]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Improve partition table code.
[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;
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 (auto_assem && st == NULL &&
264 !conf_test_metadata(tst->ss->name,
265 tst->ss->match_home(tst, homehost) == 1)) {
266 if (report_missmatch)
267 fprintf(stderr, Name ": %s has metadata type %s for which "
268 "auto-assembly is disabled\n",
269 devname, tst->ss->name);
270 tmpdev->used = 2;
271 } else if (tst->ss->load_super(tst,dfd, NULL)) {
272 if (report_missmatch)
273 fprintf( stderr, Name ": no RAID superblock on %s\n",
274 devname);
275 } else {
276 content = &info;
277 memset(content, 0, sizeof(*content));
278 tst->ss->getinfo_super(tst, content);
279 }
280 if (dfd >= 0) close(dfd);
281
282 if (tst && tst->sb && tst->ss->container_content
283 && tst->loaded_container) {
284 /* tmpdev is a container. We need to be either
285 * looking for a member, or auto-assembling
286 */
287 if (st) {
288 /* already found some components, this cannot
289 * be another one.
290 */
291 if (report_missmatch)
292 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
293 devname);
294 goto loop;
295 }
296
297 if (ident->container) {
298 if (ident->container[0] == '/' &&
299 !same_dev(ident->container, devname)) {
300 if (report_missmatch)
301 fprintf(stderr, Name ": %s is not the container required (%s)\n",
302 devname, ident->container);
303 goto loop;
304 }
305 if (ident->container[0] != '/') {
306 /* we have a uuid */
307 int uuid[4];
308 if (!parse_uuid(ident->container, uuid) ||
309 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
310 if (report_missmatch)
311 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
312 devname);
313 goto loop;
314 }
315 }
316 }
317 /* It is worth looking inside this container.
318 */
319 if (verbose > 0)
320 fprintf(stderr, Name ": looking in container %s\n",
321 devname);
322 next_member:
323 if (tmpdev->content)
324 content = tmpdev->content;
325 else
326 content = tst->ss->container_content(tst);
327 if (!content)
328 goto loop; /* empty container */
329
330 tmpdev->content = content->next;
331 if (tmpdev->content == NULL)
332 tmpdev->used = 2;
333
334 } else if (ident->container || ident->member) {
335 /* No chance of this matching if we don't have
336 * a container */
337 if (report_missmatch)
338 fprintf(stderr, Name "%s is not a container, and one is required.\n",
339 devname);
340 goto loop;
341 }
342
343 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
344 (!tst || !tst->sb ||
345 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
346 if (report_missmatch)
347 fprintf(stderr, Name ": %s has wrong uuid.\n",
348 devname);
349 goto loop;
350 }
351 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
352 (!tst || !tst->sb ||
353 name_matches(content->name, ident->name, homehost)==0)) {
354 if (report_missmatch)
355 fprintf(stderr, Name ": %s has wrong name.\n",
356 devname);
357 goto loop;
358 }
359 if (ident->super_minor != UnSet &&
360 (!tst || !tst->sb ||
361 ident->super_minor != content->array.md_minor)) {
362 if (report_missmatch)
363 fprintf(stderr, Name ": %s has wrong super-minor.\n",
364 devname);
365 goto loop;
366 }
367 if (ident->level != UnSet &&
368 (!tst || !tst->sb ||
369 ident->level != content->array.level)) {
370 if (report_missmatch)
371 fprintf(stderr, Name ": %s has wrong raid level.\n",
372 devname);
373 goto loop;
374 }
375 if (ident->raid_disks != UnSet &&
376 (!tst || !tst->sb ||
377 ident->raid_disks!= content->array.raid_disks)) {
378 if (report_missmatch)
379 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
380 devname);
381 goto loop;
382 }
383 if (auto_assem) {
384 if (tst == NULL || tst->sb == NULL)
385 continue;
386 }
387 /* If we are this far, then we are nearly commited to this device.
388 * If the super_block doesn't exist, or doesn't match others,
389 * then we probably cannot continue
390 * However if one of the arrays is for the homehost, and
391 * the other isn't that can disambiguate.
392 */
393
394 if (!tst || !tst->sb) {
395 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
396 devname);
397 if (st)
398 st->ss->free_super(st);
399 return 1;
400 }
401
402 if (tst && tst->sb && tst->ss->container_content
403 && tst->loaded_container) {
404 /* we have the one container we need, don't keep
405 * looking. If the chosen member is active, skip.
406 */
407 if (is_member_busy(content->text_version)) {
408 if (report_missmatch)
409 fprintf(stderr, Name ": member %s in %s is already assembled\n",
410 content->text_version,
411 devname);
412 skip:
413 if (tmpdev->content)
414 goto next_member;
415 tst->ss->free_super(tst);
416 tst = NULL;
417 content = NULL;
418 if (auto_assem)
419 goto loop;
420 return 1;
421 }
422 if (ident->member && ident->member[0]) {
423 char *s = strchr(content->text_version+1, '/');
424 if (s == NULL) {
425 fprintf(stderr, Name ": badly formatted version: %s\n",
426 content->text_version);
427 goto skip;
428 }
429 if (strcmp(ident->member, s+1) != 0) {
430 if (report_missmatch)
431 fprintf(stderr,
432 Name ": skipping wrong member %s\n",
433 content->text_version);
434 goto skip;
435 }
436 }
437 st = tst; tst = NULL;
438 if (!auto_assem && inargv && tmpdev->next != NULL) {
439 fprintf(stderr, Name ": %s is a container, but is not "
440 "only device given: confused and aborting\n",
441 devname);
442 st->ss->free_super(st);
443 return 1;
444 }
445 if (verbose > 0)
446 fprintf(stderr, Name ": found match on member %s in %s\n",
447 content->text_version, devname);
448 break;
449 }
450 if (st == NULL)
451 st = dup_super(tst);
452 if (st->minor_version == -1)
453 st->minor_version = tst->minor_version;
454 if (st->ss != tst->ss ||
455 st->minor_version != tst->minor_version ||
456 st->ss->compare_super(st, tst) != 0) {
457 /* Some mismatch. If exactly one array matches this host,
458 * we can resolve on that one.
459 * Or, if we are auto assembling, we just ignore the second
460 * for now.
461 */
462 if (auto_assem)
463 goto loop;
464 if (homehost) {
465 int first = st->ss->match_home(st, homehost);
466 int last = tst->ss->match_home(tst, homehost);
467 if (first != last &&
468 (first == 1 || last == 1)) {
469 /* We can do something */
470 if (first) {/* just ignore this one */
471 if (report_missmatch)
472 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
473 devname);
474 goto loop;
475 } else { /* reject all those sofar */
476 mddev_dev_t td;
477 if (report_missmatch)
478 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
479 devname);
480 for (td=devlist; td != tmpdev; td=td->next)
481 if (td->used == 1)
482 td->used = 0;
483 tmpdev->used = 1;
484 goto loop;
485 }
486 }
487 }
488 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
489 devname);
490 tst->ss->free_super(tst);
491 st->ss->free_super(st);
492 return 1;
493 }
494
495 tmpdev->used = 1;
496
497 loop:
498 if (tmpdev->content)
499 goto next_member;
500 if (tst)
501 tst->ss->free_super(tst);
502 }
503
504 if (!st || !st->sb || !content)
505 return 2;
506
507 /* Now need to open the array device. Use create_mddev */
508 if (content == &info)
509 st->ss->getinfo_super(st, content);
510
511 trustworthy = FOREIGN;
512 name = content->name;
513 switch (st->ss->match_home(st, homehost)
514 ?: st->ss->match_home(st, "any")) {
515 case 1:
516 trustworthy = LOCAL;
517 name = strchr(content->name, ':');
518 if (name)
519 name++;
520 else
521 name = content->name;
522 break;
523 }
524 if (!auto_assem)
525 /* If the array is listed in mdadm.conf or on
526 * command line, then we trust the name
527 * even if the array doesn't look local
528 */
529 trustworthy = LOCAL;
530
531 if (name[0] == 0 &&
532 content->array.level == LEVEL_CONTAINER) {
533 name = content->text_version;
534 trustworthy = METADATA;
535 }
536
537 if (name[0] && trustworthy != LOCAL &&
538 ! require_homehost &&
539 conf_name_is_free(name))
540 trustworthy = LOCAL;
541
542 if (trustworthy == LOCAL &&
543 strchr(name, ':'))
544 /* Ignore 'host:' prefix of name */
545 name = strchr(name, ':')+1;
546
547 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
548 chosen_name);
549 if (mdfd < 0) {
550 st->ss->free_super(st);
551 free(devices);
552 if (auto_assem)
553 goto try_again;
554 return 1;
555 }
556 mddev = chosen_name;
557 vers = md_get_version(mdfd);
558 if (vers < 9000) {
559 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
560 " Upgrade your kernel or try --build\n");
561 close(mdfd);
562 return 1;
563 }
564 if (mddev_busy(fd2devnum(mdfd))) {
565 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
566 mddev);
567 for (tmpdev = devlist ;
568 tmpdev && tmpdev->used != 1;
569 tmpdev = tmpdev->next)
570 ;
571 if (tmpdev && auto_assem)
572 fprintf(stderr, Name ": %s needed for %s...\n",
573 mddev, tmpdev->devname);
574 close(mdfd);
575 mdfd = -3;
576 st->ss->free_super(st);
577 free(devices);
578 if (auto_assem)
579 goto try_again;
580 return 1;
581 }
582 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
583
584 #ifndef MDASSEMBLE
585 if (content != &info) {
586 /* This is a member of a container. Try starting the array. */
587 return assemble_container_content(st, mdfd, content, runstop,
588 chosen_name, verbose);
589 }
590 #endif
591 /* Ok, no bad inconsistancy, we can try updating etc */
592 bitmap_done = 0;
593 content->update_private = NULL;
594 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
595 char *devname = tmpdev->devname;
596 struct stat stb;
597 /* looks like a good enough match to update the super block if needed */
598 #ifndef MDASSEMBLE
599 if (update) {
600 int dfd;
601 /* prepare useful information in info structures */
602 struct stat stb2;
603 struct supertype *tst;
604 fstat(mdfd, &stb2);
605
606 if (strcmp(update, "uuid")==0 &&
607 !ident->uuid_set) {
608 int rfd;
609 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
610 read(rfd, ident->uuid, 16) != 16) {
611 *(__u32*)(ident->uuid) = random();
612 *(__u32*)(ident->uuid+1) = random();
613 *(__u32*)(ident->uuid+2) = random();
614 *(__u32*)(ident->uuid+3) = random();
615 }
616 if (rfd >= 0) close(rfd);
617 }
618 dfd = dev_open(devname, O_RDWR|O_EXCL);
619
620 remove_partitions(dfd);
621
622 tst = dup_super(st);
623 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
624 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
625 devname);
626 if (dfd >= 0)
627 close(dfd);
628 close(mdfd);
629 return 1;
630 }
631 tst->ss->getinfo_super(tst, content);
632
633 memcpy(content->uuid, ident->uuid, 16);
634 strcpy(content->name, ident->name);
635 content->array.md_minor = minor(stb2.st_rdev);
636
637 tst->ss->update_super(tst, content, update,
638 devname, verbose,
639 ident->uuid_set, homehost);
640 if (strcmp(update, "uuid")==0 &&
641 !ident->uuid_set) {
642 ident->uuid_set = 1;
643 memcpy(ident->uuid, content->uuid, 16);
644 }
645 if (dfd < 0)
646 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
647 devname);
648 else if (tst->ss->store_super(tst, dfd))
649 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
650 devname);
651 if (dfd >= 0)
652 close(dfd);
653
654 if (strcmp(update, "uuid")==0 &&
655 ident->bitmap_fd >= 0 && !bitmap_done) {
656 if (bitmap_update_uuid(ident->bitmap_fd,
657 content->uuid,
658 tst->ss->swapuuid) != 0)
659 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
660 else
661 bitmap_done = 1;
662 }
663 tst->ss->free_super(tst);
664 } else
665 #endif
666 {
667 struct supertype *tst = dup_super(st);
668 int dfd;
669 dfd = dev_open(devname, O_RDWR|O_EXCL);
670
671 remove_partitions(dfd);
672
673 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
674 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
675 devname);
676 if (dfd >= 0)
677 close(dfd);
678 close(mdfd);
679 return 1;
680 }
681 tst->ss->getinfo_super(tst, content);
682 tst->ss->free_super(tst);
683 close(dfd);
684 }
685
686 stat(devname, &stb);
687
688 if (verbose > 0)
689 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
690 devname, mddev, content->disk.raid_disk);
691 devices[devcnt].devname = devname;
692 devices[devcnt].uptodate = 0;
693 devices[devcnt].i = *content;
694 devices[devcnt].i.disk.major = major(stb.st_rdev);
695 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
696 if (most_recent < devcnt) {
697 if (devices[devcnt].i.events
698 > devices[most_recent].i.events)
699 most_recent = devcnt;
700 }
701 if (content->array.level == LEVEL_MULTIPATH)
702 /* with multipath, the raid_disk from the superblock is meaningless */
703 i = devcnt;
704 else
705 i = devices[devcnt].i.disk.raid_disk;
706 if (i+1 == 0) {
707 if (nextspare < content->array.raid_disks)
708 nextspare = content->array.raid_disks;
709 i = nextspare++;
710 } else {
711 if (i >= content->array.raid_disks &&
712 i >= nextspare)
713 nextspare = i+1;
714 }
715 if (i < 10000) {
716 if (i >= bestcnt) {
717 unsigned int newbestcnt = i+10;
718 int *newbest = malloc(sizeof(int)*newbestcnt);
719 unsigned int c;
720 for (c=0; c < newbestcnt; c++)
721 if (c < bestcnt)
722 newbest[c] = best[c];
723 else
724 newbest[c] = -1;
725 if (best)free(best);
726 best = newbest;
727 bestcnt = newbestcnt;
728 }
729 if (best[i] >=0 &&
730 devices[best[i]].i.events
731 == devices[devcnt].i.events
732 && (devices[best[i]].i.disk.minor
733 != devices[devcnt].i.disk.minor)
734 && st->ss == &super0
735 && content->array.level != LEVEL_MULTIPATH) {
736 /* two different devices with identical superblock.
737 * Could be a mis-detection caused by overlapping
738 * partitions. fail-safe.
739 */
740 fprintf(stderr, Name ": WARNING %s and %s appear"
741 " to have very similar superblocks.\n"
742 " If they are really different, "
743 "please --zero the superblock on one\n"
744 " If they are the same or overlap,"
745 " please remove one from %s.\n",
746 devices[best[i]].devname, devname,
747 inargv ? "the list" :
748 "the\n DEVICE list in mdadm.conf"
749 );
750 close(mdfd);
751 return 1;
752 }
753 if (best[i] == -1
754 || (devices[best[i]].i.events
755 < devices[devcnt].i.events))
756 best[i] = devcnt;
757 }
758 devcnt++;
759 }
760 free(content->update_private);
761 content->update_private = NULL;
762
763 if (devcnt == 0) {
764 fprintf(stderr, Name ": no devices found for %s\n",
765 mddev);
766 if (st)
767 st->ss->free_super(st);
768 close(mdfd);
769 return 1;
770 }
771
772 if (update && strcmp(update, "byteorder")==0)
773 st->minor_version = 90;
774
775 st->ss->getinfo_super(st, content);
776 clean = content->array.state & 1;
777
778 /* now we have some devices that might be suitable.
779 * I wonder how many
780 */
781 avail = malloc(content->array.raid_disks);
782 memset(avail, 0, content->array.raid_disks);
783 okcnt = 0;
784 sparecnt=0;
785 for (i=0; i< bestcnt ;i++) {
786 int j = best[i];
787 int event_margin = 1; /* always allow a difference of '1'
788 * like the kernel does
789 */
790 if (j < 0) continue;
791 /* note: we ignore error flags in multipath arrays
792 * as they don't make sense
793 */
794 if (content->array.level != LEVEL_MULTIPATH)
795 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
796 if (!(devices[j].i.disk.state
797 & (1<<MD_DISK_FAULTY)))
798 sparecnt++;
799 continue;
800 }
801 if (devices[j].i.events+event_margin >=
802 devices[most_recent].i.events) {
803 devices[j].uptodate = 1;
804 if (i < content->array.raid_disks &&
805 devices[j].i.recovery_start == MaxSector) {
806 okcnt++;
807 avail[i]=1;
808 } else
809 sparecnt++;
810 }
811 }
812 while (force && !enough(content->array.level, content->array.raid_disks,
813 content->array.layout, 1,
814 avail, okcnt)) {
815 /* Choose the newest best drive which is
816 * not up-to-date, update the superblock
817 * and add it.
818 */
819 int fd;
820 struct supertype *tst;
821 long long current_events;
822 chosen_drive = -1;
823 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
824 int j = best[i];
825 if (j>=0 &&
826 !devices[j].uptodate &&
827 devices[j].i.recovery_start == MaxSector &&
828 (chosen_drive < 0 ||
829 devices[j].i.events
830 > devices[chosen_drive].i.events))
831 chosen_drive = j;
832 }
833 if (chosen_drive < 0)
834 break;
835 current_events = devices[chosen_drive].i.events;
836 add_another:
837 if (verbose >= 0)
838 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
839 devices[chosen_drive].devname,
840 devices[chosen_drive].i.disk.raid_disk,
841 (int)(devices[chosen_drive].i.events),
842 (int)(devices[most_recent].i.events));
843 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
844 if (fd < 0) {
845 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
846 devices[chosen_drive].devname);
847 devices[chosen_drive].i.events = 0;
848 continue;
849 }
850 tst = dup_super(st);
851 if (tst->ss->load_super(tst,fd, NULL)) {
852 close(fd);
853 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
854 devices[chosen_drive].devname);
855 devices[chosen_drive].i.events = 0;
856 continue;
857 }
858 content->events = devices[most_recent].i.events;
859 tst->ss->update_super(tst, content, "force-one",
860 devices[chosen_drive].devname, verbose,
861 0, NULL);
862
863 if (tst->ss->store_super(tst, fd)) {
864 close(fd);
865 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
866 devices[chosen_drive].devname);
867 devices[chosen_drive].i.events = 0;
868 tst->ss->free_super(tst);
869 continue;
870 }
871 close(fd);
872 devices[chosen_drive].i.events = devices[most_recent].i.events;
873 devices[chosen_drive].uptodate = 1;
874 avail[chosen_drive] = 1;
875 okcnt++;
876 tst->ss->free_super(tst);
877
878 /* If there are any other drives of the same vintage,
879 * add them in as well. We can't lose and we might gain
880 */
881 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
882 int j = best[i];
883 if (j >= 0 &&
884 !devices[j].uptodate &&
885 devices[j].i.events == current_events) {
886 chosen_drive = j;
887 goto add_another;
888 }
889 }
890 }
891
892 /* Now we want to look at the superblock which the kernel will base things on
893 * and compare the devices that we think are working with the devices that the
894 * superblock thinks are working.
895 * If there are differences and --force is given, then update this chosen
896 * superblock.
897 */
898 chosen_drive = -1;
899 st->ss->free_super(st);
900 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
901 int j = best[i];
902 int fd;
903
904 if (j<0)
905 continue;
906 if (!devices[j].uptodate)
907 continue;
908 chosen_drive = j;
909 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
910 fprintf(stderr, Name ": Cannot open %s: %s\n",
911 devices[j].devname, strerror(errno));
912 close(mdfd);
913 return 1;
914 }
915 if (st->ss->load_super(st,fd, NULL)) {
916 close(fd);
917 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
918 devices[j].devname);
919 close(mdfd);
920 return 1;
921 }
922 close(fd);
923 }
924 if (st->sb == NULL) {
925 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
926 close(mdfd);
927 return 1;
928 }
929 st->ss->getinfo_super(st, content);
930 #ifndef MDASSEMBLE
931 sysfs_init(content, mdfd, 0);
932 #endif
933 for (i=0; i<bestcnt; i++) {
934 int j = best[i];
935 unsigned int desired_state;
936
937 if (i < content->array.raid_disks)
938 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
939 else
940 desired_state = 0;
941
942 if (j<0)
943 continue;
944 if (!devices[j].uptodate)
945 continue;
946
947 devices[j].i.disk.state = desired_state;
948 if (!(devices[j].i.array.state & 1))
949 clean = 0;
950
951 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
952 verbose, 0, NULL)) {
953 if (force) {
954 if (verbose >= 0)
955 fprintf(stderr, Name ": "
956 "clearing FAULTY flag for device %d in %s for %s\n",
957 j, mddev, devices[j].devname);
958 change = 1;
959 } else {
960 if (verbose >= -1)
961 fprintf(stderr, Name ": "
962 "device %d in %s has wrong state in superblock, but %s seems ok\n",
963 i, mddev, devices[j].devname);
964 }
965 }
966 #if 0
967 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
968 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
969 i, mddev);
970 }
971 #endif
972 }
973 if (force && !clean &&
974 !enough(content->array.level, content->array.raid_disks,
975 content->array.layout, clean,
976 avail, okcnt)) {
977 change += st->ss->update_super(st, content, "force-array",
978 devices[chosen_drive].devname, verbose,
979 0, NULL);
980 clean = 1;
981 }
982
983 if (change) {
984 int fd;
985 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
986 if (fd < 0) {
987 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
988 devices[chosen_drive].devname);
989 close(mdfd);
990 return 1;
991 }
992 if (st->ss->store_super(st, fd)) {
993 close(fd);
994 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
995 devices[chosen_drive].devname);
996 close(mdfd);
997 return 1;
998 }
999 close(fd);
1000 }
1001
1002 /* If we are in the middle of a reshape we may need to restore saved data
1003 * that was moved aside due to the reshape overwriting live data
1004 * The code of doing this lives in Grow.c
1005 */
1006 #ifndef MDASSEMBLE
1007 if (content->reshape_active) {
1008 int err = 0;
1009 int *fdlist = malloc(sizeof(int)* bestcnt);
1010 if (verbose > 0)
1011 fprintf(stderr, Name ":%s has an active reshape - checking "
1012 "if critical section needs to be restored\n",
1013 chosen_name);
1014 for (i=0; i<bestcnt; i++) {
1015 int j = best[i];
1016 if (j >= 0) {
1017 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1018 if (fdlist[i] < 0) {
1019 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1020 devices[j].devname);
1021 err = 1;
1022 break;
1023 }
1024 } else
1025 fdlist[i] = -1;
1026 }
1027 if (!err)
1028 err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0);
1029 while (i>0) {
1030 i--;
1031 if (fdlist[i]>=0) close(fdlist[i]);
1032 }
1033 if (err) {
1034 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1035 if (backup_file == NULL)
1036 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1037 close(mdfd);
1038 return err;
1039 }
1040 }
1041 #endif
1042 /* count number of in-sync devices according to the superblock.
1043 * We must have this number to start the array without -s or -R
1044 */
1045 req_cnt = content->array.working_disks;
1046
1047 /* Almost ready to actually *do* something */
1048 if (!old_linux) {
1049 int rv;
1050
1051 /* First, fill in the map, so that udev can find our name
1052 * as soon as we become active.
1053 */
1054 map_update(NULL, fd2devnum(mdfd), content->text_version,
1055 content->uuid, chosen_name);
1056
1057 rv = set_array_info(mdfd, st, content);
1058 if (rv) {
1059 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1060 mddev, strerror(errno));
1061 ioctl(mdfd, STOP_ARRAY, NULL);
1062 close(mdfd);
1063 return 1;
1064 }
1065 if (ident->bitmap_fd >= 0) {
1066 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1067 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1068 ioctl(mdfd, STOP_ARRAY, NULL);
1069 close(mdfd);
1070 return 1;
1071 }
1072 } else if (ident->bitmap_file) {
1073 /* From config file */
1074 int bmfd = open(ident->bitmap_file, O_RDWR);
1075 if (bmfd < 0) {
1076 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1077 ident->bitmap_file);
1078 ioctl(mdfd, STOP_ARRAY, NULL);
1079 close(mdfd);
1080 return 1;
1081 }
1082 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1083 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1084 close(bmfd);
1085 ioctl(mdfd, STOP_ARRAY, NULL);
1086 close(mdfd);
1087 return 1;
1088 }
1089 close(bmfd);
1090 }
1091
1092 /* First, add the raid disks, but add the chosen one last */
1093 for (i=0; i<= bestcnt; i++) {
1094 int j;
1095 if (i < bestcnt) {
1096 j = best[i];
1097 if (j == chosen_drive)
1098 continue;
1099 } else
1100 j = chosen_drive;
1101
1102 if (j >= 0 /* && devices[j].uptodate */) {
1103 rv = add_disk(mdfd, st, content, &devices[j].i);
1104
1105 if (rv) {
1106 fprintf(stderr, Name ": failed to add "
1107 "%s to %s: %s\n",
1108 devices[j].devname,
1109 mddev,
1110 strerror(errno));
1111 if (i < content->array.raid_disks
1112 || i == bestcnt)
1113 okcnt--;
1114 else
1115 sparecnt--;
1116 } else if (verbose > 0)
1117 fprintf(stderr, Name ": added %s "
1118 "to %s as %d\n",
1119 devices[j].devname, mddev,
1120 devices[j].i.disk.raid_disk);
1121 } else if (verbose > 0 && i < content->array.raid_disks)
1122 fprintf(stderr, Name ": no uptodate device for "
1123 "slot %d of %s\n",
1124 i, mddev);
1125 }
1126
1127 if (content->array.level == LEVEL_CONTAINER) {
1128 if (verbose >= 0) {
1129 fprintf(stderr, Name ": Container %s has been "
1130 "assembled with %d drive%s",
1131 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1132 if (okcnt < content->array.raid_disks)
1133 fprintf(stderr, " (out of %d)",
1134 content->array.raid_disks);
1135 fprintf(stderr, "\n");
1136 }
1137 sysfs_uevent(content, "change");
1138 wait_for(chosen_name, mdfd);
1139 close(mdfd);
1140 return 0;
1141 }
1142
1143 if (runstop == 1 ||
1144 (runstop <= 0 &&
1145 ( enough(content->array.level, content->array.raid_disks,
1146 content->array.layout, clean, avail, okcnt) &&
1147 (okcnt >= req_cnt || start_partial_ok)
1148 ))) {
1149 /* This array is good-to-go.
1150 * If a reshape is in progress then we might need to
1151 * continue monitoring it. In that case we start
1152 * it read-only and let the grow code make it writable.
1153 */
1154 int rv;
1155 #ifndef MDASSEMBLE
1156 if (content->reshape_active &&
1157 content->delta_disks <= 0)
1158 rv = Grow_continue(mdfd, st, content, backup_file);
1159 else
1160 #endif
1161 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1162 if (rv == 0) {
1163 if (verbose >= 0) {
1164 fprintf(stderr, Name ": %s has been started with %d drive%s",
1165 mddev, okcnt, okcnt==1?"":"s");
1166 if (okcnt < content->array.raid_disks)
1167 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1168 if (sparecnt)
1169 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1170 fprintf(stderr, ".\n");
1171 }
1172 if (content->reshape_active &&
1173 content->array.level >= 4 &&
1174 content->array.level <= 6) {
1175 /* might need to increase the size
1176 * of the stripe cache - default is 256
1177 */
1178 if (256 < 4 * (content->array.chunk_size/4096)) {
1179 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1180 if (sra)
1181 sysfs_set_num(sra, NULL,
1182 "stripe_cache_size",
1183 (4 * content->array.chunk_size / 4096) + 1);
1184 }
1185 }
1186 wait_for(mddev, mdfd);
1187 close(mdfd);
1188 if (auto_assem) {
1189 int usecs = 1;
1190 /* There is a nasty race with 'mdadm --monitor'.
1191 * If it opens this device before we close it,
1192 * it gets an incomplete open on which IO
1193 * doesn't work and the capacity is
1194 * wrong.
1195 * If we reopen (to check for layered devices)
1196 * before --monitor closes, we loose.
1197 *
1198 * So: wait upto 1 second for there to be
1199 * a non-zero capacity.
1200 */
1201 while (usecs < 1000) {
1202 mdfd = open(mddev, O_RDONLY);
1203 if (mdfd >= 0) {
1204 unsigned long long size;
1205 if (get_dev_size(mdfd, NULL, &size) &&
1206 size > 0)
1207 break;
1208 close(mdfd);
1209 }
1210 usleep(usecs);
1211 usecs <<= 1;
1212 }
1213 }
1214 return 0;
1215 }
1216 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1217 mddev, strerror(errno));
1218
1219 if (!enough(content->array.level, content->array.raid_disks,
1220 content->array.layout, 1, avail, okcnt))
1221 fprintf(stderr, Name ": Not enough devices to "
1222 "start the array.\n");
1223 else if (!enough(content->array.level,
1224 content->array.raid_disks,
1225 content->array.layout, clean,
1226 avail, okcnt))
1227 fprintf(stderr, Name ": Not enough devices to "
1228 "start the array while not clean "
1229 "- consider --force.\n");
1230
1231 if (auto_assem)
1232 ioctl(mdfd, STOP_ARRAY, NULL);
1233 close(mdfd);
1234 return 1;
1235 }
1236 if (runstop == -1) {
1237 fprintf(stderr, Name ": %s assembled from %d drive%s",
1238 mddev, okcnt, okcnt==1?"":"s");
1239 if (okcnt != content->array.raid_disks)
1240 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1241 fprintf(stderr, ", but not started.\n");
1242 close(mdfd);
1243 return 0;
1244 }
1245 if (verbose >= -1) {
1246 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1247 if (sparecnt)
1248 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1249 if (!enough(content->array.level, content->array.raid_disks,
1250 content->array.layout, 1, avail, okcnt))
1251 fprintf(stderr, " - not enough to start the array.\n");
1252 else if (!enough(content->array.level,
1253 content->array.raid_disks,
1254 content->array.layout, clean,
1255 avail, okcnt))
1256 fprintf(stderr, " - not enough to start the "
1257 "array while not clean - consider "
1258 "--force.\n");
1259 else {
1260 if (req_cnt == content->array.raid_disks)
1261 fprintf(stderr, " - need all %d to start it", req_cnt);
1262 else
1263 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1264 fprintf(stderr, " (use --run to insist).\n");
1265 }
1266 }
1267 if (auto_assem)
1268 ioctl(mdfd, STOP_ARRAY, NULL);
1269 close(mdfd);
1270 return 1;
1271 } else {
1272 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1273 * been updated to point to the current locations of devices.
1274 * so we can just start the array
1275 */
1276 unsigned long dev;
1277 dev = makedev(devices[chosen_drive].i.disk.major,
1278 devices[chosen_drive].i.disk.minor);
1279 if (ioctl(mdfd, START_ARRAY, dev)) {
1280 fprintf(stderr, Name ": Cannot start array: %s\n",
1281 strerror(errno));
1282 }
1283
1284 }
1285 close(mdfd);
1286 return 0;
1287 }
1288
1289 #ifndef MDASSEMBLE
1290 int assemble_container_content(struct supertype *st, int mdfd,
1291 struct mdinfo *content, int runstop,
1292 char *chosen_name, int verbose)
1293 {
1294 struct mdinfo *dev, *sra;
1295 int working = 0, preexist = 0;
1296 struct map_ent *map = NULL;
1297
1298 sysfs_init(content, mdfd, 0);
1299
1300 sra = sysfs_read(mdfd, 0, GET_VERSION);
1301 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1302 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1303 close(mdfd);
1304 return 1;
1305 }
1306 if (sra)
1307 sysfs_free(sra);
1308
1309 for (dev = content->devs; dev; dev = dev->next)
1310 if (sysfs_add_disk(content, dev, 1) == 0)
1311 working++;
1312 else if (errno == EEXIST)
1313 preexist++;
1314 if (working == 0) {
1315 close(mdfd);
1316 return 1;/* Nothing new, don't try to start */
1317 }
1318
1319 map_update(&map, fd2devnum(mdfd),
1320 content->text_version,
1321 content->uuid, chosen_name);
1322
1323 if (runstop > 0 ||
1324 (working + preexist) >= content->array.working_disks) {
1325 int err;
1326
1327 switch(content->array.level) {
1328 case LEVEL_LINEAR:
1329 case LEVEL_MULTIPATH:
1330 case 0:
1331 err = sysfs_set_str(content, NULL, "array_state",
1332 "active");
1333 break;
1334 default:
1335 err = sysfs_set_str(content, NULL, "array_state",
1336 "readonly");
1337 /* start mdmon if needed. */
1338 if (!err) {
1339 if (!mdmon_running(st->container_dev))
1340 start_mdmon(st->container_dev);
1341 ping_monitor(devnum2devname(st->container_dev));
1342 }
1343 break;
1344 }
1345 if (!err)
1346 sysfs_set_safemode(content, content->safe_mode_delay);
1347 if (verbose >= 0) {
1348 if (err)
1349 fprintf(stderr, Name
1350 ": array %s now has %d devices",
1351 chosen_name, working + preexist);
1352 else
1353 fprintf(stderr, Name
1354 ": Started %s with %d devices",
1355 chosen_name, working + preexist);
1356 if (preexist)
1357 fprintf(stderr, " (%d new)", working);
1358 fprintf(stderr, "\n");
1359 }
1360 if (!err)
1361 wait_for(chosen_name, mdfd);
1362 close(mdfd);
1363 return 0;
1364 /* FIXME should have an O_EXCL and wait for read-auto */
1365 } else {
1366 if (verbose >= 0)
1367 fprintf(stderr, Name
1368 ": %s assembled with %d devices but "
1369 "not started\n",
1370 chosen_name, working);
1371 close(mdfd);
1372 return 1;
1373 }
1374 }
1375 #endif
1376
Unnamed repository; edit this file 'description' to name the repository.