]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Avoid misdetection of overlapping partitions...
[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, char *conffile,
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 must_close = 0;
115 int old_linux = 0;
116 int vers = 0; /* Keep gcc quite - it really is initialised */
117 void *first_super = NULL, *super = NULL;
118 struct {
119 char *devname;
120 unsigned int major, minor;
121 unsigned int oldmajor, oldminor;
122 long long events;
123 int uptodate;
124 int state;
125 int raid_disk;
126 int disk_nr;
127 } *devices;
128 int *best = NULL; /* indexed by raid_disk */
129 unsigned int bestcnt = 0;
130 int devcnt = 0;
131 unsigned int okcnt, sparecnt;
132 unsigned int req_cnt;
133 unsigned int i;
134 int most_recent = 0;
135 int chosen_drive;
136 int change = 0;
137 int inargv = 0;
138 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
139 unsigned int num_devs;
140 mddev_dev_t tmpdev;
141 struct mdinfo info;
142 char *avail;
143 int nextspare = 0;
144
145 if (get_linux_version() < 2004000)
146 old_linux = 1;
147
148 if (mdfd >= 0) {
149 vers = md_get_version(mdfd);
150 if (vers <= 0) {
151 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
152 return 1;
153 }
154 if (vers < 9000) {
155 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
156 " Upgrade your kernel or try --build\n");
157 return 1;
158 }
159
160 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
161 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
162 mddev);
163 return 1;
164 }
165 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
166 }
167 /*
168 * If any subdevs are listed, then any that don't
169 * match ident are discarded. Remainder must all match and
170 * become the array.
171 * If no subdevs, then we scan all devices in the config file, but
172 * there must be something in the identity
173 */
174
175 if (!devlist &&
176 ident->uuid_set == 0 &&
177 ident->super_minor < 0 &&
178 ident->devices == NULL) {
179 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
180 mddev ? mddev : "further assembly");
181 return 1;
182 }
183 if (devlist == NULL)
184 devlist = conf_get_devs(conffile);
185 else if (mdfd >= 0)
186 inargv = 1;
187
188 tmpdev = devlist; num_devs = 0;
189 while (tmpdev) {
190 if (tmpdev->used)
191 tmpdev->used = 2;
192 else
193 num_devs++;
194 tmpdev = tmpdev->next;
195 }
196 devices = malloc(num_devs * sizeof(*devices));
197
198 if (!st && ident->st) st = ident->st;
199
200 if (verbose>0)
201 fprintf(stderr, Name ": looking for devices for %s\n",
202 mddev ? mddev : "further assembly");
203
204 /* first walk the list of devices to find a consistent set
205 * that match the criterea, if that is possible.
206 * We flag the one we like with 'used'.
207 */
208 for (tmpdev = devlist;
209 tmpdev;
210 tmpdev = tmpdev->next) {
211 char *devname = tmpdev->devname;
212 int dfd;
213 struct stat stb;
214 struct supertype *tst = st;
215
216 if (tmpdev->used > 1) continue;
217
218 if (ident->devices &&
219 !match_oneof(ident->devices, devname)) {
220 if ((inargv && verbose>=0) || verbose > 0)
221 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
222 continue;
223 }
224
225 if (super) {
226 free(super);
227 super = NULL;
228 }
229
230 dfd = dev_open(devname, O_RDONLY|O_EXCL);
231 if (dfd < 0) {
232 if ((inargv && verbose >= 0) || verbose > 0)
233 fprintf(stderr, Name ": cannot open device %s: %s\n",
234 devname, strerror(errno));
235 tmpdev->used = 2;
236 } else if (fstat(dfd, &stb)< 0) {
237 /* Impossible! */
238 fprintf(stderr, Name ": fstat failed for %s: %s\n",
239 devname, strerror(errno));
240 tmpdev->used = 2;
241 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
242 fprintf(stderr, Name ": %s is not a block device.\n",
243 devname);
244 tmpdev->used = 2;
245 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
246 if ((inargv && verbose >= 0) || verbose > 0)
247 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
248 devname);
249 tmpdev->used = 2;
250 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
251 if ((inargv && verbose >= 0) || verbose > 0)
252 fprintf( stderr, Name ": no RAID superblock on %s\n",
253 devname);
254 } else {
255 tst->ss->getinfo_super(&info, super);
256 }
257 if (dfd >= 0) close(dfd);
258
259 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
260 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
261 if ((inargv && verbose >= 0) || verbose > 0)
262 fprintf(stderr, Name ": %s has wrong uuid.\n",
263 devname);
264 continue;
265 }
266 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
267 (!super || name_matches(info.name, ident->name, homehost)==0)) {
268 if ((inargv && verbose >= 0) || verbose > 0)
269 fprintf(stderr, Name ": %s has wrong name.\n",
270 devname);
271 continue;
272 }
273 if (ident->super_minor != UnSet &&
274 (!super || 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 continue;
279 }
280 if (ident->level != UnSet &&
281 (!super|| ident->level != info.array.level)) {
282 if ((inargv && verbose >= 0) || verbose > 0)
283 fprintf(stderr, Name ": %s has wrong raid level.\n",
284 devname);
285 continue;
286 }
287 if (ident->raid_disks != UnSet &&
288 (!super || ident->raid_disks!= info.array.raid_disks)) {
289 if ((inargv && verbose >= 0) || verbose > 0)
290 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
291 devname);
292 continue;
293 }
294 if (mdfd < 0) {
295 if (tst == NULL || super == NULL)
296 continue;
297 if (update == NULL &&
298 tst->ss->match_home(super, homehost)==0) {
299 if ((inargv && verbose >= 0) || verbose > 0)
300 fprintf(stderr, Name ": %s is not built for host %s.\n",
301 devname, homehost);
302 /* Auto-assemble, and this is not a usable host */
303 /* if update != NULL, we are updating the host
304 * name... */
305 continue;
306 }
307 }
308 /* If we are this far, then we are nearly commited to this device.
309 * If the super_block doesn't exist, or doesn't match others,
310 * then we probably cannot continue
311 * However if one of the arrays is for the homehost, and
312 * the other isn't that can disambiguate.
313 */
314
315 if (!super) {
316 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
317 devname);
318 free(first_super);
319 return 1;
320 }
321
322 if (st == NULL)
323 st = tst;
324 if (st->ss != tst->ss ||
325 st->minor_version != tst->minor_version ||
326 st->ss->compare_super(&first_super, super) != 0) {
327 /* Some mismatch. If exactly one array matches this host,
328 * we can resolve on that one.
329 * Or, if we are auto assembling, we just ignore the second
330 * for now.
331 */
332 if (mdfd < 0)
333 continue;
334 if (homehost) {
335 int first = st->ss->match_home(first_super, homehost);
336 int last = tst->ss->match_home(super, homehost);
337 if (first+last == 1) {
338 /* We can do something */
339 if (first) {/* just ignore this one */
340 if ((inargv && verbose >= 0) || verbose > 0)
341 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
342 devname);
343 continue;
344 } else { /* reject all those sofar */
345 mddev_dev_t td;
346 if ((inargv && verbose >= 0) || verbose > 0)
347 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
348 devname);
349 for (td=devlist; td != tmpdev; td=td->next)
350 if (td->used == 1)
351 td->used = 0;
352 tmpdev->used = 1;
353 continue;
354 }
355 }
356 }
357 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
358 devname);
359 free(super);
360 free(first_super);
361 return 1;
362 }
363
364 tmpdev->used = 1;
365 }
366
367 if (mdfd < 0) {
368 /* So... it is up to me to open the device.
369 * We create a name '/dev/md/XXX' based on the info in the
370 * superblock, and call open_mddev on that
371 */
372 mdu_array_info_t inf;
373 char *c;
374 if (!first_super) {
375 return 2;
376 }
377 st->ss->getinfo_super(&info, first_super);
378 c = strchr(info.name, ':');
379 if (c) c++; else c= info.name;
380 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
381 /* /dev/md/d0 style for partitionable */
382 asprintf(&mddev, "/dev/md/d%s", c);
383 else
384 asprintf(&mddev, "/dev/md/%s", c);
385 mdfd = open_mddev(mddev, ident->autof);
386 if (mdfd < 0)
387 return mdfd;
388 vers = md_get_version(mdfd);
389 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
390 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
391 close(mdfd);
392 free(first_super);
393 return 1;
394 }
395 must_close = 1;
396 }
397
398 /* Ok, no bad inconsistancy, we can try updating etc */
399 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
400 char *devname = tmpdev->devname;
401 struct stat stb;
402 /* looks like a good enough match to update the super block if needed */
403 #ifndef MDASSEMBLE
404 if (update) {
405 int dfd;
406 /* prepare useful information in info structures */
407 struct stat stb2;
408 fstat(mdfd, &stb2);
409
410 if (strcmp(update, "uuid")==0 &&
411 !ident->uuid_set) {
412 int rfd;
413 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
414 read(rfd, ident->uuid, 16) != 16) {
415 *(__u32*)(ident->uuid) = random();
416 *(__u32*)(ident->uuid+1) = random();
417 *(__u32*)(ident->uuid+2) = random();
418 *(__u32*)(ident->uuid+3) = random();
419 }
420 if (rfd >= 0) close(rfd);
421 }
422 dfd = dev_open(devname, O_RDWR|O_EXCL);
423
424 if (super) {
425 free(super);
426 super = NULL;
427 }
428
429 st->ss->load_super(st, dfd, &super, NULL);
430 st->ss->getinfo_super(&info, super);
431
432 memcpy(info.uuid, ident->uuid, 16);
433 strcpy(info.name, ident->name);
434 info.array.md_minor = minor(stb2.st_rdev);
435
436 st->ss->update_super(&info, super, update, devname, verbose,
437 ident->uuid_set, homehost);
438 if (strcmp(update, "uuid")==0 &&
439 !ident->uuid_set) {
440 ident->uuid_set = 1;
441 memcpy(ident->uuid, info.uuid, 16);
442 }
443 if (dfd < 0)
444 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
445 devname);
446 else if (st->ss->store_super(st, dfd, super))
447 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
448 devname);
449 if (dfd >= 0)
450 close(dfd);
451
452 if (strcmp(update, "uuid")==0 &&
453 ident->bitmap_fd)
454 if (bitmap_update_uuid(ident->bitmap_fd, info.uuid) != 0)
455 fprintf(stderr, Name ": Could not update uuid on %s.\n",
456 devname);
457 } else
458 #endif
459 {
460 int dfd;
461 dfd = dev_open(devname, O_RDWR|O_EXCL);
462
463 if (super) {
464 free(super);
465 super = NULL;
466 }
467
468 st->ss->load_super(st, dfd, &super, NULL);
469 st->ss->getinfo_super(&info, super);
470 close(dfd);
471 }
472
473 stat(devname, &stb);
474
475 if (verbose > 0)
476 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
477 devname, mddev, info.disk.raid_disk);
478 devices[devcnt].devname = devname;
479 devices[devcnt].major = major(stb.st_rdev);
480 devices[devcnt].minor = minor(stb.st_rdev);
481 devices[devcnt].oldmajor = info.disk.major;
482 devices[devcnt].oldminor = info.disk.minor;
483 devices[devcnt].events = info.events;
484 devices[devcnt].raid_disk = info.disk.raid_disk;
485 devices[devcnt].disk_nr = info.disk.number;
486 devices[devcnt].uptodate = 0;
487 devices[devcnt].state = info.disk.state;
488 if (most_recent < devcnt) {
489 if (devices[devcnt].events
490 > devices[most_recent].events)
491 most_recent = devcnt;
492 }
493 if (info.array.level == -4)
494 /* with multipath, the raid_disk from the superblock is meaningless */
495 i = devcnt;
496 else
497 i = devices[devcnt].raid_disk;
498 if (i+1 == 0) {
499 if (nextspare < info.array.raid_disks)
500 nextspare = info.array.raid_disks;
501 i = nextspare++;
502 }
503 if (i < 10000) {
504 if (i >= bestcnt) {
505 unsigned int newbestcnt = i+10;
506 int *newbest = malloc(sizeof(int)*newbestcnt);
507 unsigned int c;
508 for (c=0; c < newbestcnt; c++)
509 if (c < bestcnt)
510 newbest[c] = best[c];
511 else
512 newbest[c] = -1;
513 if (best)free(best);
514 best = newbest;
515 bestcnt = newbestcnt;
516 }
517 if (best[i] >=0 &&
518 devices[best[i]].events == devices[devcnt].events &&
519 devices[best[i]].minor != devices[devcnt].minor &&
520 st->ss->major == 0 &&
521 info.array.level != -4) {
522 /* two different devices with identical superblock.
523 * Could be a mis-detection caused by overlapping
524 * partitions. fail-safe.
525 */
526 fprintf(stderr, Name ": WARNING %s and %s appear"
527 " to have very similar superblocks.\n"
528 " If they are really different, "
529 "please --zero the superblock on one\n"
530 " If they are the same, please remove "
531 "one from the list.\n",
532 devices[best[i]].devname, devname);
533 if (must_close) close(mdfd);
534 return 1;
535 }
536 if (best[i] == -1
537 || devices[best[i]].events < devices[devcnt].events)
538 best[i] = devcnt;
539 }
540 devcnt++;
541 }
542
543 if (super)
544 free(super);
545 super = NULL;
546
547 if (update && strcmp(update, "byteorder")==0)
548 st->minor_version = 90;
549
550 if (devcnt == 0) {
551 fprintf(stderr, Name ": no devices found for %s\n",
552 mddev);
553 free(first_super);
554 if (must_close) close(mdfd);
555 return 1;
556 }
557
558 st->ss->getinfo_super(&info, first_super);
559
560 /* now we have some devices that might be suitable.
561 * I wonder how many
562 */
563 avail = malloc(info.array.raid_disks);
564 memset(avail, 0, info.array.raid_disks);
565 okcnt = 0;
566 sparecnt=0;
567 for (i=0; i< bestcnt ;i++) {
568 int j = best[i];
569 int event_margin = 1; /* always allow a difference of '1'
570 * like the kernel does
571 */
572 if (j < 0) continue;
573 /* note: we ignore error flags in multipath arrays
574 * as they don't make sense
575 */
576 if (info.array.level != -4)
577 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
578 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
579 sparecnt++;
580 continue;
581 }
582 if (devices[j].events+event_margin >=
583 devices[most_recent].events) {
584 devices[j].uptodate = 1;
585 if (i < info.array.raid_disks) {
586 okcnt++;
587 avail[i]=1;
588 } else
589 sparecnt++;
590 }
591 }
592 while (force && !enough(info.array.level, info.array.raid_disks,
593 info.array.layout,
594 avail, okcnt)) {
595 /* Choose the newest best drive which is
596 * not up-to-date, update the superblock
597 * and add it.
598 */
599 int fd;
600 chosen_drive = -1;
601 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
602 int j = best[i];
603 if (j>=0 &&
604 !devices[j].uptodate &&
605 devices[j].events > 0 &&
606 (chosen_drive < 0 ||
607 devices[j].events > devices[chosen_drive].events))
608 chosen_drive = j;
609 }
610 if (chosen_drive < 0)
611 break;
612 if (verbose >= 0)
613 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
614 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
615 (int)(devices[chosen_drive].events),
616 (int)(devices[most_recent].events));
617 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
618 if (fd < 0) {
619 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
620 devices[chosen_drive].devname);
621 devices[chosen_drive].events = 0;
622 continue;
623 }
624 if (st->ss->load_super(st,fd, &super, NULL)) {
625 close(fd);
626 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
627 devices[chosen_drive].devname);
628 devices[chosen_drive].events = 0;
629 continue;
630 }
631 info.events = devices[most_recent].events;
632 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
633
634 if (st->ss->store_super(st, fd, super)) {
635 close(fd);
636 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
637 devices[chosen_drive].devname);
638 devices[chosen_drive].events = 0;
639 free(super);
640 continue;
641 }
642 close(fd);
643 devices[chosen_drive].events = devices[most_recent].events;
644 devices[chosen_drive].uptodate = 1;
645 avail[chosen_drive] = 1;
646 okcnt++;
647 free(super);
648 }
649
650 /* Now we want to look at the superblock which the kernel will base things on
651 * and compare the devices that we think are working with the devices that the
652 * superblock thinks are working.
653 * If there are differences and --force is given, then update this chosen
654 * superblock.
655 */
656 chosen_drive = -1;
657 super = NULL;
658 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
659 int j = best[i];
660 int fd;
661
662 if (j<0)
663 continue;
664 if (!devices[j].uptodate)
665 continue;
666 chosen_drive = j;
667 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
668 fprintf(stderr, Name ": Cannot open %s: %s\n",
669 devices[j].devname, strerror(errno));
670 if (must_close) close(mdfd);
671 return 1;
672 }
673 if (st->ss->load_super(st,fd, &super, NULL)) {
674 close(fd);
675 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
676 devices[j].devname);
677 if (must_close) close(mdfd);
678 return 1;
679 }
680 close(fd);
681 }
682 if (super == NULL) {
683 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
684 if (must_close) close(mdfd);
685 return 1;
686 }
687 st->ss->getinfo_super(&info, super);
688 for (i=0; i<bestcnt; i++) {
689 int j = best[i];
690 unsigned int desired_state;
691
692 if (i < info.array.raid_disks)
693 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
694 else
695 desired_state = 0;
696
697 if (j<0)
698 continue;
699 if (!devices[j].uptodate)
700 continue;
701 info.disk.number = devices[j].disk_nr;
702 info.disk.raid_disk = i;
703 info.disk.state = desired_state;
704
705 if (devices[j].uptodate &&
706 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
707 if (force) {
708 if (verbose >= 0)
709 fprintf(stderr, Name ": "
710 "clearing FAULTY flag for device %d in %s for %s\n",
711 j, mddev, devices[j].devname);
712 change = 1;
713 } else {
714 if (verbose >= -1)
715 fprintf(stderr, Name ": "
716 "device %d in %s has wrong state in superblock, but %s seems ok\n",
717 i, mddev, devices[j].devname);
718 }
719 }
720 #if 0
721 if (!devices[j].uptodate &&
722 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
723 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
724 i, mddev);
725 }
726 #endif
727 }
728 if (force && okcnt == info.array.raid_disks-1) {
729 /* FIXME check event count */
730 change += st->ss->update_super(&info, super, "force",
731 devices[chosen_drive].devname, verbose, 0, NULL);
732 }
733
734 if (change) {
735 int fd;
736 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
737 if (fd < 0) {
738 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
739 devices[chosen_drive].devname);
740 if (must_close) close(mdfd);
741 return 1;
742 }
743 if (st->ss->store_super(st, fd, super)) {
744 close(fd);
745 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
746 devices[chosen_drive].devname);
747 if (must_close) close(mdfd);
748 return 1;
749 }
750 close(fd);
751 }
752
753 /* If we are in the middle of a reshape we may need to restore saved data
754 * that was moved aside due to the reshape overwriting live data
755 * The code of doing this lives in Grow.c
756 */
757 #ifndef MDASSEMBLE
758 if (info.reshape_active) {
759 int err = 0;
760 int *fdlist = malloc(sizeof(int)* bestcnt);
761 for (i=0; i<bestcnt; i++) {
762 int j = best[i];
763 if (j >= 0) {
764 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
765 if (fdlist[i] < 0) {
766 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
767 devices[j].devname);
768 err = 1;
769 break;
770 }
771 } else
772 fdlist[i] = -1;
773 }
774 if (!err)
775 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
776 while (i>0) {
777 i--;
778 if (fdlist[i]>=0) close(fdlist[i]);
779 }
780 if (err) {
781 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
782 if (must_close) close(mdfd);
783 return err;
784 }
785 }
786 #endif
787 /* count number of in-sync devices according to the superblock.
788 * We must have this number to start the array without -s or -R
789 */
790 req_cnt = info.array.working_disks;
791
792 /* Almost ready to actually *do* something */
793 if (!old_linux) {
794 int rv;
795 if ((vers % 100) >= 1) { /* can use different versions */
796 mdu_array_info_t inf;
797 memset(&inf, 0, sizeof(inf));
798 inf.major_version = st->ss->major;
799 inf.minor_version = st->minor_version;
800 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
801 } else
802 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
803
804 if (rv) {
805 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
806 mddev, strerror(errno));
807 if (must_close) close(mdfd);
808 return 1;
809 }
810 if (ident->bitmap_fd >= 0) {
811 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
812 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
813 if (must_close) close(mdfd);
814 return 1;
815 }
816 } else if (ident->bitmap_file) {
817 /* From config file */
818 int bmfd = open(ident->bitmap_file, O_RDWR);
819 if (bmfd < 0) {
820 fprintf(stderr, Name ": Could not open bitmap file %s\n",
821 ident->bitmap_file);
822 if (must_close) close(mdfd);
823 return 1;
824 }
825 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
826 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
827 close(bmfd);
828 if (must_close) close(mdfd);
829 return 1;
830 }
831 close(bmfd);
832 }
833
834 /* First, add the raid disks, but add the chosen one last */
835 for (i=0; i<= bestcnt; i++) {
836 int j;
837 if (i < bestcnt) {
838 j = best[i];
839 if (j == chosen_drive)
840 continue;
841 } else
842 j = chosen_drive;
843
844 if (j >= 0 /* && devices[j].uptodate */) {
845 mdu_disk_info_t disk;
846 memset(&disk, 0, sizeof(disk));
847 disk.major = devices[j].major;
848 disk.minor = devices[j].minor;
849 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
850 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
851 devices[j].devname,
852 mddev,
853 strerror(errno));
854 if (i < info.array.raid_disks || i == bestcnt)
855 okcnt--;
856 else
857 sparecnt--;
858 } else if (verbose > 0)
859 fprintf(stderr, Name ": added %s to %s as %d\n",
860 devices[j].devname, mddev, devices[j].raid_disk);
861 } else if (verbose > 0 && i < info.array.raid_disks)
862 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
863 i, mddev);
864 }
865
866 if (runstop == 1 ||
867 (runstop <= 0 &&
868 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
869 (okcnt >= req_cnt || start_partial_ok)
870 ))) {
871 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
872 if (verbose >= 0) {
873 fprintf(stderr, Name ": %s has been started with %d drive%s",
874 mddev, okcnt, okcnt==1?"":"s");
875 if (okcnt < info.array.raid_disks)
876 fprintf(stderr, " (out of %d)", info.array.raid_disks);
877 if (sparecnt)
878 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
879 fprintf(stderr, ".\n");
880 }
881 if (must_close) {
882 int usecs = 1;
883 close(mdfd);
884 /* There is a nasty race with 'mdadm --monitor'.
885 * If it opens this device before we close it,
886 * it gets an incomplete open on which IO
887 * doesn't work and the capacity if wrong.
888 * If we reopen (to check for layered devices)
889 * before --monitor closes, we loose.
890 *
891 * So: wait upto 1 second for there to be
892 * a non-zero capacity.
893 */
894 while (usecs < 1000) {
895 mdfd = open(mddev, O_RDONLY);
896 if (mdfd >= 0) {
897 unsigned long size;
898 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
899 size > 0)
900 break;
901 close(mdfd);
902 }
903 usleep(usecs);
904 usecs <<= 1;
905 }
906 }
907 return 0;
908 }
909 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
910 mddev, strerror(errno));
911 if (must_close) close(mdfd);
912 return 1;
913 }
914 if (runstop == -1) {
915 fprintf(stderr, Name ": %s assembled from %d drive%s",
916 mddev, okcnt, okcnt==1?"":"s");
917 if (okcnt != info.array.raid_disks)
918 fprintf(stderr, " (out of %d)", info.array.raid_disks);
919 fprintf(stderr, ", but not started.\n");
920 if (must_close) close(mdfd);
921 return 0;
922 }
923 if (verbose >= 0) {
924 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
925 if (sparecnt)
926 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
927 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
928 fprintf(stderr, " - not enough to start the array.\n");
929 else {
930 if (req_cnt == info.array.raid_disks)
931 fprintf(stderr, " - need all %d to start it", req_cnt);
932 else
933 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
934 fprintf(stderr, " (use --run to insist).\n");
935 }
936 }
937 if (must_close) close(mdfd);
938 return 1;
939 } else {
940 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
941 * been updated to point to the current locations of devices.
942 * so we can just start the array
943 */
944 unsigned long dev;
945 dev = makedev(devices[chosen_drive].major,
946 devices[chosen_drive].minor);
947 if (ioctl(mdfd, START_ARRAY, dev)) {
948 fprintf(stderr, Name ": Cannot start array: %s\n",
949 strerror(errno));
950 }
951
952 }
953 if (must_close) close(mdfd);
954 return 0;
955 }