]> 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
Assemble: don't assemble IMSM array without OROM.
[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
639 err = tst->ss->update_super(tst, content, c->update,
640 devname, c->verbose,
641 ident->uuid_set,
642 c->homehost);
643 if (err < 0) {
644 if (err == -1)
645 pr_err("--update=%s not understood for %s metadata\n",
646 c->update, tst->ss->name);
647 tst->ss->free_super(tst);
648 free(tst);
649 close(mdfd);
650 close(dfd);
651 free(devices);
652 free(devmap);
653 *stp = st;
654 return -1;
655 }
656 if (strcmp(c->update, "uuid")==0 &&
657 !ident->uuid_set) {
658 ident->uuid_set = 1;
659 memcpy(ident->uuid, content->uuid, 16);
660 }
661 if (tst->ss->store_super(tst, dfd))
662 pr_err("Could not re-write superblock on %s.\n",
663 devname);
664
665 if (strcmp(c->update, "uuid")==0 &&
666 ident->bitmap_fd >= 0 && !bitmap_done) {
667 if (bitmap_update_uuid(ident->bitmap_fd,
668 content->uuid,
669 tst->ss->swapuuid) != 0)
670 pr_err("Could not update uuid on external bitmap.\n");
671 else
672 bitmap_done = 1;
673 }
674 } else
675 #endif
676 {
677 dfd = dev_open(devname,
678 tmpdev->disposition == 'I'
679 ? O_RDWR : (O_RDWR|O_EXCL));
680 tst = dup_super(st);
681
682 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
683 pr_err("cannot re-read metadata from %s - aborting\n",
684 devname);
685 if (dfd >= 0)
686 close(dfd);
687 close(mdfd);
688 free(devices);
689 free(devmap);
690 tst->ss->free_super(tst);
691 free(tst);
692 *stp = st;
693 return -1;
694 }
695 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
696 }
697
698 fstat(dfd, &stb);
699 close(dfd);
700
701 if (c->verbose > 0)
702 pr_err("%s is identified as a member of %s, slot %d%s.\n",
703 devname, mddev, content->disk.raid_disk,
704 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
705 devices[devcnt].devname = devname;
706 devices[devcnt].uptodate = 0;
707 devices[devcnt].included = (tmpdev->disposition == 'I');
708 devices[devcnt].i = *content;
709 devices[devcnt].i.disk.major = major(stb.st_rdev);
710 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
711
712 if (devices[devcnt].i.disk.state == 6) {
713 if (most_recent < 0 ||
714 devices[devcnt].i.events
715 > devices[most_recent].i.events) {
716 struct supertype *tmp = tst;
717 tst = st;
718 st = tmp;
719 most_recent = devcnt;
720 }
721 }
722 tst->ss->free_super(tst);
723 free(tst);
724
725 if (content->array.level == LEVEL_MULTIPATH)
726 /* with multipath, the raid_disk from the superblock is meaningless */
727 i = devcnt;
728 else
729 i = devices[devcnt].i.disk.raid_disk;
730 if (i+1 == 0) {
731 if (nextspare < content->array.raid_disks*2)
732 nextspare = content->array.raid_disks*2;
733 i = nextspare++;
734 } else {
735 /* i is raid_disk - double it so there is room for
736 * replacements */
737 i *= 2;
738 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
739 i++;
740 if (i >= content->array.raid_disks*2 &&
741 i >= nextspare)
742 nextspare = i+1;
743 }
744 if (i < 10000) {
745 if (i >= bestcnt) {
746 int newbestcnt = i+10;
747 int *newbest = xmalloc(sizeof(int)*newbestcnt);
748 int c;
749 for (c=0; c < newbestcnt; c++)
750 if (c < bestcnt)
751 newbest[c] = best[c];
752 else
753 newbest[c] = -1;
754 if (best)free(best);
755 best = newbest;
756 bestcnt = newbestcnt;
757 }
758 if (best[i] >=0 &&
759 devices[best[i]].i.events
760 == devices[devcnt].i.events
761 && (devices[best[i]].i.disk.minor
762 != devices[devcnt].i.disk.minor)
763 && st->ss == &super0
764 && content->array.level != LEVEL_MULTIPATH) {
765 /* two different devices with identical superblock.
766 * Could be a mis-detection caused by overlapping
767 * partitions. fail-safe.
768 */
769 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
770 " If they are really different, please --zero the superblock on one\n"
771 " If they are the same or overlap, please remove one from %s.\n",
772 devices[best[i]].devname, devname,
773 inargv ? "the list" :
774 "the\n DEVICE list in mdadm.conf"
775 );
776 close(mdfd);
777 free(devices);
778 free(devmap);
779 *stp = st;
780 return -1;
781 }
782 if (best[i] == -1
783 || (devices[best[i]].i.events
784 < devices[devcnt].i.events))
785 best[i] = devcnt;
786 }
787 devcnt++;
788 }
789 if (most_recent >= 0)
790 *most_recentp = most_recent;
791 *bestcntp = bestcnt;
792 *bestp = best;
793 *stp = st;
794 return devcnt;
795 }
796
797 static int force_array(struct mdinfo *content,
798 struct devs *devices,
799 int *best, int bestcnt, char *avail,
800 int most_recent,
801 struct supertype *st,
802 struct context *c)
803 {
804 int okcnt = 0;
805 while (!enough(content->array.level, content->array.raid_disks,
806 content->array.layout, 1,
807 avail)
808 ||
809 (content->reshape_active && content->delta_disks > 0 &&
810 !enough(content->array.level, (content->array.raid_disks
811 - content->delta_disks),
812 content->new_layout, 1,
813 avail)
814 )) {
815 /* Choose the newest best drive which is
816 * not up-to-date, update the superblock
817 * and add it.
818 */
819 int fd;
820 struct supertype *tst;
821 unsigned long long current_events;
822 int chosen_drive = -1;
823 int i;
824
825 for (i = 0;
826 i < content->array.raid_disks * 2 && i < bestcnt;
827 i += 2) {
828 int j = best[i];
829 if (j < 0)
830 continue;
831 if (devices[j].uptodate)
832 continue;
833 if (devices[j].i.recovery_start != MaxSector) {
834 int delta;
835 if (!devices[j].i.reshape_active ||
836 devices[j].i.delta_disks <= 0)
837 continue;
838 /* When increasing number of devices, an
839 * added device also appears to be
840 * recovering. It is safe to include it
841 * as long as it won't be a source of
842 * data.
843 * For now, just allow for last data
844 * devices in RAID4 or last devices in RAID4/5/6.
845 */
846 delta = devices[j].i.delta_disks;
847 if (devices[j].i.array.level >= 4 &&
848 devices[j].i.array.level <= 6 &&
849 i/2 >= content->array.raid_disks - delta)
850 /* OK */;
851 else if (devices[j].i.array.level == 4 &&
852 i/2 >= content->array.raid_disks - delta - 1)
853 /* OK */;
854 else
855 continue;
856 }
857 if (chosen_drive < 0 ||
858 devices[j].i.events
859 > devices[chosen_drive].i.events)
860 chosen_drive = j;
861 }
862 if (chosen_drive < 0)
863 break;
864 current_events = devices[chosen_drive].i.events;
865 add_another:
866 if (c->verbose >= 0)
867 pr_err("forcing event count in %s(%d) from %d upto %d\n",
868 devices[chosen_drive].devname,
869 devices[chosen_drive].i.disk.raid_disk,
870 (int)(devices[chosen_drive].i.events),
871 (int)(devices[most_recent].i.events));
872 fd = dev_open(devices[chosen_drive].devname,
873 devices[chosen_drive].included ? O_RDWR
874 : (O_RDWR|O_EXCL));
875 if (fd < 0) {
876 pr_err("Couldn't open %s for write - not updating\n",
877 devices[chosen_drive].devname);
878 devices[chosen_drive].i.events = 0;
879 continue;
880 }
881 tst = dup_super(st);
882 if (tst->ss->load_super(tst,fd, NULL)) {
883 close(fd);
884 pr_err("RAID superblock disappeared from %s - not updating.\n",
885 devices[chosen_drive].devname);
886 devices[chosen_drive].i.events = 0;
887 continue;
888 }
889 content->events = devices[most_recent].i.events;
890 tst->ss->update_super(tst, content, "force-one",
891 devices[chosen_drive].devname, c->verbose,
892 0, NULL);
893
894 if (tst->ss->store_super(tst, fd)) {
895 close(fd);
896 pr_err("Could not re-write superblock on %s\n",
897 devices[chosen_drive].devname);
898 devices[chosen_drive].i.events = 0;
899 tst->ss->free_super(tst);
900 continue;
901 }
902 close(fd);
903 devices[chosen_drive].i.events = devices[most_recent].i.events;
904 devices[chosen_drive].uptodate = 1;
905 avail[chosen_drive] = 1;
906 okcnt++;
907 tst->ss->free_super(tst);
908
909 /* If there are any other drives of the same vintage,
910 * add them in as well. We can't lose and we might gain
911 */
912 for (i = 0;
913 i < content->array.raid_disks * 2 && i < bestcnt ;
914 i += 2) {
915 int j = best[i];
916 if (j >= 0 &&
917 !devices[j].uptodate &&
918 devices[j].i.recovery_start == MaxSector &&
919 devices[j].i.events == current_events) {
920 chosen_drive = j;
921 goto add_another;
922 }
923 }
924 }
925 return okcnt;
926 }
927
928 static int start_array(int mdfd,
929 char *mddev,
930 struct mdinfo *content,
931 struct supertype *st,
932 struct mddev_ident *ident,
933 int *best, int bestcnt,
934 int chosen_drive,
935 struct devs *devices,
936 unsigned int okcnt,
937 unsigned int sparecnt,
938 unsigned int rebuilding_cnt,
939 struct context *c,
940 int clean, char *avail,
941 int start_partial_ok,
942 int err_ok,
943 int was_forced
944 )
945 {
946 int rv;
947 int i;
948 unsigned int req_cnt;
949
950 rv = set_array_info(mdfd, st, content);
951 if (rv && !err_ok) {
952 pr_err("failed to set array info for %s: %s\n",
953 mddev, strerror(errno));
954 return 1;
955 }
956 if (ident->bitmap_fd >= 0) {
957 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
958 pr_err("SET_BITMAP_FILE failed.\n");
959 return 1;
960 }
961 } else if (ident->bitmap_file) {
962 /* From config file */
963 int bmfd = open(ident->bitmap_file, O_RDWR);
964 if (bmfd < 0) {
965 pr_err("Could not open bitmap file %s\n",
966 ident->bitmap_file);
967 return 1;
968 }
969 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
970 pr_err("Failed to set bitmapfile for %s\n", mddev);
971 close(bmfd);
972 return 1;
973 }
974 close(bmfd);
975 }
976
977 /* First, add the raid disks, but add the chosen one last */
978 for (i=0; i<= bestcnt; i++) {
979 int j;
980 if (i < bestcnt) {
981 j = best[i];
982 if (j == chosen_drive)
983 continue;
984 } else
985 j = chosen_drive;
986
987 if (j >= 0 && !devices[j].included) {
988 int dfd = dev_open(devices[j].devname,
989 O_RDWR|O_EXCL);
990 if (dfd >= 0) {
991 remove_partitions(dfd);
992 close(dfd);
993 }
994 rv = add_disk(mdfd, st, content, &devices[j].i);
995
996 if (rv) {
997 pr_err("failed to add %s to %s: %s\n",
998 devices[j].devname,
999 mddev,
1000 strerror(errno));
1001 if (i < content->array.raid_disks * 2
1002 || i == bestcnt)
1003 okcnt--;
1004 else
1005 sparecnt--;
1006 } else if (c->verbose > 0)
1007 pr_err("added %s to %s as %d%s%s\n",
1008 devices[j].devname, mddev,
1009 devices[j].i.disk.raid_disk,
1010 devices[j].uptodate?"":
1011 " (possibly out of date)",
1012 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1013 } else if (j >= 0) {
1014 if (c->verbose > 0)
1015 pr_err("%s is already in %s as %d\n",
1016 devices[j].devname, mddev,
1017 devices[j].i.disk.raid_disk);
1018 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1019 && (i&1) == 0)
1020 pr_err("no uptodate device for slot %d of %s\n",
1021 i/2, mddev);
1022 }
1023
1024 if (content->array.level == LEVEL_CONTAINER) {
1025 if (c->verbose >= 0) {
1026 pr_err("Container %s has been assembled with %d drive%s",
1027 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1028 if (okcnt < (unsigned)content->array.raid_disks)
1029 fprintf(stderr, " (out of %d)",
1030 content->array.raid_disks);
1031 fprintf(stderr, "\n");
1032 }
1033
1034 if (st->ss->validate_container) {
1035 struct mdinfo *devices_list;
1036 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1037 unsigned int count;
1038 devices_list = NULL;
1039 for (count = 0; count < okcnt+sparecnt; count++) {
1040 info_devices[count] = devices[count].i;
1041 info_devices[count].next = devices_list;
1042 devices_list = &info_devices[count];
1043 }
1044 if (st->ss->validate_container(devices_list))
1045 pr_err("Mismatch detected!\n");
1046 free(info_devices);
1047 }
1048
1049 st->ss->free_super(st);
1050 sysfs_uevent(content, "change");
1051 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1052 /* Was partial, is still partial, so signal an error
1053 * to ensure we don't retry */
1054 return 1;
1055 return 0;
1056 }
1057
1058 /* Get number of in-sync devices according to the superblock.
1059 * We must have this number to start the array without -s or -R
1060 */
1061 req_cnt = content->array.working_disks;
1062
1063 if (c->runstop == 1 ||
1064 (c->runstop <= 0 &&
1065 ( enough(content->array.level, content->array.raid_disks,
1066 content->array.layout, clean, avail) &&
1067 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1068 ))) {
1069 /* This array is good-to-go.
1070 * If a reshape is in progress then we might need to
1071 * continue monitoring it. In that case we start
1072 * it read-only and let the grow code make it writable.
1073 */
1074 int rv;
1075 #ifndef MDASSEMBLE
1076 if (content->reshape_active &&
1077 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1078 content->delta_disks <= 0) {
1079 if (!c->backup_file) {
1080 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1081 mddev);
1082 pr_err("Please provided one with \"--backup-file=...\"\n");
1083 if (c->update &&
1084 strcmp(c->update, "revert-reshape") == 0)
1085 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1086 return 1;
1087 }
1088 rv = sysfs_set_str(content, NULL,
1089 "array_state", "readonly");
1090 if (rv == 0)
1091 rv = Grow_continue(mdfd, st, content,
1092 c->backup_file, 0,
1093 c->freeze_reshape);
1094 } else if (c->readonly &&
1095 sysfs_attribute_available(
1096 content, NULL, "array_state")) {
1097 rv = sysfs_set_str(content, NULL,
1098 "array_state", "readonly");
1099 } else
1100 #endif
1101 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1102 reopen_mddev(mdfd); /* drop O_EXCL */
1103 if (rv == 0) {
1104 if (c->verbose >= 0) {
1105 pr_err("%s has been started with %d drive%s",
1106 mddev, okcnt, okcnt==1?"":"s");
1107 if (okcnt < (unsigned)content->array.raid_disks)
1108 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1109 if (rebuilding_cnt)
1110 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1111 if (sparecnt)
1112 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1113 fprintf(stderr, ".\n");
1114 }
1115 if (content->reshape_active &&
1116 content->array.level >= 4 &&
1117 content->array.level <= 6) {
1118 /* might need to increase the size
1119 * of the stripe cache - default is 256
1120 */
1121 int chunk_size = content->array.chunk_size;
1122 if (content->reshape_active &&
1123 content->new_chunk > chunk_size)
1124 chunk_size = content->new_chunk;
1125 if (256 < 4 * ((chunk_size+4065)/4096)) {
1126 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1127 if (sra)
1128 sysfs_set_num(sra, NULL,
1129 "stripe_cache_size",
1130 (4 * chunk_size / 4096) + 1);
1131 sysfs_free(sra);
1132 }
1133 }
1134 if (okcnt < (unsigned)content->array.raid_disks) {
1135 /* If any devices did not get added
1136 * because the kernel rejected them based
1137 * on event count, try adding them
1138 * again providing the action policy is
1139 * 're-add' or greater. The bitmap
1140 * might allow them to be included, or
1141 * they will become spares.
1142 */
1143 for (i = 0; i < bestcnt; i++) {
1144 int j = best[i];
1145 if (j >= 0 && !devices[j].uptodate) {
1146 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1147 continue;
1148 rv = add_disk(mdfd, st, content,
1149 &devices[j].i);
1150 if (rv == 0 && c->verbose >= 0)
1151 pr_err("%s has been re-added.\n",
1152 devices[j].devname);
1153 }
1154 }
1155 }
1156 if (content->array.level == 6 &&
1157 okcnt + 1 == (unsigned)content->array.raid_disks &&
1158 was_forced) {
1159 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1160 if (sra)
1161 sysfs_set_str(sra, NULL,
1162 "sync_action", "repair");
1163 sysfs_free(sra);
1164 }
1165 return 0;
1166 }
1167 pr_err("failed to RUN_ARRAY %s: %s\n",
1168 mddev, strerror(errno));
1169
1170 if (!enough(content->array.level, content->array.raid_disks,
1171 content->array.layout, 1, avail))
1172 pr_err("Not enough devices to start the array.\n");
1173 else if (!enough(content->array.level,
1174 content->array.raid_disks,
1175 content->array.layout, clean,
1176 avail))
1177 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1178
1179 return 1;
1180 }
1181 if (c->runstop == -1) {
1182 pr_err("%s assembled from %d drive%s",
1183 mddev, okcnt, okcnt==1?"":"s");
1184 if (okcnt != (unsigned)content->array.raid_disks)
1185 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1186 fprintf(stderr, ", but not started.\n");
1187 return 2;
1188 }
1189 if (c->verbose >= -1) {
1190 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1191 if (rebuilding_cnt)
1192 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1193 if (sparecnt)
1194 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1195 if (!enough(content->array.level, content->array.raid_disks,
1196 content->array.layout, 1, avail))
1197 fprintf(stderr, " - not enough to start the array.\n");
1198 else if (!enough(content->array.level,
1199 content->array.raid_disks,
1200 content->array.layout, clean,
1201 avail))
1202 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1203 else {
1204 if (req_cnt == (unsigned)content->array.raid_disks)
1205 fprintf(stderr, " - need all %d to start it", req_cnt);
1206 else
1207 fprintf(stderr, " - need %d to start", req_cnt);
1208 fprintf(stderr, " (use --run to insist).\n");
1209 }
1210 }
1211 return 1;
1212 }
1213
1214 int Assemble(struct supertype *st, char *mddev,
1215 struct mddev_ident *ident,
1216 struct mddev_dev *devlist,
1217 struct context *c)
1218 {
1219 /*
1220 * The task of Assemble is to find a collection of
1221 * devices that should (according to their superblocks)
1222 * form an array, and to give this collection to the MD driver.
1223 * In Linux-2.4 and later, this involves submitting a
1224 * SET_ARRAY_INFO ioctl with no arg - to prepare
1225 * the array - and then submit a number of
1226 * ADD_NEW_DISK ioctls to add disks into
1227 * the array. Finally RUN_ARRAY might
1228 * be submitted to start the array.
1229 *
1230 * Much of the work of Assemble is in finding and/or
1231 * checking the disks to make sure they look right.
1232 *
1233 * If mddev is not set, then scan must be set and we
1234 * read through the config file for dev+uuid mapping
1235 * We recurse, setting mddev, for each device that
1236 * - isn't running
1237 * - has a valid uuid (or any uuid if !uuidset)
1238 *
1239 * If mddev is set, we try to determine state of md.
1240 * check version - must be at least 0.90.0
1241 * check kernel version. must be at least 2.4.
1242 * If not, we can possibly fall back on START_ARRAY
1243 * Try to GET_ARRAY_INFO.
1244 * If possible, give up
1245 * If not, try to STOP_ARRAY just to make sure
1246 *
1247 * If !uuidset and scan, look in conf-file for uuid
1248 * If not found, give up
1249 * If !devlist and scan and uuidset, get list of devs from conf-file
1250 *
1251 * For each device:
1252 * Check superblock - discard if bad
1253 * Check uuid (set if we don't have one) - discard if no match
1254 * Check superblock similarity if we have a superblock - discard if different
1255 * Record events, devicenum
1256 * This should give us a list of devices for the array
1257 * We should collect the most recent event number
1258 *
1259 * Count disks with recent enough event count
1260 * While force && !enough disks
1261 * Choose newest rejected disks, update event count
1262 * mark clean and rewrite superblock
1263 * If recent kernel:
1264 * SET_ARRAY_INFO
1265 * foreach device with recent events : ADD_NEW_DISK
1266 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1267 * If old kernel:
1268 * Check the device numbers in superblock are right
1269 * update superblock if any changes
1270 * START_ARRAY
1271 *
1272 */
1273 int rv;
1274 int mdfd;
1275 int clean;
1276 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1277 ident->super_minor == UnSet && ident->name[0] == 0
1278 && (ident->container == NULL || ident->member == NULL));
1279 struct devs *devices;
1280 char *devmap;
1281 int *best = NULL; /* indexed by raid_disk */
1282 int bestcnt = 0;
1283 int devcnt;
1284 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt;
1285 int i;
1286 int was_forced = 0;
1287 int most_recent = 0;
1288 int chosen_drive;
1289 int change = 0;
1290 int inargv = 0;
1291 int start_partial_ok = (c->runstop >= 0) &&
1292 (c->force || devlist==NULL || auto_assem);
1293 int num_devs;
1294 struct mddev_dev *tmpdev;
1295 struct mdinfo info;
1296 struct mdinfo *content = NULL;
1297 struct mdinfo *pre_exist = NULL;
1298 char *avail;
1299 char *name = NULL;
1300 char chosen_name[1024];
1301 struct map_ent *map = NULL;
1302 struct map_ent *mp;
1303
1304 /*
1305 * If any subdevs are listed, then any that don't
1306 * match ident are discarded. Remainder must all match and
1307 * become the array.
1308 * If no subdevs, then we scan all devices in the config file, but
1309 * there must be something in the identity
1310 */
1311
1312 if (!devlist &&
1313 ident->uuid_set == 0 &&
1314 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1315 ident->name[0] == 0 &&
1316 (ident->container == NULL || ident->member == NULL) &&
1317 ident->devices == NULL) {
1318 pr_err("No identity information available for %s - cannot assemble.\n",
1319 mddev ? mddev : "further assembly");
1320 return 1;
1321 }
1322
1323 if (devlist == NULL)
1324 devlist = conf_get_devs();
1325 else if (mddev)
1326 inargv = 1;
1327
1328 try_again:
1329 /* We come back here when doing auto-assembly and attempting some
1330 * set of devices failed. Those are now marked as ->used==2 and
1331 * we ignore them and try again
1332 */
1333 if (!st && ident->st)
1334 st = ident->st;
1335 if (c->verbose>0)
1336 pr_err("looking for devices for %s\n",
1337 mddev ? mddev : "further assembly");
1338
1339 content = &info;
1340 if (st && c->force)
1341 st->ignore_hw_compat = 1;
1342 num_devs = select_devices(devlist, ident, &st, &content, c,
1343 inargv, auto_assem);
1344 if (num_devs < 0)
1345 return 1;
1346
1347 if (!st || !st->sb || !content)
1348 return 2;
1349
1350 /* We have a full set of devices - we now need to find the
1351 * array device.
1352 * However there is a risk that we are racing with "mdadm -I"
1353 * and the array is already partially assembled - we will have
1354 * rejected any devices already in this address.
1355 * So we take a lock on the map file - to prevent further races -
1356 * and look for the uuid in there. If found and the array is
1357 * active, we abort. If found and the array is not active
1358 * we commit to that md device and add all the contained devices
1359 * to our list. We flag them so that we don't try to re-add,
1360 * but can remove if they turn out to not be wanted.
1361 */
1362 if (map_lock(&map))
1363 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1364 if (c->update && strcmp(c->update,"uuid") == 0)
1365 mp = NULL;
1366 else
1367 mp = map_by_uuid(&map, content->uuid);
1368 if (mp) {
1369 struct mdinfo *dv;
1370 /* array already exists. */
1371 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1372 if (pre_exist->array.level != UnSet) {
1373 pr_err("Found some drive for an array that is already active: %s\n",
1374 mp->path);
1375 pr_err("giving up.\n");
1376 return 1;
1377 }
1378 for (dv = pre_exist->devs; dv; dv = dv->next) {
1379 /* We want to add this device to our list,
1380 * but it could already be there if "mdadm -I"
1381 * started *after* we checked for O_EXCL.
1382 * If we add it to the top of the list
1383 * it will be preferred over later copies.
1384 */
1385 struct mddev_dev *newdev;
1386 char *devname = map_dev(dv->disk.major,
1387 dv->disk.minor,
1388 0);
1389 if (!devname)
1390 continue;
1391 newdev = xmalloc(sizeof(*newdev));
1392 newdev->devname = devname;
1393 newdev->disposition = 'I';
1394 newdev->used = 1;
1395 newdev->next = devlist;
1396 devlist = newdev;
1397 num_devs++;
1398 }
1399 strcpy(chosen_name, mp->path);
1400 if (c->verbose > 0 || mddev == NULL ||
1401 strcmp(mddev, chosen_name) != 0)
1402 pr_err("Merging with already-assembled %s\n",
1403 chosen_name);
1404 mdfd = open_dev_excl(mp->devnm);
1405 } else {
1406 int trustworthy = FOREIGN;
1407 name = content->name;
1408 switch (st->ss->match_home(st, c->homehost)
1409 ?: st->ss->match_home(st, "any")) {
1410 case 1:
1411 trustworthy = LOCAL;
1412 name = strchr(content->name, ':');
1413 if (name)
1414 name++;
1415 else
1416 name = content->name;
1417 break;
1418 }
1419 if (!auto_assem)
1420 /* If the array is listed in mdadm.conf or on
1421 * command line, then we trust the name
1422 * even if the array doesn't look local
1423 */
1424 trustworthy = LOCAL;
1425
1426 if (name[0] == 0 &&
1427 content->array.level == LEVEL_CONTAINER) {
1428 name = content->text_version;
1429 trustworthy = METADATA;
1430 }
1431
1432 if (name[0] && trustworthy != LOCAL &&
1433 ! c->require_homehost &&
1434 conf_name_is_free(name))
1435 trustworthy = LOCAL;
1436
1437 if (trustworthy == LOCAL &&
1438 strchr(name, ':'))
1439 /* Ignore 'host:' prefix of name */
1440 name = strchr(name, ':')+1;
1441
1442 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1443 chosen_name);
1444 }
1445 if (mdfd < 0) {
1446 st->ss->free_super(st);
1447 if (auto_assem)
1448 goto try_again;
1449 return 1;
1450 }
1451 mddev = chosen_name;
1452 if (get_linux_version() < 2004000 ||
1453 md_get_version(mdfd) < 9000) {
1454 pr_err("Assemble requires Linux 2.4 or later, and\n"
1455 " md driver version 0.90.0 or later.\n"
1456 " Upgrade your kernel or try --build\n");
1457 close(mdfd);
1458 return 1;
1459 }
1460 if (pre_exist == NULL) {
1461 if (mddev_busy(fd2devnm(mdfd))) {
1462 pr_err("%s already active, cannot restart it!\n",
1463 mddev);
1464 for (tmpdev = devlist ;
1465 tmpdev && tmpdev->used != 1;
1466 tmpdev = tmpdev->next)
1467 ;
1468 if (tmpdev && auto_assem)
1469 pr_err("%s needed for %s...\n",
1470 mddev, tmpdev->devname);
1471 close(mdfd);
1472 mdfd = -3;
1473 st->ss->free_super(st);
1474 if (auto_assem)
1475 goto try_again;
1476 return 1;
1477 }
1478 /* just incase it was started but has no content */
1479 ioctl(mdfd, STOP_ARRAY, NULL);
1480 }
1481
1482 #ifndef MDASSEMBLE
1483 if (content != &info) {
1484 /* This is a member of a container. Try starting the array. */
1485 int err;
1486 err = assemble_container_content(st, mdfd, content, c,
1487 chosen_name, NULL);
1488 close(mdfd);
1489 return err;
1490 }
1491 #endif
1492 /* Ok, no bad inconsistancy, we can try updating etc */
1493 devices = xcalloc(num_devs, sizeof(*devices));
1494 devmap = xcalloc(num_devs, content->array.raid_disks);
1495 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1496 c, content, mdfd, mddev,
1497 &most_recent, &bestcnt, &best, inargv);
1498 if (devcnt < 0)
1499 return 1;
1500
1501 if (devcnt == 0) {
1502 pr_err("no devices found for %s\n",
1503 mddev);
1504 if (st)
1505 st->ss->free_super(st);
1506 close(mdfd);
1507 free(devices);
1508 free(devmap);
1509 return 1;
1510 }
1511
1512 if (c->update && strcmp(c->update, "byteorder")==0)
1513 st->minor_version = 90;
1514
1515 st->ss->getinfo_super(st, content, NULL);
1516 clean = content->array.state & 1;
1517
1518 /* now we have some devices that might be suitable.
1519 * I wonder how many
1520 */
1521 avail = xcalloc(content->array.raid_disks, 1);
1522 okcnt = 0;
1523 replcnt = 0;
1524 sparecnt=0;
1525 rebuilding_cnt=0;
1526 for (i=0; i< bestcnt; i++) {
1527 int j = best[i];
1528 int event_margin = 1; /* always allow a difference of '1'
1529 * like the kernel does
1530 */
1531 if (j < 0) continue;
1532 /* note: we ignore error flags in multipath arrays
1533 * as they don't make sense
1534 */
1535 if (content->array.level != LEVEL_MULTIPATH)
1536 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1537 if (!(devices[j].i.disk.state
1538 & (1<<MD_DISK_FAULTY))) {
1539 devices[j].uptodate = 1;
1540 sparecnt++;
1541 }
1542 continue;
1543 }
1544 /* If this device thinks that 'most_recent' has failed, then
1545 * we must reject this device.
1546 */
1547 if (j != most_recent && !c->force &&
1548 content->array.raid_disks > 0 &&
1549 devices[most_recent].i.disk.raid_disk >= 0 &&
1550 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1551 if (c->verbose > -1)
1552 pr_err("ignoring %s as it reports %s as failed\n",
1553 devices[j].devname, devices[most_recent].devname);
1554 best[i] = -1;
1555 continue;
1556 }
1557 /* Require event counter to be same as, or just less than,
1558 * most recent. If it is bigger, it must be a stray spare and
1559 * should be ignored.
1560 */
1561 if (devices[j].i.events+event_margin >=
1562 devices[most_recent].i.events &&
1563 devices[j].i.events <=
1564 devices[most_recent].i.events
1565 ) {
1566 devices[j].uptodate = 1;
1567 if (i < content->array.raid_disks * 2) {
1568 if (devices[j].i.recovery_start == MaxSector ||
1569 (content->reshape_active &&
1570 i >= content->array.raid_disks - content->delta_disks)) {
1571 if (!avail[i/2]) {
1572 okcnt++;
1573 avail[i/2]=1;
1574 } else
1575 replcnt++;
1576 } else
1577 rebuilding_cnt++;
1578 } else
1579 sparecnt++;
1580 }
1581 }
1582 free(devmap);
1583 if (c->force) {
1584 int force_ok = force_array(content, devices, best, bestcnt,
1585 avail, most_recent, st, c);
1586 okcnt += force_ok;
1587 if (force_ok)
1588 was_forced = 1;
1589 }
1590 /* Now we want to look at the superblock which the kernel will base things on
1591 * and compare the devices that we think are working with the devices that the
1592 * superblock thinks are working.
1593 * If there are differences and --force is given, then update this chosen
1594 * superblock.
1595 */
1596 chosen_drive = -1;
1597 st->ss->free_super(st);
1598 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1599 int j = best[i];
1600 int fd;
1601
1602 if (j<0)
1603 continue;
1604 if (!devices[j].uptodate)
1605 continue;
1606 if (devices[j].i.events < devices[most_recent].i.events)
1607 continue;
1608 chosen_drive = j;
1609 if ((fd=dev_open(devices[j].devname,
1610 devices[j].included ? O_RDONLY
1611 : (O_RDONLY|O_EXCL)))< 0) {
1612 pr_err("Cannot open %s: %s\n",
1613 devices[j].devname, strerror(errno));
1614 close(mdfd);
1615 free(devices);
1616 return 1;
1617 }
1618 if (st->ss->load_super(st,fd, NULL)) {
1619 close(fd);
1620 pr_err("RAID superblock has disappeared from %s\n",
1621 devices[j].devname);
1622 close(mdfd);
1623 free(devices);
1624 return 1;
1625 }
1626 close(fd);
1627 }
1628 if (st->sb == NULL) {
1629 pr_err("No suitable drives found for %s\n", mddev);
1630 close(mdfd);
1631 free(devices);
1632 return 1;
1633 }
1634 st->ss->getinfo_super(st, content, NULL);
1635 #ifndef MDASSEMBLE
1636 sysfs_init(content, mdfd, NULL);
1637 #endif
1638 for (i=0; i<bestcnt; i++) {
1639 int j = best[i];
1640 unsigned int desired_state;
1641
1642 if (i >= content->array.raid_disks * 2)
1643 desired_state = 0;
1644 else if (i & 1)
1645 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1646 else
1647 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1648
1649 if (j<0)
1650 continue;
1651 if (!devices[j].uptodate)
1652 continue;
1653
1654 devices[j].i.disk.state = desired_state;
1655 if (!(devices[j].i.array.state & 1))
1656 clean = 0;
1657
1658 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1659 c->verbose, 0, NULL)) {
1660 if (c->force) {
1661 if (c->verbose >= 0)
1662 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1663 j, mddev, devices[j].devname);
1664 change = 1;
1665 } else {
1666 if (c->verbose >= -1)
1667 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1668 i, mddev, devices[j].devname);
1669 }
1670 }
1671 #if 0
1672 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1673 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1674 i, mddev);
1675 }
1676 #endif
1677 }
1678 if (c->force && !clean &&
1679 !enough(content->array.level, content->array.raid_disks,
1680 content->array.layout, clean,
1681 avail)) {
1682 change += st->ss->update_super(st, content, "force-array",
1683 devices[chosen_drive].devname, c->verbose,
1684 0, NULL);
1685 was_forced = 1;
1686 clean = 1;
1687 }
1688
1689 if (change) {
1690 int fd;
1691 fd = dev_open(devices[chosen_drive].devname,
1692 devices[chosen_drive].included ?
1693 O_RDWR : (O_RDWR|O_EXCL));
1694 if (fd < 0) {
1695 pr_err("Could not open %s for write - cannot Assemble array.\n",
1696 devices[chosen_drive].devname);
1697 close(mdfd);
1698 free(devices);
1699 return 1;
1700 }
1701 if (st->ss->store_super(st, fd)) {
1702 close(fd);
1703 pr_err("Could not re-write superblock on %s\n",
1704 devices[chosen_drive].devname);
1705 close(mdfd);
1706 free(devices);
1707 return 1;
1708 }
1709 if (c->verbose >= 0)
1710 pr_err("Marking array %s as 'clean'\n",
1711 mddev);
1712 close(fd);
1713 }
1714
1715 /* If we are in the middle of a reshape we may need to restore saved data
1716 * that was moved aside due to the reshape overwriting live data
1717 * The code of doing this lives in Grow.c
1718 */
1719 #ifndef MDASSEMBLE
1720 if (content->reshape_active &&
1721 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1722 int err = 0;
1723 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1724 if (c->verbose > 0)
1725 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1726 chosen_name);
1727 if (!c->backup_file)
1728 c->backup_file = locate_backup(content->sys_name);
1729 enable_fds(bestcnt/2);
1730 for (i = 0; i < bestcnt/2; i++) {
1731 int j = best[i*2];
1732 if (j >= 0) {
1733 fdlist[i] = dev_open(devices[j].devname,
1734 devices[j].included
1735 ? O_RDWR : (O_RDWR|O_EXCL));
1736 if (fdlist[i] < 0) {
1737 pr_err("Could not open %s for write - cannot Assemble array.\n",
1738 devices[j].devname);
1739 err = 1;
1740 break;
1741 }
1742 } else
1743 fdlist[i] = -1;
1744 }
1745 if (!err) {
1746 if (st->ss->external && st->ss->recover_backup)
1747 err = st->ss->recover_backup(st, content);
1748 else
1749 err = Grow_restart(st, content, fdlist, bestcnt/2,
1750 c->backup_file, c->verbose > 0);
1751 if (err && c->invalid_backup) {
1752 if (c->verbose > 0)
1753 pr_err("continuing without restoring backup\n");
1754 err = 0;
1755 }
1756 }
1757 while (i>0) {
1758 i--;
1759 if (fdlist[i]>=0) close(fdlist[i]);
1760 }
1761 free(fdlist);
1762 if (err) {
1763 pr_err("Failed to restore critical section for reshape, sorry.\n");
1764 if (c->backup_file == NULL)
1765 cont_err("Possibly you needed to specify the --backup-file\n");
1766 close(mdfd);
1767 free(devices);
1768 return err;
1769 }
1770 }
1771 #endif
1772
1773 /* Almost ready to actually *do* something */
1774 /* First, fill in the map, so that udev can find our name
1775 * as soon as we become active.
1776 */
1777 if (c->update && strcmp(c->update, "metadata")==0) {
1778 content->array.major_version = 1;
1779 content->array.minor_version = 0;
1780 strcpy(content->text_version, "1.0");
1781 }
1782
1783 map_update(&map, fd2devnm(mdfd), content->text_version,
1784 content->uuid, chosen_name);
1785
1786 rv = start_array(mdfd, mddev, content,
1787 st, ident, best, bestcnt,
1788 chosen_drive, devices, okcnt, sparecnt,
1789 rebuilding_cnt,
1790 c,
1791 clean, avail, start_partial_ok,
1792 pre_exist != NULL,
1793 was_forced);
1794 if (rv == 1 && !pre_exist)
1795 ioctl(mdfd, STOP_ARRAY, NULL);
1796 free(devices);
1797 map_unlock(&map);
1798 if (rv == 0) {
1799 wait_for(chosen_name, mdfd);
1800 close(mdfd);
1801 if (auto_assem) {
1802 int usecs = 1;
1803 /* There is a nasty race with 'mdadm --monitor'.
1804 * If it opens this device before we close it,
1805 * it gets an incomplete open on which IO
1806 * doesn't work and the capacity is
1807 * wrong.
1808 * If we reopen (to check for layered devices)
1809 * before --monitor closes, we loose.
1810 *
1811 * So: wait upto 1 second for there to be
1812 * a non-zero capacity.
1813 */
1814 while (usecs < 1000) {
1815 mdfd = open(mddev, O_RDONLY);
1816 if (mdfd >= 0) {
1817 unsigned long long size;
1818 if (get_dev_size(mdfd, NULL, &size) &&
1819 size > 0)
1820 break;
1821 close(mdfd);
1822 }
1823 usleep(usecs);
1824 usecs <<= 1;
1825 }
1826 }
1827 } else
1828 close(mdfd);
1829
1830 /* '2' means 'OK, but not started yet' */
1831 return rv == 2 ? 0 : rv;
1832 }
1833
1834 #ifndef MDASSEMBLE
1835 int assemble_container_content(struct supertype *st, int mdfd,
1836 struct mdinfo *content, struct context *c,
1837 char *chosen_name, int *result)
1838 {
1839 struct mdinfo *dev, *sra, *dev2;
1840 int working = 0, preexist = 0;
1841 int expansion = 0;
1842 struct map_ent *map = NULL;
1843 int old_raid_disks;
1844 int start_reshape;
1845 char *avail = NULL;
1846 int err;
1847
1848 sysfs_init(content, mdfd, NULL);
1849
1850 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1851 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1852 if (content->array.major_version == -1 &&
1853 content->array.minor_version == -2 &&
1854 c->readonly &&
1855 content->text_version[0] == '/')
1856 content->text_version[0] = '-';
1857 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1858 if (sra)
1859 sysfs_free(sra);
1860 return 1;
1861 }
1862 }
1863
1864 /* There are two types of reshape: container wide or sub-array specific
1865 * Check if metadata requests blocking container wide reshapes
1866 */
1867 start_reshape = (content->reshape_active &&
1868 !((content->reshape_active == CONTAINER_RESHAPE) &&
1869 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1870
1871 /* Block subarray here if it is under reshape now
1872 * Do not allow for any changes in this array
1873 */
1874 if (st->ss->external && content->recovery_blocked && start_reshape)
1875 block_subarray(content);
1876
1877 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1878 for (dev = content->devs; dev; dev = dev->next)
1879 if (dev2->disk.major == dev->disk.major &&
1880 dev2->disk.minor == dev->disk.minor)
1881 break;
1882 if (dev)
1883 continue;
1884 /* Don't want this one any more */
1885 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1886 errno == EBUSY) {
1887 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1888 sysfs_free(sra);
1889 return 1;
1890 }
1891 sysfs_set_str(sra, dev2, "state", "remove");
1892 }
1893 old_raid_disks = content->array.raid_disks - content->delta_disks;
1894 avail = xcalloc(content->array.raid_disks, 1);
1895 for (dev = content->devs; dev; dev = dev->next) {
1896 if (dev->disk.raid_disk >= 0)
1897 avail[dev->disk.raid_disk] = 1;
1898 if (sysfs_add_disk(content, dev, 1) == 0) {
1899 if (dev->disk.raid_disk >= old_raid_disks &&
1900 content->reshape_active)
1901 expansion++;
1902 else
1903 working++;
1904 } else if (errno == EEXIST)
1905 preexist++;
1906 }
1907 sysfs_free(sra);
1908 if (working + expansion == 0 && c->runstop <= 0) {
1909 free(avail);
1910 return 1;/* Nothing new, don't try to start */
1911 }
1912 map_update(&map, fd2devnm(mdfd),
1913 content->text_version,
1914 content->uuid, chosen_name);
1915
1916
1917 if (enough(content->array.level, content->array.raid_disks,
1918 content->array.layout, content->array.state & 1, avail) == 0) {
1919 if (c->export && result)
1920 *result |= INCR_NO;
1921 else if (c->verbose >= 0) {
1922 pr_err("%s assembled with %d device%s",
1923 chosen_name, preexist + working,
1924 preexist + working == 1 ? "":"s");
1925 if (preexist)
1926 fprintf(stderr, " (%d new)", working);
1927 fprintf(stderr, " but not started\n");
1928 }
1929 free(avail);
1930 return 1;
1931 }
1932 free(avail);
1933
1934 if (c->runstop <= 0 &&
1935 (working + preexist + expansion) <
1936 content->array.working_disks) {
1937 if (c->export && result)
1938 *result |= INCR_UNSAFE;
1939 else if (c->verbose >= 0) {
1940 pr_err("%s assembled with %d device%s",
1941 chosen_name, preexist + working,
1942 preexist + working == 1 ? "":"s");
1943 if (preexist)
1944 fprintf(stderr, " (%d new)", working);
1945 fprintf(stderr, " but not safe to start\n");
1946 }
1947 return 1;
1948 }
1949
1950
1951 if (start_reshape) {
1952 int spare = content->array.raid_disks + expansion;
1953 if (restore_backup(st, content,
1954 working,
1955 spare, &c->backup_file, c->verbose) == 1)
1956 return 1;
1957
1958 err = sysfs_set_str(content, NULL,
1959 "array_state", "readonly");
1960 if (err)
1961 return 1;
1962
1963 if (st->ss->external) {
1964 if (!mdmon_running(st->container_devnm))
1965 start_mdmon(st->container_devnm);
1966 ping_monitor(st->container_devnm);
1967 if (mdmon_running(st->container_devnm) &&
1968 st->update_tail == NULL)
1969 st->update_tail = &st->updates;
1970 }
1971
1972 err = Grow_continue(mdfd, st, content, c->backup_file,
1973 0, c->freeze_reshape);
1974 } else switch(content->array.level) {
1975 case LEVEL_LINEAR:
1976 case LEVEL_MULTIPATH:
1977 case 0:
1978 err = sysfs_set_str(content, NULL, "array_state",
1979 c->readonly ? "readonly" : "active");
1980 break;
1981 default:
1982 err = sysfs_set_str(content, NULL, "array_state",
1983 "readonly");
1984 /* start mdmon if needed. */
1985 if (!err) {
1986 if (!mdmon_running(st->container_devnm))
1987 start_mdmon(st->container_devnm);
1988 ping_monitor(st->container_devnm);
1989 }
1990 break;
1991 }
1992 if (!err)
1993 sysfs_set_safemode(content, content->safe_mode_delay);
1994
1995 /* Block subarray here if it is not reshaped now
1996 * It has be blocked a little later to allow mdmon to switch in
1997 * in to R/W state
1998 */
1999 if (st->ss->external && content->recovery_blocked &&
2000 !start_reshape)
2001 block_subarray(content);
2002
2003 if (c->export && result) {
2004 if (err)
2005 *result |= INCR_NO;
2006 else
2007 *result |= INCR_YES;
2008 } else if (c->verbose >= 0) {
2009 if (err)
2010 pr_err("array %s now has %d device%s",
2011 chosen_name, working + preexist,
2012 working + preexist == 1 ? "":"s");
2013 else
2014 pr_err("Started %s with %d device%s",
2015 chosen_name, working + preexist,
2016 working + preexist == 1 ? "":"s");
2017 if (preexist)
2018 fprintf(stderr, " (%d new)", working);
2019 if (expansion)
2020 fprintf(stderr, " ( + %d for expansion)",
2021 expansion);
2022 fprintf(stderr, "\n");
2023 }
2024 if (!err)
2025 wait_for(chosen_name, mdfd);
2026 return err;
2027 /* FIXME should have an O_EXCL and wait for read-auto */
2028 }
2029 #endif
Unnamed repository; edit this file 'description' to name the repository.