]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Report uuid in --detail --brief for ddf and intel
[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 int uptodate; /* set once we decide that this device is as
121 * recent as everything else in the array.
122 */
123 struct mdinfo i;
124 } *devices;
125 int *best = NULL; /* indexed by raid_disk */
126 unsigned int bestcnt = 0;
127 int devcnt = 0;
128 unsigned int okcnt, sparecnt;
129 unsigned int req_cnt;
130 unsigned int i;
131 int most_recent = 0;
132 int chosen_drive;
133 int change = 0;
134 int inargv = 0;
135 int bitmap_done;
136 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
137 unsigned int num_devs;
138 mddev_dev_t tmpdev;
139 struct mdinfo info;
140 char *avail;
141 int nextspare = 0;
142 int uuid_for_name = 0;
143
144 memset(&info, 0, sizeof(info));
145
146 if (get_linux_version() < 2004000)
147 old_linux = 1;
148
149 if (mdfd >= 0) {
150 vers = md_get_version(mdfd);
151 if (vers <= 0) {
152 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
153 return 1;
154 }
155 if (vers < 9000) {
156 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
157 " Upgrade your kernel or try --build\n");
158 return 1;
159 }
160
161 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
162 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
163 mddev);
164 return 1;
165 }
166 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
167 }
168 /*
169 * If any subdevs are listed, then any that don't
170 * match ident are discarded. Remainder must all match and
171 * become the array.
172 * If no subdevs, then we scan all devices in the config file, but
173 * there must be something in the identity
174 */
175
176 if (!devlist &&
177 ident->uuid_set == 0 &&
178 ident->super_minor < 0 &&
179 ident->devices == NULL) {
180 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
181 mddev ? mddev : "further assembly");
182 return 1;
183 }
184 if (devlist == NULL)
185 devlist = conf_get_devs();
186 else if (mdfd >= 0)
187 inargv = 1;
188
189 try_again:
190
191 tmpdev = devlist; num_devs = 0;
192 while (tmpdev) {
193 if (tmpdev->used)
194 tmpdev->used = 2;
195 else
196 num_devs++;
197 tmpdev = tmpdev->next;
198 }
199 devices = malloc(num_devs * sizeof(*devices));
200
201 if (!st && ident->st) st = ident->st;
202
203 if (verbose>0)
204 fprintf(stderr, Name ": looking for devices for %s\n",
205 mddev ? mddev : "further assembly");
206
207 /* first walk the list of devices to find a consistent set
208 * that match the criterea, if that is possible.
209 * We flag the one we like with 'used'.
210 */
211 for (tmpdev = devlist;
212 tmpdev;
213 tmpdev = tmpdev->next) {
214 char *devname = tmpdev->devname;
215 int dfd;
216 struct stat stb;
217 struct supertype *tst = dup_super(st);
218
219 if (tmpdev->used > 1) continue;
220
221 if (ident->devices &&
222 !match_oneof(ident->devices, devname)) {
223 if ((inargv && verbose>=0) || verbose > 0)
224 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
225 continue;
226 }
227
228 dfd = dev_open(devname, O_RDONLY|O_EXCL);
229 if (dfd < 0) {
230 if ((inargv && verbose >= 0) || verbose > 0)
231 fprintf(stderr, Name ": cannot open device %s: %s\n",
232 devname, strerror(errno));
233 tmpdev->used = 2;
234 } else if (fstat(dfd, &stb)< 0) {
235 /* Impossible! */
236 fprintf(stderr, Name ": fstat failed for %s: %s\n",
237 devname, strerror(errno));
238 tmpdev->used = 2;
239 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
240 fprintf(stderr, Name ": %s is not a block device.\n",
241 devname);
242 tmpdev->used = 2;
243 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
244 if ((inargv && verbose >= 0) || verbose > 0)
245 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
246 devname);
247 tmpdev->used = 2;
248 } else if (tst->ss->load_super(tst,dfd, NULL)) {
249 if ((inargv && verbose >= 0) || verbose > 0)
250 fprintf( stderr, Name ": no RAID superblock on %s\n",
251 devname);
252 } else {
253 tst->ss->getinfo_super(tst, &info);
254 }
255 if (dfd >= 0) close(dfd);
256
257 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
258 (!tst || !tst->sb ||
259 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 (!tst || !tst->sb ||
267 name_matches(info.name, ident->name, homehost)==0)) {
268 if ((inargv && verbose >= 0) || verbose > 0)
269 fprintf(stderr, Name ": %s has wrong name.\n",
270 devname);
271 goto loop;
272 }
273 if (ident->super_minor != UnSet &&
274 (!tst || !tst->sb ||
275 ident->super_minor != info.array.md_minor)) {
276 if ((inargv && verbose >= 0) || verbose > 0)
277 fprintf(stderr, Name ": %s has wrong super-minor.\n",
278 devname);
279 goto loop;
280 }
281 if (ident->level != UnSet &&
282 (!tst || !tst->sb ||
283 ident->level != info.array.level)) {
284 if ((inargv && verbose >= 0) || verbose > 0)
285 fprintf(stderr, Name ": %s has wrong raid level.\n",
286 devname);
287 goto loop;
288 }
289 if (ident->raid_disks != UnSet &&
290 (!tst || !tst->sb ||
291 ident->raid_disks!= info.array.raid_disks)) {
292 if ((inargv && verbose >= 0) || verbose > 0)
293 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
294 devname);
295 goto loop;
296 }
297 if (mdfd < 0) {
298 if (tst == NULL || tst->sb == NULL)
299 continue;
300 switch(tst->ss->match_home(tst, homehost))
301 {
302 case 1: /* happy with match. */
303 break;
304 case -1: /* cannot match */
305 uuid_for_name = 1;
306 break;
307 case 0: /* Doesn't match */
308 if (update)
309 /* We are changing the name*/
310 break;
311 if ((inargv && verbose >= 0) || verbose > 0)
312 fprintf(stderr, Name ": %s is not built for "
313 "host %s - using UUID for "
314 "device name.\n",
315 devname, homehost);
316
317 /* Auto-assemble, and this is not a usable host */
318 /* if update != NULL, we are updating the host
319 * name... */
320 uuid_for_name = 1;
321 break;
322 }
323 }
324 /* If we are this far, then we are nearly commited to this device.
325 * If the super_block doesn't exist, or doesn't match others,
326 * then we probably cannot continue
327 * However if one of the arrays is for the homehost, and
328 * the other isn't that can disambiguate.
329 */
330
331 if (!tst || !tst->sb) {
332 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
333 devname);
334 if (st)
335 st->ss->free_super(st);
336 return 1;
337 }
338
339 if (st == NULL)
340 st = dup_super(tst);
341 if (st->minor_version == -1)
342 st->minor_version = tst->minor_version;
343 if (st->ss != tst->ss ||
344 st->minor_version != tst->minor_version ||
345 st->ss->compare_super(st, tst) != 0) {
346 /* Some mismatch. If exactly one array matches this host,
347 * we can resolve on that one.
348 * Or, if we are auto assembling, we just ignore the second
349 * for now.
350 */
351 if (mdfd < 0)
352 goto loop;
353 if (homehost) {
354 int first = st->ss->match_home(st, homehost);
355 int last = tst->ss->match_home(tst, homehost);
356 if (first != last &&
357 (first == 1 || last == 1)) {
358 /* We can do something */
359 if (first) {/* just ignore this one */
360 if ((inargv && verbose >= 0) || verbose > 0)
361 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
362 devname);
363 goto loop;
364 } else { /* reject all those sofar */
365 mddev_dev_t td;
366 if ((inargv && verbose >= 0) || verbose > 0)
367 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
368 devname);
369 for (td=devlist; td != tmpdev; td=td->next)
370 if (td->used == 1)
371 td->used = 0;
372 tmpdev->used = 1;
373 goto loop;
374 }
375 }
376 }
377 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
378 devname);
379 tst->ss->free_super(tst);
380 st->ss->free_super(st);
381 return 1;
382 }
383
384 tmpdev->used = 1;
385
386 loop:
387 if (tst)
388 tst->ss->free_super(tst);
389 }
390
391 if (mdfd < 0) {
392 /* So... it is up to me to open the device.
393 * We create a name '/dev/md/XXX' based on the info in the
394 * superblock, and call open_mddev on that
395 */
396 mdu_array_info_t inf;
397 char *c;
398 char nbuf[64];
399 if (!st || !st->sb) {
400 return 2;
401 }
402 st->ss->getinfo_super(st, &info);
403 if (uuid_for_name)
404 c = fname_from_uuid(st, &info, nbuf, '-');
405 else {
406 c = strchr(info.name, ':');
407 if (c) c++; else c= info.name;
408 }
409 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
410 /* /dev/md/d0 style for partitionable */
411 asprintf(&mddev, "/dev/md/d%s", c);
412 else
413 asprintf(&mddev, "/dev/md/%s", c);
414 mdfd = open_mddev(mddev, ident->autof);
415 if (mdfd < 0) {
416 st->ss->free_super(st);
417 free(devices);
418 goto try_again;
419 }
420 vers = md_get_version(mdfd);
421 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
422 for (tmpdev = devlist ;
423 tmpdev && tmpdev->used != 1;
424 tmpdev = tmpdev->next)
425 ;
426 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
427 if (tmpdev)
428 fprintf(stderr, Name ": %s needed for %s...\n",
429 mddev, tmpdev->devname);
430 close(mdfd);
431 mdfd = -1;
432 st->ss->free_super(st);
433 free(devices);
434 goto try_again;
435 }
436 must_close = 1;
437 }
438
439 /* Ok, no bad inconsistancy, we can try updating etc */
440 bitmap_done = 0;
441 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
442 char *devname = tmpdev->devname;
443 struct stat stb;
444 /* looks like a good enough match to update the super block if needed */
445 #ifndef MDASSEMBLE
446 if (update) {
447 int dfd;
448 /* prepare useful information in info structures */
449 struct stat stb2;
450 struct supertype *tst;
451 fstat(mdfd, &stb2);
452
453 if (strcmp(update, "uuid")==0 &&
454 !ident->uuid_set) {
455 int rfd;
456 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
457 read(rfd, ident->uuid, 16) != 16) {
458 *(__u32*)(ident->uuid) = random();
459 *(__u32*)(ident->uuid+1) = random();
460 *(__u32*)(ident->uuid+2) = random();
461 *(__u32*)(ident->uuid+3) = random();
462 }
463 if (rfd >= 0) close(rfd);
464 }
465 dfd = dev_open(devname, O_RDWR|O_EXCL);
466
467 remove_partitions(dfd);
468
469 tst = dup_super(st);
470 tst->ss->load_super(tst, dfd, NULL);
471 tst->ss->getinfo_super(tst, &info);
472
473 memcpy(info.uuid, ident->uuid, 16);
474 strcpy(info.name, ident->name);
475 info.array.md_minor = minor(stb2.st_rdev);
476
477 tst->ss->update_super(tst, &info, update,
478 devname, verbose,
479 ident->uuid_set, homehost);
480 if (strcmp(update, "uuid")==0 &&
481 !ident->uuid_set) {
482 ident->uuid_set = 1;
483 memcpy(ident->uuid, info.uuid, 16);
484 }
485 if (dfd < 0)
486 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
487 devname);
488 else if (tst->ss->store_super(tst, dfd))
489 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
490 devname);
491 if (dfd >= 0)
492 close(dfd);
493
494 if (strcmp(update, "uuid")==0 &&
495 ident->bitmap_fd >= 0 && !bitmap_done) {
496 if (bitmap_update_uuid(ident->bitmap_fd,
497 info.uuid,
498 tst->ss->swapuuid) != 0)
499 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
500 else
501 bitmap_done = 1;
502 }
503 tst->ss->free_super(tst);
504 } else
505 #endif
506 {
507 struct supertype *tst = dup_super(st);
508 int dfd;
509 dfd = dev_open(devname, O_RDWR|O_EXCL);
510
511 remove_partitions(dfd);
512
513 tst->ss->load_super(tst, dfd, NULL);
514 tst->ss->getinfo_super(tst, &info);
515 tst->ss->free_super(tst);
516 close(dfd);
517 }
518
519 stat(devname, &stb);
520
521 if (verbose > 0)
522 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
523 devname, mddev, info.disk.raid_disk);
524 devices[devcnt].devname = devname;
525 devices[devcnt].uptodate = 0;
526 devices[devcnt].i = info;
527 devices[devcnt].i.disk.major = major(stb.st_rdev);
528 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
529 if (most_recent < devcnt) {
530 if (devices[devcnt].i.events
531 > devices[most_recent].i.events)
532 most_recent = devcnt;
533 }
534 if (info.array.level == -4)
535 /* with multipath, the raid_disk from the superblock is meaningless */
536 i = devcnt;
537 else
538 i = devices[devcnt].i.disk.raid_disk;
539 if (i+1 == 0) {
540 if (nextspare < info.array.raid_disks)
541 nextspare = info.array.raid_disks;
542 i = nextspare++;
543 } else {
544 if (i >= info.array.raid_disks &&
545 i >= nextspare)
546 nextspare = i+1;
547 }
548 if (i < 10000) {
549 if (i >= bestcnt) {
550 unsigned int newbestcnt = i+10;
551 int *newbest = malloc(sizeof(int)*newbestcnt);
552 unsigned int c;
553 for (c=0; c < newbestcnt; c++)
554 if (c < bestcnt)
555 newbest[c] = best[c];
556 else
557 newbest[c] = -1;
558 if (best)free(best);
559 best = newbest;
560 bestcnt = newbestcnt;
561 }
562 if (best[i] >=0 &&
563 devices[best[i]].i.events
564 == devices[devcnt].i.events
565 && (devices[best[i]].i.disk.minor
566 != devices[devcnt].i.disk.minor)
567 && st->ss == &super0
568 && info.array.level != LEVEL_MULTIPATH) {
569 /* two different devices with identical superblock.
570 * Could be a mis-detection caused by overlapping
571 * partitions. fail-safe.
572 */
573 fprintf(stderr, Name ": WARNING %s and %s appear"
574 " to have very similar superblocks.\n"
575 " If they are really different, "
576 "please --zero the superblock on one\n"
577 " If they are the same or overlap,"
578 " please remove one from %s.\n",
579 devices[best[i]].devname, devname,
580 inargv ? "the list" :
581 "the\n DEVICE list in mdadm.conf"
582 );
583 if (must_close) close(mdfd);
584 return 1;
585 }
586 if (best[i] == -1
587 || (devices[best[i]].i.events
588 < devices[devcnt].i.events))
589 best[i] = devcnt;
590 }
591 devcnt++;
592 }
593
594 if (devcnt == 0) {
595 fprintf(stderr, Name ": no devices found for %s\n",
596 mddev);
597 if (st)
598 st->ss->free_super(st);
599 if (must_close) close(mdfd);
600 return 1;
601 }
602
603 if (update && strcmp(update, "byteorder")==0)
604 st->minor_version = 90;
605
606 st->ss->getinfo_super(st, &info);
607 clean = info.array.state & 1;
608
609 /* now we have some devices that might be suitable.
610 * I wonder how many
611 */
612 avail = malloc(info.array.raid_disks);
613 memset(avail, 0, info.array.raid_disks);
614 okcnt = 0;
615 sparecnt=0;
616 for (i=0; i< bestcnt ;i++) {
617 int j = best[i];
618 int event_margin = 1; /* always allow a difference of '1'
619 * like the kernel does
620 */
621 if (j < 0) continue;
622 /* note: we ignore error flags in multipath arrays
623 * as they don't make sense
624 */
625 if (info.array.level != -4)
626 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
627 if (!(devices[j].i.disk.state
628 & (1<<MD_DISK_FAULTY)))
629 sparecnt++;
630 continue;
631 }
632 if (devices[j].i.events+event_margin >=
633 devices[most_recent].i.events) {
634 devices[j].uptodate = 1;
635 if (i < info.array.raid_disks) {
636 okcnt++;
637 avail[i]=1;
638 } else
639 sparecnt++;
640 }
641 }
642 while (force && !enough(info.array.level, info.array.raid_disks,
643 info.array.layout, 1,
644 avail, okcnt)) {
645 /* Choose the newest best drive which is
646 * not up-to-date, update the superblock
647 * and add it.
648 */
649 int fd;
650 struct supertype *tst;
651 long long current_events;
652 chosen_drive = -1;
653 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
654 int j = best[i];
655 if (j>=0 &&
656 !devices[j].uptodate &&
657 devices[j].i.events > 0 &&
658 (chosen_drive < 0 ||
659 devices[j].i.events
660 > devices[chosen_drive].i.events))
661 chosen_drive = j;
662 }
663 if (chosen_drive < 0)
664 break;
665 current_events = devices[chosen_drive].i.events;
666 add_another:
667 if (verbose >= 0)
668 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
669 devices[chosen_drive].devname,
670 devices[chosen_drive].i.disk.raid_disk,
671 (int)(devices[chosen_drive].i.events),
672 (int)(devices[most_recent].i.events));
673 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
674 if (fd < 0) {
675 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
676 devices[chosen_drive].devname);
677 devices[chosen_drive].i.events = 0;
678 continue;
679 }
680 tst = dup_super(st);
681 if (tst->ss->load_super(tst,fd, NULL)) {
682 close(fd);
683 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
684 devices[chosen_drive].devname);
685 devices[chosen_drive].i.events = 0;
686 continue;
687 }
688 info.events = devices[most_recent].i.events;
689 tst->ss->update_super(tst, &info, "force-one",
690 devices[chosen_drive].devname, verbose,
691 0, NULL);
692
693 if (tst->ss->store_super(tst, fd)) {
694 close(fd);
695 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
696 devices[chosen_drive].devname);
697 devices[chosen_drive].i.events = 0;
698 tst->ss->free_super(tst);
699 continue;
700 }
701 close(fd);
702 devices[chosen_drive].i.events = devices[most_recent].i.events;
703 devices[chosen_drive].uptodate = 1;
704 avail[chosen_drive] = 1;
705 okcnt++;
706 tst->ss->free_super(tst);
707
708 /* If there are any other drives of the same vintage,
709 * add them in as well. We can't lose and we might gain
710 */
711 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
712 int j = best[i];
713 if (j >= 0 &&
714 !devices[j].uptodate &&
715 devices[j].i.events > 0 &&
716 devices[j].i.events == current_events) {
717 chosen_drive = j;
718 goto add_another;
719 }
720 }
721 }
722
723 /* Now we want to look at the superblock which the kernel will base things on
724 * and compare the devices that we think are working with the devices that the
725 * superblock thinks are working.
726 * If there are differences and --force is given, then update this chosen
727 * superblock.
728 */
729 chosen_drive = -1;
730 st->ss->free_super(st);
731 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
732 int j = best[i];
733 int fd;
734
735 if (j<0)
736 continue;
737 if (!devices[j].uptodate)
738 continue;
739 chosen_drive = j;
740 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
741 fprintf(stderr, Name ": Cannot open %s: %s\n",
742 devices[j].devname, strerror(errno));
743 if (must_close) close(mdfd);
744 return 1;
745 }
746 if (st->ss->load_super(st,fd, NULL)) {
747 close(fd);
748 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
749 devices[j].devname);
750 if (must_close) close(mdfd);
751 return 1;
752 }
753 close(fd);
754 }
755 if (st->sb == NULL) {
756 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
757 if (must_close) close(mdfd);
758 return 1;
759 }
760 st->ss->getinfo_super(st, &info);
761 #ifndef MDASSEMBLE
762 sysfs_init(&info, mdfd, 0);
763 #endif
764 for (i=0; i<bestcnt; i++) {
765 int j = best[i];
766 unsigned int desired_state;
767
768 if (i < info.array.raid_disks)
769 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
770 else
771 desired_state = 0;
772
773 if (j<0)
774 continue;
775 if (!devices[j].uptodate)
776 continue;
777
778 devices[j].i.disk.state = desired_state;
779
780 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
781 verbose, 0, NULL)) {
782 if (force) {
783 if (verbose >= 0)
784 fprintf(stderr, Name ": "
785 "clearing FAULTY flag for device %d in %s for %s\n",
786 j, mddev, devices[j].devname);
787 change = 1;
788 } else {
789 if (verbose >= -1)
790 fprintf(stderr, Name ": "
791 "device %d in %s has wrong state in superblock, but %s seems ok\n",
792 i, mddev, devices[j].devname);
793 }
794 }
795 #if 0
796 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
797 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
798 i, mddev);
799 }
800 #endif
801 }
802 if (force && !clean &&
803 !enough(info.array.level, info.array.raid_disks,
804 info.array.layout, clean,
805 avail, okcnt)) {
806 change += st->ss->update_super(st, &info, "force-array",
807 devices[chosen_drive].devname, verbose,
808 0, NULL);
809 clean = 1;
810 }
811
812 if (change) {
813 int fd;
814 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
815 if (fd < 0) {
816 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
817 devices[chosen_drive].devname);
818 if (must_close) close(mdfd);
819 return 1;
820 }
821 if (st->ss->store_super(st, fd)) {
822 close(fd);
823 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
824 devices[chosen_drive].devname);
825 if (must_close) close(mdfd);
826 return 1;
827 }
828 close(fd);
829 }
830
831 /* If we are in the middle of a reshape we may need to restore saved data
832 * that was moved aside due to the reshape overwriting live data
833 * The code of doing this lives in Grow.c
834 */
835 #ifndef MDASSEMBLE
836 if (info.reshape_active) {
837 int err = 0;
838 int *fdlist = malloc(sizeof(int)* bestcnt);
839 for (i=0; i<bestcnt; i++) {
840 int j = best[i];
841 if (j >= 0) {
842 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
843 if (fdlist[i] < 0) {
844 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
845 devices[j].devname);
846 err = 1;
847 break;
848 }
849 } else
850 fdlist[i] = -1;
851 }
852 if (!err)
853 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
854 while (i>0) {
855 i--;
856 if (fdlist[i]>=0) close(fdlist[i]);
857 }
858 if (err) {
859 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
860 if (must_close) close(mdfd);
861 return err;
862 }
863 }
864 #endif
865 /* count number of in-sync devices according to the superblock.
866 * We must have this number to start the array without -s or -R
867 */
868 req_cnt = info.array.working_disks;
869
870 /* Almost ready to actually *do* something */
871 if (!old_linux) {
872 int rv;
873
874 rv = set_array_info(mdfd, st, &info);
875 if (rv) {
876 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
877 mddev, strerror(errno));
878 if (must_close) close(mdfd);
879 return 1;
880 }
881 if (ident->bitmap_fd >= 0) {
882 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
883 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
884 if (must_close) close(mdfd);
885 return 1;
886 }
887 } else if (ident->bitmap_file) {
888 /* From config file */
889 int bmfd = open(ident->bitmap_file, O_RDWR);
890 if (bmfd < 0) {
891 fprintf(stderr, Name ": Could not open bitmap file %s\n",
892 ident->bitmap_file);
893 if (must_close) close(mdfd);
894 return 1;
895 }
896 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
897 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
898 close(bmfd);
899 if (must_close) close(mdfd);
900 return 1;
901 }
902 close(bmfd);
903 }
904
905 /* First, add the raid disks, but add the chosen one last */
906 for (i=0; i<= bestcnt; i++) {
907 int j;
908 if (i < bestcnt) {
909 j = best[i];
910 if (j == chosen_drive)
911 continue;
912 } else
913 j = chosen_drive;
914
915 if (j >= 0 /* && devices[j].uptodate */) {
916 rv = add_disk(mdfd, st, &info, &devices[j].i);
917
918 if (rv) {
919 fprintf(stderr, Name ": failed to add "
920 "%s to %s: %s\n",
921 devices[j].devname,
922 mddev,
923 strerror(errno));
924 if (i < info.array.raid_disks
925 || i == bestcnt)
926 okcnt--;
927 else
928 sparecnt--;
929 } else if (verbose > 0)
930 fprintf(stderr, Name ": added %s "
931 "to %s as %d\n",
932 devices[j].devname, mddev,
933 devices[j].i.disk.raid_disk);
934 } else if (verbose > 0 && i < info.array.raid_disks)
935 fprintf(stderr, Name ": no uptodate device for "
936 "slot %d of %s\n",
937 i, mddev);
938 }
939
940 if (info.array.level == LEVEL_CONTAINER) {
941 if (verbose >= 0) {
942 fprintf(stderr, Name ": Container %s has been "
943 "assembled with %d drive%s",
944 mddev, okcnt, okcnt==1?"":"s");
945 if (okcnt < info.array.raid_disks)
946 fprintf(stderr, " (out of %d)",
947 info.array.raid_disks);
948 fprintf(stderr, "\n");
949 }
950 if (must_close)
951 close(mdfd);
952 return 0;
953 }
954
955 if (runstop == 1 ||
956 (runstop <= 0 &&
957 ( enough(info.array.level, info.array.raid_disks,
958 info.array.layout, clean, avail, okcnt) &&
959 (okcnt >= req_cnt || start_partial_ok)
960 ))) {
961 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
962 if (verbose >= 0) {
963 fprintf(stderr, Name ": %s has been started with %d drive%s",
964 mddev, okcnt, okcnt==1?"":"s");
965 if (okcnt < info.array.raid_disks)
966 fprintf(stderr, " (out of %d)", info.array.raid_disks);
967 if (sparecnt)
968 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
969 fprintf(stderr, ".\n");
970 }
971 if (must_close) {
972 int usecs = 1;
973 close(mdfd);
974 /* There is a nasty race with 'mdadm --monitor'.
975 * If it opens this device before we close it,
976 * it gets an incomplete open on which IO
977 * doesn't work and the capacity is
978 * wrong.
979 * If we reopen (to check for layered devices)
980 * before --monitor closes, we loose.
981 *
982 * So: wait upto 1 second for there to be
983 * a non-zero capacity.
984 */
985 while (usecs < 1000) {
986 mdfd = open(mddev, O_RDONLY);
987 if (mdfd >= 0) {
988 unsigned long long size;
989 if (get_dev_size(mdfd, NULL, &size) &&
990 size > 0)
991 break;
992 close(mdfd);
993 }
994 usleep(usecs);
995 usecs <<= 1;
996 }
997 }
998 return 0;
999 }
1000 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1001 mddev, strerror(errno));
1002
1003 if (!enough(info.array.level, info.array.raid_disks,
1004 info.array.layout, 1, avail, okcnt))
1005 fprintf(stderr, Name ": Not enough devices to "
1006 "start the array.\n");
1007 else if (!enough(info.array.level,
1008 info.array.raid_disks,
1009 info.array.layout, clean,
1010 avail, okcnt))
1011 fprintf(stderr, Name ": Not enough devices to "
1012 "start the array while not clean "
1013 "- consider --force.\n");
1014
1015 if (must_close) {
1016 ioctl(mdfd, STOP_ARRAY, NULL);
1017 close(mdfd);
1018 }
1019 return 1;
1020 }
1021 if (runstop == -1) {
1022 fprintf(stderr, Name ": %s assembled from %d drive%s",
1023 mddev, okcnt, okcnt==1?"":"s");
1024 if (okcnt != info.array.raid_disks)
1025 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1026 fprintf(stderr, ", but not started.\n");
1027 if (must_close) close(mdfd);
1028 return 0;
1029 }
1030 if (verbose >= -1) {
1031 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1032 if (sparecnt)
1033 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1034 if (!enough(info.array.level, info.array.raid_disks,
1035 info.array.layout, 1, avail, okcnt))
1036 fprintf(stderr, " - not enough to start the array.\n");
1037 else if (!enough(info.array.level,
1038 info.array.raid_disks,
1039 info.array.layout, clean,
1040 avail, okcnt))
1041 fprintf(stderr, " - not enough to start the "
1042 "array while not clean - consider "
1043 "--force.\n");
1044 else {
1045 if (req_cnt == info.array.raid_disks)
1046 fprintf(stderr, " - need all %d to start it", req_cnt);
1047 else
1048 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1049 fprintf(stderr, " (use --run to insist).\n");
1050 }
1051 }
1052 if (must_close) {
1053 ioctl(mdfd, STOP_ARRAY, NULL);
1054 close(mdfd);
1055 }
1056 return 1;
1057 } else {
1058 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1059 * been updated to point to the current locations of devices.
1060 * so we can just start the array
1061 */
1062 unsigned long dev;
1063 dev = makedev(devices[chosen_drive].i.disk.major,
1064 devices[chosen_drive].i.disk.minor);
1065 if (ioctl(mdfd, START_ARRAY, dev)) {
1066 fprintf(stderr, Name ": Cannot start array: %s\n",
1067 strerror(errno));
1068 }
1069
1070 }
1071 if (must_close) close(mdfd);
1072 return 0;
1073 }