]> 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
fix examine_brief segfault
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(1, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 int Assemble(struct supertype *st, char *mddev,
74 mddev_ident_t ident,
75 mddev_dev_t devlist, char *backup_file,
76 int readonly, int runstop,
77 char *update, char *homehost, int require_homehost,
78 int verbose, int force)
79 {
80 /*
81 * The task of Assemble is to find a collection of
82 * devices that should (according to their superblocks)
83 * form an array, and to give this collection to the MD driver.
84 * In Linux-2.4 and later, this involves submitting a
85 * SET_ARRAY_INFO ioctl with no arg - to prepare
86 * the array - and then submit a number of
87 * ADD_NEW_DISK ioctls to add disks into
88 * the array. Finally RUN_ARRAY might
89 * be submitted to start the array.
90 *
91 * Much of the work of Assemble is in finding and/or
92 * checking the disks to make sure they look right.
93 *
94 * If mddev is not set, then scan must be set and we
95 * read through the config file for dev+uuid mapping
96 * We recurse, setting mddev, for each device that
97 * - isn't running
98 * - has a valid uuid (or any uuid if !uuidset)
99 *
100 * If mddev is set, we try to determine state of md.
101 * check version - must be at least 0.90.0
102 * check kernel version. must be at least 2.4.
103 * If not, we can possibly fall back on START_ARRAY
104 * Try to GET_ARRAY_INFO.
105 * If possible, give up
106 * If not, try to STOP_ARRAY just to make sure
107 *
108 * If !uuidset and scan, look in conf-file for uuid
109 * If not found, give up
110 * If !devlist and scan and uuidset, get list of devs from conf-file
111 *
112 * For each device:
113 * Check superblock - discard if bad
114 * Check uuid (set if we don't have one) - discard if no match
115 * Check superblock similarity if we have a superblock - discard if different
116 * Record events, devicenum
117 * This should give us a list of devices for the array
118 * We should collect the most recent event number
119 *
120 * Count disks with recent enough event count
121 * While force && !enough disks
122 * Choose newest rejected disks, update event count
123 * mark clean and rewrite superblock
124 * If recent kernel:
125 * SET_ARRAY_INFO
126 * foreach device with recent events : ADD_NEW_DISK
127 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
128 * If old kernel:
129 * Check the device numbers in superblock are right
130 * update superblock if any changes
131 * START_ARRAY
132 *
133 */
134 int mdfd;
135 int clean;
136 int auto_assem = (mddev == NULL && !ident->uuid_set &&
137 ident->super_minor == UnSet && ident->name[0] == 0
138 && (ident->container == NULL || ident->member == NULL));
139 int old_linux = 0;
140 int vers = vers; /* Keep gcc quite - it really is initialised */
141 struct {
142 char *devname;
143 int uptodate; /* set once we decide that this device is as
144 * recent as everything else in the array.
145 */
146 struct mdinfo i;
147 } *devices;
148 int *best = NULL; /* indexed by raid_disk */
149 unsigned int bestcnt = 0;
150 int devcnt = 0;
151 unsigned int okcnt, sparecnt;
152 unsigned int req_cnt;
153 unsigned int i;
154 int most_recent = 0;
155 int chosen_drive;
156 int change = 0;
157 int inargv = 0;
158 int report_missmatch;
159 int bitmap_done;
160 int start_partial_ok = (runstop >= 0) &&
161 (force || devlist==NULL || auto_assem);
162 unsigned int num_devs;
163 mddev_dev_t tmpdev;
164 struct mdinfo info;
165 struct mdinfo *content = NULL;
166 char *avail;
167 int nextspare = 0;
168 char *name = NULL;
169 int trustworthy;
170 char chosen_name[1024];
171
172 if (get_linux_version() < 2004000)
173 old_linux = 1;
174
175 /*
176 * If any subdevs are listed, then any that don't
177 * match ident are discarded. Remainder must all match and
178 * become the array.
179 * If no subdevs, then we scan all devices in the config file, but
180 * there must be something in the identity
181 */
182
183 if (!devlist &&
184 ident->uuid_set == 0 &&
185 ident->super_minor < 0 &&
186 ident->name[0] == 0 &&
187 (ident->container == NULL || ident->member == NULL) &&
188 ident->devices == NULL) {
189 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
190 mddev ? mddev : "further assembly");
191 return 1;
192 }
193
194 if (devlist == NULL)
195 devlist = conf_get_devs();
196 else if (mddev)
197 inargv = 1;
198
199 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
200 try_again:
201 /* We come back here when doing auto-assembly and attempting some
202 * set of devices failed. Those are now marked as ->used==2 and
203 * we ignore them and try again
204 */
205
206 tmpdev = devlist; num_devs = 0;
207 while (tmpdev) {
208 if (tmpdev->used)
209 tmpdev->used = 2;
210 else
211 num_devs++;
212 tmpdev = tmpdev->next;
213 }
214 devices = malloc(num_devs * sizeof(*devices));
215
216 if (!st && ident->st) st = ident->st;
217
218 if (verbose>0)
219 fprintf(stderr, Name ": looking for devices for %s\n",
220 mddev ? mddev : "further assembly");
221
222 /* first walk the list of devices to find a consistent set
223 * that match the criterea, if that is possible.
224 * We flag the ones we like with 'used'.
225 */
226 for (tmpdev = devlist;
227 tmpdev;
228 tmpdev = tmpdev->next) {
229 char *devname = tmpdev->devname;
230 int dfd;
231 struct stat stb;
232 struct supertype *tst = dup_super(st);
233
234 if (tmpdev->used > 1) continue;
235
236 if (ident->devices &&
237 !match_oneof(ident->devices, devname)) {
238 if (report_missmatch)
239 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
240 continue;
241 }
242
243 dfd = dev_open(devname, O_RDONLY|O_EXCL);
244 if (dfd < 0) {
245 if (report_missmatch)
246 fprintf(stderr, Name ": cannot open device %s: %s\n",
247 devname, strerror(errno));
248 tmpdev->used = 2;
249 } else if (fstat(dfd, &stb)< 0) {
250 /* Impossible! */
251 fprintf(stderr, Name ": fstat failed for %s: %s\n",
252 devname, strerror(errno));
253 tmpdev->used = 2;
254 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
255 fprintf(stderr, Name ": %s is not a block device.\n",
256 devname);
257 tmpdev->used = 2;
258 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
259 if (report_missmatch)
260 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
261 devname);
262 tmpdev->used = 2;
263 } else if (auto_assem && st == NULL &&
264 !conf_test_metadata(tst->ss->name)) {
265 if (report_missmatch)
266 fprintf(stderr, Name ": %s has metadata type %s for which "
267 "auto-assembly is disabled\n",
268 devname, tst->ss->name);
269 tmpdev->used = 2;
270 } else if (tst->ss->load_super(tst,dfd, NULL)) {
271 if (report_missmatch)
272 fprintf( stderr, Name ": no RAID superblock on %s\n",
273 devname);
274 } else {
275 content = &info;
276 memset(content, 0, sizeof(*content));
277 tst->ss->getinfo_super(tst, content);
278 }
279 if (dfd >= 0) close(dfd);
280
281 if (tst && tst->sb && tst->ss->container_content
282 && tst->loaded_container) {
283 /* tmpdev is a container. We need to be either
284 * looking for a member, or auto-assembling
285 */
286 if (st) {
287 /* already found some components, this cannot
288 * be another one.
289 */
290 if (report_missmatch)
291 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
292 devname);
293 goto loop;
294 }
295
296 if (ident->container) {
297 if (ident->container[0] == '/' &&
298 !same_dev(ident->container, devname)) {
299 if (report_missmatch)
300 fprintf(stderr, Name ": %s is not the container required (%s)\n",
301 devname, ident->container);
302 goto loop;
303 }
304 if (ident->container[0] != '/') {
305 /* we have a uuid */
306 int uuid[4];
307 if (!parse_uuid(ident->container, uuid) ||
308 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
309 if (report_missmatch)
310 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
311 devname);
312 goto loop;
313 }
314 }
315 }
316 /* It is worth looking inside this container.
317 */
318 next_member:
319 if (tmpdev->content)
320 content = tmpdev->content;
321 else
322 content = tst->ss->container_content(tst);
323
324 tmpdev->content = content->next;
325 if (tmpdev->content == NULL)
326 tmpdev->used = 2;
327
328 } else if (ident->container || ident->member) {
329 /* No chance of this matching if we don't have
330 * a container */
331 if (report_missmatch)
332 fprintf(stderr, Name "%s is not a container, and one is required.\n",
333 devname);
334 goto loop;
335 }
336
337 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
338 (!tst || !tst->sb ||
339 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
340 if (report_missmatch)
341 fprintf(stderr, Name ": %s has wrong uuid.\n",
342 devname);
343 goto loop;
344 }
345 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
346 (!tst || !tst->sb ||
347 name_matches(content->name, ident->name, homehost)==0)) {
348 if (report_missmatch)
349 fprintf(stderr, Name ": %s has wrong name.\n",
350 devname);
351 goto loop;
352 }
353 if (ident->super_minor != UnSet &&
354 (!tst || !tst->sb ||
355 ident->super_minor != content->array.md_minor)) {
356 if (report_missmatch)
357 fprintf(stderr, Name ": %s has wrong super-minor.\n",
358 devname);
359 goto loop;
360 }
361 if (ident->level != UnSet &&
362 (!tst || !tst->sb ||
363 ident->level != content->array.level)) {
364 if (report_missmatch)
365 fprintf(stderr, Name ": %s has wrong raid level.\n",
366 devname);
367 goto loop;
368 }
369 if (ident->raid_disks != UnSet &&
370 (!tst || !tst->sb ||
371 ident->raid_disks!= content->array.raid_disks)) {
372 if (report_missmatch)
373 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
374 devname);
375 goto loop;
376 }
377 if (auto_assem) {
378 if (tst == NULL || tst->sb == NULL)
379 continue;
380 }
381 /* If we are this far, then we are nearly commited to this device.
382 * If the super_block doesn't exist, or doesn't match others,
383 * then we probably cannot continue
384 * However if one of the arrays is for the homehost, and
385 * the other isn't that can disambiguate.
386 */
387
388 if (!tst || !tst->sb) {
389 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
390 devname);
391 if (st)
392 st->ss->free_super(st);
393 return 1;
394 }
395
396 if (tst && tst->sb && tst->ss->container_content
397 && tst->loaded_container) {
398 /* we have the one container we need, don't keep
399 * looking. If the chosen member is active, skip.
400 */
401 if (is_member_busy(content->text_version)) {
402 if (report_missmatch)
403 fprintf(stderr, Name ": member %s in %s is already assembled\n",
404 content->text_version,
405 devname);
406 tst->ss->free_super(tst);
407 tst = NULL;
408 content = NULL;
409 if (auto_assem)
410 goto loop;
411 return 1;
412 }
413 st = tst; tst = NULL;
414 if (!auto_assem && tmpdev->next != NULL) {
415 fprintf(stderr, Name ": %s is a container, but is not "
416 "only device given: confused and aborting\n",
417 devname);
418 st->ss->free_super(st);
419 return 1;
420 }
421 break;
422 }
423 if (st == NULL)
424 st = dup_super(tst);
425 if (st->minor_version == -1)
426 st->minor_version = tst->minor_version;
427 if (st->ss != tst->ss ||
428 st->minor_version != tst->minor_version ||
429 st->ss->compare_super(st, tst) != 0) {
430 /* Some mismatch. If exactly one array matches this host,
431 * we can resolve on that one.
432 * Or, if we are auto assembling, we just ignore the second
433 * for now.
434 */
435 if (auto_assem)
436 goto loop;
437 if (homehost) {
438 int first = st->ss->match_home(st, homehost);
439 int last = tst->ss->match_home(tst, homehost);
440 if (first != last &&
441 (first == 1 || last == 1)) {
442 /* We can do something */
443 if (first) {/* just ignore this one */
444 if (report_missmatch)
445 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
446 devname);
447 goto loop;
448 } else { /* reject all those sofar */
449 mddev_dev_t td;
450 if (report_missmatch)
451 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
452 devname);
453 for (td=devlist; td != tmpdev; td=td->next)
454 if (td->used == 1)
455 td->used = 0;
456 tmpdev->used = 1;
457 goto loop;
458 }
459 }
460 }
461 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
462 devname);
463 tst->ss->free_super(tst);
464 st->ss->free_super(st);
465 return 1;
466 }
467
468 tmpdev->used = 1;
469
470 loop:
471 if (tmpdev->content)
472 goto next_member;
473 if (tst)
474 tst->ss->free_super(tst);
475 }
476
477 if (!st || !st->sb || !content)
478 return 2;
479
480 /* Now need to open the array device. Use create_mddev */
481 if (content == &info)
482 st->ss->getinfo_super(st, content);
483
484 trustworthy = FOREIGN;
485 name = content->name;
486 switch (st->ss->match_home(st, homehost)
487 ?: st->ss->match_home(st, "any")) {
488 case 1:
489 trustworthy = LOCAL;
490 name = strchr(content->name, ':');
491 if (name)
492 name++;
493 else
494 name = content->name;
495 break;
496 }
497 if (!auto_assem)
498 /* If the array is listed in mdadm.conf or on
499 * command line, then we trust the name
500 * even if the array doesn't look local
501 */
502 trustworthy = LOCAL;
503
504 if (name[0] == 0 &&
505 content->array.level == LEVEL_CONTAINER) {
506 name = content->text_version;
507 trustworthy = METADATA;
508 }
509
510 if (name[0] && trustworthy != LOCAL &&
511 ! require_homehost &&
512 conf_name_is_free(name))
513 trustworthy = LOCAL;
514
515 if (trustworthy == LOCAL &&
516 strchr(name, ':'))
517 /* Ignore 'host:' prefix of name */
518 name = strchr(name, ':')+1;
519
520 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
521 chosen_name);
522 if (mdfd < 0) {
523 st->ss->free_super(st);
524 free(devices);
525 if (auto_assem)
526 goto try_again;
527 return 1;
528 }
529 mddev = chosen_name;
530 vers = md_get_version(mdfd);
531 if (vers < 9000) {
532 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
533 " Upgrade your kernel or try --build\n");
534 close(mdfd);
535 return 1;
536 }
537 if (mddev_busy(fd2devnum(mdfd))) {
538 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
539 mddev);
540 for (tmpdev = devlist ;
541 tmpdev && tmpdev->used != 1;
542 tmpdev = tmpdev->next)
543 ;
544 if (tmpdev && auto_assem)
545 fprintf(stderr, Name ": %s needed for %s...\n",
546 mddev, tmpdev->devname);
547 close(mdfd);
548 mdfd = -3;
549 st->ss->free_super(st);
550 free(devices);
551 if (auto_assem)
552 goto try_again;
553 return 1;
554 }
555 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
556
557 #ifndef MDASSEMBLE
558 if (content != &info) {
559 /* This is a member of a container. Try starting the array. */
560 return assemble_container_content(st, mdfd, content, runstop,
561 chosen_name, verbose);
562 }
563 #endif
564 /* Ok, no bad inconsistancy, we can try updating etc */
565 bitmap_done = 0;
566 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
567 char *devname = tmpdev->devname;
568 struct stat stb;
569 /* looks like a good enough match to update the super block if needed */
570 #ifndef MDASSEMBLE
571 if (update) {
572 int dfd;
573 /* prepare useful information in info structures */
574 struct stat stb2;
575 struct supertype *tst;
576 fstat(mdfd, &stb2);
577
578 if (strcmp(update, "uuid")==0 &&
579 !ident->uuid_set) {
580 int rfd;
581 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
582 read(rfd, ident->uuid, 16) != 16) {
583 *(__u32*)(ident->uuid) = random();
584 *(__u32*)(ident->uuid+1) = random();
585 *(__u32*)(ident->uuid+2) = random();
586 *(__u32*)(ident->uuid+3) = random();
587 }
588 if (rfd >= 0) close(rfd);
589 }
590 dfd = dev_open(devname, O_RDWR|O_EXCL);
591
592 remove_partitions(dfd);
593
594 tst = dup_super(st);
595 tst->ss->load_super(tst, dfd, NULL);
596 tst->ss->getinfo_super(tst, content);
597
598 memcpy(content->uuid, ident->uuid, 16);
599 strcpy(content->name, ident->name);
600 content->array.md_minor = minor(stb2.st_rdev);
601
602 tst->ss->update_super(tst, content, update,
603 devname, verbose,
604 ident->uuid_set, homehost);
605 if (strcmp(update, "uuid")==0 &&
606 !ident->uuid_set) {
607 ident->uuid_set = 1;
608 memcpy(ident->uuid, content->uuid, 16);
609 }
610 if (dfd < 0)
611 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
612 devname);
613 else if (tst->ss->store_super(tst, dfd))
614 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
615 devname);
616 if (dfd >= 0)
617 close(dfd);
618
619 if (strcmp(update, "uuid")==0 &&
620 ident->bitmap_fd >= 0 && !bitmap_done) {
621 if (bitmap_update_uuid(ident->bitmap_fd,
622 content->uuid,
623 tst->ss->swapuuid) != 0)
624 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
625 else
626 bitmap_done = 1;
627 }
628 tst->ss->free_super(tst);
629 } else
630 #endif
631 {
632 struct supertype *tst = dup_super(st);
633 int dfd;
634 dfd = dev_open(devname, O_RDWR|O_EXCL);
635
636 remove_partitions(dfd);
637
638 tst->ss->load_super(tst, dfd, NULL);
639 tst->ss->getinfo_super(tst, content);
640 tst->ss->free_super(tst);
641 close(dfd);
642 }
643
644 stat(devname, &stb);
645
646 if (verbose > 0)
647 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
648 devname, mddev, content->disk.raid_disk);
649 devices[devcnt].devname = devname;
650 devices[devcnt].uptodate = 0;
651 devices[devcnt].i = *content;
652 devices[devcnt].i.disk.major = major(stb.st_rdev);
653 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
654 if (most_recent < devcnt) {
655 if (devices[devcnt].i.events
656 > devices[most_recent].i.events)
657 most_recent = devcnt;
658 }
659 if (content->array.level == -4)
660 /* with multipath, the raid_disk from the superblock is meaningless */
661 i = devcnt;
662 else
663 i = devices[devcnt].i.disk.raid_disk;
664 if (i+1 == 0) {
665 if (nextspare < content->array.raid_disks)
666 nextspare = content->array.raid_disks;
667 i = nextspare++;
668 } else {
669 if (i >= content->array.raid_disks &&
670 i >= nextspare)
671 nextspare = i+1;
672 }
673 if (i < 10000) {
674 if (i >= bestcnt) {
675 unsigned int newbestcnt = i+10;
676 int *newbest = malloc(sizeof(int)*newbestcnt);
677 unsigned int c;
678 for (c=0; c < newbestcnt; c++)
679 if (c < bestcnt)
680 newbest[c] = best[c];
681 else
682 newbest[c] = -1;
683 if (best)free(best);
684 best = newbest;
685 bestcnt = newbestcnt;
686 }
687 if (best[i] >=0 &&
688 devices[best[i]].i.events
689 == devices[devcnt].i.events
690 && (devices[best[i]].i.disk.minor
691 != devices[devcnt].i.disk.minor)
692 && st->ss == &super0
693 && content->array.level != LEVEL_MULTIPATH) {
694 /* two different devices with identical superblock.
695 * Could be a mis-detection caused by overlapping
696 * partitions. fail-safe.
697 */
698 fprintf(stderr, Name ": WARNING %s and %s appear"
699 " to have very similar superblocks.\n"
700 " If they are really different, "
701 "please --zero the superblock on one\n"
702 " If they are the same or overlap,"
703 " please remove one from %s.\n",
704 devices[best[i]].devname, devname,
705 inargv ? "the list" :
706 "the\n DEVICE list in mdadm.conf"
707 );
708 close(mdfd);
709 return 1;
710 }
711 if (best[i] == -1
712 || (devices[best[i]].i.events
713 < devices[devcnt].i.events))
714 best[i] = devcnt;
715 }
716 devcnt++;
717 }
718
719 if (devcnt == 0) {
720 fprintf(stderr, Name ": no devices found for %s\n",
721 mddev);
722 if (st)
723 st->ss->free_super(st);
724 close(mdfd);
725 return 1;
726 }
727
728 if (update && strcmp(update, "byteorder")==0)
729 st->minor_version = 90;
730
731 st->ss->getinfo_super(st, content);
732 clean = content->array.state & 1;
733
734 /* now we have some devices that might be suitable.
735 * I wonder how many
736 */
737 avail = malloc(content->array.raid_disks);
738 memset(avail, 0, content->array.raid_disks);
739 okcnt = 0;
740 sparecnt=0;
741 for (i=0; i< bestcnt ;i++) {
742 int j = best[i];
743 int event_margin = 1; /* always allow a difference of '1'
744 * like the kernel does
745 */
746 if (j < 0) continue;
747 /* note: we ignore error flags in multipath arrays
748 * as they don't make sense
749 */
750 if (content->array.level != -4)
751 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
752 if (!(devices[j].i.disk.state
753 & (1<<MD_DISK_FAULTY)))
754 sparecnt++;
755 continue;
756 }
757 if (devices[j].i.events+event_margin >=
758 devices[most_recent].i.events) {
759 devices[j].uptodate = 1;
760 if (i < content->array.raid_disks) {
761 okcnt++;
762 avail[i]=1;
763 } else
764 sparecnt++;
765 }
766 }
767 while (force && !enough(content->array.level, content->array.raid_disks,
768 content->array.layout, 1,
769 avail, okcnt)) {
770 /* Choose the newest best drive which is
771 * not up-to-date, update the superblock
772 * and add it.
773 */
774 int fd;
775 struct supertype *tst;
776 long long current_events;
777 chosen_drive = -1;
778 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
779 int j = best[i];
780 if (j>=0 &&
781 !devices[j].uptodate &&
782 (chosen_drive < 0 ||
783 devices[j].i.events
784 > devices[chosen_drive].i.events))
785 chosen_drive = j;
786 }
787 if (chosen_drive < 0)
788 break;
789 current_events = devices[chosen_drive].i.events;
790 add_another:
791 if (verbose >= 0)
792 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
793 devices[chosen_drive].devname,
794 devices[chosen_drive].i.disk.raid_disk,
795 (int)(devices[chosen_drive].i.events),
796 (int)(devices[most_recent].i.events));
797 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
798 if (fd < 0) {
799 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
800 devices[chosen_drive].devname);
801 devices[chosen_drive].i.events = 0;
802 continue;
803 }
804 tst = dup_super(st);
805 if (tst->ss->load_super(tst,fd, NULL)) {
806 close(fd);
807 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
808 devices[chosen_drive].devname);
809 devices[chosen_drive].i.events = 0;
810 continue;
811 }
812 content->events = devices[most_recent].i.events;
813 tst->ss->update_super(tst, content, "force-one",
814 devices[chosen_drive].devname, verbose,
815 0, NULL);
816
817 if (tst->ss->store_super(tst, fd)) {
818 close(fd);
819 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
820 devices[chosen_drive].devname);
821 devices[chosen_drive].i.events = 0;
822 tst->ss->free_super(tst);
823 continue;
824 }
825 close(fd);
826 devices[chosen_drive].i.events = devices[most_recent].i.events;
827 devices[chosen_drive].uptodate = 1;
828 avail[chosen_drive] = 1;
829 okcnt++;
830 tst->ss->free_super(tst);
831
832 /* If there are any other drives of the same vintage,
833 * add them in as well. We can't lose and we might gain
834 */
835 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
836 int j = best[i];
837 if (j >= 0 &&
838 !devices[j].uptodate &&
839 devices[j].i.events == current_events) {
840 chosen_drive = j;
841 goto add_another;
842 }
843 }
844 }
845
846 /* Now we want to look at the superblock which the kernel will base things on
847 * and compare the devices that we think are working with the devices that the
848 * superblock thinks are working.
849 * If there are differences and --force is given, then update this chosen
850 * superblock.
851 */
852 chosen_drive = -1;
853 st->ss->free_super(st);
854 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
855 int j = best[i];
856 int fd;
857
858 if (j<0)
859 continue;
860 if (!devices[j].uptodate)
861 continue;
862 chosen_drive = j;
863 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
864 fprintf(stderr, Name ": Cannot open %s: %s\n",
865 devices[j].devname, strerror(errno));
866 close(mdfd);
867 return 1;
868 }
869 if (st->ss->load_super(st,fd, NULL)) {
870 close(fd);
871 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
872 devices[j].devname);
873 close(mdfd);
874 return 1;
875 }
876 close(fd);
877 }
878 if (st->sb == NULL) {
879 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
880 close(mdfd);
881 return 1;
882 }
883 st->ss->getinfo_super(st, content);
884 #ifndef MDASSEMBLE
885 sysfs_init(content, mdfd, 0);
886 #endif
887 for (i=0; i<bestcnt; i++) {
888 int j = best[i];
889 unsigned int desired_state;
890
891 if (i < content->array.raid_disks)
892 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
893 else
894 desired_state = 0;
895
896 if (j<0)
897 continue;
898 if (!devices[j].uptodate)
899 continue;
900
901 devices[j].i.disk.state = desired_state;
902 if (!(devices[j].i.array.state & 1))
903 clean = 0;
904
905 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
906 verbose, 0, NULL)) {
907 if (force) {
908 if (verbose >= 0)
909 fprintf(stderr, Name ": "
910 "clearing FAULTY flag for device %d in %s for %s\n",
911 j, mddev, devices[j].devname);
912 change = 1;
913 } else {
914 if (verbose >= -1)
915 fprintf(stderr, Name ": "
916 "device %d in %s has wrong state in superblock, but %s seems ok\n",
917 i, mddev, devices[j].devname);
918 }
919 }
920 #if 0
921 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
922 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
923 i, mddev);
924 }
925 #endif
926 }
927 if (force && !clean &&
928 !enough(content->array.level, content->array.raid_disks,
929 content->array.layout, clean,
930 avail, okcnt)) {
931 change += st->ss->update_super(st, content, "force-array",
932 devices[chosen_drive].devname, verbose,
933 0, NULL);
934 clean = 1;
935 }
936
937 if (change) {
938 int fd;
939 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
940 if (fd < 0) {
941 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
942 devices[chosen_drive].devname);
943 close(mdfd);
944 return 1;
945 }
946 if (st->ss->store_super(st, fd)) {
947 close(fd);
948 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
949 devices[chosen_drive].devname);
950 close(mdfd);
951 return 1;
952 }
953 close(fd);
954 }
955
956 /* If we are in the middle of a reshape we may need to restore saved data
957 * that was moved aside due to the reshape overwriting live data
958 * The code of doing this lives in Grow.c
959 */
960 #ifndef MDASSEMBLE
961 if (content->reshape_active) {
962 int err = 0;
963 int *fdlist = malloc(sizeof(int)* bestcnt);
964 for (i=0; i<bestcnt; i++) {
965 int j = best[i];
966 if (j >= 0) {
967 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
968 if (fdlist[i] < 0) {
969 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
970 devices[j].devname);
971 err = 1;
972 break;
973 }
974 } else
975 fdlist[i] = -1;
976 }
977 if (!err)
978 err = Grow_restart(st, content, fdlist, bestcnt, backup_file);
979 while (i>0) {
980 i--;
981 if (fdlist[i]>=0) close(fdlist[i]);
982 }
983 if (err) {
984 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
985 close(mdfd);
986 return err;
987 }
988 }
989 #endif
990 /* count number of in-sync devices according to the superblock.
991 * We must have this number to start the array without -s or -R
992 */
993 req_cnt = content->array.working_disks;
994
995 /* Almost ready to actually *do* something */
996 if (!old_linux) {
997 int rv;
998
999 /* First, fill in the map, so that udev can find our name
1000 * as soon as we become active.
1001 */
1002 map_update(NULL, fd2devnum(mdfd), content->text_version,
1003 content->uuid, chosen_name);
1004
1005 rv = set_array_info(mdfd, st, content);
1006 if (rv) {
1007 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1008 mddev, strerror(errno));
1009 close(mdfd);
1010 return 1;
1011 }
1012 if (ident->bitmap_fd >= 0) {
1013 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1014 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1015 close(mdfd);
1016 return 1;
1017 }
1018 } else if (ident->bitmap_file) {
1019 /* From config file */
1020 int bmfd = open(ident->bitmap_file, O_RDWR);
1021 if (bmfd < 0) {
1022 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1023 ident->bitmap_file);
1024 close(mdfd);
1025 return 1;
1026 }
1027 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1028 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1029 close(bmfd);
1030 close(mdfd);
1031 return 1;
1032 }
1033 close(bmfd);
1034 }
1035
1036 /* First, add the raid disks, but add the chosen one last */
1037 for (i=0; i<= bestcnt; i++) {
1038 int j;
1039 if (i < bestcnt) {
1040 j = best[i];
1041 if (j == chosen_drive)
1042 continue;
1043 } else
1044 j = chosen_drive;
1045
1046 if (j >= 0 /* && devices[j].uptodate */) {
1047 rv = add_disk(mdfd, st, content, &devices[j].i);
1048
1049 if (rv) {
1050 fprintf(stderr, Name ": failed to add "
1051 "%s to %s: %s\n",
1052 devices[j].devname,
1053 mddev,
1054 strerror(errno));
1055 if (i < content->array.raid_disks
1056 || i == bestcnt)
1057 okcnt--;
1058 else
1059 sparecnt--;
1060 } else if (verbose > 0)
1061 fprintf(stderr, Name ": added %s "
1062 "to %s as %d\n",
1063 devices[j].devname, mddev,
1064 devices[j].i.disk.raid_disk);
1065 } else if (verbose > 0 && i < content->array.raid_disks)
1066 fprintf(stderr, Name ": no uptodate device for "
1067 "slot %d of %s\n",
1068 i, mddev);
1069 }
1070
1071 if (content->array.level == LEVEL_CONTAINER) {
1072 if (verbose >= 0) {
1073 fprintf(stderr, Name ": Container %s has been "
1074 "assembled with %d drive%s",
1075 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1076 if (okcnt < content->array.raid_disks)
1077 fprintf(stderr, " (out of %d)",
1078 content->array.raid_disks);
1079 fprintf(stderr, "\n");
1080 }
1081 sysfs_uevent(content, "change");
1082 wait_for(chosen_name, mdfd);
1083 close(mdfd);
1084 return 0;
1085 }
1086
1087 if (runstop == 1 ||
1088 (runstop <= 0 &&
1089 ( enough(content->array.level, content->array.raid_disks,
1090 content->array.layout, clean, avail, okcnt) &&
1091 (okcnt >= req_cnt || start_partial_ok)
1092 ))) {
1093 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1094 if (verbose >= 0) {
1095 fprintf(stderr, Name ": %s has been started with %d drive%s",
1096 mddev, okcnt, okcnt==1?"":"s");
1097 if (okcnt < content->array.raid_disks)
1098 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1099 if (sparecnt)
1100 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1101 fprintf(stderr, ".\n");
1102 }
1103 if (content->reshape_active &&
1104 content->array.level >= 4 &&
1105 content->array.level <= 6) {
1106 /* might need to increase the size
1107 * of the stripe cache - default is 256
1108 */
1109 if (256 < 4 * (content->array.chunk_size/4096)) {
1110 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1111 if (sra)
1112 sysfs_set_num(sra, NULL,
1113 "stripe_cache_size",
1114 (4 * content->array.chunk_size / 4096) + 1);
1115 }
1116 }
1117 wait_for(mddev, mdfd);
1118 close(mdfd);
1119 if (auto_assem) {
1120 int usecs = 1;
1121 /* There is a nasty race with 'mdadm --monitor'.
1122 * If it opens this device before we close it,
1123 * it gets an incomplete open on which IO
1124 * doesn't work and the capacity is
1125 * wrong.
1126 * If we reopen (to check for layered devices)
1127 * before --monitor closes, we loose.
1128 *
1129 * So: wait upto 1 second for there to be
1130 * a non-zero capacity.
1131 */
1132 while (usecs < 1000) {
1133 mdfd = open(mddev, O_RDONLY);
1134 if (mdfd >= 0) {
1135 unsigned long long size;
1136 if (get_dev_size(mdfd, NULL, &size) &&
1137 size > 0)
1138 break;
1139 close(mdfd);
1140 }
1141 usleep(usecs);
1142 usecs <<= 1;
1143 }
1144 }
1145 return 0;
1146 }
1147 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1148 mddev, strerror(errno));
1149
1150 if (!enough(content->array.level, content->array.raid_disks,
1151 content->array.layout, 1, avail, okcnt))
1152 fprintf(stderr, Name ": Not enough devices to "
1153 "start the array.\n");
1154 else if (!enough(content->array.level,
1155 content->array.raid_disks,
1156 content->array.layout, clean,
1157 avail, okcnt))
1158 fprintf(stderr, Name ": Not enough devices to "
1159 "start the array while not clean "
1160 "- consider --force.\n");
1161
1162 if (auto_assem)
1163 ioctl(mdfd, STOP_ARRAY, NULL);
1164 close(mdfd);
1165 return 1;
1166 }
1167 if (runstop == -1) {
1168 fprintf(stderr, Name ": %s assembled from %d drive%s",
1169 mddev, okcnt, okcnt==1?"":"s");
1170 if (okcnt != content->array.raid_disks)
1171 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1172 fprintf(stderr, ", but not started.\n");
1173 close(mdfd);
1174 return 0;
1175 }
1176 if (verbose >= -1) {
1177 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1178 if (sparecnt)
1179 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1180 if (!enough(content->array.level, content->array.raid_disks,
1181 content->array.layout, 1, avail, okcnt))
1182 fprintf(stderr, " - not enough to start the array.\n");
1183 else if (!enough(content->array.level,
1184 content->array.raid_disks,
1185 content->array.layout, clean,
1186 avail, okcnt))
1187 fprintf(stderr, " - not enough to start the "
1188 "array while not clean - consider "
1189 "--force.\n");
1190 else {
1191 if (req_cnt == content->array.raid_disks)
1192 fprintf(stderr, " - need all %d to start it", req_cnt);
1193 else
1194 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1195 fprintf(stderr, " (use --run to insist).\n");
1196 }
1197 }
1198 if (auto_assem)
1199 ioctl(mdfd, STOP_ARRAY, NULL);
1200 close(mdfd);
1201 return 1;
1202 } else {
1203 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1204 * been updated to point to the current locations of devices.
1205 * so we can just start the array
1206 */
1207 unsigned long dev;
1208 dev = makedev(devices[chosen_drive].i.disk.major,
1209 devices[chosen_drive].i.disk.minor);
1210 if (ioctl(mdfd, START_ARRAY, dev)) {
1211 fprintf(stderr, Name ": Cannot start array: %s\n",
1212 strerror(errno));
1213 }
1214
1215 }
1216 close(mdfd);
1217 return 0;
1218 }
1219
1220 #ifndef MDASSEMBLE
1221 int assemble_container_content(struct supertype *st, int mdfd,
1222 struct mdinfo *content, int runstop,
1223 char *chosen_name, int verbose)
1224 {
1225 struct mdinfo *dev, *sra;
1226 int working = 0, preexist = 0;
1227 struct map_ent *map = NULL;
1228
1229 sysfs_init(content, mdfd, 0);
1230
1231 sra = sysfs_read(mdfd, 0, GET_VERSION);
1232 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1233 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1234 close(mdfd);
1235 return 1;
1236 }
1237 if (sra)
1238 sysfs_free(sra);
1239
1240 for (dev = content->devs; dev; dev = dev->next)
1241 if (sysfs_add_disk(content, dev, 1) == 0)
1242 working++;
1243 else if (errno == EEXIST)
1244 preexist++;
1245 if (working == 0) {
1246 close(mdfd);
1247 return 1;/* Nothing new, don't try to start */
1248 }
1249
1250 map_update(&map, fd2devnum(mdfd),
1251 content->text_version,
1252 content->uuid, chosen_name);
1253
1254 if (runstop > 0 ||
1255 (working + preexist) >= content->array.working_disks) {
1256 int err;
1257
1258 switch(content->array.level) {
1259 case LEVEL_LINEAR:
1260 case LEVEL_MULTIPATH:
1261 case 0:
1262 err = sysfs_set_str(content, NULL, "array_state",
1263 "active");
1264 break;
1265 default:
1266 err = sysfs_set_str(content, NULL, "array_state",
1267 "readonly");
1268 /* start mdmon if needed. */
1269 if (!err) {
1270 if (!mdmon_running(st->container_dev))
1271 start_mdmon(st->container_dev);
1272 ping_monitor(devnum2devname(st->container_dev));
1273 }
1274 break;
1275 }
1276 if (!err)
1277 sysfs_set_safemode(content, content->safe_mode_delay);
1278 if (verbose >= 0) {
1279 if (err)
1280 fprintf(stderr, Name
1281 ": array %s now has %d devices",
1282 chosen_name, working + preexist);
1283 else
1284 fprintf(stderr, Name
1285 ": Started %s with %d devices",
1286 chosen_name, working + preexist);
1287 if (preexist)
1288 fprintf(stderr, " (%d new)", working);
1289 fprintf(stderr, "\n");
1290 }
1291 if (!err)
1292 wait_for(chosen_name, mdfd);
1293 close(mdfd);
1294 return 0;
1295 /* FIXME should have an O_EXCL and wait for read-auto */
1296 } else {
1297 if (verbose >= 0)
1298 fprintf(stderr, Name
1299 ": %s assembled with %d devices but "
1300 "not started\n",
1301 chosen_name, working);
1302 close(mdfd);
1303 return 1;
1304 }
1305 }
1306 #endif
1307
Unnamed repository; edit this file 'description' to name the repository.