]> 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
main: factor out code to parse layout for raid10 and faulty.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
23 * Paper: Neil Brown
24 * School of Computer Science and Engineering
25 * The University of New South Wales
26 * Sydney, 2052
27 * Australia
28 */
29
30 #include "mdadm.h"
31 #include <ctype.h>
32
33 static int name_matches(char *found, char *required, char *homehost)
34 {
35 /* See if the name found matches the required name, possibly
36 * prefixed with 'homehost'
37 */
38 char fnd[33];
39
40 strncpy(fnd, found, 32);
41 fnd[32] = 0;
42 if (strcmp(found, required)==0)
43 return 1;
44 if (homehost) {
45 int l = strlen(homehost);
46 if (l < 32 && fnd[l] == ':' &&
47 strcmp(fnd+l+1, required)==0)
48 return 1;
49 }
50 return 0;
51 }
52
53 static int is_member_busy(char *metadata_version)
54 {
55 /* check if the given member array is active */
56 struct mdstat_ent *mdstat = mdstat_read(1, 0);
57 struct mdstat_ent *ent;
58 int busy = 0;
59
60 for (ent = mdstat; ent; ent = ent->next) {
61 if (ent->metadata_version == NULL)
62 continue;
63 if (strncmp(ent->metadata_version, "external:", 9) != 0)
64 continue;
65 if (!is_subarray(&ent->metadata_version[9]))
66 continue;
67 /* Skip first char - it can be '/' or '-' */
68 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
69 busy = 1;
70 break;
71 }
72 }
73 free_mdstat(mdstat);
74
75 return busy;
76 }
77
78 int Assemble(struct supertype *st, char *mddev,
79 mddev_ident_t ident,
80 mddev_dev_t devlist, char *backup_file,
81 int readonly, int runstop,
82 char *update, char *homehost, int require_homehost,
83 int verbose, int force)
84 {
85 /*
86 * The task of Assemble is to find a collection of
87 * devices that should (according to their superblocks)
88 * form an array, and to give this collection to the MD driver.
89 * In Linux-2.4 and later, this involves submitting a
90 * SET_ARRAY_INFO ioctl with no arg - to prepare
91 * the array - and then submit a number of
92 * ADD_NEW_DISK ioctls to add disks into
93 * the array. Finally RUN_ARRAY might
94 * be submitted to start the array.
95 *
96 * Much of the work of Assemble is in finding and/or
97 * checking the disks to make sure they look right.
98 *
99 * If mddev is not set, then scan must be set and we
100 * read through the config file for dev+uuid mapping
101 * We recurse, setting mddev, for each device that
102 * - isn't running
103 * - has a valid uuid (or any uuid if !uuidset)
104 *
105 * If mddev is set, we try to determine state of md.
106 * check version - must be at least 0.90.0
107 * check kernel version. must be at least 2.4.
108 * If not, we can possibly fall back on START_ARRAY
109 * Try to GET_ARRAY_INFO.
110 * If possible, give up
111 * If not, try to STOP_ARRAY just to make sure
112 *
113 * If !uuidset and scan, look in conf-file for uuid
114 * If not found, give up
115 * If !devlist and scan and uuidset, get list of devs from conf-file
116 *
117 * For each device:
118 * Check superblock - discard if bad
119 * Check uuid (set if we don't have one) - discard if no match
120 * Check superblock similarity if we have a superblock - discard if different
121 * Record events, devicenum
122 * This should give us a list of devices for the array
123 * We should collect the most recent event number
124 *
125 * Count disks with recent enough event count
126 * While force && !enough disks
127 * Choose newest rejected disks, update event count
128 * mark clean and rewrite superblock
129 * If recent kernel:
130 * SET_ARRAY_INFO
131 * foreach device with recent events : ADD_NEW_DISK
132 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
133 * If old kernel:
134 * Check the device numbers in superblock are right
135 * update superblock if any changes
136 * START_ARRAY
137 *
138 */
139 int mdfd;
140 int clean;
141 int auto_assem = (mddev == NULL && !ident->uuid_set &&
142 ident->super_minor == UnSet && ident->name[0] == 0
143 && (ident->container == NULL || ident->member == NULL));
144 int old_linux = 0;
145 int vers = vers; /* Keep gcc quite - it really is initialised */
146 struct {
147 char *devname;
148 int uptodate; /* set once we decide that this device is as
149 * recent as everything else in the array.
150 */
151 struct mdinfo i;
152 } *devices;
153 int *best = NULL; /* indexed by raid_disk */
154 unsigned int bestcnt = 0;
155 int devcnt = 0;
156 unsigned int okcnt, sparecnt;
157 unsigned int req_cnt;
158 unsigned int i;
159 int most_recent = 0;
160 int chosen_drive;
161 int change = 0;
162 int inargv = 0;
163 int report_missmatch;
164 int bitmap_done;
165 int start_partial_ok = (runstop >= 0) &&
166 (force || devlist==NULL || auto_assem);
167 unsigned int num_devs;
168 mddev_dev_t tmpdev;
169 struct mdinfo info;
170 struct mdinfo *content = NULL;
171 char *avail;
172 int nextspare = 0;
173 char *name = NULL;
174 int trustworthy;
175 char chosen_name[1024];
176
177 if (get_linux_version() < 2004000)
178 old_linux = 1;
179
180 /*
181 * If any subdevs are listed, then any that don't
182 * match ident are discarded. Remainder must all match and
183 * become the array.
184 * If no subdevs, then we scan all devices in the config file, but
185 * there must be something in the identity
186 */
187
188 if (!devlist &&
189 ident->uuid_set == 0 &&
190 ident->super_minor < 0 &&
191 ident->devices == NULL) {
192 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
193 mddev ? mddev : "further assembly");
194 return 1;
195 }
196
197 if (devlist == NULL)
198 devlist = conf_get_devs();
199 else if (mddev)
200 inargv = 1;
201
202 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
203 try_again:
204 /* We come back here when doing auto-assembly and attempting some
205 * set of devices failed. Those are now marked as ->used==2 and
206 * we ignore them and try again
207 */
208
209 tmpdev = devlist; num_devs = 0;
210 while (tmpdev) {
211 if (tmpdev->used)
212 tmpdev->used = 2;
213 else
214 num_devs++;
215 tmpdev = tmpdev->next;
216 }
217 devices = malloc(num_devs * sizeof(*devices));
218
219 if (!st && ident->st) st = ident->st;
220
221 if (verbose>0)
222 fprintf(stderr, Name ": looking for devices for %s\n",
223 mddev ? mddev : "further assembly");
224
225 /* first walk the list of devices to find a consistent set
226 * that match the criterea, if that is possible.
227 * We flag the ones we like with 'used'.
228 */
229 for (tmpdev = devlist;
230 tmpdev;
231 tmpdev = tmpdev->next) {
232 char *devname = tmpdev->devname;
233 int dfd;
234 struct stat stb;
235 struct supertype *tst = dup_super(st);
236
237 if (tmpdev->used > 1) continue;
238
239 if (ident->devices &&
240 !match_oneof(ident->devices, devname)) {
241 if (report_missmatch)
242 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
243 continue;
244 }
245
246 dfd = dev_open(devname, O_RDONLY|O_EXCL);
247 if (dfd < 0) {
248 if (report_missmatch)
249 fprintf(stderr, Name ": cannot open device %s: %s\n",
250 devname, strerror(errno));
251 tmpdev->used = 2;
252 } else if (fstat(dfd, &stb)< 0) {
253 /* Impossible! */
254 fprintf(stderr, Name ": fstat failed for %s: %s\n",
255 devname, strerror(errno));
256 tmpdev->used = 2;
257 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
258 fprintf(stderr, Name ": %s is not a block device.\n",
259 devname);
260 tmpdev->used = 2;
261 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
262 if (report_missmatch)
263 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
264 devname);
265 tmpdev->used = 2;
266 } else if (auto_assem && st == NULL &&
267 !conf_test_metadata(tst->ss->name)) {
268 if (report_missmatch)
269 fprintf(stderr, Name ": %s has metadata type %s for which "
270 "auto-assembly is disabled\n",
271 devname, tst->ss->name);
272 tmpdev->used = 2;
273 } else if (tst->ss->load_super(tst,dfd, NULL)) {
274 if (report_missmatch)
275 fprintf( stderr, Name ": no RAID superblock on %s\n",
276 devname);
277 } else {
278 content = &info;
279 memset(content, 0, sizeof(*content));
280 tst->ss->getinfo_super(tst, content);
281 }
282 if (dfd >= 0) close(dfd);
283
284 if (tst && tst->sb && tst->ss->container_content
285 && tst->loaded_container) {
286 /* tmpdev is a container. We need to be either
287 * looking for a member, or auto-assembling
288 */
289 if (st) {
290 /* already found some components, this cannot
291 * be another one.
292 */
293 if (report_missmatch)
294 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
295 devname);
296 goto loop;
297 }
298
299 if (ident->container) {
300 if (ident->container[0] == '/' &&
301 !same_dev(ident->container, devname)) {
302 if (report_missmatch)
303 fprintf(stderr, Name ": %s is not the container required (%s)\n",
304 devname, ident->container);
305 goto loop;
306 }
307 if (ident->container[0] != '/') {
308 /* we have a uuid */
309 int uuid[4];
310 if (!parse_uuid(ident->container, uuid) ||
311 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
312 if (report_missmatch)
313 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
314 devname);
315 goto loop;
316 }
317 }
318 }
319 /* It is worth looking inside this container.
320 */
321 next_member:
322 if (tmpdev->content)
323 content = tmpdev->content;
324 else
325 content = tst->ss->container_content(tst);
326
327 tmpdev->content = content->next;
328 if (tmpdev->content == NULL)
329 tmpdev->used = 2;
330
331 } else if (ident->container || ident->member) {
332 /* No chance of this matching if we don't have
333 * a container */
334 if (report_missmatch)
335 fprintf(stderr, Name "%s is not a container, and one is required.\n",
336 devname);
337 goto loop;
338 }
339
340 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
341 (!tst || !tst->sb ||
342 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
343 if (report_missmatch)
344 fprintf(stderr, Name ": %s has wrong uuid.\n",
345 devname);
346 goto loop;
347 }
348 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
349 (!tst || !tst->sb ||
350 name_matches(content->name, ident->name, homehost)==0)) {
351 if (report_missmatch)
352 fprintf(stderr, Name ": %s has wrong name.\n",
353 devname);
354 goto loop;
355 }
356 if (ident->super_minor != UnSet &&
357 (!tst || !tst->sb ||
358 ident->super_minor != content->array.md_minor)) {
359 if (report_missmatch)
360 fprintf(stderr, Name ": %s has wrong super-minor.\n",
361 devname);
362 goto loop;
363 }
364 if (ident->level != UnSet &&
365 (!tst || !tst->sb ||
366 ident->level != content->array.level)) {
367 if (report_missmatch)
368 fprintf(stderr, Name ": %s has wrong raid level.\n",
369 devname);
370 goto loop;
371 }
372 if (ident->raid_disks != UnSet &&
373 (!tst || !tst->sb ||
374 ident->raid_disks!= content->array.raid_disks)) {
375 if (report_missmatch)
376 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
377 devname);
378 goto loop;
379 }
380 if (auto_assem) {
381 if (tst == NULL || tst->sb == NULL)
382 continue;
383 }
384 /* If we are this far, then we are nearly commited to this device.
385 * If the super_block doesn't exist, or doesn't match others,
386 * then we probably cannot continue
387 * However if one of the arrays is for the homehost, and
388 * the other isn't that can disambiguate.
389 */
390
391 if (!tst || !tst->sb) {
392 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
393 devname);
394 if (st)
395 st->ss->free_super(st);
396 return 1;
397 }
398
399 if (tst && tst->sb && tst->ss->container_content
400 && tst->loaded_container) {
401 /* we have the one container we need, don't keep
402 * looking. If the chosen member is active, skip.
403 */
404 if (is_member_busy(content->text_version)) {
405 if (report_missmatch)
406 fprintf(stderr, Name ": member %s in %s is already assembled\n",
407 content->text_version,
408 devname);
409 tst->ss->free_super(tst);
410 tst = NULL;
411 content = NULL;
412 if (auto_assem)
413 goto loop;
414 return 1;
415 }
416 st = tst; tst = NULL;
417 if (!auto_assem && tmpdev->next != NULL) {
418 fprintf(stderr, Name ": %s is a container, but is not "
419 "only device given: confused and aborting\n",
420 devname);
421 st->ss->free_super(st);
422 return 1;
423 }
424 break;
425 }
426 if (st == NULL)
427 st = dup_super(tst);
428 if (st->minor_version == -1)
429 st->minor_version = tst->minor_version;
430 if (st->ss != tst->ss ||
431 st->minor_version != tst->minor_version ||
432 st->ss->compare_super(st, tst) != 0) {
433 /* Some mismatch. If exactly one array matches this host,
434 * we can resolve on that one.
435 * Or, if we are auto assembling, we just ignore the second
436 * for now.
437 */
438 if (auto_assem)
439 goto loop;
440 if (homehost) {
441 int first = st->ss->match_home(st, homehost);
442 int last = tst->ss->match_home(tst, homehost);
443 if (first != last &&
444 (first == 1 || last == 1)) {
445 /* We can do something */
446 if (first) {/* just ignore this one */
447 if (report_missmatch)
448 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
449 devname);
450 goto loop;
451 } else { /* reject all those sofar */
452 mddev_dev_t td;
453 if (report_missmatch)
454 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
455 devname);
456 for (td=devlist; td != tmpdev; td=td->next)
457 if (td->used == 1)
458 td->used = 0;
459 tmpdev->used = 1;
460 goto loop;
461 }
462 }
463 }
464 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
465 devname);
466 tst->ss->free_super(tst);
467 st->ss->free_super(st);
468 return 1;
469 }
470
471 tmpdev->used = 1;
472
473 loop:
474 if (tmpdev->content)
475 goto next_member;
476 if (tst)
477 tst->ss->free_super(tst);
478 }
479
480 if (!st || !st->sb || !content)
481 return 2;
482
483 /* Now need to open the array device. Use create_mddev */
484 if (content == &info)
485 st->ss->getinfo_super(st, content);
486
487 trustworthy = FOREIGN;
488 name = content->name;
489 switch (st->ss->match_home(st, homehost)
490 ?: st->ss->match_home(st, "any")) {
491 case 1:
492 trustworthy = LOCAL;
493 name = strchr(content->name, ':');
494 if (name)
495 name++;
496 else
497 name = content->name;
498 break;
499 }
500 if (!auto_assem)
501 /* If the array is listed in mdadm.conf or on
502 * command line, then we trust the name
503 * even if the array doesn't look local
504 */
505 trustworthy = LOCAL;
506
507 if (name[0] == 0 &&
508 content->array.level == LEVEL_CONTAINER) {
509 name = content->text_version;
510 trustworthy = METADATA;
511 }
512
513 if (name[0] && trustworthy != LOCAL &&
514 ! require_homehost &&
515 conf_name_is_free(name))
516 trustworthy = LOCAL;
517
518 if (trustworthy == LOCAL &&
519 strchr(name, ':'))
520 /* Ignore 'host:' prefix of name */
521 name = strchr(name, ':')+1;
522
523 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
524 chosen_name);
525 if (mdfd < 0) {
526 st->ss->free_super(st);
527 free(devices);
528 if (auto_assem)
529 goto try_again;
530 return 1;
531 }
532 mddev = chosen_name;
533 vers = md_get_version(mdfd);
534 if (vers < 9000) {
535 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
536 " Upgrade your kernel or try --build\n");
537 close(mdfd);
538 return 1;
539 }
540 if (mddev_busy(fd2devnum(mdfd))) {
541 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
542 mddev);
543 for (tmpdev = devlist ;
544 tmpdev && tmpdev->used != 1;
545 tmpdev = tmpdev->next)
546 ;
547 if (tmpdev && auto_assem)
548 fprintf(stderr, Name ": %s needed for %s...\n",
549 mddev, tmpdev->devname);
550 close(mdfd);
551 mdfd = -3;
552 st->ss->free_super(st);
553 free(devices);
554 if (auto_assem)
555 goto try_again;
556 return 1;
557 }
558 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
559
560 #ifndef MDASSEMBLE
561 if (content != &info) {
562 /* This is a member of a container. Try starting the array. */
563 return assemble_container_content(st, mdfd, content, runstop,
564 chosen_name, verbose);
565 }
566 #endif
567 /* Ok, no bad inconsistancy, we can try updating etc */
568 bitmap_done = 0;
569 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
570 char *devname = tmpdev->devname;
571 struct stat stb;
572 /* looks like a good enough match to update the super block if needed */
573 #ifndef MDASSEMBLE
574 if (update) {
575 int dfd;
576 /* prepare useful information in info structures */
577 struct stat stb2;
578 struct supertype *tst;
579 fstat(mdfd, &stb2);
580
581 if (strcmp(update, "uuid")==0 &&
582 !ident->uuid_set) {
583 int rfd;
584 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
585 read(rfd, ident->uuid, 16) != 16) {
586 *(__u32*)(ident->uuid) = random();
587 *(__u32*)(ident->uuid+1) = random();
588 *(__u32*)(ident->uuid+2) = random();
589 *(__u32*)(ident->uuid+3) = random();
590 }
591 if (rfd >= 0) close(rfd);
592 }
593 dfd = dev_open(devname, O_RDWR|O_EXCL);
594
595 remove_partitions(dfd);
596
597 tst = dup_super(st);
598 tst->ss->load_super(tst, dfd, NULL);
599 tst->ss->getinfo_super(tst, content);
600
601 memcpy(content->uuid, ident->uuid, 16);
602 strcpy(content->name, ident->name);
603 content->array.md_minor = minor(stb2.st_rdev);
604
605 tst->ss->update_super(tst, content, update,
606 devname, verbose,
607 ident->uuid_set, homehost);
608 if (strcmp(update, "uuid")==0 &&
609 !ident->uuid_set) {
610 ident->uuid_set = 1;
611 memcpy(ident->uuid, content->uuid, 16);
612 }
613 if (dfd < 0)
614 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
615 devname);
616 else if (tst->ss->store_super(tst, dfd))
617 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
618 devname);
619 if (dfd >= 0)
620 close(dfd);
621
622 if (strcmp(update, "uuid")==0 &&
623 ident->bitmap_fd >= 0 && !bitmap_done) {
624 if (bitmap_update_uuid(ident->bitmap_fd,
625 content->uuid,
626 tst->ss->swapuuid) != 0)
627 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
628 else
629 bitmap_done = 1;
630 }
631 tst->ss->free_super(tst);
632 } else
633 #endif
634 {
635 struct supertype *tst = dup_super(st);
636 int dfd;
637 dfd = dev_open(devname, O_RDWR|O_EXCL);
638
639 remove_partitions(dfd);
640
641 tst->ss->load_super(tst, dfd, NULL);
642 tst->ss->getinfo_super(tst, content);
643 tst->ss->free_super(tst);
644 close(dfd);
645 }
646
647 stat(devname, &stb);
648
649 if (verbose > 0)
650 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
651 devname, mddev, content->disk.raid_disk);
652 devices[devcnt].devname = devname;
653 devices[devcnt].uptodate = 0;
654 devices[devcnt].i = *content;
655 devices[devcnt].i.disk.major = major(stb.st_rdev);
656 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
657 if (most_recent < devcnt) {
658 if (devices[devcnt].i.events
659 > devices[most_recent].i.events)
660 most_recent = devcnt;
661 }
662 if (content->array.level == -4)
663 /* with multipath, the raid_disk from the superblock is meaningless */
664 i = devcnt;
665 else
666 i = devices[devcnt].i.disk.raid_disk;
667 if (i+1 == 0) {
668 if (nextspare < content->array.raid_disks)
669 nextspare = content->array.raid_disks;
670 i = nextspare++;
671 } else {
672 if (i >= content->array.raid_disks &&
673 i >= nextspare)
674 nextspare = i+1;
675 }
676 if (i < 10000) {
677 if (i >= bestcnt) {
678 unsigned int newbestcnt = i+10;
679 int *newbest = malloc(sizeof(int)*newbestcnt);
680 unsigned int c;
681 for (c=0; c < newbestcnt; c++)
682 if (c < bestcnt)
683 newbest[c] = best[c];
684 else
685 newbest[c] = -1;
686 if (best)free(best);
687 best = newbest;
688 bestcnt = newbestcnt;
689 }
690 if (best[i] >=0 &&
691 devices[best[i]].i.events
692 == devices[devcnt].i.events
693 && (devices[best[i]].i.disk.minor
694 != devices[devcnt].i.disk.minor)
695 && st->ss == &super0
696 && content->array.level != LEVEL_MULTIPATH) {
697 /* two different devices with identical superblock.
698 * Could be a mis-detection caused by overlapping
699 * partitions. fail-safe.
700 */
701 fprintf(stderr, Name ": WARNING %s and %s appear"
702 " to have very similar superblocks.\n"
703 " If they are really different, "
704 "please --zero the superblock on one\n"
705 " If they are the same or overlap,"
706 " please remove one from %s.\n",
707 devices[best[i]].devname, devname,
708 inargv ? "the list" :
709 "the\n DEVICE list in mdadm.conf"
710 );
711 close(mdfd);
712 return 1;
713 }
714 if (best[i] == -1
715 || (devices[best[i]].i.events
716 < devices[devcnt].i.events))
717 best[i] = devcnt;
718 }
719 devcnt++;
720 }
721
722 if (devcnt == 0) {
723 fprintf(stderr, Name ": no devices found for %s\n",
724 mddev);
725 if (st)
726 st->ss->free_super(st);
727 close(mdfd);
728 return 1;
729 }
730
731 if (update && strcmp(update, "byteorder")==0)
732 st->minor_version = 90;
733
734 st->ss->getinfo_super(st, content);
735 clean = content->array.state & 1;
736
737 /* now we have some devices that might be suitable.
738 * I wonder how many
739 */
740 avail = malloc(content->array.raid_disks);
741 memset(avail, 0, content->array.raid_disks);
742 okcnt = 0;
743 sparecnt=0;
744 for (i=0; i< bestcnt ;i++) {
745 int j = best[i];
746 int event_margin = 1; /* always allow a difference of '1'
747 * like the kernel does
748 */
749 if (j < 0) continue;
750 /* note: we ignore error flags in multipath arrays
751 * as they don't make sense
752 */
753 if (content->array.level != -4)
754 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
755 if (!(devices[j].i.disk.state
756 & (1<<MD_DISK_FAULTY)))
757 sparecnt++;
758 continue;
759 }
760 if (devices[j].i.events+event_margin >=
761 devices[most_recent].i.events) {
762 devices[j].uptodate = 1;
763 if (i < content->array.raid_disks) {
764 okcnt++;
765 avail[i]=1;
766 } else
767 sparecnt++;
768 }
769 }
770 while (force && !enough(content->array.level, content->array.raid_disks,
771 content->array.layout, 1,
772 avail, okcnt)) {
773 /* Choose the newest best drive which is
774 * not up-to-date, update the superblock
775 * and add it.
776 */
777 int fd;
778 struct supertype *tst;
779 long long current_events;
780 chosen_drive = -1;
781 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
782 int j = best[i];
783 if (j>=0 &&
784 !devices[j].uptodate &&
785 (chosen_drive < 0 ||
786 devices[j].i.events
787 > devices[chosen_drive].i.events))
788 chosen_drive = j;
789 }
790 if (chosen_drive < 0)
791 break;
792 current_events = devices[chosen_drive].i.events;
793 add_another:
794 if (verbose >= 0)
795 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
796 devices[chosen_drive].devname,
797 devices[chosen_drive].i.disk.raid_disk,
798 (int)(devices[chosen_drive].i.events),
799 (int)(devices[most_recent].i.events));
800 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
801 if (fd < 0) {
802 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
803 devices[chosen_drive].devname);
804 devices[chosen_drive].i.events = 0;
805 continue;
806 }
807 tst = dup_super(st);
808 if (tst->ss->load_super(tst,fd, NULL)) {
809 close(fd);
810 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
811 devices[chosen_drive].devname);
812 devices[chosen_drive].i.events = 0;
813 continue;
814 }
815 content->events = devices[most_recent].i.events;
816 tst->ss->update_super(tst, content, "force-one",
817 devices[chosen_drive].devname, verbose,
818 0, NULL);
819
820 if (tst->ss->store_super(tst, fd)) {
821 close(fd);
822 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
823 devices[chosen_drive].devname);
824 devices[chosen_drive].i.events = 0;
825 tst->ss->free_super(tst);
826 continue;
827 }
828 close(fd);
829 devices[chosen_drive].i.events = devices[most_recent].i.events;
830 devices[chosen_drive].uptodate = 1;
831 avail[chosen_drive] = 1;
832 okcnt++;
833 tst->ss->free_super(tst);
834
835 /* If there are any other drives of the same vintage,
836 * add them in as well. We can't lose and we might gain
837 */
838 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
839 int j = best[i];
840 if (j >= 0 &&
841 !devices[j].uptodate &&
842 devices[j].i.events == current_events) {
843 chosen_drive = j;
844 goto add_another;
845 }
846 }
847 }
848
849 /* Now we want to look at the superblock which the kernel will base things on
850 * and compare the devices that we think are working with the devices that the
851 * superblock thinks are working.
852 * If there are differences and --force is given, then update this chosen
853 * superblock.
854 */
855 chosen_drive = -1;
856 st->ss->free_super(st);
857 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
858 int j = best[i];
859 int fd;
860
861 if (j<0)
862 continue;
863 if (!devices[j].uptodate)
864 continue;
865 chosen_drive = j;
866 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
867 fprintf(stderr, Name ": Cannot open %s: %s\n",
868 devices[j].devname, strerror(errno));
869 close(mdfd);
870 return 1;
871 }
872 if (st->ss->load_super(st,fd, NULL)) {
873 close(fd);
874 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
875 devices[j].devname);
876 close(mdfd);
877 return 1;
878 }
879 close(fd);
880 }
881 if (st->sb == NULL) {
882 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
883 close(mdfd);
884 return 1;
885 }
886 st->ss->getinfo_super(st, content);
887 #ifndef MDASSEMBLE
888 sysfs_init(content, mdfd, 0);
889 #endif
890 for (i=0; i<bestcnt; i++) {
891 int j = best[i];
892 unsigned int desired_state;
893
894 if (i < content->array.raid_disks)
895 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
896 else
897 desired_state = 0;
898
899 if (j<0)
900 continue;
901 if (!devices[j].uptodate)
902 continue;
903
904 devices[j].i.disk.state = desired_state;
905 if (!(devices[j].i.array.state & 1))
906 clean = 0;
907
908 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
909 verbose, 0, NULL)) {
910 if (force) {
911 if (verbose >= 0)
912 fprintf(stderr, Name ": "
913 "clearing FAULTY flag for device %d in %s for %s\n",
914 j, mddev, devices[j].devname);
915 change = 1;
916 } else {
917 if (verbose >= -1)
918 fprintf(stderr, Name ": "
919 "device %d in %s has wrong state in superblock, but %s seems ok\n",
920 i, mddev, devices[j].devname);
921 }
922 }
923 #if 0
924 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
925 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
926 i, mddev);
927 }
928 #endif
929 }
930 if (force && !clean &&
931 !enough(content->array.level, content->array.raid_disks,
932 content->array.layout, clean,
933 avail, okcnt)) {
934 change += st->ss->update_super(st, content, "force-array",
935 devices[chosen_drive].devname, verbose,
936 0, NULL);
937 clean = 1;
938 }
939
940 if (change) {
941 int fd;
942 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
943 if (fd < 0) {
944 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
945 devices[chosen_drive].devname);
946 close(mdfd);
947 return 1;
948 }
949 if (st->ss->store_super(st, fd)) {
950 close(fd);
951 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
952 devices[chosen_drive].devname);
953 close(mdfd);
954 return 1;
955 }
956 close(fd);
957 }
958
959 /* If we are in the middle of a reshape we may need to restore saved data
960 * that was moved aside due to the reshape overwriting live data
961 * The code of doing this lives in Grow.c
962 */
963 #ifndef MDASSEMBLE
964 if (content->reshape_active) {
965 int err = 0;
966 int *fdlist = malloc(sizeof(int)* bestcnt);
967 for (i=0; i<bestcnt; i++) {
968 int j = best[i];
969 if (j >= 0) {
970 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
971 if (fdlist[i] < 0) {
972 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
973 devices[j].devname);
974 err = 1;
975 break;
976 }
977 } else
978 fdlist[i] = -1;
979 }
980 if (!err)
981 err = Grow_restart(st, content, fdlist, bestcnt, backup_file);
982 while (i>0) {
983 i--;
984 if (fdlist[i]>=0) close(fdlist[i]);
985 }
986 if (err) {
987 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
988 close(mdfd);
989 return err;
990 }
991 }
992 #endif
993 /* count number of in-sync devices according to the superblock.
994 * We must have this number to start the array without -s or -R
995 */
996 req_cnt = content->array.working_disks;
997
998 /* Almost ready to actually *do* something */
999 if (!old_linux) {
1000 int rv;
1001
1002 /* First, fill in the map, so that udev can find our name
1003 * as soon as we become active.
1004 */
1005 map_update(NULL, fd2devnum(mdfd), content->text_version,
1006 content->uuid, chosen_name);
1007
1008 rv = set_array_info(mdfd, st, content);
1009 if (rv) {
1010 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1011 mddev, strerror(errno));
1012 close(mdfd);
1013 return 1;
1014 }
1015 if (ident->bitmap_fd >= 0) {
1016 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1017 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1018 close(mdfd);
1019 return 1;
1020 }
1021 } else if (ident->bitmap_file) {
1022 /* From config file */
1023 int bmfd = open(ident->bitmap_file, O_RDWR);
1024 if (bmfd < 0) {
1025 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1026 ident->bitmap_file);
1027 close(mdfd);
1028 return 1;
1029 }
1030 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1031 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1032 close(bmfd);
1033 close(mdfd);
1034 return 1;
1035 }
1036 close(bmfd);
1037 }
1038
1039 /* First, add the raid disks, but add the chosen one last */
1040 for (i=0; i<= bestcnt; i++) {
1041 int j;
1042 if (i < bestcnt) {
1043 j = best[i];
1044 if (j == chosen_drive)
1045 continue;
1046 } else
1047 j = chosen_drive;
1048
1049 if (j >= 0 /* && devices[j].uptodate */) {
1050 rv = add_disk(mdfd, st, content, &devices[j].i);
1051
1052 if (rv) {
1053 fprintf(stderr, Name ": failed to add "
1054 "%s to %s: %s\n",
1055 devices[j].devname,
1056 mddev,
1057 strerror(errno));
1058 if (i < content->array.raid_disks
1059 || i == bestcnt)
1060 okcnt--;
1061 else
1062 sparecnt--;
1063 } else if (verbose > 0)
1064 fprintf(stderr, Name ": added %s "
1065 "to %s as %d\n",
1066 devices[j].devname, mddev,
1067 devices[j].i.disk.raid_disk);
1068 } else if (verbose > 0 && i < content->array.raid_disks)
1069 fprintf(stderr, Name ": no uptodate device for "
1070 "slot %d of %s\n",
1071 i, mddev);
1072 }
1073
1074 if (content->array.level == LEVEL_CONTAINER) {
1075 if (verbose >= 0) {
1076 fprintf(stderr, Name ": Container %s has been "
1077 "assembled with %d drive%s",
1078 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1079 if (okcnt < content->array.raid_disks)
1080 fprintf(stderr, " (out of %d)",
1081 content->array.raid_disks);
1082 fprintf(stderr, "\n");
1083 }
1084 sysfs_uevent(content, "change");
1085 wait_for(chosen_name, mdfd);
1086 close(mdfd);
1087 return 0;
1088 }
1089
1090 if (runstop == 1 ||
1091 (runstop <= 0 &&
1092 ( enough(content->array.level, content->array.raid_disks,
1093 content->array.layout, clean, avail, okcnt) &&
1094 (okcnt >= req_cnt || start_partial_ok)
1095 ))) {
1096 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1097 if (verbose >= 0) {
1098 fprintf(stderr, Name ": %s has been started with %d drive%s",
1099 mddev, okcnt, okcnt==1?"":"s");
1100 if (okcnt < content->array.raid_disks)
1101 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1102 if (sparecnt)
1103 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1104 fprintf(stderr, ".\n");
1105 }
1106 if (content->reshape_active &&
1107 content->array.level >= 4 &&
1108 content->array.level <= 6) {
1109 /* might need to increase the size
1110 * of the stripe cache - default is 256
1111 */
1112 if (256 < 4 * (content->array.chunk_size/4096)) {
1113 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1114 if (sra)
1115 sysfs_set_num(sra, NULL,
1116 "stripe_cache_size",
1117 (4 * content->array.chunk_size / 4096) + 1);
1118 }
1119 }
1120 wait_for(mddev, mdfd);
1121 close(mdfd);
1122 if (auto_assem) {
1123 int usecs = 1;
1124 /* There is a nasty race with 'mdadm --monitor'.
1125 * If it opens this device before we close it,
1126 * it gets an incomplete open on which IO
1127 * doesn't work and the capacity is
1128 * wrong.
1129 * If we reopen (to check for layered devices)
1130 * before --monitor closes, we loose.
1131 *
1132 * So: wait upto 1 second for there to be
1133 * a non-zero capacity.
1134 */
1135 while (usecs < 1000) {
1136 mdfd = open(mddev, O_RDONLY);
1137 if (mdfd >= 0) {
1138 unsigned long long size;
1139 if (get_dev_size(mdfd, NULL, &size) &&
1140 size > 0)
1141 break;
1142 close(mdfd);
1143 }
1144 usleep(usecs);
1145 usecs <<= 1;
1146 }
1147 }
1148 return 0;
1149 }
1150 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1151 mddev, strerror(errno));
1152
1153 if (!enough(content->array.level, content->array.raid_disks,
1154 content->array.layout, 1, avail, okcnt))
1155 fprintf(stderr, Name ": Not enough devices to "
1156 "start the array.\n");
1157 else if (!enough(content->array.level,
1158 content->array.raid_disks,
1159 content->array.layout, clean,
1160 avail, okcnt))
1161 fprintf(stderr, Name ": Not enough devices to "
1162 "start the array while not clean "
1163 "- consider --force.\n");
1164
1165 if (auto_assem)
1166 ioctl(mdfd, STOP_ARRAY, NULL);
1167 close(mdfd);
1168 return 1;
1169 }
1170 if (runstop == -1) {
1171 fprintf(stderr, Name ": %s assembled from %d drive%s",
1172 mddev, okcnt, okcnt==1?"":"s");
1173 if (okcnt != content->array.raid_disks)
1174 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1175 fprintf(stderr, ", but not started.\n");
1176 close(mdfd);
1177 return 0;
1178 }
1179 if (verbose >= -1) {
1180 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1181 if (sparecnt)
1182 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1183 if (!enough(content->array.level, content->array.raid_disks,
1184 content->array.layout, 1, avail, okcnt))
1185 fprintf(stderr, " - not enough to start the array.\n");
1186 else if (!enough(content->array.level,
1187 content->array.raid_disks,
1188 content->array.layout, clean,
1189 avail, okcnt))
1190 fprintf(stderr, " - not enough to start the "
1191 "array while not clean - consider "
1192 "--force.\n");
1193 else {
1194 if (req_cnt == content->array.raid_disks)
1195 fprintf(stderr, " - need all %d to start it", req_cnt);
1196 else
1197 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1198 fprintf(stderr, " (use --run to insist).\n");
1199 }
1200 }
1201 if (auto_assem)
1202 ioctl(mdfd, STOP_ARRAY, NULL);
1203 close(mdfd);
1204 return 1;
1205 } else {
1206 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1207 * been updated to point to the current locations of devices.
1208 * so we can just start the array
1209 */
1210 unsigned long dev;
1211 dev = makedev(devices[chosen_drive].i.disk.major,
1212 devices[chosen_drive].i.disk.minor);
1213 if (ioctl(mdfd, START_ARRAY, dev)) {
1214 fprintf(stderr, Name ": Cannot start array: %s\n",
1215 strerror(errno));
1216 }
1217
1218 }
1219 close(mdfd);
1220 return 0;
1221 }
1222
1223 #ifndef MDASSEMBLE
1224 int assemble_container_content(struct supertype *st, int mdfd,
1225 struct mdinfo *content, int runstop,
1226 char *chosen_name, int verbose)
1227 {
1228 struct mdinfo *dev, *sra;
1229 int working = 0, preexist = 0;
1230 struct map_ent *map = NULL;
1231
1232 sysfs_init(content, mdfd, 0);
1233
1234 sra = sysfs_read(mdfd, 0, GET_VERSION);
1235 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1236 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1237 close(mdfd);
1238 return 1;
1239 }
1240 if (sra)
1241 sysfs_free(sra);
1242
1243 for (dev = content->devs; dev; dev = dev->next)
1244 if (sysfs_add_disk(content, dev, 1) == 0)
1245 working++;
1246 else if (errno == EEXIST)
1247 preexist++;
1248 if (working == 0) {
1249 close(mdfd);
1250 return 1;/* Nothing new, don't try to start */
1251 }
1252
1253 map_update(&map, fd2devnum(mdfd),
1254 content->text_version,
1255 content->uuid, chosen_name);
1256
1257 if (runstop > 0 ||
1258 (working + preexist) >= content->array.working_disks) {
1259 int err;
1260
1261 switch(content->array.level) {
1262 case LEVEL_LINEAR:
1263 case LEVEL_MULTIPATH:
1264 case 0:
1265 err = sysfs_set_str(content, NULL, "array_state",
1266 "active");
1267 break;
1268 default:
1269 err = sysfs_set_str(content, NULL, "array_state",
1270 "readonly");
1271 /* start mdmon if needed. */
1272 if (!err) {
1273 if (!mdmon_running(st->container_dev))
1274 start_mdmon(st->container_dev);
1275 ping_monitor(devnum2devname(st->container_dev));
1276 }
1277 break;
1278 }
1279 if (!err)
1280 sysfs_set_safemode(content, content->safe_mode_delay);
1281 if (verbose >= 0) {
1282 if (err)
1283 fprintf(stderr, Name
1284 ": array %s now has %d devices",
1285 chosen_name, working + preexist);
1286 else
1287 fprintf(stderr, Name
1288 ": Started %s with %d devices",
1289 chosen_name, working + preexist);
1290 if (preexist)
1291 fprintf(stderr, " (%d new)", working);
1292 fprintf(stderr, "\n");
1293 }
1294 if (!err)
1295 wait_for(chosen_name, mdfd);
1296 close(mdfd);
1297 return 0;
1298 /* FIXME should have an O_EXCL and wait for read-auto */
1299 } else {
1300 if (verbose >= 0)
1301 fprintf(stderr, Name
1302 ": %s assembled with %d devices but "
1303 "not started\n",
1304 chosen_name, working);
1305 close(mdfd);
1306 return 1;
1307 }
1308 }
1309 #endif
1310
Unnamed repository; edit this file 'description' to name the repository.