] = devnum;
687 st->ss->getinfo_super(st, bestinfo, NULL);
688 } else { /* info.events much bigger */
689 memset(avail, 0, raid_disks);
690 max_events = info.events;
691 avail[info.disk.raid_disk] = 2;
692 best[info.disk.raid_disk] = devnum;
693 st->ss->getinfo_super(st, bestinfo, NULL);
694 }
695 }
696 st->ss->free_super(st);
697 }
698 if (!avail)
699 return 0;
700 /* We need to reject any device that thinks the best device is
701 * failed or missing */
702 for (b = 0; b < raid_disks; b++)
703 if (avail[b] == 2)
704 break;
705 cnt = 0;
706 for (i = 0 ; i < raid_disks ; i++) {
707 if (i != b && avail[i])
708 if (devmap[raid_disks * best[i] + b] == 0) {
709 /* This device thinks 'b' is failed -
710 * don't use it */
711 devnum = best[i];
712 for (d=sra->devs ; devnum; d = d->next)
713 devnum--;
714 d->disk.state |= (1 << MD_DISK_REMOVED);
715 avail[i] = 0;
716 }
717 if (avail[i])
718 cnt++;
719 }
720 free(best);
721 free(devmap);
722 return cnt;
723 }
724
725 /* test if container has degraded member(s) */
726 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
727 {
728 mdu_array_info_t array;
729 int afd;
730 int max_degraded = 0;
731
732 for(; map; map = map->next) {
733 if (!is_subarray(map->metadata) ||
734 devname2devnum(map->metadata+1) != me->devnum)
735 continue;
736 afd = open_dev(map->devnum);
737 if (afd < 0)
738 continue;
739 /* most accurate information regarding array degradation */
740 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
741 int degraded = array.raid_disks - array.active_disks -
742 array.spare_disks;
743 if (degraded > max_degraded)
744 max_degraded = degraded;
745 }
746 close(afd);
747 }
748 return (max_degraded);
749 }
750
751 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
752 struct map_ent *target, int bare,
753 struct supertype *st, int verbose)
754 {
755 /* This device doesn't have any md metadata
756 * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
757 * If 'st' is not set, then we only know that some metadata allows this,
758 * others possibly don't.
759 * So look for a container or array to attach the device to.
760 * Prefer 'target' if that is set and the array is found.
761 *
762 * If st is set, then only arrays of that type are considered
763 * Return 0 on success, or some exit code on failure, probably 1.
764 */
765 int rv = 1;
766 struct stat stb;
767 struct map_ent *mp, *map = NULL;
768 struct mdinfo *chosen = NULL;
769 int dfd = *dfdp;
770
771 if (fstat(dfd, &stb) != 0)
772 return 1;
773
774 /*
775 * Now we need to find a suitable array to add this to.
776 * We only accept arrays that:
777 * - match 'st'
778 * - are in the same domains as the device
779 * - are of an size for which the device will be useful
780 * and we choose the one that is the most degraded
781 */
782
783 if (map_lock(&map)) {
784 pr_err("failed to get exclusive lock on "
785 "mapfile\n");
786 return 1;
787 }
788 for (mp = map ; mp ; mp = mp->next) {
789 struct supertype *st2;
790 struct domainlist *dl = NULL;
791 struct mdinfo *sra;
792 unsigned long long devsize;
793 unsigned long long component_size = 0;
794
795 if (is_subarray(mp->metadata))
796 continue;
797 if (st) {
798 st2 = st->ss->match_metadata_desc(mp->metadata);
799 if (!st2 ||
800 (st->minor_version >= 0 &&
801 st->minor_version != st2->minor_version)) {
802 if (verbose > 1)
803 pr_err("not adding %s to %s as metadata type doesn't match\n",
804 devname, mp->path);
805 free(st2);
806 continue;
807 }
808 free(st2);
809 }
810 sra = sysfs_read(-1, mp->devnum,
811 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
812 GET_DEGRADED|GET_COMPONENT|GET_VERSION);
813 if (!sra) {
814 /* Probably a container - no degraded info */
815 sra = sysfs_read(-1, mp->devnum,
816 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
817 GET_COMPONENT|GET_VERSION);
818 if (sra)
819 sra->array.failed_disks = -1;
820 }
821 if (!sra)
822 continue;
823 if (st == NULL) {
824 int i;
825 st2 = NULL;
826 for(i = 0; !st2 && superlist[i]; i++)
827 st2 = superlist[i]->match_metadata_desc(
828 sra->text_version);
829 if (!st2) {
830 if (verbose > 1)
831 pr_err("not adding %s to %s"
832 " as metadata not recognised.\n",
833 devname, mp->path);
834 goto next;
835 }
836 /* Need to double check the 'act_spare' permissions applies
837 * to this metadata.
838 */
839 if (!policy_action_allows(pol, st2->ss->name, act_spare))
840 goto next;
841 if (!bare && !policy_action_allows(pol, st2->ss->name,
842 act_spare_same_slot))
843 goto next;
844 } else
845 st2 = st;
846 /* update number of failed disks for mostly degraded
847 * container member */
848 if (sra->array.failed_disks == -1)
849 sra->array.failed_disks = container_members_max_degradation(map, mp);
850
851 get_dev_size(dfd, NULL, &devsize);
852 if (sra->component_size == 0) {
853 /* true for containers, here we must read superblock
854 * to obtain minimum spare size */
855 struct supertype *st3 = dup_super(st2);
856 int mdfd = open_dev(mp->devnum);
857 if (mdfd < 0) {
858 free(st3);
859 goto next;
860 }
861 if (st3->ss->load_container &&
862 !st3->ss->load_container(st3, mdfd, mp->path)) {
863 component_size = st3->ss->min_acceptable_spare_size(st3);
864 st3->ss->free_super(st3);
865 }
866 free(st3);
867 close(mdfd);
868 }
869 if ((sra->component_size > 0 &&
870 st2->ss->avail_size(st2, devsize,
871 sra->devs
872 ? sra->devs->data_offset
873 : INVALID_SECTORS)
874 < sra->component_size)
875 ||
876 (sra->component_size == 0 && devsize < component_size)) {
877 if (verbose > 1)
878 pr_err("not adding %s to %s as it is too small\n",
879 devname, mp->path);
880 goto next;
881 }
882 /* test against target.
883 * If 'target' is set and 'bare' is false, we only accept
884 * arrays/containers that match 'target'.
885 * If 'target' is set and 'bare' is true, we prefer the
886 * array which matches 'target'.
887 * target is considered only if we deal with degraded array
888 */
889 if (target && policy_action_allows(pol, st2->ss->name,
890 act_spare_same_slot)) {
891 if (strcmp(target->metadata, mp->metadata) == 0 &&
892 memcmp(target->uuid, mp->uuid,
893 sizeof(target->uuid)) == 0 &&
894 sra->array.failed_disks > 0) {
895 /* This is our target!! */
896 if (chosen)
897 sysfs_free(chosen);
898 chosen = sra;
899 sra = NULL;
900 /* skip to end so we don't check any more */
901 while (mp->next)
902 mp = mp->next;
903 goto next;
904 }
905 /* not our target */
906 if (!bare)
907 goto next;
908 }
909
910 dl = domain_from_array(sra, st2->ss->name);
911 if (domain_test(dl, pol, st2->ss->name) != 1) {
912 /* domain test fails */
913 if (verbose > 1)
914 pr_err("not adding %s to %s as"
915 " it is not in a compatible domain\n",
916 devname, mp->path);
917
918 goto next;
919 }
920 /* all tests passed, OK to add to this array */
921 if (!chosen) {
922 chosen = sra;
923 sra = NULL;
924 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
925 sysfs_free(chosen);
926 chosen = sra;
927 sra = NULL;
928 }
929 next:
930 if (sra)
931 sysfs_free(sra);
932 if (st != st2)
933 free(st2);
934 if (dl)
935 domain_free(dl);
936 }
937 if (chosen) {
938 /* add current device to chosen array as a spare */
939 int mdfd = open_dev(devname2devnum(chosen->sys_name));
940 if (mdfd >= 0) {
941 struct mddev_dev devlist;
942 char devname[20];
943 devlist.next = NULL;
944 devlist.used = 0;
945 devlist.writemostly = 0;
946 devlist.devname = devname;
947 sprintf(devname, "%d:%d", major(stb.st_rdev),
948 minor(stb.st_rdev));
949 devlist.disposition = 'a';
950 close(dfd);
951 *dfdp = -1;
952 rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
953 -1, 0, NULL, 0);
954 close(mdfd);
955 }
956 if (verbose > 0) {
957 if (rv == 0)
958 pr_err("added %s as spare for %s\n",
959 devname, chosen->sys_name);
960 else
961 pr_err("failed to add %s as spare for %s\n",
962 devname, chosen->sys_name);
963 }
964 sysfs_free(chosen);
965 }
966 map_unlock(&map);
967 return rv;
968 }
969
970 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
971 struct supertype *st, int verbose)
972 {
973 /* we know that at least one partition virtual-metadata is
974 * allowed to incorporate spares like this device. We need to
975 * find a suitable device to copy partition information from.
976 *
977 * Getting a list of all disk (not partition) devices is
978 * slightly non-trivial. We could look at /sys/block, but
979 * that is theoretically due to be removed. Maybe best to use
980 * /dev/disk/by-path/?* and ignore names ending '-partNN' as
981 * we depend on this directory of 'path' info. But that fails
982 * to find loop devices and probably others. Maybe don't
983 * worry about that, they aren't the real target.
984 *
985 * So: check things in /dev/disk/by-path to see if they are in
986 * a compatible domain, then load the partition table and see
987 * if it is OK for the new device, and choose the largest
988 * partition table that fits.
989 */
990 DIR *dir;
991 struct dirent *de;
992 char *chosen = NULL;
993 unsigned long long chosen_size = 0;
994 struct supertype *chosen_st = NULL;
995 int fd;
996
997 dir = opendir("/dev/disk/by-path");
998 if (!dir)
999 return 1;
1000 while ((de = readdir(dir)) != NULL) {
1001 char *ep;
1002 struct dev_policy *pol2 = NULL;
1003 struct domainlist *domlist = NULL;
1004 int fd = -1;
1005 struct mdinfo info;
1006 struct supertype *st2 = NULL;
1007 char *devname = NULL;
1008 unsigned long long devsectors;
1009
1010 if (de->d_ino == 0 ||
1011 de->d_name[0] == '.' ||
1012 (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1013 goto next;
1014
1015 ep = de->d_name + strlen(de->d_name);
1016 while (ep > de->d_name &&
1017 isdigit(ep[-1]))
1018 ep--;
1019 if (ep > de->d_name + 5 &&
1020 strncmp(ep-5, "-part", 5) == 0)
1021 /* This is a partition - skip it */
1022 goto next;
1023
1024 pol2 = path_policy(de->d_name, type_disk);
1025
1026 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1027 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1028 /* new device is incompatible with this device. */
1029 goto next;
1030
1031 domain_free(domlist);
1032 domlist = NULL;
1033
1034 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1035 devname = NULL;
1036 goto next;
1037 }
1038 fd = open(devname, O_RDONLY);
1039 if (fd < 0)
1040 goto next;
1041 if (get_dev_size(fd, devname, &devsectors) == 0)
1042 goto next;
1043 devsectors >>= 9;
1044
1045 if (st)
1046 st2 = dup_super(st);
1047 else
1048 st2 = guess_super_type(fd, guess_partitions);
1049 if (st2 == NULL ||
1050 st2->ss->load_super(st2, fd, NULL) < 0)
1051 goto next;
1052
1053 if (!st) {
1054 /* Check domain policy again, this time referring to metadata */
1055 domain_merge(&domlist, pol2, st2->ss->name);
1056 if (domain_test(domlist, pol, st2->ss->name) != 1)
1057 /* Incompatible devices for this metadata type */
1058 goto next;
1059 if (!policy_action_allows(pol, st2->ss->name, act_spare))
1060 /* Some partition types allow sparing, but not
1061 * this one.
1062 */
1063 goto next;
1064 }
1065
1066 st2->ss->getinfo_super(st2, &info, NULL);
1067 if (info.component_size > devsectors)
1068 /* This partitioning doesn't fit in the device */
1069 goto next;
1070
1071 /* This is an acceptable device to copy partition
1072 * metadata from. We could just stop here, but I
1073 * think I want to keep looking incase a larger
1074 * metadata which makes better use of the device can
1075 * be found.
1076 */
1077 if (chosen == NULL ||
1078 chosen_size < info.component_size) {
1079 chosen_size = info.component_size;
1080 free(chosen);
1081 chosen = devname;
1082 devname = NULL;
1083 if (chosen_st) {
1084 chosen_st->ss->free_super(chosen_st);
1085 free(chosen_st);
1086 }
1087 chosen_st = st2;
1088 st2 = NULL;
1089 }
1090
1091 next:
1092 free(devname);
1093 domain_free(domlist);
1094 dev_policy_free(pol2);
1095 if (st2)
1096 st2->ss->free_super(st2);
1097 free(st2);
1098
1099 if (fd >= 0)
1100 close(fd);
1101 }
1102
1103 closedir(dir);
1104
1105 if (!chosen)
1106 return 1;
1107
1108 /* 'chosen' is the best device we can find. Let's write its
1109 * metadata to devname dfd is read-only so don't use that
1110 */
1111 fd = open(devname, O_RDWR);
1112 if (fd >= 0) {
1113 chosen_st->ss->store_super(chosen_st, fd);
1114 close(fd);
1115 }
1116 free(chosen);
1117 chosen_st->ss->free_super(chosen_st);
1118 free(chosen_st);
1119 return 0;
1120 }
1121
1122 static int is_bare(int dfd)
1123 {
1124 unsigned long long size = 0;
1125 char bufpad[4096 + 4096];
1126 char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1127
1128 if (lseek(dfd, 0, SEEK_SET) != 0 ||
1129 read(dfd, buf, 4096) != 4096)
1130 return 0;
1131
1132 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1133 return 0;
1134 if (memcmp(buf, buf+1, 4095) != 0)
1135 return 0;
1136
1137 /* OK, first 4K appear blank, try the end. */
1138 get_dev_size(dfd, NULL, &size);
1139 if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1140 read(dfd, buf, 4096) != 4096)
1141 return 0;
1142
1143 if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1144 return 0;
1145 if (memcmp(buf, buf+1, 4095) != 0)
1146 return 0;
1147
1148 return 1;
1149 }
1150
1151 /* adding a spare to a regular array is quite different from adding one to
1152 * a set-of-partitions virtual array.
1153 * This function determines which is worth trying and tries as appropriate.
1154 * Arrays are given priority over partitions.
1155 */
1156 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1157 struct map_ent *target,
1158 struct supertype *st, int verbose)
1159 {
1160 int i;
1161 int rv;
1162 int arrays_ok = 0;
1163 int partitions_ok = 0;
1164 int dfd = *dfdp;
1165 int bare;
1166
1167 /* Can only add a spare if device has at least one domain */
1168 if (pol_find(pol, pol_domain) == NULL)
1169 return 1;
1170 /* And only if some action allows spares */
1171 if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1172 return 1;
1173
1174 /* Now check if the device is bare.
1175 * bare devices can always be added as a spare
1176 * non-bare devices can only be added if spare-same-slot is permitted,
1177 * and this device is replacing a previous device - in which case 'target'
1178 * will be set.
1179 */
1180 if (!is_bare(dfd)) {
1181 /* Must have a target and allow same_slot */
1182 /* Later - may allow force_spare without target */
1183 if (!target ||
1184 !policy_action_allows(pol, st?st->ss->name:NULL,
1185 act_spare_same_slot)) {
1186 if (verbose > 1)
1187 pr_err("%s is not bare, so not "
1188 "considering as a spare\n",
1189 devname);
1190 return 1;
1191 }
1192 bare = 0;
1193 } else
1194 bare = 1;
1195
1196 /* It might be OK to add this device to an array - need to see
1197 * what arrays might be candidates.
1198 */
1199 if (st) {
1200 /* just try try 'array' or 'partition' based on this metadata */
1201 if (st->ss->add_to_super)
1202 return array_try_spare(devname, dfdp, pol, target, bare,
1203 st, verbose);
1204 else
1205 return partition_try_spare(devname, dfdp, pol,
1206 st, verbose);
1207 }
1208 /* No metadata was specified or found so options are open.
1209 * Check for whether any array metadata, or any partition metadata
1210 * might allow adding the spare. This check is just help to avoid
1211 * a more costly scan of all arrays when we can be sure that will
1212 * fail.
1213 */
1214 for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1215 if (superlist[i]->add_to_super && !arrays_ok &&
1216 policy_action_allows(pol, superlist[i]->name, act_spare))
1217 arrays_ok = 1;
1218 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1219 policy_action_allows(pol, superlist[i]->name, act_spare))
1220 partitions_ok = 1;
1221 }
1222 rv = 1;
1223 if (arrays_ok)
1224 rv = array_try_spare(devname, dfdp, pol, target, bare,
1225 st, verbose);
1226 if (rv != 0 && partitions_ok)
1227 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1228 return rv;
1229 }
1230
1231 int IncrementalScan(int verbose)
1232 {
1233 /* look at every device listed in the 'map' file.
1234 * If one is found that is not running then:
1235 * look in mdadm.conf for bitmap file.
1236 * if one exists, but array has none, add it.
1237 * try to start array in auto-readonly mode
1238 */
1239 struct map_ent *mapl = NULL;
1240 struct map_ent *me;
1241 struct mddev_ident *devs, *mddev;
1242 int rv = 0;
1243
1244 map_read(&mapl);
1245 devs = conf_get_ident(NULL);
1246
1247 for (me = mapl ; me ; me = me->next) {
1248 mdu_array_info_t array;
1249 mdu_bitmap_file_t bmf;
1250 struct mdinfo *sra;
1251 int mdfd = open_dev(me->devnum);
1252
1253 if (mdfd < 0)
1254 continue;
1255 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1256 errno != ENODEV) {
1257 close(mdfd);
1258 continue;
1259 }
1260 /* Ok, we can try this one. Maybe it needs a bitmap */
1261 for (mddev = devs ; mddev ; mddev = mddev->next)
1262 if (mddev->devname && me->path
1263 && devname_matches(mddev->devname, me->path))
1264 break;
1265 if (mddev && mddev->bitmap_file) {
1266 /*
1267 * Note: early kernels will wrongly fail this, so it
1268 * is a hint only
1269 */
1270 int added = -1;
1271 if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1272 int bmfd = open(mddev->bitmap_file, O_RDWR);
1273 if (bmfd >= 0) {
1274 added = ioctl(mdfd, SET_BITMAP_FILE,
1275 bmfd);
1276 close(bmfd);
1277 }
1278 }
1279 if (verbose >= 0) {
1280 if (added == 0)
1281 pr_err("Added bitmap %s to %s\n",
1282 mddev->bitmap_file, me->path);
1283 else if (errno != EEXIST)
1284 pr_err("Failed to add bitmap to %s: %s\n",
1285 me->path, strerror(errno));
1286 }
1287 }
1288 sra = sysfs_read(mdfd, 0, 0);
1289 if (sra) {
1290 if (sysfs_set_str(sra, NULL,
1291 "array_state", "read-auto") == 0) {
1292 if (verbose >= 0)
1293 pr_err("started array %s\n",
1294 me->path ?: devnum2devname(me->devnum));
1295 } else {
1296 pr_err("failed to start array %s: %s\n",
1297 me->path ?: devnum2devname(me->devnum),
1298 strerror(errno));
1299 rv = 1;
1300 }
1301 sysfs_free(sra);
1302 }
1303 }
1304 return rv;
1305 }
1306
1307 static char *container2devname(char *devname)
1308 {
1309 char *mdname = NULL;
1310
1311 if (devname[0] == '/') {
1312 int fd = open(devname, O_RDONLY);
1313 if (fd >= 0) {
1314 mdname = devnum2devname(fd2devnum(fd));
1315 close(fd);
1316 }
1317 } else {
1318 int uuid[4];
1319 struct map_ent *mp, *map = NULL;
1320
1321 if (!parse_uuid(devname, uuid))
1322 return mdname;
1323 mp = map_by_uuid(&map, uuid);
1324 if (mp)
1325 mdname = devnum2devname(mp->devnum);
1326 map_free(map);
1327 }
1328
1329 return mdname;
1330 }
1331
1332 static int Incremental_container(struct supertype *st, char *devname,
1333 struct context *c)
1334 {
1335 /* Collect the contents of this container and for each
1336 * array, choose a device name and assemble the array.
1337 */
1338
1339 struct mdinfo *list;
1340 struct mdinfo *ra;
1341 struct map_ent *map = NULL;
1342 struct mdinfo info;
1343 int trustworthy;
1344 struct mddev_ident *match;
1345 int rv = 0;
1346 struct domainlist *domains;
1347 struct map_ent *smp;
1348 int suuid[4];
1349 int sfd;
1350 int ra_blocked = 0;
1351 int ra_all = 0;
1352
1353 st->ss->getinfo_super(st, &info, NULL);
1354
1355 if ((c->runstop > 0 && info.container_enough >= 0) ||
1356 info.container_enough > 0)
1357 /* pass */;
1358 else {
1359 if (c->verbose)
1360 pr_err("not enough devices to start the container\n");
1361 return 0;
1362 }
1363
1364 match = conf_match(st, &info, devname, c->verbose, &rv);
1365 if (match == NULL && rv == 2)
1366 return rv;
1367
1368 /* Need to compute 'trustworthy' */
1369 if (match)
1370 trustworthy = LOCAL;
1371 else if (st->ss->match_home(st, c->homehost) == 1)
1372 trustworthy = LOCAL;
1373 else if (st->ss->match_home(st, "any") == 1)
1374 trustworthy = LOCAL;
1375 else
1376 trustworthy = FOREIGN;
1377
1378 list = st->ss->container_content(st, NULL);
1379 /* when nothing to activate - quit */
1380 if (list == NULL)
1381 return 0;
1382 for (ra = list ; ra ; ra = ra->next) {
1383 int mdfd;
1384 char chosen_name[1024];
1385 struct map_ent *mp;
1386 struct mddev_ident *match = NULL;
1387
1388 ra_all++;
1389 /* do not activate arrays blocked by metadata handler */
1390 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1391 pr_err("Cannot activate array %s in %s.\n",
1392 ra->text_version, devname);
1393 ra_blocked++;
1394 continue;
1395 }
1396 mp = map_by_uuid(&map, ra->uuid);
1397
1398 if (mp) {
1399 mdfd = open_dev(mp->devnum);
1400 if (mp->path)
1401 strcpy(chosen_name, mp->path);
1402 else
1403 strcpy(chosen_name, devnum2devname(mp->devnum));
1404 } else {
1405
1406 /* Check in mdadm.conf for container == devname and
1407 * member == ra->text_version after second slash.
1408 */
1409 char *sub = strchr(ra->text_version+1, '/');
1410 struct mddev_ident *array_list;
1411 if (sub) {
1412 sub++;
1413 array_list = conf_get_ident(NULL);
1414 } else
1415 array_list = NULL;
1416 for(; array_list ; array_list = array_list->next) {
1417 char *dn;
1418 if (array_list->member == NULL ||
1419 array_list->container == NULL)
1420 continue;
1421 if (strcmp(array_list->member, sub) != 0)
1422 continue;
1423 if (array_list->uuid_set &&
1424 !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1425 continue;
1426 dn = container2devname(array_list->container);
1427 if (dn == NULL)
1428 continue;
1429 if (strncmp(dn, ra->text_version+1,
1430 strlen(dn)) != 0 ||
1431 ra->text_version[strlen(dn)+1] != '/') {
1432 free(dn);
1433 continue;
1434 }
1435 free(dn);
1436 /* we have a match */
1437 match = array_list;
1438 if (c->verbose>0)
1439 pr_err("match found for member %s\n",
1440 array_list->member);
1441 break;
1442 }
1443
1444 if (match && match->devname &&
1445 strcasecmp(match->devname, "<ignore>") == 0) {
1446 if (c->verbose > 0)
1447 pr_err("array %s/%s is "
1448 "explicitly ignored by mdadm.conf\n",
1449 match->container, match->member);
1450 return 2;
1451 }
1452 if (match)
1453 trustworthy = LOCAL;
1454
1455 mdfd = create_mddev(match ? match->devname : NULL,
1456 ra->name,
1457 c->autof,
1458 trustworthy,
1459 chosen_name);
1460 }
1461
1462 if (mdfd < 0) {
1463 pr_err("failed to open %s: %s.\n",
1464 chosen_name, strerror(errno));
1465 return 2;
1466 }
1467
1468 assemble_container_content(st, mdfd, ra, c,
1469 chosen_name);
1470 close(mdfd);
1471 }
1472
1473 /* don't move spares to container with volume being activated
1474 when all volumes are blocked */
1475 if (ra_all == ra_blocked)
1476 return 0;
1477
1478 /* Now move all suitable spares from spare container */
1479 domains = domain_from_array(list, st->ss->name);
1480 memcpy(suuid, uuid_zero, sizeof(int[4]));
1481 if (domains &&
1482 (smp = map_by_uuid(&map, suuid)) != NULL &&
1483 (sfd = open(smp->path, O_RDONLY)) >= 0) {
1484 /* spare container found */
1485 struct supertype *sst =
1486 super_imsm.match_metadata_desc("imsm");
1487 struct mdinfo *sinfo;
1488 unsigned long long min_size = 0;
1489 if (st->ss->min_acceptable_spare_size)
1490 min_size = st->ss->min_acceptable_spare_size(st);
1491 if (!sst->ss->load_container(sst, sfd, NULL)) {
1492 close(sfd);
1493 sinfo = container_choose_spares(sst, min_size,
1494 domains, NULL,
1495 st->ss->name, 0);
1496 sst->ss->free_super(sst);
1497 if (sinfo){
1498 int count = 0;
1499 struct mdinfo *disks = sinfo->devs;
1500 while (disks) {
1501 /* move spare from spare
1502 * container to currently
1503 * assembled one
1504 */
1505 if (move_spare(
1506 smp->path,
1507 devname,
1508 makedev(disks->disk.major,
1509 disks->disk.minor)))
1510 count++;
1511 disks = disks->next;
1512 }
1513 if (count)
1514 pr_err("Added %d spare%s to %s\n",
1515 count, count>1?"s":"", devname);
1516 }
1517 sysfs_free(sinfo);
1518 } else
1519 close(sfd);
1520 }
1521 domain_free(domains);
1522 return 0;
1523 }
1524
1525 /*
1526 * IncrementalRemove - Attempt to see if the passed in device belongs to any
1527 * raid arrays, and if so first fail (if needed) and then remove the device.
1528 *
1529 * @devname - The device we want to remove
1530 * @id_path - name as found in /dev/disk/by-path for this device
1531 *
1532 * Note: the device name must be a kernel name like "sda", so
1533 * that we can find it in /proc/mdstat
1534 */
1535 int IncrementalRemove(char *devname, char *id_path, int verbose)
1536 {
1537 int mdfd;
1538 int rv;
1539 struct mdstat_ent *ent;
1540 struct mddev_dev devlist;
1541
1542 if (!id_path)
1543 dprintf(Name ": incremental removal without --path <id_path> "
1544 "lacks the possibility to re-add new device in this "
1545 "port\n");
1546
1547 if (strchr(devname, '/')) {
1548 pr_err("incremental removal requires a "
1549 "kernel device name, not a file: %s\n", devname);
1550 return 1;
1551 }
1552 ent = mdstat_by_component(devname);
1553 if (!ent) {
1554 pr_err("%s does not appear to be a component "
1555 "of any array\n", devname);
1556 return 1;
1557 }
1558 mdfd = open_dev(ent->devnum);
1559 if (mdfd < 0) {
1560 pr_err("Cannot open array %s!!\n", ent->dev);
1561 free_mdstat(ent);
1562 return 1;
1563 }
1564
1565 if (id_path) {
1566 struct map_ent *map = NULL, *me;
1567 me = map_by_devnum(&map, ent->devnum);
1568 if (me)
1569 policy_save_path(id_path, me);
1570 map_free(map);
1571 }
1572
1573 memset(&devlist, 0, sizeof(devlist));
1574 devlist.devname = devname;
1575 devlist.disposition = 'f';
1576 /* for a container, we must fail each member array */
1577 if (ent->metadata_version &&
1578 strncmp(ent->metadata_version, "external:", 9) == 0) {
1579 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1580 struct mdstat_ent *memb;
1581 for (memb = mdstat ; memb ; memb = memb->next)
1582 if (is_container_member(memb, ent->dev)) {
1583 int subfd = open_dev(memb->devnum);
1584 if (subfd >= 0) {
1585 Manage_subdevs(memb->dev, subfd,
1586 &devlist, verbose, 0,
1587 NULL, 0);
1588 close(subfd);
1589 }
1590 }
1591 free_mdstat(mdstat);
1592 } else
1593 Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1594 devlist.disposition = 'r';
1595 rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0, NULL, 0);
1596 close(mdfd);
1597 free_mdstat(ent);
1598 return rv;
1599 }
Unnamed repository; edit this file 'description' to name the repository.