]> 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
Free map to avoid resource leak issues
[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 return -1;
285 }
286
287 if (found_container) {
288 /* tmpdev is a container. We need to be either
289 * looking for a member, or auto-assembling
290 */
291 /* should be safe to try an exclusive open now, we
292 * have rejected anything that some other mdadm might
293 * be looking at
294 */
295 dfd = dev_open(devname, O_RDONLY | O_EXCL);
296 if (dfd < 0) {
297 if (report_mismatch)
298 pr_err("%s is busy - skipping\n", devname);
299 goto loop;
300 }
301 close(dfd);
302
303 if (ident->container && ident->container[0] != '/') {
304 /* we have a uuid */
305 int uuid[4];
306
307 content = *contentp;
308 tst->ss->getinfo_super(tst, content, NULL);
309
310 if (!parse_uuid(ident->container, uuid) ||
311 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
312 if (report_mismatch)
313 pr_err("%s has wrong UUID to be required container\n",
314 devname);
315 goto loop;
316 }
317 }
318 /* It is worth looking inside this container.
319 */
320 if (c->verbose > 0)
321 pr_err("looking in container %s\n",
322 devname);
323
324 for (content = tst->ss->container_content(tst, NULL);
325 content;
326 content = content->next) {
327
328 if (!ident_matches(ident, content, tst,
329 c->homehost, c->require_homehost,
330 c->update,
331 report_mismatch ? devname : NULL))
332 /* message already printed */;
333 else if (is_member_busy(content->text_version)) {
334 if (report_mismatch)
335 pr_err("member %s in %s is already assembled\n",
336 content->text_version,
337 devname);
338 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
339 /* do not assemble arrays with unsupported configurations */
340 pr_err("Cannot activate member %s in %s.\n",
341 content->text_version,
342 devname);
343 } else
344 break;
345 }
346 if (!content) {
347 tmpdev->used = 2;
348 goto loop; /* empty container */
349 }
350
351 st = tst; tst = NULL;
352 if (!auto_assem && inargv && tmpdev->next != NULL) {
353 pr_err("%s is a container, but is not only device given: confused and aborting\n",
354 devname);
355 st->ss->free_super(st);
356 dev_policy_free(pol);
357 domain_free(domains);
358 return -1;
359 }
360 if (c->verbose > 0)
361 pr_err("found match on member %s in %s\n",
362 content->text_version, devname);
363
364 /* make sure we finished the loop */
365 tmpdev = NULL;
366 goto loop;
367 } else {
368 content = *contentp;
369 tst->ss->getinfo_super(tst, content, NULL);
370
371 if (!ident_matches(ident, content, tst,
372 c->homehost, c->require_homehost,
373 c->update,
374 report_mismatch ? devname : NULL))
375 goto loop;
376
377 if (auto_assem) {
378 /* Never auto-assemble things that conflict
379 * with mdadm.conf in some way
380 */
381 struct mddev_ident *match;
382 int rv = 0;
383
384 match = conf_match(tst, content, devname,
385 report_mismatch ? c->verbose : -1,
386 &rv);
387 if (!match && rv == 2)
388 goto loop;
389 if (match && match->devname &&
390 strcasecmp(match->devname, "<ignore>") == 0) {
391 if (report_mismatch)
392 pr_err("%s is a member of an explicitly ignored array\n",
393 devname);
394 goto loop;
395 }
396 if (match && !ident_matches(match, content, tst,
397 c->homehost, c->require_homehost,
398 c->update,
399 report_mismatch ? devname : NULL))
400 /* Array exists in mdadm.conf but some
401 * details don't match, so reject it
402 */
403 goto loop;
404 }
405
406 /* should be safe to try an exclusive open now, we
407 * have rejected anything that some other mdadm might
408 * be looking at
409 */
410 dfd = dev_open(devname, O_RDONLY | O_EXCL);
411 if (dfd < 0) {
412 if (report_mismatch)
413 pr_err("%s is busy - skipping\n", devname);
414 goto loop;
415 }
416 close(dfd);
417
418 if (st == NULL)
419 st = dup_super(tst);
420 if (st->minor_version == -1)
421 st->minor_version = tst->minor_version;
422
423 if (memcmp(content->uuid, uuid_zero,
424 sizeof(int[4])) == 0) {
425 /* this is a floating spare. It cannot define
426 * an array unless there are no more arrays of
427 * this type to be found. It can be included
428 * in an array of this type though.
429 */
430 tmpdev->used = 3;
431 goto loop;
432 }
433
434 if (st->ss != tst->ss ||
435 st->minor_version != tst->minor_version ||
436 st->ss->compare_super(st, tst) != 0) {
437 /* Some mismatch. If exactly one array matches this host,
438 * we can resolve on that one.
439 * Or, if we are auto assembling, we just ignore the second
440 * for now.
441 */
442 if (auto_assem)
443 goto loop;
444 if (c->homehost) {
445 int first = st->ss->match_home(st, c->homehost);
446 int last = tst->ss->match_home(tst, c->homehost);
447 if (first != last &&
448 (first == 1 || last == 1)) {
449 /* We can do something */
450 if (first) {/* just ignore this one */
451 if (report_mismatch)
452 pr_err("%s misses out due to wrong homehost\n",
453 devname);
454 goto loop;
455 } else { /* reject all those sofar */
456 struct mddev_dev *td;
457 if (report_mismatch)
458 pr_err("%s overrides previous devices due to good homehost\n",
459 devname);
460 for (td=devlist; td != tmpdev; td=td->next)
461 if (td->used == 1)
462 td->used = 0;
463 tmpdev->used = 1;
464 goto loop;
465 }
466 }
467 }
468 pr_err("superblock on %s doesn't match others - assembly aborted\n",
469 devname);
470 tst->ss->free_super(tst);
471 st->ss->free_super(st);
472 dev_policy_free(pol);
473 domain_free(domains);
474 return -1;
475 }
476 tmpdev->used = 1;
477 }
478 loop:
479 /* Collect domain information from members only */
480 if (tmpdev && tmpdev->used == 1) {
481 if (!pol)
482 pol = devid_policy(rdev);
483 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
484 }
485 dev_policy_free(pol);
486 pol = NULL;
487 if (tst)
488 tst->ss->free_super(tst);
489 }
490
491 /* Check if we found some imsm spares but no members */
492 if ((auto_assem ||
493 (ident->uuid_set &&
494 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
495 (!st || !st->sb))
496 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
497 if (tmpdev->used != 3)
498 continue;
499 tmpdev->used = 1;
500 content = *contentp;
501
502 if (!st->sb) {
503 /* we need sb from one of the spares */
504 int dfd = dev_open(tmpdev->devname, O_RDONLY);
505 if (dfd < 0 ||
506 st->ss->load_super(st, dfd, NULL))
507 tmpdev->used = 2;
508 if (dfd > 0)
509 close(dfd);
510 }
511 }
512
513 /* Now reject spares that don't match domains of identified members */
514 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
515 if (tmpdev->used != 3)
516 continue;
517 if (!stat_is_blkdev(tmpdev->devname, &rdev)) {
518 tmpdev->used = 2;
519 } else {
520 struct dev_policy *pol = devid_policy(rdev);
521 int dt = domain_test(domains, pol, NULL);
522 if (inargv && dt != 0)
523 /* take this spare as domains match
524 * if there are any */
525 tmpdev->used = 1;
526 else if (!inargv && dt == 1)
527 /* device wasn't explicitly listed, so need
528 * explicit domain match - which we have */
529 tmpdev->used = 1;
530 else
531 /* if domains don't match mark as unused */
532 tmpdev->used = 0;
533 dev_policy_free(pol);
534 }
535 }
536 domain_free(domains);
537 *stp = st;
538 if (st && st->sb && content == *contentp)
539 st->ss->getinfo_super(st, content, NULL);
540 *contentp = content;
541
542 return num_devs;
543 }
544
545 struct devs {
546 char *devname;
547 int uptodate; /* set once we decide that this device is as
548 * recent as everything else in the array.
549 */
550 int included; /* set if the device is already in the array
551 * due to a previous '-I'
552 */
553 struct mdinfo i;
554 };
555
556 static int load_devices(struct devs *devices, char *devmap,
557 struct mddev_ident *ident, struct supertype **stp,
558 struct mddev_dev *devlist, struct context *c,
559 struct mdinfo *content,
560 int mdfd, char *mddev,
561 int *most_recentp, int *bestcntp, int **bestp,
562 int inargv)
563 {
564 struct mddev_dev *tmpdev;
565 int devcnt = 0;
566 int nextspare = 0;
567 int bitmap_done = 0;
568 int most_recent = -1;
569 int bestcnt = 0;
570 int *best = *bestp;
571 struct supertype *st = *stp;
572
573 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
574 char *devname = tmpdev->devname;
575 struct stat stb;
576 struct supertype *tst;
577 int i;
578 int dfd;
579
580 if (tmpdev->used != 1)
581 continue;
582 /* looks like a good enough match to update the super block if needed */
583 if (c->update) {
584 /* prepare useful information in info structures */
585 struct stat stb2;
586 int err;
587 fstat(mdfd, &stb2);
588
589 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
590 random_uuid((__u8 *)ident->uuid);
591
592 if (strcmp(c->update, "ppl") == 0 &&
593 ident->bitmap_fd >= 0) {
594 pr_err("PPL is not compatible with bitmap\n");
595 return -1;
596 }
597
598 dfd = dev_open(devname,
599 tmpdev->disposition == 'I'
600 ? O_RDWR : (O_RDWR|O_EXCL));
601
602 tst = dup_super(st);
603 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
604 pr_err("cannot re-read metadata from %s - aborting\n",
605 devname);
606 if (dfd >= 0)
607 close(dfd);
608 close(mdfd);
609 free(devices);
610 free(devmap);
611 tst->ss->free_super(tst);
612 free(tst);
613 *stp = st;
614 return -1;
615 }
616 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
617
618 memcpy(content->uuid, ident->uuid, 16);
619 strcpy(content->name, ident->name);
620 content->array.md_minor = minor(stb2.st_rdev);
621
622 if (strcmp(c->update, "byteorder") == 0)
623 err = 0;
624 else if (strcmp(c->update, "home-cluster") == 0) {
625 tst->cluster_name = c->homecluster;
626 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
627 } else if (strcmp(c->update, "nodes") == 0) {
628 tst->nodes = c->nodes;
629 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
630 } else if (strcmp(c->update, "revert-reshape") == 0 &&
631 c->invalid_backup)
632 err = tst->ss->update_super(tst, content,
633 "revert-reshape-nobackup",
634 devname, c->verbose,
635 ident->uuid_set,
636 c->homehost);
637 else
638 err = tst->ss->update_super(tst, content, c->update,
639 devname, c->verbose,
640 ident->uuid_set,
641 c->homehost);
642 if (err < 0) {
643 if (err == -1)
644 pr_err("--update=%s not understood for %s metadata\n",
645 c->update, tst->ss->name);
646 tst->ss->free_super(tst);
647 free(tst);
648 close(mdfd);
649 close(dfd);
650 free(devices);
651 free(devmap);
652 *stp = st;
653 return -1;
654 }
655 if (strcmp(c->update, "uuid")==0 &&
656 !ident->uuid_set) {
657 ident->uuid_set = 1;
658 memcpy(ident->uuid, content->uuid, 16);
659 }
660 if (tst->ss->store_super(tst, dfd))
661 pr_err("Could not re-write superblock on %s.\n",
662 devname);
663
664 if (strcmp(c->update, "uuid")==0 &&
665 ident->bitmap_fd >= 0 && !bitmap_done) {
666 if (bitmap_update_uuid(ident->bitmap_fd,
667 content->uuid,
668 tst->ss->swapuuid) != 0)
669 pr_err("Could not update uuid on external bitmap.\n");
670 else
671 bitmap_done = 1;
672 }
673 } else {
674 dfd = dev_open(devname,
675 tmpdev->disposition == 'I'
676 ? O_RDWR : (O_RDWR|O_EXCL));
677 tst = dup_super(st);
678
679 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
680 pr_err("cannot re-read metadata from %s - aborting\n",
681 devname);
682 if (dfd >= 0)
683 close(dfd);
684 close(mdfd);
685 free(devices);
686 free(devmap);
687 tst->ss->free_super(tst);
688 free(tst);
689 *stp = st;
690 return -1;
691 }
692 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
693 }
694
695 fstat(dfd, &stb);
696 close(dfd);
697
698 if (c->verbose > 0)
699 pr_err("%s is identified as a member of %s, slot %d%s.\n",
700 devname, mddev, content->disk.raid_disk,
701 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
702 devices[devcnt].devname = devname;
703 devices[devcnt].uptodate = 0;
704 devices[devcnt].included = (tmpdev->disposition == 'I');
705 devices[devcnt].i = *content;
706 devices[devcnt].i.disk.major = major(stb.st_rdev);
707 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
708
709 if (devices[devcnt].i.disk.state == 6) {
710 if (most_recent < 0 ||
711 devices[devcnt].i.events
712 > devices[most_recent].i.events) {
713 struct supertype *tmp = tst;
714 tst = st;
715 st = tmp;
716 most_recent = devcnt;
717 }
718 }
719 tst->ss->free_super(tst);
720 free(tst);
721
722 if (content->array.level == LEVEL_MULTIPATH)
723 /* with multipath, the raid_disk from the superblock is meaningless */
724 i = devcnt;
725 else
726 i = devices[devcnt].i.disk.raid_disk;
727 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
728 if (nextspare < content->array.raid_disks*2)
729 nextspare = content->array.raid_disks*2;
730 i = nextspare++;
731 } else {
732 /* i is raid_disk - double it so there is room for
733 * replacements */
734 i *= 2;
735 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
736 i++;
737 if (i >= content->array.raid_disks*2 &&
738 i >= nextspare)
739 nextspare = i+1;
740 }
741 if (i < 10000) {
742 if (i >= bestcnt) {
743 int newbestcnt = i+10;
744 int *newbest = xmalloc(sizeof(int)*newbestcnt);
745 int c;
746 for (c=0; c < newbestcnt; c++)
747 if (c < bestcnt)
748 newbest[c] = best[c];
749 else
750 newbest[c] = -1;
751 if (best)free(best);
752 best = newbest;
753 bestcnt = newbestcnt;
754 }
755 if (best[i] >=0 &&
756 devices[best[i]].i.events ==
757 devices[devcnt].i.events &&
758 (devices[best[i]].i.disk.minor !=
759 devices[devcnt].i.disk.minor) &&
760 st->ss == &super0 &&
761 content->array.level != LEVEL_MULTIPATH) {
762 /* two different devices with identical superblock.
763 * Could be a mis-detection caused by overlapping
764 * partitions. fail-safe.
765 */
766 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
767 " If they are really different, please --zero the superblock on one\n"
768 " If they are the same or overlap, please remove one from %s.\n",
769 devices[best[i]].devname, devname,
770 inargv ? "the list" :
771 "the\n DEVICE list in mdadm.conf"
772 );
773 close(mdfd);
774 free(devices);
775 free(devmap);
776 *stp = st;
777 return -1;
778 }
779 if (best[i] == -1 || (devices[best[i]].i.events
780 < devices[devcnt].i.events))
781 best[i] = devcnt;
782 else if (st->ss == &super_imsm)
783 best[i+1] = devcnt;
784 }
785 devcnt++;
786 }
787 if (most_recent >= 0)
788 *most_recentp = most_recent;
789 *bestcntp = bestcnt;
790 *bestp = best;
791 *stp = st;
792 return devcnt;
793 }
794
795 static int force_array(struct mdinfo *content,
796 struct devs *devices,
797 int *best, int bestcnt, char *avail,
798 int most_recent,
799 struct supertype *st,
800 struct context *c)
801 {
802 int okcnt = 0;
803 while (!enough(content->array.level, content->array.raid_disks,
804 content->array.layout, 1,
805 avail) ||
806 (content->reshape_active && content->delta_disks > 0 &&
807 !enough(content->array.level, (content->array.raid_disks
808 - content->delta_disks),
809 content->new_layout, 1, avail))) {
810 /* Choose the newest best drive which is
811 * not up-to-date, update the superblock
812 * and add it.
813 */
814 int fd;
815 struct supertype *tst;
816 unsigned long long current_events;
817 int chosen_drive = -1;
818 int i;
819
820 for (i = 0;
821 i < content->array.raid_disks * 2 && i < bestcnt;
822 i += 2) {
823 int j = best[i];
824 if (j < 0)
825 continue;
826 if (devices[j].uptodate)
827 continue;
828 if (devices[j].i.recovery_start != MaxSector) {
829 int delta;
830 if (!devices[j].i.reshape_active ||
831 devices[j].i.delta_disks <= 0)
832 continue;
833 /* When increasing number of devices, an
834 * added device also appears to be
835 * recovering. It is safe to include it
836 * as long as it won't be a source of
837 * data.
838 * For now, just allow for last data
839 * devices in RAID4 or last devices in RAID4/5/6.
840 */
841 delta = devices[j].i.delta_disks;
842 if (devices[j].i.array.level >= 4 &&
843 devices[j].i.array.level <= 6 &&
844 i/2 >= content->array.raid_disks - delta)
845 /* OK */;
846 else if (devices[j].i.array.level == 4 &&
847 i/2 >= content->array.raid_disks - delta - 1)
848 /* OK */;
849 else
850 continue;
851 } else if (devices[j].i.reshape_active !=
852 content->reshape_active ||
853 (devices[j].i.reshape_active &&
854 devices[j].i.reshape_progress !=
855 content->reshape_progress))
856 /* Here, it may be a source of data. If two
857 * devices claim different progresses, it
858 * means that reshape boundaries differ for
859 * their own devices. Kernel will only treat
860 * the first one as reshape progress and
861 * go on. It may cause disaster, so avoid it.
862 */
863 continue;
864 if (chosen_drive < 0 ||
865 devices[j].i.events
866 > devices[chosen_drive].i.events)
867 chosen_drive = j;
868 }
869 if (chosen_drive < 0)
870 break;
871 current_events = devices[chosen_drive].i.events;
872 add_another:
873 if (c->verbose >= 0)
874 pr_err("forcing event count in %s(%d) from %d upto %d\n",
875 devices[chosen_drive].devname,
876 devices[chosen_drive].i.disk.raid_disk,
877 (int)(devices[chosen_drive].i.events),
878 (int)(devices[most_recent].i.events));
879 fd = dev_open(devices[chosen_drive].devname,
880 devices[chosen_drive].included ? O_RDWR
881 : (O_RDWR|O_EXCL));
882 if (fd < 0) {
883 pr_err("Couldn't open %s for write - not updating\n",
884 devices[chosen_drive].devname);
885 devices[chosen_drive].i.events = 0;
886 continue;
887 }
888 tst = dup_super(st);
889 if (tst->ss->load_super(tst,fd, NULL)) {
890 close(fd);
891 pr_err("RAID superblock disappeared from %s - not updating.\n",
892 devices[chosen_drive].devname);
893 devices[chosen_drive].i.events = 0;
894 continue;
895 }
896 content->events = devices[most_recent].i.events;
897 tst->ss->update_super(tst, content, "force-one",
898 devices[chosen_drive].devname, c->verbose,
899 0, NULL);
900
901 if (tst->ss->store_super(tst, fd)) {
902 close(fd);
903 pr_err("Could not re-write superblock on %s\n",
904 devices[chosen_drive].devname);
905 devices[chosen_drive].i.events = 0;
906 tst->ss->free_super(tst);
907 continue;
908 }
909 close(fd);
910 devices[chosen_drive].i.events = devices[most_recent].i.events;
911 devices[chosen_drive].uptodate = 1;
912 avail[chosen_drive] = 1;
913 okcnt++;
914 tst->ss->free_super(tst);
915 /* If there are any other drives of the same vintage,
916 * add them in as well. We can't lose and we might gain
917 */
918 for (i = 0;
919 i < content->array.raid_disks * 2 && i < bestcnt ;
920 i += 2) {
921 int j = best[i];
922 if (j >= 0 &&
923 !devices[j].uptodate &&
924 devices[j].i.recovery_start == MaxSector &&
925 devices[j].i.events == current_events &&
926 ((!devices[j].i.reshape_active &&
927 !content->reshape_active) ||
928 (devices[j].i.reshape_active ==
929 content->reshape_active &&
930 devices[j].i.reshape_progress ==
931 content->reshape_progress))) {
932 chosen_drive = j;
933 goto add_another;
934 }
935 }
936 }
937 return okcnt;
938 }
939
940 static int start_array(int mdfd,
941 char *mddev,
942 struct mdinfo *content,
943 struct supertype *st,
944 struct mddev_ident *ident,
945 int *best, int bestcnt,
946 int chosen_drive,
947 struct devs *devices,
948 unsigned int okcnt,
949 unsigned int sparecnt,
950 unsigned int rebuilding_cnt,
951 unsigned int journalcnt,
952 struct context *c,
953 int clean, char *avail,
954 int start_partial_ok,
955 int err_ok,
956 int was_forced
957 )
958 {
959 int rv;
960 int i;
961 unsigned int req_cnt;
962
963 if (content->journal_device_required && (content->journal_clean == 0)) {
964 if (!c->force) {
965 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
966 return 1;
967 }
968 pr_err("Journal is missing or stale, starting array read only.\n");
969 c->readonly = 1;
970 }
971
972 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
973 clean = 1;
974
975 rv = set_array_info(mdfd, st, content);
976 if (rv && !err_ok) {
977 pr_err("failed to set array info for %s: %s\n",
978 mddev, strerror(errno));
979 return 1;
980 }
981 if (ident->bitmap_fd >= 0) {
982 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
983 pr_err("SET_BITMAP_FILE failed.\n");
984 return 1;
985 }
986 } else if (ident->bitmap_file) {
987 /* From config file */
988 int bmfd = open(ident->bitmap_file, O_RDWR);
989 if (bmfd < 0) {
990 pr_err("Could not open bitmap file %s\n",
991 ident->bitmap_file);
992 return 1;
993 }
994 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
995 pr_err("Failed to set bitmapfile for %s\n", mddev);
996 close(bmfd);
997 return 1;
998 }
999 close(bmfd);
1000 }
1001
1002 /* First, add the raid disks, but add the chosen one last */
1003 for (i = 0; i <= bestcnt; i++) {
1004 int j;
1005 if (i < bestcnt) {
1006 j = best[i];
1007 if (j == chosen_drive)
1008 continue;
1009 } else
1010 j = chosen_drive;
1011
1012 if (j >= 0 && !devices[j].included) {
1013 int dfd;
1014
1015 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1016 if (dfd >= 0) {
1017 remove_partitions(dfd);
1018 close(dfd);
1019 }
1020 rv = add_disk(mdfd, st, content, &devices[j].i);
1021
1022 if (rv) {
1023 pr_err("failed to add %s to %s: %s\n",
1024 devices[j].devname, mddev,
1025 strerror(errno));
1026 if (i < content->array.raid_disks * 2 ||
1027 i == bestcnt)
1028 okcnt--;
1029 else
1030 sparecnt--;
1031 } else if (c->verbose > 0) {
1032 pr_err("added %s to %s as %d%s%s\n",
1033 devices[j].devname, mddev,
1034 devices[j].i.disk.raid_disk,
1035 devices[j].uptodate?"":
1036 " (possibly out of date)",
1037 (devices[j].i.disk.state &
1038 (1<<MD_DISK_REPLACEMENT)) ?
1039 " replacement":"");
1040 }
1041 } else if (j >= 0) {
1042 if (c->verbose > 0)
1043 pr_err("%s is already in %s as %d\n",
1044 devices[j].devname, mddev,
1045 devices[j].i.disk.raid_disk);
1046 } else if (c->verbose > 0 &&
1047 i < content->array.raid_disks * 2 && (i & 1) == 0)
1048 pr_err("no uptodate device for slot %d of %s\n",
1049 i/2, mddev);
1050 }
1051
1052 if (content->array.level == LEVEL_CONTAINER) {
1053 if (c->verbose >= 0) {
1054 pr_err("Container %s has been assembled with %d drive%s",
1055 mddev, okcnt + sparecnt + journalcnt,
1056 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1057 if (okcnt < (unsigned)content->array.raid_disks)
1058 fprintf(stderr, " (out of %d)",
1059 content->array.raid_disks);
1060 fprintf(stderr, "\n");
1061 }
1062
1063 if (st->ss->validate_container) {
1064 struct mdinfo *devices_list;
1065 struct mdinfo *info_devices;
1066 unsigned int count;
1067
1068 devices_list = NULL;
1069 info_devices = xmalloc(sizeof(struct mdinfo) *
1070 (okcnt + sparecnt));
1071 for (count = 0; count < okcnt + sparecnt; count++) {
1072 info_devices[count] = devices[count].i;
1073 info_devices[count].next = devices_list;
1074 devices_list = &info_devices[count];
1075 }
1076 if (st->ss->validate_container(devices_list))
1077 pr_err("Mismatch detected!\n");
1078 free(info_devices);
1079 }
1080
1081 st->ss->free_super(st);
1082 sysfs_uevent(content, "change");
1083 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1084 /* Was partial, is still partial, so signal an error
1085 * to ensure we don't retry */
1086 return 1;
1087 return 0;
1088 }
1089
1090 /* Get number of in-sync devices according to the superblock.
1091 * We must have this number to start the array without -s or -R
1092 */
1093 req_cnt = content->array.working_disks;
1094
1095 if (c->runstop == 1 ||
1096 (c->runstop <= 0 &&
1097 (enough(content->array.level, content->array.raid_disks,
1098 content->array.layout, clean, avail) &&
1099 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1100 /* This array is good-to-go.
1101 * If a reshape is in progress then we might need to
1102 * continue monitoring it. In that case we start
1103 * it read-only and let the grow code make it writable.
1104 */
1105 int rv;
1106
1107 if (content->reshape_active &&
1108 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1109 content->delta_disks <= 0) {
1110 if (!c->backup_file) {
1111 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1112 mddev);
1113 pr_err("Please provided one with \"--backup-file=...\"\n");
1114 if (c->update &&
1115 strcmp(c->update, "revert-reshape") == 0)
1116 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1117 return 1;
1118 }
1119 rv = sysfs_set_str(content, NULL,
1120 "array_state", "readonly");
1121 if (rv == 0)
1122 rv = Grow_continue(mdfd, st, content,
1123 c->backup_file, 0,
1124 c->freeze_reshape);
1125 } else if (c->readonly &&
1126 sysfs_attribute_available(content, NULL,
1127 "array_state")) {
1128 rv = sysfs_set_str(content, NULL,
1129 "array_state", "readonly");
1130 } else
1131 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1132 reopen_mddev(mdfd); /* drop O_EXCL */
1133 if (rv == 0) {
1134 if (c->verbose >= 0) {
1135 pr_err("%s has been started with %d drive%s",
1136 mddev, okcnt, okcnt==1?"":"s");
1137 if (okcnt < (unsigned)content->array.raid_disks)
1138 fprintf(stderr, " (out of %d)",
1139 content->array.raid_disks);
1140 if (rebuilding_cnt)
1141 fprintf(stderr, "%s %d rebuilding",
1142 sparecnt?",":" and",
1143 rebuilding_cnt);
1144 if (sparecnt)
1145 fprintf(stderr, " and %d spare%s",
1146 sparecnt,
1147 sparecnt == 1 ? "" : "s");
1148 if (content->journal_clean)
1149 fprintf(stderr, " and %d journal",
1150 journalcnt);
1151 fprintf(stderr, ".\n");
1152 }
1153 if (content->reshape_active &&
1154 content->array.level >= 4 &&
1155 content->array.level <= 6) {
1156 /* might need to increase the size
1157 * of the stripe cache - default is 256
1158 */
1159 int chunk_size = content->array.chunk_size;
1160
1161 if (content->reshape_active &&
1162 content->new_chunk > chunk_size)
1163 chunk_size = content->new_chunk;
1164 if (256 < 4 * ((chunk_size+4065)/4096)) {
1165 struct mdinfo *sra;
1166
1167 sra = sysfs_read(mdfd, NULL, 0);
1168 if (sra)
1169 sysfs_set_num(sra, NULL,
1170 "stripe_cache_size",
1171 (4 * chunk_size / 4096) + 1);
1172 sysfs_free(sra);
1173 }
1174 }
1175 if (okcnt < (unsigned)content->array.raid_disks) {
1176 /* If any devices did not get added
1177 * because the kernel rejected them based
1178 * on event count, try adding them
1179 * again providing the action policy is
1180 * 're-add' or greater. The bitmap
1181 * might allow them to be included, or
1182 * they will become spares.
1183 */
1184 for (i = 0; i < bestcnt; i++) {
1185 int j = best[i];
1186 if (j >= 0 && !devices[j].uptodate) {
1187 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1188 continue;
1189 rv = add_disk(mdfd, st, content,
1190 &devices[j].i);
1191 if (rv == 0 && c->verbose >= 0)
1192 pr_err("%s has been re-added.\n",
1193 devices[j].devname);
1194 }
1195 }
1196 }
1197 if (content->array.level == 6 &&
1198 okcnt + 1 == (unsigned)content->array.raid_disks &&
1199 was_forced) {
1200 struct mdinfo *sra;
1201
1202 sra = sysfs_read(mdfd, NULL, 0);
1203 if (sra)
1204 sysfs_set_str(sra, NULL,
1205 "sync_action", "repair");
1206 sysfs_free(sra);
1207 }
1208 return 0;
1209 }
1210 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1211
1212 if (!enough(content->array.level, content->array.raid_disks,
1213 content->array.layout, 1, avail))
1214 pr_err("Not enough devices to start the array.\n");
1215 else if (!enough(content->array.level,
1216 content->array.raid_disks,
1217 content->array.layout, clean, avail))
1218 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1219
1220 return 1;
1221 }
1222 if (c->runstop == -1) {
1223 pr_err("%s assembled from %d drive%s",
1224 mddev, okcnt, okcnt == 1 ? "" : "s");
1225 if (okcnt != (unsigned)content->array.raid_disks)
1226 fprintf(stderr, " (out of %d)",
1227 content->array.raid_disks);
1228 fprintf(stderr, ", but not started.\n");
1229 return 2;
1230 }
1231 if (c->verbose >= -1) {
1232 pr_err("%s assembled from %d drive%s",
1233 mddev, okcnt, okcnt == 1 ? "" : "s");
1234 if (rebuilding_cnt)
1235 fprintf(stderr, "%s %d rebuilding",
1236 sparecnt ? "," : " and", rebuilding_cnt);
1237 if (sparecnt)
1238 fprintf(stderr, " and %d spare%s", sparecnt,
1239 sparecnt == 1 ? "" : "s");
1240 if (!enough(content->array.level, content->array.raid_disks,
1241 content->array.layout, 1, avail))
1242 fprintf(stderr, " - not enough to start the array.\n");
1243 else if (!enough(content->array.level,
1244 content->array.raid_disks,
1245 content->array.layout, clean, avail))
1246 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1247 else {
1248 if (req_cnt == (unsigned)content->array.raid_disks)
1249 fprintf(stderr, " - need all %d to start it",
1250 req_cnt);
1251 else
1252 fprintf(stderr, " - need %d to start", req_cnt);
1253 fprintf(stderr, " (use --run to insist).\n");
1254 }
1255 }
1256 return 1;
1257 }
1258
1259 int Assemble(struct supertype *st, char *mddev,
1260 struct mddev_ident *ident,
1261 struct mddev_dev *devlist,
1262 struct context *c)
1263 {
1264 /*
1265 * The task of Assemble is to find a collection of
1266 * devices that should (according to their superblocks)
1267 * form an array, and to give this collection to the MD driver.
1268 * In Linux-2.4 and later, this involves submitting a
1269 * SET_ARRAY_INFO ioctl with no arg - to prepare
1270 * the array - and then submit a number of
1271 * ADD_NEW_DISK ioctls to add disks into
1272 * the array. Finally RUN_ARRAY might
1273 * be submitted to start the array.
1274 *
1275 * Much of the work of Assemble is in finding and/or
1276 * checking the disks to make sure they look right.
1277 *
1278 * If mddev is not set, then scan must be set and we
1279 * read through the config file for dev+uuid mapping
1280 * We recurse, setting mddev, for each device that
1281 * - isn't running
1282 * - has a valid uuid (or any uuid if !uuidset)
1283 *
1284 * If mddev is set, we try to determine state of md.
1285 * check version - must be at least 0.90.0
1286 * check kernel version. must be at least 2.4.
1287 * If not, we can possibly fall back on START_ARRAY
1288 * Try to GET_ARRAY_INFO.
1289 * If possible, give up
1290 * If not, try to STOP_ARRAY just to make sure
1291 *
1292 * If !uuidset and scan, look in conf-file for uuid
1293 * If not found, give up
1294 * If !devlist and scan and uuidset, get list of devs from conf-file
1295 *
1296 * For each device:
1297 * Check superblock - discard if bad
1298 * Check uuid (set if we don't have one) - discard if no match
1299 * Check superblock similarity if we have a superblock - discard if different
1300 * Record events, devicenum
1301 * This should give us a list of devices for the array
1302 * We should collect the most recent event number
1303 *
1304 * Count disks with recent enough event count
1305 * While force && !enough disks
1306 * Choose newest rejected disks, update event count
1307 * mark clean and rewrite superblock
1308 * If recent kernel:
1309 * SET_ARRAY_INFO
1310 * foreach device with recent events : ADD_NEW_DISK
1311 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1312 * If old kernel:
1313 * Check the device numbers in superblock are right
1314 * update superblock if any changes
1315 * START_ARRAY
1316 *
1317 */
1318 int rv = -1;
1319 int mdfd = -1;
1320 int clean;
1321 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1322 ident->super_minor == UnSet && ident->name[0] == 0 &&
1323 (ident->container == NULL || ident->member == NULL));
1324 struct devs *devices = NULL;
1325 char *devmap;
1326 int *best = NULL; /* indexed by raid_disk */
1327 int bestcnt = 0;
1328 int devcnt;
1329 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1330 int journal_clean = 0;
1331 int i;
1332 int was_forced = 0;
1333 int most_recent = 0;
1334 int chosen_drive;
1335 int change = 0;
1336 int inargv = 0;
1337 int start_partial_ok = (c->runstop >= 0) &&
1338 (c->force || devlist==NULL || auto_assem);
1339 int num_devs;
1340 struct mddev_dev *tmpdev;
1341 struct mdinfo info;
1342 struct mdinfo *content = NULL;
1343 struct mdinfo *pre_exist = NULL;
1344 char *avail;
1345 char *name = NULL;
1346 char chosen_name[1024];
1347 struct map_ent *map = NULL;
1348 struct map_ent *mp;
1349 int locked = 0;
1350 struct mdp_superblock_1 *sb;
1351 bitmap_super_t *bms;
1352
1353 /*
1354 * If any subdevs are listed, then any that don't
1355 * match ident are discarded. Remainder must all match and
1356 * become the array.
1357 * If no subdevs, then we scan all devices in the config file, but
1358 * there must be something in the identity
1359 */
1360
1361 if (!devlist &&
1362 ident->uuid_set == 0 &&
1363 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1364 ident->name[0] == 0 &&
1365 (ident->container == NULL || ident->member == NULL) &&
1366 ident->devices == NULL) {
1367 pr_err("No identity information available for %s - cannot assemble.\n",
1368 mddev ? mddev : "further assembly");
1369 return 1;
1370 }
1371
1372 if (devlist == NULL)
1373 devlist = conf_get_devs();
1374 else if (mddev)
1375 inargv = 1;
1376
1377 try_again:
1378 /* We come back here when doing auto-assembly and attempting some
1379 * set of devices failed. Those are now marked as ->used==2 and
1380 * we ignore them and try again
1381 */
1382 if (locked)
1383 /*
1384 * if come back try_again is called, then need to unlock first,
1385 * and lock again since the metadate is re-read.
1386 */
1387 cluster_release_dlmlock();
1388 if (!st && ident->st)
1389 st = ident->st;
1390 if (c->verbose>0)
1391 pr_err("looking for devices for %s\n",
1392 mddev ? mddev : "further assembly");
1393
1394 content = &info;
1395 if (st && c->force)
1396 st->ignore_hw_compat = 1;
1397 num_devs = select_devices(devlist, ident, &st, &content, c,
1398 inargv, auto_assem);
1399 if (num_devs < 0)
1400 return 1;
1401
1402 if (!st || !st->sb || !content)
1403 return 2;
1404
1405 sb = st->sb;
1406 bms = (bitmap_super_t*)(((char*)sb) + 4096);
1407 if (sb && bms->version == BITMAP_MAJOR_CLUSTERED) {
1408 locked = cluster_get_dlmlock();
1409 if (locked != 1)
1410 return 1;
1411 }
1412
1413 /* We have a full set of devices - we now need to find the
1414 * array device.
1415 * However there is a risk that we are racing with "mdadm -I"
1416 * and the array is already partially assembled - we will have
1417 * rejected any devices already in this address.
1418 * So we take a lock on the map file - to prevent further races -
1419 * and look for the uuid in there. If found and the array is
1420 * active, we abort. If found and the array is not active
1421 * we commit to that md device and add all the contained devices
1422 * to our list. We flag them so that we don't try to re-add,
1423 * but can remove if they turn out to not be wanted.
1424 */
1425 if (map_lock(&map))
1426 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1427 if (c->update && strcmp(c->update,"uuid") == 0)
1428 mp = NULL;
1429 else
1430 mp = map_by_uuid(&map, content->uuid);
1431 if (mp) {
1432 struct mdinfo *dv;
1433 /* array already exists. */
1434 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1435 if (pre_exist->array.level != UnSet) {
1436 pr_err("Found some drive for an array that is already active: %s\n",
1437 mp->path);
1438 pr_err("giving up.\n");
1439 goto out;
1440 }
1441 for (dv = pre_exist->devs; dv; dv = dv->next) {
1442 /* We want to add this device to our list,
1443 * but it could already be there if "mdadm -I"
1444 * started *after* we checked for O_EXCL.
1445 * If we add it to the top of the list
1446 * it will be preferred over later copies.
1447 */
1448 struct mddev_dev *newdev;
1449 char *devname = map_dev(dv->disk.major,
1450 dv->disk.minor,
1451 0);
1452 if (!devname)
1453 continue;
1454 newdev = xmalloc(sizeof(*newdev));
1455 newdev->devname = devname;
1456 newdev->disposition = 'I';
1457 newdev->used = 1;
1458 newdev->next = devlist;
1459 devlist = newdev;
1460 num_devs++;
1461 }
1462 strcpy(chosen_name, mp->path);
1463 if (c->verbose > 0 || mddev == NULL ||
1464 strcmp(mddev, chosen_name) != 0)
1465 pr_err("Merging with already-assembled %s\n",
1466 chosen_name);
1467 mdfd = open_dev_excl(mp->devnm);
1468 } else {
1469 int trustworthy = FOREIGN;
1470 name = content->name;
1471 switch (st->ss->match_home(st, c->homehost)
1472 ?: st->ss->match_home(st, "any")) {
1473 case 1:
1474 trustworthy = LOCAL;
1475 name = strchr(content->name, ':');
1476 if (name)
1477 name++;
1478 else
1479 name = content->name;
1480 break;
1481 }
1482 if (!auto_assem)
1483 /* If the array is listed in mdadm.conf or on
1484 * command line, then we trust the name
1485 * even if the array doesn't look local
1486 */
1487 trustworthy = LOCAL;
1488
1489 if (name[0] == 0 &&
1490 content->array.level == LEVEL_CONTAINER) {
1491 name = content->text_version;
1492 trustworthy = METADATA;
1493 }
1494
1495 if (name[0] && trustworthy != LOCAL &&
1496 ! c->require_homehost &&
1497 conf_name_is_free(name))
1498 trustworthy = LOCAL;
1499
1500 if (trustworthy == LOCAL &&
1501 strchr(name, ':'))
1502 /* Ignore 'host:' prefix of name */
1503 name = strchr(name, ':')+1;
1504
1505 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1506 chosen_name, 0);
1507 }
1508 if (mdfd < 0) {
1509 st->ss->free_super(st);
1510 if (auto_assem)
1511 goto try_again;
1512 goto out;
1513 }
1514 mddev = chosen_name;
1515 if (pre_exist == NULL) {
1516 if (mddev_busy(fd2devnm(mdfd))) {
1517 pr_err("%s already active, cannot restart it!\n",
1518 mddev);
1519 for (tmpdev = devlist ;
1520 tmpdev && tmpdev->used != 1;
1521 tmpdev = tmpdev->next)
1522 ;
1523 if (tmpdev && auto_assem)
1524 pr_err("%s needed for %s...\n",
1525 mddev, tmpdev->devname);
1526 close(mdfd);
1527 mdfd = -3;
1528 st->ss->free_super(st);
1529 if (auto_assem)
1530 goto try_again;
1531 goto out;
1532 }
1533 /* just incase it was started but has no content */
1534 ioctl(mdfd, STOP_ARRAY, NULL);
1535 }
1536
1537 if (content != &info) {
1538 /* This is a member of a container. Try starting the array. */
1539 int err;
1540 err = assemble_container_content(st, mdfd, content, c,
1541 chosen_name, NULL);
1542 close(mdfd);
1543 if (locked == 1)
1544 cluster_release_dlmlock();
1545 return err;
1546 }
1547
1548 /* Ok, no bad inconsistancy, we can try updating etc */
1549 devices = xcalloc(num_devs, sizeof(*devices));
1550 devmap = xcalloc(num_devs, content->array.raid_disks);
1551 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1552 c, content, mdfd, mddev,
1553 &most_recent, &bestcnt, &best, inargv);
1554 if (devcnt < 0) {
1555 mdfd = -3;
1556 goto out;
1557 }
1558
1559 if (devcnt == 0) {
1560 pr_err("no devices found for %s\n",
1561 mddev);
1562 if (st)
1563 st->ss->free_super(st);
1564 free(devmap);
1565 goto out;
1566 }
1567
1568 if (c->update && strcmp(c->update, "byteorder")==0)
1569 st->minor_version = 90;
1570
1571 st->ss->getinfo_super(st, content, NULL);
1572 clean = content->array.state & 1;
1573
1574 /* now we have some devices that might be suitable.
1575 * I wonder how many
1576 */
1577 avail = xcalloc(content->array.raid_disks, 1);
1578 okcnt = 0;
1579 replcnt = 0;
1580 sparecnt=0;
1581 journalcnt=0;
1582 rebuilding_cnt=0;
1583 for (i=0; i< bestcnt; i++) {
1584 int j = best[i];
1585 int event_margin = 1; /* always allow a difference of '1'
1586 * like the kernel does
1587 */
1588 if (j < 0) continue;
1589 /* note: we ignore error flags in multipath arrays
1590 * as they don't make sense
1591 */
1592 if (content->array.level != LEVEL_MULTIPATH) {
1593 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1594 if (content->journal_device_required)
1595 journalcnt++;
1596 else /* unexpected journal, mark as faulty */
1597 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1598 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1599 if (!(devices[j].i.disk.state
1600 & (1<<MD_DISK_FAULTY))) {
1601 devices[j].uptodate = 1;
1602 sparecnt++;
1603 }
1604 continue;
1605 }
1606 }
1607 /* If this device thinks that 'most_recent' has failed, then
1608 * we must reject this device.
1609 */
1610 if (j != most_recent && !c->force &&
1611 content->array.raid_disks > 0 &&
1612 devices[most_recent].i.disk.raid_disk >= 0 &&
1613 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1614 if (c->verbose > -1)
1615 pr_err("ignoring %s as it reports %s as failed\n",
1616 devices[j].devname, devices[most_recent].devname);
1617 best[i] = -1;
1618 continue;
1619 }
1620 /* Require event counter to be same as, or just less than,
1621 * most recent. If it is bigger, it must be a stray spare and
1622 * should be ignored.
1623 */
1624 if (devices[j].i.events+event_margin >=
1625 devices[most_recent].i.events &&
1626 devices[j].i.events <=
1627 devices[most_recent].i.events
1628 ) {
1629 devices[j].uptodate = 1;
1630 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1631 journal_clean = 1;
1632 if (i < content->array.raid_disks * 2) {
1633 if (devices[j].i.recovery_start == MaxSector ||
1634 (content->reshape_active &&
1635 i >= content->array.raid_disks - content->delta_disks)) {
1636 if (!avail[i/2]) {
1637 okcnt++;
1638 avail[i/2]=1;
1639 } else
1640 replcnt++;
1641 } else
1642 rebuilding_cnt++;
1643 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1644 sparecnt++;
1645 }
1646 }
1647 free(devmap);
1648 if (c->force) {
1649 int force_ok = force_array(content, devices, best, bestcnt,
1650 avail, most_recent, st, c);
1651 okcnt += force_ok;
1652 if (force_ok)
1653 was_forced = 1;
1654 }
1655 /* Now we want to look at the superblock which the kernel will base things on
1656 * and compare the devices that we think are working with the devices that the
1657 * superblock thinks are working.
1658 * If there are differences and --force is given, then update this chosen
1659 * superblock.
1660 */
1661 chosen_drive = -1;
1662 st->ss->free_super(st);
1663 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1664 int j = best[i];
1665 int fd;
1666
1667 if (j<0)
1668 continue;
1669 if (!devices[j].uptodate)
1670 continue;
1671 if (devices[j].i.events < devices[most_recent].i.events)
1672 continue;
1673 chosen_drive = j;
1674 if ((fd=dev_open(devices[j].devname,
1675 devices[j].included ? O_RDONLY
1676 : (O_RDONLY|O_EXCL)))< 0) {
1677 pr_err("Cannot open %s: %s\n",
1678 devices[j].devname, strerror(errno));
1679 goto out;
1680 }
1681 if (st->ss->load_super(st,fd, NULL)) {
1682 close(fd);
1683 pr_err("RAID superblock has disappeared from %s\n",
1684 devices[j].devname);
1685 goto out;
1686 }
1687 close(fd);
1688 }
1689 if (st->sb == NULL) {
1690 pr_err("No suitable drives found for %s\n", mddev);
1691 goto out;
1692 }
1693 st->ss->getinfo_super(st, content, NULL);
1694 if (sysfs_init(content, mdfd, NULL)) {
1695 pr_err("Unable to initialize sysfs\n");
1696 goto out;
1697 }
1698
1699 /* after reload context, store journal_clean in context */
1700 content->journal_clean = journal_clean;
1701 for (i=0; i<bestcnt; i++) {
1702 int j = best[i];
1703 unsigned int desired_state;
1704
1705 if (j < 0)
1706 continue;
1707 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1708 desired_state = (1<<MD_DISK_JOURNAL);
1709 else if (i >= content->array.raid_disks * 2)
1710 desired_state = 0;
1711 else if (i & 1)
1712 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1713 else
1714 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1715
1716 if (!devices[j].uptodate)
1717 continue;
1718
1719 devices[j].i.disk.state = desired_state;
1720 if (!(devices[j].i.array.state & 1))
1721 clean = 0;
1722
1723 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1724 c->verbose, 0, NULL)) {
1725 if (c->force) {
1726 if (c->verbose >= 0)
1727 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1728 j, mddev, devices[j].devname);
1729 change = 1;
1730 } else {
1731 if (c->verbose >= -1)
1732 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1733 i, mddev, devices[j].devname);
1734 }
1735 }
1736 #if 0
1737 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1738 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1739 i, mddev);
1740 }
1741 #endif
1742 }
1743 if (c->force && !clean &&
1744 !enough(content->array.level, content->array.raid_disks,
1745 content->array.layout, clean,
1746 avail)) {
1747 change += st->ss->update_super(st, content, "force-array",
1748 devices[chosen_drive].devname, c->verbose,
1749 0, NULL);
1750 was_forced = 1;
1751 clean = 1;
1752 }
1753
1754 if (change) {
1755 int fd;
1756 fd = dev_open(devices[chosen_drive].devname,
1757 devices[chosen_drive].included ?
1758 O_RDWR : (O_RDWR|O_EXCL));
1759 if (fd < 0) {
1760 pr_err("Could not open %s for write - cannot Assemble array.\n",
1761 devices[chosen_drive].devname);
1762 goto out;
1763 }
1764 if (st->ss->store_super(st, fd)) {
1765 close(fd);
1766 pr_err("Could not re-write superblock on %s\n",
1767 devices[chosen_drive].devname);
1768 goto out;
1769 }
1770 if (c->verbose >= 0)
1771 pr_err("Marking array %s as 'clean'\n",
1772 mddev);
1773 close(fd);
1774 }
1775
1776 /* If we are in the middle of a reshape we may need to restore saved data
1777 * that was moved aside due to the reshape overwriting live data
1778 * The code of doing this lives in Grow.c
1779 */
1780 if (content->reshape_active &&
1781 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1782 int err = 0;
1783 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1784 if (c->verbose > 0)
1785 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1786 chosen_name);
1787 if (!c->backup_file)
1788 c->backup_file = locate_backup(content->sys_name);
1789 enable_fds(bestcnt/2);
1790 for (i = 0; i < bestcnt/2; i++) {
1791 int j = best[i*2];
1792 if (j >= 0) {
1793 fdlist[i] = dev_open(devices[j].devname,
1794 devices[j].included
1795 ? O_RDWR : (O_RDWR|O_EXCL));
1796 if (fdlist[i] < 0) {
1797 pr_err("Could not open %s for write - cannot Assemble array.\n",
1798 devices[j].devname);
1799 err = 1;
1800 break;
1801 }
1802 } else
1803 fdlist[i] = -1;
1804 }
1805 if (!err) {
1806 if (st->ss->external && st->ss->recover_backup)
1807 err = st->ss->recover_backup(st, content);
1808 else
1809 err = Grow_restart(st, content, fdlist, bestcnt/2,
1810 c->backup_file, c->verbose > 0);
1811 if (err && c->invalid_backup) {
1812 if (c->verbose > 0)
1813 pr_err("continuing without restoring backup\n");
1814 err = 0;
1815 }
1816 }
1817 while (i>0) {
1818 i--;
1819 if (fdlist[i]>=0) close(fdlist[i]);
1820 }
1821 free(fdlist);
1822 if (err) {
1823 pr_err("Failed to restore critical section for reshape, sorry.\n");
1824 if (c->backup_file == NULL)
1825 cont_err("Possibly you needed to specify the --backup-file\n");
1826 goto out;
1827 }
1828 }
1829
1830 /* Almost ready to actually *do* something */
1831 /* First, fill in the map, so that udev can find our name
1832 * as soon as we become active.
1833 */
1834 if (c->update && strcmp(c->update, "metadata")==0) {
1835 content->array.major_version = 1;
1836 content->array.minor_version = 0;
1837 strcpy(content->text_version, "1.0");
1838 }
1839
1840 map_update(&map, fd2devnm(mdfd), content->text_version,
1841 content->uuid, chosen_name);
1842
1843 rv = start_array(mdfd, mddev, content,
1844 st, ident, best, bestcnt,
1845 chosen_drive, devices, okcnt, sparecnt,
1846 rebuilding_cnt, journalcnt,
1847 c,
1848 clean, avail, start_partial_ok,
1849 pre_exist != NULL,
1850 was_forced);
1851 if (rv == 1 && !pre_exist)
1852 ioctl(mdfd, STOP_ARRAY, NULL);
1853 free(devices);
1854 out:
1855 map_unlock(&map);
1856 if (rv == 0) {
1857 wait_for(chosen_name, mdfd);
1858 close(mdfd);
1859 if (auto_assem) {
1860 int usecs = 1;
1861 /* There is a nasty race with 'mdadm --monitor'.
1862 * If it opens this device before we close it,
1863 * it gets an incomplete open on which IO
1864 * doesn't work and the capacity is
1865 * wrong.
1866 * If we reopen (to check for layered devices)
1867 * before --monitor closes, we loose.
1868 *
1869 * So: wait upto 1 second for there to be
1870 * a non-zero capacity.
1871 */
1872 while (usecs < 1000) {
1873 mdfd = open(mddev, O_RDONLY);
1874 if (mdfd >= 0) {
1875 unsigned long long size;
1876 if (get_dev_size(mdfd, NULL, &size) &&
1877 size > 0)
1878 break;
1879 close(mdfd);
1880 }
1881 usleep(usecs);
1882 usecs <<= 1;
1883 }
1884 }
1885 } else if (mdfd >= 0)
1886 close(mdfd);
1887
1888 /* '2' means 'OK, but not started yet' */
1889 if (locked == 1)
1890 cluster_release_dlmlock();
1891 if (rv == -1) {
1892 free(devices);
1893 return 1;
1894 }
1895 return rv == 2 ? 0 : rv;
1896 }
1897
1898 int assemble_container_content(struct supertype *st, int mdfd,
1899 struct mdinfo *content, struct context *c,
1900 char *chosen_name, int *result)
1901 {
1902 struct mdinfo *dev, *sra, *dev2;
1903 int working = 0, preexist = 0;
1904 int expansion = 0;
1905 int old_raid_disks;
1906 int start_reshape;
1907 char *avail;
1908 int err;
1909
1910 if (sysfs_init(content, mdfd, NULL)) {
1911 pr_err("Unable to initialize sysfs\n");
1912 return 1;
1913 }
1914
1915 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1916 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1917 if (content->array.major_version == -1 &&
1918 content->array.minor_version == -2 &&
1919 c->readonly &&
1920 content->text_version[0] == '/')
1921 content->text_version[0] = '-';
1922 if (sysfs_set_array(content, 9003) != 0) {
1923 sysfs_free(sra);
1924 return 1;
1925 }
1926 }
1927
1928 /* There are two types of reshape: container wide or sub-array specific
1929 * Check if metadata requests blocking container wide reshapes
1930 */
1931 start_reshape = (content->reshape_active &&
1932 !((content->reshape_active == CONTAINER_RESHAPE) &&
1933 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1934
1935 /* Block subarray here if it is under reshape now
1936 * Do not allow for any changes in this array
1937 */
1938 if (st->ss->external && content->recovery_blocked && start_reshape)
1939 block_subarray(content);
1940
1941 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1942 for (dev = content->devs; dev; dev = dev->next)
1943 if (dev2->disk.major == dev->disk.major &&
1944 dev2->disk.minor == dev->disk.minor)
1945 break;
1946 if (dev)
1947 continue;
1948 /* Don't want this one any more */
1949 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1950 errno == EBUSY) {
1951 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1952 sysfs_free(sra);
1953 return 1;
1954 }
1955 sysfs_set_str(sra, dev2, "state", "remove");
1956 }
1957 old_raid_disks = content->array.raid_disks - content->delta_disks;
1958 avail = xcalloc(content->array.raid_disks, 1);
1959 for (dev = content->devs; dev; dev = dev->next) {
1960 if (dev->disk.raid_disk >= 0)
1961 avail[dev->disk.raid_disk] = 1;
1962 if (sysfs_add_disk(content, dev, 1) == 0) {
1963 if (dev->disk.raid_disk >= old_raid_disks &&
1964 content->reshape_active)
1965 expansion++;
1966 else
1967 working++;
1968 } else if (errno == EEXIST)
1969 preexist++;
1970 }
1971 sysfs_free(sra);
1972 if (working + expansion == 0 && c->runstop <= 0) {
1973 free(avail);
1974 return 1;/* Nothing new, don't try to start */
1975 }
1976 map_update(NULL, fd2devnm(mdfd), content->text_version,
1977 content->uuid, chosen_name);
1978
1979 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1980 st->ss->validate_ppl) {
1981 content->array.state |= 1;
1982 err = 0;
1983
1984 for (dev = content->devs; dev; dev = dev->next) {
1985 int dfd;
1986 char *devpath;
1987 int ret;
1988
1989 ret = st->ss->validate_ppl(st, content, dev);
1990 if (ret == 0)
1991 continue;
1992
1993 if (ret < 0) {
1994 err = 1;
1995 break;
1996 }
1997
1998 if (!c->force) {
1999 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
2000 chosen_name);
2001 content->array.state &= ~1;
2002 avail[dev->disk.raid_disk] = 0;
2003 break;
2004 }
2005
2006 /* have --force - overwrite the invalid ppl */
2007 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
2008 dfd = dev_open(devpath, O_RDWR);
2009 if (dfd < 0) {
2010 pr_err("Failed to open %s\n", devpath);
2011 err = 1;
2012 break;
2013 }
2014
2015 err = st->ss->write_init_ppl(st, content, dfd);
2016 close(dfd);
2017
2018 if (err)
2019 break;
2020 }
2021
2022 if (err) {
2023 free(avail);
2024 return err;
2025 }
2026 }
2027
2028 if (enough(content->array.level, content->array.raid_disks,
2029 content->array.layout, content->array.state & 1, avail) == 0) {
2030 if (c->export && result)
2031 *result |= INCR_NO;
2032 else if (c->verbose >= 0) {
2033 pr_err("%s assembled with %d device%s",
2034 chosen_name, preexist + working,
2035 preexist + working == 1 ? "":"s");
2036 if (preexist)
2037 fprintf(stderr, " (%d new)", working);
2038 fprintf(stderr, " but not started\n");
2039 }
2040 free(avail);
2041 return 1;
2042 }
2043 free(avail);
2044
2045 if (c->runstop <= 0 &&
2046 (working + preexist + expansion) <
2047 content->array.working_disks) {
2048 if (c->export && result)
2049 *result |= INCR_UNSAFE;
2050 else if (c->verbose >= 0) {
2051 pr_err("%s assembled with %d device%s",
2052 chosen_name, preexist + working,
2053 preexist + working == 1 ? "":"s");
2054 if (preexist)
2055 fprintf(stderr, " (%d new)", working);
2056 fprintf(stderr, " but not safe to start\n");
2057 }
2058 return 1;
2059 }
2060
2061
2062 if (start_reshape) {
2063 int spare = content->array.raid_disks + expansion;
2064 if (restore_backup(st, content,
2065 working,
2066 spare, &c->backup_file, c->verbose) == 1)
2067 return 1;
2068
2069 err = sysfs_set_str(content, NULL,
2070 "array_state", "readonly");
2071 if (err)
2072 return 1;
2073
2074 if (st->ss->external) {
2075 if (!mdmon_running(st->container_devnm))
2076 start_mdmon(st->container_devnm);
2077 ping_monitor(st->container_devnm);
2078 if (mdmon_running(st->container_devnm) &&
2079 st->update_tail == NULL)
2080 st->update_tail = &st->updates;
2081 }
2082
2083 err = Grow_continue(mdfd, st, content, c->backup_file,
2084 0, c->freeze_reshape);
2085 } else switch(content->array.level) {
2086 case LEVEL_LINEAR:
2087 case LEVEL_MULTIPATH:
2088 case 0:
2089 err = sysfs_set_str(content, NULL, "array_state",
2090 c->readonly ? "readonly" : "active");
2091 break;
2092 default:
2093 err = sysfs_set_str(content, NULL, "array_state",
2094 "readonly");
2095 /* start mdmon if needed. */
2096 if (!err) {
2097 if (!mdmon_running(st->container_devnm))
2098 start_mdmon(st->container_devnm);
2099 ping_monitor(st->container_devnm);
2100 }
2101 break;
2102 }
2103 if (!err)
2104 sysfs_set_safemode(content, content->safe_mode_delay);
2105
2106 /* Block subarray here if it is not reshaped now
2107 * It has be blocked a little later to allow mdmon to switch in
2108 * in to R/W state
2109 */
2110 if (st->ss->external && content->recovery_blocked &&
2111 !start_reshape)
2112 block_subarray(content);
2113
2114 if (c->export && result) {
2115 if (err)
2116 *result |= INCR_NO;
2117 else
2118 *result |= INCR_YES;
2119 } else if (c->verbose >= 0) {
2120 if (err)
2121 pr_err("array %s now has %d device%s",
2122 chosen_name, working + preexist,
2123 working + preexist == 1 ? "":"s");
2124 else
2125 pr_err("Started %s with %d device%s",
2126 chosen_name, working + preexist,
2127 working + preexist == 1 ? "":"s");
2128 if (preexist)
2129 fprintf(stderr, " (%d new)", working);
2130 if (expansion)
2131 fprintf(stderr, " ( + %d for expansion)",
2132 expansion);
2133 fprintf(stderr, "\n");
2134 }
2135 if (!err)
2136 wait_for(chosen_name, mdfd);
2137 return err;
2138 /* FIXME should have an O_EXCL and wait for read-auto */
2139 }
Unnamed repository; edit this file 'description' to name the repository.