]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
mdadm: refactor write journal code in Assemble and Incremental
[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->ignore_hw_compat = 0),
254 tst->ss->load_super(tst, dfd,
255 report_mismatch ? devname : NULL)) {
256 if (report_mismatch)
257 pr_err("no RAID superblock on %s\n",
258 devname);
259 tmpdev->used = 2;
260 } else if (tst->ss->compare_super == NULL) {
261 if (report_mismatch)
262 pr_err("Cannot assemble %s metadata on %s\n",
263 tst->ss->name, devname);
264 tmpdev->used = 2;
265 } else if (auto_assem && st == NULL &&
266 !conf_test_metadata(tst->ss->name, (pol = devid_policy(stb.st_rdev)),
267 tst->ss->match_home(tst, c->homehost) == 1)) {
268 if (report_mismatch)
269 pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
270 devname, tst->ss->name);
271 tmpdev->used = 2;
272 }
273 }
274 if (dfd >= 0) close(dfd);
275 if (tmpdev->used == 2) {
276 if (auto_assem || !inargv)
277 /* Ignore unrecognised devices during auto-assembly */
278 goto loop;
279 if (ident->uuid_set || ident->name[0] ||
280 ident->super_minor != UnSet)
281 /* Ignore unrecognised device if looking for
282 * specific array */
283 goto loop;
284
285 pr_err("%s has no superblock - assembly aborted\n",
286 devname);
287 if (st)
288 st->ss->free_super(st);
289 dev_policy_free(pol);
290 domain_free(domains);
291 return -1;
292 }
293
294 if (found_container) {
295 /* tmpdev is a container. We need to be either
296 * looking for a member, or auto-assembling
297 */
298 /* should be safe to try an exclusive open now, we
299 * have rejected anything that some other mdadm might
300 * be looking at
301 */
302 dfd = dev_open(devname, O_RDONLY | O_EXCL);
303 if (dfd < 0) {
304 if (report_mismatch)
305 pr_err("%s is busy - skipping\n", devname);
306 goto loop;
307 }
308 close(dfd);
309
310 if (ident->container && ident->container[0] != '/') {
311 /* we have a uuid */
312 int uuid[4];
313
314 content = *contentp;
315 tst->ss->getinfo_super(tst, content, NULL);
316
317 if (!parse_uuid(ident->container, uuid) ||
318 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
319 if (report_mismatch)
320 pr_err("%s has wrong UUID to be required container\n",
321 devname);
322 goto loop;
323 }
324 }
325 /* It is worth looking inside this container.
326 */
327 if (c->verbose > 0)
328 pr_err("looking in container %s\n",
329 devname);
330
331 for (content = tst->ss->container_content(tst, NULL);
332 content;
333 content = content->next) {
334
335 if (!ident_matches(ident, content, tst,
336 c->homehost, c->require_homehost,
337 c->update,
338 report_mismatch ? devname : NULL))
339 /* message already printed */;
340 else if (is_member_busy(content->text_version)) {
341 if (report_mismatch)
342 pr_err("member %s in %s is already assembled\n",
343 content->text_version,
344 devname);
345 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
346 /* do not assemble arrays with unsupported configurations */
347 pr_err("Cannot activate member %s in %s.\n",
348 content->text_version,
349 devname);
350 } else
351 break;
352 }
353 if (!content) {
354 tmpdev->used = 2;
355 goto loop; /* empty container */
356 }
357
358 st = tst; tst = NULL;
359 if (!auto_assem && inargv && tmpdev->next != NULL) {
360 pr_err("%s is a container, but is not only device given: confused and aborting\n",
361 devname);
362 st->ss->free_super(st);
363 dev_policy_free(pol);
364 domain_free(domains);
365 return -1;
366 }
367 if (c->verbose > 0)
368 pr_err("found match on member %s in %s\n",
369 content->text_version, devname);
370
371 /* make sure we finished the loop */
372 tmpdev = NULL;
373 goto loop;
374 } else {
375 content = *contentp;
376 tst->ss->getinfo_super(tst, content, NULL);
377
378 if (!ident_matches(ident, content, tst,
379 c->homehost, c->require_homehost,
380 c->update,
381 report_mismatch ? devname : NULL))
382 goto loop;
383
384 if (auto_assem) {
385 /* Never auto-assemble things that conflict
386 * with mdadm.conf in some way
387 */
388 struct mddev_ident *match;
389 int rv = 0;
390
391 match = conf_match(tst, content, devname,
392 report_mismatch ? c->verbose : -1,
393 &rv);
394 if (!match && rv == 2)
395 goto loop;
396 if (match && match->devname &&
397 strcasecmp(match->devname, "<ignore>") == 0) {
398 if (report_mismatch)
399 pr_err("%s is a member of an explicitly ignored array\n",
400 devname);
401 goto loop;
402 }
403 if (match && !ident_matches(match, content, tst,
404 c->homehost, c->require_homehost,
405 c->update,
406 report_mismatch ? devname : NULL))
407 /* Array exists in mdadm.conf but some
408 * details don't match, so reject it
409 */
410 goto loop;
411 }
412
413 /* should be safe to try an exclusive open now, we
414 * have rejected anything that some other mdadm might
415 * be looking at
416 */
417 dfd = dev_open(devname, O_RDONLY | O_EXCL);
418 if (dfd < 0) {
419 if (report_mismatch)
420 pr_err("%s is busy - skipping\n", devname);
421 goto loop;
422 }
423 close(dfd);
424
425 if (st == NULL)
426 st = dup_super(tst);
427 if (st->minor_version == -1)
428 st->minor_version = tst->minor_version;
429
430 if (memcmp(content->uuid, uuid_zero,
431 sizeof(int[4])) == 0) {
432 /* this is a floating spare. It cannot define
433 * an array unless there are no more arrays of
434 * this type to be found. It can be included
435 * in an array of this type though.
436 */
437 tmpdev->used = 3;
438 goto loop;
439 }
440
441 if (st->ss != tst->ss ||
442 st->minor_version != tst->minor_version ||
443 st->ss->compare_super(st, tst) != 0) {
444 /* Some mismatch. If exactly one array matches this host,
445 * we can resolve on that one.
446 * Or, if we are auto assembling, we just ignore the second
447 * for now.
448 */
449 if (auto_assem)
450 goto loop;
451 if (c->homehost) {
452 int first = st->ss->match_home(st, c->homehost);
453 int last = tst->ss->match_home(tst, c->homehost);
454 if (first != last &&
455 (first == 1 || last == 1)) {
456 /* We can do something */
457 if (first) {/* just ignore this one */
458 if (report_mismatch)
459 pr_err("%s misses out due to wrong homehost\n",
460 devname);
461 goto loop;
462 } else { /* reject all those sofar */
463 struct mddev_dev *td;
464 if (report_mismatch)
465 pr_err("%s overrides previous devices due to good homehost\n",
466 devname);
467 for (td=devlist; td != tmpdev; td=td->next)
468 if (td->used == 1)
469 td->used = 0;
470 tmpdev->used = 1;
471 goto loop;
472 }
473 }
474 }
475 pr_err("superblock on %s doesn't match others - assembly aborted\n",
476 devname);
477 tst->ss->free_super(tst);
478 st->ss->free_super(st);
479 dev_policy_free(pol);
480 domain_free(domains);
481 return -1;
482 }
483 tmpdev->used = 1;
484 }
485 loop:
486 /* Collect domain information from members only */
487 if (tmpdev && tmpdev->used == 1) {
488 if (!pol)
489 pol = devid_policy(stb.st_rdev);
490 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
491 }
492 dev_policy_free(pol);
493 pol = NULL;
494 if (tst)
495 tst->ss->free_super(tst);
496 }
497
498 /* Check if we found some imsm spares but no members */
499 if ((auto_assem ||
500 (ident->uuid_set &&
501 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
502 (!st || !st->sb))
503 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
504 if (tmpdev->used != 3)
505 continue;
506 tmpdev->used = 1;
507 content = *contentp;
508
509 if (!st->sb) {
510 /* we need sb from one of the spares */
511 int dfd = dev_open(tmpdev->devname, O_RDONLY);
512 if (dfd < 0 ||
513 st->ss->load_super(st, dfd, NULL))
514 tmpdev->used = 2;
515 if (dfd > 0)
516 close(dfd);
517 }
518 }
519
520 /* Now reject spares that don't match domains of identified members */
521 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
522 struct stat stb;
523 if (tmpdev->used != 3)
524 continue;
525 if (stat(tmpdev->devname, &stb)< 0) {
526 pr_err("fstat failed for %s: %s\n",
527 tmpdev->devname, strerror(errno));
528 tmpdev->used = 2;
529 } else {
530 struct dev_policy *pol = devid_policy(stb.st_rdev);
531 int dt = domain_test(domains, pol, NULL);
532 if (inargv && dt != 0)
533 /* take this spare as domains match
534 * if there are any */
535 tmpdev->used = 1;
536 else if (!inargv && dt == 1)
537 /* device wasn't explicitly listed, so need
538 * explicit domain match - which we have */
539 tmpdev->used = 1;
540 else
541 /* if domains don't match mark as unused */
542 tmpdev->used = 0;
543 dev_policy_free(pol);
544 }
545 }
546 domain_free(domains);
547 *stp = st;
548 if (st && st->sb && content == *contentp)
549 st->ss->getinfo_super(st, content, NULL);
550 *contentp = content;
551
552 return num_devs;
553 }
554
555 struct devs {
556 char *devname;
557 int uptodate; /* set once we decide that this device is as
558 * recent as everything else in the array.
559 */
560 int included; /* set if the device is already in the array
561 * due to a previous '-I'
562 */
563 struct mdinfo i;
564 };
565
566 static int load_devices(struct devs *devices, char *devmap,
567 struct mddev_ident *ident, struct supertype **stp,
568 struct mddev_dev *devlist, struct context *c,
569 struct mdinfo *content,
570 int mdfd, char *mddev,
571 int *most_recentp, int *bestcntp, int **bestp,
572 int inargv)
573 {
574 struct mddev_dev *tmpdev;
575 int devcnt = 0;
576 int nextspare = 0;
577 #ifndef MDASSEMBLE
578 int bitmap_done = 0;
579 #endif
580 int most_recent = -1;
581 int bestcnt = 0;
582 int *best = *bestp;
583 struct supertype *st = *stp;
584
585 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) {
586 char *devname = tmpdev->devname;
587 struct stat stb;
588 struct supertype *tst;
589 int i;
590 int dfd;
591
592 if (tmpdev->used != 1)
593 continue;
594 /* looks like a good enough match to update the super block if needed */
595 #ifndef MDASSEMBLE
596 if (c->update) {
597 /* prepare useful information in info structures */
598 struct stat stb2;
599 int err;
600 fstat(mdfd, &stb2);
601
602 if (strcmp(c->update, "uuid")==0 &&
603 !ident->uuid_set) {
604 int rfd;
605 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
606 read(rfd, ident->uuid, 16) != 16) {
607 *(__u32*)(ident->uuid) = random();
608 *(__u32*)(ident->uuid+1) = random();
609 *(__u32*)(ident->uuid+2) = random();
610 *(__u32*)(ident->uuid+3) = random();
611 }
612 if (rfd >= 0) close(rfd);
613 }
614 dfd = dev_open(devname,
615 tmpdev->disposition == 'I'
616 ? O_RDWR : (O_RDWR|O_EXCL));
617
618 tst = dup_super(st);
619 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
620 pr_err("cannot re-read metadata from %s - aborting\n",
621 devname);
622 if (dfd >= 0)
623 close(dfd);
624 close(mdfd);
625 free(devices);
626 free(devmap);
627 tst->ss->free_super(tst);
628 free(tst);
629 *stp = st;
630 return -1;
631 }
632 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
633
634 memcpy(content->uuid, ident->uuid, 16);
635 strcpy(content->name, ident->name);
636 content->array.md_minor = minor(stb2.st_rdev);
637
638 if (strcmp(c->update, "byteorder") == 0)
639 err = 0;
640 else if (strcmp(c->update, "home-cluster") == 0) {
641 tst->cluster_name = c->homecluster;
642 err = tst->ss->write_bitmap(tst, dfd, NameUpdate);
643 } else if (strcmp(c->update, "nodes") == 0) {
644 tst->nodes = c->nodes;
645 err = tst->ss->write_bitmap(tst, dfd, NodeNumUpdate);
646 } else
647 err = tst->ss->update_super(tst, content, c->update,
648 devname, c->verbose,
649 ident->uuid_set,
650 c->homehost);
651 if (err < 0) {
652 if (err == -1)
653 pr_err("--update=%s not understood for %s metadata\n",
654 c->update, tst->ss->name);
655 tst->ss->free_super(tst);
656 free(tst);
657 close(mdfd);
658 close(dfd);
659 free(devices);
660 free(devmap);
661 *stp = st;
662 return -1;
663 }
664 if (strcmp(c->update, "uuid")==0 &&
665 !ident->uuid_set) {
666 ident->uuid_set = 1;
667 memcpy(ident->uuid, content->uuid, 16);
668 }
669 if (tst->ss->store_super(tst, dfd))
670 pr_err("Could not re-write superblock on %s.\n",
671 devname);
672
673 if (strcmp(c->update, "uuid")==0 &&
674 ident->bitmap_fd >= 0 && !bitmap_done) {
675 if (bitmap_update_uuid(ident->bitmap_fd,
676 content->uuid,
677 tst->ss->swapuuid) != 0)
678 pr_err("Could not update uuid on external bitmap.\n");
679 else
680 bitmap_done = 1;
681 }
682 } else
683 #endif
684 {
685 dfd = dev_open(devname,
686 tmpdev->disposition == 'I'
687 ? O_RDWR : (O_RDWR|O_EXCL));
688 tst = dup_super(st);
689
690 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
691 pr_err("cannot re-read metadata from %s - aborting\n",
692 devname);
693 if (dfd >= 0)
694 close(dfd);
695 close(mdfd);
696 free(devices);
697 free(devmap);
698 tst->ss->free_super(tst);
699 free(tst);
700 *stp = st;
701 return -1;
702 }
703 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
704 }
705
706 fstat(dfd, &stb);
707 close(dfd);
708
709 if (c->verbose > 0)
710 pr_err("%s is identified as a member of %s, slot %d%s.\n",
711 devname, mddev, content->disk.raid_disk,
712 (content->disk.state & (1<<MD_DISK_REPLACEMENT)) ? " replacement":"");
713 devices[devcnt].devname = devname;
714 devices[devcnt].uptodate = 0;
715 devices[devcnt].included = (tmpdev->disposition == 'I');
716 devices[devcnt].i = *content;
717 devices[devcnt].i.disk.major = major(stb.st_rdev);
718 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
719
720 if (devices[devcnt].i.disk.state == 6) {
721 if (most_recent < 0 ||
722 devices[devcnt].i.events
723 > devices[most_recent].i.events) {
724 struct supertype *tmp = tst;
725 tst = st;
726 st = tmp;
727 most_recent = devcnt;
728 }
729 }
730 tst->ss->free_super(tst);
731 free(tst);
732
733 if (content->array.level == LEVEL_MULTIPATH)
734 /* with multipath, the raid_disk from the superblock is meaningless */
735 i = devcnt;
736 else
737 i = devices[devcnt].i.disk.raid_disk;
738 if (i+1 == 0 || i == MD_DISK_ROLE_JOURNAL) {
739 if (nextspare < content->array.raid_disks*2)
740 nextspare = content->array.raid_disks*2;
741 i = nextspare++;
742 } else {
743 /* i is raid_disk - double it so there is room for
744 * replacements */
745 i *= 2;
746 if (devices[devcnt].i.disk.state & (1<<MD_DISK_REPLACEMENT))
747 i++;
748 if (i >= content->array.raid_disks*2 &&
749 i >= nextspare)
750 nextspare = i+1;
751 }
752 if (i < 10000) {
753 if (i >= bestcnt) {
754 int newbestcnt = i+10;
755 int *newbest = xmalloc(sizeof(int)*newbestcnt);
756 int c;
757 for (c=0; c < newbestcnt; c++)
758 if (c < bestcnt)
759 newbest[c] = best[c];
760 else
761 newbest[c] = -1;
762 if (best)free(best);
763 best = newbest;
764 bestcnt = newbestcnt;
765 }
766 if (best[i] >=0 &&
767 devices[best[i]].i.events
768 == devices[devcnt].i.events
769 && (devices[best[i]].i.disk.minor
770 != devices[devcnt].i.disk.minor)
771 && st->ss == &super0
772 && content->array.level != LEVEL_MULTIPATH) {
773 /* two different devices with identical superblock.
774 * Could be a mis-detection caused by overlapping
775 * partitions. fail-safe.
776 */
777 pr_err("WARNING %s and %s appear to have very similar superblocks.\n"
778 " If they are really different, please --zero the superblock on one\n"
779 " If they are the same or overlap, please remove one from %s.\n",
780 devices[best[i]].devname, devname,
781 inargv ? "the list" :
782 "the\n DEVICE list in mdadm.conf"
783 );
784 close(mdfd);
785 free(devices);
786 free(devmap);
787 *stp = st;
788 return -1;
789 }
790 if (best[i] == -1
791 || (devices[best[i]].i.events
792 < devices[devcnt].i.events))
793 best[i] = devcnt;
794 }
795 devcnt++;
796 }
797 if (most_recent >= 0)
798 *most_recentp = most_recent;
799 *bestcntp = bestcnt;
800 *bestp = best;
801 *stp = st;
802 return devcnt;
803 }
804
805 static int force_array(struct mdinfo *content,
806 struct devs *devices,
807 int *best, int bestcnt, char *avail,
808 int most_recent,
809 struct supertype *st,
810 struct context *c)
811 {
812 int okcnt = 0;
813 while (!enough(content->array.level, content->array.raid_disks,
814 content->array.layout, 1,
815 avail)
816 ||
817 (content->reshape_active && content->delta_disks > 0 &&
818 !enough(content->array.level, (content->array.raid_disks
819 - content->delta_disks),
820 content->new_layout, 1,
821 avail)
822 )) {
823 /* Choose the newest best drive which is
824 * not up-to-date, update the superblock
825 * and add it.
826 */
827 int fd;
828 struct supertype *tst;
829 unsigned long long current_events;
830 int chosen_drive = -1;
831 int i;
832
833 for (i = 0;
834 i < content->array.raid_disks * 2 && i < bestcnt;
835 i += 2) {
836 int j = best[i];
837 if (j < 0)
838 continue;
839 if (devices[j].uptodate)
840 continue;
841 if (devices[j].i.recovery_start != MaxSector) {
842 int delta;
843 if (!devices[j].i.reshape_active ||
844 devices[j].i.delta_disks <= 0)
845 continue;
846 /* When increasing number of devices, an
847 * added device also appears to be
848 * recovering. It is safe to include it
849 * as long as it won't be a source of
850 * data.
851 * For now, just allow for last data
852 * devices in RAID4 or last devices in RAID4/5/6.
853 */
854 delta = devices[j].i.delta_disks;
855 if (devices[j].i.array.level >= 4 &&
856 devices[j].i.array.level <= 6 &&
857 i/2 >= content->array.raid_disks - delta)
858 /* OK */;
859 else if (devices[j].i.array.level == 4 &&
860 i/2 >= content->array.raid_disks - delta - 1)
861 /* OK */;
862 else
863 continue;
864 }
865 if (chosen_drive < 0 ||
866 devices[j].i.events
867 > devices[chosen_drive].i.events)
868 chosen_drive = j;
869 }
870 if (chosen_drive < 0)
871 break;
872 current_events = devices[chosen_drive].i.events;
873 add_another:
874 if (c->verbose >= 0)
875 pr_err("forcing event count in %s(%d) from %d upto %d\n",
876 devices[chosen_drive].devname,
877 devices[chosen_drive].i.disk.raid_disk,
878 (int)(devices[chosen_drive].i.events),
879 (int)(devices[most_recent].i.events));
880 fd = dev_open(devices[chosen_drive].devname,
881 devices[chosen_drive].included ? O_RDWR
882 : (O_RDWR|O_EXCL));
883 if (fd < 0) {
884 pr_err("Couldn't open %s for write - not updating\n",
885 devices[chosen_drive].devname);
886 devices[chosen_drive].i.events = 0;
887 continue;
888 }
889 tst = dup_super(st);
890 if (tst->ss->load_super(tst,fd, NULL)) {
891 close(fd);
892 pr_err("RAID superblock disappeared from %s - not updating.\n",
893 devices[chosen_drive].devname);
894 devices[chosen_drive].i.events = 0;
895 continue;
896 }
897 content->events = devices[most_recent].i.events;
898 tst->ss->update_super(tst, content, "force-one",
899 devices[chosen_drive].devname, c->verbose,
900 0, NULL);
901
902 if (tst->ss->store_super(tst, fd)) {
903 close(fd);
904 pr_err("Could not re-write superblock on %s\n",
905 devices[chosen_drive].devname);
906 devices[chosen_drive].i.events = 0;
907 tst->ss->free_super(tst);
908 continue;
909 }
910 close(fd);
911 devices[chosen_drive].i.events = devices[most_recent].i.events;
912 devices[chosen_drive].uptodate = 1;
913 avail[chosen_drive] = 1;
914 okcnt++;
915 tst->ss->free_super(tst);
916 /* If there are any other drives of the same vintage,
917 * add them in as well. We can't lose and we might gain
918 */
919 for (i = 0;
920 i < content->array.raid_disks * 2 && i < bestcnt ;
921 i += 2) {
922 int j = best[i];
923 if (j >= 0 &&
924 !devices[j].uptodate &&
925 devices[j].i.recovery_start == MaxSector &&
926 devices[j].i.events == current_events) {
927 chosen_drive = j;
928 goto add_another;
929 }
930 }
931 }
932 return okcnt;
933 }
934
935 static int start_array(int mdfd,
936 char *mddev,
937 struct mdinfo *content,
938 struct supertype *st,
939 struct mddev_ident *ident,
940 int *best, int bestcnt,
941 int chosen_drive,
942 struct devs *devices,
943 unsigned int okcnt,
944 unsigned int sparecnt,
945 unsigned int rebuilding_cnt,
946 unsigned int journalcnt,
947 struct context *c,
948 int clean, char *avail,
949 int start_partial_ok,
950 int err_ok,
951 int was_forced
952 )
953 {
954 int rv;
955 int i;
956 unsigned int req_cnt;
957
958 if (content->journal_device_required && (content->journal_clean == 0)) {
959 if (!c->force) {
960 pr_err("Not safe to assemble with missing or stale journal device, consider --force.\n");
961 return 1;
962 }
963 pr_err("Journal is missing or stale, starting array read only.\n");
964 c->readonly = 1;
965 }
966
967 rv = set_array_info(mdfd, st, content);
968 if (rv && !err_ok) {
969 pr_err("failed to set array info for %s: %s\n",
970 mddev, strerror(errno));
971 return 1;
972 }
973 if (ident->bitmap_fd >= 0) {
974 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
975 pr_err("SET_BITMAP_FILE failed.\n");
976 return 1;
977 }
978 } else if (ident->bitmap_file) {
979 /* From config file */
980 int bmfd = open(ident->bitmap_file, O_RDWR);
981 if (bmfd < 0) {
982 pr_err("Could not open bitmap file %s\n",
983 ident->bitmap_file);
984 return 1;
985 }
986 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
987 pr_err("Failed to set bitmapfile for %s\n", mddev);
988 close(bmfd);
989 return 1;
990 }
991 close(bmfd);
992 }
993
994 /* First, add the raid disks, but add the chosen one last */
995 for (i=0; i<= bestcnt; i++) {
996 int j;
997 if (i < bestcnt) {
998 j = best[i];
999 if (j == chosen_drive)
1000 continue;
1001 } else
1002 j = chosen_drive;
1003
1004 if (j >= 0 && !devices[j].included) {
1005 int dfd = dev_open(devices[j].devname,
1006 O_RDWR|O_EXCL);
1007 if (dfd >= 0) {
1008 remove_partitions(dfd);
1009 close(dfd);
1010 }
1011 rv = add_disk(mdfd, st, content, &devices[j].i);
1012
1013 if (rv) {
1014 pr_err("failed to add %s to %s: %s\n",
1015 devices[j].devname,
1016 mddev,
1017 strerror(errno));
1018 if (i < content->array.raid_disks * 2
1019 || i == bestcnt)
1020 okcnt--;
1021 else
1022 sparecnt--;
1023 } else if (c->verbose > 0)
1024 pr_err("added %s to %s as %d%s%s\n",
1025 devices[j].devname, mddev,
1026 devices[j].i.disk.raid_disk,
1027 devices[j].uptodate?"":
1028 " (possibly out of date)",
1029 (devices[j].i.disk.state & (1<<MD_DISK_REPLACEMENT))?" replacement":"");
1030 } else if (j >= 0) {
1031 if (c->verbose > 0)
1032 pr_err("%s is already in %s as %d\n",
1033 devices[j].devname, mddev,
1034 devices[j].i.disk.raid_disk);
1035 } else if (c->verbose > 0 && i < content->array.raid_disks*2
1036 && (i&1) == 0)
1037 pr_err("no uptodate device for slot %d of %s\n",
1038 i/2, mddev);
1039 }
1040
1041 if (content->array.level == LEVEL_CONTAINER) {
1042 if (c->verbose >= 0) {
1043 pr_err("Container %s has been assembled with %d drive%s",
1044 mddev, okcnt+sparecnt+journalcnt,
1045 okcnt+sparecnt+journalcnt==1?"":"s");
1046 if (okcnt < (unsigned)content->array.raid_disks)
1047 fprintf(stderr, " (out of %d)",
1048 content->array.raid_disks);
1049 fprintf(stderr, "\n");
1050 }
1051
1052 if (st->ss->validate_container) {
1053 struct mdinfo *devices_list;
1054 struct mdinfo *info_devices = xmalloc(sizeof(struct mdinfo)*(okcnt+sparecnt));
1055 unsigned int count;
1056 devices_list = NULL;
1057 for (count = 0; count < okcnt+sparecnt; count++) {
1058 info_devices[count] = devices[count].i;
1059 info_devices[count].next = devices_list;
1060 devices_list = &info_devices[count];
1061 }
1062 if (st->ss->validate_container(devices_list))
1063 pr_err("Mismatch detected!\n");
1064 free(info_devices);
1065 }
1066
1067 st->ss->free_super(st);
1068 sysfs_uevent(content, "change");
1069 if (err_ok && okcnt < (unsigned)content->array.raid_disks)
1070 /* Was partial, is still partial, so signal an error
1071 * to ensure we don't retry */
1072 return 1;
1073 return 0;
1074 }
1075
1076 /* Get number of in-sync devices according to the superblock.
1077 * We must have this number to start the array without -s or -R
1078 */
1079 req_cnt = content->array.working_disks;
1080
1081 if (c->runstop == 1 ||
1082 (c->runstop <= 0 &&
1083 ( enough(content->array.level, content->array.raid_disks,
1084 content->array.layout, clean, avail) &&
1085 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1086 ))) {
1087 /* This array is good-to-go.
1088 * If a reshape is in progress then we might need to
1089 * continue monitoring it. In that case we start
1090 * it read-only and let the grow code make it writable.
1091 */
1092 int rv;
1093 #ifndef MDASSEMBLE
1094 if (content->reshape_active &&
1095 !(content->reshape_active & RESHAPE_NO_BACKUP) &&
1096 content->delta_disks <= 0) {
1097 if (!c->backup_file) {
1098 pr_err("%s: Need a backup file to complete reshape of this array.\n",
1099 mddev);
1100 pr_err("Please provided one with \"--backup-file=...\"\n");
1101 if (c->update &&
1102 strcmp(c->update, "revert-reshape") == 0)
1103 pr_err("(Don't specify --update=revert-reshape again, that part succeeded.)\n");
1104 return 1;
1105 }
1106 rv = sysfs_set_str(content, NULL,
1107 "array_state", "readonly");
1108 if (rv == 0)
1109 rv = Grow_continue(mdfd, st, content,
1110 c->backup_file, 0,
1111 c->freeze_reshape);
1112 } else if (c->readonly &&
1113 sysfs_attribute_available(
1114 content, NULL, "array_state")) {
1115 rv = sysfs_set_str(content, NULL,
1116 "array_state", "readonly");
1117 } else
1118 #endif
1119 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1120 reopen_mddev(mdfd); /* drop O_EXCL */
1121 if (rv == 0) {
1122 if (c->verbose >= 0) {
1123 pr_err("%s has been started with %d drive%s",
1124 mddev, okcnt, okcnt==1?"":"s");
1125 if (okcnt < (unsigned)content->array.raid_disks)
1126 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1127 if (rebuilding_cnt)
1128 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1129 if (sparecnt)
1130 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1131 if (content->journal_clean)
1132 fprintf(stderr, " and %d journal", journalcnt);
1133 fprintf(stderr, ".\n");
1134 }
1135 if (content->reshape_active &&
1136 content->array.level >= 4 &&
1137 content->array.level <= 6) {
1138 /* might need to increase the size
1139 * of the stripe cache - default is 256
1140 */
1141 int chunk_size = content->array.chunk_size;
1142 if (content->reshape_active &&
1143 content->new_chunk > chunk_size)
1144 chunk_size = content->new_chunk;
1145 if (256 < 4 * ((chunk_size+4065)/4096)) {
1146 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1147 if (sra)
1148 sysfs_set_num(sra, NULL,
1149 "stripe_cache_size",
1150 (4 * chunk_size / 4096) + 1);
1151 sysfs_free(sra);
1152 }
1153 }
1154 if (okcnt < (unsigned)content->array.raid_disks) {
1155 /* If any devices did not get added
1156 * because the kernel rejected them based
1157 * on event count, try adding them
1158 * again providing the action policy is
1159 * 're-add' or greater. The bitmap
1160 * might allow them to be included, or
1161 * they will become spares.
1162 */
1163 for (i = 0; i < bestcnt; i++) {
1164 int j = best[i];
1165 if (j >= 0 && !devices[j].uptodate) {
1166 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1167 continue;
1168 rv = add_disk(mdfd, st, content,
1169 &devices[j].i);
1170 if (rv == 0 && c->verbose >= 0)
1171 pr_err("%s has been re-added.\n",
1172 devices[j].devname);
1173 }
1174 }
1175 }
1176 if (content->array.level == 6 &&
1177 okcnt + 1 == (unsigned)content->array.raid_disks &&
1178 was_forced) {
1179 struct mdinfo *sra = sysfs_read(mdfd, NULL, 0);
1180 if (sra)
1181 sysfs_set_str(sra, NULL,
1182 "sync_action", "repair");
1183 sysfs_free(sra);
1184 }
1185 return 0;
1186 }
1187 pr_err("failed to RUN_ARRAY %s: %s\n",
1188 mddev, strerror(errno));
1189
1190 if (!enough(content->array.level, content->array.raid_disks,
1191 content->array.layout, 1, avail))
1192 pr_err("Not enough devices to start the array.\n");
1193 else if (!enough(content->array.level,
1194 content->array.raid_disks,
1195 content->array.layout, clean,
1196 avail))
1197 pr_err("Not enough devices to start the array while not clean - consider --force.\n");
1198
1199 return 1;
1200 }
1201 if (c->runstop == -1) {
1202 pr_err("%s assembled from %d drive%s",
1203 mddev, okcnt, okcnt==1?"":"s");
1204 if (okcnt != (unsigned)content->array.raid_disks)
1205 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1206 fprintf(stderr, ", but not started.\n");
1207 return 2;
1208 }
1209 if (c->verbose >= -1) {
1210 pr_err("%s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1211 if (rebuilding_cnt)
1212 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1213 if (sparecnt)
1214 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1215 if (!enough(content->array.level, content->array.raid_disks,
1216 content->array.layout, 1, avail))
1217 fprintf(stderr, " - not enough to start the array.\n");
1218 else if (!enough(content->array.level,
1219 content->array.raid_disks,
1220 content->array.layout, clean,
1221 avail))
1222 fprintf(stderr, " - not enough to start the array while not clean - consider --force.\n");
1223 else {
1224 if (req_cnt == (unsigned)content->array.raid_disks)
1225 fprintf(stderr, " - need all %d to start it", req_cnt);
1226 else
1227 fprintf(stderr, " - need %d to start", req_cnt);
1228 fprintf(stderr, " (use --run to insist).\n");
1229 }
1230 }
1231 return 1;
1232 }
1233
1234 int Assemble(struct supertype *st, char *mddev,
1235 struct mddev_ident *ident,
1236 struct mddev_dev *devlist,
1237 struct context *c)
1238 {
1239 /*
1240 * The task of Assemble is to find a collection of
1241 * devices that should (according to their superblocks)
1242 * form an array, and to give this collection to the MD driver.
1243 * In Linux-2.4 and later, this involves submitting a
1244 * SET_ARRAY_INFO ioctl with no arg - to prepare
1245 * the array - and then submit a number of
1246 * ADD_NEW_DISK ioctls to add disks into
1247 * the array. Finally RUN_ARRAY might
1248 * be submitted to start the array.
1249 *
1250 * Much of the work of Assemble is in finding and/or
1251 * checking the disks to make sure they look right.
1252 *
1253 * If mddev is not set, then scan must be set and we
1254 * read through the config file for dev+uuid mapping
1255 * We recurse, setting mddev, for each device that
1256 * - isn't running
1257 * - has a valid uuid (or any uuid if !uuidset)
1258 *
1259 * If mddev is set, we try to determine state of md.
1260 * check version - must be at least 0.90.0
1261 * check kernel version. must be at least 2.4.
1262 * If not, we can possibly fall back on START_ARRAY
1263 * Try to GET_ARRAY_INFO.
1264 * If possible, give up
1265 * If not, try to STOP_ARRAY just to make sure
1266 *
1267 * If !uuidset and scan, look in conf-file for uuid
1268 * If not found, give up
1269 * If !devlist and scan and uuidset, get list of devs from conf-file
1270 *
1271 * For each device:
1272 * Check superblock - discard if bad
1273 * Check uuid (set if we don't have one) - discard if no match
1274 * Check superblock similarity if we have a superblock - discard if different
1275 * Record events, devicenum
1276 * This should give us a list of devices for the array
1277 * We should collect the most recent event number
1278 *
1279 * Count disks with recent enough event count
1280 * While force && !enough disks
1281 * Choose newest rejected disks, update event count
1282 * mark clean and rewrite superblock
1283 * If recent kernel:
1284 * SET_ARRAY_INFO
1285 * foreach device with recent events : ADD_NEW_DISK
1286 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
1287 * If old kernel:
1288 * Check the device numbers in superblock are right
1289 * update superblock if any changes
1290 * START_ARRAY
1291 *
1292 */
1293 int rv;
1294 int mdfd;
1295 int clean;
1296 int auto_assem = (mddev == NULL && !ident->uuid_set &&
1297 ident->super_minor == UnSet && ident->name[0] == 0
1298 && (ident->container == NULL || ident->member == NULL));
1299 struct devs *devices;
1300 char *devmap;
1301 int *best = NULL; /* indexed by raid_disk */
1302 int bestcnt = 0;
1303 int devcnt;
1304 unsigned int okcnt, sparecnt, rebuilding_cnt, replcnt, journalcnt;
1305 int i;
1306 int was_forced = 0;
1307 int most_recent = 0;
1308 int chosen_drive;
1309 int change = 0;
1310 int inargv = 0;
1311 int start_partial_ok = (c->runstop >= 0) &&
1312 (c->force || devlist==NULL || auto_assem);
1313 int num_devs;
1314 struct mddev_dev *tmpdev;
1315 struct mdinfo info;
1316 struct mdinfo *content = NULL;
1317 struct mdinfo *pre_exist = NULL;
1318 char *avail;
1319 char *name = NULL;
1320 char chosen_name[1024];
1321 struct map_ent *map = NULL;
1322 struct map_ent *mp;
1323
1324 /*
1325 * If any subdevs are listed, then any that don't
1326 * match ident are discarded. Remainder must all match and
1327 * become the array.
1328 * If no subdevs, then we scan all devices in the config file, but
1329 * there must be something in the identity
1330 */
1331
1332 if (!devlist &&
1333 ident->uuid_set == 0 &&
1334 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
1335 ident->name[0] == 0 &&
1336 (ident->container == NULL || ident->member == NULL) &&
1337 ident->devices == NULL) {
1338 pr_err("No identity information available for %s - cannot assemble.\n",
1339 mddev ? mddev : "further assembly");
1340 return 1;
1341 }
1342
1343 if (devlist == NULL)
1344 devlist = conf_get_devs();
1345 else if (mddev)
1346 inargv = 1;
1347
1348 try_again:
1349 /* We come back here when doing auto-assembly and attempting some
1350 * set of devices failed. Those are now marked as ->used==2 and
1351 * we ignore them and try again
1352 */
1353 if (!st && ident->st)
1354 st = ident->st;
1355 if (c->verbose>0)
1356 pr_err("looking for devices for %s\n",
1357 mddev ? mddev : "further assembly");
1358
1359 content = &info;
1360 if (st && c->force)
1361 st->ignore_hw_compat = 1;
1362 num_devs = select_devices(devlist, ident, &st, &content, c,
1363 inargv, auto_assem);
1364 if (num_devs < 0)
1365 return 1;
1366
1367 if (!st || !st->sb || !content)
1368 return 2;
1369
1370 /* We have a full set of devices - we now need to find the
1371 * array device.
1372 * However there is a risk that we are racing with "mdadm -I"
1373 * and the array is already partially assembled - we will have
1374 * rejected any devices already in this address.
1375 * So we take a lock on the map file - to prevent further races -
1376 * and look for the uuid in there. If found and the array is
1377 * active, we abort. If found and the array is not active
1378 * we commit to that md device and add all the contained devices
1379 * to our list. We flag them so that we don't try to re-add,
1380 * but can remove if they turn out to not be wanted.
1381 */
1382 if (map_lock(&map))
1383 pr_err("failed to get exclusive lock on mapfile - continue anyway...\n");
1384 if (c->update && strcmp(c->update,"uuid") == 0)
1385 mp = NULL;
1386 else
1387 mp = map_by_uuid(&map, content->uuid);
1388 if (mp) {
1389 struct mdinfo *dv;
1390 /* array already exists. */
1391 pre_exist = sysfs_read(-1, mp->devnm, GET_LEVEL|GET_DEVS);
1392 if (pre_exist->array.level != UnSet) {
1393 pr_err("Found some drive for an array that is already active: %s\n",
1394 mp->path);
1395 pr_err("giving up.\n");
1396 return 1;
1397 }
1398 for (dv = pre_exist->devs; dv; dv = dv->next) {
1399 /* We want to add this device to our list,
1400 * but it could already be there if "mdadm -I"
1401 * started *after* we checked for O_EXCL.
1402 * If we add it to the top of the list
1403 * it will be preferred over later copies.
1404 */
1405 struct mddev_dev *newdev;
1406 char *devname = map_dev(dv->disk.major,
1407 dv->disk.minor,
1408 0);
1409 if (!devname)
1410 continue;
1411 newdev = xmalloc(sizeof(*newdev));
1412 newdev->devname = devname;
1413 newdev->disposition = 'I';
1414 newdev->used = 1;
1415 newdev->next = devlist;
1416 devlist = newdev;
1417 num_devs++;
1418 }
1419 strcpy(chosen_name, mp->path);
1420 if (c->verbose > 0 || mddev == NULL ||
1421 strcmp(mddev, chosen_name) != 0)
1422 pr_err("Merging with already-assembled %s\n",
1423 chosen_name);
1424 mdfd = open_dev_excl(mp->devnm);
1425 } else {
1426 int trustworthy = FOREIGN;
1427 name = content->name;
1428 switch (st->ss->match_home(st, c->homehost)
1429 ?: st->ss->match_home(st, "any")) {
1430 case 1:
1431 trustworthy = LOCAL;
1432 name = strchr(content->name, ':');
1433 if (name)
1434 name++;
1435 else
1436 name = content->name;
1437 break;
1438 }
1439 if (!auto_assem)
1440 /* If the array is listed in mdadm.conf or on
1441 * command line, then we trust the name
1442 * even if the array doesn't look local
1443 */
1444 trustworthy = LOCAL;
1445
1446 if (name[0] == 0 &&
1447 content->array.level == LEVEL_CONTAINER) {
1448 name = content->text_version;
1449 trustworthy = METADATA;
1450 }
1451
1452 if (name[0] && trustworthy != LOCAL &&
1453 ! c->require_homehost &&
1454 conf_name_is_free(name))
1455 trustworthy = LOCAL;
1456
1457 if (trustworthy == LOCAL &&
1458 strchr(name, ':'))
1459 /* Ignore 'host:' prefix of name */
1460 name = strchr(name, ':')+1;
1461
1462 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
1463 chosen_name);
1464 }
1465 if (mdfd < 0) {
1466 st->ss->free_super(st);
1467 if (auto_assem)
1468 goto try_again;
1469 return 1;
1470 }
1471 mddev = chosen_name;
1472 if (get_linux_version() < 2004000 ||
1473 md_get_version(mdfd) < 9000) {
1474 pr_err("Assemble requires Linux 2.4 or later, and\n"
1475 " md driver version 0.90.0 or later.\n"
1476 " Upgrade your kernel or try --build\n");
1477 close(mdfd);
1478 return 1;
1479 }
1480 if (pre_exist == NULL) {
1481 if (mddev_busy(fd2devnm(mdfd))) {
1482 pr_err("%s already active, cannot restart it!\n",
1483 mddev);
1484 for (tmpdev = devlist ;
1485 tmpdev && tmpdev->used != 1;
1486 tmpdev = tmpdev->next)
1487 ;
1488 if (tmpdev && auto_assem)
1489 pr_err("%s needed for %s...\n",
1490 mddev, tmpdev->devname);
1491 close(mdfd);
1492 mdfd = -3;
1493 st->ss->free_super(st);
1494 if (auto_assem)
1495 goto try_again;
1496 return 1;
1497 }
1498 /* just incase it was started but has no content */
1499 ioctl(mdfd, STOP_ARRAY, NULL);
1500 }
1501
1502 #ifndef MDASSEMBLE
1503 if (content != &info) {
1504 /* This is a member of a container. Try starting the array. */
1505 int err;
1506 err = assemble_container_content(st, mdfd, content, c,
1507 chosen_name, NULL);
1508 close(mdfd);
1509 return err;
1510 }
1511 #endif
1512 /* Ok, no bad inconsistancy, we can try updating etc */
1513 devices = xcalloc(num_devs, sizeof(*devices));
1514 devmap = xcalloc(num_devs, content->array.raid_disks);
1515 devcnt = load_devices(devices, devmap, ident, &st, devlist,
1516 c, content, mdfd, mddev,
1517 &most_recent, &bestcnt, &best, inargv);
1518 if (devcnt < 0)
1519 return 1;
1520
1521 if (devcnt == 0) {
1522 pr_err("no devices found for %s\n",
1523 mddev);
1524 if (st)
1525 st->ss->free_super(st);
1526 close(mdfd);
1527 free(devices);
1528 free(devmap);
1529 return 1;
1530 }
1531
1532 if (c->update && strcmp(c->update, "byteorder")==0)
1533 st->minor_version = 90;
1534
1535 st->ss->getinfo_super(st, content, NULL);
1536 clean = content->array.state & 1;
1537
1538 /* now we have some devices that might be suitable.
1539 * I wonder how many
1540 */
1541 avail = xcalloc(content->array.raid_disks, 1);
1542 okcnt = 0;
1543 replcnt = 0;
1544 sparecnt=0;
1545 journalcnt=0;
1546 rebuilding_cnt=0;
1547 for (i=0; i< bestcnt; i++) {
1548 int j = best[i];
1549 int event_margin = 1; /* always allow a difference of '1'
1550 * like the kernel does
1551 */
1552 if (j < 0) continue;
1553 /* note: we ignore error flags in multipath arrays
1554 * as they don't make sense
1555 */
1556 if (content->array.level != LEVEL_MULTIPATH) {
1557 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL)) {
1558 if (content->journal_device_required)
1559 journalcnt++;
1560 else /* unexpected journal, mark as faulty */
1561 devices[j].i.disk.state |= (1<<MD_DISK_FAULTY);
1562 } else if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
1563 if (!(devices[j].i.disk.state
1564 & (1<<MD_DISK_FAULTY))) {
1565 devices[j].uptodate = 1;
1566 sparecnt++;
1567 }
1568 continue;
1569 }
1570 }
1571 /* If this device thinks that 'most_recent' has failed, then
1572 * we must reject this device.
1573 */
1574 if (j != most_recent && !c->force &&
1575 content->array.raid_disks > 0 &&
1576 devices[most_recent].i.disk.raid_disk >= 0 &&
1577 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
1578 if (c->verbose > -1)
1579 pr_err("ignoring %s as it reports %s as failed\n",
1580 devices[j].devname, devices[most_recent].devname);
1581 best[i] = -1;
1582 continue;
1583 }
1584 /* Require event counter to be same as, or just less than,
1585 * most recent. If it is bigger, it must be a stray spare and
1586 * should be ignored.
1587 */
1588 if (devices[j].i.events+event_margin >=
1589 devices[most_recent].i.events &&
1590 devices[j].i.events <=
1591 devices[most_recent].i.events
1592 ) {
1593 devices[j].uptodate = 1;
1594 if (devices[j].i.disk.state & (1<<MD_DISK_JOURNAL))
1595 content->journal_clean = 1;
1596 if (i < content->array.raid_disks * 2) {
1597 if (devices[j].i.recovery_start == MaxSector ||
1598 (content->reshape_active &&
1599 i >= content->array.raid_disks - content->delta_disks)) {
1600 if (!avail[i/2]) {
1601 okcnt++;
1602 avail[i/2]=1;
1603 } else
1604 replcnt++;
1605 } else
1606 rebuilding_cnt++;
1607 } else if (devices[j].i.disk.raid_disk != MD_DISK_ROLE_JOURNAL)
1608 sparecnt++;
1609 }
1610 }
1611 free(devmap);
1612 if (c->force) {
1613 int force_ok = force_array(content, devices, best, bestcnt,
1614 avail, most_recent, st, c);
1615 okcnt += force_ok;
1616 if (force_ok)
1617 was_forced = 1;
1618 }
1619 /* Now we want to look at the superblock which the kernel will base things on
1620 * and compare the devices that we think are working with the devices that the
1621 * superblock thinks are working.
1622 * If there are differences and --force is given, then update this chosen
1623 * superblock.
1624 */
1625 chosen_drive = -1;
1626 st->ss->free_super(st);
1627 for (i=0; chosen_drive < 0 && i<bestcnt; i+=2) {
1628 int j = best[i];
1629 int fd;
1630
1631 if (j<0)
1632 continue;
1633 if (!devices[j].uptodate)
1634 continue;
1635 if (devices[j].i.events < devices[most_recent].i.events)
1636 continue;
1637 chosen_drive = j;
1638 if ((fd=dev_open(devices[j].devname,
1639 devices[j].included ? O_RDONLY
1640 : (O_RDONLY|O_EXCL)))< 0) {
1641 pr_err("Cannot open %s: %s\n",
1642 devices[j].devname, strerror(errno));
1643 close(mdfd);
1644 free(devices);
1645 return 1;
1646 }
1647 if (st->ss->load_super(st,fd, NULL)) {
1648 close(fd);
1649 pr_err("RAID superblock has disappeared from %s\n",
1650 devices[j].devname);
1651 close(mdfd);
1652 free(devices);
1653 return 1;
1654 }
1655 close(fd);
1656 }
1657 if (st->sb == NULL) {
1658 pr_err("No suitable drives found for %s\n", mddev);
1659 close(mdfd);
1660 free(devices);
1661 return 1;
1662 }
1663 st->ss->getinfo_super(st, content, NULL);
1664 #ifndef MDASSEMBLE
1665 sysfs_init(content, mdfd, NULL);
1666 #endif
1667 for (i=0; i<bestcnt; i++) {
1668 int j = best[i];
1669 unsigned int desired_state;
1670
1671 if (devices[j].i.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
1672 desired_state = (1<<MD_DISK_JOURNAL);
1673 else if (i >= content->array.raid_disks * 2)
1674 desired_state = 0;
1675 else if (i & 1)
1676 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_REPLACEMENT);
1677 else
1678 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1679
1680 if (j<0)
1681 continue;
1682 if (!devices[j].uptodate)
1683 continue;
1684
1685 devices[j].i.disk.state = desired_state;
1686 if (!(devices[j].i.array.state & 1))
1687 clean = 0;
1688
1689 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1690 c->verbose, 0, NULL)) {
1691 if (c->force) {
1692 if (c->verbose >= 0)
1693 pr_err("clearing FAULTY flag for device %d in %s for %s\n",
1694 j, mddev, devices[j].devname);
1695 change = 1;
1696 } else {
1697 if (c->verbose >= -1)
1698 pr_err("device %d in %s has wrong state in superblock, but %s seems ok\n",
1699 i, mddev, devices[j].devname);
1700 }
1701 }
1702 #if 0
1703 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1704 pr_err("devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1705 i, mddev);
1706 }
1707 #endif
1708 }
1709 if (c->force && !clean &&
1710 !enough(content->array.level, content->array.raid_disks,
1711 content->array.layout, clean,
1712 avail)) {
1713 change += st->ss->update_super(st, content, "force-array",
1714 devices[chosen_drive].devname, c->verbose,
1715 0, NULL);
1716 was_forced = 1;
1717 clean = 1;
1718 }
1719
1720 if (change) {
1721 int fd;
1722 fd = dev_open(devices[chosen_drive].devname,
1723 devices[chosen_drive].included ?
1724 O_RDWR : (O_RDWR|O_EXCL));
1725 if (fd < 0) {
1726 pr_err("Could not open %s for write - cannot Assemble array.\n",
1727 devices[chosen_drive].devname);
1728 close(mdfd);
1729 free(devices);
1730 return 1;
1731 }
1732 if (st->ss->store_super(st, fd)) {
1733 close(fd);
1734 pr_err("Could not re-write superblock on %s\n",
1735 devices[chosen_drive].devname);
1736 close(mdfd);
1737 free(devices);
1738 return 1;
1739 }
1740 if (c->verbose >= 0)
1741 pr_err("Marking array %s as 'clean'\n",
1742 mddev);
1743 close(fd);
1744 }
1745
1746 /* If we are in the middle of a reshape we may need to restore saved data
1747 * that was moved aside due to the reshape overwriting live data
1748 * The code of doing this lives in Grow.c
1749 */
1750 #ifndef MDASSEMBLE
1751 if (content->reshape_active &&
1752 !(content->reshape_active & RESHAPE_NO_BACKUP)) {
1753 int err = 0;
1754 int *fdlist = xmalloc(sizeof(int)* bestcnt);
1755 if (c->verbose > 0)
1756 pr_err("%s has an active reshape - checking if critical section needs to be restored\n",
1757 chosen_name);
1758 if (!c->backup_file)
1759 c->backup_file = locate_backup(content->sys_name);
1760 enable_fds(bestcnt/2);
1761 for (i = 0; i < bestcnt/2; i++) {
1762 int j = best[i*2];
1763 if (j >= 0) {
1764 fdlist[i] = dev_open(devices[j].devname,
1765 devices[j].included
1766 ? O_RDWR : (O_RDWR|O_EXCL));
1767 if (fdlist[i] < 0) {
1768 pr_err("Could not open %s for write - cannot Assemble array.\n",
1769 devices[j].devname);
1770 err = 1;
1771 break;
1772 }
1773 } else
1774 fdlist[i] = -1;
1775 }
1776 if (!err) {
1777 if (st->ss->external && st->ss->recover_backup)
1778 err = st->ss->recover_backup(st, content);
1779 else
1780 err = Grow_restart(st, content, fdlist, bestcnt/2,
1781 c->backup_file, c->verbose > 0);
1782 if (err && c->invalid_backup) {
1783 if (c->verbose > 0)
1784 pr_err("continuing without restoring backup\n");
1785 err = 0;
1786 }
1787 }
1788 while (i>0) {
1789 i--;
1790 if (fdlist[i]>=0) close(fdlist[i]);
1791 }
1792 free(fdlist);
1793 if (err) {
1794 pr_err("Failed to restore critical section for reshape, sorry.\n");
1795 if (c->backup_file == NULL)
1796 cont_err("Possibly you needed to specify the --backup-file\n");
1797 close(mdfd);
1798 free(devices);
1799 return err;
1800 }
1801 }
1802 #endif
1803
1804 /* Almost ready to actually *do* something */
1805 /* First, fill in the map, so that udev can find our name
1806 * as soon as we become active.
1807 */
1808 if (c->update && strcmp(c->update, "metadata")==0) {
1809 content->array.major_version = 1;
1810 content->array.minor_version = 0;
1811 strcpy(content->text_version, "1.0");
1812 }
1813
1814 map_update(&map, fd2devnm(mdfd), content->text_version,
1815 content->uuid, chosen_name);
1816
1817 rv = start_array(mdfd, mddev, content,
1818 st, ident, best, bestcnt,
1819 chosen_drive, devices, okcnt, sparecnt,
1820 rebuilding_cnt, journalcnt,
1821 c,
1822 clean, avail, start_partial_ok,
1823 pre_exist != NULL,
1824 was_forced);
1825 if (rv == 1 && !pre_exist)
1826 ioctl(mdfd, STOP_ARRAY, NULL);
1827 free(devices);
1828 map_unlock(&map);
1829 if (rv == 0) {
1830 wait_for(chosen_name, mdfd);
1831 close(mdfd);
1832 if (auto_assem) {
1833 int usecs = 1;
1834 /* There is a nasty race with 'mdadm --monitor'.
1835 * If it opens this device before we close it,
1836 * it gets an incomplete open on which IO
1837 * doesn't work and the capacity is
1838 * wrong.
1839 * If we reopen (to check for layered devices)
1840 * before --monitor closes, we loose.
1841 *
1842 * So: wait upto 1 second for there to be
1843 * a non-zero capacity.
1844 */
1845 while (usecs < 1000) {
1846 mdfd = open(mddev, O_RDONLY);
1847 if (mdfd >= 0) {
1848 unsigned long long size;
1849 if (get_dev_size(mdfd, NULL, &size) &&
1850 size > 0)
1851 break;
1852 close(mdfd);
1853 }
1854 usleep(usecs);
1855 usecs <<= 1;
1856 }
1857 }
1858 } else
1859 close(mdfd);
1860
1861 /* '2' means 'OK, but not started yet' */
1862 return rv == 2 ? 0 : rv;
1863 }
1864
1865 #ifndef MDASSEMBLE
1866 int assemble_container_content(struct supertype *st, int mdfd,
1867 struct mdinfo *content, struct context *c,
1868 char *chosen_name, int *result)
1869 {
1870 struct mdinfo *dev, *sra, *dev2;
1871 int working = 0, preexist = 0;
1872 int expansion = 0;
1873 struct map_ent *map = NULL;
1874 int old_raid_disks;
1875 int start_reshape;
1876 char *avail = NULL;
1877 int err;
1878
1879 sysfs_init(content, mdfd, NULL);
1880
1881 sra = sysfs_read(mdfd, NULL, GET_VERSION|GET_DEVS);
1882 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0) {
1883 if (content->array.major_version == -1 &&
1884 content->array.minor_version == -2 &&
1885 c->readonly &&
1886 content->text_version[0] == '/')
1887 content->text_version[0] = '-';
1888 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1889 if (sra)
1890 sysfs_free(sra);
1891 return 1;
1892 }
1893 }
1894
1895 /* There are two types of reshape: container wide or sub-array specific
1896 * Check if metadata requests blocking container wide reshapes
1897 */
1898 start_reshape = (content->reshape_active &&
1899 !((content->reshape_active == CONTAINER_RESHAPE) &&
1900 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1901
1902 /* Block subarray here if it is under reshape now
1903 * Do not allow for any changes in this array
1904 */
1905 if (st->ss->external && content->recovery_blocked && start_reshape)
1906 block_subarray(content);
1907
1908 for (dev2 = sra->devs; dev2; dev2 = dev2->next) {
1909 for (dev = content->devs; dev; dev = dev->next)
1910 if (dev2->disk.major == dev->disk.major &&
1911 dev2->disk.minor == dev->disk.minor)
1912 break;
1913 if (dev)
1914 continue;
1915 /* Don't want this one any more */
1916 if (sysfs_set_str(sra, dev2, "slot", "none") < 0 &&
1917 errno == EBUSY) {
1918 pr_err("Cannot remove old device %s: not updating %s\n", dev2->sys_name, sra->sys_name);
1919 sysfs_free(sra);
1920 return 1;
1921 }
1922 sysfs_set_str(sra, dev2, "state", "remove");
1923 }
1924 old_raid_disks = content->array.raid_disks - content->delta_disks;
1925 avail = xcalloc(content->array.raid_disks, 1);
1926 for (dev = content->devs; dev; dev = dev->next) {
1927 if (dev->disk.raid_disk >= 0)
1928 avail[dev->disk.raid_disk] = 1;
1929 if (sysfs_add_disk(content, dev, 1) == 0) {
1930 if (dev->disk.raid_disk >= old_raid_disks &&
1931 content->reshape_active)
1932 expansion++;
1933 else
1934 working++;
1935 } else if (errno == EEXIST)
1936 preexist++;
1937 }
1938 sysfs_free(sra);
1939 if (working + expansion == 0 && c->runstop <= 0) {
1940 free(avail);
1941 return 1;/* Nothing new, don't try to start */
1942 }
1943 map_update(&map, fd2devnm(mdfd),
1944 content->text_version,
1945 content->uuid, chosen_name);
1946
1947
1948 if (enough(content->array.level, content->array.raid_disks,
1949 content->array.layout, content->array.state & 1, avail) == 0) {
1950 if (c->export && result)
1951 *result |= INCR_NO;
1952 else if (c->verbose >= 0) {
1953 pr_err("%s assembled with %d device%s",
1954 chosen_name, preexist + working,
1955 preexist + working == 1 ? "":"s");
1956 if (preexist)
1957 fprintf(stderr, " (%d new)", working);
1958 fprintf(stderr, " but not started\n");
1959 }
1960 free(avail);
1961 return 1;
1962 }
1963 free(avail);
1964
1965 if (c->runstop <= 0 &&
1966 (working + preexist + expansion) <
1967 content->array.working_disks) {
1968 if (c->export && result)
1969 *result |= INCR_UNSAFE;
1970 else if (c->verbose >= 0) {
1971 pr_err("%s assembled with %d device%s",
1972 chosen_name, preexist + working,
1973 preexist + working == 1 ? "":"s");
1974 if (preexist)
1975 fprintf(stderr, " (%d new)", working);
1976 fprintf(stderr, " but not safe to start\n");
1977 }
1978 return 1;
1979 }
1980
1981
1982 if (start_reshape) {
1983 int spare = content->array.raid_disks + expansion;
1984 if (restore_backup(st, content,
1985 working,
1986 spare, &c->backup_file, c->verbose) == 1)
1987 return 1;
1988
1989 err = sysfs_set_str(content, NULL,
1990 "array_state", "readonly");
1991 if (err)
1992 return 1;
1993
1994 if (st->ss->external) {
1995 if (!mdmon_running(st->container_devnm))
1996 start_mdmon(st->container_devnm);
1997 ping_monitor(st->container_devnm);
1998 if (mdmon_running(st->container_devnm) &&
1999 st->update_tail == NULL)
2000 st->update_tail = &st->updates;
2001 }
2002
2003 err = Grow_continue(mdfd, st, content, c->backup_file,
2004 0, c->freeze_reshape);
2005 } else switch(content->array.level) {
2006 case LEVEL_LINEAR:
2007 case LEVEL_MULTIPATH:
2008 case 0:
2009 err = sysfs_set_str(content, NULL, "array_state",
2010 c->readonly ? "readonly" : "active");
2011 break;
2012 default:
2013 err = sysfs_set_str(content, NULL, "array_state",
2014 "readonly");
2015 /* start mdmon if needed. */
2016 if (!err) {
2017 if (!mdmon_running(st->container_devnm))
2018 start_mdmon(st->container_devnm);
2019 ping_monitor(st->container_devnm);
2020 }
2021 break;
2022 }
2023 if (!err)
2024 sysfs_set_safemode(content, content->safe_mode_delay);
2025
2026 /* Block subarray here if it is not reshaped now
2027 * It has be blocked a little later to allow mdmon to switch in
2028 * in to R/W state
2029 */
2030 if (st->ss->external && content->recovery_blocked &&
2031 !start_reshape)
2032 block_subarray(content);
2033
2034 if (c->export && result) {
2035 if (err)
2036 *result |= INCR_NO;
2037 else
2038 *result |= INCR_YES;
2039 } else if (c->verbose >= 0) {
2040 if (err)
2041 pr_err("array %s now has %d device%s",
2042 chosen_name, working + preexist,
2043 working + preexist == 1 ? "":"s");
2044 else
2045 pr_err("Started %s with %d device%s",
2046 chosen_name, working + preexist,
2047 working + preexist == 1 ? "":"s");
2048 if (preexist)
2049 fprintf(stderr, " (%d new)", working);
2050 if (expansion)
2051 fprintf(stderr, " ( + %d for expansion)",
2052 expansion);
2053 fprintf(stderr, "\n");
2054 }
2055 if (!err)
2056 wait_for(chosen_name, mdfd);
2057 return err;
2058 /* FIXME should have an O_EXCL and wait for read-auto */
2059 }
2060 #endif