]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Incremental: adjust to the new naming scheme.
[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
148 memset(&info, 0, sizeof(info));
149
150 if (get_linux_version() < 2004000)
151 old_linux = 1;
152
153 /*
154 * If any subdevs are listed, then any that don't
155 * match ident are discarded. Remainder must all match and
156 * become the array.
157 * If no subdevs, then we scan all devices in the config file, but
158 * there must be something in the identity
159 */
160
161 if (!devlist &&
162 ident->uuid_set == 0 &&
163 ident->super_minor < 0 &&
164 ident->devices == NULL) {
165 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
166 mddev ? mddev : "further assembly");
167 return 1;
168 }
169
170 /* if the configuration specifies a container then we use that to
171 * determine the devices and retrieve the array configuration
172 */
173 #ifndef MDASSEMBLE
174 if (ident->container && ident->member) {
175 int cfd = open(ident->container, O_RDWR);
176 struct mdinfo *mdi;
177 struct supertype container;
178
179 if (verbose>0)
180 fprintf(stderr, Name ": looking to assemble member array %s"
181 " inside container %s\n", ident->member, ident->container);
182 if (cfd < 0) {
183 if (verbose>0)
184 fprintf(stderr, Name ": unable to open container %s: %s\n",
185 ident->container, strerror(errno));
186 return 1;
187 }
188
189 mdi = sysfs_read(cfd, fd2devnum(cfd), GET_VERSION);
190 if (!mdi) {
191 close(cfd);
192 if (verbose>0)
193 fprintf(stderr, Name ": unable to read container %s\n",
194 ident->container);
195 return 1;
196 }
197 container.ss = find_metadata_methods(mdi->text_version);
198 sysfs_free(mdi);
199 if (!container.ss) {
200 close(cfd);
201 fprintf(stderr, Name ": %s uses unknown metadata: %s\n",
202 ident->container, mdi->text_version);
203 return 1;
204 }
205 if (container.ss->load_super(&container, cfd, ident->container)) {
206 fprintf(stderr, Name ": Cannot load metadata for container %s\n",
207 ident->container);
208 return 1;
209 }
210
211 return Incremental_container(&container, ident->container,
212 verbose, runstop, ident->autof,
213 LOCAL);
214 }
215 #endif
216 if (devlist == NULL)
217 devlist = conf_get_devs();
218 else if (mddev)
219 inargv = 1;
220
221 try_again:
222 /* We come back here when doing auto-assembly and attempting some
223 * set of devices failed. Those are now marked as ->used==2 and
224 * we ignore them and try again
225 */
226
227 tmpdev = devlist; num_devs = 0;
228 while (tmpdev) {
229 if (tmpdev->used)
230 tmpdev->used = 2;
231 else
232 num_devs++;
233 tmpdev = tmpdev->next;
234 }
235 devices = malloc(num_devs * sizeof(*devices));
236
237 if (!st && ident->st) st = ident->st;
238
239 if (verbose>0)
240 fprintf(stderr, Name ": looking for devices for %s\n",
241 mddev ? mddev : "further assembly");
242
243 /* first walk the list of devices to find a consistent set
244 * that match the criterea, if that is possible.
245 * We flag the ones we like with 'used'.
246 */
247 for (tmpdev = devlist;
248 tmpdev;
249 tmpdev = tmpdev->next) {
250 char *devname = tmpdev->devname;
251 int dfd;
252 struct stat stb;
253 struct supertype *tst = dup_super(st);
254
255 if (tmpdev->used > 1) continue;
256
257 if (ident->devices &&
258 !match_oneof(ident->devices, devname)) {
259 if ((inargv && verbose>=0) || verbose > 0)
260 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
261 continue;
262 }
263
264 dfd = dev_open(devname, O_RDONLY|O_EXCL);
265 if (dfd < 0) {
266 if ((inargv && verbose >= 0) || verbose > 0)
267 fprintf(stderr, Name ": cannot open device %s: %s\n",
268 devname, strerror(errno));
269 tmpdev->used = 2;
270 } else if (fstat(dfd, &stb)< 0) {
271 /* Impossible! */
272 fprintf(stderr, Name ": fstat failed for %s: %s\n",
273 devname, strerror(errno));
274 tmpdev->used = 2;
275 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
276 fprintf(stderr, Name ": %s is not a block device.\n",
277 devname);
278 tmpdev->used = 2;
279 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
280 if ((inargv && verbose >= 0) || verbose > 0)
281 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
282 devname);
283 tmpdev->used = 2;
284 } else if (tst->ss->load_super(tst,dfd, NULL)) {
285 if ((inargv && verbose >= 0) || verbose > 0)
286 fprintf( stderr, Name ": no RAID superblock on %s\n",
287 devname);
288 } else {
289 tst->ss->getinfo_super(tst, &info);
290 }
291 if (dfd >= 0) close(dfd);
292
293 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
294 (!tst || !tst->sb ||
295 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
296 if ((inargv && verbose >= 0) || verbose > 0)
297 fprintf(stderr, Name ": %s has wrong uuid.\n",
298 devname);
299 goto loop;
300 }
301 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
302 (!tst || !tst->sb ||
303 name_matches(info.name, ident->name, homehost)==0)) {
304 if ((inargv && verbose >= 0) || verbose > 0)
305 fprintf(stderr, Name ": %s has wrong name.\n",
306 devname);
307 goto loop;
308 }
309 if (ident->super_minor != UnSet &&
310 (!tst || !tst->sb ||
311 ident->super_minor != info.array.md_minor)) {
312 if ((inargv && verbose >= 0) || verbose > 0)
313 fprintf(stderr, Name ": %s has wrong super-minor.\n",
314 devname);
315 goto loop;
316 }
317 if (ident->level != UnSet &&
318 (!tst || !tst->sb ||
319 ident->level != info.array.level)) {
320 if ((inargv && verbose >= 0) || verbose > 0)
321 fprintf(stderr, Name ": %s has wrong raid level.\n",
322 devname);
323 goto loop;
324 }
325 if (ident->raid_disks != UnSet &&
326 (!tst || !tst->sb ||
327 ident->raid_disks!= info.array.raid_disks)) {
328 if ((inargv && verbose >= 0) || verbose > 0)
329 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
330 devname);
331 goto loop;
332 }
333 if (auto_assem) {
334 if (tst == NULL || tst->sb == NULL)
335 continue;
336 }
337 /* If we are this far, then we are nearly commited to this device.
338 * If the super_block doesn't exist, or doesn't match others,
339 * then we probably cannot continue
340 * However if one of the arrays is for the homehost, and
341 * the other isn't that can disambiguate.
342 */
343
344 if (!tst || !tst->sb) {
345 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
346 devname);
347 if (st)
348 st->ss->free_super(st);
349 return 1;
350 }
351
352 if (st == NULL)
353 st = dup_super(tst);
354 if (st->minor_version == -1)
355 st->minor_version = tst->minor_version;
356 if (st->ss != tst->ss ||
357 st->minor_version != tst->minor_version ||
358 st->ss->compare_super(st, tst) != 0) {
359 /* Some mismatch. If exactly one array matches this host,
360 * we can resolve on that one.
361 * Or, if we are auto assembling, we just ignore the second
362 * for now.
363 */
364 if (auto_assem)
365 goto loop;
366 if (homehost) {
367 int first = st->ss->match_home(st, homehost);
368 int last = tst->ss->match_home(tst, homehost);
369 if (first != last &&
370 (first == 1 || last == 1)) {
371 /* We can do something */
372 if (first) {/* just ignore this one */
373 if ((inargv && verbose >= 0) || verbose > 0)
374 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
375 devname);
376 goto loop;
377 } else { /* reject all those sofar */
378 mddev_dev_t td;
379 if ((inargv && verbose >= 0) || verbose > 0)
380 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
381 devname);
382 for (td=devlist; td != tmpdev; td=td->next)
383 if (td->used == 1)
384 td->used = 0;
385 tmpdev->used = 1;
386 goto loop;
387 }
388 }
389 }
390 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
391 devname);
392 tst->ss->free_super(tst);
393 st->ss->free_super(st);
394 return 1;
395 }
396
397 tmpdev->used = 1;
398
399 loop:
400 if (tst)
401 tst->ss->free_super(tst);
402 }
403
404 if (!st || !st->sb)
405 return 2;
406
407 /* Now need to open array the device. Use create_mddev */
408 st->ss->getinfo_super(st, &info);
409
410 trustworthy = FOREIGN;
411 switch (st->ss->match_home(st, homehost)) {
412 case 0:
413 trustworthy = FOREIGN;
414 name = info.name;
415 break;
416 case 1:
417 trustworthy = LOCAL;
418 name = strchr(info.name, ':');
419 if (name)
420 name++;
421 else
422 name = info.name;
423 break;
424 case -1:
425 if (info.name[0] == 0 && info.array.level == LEVEL_CONTAINER) {
426 name = info.text_version;
427 trustworthy = METADATA;
428 } else
429 trustworthy = FOREIGN;
430 break;
431 }
432 mdfd = create_mddev(mddev, name, ident->autof, trustworthy, NULL);
433 if (mdfd < 0) {
434 st->ss->free_super(st);
435 free(devices);
436 if (auto_assem)
437 goto try_again;
438 return 1;
439 }
440 vers = md_get_version(mdfd);
441 if (vers < 9000) {
442 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
443 " Upgrade your kernel or try --build\n");
444 close(mdfd);
445 return 1;
446 }
447 if (ioctl(mdfd, GET_ARRAY_INFO, &tmp_inf)==0) {
448 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
449 mddev);
450 for (tmpdev = devlist ;
451 tmpdev && tmpdev->used != 1;
452 tmpdev = tmpdev->next)
453 ;
454 if (tmpdev && auto_assem)
455 fprintf(stderr, Name ": %s needed for %s...\n",
456 mddev, tmpdev->devname);
457 close(mdfd);
458 mdfd = -3;
459 st->ss->free_super(st);
460 free(devices);
461 if (auto_assem)
462 goto try_again;
463 return 1;
464 }
465 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
466
467 /* Ok, no bad inconsistancy, we can try updating etc */
468 bitmap_done = 0;
469 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
470 char *devname = tmpdev->devname;
471 struct stat stb;
472 /* looks like a good enough match to update the super block if needed */
473 #ifndef MDASSEMBLE
474 if (update) {
475 int dfd;
476 /* prepare useful information in info structures */
477 struct stat stb2;
478 struct supertype *tst;
479 fstat(mdfd, &stb2);
480
481 if (strcmp(update, "uuid")==0 &&
482 !ident->uuid_set) {
483 int rfd;
484 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
485 read(rfd, ident->uuid, 16) != 16) {
486 *(__u32*)(ident->uuid) = random();
487 *(__u32*)(ident->uuid+1) = random();
488 *(__u32*)(ident->uuid+2) = random();
489 *(__u32*)(ident->uuid+3) = random();
490 }
491 if (rfd >= 0) close(rfd);
492 }
493 dfd = dev_open(devname, O_RDWR|O_EXCL);
494
495 remove_partitions(dfd);
496
497 tst = dup_super(st);
498 tst->ss->load_super(tst, dfd, NULL);
499 tst->ss->getinfo_super(tst, &info);
500
501 memcpy(info.uuid, ident->uuid, 16);
502 strcpy(info.name, ident->name);
503 info.array.md_minor = minor(stb2.st_rdev);
504
505 tst->ss->update_super(tst, &info, update,
506 devname, verbose,
507 ident->uuid_set, homehost);
508 if (strcmp(update, "uuid")==0 &&
509 !ident->uuid_set) {
510 ident->uuid_set = 1;
511 memcpy(ident->uuid, info.uuid, 16);
512 }
513 if (dfd < 0)
514 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
515 devname);
516 else if (tst->ss->store_super(tst, dfd))
517 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
518 devname);
519 if (dfd >= 0)
520 close(dfd);
521
522 if (strcmp(update, "uuid")==0 &&
523 ident->bitmap_fd >= 0 && !bitmap_done) {
524 if (bitmap_update_uuid(ident->bitmap_fd,
525 info.uuid,
526 tst->ss->swapuuid) != 0)
527 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
528 else
529 bitmap_done = 1;
530 }
531 tst->ss->free_super(tst);
532 } else
533 #endif
534 {
535 struct supertype *tst = dup_super(st);
536 int dfd;
537 dfd = dev_open(devname, O_RDWR|O_EXCL);
538
539 remove_partitions(dfd);
540
541 tst->ss->load_super(tst, dfd, NULL);
542 tst->ss->getinfo_super(tst, &info);
543 tst->ss->free_super(tst);
544 close(dfd);
545 }
546
547 stat(devname, &stb);
548
549 if (verbose > 0)
550 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
551 devname, mddev, info.disk.raid_disk);
552 devices[devcnt].devname = devname;
553 devices[devcnt].uptodate = 0;
554 devices[devcnt].i = info;
555 devices[devcnt].i.disk.major = major(stb.st_rdev);
556 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
557 if (most_recent < devcnt) {
558 if (devices[devcnt].i.events
559 > devices[most_recent].i.events)
560 most_recent = devcnt;
561 }
562 if (info.array.level == -4)
563 /* with multipath, the raid_disk from the superblock is meaningless */
564 i = devcnt;
565 else
566 i = devices[devcnt].i.disk.raid_disk;
567 if (i+1 == 0) {
568 if (nextspare < info.array.raid_disks)
569 nextspare = info.array.raid_disks;
570 i = nextspare++;
571 } else {
572 if (i >= info.array.raid_disks &&
573 i >= nextspare)
574 nextspare = i+1;
575 }
576 if (i < 10000) {
577 if (i >= bestcnt) {
578 unsigned int newbestcnt = i+10;
579 int *newbest = malloc(sizeof(int)*newbestcnt);
580 unsigned int c;
581 for (c=0; c < newbestcnt; c++)
582 if (c < bestcnt)
583 newbest[c] = best[c];
584 else
585 newbest[c] = -1;
586 if (best)free(best);
587 best = newbest;
588 bestcnt = newbestcnt;
589 }
590 if (best[i] >=0 &&
591 devices[best[i]].i.events
592 == devices[devcnt].i.events
593 && (devices[best[i]].i.disk.minor
594 != devices[devcnt].i.disk.minor)
595 && st->ss == &super0
596 && info.array.level != LEVEL_MULTIPATH) {
597 /* two different devices with identical superblock.
598 * Could be a mis-detection caused by overlapping
599 * partitions. fail-safe.
600 */
601 fprintf(stderr, Name ": WARNING %s and %s appear"
602 " to have very similar superblocks.\n"
603 " If they are really different, "
604 "please --zero the superblock on one\n"
605 " If they are the same or overlap,"
606 " please remove one from %s.\n",
607 devices[best[i]].devname, devname,
608 inargv ? "the list" :
609 "the\n DEVICE list in mdadm.conf"
610 );
611 close(mdfd);
612 return 1;
613 }
614 if (best[i] == -1
615 || (devices[best[i]].i.events
616 < devices[devcnt].i.events))
617 best[i] = devcnt;
618 }
619 devcnt++;
620 }
621
622 if (devcnt == 0) {
623 fprintf(stderr, Name ": no devices found for %s\n",
624 mddev);
625 if (st)
626 st->ss->free_super(st);
627 close(mdfd);
628 return 1;
629 }
630
631 if (update && strcmp(update, "byteorder")==0)
632 st->minor_version = 90;
633
634 st->ss->getinfo_super(st, &info);
635 clean = info.array.state & 1;
636
637 /* now we have some devices that might be suitable.
638 * I wonder how many
639 */
640 avail = malloc(info.array.raid_disks);
641 memset(avail, 0, info.array.raid_disks);
642 okcnt = 0;
643 sparecnt=0;
644 for (i=0; i< bestcnt ;i++) {
645 int j = best[i];
646 int event_margin = 1; /* always allow a difference of '1'
647 * like the kernel does
648 */
649 if (j < 0) continue;
650 /* note: we ignore error flags in multipath arrays
651 * as they don't make sense
652 */
653 if (info.array.level != -4)
654 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
655 if (!(devices[j].i.disk.state
656 & (1<<MD_DISK_FAULTY)))
657 sparecnt++;
658 continue;
659 }
660 if (devices[j].i.events+event_margin >=
661 devices[most_recent].i.events) {
662 devices[j].uptodate = 1;
663 if (i < info.array.raid_disks) {
664 okcnt++;
665 avail[i]=1;
666 } else
667 sparecnt++;
668 }
669 }
670 while (force && !enough(info.array.level, info.array.raid_disks,
671 info.array.layout, 1,
672 avail, okcnt)) {
673 /* Choose the newest best drive which is
674 * not up-to-date, update the superblock
675 * and add it.
676 */
677 int fd;
678 struct supertype *tst;
679 long long current_events;
680 chosen_drive = -1;
681 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
682 int j = best[i];
683 if (j>=0 &&
684 !devices[j].uptodate &&
685 devices[j].i.events > 0 &&
686 (chosen_drive < 0 ||
687 devices[j].i.events
688 > devices[chosen_drive].i.events))
689 chosen_drive = j;
690 }
691 if (chosen_drive < 0)
692 break;
693 current_events = devices[chosen_drive].i.events;
694 add_another:
695 if (verbose >= 0)
696 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
697 devices[chosen_drive].devname,
698 devices[chosen_drive].i.disk.raid_disk,
699 (int)(devices[chosen_drive].i.events),
700 (int)(devices[most_recent].i.events));
701 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
702 if (fd < 0) {
703 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
704 devices[chosen_drive].devname);
705 devices[chosen_drive].i.events = 0;
706 continue;
707 }
708 tst = dup_super(st);
709 if (tst->ss->load_super(tst,fd, NULL)) {
710 close(fd);
711 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
712 devices[chosen_drive].devname);
713 devices[chosen_drive].i.events = 0;
714 continue;
715 }
716 info.events = devices[most_recent].i.events;
717 tst->ss->update_super(tst, &info, "force-one",
718 devices[chosen_drive].devname, verbose,
719 0, NULL);
720
721 if (tst->ss->store_super(tst, fd)) {
722 close(fd);
723 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
724 devices[chosen_drive].devname);
725 devices[chosen_drive].i.events = 0;
726 tst->ss->free_super(tst);
727 continue;
728 }
729 close(fd);
730 devices[chosen_drive].i.events = devices[most_recent].i.events;
731 devices[chosen_drive].uptodate = 1;
732 avail[chosen_drive] = 1;
733 okcnt++;
734 tst->ss->free_super(tst);
735
736 /* If there are any other drives of the same vintage,
737 * add them in as well. We can't lose and we might gain
738 */
739 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
740 int j = best[i];
741 if (j >= 0 &&
742 !devices[j].uptodate &&
743 devices[j].i.events > 0 &&
744 devices[j].i.events == current_events) {
745 chosen_drive = j;
746 goto add_another;
747 }
748 }
749 }
750
751 /* Now we want to look at the superblock which the kernel will base things on
752 * and compare the devices that we think are working with the devices that the
753 * superblock thinks are working.
754 * If there are differences and --force is given, then update this chosen
755 * superblock.
756 */
757 chosen_drive = -1;
758 st->ss->free_super(st);
759 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
760 int j = best[i];
761 int fd;
762
763 if (j<0)
764 continue;
765 if (!devices[j].uptodate)
766 continue;
767 chosen_drive = j;
768 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
769 fprintf(stderr, Name ": Cannot open %s: %s\n",
770 devices[j].devname, strerror(errno));
771 close(mdfd);
772 return 1;
773 }
774 if (st->ss->load_super(st,fd, NULL)) {
775 close(fd);
776 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
777 devices[j].devname);
778 close(mdfd);
779 return 1;
780 }
781 close(fd);
782 }
783 if (st->sb == NULL) {
784 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
785 close(mdfd);
786 return 1;
787 }
788 st->ss->getinfo_super(st, &info);
789 #ifndef MDASSEMBLE
790 sysfs_init(&info, mdfd, 0);
791 #endif
792 for (i=0; i<bestcnt; i++) {
793 int j = best[i];
794 unsigned int desired_state;
795
796 if (i < info.array.raid_disks)
797 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
798 else
799 desired_state = 0;
800
801 if (j<0)
802 continue;
803 if (!devices[j].uptodate)
804 continue;
805
806 devices[j].i.disk.state = desired_state;
807
808 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
809 verbose, 0, NULL)) {
810 if (force) {
811 if (verbose >= 0)
812 fprintf(stderr, Name ": "
813 "clearing FAULTY flag for device %d in %s for %s\n",
814 j, mddev, devices[j].devname);
815 change = 1;
816 } else {
817 if (verbose >= -1)
818 fprintf(stderr, Name ": "
819 "device %d in %s has wrong state in superblock, but %s seems ok\n",
820 i, mddev, devices[j].devname);
821 }
822 }
823 #if 0
824 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
825 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
826 i, mddev);
827 }
828 #endif
829 }
830 if (force && !clean &&
831 !enough(info.array.level, info.array.raid_disks,
832 info.array.layout, clean,
833 avail, okcnt)) {
834 change += st->ss->update_super(st, &info, "force-array",
835 devices[chosen_drive].devname, verbose,
836 0, NULL);
837 clean = 1;
838 }
839
840 if (change) {
841 int fd;
842 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
843 if (fd < 0) {
844 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
845 devices[chosen_drive].devname);
846 close(mdfd);
847 return 1;
848 }
849 if (st->ss->store_super(st, fd)) {
850 close(fd);
851 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
852 devices[chosen_drive].devname);
853 close(mdfd);
854 return 1;
855 }
856 close(fd);
857 }
858
859 /* If we are in the middle of a reshape we may need to restore saved data
860 * that was moved aside due to the reshape overwriting live data
861 * The code of doing this lives in Grow.c
862 */
863 #ifndef MDASSEMBLE
864 if (info.reshape_active) {
865 int err = 0;
866 int *fdlist = malloc(sizeof(int)* bestcnt);
867 for (i=0; i<bestcnt; i++) {
868 int j = best[i];
869 if (j >= 0) {
870 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
871 if (fdlist[i] < 0) {
872 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
873 devices[j].devname);
874 err = 1;
875 break;
876 }
877 } else
878 fdlist[i] = -1;
879 }
880 if (!err)
881 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
882 while (i>0) {
883 i--;
884 if (fdlist[i]>=0) close(fdlist[i]);
885 }
886 if (err) {
887 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
888 close(mdfd);
889 return err;
890 }
891 }
892 #endif
893 /* count number of in-sync devices according to the superblock.
894 * We must have this number to start the array without -s or -R
895 */
896 req_cnt = info.array.working_disks;
897
898 /* Almost ready to actually *do* something */
899 if (!old_linux) {
900 int rv;
901
902 rv = set_array_info(mdfd, st, &info);
903 if (rv) {
904 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
905 mddev, strerror(errno));
906 close(mdfd);
907 return 1;
908 }
909 if (ident->bitmap_fd >= 0) {
910 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
911 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
912 close(mdfd);
913 return 1;
914 }
915 } else if (ident->bitmap_file) {
916 /* From config file */
917 int bmfd = open(ident->bitmap_file, O_RDWR);
918 if (bmfd < 0) {
919 fprintf(stderr, Name ": Could not open bitmap file %s\n",
920 ident->bitmap_file);
921 close(mdfd);
922 return 1;
923 }
924 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
925 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
926 close(bmfd);
927 close(mdfd);
928 return 1;
929 }
930 close(bmfd);
931 }
932
933 /* First, add the raid disks, but add the chosen one last */
934 for (i=0; i<= bestcnt; i++) {
935 int j;
936 if (i < bestcnt) {
937 j = best[i];
938 if (j == chosen_drive)
939 continue;
940 } else
941 j = chosen_drive;
942
943 if (j >= 0 /* && devices[j].uptodate */) {
944 rv = add_disk(mdfd, st, &info, &devices[j].i);
945
946 if (rv) {
947 fprintf(stderr, Name ": failed to add "
948 "%s to %s: %s\n",
949 devices[j].devname,
950 mddev,
951 strerror(errno));
952 if (i < info.array.raid_disks
953 || i == bestcnt)
954 okcnt--;
955 else
956 sparecnt--;
957 } else if (verbose > 0)
958 fprintf(stderr, Name ": added %s "
959 "to %s as %d\n",
960 devices[j].devname, mddev,
961 devices[j].i.disk.raid_disk);
962 } else if (verbose > 0 && i < info.array.raid_disks)
963 fprintf(stderr, Name ": no uptodate device for "
964 "slot %d of %s\n",
965 i, mddev);
966 }
967
968 if (info.array.level == LEVEL_CONTAINER) {
969 if (verbose >= 0) {
970 fprintf(stderr, Name ": Container %s has been "
971 "assembled with %d drive%s",
972 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
973 if (okcnt < info.array.raid_disks)
974 fprintf(stderr, " (out of %d)",
975 info.array.raid_disks);
976 fprintf(stderr, "\n");
977 }
978 close(mdfd);
979 return 0;
980 }
981
982 if (runstop == 1 ||
983 (runstop <= 0 &&
984 ( enough(info.array.level, info.array.raid_disks,
985 info.array.layout, clean, avail, okcnt) &&
986 (okcnt >= req_cnt || start_partial_ok)
987 ))) {
988 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
989 if (verbose >= 0) {
990 fprintf(stderr, Name ": %s has been started with %d drive%s",
991 mddev, okcnt, okcnt==1?"":"s");
992 if (okcnt < info.array.raid_disks)
993 fprintf(stderr, " (out of %d)", info.array.raid_disks);
994 if (sparecnt)
995 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
996 fprintf(stderr, ".\n");
997 }
998 close(mdfd);
999 if (auto_assem) {
1000 int usecs = 1;
1001 /* There is a nasty race with 'mdadm --monitor'.
1002 * If it opens this device before we close it,
1003 * it gets an incomplete open on which IO
1004 * doesn't work and the capacity is
1005 * wrong.
1006 * If we reopen (to check for layered devices)
1007 * before --monitor closes, we loose.
1008 *
1009 * So: wait upto 1 second for there to be
1010 * a non-zero capacity.
1011 */
1012 while (usecs < 1000) {
1013 mdfd = open(mddev, O_RDONLY);
1014 if (mdfd >= 0) {
1015 unsigned long long size;
1016 if (get_dev_size(mdfd, NULL, &size) &&
1017 size > 0)
1018 break;
1019 close(mdfd);
1020 }
1021 usleep(usecs);
1022 usecs <<= 1;
1023 }
1024 }
1025 return 0;
1026 }
1027 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1028 mddev, strerror(errno));
1029
1030 if (!enough(info.array.level, info.array.raid_disks,
1031 info.array.layout, 1, avail, okcnt))
1032 fprintf(stderr, Name ": Not enough devices to "
1033 "start the array.\n");
1034 else if (!enough(info.array.level,
1035 info.array.raid_disks,
1036 info.array.layout, clean,
1037 avail, okcnt))
1038 fprintf(stderr, Name ": Not enough devices to "
1039 "start the array while not clean "
1040 "- consider --force.\n");
1041
1042 if (auto_assem)
1043 ioctl(mdfd, STOP_ARRAY, NULL);
1044 close(mdfd);
1045 return 1;
1046 }
1047 if (runstop == -1) {
1048 fprintf(stderr, Name ": %s assembled from %d drive%s",
1049 mddev, okcnt, okcnt==1?"":"s");
1050 if (okcnt != info.array.raid_disks)
1051 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1052 fprintf(stderr, ", but not started.\n");
1053 close(mdfd);
1054 return 0;
1055 }
1056 if (verbose >= -1) {
1057 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1058 if (sparecnt)
1059 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1060 if (!enough(info.array.level, info.array.raid_disks,
1061 info.array.layout, 1, avail, okcnt))
1062 fprintf(stderr, " - not enough to start the array.\n");
1063 else if (!enough(info.array.level,
1064 info.array.raid_disks,
1065 info.array.layout, clean,
1066 avail, okcnt))
1067 fprintf(stderr, " - not enough to start the "
1068 "array while not clean - consider "
1069 "--force.\n");
1070 else {
1071 if (req_cnt == info.array.raid_disks)
1072 fprintf(stderr, " - need all %d to start it", req_cnt);
1073 else
1074 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1075 fprintf(stderr, " (use --run to insist).\n");
1076 }
1077 }
1078 if (auto_assem)
1079 ioctl(mdfd, STOP_ARRAY, NULL);
1080 return 1;
1081 } else {
1082 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1083 * been updated to point to the current locations of devices.
1084 * so we can just start the array
1085 */
1086 unsigned long dev;
1087 dev = makedev(devices[chosen_drive].i.disk.major,
1088 devices[chosen_drive].i.disk.minor);
1089 if (ioctl(mdfd, START_ARRAY, dev)) {
1090 fprintf(stderr, Name ": Cannot start array: %s\n",
1091 strerror(errno));
1092 }
1093
1094 }
1095 close(mdfd);
1096 return 0;
1097 }
Unnamed repository; edit this file 'description' to name the repository.