]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Allow metadata handlers to free their own superblock.
[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 int Assemble(struct supertype *st, char *mddev, int mdfd,
54 mddev_ident_t ident,
55 mddev_dev_t devlist, char *backup_file,
56 int readonly, int runstop,
57 char *update, char *homehost,
58 int verbose, int force)
59 {
60 /*
61 * The task of Assemble is to find a collection of
62 * devices that should (according to their superblocks)
63 * form an array, and to give this collection to the MD driver.
64 * In Linux-2.4 and later, this involves submitting a
65 * SET_ARRAY_INFO ioctl with no arg - to prepare
66 * the array - and then submit a number of
67 * ADD_NEW_DISK ioctls to add disks into
68 * the array. Finally RUN_ARRAY might
69 * be submitted to start the array.
70 *
71 * Much of the work of Assemble is in finding and/or
72 * checking the disks to make sure they look right.
73 *
74 * If mddev is not set, then scan must be set and we
75 * read through the config file for dev+uuid mapping
76 * We recurse, setting mddev, for each device that
77 * - isn't running
78 * - has a valid uuid (or any uuid if !uuidset)
79 *
80 * If mddev is set, we try to determine state of md.
81 * check version - must be at least 0.90.0
82 * check kernel version. must be at least 2.4.
83 * If not, we can possibly fall back on START_ARRAY
84 * Try to GET_ARRAY_INFO.
85 * If possible, give up
86 * If not, try to STOP_ARRAY just to make sure
87 *
88 * If !uuidset and scan, look in conf-file for uuid
89 * If not found, give up
90 * If !devlist and scan and uuidset, get list of devs from conf-file
91 *
92 * For each device:
93 * Check superblock - discard if bad
94 * Check uuid (set if we don't have one) - discard if no match
95 * Check superblock similarity if we have a superblock - discard if different
96 * Record events, devicenum
97 * This should give us a list of devices for the array
98 * We should collect the most recent event number
99 *
100 * Count disks with recent enough event count
101 * While force && !enough disks
102 * Choose newest rejected disks, update event count
103 * mark clean and rewrite superblock
104 * If recent kernel:
105 * SET_ARRAY_INFO
106 * foreach device with recent events : ADD_NEW_DISK
107 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108 * If old kernel:
109 * Check the device numbers in superblock are right
110 * update superblock if any changes
111 * START_ARRAY
112 *
113 */
114 int clean = 0;
115 int must_close = 0;
116 int old_linux = 0;
117 int vers = 0; /* Keep gcc quite - it really is initialised */
118 void *first_super = NULL, *super = NULL;
119 struct {
120 char *devname;
121 unsigned int major, minor;
122 unsigned int oldmajor, oldminor;
123 long long events;
124 int uptodate;
125 int state;
126 int raid_disk;
127 int disk_nr;
128 } *devices;
129 int *best = NULL; /* indexed by raid_disk */
130 unsigned int bestcnt = 0;
131 int devcnt = 0;
132 unsigned int okcnt, sparecnt;
133 unsigned int req_cnt;
134 unsigned int i;
135 int most_recent = 0;
136 int chosen_drive;
137 int change = 0;
138 int inargv = 0;
139 int bitmap_done;
140 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
141 unsigned int num_devs;
142 mddev_dev_t tmpdev;
143 struct mdinfo info;
144 char *avail;
145 int nextspare = 0;
146
147 if (get_linux_version() < 2004000)
148 old_linux = 1;
149
150 if (mdfd >= 0) {
151 vers = md_get_version(mdfd);
152 if (vers <= 0) {
153 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
154 return 1;
155 }
156 if (vers < 9000) {
157 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
158 " Upgrade your kernel or try --build\n");
159 return 1;
160 }
161
162 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
163 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
164 mddev);
165 return 1;
166 }
167 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
168 }
169 /*
170 * If any subdevs are listed, then any that don't
171 * match ident are discarded. Remainder must all match and
172 * become the array.
173 * If no subdevs, then we scan all devices in the config file, but
174 * there must be something in the identity
175 */
176
177 if (!devlist &&
178 ident->uuid_set == 0 &&
179 ident->super_minor < 0 &&
180 ident->devices == NULL) {
181 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
182 mddev ? mddev : "further assembly");
183 return 1;
184 }
185 if (devlist == NULL)
186 devlist = conf_get_devs();
187 else if (mdfd >= 0)
188 inargv = 1;
189
190 try_again:
191
192 tmpdev = devlist; num_devs = 0;
193 while (tmpdev) {
194 if (tmpdev->used)
195 tmpdev->used = 2;
196 else
197 num_devs++;
198 tmpdev = tmpdev->next;
199 }
200 devices = malloc(num_devs * sizeof(*devices));
201
202 if (!st && ident->st) st = ident->st;
203
204 if (verbose>0)
205 fprintf(stderr, Name ": looking for devices for %s\n",
206 mddev ? mddev : "further assembly");
207
208 /* first walk the list of devices to find a consistent set
209 * that match the criterea, if that is possible.
210 * We flag the one we like with 'used'.
211 */
212 for (tmpdev = devlist;
213 tmpdev;
214 tmpdev = tmpdev->next) {
215 char *devname = tmpdev->devname;
216 int dfd;
217 struct stat stb;
218 struct supertype *tst = st;
219
220 if (tmpdev->used > 1) continue;
221
222 if (ident->devices &&
223 !match_oneof(ident->devices, devname)) {
224 if ((inargv && verbose>=0) || verbose > 0)
225 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
226 continue;
227 }
228
229 dfd = dev_open(devname, O_RDONLY|O_EXCL);
230 if (dfd < 0) {
231 if ((inargv && verbose >= 0) || verbose > 0)
232 fprintf(stderr, Name ": cannot open device %s: %s\n",
233 devname, strerror(errno));
234 tmpdev->used = 2;
235 } else if (fstat(dfd, &stb)< 0) {
236 /* Impossible! */
237 fprintf(stderr, Name ": fstat failed for %s: %s\n",
238 devname, strerror(errno));
239 tmpdev->used = 2;
240 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
241 fprintf(stderr, Name ": %s is not a block device.\n",
242 devname);
243 tmpdev->used = 2;
244 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
245 if ((inargv && verbose >= 0) || verbose > 0)
246 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
247 devname);
248 tmpdev->used = 2;
249 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
250 if ((inargv && verbose >= 0) || verbose > 0)
251 fprintf( stderr, Name ": no RAID superblock on %s\n",
252 devname);
253 } else {
254 tst->ss->getinfo_super(&info, super);
255 }
256 if (dfd >= 0) close(dfd);
257
258 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
259 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
260 if ((inargv && verbose >= 0) || verbose > 0)
261 fprintf(stderr, Name ": %s has wrong uuid.\n",
262 devname);
263 goto loop;
264 }
265 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
266 (!super || name_matches(info.name, ident->name, homehost)==0)) {
267 if ((inargv && verbose >= 0) || verbose > 0)
268 fprintf(stderr, Name ": %s has wrong name.\n",
269 devname);
270 goto loop;
271 }
272 if (ident->super_minor != UnSet &&
273 (!super || ident->super_minor != info.array.md_minor)) {
274 if ((inargv && verbose >= 0) || verbose > 0)
275 fprintf(stderr, Name ": %s has wrong super-minor.\n",
276 devname);
277 goto loop;
278 }
279 if (ident->level != UnSet &&
280 (!super|| ident->level != info.array.level)) {
281 if ((inargv && verbose >= 0) || verbose > 0)
282 fprintf(stderr, Name ": %s has wrong raid level.\n",
283 devname);
284 goto loop;
285 }
286 if (ident->raid_disks != UnSet &&
287 (!super || ident->raid_disks!= info.array.raid_disks)) {
288 if ((inargv && verbose >= 0) || verbose > 0)
289 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
290 devname);
291 goto loop;
292 }
293 if (mdfd < 0) {
294 if (tst == NULL || super == NULL)
295 continue;
296 if (update == NULL &&
297 tst->ss->match_home(super, homehost)==0) {
298 if ((inargv && verbose >= 0) || verbose > 0)
299 fprintf(stderr, Name ": %s is not built for host %s.\n",
300 devname, homehost);
301 /* Auto-assemble, and this is not a usable host */
302 /* if update != NULL, we are updating the host
303 * name... */
304 goto loop;
305 }
306 }
307 /* If we are this far, then we are nearly commited to this device.
308 * If the super_block doesn't exist, or doesn't match others,
309 * then we probably cannot continue
310 * However if one of the arrays is for the homehost, and
311 * the other isn't that can disambiguate.
312 */
313
314 if (!super) {
315 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
316 devname);
317 st->ss->free_super(first_super);
318 return 1;
319 }
320
321 if (st == NULL)
322 st = tst;
323 if (st->ss != tst->ss ||
324 st->minor_version != tst->minor_version ||
325 st->ss->compare_super(&first_super, super) != 0) {
326 /* Some mismatch. If exactly one array matches this host,
327 * we can resolve on that one.
328 * Or, if we are auto assembling, we just ignore the second
329 * for now.
330 */
331 if (mdfd < 0)
332 goto loop;
333 if (homehost) {
334 int first = st->ss->match_home(first_super, homehost);
335 int last = tst->ss->match_home(super, homehost);
336 if (first+last == 1) {
337 /* We can do something */
338 if (first) {/* just ignore this one */
339 if ((inargv && verbose >= 0) || verbose > 0)
340 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
341 devname);
342 goto loop;
343 } else { /* reject all those sofar */
344 mddev_dev_t td;
345 if ((inargv && verbose >= 0) || verbose > 0)
346 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
347 devname);
348 for (td=devlist; td != tmpdev; td=td->next)
349 if (td->used == 1)
350 td->used = 0;
351 tmpdev->used = 1;
352 goto loop;
353 }
354 }
355 }
356 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
357 devname);
358 tst->ss->free_super(super);
359 st->ss->free_super(first_super);
360 return 1;
361 }
362
363 tmpdev->used = 1;
364
365 loop:
366 if (super)
367 tst->ss->free_super(super);
368 super = NULL;
369 }
370
371 if (mdfd < 0) {
372 /* So... it is up to me to open the device.
373 * We create a name '/dev/md/XXX' based on the info in the
374 * superblock, and call open_mddev on that
375 */
376 mdu_array_info_t inf;
377 char *c;
378 if (!first_super) {
379 return 2;
380 }
381 st->ss->getinfo_super(&info, first_super);
382 c = strchr(info.name, ':');
383 if (c) c++; else c= info.name;
384 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
385 /* /dev/md/d0 style for partitionable */
386 asprintf(&mddev, "/dev/md/d%s", c);
387 else
388 asprintf(&mddev, "/dev/md/%s", c);
389 mdfd = open_mddev(mddev, ident->autof);
390 if (mdfd < 0) {
391 st->ss->free_super(first_super);
392 free(devices);
393 first_super = NULL;
394 goto try_again;
395 }
396 vers = md_get_version(mdfd);
397 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
398 for (tmpdev = devlist ;
399 tmpdev && tmpdev->used != 1;
400 tmpdev = tmpdev->next)
401 ;
402 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
403 if (tmpdev)
404 fprintf(stderr, Name ": %s needed for %s...\n",
405 mddev, tmpdev->devname);
406 close(mdfd);
407 mdfd = -1;
408 st->ss->free_super(first_super);
409 free(devices);
410 first_super = NULL;
411 goto try_again;
412 }
413 must_close = 1;
414 }
415
416 /* Ok, no bad inconsistancy, we can try updating etc */
417 bitmap_done = 0;
418 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
419 char *devname = tmpdev->devname;
420 struct stat stb;
421 /* looks like a good enough match to update the super block if needed */
422 #ifndef MDASSEMBLE
423 if (update) {
424 int dfd;
425 /* prepare useful information in info structures */
426 struct stat stb2;
427 fstat(mdfd, &stb2);
428
429 if (strcmp(update, "uuid")==0 &&
430 !ident->uuid_set) {
431 int rfd;
432 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
433 read(rfd, ident->uuid, 16) != 16) {
434 *(__u32*)(ident->uuid) = random();
435 *(__u32*)(ident->uuid+1) = random();
436 *(__u32*)(ident->uuid+2) = random();
437 *(__u32*)(ident->uuid+3) = random();
438 }
439 if (rfd >= 0) close(rfd);
440 }
441 dfd = dev_open(devname, O_RDWR|O_EXCL);
442
443 remove_partitions(dfd);
444
445 st->ss->load_super(st, dfd, &super, NULL);
446 st->ss->getinfo_super(&info, super);
447
448 memcpy(info.uuid, ident->uuid, 16);
449 strcpy(info.name, ident->name);
450 info.array.md_minor = minor(stb2.st_rdev);
451
452 st->ss->update_super(&info, super, update, devname, verbose,
453 ident->uuid_set, homehost);
454 if (strcmp(update, "uuid")==0 &&
455 !ident->uuid_set) {
456 ident->uuid_set = 1;
457 memcpy(ident->uuid, info.uuid, 16);
458 }
459 if (dfd < 0)
460 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
461 devname);
462 else if (st->ss->store_super(st, dfd, super))
463 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
464 devname);
465 if (dfd >= 0)
466 close(dfd);
467
468 if (strcmp(update, "uuid")==0 &&
469 ident->bitmap_fd >= 0 && !bitmap_done) {
470 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid, st->ss->swapuuid) != 0)
471 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
472 else
473 bitmap_done = 1;
474 }
475 } else
476 #endif
477 {
478 int dfd;
479 dfd = dev_open(devname, O_RDWR|O_EXCL);
480
481 remove_partitions(dfd);
482
483 st->ss->load_super(st, dfd, &super, NULL);
484 st->ss->getinfo_super(&info, super);
485 close(dfd);
486 }
487
488 stat(devname, &stb);
489
490 if (verbose > 0)
491 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
492 devname, mddev, info.disk.raid_disk);
493 devices[devcnt].devname = devname;
494 devices[devcnt].major = major(stb.st_rdev);
495 devices[devcnt].minor = minor(stb.st_rdev);
496 devices[devcnt].oldmajor = info.disk.major;
497 devices[devcnt].oldminor = info.disk.minor;
498 devices[devcnt].events = info.events;
499 devices[devcnt].raid_disk = info.disk.raid_disk;
500 devices[devcnt].disk_nr = info.disk.number;
501 devices[devcnt].uptodate = 0;
502 devices[devcnt].state = info.disk.state;
503 if (most_recent < devcnt) {
504 if (devices[devcnt].events
505 > devices[most_recent].events)
506 most_recent = devcnt;
507 }
508 if (info.array.level == -4)
509 /* with multipath, the raid_disk from the superblock is meaningless */
510 i = devcnt;
511 else
512 i = devices[devcnt].raid_disk;
513 if (i+1 == 0) {
514 if (nextspare < info.array.raid_disks)
515 nextspare = info.array.raid_disks;
516 i = nextspare++;
517 } else {
518 if (i >= info.array.raid_disks &&
519 i >= nextspare)
520 nextspare = i+1;
521 }
522 if (i < 10000) {
523 if (i >= bestcnt) {
524 unsigned int newbestcnt = i+10;
525 int *newbest = malloc(sizeof(int)*newbestcnt);
526 unsigned int c;
527 for (c=0; c < newbestcnt; c++)
528 if (c < bestcnt)
529 newbest[c] = best[c];
530 else
531 newbest[c] = -1;
532 if (best)free(best);
533 best = newbest;
534 bestcnt = newbestcnt;
535 }
536 if (best[i] >=0 &&
537 devices[best[i]].events == devices[devcnt].events &&
538 devices[best[i]].minor != devices[devcnt].minor &&
539 st->ss->major == 0 &&
540 info.array.level != -4) {
541 /* two different devices with identical superblock.
542 * Could be a mis-detection caused by overlapping
543 * partitions. fail-safe.
544 */
545 fprintf(stderr, Name ": WARNING %s and %s appear"
546 " to have very similar superblocks.\n"
547 " If they are really different, "
548 "please --zero the superblock on one\n"
549 " If they are the same or overlap,"
550 " please remove one from %s.\n",
551 devices[best[i]].devname, devname,
552 inargv ? "the list" :
553 "the\n DEVICE list in mdadm.conf"
554 );
555 if (must_close) close(mdfd);
556 return 1;
557 }
558 if (best[i] == -1
559 || devices[best[i]].events < devices[devcnt].events)
560 best[i] = devcnt;
561 }
562 devcnt++;
563
564 if (super)
565 st->ss->free_super(super);
566 super = NULL;
567 }
568
569 if (update && strcmp(update, "byteorder")==0)
570 st->minor_version = 90;
571
572 if (devcnt == 0) {
573 fprintf(stderr, Name ": no devices found for %s\n",
574 mddev);
575 st->ss->free_super(first_super);
576 if (must_close) close(mdfd);
577 return 1;
578 }
579
580 st->ss->getinfo_super(&info, first_super);
581 clean = info.array.state & 1;
582
583 /* now we have some devices that might be suitable.
584 * I wonder how many
585 */
586 avail = malloc(info.array.raid_disks);
587 memset(avail, 0, info.array.raid_disks);
588 okcnt = 0;
589 sparecnt=0;
590 for (i=0; i< bestcnt ;i++) {
591 int j = best[i];
592 int event_margin = 1; /* always allow a difference of '1'
593 * like the kernel does
594 */
595 if (j < 0) continue;
596 /* note: we ignore error flags in multipath arrays
597 * as they don't make sense
598 */
599 if (info.array.level != -4)
600 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
601 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
602 sparecnt++;
603 continue;
604 }
605 if (devices[j].events+event_margin >=
606 devices[most_recent].events) {
607 devices[j].uptodate = 1;
608 if (i < info.array.raid_disks) {
609 okcnt++;
610 avail[i]=1;
611 } else
612 sparecnt++;
613 }
614 }
615 while (force && !enough(info.array.level, info.array.raid_disks,
616 info.array.layout, 1,
617 avail, okcnt)) {
618 /* Choose the newest best drive which is
619 * not up-to-date, update the superblock
620 * and add it.
621 */
622 int fd;
623 long long current_events;
624 chosen_drive = -1;
625 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
626 int j = best[i];
627 if (j>=0 &&
628 !devices[j].uptodate &&
629 devices[j].events > 0 &&
630 (chosen_drive < 0 ||
631 devices[j].events > devices[chosen_drive].events))
632 chosen_drive = j;
633 }
634 if (chosen_drive < 0)
635 break;
636 current_events = devices[chosen_drive].events;
637 add_another:
638 if (verbose >= 0)
639 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
640 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
641 (int)(devices[chosen_drive].events),
642 (int)(devices[most_recent].events));
643 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
644 if (fd < 0) {
645 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
646 devices[chosen_drive].devname);
647 devices[chosen_drive].events = 0;
648 continue;
649 }
650 if (st->ss->load_super(st,fd, &super, NULL)) {
651 close(fd);
652 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
653 devices[chosen_drive].devname);
654 devices[chosen_drive].events = 0;
655 continue;
656 }
657 info.events = devices[most_recent].events;
658 st->ss->update_super(&info, super, "force-one",
659 devices[chosen_drive].devname, verbose,
660 0, NULL);
661
662 if (st->ss->store_super(st, fd, super)) {
663 close(fd);
664 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
665 devices[chosen_drive].devname);
666 devices[chosen_drive].events = 0;
667 st->ss->free_super(super);
668 continue;
669 }
670 close(fd);
671 devices[chosen_drive].events = devices[most_recent].events;
672 devices[chosen_drive].uptodate = 1;
673 avail[chosen_drive] = 1;
674 okcnt++;
675 st->ss->free_super(super);
676
677 /* If there are any other drives of the same vintage,
678 * add them in as well. We can't lose and we might gain
679 */
680 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
681 int j = best[i];
682 if (j >= 0 &&
683 !devices[j].uptodate &&
684 devices[j].events > 0 &&
685 devices[j].events == current_events) {
686 chosen_drive = j;
687 goto add_another;
688 }
689 }
690 }
691
692 /* Now we want to look at the superblock which the kernel will base things on
693 * and compare the devices that we think are working with the devices that the
694 * superblock thinks are working.
695 * If there are differences and --force is given, then update this chosen
696 * superblock.
697 */
698 chosen_drive = -1;
699 super = NULL;
700 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
701 int j = best[i];
702 int fd;
703
704 if (j<0)
705 continue;
706 if (!devices[j].uptodate)
707 continue;
708 chosen_drive = j;
709 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
710 fprintf(stderr, Name ": Cannot open %s: %s\n",
711 devices[j].devname, strerror(errno));
712 if (must_close) close(mdfd);
713 return 1;
714 }
715 if (st->ss->load_super(st,fd, &super, NULL)) {
716 close(fd);
717 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
718 devices[j].devname);
719 if (must_close) close(mdfd);
720 return 1;
721 }
722 close(fd);
723 }
724 if (super == NULL) {
725 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
726 if (must_close) close(mdfd);
727 return 1;
728 }
729 st->ss->getinfo_super(&info, super);
730 for (i=0; i<bestcnt; i++) {
731 int j = best[i];
732 unsigned int desired_state;
733
734 if (i < info.array.raid_disks)
735 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
736 else
737 desired_state = 0;
738
739 if (j<0)
740 continue;
741 if (!devices[j].uptodate)
742 continue;
743 info.disk.number = devices[j].disk_nr;
744 info.disk.raid_disk = i;
745 info.disk.state = desired_state;
746
747 if (devices[j].uptodate &&
748 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
749 if (force) {
750 if (verbose >= 0)
751 fprintf(stderr, Name ": "
752 "clearing FAULTY flag for device %d in %s for %s\n",
753 j, mddev, devices[j].devname);
754 change = 1;
755 } else {
756 if (verbose >= -1)
757 fprintf(stderr, Name ": "
758 "device %d in %s has wrong state in superblock, but %s seems ok\n",
759 i, mddev, devices[j].devname);
760 }
761 }
762 #if 0
763 if (!devices[j].uptodate &&
764 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
765 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
766 i, mddev);
767 }
768 #endif
769 }
770 if (force && !clean &&
771 !enough(info.array.level, info.array.raid_disks,
772 info.array.layout, clean,
773 avail, okcnt)) {
774 change += st->ss->update_super(&info, super, "force-array",
775 devices[chosen_drive].devname, verbose,
776 0, NULL);
777 clean = 1;
778 }
779
780 if (change) {
781 int fd;
782 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
783 if (fd < 0) {
784 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
785 devices[chosen_drive].devname);
786 if (must_close) close(mdfd);
787 return 1;
788 }
789 if (st->ss->store_super(st, fd, super)) {
790 close(fd);
791 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
792 devices[chosen_drive].devname);
793 if (must_close) close(mdfd);
794 return 1;
795 }
796 close(fd);
797 }
798
799 /* If we are in the middle of a reshape we may need to restore saved data
800 * that was moved aside due to the reshape overwriting live data
801 * The code of doing this lives in Grow.c
802 */
803 #ifndef MDASSEMBLE
804 if (info.reshape_active) {
805 int err = 0;
806 int *fdlist = malloc(sizeof(int)* bestcnt);
807 for (i=0; i<bestcnt; i++) {
808 int j = best[i];
809 if (j >= 0) {
810 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
811 if (fdlist[i] < 0) {
812 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
813 devices[j].devname);
814 err = 1;
815 break;
816 }
817 } else
818 fdlist[i] = -1;
819 }
820 if (!err)
821 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
822 while (i>0) {
823 i--;
824 if (fdlist[i]>=0) close(fdlist[i]);
825 }
826 if (err) {
827 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
828 if (must_close) close(mdfd);
829 return err;
830 }
831 }
832 #endif
833 /* count number of in-sync devices according to the superblock.
834 * We must have this number to start the array without -s or -R
835 */
836 req_cnt = info.array.working_disks;
837
838 /* Almost ready to actually *do* something */
839 if (!old_linux) {
840 int rv;
841 if ((vers % 100) >= 1) { /* can use different versions */
842 mdu_array_info_t inf;
843 memset(&inf, 0, sizeof(inf));
844 inf.major_version = st->ss->major;
845 inf.minor_version = st->minor_version;
846 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
847 } else
848 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
849
850 if (rv) {
851 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
852 mddev, strerror(errno));
853 if (must_close) close(mdfd);
854 return 1;
855 }
856 if (ident->bitmap_fd >= 0) {
857 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
858 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
859 if (must_close) close(mdfd);
860 return 1;
861 }
862 } else if (ident->bitmap_file) {
863 /* From config file */
864 int bmfd = open(ident->bitmap_file, O_RDWR);
865 if (bmfd < 0) {
866 fprintf(stderr, Name ": Could not open bitmap file %s\n",
867 ident->bitmap_file);
868 if (must_close) close(mdfd);
869 return 1;
870 }
871 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
872 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
873 close(bmfd);
874 if (must_close) close(mdfd);
875 return 1;
876 }
877 close(bmfd);
878 }
879
880 /* First, add the raid disks, but add the chosen one last */
881 for (i=0; i<= bestcnt; i++) {
882 int j;
883 if (i < bestcnt) {
884 j = best[i];
885 if (j == chosen_drive)
886 continue;
887 } else
888 j = chosen_drive;
889
890 if (j >= 0 /* && devices[j].uptodate */) {
891 mdu_disk_info_t disk;
892 memset(&disk, 0, sizeof(disk));
893 disk.major = devices[j].major;
894 disk.minor = devices[j].minor;
895 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
896 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
897 devices[j].devname,
898 mddev,
899 strerror(errno));
900 if (i < info.array.raid_disks || i == bestcnt)
901 okcnt--;
902 else
903 sparecnt--;
904 } else if (verbose > 0)
905 fprintf(stderr, Name ": added %s to %s as %d\n",
906 devices[j].devname, mddev, devices[j].raid_disk);
907 } else if (verbose > 0 && i < info.array.raid_disks)
908 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
909 i, mddev);
910 }
911
912 if (runstop == 1 ||
913 (runstop <= 0 &&
914 ( enough(info.array.level, info.array.raid_disks,
915 info.array.layout, clean, avail, okcnt) &&
916 (okcnt >= req_cnt || start_partial_ok)
917 ))) {
918 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
919 if (verbose >= 0) {
920 fprintf(stderr, Name ": %s has been started with %d drive%s",
921 mddev, okcnt, okcnt==1?"":"s");
922 if (okcnt < info.array.raid_disks)
923 fprintf(stderr, " (out of %d)", info.array.raid_disks);
924 if (sparecnt)
925 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
926 fprintf(stderr, ".\n");
927 }
928 if (must_close) {
929 int usecs = 1;
930 close(mdfd);
931 /* There is a nasty race with 'mdadm --monitor'.
932 * If it opens this device before we close it,
933 * it gets an incomplete open on which IO
934 * doesn't work and the capacity if wrong.
935 * If we reopen (to check for layered devices)
936 * before --monitor closes, we loose.
937 *
938 * So: wait upto 1 second for there to be
939 * a non-zero capacity.
940 */
941 while (usecs < 1000) {
942 mdfd = open(mddev, O_RDONLY);
943 if (mdfd >= 0) {
944 unsigned long long size;
945 if (get_dev_size(mdfd, NULL, &size) &&
946 size > 0)
947 break;
948 close(mdfd);
949 }
950 usleep(usecs);
951 usecs <<= 1;
952 }
953 }
954 return 0;
955 }
956 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
957 mddev, strerror(errno));
958
959 if (!enough(info.array.level, info.array.raid_disks,
960 info.array.layout, 1, avail, okcnt))
961 fprintf(stderr, Name ": Not enough devices to "
962 "start the array.\n");
963 else if (!enough(info.array.level,
964 info.array.raid_disks,
965 info.array.layout, clean,
966 avail, okcnt))
967 fprintf(stderr, Name ": Not enough devices to "
968 "start the array while not clean "
969 "- consider --force.\n");
970
971 if (must_close) close(mdfd);
972 return 1;
973 }
974 if (runstop == -1) {
975 fprintf(stderr, Name ": %s assembled from %d drive%s",
976 mddev, okcnt, okcnt==1?"":"s");
977 if (okcnt != info.array.raid_disks)
978 fprintf(stderr, " (out of %d)", info.array.raid_disks);
979 fprintf(stderr, ", but not started.\n");
980 if (must_close) close(mdfd);
981 return 0;
982 }
983 if (verbose >= -1) {
984 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
985 if (sparecnt)
986 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
987 if (!enough(info.array.level, info.array.raid_disks,
988 info.array.layout, 1, avail, okcnt))
989 fprintf(stderr, " - not enough to start the array.\n");
990 else if (!enough(info.array.level,
991 info.array.raid_disks,
992 info.array.layout, clean,
993 avail, okcnt))
994 fprintf(stderr, " - not enough to start the "
995 "array while not clean - consider "
996 "--force.\n");
997 else {
998 if (req_cnt == info.array.raid_disks)
999 fprintf(stderr, " - need all %d to start it", req_cnt);
1000 else
1001 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1002 fprintf(stderr, " (use --run to insist).\n");
1003 }
1004 }
1005 if (must_close) close(mdfd);
1006 return 1;
1007 } else {
1008 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1009 * been updated to point to the current locations of devices.
1010 * so we can just start the array
1011 */
1012 unsigned long dev;
1013 dev = makedev(devices[chosen_drive].major,
1014 devices[chosen_drive].minor);
1015 if (ioctl(mdfd, START_ARRAY, dev)) {
1016 fprintf(stderr, Name ": Cannot start array: %s\n",
1017 strerror(errno));
1018 }
1019
1020 }
1021 if (must_close) close(mdfd);
1022 return 0;
1023 }