]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
fix mdmon takeover
[thirdparty/mdadm.git] / Assemble.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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@suse.de>
23 */
24
25 #include "mdadm.h"
26 #include <ctype.h>
27
28 static int name_matches(char *found, char *required, char *homehost)
29 {
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
32 */
33 char fnd[33];
34
35 strncpy(fnd, found, 32);
36 fnd[32] = 0;
37 if (strcmp(found, required)==0)
38 return 1;
39 if (homehost) {
40 int l = strlen(homehost);
41 if (l < 32 && fnd[l] == ':' &&
42 strcmp(fnd+l+1, required)==0)
43 return 1;
44 }
45 return 0;
46 }
47
48 static int is_member_busy(char *metadata_version)
49 {
50 /* check if the given member array is active */
51 struct mdstat_ent *mdstat = mdstat_read(1, 0);
52 struct mdstat_ent *ent;
53 int busy = 0;
54
55 for (ent = mdstat; ent; ent = ent->next) {
56 if (ent->metadata_version == NULL)
57 continue;
58 if (strncmp(ent->metadata_version, "external:", 9) != 0)
59 continue;
60 if (!is_subarray(&ent->metadata_version[9]))
61 continue;
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent->metadata_version[10], metadata_version+1) == 0) {
64 busy = 1;
65 break;
66 }
67 }
68 free_mdstat(mdstat);
69
70 return busy;
71 }
72
73 int Assemble(struct supertype *st, char *mddev,
74 mddev_ident_t ident,
75 mddev_dev_t devlist, char *backup_file,
76 int readonly, int runstop,
77 char *update, char *homehost, int require_homehost,
78 int verbose, int force)
79 {
80 /*
81 * The task of Assemble is to find a collection of
82 * devices that should (according to their superblocks)
83 * form an array, and to give this collection to the MD driver.
84 * In Linux-2.4 and later, this involves submitting a
85 * SET_ARRAY_INFO ioctl with no arg - to prepare
86 * the array - and then submit a number of
87 * ADD_NEW_DISK ioctls to add disks into
88 * the array. Finally RUN_ARRAY might
89 * be submitted to start the array.
90 *
91 * Much of the work of Assemble is in finding and/or
92 * checking the disks to make sure they look right.
93 *
94 * If mddev is not set, then scan must be set and we
95 * read through the config file for dev+uuid mapping
96 * We recurse, setting mddev, for each device that
97 * - isn't running
98 * - has a valid uuid (or any uuid if !uuidset)
99 *
100 * If mddev is set, we try to determine state of md.
101 * check version - must be at least 0.90.0
102 * check kernel version. must be at least 2.4.
103 * If not, we can possibly fall back on START_ARRAY
104 * Try to GET_ARRAY_INFO.
105 * If possible, give up
106 * If not, try to STOP_ARRAY just to make sure
107 *
108 * If !uuidset and scan, look in conf-file for uuid
109 * If not found, give up
110 * If !devlist and scan and uuidset, get list of devs from conf-file
111 *
112 * For each device:
113 * Check superblock - discard if bad
114 * Check uuid (set if we don't have one) - discard if no match
115 * Check superblock similarity if we have a superblock - discard if different
116 * Record events, devicenum
117 * This should give us a list of devices for the array
118 * We should collect the most recent event number
119 *
120 * Count disks with recent enough event count
121 * While force && !enough disks
122 * Choose newest rejected disks, update event count
123 * mark clean and rewrite superblock
124 * If recent kernel:
125 * SET_ARRAY_INFO
126 * foreach device with recent events : ADD_NEW_DISK
127 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
128 * If old kernel:
129 * Check the device numbers in superblock are right
130 * update superblock if any changes
131 * START_ARRAY
132 *
133 */
134 int mdfd;
135 int clean;
136 int auto_assem = (mddev == NULL && !ident->uuid_set &&
137 ident->super_minor == UnSet && ident->name[0] == 0
138 && (ident->container == NULL || ident->member == NULL));
139 int old_linux = 0;
140 int vers = vers; /* Keep gcc quite - it really is initialised */
141 struct {
142 char *devname;
143 int uptodate; /* set once we decide that this device is as
144 * recent as everything else in the array.
145 */
146 struct mdinfo i;
147 } *devices;
148 int *best = NULL; /* indexed by raid_disk */
149 unsigned int bestcnt = 0;
150 int devcnt = 0;
151 unsigned int okcnt, sparecnt;
152 unsigned int req_cnt;
153 unsigned int i;
154 int most_recent = 0;
155 int chosen_drive;
156 int change = 0;
157 int inargv = 0;
158 int report_missmatch;
159 int bitmap_done;
160 int start_partial_ok = (runstop >= 0) &&
161 (force || devlist==NULL || auto_assem);
162 unsigned int num_devs;
163 mddev_dev_t tmpdev;
164 struct mdinfo info;
165 struct mdinfo *content = NULL;
166 char *avail;
167 int nextspare = 0;
168 char *name = NULL;
169 int trustworthy;
170 char chosen_name[1024];
171
172 if (get_linux_version() < 2004000)
173 old_linux = 1;
174
175 /*
176 * If any subdevs are listed, then any that don't
177 * match ident are discarded. Remainder must all match and
178 * become the array.
179 * If no subdevs, then we scan all devices in the config file, but
180 * there must be something in the identity
181 */
182
183 if (!devlist &&
184 ident->uuid_set == 0 &&
185 ident->super_minor < 0 &&
186 ident->name[0] == 0 &&
187 (ident->container == NULL || ident->member == NULL) &&
188 ident->devices == NULL) {
189 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
190 mddev ? mddev : "further assembly");
191 return 1;
192 }
193
194 if (devlist == NULL)
195 devlist = conf_get_devs();
196 else if (mddev)
197 inargv = 1;
198
199 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
200 try_again:
201 /* We come back here when doing auto-assembly and attempting some
202 * set of devices failed. Those are now marked as ->used==2 and
203 * we ignore them and try again
204 */
205
206 tmpdev = devlist; num_devs = 0;
207 while (tmpdev) {
208 if (tmpdev->used)
209 tmpdev->used = 2;
210 else
211 num_devs++;
212 tmpdev = tmpdev->next;
213 }
214 devices = malloc(num_devs * sizeof(*devices));
215
216 if (!st && ident->st) st = ident->st;
217
218 if (verbose>0)
219 fprintf(stderr, Name ": looking for devices for %s\n",
220 mddev ? mddev : "further assembly");
221
222 /* first walk the list of devices to find a consistent set
223 * that match the criterea, if that is possible.
224 * We flag the ones we like with 'used'.
225 */
226 for (tmpdev = devlist;
227 tmpdev;
228 tmpdev = tmpdev->next) {
229 char *devname = tmpdev->devname;
230 int dfd;
231 struct stat stb;
232 struct supertype *tst = dup_super(st);
233
234 if (tmpdev->used > 1) continue;
235
236 if (ident->devices &&
237 !match_oneof(ident->devices, devname)) {
238 if (report_missmatch)
239 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
240 continue;
241 }
242
243 dfd = dev_open(devname, O_RDONLY|O_EXCL);
244 if (dfd < 0) {
245 if (report_missmatch)
246 fprintf(stderr, Name ": cannot open device %s: %s\n",
247 devname, strerror(errno));
248 tmpdev->used = 2;
249 } else if (fstat(dfd, &stb)< 0) {
250 /* Impossible! */
251 fprintf(stderr, Name ": fstat failed for %s: %s\n",
252 devname, strerror(errno));
253 tmpdev->used = 2;
254 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
255 fprintf(stderr, Name ": %s is not a block device.\n",
256 devname);
257 tmpdev->used = 2;
258 } else if (!tst && (tst = guess_super(dfd)) == NULL) {
259 if (report_missmatch)
260 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
261 devname);
262 tmpdev->used = 2;
263 } else if (auto_assem && st == NULL &&
264 !conf_test_metadata(tst->ss->name)) {
265 if (report_missmatch)
266 fprintf(stderr, Name ": %s has metadata type %s for which "
267 "auto-assembly is disabled\n",
268 devname, tst->ss->name);
269 tmpdev->used = 2;
270 } else if (tst->ss->load_super(tst,dfd, NULL)) {
271 if (report_missmatch)
272 fprintf( stderr, Name ": no RAID superblock on %s\n",
273 devname);
274 } else {
275 content = &info;
276 memset(content, 0, sizeof(*content));
277 tst->ss->getinfo_super(tst, content);
278 }
279 if (dfd >= 0) close(dfd);
280
281 if (tst && tst->sb && tst->ss->container_content
282 && tst->loaded_container) {
283 /* tmpdev is a container. We need to be either
284 * looking for a member, or auto-assembling
285 */
286 if (st) {
287 /* already found some components, this cannot
288 * be another one.
289 */
290 if (report_missmatch)
291 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
292 devname);
293 goto loop;
294 }
295
296 if (ident->container) {
297 if (ident->container[0] == '/' &&
298 !same_dev(ident->container, devname)) {
299 if (report_missmatch)
300 fprintf(stderr, Name ": %s is not the container required (%s)\n",
301 devname, ident->container);
302 goto loop;
303 }
304 if (ident->container[0] != '/') {
305 /* we have a uuid */
306 int uuid[4];
307 if (!parse_uuid(ident->container, uuid) ||
308 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
309 if (report_missmatch)
310 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
311 devname);
312 goto loop;
313 }
314 }
315 }
316 /* It is worth looking inside this container.
317 */
318 if (verbose > 0)
319 fprintf(stderr, Name ": looking in container %s\n",
320 devname);
321 next_member:
322 if (tmpdev->content)
323 content = tmpdev->content;
324 else
325 content = tst->ss->container_content(tst);
326 if (!content)
327 goto loop; /* empty container */
328
329 tmpdev->content = content->next;
330 if (tmpdev->content == NULL)
331 tmpdev->used = 2;
332
333 } else if (ident->container || ident->member) {
334 /* No chance of this matching if we don't have
335 * a container */
336 if (report_missmatch)
337 fprintf(stderr, Name "%s is not a container, and one is required.\n",
338 devname);
339 goto loop;
340 }
341
342 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
343 (!tst || !tst->sb ||
344 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0)) {
345 if (report_missmatch)
346 fprintf(stderr, Name ": %s has wrong uuid.\n",
347 devname);
348 goto loop;
349 }
350 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
351 (!tst || !tst->sb ||
352 name_matches(content->name, ident->name, homehost)==0)) {
353 if (report_missmatch)
354 fprintf(stderr, Name ": %s has wrong name.\n",
355 devname);
356 goto loop;
357 }
358 if (ident->super_minor != UnSet &&
359 (!tst || !tst->sb ||
360 ident->super_minor != content->array.md_minor)) {
361 if (report_missmatch)
362 fprintf(stderr, Name ": %s has wrong super-minor.\n",
363 devname);
364 goto loop;
365 }
366 if (ident->level != UnSet &&
367 (!tst || !tst->sb ||
368 ident->level != content->array.level)) {
369 if (report_missmatch)
370 fprintf(stderr, Name ": %s has wrong raid level.\n",
371 devname);
372 goto loop;
373 }
374 if (ident->raid_disks != UnSet &&
375 (!tst || !tst->sb ||
376 ident->raid_disks!= content->array.raid_disks)) {
377 if (report_missmatch)
378 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
379 devname);
380 goto loop;
381 }
382 if (auto_assem) {
383 if (tst == NULL || tst->sb == NULL)
384 continue;
385 }
386 /* If we are this far, then we are nearly commited to this device.
387 * If the super_block doesn't exist, or doesn't match others,
388 * then we probably cannot continue
389 * However if one of the arrays is for the homehost, and
390 * the other isn't that can disambiguate.
391 */
392
393 if (!tst || !tst->sb) {
394 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
395 devname);
396 if (st)
397 st->ss->free_super(st);
398 return 1;
399 }
400
401 if (tst && tst->sb && tst->ss->container_content
402 && tst->loaded_container) {
403 /* we have the one container we need, don't keep
404 * looking. If the chosen member is active, skip.
405 */
406 if (is_member_busy(content->text_version)) {
407 if (report_missmatch)
408 fprintf(stderr, Name ": member %s in %s is already assembled\n",
409 content->text_version,
410 devname);
411 skip:
412 if (tmpdev->content)
413 goto next_member;
414 tst->ss->free_super(tst);
415 tst = NULL;
416 content = NULL;
417 if (auto_assem)
418 goto loop;
419 return 1;
420 }
421 if (ident->member && ident->member[0]) {
422 char *s = strchr(content->text_version+1, '/');
423 if (s == NULL) {
424 fprintf(stderr, Name ": badly formatted version: %s\n",
425 content->text_version);
426 goto skip;
427 }
428 if (strcmp(ident->member, s+1) != 0) {
429 if (report_missmatch)
430 fprintf(stderr,
431 Name ": skipping wrong member %s\n",
432 content->text_version);
433 goto skip;
434 }
435 }
436 st = tst; tst = NULL;
437 if (!auto_assem && inargv && tmpdev->next != NULL) {
438 fprintf(stderr, Name ": %s is a container, but is not "
439 "only device given: confused and aborting\n",
440 devname);
441 st->ss->free_super(st);
442 return 1;
443 }
444 if (verbose > 0)
445 fprintf(stderr, Name ": found match on member %s in %s\n",
446 content->text_version, devname);
447 break;
448 }
449 if (st == NULL)
450 st = dup_super(tst);
451 if (st->minor_version == -1)
452 st->minor_version = tst->minor_version;
453 if (st->ss != tst->ss ||
454 st->minor_version != tst->minor_version ||
455 st->ss->compare_super(st, tst) != 0) {
456 /* Some mismatch. If exactly one array matches this host,
457 * we can resolve on that one.
458 * Or, if we are auto assembling, we just ignore the second
459 * for now.
460 */
461 if (auto_assem)
462 goto loop;
463 if (homehost) {
464 int first = st->ss->match_home(st, homehost);
465 int last = tst->ss->match_home(tst, homehost);
466 if (first != last &&
467 (first == 1 || last == 1)) {
468 /* We can do something */
469 if (first) {/* just ignore this one */
470 if (report_missmatch)
471 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
472 devname);
473 goto loop;
474 } else { /* reject all those sofar */
475 mddev_dev_t td;
476 if (report_missmatch)
477 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
478 devname);
479 for (td=devlist; td != tmpdev; td=td->next)
480 if (td->used == 1)
481 td->used = 0;
482 tmpdev->used = 1;
483 goto loop;
484 }
485 }
486 }
487 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
488 devname);
489 tst->ss->free_super(tst);
490 st->ss->free_super(st);
491 return 1;
492 }
493
494 tmpdev->used = 1;
495
496 loop:
497 if (tmpdev->content)
498 goto next_member;
499 if (tst)
500 tst->ss->free_super(tst);
501 }
502
503 if (!st || !st->sb || !content)
504 return 2;
505
506 /* Now need to open the array device. Use create_mddev */
507 if (content == &info)
508 st->ss->getinfo_super(st, content);
509
510 trustworthy = FOREIGN;
511 name = content->name;
512 switch (st->ss->match_home(st, homehost)
513 ?: st->ss->match_home(st, "any")) {
514 case 1:
515 trustworthy = LOCAL;
516 name = strchr(content->name, ':');
517 if (name)
518 name++;
519 else
520 name = content->name;
521 break;
522 }
523 if (!auto_assem)
524 /* If the array is listed in mdadm.conf or on
525 * command line, then we trust the name
526 * even if the array doesn't look local
527 */
528 trustworthy = LOCAL;
529
530 if (name[0] == 0 &&
531 content->array.level == LEVEL_CONTAINER) {
532 name = content->text_version;
533 trustworthy = METADATA;
534 }
535
536 if (name[0] && trustworthy != LOCAL &&
537 ! require_homehost &&
538 conf_name_is_free(name))
539 trustworthy = LOCAL;
540
541 if (trustworthy == LOCAL &&
542 strchr(name, ':'))
543 /* Ignore 'host:' prefix of name */
544 name = strchr(name, ':')+1;
545
546 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
547 chosen_name);
548 if (mdfd < 0) {
549 st->ss->free_super(st);
550 free(devices);
551 if (auto_assem)
552 goto try_again;
553 return 1;
554 }
555 mddev = chosen_name;
556 vers = md_get_version(mdfd);
557 if (vers < 9000) {
558 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
559 " Upgrade your kernel or try --build\n");
560 close(mdfd);
561 return 1;
562 }
563 if (mddev_busy(fd2devnum(mdfd))) {
564 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
565 mddev);
566 for (tmpdev = devlist ;
567 tmpdev && tmpdev->used != 1;
568 tmpdev = tmpdev->next)
569 ;
570 if (tmpdev && auto_assem)
571 fprintf(stderr, Name ": %s needed for %s...\n",
572 mddev, tmpdev->devname);
573 close(mdfd);
574 mdfd = -3;
575 st->ss->free_super(st);
576 free(devices);
577 if (auto_assem)
578 goto try_again;
579 return 1;
580 }
581 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
582
583 #ifndef MDASSEMBLE
584 if (content != &info) {
585 /* This is a member of a container. Try starting the array. */
586 return assemble_container_content(st, mdfd, content, runstop,
587 chosen_name, verbose);
588 }
589 #endif
590 /* Ok, no bad inconsistancy, we can try updating etc */
591 bitmap_done = 0;
592 content->update_private = NULL;
593 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
594 char *devname = tmpdev->devname;
595 struct stat stb;
596 /* looks like a good enough match to update the super block if needed */
597 #ifndef MDASSEMBLE
598 if (update) {
599 int dfd;
600 /* prepare useful information in info structures */
601 struct stat stb2;
602 struct supertype *tst;
603 fstat(mdfd, &stb2);
604
605 if (strcmp(update, "uuid")==0 &&
606 !ident->uuid_set) {
607 int rfd;
608 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
609 read(rfd, ident->uuid, 16) != 16) {
610 *(__u32*)(ident->uuid) = random();
611 *(__u32*)(ident->uuid+1) = random();
612 *(__u32*)(ident->uuid+2) = random();
613 *(__u32*)(ident->uuid+3) = random();
614 }
615 if (rfd >= 0) close(rfd);
616 }
617 dfd = dev_open(devname, O_RDWR|O_EXCL);
618
619 remove_partitions(dfd);
620
621 tst = dup_super(st);
622 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
623 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
624 devname);
625 if (dfd >= 0)
626 close(dfd);
627 close(mdfd);
628 return 1;
629 }
630 tst->ss->getinfo_super(tst, content);
631
632 memcpy(content->uuid, ident->uuid, 16);
633 strcpy(content->name, ident->name);
634 content->array.md_minor = minor(stb2.st_rdev);
635
636 tst->ss->update_super(tst, content, update,
637 devname, verbose,
638 ident->uuid_set, homehost);
639 if (strcmp(update, "uuid")==0 &&
640 !ident->uuid_set) {
641 ident->uuid_set = 1;
642 memcpy(ident->uuid, content->uuid, 16);
643 }
644 if (dfd < 0)
645 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
646 devname);
647 else if (tst->ss->store_super(tst, dfd))
648 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
649 devname);
650 if (dfd >= 0)
651 close(dfd);
652
653 if (strcmp(update, "uuid")==0 &&
654 ident->bitmap_fd >= 0 && !bitmap_done) {
655 if (bitmap_update_uuid(ident->bitmap_fd,
656 content->uuid,
657 tst->ss->swapuuid) != 0)
658 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
659 else
660 bitmap_done = 1;
661 }
662 tst->ss->free_super(tst);
663 } else
664 #endif
665 {
666 struct supertype *tst = dup_super(st);
667 int dfd;
668 dfd = dev_open(devname, O_RDWR|O_EXCL);
669
670 remove_partitions(dfd);
671
672 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
673 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
674 devname);
675 if (dfd >= 0)
676 close(dfd);
677 close(mdfd);
678 return 1;
679 }
680 tst->ss->getinfo_super(tst, content);
681 tst->ss->free_super(tst);
682 close(dfd);
683 }
684
685 stat(devname, &stb);
686
687 if (verbose > 0)
688 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
689 devname, mddev, content->disk.raid_disk);
690 devices[devcnt].devname = devname;
691 devices[devcnt].uptodate = 0;
692 devices[devcnt].i = *content;
693 devices[devcnt].i.disk.major = major(stb.st_rdev);
694 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
695 if (most_recent < devcnt) {
696 if (devices[devcnt].i.events
697 > devices[most_recent].i.events)
698 most_recent = devcnt;
699 }
700 if (content->array.level == LEVEL_MULTIPATH)
701 /* with multipath, the raid_disk from the superblock is meaningless */
702 i = devcnt;
703 else
704 i = devices[devcnt].i.disk.raid_disk;
705 if (i+1 == 0) {
706 if (nextspare < content->array.raid_disks)
707 nextspare = content->array.raid_disks;
708 i = nextspare++;
709 } else {
710 if (i >= content->array.raid_disks &&
711 i >= nextspare)
712 nextspare = i+1;
713 }
714 if (i < 10000) {
715 if (i >= bestcnt) {
716 unsigned int newbestcnt = i+10;
717 int *newbest = malloc(sizeof(int)*newbestcnt);
718 unsigned int c;
719 for (c=0; c < newbestcnt; c++)
720 if (c < bestcnt)
721 newbest[c] = best[c];
722 else
723 newbest[c] = -1;
724 if (best)free(best);
725 best = newbest;
726 bestcnt = newbestcnt;
727 }
728 if (best[i] >=0 &&
729 devices[best[i]].i.events
730 == devices[devcnt].i.events
731 && (devices[best[i]].i.disk.minor
732 != devices[devcnt].i.disk.minor)
733 && st->ss == &super0
734 && content->array.level != LEVEL_MULTIPATH) {
735 /* two different devices with identical superblock.
736 * Could be a mis-detection caused by overlapping
737 * partitions. fail-safe.
738 */
739 fprintf(stderr, Name ": WARNING %s and %s appear"
740 " to have very similar superblocks.\n"
741 " If they are really different, "
742 "please --zero the superblock on one\n"
743 " If they are the same or overlap,"
744 " please remove one from %s.\n",
745 devices[best[i]].devname, devname,
746 inargv ? "the list" :
747 "the\n DEVICE list in mdadm.conf"
748 );
749 close(mdfd);
750 return 1;
751 }
752 if (best[i] == -1
753 || (devices[best[i]].i.events
754 < devices[devcnt].i.events))
755 best[i] = devcnt;
756 }
757 devcnt++;
758 }
759 free(content->update_private);
760 content->update_private = NULL;
761
762 if (devcnt == 0) {
763 fprintf(stderr, Name ": no devices found for %s\n",
764 mddev);
765 if (st)
766 st->ss->free_super(st);
767 close(mdfd);
768 return 1;
769 }
770
771 if (update && strcmp(update, "byteorder")==0)
772 st->minor_version = 90;
773
774 st->ss->getinfo_super(st, content);
775 clean = content->array.state & 1;
776
777 /* now we have some devices that might be suitable.
778 * I wonder how many
779 */
780 avail = malloc(content->array.raid_disks);
781 memset(avail, 0, content->array.raid_disks);
782 okcnt = 0;
783 sparecnt=0;
784 for (i=0; i< bestcnt ;i++) {
785 int j = best[i];
786 int event_margin = 1; /* always allow a difference of '1'
787 * like the kernel does
788 */
789 if (j < 0) continue;
790 /* note: we ignore error flags in multipath arrays
791 * as they don't make sense
792 */
793 if (content->array.level != LEVEL_MULTIPATH)
794 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
795 if (!(devices[j].i.disk.state
796 & (1<<MD_DISK_FAULTY)))
797 sparecnt++;
798 continue;
799 }
800 if (devices[j].i.events+event_margin >=
801 devices[most_recent].i.events) {
802 devices[j].uptodate = 1;
803 if (i < content->array.raid_disks &&
804 devices[j].i.recovery_start == MaxSector) {
805 okcnt++;
806 avail[i]=1;
807 } else
808 sparecnt++;
809 }
810 }
811 while (force && !enough(content->array.level, content->array.raid_disks,
812 content->array.layout, 1,
813 avail, okcnt)) {
814 /* Choose the newest best drive which is
815 * not up-to-date, update the superblock
816 * and add it.
817 */
818 int fd;
819 struct supertype *tst;
820 long long current_events;
821 chosen_drive = -1;
822 for (i=0; i<content->array.raid_disks && i < bestcnt; i++) {
823 int j = best[i];
824 if (j>=0 &&
825 !devices[j].uptodate &&
826 devices[j].i.recovery_start == MaxSector &&
827 (chosen_drive < 0 ||
828 devices[j].i.events
829 > devices[chosen_drive].i.events))
830 chosen_drive = j;
831 }
832 if (chosen_drive < 0)
833 break;
834 current_events = devices[chosen_drive].i.events;
835 add_another:
836 if (verbose >= 0)
837 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
838 devices[chosen_drive].devname,
839 devices[chosen_drive].i.disk.raid_disk,
840 (int)(devices[chosen_drive].i.events),
841 (int)(devices[most_recent].i.events));
842 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
843 if (fd < 0) {
844 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
845 devices[chosen_drive].devname);
846 devices[chosen_drive].i.events = 0;
847 continue;
848 }
849 tst = dup_super(st);
850 if (tst->ss->load_super(tst,fd, NULL)) {
851 close(fd);
852 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
853 devices[chosen_drive].devname);
854 devices[chosen_drive].i.events = 0;
855 continue;
856 }
857 content->events = devices[most_recent].i.events;
858 tst->ss->update_super(tst, content, "force-one",
859 devices[chosen_drive].devname, verbose,
860 0, NULL);
861
862 if (tst->ss->store_super(tst, fd)) {
863 close(fd);
864 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
865 devices[chosen_drive].devname);
866 devices[chosen_drive].i.events = 0;
867 tst->ss->free_super(tst);
868 continue;
869 }
870 close(fd);
871 devices[chosen_drive].i.events = devices[most_recent].i.events;
872 devices[chosen_drive].uptodate = 1;
873 avail[chosen_drive] = 1;
874 okcnt++;
875 tst->ss->free_super(tst);
876
877 /* If there are any other drives of the same vintage,
878 * add them in as well. We can't lose and we might gain
879 */
880 for (i=0; i<content->array.raid_disks && i < bestcnt ; i++) {
881 int j = best[i];
882 if (j >= 0 &&
883 !devices[j].uptodate &&
884 devices[j].i.events == current_events) {
885 chosen_drive = j;
886 goto add_another;
887 }
888 }
889 }
890
891 /* Now we want to look at the superblock which the kernel will base things on
892 * and compare the devices that we think are working with the devices that the
893 * superblock thinks are working.
894 * If there are differences and --force is given, then update this chosen
895 * superblock.
896 */
897 chosen_drive = -1;
898 st->ss->free_super(st);
899 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
900 int j = best[i];
901 int fd;
902
903 if (j<0)
904 continue;
905 if (!devices[j].uptodate)
906 continue;
907 chosen_drive = j;
908 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
909 fprintf(stderr, Name ": Cannot open %s: %s\n",
910 devices[j].devname, strerror(errno));
911 close(mdfd);
912 return 1;
913 }
914 if (st->ss->load_super(st,fd, NULL)) {
915 close(fd);
916 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
917 devices[j].devname);
918 close(mdfd);
919 return 1;
920 }
921 close(fd);
922 }
923 if (st->sb == NULL) {
924 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
925 close(mdfd);
926 return 1;
927 }
928 st->ss->getinfo_super(st, content);
929 #ifndef MDASSEMBLE
930 sysfs_init(content, mdfd, 0);
931 #endif
932 for (i=0; i<bestcnt; i++) {
933 int j = best[i];
934 unsigned int desired_state;
935
936 if (i < content->array.raid_disks)
937 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
938 else
939 desired_state = 0;
940
941 if (j<0)
942 continue;
943 if (!devices[j].uptodate)
944 continue;
945
946 devices[j].i.disk.state = desired_state;
947 if (!(devices[j].i.array.state & 1))
948 clean = 0;
949
950 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
951 verbose, 0, NULL)) {
952 if (force) {
953 if (verbose >= 0)
954 fprintf(stderr, Name ": "
955 "clearing FAULTY flag for device %d in %s for %s\n",
956 j, mddev, devices[j].devname);
957 change = 1;
958 } else {
959 if (verbose >= -1)
960 fprintf(stderr, Name ": "
961 "device %d in %s has wrong state in superblock, but %s seems ok\n",
962 i, mddev, devices[j].devname);
963 }
964 }
965 #if 0
966 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
967 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
968 i, mddev);
969 }
970 #endif
971 }
972 if (force && !clean &&
973 !enough(content->array.level, content->array.raid_disks,
974 content->array.layout, clean,
975 avail, okcnt)) {
976 change += st->ss->update_super(st, content, "force-array",
977 devices[chosen_drive].devname, verbose,
978 0, NULL);
979 clean = 1;
980 }
981
982 if (change) {
983 int fd;
984 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
985 if (fd < 0) {
986 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
987 devices[chosen_drive].devname);
988 close(mdfd);
989 return 1;
990 }
991 if (st->ss->store_super(st, fd)) {
992 close(fd);
993 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
994 devices[chosen_drive].devname);
995 close(mdfd);
996 return 1;
997 }
998 close(fd);
999 }
1000
1001 /* If we are in the middle of a reshape we may need to restore saved data
1002 * that was moved aside due to the reshape overwriting live data
1003 * The code of doing this lives in Grow.c
1004 */
1005 #ifndef MDASSEMBLE
1006 if (content->reshape_active) {
1007 int err = 0;
1008 int *fdlist = malloc(sizeof(int)* bestcnt);
1009 if (verbose > 0)
1010 fprintf(stderr, Name ":%s has an active reshape - checking "
1011 "if critical section needs to be restored\n",
1012 chosen_name);
1013 for (i=0; i<bestcnt; i++) {
1014 int j = best[i];
1015 if (j >= 0) {
1016 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1017 if (fdlist[i] < 0) {
1018 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1019 devices[j].devname);
1020 err = 1;
1021 break;
1022 }
1023 } else
1024 fdlist[i] = -1;
1025 }
1026 if (!err)
1027 err = Grow_restart(st, content, fdlist, bestcnt, backup_file, verbose > 0);
1028 while (i>0) {
1029 i--;
1030 if (fdlist[i]>=0) close(fdlist[i]);
1031 }
1032 if (err) {
1033 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1034 if (backup_file == NULL)
1035 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1036 close(mdfd);
1037 return err;
1038 }
1039 }
1040 #endif
1041 /* count number of in-sync devices according to the superblock.
1042 * We must have this number to start the array without -s or -R
1043 */
1044 req_cnt = content->array.working_disks;
1045
1046 /* Almost ready to actually *do* something */
1047 if (!old_linux) {
1048 int rv;
1049
1050 /* First, fill in the map, so that udev can find our name
1051 * as soon as we become active.
1052 */
1053 map_update(NULL, fd2devnum(mdfd), content->text_version,
1054 content->uuid, chosen_name);
1055
1056 rv = set_array_info(mdfd, st, content);
1057 if (rv) {
1058 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1059 mddev, strerror(errno));
1060 ioctl(mdfd, STOP_ARRAY, NULL);
1061 close(mdfd);
1062 return 1;
1063 }
1064 if (ident->bitmap_fd >= 0) {
1065 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1066 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1067 ioctl(mdfd, STOP_ARRAY, NULL);
1068 close(mdfd);
1069 return 1;
1070 }
1071 } else if (ident->bitmap_file) {
1072 /* From config file */
1073 int bmfd = open(ident->bitmap_file, O_RDWR);
1074 if (bmfd < 0) {
1075 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1076 ident->bitmap_file);
1077 ioctl(mdfd, STOP_ARRAY, NULL);
1078 close(mdfd);
1079 return 1;
1080 }
1081 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1082 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1083 close(bmfd);
1084 ioctl(mdfd, STOP_ARRAY, NULL);
1085 close(mdfd);
1086 return 1;
1087 }
1088 close(bmfd);
1089 }
1090
1091 /* First, add the raid disks, but add the chosen one last */
1092 for (i=0; i<= bestcnt; i++) {
1093 int j;
1094 if (i < bestcnt) {
1095 j = best[i];
1096 if (j == chosen_drive)
1097 continue;
1098 } else
1099 j = chosen_drive;
1100
1101 if (j >= 0 /* && devices[j].uptodate */) {
1102 rv = add_disk(mdfd, st, content, &devices[j].i);
1103
1104 if (rv) {
1105 fprintf(stderr, Name ": failed to add "
1106 "%s to %s: %s\n",
1107 devices[j].devname,
1108 mddev,
1109 strerror(errno));
1110 if (i < content->array.raid_disks
1111 || i == bestcnt)
1112 okcnt--;
1113 else
1114 sparecnt--;
1115 } else if (verbose > 0)
1116 fprintf(stderr, Name ": added %s "
1117 "to %s as %d\n",
1118 devices[j].devname, mddev,
1119 devices[j].i.disk.raid_disk);
1120 } else if (verbose > 0 && i < content->array.raid_disks)
1121 fprintf(stderr, Name ": no uptodate device for "
1122 "slot %d of %s\n",
1123 i, mddev);
1124 }
1125
1126 if (content->array.level == LEVEL_CONTAINER) {
1127 if (verbose >= 0) {
1128 fprintf(stderr, Name ": Container %s has been "
1129 "assembled with %d drive%s",
1130 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1131 if (okcnt < content->array.raid_disks)
1132 fprintf(stderr, " (out of %d)",
1133 content->array.raid_disks);
1134 fprintf(stderr, "\n");
1135 }
1136 sysfs_uevent(content, "change");
1137 wait_for(chosen_name, mdfd);
1138 close(mdfd);
1139 return 0;
1140 }
1141
1142 if (runstop == 1 ||
1143 (runstop <= 0 &&
1144 ( enough(content->array.level, content->array.raid_disks,
1145 content->array.layout, clean, avail, okcnt) &&
1146 (okcnt >= req_cnt || start_partial_ok)
1147 ))) {
1148 /* This array is good-to-go.
1149 * If a reshape is in progress then we might need to
1150 * continue monitoring it. In that case we start
1151 * it read-only and let the grow code make it writable.
1152 */
1153 int rv;
1154 #ifndef MDASSEMBLE
1155 if (content->reshape_active &&
1156 content->delta_disks <= 0)
1157 rv = Grow_continue(mdfd, st, content, backup_file);
1158 else
1159 #endif
1160 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1161 if (rv == 0) {
1162 if (verbose >= 0) {
1163 fprintf(stderr, Name ": %s has been started with %d drive%s",
1164 mddev, okcnt, okcnt==1?"":"s");
1165 if (okcnt < content->array.raid_disks)
1166 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1167 if (sparecnt)
1168 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1169 fprintf(stderr, ".\n");
1170 }
1171 if (content->reshape_active &&
1172 content->array.level >= 4 &&
1173 content->array.level <= 6) {
1174 /* might need to increase the size
1175 * of the stripe cache - default is 256
1176 */
1177 if (256 < 4 * (content->array.chunk_size/4096)) {
1178 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1179 if (sra)
1180 sysfs_set_num(sra, NULL,
1181 "stripe_cache_size",
1182 (4 * content->array.chunk_size / 4096) + 1);
1183 }
1184 }
1185 wait_for(mddev, mdfd);
1186 close(mdfd);
1187 if (auto_assem) {
1188 int usecs = 1;
1189 /* There is a nasty race with 'mdadm --monitor'.
1190 * If it opens this device before we close it,
1191 * it gets an incomplete open on which IO
1192 * doesn't work and the capacity is
1193 * wrong.
1194 * If we reopen (to check for layered devices)
1195 * before --monitor closes, we loose.
1196 *
1197 * So: wait upto 1 second for there to be
1198 * a non-zero capacity.
1199 */
1200 while (usecs < 1000) {
1201 mdfd = open(mddev, O_RDONLY);
1202 if (mdfd >= 0) {
1203 unsigned long long size;
1204 if (get_dev_size(mdfd, NULL, &size) &&
1205 size > 0)
1206 break;
1207 close(mdfd);
1208 }
1209 usleep(usecs);
1210 usecs <<= 1;
1211 }
1212 }
1213 return 0;
1214 }
1215 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1216 mddev, strerror(errno));
1217
1218 if (!enough(content->array.level, content->array.raid_disks,
1219 content->array.layout, 1, avail, okcnt))
1220 fprintf(stderr, Name ": Not enough devices to "
1221 "start the array.\n");
1222 else if (!enough(content->array.level,
1223 content->array.raid_disks,
1224 content->array.layout, clean,
1225 avail, okcnt))
1226 fprintf(stderr, Name ": Not enough devices to "
1227 "start the array while not clean "
1228 "- consider --force.\n");
1229
1230 if (auto_assem)
1231 ioctl(mdfd, STOP_ARRAY, NULL);
1232 close(mdfd);
1233 return 1;
1234 }
1235 if (runstop == -1) {
1236 fprintf(stderr, Name ": %s assembled from %d drive%s",
1237 mddev, okcnt, okcnt==1?"":"s");
1238 if (okcnt != content->array.raid_disks)
1239 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1240 fprintf(stderr, ", but not started.\n");
1241 close(mdfd);
1242 return 0;
1243 }
1244 if (verbose >= -1) {
1245 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1246 if (sparecnt)
1247 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1248 if (!enough(content->array.level, content->array.raid_disks,
1249 content->array.layout, 1, avail, okcnt))
1250 fprintf(stderr, " - not enough to start the array.\n");
1251 else if (!enough(content->array.level,
1252 content->array.raid_disks,
1253 content->array.layout, clean,
1254 avail, okcnt))
1255 fprintf(stderr, " - not enough to start the "
1256 "array while not clean - consider "
1257 "--force.\n");
1258 else {
1259 if (req_cnt == content->array.raid_disks)
1260 fprintf(stderr, " - need all %d to start it", req_cnt);
1261 else
1262 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1263 fprintf(stderr, " (use --run to insist).\n");
1264 }
1265 }
1266 if (auto_assem)
1267 ioctl(mdfd, STOP_ARRAY, NULL);
1268 close(mdfd);
1269 return 1;
1270 } else {
1271 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1272 * been updated to point to the current locations of devices.
1273 * so we can just start the array
1274 */
1275 unsigned long dev;
1276 dev = makedev(devices[chosen_drive].i.disk.major,
1277 devices[chosen_drive].i.disk.minor);
1278 if (ioctl(mdfd, START_ARRAY, dev)) {
1279 fprintf(stderr, Name ": Cannot start array: %s\n",
1280 strerror(errno));
1281 }
1282
1283 }
1284 close(mdfd);
1285 return 0;
1286 }
1287
1288 #ifndef MDASSEMBLE
1289 int assemble_container_content(struct supertype *st, int mdfd,
1290 struct mdinfo *content, int runstop,
1291 char *chosen_name, int verbose)
1292 {
1293 struct mdinfo *dev, *sra;
1294 int working = 0, preexist = 0;
1295 struct map_ent *map = NULL;
1296
1297 sysfs_init(content, mdfd, 0);
1298
1299 sra = sysfs_read(mdfd, 0, GET_VERSION);
1300 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1301 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1302 close(mdfd);
1303 return 1;
1304 }
1305 if (sra)
1306 sysfs_free(sra);
1307
1308 for (dev = content->devs; dev; dev = dev->next)
1309 if (sysfs_add_disk(content, dev, 1) == 0)
1310 working++;
1311 else if (errno == EEXIST)
1312 preexist++;
1313 if (working == 0) {
1314 close(mdfd);
1315 return 1;/* Nothing new, don't try to start */
1316 }
1317
1318 map_update(&map, fd2devnum(mdfd),
1319 content->text_version,
1320 content->uuid, chosen_name);
1321
1322 if (runstop > 0 ||
1323 (working + preexist) >= content->array.working_disks) {
1324 int err;
1325
1326 switch(content->array.level) {
1327 case LEVEL_LINEAR:
1328 case LEVEL_MULTIPATH:
1329 case 0:
1330 err = sysfs_set_str(content, NULL, "array_state",
1331 "active");
1332 break;
1333 default:
1334 err = sysfs_set_str(content, NULL, "array_state",
1335 "readonly");
1336 /* start mdmon if needed. */
1337 if (!err) {
1338 if (!mdmon_running(st->container_dev))
1339 start_mdmon(st->container_dev);
1340 ping_monitor(devnum2devname(st->container_dev));
1341 }
1342 break;
1343 }
1344 if (!err)
1345 sysfs_set_safemode(content, content->safe_mode_delay);
1346 if (verbose >= 0) {
1347 if (err)
1348 fprintf(stderr, Name
1349 ": array %s now has %d devices",
1350 chosen_name, working + preexist);
1351 else
1352 fprintf(stderr, Name
1353 ": Started %s with %d devices",
1354 chosen_name, working + preexist);
1355 if (preexist)
1356 fprintf(stderr, " (%d new)", working);
1357 fprintf(stderr, "\n");
1358 }
1359 if (!err)
1360 wait_for(chosen_name, mdfd);
1361 close(mdfd);
1362 return 0;
1363 /* FIXME should have an O_EXCL and wait for read-auto */
1364 } else {
1365 if (verbose >= 0)
1366 fprintf(stderr, Name
1367 ": %s assembled with %d devices but "
1368 "not started\n",
1369 chosen_name, working);
1370 close(mdfd);
1371 return 1;
1372 }
1373 }
1374 #endif
1375