]> 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: don't ignore the return value from stat.
[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 auto-assembly is disabled\n",
237 devname, tst->ss->name);
238 tmpdev->used = 2;
239 } else
240 found_container = 1;
241 } else {
242 if (!tst && (tst = guess_super(dfd)) == NULL) {
243 if (report_mismatch)
244 pr_err("no recogniseable superblock on %s\n",
245 devname);
246 tmpdev->used = 2;
247 } else if (tst->ss->load_super(tst,dfd, NULL)) {
248 if (report_mismatch)
249 pr_err("no RAID superblock on %s\n",
250 devname);
251 tmpdev->used = 2;
252 } else if (tst->ss->compare_super == NULL) {
253 if (report_mismatch)
254 pr_err("Cannot assemble %s metadata on %s\n",
255 tst->ss->name, devname);
256 tmpdev->used = 2;
257 } else if (auto_assem && st == NULL &&
258 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
259 tst->ss->match_home(tst, c->homehost) == 1)) {
260 if (report_mismatch)
261 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
262 devname, tst->ss->name);
263 tmpdev->used = 2;
264 }
265 }
266 if (dfd >= 0) close(dfd);
267 if (tmpdev->used == 2) {
268 if (auto_assem || !inargv)
269 /* Ignore unrecognised devices during auto-assembly */
270 goto loop;
271 if (ident->uuid_set || ident->name[0] ||
272 ident->super_minor != UnSet)
273 /* Ignore unrecognised device if looking for
274 * specific array */
275 goto loop;
276
277 pr_err("%s has no superblock - assembly aborted\n",
278 devname);
279 if (st)
280 st->ss->free_super(st);
281 dev_policy_free(pol);
282 domain_free(domains);
283 return -1;
284 }
285
286 if (found_container) {
287 /* tmpdev is a container. We need to be either
288 * looking for a member, or auto-assembling
289 */
290 /* should be safe to try an exclusive open now, we
291 * have rejected anything that some other mdadm might
292 * be looking at
293 */
294 dfd = dev_open(devname, O_RDONLY | O_EXCL);
295 if (dfd < 0) {
296 if (report_mismatch)
297 pr_err("%s is busy - skipping\n", devname);
298 goto loop;
299 }
300 close(dfd);
301
302 if (ident->container && ident->container[0] != '/') {
303 /* we have a uuid */
304 int uuid[4];
305
306 content = *contentp;
307 tst->ss->getinfo_super(tst, content, NULL);
308
309 if (!parse_uuid(ident->container, uuid) ||
310 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
311 if (report_mismatch)
312 pr_err("%s has wrong UUID to be required container\n",
313 devname);
314 goto loop;
315 }
316 }
317 /* It is worth looking inside this container.
318 */
319 if (c->verbose > 0)
320 pr_err("looking in container %s\n",
321 devname);
322
323 for (content = tst->ss->container_content(tst, NULL);
324 content;
325 content = content->next) {
326
327 if (!ident_matches(ident, content, tst,
328 c->homehost, c->update,
329 report_mismatch ? devname : NULL))
330 /* message already printed */;
331 else if (is_member_busy(content->text_version)) {
332 if (report_mismatch)
333 pr_err("member %s in %s is already assembled\n",
334 content->text_version,
335 devname);
336 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
337 /* do not assemble arrays with unsupported configurations */
338 pr_err("Cannot activate member %s in %s.\n",
339 content->text_version,
340 devname);
341 } else
342 break;
343 }
344 if (!content) {
345 tmpdev->used = 2;
346 goto loop; /* empty container */
347 }
348
349 st = tst; tst = NULL;
350 if (!auto_assem && inargv && tmpdev->next != NULL) {
351 pr_err("%s is a container, but is not 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 content = *contentp;
367 tst->ss->getinfo_super(tst, content, NULL);
368
369 if (!ident_matches(ident, content, tst,
370 c->homehost, c->update,
371 report_mismatch ? devname : NULL))
372 goto loop;
373
374 if (auto_assem) {
375 /* Never auto-assemble things that conflict
376 * with mdadm.conf in some way
377 */
378 struct mddev_ident *match;
379 int rv = 0;
380
381 match = conf_match(tst, content, devname,
382 report_mismatch ? c->verbose : -1,
383 &rv);
384 if (!match && rv == 2)
385 goto loop;
386 if (match && match->devname &&
387 strcasecmp(match->devname, "<ignore>") == 0) {
388 if (report_mismatch)
389 pr_err("%s is a member of an explicitly ignored array\n",
390 devname);
391 goto loop;
392 }
393 if (match && !ident_matches(match, content, tst,
394 c->homehost, c->update,
395 report_mismatch ? devname : NULL))
396 /* Array exists in mdadm.conf but some
397 * details don't match, so reject it
398 */
399 goto loop;
400 }
401
402 /* should be safe to try an exclusive open now, we
403 * have rejected anything that some other mdadm might
404 * be looking at
405 */
406 dfd = dev_open(devname, O_RDONLY | O_EXCL);
407 if (dfd < 0) {
408 if (report_mismatch)
409 pr_err("%s is busy - skipping\n", devname);
410 goto loop;
411 }
412 close(dfd);
413
414 if (st == NULL)
415 st = dup_super(tst);
416 if (st->minor_version == -1)
417 st->minor_version = tst->minor_version;
418
419 if (memcmp(content->uuid, uuid_zero,
420 sizeof(int[4])) == 0) {
421 /* this is a floating spare. It cannot define
422 * an array unless there are no more arrays of
423 * this type to be found. It can be included
424 * in an array of this type though.
425 */
426 tmpdev->used = 3;
427 goto loop;
428 }
429
430 if (st->ss != tst->ss ||
431 st->minor_version != tst->minor_version ||
432 st->ss->compare_super(st, tst) != 0) {
433 /* Some mismatch. If exactly one array matches this host,
434 * we can resolve on that one.
435 * Or, if we are auto assembling, we just ignore the second
436 * for now.
437 */
438 if (auto_assem)
439 goto loop;
440 if (c->homehost) {
441 int first = st->ss->match_home(st, c->homehost);
442 int last = tst->ss->match_home(tst, c->homehost);
443 if (first != last &&
444 (first == 1 || last == 1)) {
445 /* We can do something */
446 if (first) {/* just ignore this one */
447 if (report_mismatch)
448 pr_err("%s misses out due to wrong homehost\n",
449 devname);
450 goto loop;
451 } else { /* reject all those sofar */
452 struct mddev_dev *td;
453 if (report_mismatch)
454 pr_err("%s overrides previous devices due to good homehost\n",
455 devname);
456 for (td=devlist; td != tmpdev; td=td->next)
457 if (td->used == 1)
458 td->used = 0;
459 tmpdev->used = 1;
460 goto loop;
461 }
462 }
463 }
464 pr_err("superblock on %s doesn't match others - assembly aborted\n",
465 devname);
466 tst->ss->free_super(tst);
467 st->ss->free_super(st);
468 dev_policy_free(pol);
469 domain_free(domains);
470 return -1;
471 }
472 tmpdev->used = 1;
473 }
474 loop:
475 /* Collect domain information from members only */
476 if (tmpdev && tmpdev->used == 1) {
477 if (!pol)
478 pol = devid_policy(stb.st_rdev);
479 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
480 }
481 dev_policy_free(pol);
482 pol = NULL;
483 if (tst)
484 tst->ss->free_super(tst);
485 }
486
487 /* Check if we found some imsm spares but no members */
488 if ((auto_assem ||
489 (ident->uuid_set &&
490 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
491 (!st || !st->sb))
492 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
493 if (tmpdev->used != 3)
494 continue;
495 tmpdev->used = 1;
496 content = *contentp;
497
498 if (!st->sb) {
499 /* we need sb from one of the spares */
500 int dfd = dev_open(tmpdev->devname, O_RDONLY);
501 if (dfd < 0 ||
502 st->ss->load_super(st, dfd, NULL))
503 tmpdev->used = 2;
504 if (dfd > 0)
505 close(dfd);
506 }
507 }
508
509 /* Now reject spares that don't match domains of identified members */
510 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
511 struct stat stb;
512 if (tmpdev->used != 3)
513 continue;
514 if (stat(tmpdev->devname, &stb)< 0) {
515 pr_err("fstat failed for %s: %s\n",
516 tmpdev->devname, strerror(errno));
517 tmpdev->used = 2;
518 } else {
519 struct dev_policy *pol = devid_policy(stb.st_rdev);
520 int dt = domain_test(domains, pol, NULL);
521 if (inargv && dt != 0)
522 /* take this spare as domains match
523 * if there are any */
524 tmpdev->used = 1;
525 else if (!inargv && dt == 1)
526 /* device wasn't explicitly listed, so need
527 * explicit domain match - which we have */
528 tmpdev->used = 1;
529 else
530 /* if domains don't match mark as unused */
531 tmpdev->used = 0;
532 dev_policy_free(pol);
533 }
534 }
535 domain_free(domains);
536 *stp = st;
537 if (st && st->sb && content == *contentp)
538 st->ss->getinfo_super(st, content, NULL);
539 *contentp = content;
540
541 return num_devs;
542 }
543
544 struct devs {
545 char *devname;
546 int uptodate; /* set once we decide that this device is as
547 * recent as everything else in the array.
548 */
549 int included; /* set if the device is already in the array
550 * due to a previous '-I'
551 */
552 struct mdinfo i;
553 };
554
555 static int load_devices(struct devs *devices, char *devmap,
556 struct mddev_ident *ident, struct supertype **stp,
557 struct mddev_dev *devlist, struct context *c,
558 struct mdinfo *content,
559 int mdfd, char *mddev,
560 int *most_recentp, int *bestcntp, int **bestp,
561 int inargv)
562 {
563 struct mddev_dev *tmpdev;
564 int devcnt = 0;
565 int nextspare = 0;
566 #ifndef MDASSEMBLE
567 int bitmap_done = 0;
568 #endif
569 int most_recent = -1;
570 int bestcnt = 0;
571 int *best = *bestp;
572 struct supertype *st = *stp;
573
574 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
575 char *devname = tmpdev->devname;
576 struct stat stb;
577 struct supertype *tst;
578 int i;
579 int dfd;
580
581 if (tmpdev->used != 1)
582 continue;
583 /* looks like a good enough match to update the super block if needed */
584 #ifndef MDASSEMBLE
585 if (c->update) {
586 /* prepare useful information in info structures */
587 struct stat stb2;
588 int err;
589 fstat(mdfd, &stb2);
590
591 if (strcmp(c->update, "uuid")==0 &&
592 !ident->uuid_set) {
593 int rfd;
594 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
595 read(rfd, ident->uuid, 16) != 16) {
596 *(__u32*)(ident->uuid) = random();
597 *(__u32*)(ident->uuid+1) = random();
598 *(__u32*)(ident->uuid+2) = random();
599 *(__u32*)(ident->uuid+3) = random();
600 }
601 if (rfd >= 0) close(rfd);
602 }
603 dfd = dev_open(devname,
604 tmpdev->disposition == 'I'
605 ? O_RDWR : (O_RDWR|O_EXCL));
606
607 tst = dup_super(st);
608 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
609 pr_err("cannot re-read metadata from %s - aborting\n",
610 devname);
611 if (dfd >= 0)
612 close(dfd);
613 close(mdfd);
614 free(devices);
615 free(devmap);
616 tst->ss->free_super(tst);
617 free(tst);
618 *stp = st;
619 return -1;
620 }
621 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
622
623 memcpy(content->uuid, ident->uuid, 16);
624 strcpy(content->name, ident->name);
625 content->array.md_minor = minor(stb2.st_rdev);
626
627 if (strcmp(c->update, "byteorder") == 0)
628 err = 0;
629 else
630 err = tst->ss->update_super(tst, content, c->update,
631 devname, c->verbose,
632 ident->uuid_set,
633 c->homehost);
634 if (err < 0) {
635 if (err == -1)
636 pr_err("--update=%s not understood for %s metadata\n",
637 c->update, tst->ss->name);
638 tst->ss->free_super(tst);
639 free(tst);
640 close(mdfd);
641 close(dfd);
642 free(devices);
643 free(devmap);
644 *stp = st;
645 return -1;
646 }
647 if (strcmp(c->update, "uuid")==0 &&
648 !ident->uuid_set) {
649 ident->uuid_set = 1;
650 memcpy(ident->uuid, content->uuid, 16);
651 }
652 if (tst->ss->store_super(tst, dfd))
653 pr_err("Could not re-write superblock on %s.\n",
654 devname);
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 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 }
688
689 fstat(dfd, &stb);
690 close(dfd);
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 to have very similar superblocks.\n"
761 " If they are really different, please --zero the superblock on one\n"
762 " If they are the same or overlap, please remove one from %s.\n",
763 devices[best[i]].devname, devname,
764 inargv ? "the list" :
765 "the\n DEVICE list in mdadm.conf"
766 );
767 close(mdfd);
768 free(devices);
769 free(devmap);
770 *stp = st;
771 return -1;
772 }
773 if (best[i] == -1
774 || (devices[best[i]].i.events
775 < devices[devcnt].i.events))
776 best[i] = devcnt;
777 }
778 devcnt++;
779 }
780 if (most_recent >= 0)
781 *most_recentp = most_recent;
782 *bestcntp = bestcnt;
783 *bestp = best;
784 *stp = st;
785 return devcnt;
786 }
787
788 static int force_array(struct mdinfo *content,
789 struct devs *devices,
790 int *best, int bestcnt, char *avail,
791 int most_recent,
792 struct supertype *st,
793 struct context *c)
794 {
795 int okcnt = 0;
796 while (!enough(content->array.level, content->array.raid_disks,
797 content->array.layout, 1,
798 avail)
799 ||
800 (content->reshape_active && content->delta_disks > 0 &&
801 !enough(content->array.level, (content->array.raid_disks
802 - content->delta_disks),
803 content->new_layout, 1,
804 avail)
805 )) {
806 /* Choose the newest best drive which is
807 * not up-to-date, update the superblock
808 * and add it.
809 */
810 int fd;
811 struct supertype *tst;
812 unsigned long long current_events;
813 int chosen_drive = -1;
814 int i;
815
816 for (i = 0;
817 i < content->array.raid_disks * 2 && i < bestcnt;
818 i += 2) {
819 int j = best[i];
820 if (j>=0 &&
821 !devices[j].uptodate &&
822 devices[j].i.recovery_start == MaxSector &&
823 (chosen_drive < 0 ||
824 devices[j].i.events
825 > devices[chosen_drive].i.events))
826 chosen_drive = j;
827 }
828 if (chosen_drive < 0)
829 break;
830 current_events = devices[chosen_drive].i.events;
831 add_another:
832 if (c->verbose >= 0)
833 pr_err("forcing event count in %s(%d) from %d upto %d\n",
834 devices[chosen_drive].devname,
835 devices[chosen_drive].i.disk.raid_disk,
836 (int)(devices[chosen_drive].i.events),
837 (int)(devices[most_recent].i.events));
838 fd = dev_open(devices[chosen_drive].devname,
839 devices[chosen_drive].included ? O_RDWR
840 : (O_RDWR|O_EXCL));
841 if (fd < 0) {
842 pr_err("Couldn't open %s for write - not updating\n",
843 devices[chosen_drive].devname);
844 devices[chosen_drive].i.events = 0;
845 continue;
846 }
847 tst = dup_super(st);
848 if (tst->ss->load_super(tst,fd, NULL)) {
849 close(fd);
850 pr_err("RAID superblock disappeared from %s - not updating.\n",
851 devices[chosen_drive].devname);
852 devices[chosen_drive].i.events = 0;
853 continue;
854 }
855 content->events = devices[most_recent].i.events;
856 tst->ss->update_super(tst, content, "force-one",
857 devices[chosen_drive].devname, c->verbose,
858 0, NULL);
859
860 if (tst->ss->store_super(tst, fd)) {
861 close(fd);
862 pr_err("Could not re-write superblock on %s\n",
863 devices[chosen_drive].devname);
864 devices[chosen_drive].i.events = 0;
865 tst->ss->free_super(tst);
866 continue;
867 }
868 close(fd);
869 devices[chosen_drive].i.events = devices[most_recent].i.events;
870 devices[chosen_drive].uptodate = 1;
871 avail[chosen_drive] = 1;
872 okcnt++;
873 tst->ss->free_super(tst);
874
875 /* If there are any other drives of the same vintage,
876 * add them in as well. We can't lose and we might gain
877 */
878 for (i = 0;
879 i < content->array.raid_disks * 2 && i < bestcnt ;
880 i += 2) {
881 int j = best[i];
882 if (j >= 0 &&
883 !devices[j].uptodate &&
884 devices[j].i.recovery_start == MaxSector &&
885 devices[j].i.events == current_events) {
886 chosen_drive = j;
887 goto add_another;
888 }
889 }
890 }
891 return okcnt;
892 }
893
894 static int start_array(int mdfd,
895 char *mddev,
896 struct mdinfo *content,
897 struct supertype *st,
898 struct mddev_ident *ident,
899 int *best, int bestcnt,
900 int chosen_drive,
901 struct devs *devices,
902 unsigned int okcnt,
903 unsigned int sparecnt,
904 unsigned int rebuilding_cnt,
905 struct context *c,
906 int clean, char *avail,
907 int start_partial_ok,
908 int err_ok,
909 int was_forced
910 )
911 {
912 int rv;
913 int i;
914 unsigned int req_cnt;
915
916 rv = set_array_info(mdfd, st, content);
917 if (rv && !err_ok) {
918 pr_err("failed to set array info for %s: %s\n",
919 mddev, strerror(errno));
920 return 1;
921 }
922 if (ident->bitmap_fd >= 0) {
923 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
924 pr_err("SET_BITMAP_FILE failed.\n");
925 return 1;
926 }
927 } else if (ident->bitmap_file) {
928 /* From config file */
929 int bmfd = open(ident->bitmap_file, O_RDWR);
930 if (bmfd < 0) {
931 pr_err("Could not open bitmap file %s\n",
932 ident->bitmap_file);
933 return 1;
934 }
935 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
936 pr_err("Failed to set bitmapfile for %s\n", mddev);
937 close(bmfd);
938 return 1;
939 }
940 close(bmfd);
941 }
942
943 /* First, add the raid disks, but add the chosen one last */
944 for (i=0; i<= bestcnt; i++) {
945 int j;
946 if (i < bestcnt) {
947 j = best[i];
948 if (j == chosen_drive)
949 continue;
950 } else
951 j = chosen_drive;
952
953 if (j >= 0 && !devices[j].included) {
954 int dfd = dev_open(devices[j].devname,
955 O_RDWR|O_EXCL);
956 if (dfd >= 0) {
957 remove_partitions(dfd);
958 close(dfd);
959 }
960 rv = add_disk(mdfd, st, content, &devices[j].i);
961
962 if (rv) {
963 pr_err("failed to add %s to %s: %s\n",
964 devices[j].devname,
965 mddev,
966 strerror(errno));
967 if (i < content->array.raid_disks * 2
968 || i == bestcnt)
969 okcnt--;
970 else
971 sparecnt--;
972 } else if (c->verbose > 0)
973 pr_err("added %s to %s as %d%s%s\n",
974 devices[j].devname, mddev,
975 devices[j].i.disk.raid_disk,
976 devices[j].uptodate?"":
977 " (possibly out of date)",
978 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
979 } else if (j >= 0) {
980 if (c->verbose > 0)
981 pr_err("%s is already in %s as %d\n",
982 devices[j].devname, mddev,
983 devices[j].i.disk.raid_disk);
984 } else if (c->verbose > 0 && i < content->array.raid_disks*2
985 && (i&1) == 0)
986 pr_err("no uptodate device for slot %d of %s\n",
987 i, mddev);
988 }
989
990 if (content->array.level == LEVEL_CONTAINER) {
991 if (c->verbose >= 0) {
992 pr_err("Container %s has been assembled with %d drive%s",
993 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
994 if (okcnt < (unsigned)content->array.raid_disks)
995 fprintf(stderr, " (out of %d)",
996 content->array.raid_disks);
997 fprintf(stderr, "\n");
998 }
999
1000 if (st->ss->validate_container) {
1001 struct mdinfo *devices_list;
1002 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1003 unsigned int count;
1004 devices_list = NULL;
1005 for (count = 0; count < okcnt+sparecnt; count++) {
1006 info_devices[count] = devices[count].i;
1007 info_devices[count].next = devices_list;
1008 devices_list = &info_devices[count];
1009 }
1010 if (st->ss->validate_container(devices_list))
1011 pr_err("Mismatch detected!\n");
1012 free(info_devices);
1013 }
1014
1015 st->ss->free_super(st);
1016 sysfs_uevent(content, "change");
1017 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1018 /* Was partial, is still partial, so signal an error
1019 * to ensure we don't retry */
1020 return 1;
1021 return 0;
1022 }
1023
1024 /* Get number of in-sync devices according to the superblock.
1025 * We must have this number to start the array without -s or -R
1026 */
1027 req_cnt = content->array.working_disks;
1028
1029 if (c->runstop == 1 ||
1030 (c->runstop <= 0 &&
1031 ( enough(content->array.level, content->array.raid_disks,
1032 content->array.layout, clean, avail) &&
1033 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1034 ))) {
1035 /* This array is good-to-go.
1036 * If a reshape is in progress then we might need to
1037 * continue monitoring it. In that case we start
1038 * it read-only and let the grow code make it writable.
1039 */
1040 int rv;
1041 #ifndef MDASSEMBLE
1042 if (content->reshape_active &&
1043 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1044 content->delta_disks <= 0) {
1045 if (!c->backup_file) {
1046 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1047 mddev);
1048 pr_err("Please provided one with \"--backup-file=...\"\n");
1049 if (c->update &&
1050 strcmp(c->update, "revert-reshape") == 0)
1051 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1052 return 1;
1053 }
1054 rv = sysfs_set_str(content, NULL,
1055 "array_state", "readonly");
1056 if (rv == 0)
1057 rv = Grow_continue(mdfd, st, content,
1058 c->backup_file, 0,
1059 c->freeze_reshape);
1060 } else if (c->readonly &&
1061 sysfs_attribute_available(
1062 content, NULL, "array_state")) {
1063 rv = sysfs_set_str(content, NULL,
1064 "array_state", "readonly");
1065 } else
1066 #endif
1067 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1068 reopen_mddev(mdfd); /* drop O_EXCL */
1069 if (rv == 0) {
1070 if (c->verbose >= 0) {
1071 pr_err("%s has been started with %d drive%s",
1072 mddev, okcnt, okcnt==1?"":"s");
1073 if (okcnt < (unsigned)content->array.raid_disks)
1074 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1075 if (rebuilding_cnt)
1076 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1077 if (sparecnt)
1078 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1079 fprintf(stderr, ".\n");
1080 }
1081 if (content->reshape_active &&
1082 content->array.level >= 4 &&
1083 content->array.level <= 6) {
1084 /* might need to increase the size
1085 * of the stripe cache - default is 256
1086 */
1087 if (256 < 4 * (content->array.chunk_size/4096)) {
1088 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1089 if (sra)
1090 sysfs_set_num(sra, NULL,
1091 "stripe_cache_size",
1092 (4 * content->array.chunk_size / 4096) + 1);
1093 sysfs_free(sra);
1094 }
1095 }
1096 if (okcnt < (unsigned)content->array.raid_disks) {
1097 /* If any devices did not get added
1098 * because the kernel rejected them based
1099 * on event count, try adding them
1100 * again providing the action policy is
1101 * 're-add' or greater. The bitmap
1102 * might allow them to be included, or
1103 * they will become spares.
1104 */
1105 for (i = 0; i < bestcnt; i++) {
1106 int j = best[i];
1107 if (j >= 0 && !devices[j].uptodate) {
1108 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1109 continue;
1110 rv = add_disk(mdfd, st, content,
1111 &devices[j].i);
1112 if (rv == 0 && c->verbose >= 0)
1113 pr_err("%s has been re-added.\n",
1114 devices[j].devname);
1115 }
1116 }
1117 }
1118 if (content->array.level == 6 &&
1119 okcnt + 1 == (unsigned)content->array.raid_disks &&
1120 was_forced) {
1121 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1122 if (sra)
1123 sysfs_set_str(sra, NULL,
1124 "sync_action", "repair");
1125 sysfs_free(sra);
1126 }
1127 return 0;
1128 }
1129 pr_err("failed to RUN_ARRAY %s: %s\n",
1130 mddev, strerror(errno));
1131
1132 if (!enough(content->array.level, content->array.raid_disks,
1133 content->array.layout, 1, avail))
1134 pr_err("Not enough devices to start the array.\n");
1135 else if (!enough(content->array.level,
1136 content->array.raid_disks,
1137 content->array.layout, clean,
1138 avail))
1139 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1140
1141 return 1;
1142 }
1143 if (c->runstop == -1) {
1144 pr_err("%s assembled from %d drive%s",
1145 mddev, okcnt, okcnt==1?"":"s");
1146 if (okcnt != (unsigned)content->array.raid_disks)
1147 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1148 fprintf(stderr, ", but not started.\n");
1149 return 2;
1150 }
1151 if (c->verbose >= -1) {
1152 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1153 if (rebuilding_cnt)
1154 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1155 if (sparecnt)
1156 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1157 if (!enough(content->array.level, content->array.raid_disks,
1158 content->array.layout, 1, avail))
1159 fprintf(stderr, " - not enough to start the array.\n");
1160 else if (!enough(content->array.level,
1161 content->array.raid_disks,
1162 content->array.layout, clean,
1163 avail))
1164 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1165 else {
1166 if (req_cnt == (unsigned)content->array.raid_disks)
1167 fprintf(stderr, " - need all %d to start it", req_cnt);
1168 else
1169 fprintf(stderr, " - need %d to start", req_cnt);
1170 fprintf(stderr, " (use --run to insist).\n");
1171 }
1172 }
1173 return 1;
1174 }
1175
1176 int Assemble(struct supertype *st, char *mddev,
1177 struct mddev_ident *ident,
1178 struct mddev_dev *devlist,
1179 struct context *c)
1180 {
1181 /*
1182 * The task of Assemble is to find a collection of
1183 * devices that should (according to their superblocks)
1184 * form an array, and to give this collection to the MD driver.
1185 * In Linux-2.4 and later, this involves submitting a
1186 * SET_ARRAY_INFO ioctl with no arg - to prepare
1187 * the array - and then submit a number of
1188 * ADD_NEW_DISK ioctls to add disks into
1189 * the array. Finally RUN_ARRAY might
1190 * be submitted to start the array.
1191 *
1192 * Much of the work of Assemble is in finding and/or
1193 * checking the disks to make sure they look right.
1194 *
1195 * If mddev is not set, then scan must be set and we
1196 * read through the config file for dev+uuid mapping
1197 * We recurse, setting mddev, for each device that
1198 * - isn't running
1199 * - has a valid uuid (or any uuid if !uuidset)
1200 *
1201 * If mddev is set, we try to determine state of md.
1202 * check version - must be at least 0.90.0
1203 * check kernel version. must be at least 2.4.
1204 * If not, we can possibly fall back on START_ARRAY
1205 * Try to GET_ARRAY_INFO.
1206 * If possible, give up
1207 * If not, try to STOP_ARRAY just to make sure
1208 *
1209 * If !uuidset and scan, look in conf-file for uuid
1210 * If not found, give up
1211 * If !devlist and scan and uuidset, get list of devs from conf-file
1212 *
1213 * For each device:
1214 * Check superblock - discard if bad
1215 * Check uuid (set if we don't have one) - discard if no match
1216 * Check superblock similarity if we have a superblock - discard if different
1217 * Record events, devicenum
1218 * This should give us a list of devices for the array
1219 * We should collect the most recent event number
1220 *
1221 * Count disks with recent enough event count
1222 * While force && !enough disks
1223 * Choose newest rejected disks, update event count
1224 * mark clean and rewrite superblock
1225 * If recent kernel:
1226 * SET_ARRAY_INFO
1227 * foreach device with recent events : ADD_NEW_DISK
1228 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1229 * If old kernel:
1230 * Check the device numbers in superblock are right
1231 * update superblock if any changes
1232 * START_ARRAY
1233 *
1234 */
1235 int rv;
1236 int mdfd;
1237 int clean;
1238 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1239 ident->super_minor == UnSet && ident->name[0] == 0
1240 && (ident->container == NULL || ident->member == NULL));
1241 struct devs *devices;
1242 char *devmap;
1243 int *best = NULL; /* indexed by raid_disk */
1244 int bestcnt = 0;
1245 int devcnt;
1246 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1247 int i;
1248 int was_forced = 0;
1249 int most_recent = 0;
1250 int chosen_drive;
1251 int change = 0;
1252 int inargv = 0;
1253 int start_partial_ok = (c->runstop >= 0) &&
1254 (c->force || devlist==NULL || auto_assem);
1255 int num_devs;
1256 struct mddev_dev *tmpdev;
1257 struct mdinfo info;
1258 struct mdinfo *content = NULL;
1259 struct mdinfo *pre_exist = NULL;
1260 char *avail;
1261 char *name = NULL;
1262 char chosen_name[1024];
1263 struct map_ent *map = NULL;
1264 struct map_ent *mp;
1265
1266 /*
1267 * If any subdevs are listed, then any that don't
1268 * match ident are discarded. Remainder must all match and
1269 * become the array.
1270 * If no subdevs, then we scan all devices in the config file, but
1271 * there must be something in the identity
1272 */
1273
1274 if (!devlist &&
1275 ident->uuid_set == 0 &&
1276 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1277 ident->name[0] == 0 &&
1278 (ident->container == NULL || ident->member == NULL) &&
1279 ident->devices == NULL) {
1280 pr_err("No identity information available for %s - cannot assemble.\n",
1281 mddev ? mddev : "further assembly");
1282 return 1;
1283 }
1284
1285 if (devlist == NULL)
1286 devlist = conf_get_devs();
1287 else if (mddev)
1288 inargv = 1;
1289
1290 try_again:
1291 /* We come back here when doing auto-assembly and attempting some
1292 * set of devices failed. Those are now marked as ->used==2 and
1293 * we ignore them and try again
1294 */
1295 if (!st && ident->st)
1296 st = ident->st;
1297 if (c->verbose>0)
1298 pr_err("looking for devices for %s\n",
1299 mddev ? mddev : "further assembly");
1300
1301 content = &info;
1302 if (st)
1303 st->ignore_hw_compat = 1;
1304 num_devs = select_devices(devlist, ident, &st, &content, c,
1305 inargv, auto_assem);
1306 if (num_devs < 0)
1307 return 1;
1308
1309 if (!st || !st->sb || !content)
1310 return 2;
1311
1312 /* We have a full set of devices - we now need to find the
1313 * array device.
1314 * However there is a risk that we are racing with "mdadm -I"
1315 * and the array is already partially assembled - we will have
1316 * rejected any devices already in this address.
1317 * So we take a lock on the map file - to prevent further races -
1318 * and look for the uuid in there. If found and the array is
1319 * active, we abort. If found and the array is not active
1320 * we commit to that md device and add all the contained devices
1321 * to our list. We flag them so that we don't try to re-add,
1322 * but can remove if they turn out to not be wanted.
1323 */
1324 if (map_lock(&map))
1325 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1326 mp = map_by_uuid(&map, content->uuid);
1327 if (mp) {
1328 struct mdinfo *dv;
1329 /* array already exists. */
1330 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1331 if (pre_exist->array.level != UnSet) {
1332 pr_err("Found some drive for an array that is already active: %s\n",
1333 mp->path);
1334 pr_err("giving up.\n");
1335 return 1;
1336 }
1337 for (dv = pre_exist->devs; dv; dv = dv->next) {
1338 /* We want to add this device to our list,
1339 * but it could already be there if "mdadm -I"
1340 * started *after* we checked for O_EXCL.
1341 * If we add it to the top of the list
1342 * it will be preferred over later copies.
1343 */
1344 struct mddev_dev *newdev;
1345 char *devname = map_dev(dv->disk.major,
1346 dv->disk.minor,
1347 0);
1348 if (!devname)
1349 continue;
1350 newdev = xmalloc(sizeof(*newdev));
1351 newdev->devname = devname;
1352 newdev->disposition = 'I';
1353 newdev->used = 1;
1354 newdev->next = devlist;
1355 devlist = newdev;
1356 num_devs++;
1357 }
1358 strcpy(chosen_name, mp->path);
1359 if (c->verbose > 0 || mddev == NULL ||
1360 strcmp(mddev, chosen_name) != 0)
1361 pr_err("Merging with already-assembled %s\n",
1362 chosen_name);
1363 mdfd = open_dev_excl(mp->devnm);
1364 } else {
1365 int trustworthy = FOREIGN;
1366 name = content->name;
1367 switch (st->ss->match_home(st, c->homehost)
1368 ?: st->ss->match_home(st, "any")) {
1369 case 1:
1370 trustworthy = LOCAL;
1371 name = strchr(content->name, ':');
1372 if (name)
1373 name++;
1374 else
1375 name = content->name;
1376 break;
1377 }
1378 if (!auto_assem)
1379 /* If the array is listed in mdadm.conf or on
1380 * command line, then we trust the name
1381 * even if the array doesn't look local
1382 */
1383 trustworthy = LOCAL;
1384
1385 if (name[0] == 0 &&
1386 content->array.level == LEVEL_CONTAINER) {
1387 name = content->text_version;
1388 trustworthy = METADATA;
1389 }
1390
1391 if (name[0] && trustworthy != LOCAL &&
1392 ! c->require_homehost &&
1393 conf_name_is_free(name))
1394 trustworthy = LOCAL;
1395
1396 if (trustworthy == LOCAL &&
1397 strchr(name, ':'))
1398 /* Ignore 'host:' prefix of name */
1399 name = strchr(name, ':')+1;
1400
1401 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1402 chosen_name);
1403 }
1404 if (mdfd < 0) {
1405 st->ss->free_super(st);
1406 if (auto_assem)
1407 goto try_again;
1408 return 1;
1409 }
1410 mddev = chosen_name;
1411 if (get_linux_version() < 2004000 ||
1412 md_get_version(mdfd) < 9000) {
1413 pr_err("Assemble requires Linux 2.4 or later, and\n"
1414 " md driver version 0.90.0 or later.\n"
1415 " Upgrade your kernel or try --build\n");
1416 close(mdfd);
1417 return 1;
1418 }
1419 if (pre_exist == NULL) {
1420 if (mddev_busy(fd2devnm(mdfd))) {
1421 pr_err("%s already active, cannot restart it!\n",
1422 mddev);
1423 for (tmpdev = devlist ;
1424 tmpdev && tmpdev->used != 1;
1425 tmpdev = tmpdev->next)
1426 ;
1427 if (tmpdev && auto_assem)
1428 pr_err("%s needed for %s...\n",
1429 mddev, tmpdev->devname);
1430 close(mdfd);
1431 mdfd = -3;
1432 st->ss->free_super(st);
1433 if (auto_assem)
1434 goto try_again;
1435 return 1;
1436 }
1437 /* just incase it was started but has no content */
1438 ioctl(mdfd, STOP_ARRAY, NULL);
1439 }
1440
1441 #ifndef MDASSEMBLE
1442 if (content != &info) {
1443 /* This is a member of a container. Try starting the array. */
1444 int err;
1445 err = assemble_container_content(st, mdfd, content, c,
1446 chosen_name, NULL);
1447 close(mdfd);
1448 return err;
1449 }
1450 #endif
1451 /* Ok, no bad inconsistancy, we can try updating etc */
1452 devices = xcalloc(num_devs, sizeof(*devices));
1453 devmap = xcalloc(num_devs, content->array.raid_disks);
1454 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1455 c, content, mdfd, mddev,
1456 &most_recent, &bestcnt, &best, inargv);
1457 if (devcnt < 0)
1458 return 1;
1459
1460 if (devcnt == 0) {
1461 pr_err("no devices found for %s\n",
1462 mddev);
1463 if (st)
1464 st->ss->free_super(st);
1465 close(mdfd);
1466 free(devices);
1467 free(devmap);
1468 return 1;
1469 }
1470
1471 if (c->update && strcmp(c->update, "byteorder")==0)
1472 st->minor_version = 90;
1473
1474 st->ss->getinfo_super(st, content, NULL);
1475 clean = content->array.state & 1;
1476
1477 /* now we have some devices that might be suitable.
1478 * I wonder how many
1479 */
1480 avail = xcalloc(content->array.raid_disks, 1);
1481 okcnt = 0;
1482 replcnt = 0;
1483 sparecnt=0;
1484 rebuilding_cnt=0;
1485 for (i=0; i< bestcnt; i++) {
1486 int j = best[i];
1487 int event_margin = 1; /* always allow a difference of '1'
1488 * like the kernel does
1489 */
1490 if (j < 0) continue;
1491 /* note: we ignore error flags in multipath arrays
1492 * as they don't make sense
1493 */
1494 if (content->array.level != LEVEL_MULTIPATH)
1495 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1496 if (!(devices[j].i.disk.state
1497 & (1<<MD_DISK_FAULTY))) {
1498 devices[j].uptodate = 1;
1499 sparecnt++;
1500 }
1501 continue;
1502 }
1503 /* If this device thinks that 'most_recent' has failed, then
1504 * we must reject this device.
1505 */
1506 if (j != most_recent && !c->force &&
1507 content->array.raid_disks > 0 &&
1508 devices[most_recent].i.disk.raid_disk >= 0 &&
1509 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1510 if (c->verbose > -1)
1511 pr_err("ignoring %s as it reports %s as failed\n",
1512 devices[j].devname, devices[most_recent].devname);
1513 best[i] = -1;
1514 continue;
1515 }
1516 /* Require event counter to be same as, or just less than,
1517 * most recent. If it is bigger, it must be a stray spare and
1518 * should be ignored.
1519 */
1520 if (devices[j].i.events+event_margin >=
1521 devices[most_recent].i.events &&
1522 devices[j].i.events <=
1523 devices[most_recent].i.events
1524 ) {
1525 devices[j].uptodate = 1;
1526 if (i < content->array.raid_disks * 2) {
1527 if (devices[j].i.recovery_start == MaxSector ||
1528 (content->reshape_active &&
1529 ((i >= content->array.raid_disks - content->delta_disks) ||
1530 (i >= content->array.raid_disks - content->delta_disks - 1
1531 && content->array.level == 4)))) {
1532 if (!avail[i/2]) {
1533 okcnt++;
1534 avail[i/2]=1;
1535 } else
1536 replcnt++;
1537 } else
1538 rebuilding_cnt++;
1539 } else
1540 sparecnt++;
1541 }
1542 }
1543 free(devmap);
1544 if (c->force) {
1545 int force_ok = force_array(content, devices, best, bestcnt,
1546 avail, most_recent, st, c);
1547 okcnt += force_ok;
1548 if (force_ok)
1549 was_forced = 1;
1550 }
1551 /* Now we want to look at the superblock which the kernel will base things on
1552 * and compare the devices that we think are working with the devices that the
1553 * superblock thinks are working.
1554 * If there are differences and --force is given, then update this chosen
1555 * superblock.
1556 */
1557 chosen_drive = -1;
1558 st->ss->free_super(st);
1559 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1560 int j = best[i];
1561 int fd;
1562
1563 if (j<0)
1564 continue;
1565 if (!devices[j].uptodate)
1566 continue;
1567 if (devices[j].i.events < devices[most_recent].i.events)
1568 continue;
1569 chosen_drive = j;
1570 if ((fd=dev_open(devices[j].devname,
1571 devices[j].included ? O_RDONLY
1572 : (O_RDONLY|O_EXCL)))< 0) {
1573 pr_err("Cannot open %s: %s\n",
1574 devices[j].devname, strerror(errno));
1575 close(mdfd);
1576 free(devices);
1577 return 1;
1578 }
1579 if (st->ss->load_super(st,fd, NULL)) {
1580 close(fd);
1581 pr_err("RAID superblock has disappeared from %s\n",
1582 devices[j].devname);
1583 close(mdfd);
1584 free(devices);
1585 return 1;
1586 }
1587 close(fd);
1588 }
1589 if (st->sb == NULL) {
1590 pr_err("No suitable drives found for %s\n", mddev);
1591 close(mdfd);
1592 free(devices);
1593 return 1;
1594 }
1595 st->ss->getinfo_super(st, content, NULL);
1596 #ifndef MDASSEMBLE
1597 sysfs_init(content, mdfd, NULL);
1598 #endif
1599 for (i=0; i<bestcnt; i++) {
1600 int j = best[i];
1601 unsigned int desired_state;
1602
1603 if (i >= content->array.raid_disks * 2)
1604 desired_state = 0;
1605 else if (i & 1)
1606 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1607 else
1608 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1609
1610 if (j<0)
1611 continue;
1612 if (!devices[j].uptodate)
1613 continue;
1614
1615 devices[j].i.disk.state = desired_state;
1616 if (!(devices[j].i.array.state & 1))
1617 clean = 0;
1618
1619 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1620 c->verbose, 0, NULL)) {
1621 if (c->force) {
1622 if (c->verbose >= 0)
1623 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1624 j, mddev, devices[j].devname);
1625 change = 1;
1626 } else {
1627 if (c->verbose >= -1)
1628 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1629 i, mddev, devices[j].devname);
1630 }
1631 }
1632 #if 0
1633 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1634 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1635 i, mddev);
1636 }
1637 #endif
1638 }
1639 if (c->force && !clean &&
1640 !enough(content->array.level, content->array.raid_disks,
1641 content->array.layout, clean,
1642 avail)) {
1643 change += st->ss->update_super(st, content, "force-array",
1644 devices[chosen_drive].devname, c->verbose,
1645 0, NULL);
1646 was_forced = 1;
1647 clean = 1;
1648 }
1649
1650 if (change) {
1651 int fd;
1652 fd = dev_open(devices[chosen_drive].devname,
1653 devices[chosen_drive].included ?
1654 O_RDWR : (O_RDWR|O_EXCL));
1655 if (fd < 0) {
1656 pr_err("Could not open %s for write - cannot Assemble array.\n",
1657 devices[chosen_drive].devname);
1658 close(mdfd);
1659 free(devices);
1660 return 1;
1661 }
1662 if (st->ss->store_super(st, fd)) {
1663 close(fd);
1664 pr_err("Could not re-write superblock on %s\n",
1665 devices[chosen_drive].devname);
1666 close(mdfd);
1667 free(devices);
1668 return 1;
1669 }
1670 if (c->verbose >= 0)
1671 pr_err("Marking array %s as 'clean'\n",
1672 mddev);
1673 close(fd);
1674 }
1675
1676 /* If we are in the middle of a reshape we may need to restore saved data
1677 * that was moved aside due to the reshape overwriting live data
1678 * The code of doing this lives in Grow.c
1679 */
1680 #ifndef MDASSEMBLE
1681 if (content->reshape_active &&
1682 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1683 int err = 0;
1684 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1685 if (c->verbose > 0)
1686 pr_err(":%s has an active reshape - checking if critical section needs to be restored\n",
1687 chosen_name);
1688 if (!c->backup_file)
1689 c->backup_file = locate_backup(content->sys_name);
1690 enable_fds(bestcnt/2);
1691 for (i = 0; i < bestcnt/2; i++) {
1692 int j = best[i*2];
1693 if (j >= 0) {
1694 fdlist[i] = dev_open(devices[j].devname,
1695 devices[j].included
1696 ? O_RDWR : (O_RDWR|O_EXCL));
1697 if (fdlist[i] < 0) {
1698 pr_err("Could not open %s for write - cannot Assemble array.\n",
1699 devices[j].devname);
1700 err = 1;
1701 break;
1702 }
1703 } else
1704 fdlist[i] = -1;
1705 }
1706 if (!err) {
1707 if (st->ss->external && st->ss->recover_backup)
1708 err = st->ss->recover_backup(st, content);
1709 else
1710 err = Grow_restart(st, content, fdlist, bestcnt/2,
1711 c->backup_file, c->verbose > 0);
1712 if (err && c->invalid_backup) {
1713 if (c->verbose > 0)
1714 pr_err("continuing without restoring backup\n");
1715 err = 0;
1716 }
1717 }
1718 while (i>0) {
1719 i--;
1720 if (fdlist[i]>=0) close(fdlist[i]);
1721 }
1722 free(fdlist);
1723 if (err) {
1724 pr_err("Failed to restore critical section for reshape, sorry.\n");
1725 if (c->backup_file == NULL)
1726 cont_err("Possibly you needed to specify the --backup-file\n");
1727 close(mdfd);
1728 free(devices);
1729 return err;
1730 }
1731 }
1732 #endif
1733
1734 /* Almost ready to actually *do* something */
1735 /* First, fill in the map, so that udev can find our name
1736 * as soon as we become active.
1737 */
1738 if (c->update && strcmp(c->update, "metadata")==0) {
1739 content->array.major_version = 1;
1740 content->array.minor_version = 0;
1741 strcpy(content->text_version, "1.0");
1742 }
1743
1744 map_update(&map, fd2devnm(mdfd), content->text_version,
1745 content->uuid, chosen_name);
1746
1747 rv = start_array(mdfd, mddev, content,
1748 st, ident, best, bestcnt,
1749 chosen_drive, devices, okcnt, sparecnt,
1750 rebuilding_cnt,
1751 c,
1752 clean, avail, start_partial_ok,
1753 pre_exist != NULL,
1754 was_forced);
1755 if (rv == 1 && !pre_exist)
1756 ioctl(mdfd, STOP_ARRAY, NULL);
1757 free(devices);
1758 map_unlock(&map);
1759 if (rv == 0) {
1760 wait_for(chosen_name, mdfd);
1761 close(mdfd);
1762 if (auto_assem) {
1763 int usecs = 1;
1764 /* There is a nasty race with 'mdadm --monitor'.
1765 * If it opens this device before we close it,
1766 * it gets an incomplete open on which IO
1767 * doesn't work and the capacity is
1768 * wrong.
1769 * If we reopen (to check for layered devices)
1770 * before --monitor closes, we loose.
1771 *
1772 * So: wait upto 1 second for there to be
1773 * a non-zero capacity.
1774 */
1775 while (usecs < 1000) {
1776 mdfd = open(mddev, O_RDONLY);
1777 if (mdfd >= 0) {
1778 unsigned long long size;
1779 if (get_dev_size(mdfd, NULL, &size) &&
1780 size > 0)
1781 break;
1782 close(mdfd);
1783 }
1784 usleep(usecs);
1785 usecs <<= 1;
1786 }
1787 }
1788 } else
1789 close(mdfd);
1790
1791 /* '2' means 'OK, but not started yet' */
1792 return rv == 2 ? 0 : rv;
1793 }
1794
1795 #ifndef MDASSEMBLE
1796 int assemble_container_content(struct supertype *st, int mdfd,
1797 struct mdinfo *content, struct context *c,
1798 char *chosen_name, int *result)
1799 {
1800 struct mdinfo *dev, *sra, *dev2;
1801 int working = 0, preexist = 0;
1802 int expansion = 0;
1803 struct map_ent *map = NULL;
1804 int old_raid_disks;
1805 int start_reshape;
1806 char *avail = NULL;
1807 int err;
1808
1809 sysfs_init(content, mdfd, NULL);
1810
1811 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1812 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1813 if (content->array.major_version == -1 &&
1814 content->array.minor_version == -2 &&
1815 c->readonly &&
1816 content->text_version[0] == '/')
1817 content->text_version[0] = '-';
1818 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1819 if (sra)
1820 sysfs_free(sra);
1821 return 1;
1822 }
1823 }
1824
1825 /* There are two types of reshape: container wide or sub-array specific
1826 * Check if metadata requests blocking container wide reshapes
1827 */
1828 start_reshape = (content->reshape_active &&
1829 !((content->reshape_active == CONTAINER_RESHAPE) &&
1830 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1831
1832 /* Block subarray here if it is under reshape now
1833 * Do not allow for any changes in this array
1834 */
1835 if (st->ss->external && content->recovery_blocked && start_reshape)
1836 block_subarray(content);
1837
1838 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1839 for (dev = content->devs; dev; dev = dev->next)
1840 if (dev2->disk.major == dev->disk.major &&
1841 dev2->disk.minor == dev->disk.minor)
1842 break;
1843 if (dev)
1844 continue;
1845 /* Don't want this one any more */
1846 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1847 errno == EBUSY) {
1848 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1849 sysfs_free(sra);
1850 return 1;
1851 }
1852 sysfs_set_str(sra, dev2, "state", "remove");
1853 }
1854 old_raid_disks = content->array.raid_disks - content->delta_disks;
1855 avail = xcalloc(content->array.raid_disks, 1);
1856 for (dev = content->devs; dev; dev = dev->next) {
1857 if (dev->disk.raid_disk >= 0)
1858 avail[dev->disk.raid_disk] = 1;
1859 if (sysfs_add_disk(content, dev, 1) == 0) {
1860 if (dev->disk.raid_disk >= old_raid_disks &&
1861 content->reshape_active)
1862 expansion++;
1863 else
1864 working++;
1865 } else if (errno == EEXIST)
1866 preexist++;
1867 }
1868 sysfs_free(sra);
1869 if (working + expansion == 0 && c->runstop <= 0) {
1870 free(avail);
1871 return 1;/* Nothing new, don't try to start */
1872 }
1873 map_update(&map, fd2devnm(mdfd),
1874 content->text_version,
1875 content->uuid, chosen_name);
1876
1877
1878 if (enough(content->array.level, content->array.raid_disks,
1879 content->array.layout, content->array.state & 1, avail) == 0) {
1880 if (c->export && result)
1881 *result |= INCR_NO;
1882 else if (c->verbose >= 0) {
1883 pr_err("%s assembled with %d device%s",
1884 chosen_name, preexist + working,
1885 preexist + working == 1 ? "":"s");
1886 if (preexist)
1887 fprintf(stderr, " (%d new)", working);
1888 fprintf(stderr, " but not started\n");
1889 }
1890 free(avail);
1891 return 1;
1892 }
1893 free(avail);
1894
1895 if (c->runstop <= 0 &&
1896 (working + preexist + expansion) <
1897 content->array.working_disks) {
1898 if (c->export && result)
1899 *result |= INCR_UNSAFE;
1900 else if (c->verbose >= 0) {
1901 pr_err("%s assembled with %d device%s",
1902 chosen_name, preexist + working,
1903 preexist + working == 1 ? "":"s");
1904 if (preexist)
1905 fprintf(stderr, " (%d new)", working);
1906 fprintf(stderr, " but not safe to start\n");
1907 }
1908 return 1;
1909 }
1910
1911
1912 if (start_reshape) {
1913 int spare = content->array.raid_disks + expansion;
1914 if (restore_backup(st, content,
1915 working,
1916 spare, &c->backup_file, c->verbose) == 1)
1917 return 1;
1918
1919 err = sysfs_set_str(content, NULL,
1920 "array_state", "readonly");
1921 if (err)
1922 return 1;
1923
1924 if (st->ss->external) {
1925 if (!mdmon_running(st->container_devnm))
1926 start_mdmon(st->container_devnm);
1927 ping_monitor(st->container_devnm);
1928 if (mdmon_running(st->container_devnm) &&
1929 st->update_tail == NULL)
1930 st->update_tail = &st->updates;
1931 }
1932
1933 err = Grow_continue(mdfd, st, content, c->backup_file,
1934 0, c->freeze_reshape);
1935 } else switch(content->array.level) {
1936 case LEVEL_LINEAR:
1937 case LEVEL_MULTIPATH:
1938 case 0:
1939 err = sysfs_set_str(content, NULL, "array_state",
1940 c->readonly ? "readonly" : "active");
1941 break;
1942 default:
1943 err = sysfs_set_str(content, NULL, "array_state",
1944 "readonly");
1945 /* start mdmon if needed. */
1946 if (!err) {
1947 if (!mdmon_running(st->container_devnm))
1948 start_mdmon(st->container_devnm);
1949 ping_monitor(st->container_devnm);
1950 }
1951 break;
1952 }
1953 if (!err)
1954 sysfs_set_safemode(content, content->safe_mode_delay);
1955
1956 /* Block subarray here if it is not reshaped now
1957 * It has be blocked a little later to allow mdmon to switch in
1958 * in to R/W state
1959 */
1960 if (st->ss->external && content->recovery_blocked &&
1961 !start_reshape)
1962 block_subarray(content);
1963
1964 if (c->export && result) {
1965 if (err)
1966 *result |= INCR_NO;
1967 else
1968 *result |= INCR_YES;
1969 } else if (c->verbose >= 0) {
1970 if (err)
1971 pr_err("array %s now has %d device%s",
1972 chosen_name, working + preexist,
1973 working + preexist == 1 ? "":"s");
1974 else
1975 pr_err("Started %s with %d device%s",
1976 chosen_name, working + preexist,
1977 working + preexist == 1 ? "":"s");
1978 if (preexist)
1979 fprintf(stderr, " (%d new)", working);
1980 if (expansion)
1981 fprintf(stderr, " ( + %d for expansion)",
1982 expansion);
1983 fprintf(stderr, "\n");
1984 }
1985 if (!err)
1986 wait_for(chosen_name, mdfd);
1987 return err;
1988 /* FIXME should have an O_EXCL and wait for read-auto */
1989 }
1990 #endif
Unnamed repository; edit this file 'description' to name the repository.