]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Assemble.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Remove avail_disks arg from 'enough'.
[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 static int ident_matches(struct mddev_ident *ident,
74 struct mdinfo *content,
75 struct supertype *tst,
76 char *homehost,
77 char *update, char *devname)
78 {
79
80 if (ident->uuid_set && (!update || strcmp(update, "uuid")!= 0) &&
81 same_uuid(content->uuid, ident->uuid, tst->ss->swapuuid)==0 &&
82 memcmp(content->uuid, uuid_zero, sizeof(int[4])) != 0) {
83 if (devname)
84 fprintf(stderr, Name ": %s has wrong uuid.\n",
85 devname);
86 return 0;
87 }
88 if (ident->name[0] && (!update || strcmp(update, "name")!= 0) &&
89 name_matches(content->name, ident->name, homehost)==0) {
90 if (devname)
91 fprintf(stderr, Name ": %s has wrong name.\n",
92 devname);
93 return 0;
94 }
95 if (ident->super_minor != UnSet &&
96 ident->super_minor != content->array.md_minor) {
97 if (devname)
98 fprintf(stderr, Name ": %s has wrong super-minor.\n",
99 devname);
100 return 0;
101 }
102 if (ident->level != UnSet &&
103 ident->level != content->array.level) {
104 if (devname)
105 fprintf(stderr, Name ": %s has wrong raid level.\n",
106 devname);
107 return 0;
108 }
109 if (ident->raid_disks != UnSet &&
110 ident->raid_disks!= content->array.raid_disks) {
111 if (devname)
112 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
113 devname);
114 return 0;
115 }
116 if (ident->member && ident->member[0]) {
117 /* content->text_version must match */
118 char *s = strchr(content->text_version+1, '/');
119 if (s == NULL) {
120 if (devname)
121 fprintf(stderr, Name ": %s is not a container and one is required.\n",
122 devname);
123 return 0;
124 } else if (strcmp(ident->member, s+1) != 0) {
125 if (devname)
126 fprintf(stderr, Name ": skipping wrong member %s is %s\n",
127 content->text_version, devname);
128 return 0;
129 }
130 }
131 return 1;
132 }
133
134
135 int Assemble(struct supertype *st, char *mddev,
136 struct mddev_ident *ident,
137 struct mddev_dev *devlist,
138 char *backup_file, int invalid_backup,
139 int readonly, int runstop,
140 char *update, char *homehost, int require_homehost,
141 int verbose, int force, int freeze_reshape)
142 {
143 /*
144 * The task of Assemble is to find a collection of
145 * devices that should (according to their superblocks)
146 * form an array, and to give this collection to the MD driver.
147 * In Linux-2.4 and later, this involves submitting a
148 * SET_ARRAY_INFO ioctl with no arg - to prepare
149 * the array - and then submit a number of
150 * ADD_NEW_DISK ioctls to add disks into
151 * the array. Finally RUN_ARRAY might
152 * be submitted to start the array.
153 *
154 * Much of the work of Assemble is in finding and/or
155 * checking the disks to make sure they look right.
156 *
157 * If mddev is not set, then scan must be set and we
158 * read through the config file for dev+uuid mapping
159 * We recurse, setting mddev, for each device that
160 * - isn't running
161 * - has a valid uuid (or any uuid if !uuidset)
162 *
163 * If mddev is set, we try to determine state of md.
164 * check version - must be at least 0.90.0
165 * check kernel version. must be at least 2.4.
166 * If not, we can possibly fall back on START_ARRAY
167 * Try to GET_ARRAY_INFO.
168 * If possible, give up
169 * If not, try to STOP_ARRAY just to make sure
170 *
171 * If !uuidset and scan, look in conf-file for uuid
172 * If not found, give up
173 * If !devlist and scan and uuidset, get list of devs from conf-file
174 *
175 * For each device:
176 * Check superblock - discard if bad
177 * Check uuid (set if we don't have one) - discard if no match
178 * Check superblock similarity if we have a superblock - discard if different
179 * Record events, devicenum
180 * This should give us a list of devices for the array
181 * We should collect the most recent event number
182 *
183 * Count disks with recent enough event count
184 * While force && !enough disks
185 * Choose newest rejected disks, update event count
186 * mark clean and rewrite superblock
187 * If recent kernel:
188 * SET_ARRAY_INFO
189 * foreach device with recent events : ADD_NEW_DISK
190 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
191 * If old kernel:
192 * Check the device numbers in superblock are right
193 * update superblock if any changes
194 * START_ARRAY
195 *
196 */
197 int mdfd;
198 int clean;
199 int auto_assem = (mddev == NULL && !ident->uuid_set &&
200 ident->super_minor == UnSet && ident->name[0] == 0
201 && (ident->container == NULL || ident->member == NULL));
202 int old_linux = 0;
203 int vers = vers; /* Keep gcc quite - it really is initialised */
204 struct {
205 char *devname;
206 int uptodate; /* set once we decide that this device is as
207 * recent as everything else in the array.
208 */
209 struct mdinfo i;
210 } *devices;
211 char *devmap;
212 int *best = NULL; /* indexed by raid_disk */
213 int bestcnt = 0;
214 int devcnt = 0;
215 unsigned int okcnt, sparecnt, rebuilding_cnt;
216 unsigned int req_cnt;
217 int i;
218 int most_recent = 0;
219 int chosen_drive;
220 int change = 0;
221 int inargv = 0;
222 int report_missmatch;
223 #ifndef MDASSEMBLE
224 int bitmap_done;
225 #endif
226 int start_partial_ok = (runstop >= 0) &&
227 (force || devlist==NULL || auto_assem);
228 unsigned int num_devs;
229 struct mddev_dev *tmpdev;
230 struct mdinfo info;
231 struct mdinfo *content = NULL;
232 char *avail;
233 int nextspare = 0;
234 char *name = NULL;
235 int trustworthy;
236 char chosen_name[1024];
237 struct domainlist *domains = NULL;
238
239 if (get_linux_version() < 2004000)
240 old_linux = 1;
241
242 /*
243 * If any subdevs are listed, then any that don't
244 * match ident are discarded. Remainder must all match and
245 * become the array.
246 * If no subdevs, then we scan all devices in the config file, but
247 * there must be something in the identity
248 */
249
250 if (!devlist &&
251 ident->uuid_set == 0 &&
252 (ident->super_minor < 0 || ident->super_minor == UnSet) &&
253 ident->name[0] == 0 &&
254 (ident->container == NULL || ident->member == NULL) &&
255 ident->devices == NULL) {
256 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
257 mddev ? mddev : "further assembly");
258 return 1;
259 }
260
261 if (devlist == NULL)
262 devlist = conf_get_devs();
263 else if (mddev)
264 inargv = 1;
265
266 report_missmatch = ((inargv && verbose >= 0) || verbose > 0);
267 try_again:
268 /* We come back here when doing auto-assembly and attempting some
269 * set of devices failed. Those are now marked as ->used==2 and
270 * we ignore them and try again
271 */
272
273 tmpdev = devlist; num_devs = 0;
274 while (tmpdev) {
275 if (tmpdev->used)
276 tmpdev->used = 2;
277 else
278 num_devs++;
279 tmpdev = tmpdev->next;
280 }
281
282 if (!st && ident->st) st = ident->st;
283
284 if (verbose>0)
285 fprintf(stderr, Name ": looking for devices for %s\n",
286 mddev ? mddev : "further assembly");
287
288 /* first walk the list of devices to find a consistent set
289 * that match the criterea, if that is possible.
290 * We flag the ones we like with 'used'.
291 */
292 for (tmpdev = devlist;
293 tmpdev;
294 tmpdev = tmpdev ? tmpdev->next : NULL) {
295 char *devname = tmpdev->devname;
296 int dfd;
297 struct stat stb;
298 struct supertype *tst;
299 struct dev_policy *pol = NULL;
300 int found_container = 0;
301
302 if (tmpdev->used > 1) continue;
303
304 if (ident->devices &&
305 !match_oneof(ident->devices, devname)) {
306 if (report_missmatch)
307 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
308 continue;
309 }
310
311 tst = dup_super(st);
312
313 dfd = dev_open(devname, O_RDONLY|O_EXCL);
314 if (dfd < 0) {
315 if (report_missmatch)
316 fprintf(stderr, Name ": cannot open device %s: %s\n",
317 devname, strerror(errno));
318 tmpdev->used = 2;
319 } else if (fstat(dfd, &stb)< 0) {
320 /* Impossible! */
321 fprintf(stderr, Name ": fstat failed for %s: %s\n",
322 devname, strerror(errno));
323 tmpdev->used = 2;
324 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
325 fprintf(stderr, Name ": %s is not a block device.\n",
326 devname);
327 tmpdev->used = 2;
328 } else if (must_be_container(dfd)) {
329 if (st) {
330 /* already found some components, this cannot
331 * be another one.
332 */
333 if (report_missmatch)
334 fprintf(stderr, Name ": %s is a container, but we are looking for components\n",
335 devname);
336 tmpdev->used = 2;
337 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
338 } if (!tst && (tst = super_by_fd(dfd, NULL)) == NULL) {
339 if (report_missmatch)
340 fprintf(stderr, Name ": not a recognisable container: %s\n",
341 devname);
342 tmpdev->used = 2;
343 #endif
344 } else if (!tst->ss->load_container
345 || tst->ss->load_container(tst, dfd, NULL)) {
346 if (report_missmatch)
347 fprintf(stderr, Name ": no correct container type: %s\n",
348 devname);
349 tmpdev->used = 2;
350 } else if (auto_assem &&
351 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
352 tst->ss->match_home(tst, homehost) == 1)) {
353 if (report_missmatch)
354 fprintf(stderr, Name ": %s has metadata type %s for which "
355 "auto-assembly is disabled\n",
356 devname, tst->ss->name);
357 tmpdev->used = 2;
358 } else
359 found_container = 1;
360 } else {
361 if (!tst && (tst = guess_super(dfd)) == NULL) {
362 if (report_missmatch)
363 fprintf(stderr, Name ": no recogniseable superblock on %s\n",
364 devname);
365 tmpdev->used = 2;
366 } else if (tst->ss->load_super(tst,dfd, NULL)) {
367 if (report_missmatch)
368 fprintf(stderr, Name ": no RAID superblock on %s\n",
369 devname);
370 tmpdev->used = 2;
371 } else if (tst->ss->compare_super == NULL) {
372 if (report_missmatch)
373 fprintf(stderr, Name ": Cannot assemble %s metadata on %s\n",
374 tst->ss->name, devname);
375 tmpdev->used = 2;
376 } else if (auto_assem && st == NULL &&
377 !conf_test_metadata(tst->ss->name, (pol = devnum_policy(stb.st_rdev)),
378 tst->ss->match_home(tst, homehost) == 1)) {
379 if (report_missmatch)
380 fprintf(stderr, Name ": %s has metadata type %s for which "
381 "auto-assembly is disabled\n",
382 devname, tst->ss->name);
383 tmpdev->used = 2;
384 }
385 }
386 if (dfd >= 0) close(dfd);
387 if (tmpdev->used == 2) {
388 if (auto_assem || !inargv)
389 /* Ignore unrecognised devices during auto-assembly */
390 goto loop;
391 if (ident->uuid_set || ident->name[0] ||
392 ident->super_minor != UnSet)
393 /* Ignore unrecognised device if looking for
394 * specific array */
395 goto loop;
396
397
398 fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
399 devname);
400 if (st)
401 st->ss->free_super(st);
402 dev_policy_free(pol);
403 domain_free(domains);
404 return 1;
405 }
406
407 if (found_container) {
408 /* tmpdev is a container. We need to be either
409 * looking for a member, or auto-assembling
410 */
411
412 if (ident->container) {
413 if (ident->container[0] == '/' &&
414 !same_dev(ident->container, devname)) {
415 if (report_missmatch)
416 fprintf(stderr, Name ": %s is not the container required (%s)\n",
417 devname, ident->container);
418 goto loop;
419 }
420 if (ident->container[0] != '/') {
421 /* we have a uuid */
422 int uuid[4];
423
424 content = &info;
425 tst->ss->getinfo_super(tst, content, NULL);
426
427 if (!parse_uuid(ident->container, uuid) ||
428 !same_uuid(content->uuid, uuid, tst->ss->swapuuid)) {
429 if (report_missmatch)
430 fprintf(stderr, Name ": %s has wrong UUID to be required container\n",
431 devname);
432 goto loop;
433 }
434 }
435 }
436 /* It is worth looking inside this container.
437 */
438 if (verbose > 0)
439 fprintf(stderr, Name ": looking in container %s\n",
440 devname);
441
442 for (content = tst->ss->container_content(tst, NULL);
443 content;
444 content = content->next) {
445
446 if (!ident_matches(ident, content, tst,
447 homehost, update,
448 report_missmatch ? devname : NULL))
449 /* message already printed */;
450 else if (is_member_busy(content->text_version)) {
451 if (report_missmatch)
452 fprintf(stderr, Name ": member %s in %s is already assembled\n",
453 content->text_version,
454 devname);
455 } else if (content->array.state & (1<<MD_SB_BLOCK_VOLUME)) {
456 /* do not assemble arrays with unsupported configurations */
457 fprintf(stderr, Name ": Cannot activate member %s in %s.\n",
458 content->text_version,
459 devname);
460 } else
461 break;
462 }
463 if (!content) {
464 tmpdev->used = 2;
465 goto loop; /* empty container */
466 }
467
468 st = tst; tst = NULL;
469 if (!auto_assem && inargv && tmpdev->next != NULL) {
470 fprintf(stderr, Name ": %s is a container, but is not "
471 "only device given: confused and aborting\n",
472 devname);
473 st->ss->free_super(st);
474 dev_policy_free(pol);
475 domain_free(domains);
476 return 1;
477 }
478 if (verbose > 0)
479 fprintf(stderr, Name ": found match on member %s in %s\n",
480 content->text_version, devname);
481
482 /* make sure we finished the loop */
483 tmpdev = NULL;
484 goto loop;
485 } else {
486
487 content = &info;
488 tst->ss->getinfo_super(tst, content, NULL);
489
490 if (!ident_matches(ident, content, tst,
491 homehost, update,
492 report_missmatch ? devname : NULL))
493 goto loop;
494
495 if (st == NULL)
496 st = dup_super(tst);
497 if (st->minor_version == -1)
498 st->minor_version = tst->minor_version;
499
500 if (memcmp(content->uuid, uuid_zero,
501 sizeof(int[4])) == 0) {
502 /* this is a floating spare. It cannot define
503 * an array unless there are no more arrays of
504 * this type to be found. It can be included
505 * in an array of this type though.
506 */
507 tmpdev->used = 3;
508 goto loop;
509 }
510
511 if (st->ss != tst->ss ||
512 st->minor_version != tst->minor_version ||
513 st->ss->compare_super(st, tst) != 0) {
514 /* Some mismatch. If exactly one array matches this host,
515 * we can resolve on that one.
516 * Or, if we are auto assembling, we just ignore the second
517 * for now.
518 */
519 if (auto_assem)
520 goto loop;
521 if (homehost) {
522 int first = st->ss->match_home(st, homehost);
523 int last = tst->ss->match_home(tst, homehost);
524 if (first != last &&
525 (first == 1 || last == 1)) {
526 /* We can do something */
527 if (first) {/* just ignore this one */
528 if (report_missmatch)
529 fprintf(stderr, Name ": %s misses out due to wrong homehost\n",
530 devname);
531 goto loop;
532 } else { /* reject all those sofar */
533 struct mddev_dev *td;
534 if (report_missmatch)
535 fprintf(stderr, Name ": %s overrides previous devices due to good homehost\n",
536 devname);
537 for (td=devlist; td != tmpdev; td=td->next)
538 if (td->used == 1)
539 td->used = 0;
540 tmpdev->used = 1;
541 goto loop;
542 }
543 }
544 }
545 fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
546 devname);
547 tst->ss->free_super(tst);
548 st->ss->free_super(st);
549 dev_policy_free(pol);
550 domain_free(domains);
551 return 1;
552 }
553 tmpdev->used = 1;
554 }
555 loop:
556 /* Collect domain information from members only */
557 if (tmpdev && tmpdev->used == 1) {
558 if (!pol)
559 pol = devnum_policy(stb.st_rdev);
560 domain_merge(&domains, pol, tst?tst->ss->name:NULL);
561 }
562 dev_policy_free(pol);
563 pol = NULL;
564 if (tst)
565 tst->ss->free_super(tst);
566 }
567
568 /* Check if we found some imsm spares but no members */
569 if ((auto_assem ||
570 (ident->uuid_set &&
571 memcmp(uuid_zero, ident->uuid,sizeof(uuid_zero)) == 0)) &&
572 (!st || !st->sb))
573 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
574 if (tmpdev->used != 3)
575 continue;
576 tmpdev->used = 1;
577 content = &info;
578
579 if (!st->sb) {
580 /* we need sb from one of the spares */
581 int dfd = dev_open(tmpdev->devname, O_RDONLY);
582 if (dfd < 0 ||
583 st->ss->load_super(st, dfd, NULL))
584 tmpdev->used = 2;
585 if (dfd > 0)
586 close(dfd);
587 }
588 }
589
590 /* Now reject spares that don't match domains of identified members */
591 for (tmpdev = devlist; tmpdev; tmpdev = tmpdev->next) {
592 struct stat stb;
593 if (tmpdev->used != 3)
594 continue;
595 if (stat(tmpdev->devname, &stb)< 0) {
596 fprintf(stderr, Name ": fstat failed for %s: %s\n",
597 tmpdev->devname, strerror(errno));
598 tmpdev->used = 2;
599 } else {
600 struct dev_policy *pol = devnum_policy(stb.st_rdev);
601 int dt = domain_test(domains, pol, NULL);
602 if (inargv && dt != 0)
603 /* take this spare as domains match
604 * if there are any */
605 tmpdev->used = 1;
606 else if (!inargv && dt == 1)
607 /* device wasn't explicitly listed, so need
608 * explicit domain match - which we have */
609 tmpdev->used = 1;
610 else
611 /* if domains don't match mark as unused */
612 tmpdev->used = 0;
613 dev_policy_free(pol);
614 }
615 }
616 domain_free(domains);
617
618 if (!st || !st->sb || !content)
619 return 2;
620
621 /* Now need to open the array device. Use create_mddev */
622 if (content == &info)
623 st->ss->getinfo_super(st, content, NULL);
624
625 trustworthy = FOREIGN;
626 name = content->name;
627 switch (st->ss->match_home(st, homehost)
628 ?: st->ss->match_home(st, "any")) {
629 case 1:
630 trustworthy = LOCAL;
631 name = strchr(content->name, ':');
632 if (name)
633 name++;
634 else
635 name = content->name;
636 break;
637 }
638 if (!auto_assem)
639 /* If the array is listed in mdadm.conf or on
640 * command line, then we trust the name
641 * even if the array doesn't look local
642 */
643 trustworthy = LOCAL;
644
645 if (name[0] == 0 &&
646 content->array.level == LEVEL_CONTAINER) {
647 name = content->text_version;
648 trustworthy = METADATA;
649 }
650
651 if (name[0] && trustworthy != LOCAL &&
652 ! require_homehost &&
653 conf_name_is_free(name))
654 trustworthy = LOCAL;
655
656 if (trustworthy == LOCAL &&
657 strchr(name, ':'))
658 /* Ignore 'host:' prefix of name */
659 name = strchr(name, ':')+1;
660
661 mdfd = create_mddev(mddev, name, ident->autof, trustworthy,
662 chosen_name);
663 if (mdfd < 0) {
664 st->ss->free_super(st);
665 if (auto_assem)
666 goto try_again;
667 return 1;
668 }
669 mddev = chosen_name;
670 vers = md_get_version(mdfd);
671 if (vers < 9000) {
672 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
673 " Upgrade your kernel or try --build\n");
674 close(mdfd);
675 return 1;
676 }
677 if (mddev_busy(fd2devnum(mdfd))) {
678 fprintf(stderr, Name ": %s already active, cannot restart it!\n",
679 mddev);
680 for (tmpdev = devlist ;
681 tmpdev && tmpdev->used != 1;
682 tmpdev = tmpdev->next)
683 ;
684 if (tmpdev && auto_assem)
685 fprintf(stderr, Name ": %s needed for %s...\n",
686 mddev, tmpdev->devname);
687 close(mdfd);
688 mdfd = -3;
689 st->ss->free_super(st);
690 if (auto_assem)
691 goto try_again;
692 return 1;
693 }
694 ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
695
696 #ifndef MDASSEMBLE
697 if (content != &info) {
698 /* This is a member of a container. Try starting the array. */
699 int err;
700 err = assemble_container_content(st, mdfd, content, runstop,
701 chosen_name, verbose,
702 backup_file, freeze_reshape);
703 close(mdfd);
704 return err;
705 }
706 bitmap_done = 0;
707 #endif
708 /* Ok, no bad inconsistancy, we can try updating etc */
709 devices = malloc(num_devs * sizeof(*devices));
710 devmap = calloc(num_devs * content->array.raid_disks, 1);
711 for (tmpdev = devlist; tmpdev; tmpdev=tmpdev->next) if (tmpdev->used == 1) {
712 char *devname = tmpdev->devname;
713 struct stat stb;
714 /* looks like a good enough match to update the super block if needed */
715 #ifndef MDASSEMBLE
716 if (update) {
717 int dfd;
718 /* prepare useful information in info structures */
719 struct stat stb2;
720 struct supertype *tst;
721 int err;
722 fstat(mdfd, &stb2);
723
724 if (strcmp(update, "uuid")==0 &&
725 !ident->uuid_set) {
726 int rfd;
727 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
728 read(rfd, ident->uuid, 16) != 16) {
729 *(__u32*)(ident->uuid) = random();
730 *(__u32*)(ident->uuid+1) = random();
731 *(__u32*)(ident->uuid+2) = random();
732 *(__u32*)(ident->uuid+3) = random();
733 }
734 if (rfd >= 0) close(rfd);
735 }
736 dfd = dev_open(devname, O_RDWR|O_EXCL);
737
738 tst = dup_super(st);
739 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
740 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
741 devname);
742 if (dfd >= 0)
743 close(dfd);
744 close(mdfd);
745 free(devices);
746 free(devmap);
747 return 1;
748 }
749 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
750
751 memcpy(content->uuid, ident->uuid, 16);
752 strcpy(content->name, ident->name);
753 content->array.md_minor = minor(stb2.st_rdev);
754
755 if (strcmp(update, "byteorder") == 0)
756 err = 0;
757 else
758 err = tst->ss->update_super(tst, content, update,
759 devname, verbose,
760 ident->uuid_set,
761 homehost);
762 if (err < 0) {
763 fprintf(stderr,
764 Name ": --update=%s not understood"
765 " for %s metadata\n",
766 update, tst->ss->name);
767 tst->ss->free_super(tst);
768 free(tst);
769 close(mdfd);
770 close(dfd);
771 free(devices);
772 free(devmap);
773 return 1;
774 }
775 if (strcmp(update, "uuid")==0 &&
776 !ident->uuid_set) {
777 ident->uuid_set = 1;
778 memcpy(ident->uuid, content->uuid, 16);
779 }
780 if (tst->ss->store_super(tst, dfd))
781 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
782 devname);
783 close(dfd);
784
785 if (strcmp(update, "uuid")==0 &&
786 ident->bitmap_fd >= 0 && !bitmap_done) {
787 if (bitmap_update_uuid(ident->bitmap_fd,
788 content->uuid,
789 tst->ss->swapuuid) != 0)
790 fprintf(stderr, Name ": Could not update uuid on external bitmap.\n");
791 else
792 bitmap_done = 1;
793 }
794 tst->ss->free_super(tst);
795 } else
796 #endif
797 {
798 struct supertype *tst = dup_super(st);
799 int dfd;
800 dfd = dev_open(devname, O_RDWR|O_EXCL);
801
802 if (dfd < 0 || tst->ss->load_super(tst, dfd, NULL) != 0) {
803 fprintf(stderr, Name ": cannot re-read metadata from %s - aborting\n",
804 devname);
805 if (dfd >= 0)
806 close(dfd);
807 close(mdfd);
808 free(devices);
809 free(devmap);
810 return 1;
811 }
812 tst->ss->getinfo_super(tst, content, devmap + devcnt * content->array.raid_disks);
813 tst->ss->free_super(tst);
814 close(dfd);
815 }
816
817 stat(devname, &stb);
818
819 if (verbose > 0)
820 fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
821 devname, mddev, content->disk.raid_disk);
822 devices[devcnt].devname = devname;
823 devices[devcnt].uptodate = 0;
824 devices[devcnt].i = *content;
825 devices[devcnt].i.disk.major = major(stb.st_rdev);
826 devices[devcnt].i.disk.minor = minor(stb.st_rdev);
827 if (most_recent < devcnt) {
828 if (devices[devcnt].i.events
829 > devices[most_recent].i.events)
830 most_recent = devcnt;
831 }
832 if (content->array.level == LEVEL_MULTIPATH)
833 /* with multipath, the raid_disk from the superblock is meaningless */
834 i = devcnt;
835 else
836 i = devices[devcnt].i.disk.raid_disk;
837 if (i+1 == 0) {
838 if (nextspare < content->array.raid_disks)
839 nextspare = content->array.raid_disks;
840 i = nextspare++;
841 } else {
842 if (i >= content->array.raid_disks &&
843 i >= nextspare)
844 nextspare = i+1;
845 }
846 if (i < 10000) {
847 if (i >= bestcnt) {
848 int newbestcnt = i+10;
849 int *newbest = malloc(sizeof(int)*newbestcnt);
850 int c;
851 for (c=0; c < newbestcnt; c++)
852 if (c < bestcnt)
853 newbest[c] = best[c];
854 else
855 newbest[c] = -1;
856 if (best)free(best);
857 best = newbest;
858 bestcnt = newbestcnt;
859 }
860 if (best[i] >=0 &&
861 devices[best[i]].i.events
862 == devices[devcnt].i.events
863 && (devices[best[i]].i.disk.minor
864 != devices[devcnt].i.disk.minor)
865 && st->ss == &super0
866 && content->array.level != LEVEL_MULTIPATH) {
867 /* two different devices with identical superblock.
868 * Could be a mis-detection caused by overlapping
869 * partitions. fail-safe.
870 */
871 fprintf(stderr, Name ": WARNING %s and %s appear"
872 " to have very similar superblocks.\n"
873 " If they are really different, "
874 "please --zero the superblock on one\n"
875 " If they are the same or overlap,"
876 " please remove one from %s.\n",
877 devices[best[i]].devname, devname,
878 inargv ? "the list" :
879 "the\n DEVICE list in mdadm.conf"
880 );
881 close(mdfd);
882 free(devices);
883 free(devmap);
884 return 1;
885 }
886 if (best[i] == -1
887 || (devices[best[i]].i.events
888 < devices[devcnt].i.events))
889 best[i] = devcnt;
890 }
891 devcnt++;
892 }
893
894 if (devcnt == 0) {
895 fprintf(stderr, Name ": no devices found for %s\n",
896 mddev);
897 if (st)
898 st->ss->free_super(st);
899 close(mdfd);
900 free(devices);
901 free(devmap);
902 return 1;
903 }
904
905 if (update && strcmp(update, "byteorder")==0)
906 st->minor_version = 90;
907
908 st->ss->getinfo_super(st, content, NULL);
909 clean = content->array.state & 1;
910
911 /* now we have some devices that might be suitable.
912 * I wonder how many
913 */
914 avail = malloc(content->array.raid_disks);
915 memset(avail, 0, content->array.raid_disks);
916 okcnt = 0;
917 sparecnt=0;
918 rebuilding_cnt=0;
919 for (i=0; i< bestcnt; i++) {
920 int j = best[i];
921 int event_margin = 1; /* always allow a difference of '1'
922 * like the kernel does
923 */
924 if (j < 0) continue;
925 /* note: we ignore error flags in multipath arrays
926 * as they don't make sense
927 */
928 if (content->array.level != LEVEL_MULTIPATH)
929 if (!(devices[j].i.disk.state & (1<<MD_DISK_ACTIVE))) {
930 if (!(devices[j].i.disk.state
931 & (1<<MD_DISK_FAULTY))) {
932 devices[j].uptodate = 1;
933 sparecnt++;
934 }
935 continue;
936 }
937 /* If this device thinks that 'most_recent' has failed, then
938 * we must reject this device.
939 */
940 if (j != most_recent &&
941 content->array.raid_disks > 0 &&
942 devices[most_recent].i.disk.raid_disk >= 0 &&
943 devmap[j * content->array.raid_disks + devices[most_recent].i.disk.raid_disk] == 0) {
944 if (verbose > -1)
945 fprintf(stderr, Name ": ignoring %s as it reports %s as failed\n",
946 devices[j].devname, devices[most_recent].devname);
947 best[i] = -1;
948 continue;
949 }
950 if (devices[j].i.events+event_margin >=
951 devices[most_recent].i.events) {
952 devices[j].uptodate = 1;
953 if (i < content->array.raid_disks) {
954 if (devices[j].i.recovery_start == MaxSector ||
955 (content->reshape_active &&
956 j >= content->array.raid_disks - content->delta_disks)) {
957 okcnt++;
958 avail[i]=1;
959 } else
960 rebuilding_cnt++;
961 } else
962 sparecnt++;
963 }
964 }
965 free(devmap);
966 while (force && !enough(content->array.level, content->array.raid_disks,
967 content->array.layout, 1,
968 avail)) {
969 /* Choose the newest best drive which is
970 * not up-to-date, update the superblock
971 * and add it.
972 */
973 int fd;
974 struct supertype *tst;
975 unsigned long long current_events;
976 chosen_drive = -1;
977 for (i = 0; i < content->array.raid_disks && i < bestcnt; i++) {
978 int j = best[i];
979 if (j>=0 &&
980 !devices[j].uptodate &&
981 devices[j].i.recovery_start == MaxSector &&
982 (chosen_drive < 0 ||
983 devices[j].i.events
984 > devices[chosen_drive].i.events))
985 chosen_drive = j;
986 }
987 if (chosen_drive < 0)
988 break;
989 current_events = devices[chosen_drive].i.events;
990 add_another:
991 if (verbose >= 0)
992 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
993 devices[chosen_drive].devname,
994 devices[chosen_drive].i.disk.raid_disk,
995 (int)(devices[chosen_drive].i.events),
996 (int)(devices[most_recent].i.events));
997 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
998 if (fd < 0) {
999 fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
1000 devices[chosen_drive].devname);
1001 devices[chosen_drive].i.events = 0;
1002 continue;
1003 }
1004 tst = dup_super(st);
1005 if (tst->ss->load_super(tst,fd, NULL)) {
1006 close(fd);
1007 fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
1008 devices[chosen_drive].devname);
1009 devices[chosen_drive].i.events = 0;
1010 continue;
1011 }
1012 content->events = devices[most_recent].i.events;
1013 tst->ss->update_super(tst, content, "force-one",
1014 devices[chosen_drive].devname, verbose,
1015 0, NULL);
1016
1017 if (tst->ss->store_super(tst, fd)) {
1018 close(fd);
1019 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1020 devices[chosen_drive].devname);
1021 devices[chosen_drive].i.events = 0;
1022 tst->ss->free_super(tst);
1023 continue;
1024 }
1025 close(fd);
1026 devices[chosen_drive].i.events = devices[most_recent].i.events;
1027 devices[chosen_drive].uptodate = 1;
1028 avail[chosen_drive] = 1;
1029 okcnt++;
1030 tst->ss->free_super(tst);
1031
1032 /* If there are any other drives of the same vintage,
1033 * add them in as well. We can't lose and we might gain
1034 */
1035 for (i = 0; i < content->array.raid_disks && i < bestcnt ; i++) {
1036 int j = best[i];
1037 if (j >= 0 &&
1038 !devices[j].uptodate &&
1039 devices[j].i.events == current_events) {
1040 chosen_drive = j;
1041 goto add_another;
1042 }
1043 }
1044 }
1045
1046 /* Now we want to look at the superblock which the kernel will base things on
1047 * and compare the devices that we think are working with the devices that the
1048 * superblock thinks are working.
1049 * If there are differences and --force is given, then update this chosen
1050 * superblock.
1051 */
1052 chosen_drive = -1;
1053 st->ss->free_super(st);
1054 for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
1055 int j = best[i];
1056 int fd;
1057
1058 if (j<0)
1059 continue;
1060 if (!devices[j].uptodate)
1061 continue;
1062 if (devices[j].i.events < devices[most_recent].i.events)
1063 continue;
1064 chosen_drive = j;
1065 if ((fd=dev_open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
1066 fprintf(stderr, Name ": Cannot open %s: %s\n",
1067 devices[j].devname, strerror(errno));
1068 close(mdfd);
1069 free(devices);
1070 return 1;
1071 }
1072 if (st->ss->load_super(st,fd, NULL)) {
1073 close(fd);
1074 fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
1075 devices[j].devname);
1076 close(mdfd);
1077 free(devices);
1078 return 1;
1079 }
1080 close(fd);
1081 }
1082 if (st->sb == NULL) {
1083 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
1084 close(mdfd);
1085 free(devices);
1086 return 1;
1087 }
1088 st->ss->getinfo_super(st, content, NULL);
1089 #ifndef MDASSEMBLE
1090 sysfs_init(content, mdfd, 0);
1091 #endif
1092 for (i=0; i<bestcnt; i++) {
1093 int j = best[i];
1094 unsigned int desired_state;
1095
1096 if (i < content->array.raid_disks)
1097 desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
1098 else
1099 desired_state = 0;
1100
1101 if (j<0)
1102 continue;
1103 if (!devices[j].uptodate)
1104 continue;
1105
1106 devices[j].i.disk.state = desired_state;
1107 if (!(devices[j].i.array.state & 1))
1108 clean = 0;
1109
1110 if (st->ss->update_super(st, &devices[j].i, "assemble", NULL,
1111 verbose, 0, NULL)) {
1112 if (force) {
1113 if (verbose >= 0)
1114 fprintf(stderr, Name ": "
1115 "clearing FAULTY flag for device %d in %s for %s\n",
1116 j, mddev, devices[j].devname);
1117 change = 1;
1118 } else {
1119 if (verbose >= -1)
1120 fprintf(stderr, Name ": "
1121 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1122 i, mddev, devices[j].devname);
1123 }
1124 }
1125 #if 0
1126 if (!(super.disks[i].i.disk.state & (1 << MD_DISK_FAULTY))) {
1127 fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1128 i, mddev);
1129 }
1130 #endif
1131 }
1132 if (force && !clean &&
1133 !enough(content->array.level, content->array.raid_disks,
1134 content->array.layout, clean,
1135 avail)) {
1136 change += st->ss->update_super(st, content, "force-array",
1137 devices[chosen_drive].devname, verbose,
1138 0, NULL);
1139 clean = 1;
1140 }
1141
1142 if (change) {
1143 int fd;
1144 fd = dev_open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
1145 if (fd < 0) {
1146 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1147 devices[chosen_drive].devname);
1148 close(mdfd);
1149 free(devices);
1150 return 1;
1151 }
1152 if (st->ss->store_super(st, fd)) {
1153 close(fd);
1154 fprintf(stderr, Name ": Could not re-write superblock on %s\n",
1155 devices[chosen_drive].devname);
1156 close(mdfd);
1157 free(devices);
1158 return 1;
1159 }
1160 if (verbose >= 0)
1161 fprintf(stderr, Name ": Marking array %s as 'clean'\n",
1162 mddev);
1163 close(fd);
1164 }
1165
1166 /* If we are in the middle of a reshape we may need to restore saved data
1167 * that was moved aside due to the reshape overwriting live data
1168 * The code of doing this lives in Grow.c
1169 */
1170 #ifndef MDASSEMBLE
1171 if (content->reshape_active) {
1172 int err = 0;
1173 int *fdlist = malloc(sizeof(int)* bestcnt);
1174 if (verbose > 0)
1175 fprintf(stderr, Name ":%s has an active reshape - checking "
1176 "if critical section needs to be restored\n",
1177 chosen_name);
1178 for (i=0; i<bestcnt; i++) {
1179 int j = best[i];
1180 if (j >= 0) {
1181 fdlist[i] = dev_open(devices[j].devname, O_RDWR|O_EXCL);
1182 if (fdlist[i] < 0) {
1183 fprintf(stderr, Name ": Could not open %s for write - cannot Assemble array.\n",
1184 devices[j].devname);
1185 err = 1;
1186 break;
1187 }
1188 } else
1189 fdlist[i] = -1;
1190 }
1191 if (!err) {
1192 if (st->ss->external && st->ss->recover_backup)
1193 err = st->ss->recover_backup(st, content);
1194 else
1195 err = Grow_restart(st, content, fdlist, bestcnt,
1196 backup_file, verbose > 0);
1197 if (err && invalid_backup) {
1198 if (verbose > 0)
1199 fprintf(stderr, Name ": continuing"
1200 " without restoring backup\n");
1201 err = 0;
1202 }
1203 }
1204 while (i>0) {
1205 i--;
1206 if (fdlist[i]>=0) close(fdlist[i]);
1207 }
1208 if (err) {
1209 fprintf(stderr, Name ": Failed to restore critical section for reshape, sorry.\n");
1210 if (backup_file == NULL)
1211 fprintf(stderr," Possibly you needed to specify the --backup-file\n");
1212 close(mdfd);
1213 free(devices);
1214 return err;
1215 }
1216 }
1217 #endif
1218 /* count number of in-sync devices according to the superblock.
1219 * We must have this number to start the array without -s or -R
1220 */
1221 req_cnt = content->array.working_disks;
1222
1223 /* Almost ready to actually *do* something */
1224 if (!old_linux) {
1225 int rv;
1226
1227 /* First, fill in the map, so that udev can find our name
1228 * as soon as we become active.
1229 */
1230 map_update(NULL, fd2devnum(mdfd), content->text_version,
1231 content->uuid, chosen_name);
1232
1233 rv = set_array_info(mdfd, st, content);
1234 if (rv) {
1235 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
1236 mddev, strerror(errno));
1237 ioctl(mdfd, STOP_ARRAY, NULL);
1238 close(mdfd);
1239 free(devices);
1240 return 1;
1241 }
1242 if (ident->bitmap_fd >= 0) {
1243 if (ioctl(mdfd, SET_BITMAP_FILE, ident->bitmap_fd) != 0) {
1244 fprintf(stderr, Name ": SET_BITMAP_FILE failed.\n");
1245 ioctl(mdfd, STOP_ARRAY, NULL);
1246 close(mdfd);
1247 free(devices);
1248 return 1;
1249 }
1250 } else if (ident->bitmap_file) {
1251 /* From config file */
1252 int bmfd = open(ident->bitmap_file, O_RDWR);
1253 if (bmfd < 0) {
1254 fprintf(stderr, Name ": Could not open bitmap file %s\n",
1255 ident->bitmap_file);
1256 ioctl(mdfd, STOP_ARRAY, NULL);
1257 close(mdfd);
1258 free(devices);
1259 return 1;
1260 }
1261 if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
1262 fprintf(stderr, Name ": Failed to set bitmapfile for %s\n", mddev);
1263 close(bmfd);
1264 ioctl(mdfd, STOP_ARRAY, NULL);
1265 close(mdfd);
1266 free(devices);
1267 return 1;
1268 }
1269 close(bmfd);
1270 }
1271
1272 /* First, add the raid disks, but add the chosen one last */
1273 for (i=0; i<= bestcnt; i++) {
1274 int j;
1275 if (i < bestcnt) {
1276 j = best[i];
1277 if (j == chosen_drive)
1278 continue;
1279 } else
1280 j = chosen_drive;
1281
1282 if (j >= 0 /* && devices[j].uptodate */) {
1283 int dfd = dev_open(devices[j].devname,
1284 O_RDWR|O_EXCL);
1285 if (dfd >= 0) {
1286 remove_partitions(dfd);
1287 close(dfd);
1288 }
1289 rv = add_disk(mdfd, st, content, &devices[j].i);
1290
1291 if (rv) {
1292 fprintf(stderr, Name ": failed to add "
1293 "%s to %s: %s\n",
1294 devices[j].devname,
1295 mddev,
1296 strerror(errno));
1297 if (i < content->array.raid_disks
1298 || i == bestcnt)
1299 okcnt--;
1300 else
1301 sparecnt--;
1302 } else if (verbose > 0)
1303 fprintf(stderr, Name ": added %s "
1304 "to %s as %d\n",
1305 devices[j].devname, mddev,
1306 devices[j].i.disk.raid_disk);
1307 } else if (verbose > 0 && i < content->array.raid_disks)
1308 fprintf(stderr, Name ": no uptodate device for "
1309 "slot %d of %s\n",
1310 i, mddev);
1311 }
1312
1313 if (content->array.level == LEVEL_CONTAINER) {
1314 if (verbose >= 0) {
1315 fprintf(stderr, Name ": Container %s has been "
1316 "assembled with %d drive%s",
1317 mddev, okcnt+sparecnt, okcnt+sparecnt==1?"":"s");
1318 if (okcnt < (unsigned)content->array.raid_disks)
1319 fprintf(stderr, " (out of %d)",
1320 content->array.raid_disks);
1321 fprintf(stderr, "\n");
1322 }
1323 st->ss->free_super(st);
1324 sysfs_uevent(content, "change");
1325 wait_for(chosen_name, mdfd);
1326 close(mdfd);
1327 free(devices);
1328 return 0;
1329 }
1330
1331 if (runstop == 1 ||
1332 (runstop <= 0 &&
1333 ( enough(content->array.level, content->array.raid_disks,
1334 content->array.layout, clean, avail) &&
1335 (okcnt + rebuilding_cnt >= req_cnt || start_partial_ok)
1336 ))) {
1337 /* This array is good-to-go.
1338 * If a reshape is in progress then we might need to
1339 * continue monitoring it. In that case we start
1340 * it read-only and let the grow code make it writable.
1341 */
1342 int rv;
1343 #ifndef MDASSEMBLE
1344 if (content->reshape_active &&
1345 content->delta_disks <= 0) {
1346 rv = sysfs_set_str(content, NULL,
1347 "array_state", "readonly");
1348 if (rv == 0)
1349 rv = Grow_continue(mdfd, st, content,
1350 backup_file,
1351 freeze_reshape);
1352 } else
1353 #endif
1354 rv = ioctl(mdfd, RUN_ARRAY, NULL);
1355 if (rv == 0) {
1356 if (verbose >= 0) {
1357 fprintf(stderr, Name ": %s has been started with %d drive%s",
1358 mddev, okcnt, okcnt==1?"":"s");
1359 if (okcnt < (unsigned)content->array.raid_disks)
1360 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1361 if (rebuilding_cnt)
1362 fprintf(stderr, "%s %d rebuilding", sparecnt?",":" and", rebuilding_cnt);
1363 if (sparecnt)
1364 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1365 fprintf(stderr, ".\n");
1366 }
1367 if (content->reshape_active &&
1368 content->array.level >= 4 &&
1369 content->array.level <= 6) {
1370 /* might need to increase the size
1371 * of the stripe cache - default is 256
1372 */
1373 if (256 < 4 * (content->array.chunk_size/4096)) {
1374 struct mdinfo *sra = sysfs_read(mdfd, 0, 0);
1375 if (sra)
1376 sysfs_set_num(sra, NULL,
1377 "stripe_cache_size",
1378 (4 * content->array.chunk_size / 4096) + 1);
1379 sysfs_free(sra);
1380 }
1381 }
1382 if (okcnt < (unsigned)content->array.raid_disks) {
1383 /* If any devices did not get added
1384 * because the kernel rejected them based
1385 * on event count, try adding them
1386 * again providing the action policy is
1387 * 're-add' or greater. The bitmap
1388 * might allow them to be included, or
1389 * they will become spares.
1390 */
1391 for (i = 0; i < bestcnt; i++) {
1392 int j = best[i];
1393 if (j >= 0 && !devices[j].uptodate) {
1394 if (!disk_action_allows(&devices[j].i, st->ss->name, act_re_add))
1395 continue;
1396 rv = add_disk(mdfd, st, content,
1397 &devices[j].i);
1398 if (rv == 0 && verbose >= 0)
1399 fprintf(stderr,
1400 Name ": %s has been re-added.\n",
1401 devices[j].devname);
1402 }
1403 }
1404 }
1405 wait_for(mddev, mdfd);
1406 close(mdfd);
1407 if (auto_assem) {
1408 int usecs = 1;
1409 /* There is a nasty race with 'mdadm --monitor'.
1410 * If it opens this device before we close it,
1411 * it gets an incomplete open on which IO
1412 * doesn't work and the capacity is
1413 * wrong.
1414 * If we reopen (to check for layered devices)
1415 * before --monitor closes, we loose.
1416 *
1417 * So: wait upto 1 second for there to be
1418 * a non-zero capacity.
1419 */
1420 while (usecs < 1000) {
1421 mdfd = open(mddev, O_RDONLY);
1422 if (mdfd >= 0) {
1423 unsigned long long size;
1424 if (get_dev_size(mdfd, NULL, &size) &&
1425 size > 0)
1426 break;
1427 close(mdfd);
1428 }
1429 usleep(usecs);
1430 usecs <<= 1;
1431 }
1432 }
1433 free(devices);
1434 return 0;
1435 }
1436 fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
1437 mddev, strerror(errno));
1438
1439 if (!enough(content->array.level, content->array.raid_disks,
1440 content->array.layout, 1, avail))
1441 fprintf(stderr, Name ": Not enough devices to "
1442 "start the array.\n");
1443 else if (!enough(content->array.level,
1444 content->array.raid_disks,
1445 content->array.layout, clean,
1446 avail))
1447 fprintf(stderr, Name ": Not enough devices to "
1448 "start the array while not clean "
1449 "- consider --force.\n");
1450
1451 if (auto_assem)
1452 ioctl(mdfd, STOP_ARRAY, NULL);
1453 close(mdfd);
1454 free(devices);
1455 return 1;
1456 }
1457 if (runstop == -1) {
1458 fprintf(stderr, Name ": %s assembled from %d drive%s",
1459 mddev, okcnt, okcnt==1?"":"s");
1460 if (okcnt != (unsigned)content->array.raid_disks)
1461 fprintf(stderr, " (out of %d)", content->array.raid_disks);
1462 fprintf(stderr, ", but not started.\n");
1463 close(mdfd);
1464 free(devices);
1465 return 0;
1466 }
1467 if (verbose >= -1) {
1468 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
1469 if (rebuilding_cnt)
1470 fprintf(stderr, "%s %d rebuilding", sparecnt?", ":" and ", rebuilding_cnt);
1471 if (sparecnt)
1472 fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
1473 if (!enough(content->array.level, content->array.raid_disks,
1474 content->array.layout, 1, avail))
1475 fprintf(stderr, " - not enough to start the array.\n");
1476 else if (!enough(content->array.level,
1477 content->array.raid_disks,
1478 content->array.layout, clean,
1479 avail))
1480 fprintf(stderr, " - not enough to start the "
1481 "array while not clean - consider "
1482 "--force.\n");
1483 else {
1484 if (req_cnt == (unsigned)content->array.raid_disks)
1485 fprintf(stderr, " - need all %d to start it", req_cnt);
1486 else
1487 fprintf(stderr, " - need %d of %d to start", req_cnt, content->array.raid_disks);
1488 fprintf(stderr, " (use --run to insist).\n");
1489 }
1490 }
1491 if (auto_assem)
1492 ioctl(mdfd, STOP_ARRAY, NULL);
1493 close(mdfd);
1494 free(devices);
1495 return 1;
1496 } else {
1497 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1498 * been updated to point to the current locations of devices.
1499 * so we can just start the array
1500 */
1501 unsigned long dev;
1502 dev = makedev(devices[chosen_drive].i.disk.major,
1503 devices[chosen_drive].i.disk.minor);
1504 if (ioctl(mdfd, START_ARRAY, dev)) {
1505 fprintf(stderr, Name ": Cannot start array: %s\n",
1506 strerror(errno));
1507 }
1508
1509 }
1510 close(mdfd);
1511 free(devices);
1512 return 0;
1513 }
1514
1515 #ifndef MDASSEMBLE
1516 int assemble_container_content(struct supertype *st, int mdfd,
1517 struct mdinfo *content, int runstop,
1518 char *chosen_name, int verbose,
1519 char *backup_file, int freeze_reshape)
1520 {
1521 struct mdinfo *dev, *sra;
1522 int working = 0, preexist = 0;
1523 int expansion = 0;
1524 struct map_ent *map = NULL;
1525 int old_raid_disks;
1526
1527 sysfs_init(content, mdfd, 0);
1528
1529 sra = sysfs_read(mdfd, 0, GET_VERSION);
1530 if (sra == NULL || strcmp(sra->text_version, content->text_version) != 0)
1531 if (sysfs_set_array(content, md_get_version(mdfd)) != 0) {
1532 if (sra)
1533 sysfs_free(sra);
1534 return 1;
1535 }
1536
1537 if (st->ss->external && content->recovery_blocked)
1538 block_subarray(content);
1539
1540 if (sra)
1541 sysfs_free(sra);
1542 old_raid_disks = content->array.raid_disks - content->delta_disks;
1543 for (dev = content->devs; dev; dev = dev->next)
1544 if (sysfs_add_disk(content, dev, 1) == 0) {
1545 if (dev->disk.raid_disk >= old_raid_disks &&
1546 content->reshape_active)
1547 expansion++;
1548 else
1549 working++;
1550 } else if (errno == EEXIST)
1551 preexist++;
1552 if (working == 0)
1553 return 1;/* Nothing new, don't try to start */
1554
1555 map_update(&map, fd2devnum(mdfd),
1556 content->text_version,
1557 content->uuid, chosen_name);
1558
1559 if (runstop > 0 ||
1560 (working + preexist + expansion) >=
1561 content->array.working_disks) {
1562 int err;
1563 int start_reshape;
1564
1565 /* There are two types of reshape: container wide or sub-array specific
1566 * Check if metadata requests blocking container wide reshapes
1567 */
1568 start_reshape = (content->reshape_active &&
1569 !((content->reshape_active == CONTAINER_RESHAPE) &&
1570 (content->array.state & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))));
1571 if (start_reshape) {
1572 int spare = content->array.raid_disks + expansion;
1573 if (restore_backup(st, content,
1574 working,
1575 spare, backup_file, verbose) == 1)
1576 return 1;
1577
1578 err = sysfs_set_str(content, NULL,
1579 "array_state", "readonly");
1580 if (err)
1581 return 1;
1582
1583 if (st->ss->external) {
1584 if (!mdmon_running(st->container_dev))
1585 start_mdmon(st->container_dev);
1586 ping_monitor_by_id(st->container_dev);
1587 if (mdmon_running(st->container_dev) &&
1588 st->update_tail == NULL)
1589 st->update_tail = &st->updates;
1590 }
1591
1592 err = Grow_continue(mdfd, st, content, backup_file,
1593 freeze_reshape);
1594 } else switch(content->array.level) {
1595 case LEVEL_LINEAR:
1596 case LEVEL_MULTIPATH:
1597 case 0:
1598 err = sysfs_set_str(content, NULL, "array_state",
1599 "active");
1600 break;
1601 default:
1602 err = sysfs_set_str(content, NULL, "array_state",
1603 "readonly");
1604 /* start mdmon if needed. */
1605 if (!err) {
1606 if (!mdmon_running(st->container_dev))
1607 start_mdmon(st->container_dev);
1608 ping_monitor_by_id(st->container_dev);
1609 }
1610 break;
1611 }
1612 if (!err)
1613 sysfs_set_safemode(content, content->safe_mode_delay);
1614 if (verbose >= 0) {
1615 if (err)
1616 fprintf(stderr, Name
1617 ": array %s now has %d device%s",
1618 chosen_name, working + preexist,
1619 working + preexist == 1 ? "":"s");
1620 else
1621 fprintf(stderr, Name
1622 ": Started %s with %d device%s",
1623 chosen_name, working + preexist,
1624 working + preexist == 1 ? "":"s");
1625 if (preexist)
1626 fprintf(stderr, " (%d new)", working);
1627 if (expansion)
1628 fprintf(stderr, " ( + %d for expansion)",
1629 expansion);
1630 fprintf(stderr, "\n");
1631 }
1632 if (!err)
1633 wait_for(chosen_name, mdfd);
1634 return err;
1635 /* FIXME should have an O_EXCL and wait for read-auto */
1636 } else {
1637 if (verbose >= 0) {
1638 fprintf(stderr, Name
1639 ": %s assembled with %d device%s",
1640 chosen_name, preexist + working,
1641 preexist + working == 1 ? "":"s");
1642 if (preexist)
1643 fprintf(stderr, " (%d new)", working);
1644 fprintf(stderr, " but not started\n");
1645 }
1646 return 1;
1647 }
1648 }
1649 #endif
1650
Unnamed repository; edit this file 'description' to name the repository.