]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Add failfast support.
[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
153 tmpdev = devlist; num_devs = 0;
154 while (tmpdev) {
155 if (tmpdev->used)
156 tmpdev->used = 2;
157 else
158 num_devs++;
159 tmpdev->disposition = 0;
160 tmpdev = tmpdev->next;
161 }
162
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
166 */
167 for (tmpdev = devlist;
168 tmpdev;
169 tmpdev = tmpdev ? tmpdev->next : NULL) {
170 char *devname = tmpdev->devname;
171 int dfd;
172 struct stat stb;
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(dfd, &stb)< 0) {
208 /* Impossible! */
209 pr_err("fstat failed for %s: %s\n",
210 devname, strerror(errno));
211 tmpdev->used = 2;
212 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
213 pr_err("%s is not a block device.\n",
214 devname);
215 tmpdev->used = 2;
216 } else if (must_be_container(dfd)) {
217 if (st) {
218 /* already found some components, this cannot
219 * be another one.
220 */
221 if (report_mismatch)
222 pr_err("%s is a container, but we are looking for components\n",
223 devname);
224 tmpdev->used = 2;
225 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
226 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
227 if (report_mismatch)
228 pr_err("not a recognisable container: %s\n",
229 devname);
230 tmpdev->used = 2;
231 #endif
232 } else if (!tst->ss->load_container
233 || tst->ss->load_container(tst, dfd, NULL)) {
234 if (report_mismatch)
235 pr_err("no correct container type: %s\n",
236 devname);
237 tmpdev->used = 2;
238 } else if (auto_assem &&
239 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
240 tst->ss->match_home(tst, c->homehost) == 1)) {
241 if (report_mismatch)
242 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
243 devname, tst->ss->name);
244 tmpdev->used = 2;
245 } else
246 found_container = 1;
247 } else {
248 if (!tst && (tst = guess_super(dfd)) == NULL) {
249 if (report_mismatch)
250 pr_err("no recogniseable superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if ((tst->ignore_hw_compat = 0),
254 tst->ss->load_super(tst, dfd,
255 report_mismatch ? devname : NULL)) {
256 if (report_mismatch)
257 pr_err("no RAID superblock on %s\n",
258 devname);
259 tmpdev->used = 2;
260 } else if (tst->ss->compare_super == NULL) {
261 if (report_mismatch)
262 pr_err("Cannot assemble %s metadata on %s\n",
263 tst->ss->name, devname);
264 tmpdev->used = 2;
265 } else if (auto_assem && st == NULL &&
266 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
267 tst->ss->match_home(tst, c->homehost) == 1)) {
268 if (report_mismatch)
269 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
270 devname, tst->ss->name);
271 tmpdev->used = 2;
272 }
273 }
274 if (dfd >= 0) close(dfd);
275 if (tmpdev->used == 2) {
276 if (auto_assem || !inargv)
277 /* Ignore unrecognised devices during auto-assembly */
278 goto loop;
279 if (ident->uuid_set || ident->name[0] ||
280 ident->super_minor != UnSet)
281 /* Ignore unrecognised device if looking for
282 * specific array */
283 goto loop;
284
285 pr_err("%s has no superblock - assembly aborted\n",
286 devname);
287 if (st)
288 st->ss->free_super(st);
289 dev_policy_free(pol);
290 domain_free(domains);
291 return -1;
292 }
293
294 if (found_container) {
295 /* tmpdev is a container. We need to be either
296 * looking for a member, or auto-assembling
297 */
298 /* should be safe to try an exclusive open now, we
299 * have rejected anything that some other mdadm might
300 * be looking at
301 */
302 dfd = dev_open(devname, O_RDONLY | O_EXCL);
303 if (dfd < 0) {
304 if (report_mismatch)
305 pr_err("%s is busy - skipping\n", devname);
306 goto loop;
307 }
308 close(dfd);
309
310 if (ident->container && ident->container[0] != '/') {
311 /* we have a uuid */
312 int uuid[4];
313
314 content = *contentp;
315 tst->ss->getinfo_super(tst, content, NULL);
316
317 if (!parse_uuid(ident->container, uuid) ||
318 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
319 if (report_mismatch)
320 pr_err("%s has wrong UUID to be required container\n",
321 devname);
322 goto loop;
323 }
324 }
325 /* It is worth looking inside this container.
326 */
327 if (c->verbose > 0)
328 pr_err("looking in container %s\n",
329 devname);
330
331 for (content = tst->ss->container_content(tst, NULL);
332 content;
333 content = content->next) {
334
335 if (!ident_matches(ident, content, tst,
336 c->homehost, c->require_homehost,
337 c->update,
338 report_mismatch ? devname : NULL))
339 /* message already printed */;
340 else if (is_member_busy(content->text_version)) {
341 if (report_mismatch)
342 pr_err("member %s in %s is already assembled\n",
343 content->text_version,
344 devname);
345 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
346 /* do not assemble arrays with unsupported configurations */
347 pr_err("Cannot activate member %s in %s.\n",
348 content->text_version,
349 devname);
350 } else
351 break;
352 }
353 if (!content) {
354 tmpdev->used = 2;
355 goto loop; /* empty container */
356 }
357
358 st = tst; tst = NULL;
359 if (!auto_assem && inargv && tmpdev->next != NULL) {
360 pr_err("%s is a container, but is not only device given: confused and aborting\n",
361 devname);
362 st->ss->free_super(st);
363 dev_policy_free(pol);
364 domain_free(domains);
365 return -1;
366 }
367 if (c->verbose > 0)
368 pr_err("found match on member %s in %s\n",
369 content->text_version, devname);
370
371 /* make sure we finished the loop */
372 tmpdev = NULL;
373 goto loop;
374 } else {
375 content = *contentp;
376 tst->ss->getinfo_super(tst, content, NULL);
377
378 if (!ident_matches(ident, content, tst,
379 c->homehost, c->require_homehost,
380 c->update,
381 report_mismatch ? devname : NULL))
382 goto loop;
383
384 if (auto_assem) {
385 /* Never auto-assemble things that conflict
386 * with mdadm.conf in some way
387 */
388 struct mddev_ident *match;
389 int rv = 0;
390
391 match = conf_match(tst, content, devname,
392 report_mismatch ? c->verbose : -1,
393 &rv);
394 if (!match && rv == 2)
395 goto loop;
396 if (match && match->devname &&
397 strcasecmp(match->devname, "<ignore>") == 0) {
398 if (report_mismatch)
399 pr_err("%s is a member of an explicitly ignored array\n",
400 devname);
401 goto loop;
402 }
403 if (match && !ident_matches(match, content, tst,
404 c->homehost, c->require_homehost,
405 c->update,
406 report_mismatch ? devname : NULL))
407 /* Array exists in mdadm.conf but some
408 * details don't match, so reject it
409 */
410 goto loop;
411 }
412
413 /* should be safe to try an exclusive open now, we
414 * have rejected anything that some other mdadm might
415 * be looking at
416 */
417 dfd = dev_open(devname, O_RDONLY | O_EXCL);
418 if (dfd < 0) {
419 if (report_mismatch)
420 pr_err("%s is busy - skipping\n", devname);
421 goto loop;
422 }
423 close(dfd);
424
425 if (st == NULL)
426 st = dup_super(tst);
427 if (st->minor_version == -1)
428 st->minor_version = tst->minor_version;
429
430 if (memcmp(content->uuid, uuid_zero,
431 sizeof(int[4])) == 0) {
432 /* this is a floating spare. It cannot define
433 * an array unless there are no more arrays of
434 * this type to be found. It can be included
435 * in an array of this type though.
436 */
437 tmpdev->used = 3;
438 goto loop;
439 }
440
441 if (st->ss != tst->ss ||
442 st->minor_version != tst->minor_version ||
443 st->ss->compare_super(st, tst) != 0) {
444 /* Some mismatch. If exactly one array matches this host,
445 * we can resolve on that one.
446 * Or, if we are auto assembling, we just ignore the second
447 * for now.
448 */
449 if (auto_assem)
450 goto loop;
451 if (c->homehost) {
452 int first = st->ss->match_home(st, c->homehost);
453 int last = tst->ss->match_home(tst, c->homehost);
454 if (first != last &&
455 (first == 1 || last == 1)) {
456 /* We can do something */
457 if (first) {/* just ignore this one */
458 if (report_mismatch)
459 pr_err("%s misses out due to wrong homehost\n",
460 devname);
461 goto loop;
462 } else { /* reject all those sofar */
463 struct mddev_dev *td;
464 if (report_mismatch)
465 pr_err("%s overrides previous devices due to good homehost\n",
466 devname);
467 for (td=devlist; td != tmpdev; td=td->next)
468 if (td->used == 1)
469 td->used = 0;
470 tmpdev->used = 1;
471 goto loop;
472 }
473 }
474 }
475 pr_err("superblock on %s doesn't match others - assembly aborted\n",
476 devname);
477 tst->ss->free_super(tst);
478 st->ss->free_super(st);
479 dev_policy_free(pol);
480 domain_free(domains);
481 return -1;
482 }
483 tmpdev->used = 1;
484 }
485 loop:
486 /* Collect domain information from members only */
487 if (tmpdev && tmpdev->used == 1) {
488 if (!pol)
489 pol = devid_policy(stb.st_rdev);
490 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
491 }
492 dev_policy_free(pol);
493 pol = NULL;
494 if (tst)
495 tst->ss->free_super(tst);
496 }
497
498 /* Check if we found some imsm spares but no members */
499 if ((auto_assem ||
500 (ident->uuid_set &&
501 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
502 (!st || !st->sb))
503 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
504 if (tmpdev->used != 3)
505 continue;
506 tmpdev->used = 1;
507 content = *contentp;
508
509 if (!st->sb) {
510 /* we need sb from one of the spares */
511 int dfd = dev_open(tmpdev->devname, O_RDONLY);
512 if (dfd < 0 ||
513 st->ss->load_super(st, dfd, NULL))
514 tmpdev->used = 2;
515 if (dfd > 0)
516 close(dfd);
517 }
518 }
519
520 /* Now reject spares that don't match domains of identified members */
521 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
522 struct stat stb;
523 if (tmpdev->used != 3)
524 continue;
525 if (stat(tmpdev->devname, &stb)< 0) {
526 pr_err("fstat failed for %s: %s\n",
527 tmpdev->devname, strerror(errno));
528 tmpdev->used = 2;
529 } else {
530 struct dev_policy *pol = devid_policy(stb.st_rdev);
531 int dt = domain_test(domains, pol, NULL);
532 if (inargv && dt != 0)
533 /* take this spare as domains match
534 * if there are any */
535 tmpdev->used = 1;
536 else if (!inargv && dt == 1)
537 /* device wasn't explicitly listed, so need
538 * explicit domain match - which we have */
539 tmpdev->used = 1;
540 else
541 /* if domains don't match mark as unused */
542 tmpdev->used = 0;
543 dev_policy_free(pol);
544 }
545 }
546 domain_free(domains);
547 *stp = st;
548 if (st && st->sb && content == *contentp)
549 st->ss->getinfo_super(st, content, NULL);
550 *contentp = content;
551
552 return num_devs;
553 }
554
555 struct devs {
556 char *devname;
557 int uptodate; /* set once we decide that this device is as
558 * recent as everything else in the array.
559 */
560 int included; /* set if the device is already in the array
561 * due to a previous '-I'
562 */
563 struct mdinfo i;
564 };
565
566 static int load_devices(struct devs *devices, char *devmap,
567 struct mddev_ident *ident, struct supertype **stp,
568 struct mddev_dev *devlist, struct context *c,
569 struct mdinfo *content,
570 int mdfd, char *mddev,
571 int *most_recentp, int *bestcntp, int **bestp,
572 int inargv)
573 {
574 struct mddev_dev *tmpdev;
575 int devcnt = 0;
576 int nextspare = 0;
577 #ifndef MDASSEMBLE
578 int bitmap_done = 0;
579 #endif
580 int most_recent = -1;
581 int bestcnt = 0;
582 int *best = *bestp;
583 struct supertype *st = *stp;
584
585 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
586 char *devname = tmpdev->devname;
587 struct stat stb;
588 struct supertype *tst;
589 int i;
590 int dfd;
591
592 if (tmpdev->used != 1)
593 continue;
594 /* looks like a good enough match to update the super block if needed */
595 #ifndef MDASSEMBLE
596 if (c->update) {
597 /* prepare useful information in info structures */
598 struct stat stb2;
599 int err;
600 fstat(mdfd, &stb2);
601
602 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
603 random_uuid((__u8 *)ident->uuid);
604
605 dfd = dev_open(devname,
606 tmpdev->disposition == 'I'
607 ? O_RDWR : (O_RDWR|O_EXCL));
608
609 tst = dup_super(st);
610 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
611 pr_err("cannot re-read metadata from %s - aborting\n",
612 devname);
613 if (dfd >= 0)
614 close(dfd);
615 close(mdfd);
616 free(devices);
617 free(devmap);
618 tst->ss->free_super(tst);
619 free(tst);
620 *stp = st;
621 return -1;
622 }
623 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
624
625 memcpy(content->uuid, ident->uuid, 16);
626 strcpy(content->name, ident->name);
627 content->array.md_minor = minor(stb2.st_rdev);
628
629 if (strcmp(c->update, "byteorder") == 0)
630 err = 0;
631 else if (strcmp(c->update, "home-cluster") == 0) {
632 tst->cluster_name = c->homecluster;
633 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
634 } else if (strcmp(c->update, "nodes") == 0) {
635 tst->nodes = c->nodes;
636 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
637 } else if (strcmp(c->update, "revert-reshape") == 0 &&
638 c->invalid_backup)
639 err = tst->ss->update_super(tst, content,
640 "revert-reshape-nobackup",
641 devname, c->verbose,
642 ident->uuid_set,
643 c->homehost);
644 else
645 err = tst->ss->update_super(tst, content, c->update,
646 devname, c->verbose,
647 ident->uuid_set,
648 c->homehost);
649 if (err < 0) {
650 if (err == -1)
651 pr_err("--update=%s not understood for %s metadata\n",
652 c->update, tst->ss->name);
653 tst->ss->free_super(tst);
654 free(tst);
655 close(mdfd);
656 close(dfd);
657 free(devices);
658 free(devmap);
659 *stp = st;
660 return -1;
661 }
662 if (strcmp(c->update, "uuid")==0 &&
663 !ident->uuid_set) {
664 ident->uuid_set = 1;
665 memcpy(ident->uuid, content->uuid, 16);
666 }
667 if (tst->ss->store_super(tst, dfd))
668 pr_err("Could not re-write superblock on %s.\n",
669 devname);
670
671 if (strcmp(c->update, "uuid")==0 &&
672 ident->bitmap_fd >= 0 && !bitmap_done) {
673 if (bitmap_update_uuid(ident->bitmap_fd,
674 content->uuid,
675 tst->ss->swapuuid) != 0)
676 pr_err("Could not update uuid on external bitmap.\n");
677 else
678 bitmap_done = 1;
679 }
680 } else
681 #endif
682 {
683 dfd = dev_open(devname,
684 tmpdev->disposition == 'I'
685 ? O_RDWR : (O_RDWR|O_EXCL));
686 tst = dup_super(st);
687
688 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
689 pr_err("cannot re-read metadata from %s - aborting\n",
690 devname);
691 if (dfd >= 0)
692 close(dfd);
693 close(mdfd);
694 free(devices);
695 free(devmap);
696 tst->ss->free_super(tst);
697 free(tst);
698 *stp = st;
699 return -1;
700 }
701 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
702 }
703
704 fstat(dfd, &stb);
705 close(dfd);
706
707 if (c->verbose > 0)
708 pr_err("%s is identified as a member of %s, slot %d%s.\n",
709 devname, mddev, content->disk.raid_disk,
710 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
711 devices[devcnt].devname = devname;
712 devices[devcnt].uptodate = 0;
713 devices[devcnt].included = (tmpdev->disposition == 'I');
714 devices[devcnt].i = *content;
715 devices[devcnt].i.disk.major = major(stb.st_rdev);
716 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
717
718 if (devices[devcnt].i.disk.state == 6) {
719 if (most_recent < 0 ||
720 devices[devcnt].i.events
721 > devices[most_recent].i.events) {
722 struct supertype *tmp = tst;
723 tst = st;
724 st = tmp;
725 most_recent = devcnt;
726 }
727 }
728 tst->ss->free_super(tst);
729 free(tst);
730
731 if (content->array.level == LEVEL_MULTIPATH)
732 /* with multipath, the raid_disk from the superblock is meaningless */
733 i = devcnt;
734 else
735 i = devices[devcnt].i.disk.raid_disk;
736 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
737 if (nextspare < content->array.raid_disks*2)
738 nextspare = content->array.raid_disks*2;
739 i = nextspare++;
740 } else {
741 /* i is raid_disk - double it so there is room for
742 * replacements */
743 i *= 2;
744 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
745 i++;
746 if (i >= content->array.raid_disks*2 &&
747 i >= nextspare)
748 nextspare = i+1;
749 }
750 if (i < 10000) {
751 if (i >= bestcnt) {
752 int newbestcnt = i+10;
753 int *newbest = xmalloc(sizeof(int)*newbestcnt);
754 int c;
755 for (c=0; c < newbestcnt; c++)
756 if (c < bestcnt)
757 newbest[c] = best[c];
758 else
759 newbest[c] = -1;
760 if (best)free(best);
761 best = newbest;
762 bestcnt = newbestcnt;
763 }
764 if (best[i] >=0 &&
765 devices[best[i]].i.events
766 == devices[devcnt].i.events
767 && (devices[best[i]].i.disk.minor
768 != devices[devcnt].i.disk.minor)
769 && st->ss == &super0
770 && content->array.level != LEVEL_MULTIPATH) {
771 /* two different devices with identical superblock.
772 * Could be a mis-detection caused by overlapping
773 * partitions. fail-safe.
774 */
775 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
776 " If they are really different, please --zero the superblock on one\n"
777 " If they are the same or overlap, please remove one from %s.\n",
778 devices[best[i]].devname, devname,
779 inargv ? "the list" :
780 "the\n DEVICE list in mdadm.conf"
781 );
782 close(mdfd);
783 free(devices);
784 free(devmap);
785 *stp = st;
786 return -1;
787 }
788 if (best[i] == -1
789 || (devices[best[i]].i.events
790 < devices[devcnt].i.events))
791 best[i] = 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 ||
815 (content->reshape_active && content->delta_disks > 0 &&
816 !enough(content->array.level, (content->array.raid_disks
817 - content->delta_disks),
818 content->new_layout, 1,
819 avail)
820 )) {
821 /* Choose the newest best drive which is
822 * not up-to-date, update the superblock
823 * and add it.
824 */
825 int fd;
826 struct supertype *tst;
827 unsigned long long current_events;
828 int chosen_drive = -1;
829 int i;
830
831 for (i = 0;
832 i < content->array.raid_disks * 2 && i < bestcnt;
833 i += 2) {
834 int j = best[i];
835 if (j < 0)
836 continue;
837 if (devices[j].uptodate)
838 continue;
839 if (devices[j].i.recovery_start != MaxSector) {
840 int delta;
841 if (!devices[j].i.reshape_active ||
842 devices[j].i.delta_disks <= 0)
843 continue;
844 /* When increasing number of devices, an
845 * added device also appears to be
846 * recovering. It is safe to include it
847 * as long as it won't be a source of
848 * data.
849 * For now, just allow for last data
850 * devices in RAID4 or last devices in RAID4/5/6.
851 */
852 delta = devices[j].i.delta_disks;
853 if (devices[j].i.array.level >= 4 &&
854 devices[j].i.array.level <= 6 &&
855 i/2 >= content->array.raid_disks - delta)
856 /* OK */;
857 else if (devices[j].i.array.level == 4 &&
858 i/2 >= content->array.raid_disks - delta - 1)
859 /* OK */;
860 else
861 continue;
862 }
863 if (chosen_drive < 0 ||
864 devices[j].i.events
865 > devices[chosen_drive].i.events)
866 chosen_drive = j;
867 }
868 if (chosen_drive < 0)
869 break;
870 current_events = devices[chosen_drive].i.events;
871 add_another:
872 if (c->verbose >= 0)
873 pr_err("forcing event count in %s(%d) from %d upto %d\n",
874 devices[chosen_drive].devname,
875 devices[chosen_drive].i.disk.raid_disk,
876 (int)(devices[chosen_drive].i.events),
877 (int)(devices[most_recent].i.events));
878 fd = dev_open(devices[chosen_drive].devname,
879 devices[chosen_drive].included ? O_RDWR
880 : (O_RDWR|O_EXCL));
881 if (fd < 0) {
882 pr_err("Couldn't open %s for write - not updating\n",
883 devices[chosen_drive].devname);
884 devices[chosen_drive].i.events = 0;
885 continue;
886 }
887 tst = dup_super(st);
888 if (tst->ss->load_super(tst,fd, NULL)) {
889 close(fd);
890 pr_err("RAID superblock disappeared from %s - not updating.\n",
891 devices[chosen_drive].devname);
892 devices[chosen_drive].i.events = 0;
893 continue;
894 }
895 content->events = devices[most_recent].i.events;
896 tst->ss->update_super(tst, content, "force-one",
897 devices[chosen_drive].devname, c->verbose,
898 0, NULL);
899
900 if (tst->ss->store_super(tst, fd)) {
901 close(fd);
902 pr_err("Could not re-write superblock on %s\n",
903 devices[chosen_drive].devname);
904 devices[chosen_drive].i.events = 0;
905 tst->ss->free_super(tst);
906 continue;
907 }
908 close(fd);
909 devices[chosen_drive].i.events = devices[most_recent].i.events;
910 devices[chosen_drive].uptodate = 1;
911 avail[chosen_drive] = 1;
912 okcnt++;
913 tst->ss->free_super(tst);
914 /* If there are any other drives of the same vintage,
915 * add them in as well. We can't lose and we might gain
916 */
917 for (i = 0;
918 i < content->array.raid_disks * 2 && i < bestcnt ;
919 i += 2) {
920 int j = best[i];
921 if (j >= 0 &&
922 !devices[j].uptodate &&
923 devices[j].i.recovery_start == MaxSector &&
924 devices[j].i.events == current_events) {
925 chosen_drive = j;
926 goto add_another;
927 }
928 }
929 }
930 return okcnt;
931 }
932
933 static int start_array(int mdfd,
934 char *mddev,
935 struct mdinfo *content,
936 struct supertype *st,
937 struct mddev_ident *ident,
938 int *best, int bestcnt,
939 int chosen_drive,
940 struct devs *devices,
941 unsigned int okcnt,
942 unsigned int sparecnt,
943 unsigned int rebuilding_cnt,
944 unsigned int journalcnt,
945 struct context *c,
946 int clean, char *avail,
947 int start_partial_ok,
948 int err_ok,
949 int was_forced
950 )
951 {
952 int rv;
953 int i;
954 unsigned int req_cnt;
955
956 if (content->journal_device_required && (content->journal_clean == 0)) {
957 if (!c->force) {
958 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
959 return 1;
960 }
961 pr_err("Journal is missing or stale, starting array read only.\n");
962 c->readonly = 1;
963 }
964
965 rv = set_array_info(mdfd, st, content);
966 if (rv && !err_ok) {
967 pr_err("failed to set array info for %s: %s\n",
968 mddev, strerror(errno));
969 return 1;
970 }
971 if (ident->bitmap_fd >= 0) {
972 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
973 pr_err("SET_BITMAP_FILE failed.\n");
974 return 1;
975 }
976 } else if (ident->bitmap_file) {
977 /* From config file */
978 int bmfd = open(ident->bitmap_file, O_RDWR);
979 if (bmfd < 0) {
980 pr_err("Could not open bitmap file %s\n",
981 ident->bitmap_file);
982 return 1;
983 }
984 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
985 pr_err("Failed to set bitmapfile for %s\n", mddev);
986 close(bmfd);
987 return 1;
988 }
989 close(bmfd);
990 }
991
992 /* First, add the raid disks, but add the chosen one last */
993 for (i=0; i<= bestcnt; i++) {
994 int j;
995 if (i < bestcnt) {
996 j = best[i];
997 if (j == chosen_drive)
998 continue;
999 } else
1000 j = chosen_drive;
1001
1002 if (j >= 0 && !devices[j].included) {
1003 int dfd = dev_open(devices[j].devname,
1004 O_RDWR|O_EXCL);
1005 if (dfd >= 0) {
1006 remove_partitions(dfd);
1007 close(dfd);
1008 }
1009 rv = add_disk(mdfd, st, content, &devices[j].i);
1010
1011 if (rv) {
1012 pr_err("failed to add %s to %s: %s\n",
1013 devices[j].devname,
1014 mddev,
1015 strerror(errno));
1016 if (i < content->array.raid_disks * 2
1017 || i == bestcnt)
1018 okcnt--;
1019 else
1020 sparecnt--;
1021 } else if (c->verbose > 0)
1022 pr_err("added %s to %s as %d%s%s\n",
1023 devices[j].devname, mddev,
1024 devices[j].i.disk.raid_disk,
1025 devices[j].uptodate?"":
1026 " (possibly out of date)",
1027 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1028 } else if (j >= 0) {
1029 if (c->verbose > 0)
1030 pr_err("%s is already in %s as %d\n",
1031 devices[j].devname, mddev,
1032 devices[j].i.disk.raid_disk);
1033 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1034 && (i&1) == 0)
1035 pr_err("no uptodate device for slot %d of %s\n",
1036 i/2, mddev);
1037 }
1038
1039 if (content->array.level == LEVEL_CONTAINER) {
1040 if (c->verbose >= 0) {
1041 pr_err("Container %s has been assembled with %d drive%s",
1042 mddev, okcnt+sparecnt+journalcnt,
1043 okcnt+sparecnt+journalcnt==1?"":"s");
1044 if (okcnt < (unsigned)content->array.raid_disks)
1045 fprintf(stderr, " (out of %d)",
1046 content->array.raid_disks);
1047 fprintf(stderr, "\n");
1048 }
1049
1050 if (st->ss->validate_container) {
1051 struct mdinfo *devices_list;
1052 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1053 unsigned int count;
1054 devices_list = NULL;
1055 for (count = 0; count < okcnt+sparecnt; count++) {
1056 info_devices[count] = devices[count].i;
1057 info_devices[count].next = devices_list;
1058 devices_list = &info_devices[count];
1059 }
1060 if (st->ss->validate_container(devices_list))
1061 pr_err("Mismatch detected!\n");
1062 free(info_devices);
1063 }
1064
1065 st->ss->free_super(st);
1066 sysfs_uevent(content, "change");
1067 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1068 /* Was partial, is still partial, so signal an error
1069 * to ensure we don't retry */
1070 return 1;
1071 return 0;
1072 }
1073
1074 /* Get number of in-sync devices according to the superblock.
1075 * We must have this number to start the array without -s or -R
1076 */
1077 req_cnt = content->array.working_disks;
1078
1079 if (c->runstop == 1 ||
1080 (c->runstop <= 0 &&
1081 ( enough(content->array.level, content->array.raid_disks,
1082 content->array.layout, clean, avail) &&
1083 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1084 ))) {
1085 /* This array is good-to-go.
1086 * If a reshape is in progress then we might need to
1087 * continue monitoring it. In that case we start
1088 * it read-only and let the grow code make it writable.
1089 */
1090 int rv;
1091 #ifndef MDASSEMBLE
1092 if (content->reshape_active &&
1093 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1094 content->delta_disks <= 0) {
1095 if (!c->backup_file) {
1096 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1097 mddev);
1098 pr_err("Please provided one with \"--backup-file=...\"\n");
1099 if (c->update &&
1100 strcmp(c->update, "revert-reshape") == 0)
1101 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1102 return 1;
1103 }
1104 rv = sysfs_set_str(content, NULL,
1105 "array_state", "readonly");
1106 if (rv == 0)
1107 rv = Grow_continue(mdfd, st, content,
1108 c->backup_file, 0,
1109 c->freeze_reshape);
1110 } else if (c->readonly &&
1111 sysfs_attribute_available(
1112 content, NULL, "array_state")) {
1113 rv = sysfs_set_str(content, NULL,
1114 "array_state", "readonly");
1115 } else
1116 #endif
1117 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1118 reopen_mddev(mdfd); /* drop O_EXCL */
1119 if (rv == 0) {
1120 if (c->verbose >= 0) {
1121 pr_err("%s has been started with %d drive%s",
1122 mddev, okcnt, okcnt==1?"":"s");
1123 if (okcnt < (unsigned)content->array.raid_disks)
1124 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1125 if (rebuilding_cnt)
1126 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1127 if (sparecnt)
1128 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1129 if (content->journal_clean)
1130 fprintf(stderr, " and %d journal", journalcnt);
1131 fprintf(stderr, ".\n");
1132 }
1133 if (content->reshape_active &&
1134 content->array.level >= 4 &&
1135 content->array.level <= 6) {
1136 /* might need to increase the size
1137 * of the stripe cache - default is 256
1138 */
1139 int chunk_size = content->array.chunk_size;
1140 if (content->reshape_active &&
1141 content->new_chunk > chunk_size)
1142 chunk_size = content->new_chunk;
1143 if (256 < 4 * ((chunk_size+4065)/4096)) {
1144 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1145 if (sra)
1146 sysfs_set_num(sra, NULL,
1147 "stripe_cache_size",
1148 (4 * chunk_size / 4096) + 1);
1149 sysfs_free(sra);
1150 }
1151 }
1152 if (okcnt < (unsigned)content->array.raid_disks) {
1153 /* If any devices did not get added
1154 * because the kernel rejected them based
1155 * on event count, try adding them
1156 * again providing the action policy is
1157 * 're-add' or greater. The bitmap
1158 * might allow them to be included, or
1159 * they will become spares.
1160 */
1161 for (i = 0; i < bestcnt; i++) {
1162 int j = best[i];
1163 if (j >= 0 && !devices[j].uptodate) {
1164 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1165 continue;
1166 rv = add_disk(mdfd, st, content,
1167 &devices[j].i);
1168 if (rv == 0 && c->verbose >= 0)
1169 pr_err("%s has been re-added.\n",
1170 devices[j].devname);
1171 }
1172 }
1173 }
1174 if (content->array.level == 6 &&
1175 okcnt + 1 == (unsigned)content->array.raid_disks &&
1176 was_forced) {
1177 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1178 if (sra)
1179 sysfs_set_str(sra, NULL,
1180 "sync_action", "repair");
1181 sysfs_free(sra);
1182 }
1183 return 0;
1184 }
1185 pr_err("failed to RUN_ARRAY %s: %s\n",
1186 mddev, strerror(errno));
1187
1188 if (!enough(content->array.level, content->array.raid_disks,
1189 content->array.layout, 1, avail))
1190 pr_err("Not enough devices to start the array.\n");
1191 else if (!enough(content->array.level,
1192 content->array.raid_disks,
1193 content->array.layout, clean,
1194 avail))
1195 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1196
1197 return 1;
1198 }
1199 if (c->runstop == -1) {
1200 pr_err("%s assembled from %d drive%s",
1201 mddev, okcnt, okcnt==1?"":"s");
1202 if (okcnt != (unsigned)content->array.raid_disks)
1203 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1204 fprintf(stderr, ", but not started.\n");
1205 return 2;
1206 }
1207 if (c->verbose >= -1) {
1208 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1209 if (rebuilding_cnt)
1210 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1211 if (sparecnt)
1212 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1213 if (!enough(content->array.level, content->array.raid_disks,
1214 content->array.layout, 1, avail))
1215 fprintf(stderr, " - not enough to start the array.\n");
1216 else if (!enough(content->array.level,
1217 content->array.raid_disks,
1218 content->array.layout, clean,
1219 avail))
1220 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1221 else {
1222 if (req_cnt == (unsigned)content->array.raid_disks)
1223 fprintf(stderr, " - need all %d to start it", req_cnt);
1224 else
1225 fprintf(stderr, " - need %d to start", req_cnt);
1226 fprintf(stderr, " (use --run to insist).\n");
1227 }
1228 }
1229 return 1;
1230 }
1231
1232 int Assemble(struct supertype *st, char *mddev,
1233 struct mddev_ident *ident,
1234 struct mddev_dev *devlist,
1235 struct context *c)
1236 {
1237 /*
1238 * The task of Assemble is to find a collection of
1239 * devices that should (according to their superblocks)
1240 * form an array, and to give this collection to the MD driver.
1241 * In Linux-2.4 and later, this involves submitting a
1242 * SET_ARRAY_INFO ioctl with no arg - to prepare
1243 * the array - and then submit a number of
1244 * ADD_NEW_DISK ioctls to add disks into
1245 * the array. Finally RUN_ARRAY might
1246 * be submitted to start the array.
1247 *
1248 * Much of the work of Assemble is in finding and/or
1249 * checking the disks to make sure they look right.
1250 *
1251 * If mddev is not set, then scan must be set and we
1252 * read through the config file for dev+uuid mapping
1253 * We recurse, setting mddev, for each device that
1254 * - isn't running
1255 * - has a valid uuid (or any uuid if !uuidset)
1256 *
1257 * If mddev is set, we try to determine state of md.
1258 * check version - must be at least 0.90.0
1259 * check kernel version. must be at least 2.4.
1260 * If not, we can possibly fall back on START_ARRAY
1261 * Try to GET_ARRAY_INFO.
1262 * If possible, give up
1263 * If not, try to STOP_ARRAY just to make sure
1264 *
1265 * If !uuidset and scan, look in conf-file for uuid
1266 * If not found, give up
1267 * If !devlist and scan and uuidset, get list of devs from conf-file
1268 *
1269 * For each device:
1270 * Check superblock - discard if bad
1271 * Check uuid (set if we don't have one) - discard if no match
1272 * Check superblock similarity if we have a superblock - discard if different
1273 * Record events, devicenum
1274 * This should give us a list of devices for the array
1275 * We should collect the most recent event number
1276 *
1277 * Count disks with recent enough event count
1278 * While force && !enough disks
1279 * Choose newest rejected disks, update event count
1280 * mark clean and rewrite superblock
1281 * If recent kernel:
1282 * SET_ARRAY_INFO
1283 * foreach device with recent events : ADD_NEW_DISK
1284 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1285 * If old kernel:
1286 * Check the device numbers in superblock are right
1287 * update superblock if any changes
1288 * START_ARRAY
1289 *
1290 */
1291 int rv;
1292 int mdfd;
1293 int clean;
1294 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1295 ident->super_minor == UnSet && ident->name[0] == 0
1296 && (ident->container == NULL || ident->member == NULL));
1297 struct devs *devices;
1298 char *devmap;
1299 int *best = NULL; /* indexed by raid_disk */
1300 int bestcnt = 0;
1301 int devcnt;
1302 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1303 int journal_clean = 0;
1304 int i;
1305 int was_forced = 0;
1306 int most_recent = 0;
1307 int chosen_drive;
1308 int change = 0;
1309 int inargv = 0;
1310 int start_partial_ok = (c->runstop >= 0) &&
1311 (c->force || devlist==NULL || auto_assem);
1312 int num_devs;
1313 struct mddev_dev *tmpdev;
1314 struct mdinfo info;
1315 struct mdinfo *content = NULL;
1316 struct mdinfo *pre_exist = NULL;
1317 char *avail;
1318 char *name = NULL;
1319 char chosen_name[1024];
1320 struct map_ent *map = NULL;
1321 struct map_ent *mp;
1322
1323 /*
1324 * If any subdevs are listed, then any that don't
1325 * match ident are discarded. Remainder must all match and
1326 * become the array.
1327 * If no subdevs, then we scan all devices in the config file, but
1328 * there must be something in the identity
1329 */
1330
1331 if (!devlist &&
1332 ident->uuid_set == 0 &&
1333 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1334 ident->name[0] == 0 &&
1335 (ident->container == NULL || ident->member == NULL) &&
1336 ident->devices == NULL) {
1337 pr_err("No identity information available for %s - cannot assemble.\n",
1338 mddev ? mddev : "further assembly");
1339 return 1;
1340 }
1341
1342 if (devlist == NULL)
1343 devlist = conf_get_devs();
1344 else if (mddev)
1345 inargv = 1;
1346
1347 try_again:
1348 /* We come back here when doing auto-assembly and attempting some
1349 * set of devices failed. Those are now marked as ->used==2 and
1350 * we ignore them and try again
1351 */
1352 if (!st && ident->st)
1353 st = ident->st;
1354 if (c->verbose>0)
1355 pr_err("looking for devices for %s\n",
1356 mddev ? mddev : "further assembly");
1357
1358 content = &info;
1359 if (st && c->force)
1360 st->ignore_hw_compat = 1;
1361 num_devs = select_devices(devlist, ident, &st, &content, c,
1362 inargv, auto_assem);
1363 if (num_devs < 0)
1364 return 1;
1365
1366 if (!st || !st->sb || !content)
1367 return 2;
1368
1369 /* We have a full set of devices - we now need to find the
1370 * array device.
1371 * However there is a risk that we are racing with "mdadm -I"
1372 * and the array is already partially assembled - we will have
1373 * rejected any devices already in this address.
1374 * So we take a lock on the map file - to prevent further races -
1375 * and look for the uuid in there. If found and the array is
1376 * active, we abort. If found and the array is not active
1377 * we commit to that md device and add all the contained devices
1378 * to our list. We flag them so that we don't try to re-add,
1379 * but can remove if they turn out to not be wanted.
1380 */
1381 if (map_lock(&map))
1382 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1383 if (c->update && strcmp(c->update,"uuid") == 0)
1384 mp = NULL;
1385 else
1386 mp = map_by_uuid(&map, content->uuid);
1387 if (mp) {
1388 struct mdinfo *dv;
1389 /* array already exists. */
1390 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1391 if (pre_exist->array.level != UnSet) {
1392 pr_err("Found some drive for an array that is already active: %s\n",
1393 mp->path);
1394 pr_err("giving up.\n");
1395 return 1;
1396 }
1397 for (dv = pre_exist->devs; dv; dv = dv->next) {
1398 /* We want to add this device to our list,
1399 * but it could already be there if "mdadm -I"
1400 * started *after* we checked for O_EXCL.
1401 * If we add it to the top of the list
1402 * it will be preferred over later copies.
1403 */
1404 struct mddev_dev *newdev;
1405 char *devname = map_dev(dv->disk.major,
1406 dv->disk.minor,
1407 0);
1408 if (!devname)
1409 continue;
1410 newdev = xmalloc(sizeof(*newdev));
1411 newdev->devname = devname;
1412 newdev->disposition = 'I';
1413 newdev->used = 1;
1414 newdev->next = devlist;
1415 devlist = newdev;
1416 num_devs++;
1417 }
1418 strcpy(chosen_name, mp->path);
1419 if (c->verbose > 0 || mddev == NULL ||
1420 strcmp(mddev, chosen_name) != 0)
1421 pr_err("Merging with already-assembled %s\n",
1422 chosen_name);
1423 mdfd = open_dev_excl(mp->devnm);
1424 } else {
1425 int trustworthy = FOREIGN;
1426 name = content->name;
1427 switch (st->ss->match_home(st, c->homehost)
1428 ?: st->ss->match_home(st, "any")) {
1429 case 1:
1430 trustworthy = LOCAL;
1431 name = strchr(content->name, ':');
1432 if (name)
1433 name++;
1434 else
1435 name = content->name;
1436 break;
1437 }
1438 if (!auto_assem)
1439 /* If the array is listed in mdadm.conf or on
1440 * command line, then we trust the name
1441 * even if the array doesn't look local
1442 */
1443 trustworthy = LOCAL;
1444
1445 if (name[0] == 0 &&
1446 content->array.level == LEVEL_CONTAINER) {
1447 name = content->text_version;
1448 trustworthy = METADATA;
1449 }
1450
1451 if (name[0] && trustworthy != LOCAL &&
1452 ! c->require_homehost &&
1453 conf_name_is_free(name))
1454 trustworthy = LOCAL;
1455
1456 if (trustworthy == LOCAL &&
1457 strchr(name, ':'))
1458 /* Ignore 'host:' prefix of name */
1459 name = strchr(name, ':')+1;
1460
1461 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1462 chosen_name);
1463 }
1464 if (mdfd < 0) {
1465 st->ss->free_super(st);
1466 if (auto_assem)
1467 goto try_again;
1468 return 1;
1469 }
1470 mddev = chosen_name;
1471 if (get_linux_version() < 2004000 ||
1472 md_get_version(mdfd) < 9000) {
1473 pr_err("Assemble requires Linux 2.4 or later, and\n"
1474 " md driver version 0.90.0 or later.\n"
1475 " Upgrade your kernel or try --build\n");
1476 close(mdfd);
1477 return 1;
1478 }
1479 if (pre_exist == NULL) {
1480 if (mddev_busy(fd2devnm(mdfd))) {
1481 pr_err("%s already active, cannot restart it!\n",
1482 mddev);
1483 for (tmpdev = devlist ;
1484 tmpdev && tmpdev->used != 1;
1485 tmpdev = tmpdev->next)
1486 ;
1487 if (tmpdev && auto_assem)
1488 pr_err("%s needed for %s...\n",
1489 mddev, tmpdev->devname);
1490 close(mdfd);
1491 mdfd = -3;
1492 st->ss->free_super(st);
1493 if (auto_assem)
1494 goto try_again;
1495 return 1;
1496 }
1497 /* just incase it was started but has no content */
1498 ioctl(mdfd, STOP_ARRAY, NULL);
1499 }
1500
1501 #ifndef MDASSEMBLE
1502 if (content != &info) {
1503 /* This is a member of a container. Try starting the array. */
1504 int err;
1505 err = assemble_container_content(st, mdfd, content, c,
1506 chosen_name, NULL);
1507 close(mdfd);
1508 return err;
1509 }
1510 #endif
1511 /* Ok, no bad inconsistancy, we can try updating etc */
1512 devices = xcalloc(num_devs, sizeof(*devices));
1513 devmap = xcalloc(num_devs, content->array.raid_disks);
1514 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1515 c, content, mdfd, mddev,
1516 &most_recent, &bestcnt, &best, inargv);
1517 if (devcnt < 0)
1518 return 1;
1519
1520 if (devcnt == 0) {
1521 pr_err("no devices found for %s\n",
1522 mddev);
1523 if (st)
1524 st->ss->free_super(st);
1525 close(mdfd);
1526 free(devices);
1527 free(devmap);
1528 return 1;
1529 }
1530
1531 if (c->update && strcmp(c->update, "byteorder")==0)
1532 st->minor_version = 90;
1533
1534 st->ss->getinfo_super(st, content, NULL);
1535 clean = content->array.state & 1;
1536
1537 /* now we have some devices that might be suitable.
1538 * I wonder how many
1539 */
1540 avail = xcalloc(content->array.raid_disks, 1);
1541 okcnt = 0;
1542 replcnt = 0;
1543 sparecnt=0;
1544 journalcnt=0;
1545 rebuilding_cnt=0;
1546 for (i=0; i< bestcnt; i++) {
1547 int j = best[i];
1548 int event_margin = 1; /* always allow a difference of '1'
1549 * like the kernel does
1550 */
1551 if (j < 0) continue;
1552 /* note: we ignore error flags in multipath arrays
1553 * as they don't make sense
1554 */
1555 if (content->array.level != LEVEL_MULTIPATH) {
1556 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1557 if (content->journal_device_required)
1558 journalcnt++;
1559 else /* unexpected journal, mark as faulty */
1560 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1561 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1562 if (!(devices[j].i.disk.state
1563 & (1<<MD_DISK_FAULTY))) {
1564 devices[j].uptodate = 1;
1565 sparecnt++;
1566 }
1567 continue;
1568 }
1569 }
1570 /* If this device thinks that 'most_recent' has failed, then
1571 * we must reject this device.
1572 */
1573 if (j != most_recent && !c->force &&
1574 content->array.raid_disks > 0 &&
1575 devices[most_recent].i.disk.raid_disk >= 0 &&
1576 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1577 if (c->verbose > -1)
1578 pr_err("ignoring %s as it reports %s as failed\n",
1579 devices[j].devname, devices[most_recent].devname);
1580 best[i] = -1;
1581 continue;
1582 }
1583 /* Require event counter to be same as, or just less than,
1584 * most recent. If it is bigger, it must be a stray spare and
1585 * should be ignored.
1586 */
1587 if (devices[j].i.events+event_margin >=
1588 devices[most_recent].i.events &&
1589 devices[j].i.events <=
1590 devices[most_recent].i.events
1591 ) {
1592 devices[j].uptodate = 1;
1593 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1594 journal_clean = 1;
1595 if (i < content->array.raid_disks * 2) {
1596 if (devices[j].i.recovery_start == MaxSector ||
1597 (content->reshape_active &&
1598 i >= content->array.raid_disks - content->delta_disks)) {
1599 if (!avail[i/2]) {
1600 okcnt++;
1601 avail[i/2]=1;
1602 } else
1603 replcnt++;
1604 } else
1605 rebuilding_cnt++;
1606 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1607 sparecnt++;
1608 }
1609 }
1610 free(devmap);
1611 if (c->force) {
1612 int force_ok = force_array(content, devices, best, bestcnt,
1613 avail, most_recent, st, c);
1614 okcnt += force_ok;
1615 if (force_ok)
1616 was_forced = 1;
1617 }
1618 /* Now we want to look at the superblock which the kernel will base things on
1619 * and compare the devices that we think are working with the devices that the
1620 * superblock thinks are working.
1621 * If there are differences and --force is given, then update this chosen
1622 * superblock.
1623 */
1624 chosen_drive = -1;
1625 st->ss->free_super(st);
1626 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1627 int j = best[i];
1628 int fd;
1629
1630 if (j<0)
1631 continue;
1632 if (!devices[j].uptodate)
1633 continue;
1634 if (devices[j].i.events < devices[most_recent].i.events)
1635 continue;
1636 chosen_drive = j;
1637 if ((fd=dev_open(devices[j].devname,
1638 devices[j].included ? O_RDONLY
1639 : (O_RDONLY|O_EXCL)))< 0) {
1640 pr_err("Cannot open %s: %s\n",
1641 devices[j].devname, strerror(errno));
1642 close(mdfd);
1643 free(devices);
1644 return 1;
1645 }
1646 if (st->ss->load_super(st,fd, NULL)) {
1647 close(fd);
1648 pr_err("RAID superblock has disappeared from %s\n",
1649 devices[j].devname);
1650 close(mdfd);
1651 free(devices);
1652 return 1;
1653 }
1654 close(fd);
1655 }
1656 if (st->sb == NULL) {
1657 pr_err("No suitable drives found for %s\n", mddev);
1658 close(mdfd);
1659 free(devices);
1660 return 1;
1661 }
1662 st->ss->getinfo_super(st, content, NULL);
1663 #ifndef MDASSEMBLE
1664 sysfs_init(content, mdfd, NULL);
1665 #endif
1666 /* after reload context, store journal_clean in context */
1667 content->journal_clean = journal_clean;
1668 for (i=0; i<bestcnt; i++) {
1669 int j = best[i];
1670 unsigned int desired_state;
1671
1672 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1673 desired_state = (1<<MD_DISK_JOURNAL);
1674 else if (i >= content->array.raid_disks * 2)
1675 desired_state = 0;
1676 else if (i & 1)
1677 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1678 else
1679 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1680
1681 if (j<0)
1682 continue;
1683 if (!devices[j].uptodate)
1684 continue;
1685
1686 devices[j].i.disk.state = desired_state;
1687 if (!(devices[j].i.array.state & 1))
1688 clean = 0;
1689
1690 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1691 c->verbose, 0, NULL)) {
1692 if (c->force) {
1693 if (c->verbose >= 0)
1694 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1695 j, mddev, devices[j].devname);
1696 change = 1;
1697 } else {
1698 if (c->verbose >= -1)
1699 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1700 i, mddev, devices[j].devname);
1701 }
1702 }
1703 #if 0
1704 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1705 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1706 i, mddev);
1707 }
1708 #endif
1709 }
1710 if (c->force && !clean &&
1711 !enough(content->array.level, content->array.raid_disks,
1712 content->array.layout, clean,
1713 avail)) {
1714 change += st->ss->update_super(st, content, "force-array",
1715 devices[chosen_drive].devname, c->verbose,
1716 0, NULL);
1717 was_forced = 1;
1718 clean = 1;
1719 }
1720
1721 if (change) {
1722 int fd;
1723 fd = dev_open(devices[chosen_drive].devname,
1724 devices[chosen_drive].included ?
1725 O_RDWR : (O_RDWR|O_EXCL));
1726 if (fd < 0) {
1727 pr_err("Could not open %s for write - cannot Assemble array.\n",
1728 devices[chosen_drive].devname);
1729 close(mdfd);
1730 free(devices);
1731 return 1;
1732 }
1733 if (st->ss->store_super(st, fd)) {
1734 close(fd);
1735 pr_err("Could not re-write superblock on %s\n",
1736 devices[chosen_drive].devname);
1737 close(mdfd);
1738 free(devices);
1739 return 1;
1740 }
1741 if (c->verbose >= 0)
1742 pr_err("Marking array %s as 'clean'\n",
1743 mddev);
1744 close(fd);
1745 }
1746
1747 /* If we are in the middle of a reshape we may need to restore saved data
1748 * that was moved aside due to the reshape overwriting live data
1749 * The code of doing this lives in Grow.c
1750 */
1751 #ifndef MDASSEMBLE
1752 if (content->reshape_active &&
1753 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1754 int err = 0;
1755 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1756 if (c->verbose > 0)
1757 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1758 chosen_name);
1759 if (!c->backup_file)
1760 c->backup_file = locate_backup(content->sys_name);
1761 enable_fds(bestcnt/2);
1762 for (i = 0; i < bestcnt/2; i++) {
1763 int j = best[i*2];
1764 if (j >= 0) {
1765 fdlist[i] = dev_open(devices[j].devname,
1766 devices[j].included
1767 ? O_RDWR : (O_RDWR|O_EXCL));
1768 if (fdlist[i] < 0) {
1769 pr_err("Could not open %s for write - cannot Assemble array.\n",
1770 devices[j].devname);
1771 err = 1;
1772 break;
1773 }
1774 } else
1775 fdlist[i] = -1;
1776 }
1777 if (!err) {
1778 if (st->ss->external && st->ss->recover_backup)
1779 err = st->ss->recover_backup(st, content);
1780 else
1781 err = Grow_restart(st, content, fdlist, bestcnt/2,
1782 c->backup_file, c->verbose > 0);
1783 if (err && c->invalid_backup) {
1784 if (c->verbose > 0)
1785 pr_err("continuing without restoring backup\n");
1786 err = 0;
1787 }
1788 }
1789 while (i>0) {
1790 i--;
1791 if (fdlist[i]>=0) close(fdlist[i]);
1792 }
1793 free(fdlist);
1794 if (err) {
1795 pr_err("Failed to restore critical section for reshape, sorry.\n");
1796 if (c->backup_file == NULL)
1797 cont_err("Possibly you needed to specify the --backup-file\n");
1798 close(mdfd);
1799 free(devices);
1800 return err;
1801 }
1802 }
1803 #endif
1804
1805 /* Almost ready to actually *do* something */
1806 /* First, fill in the map, so that udev can find our name
1807 * as soon as we become active.
1808 */
1809 if (c->update && strcmp(c->update, "metadata")==0) {
1810 content->array.major_version = 1;
1811 content->array.minor_version = 0;
1812 strcpy(content->text_version, "1.0");
1813 }
1814
1815 map_update(&map, fd2devnm(mdfd), content->text_version,
1816 content->uuid, chosen_name);
1817
1818 rv = start_array(mdfd, mddev, content,
1819 st, ident, best, bestcnt,
1820 chosen_drive, devices, okcnt, sparecnt,
1821 rebuilding_cnt, journalcnt,
1822 c,
1823 clean, avail, start_partial_ok,
1824 pre_exist != NULL,
1825 was_forced);
1826 if (rv == 1 && !pre_exist)
1827 ioctl(mdfd, STOP_ARRAY, NULL);
1828 free(devices);
1829 map_unlock(&map);
1830 if (rv == 0) {
1831 wait_for(chosen_name, mdfd);
1832 close(mdfd);
1833 if (auto_assem) {
1834 int usecs = 1;
1835 /* There is a nasty race with 'mdadm --monitor'.
1836 * If it opens this device before we close it,
1837 * it gets an incomplete open on which IO
1838 * doesn't work and the capacity is
1839 * wrong.
1840 * If we reopen (to check for layered devices)
1841 * before --monitor closes, we loose.
1842 *
1843 * So: wait upto 1 second for there to be
1844 * a non-zero capacity.
1845 */
1846 while (usecs < 1000) {
1847 mdfd = open(mddev, O_RDONLY);
1848 if (mdfd >= 0) {
1849 unsigned long long size;
1850 if (get_dev_size(mdfd, NULL, &size) &&
1851 size > 0)
1852 break;
1853 close(mdfd);
1854 }
1855 usleep(usecs);
1856 usecs <<= 1;
1857 }
1858 }
1859 } else
1860 close(mdfd);
1861
1862 /* '2' means 'OK, but not started yet' */
1863 return rv == 2 ? 0 : rv;
1864 }
1865
1866 #ifndef MDASSEMBLE
1867 int assemble_container_content(struct supertype *st, int mdfd,
1868 struct mdinfo *content, struct context *c,
1869 char *chosen_name, int *result)
1870 {
1871 struct mdinfo *dev, *sra, *dev2;
1872 int working = 0, preexist = 0;
1873 int expansion = 0;
1874 int old_raid_disks;
1875 int start_reshape;
1876 char *avail;
1877 int err;
1878
1879 sysfs_init(content, mdfd, NULL);
1880
1881 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1882 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1883 if (content->array.major_version == -1 &&
1884 content->array.minor_version == -2 &&
1885 c->readonly &&
1886 content->text_version[0] == '/')
1887 content->text_version[0] = '-';
1888 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1889 sysfs_free(sra);
1890 return 1;
1891 }
1892 }
1893
1894 /* There are two types of reshape: container wide or sub-array specific
1895 * Check if metadata requests blocking container wide reshapes
1896 */
1897 start_reshape = (content->reshape_active &&
1898 !((content->reshape_active == CONTAINER_RESHAPE) &&
1899 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1900
1901 /* Block subarray here if it is under reshape now
1902 * Do not allow for any changes in this array
1903 */
1904 if (st->ss->external && content->recovery_blocked && start_reshape)
1905 block_subarray(content);
1906
1907 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1908 for (dev = content->devs; dev; dev = dev->next)
1909 if (dev2->disk.major == dev->disk.major &&
1910 dev2->disk.minor == dev->disk.minor)
1911 break;
1912 if (dev)
1913 continue;
1914 /* Don't want this one any more */
1915 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1916 errno == EBUSY) {
1917 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1918 sysfs_free(sra);
1919 return 1;
1920 }
1921 sysfs_set_str(sra, dev2, "state", "remove");
1922 }
1923 old_raid_disks = content->array.raid_disks - content->delta_disks;
1924 avail = xcalloc(content->array.raid_disks, 1);
1925 for (dev = content->devs; dev; dev = dev->next) {
1926 if (dev->disk.raid_disk >= 0)
1927 avail[dev->disk.raid_disk] = 1;
1928 if (sysfs_add_disk(content, dev, 1) == 0) {
1929 if (dev->disk.raid_disk >= old_raid_disks &&
1930 content->reshape_active)
1931 expansion++;
1932 else
1933 working++;
1934 } else if (errno == EEXIST)
1935 preexist++;
1936 }
1937 sysfs_free(sra);
1938 if (working + expansion == 0 && c->runstop <= 0) {
1939 free(avail);
1940 return 1;/* Nothing new, don't try to start */
1941 }
1942 map_update(NULL, fd2devnm(mdfd), content->text_version,
1943 content->uuid, chosen_name);
1944
1945 if (enough(content->array.level, content->array.raid_disks,
1946 content->array.layout, content->array.state & 1, avail) == 0) {
1947 if (c->export && result)
1948 *result |= INCR_NO;
1949 else if (c->verbose >= 0) {
1950 pr_err("%s assembled with %d device%s",
1951 chosen_name, preexist + working,
1952 preexist + working == 1 ? "":"s");
1953 if (preexist)
1954 fprintf(stderr, " (%d new)", working);
1955 fprintf(stderr, " but not started\n");
1956 }
1957 free(avail);
1958 return 1;
1959 }
1960 free(avail);
1961
1962 if (c->runstop <= 0 &&
1963 (working + preexist + expansion) <
1964 content->array.working_disks) {
1965 if (c->export && result)
1966 *result |= INCR_UNSAFE;
1967 else if (c->verbose >= 0) {
1968 pr_err("%s assembled with %d device%s",
1969 chosen_name, preexist + working,
1970 preexist + working == 1 ? "":"s");
1971 if (preexist)
1972 fprintf(stderr, " (%d new)", working);
1973 fprintf(stderr, " but not safe to start\n");
1974 }
1975 return 1;
1976 }
1977
1978
1979 if (start_reshape) {
1980 int spare = content->array.raid_disks + expansion;
1981 if (restore_backup(st, content,
1982 working,
1983 spare, &c->backup_file, c->verbose) == 1)
1984 return 1;
1985
1986 err = sysfs_set_str(content, NULL,
1987 "array_state", "readonly");
1988 if (err)
1989 return 1;
1990
1991 if (st->ss->external) {
1992 if (!mdmon_running(st->container_devnm))
1993 start_mdmon(st->container_devnm);
1994 ping_monitor(st->container_devnm);
1995 if (mdmon_running(st->container_devnm) &&
1996 st->update_tail == NULL)
1997 st->update_tail = &st->updates;
1998 }
1999
2000 err = Grow_continue(mdfd, st, content, c->backup_file,
2001 0, c->freeze_reshape);
2002 } else switch(content->array.level) {
2003 case LEVEL_LINEAR:
2004 case LEVEL_MULTIPATH:
2005 case 0:
2006 err = sysfs_set_str(content, NULL, "array_state",
2007 c->readonly ? "readonly" : "active");
2008 break;
2009 default:
2010 err = sysfs_set_str(content, NULL, "array_state",
2011 "readonly");
2012 /* start mdmon if needed. */
2013 if (!err) {
2014 if (!mdmon_running(st->container_devnm))
2015 start_mdmon(st->container_devnm);
2016 ping_monitor(st->container_devnm);
2017 }
2018 break;
2019 }
2020 if (!err)
2021 sysfs_set_safemode(content, content->safe_mode_delay);
2022
2023 /* Block subarray here if it is not reshaped now
2024 * It has be blocked a little later to allow mdmon to switch in
2025 * in to R/W state
2026 */
2027 if (st->ss->external && content->recovery_blocked &&
2028 !start_reshape)
2029 block_subarray(content);
2030
2031 if (c->export && result) {
2032 if (err)
2033 *result |= INCR_NO;
2034 else
2035 *result |= INCR_YES;
2036 } else if (c->verbose >= 0) {
2037 if (err)
2038 pr_err("array %s now has %d device%s",
2039 chosen_name, working + preexist,
2040 working + preexist == 1 ? "":"s");
2041 else
2042 pr_err("Started %s with %d device%s",
2043 chosen_name, working + preexist,
2044 working + preexist == 1 ? "":"s");
2045 if (preexist)
2046 fprintf(stderr, " (%d new)", working);
2047 if (expansion)
2048 fprintf(stderr, " ( + %d for expansion)",
2049 expansion);
2050 fprintf(stderr, "\n");
2051 }
2052 if (!err)
2053 wait_for(chosen_name, mdfd);
2054 return err;
2055 /* FIXME should have an O_EXCL and wait for read-auto */
2056 }
2057 #endif