]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Don't record oldmajor/oldminor any more during assemble.
[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 struct {
119 char *devname;
120 unsigned int major, minor;
121 long long events;
122 int uptodate;
123 int state;
124 int raid_disk;
125 int disk_nr;
126 } *devices;
127 int *best = NULL; /* indexed by raid_disk */
128 unsigned int bestcnt = 0;
129 int devcnt = 0;
130 unsigned int okcnt, sparecnt;
131 unsigned int req_cnt;
132 unsigned int i;
133 int most_recent = 0;
134 int chosen_drive;
135 int change = 0;
136 int inargv = 0;
137 int bitmap_done;
138 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
139 unsigned int num_devs;
140 mddev_dev_t tmpdev;
141 struct mdinfo info;
142 char *avail;
143 int nextspare = 0;
144
145 if (get_linux_version() < 2004000)
146 old_linux = 1;
147
148 if (mdfd >= 0) {
149 vers = md_get_version(mdfd);
150 if (vers <= 0) {
151 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
152 return 1;
153 }
154 if (vers < 9000) {
155 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
156 " Upgrade your kernel or try --build\n");
157 return 1;
158 }
159
160 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
161 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
162 mddev);
163 return 1;
164 }
165 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
166 }
167 /*
168 * If any subdevs are listed, then any that don't
169 * match ident are discarded. Remainder must all match and
170 * become the array.
171 * If no subdevs, then we scan all devices in the config file, but
172 * there must be something in the identity
173 */
174
175 if (!devlist &&
176 ident->uuid_set == 0 &&
177 ident->super_minor < 0 &&
178 ident->devices == NULL) {
179 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
180 mddev ? mddev : "further assembly");
181 return 1;
182 }
183 if (devlist == NULL)
184 devlist = conf_get_devs();
185 else if (mdfd >= 0)
186 inargv = 1;
187
188 try_again:
189
190 tmpdev = devlist; num_devs = 0;
191 while (tmpdev) {
192 if (tmpdev->used)
193 tmpdev->used = 2;
194 else
195 num_devs++;
196 tmpdev = tmpdev->next;
197 }
198 devices = malloc(num_devs * sizeof(*devices));
199
200 if (!st && ident->st) st = ident->st;
201
202 if (verbose>0)
203 fprintf(stderr, Name ": looking for devices for %s\n",
204 mddev ? mddev : "further assembly");
205
206 /* first walk the list of devices to find a consistent set
207 * that match the criterea, if that is possible.
208 * We flag the one we like with 'used'.
209 */
210 for (tmpdev = devlist;
211 tmpdev;
212 tmpdev = tmpdev->next) {
213 char *devname = tmpdev->devname;
214 int dfd;
215 struct stat stb;
216 struct supertype *tst = dup_super(st);
217
218 if (tmpdev->used > 1) continue;
219
220 if (ident->devices &&
221 !match_oneof(ident->devices, devname)) {
222 if ((inargv && verbose>=0) || verbose > 0)
223 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
224 continue;
225 }
226
227 dfd = dev_open(devname, O_RDONLY|O_EXCL);
228 if (dfd < 0) {
229 if ((inargv && verbose >= 0) || verbose > 0)
230 fprintf(stderr, Name ": cannot open device %s: %s\n",
231 devname, strerror(errno));
232 tmpdev->used = 2;
233 } else if (fstat(dfd, &stb)< 0) {
234 /* Impossible! */
235 fprintf(stderr, Name ": fstat failed for %s: %s\n",
236 devname, strerror(errno));
237 tmpdev->used = 2;
238 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
239 fprintf(stderr, Name ": %s is not a block device.\n",
240 devname);
241 tmpdev->used = 2;
242 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
243 if ((inargv && verbose >= 0) || verbose > 0)
244 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
245 devname);
246 tmpdev->used = 2;
247 } else if (tst->ss->load_super(tst,dfd, NULL)) {
248 if ((inargv && verbose >= 0) || verbose > 0)
249 fprintf( stderr, Name ": no RAID superblock on %s\n",
250 devname);
251 } else {
252 tst->ss->getinfo_super(tst, &info);
253 }
254 if (dfd >= 0) close(dfd);
255
256 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
257 (!tst || !tst->sb ||
258 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
259 if ((inargv && verbose >= 0) || verbose > 0)
260 fprintf(stderr, Name ": %s has wrong uuid.\n",
261 devname);
262 goto loop;
263 }
264 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
265 (!tst || !tst->sb ||
266 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 (!tst || !tst->sb ||
274 ident->super_minor != info.array.md_minor)) {
275 if ((inargv && verbose >= 0) || verbose > 0)
276 fprintf(stderr, Name ": %s has wrong super-minor.\n",
277 devname);
278 goto loop;
279 }
280 if (ident->level != UnSet &&
281 (!tst || !tst->sb ||
282 ident->level != info.array.level)) {
283 if ((inargv && verbose >= 0) || verbose > 0)
284 fprintf(stderr, Name ": %s has wrong raid level.\n",
285 devname);
286 goto loop;
287 }
288 if (ident->raid_disks != UnSet &&
289 (!tst || !tst->sb ||
290 ident->raid_disks!= info.array.raid_disks)) {
291 if ((inargv && verbose >= 0) || verbose > 0)
292 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
293 devname);
294 goto loop;
295 }
296 if (mdfd < 0) {
297 if (tst == NULL || tst->sb == NULL)
298 continue;
299 if (update == NULL &&
300 tst->ss->match_home(tst, homehost)==0) {
301 if ((inargv && verbose >= 0) || verbose > 0)
302 fprintf(stderr, Name ": %s is not built for host %s.\n",
303 devname, homehost);
304 /* Auto-assemble, and this is not a usable host */
305 /* if update != NULL, we are updating the host
306 * name... */
307 goto loop;
308 }
309 }
310 /* If we are this far, then we are nearly commited to this device.
311 * If the super_block doesn't exist, or doesn't match others,
312 * then we probably cannot continue
313 * However if one of the arrays is for the homehost, and
314 * the other isn't that can disambiguate.
315 */
316
317 if (!tst || !tst->sb) {
318 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
319 devname);
320 st->ss->free_super(st);
321 return 1;
322 }
323
324 if (st == NULL)
325 st = dup_super(tst);
326 if (st->minor_version == -1)
327 st->minor_version = tst->minor_version;
328 if (st->ss != tst->ss ||
329 st->minor_version != tst->minor_version ||
330 st->ss->compare_super(st, tst) != 0) {
331 /* Some mismatch. If exactly one array matches this host,
332 * we can resolve on that one.
333 * Or, if we are auto assembling, we just ignore the second
334 * for now.
335 */
336 if (mdfd < 0)
337 goto loop;
338 if (homehost) {
339 int first = st->ss->match_home(st, homehost);
340 int last = tst->ss->match_home(tst, homehost);
341 if (first+last == 1) {
342 /* We can do something */
343 if (first) {/* just ignore this one */
344 if ((inargv && verbose >= 0) || verbose > 0)
345 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
346 devname);
347 goto loop;
348 } else { /* reject all those sofar */
349 mddev_dev_t td;
350 if ((inargv && verbose >= 0) || verbose > 0)
351 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
352 devname);
353 for (td=devlist; td != tmpdev; td=td->next)
354 if (td->used == 1)
355 td->used = 0;
356 tmpdev->used = 1;
357 goto loop;
358 }
359 }
360 }
361 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
362 devname);
363 tst->ss->free_super(tst);
364 st->ss->free_super(st);
365 return 1;
366 }
367
368 tmpdev->used = 1;
369
370 loop:
371 tst->ss->free_super(tst);
372 }
373
374 if (mdfd < 0) {
375 /* So... it is up to me to open the device.
376 * We create a name '/dev/md/XXX' based on the info in the
377 * superblock, and call open_mddev on that
378 */
379 mdu_array_info_t inf;
380 char *c;
381 if (!st->sb) {
382 return 2;
383 }
384 st->ss->getinfo_super(st, &info);
385 c = strchr(info.name, ':');
386 if (c) c++; else c= info.name;
387 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
388 /* /dev/md/d0 style for partitionable */
389 asprintf(&mddev, "/dev/md/d%s", c);
390 else
391 asprintf(&mddev, "/dev/md/%s", c);
392 mdfd = open_mddev(mddev, ident->autof);
393 if (mdfd < 0) {
394 st->ss->free_super(st);
395 free(devices);
396 goto try_again;
397 }
398 vers = md_get_version(mdfd);
399 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
400 for (tmpdev = devlist ;
401 tmpdev && tmpdev->used != 1;
402 tmpdev = tmpdev->next)
403 ;
404 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
405 if (tmpdev)
406 fprintf(stderr, Name ": %s needed for %s...\n",
407 mddev, tmpdev->devname);
408 close(mdfd);
409 mdfd = -1;
410 st->ss->free_super(st);
411 free(devices);
412 goto try_again;
413 }
414 must_close = 1;
415 }
416
417 /* Ok, no bad inconsistancy, we can try updating etc */
418 bitmap_done = 0;
419 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
420 char *devname = tmpdev->devname;
421 struct stat stb;
422 /* looks like a good enough match to update the super block if needed */
423 #ifndef MDASSEMBLE
424 if (update) {
425 int dfd;
426 /* prepare useful information in info structures */
427 struct stat stb2;
428 struct supertype *tst;
429 fstat(mdfd, &stb2);
430
431 if (strcmp(update, "uuid")==0 &&
432 !ident->uuid_set) {
433 int rfd;
434 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
435 read(rfd, ident->uuid, 16) != 16) {
436 *(__u32*)(ident->uuid) = random();
437 *(__u32*)(ident->uuid+1) = random();
438 *(__u32*)(ident->uuid+2) = random();
439 *(__u32*)(ident->uuid+3) = random();
440 }
441 if (rfd >= 0) close(rfd);
442 }
443 dfd = dev_open(devname, O_RDWR|O_EXCL);
444
445 remove_partitions(dfd);
446
447 tst = dup_super(st);
448 tst->ss->load_super(tst, dfd, NULL);
449 tst->ss->getinfo_super(tst, &info);
450
451 memcpy(info.uuid, ident->uuid, 16);
452 strcpy(info.name, ident->name);
453 info.array.md_minor = minor(stb2.st_rdev);
454
455 tst->ss->update_super(tst, &info, update,
456 devname, verbose,
457 ident->uuid_set, homehost);
458 if (strcmp(update, "uuid")==0 &&
459 !ident->uuid_set) {
460 ident->uuid_set = 1;
461 memcpy(ident->uuid, info.uuid, 16);
462 }
463 if (dfd < 0)
464 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
465 devname);
466 else if (tst->ss->store_super(tst, dfd))
467 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
468 devname);
469 if (dfd >= 0)
470 close(dfd);
471
472 if (strcmp(update, "uuid")==0 &&
473 ident->bitmap_fd >= 0 && !bitmap_done) {
474 if (bitmap_update_uuid(ident->bitmap_fd,
475 info.uuid,
476 tst->ss->swapuuid) != 0)
477 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
478 else
479 bitmap_done = 1;
480 }
481 tst->ss->free_super(tst);
482 } else
483 #endif
484 {
485 struct supertype *tst = dup_super(st);;
486 int dfd;
487 dfd = dev_open(devname, O_RDWR|O_EXCL);
488
489 remove_partitions(dfd);
490
491 tst->ss->load_super(tst, dfd, NULL);
492 tst->ss->getinfo_super(tst, &info);
493 tst->ss->free_super(tst);
494 close(dfd);
495 }
496
497 stat(devname, &stb);
498
499 if (verbose > 0)
500 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
501 devname, mddev, info.disk.raid_disk);
502 devices[devcnt].devname = devname;
503 devices[devcnt].major = major(stb.st_rdev);
504 devices[devcnt].minor = minor(stb.st_rdev);
505 devices[devcnt].events = info.events;
506 devices[devcnt].raid_disk = info.disk.raid_disk;
507 devices[devcnt].disk_nr = info.disk.number;
508 devices[devcnt].uptodate = 0;
509 devices[devcnt].state = info.disk.state;
510 if (most_recent < devcnt) {
511 if (devices[devcnt].events
512 > devices[most_recent].events)
513 most_recent = devcnt;
514 }
515 if (info.array.level == -4)
516 /* with multipath, the raid_disk from the superblock is meaningless */
517 i = devcnt;
518 else
519 i = devices[devcnt].raid_disk;
520 if (i+1 == 0) {
521 if (nextspare < info.array.raid_disks)
522 nextspare = info.array.raid_disks;
523 i = nextspare++;
524 } else {
525 if (i >= info.array.raid_disks &&
526 i >= nextspare)
527 nextspare = i+1;
528 }
529 if (i < 10000) {
530 if (i >= bestcnt) {
531 unsigned int newbestcnt = i+10;
532 int *newbest = malloc(sizeof(int)*newbestcnt);
533 unsigned int c;
534 for (c=0; c < newbestcnt; c++)
535 if (c < bestcnt)
536 newbest[c] = best[c];
537 else
538 newbest[c] = -1;
539 if (best)free(best);
540 best = newbest;
541 bestcnt = newbestcnt;
542 }
543 if (best[i] >=0 &&
544 devices[best[i]].events == devices[devcnt].events &&
545 devices[best[i]].minor != devices[devcnt].minor &&
546 st->ss->major == 0 &&
547 info.array.level != -4) {
548 /* two different devices with identical superblock.
549 * Could be a mis-detection caused by overlapping
550 * partitions. fail-safe.
551 */
552 fprintf(stderr, Name ": WARNING %s and %s appear"
553 " to have very similar superblocks.\n"
554 " If they are really different, "
555 "please --zero the superblock on one\n"
556 " If they are the same or overlap,"
557 " please remove one from %s.\n",
558 devices[best[i]].devname, devname,
559 inargv ? "the list" :
560 "the\n DEVICE list in mdadm.conf"
561 );
562 if (must_close) close(mdfd);
563 return 1;
564 }
565 if (best[i] == -1
566 || devices[best[i]].events < devices[devcnt].events)
567 best[i] = devcnt;
568 }
569 devcnt++;
570 }
571
572 if (update && strcmp(update, "byteorder")==0)
573 st->minor_version = 90;
574
575 if (devcnt == 0) {
576 fprintf(stderr, Name ": no devices found for %s\n",
577 mddev);
578 st->ss->free_super(st);
579 if (must_close) close(mdfd);
580 return 1;
581 }
582
583 st->ss->getinfo_super(st, &info);
584 clean = info.array.state & 1;
585
586 /* now we have some devices that might be suitable.
587 * I wonder how many
588 */
589 avail = malloc(info.array.raid_disks);
590 memset(avail, 0, info.array.raid_disks);
591 okcnt = 0;
592 sparecnt=0;
593 for (i=0; i< bestcnt ;i++) {
594 int j = best[i];
595 int event_margin = 1; /* always allow a difference of '1'
596 * like the kernel does
597 */
598 if (j < 0) continue;
599 /* note: we ignore error flags in multipath arrays
600 * as they don't make sense
601 */
602 if (info.array.level != -4)
603 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
604 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
605 sparecnt++;
606 continue;
607 }
608 if (devices[j].events+event_margin >=
609 devices[most_recent].events) {
610 devices[j].uptodate = 1;
611 if (i < info.array.raid_disks) {
612 okcnt++;
613 avail[i]=1;
614 } else
615 sparecnt++;
616 }
617 }
618 while (force && !enough(info.array.level, info.array.raid_disks,
619 info.array.layout, 1,
620 avail, okcnt)) {
621 /* Choose the newest best drive which is
622 * not up-to-date, update the superblock
623 * and add it.
624 */
625 int fd;
626 struct supertype *tst;
627 long long current_events;
628 chosen_drive = -1;
629 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
630 int j = best[i];
631 if (j>=0 &&
632 !devices[j].uptodate &&
633 devices[j].events > 0 &&
634 (chosen_drive < 0 ||
635 devices[j].events > devices[chosen_drive].events))
636 chosen_drive = j;
637 }
638 if (chosen_drive < 0)
639 break;
640 current_events = devices[chosen_drive].events;
641 add_another:
642 if (verbose >= 0)
643 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
644 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
645 (int)(devices[chosen_drive].events),
646 (int)(devices[most_recent].events));
647 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
648 if (fd < 0) {
649 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
650 devices[chosen_drive].devname);
651 devices[chosen_drive].events = 0;
652 continue;
653 }
654 tst = dup_super(st);
655 if (tst->ss->load_super(st,fd, NULL)) {
656 close(fd);
657 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
658 devices[chosen_drive].devname);
659 devices[chosen_drive].events = 0;
660 continue;
661 }
662 info.events = devices[most_recent].events;
663 tst->ss->update_super(tst, &info, "force-one",
664 devices[chosen_drive].devname, verbose,
665 0, NULL);
666
667 if (tst->ss->store_super(tst, fd)) {
668 close(fd);
669 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
670 devices[chosen_drive].devname);
671 devices[chosen_drive].events = 0;
672 tst->ss->free_super(tst);
673 continue;
674 }
675 close(fd);
676 devices[chosen_drive].events = devices[most_recent].events;
677 devices[chosen_drive].uptodate = 1;
678 avail[chosen_drive] = 1;
679 okcnt++;
680 tst->ss->free_super(tst);
681
682 /* If there are any other drives of the same vintage,
683 * add them in as well. We can't lose and we might gain
684 */
685 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
686 int j = best[i];
687 if (j >= 0 &&
688 !devices[j].uptodate &&
689 devices[j].events > 0 &&
690 devices[j].events == current_events) {
691 chosen_drive = j;
692 goto add_another;
693 }
694 }
695 }
696
697 /* Now we want to look at the superblock which the kernel will base things on
698 * and compare the devices that we think are working with the devices that the
699 * superblock thinks are working.
700 * If there are differences and --force is given, then update this chosen
701 * superblock.
702 */
703 chosen_drive = -1;
704 st->ss->free_super(st);
705 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
706 int j = best[i];
707 int fd;
708
709 if (j<0)
710 continue;
711 if (!devices[j].uptodate)
712 continue;
713 chosen_drive = j;
714 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
715 fprintf(stderr, Name ": Cannot open %s: %s\n",
716 devices[j].devname, strerror(errno));
717 if (must_close) close(mdfd);
718 return 1;
719 }
720 if (st->ss->load_super(st,fd, NULL)) {
721 close(fd);
722 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
723 devices[j].devname);
724 if (must_close) close(mdfd);
725 return 1;
726 }
727 close(fd);
728 }
729 if (st->sb == NULL) {
730 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
731 if (must_close) close(mdfd);
732 return 1;
733 }
734 st->ss->getinfo_super(st, &info);
735 for (i=0; i<bestcnt; i++) {
736 int j = best[i];
737 unsigned int desired_state;
738
739 if (i < info.array.raid_disks)
740 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
741 else
742 desired_state = 0;
743
744 if (j<0)
745 continue;
746 if (!devices[j].uptodate)
747 continue;
748 info.disk.number = devices[j].disk_nr;
749 info.disk.raid_disk = i;
750 info.disk.state = desired_state;
751
752 if (devices[j].uptodate &&
753 st->ss->update_super(st, &info, "assemble", NULL,
754 verbose, 0, NULL)) {
755 if (force) {
756 if (verbose >= 0)
757 fprintf(stderr, Name ": "
758 "clearing FAULTY flag for device %d in %s for %s\n",
759 j, mddev, devices[j].devname);
760 change = 1;
761 } else {
762 if (verbose >= -1)
763 fprintf(stderr, Name ": "
764 "device %d in %s has wrong state in superblock, but %s seems ok\n",
765 i, mddev, devices[j].devname);
766 }
767 }
768 #if 0
769 if (!devices[j].uptodate &&
770 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
771 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
772 i, mddev);
773 }
774 #endif
775 }
776 if (force && !clean &&
777 !enough(info.array.level, info.array.raid_disks,
778 info.array.layout, clean,
779 avail, okcnt)) {
780 change += st->ss->update_super(st, &info, "force-array",
781 devices[chosen_drive].devname, verbose,
782 0, NULL);
783 clean = 1;
784 }
785
786 if (change) {
787 int fd;
788 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
789 if (fd < 0) {
790 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
791 devices[chosen_drive].devname);
792 if (must_close) close(mdfd);
793 return 1;
794 }
795 if (st->ss->store_super(st, fd)) {
796 close(fd);
797 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
798 devices[chosen_drive].devname);
799 if (must_close) close(mdfd);
800 return 1;
801 }
802 close(fd);
803 }
804
805 /* If we are in the middle of a reshape we may need to restore saved data
806 * that was moved aside due to the reshape overwriting live data
807 * The code of doing this lives in Grow.c
808 */
809 #ifndef MDASSEMBLE
810 if (info.reshape_active) {
811 int err = 0;
812 int *fdlist = malloc(sizeof(int)* bestcnt);
813 for (i=0; i<bestcnt; i++) {
814 int j = best[i];
815 if (j >= 0) {
816 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
817 if (fdlist[i] < 0) {
818 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
819 devices[j].devname);
820 err = 1;
821 break;
822 }
823 } else
824 fdlist[i] = -1;
825 }
826 if (!err)
827 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
828 while (i>0) {
829 i--;
830 if (fdlist[i]>=0) close(fdlist[i]);
831 }
832 if (err) {
833 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
834 if (must_close) close(mdfd);
835 return err;
836 }
837 }
838 #endif
839 /* count number of in-sync devices according to the superblock.
840 * We must have this number to start the array without -s or -R
841 */
842 req_cnt = info.array.working_disks;
843
844 /* Almost ready to actually *do* something */
845 if (!old_linux) {
846 int rv;
847 if ((vers % 100) >= 1) { /* can use different versions */
848 mdu_array_info_t inf;
849 memset(&inf, 0, sizeof(inf));
850 inf.major_version = st->ss->major;
851 inf.minor_version = st->minor_version;
852 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
853 } else
854 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
855
856 if (rv) {
857 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
858 mddev, strerror(errno));
859 if (must_close) close(mdfd);
860 return 1;
861 }
862 if (ident->bitmap_fd >= 0) {
863 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
864 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
865 if (must_close) close(mdfd);
866 return 1;
867 }
868 } else if (ident->bitmap_file) {
869 /* From config file */
870 int bmfd = open(ident->bitmap_file, O_RDWR);
871 if (bmfd < 0) {
872 fprintf(stderr, Name ": Could not open bitmap file %s\n",
873 ident->bitmap_file);
874 if (must_close) close(mdfd);
875 return 1;
876 }
877 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
878 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
879 close(bmfd);
880 if (must_close) close(mdfd);
881 return 1;
882 }
883 close(bmfd);
884 }
885
886 /* First, add the raid disks, but add the chosen one last */
887 for (i=0; i<= bestcnt; i++) {
888 int j;
889 if (i < bestcnt) {
890 j = best[i];
891 if (j == chosen_drive)
892 continue;
893 } else
894 j = chosen_drive;
895
896 if (j >= 0 /* && devices[j].uptodate */) {
897 mdu_disk_info_t disk;
898 memset(&disk, 0, sizeof(disk));
899 disk.major = devices[j].major;
900 disk.minor = devices[j].minor;
901 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
902 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
903 devices[j].devname,
904 mddev,
905 strerror(errno));
906 if (i < info.array.raid_disks || i == bestcnt)
907 okcnt--;
908 else
909 sparecnt--;
910 } else if (verbose > 0)
911 fprintf(stderr, Name ": added %s to %s as %d\n",
912 devices[j].devname, mddev, devices[j].raid_disk);
913 } else if (verbose > 0 && i < info.array.raid_disks)
914 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
915 i, mddev);
916 }
917
918 if (runstop == 1 ||
919 (runstop <= 0 &&
920 ( enough(info.array.level, info.array.raid_disks,
921 info.array.layout, clean, avail, okcnt) &&
922 (okcnt >= req_cnt || start_partial_ok)
923 ))) {
924 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
925 if (verbose >= 0) {
926 fprintf(stderr, Name ": %s has been started with %d drive%s",
927 mddev, okcnt, okcnt==1?"":"s");
928 if (okcnt < info.array.raid_disks)
929 fprintf(stderr, " (out of %d)", info.array.raid_disks);
930 if (sparecnt)
931 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
932 fprintf(stderr, ".\n");
933 }
934 if (must_close) {
935 int usecs = 1;
936 close(mdfd);
937 /* There is a nasty race with 'mdadm --monitor'.
938 * If it opens this device before we close it,
939 * it gets an incomplete open on which IO
940 * doesn't work and the capacity if wrong.
941 * If we reopen (to check for layered devices)
942 * before --monitor closes, we loose.
943 *
944 * So: wait upto 1 second for there to be
945 * a non-zero capacity.
946 */
947 while (usecs < 1000) {
948 mdfd = open(mddev, O_RDONLY);
949 if (mdfd >= 0) {
950 unsigned long long size;
951 if (get_dev_size(mdfd, NULL, &size) &&
952 size > 0)
953 break;
954 close(mdfd);
955 }
956 usleep(usecs);
957 usecs <<= 1;
958 }
959 }
960 return 0;
961 }
962 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
963 mddev, strerror(errno));
964
965 if (!enough(info.array.level, info.array.raid_disks,
966 info.array.layout, 1, avail, okcnt))
967 fprintf(stderr, Name ": Not enough devices to "
968 "start the array.\n");
969 else if (!enough(info.array.level,
970 info.array.raid_disks,
971 info.array.layout, clean,
972 avail, okcnt))
973 fprintf(stderr, Name ": Not enough devices to "
974 "start the array while not clean "
975 "- consider --force.\n");
976
977 if (must_close) close(mdfd);
978 return 1;
979 }
980 if (runstop == -1) {
981 fprintf(stderr, Name ": %s assembled from %d drive%s",
982 mddev, okcnt, okcnt==1?"":"s");
983 if (okcnt != info.array.raid_disks)
984 fprintf(stderr, " (out of %d)", info.array.raid_disks);
985 fprintf(stderr, ", but not started.\n");
986 if (must_close) close(mdfd);
987 return 0;
988 }
989 if (verbose >= -1) {
990 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
991 if (sparecnt)
992 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
993 if (!enough(info.array.level, info.array.raid_disks,
994 info.array.layout, 1, avail, okcnt))
995 fprintf(stderr, " - not enough to start the array.\n");
996 else if (!enough(info.array.level,
997 info.array.raid_disks,
998 info.array.layout, clean,
999 avail, okcnt))
1000 fprintf(stderr, " - not enough to start the "
1001 "array while not clean - consider "
1002 "--force.\n");
1003 else {
1004 if (req_cnt == info.array.raid_disks)
1005 fprintf(stderr, " - need all %d to start it", req_cnt);
1006 else
1007 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1008 fprintf(stderr, " (use --run to insist).\n");
1009 }
1010 }
1011 if (must_close) close(mdfd);
1012 return 1;
1013 } else {
1014 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1015 * been updated to point to the current locations of devices.
1016 * so we can just start the array
1017 */
1018 unsigned long dev;
1019 dev = makedev(devices[chosen_drive].major,
1020 devices[chosen_drive].minor);
1021 if (ioctl(mdfd, START_ARRAY, dev)) {
1022 fprintf(stderr, Name ": Cannot start array: %s\n",
1023 strerror(errno));
1024 }
1025
1026 }
1027 if (must_close) close(mdfd);
1028 return 0;
1029 }