]> 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
Release mdadm-3.4
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
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, int require_homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char *sep;
34 unsigned int l;
35
36 if (strcmp(found, required)==0)
37 return 1;
38 sep = strchr(found, ':');
39 if (!sep)
40 return 0;
41 l = sep - found;
42 if (strncmp(found, "any:", 4) == 0 ||
43 (homehost && strcmp(homehost, "any") == 0) ||
44 !require_homehost ||
45 (homehost && strlen(homehost) == l &&
46 strncmp(found, homehost, l) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep+1, required) == 0)
49 return 1;
50 }
51 return 0;
52 }
53
54 static int is_member_busy(char *metadata_version)
55 {
56 /* check if the given member array is active */
57 struct mdstat_ent *mdstat = mdstat_read(0, 0);
58 struct mdstat_ent *ent;
59 int busy = 0;
60
61 for (ent = mdstat; ent; ent = ent->next) {
62 if (ent->metadata_version == NULL)
63 continue;
64 if (strncmp(ent->metadata_version, "external:", 9) != 0)
65 continue;
66 if (!is_subarray(&ent->metadata_version[9]))
67 continue;
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
70 busy = 1;
71 break;
72 }
73 }
74 free_mdstat(mdstat);
75
76 return busy;
77 }
78
79 static int ident_matches(struct mddev_ident *ident,
80 struct mdinfo *content,
81 struct supertype *tst,
82 char *homehost, int require_homehost,
83 char *update, char *devname)
84 {
85
86 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
87 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
88 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
89 if (devname)
90 pr_err("%s has wrong uuid.\n", devname);
91 return 0;
92 }
93 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
94 name_matches(content->name, ident->name, homehost, require_homehost)==0) {
95 if (devname)
96 pr_err("%s has wrong name.\n", devname);
97 return 0;
98 }
99 if (ident->super_minor != UnSet &&
100 ident->super_minor != content->array.md_minor) {
101 if (devname)
102 pr_err("%s has wrong super-minor.\n",
103 devname);
104 return 0;
105 }
106 if (ident->level != UnSet &&
107 ident->level != content->array.level) {
108 if (devname)
109 pr_err("%s has wrong raid level.\n",
110 devname);
111 return 0;
112 }
113 if (ident->raid_disks != UnSet &&
114 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
115 ident->raid_disks!= content->array.raid_disks) {
116 if (devname)
117 pr_err("%s requires wrong number of drives.\n",
118 devname);
119 return 0;
120 }
121 if (ident->member && ident->member[0]) {
122 /* content->text_version must match */
123 char *s = strchr(content->text_version+1, '/');
124 if (s == NULL) {
125 if (devname)
126 pr_err("%s is not a container and one is required.\n",
127 devname);
128 return 0;
129 } else if (strcmp(ident->member, s+1) != 0) {
130 if (devname)
131 pr_err("skipping wrong member %s is %s\n",
132 content->text_version, devname);
133 return 0;
134 }
135 }
136 return 1;
137 }
138
139 static int select_devices(struct mddev_dev *devlist,
140 struct mddev_ident *ident,
141 struct supertype **stp,
142 struct mdinfo **contentp,
143 struct context *c,
144 int inargv, int auto_assem)
145 {
146 struct mddev_dev *tmpdev;
147 int num_devs;
148 struct supertype *st = *stp;
149 struct mdinfo *content = NULL;
150 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
151 struct domainlist *domains = NULL;
152
153 tmpdev = devlist; num_devs = 0;
154 while (tmpdev) {
155 if (tmpdev->used)
156 tmpdev->used = 2;
157 else
158 num_devs++;
159 tmpdev->disposition = 0;
160 tmpdev = tmpdev->next;
161 }
162
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
166 */
167 for (tmpdev = devlist;
168 tmpdev;
169 tmpdev = tmpdev ? tmpdev->next : NULL) {
170 char *devname = tmpdev->devname;
171 int dfd;
172 struct stat stb;
173 struct supertype *tst;
174 struct dev_policy *pol = NULL;
175 int found_container = 0;
176
177 if (tmpdev->used > 1)
178 continue;
179
180 if (ident->container) {
181 if (ident->container[0] == '/' &&
182 !same_dev(ident->container, devname)) {
183 if (report_mismatch)
184 pr_err("%s is not the container required (%s)\n",
185 devname, ident->container);
186 continue;
187 }
188 } else if (ident->devices &&
189 !match_oneof(ident->devices, devname)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
192 * complicated.
193 */
194 if (report_mismatch)
195 pr_err("%s is not one of %s\n", devname, ident->devices);
196 continue;
197 }
198
199 tst = dup_super(st);
200
201 dfd = dev_open(devname, O_RDONLY);
202 if (dfd < 0) {
203 if (report_mismatch)
204 pr_err("cannot open device %s: %s\n",
205 devname, strerror(errno));
206 tmpdev->used = 2;
207 } else if (fstat(dfd, &stb)< 0) {
208 /* Impossible! */
209 pr_err("fstat failed for %s: %s\n",
210 devname, strerror(errno));
211 tmpdev->used = 2;
212 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
213 pr_err("%s is not a block device.\n",
214 devname);
215 tmpdev->used = 2;
216 } else if (must_be_container(dfd)) {
217 if (st) {
218 /* already found some components, this cannot
219 * be another one.
220 */
221 if (report_mismatch)
222 pr_err("%s is a container, but we are looking for components\n",
223 devname);
224 tmpdev->used = 2;
225 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
226 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
227 if (report_mismatch)
228 pr_err("not a recognisable container: %s\n",
229 devname);
230 tmpdev->used = 2;
231 #endif
232 } else if (!tst->ss->load_container
233 || tst->ss->load_container(tst, dfd, NULL)) {
234 if (report_mismatch)
235 pr_err("no correct container type: %s\n",
236 devname);
237 tmpdev->used = 2;
238 } else if (auto_assem &&
239 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
240 tst->ss->match_home(tst, c->homehost) == 1)) {
241 if (report_mismatch)
242 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
243 devname, tst->ss->name);
244 tmpdev->used = 2;
245 } else
246 found_container = 1;
247 } else {
248 if (!tst && (tst = guess_super(dfd)) == NULL) {
249 if (report_mismatch)
250 pr_err("no recogniseable superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if ((tst->ignore_hw_compat = 0),
254 tst->ss->load_super(tst, dfd,
255 report_mismatch ? devname : NULL)) {
256 if (report_mismatch)
257 pr_err("no RAID superblock on %s\n",
258 devname);
259 tmpdev->used = 2;
260 } else if (tst->ss->compare_super == NULL) {
261 if (report_mismatch)
262 pr_err("Cannot assemble %s metadata on %s\n",
263 tst->ss->name, devname);
264 tmpdev->used = 2;
265 } else if (auto_assem && st == NULL &&
266 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
267 tst->ss->match_home(tst, c->homehost) == 1)) {
268 if (report_mismatch)
269 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
270 devname, tst->ss->name);
271 tmpdev->used = 2;
272 }
273 }
274 if (dfd >= 0) close(dfd);
275 if (tmpdev->used == 2) {
276 if (auto_assem || !inargv)
277 /* Ignore unrecognised devices during auto-assembly */
278 goto loop;
279 if (ident->uuid_set || ident->name[0] ||
280 ident->super_minor != UnSet)
281 /* Ignore unrecognised device if looking for
282 * specific array */
283 goto loop;
284
285 pr_err("%s has no superblock - assembly aborted\n",
286 devname);
287 if (st)
288 st->ss->free_super(st);
289 dev_policy_free(pol);
290 domain_free(domains);
291 return -1;
292 }
293
294 if (found_container) {
295 /* tmpdev is a container. We need to be either
296 * looking for a member, or auto-assembling
297 */
298 /* should be safe to try an exclusive open now, we
299 * have rejected anything that some other mdadm might
300 * be looking at
301 */
302 dfd = dev_open(devname, O_RDONLY | O_EXCL);
303 if (dfd < 0) {
304 if (report_mismatch)
305 pr_err("%s is busy - skipping\n", devname);
306 goto loop;
307 }
308 close(dfd);
309
310 if (ident->container && ident->container[0] != '/') {
311 /* we have a uuid */
312 int uuid[4];
313
314 content = *contentp;
315 tst->ss->getinfo_super(tst, content, NULL);
316
317 if (!parse_uuid(ident->container, uuid) ||
318 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
319 if (report_mismatch)
320 pr_err("%s has wrong UUID to be required container\n",
321 devname);
322 goto loop;
323 }
324 }
325 /* It is worth looking inside this container.
326 */
327 if (c->verbose > 0)
328 pr_err("looking in container %s\n",
329 devname);
330
331 for (content = tst->ss->container_content(tst, NULL);
332 content;
333 content = content->next) {
334
335 if (!ident_matches(ident, content, tst,
336 c->homehost, c->require_homehost,
337 c->update,
338 report_mismatch ? devname : NULL))
339 /* message already printed */;
340 else if (is_member_busy(content->text_version)) {
341 if (report_mismatch)
342 pr_err("member %s in %s is already assembled\n",
343 content->text_version,
344 devname);
345 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
346 /* do not assemble arrays with unsupported configurations */
347 pr_err("Cannot activate member %s in %s.\n",
348 content->text_version,
349 devname);
350 } else
351 break;
352 }
353 if (!content) {
354 tmpdev->used = 2;
355 goto loop; /* empty container */
356 }
357
358 st = tst; tst = NULL;
359 if (!auto_assem && inargv && tmpdev->next != NULL) {
360 pr_err("%s is a container, but is not only device given: confused and aborting\n",
361 devname);
362 st->ss->free_super(st);
363 dev_policy_free(pol);
364 domain_free(domains);
365 return -1;
366 }
367 if (c->verbose > 0)
368 pr_err("found match on member %s in %s\n",
369 content->text_version, devname);
370
371 /* make sure we finished the loop */
372 tmpdev = NULL;
373 goto loop;
374 } else {
375 content = *contentp;
376 tst->ss->getinfo_super(tst, content, NULL);
377
378 if (!ident_matches(ident, content, tst,
379 c->homehost, c->require_homehost,
380 c->update,
381 report_mismatch ? devname : NULL))
382 goto loop;
383
384 if (auto_assem) {
385 /* Never auto-assemble things that conflict
386 * with mdadm.conf in some way
387 */
388 struct mddev_ident *match;
389 int rv = 0;
390
391 match = conf_match(tst, content, devname,
392 report_mismatch ? c->verbose : -1,
393 &rv);
394 if (!match && rv == 2)
395 goto loop;
396 if (match && match->devname &&
397 strcasecmp(match->devname, "<ignore>") == 0) {
398 if (report_mismatch)
399 pr_err("%s is a member of an explicitly ignored array\n",
400 devname);
401 goto loop;
402 }
403 if (match && !ident_matches(match, content, tst,
404 c->homehost, c->require_homehost,
405 c->update,
406 report_mismatch ? devname : NULL))
407 /* Array exists in mdadm.conf but some
408 * details don't match, so reject it
409 */
410 goto loop;
411 }
412
413 /* should be safe to try an exclusive open now, we
414 * have rejected anything that some other mdadm might
415 * be looking at
416 */
417 dfd = dev_open(devname, O_RDONLY | O_EXCL);
418 if (dfd < 0) {
419 if (report_mismatch)
420 pr_err("%s is busy - skipping\n", devname);
421 goto loop;
422 }
423 close(dfd);
424
425 if (st == NULL)
426 st = dup_super(tst);
427 if (st->minor_version == -1)
428 st->minor_version = tst->minor_version;
429
430 if (memcmp(content->uuid, uuid_zero,
431 sizeof(int[4])) == 0) {
432 /* this is a floating spare. It cannot define
433 * an array unless there are no more arrays of
434 * this type to be found. It can be included
435 * in an array of this type though.
436 */
437 tmpdev->used = 3;
438 goto loop;
439 }
440
441 if (st->ss != tst->ss ||
442 st->minor_version != tst->minor_version ||
443 st->ss->compare_super(st, tst) != 0) {
444 /* Some mismatch. If exactly one array matches this host,
445 * we can resolve on that one.
446 * Or, if we are auto assembling, we just ignore the second
447 * for now.
448 */
449 if (auto_assem)
450 goto loop;
451 if (c->homehost) {
452 int first = st->ss->match_home(st, c->homehost);
453 int last = tst->ss->match_home(tst, c->homehost);
454 if (first != last &&
455 (first == 1 || last == 1)) {
456 /* We can do something */
457 if (first) {/* just ignore this one */
458 if (report_mismatch)
459 pr_err("%s misses out due to wrong homehost\n",
460 devname);
461 goto loop;
462 } else { /* reject all those sofar */
463 struct mddev_dev *td;
464 if (report_mismatch)
465 pr_err("%s overrides previous devices due to good homehost\n",
466 devname);
467 for (td=devlist; td != tmpdev; td=td->next)
468 if (td->used == 1)
469 td->used = 0;
470 tmpdev->used = 1;
471 goto loop;
472 }
473 }
474 }
475 pr_err("superblock on %s doesn't match others - assembly aborted\n",
476 devname);
477 tst->ss->free_super(tst);
478 st->ss->free_super(st);
479 dev_policy_free(pol);
480 domain_free(domains);
481 return -1;
482 }
483 tmpdev->used = 1;
484 }
485 loop:
486 /* Collect domain information from members only */
487 if (tmpdev && tmpdev->used == 1) {
488 if (!pol)
489 pol = devid_policy(stb.st_rdev);
490 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
491 }
492 dev_policy_free(pol);
493 pol = NULL;
494 if (tst)
495 tst->ss->free_super(tst);
496 }
497
498 /* Check if we found some imsm spares but no members */
499 if ((auto_assem ||
500 (ident->uuid_set &&
501 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
502 (!st || !st->sb))
503 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
504 if (tmpdev->used != 3)
505 continue;
506 tmpdev->used = 1;
507 content = *contentp;
508
509 if (!st->sb) {
510 /* we need sb from one of the spares */
511 int dfd = dev_open(tmpdev->devname, O_RDONLY);
512 if (dfd < 0 ||
513 st->ss->load_super(st, dfd, NULL))
514 tmpdev->used = 2;
515 if (dfd > 0)
516 close(dfd);
517 }
518 }
519
520 /* Now reject spares that don't match domains of identified members */
521 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
522 struct stat stb;
523 if (tmpdev->used != 3)
524 continue;
525 if (stat(tmpdev->devname, &stb)< 0) {
526 pr_err("fstat failed for %s: %s\n",
527 tmpdev->devname, strerror(errno));
528 tmpdev->used = 2;
529 } else {
530 struct dev_policy *pol = devid_policy(stb.st_rdev);
531 int dt = domain_test(domains, pol, NULL);
532 if (inargv && dt != 0)
533 /* take this spare as domains match
534 * if there are any */
535 tmpdev->used = 1;
536 else if (!inargv && dt == 1)
537 /* device wasn't explicitly listed, so need
538 * explicit domain match - which we have */
539 tmpdev->used = 1;
540 else
541 /* if domains don't match mark as unused */
542 tmpdev->used = 0;
543 dev_policy_free(pol);
544 }
545 }
546 domain_free(domains);
547 *stp = st;
548 if (st && st->sb && content == *contentp)
549 st->ss->getinfo_super(st, content, NULL);
550 *contentp = content;
551
552 return num_devs;
553 }
554
555 struct devs {
556 char *devname;
557 int uptodate; /* set once we decide that this device is as
558 * recent as everything else in the array.
559 */
560 int included; /* set if the device is already in the array
561 * due to a previous '-I'
562 */
563 struct mdinfo i;
564 };
565
566 static int load_devices(struct devs *devices, char *devmap,
567 struct mddev_ident *ident, struct supertype **stp,
568 struct mddev_dev *devlist, struct context *c,
569 struct mdinfo *content,
570 int mdfd, char *mddev,
571 int *most_recentp, int *bestcntp, int **bestp,
572 int inargv)
573 {
574 struct mddev_dev *tmpdev;
575 int devcnt = 0;
576 int nextspare = 0;
577 #ifndef MDASSEMBLE
578 int bitmap_done = 0;
579 #endif
580 int most_recent = -1;
581 int bestcnt = 0;
582 int *best = *bestp;
583 struct supertype *st = *stp;
584
585 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
586 char *devname = tmpdev->devname;
587 struct stat stb;
588 struct supertype *tst;
589 int i;
590 int dfd;
591
592 if (tmpdev->used != 1)
593 continue;
594 /* looks like a good enough match to update the super block if needed */
595 #ifndef MDASSEMBLE
596 if (c->update) {
597 /* prepare useful information in info structures */
598 struct stat stb2;
599 int err;
600 fstat(mdfd, &stb2);
601
602 if (strcmp(c->update, "uuid")==0 &&
603 !ident->uuid_set) {
604 int rfd;
605 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
606 read(rfd, ident->uuid, 16) != 16) {
607 *(__u32*)(ident->uuid) = random();
608 *(__u32*)(ident->uuid+1) = random();
609 *(__u32*)(ident->uuid+2) = random();
610 *(__u32*)(ident->uuid+3) = random();
611 }
612 if (rfd >= 0) close(rfd);
613 }
614 dfd = dev_open(devname,
615 tmpdev->disposition == 'I'
616 ? O_RDWR : (O_RDWR|O_EXCL));
617
618 tst = dup_super(st);
619 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
620 pr_err("cannot re-read metadata from %s - aborting\n",
621 devname);
622 if (dfd >= 0)
623 close(dfd);
624 close(mdfd);
625 free(devices);
626 free(devmap);
627 tst->ss->free_super(tst);
628 free(tst);
629 *stp = st;
630 return -1;
631 }
632 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
633
634 memcpy(content->uuid, ident->uuid, 16);
635 strcpy(content->name, ident->name);
636 content->array.md_minor = minor(stb2.st_rdev);
637
638 if (strcmp(c->update, "byteorder") == 0)
639 err = 0;
640 else if (strcmp(c->update, "home-cluster") == 0) {
641 tst->cluster_name = c->homecluster;
642 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
643 } else if (strcmp(c->update, "nodes") == 0) {
644 tst->nodes = c->nodes;
645 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
646 } else if (strcmp(c->update, "revert-reshape") == 0 &&
647 c->invalid_backup)
648 err = tst->ss->update_super(tst, content,
649 "revert-reshape-nobackup",
650 devname, c->verbose,
651 ident->uuid_set,
652 c->homehost);
653 else
654 err = tst->ss->update_super(tst, content, c->update,
655 devname, c->verbose,
656 ident->uuid_set,
657 c->homehost);
658 if (err < 0) {
659 if (err == -1)
660 pr_err("--update=%s not understood for %s metadata\n",
661 c->update, tst->ss->name);
662 tst->ss->free_super(tst);
663 free(tst);
664 close(mdfd);
665 close(dfd);
666 free(devices);
667 free(devmap);
668 *stp = st;
669 return -1;
670 }
671 if (strcmp(c->update, "uuid")==0 &&
672 !ident->uuid_set) {
673 ident->uuid_set = 1;
674 memcpy(ident->uuid, content->uuid, 16);
675 }
676 if (tst->ss->store_super(tst, dfd))
677 pr_err("Could not re-write superblock on %s.\n",
678 devname);
679
680 if (strcmp(c->update, "uuid")==0 &&
681 ident->bitmap_fd >= 0 && !bitmap_done) {
682 if (bitmap_update_uuid(ident->bitmap_fd,
683 content->uuid,
684 tst->ss->swapuuid) != 0)
685 pr_err("Could not update uuid on external bitmap.\n");
686 else
687 bitmap_done = 1;
688 }
689 } else
690 #endif
691 {
692 dfd = dev_open(devname,
693 tmpdev->disposition == 'I'
694 ? O_RDWR : (O_RDWR|O_EXCL));
695 tst = dup_super(st);
696
697 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
698 pr_err("cannot re-read metadata from %s - aborting\n",
699 devname);
700 if (dfd >= 0)
701 close(dfd);
702 close(mdfd);
703 free(devices);
704 free(devmap);
705 tst->ss->free_super(tst);
706 free(tst);
707 *stp = st;
708 return -1;
709 }
710 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
711 }
712
713 fstat(dfd, &stb);
714 close(dfd);
715
716 if (c->verbose > 0)
717 pr_err("%s is identified as a member of %s, slot %d%s.\n",
718 devname, mddev, content->disk.raid_disk,
719 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
720 devices[devcnt].devname = devname;
721 devices[devcnt].uptodate = 0;
722 devices[devcnt].included = (tmpdev->disposition == 'I');
723 devices[devcnt].i = *content;
724 devices[devcnt].i.disk.major = major(stb.st_rdev);
725 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
726
727 if (devices[devcnt].i.disk.state == 6) {
728 if (most_recent < 0 ||
729 devices[devcnt].i.events
730 > devices[most_recent].i.events) {
731 struct supertype *tmp = tst;
732 tst = st;
733 st = tmp;
734 most_recent = devcnt;
735 }
736 }
737 tst->ss->free_super(tst);
738 free(tst);
739
740 if (content->array.level == LEVEL_MULTIPATH)
741 /* with multipath, the raid_disk from the superblock is meaningless */
742 i = devcnt;
743 else
744 i = devices[devcnt].i.disk.raid_disk;
745 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
746 if (nextspare < content->array.raid_disks*2)
747 nextspare = content->array.raid_disks*2;
748 i = nextspare++;
749 } else {
750 /* i is raid_disk - double it so there is room for
751 * replacements */
752 i *= 2;
753 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
754 i++;
755 if (i >= content->array.raid_disks*2 &&
756 i >= nextspare)
757 nextspare = i+1;
758 }
759 if (i < 10000) {
760 if (i >= bestcnt) {
761 int newbestcnt = i+10;
762 int *newbest = xmalloc(sizeof(int)*newbestcnt);
763 int c;
764 for (c=0; c < newbestcnt; c++)
765 if (c < bestcnt)
766 newbest[c] = best[c];
767 else
768 newbest[c] = -1;
769 if (best)free(best);
770 best = newbest;
771 bestcnt = newbestcnt;
772 }
773 if (best[i] >=0 &&
774 devices[best[i]].i.events
775 == devices[devcnt].i.events
776 && (devices[best[i]].i.disk.minor
777 != devices[devcnt].i.disk.minor)
778 && st->ss == &super0
779 && content->array.level != LEVEL_MULTIPATH) {
780 /* two different devices with identical superblock.
781 * Could be a mis-detection caused by overlapping
782 * partitions. fail-safe.
783 */
784 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
785 " If they are really different, please --zero the superblock on one\n"
786 " If they are the same or overlap, please remove one from %s.\n",
787 devices[best[i]].devname, devname,
788 inargv ? "the list" :
789 "the\n DEVICE list in mdadm.conf"
790 );
791 close(mdfd);
792 free(devices);
793 free(devmap);
794 *stp = st;
795 return -1;
796 }
797 if (best[i] == -1
798 || (devices[best[i]].i.events
799 < devices[devcnt].i.events))
800 best[i] = devcnt;
801 }
802 devcnt++;
803 }
804 if (most_recent >= 0)
805 *most_recentp = most_recent;
806 *bestcntp = bestcnt;
807 *bestp = best;
808 *stp = st;
809 return devcnt;
810 }
811
812 static int force_array(struct mdinfo *content,
813 struct devs *devices,
814 int *best, int bestcnt, char *avail,
815 int most_recent,
816 struct supertype *st,
817 struct context *c)
818 {
819 int okcnt = 0;
820 while (!enough(content->array.level, content->array.raid_disks,
821 content->array.layout, 1,
822 avail)
823 ||
824 (content->reshape_active && content->delta_disks > 0 &&
825 !enough(content->array.level, (content->array.raid_disks
826 - content->delta_disks),
827 content->new_layout, 1,
828 avail)
829 )) {
830 /* Choose the newest best drive which is
831 * not up-to-date, update the superblock
832 * and add it.
833 */
834 int fd;
835 struct supertype *tst;
836 unsigned long long current_events;
837 int chosen_drive = -1;
838 int i;
839
840 for (i = 0;
841 i < content->array.raid_disks * 2 && i < bestcnt;
842 i += 2) {
843 int j = best[i];
844 if (j < 0)
845 continue;
846 if (devices[j].uptodate)
847 continue;
848 if (devices[j].i.recovery_start != MaxSector) {
849 int delta;
850 if (!devices[j].i.reshape_active ||
851 devices[j].i.delta_disks <= 0)
852 continue;
853 /* When increasing number of devices, an
854 * added device also appears to be
855 * recovering. It is safe to include it
856 * as long as it won't be a source of
857 * data.
858 * For now, just allow for last data
859 * devices in RAID4 or last devices in RAID4/5/6.
860 */
861 delta = devices[j].i.delta_disks;
862 if (devices[j].i.array.level >= 4 &&
863 devices[j].i.array.level <= 6 &&
864 i/2 >= content->array.raid_disks - delta)
865 /* OK */;
866 else if (devices[j].i.array.level == 4 &&
867 i/2 >= content->array.raid_disks - delta - 1)
868 /* OK */;
869 else
870 continue;
871 }
872 if (chosen_drive < 0 ||
873 devices[j].i.events
874 > devices[chosen_drive].i.events)
875 chosen_drive = j;
876 }
877 if (chosen_drive < 0)
878 break;
879 current_events = devices[chosen_drive].i.events;
880 add_another:
881 if (c->verbose >= 0)
882 pr_err("forcing event count in %s(%d) from %d upto %d\n",
883 devices[chosen_drive].devname,
884 devices[chosen_drive].i.disk.raid_disk,
885 (int)(devices[chosen_drive].i.events),
886 (int)(devices[most_recent].i.events));
887 fd = dev_open(devices[chosen_drive].devname,
888 devices[chosen_drive].included ? O_RDWR
889 : (O_RDWR|O_EXCL));
890 if (fd < 0) {
891 pr_err("Couldn't open %s for write - not updating\n",
892 devices[chosen_drive].devname);
893 devices[chosen_drive].i.events = 0;
894 continue;
895 }
896 tst = dup_super(st);
897 if (tst->ss->load_super(tst,fd, NULL)) {
898 close(fd);
899 pr_err("RAID superblock disappeared from %s - not updating.\n",
900 devices[chosen_drive].devname);
901 devices[chosen_drive].i.events = 0;
902 continue;
903 }
904 content->events = devices[most_recent].i.events;
905 tst->ss->update_super(tst, content, "force-one",
906 devices[chosen_drive].devname, c->verbose,
907 0, NULL);
908
909 if (tst->ss->store_super(tst, fd)) {
910 close(fd);
911 pr_err("Could not re-write superblock on %s\n",
912 devices[chosen_drive].devname);
913 devices[chosen_drive].i.events = 0;
914 tst->ss->free_super(tst);
915 continue;
916 }
917 close(fd);
918 devices[chosen_drive].i.events = devices[most_recent].i.events;
919 devices[chosen_drive].uptodate = 1;
920 avail[chosen_drive] = 1;
921 okcnt++;
922 tst->ss->free_super(tst);
923 /* If there are any other drives of the same vintage,
924 * add them in as well. We can't lose and we might gain
925 */
926 for (i = 0;
927 i < content->array.raid_disks * 2 && i < bestcnt ;
928 i += 2) {
929 int j = best[i];
930 if (j >= 0 &&
931 !devices[j].uptodate &&
932 devices[j].i.recovery_start == MaxSector &&
933 devices[j].i.events == current_events) {
934 chosen_drive = j;
935 goto add_another;
936 }
937 }
938 }
939 return okcnt;
940 }
941
942 static int start_array(int mdfd,
943 char *mddev,
944 struct mdinfo *content,
945 struct supertype *st,
946 struct mddev_ident *ident,
947 int *best, int bestcnt,
948 int chosen_drive,
949 struct devs *devices,
950 unsigned int okcnt,
951 unsigned int sparecnt,
952 unsigned int rebuilding_cnt,
953 unsigned int journalcnt,
954 struct context *c,
955 int clean, char *avail,
956 int start_partial_ok,
957 int err_ok,
958 int was_forced
959 )
960 {
961 int rv;
962 int i;
963 unsigned int req_cnt;
964
965 if (content->journal_device_required && (content->journal_clean == 0)) {
966 if (!c->force) {
967 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
968 return 1;
969 }
970 pr_err("Journal is missing or stale, starting array read only.\n");
971 c->readonly = 1;
972 }
973
974 rv = set_array_info(mdfd, st, content);
975 if (rv && !err_ok) {
976 pr_err("failed to set array info for %s: %s\n",
977 mddev, strerror(errno));
978 return 1;
979 }
980 if (ident->bitmap_fd >= 0) {
981 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
982 pr_err("SET_BITMAP_FILE failed.\n");
983 return 1;
984 }
985 } else if (ident->bitmap_file) {
986 /* From config file */
987 int bmfd = open(ident->bitmap_file, O_RDWR);
988 if (bmfd < 0) {
989 pr_err("Could not open bitmap file %s\n",
990 ident->bitmap_file);
991 return 1;
992 }
993 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
994 pr_err("Failed to set bitmapfile for %s\n", mddev);
995 close(bmfd);
996 return 1;
997 }
998 close(bmfd);
999 }
1000
1001 /* First, add the raid disks, but add the chosen one last */
1002 for (i=0; i<= bestcnt; i++) {
1003 int j;
1004 if (i < bestcnt) {
1005 j = best[i];
1006 if (j == chosen_drive)
1007 continue;
1008 } else
1009 j = chosen_drive;
1010
1011 if (j >= 0 && !devices[j].included) {
1012 int dfd = dev_open(devices[j].devname,
1013 O_RDWR|O_EXCL);
1014 if (dfd >= 0) {
1015 remove_partitions(dfd);
1016 close(dfd);
1017 }
1018 rv = add_disk(mdfd, st, content, &devices[j].i);
1019
1020 if (rv) {
1021 pr_err("failed to add %s to %s: %s\n",
1022 devices[j].devname,
1023 mddev,
1024 strerror(errno));
1025 if (i < content->array.raid_disks * 2
1026 || i == bestcnt)
1027 okcnt--;
1028 else
1029 sparecnt--;
1030 } else if (c->verbose > 0)
1031 pr_err("added %s to %s as %d%s%s\n",
1032 devices[j].devname, mddev,
1033 devices[j].i.disk.raid_disk,
1034 devices[j].uptodate?"":
1035 " (possibly out of date)",
1036 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1037 } else if (j >= 0) {
1038 if (c->verbose > 0)
1039 pr_err("%s is already in %s as %d\n",
1040 devices[j].devname, mddev,
1041 devices[j].i.disk.raid_disk);
1042 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1043 && (i&1) == 0)
1044 pr_err("no uptodate device for slot %d of %s\n",
1045 i/2, mddev);
1046 }
1047
1048 if (content->array.level == LEVEL_CONTAINER) {
1049 if (c->verbose >= 0) {
1050 pr_err("Container %s has been assembled with %d drive%s",
1051 mddev, okcnt+sparecnt+journalcnt,
1052 okcnt+sparecnt+journalcnt==1?"":"s");
1053 if (okcnt < (unsigned)content->array.raid_disks)
1054 fprintf(stderr, " (out of %d)",
1055 content->array.raid_disks);
1056 fprintf(stderr, "\n");
1057 }
1058
1059 if (st->ss->validate_container) {
1060 struct mdinfo *devices_list;
1061 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1062 unsigned int count;
1063 devices_list = NULL;
1064 for (count = 0; count < okcnt+sparecnt; count++) {
1065 info_devices[count] = devices[count].i;
1066 info_devices[count].next = devices_list;
1067 devices_list = &info_devices[count];
1068 }
1069 if (st->ss->validate_container(devices_list))
1070 pr_err("Mismatch detected!\n");
1071 free(info_devices);
1072 }
1073
1074 st->ss->free_super(st);
1075 sysfs_uevent(content, "change");
1076 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1077 /* Was partial, is still partial, so signal an error
1078 * to ensure we don't retry */
1079 return 1;
1080 return 0;
1081 }
1082
1083 /* Get number of in-sync devices according to the superblock.
1084 * We must have this number to start the array without -s or -R
1085 */
1086 req_cnt = content->array.working_disks;
1087
1088 if (c->runstop == 1 ||
1089 (c->runstop <= 0 &&
1090 ( enough(content->array.level, content->array.raid_disks,
1091 content->array.layout, clean, avail) &&
1092 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1093 ))) {
1094 /* This array is good-to-go.
1095 * If a reshape is in progress then we might need to
1096 * continue monitoring it. In that case we start
1097 * it read-only and let the grow code make it writable.
1098 */
1099 int rv;
1100 #ifndef MDASSEMBLE
1101 if (content->reshape_active &&
1102 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1103 content->delta_disks <= 0) {
1104 if (!c->backup_file) {
1105 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1106 mddev);
1107 pr_err("Please provided one with \"--backup-file=...\"\n");
1108 if (c->update &&
1109 strcmp(c->update, "revert-reshape") == 0)
1110 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1111 return 1;
1112 }
1113 rv = sysfs_set_str(content, NULL,
1114 "array_state", "readonly");
1115 if (rv == 0)
1116 rv = Grow_continue(mdfd, st, content,
1117 c->backup_file, 0,
1118 c->freeze_reshape);
1119 } else if (c->readonly &&
1120 sysfs_attribute_available(
1121 content, NULL, "array_state")) {
1122 rv = sysfs_set_str(content, NULL,
1123 "array_state", "readonly");
1124 } else
1125 #endif
1126 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1127 reopen_mddev(mdfd); /* drop O_EXCL */
1128 if (rv == 0) {
1129 if (c->verbose >= 0) {
1130 pr_err("%s has been started with %d drive%s",
1131 mddev, okcnt, okcnt==1?"":"s");
1132 if (okcnt < (unsigned)content->array.raid_disks)
1133 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1134 if (rebuilding_cnt)
1135 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1136 if (sparecnt)
1137 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1138 if (content->journal_clean)
1139 fprintf(stderr, " and %d journal", journalcnt);
1140 fprintf(stderr, ".\n");
1141 }
1142 if (content->reshape_active &&
1143 content->array.level >= 4 &&
1144 content->array.level <= 6) {
1145 /* might need to increase the size
1146 * of the stripe cache - default is 256
1147 */
1148 int chunk_size = content->array.chunk_size;
1149 if (content->reshape_active &&
1150 content->new_chunk > chunk_size)
1151 chunk_size = content->new_chunk;
1152 if (256 < 4 * ((chunk_size+4065)/4096)) {
1153 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1154 if (sra)
1155 sysfs_set_num(sra, NULL,
1156 "stripe_cache_size",
1157 (4 * chunk_size / 4096) + 1);
1158 sysfs_free(sra);
1159 }
1160 }
1161 if (okcnt < (unsigned)content->array.raid_disks) {
1162 /* If any devices did not get added
1163 * because the kernel rejected them based
1164 * on event count, try adding them
1165 * again providing the action policy is
1166 * 're-add' or greater. The bitmap
1167 * might allow them to be included, or
1168 * they will become spares.
1169 */
1170 for (i = 0; i < bestcnt; i++) {
1171 int j = best[i];
1172 if (j >= 0 && !devices[j].uptodate) {
1173 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1174 continue;
1175 rv = add_disk(mdfd, st, content,
1176 &devices[j].i);
1177 if (rv == 0 && c->verbose >= 0)
1178 pr_err("%s has been re-added.\n",
1179 devices[j].devname);
1180 }
1181 }
1182 }
1183 if (content->array.level == 6 &&
1184 okcnt + 1 == (unsigned)content->array.raid_disks &&
1185 was_forced) {
1186 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1187 if (sra)
1188 sysfs_set_str(sra, NULL,
1189 "sync_action", "repair");
1190 sysfs_free(sra);
1191 }
1192 return 0;
1193 }
1194 pr_err("failed to RUN_ARRAY %s: %s\n",
1195 mddev, strerror(errno));
1196
1197 if (!enough(content->array.level, content->array.raid_disks,
1198 content->array.layout, 1, avail))
1199 pr_err("Not enough devices to start the array.\n");
1200 else if (!enough(content->array.level,
1201 content->array.raid_disks,
1202 content->array.layout, clean,
1203 avail))
1204 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1205
1206 return 1;
1207 }
1208 if (c->runstop == -1) {
1209 pr_err("%s assembled from %d drive%s",
1210 mddev, okcnt, okcnt==1?"":"s");
1211 if (okcnt != (unsigned)content->array.raid_disks)
1212 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1213 fprintf(stderr, ", but not started.\n");
1214 return 2;
1215 }
1216 if (c->verbose >= -1) {
1217 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1218 if (rebuilding_cnt)
1219 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1220 if (sparecnt)
1221 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1222 if (!enough(content->array.level, content->array.raid_disks,
1223 content->array.layout, 1, avail))
1224 fprintf(stderr, " - not enough to start the array.\n");
1225 else if (!enough(content->array.level,
1226 content->array.raid_disks,
1227 content->array.layout, clean,
1228 avail))
1229 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1230 else {
1231 if (req_cnt == (unsigned)content->array.raid_disks)
1232 fprintf(stderr, " - need all %d to start it", req_cnt);
1233 else
1234 fprintf(stderr, " - need %d to start", req_cnt);
1235 fprintf(stderr, " (use --run to insist).\n");
1236 }
1237 }
1238 return 1;
1239 }
1240
1241 int Assemble(struct supertype *st, char *mddev,
1242 struct mddev_ident *ident,
1243 struct mddev_dev *devlist,
1244 struct context *c)
1245 {
1246 /*
1247 * The task of Assemble is to find a collection of
1248 * devices that should (according to their superblocks)
1249 * form an array, and to give this collection to the MD driver.
1250 * In Linux-2.4 and later, this involves submitting a
1251 * SET_ARRAY_INFO ioctl with no arg - to prepare
1252 * the array - and then submit a number of
1253 * ADD_NEW_DISK ioctls to add disks into
1254 * the array. Finally RUN_ARRAY might
1255 * be submitted to start the array.
1256 *
1257 * Much of the work of Assemble is in finding and/or
1258 * checking the disks to make sure they look right.
1259 *
1260 * If mddev is not set, then scan must be set and we
1261 * read through the config file for dev+uuid mapping
1262 * We recurse, setting mddev, for each device that
1263 * - isn't running
1264 * - has a valid uuid (or any uuid if !uuidset)
1265 *
1266 * If mddev is set, we try to determine state of md.
1267 * check version - must be at least 0.90.0
1268 * check kernel version. must be at least 2.4.
1269 * If not, we can possibly fall back on START_ARRAY
1270 * Try to GET_ARRAY_INFO.
1271 * If possible, give up
1272 * If not, try to STOP_ARRAY just to make sure
1273 *
1274 * If !uuidset and scan, look in conf-file for uuid
1275 * If not found, give up
1276 * If !devlist and scan and uuidset, get list of devs from conf-file
1277 *
1278 * For each device:
1279 * Check superblock - discard if bad
1280 * Check uuid (set if we don't have one) - discard if no match
1281 * Check superblock similarity if we have a superblock - discard if different
1282 * Record events, devicenum
1283 * This should give us a list of devices for the array
1284 * We should collect the most recent event number
1285 *
1286 * Count disks with recent enough event count
1287 * While force && !enough disks
1288 * Choose newest rejected disks, update event count
1289 * mark clean and rewrite superblock
1290 * If recent kernel:
1291 * SET_ARRAY_INFO
1292 * foreach device with recent events : ADD_NEW_DISK
1293 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1294 * If old kernel:
1295 * Check the device numbers in superblock are right
1296 * update superblock if any changes
1297 * START_ARRAY
1298 *
1299 */
1300 int rv;
1301 int mdfd;
1302 int clean;
1303 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1304 ident->super_minor == UnSet && ident->name[0] == 0
1305 && (ident->container == NULL || ident->member == NULL));
1306 struct devs *devices;
1307 char *devmap;
1308 int *best = NULL; /* indexed by raid_disk */
1309 int bestcnt = 0;
1310 int devcnt;
1311 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1312 int journal_clean = 0;
1313 int i;
1314 int was_forced = 0;
1315 int most_recent = 0;
1316 int chosen_drive;
1317 int change = 0;
1318 int inargv = 0;
1319 int start_partial_ok = (c->runstop >= 0) &&
1320 (c->force || devlist==NULL || auto_assem);
1321 int num_devs;
1322 struct mddev_dev *tmpdev;
1323 struct mdinfo info;
1324 struct mdinfo *content = NULL;
1325 struct mdinfo *pre_exist = NULL;
1326 char *avail;
1327 char *name = NULL;
1328 char chosen_name[1024];
1329 struct map_ent *map = NULL;
1330 struct map_ent *mp;
1331
1332 /*
1333 * If any subdevs are listed, then any that don't
1334 * match ident are discarded. Remainder must all match and
1335 * become the array.
1336 * If no subdevs, then we scan all devices in the config file, but
1337 * there must be something in the identity
1338 */
1339
1340 if (!devlist &&
1341 ident->uuid_set == 0 &&
1342 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1343 ident->name[0] == 0 &&
1344 (ident->container == NULL || ident->member == NULL) &&
1345 ident->devices == NULL) {
1346 pr_err("No identity information available for %s - cannot assemble.\n",
1347 mddev ? mddev : "further assembly");
1348 return 1;
1349 }
1350
1351 if (devlist == NULL)
1352 devlist = conf_get_devs();
1353 else if (mddev)
1354 inargv = 1;
1355
1356 try_again:
1357 /* We come back here when doing auto-assembly and attempting some
1358 * set of devices failed. Those are now marked as ->used==2 and
1359 * we ignore them and try again
1360 */
1361 if (!st && ident->st)
1362 st = ident->st;
1363 if (c->verbose>0)
1364 pr_err("looking for devices for %s\n",
1365 mddev ? mddev : "further assembly");
1366
1367 content = &info;
1368 if (st && c->force)
1369 st->ignore_hw_compat = 1;
1370 num_devs = select_devices(devlist, ident, &st, &content, c,
1371 inargv, auto_assem);
1372 if (num_devs < 0)
1373 return 1;
1374
1375 if (!st || !st->sb || !content)
1376 return 2;
1377
1378 /* We have a full set of devices - we now need to find the
1379 * array device.
1380 * However there is a risk that we are racing with "mdadm -I"
1381 * and the array is already partially assembled - we will have
1382 * rejected any devices already in this address.
1383 * So we take a lock on the map file - to prevent further races -
1384 * and look for the uuid in there. If found and the array is
1385 * active, we abort. If found and the array is not active
1386 * we commit to that md device and add all the contained devices
1387 * to our list. We flag them so that we don't try to re-add,
1388 * but can remove if they turn out to not be wanted.
1389 */
1390 if (map_lock(&map))
1391 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1392 if (c->update && strcmp(c->update,"uuid") == 0)
1393 mp = NULL;
1394 else
1395 mp = map_by_uuid(&map, content->uuid);
1396 if (mp) {
1397 struct mdinfo *dv;
1398 /* array already exists. */
1399 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1400 if (pre_exist->array.level != UnSet) {
1401 pr_err("Found some drive for an array that is already active: %s\n",
1402 mp->path);
1403 pr_err("giving up.\n");
1404 return 1;
1405 }
1406 for (dv = pre_exist->devs; dv; dv = dv->next) {
1407 /* We want to add this device to our list,
1408 * but it could already be there if "mdadm -I"
1409 * started *after* we checked for O_EXCL.
1410 * If we add it to the top of the list
1411 * it will be preferred over later copies.
1412 */
1413 struct mddev_dev *newdev;
1414 char *devname = map_dev(dv->disk.major,
1415 dv->disk.minor,
1416 0);
1417 if (!devname)
1418 continue;
1419 newdev = xmalloc(sizeof(*newdev));
1420 newdev->devname = devname;
1421 newdev->disposition = 'I';
1422 newdev->used = 1;
1423 newdev->next = devlist;
1424 devlist = newdev;
1425 num_devs++;
1426 }
1427 strcpy(chosen_name, mp->path);
1428 if (c->verbose > 0 || mddev == NULL ||
1429 strcmp(mddev, chosen_name) != 0)
1430 pr_err("Merging with already-assembled %s\n",
1431 chosen_name);
1432 mdfd = open_dev_excl(mp->devnm);
1433 } else {
1434 int trustworthy = FOREIGN;
1435 name = content->name;
1436 switch (st->ss->match_home(st, c->homehost)
1437 ?: st->ss->match_home(st, "any")) {
1438 case 1:
1439 trustworthy = LOCAL;
1440 name = strchr(content->name, ':');
1441 if (name)
1442 name++;
1443 else
1444 name = content->name;
1445 break;
1446 }
1447 if (!auto_assem)
1448 /* If the array is listed in mdadm.conf or on
1449 * command line, then we trust the name
1450 * even if the array doesn't look local
1451 */
1452 trustworthy = LOCAL;
1453
1454 if (name[0] == 0 &&
1455 content->array.level == LEVEL_CONTAINER) {
1456 name = content->text_version;
1457 trustworthy = METADATA;
1458 }
1459
1460 if (name[0] && trustworthy != LOCAL &&
1461 ! c->require_homehost &&
1462 conf_name_is_free(name))
1463 trustworthy = LOCAL;
1464
1465 if (trustworthy == LOCAL &&
1466 strchr(name, ':'))
1467 /* Ignore 'host:' prefix of name */
1468 name = strchr(name, ':')+1;
1469
1470 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1471 chosen_name);
1472 }
1473 if (mdfd < 0) {
1474 st->ss->free_super(st);
1475 if (auto_assem)
1476 goto try_again;
1477 return 1;
1478 }
1479 mddev = chosen_name;
1480 if (get_linux_version() < 2004000 ||
1481 md_get_version(mdfd) < 9000) {
1482 pr_err("Assemble requires Linux 2.4 or later, and\n"
1483 " md driver version 0.90.0 or later.\n"
1484 " Upgrade your kernel or try --build\n");
1485 close(mdfd);
1486 return 1;
1487 }
1488 if (pre_exist == NULL) {
1489 if (mddev_busy(fd2devnm(mdfd))) {
1490 pr_err("%s already active, cannot restart it!\n",
1491 mddev);
1492 for (tmpdev = devlist ;
1493 tmpdev && tmpdev->used != 1;
1494 tmpdev = tmpdev->next)
1495 ;
1496 if (tmpdev && auto_assem)
1497 pr_err("%s needed for %s...\n",
1498 mddev, tmpdev->devname);
1499 close(mdfd);
1500 mdfd = -3;
1501 st->ss->free_super(st);
1502 if (auto_assem)
1503 goto try_again;
1504 return 1;
1505 }
1506 /* just incase it was started but has no content */
1507 ioctl(mdfd, STOP_ARRAY, NULL);
1508 }
1509
1510 #ifndef MDASSEMBLE
1511 if (content != &info) {
1512 /* This is a member of a container. Try starting the array. */
1513 int err;
1514 err = assemble_container_content(st, mdfd, content, c,
1515 chosen_name, NULL);
1516 close(mdfd);
1517 return err;
1518 }
1519 #endif
1520 /* Ok, no bad inconsistancy, we can try updating etc */
1521 devices = xcalloc(num_devs, sizeof(*devices));
1522 devmap = xcalloc(num_devs, content->array.raid_disks);
1523 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1524 c, content, mdfd, mddev,
1525 &most_recent, &bestcnt, &best, inargv);
1526 if (devcnt < 0)
1527 return 1;
1528
1529 if (devcnt == 0) {
1530 pr_err("no devices found for %s\n",
1531 mddev);
1532 if (st)
1533 st->ss->free_super(st);
1534 close(mdfd);
1535 free(devices);
1536 free(devmap);
1537 return 1;
1538 }
1539
1540 if (c->update && strcmp(c->update, "byteorder")==0)
1541 st->minor_version = 90;
1542
1543 st->ss->getinfo_super(st, content, NULL);
1544 clean = content->array.state & 1;
1545
1546 /* now we have some devices that might be suitable.
1547 * I wonder how many
1548 */
1549 avail = xcalloc(content->array.raid_disks, 1);
1550 okcnt = 0;
1551 replcnt = 0;
1552 sparecnt=0;
1553 journalcnt=0;
1554 rebuilding_cnt=0;
1555 for (i=0; i< bestcnt; i++) {
1556 int j = best[i];
1557 int event_margin = 1; /* always allow a difference of '1'
1558 * like the kernel does
1559 */
1560 if (j < 0) continue;
1561 /* note: we ignore error flags in multipath arrays
1562 * as they don't make sense
1563 */
1564 if (content->array.level != LEVEL_MULTIPATH) {
1565 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1566 if (content->journal_device_required)
1567 journalcnt++;
1568 else /* unexpected journal, mark as faulty */
1569 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1570 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1571 if (!(devices[j].i.disk.state
1572 & (1<<MD_DISK_FAULTY))) {
1573 devices[j].uptodate = 1;
1574 sparecnt++;
1575 }
1576 continue;
1577 }
1578 }
1579 /* If this device thinks that 'most_recent' has failed, then
1580 * we must reject this device.
1581 */
1582 if (j != most_recent && !c->force &&
1583 content->array.raid_disks > 0 &&
1584 devices[most_recent].i.disk.raid_disk >= 0 &&
1585 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1586 if (c->verbose > -1)
1587 pr_err("ignoring %s as it reports %s as failed\n",
1588 devices[j].devname, devices[most_recent].devname);
1589 best[i] = -1;
1590 continue;
1591 }
1592 /* Require event counter to be same as, or just less than,
1593 * most recent. If it is bigger, it must be a stray spare and
1594 * should be ignored.
1595 */
1596 if (devices[j].i.events+event_margin >=
1597 devices[most_recent].i.events &&
1598 devices[j].i.events <=
1599 devices[most_recent].i.events
1600 ) {
1601 devices[j].uptodate = 1;
1602 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1603 journal_clean = 1;
1604 if (i < content->array.raid_disks * 2) {
1605 if (devices[j].i.recovery_start == MaxSector ||
1606 (content->reshape_active &&
1607 i >= content->array.raid_disks - content->delta_disks)) {
1608 if (!avail[i/2]) {
1609 okcnt++;
1610 avail[i/2]=1;
1611 } else
1612 replcnt++;
1613 } else
1614 rebuilding_cnt++;
1615 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1616 sparecnt++;
1617 }
1618 }
1619 free(devmap);
1620 if (c->force) {
1621 int force_ok = force_array(content, devices, best, bestcnt,
1622 avail, most_recent, st, c);
1623 okcnt += force_ok;
1624 if (force_ok)
1625 was_forced = 1;
1626 }
1627 /* Now we want to look at the superblock which the kernel will base things on
1628 * and compare the devices that we think are working with the devices that the
1629 * superblock thinks are working.
1630 * If there are differences and --force is given, then update this chosen
1631 * superblock.
1632 */
1633 chosen_drive = -1;
1634 st->ss->free_super(st);
1635 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1636 int j = best[i];
1637 int fd;
1638
1639 if (j<0)
1640 continue;
1641 if (!devices[j].uptodate)
1642 continue;
1643 if (devices[j].i.events < devices[most_recent].i.events)
1644 continue;
1645 chosen_drive = j;
1646 if ((fd=dev_open(devices[j].devname,
1647 devices[j].included ? O_RDONLY
1648 : (O_RDONLY|O_EXCL)))< 0) {
1649 pr_err("Cannot open %s: %s\n",
1650 devices[j].devname, strerror(errno));
1651 close(mdfd);
1652 free(devices);
1653 return 1;
1654 }
1655 if (st->ss->load_super(st,fd, NULL)) {
1656 close(fd);
1657 pr_err("RAID superblock has disappeared from %s\n",
1658 devices[j].devname);
1659 close(mdfd);
1660 free(devices);
1661 return 1;
1662 }
1663 close(fd);
1664 }
1665 if (st->sb == NULL) {
1666 pr_err("No suitable drives found for %s\n", mddev);
1667 close(mdfd);
1668 free(devices);
1669 return 1;
1670 }
1671 st->ss->getinfo_super(st, content, NULL);
1672 #ifndef MDASSEMBLE
1673 sysfs_init(content, mdfd, NULL);
1674 #endif
1675 /* after reload context, store journal_clean in context */
1676 content->journal_clean = journal_clean;
1677 for (i=0; i<bestcnt; i++) {
1678 int j = best[i];
1679 unsigned int desired_state;
1680
1681 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1682 desired_state = (1<<MD_DISK_JOURNAL);
1683 else if (i >= content->array.raid_disks * 2)
1684 desired_state = 0;
1685 else if (i & 1)
1686 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1687 else
1688 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1689
1690 if (j<0)
1691 continue;
1692 if (!devices[j].uptodate)
1693 continue;
1694
1695 devices[j].i.disk.state = desired_state;
1696 if (!(devices[j].i.array.state & 1))
1697 clean = 0;
1698
1699 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1700 c->verbose, 0, NULL)) {
1701 if (c->force) {
1702 if (c->verbose >= 0)
1703 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1704 j, mddev, devices[j].devname);
1705 change = 1;
1706 } else {
1707 if (c->verbose >= -1)
1708 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1709 i, mddev, devices[j].devname);
1710 }
1711 }
1712 #if 0
1713 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1714 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1715 i, mddev);
1716 }
1717 #endif
1718 }
1719 if (c->force && !clean &&
1720 !enough(content->array.level, content->array.raid_disks,
1721 content->array.layout, clean,
1722 avail)) {
1723 change += st->ss->update_super(st, content, "force-array",
1724 devices[chosen_drive].devname, c->verbose,
1725 0, NULL);
1726 was_forced = 1;
1727 clean = 1;
1728 }
1729
1730 if (change) {
1731 int fd;
1732 fd = dev_open(devices[chosen_drive].devname,
1733 devices[chosen_drive].included ?
1734 O_RDWR : (O_RDWR|O_EXCL));
1735 if (fd < 0) {
1736 pr_err("Could not open %s for write - cannot Assemble array.\n",
1737 devices[chosen_drive].devname);
1738 close(mdfd);
1739 free(devices);
1740 return 1;
1741 }
1742 if (st->ss->store_super(st, fd)) {
1743 close(fd);
1744 pr_err("Could not re-write superblock on %s\n",
1745 devices[chosen_drive].devname);
1746 close(mdfd);
1747 free(devices);
1748 return 1;
1749 }
1750 if (c->verbose >= 0)
1751 pr_err("Marking array %s as 'clean'\n",
1752 mddev);
1753 close(fd);
1754 }
1755
1756 /* If we are in the middle of a reshape we may need to restore saved data
1757 * that was moved aside due to the reshape overwriting live data
1758 * The code of doing this lives in Grow.c
1759 */
1760 #ifndef MDASSEMBLE
1761 if (content->reshape_active &&
1762 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1763 int err = 0;
1764 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1765 if (c->verbose > 0)
1766 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1767 chosen_name);
1768 if (!c->backup_file)
1769 c->backup_file = locate_backup(content->sys_name);
1770 enable_fds(bestcnt/2);
1771 for (i = 0; i < bestcnt/2; i++) {
1772 int j = best[i*2];
1773 if (j >= 0) {
1774 fdlist[i] = dev_open(devices[j].devname,
1775 devices[j].included
1776 ? O_RDWR : (O_RDWR|O_EXCL));
1777 if (fdlist[i] < 0) {
1778 pr_err("Could not open %s for write - cannot Assemble array.\n",
1779 devices[j].devname);
1780 err = 1;
1781 break;
1782 }
1783 } else
1784 fdlist[i] = -1;
1785 }
1786 if (!err) {
1787 if (st->ss->external && st->ss->recover_backup)
1788 err = st->ss->recover_backup(st, content);
1789 else
1790 err = Grow_restart(st, content, fdlist, bestcnt/2,
1791 c->backup_file, c->verbose > 0);
1792 if (err && c->invalid_backup) {
1793 if (c->verbose > 0)
1794 pr_err("continuing without restoring backup\n");
1795 err = 0;
1796 }
1797 }
1798 while (i>0) {
1799 i--;
1800 if (fdlist[i]>=0) close(fdlist[i]);
1801 }
1802 free(fdlist);
1803 if (err) {
1804 pr_err("Failed to restore critical section for reshape, sorry.\n");
1805 if (c->backup_file == NULL)
1806 cont_err("Possibly you needed to specify the --backup-file\n");
1807 close(mdfd);
1808 free(devices);
1809 return err;
1810 }
1811 }
1812 #endif
1813
1814 /* Almost ready to actually *do* something */
1815 /* First, fill in the map, so that udev can find our name
1816 * as soon as we become active.
1817 */
1818 if (c->update && strcmp(c->update, "metadata")==0) {
1819 content->array.major_version = 1;
1820 content->array.minor_version = 0;
1821 strcpy(content->text_version, "1.0");
1822 }
1823
1824 map_update(&map, fd2devnm(mdfd), content->text_version,
1825 content->uuid, chosen_name);
1826
1827 rv = start_array(mdfd, mddev, content,
1828 st, ident, best, bestcnt,
1829 chosen_drive, devices, okcnt, sparecnt,
1830 rebuilding_cnt, journalcnt,
1831 c,
1832 clean, avail, start_partial_ok,
1833 pre_exist != NULL,
1834 was_forced);
1835 if (rv == 1 && !pre_exist)
1836 ioctl(mdfd, STOP_ARRAY, NULL);
1837 free(devices);
1838 map_unlock(&map);
1839 if (rv == 0) {
1840 wait_for(chosen_name, mdfd);
1841 close(mdfd);
1842 if (auto_assem) {
1843 int usecs = 1;
1844 /* There is a nasty race with 'mdadm --monitor'.
1845 * If it opens this device before we close it,
1846 * it gets an incomplete open on which IO
1847 * doesn't work and the capacity is
1848 * wrong.
1849 * If we reopen (to check for layered devices)
1850 * before --monitor closes, we loose.
1851 *
1852 * So: wait upto 1 second for there to be
1853 * a non-zero capacity.
1854 */
1855 while (usecs < 1000) {
1856 mdfd = open(mddev, O_RDONLY);
1857 if (mdfd >= 0) {
1858 unsigned long long size;
1859 if (get_dev_size(mdfd, NULL, &size) &&
1860 size > 0)
1861 break;
1862 close(mdfd);
1863 }
1864 usleep(usecs);
1865 usecs <<= 1;
1866 }
1867 }
1868 } else
1869 close(mdfd);
1870
1871 /* '2' means 'OK, but not started yet' */
1872 return rv == 2 ? 0 : rv;
1873 }
1874
1875 #ifndef MDASSEMBLE
1876 int assemble_container_content(struct supertype *st, int mdfd,
1877 struct mdinfo *content, struct context *c,
1878 char *chosen_name, int *result)
1879 {
1880 struct mdinfo *dev, *sra, *dev2;
1881 int working = 0, preexist = 0;
1882 int expansion = 0;
1883 struct map_ent *map = NULL;
1884 int old_raid_disks;
1885 int start_reshape;
1886 char *avail = NULL;
1887 int err;
1888
1889 sysfs_init(content, mdfd, NULL);
1890
1891 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1892 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1893 if (content->array.major_version == -1 &&
1894 content->array.minor_version == -2 &&
1895 c->readonly &&
1896 content->text_version[0] == '/')
1897 content->text_version[0] = '-';
1898 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1899 if (sra)
1900 sysfs_free(sra);
1901 return 1;
1902 }
1903 }
1904
1905 /* There are two types of reshape: container wide or sub-array specific
1906 * Check if metadata requests blocking container wide reshapes
1907 */
1908 start_reshape = (content->reshape_active &&
1909 !((content->reshape_active == CONTAINER_RESHAPE) &&
1910 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1911
1912 /* Block subarray here if it is under reshape now
1913 * Do not allow for any changes in this array
1914 */
1915 if (st->ss->external && content->recovery_blocked && start_reshape)
1916 block_subarray(content);
1917
1918 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1919 for (dev = content->devs; dev; dev = dev->next)
1920 if (dev2->disk.major == dev->disk.major &&
1921 dev2->disk.minor == dev->disk.minor)
1922 break;
1923 if (dev)
1924 continue;
1925 /* Don't want this one any more */
1926 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1927 errno == EBUSY) {
1928 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1929 sysfs_free(sra);
1930 return 1;
1931 }
1932 sysfs_set_str(sra, dev2, "state", "remove");
1933 }
1934 old_raid_disks = content->array.raid_disks - content->delta_disks;
1935 avail = xcalloc(content->array.raid_disks, 1);
1936 for (dev = content->devs; dev; dev = dev->next) {
1937 if (dev->disk.raid_disk >= 0)
1938 avail[dev->disk.raid_disk] = 1;
1939 if (sysfs_add_disk(content, dev, 1) == 0) {
1940 if (dev->disk.raid_disk >= old_raid_disks &&
1941 content->reshape_active)
1942 expansion++;
1943 else
1944 working++;
1945 } else if (errno == EEXIST)
1946 preexist++;
1947 }
1948 sysfs_free(sra);
1949 if (working + expansion == 0 && c->runstop <= 0) {
1950 free(avail);
1951 return 1;/* Nothing new, don't try to start */
1952 }
1953 map_update(&map, fd2devnm(mdfd),
1954 content->text_version,
1955 content->uuid, chosen_name);
1956
1957
1958 if (enough(content->array.level, content->array.raid_disks,
1959 content->array.layout, content->array.state & 1, avail) == 0) {
1960 if (c->export && result)
1961 *result |= INCR_NO;
1962 else if (c->verbose >= 0) {
1963 pr_err("%s assembled with %d device%s",
1964 chosen_name, preexist + working,
1965 preexist + working == 1 ? "":"s");
1966 if (preexist)
1967 fprintf(stderr, " (%d new)", working);
1968 fprintf(stderr, " but not started\n");
1969 }
1970 free(avail);
1971 return 1;
1972 }
1973 free(avail);
1974
1975 if (c->runstop <= 0 &&
1976 (working + preexist + expansion) <
1977 content->array.working_disks) {
1978 if (c->export && result)
1979 *result |= INCR_UNSAFE;
1980 else if (c->verbose >= 0) {
1981 pr_err("%s assembled with %d device%s",
1982 chosen_name, preexist + working,
1983 preexist + working == 1 ? "":"s");
1984 if (preexist)
1985 fprintf(stderr, " (%d new)", working);
1986 fprintf(stderr, " but not safe to start\n");
1987 }
1988 return 1;
1989 }
1990
1991
1992 if (start_reshape) {
1993 int spare = content->array.raid_disks + expansion;
1994 if (restore_backup(st, content,
1995 working,
1996 spare, &c->backup_file, c->verbose) == 1)
1997 return 1;
1998
1999 err = sysfs_set_str(content, NULL,
2000 "array_state", "readonly");
2001 if (err)
2002 return 1;
2003
2004 if (st->ss->external) {
2005 if (!mdmon_running(st->container_devnm))
2006 start_mdmon(st->container_devnm);
2007 ping_monitor(st->container_devnm);
2008 if (mdmon_running(st->container_devnm) &&
2009 st->update_tail == NULL)
2010 st->update_tail = &st->updates;
2011 }
2012
2013 err = Grow_continue(mdfd, st, content, c->backup_file,
2014 0, c->freeze_reshape);
2015 } else switch(content->array.level) {
2016 case LEVEL_LINEAR:
2017 case LEVEL_MULTIPATH:
2018 case 0:
2019 err = sysfs_set_str(content, NULL, "array_state",
2020 c->readonly ? "readonly" : "active");
2021 break;
2022 default:
2023 err = sysfs_set_str(content, NULL, "array_state",
2024 "readonly");
2025 /* start mdmon if needed. */
2026 if (!err) {
2027 if (!mdmon_running(st->container_devnm))
2028 start_mdmon(st->container_devnm);
2029 ping_monitor(st->container_devnm);
2030 }
2031 break;
2032 }
2033 if (!err)
2034 sysfs_set_safemode(content, content->safe_mode_delay);
2035
2036 /* Block subarray here if it is not reshaped now
2037 * It has be blocked a little later to allow mdmon to switch in
2038 * in to R/W state
2039 */
2040 if (st->ss->external && content->recovery_blocked &&
2041 !start_reshape)
2042 block_subarray(content);
2043
2044 if (c->export && result) {
2045 if (err)
2046 *result |= INCR_NO;
2047 else
2048 *result |= INCR_YES;
2049 } else if (c->verbose >= 0) {
2050 if (err)
2051 pr_err("array %s now has %d device%s",
2052 chosen_name, working + preexist,
2053 working + preexist == 1 ? "":"s");
2054 else
2055 pr_err("Started %s with %d device%s",
2056 chosen_name, working + preexist,
2057 working + preexist == 1 ? "":"s");
2058 if (preexist)
2059 fprintf(stderr, " (%d new)", working);
2060 if (expansion)
2061 fprintf(stderr, " ( + %d for expansion)",
2062 expansion);
2063 fprintf(stderr, "\n");
2064 }
2065 if (!err)
2066 wait_for(chosen_name, mdfd);
2067 return err;
2068 /* FIXME should have an O_EXCL and wait for read-auto */
2069 }
2070 #endif
Unnamed repository; edit this file 'description' to name the repository.