]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
assemble: fix setting of 'foreign' trustability.
[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,
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 mdfd;
115 int clean;
116 int auto_assem = (mddev == NULL);
117 int old_linux = 0;
118 int vers = vers; /* Keep gcc quite - it really is initialised */
119 struct {
120 char *devname;
121 int uptodate; /* set once we decide that this device is as
122 * recent as everything else in the array.
123 */
124 struct mdinfo i;
125 } *devices;
126 int *best = NULL; /* indexed by raid_disk */
127 unsigned int bestcnt = 0;
128 int devcnt = 0;
129 unsigned int okcnt, sparecnt;
130 unsigned int req_cnt;
131 unsigned int i;
132 int most_recent = 0;
133 int chosen_drive;
134 int change = 0;
135 int inargv = 0;
136 int bitmap_done;
137 int start_partial_ok = (runstop >= 0) &&
138 (force || devlist==NULL || auto_assem);
139 unsigned int num_devs;
140 mddev_dev_t tmpdev;
141 struct mdinfo info;
142 mdu_array_info_t tmp_inf;
143 char *avail;
144 int nextspare = 0;
145 char *name;
146 int trustworthy;
147 char chosen_name[1024];
148
149 memset(&info, 0, sizeof(info));
150
151 if (get_linux_version() < 2004000)
152 old_linux = 1;
153
154 /*
155 * If any subdevs are listed, then any that don't
156 * match ident are discarded. Remainder must all match and
157 * become the array.
158 * If no subdevs, then we scan all devices in the config file, but
159 * there must be something in the identity
160 */
161
162 if (!devlist &&
163 ident->uuid_set == 0 &&
164 ident->super_minor < 0 &&
165 ident->devices == NULL) {
166 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
167 mddev ? mddev : "further assembly");
168 return 1;
169 }
170
171 /* if the configuration specifies a container then we use that to
172 * determine the devices and retrieve the array configuration
173 */
174 #ifndef MDASSEMBLE
175 if (ident->container && ident->member) {
176 int cfd = open(ident->container, O_RDWR);
177 struct mdinfo *mdi;
178 struct supertype container;
179
180 if (verbose>0)
181 fprintf(stderr, Name ": looking to assemble member array %s"
182 " inside container %s\n", ident->member, ident->container);
183 if (cfd < 0) {
184 if (verbose>0)
185 fprintf(stderr, Name ": unable to open container %s: %s\n",
186 ident->container, strerror(errno));
187 return 1;
188 }
189
190 mdi = sysfs_read(cfd, fd2devnum(cfd), GET_VERSION);
191 if (!mdi) {
192 close(cfd);
193 if (verbose>0)
194 fprintf(stderr, Name ": unable to read container %s\n",
195 ident->container);
196 return 1;
197 }
198 container.ss = find_metadata_methods(mdi->text_version);
199 sysfs_free(mdi);
200 if (!container.ss) {
201 close(cfd);
202 fprintf(stderr, Name ": %s uses unknown metadata: %s\n",
203 ident->container, mdi->text_version);
204 return 1;
205 }
206 if (container.ss->load_super(&container, cfd, ident->container)) {
207 fprintf(stderr, Name ": Cannot load metadata for container %s\n",
208 ident->container);
209 return 1;
210 }
211
212 return Incremental_container(&container, ident->container,
213 verbose, runstop, ident->autof,
214 LOCAL);
215 }
216 #endif
217 if (devlist == NULL)
218 devlist = conf_get_devs();
219 else if (mddev)
220 inargv = 1;
221
222 try_again:
223 /* We come back here when doing auto-assembly and attempting some
224 * set of devices failed. Those are now marked as ->used==2 and
225 * we ignore them and try again
226 */
227
228 tmpdev = devlist; num_devs = 0;
229 while (tmpdev) {
230 if (tmpdev->used)
231 tmpdev->used = 2;
232 else
233 num_devs++;
234 tmpdev = tmpdev->next;
235 }
236 devices = malloc(num_devs * sizeof(*devices));
237
238 if (!st && ident->st) st = ident->st;
239
240 if (verbose>0)
241 fprintf(stderr, Name ": looking for devices for %s\n",
242 mddev ? mddev : "further assembly");
243
244 /* first walk the list of devices to find a consistent set
245 * that match the criterea, if that is possible.
246 * We flag the ones we like with 'used'.
247 */
248 for (tmpdev = devlist;
249 tmpdev;
250 tmpdev = tmpdev->next) {
251 char *devname = tmpdev->devname;
252 int dfd;
253 struct stat stb;
254 struct supertype *tst = dup_super(st);
255
256 if (tmpdev->used > 1) continue;
257
258 if (ident->devices &&
259 !match_oneof(ident->devices, devname)) {
260 if ((inargv && verbose>=0) || verbose > 0)
261 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
262 continue;
263 }
264
265 dfd = dev_open(devname, O_RDONLY|O_EXCL);
266 if (dfd < 0) {
267 if ((inargv && verbose >= 0) || verbose > 0)
268 fprintf(stderr, Name ": cannot open device %s: %s\n",
269 devname, strerror(errno));
270 tmpdev->used = 2;
271 } else if (fstat(dfd, &stb)< 0) {
272 /* Impossible! */
273 fprintf(stderr, Name ": fstat failed for %s: %s\n",
274 devname, strerror(errno));
275 tmpdev->used = 2;
276 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
277 fprintf(stderr, Name ": %s is not a block device.\n",
278 devname);
279 tmpdev->used = 2;
280 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
281 if ((inargv && verbose >= 0) || verbose > 0)
282 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
283 devname);
284 tmpdev->used = 2;
285 } else if (tst->ss->load_super(tst,dfd, NULL)) {
286 if ((inargv && verbose >= 0) || verbose > 0)
287 fprintf( stderr, Name ": no RAID superblock on %s\n",
288 devname);
289 } else {
290 tst->ss->getinfo_super(tst, &info);
291 }
292 if (dfd >= 0) close(dfd);
293
294 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
295 (!tst || !tst->sb ||
296 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
297 if ((inargv && verbose >= 0) || verbose > 0)
298 fprintf(stderr, Name ": %s has wrong uuid.\n",
299 devname);
300 goto loop;
301 }
302 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
303 (!tst || !tst->sb ||
304 name_matches(info.name, ident->name, homehost)==0)) {
305 if ((inargv && verbose >= 0) || verbose > 0)
306 fprintf(stderr, Name ": %s has wrong name.\n",
307 devname);
308 goto loop;
309 }
310 if (ident->super_minor != UnSet &&
311 (!tst || !tst->sb ||
312 ident->super_minor != info.array.md_minor)) {
313 if ((inargv && verbose >= 0) || verbose > 0)
314 fprintf(stderr, Name ": %s has wrong super-minor.\n",
315 devname);
316 goto loop;
317 }
318 if (ident->level != UnSet &&
319 (!tst || !tst->sb ||
320 ident->level != info.array.level)) {
321 if ((inargv && verbose >= 0) || verbose > 0)
322 fprintf(stderr, Name ": %s has wrong raid level.\n",
323 devname);
324 goto loop;
325 }
326 if (ident->raid_disks != UnSet &&
327 (!tst || !tst->sb ||
328 ident->raid_disks!= info.array.raid_disks)) {
329 if ((inargv && verbose >= 0) || verbose > 0)
330 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
331 devname);
332 goto loop;
333 }
334 if (auto_assem) {
335 if (tst == NULL || tst->sb == NULL)
336 continue;
337 }
338 /* If we are this far, then we are nearly commited to this device.
339 * If the super_block doesn't exist, or doesn't match others,
340 * then we probably cannot continue
341 * However if one of the arrays is for the homehost, and
342 * the other isn't that can disambiguate.
343 */
344
345 if (!tst || !tst->sb) {
346 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
347 devname);
348 if (st)
349 st->ss->free_super(st);
350 return 1;
351 }
352
353 if (st == NULL)
354 st = dup_super(tst);
355 if (st->minor_version == -1)
356 st->minor_version = tst->minor_version;
357 if (st->ss != tst->ss ||
358 st->minor_version != tst->minor_version ||
359 st->ss->compare_super(st, tst) != 0) {
360 /* Some mismatch. If exactly one array matches this host,
361 * we can resolve on that one.
362 * Or, if we are auto assembling, we just ignore the second
363 * for now.
364 */
365 if (auto_assem)
366 goto loop;
367 if (homehost) {
368 int first = st->ss->match_home(st, homehost);
369 int last = tst->ss->match_home(tst, homehost);
370 if (first != last &&
371 (first == 1 || last == 1)) {
372 /* We can do something */
373 if (first) {/* just ignore this one */
374 if ((inargv && verbose >= 0) || verbose > 0)
375 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
376 devname);
377 goto loop;
378 } else { /* reject all those sofar */
379 mddev_dev_t td;
380 if ((inargv && verbose >= 0) || verbose > 0)
381 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
382 devname);
383 for (td=devlist; td != tmpdev; td=td->next)
384 if (td->used == 1)
385 td->used = 0;
386 tmpdev->used = 1;
387 goto loop;
388 }
389 }
390 }
391 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
392 devname);
393 tst->ss->free_super(tst);
394 st->ss->free_super(st);
395 return 1;
396 }
397
398 tmpdev->used = 1;
399
400 loop:
401 if (tst)
402 tst->ss->free_super(tst);
403 }
404
405 if (!st || !st->sb)
406 return 2;
407
408 /* Now need to open array the device. Use create_mddev */
409 st->ss->getinfo_super(st, &info);
410
411 trustworthy = FOREIGN;
412 switch (st->ss->match_home(st, homehost)) {
413 case 0:
414 trustworthy = FOREIGN;
415 name = info.name;
416 break;
417 case 1:
418 trustworthy = LOCAL;
419 name = strchr(info.name, ':');
420 if (name)
421 name++;
422 else
423 name = info.name;
424 break;
425 case -1:
426 trustworthy = FOREIGN;
427 break;
428 }
429 if (!auto_assem && trustworthy == FOREIGN)
430 /* If the array is listed in mdadm or on
431 * command line, then we trust the name
432 * even if the array doesn't look local
433 */
434 trustworthy = LOCAL;
435
436 if (info.name[0] == 0 &&
437 info.array.level == LEVEL_CONTAINER) {
438 name = info.text_version;
439 trustworthy = METADATA;
440 }
441 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
442 chosen_name);
443 if (mdfd < 0) {
444 st->ss->free_super(st);
445 free(devices);
446 if (auto_assem)
447 goto try_again;
448 return 1;
449 }
450 mddev = chosen_name;
451 vers = md_get_version(mdfd);
452 if (vers < 9000) {
453 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
454 " Upgrade your kernel or try --build\n");
455 close(mdfd);
456 return 1;
457 }
458 if (ioctl(mdfd, GET_ARRAY_INFO, &tmp_inf)==0) {
459 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
460 mddev);
461 for (tmpdev = devlist ;
462 tmpdev && tmpdev->used != 1;
463 tmpdev = tmpdev->next)
464 ;
465 if (tmpdev && auto_assem)
466 fprintf(stderr, Name ": %s needed for %s...\n",
467 mddev, tmpdev->devname);
468 close(mdfd);
469 mdfd = -3;
470 st->ss->free_super(st);
471 free(devices);
472 if (auto_assem)
473 goto try_again;
474 return 1;
475 }
476 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
477
478 /* Ok, no bad inconsistancy, we can try updating etc */
479 bitmap_done = 0;
480 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
481 char *devname = tmpdev->devname;
482 struct stat stb;
483 /* looks like a good enough match to update the super block if needed */
484 #ifndef MDASSEMBLE
485 if (update) {
486 int dfd;
487 /* prepare useful information in info structures */
488 struct stat stb2;
489 struct supertype *tst;
490 fstat(mdfd, &stb2);
491
492 if (strcmp(update, "uuid")==0 &&
493 !ident->uuid_set) {
494 int rfd;
495 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
496 read(rfd, ident->uuid, 16) != 16) {
497 *(__u32*)(ident->uuid) = random();
498 *(__u32*)(ident->uuid+1) = random();
499 *(__u32*)(ident->uuid+2) = random();
500 *(__u32*)(ident->uuid+3) = random();
501 }
502 if (rfd >= 0) close(rfd);
503 }
504 dfd = dev_open(devname, O_RDWR|O_EXCL);
505
506 remove_partitions(dfd);
507
508 tst = dup_super(st);
509 tst->ss->load_super(tst, dfd, NULL);
510 tst->ss->getinfo_super(tst, &info);
511
512 memcpy(info.uuid, ident->uuid, 16);
513 strcpy(info.name, ident->name);
514 info.array.md_minor = minor(stb2.st_rdev);
515
516 tst->ss->update_super(tst, &info, update,
517 devname, verbose,
518 ident->uuid_set, homehost);
519 if (strcmp(update, "uuid")==0 &&
520 !ident->uuid_set) {
521 ident->uuid_set = 1;
522 memcpy(ident->uuid, info.uuid, 16);
523 }
524 if (dfd < 0)
525 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
526 devname);
527 else if (tst->ss->store_super(tst, dfd))
528 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
529 devname);
530 if (dfd >= 0)
531 close(dfd);
532
533 if (strcmp(update, "uuid")==0 &&
534 ident->bitmap_fd >= 0 && !bitmap_done) {
535 if (bitmap_update_uuid(ident->bitmap_fd,
536 info.uuid,
537 tst->ss->swapuuid) != 0)
538 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
539 else
540 bitmap_done = 1;
541 }
542 tst->ss->free_super(tst);
543 } else
544 #endif
545 {
546 struct supertype *tst = dup_super(st);
547 int dfd;
548 dfd = dev_open(devname, O_RDWR|O_EXCL);
549
550 remove_partitions(dfd);
551
552 tst->ss->load_super(tst, dfd, NULL);
553 tst->ss->getinfo_super(tst, &info);
554 tst->ss->free_super(tst);
555 close(dfd);
556 }
557
558 stat(devname, &stb);
559
560 if (verbose > 0)
561 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
562 devname, mddev, info.disk.raid_disk);
563 devices[devcnt].devname = devname;
564 devices[devcnt].uptodate = 0;
565 devices[devcnt].i = info;
566 devices[devcnt].i.disk.major = major(stb.st_rdev);
567 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
568 if (most_recent < devcnt) {
569 if (devices[devcnt].i.events
570 > devices[most_recent].i.events)
571 most_recent = devcnt;
572 }
573 if (info.array.level == -4)
574 /* with multipath, the raid_disk from the superblock is meaningless */
575 i = devcnt;
576 else
577 i = devices[devcnt].i.disk.raid_disk;
578 if (i+1 == 0) {
579 if (nextspare < info.array.raid_disks)
580 nextspare = info.array.raid_disks;
581 i = nextspare++;
582 } else {
583 if (i >= info.array.raid_disks &&
584 i >= nextspare)
585 nextspare = i+1;
586 }
587 if (i < 10000) {
588 if (i >= bestcnt) {
589 unsigned int newbestcnt = i+10;
590 int *newbest = malloc(sizeof(int)*newbestcnt);
591 unsigned int c;
592 for (c=0; c < newbestcnt; c++)
593 if (c < bestcnt)
594 newbest[c] = best[c];
595 else
596 newbest[c] = -1;
597 if (best)free(best);
598 best = newbest;
599 bestcnt = newbestcnt;
600 }
601 if (best[i] >=0 &&
602 devices[best[i]].i.events
603 == devices[devcnt].i.events
604 && (devices[best[i]].i.disk.minor
605 != devices[devcnt].i.disk.minor)
606 && st->ss == &super0
607 && info.array.level != LEVEL_MULTIPATH) {
608 /* two different devices with identical superblock.
609 * Could be a mis-detection caused by overlapping
610 * partitions. fail-safe.
611 */
612 fprintf(stderr, Name ": WARNING %s and %s appear"
613 " to have very similar superblocks.\n"
614 " If they are really different, "
615 "please --zero the superblock on one\n"
616 " If they are the same or overlap,"
617 " please remove one from %s.\n",
618 devices[best[i]].devname, devname,
619 inargv ? "the list" :
620 "the\n DEVICE list in mdadm.conf"
621 );
622 close(mdfd);
623 return 1;
624 }
625 if (best[i] == -1
626 || (devices[best[i]].i.events
627 < devices[devcnt].i.events))
628 best[i] = devcnt;
629 }
630 devcnt++;
631 }
632
633 if (devcnt == 0) {
634 fprintf(stderr, Name ": no devices found for %s\n",
635 mddev);
636 if (st)
637 st->ss->free_super(st);
638 close(mdfd);
639 return 1;
640 }
641
642 if (update && strcmp(update, "byteorder")==0)
643 st->minor_version = 90;
644
645 st->ss->getinfo_super(st, &info);
646 clean = info.array.state & 1;
647
648 /* now we have some devices that might be suitable.
649 * I wonder how many
650 */
651 avail = malloc(info.array.raid_disks);
652 memset(avail, 0, info.array.raid_disks);
653 okcnt = 0;
654 sparecnt=0;
655 for (i=0; i< bestcnt ;i++) {
656 int j = best[i];
657 int event_margin = 1; /* always allow a difference of '1'
658 * like the kernel does
659 */
660 if (j < 0) continue;
661 /* note: we ignore error flags in multipath arrays
662 * as they don't make sense
663 */
664 if (info.array.level != -4)
665 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
666 if (!(devices[j].i.disk.state
667 & (1<<MD_DISK_FAULTY)))
668 sparecnt++;
669 continue;
670 }
671 if (devices[j].i.events+event_margin >=
672 devices[most_recent].i.events) {
673 devices[j].uptodate = 1;
674 if (i < info.array.raid_disks) {
675 okcnt++;
676 avail[i]=1;
677 } else
678 sparecnt++;
679 }
680 }
681 while (force && !enough(info.array.level, info.array.raid_disks,
682 info.array.layout, 1,
683 avail, okcnt)) {
684 /* Choose the newest best drive which is
685 * not up-to-date, update the superblock
686 * and add it.
687 */
688 int fd;
689 struct supertype *tst;
690 long long current_events;
691 chosen_drive = -1;
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 (chosen_drive < 0 ||
698 devices[j].i.events
699 > devices[chosen_drive].i.events))
700 chosen_drive = j;
701 }
702 if (chosen_drive < 0)
703 break;
704 current_events = devices[chosen_drive].i.events;
705 add_another:
706 if (verbose >= 0)
707 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
708 devices[chosen_drive].devname,
709 devices[chosen_drive].i.disk.raid_disk,
710 (int)(devices[chosen_drive].i.events),
711 (int)(devices[most_recent].i.events));
712 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
713 if (fd < 0) {
714 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
715 devices[chosen_drive].devname);
716 devices[chosen_drive].i.events = 0;
717 continue;
718 }
719 tst = dup_super(st);
720 if (tst->ss->load_super(tst,fd, NULL)) {
721 close(fd);
722 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
723 devices[chosen_drive].devname);
724 devices[chosen_drive].i.events = 0;
725 continue;
726 }
727 info.events = devices[most_recent].i.events;
728 tst->ss->update_super(tst, &info, "force-one",
729 devices[chosen_drive].devname, verbose,
730 0, NULL);
731
732 if (tst->ss->store_super(tst, fd)) {
733 close(fd);
734 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
735 devices[chosen_drive].devname);
736 devices[chosen_drive].i.events = 0;
737 tst->ss->free_super(tst);
738 continue;
739 }
740 close(fd);
741 devices[chosen_drive].i.events = devices[most_recent].i.events;
742 devices[chosen_drive].uptodate = 1;
743 avail[chosen_drive] = 1;
744 okcnt++;
745 tst->ss->free_super(tst);
746
747 /* If there are any other drives of the same vintage,
748 * add them in as well. We can't lose and we might gain
749 */
750 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
751 int j = best[i];
752 if (j >= 0 &&
753 !devices[j].uptodate &&
754 devices[j].i.events > 0 &&
755 devices[j].i.events == current_events) {
756 chosen_drive = j;
757 goto add_another;
758 }
759 }
760 }
761
762 /* Now we want to look at the superblock which the kernel will base things on
763 * and compare the devices that we think are working with the devices that the
764 * superblock thinks are working.
765 * If there are differences and --force is given, then update this chosen
766 * superblock.
767 */
768 chosen_drive = -1;
769 st->ss->free_super(st);
770 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
771 int j = best[i];
772 int fd;
773
774 if (j<0)
775 continue;
776 if (!devices[j].uptodate)
777 continue;
778 chosen_drive = j;
779 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
780 fprintf(stderr, Name ": Cannot open %s: %s\n",
781 devices[j].devname, strerror(errno));
782 close(mdfd);
783 return 1;
784 }
785 if (st->ss->load_super(st,fd, NULL)) {
786 close(fd);
787 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
788 devices[j].devname);
789 close(mdfd);
790 return 1;
791 }
792 close(fd);
793 }
794 if (st->sb == NULL) {
795 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
796 close(mdfd);
797 return 1;
798 }
799 st->ss->getinfo_super(st, &info);
800 #ifndef MDASSEMBLE
801 sysfs_init(&info, mdfd, 0);
802 #endif
803 for (i=0; i<bestcnt; i++) {
804 int j = best[i];
805 unsigned int desired_state;
806
807 if (i < info.array.raid_disks)
808 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
809 else
810 desired_state = 0;
811
812 if (j<0)
813 continue;
814 if (!devices[j].uptodate)
815 continue;
816
817 devices[j].i.disk.state = desired_state;
818
819 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
820 verbose, 0, NULL)) {
821 if (force) {
822 if (verbose >= 0)
823 fprintf(stderr, Name ": "
824 "clearing FAULTY flag for device %d in %s for %s\n",
825 j, mddev, devices[j].devname);
826 change = 1;
827 } else {
828 if (verbose >= -1)
829 fprintf(stderr, Name ": "
830 "device %d in %s has wrong state in superblock, but %s seems ok\n",
831 i, mddev, devices[j].devname);
832 }
833 }
834 #if 0
835 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
836 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
837 i, mddev);
838 }
839 #endif
840 }
841 if (force && !clean &&
842 !enough(info.array.level, info.array.raid_disks,
843 info.array.layout, clean,
844 avail, okcnt)) {
845 change += st->ss->update_super(st, &info, "force-array",
846 devices[chosen_drive].devname, verbose,
847 0, NULL);
848 clean = 1;
849 }
850
851 if (change) {
852 int fd;
853 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
854 if (fd < 0) {
855 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
856 devices[chosen_drive].devname);
857 close(mdfd);
858 return 1;
859 }
860 if (st->ss->store_super(st, fd)) {
861 close(fd);
862 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
863 devices[chosen_drive].devname);
864 close(mdfd);
865 return 1;
866 }
867 close(fd);
868 }
869
870 /* If we are in the middle of a reshape we may need to restore saved data
871 * that was moved aside due to the reshape overwriting live data
872 * The code of doing this lives in Grow.c
873 */
874 #ifndef MDASSEMBLE
875 if (info.reshape_active) {
876 int err = 0;
877 int *fdlist = malloc(sizeof(int)* bestcnt);
878 for (i=0; i<bestcnt; i++) {
879 int j = best[i];
880 if (j >= 0) {
881 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
882 if (fdlist[i] < 0) {
883 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
884 devices[j].devname);
885 err = 1;
886 break;
887 }
888 } else
889 fdlist[i] = -1;
890 }
891 if (!err)
892 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
893 while (i>0) {
894 i--;
895 if (fdlist[i]>=0) close(fdlist[i]);
896 }
897 if (err) {
898 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
899 close(mdfd);
900 return err;
901 }
902 }
903 #endif
904 /* count number of in-sync devices according to the superblock.
905 * We must have this number to start the array without -s or -R
906 */
907 req_cnt = info.array.working_disks;
908
909 /* Almost ready to actually *do* something */
910 if (!old_linux) {
911 int rv;
912
913 /* First, fill in the map, so that udev can find our name
914 * as soon as we become active.
915 */
916 map_update(NULL, fd2devnum(mdfd), info.text_version,
917 info.uuid, chosen_name);
918
919 rv = set_array_info(mdfd, st, &info);
920 if (rv) {
921 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
922 mddev, strerror(errno));
923 close(mdfd);
924 return 1;
925 }
926 if (ident->bitmap_fd >= 0) {
927 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
928 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
929 close(mdfd);
930 return 1;
931 }
932 } else if (ident->bitmap_file) {
933 /* From config file */
934 int bmfd = open(ident->bitmap_file, O_RDWR);
935 if (bmfd < 0) {
936 fprintf(stderr, Name ": Could not open bitmap file %s\n",
937 ident->bitmap_file);
938 close(mdfd);
939 return 1;
940 }
941 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
942 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
943 close(bmfd);
944 close(mdfd);
945 return 1;
946 }
947 close(bmfd);
948 }
949
950 /* First, add the raid disks, but add the chosen one last */
951 for (i=0; i<= bestcnt; i++) {
952 int j;
953 if (i < bestcnt) {
954 j = best[i];
955 if (j == chosen_drive)
956 continue;
957 } else
958 j = chosen_drive;
959
960 if (j >= 0 /* && devices[j].uptodate */) {
961 rv = add_disk(mdfd, st, &info, &devices[j].i);
962
963 if (rv) {
964 fprintf(stderr, Name ": failed to add "
965 "%s to %s: %s\n",
966 devices[j].devname,
967 mddev,
968 strerror(errno));
969 if (i < info.array.raid_disks
970 || i == bestcnt)
971 okcnt--;
972 else
973 sparecnt--;
974 } else if (verbose > 0)
975 fprintf(stderr, Name ": added %s "
976 "to %s as %d\n",
977 devices[j].devname, mddev,
978 devices[j].i.disk.raid_disk);
979 } else if (verbose > 0 && i < info.array.raid_disks)
980 fprintf(stderr, Name ": no uptodate device for "
981 "slot %d of %s\n",
982 i, mddev);
983 }
984
985 if (info.array.level == LEVEL_CONTAINER) {
986 if (verbose >= 0) {
987 fprintf(stderr, Name ": Container %s has been "
988 "assembled with %d drive%s",
989 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
990 if (okcnt < info.array.raid_disks)
991 fprintf(stderr, " (out of %d)",
992 info.array.raid_disks);
993 fprintf(stderr, "\n");
994 }
995 sysfs_uevent(&info, "change");
996 close(mdfd);
997 return 0;
998 }
999
1000 if (runstop == 1 ||
1001 (runstop <= 0 &&
1002 ( enough(info.array.level, info.array.raid_disks,
1003 info.array.layout, clean, avail, okcnt) &&
1004 (okcnt >= req_cnt || start_partial_ok)
1005 ))) {
1006 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1007 if (verbose >= 0) {
1008 fprintf(stderr, Name ": %s has been started with %d drive%s",
1009 mddev, okcnt, okcnt==1?"":"s");
1010 if (okcnt < info.array.raid_disks)
1011 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1012 if (sparecnt)
1013 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1014 fprintf(stderr, ".\n");
1015 }
1016 close(mdfd);
1017 if (auto_assem) {
1018 int usecs = 1;
1019 /* There is a nasty race with 'mdadm --monitor'.
1020 * If it opens this device before we close it,
1021 * it gets an incomplete open on which IO
1022 * doesn't work and the capacity is
1023 * wrong.
1024 * If we reopen (to check for layered devices)
1025 * before --monitor closes, we loose.
1026 *
1027 * So: wait upto 1 second for there to be
1028 * a non-zero capacity.
1029 */
1030 while (usecs < 1000) {
1031 mdfd = open(mddev, O_RDONLY);
1032 if (mdfd >= 0) {
1033 unsigned long long size;
1034 if (get_dev_size(mdfd, NULL, &size) &&
1035 size > 0)
1036 break;
1037 close(mdfd);
1038 }
1039 usleep(usecs);
1040 usecs <<= 1;
1041 }
1042 }
1043 return 0;
1044 }
1045 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1046 mddev, strerror(errno));
1047
1048 if (!enough(info.array.level, info.array.raid_disks,
1049 info.array.layout, 1, avail, okcnt))
1050 fprintf(stderr, Name ": Not enough devices to "
1051 "start the array.\n");
1052 else if (!enough(info.array.level,
1053 info.array.raid_disks,
1054 info.array.layout, clean,
1055 avail, okcnt))
1056 fprintf(stderr, Name ": Not enough devices to "
1057 "start the array while not clean "
1058 "- consider --force.\n");
1059
1060 if (auto_assem)
1061 ioctl(mdfd, STOP_ARRAY, NULL);
1062 close(mdfd);
1063 return 1;
1064 }
1065 if (runstop == -1) {
1066 fprintf(stderr, Name ": %s assembled from %d drive%s",
1067 mddev, okcnt, okcnt==1?"":"s");
1068 if (okcnt != info.array.raid_disks)
1069 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1070 fprintf(stderr, ", but not started.\n");
1071 close(mdfd);
1072 return 0;
1073 }
1074 if (verbose >= -1) {
1075 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1076 if (sparecnt)
1077 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1078 if (!enough(info.array.level, info.array.raid_disks,
1079 info.array.layout, 1, avail, okcnt))
1080 fprintf(stderr, " - not enough to start the array.\n");
1081 else if (!enough(info.array.level,
1082 info.array.raid_disks,
1083 info.array.layout, clean,
1084 avail, okcnt))
1085 fprintf(stderr, " - not enough to start the "
1086 "array while not clean - consider "
1087 "--force.\n");
1088 else {
1089 if (req_cnt == info.array.raid_disks)
1090 fprintf(stderr, " - need all %d to start it", req_cnt);
1091 else
1092 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1093 fprintf(stderr, " (use --run to insist).\n");
1094 }
1095 }
1096 if (auto_assem)
1097 ioctl(mdfd, STOP_ARRAY, NULL);
1098 return 1;
1099 } else {
1100 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1101 * been updated to point to the current locations of devices.
1102 * so we can just start the array
1103 */
1104 unsigned long dev;
1105 dev = makedev(devices[chosen_drive].i.disk.major,
1106 devices[chosen_drive].i.disk.minor);
1107 if (ioctl(mdfd, START_ARRAY, dev)) {
1108 fprintf(stderr, Name ": Cannot start array: %s\n",
1109 strerror(errno));
1110 }
1111
1112 }
1113 close(mdfd);
1114 return 0;
1115 }