]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
Support --auto-update-homehost
[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
32 static int name_matches(char *found, char *required, char *homehost)
33 {
34 /* See if the name found matches the required name, possibly
35 * prefixed with 'homehost'
36 */
37 char fnd[33];
38
39 strncpy(fnd, found, 32);
40 fnd[32] = 0;
41 if (strcmp(found, required)==0)
42 return 1;
43 if (homehost) {
44 int l = strlen(homehost);
45 if (l < 32 && fnd[l] == ':' &&
46 strcmp(fnd+l+1, required)==0)
47 return 1;
48 }
49 return 0;
50 }
51
52 int Assemble(struct supertype *st, char *mddev, int mdfd,
53 mddev_ident_t ident, char *conffile,
54 mddev_dev_t devlist, char *backup_file,
55 int readonly, int runstop,
56 char *update, char *homehost,
57 int verbose, int force)
58 {
59 /*
60 * The task of Assemble is to find a collection of
61 * devices that should (according to their superblocks)
62 * form an array, and to give this collection to the MD driver.
63 * In Linux-2.4 and later, this involves submitting a
64 * SET_ARRAY_INFO ioctl with no arg - to prepare
65 * the array - and then submit a number of
66 * ADD_NEW_DISK ioctls to add disks into
67 * the array. Finally RUN_ARRAY might
68 * be submitted to start the array.
69 *
70 * Much of the work of Assemble is in finding and/or
71 * checking the disks to make sure they look right.
72 *
73 * If mddev is not set, then scan must be set and we
74 * read through the config file for dev+uuid mapping
75 * We recurse, setting mddev, for each device that
76 * - isn't running
77 * - has a valid uuid (or any uuid if !uuidset)
78 *
79 * If mddev is set, we try to determine state of md.
80 * check version - must be at least 0.90.0
81 * check kernel version. must be at least 2.4.
82 * If not, we can possibly fall back on START_ARRAY
83 * Try to GET_ARRAY_INFO.
84 * If possible, give up
85 * If not, try to STOP_ARRAY just to make sure
86 *
87 * If !uuidset and scan, look in conf-file for uuid
88 * If not found, give up
89 * If !devlist and scan and uuidset, get list of devs from conf-file
90 *
91 * For each device:
92 * Check superblock - discard if bad
93 * Check uuid (set if we don't have one) - discard if no match
94 * Check superblock similarity if we have a superblock - discard if different
95 * Record events, devicenum
96 * This should give us a list of devices for the array
97 * We should collect the most recent event number
98 *
99 * Count disks with recent enough event count
100 * While force && !enough disks
101 * Choose newest rejected disks, update event count
102 * mark clean and rewrite superblock
103 * If recent kernel:
104 * SET_ARRAY_INFO
105 * foreach device with recent events : ADD_NEW_DISK
106 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
107 * If old kernel:
108 * Check the device numbers in superblock are right
109 * update superblock if any changes
110 * START_ARRAY
111 *
112 */
113 int must_close = 0;
114 int old_linux = 0;
115 int vers;
116 void *first_super = NULL, *super = NULL;
117 struct {
118 char *devname;
119 unsigned int major, minor;
120 unsigned int oldmajor, oldminor;
121 long long events;
122 int uptodate;
123 int state;
124 int raid_disk;
125 int disk_nr;
126 } *devices;
127 int *best = NULL; /* indexed by raid_disk */
128 unsigned int bestcnt = 0;
129 int devcnt = 0;
130 unsigned int okcnt, sparecnt;
131 unsigned int req_cnt;
132 unsigned int i;
133 int most_recent = 0;
134 int chosen_drive;
135 int change = 0;
136 int inargv = 0;
137 int start_partial_ok = (runstop >= 0) && (force || devlist==NULL || mdfd < 0);
138 unsigned int num_devs;
139 mddev_dev_t tmpdev;
140 struct mdinfo info;
141 char *avail;
142 int nextspare = 0;
143
144 if (get_linux_version() < 2004000)
145 old_linux = 1;
146
147 if (mdfd >= 0) {
148 vers = md_get_version(mdfd);
149 if (vers <= 0) {
150 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
151 return 1;
152 }
153 if (vers < 9000) {
154 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
155 " Upgrade your kernel or try --build\n");
156 return 1;
157 }
158
159 if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
160 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
161 mddev);
162 return 1;
163 }
164 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
165 }
166 /*
167 * If any subdevs are listed, then any that don't
168 * match ident are discarded. Remainder must all match and
169 * become the array.
170 * If no subdevs, then we scan all devices in the config file, but
171 * there must be something in the identity
172 */
173
174 if (!devlist &&
175 ident->uuid_set == 0 &&
176 ident->super_minor < 0 &&
177 ident->devices == NULL) {
178 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
179 mddev ? mddev : "further assembly");
180 return 1;
181 }
182 if (devlist == NULL)
183 devlist = conf_get_devs(conffile);
184 else if (mdfd >= 0)
185 inargv = 1;
186
187 tmpdev = devlist; num_devs = 0;
188 while (tmpdev) {
189 if (tmpdev->used)
190 tmpdev->used = 2;
191 else
192 num_devs++;
193 tmpdev = tmpdev->next;
194 }
195 devices = malloc(num_devs * sizeof(*devices));
196
197 if (!st && ident->st) st = ident->st;
198
199 if (verbose>0)
200 fprintf(stderr, Name ": looking for devices for %s\n",
201 mddev ? mddev : "further assembly");
202
203 /* first walk the list of devices to find a consistent set
204 * that match the criterea, if that is possible.
205 * We flag the one we like with 'used'.
206 */
207 for (tmpdev = devlist;
208 tmpdev;
209 tmpdev = tmpdev->next) {
210 char *devname = tmpdev->devname;
211 int dfd;
212 struct stat stb;
213 struct supertype *tst = st;
214
215 if (tmpdev->used > 1) continue;
216
217 if (ident->devices &&
218 !match_oneof(ident->devices, devname)) {
219 if ((inargv && verbose>=0) || verbose > 0)
220 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
221 continue;
222 }
223
224 if (super) {
225 free(super);
226 super = NULL;
227 }
228
229 dfd = dev_open(devname, O_RDONLY|O_EXCL);
230 if (dfd < 0) {
231 if ((inargv && verbose >= 0) || verbose > 0)
232 fprintf(stderr, Name ": cannot open device %s: %s\n",
233 devname, strerror(errno));
234 tmpdev->used = 2;
235 } else if (fstat(dfd, &stb)< 0) {
236 /* Impossible! */
237 fprintf(stderr, Name ": fstat failed for %s: %s\n",
238 devname, strerror(errno));
239 tmpdev->used = 2;
240 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
241 fprintf(stderr, Name ": %s is not a block device.\n",
242 devname);
243 tmpdev->used = 2;
244 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
245 if ((inargv && verbose >= 0) || verbose > 0)
246 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
247 devname);
248 tmpdev->used = 2;
249 } else if (tst->ss->load_super(tst,dfd, &super, NULL)) {
250 if ((inargv && verbose >= 0) || verbose > 0)
251 fprintf( stderr, Name ": no RAID superblock on %s\n",
252 devname);
253 } else {
254 tst->ss->getinfo_super(&info, super);
255 }
256 if (dfd >= 0) close(dfd);
257
258 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
259 (!super || same_uuid(info.uuid, ident->uuid, tst->ss->swapuuid)==0)) {
260 if ((inargv && verbose >= 0) || verbose > 0)
261 fprintf(stderr, Name ": %s has wrong uuid.\n",
262 devname);
263 continue;
264 }
265 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
266 (!super || name_matches(info.name, ident->name, homehost)==0)) {
267 if ((inargv && verbose >= 0) || verbose > 0)
268 fprintf(stderr, Name ": %s has wrong name.\n",
269 devname);
270 continue;
271 }
272 if (ident->super_minor != UnSet &&
273 (!super || ident->super_minor != info.array.md_minor)) {
274 if ((inargv && verbose >= 0) || verbose > 0)
275 fprintf(stderr, Name ": %s has wrong super-minor.\n",
276 devname);
277 continue;
278 }
279 if (ident->level != UnSet &&
280 (!super|| ident->level != info.array.level)) {
281 if ((inargv && verbose >= 0) || verbose > 0)
282 fprintf(stderr, Name ": %s has wrong raid level.\n",
283 devname);
284 continue;
285 }
286 if (ident->raid_disks != UnSet &&
287 (!super || ident->raid_disks!= info.array.raid_disks)) {
288 if ((inargv && verbose >= 0) || verbose > 0)
289 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
290 devname);
291 continue;
292 }
293 if (mdfd < 0) {
294 if (tst == NULL || super == NULL)
295 continue;
296 if (update == NULL &&
297 tst->ss->match_home(super, homehost)==0) {
298 if ((inargv && verbose >= 0) || verbose > 0)
299 fprintf(stderr, Name ": %s is not built for host %s.\n",
300 devname, homehost);
301 /* Auto-assemble, and this is not a usable host */
302 /* if update != NULL, we are updating the host
303 * name... */
304 continue;
305 }
306 }
307 /* If we are this far, then we are nearly commited to this device.
308 * If the super_block doesn't exist, or doesn't match others,
309 * then we probably cannot continue
310 * However if one of the arrays is for the homehost, and
311 * the other isn't that can disambiguate.
312 */
313
314 if (!super) {
315 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
316 devname);
317 free(first_super);
318 return 1;
319 }
320
321 if (st == NULL)
322 st = tst;
323 if (st->ss != tst->ss ||
324 st->minor_version != tst->minor_version ||
325 st->ss->compare_super(&first_super, super) != 0) {
326 /* Some mismatch. If exactly one array matches this host,
327 * we can resolve on that one.
328 * Or, if we are auto assembling, we just ignore the second
329 * for now.
330 */
331 if (mdfd < 0)
332 continue;
333 if (homehost) {
334 int first = st->ss->match_home(first_super, homehost);
335 int last = tst->ss->match_home(super, homehost);
336 if (first+last == 1) {
337 /* We can do something */
338 if (first) {/* just ignore this one */
339 if ((inargv && verbose >= 0) || verbose > 0)
340 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
341 devname);
342 continue;
343 } else { /* reject all those sofar */
344 mddev_dev_t td;
345 if ((inargv && verbose >= 0) || verbose > 0)
346 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
347 devname);
348 for (td=devlist; td != tmpdev; td=td->next)
349 if (td->used == 1)
350 td->used = 0;
351 tmpdev->used = 1;
352 continue;
353 }
354 }
355 }
356 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
357 devname);
358 free(super);
359 free(first_super);
360 return 1;
361 }
362
363 tmpdev->used = 1;
364 }
365
366 if (mdfd < 0) {
367 /* So... it is up to me to open the device.
368 * We create a name '/dev/md/XXX' based on the info in the
369 * superblock, and call open_mddev on that
370 */
371 mdu_array_info_t inf;
372 char *c;
373 if (!first_super) {
374 return 2;
375 }
376 st->ss->getinfo_super(&info, first_super);
377 c = strchr(info.name, ':');
378 if (c) c++; else c= info.name;
379 asprintf(&mddev, "/dev/md/%s", c);
380 mdfd = open_mddev(mddev, ident->autof);
381 if (mdfd < 0)
382 return mdfd;
383 vers = md_get_version(mdfd);
384 if (ioctl(mdfd, GET_ARRAY_INFO, &inf)==0) {
385 fprintf(stderr, Name ": %s already active, cannot restart it!\n", mddev);
386 close(mdfd);
387 free(first_super);
388 return 1;
389 }
390 must_close = 1;
391 }
392
393 /* Ok, no bad inconsistancy, we can try updating etc */
394 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
395 char *devname = tmpdev->devname;
396 struct stat stb;
397 /* looks like a good enough match to update the super block if needed */
398 if (update) {
399 int dfd;
400 /* prepare useful information in info structures */
401 struct stat stb2;
402 fstat(mdfd, &stb2);
403
404 if (strcmp(update, "uuid")==0 &&
405 !ident->uuid_set) {
406 int rfd;
407 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
408 read(rfd, ident->uuid, 16) != 16) {
409 *(__u32*)(ident->uuid) = random();
410 *(__u32*)(ident->uuid+1) = random();
411 *(__u32*)(ident->uuid+2) = random();
412 *(__u32*)(ident->uuid+3) = random();
413 }
414 if (rfd >= 0) close(rfd);
415 }
416 dfd = dev_open(devname, O_RDWR|O_EXCL);
417
418 if (super) {
419 free(super);
420 super = NULL;
421 }
422
423 st->ss->load_super(st, dfd, &super, NULL);
424 st->ss->getinfo_super(&info, super);
425
426 memcpy(info.uuid, ident->uuid, 16);
427 strcpy(info.name, ident->name);
428 info.array.md_minor = minor(stb2.st_rdev);
429
430 st->ss->update_super(&info, super, update, devname, verbose,
431 ident->uuid_set, homehost);
432 if (strcmp(update, "uuid")==0 &&
433 !ident->uuid_set) {
434 ident->uuid_set = 1;
435 memcpy(ident->uuid, info.uuid, 16);
436 }
437 if (dfd < 0)
438 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
439 devname);
440 else if (st->ss->store_super(st, dfd, super))
441 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
442 devname);
443 if (dfd >= 0)
444 close(dfd);
445
446 if (strcmp(update, "uuid")==0 &&
447 ident->bitmap_fd)
448 bitmap_update_uuid(ident->bitmap_fd, info.uuid);
449 } else {
450 int dfd;
451 dfd = dev_open(devname, O_RDWR|O_EXCL);
452
453 if (super) {
454 free(super);
455 super = NULL;
456 }
457
458 st->ss->load_super(st, dfd, &super, NULL);
459 st->ss->getinfo_super(&info, super);
460 close(dfd);
461 }
462
463 stat(devname, &stb);
464
465 if (verbose > 0)
466 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
467 devname, mddev, info.disk.raid_disk);
468 devices[devcnt].devname = devname;
469 devices[devcnt].major = major(stb.st_rdev);
470 devices[devcnt].minor = minor(stb.st_rdev);
471 devices[devcnt].oldmajor = info.disk.major;
472 devices[devcnt].oldminor = info.disk.minor;
473 devices[devcnt].events = info.events;
474 devices[devcnt].raid_disk = info.disk.raid_disk;
475 devices[devcnt].disk_nr = info.disk.number;
476 devices[devcnt].uptodate = 0;
477 devices[devcnt].state = info.disk.state;
478 if (most_recent < devcnt) {
479 if (devices[devcnt].events
480 > devices[most_recent].events)
481 most_recent = devcnt;
482 }
483 if (info.array.level == -4)
484 /* with multipath, the raid_disk from the superblock is meaningless */
485 i = devcnt;
486 else
487 i = devices[devcnt].raid_disk;
488 if (i+1 == 0) {
489 if (nextspare < info.array.raid_disks)
490 nextspare = info.array.raid_disks;
491 i = nextspare++;
492 }
493 if (i < 10000) {
494 if (i >= bestcnt) {
495 unsigned int newbestcnt = i+10;
496 int *newbest = malloc(sizeof(int)*newbestcnt);
497 unsigned int c;
498 for (c=0; c < newbestcnt; c++)
499 if (c < bestcnt)
500 newbest[c] = best[c];
501 else
502 newbest[c] = -1;
503 if (best)free(best);
504 best = newbest;
505 bestcnt = newbestcnt;
506 }
507 if (best[i] == -1
508 || devices[best[i]].events < devices[devcnt].events)
509 best[i] = devcnt;
510 }
511 devcnt++;
512 }
513
514 if (super)
515 free(super);
516 super = NULL;
517
518 if (update && strcmp(update, "byteorder")==0)
519 st->minor_version = 90;
520
521 if (devcnt == 0) {
522 fprintf(stderr, Name ": no devices found for %s\n",
523 mddev);
524 free(first_super);
525 if (must_close) close(mdfd);
526 return 1;
527 }
528
529 st->ss->getinfo_super(&info, first_super);
530
531 /* now we have some devices that might be suitable.
532 * I wonder how many
533 */
534 avail = malloc(info.array.raid_disks);
535 memset(avail, 0, info.array.raid_disks);
536 okcnt = 0;
537 sparecnt=0;
538 for (i=0; i< bestcnt ;i++) {
539 int j = best[i];
540 int event_margin = 1; /* always allow a difference of '1'
541 * like the kernel does
542 */
543 if (j < 0) continue;
544 /* note: we ignore error flags in multipath arrays
545 * as they don't make sense
546 */
547 if (info.array.level != -4)
548 if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
549 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
550 sparecnt++;
551 continue;
552 }
553 if (devices[j].events+event_margin >=
554 devices[most_recent].events) {
555 devices[j].uptodate = 1;
556 if (i < info.array.raid_disks) {
557 okcnt++;
558 avail[i]=1;
559 } else
560 sparecnt++;
561 }
562 }
563 while (force && !enough(info.array.level, info.array.raid_disks,
564 info.array.layout,
565 avail, okcnt)) {
566 /* Choose the newest best drive which is
567 * not up-to-date, update the superblock
568 * and add it.
569 */
570 int fd;
571 chosen_drive = -1;
572 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
573 int j = best[i];
574 if (j>=0 &&
575 !devices[j].uptodate &&
576 devices[j].events > 0 &&
577 (chosen_drive < 0 ||
578 devices[j].events > devices[chosen_drive].events))
579 chosen_drive = j;
580 }
581 if (chosen_drive < 0)
582 break;
583 if (verbose >= 0)
584 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
585 devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
586 (int)(devices[chosen_drive].events),
587 (int)(devices[most_recent].events));
588 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
589 if (fd < 0) {
590 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
591 devices[chosen_drive].devname);
592 devices[chosen_drive].events = 0;
593 continue;
594 }
595 if (st->ss->load_super(st,fd, &super, NULL)) {
596 close(fd);
597 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
598 devices[chosen_drive].devname);
599 devices[chosen_drive].events = 0;
600 continue;
601 }
602 info.events = devices[most_recent].events;
603 st->ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose, 0, NULL);
604
605 if (st->ss->store_super(st, fd, super)) {
606 close(fd);
607 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
608 devices[chosen_drive].devname);
609 devices[chosen_drive].events = 0;
610 free(super);
611 continue;
612 }
613 close(fd);
614 devices[chosen_drive].events = devices[most_recent].events;
615 devices[chosen_drive].uptodate = 1;
616 avail[chosen_drive] = 1;
617 okcnt++;
618 free(super);
619 }
620
621 /* Now we want to look at the superblock which the kernel will base things on
622 * and compare the devices that we think are working with the devices that the
623 * superblock thinks are working.
624 * If there are differences and --force is given, then update this chosen
625 * superblock.
626 */
627 chosen_drive = -1;
628 super = NULL;
629 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
630 int j = best[i];
631 int fd;
632
633 if (j<0)
634 continue;
635 if (!devices[j].uptodate)
636 continue;
637 chosen_drive = j;
638 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
639 fprintf(stderr, Name ": Cannot open %s: %s\n",
640 devices[j].devname, strerror(errno));
641 if (must_close) close(mdfd);
642 return 1;
643 }
644 if (st->ss->load_super(st,fd, &super, NULL)) {
645 close(fd);
646 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
647 devices[j].devname);
648 if (must_close) close(mdfd);
649 return 1;
650 }
651 close(fd);
652 }
653 if (super == NULL) {
654 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
655 if (must_close) close(mdfd);
656 return 1;
657 }
658 st->ss->getinfo_super(&info, super);
659 for (i=0; i<bestcnt; i++) {
660 int j = best[i];
661 unsigned int desired_state;
662
663 if (i < info.array.raid_disks)
664 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
665 else
666 desired_state = 0;
667
668 if (j<0)
669 continue;
670 if (!devices[j].uptodate)
671 continue;
672 info.disk.number = devices[j].disk_nr;
673 info.disk.raid_disk = i;
674 info.disk.state = desired_state;
675
676 if (devices[j].uptodate &&
677 st->ss->update_super(&info, super, "assemble", NULL, verbose, 0, NULL)) {
678 if (force) {
679 if (verbose >= 0)
680 fprintf(stderr, Name ": "
681 "clearing FAULTY flag for device %d in %s for %s\n",
682 j, mddev, devices[j].devname);
683 change = 1;
684 } else {
685 if (verbose >= -1)
686 fprintf(stderr, Name ": "
687 "device %d in %s has wrong state in superblock, but %s seems ok\n",
688 i, mddev, devices[j].devname);
689 }
690 }
691 #if 0
692 if (!devices[j].uptodate &&
693 !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
694 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
695 i, mddev);
696 }
697 #endif
698 }
699 if (force && okcnt == info.array.raid_disks-1) {
700 /* FIXME check event count */
701 change += st->ss->update_super(&info, super, "force",
702 devices[chosen_drive].devname, verbose, 0, NULL);
703 }
704
705 if (change) {
706 int fd;
707 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
708 if (fd < 0) {
709 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
710 devices[chosen_drive].devname);
711 if (must_close) close(mdfd);
712 return 1;
713 }
714 if (st->ss->store_super(st, fd, super)) {
715 close(fd);
716 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
717 devices[chosen_drive].devname);
718 if (must_close) close(mdfd);
719 return 1;
720 }
721 close(fd);
722 }
723
724 /* If we are in the middle of a reshape we may need to restore saved data
725 * that was moved aside due to the reshape overwriting live data
726 * The code of doing this lives in Grow.c
727 */
728 #ifndef MDASSEMBLE
729 if (info.reshape_active) {
730 int err = 0;
731 int *fdlist = malloc(sizeof(int)* bestcnt);
732 for (i=0; i<bestcnt; i++) {
733 int j = best[i];
734 if (j >= 0) {
735 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
736 if (fdlist[i] < 0) {
737 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
738 devices[j].devname);
739 err = 1;
740 break;
741 }
742 } else
743 fdlist[i] = -1;
744 }
745 if (!err)
746 err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
747 while (i>0) {
748 i--;
749 if (fdlist[i]>=0) close(fdlist[i]);
750 }
751 if (err) {
752 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
753 if (must_close) close(mdfd);
754 return err;
755 }
756 }
757 #endif
758 /* count number of in-sync devices according to the superblock.
759 * We must have this number to start the array without -s or -R
760 */
761 req_cnt = info.array.working_disks;
762
763 /* Almost ready to actually *do* something */
764 if (!old_linux) {
765 int rv;
766 if ((vers % 100) >= 1) { /* can use different versions */
767 mdu_array_info_t inf;
768 memset(&inf, 0, sizeof(inf));
769 inf.major_version = st->ss->major;
770 inf.minor_version = st->minor_version;
771 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
772 } else
773 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
774
775 if (rv) {
776 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
777 mddev, strerror(errno));
778 if (must_close) close(mdfd);
779 return 1;
780 }
781 if (ident->bitmap_fd >= 0) {
782 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
783 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
784 if (must_close) close(mdfd);
785 return 1;
786 }
787 } else if (ident->bitmap_file) {
788 /* From config file */
789 int bmfd = open(ident->bitmap_file, O_RDWR);
790 if (bmfd < 0) {
791 fprintf(stderr, Name ": Could not open bitmap file %s\n",
792 ident->bitmap_file);
793 if (must_close) close(mdfd);
794 return 1;
795 }
796 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
797 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
798 close(bmfd);
799 if (must_close) close(mdfd);
800 return 1;
801 }
802 close(bmfd);
803 }
804
805 /* First, add the raid disks, but add the chosen one last */
806 for (i=0; i<= bestcnt; i++) {
807 int j;
808 if (i < bestcnt) {
809 j = best[i];
810 if (j == chosen_drive)
811 continue;
812 } else
813 j = chosen_drive;
814
815 if (j >= 0 /* && devices[j].uptodate */) {
816 mdu_disk_info_t disk;
817 memset(&disk, 0, sizeof(disk));
818 disk.major = devices[j].major;
819 disk.minor = devices[j].minor;
820 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
821 fprintf(stderr, Name ": failed to add %s to %s: %s\n",
822 devices[j].devname,
823 mddev,
824 strerror(errno));
825 if (i < info.array.raid_disks || i == bestcnt)
826 okcnt--;
827 else
828 sparecnt--;
829 } else if (verbose > 0)
830 fprintf(stderr, Name ": added %s to %s as %d\n",
831 devices[j].devname, mddev, devices[j].raid_disk);
832 } else if (verbose > 0 && i < info.array.raid_disks)
833 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
834 i, mddev);
835 }
836
837 if (runstop == 1 ||
838 (runstop <= 0 &&
839 ( enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt) &&
840 (okcnt >= req_cnt || start_partial_ok)
841 ))) {
842 if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
843 if (verbose >= 0) {
844 fprintf(stderr, Name ": %s has been started with %d drive%s",
845 mddev, okcnt, okcnt==1?"":"s");
846 if (okcnt < info.array.raid_disks)
847 fprintf(stderr, " (out of %d)", info.array.raid_disks);
848 if (sparecnt)
849 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
850 fprintf(stderr, ".\n");
851 }
852 if (must_close) {
853 int usecs = 1;
854 close(mdfd);
855 /* There is a nasty race with 'mdadm --monitor'.
856 * If it opens this device before we close it,
857 * it gets an incomplete open on which IO
858 * doesn't work and the capacity if wrong.
859 * If we reopen (to check for layered devices)
860 * before --monitor closes, we loose.
861 *
862 * So: wait upto 1 second for there to be
863 * a non-zero capacity.
864 */
865 while (usecs < 1000) {
866 mdfd = open(mddev, O_RDONLY);
867 if (mdfd >= 0) {
868 unsigned long size;
869 if (ioctl(mdfd, BLKGETSIZE, &size) == 0 &&
870 size > 0)
871 break;
872 close(mdfd);
873 }
874 usleep(usecs);
875 usecs <<= 1;
876 }
877 }
878 return 0;
879 }
880 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
881 mddev, strerror(errno));
882 if (must_close) close(mdfd);
883 return 1;
884 }
885 if (runstop == -1) {
886 fprintf(stderr, Name ": %s assembled from %d drive%s",
887 mddev, okcnt, okcnt==1?"":"s");
888 if (okcnt != info.array.raid_disks)
889 fprintf(stderr, " (out of %d)", info.array.raid_disks);
890 fprintf(stderr, ", but not started.\n");
891 if (must_close) close(mdfd);
892 return 0;
893 }
894 if (verbose >= 0) {
895 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
896 if (sparecnt)
897 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
898 if (!enough(info.array.level, info.array.raid_disks, info.array.layout, avail, okcnt))
899 fprintf(stderr, " - not enough to start the array.\n");
900 else {
901 if (req_cnt == info.array.raid_disks)
902 fprintf(stderr, " - need all %d to start it", req_cnt);
903 else
904 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
905 fprintf(stderr, " (use --run to insist).\n");
906 }
907 }
908 if (must_close) close(mdfd);
909 return 1;
910 } else {
911 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
912 * been updated to point to the current locations of devices.
913 * so we can just start the array
914 */
915 unsigned long dev;
916 dev = makedev(devices[chosen_drive].major,
917 devices[chosen_drive].minor);
918 if (ioctl(mdfd, START_ARRAY, dev)) {
919 fprintf(stderr, Name ": Cannot start array: %s\n",
920 strerror(errno));
921 }
922
923 }
924 if (must_close) close(mdfd);
925 return 0;
926 }