]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
remove ANNOUNCE-2.0-devel-?
[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 int disk_nr;
105 } *devices;
106 int *best = NULL; /* indexed by raid_disk */
107 unsigned int bestcnt = 0;
108 int devcnt = 0;
109 unsigned int okcnt, sparecnt;
110 unsigned int req_cnt;
111 unsigned int i;
112 int most_recent = 0;
113 int chosen_drive;
114 int change = 0;
115 int inargv = 0;
116 int start_partial_ok = force || devlist==NULL;
117 unsigned int num_devs;
118 mddev_dev_t tmpdev;
119 struct mdinfo info;
120 struct mddev_ident_s ident2;
121
122 vers = md_get_version(mdfd);
123 if (vers <= 0) {
124 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
125 return 1;
126 }
127 if (vers < 9000) {
128 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
129 " Upgrade your kernel or try --build\n");
130 return 1;
131 }
132 if (get_linux_version() < 2004000)
133 old_linux = 1;
134
135 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
136 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
137 mddev);
138 return 1;
139 }
140 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
141
142 /*
143 * If any subdevs are listed, then any that don't
144 * match ident are discarded. Remainder must all match and
145 * become the array.
146 * If no subdevs, then we scan all devices in the config file, but
147 * there must be something in the identity
148 */
149
150 if (!devlist &&
151 ident->uuid_set == 0 &&
152 ident->super_minor < 0 &&
153 ident->devices == NULL) {
154 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
155 mddev);
156 return 1;
157 }
158 if (devlist == NULL)
159 devlist = conf_get_devs(conffile);
160 else inargv = 1;
161
162 tmpdev = devlist; num_devs = 0;
163 while (tmpdev) {
164 num_devs++;
165 tmpdev = tmpdev->next;
166 }
167 devices = malloc(num_devs * sizeof(*devices));
168
169 if (!st && ident->st) st = ident->st;
170
171 if (verbose>0)
172 fprintf(stderr, Name ": looking for devices for %s\n",
173 mddev);
174
175 while ( devlist) {
176 char *devname;
177 int dfd;
178 struct stat stb;
179 struct supertype *tst = st;
180
181 devname = devlist->devname;
182 devlist = devlist->next;
183
184 if (ident->devices &&
185 !match_oneof(ident->devices, devname)) {
186 if ((inargv && verbose>=0) || verbose > 0)
187 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
188 continue;
189 }
190
191 if (super) {
192 free(super);
193 super = NULL;
194 }
195
196 dfd = open(devname, O_RDONLY|O_EXCL, 0);
197 if (dfd < 0) {
198 if ((inargv && verbose >= 0) || verbose > 0)
199 fprintf(stderr, Name ": cannot open device %s: %s\n",
200 devname, strerror(errno));
201 } else if (fstat(dfd, &stb)< 0) {
202 /* Impossible! */
203 fprintf(stderr, Name ": fstat failed for %s: %s\n",
204 devname, strerror(errno));
205 close(dfd);
206 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
207 fprintf(stderr, Name ": %s is not a block device.\n",
208 devname);
209 close(dfd);
210 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
211 if ((inargv && verbose >= 0) || verbose > 0)
212 fprintf(stderr, Name ": no recogniseable superblock\n");
213 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
214 if ((inargv && verbose >= 0) || verbose > 0)
215 fprintf( stderr, Name ": no RAID superblock on %s\n",
216 devname);
217 close(dfd);
218 } else {
219 tst->ss->getinfo_super(&info, &ident2, super);
220 close(dfd);
221 }
222
223 if (ident->uuid_set &&
224 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
225 if ((inargv && verbose >= 0) || verbose > 0)
226 fprintf(stderr, Name ": %s has wrong uuid.\n",
227 devname);
228 continue;
229 }
230 if (ident->name[0] &&
231 (!super || strncmp(ident2.name, ident->name, 32)!=0)) {
232 if ((inargv && verbose >= 0) || verbose > 0)
233 fprintf(stderr, Name ": %s has wrong name.\n",
234 devname);
235 continue;
236 }
237 if (ident->super_minor != UnSet &&
238 (!super || ident->super_minor != info.array.md_minor)) {
239 if ((inargv && verbose >= 0) || verbose > 0)
240 fprintf(stderr, Name ": %s has wrong super-minor.\n",
241 devname);
242 continue;
243 }
244 if (ident->level != UnSet &&
245 (!super|| ident->level != info.array.level)) {
246 if ((inargv && verbose >= 0) || verbose > 0)
247 fprintf(stderr, Name ": %s has wrong raid level.\n",
248 devname);
249 continue;
250 }
251 if (ident->raid_disks != UnSet &&
252 (!super || ident->raid_disks!= info.array.raid_disks)) {
253 if ((inargv && verbose >= 0) || verbose > 0)
254 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
255 devname);
256 continue;
257 }
258
259 /* If we are this far, then we are commited to this device.
260 * If the super_block doesn't exist, or doesn't match others,
261 * then we cannot continue
262 */
263
264 if (!super) {
265 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
266 devname);
267 free(first_super);
268 return 1;
269 }
270 st = tst; /* commit to this format, if haven't already */
271 if (st->ss->compare_super(&first_super, super)) {
272 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
273 devname);
274 free(super);
275 free(first_super);
276 return 1;
277 }
278
279 /* looks like a good enough match to update the super block if needed */
280 if (update) {
281 /* prepare useful information in info structures */
282 struct stat stb2;
283 fstat(mdfd, &stb2);
284 info.array.md_minor = minor(stb2.st_rdev);
285
286 st->ss->update_super(&info, super, update, devname, verbose);
287
288 dfd = open(devname, O_RDWR|O_EXCL, 0);
289 if (dfd < 0)
290 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
291 devname);
292 else if (st->ss->store_super(st, dfd, super))
293 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
294 devname);
295 if (dfd >= 0)
296 close(dfd);
297 }
298
299 if (verbose > 0)
300 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
301 devname, mddev, info.disk.raid_disk);
302 devices[devcnt].devname = devname;
303 devices[devcnt].major = major(stb.st_rdev);
304 devices[devcnt].minor = minor(stb.st_rdev);
305 devices[devcnt].oldmajor = info.disk.major;
306 devices[devcnt].oldminor = info.disk.minor;
307 devices[devcnt].events = info.events;
308 devices[devcnt].raid_disk = info.disk.raid_disk;
309 devices[devcnt].disk_nr = info.disk.number;
310 devices[devcnt].uptodate = 0;
311 devices[devcnt].state = info.disk.state;
312 if (most_recent < devcnt) {
313 if (devices[devcnt].events
314 > devices[most_recent].events)
315 most_recent = devcnt;
316 }
317 if (info.array.level == -4)
318 /* with multipath, the raid_disk from the superblock is meaningless */
319 i = devcnt;
320 else
321 i = devices[devcnt].raid_disk;
322 if (i < 10000) {
323 if (i >= bestcnt) {
324 unsigned int newbestcnt = i+10;
325 int *newbest = malloc(sizeof(int)*newbestcnt);
326 unsigned int c;
327 for (c=0; c < newbestcnt; c++)
328 if (c < bestcnt)
329 newbest[c] = best[c];
330 else
331 newbest[c] = -1;
332 if (best)free(best);
333 best = newbest;
334 bestcnt = newbestcnt;
335 }
336 if (best[i] == -1
337 || devices[best[i]].events < devices[devcnt].events)
338 best[i] = devcnt;
339 }
340 devcnt++;
341 }
342
343 if (super)
344 free(super);
345 super = NULL;
346
347 if (update && strcmp(update, "byteorder")==0)
348 st->minor_version = 90;
349
350 if (devcnt == 0) {
351 fprintf(stderr, Name ": no devices found for %s\n",
352 mddev);
353 free(first_super);
354 return 1;
355 }
356
357 st->ss->getinfo_super(&info, &ident2, first_super);
358
359 /* now we have some devices that might be suitable.
360 * I wonder how many
361 */
362 okcnt = 0;
363 sparecnt=0;
364 for (i=0; i< bestcnt ;i++) {
365 int j = best[i];
366 int event_margin = !force;
367 if (j < 0) continue;
368 /* note: we ignore error flags in multipath arrays
369 * as they don't make sense
370 */
371 if (info.array.level != -4)
372 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
373 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
374 sparecnt++;
375 continue;
376 }
377 if (devices[j].events+event_margin >=
378 devices[most_recent].events) {
379 devices[j].uptodate = 1;
380 if (i < info.array.raid_disks)
381 okcnt++;
382 else
383 sparecnt++;
384 }
385 }
386 while (force && !enough(info.array.level, info.array.raid_disks, okcnt)) {
387 /* Choose the newest best drive which is
388 * not up-to-date, update the superblock
389 * and add it.
390 */
391 int fd;
392 chosen_drive = -1;
393 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
394 int j = best[i];
395 if (j>=0 &&
396 !devices[j].uptodate &&
397 devices[j].events > 0 &&
398 (chosen_drive < 0 ||
399 devices[j].events > devices[chosen_drive].events))
400 chosen_drive = j;
401 }
402 if (chosen_drive < 0)
403 break;
404 if (verbose >= 0)
405 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
406 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
407 (int)(devices[chosen_drive].events),
408 (int)(devices[most_recent].events));
409 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
410 if (fd < 0) {
411 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
412 devices[chosen_drive].devname);
413 devices[chosen_drive].events = 0;
414 continue;
415 }
416 if (st->ss->load_super(st,fd, &super, NULL)) {
417 close(fd);
418 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
419 devices[chosen_drive].devname);
420 devices[chosen_drive].events = 0;
421 continue;
422 }
423 info.events = devices[most_recent].events;
424 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose);
425
426 if (st->ss->store_super(st, fd, super)) {
427 close(fd);
428 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
429 devices[chosen_drive].devname);
430 devices[chosen_drive].events = 0;
431 free(super);
432 continue;
433 }
434 close(fd);
435 devices[chosen_drive].events = devices[most_recent].events;
436 devices[chosen_drive].uptodate = 1;
437 okcnt++;
438 free(super);
439 }
440
441 /* Now we want to look at the superblock which the kernel will base things on
442 * and compare the devices that we think are working with the devices that the
443 * superblock thinks are working.
444 * If there are differences and --force is given, then update this chosen
445 * superblock.
446 */
447 chosen_drive = -1;
448 super = NULL;
449 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
450 int j = best[i];
451 int fd;
452
453 if (j<0)
454 continue;
455 if (!devices[j].uptodate)
456 continue;
457 chosen_drive = j;
458 if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
459 fprintf(stderr, Name ": Cannot open %s: %s\n",
460 devices[j].devname, strerror(errno));
461 return 1;
462 }
463 if (st->ss->load_super(st,fd, &super, NULL)) {
464 close(fd);
465 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
466 devices[j].devname);
467 return 1;
468 }
469 close(fd);
470 }
471 if (super == NULL) {
472 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
473 return 1;
474 }
475 st->ss->getinfo_super(&info, &ident2, super);
476 for (i=0; i<bestcnt; i++) {
477 int j = best[i];
478 unsigned int desired_state;
479
480 if (i < info.array.raid_disks)
481 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
482 else
483 desired_state = 0;
484
485 if (j<0)
486 continue;
487 if (!devices[j].uptodate)
488 continue;
489 info.disk.number = devices[j].disk_nr;
490 info.disk.raid_disk = i;
491 info.disk.state = desired_state;
492
493 if (devices[j].uptodate &&
494 st->ss->update_super(&info, super, "assemble", NULL, verbose)) {
495 if (force) {
496 if (verbose >= 0)
497 fprintf(stderr, Name ": "
498 "clearing FAULTY flag for device %d in %s for %s\n",
499 j, mddev, devices[j].devname);
500 change = 1;
501 } else {
502 if (verbose >= -1)
503 fprintf(stderr, Name ": "
504 "device %d in %s has wrong state in superblock, but %s seems ok\n",
505 i, mddev, devices[j].devname);
506 }
507 }
508 #if 0
509 if (!devices[j].uptodate &&
510 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
511 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
512 i, mddev);
513 }
514 #endif
515 }
516 if (force && okcnt == info.array.raid_disks-1) {
517 /* FIXME check event count */
518 change += st->ss->update_super(&info, super, "force",
519 devices[chosen_drive].devname, verbose);
520 }
521
522 if (change) {
523 int fd;
524 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
525 if (fd < 0) {
526 fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
527 devices[chosen_drive].devname);
528 return 1;
529 }
530 if (st->ss->store_super(st, fd, super)) {
531 close(fd);
532 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
533 devices[chosen_drive].devname);
534 return 1;
535 }
536 close(fd);
537 }
538
539 /* count number of in-sync devices according to the superblock.
540 * We must have this number to start the array without -s or -R
541 */
542 req_cnt = info.array.working_disks;
543
544 /* Almost ready to actually *do* something */
545 if (!old_linux) {
546 int rv;
547 if ((vers % 100) >= 1) { /* can use different versions */
548 mdu_array_info_t inf;
549 memset(&inf, 0, sizeof(inf));
550 inf.major_version = st->ss->major;
551 inf.minor_version = st->minor_version;
552 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
553 } else
554 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
555
556 if (rv) {
557 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
558 mddev, strerror(errno));
559 return 1;
560 }
561 if (ident->bitmap_fd >= 0) {
562 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
563 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
564 return 1;
565 }
566 }
567
568 /* First, add the raid disks, but add the chosen one last */
569 for (i=0; i<= bestcnt; i++) {
570 int j;
571 if (i < bestcnt) {
572 j = best[i];
573 if (j == chosen_drive)
574 continue;
575 } else
576 j = chosen_drive;
577
578 if (j >= 0 /* && devices[j].uptodate */) {
579 mdu_disk_info_t disk;
580 memset(&disk, 0, sizeof(disk));
581 disk.major = devices[j].major;
582 disk.minor = devices[j].minor;
583 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
584 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
585 devices[j].devname,
586 mddev,
587 strerror(errno));
588 if (i < info.array.raid_disks || i == bestcnt)
589 okcnt--;
590 else
591 sparecnt--;
592 } else if (verbose > 0)
593 fprintf(stderr, Name ": added %s to %s as %d\n",
594 devices[j].devname, mddev, devices[j].raid_disk);
595 } else if (verbose > 0 && i < info.array.raid_disks)
596 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
597 i, mddev);
598 }
599
600 if (runstop == 1 ||
601 (runstop == 0 &&
602 ( enough(info.array.level, info.array.raid_disks, okcnt) &&
603 (okcnt >= req_cnt || start_partial_ok)
604 ))) {
605 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
606 if (verbose >= 0) {
607 fprintf(stderr, Name ": %s has been started with %d drive%s",
608 mddev, okcnt, okcnt==1?"":"s");
609 if (okcnt < info.array.raid_disks)
610 fprintf(stderr, " (out of %d)", info.array.raid_disks);
611 if (sparecnt)
612 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
613 fprintf(stderr, ".\n");
614 }
615 return 0;
616 }
617 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
618 mddev, strerror(errno));
619 return 1;
620 }
621 if (runstop == -1) {
622 fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
623 mddev, okcnt, okcnt==1?"":"s");
624 return 0;
625 }
626 if (verbose >= 0) {
627 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
628 if (sparecnt)
629 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
630 if (!enough(info.array.level, info.array.raid_disks, okcnt))
631 fprintf(stderr, " - not enough to start the array.\n");
632 else {
633 if (req_cnt == info.array.raid_disks)
634 fprintf(stderr, " - need all %d to start it", req_cnt);
635 else
636 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
637 fprintf(stderr, " (use --run to insist).\n");
638 }
639 }
640 return 1;
641 } else {
642 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
643 * been updated to point to the current locations of devices.
644 * so we can just start the array
645 */
646 unsigned long dev;
647 dev = makedev(devices[chosen_drive].major,
648 devices[chosen_drive].minor);
649 if (ioctl(mdfd, START_ARRAY, dev)) {
650 fprintf(stderr, Name ": Cannot start array: %s\n",
651 strerror(errno));
652 }
653
654 }
655 return 0;
656 }