]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Assemble: allow to assemble spares on their own
[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)
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 && (!st || !st->sb))
568 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
569 if (tmpdev->used != 3)
570 continue;
571 tmpdev->used = 1;
572 content = &info;
573
574 if (!st->sb) {
575 /* we need sb from one of the spares */
576 int dfd = dev_open(tmpdev->devname, O_RDONLY);
577 if (dfd < 0 ||
578 st->ss->load_super(st, dfd, NULL))
579 tmpdev->used = 2;
580 if (dfd > 0)
581 close(dfd);
582 }
583 }
584
585 /* Now reject spares that don't match domains of identified members */
586 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
587 struct stat stb;
588 if (tmpdev->used != 3)
589 continue;
590 if (stat(tmpdev->devname, &stb)< 0) {
591 fprintf(stderr, Name ": fstat failed for %s: %s\n",
592 tmpdev->devname, strerror(errno));
593 tmpdev->used = 2;
594 } else {
595 struct dev_policy *pol = NULL;
596 pol = devnum_policy(stb.st_rdev);
597 if (domain_test(domains, pol, NULL))
598 /* take this spare if domains match */
599 tmpdev->used = 1;
600 else
601 /* if domains don't match mark as unused */
602 tmpdev->used = 0;
603 dev_policy_free(pol);
604 }
605 }
606 domain_free(domains);
607
608 if (!st || !st->sb || !content)
609 return 2;
610
611 /* Now need to open the array device. Use create_mddev */
612 if (content == &info)
613 st->ss->getinfo_super(st, content, NULL);
614
615 trustworthy = FOREIGN;
616 name = content->name;
617 switch (st->ss->match_home(st, homehost)
618 ?: st->ss->match_home(st, "any")) {
619 case 1:
620 trustworthy = LOCAL;
621 name = strchr(content->name, ':');
622 if (name)
623 name++;
624 else
625 name = content->name;
626 break;
627 }
628 if (!auto_assem)
629 /* If the array is listed in mdadm.conf or on
630 * command line, then we trust the name
631 * even if the array doesn't look local
632 */
633 trustworthy = LOCAL;
634
635 if (name[0] == 0 &&
636 content->array.level == LEVEL_CONTAINER) {
637 name = content->text_version;
638 trustworthy = METADATA;
639 }
640
641 if (name[0] && trustworthy != LOCAL &&
642 ! require_homehost &&
643 conf_name_is_free(name))
644 trustworthy = LOCAL;
645
646 if (trustworthy == LOCAL &&
647 strchr(name, ':'))
648 /* Ignore 'host:' prefix of name */
649 name = strchr(name, ':')+1;
650
651 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
652 chosen_name);
653 if (mdfd < 0) {
654 st->ss->free_super(st);
655 if (auto_assem)
656 goto try_again;
657 return 1;
658 }
659 mddev = chosen_name;
660 vers = md_get_version(mdfd);
661 if (vers < 9000) {
662 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
663 " Upgrade your kernel or try --build\n");
664 close(mdfd);
665 return 1;
666 }
667 if (mddev_busy(fd2devnum(mdfd))) {
668 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
669 mddev);
670 for (tmpdev = devlist ;
671 tmpdev && tmpdev->used != 1;
672 tmpdev = tmpdev->next)
673 ;
674 if (tmpdev && auto_assem)
675 fprintf(stderr, Name ": %s needed for %s...\n",
676 mddev, tmpdev->devname);
677 close(mdfd);
678 mdfd = -3;
679 st->ss->free_super(st);
680 if (auto_assem)
681 goto try_again;
682 return 1;
683 }
684 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
685
686 #ifndef MDASSEMBLE
687 if (content != &info) {
688 /* This is a member of a container. Try starting the array. */
689 return assemble_container_content(st, mdfd, content, runstop,
690 chosen_name, verbose);
691 }
692 #endif
693 /* Ok, no bad inconsistancy, we can try updating etc */
694 bitmap_done = 0;
695 content->update_private = NULL;
696 devices = malloc(num_devs * sizeof(*devices));
697 devmap = calloc(num_devs * content->array.raid_disks, 1);
698 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
699 char *devname = tmpdev->devname;
700 struct stat stb;
701 /* looks like a good enough match to update the super block if needed */
702 #ifndef MDASSEMBLE
703 if (update) {
704 int dfd;
705 /* prepare useful information in info structures */
706 struct stat stb2;
707 struct supertype *tst;
708 int err;
709 fstat(mdfd, &stb2);
710
711 if (strcmp(update, "uuid")==0 &&
712 !ident->uuid_set) {
713 int rfd;
714 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
715 read(rfd, ident->uuid, 16) != 16) {
716 *(__u32*)(ident->uuid) = random();
717 *(__u32*)(ident->uuid+1) = random();
718 *(__u32*)(ident->uuid+2) = random();
719 *(__u32*)(ident->uuid+3) = random();
720 }
721 if (rfd >= 0) close(rfd);
722 }
723 dfd = dev_open(devname, O_RDWR|O_EXCL);
724
725 tst = dup_super(st);
726 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
727 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
728 devname);
729 if (dfd >= 0)
730 close(dfd);
731 close(mdfd);
732 free(devices);
733 free(devmap);
734 return 1;
735 }
736 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
737
738 memcpy(content->uuid, ident->uuid, 16);
739 strcpy(content->name, ident->name);
740 content->array.md_minor = minor(stb2.st_rdev);
741
742 if (strcmp(update, "byteorder") == 0)
743 err = 0;
744 else
745 err = tst->ss->update_super(tst, content, update,
746 devname, verbose,
747 ident->uuid_set,
748 homehost);
749 if (err < 0) {
750 fprintf(stderr,
751 Name ": --update=%s not understood"
752 " for %s metadata\n",
753 update, tst->ss->name);
754 tst->ss->free_super(tst);
755 free(tst);
756 close(mdfd);
757 close(dfd);
758 free(devices);
759 free(devmap);
760 return 1;
761 }
762 if (strcmp(update, "uuid")==0 &&
763 !ident->uuid_set) {
764 ident->uuid_set = 1;
765 memcpy(ident->uuid, content->uuid, 16);
766 }
767 if (tst->ss->store_super(tst, dfd))
768 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
769 devname);
770 close(dfd);
771
772 if (strcmp(update, "uuid")==0 &&
773 ident->bitmap_fd >= 0 && !bitmap_done) {
774 if (bitmap_update_uuid(ident->bitmap_fd,
775 content->uuid,
776 tst->ss->swapuuid) != 0)
777 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
778 else
779 bitmap_done = 1;
780 }
781 tst->ss->free_super(tst);
782 } else
783 #endif
784 {
785 struct supertype *tst = dup_super(st);
786 int dfd;
787 dfd = dev_open(devname, O_RDWR|O_EXCL);
788
789 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
790 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
791 devname);
792 if (dfd >= 0)
793 close(dfd);
794 close(mdfd);
795 free(devices);
796 free(devmap);
797 return 1;
798 }
799 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
800 tst->ss->free_super(tst);
801 close(dfd);
802 }
803
804 stat(devname, &stb);
805
806 if (verbose > 0)
807 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
808 devname, mddev, content->disk.raid_disk);
809 devices[devcnt].devname = devname;
810 devices[devcnt].uptodate = 0;
811 devices[devcnt].i = *content;
812 devices[devcnt].i.disk.major = major(stb.st_rdev);
813 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
814 if (most_recent < devcnt) {
815 if (devices[devcnt].i.events
816 > devices[most_recent].i.events)
817 most_recent = devcnt;
818 }
819 if (content->array.level == LEVEL_MULTIPATH)
820 /* with multipath, the raid_disk from the superblock is meaningless */
821 i = devcnt;
822 else
823 i = devices[devcnt].i.disk.raid_disk;
824 if (i+1 == 0) {
825 if (nextspare < content->array.raid_disks)
826 nextspare = content->array.raid_disks;
827 i = nextspare++;
828 } else {
829 if (i >= content->array.raid_disks &&
830 i >= nextspare)
831 nextspare = i+1;
832 }
833 if (i < 10000) {
834 if (i >= bestcnt) {
835 int newbestcnt = i+10;
836 int *newbest = malloc(sizeof(int)*newbestcnt);
837 int c;
838 for (c=0; c < newbestcnt; c++)
839 if (c < bestcnt)
840 newbest[c] = best[c];
841 else
842 newbest[c] = -1;
843 if (best)free(best);
844 best = newbest;
845 bestcnt = newbestcnt;
846 }
847 if (best[i] >=0 &&
848 devices[best[i]].i.events
849 == devices[devcnt].i.events
850 && (devices[best[i]].i.disk.minor
851 != devices[devcnt].i.disk.minor)
852 && st->ss == &super0
853 && content->array.level != LEVEL_MULTIPATH) {
854 /* two different devices with identical superblock.
855 * Could be a mis-detection caused by overlapping
856 * partitions. fail-safe.
857 */
858 fprintf(stderr, Name ": WARNING %s and %s appear"
859 " to have very similar superblocks.\n"
860 " If they are really different, "
861 "please --zero the superblock on one\n"
862 " If they are the same or overlap,"
863 " please remove one from %s.\n",
864 devices[best[i]].devname, devname,
865 inargv ? "the list" :
866 "the\n DEVICE list in mdadm.conf"
867 );
868 close(mdfd);
869 free(devices);
870 free(devmap);
871 return 1;
872 }
873 if (best[i] == -1
874 || (devices[best[i]].i.events
875 < devices[devcnt].i.events))
876 best[i] = devcnt;
877 }
878 devcnt++;
879 }
880 free(content->update_private);
881 content->update_private = NULL;
882
883 if (devcnt == 0) {
884 fprintf(stderr, Name ": no devices found for %s\n",
885 mddev);
886 if (st)
887 st->ss->free_super(st);
888 close(mdfd);
889 free(devices);
890 free(devmap);
891 return 1;
892 }
893
894 if (update && strcmp(update, "byteorder")==0)
895 st->minor_version = 90;
896
897 st->ss->getinfo_super(st, content, NULL);
898 clean = content->array.state & 1;
899
900 /* now we have some devices that might be suitable.
901 * I wonder how many
902 */
903 avail = malloc(content->array.raid_disks);
904 memset(avail, 0, content->array.raid_disks);
905 okcnt = 0;
906 sparecnt=0;
907 rebuilding_cnt=0;
908 for (i=0; i< bestcnt; i++) {
909 int j = best[i];
910 int event_margin = 1; /* always allow a difference of '1'
911 * like the kernel does
912 */
913 if (j < 0) continue;
914 /* note: we ignore error flags in multipath arrays
915 * as they don't make sense
916 */
917 if (content->array.level != LEVEL_MULTIPATH)
918 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
919 if (!(devices[j].i.disk.state
920 & (1<<MD_DISK_FAULTY))) {
921 devices[j].uptodate = 1;
922 sparecnt++;
923 }
924 continue;
925 }
926 /* If this devices thinks that 'most_recent' has failed, then
927 * we must reject this device.
928 */
929 if (j != most_recent &&
930 content->array.raid_disks > 0 &&
931 devices[most_recent].i.disk.raid_disk >= 0 &&
932 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
933 if (verbose > -1)
934 fprintf(stderr, Name ": ignoring %s as it reports %s as failed\n",
935 devices[j].devname, devices[most_recent].devname);
936 best[i] = -1;
937 continue;
938 }
939 if (devices[j].i.events+event_margin >=
940 devices[most_recent].i.events) {
941 devices[j].uptodate = 1;
942 if (i < content->array.raid_disks) {
943 if (devices[j].i.recovery_start == MaxSector ||
944 (content->reshape_active &&
945 j >= content->array.raid_disks - content->delta_disks)) {
946 okcnt++;
947 avail[i]=1;
948 } else
949 rebuilding_cnt++;
950 } else
951 sparecnt++;
952 }
953 }
954 free(devmap);
955 while (force && !enough(content->array.level, content->array.raid_disks,
956 content->array.layout, 1,
957 avail, okcnt)) {
958 /* Choose the newest best drive which is
959 * not up-to-date, update the superblock
960 * and add it.
961 */
962 int fd;
963 struct supertype *tst;
964 unsigned long long current_events;
965 chosen_drive = -1;
966 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
967 int j = best[i];
968 if (j>=0 &&
969 !devices[j].uptodate &&
970 devices[j].i.recovery_start == MaxSector &&
971 (chosen_drive < 0 ||
972 devices[j].i.events
973 > devices[chosen_drive].i.events))
974 chosen_drive = j;
975 }
976 if (chosen_drive < 0)
977 break;
978 current_events = devices[chosen_drive].i.events;
979 add_another:
980 if (verbose >= 0)
981 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
982 devices[chosen_drive].devname,
983 devices[chosen_drive].i.disk.raid_disk,
984 (int)(devices[chosen_drive].i.events),
985 (int)(devices[most_recent].i.events));
986 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
987 if (fd < 0) {
988 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
989 devices[chosen_drive].devname);
990 devices[chosen_drive].i.events = 0;
991 continue;
992 }
993 tst = dup_super(st);
994 if (tst->ss->load_super(tst,fd, NULL)) {
995 close(fd);
996 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
997 devices[chosen_drive].devname);
998 devices[chosen_drive].i.events = 0;
999 continue;
1000 }
1001 content->events = devices[most_recent].i.events;
1002 tst->ss->update_super(tst, content, "force-one",
1003 devices[chosen_drive].devname, verbose,
1004 0, NULL);
1005
1006 if (tst->ss->store_super(tst, fd)) {
1007 close(fd);
1008 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1009 devices[chosen_drive].devname);
1010 devices[chosen_drive].i.events = 0;
1011 tst->ss->free_super(tst);
1012 continue;
1013 }
1014 close(fd);
1015 devices[chosen_drive].i.events = devices[most_recent].i.events;
1016 devices[chosen_drive].uptodate = 1;
1017 avail[chosen_drive] = 1;
1018 okcnt++;
1019 tst->ss->free_super(tst);
1020
1021 /* If there are any other drives of the same vintage,
1022 * add them in as well. We can't lose and we might gain
1023 */
1024 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
1025 int j = best[i];
1026 if (j >= 0 &&
1027 !devices[j].uptodate &&
1028 devices[j].i.events == current_events) {
1029 chosen_drive = j;
1030 goto add_another;
1031 }
1032 }
1033 }
1034
1035 /* Now we want to look at the superblock which the kernel will base things on
1036 * and compare the devices that we think are working with the devices that the
1037 * superblock thinks are working.
1038 * If there are differences and --force is given, then update this chosen
1039 * superblock.
1040 */
1041 chosen_drive = -1;
1042 st->ss->free_super(st);
1043 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
1044 int j = best[i];
1045 int fd;
1046
1047 if (j<0)
1048 continue;
1049 if (!devices[j].uptodate)
1050 continue;
1051 chosen_drive = j;
1052 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
1053 fprintf(stderr, Name ": Cannot open %s: %s\n",
1054 devices[j].devname, strerror(errno));
1055 close(mdfd);
1056 free(devices);
1057 return 1;
1058 }
1059 if (st->ss->load_super(st,fd, NULL)) {
1060 close(fd);
1061 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
1062 devices[j].devname);
1063 close(mdfd);
1064 free(devices);
1065 return 1;
1066 }
1067 close(fd);
1068 }
1069 if (st->sb == NULL) {
1070 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
1071 close(mdfd);
1072 free(devices);
1073 return 1;
1074 }
1075 st->ss->getinfo_super(st, content, NULL);
1076 #ifndef MDASSEMBLE
1077 sysfs_init(content, mdfd, 0);
1078 #endif
1079 for (i=0; i<bestcnt; i++) {
1080 int j = best[i];
1081 unsigned int desired_state;
1082
1083 if (i < content->array.raid_disks)
1084 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1085 else
1086 desired_state = 0;
1087
1088 if (j<0)
1089 continue;
1090 if (!devices[j].uptodate)
1091 continue;
1092
1093 devices[j].i.disk.state = desired_state;
1094 if (!(devices[j].i.array.state & 1))
1095 clean = 0;
1096
1097 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1098 verbose, 0, NULL)) {
1099 if (force) {
1100 if (verbose >= 0)
1101 fprintf(stderr, Name ": "
1102 "clearing FAULTY flag for device %d in %s for %s\n",
1103 j, mddev, devices[j].devname);
1104 change = 1;
1105 } else {
1106 if (verbose >= -1)
1107 fprintf(stderr, Name ": "
1108 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1109 i, mddev, devices[j].devname);
1110 }
1111 }
1112 #if 0
1113 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1114 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1115 i, mddev);
1116 }
1117 #endif
1118 }
1119 if (force && !clean &&
1120 !enough(content->array.level, content->array.raid_disks,
1121 content->array.layout, clean,
1122 avail, okcnt)) {
1123 change += st->ss->update_super(st, content, "force-array",
1124 devices[chosen_drive].devname, verbose,
1125 0, NULL);
1126 clean = 1;
1127 }
1128
1129 if (change) {
1130 int fd;
1131 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
1132 if (fd < 0) {
1133 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1134 devices[chosen_drive].devname);
1135 close(mdfd);
1136 free(devices);
1137 return 1;
1138 }
1139 if (st->ss->store_super(st, fd)) {
1140 close(fd);
1141 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1142 devices[chosen_drive].devname);
1143 close(mdfd);
1144 free(devices);
1145 return 1;
1146 }
1147 close(fd);
1148 }
1149
1150 /* If we are in the middle of a reshape we may need to restore saved data
1151 * that was moved aside due to the reshape overwriting live data
1152 * The code of doing this lives in Grow.c
1153 */
1154 #ifndef MDASSEMBLE
1155 if (content->reshape_active) {
1156 int err = 0;
1157 int *fdlist = malloc(sizeof(int)* bestcnt);
1158 if (verbose > 0)
1159 fprintf(stderr, Name ":%s has an active reshape - checking "
1160 "if critical section needs to be restored\n",
1161 chosen_name);
1162 for (i=0; i<bestcnt; i++) {
1163 int j = best[i];
1164 if (j >= 0) {
1165 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1166 if (fdlist[i] < 0) {
1167 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1168 devices[j].devname);
1169 err = 1;
1170 break;
1171 }
1172 } else
1173 fdlist[i] = -1;
1174 }
1175 if (!err) {
1176 err = Grow_restart(st, content, fdlist, bestcnt,
1177 backup_file, verbose > 0);
1178 if (err && invalid_backup) {
1179 if (verbose > 0)
1180 fprintf(stderr, Name ": continuing"
1181 " without restoring backup\n");
1182 err = 0;
1183 }
1184 }
1185 while (i>0) {
1186 i--;
1187 if (fdlist[i]>=0) close(fdlist[i]);
1188 }
1189 if (err) {
1190 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1191 if (backup_file == NULL)
1192 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1193 close(mdfd);
1194 free(devices);
1195 return err;
1196 }
1197 }
1198 #endif
1199 /* count number of in-sync devices according to the superblock.
1200 * We must have this number to start the array without -s or -R
1201 */
1202 req_cnt = content->array.working_disks;
1203
1204 /* Almost ready to actually *do* something */
1205 if (!old_linux) {
1206 int rv;
1207
1208 /* First, fill in the map, so that udev can find our name
1209 * as soon as we become active.
1210 */
1211 map_update(NULL, fd2devnum(mdfd), content->text_version,
1212 content->uuid, chosen_name);
1213
1214 rv = set_array_info(mdfd, st, content);
1215 if (rv) {
1216 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1217 mddev, strerror(errno));
1218 ioctl(mdfd, STOP_ARRAY, NULL);
1219 close(mdfd);
1220 free(devices);
1221 return 1;
1222 }
1223 if (ident->bitmap_fd >= 0) {
1224 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1225 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1226 ioctl(mdfd, STOP_ARRAY, NULL);
1227 close(mdfd);
1228 free(devices);
1229 return 1;
1230 }
1231 } else if (ident->bitmap_file) {
1232 /* From config file */
1233 int bmfd = open(ident->bitmap_file, O_RDWR);
1234 if (bmfd < 0) {
1235 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1236 ident->bitmap_file);
1237 ioctl(mdfd, STOP_ARRAY, NULL);
1238 close(mdfd);
1239 free(devices);
1240 return 1;
1241 }
1242 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1243 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1244 close(bmfd);
1245 ioctl(mdfd, STOP_ARRAY, NULL);
1246 close(mdfd);
1247 free(devices);
1248 return 1;
1249 }
1250 close(bmfd);
1251 }
1252
1253 /* First, add the raid disks, but add the chosen one last */
1254 for (i=0; i<= bestcnt; i++) {
1255 int j;
1256 if (i < bestcnt) {
1257 j = best[i];
1258 if (j == chosen_drive)
1259 continue;
1260 } else
1261 j = chosen_drive;
1262
1263 if (j >= 0 /* && devices[j].uptodate */) {
1264 int dfd = dev_open(devices[j].devname,
1265 O_RDWR|O_EXCL);
1266 if (dfd >= 0) {
1267 remove_partitions(dfd);
1268 close(dfd);
1269 }
1270 rv = add_disk(mdfd, st, content, &devices[j].i);
1271
1272 if (rv) {
1273 fprintf(stderr, Name ": failed to add "
1274 "%s to %s: %s\n",
1275 devices[j].devname,
1276 mddev,
1277 strerror(errno));
1278 if (i < content->array.raid_disks
1279 || i == bestcnt)
1280 okcnt--;
1281 else
1282 sparecnt--;
1283 } else if (verbose > 0)
1284 fprintf(stderr, Name ": added %s "
1285 "to %s as %d\n",
1286 devices[j].devname, mddev,
1287 devices[j].i.disk.raid_disk);
1288 } else if (verbose > 0 && i < content->array.raid_disks)
1289 fprintf(stderr, Name ": no uptodate device for "
1290 "slot %d of %s\n",
1291 i, mddev);
1292 }
1293
1294 if (content->array.level == LEVEL_CONTAINER) {
1295 if (verbose >= 0) {
1296 fprintf(stderr, Name ": Container %s has been "
1297 "assembled with %d drive%s",
1298 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1299 if (okcnt < (unsigned)content->array.raid_disks)
1300 fprintf(stderr, " (out of %d)",
1301 content->array.raid_disks);
1302 fprintf(stderr, "\n");
1303 }
1304 sysfs_uevent(content, "change");
1305 wait_for(chosen_name, mdfd);
1306 close(mdfd);
1307 free(devices);
1308 return 0;
1309 }
1310
1311 if (runstop == 1 ||
1312 (runstop <= 0 &&
1313 ( enough(content->array.level, content->array.raid_disks,
1314 content->array.layout, clean, avail, okcnt) &&
1315 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1316 ))) {
1317 /* This array is good-to-go.
1318 * If a reshape is in progress then we might need to
1319 * continue monitoring it. In that case we start
1320 * it read-only and let the grow code make it writable.
1321 */
1322 int rv;
1323 #ifndef MDASSEMBLE
1324 if (content->reshape_active &&
1325 content->delta_disks <= 0)
1326 rv = Grow_continue(mdfd, st, content, backup_file);
1327 else
1328 #endif
1329 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1330 if (rv == 0) {
1331 if (verbose >= 0) {
1332 fprintf(stderr, Name ": %s has been started with %d drive%s",
1333 mddev, okcnt, okcnt==1?"":"s");
1334 if (okcnt < (unsigned)content->array.raid_disks)
1335 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1336 if (rebuilding_cnt)
1337 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1338 if (sparecnt)
1339 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1340 fprintf(stderr, ".\n");
1341 }
1342 if (content->reshape_active &&
1343 content->array.level >= 4 &&
1344 content->array.level <= 6) {
1345 /* might need to increase the size
1346 * of the stripe cache - default is 256
1347 */
1348 if (256 < 4 * (content->array.chunk_size/4096)) {
1349 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1350 if (sra)
1351 sysfs_set_num(sra, NULL,
1352 "stripe_cache_size",
1353 (4 * content->array.chunk_size / 4096) + 1);
1354 }
1355 }
1356 if (okcnt < (unsigned)content->array.raid_disks) {
1357 /* If any devices did not get added
1358 * because the kernel rejected them based
1359 * on event count, try adding them
1360 * again providing the action policy is
1361 * 're-add' or greater. The bitmap
1362 * might allow them to be included, or
1363 * they will become spares.
1364 */
1365 for (i = 0; i <= bestcnt; i++) {
1366 int j = best[i];
1367 if (j >= 0 && !devices[j].uptodate) {
1368 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1369 continue;
1370 rv = add_disk(mdfd, st, content,
1371 &devices[j].i);
1372 if (rv == 0 && verbose >= 0)
1373 fprintf(stderr,
1374 Name ": %s has been re-added.\n",
1375 devices[j].devname);
1376 }
1377 }
1378 }
1379 wait_for(mddev, mdfd);
1380 close(mdfd);
1381 if (auto_assem) {
1382 int usecs = 1;
1383 /* There is a nasty race with 'mdadm --monitor'.
1384 * If it opens this device before we close it,
1385 * it gets an incomplete open on which IO
1386 * doesn't work and the capacity is
1387 * wrong.
1388 * If we reopen (to check for layered devices)
1389 * before --monitor closes, we loose.
1390 *
1391 * So: wait upto 1 second for there to be
1392 * a non-zero capacity.
1393 */
1394 while (usecs < 1000) {
1395 mdfd = open(mddev, O_RDONLY);
1396 if (mdfd >= 0) {
1397 unsigned long long size;
1398 if (get_dev_size(mdfd, NULL, &size) &&
1399 size > 0)
1400 break;
1401 close(mdfd);
1402 }
1403 usleep(usecs);
1404 usecs <<= 1;
1405 }
1406 }
1407 free(devices);
1408 return 0;
1409 }
1410 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1411 mddev, strerror(errno));
1412
1413 if (!enough(content->array.level, content->array.raid_disks,
1414 content->array.layout, 1, avail, okcnt))
1415 fprintf(stderr, Name ": Not enough devices to "
1416 "start the array.\n");
1417 else if (!enough(content->array.level,
1418 content->array.raid_disks,
1419 content->array.layout, clean,
1420 avail, okcnt))
1421 fprintf(stderr, Name ": Not enough devices to "
1422 "start the array while not clean "
1423 "- consider --force.\n");
1424
1425 if (auto_assem)
1426 ioctl(mdfd, STOP_ARRAY, NULL);
1427 close(mdfd);
1428 free(devices);
1429 return 1;
1430 }
1431 if (runstop == -1) {
1432 fprintf(stderr, Name ": %s assembled from %d drive%s",
1433 mddev, okcnt, okcnt==1?"":"s");
1434 if (okcnt != (unsigned)content->array.raid_disks)
1435 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1436 fprintf(stderr, ", but not started.\n");
1437 close(mdfd);
1438 free(devices);
1439 return 0;
1440 }
1441 if (verbose >= -1) {
1442 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1443 if (rebuilding_cnt)
1444 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
1445 if (sparecnt)
1446 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1447 if (!enough(content->array.level, content->array.raid_disks,
1448 content->array.layout, 1, avail, okcnt))
1449 fprintf(stderr, " - not enough to start the array.\n");
1450 else if (!enough(content->array.level,
1451 content->array.raid_disks,
1452 content->array.layout, clean,
1453 avail, okcnt))
1454 fprintf(stderr, " - not enough to start the "
1455 "array while not clean - consider "
1456 "--force.\n");
1457 else {
1458 if (req_cnt == (unsigned)content->array.raid_disks)
1459 fprintf(stderr, " - need all %d to start it", req_cnt);
1460 else
1461 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1462 fprintf(stderr, " (use --run to insist).\n");
1463 }
1464 }
1465 if (auto_assem)
1466 ioctl(mdfd, STOP_ARRAY, NULL);
1467 close(mdfd);
1468 free(devices);
1469 return 1;
1470 } else {
1471 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1472 * been updated to point to the current locations of devices.
1473 * so we can just start the array
1474 */
1475 unsigned long dev;
1476 dev = makedev(devices[chosen_drive].i.disk.major,
1477 devices[chosen_drive].i.disk.minor);
1478 if (ioctl(mdfd, START_ARRAY, dev)) {
1479 fprintf(stderr, Name ": Cannot start array: %s\n",
1480 strerror(errno));
1481 }
1482
1483 }
1484 close(mdfd);
1485 free(devices);
1486 return 0;
1487 }
1488
1489 #ifndef MDASSEMBLE
1490 int assemble_container_content(struct supertype *st, int mdfd,
1491 struct mdinfo *content, int runstop,
1492 char *chosen_name, int verbose)
1493 {
1494 struct mdinfo *dev, *sra;
1495 int working = 0, preexist = 0;
1496 struct map_ent *map = NULL;
1497
1498 sysfs_init(content, mdfd, 0);
1499
1500 sra = sysfs_read(mdfd, 0, GET_VERSION);
1501 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1502 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1503 close(mdfd);
1504 return 1;
1505 }
1506 if (sra)
1507 sysfs_free(sra);
1508
1509 for (dev = content->devs; dev; dev = dev->next)
1510 if (sysfs_add_disk(content, dev, 1) == 0)
1511 working++;
1512 else if (errno == EEXIST)
1513 preexist++;
1514 if (working == 0) {
1515 close(mdfd);
1516 return 1;/* Nothing new, don't try to start */
1517 }
1518
1519 map_update(&map, fd2devnum(mdfd),
1520 content->text_version,
1521 content->uuid, chosen_name);
1522
1523 if (runstop > 0 ||
1524 (working + preexist) >= content->array.working_disks) {
1525 int err;
1526
1527 switch(content->array.level) {
1528 case LEVEL_LINEAR:
1529 case LEVEL_MULTIPATH:
1530 case 0:
1531 err = sysfs_set_str(content, NULL, "array_state",
1532 "active");
1533 break;
1534 default:
1535 err = sysfs_set_str(content, NULL, "array_state",
1536 "readonly");
1537 /* start mdmon if needed. */
1538 if (!err) {
1539 if (!mdmon_running(st->container_dev))
1540 start_mdmon(st->container_dev);
1541 ping_monitor(devnum2devname(st->container_dev));
1542 }
1543 break;
1544 }
1545 if (!err)
1546 sysfs_set_safemode(content, content->safe_mode_delay);
1547 if (verbose >= 0) {
1548 if (err)
1549 fprintf(stderr, Name
1550 ": array %s now has %d devices",
1551 chosen_name, working + preexist);
1552 else
1553 fprintf(stderr, Name
1554 ": Started %s with %d devices",
1555 chosen_name, working + preexist);
1556 if (preexist)
1557 fprintf(stderr, " (%d new)", working);
1558 fprintf(stderr, "\n");
1559 }
1560 if (!err)
1561 wait_for(chosen_name, mdfd);
1562 close(mdfd);
1563 return 0;
1564 /* FIXME should have an O_EXCL and wait for read-auto */
1565 } else {
1566 if (verbose >= 0)
1567 fprintf(stderr, Name
1568 ": %s assembled with %d devices but "
1569 "not started\n",
1570 chosen_name, working);
1571 close(mdfd);
1572 return 1;
1573 }
1574 }
1575 #endif
1576