]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
fix: assemble for external metadata generates segfault if invalid device 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 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
336 || tst->ss->load_container(tst, dfd, NULL)) {
337 if (report_missmatch)
338 fprintf(stderr, Name ": no correct container type: %s\n",
339 devname);
340 tmpdev->used = 2;
341 } else if (auto_assem &&
342 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
343 tst->ss->match_home(tst, homehost) == 1)) {
344 if (report_missmatch)
345 fprintf(stderr, Name ": %s has metadata type %s for which "
346 "auto-assembly is disabled\n",
347 devname, tst->ss->name);
348 tmpdev->used = 2;
349 } else
350 found_container = 1;
351 } else {
352 if (!tst && (tst = guess_super(dfd)) == NULL) {
353 if (report_missmatch)
354 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
355 devname);
356 tmpdev->used = 2;
357 } else if (tst->ss->load_super(tst,dfd, NULL)) {
358 if (report_missmatch)
359 fprintf(stderr, Name ": no RAID superblock on %s\n",
360 devname);
361 tmpdev->used = 2;
362 } else if (tst->ss->compare_super == NULL) {
363 if (report_missmatch)
364 fprintf(stderr, Name ": Cannot assemble %s metadata on %s\n",
365 tst->ss->name, devname);
366 tmpdev->used = 2;
367 } else if (auto_assem && st == NULL &&
368 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
369 tst->ss->match_home(tst, homehost) == 1)) {
370 if (report_missmatch)
371 fprintf(stderr, Name ": %s has metadata type %s for which "
372 "auto-assembly is disabled\n",
373 devname, tst->ss->name);
374 tmpdev->used = 2;
375 }
376 }
377 if (dfd >= 0) close(dfd);
378 if (tmpdev->used == 2) {
379 if (auto_assem)
380 /* Ignore unrecognised devices during auto-assembly */
381 goto loop;
382 if (ident->uuid_set || ident->name[0] ||
383 ident->super_minor != UnSet)
384 /* Ignore unrecognised device if looking for
385 * specific array */
386 goto loop;
387
388
389 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
390 devname);
391 if (st)
392 st->ss->free_super(st);
393 dev_policy_free(pol);
394 return 1;
395 }
396
397 if (found_container) {
398 /* tmpdev is a container. We need to be either
399 * looking for a member, or auto-assembling
400 */
401
402 if (ident->container) {
403 if (ident->container[0] == '/' &&
404 !same_dev(ident->container, devname)) {
405 if (report_missmatch)
406 fprintf(stderr, Name ": %s is not the container required (%s)\n",
407 devname, ident->container);
408 goto loop;
409 }
410 if (ident->container[0] != '/') {
411 /* we have a uuid */
412 int uuid[4];
413
414 content = &info;
415 memset(content, 0, sizeof(*content));
416 tst->ss->getinfo_super(tst, content, NULL);
417
418 if (!parse_uuid(ident->container, uuid) ||
419 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
420 if (report_missmatch)
421 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
422 devname);
423 goto loop;
424 }
425 }
426 }
427 /* It is worth looking inside this container.
428 */
429 if (verbose > 0)
430 fprintf(stderr, Name ": looking in container %s\n",
431 devname);
432
433 for (content = tst->ss->container_content(tst, NULL);
434 content;
435 content = content->next) {
436
437 if (!ident_matches(ident, content, tst,
438 homehost, update,
439 report_missmatch ? devname : NULL))
440 /* message already printed */;
441 else if (is_member_busy(content->text_version)) {
442 if (report_missmatch)
443 fprintf(stderr, Name ": member %s in %s is already assembled\n",
444 content->text_version,
445 devname);
446 } else
447 break;
448 }
449 if (!content) {
450 tmpdev->used = 2;
451 goto loop; /* empty container */
452 }
453
454 st = tst; tst = NULL;
455 if (!auto_assem && inargv && tmpdev->next != NULL) {
456 fprintf(stderr, Name ": %s is a container, but is not "
457 "only device given: confused and aborting\n",
458 devname);
459 st->ss->free_super(st);
460 dev_policy_free(pol);
461 return 1;
462 }
463 if (verbose > 0)
464 fprintf(stderr, Name ": found match on member %s in %s\n",
465 content->text_version, devname);
466
467 /* make sure we finished the loop */
468 tmpdev = NULL;
469 goto loop;
470 } else {
471
472 content = &info;
473 memset(content, 0, sizeof(*content));
474 tst->ss->getinfo_super(tst, content, NULL);
475
476 if (!ident_matches(ident, content, tst,
477 homehost, update,
478 report_missmatch ? devname : NULL))
479 goto loop;
480
481 if (st == NULL)
482 st = dup_super(tst);
483 if (st->minor_version == -1)
484 st->minor_version = tst->minor_version;
485 if (st->ss != tst->ss ||
486 st->minor_version != tst->minor_version ||
487 st->ss->compare_super(st, tst) != 0) {
488 /* Some mismatch. If exactly one array matches this host,
489 * we can resolve on that one.
490 * Or, if we are auto assembling, we just ignore the second
491 * for now.
492 */
493 if (auto_assem)
494 goto loop;
495 if (homehost) {
496 int first = st->ss->match_home(st, homehost);
497 int last = tst->ss->match_home(tst, homehost);
498 if (first != last &&
499 (first == 1 || last == 1)) {
500 /* We can do something */
501 if (first) {/* just ignore this one */
502 if (report_missmatch)
503 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
504 devname);
505 goto loop;
506 } else { /* reject all those sofar */
507 struct mddev_dev *td;
508 if (report_missmatch)
509 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
510 devname);
511 for (td=devlist; td != tmpdev; td=td->next)
512 if (td->used == 1)
513 td->used = 0;
514 tmpdev->used = 1;
515 goto loop;
516 }
517 }
518 }
519 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
520 devname);
521 tst->ss->free_super(tst);
522 st->ss->free_super(st);
523 dev_policy_free(pol);
524 return 1;
525 }
526 tmpdev->used = 1;
527 }
528 loop:
529 dev_policy_free(pol);
530 pol = NULL;
531 if (tst)
532 tst->ss->free_super(tst);
533 }
534
535 if (!st || !st->sb || !content)
536 return 2;
537
538 /* Now need to open the array device. Use create_mddev */
539 if (content == &info)
540 st->ss->getinfo_super(st, content, NULL);
541
542 trustworthy = FOREIGN;
543 name = content->name;
544 switch (st->ss->match_home(st, homehost)
545 ?: st->ss->match_home(st, "any")) {
546 case 1:
547 trustworthy = LOCAL;
548 name = strchr(content->name, ':');
549 if (name)
550 name++;
551 else
552 name = content->name;
553 break;
554 }
555 if (!auto_assem)
556 /* If the array is listed in mdadm.conf or on
557 * command line, then we trust the name
558 * even if the array doesn't look local
559 */
560 trustworthy = LOCAL;
561
562 if (name[0] == 0 &&
563 content->array.level == LEVEL_CONTAINER) {
564 name = content->text_version;
565 trustworthy = METADATA;
566 }
567
568 if (name[0] && trustworthy != LOCAL &&
569 ! require_homehost &&
570 conf_name_is_free(name))
571 trustworthy = LOCAL;
572
573 if (trustworthy == LOCAL &&
574 strchr(name, ':'))
575 /* Ignore 'host:' prefix of name */
576 name = strchr(name, ':')+1;
577
578 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
579 chosen_name);
580 if (mdfd < 0) {
581 st->ss->free_super(st);
582 if (auto_assem)
583 goto try_again;
584 return 1;
585 }
586 mddev = chosen_name;
587 vers = md_get_version(mdfd);
588 if (vers < 9000) {
589 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
590 " Upgrade your kernel or try --build\n");
591 close(mdfd);
592 return 1;
593 }
594 if (mddev_busy(fd2devnum(mdfd))) {
595 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
596 mddev);
597 for (tmpdev = devlist ;
598 tmpdev && tmpdev->used != 1;
599 tmpdev = tmpdev->next)
600 ;
601 if (tmpdev && auto_assem)
602 fprintf(stderr, Name ": %s needed for %s...\n",
603 mddev, tmpdev->devname);
604 close(mdfd);
605 mdfd = -3;
606 st->ss->free_super(st);
607 if (auto_assem)
608 goto try_again;
609 return 1;
610 }
611 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
612
613 #ifndef MDASSEMBLE
614 if (content != &info) {
615 /* This is a member of a container. Try starting the array. */
616 return assemble_container_content(st, mdfd, content, runstop,
617 chosen_name, verbose);
618 }
619 #endif
620 /* Ok, no bad inconsistancy, we can try updating etc */
621 bitmap_done = 0;
622 content->update_private = NULL;
623 devices = malloc(num_devs * sizeof(*devices));
624 devmap = calloc(num_devs * content->array.raid_disks, 1);
625 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
626 char *devname = tmpdev->devname;
627 struct stat stb;
628 /* looks like a good enough match to update the super block if needed */
629 #ifndef MDASSEMBLE
630 if (update) {
631 int dfd;
632 /* prepare useful information in info structures */
633 struct stat stb2;
634 struct supertype *tst;
635 int err;
636 fstat(mdfd, &stb2);
637
638 if (strcmp(update, "uuid")==0 &&
639 !ident->uuid_set) {
640 int rfd;
641 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
642 read(rfd, ident->uuid, 16) != 16) {
643 *(__u32*)(ident->uuid) = random();
644 *(__u32*)(ident->uuid+1) = random();
645 *(__u32*)(ident->uuid+2) = random();
646 *(__u32*)(ident->uuid+3) = random();
647 }
648 if (rfd >= 0) close(rfd);
649 }
650 dfd = dev_open(devname, O_RDWR|O_EXCL);
651
652 tst = dup_super(st);
653 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
654 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
655 devname);
656 if (dfd >= 0)
657 close(dfd);
658 close(mdfd);
659 free(devices);
660 free(devmap);
661 return 1;
662 }
663 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
664
665 memcpy(content->uuid, ident->uuid, 16);
666 strcpy(content->name, ident->name);
667 content->array.md_minor = minor(stb2.st_rdev);
668
669 if (strcmp(update, "byteorder") == 0)
670 err = 0;
671 else
672 err = tst->ss->update_super(tst, content, update,
673 devname, verbose,
674 ident->uuid_set,
675 homehost);
676 if (err < 0) {
677 fprintf(stderr,
678 Name ": --update=%s not understood"
679 " for %s metadata\n",
680 update, tst->ss->name);
681 tst->ss->free_super(tst);
682 free(tst);
683 close(mdfd);
684 close(dfd);
685 free(devices);
686 free(devmap);
687 return 1;
688 }
689 if (strcmp(update, "uuid")==0 &&
690 !ident->uuid_set) {
691 ident->uuid_set = 1;
692 memcpy(ident->uuid, content->uuid, 16);
693 }
694 if (tst->ss->store_super(tst, dfd))
695 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
696 devname);
697 close(dfd);
698
699 if (strcmp(update, "uuid")==0 &&
700 ident->bitmap_fd >= 0 && !bitmap_done) {
701 if (bitmap_update_uuid(ident->bitmap_fd,
702 content->uuid,
703 tst->ss->swapuuid) != 0)
704 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
705 else
706 bitmap_done = 1;
707 }
708 tst->ss->free_super(tst);
709 } else
710 #endif
711 {
712 struct supertype *tst = dup_super(st);
713 int dfd;
714 dfd = dev_open(devname, O_RDWR|O_EXCL);
715
716 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
717 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
718 devname);
719 if (dfd >= 0)
720 close(dfd);
721 close(mdfd);
722 free(devices);
723 free(devmap);
724 return 1;
725 }
726 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
727 tst->ss->free_super(tst);
728 close(dfd);
729 }
730
731 stat(devname, &stb);
732
733 if (verbose > 0)
734 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
735 devname, mddev, content->disk.raid_disk);
736 devices[devcnt].devname = devname;
737 devices[devcnt].uptodate = 0;
738 devices[devcnt].i = *content;
739 devices[devcnt].i.disk.major = major(stb.st_rdev);
740 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
741 if (most_recent < devcnt) {
742 if (devices[devcnt].i.events
743 > devices[most_recent].i.events)
744 most_recent = devcnt;
745 }
746 if (content->array.level == LEVEL_MULTIPATH)
747 /* with multipath, the raid_disk from the superblock is meaningless */
748 i = devcnt;
749 else
750 i = devices[devcnt].i.disk.raid_disk;
751 if (i+1 == 0) {
752 if (nextspare < content->array.raid_disks)
753 nextspare = content->array.raid_disks;
754 i = nextspare++;
755 } else {
756 if (i >= content->array.raid_disks &&
757 i >= nextspare)
758 nextspare = i+1;
759 }
760 if (i < 10000) {
761 if (i >= bestcnt) {
762 int newbestcnt = i+10;
763 int *newbest = malloc(sizeof(int)*newbestcnt);
764 int c;
765 for (c=0; c < newbestcnt; c++)
766 if (c < bestcnt)
767 newbest[c] = best[c];
768 else
769 newbest[c] = -1;
770 if (best)free(best);
771 best = newbest;
772 bestcnt = newbestcnt;
773 }
774 if (best[i] >=0 &&
775 devices[best[i]].i.events
776 == devices[devcnt].i.events
777 && (devices[best[i]].i.disk.minor
778 != devices[devcnt].i.disk.minor)
779 && st->ss == &super0
780 && content->array.level != LEVEL_MULTIPATH) {
781 /* two different devices with identical superblock.
782 * Could be a mis-detection caused by overlapping
783 * partitions. fail-safe.
784 */
785 fprintf(stderr, Name ": WARNING %s and %s appear"
786 " to have very similar superblocks.\n"
787 " If they are really different, "
788 "please --zero the superblock on one\n"
789 " If they are the same or overlap,"
790 " please remove one from %s.\n",
791 devices[best[i]].devname, devname,
792 inargv ? "the list" :
793 "the\n DEVICE list in mdadm.conf"
794 );
795 close(mdfd);
796 free(devices);
797 free(devmap);
798 return 1;
799 }
800 if (best[i] == -1
801 || (devices[best[i]].i.events
802 < devices[devcnt].i.events))
803 best[i] = devcnt;
804 }
805 devcnt++;
806 }
807 free(content->update_private);
808 content->update_private = NULL;
809
810 if (devcnt == 0) {
811 fprintf(stderr, Name ": no devices found for %s\n",
812 mddev);
813 if (st)
814 st->ss->free_super(st);
815 close(mdfd);
816 free(devices);
817 free(devmap);
818 return 1;
819 }
820
821 if (update && strcmp(update, "byteorder")==0)
822 st->minor_version = 90;
823
824 st->ss->getinfo_super(st, content, NULL);
825 clean = content->array.state & 1;
826
827 /* now we have some devices that might be suitable.
828 * I wonder how many
829 */
830 avail = malloc(content->array.raid_disks);
831 memset(avail, 0, content->array.raid_disks);
832 okcnt = 0;
833 sparecnt=0;
834 rebuilding_cnt=0;
835 for (i=0; i< bestcnt; i++) {
836 int j = best[i];
837 int event_margin = 1; /* always allow a difference of '1'
838 * like the kernel does
839 */
840 if (j < 0) continue;
841 /* note: we ignore error flags in multipath arrays
842 * as they don't make sense
843 */
844 if (content->array.level != LEVEL_MULTIPATH)
845 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
846 if (!(devices[j].i.disk.state
847 & (1<<MD_DISK_FAULTY)))
848 sparecnt++;
849 continue;
850 }
851 /* If this devices thinks that 'most_recent' has failed, then
852 * we must reject this device.
853 */
854 if (j != most_recent &&
855 content->array.raid_disks > 0 &&
856 devices[most_recent].i.disk.raid_disk >= 0 &&
857 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
858 if (verbose > -1)
859 fprintf(stderr, Name ": ignoring %s as it reports %s as failed\n",
860 devices[j].devname, devices[most_recent].devname);
861 best[i] = -1;
862 continue;
863 }
864 if (devices[j].i.events+event_margin >=
865 devices[most_recent].i.events) {
866 devices[j].uptodate = 1;
867 if (i < content->array.raid_disks) {
868 if (devices[j].i.recovery_start == MaxSector ||
869 (content->reshape_active &&
870 j >= content->array.raid_disks - content->delta_disks)) {
871 okcnt++;
872 avail[i]=1;
873 } else
874 rebuilding_cnt++;
875 } else
876 sparecnt++;
877 }
878 }
879 free(devmap);
880 while (force && !enough(content->array.level, content->array.raid_disks,
881 content->array.layout, 1,
882 avail, okcnt)) {
883 /* Choose the newest best drive which is
884 * not up-to-date, update the superblock
885 * and add it.
886 */
887 int fd;
888 struct supertype *tst;
889 unsigned long long current_events;
890 chosen_drive = -1;
891 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
892 int j = best[i];
893 if (j>=0 &&
894 !devices[j].uptodate &&
895 devices[j].i.recovery_start == MaxSector &&
896 (chosen_drive < 0 ||
897 devices[j].i.events
898 > devices[chosen_drive].i.events))
899 chosen_drive = j;
900 }
901 if (chosen_drive < 0)
902 break;
903 current_events = devices[chosen_drive].i.events;
904 add_another:
905 if (verbose >= 0)
906 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
907 devices[chosen_drive].devname,
908 devices[chosen_drive].i.disk.raid_disk,
909 (int)(devices[chosen_drive].i.events),
910 (int)(devices[most_recent].i.events));
911 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
912 if (fd < 0) {
913 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
914 devices[chosen_drive].devname);
915 devices[chosen_drive].i.events = 0;
916 continue;
917 }
918 tst = dup_super(st);
919 if (tst->ss->load_super(tst,fd, NULL)) {
920 close(fd);
921 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
922 devices[chosen_drive].devname);
923 devices[chosen_drive].i.events = 0;
924 continue;
925 }
926 content->events = devices[most_recent].i.events;
927 tst->ss->update_super(tst, content, "force-one",
928 devices[chosen_drive].devname, verbose,
929 0, NULL);
930
931 if (tst->ss->store_super(tst, fd)) {
932 close(fd);
933 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
934 devices[chosen_drive].devname);
935 devices[chosen_drive].i.events = 0;
936 tst->ss->free_super(tst);
937 continue;
938 }
939 close(fd);
940 devices[chosen_drive].i.events = devices[most_recent].i.events;
941 devices[chosen_drive].uptodate = 1;
942 avail[chosen_drive] = 1;
943 okcnt++;
944 tst->ss->free_super(tst);
945
946 /* If there are any other drives of the same vintage,
947 * add them in as well. We can't lose and we might gain
948 */
949 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
950 int j = best[i];
951 if (j >= 0 &&
952 !devices[j].uptodate &&
953 devices[j].i.events == current_events) {
954 chosen_drive = j;
955 goto add_another;
956 }
957 }
958 }
959
960 /* Now we want to look at the superblock which the kernel will base things on
961 * and compare the devices that we think are working with the devices that the
962 * superblock thinks are working.
963 * If there are differences and --force is given, then update this chosen
964 * superblock.
965 */
966 chosen_drive = -1;
967 st->ss->free_super(st);
968 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
969 int j = best[i];
970 int fd;
971
972 if (j<0)
973 continue;
974 if (!devices[j].uptodate)
975 continue;
976 chosen_drive = j;
977 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
978 fprintf(stderr, Name ": Cannot open %s: %s\n",
979 devices[j].devname, strerror(errno));
980 close(mdfd);
981 free(devices);
982 return 1;
983 }
984 if (st->ss->load_super(st,fd, NULL)) {
985 close(fd);
986 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
987 devices[j].devname);
988 close(mdfd);
989 free(devices);
990 return 1;
991 }
992 close(fd);
993 }
994 if (st->sb == NULL) {
995 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
996 close(mdfd);
997 free(devices);
998 return 1;
999 }
1000 st->ss->getinfo_super(st, content, NULL);
1001 #ifndef MDASSEMBLE
1002 sysfs_init(content, mdfd, 0);
1003 #endif
1004 for (i=0; i<bestcnt; i++) {
1005 int j = best[i];
1006 unsigned int desired_state;
1007
1008 if (i < content->array.raid_disks)
1009 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1010 else
1011 desired_state = 0;
1012
1013 if (j<0)
1014 continue;
1015 if (!devices[j].uptodate)
1016 continue;
1017
1018 devices[j].i.disk.state = desired_state;
1019 if (!(devices[j].i.array.state & 1))
1020 clean = 0;
1021
1022 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1023 verbose, 0, NULL)) {
1024 if (force) {
1025 if (verbose >= 0)
1026 fprintf(stderr, Name ": "
1027 "clearing FAULTY flag for device %d in %s for %s\n",
1028 j, mddev, devices[j].devname);
1029 change = 1;
1030 } else {
1031 if (verbose >= -1)
1032 fprintf(stderr, Name ": "
1033 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1034 i, mddev, devices[j].devname);
1035 }
1036 }
1037 #if 0
1038 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1039 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1040 i, mddev);
1041 }
1042 #endif
1043 }
1044 if (force && !clean &&
1045 !enough(content->array.level, content->array.raid_disks,
1046 content->array.layout, clean,
1047 avail, okcnt)) {
1048 change += st->ss->update_super(st, content, "force-array",
1049 devices[chosen_drive].devname, verbose,
1050 0, NULL);
1051 clean = 1;
1052 }
1053
1054 if (change) {
1055 int fd;
1056 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
1057 if (fd < 0) {
1058 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1059 devices[chosen_drive].devname);
1060 close(mdfd);
1061 free(devices);
1062 return 1;
1063 }
1064 if (st->ss->store_super(st, fd)) {
1065 close(fd);
1066 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1067 devices[chosen_drive].devname);
1068 close(mdfd);
1069 free(devices);
1070 return 1;
1071 }
1072 close(fd);
1073 }
1074
1075 /* If we are in the middle of a reshape we may need to restore saved data
1076 * that was moved aside due to the reshape overwriting live data
1077 * The code of doing this lives in Grow.c
1078 */
1079 #ifndef MDASSEMBLE
1080 if (content->reshape_active) {
1081 int err = 0;
1082 int *fdlist = malloc(sizeof(int)* bestcnt);
1083 if (verbose > 0)
1084 fprintf(stderr, Name ":%s has an active reshape - checking "
1085 "if critical section needs to be restored\n",
1086 chosen_name);
1087 for (i=0; i<bestcnt; i++) {
1088 int j = best[i];
1089 if (j >= 0) {
1090 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1091 if (fdlist[i] < 0) {
1092 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1093 devices[j].devname);
1094 err = 1;
1095 break;
1096 }
1097 } else
1098 fdlist[i] = -1;
1099 }
1100 if (!err)
1101 err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0);
1102 while (i>0) {
1103 i--;
1104 if (fdlist[i]>=0) close(fdlist[i]);
1105 }
1106 if (err) {
1107 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1108 if (backup_file == NULL)
1109 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1110 close(mdfd);
1111 free(devices);
1112 return err;
1113 }
1114 }
1115 #endif
1116 /* count number of in-sync devices according to the superblock.
1117 * We must have this number to start the array without -s or -R
1118 */
1119 req_cnt = content->array.working_disks;
1120
1121 /* Almost ready to actually *do* something */
1122 if (!old_linux) {
1123 int rv;
1124
1125 /* First, fill in the map, so that udev can find our name
1126 * as soon as we become active.
1127 */
1128 map_update(NULL, fd2devnum(mdfd), content->text_version,
1129 content->uuid, chosen_name);
1130
1131 rv = set_array_info(mdfd, st, content);
1132 if (rv) {
1133 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1134 mddev, strerror(errno));
1135 ioctl(mdfd, STOP_ARRAY, NULL);
1136 close(mdfd);
1137 free(devices);
1138 return 1;
1139 }
1140 if (ident->bitmap_fd >= 0) {
1141 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1142 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1143 ioctl(mdfd, STOP_ARRAY, NULL);
1144 close(mdfd);
1145 free(devices);
1146 return 1;
1147 }
1148 } else if (ident->bitmap_file) {
1149 /* From config file */
1150 int bmfd = open(ident->bitmap_file, O_RDWR);
1151 if (bmfd < 0) {
1152 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1153 ident->bitmap_file);
1154 ioctl(mdfd, STOP_ARRAY, NULL);
1155 close(mdfd);
1156 free(devices);
1157 return 1;
1158 }
1159 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1160 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1161 close(bmfd);
1162 ioctl(mdfd, STOP_ARRAY, NULL);
1163 close(mdfd);
1164 free(devices);
1165 return 1;
1166 }
1167 close(bmfd);
1168 }
1169
1170 /* First, add the raid disks, but add the chosen one last */
1171 for (i=0; i<= bestcnt; i++) {
1172 int j;
1173 if (i < bestcnt) {
1174 j = best[i];
1175 if (j == chosen_drive)
1176 continue;
1177 } else
1178 j = chosen_drive;
1179
1180 if (j >= 0 /* && devices[j].uptodate */) {
1181 int dfd = dev_open(devices[j].devname,
1182 O_RDWR|O_EXCL);
1183 if (dfd >= 0) {
1184 remove_partitions(dfd);
1185 close(dfd);
1186 }
1187 rv = add_disk(mdfd, st, content, &devices[j].i);
1188
1189 if (rv) {
1190 fprintf(stderr, Name ": failed to add "
1191 "%s to %s: %s\n",
1192 devices[j].devname,
1193 mddev,
1194 strerror(errno));
1195 if (i < content->array.raid_disks
1196 || i == bestcnt)
1197 okcnt--;
1198 else
1199 sparecnt--;
1200 } else if (verbose > 0)
1201 fprintf(stderr, Name ": added %s "
1202 "to %s as %d\n",
1203 devices[j].devname, mddev,
1204 devices[j].i.disk.raid_disk);
1205 } else if (verbose > 0 && i < content->array.raid_disks)
1206 fprintf(stderr, Name ": no uptodate device for "
1207 "slot %d of %s\n",
1208 i, mddev);
1209 }
1210
1211 if (content->array.level == LEVEL_CONTAINER) {
1212 if (verbose >= 0) {
1213 fprintf(stderr, Name ": Container %s has been "
1214 "assembled with %d drive%s",
1215 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1216 if (okcnt < (unsigned)content->array.raid_disks)
1217 fprintf(stderr, " (out of %d)",
1218 content->array.raid_disks);
1219 fprintf(stderr, "\n");
1220 }
1221 sysfs_uevent(content, "change");
1222 wait_for(chosen_name, mdfd);
1223 close(mdfd);
1224 free(devices);
1225 return 0;
1226 }
1227
1228 if (runstop == 1 ||
1229 (runstop <= 0 &&
1230 ( enough(content->array.level, content->array.raid_disks,
1231 content->array.layout, clean, avail, okcnt) &&
1232 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1233 ))) {
1234 /* This array is good-to-go.
1235 * If a reshape is in progress then we might need to
1236 * continue monitoring it. In that case we start
1237 * it read-only and let the grow code make it writable.
1238 */
1239 int rv;
1240 #ifndef MDASSEMBLE
1241 if (content->reshape_active &&
1242 content->delta_disks <= 0)
1243 rv = Grow_continue(mdfd, st, content, backup_file);
1244 else
1245 #endif
1246 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1247 if (rv == 0) {
1248 if (verbose >= 0) {
1249 fprintf(stderr, Name ": %s has been started with %d drive%s",
1250 mddev, okcnt, okcnt==1?"":"s");
1251 if (okcnt < (unsigned)content->array.raid_disks)
1252 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1253 if (rebuilding_cnt)
1254 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1255 if (sparecnt)
1256 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1257 fprintf(stderr, ".\n");
1258 }
1259 if (content->reshape_active &&
1260 content->array.level >= 4 &&
1261 content->array.level <= 6) {
1262 /* might need to increase the size
1263 * of the stripe cache - default is 256
1264 */
1265 if (256 < 4 * (content->array.chunk_size/4096)) {
1266 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1267 if (sra)
1268 sysfs_set_num(sra, NULL,
1269 "stripe_cache_size",
1270 (4 * content->array.chunk_size / 4096) + 1);
1271 }
1272 }
1273 if (okcnt < (unsigned)content->array.raid_disks) {
1274 /* If any devices did not get added
1275 * because the kernel rejected them based
1276 * on event count, try adding them
1277 * again providing the action policy is
1278 * 're-add' or greater. The bitmap
1279 * might allow them to be included, or
1280 * they will become spares.
1281 */
1282 for (i = 0; i <= bestcnt; i++) {
1283 int j = best[i];
1284 if (j >= 0 && !devices[j].uptodate) {
1285 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1286 continue;
1287 rv = add_disk(mdfd, st, content,
1288 &devices[j].i);
1289 if (rv == 0 && verbose >= 0)
1290 fprintf(stderr,
1291 Name ": %s has been re-added.\n",
1292 devices[j].devname);
1293 }
1294 }
1295 }
1296 wait_for(mddev, mdfd);
1297 close(mdfd);
1298 if (auto_assem) {
1299 int usecs = 1;
1300 /* There is a nasty race with 'mdadm --monitor'.
1301 * If it opens this device before we close it,
1302 * it gets an incomplete open on which IO
1303 * doesn't work and the capacity is
1304 * wrong.
1305 * If we reopen (to check for layered devices)
1306 * before --monitor closes, we loose.
1307 *
1308 * So: wait upto 1 second for there to be
1309 * a non-zero capacity.
1310 */
1311 while (usecs < 1000) {
1312 mdfd = open(mddev, O_RDONLY);
1313 if (mdfd >= 0) {
1314 unsigned long long size;
1315 if (get_dev_size(mdfd, NULL, &size) &&
1316 size > 0)
1317 break;
1318 close(mdfd);
1319 }
1320 usleep(usecs);
1321 usecs <<= 1;
1322 }
1323 }
1324 free(devices);
1325 return 0;
1326 }
1327 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1328 mddev, strerror(errno));
1329
1330 if (!enough(content->array.level, content->array.raid_disks,
1331 content->array.layout, 1, avail, okcnt))
1332 fprintf(stderr, Name ": Not enough devices to "
1333 "start the array.\n");
1334 else if (!enough(content->array.level,
1335 content->array.raid_disks,
1336 content->array.layout, clean,
1337 avail, okcnt))
1338 fprintf(stderr, Name ": Not enough devices to "
1339 "start the array while not clean "
1340 "- consider --force.\n");
1341
1342 if (auto_assem)
1343 ioctl(mdfd, STOP_ARRAY, NULL);
1344 close(mdfd);
1345 free(devices);
1346 return 1;
1347 }
1348 if (runstop == -1) {
1349 fprintf(stderr, Name ": %s assembled from %d drive%s",
1350 mddev, okcnt, okcnt==1?"":"s");
1351 if (okcnt != (unsigned)content->array.raid_disks)
1352 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1353 fprintf(stderr, ", but not started.\n");
1354 close(mdfd);
1355 free(devices);
1356 return 0;
1357 }
1358 if (verbose >= -1) {
1359 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1360 if (rebuilding_cnt)
1361 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
1362 if (sparecnt)
1363 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1364 if (!enough(content->array.level, content->array.raid_disks,
1365 content->array.layout, 1, avail, okcnt))
1366 fprintf(stderr, " - not enough to start the array.\n");
1367 else if (!enough(content->array.level,
1368 content->array.raid_disks,
1369 content->array.layout, clean,
1370 avail, okcnt))
1371 fprintf(stderr, " - not enough to start the "
1372 "array while not clean - consider "
1373 "--force.\n");
1374 else {
1375 if (req_cnt == (unsigned)content->array.raid_disks)
1376 fprintf(stderr, " - need all %d to start it", req_cnt);
1377 else
1378 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1379 fprintf(stderr, " (use --run to insist).\n");
1380 }
1381 }
1382 if (auto_assem)
1383 ioctl(mdfd, STOP_ARRAY, NULL);
1384 close(mdfd);
1385 free(devices);
1386 return 1;
1387 } else {
1388 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1389 * been updated to point to the current locations of devices.
1390 * so we can just start the array
1391 */
1392 unsigned long dev;
1393 dev = makedev(devices[chosen_drive].i.disk.major,
1394 devices[chosen_drive].i.disk.minor);
1395 if (ioctl(mdfd, START_ARRAY, dev)) {
1396 fprintf(stderr, Name ": Cannot start array: %s\n",
1397 strerror(errno));
1398 }
1399
1400 }
1401 close(mdfd);
1402 free(devices);
1403 return 0;
1404 }
1405
1406 #ifndef MDASSEMBLE
1407 int assemble_container_content(struct supertype *st, int mdfd,
1408 struct mdinfo *content, int runstop,
1409 char *chosen_name, int verbose)
1410 {
1411 struct mdinfo *dev, *sra;
1412 int working = 0, preexist = 0;
1413 struct map_ent *map = NULL;
1414
1415 sysfs_init(content, mdfd, 0);
1416
1417 sra = sysfs_read(mdfd, 0, GET_VERSION);
1418 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1419 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1420 close(mdfd);
1421 return 1;
1422 }
1423 if (sra)
1424 sysfs_free(sra);
1425
1426 for (dev = content->devs; dev; dev = dev->next)
1427 if (sysfs_add_disk(content, dev, 1) == 0)
1428 working++;
1429 else if (errno == EEXIST)
1430 preexist++;
1431 if (working == 0) {
1432 close(mdfd);
1433 return 1;/* Nothing new, don't try to start */
1434 }
1435
1436 map_update(&map, fd2devnum(mdfd),
1437 content->text_version,
1438 content->uuid, chosen_name);
1439
1440 if (runstop > 0 ||
1441 (working + preexist) >= content->array.working_disks) {
1442 int err;
1443
1444 switch(content->array.level) {
1445 case LEVEL_LINEAR:
1446 case LEVEL_MULTIPATH:
1447 case 0:
1448 err = sysfs_set_str(content, NULL, "array_state",
1449 "active");
1450 break;
1451 default:
1452 err = sysfs_set_str(content, NULL, "array_state",
1453 "readonly");
1454 /* start mdmon if needed. */
1455 if (!err) {
1456 if (!mdmon_running(st->container_dev))
1457 start_mdmon(st->container_dev);
1458 ping_monitor(devnum2devname(st->container_dev));
1459 }
1460 break;
1461 }
1462 if (!err)
1463 sysfs_set_safemode(content, content->safe_mode_delay);
1464 if (verbose >= 0) {
1465 if (err)
1466 fprintf(stderr, Name
1467 ": array %s now has %d devices",
1468 chosen_name, working + preexist);
1469 else
1470 fprintf(stderr, Name
1471 ": Started %s with %d devices",
1472 chosen_name, working + preexist);
1473 if (preexist)
1474 fprintf(stderr, " (%d new)", working);
1475 fprintf(stderr, "\n");
1476 }
1477 if (!err)
1478 wait_for(chosen_name, mdfd);
1479 close(mdfd);
1480 return 0;
1481 /* FIXME should have an O_EXCL and wait for read-auto */
1482 } else {
1483 if (verbose >= 0)
1484 fprintf(stderr, Name
1485 ": %s assembled with %d devices but "
1486 "not started\n",
1487 chosen_name, working);
1488 close(mdfd);
1489 return 1;
1490 }
1491 }
1492 #endif
1493