]> 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
mdadm.8: Man page updates
[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->name[0] == 0 &&
192 (ident->container == NULL || ident->member == NULL) &&
193 ident->devices == NULL) {
194 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
195 mddev ? mddev : "further assembly");
196 return 1;
197 }
198
199 if (devlist == NULL)
200 devlist = conf_get_devs();
201 else if (mddev)
202 inargv = 1;
203
204 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
205 try_again:
206 /* We come back here when doing auto-assembly and attempting some
207 * set of devices failed. Those are now marked as ->used==2 and
208 * we ignore them and try again
209 */
210
211 tmpdev = devlist; num_devs = 0;
212 while (tmpdev) {
213 if (tmpdev->used)
214 tmpdev->used = 2;
215 else
216 num_devs++;
217 tmpdev = tmpdev->next;
218 }
219 devices = malloc(num_devs * sizeof(*devices));
220
221 if (!st && ident->st) st = ident->st;
222
223 if (verbose>0)
224 fprintf(stderr, Name ": looking for devices for %s\n",
225 mddev ? mddev : "further assembly");
226
227 /* first walk the list of devices to find a consistent set
228 * that match the criterea, if that is possible.
229 * We flag the ones we like with 'used'.
230 */
231 for (tmpdev = devlist;
232 tmpdev;
233 tmpdev = tmpdev->next) {
234 char *devname = tmpdev->devname;
235 int dfd;
236 struct stat stb;
237 struct supertype *tst = dup_super(st);
238
239 if (tmpdev->used > 1) continue;
240
241 if (ident->devices &&
242 !match_oneof(ident->devices, devname)) {
243 if (report_missmatch)
244 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
245 continue;
246 }
247
248 dfd = dev_open(devname, O_RDONLY|O_EXCL);
249 if (dfd < 0) {
250 if (report_missmatch)
251 fprintf(stderr, Name ": cannot open device %s: %s\n",
252 devname, strerror(errno));
253 tmpdev->used = 2;
254 } else if (fstat(dfd, &stb)< 0) {
255 /* Impossible! */
256 fprintf(stderr, Name ": fstat failed for %s: %s\n",
257 devname, strerror(errno));
258 tmpdev->used = 2;
259 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
260 fprintf(stderr, Name ": %s is not a block device.\n",
261 devname);
262 tmpdev->used = 2;
263 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
264 if (report_missmatch)
265 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
266 devname);
267 tmpdev->used = 2;
268 } else if (auto_assem && st == NULL &&
269 !conf_test_metadata(tst->ss->name)) {
270 if (report_missmatch)
271 fprintf(stderr, Name ": %s has metadata type %s for which "
272 "auto-assembly is disabled\n",
273 devname, tst->ss->name);
274 tmpdev->used = 2;
275 } else if (tst->ss->load_super(tst,dfd, NULL)) {
276 if (report_missmatch)
277 fprintf( stderr, Name ": no RAID superblock on %s\n",
278 devname);
279 } else {
280 content = &info;
281 memset(content, 0, sizeof(*content));
282 tst->ss->getinfo_super(tst, content);
283 }
284 if (dfd >= 0) close(dfd);
285
286 if (tst && tst->sb && tst->ss->container_content
287 && tst->loaded_container) {
288 /* tmpdev is a container. We need to be either
289 * looking for a member, or auto-assembling
290 */
291 if (st) {
292 /* already found some components, this cannot
293 * be another one.
294 */
295 if (report_missmatch)
296 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
297 devname);
298 goto loop;
299 }
300
301 if (ident->container) {
302 if (ident->container[0] == '/' &&
303 !same_dev(ident->container, devname)) {
304 if (report_missmatch)
305 fprintf(stderr, Name ": %s is not the container required (%s)\n",
306 devname, ident->container);
307 goto loop;
308 }
309 if (ident->container[0] != '/') {
310 /* we have a uuid */
311 int uuid[4];
312 if (!parse_uuid(ident->container, uuid) ||
313 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
314 if (report_missmatch)
315 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
316 devname);
317 goto loop;
318 }
319 }
320 }
321 /* It is worth looking inside this container.
322 */
323 next_member:
324 if (tmpdev->content)
325 content = tmpdev->content;
326 else
327 content = tst->ss->container_content(tst);
328
329 tmpdev->content = content->next;
330 if (tmpdev->content == NULL)
331 tmpdev->used = 2;
332
333 } else if (ident->container || ident->member) {
334 /* No chance of this matching if we don't have
335 * a container */
336 if (report_missmatch)
337 fprintf(stderr, Name "%s is not a container, and one is required.\n",
338 devname);
339 goto loop;
340 }
341
342 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
343 (!tst || !tst->sb ||
344 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
345 if (report_missmatch)
346 fprintf(stderr, Name ": %s has wrong uuid.\n",
347 devname);
348 goto loop;
349 }
350 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
351 (!tst || !tst->sb ||
352 name_matches(content->name, ident->name, homehost)==0)) {
353 if (report_missmatch)
354 fprintf(stderr, Name ": %s has wrong name.\n",
355 devname);
356 goto loop;
357 }
358 if (ident->super_minor != UnSet &&
359 (!tst || !tst->sb ||
360 ident->super_minor != content->array.md_minor)) {
361 if (report_missmatch)
362 fprintf(stderr, Name ": %s has wrong super-minor.\n",
363 devname);
364 goto loop;
365 }
366 if (ident->level != UnSet &&
367 (!tst || !tst->sb ||
368 ident->level != content->array.level)) {
369 if (report_missmatch)
370 fprintf(stderr, Name ": %s has wrong raid level.\n",
371 devname);
372 goto loop;
373 }
374 if (ident->raid_disks != UnSet &&
375 (!tst || !tst->sb ||
376 ident->raid_disks!= content->array.raid_disks)) {
377 if (report_missmatch)
378 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
379 devname);
380 goto loop;
381 }
382 if (auto_assem) {
383 if (tst == NULL || tst->sb == NULL)
384 continue;
385 }
386 /* If we are this far, then we are nearly commited to this device.
387 * If the super_block doesn't exist, or doesn't match others,
388 * then we probably cannot continue
389 * However if one of the arrays is for the homehost, and
390 * the other isn't that can disambiguate.
391 */
392
393 if (!tst || !tst->sb) {
394 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
395 devname);
396 if (st)
397 st->ss->free_super(st);
398 return 1;
399 }
400
401 if (tst && tst->sb && tst->ss->container_content
402 && tst->loaded_container) {
403 /* we have the one container we need, don't keep
404 * looking. If the chosen member is active, skip.
405 */
406 if (is_member_busy(content->text_version)) {
407 if (report_missmatch)
408 fprintf(stderr, Name ": member %s in %s is already assembled\n",
409 content->text_version,
410 devname);
411 tst->ss->free_super(tst);
412 tst = NULL;
413 content = NULL;
414 if (auto_assem)
415 goto loop;
416 return 1;
417 }
418 st = tst; tst = NULL;
419 if (!auto_assem && tmpdev->next != NULL) {
420 fprintf(stderr, Name ": %s is a container, but is not "
421 "only device given: confused and aborting\n",
422 devname);
423 st->ss->free_super(st);
424 return 1;
425 }
426 break;
427 }
428 if (st == NULL)
429 st = dup_super(tst);
430 if (st->minor_version == -1)
431 st->minor_version = tst->minor_version;
432 if (st->ss != tst->ss ||
433 st->minor_version != tst->minor_version ||
434 st->ss->compare_super(st, tst) != 0) {
435 /* Some mismatch. If exactly one array matches this host,
436 * we can resolve on that one.
437 * Or, if we are auto assembling, we just ignore the second
438 * for now.
439 */
440 if (auto_assem)
441 goto loop;
442 if (homehost) {
443 int first = st->ss->match_home(st, homehost);
444 int last = tst->ss->match_home(tst, homehost);
445 if (first != last &&
446 (first == 1 || last == 1)) {
447 /* We can do something */
448 if (first) {/* just ignore this one */
449 if (report_missmatch)
450 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
451 devname);
452 goto loop;
453 } else { /* reject all those sofar */
454 mddev_dev_t td;
455 if (report_missmatch)
456 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
457 devname);
458 for (td=devlist; td != tmpdev; td=td->next)
459 if (td->used == 1)
460 td->used = 0;
461 tmpdev->used = 1;
462 goto loop;
463 }
464 }
465 }
466 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
467 devname);
468 tst->ss->free_super(tst);
469 st->ss->free_super(st);
470 return 1;
471 }
472
473 tmpdev->used = 1;
474
475 loop:
476 if (tmpdev->content)
477 goto next_member;
478 if (tst)
479 tst->ss->free_super(tst);
480 }
481
482 if (!st || !st->sb || !content)
483 return 2;
484
485 /* Now need to open the array device. Use create_mddev */
486 if (content == &info)
487 st->ss->getinfo_super(st, content);
488
489 trustworthy = FOREIGN;
490 name = content->name;
491 switch (st->ss->match_home(st, homehost)
492 ?: st->ss->match_home(st, "any")) {
493 case 1:
494 trustworthy = LOCAL;
495 name = strchr(content->name, ':');
496 if (name)
497 name++;
498 else
499 name = content->name;
500 break;
501 }
502 if (!auto_assem)
503 /* If the array is listed in mdadm.conf or on
504 * command line, then we trust the name
505 * even if the array doesn't look local
506 */
507 trustworthy = LOCAL;
508
509 if (name[0] == 0 &&
510 content->array.level == LEVEL_CONTAINER) {
511 name = content->text_version;
512 trustworthy = METADATA;
513 }
514
515 if (name[0] && trustworthy != LOCAL &&
516 ! require_homehost &&
517 conf_name_is_free(name))
518 trustworthy = LOCAL;
519
520 if (trustworthy == LOCAL &&
521 strchr(name, ':'))
522 /* Ignore 'host:' prefix of name */
523 name = strchr(name, ':')+1;
524
525 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
526 chosen_name);
527 if (mdfd < 0) {
528 st->ss->free_super(st);
529 free(devices);
530 if (auto_assem)
531 goto try_again;
532 return 1;
533 }
534 mddev = chosen_name;
535 vers = md_get_version(mdfd);
536 if (vers < 9000) {
537 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
538 " Upgrade your kernel or try --build\n");
539 close(mdfd);
540 return 1;
541 }
542 if (mddev_busy(fd2devnum(mdfd))) {
543 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
544 mddev);
545 for (tmpdev = devlist ;
546 tmpdev && tmpdev->used != 1;
547 tmpdev = tmpdev->next)
548 ;
549 if (tmpdev && auto_assem)
550 fprintf(stderr, Name ": %s needed for %s...\n",
551 mddev, tmpdev->devname);
552 close(mdfd);
553 mdfd = -3;
554 st->ss->free_super(st);
555 free(devices);
556 if (auto_assem)
557 goto try_again;
558 return 1;
559 }
560 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
561
562 #ifndef MDASSEMBLE
563 if (content != &info) {
564 /* This is a member of a container. Try starting the array. */
565 return assemble_container_content(st, mdfd, content, runstop,
566 chosen_name, verbose);
567 }
568 #endif
569 /* Ok, no bad inconsistancy, we can try updating etc */
570 bitmap_done = 0;
571 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
572 char *devname = tmpdev->devname;
573 struct stat stb;
574 /* looks like a good enough match to update the super block if needed */
575 #ifndef MDASSEMBLE
576 if (update) {
577 int dfd;
578 /* prepare useful information in info structures */
579 struct stat stb2;
580 struct supertype *tst;
581 fstat(mdfd, &stb2);
582
583 if (strcmp(update, "uuid")==0 &&
584 !ident->uuid_set) {
585 int rfd;
586 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
587 read(rfd, ident->uuid, 16) != 16) {
588 *(__u32*)(ident->uuid) = random();
589 *(__u32*)(ident->uuid+1) = random();
590 *(__u32*)(ident->uuid+2) = random();
591 *(__u32*)(ident->uuid+3) = random();
592 }
593 if (rfd >= 0) close(rfd);
594 }
595 dfd = dev_open(devname, O_RDWR|O_EXCL);
596
597 remove_partitions(dfd);
598
599 tst = dup_super(st);
600 tst->ss->load_super(tst, dfd, NULL);
601 tst->ss->getinfo_super(tst, content);
602
603 memcpy(content->uuid, ident->uuid, 16);
604 strcpy(content->name, ident->name);
605 content->array.md_minor = minor(stb2.st_rdev);
606
607 tst->ss->update_super(tst, content, update,
608 devname, verbose,
609 ident->uuid_set, homehost);
610 if (strcmp(update, "uuid")==0 &&
611 !ident->uuid_set) {
612 ident->uuid_set = 1;
613 memcpy(ident->uuid, content->uuid, 16);
614 }
615 if (dfd < 0)
616 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
617 devname);
618 else if (tst->ss->store_super(tst, dfd))
619 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
620 devname);
621 if (dfd >= 0)
622 close(dfd);
623
624 if (strcmp(update, "uuid")==0 &&
625 ident->bitmap_fd >= 0 && !bitmap_done) {
626 if (bitmap_update_uuid(ident->bitmap_fd,
627 content->uuid,
628 tst->ss->swapuuid) != 0)
629 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
630 else
631 bitmap_done = 1;
632 }
633 tst->ss->free_super(tst);
634 } else
635 #endif
636 {
637 struct supertype *tst = dup_super(st);
638 int dfd;
639 dfd = dev_open(devname, O_RDWR|O_EXCL);
640
641 remove_partitions(dfd);
642
643 tst->ss->load_super(tst, dfd, NULL);
644 tst->ss->getinfo_super(tst, content);
645 tst->ss->free_super(tst);
646 close(dfd);
647 }
648
649 stat(devname, &stb);
650
651 if (verbose > 0)
652 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
653 devname, mddev, content->disk.raid_disk);
654 devices[devcnt].devname = devname;
655 devices[devcnt].uptodate = 0;
656 devices[devcnt].i = *content;
657 devices[devcnt].i.disk.major = major(stb.st_rdev);
658 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
659 if (most_recent < devcnt) {
660 if (devices[devcnt].i.events
661 > devices[most_recent].i.events)
662 most_recent = devcnt;
663 }
664 if (content->array.level == -4)
665 /* with multipath, the raid_disk from the superblock is meaningless */
666 i = devcnt;
667 else
668 i = devices[devcnt].i.disk.raid_disk;
669 if (i+1 == 0) {
670 if (nextspare < content->array.raid_disks)
671 nextspare = content->array.raid_disks;
672 i = nextspare++;
673 } else {
674 if (i >= content->array.raid_disks &&
675 i >= nextspare)
676 nextspare = i+1;
677 }
678 if (i < 10000) {
679 if (i >= bestcnt) {
680 unsigned int newbestcnt = i+10;
681 int *newbest = malloc(sizeof(int)*newbestcnt);
682 unsigned int c;
683 for (c=0; c < newbestcnt; c++)
684 if (c < bestcnt)
685 newbest[c] = best[c];
686 else
687 newbest[c] = -1;
688 if (best)free(best);
689 best = newbest;
690 bestcnt = newbestcnt;
691 }
692 if (best[i] >=0 &&
693 devices[best[i]].i.events
694 == devices[devcnt].i.events
695 && (devices[best[i]].i.disk.minor
696 != devices[devcnt].i.disk.minor)
697 && st->ss == &super0
698 && content->array.level != LEVEL_MULTIPATH) {
699 /* two different devices with identical superblock.
700 * Could be a mis-detection caused by overlapping
701 * partitions. fail-safe.
702 */
703 fprintf(stderr, Name ": WARNING %s and %s appear"
704 " to have very similar superblocks.\n"
705 " If they are really different, "
706 "please --zero the superblock on one\n"
707 " If they are the same or overlap,"
708 " please remove one from %s.\n",
709 devices[best[i]].devname, devname,
710 inargv ? "the list" :
711 "the\n DEVICE list in mdadm.conf"
712 );
713 close(mdfd);
714 return 1;
715 }
716 if (best[i] == -1
717 || (devices[best[i]].i.events
718 < devices[devcnt].i.events))
719 best[i] = devcnt;
720 }
721 devcnt++;
722 }
723
724 if (devcnt == 0) {
725 fprintf(stderr, Name ": no devices found for %s\n",
726 mddev);
727 if (st)
728 st->ss->free_super(st);
729 close(mdfd);
730 return 1;
731 }
732
733 if (update && strcmp(update, "byteorder")==0)
734 st->minor_version = 90;
735
736 st->ss->getinfo_super(st, content);
737 clean = content->array.state & 1;
738
739 /* now we have some devices that might be suitable.
740 * I wonder how many
741 */
742 avail = malloc(content->array.raid_disks);
743 memset(avail, 0, content->array.raid_disks);
744 okcnt = 0;
745 sparecnt=0;
746 for (i=0; i< bestcnt ;i++) {
747 int j = best[i];
748 int event_margin = 1; /* always allow a difference of '1'
749 * like the kernel does
750 */
751 if (j < 0) continue;
752 /* note: we ignore error flags in multipath arrays
753 * as they don't make sense
754 */
755 if (content->array.level != -4)
756 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
757 if (!(devices[j].i.disk.state
758 & (1<<MD_DISK_FAULTY)))
759 sparecnt++;
760 continue;
761 }
762 if (devices[j].i.events+event_margin >=
763 devices[most_recent].i.events) {
764 devices[j].uptodate = 1;
765 if (i < content->array.raid_disks) {
766 okcnt++;
767 avail[i]=1;
768 } else
769 sparecnt++;
770 }
771 }
772 while (force && !enough(content->array.level, content->array.raid_disks,
773 content->array.layout, 1,
774 avail, okcnt)) {
775 /* Choose the newest best drive which is
776 * not up-to-date, update the superblock
777 * and add it.
778 */
779 int fd;
780 struct supertype *tst;
781 long long current_events;
782 chosen_drive = -1;
783 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
784 int j = best[i];
785 if (j>=0 &&
786 !devices[j].uptodate &&
787 (chosen_drive < 0 ||
788 devices[j].i.events
789 > devices[chosen_drive].i.events))
790 chosen_drive = j;
791 }
792 if (chosen_drive < 0)
793 break;
794 current_events = devices[chosen_drive].i.events;
795 add_another:
796 if (verbose >= 0)
797 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
798 devices[chosen_drive].devname,
799 devices[chosen_drive].i.disk.raid_disk,
800 (int)(devices[chosen_drive].i.events),
801 (int)(devices[most_recent].i.events));
802 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
803 if (fd < 0) {
804 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
805 devices[chosen_drive].devname);
806 devices[chosen_drive].i.events = 0;
807 continue;
808 }
809 tst = dup_super(st);
810 if (tst->ss->load_super(tst,fd, NULL)) {
811 close(fd);
812 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
813 devices[chosen_drive].devname);
814 devices[chosen_drive].i.events = 0;
815 continue;
816 }
817 content->events = devices[most_recent].i.events;
818 tst->ss->update_super(tst, content, "force-one",
819 devices[chosen_drive].devname, verbose,
820 0, NULL);
821
822 if (tst->ss->store_super(tst, fd)) {
823 close(fd);
824 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
825 devices[chosen_drive].devname);
826 devices[chosen_drive].i.events = 0;
827 tst->ss->free_super(tst);
828 continue;
829 }
830 close(fd);
831 devices[chosen_drive].i.events = devices[most_recent].i.events;
832 devices[chosen_drive].uptodate = 1;
833 avail[chosen_drive] = 1;
834 okcnt++;
835 tst->ss->free_super(tst);
836
837 /* If there are any other drives of the same vintage,
838 * add them in as well. We can't lose and we might gain
839 */
840 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
841 int j = best[i];
842 if (j >= 0 &&
843 !devices[j].uptodate &&
844 devices[j].i.events == current_events) {
845 chosen_drive = j;
846 goto add_another;
847 }
848 }
849 }
850
851 /* Now we want to look at the superblock which the kernel will base things on
852 * and compare the devices that we think are working with the devices that the
853 * superblock thinks are working.
854 * If there are differences and --force is given, then update this chosen
855 * superblock.
856 */
857 chosen_drive = -1;
858 st->ss->free_super(st);
859 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
860 int j = best[i];
861 int fd;
862
863 if (j<0)
864 continue;
865 if (!devices[j].uptodate)
866 continue;
867 chosen_drive = j;
868 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
869 fprintf(stderr, Name ": Cannot open %s: %s\n",
870 devices[j].devname, strerror(errno));
871 close(mdfd);
872 return 1;
873 }
874 if (st->ss->load_super(st,fd, NULL)) {
875 close(fd);
876 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
877 devices[j].devname);
878 close(mdfd);
879 return 1;
880 }
881 close(fd);
882 }
883 if (st->sb == NULL) {
884 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
885 close(mdfd);
886 return 1;
887 }
888 st->ss->getinfo_super(st, content);
889 #ifndef MDASSEMBLE
890 sysfs_init(content, mdfd, 0);
891 #endif
892 for (i=0; i<bestcnt; i++) {
893 int j = best[i];
894 unsigned int desired_state;
895
896 if (i < content->array.raid_disks)
897 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
898 else
899 desired_state = 0;
900
901 if (j<0)
902 continue;
903 if (!devices[j].uptodate)
904 continue;
905
906 devices[j].i.disk.state = desired_state;
907 if (!(devices[j].i.array.state & 1))
908 clean = 0;
909
910 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
911 verbose, 0, NULL)) {
912 if (force) {
913 if (verbose >= 0)
914 fprintf(stderr, Name ": "
915 "clearing FAULTY flag for device %d in %s for %s\n",
916 j, mddev, devices[j].devname);
917 change = 1;
918 } else {
919 if (verbose >= -1)
920 fprintf(stderr, Name ": "
921 "device %d in %s has wrong state in superblock, but %s seems ok\n",
922 i, mddev, devices[j].devname);
923 }
924 }
925 #if 0
926 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
927 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
928 i, mddev);
929 }
930 #endif
931 }
932 if (force && !clean &&
933 !enough(content->array.level, content->array.raid_disks,
934 content->array.layout, clean,
935 avail, okcnt)) {
936 change += st->ss->update_super(st, content, "force-array",
937 devices[chosen_drive].devname, verbose,
938 0, NULL);
939 clean = 1;
940 }
941
942 if (change) {
943 int fd;
944 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
945 if (fd < 0) {
946 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
947 devices[chosen_drive].devname);
948 close(mdfd);
949 return 1;
950 }
951 if (st->ss->store_super(st, fd)) {
952 close(fd);
953 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
954 devices[chosen_drive].devname);
955 close(mdfd);
956 return 1;
957 }
958 close(fd);
959 }
960
961 /* If we are in the middle of a reshape we may need to restore saved data
962 * that was moved aside due to the reshape overwriting live data
963 * The code of doing this lives in Grow.c
964 */
965 #ifndef MDASSEMBLE
966 if (content->reshape_active) {
967 int err = 0;
968 int *fdlist = malloc(sizeof(int)* bestcnt);
969 for (i=0; i<bestcnt; i++) {
970 int j = best[i];
971 if (j >= 0) {
972 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
973 if (fdlist[i] < 0) {
974 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
975 devices[j].devname);
976 err = 1;
977 break;
978 }
979 } else
980 fdlist[i] = -1;
981 }
982 if (!err)
983 err = Grow_restart(st, content, fdlist, bestcnt, backup_file);
984 while (i>0) {
985 i--;
986 if (fdlist[i]>=0) close(fdlist[i]);
987 }
988 if (err) {
989 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
990 close(mdfd);
991 return err;
992 }
993 }
994 #endif
995 /* count number of in-sync devices according to the superblock.
996 * We must have this number to start the array without -s or -R
997 */
998 req_cnt = content->array.working_disks;
999
1000 /* Almost ready to actually *do* something */
1001 if (!old_linux) {
1002 int rv;
1003
1004 /* First, fill in the map, so that udev can find our name
1005 * as soon as we become active.
1006 */
1007 map_update(NULL, fd2devnum(mdfd), content->text_version,
1008 content->uuid, chosen_name);
1009
1010 rv = set_array_info(mdfd, st, content);
1011 if (rv) {
1012 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1013 mddev, strerror(errno));
1014 close(mdfd);
1015 return 1;
1016 }
1017 if (ident->bitmap_fd >= 0) {
1018 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1019 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1020 close(mdfd);
1021 return 1;
1022 }
1023 } else if (ident->bitmap_file) {
1024 /* From config file */
1025 int bmfd = open(ident->bitmap_file, O_RDWR);
1026 if (bmfd < 0) {
1027 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1028 ident->bitmap_file);
1029 close(mdfd);
1030 return 1;
1031 }
1032 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1033 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1034 close(bmfd);
1035 close(mdfd);
1036 return 1;
1037 }
1038 close(bmfd);
1039 }
1040
1041 /* First, add the raid disks, but add the chosen one last */
1042 for (i=0; i<= bestcnt; i++) {
1043 int j;
1044 if (i < bestcnt) {
1045 j = best[i];
1046 if (j == chosen_drive)
1047 continue;
1048 } else
1049 j = chosen_drive;
1050
1051 if (j >= 0 /* && devices[j].uptodate */) {
1052 rv = add_disk(mdfd, st, content, &devices[j].i);
1053
1054 if (rv) {
1055 fprintf(stderr, Name ": failed to add "
1056 "%s to %s: %s\n",
1057 devices[j].devname,
1058 mddev,
1059 strerror(errno));
1060 if (i < content->array.raid_disks
1061 || i == bestcnt)
1062 okcnt--;
1063 else
1064 sparecnt--;
1065 } else if (verbose > 0)
1066 fprintf(stderr, Name ": added %s "
1067 "to %s as %d\n",
1068 devices[j].devname, mddev,
1069 devices[j].i.disk.raid_disk);
1070 } else if (verbose > 0 && i < content->array.raid_disks)
1071 fprintf(stderr, Name ": no uptodate device for "
1072 "slot %d of %s\n",
1073 i, mddev);
1074 }
1075
1076 if (content->array.level == LEVEL_CONTAINER) {
1077 if (verbose >= 0) {
1078 fprintf(stderr, Name ": Container %s has been "
1079 "assembled with %d drive%s",
1080 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1081 if (okcnt < content->array.raid_disks)
1082 fprintf(stderr, " (out of %d)",
1083 content->array.raid_disks);
1084 fprintf(stderr, "\n");
1085 }
1086 sysfs_uevent(content, "change");
1087 wait_for(chosen_name, mdfd);
1088 close(mdfd);
1089 return 0;
1090 }
1091
1092 if (runstop == 1 ||
1093 (runstop <= 0 &&
1094 ( enough(content->array.level, content->array.raid_disks,
1095 content->array.layout, clean, avail, okcnt) &&
1096 (okcnt >= req_cnt || start_partial_ok)
1097 ))) {
1098 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1099 if (verbose >= 0) {
1100 fprintf(stderr, Name ": %s has been started with %d drive%s",
1101 mddev, okcnt, okcnt==1?"":"s");
1102 if (okcnt < content->array.raid_disks)
1103 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1104 if (sparecnt)
1105 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1106 fprintf(stderr, ".\n");
1107 }
1108 if (content->reshape_active &&
1109 content->array.level >= 4 &&
1110 content->array.level <= 6) {
1111 /* might need to increase the size
1112 * of the stripe cache - default is 256
1113 */
1114 if (256 < 4 * (content->array.chunk_size/4096)) {
1115 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1116 if (sra)
1117 sysfs_set_num(sra, NULL,
1118 "stripe_cache_size",
1119 (4 * content->array.chunk_size / 4096) + 1);
1120 }
1121 }
1122 wait_for(mddev, mdfd);
1123 close(mdfd);
1124 if (auto_assem) {
1125 int usecs = 1;
1126 /* There is a nasty race with 'mdadm --monitor'.
1127 * If it opens this device before we close it,
1128 * it gets an incomplete open on which IO
1129 * doesn't work and the capacity is
1130 * wrong.
1131 * If we reopen (to check for layered devices)
1132 * before --monitor closes, we loose.
1133 *
1134 * So: wait upto 1 second for there to be
1135 * a non-zero capacity.
1136 */
1137 while (usecs < 1000) {
1138 mdfd = open(mddev, O_RDONLY);
1139 if (mdfd >= 0) {
1140 unsigned long long size;
1141 if (get_dev_size(mdfd, NULL, &size) &&
1142 size > 0)
1143 break;
1144 close(mdfd);
1145 }
1146 usleep(usecs);
1147 usecs <<= 1;
1148 }
1149 }
1150 return 0;
1151 }
1152 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1153 mddev, strerror(errno));
1154
1155 if (!enough(content->array.level, content->array.raid_disks,
1156 content->array.layout, 1, avail, okcnt))
1157 fprintf(stderr, Name ": Not enough devices to "
1158 "start the array.\n");
1159 else if (!enough(content->array.level,
1160 content->array.raid_disks,
1161 content->array.layout, clean,
1162 avail, okcnt))
1163 fprintf(stderr, Name ": Not enough devices to "
1164 "start the array while not clean "
1165 "- consider --force.\n");
1166
1167 if (auto_assem)
1168 ioctl(mdfd, STOP_ARRAY, NULL);
1169 close(mdfd);
1170 return 1;
1171 }
1172 if (runstop == -1) {
1173 fprintf(stderr, Name ": %s assembled from %d drive%s",
1174 mddev, okcnt, okcnt==1?"":"s");
1175 if (okcnt != content->array.raid_disks)
1176 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1177 fprintf(stderr, ", but not started.\n");
1178 close(mdfd);
1179 return 0;
1180 }
1181 if (verbose >= -1) {
1182 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1183 if (sparecnt)
1184 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1185 if (!enough(content->array.level, content->array.raid_disks,
1186 content->array.layout, 1, avail, okcnt))
1187 fprintf(stderr, " - not enough to start the array.\n");
1188 else if (!enough(content->array.level,
1189 content->array.raid_disks,
1190 content->array.layout, clean,
1191 avail, okcnt))
1192 fprintf(stderr, " - not enough to start the "
1193 "array while not clean - consider "
1194 "--force.\n");
1195 else {
1196 if (req_cnt == content->array.raid_disks)
1197 fprintf(stderr, " - need all %d to start it", req_cnt);
1198 else
1199 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1200 fprintf(stderr, " (use --run to insist).\n");
1201 }
1202 }
1203 if (auto_assem)
1204 ioctl(mdfd, STOP_ARRAY, NULL);
1205 close(mdfd);
1206 return 1;
1207 } else {
1208 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1209 * been updated to point to the current locations of devices.
1210 * so we can just start the array
1211 */
1212 unsigned long dev;
1213 dev = makedev(devices[chosen_drive].i.disk.major,
1214 devices[chosen_drive].i.disk.minor);
1215 if (ioctl(mdfd, START_ARRAY, dev)) {
1216 fprintf(stderr, Name ": Cannot start array: %s\n",
1217 strerror(errno));
1218 }
1219
1220 }
1221 close(mdfd);
1222 return 0;
1223 }
1224
1225 #ifndef MDASSEMBLE
1226 int assemble_container_content(struct supertype *st, int mdfd,
1227 struct mdinfo *content, int runstop,
1228 char *chosen_name, int verbose)
1229 {
1230 struct mdinfo *dev, *sra;
1231 int working = 0, preexist = 0;
1232 struct map_ent *map = NULL;
1233
1234 sysfs_init(content, mdfd, 0);
1235
1236 sra = sysfs_read(mdfd, 0, GET_VERSION);
1237 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1238 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1239 close(mdfd);
1240 return 1;
1241 }
1242 if (sra)
1243 sysfs_free(sra);
1244
1245 for (dev = content->devs; dev; dev = dev->next)
1246 if (sysfs_add_disk(content, dev, 1) == 0)
1247 working++;
1248 else if (errno == EEXIST)
1249 preexist++;
1250 if (working == 0) {
1251 close(mdfd);
1252 return 1;/* Nothing new, don't try to start */
1253 }
1254
1255 map_update(&map, fd2devnum(mdfd),
1256 content->text_version,
1257 content->uuid, chosen_name);
1258
1259 if (runstop > 0 ||
1260 (working + preexist) >= content->array.working_disks) {
1261 int err;
1262
1263 switch(content->array.level) {
1264 case LEVEL_LINEAR:
1265 case LEVEL_MULTIPATH:
1266 case 0:
1267 err = sysfs_set_str(content, NULL, "array_state",
1268 "active");
1269 break;
1270 default:
1271 err = sysfs_set_str(content, NULL, "array_state",
1272 "readonly");
1273 /* start mdmon if needed. */
1274 if (!err) {
1275 if (!mdmon_running(st->container_dev))
1276 start_mdmon(st->container_dev);
1277 ping_monitor(devnum2devname(st->container_dev));
1278 }
1279 break;
1280 }
1281 if (!err)
1282 sysfs_set_safemode(content, content->safe_mode_delay);
1283 if (verbose >= 0) {
1284 if (err)
1285 fprintf(stderr, Name
1286 ": array %s now has %d devices",
1287 chosen_name, working + preexist);
1288 else
1289 fprintf(stderr, Name
1290 ": Started %s with %d devices",
1291 chosen_name, working + preexist);
1292 if (preexist)
1293 fprintf(stderr, " (%d new)", working);
1294 fprintf(stderr, "\n");
1295 }
1296 if (!err)
1297 wait_for(chosen_name, mdfd);
1298 close(mdfd);
1299 return 0;
1300 /* FIXME should have an O_EXCL and wait for read-auto */
1301 } else {
1302 if (verbose >= 0)
1303 fprintf(stderr, Name
1304 ": %s assembled with %d devices but "
1305 "not started\n",
1306 chosen_name, working);
1307 close(mdfd);
1308 return 1;
1309 }
1310 }
1311 #endif
1312
Unnamed repository; edit this file 'description' to name the repository.