2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
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.
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.
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
22 * Email: <neilb@suse.de>
28 static int name_matches(char *found
, char *required
, char *homehost
)
30 /* See if the name found matches the required name, possibly
31 * prefixed with 'homehost'
35 strncpy(fnd
, found
, 32);
37 if (strcmp(found
, required
)==0)
40 int l
= strlen(homehost
);
41 if (l
< 32 && fnd
[l
] == ':' &&
42 strcmp(fnd
+l
+1, required
)==0)
48 static int is_member_busy(char *metadata_version
)
50 /* check if the given member array is active */
51 struct mdstat_ent
*mdstat
= mdstat_read(1, 0);
52 struct mdstat_ent
*ent
;
55 for (ent
= mdstat
; ent
; ent
= ent
->next
) {
56 if (ent
->metadata_version
== NULL
)
58 if (strncmp(ent
->metadata_version
, "external:", 9) != 0)
60 if (!is_subarray(&ent
->metadata_version
[9]))
62 /* Skip first char - it can be '/' or '-' */
63 if (strcmp(&ent
->metadata_version
[10], metadata_version
+1) == 0) {
73 static int ident_matches(struct mddev_ident
*ident
,
74 struct mdinfo
*content
,
75 struct supertype
*tst
,
77 char *update
, char *devname
)
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) {
84 fprintf(stderr
, Name
": %s has wrong uuid.\n",
88 if (ident
->name
[0] && (!update
|| strcmp(update
, "name")!= 0) &&
89 name_matches(content
->name
, ident
->name
, homehost
)==0) {
91 fprintf(stderr
, Name
": %s has wrong name.\n",
95 if (ident
->super_minor
!= UnSet
&&
96 ident
->super_minor
!= content
->array
.md_minor
) {
98 fprintf(stderr
, Name
": %s has wrong super-minor.\n",
102 if (ident
->level
!= UnSet
&&
103 ident
->level
!= content
->array
.level
) {
105 fprintf(stderr
, Name
": %s has wrong raid level.\n",
109 if (ident
->raid_disks
!= UnSet
&&
110 ident
->raid_disks
!= content
->array
.raid_disks
) {
112 fprintf(stderr
, Name
": %s requires wrong number of drives.\n",
116 if (ident
->member
&& ident
->member
[0]) {
117 /* content->text_version must match */
118 char *s
= strchr(content
->text_version
+1, '/');
121 fprintf(stderr
, Name
": %s is not a container and one is required.\n",
124 } else if (strcmp(ident
->member
, s
+1) != 0) {
126 fprintf(stderr
, Name
": skipping wrong member %s is %s\n",
127 content
->text_version
, devname
);
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
)
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.
154 * Much of the work of Assemble is in finding and/or
155 * checking the disks to make sure they look right.
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
161 * - has a valid uuid (or any uuid if !uuidset)
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
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
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
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
189 * foreach device with recent events : ADD_NEW_DISK
190 * if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
192 * Check the device numbers in superblock are right
193 * update superblock if any changes
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
));
203 int vers
= vers
; /* Keep gcc quite - it really is initialised */
206 int uptodate
; /* set once we decide that this device is as
207 * recent as everything else in the array.
212 int *best
= NULL
; /* indexed by raid_disk */
215 unsigned int okcnt
, sparecnt
, rebuilding_cnt
;
216 unsigned int req_cnt
;
222 int report_missmatch
;
226 int start_partial_ok
= (runstop
>= 0) &&
227 (force
|| devlist
==NULL
|| auto_assem
);
228 unsigned int num_devs
;
229 struct mddev_dev
*tmpdev
;
231 struct mdinfo
*content
= NULL
;
236 char chosen_name
[1024];
237 struct domainlist
*domains
= NULL
;
239 if (get_linux_version() < 2004000)
243 * If any subdevs are listed, then any that don't
244 * match ident are discarded. Remainder must all match and
246 * If no subdevs, then we scan all devices in the config file, but
247 * there must be something in the identity
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");
262 devlist
= conf_get_devs();
266 report_missmatch
= ((inargv
&& verbose
>= 0) || verbose
> 0);
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
273 tmpdev
= devlist
; num_devs
= 0;
279 tmpdev
= tmpdev
->next
;
282 if (!st
&& ident
->st
) st
= ident
->st
;
285 fprintf(stderr
, Name
": looking for devices for %s\n",
286 mddev
? mddev
: "further assembly");
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'.
292 for (tmpdev
= devlist
;
294 tmpdev
= tmpdev
? tmpdev
->next
: NULL
) {
295 char *devname
= tmpdev
->devname
;
298 struct supertype
*tst
;
299 struct dev_policy
*pol
= NULL
;
300 int found_container
= 0;
302 if (tmpdev
->used
> 1) continue;
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
);
313 dfd
= dev_open(devname
, O_RDONLY
);
315 if (report_missmatch
)
316 fprintf(stderr
, Name
": cannot open device %s: %s\n",
317 devname
, strerror(errno
));
319 } else if (fstat(dfd
, &stb
)< 0) {
321 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
322 devname
, strerror(errno
));
324 } else if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
325 fprintf(stderr
, Name
": %s is not a block device.\n",
328 } else if (must_be_container(dfd
)) {
330 /* already found some components, this cannot
333 if (report_missmatch
)
334 fprintf(stderr
, Name
": %s is a container, but we are looking for components\n",
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",
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",
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
);
361 if (!tst
&& (tst
= guess_super(dfd
)) == NULL
) {
362 if (report_missmatch
)
363 fprintf(stderr
, Name
": no recogniseable superblock on %s\n",
366 } else if (tst
->ss
->load_super(tst
,dfd
, NULL
)) {
367 if (report_missmatch
)
368 fprintf(stderr
, Name
": no RAID superblock on %s\n",
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
);
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
);
386 if (dfd
>= 0) close(dfd
);
387 if (tmpdev
->used
== 2) {
388 if (auto_assem
|| !inargv
)
389 /* Ignore unrecognised devices during auto-assembly */
391 if (ident
->uuid_set
|| ident
->name
[0] ||
392 ident
->super_minor
!= UnSet
)
393 /* Ignore unrecognised device if looking for
398 fprintf(stderr
, Name
": %s has no superblock - assembly aborted\n",
401 st
->ss
->free_super(st
);
402 dev_policy_free(pol
);
403 domain_free(domains
);
407 if (found_container
) {
408 /* tmpdev is a container. We need to be either
409 * looking for a member, or auto-assembling
411 /* should be safe to try an exclusive open now, we
412 * have rejected anything that some other mdadm might
415 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
417 if (report_missmatch
)
418 fprintf(stderr
, Name
": %s is busy - skipping\n", devname
);
423 if (ident
->container
) {
424 if (ident
->container
[0] == '/' &&
425 !same_dev(ident
->container
, devname
)) {
426 if (report_missmatch
)
427 fprintf(stderr
, Name
": %s is not the container required (%s)\n",
428 devname
, ident
->container
);
431 if (ident
->container
[0] != '/') {
436 tst
->ss
->getinfo_super(tst
, content
, NULL
);
438 if (!parse_uuid(ident
->container
, uuid
) ||
439 !same_uuid(content
->uuid
, uuid
, tst
->ss
->swapuuid
)) {
440 if (report_missmatch
)
441 fprintf(stderr
, Name
": %s has wrong UUID to be required container\n",
447 /* It is worth looking inside this container.
450 fprintf(stderr
, Name
": looking in container %s\n",
453 for (content
= tst
->ss
->container_content(tst
, NULL
);
455 content
= content
->next
) {
457 if (!ident_matches(ident
, content
, tst
,
459 report_missmatch
? devname
: NULL
))
460 /* message already printed */;
461 else if (is_member_busy(content
->text_version
)) {
462 if (report_missmatch
)
463 fprintf(stderr
, Name
": member %s in %s is already assembled\n",
464 content
->text_version
,
466 } else if (content
->array
.state
& (1<<MD_SB_BLOCK_VOLUME
)) {
467 /* do not assemble arrays with unsupported configurations */
468 fprintf(stderr
, Name
": Cannot activate member %s in %s.\n",
469 content
->text_version
,
476 goto loop
; /* empty container */
479 st
= tst
; tst
= NULL
;
480 if (!auto_assem
&& inargv
&& tmpdev
->next
!= NULL
) {
481 fprintf(stderr
, Name
": %s is a container, but is not "
482 "only device given: confused and aborting\n",
484 st
->ss
->free_super(st
);
485 dev_policy_free(pol
);
486 domain_free(domains
);
490 fprintf(stderr
, Name
": found match on member %s in %s\n",
491 content
->text_version
, devname
);
493 /* make sure we finished the loop */
499 tst
->ss
->getinfo_super(tst
, content
, NULL
);
501 if (!ident_matches(ident
, content
, tst
,
503 report_missmatch
? devname
: NULL
))
506 /* should be safe to try an exclusive open now, we
507 * have rejected anything that some other mdadm might
510 dfd
= dev_open(devname
, O_RDONLY
| O_EXCL
);
512 if (report_missmatch
)
513 fprintf(stderr
, Name
": %s is busy - skipping\n", devname
);
520 if (st
->minor_version
== -1)
521 st
->minor_version
= tst
->minor_version
;
523 if (memcmp(content
->uuid
, uuid_zero
,
524 sizeof(int[4])) == 0) {
525 /* this is a floating spare. It cannot define
526 * an array unless there are no more arrays of
527 * this type to be found. It can be included
528 * in an array of this type though.
534 if (st
->ss
!= tst
->ss
||
535 st
->minor_version
!= tst
->minor_version
||
536 st
->ss
->compare_super(st
, tst
) != 0) {
537 /* Some mismatch. If exactly one array matches this host,
538 * we can resolve on that one.
539 * Or, if we are auto assembling, we just ignore the second
545 int first
= st
->ss
->match_home(st
, homehost
);
546 int last
= tst
->ss
->match_home(tst
, homehost
);
548 (first
== 1 || last
== 1)) {
549 /* We can do something */
550 if (first
) {/* just ignore this one */
551 if (report_missmatch
)
552 fprintf(stderr
, Name
": %s misses out due to wrong homehost\n",
555 } else { /* reject all those sofar */
556 struct mddev_dev
*td
;
557 if (report_missmatch
)
558 fprintf(stderr
, Name
": %s overrides previous devices due to good homehost\n",
560 for (td
=devlist
; td
!= tmpdev
; td
=td
->next
)
568 fprintf(stderr
, Name
": superblock on %s doesn't match others - assembly aborted\n",
570 tst
->ss
->free_super(tst
);
571 st
->ss
->free_super(st
);
572 dev_policy_free(pol
);
573 domain_free(domains
);
579 /* Collect domain information from members only */
580 if (tmpdev
&& tmpdev
->used
== 1) {
582 pol
= devnum_policy(stb
.st_rdev
);
583 domain_merge(&domains
, pol
, tst
?tst
->ss
->name
:NULL
);
585 dev_policy_free(pol
);
588 tst
->ss
->free_super(tst
);
591 /* Check if we found some imsm spares but no members */
594 memcmp(uuid_zero
, ident
->uuid
,sizeof(uuid_zero
)) == 0)) &&
596 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
597 if (tmpdev
->used
!= 3)
603 /* we need sb from one of the spares */
604 int dfd
= dev_open(tmpdev
->devname
, O_RDONLY
);
606 st
->ss
->load_super(st
, dfd
, NULL
))
613 /* Now reject spares that don't match domains of identified members */
614 for (tmpdev
= devlist
; tmpdev
; tmpdev
= tmpdev
->next
) {
616 if (tmpdev
->used
!= 3)
618 if (stat(tmpdev
->devname
, &stb
)< 0) {
619 fprintf(stderr
, Name
": fstat failed for %s: %s\n",
620 tmpdev
->devname
, strerror(errno
));
623 struct dev_policy
*pol
= devnum_policy(stb
.st_rdev
);
624 int dt
= domain_test(domains
, pol
, NULL
);
625 if (inargv
&& dt
!= 0)
626 /* take this spare as domains match
627 * if there are any */
629 else if (!inargv
&& dt
== 1)
630 /* device wasn't explicitly listed, so need
631 * explicit domain match - which we have */
634 /* if domains don't match mark as unused */
636 dev_policy_free(pol
);
639 domain_free(domains
);
641 if (!st
|| !st
->sb
|| !content
)
644 /* Now need to open the array device. Use create_mddev */
645 if (content
== &info
)
646 st
->ss
->getinfo_super(st
, content
, NULL
);
648 trustworthy
= FOREIGN
;
649 name
= content
->name
;
650 switch (st
->ss
->match_home(st
, homehost
)
651 ?: st
->ss
->match_home(st
, "any")) {
654 name
= strchr(content
->name
, ':');
658 name
= content
->name
;
662 /* If the array is listed in mdadm.conf or on
663 * command line, then we trust the name
664 * even if the array doesn't look local
669 content
->array
.level
== LEVEL_CONTAINER
) {
670 name
= content
->text_version
;
671 trustworthy
= METADATA
;
674 if (name
[0] && trustworthy
!= LOCAL
&&
675 ! require_homehost
&&
676 conf_name_is_free(name
))
679 if (trustworthy
== LOCAL
&&
681 /* Ignore 'host:' prefix of name */
682 name
= strchr(name
, ':')+1;
684 mdfd
= create_mddev(mddev
, name
, ident
->autof
, trustworthy
,
687 st
->ss
->free_super(st
);
693 vers
= md_get_version(mdfd
);
695 fprintf(stderr
, Name
": Assemble requires driver version 0.90.0 or later.\n"
696 " Upgrade your kernel or try --build\n");
700 if (mddev_busy(fd2devnum(mdfd
))) {
701 fprintf(stderr
, Name
": %s already active, cannot restart it!\n",
703 for (tmpdev
= devlist
;
704 tmpdev
&& tmpdev
->used
!= 1;
705 tmpdev
= tmpdev
->next
)
707 if (tmpdev
&& auto_assem
)
708 fprintf(stderr
, Name
": %s needed for %s...\n",
709 mddev
, tmpdev
->devname
);
712 st
->ss
->free_super(st
);
717 ioctl(mdfd
, STOP_ARRAY
, NULL
); /* just incase it was started but has no content */
720 if (content
!= &info
) {
721 /* This is a member of a container. Try starting the array. */
723 err
= assemble_container_content(st
, mdfd
, content
, runstop
,
724 chosen_name
, verbose
,
725 backup_file
, freeze_reshape
);
731 /* Ok, no bad inconsistancy, we can try updating etc */
732 devices
= malloc(num_devs
* sizeof(*devices
));
733 devmap
= calloc(num_devs
* content
->array
.raid_disks
, 1);
734 for (tmpdev
= devlist
; tmpdev
; tmpdev
=tmpdev
->next
) if (tmpdev
->used
== 1) {
735 char *devname
= tmpdev
->devname
;
737 /* looks like a good enough match to update the super block if needed */
741 /* prepare useful information in info structures */
743 struct supertype
*tst
;
747 if (strcmp(update
, "uuid")==0 &&
750 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
751 read(rfd
, ident
->uuid
, 16) != 16) {
752 *(__u32
*)(ident
->uuid
) = random();
753 *(__u32
*)(ident
->uuid
+1) = random();
754 *(__u32
*)(ident
->uuid
+2) = random();
755 *(__u32
*)(ident
->uuid
+3) = random();
757 if (rfd
>= 0) close(rfd
);
759 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
762 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
763 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
772 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
774 memcpy(content
->uuid
, ident
->uuid
, 16);
775 strcpy(content
->name
, ident
->name
);
776 content
->array
.md_minor
= minor(stb2
.st_rdev
);
778 if (strcmp(update
, "byteorder") == 0)
781 err
= tst
->ss
->update_super(tst
, content
, update
,
787 Name
": --update=%s not understood"
788 " for %s metadata\n",
789 update
, tst
->ss
->name
);
790 tst
->ss
->free_super(tst
);
798 if (strcmp(update
, "uuid")==0 &&
801 memcpy(ident
->uuid
, content
->uuid
, 16);
803 if (tst
->ss
->store_super(tst
, dfd
))
804 fprintf(stderr
, Name
": Could not re-write superblock on %s.\n",
808 if (strcmp(update
, "uuid")==0 &&
809 ident
->bitmap_fd
>= 0 && !bitmap_done
) {
810 if (bitmap_update_uuid(ident
->bitmap_fd
,
812 tst
->ss
->swapuuid
) != 0)
813 fprintf(stderr
, Name
": Could not update uuid on external bitmap.\n");
817 tst
->ss
->free_super(tst
);
821 struct supertype
*tst
= dup_super(st
);
823 dfd
= dev_open(devname
, O_RDWR
|O_EXCL
);
825 if (dfd
< 0 || tst
->ss
->load_super(tst
, dfd
, NULL
) != 0) {
826 fprintf(stderr
, Name
": cannot re-read metadata from %s - aborting\n",
835 tst
->ss
->getinfo_super(tst
, content
, devmap
+ devcnt
* content
->array
.raid_disks
);
836 tst
->ss
->free_super(tst
);
843 fprintf(stderr
, Name
": %s is identified as a member of %s, slot %d.\n",
844 devname
, mddev
, content
->disk
.raid_disk
);
845 devices
[devcnt
].devname
= devname
;
846 devices
[devcnt
].uptodate
= 0;
847 devices
[devcnt
].i
= *content
;
848 devices
[devcnt
].i
.disk
.major
= major(stb
.st_rdev
);
849 devices
[devcnt
].i
.disk
.minor
= minor(stb
.st_rdev
);
850 if (most_recent
< devcnt
) {
851 if (devices
[devcnt
].i
.events
852 > devices
[most_recent
].i
.events
)
853 most_recent
= devcnt
;
855 if (content
->array
.level
== LEVEL_MULTIPATH
)
856 /* with multipath, the raid_disk from the superblock is meaningless */
859 i
= devices
[devcnt
].i
.disk
.raid_disk
;
861 if (nextspare
< content
->array
.raid_disks
)
862 nextspare
= content
->array
.raid_disks
;
865 if (i
>= content
->array
.raid_disks
&&
871 int newbestcnt
= i
+10;
872 int *newbest
= malloc(sizeof(int)*newbestcnt
);
874 for (c
=0; c
< newbestcnt
; c
++)
876 newbest
[c
] = best
[c
];
881 bestcnt
= newbestcnt
;
884 devices
[best
[i
]].i
.events
885 == devices
[devcnt
].i
.events
886 && (devices
[best
[i
]].i
.disk
.minor
887 != devices
[devcnt
].i
.disk
.minor
)
889 && content
->array
.level
!= LEVEL_MULTIPATH
) {
890 /* two different devices with identical superblock.
891 * Could be a mis-detection caused by overlapping
892 * partitions. fail-safe.
894 fprintf(stderr
, Name
": WARNING %s and %s appear"
895 " to have very similar superblocks.\n"
896 " If they are really different, "
897 "please --zero the superblock on one\n"
898 " If they are the same or overlap,"
899 " please remove one from %s.\n",
900 devices
[best
[i
]].devname
, devname
,
901 inargv
? "the list" :
902 "the\n DEVICE list in mdadm.conf"
910 || (devices
[best
[i
]].i
.events
911 < devices
[devcnt
].i
.events
))
918 fprintf(stderr
, Name
": no devices found for %s\n",
921 st
->ss
->free_super(st
);
928 if (update
&& strcmp(update
, "byteorder")==0)
929 st
->minor_version
= 90;
931 st
->ss
->getinfo_super(st
, content
, NULL
);
932 clean
= content
->array
.state
& 1;
934 /* now we have some devices that might be suitable.
937 avail
= malloc(content
->array
.raid_disks
);
938 memset(avail
, 0, content
->array
.raid_disks
);
942 for (i
=0; i
< bestcnt
; i
++) {
944 int event_margin
= 1; /* always allow a difference of '1'
945 * like the kernel does
948 /* note: we ignore error flags in multipath arrays
949 * as they don't make sense
951 if (content
->array
.level
!= LEVEL_MULTIPATH
)
952 if (!(devices
[j
].i
.disk
.state
& (1<<MD_DISK_ACTIVE
))) {
953 if (!(devices
[j
].i
.disk
.state
954 & (1<<MD_DISK_FAULTY
))) {
955 devices
[j
].uptodate
= 1;
960 /* If this device thinks that 'most_recent' has failed, then
961 * we must reject this device.
963 if (j
!= most_recent
&&
964 content
->array
.raid_disks
> 0 &&
965 devices
[most_recent
].i
.disk
.raid_disk
>= 0 &&
966 devmap
[j
* content
->array
.raid_disks
+ devices
[most_recent
].i
.disk
.raid_disk
] == 0) {
968 fprintf(stderr
, Name
": ignoring %s as it reports %s as failed\n",
969 devices
[j
].devname
, devices
[most_recent
].devname
);
973 if (devices
[j
].i
.events
+event_margin
>=
974 devices
[most_recent
].i
.events
) {
975 devices
[j
].uptodate
= 1;
976 if (i
< content
->array
.raid_disks
) {
977 if (devices
[j
].i
.recovery_start
== MaxSector
||
978 (content
->reshape_active
&&
979 ((i
>= content
->array
.raid_disks
- content
->delta_disks
) ||
980 (i
>= content
->array
.raid_disks
- content
->delta_disks
- 1
981 && content
->array
.level
== 4)))) {
992 (!enough(content
->array
.level
, content
->array
.raid_disks
,
993 content
->array
.layout
, 1,
996 (content
->reshape_active
&& content
->delta_disks
> 0 &&
997 !enough(content
->array
.level
, (content
->array
.raid_disks
998 - content
->delta_disks
),
999 content
->new_layout
, 1,
1002 /* Choose the newest best drive which is
1003 * not up-to-date, update the superblock
1007 struct supertype
*tst
;
1008 unsigned long long current_events
;
1010 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
1013 !devices
[j
].uptodate
&&
1014 devices
[j
].i
.recovery_start
== MaxSector
&&
1015 (chosen_drive
< 0 ||
1017 > devices
[chosen_drive
].i
.events
))
1020 if (chosen_drive
< 0)
1022 current_events
= devices
[chosen_drive
].i
.events
;
1025 fprintf(stderr
, Name
": forcing event count in %s(%d) from %d upto %d\n",
1026 devices
[chosen_drive
].devname
,
1027 devices
[chosen_drive
].i
.disk
.raid_disk
,
1028 (int)(devices
[chosen_drive
].i
.events
),
1029 (int)(devices
[most_recent
].i
.events
));
1030 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1032 fprintf(stderr
, Name
": Couldn't open %s for write - not updating\n",
1033 devices
[chosen_drive
].devname
);
1034 devices
[chosen_drive
].i
.events
= 0;
1037 tst
= dup_super(st
);
1038 if (tst
->ss
->load_super(tst
,fd
, NULL
)) {
1040 fprintf(stderr
, Name
": RAID superblock disappeared from %s - not updating.\n",
1041 devices
[chosen_drive
].devname
);
1042 devices
[chosen_drive
].i
.events
= 0;
1045 content
->events
= devices
[most_recent
].i
.events
;
1046 tst
->ss
->update_super(tst
, content
, "force-one",
1047 devices
[chosen_drive
].devname
, verbose
,
1050 if (tst
->ss
->store_super(tst
, fd
)) {
1052 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1053 devices
[chosen_drive
].devname
);
1054 devices
[chosen_drive
].i
.events
= 0;
1055 tst
->ss
->free_super(tst
);
1059 devices
[chosen_drive
].i
.events
= devices
[most_recent
].i
.events
;
1060 devices
[chosen_drive
].uptodate
= 1;
1061 avail
[chosen_drive
] = 1;
1063 tst
->ss
->free_super(tst
);
1065 /* If there are any other drives of the same vintage,
1066 * add them in as well. We can't lose and we might gain
1068 for (i
= 0; i
< content
->array
.raid_disks
&& i
< bestcnt
; i
++) {
1071 !devices
[j
].uptodate
&&
1072 devices
[j
].i
.events
== current_events
) {
1079 /* Now we want to look at the superblock which the kernel will base things on
1080 * and compare the devices that we think are working with the devices that the
1081 * superblock thinks are working.
1082 * If there are differences and --force is given, then update this chosen
1086 st
->ss
->free_super(st
);
1087 for (i
=0; chosen_drive
< 0 && i
<bestcnt
; i
++) {
1093 if (!devices
[j
].uptodate
)
1095 if (devices
[j
].i
.events
< devices
[most_recent
].i
.events
)
1098 if ((fd
=dev_open(devices
[j
].devname
, O_RDONLY
|O_EXCL
))< 0) {
1099 fprintf(stderr
, Name
": Cannot open %s: %s\n",
1100 devices
[j
].devname
, strerror(errno
));
1105 if (st
->ss
->load_super(st
,fd
, NULL
)) {
1107 fprintf(stderr
, Name
": RAID superblock has disappeared from %s\n",
1108 devices
[j
].devname
);
1115 if (st
->sb
== NULL
) {
1116 fprintf(stderr
, Name
": No suitable drives found for %s\n", mddev
);
1121 st
->ss
->getinfo_super(st
, content
, NULL
);
1123 sysfs_init(content
, mdfd
, 0);
1125 for (i
=0; i
<bestcnt
; i
++) {
1127 unsigned int desired_state
;
1129 if (i
< content
->array
.raid_disks
)
1130 desired_state
= (1<<MD_DISK_ACTIVE
) | (1<<MD_DISK_SYNC
);
1136 if (!devices
[j
].uptodate
)
1139 devices
[j
].i
.disk
.state
= desired_state
;
1140 if (!(devices
[j
].i
.array
.state
& 1))
1143 if (st
->ss
->update_super(st
, &devices
[j
].i
, "assemble", NULL
,
1144 verbose
, 0, NULL
)) {
1147 fprintf(stderr
, Name
": "
1148 "clearing FAULTY flag for device %d in %s for %s\n",
1149 j
, mddev
, devices
[j
].devname
);
1153 fprintf(stderr
, Name
": "
1154 "device %d in %s has wrong state in superblock, but %s seems ok\n",
1155 i
, mddev
, devices
[j
].devname
);
1159 if (!(super
.disks
[i
].i
.disk
.state
& (1 << MD_DISK_FAULTY
))) {
1160 fprintf(stderr
, Name
": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
1165 if (force
&& !clean
&&
1166 !enough(content
->array
.level
, content
->array
.raid_disks
,
1167 content
->array
.layout
, clean
,
1169 change
+= st
->ss
->update_super(st
, content
, "force-array",
1170 devices
[chosen_drive
].devname
, verbose
,
1177 fd
= dev_open(devices
[chosen_drive
].devname
, O_RDWR
|O_EXCL
);
1179 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1180 devices
[chosen_drive
].devname
);
1185 if (st
->ss
->store_super(st
, fd
)) {
1187 fprintf(stderr
, Name
": Could not re-write superblock on %s\n",
1188 devices
[chosen_drive
].devname
);
1194 fprintf(stderr
, Name
": Marking array %s as 'clean'\n",
1199 /* If we are in the middle of a reshape we may need to restore saved data
1200 * that was moved aside due to the reshape overwriting live data
1201 * The code of doing this lives in Grow.c
1204 if (content
->reshape_active
) {
1206 int *fdlist
= malloc(sizeof(int)* bestcnt
);
1208 fprintf(stderr
, Name
":%s has an active reshape - checking "
1209 "if critical section needs to be restored\n",
1211 for (i
=0; i
<bestcnt
; i
++) {
1214 fdlist
[i
] = dev_open(devices
[j
].devname
, O_RDWR
|O_EXCL
);
1215 if (fdlist
[i
] < 0) {
1216 fprintf(stderr
, Name
": Could not open %s for write - cannot Assemble array.\n",
1217 devices
[j
].devname
);
1225 if (st
->ss
->external
&& st
->ss
->recover_backup
)
1226 err
= st
->ss
->recover_backup(st
, content
);
1228 err
= Grow_restart(st
, content
, fdlist
, bestcnt
,
1229 backup_file
, verbose
> 0);
1230 if (err
&& invalid_backup
) {
1232 fprintf(stderr
, Name
": continuing"
1233 " without restoring backup\n");
1239 if (fdlist
[i
]>=0) close(fdlist
[i
]);
1242 fprintf(stderr
, Name
": Failed to restore critical section for reshape, sorry.\n");
1243 if (backup_file
== NULL
)
1244 fprintf(stderr
," Possibly you needed to specify the --backup-file\n");
1251 /* count number of in-sync devices according to the superblock.
1252 * We must have this number to start the array without -s or -R
1254 req_cnt
= content
->array
.working_disks
;
1256 /* Almost ready to actually *do* something */
1260 /* First, fill in the map, so that udev can find our name
1261 * as soon as we become active.
1263 map_update(NULL
, fd2devnum(mdfd
), content
->text_version
,
1264 content
->uuid
, chosen_name
);
1266 rv
= set_array_info(mdfd
, st
, content
);
1268 fprintf(stderr
, Name
": failed to set array info for %s: %s\n",
1269 mddev
, strerror(errno
));
1270 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1275 if (ident
->bitmap_fd
>= 0) {
1276 if (ioctl(mdfd
, SET_BITMAP_FILE
, ident
->bitmap_fd
) != 0) {
1277 fprintf(stderr
, Name
": SET_BITMAP_FILE failed.\n");
1278 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1283 } else if (ident
->bitmap_file
) {
1284 /* From config file */
1285 int bmfd
= open(ident
->bitmap_file
, O_RDWR
);
1287 fprintf(stderr
, Name
": Could not open bitmap file %s\n",
1288 ident
->bitmap_file
);
1289 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1294 if (ioctl(mdfd
, SET_BITMAP_FILE
, bmfd
) != 0) {
1295 fprintf(stderr
, Name
": Failed to set bitmapfile for %s\n", mddev
);
1297 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1305 /* First, add the raid disks, but add the chosen one last */
1306 for (i
=0; i
<= bestcnt
; i
++) {
1310 if (j
== chosen_drive
)
1315 if (j
>= 0 /* && devices[j].uptodate */) {
1316 int dfd
= dev_open(devices
[j
].devname
,
1319 remove_partitions(dfd
);
1322 rv
= add_disk(mdfd
, st
, content
, &devices
[j
].i
);
1325 fprintf(stderr
, Name
": failed to add "
1330 if (i
< content
->array
.raid_disks
1335 } else if (verbose
> 0)
1336 fprintf(stderr
, Name
": added %s "
1338 devices
[j
].devname
, mddev
,
1339 devices
[j
].i
.disk
.raid_disk
,
1340 devices
[j
].uptodate
?"":
1341 " (possibly out of date)");
1342 } else if (verbose
> 0 && i
< content
->array
.raid_disks
)
1343 fprintf(stderr
, Name
": no uptodate device for "
1348 if (content
->array
.level
== LEVEL_CONTAINER
) {
1350 fprintf(stderr
, Name
": Container %s has been "
1351 "assembled with %d drive%s",
1352 mddev
, okcnt
+sparecnt
, okcnt
+sparecnt
==1?"":"s");
1353 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1354 fprintf(stderr
, " (out of %d)",
1355 content
->array
.raid_disks
);
1356 fprintf(stderr
, "\n");
1358 st
->ss
->free_super(st
);
1359 sysfs_uevent(content
, "change");
1360 wait_for(chosen_name
, mdfd
);
1368 ( enough(content
->array
.level
, content
->array
.raid_disks
,
1369 content
->array
.layout
, clean
, avail
) &&
1370 (okcnt
+ rebuilding_cnt
>= req_cnt
|| start_partial_ok
)
1372 /* This array is good-to-go.
1373 * If a reshape is in progress then we might need to
1374 * continue monitoring it. In that case we start
1375 * it read-only and let the grow code make it writable.
1379 if (content
->reshape_active
&&
1380 content
->delta_disks
<= 0) {
1381 rv
= sysfs_set_str(content
, NULL
,
1382 "array_state", "readonly");
1384 rv
= Grow_continue(mdfd
, st
, content
,
1389 rv
= ioctl(mdfd
, RUN_ARRAY
, NULL
);
1392 fprintf(stderr
, Name
": %s has been started with %d drive%s",
1393 mddev
, okcnt
, okcnt
==1?"":"s");
1394 if (okcnt
< (unsigned)content
->array
.raid_disks
)
1395 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1397 fprintf(stderr
, "%s %d rebuilding", sparecnt
?",":" and", rebuilding_cnt
);
1399 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1400 fprintf(stderr
, ".\n");
1402 if (content
->reshape_active
&&
1403 content
->array
.level
>= 4 &&
1404 content
->array
.level
<= 6) {
1405 /* might need to increase the size
1406 * of the stripe cache - default is 256
1408 if (256 < 4 * (content
->array
.chunk_size
/4096)) {
1409 struct mdinfo
*sra
= sysfs_read(mdfd
, 0, 0);
1411 sysfs_set_num(sra
, NULL
,
1412 "stripe_cache_size",
1413 (4 * content
->array
.chunk_size
/ 4096) + 1);
1417 if (okcnt
< (unsigned)content
->array
.raid_disks
) {
1418 /* If any devices did not get added
1419 * because the kernel rejected them based
1420 * on event count, try adding them
1421 * again providing the action policy is
1422 * 're-add' or greater. The bitmap
1423 * might allow them to be included, or
1424 * they will become spares.
1426 for (i
= 0; i
< bestcnt
; i
++) {
1428 if (j
>= 0 && !devices
[j
].uptodate
) {
1429 if (!disk_action_allows(&devices
[j
].i
, st
->ss
->name
, act_re_add
))
1431 rv
= add_disk(mdfd
, st
, content
,
1433 if (rv
== 0 && verbose
>= 0)
1435 Name
": %s has been re-added.\n",
1436 devices
[j
].devname
);
1440 wait_for(mddev
, mdfd
);
1444 /* There is a nasty race with 'mdadm --monitor'.
1445 * If it opens this device before we close it,
1446 * it gets an incomplete open on which IO
1447 * doesn't work and the capacity is
1449 * If we reopen (to check for layered devices)
1450 * before --monitor closes, we loose.
1452 * So: wait upto 1 second for there to be
1453 * a non-zero capacity.
1455 while (usecs
< 1000) {
1456 mdfd
= open(mddev
, O_RDONLY
);
1458 unsigned long long size
;
1459 if (get_dev_size(mdfd
, NULL
, &size
) &&
1471 fprintf(stderr
, Name
": failed to RUN_ARRAY %s: %s\n",
1472 mddev
, strerror(errno
));
1474 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1475 content
->array
.layout
, 1, avail
))
1476 fprintf(stderr
, Name
": Not enough devices to "
1477 "start the array.\n");
1478 else if (!enough(content
->array
.level
,
1479 content
->array
.raid_disks
,
1480 content
->array
.layout
, clean
,
1482 fprintf(stderr
, Name
": Not enough devices to "
1483 "start the array while not clean "
1484 "- consider --force.\n");
1487 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1492 if (runstop
== -1) {
1493 fprintf(stderr
, Name
": %s assembled from %d drive%s",
1494 mddev
, okcnt
, okcnt
==1?"":"s");
1495 if (okcnt
!= (unsigned)content
->array
.raid_disks
)
1496 fprintf(stderr
, " (out of %d)", content
->array
.raid_disks
);
1497 fprintf(stderr
, ", but not started.\n");
1502 if (verbose
>= -1) {
1503 fprintf(stderr
, Name
": %s assembled from %d drive%s", mddev
, okcnt
, okcnt
==1?"":"s");
1505 fprintf(stderr
, "%s %d rebuilding", sparecnt
?", ":" and ", rebuilding_cnt
);
1507 fprintf(stderr
, " and %d spare%s", sparecnt
, sparecnt
==1?"":"s");
1508 if (!enough(content
->array
.level
, content
->array
.raid_disks
,
1509 content
->array
.layout
, 1, avail
))
1510 fprintf(stderr
, " - not enough to start the array.\n");
1511 else if (!enough(content
->array
.level
,
1512 content
->array
.raid_disks
,
1513 content
->array
.layout
, clean
,
1515 fprintf(stderr
, " - not enough to start the "
1516 "array while not clean - consider "
1519 if (req_cnt
== (unsigned)content
->array
.raid_disks
)
1520 fprintf(stderr
, " - need all %d to start it", req_cnt
);
1522 fprintf(stderr
, " - need %d of %d to start", req_cnt
, content
->array
.raid_disks
);
1523 fprintf(stderr
, " (use --run to insist).\n");
1527 ioctl(mdfd
, STOP_ARRAY
, NULL
);
1532 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
1533 * been updated to point to the current locations of devices.
1534 * so we can just start the array
1537 dev
= makedev(devices
[chosen_drive
].i
.disk
.major
,
1538 devices
[chosen_drive
].i
.disk
.minor
);
1539 if (ioctl(mdfd
, START_ARRAY
, dev
)) {
1540 fprintf(stderr
, Name
": Cannot start array: %s\n",
1551 int assemble_container_content(struct supertype
*st
, int mdfd
,
1552 struct mdinfo
*content
, int runstop
,
1553 char *chosen_name
, int verbose
,
1554 char *backup_file
, int freeze_reshape
)
1556 struct mdinfo
*dev
, *sra
;
1557 int working
= 0, preexist
= 0;
1559 struct map_ent
*map
= NULL
;
1562 sysfs_init(content
, mdfd
, 0);
1564 sra
= sysfs_read(mdfd
, 0, GET_VERSION
);
1565 if (sra
== NULL
|| strcmp(sra
->text_version
, content
->text_version
) != 0)
1566 if (sysfs_set_array(content
, md_get_version(mdfd
)) != 0) {
1572 if (st
->ss
->external
&& content
->recovery_blocked
)
1573 block_subarray(content
);
1577 old_raid_disks
= content
->array
.raid_disks
- content
->delta_disks
;
1578 for (dev
= content
->devs
; dev
; dev
= dev
->next
)
1579 if (sysfs_add_disk(content
, dev
, 1) == 0) {
1580 if (dev
->disk
.raid_disk
>= old_raid_disks
&&
1581 content
->reshape_active
)
1585 } else if (errno
== EEXIST
)
1587 if (working
+ expansion
== 0)
1588 return 1;/* Nothing new, don't try to start */
1590 map_update(&map
, fd2devnum(mdfd
),
1591 content
->text_version
,
1592 content
->uuid
, chosen_name
);
1595 (working
+ preexist
+ expansion
) >=
1596 content
->array
.working_disks
) {
1600 /* There are two types of reshape: container wide or sub-array specific
1601 * Check if metadata requests blocking container wide reshapes
1603 start_reshape
= (content
->reshape_active
&&
1604 !((content
->reshape_active
== CONTAINER_RESHAPE
) &&
1605 (content
->array
.state
& (1<<MD_SB_BLOCK_CONTAINER_RESHAPE
))));
1606 if (start_reshape
) {
1607 int spare
= content
->array
.raid_disks
+ expansion
;
1608 if (restore_backup(st
, content
,
1610 spare
, backup_file
, verbose
) == 1)
1613 err
= sysfs_set_str(content
, NULL
,
1614 "array_state", "readonly");
1618 if (st
->ss
->external
) {
1619 if (!mdmon_running(st
->container_dev
))
1620 start_mdmon(st
->container_dev
);
1621 ping_monitor_by_id(st
->container_dev
);
1622 if (mdmon_running(st
->container_dev
) &&
1623 st
->update_tail
== NULL
)
1624 st
->update_tail
= &st
->updates
;
1627 err
= Grow_continue(mdfd
, st
, content
, backup_file
,
1629 } else switch(content
->array
.level
) {
1631 case LEVEL_MULTIPATH
:
1633 err
= sysfs_set_str(content
, NULL
, "array_state",
1637 err
= sysfs_set_str(content
, NULL
, "array_state",
1639 /* start mdmon if needed. */
1641 if (!mdmon_running(st
->container_dev
))
1642 start_mdmon(st
->container_dev
);
1643 ping_monitor_by_id(st
->container_dev
);
1648 sysfs_set_safemode(content
, content
->safe_mode_delay
);
1651 fprintf(stderr
, Name
1652 ": array %s now has %d device%s",
1653 chosen_name
, working
+ preexist
,
1654 working
+ preexist
== 1 ? "":"s");
1656 fprintf(stderr
, Name
1657 ": Started %s with %d device%s",
1658 chosen_name
, working
+ preexist
,
1659 working
+ preexist
== 1 ? "":"s");
1661 fprintf(stderr
, " (%d new)", working
);
1663 fprintf(stderr
, " ( + %d for expansion)",
1665 fprintf(stderr
, "\n");
1668 wait_for(chosen_name
, mdfd
);
1670 /* FIXME should have an O_EXCL and wait for read-auto */
1673 fprintf(stderr
, Name
1674 ": %s assembled with %d device%s",
1675 chosen_name
, preexist
+ working
,
1676 preexist
+ working
== 1 ? "":"s");
1678 fprintf(stderr
, " (%d new)", working
);
1679 fprintf(stderr
, " but not started\n");