]> 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
Incremental: add --export handling.
[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 *st,
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
571 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
572 char *devname = tmpdev->devname;
573 struct stat stb;
574 int i;
575
576 if (tmpdev->used != 1)
577 continue;
578 /* looks like a good enough match to update the super block if needed */
579 #ifndef MDASSEMBLE
580 if (c->update) {
581 int dfd;
582 /* prepare useful information in info structures */
583 struct stat stb2;
584 struct supertype *tst;
585 int err;
586 fstat(mdfd, &stb2);
587
588 if (strcmp(c->update, "uuid")==0 &&
589 !ident->uuid_set) {
590 int rfd;
591 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
592 read(rfd, ident->uuid, 16) != 16) {
593 *(__u32*)(ident->uuid) = random();
594 *(__u32*)(ident->uuid+1) = random();
595 *(__u32*)(ident->uuid+2) = random();
596 *(__u32*)(ident->uuid+3) = random();
597 }
598 if (rfd >= 0) close(rfd);
599 }
600 dfd = dev_open(devname,
601 tmpdev->disposition == 'I'
602 ? O_RDWR : (O_RDWR|O_EXCL));
603
604 tst = dup_super(st);
605 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
606 pr_err("cannot re-read metadata from %s - aborting\n",
607 devname);
608 if (dfd >= 0)
609 close(dfd);
610 close(mdfd);
611 free(devices);
612 free(devmap);
613 return -1;
614 }
615 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
616
617 memcpy(content->uuid, ident->uuid, 16);
618 strcpy(content->name, ident->name);
619 content->array.md_minor = minor(stb2.st_rdev);
620
621 if (strcmp(c->update, "byteorder") == 0)
622 err = 0;
623 else
624 err = tst->ss->update_super(tst, content, c->update,
625 devname, c->verbose,
626 ident->uuid_set,
627 c->homehost);
628 if (err < 0) {
629 if (err == -1)
630 pr_err("--update=%s not understood"
631 " for %s metadata\n",
632 c->update, tst->ss->name);
633 tst->ss->free_super(tst);
634 free(tst);
635 close(mdfd);
636 close(dfd);
637 free(devices);
638 free(devmap);
639 return -1;
640 }
641 if (strcmp(c->update, "uuid")==0 &&
642 !ident->uuid_set) {
643 ident->uuid_set = 1;
644 memcpy(ident->uuid, content->uuid, 16);
645 }
646 if (tst->ss->store_super(tst, dfd))
647 pr_err("Could not re-write superblock on %s.\n",
648 devname);
649 close(dfd);
650
651 if (strcmp(c->update, "uuid")==0 &&
652 ident->bitmap_fd >= 0 && !bitmap_done) {
653 if (bitmap_update_uuid(ident->bitmap_fd,
654 content->uuid,
655 tst->ss->swapuuid) != 0)
656 pr_err("Could not update uuid on external bitmap.\n");
657 else
658 bitmap_done = 1;
659 }
660 tst->ss->free_super(tst);
661 } else
662 #endif
663 {
664 struct supertype *tst = dup_super(st);
665 int dfd;
666 dfd = dev_open(devname,
667 tmpdev->disposition == 'I'
668 ? O_RDWR : (O_RDWR|O_EXCL));
669
670 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
671 pr_err("cannot re-read metadata from %s - aborting\n",
672 devname);
673 if (dfd >= 0)
674 close(dfd);
675 close(mdfd);
676 free(devices);
677 free(devmap);
678 return -1;
679 }
680 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
681 tst->ss->free_super(tst);
682 close(dfd);
683 }
684
685 stat(devname, &stb);
686
687 if (c->verbose > 0)
688 pr_err("%s is identified as a member of %s, slot %d%s.\n",
689 devname, mddev, content->disk.raid_disk,
690 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
691 devices[devcnt].devname = devname;
692 devices[devcnt].uptodate = 0;
693 devices[devcnt].included = (tmpdev->disposition == 'I');
694 devices[devcnt].i = *content;
695 devices[devcnt].i.disk.major = major(stb.st_rdev);
696 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
697
698 if (devices[devcnt].i.disk.state == 6) {
699 if (most_recent < 0 ||
700 devices[devcnt].i.events
701 > devices[most_recent].i.events)
702 most_recent = devcnt;
703 }
704
705 if (content->array.level == LEVEL_MULTIPATH)
706 /* with multipath, the raid_disk from the superblock is meaningless */
707 i = devcnt;
708 else
709 i = devices[devcnt].i.disk.raid_disk;
710 if (i+1 == 0) {
711 if (nextspare < content->array.raid_disks*2)
712 nextspare = content->array.raid_disks*2;
713 i = nextspare++;
714 } else {
715 /* i is raid_disk - double it so there is room for
716 * replacements */
717 i *= 2;
718 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
719 i++;
720 if (i >= content->array.raid_disks*2 &&
721 i >= nextspare)
722 nextspare = i+1;
723 }
724 if (i < 10000) {
725 if (i >= bestcnt) {
726 int newbestcnt = i+10;
727 int *newbest = xmalloc(sizeof(int)*newbestcnt);
728 int c;
729 for (c=0; c < newbestcnt; c++)
730 if (c < bestcnt)
731 newbest[c] = best[c];
732 else
733 newbest[c] = -1;
734 if (best)free(best);
735 best = newbest;
736 bestcnt = newbestcnt;
737 }
738 if (best[i] >=0 &&
739 devices[best[i]].i.events
740 == devices[devcnt].i.events
741 && (devices[best[i]].i.disk.minor
742 != devices[devcnt].i.disk.minor)
743 && st->ss == &super0
744 && content->array.level != LEVEL_MULTIPATH) {
745 /* two different devices with identical superblock.
746 * Could be a mis-detection caused by overlapping
747 * partitions. fail-safe.
748 */
749 pr_err("WARNING %s and %s appear"
750 " to have very similar superblocks.\n"
751 " If they are really different, "
752 "please --zero the superblock on one\n"
753 " If they are the same or overlap,"
754 " please remove one from %s.\n",
755 devices[best[i]].devname, devname,
756 inargv ? "the list" :
757 "the\n DEVICE list in mdadm.conf"
758 );
759 close(mdfd);
760 free(devices);
761 free(devmap);
762 return -1;
763 }
764 if (best[i] == -1
765 || (devices[best[i]].i.events
766 < devices[devcnt].i.events))
767 best[i] = devcnt;
768 }
769 devcnt++;
770 }
771 if (most_recent >= 0)
772 *most_recentp = most_recent;
773 *bestcntp = bestcnt;
774 *bestp = best;
775 return devcnt;
776 }
777
778 static int force_array(struct mdinfo *content,
779 struct devs *devices,
780 int *best, int bestcnt, char *avail,
781 int most_recent,
782 struct supertype *st,
783 struct context *c)
784 {
785 int okcnt = 0;
786 while (!enough(content->array.level, content->array.raid_disks,
787 content->array.layout, 1,
788 avail)
789 ||
790 (content->reshape_active && content->delta_disks > 0 &&
791 !enough(content->array.level, (content->array.raid_disks
792 - content->delta_disks),
793 content->new_layout, 1,
794 avail)
795 )) {
796 /* Choose the newest best drive which is
797 * not up-to-date, update the superblock
798 * and add it.
799 */
800 int fd;
801 struct supertype *tst;
802 unsigned long long current_events;
803 int chosen_drive = -1;
804 int i;
805
806 for (i = 0;
807 i < content->array.raid_disks * 2 && i < bestcnt;
808 i += 2) {
809 int j = best[i];
810 if (j>=0 &&
811 !devices[j].uptodate &&
812 devices[j].i.recovery_start == MaxSector &&
813 (chosen_drive < 0 ||
814 devices[j].i.events
815 > devices[chosen_drive].i.events))
816 chosen_drive = j;
817 }
818 if (chosen_drive < 0)
819 break;
820 current_events = devices[chosen_drive].i.events;
821 add_another:
822 if (c->verbose >= 0)
823 pr_err("forcing event count in %s(%d) from %d upto %d\n",
824 devices[chosen_drive].devname,
825 devices[chosen_drive].i.disk.raid_disk,
826 (int)(devices[chosen_drive].i.events),
827 (int)(devices[most_recent].i.events));
828 fd = dev_open(devices[chosen_drive].devname,
829 devices[chosen_drive].included ? O_RDWR
830 : (O_RDWR|O_EXCL));
831 if (fd < 0) {
832 pr_err("Couldn't open %s for write - not updating\n",
833 devices[chosen_drive].devname);
834 devices[chosen_drive].i.events = 0;
835 continue;
836 }
837 tst = dup_super(st);
838 if (tst->ss->load_super(tst,fd, NULL)) {
839 close(fd);
840 pr_err("RAID superblock disappeared from %s - not updating.\n",
841 devices[chosen_drive].devname);
842 devices[chosen_drive].i.events = 0;
843 continue;
844 }
845 content->events = devices[most_recent].i.events;
846 tst->ss->update_super(tst, content, "force-one",
847 devices[chosen_drive].devname, c->verbose,
848 0, NULL);
849
850 if (tst->ss->store_super(tst, fd)) {
851 close(fd);
852 pr_err("Could not re-write superblock on %s\n",
853 devices[chosen_drive].devname);
854 devices[chosen_drive].i.events = 0;
855 tst->ss->free_super(tst);
856 continue;
857 }
858 close(fd);
859 devices[chosen_drive].i.events = devices[most_recent].i.events;
860 devices[chosen_drive].uptodate = 1;
861 avail[chosen_drive] = 1;
862 okcnt++;
863 tst->ss->free_super(tst);
864
865 /* If there are any other drives of the same vintage,
866 * add them in as well. We can't lose and we might gain
867 */
868 for (i = 0;
869 i < content->array.raid_disks * 2 && i < bestcnt ;
870 i += 2) {
871 int j = best[i];
872 if (j >= 0 &&
873 !devices[j].uptodate &&
874 devices[j].i.recovery_start == MaxSector &&
875 devices[j].i.events == current_events) {
876 chosen_drive = j;
877 goto add_another;
878 }
879 }
880 }
881 return okcnt;
882 }
883
884 static int start_array(int mdfd,
885 char *mddev,
886 struct mdinfo *content,
887 struct supertype *st,
888 struct mddev_ident *ident,
889 int *best, int bestcnt,
890 int chosen_drive,
891 struct devs *devices,
892 unsigned int okcnt,
893 unsigned int sparecnt,
894 unsigned int rebuilding_cnt,
895 struct context *c,
896 int clean, char *avail,
897 int start_partial_ok,
898 int err_ok,
899 int was_forced
900 )
901 {
902 int rv;
903 int i;
904 unsigned int req_cnt;
905
906 rv = set_array_info(mdfd, st, content);
907 if (rv && !err_ok) {
908 pr_err("failed to set array info for %s: %s\n",
909 mddev, strerror(errno));
910 return 1;
911 }
912 if (ident->bitmap_fd >= 0) {
913 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
914 pr_err("SET_BITMAP_FILE failed.\n");
915 return 1;
916 }
917 } else if (ident->bitmap_file) {
918 /* From config file */
919 int bmfd = open(ident->bitmap_file, O_RDWR);
920 if (bmfd < 0) {
921 pr_err("Could not open bitmap file %s\n",
922 ident->bitmap_file);
923 return 1;
924 }
925 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
926 pr_err("Failed to set bitmapfile for %s\n", mddev);
927 close(bmfd);
928 return 1;
929 }
930 close(bmfd);
931 }
932
933 /* First, add the raid disks, but add the chosen one last */
934 for (i=0; i<= bestcnt; i++) {
935 int j;
936 if (i < bestcnt) {
937 j = best[i];
938 if (j == chosen_drive)
939 continue;
940 } else
941 j = chosen_drive;
942
943 if (j >= 0 && !devices[j].included) {
944 int dfd = dev_open(devices[j].devname,
945 O_RDWR|O_EXCL);
946 if (dfd >= 0) {
947 remove_partitions(dfd);
948 close(dfd);
949 }
950 rv = add_disk(mdfd, st, content, &devices[j].i);
951
952 if (rv) {
953 pr_err("failed to add "
954 "%s to %s: %s\n",
955 devices[j].devname,
956 mddev,
957 strerror(errno));
958 if (i < content->array.raid_disks * 2
959 || i == bestcnt)
960 okcnt--;
961 else
962 sparecnt--;
963 } else if (c->verbose > 0)
964 pr_err("added %s to %s as %d%s%s\n",
965 devices[j].devname, mddev,
966 devices[j].i.disk.raid_disk,
967 devices[j].uptodate?"":
968 " (possibly out of date)",
969 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
970 } else if (j >= 0) {
971 if (c->verbose > 0)
972 pr_err("%s is already in %s as %d\n",
973 devices[j].devname, mddev,
974 devices[j].i.disk.raid_disk);
975 } else if (c->verbose > 0 && i < content->array.raid_disks*2
976 && (i&1) == 0)
977 pr_err("no uptodate device for slot %d of %s\n",
978 i, mddev);
979 }
980
981 if (content->array.level == LEVEL_CONTAINER) {
982 if (c->verbose >= 0) {
983 pr_err("Container %s has been "
984 "assembled with %d drive%s",
985 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
986 if (okcnt < (unsigned)content->array.raid_disks)
987 fprintf(stderr, " (out of %d)",
988 content->array.raid_disks);
989 fprintf(stderr, "\n");
990 }
991 st->ss->free_super(st);
992 sysfs_uevent(content, "change");
993 return 0;
994 }
995
996 /* Get number of in-sync devices according to the superblock.
997 * We must have this number to start the array without -s or -R
998 */
999 req_cnt = content->array.working_disks;
1000
1001 if (c->runstop == 1 ||
1002 (c->runstop <= 0 &&
1003 ( enough(content->array.level, content->array.raid_disks,
1004 content->array.layout, clean, avail) &&
1005 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1006 ))) {
1007 /* This array is good-to-go.
1008 * If a reshape is in progress then we might need to
1009 * continue monitoring it. In that case we start
1010 * it read-only and let the grow code make it writable.
1011 */
1012 int rv;
1013 #ifndef MDASSEMBLE
1014 if (content->reshape_active &&
1015 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1016 content->delta_disks <= 0) {
1017 if (!c->backup_file) {
1018 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1019 mddev);
1020 pr_err("Please provided one with \"--backup-file=...\"\n");
1021 if (c->update &&
1022 strcmp(c->update, "revert-reshape") == 0)
1023 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1024 return 1;
1025 }
1026 rv = sysfs_set_str(content, NULL,
1027 "array_state", "readonly");
1028 if (rv == 0)
1029 rv = Grow_continue(mdfd, st, content,
1030 c->backup_file,
1031 c->freeze_reshape);
1032 } else if (c->readonly &&
1033 sysfs_attribute_available(
1034 content, NULL, "array_state")) {
1035 rv = sysfs_set_str(content, NULL,
1036 "array_state", "readonly");
1037 } else
1038 #endif
1039 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1040 if (rv == 0) {
1041 if (c->verbose >= 0) {
1042 pr_err("%s has been started with %d drive%s",
1043 mddev, okcnt, okcnt==1?"":"s");
1044 if (okcnt < (unsigned)content->array.raid_disks)
1045 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1046 if (rebuilding_cnt)
1047 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1048 if (sparecnt)
1049 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1050 fprintf(stderr, ".\n");
1051 }
1052 if (content->reshape_active &&
1053 content->array.level >= 4 &&
1054 content->array.level <= 6) {
1055 /* might need to increase the size
1056 * of the stripe cache - default is 256
1057 */
1058 if (256 < 4 * (content->array.chunk_size/4096)) {
1059 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1060 if (sra)
1061 sysfs_set_num(sra, NULL,
1062 "stripe_cache_size",
1063 (4 * content->array.chunk_size / 4096) + 1);
1064 sysfs_free(sra);
1065 }
1066 }
1067 if (okcnt < (unsigned)content->array.raid_disks) {
1068 /* If any devices did not get added
1069 * because the kernel rejected them based
1070 * on event count, try adding them
1071 * again providing the action policy is
1072 * 're-add' or greater. The bitmap
1073 * might allow them to be included, or
1074 * they will become spares.
1075 */
1076 for (i = 0; i < bestcnt; i++) {
1077 int j = best[i];
1078 if (j >= 0 && !devices[j].uptodate) {
1079 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1080 continue;
1081 rv = add_disk(mdfd, st, content,
1082 &devices[j].i);
1083 if (rv == 0 && c->verbose >= 0)
1084 pr_err("%s has been re-added.\n",
1085 devices[j].devname);
1086 }
1087 }
1088 }
1089 if (content->array.level == 6 &&
1090 okcnt + 1 == (unsigned)content->array.raid_disks &&
1091 was_forced) {
1092 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1093 if (sra)
1094 sysfs_set_str(sra, NULL,
1095 "sync_action", "repair");
1096 sysfs_free(sra);
1097 }
1098 return 0;
1099 }
1100 pr_err("failed to RUN_ARRAY %s: %s\n",
1101 mddev, strerror(errno));
1102
1103 if (!enough(content->array.level, content->array.raid_disks,
1104 content->array.layout, 1, avail))
1105 pr_err("Not enough devices to "
1106 "start the array.\n");
1107 else if (!enough(content->array.level,
1108 content->array.raid_disks,
1109 content->array.layout, clean,
1110 avail))
1111 pr_err("Not enough devices to "
1112 "start the array while not clean "
1113 "- consider --force.\n");
1114
1115 return 1;
1116 }
1117 if (c->runstop == -1) {
1118 pr_err("%s assembled from %d drive%s",
1119 mddev, okcnt, okcnt==1?"":"s");
1120 if (okcnt != (unsigned)content->array.raid_disks)
1121 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1122 fprintf(stderr, ", but not started.\n");
1123 return 2;
1124 }
1125 if (c->verbose >= -1) {
1126 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1127 if (rebuilding_cnt)
1128 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1129 if (sparecnt)
1130 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1131 if (!enough(content->array.level, content->array.raid_disks,
1132 content->array.layout, 1, avail))
1133 fprintf(stderr, " - not enough to start the array.\n");
1134 else if (!enough(content->array.level,
1135 content->array.raid_disks,
1136 content->array.layout, clean,
1137 avail))
1138 fprintf(stderr, " - not enough to start the "
1139 "array while not clean - consider "
1140 "--force.\n");
1141 else {
1142 if (req_cnt == (unsigned)content->array.raid_disks)
1143 fprintf(stderr, " - need all %d to start it", req_cnt);
1144 else
1145 fprintf(stderr, " - need %d to start", req_cnt);
1146 fprintf(stderr, " (use --run to insist).\n");
1147 }
1148 }
1149 return 1;
1150 }
1151
1152 int Assemble(struct supertype *st, char *mddev,
1153 struct mddev_ident *ident,
1154 struct mddev_dev *devlist,
1155 struct context *c)
1156 {
1157 /*
1158 * The task of Assemble is to find a collection of
1159 * devices that should (according to their superblocks)
1160 * form an array, and to give this collection to the MD driver.
1161 * In Linux-2.4 and later, this involves submitting a
1162 * SET_ARRAY_INFO ioctl with no arg - to prepare
1163 * the array - and then submit a number of
1164 * ADD_NEW_DISK ioctls to add disks into
1165 * the array. Finally RUN_ARRAY might
1166 * be submitted to start the array.
1167 *
1168 * Much of the work of Assemble is in finding and/or
1169 * checking the disks to make sure they look right.
1170 *
1171 * If mddev is not set, then scan must be set and we
1172 * read through the config file for dev+uuid mapping
1173 * We recurse, setting mddev, for each device that
1174 * - isn't running
1175 * - has a valid uuid (or any uuid if !uuidset)
1176 *
1177 * If mddev is set, we try to determine state of md.
1178 * check version - must be at least 0.90.0
1179 * check kernel version. must be at least 2.4.
1180 * If not, we can possibly fall back on START_ARRAY
1181 * Try to GET_ARRAY_INFO.
1182 * If possible, give up
1183 * If not, try to STOP_ARRAY just to make sure
1184 *
1185 * If !uuidset and scan, look in conf-file for uuid
1186 * If not found, give up
1187 * If !devlist and scan and uuidset, get list of devs from conf-file
1188 *
1189 * For each device:
1190 * Check superblock - discard if bad
1191 * Check uuid (set if we don't have one) - discard if no match
1192 * Check superblock similarity if we have a superblock - discard if different
1193 * Record events, devicenum
1194 * This should give us a list of devices for the array
1195 * We should collect the most recent event number
1196 *
1197 * Count disks with recent enough event count
1198 * While force && !enough disks
1199 * Choose newest rejected disks, update event count
1200 * mark clean and rewrite superblock
1201 * If recent kernel:
1202 * SET_ARRAY_INFO
1203 * foreach device with recent events : ADD_NEW_DISK
1204 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1205 * If old kernel:
1206 * Check the device numbers in superblock are right
1207 * update superblock if any changes
1208 * START_ARRAY
1209 *
1210 */
1211 int rv;
1212 int mdfd;
1213 int clean;
1214 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1215 ident->super_minor == UnSet && ident->name[0] == 0
1216 && (ident->container == NULL || ident->member == NULL));
1217 struct devs *devices;
1218 char *devmap;
1219 int *best = NULL; /* indexed by raid_disk */
1220 int bestcnt = 0;
1221 int devcnt;
1222 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1223 int i;
1224 int was_forced = 0;
1225 int most_recent = 0;
1226 int chosen_drive;
1227 int change = 0;
1228 int inargv = 0;
1229 int start_partial_ok = (c->runstop >= 0) &&
1230 (c->force || devlist==NULL || auto_assem);
1231 int num_devs;
1232 struct mddev_dev *tmpdev;
1233 struct mdinfo info;
1234 struct mdinfo *content = NULL;
1235 struct mdinfo *pre_exist = NULL;
1236 char *avail;
1237 char *name = NULL;
1238 char chosen_name[1024];
1239 struct map_ent *map = NULL;
1240 struct map_ent *mp;
1241
1242 /*
1243 * If any subdevs are listed, then any that don't
1244 * match ident are discarded. Remainder must all match and
1245 * become the array.
1246 * If no subdevs, then we scan all devices in the config file, but
1247 * there must be something in the identity
1248 */
1249
1250 if (!devlist &&
1251 ident->uuid_set == 0 &&
1252 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1253 ident->name[0] == 0 &&
1254 (ident->container == NULL || ident->member == NULL) &&
1255 ident->devices == NULL) {
1256 pr_err("No identity information available for %s - cannot assemble.\n",
1257 mddev ? mddev : "further assembly");
1258 return 1;
1259 }
1260
1261 if (devlist == NULL)
1262 devlist = conf_get_devs();
1263 else if (mddev)
1264 inargv = 1;
1265
1266 try_again:
1267 /* We come back here when doing auto-assembly and attempting some
1268 * set of devices failed. Those are now marked as ->used==2 and
1269 * we ignore them and try again
1270 */
1271 if (!st && ident->st)
1272 st = ident->st;
1273 if (c->verbose>0)
1274 pr_err("looking for devices for %s\n",
1275 mddev ? mddev : "further assembly");
1276
1277 content = &info;
1278 if (st)
1279 st->ignore_hw_compat = 1;
1280 num_devs = select_devices(devlist, ident, &st, &content, c,
1281 inargv, auto_assem);
1282 if (num_devs < 0)
1283 return 1;
1284
1285 if (!st || !st->sb || !content)
1286 return 2;
1287
1288 /* We have a full set of devices - we now need to find the
1289 * array device.
1290 * However there is a risk that we are racing with "mdadm -I"
1291 * and the array is already partially assembled - we will have
1292 * rejected any devices already in this address.
1293 * So we take a lock on the map file - to prevent further races -
1294 * and look for the uuid in there. If found and the array is
1295 * active, we abort. If found and the array is not active
1296 * we commit to that md device and add all the contained devices
1297 * to our list. We flag them so that we don't try to re-add,
1298 * but can remove if they turn out to not be wanted.
1299 */
1300 if (map_lock(&map))
1301 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1302 mp = map_by_uuid(&map, content->uuid);
1303 if (mp) {
1304 struct mdinfo *dv;
1305 /* array already exists. */
1306 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1307 if (pre_exist->array.level != UnSet) {
1308 pr_err("Found some drive for an array that is already active: %s\n",
1309 mp->path);
1310 pr_err("giving up.\n");
1311 return 1;
1312 }
1313 for (dv = pre_exist->devs; dv; dv = dv->next) {
1314 /* We want to add this device to our list,
1315 * but it could already be there if "mdadm -I"
1316 * started *after* we checked for O_EXCL.
1317 * If we add it to the top of the list
1318 * it will be preferred over later copies.
1319 */
1320 struct mddev_dev *newdev;
1321 char *devname = map_dev(dv->disk.major,
1322 dv->disk.minor,
1323 0);
1324 if (!devname)
1325 continue;
1326 newdev = xmalloc(sizeof(*newdev));
1327 newdev->devname = devname;
1328 newdev->disposition = 'I';
1329 newdev->used = 1;
1330 newdev->next = devlist;
1331 devlist = newdev;
1332 num_devs++;
1333 }
1334 strcpy(chosen_name, mp->path);
1335 if (c->verbose > 0 || mddev == NULL ||
1336 strcmp(mddev, chosen_name) != 0)
1337 pr_err("Merging with already-assembled %s\n",
1338 chosen_name);
1339 mdfd = open_dev_excl(mp->devnm);
1340 } else {
1341 int trustworthy = FOREIGN;
1342 name = content->name;
1343 switch (st->ss->match_home(st, c->homehost)
1344 ?: st->ss->match_home(st, "any")) {
1345 case 1:
1346 trustworthy = LOCAL;
1347 name = strchr(content->name, ':');
1348 if (name)
1349 name++;
1350 else
1351 name = content->name;
1352 break;
1353 }
1354 if (!auto_assem)
1355 /* If the array is listed in mdadm.conf or on
1356 * command line, then we trust the name
1357 * even if the array doesn't look local
1358 */
1359 trustworthy = LOCAL;
1360
1361 if (name[0] == 0 &&
1362 content->array.level == LEVEL_CONTAINER) {
1363 name = content->text_version;
1364 trustworthy = METADATA;
1365 }
1366
1367 if (name[0] && trustworthy != LOCAL &&
1368 ! c->require_homehost &&
1369 conf_name_is_free(name))
1370 trustworthy = LOCAL;
1371
1372 if (trustworthy == LOCAL &&
1373 strchr(name, ':'))
1374 /* Ignore 'host:' prefix of name */
1375 name = strchr(name, ':')+1;
1376
1377 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1378 chosen_name);
1379 }
1380 if (mdfd < 0) {
1381 st->ss->free_super(st);
1382 if (auto_assem)
1383 goto try_again;
1384 return 1;
1385 }
1386 mddev = chosen_name;
1387 if (get_linux_version() < 2004000 ||
1388 md_get_version(mdfd) < 9000) {
1389 pr_err("Assemble requires Linux 2.4 or later, and\n"
1390 " md driver version 0.90.0 or later.\n"
1391 " Upgrade your kernel or try --build\n");
1392 close(mdfd);
1393 return 1;
1394 }
1395 if (pre_exist == NULL) {
1396 if (mddev_busy(fd2devnm(mdfd))) {
1397 pr_err("%s already active, cannot restart it!\n",
1398 mddev);
1399 for (tmpdev = devlist ;
1400 tmpdev && tmpdev->used != 1;
1401 tmpdev = tmpdev->next)
1402 ;
1403 if (tmpdev && auto_assem)
1404 pr_err("%s needed for %s...\n",
1405 mddev, tmpdev->devname);
1406 close(mdfd);
1407 mdfd = -3;
1408 st->ss->free_super(st);
1409 if (auto_assem)
1410 goto try_again;
1411 return 1;
1412 }
1413 /* just incase it was started but has no content */
1414 ioctl(mdfd, STOP_ARRAY, NULL);
1415 }
1416
1417 #ifndef MDASSEMBLE
1418 if (content != &info) {
1419 /* This is a member of a container. Try starting the array. */
1420 int err;
1421 err = assemble_container_content(st, mdfd, content, c,
1422 chosen_name, NULL);
1423 close(mdfd);
1424 return err;
1425 }
1426 #endif
1427 /* Ok, no bad inconsistancy, we can try updating etc */
1428 devices = xcalloc(num_devs, sizeof(*devices));
1429 devmap = xcalloc(num_devs, content->array.raid_disks);
1430 devcnt = load_devices(devices, devmap, ident, st, devlist,
1431 c, content, mdfd, mddev,
1432 &most_recent, &bestcnt, &best, inargv);
1433 if (devcnt < 0)
1434 return 1;
1435
1436 if (devcnt == 0) {
1437 pr_err("no devices found for %s\n",
1438 mddev);
1439 if (st)
1440 st->ss->free_super(st);
1441 close(mdfd);
1442 free(devices);
1443 free(devmap);
1444 return 1;
1445 }
1446
1447 if (c->update && strcmp(c->update, "byteorder")==0)
1448 st->minor_version = 90;
1449
1450 st->ss->getinfo_super(st, content, NULL);
1451 clean = content->array.state & 1;
1452
1453 /* now we have some devices that might be suitable.
1454 * I wonder how many
1455 */
1456 avail = xcalloc(content->array.raid_disks, 1);
1457 okcnt = 0;
1458 replcnt = 0;
1459 sparecnt=0;
1460 rebuilding_cnt=0;
1461 for (i=0; i< bestcnt; i++) {
1462 int j = best[i];
1463 int event_margin = 1; /* always allow a difference of '1'
1464 * like the kernel does
1465 */
1466 if (j < 0) continue;
1467 /* note: we ignore error flags in multipath arrays
1468 * as they don't make sense
1469 */
1470 if (content->array.level != LEVEL_MULTIPATH)
1471 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1472 if (!(devices[j].i.disk.state
1473 & (1<<MD_DISK_FAULTY))) {
1474 devices[j].uptodate = 1;
1475 sparecnt++;
1476 }
1477 continue;
1478 }
1479 /* If this device thinks that 'most_recent' has failed, then
1480 * we must reject this device.
1481 */
1482 if (j != most_recent && !c->force &&
1483 content->array.raid_disks > 0 &&
1484 devices[most_recent].i.disk.raid_disk >= 0 &&
1485 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1486 if (c->verbose > -1)
1487 pr_err("ignoring %s as it reports %s as failed\n",
1488 devices[j].devname, devices[most_recent].devname);
1489 best[i] = -1;
1490 continue;
1491 }
1492 /* Require event counter to be same as, or just less than,
1493 * most recent. If it is bigger, it must be a stray spare and
1494 * should be ignored.
1495 */
1496 if (devices[j].i.events+event_margin >=
1497 devices[most_recent].i.events &&
1498 devices[j].i.events <=
1499 devices[most_recent].i.events
1500 ) {
1501 devices[j].uptodate = 1;
1502 if (i < content->array.raid_disks * 2) {
1503 if (devices[j].i.recovery_start == MaxSector ||
1504 (content->reshape_active &&
1505 ((i >= content->array.raid_disks - content->delta_disks) ||
1506 (i >= content->array.raid_disks - content->delta_disks - 1
1507 && content->array.level == 4)))) {
1508 if (!avail[i/2]) {
1509 okcnt++;
1510 avail[i/2]=1;
1511 } else
1512 replcnt++;
1513 } else
1514 rebuilding_cnt++;
1515 } else
1516 sparecnt++;
1517 }
1518 }
1519 free(devmap);
1520 if (c->force) {
1521 int force_ok = force_array(content, devices, best, bestcnt,
1522 avail, most_recent, st, c);
1523 okcnt += force_ok;
1524 if (force_ok)
1525 was_forced = 1;
1526 }
1527 /* Now we want to look at the superblock which the kernel will base things on
1528 * and compare the devices that we think are working with the devices that the
1529 * superblock thinks are working.
1530 * If there are differences and --force is given, then update this chosen
1531 * superblock.
1532 */
1533 chosen_drive = -1;
1534 st->ss->free_super(st);
1535 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1536 int j = best[i];
1537 int fd;
1538
1539 if (j<0)
1540 continue;
1541 if (!devices[j].uptodate)
1542 continue;
1543 if (devices[j].i.events < devices[most_recent].i.events)
1544 continue;
1545 chosen_drive = j;
1546 if ((fd=dev_open(devices[j].devname,
1547 devices[j].included ? O_RDONLY
1548 : (O_RDONLY|O_EXCL)))< 0) {
1549 pr_err("Cannot open %s: %s\n",
1550 devices[j].devname, strerror(errno));
1551 close(mdfd);
1552 free(devices);
1553 return 1;
1554 }
1555 if (st->ss->load_super(st,fd, NULL)) {
1556 close(fd);
1557 pr_err("RAID superblock has disappeared from %s\n",
1558 devices[j].devname);
1559 close(mdfd);
1560 free(devices);
1561 return 1;
1562 }
1563 close(fd);
1564 }
1565 if (st->sb == NULL) {
1566 pr_err("No suitable drives found for %s\n", mddev);
1567 close(mdfd);
1568 free(devices);
1569 return 1;
1570 }
1571 st->ss->getinfo_super(st, content, NULL);
1572 #ifndef MDASSEMBLE
1573 sysfs_init(content, mdfd, NULL);
1574 #endif
1575 for (i=0; i<bestcnt; i++) {
1576 int j = best[i];
1577 unsigned int desired_state;
1578
1579 if (i >= content->array.raid_disks * 2)
1580 desired_state = 0;
1581 else if (i & 1)
1582 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1583 else
1584 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1585
1586 if (j<0)
1587 continue;
1588 if (!devices[j].uptodate)
1589 continue;
1590
1591 devices[j].i.disk.state = desired_state;
1592 if (!(devices[j].i.array.state & 1))
1593 clean = 0;
1594
1595 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1596 c->verbose, 0, NULL)) {
1597 if (c->force) {
1598 if (c->verbose >= 0)
1599 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1600 j, mddev, devices[j].devname);
1601 change = 1;
1602 } else {
1603 if (c->verbose >= -1)
1604 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1605 i, mddev, devices[j].devname);
1606 }
1607 }
1608 #if 0
1609 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1610 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1611 i, mddev);
1612 }
1613 #endif
1614 }
1615 if (c->force && !clean &&
1616 !enough(content->array.level, content->array.raid_disks,
1617 content->array.layout, clean,
1618 avail)) {
1619 change += st->ss->update_super(st, content, "force-array",
1620 devices[chosen_drive].devname, c->verbose,
1621 0, NULL);
1622 was_forced = 1;
1623 clean = 1;
1624 }
1625
1626 if (change) {
1627 int fd;
1628 fd = dev_open(devices[chosen_drive].devname,
1629 devices[chosen_drive].included ?
1630 O_RDWR : (O_RDWR|O_EXCL));
1631 if (fd < 0) {
1632 pr_err("Could not open %s for write - cannot Assemble array.\n",
1633 devices[chosen_drive].devname);
1634 close(mdfd);
1635 free(devices);
1636 return 1;
1637 }
1638 if (st->ss->store_super(st, fd)) {
1639 close(fd);
1640 pr_err("Could not re-write superblock on %s\n",
1641 devices[chosen_drive].devname);
1642 close(mdfd);
1643 free(devices);
1644 return 1;
1645 }
1646 if (c->verbose >= 0)
1647 pr_err("Marking array %s as 'clean'\n",
1648 mddev);
1649 close(fd);
1650 }
1651
1652 /* If we are in the middle of a reshape we may need to restore saved data
1653 * that was moved aside due to the reshape overwriting live data
1654 * The code of doing this lives in Grow.c
1655 */
1656 #ifndef MDASSEMBLE
1657 if (content->reshape_active &&
1658 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1659 int err = 0;
1660 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1661 if (c->verbose > 0)
1662 pr_err(":%s has an active reshape - checking "
1663 "if critical section needs to be restored\n",
1664 chosen_name);
1665 enable_fds(bestcnt/2);
1666 for (i = 0; i < bestcnt/2; i++) {
1667 int j = best[i*2];
1668 if (j >= 0) {
1669 fdlist[i] = dev_open(devices[j].devname,
1670 devices[j].included
1671 ? O_RDWR : (O_RDWR|O_EXCL));
1672 if (fdlist[i] < 0) {
1673 pr_err("Could not open %s for write - cannot Assemble array.\n",
1674 devices[j].devname);
1675 err = 1;
1676 break;
1677 }
1678 } else
1679 fdlist[i] = -1;
1680 }
1681 if (!err) {
1682 if (st->ss->external && st->ss->recover_backup)
1683 err = st->ss->recover_backup(st, content);
1684 else
1685 err = Grow_restart(st, content, fdlist, bestcnt/2,
1686 c->backup_file, c->verbose > 0);
1687 if (err && c->invalid_backup) {
1688 if (c->verbose > 0)
1689 pr_err("continuing"
1690 " without restoring backup\n");
1691 err = 0;
1692 }
1693 }
1694 while (i>0) {
1695 i--;
1696 if (fdlist[i]>=0) close(fdlist[i]);
1697 }
1698 free(fdlist);
1699 if (err) {
1700 pr_err("Failed to restore critical section for reshape, sorry.\n");
1701 if (c->backup_file == NULL)
1702 cont_err("Possibly you needed to specify the --backup-file\n");
1703 close(mdfd);
1704 free(devices);
1705 return err;
1706 }
1707 }
1708 #endif
1709
1710 /* Almost ready to actually *do* something */
1711 /* First, fill in the map, so that udev can find our name
1712 * as soon as we become active.
1713 */
1714 if (c->update && strcmp(c->update, "metadata")==0) {
1715 content->array.major_version = 1;
1716 content->array.minor_version = 0;
1717 strcpy(content->text_version, "1.0");
1718 }
1719
1720 map_update(&map, fd2devnm(mdfd), content->text_version,
1721 content->uuid, chosen_name);
1722
1723 rv = start_array(mdfd, mddev, content,
1724 st, ident, best, bestcnt,
1725 chosen_drive, devices, okcnt, sparecnt,
1726 rebuilding_cnt,
1727 c,
1728 clean, avail, start_partial_ok,
1729 pre_exist != NULL,
1730 was_forced);
1731 if (rv == 1 && !pre_exist)
1732 ioctl(mdfd, STOP_ARRAY, NULL);
1733 free(devices);
1734 map_unlock(&map);
1735 if (rv == 0) {
1736 wait_for(chosen_name, mdfd);
1737 close(mdfd);
1738 if (auto_assem) {
1739 int usecs = 1;
1740 /* There is a nasty race with 'mdadm --monitor'.
1741 * If it opens this device before we close it,
1742 * it gets an incomplete open on which IO
1743 * doesn't work and the capacity is
1744 * wrong.
1745 * If we reopen (to check for layered devices)
1746 * before --monitor closes, we loose.
1747 *
1748 * So: wait upto 1 second for there to be
1749 * a non-zero capacity.
1750 */
1751 while (usecs < 1000) {
1752 mdfd = open(mddev, O_RDONLY);
1753 if (mdfd >= 0) {
1754 unsigned long long size;
1755 if (get_dev_size(mdfd, NULL, &size) &&
1756 size > 0)
1757 break;
1758 close(mdfd);
1759 }
1760 usleep(usecs);
1761 usecs <<= 1;
1762 }
1763 }
1764 } else
1765 close(mdfd);
1766
1767 /* '2' means 'OK, but not started yet' */
1768 return rv == 2 ? 0 : rv;
1769 }
1770
1771 #ifndef MDASSEMBLE
1772 int assemble_container_content(struct supertype *st, int mdfd,
1773 struct mdinfo *content, struct context *c,
1774 char *chosen_name, int *result)
1775 {
1776 struct mdinfo *dev, *sra;
1777 int working = 0, preexist = 0;
1778 int expansion = 0;
1779 struct map_ent *map = NULL;
1780 int old_raid_disks;
1781 int start_reshape;
1782 char *avail = NULL;
1783 int err;
1784
1785 sysfs_init(content, mdfd, NULL);
1786
1787 sra = sysfs_read(mdfd, NULL, GET_VERSION);
1788 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1789 if (content->array.major_version == -1 &&
1790 content->array.minor_version == -2 &&
1791 c->readonly &&
1792 content->text_version[0] == '/')
1793 content->text_version[0] = '-';
1794 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1795 if (sra)
1796 sysfs_free(sra);
1797 return 1;
1798 }
1799 }
1800
1801 /* There are two types of reshape: container wide or sub-array specific
1802 * Check if metadata requests blocking container wide reshapes
1803 */
1804 start_reshape = (content->reshape_active &&
1805 !((content->reshape_active == CONTAINER_RESHAPE) &&
1806 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1807
1808 /* Block subarray here if it is under reshape now
1809 * Do not allow for any changes in this array
1810 */
1811 if (st->ss->external && content->recovery_blocked && start_reshape)
1812 block_subarray(content);
1813
1814 if (sra)
1815 sysfs_free(sra);
1816 old_raid_disks = content->array.raid_disks - content->delta_disks;
1817 avail = xcalloc(content->array.raid_disks, 1);
1818 for (dev = content->devs; dev; dev = dev->next) {
1819 if (dev->disk.raid_disk >= 0)
1820 avail[dev->disk.raid_disk] = 1;
1821 if (sysfs_add_disk(content, dev, 1) == 0) {
1822 if (dev->disk.raid_disk >= old_raid_disks &&
1823 content->reshape_active)
1824 expansion++;
1825 else
1826 working++;
1827 } else if (errno == EEXIST)
1828 preexist++;
1829 }
1830 if (working + expansion == 0 && c->runstop <= 0) {
1831 free(avail);
1832 return 1;/* Nothing new, don't try to start */
1833 }
1834 map_update(&map, fd2devnm(mdfd),
1835 content->text_version,
1836 content->uuid, chosen_name);
1837
1838
1839 if (enough(content->array.level, content->array.raid_disks,
1840 content->array.layout, content->array.state & 1, avail) == 0) {
1841 if (c->export && result)
1842 *result |= INCR_NO;
1843 else if (c->verbose >= 0) {
1844 pr_err("%s assembled with %d device%s",
1845 chosen_name, preexist + working,
1846 preexist + working == 1 ? "":"s");
1847 if (preexist)
1848 fprintf(stderr, " (%d new)", working);
1849 fprintf(stderr, " but not started\n");
1850 }
1851 free(avail);
1852 return 1;
1853 }
1854 free(avail);
1855
1856 if (c->runstop <= 0 &&
1857 (working + preexist + expansion) <
1858 content->array.working_disks) {
1859 if (c->export && result)
1860 *result |= INCR_UNSAFE;
1861 else if (c->verbose >= 0) {
1862 pr_err("%s assembled with %d device%s",
1863 chosen_name, preexist + working,
1864 preexist + working == 1 ? "":"s");
1865 if (preexist)
1866 fprintf(stderr, " (%d new)", working);
1867 fprintf(stderr, " but not safe to start\n");
1868 }
1869 return 1;
1870 }
1871
1872
1873 if (start_reshape) {
1874 int spare = content->array.raid_disks + expansion;
1875 if (restore_backup(st, content,
1876 working,
1877 spare, c->backup_file, c->verbose) == 1)
1878 return 1;
1879
1880 err = sysfs_set_str(content, NULL,
1881 "array_state", "readonly");
1882 if (err)
1883 return 1;
1884
1885 if (st->ss->external) {
1886 if (!mdmon_running(st->container_devnm))
1887 start_mdmon(st->container_devnm);
1888 ping_monitor(st->container_devnm);
1889 if (mdmon_running(st->container_devnm) &&
1890 st->update_tail == NULL)
1891 st->update_tail = &st->updates;
1892 }
1893
1894 err = Grow_continue(mdfd, st, content, c->backup_file,
1895 c->freeze_reshape);
1896 } else switch(content->array.level) {
1897 case LEVEL_LINEAR:
1898 case LEVEL_MULTIPATH:
1899 case 0:
1900 err = sysfs_set_str(content, NULL, "array_state",
1901 c->readonly ? "readonly" : "active");
1902 break;
1903 default:
1904 err = sysfs_set_str(content, NULL, "array_state",
1905 "readonly");
1906 /* start mdmon if needed. */
1907 if (!err) {
1908 if (!mdmon_running(st->container_devnm))
1909 start_mdmon(st->container_devnm);
1910 ping_monitor(st->container_devnm);
1911 }
1912 break;
1913 }
1914 if (!err)
1915 sysfs_set_safemode(content, content->safe_mode_delay);
1916
1917 /* Block subarray here if it is not reshaped now
1918 * It has be blocked a little later to allow mdmon to switch in
1919 * in to R/W state
1920 */
1921 if (st->ss->external && content->recovery_blocked &&
1922 !start_reshape)
1923 block_subarray(content);
1924
1925 if (c->export && result) {
1926 if (err)
1927 *result |= INCR_NO;
1928 else
1929 *result |= INCR_YES;
1930 } else if (c->verbose >= 0) {
1931 if (err)
1932 pr_err("array %s now has %d device%s",
1933 chosen_name, working + preexist,
1934 working + preexist == 1 ? "":"s");
1935 else
1936 pr_err("Started %s with %d device%s",
1937 chosen_name, working + preexist,
1938 working + preexist == 1 ? "":"s");
1939 if (preexist)
1940 fprintf(stderr, " (%d new)", working);
1941 if (expansion)
1942 fprintf(stderr, " ( + %d for expansion)",
1943 expansion);
1944 fprintf(stderr, "\n");
1945 }
1946 if (!err)
1947 wait_for(chosen_name, mdfd);
1948 return err;
1949 /* FIXME should have an O_EXCL and wait for read-auto */
1950 }
1951 #endif
Unnamed repository; edit this file 'description' to name the repository.