]> 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
mdadm: change the num of cluster node
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(0, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 static int ident_matches(struct mddev_ident *ident,
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78 {
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
82 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
83 if (devname)
84 pr_err("%s has wrong uuid.\n", devname);
85 return 0;
86 }
87 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
88 name_matches(content->name, ident->name, homehost)==0) {
89 if (devname)
90 pr_err("%s has wrong name.\n", devname);
91 return 0;
92 }
93 if (ident->super_minor != UnSet &&
94 ident->super_minor != content->array.md_minor) {
95 if (devname)
96 pr_err("%s has wrong super-minor.\n",
97 devname);
98 return 0;
99 }
100 if (ident->level != UnSet &&
101 ident->level != content->array.level) {
102 if (devname)
103 pr_err("%s has wrong raid level.\n",
104 devname);
105 return 0;
106 }
107 if (ident->raid_disks != UnSet &&
108 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
109 ident->raid_disks!= content->array.raid_disks) {
110 if (devname)
111 pr_err("%s requires wrong number of drives.\n",
112 devname);
113 return 0;
114 }
115 if (ident->member && ident->member[0]) {
116 /* content->text_version must match */
117 char *s = strchr(content->text_version+1, '/');
118 if (s == NULL) {
119 if (devname)
120 pr_err("%s is not a container and one is required.\n",
121 devname);
122 return 0;
123 } else if (strcmp(ident->member, s+1) != 0) {
124 if (devname)
125 pr_err("skipping wrong member %s is %s\n",
126 content->text_version, devname);
127 return 0;
128 }
129 }
130 return 1;
131 }
132
133 static int select_devices(struct mddev_dev *devlist,
134 struct mddev_ident *ident,
135 struct supertype **stp,
136 struct mdinfo **contentp,
137 struct context *c,
138 int inargv, int auto_assem)
139 {
140 struct mddev_dev *tmpdev;
141 int num_devs;
142 struct supertype *st = *stp;
143 struct mdinfo *content = NULL;
144 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
145 struct domainlist *domains = NULL;
146
147 tmpdev = devlist; num_devs = 0;
148 while (tmpdev) {
149 if (tmpdev->used)
150 tmpdev->used = 2;
151 else
152 num_devs++;
153 tmpdev->disposition = 0;
154 tmpdev = tmpdev->next;
155 }
156
157 /* first walk the list of devices to find a consistent set
158 * that match the criterea, if that is possible.
159 * We flag the ones we like with 'used'.
160 */
161 for (tmpdev = devlist;
162 tmpdev;
163 tmpdev = tmpdev ? tmpdev->next : NULL) {
164 char *devname = tmpdev->devname;
165 int dfd;
166 struct stat stb;
167 struct supertype *tst;
168 struct dev_policy *pol = NULL;
169 int found_container = 0;
170
171 if (tmpdev->used > 1)
172 continue;
173
174 if (ident->container) {
175 if (ident->container[0] == '/' &&
176 !same_dev(ident->container, devname)) {
177 if (report_mismatch)
178 pr_err("%s is not the container required (%s)\n",
179 devname, ident->container);
180 continue;
181 }
182 } else if (ident->devices &&
183 !match_oneof(ident->devices, devname)) {
184 /* Note that we ignore the "device=" identifier if a
185 * "container=" is given. Checking both is unnecessarily
186 * complicated.
187 */
188 if (report_mismatch)
189 pr_err("%s is not one of %s\n", devname, ident->devices);
190 continue;
191 }
192
193 tst = dup_super(st);
194
195 dfd = dev_open(devname, O_RDONLY);
196 if (dfd < 0) {
197 if (report_mismatch)
198 pr_err("cannot open device %s: %s\n",
199 devname, strerror(errno));
200 tmpdev->used = 2;
201 } else if (fstat(dfd, &stb)< 0) {
202 /* Impossible! */
203 pr_err("fstat failed for %s: %s\n",
204 devname, strerror(errno));
205 tmpdev->used = 2;
206 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
207 pr_err("%s is not a block device.\n",
208 devname);
209 tmpdev->used = 2;
210 } else if (must_be_container(dfd)) {
211 if (st) {
212 /* already found some components, this cannot
213 * be another one.
214 */
215 if (report_mismatch)
216 pr_err("%s is a container, but we are looking for components\n",
217 devname);
218 tmpdev->used = 2;
219 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
220 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
221 if (report_mismatch)
222 pr_err("not a recognisable container: %s\n",
223 devname);
224 tmpdev->used = 2;
225 #endif
226 } else if (!tst->ss->load_container
227 || tst->ss->load_container(tst, dfd, NULL)) {
228 if (report_mismatch)
229 pr_err("no correct container type: %s\n",
230 devname);
231 tmpdev->used = 2;
232 } else if (auto_assem &&
233 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
234 tst->ss->match_home(tst, c->homehost) == 1)) {
235 if (report_mismatch)
236 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
237 devname, tst->ss->name);
238 tmpdev->used = 2;
239 } else
240 found_container = 1;
241 } else {
242 if (!tst && (tst = guess_super(dfd)) == NULL) {
243 if (report_mismatch)
244 pr_err("no recogniseable superblock on %s\n",
245 devname);
246 tmpdev->used = 2;
247 } else if (tst->ss->load_super(tst,dfd, NULL)) {
248 if (report_mismatch)
249 pr_err("no RAID superblock on %s\n",
250 devname);
251 tmpdev->used = 2;
252 } else if (tst->ss->compare_super == NULL) {
253 if (report_mismatch)
254 pr_err("Cannot assemble %s metadata on %s\n",
255 tst->ss->name, devname);
256 tmpdev->used = 2;
257 } else if (auto_assem && st == NULL &&
258 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
259 tst->ss->match_home(tst, c->homehost) == 1)) {
260 if (report_mismatch)
261 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
262 devname, tst->ss->name);
263 tmpdev->used = 2;
264 }
265 }
266 if (dfd >= 0) close(dfd);
267 if (tmpdev->used == 2) {
268 if (auto_assem || !inargv)
269 /* Ignore unrecognised devices during auto-assembly */
270 goto loop;
271 if (ident->uuid_set || ident->name[0] ||
272 ident->super_minor != UnSet)
273 /* Ignore unrecognised device if looking for
274 * specific array */
275 goto loop;
276
277 pr_err("%s has no superblock - assembly aborted\n",
278 devname);
279 if (st)
280 st->ss->free_super(st);
281 dev_policy_free(pol);
282 domain_free(domains);
283 return -1;
284 }
285
286 if (found_container) {
287 /* tmpdev is a container. We need to be either
288 * looking for a member, or auto-assembling
289 */
290 /* should be safe to try an exclusive open now, we
291 * have rejected anything that some other mdadm might
292 * be looking at
293 */
294 dfd = dev_open(devname, O_RDONLY | O_EXCL);
295 if (dfd < 0) {
296 if (report_mismatch)
297 pr_err("%s is busy - skipping\n", devname);
298 goto loop;
299 }
300 close(dfd);
301
302 if (ident->container && ident->container[0] != '/') {
303 /* we have a uuid */
304 int uuid[4];
305
306 content = *contentp;
307 tst->ss->getinfo_super(tst, content, NULL);
308
309 if (!parse_uuid(ident->container, uuid) ||
310 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
311 if (report_mismatch)
312 pr_err("%s has wrong UUID to be required container\n",
313 devname);
314 goto loop;
315 }
316 }
317 /* It is worth looking inside this container.
318 */
319 if (c->verbose > 0)
320 pr_err("looking in container %s\n",
321 devname);
322
323 for (content = tst->ss->container_content(tst, NULL);
324 content;
325 content = content->next) {
326
327 if (!ident_matches(ident, content, tst,
328 c->homehost, c->update,
329 report_mismatch ? devname : NULL))
330 /* message already printed */;
331 else if (is_member_busy(content->text_version)) {
332 if (report_mismatch)
333 pr_err("member %s in %s is already assembled\n",
334 content->text_version,
335 devname);
336 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
337 /* do not assemble arrays with unsupported configurations */
338 pr_err("Cannot activate member %s in %s.\n",
339 content->text_version,
340 devname);
341 } else
342 break;
343 }
344 if (!content) {
345 tmpdev->used = 2;
346 goto loop; /* empty container */
347 }
348
349 st = tst; tst = NULL;
350 if (!auto_assem && inargv && tmpdev->next != NULL) {
351 pr_err("%s is a container, but is not only device given: confused and aborting\n",
352 devname);
353 st->ss->free_super(st);
354 dev_policy_free(pol);
355 domain_free(domains);
356 return -1;
357 }
358 if (c->verbose > 0)
359 pr_err("found match on member %s in %s\n",
360 content->text_version, devname);
361
362 /* make sure we finished the loop */
363 tmpdev = NULL;
364 goto loop;
365 } else {
366 content = *contentp;
367 tst->ss->getinfo_super(tst, content, NULL);
368
369 if (!ident_matches(ident, content, tst,
370 c->homehost, c->update,
371 report_mismatch ? devname : NULL))
372 goto loop;
373
374 if (auto_assem) {
375 /* Never auto-assemble things that conflict
376 * with mdadm.conf in some way
377 */
378 struct mddev_ident *match;
379 int rv = 0;
380
381 match = conf_match(tst, content, devname,
382 report_mismatch ? c->verbose : -1,
383 &rv);
384 if (!match && rv == 2)
385 goto loop;
386 if (match && match->devname &&
387 strcasecmp(match->devname, "<ignore>") == 0) {
388 if (report_mismatch)
389 pr_err("%s is a member of an explicitly ignored array\n",
390 devname);
391 goto loop;
392 }
393 if (match && !ident_matches(match, content, tst,
394 c->homehost, c->update,
395 report_mismatch ? devname : NULL))
396 /* Array exists in mdadm.conf but some
397 * details don't match, so reject it
398 */
399 goto loop;
400 }
401
402 /* should be safe to try an exclusive open now, we
403 * have rejected anything that some other mdadm might
404 * be looking at
405 */
406 dfd = dev_open(devname, O_RDONLY | O_EXCL);
407 if (dfd < 0) {
408 if (report_mismatch)
409 pr_err("%s is busy - skipping\n", devname);
410 goto loop;
411 }
412 close(dfd);
413
414 if (st == NULL)
415 st = dup_super(tst);
416 if (st->minor_version == -1)
417 st->minor_version = tst->minor_version;
418
419 if (memcmp(content->uuid, uuid_zero,
420 sizeof(int[4])) == 0) {
421 /* this is a floating spare. It cannot define
422 * an array unless there are no more arrays of
423 * this type to be found. It can be included
424 * in an array of this type though.
425 */
426 tmpdev->used = 3;
427 goto loop;
428 }
429
430 if (st->ss != tst->ss ||
431 st->minor_version != tst->minor_version ||
432 st->ss->compare_super(st, tst) != 0) {
433 /* Some mismatch. If exactly one array matches this host,
434 * we can resolve on that one.
435 * Or, if we are auto assembling, we just ignore the second
436 * for now.
437 */
438 if (auto_assem)
439 goto loop;
440 if (c->homehost) {
441 int first = st->ss->match_home(st, c->homehost);
442 int last = tst->ss->match_home(tst, c->homehost);
443 if (first != last &&
444 (first == 1 || last == 1)) {
445 /* We can do something */
446 if (first) {/* just ignore this one */
447 if (report_mismatch)
448 pr_err("%s misses out due to wrong homehost\n",
449 devname);
450 goto loop;
451 } else { /* reject all those sofar */
452 struct mddev_dev *td;
453 if (report_mismatch)
454 pr_err("%s overrides previous devices due to good homehost\n",
455 devname);
456 for (td=devlist; td != tmpdev; td=td->next)
457 if (td->used == 1)
458 td->used = 0;
459 tmpdev->used = 1;
460 goto loop;
461 }
462 }
463 }
464 pr_err("superblock on %s doesn't match others - assembly aborted\n",
465 devname);
466 tst->ss->free_super(tst);
467 st->ss->free_super(st);
468 dev_policy_free(pol);
469 domain_free(domains);
470 return -1;
471 }
472 tmpdev->used = 1;
473 }
474 loop:
475 /* Collect domain information from members only */
476 if (tmpdev && tmpdev->used == 1) {
477 if (!pol)
478 pol = devid_policy(stb.st_rdev);
479 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
480 }
481 dev_policy_free(pol);
482 pol = NULL;
483 if (tst)
484 tst->ss->free_super(tst);
485 }
486
487 /* Check if we found some imsm spares but no members */
488 if ((auto_assem ||
489 (ident->uuid_set &&
490 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
491 (!st || !st->sb))
492 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
493 if (tmpdev->used != 3)
494 continue;
495 tmpdev->used = 1;
496 content = *contentp;
497
498 if (!st->sb) {
499 /* we need sb from one of the spares */
500 int dfd = dev_open(tmpdev->devname, O_RDONLY);
501 if (dfd < 0 ||
502 st->ss->load_super(st, dfd, NULL))
503 tmpdev->used = 2;
504 if (dfd > 0)
505 close(dfd);
506 }
507 }
508
509 /* Now reject spares that don't match domains of identified members */
510 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
511 struct stat stb;
512 if (tmpdev->used != 3)
513 continue;
514 if (stat(tmpdev->devname, &stb)< 0) {
515 pr_err("fstat failed for %s: %s\n",
516 tmpdev->devname, strerror(errno));
517 tmpdev->used = 2;
518 } else {
519 struct dev_policy *pol = devid_policy(stb.st_rdev);
520 int dt = domain_test(domains, pol, NULL);
521 if (inargv && dt != 0)
522 /* take this spare as domains match
523 * if there are any */
524 tmpdev->used = 1;
525 else if (!inargv && dt == 1)
526 /* device wasn't explicitly listed, so need
527 * explicit domain match - which we have */
528 tmpdev->used = 1;
529 else
530 /* if domains don't match mark as unused */
531 tmpdev->used = 0;
532 dev_policy_free(pol);
533 }
534 }
535 domain_free(domains);
536 *stp = st;
537 if (st && st->sb && content == *contentp)
538 st->ss->getinfo_super(st, content, NULL);
539 *contentp = content;
540
541 return num_devs;
542 }
543
544 struct devs {
545 char *devname;
546 int uptodate; /* set once we decide that this device is as
547 * recent as everything else in the array.
548 */
549 int included; /* set if the device is already in the array
550 * due to a previous '-I'
551 */
552 struct mdinfo i;
553 };
554
555 static int load_devices(struct devs *devices, char *devmap,
556 struct mddev_ident *ident, struct supertype **stp,
557 struct mddev_dev *devlist, struct context *c,
558 struct mdinfo *content,
559 int mdfd, char *mddev,
560 int *most_recentp, int *bestcntp, int **bestp,
561 int inargv)
562 {
563 struct mddev_dev *tmpdev;
564 int devcnt = 0;
565 int nextspare = 0;
566 #ifndef MDASSEMBLE
567 int bitmap_done = 0;
568 #endif
569 int most_recent = -1;
570 int bestcnt = 0;
571 int *best = *bestp;
572 struct supertype *st = *stp;
573
574 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
575 char *devname = tmpdev->devname;
576 struct stat stb;
577 struct supertype *tst;
578 int i;
579 int dfd;
580
581 if (tmpdev->used != 1)
582 continue;
583 /* looks like a good enough match to update the super block if needed */
584 #ifndef MDASSEMBLE
585 if (c->update) {
586 /* prepare useful information in info structures */
587 struct stat stb2;
588 int err;
589 fstat(mdfd, &stb2);
590
591 if (strcmp(c->update, "uuid")==0 &&
592 !ident->uuid_set) {
593 int rfd;
594 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
595 read(rfd, ident->uuid, 16) != 16) {
596 *(__u32*)(ident->uuid) = random();
597 *(__u32*)(ident->uuid+1) = random();
598 *(__u32*)(ident->uuid+2) = random();
599 *(__u32*)(ident->uuid+3) = random();
600 }
601 if (rfd >= 0) close(rfd);
602 }
603 dfd = dev_open(devname,
604 tmpdev->disposition == 'I'
605 ? O_RDWR : (O_RDWR|O_EXCL));
606
607 tst = dup_super(st);
608 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
609 pr_err("cannot re-read metadata from %s - aborting\n",
610 devname);
611 if (dfd >= 0)
612 close(dfd);
613 close(mdfd);
614 free(devices);
615 free(devmap);
616 tst->ss->free_super(tst);
617 free(tst);
618 *stp = st;
619 return -1;
620 }
621 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
622
623 memcpy(content->uuid, ident->uuid, 16);
624 strcpy(content->name, ident->name);
625 content->array.md_minor = minor(stb2.st_rdev);
626
627 if (strcmp(c->update, "byteorder") == 0)
628 err = 0;
629 else if (strcmp(c->update, "home-cluster") == 0) {
630 tst->cluster_name = c->homecluster;
631 tst->ss->write_bitmap(tst, dfd, NameUpdate);
632 } else if (strcmp(c->update, "nodes") == 0) {
633 tst->nodes = c->nodes;
634 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
635 } else
636 err = tst->ss->update_super(tst, content, c->update,
637 devname, c->verbose,
638 ident->uuid_set,
639 c->homehost);
640 if (err < 0) {
641 if (err == -1)
642 pr_err("--update=%s not understood for %s metadata\n",
643 c->update, tst->ss->name);
644 tst->ss->free_super(tst);
645 free(tst);
646 close(mdfd);
647 close(dfd);
648 free(devices);
649 free(devmap);
650 *stp = st;
651 return -1;
652 }
653 if (strcmp(c->update, "uuid")==0 &&
654 !ident->uuid_set) {
655 ident->uuid_set = 1;
656 memcpy(ident->uuid, content->uuid, 16);
657 }
658 if (tst->ss->store_super(tst, dfd))
659 pr_err("Could not re-write superblock on %s.\n",
660 devname);
661
662 if (strcmp(c->update, "uuid")==0 &&
663 ident->bitmap_fd >= 0 && !bitmap_done) {
664 if (bitmap_update_uuid(ident->bitmap_fd,
665 content->uuid,
666 tst->ss->swapuuid) != 0)
667 pr_err("Could not update uuid on external bitmap.\n");
668 else
669 bitmap_done = 1;
670 }
671 } else
672 #endif
673 {
674 dfd = dev_open(devname,
675 tmpdev->disposition == 'I'
676 ? O_RDWR : (O_RDWR|O_EXCL));
677 tst = dup_super(st);
678
679 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
680 pr_err("cannot re-read metadata from %s - aborting\n",
681 devname);
682 if (dfd >= 0)
683 close(dfd);
684 close(mdfd);
685 free(devices);
686 free(devmap);
687 tst->ss->free_super(tst);
688 free(tst);
689 *stp = st;
690 return -1;
691 }
692 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
693 }
694
695 fstat(dfd, &stb);
696 close(dfd);
697
698 if (c->verbose > 0)
699 pr_err("%s is identified as a member of %s, slot %d%s.\n",
700 devname, mddev, content->disk.raid_disk,
701 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
702 devices[devcnt].devname = devname;
703 devices[devcnt].uptodate = 0;
704 devices[devcnt].included = (tmpdev->disposition == 'I');
705 devices[devcnt].i = *content;
706 devices[devcnt].i.disk.major = major(stb.st_rdev);
707 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
708
709 if (devices[devcnt].i.disk.state == 6) {
710 if (most_recent < 0 ||
711 devices[devcnt].i.events
712 > devices[most_recent].i.events) {
713 struct supertype *tmp = tst;
714 tst = st;
715 st = tmp;
716 most_recent = devcnt;
717 }
718 }
719 tst->ss->free_super(tst);
720 free(tst);
721
722 if (content->array.level == LEVEL_MULTIPATH)
723 /* with multipath, the raid_disk from the superblock is meaningless */
724 i = devcnt;
725 else
726 i = devices[devcnt].i.disk.raid_disk;
727 if (i+1 == 0) {
728 if (nextspare < content->array.raid_disks*2)
729 nextspare = content->array.raid_disks*2;
730 i = nextspare++;
731 } else {
732 /* i is raid_disk - double it so there is room for
733 * replacements */
734 i *= 2;
735 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
736 i++;
737 if (i >= content->array.raid_disks*2 &&
738 i >= nextspare)
739 nextspare = i+1;
740 }
741 if (i < 10000) {
742 if (i >= bestcnt) {
743 int newbestcnt = i+10;
744 int *newbest = xmalloc(sizeof(int)*newbestcnt);
745 int c;
746 for (c=0; c < newbestcnt; c++)
747 if (c < bestcnt)
748 newbest[c] = best[c];
749 else
750 newbest[c] = -1;
751 if (best)free(best);
752 best = newbest;
753 bestcnt = newbestcnt;
754 }
755 if (best[i] >=0 &&
756 devices[best[i]].i.events
757 == devices[devcnt].i.events
758 && (devices[best[i]].i.disk.minor
759 != devices[devcnt].i.disk.minor)
760 && st->ss == &super0
761 && content->array.level != LEVEL_MULTIPATH) {
762 /* two different devices with identical superblock.
763 * Could be a mis-detection caused by overlapping
764 * partitions. fail-safe.
765 */
766 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
767 " If they are really different, please --zero the superblock on one\n"
768 " If they are the same or overlap, please remove one from %s.\n",
769 devices[best[i]].devname, devname,
770 inargv ? "the list" :
771 "the\n DEVICE list in mdadm.conf"
772 );
773 close(mdfd);
774 free(devices);
775 free(devmap);
776 *stp = st;
777 return -1;
778 }
779 if (best[i] == -1
780 || (devices[best[i]].i.events
781 < devices[devcnt].i.events))
782 best[i] = devcnt;
783 }
784 devcnt++;
785 }
786 if (most_recent >= 0)
787 *most_recentp = most_recent;
788 *bestcntp = bestcnt;
789 *bestp = best;
790 *stp = st;
791 return devcnt;
792 }
793
794 static int force_array(struct mdinfo *content,
795 struct devs *devices,
796 int *best, int bestcnt, char *avail,
797 int most_recent,
798 struct supertype *st,
799 struct context *c)
800 {
801 int okcnt = 0;
802 while (!enough(content->array.level, content->array.raid_disks,
803 content->array.layout, 1,
804 avail)
805 ||
806 (content->reshape_active && content->delta_disks > 0 &&
807 !enough(content->array.level, (content->array.raid_disks
808 - content->delta_disks),
809 content->new_layout, 1,
810 avail)
811 )) {
812 /* Choose the newest best drive which is
813 * not up-to-date, update the superblock
814 * and add it.
815 */
816 int fd;
817 struct supertype *tst;
818 unsigned long long current_events;
819 int chosen_drive = -1;
820 int i;
821
822 for (i = 0;
823 i < content->array.raid_disks * 2 && i < bestcnt;
824 i += 2) {
825 int j = best[i];
826 if (j < 0)
827 continue;
828 if (devices[j].uptodate)
829 continue;
830 if (devices[j].i.recovery_start != MaxSector) {
831 int delta;
832 if (!devices[j].i.reshape_active ||
833 devices[j].i.delta_disks <= 0)
834 continue;
835 /* When increasing number of devices, an
836 * added device also appears to be
837 * recovering. It is safe to include it
838 * as long as it won't be a source of
839 * data.
840 * For now, just allow for last data
841 * devices in RAID4 or last devices in RAID4/5/6.
842 */
843 delta = devices[j].i.delta_disks;
844 if (devices[j].i.array.level >= 4 &&
845 devices[j].i.array.level <= 6 &&
846 i/2 >= content->array.raid_disks - delta)
847 /* OK */;
848 else if (devices[j].i.array.level == 4 &&
849 i/2 >= content->array.raid_disks - delta - 1)
850 /* OK */;
851 else
852 continue;
853 }
854 if (chosen_drive < 0 ||
855 devices[j].i.events
856 > devices[chosen_drive].i.events)
857 chosen_drive = j;
858 }
859 if (chosen_drive < 0)
860 break;
861 current_events = devices[chosen_drive].i.events;
862 add_another:
863 if (c->verbose >= 0)
864 pr_err("forcing event count in %s(%d) from %d upto %d\n",
865 devices[chosen_drive].devname,
866 devices[chosen_drive].i.disk.raid_disk,
867 (int)(devices[chosen_drive].i.events),
868 (int)(devices[most_recent].i.events));
869 fd = dev_open(devices[chosen_drive].devname,
870 devices[chosen_drive].included ? O_RDWR
871 : (O_RDWR|O_EXCL));
872 if (fd < 0) {
873 pr_err("Couldn't open %s for write - not updating\n",
874 devices[chosen_drive].devname);
875 devices[chosen_drive].i.events = 0;
876 continue;
877 }
878 tst = dup_super(st);
879 if (tst->ss->load_super(tst,fd, NULL)) {
880 close(fd);
881 pr_err("RAID superblock disappeared from %s - not updating.\n",
882 devices[chosen_drive].devname);
883 devices[chosen_drive].i.events = 0;
884 continue;
885 }
886 content->events = devices[most_recent].i.events;
887 tst->ss->update_super(tst, content, "force-one",
888 devices[chosen_drive].devname, c->verbose,
889 0, NULL);
890
891 if (tst->ss->store_super(tst, fd)) {
892 close(fd);
893 pr_err("Could not re-write superblock on %s\n",
894 devices[chosen_drive].devname);
895 devices[chosen_drive].i.events = 0;
896 tst->ss->free_super(tst);
897 continue;
898 }
899 close(fd);
900 devices[chosen_drive].i.events = devices[most_recent].i.events;
901 devices[chosen_drive].uptodate = 1;
902 avail[chosen_drive] = 1;
903 okcnt++;
904 tst->ss->free_super(tst);
905
906 /* If there are any other drives of the same vintage,
907 * add them in as well. We can't lose and we might gain
908 */
909 for (i = 0;
910 i < content->array.raid_disks * 2 && i < bestcnt ;
911 i += 2) {
912 int j = best[i];
913 if (j >= 0 &&
914 !devices[j].uptodate &&
915 devices[j].i.recovery_start == MaxSector &&
916 devices[j].i.events == current_events) {
917 chosen_drive = j;
918 goto add_another;
919 }
920 }
921 }
922 return okcnt;
923 }
924
925 static int start_array(int mdfd,
926 char *mddev,
927 struct mdinfo *content,
928 struct supertype *st,
929 struct mddev_ident *ident,
930 int *best, int bestcnt,
931 int chosen_drive,
932 struct devs *devices,
933 unsigned int okcnt,
934 unsigned int sparecnt,
935 unsigned int rebuilding_cnt,
936 struct context *c,
937 int clean, char *avail,
938 int start_partial_ok,
939 int err_ok,
940 int was_forced
941 )
942 {
943 int rv;
944 int i;
945 unsigned int req_cnt;
946
947 rv = set_array_info(mdfd, st, content);
948 if (rv && !err_ok) {
949 pr_err("failed to set array info for %s: %s\n",
950 mddev, strerror(errno));
951 return 1;
952 }
953 if (ident->bitmap_fd >= 0) {
954 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
955 pr_err("SET_BITMAP_FILE failed.\n");
956 return 1;
957 }
958 } else if (ident->bitmap_file) {
959 /* From config file */
960 int bmfd = open(ident->bitmap_file, O_RDWR);
961 if (bmfd < 0) {
962 pr_err("Could not open bitmap file %s\n",
963 ident->bitmap_file);
964 return 1;
965 }
966 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
967 pr_err("Failed to set bitmapfile for %s\n", mddev);
968 close(bmfd);
969 return 1;
970 }
971 close(bmfd);
972 }
973
974 /* First, add the raid disks, but add the chosen one last */
975 for (i=0; i<= bestcnt; i++) {
976 int j;
977 if (i < bestcnt) {
978 j = best[i];
979 if (j == chosen_drive)
980 continue;
981 } else
982 j = chosen_drive;
983
984 if (j >= 0 && !devices[j].included) {
985 int dfd = dev_open(devices[j].devname,
986 O_RDWR|O_EXCL);
987 if (dfd >= 0) {
988 remove_partitions(dfd);
989 close(dfd);
990 }
991 rv = add_disk(mdfd, st, content, &devices[j].i);
992
993 if (rv) {
994 pr_err("failed to add %s to %s: %s\n",
995 devices[j].devname,
996 mddev,
997 strerror(errno));
998 if (i < content->array.raid_disks * 2
999 || i == bestcnt)
1000 okcnt--;
1001 else
1002 sparecnt--;
1003 } else if (c->verbose > 0)
1004 pr_err("added %s to %s as %d%s%s\n",
1005 devices[j].devname, mddev,
1006 devices[j].i.disk.raid_disk,
1007 devices[j].uptodate?"":
1008 " (possibly out of date)",
1009 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1010 } else if (j >= 0) {
1011 if (c->verbose > 0)
1012 pr_err("%s is already in %s as %d\n",
1013 devices[j].devname, mddev,
1014 devices[j].i.disk.raid_disk);
1015 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1016 && (i&1) == 0)
1017 pr_err("no uptodate device for slot %d of %s\n",
1018 i/2, mddev);
1019 }
1020
1021 if (content->array.level == LEVEL_CONTAINER) {
1022 if (c->verbose >= 0) {
1023 pr_err("Container %s has been assembled with %d drive%s",
1024 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1025 if (okcnt < (unsigned)content->array.raid_disks)
1026 fprintf(stderr, " (out of %d)",
1027 content->array.raid_disks);
1028 fprintf(stderr, "\n");
1029 }
1030
1031 if (st->ss->validate_container) {
1032 struct mdinfo *devices_list;
1033 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1034 unsigned int count;
1035 devices_list = NULL;
1036 for (count = 0; count < okcnt+sparecnt; count++) {
1037 info_devices[count] = devices[count].i;
1038 info_devices[count].next = devices_list;
1039 devices_list = &info_devices[count];
1040 }
1041 if (st->ss->validate_container(devices_list))
1042 pr_err("Mismatch detected!\n");
1043 free(info_devices);
1044 }
1045
1046 st->ss->free_super(st);
1047 sysfs_uevent(content, "change");
1048 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1049 /* Was partial, is still partial, so signal an error
1050 * to ensure we don't retry */
1051 return 1;
1052 return 0;
1053 }
1054
1055 /* Get number of in-sync devices according to the superblock.
1056 * We must have this number to start the array without -s or -R
1057 */
1058 req_cnt = content->array.working_disks;
1059
1060 if (c->runstop == 1 ||
1061 (c->runstop <= 0 &&
1062 ( enough(content->array.level, content->array.raid_disks,
1063 content->array.layout, clean, avail) &&
1064 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1065 ))) {
1066 /* This array is good-to-go.
1067 * If a reshape is in progress then we might need to
1068 * continue monitoring it. In that case we start
1069 * it read-only and let the grow code make it writable.
1070 */
1071 int rv;
1072 #ifndef MDASSEMBLE
1073 if (content->reshape_active &&
1074 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1075 content->delta_disks <= 0) {
1076 if (!c->backup_file) {
1077 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1078 mddev);
1079 pr_err("Please provided one with \"--backup-file=...\"\n");
1080 if (c->update &&
1081 strcmp(c->update, "revert-reshape") == 0)
1082 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1083 return 1;
1084 }
1085 rv = sysfs_set_str(content, NULL,
1086 "array_state", "readonly");
1087 if (rv == 0)
1088 rv = Grow_continue(mdfd, st, content,
1089 c->backup_file, 0,
1090 c->freeze_reshape);
1091 } else if (c->readonly &&
1092 sysfs_attribute_available(
1093 content, NULL, "array_state")) {
1094 rv = sysfs_set_str(content, NULL,
1095 "array_state", "readonly");
1096 } else
1097 #endif
1098 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1099 reopen_mddev(mdfd); /* drop O_EXCL */
1100 if (rv == 0) {
1101 if (c->verbose >= 0) {
1102 pr_err("%s has been started with %d drive%s",
1103 mddev, okcnt, okcnt==1?"":"s");
1104 if (okcnt < (unsigned)content->array.raid_disks)
1105 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1106 if (rebuilding_cnt)
1107 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1108 if (sparecnt)
1109 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1110 fprintf(stderr, ".\n");
1111 }
1112 if (content->reshape_active &&
1113 content->array.level >= 4 &&
1114 content->array.level <= 6) {
1115 /* might need to increase the size
1116 * of the stripe cache - default is 256
1117 */
1118 if (256 < 4 * (content->array.chunk_size/4096)) {
1119 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1120 if (sra)
1121 sysfs_set_num(sra, NULL,
1122 "stripe_cache_size",
1123 (4 * content->array.chunk_size / 4096) + 1);
1124 sysfs_free(sra);
1125 }
1126 }
1127 if (okcnt < (unsigned)content->array.raid_disks) {
1128 /* If any devices did not get added
1129 * because the kernel rejected them based
1130 * on event count, try adding them
1131 * again providing the action policy is
1132 * 're-add' or greater. The bitmap
1133 * might allow them to be included, or
1134 * they will become spares.
1135 */
1136 for (i = 0; i < bestcnt; i++) {
1137 int j = best[i];
1138 if (j >= 0 && !devices[j].uptodate) {
1139 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1140 continue;
1141 rv = add_disk(mdfd, st, content,
1142 &devices[j].i);
1143 if (rv == 0 && c->verbose >= 0)
1144 pr_err("%s has been re-added.\n",
1145 devices[j].devname);
1146 }
1147 }
1148 }
1149 if (content->array.level == 6 &&
1150 okcnt + 1 == (unsigned)content->array.raid_disks &&
1151 was_forced) {
1152 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1153 if (sra)
1154 sysfs_set_str(sra, NULL,
1155 "sync_action", "repair");
1156 sysfs_free(sra);
1157 }
1158 return 0;
1159 }
1160 pr_err("failed to RUN_ARRAY %s: %s\n",
1161 mddev, strerror(errno));
1162
1163 if (!enough(content->array.level, content->array.raid_disks,
1164 content->array.layout, 1, avail))
1165 pr_err("Not enough devices to start the array.\n");
1166 else if (!enough(content->array.level,
1167 content->array.raid_disks,
1168 content->array.layout, clean,
1169 avail))
1170 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1171
1172 return 1;
1173 }
1174 if (c->runstop == -1) {
1175 pr_err("%s assembled from %d drive%s",
1176 mddev, okcnt, okcnt==1?"":"s");
1177 if (okcnt != (unsigned)content->array.raid_disks)
1178 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1179 fprintf(stderr, ", but not started.\n");
1180 return 2;
1181 }
1182 if (c->verbose >= -1) {
1183 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1184 if (rebuilding_cnt)
1185 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1186 if (sparecnt)
1187 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1188 if (!enough(content->array.level, content->array.raid_disks,
1189 content->array.layout, 1, avail))
1190 fprintf(stderr, " - not enough to start the array.\n");
1191 else if (!enough(content->array.level,
1192 content->array.raid_disks,
1193 content->array.layout, clean,
1194 avail))
1195 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1196 else {
1197 if (req_cnt == (unsigned)content->array.raid_disks)
1198 fprintf(stderr, " - need all %d to start it", req_cnt);
1199 else
1200 fprintf(stderr, " - need %d to start", req_cnt);
1201 fprintf(stderr, " (use --run to insist).\n");
1202 }
1203 }
1204 return 1;
1205 }
1206
1207 int Assemble(struct supertype *st, char *mddev,
1208 struct mddev_ident *ident,
1209 struct mddev_dev *devlist,
1210 struct context *c)
1211 {
1212 /*
1213 * The task of Assemble is to find a collection of
1214 * devices that should (according to their superblocks)
1215 * form an array, and to give this collection to the MD driver.
1216 * In Linux-2.4 and later, this involves submitting a
1217 * SET_ARRAY_INFO ioctl with no arg - to prepare
1218 * the array - and then submit a number of
1219 * ADD_NEW_DISK ioctls to add disks into
1220 * the array. Finally RUN_ARRAY might
1221 * be submitted to start the array.
1222 *
1223 * Much of the work of Assemble is in finding and/or
1224 * checking the disks to make sure they look right.
1225 *
1226 * If mddev is not set, then scan must be set and we
1227 * read through the config file for dev+uuid mapping
1228 * We recurse, setting mddev, for each device that
1229 * - isn't running
1230 * - has a valid uuid (or any uuid if !uuidset)
1231 *
1232 * If mddev is set, we try to determine state of md.
1233 * check version - must be at least 0.90.0
1234 * check kernel version. must be at least 2.4.
1235 * If not, we can possibly fall back on START_ARRAY
1236 * Try to GET_ARRAY_INFO.
1237 * If possible, give up
1238 * If not, try to STOP_ARRAY just to make sure
1239 *
1240 * If !uuidset and scan, look in conf-file for uuid
1241 * If not found, give up
1242 * If !devlist and scan and uuidset, get list of devs from conf-file
1243 *
1244 * For each device:
1245 * Check superblock - discard if bad
1246 * Check uuid (set if we don't have one) - discard if no match
1247 * Check superblock similarity if we have a superblock - discard if different
1248 * Record events, devicenum
1249 * This should give us a list of devices for the array
1250 * We should collect the most recent event number
1251 *
1252 * Count disks with recent enough event count
1253 * While force && !enough disks
1254 * Choose newest rejected disks, update event count
1255 * mark clean and rewrite superblock
1256 * If recent kernel:
1257 * SET_ARRAY_INFO
1258 * foreach device with recent events : ADD_NEW_DISK
1259 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1260 * If old kernel:
1261 * Check the device numbers in superblock are right
1262 * update superblock if any changes
1263 * START_ARRAY
1264 *
1265 */
1266 int rv;
1267 int mdfd;
1268 int clean;
1269 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1270 ident->super_minor == UnSet && ident->name[0] == 0
1271 && (ident->container == NULL || ident->member == NULL));
1272 struct devs *devices;
1273 char *devmap;
1274 int *best = NULL; /* indexed by raid_disk */
1275 int bestcnt = 0;
1276 int devcnt;
1277 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1278 int i;
1279 int was_forced = 0;
1280 int most_recent = 0;
1281 int chosen_drive;
1282 int change = 0;
1283 int inargv = 0;
1284 int start_partial_ok = (c->runstop >= 0) &&
1285 (c->force || devlist==NULL || auto_assem);
1286 int num_devs;
1287 struct mddev_dev *tmpdev;
1288 struct mdinfo info;
1289 struct mdinfo *content = NULL;
1290 struct mdinfo *pre_exist = NULL;
1291 char *avail;
1292 char *name = NULL;
1293 char chosen_name[1024];
1294 struct map_ent *map = NULL;
1295 struct map_ent *mp;
1296
1297 /*
1298 * If any subdevs are listed, then any that don't
1299 * match ident are discarded. Remainder must all match and
1300 * become the array.
1301 * If no subdevs, then we scan all devices in the config file, but
1302 * there must be something in the identity
1303 */
1304
1305 if (!devlist &&
1306 ident->uuid_set == 0 &&
1307 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1308 ident->name[0] == 0 &&
1309 (ident->container == NULL || ident->member == NULL) &&
1310 ident->devices == NULL) {
1311 pr_err("No identity information available for %s - cannot assemble.\n",
1312 mddev ? mddev : "further assembly");
1313 return 1;
1314 }
1315
1316 if (devlist == NULL)
1317 devlist = conf_get_devs();
1318 else if (mddev)
1319 inargv = 1;
1320
1321 try_again:
1322 /* We come back here when doing auto-assembly and attempting some
1323 * set of devices failed. Those are now marked as ->used==2 and
1324 * we ignore them and try again
1325 */
1326 if (!st && ident->st)
1327 st = ident->st;
1328 if (c->verbose>0)
1329 pr_err("looking for devices for %s\n",
1330 mddev ? mddev : "further assembly");
1331
1332 content = &info;
1333 if (st)
1334 st->ignore_hw_compat = 1;
1335 num_devs = select_devices(devlist, ident, &st, &content, c,
1336 inargv, auto_assem);
1337 if (num_devs < 0)
1338 return 1;
1339
1340 if (!st || !st->sb || !content)
1341 return 2;
1342
1343 /* We have a full set of devices - we now need to find the
1344 * array device.
1345 * However there is a risk that we are racing with "mdadm -I"
1346 * and the array is already partially assembled - we will have
1347 * rejected any devices already in this address.
1348 * So we take a lock on the map file - to prevent further races -
1349 * and look for the uuid in there. If found and the array is
1350 * active, we abort. If found and the array is not active
1351 * we commit to that md device and add all the contained devices
1352 * to our list. We flag them so that we don't try to re-add,
1353 * but can remove if they turn out to not be wanted.
1354 */
1355 if (map_lock(&map))
1356 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1357 if (c->update && strcmp(c->update,"uuid") == 0)
1358 mp = NULL;
1359 else
1360 mp = map_by_uuid(&map, content->uuid);
1361 if (mp) {
1362 struct mdinfo *dv;
1363 /* array already exists. */
1364 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1365 if (pre_exist->array.level != UnSet) {
1366 pr_err("Found some drive for an array that is already active: %s\n",
1367 mp->path);
1368 pr_err("giving up.\n");
1369 return 1;
1370 }
1371 for (dv = pre_exist->devs; dv; dv = dv->next) {
1372 /* We want to add this device to our list,
1373 * but it could already be there if "mdadm -I"
1374 * started *after* we checked for O_EXCL.
1375 * If we add it to the top of the list
1376 * it will be preferred over later copies.
1377 */
1378 struct mddev_dev *newdev;
1379 char *devname = map_dev(dv->disk.major,
1380 dv->disk.minor,
1381 0);
1382 if (!devname)
1383 continue;
1384 newdev = xmalloc(sizeof(*newdev));
1385 newdev->devname = devname;
1386 newdev->disposition = 'I';
1387 newdev->used = 1;
1388 newdev->next = devlist;
1389 devlist = newdev;
1390 num_devs++;
1391 }
1392 strcpy(chosen_name, mp->path);
1393 if (c->verbose > 0 || mddev == NULL ||
1394 strcmp(mddev, chosen_name) != 0)
1395 pr_err("Merging with already-assembled %s\n",
1396 chosen_name);
1397 mdfd = open_dev_excl(mp->devnm);
1398 } else {
1399 int trustworthy = FOREIGN;
1400 name = content->name;
1401 switch (st->ss->match_home(st, c->homehost)
1402 ?: st->ss->match_home(st, "any")) {
1403 case 1:
1404 trustworthy = LOCAL;
1405 name = strchr(content->name, ':');
1406 if (name)
1407 name++;
1408 else
1409 name = content->name;
1410 break;
1411 }
1412 if (!auto_assem)
1413 /* If the array is listed in mdadm.conf or on
1414 * command line, then we trust the name
1415 * even if the array doesn't look local
1416 */
1417 trustworthy = LOCAL;
1418
1419 if (name[0] == 0 &&
1420 content->array.level == LEVEL_CONTAINER) {
1421 name = content->text_version;
1422 trustworthy = METADATA;
1423 }
1424
1425 if (name[0] && trustworthy != LOCAL &&
1426 ! c->require_homehost &&
1427 conf_name_is_free(name))
1428 trustworthy = LOCAL;
1429
1430 if (trustworthy == LOCAL &&
1431 strchr(name, ':'))
1432 /* Ignore 'host:' prefix of name */
1433 name = strchr(name, ':')+1;
1434
1435 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1436 chosen_name);
1437 }
1438 if (mdfd < 0) {
1439 st->ss->free_super(st);
1440 if (auto_assem)
1441 goto try_again;
1442 return 1;
1443 }
1444 mddev = chosen_name;
1445 if (get_linux_version() < 2004000 ||
1446 md_get_version(mdfd) < 9000) {
1447 pr_err("Assemble requires Linux 2.4 or later, and\n"
1448 " md driver version 0.90.0 or later.\n"
1449 " Upgrade your kernel or try --build\n");
1450 close(mdfd);
1451 return 1;
1452 }
1453 if (pre_exist == NULL) {
1454 if (mddev_busy(fd2devnm(mdfd))) {
1455 pr_err("%s already active, cannot restart it!\n",
1456 mddev);
1457 for (tmpdev = devlist ;
1458 tmpdev && tmpdev->used != 1;
1459 tmpdev = tmpdev->next)
1460 ;
1461 if (tmpdev && auto_assem)
1462 pr_err("%s needed for %s...\n",
1463 mddev, tmpdev->devname);
1464 close(mdfd);
1465 mdfd = -3;
1466 st->ss->free_super(st);
1467 if (auto_assem)
1468 goto try_again;
1469 return 1;
1470 }
1471 /* just incase it was started but has no content */
1472 ioctl(mdfd, STOP_ARRAY, NULL);
1473 }
1474
1475 #ifndef MDASSEMBLE
1476 if (content != &info) {
1477 /* This is a member of a container. Try starting the array. */
1478 int err;
1479 err = assemble_container_content(st, mdfd, content, c,
1480 chosen_name, NULL);
1481 close(mdfd);
1482 return err;
1483 }
1484 #endif
1485 /* Ok, no bad inconsistancy, we can try updating etc */
1486 devices = xcalloc(num_devs, sizeof(*devices));
1487 devmap = xcalloc(num_devs, content->array.raid_disks);
1488 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1489 c, content, mdfd, mddev,
1490 &most_recent, &bestcnt, &best, inargv);
1491 if (devcnt < 0)
1492 return 1;
1493
1494 if (devcnt == 0) {
1495 pr_err("no devices found for %s\n",
1496 mddev);
1497 if (st)
1498 st->ss->free_super(st);
1499 close(mdfd);
1500 free(devices);
1501 free(devmap);
1502 return 1;
1503 }
1504
1505 if (c->update && strcmp(c->update, "byteorder")==0)
1506 st->minor_version = 90;
1507
1508 st->ss->getinfo_super(st, content, NULL);
1509 clean = content->array.state & 1;
1510
1511 /* now we have some devices that might be suitable.
1512 * I wonder how many
1513 */
1514 avail = xcalloc(content->array.raid_disks, 1);
1515 okcnt = 0;
1516 replcnt = 0;
1517 sparecnt=0;
1518 rebuilding_cnt=0;
1519 for (i=0; i< bestcnt; i++) {
1520 int j = best[i];
1521 int event_margin = 1; /* always allow a difference of '1'
1522 * like the kernel does
1523 */
1524 if (j < 0) continue;
1525 /* note: we ignore error flags in multipath arrays
1526 * as they don't make sense
1527 */
1528 if (content->array.level != LEVEL_MULTIPATH)
1529 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1530 if (!(devices[j].i.disk.state
1531 & (1<<MD_DISK_FAULTY))) {
1532 devices[j].uptodate = 1;
1533 sparecnt++;
1534 }
1535 continue;
1536 }
1537 /* If this device thinks that 'most_recent' has failed, then
1538 * we must reject this device.
1539 */
1540 if (j != most_recent && !c->force &&
1541 content->array.raid_disks > 0 &&
1542 devices[most_recent].i.disk.raid_disk >= 0 &&
1543 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1544 if (c->verbose > -1)
1545 pr_err("ignoring %s as it reports %s as failed\n",
1546 devices[j].devname, devices[most_recent].devname);
1547 best[i] = -1;
1548 continue;
1549 }
1550 /* Require event counter to be same as, or just less than,
1551 * most recent. If it is bigger, it must be a stray spare and
1552 * should be ignored.
1553 */
1554 if (devices[j].i.events+event_margin >=
1555 devices[most_recent].i.events &&
1556 devices[j].i.events <=
1557 devices[most_recent].i.events
1558 ) {
1559 devices[j].uptodate = 1;
1560 if (i < content->array.raid_disks * 2) {
1561 if (devices[j].i.recovery_start == MaxSector ||
1562 (content->reshape_active &&
1563 i >= content->array.raid_disks - content->delta_disks)) {
1564 if (!avail[i/2]) {
1565 okcnt++;
1566 avail[i/2]=1;
1567 } else
1568 replcnt++;
1569 } else
1570 rebuilding_cnt++;
1571 } else
1572 sparecnt++;
1573 }
1574 }
1575 free(devmap);
1576 if (c->force) {
1577 int force_ok = force_array(content, devices, best, bestcnt,
1578 avail, most_recent, st, c);
1579 okcnt += force_ok;
1580 if (force_ok)
1581 was_forced = 1;
1582 }
1583 /* Now we want to look at the superblock which the kernel will base things on
1584 * and compare the devices that we think are working with the devices that the
1585 * superblock thinks are working.
1586 * If there are differences and --force is given, then update this chosen
1587 * superblock.
1588 */
1589 chosen_drive = -1;
1590 st->ss->free_super(st);
1591 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1592 int j = best[i];
1593 int fd;
1594
1595 if (j<0)
1596 continue;
1597 if (!devices[j].uptodate)
1598 continue;
1599 if (devices[j].i.events < devices[most_recent].i.events)
1600 continue;
1601 chosen_drive = j;
1602 if ((fd=dev_open(devices[j].devname,
1603 devices[j].included ? O_RDONLY
1604 : (O_RDONLY|O_EXCL)))< 0) {
1605 pr_err("Cannot open %s: %s\n",
1606 devices[j].devname, strerror(errno));
1607 close(mdfd);
1608 free(devices);
1609 return 1;
1610 }
1611 if (st->ss->load_super(st,fd, NULL)) {
1612 close(fd);
1613 pr_err("RAID superblock has disappeared from %s\n",
1614 devices[j].devname);
1615 close(mdfd);
1616 free(devices);
1617 return 1;
1618 }
1619 close(fd);
1620 }
1621 if (st->sb == NULL) {
1622 pr_err("No suitable drives found for %s\n", mddev);
1623 close(mdfd);
1624 free(devices);
1625 return 1;
1626 }
1627 st->ss->getinfo_super(st, content, NULL);
1628 #ifndef MDASSEMBLE
1629 sysfs_init(content, mdfd, NULL);
1630 #endif
1631 for (i=0; i<bestcnt; i++) {
1632 int j = best[i];
1633 unsigned int desired_state;
1634
1635 if (i >= content->array.raid_disks * 2)
1636 desired_state = 0;
1637 else if (i & 1)
1638 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1639 else
1640 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1641
1642 if (j<0)
1643 continue;
1644 if (!devices[j].uptodate)
1645 continue;
1646
1647 devices[j].i.disk.state = desired_state;
1648 if (!(devices[j].i.array.state & 1))
1649 clean = 0;
1650
1651 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1652 c->verbose, 0, NULL)) {
1653 if (c->force) {
1654 if (c->verbose >= 0)
1655 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1656 j, mddev, devices[j].devname);
1657 change = 1;
1658 } else {
1659 if (c->verbose >= -1)
1660 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1661 i, mddev, devices[j].devname);
1662 }
1663 }
1664 #if 0
1665 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1666 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1667 i, mddev);
1668 }
1669 #endif
1670 }
1671 if (c->force && !clean &&
1672 !enough(content->array.level, content->array.raid_disks,
1673 content->array.layout, clean,
1674 avail)) {
1675 change += st->ss->update_super(st, content, "force-array",
1676 devices[chosen_drive].devname, c->verbose,
1677 0, NULL);
1678 was_forced = 1;
1679 clean = 1;
1680 }
1681
1682 if (change) {
1683 int fd;
1684 fd = dev_open(devices[chosen_drive].devname,
1685 devices[chosen_drive].included ?
1686 O_RDWR : (O_RDWR|O_EXCL));
1687 if (fd < 0) {
1688 pr_err("Could not open %s for write - cannot Assemble array.\n",
1689 devices[chosen_drive].devname);
1690 close(mdfd);
1691 free(devices);
1692 return 1;
1693 }
1694 if (st->ss->store_super(st, fd)) {
1695 close(fd);
1696 pr_err("Could not re-write superblock on %s\n",
1697 devices[chosen_drive].devname);
1698 close(mdfd);
1699 free(devices);
1700 return 1;
1701 }
1702 if (c->verbose >= 0)
1703 pr_err("Marking array %s as 'clean'\n",
1704 mddev);
1705 close(fd);
1706 }
1707
1708 /* If we are in the middle of a reshape we may need to restore saved data
1709 * that was moved aside due to the reshape overwriting live data
1710 * The code of doing this lives in Grow.c
1711 */
1712 #ifndef MDASSEMBLE
1713 if (content->reshape_active &&
1714 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1715 int err = 0;
1716 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1717 if (c->verbose > 0)
1718 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1719 chosen_name);
1720 if (!c->backup_file)
1721 c->backup_file = locate_backup(content->sys_name);
1722 enable_fds(bestcnt/2);
1723 for (i = 0; i < bestcnt/2; i++) {
1724 int j = best[i*2];
1725 if (j >= 0) {
1726 fdlist[i] = dev_open(devices[j].devname,
1727 devices[j].included
1728 ? O_RDWR : (O_RDWR|O_EXCL));
1729 if (fdlist[i] < 0) {
1730 pr_err("Could not open %s for write - cannot Assemble array.\n",
1731 devices[j].devname);
1732 err = 1;
1733 break;
1734 }
1735 } else
1736 fdlist[i] = -1;
1737 }
1738 if (!err) {
1739 if (st->ss->external && st->ss->recover_backup)
1740 err = st->ss->recover_backup(st, content);
1741 else
1742 err = Grow_restart(st, content, fdlist, bestcnt/2,
1743 c->backup_file, c->verbose > 0);
1744 if (err && c->invalid_backup) {
1745 if (c->verbose > 0)
1746 pr_err("continuing without restoring backup\n");
1747 err = 0;
1748 }
1749 }
1750 while (i>0) {
1751 i--;
1752 if (fdlist[i]>=0) close(fdlist[i]);
1753 }
1754 free(fdlist);
1755 if (err) {
1756 pr_err("Failed to restore critical section for reshape, sorry.\n");
1757 if (c->backup_file == NULL)
1758 cont_err("Possibly you needed to specify the --backup-file\n");
1759 close(mdfd);
1760 free(devices);
1761 return err;
1762 }
1763 }
1764 #endif
1765
1766 /* Almost ready to actually *do* something */
1767 /* First, fill in the map, so that udev can find our name
1768 * as soon as we become active.
1769 */
1770 if (c->update && strcmp(c->update, "metadata")==0) {
1771 content->array.major_version = 1;
1772 content->array.minor_version = 0;
1773 strcpy(content->text_version, "1.0");
1774 }
1775
1776 map_update(&map, fd2devnm(mdfd), content->text_version,
1777 content->uuid, chosen_name);
1778
1779 rv = start_array(mdfd, mddev, content,
1780 st, ident, best, bestcnt,
1781 chosen_drive, devices, okcnt, sparecnt,
1782 rebuilding_cnt,
1783 c,
1784 clean, avail, start_partial_ok,
1785 pre_exist != NULL,
1786 was_forced);
1787 if (rv == 1 && !pre_exist)
1788 ioctl(mdfd, STOP_ARRAY, NULL);
1789 free(devices);
1790 map_unlock(&map);
1791 if (rv == 0) {
1792 wait_for(chosen_name, mdfd);
1793 close(mdfd);
1794 if (auto_assem) {
1795 int usecs = 1;
1796 /* There is a nasty race with 'mdadm --monitor'.
1797 * If it opens this device before we close it,
1798 * it gets an incomplete open on which IO
1799 * doesn't work and the capacity is
1800 * wrong.
1801 * If we reopen (to check for layered devices)
1802 * before --monitor closes, we loose.
1803 *
1804 * So: wait upto 1 second for there to be
1805 * a non-zero capacity.
1806 */
1807 while (usecs < 1000) {
1808 mdfd = open(mddev, O_RDONLY);
1809 if (mdfd >= 0) {
1810 unsigned long long size;
1811 if (get_dev_size(mdfd, NULL, &size) &&
1812 size > 0)
1813 break;
1814 close(mdfd);
1815 }
1816 usleep(usecs);
1817 usecs <<= 1;
1818 }
1819 }
1820 } else
1821 close(mdfd);
1822
1823 /* '2' means 'OK, but not started yet' */
1824 return rv == 2 ? 0 : rv;
1825 }
1826
1827 #ifndef MDASSEMBLE
1828 int assemble_container_content(struct supertype *st, int mdfd,
1829 struct mdinfo *content, struct context *c,
1830 char *chosen_name, int *result)
1831 {
1832 struct mdinfo *dev, *sra, *dev2;
1833 int working = 0, preexist = 0;
1834 int expansion = 0;
1835 struct map_ent *map = NULL;
1836 int old_raid_disks;
1837 int start_reshape;
1838 char *avail = NULL;
1839 int err;
1840
1841 sysfs_init(content, mdfd, NULL);
1842
1843 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1844 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1845 if (content->array.major_version == -1 &&
1846 content->array.minor_version == -2 &&
1847 c->readonly &&
1848 content->text_version[0] == '/')
1849 content->text_version[0] = '-';
1850 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1851 if (sra)
1852 sysfs_free(sra);
1853 return 1;
1854 }
1855 }
1856
1857 /* There are two types of reshape: container wide or sub-array specific
1858 * Check if metadata requests blocking container wide reshapes
1859 */
1860 start_reshape = (content->reshape_active &&
1861 !((content->reshape_active == CONTAINER_RESHAPE) &&
1862 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1863
1864 /* Block subarray here if it is under reshape now
1865 * Do not allow for any changes in this array
1866 */
1867 if (st->ss->external && content->recovery_blocked && start_reshape)
1868 block_subarray(content);
1869
1870 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1871 for (dev = content->devs; dev; dev = dev->next)
1872 if (dev2->disk.major == dev->disk.major &&
1873 dev2->disk.minor == dev->disk.minor)
1874 break;
1875 if (dev)
1876 continue;
1877 /* Don't want this one any more */
1878 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1879 errno == EBUSY) {
1880 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1881 sysfs_free(sra);
1882 return 1;
1883 }
1884 sysfs_set_str(sra, dev2, "state", "remove");
1885 }
1886 old_raid_disks = content->array.raid_disks - content->delta_disks;
1887 avail = xcalloc(content->array.raid_disks, 1);
1888 for (dev = content->devs; dev; dev = dev->next) {
1889 if (dev->disk.raid_disk >= 0)
1890 avail[dev->disk.raid_disk] = 1;
1891 if (sysfs_add_disk(content, dev, 1) == 0) {
1892 if (dev->disk.raid_disk >= old_raid_disks &&
1893 content->reshape_active)
1894 expansion++;
1895 else
1896 working++;
1897 } else if (errno == EEXIST)
1898 preexist++;
1899 }
1900 sysfs_free(sra);
1901 if (working + expansion == 0 && c->runstop <= 0) {
1902 free(avail);
1903 return 1;/* Nothing new, don't try to start */
1904 }
1905 map_update(&map, fd2devnm(mdfd),
1906 content->text_version,
1907 content->uuid, chosen_name);
1908
1909
1910 if (enough(content->array.level, content->array.raid_disks,
1911 content->array.layout, content->array.state & 1, avail) == 0) {
1912 if (c->export && result)
1913 *result |= INCR_NO;
1914 else if (c->verbose >= 0) {
1915 pr_err("%s assembled with %d device%s",
1916 chosen_name, preexist + working,
1917 preexist + working == 1 ? "":"s");
1918 if (preexist)
1919 fprintf(stderr, " (%d new)", working);
1920 fprintf(stderr, " but not started\n");
1921 }
1922 free(avail);
1923 return 1;
1924 }
1925 free(avail);
1926
1927 if (c->runstop <= 0 &&
1928 (working + preexist + expansion) <
1929 content->array.working_disks) {
1930 if (c->export && result)
1931 *result |= INCR_UNSAFE;
1932 else if (c->verbose >= 0) {
1933 pr_err("%s assembled with %d device%s",
1934 chosen_name, preexist + working,
1935 preexist + working == 1 ? "":"s");
1936 if (preexist)
1937 fprintf(stderr, " (%d new)", working);
1938 fprintf(stderr, " but not safe to start\n");
1939 }
1940 return 1;
1941 }
1942
1943
1944 if (start_reshape) {
1945 int spare = content->array.raid_disks + expansion;
1946 if (restore_backup(st, content,
1947 working,
1948 spare, &c->backup_file, c->verbose) == 1)
1949 return 1;
1950
1951 err = sysfs_set_str(content, NULL,
1952 "array_state", "readonly");
1953 if (err)
1954 return 1;
1955
1956 if (st->ss->external) {
1957 if (!mdmon_running(st->container_devnm))
1958 start_mdmon(st->container_devnm);
1959 ping_monitor(st->container_devnm);
1960 if (mdmon_running(st->container_devnm) &&
1961 st->update_tail == NULL)
1962 st->update_tail = &st->updates;
1963 }
1964
1965 err = Grow_continue(mdfd, st, content, c->backup_file,
1966 0, c->freeze_reshape);
1967 } else switch(content->array.level) {
1968 case LEVEL_LINEAR:
1969 case LEVEL_MULTIPATH:
1970 case 0:
1971 err = sysfs_set_str(content, NULL, "array_state",
1972 c->readonly ? "readonly" : "active");
1973 break;
1974 default:
1975 err = sysfs_set_str(content, NULL, "array_state",
1976 "readonly");
1977 /* start mdmon if needed. */
1978 if (!err) {
1979 if (!mdmon_running(st->container_devnm))
1980 start_mdmon(st->container_devnm);
1981 ping_monitor(st->container_devnm);
1982 }
1983 break;
1984 }
1985 if (!err)
1986 sysfs_set_safemode(content, content->safe_mode_delay);
1987
1988 /* Block subarray here if it is not reshaped now
1989 * It has be blocked a little later to allow mdmon to switch in
1990 * in to R/W state
1991 */
1992 if (st->ss->external && content->recovery_blocked &&
1993 !start_reshape)
1994 block_subarray(content);
1995
1996 if (c->export && result) {
1997 if (err)
1998 *result |= INCR_NO;
1999 else
2000 *result |= INCR_YES;
2001 } else if (c->verbose >= 0) {
2002 if (err)
2003 pr_err("array %s now has %d device%s",
2004 chosen_name, working + preexist,
2005 working + preexist == 1 ? "":"s");
2006 else
2007 pr_err("Started %s with %d device%s",
2008 chosen_name, working + preexist,
2009 working + preexist == 1 ? "":"s");
2010 if (preexist)
2011 fprintf(stderr, " (%d new)", working);
2012 if (expansion)
2013 fprintf(stderr, " ( + %d for expansion)",
2014 expansion);
2015 fprintf(stderr, "\n");
2016 }
2017 if (!err)
2018 wait_for(chosen_name, mdfd);
2019 return err;
2020 /* FIXME should have an O_EXCL and wait for read-auto */
2021 }
2022 #endif
Unnamed repository; edit this file 'description' to name the repository.