]> 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: Fixup a large number of bad formatting of logical operators
[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 (content->reshape_active && content->delta_disks > 0 &&
806 !enough(content->array.level, (content->array.raid_disks
807 - content->delta_disks),
808 content->new_layout, 1, avail))) {
809 /* Choose the newest best drive which is
810 * not up-to-date, update the superblock
811 * and add it.
812 */
813 int fd;
814 struct supertype *tst;
815 unsigned long long current_events;
816 int chosen_drive = -1;
817 int i;
818
819 for (i = 0;
820 i < content->array.raid_disks * 2 && i < bestcnt;
821 i += 2) {
822 int j = best[i];
823 if (j < 0)
824 continue;
825 if (devices[j].uptodate)
826 continue;
827 if (devices[j].i.recovery_start != MaxSector) {
828 int delta;
829 if (!devices[j].i.reshape_active ||
830 devices[j].i.delta_disks <= 0)
831 continue;
832 /* When increasing number of devices, an
833 * added device also appears to be
834 * recovering. It is safe to include it
835 * as long as it won't be a source of
836 * data.
837 * For now, just allow for last data
838 * devices in RAID4 or last devices in RAID4/5/6.
839 */
840 delta = devices[j].i.delta_disks;
841 if (devices[j].i.array.level >= 4 &&
842 devices[j].i.array.level <= 6 &&
843 i/2 >= content->array.raid_disks - delta)
844 /* OK */;
845 else if (devices[j].i.array.level == 4 &&
846 i/2 >= content->array.raid_disks - delta - 1)
847 /* OK */;
848 else
849 continue;
850 }
851 if (chosen_drive < 0 ||
852 devices[j].i.events
853 > devices[chosen_drive].i.events)
854 chosen_drive = j;
855 }
856 if (chosen_drive < 0)
857 break;
858 current_events = devices[chosen_drive].i.events;
859 add_another:
860 if (c->verbose >= 0)
861 pr_err("forcing event count in %s(%d) from %d upto %d\n",
862 devices[chosen_drive].devname,
863 devices[chosen_drive].i.disk.raid_disk,
864 (int)(devices[chosen_drive].i.events),
865 (int)(devices[most_recent].i.events));
866 fd = dev_open(devices[chosen_drive].devname,
867 devices[chosen_drive].included ? O_RDWR
868 : (O_RDWR|O_EXCL));
869 if (fd < 0) {
870 pr_err("Couldn't open %s for write - not updating\n",
871 devices[chosen_drive].devname);
872 devices[chosen_drive].i.events = 0;
873 continue;
874 }
875 tst = dup_super(st);
876 if (tst->ss->load_super(tst,fd, NULL)) {
877 close(fd);
878 pr_err("RAID superblock disappeared from %s - not updating.\n",
879 devices[chosen_drive].devname);
880 devices[chosen_drive].i.events = 0;
881 continue;
882 }
883 content->events = devices[most_recent].i.events;
884 tst->ss->update_super(tst, content, "force-one",
885 devices[chosen_drive].devname, c->verbose,
886 0, NULL);
887
888 if (tst->ss->store_super(tst, fd)) {
889 close(fd);
890 pr_err("Could not re-write superblock on %s\n",
891 devices[chosen_drive].devname);
892 devices[chosen_drive].i.events = 0;
893 tst->ss->free_super(tst);
894 continue;
895 }
896 close(fd);
897 devices[chosen_drive].i.events = devices[most_recent].i.events;
898 devices[chosen_drive].uptodate = 1;
899 avail[chosen_drive] = 1;
900 okcnt++;
901 tst->ss->free_super(tst);
902 /* If there are any other drives of the same vintage,
903 * add them in as well. We can't lose and we might gain
904 */
905 for (i = 0;
906 i < content->array.raid_disks * 2 && i < bestcnt ;
907 i += 2) {
908 int j = best[i];
909 if (j >= 0 &&
910 !devices[j].uptodate &&
911 devices[j].i.recovery_start == MaxSector &&
912 devices[j].i.events == current_events) {
913 chosen_drive = j;
914 goto add_another;
915 }
916 }
917 }
918 return okcnt;
919 }
920
921 static int start_array(int mdfd,
922 char *mddev,
923 struct mdinfo *content,
924 struct supertype *st,
925 struct mddev_ident *ident,
926 int *best, int bestcnt,
927 int chosen_drive,
928 struct devs *devices,
929 unsigned int okcnt,
930 unsigned int sparecnt,
931 unsigned int rebuilding_cnt,
932 unsigned int journalcnt,
933 struct context *c,
934 int clean, char *avail,
935 int start_partial_ok,
936 int err_ok,
937 int was_forced
938 )
939 {
940 int rv;
941 int i;
942 unsigned int req_cnt;
943
944 if (content->journal_device_required && (content->journal_clean == 0)) {
945 if (!c->force) {
946 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
947 return 1;
948 }
949 pr_err("Journal is missing or stale, starting array read only.\n");
950 c->readonly = 1;
951 }
952
953 if (content->consistency_policy == CONSISTENCY_POLICY_PPL)
954 clean = 1;
955
956 rv = set_array_info(mdfd, st, content);
957 if (rv && !err_ok) {
958 pr_err("failed to set array info for %s: %s\n",
959 mddev, strerror(errno));
960 return 1;
961 }
962 if (ident->bitmap_fd >= 0) {
963 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
964 pr_err("SET_BITMAP_FILE failed.\n");
965 return 1;
966 }
967 } else if (ident->bitmap_file) {
968 /* From config file */
969 int bmfd = open(ident->bitmap_file, O_RDWR);
970 if (bmfd < 0) {
971 pr_err("Could not open bitmap file %s\n",
972 ident->bitmap_file);
973 return 1;
974 }
975 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
976 pr_err("Failed to set bitmapfile for %s\n", mddev);
977 close(bmfd);
978 return 1;
979 }
980 close(bmfd);
981 }
982
983 /* First, add the raid disks, but add the chosen one last */
984 for (i = 0; i <= bestcnt; i++) {
985 int j;
986 if (i < bestcnt) {
987 j = best[i];
988 if (j == chosen_drive)
989 continue;
990 } else
991 j = chosen_drive;
992
993 if (j >= 0 && !devices[j].included) {
994 int dfd;
995
996 dfd = dev_open(devices[j].devname, O_RDWR|O_EXCL);
997 if (dfd >= 0) {
998 remove_partitions(dfd);
999 close(dfd);
1000 }
1001 rv = add_disk(mdfd, st, content, &devices[j].i);
1002
1003 if (rv) {
1004 pr_err("failed to add %s to %s: %s\n",
1005 devices[j].devname, mddev,
1006 strerror(errno));
1007 if (i < content->array.raid_disks * 2 ||
1008 i == bestcnt)
1009 okcnt--;
1010 else
1011 sparecnt--;
1012 } else if (c->verbose > 0) {
1013 pr_err("added %s to %s as %d%s%s\n",
1014 devices[j].devname, mddev,
1015 devices[j].i.disk.raid_disk,
1016 devices[j].uptodate?"":
1017 " (possibly out of date)",
1018 (devices[j].i.disk.state &
1019 (1<<MD_DISK_REPLACEMENT)) ?
1020 " replacement":"");
1021 }
1022 } else if (j >= 0) {
1023 if (c->verbose > 0)
1024 pr_err("%s is already in %s as %d\n",
1025 devices[j].devname, mddev,
1026 devices[j].i.disk.raid_disk);
1027 } else if (c->verbose > 0 &&
1028 i < content->array.raid_disks * 2 && (i & 1) == 0)
1029 pr_err("no uptodate device for slot %d of %s\n",
1030 i/2, mddev);
1031 }
1032
1033 if (content->array.level == LEVEL_CONTAINER) {
1034 if (c->verbose >= 0) {
1035 pr_err("Container %s has been assembled with %d drive%s",
1036 mddev, okcnt + sparecnt + journalcnt,
1037 okcnt + sparecnt + journalcnt == 1 ? "" : "s");
1038 if (okcnt < (unsigned)content->array.raid_disks)
1039 fprintf(stderr, " (out of %d)",
1040 content->array.raid_disks);
1041 fprintf(stderr, "\n");
1042 }
1043
1044 if (st->ss->validate_container) {
1045 struct mdinfo *devices_list;
1046 struct mdinfo *info_devices;
1047 unsigned int count;
1048
1049 devices_list = NULL;
1050 info_devices = xmalloc(sizeof(struct mdinfo) *
1051 (okcnt + sparecnt));
1052 for (count = 0; count < okcnt + sparecnt; count++) {
1053 info_devices[count] = devices[count].i;
1054 info_devices[count].next = devices_list;
1055 devices_list = &info_devices[count];
1056 }
1057 if (st->ss->validate_container(devices_list))
1058 pr_err("Mismatch detected!\n");
1059 free(info_devices);
1060 }
1061
1062 st->ss->free_super(st);
1063 sysfs_uevent(content, "change");
1064 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1065 /* Was partial, is still partial, so signal an error
1066 * to ensure we don't retry */
1067 return 1;
1068 return 0;
1069 }
1070
1071 /* Get number of in-sync devices according to the superblock.
1072 * We must have this number to start the array without -s or -R
1073 */
1074 req_cnt = content->array.working_disks;
1075
1076 if (c->runstop == 1 ||
1077 (c->runstop <= 0 &&
1078 (enough(content->array.level, content->array.raid_disks,
1079 content->array.layout, clean, avail) &&
1080 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)))) {
1081 /* This array is good-to-go.
1082 * If a reshape is in progress then we might need to
1083 * continue monitoring it. In that case we start
1084 * it read-only and let the grow code make it writable.
1085 */
1086 int rv;
1087
1088 if (content->reshape_active &&
1089 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1090 content->delta_disks <= 0) {
1091 if (!c->backup_file) {
1092 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1093 mddev);
1094 pr_err("Please provided one with \"--backup-file=...\"\n");
1095 if (c->update &&
1096 strcmp(c->update, "revert-reshape") == 0)
1097 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1098 return 1;
1099 }
1100 rv = sysfs_set_str(content, NULL,
1101 "array_state", "readonly");
1102 if (rv == 0)
1103 rv = Grow_continue(mdfd, st, content,
1104 c->backup_file, 0,
1105 c->freeze_reshape);
1106 } else if (c->readonly &&
1107 sysfs_attribute_available(content, NULL,
1108 "array_state")) {
1109 rv = sysfs_set_str(content, NULL,
1110 "array_state", "readonly");
1111 } else
1112 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1113 reopen_mddev(mdfd); /* drop O_EXCL */
1114 if (rv == 0) {
1115 if (c->verbose >= 0) {
1116 pr_err("%s has been started with %d drive%s",
1117 mddev, okcnt, okcnt==1?"":"s");
1118 if (okcnt < (unsigned)content->array.raid_disks)
1119 fprintf(stderr, " (out of %d)",
1120 content->array.raid_disks);
1121 if (rebuilding_cnt)
1122 fprintf(stderr, "%s %d rebuilding",
1123 sparecnt?",":" and",
1124 rebuilding_cnt);
1125 if (sparecnt)
1126 fprintf(stderr, " and %d spare%s",
1127 sparecnt,
1128 sparecnt == 1 ? "" : "s");
1129 if (content->journal_clean)
1130 fprintf(stderr, " and %d journal",
1131 journalcnt);
1132 fprintf(stderr, ".\n");
1133 }
1134 if (content->reshape_active &&
1135 content->array.level >= 4 &&
1136 content->array.level <= 6) {
1137 /* might need to increase the size
1138 * of the stripe cache - default is 256
1139 */
1140 int chunk_size = content->array.chunk_size;
1141
1142 if (content->reshape_active &&
1143 content->new_chunk > chunk_size)
1144 chunk_size = content->new_chunk;
1145 if (256 < 4 * ((chunk_size+4065)/4096)) {
1146 struct mdinfo *sra;
1147
1148 sra = sysfs_read(mdfd, NULL, 0);
1149 if (sra)
1150 sysfs_set_num(sra, NULL,
1151 "stripe_cache_size",
1152 (4 * chunk_size / 4096) + 1);
1153 sysfs_free(sra);
1154 }
1155 }
1156 if (okcnt < (unsigned)content->array.raid_disks) {
1157 /* If any devices did not get added
1158 * because the kernel rejected them based
1159 * on event count, try adding them
1160 * again providing the action policy is
1161 * 're-add' or greater. The bitmap
1162 * might allow them to be included, or
1163 * they will become spares.
1164 */
1165 for (i = 0; i < bestcnt; i++) {
1166 int j = best[i];
1167 if (j >= 0 && !devices[j].uptodate) {
1168 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1169 continue;
1170 rv = add_disk(mdfd, st, content,
1171 &devices[j].i);
1172 if (rv == 0 && c->verbose >= 0)
1173 pr_err("%s has been re-added.\n",
1174 devices[j].devname);
1175 }
1176 }
1177 }
1178 if (content->array.level == 6 &&
1179 okcnt + 1 == (unsigned)content->array.raid_disks &&
1180 was_forced) {
1181 struct mdinfo *sra;
1182
1183 sra = sysfs_read(mdfd, NULL, 0);
1184 if (sra)
1185 sysfs_set_str(sra, NULL,
1186 "sync_action", "repair");
1187 sysfs_free(sra);
1188 }
1189 return 0;
1190 }
1191 pr_err("failed to RUN_ARRAY %s: %s\n", mddev, strerror(errno));
1192
1193 if (!enough(content->array.level, content->array.raid_disks,
1194 content->array.layout, 1, avail))
1195 pr_err("Not enough devices to start the array.\n");
1196 else if (!enough(content->array.level,
1197 content->array.raid_disks,
1198 content->array.layout, clean, avail))
1199 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1200
1201 return 1;
1202 }
1203 if (c->runstop == -1) {
1204 pr_err("%s assembled from %d drive%s",
1205 mddev, okcnt, okcnt == 1 ? "" : "s");
1206 if (okcnt != (unsigned)content->array.raid_disks)
1207 fprintf(stderr, " (out of %d)",
1208 content->array.raid_disks);
1209 fprintf(stderr, ", but not started.\n");
1210 return 2;
1211 }
1212 if (c->verbose >= -1) {
1213 pr_err("%s assembled from %d drive%s",
1214 mddev, okcnt, okcnt == 1 ? "" : "s");
1215 if (rebuilding_cnt)
1216 fprintf(stderr, "%s %d rebuilding",
1217 sparecnt ? "," : " and", rebuilding_cnt);
1218 if (sparecnt)
1219 fprintf(stderr, " and %d spare%s", sparecnt,
1220 sparecnt == 1 ? "" : "s");
1221 if (!enough(content->array.level, content->array.raid_disks,
1222 content->array.layout, 1, avail))
1223 fprintf(stderr, " - not enough to start the array.\n");
1224 else if (!enough(content->array.level,
1225 content->array.raid_disks,
1226 content->array.layout, clean, avail))
1227 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1228 else {
1229 if (req_cnt == (unsigned)content->array.raid_disks)
1230 fprintf(stderr, " - need all %d to start it",
1231 req_cnt);
1232 else
1233 fprintf(stderr, " - need %d to start", req_cnt);
1234 fprintf(stderr, " (use --run to insist).\n");
1235 }
1236 }
1237 return 1;
1238 }
1239
1240 int Assemble(struct supertype *st, char *mddev,
1241 struct mddev_ident *ident,
1242 struct mddev_dev *devlist,
1243 struct context *c)
1244 {
1245 /*
1246 * The task of Assemble is to find a collection of
1247 * devices that should (according to their superblocks)
1248 * form an array, and to give this collection to the MD driver.
1249 * In Linux-2.4 and later, this involves submitting a
1250 * SET_ARRAY_INFO ioctl with no arg - to prepare
1251 * the array - and then submit a number of
1252 * ADD_NEW_DISK ioctls to add disks into
1253 * the array. Finally RUN_ARRAY might
1254 * be submitted to start the array.
1255 *
1256 * Much of the work of Assemble is in finding and/or
1257 * checking the disks to make sure they look right.
1258 *
1259 * If mddev is not set, then scan must be set and we
1260 * read through the config file for dev+uuid mapping
1261 * We recurse, setting mddev, for each device that
1262 * - isn't running
1263 * - has a valid uuid (or any uuid if !uuidset)
1264 *
1265 * If mddev is set, we try to determine state of md.
1266 * check version - must be at least 0.90.0
1267 * check kernel version. must be at least 2.4.
1268 * If not, we can possibly fall back on START_ARRAY
1269 * Try to GET_ARRAY_INFO.
1270 * If possible, give up
1271 * If not, try to STOP_ARRAY just to make sure
1272 *
1273 * If !uuidset and scan, look in conf-file for uuid
1274 * If not found, give up
1275 * If !devlist and scan and uuidset, get list of devs from conf-file
1276 *
1277 * For each device:
1278 * Check superblock - discard if bad
1279 * Check uuid (set if we don't have one) - discard if no match
1280 * Check superblock similarity if we have a superblock - discard if different
1281 * Record events, devicenum
1282 * This should give us a list of devices for the array
1283 * We should collect the most recent event number
1284 *
1285 * Count disks with recent enough event count
1286 * While force && !enough disks
1287 * Choose newest rejected disks, update event count
1288 * mark clean and rewrite superblock
1289 * If recent kernel:
1290 * SET_ARRAY_INFO
1291 * foreach device with recent events : ADD_NEW_DISK
1292 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1293 * If old kernel:
1294 * Check the device numbers in superblock are right
1295 * update superblock if any changes
1296 * START_ARRAY
1297 *
1298 */
1299 int rv;
1300 int mdfd;
1301 int clean;
1302 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1303 ident->super_minor == UnSet && ident->name[0] == 0 &&
1304 (ident->container == NULL || ident->member == NULL));
1305 struct devs *devices;
1306 char *devmap;
1307 int *best = NULL; /* indexed by raid_disk */
1308 int bestcnt = 0;
1309 int devcnt;
1310 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1311 int journal_clean = 0;
1312 int i;
1313 int was_forced = 0;
1314 int most_recent = 0;
1315 int chosen_drive;
1316 int change = 0;
1317 int inargv = 0;
1318 int start_partial_ok = (c->runstop >= 0) &&
1319 (c->force || devlist==NULL || auto_assem);
1320 int num_devs;
1321 struct mddev_dev *tmpdev;
1322 struct mdinfo info;
1323 struct mdinfo *content = NULL;
1324 struct mdinfo *pre_exist = NULL;
1325 char *avail;
1326 char *name = NULL;
1327 char chosen_name[1024];
1328 struct map_ent *map = NULL;
1329 struct map_ent *mp;
1330
1331 /*
1332 * If any subdevs are listed, then any that don't
1333 * match ident are discarded. Remainder must all match and
1334 * become the array.
1335 * If no subdevs, then we scan all devices in the config file, but
1336 * there must be something in the identity
1337 */
1338
1339 if (!devlist &&
1340 ident->uuid_set == 0 &&
1341 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1342 ident->name[0] == 0 &&
1343 (ident->container == NULL || ident->member == NULL) &&
1344 ident->devices == NULL) {
1345 pr_err("No identity information available for %s - cannot assemble.\n",
1346 mddev ? mddev : "further assembly");
1347 return 1;
1348 }
1349
1350 if (devlist == NULL)
1351 devlist = conf_get_devs();
1352 else if (mddev)
1353 inargv = 1;
1354
1355 try_again:
1356 /* We come back here when doing auto-assembly and attempting some
1357 * set of devices failed. Those are now marked as ->used==2 and
1358 * we ignore them and try again
1359 */
1360 if (!st && ident->st)
1361 st = ident->st;
1362 if (c->verbose>0)
1363 pr_err("looking for devices for %s\n",
1364 mddev ? mddev : "further assembly");
1365
1366 content = &info;
1367 if (st && c->force)
1368 st->ignore_hw_compat = 1;
1369 num_devs = select_devices(devlist, ident, &st, &content, c,
1370 inargv, auto_assem);
1371 if (num_devs < 0)
1372 return 1;
1373
1374 if (!st || !st->sb || !content)
1375 return 2;
1376
1377 /* We have a full set of devices - we now need to find the
1378 * array device.
1379 * However there is a risk that we are racing with "mdadm -I"
1380 * and the array is already partially assembled - we will have
1381 * rejected any devices already in this address.
1382 * So we take a lock on the map file - to prevent further races -
1383 * and look for the uuid in there. If found and the array is
1384 * active, we abort. If found and the array is not active
1385 * we commit to that md device and add all the contained devices
1386 * to our list. We flag them so that we don't try to re-add,
1387 * but can remove if they turn out to not be wanted.
1388 */
1389 if (map_lock(&map))
1390 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1391 if (c->update && strcmp(c->update,"uuid") == 0)
1392 mp = NULL;
1393 else
1394 mp = map_by_uuid(&map, content->uuid);
1395 if (mp) {
1396 struct mdinfo *dv;
1397 /* array already exists. */
1398 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1399 if (pre_exist->array.level != UnSet) {
1400 pr_err("Found some drive for an array that is already active: %s\n",
1401 mp->path);
1402 pr_err("giving up.\n");
1403 return 1;
1404 }
1405 for (dv = pre_exist->devs; dv; dv = dv->next) {
1406 /* We want to add this device to our list,
1407 * but it could already be there if "mdadm -I"
1408 * started *after* we checked for O_EXCL.
1409 * If we add it to the top of the list
1410 * it will be preferred over later copies.
1411 */
1412 struct mddev_dev *newdev;
1413 char *devname = map_dev(dv->disk.major,
1414 dv->disk.minor,
1415 0);
1416 if (!devname)
1417 continue;
1418 newdev = xmalloc(sizeof(*newdev));
1419 newdev->devname = devname;
1420 newdev->disposition = 'I';
1421 newdev->used = 1;
1422 newdev->next = devlist;
1423 devlist = newdev;
1424 num_devs++;
1425 }
1426 strcpy(chosen_name, mp->path);
1427 if (c->verbose > 0 || mddev == NULL ||
1428 strcmp(mddev, chosen_name) != 0)
1429 pr_err("Merging with already-assembled %s\n",
1430 chosen_name);
1431 mdfd = open_dev_excl(mp->devnm);
1432 } else {
1433 int trustworthy = FOREIGN;
1434 name = content->name;
1435 switch (st->ss->match_home(st, c->homehost)
1436 ?: st->ss->match_home(st, "any")) {
1437 case 1:
1438 trustworthy = LOCAL;
1439 name = strchr(content->name, ':');
1440 if (name)
1441 name++;
1442 else
1443 name = content->name;
1444 break;
1445 }
1446 if (!auto_assem)
1447 /* If the array is listed in mdadm.conf or on
1448 * command line, then we trust the name
1449 * even if the array doesn't look local
1450 */
1451 trustworthy = LOCAL;
1452
1453 if (name[0] == 0 &&
1454 content->array.level == LEVEL_CONTAINER) {
1455 name = content->text_version;
1456 trustworthy = METADATA;
1457 }
1458
1459 if (name[0] && trustworthy != LOCAL &&
1460 ! c->require_homehost &&
1461 conf_name_is_free(name))
1462 trustworthy = LOCAL;
1463
1464 if (trustworthy == LOCAL &&
1465 strchr(name, ':'))
1466 /* Ignore 'host:' prefix of name */
1467 name = strchr(name, ':')+1;
1468
1469 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1470 chosen_name, 0);
1471 }
1472 if (mdfd < 0) {
1473 st->ss->free_super(st);
1474 if (auto_assem)
1475 goto try_again;
1476 return 1;
1477 }
1478 mddev = chosen_name;
1479 if (pre_exist == NULL) {
1480 if (mddev_busy(fd2devnm(mdfd))) {
1481 pr_err("%s already active, cannot restart it!\n",
1482 mddev);
1483 for (tmpdev = devlist ;
1484 tmpdev && tmpdev->used != 1;
1485 tmpdev = tmpdev->next)
1486 ;
1487 if (tmpdev && auto_assem)
1488 pr_err("%s needed for %s...\n",
1489 mddev, tmpdev->devname);
1490 close(mdfd);
1491 mdfd = -3;
1492 st->ss->free_super(st);
1493 if (auto_assem)
1494 goto try_again;
1495 return 1;
1496 }
1497 /* just incase it was started but has no content */
1498 ioctl(mdfd, STOP_ARRAY, NULL);
1499 }
1500
1501 if (content != &info) {
1502 /* This is a member of a container. Try starting the array. */
1503 int err;
1504 err = assemble_container_content(st, mdfd, content, c,
1505 chosen_name, NULL);
1506 close(mdfd);
1507 return err;
1508 }
1509
1510 /* Ok, no bad inconsistancy, we can try updating etc */
1511 devices = xcalloc(num_devs, sizeof(*devices));
1512 devmap = xcalloc(num_devs, content->array.raid_disks);
1513 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1514 c, content, mdfd, mddev,
1515 &most_recent, &bestcnt, &best, inargv);
1516 if (devcnt < 0)
1517 return 1;
1518
1519 if (devcnt == 0) {
1520 pr_err("no devices found for %s\n",
1521 mddev);
1522 if (st)
1523 st->ss->free_super(st);
1524 close(mdfd);
1525 free(devices);
1526 free(devmap);
1527 return 1;
1528 }
1529
1530 if (c->update && strcmp(c->update, "byteorder")==0)
1531 st->minor_version = 90;
1532
1533 st->ss->getinfo_super(st, content, NULL);
1534 clean = content->array.state & 1;
1535
1536 /* now we have some devices that might be suitable.
1537 * I wonder how many
1538 */
1539 avail = xcalloc(content->array.raid_disks, 1);
1540 okcnt = 0;
1541 replcnt = 0;
1542 sparecnt=0;
1543 journalcnt=0;
1544 rebuilding_cnt=0;
1545 for (i=0; i< bestcnt; i++) {
1546 int j = best[i];
1547 int event_margin = 1; /* always allow a difference of '1'
1548 * like the kernel does
1549 */
1550 if (j < 0) continue;
1551 /* note: we ignore error flags in multipath arrays
1552 * as they don't make sense
1553 */
1554 if (content->array.level != LEVEL_MULTIPATH) {
1555 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1556 if (content->journal_device_required)
1557 journalcnt++;
1558 else /* unexpected journal, mark as faulty */
1559 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1560 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1561 if (!(devices[j].i.disk.state
1562 & (1<<MD_DISK_FAULTY))) {
1563 devices[j].uptodate = 1;
1564 sparecnt++;
1565 }
1566 continue;
1567 }
1568 }
1569 /* If this device thinks that 'most_recent' has failed, then
1570 * we must reject this device.
1571 */
1572 if (j != most_recent && !c->force &&
1573 content->array.raid_disks > 0 &&
1574 devices[most_recent].i.disk.raid_disk >= 0 &&
1575 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1576 if (c->verbose > -1)
1577 pr_err("ignoring %s as it reports %s as failed\n",
1578 devices[j].devname, devices[most_recent].devname);
1579 best[i] = -1;
1580 continue;
1581 }
1582 /* Require event counter to be same as, or just less than,
1583 * most recent. If it is bigger, it must be a stray spare and
1584 * should be ignored.
1585 */
1586 if (devices[j].i.events+event_margin >=
1587 devices[most_recent].i.events &&
1588 devices[j].i.events <=
1589 devices[most_recent].i.events
1590 ) {
1591 devices[j].uptodate = 1;
1592 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1593 journal_clean = 1;
1594 if (i < content->array.raid_disks * 2) {
1595 if (devices[j].i.recovery_start == MaxSector ||
1596 (content->reshape_active &&
1597 i >= content->array.raid_disks - content->delta_disks)) {
1598 if (!avail[i/2]) {
1599 okcnt++;
1600 avail[i/2]=1;
1601 } else
1602 replcnt++;
1603 } else
1604 rebuilding_cnt++;
1605 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1606 sparecnt++;
1607 }
1608 }
1609 free(devmap);
1610 if (c->force) {
1611 int force_ok = force_array(content, devices, best, bestcnt,
1612 avail, most_recent, st, c);
1613 okcnt += force_ok;
1614 if (force_ok)
1615 was_forced = 1;
1616 }
1617 /* Now we want to look at the superblock which the kernel will base things on
1618 * and compare the devices that we think are working with the devices that the
1619 * superblock thinks are working.
1620 * If there are differences and --force is given, then update this chosen
1621 * superblock.
1622 */
1623 chosen_drive = -1;
1624 st->ss->free_super(st);
1625 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1626 int j = best[i];
1627 int fd;
1628
1629 if (j<0)
1630 continue;
1631 if (!devices[j].uptodate)
1632 continue;
1633 if (devices[j].i.events < devices[most_recent].i.events)
1634 continue;
1635 chosen_drive = j;
1636 if ((fd=dev_open(devices[j].devname,
1637 devices[j].included ? O_RDONLY
1638 : (O_RDONLY|O_EXCL)))< 0) {
1639 pr_err("Cannot open %s: %s\n",
1640 devices[j].devname, strerror(errno));
1641 close(mdfd);
1642 free(devices);
1643 return 1;
1644 }
1645 if (st->ss->load_super(st,fd, NULL)) {
1646 close(fd);
1647 pr_err("RAID superblock has disappeared from %s\n",
1648 devices[j].devname);
1649 close(mdfd);
1650 free(devices);
1651 return 1;
1652 }
1653 close(fd);
1654 }
1655 if (st->sb == NULL) {
1656 pr_err("No suitable drives found for %s\n", mddev);
1657 close(mdfd);
1658 free(devices);
1659 return 1;
1660 }
1661 st->ss->getinfo_super(st, content, NULL);
1662 if (sysfs_init(content, mdfd, NULL)) {
1663 pr_err("Unable to initialize sysfs\n");
1664 close(mdfd);
1665 free(devices);
1666 return 1;
1667 }
1668
1669 /* after reload context, store journal_clean in context */
1670 content->journal_clean = journal_clean;
1671 for (i=0; i<bestcnt; i++) {
1672 int j = best[i];
1673 unsigned int desired_state;
1674
1675 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1676 desired_state = (1<<MD_DISK_JOURNAL);
1677 else if (i >= content->array.raid_disks * 2)
1678 desired_state = 0;
1679 else if (i & 1)
1680 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1681 else
1682 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1683
1684 if (j<0)
1685 continue;
1686 if (!devices[j].uptodate)
1687 continue;
1688
1689 devices[j].i.disk.state = desired_state;
1690 if (!(devices[j].i.array.state & 1))
1691 clean = 0;
1692
1693 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1694 c->verbose, 0, NULL)) {
1695 if (c->force) {
1696 if (c->verbose >= 0)
1697 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1698 j, mddev, devices[j].devname);
1699 change = 1;
1700 } else {
1701 if (c->verbose >= -1)
1702 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1703 i, mddev, devices[j].devname);
1704 }
1705 }
1706 #if 0
1707 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1708 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1709 i, mddev);
1710 }
1711 #endif
1712 }
1713 if (c->force && !clean &&
1714 !enough(content->array.level, content->array.raid_disks,
1715 content->array.layout, clean,
1716 avail)) {
1717 change += st->ss->update_super(st, content, "force-array",
1718 devices[chosen_drive].devname, c->verbose,
1719 0, NULL);
1720 was_forced = 1;
1721 clean = 1;
1722 }
1723
1724 if (change) {
1725 int fd;
1726 fd = dev_open(devices[chosen_drive].devname,
1727 devices[chosen_drive].included ?
1728 O_RDWR : (O_RDWR|O_EXCL));
1729 if (fd < 0) {
1730 pr_err("Could not open %s for write - cannot Assemble array.\n",
1731 devices[chosen_drive].devname);
1732 close(mdfd);
1733 free(devices);
1734 return 1;
1735 }
1736 if (st->ss->store_super(st, fd)) {
1737 close(fd);
1738 pr_err("Could not re-write superblock on %s\n",
1739 devices[chosen_drive].devname);
1740 close(mdfd);
1741 free(devices);
1742 return 1;
1743 }
1744 if (c->verbose >= 0)
1745 pr_err("Marking array %s as 'clean'\n",
1746 mddev);
1747 close(fd);
1748 }
1749
1750 /* If we are in the middle of a reshape we may need to restore saved data
1751 * that was moved aside due to the reshape overwriting live data
1752 * The code of doing this lives in Grow.c
1753 */
1754 if (content->reshape_active &&
1755 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1756 int err = 0;
1757 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1758 if (c->verbose > 0)
1759 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1760 chosen_name);
1761 if (!c->backup_file)
1762 c->backup_file = locate_backup(content->sys_name);
1763 enable_fds(bestcnt/2);
1764 for (i = 0; i < bestcnt/2; i++) {
1765 int j = best[i*2];
1766 if (j >= 0) {
1767 fdlist[i] = dev_open(devices[j].devname,
1768 devices[j].included
1769 ? O_RDWR : (O_RDWR|O_EXCL));
1770 if (fdlist[i] < 0) {
1771 pr_err("Could not open %s for write - cannot Assemble array.\n",
1772 devices[j].devname);
1773 err = 1;
1774 break;
1775 }
1776 } else
1777 fdlist[i] = -1;
1778 }
1779 if (!err) {
1780 if (st->ss->external && st->ss->recover_backup)
1781 err = st->ss->recover_backup(st, content);
1782 else
1783 err = Grow_restart(st, content, fdlist, bestcnt/2,
1784 c->backup_file, c->verbose > 0);
1785 if (err && c->invalid_backup) {
1786 if (c->verbose > 0)
1787 pr_err("continuing without restoring backup\n");
1788 err = 0;
1789 }
1790 }
1791 while (i>0) {
1792 i--;
1793 if (fdlist[i]>=0) close(fdlist[i]);
1794 }
1795 free(fdlist);
1796 if (err) {
1797 pr_err("Failed to restore critical section for reshape, sorry.\n");
1798 if (c->backup_file == NULL)
1799 cont_err("Possibly you needed to specify the --backup-file\n");
1800 close(mdfd);
1801 free(devices);
1802 return err;
1803 }
1804 }
1805
1806 /* Almost ready to actually *do* something */
1807 /* First, fill in the map, so that udev can find our name
1808 * as soon as we become active.
1809 */
1810 if (c->update && strcmp(c->update, "metadata")==0) {
1811 content->array.major_version = 1;
1812 content->array.minor_version = 0;
1813 strcpy(content->text_version, "1.0");
1814 }
1815
1816 map_update(&map, fd2devnm(mdfd), content->text_version,
1817 content->uuid, chosen_name);
1818
1819 rv = start_array(mdfd, mddev, content,
1820 st, ident, best, bestcnt,
1821 chosen_drive, devices, okcnt, sparecnt,
1822 rebuilding_cnt, journalcnt,
1823 c,
1824 clean, avail, start_partial_ok,
1825 pre_exist != NULL,
1826 was_forced);
1827 if (rv == 1 && !pre_exist)
1828 ioctl(mdfd, STOP_ARRAY, NULL);
1829 free(devices);
1830 map_unlock(&map);
1831 if (rv == 0) {
1832 wait_for(chosen_name, mdfd);
1833 close(mdfd);
1834 if (auto_assem) {
1835 int usecs = 1;
1836 /* There is a nasty race with 'mdadm --monitor'.
1837 * If it opens this device before we close it,
1838 * it gets an incomplete open on which IO
1839 * doesn't work and the capacity is
1840 * wrong.
1841 * If we reopen (to check for layered devices)
1842 * before --monitor closes, we loose.
1843 *
1844 * So: wait upto 1 second for there to be
1845 * a non-zero capacity.
1846 */
1847 while (usecs < 1000) {
1848 mdfd = open(mddev, O_RDONLY);
1849 if (mdfd >= 0) {
1850 unsigned long long size;
1851 if (get_dev_size(mdfd, NULL, &size) &&
1852 size > 0)
1853 break;
1854 close(mdfd);
1855 }
1856 usleep(usecs);
1857 usecs <<= 1;
1858 }
1859 }
1860 } else
1861 close(mdfd);
1862
1863 /* '2' means 'OK, but not started yet' */
1864 return rv == 2 ? 0 : rv;
1865 }
1866
1867 int assemble_container_content(struct supertype *st, int mdfd,
1868 struct mdinfo *content, struct context *c,
1869 char *chosen_name, int *result)
1870 {
1871 struct mdinfo *dev, *sra, *dev2;
1872 int working = 0, preexist = 0;
1873 int expansion = 0;
1874 int old_raid_disks;
1875 int start_reshape;
1876 char *avail;
1877 int err;
1878
1879 if (sysfs_init(content, mdfd, NULL)) {
1880 pr_err("Unable to initialize sysfs\n");
1881 return 1;
1882 }
1883
1884 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1885 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1886 if (content->array.major_version == -1 &&
1887 content->array.minor_version == -2 &&
1888 c->readonly &&
1889 content->text_version[0] == '/')
1890 content->text_version[0] = '-';
1891 if (sysfs_set_array(content, 9003) != 0) {
1892 sysfs_free(sra);
1893 return 1;
1894 }
1895 }
1896
1897 /* There are two types of reshape: container wide or sub-array specific
1898 * Check if metadata requests blocking container wide reshapes
1899 */
1900 start_reshape = (content->reshape_active &&
1901 !((content->reshape_active == CONTAINER_RESHAPE) &&
1902 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1903
1904 /* Block subarray here if it is under reshape now
1905 * Do not allow for any changes in this array
1906 */
1907 if (st->ss->external && content->recovery_blocked && start_reshape)
1908 block_subarray(content);
1909
1910 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1911 for (dev = content->devs; dev; dev = dev->next)
1912 if (dev2->disk.major == dev->disk.major &&
1913 dev2->disk.minor == dev->disk.minor)
1914 break;
1915 if (dev)
1916 continue;
1917 /* Don't want this one any more */
1918 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1919 errno == EBUSY) {
1920 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1921 sysfs_free(sra);
1922 return 1;
1923 }
1924 sysfs_set_str(sra, dev2, "state", "remove");
1925 }
1926 old_raid_disks = content->array.raid_disks - content->delta_disks;
1927 avail = xcalloc(content->array.raid_disks, 1);
1928 for (dev = content->devs; dev; dev = dev->next) {
1929 if (dev->disk.raid_disk >= 0)
1930 avail[dev->disk.raid_disk] = 1;
1931 if (sysfs_add_disk(content, dev, 1) == 0) {
1932 if (dev->disk.raid_disk >= old_raid_disks &&
1933 content->reshape_active)
1934 expansion++;
1935 else
1936 working++;
1937 } else if (errno == EEXIST)
1938 preexist++;
1939 }
1940 sysfs_free(sra);
1941 if (working + expansion == 0 && c->runstop <= 0) {
1942 free(avail);
1943 return 1;/* Nothing new, don't try to start */
1944 }
1945 map_update(NULL, fd2devnm(mdfd), content->text_version,
1946 content->uuid, chosen_name);
1947
1948 if (content->consistency_policy == CONSISTENCY_POLICY_PPL &&
1949 st->ss->validate_ppl) {
1950 content->array.state |= 1;
1951 err = 0;
1952
1953 for (dev = content->devs; dev; dev = dev->next) {
1954 int dfd;
1955 char *devpath;
1956 int ret;
1957
1958 ret = st->ss->validate_ppl(st, content, dev);
1959 if (ret == 0)
1960 continue;
1961
1962 if (ret < 0) {
1963 err = 1;
1964 break;
1965 }
1966
1967 if (!c->force) {
1968 pr_err("%s contains invalid PPL - consider --force or --update-subarray with --update=no-ppl\n",
1969 chosen_name);
1970 content->array.state &= ~1;
1971 avail[dev->disk.raid_disk] = 0;
1972 break;
1973 }
1974
1975 /* have --force - overwrite the invalid ppl */
1976 devpath = map_dev(dev->disk.major, dev->disk.minor, 0);
1977 dfd = dev_open(devpath, O_RDWR);
1978 if (dfd < 0) {
1979 pr_err("Failed to open %s\n", devpath);
1980 err = 1;
1981 break;
1982 }
1983
1984 err = st->ss->write_init_ppl(st, content, dfd);
1985 close(dfd);
1986
1987 if (err)
1988 break;
1989 }
1990
1991 if (err) {
1992 free(avail);
1993 return err;
1994 }
1995 }
1996
1997 if (enough(content->array.level, content->array.raid_disks,
1998 content->array.layout, content->array.state & 1, avail) == 0) {
1999 if (c->export && result)
2000 *result |= INCR_NO;
2001 else if (c->verbose >= 0) {
2002 pr_err("%s assembled with %d device%s",
2003 chosen_name, preexist + working,
2004 preexist + working == 1 ? "":"s");
2005 if (preexist)
2006 fprintf(stderr, " (%d new)", working);
2007 fprintf(stderr, " but not started\n");
2008 }
2009 free(avail);
2010 return 1;
2011 }
2012 free(avail);
2013
2014 if (c->runstop <= 0 &&
2015 (working + preexist + expansion) <
2016 content->array.working_disks) {
2017 if (c->export && result)
2018 *result |= INCR_UNSAFE;
2019 else if (c->verbose >= 0) {
2020 pr_err("%s assembled with %d device%s",
2021 chosen_name, preexist + working,
2022 preexist + working == 1 ? "":"s");
2023 if (preexist)
2024 fprintf(stderr, " (%d new)", working);
2025 fprintf(stderr, " but not safe to start\n");
2026 }
2027 return 1;
2028 }
2029
2030
2031 if (start_reshape) {
2032 int spare = content->array.raid_disks + expansion;
2033 if (restore_backup(st, content,
2034 working,
2035 spare, &c->backup_file, c->verbose) == 1)
2036 return 1;
2037
2038 err = sysfs_set_str(content, NULL,
2039 "array_state", "readonly");
2040 if (err)
2041 return 1;
2042
2043 if (st->ss->external) {
2044 if (!mdmon_running(st->container_devnm))
2045 start_mdmon(st->container_devnm);
2046 ping_monitor(st->container_devnm);
2047 if (mdmon_running(st->container_devnm) &&
2048 st->update_tail == NULL)
2049 st->update_tail = &st->updates;
2050 }
2051
2052 err = Grow_continue(mdfd, st, content, c->backup_file,
2053 0, c->freeze_reshape);
2054 } else switch(content->array.level) {
2055 case LEVEL_LINEAR:
2056 case LEVEL_MULTIPATH:
2057 case 0:
2058 err = sysfs_set_str(content, NULL, "array_state",
2059 c->readonly ? "readonly" : "active");
2060 break;
2061 default:
2062 err = sysfs_set_str(content, NULL, "array_state",
2063 "readonly");
2064 /* start mdmon if needed. */
2065 if (!err) {
2066 if (!mdmon_running(st->container_devnm))
2067 start_mdmon(st->container_devnm);
2068 ping_monitor(st->container_devnm);
2069 }
2070 break;
2071 }
2072 if (!err)
2073 sysfs_set_safemode(content, content->safe_mode_delay);
2074
2075 /* Block subarray here if it is not reshaped now
2076 * It has be blocked a little later to allow mdmon to switch in
2077 * in to R/W state
2078 */
2079 if (st->ss->external && content->recovery_blocked &&
2080 !start_reshape)
2081 block_subarray(content);
2082
2083 if (c->export && result) {
2084 if (err)
2085 *result |= INCR_NO;
2086 else
2087 *result |= INCR_YES;
2088 } else if (c->verbose >= 0) {
2089 if (err)
2090 pr_err("array %s now has %d device%s",
2091 chosen_name, working + preexist,
2092 working + preexist == 1 ? "":"s");
2093 else
2094 pr_err("Started %s with %d device%s",
2095 chosen_name, working + preexist,
2096 working + preexist == 1 ? "":"s");
2097 if (preexist)
2098 fprintf(stderr, " (%d new)", working);
2099 if (expansion)
2100 fprintf(stderr, " ( + %d for expansion)",
2101 expansion);
2102 fprintf(stderr, "\n");
2103 }
2104 if (!err)
2105 wait_for(chosen_name, mdfd);
2106 return err;
2107 /* FIXME should have an O_EXCL and wait for read-auto */
2108 }
Unnamed repository; edit this file 'description' to name the repository.