]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Support nameing of version-1 arrays.
[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 struct mddev_ident_s ident2;
120
121 vers = md_get_version(mdfd);
122 if (vers <= 0) {
123 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
124 return 1;
125 }
126 if (vers < 9000) {
127 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
128 " Upgrade your kernel or try --build\n");
129 return 1;
130 }
131 if (get_linux_version() < 2004000)
132 old_linux = 1;
133
134 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
135 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
136 mddev);
137 return 1;
138 }
139 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
140
141 /*
142 * If any subdevs are listed, then any that don't
143 * match ident are discarded. Remainder must all match and
144 * become the array.
145 * If no subdevs, then we scan all devices in the config file, but
146 * there must be something in the identity
147 */
148
149 if (!devlist &&
150 ident->uuid_set == 0 &&
151 ident->super_minor < 0 &&
152 ident->devices == NULL) {
153 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
154 mddev);
155 return 1;
156 }
157 if (devlist == NULL)
158 devlist = conf_get_devs(conffile);
159 else inargv = 1;
160
161 tmpdev = devlist; num_devs = 0;
162 while (tmpdev) {
163 num_devs++;
164 tmpdev = tmpdev->next;
165 }
166 devices = malloc(num_devs * sizeof(*devices));
167
168 if (!st && ident->st) st = ident->st;
169
170 if (verbose)
171 fprintf(stderr, Name ": looking for devices for %s\n",
172 mddev);
173
174 while ( devlist) {
175 char *devname;
176 int dfd;
177 struct stat stb;
178 struct supertype *tst = st;
179
180 devname = devlist->devname;
181 devlist = devlist->next;
182
183 if (ident->devices &&
184 !match_oneof(ident->devices, devname)) {
185 if (inargv || verbose)
186 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
187 continue;
188 }
189
190 if (super) {
191 free(super);
192 super = NULL;
193 }
194
195 dfd = open(devname, O_RDONLY|O_EXCL, 0);
196 if (dfd < 0) {
197 if (inargv || verbose)
198 fprintf(stderr, Name ": cannot open device %s: %s\n",
199 devname, strerror(errno));
200 } else if (fstat(dfd, &stb)< 0) {
201 /* Impossible! */
202 fprintf(stderr, Name ": fstat failed for %s: %s\n",
203 devname, strerror(errno));
204 close(dfd);
205 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
206 fprintf(stderr, Name ": %s is not a block device.\n",
207 devname);
208 close(dfd);
209 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
210 if (inargv || verbose)
211 fprintf(stderr, Name ": no recogniseable superblock\n");
212 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
213 if (inargv || verbose)
214 fprintf( stderr, Name ": no RAID superblock on %s\n",
215 devname);
216 close(dfd);
217 } else {
218 tst->ss->getinfo_super(&info, &ident2, super);
219 close(dfd);
220 }
221
222 if (ident->uuid_set &&
223 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
224 if (inargv || verbose)
225 fprintf(stderr, Name ": %s has wrong uuid.\n",
226 devname);
227 continue;
228 }
229 if (ident->name[0] &&
230 (!super || strncmp(ident2.name, ident->name, 32)!=0)) {
231 if (inargv || verbose)
232 fprintf(stderr, Name ": %s has wrong name.\n",
233 devname);
234 continue;
235 }
236 if (ident->super_minor != UnSet &&
237 (!super || ident->super_minor != info.array.md_minor)) {
238 if (inargv || verbose)
239 fprintf(stderr, Name ": %s has wrong super-minor.\n",
240 devname);
241 continue;
242 }
243 if (ident->level != UnSet &&
244 (!super|| ident->level != info.array.level)) {
245 if (inargv || verbose)
246 fprintf(stderr, Name ": %s has wrong raid level.\n",
247 devname);
248 continue;
249 }
250 if (ident->raid_disks != UnSet &&
251 (!super || ident->raid_disks!= info.array.raid_disks)) {
252 if (inargv || verbose)
253 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
254 devname);
255 continue;
256 }
257
258 /* If we are this far, then we are commited to this device.
259 * If the super_block doesn't exist, or doesn't match others,
260 * then we cannot continue
261 */
262
263 if (!super) {
264 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
265 devname);
266 free(first_super);
267 return 1;
268 }
269 st = tst; /* commit to this format, if haven't already */
270 if (st->ss->compare_super(&first_super, super)) {
271 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
272 devname);
273 free(super);
274 free(first_super);
275 return 1;
276 }
277
278 /* looks like a good enough match to update the super block if needed */
279 if (update) {
280 /* prepare useful information in info structures */
281 struct stat stb2;
282 fstat(mdfd, &stb2);
283 info.array.md_minor = minor(stb2.st_rdev);
284
285 st->ss->update_super(&info, super, update, devname, verbose);
286
287 dfd = open(devname, O_RDWR|O_EXCL, 0);
288 if (dfd < 0)
289 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
290 devname);
291 else if (st->ss->store_super(st, dfd, super))
292 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
293 devname);
294 if (dfd >= 0)
295 close(dfd);
296 }
297
298 if (verbose)
299 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
300 devname, mddev, info.disk.raid_disk);
301 devices[devcnt].devname = devname;
302 devices[devcnt].major = major(stb.st_rdev);
303 devices[devcnt].minor = minor(stb.st_rdev);
304 devices[devcnt].oldmajor = info.disk.major;
305 devices[devcnt].oldminor = info.disk.minor;
306 devices[devcnt].events = info.events;
307 devices[devcnt].raid_disk = info.disk.raid_disk;
308 devices[devcnt].uptodate = 0;
309 devices[devcnt].state = info.disk.state;
310 if (most_recent < devcnt) {
311 if (devices[devcnt].events
312 > devices[most_recent].events)
313 most_recent = devcnt;
314 }
315 if (info.array.level == -4)
316 /* with multipath, the raid_disk from the superblock is meaningless */
317 i = devcnt;
318 else
319 i = devices[devcnt].raid_disk;
320 if (i < 10000) {
321 if (i >= bestcnt) {
322 unsigned int newbestcnt = i+10;
323 int *newbest = malloc(sizeof(int)*newbestcnt);
324 unsigned int c;
325 for (c=0; c < newbestcnt; c++)
326 if (c < bestcnt)
327 newbest[c] = best[c];
328 else
329 newbest[c] = -1;
330 if (best)free(best);
331 best = newbest;
332 bestcnt = newbestcnt;
333 }
334 if (best[i] == -1
335 || devices[best[i]].events < devices[devcnt].events)
336 best[i] = devcnt;
337 }
338 devcnt++;
339 }
340
341 if (super)
342 free(super);
343 super = NULL;
344
345 if (update && strcmp(update, "byteorder")==0)
346 st->minor_version = 90;
347
348 if (devcnt == 0) {
349 fprintf(stderr, Name ": no devices found for %s\n",
350 mddev);
351 free(first_super);
352 return 1;
353 }
354
355 st->ss->getinfo_super(&info, &ident2, first_super);
356
357 /* now we have some devices that might be suitable.
358 * I wonder how many
359 */
360 okcnt = 0;
361 sparecnt=0;
362 for (i=0; i< bestcnt ;i++) {
363 int j = best[i];
364 int event_margin = !force;
365 if (j < 0) continue;
366 /* note: we ignore error flags in multipath arrays
367 * as they don't make sense
368 */
369 if (info.array.level != -4)
370 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
371 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
372 sparecnt++;
373 continue;
374 }
375 if (devices[j].events+event_margin >=
376 devices[most_recent].events) {
377 devices[j].uptodate = 1;
378 if (i < info.array.raid_disks)
379 okcnt++;
380 else
381 sparecnt++;
382 }
383 }
384 while (force && !enough(info.array.level, info.array.raid_disks, okcnt)) {
385 /* Choose the newest best drive which is
386 * not up-to-date, update the superblock
387 * and add it.
388 */
389 int fd;
390 chosen_drive = -1;
391 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
392 int j = best[i];
393 if (j>=0 &&
394 !devices[j].uptodate &&
395 devices[j].events > 0 &&
396 (chosen_drive < 0 ||
397 devices[j].events > devices[chosen_drive].events))
398 chosen_drive = j;
399 }
400 if (chosen_drive < 0)
401 break;
402 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
403 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
404 (int)(devices[chosen_drive].events),
405 (int)(devices[most_recent].events));
406 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
407 if (fd < 0) {
408 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
409 devices[chosen_drive].devname);
410 devices[chosen_drive].events = 0;
411 continue;
412 }
413 if (st->ss->load_super(st,fd, &super, NULL)) {
414 close(fd);
415 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
416 devices[chosen_drive].devname);
417 devices[chosen_drive].events = 0;
418 continue;
419 }
420 info.events = devices[most_recent].events;
421 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose);
422
423 if (st->ss->store_super(st, fd, super)) {
424 close(fd);
425 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
426 devices[chosen_drive].devname);
427 devices[chosen_drive].events = 0;
428 free(super);
429 continue;
430 }
431 close(fd);
432 devices[chosen_drive].events = devices[most_recent].events;
433 devices[chosen_drive].uptodate = 1;
434 okcnt++;
435 free(super);
436 }
437
438 /* Now we want to look at the superblock which the kernel will base things on
439 * and compare the devices that we think are working with the devices that the
440 * superblock thinks are working.
441 * If there are differences and --force is given, then update this chosen
442 * superblock.
443 */
444 chosen_drive = -1;
445 super = NULL;
446 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
447 int j = best[i];
448 int fd;
449
450 if (j<0)
451 continue;
452 if (!devices[j].uptodate)
453 continue;
454 chosen_drive = j;
455 if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
456 fprintf(stderr, Name ": Cannot open %s: %s\n",
457 devices[j].devname, strerror(errno));
458 return 1;
459 }
460 if (st->ss->load_super(st,fd, &super, NULL)) {
461 close(fd);
462 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
463 devices[j].devname);
464 return 1;
465 }
466 close(fd);
467 }
468 if (super == NULL) {
469 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
470 return 1;
471 }
472 st->ss->getinfo_super(&info, &ident2, super);
473 for (i=0; i<bestcnt; i++) {
474 int j = best[i];
475 unsigned int desired_state;
476
477 if (i < info.array.raid_disks)
478 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
479 else
480 desired_state = 0;
481
482 if (j<0)
483 continue;
484 if (!devices[j].uptodate)
485 continue;
486 info.disk.number = i;
487 info.disk.raid_disk = i;
488 info.disk.state = desired_state;
489
490 if (devices[j].uptodate &&
491 st->ss->update_super(&info, super, "assemble", NULL, 0)) {
492 if (force) {
493 fprintf(stderr, Name ": "
494 "clearing FAULTY flag for device %d in %s for %s\n",
495 j, mddev, devices[j].devname);
496 change = 1;
497 } else {
498 fprintf(stderr, Name ": "
499 "device %d in %s has wrong state in superblock, but %s seems ok\n",
500 i, mddev, devices[j].devname);
501 }
502 }
503 #if 0
504 if (!devices[j].uptodate &&
505 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
506 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
507 i, mddev);
508 }
509 #endif
510 }
511 if (force && okcnt == info.array.raid_disks-1) {
512 /* FIXME check event count */
513 change += st->ss->update_super(&info, super, "force",
514 devices[chosen_drive].devname, 0);
515 }
516
517 if (change) {
518 int fd;
519 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
520 if (fd < 0) {
521 fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
522 devices[chosen_drive].devname);
523 return 1;
524 }
525 if (st->ss->store_super(st, fd, super)) {
526 close(fd);
527 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
528 devices[chosen_drive].devname);
529 return 1;
530 }
531 close(fd);
532 }
533
534 /* count number of in-sync devices according to the superblock.
535 * We must have this number to start the array without -s or -R
536 */
537 req_cnt = info.array.working_disks;
538
539 /* Almost ready to actually *do* something */
540 if (!old_linux) {
541 int rv;
542 if ((vers % 100) >= 1) { /* can use different versions */
543 mdu_array_info_t inf;
544 memset(&inf, 0, sizeof(inf));
545 inf.major_version = st->ss->major;
546 inf.minor_version = st->minor_version;
547 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
548 } else
549 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
550
551 if (rv) {
552 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
553 mddev, strerror(errno));
554 return 1;
555 }
556 if (ident->bitmap_fd >= 0) {
557 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
558 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
559 return 1;
560 }
561 }
562
563 /* First, add the raid disks, but add the chosen one last */
564 for (i=0; i<= bestcnt; i++) {
565 int j;
566 if (i < bestcnt) {
567 j = best[i];
568 if (j == chosen_drive)
569 continue;
570 } else
571 j = chosen_drive;
572
573 if (j >= 0 /* && devices[j].uptodate */) {
574 mdu_disk_info_t disk;
575 memset(&disk, 0, sizeof(disk));
576 disk.major = devices[j].major;
577 disk.minor = devices[j].minor;
578 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
579 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
580 devices[j].devname,
581 mddev,
582 strerror(errno));
583 if (i < info.array.raid_disks || i == bestcnt)
584 okcnt--;
585 else
586 sparecnt--;
587 } else if (verbose)
588 fprintf(stderr, Name ": added %s to %s as %d\n",
589 devices[j].devname, mddev, devices[j].raid_disk);
590 } else if (verbose && i < info.array.raid_disks)
591 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
592 i, mddev);
593 }
594
595 if (runstop == 1 ||
596 (runstop == 0 &&
597 ( enough(info.array.level, info.array.raid_disks, okcnt) &&
598 (okcnt >= req_cnt || start_partial_ok)
599 ))) {
600 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
601 fprintf(stderr, Name ": %s has been started with %d drive%s",
602 mddev, okcnt, okcnt==1?"":"s");
603 if (okcnt < info.array.raid_disks)
604 fprintf(stderr, " (out of %d)", info.array.raid_disks);
605 if (sparecnt)
606 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
607 fprintf(stderr, ".\n");
608 return 0;
609 }
610 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
611 mddev, strerror(errno));
612 return 1;
613 }
614 if (runstop == -1) {
615 fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
616 mddev, okcnt, okcnt==1?"":"s");
617 return 0;
618 }
619 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
620 if (sparecnt)
621 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
622 if (!enough(info.array.level, info.array.raid_disks, okcnt))
623 fprintf(stderr, " - not enough to start the array.\n");
624 else {
625 if (req_cnt == info.array.raid_disks)
626 fprintf(stderr, " - need all %d to start it", req_cnt);
627 else
628 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
629 fprintf(stderr, " (use --run to insist).\n");
630 }
631 return 1;
632 } else {
633 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
634 * been updated to point to the current locations of devices.
635 * so we can just start the array
636 */
637 unsigned long dev;
638 dev = makedev(devices[chosen_drive].major,
639 devices[chosen_drive].minor);
640 if (ioctl(mdfd, START_ARRAY, dev)) {
641 fprintf(stderr, Name ": Cannot start array: %s\n",
642 strerror(errno));
643 }
644
645 }
646 return 0;
647 }