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