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


a57d384daeff06b77d75580f9ad86723233c26f4
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(0, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 static int ident_matches(struct mddev_ident *ident,
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78 {
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
82 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
83 if (devname)
84 pr_err("%s has wrong uuid.\n", devname);
85 return 0;
86 }
87 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
88 name_matches(content->name, ident->name, homehost)==0) {
89 if (devname)
90 pr_err("%s has wrong name.\n", devname);
91 return 0;
92 }
93 if (ident->super_minor != UnSet &&
94 ident->super_minor != content->array.md_minor) {
95 if (devname)
96 pr_err("%s has wrong super-minor.\n",
97 devname);
98 return 0;
99 }
100 if (ident->level != UnSet &&
101 ident->level != content->array.level) {
102 if (devname)
103 pr_err("%s has wrong raid level.\n",
104 devname);
105 return 0;
106 }
107 if (ident->raid_disks != UnSet &&
108 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
109 ident->raid_disks!= content->array.raid_disks) {
110 if (devname)
111 pr_err("%s requires wrong number of drives.\n",
112 devname);
113 return 0;
114 }
115 if (ident->member && ident->member[0]) {
116 /* content->text_version must match */
117 char *s = strchr(content->text_version+1, '/');
118 if (s == NULL) {
119 if (devname)
120 pr_err("%s is not a container and one is required.\n",
121 devname);
122 return 0;
123 } else if (strcmp(ident->member, s+1) != 0) {
124 if (devname)
125 pr_err("skipping wrong member %s is %s\n",
126 content->text_version, devname);
127 return 0;
128 }
129 }
130 return 1;
131 }
132
133 static int select_devices(struct mddev_dev *devlist,
134 struct mddev_ident *ident,
135 struct supertype **stp,
136 struct mdinfo **contentp,
137 struct context *c,
138 int inargv, int auto_assem)
139 {
140 struct mddev_dev *tmpdev;
141 int num_devs;
142 struct supertype *st = *stp;
143 struct mdinfo *content = NULL;
144 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
145 struct domainlist *domains = NULL;
146
147 tmpdev = devlist; num_devs = 0;
148 while (tmpdev) {
149 if (tmpdev->used)
150 tmpdev->used = 2;
151 else
152 num_devs++;
153 tmpdev->disposition = 0;
154 tmpdev = tmpdev->next;
155 }
156
157 /* first walk the list of devices to find a consistent set
158 * that match the criterea, if that is possible.
159 * We flag the ones we like with 'used'.
160 */
161 for (tmpdev = devlist;
162 tmpdev;
163 tmpdev = tmpdev ? tmpdev->next : NULL) {
164 char *devname = tmpdev->devname;
165 int dfd;
166 struct stat stb;
167 struct supertype *tst;
168 struct dev_policy *pol = NULL;
169 int found_container = 0;
170
171 if (tmpdev->used > 1)
172 continue;
173
174 if (ident->container) {
175 if (ident->container[0] == '/' &&
176 !same_dev(ident->container, devname)) {
177 if (report_mismatch)
178 pr_err("%s is not the container required (%s)\n",
179 devname, ident->container);
180 continue;
181 }
182 } else if (ident->devices &&
183 !match_oneof(ident->devices, devname)) {
184 /* Note that we ignore the "device=" identifier if a
185 * "container=" is given. Checking both is unnecessarily
186 * complicated.
187 */
188 if (report_mismatch)
189 pr_err("%s is not one of %s\n", devname, ident->devices);
190 continue;
191 }
192
193 tst = dup_super(st);
194
195 dfd = dev_open(devname, O_RDONLY);
196 if (dfd < 0) {
197 if (report_mismatch)
198 pr_err("cannot open device %s: %s\n",
199 devname, strerror(errno));
200 tmpdev->used = 2;
201 } else if (fstat(dfd, &stb)< 0) {
202 /* Impossible! */
203 pr_err("fstat failed for %s: %s\n",
204 devname, strerror(errno));
205 tmpdev->used = 2;
206 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
207 pr_err("%s is not a block device.\n",
208 devname);
209 tmpdev->used = 2;
210 } else if (must_be_container(dfd)) {
211 if (st) {
212 /* already found some components, this cannot
213 * be another one.
214 */
215 if (report_mismatch)
216 pr_err("%s is a container, but we are looking for components\n",
217 devname);
218 tmpdev->used = 2;
219 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
220 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
221 if (report_mismatch)
222 pr_err("not a recognisable container: %s\n",
223 devname);
224 tmpdev->used = 2;
225 #endif
226 } else if (!tst->ss->load_container
227 || tst->ss->load_container(tst, dfd, NULL)) {
228 if (report_mismatch)
229 pr_err("no correct container type: %s\n",
230 devname);
231 tmpdev->used = 2;
232 } else if (auto_assem &&
233 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
234 tst->ss->match_home(tst, c->homehost) == 1)) {
235 if (report_mismatch)
236 pr_err("%s has metadata type %s for which "
237 "auto-assembly is disabled\n",
238 devname, tst->ss->name);
239 tmpdev->used = 2;
240 } else
241 found_container = 1;
242 } else {
243 if (!tst && (tst = guess_super(dfd)) == NULL) {
244 if (report_mismatch)
245 pr_err("no recogniseable superblock on %s\n",
246 devname);
247 tmpdev->used = 2;
248 } else if (tst->ss->load_super(tst,dfd, NULL)) {
249 if (report_mismatch)
250 pr_err("no RAID superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if (tst->ss->compare_super == NULL) {
254 if (report_mismatch)
255 pr_err("Cannot assemble %s metadata on %s\n",
256 tst->ss->name, devname);
257 tmpdev->used = 2;
258 } else if (auto_assem && st == NULL &&
259 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
260 tst->ss->match_home(tst, c->homehost) == 1)) {
261 if (report_mismatch)
262 pr_err("%s has metadata type %s for which "
263 "auto-assembly is disabled\n",
264 devname, tst->ss->name);
265 tmpdev->used = 2;
266 }
267 }
268 if (dfd >= 0) close(dfd);
269 if (tmpdev->used == 2) {
270 if (auto_assem || !inargv)
271 /* Ignore unrecognised devices during auto-assembly */
272 goto loop;
273 if (ident->uuid_set || ident->name[0] ||
274 ident->super_minor != UnSet)
275 /* Ignore unrecognised device if looking for
276 * specific array */
277 goto loop;
278
279 pr_err("%s has no superblock - assembly aborted\n",
280 devname);
281 if (st)
282 st->ss->free_super(st);
283 dev_policy_free(pol);
284 domain_free(domains);
285 return -1;
286 }
287
288 if (found_container) {
289 /* tmpdev is a container. We need to be either
290 * looking for a member, or auto-assembling
291 */
292 /* should be safe to try an exclusive open now, we
293 * have rejected anything that some other mdadm might
294 * be looking at
295 */
296 dfd = dev_open(devname, O_RDONLY | O_EXCL);
297 if (dfd < 0) {
298 if (report_mismatch)
299 pr_err("%s is busy - skipping\n", devname);
300 goto loop;
301 }
302 close(dfd);
303
304 if (ident->container && ident->container[0] != '/') {
305 /* we have a uuid */
306 int uuid[4];
307
308 content = *contentp;
309 tst->ss->getinfo_super(tst, content, NULL);
310
311 if (!parse_uuid(ident->container, uuid) ||
312 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
313 if (report_mismatch)
314 pr_err("%s has wrong UUID to be required container\n",
315 devname);
316 goto loop;
317 }
318 }
319 /* It is worth looking inside this container.
320 */
321 if (c->verbose > 0)
322 pr_err("looking in container %s\n",
323 devname);
324
325 for (content = tst->ss->container_content(tst, NULL);
326 content;
327 content = content->next) {
328
329 if (!ident_matches(ident, content, tst,
330 c->homehost, c->update,
331 report_mismatch ? devname : NULL))
332 /* message already printed */;
333 else if (is_member_busy(content->text_version)) {
334 if (report_mismatch)
335 pr_err("member %s in %s is already assembled\n",
336 content->text_version,
337 devname);
338 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
339 /* do not assemble arrays with unsupported configurations */
340 pr_err("Cannot activate member %s in %s.\n",
341 content->text_version,
342 devname);
343 } else
344 break;
345 }
346 if (!content) {
347 tmpdev->used = 2;
348 goto loop; /* empty container */
349 }
350
351 st = tst; tst = NULL;
352 if (!auto_assem && inargv && tmpdev->next != NULL) {
353 pr_err("%s is a container, but is not "
354 "only device given: confused and aborting\n",
355 devname);
356 st->ss->free_super(st);
357 dev_policy_free(pol);
358 domain_free(domains);
359 return -1;
360 }
361 if (c->verbose > 0)
362 pr_err("found match on member %s in %s\n",
363 content->text_version, devname);
364
365 /* make sure we finished the loop */
366 tmpdev = NULL;
367 goto loop;
368 } else {
369 int rv = 0;
370 struct mddev_ident *match;
371
372 content = *contentp;
373 tst->ss->getinfo_super(tst, content, NULL);
374
375 if (!ident_matches(ident, content, tst,
376 c->homehost, c->update,
377 report_mismatch ? devname : NULL))
378 goto loop;
379
380 match = conf_match(tst, content, devname,
381 report_mismatch ? c->verbose : -1,
382 &rv);
383 if (!match && rv == 2)
384 goto loop;
385 if (match && match->devname &&
386 strcasecmp(match->devname, "<ignore>") == 0) {
387 if (report_mismatch)
388 pr_err("%s is a member of an explicitly ignored array\n",
389 devname);
390 goto loop;
391 }
392 if (match && !ident_matches(match, content, tst,
393 c->homehost, c->update,
394 report_mismatch ? devname : NULL))
395 /* Array exists in mdadm.conf but some
396 * details don't match, so reject it
397 */
398 goto loop;
399
400 /* should be safe to try an exclusive open now, we
401 * have rejected anything that some other mdadm might
402 * be looking at
403 */
404 dfd = dev_open(devname, O_RDONLY | O_EXCL);
405 if (dfd < 0) {
406 if (report_mismatch)
407 pr_err("%s is busy - skipping\n", devname);
408 goto loop;
409 }
410 close(dfd);
411
412 if (st == NULL)
413 st = dup_super(tst);
414 if (st->minor_version == -1)
415 st->minor_version = tst->minor_version;
416
417 if (memcmp(content->uuid, uuid_zero,
418 sizeof(int[4])) == 0) {
419 /* this is a floating spare. It cannot define
420 * an array unless there are no more arrays of
421 * this type to be found. It can be included
422 * in an array of this type though.
423 */
424 tmpdev->used = 3;
425 goto loop;
426 }
427
428 if (st->ss != tst->ss ||
429 st->minor_version != tst->minor_version ||
430 st->ss->compare_super(st, tst) != 0) {
431 /* Some mismatch. If exactly one array matches this host,
432 * we can resolve on that one.
433 * Or, if we are auto assembling, we just ignore the second
434 * for now.
435 */
436 if (auto_assem)
437 goto loop;
438 if (c->homehost) {
439 int first = st->ss->match_home(st, c->homehost);
440 int last = tst->ss->match_home(tst, c->homehost);
441 if (first != last &&
442 (first == 1 || last == 1)) {
443 /* We can do something */
444 if (first) {/* just ignore this one */
445 if (report_mismatch)
446 pr_err("%s misses out due to wrong homehost\n",
447 devname);
448 goto loop;
449 } else { /* reject all those sofar */
450 struct mddev_dev *td;
451 if (report_mismatch)
452 pr_err("%s overrides previous devices due to good homehost\n",
453 devname);
454 for (td=devlist; td != tmpdev; td=td->next)
455 if (td->used == 1)
456 td->used = 0;
457 tmpdev->used = 1;
458 goto loop;
459 }
460 }
461 }
462 pr_err("superblock on %s doesn't match others - assembly aborted\n",
463 devname);
464 tst->ss->free_super(tst);
465 st->ss->free_super(st);
466 dev_policy_free(pol);
467 domain_free(domains);
468 return -1;
469 }
470 tmpdev->used = 1;
471 }
472 loop:
473 /* Collect domain information from members only */
474 if (tmpdev && tmpdev->used == 1) {
475 if (!pol)
476 pol = devid_policy(stb.st_rdev);
477 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
478 }
479 dev_policy_free(pol);
480 pol = NULL;
481 if (tst)
482 tst->ss->free_super(tst);
483 }
484
485 /* Check if we found some imsm spares but no members */
486 if ((auto_assem ||
487 (ident->uuid_set &&
488 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
489 (!st || !st->sb))
490 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
491 if (tmpdev->used != 3)
492 continue;
493 tmpdev->used = 1;
494 content = *contentp;
495
496 if (!st->sb) {
497 /* we need sb from one of the spares */
498 int dfd = dev_open(tmpdev->devname, O_RDONLY);
499 if (dfd < 0 ||
500 st->ss->load_super(st, dfd, NULL))
501 tmpdev->used = 2;
502 if (dfd > 0)
503 close(dfd);
504 }
505 }
506
507 /* Now reject spares that don't match domains of identified members */
508 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
509 struct stat stb;
510 if (tmpdev->used != 3)
511 continue;
512 if (stat(tmpdev->devname, &stb)< 0) {
513 pr_err("fstat failed for %s: %s\n",
514 tmpdev->devname, strerror(errno));
515 tmpdev->used = 2;
516 } else {
517 struct dev_policy *pol = devid_policy(stb.st_rdev);
518 int dt = domain_test(domains, pol, NULL);
519 if (inargv && dt != 0)
520 /* take this spare as domains match
521 * if there are any */
522 tmpdev->used = 1;
523 else if (!inargv && dt == 1)
524 /* device wasn't explicitly listed, so need
525 * explicit domain match - which we have */
526 tmpdev->used = 1;
527 else
528 /* if domains don't match mark as unused */
529 tmpdev->used = 0;
530 dev_policy_free(pol);
531 }
532 }
533 domain_free(domains);
534 *stp = st;
535 if (st && st->sb && content == *contentp)
536 st->ss->getinfo_super(st, content, NULL);
537 *contentp = content;
538
539 return num_devs;
540 }
541
542 struct devs {
543 char *devname;
544 int uptodate; /* set once we decide that this device is as
545 * recent as everything else in the array.
546 */
547 int included; /* set if the device is already in the array
548 * due to a previous '-I'
549 */
550 struct mdinfo i;
551 };
552
553 static int load_devices(struct devs *devices, char *devmap,
554 struct mddev_ident *ident, struct supertype **stp,
555 struct mddev_dev *devlist, struct context *c,
556 struct mdinfo *content,
557 int mdfd, char *mddev,
558 int *most_recentp, int *bestcntp, int **bestp,
559 int inargv)
560 {
561 struct mddev_dev *tmpdev;
562 int devcnt = 0;
563 int nextspare = 0;
564 #ifndef MDASSEMBLE
565 int bitmap_done = 0;
566 #endif
567 int most_recent = -1;
568 int bestcnt = 0;
569 int *best = *bestp;
570 struct supertype *st = *stp;
571
572 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
573 char *devname = tmpdev->devname;
574 struct stat stb;
575 struct supertype *tst;
576 int i;
577
578 if (tmpdev->used != 1)
579 continue;
580 /* looks like a good enough match to update the super block if needed */
581 #ifndef MDASSEMBLE
582 if (c->update) {
583 int dfd;
584 /* prepare useful information in info structures */
585 struct stat stb2;
586 int err;
587 fstat(mdfd, &stb2);
588
589 if (strcmp(c->update, "uuid")==0 &&
590 !ident->uuid_set) {
591 int rfd;
592 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
593 read(rfd, ident->uuid, 16) != 16) {
594 *(__u32*)(ident->uuid) = random();
595 *(__u32*)(ident->uuid+1) = random();
596 *(__u32*)(ident->uuid+2) = random();
597 *(__u32*)(ident->uuid+3) = random();
598 }
599 if (rfd >= 0) close(rfd);
600 }
601 dfd = dev_open(devname,
602 tmpdev->disposition == 'I'
603 ? O_RDWR : (O_RDWR|O_EXCL));
604
605 tst = dup_super(st);
606 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
607 pr_err("cannot re-read metadata from %s - aborting\n",
608 devname);
609 if (dfd >= 0)
610 close(dfd);
611 close(mdfd);
612 free(devices);
613 free(devmap);
614 tst->ss->free_super(tst);
615 free(tst);
616 *stp = st;
617 return -1;
618 }
619 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
620
621 memcpy(content->uuid, ident->uuid, 16);
622 strcpy(content->name, ident->name);
623 content->array.md_minor = minor(stb2.st_rdev);
624
625 if (strcmp(c->update, "byteorder") == 0)
626 err = 0;
627 else
628 err = tst->ss->update_super(tst, content, c->update,
629 devname, c->verbose,
630 ident->uuid_set,
631 c->homehost);
632 if (err < 0) {
633 if (err == -1)
634 pr_err("--update=%s not understood"
635 " for %s metadata\n",
636 c->update, tst->ss->name);
637 tst->ss->free_super(tst);
638 free(tst);
639 close(mdfd);
640 close(dfd);
641 free(devices);
642 free(devmap);
643 *stp = st;
644 return -1;
645 }
646 if (strcmp(c->update, "uuid")==0 &&
647 !ident->uuid_set) {
648 ident->uuid_set = 1;
649 memcpy(ident->uuid, content->uuid, 16);
650 }
651 if (tst->ss->store_super(tst, dfd))
652 pr_err("Could not re-write superblock on %s.\n",
653 devname);
654 close(dfd);
655
656 if (strcmp(c->update, "uuid")==0 &&
657 ident->bitmap_fd >= 0 && !bitmap_done) {
658 if (bitmap_update_uuid(ident->bitmap_fd,
659 content->uuid,
660 tst->ss->swapuuid) != 0)
661 pr_err("Could not update uuid on external bitmap.\n");
662 else
663 bitmap_done = 1;
664 }
665 } else
666 #endif
667 {
668 int dfd = dev_open(devname,
669 tmpdev->disposition == 'I'
670 ? O_RDWR : (O_RDWR|O_EXCL));
671 tst = dup_super(st);
672
673 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
674 pr_err("cannot re-read metadata from %s - aborting\n",
675 devname);
676 if (dfd >= 0)
677 close(dfd);
678 close(mdfd);
679 free(devices);
680 free(devmap);
681 tst->ss->free_super(tst);
682 free(tst);
683 *stp = st;
684 return -1;
685 }
686 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
687 close(dfd);
688 }
689
690 stat(devname, &stb);
691
692 if (c->verbose > 0)
693 pr_err("%s is identified as a member of %s, slot %d%s.\n",
694 devname, mddev, content->disk.raid_disk,
695 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
696 devices[devcnt].devname = devname;
697 devices[devcnt].uptodate = 0;
698 devices[devcnt].included = (tmpdev->disposition == 'I');
699 devices[devcnt].i = *content;
700 devices[devcnt].i.disk.major = major(stb.st_rdev);
701 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
702
703 if (devices[devcnt].i.disk.state == 6) {
704 if (most_recent < 0 ||
705 devices[devcnt].i.events
706 > devices[most_recent].i.events) {
707 struct supertype *tmp = tst;
708 tst = st;
709 st = tmp;
710 most_recent = devcnt;
711 }
712 }
713 tst->ss->free_super(tst);
714 free(tst);
715
716 if (content->array.level == LEVEL_MULTIPATH)
717 /* with multipath, the raid_disk from the superblock is meaningless */
718 i = devcnt;
719 else
720 i = devices[devcnt].i.disk.raid_disk;
721 if (i+1 == 0) {
722 if (nextspare < content->array.raid_disks*2)
723 nextspare = content->array.raid_disks*2;
724 i = nextspare++;
725 } else {
726 /* i is raid_disk - double it so there is room for
727 * replacements */
728 i *= 2;
729 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
730 i++;
731 if (i >= content->array.raid_disks*2 &&
732 i >= nextspare)
733 nextspare = i+1;
734 }
735 if (i < 10000) {
736 if (i >= bestcnt) {
737 int newbestcnt = i+10;
738 int *newbest = xmalloc(sizeof(int)*newbestcnt);
739 int c;
740 for (c=0; c < newbestcnt; c++)
741 if (c < bestcnt)
742 newbest[c] = best[c];
743 else
744 newbest[c] = -1;
745 if (best)free(best);
746 best = newbest;
747 bestcnt = newbestcnt;
748 }
749 if (best[i] >=0 &&
750 devices[best[i]].i.events
751 == devices[devcnt].i.events
752 && (devices[best[i]].i.disk.minor
753 != devices[devcnt].i.disk.minor)
754 && st->ss == &super0
755 && content->array.level != LEVEL_MULTIPATH) {
756 /* two different devices with identical superblock.
757 * Could be a mis-detection caused by overlapping
758 * partitions. fail-safe.
759 */
760 pr_err("WARNING %s and %s appear"
761 " to have very similar superblocks.\n"
762 " If they are really different, "
763 "please --zero the superblock on one\n"
764 " If they are the same or overlap,"
765 " please remove one from %s.\n",
766 devices[best[i]].devname, devname,
767 inargv ? "the list" :
768 "the\n DEVICE list in mdadm.conf"
769 );
770 close(mdfd);
771 free(devices);
772 free(devmap);
773 *stp = st;
774 return -1;
775 }
776 if (best[i] == -1
777 || (devices[best[i]].i.events
778 < devices[devcnt].i.events))
779 best[i] = devcnt;
780 }
781 devcnt++;
782 }
783 if (most_recent >= 0)
784 *most_recentp = most_recent;
785 *bestcntp = bestcnt;
786 *bestp = best;
787 *stp = st;
788 return devcnt;
789 }
790
791 static int force_array(struct mdinfo *content,
792 struct devs *devices,
793 int *best, int bestcnt, char *avail,
794 int most_recent,
795 struct supertype *st,
796 struct context *c)
797 {
798 int okcnt = 0;
799 while (!enough(content->array.level, content->array.raid_disks,
800 content->array.layout, 1,
801 avail)
802 ||
803 (content->reshape_active && content->delta_disks > 0 &&
804 !enough(content->array.level, (content->array.raid_disks
805 - content->delta_disks),
806 content->new_layout, 1,
807 avail)
808 )) {
809 /* Choose the newest best drive which is
810 * not up-to-date, update the superblock
811 * and add it.
812 */
813 int fd;
814 struct supertype *tst;
815 unsigned long long current_events;
816 int chosen_drive = -1;
817 int i;
818
819 for (i = 0;
820 i < content->array.raid_disks * 2 && i < bestcnt;
821 i += 2) {
822 int j = best[i];
823 if (j>=0 &&
824 !devices[j].uptodate &&
825 devices[j].i.recovery_start == MaxSector &&
826 (chosen_drive < 0 ||
827 devices[j].i.events
828 > devices[chosen_drive].i.events))
829 chosen_drive = j;
830 }
831 if (chosen_drive < 0)
832 break;
833 current_events = devices[chosen_drive].i.events;
834 add_another:
835 if (c->verbose >= 0)
836 pr_err("forcing event count in %s(%d) from %d upto %d\n",
837 devices[chosen_drive].devname,
838 devices[chosen_drive].i.disk.raid_disk,
839 (int)(devices[chosen_drive].i.events),
840 (int)(devices[most_recent].i.events));
841 fd = dev_open(devices[chosen_drive].devname,
842 devices[chosen_drive].included ? O_RDWR
843 : (O_RDWR|O_EXCL));
844 if (fd < 0) {
845 pr_err("Couldn't open %s for write - not updating\n",
846 devices[chosen_drive].devname);
847 devices[chosen_drive].i.events = 0;
848 continue;
849 }
850 tst = dup_super(st);
851 if (tst->ss->load_super(tst,fd, NULL)) {
852 close(fd);
853 pr_err("RAID superblock disappeared from %s - not updating.\n",
854 devices[chosen_drive].devname);
855 devices[chosen_drive].i.events = 0;
856 continue;
857 }
858 content->events = devices[most_recent].i.events;
859 tst->ss->update_super(tst, content, "force-one",
860 devices[chosen_drive].devname, c->verbose,
861 0, NULL);
862
863 if (tst->ss->store_super(tst, fd)) {
864 close(fd);
865 pr_err("Could not re-write superblock on %s\n",
866 devices[chosen_drive].devname);
867 devices[chosen_drive].i.events = 0;
868 tst->ss->free_super(tst);
869 continue;
870 }
871 close(fd);
872 devices[chosen_drive].i.events = devices[most_recent].i.events;
873 devices[chosen_drive].uptodate = 1;
874 avail[chosen_drive] = 1;
875 okcnt++;
876 tst->ss->free_super(tst);
877
878 /* If there are any other drives of the same vintage,
879 * add them in as well. We can't lose and we might gain
880 */
881 for (i = 0;
882 i < content->array.raid_disks * 2 && i < bestcnt ;
883 i += 2) {
884 int j = best[i];
885 if (j >= 0 &&
886 !devices[j].uptodate &&
887 devices[j].i.recovery_start == MaxSector &&
888 devices[j].i.events == current_events) {
889 chosen_drive = j;
890 goto add_another;
891 }
892 }
893 }
894 return okcnt;
895 }
896
897 static int start_array(int mdfd,
898 char *mddev,
899 struct mdinfo *content,
900 struct supertype *st,
901 struct mddev_ident *ident,
902 int *best, int bestcnt,
903 int chosen_drive,
904 struct devs *devices,
905 unsigned int okcnt,
906 unsigned int sparecnt,
907 unsigned int rebuilding_cnt,
908 struct context *c,
909 int clean, char *avail,
910 int start_partial_ok,
911 int err_ok,
912 int was_forced
913 )
914 {
915 int rv;
916 int i;
917 unsigned int req_cnt;
918
919 rv = set_array_info(mdfd, st, content);
920 if (rv && !err_ok) {
921 pr_err("failed to set array info for %s: %s\n",
922 mddev, strerror(errno));
923 return 1;
924 }
925 if (ident->bitmap_fd >= 0) {
926 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
927 pr_err("SET_BITMAP_FILE failed.\n");
928 return 1;
929 }
930 } else if (ident->bitmap_file) {
931 /* From config file */
932 int bmfd = open(ident->bitmap_file, O_RDWR);
933 if (bmfd < 0) {
934 pr_err("Could not open bitmap file %s\n",
935 ident->bitmap_file);
936 return 1;
937 }
938 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
939 pr_err("Failed to set bitmapfile for %s\n", mddev);
940 close(bmfd);
941 return 1;
942 }
943 close(bmfd);
944 }
945
946 /* First, add the raid disks, but add the chosen one last */
947 for (i=0; i<= bestcnt; i++) {
948 int j;
949 if (i < bestcnt) {
950 j = best[i];
951 if (j == chosen_drive)
952 continue;
953 } else
954 j = chosen_drive;
955
956 if (j >= 0 && !devices[j].included) {
957 int dfd = dev_open(devices[j].devname,
958 O_RDWR|O_EXCL);
959 if (dfd >= 0) {
960 remove_partitions(dfd);
961 close(dfd);
962 }
963 rv = add_disk(mdfd, st, content, &devices[j].i);
964
965 if (rv) {
966 pr_err("failed to add "
967 "%s to %s: %s\n",
968 devices[j].devname,
969 mddev,
970 strerror(errno));
971 if (i < content->array.raid_disks * 2
972 || i == bestcnt)
973 okcnt--;
974 else
975 sparecnt--;
976 } else if (c->verbose > 0)
977 pr_err("added %s to %s as %d%s%s\n",
978 devices[j].devname, mddev,
979 devices[j].i.disk.raid_disk,
980 devices[j].uptodate?"":
981 " (possibly out of date)",
982 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
983 } else if (j >= 0) {
984 if (c->verbose > 0)
985 pr_err("%s is already in %s as %d\n",
986 devices[j].devname, mddev,
987 devices[j].i.disk.raid_disk);
988 } else if (c->verbose > 0 && i < content->array.raid_disks*2
989 && (i&1) == 0)
990 pr_err("no uptodate device for slot %d of %s\n",
991 i, mddev);
992 }
993
994 if (content->array.level == LEVEL_CONTAINER) {
995 if (c->verbose >= 0) {
996 pr_err("Container %s has been "
997 "assembled with %d drive%s",
998 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
999 if (okcnt < (unsigned)content->array.raid_disks)
1000 fprintf(stderr, " (out of %d)",
1001 content->array.raid_disks);
1002 fprintf(stderr, "\n");
1003 }
1004 st->ss->free_super(st);
1005 sysfs_uevent(content, "change");
1006 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1007 /* Was partial, is still partial, so signal an error
1008 * to ensure we don't retry */
1009 return 1;
1010 return 0;
1011 }
1012
1013 /* Get number of in-sync devices according to the superblock.
1014 * We must have this number to start the array without -s or -R
1015 */
1016 req_cnt = content->array.working_disks;
1017
1018 if (c->runstop == 1 ||
1019 (c->runstop <= 0 &&
1020 ( enough(content->array.level, content->array.raid_disks,
1021 content->array.layout, clean, avail) &&
1022 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1023 ))) {
1024 /* This array is good-to-go.
1025 * If a reshape is in progress then we might need to
1026 * continue monitoring it. In that case we start
1027 * it read-only and let the grow code make it writable.
1028 */
1029 int rv;
1030 #ifndef MDASSEMBLE
1031 if (content->reshape_active &&
1032 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1033 content->delta_disks <= 0) {
1034 if (!c->backup_file) {
1035 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1036 mddev);
1037 pr_err("Please provided one with \"--backup-file=...\"\n");
1038 if (c->update &&
1039 strcmp(c->update, "revert-reshape") == 0)
1040 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1041 return 1;
1042 }
1043 rv = sysfs_set_str(content, NULL,
1044 "array_state", "readonly");
1045 if (rv == 0)
1046 rv = Grow_continue(mdfd, st, content,
1047 c->backup_file, 0,
1048 c->freeze_reshape);
1049 } else if (c->readonly &&
1050 sysfs_attribute_available(
1051 content, NULL, "array_state")) {
1052 rv = sysfs_set_str(content, NULL,
1053 "array_state", "readonly");
1054 } else
1055 #endif
1056 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1057 reopen_mddev(mdfd); /* drop O_EXCL */
1058 if (rv == 0) {
1059 if (c->verbose >= 0) {
1060 pr_err("%s has been started with %d drive%s",
1061 mddev, okcnt, okcnt==1?"":"s");
1062 if (okcnt < (unsigned)content->array.raid_disks)
1063 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1064 if (rebuilding_cnt)
1065 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1066 if (sparecnt)
1067 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1068 fprintf(stderr, ".\n");
1069 }
1070 if (content->reshape_active &&
1071 content->array.level >= 4 &&
1072 content->array.level <= 6) {
1073 /* might need to increase the size
1074 * of the stripe cache - default is 256
1075 */
1076 if (256 < 4 * (content->array.chunk_size/4096)) {
1077 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1078 if (sra)
1079 sysfs_set_num(sra, NULL,
1080 "stripe_cache_size",
1081 (4 * content->array.chunk_size / 4096) + 1);
1082 sysfs_free(sra);
1083 }
1084 }
1085 if (okcnt < (unsigned)content->array.raid_disks) {
1086 /* If any devices did not get added
1087 * because the kernel rejected them based
1088 * on event count, try adding them
1089 * again providing the action policy is
1090 * 're-add' or greater. The bitmap
1091 * might allow them to be included, or
1092 * they will become spares.
1093 */
1094 for (i = 0; i < bestcnt; i++) {
1095 int j = best[i];
1096 if (j >= 0 && !devices[j].uptodate) {
1097 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1098 continue;
1099 rv = add_disk(mdfd, st, content,
1100 &devices[j].i);
1101 if (rv == 0 && c->verbose >= 0)
1102 pr_err("%s has been re-added.\n",
1103 devices[j].devname);
1104 }
1105 }
1106 }
1107 if (content->array.level == 6 &&
1108 okcnt + 1 == (unsigned)content->array.raid_disks &&
1109 was_forced) {
1110 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1111 if (sra)
1112 sysfs_set_str(sra, NULL,
1113 "sync_action", "repair");
1114 sysfs_free(sra);
1115 }
1116 return 0;
1117 }
1118 pr_err("failed to RUN_ARRAY %s: %s\n",
1119 mddev, strerror(errno));
1120
1121 if (!enough(content->array.level, content->array.raid_disks,
1122 content->array.layout, 1, avail))
1123 pr_err("Not enough devices to "
1124 "start the array.\n");
1125 else if (!enough(content->array.level,
1126 content->array.raid_disks,
1127 content->array.layout, clean,
1128 avail))
1129 pr_err("Not enough devices to "
1130 "start the array while not clean "
1131 "- consider --force.\n");
1132
1133 return 1;
1134 }
1135 if (c->runstop == -1) {
1136 pr_err("%s assembled from %d drive%s",
1137 mddev, okcnt, okcnt==1?"":"s");
1138 if (okcnt != (unsigned)content->array.raid_disks)
1139 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1140 fprintf(stderr, ", but not started.\n");
1141 return 2;
1142 }
1143 if (c->verbose >= -1) {
1144 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1145 if (rebuilding_cnt)
1146 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1147 if (sparecnt)
1148 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1149 if (!enough(content->array.level, content->array.raid_disks,
1150 content->array.layout, 1, avail))
1151 fprintf(stderr, " - not enough to start the array.\n");
1152 else if (!enough(content->array.level,
1153 content->array.raid_disks,
1154 content->array.layout, clean,
1155 avail))
1156 fprintf(stderr, " - not enough to start the "
1157 "array while not clean - consider "
1158 "--force.\n");
1159 else {
1160 if (req_cnt == (unsigned)content->array.raid_disks)
1161 fprintf(stderr, " - need all %d to start it", req_cnt);
1162 else
1163 fprintf(stderr, " - need %d to start", req_cnt);
1164 fprintf(stderr, " (use --run to insist).\n");
1165 }
1166 }
1167 return 1;
1168 }
1169
1170 int Assemble(struct supertype *st, char *mddev,
1171 struct mddev_ident *ident,
1172 struct mddev_dev *devlist,
1173 struct context *c)
1174 {
1175 /*
1176 * The task of Assemble is to find a collection of
1177 * devices that should (according to their superblocks)
1178 * form an array, and to give this collection to the MD driver.
1179 * In Linux-2.4 and later, this involves submitting a
1180 * SET_ARRAY_INFO ioctl with no arg - to prepare
1181 * the array - and then submit a number of
1182 * ADD_NEW_DISK ioctls to add disks into
1183 * the array. Finally RUN_ARRAY might
1184 * be submitted to start the array.
1185 *
1186 * Much of the work of Assemble is in finding and/or
1187 * checking the disks to make sure they look right.
1188 *
1189 * If mddev is not set, then scan must be set and we
1190 * read through the config file for dev+uuid mapping
1191 * We recurse, setting mddev, for each device that
1192 * - isn't running
1193 * - has a valid uuid (or any uuid if !uuidset)
1194 *
1195 * If mddev is set, we try to determine state of md.
1196 * check version - must be at least 0.90.0
1197 * check kernel version. must be at least 2.4.
1198 * If not, we can possibly fall back on START_ARRAY
1199 * Try to GET_ARRAY_INFO.
1200 * If possible, give up
1201 * If not, try to STOP_ARRAY just to make sure
1202 *
1203 * If !uuidset and scan, look in conf-file for uuid
1204 * If not found, give up
1205 * If !devlist and scan and uuidset, get list of devs from conf-file
1206 *
1207 * For each device:
1208 * Check superblock - discard if bad
1209 * Check uuid (set if we don't have one) - discard if no match
1210 * Check superblock similarity if we have a superblock - discard if different
1211 * Record events, devicenum
1212 * This should give us a list of devices for the array
1213 * We should collect the most recent event number
1214 *
1215 * Count disks with recent enough event count
1216 * While force && !enough disks
1217 * Choose newest rejected disks, update event count
1218 * mark clean and rewrite superblock
1219 * If recent kernel:
1220 * SET_ARRAY_INFO
1221 * foreach device with recent events : ADD_NEW_DISK
1222 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1223 * If old kernel:
1224 * Check the device numbers in superblock are right
1225 * update superblock if any changes
1226 * START_ARRAY
1227 *
1228 */
1229 int rv;
1230 int mdfd;
1231 int clean;
1232 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1233 ident->super_minor == UnSet && ident->name[0] == 0
1234 && (ident->container == NULL || ident->member == NULL));
1235 struct devs *devices;
1236 char *devmap;
1237 int *best = NULL; /* indexed by raid_disk */
1238 int bestcnt = 0;
1239 int devcnt;
1240 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1241 int i;
1242 int was_forced = 0;
1243 int most_recent = 0;
1244 int chosen_drive;
1245 int change = 0;
1246 int inargv = 0;
1247 int start_partial_ok = (c->runstop >= 0) &&
1248 (c->force || devlist==NULL || auto_assem);
1249 int num_devs;
1250 struct mddev_dev *tmpdev;
1251 struct mdinfo info;
1252 struct mdinfo *content = NULL;
1253 struct mdinfo *pre_exist = NULL;
1254 char *avail;
1255 char *name = NULL;
1256 char chosen_name[1024];
1257 struct map_ent *map = NULL;
1258 struct map_ent *mp;
1259
1260 /*
1261 * If any subdevs are listed, then any that don't
1262 * match ident are discarded. Remainder must all match and
1263 * become the array.
1264 * If no subdevs, then we scan all devices in the config file, but
1265 * there must be something in the identity
1266 */
1267
1268 if (!devlist &&
1269 ident->uuid_set == 0 &&
1270 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1271 ident->name[0] == 0 &&
1272 (ident->container == NULL || ident->member == NULL) &&
1273 ident->devices == NULL) {
1274 pr_err("No identity information available for %s - cannot assemble.\n",
1275 mddev ? mddev : "further assembly");
1276 return 1;
1277 }
1278
1279 if (devlist == NULL)
1280 devlist = conf_get_devs();
1281 else if (mddev)
1282 inargv = 1;
1283
1284 try_again:
1285 /* We come back here when doing auto-assembly and attempting some
1286 * set of devices failed. Those are now marked as ->used==2 and
1287 * we ignore them and try again
1288 */
1289 if (!st && ident->st)
1290 st = ident->st;
1291 if (c->verbose>0)
1292 pr_err("looking for devices for %s\n",
1293 mddev ? mddev : "further assembly");
1294
1295 content = &info;
1296 if (st)
1297 st->ignore_hw_compat = 1;
1298 num_devs = select_devices(devlist, ident, &st, &content, c,
1299 inargv, auto_assem);
1300 if (num_devs < 0)
1301 return 1;
1302
1303 if (!st || !st->sb || !content)
1304 return 2;
1305
1306 /* We have a full set of devices - we now need to find the
1307 * array device.
1308 * However there is a risk that we are racing with "mdadm -I"
1309 * and the array is already partially assembled - we will have
1310 * rejected any devices already in this address.
1311 * So we take a lock on the map file - to prevent further races -
1312 * and look for the uuid in there. If found and the array is
1313 * active, we abort. If found and the array is not active
1314 * we commit to that md device and add all the contained devices
1315 * to our list. We flag them so that we don't try to re-add,
1316 * but can remove if they turn out to not be wanted.
1317 */
1318 if (map_lock(&map))
1319 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1320 mp = map_by_uuid(&map, content->uuid);
1321 if (mp) {
1322 struct mdinfo *dv;
1323 /* array already exists. */
1324 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1325 if (pre_exist->array.level != UnSet) {
1326 pr_err("Found some drive for an array that is already active: %s\n",
1327 mp->path);
1328 pr_err("giving up.\n");
1329 return 1;
1330 }
1331 for (dv = pre_exist->devs; dv; dv = dv->next) {
1332 /* We want to add this device to our list,
1333 * but it could already be there if "mdadm -I"
1334 * started *after* we checked for O_EXCL.
1335 * If we add it to the top of the list
1336 * it will be preferred over later copies.
1337 */
1338 struct mddev_dev *newdev;
1339 char *devname = map_dev(dv->disk.major,
1340 dv->disk.minor,
1341 0);
1342 if (!devname)
1343 continue;
1344 newdev = xmalloc(sizeof(*newdev));
1345 newdev->devname = devname;
1346 newdev->disposition = 'I';
1347 newdev->used = 1;
1348 newdev->next = devlist;
1349 devlist = newdev;
1350 num_devs++;
1351 }
1352 strcpy(chosen_name, mp->path);
1353 if (c->verbose > 0 || mddev == NULL ||
1354 strcmp(mddev, chosen_name) != 0)
1355 pr_err("Merging with already-assembled %s\n",
1356 chosen_name);
1357 mdfd = open_dev_excl(mp->devnm);
1358 } else {
1359 int trustworthy = FOREIGN;
1360 name = content->name;
1361 switch (st->ss->match_home(st, c->homehost)
1362 ?: st->ss->match_home(st, "any")) {
1363 case 1:
1364 trustworthy = LOCAL;
1365 name = strchr(content->name, ':');
1366 if (name)
1367 name++;
1368 else
1369 name = content->name;
1370 break;
1371 }
1372 if (!auto_assem)
1373 /* If the array is listed in mdadm.conf or on
1374 * command line, then we trust the name
1375 * even if the array doesn't look local
1376 */
1377 trustworthy = LOCAL;
1378
1379 if (name[0] == 0 &&
1380 content->array.level == LEVEL_CONTAINER) {
1381 name = content->text_version;
1382 trustworthy = METADATA;
1383 }
1384
1385 if (name[0] && trustworthy != LOCAL &&
1386 ! c->require_homehost &&
1387 conf_name_is_free(name))
1388 trustworthy = LOCAL;
1389
1390 if (trustworthy == LOCAL &&
1391 strchr(name, ':'))
1392 /* Ignore 'host:' prefix of name */
1393 name = strchr(name, ':')+1;
1394
1395 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1396 chosen_name);
1397 }
1398 if (mdfd < 0) {
1399 st->ss->free_super(st);
1400 if (auto_assem)
1401 goto try_again;
1402 return 1;
1403 }
1404 mddev = chosen_name;
1405 if (get_linux_version() < 2004000 ||
1406 md_get_version(mdfd) < 9000) {
1407 pr_err("Assemble requires Linux 2.4 or later, and\n"
1408 " md driver version 0.90.0 or later.\n"
1409 " Upgrade your kernel or try --build\n");
1410 close(mdfd);
1411 return 1;
1412 }
1413 if (pre_exist == NULL) {
1414 if (mddev_busy(fd2devnm(mdfd))) {
1415 pr_err("%s already active, cannot restart it!\n",
1416 mddev);
1417 for (tmpdev = devlist ;
1418 tmpdev && tmpdev->used != 1;
1419 tmpdev = tmpdev->next)
1420 ;
1421 if (tmpdev && auto_assem)
1422 pr_err("%s needed for %s...\n",
1423 mddev, tmpdev->devname);
1424 close(mdfd);
1425 mdfd = -3;
1426 st->ss->free_super(st);
1427 if (auto_assem)
1428 goto try_again;
1429 return 1;
1430 }
1431 /* just incase it was started but has no content */
1432 ioctl(mdfd, STOP_ARRAY, NULL);
1433 }
1434
1435 #ifndef MDASSEMBLE
1436 if (content != &info) {
1437 /* This is a member of a container. Try starting the array. */
1438 int err;
1439 err = assemble_container_content(st, mdfd, content, c,
1440 chosen_name, NULL);
1441 close(mdfd);
1442 return err;
1443 }
1444 #endif
1445 /* Ok, no bad inconsistancy, we can try updating etc */
1446 devices = xcalloc(num_devs, sizeof(*devices));
1447 devmap = xcalloc(num_devs, content->array.raid_disks);
1448 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1449 c, content, mdfd, mddev,
1450 &most_recent, &bestcnt, &best, inargv);
1451 if (devcnt < 0)
1452 return 1;
1453
1454 if (devcnt == 0) {
1455 pr_err("no devices found for %s\n",
1456 mddev);
1457 if (st)
1458 st->ss->free_super(st);
1459 close(mdfd);
1460 free(devices);
1461 free(devmap);
1462 return 1;
1463 }
1464
1465 if (c->update && strcmp(c->update, "byteorder")==0)
1466 st->minor_version = 90;
1467
1468 st->ss->getinfo_super(st, content, NULL);
1469 clean = content->array.state & 1;
1470
1471 /* now we have some devices that might be suitable.
1472 * I wonder how many
1473 */
1474 avail = xcalloc(content->array.raid_disks, 1);
1475 okcnt = 0;
1476 replcnt = 0;
1477 sparecnt=0;
1478 rebuilding_cnt=0;
1479 for (i=0; i< bestcnt; i++) {
1480 int j = best[i];
1481 int event_margin = 1; /* always allow a difference of '1'
1482 * like the kernel does
1483 */
1484 if (j < 0) continue;
1485 /* note: we ignore error flags in multipath arrays
1486 * as they don't make sense
1487 */
1488 if (content->array.level != LEVEL_MULTIPATH)
1489 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1490 if (!(devices[j].i.disk.state
1491 & (1<<MD_DISK_FAULTY))) {
1492 devices[j].uptodate = 1;
1493 sparecnt++;
1494 }
1495 continue;
1496 }
1497 /* If this device thinks that 'most_recent' has failed, then
1498 * we must reject this device.
1499 */
1500 if (j != most_recent && !c->force &&
1501 content->array.raid_disks > 0 &&
1502 devices[most_recent].i.disk.raid_disk >= 0 &&
1503 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1504 if (c->verbose > -1)
1505 pr_err("ignoring %s as it reports %s as failed\n",
1506 devices[j].devname, devices[most_recent].devname);
1507 best[i] = -1;
1508 continue;
1509 }
1510 /* Require event counter to be same as, or just less than,
1511 * most recent. If it is bigger, it must be a stray spare and
1512 * should be ignored.
1513 */
1514 if (devices[j].i.events+event_margin >=
1515 devices[most_recent].i.events &&
1516 devices[j].i.events <=
1517 devices[most_recent].i.events
1518 ) {
1519 devices[j].uptodate = 1;
1520 if (i < content->array.raid_disks * 2) {
1521 if (devices[j].i.recovery_start == MaxSector ||
1522 (content->reshape_active &&
1523 ((i >= content->array.raid_disks - content->delta_disks) ||
1524 (i >= content->array.raid_disks - content->delta_disks - 1
1525 && content->array.level == 4)))) {
1526 if (!avail[i/2]) {
1527 okcnt++;
1528 avail[i/2]=1;
1529 } else
1530 replcnt++;
1531 } else
1532 rebuilding_cnt++;
1533 } else
1534 sparecnt++;
1535 }
1536 }
1537 free(devmap);
1538 if (c->force) {
1539 int force_ok = force_array(content, devices, best, bestcnt,
1540 avail, most_recent, st, c);
1541 okcnt += force_ok;
1542 if (force_ok)
1543 was_forced = 1;
1544 }
1545 /* Now we want to look at the superblock which the kernel will base things on
1546 * and compare the devices that we think are working with the devices that the
1547 * superblock thinks are working.
1548 * If there are differences and --force is given, then update this chosen
1549 * superblock.
1550 */
1551 chosen_drive = -1;
1552 st->ss->free_super(st);
1553 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1554 int j = best[i];
1555 int fd;
1556
1557 if (j<0)
1558 continue;
1559 if (!devices[j].uptodate)
1560 continue;
1561 if (devices[j].i.events < devices[most_recent].i.events)
1562 continue;
1563 chosen_drive = j;
1564 if ((fd=dev_open(devices[j].devname,
1565 devices[j].included ? O_RDONLY
1566 : (O_RDONLY|O_EXCL)))< 0) {
1567 pr_err("Cannot open %s: %s\n",
1568 devices[j].devname, strerror(errno));
1569 close(mdfd);
1570 free(devices);
1571 return 1;
1572 }
1573 if (st->ss->load_super(st,fd, NULL)) {
1574 close(fd);
1575 pr_err("RAID superblock has disappeared from %s\n",
1576 devices[j].devname);
1577 close(mdfd);
1578 free(devices);
1579 return 1;
1580 }
1581 close(fd);
1582 }
1583 if (st->sb == NULL) {
1584 pr_err("No suitable drives found for %s\n", mddev);
1585 close(mdfd);
1586 free(devices);
1587 return 1;
1588 }
1589 st->ss->getinfo_super(st, content, NULL);
1590 #ifndef MDASSEMBLE
1591 sysfs_init(content, mdfd, NULL);
1592 #endif
1593 for (i=0; i<bestcnt; i++) {
1594 int j = best[i];
1595 unsigned int desired_state;
1596
1597 if (i >= content->array.raid_disks * 2)
1598 desired_state = 0;
1599 else if (i & 1)
1600 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1601 else
1602 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1603
1604 if (j<0)
1605 continue;
1606 if (!devices[j].uptodate)
1607 continue;
1608
1609 devices[j].i.disk.state = desired_state;
1610 if (!(devices[j].i.array.state & 1))
1611 clean = 0;
1612
1613 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1614 c->verbose, 0, NULL)) {
1615 if (c->force) {
1616 if (c->verbose >= 0)
1617 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1618 j, mddev, devices[j].devname);
1619 change = 1;
1620 } else {
1621 if (c->verbose >= -1)
1622 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1623 i, mddev, devices[j].devname);
1624 }
1625 }
1626 #if 0
1627 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1628 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1629 i, mddev);
1630 }
1631 #endif
1632 }
1633 if (c->force && !clean &&
1634 !enough(content->array.level, content->array.raid_disks,
1635 content->array.layout, clean,
1636 avail)) {
1637 change += st->ss->update_super(st, content, "force-array",
1638 devices[chosen_drive].devname, c->verbose,
1639 0, NULL);
1640 was_forced = 1;
1641 clean = 1;
1642 }
1643
1644 if (change) {
1645 int fd;
1646 fd = dev_open(devices[chosen_drive].devname,
1647 devices[chosen_drive].included ?
1648 O_RDWR : (O_RDWR|O_EXCL));
1649 if (fd < 0) {
1650 pr_err("Could not open %s for write - cannot Assemble array.\n",
1651 devices[chosen_drive].devname);
1652 close(mdfd);
1653 free(devices);
1654 return 1;
1655 }
1656 if (st->ss->store_super(st, fd)) {
1657 close(fd);
1658 pr_err("Could not re-write superblock on %s\n",
1659 devices[chosen_drive].devname);
1660 close(mdfd);
1661 free(devices);
1662 return 1;
1663 }
1664 if (c->verbose >= 0)
1665 pr_err("Marking array %s as 'clean'\n",
1666 mddev);
1667 close(fd);
1668 }
1669
1670 /* If we are in the middle of a reshape we may need to restore saved data
1671 * that was moved aside due to the reshape overwriting live data
1672 * The code of doing this lives in Grow.c
1673 */
1674 #ifndef MDASSEMBLE
1675 if (content->reshape_active &&
1676 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1677 int err = 0;
1678 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1679 if (c->verbose > 0)
1680 pr_err(":%s has an active reshape - checking "
1681 "if critical section needs to be restored\n",
1682 chosen_name);
1683 if (!c->backup_file)
1684 c->backup_file = locate_backup(content->sys_name);
1685 enable_fds(bestcnt/2);
1686 for (i = 0; i < bestcnt/2; i++) {
1687 int j = best[i*2];
1688 if (j >= 0) {
1689 fdlist[i] = dev_open(devices[j].devname,
1690 devices[j].included
1691 ? O_RDWR : (O_RDWR|O_EXCL));
1692 if (fdlist[i] < 0) {
1693 pr_err("Could not open %s for write - cannot Assemble array.\n",
1694 devices[j].devname);
1695 err = 1;
1696 break;
1697 }
1698 } else
1699 fdlist[i] = -1;
1700 }
1701 if (!err) {
1702 if (st->ss->external && st->ss->recover_backup)
1703 err = st->ss->recover_backup(st, content);
1704 else
1705 err = Grow_restart(st, content, fdlist, bestcnt/2,
1706 c->backup_file, c->verbose > 0);
1707 if (err && c->invalid_backup) {
1708 if (c->verbose > 0)
1709 pr_err("continuing"
1710 " without restoring backup\n");
1711 err = 0;
1712 }
1713 }
1714 while (i>0) {
1715 i--;
1716 if (fdlist[i]>=0) close(fdlist[i]);
1717 }
1718 free(fdlist);
1719 if (err) {
1720 pr_err("Failed to restore critical section for reshape, sorry.\n");
1721 if (c->backup_file == NULL)
1722 cont_err("Possibly you needed to specify the --backup-file\n");
1723 close(mdfd);
1724 free(devices);
1725 return err;
1726 }
1727 }
1728 #endif
1729
1730 /* Almost ready to actually *do* something */
1731 /* First, fill in the map, so that udev can find our name
1732 * as soon as we become active.
1733 */
1734 if (c->update && strcmp(c->update, "metadata")==0) {
1735 content->array.major_version = 1;
1736 content->array.minor_version = 0;
1737 strcpy(content->text_version, "1.0");
1738 }
1739
1740 map_update(&map, fd2devnm(mdfd), content->text_version,
1741 content->uuid, chosen_name);
1742
1743 rv = start_array(mdfd, mddev, content,
1744 st, ident, best, bestcnt,
1745 chosen_drive, devices, okcnt, sparecnt,
1746 rebuilding_cnt,
1747 c,
1748 clean, avail, start_partial_ok,
1749 pre_exist != NULL,
1750 was_forced);
1751 if (rv == 1 && !pre_exist)
1752 ioctl(mdfd, STOP_ARRAY, NULL);
1753 free(devices);
1754 map_unlock(&map);
1755 if (rv == 0) {
1756 wait_for(chosen_name, mdfd);
1757 close(mdfd);
1758 if (auto_assem) {
1759 int usecs = 1;
1760 /* There is a nasty race with 'mdadm --monitor'.
1761 * If it opens this device before we close it,
1762 * it gets an incomplete open on which IO
1763 * doesn't work and the capacity is
1764 * wrong.
1765 * If we reopen (to check for layered devices)
1766 * before --monitor closes, we loose.
1767 *
1768 * So: wait upto 1 second for there to be
1769 * a non-zero capacity.
1770 */
1771 while (usecs < 1000) {
1772 mdfd = open(mddev, O_RDONLY);
1773 if (mdfd >= 0) {
1774 unsigned long long size;
1775 if (get_dev_size(mdfd, NULL, &size) &&
1776 size > 0)
1777 break;
1778 close(mdfd);
1779 }
1780 usleep(usecs);
1781 usecs <<= 1;
1782 }
1783 }
1784 } else
1785 close(mdfd);
1786
1787 /* '2' means 'OK, but not started yet' */
1788 return rv == 2 ? 0 : rv;
1789 }
1790
1791 #ifndef MDASSEMBLE
1792 int assemble_container_content(struct supertype *st, int mdfd,
1793 struct mdinfo *content, struct context *c,
1794 char *chosen_name, int *result)
1795 {
1796 struct mdinfo *dev, *sra;
1797 int working = 0, preexist = 0;
1798 int expansion = 0;
1799 struct map_ent *map = NULL;
1800 int old_raid_disks;
1801 int start_reshape;
1802 char *avail = NULL;
1803 int err;
1804
1805 sysfs_init(content, mdfd, NULL);
1806
1807 sra = sysfs_read(mdfd, NULL, GET_VERSION);
1808 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1809 if (content->array.major_version == -1 &&
1810 content->array.minor_version == -2 &&
1811 c->readonly &&
1812 content->text_version[0] == '/')
1813 content->text_version[0] = '-';
1814 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1815 if (sra)
1816 sysfs_free(sra);
1817 return 1;
1818 }
1819 }
1820
1821 /* There are two types of reshape: container wide or sub-array specific
1822 * Check if metadata requests blocking container wide reshapes
1823 */
1824 start_reshape = (content->reshape_active &&
1825 !((content->reshape_active == CONTAINER_RESHAPE) &&
1826 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1827
1828 /* Block subarray here if it is under reshape now
1829 * Do not allow for any changes in this array
1830 */
1831 if (st->ss->external && content->recovery_blocked && start_reshape)
1832 block_subarray(content);
1833
1834 if (sra)
1835 sysfs_free(sra);
1836 old_raid_disks = content->array.raid_disks - content->delta_disks;
1837 avail = xcalloc(content->array.raid_disks, 1);
1838 for (dev = content->devs; dev; dev = dev->next) {
1839 if (dev->disk.raid_disk >= 0)
1840 avail[dev->disk.raid_disk] = 1;
1841 if (sysfs_add_disk(content, dev, 1) == 0) {
1842 if (dev->disk.raid_disk >= old_raid_disks &&
1843 content->reshape_active)
1844 expansion++;
1845 else
1846 working++;
1847 } else if (errno == EEXIST)
1848 preexist++;
1849 }
1850 if (working + expansion == 0 && c->runstop <= 0) {
1851 free(avail);
1852 return 1;/* Nothing new, don't try to start */
1853 }
1854 map_update(&map, fd2devnm(mdfd),
1855 content->text_version,
1856 content->uuid, chosen_name);
1857
1858
1859 if (enough(content->array.level, content->array.raid_disks,
1860 content->array.layout, content->array.state & 1, avail) == 0) {
1861 if (c->export && result)
1862 *result |= INCR_NO;
1863 else if (c->verbose >= 0) {
1864 pr_err("%s assembled with %d device%s",
1865 chosen_name, preexist + working,
1866 preexist + working == 1 ? "":"s");
1867 if (preexist)
1868 fprintf(stderr, " (%d new)", working);
1869 fprintf(stderr, " but not started\n");
1870 }
1871 free(avail);
1872 return 1;
1873 }
1874 free(avail);
1875
1876 if (c->runstop <= 0 &&
1877 (working + preexist + expansion) <
1878 content->array.working_disks) {
1879 if (c->export && result)
1880 *result |= INCR_UNSAFE;
1881 else if (c->verbose >= 0) {
1882 pr_err("%s assembled with %d device%s",
1883 chosen_name, preexist + working,
1884 preexist + working == 1 ? "":"s");
1885 if (preexist)
1886 fprintf(stderr, " (%d new)", working);
1887 fprintf(stderr, " but not safe to start\n");
1888 }
1889 return 1;
1890 }
1891
1892
1893 if (start_reshape) {
1894 int spare = content->array.raid_disks + expansion;
1895 if (restore_backup(st, content,
1896 working,
1897 spare, &c->backup_file, c->verbose) == 1)
1898 return 1;
1899
1900 err = sysfs_set_str(content, NULL,
1901 "array_state", "readonly");
1902 if (err)
1903 return 1;
1904
1905 if (st->ss->external) {
1906 if (!mdmon_running(st->container_devnm))
1907 start_mdmon(st->container_devnm);
1908 ping_monitor(st->container_devnm);
1909 if (mdmon_running(st->container_devnm) &&
1910 st->update_tail == NULL)
1911 st->update_tail = &st->updates;
1912 }
1913
1914 err = Grow_continue(mdfd, st, content, c->backup_file,
1915 0, c->freeze_reshape);
1916 } else switch(content->array.level) {
1917 case LEVEL_LINEAR:
1918 case LEVEL_MULTIPATH:
1919 case 0:
1920 err = sysfs_set_str(content, NULL, "array_state",
1921 c->readonly ? "readonly" : "active");
1922 break;
1923 default:
1924 err = sysfs_set_str(content, NULL, "array_state",
1925 "readonly");
1926 /* start mdmon if needed. */
1927 if (!err) {
1928 if (!mdmon_running(st->container_devnm))
1929 start_mdmon(st->container_devnm);
1930 ping_monitor(st->container_devnm);
1931 }
1932 break;
1933 }
1934 if (!err)
1935 sysfs_set_safemode(content, content->safe_mode_delay);
1936
1937 /* Block subarray here if it is not reshaped now
1938 * It has be blocked a little later to allow mdmon to switch in
1939 * in to R/W state
1940 */
1941 if (st->ss->external && content->recovery_blocked &&
1942 !start_reshape)
1943 block_subarray(content);
1944
1945 if (c->export && result) {
1946 if (err)
1947 *result |= INCR_NO;
1948 else
1949 *result |= INCR_YES;
1950 } else if (c->verbose >= 0) {
1951 if (err)
1952 pr_err("array %s now has %d device%s",
1953 chosen_name, working + preexist,
1954 working + preexist == 1 ? "":"s");
1955 else
1956 pr_err("Started %s with %d device%s",
1957 chosen_name, working + preexist,
1958 working + preexist == 1 ? "":"s");
1959 if (preexist)
1960 fprintf(stderr, " (%d new)", working);
1961 if (expansion)
1962 fprintf(stderr, " ( + %d for expansion)",
1963 expansion);
1964 fprintf(stderr, "\n");
1965 }
1966 if (!err)
1967 wait_for(chosen_name, mdfd);
1968 return err;
1969 /* FIXME should have an O_EXCL and wait for read-auto */
1970 }
1971 #endif
Unnamed repository; edit this file 'description' to name the repository.