]> 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
--update: add 'bbl' and 'no-bbl' to the list of known updates.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
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)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(0, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 static int ident_matches(struct mddev_ident *ident,
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78 {
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
82 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
83 if (devname)
84 pr_err("%s has wrong uuid.\n", devname);
85 return 0;
86 }
87 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
88 name_matches(content->name, ident->name, homehost)==0) {
89 if (devname)
90 pr_err("%s has wrong name.\n", devname);
91 return 0;
92 }
93 if (ident->super_minor != UnSet &&
94 ident->super_minor != content->array.md_minor) {
95 if (devname)
96 pr_err("%s has wrong super-minor.\n",
97 devname);
98 return 0;
99 }
100 if (ident->level != UnSet &&
101 ident->level != content->array.level) {
102 if (devname)
103 pr_err("%s has wrong raid level.\n",
104 devname);
105 return 0;
106 }
107 if (ident->raid_disks != UnSet &&
108 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
109 ident->raid_disks!= content->array.raid_disks) {
110 if (devname)
111 pr_err("%s requires wrong number of drives.\n",
112 devname);
113 return 0;
114 }
115 if (ident->member && ident->member[0]) {
116 /* content->text_version must match */
117 char *s = strchr(content->text_version+1, '/');
118 if (s == NULL) {
119 if (devname)
120 pr_err("%s is not a container and one is required.\n",
121 devname);
122 return 0;
123 } else if (strcmp(ident->member, s+1) != 0) {
124 if (devname)
125 pr_err("skipping wrong member %s is %s\n",
126 content->text_version, devname);
127 return 0;
128 }
129 }
130 return 1;
131 }
132
133 static int select_devices(struct mddev_dev *devlist,
134 struct mddev_ident *ident,
135 struct supertype **stp,
136 struct mdinfo **contentp,
137 struct context *c,
138 int inargv, int auto_assem)
139 {
140 struct mddev_dev *tmpdev;
141 int num_devs;
142 struct supertype *st = *stp;
143 struct mdinfo *content = NULL;
144 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
145 struct domainlist *domains = NULL;
146
147 tmpdev = devlist; num_devs = 0;
148 while (tmpdev) {
149 if (tmpdev->used)
150 tmpdev->used = 2;
151 else
152 num_devs++;
153 tmpdev->disposition = 0;
154 tmpdev = tmpdev->next;
155 }
156
157 /* first walk the list of devices to find a consistent set
158 * that match the criterea, if that is possible.
159 * We flag the ones we like with 'used'.
160 */
161 for (tmpdev = devlist;
162 tmpdev;
163 tmpdev = tmpdev ? tmpdev->next : NULL) {
164 char *devname = tmpdev->devname;
165 int dfd;
166 struct stat stb;
167 struct supertype *tst;
168 struct dev_policy *pol = NULL;
169 int found_container = 0;
170
171 if (tmpdev->used > 1)
172 continue;
173
174 if (ident->container) {
175 if (ident->container[0] == '/' &&
176 !same_dev(ident->container, devname)) {
177 if (report_mismatch)
178 pr_err("%s is not the container required (%s)\n",
179 devname, ident->container);
180 continue;
181 }
182 } else if (ident->devices &&
183 !match_oneof(ident->devices, devname)) {
184 /* Note that we ignore the "device=" identifier if a
185 * "container=" is given. Checking both is unnecessarily
186 * complicated.
187 */
188 if (report_mismatch)
189 pr_err("%s is not one of %s\n", devname, ident->devices);
190 continue;
191 }
192
193 tst = dup_super(st);
194
195 dfd = dev_open(devname, O_RDONLY);
196 if (dfd < 0) {
197 if (report_mismatch)
198 pr_err("cannot open device %s: %s\n",
199 devname, strerror(errno));
200 tmpdev->used = 2;
201 } else if (fstat(dfd, &stb)< 0) {
202 /* Impossible! */
203 pr_err("fstat failed for %s: %s\n",
204 devname, strerror(errno));
205 tmpdev->used = 2;
206 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
207 pr_err("%s is not a block device.\n",
208 devname);
209 tmpdev->used = 2;
210 } else if (must_be_container(dfd)) {
211 if (st) {
212 /* already found some components, this cannot
213 * be another one.
214 */
215 if (report_mismatch)
216 pr_err("%s is a container, but we are looking for components\n",
217 devname);
218 tmpdev->used = 2;
219 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
220 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
221 if (report_mismatch)
222 pr_err("not a recognisable container: %s\n",
223 devname);
224 tmpdev->used = 2;
225 #endif
226 } else if (!tst->ss->load_container
227 || tst->ss->load_container(tst, dfd, NULL)) {
228 if (report_mismatch)
229 pr_err("no correct container type: %s\n",
230 devname);
231 tmpdev->used = 2;
232 } else if (auto_assem &&
233 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
234 tst->ss->match_home(tst, c->homehost) == 1)) {
235 if (report_mismatch)
236 pr_err("%s has metadata type %s for which "
237 "auto-assembly is disabled\n",
238 devname, tst->ss->name);
239 tmpdev->used = 2;
240 } else
241 found_container = 1;
242 } else {
243 if (!tst && (tst = guess_super(dfd)) == NULL) {
244 if (report_mismatch)
245 pr_err("no recogniseable superblock on %s\n",
246 devname);
247 tmpdev->used = 2;
248 } else if (tst->ss->load_super(tst,dfd, NULL)) {
249 if (report_mismatch)
250 pr_err("no RAID superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if (tst->ss->compare_super == NULL) {
254 if (report_mismatch)
255 pr_err("Cannot assemble %s metadata on %s\n",
256 tst->ss->name, devname);
257 tmpdev->used = 2;
258 } else if (auto_assem && st == NULL &&
259 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
260 tst->ss->match_home(tst, c->homehost) == 1)) {
261 if (report_mismatch)
262 pr_err("%s has metadata type %s for which "
263 "auto-assembly is disabled\n",
264 devname, tst->ss->name);
265 tmpdev->used = 2;
266 }
267 }
268 if (dfd >= 0) close(dfd);
269 if (tmpdev->used == 2) {
270 if (auto_assem || !inargv)
271 /* Ignore unrecognised devices during auto-assembly */
272 goto loop;
273 if (ident->uuid_set || ident->name[0] ||
274 ident->super_minor != UnSet)
275 /* Ignore unrecognised device if looking for
276 * specific array */
277 goto loop;
278
279 pr_err("%s has no superblock - assembly aborted\n",
280 devname);
281 if (st)
282 st->ss->free_super(st);
283 dev_policy_free(pol);
284 domain_free(domains);
285 return -1;
286 }
287
288 if (found_container) {
289 /* tmpdev is a container. We need to be either
290 * looking for a member, or auto-assembling
291 */
292 /* should be safe to try an exclusive open now, we
293 * have rejected anything that some other mdadm might
294 * be looking at
295 */
296 dfd = dev_open(devname, O_RDONLY | O_EXCL);
297 if (dfd < 0) {
298 if (report_mismatch)
299 pr_err("%s is busy - skipping\n", devname);
300 goto loop;
301 }
302 close(dfd);
303
304 if (ident->container && ident->container[0] != '/') {
305 /* we have a uuid */
306 int uuid[4];
307
308 content = *contentp;
309 tst->ss->getinfo_super(tst, content, NULL);
310
311 if (!parse_uuid(ident->container, uuid) ||
312 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
313 if (report_mismatch)
314 pr_err("%s has wrong UUID to be required container\n",
315 devname);
316 goto loop;
317 }
318 }
319 /* It is worth looking inside this container.
320 */
321 if (c->verbose > 0)
322 pr_err("looking in container %s\n",
323 devname);
324
325 for (content = tst->ss->container_content(tst, NULL);
326 content;
327 content = content->next) {
328
329 if (!ident_matches(ident, content, tst,
330 c->homehost, 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 "
354 "only device given: confused and aborting\n",
355 devname);
356 st->ss->free_super(st);
357 dev_policy_free(pol);
358 domain_free(domains);
359 return -1;
360 }
361 if (c->verbose > 0)
362 pr_err("found match on member %s in %s\n",
363 content->text_version, devname);
364
365 /* make sure we finished the loop */
366 tmpdev = NULL;
367 goto loop;
368 } else {
369 content = *contentp;
370 tst->ss->getinfo_super(tst, content, NULL);
371
372 if (!ident_matches(ident, content, tst,
373 c->homehost, 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->update,
398 report_mismatch ? devname : NULL))
399 /* Array exists in mdadm.conf but some
400 * details don't match, so reject it
401 */
402 goto loop;
403 }
404
405 /* should be safe to try an exclusive open now, we
406 * have rejected anything that some other mdadm might
407 * be looking at
408 */
409 dfd = dev_open(devname, O_RDONLY | O_EXCL);
410 if (dfd < 0) {
411 if (report_mismatch)
412 pr_err("%s is busy - skipping\n", devname);
413 goto loop;
414 }
415 close(dfd);
416
417 if (st == NULL)
418 st = dup_super(tst);
419 if (st->minor_version == -1)
420 st->minor_version = tst->minor_version;
421
422 if (memcmp(content->uuid, uuid_zero,
423 sizeof(int[4])) == 0) {
424 /* this is a floating spare. It cannot define
425 * an array unless there are no more arrays of
426 * this type to be found. It can be included
427 * in an array of this type though.
428 */
429 tmpdev->used = 3;
430 goto loop;
431 }
432
433 if (st->ss != tst->ss ||
434 st->minor_version != tst->minor_version ||
435 st->ss->compare_super(st, tst) != 0) {
436 /* Some mismatch. If exactly one array matches this host,
437 * we can resolve on that one.
438 * Or, if we are auto assembling, we just ignore the second
439 * for now.
440 */
441 if (auto_assem)
442 goto loop;
443 if (c->homehost) {
444 int first = st->ss->match_home(st, c->homehost);
445 int last = tst->ss->match_home(tst, c->homehost);
446 if (first != last &&
447 (first == 1 || last == 1)) {
448 /* We can do something */
449 if (first) {/* just ignore this one */
450 if (report_mismatch)
451 pr_err("%s misses out due to wrong homehost\n",
452 devname);
453 goto loop;
454 } else { /* reject all those sofar */
455 struct mddev_dev *td;
456 if (report_mismatch)
457 pr_err("%s overrides previous devices due to good homehost\n",
458 devname);
459 for (td=devlist; td != tmpdev; td=td->next)
460 if (td->used == 1)
461 td->used = 0;
462 tmpdev->used = 1;
463 goto loop;
464 }
465 }
466 }
467 pr_err("superblock on %s doesn't match others - assembly aborted\n",
468 devname);
469 tst->ss->free_super(tst);
470 st->ss->free_super(st);
471 dev_policy_free(pol);
472 domain_free(domains);
473 return -1;
474 }
475 tmpdev->used = 1;
476 }
477 loop:
478 /* Collect domain information from members only */
479 if (tmpdev && tmpdev->used == 1) {
480 if (!pol)
481 pol = devid_policy(stb.st_rdev);
482 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
483 }
484 dev_policy_free(pol);
485 pol = NULL;
486 if (tst)
487 tst->ss->free_super(tst);
488 }
489
490 /* Check if we found some imsm spares but no members */
491 if ((auto_assem ||
492 (ident->uuid_set &&
493 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
494 (!st || !st->sb))
495 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
496 if (tmpdev->used != 3)
497 continue;
498 tmpdev->used = 1;
499 content = *contentp;
500
501 if (!st->sb) {
502 /* we need sb from one of the spares */
503 int dfd = dev_open(tmpdev->devname, O_RDONLY);
504 if (dfd < 0 ||
505 st->ss->load_super(st, dfd, NULL))
506 tmpdev->used = 2;
507 if (dfd > 0)
508 close(dfd);
509 }
510 }
511
512 /* Now reject spares that don't match domains of identified members */
513 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
514 struct stat stb;
515 if (tmpdev->used != 3)
516 continue;
517 if (stat(tmpdev->devname, &stb)< 0) {
518 pr_err("fstat failed for %s: %s\n",
519 tmpdev->devname, strerror(errno));
520 tmpdev->used = 2;
521 } else {
522 struct dev_policy *pol = devid_policy(stb.st_rdev);
523 int dt = domain_test(domains, pol, NULL);
524 if (inargv && dt != 0)
525 /* take this spare as domains match
526 * if there are any */
527 tmpdev->used = 1;
528 else if (!inargv && dt == 1)
529 /* device wasn't explicitly listed, so need
530 * explicit domain match - which we have */
531 tmpdev->used = 1;
532 else
533 /* if domains don't match mark as unused */
534 tmpdev->used = 0;
535 dev_policy_free(pol);
536 }
537 }
538 domain_free(domains);
539 *stp = st;
540 if (st && st->sb && content == *contentp)
541 st->ss->getinfo_super(st, content, NULL);
542 *contentp = content;
543
544 return num_devs;
545 }
546
547 struct devs {
548 char *devname;
549 int uptodate; /* set once we decide that this device is as
550 * recent as everything else in the array.
551 */
552 int included; /* set if the device is already in the array
553 * due to a previous '-I'
554 */
555 struct mdinfo i;
556 };
557
558 static int load_devices(struct devs *devices, char *devmap,
559 struct mddev_ident *ident, struct supertype **stp,
560 struct mddev_dev *devlist, struct context *c,
561 struct mdinfo *content,
562 int mdfd, char *mddev,
563 int *most_recentp, int *bestcntp, int **bestp,
564 int inargv)
565 {
566 struct mddev_dev *tmpdev;
567 int devcnt = 0;
568 int nextspare = 0;
569 #ifndef MDASSEMBLE
570 int bitmap_done = 0;
571 #endif
572 int most_recent = -1;
573 int bestcnt = 0;
574 int *best = *bestp;
575 struct supertype *st = *stp;
576
577 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
578 char *devname = tmpdev->devname;
579 struct stat stb;
580 struct supertype *tst;
581 int i;
582
583 if (tmpdev->used != 1)
584 continue;
585 /* looks like a good enough match to update the super block if needed */
586 #ifndef MDASSEMBLE
587 if (c->update) {
588 int dfd;
589 /* prepare useful information in info structures */
590 struct stat stb2;
591 int err;
592 fstat(mdfd, &stb2);
593
594 if (strcmp(c->update, "uuid")==0 &&
595 !ident->uuid_set) {
596 int rfd;
597 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
598 read(rfd, ident->uuid, 16) != 16) {
599 *(__u32*)(ident->uuid) = random();
600 *(__u32*)(ident->uuid+1) = random();
601 *(__u32*)(ident->uuid+2) = random();
602 *(__u32*)(ident->uuid+3) = random();
603 }
604 if (rfd >= 0) close(rfd);
605 }
606 dfd = dev_open(devname,
607 tmpdev->disposition == 'I'
608 ? O_RDWR : (O_RDWR|O_EXCL));
609
610 tst = dup_super(st);
611 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
612 pr_err("cannot re-read metadata from %s - aborting\n",
613 devname);
614 if (dfd >= 0)
615 close(dfd);
616 close(mdfd);
617 free(devices);
618 free(devmap);
619 tst->ss->free_super(tst);
620 free(tst);
621 *stp = st;
622 return -1;
623 }
624 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
625
626 memcpy(content->uuid, ident->uuid, 16);
627 strcpy(content->name, ident->name);
628 content->array.md_minor = minor(stb2.st_rdev);
629
630 if (strcmp(c->update, "byteorder") == 0)
631 err = 0;
632 else
633 err = tst->ss->update_super(tst, content, c->update,
634 devname, c->verbose,
635 ident->uuid_set,
636 c->homehost);
637 if (err < 0) {
638 if (err == -1)
639 pr_err("--update=%s not understood"
640 " for %s metadata\n",
641 c->update, tst->ss->name);
642 tst->ss->free_super(tst);
643 free(tst);
644 close(mdfd);
645 close(dfd);
646 free(devices);
647 free(devmap);
648 *stp = st;
649 return -1;
650 }
651 if (strcmp(c->update, "uuid")==0 &&
652 !ident->uuid_set) {
653 ident->uuid_set = 1;
654 memcpy(ident->uuid, content->uuid, 16);
655 }
656 if (tst->ss->store_super(tst, dfd))
657 pr_err("Could not re-write superblock on %s.\n",
658 devname);
659 close(dfd);
660
661 if (strcmp(c->update, "uuid")==0 &&
662 ident->bitmap_fd >= 0 && !bitmap_done) {
663 if (bitmap_update_uuid(ident->bitmap_fd,
664 content->uuid,
665 tst->ss->swapuuid) != 0)
666 pr_err("Could not update uuid on external bitmap.\n");
667 else
668 bitmap_done = 1;
669 }
670 } else
671 #endif
672 {
673 int dfd = dev_open(devname,
674 tmpdev->disposition == 'I'
675 ? O_RDWR : (O_RDWR|O_EXCL));
676 tst = dup_super(st);
677
678 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
679 pr_err("cannot re-read metadata from %s - aborting\n",
680 devname);
681 if (dfd >= 0)
682 close(dfd);
683 close(mdfd);
684 free(devices);
685 free(devmap);
686 tst->ss->free_super(tst);
687 free(tst);
688 *stp = st;
689 return -1;
690 }
691 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
692 close(dfd);
693 }
694
695 stat(devname, &stb);
696
697 if (c->verbose > 0)
698 pr_err("%s is identified as a member of %s, slot %d%s.\n",
699 devname, mddev, content->disk.raid_disk,
700 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
701 devices[devcnt].devname = devname;
702 devices[devcnt].uptodate = 0;
703 devices[devcnt].included = (tmpdev->disposition == 'I');
704 devices[devcnt].i = *content;
705 devices[devcnt].i.disk.major = major(stb.st_rdev);
706 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
707
708 if (devices[devcnt].i.disk.state == 6) {
709 if (most_recent < 0 ||
710 devices[devcnt].i.events
711 > devices[most_recent].i.events) {
712 struct supertype *tmp = tst;
713 tst = st;
714 st = tmp;
715 most_recent = devcnt;
716 }
717 }
718 tst->ss->free_super(tst);
719 free(tst);
720
721 if (content->array.level == LEVEL_MULTIPATH)
722 /* with multipath, the raid_disk from the superblock is meaningless */
723 i = devcnt;
724 else
725 i = devices[devcnt].i.disk.raid_disk;
726 if (i+1 == 0) {
727 if (nextspare < content->array.raid_disks*2)
728 nextspare = content->array.raid_disks*2;
729 i = nextspare++;
730 } else {
731 /* i is raid_disk - double it so there is room for
732 * replacements */
733 i *= 2;
734 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
735 i++;
736 if (i >= content->array.raid_disks*2 &&
737 i >= nextspare)
738 nextspare = i+1;
739 }
740 if (i < 10000) {
741 if (i >= bestcnt) {
742 int newbestcnt = i+10;
743 int *newbest = xmalloc(sizeof(int)*newbestcnt);
744 int c;
745 for (c=0; c < newbestcnt; c++)
746 if (c < bestcnt)
747 newbest[c] = best[c];
748 else
749 newbest[c] = -1;
750 if (best)free(best);
751 best = newbest;
752 bestcnt = newbestcnt;
753 }
754 if (best[i] >=0 &&
755 devices[best[i]].i.events
756 == devices[devcnt].i.events
757 && (devices[best[i]].i.disk.minor
758 != devices[devcnt].i.disk.minor)
759 && st->ss == &super0
760 && content->array.level != LEVEL_MULTIPATH) {
761 /* two different devices with identical superblock.
762 * Could be a mis-detection caused by overlapping
763 * partitions. fail-safe.
764 */
765 pr_err("WARNING %s and %s appear"
766 " to have very similar superblocks.\n"
767 " If they are really different, "
768 "please --zero the superblock on one\n"
769 " If they are the same or overlap,"
770 " please remove one from %s.\n",
771 devices[best[i]].devname, devname,
772 inargv ? "the list" :
773 "the\n DEVICE list in mdadm.conf"
774 );
775 close(mdfd);
776 free(devices);
777 free(devmap);
778 *stp = st;
779 return -1;
780 }
781 if (best[i] == -1
782 || (devices[best[i]].i.events
783 < devices[devcnt].i.events))
784 best[i] = devcnt;
785 }
786 devcnt++;
787 }
788 if (most_recent >= 0)
789 *most_recentp = most_recent;
790 *bestcntp = bestcnt;
791 *bestp = best;
792 *stp = st;
793 return devcnt;
794 }
795
796 static int force_array(struct mdinfo *content,
797 struct devs *devices,
798 int *best, int bestcnt, char *avail,
799 int most_recent,
800 struct supertype *st,
801 struct context *c)
802 {
803 int okcnt = 0;
804 while (!enough(content->array.level, content->array.raid_disks,
805 content->array.layout, 1,
806 avail)
807 ||
808 (content->reshape_active && content->delta_disks > 0 &&
809 !enough(content->array.level, (content->array.raid_disks
810 - content->delta_disks),
811 content->new_layout, 1,
812 avail)
813 )) {
814 /* Choose the newest best drive which is
815 * not up-to-date, update the superblock
816 * and add it.
817 */
818 int fd;
819 struct supertype *tst;
820 unsigned long long current_events;
821 int chosen_drive = -1;
822 int i;
823
824 for (i = 0;
825 i < content->array.raid_disks * 2 && i < bestcnt;
826 i += 2) {
827 int j = best[i];
828 if (j>=0 &&
829 !devices[j].uptodate &&
830 devices[j].i.recovery_start == MaxSector &&
831 (chosen_drive < 0 ||
832 devices[j].i.events
833 > devices[chosen_drive].i.events))
834 chosen_drive = j;
835 }
836 if (chosen_drive < 0)
837 break;
838 current_events = devices[chosen_drive].i.events;
839 add_another:
840 if (c->verbose >= 0)
841 pr_err("forcing event count in %s(%d) from %d upto %d\n",
842 devices[chosen_drive].devname,
843 devices[chosen_drive].i.disk.raid_disk,
844 (int)(devices[chosen_drive].i.events),
845 (int)(devices[most_recent].i.events));
846 fd = dev_open(devices[chosen_drive].devname,
847 devices[chosen_drive].included ? O_RDWR
848 : (O_RDWR|O_EXCL));
849 if (fd < 0) {
850 pr_err("Couldn't open %s for write - not updating\n",
851 devices[chosen_drive].devname);
852 devices[chosen_drive].i.events = 0;
853 continue;
854 }
855 tst = dup_super(st);
856 if (tst->ss->load_super(tst,fd, NULL)) {
857 close(fd);
858 pr_err("RAID superblock disappeared from %s - not updating.\n",
859 devices[chosen_drive].devname);
860 devices[chosen_drive].i.events = 0;
861 continue;
862 }
863 content->events = devices[most_recent].i.events;
864 tst->ss->update_super(tst, content, "force-one",
865 devices[chosen_drive].devname, c->verbose,
866 0, NULL);
867
868 if (tst->ss->store_super(tst, fd)) {
869 close(fd);
870 pr_err("Could not re-write superblock on %s\n",
871 devices[chosen_drive].devname);
872 devices[chosen_drive].i.events = 0;
873 tst->ss->free_super(tst);
874 continue;
875 }
876 close(fd);
877 devices[chosen_drive].i.events = devices[most_recent].i.events;
878 devices[chosen_drive].uptodate = 1;
879 avail[chosen_drive] = 1;
880 okcnt++;
881 tst->ss->free_super(tst);
882
883 /* If there are any other drives of the same vintage,
884 * add them in as well. We can't lose and we might gain
885 */
886 for (i = 0;
887 i < content->array.raid_disks * 2 && i < bestcnt ;
888 i += 2) {
889 int j = best[i];
890 if (j >= 0 &&
891 !devices[j].uptodate &&
892 devices[j].i.recovery_start == MaxSector &&
893 devices[j].i.events == current_events) {
894 chosen_drive = j;
895 goto add_another;
896 }
897 }
898 }
899 return okcnt;
900 }
901
902 static int start_array(int mdfd,
903 char *mddev,
904 struct mdinfo *content,
905 struct supertype *st,
906 struct mddev_ident *ident,
907 int *best, int bestcnt,
908 int chosen_drive,
909 struct devs *devices,
910 unsigned int okcnt,
911 unsigned int sparecnt,
912 unsigned int rebuilding_cnt,
913 struct context *c,
914 int clean, char *avail,
915 int start_partial_ok,
916 int err_ok,
917 int was_forced
918 )
919 {
920 int rv;
921 int i;
922 unsigned int req_cnt;
923
924 rv = set_array_info(mdfd, st, content);
925 if (rv && !err_ok) {
926 pr_err("failed to set array info for %s: %s\n",
927 mddev, strerror(errno));
928 return 1;
929 }
930 if (ident->bitmap_fd >= 0) {
931 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
932 pr_err("SET_BITMAP_FILE failed.\n");
933 return 1;
934 }
935 } else if (ident->bitmap_file) {
936 /* From config file */
937 int bmfd = open(ident->bitmap_file, O_RDWR);
938 if (bmfd < 0) {
939 pr_err("Could not open bitmap file %s\n",
940 ident->bitmap_file);
941 return 1;
942 }
943 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
944 pr_err("Failed to set bitmapfile for %s\n", mddev);
945 close(bmfd);
946 return 1;
947 }
948 close(bmfd);
949 }
950
951 /* First, add the raid disks, but add the chosen one last */
952 for (i=0; i<= bestcnt; i++) {
953 int j;
954 if (i < bestcnt) {
955 j = best[i];
956 if (j == chosen_drive)
957 continue;
958 } else
959 j = chosen_drive;
960
961 if (j >= 0 && !devices[j].included) {
962 int dfd = dev_open(devices[j].devname,
963 O_RDWR|O_EXCL);
964 if (dfd >= 0) {
965 remove_partitions(dfd);
966 close(dfd);
967 }
968 rv = add_disk(mdfd, st, content, &devices[j].i);
969
970 if (rv) {
971 pr_err("failed to add "
972 "%s to %s: %s\n",
973 devices[j].devname,
974 mddev,
975 strerror(errno));
976 if (i < content->array.raid_disks * 2
977 || i == bestcnt)
978 okcnt--;
979 else
980 sparecnt--;
981 } else if (c->verbose > 0)
982 pr_err("added %s to %s as %d%s%s\n",
983 devices[j].devname, mddev,
984 devices[j].i.disk.raid_disk,
985 devices[j].uptodate?"":
986 " (possibly out of date)",
987 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
988 } else if (j >= 0) {
989 if (c->verbose > 0)
990 pr_err("%s is already in %s as %d\n",
991 devices[j].devname, mddev,
992 devices[j].i.disk.raid_disk);
993 } else if (c->verbose > 0 && i < content->array.raid_disks*2
994 && (i&1) == 0)
995 pr_err("no uptodate device for slot %d of %s\n",
996 i, mddev);
997 }
998
999 if (content->array.level == LEVEL_CONTAINER) {
1000 if (c->verbose >= 0) {
1001 pr_err("Container %s has been "
1002 "assembled with %d drive%s",
1003 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1004 if (okcnt < (unsigned)content->array.raid_disks)
1005 fprintf(stderr, " (out of %d)",
1006 content->array.raid_disks);
1007 fprintf(stderr, "\n");
1008 }
1009
1010 if (st->ss->validate_container) {
1011 struct mdinfo *devices_list;
1012 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1013 unsigned int count;
1014 devices_list = NULL;
1015 for (count = 0; count < okcnt+sparecnt; count++) {
1016 info_devices[count] = devices[count].i;
1017 info_devices[count].next = devices_list;
1018 devices_list = &info_devices[count];
1019 }
1020 if (st->ss->validate_container(devices_list))
1021 pr_err("Mismatch detected!\n");
1022 free(info_devices);
1023 }
1024
1025 st->ss->free_super(st);
1026 sysfs_uevent(content, "change");
1027 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1028 /* Was partial, is still partial, so signal an error
1029 * to ensure we don't retry */
1030 return 1;
1031 return 0;
1032 }
1033
1034 /* Get number of in-sync devices according to the superblock.
1035 * We must have this number to start the array without -s or -R
1036 */
1037 req_cnt = content->array.working_disks;
1038
1039 if (c->runstop == 1 ||
1040 (c->runstop <= 0 &&
1041 ( enough(content->array.level, content->array.raid_disks,
1042 content->array.layout, clean, avail) &&
1043 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1044 ))) {
1045 /* This array is good-to-go.
1046 * If a reshape is in progress then we might need to
1047 * continue monitoring it. In that case we start
1048 * it read-only and let the grow code make it writable.
1049 */
1050 int rv;
1051 #ifndef MDASSEMBLE
1052 if (content->reshape_active &&
1053 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1054 content->delta_disks <= 0) {
1055 if (!c->backup_file) {
1056 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1057 mddev);
1058 pr_err("Please provided one with \"--backup-file=...\"\n");
1059 if (c->update &&
1060 strcmp(c->update, "revert-reshape") == 0)
1061 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1062 return 1;
1063 }
1064 rv = sysfs_set_str(content, NULL,
1065 "array_state", "readonly");
1066 if (rv == 0)
1067 rv = Grow_continue(mdfd, st, content,
1068 c->backup_file, 0,
1069 c->freeze_reshape);
1070 } else if (c->readonly &&
1071 sysfs_attribute_available(
1072 content, NULL, "array_state")) {
1073 rv = sysfs_set_str(content, NULL,
1074 "array_state", "readonly");
1075 } else
1076 #endif
1077 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1078 reopen_mddev(mdfd); /* drop O_EXCL */
1079 if (rv == 0) {
1080 if (c->verbose >= 0) {
1081 pr_err("%s has been started with %d drive%s",
1082 mddev, okcnt, okcnt==1?"":"s");
1083 if (okcnt < (unsigned)content->array.raid_disks)
1084 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1085 if (rebuilding_cnt)
1086 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1087 if (sparecnt)
1088 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1089 fprintf(stderr, ".\n");
1090 }
1091 if (content->reshape_active &&
1092 content->array.level >= 4 &&
1093 content->array.level <= 6) {
1094 /* might need to increase the size
1095 * of the stripe cache - default is 256
1096 */
1097 if (256 < 4 * (content->array.chunk_size/4096)) {
1098 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1099 if (sra)
1100 sysfs_set_num(sra, NULL,
1101 "stripe_cache_size",
1102 (4 * content->array.chunk_size / 4096) + 1);
1103 sysfs_free(sra);
1104 }
1105 }
1106 if (okcnt < (unsigned)content->array.raid_disks) {
1107 /* If any devices did not get added
1108 * because the kernel rejected them based
1109 * on event count, try adding them
1110 * again providing the action policy is
1111 * 're-add' or greater. The bitmap
1112 * might allow them to be included, or
1113 * they will become spares.
1114 */
1115 for (i = 0; i < bestcnt; i++) {
1116 int j = best[i];
1117 if (j >= 0 && !devices[j].uptodate) {
1118 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1119 continue;
1120 rv = add_disk(mdfd, st, content,
1121 &devices[j].i);
1122 if (rv == 0 && c->verbose >= 0)
1123 pr_err("%s has been re-added.\n",
1124 devices[j].devname);
1125 }
1126 }
1127 }
1128 if (content->array.level == 6 &&
1129 okcnt + 1 == (unsigned)content->array.raid_disks &&
1130 was_forced) {
1131 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1132 if (sra)
1133 sysfs_set_str(sra, NULL,
1134 "sync_action", "repair");
1135 sysfs_free(sra);
1136 }
1137 return 0;
1138 }
1139 pr_err("failed to RUN_ARRAY %s: %s\n",
1140 mddev, strerror(errno));
1141
1142 if (!enough(content->array.level, content->array.raid_disks,
1143 content->array.layout, 1, avail))
1144 pr_err("Not enough devices to "
1145 "start the array.\n");
1146 else if (!enough(content->array.level,
1147 content->array.raid_disks,
1148 content->array.layout, clean,
1149 avail))
1150 pr_err("Not enough devices to "
1151 "start the array while not clean "
1152 "- consider --force.\n");
1153
1154 return 1;
1155 }
1156 if (c->runstop == -1) {
1157 pr_err("%s assembled from %d drive%s",
1158 mddev, okcnt, okcnt==1?"":"s");
1159 if (okcnt != (unsigned)content->array.raid_disks)
1160 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1161 fprintf(stderr, ", but not started.\n");
1162 return 2;
1163 }
1164 if (c->verbose >= -1) {
1165 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1166 if (rebuilding_cnt)
1167 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1168 if (sparecnt)
1169 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1170 if (!enough(content->array.level, content->array.raid_disks,
1171 content->array.layout, 1, avail))
1172 fprintf(stderr, " - not enough to start the array.\n");
1173 else if (!enough(content->array.level,
1174 content->array.raid_disks,
1175 content->array.layout, clean,
1176 avail))
1177 fprintf(stderr, " - not enough to start the "
1178 "array while not clean - consider "
1179 "--force.\n");
1180 else {
1181 if (req_cnt == (unsigned)content->array.raid_disks)
1182 fprintf(stderr, " - need all %d to start it", req_cnt);
1183 else
1184 fprintf(stderr, " - need %d to start", req_cnt);
1185 fprintf(stderr, " (use --run to insist).\n");
1186 }
1187 }
1188 return 1;
1189 }
1190
1191 int Assemble(struct supertype *st, char *mddev,
1192 struct mddev_ident *ident,
1193 struct mddev_dev *devlist,
1194 struct context *c)
1195 {
1196 /*
1197 * The task of Assemble is to find a collection of
1198 * devices that should (according to their superblocks)
1199 * form an array, and to give this collection to the MD driver.
1200 * In Linux-2.4 and later, this involves submitting a
1201 * SET_ARRAY_INFO ioctl with no arg - to prepare
1202 * the array - and then submit a number of
1203 * ADD_NEW_DISK ioctls to add disks into
1204 * the array. Finally RUN_ARRAY might
1205 * be submitted to start the array.
1206 *
1207 * Much of the work of Assemble is in finding and/or
1208 * checking the disks to make sure they look right.
1209 *
1210 * If mddev is not set, then scan must be set and we
1211 * read through the config file for dev+uuid mapping
1212 * We recurse, setting mddev, for each device that
1213 * - isn't running
1214 * - has a valid uuid (or any uuid if !uuidset)
1215 *
1216 * If mddev is set, we try to determine state of md.
1217 * check version - must be at least 0.90.0
1218 * check kernel version. must be at least 2.4.
1219 * If not, we can possibly fall back on START_ARRAY
1220 * Try to GET_ARRAY_INFO.
1221 * If possible, give up
1222 * If not, try to STOP_ARRAY just to make sure
1223 *
1224 * If !uuidset and scan, look in conf-file for uuid
1225 * If not found, give up
1226 * If !devlist and scan and uuidset, get list of devs from conf-file
1227 *
1228 * For each device:
1229 * Check superblock - discard if bad
1230 * Check uuid (set if we don't have one) - discard if no match
1231 * Check superblock similarity if we have a superblock - discard if different
1232 * Record events, devicenum
1233 * This should give us a list of devices for the array
1234 * We should collect the most recent event number
1235 *
1236 * Count disks with recent enough event count
1237 * While force && !enough disks
1238 * Choose newest rejected disks, update event count
1239 * mark clean and rewrite superblock
1240 * If recent kernel:
1241 * SET_ARRAY_INFO
1242 * foreach device with recent events : ADD_NEW_DISK
1243 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1244 * If old kernel:
1245 * Check the device numbers in superblock are right
1246 * update superblock if any changes
1247 * START_ARRAY
1248 *
1249 */
1250 int rv;
1251 int mdfd;
1252 int clean;
1253 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1254 ident->super_minor == UnSet && ident->name[0] == 0
1255 && (ident->container == NULL || ident->member == NULL));
1256 struct devs *devices;
1257 char *devmap;
1258 int *best = NULL; /* indexed by raid_disk */
1259 int bestcnt = 0;
1260 int devcnt;
1261 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1262 int i;
1263 int was_forced = 0;
1264 int most_recent = 0;
1265 int chosen_drive;
1266 int change = 0;
1267 int inargv = 0;
1268 int start_partial_ok = (c->runstop >= 0) &&
1269 (c->force || devlist==NULL || auto_assem);
1270 int num_devs;
1271 struct mddev_dev *tmpdev;
1272 struct mdinfo info;
1273 struct mdinfo *content = NULL;
1274 struct mdinfo *pre_exist = NULL;
1275 char *avail;
1276 char *name = NULL;
1277 char chosen_name[1024];
1278 struct map_ent *map = NULL;
1279 struct map_ent *mp;
1280
1281 /*
1282 * If any subdevs are listed, then any that don't
1283 * match ident are discarded. Remainder must all match and
1284 * become the array.
1285 * If no subdevs, then we scan all devices in the config file, but
1286 * there must be something in the identity
1287 */
1288
1289 if (!devlist &&
1290 ident->uuid_set == 0 &&
1291 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1292 ident->name[0] == 0 &&
1293 (ident->container == NULL || ident->member == NULL) &&
1294 ident->devices == NULL) {
1295 pr_err("No identity information available for %s - cannot assemble.\n",
1296 mddev ? mddev : "further assembly");
1297 return 1;
1298 }
1299
1300 if (devlist == NULL)
1301 devlist = conf_get_devs();
1302 else if (mddev)
1303 inargv = 1;
1304
1305 try_again:
1306 /* We come back here when doing auto-assembly and attempting some
1307 * set of devices failed. Those are now marked as ->used==2 and
1308 * we ignore them and try again
1309 */
1310 if (!st && ident->st)
1311 st = ident->st;
1312 if (c->verbose>0)
1313 pr_err("looking for devices for %s\n",
1314 mddev ? mddev : "further assembly");
1315
1316 content = &info;
1317 if (st)
1318 st->ignore_hw_compat = 1;
1319 num_devs = select_devices(devlist, ident, &st, &content, c,
1320 inargv, auto_assem);
1321 if (num_devs < 0)
1322 return 1;
1323
1324 if (!st || !st->sb || !content)
1325 return 2;
1326
1327 /* We have a full set of devices - we now need to find the
1328 * array device.
1329 * However there is a risk that we are racing with "mdadm -I"
1330 * and the array is already partially assembled - we will have
1331 * rejected any devices already in this address.
1332 * So we take a lock on the map file - to prevent further races -
1333 * and look for the uuid in there. If found and the array is
1334 * active, we abort. If found and the array is not active
1335 * we commit to that md device and add all the contained devices
1336 * to our list. We flag them so that we don't try to re-add,
1337 * but can remove if they turn out to not be wanted.
1338 */
1339 if (map_lock(&map))
1340 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1341 mp = map_by_uuid(&map, content->uuid);
1342 if (mp) {
1343 struct mdinfo *dv;
1344 /* array already exists. */
1345 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1346 if (pre_exist->array.level != UnSet) {
1347 pr_err("Found some drive for an array that is already active: %s\n",
1348 mp->path);
1349 pr_err("giving up.\n");
1350 return 1;
1351 }
1352 for (dv = pre_exist->devs; dv; dv = dv->next) {
1353 /* We want to add this device to our list,
1354 * but it could already be there if "mdadm -I"
1355 * started *after* we checked for O_EXCL.
1356 * If we add it to the top of the list
1357 * it will be preferred over later copies.
1358 */
1359 struct mddev_dev *newdev;
1360 char *devname = map_dev(dv->disk.major,
1361 dv->disk.minor,
1362 0);
1363 if (!devname)
1364 continue;
1365 newdev = xmalloc(sizeof(*newdev));
1366 newdev->devname = devname;
1367 newdev->disposition = 'I';
1368 newdev->used = 1;
1369 newdev->next = devlist;
1370 devlist = newdev;
1371 num_devs++;
1372 }
1373 strcpy(chosen_name, mp->path);
1374 if (c->verbose > 0 || mddev == NULL ||
1375 strcmp(mddev, chosen_name) != 0)
1376 pr_err("Merging with already-assembled %s\n",
1377 chosen_name);
1378 mdfd = open_dev_excl(mp->devnm);
1379 } else {
1380 int trustworthy = FOREIGN;
1381 name = content->name;
1382 switch (st->ss->match_home(st, c->homehost)
1383 ?: st->ss->match_home(st, "any")) {
1384 case 1:
1385 trustworthy = LOCAL;
1386 name = strchr(content->name, ':');
1387 if (name)
1388 name++;
1389 else
1390 name = content->name;
1391 break;
1392 }
1393 if (!auto_assem)
1394 /* If the array is listed in mdadm.conf or on
1395 * command line, then we trust the name
1396 * even if the array doesn't look local
1397 */
1398 trustworthy = LOCAL;
1399
1400 if (name[0] == 0 &&
1401 content->array.level == LEVEL_CONTAINER) {
1402 name = content->text_version;
1403 trustworthy = METADATA;
1404 }
1405
1406 if (name[0] && trustworthy != LOCAL &&
1407 ! c->require_homehost &&
1408 conf_name_is_free(name))
1409 trustworthy = LOCAL;
1410
1411 if (trustworthy == LOCAL &&
1412 strchr(name, ':'))
1413 /* Ignore 'host:' prefix of name */
1414 name = strchr(name, ':')+1;
1415
1416 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1417 chosen_name);
1418 }
1419 if (mdfd < 0) {
1420 st->ss->free_super(st);
1421 if (auto_assem)
1422 goto try_again;
1423 return 1;
1424 }
1425 mddev = chosen_name;
1426 if (get_linux_version() < 2004000 ||
1427 md_get_version(mdfd) < 9000) {
1428 pr_err("Assemble requires Linux 2.4 or later, and\n"
1429 " md driver version 0.90.0 or later.\n"
1430 " Upgrade your kernel or try --build\n");
1431 close(mdfd);
1432 return 1;
1433 }
1434 if (pre_exist == NULL) {
1435 if (mddev_busy(fd2devnm(mdfd))) {
1436 pr_err("%s already active, cannot restart it!\n",
1437 mddev);
1438 for (tmpdev = devlist ;
1439 tmpdev && tmpdev->used != 1;
1440 tmpdev = tmpdev->next)
1441 ;
1442 if (tmpdev && auto_assem)
1443 pr_err("%s needed for %s...\n",
1444 mddev, tmpdev->devname);
1445 close(mdfd);
1446 mdfd = -3;
1447 st->ss->free_super(st);
1448 if (auto_assem)
1449 goto try_again;
1450 return 1;
1451 }
1452 /* just incase it was started but has no content */
1453 ioctl(mdfd, STOP_ARRAY, NULL);
1454 }
1455
1456 #ifndef MDASSEMBLE
1457 if (content != &info) {
1458 /* This is a member of a container. Try starting the array. */
1459 int err;
1460 err = assemble_container_content(st, mdfd, content, c,
1461 chosen_name, NULL);
1462 close(mdfd);
1463 return err;
1464 }
1465 #endif
1466 /* Ok, no bad inconsistancy, we can try updating etc */
1467 devices = xcalloc(num_devs, sizeof(*devices));
1468 devmap = xcalloc(num_devs, content->array.raid_disks);
1469 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1470 c, content, mdfd, mddev,
1471 &most_recent, &bestcnt, &best, inargv);
1472 if (devcnt < 0)
1473 return 1;
1474
1475 if (devcnt == 0) {
1476 pr_err("no devices found for %s\n",
1477 mddev);
1478 if (st)
1479 st->ss->free_super(st);
1480 close(mdfd);
1481 free(devices);
1482 free(devmap);
1483 return 1;
1484 }
1485
1486 if (c->update && strcmp(c->update, "byteorder")==0)
1487 st->minor_version = 90;
1488
1489 st->ss->getinfo_super(st, content, NULL);
1490 clean = content->array.state & 1;
1491
1492 /* now we have some devices that might be suitable.
1493 * I wonder how many
1494 */
1495 avail = xcalloc(content->array.raid_disks, 1);
1496 okcnt = 0;
1497 replcnt = 0;
1498 sparecnt=0;
1499 rebuilding_cnt=0;
1500 for (i=0; i< bestcnt; i++) {
1501 int j = best[i];
1502 int event_margin = 1; /* always allow a difference of '1'
1503 * like the kernel does
1504 */
1505 if (j < 0) continue;
1506 /* note: we ignore error flags in multipath arrays
1507 * as they don't make sense
1508 */
1509 if (content->array.level != LEVEL_MULTIPATH)
1510 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1511 if (!(devices[j].i.disk.state
1512 & (1<<MD_DISK_FAULTY))) {
1513 devices[j].uptodate = 1;
1514 sparecnt++;
1515 }
1516 continue;
1517 }
1518 /* If this device thinks that 'most_recent' has failed, then
1519 * we must reject this device.
1520 */
1521 if (j != most_recent && !c->force &&
1522 content->array.raid_disks > 0 &&
1523 devices[most_recent].i.disk.raid_disk >= 0 &&
1524 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1525 if (c->verbose > -1)
1526 pr_err("ignoring %s as it reports %s as failed\n",
1527 devices[j].devname, devices[most_recent].devname);
1528 best[i] = -1;
1529 continue;
1530 }
1531 /* Require event counter to be same as, or just less than,
1532 * most recent. If it is bigger, it must be a stray spare and
1533 * should be ignored.
1534 */
1535 if (devices[j].i.events+event_margin >=
1536 devices[most_recent].i.events &&
1537 devices[j].i.events <=
1538 devices[most_recent].i.events
1539 ) {
1540 devices[j].uptodate = 1;
1541 if (i < content->array.raid_disks * 2) {
1542 if (devices[j].i.recovery_start == MaxSector ||
1543 (content->reshape_active &&
1544 ((i >= content->array.raid_disks - content->delta_disks) ||
1545 (i >= content->array.raid_disks - content->delta_disks - 1
1546 && content->array.level == 4)))) {
1547 if (!avail[i/2]) {
1548 okcnt++;
1549 avail[i/2]=1;
1550 } else
1551 replcnt++;
1552 } else
1553 rebuilding_cnt++;
1554 } else
1555 sparecnt++;
1556 }
1557 }
1558 free(devmap);
1559 if (c->force) {
1560 int force_ok = force_array(content, devices, best, bestcnt,
1561 avail, most_recent, st, c);
1562 okcnt += force_ok;
1563 if (force_ok)
1564 was_forced = 1;
1565 }
1566 /* Now we want to look at the superblock which the kernel will base things on
1567 * and compare the devices that we think are working with the devices that the
1568 * superblock thinks are working.
1569 * If there are differences and --force is given, then update this chosen
1570 * superblock.
1571 */
1572 chosen_drive = -1;
1573 st->ss->free_super(st);
1574 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1575 int j = best[i];
1576 int fd;
1577
1578 if (j<0)
1579 continue;
1580 if (!devices[j].uptodate)
1581 continue;
1582 if (devices[j].i.events < devices[most_recent].i.events)
1583 continue;
1584 chosen_drive = j;
1585 if ((fd=dev_open(devices[j].devname,
1586 devices[j].included ? O_RDONLY
1587 : (O_RDONLY|O_EXCL)))< 0) {
1588 pr_err("Cannot open %s: %s\n",
1589 devices[j].devname, strerror(errno));
1590 close(mdfd);
1591 free(devices);
1592 return 1;
1593 }
1594 if (st->ss->load_super(st,fd, NULL)) {
1595 close(fd);
1596 pr_err("RAID superblock has disappeared from %s\n",
1597 devices[j].devname);
1598 close(mdfd);
1599 free(devices);
1600 return 1;
1601 }
1602 close(fd);
1603 }
1604 if (st->sb == NULL) {
1605 pr_err("No suitable drives found for %s\n", mddev);
1606 close(mdfd);
1607 free(devices);
1608 return 1;
1609 }
1610 st->ss->getinfo_super(st, content, NULL);
1611 #ifndef MDASSEMBLE
1612 sysfs_init(content, mdfd, NULL);
1613 #endif
1614 for (i=0; i<bestcnt; i++) {
1615 int j = best[i];
1616 unsigned int desired_state;
1617
1618 if (i >= content->array.raid_disks * 2)
1619 desired_state = 0;
1620 else if (i & 1)
1621 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1622 else
1623 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1624
1625 if (j<0)
1626 continue;
1627 if (!devices[j].uptodate)
1628 continue;
1629
1630 devices[j].i.disk.state = desired_state;
1631 if (!(devices[j].i.array.state & 1))
1632 clean = 0;
1633
1634 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1635 c->verbose, 0, NULL)) {
1636 if (c->force) {
1637 if (c->verbose >= 0)
1638 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1639 j, mddev, devices[j].devname);
1640 change = 1;
1641 } else {
1642 if (c->verbose >= -1)
1643 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1644 i, mddev, devices[j].devname);
1645 }
1646 }
1647 #if 0
1648 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1649 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1650 i, mddev);
1651 }
1652 #endif
1653 }
1654 if (c->force && !clean &&
1655 !enough(content->array.level, content->array.raid_disks,
1656 content->array.layout, clean,
1657 avail)) {
1658 change += st->ss->update_super(st, content, "force-array",
1659 devices[chosen_drive].devname, c->verbose,
1660 0, NULL);
1661 was_forced = 1;
1662 clean = 1;
1663 }
1664
1665 if (change) {
1666 int fd;
1667 fd = dev_open(devices[chosen_drive].devname,
1668 devices[chosen_drive].included ?
1669 O_RDWR : (O_RDWR|O_EXCL));
1670 if (fd < 0) {
1671 pr_err("Could not open %s for write - cannot Assemble array.\n",
1672 devices[chosen_drive].devname);
1673 close(mdfd);
1674 free(devices);
1675 return 1;
1676 }
1677 if (st->ss->store_super(st, fd)) {
1678 close(fd);
1679 pr_err("Could not re-write superblock on %s\n",
1680 devices[chosen_drive].devname);
1681 close(mdfd);
1682 free(devices);
1683 return 1;
1684 }
1685 if (c->verbose >= 0)
1686 pr_err("Marking array %s as 'clean'\n",
1687 mddev);
1688 close(fd);
1689 }
1690
1691 /* If we are in the middle of a reshape we may need to restore saved data
1692 * that was moved aside due to the reshape overwriting live data
1693 * The code of doing this lives in Grow.c
1694 */
1695 #ifndef MDASSEMBLE
1696 if (content->reshape_active &&
1697 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1698 int err = 0;
1699 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1700 if (c->verbose > 0)
1701 pr_err(":%s has an active reshape - checking "
1702 "if critical section needs to be restored\n",
1703 chosen_name);
1704 if (!c->backup_file)
1705 c->backup_file = locate_backup(content->sys_name);
1706 enable_fds(bestcnt/2);
1707 for (i = 0; i < bestcnt/2; i++) {
1708 int j = best[i*2];
1709 if (j >= 0) {
1710 fdlist[i] = dev_open(devices[j].devname,
1711 devices[j].included
1712 ? O_RDWR : (O_RDWR|O_EXCL));
1713 if (fdlist[i] < 0) {
1714 pr_err("Could not open %s for write - cannot Assemble array.\n",
1715 devices[j].devname);
1716 err = 1;
1717 break;
1718 }
1719 } else
1720 fdlist[i] = -1;
1721 }
1722 if (!err) {
1723 if (st->ss->external && st->ss->recover_backup)
1724 err = st->ss->recover_backup(st, content);
1725 else
1726 err = Grow_restart(st, content, fdlist, bestcnt/2,
1727 c->backup_file, c->verbose > 0);
1728 if (err && c->invalid_backup) {
1729 if (c->verbose > 0)
1730 pr_err("continuing"
1731 " without restoring backup\n");
1732 err = 0;
1733 }
1734 }
1735 while (i>0) {
1736 i--;
1737 if (fdlist[i]>=0) close(fdlist[i]);
1738 }
1739 free(fdlist);
1740 if (err) {
1741 pr_err("Failed to restore critical section for reshape, sorry.\n");
1742 if (c->backup_file == NULL)
1743 cont_err("Possibly you needed to specify the --backup-file\n");
1744 close(mdfd);
1745 free(devices);
1746 return err;
1747 }
1748 }
1749 #endif
1750
1751 /* Almost ready to actually *do* something */
1752 /* First, fill in the map, so that udev can find our name
1753 * as soon as we become active.
1754 */
1755 if (c->update && strcmp(c->update, "metadata")==0) {
1756 content->array.major_version = 1;
1757 content->array.minor_version = 0;
1758 strcpy(content->text_version, "1.0");
1759 }
1760
1761 map_update(&map, fd2devnm(mdfd), content->text_version,
1762 content->uuid, chosen_name);
1763
1764 rv = start_array(mdfd, mddev, content,
1765 st, ident, best, bestcnt,
1766 chosen_drive, devices, okcnt, sparecnt,
1767 rebuilding_cnt,
1768 c,
1769 clean, avail, start_partial_ok,
1770 pre_exist != NULL,
1771 was_forced);
1772 if (rv == 1 && !pre_exist)
1773 ioctl(mdfd, STOP_ARRAY, NULL);
1774 free(devices);
1775 map_unlock(&map);
1776 if (rv == 0) {
1777 wait_for(chosen_name, mdfd);
1778 close(mdfd);
1779 if (auto_assem) {
1780 int usecs = 1;
1781 /* There is a nasty race with 'mdadm --monitor'.
1782 * If it opens this device before we close it,
1783 * it gets an incomplete open on which IO
1784 * doesn't work and the capacity is
1785 * wrong.
1786 * If we reopen (to check for layered devices)
1787 * before --monitor closes, we loose.
1788 *
1789 * So: wait upto 1 second for there to be
1790 * a non-zero capacity.
1791 */
1792 while (usecs < 1000) {
1793 mdfd = open(mddev, O_RDONLY);
1794 if (mdfd >= 0) {
1795 unsigned long long size;
1796 if (get_dev_size(mdfd, NULL, &size) &&
1797 size > 0)
1798 break;
1799 close(mdfd);
1800 }
1801 usleep(usecs);
1802 usecs <<= 1;
1803 }
1804 }
1805 } else
1806 close(mdfd);
1807
1808 /* '2' means 'OK, but not started yet' */
1809 return rv == 2 ? 0 : rv;
1810 }
1811
1812 #ifndef MDASSEMBLE
1813 int assemble_container_content(struct supertype *st, int mdfd,
1814 struct mdinfo *content, struct context *c,
1815 char *chosen_name, int *result)
1816 {
1817 struct mdinfo *dev, *sra, *dev2;
1818 int working = 0, preexist = 0;
1819 int expansion = 0;
1820 struct map_ent *map = NULL;
1821 int old_raid_disks;
1822 int start_reshape;
1823 char *avail = NULL;
1824 int err;
1825
1826 sysfs_init(content, mdfd, NULL);
1827
1828 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1829 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1830 if (content->array.major_version == -1 &&
1831 content->array.minor_version == -2 &&
1832 c->readonly &&
1833 content->text_version[0] == '/')
1834 content->text_version[0] = '-';
1835 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1836 if (sra)
1837 sysfs_free(sra);
1838 return 1;
1839 }
1840 }
1841
1842 /* There are two types of reshape: container wide or sub-array specific
1843 * Check if metadata requests blocking container wide reshapes
1844 */
1845 start_reshape = (content->reshape_active &&
1846 !((content->reshape_active == CONTAINER_RESHAPE) &&
1847 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1848
1849 /* Block subarray here if it is under reshape now
1850 * Do not allow for any changes in this array
1851 */
1852 if (st->ss->external && content->recovery_blocked && start_reshape)
1853 block_subarray(content);
1854
1855 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1856 for (dev = content->devs; dev; dev = dev->next)
1857 if (dev2->disk.major == dev->disk.major &&
1858 dev2->disk.minor == dev->disk.minor)
1859 break;
1860 if (dev)
1861 continue;
1862 /* Don't want this one any more */
1863 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1864 errno == EBUSY) {
1865 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1866 sysfs_free(sra);
1867 return 1;
1868 }
1869 sysfs_set_str(sra, dev2, "state", "remove");
1870 }
1871 old_raid_disks = content->array.raid_disks - content->delta_disks;
1872 avail = xcalloc(content->array.raid_disks, 1);
1873 for (dev = content->devs; dev; dev = dev->next) {
1874 if (dev->disk.raid_disk >= 0)
1875 avail[dev->disk.raid_disk] = 1;
1876 if (sysfs_add_disk(content, dev, 1) == 0) {
1877 if (dev->disk.raid_disk >= old_raid_disks &&
1878 content->reshape_active)
1879 expansion++;
1880 else
1881 working++;
1882 } else if (errno == EEXIST)
1883 preexist++;
1884 }
1885 sysfs_free(sra);
1886 if (working + expansion == 0 && c->runstop <= 0) {
1887 free(avail);
1888 return 1;/* Nothing new, don't try to start */
1889 }
1890 map_update(&map, fd2devnm(mdfd),
1891 content->text_version,
1892 content->uuid, chosen_name);
1893
1894
1895 if (enough(content->array.level, content->array.raid_disks,
1896 content->array.layout, content->array.state & 1, avail) == 0) {
1897 if (c->export && result)
1898 *result |= INCR_NO;
1899 else if (c->verbose >= 0) {
1900 pr_err("%s assembled with %d device%s",
1901 chosen_name, preexist + working,
1902 preexist + working == 1 ? "":"s");
1903 if (preexist)
1904 fprintf(stderr, " (%d new)", working);
1905 fprintf(stderr, " but not started\n");
1906 }
1907 free(avail);
1908 return 1;
1909 }
1910 free(avail);
1911
1912 if (c->runstop <= 0 &&
1913 (working + preexist + expansion) <
1914 content->array.working_disks) {
1915 if (c->export && result)
1916 *result |= INCR_UNSAFE;
1917 else if (c->verbose >= 0) {
1918 pr_err("%s assembled with %d device%s",
1919 chosen_name, preexist + working,
1920 preexist + working == 1 ? "":"s");
1921 if (preexist)
1922 fprintf(stderr, " (%d new)", working);
1923 fprintf(stderr, " but not safe to start\n");
1924 }
1925 return 1;
1926 }
1927
1928
1929 if (start_reshape) {
1930 int spare = content->array.raid_disks + expansion;
1931 if (restore_backup(st, content,
1932 working,
1933 spare, &c->backup_file, c->verbose) == 1)
1934 return 1;
1935
1936 err = sysfs_set_str(content, NULL,
1937 "array_state", "readonly");
1938 if (err)
1939 return 1;
1940
1941 if (st->ss->external) {
1942 if (!mdmon_running(st->container_devnm))
1943 start_mdmon(st->container_devnm);
1944 ping_monitor(st->container_devnm);
1945 if (mdmon_running(st->container_devnm) &&
1946 st->update_tail == NULL)
1947 st->update_tail = &st->updates;
1948 }
1949
1950 err = Grow_continue(mdfd, st, content, c->backup_file,
1951 0, c->freeze_reshape);
1952 } else switch(content->array.level) {
1953 case LEVEL_LINEAR:
1954 case LEVEL_MULTIPATH:
1955 case 0:
1956 err = sysfs_set_str(content, NULL, "array_state",
1957 c->readonly ? "readonly" : "active");
1958 break;
1959 default:
1960 err = sysfs_set_str(content, NULL, "array_state",
1961 "readonly");
1962 /* start mdmon if needed. */
1963 if (!err) {
1964 if (!mdmon_running(st->container_devnm))
1965 start_mdmon(st->container_devnm);
1966 ping_monitor(st->container_devnm);
1967 }
1968 break;
1969 }
1970 if (!err)
1971 sysfs_set_safemode(content, content->safe_mode_delay);
1972
1973 /* Block subarray here if it is not reshaped now
1974 * It has be blocked a little later to allow mdmon to switch in
1975 * in to R/W state
1976 */
1977 if (st->ss->external && content->recovery_blocked &&
1978 !start_reshape)
1979 block_subarray(content);
1980
1981 if (c->export && result) {
1982 if (err)
1983 *result |= INCR_NO;
1984 else
1985 *result |= INCR_YES;
1986 } else if (c->verbose >= 0) {
1987 if (err)
1988 pr_err("array %s now has %d device%s",
1989 chosen_name, working + preexist,
1990 working + preexist == 1 ? "":"s");
1991 else
1992 pr_err("Started %s with %d device%s",
1993 chosen_name, working + preexist,
1994 working + preexist == 1 ? "":"s");
1995 if (preexist)
1996 fprintf(stderr, " (%d new)", working);
1997 if (expansion)
1998 fprintf(stderr, " ( + %d for expansion)",
1999 expansion);
2000 fprintf(stderr, "\n");
2001 }
2002 if (!err)
2003 wait_for(chosen_name, mdfd);
2004 return err;
2005 /* FIXME should have an O_EXCL and wait for read-auto */
2006 }
2007 #endif
Unnamed repository; edit this file 'description' to name the repository.