]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Generate 'change' uevents when arrays change in non-obvious ways.
[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 if (info.name[0] == 0 && info.array.level == LEVEL_CONTAINER) {
427 name = info.text_version;
428 trustworthy = METADATA;
429 } else
430 trustworthy = FOREIGN;
431 break;
432 }
433 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
434 chosen_name);
435 if (mdfd < 0) {
436 st->ss->free_super(st);
437 free(devices);
438 if (auto_assem)
439 goto try_again;
440 return 1;
441 }
442 mddev = chosen_name;
443 vers = md_get_version(mdfd);
444 if (vers < 9000) {
445 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
446 " Upgrade your kernel or try --build\n");
447 close(mdfd);
448 return 1;
449 }
450 if (ioctl(mdfd, GET_ARRAY_INFO, &tmp_inf)==0) {
451 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
452 mddev);
453 for (tmpdev = devlist ;
454 tmpdev && tmpdev->used != 1;
455 tmpdev = tmpdev->next)
456 ;
457 if (tmpdev && auto_assem)
458 fprintf(stderr, Name ": %s needed for %s...\n",
459 mddev, tmpdev->devname);
460 close(mdfd);
461 mdfd = -3;
462 st->ss->free_super(st);
463 free(devices);
464 if (auto_assem)
465 goto try_again;
466 return 1;
467 }
468 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
469
470 /* Ok, no bad inconsistancy, we can try updating etc */
471 bitmap_done = 0;
472 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
473 char *devname = tmpdev->devname;
474 struct stat stb;
475 /* looks like a good enough match to update the super block if needed */
476 #ifndef MDASSEMBLE
477 if (update) {
478 int dfd;
479 /* prepare useful information in info structures */
480 struct stat stb2;
481 struct supertype *tst;
482 fstat(mdfd, &stb2);
483
484 if (strcmp(update, "uuid")==0 &&
485 !ident->uuid_set) {
486 int rfd;
487 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
488 read(rfd, ident->uuid, 16) != 16) {
489 *(__u32*)(ident->uuid) = random();
490 *(__u32*)(ident->uuid+1) = random();
491 *(__u32*)(ident->uuid+2) = random();
492 *(__u32*)(ident->uuid+3) = random();
493 }
494 if (rfd >= 0) close(rfd);
495 }
496 dfd = dev_open(devname, O_RDWR|O_EXCL);
497
498 remove_partitions(dfd);
499
500 tst = dup_super(st);
501 tst->ss->load_super(tst, dfd, NULL);
502 tst->ss->getinfo_super(tst, &info);
503
504 memcpy(info.uuid, ident->uuid, 16);
505 strcpy(info.name, ident->name);
506 info.array.md_minor = minor(stb2.st_rdev);
507
508 tst->ss->update_super(tst, &info, update,
509 devname, verbose,
510 ident->uuid_set, homehost);
511 if (strcmp(update, "uuid")==0 &&
512 !ident->uuid_set) {
513 ident->uuid_set = 1;
514 memcpy(ident->uuid, info.uuid, 16);
515 }
516 if (dfd < 0)
517 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
518 devname);
519 else if (tst->ss->store_super(tst, dfd))
520 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
521 devname);
522 if (dfd >= 0)
523 close(dfd);
524
525 if (strcmp(update, "uuid")==0 &&
526 ident->bitmap_fd >= 0 && !bitmap_done) {
527 if (bitmap_update_uuid(ident->bitmap_fd,
528 info.uuid,
529 tst->ss->swapuuid) != 0)
530 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
531 else
532 bitmap_done = 1;
533 }
534 tst->ss->free_super(tst);
535 } else
536 #endif
537 {
538 struct supertype *tst = dup_super(st);
539 int dfd;
540 dfd = dev_open(devname, O_RDWR|O_EXCL);
541
542 remove_partitions(dfd);
543
544 tst->ss->load_super(tst, dfd, NULL);
545 tst->ss->getinfo_super(tst, &info);
546 tst->ss->free_super(tst);
547 close(dfd);
548 }
549
550 stat(devname, &stb);
551
552 if (verbose > 0)
553 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
554 devname, mddev, info.disk.raid_disk);
555 devices[devcnt].devname = devname;
556 devices[devcnt].uptodate = 0;
557 devices[devcnt].i = info;
558 devices[devcnt].i.disk.major = major(stb.st_rdev);
559 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
560 if (most_recent < devcnt) {
561 if (devices[devcnt].i.events
562 > devices[most_recent].i.events)
563 most_recent = devcnt;
564 }
565 if (info.array.level == -4)
566 /* with multipath, the raid_disk from the superblock is meaningless */
567 i = devcnt;
568 else
569 i = devices[devcnt].i.disk.raid_disk;
570 if (i+1 == 0) {
571 if (nextspare < info.array.raid_disks)
572 nextspare = info.array.raid_disks;
573 i = nextspare++;
574 } else {
575 if (i >= info.array.raid_disks &&
576 i >= nextspare)
577 nextspare = i+1;
578 }
579 if (i < 10000) {
580 if (i >= bestcnt) {
581 unsigned int newbestcnt = i+10;
582 int *newbest = malloc(sizeof(int)*newbestcnt);
583 unsigned int c;
584 for (c=0; c < newbestcnt; c++)
585 if (c < bestcnt)
586 newbest[c] = best[c];
587 else
588 newbest[c] = -1;
589 if (best)free(best);
590 best = newbest;
591 bestcnt = newbestcnt;
592 }
593 if (best[i] >=0 &&
594 devices[best[i]].i.events
595 == devices[devcnt].i.events
596 && (devices[best[i]].i.disk.minor
597 != devices[devcnt].i.disk.minor)
598 && st->ss == &super0
599 && info.array.level != LEVEL_MULTIPATH) {
600 /* two different devices with identical superblock.
601 * Could be a mis-detection caused by overlapping
602 * partitions. fail-safe.
603 */
604 fprintf(stderr, Name ": WARNING %s and %s appear"
605 " to have very similar superblocks.\n"
606 " If they are really different, "
607 "please --zero the superblock on one\n"
608 " If they are the same or overlap,"
609 " please remove one from %s.\n",
610 devices[best[i]].devname, devname,
611 inargv ? "the list" :
612 "the\n DEVICE list in mdadm.conf"
613 );
614 close(mdfd);
615 return 1;
616 }
617 if (best[i] == -1
618 || (devices[best[i]].i.events
619 < devices[devcnt].i.events))
620 best[i] = devcnt;
621 }
622 devcnt++;
623 }
624
625 if (devcnt == 0) {
626 fprintf(stderr, Name ": no devices found for %s\n",
627 mddev);
628 if (st)
629 st->ss->free_super(st);
630 close(mdfd);
631 return 1;
632 }
633
634 if (update && strcmp(update, "byteorder")==0)
635 st->minor_version = 90;
636
637 st->ss->getinfo_super(st, &info);
638 clean = info.array.state & 1;
639
640 /* now we have some devices that might be suitable.
641 * I wonder how many
642 */
643 avail = malloc(info.array.raid_disks);
644 memset(avail, 0, info.array.raid_disks);
645 okcnt = 0;
646 sparecnt=0;
647 for (i=0; i< bestcnt ;i++) {
648 int j = best[i];
649 int event_margin = 1; /* always allow a difference of '1'
650 * like the kernel does
651 */
652 if (j < 0) continue;
653 /* note: we ignore error flags in multipath arrays
654 * as they don't make sense
655 */
656 if (info.array.level != -4)
657 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
658 if (!(devices[j].i.disk.state
659 & (1<<MD_DISK_FAULTY)))
660 sparecnt++;
661 continue;
662 }
663 if (devices[j].i.events+event_margin >=
664 devices[most_recent].i.events) {
665 devices[j].uptodate = 1;
666 if (i < info.array.raid_disks) {
667 okcnt++;
668 avail[i]=1;
669 } else
670 sparecnt++;
671 }
672 }
673 while (force && !enough(info.array.level, info.array.raid_disks,
674 info.array.layout, 1,
675 avail, okcnt)) {
676 /* Choose the newest best drive which is
677 * not up-to-date, update the superblock
678 * and add it.
679 */
680 int fd;
681 struct supertype *tst;
682 long long current_events;
683 chosen_drive = -1;
684 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
685 int j = best[i];
686 if (j>=0 &&
687 !devices[j].uptodate &&
688 devices[j].i.events > 0 &&
689 (chosen_drive < 0 ||
690 devices[j].i.events
691 > devices[chosen_drive].i.events))
692 chosen_drive = j;
693 }
694 if (chosen_drive < 0)
695 break;
696 current_events = devices[chosen_drive].i.events;
697 add_another:
698 if (verbose >= 0)
699 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
700 devices[chosen_drive].devname,
701 devices[chosen_drive].i.disk.raid_disk,
702 (int)(devices[chosen_drive].i.events),
703 (int)(devices[most_recent].i.events));
704 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
705 if (fd < 0) {
706 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
707 devices[chosen_drive].devname);
708 devices[chosen_drive].i.events = 0;
709 continue;
710 }
711 tst = dup_super(st);
712 if (tst->ss->load_super(tst,fd, NULL)) {
713 close(fd);
714 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
715 devices[chosen_drive].devname);
716 devices[chosen_drive].i.events = 0;
717 continue;
718 }
719 info.events = devices[most_recent].i.events;
720 tst->ss->update_super(tst, &info, "force-one",
721 devices[chosen_drive].devname, verbose,
722 0, NULL);
723
724 if (tst->ss->store_super(tst, fd)) {
725 close(fd);
726 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
727 devices[chosen_drive].devname);
728 devices[chosen_drive].i.events = 0;
729 tst->ss->free_super(tst);
730 continue;
731 }
732 close(fd);
733 devices[chosen_drive].i.events = devices[most_recent].i.events;
734 devices[chosen_drive].uptodate = 1;
735 avail[chosen_drive] = 1;
736 okcnt++;
737 tst->ss->free_super(tst);
738
739 /* If there are any other drives of the same vintage,
740 * add them in as well. We can't lose and we might gain
741 */
742 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
743 int j = best[i];
744 if (j >= 0 &&
745 !devices[j].uptodate &&
746 devices[j].i.events > 0 &&
747 devices[j].i.events == current_events) {
748 chosen_drive = j;
749 goto add_another;
750 }
751 }
752 }
753
754 /* Now we want to look at the superblock which the kernel will base things on
755 * and compare the devices that we think are working with the devices that the
756 * superblock thinks are working.
757 * If there are differences and --force is given, then update this chosen
758 * superblock.
759 */
760 chosen_drive = -1;
761 st->ss->free_super(st);
762 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
763 int j = best[i];
764 int fd;
765
766 if (j<0)
767 continue;
768 if (!devices[j].uptodate)
769 continue;
770 chosen_drive = j;
771 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
772 fprintf(stderr, Name ": Cannot open %s: %s\n",
773 devices[j].devname, strerror(errno));
774 close(mdfd);
775 return 1;
776 }
777 if (st->ss->load_super(st,fd, NULL)) {
778 close(fd);
779 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
780 devices[j].devname);
781 close(mdfd);
782 return 1;
783 }
784 close(fd);
785 }
786 if (st->sb == NULL) {
787 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
788 close(mdfd);
789 return 1;
790 }
791 st->ss->getinfo_super(st, &info);
792 #ifndef MDASSEMBLE
793 sysfs_init(&info, mdfd, 0);
794 #endif
795 for (i=0; i<bestcnt; i++) {
796 int j = best[i];
797 unsigned int desired_state;
798
799 if (i < info.array.raid_disks)
800 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
801 else
802 desired_state = 0;
803
804 if (j<0)
805 continue;
806 if (!devices[j].uptodate)
807 continue;
808
809 devices[j].i.disk.state = desired_state;
810
811 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
812 verbose, 0, NULL)) {
813 if (force) {
814 if (verbose >= 0)
815 fprintf(stderr, Name ": "
816 "clearing FAULTY flag for device %d in %s for %s\n",
817 j, mddev, devices[j].devname);
818 change = 1;
819 } else {
820 if (verbose >= -1)
821 fprintf(stderr, Name ": "
822 "device %d in %s has wrong state in superblock, but %s seems ok\n",
823 i, mddev, devices[j].devname);
824 }
825 }
826 #if 0
827 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
828 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
829 i, mddev);
830 }
831 #endif
832 }
833 if (force && !clean &&
834 !enough(info.array.level, info.array.raid_disks,
835 info.array.layout, clean,
836 avail, okcnt)) {
837 change += st->ss->update_super(st, &info, "force-array",
838 devices[chosen_drive].devname, verbose,
839 0, NULL);
840 clean = 1;
841 }
842
843 if (change) {
844 int fd;
845 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
846 if (fd < 0) {
847 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
848 devices[chosen_drive].devname);
849 close(mdfd);
850 return 1;
851 }
852 if (st->ss->store_super(st, fd)) {
853 close(fd);
854 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
855 devices[chosen_drive].devname);
856 close(mdfd);
857 return 1;
858 }
859 close(fd);
860 }
861
862 /* If we are in the middle of a reshape we may need to restore saved data
863 * that was moved aside due to the reshape overwriting live data
864 * The code of doing this lives in Grow.c
865 */
866 #ifndef MDASSEMBLE
867 if (info.reshape_active) {
868 int err = 0;
869 int *fdlist = malloc(sizeof(int)* bestcnt);
870 for (i=0; i<bestcnt; i++) {
871 int j = best[i];
872 if (j >= 0) {
873 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
874 if (fdlist[i] < 0) {
875 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
876 devices[j].devname);
877 err = 1;
878 break;
879 }
880 } else
881 fdlist[i] = -1;
882 }
883 if (!err)
884 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
885 while (i>0) {
886 i--;
887 if (fdlist[i]>=0) close(fdlist[i]);
888 }
889 if (err) {
890 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
891 close(mdfd);
892 return err;
893 }
894 }
895 #endif
896 /* count number of in-sync devices according to the superblock.
897 * We must have this number to start the array without -s or -R
898 */
899 req_cnt = info.array.working_disks;
900
901 /* Almost ready to actually *do* something */
902 if (!old_linux) {
903 int rv;
904
905 /* First, fill in the map, so that udev can find our name
906 * as soon as we become active.
907 */
908 map_update(NULL, fd2devnum(mdfd), info.text_version,
909 info.uuid, chosen_name);
910
911 rv = set_array_info(mdfd, st, &info);
912 if (rv) {
913 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
914 mddev, strerror(errno));
915 close(mdfd);
916 return 1;
917 }
918 if (ident->bitmap_fd >= 0) {
919 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
920 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
921 close(mdfd);
922 return 1;
923 }
924 } else if (ident->bitmap_file) {
925 /* From config file */
926 int bmfd = open(ident->bitmap_file, O_RDWR);
927 if (bmfd < 0) {
928 fprintf(stderr, Name ": Could not open bitmap file %s\n",
929 ident->bitmap_file);
930 close(mdfd);
931 return 1;
932 }
933 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
934 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
935 close(bmfd);
936 close(mdfd);
937 return 1;
938 }
939 close(bmfd);
940 }
941
942 /* First, add the raid disks, but add the chosen one last */
943 for (i=0; i<= bestcnt; i++) {
944 int j;
945 if (i < bestcnt) {
946 j = best[i];
947 if (j == chosen_drive)
948 continue;
949 } else
950 j = chosen_drive;
951
952 if (j >= 0 /* && devices[j].uptodate */) {
953 rv = add_disk(mdfd, st, &info, &devices[j].i);
954
955 if (rv) {
956 fprintf(stderr, Name ": failed to add "
957 "%s to %s: %s\n",
958 devices[j].devname,
959 mddev,
960 strerror(errno));
961 if (i < info.array.raid_disks
962 || i == bestcnt)
963 okcnt--;
964 else
965 sparecnt--;
966 } else if (verbose > 0)
967 fprintf(stderr, Name ": added %s "
968 "to %s as %d\n",
969 devices[j].devname, mddev,
970 devices[j].i.disk.raid_disk);
971 } else if (verbose > 0 && i < info.array.raid_disks)
972 fprintf(stderr, Name ": no uptodate device for "
973 "slot %d of %s\n",
974 i, mddev);
975 }
976
977 if (info.array.level == LEVEL_CONTAINER) {
978 if (verbose >= 0) {
979 fprintf(stderr, Name ": Container %s has been "
980 "assembled with %d drive%s",
981 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
982 if (okcnt < info.array.raid_disks)
983 fprintf(stderr, " (out of %d)",
984 info.array.raid_disks);
985 fprintf(stderr, "\n");
986 }
987 sysfs_uevent(&info, "change");
988 close(mdfd);
989 return 0;
990 }
991
992 if (runstop == 1 ||
993 (runstop <= 0 &&
994 ( enough(info.array.level, info.array.raid_disks,
995 info.array.layout, clean, avail, okcnt) &&
996 (okcnt >= req_cnt || start_partial_ok)
997 ))) {
998 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
999 if (verbose >= 0) {
1000 fprintf(stderr, Name ": %s has been started with %d drive%s",
1001 mddev, okcnt, okcnt==1?"":"s");
1002 if (okcnt < info.array.raid_disks)
1003 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1004 if (sparecnt)
1005 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1006 fprintf(stderr, ".\n");
1007 }
1008 close(mdfd);
1009 if (auto_assem) {
1010 int usecs = 1;
1011 /* There is a nasty race with 'mdadm --monitor'.
1012 * If it opens this device before we close it,
1013 * it gets an incomplete open on which IO
1014 * doesn't work and the capacity is
1015 * wrong.
1016 * If we reopen (to check for layered devices)
1017 * before --monitor closes, we loose.
1018 *
1019 * So: wait upto 1 second for there to be
1020 * a non-zero capacity.
1021 */
1022 while (usecs < 1000) {
1023 mdfd = open(mddev, O_RDONLY);
1024 if (mdfd >= 0) {
1025 unsigned long long size;
1026 if (get_dev_size(mdfd, NULL, &size) &&
1027 size > 0)
1028 break;
1029 close(mdfd);
1030 }
1031 usleep(usecs);
1032 usecs <<= 1;
1033 }
1034 }
1035 return 0;
1036 }
1037 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1038 mddev, strerror(errno));
1039
1040 if (!enough(info.array.level, info.array.raid_disks,
1041 info.array.layout, 1, avail, okcnt))
1042 fprintf(stderr, Name ": Not enough devices to "
1043 "start the array.\n");
1044 else if (!enough(info.array.level,
1045 info.array.raid_disks,
1046 info.array.layout, clean,
1047 avail, okcnt))
1048 fprintf(stderr, Name ": Not enough devices to "
1049 "start the array while not clean "
1050 "- consider --force.\n");
1051
1052 if (auto_assem)
1053 ioctl(mdfd, STOP_ARRAY, NULL);
1054 close(mdfd);
1055 return 1;
1056 }
1057 if (runstop == -1) {
1058 fprintf(stderr, Name ": %s assembled from %d drive%s",
1059 mddev, okcnt, okcnt==1?"":"s");
1060 if (okcnt != info.array.raid_disks)
1061 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1062 fprintf(stderr, ", but not started.\n");
1063 close(mdfd);
1064 return 0;
1065 }
1066 if (verbose >= -1) {
1067 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1068 if (sparecnt)
1069 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1070 if (!enough(info.array.level, info.array.raid_disks,
1071 info.array.layout, 1, avail, okcnt))
1072 fprintf(stderr, " - not enough to start the array.\n");
1073 else if (!enough(info.array.level,
1074 info.array.raid_disks,
1075 info.array.layout, clean,
1076 avail, okcnt))
1077 fprintf(stderr, " - not enough to start the "
1078 "array while not clean - consider "
1079 "--force.\n");
1080 else {
1081 if (req_cnt == info.array.raid_disks)
1082 fprintf(stderr, " - need all %d to start it", req_cnt);
1083 else
1084 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1085 fprintf(stderr, " (use --run to insist).\n");
1086 }
1087 }
1088 if (auto_assem)
1089 ioctl(mdfd, STOP_ARRAY, NULL);
1090 return 1;
1091 } else {
1092 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1093 * been updated to point to the current locations of devices.
1094 * so we can just start the array
1095 */
1096 unsigned long dev;
1097 dev = makedev(devices[chosen_drive].i.disk.major,
1098 devices[chosen_drive].i.disk.minor);
1099 if (ioctl(mdfd, START_ARRAY, dev)) {
1100 fprintf(stderr, Name ": Cannot start array: %s\n",
1101 strerror(errno));
1102 }
1103
1104 }
1105 close(mdfd);
1106 return 0;
1107 }