]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Monitor/check_array: Get nr_disks, active_disks and spare_disks from sysfs
[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 dev_t rdev;
153
154 tmpdev = devlist; num_devs = 0;
155 while (tmpdev) {
156 if (tmpdev->used)
157 tmpdev->used = 2;
158 else
159 num_devs++;
160 tmpdev->disposition = 0;
161 tmpdev = tmpdev->next;
162 }
163
164 /* first walk the list of devices to find a consistent set
165 * that match the criterea, if that is possible.
166 * We flag the ones we like with 'used'.
167 */
168 for (tmpdev = devlist;
169 tmpdev;
170 tmpdev = tmpdev ? tmpdev->next : NULL) {
171 char *devname = tmpdev->devname;
172 int dfd;
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_is_blkdev(dfd, devname, &rdev)) {
208 tmpdev->used = 2;
209 } else if (must_be_container(dfd)) {
210 if (st) {
211 /* already found some components, this cannot
212 * be another one.
213 */
214 if (report_mismatch)
215 pr_err("%s is a container, but we are looking for components\n",
216 devname);
217 tmpdev->used = 2;
218 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
219 if (report_mismatch)
220 pr_err("not a recognisable container: %s\n",
221 devname);
222 tmpdev->used = 2;
223 } else if (!tst->ss->load_container
224 || tst->ss->load_container(tst, dfd, NULL)) {
225 if (report_mismatch)
226 pr_err("no correct container type: %s\n",
227 devname);
228 tmpdev->used = 2;
229 } else if (auto_assem &&
230 !conf_test_metadata(tst->ss->name,
231 (pol = devid_policy(rdev)),
232 tst->ss->match_home(tst, c->homehost) == 1)) {
233 if (report_mismatch)
234 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
235 devname, tst->ss->name);
236 tmpdev->used = 2;
237 } else
238 found_container = 1;
239 } else {
240 if (!tst && (tst = guess_super(dfd)) == NULL) {
241 if (report_mismatch)
242 pr_err("no recogniseable superblock on %s\n",
243 devname);
244 tmpdev->used = 2;
245 } else if ((tst->ignore_hw_compat = 0),
246 tst->ss->load_super(tst, dfd,
247 report_mismatch ? devname : 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,
259 (pol = devid_policy(rdev)),
260 tst->ss->match_home(tst, c->homehost) == 1)) {
261 if (report_mismatch)
262 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
263 devname, tst->ss->name);
264 tmpdev->used = 2;
265 }
266 }
267 if (dfd >= 0) close(dfd);
268 if (tmpdev->used == 2) {
269 if (auto_assem || !inargv)
270 /* Ignore unrecognised devices during auto-assembly */
271 goto loop;
272 if (ident->uuid_set || ident->name[0] ||
273 ident->super_minor != UnSet)
274 /* Ignore unrecognised device if looking for
275 * specific array */
276 goto loop;
277
278 pr_err("%s has no superblock - assembly aborted\n",
279 devname);
280 if (st)
281 st->ss->free_super(st);
282 dev_policy_free(pol);
283 domain_free(domains);
284 return -1;
285 }
286
287 if (found_container) {
288 /* tmpdev is a container. We need to be either
289 * looking for a member, or auto-assembling
290 */
291 /* should be safe to try an exclusive open now, we
292 * have rejected anything that some other mdadm might
293 * be looking at
294 */
295 dfd = dev_open(devname, O_RDONLY | O_EXCL);
296 if (dfd < 0) {
297 if (report_mismatch)
298 pr_err("%s is busy - skipping\n", devname);
299 goto loop;
300 }
301 close(dfd);
302
303 if (ident->container && ident->container[0] != '/') {
304 /* we have a uuid */
305 int uuid[4];
306
307 content = *contentp;
308 tst->ss->getinfo_super(tst, content, NULL);
309
310 if (!parse_uuid(ident->container, uuid) ||
311 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
312 if (report_mismatch)
313 pr_err("%s has wrong UUID to be required container\n",
314 devname);
315 goto loop;
316 }
317 }
318 /* It is worth looking inside this container.
319 */
320 if (c->verbose > 0)
321 pr_err("looking in container %s\n",
322 devname);
323
324 for (content = tst->ss->container_content(tst, NULL);
325 content;
326 content = content->next) {
327
328 if (!ident_matches(ident, content, tst,
329 c->homehost, c->require_homehost,
330 c->update,
331 report_mismatch ? devname : NULL))
332 /* message already printed */;
333 else if (is_member_busy(content->text_version)) {
334 if (report_mismatch)
335 pr_err("member %s in %s is already assembled\n",
336 content->text_version,
337 devname);
338 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
339 /* do not assemble arrays with unsupported configurations */
340 pr_err("Cannot activate member %s in %s.\n",
341 content->text_version,
342 devname);
343 } else
344 break;
345 }
346 if (!content) {
347 tmpdev->used = 2;
348 goto loop; /* empty container */
349 }
350
351 st = tst; tst = NULL;
352 if (!auto_assem && inargv && tmpdev->next != NULL) {
353 pr_err("%s is a container, but is not only device given: confused and aborting\n",
354 devname);
355 st->ss->free_super(st);
356 dev_policy_free(pol);
357 domain_free(domains);
358 return -1;
359 }
360 if (c->verbose > 0)
361 pr_err("found match on member %s in %s\n",
362 content->text_version, devname);
363
364 /* make sure we finished the loop */
365 tmpdev = NULL;
366 goto loop;
367 } else {
368 content = *contentp;
369 tst->ss->getinfo_super(tst, content, NULL);
370
371 if (!ident_matches(ident, content, tst,
372 c->homehost, c->require_homehost,
373 c->update,
374 report_mismatch ? devname : NULL))
375 goto loop;
376
377 if (auto_assem) {
378 /* Never auto-assemble things that conflict
379 * with mdadm.conf in some way
380 */
381 struct mddev_ident *match;
382 int rv = 0;
383
384 match = conf_match(tst, content, devname,
385 report_mismatch ? c->verbose : -1,
386 &rv);
387 if (!match && rv == 2)
388 goto loop;
389 if (match && match->devname &&
390 strcasecmp(match->devname, "<ignore>") == 0) {
391 if (report_mismatch)
392 pr_err("%s is a member of an explicitly ignored array\n",
393 devname);
394 goto loop;
395 }
396 if (match && !ident_matches(match, content, tst,
397 c->homehost, c->require_homehost,
398 c->update,
399 report_mismatch ? devname : NULL))
400 /* Array exists in mdadm.conf but some
401 * details don't match, so reject it
402 */
403 goto loop;
404 }
405
406 /* should be safe to try an exclusive open now, we
407 * have rejected anything that some other mdadm might
408 * be looking at
409 */
410 dfd = dev_open(devname, O_RDONLY | O_EXCL);
411 if (dfd < 0) {
412 if (report_mismatch)
413 pr_err("%s is busy - skipping\n", devname);
414 goto loop;
415 }
416 close(dfd);
417
418 if (st == NULL)
419 st = dup_super(tst);
420 if (st->minor_version == -1)
421 st->minor_version = tst->minor_version;
422
423 if (memcmp(content->uuid, uuid_zero,
424 sizeof(int[4])) == 0) {
425 /* this is a floating spare. It cannot define
426 * an array unless there are no more arrays of
427 * this type to be found. It can be included
428 * in an array of this type though.
429 */
430 tmpdev->used = 3;
431 goto loop;
432 }
433
434 if (st->ss != tst->ss ||
435 st->minor_version != tst->minor_version ||
436 st->ss->compare_super(st, tst) != 0) {
437 /* Some mismatch. If exactly one array matches this host,
438 * we can resolve on that one.
439 * Or, if we are auto assembling, we just ignore the second
440 * for now.
441 */
442 if (auto_assem)
443 goto loop;
444 if (c->homehost) {
445 int first = st->ss->match_home(st, c->homehost);
446 int last = tst->ss->match_home(tst, c->homehost);
447 if (first != last &&
448 (first == 1 || last == 1)) {
449 /* We can do something */
450 if (first) {/* just ignore this one */
451 if (report_mismatch)
452 pr_err("%s misses out due to wrong homehost\n",
453 devname);
454 goto loop;
455 } else { /* reject all those sofar */
456 struct mddev_dev *td;
457 if (report_mismatch)
458 pr_err("%s overrides previous devices due to good homehost\n",
459 devname);
460 for (td=devlist; td != tmpdev; td=td->next)
461 if (td->used == 1)
462 td->used = 0;
463 tmpdev->used = 1;
464 goto loop;
465 }
466 }
467 }
468 pr_err("superblock on %s doesn't match others - assembly aborted\n",
469 devname);
470 tst->ss->free_super(tst);
471 st->ss->free_super(st);
472 dev_policy_free(pol);
473 domain_free(domains);
474 return -1;
475 }
476 tmpdev->used = 1;
477 }
478 loop:
479 /* Collect domain information from members only */
480 if (tmpdev && tmpdev->used == 1) {
481 if (!pol)
482 pol = devid_policy(rdev);
483 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
484 }
485 dev_policy_free(pol);
486 pol = NULL;
487 if (tst)
488 tst->ss->free_super(tst);
489 }
490
491 /* Check if we found some imsm spares but no members */
492 if ((auto_assem ||
493 (ident->uuid_set &&
494 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
495 (!st || !st->sb))
496 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
497 if (tmpdev->used != 3)
498 continue;
499 tmpdev->used = 1;
500 content = *contentp;
501
502 if (!st->sb) {
503 /* we need sb from one of the spares */
504 int dfd = dev_open(tmpdev->devname, O_RDONLY);
505 if (dfd < 0 ||
506 st->ss->load_super(st, dfd, NULL))
507 tmpdev->used = 2;
508 if (dfd > 0)
509 close(dfd);
510 }
511 }
512
513 /* Now reject spares that don't match domains of identified members */
514 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
515 if (tmpdev->used != 3)
516 continue;
517 if (!stat_is_blkdev(tmpdev->devname, &rdev)) {
518 tmpdev->used = 2;
519 } else {
520 struct dev_policy *pol = devid_policy(rdev);
521 int dt = domain_test(domains, pol, NULL);
522 if (inargv && dt != 0)
523 /* take this spare as domains match
524 * if there are any */
525 tmpdev->used = 1;
526 else if (!inargv && dt == 1)
527 /* device wasn't explicitly listed, so need
528 * explicit domain match - which we have */
529 tmpdev->used = 1;
530 else
531 /* if domains don't match mark as unused */
532 tmpdev->used = 0;
533 dev_policy_free(pol);
534 }
535 }
536 domain_free(domains);
537 *stp = st;
538 if (st && st->sb && content == *contentp)
539 st->ss->getinfo_super(st, content, NULL);
540 *contentp = content;
541
542 return num_devs;
543 }
544
545 struct devs {
546 char *devname;
547 int uptodate; /* set once we decide that this device is as
548 * recent as everything else in the array.
549 */
550 int included; /* set if the device is already in the array
551 * due to a previous '-I'
552 */
553 struct mdinfo i;
554 };
555
556 static int load_devices(struct devs *devices, char *devmap,
557 struct mddev_ident *ident, struct supertype **stp,
558 struct mddev_dev *devlist, struct context *c,
559 struct mdinfo *content,
560 int mdfd, char *mddev,
561 int *most_recentp, int *bestcntp, int **bestp,
562 int inargv)
563 {
564 struct mddev_dev *tmpdev;
565 int devcnt = 0;
566 int nextspare = 0;
567 int bitmap_done = 0;
568 int most_recent = -1;
569 int bestcnt = 0;
570 int *best = *bestp;
571 struct supertype *st = *stp;
572
573 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
574 char *devname = tmpdev->devname;
575 struct stat stb;
576 struct supertype *tst;
577 int i;
578 int dfd;
579
580 if (tmpdev->used != 1)
581 continue;
582 /* looks like a good enough match to update the super block if needed */
583 if (c->update) {
584 /* prepare useful information in info structures */
585 struct stat stb2;
586 int err;
587 fstat(mdfd, &stb2);
588
589 if (strcmp(c->update, "uuid") == 0 && !ident->uuid_set)
590 random_uuid((__u8 *)ident->uuid);
591
592 if (strcmp(c->update, "ppl") == 0 &&
593 ident->bitmap_fd >= 0) {
594 pr_err("PPL is not compatible with bitmap\n");
595 return -1;
596 }
597
598 dfd = dev_open(devname,
599 tmpdev->disposition == 'I'
600 ? O_RDWR : (O_RDWR|O_EXCL));
601
602 tst = dup_super(st);
603 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
604 pr_err("cannot re-read metadata from %s - aborting\n",
605 devname);
606 if (dfd >= 0)
607 close(dfd);
608 close(mdfd);
609 free(devices);
610 free(devmap);
611 tst->ss->free_super(tst);
612 free(tst);
613 *stp = st;
614 return -1;
615 }
616 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
617
618 memcpy(content->uuid, ident->uuid, 16);
619 strcpy(content->name, ident->name);
620 content->array.md_minor = minor(stb2.st_rdev);
621
622 if (strcmp(c->update, "byteorder") == 0)
623 err = 0;
624 else if (strcmp(c->update, "home-cluster") == 0) {
625 tst->cluster_name = c->homecluster;
626 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
627 } else if (strcmp(c->update, "nodes") == 0) {
628 tst->nodes = c->nodes;
629 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
630 } else if (strcmp(c->update, "revert-reshape") == 0 &&
631 c->invalid_backup)
632 err = tst->ss->update_super(tst, content,
633 "revert-reshape-nobackup",
634 devname, c->verbose,
635 ident->uuid_set,
636 c->homehost);
637 else
638 err = tst->ss->update_super(tst, content, c->update,
639 devname, c->verbose,
640 ident->uuid_set,
641 c->homehost);
642 if (err < 0) {
643 if (err == -1)
644 pr_err("--update=%s not understood for %s metadata\n",
645 c->update, tst->ss->name);
646 tst->ss->free_super(tst);
647 free(tst);
648 close(mdfd);
649 close(dfd);
650 free(devices);
651 free(devmap);
652 *stp = st;
653 return -1;
654 }
655 if (strcmp(c->update, "uuid")==0 &&
656 !ident->uuid_set) {
657 ident->uuid_set = 1;
658 memcpy(ident->uuid, content->uuid, 16);
659 }
660 if (tst->ss->store_super(tst, dfd))
661 pr_err("Could not re-write superblock on %s.\n",
662 devname);
663
664 if (strcmp(c->update, "uuid")==0 &&
665 ident->bitmap_fd >= 0 && !bitmap_done) {
666 if (bitmap_update_uuid(ident->bitmap_fd,
667 content->uuid,
668 tst->ss->swapuuid) != 0)
669 pr_err("Could not update uuid on external bitmap.\n");
670 else
671 bitmap_done = 1;
672 }
673 } else {
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 || i == MD_DISK_ROLE_JOURNAL) {
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 /* If there are any other drives of the same vintage,
906 * add them in as well. We can't lose and we might gain
907 */
908 for (i = 0;
909 i < content->array.raid_disks * 2 && i < bestcnt ;
910 i += 2) {
911 int j = best[i];
912 if (j >= 0 &&
913 !devices[j].uptodate &&
914 devices[j].i.recovery_start == MaxSector &&
915 devices[j].i.events == current_events) {
916 chosen_drive = j;
917 goto add_another;
918 }
919 }
920 }
921 return okcnt;
922 }
923
924 static int start_array(int mdfd,
925 char *mddev,
926 struct mdinfo *content,
927 struct supertype *st,
928 struct mddev_ident *ident,
929 int *best, int bestcnt,
930 int chosen_drive,
931 struct devs *devices,
932 unsigned int okcnt,
933 unsigned int sparecnt,
934 unsigned int rebuilding_cnt,
935 unsigned int journalcnt,
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 if (content->journal_device_required && (content->journal_clean == 0)) {
948 if (!c->force) {
949 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
950 return 1;
951 }
952 pr_err("Journal is missing or stale, starting array read only.\n");
953 c->readonly = 1;
954 }
955
956 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
957 clean = 1;
958
959 rv = set_array_info(mdfd, st, content);
960 if (rv && !err_ok) {
961 pr_err("failed to set array info for %s: %s\n",
962 mddev, strerror(errno));
963 return 1;
964 }
965 if (ident->bitmap_fd >= 0) {
966 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
967 pr_err("SET_BITMAP_FILE failed.\n");
968 return 1;
969 }
970 } else if (ident->bitmap_file) {
971 /* From config file */
972 int bmfd = open(ident->bitmap_file, O_RDWR);
973 if (bmfd < 0) {
974 pr_err("Could not open bitmap file %s\n",
975 ident->bitmap_file);
976 return 1;
977 }
978 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
979 pr_err("Failed to set bitmapfile for %s\n", mddev);
980 close(bmfd);
981 return 1;
982 }
983 close(bmfd);
984 }
985
986 /* First, add the raid disks, but add the chosen one last */
987 for (i = 0; i <= bestcnt; i++) {
988 int j;
989 if (i < bestcnt) {
990 j = best[i];
991 if (j == chosen_drive)
992 continue;
993 } else
994 j = chosen_drive;
995
996 if (j >= 0 && !devices[j].included) {
997 int dfd;
998
999 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1000 if (dfd >= 0) {
1001 remove_partitions(dfd);
1002 close(dfd);
1003 }
1004 rv = add_disk(mdfd, st, content, &devices[j].i);
1005
1006 if (rv) {
1007 pr_err("failed to add %s to %s: %s\n",
1008 devices[j].devname, mddev,
1009 strerror(errno));
1010 if (i < content->array.raid_disks * 2 ||
1011 i == bestcnt)
1012 okcnt--;
1013 else
1014 sparecnt--;
1015 } else if (c->verbose > 0) {
1016 pr_err("added %s to %s as %d%s%s\n",
1017 devices[j].devname, mddev,
1018 devices[j].i.disk.raid_disk,
1019 devices[j].uptodate?"":
1020 " (possibly out of date)",
1021 (devices[j].i.disk.state &
1022 (1<<MD_DISK_REPLACEMENT)) ?
1023 " replacement":"");
1024 }
1025 } else if (j >= 0) {
1026 if (c->verbose > 0)
1027 pr_err("%s is already in %s as %d\n",
1028 devices[j].devname, mddev,
1029 devices[j].i.disk.raid_disk);
1030 } else if (c->verbose > 0 &&
1031 i < content->array.raid_disks * 2 && (i & 1) == 0)
1032 pr_err("no uptodate device for slot %d of %s\n",
1033 i/2, mddev);
1034 }
1035
1036 if (content->array.level == LEVEL_CONTAINER) {
1037 if (c->verbose >= 0) {
1038 pr_err("Container %s has been assembled with %d drive%s",
1039 mddev, okcnt + sparecnt + journalcnt,
1040 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1041 if (okcnt < (unsigned)content->array.raid_disks)
1042 fprintf(stderr, " (out of %d)",
1043 content->array.raid_disks);
1044 fprintf(stderr, "\n");
1045 }
1046
1047 if (st->ss->validate_container) {
1048 struct mdinfo *devices_list;
1049 struct mdinfo *info_devices;
1050 unsigned int count;
1051
1052 devices_list = NULL;
1053 info_devices = xmalloc(sizeof(struct mdinfo) *
1054 (okcnt + sparecnt));
1055 for (count = 0; count < okcnt + sparecnt; count++) {
1056 info_devices[count] = devices[count].i;
1057 info_devices[count].next = devices_list;
1058 devices_list = &info_devices[count];
1059 }
1060 if (st->ss->validate_container(devices_list))
1061 pr_err("Mismatch detected!\n");
1062 free(info_devices);
1063 }
1064
1065 st->ss->free_super(st);
1066 sysfs_uevent(content, "change");
1067 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1068 /* Was partial, is still partial, so signal an error
1069 * to ensure we don't retry */
1070 return 1;
1071 return 0;
1072 }
1073
1074 /* Get number of in-sync devices according to the superblock.
1075 * We must have this number to start the array without -s or -R
1076 */
1077 req_cnt = content->array.working_disks;
1078
1079 if (c->runstop == 1 ||
1080 (c->runstop <= 0 &&
1081 (enough(content->array.level, content->array.raid_disks,
1082 content->array.layout, clean, avail) &&
1083 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1084 /* This array is good-to-go.
1085 * If a reshape is in progress then we might need to
1086 * continue monitoring it. In that case we start
1087 * it read-only and let the grow code make it writable.
1088 */
1089 int rv;
1090
1091 if (content->reshape_active &&
1092 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1093 content->delta_disks <= 0) {
1094 if (!c->backup_file) {
1095 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1096 mddev);
1097 pr_err("Please provided one with \"--backup-file=...\"\n");
1098 if (c->update &&
1099 strcmp(c->update, "revert-reshape") == 0)
1100 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1101 return 1;
1102 }
1103 rv = sysfs_set_str(content, NULL,
1104 "array_state", "readonly");
1105 if (rv == 0)
1106 rv = Grow_continue(mdfd, st, content,
1107 c->backup_file, 0,
1108 c->freeze_reshape);
1109 } else if (c->readonly &&
1110 sysfs_attribute_available(content, NULL,
1111 "array_state")) {
1112 rv = sysfs_set_str(content, NULL,
1113 "array_state", "readonly");
1114 } else
1115 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1116 reopen_mddev(mdfd); /* drop O_EXCL */
1117 if (rv == 0) {
1118 if (c->verbose >= 0) {
1119 pr_err("%s has been started with %d drive%s",
1120 mddev, okcnt, okcnt==1?"":"s");
1121 if (okcnt < (unsigned)content->array.raid_disks)
1122 fprintf(stderr, " (out of %d)",
1123 content->array.raid_disks);
1124 if (rebuilding_cnt)
1125 fprintf(stderr, "%s %d rebuilding",
1126 sparecnt?",":" and",
1127 rebuilding_cnt);
1128 if (sparecnt)
1129 fprintf(stderr, " and %d spare%s",
1130 sparecnt,
1131 sparecnt == 1 ? "" : "s");
1132 if (content->journal_clean)
1133 fprintf(stderr, " and %d journal",
1134 journalcnt);
1135 fprintf(stderr, ".\n");
1136 }
1137 if (content->reshape_active &&
1138 content->array.level >= 4 &&
1139 content->array.level <= 6) {
1140 /* might need to increase the size
1141 * of the stripe cache - default is 256
1142 */
1143 int chunk_size = content->array.chunk_size;
1144
1145 if (content->reshape_active &&
1146 content->new_chunk > chunk_size)
1147 chunk_size = content->new_chunk;
1148 if (256 < 4 * ((chunk_size+4065)/4096)) {
1149 struct mdinfo *sra;
1150
1151 sra = sysfs_read(mdfd, NULL, 0);
1152 if (sra)
1153 sysfs_set_num(sra, NULL,
1154 "stripe_cache_size",
1155 (4 * chunk_size / 4096) + 1);
1156 sysfs_free(sra);
1157 }
1158 }
1159 if (okcnt < (unsigned)content->array.raid_disks) {
1160 /* If any devices did not get added
1161 * because the kernel rejected them based
1162 * on event count, try adding them
1163 * again providing the action policy is
1164 * 're-add' or greater. The bitmap
1165 * might allow them to be included, or
1166 * they will become spares.
1167 */
1168 for (i = 0; i < bestcnt; i++) {
1169 int j = best[i];
1170 if (j >= 0 && !devices[j].uptodate) {
1171 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1172 continue;
1173 rv = add_disk(mdfd, st, content,
1174 &devices[j].i);
1175 if (rv == 0 && c->verbose >= 0)
1176 pr_err("%s has been re-added.\n",
1177 devices[j].devname);
1178 }
1179 }
1180 }
1181 if (content->array.level == 6 &&
1182 okcnt + 1 == (unsigned)content->array.raid_disks &&
1183 was_forced) {
1184 struct mdinfo *sra;
1185
1186 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", mddev, strerror(errno));
1195
1196 if (!enough(content->array.level, content->array.raid_disks,
1197 content->array.layout, 1, avail))
1198 pr_err("Not enough devices to start the array.\n");
1199 else if (!enough(content->array.level,
1200 content->array.raid_disks,
1201 content->array.layout, clean, avail))
1202 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1203
1204 return 1;
1205 }
1206 if (c->runstop == -1) {
1207 pr_err("%s assembled from %d drive%s",
1208 mddev, okcnt, okcnt == 1 ? "" : "s");
1209 if (okcnt != (unsigned)content->array.raid_disks)
1210 fprintf(stderr, " (out of %d)",
1211 content->array.raid_disks);
1212 fprintf(stderr, ", but not started.\n");
1213 return 2;
1214 }
1215 if (c->verbose >= -1) {
1216 pr_err("%s assembled from %d drive%s",
1217 mddev, okcnt, okcnt == 1 ? "" : "s");
1218 if (rebuilding_cnt)
1219 fprintf(stderr, "%s %d rebuilding",
1220 sparecnt ? "," : " and", rebuilding_cnt);
1221 if (sparecnt)
1222 fprintf(stderr, " and %d spare%s", sparecnt,
1223 sparecnt == 1 ? "" : "s");
1224 if (!enough(content->array.level, content->array.raid_disks,
1225 content->array.layout, 1, avail))
1226 fprintf(stderr, " - not enough to start the array.\n");
1227 else if (!enough(content->array.level,
1228 content->array.raid_disks,
1229 content->array.layout, clean, avail))
1230 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1231 else {
1232 if (req_cnt == (unsigned)content->array.raid_disks)
1233 fprintf(stderr, " - need all %d to start it",
1234 req_cnt);
1235 else
1236 fprintf(stderr, " - need %d to start", req_cnt);
1237 fprintf(stderr, " (use --run to insist).\n");
1238 }
1239 }
1240 return 1;
1241 }
1242
1243 int Assemble(struct supertype *st, char *mddev,
1244 struct mddev_ident *ident,
1245 struct mddev_dev *devlist,
1246 struct context *c)
1247 {
1248 /*
1249 * The task of Assemble is to find a collection of
1250 * devices that should (according to their superblocks)
1251 * form an array, and to give this collection to the MD driver.
1252 * In Linux-2.4 and later, this involves submitting a
1253 * SET_ARRAY_INFO ioctl with no arg - to prepare
1254 * the array - and then submit a number of
1255 * ADD_NEW_DISK ioctls to add disks into
1256 * the array. Finally RUN_ARRAY might
1257 * be submitted to start the array.
1258 *
1259 * Much of the work of Assemble is in finding and/or
1260 * checking the disks to make sure they look right.
1261 *
1262 * If mddev is not set, then scan must be set and we
1263 * read through the config file for dev+uuid mapping
1264 * We recurse, setting mddev, for each device that
1265 * - isn't running
1266 * - has a valid uuid (or any uuid if !uuidset)
1267 *
1268 * If mddev is set, we try to determine state of md.
1269 * check version - must be at least 0.90.0
1270 * check kernel version. must be at least 2.4.
1271 * If not, we can possibly fall back on START_ARRAY
1272 * Try to GET_ARRAY_INFO.
1273 * If possible, give up
1274 * If not, try to STOP_ARRAY just to make sure
1275 *
1276 * If !uuidset and scan, look in conf-file for uuid
1277 * If not found, give up
1278 * If !devlist and scan and uuidset, get list of devs from conf-file
1279 *
1280 * For each device:
1281 * Check superblock - discard if bad
1282 * Check uuid (set if we don't have one) - discard if no match
1283 * Check superblock similarity if we have a superblock - discard if different
1284 * Record events, devicenum
1285 * This should give us a list of devices for the array
1286 * We should collect the most recent event number
1287 *
1288 * Count disks with recent enough event count
1289 * While force && !enough disks
1290 * Choose newest rejected disks, update event count
1291 * mark clean and rewrite superblock
1292 * If recent kernel:
1293 * SET_ARRAY_INFO
1294 * foreach device with recent events : ADD_NEW_DISK
1295 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1296 * If old kernel:
1297 * Check the device numbers in superblock are right
1298 * update superblock if any changes
1299 * START_ARRAY
1300 *
1301 */
1302 int rv;
1303 int mdfd;
1304 int clean;
1305 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1306 ident->super_minor == UnSet && ident->name[0] == 0
1307 && (ident->container == NULL || ident->member == NULL));
1308 struct devs *devices;
1309 char *devmap;
1310 int *best = NULL; /* indexed by raid_disk */
1311 int bestcnt = 0;
1312 int devcnt;
1313 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1314 int journal_clean = 0;
1315 int i;
1316 int was_forced = 0;
1317 int most_recent = 0;
1318 int chosen_drive;
1319 int change = 0;
1320 int inargv = 0;
1321 int start_partial_ok = (c->runstop >= 0) &&
1322 (c->force || devlist==NULL || auto_assem);
1323 int num_devs;
1324 struct mddev_dev *tmpdev;
1325 struct mdinfo info;
1326 struct mdinfo *content = NULL;
1327 struct mdinfo *pre_exist = NULL;
1328 char *avail;
1329 char *name = NULL;
1330 char chosen_name[1024];
1331 struct map_ent *map = NULL;
1332 struct map_ent *mp;
1333
1334 /*
1335 * If any subdevs are listed, then any that don't
1336 * match ident are discarded. Remainder must all match and
1337 * become the array.
1338 * If no subdevs, then we scan all devices in the config file, but
1339 * there must be something in the identity
1340 */
1341
1342 if (!devlist &&
1343 ident->uuid_set == 0 &&
1344 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1345 ident->name[0] == 0 &&
1346 (ident->container == NULL || ident->member == NULL) &&
1347 ident->devices == NULL) {
1348 pr_err("No identity information available for %s - cannot assemble.\n",
1349 mddev ? mddev : "further assembly");
1350 return 1;
1351 }
1352
1353 if (devlist == NULL)
1354 devlist = conf_get_devs();
1355 else if (mddev)
1356 inargv = 1;
1357
1358 try_again:
1359 /* We come back here when doing auto-assembly and attempting some
1360 * set of devices failed. Those are now marked as ->used==2 and
1361 * we ignore them and try again
1362 */
1363 if (!st && ident->st)
1364 st = ident->st;
1365 if (c->verbose>0)
1366 pr_err("looking for devices for %s\n",
1367 mddev ? mddev : "further assembly");
1368
1369 content = &info;
1370 if (st && c->force)
1371 st->ignore_hw_compat = 1;
1372 num_devs = select_devices(devlist, ident, &st, &content, c,
1373 inargv, auto_assem);
1374 if (num_devs < 0)
1375 return 1;
1376
1377 if (!st || !st->sb || !content)
1378 return 2;
1379
1380 /* We have a full set of devices - we now need to find the
1381 * array device.
1382 * However there is a risk that we are racing with "mdadm -I"
1383 * and the array is already partially assembled - we will have
1384 * rejected any devices already in this address.
1385 * So we take a lock on the map file - to prevent further races -
1386 * and look for the uuid in there. If found and the array is
1387 * active, we abort. If found and the array is not active
1388 * we commit to that md device and add all the contained devices
1389 * to our list. We flag them so that we don't try to re-add,
1390 * but can remove if they turn out to not be wanted.
1391 */
1392 if (map_lock(&map))
1393 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1394 if (c->update && strcmp(c->update,"uuid") == 0)
1395 mp = NULL;
1396 else
1397 mp = map_by_uuid(&map, content->uuid);
1398 if (mp) {
1399 struct mdinfo *dv;
1400 /* array already exists. */
1401 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1402 if (pre_exist->array.level != UnSet) {
1403 pr_err("Found some drive for an array that is already active: %s\n",
1404 mp->path);
1405 pr_err("giving up.\n");
1406 return 1;
1407 }
1408 for (dv = pre_exist->devs; dv; dv = dv->next) {
1409 /* We want to add this device to our list,
1410 * but it could already be there if "mdadm -I"
1411 * started *after* we checked for O_EXCL.
1412 * If we add it to the top of the list
1413 * it will be preferred over later copies.
1414 */
1415 struct mddev_dev *newdev;
1416 char *devname = map_dev(dv->disk.major,
1417 dv->disk.minor,
1418 0);
1419 if (!devname)
1420 continue;
1421 newdev = xmalloc(sizeof(*newdev));
1422 newdev->devname = devname;
1423 newdev->disposition = 'I';
1424 newdev->used = 1;
1425 newdev->next = devlist;
1426 devlist = newdev;
1427 num_devs++;
1428 }
1429 strcpy(chosen_name, mp->path);
1430 if (c->verbose > 0 || mddev == NULL ||
1431 strcmp(mddev, chosen_name) != 0)
1432 pr_err("Merging with already-assembled %s\n",
1433 chosen_name);
1434 mdfd = open_dev_excl(mp->devnm);
1435 } else {
1436 int trustworthy = FOREIGN;
1437 name = content->name;
1438 switch (st->ss->match_home(st, c->homehost)
1439 ?: st->ss->match_home(st, "any")) {
1440 case 1:
1441 trustworthy = LOCAL;
1442 name = strchr(content->name, ':');
1443 if (name)
1444 name++;
1445 else
1446 name = content->name;
1447 break;
1448 }
1449 if (!auto_assem)
1450 /* If the array is listed in mdadm.conf or on
1451 * command line, then we trust the name
1452 * even if the array doesn't look local
1453 */
1454 trustworthy = LOCAL;
1455
1456 if (name[0] == 0 &&
1457 content->array.level == LEVEL_CONTAINER) {
1458 name = content->text_version;
1459 trustworthy = METADATA;
1460 }
1461
1462 if (name[0] && trustworthy != LOCAL &&
1463 ! c->require_homehost &&
1464 conf_name_is_free(name))
1465 trustworthy = LOCAL;
1466
1467 if (trustworthy == LOCAL &&
1468 strchr(name, ':'))
1469 /* Ignore 'host:' prefix of name */
1470 name = strchr(name, ':')+1;
1471
1472 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1473 chosen_name, 0);
1474 }
1475 if (mdfd < 0) {
1476 st->ss->free_super(st);
1477 if (auto_assem)
1478 goto try_again;
1479 return 1;
1480 }
1481 mddev = chosen_name;
1482 if (pre_exist == NULL) {
1483 if (mddev_busy(fd2devnm(mdfd))) {
1484 pr_err("%s already active, cannot restart it!\n",
1485 mddev);
1486 for (tmpdev = devlist ;
1487 tmpdev && tmpdev->used != 1;
1488 tmpdev = tmpdev->next)
1489 ;
1490 if (tmpdev && auto_assem)
1491 pr_err("%s needed for %s...\n",
1492 mddev, tmpdev->devname);
1493 close(mdfd);
1494 mdfd = -3;
1495 st->ss->free_super(st);
1496 if (auto_assem)
1497 goto try_again;
1498 return 1;
1499 }
1500 /* just incase it was started but has no content */
1501 ioctl(mdfd, STOP_ARRAY, NULL);
1502 }
1503
1504 if (content != &info) {
1505 /* This is a member of a container. Try starting the array. */
1506 int err;
1507 err = assemble_container_content(st, mdfd, content, c,
1508 chosen_name, NULL);
1509 close(mdfd);
1510 return err;
1511 }
1512
1513 /* Ok, no bad inconsistancy, we can try updating etc */
1514 devices = xcalloc(num_devs, sizeof(*devices));
1515 devmap = xcalloc(num_devs, content->array.raid_disks);
1516 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1517 c, content, mdfd, mddev,
1518 &most_recent, &bestcnt, &best, inargv);
1519 if (devcnt < 0)
1520 return 1;
1521
1522 if (devcnt == 0) {
1523 pr_err("no devices found for %s\n",
1524 mddev);
1525 if (st)
1526 st->ss->free_super(st);
1527 close(mdfd);
1528 free(devices);
1529 free(devmap);
1530 return 1;
1531 }
1532
1533 if (c->update && strcmp(c->update, "byteorder")==0)
1534 st->minor_version = 90;
1535
1536 st->ss->getinfo_super(st, content, NULL);
1537 clean = content->array.state & 1;
1538
1539 /* now we have some devices that might be suitable.
1540 * I wonder how many
1541 */
1542 avail = xcalloc(content->array.raid_disks, 1);
1543 okcnt = 0;
1544 replcnt = 0;
1545 sparecnt=0;
1546 journalcnt=0;
1547 rebuilding_cnt=0;
1548 for (i=0; i< bestcnt; i++) {
1549 int j = best[i];
1550 int event_margin = 1; /* always allow a difference of '1'
1551 * like the kernel does
1552 */
1553 if (j < 0) continue;
1554 /* note: we ignore error flags in multipath arrays
1555 * as they don't make sense
1556 */
1557 if (content->array.level != LEVEL_MULTIPATH) {
1558 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1559 if (content->journal_device_required)
1560 journalcnt++;
1561 else /* unexpected journal, mark as faulty */
1562 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1563 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1564 if (!(devices[j].i.disk.state
1565 & (1<<MD_DISK_FAULTY))) {
1566 devices[j].uptodate = 1;
1567 sparecnt++;
1568 }
1569 continue;
1570 }
1571 }
1572 /* If this device thinks that 'most_recent' has failed, then
1573 * we must reject this device.
1574 */
1575 if (j != most_recent && !c->force &&
1576 content->array.raid_disks > 0 &&
1577 devices[most_recent].i.disk.raid_disk >= 0 &&
1578 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1579 if (c->verbose > -1)
1580 pr_err("ignoring %s as it reports %s as failed\n",
1581 devices[j].devname, devices[most_recent].devname);
1582 best[i] = -1;
1583 continue;
1584 }
1585 /* Require event counter to be same as, or just less than,
1586 * most recent. If it is bigger, it must be a stray spare and
1587 * should be ignored.
1588 */
1589 if (devices[j].i.events+event_margin >=
1590 devices[most_recent].i.events &&
1591 devices[j].i.events <=
1592 devices[most_recent].i.events
1593 ) {
1594 devices[j].uptodate = 1;
1595 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1596 journal_clean = 1;
1597 if (i < content->array.raid_disks * 2) {
1598 if (devices[j].i.recovery_start == MaxSector ||
1599 (content->reshape_active &&
1600 i >= content->array.raid_disks - content->delta_disks)) {
1601 if (!avail[i/2]) {
1602 okcnt++;
1603 avail[i/2]=1;
1604 } else
1605 replcnt++;
1606 } else
1607 rebuilding_cnt++;
1608 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1609 sparecnt++;
1610 }
1611 }
1612 free(devmap);
1613 if (c->force) {
1614 int force_ok = force_array(content, devices, best, bestcnt,
1615 avail, most_recent, st, c);
1616 okcnt += force_ok;
1617 if (force_ok)
1618 was_forced = 1;
1619 }
1620 /* Now we want to look at the superblock which the kernel will base things on
1621 * and compare the devices that we think are working with the devices that the
1622 * superblock thinks are working.
1623 * If there are differences and --force is given, then update this chosen
1624 * superblock.
1625 */
1626 chosen_drive = -1;
1627 st->ss->free_super(st);
1628 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1629 int j = best[i];
1630 int fd;
1631
1632 if (j<0)
1633 continue;
1634 if (!devices[j].uptodate)
1635 continue;
1636 if (devices[j].i.events < devices[most_recent].i.events)
1637 continue;
1638 chosen_drive = j;
1639 if ((fd=dev_open(devices[j].devname,
1640 devices[j].included ? O_RDONLY
1641 : (O_RDONLY|O_EXCL)))< 0) {
1642 pr_err("Cannot open %s: %s\n",
1643 devices[j].devname, strerror(errno));
1644 close(mdfd);
1645 free(devices);
1646 return 1;
1647 }
1648 if (st->ss->load_super(st,fd, NULL)) {
1649 close(fd);
1650 pr_err("RAID superblock has disappeared from %s\n",
1651 devices[j].devname);
1652 close(mdfd);
1653 free(devices);
1654 return 1;
1655 }
1656 close(fd);
1657 }
1658 if (st->sb == NULL) {
1659 pr_err("No suitable drives found for %s\n", mddev);
1660 close(mdfd);
1661 free(devices);
1662 return 1;
1663 }
1664 st->ss->getinfo_super(st, content, NULL);
1665 if (sysfs_init(content, mdfd, NULL)) {
1666 pr_err("Unable to initialize sysfs\n");
1667 close(mdfd);
1668 free(devices);
1669 return 1;
1670 }
1671
1672 /* after reload context, store journal_clean in context */
1673 content->journal_clean = journal_clean;
1674 for (i=0; i<bestcnt; i++) {
1675 int j = best[i];
1676 unsigned int desired_state;
1677
1678 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1679 desired_state = (1<<MD_DISK_JOURNAL);
1680 else if (i >= content->array.raid_disks * 2)
1681 desired_state = 0;
1682 else if (i & 1)
1683 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1684 else
1685 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1686
1687 if (j<0)
1688 continue;
1689 if (!devices[j].uptodate)
1690 continue;
1691
1692 devices[j].i.disk.state = desired_state;
1693 if (!(devices[j].i.array.state & 1))
1694 clean = 0;
1695
1696 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1697 c->verbose, 0, NULL)) {
1698 if (c->force) {
1699 if (c->verbose >= 0)
1700 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1701 j, mddev, devices[j].devname);
1702 change = 1;
1703 } else {
1704 if (c->verbose >= -1)
1705 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1706 i, mddev, devices[j].devname);
1707 }
1708 }
1709 #if 0
1710 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1711 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1712 i, mddev);
1713 }
1714 #endif
1715 }
1716 if (c->force && !clean &&
1717 !enough(content->array.level, content->array.raid_disks,
1718 content->array.layout, clean,
1719 avail)) {
1720 change += st->ss->update_super(st, content, "force-array",
1721 devices[chosen_drive].devname, c->verbose,
1722 0, NULL);
1723 was_forced = 1;
1724 clean = 1;
1725 }
1726
1727 if (change) {
1728 int fd;
1729 fd = dev_open(devices[chosen_drive].devname,
1730 devices[chosen_drive].included ?
1731 O_RDWR : (O_RDWR|O_EXCL));
1732 if (fd < 0) {
1733 pr_err("Could not open %s for write - cannot Assemble array.\n",
1734 devices[chosen_drive].devname);
1735 close(mdfd);
1736 free(devices);
1737 return 1;
1738 }
1739 if (st->ss->store_super(st, fd)) {
1740 close(fd);
1741 pr_err("Could not re-write superblock on %s\n",
1742 devices[chosen_drive].devname);
1743 close(mdfd);
1744 free(devices);
1745 return 1;
1746 }
1747 if (c->verbose >= 0)
1748 pr_err("Marking array %s as 'clean'\n",
1749 mddev);
1750 close(fd);
1751 }
1752
1753 /* If we are in the middle of a reshape we may need to restore saved data
1754 * that was moved aside due to the reshape overwriting live data
1755 * The code of doing this lives in Grow.c
1756 */
1757 if (content->reshape_active &&
1758 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1759 int err = 0;
1760 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1761 if (c->verbose > 0)
1762 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1763 chosen_name);
1764 if (!c->backup_file)
1765 c->backup_file = locate_backup(content->sys_name);
1766 enable_fds(bestcnt/2);
1767 for (i = 0; i < bestcnt/2; i++) {
1768 int j = best[i*2];
1769 if (j >= 0) {
1770 fdlist[i] = dev_open(devices[j].devname,
1771 devices[j].included
1772 ? O_RDWR : (O_RDWR|O_EXCL));
1773 if (fdlist[i] < 0) {
1774 pr_err("Could not open %s for write - cannot Assemble array.\n",
1775 devices[j].devname);
1776 err = 1;
1777 break;
1778 }
1779 } else
1780 fdlist[i] = -1;
1781 }
1782 if (!err) {
1783 if (st->ss->external && st->ss->recover_backup)
1784 err = st->ss->recover_backup(st, content);
1785 else
1786 err = Grow_restart(st, content, fdlist, bestcnt/2,
1787 c->backup_file, c->verbose > 0);
1788 if (err && c->invalid_backup) {
1789 if (c->verbose > 0)
1790 pr_err("continuing without restoring backup\n");
1791 err = 0;
1792 }
1793 }
1794 while (i>0) {
1795 i--;
1796 if (fdlist[i]>=0) close(fdlist[i]);
1797 }
1798 free(fdlist);
1799 if (err) {
1800 pr_err("Failed to restore critical section for reshape, sorry.\n");
1801 if (c->backup_file == NULL)
1802 cont_err("Possibly you needed to specify the --backup-file\n");
1803 close(mdfd);
1804 free(devices);
1805 return err;
1806 }
1807 }
1808
1809 /* Almost ready to actually *do* something */
1810 /* First, fill in the map, so that udev can find our name
1811 * as soon as we become active.
1812 */
1813 if (c->update && strcmp(c->update, "metadata")==0) {
1814 content->array.major_version = 1;
1815 content->array.minor_version = 0;
1816 strcpy(content->text_version, "1.0");
1817 }
1818
1819 map_update(&map, fd2devnm(mdfd), content->text_version,
1820 content->uuid, chosen_name);
1821
1822 rv = start_array(mdfd, mddev, content,
1823 st, ident, best, bestcnt,
1824 chosen_drive, devices, okcnt, sparecnt,
1825 rebuilding_cnt, journalcnt,
1826 c,
1827 clean, avail, start_partial_ok,
1828 pre_exist != NULL,
1829 was_forced);
1830 if (rv == 1 && !pre_exist)
1831 ioctl(mdfd, STOP_ARRAY, NULL);
1832 free(devices);
1833 map_unlock(&map);
1834 if (rv == 0) {
1835 wait_for(chosen_name, mdfd);
1836 close(mdfd);
1837 if (auto_assem) {
1838 int usecs = 1;
1839 /* There is a nasty race with 'mdadm --monitor'.
1840 * If it opens this device before we close it,
1841 * it gets an incomplete open on which IO
1842 * doesn't work and the capacity is
1843 * wrong.
1844 * If we reopen (to check for layered devices)
1845 * before --monitor closes, we loose.
1846 *
1847 * So: wait upto 1 second for there to be
1848 * a non-zero capacity.
1849 */
1850 while (usecs < 1000) {
1851 mdfd = open(mddev, O_RDONLY);
1852 if (mdfd >= 0) {
1853 unsigned long long size;
1854 if (get_dev_size(mdfd, NULL, &size) &&
1855 size > 0)
1856 break;
1857 close(mdfd);
1858 }
1859 usleep(usecs);
1860 usecs <<= 1;
1861 }
1862 }
1863 } else
1864 close(mdfd);
1865
1866 /* '2' means 'OK, but not started yet' */
1867 return rv == 2 ? 0 : rv;
1868 }
1869
1870 int assemble_container_content(struct supertype *st, int mdfd,
1871 struct mdinfo *content, struct context *c,
1872 char *chosen_name, int *result)
1873 {
1874 struct mdinfo *dev, *sra, *dev2;
1875 int working = 0, preexist = 0;
1876 int expansion = 0;
1877 int old_raid_disks;
1878 int start_reshape;
1879 char *avail;
1880 int err;
1881
1882 if (sysfs_init(content, mdfd, NULL)) {
1883 pr_err("Unable to initialize sysfs\n");
1884 return 1;
1885 }
1886
1887 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1888 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1889 if (content->array.major_version == -1 &&
1890 content->array.minor_version == -2 &&
1891 c->readonly &&
1892 content->text_version[0] == '/')
1893 content->text_version[0] = '-';
1894 if (sysfs_set_array(content, 9003) != 0) {
1895 sysfs_free(sra);
1896 return 1;
1897 }
1898 }
1899
1900 /* There are two types of reshape: container wide or sub-array specific
1901 * Check if metadata requests blocking container wide reshapes
1902 */
1903 start_reshape = (content->reshape_active &&
1904 !((content->reshape_active == CONTAINER_RESHAPE) &&
1905 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1906
1907 /* Block subarray here if it is under reshape now
1908 * Do not allow for any changes in this array
1909 */
1910 if (st->ss->external && content->recovery_blocked && start_reshape)
1911 block_subarray(content);
1912
1913 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1914 for (dev = content->devs; dev; dev = dev->next)
1915 if (dev2->disk.major == dev->disk.major &&
1916 dev2->disk.minor == dev->disk.minor)
1917 break;
1918 if (dev)
1919 continue;
1920 /* Don't want this one any more */
1921 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1922 errno == EBUSY) {
1923 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1924 sysfs_free(sra);
1925 return 1;
1926 }
1927 sysfs_set_str(sra, dev2, "state", "remove");
1928 }
1929 old_raid_disks = content->array.raid_disks - content->delta_disks;
1930 avail = xcalloc(content->array.raid_disks, 1);
1931 for (dev = content->devs; dev; dev = dev->next) {
1932 if (dev->disk.raid_disk >= 0)
1933 avail[dev->disk.raid_disk] = 1;
1934 if (sysfs_add_disk(content, dev, 1) == 0) {
1935 if (dev->disk.raid_disk >= old_raid_disks &&
1936 content->reshape_active)
1937 expansion++;
1938 else
1939 working++;
1940 } else if (errno == EEXIST)
1941 preexist++;
1942 }
1943 sysfs_free(sra);
1944 if (working + expansion == 0 && c->runstop <= 0) {
1945 free(avail);
1946 return 1;/* Nothing new, don't try to start */
1947 }
1948 map_update(NULL, fd2devnm(mdfd), content->text_version,
1949 content->uuid, chosen_name);
1950
1951 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1952 st->ss->validate_ppl) {
1953 content->array.state |= 1;
1954 err = 0;
1955
1956 for (dev = content->devs; dev; dev = dev->next) {
1957 int dfd;
1958 char *devpath;
1959 int ret;
1960
1961 ret = st->ss->validate_ppl(st, content, dev);
1962 if (ret == 0)
1963 continue;
1964
1965 if (ret < 0) {
1966 err = 1;
1967 break;
1968 }
1969
1970 if (!c->force) {
1971 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1972 chosen_name);
1973 content->array.state &= ~1;
1974 avail[dev->disk.raid_disk] = 0;
1975 break;
1976 }
1977
1978 /* have --force - overwrite the invalid ppl */
1979 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1980 dfd = dev_open(devpath, O_RDWR);
1981 if (dfd < 0) {
1982 pr_err("Failed to open %s\n", devpath);
1983 err = 1;
1984 break;
1985 }
1986
1987 err = st->ss->write_init_ppl(st, content, dfd);
1988 close(dfd);
1989
1990 if (err)
1991 break;
1992 }
1993
1994 if (err) {
1995 free(avail);
1996 return err;
1997 }
1998 }
1999
2000 if (enough(content->array.level, content->array.raid_disks,
2001 content->array.layout, content->array.state & 1, avail) == 0) {
2002 if (c->export && result)
2003 *result |= INCR_NO;
2004 else if (c->verbose >= 0) {
2005 pr_err("%s assembled with %d device%s",
2006 chosen_name, preexist + working,
2007 preexist + working == 1 ? "":"s");
2008 if (preexist)
2009 fprintf(stderr, " (%d new)", working);
2010 fprintf(stderr, " but not started\n");
2011 }
2012 free(avail);
2013 return 1;
2014 }
2015 free(avail);
2016
2017 if (c->runstop <= 0 &&
2018 (working + preexist + expansion) <
2019 content->array.working_disks) {
2020 if (c->export && result)
2021 *result |= INCR_UNSAFE;
2022 else if (c->verbose >= 0) {
2023 pr_err("%s assembled with %d device%s",
2024 chosen_name, preexist + working,
2025 preexist + working == 1 ? "":"s");
2026 if (preexist)
2027 fprintf(stderr, " (%d new)", working);
2028 fprintf(stderr, " but not safe to start\n");
2029 }
2030 return 1;
2031 }
2032
2033
2034 if (start_reshape) {
2035 int spare = content->array.raid_disks + expansion;
2036 if (restore_backup(st, content,
2037 working,
2038 spare, &c->backup_file, c->verbose) == 1)
2039 return 1;
2040
2041 err = sysfs_set_str(content, NULL,
2042 "array_state", "readonly");
2043 if (err)
2044 return 1;
2045
2046 if (st->ss->external) {
2047 if (!mdmon_running(st->container_devnm))
2048 start_mdmon(st->container_devnm);
2049 ping_monitor(st->container_devnm);
2050 if (mdmon_running(st->container_devnm) &&
2051 st->update_tail == NULL)
2052 st->update_tail = &st->updates;
2053 }
2054
2055 err = Grow_continue(mdfd, st, content, c->backup_file,
2056 0, c->freeze_reshape);
2057 } else switch(content->array.level) {
2058 case LEVEL_LINEAR:
2059 case LEVEL_MULTIPATH:
2060 case 0:
2061 err = sysfs_set_str(content, NULL, "array_state",
2062 c->readonly ? "readonly" : "active");
2063 break;
2064 default:
2065 err = sysfs_set_str(content, NULL, "array_state",
2066 "readonly");
2067 /* start mdmon if needed. */
2068 if (!err) {
2069 if (!mdmon_running(st->container_devnm))
2070 start_mdmon(st->container_devnm);
2071 ping_monitor(st->container_devnm);
2072 }
2073 break;
2074 }
2075 if (!err)
2076 sysfs_set_safemode(content, content->safe_mode_delay);
2077
2078 /* Block subarray here if it is not reshaped now
2079 * It has be blocked a little later to allow mdmon to switch in
2080 * in to R/W state
2081 */
2082 if (st->ss->external && content->recovery_blocked &&
2083 !start_reshape)
2084 block_subarray(content);
2085
2086 if (c->export && result) {
2087 if (err)
2088 *result |= INCR_NO;
2089 else
2090 *result |= INCR_YES;
2091 } else if (c->verbose >= 0) {
2092 if (err)
2093 pr_err("array %s now has %d device%s",
2094 chosen_name, working + preexist,
2095 working + preexist == 1 ? "":"s");
2096 else
2097 pr_err("Started %s with %d device%s",
2098 chosen_name, working + preexist,
2099 working + preexist == 1 ? "":"s");
2100 if (preexist)
2101 fprintf(stderr, " (%d new)", working);
2102 if (expansion)
2103 fprintf(stderr, " ( + %d for expansion)",
2104 expansion);
2105 fprintf(stderr, "\n");
2106 }
2107 if (!err)
2108 wait_for(chosen_name, mdfd);
2109 return err;
2110 /* FIXME should have an O_EXCL and wait for read-auto */
2111 }