]> 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
Assemble: split out force_array()
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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 {
140 struct mddev_dev *tmpdev;
141 int num_devs;
142 struct supertype *st = *stp;
143 struct mdinfo *content = NULL;
144 int report_missmatch = ((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->devices &&
175 !match_oneof(ident->devices, devname)) {
176 if (report_missmatch)
177 pr_err("%s is not one of %s\n", devname, ident->devices);
178 continue;
179 }
180
181 tst = dup_super(st);
182
183 dfd = dev_open(devname, O_RDONLY);
184 if (dfd < 0) {
185 if (report_missmatch)
186 pr_err("cannot open device %s: %s\n",
187 devname, strerror(errno));
188 tmpdev->used = 2;
189 } else if (fstat(dfd, &stb)< 0) {
190 /* Impossible! */
191 pr_err("fstat failed for %s: %s\n",
192 devname, strerror(errno));
193 tmpdev->used = 2;
194 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
195 pr_err("%s is not a block device.\n",
196 devname);
197 tmpdev->used = 2;
198 } else if (must_be_container(dfd)) {
199 if (st) {
200 /* already found some components, this cannot
201 * be another one.
202 */
203 if (report_missmatch)
204 pr_err("%s is a container, but we are looking for components\n",
205 devname);
206 tmpdev->used = 2;
207 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
208 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
209 if (report_missmatch)
210 pr_err("not a recognisable container: %s\n",
211 devname);
212 tmpdev->used = 2;
213 #endif
214 } else if (!tst->ss->load_container
215 || tst->ss->load_container(tst, dfd, NULL)) {
216 if (report_missmatch)
217 pr_err("no correct container type: %s\n",
218 devname);
219 tmpdev->used = 2;
220 } else if (auto_assem &&
221 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
222 tst->ss->match_home(tst, c->homehost) == 1)) {
223 if (report_missmatch)
224 pr_err("%s has metadata type %s for which "
225 "auto-assembly is disabled\n",
226 devname, tst->ss->name);
227 tmpdev->used = 2;
228 } else
229 found_container = 1;
230 } else {
231 if (!tst && (tst = guess_super(dfd)) == NULL) {
232 if (report_missmatch)
233 pr_err("no recogniseable superblock on %s\n",
234 devname);
235 tmpdev->used = 2;
236 } else if (tst->ss->load_super(tst,dfd, NULL)) {
237 if (report_missmatch)
238 pr_err("no RAID superblock on %s\n",
239 devname);
240 tmpdev->used = 2;
241 } else if (tst->ss->compare_super == NULL) {
242 if (report_missmatch)
243 pr_err("Cannot assemble %s metadata on %s\n",
244 tst->ss->name, devname);
245 tmpdev->used = 2;
246 } else if (auto_assem && st == NULL &&
247 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
248 tst->ss->match_home(tst, c->homehost) == 1)) {
249 if (report_missmatch)
250 pr_err("%s has metadata type %s for which "
251 "auto-assembly is disabled\n",
252 devname, tst->ss->name);
253 tmpdev->used = 2;
254 }
255 }
256 if (dfd >= 0) close(dfd);
257 if (tmpdev->used == 2) {
258 if (auto_assem || !inargv)
259 /* Ignore unrecognised devices during auto-assembly */
260 goto loop;
261 if (ident->uuid_set || ident->name[0] ||
262 ident->super_minor != UnSet)
263 /* Ignore unrecognised device if looking for
264 * specific array */
265 goto loop;
266
267 pr_err("%s has no superblock - assembly aborted\n",
268 devname);
269 if (st)
270 st->ss->free_super(st);
271 dev_policy_free(pol);
272 domain_free(domains);
273 return -1;
274 }
275
276 if (found_container) {
277 /* tmpdev is a container. We need to be either
278 * looking for a member, or auto-assembling
279 */
280 /* should be safe to try an exclusive open now, we
281 * have rejected anything that some other mdadm might
282 * be looking at
283 */
284 dfd = dev_open(devname, O_RDONLY | O_EXCL);
285 if (dfd < 0) {
286 if (report_missmatch)
287 pr_err("%s is busy - skipping\n", devname);
288 goto loop;
289 }
290 close(dfd);
291
292 if (ident->container) {
293 if (ident->container[0] == '/' &&
294 !same_dev(ident->container, devname)) {
295 if (report_missmatch)
296 pr_err("%s is not the container required (%s)\n",
297 devname, ident->container);
298 goto loop;
299 }
300 if (ident->container[0] != '/') {
301 /* we have a uuid */
302 int uuid[4];
303
304 content = *contentp;
305 tst->ss->getinfo_super(tst, content, NULL);
306
307 if (!parse_uuid(ident->container, uuid) ||
308 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
309 if (report_missmatch)
310 pr_err("%s has wrong UUID to be required container\n",
311 devname);
312 goto loop;
313 }
314 }
315 }
316 /* It is worth looking inside this container.
317 */
318 if (c->verbose > 0)
319 pr_err("looking in container %s\n",
320 devname);
321
322 for (content = tst->ss->container_content(tst, NULL);
323 content;
324 content = content->next) {
325
326 if (!ident_matches(ident, content, tst,
327 c->homehost, c->update,
328 report_missmatch ? devname : NULL))
329 /* message already printed */;
330 else if (is_member_busy(content->text_version)) {
331 if (report_missmatch)
332 pr_err("member %s in %s is already assembled\n",
333 content->text_version,
334 devname);
335 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
336 /* do not assemble arrays with unsupported configurations */
337 pr_err("Cannot activate member %s in %s.\n",
338 content->text_version,
339 devname);
340 } else
341 break;
342 }
343 if (!content) {
344 tmpdev->used = 2;
345 goto loop; /* empty container */
346 }
347
348 st = tst; tst = NULL;
349 if (!auto_assem && inargv && tmpdev->next != NULL) {
350 pr_err("%s is a container, but is not "
351 "only device given: confused and aborting\n",
352 devname);
353 st->ss->free_super(st);
354 dev_policy_free(pol);
355 domain_free(domains);
356 return -1;
357 }
358 if (c->verbose > 0)
359 pr_err("found match on member %s in %s\n",
360 content->text_version, devname);
361
362 /* make sure we finished the loop */
363 tmpdev = NULL;
364 goto loop;
365 } else {
366
367 content = *contentp;
368 tst->ss->getinfo_super(tst, content, NULL);
369
370 if (!ident_matches(ident, content, tst,
371 c->homehost, c->update,
372 report_missmatch ? devname : NULL))
373 goto loop;
374
375 /* should be safe to try an exclusive open now, we
376 * have rejected anything that some other mdadm might
377 * be looking at
378 */
379 dfd = dev_open(devname, O_RDONLY | O_EXCL);
380 if (dfd < 0) {
381 if (report_missmatch)
382 pr_err("%s is busy - skipping\n", devname);
383 goto loop;
384 }
385 close(dfd);
386
387 if (st == NULL)
388 st = dup_super(tst);
389 if (st->minor_version == -1)
390 st->minor_version = tst->minor_version;
391
392 if (memcmp(content->uuid, uuid_zero,
393 sizeof(int[4])) == 0) {
394 /* this is a floating spare. It cannot define
395 * an array unless there are no more arrays of
396 * this type to be found. It can be included
397 * in an array of this type though.
398 */
399 tmpdev->used = 3;
400 goto loop;
401 }
402
403 if (st->ss != tst->ss ||
404 st->minor_version != tst->minor_version ||
405 st->ss->compare_super(st, tst) != 0) {
406 /* Some mismatch. If exactly one array matches this host,
407 * we can resolve on that one.
408 * Or, if we are auto assembling, we just ignore the second
409 * for now.
410 */
411 if (auto_assem)
412 goto loop;
413 if (c->homehost) {
414 int first = st->ss->match_home(st, c->homehost);
415 int last = tst->ss->match_home(tst, c->homehost);
416 if (first != last &&
417 (first == 1 || last == 1)) {
418 /* We can do something */
419 if (first) {/* just ignore this one */
420 if (report_missmatch)
421 pr_err("%s misses out due to wrong homehost\n",
422 devname);
423 goto loop;
424 } else { /* reject all those sofar */
425 struct mddev_dev *td;
426 if (report_missmatch)
427 pr_err("%s overrides previous devices due to good homehost\n",
428 devname);
429 for (td=devlist; td != tmpdev; td=td->next)
430 if (td->used == 1)
431 td->used = 0;
432 tmpdev->used = 1;
433 goto loop;
434 }
435 }
436 }
437 pr_err("superblock on %s doesn't match others - assembly aborted\n",
438 devname);
439 tst->ss->free_super(tst);
440 st->ss->free_super(st);
441 dev_policy_free(pol);
442 domain_free(domains);
443 return -1;
444 }
445 tmpdev->used = 1;
446 }
447 loop:
448 /* Collect domain information from members only */
449 if (tmpdev && tmpdev->used == 1) {
450 if (!pol)
451 pol = devnum_policy(stb.st_rdev);
452 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
453 }
454 dev_policy_free(pol);
455 pol = NULL;
456 if (tst)
457 tst->ss->free_super(tst);
458 }
459
460 /* Check if we found some imsm spares but no members */
461 if ((auto_assem ||
462 (ident->uuid_set &&
463 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
464 (!st || !st->sb))
465 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
466 if (tmpdev->used != 3)
467 continue;
468 tmpdev->used = 1;
469 content = *contentp;
470
471 if (!st->sb) {
472 /* we need sb from one of the spares */
473 int dfd = dev_open(tmpdev->devname, O_RDONLY);
474 if (dfd < 0 ||
475 st->ss->load_super(st, dfd, NULL))
476 tmpdev->used = 2;
477 if (dfd > 0)
478 close(dfd);
479 }
480 }
481
482 /* Now reject spares that don't match domains of identified members */
483 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
484 struct stat stb;
485 if (tmpdev->used != 3)
486 continue;
487 if (stat(tmpdev->devname, &stb)< 0) {
488 pr_err("fstat failed for %s: %s\n",
489 tmpdev->devname, strerror(errno));
490 tmpdev->used = 2;
491 } else {
492 struct dev_policy *pol = devnum_policy(stb.st_rdev);
493 int dt = domain_test(domains, pol, NULL);
494 if (inargv && dt != 0)
495 /* take this spare as domains match
496 * if there are any */
497 tmpdev->used = 1;
498 else if (!inargv && dt == 1)
499 /* device wasn't explicitly listed, so need
500 * explicit domain match - which we have */
501 tmpdev->used = 1;
502 else
503 /* if domains don't match mark as unused */
504 tmpdev->used = 0;
505 dev_policy_free(pol);
506 }
507 }
508 domain_free(domains);
509 *stp = st;
510 if (st && st->sb && content == *contentp)
511 st->ss->getinfo_super(st, content, NULL);
512 *contentp = content;
513
514 return num_devs;
515 }
516
517 struct devs {
518 char *devname;
519 int uptodate; /* set once we decide that this device is as
520 * recent as everything else in the array.
521 */
522 int included; /* set if the device is already in the array
523 * due to a previous '-I'
524 */
525 struct mdinfo i;
526 };
527
528 static int load_devices(struct devs *devices, char *devmap,
529 struct mddev_ident *ident, struct supertype *st,
530 struct mddev_dev *devlist, struct context *c,
531 struct mdinfo *content,
532 int mdfd, char *mddev,
533 int *most_recentp, int *bestcntp, int **bestp,
534 int inargv)
535 {
536 struct mddev_dev *tmpdev;
537 int devcnt = 0;
538 int nextspare = 0;
539 #ifndef MDASSEMBLE
540 int bitmap_done = 0;
541 #endif
542 int most_recent = 0;
543 int bestcnt = 0;
544 int *best = *bestp;
545
546 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
547 char *devname = tmpdev->devname;
548 struct stat stb;
549 int i;
550
551 if (tmpdev->used != 1)
552 continue;
553 /* looks like a good enough match to update the super block if needed */
554 #ifndef MDASSEMBLE
555 if (c->update) {
556 int dfd;
557 /* prepare useful information in info structures */
558 struct stat stb2;
559 struct supertype *tst;
560 int err;
561 fstat(mdfd, &stb2);
562
563 if (strcmp(c->update, "uuid")==0 &&
564 !ident->uuid_set) {
565 int rfd;
566 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
567 read(rfd, ident->uuid, 16) != 16) {
568 *(__u32*)(ident->uuid) = random();
569 *(__u32*)(ident->uuid+1) = random();
570 *(__u32*)(ident->uuid+2) = random();
571 *(__u32*)(ident->uuid+3) = random();
572 }
573 if (rfd >= 0) close(rfd);
574 }
575 dfd = dev_open(devname,
576 tmpdev->disposition == 'I'
577 ? O_RDWR : (O_RDWR|O_EXCL));
578
579 tst = dup_super(st);
580 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
581 pr_err("cannot re-read metadata from %s - aborting\n",
582 devname);
583 if (dfd >= 0)
584 close(dfd);
585 close(mdfd);
586 free(devices);
587 free(devmap);
588 return -1;
589 }
590 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
591
592 memcpy(content->uuid, ident->uuid, 16);
593 strcpy(content->name, ident->name);
594 content->array.md_minor = minor(stb2.st_rdev);
595
596 if (strcmp(c->update, "byteorder") == 0)
597 err = 0;
598 else
599 err = tst->ss->update_super(tst, content, c->update,
600 devname, c->verbose,
601 ident->uuid_set,
602 c->homehost);
603 if (err < 0) {
604 pr_err("--update=%s not understood"
605 " for %s metadata\n",
606 c->update, tst->ss->name);
607 tst->ss->free_super(tst);
608 free(tst);
609 close(mdfd);
610 close(dfd);
611 free(devices);
612 free(devmap);
613 return -1;
614 }
615 if (strcmp(c->update, "uuid")==0 &&
616 !ident->uuid_set) {
617 ident->uuid_set = 1;
618 memcpy(ident->uuid, content->uuid, 16);
619 }
620 if (tst->ss->store_super(tst, dfd))
621 pr_err("Could not re-write superblock on %s.\n",
622 devname);
623 close(dfd);
624
625 if (strcmp(c->update, "uuid")==0 &&
626 ident->bitmap_fd >= 0 && !bitmap_done) {
627 if (bitmap_update_uuid(ident->bitmap_fd,
628 content->uuid,
629 tst->ss->swapuuid) != 0)
630 pr_err("Could not update uuid on external bitmap.\n");
631 else
632 bitmap_done = 1;
633 }
634 tst->ss->free_super(tst);
635 } else
636 #endif
637 {
638 struct supertype *tst = dup_super(st);
639 int dfd;
640 dfd = dev_open(devname,
641 tmpdev->disposition == 'I'
642 ? O_RDWR : (O_RDWR|O_EXCL));
643
644 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
645 pr_err("cannot re-read metadata from %s - aborting\n",
646 devname);
647 if (dfd >= 0)
648 close(dfd);
649 close(mdfd);
650 free(devices);
651 free(devmap);
652 return -1;
653 }
654 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
655 tst->ss->free_super(tst);
656 close(dfd);
657 }
658
659 stat(devname, &stb);
660
661 if (c->verbose > 0)
662 pr_err("%s is identified as a member of %s, slot %d.\n",
663 devname, mddev, content->disk.raid_disk);
664 devices[devcnt].devname = devname;
665 devices[devcnt].uptodate = 0;
666 devices[devcnt].included = (tmpdev->disposition == 'I');
667 devices[devcnt].i = *content;
668 devices[devcnt].i.disk.major = major(stb.st_rdev);
669 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
670 if (most_recent < devcnt) {
671 if (devices[devcnt].i.events
672 > devices[most_recent].i.events)
673 most_recent = devcnt;
674 }
675 if (content->array.level == LEVEL_MULTIPATH)
676 /* with multipath, the raid_disk from the superblock is meaningless */
677 i = devcnt;
678 else
679 i = devices[devcnt].i.disk.raid_disk;
680 if (i+1 == 0) {
681 if (nextspare < content->array.raid_disks)
682 nextspare = content->array.raid_disks;
683 i = nextspare++;
684 } else {
685 if (i >= content->array.raid_disks &&
686 i >= nextspare)
687 nextspare = i+1;
688 }
689 if (i < 10000) {
690 if (i >= bestcnt) {
691 int newbestcnt = i+10;
692 int *newbest = xmalloc(sizeof(int)*newbestcnt);
693 int c;
694 for (c=0; c < newbestcnt; c++)
695 if (c < bestcnt)
696 newbest[c] = best[c];
697 else
698 newbest[c] = -1;
699 if (best)free(best);
700 best = newbest;
701 bestcnt = newbestcnt;
702 }
703 if (best[i] >=0 &&
704 devices[best[i]].i.events
705 == devices[devcnt].i.events
706 && (devices[best[i]].i.disk.minor
707 != devices[devcnt].i.disk.minor)
708 && st->ss == &super0
709 && content->array.level != LEVEL_MULTIPATH) {
710 /* two different devices with identical superblock.
711 * Could be a mis-detection caused by overlapping
712 * partitions. fail-safe.
713 */
714 pr_err("WARNING %s and %s appear"
715 " to have very similar superblocks.\n"
716 " If they are really different, "
717 "please --zero the superblock on one\n"
718 " If they are the same or overlap,"
719 " please remove one from %s.\n",
720 devices[best[i]].devname, devname,
721 inargv ? "the list" :
722 "the\n DEVICE list in mdadm.conf"
723 );
724 close(mdfd);
725 free(devices);
726 free(devmap);
727 return -1;
728 }
729 if (best[i] == -1
730 || (devices[best[i]].i.events
731 < devices[devcnt].i.events))
732 best[i] = devcnt;
733 }
734 devcnt++;
735 }
736 *most_recentp = most_recent;
737 *bestcntp = bestcnt;
738 *bestp = best;
739 return devcnt;
740 }
741
742 static int force_array(struct mdinfo *content,
743 struct devs *devices,
744 int *best, int bestcnt, char *avail,
745 int most_recent,
746 struct supertype *st,
747 struct context *c)
748 {
749 int okcnt = 0;
750 while (!enough(content->array.level, content->array.raid_disks,
751 content->array.layout, 1,
752 avail)
753 ||
754 (content->reshape_active && content->delta_disks > 0 &&
755 !enough(content->array.level, (content->array.raid_disks
756 - content->delta_disks),
757 content->new_layout, 1,
758 avail)
759 )) {
760 /* Choose the newest best drive which is
761 * not up-to-date, update the superblock
762 * and add it.
763 */
764 int fd;
765 struct supertype *tst;
766 unsigned long long current_events;
767 int chosen_drive = -1;
768 int i;
769
770 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
771 int j = best[i];
772 if (j>=0 &&
773 !devices[j].uptodate &&
774 devices[j].i.recovery_start == MaxSector &&
775 (chosen_drive < 0 ||
776 devices[j].i.events
777 > devices[chosen_drive].i.events))
778 chosen_drive = j;
779 }
780 if (chosen_drive < 0)
781 break;
782 current_events = devices[chosen_drive].i.events;
783 add_another:
784 if (c->verbose >= 0)
785 pr_err("forcing event count in %s(%d) from %d upto %d\n",
786 devices[chosen_drive].devname,
787 devices[chosen_drive].i.disk.raid_disk,
788 (int)(devices[chosen_drive].i.events),
789 (int)(devices[most_recent].i.events));
790 fd = dev_open(devices[chosen_drive].devname,
791 devices[chosen_drive].included ? O_RDWR
792 : (O_RDWR|O_EXCL));
793 if (fd < 0) {
794 pr_err("Couldn't open %s for write - not updating\n",
795 devices[chosen_drive].devname);
796 devices[chosen_drive].i.events = 0;
797 continue;
798 }
799 tst = dup_super(st);
800 if (tst->ss->load_super(tst,fd, NULL)) {
801 close(fd);
802 pr_err("RAID superblock disappeared from %s - not updating.\n",
803 devices[chosen_drive].devname);
804 devices[chosen_drive].i.events = 0;
805 continue;
806 }
807 content->events = devices[most_recent].i.events;
808 tst->ss->update_super(tst, content, "force-one",
809 devices[chosen_drive].devname, c->verbose,
810 0, NULL);
811
812 if (tst->ss->store_super(tst, fd)) {
813 close(fd);
814 pr_err("Could not re-write superblock on %s\n",
815 devices[chosen_drive].devname);
816 devices[chosen_drive].i.events = 0;
817 tst->ss->free_super(tst);
818 continue;
819 }
820 close(fd);
821 devices[chosen_drive].i.events = devices[most_recent].i.events;
822 devices[chosen_drive].uptodate = 1;
823 avail[chosen_drive] = 1;
824 okcnt++;
825 tst->ss->free_super(tst);
826
827 /* If there are any other drives of the same vintage,
828 * add them in as well. We can't lose and we might gain
829 */
830 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
831 int j = best[i];
832 if (j >= 0 &&
833 !devices[j].uptodate &&
834 devices[j].i.recovery_start == MaxSector &&
835 devices[j].i.events == current_events) {
836 chosen_drive = j;
837 goto add_another;
838 }
839 }
840 }
841 return okcnt;
842 }
843
844 int Assemble(struct supertype *st, char *mddev,
845 struct mddev_ident *ident,
846 struct mddev_dev *devlist,
847 struct context *c)
848 {
849 /*
850 * The task of Assemble is to find a collection of
851 * devices that should (according to their superblocks)
852 * form an array, and to give this collection to the MD driver.
853 * In Linux-2.4 and later, this involves submitting a
854 * SET_ARRAY_INFO ioctl with no arg - to prepare
855 * the array - and then submit a number of
856 * ADD_NEW_DISK ioctls to add disks into
857 * the array. Finally RUN_ARRAY might
858 * be submitted to start the array.
859 *
860 * Much of the work of Assemble is in finding and/or
861 * checking the disks to make sure they look right.
862 *
863 * If mddev is not set, then scan must be set and we
864 * read through the config file for dev+uuid mapping
865 * We recurse, setting mddev, for each device that
866 * - isn't running
867 * - has a valid uuid (or any uuid if !uuidset)
868 *
869 * If mddev is set, we try to determine state of md.
870 * check version - must be at least 0.90.0
871 * check kernel version. must be at least 2.4.
872 * If not, we can possibly fall back on START_ARRAY
873 * Try to GET_ARRAY_INFO.
874 * If possible, give up
875 * If not, try to STOP_ARRAY just to make sure
876 *
877 * If !uuidset and scan, look in conf-file for uuid
878 * If not found, give up
879 * If !devlist and scan and uuidset, get list of devs from conf-file
880 *
881 * For each device:
882 * Check superblock - discard if bad
883 * Check uuid (set if we don't have one) - discard if no match
884 * Check superblock similarity if we have a superblock - discard if different
885 * Record events, devicenum
886 * This should give us a list of devices for the array
887 * We should collect the most recent event number
888 *
889 * Count disks with recent enough event count
890 * While force && !enough disks
891 * Choose newest rejected disks, update event count
892 * mark clean and rewrite superblock
893 * If recent kernel:
894 * SET_ARRAY_INFO
895 * foreach device with recent events : ADD_NEW_DISK
896 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
897 * If old kernel:
898 * Check the device numbers in superblock are right
899 * update superblock if any changes
900 * START_ARRAY
901 *
902 */
903 int mdfd;
904 int clean;
905 int auto_assem = (mddev == NULL && !ident->uuid_set &&
906 ident->super_minor == UnSet && ident->name[0] == 0
907 && (ident->container == NULL || ident->member == NULL));
908 int old_linux = 0;
909 int vers = vers; /* Keep gcc quite - it really is initialised */
910 struct devs *devices;
911 char *devmap;
912 int *best = NULL; /* indexed by raid_disk */
913 int bestcnt = 0;
914 int devcnt;
915 unsigned int okcnt, sparecnt, rebuilding_cnt;
916 unsigned int req_cnt;
917 int i;
918 int most_recent = 0;
919 int chosen_drive;
920 int change = 0;
921 int inargv = 0;
922 int start_partial_ok = (c->runstop >= 0) &&
923 (c->force || devlist==NULL || auto_assem);
924 int num_devs;
925 struct mddev_dev *tmpdev;
926 struct mdinfo info;
927 struct mdinfo *content = NULL;
928 struct mdinfo *pre_exist = NULL;
929 char *avail;
930 char *name = NULL;
931 char chosen_name[1024];
932 struct map_ent *map = NULL;
933 struct map_ent *mp;
934
935 if (get_linux_version() < 2004000)
936 old_linux = 1;
937
938 /*
939 * If any subdevs are listed, then any that don't
940 * match ident are discarded. Remainder must all match and
941 * become the array.
942 * If no subdevs, then we scan all devices in the config file, but
943 * there must be something in the identity
944 */
945
946 if (!devlist &&
947 ident->uuid_set == 0 &&
948 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
949 ident->name[0] == 0 &&
950 (ident->container == NULL || ident->member == NULL) &&
951 ident->devices == NULL) {
952 pr_err("No identity information available for %s - cannot assemble.\n",
953 mddev ? mddev : "further assembly");
954 return 1;
955 }
956
957 if (devlist == NULL)
958 devlist = conf_get_devs();
959 else if (mddev)
960 inargv = 1;
961
962 try_again:
963 /* We come back here when doing auto-assembly and attempting some
964 * set of devices failed. Those are now marked as ->used==2 and
965 * we ignore them and try again
966 */
967 if (!st && ident->st)
968 st = ident->st;
969 if (c->verbose>0)
970 pr_err("looking for devices for %s\n",
971 mddev ? mddev : "further assembly");
972
973 content = &info;
974 num_devs = select_devices(devlist, ident, &st, &content, c,
975 inargv, auto_assem);
976 if (num_devs < 0)
977 return 1;
978
979 if (!st || !st->sb || !content)
980 return 2;
981
982 /* We have a full set of devices - we now need to find the
983 * array device.
984 * However there is a risk that we are racing with "mdadm -I"
985 * and the array is already partially assembled - we will have
986 * rejected any devices already in this address.
987 * So we take a lock on the map file - to prevent further races -
988 * and look for the uuid in there. If found and the array is
989 * active, we abort. If found and the array is not active
990 * we commit to that md device and add all the contained devices
991 * to our list. We flag them so that we don't try to re-add,
992 * but can remove if they turn out to not be wanted.
993 */
994 if (map_lock(&map))
995 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
996 mp = map_by_uuid(&map, content->uuid);
997 if (mp) {
998 struct mdinfo *dv;
999 /* array already exists. */
1000 pre_exist = sysfs_read(-1, mp->devnum, GET_LEVEL|GET_DEVS);
1001 if (pre_exist->array.level != UnSet) {
1002 pr_err("Found some drive for an array that is already active: %s\n",
1003 mp->path);
1004 pr_err("giving up.\n");
1005 return 1;
1006 }
1007 for (dv = pre_exist->devs; dv; dv = dv->next) {
1008 /* We want to add this device to our list,
1009 * but it could already be there if "mdadm -I"
1010 * started *after* we checked for O_EXCL.
1011 * If we add it to the top of the list
1012 * it will be preferred over later copies.
1013 */
1014 struct mddev_dev *newdev;
1015 char *devname = map_dev(dv->disk.major,
1016 dv->disk.minor,
1017 0);
1018 if (!devname)
1019 continue;
1020 newdev = xmalloc(sizeof(*newdev));
1021 newdev->devname = devname;
1022 newdev->disposition = 'I';
1023 newdev->used = 1;
1024 newdev->next = devlist;
1025 devlist = newdev;
1026 num_devs++;
1027 }
1028 strcpy(chosen_name, mp->path);
1029 if (c->verbose > 0 || mddev == NULL ||
1030 strcmp(mddev, chosen_name) != 0)
1031 pr_err("Merging with already-assembled %s\n",
1032 chosen_name);
1033 mdfd = open_dev_excl(mp->devnum);
1034 } else {
1035 int trustworthy = FOREIGN;
1036 name = content->name;
1037 switch (st->ss->match_home(st, c->homehost)
1038 ?: st->ss->match_home(st, "any")) {
1039 case 1:
1040 trustworthy = LOCAL;
1041 name = strchr(content->name, ':');
1042 if (name)
1043 name++;
1044 else
1045 name = content->name;
1046 break;
1047 }
1048 if (!auto_assem)
1049 /* If the array is listed in mdadm.conf or on
1050 * command line, then we trust the name
1051 * even if the array doesn't look local
1052 */
1053 trustworthy = LOCAL;
1054
1055 if (name[0] == 0 &&
1056 content->array.level == LEVEL_CONTAINER) {
1057 name = content->text_version;
1058 trustworthy = METADATA;
1059 }
1060
1061 if (name[0] && trustworthy != LOCAL &&
1062 ! c->require_homehost &&
1063 conf_name_is_free(name))
1064 trustworthy = LOCAL;
1065
1066 if (trustworthy == LOCAL &&
1067 strchr(name, ':'))
1068 /* Ignore 'host:' prefix of name */
1069 name = strchr(name, ':')+1;
1070
1071 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1072 chosen_name);
1073 }
1074 if (mdfd < 0) {
1075 st->ss->free_super(st);
1076 if (auto_assem)
1077 goto try_again;
1078 return 1;
1079 }
1080 mddev = chosen_name;
1081 vers = md_get_version(mdfd);
1082 if (vers < 9000) {
1083 pr_err("Assemble requires driver version 0.90.0 or later.\n"
1084 " Upgrade your kernel or try --build\n");
1085 close(mdfd);
1086 return 1;
1087 }
1088 if (pre_exist == NULL) {
1089 if (mddev_busy(fd2devnum(mdfd))) {
1090 pr_err("%s already active, cannot restart it!\n",
1091 mddev);
1092 for (tmpdev = devlist ;
1093 tmpdev && tmpdev->used != 1;
1094 tmpdev = tmpdev->next)
1095 ;
1096 if (tmpdev && auto_assem)
1097 pr_err("%s needed for %s...\n",
1098 mddev, tmpdev->devname);
1099 close(mdfd);
1100 mdfd = -3;
1101 st->ss->free_super(st);
1102 if (auto_assem)
1103 goto try_again;
1104 return 1;
1105 }
1106 /* just incase it was started but has no content */
1107 ioctl(mdfd, STOP_ARRAY, NULL);
1108 }
1109
1110 #ifndef MDASSEMBLE
1111 if (content != &info) {
1112 /* This is a member of a container. Try starting the array. */
1113 int err;
1114 err = assemble_container_content(st, mdfd, content, c,
1115 chosen_name);
1116 close(mdfd);
1117 return err;
1118 }
1119 #endif
1120 /* Ok, no bad inconsistancy, we can try updating etc */
1121 devices = xcalloc(num_devs, sizeof(*devices));
1122 devmap = xcalloc(num_devs, content->array.raid_disks);
1123 devcnt = load_devices(devices, devmap, ident, st, devlist,
1124 c, content, mdfd, mddev,
1125 &most_recent, &bestcnt, &best, inargv);
1126 if (devcnt < 0)
1127 return 1;
1128
1129 if (devcnt == 0) {
1130 pr_err("no devices found for %s\n",
1131 mddev);
1132 if (st)
1133 st->ss->free_super(st);
1134 close(mdfd);
1135 free(devices);
1136 free(devmap);
1137 return 1;
1138 }
1139
1140 if (c->update && strcmp(c->update, "byteorder")==0)
1141 st->minor_version = 90;
1142
1143 st->ss->getinfo_super(st, content, NULL);
1144 clean = content->array.state & 1;
1145
1146 /* now we have some devices that might be suitable.
1147 * I wonder how many
1148 */
1149 avail = xcalloc(content->array.raid_disks, 1);
1150 okcnt = 0;
1151 sparecnt=0;
1152 rebuilding_cnt=0;
1153 for (i=0; i< bestcnt; i++) {
1154 int j = best[i];
1155 int event_margin = 1; /* always allow a difference of '1'
1156 * like the kernel does
1157 */
1158 if (j < 0) continue;
1159 /* note: we ignore error flags in multipath arrays
1160 * as they don't make sense
1161 */
1162 if (content->array.level != LEVEL_MULTIPATH)
1163 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1164 if (!(devices[j].i.disk.state
1165 & (1<<MD_DISK_FAULTY))) {
1166 devices[j].uptodate = 1;
1167 sparecnt++;
1168 }
1169 continue;
1170 }
1171 /* If this device thinks that 'most_recent' has failed, then
1172 * we must reject this device.
1173 */
1174 if (j != most_recent &&
1175 content->array.raid_disks > 0 &&
1176 devices[most_recent].i.disk.raid_disk >= 0 &&
1177 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1178 if (c->verbose > -1)
1179 pr_err("ignoring %s as it reports %s as failed\n",
1180 devices[j].devname, devices[most_recent].devname);
1181 best[i] = -1;
1182 continue;
1183 }
1184 if (devices[j].i.events+event_margin >=
1185 devices[most_recent].i.events) {
1186 devices[j].uptodate = 1;
1187 if (i < content->array.raid_disks) {
1188 if (devices[j].i.recovery_start == MaxSector ||
1189 (content->reshape_active &&
1190 ((i >= content->array.raid_disks - content->delta_disks) ||
1191 (i >= content->array.raid_disks - content->delta_disks - 1
1192 && content->array.level == 4)))) {
1193 okcnt++;
1194 avail[i]=1;
1195 } else
1196 rebuilding_cnt++;
1197 } else
1198 sparecnt++;
1199 }
1200 }
1201 free(devmap);
1202 if (c->force)
1203 okcnt += force_array(content, devices, best, bestcnt,
1204 avail, most_recent, st, c);
1205
1206 /* Now we want to look at the superblock which the kernel will base things on
1207 * and compare the devices that we think are working with the devices that the
1208 * superblock thinks are working.
1209 * If there are differences and --force is given, then update this chosen
1210 * superblock.
1211 */
1212 chosen_drive = -1;
1213 st->ss->free_super(st);
1214 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
1215 int j = best[i];
1216 int fd;
1217
1218 if (j<0)
1219 continue;
1220 if (!devices[j].uptodate)
1221 continue;
1222 if (devices[j].i.events < devices[most_recent].i.events)
1223 continue;
1224 chosen_drive = j;
1225 if ((fd=dev_open(devices[j].devname,
1226 devices[j].included ? O_RDONLY
1227 : (O_RDONLY|O_EXCL)))< 0) {
1228 pr_err("Cannot open %s: %s\n",
1229 devices[j].devname, strerror(errno));
1230 close(mdfd);
1231 free(devices);
1232 return 1;
1233 }
1234 if (st->ss->load_super(st,fd, NULL)) {
1235 close(fd);
1236 pr_err("RAID superblock has disappeared from %s\n",
1237 devices[j].devname);
1238 close(mdfd);
1239 free(devices);
1240 return 1;
1241 }
1242 close(fd);
1243 }
1244 if (st->sb == NULL) {
1245 pr_err("No suitable drives found for %s\n", mddev);
1246 close(mdfd);
1247 free(devices);
1248 return 1;
1249 }
1250 st->ss->getinfo_super(st, content, NULL);
1251 #ifndef MDASSEMBLE
1252 sysfs_init(content, mdfd, 0);
1253 #endif
1254 for (i=0; i<bestcnt; i++) {
1255 int j = best[i];
1256 unsigned int desired_state;
1257
1258 if (i < content->array.raid_disks)
1259 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1260 else
1261 desired_state = 0;
1262
1263 if (j<0)
1264 continue;
1265 if (!devices[j].uptodate)
1266 continue;
1267
1268 devices[j].i.disk.state = desired_state;
1269 if (!(devices[j].i.array.state & 1))
1270 clean = 0;
1271
1272 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1273 c->verbose, 0, NULL)) {
1274 if (c->force) {
1275 if (c->verbose >= 0)
1276 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1277 j, mddev, devices[j].devname);
1278 change = 1;
1279 } else {
1280 if (c->verbose >= -1)
1281 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1282 i, mddev, devices[j].devname);
1283 }
1284 }
1285 #if 0
1286 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1287 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1288 i, mddev);
1289 }
1290 #endif
1291 }
1292 if (c->force && !clean &&
1293 !enough(content->array.level, content->array.raid_disks,
1294 content->array.layout, clean,
1295 avail)) {
1296 change += st->ss->update_super(st, content, "force-array",
1297 devices[chosen_drive].devname, c->verbose,
1298 0, NULL);
1299 clean = 1;
1300 }
1301
1302 if (change) {
1303 int fd;
1304 fd = dev_open(devices[chosen_drive].devname,
1305 devices[chosen_drive].included ?
1306 O_RDWR : (O_RDWR|O_EXCL));
1307 if (fd < 0) {
1308 pr_err("Could not open %s for write - cannot Assemble array.\n",
1309 devices[chosen_drive].devname);
1310 close(mdfd);
1311 free(devices);
1312 return 1;
1313 }
1314 if (st->ss->store_super(st, fd)) {
1315 close(fd);
1316 pr_err("Could not re-write superblock on %s\n",
1317 devices[chosen_drive].devname);
1318 close(mdfd);
1319 free(devices);
1320 return 1;
1321 }
1322 if (c->verbose >= 0)
1323 pr_err("Marking array %s as 'clean'\n",
1324 mddev);
1325 close(fd);
1326 }
1327
1328 /* If we are in the middle of a reshape we may need to restore saved data
1329 * that was moved aside due to the reshape overwriting live data
1330 * The code of doing this lives in Grow.c
1331 */
1332 #ifndef MDASSEMBLE
1333 if (content->reshape_active &&
1334 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1335 int err = 0;
1336 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1337 if (c->verbose > 0)
1338 pr_err(":%s has an active reshape - checking "
1339 "if critical section needs to be restored\n",
1340 chosen_name);
1341 for (i=0; i<bestcnt; i++) {
1342 int j = best[i];
1343 if (j >= 0) {
1344 fdlist[i] = dev_open(devices[j].devname,
1345 devices[j].included
1346 ? O_RDWR : (O_RDWR|O_EXCL));
1347 if (fdlist[i] < 0) {
1348 pr_err("Could not open %s for write - cannot Assemble array.\n",
1349 devices[j].devname);
1350 err = 1;
1351 break;
1352 }
1353 } else
1354 fdlist[i] = -1;
1355 }
1356 if (!err) {
1357 if (st->ss->external && st->ss->recover_backup)
1358 err = st->ss->recover_backup(st, content);
1359 else
1360 err = Grow_restart(st, content, fdlist, bestcnt,
1361 c->backup_file, c->verbose > 0);
1362 if (err && c->invalid_backup) {
1363 if (c->verbose > 0)
1364 pr_err("continuing"
1365 " without restoring backup\n");
1366 err = 0;
1367 }
1368 }
1369 while (i>0) {
1370 i--;
1371 if (fdlist[i]>=0) close(fdlist[i]);
1372 }
1373 free(fdlist);
1374 if (err) {
1375 pr_err("Failed to restore critical section for reshape, sorry.\n");
1376 if (c->backup_file == NULL)
1377 cont_err("Possibly you needed to specify the --backup-file\n");
1378 close(mdfd);
1379 free(devices);
1380 return err;
1381 }
1382 }
1383 #endif
1384 /* count number of in-sync devices according to the superblock.
1385 * We must have this number to start the array without -s or -R
1386 */
1387 req_cnt = content->array.working_disks;
1388
1389 /* Almost ready to actually *do* something */
1390 if (!old_linux) {
1391 int rv;
1392
1393 /* First, fill in the map, so that udev can find our name
1394 * as soon as we become active.
1395 */
1396 map_update(&map, fd2devnum(mdfd), content->text_version,
1397 content->uuid, chosen_name);
1398
1399 rv = set_array_info(mdfd, st, content);
1400 if (rv && !pre_exist) {
1401 pr_err("failed to set array info for %s: %s\n",
1402 mddev, strerror(errno));
1403 ioctl(mdfd, STOP_ARRAY, NULL);
1404 close(mdfd);
1405 free(devices);
1406 map_unlock(&map);
1407 return 1;
1408 }
1409 if (ident->bitmap_fd >= 0) {
1410 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1411 pr_err("SET_BITMAP_FILE failed.\n");
1412 ioctl(mdfd, STOP_ARRAY, NULL);
1413 close(mdfd);
1414 free(devices);
1415 map_unlock(&map);
1416 return 1;
1417 }
1418 } else if (ident->bitmap_file) {
1419 /* From config file */
1420 int bmfd = open(ident->bitmap_file, O_RDWR);
1421 if (bmfd < 0) {
1422 pr_err("Could not open bitmap file %s\n",
1423 ident->bitmap_file);
1424 ioctl(mdfd, STOP_ARRAY, NULL);
1425 close(mdfd);
1426 free(devices);
1427 map_unlock(&map);
1428 return 1;
1429 }
1430 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1431 pr_err("Failed to set bitmapfile for %s\n", mddev);
1432 close(bmfd);
1433 ioctl(mdfd, STOP_ARRAY, NULL);
1434 close(mdfd);
1435 free(devices);
1436 map_unlock(&map);
1437 return 1;
1438 }
1439 close(bmfd);
1440 }
1441
1442 /* First, add the raid disks, but add the chosen one last */
1443 for (i=0; i<= bestcnt; i++) {
1444 int j;
1445 if (i < bestcnt) {
1446 j = best[i];
1447 if (j == chosen_drive)
1448 continue;
1449 } else
1450 j = chosen_drive;
1451
1452 if (j >= 0 && !devices[j].included) {
1453 int dfd = dev_open(devices[j].devname,
1454 O_RDWR|O_EXCL);
1455 if (dfd >= 0) {
1456 remove_partitions(dfd);
1457 close(dfd);
1458 }
1459 rv = add_disk(mdfd, st, content, &devices[j].i);
1460
1461 if (rv) {
1462 pr_err("failed to add "
1463 "%s to %s: %s\n",
1464 devices[j].devname,
1465 mddev,
1466 strerror(errno));
1467 if (i < content->array.raid_disks
1468 || i == bestcnt)
1469 okcnt--;
1470 else
1471 sparecnt--;
1472 } else if (c->verbose > 0)
1473 pr_err("added %s to %s as %d%s\n",
1474 devices[j].devname, mddev,
1475 devices[j].i.disk.raid_disk,
1476 devices[j].uptodate?"":
1477 " (possibly out of date)");
1478 } else if (j >= 0) {
1479 if (c->verbose > 0)
1480 pr_err("%s is already in %s as %d\n",
1481 devices[j].devname, mddev,
1482 devices[j].i.disk.raid_disk);
1483 } else if (c->verbose > 0 && i < content->array.raid_disks)
1484 pr_err("no uptodate device for slot %d of %s\n",
1485 i, mddev);
1486 }
1487
1488 if (content->array.level == LEVEL_CONTAINER) {
1489 if (c->verbose >= 0) {
1490 pr_err("Container %s has been "
1491 "assembled with %d drive%s",
1492 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1493 if (okcnt < (unsigned)content->array.raid_disks)
1494 fprintf(stderr, " (out of %d)",
1495 content->array.raid_disks);
1496 fprintf(stderr, "\n");
1497 }
1498 st->ss->free_super(st);
1499 sysfs_uevent(content, "change");
1500 map_unlock(&map);
1501 wait_for(chosen_name, mdfd);
1502 close(mdfd);
1503 free(devices);
1504 return 0;
1505 }
1506
1507 if (c->runstop == 1 ||
1508 (c->runstop <= 0 &&
1509 ( enough(content->array.level, content->array.raid_disks,
1510 content->array.layout, clean, avail) &&
1511 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1512 ))) {
1513 /* This array is good-to-go.
1514 * If a reshape is in progress then we might need to
1515 * continue monitoring it. In that case we start
1516 * it read-only and let the grow code make it writable.
1517 */
1518 int rv;
1519 #ifndef MDASSEMBLE
1520 if (content->reshape_active &&
1521 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1522 content->delta_disks <= 0) {
1523 rv = sysfs_set_str(content, NULL,
1524 "array_state", "readonly");
1525 if (rv == 0)
1526 rv = Grow_continue(mdfd, st, content,
1527 c->backup_file,
1528 c->freeze_reshape);
1529 } else if (c->readonly &&
1530 sysfs_attribute_available(
1531 content, NULL, "array_state")) {
1532 rv = sysfs_set_str(content, NULL,
1533 "array_state", "readonly");
1534 } else
1535 #endif
1536 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1537 if (rv == 0) {
1538 if (c->verbose >= 0) {
1539 pr_err("%s has been started with %d drive%s",
1540 mddev, okcnt, okcnt==1?"":"s");
1541 if (okcnt < (unsigned)content->array.raid_disks)
1542 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1543 if (rebuilding_cnt)
1544 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1545 if (sparecnt)
1546 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1547 fprintf(stderr, ".\n");
1548 }
1549 if (content->reshape_active &&
1550 content->array.level >= 4 &&
1551 content->array.level <= 6) {
1552 /* might need to increase the size
1553 * of the stripe cache - default is 256
1554 */
1555 if (256 < 4 * (content->array.chunk_size/4096)) {
1556 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1557 if (sra)
1558 sysfs_set_num(sra, NULL,
1559 "stripe_cache_size",
1560 (4 * content->array.chunk_size / 4096) + 1);
1561 sysfs_free(sra);
1562 }
1563 }
1564 if (okcnt < (unsigned)content->array.raid_disks) {
1565 /* If any devices did not get added
1566 * because the kernel rejected them based
1567 * on event count, try adding them
1568 * again providing the action policy is
1569 * 're-add' or greater. The bitmap
1570 * might allow them to be included, or
1571 * they will become spares.
1572 */
1573 for (i = 0; i < bestcnt; i++) {
1574 int j = best[i];
1575 if (j >= 0 && !devices[j].uptodate) {
1576 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1577 continue;
1578 rv = add_disk(mdfd, st, content,
1579 &devices[j].i);
1580 if (rv == 0 && c->verbose >= 0)
1581 pr_err("%s has been re-added.\n",
1582 devices[j].devname);
1583 }
1584 }
1585 }
1586 map_unlock(&map);
1587 wait_for(mddev, mdfd);
1588 close(mdfd);
1589 if (auto_assem) {
1590 int usecs = 1;
1591 /* There is a nasty race with 'mdadm --monitor'.
1592 * If it opens this device before we close it,
1593 * it gets an incomplete open on which IO
1594 * doesn't work and the capacity is
1595 * wrong.
1596 * If we reopen (to check for layered devices)
1597 * before --monitor closes, we loose.
1598 *
1599 * So: wait upto 1 second for there to be
1600 * a non-zero capacity.
1601 */
1602 while (usecs < 1000) {
1603 mdfd = open(mddev, O_RDONLY);
1604 if (mdfd >= 0) {
1605 unsigned long long size;
1606 if (get_dev_size(mdfd, NULL, &size) &&
1607 size > 0)
1608 break;
1609 close(mdfd);
1610 }
1611 usleep(usecs);
1612 usecs <<= 1;
1613 }
1614 }
1615 free(devices);
1616 return 0;
1617 }
1618 pr_err("failed to RUN_ARRAY %s: %s\n",
1619 mddev, strerror(errno));
1620
1621 if (!enough(content->array.level, content->array.raid_disks,
1622 content->array.layout, 1, avail))
1623 pr_err("Not enough devices to "
1624 "start the array.\n");
1625 else if (!enough(content->array.level,
1626 content->array.raid_disks,
1627 content->array.layout, clean,
1628 avail))
1629 pr_err("Not enough devices to "
1630 "start the array while not clean "
1631 "- consider --force.\n");
1632
1633 if (auto_assem)
1634 ioctl(mdfd, STOP_ARRAY, NULL);
1635 close(mdfd);
1636 free(devices);
1637 map_unlock(&map);
1638 return 1;
1639 }
1640 if (c->runstop == -1) {
1641 pr_err("%s assembled from %d drive%s",
1642 mddev, okcnt, okcnt==1?"":"s");
1643 if (okcnt != (unsigned)content->array.raid_disks)
1644 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1645 fprintf(stderr, ", but not started.\n");
1646 close(mdfd);
1647 free(devices);
1648 map_unlock(&map);
1649 return 0;
1650 }
1651 if (c->verbose >= -1) {
1652 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1653 if (rebuilding_cnt)
1654 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
1655 if (sparecnt)
1656 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1657 if (!enough(content->array.level, content->array.raid_disks,
1658 content->array.layout, 1, avail))
1659 fprintf(stderr, " - not enough to start the array.\n");
1660 else if (!enough(content->array.level,
1661 content->array.raid_disks,
1662 content->array.layout, clean,
1663 avail))
1664 fprintf(stderr, " - not enough to start the "
1665 "array while not clean - consider "
1666 "--force.\n");
1667 else {
1668 if (req_cnt == (unsigned)content->array.raid_disks)
1669 fprintf(stderr, " - need all %d to start it", req_cnt);
1670 else
1671 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1672 fprintf(stderr, " (use --run to insist).\n");
1673 }
1674 }
1675 if (auto_assem)
1676 ioctl(mdfd, STOP_ARRAY, NULL);
1677 close(mdfd);
1678 free(devices);
1679 map_unlock(&map);
1680 return 1;
1681 } else {
1682 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1683 * been updated to point to the current locations of devices.
1684 * so we can just start the array
1685 */
1686 unsigned long dev;
1687 dev = makedev(devices[chosen_drive].i.disk.major,
1688 devices[chosen_drive].i.disk.minor);
1689 if (ioctl(mdfd, START_ARRAY, dev)) {
1690 pr_err("Cannot start array: %s\n",
1691 strerror(errno));
1692 }
1693
1694 }
1695 close(mdfd);
1696 free(devices);
1697 map_unlock(&map);
1698 return 0;
1699 }
1700
1701 #ifndef MDASSEMBLE
1702 int assemble_container_content(struct supertype *st, int mdfd,
1703 struct mdinfo *content, struct context *c,
1704 char *chosen_name)
1705 {
1706 struct mdinfo *dev, *sra;
1707 int working = 0, preexist = 0;
1708 int expansion = 0;
1709 struct map_ent *map = NULL;
1710 int old_raid_disks;
1711 int start_reshape;
1712
1713 sysfs_init(content, mdfd, 0);
1714
1715 sra = sysfs_read(mdfd, 0, GET_VERSION);
1716 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1717 if (content->array.major_version == -1 &&
1718 content->array.minor_version == -2 &&
1719 c->readonly &&
1720 content->text_version[0] == '/')
1721 content->text_version[0] = '-';
1722 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1723 if (sra)
1724 sysfs_free(sra);
1725 return 1;
1726 }
1727 }
1728
1729 /* There are two types of reshape: container wide or sub-array specific
1730 * Check if metadata requests blocking container wide reshapes
1731 */
1732 start_reshape = (content->reshape_active &&
1733 !((content->reshape_active == CONTAINER_RESHAPE) &&
1734 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1735
1736 /* Block subarray here if it is under reshape now
1737 * Do not allow for any changes in this array
1738 */
1739 if (st->ss->external && content->recovery_blocked && start_reshape)
1740 block_subarray(content);
1741
1742 if (sra)
1743 sysfs_free(sra);
1744 old_raid_disks = content->array.raid_disks - content->delta_disks;
1745 for (dev = content->devs; dev; dev = dev->next)
1746 if (sysfs_add_disk(content, dev, 1) == 0) {
1747 if (dev->disk.raid_disk >= old_raid_disks &&
1748 content->reshape_active)
1749 expansion++;
1750 else
1751 working++;
1752 } else if (errno == EEXIST)
1753 preexist++;
1754 if (working + expansion == 0)
1755 return 1;/* Nothing new, don't try to start */
1756
1757 map_update(&map, fd2devnum(mdfd),
1758 content->text_version,
1759 content->uuid, chosen_name);
1760
1761 if (c->runstop > 0 ||
1762 (working + preexist + expansion) >=
1763 content->array.working_disks) {
1764 int err;
1765
1766 if (start_reshape) {
1767 int spare = content->array.raid_disks + expansion;
1768 if (restore_backup(st, content,
1769 working,
1770 spare, c->backup_file, c->verbose) == 1)
1771 return 1;
1772
1773 err = sysfs_set_str(content, NULL,
1774 "array_state", "readonly");
1775 if (err)
1776 return 1;
1777
1778 if (st->ss->external) {
1779 if (!mdmon_running(st->container_dev))
1780 start_mdmon(st->container_dev);
1781 ping_monitor_by_id(st->container_dev);
1782 if (mdmon_running(st->container_dev) &&
1783 st->update_tail == NULL)
1784 st->update_tail = &st->updates;
1785 }
1786
1787 err = Grow_continue(mdfd, st, content, c->backup_file,
1788 c->freeze_reshape);
1789 } else switch(content->array.level) {
1790 case LEVEL_LINEAR:
1791 case LEVEL_MULTIPATH:
1792 case 0:
1793 err = sysfs_set_str(content, NULL, "array_state",
1794 c->readonly ? "readonly" : "active");
1795 break;
1796 default:
1797 err = sysfs_set_str(content, NULL, "array_state",
1798 "readonly");
1799 /* start mdmon if needed. */
1800 if (!err) {
1801 if (!mdmon_running(st->container_dev))
1802 start_mdmon(st->container_dev);
1803 ping_monitor_by_id(st->container_dev);
1804 }
1805 break;
1806 }
1807 if (!err)
1808 sysfs_set_safemode(content, content->safe_mode_delay);
1809
1810 /* Block subarray here if it is not reshaped now
1811 * It has be blocked a little later to allow mdmon to switch in
1812 * in to R/W state
1813 */
1814 if (st->ss->external && content->recovery_blocked &&
1815 !start_reshape)
1816 block_subarray(content);
1817
1818 if (c->verbose >= 0) {
1819 if (err)
1820 pr_err("array %s now has %d device%s",
1821 chosen_name, working + preexist,
1822 working + preexist == 1 ? "":"s");
1823 else
1824 pr_err("Started %s with %d device%s",
1825 chosen_name, working + preexist,
1826 working + preexist == 1 ? "":"s");
1827 if (preexist)
1828 fprintf(stderr, " (%d new)", working);
1829 if (expansion)
1830 fprintf(stderr, " ( + %d for expansion)",
1831 expansion);
1832 fprintf(stderr, "\n");
1833 }
1834 if (!err)
1835 wait_for(chosen_name, mdfd);
1836 return err;
1837 /* FIXME should have an O_EXCL and wait for read-auto */
1838 } else {
1839 if (c->verbose >= 0) {
1840 pr_err("%s assembled with %d device%s",
1841 chosen_name, preexist + working,
1842 preexist + working == 1 ? "":"s");
1843 if (preexist)
1844 fprintf(stderr, " (%d new)", working);
1845 fprintf(stderr, " but not started\n");
1846 }
1847 return 1;
1848 }
1849 }
1850 #endif
1851
Unnamed repository; edit this file 'description' to name the repository.