]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Support fixing of byte-swapped superblocks.
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
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
32 int Assemble(struct supertype *st, char *mddev, int mdfd,
33 mddev_ident_t ident, char *conffile,
34 mddev_dev_t devlist,
35 int readonly, int runstop,
36 char *update,
37 int verbose, int force)
38 {
39 /*
40 * The task of Assemble is to find a collection of
41 * devices that should (according to their superblocks)
42 * form an array, and to give this collection to the MD driver.
43 * In Linux-2.4 and later, this involves submitting a
44 * SET_ARRAY_INFO ioctl with no arg - to prepare
45 * the array - and then submit a number of
46 * ADD_NEW_DISK ioctls to add disks into
47 * the array. Finally RUN_ARRAY might
48 * be submitted to start the array.
49 *
50 * Much of the work of Assemble is in finding and/or
51 * checking the disks to make sure they look right.
52 *
53 * If mddev is not set, then scan must be set and we
54 * read through the config file for dev+uuid mapping
55 * We recurse, setting mddev, for each device that
56 * - isn't running
57 * - has a valid uuid (or any uuid if !uuidset)
58 *
59 * If mddev is set, we try to determine state of md.
60 * check version - must be at least 0.90.0
61 * check kernel version. must be at least 2.4.
62 * If not, we can possibly fall back on START_ARRAY
63 * Try to GET_ARRAY_INFO.
64 * If possible, give up
65 * If not, try to STOP_ARRAY just to make sure
66 *
67 * If !uuidset and scan, look in conf-file for uuid
68 * If not found, give up
69 * If !devlist and scan and uuidset, get list of devs from conf-file
70 *
71 * For each device:
72 * Check superblock - discard if bad
73 * Check uuid (set if we don't have one) - discard if no match
74 * Check superblock similarity if we have a superblock - discard if different
75 * Record events, devicenum
76 * This should give us a list of devices for the array
77 * We should collect the most recent event number
78 *
79 * Count disks with recent enough event count
80 * While force && !enough disks
81 * Choose newest rejected disks, update event count
82 * mark clean and rewrite superblock
83 * If recent kernel:
84 * SET_ARRAY_INFO
85 * foreach device with recent events : ADD_NEW_DISK
86 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
87 * If old kernel:
88 * Check the device numbers in superblock are right
89 * update superblock if any changes
90 * START_ARRAY
91 *
92 */
93 int old_linux = 0;
94 int vers;
95 void *first_super = NULL, *super = NULL;
96 struct {
97 char *devname;
98 unsigned int major, minor;
99 unsigned int oldmajor, oldminor;
100 long long events;
101 int uptodate;
102 int state;
103 int raid_disk;
104 } *devices;
105 int *best = NULL; /* indexed by raid_disk */
106 unsigned int bestcnt = 0;
107 int devcnt = 0;
108 unsigned int okcnt, sparecnt;
109 unsigned int req_cnt;
110 unsigned int i;
111 int most_recent = 0;
112 int chosen_drive;
113 int change = 0;
114 int inargv = 0;
115 int start_partial_ok = force || devlist==NULL;
116 unsigned int num_devs;
117 mddev_dev_t tmpdev;
118 struct mdinfo info;
119
120 vers = md_get_version(mdfd);
121 if (vers <= 0) {
122 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
123 return 1;
124 }
125 if (vers < 9000) {
126 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
127 " Upgrade your kernel or try --build\n");
128 return 1;
129 }
130 if (get_linux_version() < 2004000)
131 old_linux = 1;
132
133 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
134 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
135 mddev);
136 return 1;
137 }
138 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
139
140 /*
141 * If any subdevs are listed, then any that don't
142 * match ident are discarded. Remainder must all match and
143 * become the array.
144 * If no subdevs, then we scan all devices in the config file, but
145 * there must be something in the identity
146 */
147
148 if (!devlist &&
149 ident->uuid_set == 0 &&
150 ident->super_minor < 0 &&
151 ident->devices == NULL) {
152 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
153 mddev);
154 return 1;
155 }
156 if (devlist == NULL)
157 devlist = conf_get_devs(conffile);
158 else inargv = 1;
159
160 tmpdev = devlist; num_devs = 0;
161 while (tmpdev) {
162 num_devs++;
163 tmpdev = tmpdev->next;
164 }
165 devices = malloc(num_devs * sizeof(*devices));
166
167 if (!st && ident->st) st = ident->st;
168
169 if (verbose)
170 fprintf(stderr, Name ": looking for devices for %s\n",
171 mddev);
172
173 while ( devlist) {
174 char *devname;
175 int dfd;
176 struct stat stb;
177 struct supertype *tst = st;
178
179 devname = devlist->devname;
180 devlist = devlist->next;
181
182 if (ident->devices &&
183 !match_oneof(ident->devices, devname)) {
184 if (inargv || verbose)
185 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
186 continue;
187 }
188
189 if (super) {
190 free(super);
191 super = NULL;
192 }
193
194 dfd = open(devname, O_RDONLY|O_EXCL, 0);
195 if (dfd < 0) {
196 if (inargv || verbose)
197 fprintf(stderr, Name ": cannot open device %s: %s\n",
198 devname, strerror(errno));
199 } else if (fstat(dfd, &stb)< 0) {
200 /* Impossible! */
201 fprintf(stderr, Name ": fstat failed for %s: %s\n",
202 devname, strerror(errno));
203 close(dfd);
204 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
205 fprintf(stderr, Name ": %s is not a block device.\n",
206 devname);
207 close(dfd);
208 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
209 if (inargv || verbose)
210 fprintf(stderr, Name ": no recogniseable superblock\n");
211 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
212 if (inargv || verbose)
213 fprintf( stderr, Name ": no RAID superblock on %s\n",
214 devname);
215 close(dfd);
216 } else {
217 tst->ss->getinfo_super(&info, super);
218 close(dfd);
219 }
220
221 if (ident->uuid_set &&
222 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
223 if (inargv || verbose)
224 fprintf(stderr, Name ": %s has wrong uuid.\n",
225 devname);
226 continue;
227 }
228 if (ident->super_minor != UnSet &&
229 (!super || ident->super_minor != info.array.md_minor)) {
230 if (inargv || verbose)
231 fprintf(stderr, Name ": %s has wrong super-minor.\n",
232 devname);
233 continue;
234 }
235 if (ident->level != UnSet &&
236 (!super|| ident->level != info.array.level)) {
237 if (inargv || verbose)
238 fprintf(stderr, Name ": %s has wrong raid level.\n",
239 devname);
240 continue;
241 }
242 if (ident->raid_disks != UnSet &&
243 (!super || ident->raid_disks!= info.array.raid_disks)) {
244 if (inargv || verbose)
245 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
246 devname);
247 continue;
248 }
249
250 /* If we are this far, then we are commited to this device.
251 * If the super_block doesn't exist, or doesn't match others,
252 * then we cannot continue
253 */
254
255 if (!super) {
256 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
257 devname);
258 free(first_super);
259 return 1;
260 }
261 st = tst; /* commit to this format, if haven't already */
262 if (st->ss->compare_super(&first_super, super)) {
263 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
264 devname);
265 free(super);
266 free(first_super);
267 return 1;
268 }
269
270 /* looks like a good enough match to update the super block if needed */
271 if (update) {
272 /* prepare useful information in info structures */
273 struct stat stb2;
274 fstat(mdfd, &stb2);
275 info.array.md_minor = minor(stb2.st_rdev);
276
277 st->ss->update_super(&info, super, update, devname, verbose);
278
279 dfd = open(devname, O_RDWR|O_EXCL, 0);
280 if (dfd < 0)
281 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
282 devname);
283 else if (st->ss->store_super(st, dfd, super))
284 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
285 devname);
286 if (dfd >= 0)
287 close(dfd);
288 }
289
290 if (verbose)
291 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
292 devname, mddev, info.disk.raid_disk);
293 devices[devcnt].devname = devname;
294 devices[devcnt].major = major(stb.st_rdev);
295 devices[devcnt].minor = minor(stb.st_rdev);
296 devices[devcnt].oldmajor = info.disk.major;
297 devices[devcnt].oldminor = info.disk.minor;
298 devices[devcnt].events = info.events;
299 devices[devcnt].raid_disk = info.disk.raid_disk;
300 devices[devcnt].uptodate = 0;
301 devices[devcnt].state = info.disk.state;
302 if (most_recent < devcnt) {
303 if (devices[devcnt].events
304 > devices[most_recent].events)
305 most_recent = devcnt;
306 }
307 if (info.array.level == -4)
308 /* with multipath, the raid_disk from the superblock is meaningless */
309 i = devcnt;
310 else
311 i = devices[devcnt].raid_disk;
312 if (i < 10000) {
313 if (i >= bestcnt) {
314 unsigned int newbestcnt = i+10;
315 int *newbest = malloc(sizeof(int)*newbestcnt);
316 unsigned int c;
317 for (c=0; c < newbestcnt; c++)
318 if (c < bestcnt)
319 newbest[c] = best[c];
320 else
321 newbest[c] = -1;
322 if (best)free(best);
323 best = newbest;
324 bestcnt = newbestcnt;
325 }
326 if (best[i] == -1
327 || devices[best[i]].events < devices[devcnt].events)
328 best[i] = devcnt;
329 }
330 devcnt++;
331 }
332
333 if (super)
334 free(super);
335 super = NULL;
336
337 if (update && strcmp(update, "byteorder")==0)
338 st->minor_version = 90;
339
340 if (devcnt == 0) {
341 fprintf(stderr, Name ": no devices found for %s\n",
342 mddev);
343 free(first_super);
344 return 1;
345 }
346
347 st->ss->getinfo_super(&info, first_super);
348
349 /* now we have some devices that might be suitable.
350 * I wonder how many
351 */
352 okcnt = 0;
353 sparecnt=0;
354 for (i=0; i< bestcnt ;i++) {
355 int j = best[i];
356 int event_margin = !force;
357 if (j < 0) continue;
358 /* note: we ignore error flags in multipath arrays
359 * as they don't make sense
360 */
361 if (info.array.level != -4)
362 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
363 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
364 sparecnt++;
365 continue;
366 }
367 if (devices[j].events+event_margin >=
368 devices[most_recent].events) {
369 devices[j].uptodate = 1;
370 if (i < info.array.raid_disks)
371 okcnt++;
372 else
373 sparecnt++;
374 }
375 }
376 while (force && !enough(info.array.level, info.array.raid_disks, okcnt)) {
377 /* Choose the newest best drive which is
378 * not up-to-date, update the superblock
379 * and add it.
380 */
381 int fd;
382 chosen_drive = -1;
383 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
384 int j = best[i];
385 if (j>=0 &&
386 !devices[j].uptodate &&
387 devices[j].events > 0 &&
388 (chosen_drive < 0 ||
389 devices[j].events > devices[chosen_drive].events))
390 chosen_drive = j;
391 }
392 if (chosen_drive < 0)
393 break;
394 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
395 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
396 (int)(devices[chosen_drive].events),
397 (int)(devices[most_recent].events));
398 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
399 if (fd < 0) {
400 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
401 devices[chosen_drive].devname);
402 devices[chosen_drive].events = 0;
403 continue;
404 }
405 if (st->ss->load_super(st,fd, &super, NULL)) {
406 close(fd);
407 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
408 devices[chosen_drive].devname);
409 devices[chosen_drive].events = 0;
410 continue;
411 }
412 info.events = devices[most_recent].events;
413 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose);
414
415 if (st->ss->store_super(st, fd, super)) {
416 close(fd);
417 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
418 devices[chosen_drive].devname);
419 devices[chosen_drive].events = 0;
420 free(super);
421 continue;
422 }
423 close(fd);
424 devices[chosen_drive].events = devices[most_recent].events;
425 devices[chosen_drive].uptodate = 1;
426 okcnt++;
427 free(super);
428 }
429
430 /* Now we want to look at the superblock which the kernel will base things on
431 * and compare the devices that we think are working with the devices that the
432 * superblock thinks are working.
433 * If there are differences and --force is given, then update this chosen
434 * superblock.
435 */
436 chosen_drive = -1;
437 super = NULL;
438 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
439 int j = best[i];
440 int fd;
441
442 if (j<0)
443 continue;
444 if (!devices[j].uptodate)
445 continue;
446 chosen_drive = j;
447 if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
448 fprintf(stderr, Name ": Cannot open %s: %s\n",
449 devices[j].devname, strerror(errno));
450 return 1;
451 }
452 if (st->ss->load_super(st,fd, &super, NULL)) {
453 close(fd);
454 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
455 devices[j].devname);
456 return 1;
457 }
458 close(fd);
459 }
460 if (super == NULL) {
461 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
462 return 1;
463 }
464 st->ss->getinfo_super(&info, super);
465 for (i=0; i<bestcnt; i++) {
466 int j = best[i];
467 unsigned int desired_state;
468
469 if (i < info.array.raid_disks)
470 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
471 else
472 desired_state = 0;
473
474 if (j<0)
475 continue;
476 if (!devices[j].uptodate)
477 continue;
478 info.disk.number = i;
479 info.disk.raid_disk = i;
480 info.disk.state = desired_state;
481
482 if (devices[j].uptodate &&
483 st->ss->update_super(&info, super, "assemble", NULL, 0)) {
484 if (force) {
485 fprintf(stderr, Name ": "
486 "clearing FAULTY flag for device %d in %s for %s\n",
487 j, mddev, devices[j].devname);
488 change = 1;
489 } else {
490 fprintf(stderr, Name ": "
491 "device %d in %s has wrong state in superblock, but %s seems ok\n",
492 i, mddev, devices[j].devname);
493 }
494 }
495 #if 0
496 if (!devices[j].uptodate &&
497 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
498 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
499 i, mddev);
500 }
501 #endif
502 }
503 if (force && okcnt == info.array.raid_disks-1) {
504 /* FIXME check event count */
505 change += st->ss->update_super(&info, super, "force",
506 devices[chosen_drive].devname, 0);
507 }
508
509 if (change) {
510 int fd;
511 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
512 if (fd < 0) {
513 fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
514 devices[chosen_drive].devname);
515 return 1;
516 }
517 if (st->ss->store_super(st, fd, super)) {
518 close(fd);
519 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
520 devices[chosen_drive].devname);
521 return 1;
522 }
523 close(fd);
524 }
525
526 /* count number of in-sync devices according to the superblock.
527 * We must have this number to start the array without -s or -R
528 */
529 req_cnt = info.array.working_disks;
530
531 /* Almost ready to actually *do* something */
532 if (!old_linux) {
533 int rv;
534 if ((vers % 100) >= 1) { /* can use different versions */
535 mdu_array_info_t inf;
536 memset(&inf, 0, sizeof(inf));
537 inf.major_version = st->ss->major;
538 inf.minor_version = st->minor_version;
539 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
540 } else
541 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
542
543 if (rv) {
544 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
545 mddev, strerror(errno));
546 return 1;
547 }
548 if (ident->bitmap_fd >= 0) {
549 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
550 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
551 return 1;
552 }
553 }
554
555 /* First, add the raid disks, but add the chosen one last */
556 for (i=0; i<= bestcnt; i++) {
557 int j;
558 if (i < bestcnt) {
559 j = best[i];
560 if (j == chosen_drive)
561 continue;
562 } else
563 j = chosen_drive;
564
565 if (j >= 0 /* && devices[j].uptodate */) {
566 mdu_disk_info_t disk;
567 memset(&disk, 0, sizeof(disk));
568 disk.major = devices[j].major;
569 disk.minor = devices[j].minor;
570 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
571 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
572 devices[j].devname,
573 mddev,
574 strerror(errno));
575 if (i < info.array.raid_disks || i == bestcnt)
576 okcnt--;
577 else
578 sparecnt--;
579 } else if (verbose)
580 fprintf(stderr, Name ": added %s to %s as %d\n",
581 devices[j].devname, mddev, devices[j].raid_disk);
582 } else if (verbose && i < info.array.raid_disks)
583 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
584 i, mddev);
585 }
586
587 if (runstop == 1 ||
588 (runstop == 0 &&
589 ( enough(info.array.level, info.array.raid_disks, okcnt) &&
590 (okcnt >= req_cnt || start_partial_ok)
591 ))) {
592 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
593 fprintf(stderr, Name ": %s has been started with %d drive%s",
594 mddev, okcnt, okcnt==1?"":"s");
595 if (okcnt < info.array.raid_disks)
596 fprintf(stderr, " (out of %d)", info.array.raid_disks);
597 if (sparecnt)
598 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
599 fprintf(stderr, ".\n");
600 return 0;
601 }
602 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
603 mddev, strerror(errno));
604 return 1;
605 }
606 if (runstop == -1) {
607 fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
608 mddev, okcnt, okcnt==1?"":"s");
609 return 0;
610 }
611 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
612 if (sparecnt)
613 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
614 if (!enough(info.array.level, info.array.raid_disks, okcnt))
615 fprintf(stderr, " - not enough to start the array.\n");
616 else {
617 if (req_cnt == info.array.raid_disks)
618 fprintf(stderr, " - need all %d to start it", req_cnt);
619 else
620 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
621 fprintf(stderr, " (use --run to insist).\n");
622 }
623 return 1;
624 } else {
625 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
626 * been updated to point to the current locations of devices.
627 * so we can just start the array
628 */
629 unsigned long dev;
630 dev = makedev(devices[chosen_drive].major,
631 devices[chosen_drive].minor);
632 if (ioctl(mdfd, START_ARRAY, dev)) {
633 fprintf(stderr, Name ": Cannot start array: %s\n",
634 strerror(errno));
635 }
636
637 }
638 return 0;
639 }