]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Manage.c
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>
30 #define REGISTER_DEV _IO (MD_MAJOR, 1)
31 #define START_MD _IO (MD_MAJOR, 2)
32 #define STOP_MD _IO (MD_MAJOR, 3)
34 int Manage_ro(char *devname
, int fd
, int readonly
)
36 /* switch to readonly or rw
39 * first check that array is runing
40 * use RESTART_ARRAY_RW or STOP_ARRAY_RO
43 mdu_array_info_t array
;
49 if (md_get_version(fd
) < 9000) {
50 pr_err("need md driver version 0.90.0 or later\n");
54 /* If this is an externally-manage array, we need to modify the
55 * metadata_version so that mdmon doesn't undo our change.
57 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
59 mdi
->array
.major_version
== -1 &&
60 is_subarray(mdi
->text_version
)) {
62 strcpy(vers
, "external:");
63 strcat(vers
, mdi
->text_version
);
66 /* We set readonly ourselves. */
68 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
71 rv
= sysfs_set_str(mdi
, NULL
, "array_state", "readonly");
74 pr_err("failed to set readonly for %s: %s\n",
75 devname
, strerror(errno
));
77 vers
[9] = mdi
->text_version
[0];
78 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
84 /* We cannot set read/write - must signal mdmon */
86 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
88 cp
= strchr(vers
+10, '/');
91 ping_monitor(vers
+10);
92 if (mdi
->array
.level
<= 0)
93 sysfs_set_str(mdi
, NULL
, "array_state", "active");
98 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
99 pr_err("%s does not appear to be active.\n",
106 if (ioctl(fd
, STOP_ARRAY_RO
, NULL
)) {
107 pr_err("failed to set readonly for %s: %s\n",
108 devname
, strerror(errno
));
112 } else if (readonly
< 0) {
113 if (ioctl(fd
, RESTART_ARRAY_RW
, NULL
)) {
114 pr_err("failed to set writable for %s: %s\n",
115 devname
, strerror(errno
));
130 static void remove_devices(int devnum
, char *path
)
133 * Remove names at 'path' - possibly with
134 * partition suffixes - which link to the 'standard'
135 * name for devnum. These were probably created
136 * by mdadm when the array was assembled.
150 sprintf(base
, "/dev/md%d", devnum
);
152 sprintf(base
, "/dev/md_d%d", -1-devnum
);
153 be
= base
+ strlen(base
);
155 path2
= xmalloc(strlen(path
)+20);
157 pe
= path2
+ strlen(path2
);
159 for (part
= 0; part
< 16; part
++) {
161 sprintf(be
, "p%d", part
);
164 sprintf(pe
, "p%d", part
);
166 sprintf(pe
, "%d", part
);
168 n
= readlink(path2
, link
, sizeof(link
));
169 if (n
> 0 && (int)strlen(base
) == n
&&
170 strncmp(link
, base
, n
) == 0)
177 int Manage_runstop(char *devname
, int fd
, int runstop
, int quiet
)
179 /* Run or stop the array. array must already be configured
181 * Only print failure messages if quiet == 0;
182 * quiet > 0 means really be quiet
183 * quiet < 0 means we will try again if it fails.
185 mdu_param_t param
; /* unused */
188 if (runstop
== -1 && md_get_version(fd
) < 9000) {
189 if (ioctl(fd
, STOP_MD
, 0)) {
190 pr_err("stopping device %s "
192 devname
, strerror(errno
));
197 if (md_get_version(fd
) < 9000) {
198 pr_err("need md driver version 0.90.0 or later\n");
202 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
203 pr_err("%s does not appear to be active.\n",
209 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
210 pr_err("failed to run array %s: %s\n",
211 devname
, strerror(errno
));
215 pr_err("started %s\n", devname
);
216 } else if (runstop
< 0){
217 struct map_ent
*map
= NULL
;
223 /* If this is an mdmon managed array, just write 'inactive'
224 * to the array state and let mdmon clear up.
226 devnum
= fd2devnum(fd
);
227 /* Get EXCL access first. If this fails, then attempting
228 * to stop is probably a bad idea.
231 fd
= open(devname
, O_RDONLY
|O_EXCL
);
232 if (fd
< 0 || fd2devnum(fd
) != devnum
) {
235 pr_err("Cannot get exclusive access to %s:"
237 "process, mounted filesystem "
238 "or active volume group?\n",
242 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
244 mdi
->array
.level
> 0 &&
245 is_subarray(mdi
->text_version
)) {
247 /* This is mdmon managed. */
252 (err
= sysfs_set_str(mdi
, NULL
,
260 pr_err("failed to stop array %s: %s\n",
261 devname
, strerror(errno
));
266 /* Give monitor a chance to act */
267 ping_monitor(mdi
->text_version
);
269 fd
= open_dev_excl(devnum
);
271 pr_err("failed to completely stop %s"
272 ": Device is busy\n",
278 mdi
->array
.major_version
== -1 &&
279 mdi
->array
.minor_version
== -2 &&
280 !is_subarray(mdi
->text_version
)) {
281 struct mdstat_ent
*mds
, *m
;
282 /* container, possibly mdmon-managed.
283 * Make sure mdmon isn't opening it, which
284 * would interfere with the 'stop'
286 ping_monitor(mdi
->sys_name
);
288 /* now check that there are no existing arrays
289 * which are members of this array
291 mds
= mdstat_read(0, 0);
292 for (m
=mds
; m
; m
=m
->next
)
293 if (m
->metadata_version
&&
294 strncmp(m
->metadata_version
, "external:", 9)==0 &&
295 is_subarray(m
->metadata_version
+9) &&
296 devname2devnum(m
->metadata_version
+10) == devnum
) {
298 pr_err("Cannot stop container %s: "
299 "member %s still active\n",
307 /* As we have an O_EXCL open, any use of the device
308 * which blocks STOP_ARRAY is probably a transient use,
309 * so it is reasonable to retry for a while - 5 seconds.
312 while (count
&& fd
>= 0
313 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
318 if (fd
>= 0 && err
) {
320 pr_err("failed to stop array %s: %s\n",
321 devname
, strerror(errno
));
323 fprintf(stderr
, "Perhaps a running "
324 "process, mounted filesystem "
325 "or active volume group?\n");
330 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
331 * was stopped, so We'll do it here just to be sure. Drop any
332 * partitions as well...
335 ioctl(fd
, BLKRRPART
, 0);
337 sysfs_uevent(mdi
, "change");
340 if (devnum
!= NoMdDev
&&
341 (stat("/dev/.udev", &stb
) != 0 ||
342 check_env("MDADM_NO_UDEV"))) {
343 struct map_ent
*mp
= map_by_devnum(&map
, devnum
);
344 remove_devices(devnum
, mp
? mp
->path
: NULL
);
349 pr_err("stopped %s\n", devname
);
351 map_remove(&map
, devnum
);
360 int Manage_resize(char *devname
, int fd
, long long size
, int raid_disks
)
362 mdu_array_info_t info
;
363 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
364 pr_err("Cannot get array information for %s: %s\n",
365 devname
, strerror(errno
));
371 info
.raid_disks
= raid_disks
;
372 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
373 pr_err("Cannot set device size/shape for %s: %s\n",
374 devname
, strerror(errno
));
380 int Manage_subdevs(char *devname
, int fd
,
381 struct mddev_dev
*devlist
, int verbose
, int test
,
382 char *update
, int force
)
384 /* do something to each dev.
386 * 'a' - add the device
388 * If that fails EINVAL, try ADD_NEW_DISK
389 * 'r' - remove the device HOT_REMOVE_DISK
390 * device can be 'faulty' or 'detached' in which case all
391 * matching devices are removed.
392 * 'f' - set the device faulty SET_DISK_FAULTY
393 * device can be 'detached' in which case any device that
394 * is inaccessible will be marked faulty.
395 * For 'f' and 'r', the device can also be a kernel-internal
396 * name such as 'sdb'.
398 struct mddev_dev
*add_devlist
= NULL
;
399 mdu_array_info_t array
;
400 mdu_disk_info_t disc
;
401 unsigned long long array_size
;
402 struct mddev_dev
*dv
, *next
= NULL
;
406 struct supertype
*st
, *tst
;
407 char *subarray
= NULL
;
412 int count
= 0; /* number of actions taken */
416 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
417 pr_err("cannot get array info for %s\n",
421 sysfs_init(&info
, fd
, 0);
423 /* array.size is only 32 bit and may be truncated.
424 * So read from sysfs if possible, and record number of sectors
427 array_size
= get_component_size(fd
);
429 array_size
= array
.size
* 2;
431 tst
= super_by_fd(fd
, &subarray
);
433 pr_err("unsupport array - version %d.%d\n",
434 array
.major_version
, array
.minor_version
);
439 for (dv
= devlist
, j
=0 ; dv
; dv
= next
, j
= jnext
) {
440 unsigned long long ldsize
;
442 char *dnprintable
= dv
->devname
;
443 char *add_dev
= dv
->devname
;
450 if (strcmp(dv
->devname
, "failed")==0 ||
451 strcmp(dv
->devname
, "faulty")==0) {
452 int remaining_disks
= array
.nr_disks
;
453 if (dv
->disposition
!= 'r') {
454 pr_err("%s only meaningful "
455 "with -r, not -%c\n",
456 dv
->devname
, dv
->disposition
);
459 for (; j
< MAX_DISKS
&& remaining_disks
> 0; j
++) {
462 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
464 if (disc
.major
== 0 && disc
.minor
== 0)
467 if ((disc
.state
& 1) == 0) /* faulty */
469 dev
= makedev(disc
.major
, disc
.minor
);
470 if (stb
.st_rdev
== dev
)
471 /* already did that one */
475 /* same slot again next time - things might
478 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
479 dnprintable
= dvname
;
484 } else if (strcmp(dv
->devname
, "detached") == 0) {
485 int remaining_disks
= array
.nr_disks
;
486 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
487 pr_err("%s only meaningful "
488 "with -r of -f, not -%c\n",
489 dv
->devname
, dv
->disposition
);
492 for (; j
< MAX_DISKS
&& remaining_disks
> 0; j
++) {
496 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
498 if (disc
.major
== 0 && disc
.minor
== 0)
501 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
502 sfd
= dev_open(dvname
, O_RDONLY
);
507 if (dv
->disposition
== 'f' &&
508 (disc
.state
& 1) == 1) /* already faulty */
512 dev
= makedev(disc
.major
, disc
.minor
);
513 if (stb
.st_rdev
== dev
)
514 /* already did that one */
518 /* same slot again next time - things might
521 dnprintable
= dvname
;
526 } else if (strcmp(dv
->devname
, "missing") == 0) {
527 if (dv
->disposition
!= 'A') {
528 pr_err("'missing' only meaningful "
532 if (add_devlist
== NULL
)
533 add_devlist
= conf_get_devs();
534 if (add_devlist
== NULL
) {
535 pr_err("no devices to scan for missing members.");
538 add_dev
= add_devlist
->devname
;
539 add_devlist
= add_devlist
->next
;
540 if (add_devlist
!= NULL
)
542 if (stat(add_dev
, &stb
) < 0)
544 } else if (strchr(dv
->devname
, '/') == NULL
&&
545 strchr(dv
->devname
, ':') == NULL
&&
546 strlen(dv
->devname
) < 50) {
547 /* Assume this is a kernel-internal name like 'sda1' */
550 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
551 pr_err("%s only meaningful "
552 "with -r or -f, not -%c\n",
553 dv
->devname
, dv
->disposition
);
557 sprintf(dname
, "dev-%s", dv
->devname
);
558 sysfd
= sysfs_open(fd2devnum(fd
), dname
, "block/dev");
562 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
563 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
564 stb
.st_rdev
= makedev(mj
,mn
);
571 sysfd
= sysfs_open(fd2devnum(fd
), dname
, "state");
573 pr_err("%s does not appear "
574 "to be a component of %s\n",
575 dv
->devname
, devname
);
582 tfd
= dev_open(dv
->devname
, O_RDONLY
);
583 if (tfd
< 0 && dv
->disposition
== 'r' &&
584 lstat(dv
->devname
, &stb
) == 0)
585 /* Be happy, the lstat worked, that is
586 * enough for --remove
590 if (tfd
< 0 || fstat(tfd
, &stb
) != 0) {
591 pr_err("cannot find %s: %s\n",
592 dv
->devname
, strerror(errno
));
600 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
601 pr_err("%s is not a "
607 switch(dv
->disposition
){
609 pr_err("internal error - devmode[%s]=%d\n",
610 dv
->devname
, dv
->disposition
);
616 pr_err("Cannot add disks to a"
617 " \'member\' array, perform this"
618 " operation on the parent container\n");
621 /* Make sure it isn't in use (in 2.6 or later) */
622 tfd
= dev_open(add_dev
, O_RDONLY
|O_EXCL
|O_DIRECT
);
623 if (tfd
< 0 && add_dev
!= dv
->devname
)
626 pr_err("Cannot open %s: %s\n",
627 dv
->devname
, strerror(errno
));
631 if (sysfs_freeze_array(&info
) == 1)
639 if (array
.not_persistent
==0)
640 st
->ss
->load_super(st
, tfd
, NULL
);
642 if (add_dev
== dv
->devname
) {
643 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
644 st
->ss
->free_super(st
);
648 } else if (!get_dev_size(tfd
, NULL
, &ldsize
)) {
649 st
->ss
->free_super(st
);
655 if (tst
->ss
->validate_geometry(
656 tst
, array
.level
, array
.layout
,
657 array
.raid_disks
, NULL
,
658 ldsize
>> 9, NULL
, NULL
, 0) == 0) {
660 pr_err("%s is larger than %s can "
662 " Add --force is you "
663 "really want to add this device.\n",
665 st
->ss
->free_super(st
);
669 pr_err("%s is larger than %s can "
671 " Adding anyway as --force "
675 if (!tst
->ss
->external
&&
676 array
.major_version
== 0 &&
677 md_get_version(fd
)%100 < 2) {
679 st
->ss
->free_super(st
);
681 if (ioctl(fd
, HOT_ADD_DISK
,
682 (unsigned long)stb
.st_rdev
)==0) {
684 pr_err("hot added %s\n",
689 pr_err("hot add failed for %s: %s\n",
690 add_dev
, strerror(errno
));
694 if (array
.not_persistent
== 0 || tst
->ss
->external
) {
696 /* need to find a sample superblock to copy, and
697 * a spare slot to use.
698 * For 'external' array (well, container based),
699 * We can just load the metadata for the array.
702 /* already loaded */;
703 else if (tst
->ss
->external
) {
704 tst
->ss
->load_container(tst
, fd
, NULL
);
705 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
709 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
711 if (disc
.major
==0 && disc
.minor
==0)
713 if ((disc
.state
& 4)==0) continue; /* sync */
714 /* Looks like a good device to try */
715 dev
= map_dev(disc
.major
, disc
.minor
, 1);
717 dfd
= dev_open(dev
, O_RDONLY
);
718 if (dfd
< 0) continue;
719 if (tst
->ss
->load_super(tst
, dfd
,
727 /* FIXME this is a bad test to be using */
728 if (!tst
->sb
&& dv
->disposition
== 'A') {
729 /* we are re-adding a device to a
730 * completely dead array - have to depend
733 } else if (!tst
->sb
) {
735 st
->ss
->free_super(st
);
736 pr_err("cannot load array metadata from %s\n", devname
);
740 /* Make sure device is large enough */
741 if (tst
->ss
->avail_size(tst
, ldsize
/512) <
745 st
->ss
->free_super(st
);
746 if (add_dev
!= dv
->devname
)
748 pr_err("%s not large enough to join array\n",
753 /* Possibly this device was recently part of the array
754 * and was temporarily removed, and is now being re-added.
755 * If so, we can simply re-add it.
760 st
->ss
->getinfo_super(st
, &mdi
, NULL
);
761 st
->ss
->uuid_from_super(st
, ouuid
);
763 tst
->ss
->uuid_from_super(tst
, duuid
);
765 /* Assume uuid matches: kernel will check */
766 memcpy(duuid
, ouuid
, sizeof(ouuid
));
767 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
768 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
769 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
770 /* look like it is worth a try. Need to
771 * make sure kernel will accept it though.
773 /* re-add doesn't work for version-1 superblocks
776 if (array
.major_version
== 1 &&
777 get_linux_version() <= 2006018)
779 disc
.number
= mdi
.disk
.number
;
780 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
781 || disc
.major
!= 0 || disc
.minor
!= 0
784 disc
.major
= major(stb
.st_rdev
);
785 disc
.minor
= minor(stb
.st_rdev
);
786 disc
.number
= mdi
.disk
.number
;
787 disc
.raid_disk
= mdi
.disk
.raid_disk
;
788 disc
.state
= mdi
.disk
.state
;
789 if (dv
->writemostly
== 1)
790 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
791 if (dv
->writemostly
== 2)
792 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
793 remove_partitions(tfd
);
796 if (update
|| dv
->writemostly
> 0) {
798 tfd
= dev_open(dv
->devname
, O_RDWR
);
800 pr_err("failed to open %s for"
801 " superblock update during re-add\n", dv
->devname
);
802 st
->ss
->free_super(st
);
806 if (dv
->writemostly
== 1)
807 rv
= st
->ss
->update_super(
808 st
, NULL
, "writemostly",
809 devname
, verbose
, 0, NULL
);
810 if (dv
->writemostly
== 2)
811 rv
= st
->ss
->update_super(
812 st
, NULL
, "readwrite",
813 devname
, verbose
, 0, NULL
);
815 rv
= st
->ss
->update_super(
817 devname
, verbose
, 0, NULL
);
819 rv
= st
->ss
->store_super(st
, tfd
);
823 pr_err("failed to update"
824 " superblock during re-add\n");
825 st
->ss
->free_super(st
);
829 /* don't even try if disk is marked as faulty */
831 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
833 pr_err("re-added %s\n", add_dev
);
835 st
->ss
->free_super(st
);
838 if (errno
== ENOMEM
|| errno
== EROFS
) {
839 pr_err("add new device failed for %s: %s\n",
840 add_dev
, strerror(errno
));
841 st
->ss
->free_super(st
);
842 if (add_dev
!= dv
->devname
)
848 st
->ss
->free_super(st
);
850 if (add_dev
!= dv
->devname
) {
852 pr_err("--re-add for %s to %s is not possible\n",
860 if (dv
->disposition
== 'A') {
863 pr_err("--re-add for %s to %s is not possible\n",
864 dv
->devname
, devname
);
867 if (array
.active_disks
< array
.raid_disks
) {
868 char *avail
= xcalloc(array
.raid_disks
, 1);
872 for (d
= 0; d
< MAX_DISKS
&& found
< array
.active_disks
; d
++) {
874 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
876 if (disc
.major
== 0 && disc
.minor
== 0)
878 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
880 avail
[disc
.raid_disk
] = 1;
883 array_failed
= !enough(array
.level
, array
.raid_disks
,
884 array
.layout
, 1, avail
);
888 pr_err("%s has failed so using --add cannot work and might destroy\n",
890 pr_err("data on %s. You should stop the array and re-assemble it.\n",
897 /* non-persistent. Must ensure that new drive
898 * is at least array.size big.
900 if (ldsize
/512 < array_size
) {
901 pr_err("%s not large enough to join array\n",
908 /* committed to really trying this device now*/
910 remove_partitions(tfd
);
914 /* in 2.6.17 and earlier, version-1 superblocks won't
915 * use the number we write, but will choose a free number.
916 * we must choose the same free number, which requires
917 * starting at 'raid_disks' and counting up
919 for (j
= array
.raid_disks
; j
< tst
->max_devs
; j
++) {
921 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
923 if (disc
.major
==0 && disc
.minor
==0)
925 if (disc
.state
& 8) /* removed */
928 disc
.major
= major(stb
.st_rdev
);
929 disc
.minor
= minor(stb
.st_rdev
);
932 if (array
.not_persistent
==0) {
934 if (dv
->writemostly
== 1)
935 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
936 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
937 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
942 if (tst
->ss
->write_init_super(tst
)) {
946 } else if (dv
->disposition
== 'A') {
947 /* this had better be raid1.
948 * As we are "--re-add"ing we must find a spare slot
951 char *used
= xcalloc(array
.raid_disks
, 1);
952 for (j
=0; j
< tst
->max_devs
; j
++) {
953 mdu_disk_info_t disc2
;
955 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
957 if (disc2
.major
==0 && disc2
.minor
==0)
959 if (disc2
.state
& 8) /* removed */
961 if (disc2
.raid_disk
< 0)
963 if (disc2
.raid_disk
> array
.raid_disks
)
965 used
[disc2
.raid_disk
] = 1;
967 for (j
=0 ; j
<array
.raid_disks
; j
++)
970 disc
.state
|= (1<<MD_DISK_SYNC
);
975 if (dv
->writemostly
== 1)
976 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
977 if (tst
->ss
->external
) {
979 * to an external metadata container */
980 struct mdinfo new_mdi
;
983 int devnum
= fd2devnum(fd
);
986 container_fd
= open_dev_excl(devnum
);
987 if (container_fd
< 0) {
988 pr_err("add failed for %s:"
989 " could not get exclusive access to container\n",
991 tst
->ss
->free_super(tst
);
995 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
996 if (mdmon_running(tst
->container_dev
))
997 tst
->update_tail
= &tst
->updates
;
998 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
1001 close(container_fd
);
1004 if (tst
->update_tail
)
1005 flush_metadata_updates(tst
);
1007 tst
->ss
->sync_metadata(tst
);
1009 sra
= sysfs_read(container_fd
, -1, 0);
1011 pr_err("add failed for %s: sysfs_read failed\n",
1013 close(container_fd
);
1014 tst
->ss
->free_super(tst
);
1017 sra
->array
.level
= LEVEL_CONTAINER
;
1018 /* Need to set data_offset and component_size */
1019 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
1020 new_mdi
.disk
.major
= disc
.major
;
1021 new_mdi
.disk
.minor
= disc
.minor
;
1022 new_mdi
.recovery_start
= 0;
1023 /* Make sure fds are closed as they are O_EXCL which
1024 * would block add_disk */
1025 tst
->ss
->free_super(tst
);
1026 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
1027 pr_err("add new device to external metadata"
1028 " failed for %s\n", dv
->devname
);
1029 close(container_fd
);
1033 ping_monitor_by_id(devnum
);
1035 close(container_fd
);
1037 tst
->ss
->free_super(tst
);
1038 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1039 pr_err("add new device failed for %s as %d: %s\n",
1040 dv
->devname
, j
, strerror(errno
));
1045 pr_err("added %s\n", dv
->devname
);
1051 pr_err("Cannot remove disks from a"
1052 " \'member\' array, perform this"
1053 " operation on the parent container\n");
1058 if (tst
->ss
->external
) {
1059 /* To remove a device from a container, we must
1060 * check that it isn't in use in an array.
1061 * This involves looking in the 'holders'
1062 * directory - there must be just one entry,
1064 * To ensure that it doesn't get used as a
1065 * hold spare while we are checking, we
1066 * get an O_EXCL open on the container
1068 int dnum
= fd2devnum(fd
);
1069 lfd
= open_dev_excl(dnum
);
1071 pr_err("Cannot get exclusive access "
1072 " to container - odd\n");
1077 /* in the detached case it is not possible to
1078 * check if we are the unique holder, so just
1079 * rely on the 'detached' checks
1081 if (strcmp(dv
->devname
, "detached") == 0 ||
1083 sysfs_unique_holder(dnum
, stb
.st_rdev
))
1086 pr_err("%s is %s, cannot remove.\n",
1088 errno
== EEXIST
? "still in use":
1094 /* FIXME check that it is a current member */
1096 /* device has been removed and we don't know
1097 * the major:minor number
1099 int n
= write(sysfd
, "remove", 6);
1107 err
= ioctl(fd
, HOT_REMOVE_DISK
, (unsigned long)stb
.st_rdev
);
1108 if (err
&& errno
== ENODEV
) {
1109 /* Old kernels rejected this if no personality
1111 struct mdinfo
*sra
= sysfs_read(fd
, 0, GET_DEVS
);
1112 struct mdinfo
*dv
= NULL
;
1115 for ( ; dv
; dv
=dv
->next
)
1116 if (dv
->disk
.major
== (int)major(stb
.st_rdev
) &&
1117 dv
->disk
.minor
== (int)minor(stb
.st_rdev
))
1120 err
= sysfs_set_str(sra
, dv
,
1129 pr_err("hot remove failed "
1130 "for %s: %s\n", dnprintable
,
1136 if (tst
->ss
->external
) {
1138 * Before dropping our exclusive open we make an
1139 * attempt at preventing mdmon from seeing an
1140 * 'add' event before reconciling this 'remove'
1143 char *name
= devnum2devname(fd2devnum(fd
));
1146 pr_err("unable to get container name\n");
1157 pr_err("hot removed %s from %s\n",
1158 dnprintable
, devname
);
1161 case 'f': /* set faulty */
1162 /* FIXME check current member */
1163 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1164 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1165 (unsigned long) stb
.st_rdev
))) {
1166 pr_err("set device faulty failed for %s: %s\n",
1167 dnprintable
, strerror(errno
));
1177 pr_err("set %s faulty in %s\n",
1178 dnprintable
, devname
);
1183 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1184 if (test
&& count
== 0)
1190 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1194 int autodetect(void)
1196 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1198 int fd
= dev_open("9:0", O_RDONLY
);
1200 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1207 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int quiet
)
1209 struct supertype supertype
, *st
= &supertype
;
1212 memset(st
, 0, sizeof(*st
));
1214 fd
= open_subarray(dev
, subarray
, st
, quiet
);
1218 if (!st
->ss
->update_subarray
) {
1220 pr_err("Operation not supported for %s metadata\n",
1225 if (mdmon_running(st
->devnum
))
1226 st
->update_tail
= &st
->updates
;
1228 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1232 pr_err("Failed to update %s of subarray-%s in %s\n",
1233 update
, subarray
, dev
);
1234 } else if (st
->update_tail
)
1235 flush_metadata_updates(st
);
1237 st
->ss
->sync_metadata(st
);
1239 if (rv
== 0 && strcmp(update
, "name") == 0 && !quiet
)
1240 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1244 st
->ss
->free_super(st
);
1250 /* Move spare from one array to another
1251 * If adding to destination array fails
1252 * add back to original array
1253 * Returns 1 on success, 0 on failure */
1254 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1256 struct mddev_dev devlist
;
1259 /* try to remove and add */
1260 int fd1
= open(to_devname
, O_RDONLY
);
1261 int fd2
= open(from_devname
, O_RDONLY
);
1263 if (fd1
< 0 || fd2
< 0) {
1264 if (fd1
>=0) close(fd1
);
1265 if (fd2
>=0) close(fd2
);
1269 devlist
.next
= NULL
;
1271 devlist
.writemostly
= 0;
1272 devlist
.devname
= devname
;
1273 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1275 devlist
.disposition
= 'r';
1276 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1277 devlist
.disposition
= 'a';
1278 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1279 /* make sure manager is aware of changes */
1280 ping_manager(to_devname
);
1281 ping_manager(from_devname
);
1286 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);