]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Allow domain_test to report that no domains were found.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(1, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 static int ident_matches(struct mddev_ident *ident,
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78 {
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
82 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
83 if (devname)
84 fprintf(stderr, Name ": %s has wrong uuid.\n",
85 devname);
86 return 0;
87 }
88 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
89 name_matches(content->name, ident->name, homehost)==0) {
90 if (devname)
91 fprintf(stderr, Name ": %s has wrong name.\n",
92 devname);
93 return 0;
94 }
95 if (ident->super_minor != UnSet &&
96 ident->super_minor != content->array.md_minor) {
97 if (devname)
98 fprintf(stderr, Name ": %s has wrong super-minor.\n",
99 devname);
100 return 0;
101 }
102 if (ident->level != UnSet &&
103 ident->level != content->array.level) {
104 if (devname)
105 fprintf(stderr, Name ": %s has wrong raid level.\n",
106 devname);
107 return 0;
108 }
109 if (ident->raid_disks != UnSet &&
110 ident->raid_disks!= content->array.raid_disks) {
111 if (devname)
112 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
113 devname);
114 return 0;
115 }
116 if (ident->member && ident->member[0]) {
117 /* content->text_version must match */
118 char *s = strchr(content->text_version+1, '/');
119 if (s == NULL) {
120 if (devname)
121 fprintf(stderr, Name ": %s is not a container and one is required.\n",
122 devname);
123 return 0;
124 } else if (strcmp(ident->member, s+1) != 0) {
125 if (devname)
126 fprintf(stderr, Name ": skipping wrong member %s is %s\n",
127 content->text_version, devname);
128 return 0;
129 }
130 }
131 return 1;
132 }
133
134
135 int Assemble(struct supertype *st, char *mddev,
136 struct mddev_ident *ident,
137 struct mddev_dev *devlist,
138 char *backup_file, int invalid_backup,
139 int readonly, int runstop,
140 char *update, char *homehost, int require_homehost,
141 int verbose, int force)
142 {
143 /*
144 * The task of Assemble is to find a collection of
145 * devices that should (according to their superblocks)
146 * form an array, and to give this collection to the MD driver.
147 * In Linux-2.4 and later, this involves submitting a
148 * SET_ARRAY_INFO ioctl with no arg - to prepare
149 * the array - and then submit a number of
150 * ADD_NEW_DISK ioctls to add disks into
151 * the array. Finally RUN_ARRAY might
152 * be submitted to start the array.
153 *
154 * Much of the work of Assemble is in finding and/or
155 * checking the disks to make sure they look right.
156 *
157 * If mddev is not set, then scan must be set and we
158 * read through the config file for dev+uuid mapping
159 * We recurse, setting mddev, for each device that
160 * - isn't running
161 * - has a valid uuid (or any uuid if !uuidset)
162 *
163 * If mddev is set, we try to determine state of md.
164 * check version - must be at least 0.90.0
165 * check kernel version. must be at least 2.4.
166 * If not, we can possibly fall back on START_ARRAY
167 * Try to GET_ARRAY_INFO.
168 * If possible, give up
169 * If not, try to STOP_ARRAY just to make sure
170 *
171 * If !uuidset and scan, look in conf-file for uuid
172 * If not found, give up
173 * If !devlist and scan and uuidset, get list of devs from conf-file
174 *
175 * For each device:
176 * Check superblock - discard if bad
177 * Check uuid (set if we don't have one) - discard if no match
178 * Check superblock similarity if we have a superblock - discard if different
179 * Record events, devicenum
180 * This should give us a list of devices for the array
181 * We should collect the most recent event number
182 *
183 * Count disks with recent enough event count
184 * While force && !enough disks
185 * Choose newest rejected disks, update event count
186 * mark clean and rewrite superblock
187 * If recent kernel:
188 * SET_ARRAY_INFO
189 * foreach device with recent events : ADD_NEW_DISK
190 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
191 * If old kernel:
192 * Check the device numbers in superblock are right
193 * update superblock if any changes
194 * START_ARRAY
195 *
196 */
197 int mdfd;
198 int clean;
199 int auto_assem = (mddev == NULL && !ident->uuid_set &&
200 ident->super_minor == UnSet && ident->name[0] == 0
201 && (ident->container == NULL || ident->member == NULL));
202 int old_linux = 0;
203 int vers = vers; /* Keep gcc quite - it really is initialised */
204 struct {
205 char *devname;
206 int uptodate; /* set once we decide that this device is as
207 * recent as everything else in the array.
208 */
209 struct mdinfo i;
210 } *devices;
211 char *devmap;
212 int *best = NULL; /* indexed by raid_disk */
213 int bestcnt = 0;
214 int devcnt = 0;
215 unsigned int okcnt, sparecnt, rebuilding_cnt;
216 unsigned int req_cnt;
217 int i;
218 int most_recent = 0;
219 int chosen_drive;
220 int change = 0;
221 int inargv = 0;
222 int report_missmatch;
223 int bitmap_done;
224 int start_partial_ok = (runstop >= 0) &&
225 (force || devlist==NULL || auto_assem);
226 unsigned int num_devs;
227 struct mddev_dev *tmpdev;
228 struct mdinfo info;
229 struct mdinfo *content = NULL;
230 char *avail;
231 int nextspare = 0;
232 char *name = NULL;
233 int trustworthy;
234 char chosen_name[1024];
235 struct domainlist *domains = NULL;
236
237 if (get_linux_version() < 2004000)
238 old_linux = 1;
239
240 /*
241 * If any subdevs are listed, then any that don't
242 * match ident are discarded. Remainder must all match and
243 * become the array.
244 * If no subdevs, then we scan all devices in the config file, but
245 * there must be something in the identity
246 */
247
248 if (!devlist &&
249 ident->uuid_set == 0 &&
250 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
251 ident->name[0] == 0 &&
252 (ident->container == NULL || ident->member == NULL) &&
253 ident->devices == NULL) {
254 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
255 mddev ? mddev : "further assembly");
256 return 1;
257 }
258
259 if (devlist == NULL)
260 devlist = conf_get_devs();
261 else if (mddev)
262 inargv = 1;
263
264 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
265 try_again:
266 /* We come back here when doing auto-assembly and attempting some
267 * set of devices failed. Those are now marked as ->used==2 and
268 * we ignore them and try again
269 */
270
271 tmpdev = devlist; num_devs = 0;
272 while (tmpdev) {
273 if (tmpdev->used)
274 tmpdev->used = 2;
275 else
276 num_devs++;
277 tmpdev = tmpdev->next;
278 }
279
280 if (!st && ident->st) st = ident->st;
281
282 if (verbose>0)
283 fprintf(stderr, Name ": looking for devices for %s\n",
284 mddev ? mddev : "further assembly");
285
286 /* first walk the list of devices to find a consistent set
287 * that match the criterea, if that is possible.
288 * We flag the ones we like with 'used'.
289 */
290 for (tmpdev = devlist;
291 tmpdev;
292 tmpdev = tmpdev ? tmpdev->next : NULL) {
293 char *devname = tmpdev->devname;
294 int dfd;
295 struct stat stb;
296 struct supertype *tst = dup_super(st);
297 struct dev_policy *pol = NULL;
298 int found_container = 0;
299
300 if (tmpdev->used > 1) continue;
301
302 if (ident->devices &&
303 !match_oneof(ident->devices, devname)) {
304 if (report_missmatch)
305 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
306 continue;
307 }
308
309 dfd = dev_open(devname, O_RDONLY|O_EXCL);
310 if (dfd < 0) {
311 if (report_missmatch)
312 fprintf(stderr, Name ": cannot open device %s: %s\n",
313 devname, strerror(errno));
314 tmpdev->used = 2;
315 } else if (fstat(dfd, &stb)< 0) {
316 /* Impossible! */
317 fprintf(stderr, Name ": fstat failed for %s: %s\n",
318 devname, strerror(errno));
319 tmpdev->used = 2;
320 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
321 fprintf(stderr, Name ": %s is not a block device.\n",
322 devname);
323 tmpdev->used = 2;
324 } else if (must_be_container(dfd)) {
325 if (st) {
326 /* already found some components, this cannot
327 * be another one.
328 */
329 if (report_missmatch)
330 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
331 devname);
332 tmpdev->used = 2;
333 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
334 if (report_missmatch)
335 fprintf(stderr, Name ": not a recognisable container: %s\n",
336 devname);
337 tmpdev->used = 2;
338 } else if (!tst->ss->load_container
339 || tst->ss->load_container(tst, dfd, NULL)) {
340 if (report_missmatch)
341 fprintf(stderr, Name ": no correct container type: %s\n",
342 devname);
343 tmpdev->used = 2;
344 } else if (auto_assem &&
345 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
346 tst->ss->match_home(tst, homehost) == 1)) {
347 if (report_missmatch)
348 fprintf(stderr, Name ": %s has metadata type %s for which "
349 "auto-assembly is disabled\n",
350 devname, tst->ss->name);
351 tmpdev->used = 2;
352 } else
353 found_container = 1;
354 } else {
355 if (!tst && (tst = guess_super(dfd)) == NULL) {
356 if (report_missmatch)
357 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
358 devname);
359 tmpdev->used = 2;
360 } else if (tst->ss->load_super(tst,dfd, NULL)) {
361 if (report_missmatch)
362 fprintf(stderr, Name ": no RAID superblock on %s\n",
363 devname);
364 tmpdev->used = 2;
365 } else if (tst->ss->compare_super == NULL) {
366 if (report_missmatch)
367 fprintf(stderr, Name ": Cannot assemble %s metadata on %s\n",
368 tst->ss->name, devname);
369 tmpdev->used = 2;
370 } else if (auto_assem && st == NULL &&
371 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
372 tst->ss->match_home(tst, homehost) == 1)) {
373 if (report_missmatch)
374 fprintf(stderr, Name ": %s has metadata type %s for which "
375 "auto-assembly is disabled\n",
376 devname, tst->ss->name);
377 tmpdev->used = 2;
378 }
379 }
380 if (dfd >= 0) close(dfd);
381 if (tmpdev->used == 2) {
382 if (auto_assem || !inargv)
383 /* Ignore unrecognised devices during auto-assembly */
384 goto loop;
385 if (ident->uuid_set || ident->name[0] ||
386 ident->super_minor != UnSet)
387 /* Ignore unrecognised device if looking for
388 * specific array */
389 goto loop;
390
391
392 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
393 devname);
394 if (st)
395 st->ss->free_super(st);
396 dev_policy_free(pol);
397 domain_free(domains);
398 return 1;
399 }
400
401 if (found_container) {
402 /* tmpdev is a container. We need to be either
403 * looking for a member, or auto-assembling
404 */
405
406 if (ident->container) {
407 if (ident->container[0] == '/' &&
408 !same_dev(ident->container, devname)) {
409 if (report_missmatch)
410 fprintf(stderr, Name ": %s is not the container required (%s)\n",
411 devname, ident->container);
412 goto loop;
413 }
414 if (ident->container[0] != '/') {
415 /* we have a uuid */
416 int uuid[4];
417
418 content = &info;
419 memset(content, 0, sizeof(*content));
420 tst->ss->getinfo_super(tst, content, NULL);
421
422 if (!parse_uuid(ident->container, uuid) ||
423 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
424 if (report_missmatch)
425 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
426 devname);
427 goto loop;
428 }
429 }
430 }
431 /* It is worth looking inside this container.
432 */
433 if (verbose > 0)
434 fprintf(stderr, Name ": looking in container %s\n",
435 devname);
436
437 for (content = tst->ss->container_content(tst, NULL);
438 content;
439 content = content->next) {
440
441 /* do not assemble arrays that might have bad blocks */
442 if (content->array.state & (1<<MD_SB_BBM_ERRORS)) {
443 fprintf(stderr, Name ": BBM log found in metadata. "
444 "Cannot activate array(s).\n");
445 tmpdev->used = 2;
446 goto loop;
447 }
448 if (!ident_matches(ident, content, tst,
449 homehost, update,
450 report_missmatch ? devname : NULL))
451 /* message already printed */;
452 else if (is_member_busy(content->text_version)) {
453 if (report_missmatch)
454 fprintf(stderr, Name ": member %s in %s is already assembled\n",
455 content->text_version,
456 devname);
457 } else
458 break;
459 }
460 if (!content) {
461 tmpdev->used = 2;
462 goto loop; /* empty container */
463 }
464
465 st = tst; tst = NULL;
466 if (!auto_assem && inargv && tmpdev->next != NULL) {
467 fprintf(stderr, Name ": %s is a container, but is not "
468 "only device given: confused and aborting\n",
469 devname);
470 st->ss->free_super(st);
471 dev_policy_free(pol);
472 domain_free(domains);
473 return 1;
474 }
475 if (verbose > 0)
476 fprintf(stderr, Name ": found match on member %s in %s\n",
477 content->text_version, devname);
478
479 /* make sure we finished the loop */
480 tmpdev = NULL;
481 goto loop;
482 } else {
483
484 content = &info;
485 memset(content, 0, sizeof(*content));
486 tst->ss->getinfo_super(tst, content, NULL);
487
488 if (!ident_matches(ident, content, tst,
489 homehost, update,
490 report_missmatch ? devname : NULL))
491 goto loop;
492
493 if (st == NULL)
494 st = dup_super(tst);
495 if (st->minor_version == -1)
496 st->minor_version = tst->minor_version;
497
498 if (memcmp(content->uuid, uuid_zero,
499 sizeof(int[4])) == 0) {
500 /* this is a floating spare. It cannot define
501 * an array unless there are no more arrays of
502 * this type to be found. It can be included
503 * in an array of this type though.
504 */
505 tmpdev->used = 3;
506 goto loop;
507 }
508
509 if (st->ss != tst->ss ||
510 st->minor_version != tst->minor_version ||
511 st->ss->compare_super(st, tst) != 0) {
512 /* Some mismatch. If exactly one array matches this host,
513 * we can resolve on that one.
514 * Or, if we are auto assembling, we just ignore the second
515 * for now.
516 */
517 if (auto_assem)
518 goto loop;
519 if (homehost) {
520 int first = st->ss->match_home(st, homehost);
521 int last = tst->ss->match_home(tst, homehost);
522 if (first != last &&
523 (first == 1 || last == 1)) {
524 /* We can do something */
525 if (first) {/* just ignore this one */
526 if (report_missmatch)
527 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
528 devname);
529 goto loop;
530 } else { /* reject all those sofar */
531 struct mddev_dev *td;
532 if (report_missmatch)
533 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
534 devname);
535 for (td=devlist; td != tmpdev; td=td->next)
536 if (td->used == 1)
537 td->used = 0;
538 tmpdev->used = 1;
539 goto loop;
540 }
541 }
542 }
543 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
544 devname);
545 tst->ss->free_super(tst);
546 st->ss->free_super(st);
547 dev_policy_free(pol);
548 domain_free(domains);
549 return 1;
550 }
551 tmpdev->used = 1;
552 }
553 loop:
554 /* Collect domain information from members only */
555 if (tmpdev && tmpdev->used == 1) {
556 if (!pol)
557 pol = devnum_policy(stb.st_rdev);
558 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
559 }
560 dev_policy_free(pol);
561 pol = NULL;
562 if (tst)
563 tst->ss->free_super(tst);
564 }
565
566 /* Check if we found some imsm spares but no members */
567 if ((auto_assem ||
568 (ident->uuid_set &&
569 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
570 (!st || !st->sb))
571 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
572 if (tmpdev->used != 3)
573 continue;
574 tmpdev->used = 1;
575 content = &info;
576
577 if (!st->sb) {
578 /* we need sb from one of the spares */
579 int dfd = dev_open(tmpdev->devname, O_RDONLY);
580 if (dfd < 0 ||
581 st->ss->load_super(st, dfd, NULL))
582 tmpdev->used = 2;
583 if (dfd > 0)
584 close(dfd);
585 }
586 }
587
588 /* Now reject spares that don't match domains of identified members */
589 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
590 struct stat stb;
591 if (tmpdev->used != 3)
592 continue;
593 if (stat(tmpdev->devname, &stb)< 0) {
594 fprintf(stderr, Name ": fstat failed for %s: %s\n",
595 tmpdev->devname, strerror(errno));
596 tmpdev->used = 2;
597 } else {
598 struct dev_policy *pol = NULL;
599 pol = devnum_policy(stb.st_rdev);
600 if (domain_test(domains, pol, NULL) == 1)
601 /* take this spare if domains match */
602 tmpdev->used = 1;
603 else
604 /* if domains don't match mark as unused */
605 tmpdev->used = 0;
606 dev_policy_free(pol);
607 }
608 }
609 domain_free(domains);
610
611 if (!st || !st->sb || !content)
612 return 2;
613
614 /* Now need to open the array device. Use create_mddev */
615 if (content == &info)
616 st->ss->getinfo_super(st, content, NULL);
617
618 trustworthy = FOREIGN;
619 name = content->name;
620 switch (st->ss->match_home(st, homehost)
621 ?: st->ss->match_home(st, "any")) {
622 case 1:
623 trustworthy = LOCAL;
624 name = strchr(content->name, ':');
625 if (name)
626 name++;
627 else
628 name = content->name;
629 break;
630 }
631 if (!auto_assem)
632 /* If the array is listed in mdadm.conf or on
633 * command line, then we trust the name
634 * even if the array doesn't look local
635 */
636 trustworthy = LOCAL;
637
638 if (name[0] == 0 &&
639 content->array.level == LEVEL_CONTAINER) {
640 name = content->text_version;
641 trustworthy = METADATA;
642 }
643
644 if (name[0] && trustworthy != LOCAL &&
645 ! require_homehost &&
646 conf_name_is_free(name))
647 trustworthy = LOCAL;
648
649 if (trustworthy == LOCAL &&
650 strchr(name, ':'))
651 /* Ignore 'host:' prefix of name */
652 name = strchr(name, ':')+1;
653
654 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
655 chosen_name);
656 if (mdfd < 0) {
657 st->ss->free_super(st);
658 if (auto_assem)
659 goto try_again;
660 return 1;
661 }
662 mddev = chosen_name;
663 vers = md_get_version(mdfd);
664 if (vers < 9000) {
665 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
666 " Upgrade your kernel or try --build\n");
667 close(mdfd);
668 return 1;
669 }
670 if (mddev_busy(fd2devnum(mdfd))) {
671 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
672 mddev);
673 for (tmpdev = devlist ;
674 tmpdev && tmpdev->used != 1;
675 tmpdev = tmpdev->next)
676 ;
677 if (tmpdev && auto_assem)
678 fprintf(stderr, Name ": %s needed for %s...\n",
679 mddev, tmpdev->devname);
680 close(mdfd);
681 mdfd = -3;
682 st->ss->free_super(st);
683 if (auto_assem)
684 goto try_again;
685 return 1;
686 }
687 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
688
689 #ifndef MDASSEMBLE
690 if (content != &info) {
691 /* This is a member of a container. Try starting the array. */
692 return assemble_container_content(st, mdfd, content, runstop,
693 chosen_name, verbose);
694 }
695 #endif
696 /* Ok, no bad inconsistancy, we can try updating etc */
697 bitmap_done = 0;
698 content->update_private = NULL;
699 devices = malloc(num_devs * sizeof(*devices));
700 devmap = calloc(num_devs * content->array.raid_disks, 1);
701 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
702 char *devname = tmpdev->devname;
703 struct stat stb;
704 /* looks like a good enough match to update the super block if needed */
705 #ifndef MDASSEMBLE
706 if (update) {
707 int dfd;
708 /* prepare useful information in info structures */
709 struct stat stb2;
710 struct supertype *tst;
711 int err;
712 fstat(mdfd, &stb2);
713
714 if (strcmp(update, "uuid")==0 &&
715 !ident->uuid_set) {
716 int rfd;
717 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
718 read(rfd, ident->uuid, 16) != 16) {
719 *(__u32*)(ident->uuid) = random();
720 *(__u32*)(ident->uuid+1) = random();
721 *(__u32*)(ident->uuid+2) = random();
722 *(__u32*)(ident->uuid+3) = random();
723 }
724 if (rfd >= 0) close(rfd);
725 }
726 dfd = dev_open(devname, O_RDWR|O_EXCL);
727
728 tst = dup_super(st);
729 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
730 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
731 devname);
732 if (dfd >= 0)
733 close(dfd);
734 close(mdfd);
735 free(devices);
736 free(devmap);
737 return 1;
738 }
739 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
740
741 memcpy(content->uuid, ident->uuid, 16);
742 strcpy(content->name, ident->name);
743 content->array.md_minor = minor(stb2.st_rdev);
744
745 if (strcmp(update, "byteorder") == 0)
746 err = 0;
747 else
748 err = tst->ss->update_super(tst, content, update,
749 devname, verbose,
750 ident->uuid_set,
751 homehost);
752 if (err < 0) {
753 fprintf(stderr,
754 Name ": --update=%s not understood"
755 " for %s metadata\n",
756 update, tst->ss->name);
757 tst->ss->free_super(tst);
758 free(tst);
759 close(mdfd);
760 close(dfd);
761 free(devices);
762 free(devmap);
763 return 1;
764 }
765 if (strcmp(update, "uuid")==0 &&
766 !ident->uuid_set) {
767 ident->uuid_set = 1;
768 memcpy(ident->uuid, content->uuid, 16);
769 }
770 if (tst->ss->store_super(tst, dfd))
771 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
772 devname);
773 close(dfd);
774
775 if (strcmp(update, "uuid")==0 &&
776 ident->bitmap_fd >= 0 && !bitmap_done) {
777 if (bitmap_update_uuid(ident->bitmap_fd,
778 content->uuid,
779 tst->ss->swapuuid) != 0)
780 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
781 else
782 bitmap_done = 1;
783 }
784 tst->ss->free_super(tst);
785 } else
786 #endif
787 {
788 struct supertype *tst = dup_super(st);
789 int dfd;
790 dfd = dev_open(devname, O_RDWR|O_EXCL);
791
792 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
793 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
794 devname);
795 if (dfd >= 0)
796 close(dfd);
797 close(mdfd);
798 free(devices);
799 free(devmap);
800 return 1;
801 }
802 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
803 tst->ss->free_super(tst);
804 close(dfd);
805 }
806
807 stat(devname, &stb);
808
809 if (verbose > 0)
810 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
811 devname, mddev, content->disk.raid_disk);
812 devices[devcnt].devname = devname;
813 devices[devcnt].uptodate = 0;
814 devices[devcnt].i = *content;
815 devices[devcnt].i.disk.major = major(stb.st_rdev);
816 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
817 if (most_recent < devcnt) {
818 if (devices[devcnt].i.events
819 > devices[most_recent].i.events)
820 most_recent = devcnt;
821 }
822 if (content->array.level == LEVEL_MULTIPATH)
823 /* with multipath, the raid_disk from the superblock is meaningless */
824 i = devcnt;
825 else
826 i = devices[devcnt].i.disk.raid_disk;
827 if (i+1 == 0) {
828 if (nextspare < content->array.raid_disks)
829 nextspare = content->array.raid_disks;
830 i = nextspare++;
831 } else {
832 if (i >= content->array.raid_disks &&
833 i >= nextspare)
834 nextspare = i+1;
835 }
836 if (i < 10000) {
837 if (i >= bestcnt) {
838 int newbestcnt = i+10;
839 int *newbest = malloc(sizeof(int)*newbestcnt);
840 int c;
841 for (c=0; c < newbestcnt; c++)
842 if (c < bestcnt)
843 newbest[c] = best[c];
844 else
845 newbest[c] = -1;
846 if (best)free(best);
847 best = newbest;
848 bestcnt = newbestcnt;
849 }
850 if (best[i] >=0 &&
851 devices[best[i]].i.events
852 == devices[devcnt].i.events
853 && (devices[best[i]].i.disk.minor
854 != devices[devcnt].i.disk.minor)
855 && st->ss == &super0
856 && content->array.level != LEVEL_MULTIPATH) {
857 /* two different devices with identical superblock.
858 * Could be a mis-detection caused by overlapping
859 * partitions. fail-safe.
860 */
861 fprintf(stderr, Name ": WARNING %s and %s appear"
862 " to have very similar superblocks.\n"
863 " If they are really different, "
864 "please --zero the superblock on one\n"
865 " If they are the same or overlap,"
866 " please remove one from %s.\n",
867 devices[best[i]].devname, devname,
868 inargv ? "the list" :
869 "the\n DEVICE list in mdadm.conf"
870 );
871 close(mdfd);
872 free(devices);
873 free(devmap);
874 return 1;
875 }
876 if (best[i] == -1
877 || (devices[best[i]].i.events
878 < devices[devcnt].i.events))
879 best[i] = devcnt;
880 }
881 devcnt++;
882 }
883 free(content->update_private);
884 content->update_private = NULL;
885
886 if (devcnt == 0) {
887 fprintf(stderr, Name ": no devices found for %s\n",
888 mddev);
889 if (st)
890 st->ss->free_super(st);
891 close(mdfd);
892 free(devices);
893 free(devmap);
894 return 1;
895 }
896
897 if (update && strcmp(update, "byteorder")==0)
898 st->minor_version = 90;
899
900 st->ss->getinfo_super(st, content, NULL);
901 clean = content->array.state & 1;
902
903 /* now we have some devices that might be suitable.
904 * I wonder how many
905 */
906 avail = malloc(content->array.raid_disks);
907 memset(avail, 0, content->array.raid_disks);
908 okcnt = 0;
909 sparecnt=0;
910 rebuilding_cnt=0;
911 for (i=0; i< bestcnt; i++) {
912 int j = best[i];
913 int event_margin = 1; /* always allow a difference of '1'
914 * like the kernel does
915 */
916 if (j < 0) continue;
917 /* note: we ignore error flags in multipath arrays
918 * as they don't make sense
919 */
920 if (content->array.level != LEVEL_MULTIPATH)
921 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
922 if (!(devices[j].i.disk.state
923 & (1<<MD_DISK_FAULTY))) {
924 devices[j].uptodate = 1;
925 sparecnt++;
926 }
927 continue;
928 }
929 /* If this devices thinks that 'most_recent' has failed, then
930 * we must reject this device.
931 */
932 if (j != most_recent &&
933 content->array.raid_disks > 0 &&
934 devices[most_recent].i.disk.raid_disk >= 0 &&
935 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
936 if (verbose > -1)
937 fprintf(stderr, Name ": ignoring %s as it reports %s as failed\n",
938 devices[j].devname, devices[most_recent].devname);
939 best[i] = -1;
940 continue;
941 }
942 if (devices[j].i.events+event_margin >=
943 devices[most_recent].i.events) {
944 devices[j].uptodate = 1;
945 if (i < content->array.raid_disks) {
946 if (devices[j].i.recovery_start == MaxSector ||
947 (content->reshape_active &&
948 j >= content->array.raid_disks - content->delta_disks)) {
949 okcnt++;
950 avail[i]=1;
951 } else
952 rebuilding_cnt++;
953 } else
954 sparecnt++;
955 }
956 }
957 free(devmap);
958 while (force && !enough(content->array.level, content->array.raid_disks,
959 content->array.layout, 1,
960 avail, okcnt)) {
961 /* Choose the newest best drive which is
962 * not up-to-date, update the superblock
963 * and add it.
964 */
965 int fd;
966 struct supertype *tst;
967 unsigned long long current_events;
968 chosen_drive = -1;
969 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
970 int j = best[i];
971 if (j>=0 &&
972 !devices[j].uptodate &&
973 devices[j].i.recovery_start == MaxSector &&
974 (chosen_drive < 0 ||
975 devices[j].i.events
976 > devices[chosen_drive].i.events))
977 chosen_drive = j;
978 }
979 if (chosen_drive < 0)
980 break;
981 current_events = devices[chosen_drive].i.events;
982 add_another:
983 if (verbose >= 0)
984 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
985 devices[chosen_drive].devname,
986 devices[chosen_drive].i.disk.raid_disk,
987 (int)(devices[chosen_drive].i.events),
988 (int)(devices[most_recent].i.events));
989 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
990 if (fd < 0) {
991 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
992 devices[chosen_drive].devname);
993 devices[chosen_drive].i.events = 0;
994 continue;
995 }
996 tst = dup_super(st);
997 if (tst->ss->load_super(tst,fd, NULL)) {
998 close(fd);
999 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
1000 devices[chosen_drive].devname);
1001 devices[chosen_drive].i.events = 0;
1002 continue;
1003 }
1004 content->events = devices[most_recent].i.events;
1005 tst->ss->update_super(tst, content, "force-one",
1006 devices[chosen_drive].devname, verbose,
1007 0, NULL);
1008
1009 if (tst->ss->store_super(tst, fd)) {
1010 close(fd);
1011 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1012 devices[chosen_drive].devname);
1013 devices[chosen_drive].i.events = 0;
1014 tst->ss->free_super(tst);
1015 continue;
1016 }
1017 close(fd);
1018 devices[chosen_drive].i.events = devices[most_recent].i.events;
1019 devices[chosen_drive].uptodate = 1;
1020 avail[chosen_drive] = 1;
1021 okcnt++;
1022 tst->ss->free_super(tst);
1023
1024 /* If there are any other drives of the same vintage,
1025 * add them in as well. We can't lose and we might gain
1026 */
1027 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
1028 int j = best[i];
1029 if (j >= 0 &&
1030 !devices[j].uptodate &&
1031 devices[j].i.events == current_events) {
1032 chosen_drive = j;
1033 goto add_another;
1034 }
1035 }
1036 }
1037
1038 /* Now we want to look at the superblock which the kernel will base things on
1039 * and compare the devices that we think are working with the devices that the
1040 * superblock thinks are working.
1041 * If there are differences and --force is given, then update this chosen
1042 * superblock.
1043 */
1044 chosen_drive = -1;
1045 st->ss->free_super(st);
1046 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
1047 int j = best[i];
1048 int fd;
1049
1050 if (j<0)
1051 continue;
1052 if (!devices[j].uptodate)
1053 continue;
1054 chosen_drive = j;
1055 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
1056 fprintf(stderr, Name ": Cannot open %s: %s\n",
1057 devices[j].devname, strerror(errno));
1058 close(mdfd);
1059 free(devices);
1060 return 1;
1061 }
1062 if (st->ss->load_super(st,fd, NULL)) {
1063 close(fd);
1064 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
1065 devices[j].devname);
1066 close(mdfd);
1067 free(devices);
1068 return 1;
1069 }
1070 close(fd);
1071 }
1072 if (st->sb == NULL) {
1073 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
1074 close(mdfd);
1075 free(devices);
1076 return 1;
1077 }
1078 st->ss->getinfo_super(st, content, NULL);
1079 #ifndef MDASSEMBLE
1080 sysfs_init(content, mdfd, 0);
1081 #endif
1082 for (i=0; i<bestcnt; i++) {
1083 int j = best[i];
1084 unsigned int desired_state;
1085
1086 if (i < content->array.raid_disks)
1087 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1088 else
1089 desired_state = 0;
1090
1091 if (j<0)
1092 continue;
1093 if (!devices[j].uptodate)
1094 continue;
1095
1096 devices[j].i.disk.state = desired_state;
1097 if (!(devices[j].i.array.state & 1))
1098 clean = 0;
1099
1100 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1101 verbose, 0, NULL)) {
1102 if (force) {
1103 if (verbose >= 0)
1104 fprintf(stderr, Name ": "
1105 "clearing FAULTY flag for device %d in %s for %s\n",
1106 j, mddev, devices[j].devname);
1107 change = 1;
1108 } else {
1109 if (verbose >= -1)
1110 fprintf(stderr, Name ": "
1111 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1112 i, mddev, devices[j].devname);
1113 }
1114 }
1115 #if 0
1116 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1117 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1118 i, mddev);
1119 }
1120 #endif
1121 }
1122 if (force && !clean &&
1123 !enough(content->array.level, content->array.raid_disks,
1124 content->array.layout, clean,
1125 avail, okcnt)) {
1126 change += st->ss->update_super(st, content, "force-array",
1127 devices[chosen_drive].devname, verbose,
1128 0, NULL);
1129 clean = 1;
1130 }
1131
1132 if (change) {
1133 int fd;
1134 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
1135 if (fd < 0) {
1136 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1137 devices[chosen_drive].devname);
1138 close(mdfd);
1139 free(devices);
1140 return 1;
1141 }
1142 if (st->ss->store_super(st, fd)) {
1143 close(fd);
1144 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1145 devices[chosen_drive].devname);
1146 close(mdfd);
1147 free(devices);
1148 return 1;
1149 }
1150 close(fd);
1151 }
1152
1153 /* If we are in the middle of a reshape we may need to restore saved data
1154 * that was moved aside due to the reshape overwriting live data
1155 * The code of doing this lives in Grow.c
1156 */
1157 #ifndef MDASSEMBLE
1158 if (content->reshape_active) {
1159 int err = 0;
1160 int *fdlist = malloc(sizeof(int)* bestcnt);
1161 if (verbose > 0)
1162 fprintf(stderr, Name ":%s has an active reshape - checking "
1163 "if critical section needs to be restored\n",
1164 chosen_name);
1165 for (i=0; i<bestcnt; i++) {
1166 int j = best[i];
1167 if (j >= 0) {
1168 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1169 if (fdlist[i] < 0) {
1170 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1171 devices[j].devname);
1172 err = 1;
1173 break;
1174 }
1175 } else
1176 fdlist[i] = -1;
1177 }
1178 if (!err) {
1179 err = Grow_restart(st, content, fdlist, bestcnt,
1180 backup_file, verbose > 0);
1181 if (err && invalid_backup) {
1182 if (verbose > 0)
1183 fprintf(stderr, Name ": continuing"
1184 " without restoring backup\n");
1185 err = 0;
1186 }
1187 }
1188 while (i>0) {
1189 i--;
1190 if (fdlist[i]>=0) close(fdlist[i]);
1191 }
1192 if (err) {
1193 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1194 if (backup_file == NULL)
1195 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1196 close(mdfd);
1197 free(devices);
1198 return err;
1199 }
1200 }
1201 #endif
1202 /* count number of in-sync devices according to the superblock.
1203 * We must have this number to start the array without -s or -R
1204 */
1205 req_cnt = content->array.working_disks;
1206
1207 /* Almost ready to actually *do* something */
1208 if (!old_linux) {
1209 int rv;
1210
1211 /* First, fill in the map, so that udev can find our name
1212 * as soon as we become active.
1213 */
1214 map_update(NULL, fd2devnum(mdfd), content->text_version,
1215 content->uuid, chosen_name);
1216
1217 rv = set_array_info(mdfd, st, content);
1218 if (rv) {
1219 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1220 mddev, strerror(errno));
1221 ioctl(mdfd, STOP_ARRAY, NULL);
1222 close(mdfd);
1223 free(devices);
1224 return 1;
1225 }
1226 if (ident->bitmap_fd >= 0) {
1227 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1228 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1229 ioctl(mdfd, STOP_ARRAY, NULL);
1230 close(mdfd);
1231 free(devices);
1232 return 1;
1233 }
1234 } else if (ident->bitmap_file) {
1235 /* From config file */
1236 int bmfd = open(ident->bitmap_file, O_RDWR);
1237 if (bmfd < 0) {
1238 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1239 ident->bitmap_file);
1240 ioctl(mdfd, STOP_ARRAY, NULL);
1241 close(mdfd);
1242 free(devices);
1243 return 1;
1244 }
1245 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1246 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1247 close(bmfd);
1248 ioctl(mdfd, STOP_ARRAY, NULL);
1249 close(mdfd);
1250 free(devices);
1251 return 1;
1252 }
1253 close(bmfd);
1254 }
1255
1256 /* First, add the raid disks, but add the chosen one last */
1257 for (i=0; i<= bestcnt; i++) {
1258 int j;
1259 if (i < bestcnt) {
1260 j = best[i];
1261 if (j == chosen_drive)
1262 continue;
1263 } else
1264 j = chosen_drive;
1265
1266 if (j >= 0 /* && devices[j].uptodate */) {
1267 int dfd = dev_open(devices[j].devname,
1268 O_RDWR|O_EXCL);
1269 if (dfd >= 0) {
1270 remove_partitions(dfd);
1271 close(dfd);
1272 }
1273 rv = add_disk(mdfd, st, content, &devices[j].i);
1274
1275 if (rv) {
1276 fprintf(stderr, Name ": failed to add "
1277 "%s to %s: %s\n",
1278 devices[j].devname,
1279 mddev,
1280 strerror(errno));
1281 if (i < content->array.raid_disks
1282 || i == bestcnt)
1283 okcnt--;
1284 else
1285 sparecnt--;
1286 } else if (verbose > 0)
1287 fprintf(stderr, Name ": added %s "
1288 "to %s as %d\n",
1289 devices[j].devname, mddev,
1290 devices[j].i.disk.raid_disk);
1291 } else if (verbose > 0 && i < content->array.raid_disks)
1292 fprintf(stderr, Name ": no uptodate device for "
1293 "slot %d of %s\n",
1294 i, mddev);
1295 }
1296
1297 if (content->array.level == LEVEL_CONTAINER) {
1298 if (verbose >= 0) {
1299 fprintf(stderr, Name ": Container %s has been "
1300 "assembled with %d drive%s",
1301 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1302 if (okcnt < (unsigned)content->array.raid_disks)
1303 fprintf(stderr, " (out of %d)",
1304 content->array.raid_disks);
1305 fprintf(stderr, "\n");
1306 }
1307 st->ss->free_super(st);
1308 sysfs_uevent(content, "change");
1309 wait_for(chosen_name, mdfd);
1310 close(mdfd);
1311 free(devices);
1312 return 0;
1313 }
1314
1315 if (runstop == 1 ||
1316 (runstop <= 0 &&
1317 ( enough(content->array.level, content->array.raid_disks,
1318 content->array.layout, clean, avail, okcnt) &&
1319 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1320 ))) {
1321 /* This array is good-to-go.
1322 * If a reshape is in progress then we might need to
1323 * continue monitoring it. In that case we start
1324 * it read-only and let the grow code make it writable.
1325 */
1326 int rv;
1327 #ifndef MDASSEMBLE
1328 if (content->reshape_active &&
1329 content->delta_disks <= 0)
1330 rv = Grow_continue(mdfd, st, content, backup_file);
1331 else
1332 #endif
1333 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1334 if (rv == 0) {
1335 if (verbose >= 0) {
1336 fprintf(stderr, Name ": %s has been started with %d drive%s",
1337 mddev, okcnt, okcnt==1?"":"s");
1338 if (okcnt < (unsigned)content->array.raid_disks)
1339 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1340 if (rebuilding_cnt)
1341 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1342 if (sparecnt)
1343 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1344 fprintf(stderr, ".\n");
1345 }
1346 if (content->reshape_active &&
1347 content->array.level >= 4 &&
1348 content->array.level <= 6) {
1349 /* might need to increase the size
1350 * of the stripe cache - default is 256
1351 */
1352 if (256 < 4 * (content->array.chunk_size/4096)) {
1353 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1354 if (sra)
1355 sysfs_set_num(sra, NULL,
1356 "stripe_cache_size",
1357 (4 * content->array.chunk_size / 4096) + 1);
1358 }
1359 }
1360 if (okcnt < (unsigned)content->array.raid_disks) {
1361 /* If any devices did not get added
1362 * because the kernel rejected them based
1363 * on event count, try adding them
1364 * again providing the action policy is
1365 * 're-add' or greater. The bitmap
1366 * might allow them to be included, or
1367 * they will become spares.
1368 */
1369 for (i = 0; i <= bestcnt; i++) {
1370 int j = best[i];
1371 if (j >= 0 && !devices[j].uptodate) {
1372 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1373 continue;
1374 rv = add_disk(mdfd, st, content,
1375 &devices[j].i);
1376 if (rv == 0 && verbose >= 0)
1377 fprintf(stderr,
1378 Name ": %s has been re-added.\n",
1379 devices[j].devname);
1380 }
1381 }
1382 }
1383 wait_for(mddev, mdfd);
1384 close(mdfd);
1385 if (auto_assem) {
1386 int usecs = 1;
1387 /* There is a nasty race with 'mdadm --monitor'.
1388 * If it opens this device before we close it,
1389 * it gets an incomplete open on which IO
1390 * doesn't work and the capacity is
1391 * wrong.
1392 * If we reopen (to check for layered devices)
1393 * before --monitor closes, we loose.
1394 *
1395 * So: wait upto 1 second for there to be
1396 * a non-zero capacity.
1397 */
1398 while (usecs < 1000) {
1399 mdfd = open(mddev, O_RDONLY);
1400 if (mdfd >= 0) {
1401 unsigned long long size;
1402 if (get_dev_size(mdfd, NULL, &size) &&
1403 size > 0)
1404 break;
1405 close(mdfd);
1406 }
1407 usleep(usecs);
1408 usecs <<= 1;
1409 }
1410 }
1411 free(devices);
1412 return 0;
1413 }
1414 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1415 mddev, strerror(errno));
1416
1417 if (!enough(content->array.level, content->array.raid_disks,
1418 content->array.layout, 1, avail, okcnt))
1419 fprintf(stderr, Name ": Not enough devices to "
1420 "start the array.\n");
1421 else if (!enough(content->array.level,
1422 content->array.raid_disks,
1423 content->array.layout, clean,
1424 avail, okcnt))
1425 fprintf(stderr, Name ": Not enough devices to "
1426 "start the array while not clean "
1427 "- consider --force.\n");
1428
1429 if (auto_assem)
1430 ioctl(mdfd, STOP_ARRAY, NULL);
1431 close(mdfd);
1432 free(devices);
1433 return 1;
1434 }
1435 if (runstop == -1) {
1436 fprintf(stderr, Name ": %s assembled from %d drive%s",
1437 mddev, okcnt, okcnt==1?"":"s");
1438 if (okcnt != (unsigned)content->array.raid_disks)
1439 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1440 fprintf(stderr, ", but not started.\n");
1441 close(mdfd);
1442 free(devices);
1443 return 0;
1444 }
1445 if (verbose >= -1) {
1446 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1447 if (rebuilding_cnt)
1448 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
1449 if (sparecnt)
1450 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1451 if (!enough(content->array.level, content->array.raid_disks,
1452 content->array.layout, 1, avail, okcnt))
1453 fprintf(stderr, " - not enough to start the array.\n");
1454 else if (!enough(content->array.level,
1455 content->array.raid_disks,
1456 content->array.layout, clean,
1457 avail, okcnt))
1458 fprintf(stderr, " - not enough to start the "
1459 "array while not clean - consider "
1460 "--force.\n");
1461 else {
1462 if (req_cnt == (unsigned)content->array.raid_disks)
1463 fprintf(stderr, " - need all %d to start it", req_cnt);
1464 else
1465 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1466 fprintf(stderr, " (use --run to insist).\n");
1467 }
1468 }
1469 if (auto_assem)
1470 ioctl(mdfd, STOP_ARRAY, NULL);
1471 close(mdfd);
1472 free(devices);
1473 return 1;
1474 } else {
1475 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1476 * been updated to point to the current locations of devices.
1477 * so we can just start the array
1478 */
1479 unsigned long dev;
1480 dev = makedev(devices[chosen_drive].i.disk.major,
1481 devices[chosen_drive].i.disk.minor);
1482 if (ioctl(mdfd, START_ARRAY, dev)) {
1483 fprintf(stderr, Name ": Cannot start array: %s\n",
1484 strerror(errno));
1485 }
1486
1487 }
1488 close(mdfd);
1489 free(devices);
1490 return 0;
1491 }
1492
1493 #ifndef MDASSEMBLE
1494 int assemble_container_content(struct supertype *st, int mdfd,
1495 struct mdinfo *content, int runstop,
1496 char *chosen_name, int verbose)
1497 {
1498 struct mdinfo *dev, *sra;
1499 int working = 0, preexist = 0;
1500 struct map_ent *map = NULL;
1501
1502 sysfs_init(content, mdfd, 0);
1503
1504 sra = sysfs_read(mdfd, 0, GET_VERSION);
1505 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1506 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1507 close(mdfd);
1508 return 1;
1509 }
1510 if (sra)
1511 sysfs_free(sra);
1512
1513 for (dev = content->devs; dev; dev = dev->next)
1514 if (sysfs_add_disk(content, dev, 1) == 0)
1515 working++;
1516 else if (errno == EEXIST)
1517 preexist++;
1518 if (working == 0) {
1519 close(mdfd);
1520 return 1;/* Nothing new, don't try to start */
1521 }
1522
1523 map_update(&map, fd2devnum(mdfd),
1524 content->text_version,
1525 content->uuid, chosen_name);
1526
1527 if (runstop > 0 ||
1528 (working + preexist) >= content->array.working_disks) {
1529 int err;
1530
1531 switch(content->array.level) {
1532 case LEVEL_LINEAR:
1533 case LEVEL_MULTIPATH:
1534 case 0:
1535 err = sysfs_set_str(content, NULL, "array_state",
1536 "active");
1537 break;
1538 default:
1539 err = sysfs_set_str(content, NULL, "array_state",
1540 "readonly");
1541 /* start mdmon if needed. */
1542 if (!err) {
1543 if (!mdmon_running(st->container_dev))
1544 start_mdmon(st->container_dev);
1545 ping_monitor(devnum2devname(st->container_dev));
1546 }
1547 break;
1548 }
1549 if (!err)
1550 sysfs_set_safemode(content, content->safe_mode_delay);
1551 if (verbose >= 0) {
1552 if (err)
1553 fprintf(stderr, Name
1554 ": array %s now has %d devices",
1555 chosen_name, working + preexist);
1556 else
1557 fprintf(stderr, Name
1558 ": Started %s with %d devices",
1559 chosen_name, working + preexist);
1560 if (preexist)
1561 fprintf(stderr, " (%d new)", working);
1562 fprintf(stderr, "\n");
1563 }
1564 if (!err)
1565 wait_for(chosen_name, mdfd);
1566 close(mdfd);
1567 return 0;
1568 /* FIXME should have an O_EXCL and wait for read-auto */
1569 } else {
1570 if (verbose >= 0)
1571 fprintf(stderr, Name
1572 ": %s assembled with %d devices but "
1573 "not started\n",
1574 chosen_name, working);
1575 close(mdfd);
1576 return 1;
1577 }
1578 }
1579 #endif
1580