]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:


7e8cdb4de8ee75a16cfedb2a7b06dded4159ba13
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2012 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(1, 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 ident->raid_disks!= content->array.raid_disks) {
109 if (devname)
110 pr_err("%s requires wrong number of drives.\n",
111 devname);
112 return 0;
113 }
114 if (ident->member && ident->member[0]) {
115 /* content->text_version must match */
116 char *s = strchr(content->text_version+1, '/');
117 if (s == NULL) {
118 if (devname)
119 pr_err("%s is not a container and one is required.\n",
120 devname);
121 return 0;
122 } else if (strcmp(ident->member, s+1) != 0) {
123 if (devname)
124 pr_err("skipping wrong member %s is %s\n",
125 content->text_version, devname);
126 return 0;
127 }
128 }
129 return 1;
130 }
131
132 static int select_devices(struct mddev_dev *devlist,
133 struct mddev_ident *ident,
134 struct supertype **stp,
135 struct mdinfo **contentp,
136 struct context *c,
137 int inargv, int auto_assem)
138 {
139 struct mddev_dev *tmpdev;
140 int num_devs;
141 struct supertype *st = *stp;
142 struct mdinfo *content = NULL;
143 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
144 struct domainlist *domains = NULL;
145
146 tmpdev = devlist; num_devs = 0;
147 while (tmpdev) {
148 if (tmpdev->used)
149 tmpdev->used = 2;
150 else
151 num_devs++;
152 tmpdev->disposition = 0;
153 tmpdev = tmpdev->next;
154 }
155
156 /* first walk the list of devices to find a consistent set
157 * that match the criterea, if that is possible.
158 * We flag the ones we like with 'used'.
159 */
160 for (tmpdev = devlist;
161 tmpdev;
162 tmpdev = tmpdev ? tmpdev->next : NULL) {
163 char *devname = tmpdev->devname;
164 int dfd;
165 struct stat stb;
166 struct supertype *tst;
167 struct dev_policy *pol = NULL;
168 int found_container = 0;
169
170 if (tmpdev->used > 1)
171 continue;
172
173 if (ident->devices &&
174 !match_oneof(ident->devices, devname)) {
175 if (report_mismatch)
176 pr_err("%s is not one of %s\n", devname, ident->devices);
177 continue;
178 }
179
180 tst = dup_super(st);
181
182 dfd = dev_open(devname, O_RDONLY);
183 if (dfd < 0) {
184 if (report_mismatch)
185 pr_err("cannot open device %s: %s\n",
186 devname, strerror(errno));
187 tmpdev->used = 2;
188 } else if (fstat(dfd, &stb)< 0) {
189 /* Impossible! */
190 pr_err("fstat failed for %s: %s\n",
191 devname, strerror(errno));
192 tmpdev->used = 2;
193 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
194 pr_err("%s is not a block device.\n",
195 devname);
196 tmpdev->used = 2;
197 } else if (must_be_container(dfd)) {
198 if (st) {
199 /* already found some components, this cannot
200 * be another one.
201 */
202 if (report_mismatch)
203 pr_err("%s is a container, but we are looking for components\n",
204 devname);
205 tmpdev->used = 2;
206 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
207 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
208 if (report_mismatch)
209 pr_err("not a recognisable container: %s\n",
210 devname);
211 tmpdev->used = 2;
212 #endif
213 } else if (!tst->ss->load_container
214 || tst->ss->load_container(tst, dfd, NULL)) {
215 if (report_mismatch)
216 pr_err("no correct container type: %s\n",
217 devname);
218 tmpdev->used = 2;
219 } else if (auto_assem &&
220 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
221 tst->ss->match_home(tst, c->homehost) == 1)) {
222 if (report_mismatch)
223 pr_err("%s has metadata type %s for which "
224 "auto-assembly is disabled\n",
225 devname, tst->ss->name);
226 tmpdev->used = 2;
227 } else
228 found_container = 1;
229 } else {
230 if (!tst && (tst = guess_super(dfd)) == NULL) {
231 if (report_mismatch)
232 pr_err("no recogniseable superblock on %s\n",
233 devname);
234 tmpdev->used = 2;
235 } else if (tst->ss->load_super(tst,dfd, NULL)) {
236 if (report_mismatch)
237 pr_err("no RAID superblock on %s\n",
238 devname);
239 tmpdev->used = 2;
240 } else if (tst->ss->compare_super == NULL) {
241 if (report_mismatch)
242 pr_err("Cannot assemble %s metadata on %s\n",
243 tst->ss->name, devname);
244 tmpdev->used = 2;
245 } else if (auto_assem && st == NULL &&
246 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
247 tst->ss->match_home(tst, c->homehost) == 1)) {
248 if (report_mismatch)
249 pr_err("%s has metadata type %s for which "
250 "auto-assembly is disabled\n",
251 devname, tst->ss->name);
252 tmpdev->used = 2;
253 }
254 }
255 if (dfd >= 0) close(dfd);
256 if (tmpdev->used == 2) {
257 if (auto_assem || !inargv)
258 /* Ignore unrecognised devices during auto-assembly */
259 goto loop;
260 if (ident->uuid_set || ident->name[0] ||
261 ident->super_minor != UnSet)
262 /* Ignore unrecognised device if looking for
263 * specific array */
264 goto loop;
265
266 pr_err("%s has no superblock - assembly aborted\n",
267 devname);
268 if (st)
269 st->ss->free_super(st);
270 dev_policy_free(pol);
271 domain_free(domains);
272 return -1;
273 }
274
275 if (found_container) {
276 /* tmpdev is a container. We need to be either
277 * looking for a member, or auto-assembling
278 */
279 /* should be safe to try an exclusive open now, we
280 * have rejected anything that some other mdadm might
281 * be looking at
282 */
283 dfd = dev_open(devname, O_RDONLY | O_EXCL);
284 if (dfd < 0) {
285 if (report_mismatch)
286 pr_err("%s is busy - skipping\n", devname);
287 goto loop;
288 }
289 close(dfd);
290
291 if (ident->container) {
292 if (ident->container[0] == '/' &&
293 !same_dev(ident->container, devname)) {
294 if (report_mismatch)
295 pr_err("%s is not the container required (%s)\n",
296 devname, ident->container);
297 goto loop;
298 }
299 if (ident->container[0] != '/') {
300 /* we have a uuid */
301 int uuid[4];
302
303 content = *contentp;
304 tst->ss->getinfo_super(tst, content, NULL);
305
306 if (!parse_uuid(ident->container, uuid) ||
307 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
308 if (report_mismatch)
309 pr_err("%s has wrong UUID to be required container\n",
310 devname);
311 goto loop;
312 }
313 }
314 }
315 /* It is worth looking inside this container.
316 */
317 if (c->verbose > 0)
318 pr_err("looking in container %s\n",
319 devname);
320
321 for (content = tst->ss->container_content(tst, NULL);
322 content;
323 content = content->next) {
324
325 if (!ident_matches(ident, content, tst,
326 c->homehost, c->update,
327 report_mismatch ? devname : NULL))
328 /* message already printed */;
329 else if (is_member_busy(content->text_version)) {
330 if (report_mismatch)
331 pr_err("member %s in %s is already assembled\n",
332 content->text_version,
333 devname);
334 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
335 /* do not assemble arrays with unsupported configurations */
336 pr_err("Cannot activate member %s in %s.\n",
337 content->text_version,
338 devname);
339 } else
340 break;
341 }
342 if (!content) {
343 tmpdev->used = 2;
344 goto loop; /* empty container */
345 }
346
347 st = tst; tst = NULL;
348 if (!auto_assem && inargv && tmpdev->next != NULL) {
349 pr_err("%s is a container, but is not "
350 "only device given: confused and aborting\n",
351 devname);
352 st->ss->free_super(st);
353 dev_policy_free(pol);
354 domain_free(domains);
355 return -1;
356 }
357 if (c->verbose > 0)
358 pr_err("found match on member %s in %s\n",
359 content->text_version, devname);
360
361 /* make sure we finished the loop */
362 tmpdev = NULL;
363 goto loop;
364 } else {
365 int rv = 0;
366 struct mddev_ident *match;
367
368 content = *contentp;
369 tst->ss->getinfo_super(tst, content, NULL);
370
371 if (!ident_matches(ident, content, tst,
372 c->homehost, c->update,
373 report_mismatch ? devname : NULL))
374 goto loop;
375
376 match = conf_match(tst, content, devname,
377 report_mismatch ? c->verbose : -1,
378 &rv);
379 if (!match && rv == 2)
380 goto loop;
381 if (match && match->devname &&
382 strcasecmp(match->devname, "<ignore>") == 0) {
383 if (report_mismatch)
384 pr_err("%s is a member of an explicitly ignored array\n",
385 devname);
386 goto loop;
387 }
388 if (match && !ident_matches(match, content, tst,
389 c->homehost, c->update,
390 report_mismatch ? devname : NULL))
391 /* Array exists in mdadm.conf but some
392 * details don't match, so reject it
393 */
394 goto loop;
395
396 /* should be safe to try an exclusive open now, we
397 * have rejected anything that some other mdadm might
398 * be looking at
399 */
400 dfd = dev_open(devname, O_RDONLY | O_EXCL);
401 if (dfd < 0) {
402 if (report_mismatch)
403 pr_err("%s is busy - skipping\n", devname);
404 goto loop;
405 }
406 close(dfd);
407
408 if (st == NULL)
409 st = dup_super(tst);
410 if (st->minor_version == -1)
411 st->minor_version = tst->minor_version;
412
413 if (memcmp(content->uuid, uuid_zero,
414 sizeof(int[4])) == 0) {
415 /* this is a floating spare. It cannot define
416 * an array unless there are no more arrays of
417 * this type to be found. It can be included
418 * in an array of this type though.
419 */
420 tmpdev->used = 3;
421 goto loop;
422 }
423
424 if (st->ss != tst->ss ||
425 st->minor_version != tst->minor_version ||
426 st->ss->compare_super(st, tst) != 0) {
427 /* Some mismatch. If exactly one array matches this host,
428 * we can resolve on that one.
429 * Or, if we are auto assembling, we just ignore the second
430 * for now.
431 */
432 if (auto_assem)
433 goto loop;
434 if (c->homehost) {
435 int first = st->ss->match_home(st, c->homehost);
436 int last = tst->ss->match_home(tst, c->homehost);
437 if (first != last &&
438 (first == 1 || last == 1)) {
439 /* We can do something */
440 if (first) {/* just ignore this one */
441 if (report_mismatch)
442 pr_err("%s misses out due to wrong homehost\n",
443 devname);
444 goto loop;
445 } else { /* reject all those sofar */
446 struct mddev_dev *td;
447 if (report_mismatch)
448 pr_err("%s overrides previous devices due to good homehost\n",
449 devname);
450 for (td=devlist; td != tmpdev; td=td->next)
451 if (td->used == 1)
452 td->used = 0;
453 tmpdev->used = 1;
454 goto loop;
455 }
456 }
457 }
458 pr_err("superblock on %s doesn't match others - assembly aborted\n",
459 devname);
460 tst->ss->free_super(tst);
461 st->ss->free_super(st);
462 dev_policy_free(pol);
463 domain_free(domains);
464 return -1;
465 }
466 tmpdev->used = 1;
467 }
468 loop:
469 /* Collect domain information from members only */
470 if (tmpdev && tmpdev->used == 1) {
471 if (!pol)
472 pol = devid_policy(stb.st_rdev);
473 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
474 }
475 dev_policy_free(pol);
476 pol = NULL;
477 if (tst)
478 tst->ss->free_super(tst);
479 }
480
481 /* Check if we found some imsm spares but no members */
482 if ((auto_assem ||
483 (ident->uuid_set &&
484 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
485 (!st || !st->sb))
486 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
487 if (tmpdev->used != 3)
488 continue;
489 tmpdev->used = 1;
490 content = *contentp;
491
492 if (!st->sb) {
493 /* we need sb from one of the spares */
494 int dfd = dev_open(tmpdev->devname, O_RDONLY);
495 if (dfd < 0 ||
496 st->ss->load_super(st, dfd, NULL))
497 tmpdev->used = 2;
498 if (dfd > 0)
499 close(dfd);
500 }
501 }
502
503 /* Now reject spares that don't match domains of identified members */
504 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
505 struct stat stb;
506 if (tmpdev->used != 3)
507 continue;
508 if (stat(tmpdev->devname, &stb)< 0) {
509 pr_err("fstat failed for %s: %s\n",
510 tmpdev->devname, strerror(errno));
511 tmpdev->used = 2;
512 } else {
513 struct dev_policy *pol = devid_policy(stb.st_rdev);
514 int dt = domain_test(domains, pol, NULL);
515 if (inargv && dt != 0)
516 /* take this spare as domains match
517 * if there are any */
518 tmpdev->used = 1;
519 else if (!inargv && dt == 1)
520 /* device wasn't explicitly listed, so need
521 * explicit domain match - which we have */
522 tmpdev->used = 1;
523 else
524 /* if domains don't match mark as unused */
525 tmpdev->used = 0;
526 dev_policy_free(pol);
527 }
528 }
529 domain_free(domains);
530 *stp = st;
531 if (st && st->sb && content == *contentp)
532 st->ss->getinfo_super(st, content, NULL);
533 *contentp = content;
534
535 return num_devs;
536 }
537
538 struct devs {
539 char *devname;
540 int uptodate; /* set once we decide that this device is as
541 * recent as everything else in the array.
542 */
543 int included; /* set if the device is already in the array
544 * due to a previous '-I'
545 */
546 struct mdinfo i;
547 };
548
549 static int load_devices(struct devs *devices, char *devmap,
550 struct mddev_ident *ident, struct supertype *st,
551 struct mddev_dev *devlist, struct context *c,
552 struct mdinfo *content,
553 int mdfd, char *mddev,
554 int *most_recentp, int *bestcntp, int **bestp,
555 int inargv)
556 {
557 struct mddev_dev *tmpdev;
558 int devcnt = 0;
559 int nextspare = 0;
560 #ifndef MDASSEMBLE
561 int bitmap_done = 0;
562 #endif
563 int most_recent = 0;
564 int bestcnt = 0;
565 int *best = *bestp;
566
567 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
568 char *devname = tmpdev->devname;
569 struct stat stb;
570 int i;
571
572 if (tmpdev->used != 1)
573 continue;
574 /* looks like a good enough match to update the super block if needed */
575 #ifndef MDASSEMBLE
576 if (c->update) {
577 int dfd;
578 /* prepare useful information in info structures */
579 struct stat stb2;
580 struct supertype *tst;
581 int err;
582 fstat(mdfd, &stb2);
583
584 if (strcmp(c->update, "uuid")==0 &&
585 !ident->uuid_set) {
586 int rfd;
587 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
588 read(rfd, ident->uuid, 16) != 16) {
589 *(__u32*)(ident->uuid) = random();
590 *(__u32*)(ident->uuid+1) = random();
591 *(__u32*)(ident->uuid+2) = random();
592 *(__u32*)(ident->uuid+3) = random();
593 }
594 if (rfd >= 0) close(rfd);
595 }
596 dfd = dev_open(devname,
597 tmpdev->disposition == 'I'
598 ? O_RDWR : (O_RDWR|O_EXCL));
599
600 tst = dup_super(st);
601 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
602 pr_err("cannot re-read metadata from %s - aborting\n",
603 devname);
604 if (dfd >= 0)
605 close(dfd);
606 close(mdfd);
607 free(devices);
608 free(devmap);
609 return -1;
610 }
611 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
612
613 memcpy(content->uuid, ident->uuid, 16);
614 strcpy(content->name, ident->name);
615 content->array.md_minor = minor(stb2.st_rdev);
616
617 if (strcmp(c->update, "byteorder") == 0)
618 err = 0;
619 else
620 err = tst->ss->update_super(tst, content, c->update,
621 devname, c->verbose,
622 ident->uuid_set,
623 c->homehost);
624 if (err < 0) {
625 if (err == -1)
626 pr_err("--update=%s not understood"
627 " for %s metadata\n",
628 c->update, tst->ss->name);
629 tst->ss->free_super(tst);
630 free(tst);
631 close(mdfd);
632 close(dfd);
633 free(devices);
634 free(devmap);
635 return -1;
636 }
637 if (strcmp(c->update, "uuid")==0 &&
638 !ident->uuid_set) {
639 ident->uuid_set = 1;
640 memcpy(ident->uuid, content->uuid, 16);
641 }
642 if (tst->ss->store_super(tst, dfd))
643 pr_err("Could not re-write superblock on %s.\n",
644 devname);
645 close(dfd);
646
647 if (strcmp(c->update, "uuid")==0 &&
648 ident->bitmap_fd >= 0 && !bitmap_done) {
649 if (bitmap_update_uuid(ident->bitmap_fd,
650 content->uuid,
651 tst->ss->swapuuid) != 0)
652 pr_err("Could not update uuid on external bitmap.\n");
653 else
654 bitmap_done = 1;
655 }
656 tst->ss->free_super(tst);
657 } else
658 #endif
659 {
660 struct supertype *tst = dup_super(st);
661 int dfd;
662 dfd = dev_open(devname,
663 tmpdev->disposition == 'I'
664 ? O_RDWR : (O_RDWR|O_EXCL));
665
666 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
667 pr_err("cannot re-read metadata from %s - aborting\n",
668 devname);
669 if (dfd >= 0)
670 close(dfd);
671 close(mdfd);
672 free(devices);
673 free(devmap);
674 return -1;
675 }
676 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
677 tst->ss->free_super(tst);
678 close(dfd);
679 }
680
681 stat(devname, &stb);
682
683 if (c->verbose > 0)
684 pr_err("%s is identified as a member of %s, slot %d%s.\n",
685 devname, mddev, content->disk.raid_disk,
686 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
687 devices[devcnt].devname = devname;
688 devices[devcnt].uptodate = 0;
689 devices[devcnt].included = (tmpdev->disposition == 'I');
690 devices[devcnt].i = *content;
691 devices[devcnt].i.disk.major = major(stb.st_rdev);
692 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
693 if (most_recent < devcnt) {
694 if (devices[devcnt].i.events
695 > devices[most_recent].i.events)
696 most_recent = devcnt;
697 }
698 if (content->array.level == LEVEL_MULTIPATH)
699 /* with multipath, the raid_disk from the superblock is meaningless */
700 i = devcnt;
701 else
702 i = devices[devcnt].i.disk.raid_disk;
703 if (i+1 == 0) {
704 if (nextspare < content->array.raid_disks*2)
705 nextspare = content->array.raid_disks*2;
706 i = nextspare++;
707 } else {
708 /* i is raid_disk - double it so there is room for
709 * replacements */
710 i *= 2;
711 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
712 i++;
713 if (i >= content->array.raid_disks*2 &&
714 i >= nextspare)
715 nextspare = i+1;
716 }
717 if (i < 10000) {
718 if (i >= bestcnt) {
719 int newbestcnt = i+10;
720 int *newbest = xmalloc(sizeof(int)*newbestcnt);
721 int c;
722 for (c=0; c < newbestcnt; c++)
723 if (c < bestcnt)
724 newbest[c] = best[c];
725 else
726 newbest[c] = -1;
727 if (best)free(best);
728 best = newbest;
729 bestcnt = newbestcnt;
730 }
731 if (best[i] >=0 &&
732 devices[best[i]].i.events
733 == devices[devcnt].i.events
734 && (devices[best[i]].i.disk.minor
735 != devices[devcnt].i.disk.minor)
736 && st->ss == &super0
737 && content->array.level != LEVEL_MULTIPATH) {
738 /* two different devices with identical superblock.
739 * Could be a mis-detection caused by overlapping
740 * partitions. fail-safe.
741 */
742 pr_err("WARNING %s and %s appear"
743 " to have very similar superblocks.\n"
744 " If they are really different, "
745 "please --zero the superblock on one\n"
746 " If they are the same or overlap,"
747 " please remove one from %s.\n",
748 devices[best[i]].devname, devname,
749 inargv ? "the list" :
750 "the\n DEVICE list in mdadm.conf"
751 );
752 close(mdfd);
753 free(devices);
754 free(devmap);
755 return -1;
756 }
757 if (best[i] == -1
758 || (devices[best[i]].i.events
759 < devices[devcnt].i.events))
760 best[i] = devcnt;
761 }
762 devcnt++;
763 }
764 *most_recentp = most_recent;
765 *bestcntp = bestcnt;
766 *bestp = best;
767 return devcnt;
768 }
769
770 static int force_array(struct mdinfo *content,
771 struct devs *devices,
772 int *best, int bestcnt, char *avail,
773 int most_recent,
774 struct supertype *st,
775 struct context *c)
776 {
777 int okcnt = 0;
778 while (!enough(content->array.level, content->array.raid_disks,
779 content->array.layout, 1,
780 avail)
781 ||
782 (content->reshape_active && content->delta_disks > 0 &&
783 !enough(content->array.level, (content->array.raid_disks
784 - content->delta_disks),
785 content->new_layout, 1,
786 avail)
787 )) {
788 /* Choose the newest best drive which is
789 * not up-to-date, update the superblock
790 * and add it.
791 */
792 int fd;
793 struct supertype *tst;
794 unsigned long long current_events;
795 int chosen_drive = -1;
796 int i;
797
798 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
799 int j = best[i];
800 if (j>=0 &&
801 !devices[j].uptodate &&
802 devices[j].i.recovery_start == MaxSector &&
803 (chosen_drive < 0 ||
804 devices[j].i.events
805 > devices[chosen_drive].i.events))
806 chosen_drive = j;
807 }
808 if (chosen_drive < 0)
809 break;
810 current_events = devices[chosen_drive].i.events;
811 add_another:
812 if (c->verbose >= 0)
813 pr_err("forcing event count in %s(%d) from %d upto %d\n",
814 devices[chosen_drive].devname,
815 devices[chosen_drive].i.disk.raid_disk,
816 (int)(devices[chosen_drive].i.events),
817 (int)(devices[most_recent].i.events));
818 fd = dev_open(devices[chosen_drive].devname,
819 devices[chosen_drive].included ? O_RDWR
820 : (O_RDWR|O_EXCL));
821 if (fd < 0) {
822 pr_err("Couldn't open %s for write - not updating\n",
823 devices[chosen_drive].devname);
824 devices[chosen_drive].i.events = 0;
825 continue;
826 }
827 tst = dup_super(st);
828 if (tst->ss->load_super(tst,fd, NULL)) {
829 close(fd);
830 pr_err("RAID superblock disappeared from %s - not updating.\n",
831 devices[chosen_drive].devname);
832 devices[chosen_drive].i.events = 0;
833 continue;
834 }
835 content->events = devices[most_recent].i.events;
836 tst->ss->update_super(tst, content, "force-one",
837 devices[chosen_drive].devname, c->verbose,
838 0, NULL);
839
840 if (tst->ss->store_super(tst, fd)) {
841 close(fd);
842 pr_err("Could not re-write superblock on %s\n",
843 devices[chosen_drive].devname);
844 devices[chosen_drive].i.events = 0;
845 tst->ss->free_super(tst);
846 continue;
847 }
848 close(fd);
849 devices[chosen_drive].i.events = devices[most_recent].i.events;
850 devices[chosen_drive].uptodate = 1;
851 avail[chosen_drive] = 1;
852 okcnt++;
853 tst->ss->free_super(tst);
854
855 /* If there are any other drives of the same vintage,
856 * add them in as well. We can't lose and we might gain
857 */
858 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
859 int j = best[i];
860 if (j >= 0 &&
861 !devices[j].uptodate &&
862 devices[j].i.recovery_start == MaxSector &&
863 devices[j].i.events == current_events) {
864 chosen_drive = j;
865 goto add_another;
866 }
867 }
868 }
869 return okcnt;
870 }
871
872 static int start_array(int mdfd,
873 char *mddev,
874 struct mdinfo *content,
875 struct supertype *st,
876 struct mddev_ident *ident,
877 int *best, int bestcnt,
878 int chosen_drive,
879 struct devs *devices,
880 unsigned int okcnt,
881 unsigned int sparecnt,
882 unsigned int rebuilding_cnt,
883 struct context *c,
884 int clean, char *avail,
885 int start_partial_ok
886 )
887 {
888 int rv;
889 int i;
890 unsigned int req_cnt;
891
892 rv = set_array_info(mdfd, st, content);
893 if (rv) {
894 pr_err("failed to set array info for %s: %s\n",
895 mddev, strerror(errno));
896 return 1;
897 }
898 if (ident->bitmap_fd >= 0) {
899 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
900 pr_err("SET_BITMAP_FILE failed.\n");
901 return 1;
902 }
903 } else if (ident->bitmap_file) {
904 /* From config file */
905 int bmfd = open(ident->bitmap_file, O_RDWR);
906 if (bmfd < 0) {
907 pr_err("Could not open bitmap file %s\n",
908 ident->bitmap_file);
909 return 1;
910 }
911 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
912 pr_err("Failed to set bitmapfile for %s\n", mddev);
913 close(bmfd);
914 return 1;
915 }
916 close(bmfd);
917 }
918
919 /* First, add the raid disks, but add the chosen one last */
920 for (i=0; i<= bestcnt; i++) {
921 int j;
922 if (i < bestcnt) {
923 j = best[i];
924 if (j == chosen_drive)
925 continue;
926 } else
927 j = chosen_drive;
928
929 if (j >= 0 && !devices[j].included) {
930 int dfd = dev_open(devices[j].devname,
931 O_RDWR|O_EXCL);
932 if (dfd >= 0) {
933 remove_partitions(dfd);
934 close(dfd);
935 }
936 rv = add_disk(mdfd, st, content, &devices[j].i);
937
938 if (rv) {
939 pr_err("failed to add "
940 "%s to %s: %s\n",
941 devices[j].devname,
942 mddev,
943 strerror(errno));
944 if (i < content->array.raid_disks * 2
945 || i == bestcnt)
946 okcnt--;
947 else
948 sparecnt--;
949 } else if (c->verbose > 0)
950 pr_err("added %s to %s as %d%s%s\n",
951 devices[j].devname, mddev,
952 devices[j].i.disk.raid_disk,
953 devices[j].uptodate?"":
954 " (possibly out of date)",
955 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
956 } else if (j >= 0) {
957 if (c->verbose > 0)
958 pr_err("%s is already in %s as %d\n",
959 devices[j].devname, mddev,
960 devices[j].i.disk.raid_disk);
961 } else if (c->verbose > 0 && i < content->array.raid_disks*2
962 && (i&1) == 0)
963 pr_err("no uptodate device for slot %d of %s\n",
964 i, mddev);
965 }
966
967 if (content->array.level == LEVEL_CONTAINER) {
968 if (c->verbose >= 0) {
969 pr_err("Container %s has been "
970 "assembled with %d drive%s",
971 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
972 if (okcnt < (unsigned)content->array.raid_disks)
973 fprintf(stderr, " (out of %d)",
974 content->array.raid_disks);
975 fprintf(stderr, "\n");
976 }
977 st->ss->free_super(st);
978 sysfs_uevent(content, "change");
979 return 0;
980 }
981
982 /* Get number of in-sync devices according to the superblock.
983 * We must have this number to start the array without -s or -R
984 */
985 req_cnt = content->array.working_disks;
986
987 if (c->runstop == 1 ||
988 (c->runstop <= 0 &&
989 ( enough(content->array.level, content->array.raid_disks,
990 content->array.layout, clean, avail) &&
991 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
992 ))) {
993 /* This array is good-to-go.
994 * If a reshape is in progress then we might need to
995 * continue monitoring it. In that case we start
996 * it read-only and let the grow code make it writable.
997 */
998 int rv;
999 #ifndef MDASSEMBLE
1000 if (content->reshape_active &&
1001 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1002 content->delta_disks <= 0) {
1003 rv = sysfs_set_str(content, NULL,
1004 "array_state", "readonly");
1005 if (rv == 0)
1006 rv = Grow_continue(mdfd, st, content,
1007 c->backup_file,
1008 c->freeze_reshape);
1009 } else if (c->readonly &&
1010 sysfs_attribute_available(
1011 content, NULL, "array_state")) {
1012 rv = sysfs_set_str(content, NULL,
1013 "array_state", "readonly");
1014 } else
1015 #endif
1016 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1017 if (rv == 0) {
1018 if (c->verbose >= 0) {
1019 pr_err("%s has been started with %d drive%s",
1020 mddev, okcnt, okcnt==1?"":"s");
1021 if (okcnt < (unsigned)content->array.raid_disks)
1022 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1023 if (rebuilding_cnt)
1024 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1025 if (sparecnt)
1026 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1027 fprintf(stderr, ".\n");
1028 }
1029 if (content->reshape_active &&
1030 content->array.level >= 4 &&
1031 content->array.level <= 6) {
1032 /* might need to increase the size
1033 * of the stripe cache - default is 256
1034 */
1035 if (256 < 4 * (content->array.chunk_size/4096)) {
1036 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1037 if (sra)
1038 sysfs_set_num(sra, NULL,
1039 "stripe_cache_size",
1040 (4 * content->array.chunk_size / 4096) + 1);
1041 sysfs_free(sra);
1042 }
1043 }
1044 if (okcnt < (unsigned)content->array.raid_disks) {
1045 /* If any devices did not get added
1046 * because the kernel rejected them based
1047 * on event count, try adding them
1048 * again providing the action policy is
1049 * 're-add' or greater. The bitmap
1050 * might allow them to be included, or
1051 * they will become spares.
1052 */
1053 for (i = 0; i < bestcnt; i++) {
1054 int j = best[i];
1055 if (j >= 0 && !devices[j].uptodate) {
1056 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1057 continue;
1058 rv = add_disk(mdfd, st, content,
1059 &devices[j].i);
1060 if (rv == 0 && c->verbose >= 0)
1061 pr_err("%s has been re-added.\n",
1062 devices[j].devname);
1063 }
1064 }
1065 }
1066 return 0;
1067 }
1068 pr_err("failed to RUN_ARRAY %s: %s\n",
1069 mddev, strerror(errno));
1070
1071 if (!enough(content->array.level, content->array.raid_disks,
1072 content->array.layout, 1, avail))
1073 pr_err("Not enough devices to "
1074 "start the array.\n");
1075 else if (!enough(content->array.level,
1076 content->array.raid_disks,
1077 content->array.layout, clean,
1078 avail))
1079 pr_err("Not enough devices to "
1080 "start the array while not clean "
1081 "- consider --force.\n");
1082
1083 return 1;
1084 }
1085 if (c->runstop == -1) {
1086 pr_err("%s assembled from %d drive%s",
1087 mddev, okcnt, okcnt==1?"":"s");
1088 if (okcnt != (unsigned)content->array.raid_disks)
1089 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1090 fprintf(stderr, ", but not started.\n");
1091 return 2;
1092 }
1093 if (c->verbose >= -1) {
1094 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1095 if (rebuilding_cnt)
1096 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1097 if (sparecnt)
1098 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1099 if (!enough(content->array.level, content->array.raid_disks,
1100 content->array.layout, 1, avail))
1101 fprintf(stderr, " - not enough to start the array.\n");
1102 else if (!enough(content->array.level,
1103 content->array.raid_disks,
1104 content->array.layout, clean,
1105 avail))
1106 fprintf(stderr, " - not enough to start the "
1107 "array while not clean - consider "
1108 "--force.\n");
1109 else {
1110 if (req_cnt == (unsigned)content->array.raid_disks)
1111 fprintf(stderr, " - need all %d to start it", req_cnt);
1112 else
1113 fprintf(stderr, " - need %d to start", req_cnt);
1114 fprintf(stderr, " (use --run to insist).\n");
1115 }
1116 }
1117 return 1;
1118 }
1119
1120 int Assemble(struct supertype *st, char *mddev,
1121 struct mddev_ident *ident,
1122 struct mddev_dev *devlist,
1123 struct context *c)
1124 {
1125 /*
1126 * The task of Assemble is to find a collection of
1127 * devices that should (according to their superblocks)
1128 * form an array, and to give this collection to the MD driver.
1129 * In Linux-2.4 and later, this involves submitting a
1130 * SET_ARRAY_INFO ioctl with no arg - to prepare
1131 * the array - and then submit a number of
1132 * ADD_NEW_DISK ioctls to add disks into
1133 * the array. Finally RUN_ARRAY might
1134 * be submitted to start the array.
1135 *
1136 * Much of the work of Assemble is in finding and/or
1137 * checking the disks to make sure they look right.
1138 *
1139 * If mddev is not set, then scan must be set and we
1140 * read through the config file for dev+uuid mapping
1141 * We recurse, setting mddev, for each device that
1142 * - isn't running
1143 * - has a valid uuid (or any uuid if !uuidset)
1144 *
1145 * If mddev is set, we try to determine state of md.
1146 * check version - must be at least 0.90.0
1147 * check kernel version. must be at least 2.4.
1148 * If not, we can possibly fall back on START_ARRAY
1149 * Try to GET_ARRAY_INFO.
1150 * If possible, give up
1151 * If not, try to STOP_ARRAY just to make sure
1152 *
1153 * If !uuidset and scan, look in conf-file for uuid
1154 * If not found, give up
1155 * If !devlist and scan and uuidset, get list of devs from conf-file
1156 *
1157 * For each device:
1158 * Check superblock - discard if bad
1159 * Check uuid (set if we don't have one) - discard if no match
1160 * Check superblock similarity if we have a superblock - discard if different
1161 * Record events, devicenum
1162 * This should give us a list of devices for the array
1163 * We should collect the most recent event number
1164 *
1165 * Count disks with recent enough event count
1166 * While force && !enough disks
1167 * Choose newest rejected disks, update event count
1168 * mark clean and rewrite superblock
1169 * If recent kernel:
1170 * SET_ARRAY_INFO
1171 * foreach device with recent events : ADD_NEW_DISK
1172 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1173 * If old kernel:
1174 * Check the device numbers in superblock are right
1175 * update superblock if any changes
1176 * START_ARRAY
1177 *
1178 */
1179 int rv;
1180 int mdfd;
1181 int clean;
1182 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1183 ident->super_minor == UnSet && ident->name[0] == 0
1184 && (ident->container == NULL || ident->member == NULL));
1185 struct devs *devices;
1186 char *devmap;
1187 int *best = NULL; /* indexed by raid_disk */
1188 int bestcnt = 0;
1189 int devcnt;
1190 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1191 int i;
1192 int most_recent = 0;
1193 int chosen_drive;
1194 int change = 0;
1195 int inargv = 0;
1196 int start_partial_ok = (c->runstop >= 0) &&
1197 (c->force || devlist==NULL || auto_assem);
1198 int num_devs;
1199 struct mddev_dev *tmpdev;
1200 struct mdinfo info;
1201 struct mdinfo *content = NULL;
1202 struct mdinfo *pre_exist = NULL;
1203 char *avail;
1204 char *name = NULL;
1205 char chosen_name[1024];
1206 struct map_ent *map = NULL;
1207 struct map_ent *mp;
1208
1209 /*
1210 * If any subdevs are listed, then any that don't
1211 * match ident are discarded. Remainder must all match and
1212 * become the array.
1213 * If no subdevs, then we scan all devices in the config file, but
1214 * there must be something in the identity
1215 */
1216
1217 if (!devlist &&
1218 ident->uuid_set == 0 &&
1219 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1220 ident->name[0] == 0 &&
1221 (ident->container == NULL || ident->member == NULL) &&
1222 ident->devices == NULL) {
1223 pr_err("No identity information available for %s - cannot assemble.\n",
1224 mddev ? mddev : "further assembly");
1225 return 1;
1226 }
1227
1228 if (devlist == NULL)
1229 devlist = conf_get_devs();
1230 else if (mddev)
1231 inargv = 1;
1232
1233 try_again:
1234 /* We come back here when doing auto-assembly and attempting some
1235 * set of devices failed. Those are now marked as ->used==2 and
1236 * we ignore them and try again
1237 */
1238 if (!st && ident->st)
1239 st = ident->st;
1240 if (c->verbose>0)
1241 pr_err("looking for devices for %s\n",
1242 mddev ? mddev : "further assembly");
1243
1244 content = &info;
1245 if (st)
1246 st->ignore_hw_compat = 1;
1247 num_devs = select_devices(devlist, ident, &st, &content, c,
1248 inargv, auto_assem);
1249 if (num_devs < 0)
1250 return 1;
1251
1252 if (!st || !st->sb || !content)
1253 return 2;
1254
1255 /* We have a full set of devices - we now need to find the
1256 * array device.
1257 * However there is a risk that we are racing with "mdadm -I"
1258 * and the array is already partially assembled - we will have
1259 * rejected any devices already in this address.
1260 * So we take a lock on the map file - to prevent further races -
1261 * and look for the uuid in there. If found and the array is
1262 * active, we abort. If found and the array is not active
1263 * we commit to that md device and add all the contained devices
1264 * to our list. We flag them so that we don't try to re-add,
1265 * but can remove if they turn out to not be wanted.
1266 */
1267 if (map_lock(&map))
1268 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1269 mp = map_by_uuid(&map, content->uuid);
1270 if (mp) {
1271 struct mdinfo *dv;
1272 /* array already exists. */
1273 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1274 if (pre_exist->array.level != UnSet) {
1275 pr_err("Found some drive for an array that is already active: %s\n",
1276 mp->path);
1277 pr_err("giving up.\n");
1278 return 1;
1279 }
1280 for (dv = pre_exist->devs; dv; dv = dv->next) {
1281 /* We want to add this device to our list,
1282 * but it could already be there if "mdadm -I"
1283 * started *after* we checked for O_EXCL.
1284 * If we add it to the top of the list
1285 * it will be preferred over later copies.
1286 */
1287 struct mddev_dev *newdev;
1288 char *devname = map_dev(dv->disk.major,
1289 dv->disk.minor,
1290 0);
1291 if (!devname)
1292 continue;
1293 newdev = xmalloc(sizeof(*newdev));
1294 newdev->devname = devname;
1295 newdev->disposition = 'I';
1296 newdev->used = 1;
1297 newdev->next = devlist;
1298 devlist = newdev;
1299 num_devs++;
1300 }
1301 strcpy(chosen_name, mp->path);
1302 if (c->verbose > 0 || mddev == NULL ||
1303 strcmp(mddev, chosen_name) != 0)
1304 pr_err("Merging with already-assembled %s\n",
1305 chosen_name);
1306 mdfd = open_dev_excl(mp->devnm);
1307 } else {
1308 int trustworthy = FOREIGN;
1309 name = content->name;
1310 switch (st->ss->match_home(st, c->homehost)
1311 ?: st->ss->match_home(st, "any")) {
1312 case 1:
1313 trustworthy = LOCAL;
1314 name = strchr(content->name, ':');
1315 if (name)
1316 name++;
1317 else
1318 name = content->name;
1319 break;
1320 }
1321 if (!auto_assem)
1322 /* If the array is listed in mdadm.conf or on
1323 * command line, then we trust the name
1324 * even if the array doesn't look local
1325 */
1326 trustworthy = LOCAL;
1327
1328 if (name[0] == 0 &&
1329 content->array.level == LEVEL_CONTAINER) {
1330 name = content->text_version;
1331 trustworthy = METADATA;
1332 }
1333
1334 if (name[0] && trustworthy != LOCAL &&
1335 ! c->require_homehost &&
1336 conf_name_is_free(name))
1337 trustworthy = LOCAL;
1338
1339 if (trustworthy == LOCAL &&
1340 strchr(name, ':'))
1341 /* Ignore 'host:' prefix of name */
1342 name = strchr(name, ':')+1;
1343
1344 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1345 chosen_name);
1346 }
1347 if (mdfd < 0) {
1348 st->ss->free_super(st);
1349 if (auto_assem)
1350 goto try_again;
1351 return 1;
1352 }
1353 mddev = chosen_name;
1354 if (get_linux_version() < 2004000 ||
1355 md_get_version(mdfd) < 9000) {
1356 pr_err("Assemble requires Linux 2.4 or later, and\n"
1357 " md driver version 0.90.0 or later.\n"
1358 " Upgrade your kernel or try --build\n");
1359 close(mdfd);
1360 return 1;
1361 }
1362 if (pre_exist == NULL) {
1363 if (mddev_busy(fd2devnm(mdfd))) {
1364 pr_err("%s already active, cannot restart it!\n",
1365 mddev);
1366 for (tmpdev = devlist ;
1367 tmpdev && tmpdev->used != 1;
1368 tmpdev = tmpdev->next)
1369 ;
1370 if (tmpdev && auto_assem)
1371 pr_err("%s needed for %s...\n",
1372 mddev, tmpdev->devname);
1373 close(mdfd);
1374 mdfd = -3;
1375 st->ss->free_super(st);
1376 if (auto_assem)
1377 goto try_again;
1378 return 1;
1379 }
1380 /* just incase it was started but has no content */
1381 ioctl(mdfd, STOP_ARRAY, NULL);
1382 }
1383
1384 #ifndef MDASSEMBLE
1385 if (content != &info) {
1386 /* This is a member of a container. Try starting the array. */
1387 int err;
1388 err = assemble_container_content(st, mdfd, content, c,
1389 chosen_name);
1390 close(mdfd);
1391 return err;
1392 }
1393 #endif
1394 /* Ok, no bad inconsistancy, we can try updating etc */
1395 devices = xcalloc(num_devs, sizeof(*devices));
1396 devmap = xcalloc(num_devs, content->array.raid_disks);
1397 devcnt = load_devices(devices, devmap, ident, st, devlist,
1398 c, content, mdfd, mddev,
1399 &most_recent, &bestcnt, &best, inargv);
1400 if (devcnt < 0)
1401 return 1;
1402
1403 if (devcnt == 0) {
1404 pr_err("no devices found for %s\n",
1405 mddev);
1406 if (st)
1407 st->ss->free_super(st);
1408 close(mdfd);
1409 free(devices);
1410 free(devmap);
1411 return 1;
1412 }
1413
1414 if (c->update && strcmp(c->update, "byteorder")==0)
1415 st->minor_version = 90;
1416
1417 st->ss->getinfo_super(st, content, NULL);
1418 clean = content->array.state & 1;
1419
1420 /* now we have some devices that might be suitable.
1421 * I wonder how many
1422 */
1423 avail = xcalloc(content->array.raid_disks, 1);
1424 okcnt = 0;
1425 replcnt = 0;
1426 sparecnt=0;
1427 rebuilding_cnt=0;
1428 for (i=0; i< bestcnt; i++) {
1429 int j = best[i];
1430 int event_margin = 1; /* always allow a difference of '1'
1431 * like the kernel does
1432 */
1433 if (j < 0) continue;
1434 /* note: we ignore error flags in multipath arrays
1435 * as they don't make sense
1436 */
1437 if (content->array.level != LEVEL_MULTIPATH)
1438 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1439 if (!(devices[j].i.disk.state
1440 & (1<<MD_DISK_FAULTY))) {
1441 devices[j].uptodate = 1;
1442 sparecnt++;
1443 }
1444 continue;
1445 }
1446 /* If this device thinks that 'most_recent' has failed, then
1447 * we must reject this device.
1448 */
1449 if (j != most_recent &&
1450 content->array.raid_disks > 0 &&
1451 devices[most_recent].i.disk.raid_disk >= 0 &&
1452 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1453 if (c->verbose > -1)
1454 pr_err("ignoring %s as it reports %s as failed\n",
1455 devices[j].devname, devices[most_recent].devname);
1456 best[i] = -1;
1457 continue;
1458 }
1459 if (devices[j].i.events+event_margin >=
1460 devices[most_recent].i.events) {
1461 devices[j].uptodate = 1;
1462 if (i < content->array.raid_disks * 2) {
1463 if (devices[j].i.recovery_start == MaxSector ||
1464 (content->reshape_active &&
1465 ((i >= content->array.raid_disks - content->delta_disks) ||
1466 (i >= content->array.raid_disks - content->delta_disks - 1
1467 && content->array.level == 4)))) {
1468 if (!avail[i/2]) {
1469 okcnt++;
1470 avail[i/2]=1;
1471 } else
1472 replcnt++;
1473 } else
1474 rebuilding_cnt++;
1475 } else
1476 sparecnt++;
1477 }
1478 }
1479 free(devmap);
1480 if (c->force)
1481 okcnt += force_array(content, devices, best, bestcnt,
1482 avail, most_recent, st, c);
1483
1484 /* Now we want to look at the superblock which the kernel will base things on
1485 * and compare the devices that we think are working with the devices that the
1486 * superblock thinks are working.
1487 * If there are differences and --force is given, then update this chosen
1488 * superblock.
1489 */
1490 chosen_drive = -1;
1491 st->ss->free_super(st);
1492 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
1493 int j = best[i];
1494 int fd;
1495
1496 if (j<0)
1497 continue;
1498 if (!devices[j].uptodate)
1499 continue;
1500 if (devices[j].i.events < devices[most_recent].i.events)
1501 continue;
1502 chosen_drive = j;
1503 if ((fd=dev_open(devices[j].devname,
1504 devices[j].included ? O_RDONLY
1505 : (O_RDONLY|O_EXCL)))< 0) {
1506 pr_err("Cannot open %s: %s\n",
1507 devices[j].devname, strerror(errno));
1508 close(mdfd);
1509 free(devices);
1510 return 1;
1511 }
1512 if (st->ss->load_super(st,fd, NULL)) {
1513 close(fd);
1514 pr_err("RAID superblock has disappeared from %s\n",
1515 devices[j].devname);
1516 close(mdfd);
1517 free(devices);
1518 return 1;
1519 }
1520 close(fd);
1521 }
1522 if (st->sb == NULL) {
1523 pr_err("No suitable drives found for %s\n", mddev);
1524 close(mdfd);
1525 free(devices);
1526 return 1;
1527 }
1528 st->ss->getinfo_super(st, content, NULL);
1529 #ifndef MDASSEMBLE
1530 sysfs_init(content, mdfd, NULL);
1531 #endif
1532 for (i=0; i<bestcnt; i++) {
1533 int j = best[i];
1534 unsigned int desired_state;
1535
1536 if (i >= content->array.raid_disks * 2)
1537 desired_state = 0;
1538 else if (i & 1)
1539 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1540 else
1541 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1542
1543 if (j<0)
1544 continue;
1545 if (!devices[j].uptodate)
1546 continue;
1547
1548 devices[j].i.disk.state = desired_state;
1549 if (!(devices[j].i.array.state & 1))
1550 clean = 0;
1551
1552 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1553 c->verbose, 0, NULL)) {
1554 if (c->force) {
1555 if (c->verbose >= 0)
1556 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1557 j, mddev, devices[j].devname);
1558 change = 1;
1559 } else {
1560 if (c->verbose >= -1)
1561 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1562 i, mddev, devices[j].devname);
1563 }
1564 }
1565 #if 0
1566 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1567 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1568 i, mddev);
1569 }
1570 #endif
1571 }
1572 if (c->force && !clean &&
1573 !enough(content->array.level, content->array.raid_disks,
1574 content->array.layout, clean,
1575 avail)) {
1576 change += st->ss->update_super(st, content, "force-array",
1577 devices[chosen_drive].devname, c->verbose,
1578 0, NULL);
1579 clean = 1;
1580 }
1581
1582 if (change) {
1583 int fd;
1584 fd = dev_open(devices[chosen_drive].devname,
1585 devices[chosen_drive].included ?
1586 O_RDWR : (O_RDWR|O_EXCL));
1587 if (fd < 0) {
1588 pr_err("Could not open %s for write - cannot Assemble array.\n",
1589 devices[chosen_drive].devname);
1590 close(mdfd);
1591 free(devices);
1592 return 1;
1593 }
1594 if (st->ss->store_super(st, fd)) {
1595 close(fd);
1596 pr_err("Could not re-write superblock on %s\n",
1597 devices[chosen_drive].devname);
1598 close(mdfd);
1599 free(devices);
1600 return 1;
1601 }
1602 if (c->verbose >= 0)
1603 pr_err("Marking array %s as 'clean'\n",
1604 mddev);
1605 close(fd);
1606 }
1607
1608 /* If we are in the middle of a reshape we may need to restore saved data
1609 * that was moved aside due to the reshape overwriting live data
1610 * The code of doing this lives in Grow.c
1611 */
1612 #ifndef MDASSEMBLE
1613 if (content->reshape_active &&
1614 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1615 int err = 0;
1616 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1617 if (c->verbose > 0)
1618 pr_err(":%s has an active reshape - checking "
1619 "if critical section needs to be restored\n",
1620 chosen_name);
1621 for (i = 0; i < bestcnt/2; i++) {
1622 int j = best[i*2];
1623 if (j >= 0) {
1624 fdlist[i] = dev_open(devices[j].devname,
1625 devices[j].included
1626 ? O_RDWR : (O_RDWR|O_EXCL));
1627 if (fdlist[i] < 0) {
1628 pr_err("Could not open %s for write - cannot Assemble array.\n",
1629 devices[j].devname);
1630 err = 1;
1631 break;
1632 }
1633 } else
1634 fdlist[i] = -1;
1635 }
1636 if (!err) {
1637 if (st->ss->external && st->ss->recover_backup)
1638 err = st->ss->recover_backup(st, content);
1639 else
1640 err = Grow_restart(st, content, fdlist, bestcnt/2,
1641 c->backup_file, c->verbose > 0);
1642 if (err && c->invalid_backup) {
1643 if (c->verbose > 0)
1644 pr_err("continuing"
1645 " without restoring backup\n");
1646 err = 0;
1647 }
1648 }
1649 while (i>0) {
1650 i--;
1651 if (fdlist[i]>=0) close(fdlist[i]);
1652 }
1653 free(fdlist);
1654 if (err) {
1655 pr_err("Failed to restore critical section for reshape, sorry.\n");
1656 if (c->backup_file == NULL)
1657 cont_err("Possibly you needed to specify the --backup-file\n");
1658 close(mdfd);
1659 free(devices);
1660 return err;
1661 }
1662 }
1663 #endif
1664
1665 /* Almost ready to actually *do* something */
1666 /* First, fill in the map, so that udev can find our name
1667 * as soon as we become active.
1668 */
1669 if (c->update && strcmp(c->update, "metadata")==0) {
1670 content->array.major_version = 1;
1671 content->array.minor_version = 0;
1672 strcpy(content->text_version, "1.0");
1673 }
1674
1675 map_update(&map, fd2devnm(mdfd), content->text_version,
1676 content->uuid, chosen_name);
1677
1678 rv = start_array(mdfd, mddev, content,
1679 st, ident, best, bestcnt,
1680 chosen_drive, devices, okcnt, sparecnt,
1681 rebuilding_cnt,
1682 c,
1683 clean, avail, start_partial_ok);
1684 if (rv == 1 && !pre_exist)
1685 ioctl(mdfd, STOP_ARRAY, NULL);
1686 free(devices);
1687 map_unlock(&map);
1688 if (rv == 0) {
1689 wait_for(chosen_name, mdfd);
1690 close(mdfd);
1691 if (auto_assem) {
1692 int usecs = 1;
1693 /* There is a nasty race with 'mdadm --monitor'.
1694 * If it opens this device before we close it,
1695 * it gets an incomplete open on which IO
1696 * doesn't work and the capacity is
1697 * wrong.
1698 * If we reopen (to check for layered devices)
1699 * before --monitor closes, we loose.
1700 *
1701 * So: wait upto 1 second for there to be
1702 * a non-zero capacity.
1703 */
1704 while (usecs < 1000) {
1705 mdfd = open(mddev, O_RDONLY);
1706 if (mdfd >= 0) {
1707 unsigned long long size;
1708 if (get_dev_size(mdfd, NULL, &size) &&
1709 size > 0)
1710 break;
1711 close(mdfd);
1712 }
1713 usleep(usecs);
1714 usecs <<= 1;
1715 }
1716 }
1717 } else
1718 close(mdfd);
1719
1720 /* '2' means 'OK, but not started yet' */
1721 return rv == 2 ? 0 : rv;
1722 }
1723
1724 #ifndef MDASSEMBLE
1725 int assemble_container_content(struct supertype *st, int mdfd,
1726 struct mdinfo *content, struct context *c,
1727 char *chosen_name)
1728 {
1729 struct mdinfo *dev, *sra;
1730 int working = 0, preexist = 0;
1731 int expansion = 0;
1732 struct map_ent *map = NULL;
1733 int old_raid_disks;
1734 int start_reshape;
1735
1736 sysfs_init(content, mdfd, NULL);
1737
1738 sra = sysfs_read(mdfd, NULL, GET_VERSION);
1739 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1740 if (content->array.major_version == -1 &&
1741 content->array.minor_version == -2 &&
1742 c->readonly &&
1743 content->text_version[0] == '/')
1744 content->text_version[0] = '-';
1745 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1746 if (sra)
1747 sysfs_free(sra);
1748 return 1;
1749 }
1750 }
1751
1752 /* There are two types of reshape: container wide or sub-array specific
1753 * Check if metadata requests blocking container wide reshapes
1754 */
1755 start_reshape = (content->reshape_active &&
1756 !((content->reshape_active == CONTAINER_RESHAPE) &&
1757 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1758
1759 /* Block subarray here if it is under reshape now
1760 * Do not allow for any changes in this array
1761 */
1762 if (st->ss->external && content->recovery_blocked && start_reshape)
1763 block_subarray(content);
1764
1765 if (sra)
1766 sysfs_free(sra);
1767 old_raid_disks = content->array.raid_disks - content->delta_disks;
1768 for (dev = content->devs; dev; dev = dev->next)
1769 if (sysfs_add_disk(content, dev, 1) == 0) {
1770 if (dev->disk.raid_disk >= old_raid_disks &&
1771 content->reshape_active)
1772 expansion++;
1773 else
1774 working++;
1775 } else if (errno == EEXIST)
1776 preexist++;
1777 if (working + expansion == 0)
1778 return 1;/* Nothing new, don't try to start */
1779
1780 map_update(&map, fd2devnm(mdfd),
1781 content->text_version,
1782 content->uuid, chosen_name);
1783
1784 if (c->runstop > 0 ||
1785 (working + preexist + expansion) >=
1786 content->array.working_disks) {
1787 int err;
1788
1789 if (start_reshape) {
1790 int spare = content->array.raid_disks + expansion;
1791 if (restore_backup(st, content,
1792 working,
1793 spare, c->backup_file, c->verbose) == 1)
1794 return 1;
1795
1796 err = sysfs_set_str(content, NULL,
1797 "array_state", "readonly");
1798 if (err)
1799 return 1;
1800
1801 if (st->ss->external) {
1802 if (!mdmon_running(st->container_devnm))
1803 start_mdmon(st->container_devnm);
1804 ping_monitor(st->container_devnm);
1805 if (mdmon_running(st->container_devnm) &&
1806 st->update_tail == NULL)
1807 st->update_tail = &st->updates;
1808 }
1809
1810 err = Grow_continue(mdfd, st, content, c->backup_file,
1811 c->freeze_reshape);
1812 } else switch(content->array.level) {
1813 case LEVEL_LINEAR:
1814 case LEVEL_MULTIPATH:
1815 case 0:
1816 err = sysfs_set_str(content, NULL, "array_state",
1817 c->readonly ? "readonly" : "active");
1818 break;
1819 default:
1820 err = sysfs_set_str(content, NULL, "array_state",
1821 "readonly");
1822 /* start mdmon if needed. */
1823 if (!err) {
1824 if (!mdmon_running(st->container_devnm))
1825 start_mdmon(st->container_devnm);
1826 ping_monitor(st->container_devnm);
1827 }
1828 break;
1829 }
1830 if (!err)
1831 sysfs_set_safemode(content, content->safe_mode_delay);
1832
1833 /* Block subarray here if it is not reshaped now
1834 * It has be blocked a little later to allow mdmon to switch in
1835 * in to R/W state
1836 */
1837 if (st->ss->external && content->recovery_blocked &&
1838 !start_reshape)
1839 block_subarray(content);
1840
1841 if (c->verbose >= 0) {
1842 if (err)
1843 pr_err("array %s now has %d device%s",
1844 chosen_name, working + preexist,
1845 working + preexist == 1 ? "":"s");
1846 else
1847 pr_err("Started %s with %d device%s",
1848 chosen_name, working + preexist,
1849 working + preexist == 1 ? "":"s");
1850 if (preexist)
1851 fprintf(stderr, " (%d new)", working);
1852 if (expansion)
1853 fprintf(stderr, " ( + %d for expansion)",
1854 expansion);
1855 fprintf(stderr, "\n");
1856 }
1857 if (!err)
1858 wait_for(chosen_name, mdfd);
1859 return err;
1860 /* FIXME should have an O_EXCL and wait for read-auto */
1861 } else {
1862 if (c->verbose >= 0) {
1863 pr_err("%s assembled with %d device%s",
1864 chosen_name, preexist + working,
1865 preexist + working == 1 ? "":"s");
1866 if (preexist)
1867 fprintf(stderr, " (%d new)", working);
1868 fprintf(stderr, " but not started\n");
1869 }
1870 return 1;
1871 }
1872 }
1873 #endif
1874
Unnamed repository; edit this file 'description' to name the repository.