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