]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Add uuid support for super-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
143 memset(&info, 0, sizeof(info));
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 if (st)
321 st->ss->free_super(st);
322 return 1;
323 }
324
325 if (st == NULL)
326 st = dup_super(tst);
327 if (st->minor_version == -1)
328 st->minor_version = tst->minor_version;
329 if (st->ss != tst->ss ||
330 st->minor_version != tst->minor_version ||
331 st->ss->compare_super(st, tst) != 0) {
332 /* Some mismatch. If exactly one array matches this host,
333 * we can resolve on that one.
334 * Or, if we are auto assembling, we just ignore the second
335 * for now.
336 */
337 if (mdfd < 0)
338 goto loop;
339 if (homehost) {
340 int first = st->ss->match_home(st, homehost);
341 int last = tst->ss->match_home(tst, homehost);
342 if (first != last &&
343 (first == 1 || last == 1)) {
344 /* We can do something */
345 if (first) {/* just ignore this one */
346 if ((inargv && verbose >= 0) || verbose > 0)
347 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
348 devname);
349 goto loop;
350 } else { /* reject all those sofar */
351 mddev_dev_t td;
352 if ((inargv && verbose >= 0) || verbose > 0)
353 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
354 devname);
355 for (td=devlist; td != tmpdev; td=td->next)
356 if (td->used == 1)
357 td->used = 0;
358 tmpdev->used = 1;
359 goto loop;
360 }
361 }
362 }
363 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
364 devname);
365 tst->ss->free_super(tst);
366 st->ss->free_super(st);
367 return 1;
368 }
369
370 tmpdev->used = 1;
371
372 loop:
373 if (tst)
374 tst->ss->free_super(tst);
375 }
376
377 if (mdfd < 0) {
378 /* So... it is up to me to open the device.
379 * We create a name '/dev/md/XXX' based on the info in the
380 * superblock, and call open_mddev on that
381 */
382 mdu_array_info_t inf;
383 char *c;
384 if (!st || !st->sb) {
385 return 2;
386 }
387 st->ss->getinfo_super(st, &info);
388 c = strchr(info.name, ':');
389 if (c) c++; else c= info.name;
390 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
391 /* /dev/md/d0 style for partitionable */
392 asprintf(&mddev, "/dev/md/d%s", c);
393 else
394 asprintf(&mddev, "/dev/md/%s", c);
395 mdfd = open_mddev(mddev, ident->autof);
396 if (mdfd < 0) {
397 st->ss->free_super(st);
398 free(devices);
399 goto try_again;
400 }
401 vers = md_get_version(mdfd);
402 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
403 for (tmpdev = devlist ;
404 tmpdev && tmpdev->used != 1;
405 tmpdev = tmpdev->next)
406 ;
407 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
408 if (tmpdev)
409 fprintf(stderr, Name ": %s needed for %s...\n",
410 mddev, tmpdev->devname);
411 close(mdfd);
412 mdfd = -1;
413 st->ss->free_super(st);
414 free(devices);
415 goto try_again;
416 }
417 must_close = 1;
418 }
419
420 /* Ok, no bad inconsistancy, we can try updating etc */
421 bitmap_done = 0;
422 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
423 char *devname = tmpdev->devname;
424 struct stat stb;
425 /* looks like a good enough match to update the super block if needed */
426 #ifndef MDASSEMBLE
427 if (update) {
428 int dfd;
429 /* prepare useful information in info structures */
430 struct stat stb2;
431 struct supertype *tst;
432 fstat(mdfd, &stb2);
433
434 if (strcmp(update, "uuid")==0 &&
435 !ident->uuid_set) {
436 int rfd;
437 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
438 read(rfd, ident->uuid, 16) != 16) {
439 *(__u32*)(ident->uuid) = random();
440 *(__u32*)(ident->uuid+1) = random();
441 *(__u32*)(ident->uuid+2) = random();
442 *(__u32*)(ident->uuid+3) = random();
443 }
444 if (rfd >= 0) close(rfd);
445 }
446 dfd = dev_open(devname, O_RDWR|O_EXCL);
447
448 remove_partitions(dfd);
449
450 tst = dup_super(st);
451 tst->ss->load_super(tst, dfd, NULL);
452 tst->ss->getinfo_super(tst, &info);
453
454 memcpy(info.uuid, ident->uuid, 16);
455 strcpy(info.name, ident->name);
456 info.array.md_minor = minor(stb2.st_rdev);
457
458 tst->ss->update_super(tst, &info, update,
459 devname, verbose,
460 ident->uuid_set, homehost);
461 if (strcmp(update, "uuid")==0 &&
462 !ident->uuid_set) {
463 ident->uuid_set = 1;
464 memcpy(ident->uuid, info.uuid, 16);
465 }
466 if (dfd < 0)
467 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
468 devname);
469 else if (tst->ss->store_super(tst, dfd))
470 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
471 devname);
472 if (dfd >= 0)
473 close(dfd);
474
475 if (strcmp(update, "uuid")==0 &&
476 ident->bitmap_fd >= 0 && !bitmap_done) {
477 if (bitmap_update_uuid(ident->bitmap_fd,
478 info.uuid,
479 tst->ss->swapuuid) != 0)
480 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
481 else
482 bitmap_done = 1;
483 }
484 tst->ss->free_super(tst);
485 } else
486 #endif
487 {
488 struct supertype *tst = dup_super(st);
489 int dfd;
490 dfd = dev_open(devname, O_RDWR|O_EXCL);
491
492 remove_partitions(dfd);
493
494 tst->ss->load_super(tst, dfd, NULL);
495 tst->ss->getinfo_super(tst, &info);
496 tst->ss->free_super(tst);
497 close(dfd);
498 }
499
500 stat(devname, &stb);
501
502 if (verbose > 0)
503 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
504 devname, mddev, info.disk.raid_disk);
505 devices[devcnt].devname = devname;
506 devices[devcnt].uptodate = 0;
507 devices[devcnt].i = info;
508 devices[devcnt].i.disk.major = major(stb.st_rdev);
509 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
510 if (most_recent < devcnt) {
511 if (devices[devcnt].i.events
512 > devices[most_recent].i.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].i.disk.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]].i.events
545 == devices[devcnt].i.events
546 && (devices[best[i]].i.disk.minor
547 != devices[devcnt].i.disk.minor)
548 && st->ss == &super0
549 && info.array.level != LEVEL_MULTIPATH) {
550 /* two different devices with identical superblock.
551 * Could be a mis-detection caused by overlapping
552 * partitions. fail-safe.
553 */
554 fprintf(stderr, Name ": WARNING %s and %s appear"
555 " to have very similar superblocks.\n"
556 " If they are really different, "
557 "please --zero the superblock on one\n"
558 " If they are the same or overlap,"
559 " please remove one from %s.\n",
560 devices[best[i]].devname, devname,
561 inargv ? "the list" :
562 "the\n DEVICE list in mdadm.conf"
563 );
564 if (must_close) close(mdfd);
565 return 1;
566 }
567 if (best[i] == -1
568 || (devices[best[i]].i.events
569 < devices[devcnt].i.events))
570 best[i] = devcnt;
571 }
572 devcnt++;
573 }
574
575 if (devcnt == 0) {
576 fprintf(stderr, Name ": no devices found for %s\n",
577 mddev);
578 if (st)
579 st->ss->free_super(st);
580 if (must_close) close(mdfd);
581 return 1;
582 }
583
584 if (update && strcmp(update, "byteorder")==0)
585 st->minor_version = 90;
586
587 st->ss->getinfo_super(st, &info);
588 clean = info.array.state & 1;
589
590 /* now we have some devices that might be suitable.
591 * I wonder how many
592 */
593 avail = malloc(info.array.raid_disks);
594 memset(avail, 0, info.array.raid_disks);
595 okcnt = 0;
596 sparecnt=0;
597 for (i=0; i< bestcnt ;i++) {
598 int j = best[i];
599 int event_margin = 1; /* always allow a difference of '1'
600 * like the kernel does
601 */
602 if (j < 0) continue;
603 /* note: we ignore error flags in multipath arrays
604 * as they don't make sense
605 */
606 if (info.array.level != -4)
607 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
608 if (!(devices[j].i.disk.state
609 & (1<<MD_DISK_FAULTY)))
610 sparecnt++;
611 continue;
612 }
613 if (devices[j].i.events+event_margin >=
614 devices[most_recent].i.events) {
615 devices[j].uptodate = 1;
616 if (i < info.array.raid_disks) {
617 okcnt++;
618 avail[i]=1;
619 } else
620 sparecnt++;
621 }
622 }
623 while (force && !enough(info.array.level, info.array.raid_disks,
624 info.array.layout, 1,
625 avail, okcnt)) {
626 /* Choose the newest best drive which is
627 * not up-to-date, update the superblock
628 * and add it.
629 */
630 int fd;
631 struct supertype *tst;
632 long long current_events;
633 chosen_drive = -1;
634 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
635 int j = best[i];
636 if (j>=0 &&
637 !devices[j].uptodate &&
638 devices[j].i.events > 0 &&
639 (chosen_drive < 0 ||
640 devices[j].i.events
641 > devices[chosen_drive].i.events))
642 chosen_drive = j;
643 }
644 if (chosen_drive < 0)
645 break;
646 current_events = devices[chosen_drive].i.events;
647 add_another:
648 if (verbose >= 0)
649 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
650 devices[chosen_drive].devname,
651 devices[chosen_drive].i.disk.raid_disk,
652 (int)(devices[chosen_drive].i.events),
653 (int)(devices[most_recent].i.events));
654 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
655 if (fd < 0) {
656 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
657 devices[chosen_drive].devname);
658 devices[chosen_drive].i.events = 0;
659 continue;
660 }
661 tst = dup_super(st);
662 if (tst->ss->load_super(tst,fd, NULL)) {
663 close(fd);
664 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
665 devices[chosen_drive].devname);
666 devices[chosen_drive].i.events = 0;
667 continue;
668 }
669 info.events = devices[most_recent].i.events;
670 tst->ss->update_super(tst, &info, "force-one",
671 devices[chosen_drive].devname, verbose,
672 0, NULL);
673
674 if (tst->ss->store_super(tst, fd)) {
675 close(fd);
676 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
677 devices[chosen_drive].devname);
678 devices[chosen_drive].i.events = 0;
679 tst->ss->free_super(tst);
680 continue;
681 }
682 close(fd);
683 devices[chosen_drive].i.events = devices[most_recent].i.events;
684 devices[chosen_drive].uptodate = 1;
685 avail[chosen_drive] = 1;
686 okcnt++;
687 tst->ss->free_super(tst);
688
689 /* If there are any other drives of the same vintage,
690 * add them in as well. We can't lose and we might gain
691 */
692 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
693 int j = best[i];
694 if (j >= 0 &&
695 !devices[j].uptodate &&
696 devices[j].i.events > 0 &&
697 devices[j].i.events == current_events) {
698 chosen_drive = j;
699 goto add_another;
700 }
701 }
702 }
703
704 /* Now we want to look at the superblock which the kernel will base things on
705 * and compare the devices that we think are working with the devices that the
706 * superblock thinks are working.
707 * If there are differences and --force is given, then update this chosen
708 * superblock.
709 */
710 chosen_drive = -1;
711 st->ss->free_super(st);
712 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
713 int j = best[i];
714 int fd;
715
716 if (j<0)
717 continue;
718 if (!devices[j].uptodate)
719 continue;
720 chosen_drive = j;
721 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
722 fprintf(stderr, Name ": Cannot open %s: %s\n",
723 devices[j].devname, strerror(errno));
724 if (must_close) close(mdfd);
725 return 1;
726 }
727 if (st->ss->load_super(st,fd, NULL)) {
728 close(fd);
729 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
730 devices[j].devname);
731 if (must_close) close(mdfd);
732 return 1;
733 }
734 close(fd);
735 }
736 if (st->sb == NULL) {
737 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
738 if (must_close) close(mdfd);
739 return 1;
740 }
741 st->ss->getinfo_super(st, &info);
742 #ifndef MDASSEMBLE
743 sysfs_init(&info, mdfd, 0);
744 #endif
745 for (i=0; i<bestcnt; i++) {
746 int j = best[i];
747 unsigned int desired_state;
748
749 if (i < info.array.raid_disks)
750 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
751 else
752 desired_state = 0;
753
754 if (j<0)
755 continue;
756 if (!devices[j].uptodate)
757 continue;
758
759 devices[j].i.disk.state = desired_state;
760
761 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
762 verbose, 0, NULL)) {
763 if (force) {
764 if (verbose >= 0)
765 fprintf(stderr, Name ": "
766 "clearing FAULTY flag for device %d in %s for %s\n",
767 j, mddev, devices[j].devname);
768 change = 1;
769 } else {
770 if (verbose >= -1)
771 fprintf(stderr, Name ": "
772 "device %d in %s has wrong state in superblock, but %s seems ok\n",
773 i, mddev, devices[j].devname);
774 }
775 }
776 #if 0
777 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
778 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
779 i, mddev);
780 }
781 #endif
782 }
783 if (force && !clean &&
784 !enough(info.array.level, info.array.raid_disks,
785 info.array.layout, clean,
786 avail, okcnt)) {
787 change += st->ss->update_super(st, &info, "force-array",
788 devices[chosen_drive].devname, verbose,
789 0, NULL);
790 clean = 1;
791 }
792
793 if (change) {
794 int fd;
795 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
796 if (fd < 0) {
797 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
798 devices[chosen_drive].devname);
799 if (must_close) close(mdfd);
800 return 1;
801 }
802 if (st->ss->store_super(st, fd)) {
803 close(fd);
804 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
805 devices[chosen_drive].devname);
806 if (must_close) close(mdfd);
807 return 1;
808 }
809 close(fd);
810 }
811
812 /* If we are in the middle of a reshape we may need to restore saved data
813 * that was moved aside due to the reshape overwriting live data
814 * The code of doing this lives in Grow.c
815 */
816 #ifndef MDASSEMBLE
817 if (info.reshape_active) {
818 int err = 0;
819 int *fdlist = malloc(sizeof(int)* bestcnt);
820 for (i=0; i<bestcnt; i++) {
821 int j = best[i];
822 if (j >= 0) {
823 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
824 if (fdlist[i] < 0) {
825 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
826 devices[j].devname);
827 err = 1;
828 break;
829 }
830 } else
831 fdlist[i] = -1;
832 }
833 if (!err)
834 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
835 while (i>0) {
836 i--;
837 if (fdlist[i]>=0) close(fdlist[i]);
838 }
839 if (err) {
840 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
841 if (must_close) close(mdfd);
842 return err;
843 }
844 }
845 #endif
846 /* count number of in-sync devices according to the superblock.
847 * We must have this number to start the array without -s or -R
848 */
849 req_cnt = info.array.working_disks;
850
851 /* Almost ready to actually *do* something */
852 if (!old_linux) {
853 int rv;
854
855 rv = set_array_info(mdfd, st, &info);
856 if (rv) {
857 fprintf(stderr, Name ": failed to set array info 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 rv = add_disk(mdfd, st, &info, &devices[j].i);
898
899 if (rv) {
900 fprintf(stderr, Name ": failed to add "
901 "%s to %s: %s\n",
902 devices[j].devname,
903 mddev,
904 strerror(errno));
905 if (i < info.array.raid_disks
906 || i == bestcnt)
907 okcnt--;
908 else
909 sparecnt--;
910 } else if (verbose > 0)
911 fprintf(stderr, Name ": added %s "
912 "to %s as %d\n",
913 devices[j].devname, mddev,
914 devices[j].i.disk.raid_disk);
915 } else if (verbose > 0 && i < info.array.raid_disks)
916 fprintf(stderr, Name ": no uptodate device for "
917 "slot %d of %s\n",
918 i, mddev);
919 }
920
921 if (info.array.level == LEVEL_CONTAINER) {
922 if (verbose >= 0) {
923 fprintf(stderr, Name ": Container %s has been "
924 "assembled with %d drive%s",
925 mddev, okcnt, okcnt==1?"":"s");
926 if (okcnt < info.array.raid_disks)
927 fprintf(stderr, " (out of %d)",
928 info.array.raid_disks);
929 fprintf(stderr, "\n");
930 }
931 if (must_close)
932 close(mdfd);
933 return 0;
934 }
935
936 if (runstop == 1 ||
937 (runstop <= 0 &&
938 ( enough(info.array.level, info.array.raid_disks,
939 info.array.layout, clean, avail, okcnt) &&
940 (okcnt >= req_cnt || start_partial_ok)
941 ))) {
942 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
943 if (verbose >= 0) {
944 fprintf(stderr, Name ": %s has been started with %d drive%s",
945 mddev, okcnt, okcnt==1?"":"s");
946 if (okcnt < info.array.raid_disks)
947 fprintf(stderr, " (out of %d)", info.array.raid_disks);
948 if (sparecnt)
949 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
950 fprintf(stderr, ".\n");
951 }
952 if (must_close) {
953 int usecs = 1;
954 close(mdfd);
955 /* There is a nasty race with 'mdadm --monitor'.
956 * If it opens this device before we close it,
957 * it gets an incomplete open on which IO
958 * doesn't work and the capacity is
959 * wrong.
960 * If we reopen (to check for layered devices)
961 * before --monitor closes, we loose.
962 *
963 * So: wait upto 1 second for there to be
964 * a non-zero capacity.
965 */
966 while (usecs < 1000) {
967 mdfd = open(mddev, O_RDONLY);
968 if (mdfd >= 0) {
969 unsigned long long size;
970 if (get_dev_size(mdfd, NULL, &size) &&
971 size > 0)
972 break;
973 close(mdfd);
974 }
975 usleep(usecs);
976 usecs <<= 1;
977 }
978 }
979 return 0;
980 }
981 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
982 mddev, strerror(errno));
983
984 if (!enough(info.array.level, info.array.raid_disks,
985 info.array.layout, 1, avail, okcnt))
986 fprintf(stderr, Name ": Not enough devices to "
987 "start the array.\n");
988 else if (!enough(info.array.level,
989 info.array.raid_disks,
990 info.array.layout, clean,
991 avail, okcnt))
992 fprintf(stderr, Name ": Not enough devices to "
993 "start the array while not clean "
994 "- consider --force.\n");
995
996 if (must_close) {
997 ioctl(mdfd, STOP_ARRAY, NULL);
998 close(mdfd);
999 }
1000 return 1;
1001 }
1002 if (runstop == -1) {
1003 fprintf(stderr, Name ": %s assembled from %d drive%s",
1004 mddev, okcnt, okcnt==1?"":"s");
1005 if (okcnt != info.array.raid_disks)
1006 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1007 fprintf(stderr, ", but not started.\n");
1008 if (must_close) close(mdfd);
1009 return 0;
1010 }
1011 if (verbose >= -1) {
1012 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1013 if (sparecnt)
1014 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1015 if (!enough(info.array.level, info.array.raid_disks,
1016 info.array.layout, 1, avail, okcnt))
1017 fprintf(stderr, " - not enough to start the array.\n");
1018 else if (!enough(info.array.level,
1019 info.array.raid_disks,
1020 info.array.layout, clean,
1021 avail, okcnt))
1022 fprintf(stderr, " - not enough to start the "
1023 "array while not clean - consider "
1024 "--force.\n");
1025 else {
1026 if (req_cnt == info.array.raid_disks)
1027 fprintf(stderr, " - need all %d to start it", req_cnt);
1028 else
1029 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1030 fprintf(stderr, " (use --run to insist).\n");
1031 }
1032 }
1033 if (must_close) {
1034 ioctl(mdfd, STOP_ARRAY, NULL);
1035 close(mdfd);
1036 }
1037 return 1;
1038 } else {
1039 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1040 * been updated to point to the current locations of devices.
1041 * so we can just start the array
1042 */
1043 unsigned long dev;
1044 dev = makedev(devices[chosen_drive].i.disk.major,
1045 devices[chosen_drive].i.disk.minor);
1046 if (ioctl(mdfd, START_ARRAY, dev)) {
1047 fprintf(stderr, Name ": Cannot start array: %s\n",
1048 strerror(errno));
1049 }
1050
1051 }
1052 if (must_close) close(mdfd);
1053 return 0;
1054 }