]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
--wait or -W will wait for resync activity to finish on the given devices.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@cse.unsw.edu.au>
23 * Paper: Neil Brown
24 * School of Computer Science and Engineering
25 * The University of New South Wales
26 * Sydney, 2052
27 * Australia
28 */
29
30 #include "mdadm.h"
31 #include <ctype.h>
32
33 static int name_matches(char *found, char *required, char *homehost)
34 {
35 /* See if the name found matches the required name, possibly
36 * prefixed with 'homehost'
37 */
38 char fnd[33];
39
40 strncpy(fnd, found, 32);
41 fnd[32] = 0;
42 if (strcmp(found, required)==0)
43 return 1;
44 if (homehost) {
45 int l = strlen(homehost);
46 if (l < 32 && fnd[l] == ':' &&
47 strcmp(fnd+l+1, required)==0)
48 return 1;
49 }
50 return 0;
51 }
52
53 int Assemble(struct supertype *st, char *mddev, int mdfd,
54 mddev_ident_t ident,
55 mddev_dev_t devlist, char *backup_file,
56 int readonly, int runstop,
57 char *update, char *homehost,
58 int verbose, int force)
59 {
60 /*
61 * The task of Assemble is to find a collection of
62 * devices that should (according to their superblocks)
63 * form an array, and to give this collection to the MD driver.
64 * In Linux-2.4 and later, this involves submitting a
65 * SET_ARRAY_INFO ioctl with no arg - to prepare
66 * the array - and then submit a number of
67 * ADD_NEW_DISK ioctls to add disks into
68 * the array. Finally RUN_ARRAY might
69 * be submitted to start the array.
70 *
71 * Much of the work of Assemble is in finding and/or
72 * checking the disks to make sure they look right.
73 *
74 * If mddev is not set, then scan must be set and we
75 * read through the config file for dev+uuid mapping
76 * We recurse, setting mddev, for each device that
77 * - isn't running
78 * - has a valid uuid (or any uuid if !uuidset)
79 *
80 * If mddev is set, we try to determine state of md.
81 * check version - must be at least 0.90.0
82 * check kernel version. must be at least 2.4.
83 * If not, we can possibly fall back on START_ARRAY
84 * Try to GET_ARRAY_INFO.
85 * If possible, give up
86 * If not, try to STOP_ARRAY just to make sure
87 *
88 * If !uuidset and scan, look in conf-file for uuid
89 * If not found, give up
90 * If !devlist and scan and uuidset, get list of devs from conf-file
91 *
92 * For each device:
93 * Check superblock - discard if bad
94 * Check uuid (set if we don't have one) - discard if no match
95 * Check superblock similarity if we have a superblock - discard if different
96 * Record events, devicenum
97 * This should give us a list of devices for the array
98 * We should collect the most recent event number
99 *
100 * Count disks with recent enough event count
101 * While force && !enough disks
102 * Choose newest rejected disks, update event count
103 * mark clean and rewrite superblock
104 * If recent kernel:
105 * SET_ARRAY_INFO
106 * foreach device with recent events : ADD_NEW_DISK
107 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
108 * If old kernel:
109 * Check the device numbers in superblock are right
110 * update superblock if any changes
111 * START_ARRAY
112 *
113 */
114 int clean = 0;
115 int must_close = 0;
116 int old_linux = 0;
117 int vers = 0; /* Keep gcc quite - it really is initialised */
118 void *first_super = NULL, *super = NULL;
119 struct {
120 char *devname;
121 unsigned int major, minor;
122 unsigned int oldmajor, oldminor;
123 long long events;
124 int uptodate;
125 int state;
126 int raid_disk;
127 int disk_nr;
128 } *devices;
129 int *best = NULL; /* indexed by raid_disk */
130 unsigned int bestcnt = 0;
131 int devcnt = 0;
132 unsigned int okcnt, sparecnt;
133 unsigned int req_cnt;
134 unsigned int i;
135 int most_recent = 0;
136 int chosen_drive;
137 int change = 0;
138 int inargv = 0;
139 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
140 unsigned int num_devs;
141 mddev_dev_t tmpdev;
142 struct mdinfo info;
143 char *avail;
144 int nextspare = 0;
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 = 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 if (super) {
229 free(super);
230 super = NULL;
231 }
232
233 dfd = dev_open(devname, O_RDONLY|O_EXCL);
234 if (dfd < 0) {
235 if ((inargv && verbose >= 0) || verbose > 0)
236 fprintf(stderr, Name ": cannot open device %s: %s\n",
237 devname, strerror(errno));
238 tmpdev->used = 2;
239 } else if (fstat(dfd, &stb)< 0) {
240 /* Impossible! */
241 fprintf(stderr, Name ": fstat failed for %s: %s\n",
242 devname, strerror(errno));
243 tmpdev->used = 2;
244 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
245 fprintf(stderr, Name ": %s is not a block device.\n",
246 devname);
247 tmpdev->used = 2;
248 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
249 if ((inargv && verbose >= 0) || verbose > 0)
250 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
251 devname);
252 tmpdev->used = 2;
253 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
254 if ((inargv && verbose >= 0) || verbose > 0)
255 fprintf( stderr, Name ": no RAID superblock on %s\n",
256 devname);
257 } else {
258 tst->ss->getinfo_super(&info, super);
259 }
260 if (dfd >= 0) close(dfd);
261
262 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
263 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
264 if ((inargv && verbose >= 0) || verbose > 0)
265 fprintf(stderr, Name ": %s has wrong uuid.\n",
266 devname);
267 continue;
268 }
269 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
270 (!super || name_matches(info.name, ident->name, homehost)==0)) {
271 if ((inargv && verbose >= 0) || verbose > 0)
272 fprintf(stderr, Name ": %s has wrong name.\n",
273 devname);
274 continue;
275 }
276 if (ident->super_minor != UnSet &&
277 (!super || ident->super_minor != info.array.md_minor)) {
278 if ((inargv && verbose >= 0) || verbose > 0)
279 fprintf(stderr, Name ": %s has wrong super-minor.\n",
280 devname);
281 continue;
282 }
283 if (ident->level != UnSet &&
284 (!super|| ident->level != info.array.level)) {
285 if ((inargv && verbose >= 0) || verbose > 0)
286 fprintf(stderr, Name ": %s has wrong raid level.\n",
287 devname);
288 continue;
289 }
290 if (ident->raid_disks != UnSet &&
291 (!super || 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 continue;
296 }
297 if (mdfd < 0) {
298 if (tst == NULL || super == NULL)
299 continue;
300 if (update == NULL &&
301 tst->ss->match_home(super, homehost)==0) {
302 if ((inargv && verbose >= 0) || verbose > 0)
303 fprintf(stderr, Name ": %s is not built for host %s.\n",
304 devname, homehost);
305 /* Auto-assemble, and this is not a usable host */
306 /* if update != NULL, we are updating the host
307 * name... */
308 continue;
309 }
310 }
311 /* If we are this far, then we are nearly commited to this device.
312 * If the super_block doesn't exist, or doesn't match others,
313 * then we probably cannot continue
314 * However if one of the arrays is for the homehost, and
315 * the other isn't that can disambiguate.
316 */
317
318 if (!super) {
319 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
320 devname);
321 free(first_super);
322 return 1;
323 }
324
325 if (st == NULL)
326 st = tst;
327 if (st->ss != tst->ss ||
328 st->minor_version != tst->minor_version ||
329 st->ss->compare_super(&first_super, super) != 0) {
330 /* Some mismatch. If exactly one array matches this host,
331 * we can resolve on that one.
332 * Or, if we are auto assembling, we just ignore the second
333 * for now.
334 */
335 if (mdfd < 0)
336 continue;
337 if (homehost) {
338 int first = st->ss->match_home(first_super, homehost);
339 int last = tst->ss->match_home(super, homehost);
340 if (first+last == 1) {
341 /* We can do something */
342 if (first) {/* just ignore this one */
343 if ((inargv && verbose >= 0) || verbose > 0)
344 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
345 devname);
346 continue;
347 } else { /* reject all those sofar */
348 mddev_dev_t td;
349 if ((inargv && verbose >= 0) || verbose > 0)
350 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
351 devname);
352 for (td=devlist; td != tmpdev; td=td->next)
353 if (td->used == 1)
354 td->used = 0;
355 tmpdev->used = 1;
356 continue;
357 }
358 }
359 }
360 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
361 devname);
362 free(super);
363 free(first_super);
364 return 1;
365 }
366
367 tmpdev->used = 1;
368 }
369
370 if (mdfd < 0) {
371 /* So... it is up to me to open the device.
372 * We create a name '/dev/md/XXX' based on the info in the
373 * superblock, and call open_mddev on that
374 */
375 mdu_array_info_t inf;
376 char *c;
377 if (!first_super) {
378 return 2;
379 }
380 st->ss->getinfo_super(&info, first_super);
381 c = strchr(info.name, ':');
382 if (c) c++; else c= info.name;
383 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
384 /* /dev/md/d0 style for partitionable */
385 asprintf(&mddev, "/dev/md/d%s", c);
386 else
387 asprintf(&mddev, "/dev/md/%s", c);
388 mdfd = open_mddev(mddev, ident->autof);
389 if (mdfd < 0) {
390 free(first_super);
391 free(devices);
392 first_super = NULL;
393 goto try_again;
394 }
395 vers = md_get_version(mdfd);
396 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
397 for (tmpdev = devlist ;
398 tmpdev && tmpdev->used != 1;
399 tmpdev = tmpdev->next)
400 ;
401 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
402 if (tmpdev)
403 fprintf(stderr, Name ": %s needed for %s...\n",
404 mddev, tmpdev->devname);
405 close(mdfd);
406 mdfd = -1;
407 free(first_super);
408 free(devices);
409 first_super = NULL;
410 goto try_again;
411 }
412 must_close = 1;
413 }
414
415 /* Ok, no bad inconsistancy, we can try updating etc */
416 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
417 char *devname = tmpdev->devname;
418 struct stat stb;
419 /* looks like a good enough match to update the super block if needed */
420 #ifndef MDASSEMBLE
421 if (update) {
422 int dfd;
423 /* prepare useful information in info structures */
424 struct stat stb2;
425 fstat(mdfd, &stb2);
426
427 if (strcmp(update, "uuid")==0 &&
428 !ident->uuid_set) {
429 int rfd;
430 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
431 read(rfd, ident->uuid, 16) != 16) {
432 *(__u32*)(ident->uuid) = random();
433 *(__u32*)(ident->uuid+1) = random();
434 *(__u32*)(ident->uuid+2) = random();
435 *(__u32*)(ident->uuid+3) = random();
436 }
437 if (rfd >= 0) close(rfd);
438 }
439 dfd = dev_open(devname, O_RDWR|O_EXCL);
440
441 remove_partitions(dfd);
442
443 if (super) {
444 free(super);
445 super = NULL;
446 }
447
448 st->ss->load_super(st, dfd, &super, NULL);
449 st->ss->getinfo_super(&info, super);
450
451 memcpy(info.uuid, ident->uuid, 16);
452 strcpy(info.name, ident->name);
453 info.array.md_minor = minor(stb2.st_rdev);
454
455 st->ss->update_super(&info, super, update, devname, verbose,
456 ident->uuid_set, homehost);
457 if (strcmp(update, "uuid")==0 &&
458 !ident->uuid_set) {
459 ident->uuid_set = 1;
460 memcpy(ident->uuid, info.uuid, 16);
461 }
462 if (dfd < 0)
463 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
464 devname);
465 else if (st->ss->store_super(st, dfd, super))
466 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
467 devname);
468 if (dfd >= 0)
469 close(dfd);
470
471 if (strcmp(update, "uuid")==0 &&
472 ident->bitmap_fd)
473 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid) != 0)
474 fprintf(stderr, Name ": Could not update uuid on %s.\n",
475 devname);
476 } else
477 #endif
478 {
479 int dfd;
480 dfd = dev_open(devname, O_RDWR|O_EXCL);
481
482 remove_partitions(dfd);
483
484 if (super) {
485 free(super);
486 super = NULL;
487 }
488
489 st->ss->load_super(st, dfd, &super, NULL);
490 st->ss->getinfo_super(&info, super);
491 close(dfd);
492 }
493
494 stat(devname, &stb);
495
496 if (verbose > 0)
497 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
498 devname, mddev, info.disk.raid_disk);
499 devices[devcnt].devname = devname;
500 devices[devcnt].major = major(stb.st_rdev);
501 devices[devcnt].minor = minor(stb.st_rdev);
502 devices[devcnt].oldmajor = info.disk.major;
503 devices[devcnt].oldminor = info.disk.minor;
504 devices[devcnt].events = info.events;
505 devices[devcnt].raid_disk = info.disk.raid_disk;
506 devices[devcnt].disk_nr = info.disk.number;
507 devices[devcnt].uptodate = 0;
508 devices[devcnt].state = info.disk.state;
509 if (most_recent < devcnt) {
510 if (devices[devcnt].events
511 > devices[most_recent].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].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]].events == devices[devcnt].events &&
544 devices[best[i]].minor != devices[devcnt].minor &&
545 st->ss->major == 0 &&
546 info.array.level != -4) {
547 /* two different devices with identical superblock.
548 * Could be a mis-detection caused by overlapping
549 * partitions. fail-safe.
550 */
551 fprintf(stderr, Name ": WARNING %s and %s appear"
552 " to have very similar superblocks.\n"
553 " If they are really different, "
554 "please --zero the superblock on one\n"
555 " If they are the same or overlap,"
556 " please remove one from %s.\n",
557 devices[best[i]].devname, devname,
558 inargv ? "the list" :
559 "the\n DEVICE list in mdadm.conf"
560 );
561 if (must_close) close(mdfd);
562 return 1;
563 }
564 if (best[i] == -1
565 || devices[best[i]].events < devices[devcnt].events)
566 best[i] = devcnt;
567 }
568 devcnt++;
569 }
570
571 if (super)
572 free(super);
573 super = NULL;
574
575 if (update && strcmp(update, "byteorder")==0)
576 st->minor_version = 90;
577
578 if (devcnt == 0) {
579 fprintf(stderr, Name ": no devices found for %s\n",
580 mddev);
581 free(first_super);
582 if (must_close) close(mdfd);
583 return 1;
584 }
585
586 st->ss->getinfo_super(&info, first_super);
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].state & (1<<MD_DISK_SYNC))) {
607 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
608 sparecnt++;
609 continue;
610 }
611 if (devices[j].events+event_margin >=
612 devices[most_recent].events) {
613 devices[j].uptodate = 1;
614 if (i < info.array.raid_disks) {
615 okcnt++;
616 avail[i]=1;
617 } else
618 sparecnt++;
619 }
620 }
621 while (force && !enough(info.array.level, info.array.raid_disks,
622 info.array.layout, 1,
623 avail, okcnt)) {
624 /* Choose the newest best drive which is
625 * not up-to-date, update the superblock
626 * and add it.
627 */
628 int fd;
629 chosen_drive = -1;
630 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
631 int j = best[i];
632 if (j>=0 &&
633 !devices[j].uptodate &&
634 devices[j].events > 0 &&
635 (chosen_drive < 0 ||
636 devices[j].events > devices[chosen_drive].events))
637 chosen_drive = j;
638 }
639 if (chosen_drive < 0)
640 break;
641 if (verbose >= 0)
642 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
643 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
644 (int)(devices[chosen_drive].events),
645 (int)(devices[most_recent].events));
646 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
647 if (fd < 0) {
648 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
649 devices[chosen_drive].devname);
650 devices[chosen_drive].events = 0;
651 continue;
652 }
653 if (st->ss->load_super(st,fd, &super, NULL)) {
654 close(fd);
655 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
656 devices[chosen_drive].devname);
657 devices[chosen_drive].events = 0;
658 continue;
659 }
660 info.events = devices[most_recent].events;
661 st->ss->update_super(&info, super, "force-one",
662 devices[chosen_drive].devname, verbose,
663 0, NULL);
664
665 if (st->ss->store_super(st, fd, super)) {
666 close(fd);
667 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
668 devices[chosen_drive].devname);
669 devices[chosen_drive].events = 0;
670 free(super);
671 continue;
672 }
673 close(fd);
674 devices[chosen_drive].events = devices[most_recent].events;
675 devices[chosen_drive].uptodate = 1;
676 avail[chosen_drive] = 1;
677 okcnt++;
678 free(super);
679 }
680
681 /* Now we want to look at the superblock which the kernel will base things on
682 * and compare the devices that we think are working with the devices that the
683 * superblock thinks are working.
684 * If there are differences and --force is given, then update this chosen
685 * superblock.
686 */
687 chosen_drive = -1;
688 super = NULL;
689 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
690 int j = best[i];
691 int fd;
692
693 if (j<0)
694 continue;
695 if (!devices[j].uptodate)
696 continue;
697 chosen_drive = j;
698 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
699 fprintf(stderr, Name ": Cannot open %s: %s\n",
700 devices[j].devname, strerror(errno));
701 if (must_close) close(mdfd);
702 return 1;
703 }
704 if (st->ss->load_super(st,fd, &super, NULL)) {
705 close(fd);
706 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
707 devices[j].devname);
708 if (must_close) close(mdfd);
709 return 1;
710 }
711 close(fd);
712 }
713 if (super == NULL) {
714 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
715 if (must_close) close(mdfd);
716 return 1;
717 }
718 st->ss->getinfo_super(&info, super);
719 for (i=0; i<bestcnt; i++) {
720 int j = best[i];
721 unsigned int desired_state;
722
723 if (i < info.array.raid_disks)
724 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
725 else
726 desired_state = 0;
727
728 if (j<0)
729 continue;
730 if (!devices[j].uptodate)
731 continue;
732 info.disk.number = devices[j].disk_nr;
733 info.disk.raid_disk = i;
734 info.disk.state = desired_state;
735
736 if (devices[j].uptodate &&
737 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
738 if (force) {
739 if (verbose >= 0)
740 fprintf(stderr, Name ": "
741 "clearing FAULTY flag for device %d in %s for %s\n",
742 j, mddev, devices[j].devname);
743 change = 1;
744 } else {
745 if (verbose >= -1)
746 fprintf(stderr, Name ": "
747 "device %d in %s has wrong state in superblock, but %s seems ok\n",
748 i, mddev, devices[j].devname);
749 }
750 }
751 #if 0
752 if (!devices[j].uptodate &&
753 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
754 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
755 i, mddev);
756 }
757 #endif
758 }
759 if (force && okcnt < info.array.raid_disks) {
760 change += st->ss->update_super(&info, super, "force-array",
761 devices[chosen_drive].devname, verbose,
762 0, NULL);
763 clean = 1;
764 }
765
766 if (change) {
767 int fd;
768 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
769 if (fd < 0) {
770 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
771 devices[chosen_drive].devname);
772 if (must_close) close(mdfd);
773 return 1;
774 }
775 if (st->ss->store_super(st, fd, super)) {
776 close(fd);
777 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
778 devices[chosen_drive].devname);
779 if (must_close) close(mdfd);
780 return 1;
781 }
782 close(fd);
783 }
784
785 /* If we are in the middle of a reshape we may need to restore saved data
786 * that was moved aside due to the reshape overwriting live data
787 * The code of doing this lives in Grow.c
788 */
789 #ifndef MDASSEMBLE
790 if (info.reshape_active) {
791 int err = 0;
792 int *fdlist = malloc(sizeof(int)* bestcnt);
793 for (i=0; i<bestcnt; i++) {
794 int j = best[i];
795 if (j >= 0) {
796 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
797 if (fdlist[i] < 0) {
798 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
799 devices[j].devname);
800 err = 1;
801 break;
802 }
803 } else
804 fdlist[i] = -1;
805 }
806 if (!err)
807 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
808 while (i>0) {
809 i--;
810 if (fdlist[i]>=0) close(fdlist[i]);
811 }
812 if (err) {
813 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
814 if (must_close) close(mdfd);
815 return err;
816 }
817 }
818 #endif
819 /* count number of in-sync devices according to the superblock.
820 * We must have this number to start the array without -s or -R
821 */
822 req_cnt = info.array.working_disks;
823
824 /* Almost ready to actually *do* something */
825 if (!old_linux) {
826 int rv;
827 if ((vers % 100) >= 1) { /* can use different versions */
828 mdu_array_info_t inf;
829 memset(&inf, 0, sizeof(inf));
830 inf.major_version = st->ss->major;
831 inf.minor_version = st->minor_version;
832 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
833 } else
834 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
835
836 if (rv) {
837 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
838 mddev, strerror(errno));
839 if (must_close) close(mdfd);
840 return 1;
841 }
842 if (ident->bitmap_fd >= 0) {
843 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
844 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
845 if (must_close) close(mdfd);
846 return 1;
847 }
848 } else if (ident->bitmap_file) {
849 /* From config file */
850 int bmfd = open(ident->bitmap_file, O_RDWR);
851 if (bmfd < 0) {
852 fprintf(stderr, Name ": Could not open bitmap file %s\n",
853 ident->bitmap_file);
854 if (must_close) close(mdfd);
855 return 1;
856 }
857 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
858 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
859 close(bmfd);
860 if (must_close) close(mdfd);
861 return 1;
862 }
863 close(bmfd);
864 }
865
866 /* First, add the raid disks, but add the chosen one last */
867 for (i=0; i<= bestcnt; i++) {
868 int j;
869 if (i < bestcnt) {
870 j = best[i];
871 if (j == chosen_drive)
872 continue;
873 } else
874 j = chosen_drive;
875
876 if (j >= 0 /* && devices[j].uptodate */) {
877 mdu_disk_info_t disk;
878 memset(&disk, 0, sizeof(disk));
879 disk.major = devices[j].major;
880 disk.minor = devices[j].minor;
881 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
882 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
883 devices[j].devname,
884 mddev,
885 strerror(errno));
886 if (i < info.array.raid_disks || i == bestcnt)
887 okcnt--;
888 else
889 sparecnt--;
890 } else if (verbose > 0)
891 fprintf(stderr, Name ": added %s to %s as %d\n",
892 devices[j].devname, mddev, devices[j].raid_disk);
893 } else if (verbose > 0 && i < info.array.raid_disks)
894 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
895 i, mddev);
896 }
897
898 if (runstop == 1 ||
899 (runstop <= 0 &&
900 ( enough(info.array.level, info.array.raid_disks,
901 info.array.layout, clean, avail, okcnt) &&
902 (okcnt >= req_cnt || start_partial_ok)
903 ))) {
904 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
905 if (verbose >= 0) {
906 fprintf(stderr, Name ": %s has been started with %d drive%s",
907 mddev, okcnt, okcnt==1?"":"s");
908 if (okcnt < info.array.raid_disks)
909 fprintf(stderr, " (out of %d)", info.array.raid_disks);
910 if (sparecnt)
911 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
912 fprintf(stderr, ".\n");
913 }
914 if (must_close) {
915 int usecs = 1;
916 close(mdfd);
917 /* There is a nasty race with 'mdadm --monitor'.
918 * If it opens this device before we close it,
919 * it gets an incomplete open on which IO
920 * doesn't work and the capacity if wrong.
921 * If we reopen (to check for layered devices)
922 * before --monitor closes, we loose.
923 *
924 * So: wait upto 1 second for there to be
925 * a non-zero capacity.
926 */
927 while (usecs < 1000) {
928 mdfd = open(mddev, O_RDONLY);
929 if (mdfd >= 0) {
930 unsigned long size;
931 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
932 size > 0)
933 break;
934 close(mdfd);
935 }
936 usleep(usecs);
937 usecs <<= 1;
938 }
939 }
940 return 0;
941 }
942 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
943 mddev, strerror(errno));
944
945 if (!enough(info.array.level, info.array.raid_disks,
946 info.array.layout, 1, avail, okcnt))
947 fprintf(stderr, Name ": Not enough devices to "
948 "start the array.\n");
949 else if (!enough(info.array.level,
950 info.array.raid_disks,
951 info.array.layout, clean,
952 avail, okcnt))
953 fprintf(stderr, Name ": Not enough devices to "
954 "start the array while not clean "
955 "- consider --force.\n");
956
957 if (must_close) close(mdfd);
958 return 1;
959 }
960 if (runstop == -1) {
961 fprintf(stderr, Name ": %s assembled from %d drive%s",
962 mddev, okcnt, okcnt==1?"":"s");
963 if (okcnt != info.array.raid_disks)
964 fprintf(stderr, " (out of %d)", info.array.raid_disks);
965 fprintf(stderr, ", but not started.\n");
966 if (must_close) close(mdfd);
967 return 0;
968 }
969 if (verbose >= 0) {
970 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
971 if (sparecnt)
972 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
973 if (!enough(info.array.level, info.array.raid_disks,
974 info.array.layout, 1, avail, okcnt))
975 fprintf(stderr, " - not enough to start the array.\n");
976 else if (!enough(info.array.level,
977 info.array.raid_disks,
978 info.array.layout, clean,
979 avail, okcnt))
980 fprintf(stderr, " - not enough to start the "
981 "array while not clean - consider "
982 "--force.\n");
983 else {
984 if (req_cnt == info.array.raid_disks)
985 fprintf(stderr, " - need all %d to start it", req_cnt);
986 else
987 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
988 fprintf(stderr, " (use --run to insist).\n");
989 }
990 }
991 if (must_close) close(mdfd);
992 return 1;
993 } else {
994 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
995 * been updated to point to the current locations of devices.
996 * so we can just start the array
997 */
998 unsigned long dev;
999 dev = makedev(devices[chosen_drive].major,
1000 devices[chosen_drive].minor);
1001 if (ioctl(mdfd, START_ARRAY, dev)) {
1002 fprintf(stderr, Name ": Cannot start array: %s\n",
1003 strerror(errno));
1004 }
1005
1006 }
1007 if (must_close) close(mdfd);
1008 return 0;
1009 }