65a679b650e1e7af29487c279f53161f51adbd7d
[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 == 1) {
343 /* We can do something */
344 if (first) {/* just ignore this one */
345 if ((inargv && verbose >= 0) || verbose > 0)
346 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
347 devname);
348 goto loop;
349 } else { /* reject all those sofar */
350 mddev_dev_t td;
351 if ((inargv && verbose >= 0) || verbose > 0)
352 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
353 devname);
354 for (td=devlist; td != tmpdev; td=td->next)
355 if (td->used == 1)
356 td->used = 0;
357 tmpdev->used = 1;
358 goto loop;
359 }
360 }
361 }
362 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
363 devname);
364 tst->ss->free_super(tst);
365 st->ss->free_super(st);
366 return 1;
367 }
368
369 tmpdev->used = 1;
370
371 loop:
372 if (tst)
373 tst->ss->free_super(tst);
374 }
375
376 if (mdfd < 0) {
377 /* So... it is up to me to open the device.
378 * We create a name '/dev/md/XXX' based on the info in the
379 * superblock, and call open_mddev on that
380 */
381 mdu_array_info_t inf;
382 char *c;
383 if (!st || !st->sb) {
384 return 2;
385 }
386 st->ss->getinfo_super(st, &info);
387 c = strchr(info.name, ':');
388 if (c) c++; else c= info.name;
389 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
390 /* /dev/md/d0 style for partitionable */
391 asprintf(&mddev, "/dev/md/d%s", c);
392 else
393 asprintf(&mddev, "/dev/md/%s", c);
394 mdfd = open_mddev(mddev, ident->autof);
395 if (mdfd < 0) {
396 st->ss->free_super(st);
397 free(devices);
398 goto try_again;
399 }
400 vers = md_get_version(mdfd);
401 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
402 for (tmpdev = devlist ;
403 tmpdev && tmpdev->used != 1;
404 tmpdev = tmpdev->next)
405 ;
406 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
407 if (tmpdev)
408 fprintf(stderr, Name ": %s needed for %s...\n",
409 mddev, tmpdev->devname);
410 close(mdfd);
411 mdfd = -1;
412 st->ss->free_super(st);
413 free(devices);
414 goto try_again;
415 }
416 must_close = 1;
417 }
418
419 /* Ok, no bad inconsistancy, we can try updating etc */
420 bitmap_done = 0;
421 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
422 char *devname = tmpdev->devname;
423 struct stat stb;
424 /* looks like a good enough match to update the super block if needed */
425 #ifndef MDASSEMBLE
426 if (update) {
427 int dfd;
428 /* prepare useful information in info structures */
429 struct stat stb2;
430 struct supertype *tst;
431 fstat(mdfd, &stb2);
432
433 if (strcmp(update, "uuid")==0 &&
434 !ident->uuid_set) {
435 int rfd;
436 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
437 read(rfd, ident->uuid, 16) != 16) {
438 *(__u32*)(ident->uuid) = random();
439 *(__u32*)(ident->uuid+1) = random();
440 *(__u32*)(ident->uuid+2) = random();
441 *(__u32*)(ident->uuid+3) = random();
442 }
443 if (rfd >= 0) close(rfd);
444 }
445 dfd = dev_open(devname, O_RDWR|O_EXCL);
446
447 remove_partitions(dfd);
448
449 tst = dup_super(st);
450 tst->ss->load_super(tst, dfd, NULL);
451 tst->ss->getinfo_super(tst, &info);
452
453 memcpy(info.uuid, ident->uuid, 16);
454 strcpy(info.name, ident->name);
455 info.array.md_minor = minor(stb2.st_rdev);
456
457 tst->ss->update_super(tst, &info, update,
458 devname, verbose,
459 ident->uuid_set, homehost);
460 if (strcmp(update, "uuid")==0 &&
461 !ident->uuid_set) {
462 ident->uuid_set = 1;
463 memcpy(ident->uuid, info.uuid, 16);
464 }
465 if (dfd < 0)
466 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
467 devname);
468 else if (tst->ss->store_super(tst, dfd))
469 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
470 devname);
471 if (dfd >= 0)
472 close(dfd);
473
474 if (strcmp(update, "uuid")==0 &&
475 ident->bitmap_fd >= 0 && !bitmap_done) {
476 if (bitmap_update_uuid(ident->bitmap_fd,
477 info.uuid,
478 tst->ss->swapuuid) != 0)
479 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
480 else
481 bitmap_done = 1;
482 }
483 tst->ss->free_super(tst);
484 } else
485 #endif
486 {
487 struct supertype *tst = dup_super(st);
488 int dfd;
489 dfd = dev_open(devname, O_RDWR|O_EXCL);
490
491 remove_partitions(dfd);
492
493 tst->ss->load_super(tst, dfd, NULL);
494 tst->ss->getinfo_super(tst, &info);
495 tst->ss->free_super(tst);
496 close(dfd);
497 }
498
499 stat(devname, &stb);
500
501 if (verbose > 0)
502 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
503 devname, mddev, info.disk.raid_disk);
504 devices[devcnt].devname = devname;
505 devices[devcnt].uptodate = 0;
506 devices[devcnt].i = info;
507 devices[devcnt].i.disk.major = major(stb.st_rdev);
508 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
509 if (most_recent < devcnt) {
510 if (devices[devcnt].i.events
511 > devices[most_recent].i.events)
512 most_recent = devcnt;
513 }
514 if (info.array.level == -4)
515 /* with multipath, the raid_disk from the superblock is meaningless */
516 i = devcnt;
517 else
518 i = devices[devcnt].i.disk.raid_disk;
519 if (i+1 == 0) {
520 if (nextspare < info.array.raid_disks)
521 nextspare = info.array.raid_disks;
522 i = nextspare++;
523 } else {
524 if (i >= info.array.raid_disks &&
525 i >= nextspare)
526 nextspare = i+1;
527 }
528 if (i < 10000) {
529 if (i >= bestcnt) {
530 unsigned int newbestcnt = i+10;
531 int *newbest = malloc(sizeof(int)*newbestcnt);
532 unsigned int c;
533 for (c=0; c < newbestcnt; c++)
534 if (c < bestcnt)
535 newbest[c] = best[c];
536 else
537 newbest[c] = -1;
538 if (best)free(best);
539 best = newbest;
540 bestcnt = newbestcnt;
541 }
542 if (best[i] >=0 &&
543 devices[best[i]].i.events
544 == devices[devcnt].i.events
545 && (devices[best[i]].i.disk.minor
546 != devices[devcnt].i.disk.minor)
547 && st->ss == &super0
548 && info.array.level != LEVEL_MULTIPATH) {
549 /* two different devices with identical superblock.
550 * Could be a mis-detection caused by overlapping
551 * partitions. fail-safe.
552 */
553 fprintf(stderr, Name ": WARNING %s and %s appear"
554 " to have very similar superblocks.\n"
555 " If they are really different, "
556 "please --zero the superblock on one\n"
557 " If they are the same or overlap,"
558 " please remove one from %s.\n",
559 devices[best[i]].devname, devname,
560 inargv ? "the list" :
561 "the\n DEVICE list in mdadm.conf"
562 );
563 if (must_close) close(mdfd);
564 return 1;
565 }
566 if (best[i] == -1
567 || (devices[best[i]].i.events
568 < devices[devcnt].i.events))
569 best[i] = devcnt;
570 }
571 devcnt++;
572 }
573
574 if (devcnt == 0) {
575 fprintf(stderr, Name ": no devices found for %s\n",
576 mddev);
577 if (st)
578 st->ss->free_super(st);
579 if (must_close) close(mdfd);
580 return 1;
581 }
582
583 if (update && strcmp(update, "byteorder")==0)
584 st->minor_version = 90;
585
586 st->ss->getinfo_super(st, &info);
587 clean = info.array.state & 1;
588
589 /* now we have some devices that might be suitable.
590 * I wonder how many
591 */
592 avail = malloc(info.array.raid_disks);
593 memset(avail, 0, info.array.raid_disks);
594 okcnt = 0;
595 sparecnt=0;
596 for (i=0; i< bestcnt ;i++) {
597 int j = best[i];
598 int event_margin = 1; /* always allow a difference of '1'
599 * like the kernel does
600 */
601 if (j < 0) continue;
602 /* note: we ignore error flags in multipath arrays
603 * as they don't make sense
604 */
605 if (info.array.level != -4)
606 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
607 if (!(devices[j].i.disk.state
608 & (1<<MD_DISK_FAULTY)))
609 sparecnt++;
610 continue;
611 }
612 if (devices[j].i.events+event_margin >=
613 devices[most_recent].i.events) {
614 devices[j].uptodate = 1;
615 if (i < info.array.raid_disks) {
616 okcnt++;
617 avail[i]=1;
618 } else
619 sparecnt++;
620 }
621 }
622 while (force && !enough(info.array.level, info.array.raid_disks,
623 info.array.layout, 1,
624 avail, okcnt)) {
625 /* Choose the newest best drive which is
626 * not up-to-date, update the superblock
627 * and add it.
628 */
629 int fd;
630 struct supertype *tst;
631 long long current_events;
632 chosen_drive = -1;
633 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
634 int j = best[i];
635 if (j>=0 &&
636 !devices[j].uptodate &&
637 devices[j].i.events > 0 &&
638 (chosen_drive < 0 ||
639 devices[j].i.events
640 > devices[chosen_drive].i.events))
641 chosen_drive = j;
642 }
643 if (chosen_drive < 0)
644 break;
645 current_events = devices[chosen_drive].i.events;
646 add_another:
647 if (verbose >= 0)
648 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
649 devices[chosen_drive].devname,
650 devices[chosen_drive].i.disk.raid_disk,
651 (int)(devices[chosen_drive].i.events),
652 (int)(devices[most_recent].i.events));
653 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
654 if (fd < 0) {
655 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
656 devices[chosen_drive].devname);
657 devices[chosen_drive].i.events = 0;
658 continue;
659 }
660 tst = dup_super(st);
661 if (tst->ss->load_super(tst,fd, NULL)) {
662 close(fd);
663 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
664 devices[chosen_drive].devname);
665 devices[chosen_drive].i.events = 0;
666 continue;
667 }
668 info.events = devices[most_recent].i.events;
669 tst->ss->update_super(tst, &info, "force-one",
670 devices[chosen_drive].devname, verbose,
671 0, NULL);
672
673 if (tst->ss->store_super(tst, fd)) {
674 close(fd);
675 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
676 devices[chosen_drive].devname);
677 devices[chosen_drive].i.events = 0;
678 tst->ss->free_super(tst);
679 continue;
680 }
681 close(fd);
682 devices[chosen_drive].i.events = devices[most_recent].i.events;
683 devices[chosen_drive].uptodate = 1;
684 avail[chosen_drive] = 1;
685 okcnt++;
686 tst->ss->free_super(tst);
687
688 /* If there are any other drives of the same vintage,
689 * add them in as well. We can't lose and we might gain
690 */
691 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
692 int j = best[i];
693 if (j >= 0 &&
694 !devices[j].uptodate &&
695 devices[j].i.events > 0 &&
696 devices[j].i.events == current_events) {
697 chosen_drive = j;
698 goto add_another;
699 }
700 }
701 }
702
703 /* Now we want to look at the superblock which the kernel will base things on
704 * and compare the devices that we think are working with the devices that the
705 * superblock thinks are working.
706 * If there are differences and --force is given, then update this chosen
707 * superblock.
708 */
709 chosen_drive = -1;
710 st->ss->free_super(st);
711 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
712 int j = best[i];
713 int fd;
714
715 if (j<0)
716 continue;
717 if (!devices[j].uptodate)
718 continue;
719 chosen_drive = j;
720 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
721 fprintf(stderr, Name ": Cannot open %s: %s\n",
722 devices[j].devname, strerror(errno));
723 if (must_close) close(mdfd);
724 return 1;
725 }
726 if (st->ss->load_super(st,fd, NULL)) {
727 close(fd);
728 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
729 devices[j].devname);
730 if (must_close) close(mdfd);
731 return 1;
732 }
733 close(fd);
734 }
735 if (st->sb == NULL) {
736 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
737 if (must_close) close(mdfd);
738 return 1;
739 }
740 st->ss->getinfo_super(st, &info);
741 #ifndef MDASSEMBLE
742 sysfs_init(&info, mdfd, 0);
743 #endif
744 for (i=0; i<bestcnt; i++) {
745 int j = best[i];
746 unsigned int desired_state;
747
748 if (i < info.array.raid_disks)
749 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
750 else
751 desired_state = 0;
752
753 if (j<0)
754 continue;
755 if (!devices[j].uptodate)
756 continue;
757
758 devices[j].i.disk.state = desired_state;
759
760 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
761 verbose, 0, NULL)) {
762 if (force) {
763 if (verbose >= 0)
764 fprintf(stderr, Name ": "
765 "clearing FAULTY flag for device %d in %s for %s\n",
766 j, mddev, devices[j].devname);
767 change = 1;
768 } else {
769 if (verbose >= -1)
770 fprintf(stderr, Name ": "
771 "device %d in %s has wrong state in superblock, but %s seems ok\n",
772 i, mddev, devices[j].devname);
773 }
774 }
775 #if 0
776 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
777 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
778 i, mddev);
779 }
780 #endif
781 }
782 if (force && !clean &&
783 !enough(info.array.level, info.array.raid_disks,
784 info.array.layout, clean,
785 avail, okcnt)) {
786 change += st->ss->update_super(st, &info, "force-array",
787 devices[chosen_drive].devname, verbose,
788 0, NULL);
789 clean = 1;
790 }
791
792 if (change) {
793 int fd;
794 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
795 if (fd < 0) {
796 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
797 devices[chosen_drive].devname);
798 if (must_close) close(mdfd);
799 return 1;
800 }
801 if (st->ss->store_super(st, fd)) {
802 close(fd);
803 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
804 devices[chosen_drive].devname);
805 if (must_close) close(mdfd);
806 return 1;
807 }
808 close(fd);
809 }
810
811 /* If we are in the middle of a reshape we may need to restore saved data
812 * that was moved aside due to the reshape overwriting live data
813 * The code of doing this lives in Grow.c
814 */
815 #ifndef MDASSEMBLE
816 if (info.reshape_active) {
817 int err = 0;
818 int *fdlist = malloc(sizeof(int)* bestcnt);
819 for (i=0; i<bestcnt; i++) {
820 int j = best[i];
821 if (j >= 0) {
822 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
823 if (fdlist[i] < 0) {
824 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
825 devices[j].devname);
826 err = 1;
827 break;
828 }
829 } else
830 fdlist[i] = -1;
831 }
832 if (!err)
833 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
834 while (i>0) {
835 i--;
836 if (fdlist[i]>=0) close(fdlist[i]);
837 }
838 if (err) {
839 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
840 if (must_close) close(mdfd);
841 return err;
842 }
843 }
844 #endif
845 /* count number of in-sync devices according to the superblock.
846 * We must have this number to start the array without -s or -R
847 */
848 req_cnt = info.array.working_disks;
849
850 /* Almost ready to actually *do* something */
851 if (!old_linux) {
852 int rv;
853
854 rv = set_array_info(mdfd, st, &info);
855 if (rv) {
856 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
857 mddev, strerror(errno));
858 if (must_close) close(mdfd);
859 return 1;
860 }
861 if (ident->bitmap_fd >= 0) {
862 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
863 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
864 if (must_close) close(mdfd);
865 return 1;
866 }
867 } else if (ident->bitmap_file) {
868 /* From config file */
869 int bmfd = open(ident->bitmap_file, O_RDWR);
870 if (bmfd < 0) {
871 fprintf(stderr, Name ": Could not open bitmap file %s\n",
872 ident->bitmap_file);
873 if (must_close) close(mdfd);
874 return 1;
875 }
876 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
877 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
878 close(bmfd);
879 if (must_close) close(mdfd);
880 return 1;
881 }
882 close(bmfd);
883 }
884
885 /* First, add the raid disks, but add the chosen one last */
886 for (i=0; i<= bestcnt; i++) {
887 int j;
888 if (i < bestcnt) {
889 j = best[i];
890 if (j == chosen_drive)
891 continue;
892 } else
893 j = chosen_drive;
894
895 if (j >= 0 /* && devices[j].uptodate */) {
896 rv = add_disk(mdfd, st, &info, &devices[j].i);
897
898 if (rv) {
899 fprintf(stderr, Name ": failed to add "
900 "%s to %s: %s\n",
901 devices[j].devname,
902 mddev,
903 strerror(errno));
904 if (i < info.array.raid_disks
905 || i == bestcnt)
906 okcnt--;
907 else
908 sparecnt--;
909 } else if (verbose > 0)
910 fprintf(stderr, Name ": added %s "
911 "to %s as %d\n",
912 devices[j].devname, mddev,
913 devices[j].i.disk.raid_disk);
914 } else if (verbose > 0 && i < info.array.raid_disks)
915 fprintf(stderr, Name ": no uptodate device for "
916 "slot %d of %s\n",
917 i, mddev);
918 }
919
920 if (info.array.level == LEVEL_CONTAINER) {
921 if (verbose >= 0) {
922 fprintf(stderr, Name ": Container %s has been "
923 "assembled with %d drive%s",
924 mddev, okcnt, okcnt==1?"":"s");
925 if (okcnt < info.array.raid_disks)
926 fprintf(stderr, " (out of %d)",
927 info.array.raid_disks);
928 fprintf(stderr, "\n");
929 }
930 if (must_close)
931 close(mdfd);
932 return 0;
933 }
934
935 if (runstop == 1 ||
936 (runstop <= 0 &&
937 ( enough(info.array.level, info.array.raid_disks,
938 info.array.layout, clean, avail, okcnt) &&
939 (okcnt >= req_cnt || start_partial_ok)
940 ))) {
941 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
942 if (verbose >= 0) {
943 fprintf(stderr, Name ": %s has been started with %d drive%s",
944 mddev, okcnt, okcnt==1?"":"s");
945 if (okcnt < info.array.raid_disks)
946 fprintf(stderr, " (out of %d)", info.array.raid_disks);
947 if (sparecnt)
948 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
949 fprintf(stderr, ".\n");
950 }
951 if (must_close) {
952 int usecs = 1;
953 close(mdfd);
954 /* There is a nasty race with 'mdadm --monitor'.
955 * If it opens this device before we close it,
956 * it gets an incomplete open on which IO
957 * doesn't work and the capacity is
958 * wrong.
959 * If we reopen (to check for layered devices)
960 * before --monitor closes, we loose.
961 *
962 * So: wait upto 1 second for there to be
963 * a non-zero capacity.
964 */
965 while (usecs < 1000) {
966 mdfd = open(mddev, O_RDONLY);
967 if (mdfd >= 0) {
968 unsigned long long size;
969 if (get_dev_size(mdfd, NULL, &size) &&
970 size > 0)
971 break;
972 close(mdfd);
973 }
974 usleep(usecs);
975 usecs <<= 1;
976 }
977 }
978 return 0;
979 }
980 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
981 mddev, strerror(errno));
982
983 if (!enough(info.array.level, info.array.raid_disks,
984 info.array.layout, 1, avail, okcnt))
985 fprintf(stderr, Name ": Not enough devices to "
986 "start the array.\n");
987 else if (!enough(info.array.level,
988 info.array.raid_disks,
989 info.array.layout, clean,
990 avail, okcnt))
991 fprintf(stderr, Name ": Not enough devices to "
992 "start the array while not clean "
993 "- consider --force.\n");
994
995 if (must_close) {
996 ioctl(mdfd, STOP_ARRAY, NULL);
997 close(mdfd);
998 }
999 return 1;
1000 }
1001 if (runstop == -1) {
1002 fprintf(stderr, Name ": %s assembled from %d drive%s",
1003 mddev, okcnt, okcnt==1?"":"s");
1004 if (okcnt != info.array.raid_disks)
1005 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1006 fprintf(stderr, ", but not started.\n");
1007 if (must_close) close(mdfd);
1008 return 0;
1009 }
1010 if (verbose >= -1) {
1011 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1012 if (sparecnt)
1013 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1014 if (!enough(info.array.level, info.array.raid_disks,
1015 info.array.layout, 1, avail, okcnt))
1016 fprintf(stderr, " - not enough to start the array.\n");
1017 else if (!enough(info.array.level,
1018 info.array.raid_disks,
1019 info.array.layout, clean,
1020 avail, okcnt))
1021 fprintf(stderr, " - not enough to start the "
1022 "array while not clean - consider "
1023 "--force.\n");
1024 else {
1025 if (req_cnt == info.array.raid_disks)
1026 fprintf(stderr, " - need all %d to start it", req_cnt);
1027 else
1028 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1029 fprintf(stderr, " (use --run to insist).\n");
1030 }
1031 }
1032 if (must_close) {
1033 ioctl(mdfd, STOP_ARRAY, NULL);
1034 close(mdfd);
1035 }
1036 return 1;
1037 } else {
1038 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1039 * been updated to point to the current locations of devices.
1040 * so we can just start the array
1041 */
1042 unsigned long dev;
1043 dev = makedev(devices[chosen_drive].i.disk.major,
1044 devices[chosen_drive].i.disk.minor);
1045 if (ioctl(mdfd, START_ARRAY, dev)) {
1046 fprintf(stderr, Name ": Cannot start array: %s\n",
1047 strerror(errno));
1048 }
1049
1050 }
1051 if (must_close) close(mdfd);
1052 return 0;
1053 }