]> 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
Merge branch 'cluster'
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost, int require_homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char *sep;
34 unsigned int l;
35
36 if (strcmp(found, required)==0)
37 return 1;
38 sep = strchr(found, ':');
39 if (!sep)
40 return 0;
41 l = sep - found;
42 if (strncmp(found, "any:", 4) == 0 ||
43 (homehost && strcmp(homehost, "any") == 0) ||
44 !require_homehost ||
45 (homehost && strlen(homehost) == l &&
46 strncmp(found, homehost, l) == 0)) {
47 /* matching homehost */
48 if (strcmp(sep+1, required) == 0)
49 return 1;
50 }
51 return 0;
52 }
53
54 static int is_member_busy(char *metadata_version)
55 {
56 /* check if the given member array is active */
57 struct mdstat_ent *mdstat = mdstat_read(0, 0);
58 struct mdstat_ent *ent;
59 int busy = 0;
60
61 for (ent = mdstat; ent; ent = ent->next) {
62 if (ent->metadata_version == NULL)
63 continue;
64 if (strncmp(ent->metadata_version, "external:", 9) != 0)
65 continue;
66 if (!is_subarray(&ent->metadata_version[9]))
67 continue;
68 /* Skip first char - it can be '/' or '-' */
69 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
70 busy = 1;
71 break;
72 }
73 }
74 free_mdstat(mdstat);
75
76 return busy;
77 }
78
79 static int ident_matches(struct mddev_ident *ident,
80 struct mdinfo *content,
81 struct supertype *tst,
82 char *homehost, int require_homehost,
83 char *update, char *devname)
84 {
85
86 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
87 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
88 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
89 if (devname)
90 pr_err("%s has wrong uuid.\n", devname);
91 return 0;
92 }
93 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
94 name_matches(content->name, ident->name, homehost, require_homehost)==0) {
95 if (devname)
96 pr_err("%s has wrong name.\n", devname);
97 return 0;
98 }
99 if (ident->super_minor != UnSet &&
100 ident->super_minor != content->array.md_minor) {
101 if (devname)
102 pr_err("%s has wrong super-minor.\n",
103 devname);
104 return 0;
105 }
106 if (ident->level != UnSet &&
107 ident->level != content->array.level) {
108 if (devname)
109 pr_err("%s has wrong raid level.\n",
110 devname);
111 return 0;
112 }
113 if (ident->raid_disks != UnSet &&
114 content->array.raid_disks != 0 && /* metadata doesn't know how many to expect */
115 ident->raid_disks!= content->array.raid_disks) {
116 if (devname)
117 pr_err("%s requires wrong number of drives.\n",
118 devname);
119 return 0;
120 }
121 if (ident->member && ident->member[0]) {
122 /* content->text_version must match */
123 char *s = strchr(content->text_version+1, '/');
124 if (s == NULL) {
125 if (devname)
126 pr_err("%s is not a container and one is required.\n",
127 devname);
128 return 0;
129 } else if (strcmp(ident->member, s+1) != 0) {
130 if (devname)
131 pr_err("skipping wrong member %s is %s\n",
132 content->text_version, devname);
133 return 0;
134 }
135 }
136 return 1;
137 }
138
139 static int select_devices(struct mddev_dev *devlist,
140 struct mddev_ident *ident,
141 struct supertype **stp,
142 struct mdinfo **contentp,
143 struct context *c,
144 int inargv, int auto_assem)
145 {
146 struct mddev_dev *tmpdev;
147 int num_devs;
148 struct supertype *st = *stp;
149 struct mdinfo *content = NULL;
150 int report_mismatch = ((inargv && c->verbose >= 0) || c->verbose > 0);
151 struct domainlist *domains = NULL;
152
153 tmpdev = devlist; num_devs = 0;
154 while (tmpdev) {
155 if (tmpdev->used)
156 tmpdev->used = 2;
157 else
158 num_devs++;
159 tmpdev->disposition = 0;
160 tmpdev = tmpdev->next;
161 }
162
163 /* first walk the list of devices to find a consistent set
164 * that match the criterea, if that is possible.
165 * We flag the ones we like with 'used'.
166 */
167 for (tmpdev = devlist;
168 tmpdev;
169 tmpdev = tmpdev ? tmpdev->next : NULL) {
170 char *devname = tmpdev->devname;
171 int dfd;
172 struct stat stb;
173 struct supertype *tst;
174 struct dev_policy *pol = NULL;
175 int found_container = 0;
176
177 if (tmpdev->used > 1)
178 continue;
179
180 if (ident->container) {
181 if (ident->container[0] == '/' &&
182 !same_dev(ident->container, devname)) {
183 if (report_mismatch)
184 pr_err("%s is not the container required (%s)\n",
185 devname, ident->container);
186 continue;
187 }
188 } else if (ident->devices &&
189 !match_oneof(ident->devices, devname)) {
190 /* Note that we ignore the "device=" identifier if a
191 * "container=" is given. Checking both is unnecessarily
192 * complicated.
193 */
194 if (report_mismatch)
195 pr_err("%s is not one of %s\n", devname, ident->devices);
196 continue;
197 }
198
199 tst = dup_super(st);
200
201 dfd = dev_open(devname, O_RDONLY);
202 if (dfd < 0) {
203 if (report_mismatch)
204 pr_err("cannot open device %s: %s\n",
205 devname, strerror(errno));
206 tmpdev->used = 2;
207 } else if (fstat(dfd, &stb)< 0) {
208 /* Impossible! */
209 pr_err("fstat failed for %s: %s\n",
210 devname, strerror(errno));
211 tmpdev->used = 2;
212 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
213 pr_err("%s is not a block device.\n",
214 devname);
215 tmpdev->used = 2;
216 } else if (must_be_container(dfd)) {
217 if (st) {
218 /* already found some components, this cannot
219 * be another one.
220 */
221 if (report_mismatch)
222 pr_err("%s is a container, but we are looking for components\n",
223 devname);
224 tmpdev->used = 2;
225 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
226 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
227 if (report_mismatch)
228 pr_err("not a recognisable container: %s\n",
229 devname);
230 tmpdev->used = 2;
231 #endif
232 } else if (!tst->ss->load_container
233 || tst->ss->load_container(tst, dfd, NULL)) {
234 if (report_mismatch)
235 pr_err("no correct container type: %s\n",
236 devname);
237 tmpdev->used = 2;
238 } else if (auto_assem &&
239 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
240 tst->ss->match_home(tst, c->homehost) == 1)) {
241 if (report_mismatch)
242 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
243 devname, tst->ss->name);
244 tmpdev->used = 2;
245 } else
246 found_container = 1;
247 } else {
248 if (!tst && (tst = guess_super(dfd)) == NULL) {
249 if (report_mismatch)
250 pr_err("no recogniseable superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if (tst->ss->load_super(tst,dfd, NULL)) {
254 if (report_mismatch)
255 pr_err("no RAID superblock on %s\n",
256 devname);
257 tmpdev->used = 2;
258 } else if (tst->ss->compare_super == NULL) {
259 if (report_mismatch)
260 pr_err("Cannot assemble %s metadata on %s\n",
261 tst->ss->name, devname);
262 tmpdev->used = 2;
263 } else if (auto_assem && st == NULL &&
264 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
265 tst->ss->match_home(tst, c->homehost) == 1)) {
266 if (report_mismatch)
267 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
268 devname, tst->ss->name);
269 tmpdev->used = 2;
270 }
271 }
272 if (dfd >= 0) close(dfd);
273 if (tmpdev->used == 2) {
274 if (auto_assem || !inargv)
275 /* Ignore unrecognised devices during auto-assembly */
276 goto loop;
277 if (ident->uuid_set || ident->name[0] ||
278 ident->super_minor != UnSet)
279 /* Ignore unrecognised device if looking for
280 * specific array */
281 goto loop;
282
283 pr_err("%s has no superblock - assembly aborted\n",
284 devname);
285 if (st)
286 st->ss->free_super(st);
287 dev_policy_free(pol);
288 domain_free(domains);
289 return -1;
290 }
291
292 if (found_container) {
293 /* tmpdev is a container. We need to be either
294 * looking for a member, or auto-assembling
295 */
296 /* should be safe to try an exclusive open now, we
297 * have rejected anything that some other mdadm might
298 * be looking at
299 */
300 dfd = dev_open(devname, O_RDONLY | O_EXCL);
301 if (dfd < 0) {
302 if (report_mismatch)
303 pr_err("%s is busy - skipping\n", devname);
304 goto loop;
305 }
306 close(dfd);
307
308 if (ident->container && ident->container[0] != '/') {
309 /* we have a uuid */
310 int uuid[4];
311
312 content = *contentp;
313 tst->ss->getinfo_super(tst, content, NULL);
314
315 if (!parse_uuid(ident->container, uuid) ||
316 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
317 if (report_mismatch)
318 pr_err("%s has wrong UUID to be required container\n",
319 devname);
320 goto loop;
321 }
322 }
323 /* It is worth looking inside this container.
324 */
325 if (c->verbose > 0)
326 pr_err("looking in container %s\n",
327 devname);
328
329 for (content = tst->ss->container_content(tst, NULL);
330 content;
331 content = content->next) {
332
333 if (!ident_matches(ident, content, tst,
334 c->homehost, c->require_homehost,
335 c->update,
336 report_mismatch ? devname : NULL))
337 /* message already printed */;
338 else if (is_member_busy(content->text_version)) {
339 if (report_mismatch)
340 pr_err("member %s in %s is already assembled\n",
341 content->text_version,
342 devname);
343 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
344 /* do not assemble arrays with unsupported configurations */
345 pr_err("Cannot activate member %s in %s.\n",
346 content->text_version,
347 devname);
348 } else
349 break;
350 }
351 if (!content) {
352 tmpdev->used = 2;
353 goto loop; /* empty container */
354 }
355
356 st = tst; tst = NULL;
357 if (!auto_assem && inargv && tmpdev->next != NULL) {
358 pr_err("%s is a container, but is not only device given: confused and aborting\n",
359 devname);
360 st->ss->free_super(st);
361 dev_policy_free(pol);
362 domain_free(domains);
363 return -1;
364 }
365 if (c->verbose > 0)
366 pr_err("found match on member %s in %s\n",
367 content->text_version, devname);
368
369 /* make sure we finished the loop */
370 tmpdev = NULL;
371 goto loop;
372 } else {
373 content = *contentp;
374 tst->ss->getinfo_super(tst, content, NULL);
375
376 if (!ident_matches(ident, content, tst,
377 c->homehost, c->require_homehost,
378 c->update,
379 report_mismatch ? devname : NULL))
380 goto loop;
381
382 if (auto_assem) {
383 /* Never auto-assemble things that conflict
384 * with mdadm.conf in some way
385 */
386 struct mddev_ident *match;
387 int rv = 0;
388
389 match = conf_match(tst, content, devname,
390 report_mismatch ? c->verbose : -1,
391 &rv);
392 if (!match && rv == 2)
393 goto loop;
394 if (match && match->devname &&
395 strcasecmp(match->devname, "<ignore>") == 0) {
396 if (report_mismatch)
397 pr_err("%s is a member of an explicitly ignored array\n",
398 devname);
399 goto loop;
400 }
401 if (match && !ident_matches(match, content, tst,
402 c->homehost, c->require_homehost,
403 c->update,
404 report_mismatch ? devname : NULL))
405 /* Array exists in mdadm.conf but some
406 * details don't match, so reject it
407 */
408 goto loop;
409 }
410
411 /* should be safe to try an exclusive open now, we
412 * have rejected anything that some other mdadm might
413 * be looking at
414 */
415 dfd = dev_open(devname, O_RDONLY | O_EXCL);
416 if (dfd < 0) {
417 if (report_mismatch)
418 pr_err("%s is busy - skipping\n", devname);
419 goto loop;
420 }
421 close(dfd);
422
423 if (st == NULL)
424 st = dup_super(tst);
425 if (st->minor_version == -1)
426 st->minor_version = tst->minor_version;
427
428 if (memcmp(content->uuid, uuid_zero,
429 sizeof(int[4])) == 0) {
430 /* this is a floating spare. It cannot define
431 * an array unless there are no more arrays of
432 * this type to be found. It can be included
433 * in an array of this type though.
434 */
435 tmpdev->used = 3;
436 goto loop;
437 }
438
439 if (st->ss != tst->ss ||
440 st->minor_version != tst->minor_version ||
441 st->ss->compare_super(st, tst) != 0) {
442 /* Some mismatch. If exactly one array matches this host,
443 * we can resolve on that one.
444 * Or, if we are auto assembling, we just ignore the second
445 * for now.
446 */
447 if (auto_assem)
448 goto loop;
449 if (c->homehost) {
450 int first = st->ss->match_home(st, c->homehost);
451 int last = tst->ss->match_home(tst, c->homehost);
452 if (first != last &&
453 (first == 1 || last == 1)) {
454 /* We can do something */
455 if (first) {/* just ignore this one */
456 if (report_mismatch)
457 pr_err("%s misses out due to wrong homehost\n",
458 devname);
459 goto loop;
460 } else { /* reject all those sofar */
461 struct mddev_dev *td;
462 if (report_mismatch)
463 pr_err("%s overrides previous devices due to good homehost\n",
464 devname);
465 for (td=devlist; td != tmpdev; td=td->next)
466 if (td->used == 1)
467 td->used = 0;
468 tmpdev->used = 1;
469 goto loop;
470 }
471 }
472 }
473 pr_err("superblock on %s doesn't match others - assembly aborted\n",
474 devname);
475 tst->ss->free_super(tst);
476 st->ss->free_super(st);
477 dev_policy_free(pol);
478 domain_free(domains);
479 return -1;
480 }
481 tmpdev->used = 1;
482 }
483 loop:
484 /* Collect domain information from members only */
485 if (tmpdev && tmpdev->used == 1) {
486 if (!pol)
487 pol = devid_policy(stb.st_rdev);
488 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
489 }
490 dev_policy_free(pol);
491 pol = NULL;
492 if (tst)
493 tst->ss->free_super(tst);
494 }
495
496 /* Check if we found some imsm spares but no members */
497 if ((auto_assem ||
498 (ident->uuid_set &&
499 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
500 (!st || !st->sb))
501 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
502 if (tmpdev->used != 3)
503 continue;
504 tmpdev->used = 1;
505 content = *contentp;
506
507 if (!st->sb) {
508 /* we need sb from one of the spares */
509 int dfd = dev_open(tmpdev->devname, O_RDONLY);
510 if (dfd < 0 ||
511 st->ss->load_super(st, dfd, NULL))
512 tmpdev->used = 2;
513 if (dfd > 0)
514 close(dfd);
515 }
516 }
517
518 /* Now reject spares that don't match domains of identified members */
519 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
520 struct stat stb;
521 if (tmpdev->used != 3)
522 continue;
523 if (stat(tmpdev->devname, &stb)< 0) {
524 pr_err("fstat failed for %s: %s\n",
525 tmpdev->devname, strerror(errno));
526 tmpdev->used = 2;
527 } else {
528 struct dev_policy *pol = devid_policy(stb.st_rdev);
529 int dt = domain_test(domains, pol, NULL);
530 if (inargv && dt != 0)
531 /* take this spare as domains match
532 * if there are any */
533 tmpdev->used = 1;
534 else if (!inargv && dt == 1)
535 /* device wasn't explicitly listed, so need
536 * explicit domain match - which we have */
537 tmpdev->used = 1;
538 else
539 /* if domains don't match mark as unused */
540 tmpdev->used = 0;
541 dev_policy_free(pol);
542 }
543 }
544 domain_free(domains);
545 *stp = st;
546 if (st && st->sb && content == *contentp)
547 st->ss->getinfo_super(st, content, NULL);
548 *contentp = content;
549
550 return num_devs;
551 }
552
553 struct devs {
554 char *devname;
555 int uptodate; /* set once we decide that this device is as
556 * recent as everything else in the array.
557 */
558 int included; /* set if the device is already in the array
559 * due to a previous '-I'
560 */
561 struct mdinfo i;
562 };
563
564 static int load_devices(struct devs *devices, char *devmap,
565 struct mddev_ident *ident, struct supertype **stp,
566 struct mddev_dev *devlist, struct context *c,
567 struct mdinfo *content,
568 int mdfd, char *mddev,
569 int *most_recentp, int *bestcntp, int **bestp,
570 int inargv)
571 {
572 struct mddev_dev *tmpdev;
573 int devcnt = 0;
574 int nextspare = 0;
575 #ifndef MDASSEMBLE
576 int bitmap_done = 0;
577 #endif
578 int most_recent = -1;
579 int bestcnt = 0;
580 int *best = *bestp;
581 struct supertype *st = *stp;
582
583 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
584 char *devname = tmpdev->devname;
585 struct stat stb;
586 struct supertype *tst;
587 int i;
588 int dfd;
589
590 if (tmpdev->used != 1)
591 continue;
592 /* looks like a good enough match to update the super block if needed */
593 #ifndef MDASSEMBLE
594 if (c->update) {
595 /* prepare useful information in info structures */
596 struct stat stb2;
597 int err;
598 fstat(mdfd, &stb2);
599
600 if (strcmp(c->update, "uuid")==0 &&
601 !ident->uuid_set) {
602 int rfd;
603 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
604 read(rfd, ident->uuid, 16) != 16) {
605 *(__u32*)(ident->uuid) = random();
606 *(__u32*)(ident->uuid+1) = random();
607 *(__u32*)(ident->uuid+2) = random();
608 *(__u32*)(ident->uuid+3) = random();
609 }
610 if (rfd >= 0) close(rfd);
611 }
612 dfd = dev_open(devname,
613 tmpdev->disposition == 'I'
614 ? O_RDWR : (O_RDWR|O_EXCL));
615
616 tst = dup_super(st);
617 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
618 pr_err("cannot re-read metadata from %s - aborting\n",
619 devname);
620 if (dfd >= 0)
621 close(dfd);
622 close(mdfd);
623 free(devices);
624 free(devmap);
625 tst->ss->free_super(tst);
626 free(tst);
627 *stp = st;
628 return -1;
629 }
630 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
631
632 memcpy(content->uuid, ident->uuid, 16);
633 strcpy(content->name, ident->name);
634 content->array.md_minor = minor(stb2.st_rdev);
635
636 if (strcmp(c->update, "byteorder") == 0)
637 err = 0;
638 else if (strcmp(c->update, "home-cluster") == 0) {
639 tst->cluster_name = c->homecluster;
640 tst->ss->write_bitmap(tst, dfd, NameUpdate);
641 } else if (strcmp(c->update, "nodes") == 0) {
642 tst->nodes = c->nodes;
643 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
644 } else
645 err = tst->ss->update_super(tst, content, c->update,
646 devname, c->verbose,
647 ident->uuid_set,
648 c->homehost);
649 if (err < 0) {
650 if (err == -1)
651 pr_err("--update=%s not understood for %s metadata\n",
652 c->update, tst->ss->name);
653 tst->ss->free_super(tst);
654 free(tst);
655 close(mdfd);
656 close(dfd);
657 free(devices);
658 free(devmap);
659 *stp = st;
660 return -1;
661 }
662 if (strcmp(c->update, "uuid")==0 &&
663 !ident->uuid_set) {
664 ident->uuid_set = 1;
665 memcpy(ident->uuid, content->uuid, 16);
666 }
667 if (tst->ss->store_super(tst, dfd))
668 pr_err("Could not re-write superblock on %s.\n",
669 devname);
670
671 if (strcmp(c->update, "uuid")==0 &&
672 ident->bitmap_fd >= 0 && !bitmap_done) {
673 if (bitmap_update_uuid(ident->bitmap_fd,
674 content->uuid,
675 tst->ss->swapuuid) != 0)
676 pr_err("Could not update uuid on external bitmap.\n");
677 else
678 bitmap_done = 1;
679 }
680 } else
681 #endif
682 {
683 dfd = dev_open(devname,
684 tmpdev->disposition == 'I'
685 ? O_RDWR : (O_RDWR|O_EXCL));
686 tst = dup_super(st);
687
688 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
689 pr_err("cannot re-read metadata from %s - aborting\n",
690 devname);
691 if (dfd >= 0)
692 close(dfd);
693 close(mdfd);
694 free(devices);
695 free(devmap);
696 tst->ss->free_super(tst);
697 free(tst);
698 *stp = st;
699 return -1;
700 }
701 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
702 }
703
704 fstat(dfd, &stb);
705 close(dfd);
706
707 if (c->verbose > 0)
708 pr_err("%s is identified as a member of %s, slot %d%s.\n",
709 devname, mddev, content->disk.raid_disk,
710 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
711 devices[devcnt].devname = devname;
712 devices[devcnt].uptodate = 0;
713 devices[devcnt].included = (tmpdev->disposition == 'I');
714 devices[devcnt].i = *content;
715 devices[devcnt].i.disk.major = major(stb.st_rdev);
716 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
717
718 if (devices[devcnt].i.disk.state == 6) {
719 if (most_recent < 0 ||
720 devices[devcnt].i.events
721 > devices[most_recent].i.events) {
722 struct supertype *tmp = tst;
723 tst = st;
724 st = tmp;
725 most_recent = devcnt;
726 }
727 }
728 tst->ss->free_super(tst);
729 free(tst);
730
731 if (content->array.level == LEVEL_MULTIPATH)
732 /* with multipath, the raid_disk from the superblock is meaningless */
733 i = devcnt;
734 else
735 i = devices[devcnt].i.disk.raid_disk;
736 if (i+1 == 0) {
737 if (nextspare < content->array.raid_disks*2)
738 nextspare = content->array.raid_disks*2;
739 i = nextspare++;
740 } else {
741 /* i is raid_disk - double it so there is room for
742 * replacements */
743 i *= 2;
744 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
745 i++;
746 if (i >= content->array.raid_disks*2 &&
747 i >= nextspare)
748 nextspare = i+1;
749 }
750 if (i < 10000) {
751 if (i >= bestcnt) {
752 int newbestcnt = i+10;
753 int *newbest = xmalloc(sizeof(int)*newbestcnt);
754 int c;
755 for (c=0; c < newbestcnt; c++)
756 if (c < bestcnt)
757 newbest[c] = best[c];
758 else
759 newbest[c] = -1;
760 if (best)free(best);
761 best = newbest;
762 bestcnt = newbestcnt;
763 }
764 if (best[i] >=0 &&
765 devices[best[i]].i.events
766 == devices[devcnt].i.events
767 && (devices[best[i]].i.disk.minor
768 != devices[devcnt].i.disk.minor)
769 && st->ss == &super0
770 && content->array.level != LEVEL_MULTIPATH) {
771 /* two different devices with identical superblock.
772 * Could be a mis-detection caused by overlapping
773 * partitions. fail-safe.
774 */
775 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
776 " If they are really different, please --zero the superblock on one\n"
777 " If they are the same or overlap, please remove one from %s.\n",
778 devices[best[i]].devname, devname,
779 inargv ? "the list" :
780 "the\n DEVICE list in mdadm.conf"
781 );
782 close(mdfd);
783 free(devices);
784 free(devmap);
785 *stp = st;
786 return -1;
787 }
788 if (best[i] == -1
789 || (devices[best[i]].i.events
790 < devices[devcnt].i.events))
791 best[i] = devcnt;
792 }
793 devcnt++;
794 }
795 if (most_recent >= 0)
796 *most_recentp = most_recent;
797 *bestcntp = bestcnt;
798 *bestp = best;
799 *stp = st;
800 return devcnt;
801 }
802
803 static int force_array(struct mdinfo *content,
804 struct devs *devices,
805 int *best, int bestcnt, char *avail,
806 int most_recent,
807 struct supertype *st,
808 struct context *c)
809 {
810 int okcnt = 0;
811 while (!enough(content->array.level, content->array.raid_disks,
812 content->array.layout, 1,
813 avail)
814 ||
815 (content->reshape_active && content->delta_disks > 0 &&
816 !enough(content->array.level, (content->array.raid_disks
817 - content->delta_disks),
818 content->new_layout, 1,
819 avail)
820 )) {
821 /* Choose the newest best drive which is
822 * not up-to-date, update the superblock
823 * and add it.
824 */
825 int fd;
826 struct supertype *tst;
827 unsigned long long current_events;
828 int chosen_drive = -1;
829 int i;
830
831 for (i = 0;
832 i < content->array.raid_disks * 2 && i < bestcnt;
833 i += 2) {
834 int j = best[i];
835 if (j < 0)
836 continue;
837 if (devices[j].uptodate)
838 continue;
839 if (devices[j].i.recovery_start != MaxSector) {
840 int delta;
841 if (!devices[j].i.reshape_active ||
842 devices[j].i.delta_disks <= 0)
843 continue;
844 /* When increasing number of devices, an
845 * added device also appears to be
846 * recovering. It is safe to include it
847 * as long as it won't be a source of
848 * data.
849 * For now, just allow for last data
850 * devices in RAID4 or last devices in RAID4/5/6.
851 */
852 delta = devices[j].i.delta_disks;
853 if (devices[j].i.array.level >= 4 &&
854 devices[j].i.array.level <= 6 &&
855 i/2 >= content->array.raid_disks - delta)
856 /* OK */;
857 else if (devices[j].i.array.level == 4 &&
858 i/2 >= content->array.raid_disks - delta - 1)
859 /* OK */;
860 else
861 continue;
862 }
863 if (chosen_drive < 0 ||
864 devices[j].i.events
865 > devices[chosen_drive].i.events)
866 chosen_drive = j;
867 }
868 if (chosen_drive < 0)
869 break;
870 current_events = devices[chosen_drive].i.events;
871 add_another:
872 if (c->verbose >= 0)
873 pr_err("forcing event count in %s(%d) from %d upto %d\n",
874 devices[chosen_drive].devname,
875 devices[chosen_drive].i.disk.raid_disk,
876 (int)(devices[chosen_drive].i.events),
877 (int)(devices[most_recent].i.events));
878 fd = dev_open(devices[chosen_drive].devname,
879 devices[chosen_drive].included ? O_RDWR
880 : (O_RDWR|O_EXCL));
881 if (fd < 0) {
882 pr_err("Couldn't open %s for write - not updating\n",
883 devices[chosen_drive].devname);
884 devices[chosen_drive].i.events = 0;
885 continue;
886 }
887 tst = dup_super(st);
888 if (tst->ss->load_super(tst,fd, NULL)) {
889 close(fd);
890 pr_err("RAID superblock disappeared from %s - not updating.\n",
891 devices[chosen_drive].devname);
892 devices[chosen_drive].i.events = 0;
893 continue;
894 }
895 content->events = devices[most_recent].i.events;
896 tst->ss->update_super(tst, content, "force-one",
897 devices[chosen_drive].devname, c->verbose,
898 0, NULL);
899
900 if (tst->ss->store_super(tst, fd)) {
901 close(fd);
902 pr_err("Could not re-write superblock on %s\n",
903 devices[chosen_drive].devname);
904 devices[chosen_drive].i.events = 0;
905 tst->ss->free_super(tst);
906 continue;
907 }
908 close(fd);
909 devices[chosen_drive].i.events = devices[most_recent].i.events;
910 devices[chosen_drive].uptodate = 1;
911 avail[chosen_drive] = 1;
912 okcnt++;
913 tst->ss->free_super(tst);
914
915 /* If there are any other drives of the same vintage,
916 * add them in as well. We can't lose and we might gain
917 */
918 for (i = 0;
919 i < content->array.raid_disks * 2 && i < bestcnt ;
920 i += 2) {
921 int j = best[i];
922 if (j >= 0 &&
923 !devices[j].uptodate &&
924 devices[j].i.recovery_start == MaxSector &&
925 devices[j].i.events == current_events) {
926 chosen_drive = j;
927 goto add_another;
928 }
929 }
930 }
931 return okcnt;
932 }
933
934 static int start_array(int mdfd,
935 char *mddev,
936 struct mdinfo *content,
937 struct supertype *st,
938 struct mddev_ident *ident,
939 int *best, int bestcnt,
940 int chosen_drive,
941 struct devs *devices,
942 unsigned int okcnt,
943 unsigned int sparecnt,
944 unsigned int rebuilding_cnt,
945 struct context *c,
946 int clean, char *avail,
947 int start_partial_ok,
948 int err_ok,
949 int was_forced
950 )
951 {
952 int rv;
953 int i;
954 unsigned int req_cnt;
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 = dev_open(devices[j].devname,
995 O_RDWR|O_EXCL);
996 if (dfd >= 0) {
997 remove_partitions(dfd);
998 close(dfd);
999 }
1000 rv = add_disk(mdfd, st, content, &devices[j].i);
1001
1002 if (rv) {
1003 pr_err("failed to add %s to %s: %s\n",
1004 devices[j].devname,
1005 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 & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1019 } else if (j >= 0) {
1020 if (c->verbose > 0)
1021 pr_err("%s is already in %s as %d\n",
1022 devices[j].devname, mddev,
1023 devices[j].i.disk.raid_disk);
1024 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1025 && (i&1) == 0)
1026 pr_err("no uptodate device for slot %d of %s\n",
1027 i/2, mddev);
1028 }
1029
1030 if (content->array.level == LEVEL_CONTAINER) {
1031 if (c->verbose >= 0) {
1032 pr_err("Container %s has been assembled with %d drive%s",
1033 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1034 if (okcnt < (unsigned)content->array.raid_disks)
1035 fprintf(stderr, " (out of %d)",
1036 content->array.raid_disks);
1037 fprintf(stderr, "\n");
1038 }
1039
1040 if (st->ss->validate_container) {
1041 struct mdinfo *devices_list;
1042 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1043 unsigned int count;
1044 devices_list = NULL;
1045 for (count = 0; count < okcnt+sparecnt; count++) {
1046 info_devices[count] = devices[count].i;
1047 info_devices[count].next = devices_list;
1048 devices_list = &info_devices[count];
1049 }
1050 if (st->ss->validate_container(devices_list))
1051 pr_err("Mismatch detected!\n");
1052 free(info_devices);
1053 }
1054
1055 st->ss->free_super(st);
1056 sysfs_uevent(content, "change");
1057 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1058 /* Was partial, is still partial, so signal an error
1059 * to ensure we don't retry */
1060 return 1;
1061 return 0;
1062 }
1063
1064 /* Get number of in-sync devices according to the superblock.
1065 * We must have this number to start the array without -s or -R
1066 */
1067 req_cnt = content->array.working_disks;
1068
1069 if (c->runstop == 1 ||
1070 (c->runstop <= 0 &&
1071 ( enough(content->array.level, content->array.raid_disks,
1072 content->array.layout, clean, avail) &&
1073 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1074 ))) {
1075 /* This array is good-to-go.
1076 * If a reshape is in progress then we might need to
1077 * continue monitoring it. In that case we start
1078 * it read-only and let the grow code make it writable.
1079 */
1080 int rv;
1081 #ifndef MDASSEMBLE
1082 if (content->reshape_active &&
1083 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1084 content->delta_disks <= 0) {
1085 if (!c->backup_file) {
1086 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1087 mddev);
1088 pr_err("Please provided one with \"--backup-file=...\"\n");
1089 if (c->update &&
1090 strcmp(c->update, "revert-reshape") == 0)
1091 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1092 return 1;
1093 }
1094 rv = sysfs_set_str(content, NULL,
1095 "array_state", "readonly");
1096 if (rv == 0)
1097 rv = Grow_continue(mdfd, st, content,
1098 c->backup_file, 0,
1099 c->freeze_reshape);
1100 } else if (c->readonly &&
1101 sysfs_attribute_available(
1102 content, NULL, "array_state")) {
1103 rv = sysfs_set_str(content, NULL,
1104 "array_state", "readonly");
1105 } else
1106 #endif
1107 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1108 reopen_mddev(mdfd); /* drop O_EXCL */
1109 if (rv == 0) {
1110 if (c->verbose >= 0) {
1111 pr_err("%s has been started with %d drive%s",
1112 mddev, okcnt, okcnt==1?"":"s");
1113 if (okcnt < (unsigned)content->array.raid_disks)
1114 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1115 if (rebuilding_cnt)
1116 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1117 if (sparecnt)
1118 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1119 fprintf(stderr, ".\n");
1120 }
1121 if (content->reshape_active &&
1122 content->array.level >= 4 &&
1123 content->array.level <= 6) {
1124 /* might need to increase the size
1125 * of the stripe cache - default is 256
1126 */
1127 int chunk_size = content->array.chunk_size;
1128 if (content->reshape_active &&
1129 content->new_chunk > chunk_size)
1130 chunk_size = content->new_chunk;
1131 if (256 < 4 * ((chunk_size+4065)/4096)) {
1132 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1133 if (sra)
1134 sysfs_set_num(sra, NULL,
1135 "stripe_cache_size",
1136 (4 * chunk_size / 4096) + 1);
1137 sysfs_free(sra);
1138 }
1139 }
1140 if (okcnt < (unsigned)content->array.raid_disks) {
1141 /* If any devices did not get added
1142 * because the kernel rejected them based
1143 * on event count, try adding them
1144 * again providing the action policy is
1145 * 're-add' or greater. The bitmap
1146 * might allow them to be included, or
1147 * they will become spares.
1148 */
1149 for (i = 0; i < bestcnt; i++) {
1150 int j = best[i];
1151 if (j >= 0 && !devices[j].uptodate) {
1152 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1153 continue;
1154 rv = add_disk(mdfd, st, content,
1155 &devices[j].i);
1156 if (rv == 0 && c->verbose >= 0)
1157 pr_err("%s has been re-added.\n",
1158 devices[j].devname);
1159 }
1160 }
1161 }
1162 if (content->array.level == 6 &&
1163 okcnt + 1 == (unsigned)content->array.raid_disks &&
1164 was_forced) {
1165 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1166 if (sra)
1167 sysfs_set_str(sra, NULL,
1168 "sync_action", "repair");
1169 sysfs_free(sra);
1170 }
1171 return 0;
1172 }
1173 pr_err("failed to RUN_ARRAY %s: %s\n",
1174 mddev, strerror(errno));
1175
1176 if (!enough(content->array.level, content->array.raid_disks,
1177 content->array.layout, 1, avail))
1178 pr_err("Not enough devices to start the array.\n");
1179 else if (!enough(content->array.level,
1180 content->array.raid_disks,
1181 content->array.layout, clean,
1182 avail))
1183 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1184
1185 return 1;
1186 }
1187 if (c->runstop == -1) {
1188 pr_err("%s assembled from %d drive%s",
1189 mddev, okcnt, okcnt==1?"":"s");
1190 if (okcnt != (unsigned)content->array.raid_disks)
1191 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1192 fprintf(stderr, ", but not started.\n");
1193 return 2;
1194 }
1195 if (c->verbose >= -1) {
1196 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1197 if (rebuilding_cnt)
1198 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1199 if (sparecnt)
1200 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1201 if (!enough(content->array.level, content->array.raid_disks,
1202 content->array.layout, 1, avail))
1203 fprintf(stderr, " - not enough to start the array.\n");
1204 else if (!enough(content->array.level,
1205 content->array.raid_disks,
1206 content->array.layout, clean,
1207 avail))
1208 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1209 else {
1210 if (req_cnt == (unsigned)content->array.raid_disks)
1211 fprintf(stderr, " - need all %d to start it", req_cnt);
1212 else
1213 fprintf(stderr, " - need %d to start", req_cnt);
1214 fprintf(stderr, " (use --run to insist).\n");
1215 }
1216 }
1217 return 1;
1218 }
1219
1220 int Assemble(struct supertype *st, char *mddev,
1221 struct mddev_ident *ident,
1222 struct mddev_dev *devlist,
1223 struct context *c)
1224 {
1225 /*
1226 * The task of Assemble is to find a collection of
1227 * devices that should (according to their superblocks)
1228 * form an array, and to give this collection to the MD driver.
1229 * In Linux-2.4 and later, this involves submitting a
1230 * SET_ARRAY_INFO ioctl with no arg - to prepare
1231 * the array - and then submit a number of
1232 * ADD_NEW_DISK ioctls to add disks into
1233 * the array. Finally RUN_ARRAY might
1234 * be submitted to start the array.
1235 *
1236 * Much of the work of Assemble is in finding and/or
1237 * checking the disks to make sure they look right.
1238 *
1239 * If mddev is not set, then scan must be set and we
1240 * read through the config file for dev+uuid mapping
1241 * We recurse, setting mddev, for each device that
1242 * - isn't running
1243 * - has a valid uuid (or any uuid if !uuidset)
1244 *
1245 * If mddev is set, we try to determine state of md.
1246 * check version - must be at least 0.90.0
1247 * check kernel version. must be at least 2.4.
1248 * If not, we can possibly fall back on START_ARRAY
1249 * Try to GET_ARRAY_INFO.
1250 * If possible, give up
1251 * If not, try to STOP_ARRAY just to make sure
1252 *
1253 * If !uuidset and scan, look in conf-file for uuid
1254 * If not found, give up
1255 * If !devlist and scan and uuidset, get list of devs from conf-file
1256 *
1257 * For each device:
1258 * Check superblock - discard if bad
1259 * Check uuid (set if we don't have one) - discard if no match
1260 * Check superblock similarity if we have a superblock - discard if different
1261 * Record events, devicenum
1262 * This should give us a list of devices for the array
1263 * We should collect the most recent event number
1264 *
1265 * Count disks with recent enough event count
1266 * While force && !enough disks
1267 * Choose newest rejected disks, update event count
1268 * mark clean and rewrite superblock
1269 * If recent kernel:
1270 * SET_ARRAY_INFO
1271 * foreach device with recent events : ADD_NEW_DISK
1272 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1273 * If old kernel:
1274 * Check the device numbers in superblock are right
1275 * update superblock if any changes
1276 * START_ARRAY
1277 *
1278 */
1279 int rv;
1280 int mdfd;
1281 int clean;
1282 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1283 ident->super_minor == UnSet && ident->name[0] == 0
1284 && (ident->container == NULL || ident->member == NULL));
1285 struct devs *devices;
1286 char *devmap;
1287 int *best = NULL; /* indexed by raid_disk */
1288 int bestcnt = 0;
1289 int devcnt;
1290 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1291 int i;
1292 int was_forced = 0;
1293 int most_recent = 0;
1294 int chosen_drive;
1295 int change = 0;
1296 int inargv = 0;
1297 int start_partial_ok = (c->runstop >= 0) &&
1298 (c->force || devlist==NULL || auto_assem);
1299 int num_devs;
1300 struct mddev_dev *tmpdev;
1301 struct mdinfo info;
1302 struct mdinfo *content = NULL;
1303 struct mdinfo *pre_exist = NULL;
1304 char *avail;
1305 char *name = NULL;
1306 char chosen_name[1024];
1307 struct map_ent *map = NULL;
1308 struct map_ent *mp;
1309
1310 /*
1311 * If any subdevs are listed, then any that don't
1312 * match ident are discarded. Remainder must all match and
1313 * become the array.
1314 * If no subdevs, then we scan all devices in the config file, but
1315 * there must be something in the identity
1316 */
1317
1318 if (!devlist &&
1319 ident->uuid_set == 0 &&
1320 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1321 ident->name[0] == 0 &&
1322 (ident->container == NULL || ident->member == NULL) &&
1323 ident->devices == NULL) {
1324 pr_err("No identity information available for %s - cannot assemble.\n",
1325 mddev ? mddev : "further assembly");
1326 return 1;
1327 }
1328
1329 if (devlist == NULL)
1330 devlist = conf_get_devs();
1331 else if (mddev)
1332 inargv = 1;
1333
1334 try_again:
1335 /* We come back here when doing auto-assembly and attempting some
1336 * set of devices failed. Those are now marked as ->used==2 and
1337 * we ignore them and try again
1338 */
1339 if (!st && ident->st)
1340 st = ident->st;
1341 if (c->verbose>0)
1342 pr_err("looking for devices for %s\n",
1343 mddev ? mddev : "further assembly");
1344
1345 content = &info;
1346 if (st)
1347 st->ignore_hw_compat = 1;
1348 num_devs = select_devices(devlist, ident, &st, &content, c,
1349 inargv, auto_assem);
1350 if (num_devs < 0)
1351 return 1;
1352
1353 if (!st || !st->sb || !content)
1354 return 2;
1355
1356 /* We have a full set of devices - we now need to find the
1357 * array device.
1358 * However there is a risk that we are racing with "mdadm -I"
1359 * and the array is already partially assembled - we will have
1360 * rejected any devices already in this address.
1361 * So we take a lock on the map file - to prevent further races -
1362 * and look for the uuid in there. If found and the array is
1363 * active, we abort. If found and the array is not active
1364 * we commit to that md device and add all the contained devices
1365 * to our list. We flag them so that we don't try to re-add,
1366 * but can remove if they turn out to not be wanted.
1367 */
1368 if (map_lock(&map))
1369 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1370 if (c->update && strcmp(c->update,"uuid") == 0)
1371 mp = NULL;
1372 else
1373 mp = map_by_uuid(&map, content->uuid);
1374 if (mp) {
1375 struct mdinfo *dv;
1376 /* array already exists. */
1377 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1378 if (pre_exist->array.level != UnSet) {
1379 pr_err("Found some drive for an array that is already active: %s\n",
1380 mp->path);
1381 pr_err("giving up.\n");
1382 return 1;
1383 }
1384 for (dv = pre_exist->devs; dv; dv = dv->next) {
1385 /* We want to add this device to our list,
1386 * but it could already be there if "mdadm -I"
1387 * started *after* we checked for O_EXCL.
1388 * If we add it to the top of the list
1389 * it will be preferred over later copies.
1390 */
1391 struct mddev_dev *newdev;
1392 char *devname = map_dev(dv->disk.major,
1393 dv->disk.minor,
1394 0);
1395 if (!devname)
1396 continue;
1397 newdev = xmalloc(sizeof(*newdev));
1398 newdev->devname = devname;
1399 newdev->disposition = 'I';
1400 newdev->used = 1;
1401 newdev->next = devlist;
1402 devlist = newdev;
1403 num_devs++;
1404 }
1405 strcpy(chosen_name, mp->path);
1406 if (c->verbose > 0 || mddev == NULL ||
1407 strcmp(mddev, chosen_name) != 0)
1408 pr_err("Merging with already-assembled %s\n",
1409 chosen_name);
1410 mdfd = open_dev_excl(mp->devnm);
1411 } else {
1412 int trustworthy = FOREIGN;
1413 name = content->name;
1414 switch (st->ss->match_home(st, c->homehost)
1415 ?: st->ss->match_home(st, "any")) {
1416 case 1:
1417 trustworthy = LOCAL;
1418 name = strchr(content->name, ':');
1419 if (name)
1420 name++;
1421 else
1422 name = content->name;
1423 break;
1424 }
1425 if (!auto_assem)
1426 /* If the array is listed in mdadm.conf or on
1427 * command line, then we trust the name
1428 * even if the array doesn't look local
1429 */
1430 trustworthy = LOCAL;
1431
1432 if (name[0] == 0 &&
1433 content->array.level == LEVEL_CONTAINER) {
1434 name = content->text_version;
1435 trustworthy = METADATA;
1436 }
1437
1438 if (name[0] && trustworthy != LOCAL &&
1439 ! c->require_homehost &&
1440 conf_name_is_free(name))
1441 trustworthy = LOCAL;
1442
1443 if (trustworthy == LOCAL &&
1444 strchr(name, ':'))
1445 /* Ignore 'host:' prefix of name */
1446 name = strchr(name, ':')+1;
1447
1448 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1449 chosen_name);
1450 }
1451 if (mdfd < 0) {
1452 st->ss->free_super(st);
1453 if (auto_assem)
1454 goto try_again;
1455 return 1;
1456 }
1457 mddev = chosen_name;
1458 if (get_linux_version() < 2004000 ||
1459 md_get_version(mdfd) < 9000) {
1460 pr_err("Assemble requires Linux 2.4 or later, and\n"
1461 " md driver version 0.90.0 or later.\n"
1462 " Upgrade your kernel or try --build\n");
1463 close(mdfd);
1464 return 1;
1465 }
1466 if (pre_exist == NULL) {
1467 if (mddev_busy(fd2devnm(mdfd))) {
1468 pr_err("%s already active, cannot restart it!\n",
1469 mddev);
1470 for (tmpdev = devlist ;
1471 tmpdev && tmpdev->used != 1;
1472 tmpdev = tmpdev->next)
1473 ;
1474 if (tmpdev && auto_assem)
1475 pr_err("%s needed for %s...\n",
1476 mddev, tmpdev->devname);
1477 close(mdfd);
1478 mdfd = -3;
1479 st->ss->free_super(st);
1480 if (auto_assem)
1481 goto try_again;
1482 return 1;
1483 }
1484 /* just incase it was started but has no content */
1485 ioctl(mdfd, STOP_ARRAY, NULL);
1486 }
1487
1488 #ifndef MDASSEMBLE
1489 if (content != &info) {
1490 /* This is a member of a container. Try starting the array. */
1491 int err;
1492 err = assemble_container_content(st, mdfd, content, c,
1493 chosen_name, NULL);
1494 close(mdfd);
1495 return err;
1496 }
1497 #endif
1498 /* Ok, no bad inconsistancy, we can try updating etc */
1499 devices = xcalloc(num_devs, sizeof(*devices));
1500 devmap = xcalloc(num_devs, content->array.raid_disks);
1501 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1502 c, content, mdfd, mddev,
1503 &most_recent, &bestcnt, &best, inargv);
1504 if (devcnt < 0)
1505 return 1;
1506
1507 if (devcnt == 0) {
1508 pr_err("no devices found for %s\n",
1509 mddev);
1510 if (st)
1511 st->ss->free_super(st);
1512 close(mdfd);
1513 free(devices);
1514 free(devmap);
1515 return 1;
1516 }
1517
1518 if (c->update && strcmp(c->update, "byteorder")==0)
1519 st->minor_version = 90;
1520
1521 st->ss->getinfo_super(st, content, NULL);
1522 clean = content->array.state & 1;
1523
1524 /* now we have some devices that might be suitable.
1525 * I wonder how many
1526 */
1527 avail = xcalloc(content->array.raid_disks, 1);
1528 okcnt = 0;
1529 replcnt = 0;
1530 sparecnt=0;
1531 rebuilding_cnt=0;
1532 for (i=0; i< bestcnt; i++) {
1533 int j = best[i];
1534 int event_margin = 1; /* always allow a difference of '1'
1535 * like the kernel does
1536 */
1537 if (j < 0) continue;
1538 /* note: we ignore error flags in multipath arrays
1539 * as they don't make sense
1540 */
1541 if (content->array.level != LEVEL_MULTIPATH)
1542 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1543 if (!(devices[j].i.disk.state
1544 & (1<<MD_DISK_FAULTY))) {
1545 devices[j].uptodate = 1;
1546 sparecnt++;
1547 }
1548 continue;
1549 }
1550 /* If this device thinks that 'most_recent' has failed, then
1551 * we must reject this device.
1552 */
1553 if (j != most_recent && !c->force &&
1554 content->array.raid_disks > 0 &&
1555 devices[most_recent].i.disk.raid_disk >= 0 &&
1556 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1557 if (c->verbose > -1)
1558 pr_err("ignoring %s as it reports %s as failed\n",
1559 devices[j].devname, devices[most_recent].devname);
1560 best[i] = -1;
1561 continue;
1562 }
1563 /* Require event counter to be same as, or just less than,
1564 * most recent. If it is bigger, it must be a stray spare and
1565 * should be ignored.
1566 */
1567 if (devices[j].i.events+event_margin >=
1568 devices[most_recent].i.events &&
1569 devices[j].i.events <=
1570 devices[most_recent].i.events
1571 ) {
1572 devices[j].uptodate = 1;
1573 if (i < content->array.raid_disks * 2) {
1574 if (devices[j].i.recovery_start == MaxSector ||
1575 (content->reshape_active &&
1576 i >= content->array.raid_disks - content->delta_disks)) {
1577 if (!avail[i/2]) {
1578 okcnt++;
1579 avail[i/2]=1;
1580 } else
1581 replcnt++;
1582 } else
1583 rebuilding_cnt++;
1584 } else
1585 sparecnt++;
1586 }
1587 }
1588 free(devmap);
1589 if (c->force) {
1590 int force_ok = force_array(content, devices, best, bestcnt,
1591 avail, most_recent, st, c);
1592 okcnt += force_ok;
1593 if (force_ok)
1594 was_forced = 1;
1595 }
1596 /* Now we want to look at the superblock which the kernel will base things on
1597 * and compare the devices that we think are working with the devices that the
1598 * superblock thinks are working.
1599 * If there are differences and --force is given, then update this chosen
1600 * superblock.
1601 */
1602 chosen_drive = -1;
1603 st->ss->free_super(st);
1604 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1605 int j = best[i];
1606 int fd;
1607
1608 if (j<0)
1609 continue;
1610 if (!devices[j].uptodate)
1611 continue;
1612 if (devices[j].i.events < devices[most_recent].i.events)
1613 continue;
1614 chosen_drive = j;
1615 if ((fd=dev_open(devices[j].devname,
1616 devices[j].included ? O_RDONLY
1617 : (O_RDONLY|O_EXCL)))< 0) {
1618 pr_err("Cannot open %s: %s\n",
1619 devices[j].devname, strerror(errno));
1620 close(mdfd);
1621 free(devices);
1622 return 1;
1623 }
1624 if (st->ss->load_super(st,fd, NULL)) {
1625 close(fd);
1626 pr_err("RAID superblock has disappeared from %s\n",
1627 devices[j].devname);
1628 close(mdfd);
1629 free(devices);
1630 return 1;
1631 }
1632 close(fd);
1633 }
1634 if (st->sb == NULL) {
1635 pr_err("No suitable drives found for %s\n", mddev);
1636 close(mdfd);
1637 free(devices);
1638 return 1;
1639 }
1640 st->ss->getinfo_super(st, content, NULL);
1641 #ifndef MDASSEMBLE
1642 sysfs_init(content, mdfd, NULL);
1643 #endif
1644 for (i=0; i<bestcnt; i++) {
1645 int j = best[i];
1646 unsigned int desired_state;
1647
1648 if (i >= content->array.raid_disks * 2)
1649 desired_state = 0;
1650 else if (i & 1)
1651 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1652 else
1653 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1654
1655 if (j<0)
1656 continue;
1657 if (!devices[j].uptodate)
1658 continue;
1659
1660 devices[j].i.disk.state = desired_state;
1661 if (!(devices[j].i.array.state & 1))
1662 clean = 0;
1663
1664 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1665 c->verbose, 0, NULL)) {
1666 if (c->force) {
1667 if (c->verbose >= 0)
1668 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1669 j, mddev, devices[j].devname);
1670 change = 1;
1671 } else {
1672 if (c->verbose >= -1)
1673 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1674 i, mddev, devices[j].devname);
1675 }
1676 }
1677 #if 0
1678 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1679 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1680 i, mddev);
1681 }
1682 #endif
1683 }
1684 if (c->force && !clean &&
1685 !enough(content->array.level, content->array.raid_disks,
1686 content->array.layout, clean,
1687 avail)) {
1688 change += st->ss->update_super(st, content, "force-array",
1689 devices[chosen_drive].devname, c->verbose,
1690 0, NULL);
1691 was_forced = 1;
1692 clean = 1;
1693 }
1694
1695 if (change) {
1696 int fd;
1697 fd = dev_open(devices[chosen_drive].devname,
1698 devices[chosen_drive].included ?
1699 O_RDWR : (O_RDWR|O_EXCL));
1700 if (fd < 0) {
1701 pr_err("Could not open %s for write - cannot Assemble array.\n",
1702 devices[chosen_drive].devname);
1703 close(mdfd);
1704 free(devices);
1705 return 1;
1706 }
1707 if (st->ss->store_super(st, fd)) {
1708 close(fd);
1709 pr_err("Could not re-write superblock on %s\n",
1710 devices[chosen_drive].devname);
1711 close(mdfd);
1712 free(devices);
1713 return 1;
1714 }
1715 if (c->verbose >= 0)
1716 pr_err("Marking array %s as 'clean'\n",
1717 mddev);
1718 close(fd);
1719 }
1720
1721 /* If we are in the middle of a reshape we may need to restore saved data
1722 * that was moved aside due to the reshape overwriting live data
1723 * The code of doing this lives in Grow.c
1724 */
1725 #ifndef MDASSEMBLE
1726 if (content->reshape_active &&
1727 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1728 int err = 0;
1729 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1730 if (c->verbose > 0)
1731 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1732 chosen_name);
1733 if (!c->backup_file)
1734 c->backup_file = locate_backup(content->sys_name);
1735 enable_fds(bestcnt/2);
1736 for (i = 0; i < bestcnt/2; i++) {
1737 int j = best[i*2];
1738 if (j >= 0) {
1739 fdlist[i] = dev_open(devices[j].devname,
1740 devices[j].included
1741 ? O_RDWR : (O_RDWR|O_EXCL));
1742 if (fdlist[i] < 0) {
1743 pr_err("Could not open %s for write - cannot Assemble array.\n",
1744 devices[j].devname);
1745 err = 1;
1746 break;
1747 }
1748 } else
1749 fdlist[i] = -1;
1750 }
1751 if (!err) {
1752 if (st->ss->external && st->ss->recover_backup)
1753 err = st->ss->recover_backup(st, content);
1754 else
1755 err = Grow_restart(st, content, fdlist, bestcnt/2,
1756 c->backup_file, c->verbose > 0);
1757 if (err && c->invalid_backup) {
1758 if (c->verbose > 0)
1759 pr_err("continuing without restoring backup\n");
1760 err = 0;
1761 }
1762 }
1763 while (i>0) {
1764 i--;
1765 if (fdlist[i]>=0) close(fdlist[i]);
1766 }
1767 free(fdlist);
1768 if (err) {
1769 pr_err("Failed to restore critical section for reshape, sorry.\n");
1770 if (c->backup_file == NULL)
1771 cont_err("Possibly you needed to specify the --backup-file\n");
1772 close(mdfd);
1773 free(devices);
1774 return err;
1775 }
1776 }
1777 #endif
1778
1779 /* Almost ready to actually *do* something */
1780 /* First, fill in the map, so that udev can find our name
1781 * as soon as we become active.
1782 */
1783 if (c->update && strcmp(c->update, "metadata")==0) {
1784 content->array.major_version = 1;
1785 content->array.minor_version = 0;
1786 strcpy(content->text_version, "1.0");
1787 }
1788
1789 map_update(&map, fd2devnm(mdfd), content->text_version,
1790 content->uuid, chosen_name);
1791
1792 rv = start_array(mdfd, mddev, content,
1793 st, ident, best, bestcnt,
1794 chosen_drive, devices, okcnt, sparecnt,
1795 rebuilding_cnt,
1796 c,
1797 clean, avail, start_partial_ok,
1798 pre_exist != NULL,
1799 was_forced);
1800 if (rv == 1 && !pre_exist)
1801 ioctl(mdfd, STOP_ARRAY, NULL);
1802 free(devices);
1803 map_unlock(&map);
1804 if (rv == 0) {
1805 wait_for(chosen_name, mdfd);
1806 close(mdfd);
1807 if (auto_assem) {
1808 int usecs = 1;
1809 /* There is a nasty race with 'mdadm --monitor'.
1810 * If it opens this device before we close it,
1811 * it gets an incomplete open on which IO
1812 * doesn't work and the capacity is
1813 * wrong.
1814 * If we reopen (to check for layered devices)
1815 * before --monitor closes, we loose.
1816 *
1817 * So: wait upto 1 second for there to be
1818 * a non-zero capacity.
1819 */
1820 while (usecs < 1000) {
1821 mdfd = open(mddev, O_RDONLY);
1822 if (mdfd >= 0) {
1823 unsigned long long size;
1824 if (get_dev_size(mdfd, NULL, &size) &&
1825 size > 0)
1826 break;
1827 close(mdfd);
1828 }
1829 usleep(usecs);
1830 usecs <<= 1;
1831 }
1832 }
1833 } else
1834 close(mdfd);
1835
1836 /* '2' means 'OK, but not started yet' */
1837 return rv == 2 ? 0 : rv;
1838 }
1839
1840 #ifndef MDASSEMBLE
1841 int assemble_container_content(struct supertype *st, int mdfd,
1842 struct mdinfo *content, struct context *c,
1843 char *chosen_name, int *result)
1844 {
1845 struct mdinfo *dev, *sra, *dev2;
1846 int working = 0, preexist = 0;
1847 int expansion = 0;
1848 struct map_ent *map = NULL;
1849 int old_raid_disks;
1850 int start_reshape;
1851 char *avail = NULL;
1852 int err;
1853
1854 sysfs_init(content, mdfd, NULL);
1855
1856 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1857 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1858 if (content->array.major_version == -1 &&
1859 content->array.minor_version == -2 &&
1860 c->readonly &&
1861 content->text_version[0] == '/')
1862 content->text_version[0] = '-';
1863 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1864 if (sra)
1865 sysfs_free(sra);
1866 return 1;
1867 }
1868 }
1869
1870 /* There are two types of reshape: container wide or sub-array specific
1871 * Check if metadata requests blocking container wide reshapes
1872 */
1873 start_reshape = (content->reshape_active &&
1874 !((content->reshape_active == CONTAINER_RESHAPE) &&
1875 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1876
1877 /* Block subarray here if it is under reshape now
1878 * Do not allow for any changes in this array
1879 */
1880 if (st->ss->external && content->recovery_blocked && start_reshape)
1881 block_subarray(content);
1882
1883 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1884 for (dev = content->devs; dev; dev = dev->next)
1885 if (dev2->disk.major == dev->disk.major &&
1886 dev2->disk.minor == dev->disk.minor)
1887 break;
1888 if (dev)
1889 continue;
1890 /* Don't want this one any more */
1891 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1892 errno == EBUSY) {
1893 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1894 sysfs_free(sra);
1895 return 1;
1896 }
1897 sysfs_set_str(sra, dev2, "state", "remove");
1898 }
1899 old_raid_disks = content->array.raid_disks - content->delta_disks;
1900 avail = xcalloc(content->array.raid_disks, 1);
1901 for (dev = content->devs; dev; dev = dev->next) {
1902 if (dev->disk.raid_disk >= 0)
1903 avail[dev->disk.raid_disk] = 1;
1904 if (sysfs_add_disk(content, dev, 1) == 0) {
1905 if (dev->disk.raid_disk >= old_raid_disks &&
1906 content->reshape_active)
1907 expansion++;
1908 else
1909 working++;
1910 } else if (errno == EEXIST)
1911 preexist++;
1912 }
1913 sysfs_free(sra);
1914 if (working + expansion == 0 && c->runstop <= 0) {
1915 free(avail);
1916 return 1;/* Nothing new, don't try to start */
1917 }
1918 map_update(&map, fd2devnm(mdfd),
1919 content->text_version,
1920 content->uuid, chosen_name);
1921
1922
1923 if (enough(content->array.level, content->array.raid_disks,
1924 content->array.layout, content->array.state & 1, avail) == 0) {
1925 if (c->export && result)
1926 *result |= INCR_NO;
1927 else if (c->verbose >= 0) {
1928 pr_err("%s assembled with %d device%s",
1929 chosen_name, preexist + working,
1930 preexist + working == 1 ? "":"s");
1931 if (preexist)
1932 fprintf(stderr, " (%d new)", working);
1933 fprintf(stderr, " but not started\n");
1934 }
1935 free(avail);
1936 return 1;
1937 }
1938 free(avail);
1939
1940 if (c->runstop <= 0 &&
1941 (working + preexist + expansion) <
1942 content->array.working_disks) {
1943 if (c->export && result)
1944 *result |= INCR_UNSAFE;
1945 else if (c->verbose >= 0) {
1946 pr_err("%s assembled with %d device%s",
1947 chosen_name, preexist + working,
1948 preexist + working == 1 ? "":"s");
1949 if (preexist)
1950 fprintf(stderr, " (%d new)", working);
1951 fprintf(stderr, " but not safe to start\n");
1952 }
1953 return 1;
1954 }
1955
1956
1957 if (start_reshape) {
1958 int spare = content->array.raid_disks + expansion;
1959 if (restore_backup(st, content,
1960 working,
1961 spare, &c->backup_file, c->verbose) == 1)
1962 return 1;
1963
1964 err = sysfs_set_str(content, NULL,
1965 "array_state", "readonly");
1966 if (err)
1967 return 1;
1968
1969 if (st->ss->external) {
1970 if (!mdmon_running(st->container_devnm))
1971 start_mdmon(st->container_devnm);
1972 ping_monitor(st->container_devnm);
1973 if (mdmon_running(st->container_devnm) &&
1974 st->update_tail == NULL)
1975 st->update_tail = &st->updates;
1976 }
1977
1978 err = Grow_continue(mdfd, st, content, c->backup_file,
1979 0, c->freeze_reshape);
1980 } else switch(content->array.level) {
1981 case LEVEL_LINEAR:
1982 case LEVEL_MULTIPATH:
1983 case 0:
1984 err = sysfs_set_str(content, NULL, "array_state",
1985 c->readonly ? "readonly" : "active");
1986 break;
1987 default:
1988 err = sysfs_set_str(content, NULL, "array_state",
1989 "readonly");
1990 /* start mdmon if needed. */
1991 if (!err) {
1992 if (!mdmon_running(st->container_devnm))
1993 start_mdmon(st->container_devnm);
1994 ping_monitor(st->container_devnm);
1995 }
1996 break;
1997 }
1998 if (!err)
1999 sysfs_set_safemode(content, content->safe_mode_delay);
2000
2001 /* Block subarray here if it is not reshaped now
2002 * It has be blocked a little later to allow mdmon to switch in
2003 * in to R/W state
2004 */
2005 if (st->ss->external && content->recovery_blocked &&
2006 !start_reshape)
2007 block_subarray(content);
2008
2009 if (c->export && result) {
2010 if (err)
2011 *result |= INCR_NO;
2012 else
2013 *result |= INCR_YES;
2014 } else if (c->verbose >= 0) {
2015 if (err)
2016 pr_err("array %s now has %d device%s",
2017 chosen_name, working + preexist,
2018 working + preexist == 1 ? "":"s");
2019 else
2020 pr_err("Started %s with %d device%s",
2021 chosen_name, working + preexist,
2022 working + preexist == 1 ? "":"s");
2023 if (preexist)
2024 fprintf(stderr, " (%d new)", working);
2025 if (expansion)
2026 fprintf(stderr, " ( + %d for expansion)",
2027 expansion);
2028 fprintf(stderr, "\n");
2029 }
2030 if (!err)
2031 wait_for(chosen_name, mdfd);
2032 return err;
2033 /* FIXME should have an O_EXCL and wait for read-auto */
2034 }
2035 #endif
Unnamed repository; edit this file 'description' to name the repository.