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