]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Preliminary -As support for container member arrays
[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, int mdfd,
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 clean = 0;
115 int must_close = 0;
116 int old_linux = 0;
117 int vers = 0; /* Keep gcc quite - it really is initialised */
118 struct {
119 char *devname;
120 int uptodate; /* set once we decide that this device is as
121 * recent as everything else in the array.
122 */
123 struct mdinfo i;
124 } *devices;
125 int *best = NULL; /* indexed by raid_disk */
126 unsigned int bestcnt = 0;
127 int devcnt = 0;
128 unsigned int okcnt, sparecnt;
129 unsigned int req_cnt;
130 unsigned int i;
131 int most_recent = 0;
132 int chosen_drive;
133 int change = 0;
134 int inargv = 0;
135 int bitmap_done;
136 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
137 unsigned int num_devs;
138 mddev_dev_t tmpdev;
139 struct mdinfo info;
140 char *avail;
141 int nextspare = 0;
142 int uuid_for_name = 0;
143
144 memset(&info, 0, sizeof(info));
145
146 if (get_linux_version() < 2004000)
147 old_linux = 1;
148
149 if (mdfd >= 0) {
150 vers = md_get_version(mdfd);
151 if (vers <= 0) {
152 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
153 return 1;
154 }
155 if (vers < 9000) {
156 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
157 " Upgrade your kernel or try --build\n");
158 return 1;
159 }
160
161 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
162 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
163 mddev);
164 return 1;
165 }
166 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
167 }
168 /*
169 * If any subdevs are listed, then any that don't
170 * match ident are discarded. Remainder must all match and
171 * become the array.
172 * If no subdevs, then we scan all devices in the config file, but
173 * there must be something in the identity
174 */
175
176 if (!devlist &&
177 ident->uuid_set == 0 &&
178 ident->super_minor < 0 &&
179 ident->devices == NULL) {
180 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
181 mddev ? mddev : "further assembly");
182 return 1;
183 }
184
185 /* if the configuration specifies a container then we use that to
186 * determine the devices and retrieve the array configuration
187 */
188 #ifndef MDASSEMBLE
189 if (ident->container && ident->member) {
190 int cfd = open(ident->container, O_RDWR);
191 struct mdinfo *mdi;
192 struct supertype container;
193
194 if (verbose>0)
195 fprintf(stderr, Name ": looking to assemble member array %s"
196 " inside container %s\n", ident->member, ident->container);
197 if (cfd < 0) {
198 if (verbose>0)
199 fprintf(stderr, Name ": unable to open container %s: %s\n",
200 ident->container, strerror(errno));
201 return 1;
202 }
203
204 mdi = sysfs_read(cfd, fd2devnum(cfd), GET_VERSION);
205 if (!mdi) {
206 close(cfd);
207 if (verbose>0)
208 fprintf(stderr, Name ": unable to read container %s\n",
209 ident->container);
210 return 1;
211 }
212 container.ss = find_metadata_methods(mdi->text_version);
213 sysfs_free(mdi);
214 if (!container.ss) {
215 close(cfd);
216 fprintf(stderr, Name ": %s uses unknown metadata: %s\n",
217 ident->container, mdi->text_version);
218 return 1;
219 }
220 if (container.ss->load_super(&container, cfd, ident->container)) {
221 fprintf(stderr, Name ": Cannot load metadata for container %s\n",
222 ident->container);
223 return 1;
224 }
225
226 return Incremental_container(&container, ident->container,
227 verbose, runstop, ident->autof);
228 }
229 #endif
230 if (devlist == NULL)
231 devlist = conf_get_devs();
232 else if (mdfd >= 0)
233 inargv = 1;
234
235 try_again:
236
237 tmpdev = devlist; num_devs = 0;
238 while (tmpdev) {
239 if (tmpdev->used)
240 tmpdev->used = 2;
241 else
242 num_devs++;
243 tmpdev = tmpdev->next;
244 }
245 devices = malloc(num_devs * sizeof(*devices));
246
247 if (!st && ident->st) st = ident->st;
248
249 if (verbose>0)
250 fprintf(stderr, Name ": looking for devices for %s\n",
251 mddev ? mddev : "further assembly");
252
253 /* first walk the list of devices to find a consistent set
254 * that match the criterea, if that is possible.
255 * We flag the one we like with 'used'.
256 */
257 for (tmpdev = devlist;
258 tmpdev;
259 tmpdev = tmpdev->next) {
260 char *devname = tmpdev->devname;
261 int dfd;
262 struct stat stb;
263 struct supertype *tst = dup_super(st);
264
265 if (tmpdev->used > 1) continue;
266
267 if (ident->devices &&
268 !match_oneof(ident->devices, devname)) {
269 if ((inargv && verbose>=0) || verbose > 0)
270 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
271 continue;
272 }
273
274 dfd = dev_open(devname, O_RDONLY|O_EXCL);
275 if (dfd < 0) {
276 if ((inargv && verbose >= 0) || verbose > 0)
277 fprintf(stderr, Name ": cannot open device %s: %s\n",
278 devname, strerror(errno));
279 tmpdev->used = 2;
280 } else if (fstat(dfd, &stb)< 0) {
281 /* Impossible! */
282 fprintf(stderr, Name ": fstat failed for %s: %s\n",
283 devname, strerror(errno));
284 tmpdev->used = 2;
285 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
286 fprintf(stderr, Name ": %s is not a block device.\n",
287 devname);
288 tmpdev->used = 2;
289 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
290 if ((inargv && verbose >= 0) || verbose > 0)
291 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
292 devname);
293 tmpdev->used = 2;
294 } else if (tst->ss->load_super(tst,dfd, NULL)) {
295 if ((inargv && verbose >= 0) || verbose > 0)
296 fprintf( stderr, Name ": no RAID superblock on %s\n",
297 devname);
298 } else {
299 tst->ss->getinfo_super(tst, &info);
300 }
301 if (dfd >= 0) close(dfd);
302
303 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
304 (!tst || !tst->sb ||
305 same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
306 if ((inargv && verbose >= 0) || verbose > 0)
307 fprintf(stderr, Name ": %s has wrong uuid.\n",
308 devname);
309 goto loop;
310 }
311 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
312 (!tst || !tst->sb ||
313 name_matches(info.name, ident->name, homehost)==0)) {
314 if ((inargv && verbose >= 0) || verbose > 0)
315 fprintf(stderr, Name ": %s has wrong name.\n",
316 devname);
317 goto loop;
318 }
319 if (ident->super_minor != UnSet &&
320 (!tst || !tst->sb ||
321 ident->super_minor != info.array.md_minor)) {
322 if ((inargv && verbose >= 0) || verbose > 0)
323 fprintf(stderr, Name ": %s has wrong super-minor.\n",
324 devname);
325 goto loop;
326 }
327 if (ident->level != UnSet &&
328 (!tst || !tst->sb ||
329 ident->level != info.array.level)) {
330 if ((inargv && verbose >= 0) || verbose > 0)
331 fprintf(stderr, Name ": %s has wrong raid level.\n",
332 devname);
333 goto loop;
334 }
335 if (ident->raid_disks != UnSet &&
336 (!tst || !tst->sb ||
337 ident->raid_disks!= info.array.raid_disks)) {
338 if ((inargv && verbose >= 0) || verbose > 0)
339 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
340 devname);
341 goto loop;
342 }
343 if (mdfd < 0) {
344 if (tst == NULL || tst->sb == NULL)
345 continue;
346 switch(tst->ss->match_home(tst, homehost))
347 {
348 case 1: /* happy with match. */
349 break;
350 case -1: /* cannot match */
351 uuid_for_name = 1;
352 break;
353 case 0: /* Doesn't match */
354 if (update)
355 /* We are changing the name*/
356 break;
357 if ((inargv && verbose >= 0) || verbose > 0)
358 fprintf(stderr, Name ": %s is not built for "
359 "host %s - using UUID for "
360 "device name.\n",
361 devname, homehost);
362
363 /* Auto-assemble, and this is not a usable host */
364 /* if update != NULL, we are updating the host
365 * name... */
366 uuid_for_name = 1;
367 break;
368 }
369 }
370 /* If we are this far, then we are nearly commited to this device.
371 * If the super_block doesn't exist, or doesn't match others,
372 * then we probably cannot continue
373 * However if one of the arrays is for the homehost, and
374 * the other isn't that can disambiguate.
375 */
376
377 if (!tst || !tst->sb) {
378 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
379 devname);
380 if (st)
381 st->ss->free_super(st);
382 return 1;
383 }
384
385 if (st == NULL)
386 st = dup_super(tst);
387 if (st->minor_version == -1)
388 st->minor_version = tst->minor_version;
389 if (st->ss != tst->ss ||
390 st->minor_version != tst->minor_version ||
391 st->ss->compare_super(st, tst) != 0) {
392 /* Some mismatch. If exactly one array matches this host,
393 * we can resolve on that one.
394 * Or, if we are auto assembling, we just ignore the second
395 * for now.
396 */
397 if (mdfd < 0)
398 goto loop;
399 if (homehost) {
400 int first = st->ss->match_home(st, homehost);
401 int last = tst->ss->match_home(tst, homehost);
402 if (first != last &&
403 (first == 1 || last == 1)) {
404 /* We can do something */
405 if (first) {/* just ignore this one */
406 if ((inargv && verbose >= 0) || verbose > 0)
407 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
408 devname);
409 goto loop;
410 } else { /* reject all those sofar */
411 mddev_dev_t td;
412 if ((inargv && verbose >= 0) || verbose > 0)
413 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
414 devname);
415 for (td=devlist; td != tmpdev; td=td->next)
416 if (td->used == 1)
417 td->used = 0;
418 tmpdev->used = 1;
419 goto loop;
420 }
421 }
422 }
423 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
424 devname);
425 tst->ss->free_super(tst);
426 st->ss->free_super(st);
427 return 1;
428 }
429
430 tmpdev->used = 1;
431
432 loop:
433 if (tst)
434 tst->ss->free_super(tst);
435 }
436
437 if (mdfd < 0) {
438 /* So... it is up to me to open the device.
439 * We create a name '/dev/md/XXX' based on the info in the
440 * superblock, and call open_mddev on that
441 */
442 mdu_array_info_t inf;
443 char *c;
444 char nbuf[64];
445 int rc;
446
447 if (!st || !st->sb) {
448 return 2;
449 }
450 st->ss->getinfo_super(st, &info);
451 if (uuid_for_name)
452 c = fname_from_uuid(st, &info, nbuf, '-');
453 else {
454 c = strchr(info.name, ':');
455 if (c) c++; else c= info.name;
456 }
457 if (isdigit(*c) && ((ident->autof & 7)==4 || (ident->autof&7)==6))
458 /* /dev/md/d0 style for partitionable */
459 rc = asprintf(&mddev, "/dev/md/d%s", c);
460 else
461 rc = asprintf(&mddev, "/dev/md/%s", c);
462 if (rc < 0)
463 mdfd = -1;
464 else
465 mdfd = open_mddev(mddev, ident->autof);
466 if (mdfd < 0) {
467 st->ss->free_super(st);
468 free(devices);
469 goto try_again;
470 }
471 vers = md_get_version(mdfd);
472 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
473 for (tmpdev = devlist ;
474 tmpdev && tmpdev->used != 1;
475 tmpdev = tmpdev->next)
476 ;
477 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
478 if (tmpdev)
479 fprintf(stderr, Name ": %s needed for %s...\n",
480 mddev, tmpdev->devname);
481 close(mdfd);
482 mdfd = -1;
483 st->ss->free_super(st);
484 free(devices);
485 goto try_again;
486 }
487 must_close = 1;
488 }
489
490 /* Ok, no bad inconsistancy, we can try updating etc */
491 bitmap_done = 0;
492 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
493 char *devname = tmpdev->devname;
494 struct stat stb;
495 /* looks like a good enough match to update the super block if needed */
496 #ifndef MDASSEMBLE
497 if (update) {
498 int dfd;
499 /* prepare useful information in info structures */
500 struct stat stb2;
501 struct supertype *tst;
502 fstat(mdfd, &stb2);
503
504 if (strcmp(update, "uuid")==0 &&
505 !ident->uuid_set) {
506 int rfd;
507 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
508 read(rfd, ident->uuid, 16) != 16) {
509 *(__u32*)(ident->uuid) = random();
510 *(__u32*)(ident->uuid+1) = random();
511 *(__u32*)(ident->uuid+2) = random();
512 *(__u32*)(ident->uuid+3) = random();
513 }
514 if (rfd >= 0) close(rfd);
515 }
516 dfd = dev_open(devname, O_RDWR|O_EXCL);
517
518 remove_partitions(dfd);
519
520 tst = dup_super(st);
521 tst->ss->load_super(tst, dfd, NULL);
522 tst->ss->getinfo_super(tst, &info);
523
524 memcpy(info.uuid, ident->uuid, 16);
525 strcpy(info.name, ident->name);
526 info.array.md_minor = minor(stb2.st_rdev);
527
528 tst->ss->update_super(tst, &info, update,
529 devname, verbose,
530 ident->uuid_set, homehost);
531 if (strcmp(update, "uuid")==0 &&
532 !ident->uuid_set) {
533 ident->uuid_set = 1;
534 memcpy(ident->uuid, info.uuid, 16);
535 }
536 if (dfd < 0)
537 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
538 devname);
539 else if (tst->ss->store_super(tst, dfd))
540 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
541 devname);
542 if (dfd >= 0)
543 close(dfd);
544
545 if (strcmp(update, "uuid")==0 &&
546 ident->bitmap_fd >= 0 && !bitmap_done) {
547 if (bitmap_update_uuid(ident->bitmap_fd,
548 info.uuid,
549 tst->ss->swapuuid) != 0)
550 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
551 else
552 bitmap_done = 1;
553 }
554 tst->ss->free_super(tst);
555 } else
556 #endif
557 {
558 struct supertype *tst = dup_super(st);
559 int dfd;
560 dfd = dev_open(devname, O_RDWR|O_EXCL);
561
562 remove_partitions(dfd);
563
564 tst->ss->load_super(tst, dfd, NULL);
565 tst->ss->getinfo_super(tst, &info);
566 tst->ss->free_super(tst);
567 close(dfd);
568 }
569
570 stat(devname, &stb);
571
572 if (verbose > 0)
573 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
574 devname, mddev, info.disk.raid_disk);
575 devices[devcnt].devname = devname;
576 devices[devcnt].uptodate = 0;
577 devices[devcnt].i = info;
578 devices[devcnt].i.disk.major = major(stb.st_rdev);
579 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
580 if (most_recent < devcnt) {
581 if (devices[devcnt].i.events
582 > devices[most_recent].i.events)
583 most_recent = devcnt;
584 }
585 if (info.array.level == -4)
586 /* with multipath, the raid_disk from the superblock is meaningless */
587 i = devcnt;
588 else
589 i = devices[devcnt].i.disk.raid_disk;
590 if (i+1 == 0) {
591 if (nextspare < info.array.raid_disks)
592 nextspare = info.array.raid_disks;
593 i = nextspare++;
594 } else {
595 if (i >= info.array.raid_disks &&
596 i >= nextspare)
597 nextspare = i+1;
598 }
599 if (i < 10000) {
600 if (i >= bestcnt) {
601 unsigned int newbestcnt = i+10;
602 int *newbest = malloc(sizeof(int)*newbestcnt);
603 unsigned int c;
604 for (c=0; c < newbestcnt; c++)
605 if (c < bestcnt)
606 newbest[c] = best[c];
607 else
608 newbest[c] = -1;
609 if (best)free(best);
610 best = newbest;
611 bestcnt = newbestcnt;
612 }
613 if (best[i] >=0 &&
614 devices[best[i]].i.events
615 == devices[devcnt].i.events
616 && (devices[best[i]].i.disk.minor
617 != devices[devcnt].i.disk.minor)
618 && st->ss == &super0
619 && info.array.level != LEVEL_MULTIPATH) {
620 /* two different devices with identical superblock.
621 * Could be a mis-detection caused by overlapping
622 * partitions. fail-safe.
623 */
624 fprintf(stderr, Name ": WARNING %s and %s appear"
625 " to have very similar superblocks.\n"
626 " If they are really different, "
627 "please --zero the superblock on one\n"
628 " If they are the same or overlap,"
629 " please remove one from %s.\n",
630 devices[best[i]].devname, devname,
631 inargv ? "the list" :
632 "the\n DEVICE list in mdadm.conf"
633 );
634 if (must_close) close(mdfd);
635 return 1;
636 }
637 if (best[i] == -1
638 || (devices[best[i]].i.events
639 < devices[devcnt].i.events))
640 best[i] = devcnt;
641 }
642 devcnt++;
643 }
644
645 if (devcnt == 0) {
646 fprintf(stderr, Name ": no devices found for %s\n",
647 mddev);
648 if (st)
649 st->ss->free_super(st);
650 if (must_close) close(mdfd);
651 return 1;
652 }
653
654 if (update && strcmp(update, "byteorder")==0)
655 st->minor_version = 90;
656
657 st->ss->getinfo_super(st, &info);
658 clean = info.array.state & 1;
659
660 /* now we have some devices that might be suitable.
661 * I wonder how many
662 */
663 avail = malloc(info.array.raid_disks);
664 memset(avail, 0, info.array.raid_disks);
665 okcnt = 0;
666 sparecnt=0;
667 for (i=0; i< bestcnt ;i++) {
668 int j = best[i];
669 int event_margin = 1; /* always allow a difference of '1'
670 * like the kernel does
671 */
672 if (j < 0) continue;
673 /* note: we ignore error flags in multipath arrays
674 * as they don't make sense
675 */
676 if (info.array.level != -4)
677 if (!(devices[j].i.disk.state & (1<<MD_DISK_SYNC))) {
678 if (!(devices[j].i.disk.state
679 & (1<<MD_DISK_FAULTY)))
680 sparecnt++;
681 continue;
682 }
683 if (devices[j].i.events+event_margin >=
684 devices[most_recent].i.events) {
685 devices[j].uptodate = 1;
686 if (i < info.array.raid_disks) {
687 okcnt++;
688 avail[i]=1;
689 } else
690 sparecnt++;
691 }
692 }
693 while (force && !enough(info.array.level, info.array.raid_disks,
694 info.array.layout, 1,
695 avail, okcnt)) {
696 /* Choose the newest best drive which is
697 * not up-to-date, update the superblock
698 * and add it.
699 */
700 int fd;
701 struct supertype *tst;
702 long long current_events;
703 chosen_drive = -1;
704 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
705 int j = best[i];
706 if (j>=0 &&
707 !devices[j].uptodate &&
708 devices[j].i.events > 0 &&
709 (chosen_drive < 0 ||
710 devices[j].i.events
711 > devices[chosen_drive].i.events))
712 chosen_drive = j;
713 }
714 if (chosen_drive < 0)
715 break;
716 current_events = devices[chosen_drive].i.events;
717 add_another:
718 if (verbose >= 0)
719 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
720 devices[chosen_drive].devname,
721 devices[chosen_drive].i.disk.raid_disk,
722 (int)(devices[chosen_drive].i.events),
723 (int)(devices[most_recent].i.events));
724 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
725 if (fd < 0) {
726 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
727 devices[chosen_drive].devname);
728 devices[chosen_drive].i.events = 0;
729 continue;
730 }
731 tst = dup_super(st);
732 if (tst->ss->load_super(tst,fd, NULL)) {
733 close(fd);
734 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
735 devices[chosen_drive].devname);
736 devices[chosen_drive].i.events = 0;
737 continue;
738 }
739 info.events = devices[most_recent].i.events;
740 tst->ss->update_super(tst, &info, "force-one",
741 devices[chosen_drive].devname, verbose,
742 0, NULL);
743
744 if (tst->ss->store_super(tst, fd)) {
745 close(fd);
746 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
747 devices[chosen_drive].devname);
748 devices[chosen_drive].i.events = 0;
749 tst->ss->free_super(tst);
750 continue;
751 }
752 close(fd);
753 devices[chosen_drive].i.events = devices[most_recent].i.events;
754 devices[chosen_drive].uptodate = 1;
755 avail[chosen_drive] = 1;
756 okcnt++;
757 tst->ss->free_super(tst);
758
759 /* If there are any other drives of the same vintage,
760 * add them in as well. We can't lose and we might gain
761 */
762 for (i=0; i<info.array.raid_disks && i < bestcnt ; i++) {
763 int j = best[i];
764 if (j >= 0 &&
765 !devices[j].uptodate &&
766 devices[j].i.events > 0 &&
767 devices[j].i.events == current_events) {
768 chosen_drive = j;
769 goto add_another;
770 }
771 }
772 }
773
774 /* Now we want to look at the superblock which the kernel will base things on
775 * and compare the devices that we think are working with the devices that the
776 * superblock thinks are working.
777 * If there are differences and --force is given, then update this chosen
778 * superblock.
779 */
780 chosen_drive = -1;
781 st->ss->free_super(st);
782 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
783 int j = best[i];
784 int fd;
785
786 if (j<0)
787 continue;
788 if (!devices[j].uptodate)
789 continue;
790 chosen_drive = j;
791 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
792 fprintf(stderr, Name ": Cannot open %s: %s\n",
793 devices[j].devname, strerror(errno));
794 if (must_close) close(mdfd);
795 return 1;
796 }
797 if (st->ss->load_super(st,fd, NULL)) {
798 close(fd);
799 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
800 devices[j].devname);
801 if (must_close) close(mdfd);
802 return 1;
803 }
804 close(fd);
805 }
806 if (st->sb == NULL) {
807 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
808 if (must_close) close(mdfd);
809 return 1;
810 }
811 st->ss->getinfo_super(st, &info);
812 #ifndef MDASSEMBLE
813 sysfs_init(&info, mdfd, 0);
814 #endif
815 for (i=0; i<bestcnt; i++) {
816 int j = best[i];
817 unsigned int desired_state;
818
819 if (i < info.array.raid_disks)
820 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
821 else
822 desired_state = 0;
823
824 if (j<0)
825 continue;
826 if (!devices[j].uptodate)
827 continue;
828
829 devices[j].i.disk.state = desired_state;
830
831 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
832 verbose, 0, NULL)) {
833 if (force) {
834 if (verbose >= 0)
835 fprintf(stderr, Name ": "
836 "clearing FAULTY flag for device %d in %s for %s\n",
837 j, mddev, devices[j].devname);
838 change = 1;
839 } else {
840 if (verbose >= -1)
841 fprintf(stderr, Name ": "
842 "device %d in %s has wrong state in superblock, but %s seems ok\n",
843 i, mddev, devices[j].devname);
844 }
845 }
846 #if 0
847 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
848 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
849 i, mddev);
850 }
851 #endif
852 }
853 if (force && !clean &&
854 !enough(info.array.level, info.array.raid_disks,
855 info.array.layout, clean,
856 avail, okcnt)) {
857 change += st->ss->update_super(st, &info, "force-array",
858 devices[chosen_drive].devname, verbose,
859 0, NULL);
860 clean = 1;
861 }
862
863 if (change) {
864 int fd;
865 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
866 if (fd < 0) {
867 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
868 devices[chosen_drive].devname);
869 if (must_close) close(mdfd);
870 return 1;
871 }
872 if (st->ss->store_super(st, fd)) {
873 close(fd);
874 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
875 devices[chosen_drive].devname);
876 if (must_close) close(mdfd);
877 return 1;
878 }
879 close(fd);
880 }
881
882 /* If we are in the middle of a reshape we may need to restore saved data
883 * that was moved aside due to the reshape overwriting live data
884 * The code of doing this lives in Grow.c
885 */
886 #ifndef MDASSEMBLE
887 if (info.reshape_active) {
888 int err = 0;
889 int *fdlist = malloc(sizeof(int)* bestcnt);
890 for (i=0; i<bestcnt; i++) {
891 int j = best[i];
892 if (j >= 0) {
893 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
894 if (fdlist[i] < 0) {
895 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
896 devices[j].devname);
897 err = 1;
898 break;
899 }
900 } else
901 fdlist[i] = -1;
902 }
903 if (!err)
904 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
905 while (i>0) {
906 i--;
907 if (fdlist[i]>=0) close(fdlist[i]);
908 }
909 if (err) {
910 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
911 if (must_close) close(mdfd);
912 return err;
913 }
914 }
915 #endif
916 /* count number of in-sync devices according to the superblock.
917 * We must have this number to start the array without -s or -R
918 */
919 req_cnt = info.array.working_disks;
920
921 /* Almost ready to actually *do* something */
922 if (!old_linux) {
923 int rv;
924
925 rv = set_array_info(mdfd, st, &info);
926 if (rv) {
927 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
928 mddev, strerror(errno));
929 if (must_close) close(mdfd);
930 return 1;
931 }
932 if (ident->bitmap_fd >= 0) {
933 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
934 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
935 if (must_close) close(mdfd);
936 return 1;
937 }
938 } else if (ident->bitmap_file) {
939 /* From config file */
940 int bmfd = open(ident->bitmap_file, O_RDWR);
941 if (bmfd < 0) {
942 fprintf(stderr, Name ": Could not open bitmap file %s\n",
943 ident->bitmap_file);
944 if (must_close) close(mdfd);
945 return 1;
946 }
947 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
948 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
949 close(bmfd);
950 if (must_close) close(mdfd);
951 return 1;
952 }
953 close(bmfd);
954 }
955
956 /* First, add the raid disks, but add the chosen one last */
957 for (i=0; i<= bestcnt; i++) {
958 int j;
959 if (i < bestcnt) {
960 j = best[i];
961 if (j == chosen_drive)
962 continue;
963 } else
964 j = chosen_drive;
965
966 if (j >= 0 /* && devices[j].uptodate */) {
967 rv = add_disk(mdfd, st, &info, &devices[j].i);
968
969 if (rv) {
970 fprintf(stderr, Name ": failed to add "
971 "%s to %s: %s\n",
972 devices[j].devname,
973 mddev,
974 strerror(errno));
975 if (i < info.array.raid_disks
976 || i == bestcnt)
977 okcnt--;
978 else
979 sparecnt--;
980 } else if (verbose > 0)
981 fprintf(stderr, Name ": added %s "
982 "to %s as %d\n",
983 devices[j].devname, mddev,
984 devices[j].i.disk.raid_disk);
985 } else if (verbose > 0 && i < info.array.raid_disks)
986 fprintf(stderr, Name ": no uptodate device for "
987 "slot %d of %s\n",
988 i, mddev);
989 }
990
991 if (info.array.level == LEVEL_CONTAINER) {
992 if (verbose >= 0) {
993 fprintf(stderr, Name ": Container %s has been "
994 "assembled with %d drive%s",
995 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
996 if (okcnt < info.array.raid_disks)
997 fprintf(stderr, " (out of %d)",
998 info.array.raid_disks);
999 fprintf(stderr, "\n");
1000 }
1001 if (must_close)
1002 close(mdfd);
1003 return 0;
1004 }
1005
1006 if (runstop == 1 ||
1007 (runstop <= 0 &&
1008 ( enough(info.array.level, info.array.raid_disks,
1009 info.array.layout, clean, avail, okcnt) &&
1010 (okcnt >= req_cnt || start_partial_ok)
1011 ))) {
1012 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
1013 if (verbose >= 0) {
1014 fprintf(stderr, Name ": %s has been started with %d drive%s",
1015 mddev, okcnt, okcnt==1?"":"s");
1016 if (okcnt < info.array.raid_disks)
1017 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1018 if (sparecnt)
1019 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1020 fprintf(stderr, ".\n");
1021 }
1022 if (must_close) {
1023 int usecs = 1;
1024 close(mdfd);
1025 /* There is a nasty race with 'mdadm --monitor'.
1026 * If it opens this device before we close it,
1027 * it gets an incomplete open on which IO
1028 * doesn't work and the capacity is
1029 * wrong.
1030 * If we reopen (to check for layered devices)
1031 * before --monitor closes, we loose.
1032 *
1033 * So: wait upto 1 second for there to be
1034 * a non-zero capacity.
1035 */
1036 while (usecs < 1000) {
1037 mdfd = open(mddev, O_RDONLY);
1038 if (mdfd >= 0) {
1039 unsigned long long size;
1040 if (get_dev_size(mdfd, NULL, &size) &&
1041 size > 0)
1042 break;
1043 close(mdfd);
1044 }
1045 usleep(usecs);
1046 usecs <<= 1;
1047 }
1048 }
1049 return 0;
1050 }
1051 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1052 mddev, strerror(errno));
1053
1054 if (!enough(info.array.level, info.array.raid_disks,
1055 info.array.layout, 1, avail, okcnt))
1056 fprintf(stderr, Name ": Not enough devices to "
1057 "start the array.\n");
1058 else if (!enough(info.array.level,
1059 info.array.raid_disks,
1060 info.array.layout, clean,
1061 avail, okcnt))
1062 fprintf(stderr, Name ": Not enough devices to "
1063 "start the array while not clean "
1064 "- consider --force.\n");
1065
1066 if (must_close) {
1067 ioctl(mdfd, STOP_ARRAY, NULL);
1068 close(mdfd);
1069 }
1070 return 1;
1071 }
1072 if (runstop == -1) {
1073 fprintf(stderr, Name ": %s assembled from %d drive%s",
1074 mddev, okcnt, okcnt==1?"":"s");
1075 if (okcnt != info.array.raid_disks)
1076 fprintf(stderr, " (out of %d)", info.array.raid_disks);
1077 fprintf(stderr, ", but not started.\n");
1078 if (must_close) close(mdfd);
1079 return 0;
1080 }
1081 if (verbose >= -1) {
1082 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1083 if (sparecnt)
1084 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1085 if (!enough(info.array.level, info.array.raid_disks,
1086 info.array.layout, 1, avail, okcnt))
1087 fprintf(stderr, " - not enough to start the array.\n");
1088 else if (!enough(info.array.level,
1089 info.array.raid_disks,
1090 info.array.layout, clean,
1091 avail, okcnt))
1092 fprintf(stderr, " - not enough to start the "
1093 "array while not clean - consider "
1094 "--force.\n");
1095 else {
1096 if (req_cnt == info.array.raid_disks)
1097 fprintf(stderr, " - need all %d to start it", req_cnt);
1098 else
1099 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
1100 fprintf(stderr, " (use --run to insist).\n");
1101 }
1102 }
1103 if (must_close) {
1104 ioctl(mdfd, STOP_ARRAY, NULL);
1105 close(mdfd);
1106 }
1107 return 1;
1108 } else {
1109 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1110 * been updated to point to the current locations of devices.
1111 * so we can just start the array
1112 */
1113 unsigned long dev;
1114 dev = makedev(devices[chosen_drive].i.disk.major,
1115 devices[chosen_drive].i.disk.minor);
1116 if (ioctl(mdfd, START_ARRAY, dev)) {
1117 fprintf(stderr, Name ": Cannot start array: %s\n",
1118 strerror(errno));
1119 }
1120
1121 }
1122 if (must_close) close(mdfd);
1123 return 0;
1124 }