]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Allow event counters to be off by one
[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 char *avail;
122 int nextspare = 0;
123
124 vers = md_get_version(mdfd);
125 if (vers <= 0) {
126 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
127 return 1;
128 }
129 if (vers < 9000) {
130 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
131 " Upgrade your kernel or try --build\n");
132 return 1;
133 }
134 if (get_linux_version() < 2004000)
135 old_linux = 1;
136
137 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
138 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
139 mddev);
140 return 1;
141 }
142 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
143
144 /*
145 * If any subdevs are listed, then any that don't
146 * match ident are discarded. Remainder must all match and
147 * become the array.
148 * If no subdevs, then we scan all devices in the config file, but
149 * there must be something in the identity
150 */
151
152 if (!devlist &&
153 ident->uuid_set == 0 &&
154 ident->super_minor < 0 &&
155 ident->devices == NULL) {
156 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
157 mddev);
158 return 1;
159 }
160 if (devlist == NULL)
161 devlist = conf_get_devs(conffile);
162 else inargv = 1;
163
164 tmpdev = devlist; num_devs = 0;
165 while (tmpdev) {
166 num_devs++;
167 tmpdev = tmpdev->next;
168 }
169 devices = malloc(num_devs * sizeof(*devices));
170
171 if (!st && ident->st) st = ident->st;
172
173 if (verbose>0)
174 fprintf(stderr, Name ": looking for devices for %s\n",
175 mddev);
176
177 while ( devlist) {
178 char *devname;
179 int dfd;
180 struct stat stb;
181 struct supertype *tst = st;
182
183 devname = devlist->devname;
184 devlist = devlist->next;
185
186 if (ident->devices &&
187 !match_oneof(ident->devices, devname)) {
188 if ((inargv && verbose>=0) || verbose > 0)
189 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
190 continue;
191 }
192
193 if (super) {
194 free(super);
195 super = NULL;
196 }
197
198 dfd = dev_open(devname, O_RDONLY|O_EXCL);
199 if (dfd < 0) {
200 if ((inargv && verbose >= 0) || verbose > 0)
201 fprintf(stderr, Name ": cannot open device %s: %s\n",
202 devname, strerror(errno));
203 } else if (fstat(dfd, &stb)< 0) {
204 /* Impossible! */
205 fprintf(stderr, Name ": fstat failed for %s: %s\n",
206 devname, strerror(errno));
207 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
208 fprintf(stderr, Name ": %s is not a block device.\n",
209 devname);
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 } else {
218 tst->ss->getinfo_super(&info, &ident2, super);
219 }
220 if (dfd >= 0) close(dfd);
221
222 if (ident->uuid_set &&
223 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
224 if ((inargv && verbose >= 0) || verbose > 0)
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 >= 0) || verbose > 0)
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 >= 0) || verbose > 0)
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 >= 0) || verbose > 0)
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 >= 0) || verbose > 0)
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 = dev_open(devname, O_RDWR|O_EXCL);
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 > 0)
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].disk_nr = info.disk.number;
309 devices[devcnt].uptodate = 0;
310 devices[devcnt].state = info.disk.state;
311 if (most_recent < devcnt) {
312 if (devices[devcnt].events
313 > devices[most_recent].events)
314 most_recent = devcnt;
315 }
316 if (info.array.level == -4)
317 /* with multipath, the raid_disk from the superblock is meaningless */
318 i = devcnt;
319 else
320 i = devices[devcnt].raid_disk;
321 if (i+1 == 0) {
322 if (nextspare < info.array.raid_disks)
323 nextspare = info.array.raid_disks;
324 i = nextspare++;
325 }
326 if (i < 10000) {
327 if (i >= bestcnt) {
328 unsigned int newbestcnt = i+10;
329 int *newbest = malloc(sizeof(int)*newbestcnt);
330 unsigned int c;
331 for (c=0; c < newbestcnt; c++)
332 if (c < bestcnt)
333 newbest[c] = best[c];
334 else
335 newbest[c] = -1;
336 if (best)free(best);
337 best = newbest;
338 bestcnt = newbestcnt;
339 }
340 if (best[i] == -1
341 || devices[best[i]].events < devices[devcnt].events)
342 best[i] = devcnt;
343 }
344 devcnt++;
345 }
346
347 if (super)
348 free(super);
349 super = NULL;
350
351 if (update && strcmp(update, "byteorder")==0)
352 st->minor_version = 90;
353
354 if (devcnt == 0) {
355 fprintf(stderr, Name ": no devices found for %s\n",
356 mddev);
357 free(first_super);
358 return 1;
359 }
360
361 st->ss->getinfo_super(&info, &ident2, first_super);
362
363 /* now we have some devices that might be suitable.
364 * I wonder how many
365 */
366 avail = malloc(info.array.raid_disks);
367 memset(avail, 0, info.array.raid_disks);
368 okcnt = 0;
369 sparecnt=0;
370 for (i=0; i< bestcnt ;i++) {
371 int j = best[i];
372 int event_margin = 1; /* always allow a difference of '1'
373 * like the kernel does
374 */
375 if (j < 0) continue;
376 /* note: we ignore error flags in multipath arrays
377 * as they don't make sense
378 */
379 if (info.array.level != -4)
380 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
381 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
382 sparecnt++;
383 continue;
384 }
385 if (devices[j].events+event_margin >=
386 devices[most_recent].events) {
387 devices[j].uptodate = 1;
388 if (i < info.array.raid_disks) {
389 okcnt++;
390 avail[i]=1;
391 } else
392 sparecnt++;
393 }
394 }
395 while (force && !enough(info.array.level, info.array.raid_disks,
396 info.array.layout,
397 avail, okcnt)) {
398 /* Choose the newest best drive which is
399 * not up-to-date, update the superblock
400 * and add it.
401 */
402 int fd;
403 chosen_drive = -1;
404 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
405 int j = best[i];
406 if (j>=0 &&
407 !devices[j].uptodate &&
408 devices[j].events > 0 &&
409 (chosen_drive < 0 ||
410 devices[j].events > devices[chosen_drive].events))
411 chosen_drive = j;
412 }
413 if (chosen_drive < 0)
414 break;
415 if (verbose >= 0)
416 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
417 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
418 (int)(devices[chosen_drive].events),
419 (int)(devices[most_recent].events));
420 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
421 if (fd < 0) {
422 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
423 devices[chosen_drive].devname);
424 devices[chosen_drive].events = 0;
425 continue;
426 }
427 if (st->ss->load_super(st,fd, &super, NULL)) {
428 close(fd);
429 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
430 devices[chosen_drive].devname);
431 devices[chosen_drive].events = 0;
432 continue;
433 }
434 info.events = devices[most_recent].events;
435 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose);
436
437 if (st->ss->store_super(st, fd, super)) {
438 close(fd);
439 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
440 devices[chosen_drive].devname);
441 devices[chosen_drive].events = 0;
442 free(super);
443 continue;
444 }
445 close(fd);
446 devices[chosen_drive].events = devices[most_recent].events;
447 devices[chosen_drive].uptodate = 1;
448 avail[chosen_drive] = 1;
449 okcnt++;
450 free(super);
451 }
452
453 /* Now we want to look at the superblock which the kernel will base things on
454 * and compare the devices that we think are working with the devices that the
455 * superblock thinks are working.
456 * If there are differences and --force is given, then update this chosen
457 * superblock.
458 */
459 chosen_drive = -1;
460 super = NULL;
461 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
462 int j = best[i];
463 int fd;
464
465 if (j<0)
466 continue;
467 if (!devices[j].uptodate)
468 continue;
469 chosen_drive = j;
470 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
471 fprintf(stderr, Name ": Cannot open %s: %s\n",
472 devices[j].devname, strerror(errno));
473 return 1;
474 }
475 if (st->ss->load_super(st,fd, &super, NULL)) {
476 close(fd);
477 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
478 devices[j].devname);
479 return 1;
480 }
481 close(fd);
482 }
483 if (super == NULL) {
484 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
485 return 1;
486 }
487 st->ss->getinfo_super(&info, &ident2, super);
488 for (i=0; i<bestcnt; i++) {
489 int j = best[i];
490 unsigned int desired_state;
491
492 if (i < info.array.raid_disks)
493 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
494 else
495 desired_state = 0;
496
497 if (j<0)
498 continue;
499 if (!devices[j].uptodate)
500 continue;
501 info.disk.number = devices[j].disk_nr;
502 info.disk.raid_disk = i;
503 info.disk.state = desired_state;
504
505 if (devices[j].uptodate &&
506 st->ss->update_super(&info, super, "assemble", NULL, verbose)) {
507 if (force) {
508 if (verbose >= 0)
509 fprintf(stderr, Name ": "
510 "clearing FAULTY flag for device %d in %s for %s\n",
511 j, mddev, devices[j].devname);
512 change = 1;
513 } else {
514 if (verbose >= -1)
515 fprintf(stderr, Name ": "
516 "device %d in %s has wrong state in superblock, but %s seems ok\n",
517 i, mddev, devices[j].devname);
518 }
519 }
520 #if 0
521 if (!devices[j].uptodate &&
522 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
523 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
524 i, mddev);
525 }
526 #endif
527 }
528 if (force && okcnt == info.array.raid_disks-1) {
529 /* FIXME check event count */
530 change += st->ss->update_super(&info, super, "force",
531 devices[chosen_drive].devname, verbose);
532 }
533
534 if (change) {
535 int fd;
536 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
537 if (fd < 0) {
538 fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
539 devices[chosen_drive].devname);
540 return 1;
541 }
542 if (st->ss->store_super(st, fd, super)) {
543 close(fd);
544 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
545 devices[chosen_drive].devname);
546 return 1;
547 }
548 close(fd);
549 }
550
551 /* count number of in-sync devices according to the superblock.
552 * We must have this number to start the array without -s or -R
553 */
554 req_cnt = info.array.working_disks;
555
556 /* Almost ready to actually *do* something */
557 if (!old_linux) {
558 int rv;
559 if ((vers % 100) >= 1) { /* can use different versions */
560 mdu_array_info_t inf;
561 memset(&inf, 0, sizeof(inf));
562 inf.major_version = st->ss->major;
563 inf.minor_version = st->minor_version;
564 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
565 } else
566 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
567
568 if (rv) {
569 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
570 mddev, strerror(errno));
571 return 1;
572 }
573 if (ident->bitmap_fd >= 0) {
574 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
575 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
576 return 1;
577 }
578 }
579
580 /* First, add the raid disks, but add the chosen one last */
581 for (i=0; i<= bestcnt; i++) {
582 int j;
583 if (i < bestcnt) {
584 j = best[i];
585 if (j == chosen_drive)
586 continue;
587 } else
588 j = chosen_drive;
589
590 if (j >= 0 /* && devices[j].uptodate */) {
591 mdu_disk_info_t disk;
592 memset(&disk, 0, sizeof(disk));
593 disk.major = devices[j].major;
594 disk.minor = devices[j].minor;
595 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
596 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
597 devices[j].devname,
598 mddev,
599 strerror(errno));
600 if (i < info.array.raid_disks || i == bestcnt)
601 okcnt--;
602 else
603 sparecnt--;
604 } else if (verbose > 0)
605 fprintf(stderr, Name ": added %s to %s as %d\n",
606 devices[j].devname, mddev, devices[j].raid_disk);
607 } else if (verbose > 0 && i < info.array.raid_disks)
608 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
609 i, mddev);
610 }
611
612 if (runstop == 1 ||
613 (runstop == 0 &&
614 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
615 (okcnt >= req_cnt || start_partial_ok)
616 ))) {
617 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
618 if (verbose >= 0) {
619 fprintf(stderr, Name ": %s has been started with %d drive%s",
620 mddev, okcnt, okcnt==1?"":"s");
621 if (okcnt < info.array.raid_disks)
622 fprintf(stderr, " (out of %d)", info.array.raid_disks);
623 if (sparecnt)
624 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
625 fprintf(stderr, ".\n");
626 }
627 return 0;
628 }
629 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
630 mddev, strerror(errno));
631 return 1;
632 }
633 if (runstop == -1) {
634 fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
635 mddev, okcnt, okcnt==1?"":"s");
636 return 0;
637 }
638 if (verbose >= 0) {
639 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
640 if (sparecnt)
641 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
642 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
643 fprintf(stderr, " - not enough to start the array.\n");
644 else {
645 if (req_cnt == info.array.raid_disks)
646 fprintf(stderr, " - need all %d to start it", req_cnt);
647 else
648 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
649 fprintf(stderr, " (use --run to insist).\n");
650 }
651 }
652 return 1;
653 } else {
654 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
655 * been updated to point to the current locations of devices.
656 * so we can just start the array
657 */
658 unsigned long dev;
659 dev = makedev(devices[chosen_drive].major,
660 devices[chosen_drive].minor);
661 if (ioctl(mdfd, START_ARRAY, dev)) {
662 fprintf(stderr, Name ": Cannot start array: %s\n",
663 strerror(errno));
664 }
665
666 }
667 return 0;
668 }