]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mdadm.h: Introduced unaligned {get,put}_unaligned{16,32}()
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost, int require_homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char *sep;
34 unsigned int l;
35
36 if (strcmp(found, required)==0)
37 return 1;
38 sep = strchr(found, ':');
39 if (!sep)
40 return 0;
41 l = sep - found;
42 if (strncmp(found, "any:", 4) == 0 ||
43 (homehost && strcmp(homehost, "any") == 0) ||
44 !require_homehost ||
45 (homehost && strlen(homehost) == l &&
46 strncmp(found, homehost, l) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep+1, required) == 0)
49 return 1;
50 }
51 return 0;
52 }
53
54 static int is_member_busy(char *metadata_version)
55 {
56 /* check if the given member array is active */
57 struct mdstat_ent *mdstat = mdstat_read(0, 0);
58 struct mdstat_ent *ent;
59 int busy = 0;
60
61 for (ent = mdstat; ent; ent = ent->next) {
62 if (ent->metadata_version == NULL)
63 continue;
64 if (strncmp(ent->metadata_version, "external:", 9) != 0)
65 continue;
66 if (!is_subarray(&ent->metadata_version[9]))
67 continue;
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
70 busy = 1;
71 break;
72 }
73 }
74 free_mdstat(mdstat);
75
76 return busy;
77 }
78
79 static int ident_matches(struct mddev_ident *ident,
80 struct mdinfo *content,
81 struct supertype *tst,
82 char *homehost, int require_homehost,
83 char *update, char *devname)
84 {
85
86 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
87 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
88 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
89 if (devname)
90 pr_err("%s has wrong uuid.\n", devname);
91 return 0;
92 }
93 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
94 name_matches(content->name, ident->name, homehost, require_homehost)==0) {
95 if (devname)
96 pr_err("%s has wrong name.\n", devname);
97 return 0;
98 }
99 if (ident->super_minor != UnSet &&
100 ident->super_minor != content->array.md_minor) {
101 if (devname)
102 pr_err("%s has wrong super-minor.\n",
103 devname);
104 return 0;
105 }
106 if (ident->level != UnSet &&
107 ident->level != content->array.level) {
108 if (devname)
109 pr_err("%s has wrong raid level.\n",
110 devname);
111 return 0;
112 }
113 if (ident->raid_disks != UnSet &&
114 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
115 ident->raid_disks!= content->array.raid_disks) {
116 if (devname)
117 pr_err("%s requires wrong number of drives.\n",
118 devname);
119 return 0;
120 }
121 if (ident->member && ident->member[0]) {
122 /* content->text_version must match */
123 char *s = strchr(content->text_version+1, '/');
124 if (s == NULL) {
125 if (devname)
126 pr_err("%s is not a container and one is required.\n",
127 devname);
128 return 0;
129 } else if (strcmp(ident->member, s+1) != 0) {
130 if (devname)
131 pr_err("skipping wrong member %s is %s\n",
132 content->text_version, devname);
133 return 0;
134 }
135 }
136 return 1;
137 }
138
139 static int select_devices(struct mddev_dev *devlist,
140 struct mddev_ident *ident,
141 struct supertype **stp,
142 struct mdinfo **contentp,
143 struct context *c,
144 int inargv, int auto_assem)
145 {
146 struct mddev_dev *tmpdev;
147 int num_devs;
148 struct supertype *st = *stp;
149 struct mdinfo *content = NULL;
150 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
151 struct domainlist *domains = NULL;
152 dev_t rdev;
153
154 tmpdev = devlist; num_devs = 0;
155 while (tmpdev) {
156 if (tmpdev->used)
157 tmpdev->used = 2;
158 else
159 num_devs++;
160 tmpdev->disposition = 0;
161 tmpdev = tmpdev->next;
162 }
163
164 /* first walk the list of devices to find a consistent set
165 * that match the criterea, if that is possible.
166 * We flag the ones we like with 'used'.
167 */
168 for (tmpdev = devlist;
169 tmpdev;
170 tmpdev = tmpdev ? tmpdev->next : NULL) {
171 char *devname = tmpdev->devname;
172 int dfd;
173 struct supertype *tst;
174 struct dev_policy *pol = NULL;
175 int found_container = 0;
176
177 if (tmpdev->used > 1)
178 continue;
179
180 if (ident->container) {
181 if (ident->container[0] == '/' &&
182 !same_dev(ident->container, devname)) {
183 if (report_mismatch)
184 pr_err("%s is not the container required (%s)\n",
185 devname, ident->container);
186 continue;
187 }
188 } else if (ident->devices &&
189 !match_oneof(ident->devices, devname)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
192 * complicated.
193 */
194 if (report_mismatch)
195 pr_err("%s is not one of %s\n", devname, ident->devices);
196 continue;
197 }
198
199 tst = dup_super(st);
200
201 dfd = dev_open(devname, O_RDONLY);
202 if (dfd < 0) {
203 if (report_mismatch)
204 pr_err("cannot open device %s: %s\n",
205 devname, strerror(errno));
206 tmpdev->used = 2;
207 } else if (!fstat_is_blkdev(dfd, devname, &rdev)) {
208 tmpdev->used = 2;
209 } else if (must_be_container(dfd)) {
210 if (st) {
211 /* already found some components, this cannot
212 * be another one.
213 */
214 if (report_mismatch)
215 pr_err("%s is a container, but we are looking for components\n",
216 devname);
217 tmpdev->used = 2;
218 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
219 if (report_mismatch)
220 pr_err("not a recognisable container: %s\n",
221 devname);
222 tmpdev->used = 2;
223 } else if (!tst->ss->load_container ||
224 tst->ss->load_container(tst, dfd, NULL)) {
225 if (report_mismatch)
226 pr_err("no correct container type: %s\n",
227 devname);
228 tmpdev->used = 2;
229 } else if (auto_assem &&
230 !conf_test_metadata(tst->ss->name,
231 (pol = devid_policy(rdev)),
232 tst->ss->match_home(tst, c->homehost) == 1)) {
233 if (report_mismatch)
234 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
235 devname, tst->ss->name);
236 tmpdev->used = 2;
237 } else
238 found_container = 1;
239 } else {
240 if (!tst && (tst = guess_super(dfd)) == NULL) {
241 if (report_mismatch)
242 pr_err("no recogniseable superblock on %s\n",
243 devname);
244 tmpdev->used = 2;
245 } else if ((tst->ignore_hw_compat = 0),
246 tst->ss->load_super(tst, dfd,
247 report_mismatch ? devname : NULL)) {
248 if (report_mismatch)
249 pr_err("no RAID superblock on %s\n",
250 devname);
251 tmpdev->used = 2;
252 } else if (tst->ss->compare_super == NULL) {
253 if (report_mismatch)
254 pr_err("Cannot assemble %s metadata on %s\n",
255 tst->ss->name, devname);
256 tmpdev->used = 2;
257 } else if (auto_assem && st == NULL &&
258 !conf_test_metadata(tst->ss->name,
259 (pol = devid_policy(rdev)),
260 tst->ss->match_home(tst, c->homehost) == 1)) {
261 if (report_mismatch)
262 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
263 devname, tst->ss->name);
264 tmpdev->used = 2;
265 }
266 }
267 if (dfd >= 0) close(dfd);
268 if (tmpdev->used == 2) {
269 if (auto_assem || !inargv)
270 /* Ignore unrecognised devices during auto-assembly */
271 goto loop;
272 if (ident->uuid_set || ident->name[0] ||
273 ident->super_minor != UnSet)
274 /* Ignore unrecognised device if looking for
275 * specific array */
276 goto loop;
277
278 pr_err("%s has no superblock - assembly aborted\n",
279 devname);
280 if (st)
281 st->ss->free_super(st);
282 dev_policy_free(pol);
283 domain_free(domains);
284 if (tst)
285 tst->ss->free_super(tst);
286 return -1;
287 }
288
289 if (found_container) {
290 /* tmpdev is a container. We need to be either
291 * looking for a member, or auto-assembling
292 */
293 /* should be safe to try an exclusive open now, we
294 * have rejected anything that some other mdadm might
295 * be looking at
296 */
297 dfd = dev_open(devname, O_RDONLY | O_EXCL);
298 if (dfd < 0) {
299 if (report_mismatch)
300 pr_err("%s is busy - skipping\n", devname);
301 goto loop;
302 }
303 close(dfd);
304
305 if (ident->container && ident->container[0] != '/') {
306 /* we have a uuid */
307 int uuid[4];
308
309 content = *contentp;
310 tst->ss->getinfo_super(tst, content, NULL);
311
312 if (!parse_uuid(ident->container, uuid) ||
313 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
314 if (report_mismatch)
315 pr_err("%s has wrong UUID to be required container\n",
316 devname);
317 goto loop;
318 }
319 }
320 /* It is worth looking inside this container.
321 */
322 if (c->verbose > 0)
323 pr_err("looking in container %s\n",
324 devname);
325
326 for (content = tst->ss->container_content(tst, NULL);
327 content;
328 content = content->next) {
329
330 if (!ident_matches(ident, content, tst,
331 c->homehost, c->require_homehost,
332 c->update,
333 report_mismatch ? devname : NULL))
334 /* message already printed */;
335 else if (is_member_busy(content->text_version)) {
336 if (report_mismatch)
337 pr_err("member %s in %s is already assembled\n",
338 content->text_version,
339 devname);
340 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
341 /* do not assemble arrays with unsupported configurations */
342 pr_err("Cannot activate member %s in %s.\n",
343 content->text_version,
344 devname);
345 } else
346 break;
347 }
348 if (!content) {
349 tmpdev->used = 2;
350 goto loop; /* empty container */
351 }
352
353 st = tst; tst = NULL;
354 if (!auto_assem && inargv && tmpdev->next != NULL) {
355 pr_err("%s is a container, but is not only device given: confused and aborting\n",
356 devname);
357 st->ss->free_super(st);
358 dev_policy_free(pol);
359 domain_free(domains);
360 return -1;
361 }
362 if (c->verbose > 0)
363 pr_err("found match on member %s in %s\n",
364 content->text_version, devname);
365
366 /* make sure we finished the loop */
367 tmpdev = NULL;
368 goto loop;
369 } else {
370 content = *contentp;
371 tst->ss->getinfo_super(tst, content, NULL);
372
373 if (!ident_matches(ident, content, tst,
374 c->homehost, c->require_homehost,
375 c->update,
376 report_mismatch ? devname : NULL))
377 goto loop;
378
379 if (auto_assem) {
380 /* Never auto-assemble things that conflict
381 * with mdadm.conf in some way
382 */
383 struct mddev_ident *match;
384 int rv = 0;
385
386 match = conf_match(tst, content, devname,
387 report_mismatch ? c->verbose : -1,
388 &rv);
389 if (!match && rv == 2)
390 goto loop;
391 if (match && match->devname &&
392 strcasecmp(match->devname, "<ignore>") == 0) {
393 if (report_mismatch)
394 pr_err("%s is a member of an explicitly ignored array\n",
395 devname);
396 goto loop;
397 }
398 if (match && !ident_matches(match, content, tst,
399 c->homehost, c->require_homehost,
400 c->update,
401 report_mismatch ? devname : NULL))
402 /* Array exists in mdadm.conf but some
403 * details don't match, so reject it
404 */
405 goto loop;
406 }
407
408 /* should be safe to try an exclusive open now, we
409 * have rejected anything that some other mdadm might
410 * be looking at
411 */
412 dfd = dev_open(devname, O_RDONLY | O_EXCL);
413 if (dfd < 0) {
414 if (report_mismatch)
415 pr_err("%s is busy - skipping\n", devname);
416 goto loop;
417 }
418 close(dfd);
419
420 if (st == NULL)
421 st = dup_super(tst);
422 if (st->minor_version == -1)
423 st->minor_version = tst->minor_version;
424
425 if (memcmp(content->uuid, uuid_zero,
426 sizeof(int[4])) == 0) {
427 /* this is a floating spare. It cannot define
428 * an array unless there are no more arrays of
429 * this type to be found. It can be included
430 * in an array of this type though.
431 */
432 tmpdev->used = 3;
433 goto loop;
434 }
435
436 if (st->ss != tst->ss ||
437 st->minor_version != tst->minor_version ||
438 st->ss->compare_super(st, tst) != 0) {
439 /* Some mismatch. If exactly one array matches this host,
440 * we can resolve on that one.
441 * Or, if we are auto assembling, we just ignore the second
442 * for now.
443 */
444 if (auto_assem)
445 goto loop;
446 if (c->homehost) {
447 int first = st->ss->match_home(st, c->homehost);
448 int last = tst->ss->match_home(tst, c->homehost);
449 if (first != last &&
450 (first == 1 || last == 1)) {
451 /* We can do something */
452 if (first) {/* just ignore this one */
453 if (report_mismatch)
454 pr_err("%s misses out due to wrong homehost\n",
455 devname);
456 goto loop;
457 } else { /* reject all those sofar */
458 struct mddev_dev *td;
459 if (report_mismatch)
460 pr_err("%s overrides previous devices due to good homehost\n",
461 devname);
462 for (td=devlist; td != tmpdev; td=td->next)
463 if (td->used == 1)
464 td->used = 0;
465 tmpdev->used = 1;
466 goto loop;
467 }
468 }
469 }
470 pr_err("superblock on %s doesn't match others - assembly aborted\n",
471 devname);
472 tst->ss->free_super(tst);
473 st->ss->free_super(st);
474 dev_policy_free(pol);
475 domain_free(domains);
476 return -1;
477 }
478 tmpdev->used = 1;
479 }
480 loop:
481 /* Collect domain information from members only */
482 if (tmpdev && tmpdev->used == 1) {
483 if (!pol)
484 pol = devid_policy(rdev);
485 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
486 }
487 dev_policy_free(pol);
488 pol = NULL;
489 if (tst)
490 tst->ss->free_super(tst);
491 }
492
493 /* Check if we found some imsm spares but no members */
494 if ((auto_assem ||
495 (ident->uuid_set &&
496 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
497 (!st || !st->sb))
498 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
499 if (tmpdev->used != 3)
500 continue;
501 tmpdev->used = 1;
502 content = *contentp;
503
504 if (!st->sb) {
505 /* we need sb from one of the spares */
506 int dfd = dev_open(tmpdev->devname, O_RDONLY);
507 if (dfd < 0 ||
508 st->ss->load_super(st, dfd, NULL))
509 tmpdev->used = 2;
510 if (dfd > 0)
511 close(dfd);
512 }
513 }
514
515 /* Now reject spares that don't match domains of identified members */
516 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
517 if (tmpdev->used != 3)
518 continue;
519 if (!stat_is_blkdev(tmpdev->devname, &rdev)) {
520 tmpdev->used = 2;
521 } else {
522 struct dev_policy *pol = devid_policy(rdev);
523 int dt = domain_test(domains, pol, NULL);
524 if (inargv && dt != 0)
525 /* take this spare as domains match
526 * if there are any */
527 tmpdev->used = 1;
528 else if (!inargv && dt == 1)
529 /* device wasn't explicitly listed, so need
530 * explicit domain match - which we have */
531 tmpdev->used = 1;
532 else
533 /* if domains don't match mark as unused */
534 tmpdev->used = 0;
535 dev_policy_free(pol);
536 }
537 }
538 domain_free(domains);
539 *stp = st;
540 if (st && st->sb && content == *contentp)
541 st->ss->getinfo_super(st, content, NULL);
542 *contentp = content;
543
544 return num_devs;
545 }
546
547 struct devs {
548 char *devname;
549 int uptodate; /* set once we decide that this device is as
550 * recent as everything else in the array.
551 */
552 int included; /* set if the device is already in the array
553 * due to a previous '-I'
554 */
555 struct mdinfo i;
556 };
557
558 static int load_devices(struct devs *devices, char *devmap,
559 struct mddev_ident *ident, struct supertype **stp,
560 struct mddev_dev *devlist, struct context *c,
561 struct mdinfo *content,
562 int mdfd, char *mddev,
563 int *most_recentp, int *bestcntp, int **bestp,
564 int inargv)
565 {
566 struct mddev_dev *tmpdev;
567 int devcnt = 0;
568 int nextspare = 0;
569 int bitmap_done = 0;
570 int most_recent = -1;
571 int bestcnt = 0;
572 int *best = *bestp;
573 struct supertype *st = *stp;
574
575 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
576 char *devname = tmpdev->devname;
577 struct stat stb;
578 struct supertype *tst;
579 int i;
580 int dfd;
581 int disk_state;
582
583 if (tmpdev->used != 1)
584 continue;
585 /* looks like a good enough match to update the super block if needed */
586 if (c->update) {
587 /* prepare useful information in info structures */
588 struct stat stb2;
589 int err;
590 fstat(mdfd, &stb2);
591
592 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
593 random_uuid((__u8 *)ident->uuid);
594
595 if (strcmp(c->update, "ppl") == 0 &&
596 ident->bitmap_fd >= 0) {
597 pr_err("PPL is not compatible with bitmap\n");
598 close(mdfd);
599 free(devices);
600 free(devmap);
601 return -1;
602 }
603
604 dfd = dev_open(devname,
605 tmpdev->disposition == 'I'
606 ? O_RDWR : (O_RDWR|O_EXCL));
607
608 tst = dup_super(st);
609 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
610 pr_err("cannot re-read metadata from %s - aborting\n",
611 devname);
612 if (dfd >= 0)
613 close(dfd);
614 close(mdfd);
615 free(devices);
616 free(devmap);
617 tst->ss->free_super(tst);
618 free(tst);
619 *stp = st;
620 return -1;
621 }
622 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
623
624 memcpy(content->uuid, ident->uuid, 16);
625 strcpy(content->name, ident->name);
626 content->array.md_minor = minor(stb2.st_rdev);
627
628 if (strcmp(c->update, "byteorder") == 0)
629 err = 0;
630 else if (strcmp(c->update, "home-cluster") == 0) {
631 tst->cluster_name = c->homecluster;
632 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
633 } else if (strcmp(c->update, "nodes") == 0) {
634 tst->nodes = c->nodes;
635 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
636 } else if (strcmp(c->update, "revert-reshape") == 0 &&
637 c->invalid_backup)
638 err = tst->ss->update_super(tst, content,
639 "revert-reshape-nobackup",
640 devname, c->verbose,
641 ident->uuid_set,
642 c->homehost);
643 else
644 err = tst->ss->update_super(tst, content, c->update,
645 devname, c->verbose,
646 ident->uuid_set,
647 c->homehost);
648 if (err < 0) {
649 if (err == -1)
650 pr_err("--update=%s not understood for %s metadata\n",
651 c->update, tst->ss->name);
652 tst->ss->free_super(tst);
653 free(tst);
654 close(mdfd);
655 close(dfd);
656 free(devices);
657 free(devmap);
658 *stp = st;
659 return -1;
660 }
661 if (strcmp(c->update, "uuid")==0 &&
662 !ident->uuid_set) {
663 ident->uuid_set = 1;
664 memcpy(ident->uuid, content->uuid, 16);
665 }
666 if (tst->ss->store_super(tst, dfd))
667 pr_err("Could not re-write superblock on %s.\n",
668 devname);
669
670 if (strcmp(c->update, "uuid")==0 &&
671 ident->bitmap_fd >= 0 && !bitmap_done) {
672 if (bitmap_update_uuid(ident->bitmap_fd,
673 content->uuid,
674 tst->ss->swapuuid) != 0)
675 pr_err("Could not update uuid on external bitmap.\n");
676 else
677 bitmap_done = 1;
678 }
679 } else {
680 dfd = dev_open(devname,
681 tmpdev->disposition == 'I'
682 ? O_RDWR : (O_RDWR|O_EXCL));
683 tst = dup_super(st);
684
685 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
686 pr_err("cannot re-read metadata from %s - aborting\n",
687 devname);
688 if (dfd >= 0)
689 close(dfd);
690 close(mdfd);
691 free(devices);
692 free(devmap);
693 tst->ss->free_super(tst);
694 free(tst);
695 *stp = st;
696 return -1;
697 }
698 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
699 }
700
701 fstat(dfd, &stb);
702 close(dfd);
703
704 if (c->verbose > 0)
705 pr_err("%s is identified as a member of %s, slot %d%s.\n",
706 devname, mddev, content->disk.raid_disk,
707 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
708 devices[devcnt].devname = devname;
709 devices[devcnt].uptodate = 0;
710 devices[devcnt].included = (tmpdev->disposition == 'I');
711 devices[devcnt].i = *content;
712 devices[devcnt].i.disk.major = major(stb.st_rdev);
713 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
714
715 disk_state = devices[devcnt].i.disk.state & ~((1<<MD_DISK_FAILFAST) |
716 (1<<MD_DISK_WRITEMOSTLY));
717 if (disk_state == ((1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC))) {
718 if (most_recent < 0 ||
719 devices[devcnt].i.events
720 > devices[most_recent].i.events) {
721 struct supertype *tmp = tst;
722 tst = st;
723 st = tmp;
724 most_recent = devcnt;
725 }
726 }
727 tst->ss->free_super(tst);
728 free(tst);
729
730 if (content->array.level == LEVEL_MULTIPATH)
731 /* with multipath, the raid_disk from the superblock is meaningless */
732 i = devcnt;
733 else
734 i = devices[devcnt].i.disk.raid_disk;
735 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
736 if (nextspare < content->array.raid_disks*2)
737 nextspare = content->array.raid_disks*2;
738 i = nextspare++;
739 } else {
740 /* i is raid_disk - double it so there is room for
741 * replacements */
742 i *= 2;
743 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
744 i++;
745 if (i >= content->array.raid_disks*2 &&
746 i >= nextspare)
747 nextspare = i+1;
748 }
749 if (i < 10000) {
750 if (i >= bestcnt) {
751 int newbestcnt = i+10;
752 int *newbest = xmalloc(sizeof(int)*newbestcnt);
753 int c;
754 for (c=0; c < newbestcnt; c++)
755 if (c < bestcnt)
756 newbest[c] = best[c];
757 else
758 newbest[c] = -1;
759 if (best)free(best);
760 best = newbest;
761 bestcnt = newbestcnt;
762 }
763 if (best[i] >=0 &&
764 devices[best[i]].i.events ==
765 devices[devcnt].i.events &&
766 (devices[best[i]].i.disk.minor !=
767 devices[devcnt].i.disk.minor) &&
768 st->ss == &super0 &&
769 content->array.level != LEVEL_MULTIPATH) {
770 /* two different devices with identical superblock.
771 * Could be a mis-detection caused by overlapping
772 * partitions. fail-safe.
773 */
774 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
775 " If they are really different, please --zero the superblock on one\n"
776 " If they are the same or overlap, please remove one from %s.\n",
777 devices[best[i]].devname, devname,
778 inargv ? "the list" :
779 "the\n DEVICE list in mdadm.conf"
780 );
781 close(mdfd);
782 free(devices);
783 free(devmap);
784 *stp = st;
785 return -1;
786 }
787 if (best[i] == -1 || (devices[best[i]].i.events
788 < devices[devcnt].i.events))
789 best[i] = devcnt;
790 else if (st->ss == &super_imsm)
791 best[i+1] = devcnt;
792 }
793 devcnt++;
794 }
795 if (most_recent >= 0)
796 *most_recentp = most_recent;
797 *bestcntp = bestcnt;
798 *bestp = best;
799 *stp = st;
800 return devcnt;
801 }
802
803 static int force_array(struct mdinfo *content,
804 struct devs *devices,
805 int *best, int bestcnt, char *avail,
806 int most_recent,
807 struct supertype *st,
808 struct context *c)
809 {
810 int okcnt = 0;
811 while (!enough(content->array.level, content->array.raid_disks,
812 content->array.layout, 1,
813 avail) ||
814 (content->reshape_active && content->delta_disks > 0 &&
815 !enough(content->array.level, (content->array.raid_disks
816 - content->delta_disks),
817 content->new_layout, 1, avail))) {
818 /* Choose the newest best drive which is
819 * not up-to-date, update the superblock
820 * and add it.
821 */
822 int fd;
823 struct supertype *tst;
824 unsigned long long current_events;
825 int chosen_drive = -1;
826 int i;
827
828 for (i = 0;
829 i < content->array.raid_disks * 2 && i < bestcnt;
830 i += 2) {
831 int j = best[i];
832 if (j < 0)
833 continue;
834 if (devices[j].uptodate)
835 continue;
836 if (devices[j].i.recovery_start != MaxSector) {
837 int delta;
838 if (!devices[j].i.reshape_active ||
839 devices[j].i.delta_disks <= 0)
840 continue;
841 /* When increasing number of devices, an
842 * added device also appears to be
843 * recovering. It is safe to include it
844 * as long as it won't be a source of
845 * data.
846 * For now, just allow for last data
847 * devices in RAID4 or last devices in RAID4/5/6.
848 */
849 delta = devices[j].i.delta_disks;
850 if (devices[j].i.array.level >= 4 &&
851 devices[j].i.array.level <= 6 &&
852 i/2 >= content->array.raid_disks - delta)
853 /* OK */;
854 else if (devices[j].i.array.level == 4 &&
855 i/2 >= content->array.raid_disks - delta - 1)
856 /* OK */;
857 else
858 continue;
859 } else if (devices[j].i.reshape_active !=
860 content->reshape_active ||
861 (devices[j].i.reshape_active &&
862 devices[j].i.reshape_progress !=
863 content->reshape_progress))
864 /* Here, it may be a source of data. If two
865 * devices claim different progresses, it
866 * means that reshape boundaries differ for
867 * their own devices. Kernel will only treat
868 * the first one as reshape progress and
869 * go on. It may cause disaster, so avoid it.
870 */
871 continue;
872 if (chosen_drive < 0 ||
873 devices[j].i.events
874 > devices[chosen_drive].i.events)
875 chosen_drive = j;
876 }
877 if (chosen_drive < 0)
878 break;
879 current_events = devices[chosen_drive].i.events;
880 add_another:
881 if (c->verbose >= 0)
882 pr_err("forcing event count in %s(%d) from %d up to %d\n",
883 devices[chosen_drive].devname,
884 devices[chosen_drive].i.disk.raid_disk,
885 (int)(devices[chosen_drive].i.events),
886 (int)(devices[most_recent].i.events));
887 fd = dev_open(devices[chosen_drive].devname,
888 devices[chosen_drive].included ? O_RDWR
889 : (O_RDWR|O_EXCL));
890 if (fd < 0) {
891 pr_err("Couldn't open %s for write - not updating\n",
892 devices[chosen_drive].devname);
893 devices[chosen_drive].i.events = 0;
894 continue;
895 }
896 tst = dup_super(st);
897 if (tst->ss->load_super(tst,fd, NULL)) {
898 close(fd);
899 pr_err("RAID superblock disappeared from %s - not updating.\n",
900 devices[chosen_drive].devname);
901 devices[chosen_drive].i.events = 0;
902 continue;
903 }
904 content->events = devices[most_recent].i.events;
905 tst->ss->update_super(tst, content, "force-one",
906 devices[chosen_drive].devname, c->verbose,
907 0, NULL);
908
909 if (tst->ss->store_super(tst, fd)) {
910 close(fd);
911 pr_err("Could not re-write superblock on %s\n",
912 devices[chosen_drive].devname);
913 devices[chosen_drive].i.events = 0;
914 tst->ss->free_super(tst);
915 continue;
916 }
917 close(fd);
918 devices[chosen_drive].i.events = devices[most_recent].i.events;
919 devices[chosen_drive].uptodate = 1;
920 avail[chosen_drive] = 1;
921 okcnt++;
922 tst->ss->free_super(tst);
923 /* If there are any other drives of the same vintage,
924 * add them in as well. We can't lose and we might gain
925 */
926 for (i = 0;
927 i < content->array.raid_disks * 2 && i < bestcnt ;
928 i += 2) {
929 int j = best[i];
930 if (j >= 0 &&
931 !devices[j].uptodate &&
932 devices[j].i.recovery_start == MaxSector &&
933 devices[j].i.events == current_events &&
934 ((!devices[j].i.reshape_active &&
935 !content->reshape_active) ||
936 (devices[j].i.reshape_active ==
937 content->reshape_active &&
938 devices[j].i.reshape_progress ==
939 content->reshape_progress))) {
940 chosen_drive = j;
941 goto add_another;
942 }
943 }
944 }
945 return okcnt;
946 }
947
948 static int start_array(int mdfd,
949 char *mddev,
950 struct mdinfo *content,
951 struct supertype *st,
952 struct mddev_ident *ident,
953 int *best, int bestcnt,
954 int chosen_drive,
955 struct devs *devices,
956 unsigned int okcnt,
957 unsigned int sparecnt,
958 unsigned int rebuilding_cnt,
959 unsigned int journalcnt,
960 struct context *c,
961 int clean, char *avail,
962 int start_partial_ok,
963 int err_ok,
964 int was_forced
965 )
966 {
967 int rv;
968 int i;
969 unsigned int req_cnt;
970
971 if (content->journal_device_required && (content->journal_clean == 0)) {
972 if (!c->force) {
973 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
974 return 1;
975 }
976 pr_err("Journal is missing or stale, starting array read only.\n");
977 c->readonly = 1;
978 }
979
980 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
981 clean = 1;
982
983 rv = set_array_info(mdfd, st, content);
984 if (rv && !err_ok) {
985 pr_err("failed to set array info for %s: %s\n",
986 mddev, strerror(errno));
987 return 1;
988 }
989 if (ident->bitmap_fd >= 0) {
990 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
991 pr_err("SET_BITMAP_FILE failed.\n");
992 return 1;
993 }
994 } else if (ident->bitmap_file) {
995 /* From config file */
996 int bmfd = open(ident->bitmap_file, O_RDWR);
997 if (bmfd < 0) {
998 pr_err("Could not open bitmap file %s\n",
999 ident->bitmap_file);
1000 return 1;
1001 }
1002 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1003 pr_err("Failed to set bitmapfile for %s\n", mddev);
1004 close(bmfd);
1005 return 1;
1006 }
1007 close(bmfd);
1008 }
1009
1010 /* First, add the raid disks, but add the chosen one last */
1011 for (i = 0; i <= bestcnt; i++) {
1012 int j;
1013 if (i < bestcnt) {
1014 j = best[i];
1015 if (j == chosen_drive)
1016 continue;
1017 } else
1018 j = chosen_drive;
1019
1020 if (j >= 0 && !devices[j].included) {
1021 int dfd;
1022
1023 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1024 if (dfd >= 0) {
1025 remove_partitions(dfd);
1026 close(dfd);
1027 }
1028 rv = add_disk(mdfd, st, content, &devices[j].i);
1029
1030 if (rv) {
1031 pr_err("failed to add %s to %s: %s\n",
1032 devices[j].devname, mddev,
1033 strerror(errno));
1034 if (i < content->array.raid_disks * 2 ||
1035 i == bestcnt)
1036 okcnt--;
1037 else
1038 sparecnt--;
1039 } else if (c->verbose > 0) {
1040 pr_err("added %s to %s as %d%s%s\n",
1041 devices[j].devname, mddev,
1042 devices[j].i.disk.raid_disk,
1043 devices[j].uptodate?"":
1044 " (possibly out of date)",
1045 (devices[j].i.disk.state &
1046 (1<<MD_DISK_REPLACEMENT)) ?
1047 " replacement":"");
1048 }
1049 } else if (j >= 0) {
1050 if (c->verbose > 0)
1051 pr_err("%s is already in %s as %d\n",
1052 devices[j].devname, mddev,
1053 devices[j].i.disk.raid_disk);
1054 } else if (c->verbose > 0 &&
1055 i < content->array.raid_disks * 2 && (i & 1) == 0)
1056 pr_err("no uptodate device for slot %d of %s\n",
1057 i/2, mddev);
1058 }
1059
1060 if (content->array.level == LEVEL_CONTAINER) {
1061 if (c->verbose >= 0) {
1062 pr_err("Container %s has been assembled with %d drive%s",
1063 mddev, okcnt + sparecnt + journalcnt,
1064 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1065 if (okcnt < (unsigned)content->array.raid_disks)
1066 fprintf(stderr, " (out of %d)",
1067 content->array.raid_disks);
1068 fprintf(stderr, "\n");
1069 }
1070
1071 if (st->ss->validate_container) {
1072 struct mdinfo *devices_list;
1073 struct mdinfo *info_devices;
1074 unsigned int count;
1075
1076 devices_list = NULL;
1077 info_devices = xmalloc(sizeof(struct mdinfo) *
1078 (okcnt + sparecnt));
1079 for (count = 0; count < okcnt + sparecnt; count++) {
1080 info_devices[count] = devices[count].i;
1081 info_devices[count].next = devices_list;
1082 devices_list = &info_devices[count];
1083 }
1084 if (st->ss->validate_container(devices_list))
1085 pr_err("Mismatch detected!\n");
1086 free(info_devices);
1087 }
1088
1089 st->ss->free_super(st);
1090 sysfs_uevent(content, "change");
1091 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1092 /* Was partial, is still partial, so signal an error
1093 * to ensure we don't retry */
1094 return 1;
1095 return 0;
1096 }
1097
1098 /* Get number of in-sync devices according to the superblock.
1099 * We must have this number to start the array without -s or -R
1100 */
1101 req_cnt = content->array.working_disks;
1102
1103 if (c->runstop == 1 ||
1104 (c->runstop <= 0 &&
1105 (enough(content->array.level, content->array.raid_disks,
1106 content->array.layout, clean, avail) &&
1107 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1108 /* This array is good-to-go.
1109 * If a reshape is in progress then we might need to
1110 * continue monitoring it. In that case we start
1111 * it read-only and let the grow code make it writable.
1112 */
1113 int rv;
1114
1115 if (content->reshape_active &&
1116 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1117 content->delta_disks <= 0) {
1118 if (!c->backup_file) {
1119 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1120 mddev);
1121 pr_err("Please provided one with \"--backup-file=...\"\n");
1122 if (c->update &&
1123 strcmp(c->update, "revert-reshape") == 0)
1124 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1125 return 1;
1126 }
1127 rv = sysfs_set_str(content, NULL,
1128 "array_state", "readonly");
1129 if (rv == 0)
1130 rv = Grow_continue(mdfd, st, content,
1131 c->backup_file, 0,
1132 c->freeze_reshape);
1133 } else if (c->readonly &&
1134 sysfs_attribute_available(content, NULL,
1135 "array_state")) {
1136 rv = sysfs_set_str(content, NULL,
1137 "array_state", "readonly");
1138 } else
1139 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1140 reopen_mddev(mdfd); /* drop O_EXCL */
1141 if (rv == 0) {
1142 if (c->verbose >= 0) {
1143 pr_err("%s has been started with %d drive%s",
1144 mddev, okcnt, okcnt==1?"":"s");
1145 if (okcnt < (unsigned)content->array.raid_disks)
1146 fprintf(stderr, " (out of %d)",
1147 content->array.raid_disks);
1148 if (rebuilding_cnt)
1149 fprintf(stderr, "%s %d rebuilding",
1150 sparecnt?",":" and",
1151 rebuilding_cnt);
1152 if (sparecnt)
1153 fprintf(stderr, " and %d spare%s",
1154 sparecnt,
1155 sparecnt == 1 ? "" : "s");
1156 if (content->journal_clean)
1157 fprintf(stderr, " and %d journal",
1158 journalcnt);
1159 fprintf(stderr, ".\n");
1160 }
1161 if (content->reshape_active &&
1162 content->array.level >= 4 &&
1163 content->array.level <= 6) {
1164 /* might need to increase the size
1165 * of the stripe cache - default is 256
1166 */
1167 int chunk_size = content->array.chunk_size;
1168
1169 if (content->reshape_active &&
1170 content->new_chunk > chunk_size)
1171 chunk_size = content->new_chunk;
1172 if (256 < 4 * ((chunk_size+4065)/4096)) {
1173 struct mdinfo *sra;
1174
1175 sra = sysfs_read(mdfd, NULL, 0);
1176 if (sra)
1177 sysfs_set_num(sra, NULL,
1178 "stripe_cache_size",
1179 (4 * chunk_size / 4096) + 1);
1180 sysfs_free(sra);
1181 }
1182 }
1183 if (okcnt < (unsigned)content->array.raid_disks) {
1184 /* If any devices did not get added
1185 * because the kernel rejected them based
1186 * on event count, try adding them
1187 * again providing the action policy is
1188 * 're-add' or greater. The bitmap
1189 * might allow them to be included, or
1190 * they will become spares.
1191 */
1192 for (i = 0; i < bestcnt; i++) {
1193 int j = best[i];
1194 if (j >= 0 && !devices[j].uptodate) {
1195 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1196 continue;
1197 rv = add_disk(mdfd, st, content,
1198 &devices[j].i);
1199 if (rv == 0 && c->verbose >= 0)
1200 pr_err("%s has been re-added.\n",
1201 devices[j].devname);
1202 }
1203 }
1204 }
1205 if (content->array.level == 6 &&
1206 okcnt + 1 == (unsigned)content->array.raid_disks &&
1207 was_forced) {
1208 struct mdinfo *sra;
1209
1210 sra = sysfs_read(mdfd, NULL, 0);
1211 if (sra)
1212 sysfs_set_str(sra, NULL,
1213 "sync_action", "repair");
1214 sysfs_free(sra);
1215 }
1216 return 0;
1217 }
1218 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1219
1220 if (!enough(content->array.level, content->array.raid_disks,
1221 content->array.layout, 1, avail))
1222 pr_err("Not enough devices to start the array.\n");
1223 else if (!enough(content->array.level,
1224 content->array.raid_disks,
1225 content->array.layout, clean, avail))
1226 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1227
1228 return 1;
1229 }
1230 if (c->runstop == -1) {
1231 pr_err("%s assembled from %d drive%s",
1232 mddev, okcnt, okcnt == 1 ? "" : "s");
1233 if (okcnt != (unsigned)content->array.raid_disks)
1234 fprintf(stderr, " (out of %d)",
1235 content->array.raid_disks);
1236 fprintf(stderr, ", but not started.\n");
1237 return 2;
1238 }
1239 if (c->verbose >= -1) {
1240 pr_err("%s assembled from %d drive%s",
1241 mddev, okcnt, okcnt == 1 ? "" : "s");
1242 if (rebuilding_cnt)
1243 fprintf(stderr, "%s %d rebuilding",
1244 sparecnt ? "," : " and", rebuilding_cnt);
1245 if (sparecnt)
1246 fprintf(stderr, " and %d spare%s", sparecnt,
1247 sparecnt == 1 ? "" : "s");
1248 if (!enough(content->array.level, content->array.raid_disks,
1249 content->array.layout, 1, avail))
1250 fprintf(stderr, " - not enough to start the array.\n");
1251 else if (!enough(content->array.level,
1252 content->array.raid_disks,
1253 content->array.layout, clean, avail))
1254 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1255 else {
1256 if (req_cnt == (unsigned)content->array.raid_disks)
1257 fprintf(stderr, " - need all %d to start it",
1258 req_cnt);
1259 else
1260 fprintf(stderr, " - need %d to start", req_cnt);
1261 fprintf(stderr, " (use --run to insist).\n");
1262 }
1263 }
1264 return 1;
1265 }
1266
1267 int Assemble(struct supertype *st, char *mddev,
1268 struct mddev_ident *ident,
1269 struct mddev_dev *devlist,
1270 struct context *c)
1271 {
1272 /*
1273 * The task of Assemble is to find a collection of
1274 * devices that should (according to their superblocks)
1275 * form an array, and to give this collection to the MD driver.
1276 * In Linux-2.4 and later, this involves submitting a
1277 * SET_ARRAY_INFO ioctl with no arg - to prepare
1278 * the array - and then submit a number of
1279 * ADD_NEW_DISK ioctls to add disks into
1280 * the array. Finally RUN_ARRAY might
1281 * be submitted to start the array.
1282 *
1283 * Much of the work of Assemble is in finding and/or
1284 * checking the disks to make sure they look right.
1285 *
1286 * If mddev is not set, then scan must be set and we
1287 * read through the config file for dev+uuid mapping
1288 * We recurse, setting mddev, for each device that
1289 * - isn't running
1290 * - has a valid uuid (or any uuid if !uuidset)
1291 *
1292 * If mddev is set, we try to determine state of md.
1293 * check version - must be at least 0.90.0
1294 * check kernel version. must be at least 2.4.
1295 * If not, we can possibly fall back on START_ARRAY
1296 * Try to GET_ARRAY_INFO.
1297 * If possible, give up
1298 * If not, try to STOP_ARRAY just to make sure
1299 *
1300 * If !uuidset and scan, look in conf-file for uuid
1301 * If not found, give up
1302 * If !devlist and scan and uuidset, get list of devs from conf-file
1303 *
1304 * For each device:
1305 * Check superblock - discard if bad
1306 * Check uuid (set if we don't have one) - discard if no match
1307 * Check superblock similarity if we have a superblock - discard if different
1308 * Record events, devicenum
1309 * This should give us a list of devices for the array
1310 * We should collect the most recent event number
1311 *
1312 * Count disks with recent enough event count
1313 * While force && !enough disks
1314 * Choose newest rejected disks, update event count
1315 * mark clean and rewrite superblock
1316 * If recent kernel:
1317 * SET_ARRAY_INFO
1318 * foreach device with recent events : ADD_NEW_DISK
1319 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1320 * If old kernel:
1321 * Check the device numbers in superblock are right
1322 * update superblock if any changes
1323 * START_ARRAY
1324 *
1325 */
1326 int rv = -1;
1327 int mdfd = -1;
1328 int clean;
1329 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1330 ident->super_minor == UnSet && ident->name[0] == 0 &&
1331 (ident->container == NULL || ident->member == NULL));
1332 struct devs *devices = NULL;
1333 char *devmap;
1334 int *best = NULL; /* indexed by raid_disk */
1335 int bestcnt = 0;
1336 int devcnt;
1337 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1338 int journal_clean = 0;
1339 int i;
1340 int was_forced = 0;
1341 int most_recent = 0;
1342 int chosen_drive;
1343 int change = 0;
1344 int inargv = 0;
1345 int start_partial_ok = (c->runstop >= 0) &&
1346 (c->force || devlist==NULL || auto_assem);
1347 int num_devs;
1348 struct mddev_dev *tmpdev;
1349 struct mdinfo info;
1350 struct mdinfo *content = NULL;
1351 struct mdinfo *pre_exist = NULL;
1352 char *avail;
1353 char *name = NULL;
1354 char chosen_name[1024];
1355 struct map_ent *map = NULL;
1356 struct map_ent *mp;
1357
1358 /*
1359 * If any subdevs are listed, then any that don't
1360 * match ident are discarded. Remainder must all match and
1361 * become the array.
1362 * If no subdevs, then we scan all devices in the config file, but
1363 * there must be something in the identity
1364 */
1365
1366 if (!devlist &&
1367 ident->uuid_set == 0 &&
1368 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1369 ident->name[0] == 0 &&
1370 (ident->container == NULL || ident->member == NULL) &&
1371 ident->devices == NULL) {
1372 pr_err("No identity information available for %s - cannot assemble.\n",
1373 mddev ? mddev : "further assembly");
1374 return 1;
1375 }
1376
1377 if (devlist == NULL)
1378 devlist = conf_get_devs();
1379 else if (mddev)
1380 inargv = 1;
1381
1382 try_again:
1383 /* We come back here when doing auto-assembly and attempting some
1384 * set of devices failed. Those are now marked as ->used==2 and
1385 * we ignore them and try again
1386 */
1387 if (!st && ident->st)
1388 st = ident->st;
1389 if (c->verbose>0)
1390 pr_err("looking for devices for %s\n",
1391 mddev ? mddev : "further assembly");
1392
1393 content = &info;
1394 if (st && c->force)
1395 st->ignore_hw_compat = 1;
1396 num_devs = select_devices(devlist, ident, &st, &content, c,
1397 inargv, auto_assem);
1398 if (num_devs < 0)
1399 return 1;
1400
1401 if (!st || !st->sb || !content)
1402 return 2;
1403
1404 /* We have a full set of devices - we now need to find the
1405 * array device.
1406 * However there is a risk that we are racing with "mdadm -I"
1407 * and the array is already partially assembled - we will have
1408 * rejected any devices already in this address.
1409 * So we take a lock on the map file - to prevent further races -
1410 * and look for the uuid in there. If found and the array is
1411 * active, we abort. If found and the array is not active
1412 * we commit to that md device and add all the contained devices
1413 * to our list. We flag them so that we don't try to re-add,
1414 * but can remove if they turn out to not be wanted.
1415 */
1416 if (map_lock(&map))
1417 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1418 if (c->update && strcmp(c->update,"uuid") == 0)
1419 mp = NULL;
1420 else
1421 mp = map_by_uuid(&map, content->uuid);
1422 if (mp) {
1423 struct mdinfo *dv;
1424 /* array already exists. */
1425 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1426 if (pre_exist->array.level != UnSet) {
1427 pr_err("Found some drive for an array that is already active: %s\n",
1428 mp->path);
1429 pr_err("giving up.\n");
1430 goto out;
1431 }
1432 for (dv = pre_exist->devs; dv; dv = dv->next) {
1433 /* We want to add this device to our list,
1434 * but it could already be there if "mdadm -I"
1435 * started *after* we checked for O_EXCL.
1436 * If we add it to the top of the list
1437 * it will be preferred over later copies.
1438 */
1439 struct mddev_dev *newdev;
1440 char *devname = map_dev(dv->disk.major,
1441 dv->disk.minor,
1442 0);
1443 if (!devname)
1444 continue;
1445 newdev = xmalloc(sizeof(*newdev));
1446 newdev->devname = devname;
1447 newdev->disposition = 'I';
1448 newdev->used = 1;
1449 newdev->next = devlist;
1450 devlist = newdev;
1451 num_devs++;
1452 }
1453 strcpy(chosen_name, mp->path);
1454 if (c->verbose > 0 || mddev == NULL ||
1455 strcmp(mddev, chosen_name) != 0)
1456 pr_err("Merging with already-assembled %s\n",
1457 chosen_name);
1458 mdfd = open_dev_excl(mp->devnm);
1459 } else {
1460 int trustworthy = FOREIGN;
1461 name = content->name;
1462 switch (st->ss->match_home(st, c->homehost)
1463 ?: st->ss->match_home(st, "any")) {
1464 case 1:
1465 trustworthy = LOCAL;
1466 name = strchr(content->name, ':');
1467 if (name)
1468 name++;
1469 else
1470 name = content->name;
1471 break;
1472 }
1473 if (!auto_assem)
1474 /* If the array is listed in mdadm.conf or on
1475 * command line, then we trust the name
1476 * even if the array doesn't look local
1477 */
1478 trustworthy = LOCAL;
1479
1480 if (name[0] == 0 &&
1481 content->array.level == LEVEL_CONTAINER) {
1482 name = content->text_version;
1483 trustworthy = METADATA;
1484 }
1485
1486 if (name[0] && trustworthy != LOCAL &&
1487 ! c->require_homehost &&
1488 conf_name_is_free(name))
1489 trustworthy = LOCAL;
1490
1491 if (trustworthy == LOCAL &&
1492 strchr(name, ':'))
1493 /* Ignore 'host:' prefix of name */
1494 name = strchr(name, ':')+1;
1495
1496 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1497 chosen_name, 0);
1498 }
1499 if (mdfd < 0) {
1500 st->ss->free_super(st);
1501 if (auto_assem)
1502 goto try_again;
1503 goto out;
1504 }
1505 mddev = chosen_name;
1506 if (pre_exist == NULL) {
1507 if (mddev_busy(fd2devnm(mdfd))) {
1508 pr_err("%s already active, cannot restart it!\n",
1509 mddev);
1510 for (tmpdev = devlist ;
1511 tmpdev && tmpdev->used != 1;
1512 tmpdev = tmpdev->next)
1513 ;
1514 if (tmpdev && auto_assem)
1515 pr_err("%s needed for %s...\n",
1516 mddev, tmpdev->devname);
1517 close(mdfd);
1518 mdfd = -3;
1519 st->ss->free_super(st);
1520 if (auto_assem)
1521 goto try_again;
1522 goto out;
1523 }
1524 /* just incase it was started but has no content */
1525 ioctl(mdfd, STOP_ARRAY, NULL);
1526 }
1527
1528 if (content != &info) {
1529 /* This is a member of a container. Try starting the array. */
1530 int err;
1531 err = assemble_container_content(st, mdfd, content, c,
1532 chosen_name, NULL);
1533 close(mdfd);
1534 return err;
1535 }
1536
1537 /* Ok, no bad inconsistancy, we can try updating etc */
1538 devices = xcalloc(num_devs, sizeof(*devices));
1539 devmap = xcalloc(num_devs, content->array.raid_disks);
1540 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1541 c, content, mdfd, mddev,
1542 &most_recent, &bestcnt, &best, inargv);
1543 if (devcnt < 0) {
1544 mdfd = -3;
1545 /*
1546 * devices is already freed in load_devices, so set devices
1547 * to NULL to avoid double free devices.
1548 */
1549 devices = NULL;
1550 goto out;
1551 }
1552
1553 if (devcnt == 0) {
1554 pr_err("no devices found for %s\n",
1555 mddev);
1556 if (st)
1557 st->ss->free_super(st);
1558 free(devmap);
1559 goto out;
1560 }
1561
1562 if (c->update && strcmp(c->update, "byteorder")==0)
1563 st->minor_version = 90;
1564
1565 st->ss->getinfo_super(st, content, NULL);
1566 clean = content->array.state & 1;
1567
1568 /* now we have some devices that might be suitable.
1569 * I wonder how many
1570 */
1571 avail = xcalloc(content->array.raid_disks, 1);
1572 okcnt = 0;
1573 replcnt = 0;
1574 sparecnt=0;
1575 journalcnt=0;
1576 rebuilding_cnt=0;
1577 for (i=0; i< bestcnt; i++) {
1578 int j = best[i];
1579 int event_margin = 1; /* always allow a difference of '1'
1580 * like the kernel does
1581 */
1582 if (j < 0) continue;
1583 /* note: we ignore error flags in multipath arrays
1584 * as they don't make sense
1585 */
1586 if (content->array.level != LEVEL_MULTIPATH) {
1587 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1588 if (content->journal_device_required)
1589 journalcnt++;
1590 else /* unexpected journal, mark as faulty */
1591 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1592 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1593 if (!(devices[j].i.disk.state
1594 & (1<<MD_DISK_FAULTY))) {
1595 devices[j].uptodate = 1;
1596 sparecnt++;
1597 }
1598 continue;
1599 }
1600 }
1601 /* If this device thinks that 'most_recent' has failed, then
1602 * we must reject this device.
1603 */
1604 if (j != most_recent && !c->force &&
1605 content->array.raid_disks > 0 &&
1606 devices[most_recent].i.disk.raid_disk >= 0 &&
1607 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1608 if (c->verbose > -1)
1609 pr_err("ignoring %s as it reports %s as failed\n",
1610 devices[j].devname, devices[most_recent].devname);
1611 best[i] = -1;
1612 continue;
1613 }
1614 /* Require event counter to be same as, or just less than,
1615 * most recent. If it is bigger, it must be a stray spare and
1616 * should be ignored.
1617 */
1618 if (devices[j].i.events+event_margin >=
1619 devices[most_recent].i.events &&
1620 devices[j].i.events <=
1621 devices[most_recent].i.events
1622 ) {
1623 devices[j].uptodate = 1;
1624 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1625 journal_clean = 1;
1626 if (i < content->array.raid_disks * 2) {
1627 if (devices[j].i.recovery_start == MaxSector ||
1628 (content->reshape_active &&
1629 i >= content->array.raid_disks - content->delta_disks)) {
1630 if (!avail[i/2]) {
1631 okcnt++;
1632 avail[i/2]=1;
1633 } else
1634 replcnt++;
1635 } else
1636 rebuilding_cnt++;
1637 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1638 sparecnt++;
1639 }
1640 }
1641 free(devmap);
1642 if (c->force) {
1643 int force_ok = force_array(content, devices, best, bestcnt,
1644 avail, most_recent, st, c);
1645 okcnt += force_ok;
1646 if (force_ok)
1647 was_forced = 1;
1648 }
1649 /* Now we want to look at the superblock which the kernel will base things on
1650 * and compare the devices that we think are working with the devices that the
1651 * superblock thinks are working.
1652 * If there are differences and --force is given, then update this chosen
1653 * superblock.
1654 */
1655 chosen_drive = -1;
1656 st->ss->free_super(st);
1657 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1658 int j = best[i];
1659 int fd;
1660
1661 if (j<0)
1662 continue;
1663 if (!devices[j].uptodate)
1664 continue;
1665 if (devices[j].i.events < devices[most_recent].i.events)
1666 continue;
1667 chosen_drive = j;
1668 if ((fd=dev_open(devices[j].devname,
1669 devices[j].included ? O_RDONLY
1670 : (O_RDONLY|O_EXCL)))< 0) {
1671 pr_err("Cannot open %s: %s\n",
1672 devices[j].devname, strerror(errno));
1673 goto out;
1674 }
1675 if (st->ss->load_super(st,fd, NULL)) {
1676 close(fd);
1677 pr_err("RAID superblock has disappeared from %s\n",
1678 devices[j].devname);
1679 goto out;
1680 }
1681 close(fd);
1682 }
1683 if (st->sb == NULL) {
1684 pr_err("No suitable drives found for %s\n", mddev);
1685 goto out;
1686 }
1687 st->ss->getinfo_super(st, content, NULL);
1688 if (sysfs_init(content, mdfd, NULL)) {
1689 pr_err("Unable to initialize sysfs\n");
1690 goto out;
1691 }
1692
1693 /* after reload context, store journal_clean in context */
1694 content->journal_clean = journal_clean;
1695 for (i=0; i<bestcnt; i++) {
1696 int j = best[i];
1697 unsigned int desired_state;
1698
1699 if (j < 0)
1700 continue;
1701 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1702 desired_state = (1<<MD_DISK_JOURNAL);
1703 else if (i >= content->array.raid_disks * 2)
1704 desired_state = 0;
1705 else if (i & 1)
1706 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1707 else
1708 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1709
1710 desired_state |= devices[j].i.disk.state & ((1<<MD_DISK_FAILFAST) |
1711 (1<<MD_DISK_WRITEMOSTLY));
1712
1713 if (!devices[j].uptodate)
1714 continue;
1715
1716 devices[j].i.disk.state = desired_state;
1717 if (!(devices[j].i.array.state & 1))
1718 clean = 0;
1719
1720 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1721 c->verbose, 0, NULL)) {
1722 if (c->force) {
1723 if (c->verbose >= 0)
1724 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1725 j, mddev, devices[j].devname);
1726 change = 1;
1727 } else {
1728 if (c->verbose >= -1)
1729 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1730 i, mddev, devices[j].devname);
1731 }
1732 }
1733 #if 0
1734 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1735 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1736 i, mddev);
1737 }
1738 #endif
1739 }
1740 if (c->force && !clean &&
1741 !enough(content->array.level, content->array.raid_disks,
1742 content->array.layout, clean,
1743 avail)) {
1744 change += st->ss->update_super(st, content, "force-array",
1745 devices[chosen_drive].devname, c->verbose,
1746 0, NULL);
1747 was_forced = 1;
1748 clean = 1;
1749 }
1750
1751 if (change) {
1752 int fd;
1753 fd = dev_open(devices[chosen_drive].devname,
1754 devices[chosen_drive].included ?
1755 O_RDWR : (O_RDWR|O_EXCL));
1756 if (fd < 0) {
1757 pr_err("Could not open %s for write - cannot Assemble array.\n",
1758 devices[chosen_drive].devname);
1759 goto out;
1760 }
1761 if (st->ss->store_super(st, fd)) {
1762 close(fd);
1763 pr_err("Could not re-write superblock on %s\n",
1764 devices[chosen_drive].devname);
1765 goto out;
1766 }
1767 if (c->verbose >= 0)
1768 pr_err("Marking array %s as 'clean'\n",
1769 mddev);
1770 close(fd);
1771 }
1772
1773 /* If we are in the middle of a reshape we may need to restore saved data
1774 * that was moved aside due to the reshape overwriting live data
1775 * The code of doing this lives in Grow.c
1776 */
1777 if (content->reshape_active &&
1778 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1779 int err = 0;
1780 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1781 if (c->verbose > 0)
1782 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1783 chosen_name);
1784 if (!c->backup_file)
1785 c->backup_file = locate_backup(content->sys_name);
1786 enable_fds(bestcnt/2);
1787 for (i = 0; i < bestcnt/2; i++) {
1788 int j = best[i*2];
1789 if (j >= 0) {
1790 fdlist[i] = dev_open(devices[j].devname,
1791 devices[j].included
1792 ? O_RDWR : (O_RDWR|O_EXCL));
1793 if (fdlist[i] < 0) {
1794 pr_err("Could not open %s for write - cannot Assemble array.\n",
1795 devices[j].devname);
1796 err = 1;
1797 break;
1798 }
1799 } else
1800 fdlist[i] = -1;
1801 }
1802 if (!err) {
1803 if (st->ss->external && st->ss->recover_backup)
1804 err = st->ss->recover_backup(st, content);
1805 else
1806 err = Grow_restart(st, content, fdlist, bestcnt/2,
1807 c->backup_file, c->verbose > 0);
1808 if (err && c->invalid_backup) {
1809 if (c->verbose > 0)
1810 pr_err("continuing without restoring backup\n");
1811 err = 0;
1812 }
1813 }
1814 while (i>0) {
1815 i--;
1816 if (fdlist[i]>=0) close(fdlist[i]);
1817 }
1818 free(fdlist);
1819 if (err) {
1820 pr_err("Failed to restore critical section for reshape, sorry.\n");
1821 if (c->backup_file == NULL)
1822 cont_err("Possibly you needed to specify the --backup-file\n");
1823 goto out;
1824 }
1825 }
1826
1827 /* Almost ready to actually *do* something */
1828 /* First, fill in the map, so that udev can find our name
1829 * as soon as we become active.
1830 */
1831 if (c->update && strcmp(c->update, "metadata")==0) {
1832 content->array.major_version = 1;
1833 content->array.minor_version = 0;
1834 strcpy(content->text_version, "1.0");
1835 }
1836
1837 map_update(&map, fd2devnm(mdfd), content->text_version,
1838 content->uuid, chosen_name);
1839
1840 rv = start_array(mdfd, mddev, content,
1841 st, ident, best, bestcnt,
1842 chosen_drive, devices, okcnt, sparecnt,
1843 rebuilding_cnt, journalcnt,
1844 c,
1845 clean, avail, start_partial_ok,
1846 pre_exist != NULL,
1847 was_forced);
1848 if (rv == 1 && !pre_exist)
1849 ioctl(mdfd, STOP_ARRAY, NULL);
1850 free(devices);
1851 out:
1852 map_unlock(&map);
1853 if (rv == 0) {
1854 wait_for(chosen_name, mdfd);
1855 close(mdfd);
1856 if (auto_assem) {
1857 int usecs = 1;
1858 /* There is a nasty race with 'mdadm --monitor'.
1859 * If it opens this device before we close it,
1860 * it gets an incomplete open on which IO
1861 * doesn't work and the capacity is
1862 * wrong.
1863 * If we reopen (to check for layered devices)
1864 * before --monitor closes, we loose.
1865 *
1866 * So: wait upto 1 second for there to be
1867 * a non-zero capacity.
1868 */
1869 while (usecs < 1000) {
1870 mdfd = open(mddev, O_RDONLY);
1871 if (mdfd >= 0) {
1872 unsigned long long size;
1873 if (get_dev_size(mdfd, NULL, &size) &&
1874 size > 0)
1875 break;
1876 close(mdfd);
1877 }
1878 usleep(usecs);
1879 usecs <<= 1;
1880 }
1881 }
1882 } else if (mdfd >= 0)
1883 close(mdfd);
1884
1885 /* '2' means 'OK, but not started yet' */
1886 if (rv == -1) {
1887 free(devices);
1888 return 1;
1889 }
1890 return rv == 2 ? 0 : rv;
1891 }
1892
1893 int assemble_container_content(struct supertype *st, int mdfd,
1894 struct mdinfo *content, struct context *c,
1895 char *chosen_name, int *result)
1896 {
1897 struct mdinfo *dev, *sra, *dev2;
1898 int working = 0, preexist = 0;
1899 int expansion = 0;
1900 int old_raid_disks;
1901 int start_reshape;
1902 char *avail;
1903 int err;
1904
1905 if (sysfs_init(content, mdfd, NULL)) {
1906 pr_err("Unable to initialize sysfs\n");
1907 return 1;
1908 }
1909
1910 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1911 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1912 if (content->array.major_version == -1 &&
1913 content->array.minor_version == -2 &&
1914 c->readonly &&
1915 content->text_version[0] == '/')
1916 content->text_version[0] = '-';
1917 if (sysfs_set_array(content, 9003) != 0) {
1918 sysfs_free(sra);
1919 return 1;
1920 }
1921 }
1922
1923 /* There are two types of reshape: container wide or sub-array specific
1924 * Check if metadata requests blocking container wide reshapes
1925 */
1926 start_reshape = (content->reshape_active &&
1927 !((content->reshape_active == CONTAINER_RESHAPE) &&
1928 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1929
1930 /* Block subarray here if it is under reshape now
1931 * Do not allow for any changes in this array
1932 */
1933 if (st->ss->external && content->recovery_blocked && start_reshape)
1934 block_subarray(content);
1935
1936 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1937 for (dev = content->devs; dev; dev = dev->next)
1938 if (dev2->disk.major == dev->disk.major &&
1939 dev2->disk.minor == dev->disk.minor)
1940 break;
1941 if (dev)
1942 continue;
1943 /* Don't want this one any more */
1944 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1945 errno == EBUSY) {
1946 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1947 sysfs_free(sra);
1948 return 1;
1949 }
1950 sysfs_set_str(sra, dev2, "state", "remove");
1951 }
1952 old_raid_disks = content->array.raid_disks - content->delta_disks;
1953 avail = xcalloc(content->array.raid_disks, 1);
1954 for (dev = content->devs; dev; dev = dev->next) {
1955 if (dev->disk.raid_disk >= 0)
1956 avail[dev->disk.raid_disk] = 1;
1957 if (sysfs_add_disk(content, dev, 1) == 0) {
1958 if (dev->disk.raid_disk >= old_raid_disks &&
1959 content->reshape_active)
1960 expansion++;
1961 else
1962 working++;
1963 } else if (errno == EEXIST)
1964 preexist++;
1965 }
1966 sysfs_free(sra);
1967 if (working + expansion == 0 && c->runstop <= 0) {
1968 free(avail);
1969 return 1;/* Nothing new, don't try to start */
1970 }
1971 map_update(NULL, fd2devnm(mdfd), content->text_version,
1972 content->uuid, chosen_name);
1973
1974 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1975 st->ss->validate_ppl) {
1976 content->array.state |= 1;
1977 err = 0;
1978
1979 for (dev = content->devs; dev; dev = dev->next) {
1980 int dfd;
1981 char *devpath;
1982 int ret;
1983
1984 ret = st->ss->validate_ppl(st, content, dev);
1985 if (ret == 0)
1986 continue;
1987
1988 if (ret < 0) {
1989 err = 1;
1990 break;
1991 }
1992
1993 if (!c->force) {
1994 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1995 chosen_name);
1996 content->array.state &= ~1;
1997 avail[dev->disk.raid_disk] = 0;
1998 break;
1999 }
2000
2001 /* have --force - overwrite the invalid ppl */
2002 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
2003 dfd = dev_open(devpath, O_RDWR);
2004 if (dfd < 0) {
2005 pr_err("Failed to open %s\n", devpath);
2006 err = 1;
2007 break;
2008 }
2009
2010 err = st->ss->write_init_ppl(st, content, dfd);
2011 close(dfd);
2012
2013 if (err)
2014 break;
2015 }
2016
2017 if (err) {
2018 free(avail);
2019 return err;
2020 }
2021 }
2022
2023 if (enough(content->array.level, content->array.raid_disks,
2024 content->array.layout, content->array.state & 1, avail) == 0) {
2025 if (c->export && result)
2026 *result |= INCR_NO;
2027 else if (c->verbose >= 0) {
2028 pr_err("%s assembled with %d device%s",
2029 chosen_name, preexist + working,
2030 preexist + working == 1 ? "":"s");
2031 if (preexist)
2032 fprintf(stderr, " (%d new)", working);
2033 fprintf(stderr, " but not started\n");
2034 }
2035 free(avail);
2036 return 1;
2037 }
2038 free(avail);
2039
2040 if (c->runstop <= 0 &&
2041 (working + preexist + expansion) <
2042 content->array.working_disks) {
2043 if (c->export && result)
2044 *result |= INCR_UNSAFE;
2045 else if (c->verbose >= 0) {
2046 pr_err("%s assembled with %d device%s",
2047 chosen_name, preexist + working,
2048 preexist + working == 1 ? "":"s");
2049 if (preexist)
2050 fprintf(stderr, " (%d new)", working);
2051 fprintf(stderr, " but not safe to start\n");
2052 }
2053 return 1;
2054 }
2055
2056
2057 if (start_reshape) {
2058 int spare = content->array.raid_disks + expansion;
2059 if (restore_backup(st, content,
2060 working,
2061 spare, &c->backup_file, c->verbose) == 1)
2062 return 1;
2063
2064 if (content->reshape_progress == 0) {
2065 /* If reshape progress is 0 - we are assembling the
2066 * array that was stopped, before reshape has started.
2067 * Array needs to be started as active, Grow_continue()
2068 * will start the reshape.
2069 */
2070 sysfs_set_num(content, NULL, "reshape_position",
2071 MaxSector);
2072 err = sysfs_set_str(content, NULL,
2073 "array_state", "active");
2074 sysfs_set_num(content, NULL, "reshape_position", 0);
2075 } else {
2076 err = sysfs_set_str(content, NULL,
2077 "array_state", "readonly");
2078 }
2079
2080 if (err)
2081 return 1;
2082
2083 if (st->ss->external) {
2084 if (!mdmon_running(st->container_devnm))
2085 start_mdmon(st->container_devnm);
2086 ping_monitor(st->container_devnm);
2087 if (mdmon_running(st->container_devnm) &&
2088 st->update_tail == NULL)
2089 st->update_tail = &st->updates;
2090 }
2091
2092 err = Grow_continue(mdfd, st, content, c->backup_file,
2093 0, c->freeze_reshape);
2094 } else switch(content->array.level) {
2095 case LEVEL_LINEAR:
2096 case LEVEL_MULTIPATH:
2097 case 0:
2098 err = sysfs_set_str(content, NULL, "array_state",
2099 c->readonly ? "readonly" : "active");
2100 break;
2101 default:
2102 err = sysfs_set_str(content, NULL, "array_state",
2103 "readonly");
2104 /* start mdmon if needed. */
2105 if (!err) {
2106 if (!mdmon_running(st->container_devnm))
2107 start_mdmon(st->container_devnm);
2108 ping_monitor(st->container_devnm);
2109 }
2110 break;
2111 }
2112 if (!err)
2113 sysfs_set_safemode(content, content->safe_mode_delay);
2114
2115 /* Block subarray here if it is not reshaped now
2116 * It has be blocked a little later to allow mdmon to switch in
2117 * in to R/W state
2118 */
2119 if (st->ss->external && content->recovery_blocked &&
2120 !start_reshape)
2121 block_subarray(content);
2122
2123 if (c->export && result) {
2124 if (err)
2125 *result |= INCR_NO;
2126 else
2127 *result |= INCR_YES;
2128 } else if (c->verbose >= 0) {
2129 if (err)
2130 pr_err("array %s now has %d device%s",
2131 chosen_name, working + preexist,
2132 working + preexist == 1 ? "":"s");
2133 else
2134 pr_err("Started %s with %d device%s",
2135 chosen_name, working + preexist,
2136 working + preexist == 1 ? "":"s");
2137 if (preexist)
2138 fprintf(stderr, " (%d new)", working);
2139 if (expansion)
2140 fprintf(stderr, " ( + %d for expansion)",
2141 expansion);
2142 fprintf(stderr, "\n");
2143 }
2144 if (!err)
2145 wait_for(chosen_name, mdfd);
2146 return err;
2147 /* FIXME should have an O_EXCL and wait for read-auto */
2148 }
Unnamed repository; edit this file 'description' to name the repository.