]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Manage.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2012 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-managed array, we need to modify the
55 * metadata_version so that mdmon doesn't undo our change.
57 mdi
= sysfs_read(fd
, NULL
, 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(char *devnm
, char *path
)
133 * Remove names at 'path' - possibly with
134 * partition suffixes - which link to the 'standard'
135 * name for devnm. These were probably created
136 * by mdadm when the array was assembled.
149 sprintf(base
, "/dev/%s", devnm
);
150 be
= base
+ strlen(base
);
152 path2
= xmalloc(strlen(path
)+20);
154 pe
= path2
+ strlen(path2
);
156 for (part
= 0; part
< 16; part
++) {
158 sprintf(be
, "p%d", part
);
161 sprintf(pe
, "p%d", part
);
163 sprintf(pe
, "%d", part
);
165 n
= readlink(path2
, link
, sizeof(link
));
166 if (n
> 0 && (int)strlen(base
) == n
&&
167 strncmp(link
, base
, n
) == 0)
173 int Manage_runstop(char *devname
, int fd
, int runstop
,
174 int verbose
, int will_retry
)
176 /* Run or stop the array. Array must already be configured
177 * 'Run' requires >= 0.90.0
178 * 'will_retry' is only relevant for 'stop', and means
179 * that error messages are not wanted.
181 mdu_param_t param
; /* unused */
184 if (will_retry
&& verbose
== 0)
187 if (runstop
== -1 && md_get_version(fd
) < 9000) {
188 if (ioctl(fd
, STOP_MD
, 0) == 0)
190 pr_err("stopping device %s "
192 devname
, strerror(errno
));
196 if (md_get_version(fd
) < 9000) {
197 pr_err("need md driver version 0.90.0 or later\n");
202 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
204 pr_err("failed to run array %s: %s\n",
205 devname
, strerror(errno
));
209 pr_err("started %s\n", devname
);
210 } else if (runstop
< 0){
211 struct map_ent
*map
= NULL
;
217 /* If this is an mdmon managed array, just write 'inactive'
218 * to the array state and let mdmon clear up.
220 strcpy(devnm
, fd2devnm(fd
));
221 /* Get EXCL access first. If this fails, then attempting
222 * to stop is probably a bad idea.
225 fd
= open(devname
, O_RDONLY
|O_EXCL
);
226 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
230 pr_err("Cannot get exclusive access to %s:"
232 "process, mounted filesystem "
233 "or active volume group?\n",
237 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
);
239 mdi
->array
.level
> 0 &&
240 is_subarray(mdi
->text_version
)) {
242 /* This is mdmon managed. */
245 /* As we have an O_EXCL open, any use of the device
246 * which blocks STOP_ARRAY is probably a transient use,
247 * so it is reasonable to retry for a while - 5 seconds.
251 (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(devnm
);
272 pr_err("failed to completely stop %s"
273 ": Device is busy\n",
279 mdi
->array
.major_version
== -1 &&
280 mdi
->array
.minor_version
== -2 &&
281 !is_subarray(mdi
->text_version
)) {
282 struct mdstat_ent
*mds
, *m
;
283 /* container, possibly mdmon-managed.
284 * Make sure mdmon isn't opening it, which
285 * would interfere with the 'stop'
287 ping_monitor(mdi
->sys_name
);
289 /* now check that there are no existing arrays
290 * which are members of this array
292 mds
= mdstat_read(0, 0);
293 for (m
= mds
; m
; m
= m
->next
)
294 if (m
->metadata_version
&&
295 strncmp(m
->metadata_version
, "external:", 9)==0 &&
296 metadata_container_matches(m
->metadata_version
+9,
299 pr_err("Cannot stop container %s: "
300 "member %s still active\n",
308 /* As we have an O_EXCL open, any use of the device
309 * which blocks STOP_ARRAY is probably a transient use,
310 * so it is reasonable to retry for a while - 5 seconds.
313 while (count
&& fd
>= 0
314 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
319 if (fd
>= 0 && err
) {
321 pr_err("failed to stop array %s: %s\n",
322 devname
, strerror(errno
));
324 fprintf(stderr
, "Perhaps a running "
325 "process, mounted filesystem "
326 "or active volume group?\n");
331 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
332 * was stopped, so We'll do it here just to be sure. Drop any
333 * partitions as well...
336 ioctl(fd
, BLKRRPART
, 0);
338 sysfs_uevent(mdi
, "change");
341 (stat("/dev/.udev", &stb
) != 0 ||
342 check_env("MDADM_NO_UDEV"))) {
343 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
344 remove_devices(devnm
, mp
? mp
->path
: NULL
);
348 pr_err("stopped %s\n", devname
);
350 map_remove(&map
, devnm
);
359 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
361 mdu_array_info_t array
;
362 mdu_disk_info_t disk
;
366 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
369 remaining_disks
= array
.nr_disks
;
370 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
371 struct mddev_dev
*new;
374 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
376 if (disk
.major
== 0 && disk
.minor
== 0)
379 if ((disk
.state
& 1) == 0) /* not faulty */
381 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
382 new = xmalloc(sizeof(*new));
383 new->devname
= xstrdup(buf
);
384 new->disposition
= disp
;
385 new->next
= dv
->next
;
391 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
393 mdu_array_info_t array
;
394 mdu_disk_info_t disk
;
398 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
401 remaining_disks
= array
.nr_disks
;
402 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
403 struct mddev_dev
*new;
407 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
409 if (disk
.major
== 0 && disk
.minor
== 0)
412 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
414 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
415 sfd
= dev_open(buf
, O_RDONLY
);
422 /* Probably not detached */
424 new = xmalloc(sizeof(*new));
425 new->devname
= xstrdup(buf
);
426 new->disposition
= disp
;
427 new->next
= dv
->next
;
433 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
434 struct supertype
*dev_st
, struct supertype
*tst
,
436 char *update
, char *devname
, int verbose
,
437 mdu_array_info_t
*array
)
443 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
444 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
446 tst
->ss
->uuid_from_super(tst
, duuid
);
448 /* Assume uuid matches: kernel will check */
449 memcpy(duuid
, ouuid
, sizeof(ouuid
));
450 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
451 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
452 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
453 /* Looks like it is worth a
454 * try. Need to make sure
455 * kernel will accept it
458 mdu_disk_info_t disc
;
459 /* re-add doesn't work for version-1 superblocks
462 if (array
->major_version
== 1 &&
463 get_linux_version() <= 2006018)
465 disc
.number
= mdi
.disk
.number
;
466 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
467 || disc
.major
!= 0 || disc
.minor
!= 0
470 disc
.major
= major(rdev
);
471 disc
.minor
= minor(rdev
);
472 disc
.number
= mdi
.disk
.number
;
473 disc
.raid_disk
= mdi
.disk
.raid_disk
;
474 disc
.state
= mdi
.disk
.state
;
475 if (dv
->writemostly
== 1)
476 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
477 if (dv
->writemostly
== 2)
478 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
479 remove_partitions(tfd
);
480 if (update
|| dv
->writemostly
> 0) {
482 tfd
= dev_open(dv
->devname
, O_RDWR
);
484 pr_err("failed to open %s for"
485 " superblock update during re-add\n", dv
->devname
);
489 if (dv
->writemostly
== 1)
490 rv
= dev_st
->ss
->update_super(
491 dev_st
, NULL
, "writemostly",
492 devname
, verbose
, 0, NULL
);
493 if (dv
->writemostly
== 2)
494 rv
= dev_st
->ss
->update_super(
495 dev_st
, NULL
, "readwrite",
496 devname
, verbose
, 0, NULL
);
498 rv
= dev_st
->ss
->update_super(
499 dev_st
, NULL
, update
,
500 devname
, verbose
, 0, NULL
);
502 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
505 pr_err("failed to update"
506 " superblock during re-add\n");
510 /* don't even try if disk is marked as faulty */
512 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
514 pr_err("re-added %s\n", dv
->devname
);
517 if (errno
== ENOMEM
|| errno
== EROFS
) {
518 pr_err("add new device failed for %s: %s\n",
519 dv
->devname
, strerror(errno
));
520 if (dv
->disposition
== 'M')
529 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
530 struct supertype
*tst
, mdu_array_info_t
*array
,
531 int force
, int verbose
, char *devname
,
532 char *update
, unsigned long rdev
, unsigned long long array_size
)
534 unsigned long long ldsize
;
535 struct supertype
*dev_st
= NULL
;
537 mdu_disk_info_t disc
;
539 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
540 if (dv
->disposition
== 'M')
546 if (tst
->ss
->validate_geometry(
547 tst
, array
->level
, array
->layout
,
548 array
->raid_disks
, NULL
,
549 ldsize
>> 9, INVALID_SECTORS
, NULL
, NULL
, 0) == 0) {
551 pr_err("%s is larger than %s can "
553 " Add --force is you "
554 "really want to add this device.\n",
555 dv
->devname
, devname
);
558 pr_err("%s is larger than %s can "
560 " Adding anyway as --force "
562 dv
->devname
, devname
);
564 if (!tst
->ss
->external
&&
565 array
->major_version
== 0 &&
566 md_get_version(fd
)%100 < 2) {
567 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
569 pr_err("hot added %s\n",
574 pr_err("hot add failed for %s: %s\n",
575 dv
->devname
, strerror(errno
));
579 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
581 /* need to find a sample superblock to copy, and
582 * a spare slot to use.
583 * For 'external' array (well, container based),
584 * We can just load the metadata for the array->
588 /* already loaded */;
589 else if (tst
->ss
->external
) {
590 tst
->ss
->load_container(tst
, fd
, NULL
);
591 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
595 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
597 if (disc
.major
==0 && disc
.minor
==0)
599 if ((disc
.state
& 4)==0) /* sync */
601 /* Looks like a good device to try */
602 dev
= map_dev(disc
.major
, disc
.minor
, 1);
605 dfd
= dev_open(dev
, O_RDONLY
);
608 if (tst
->ss
->load_super(tst
, dfd
,
616 /* FIXME this is a bad test to be using */
617 if (!tst
->sb
&& dv
->disposition
!= 'a') {
618 /* we are re-adding a device to a
619 * completely dead array - have to depend
622 } else if (!tst
->sb
) {
623 pr_err("cannot load array metadata from %s\n", devname
);
627 /* Make sure device is large enough */
628 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
630 if (dv
->disposition
== 'M')
632 pr_err("%s not large enough to join array\n",
637 /* Possibly this device was recently part of
638 * the array and was temporarily removed, and
639 * is now being re-added. If so, we can
643 if (array
->not_persistent
==0) {
644 dev_st
= dup_super(tst
);
645 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
647 if (dev_st
&& dev_st
->sb
) {
648 int rv
= attempt_re_add(fd
, tfd
, dv
,
654 dev_st
->ss
->free_super(dev_st
);
658 if (dv
->disposition
== 'M') {
660 pr_err("--re-add for %s to %s is not possible\n",
661 dv
->devname
, devname
);
664 if (dv
->disposition
== 'A') {
665 pr_err("--re-add for %s to %s is not possible\n",
666 dv
->devname
, devname
);
669 if (array
->active_disks
< array
->raid_disks
) {
670 char *avail
= xcalloc(array
->raid_disks
, 1);
674 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
676 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
678 if (disc
.major
== 0 && disc
.minor
== 0)
680 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
682 avail
[disc
.raid_disk
] = 1;
685 array_failed
= !enough(array
->level
, array
->raid_disks
,
686 array
->layout
, 1, avail
);
690 pr_err("%s has failed so using --add cannot work and might destroy\n",
692 pr_err("data on %s. You should stop the array and re-assemble it.\n",
697 /* non-persistent. Must ensure that new drive
698 * is at least array->size big.
700 if (ldsize
/512 < array_size
) {
701 pr_err("%s not large enough to join array\n",
706 /* committed to really trying this device now*/
707 remove_partitions(tfd
);
709 /* in 2.6.17 and earlier, version-1 superblocks won't
710 * use the number we write, but will choose a free number.
711 * we must choose the same free number, which requires
712 * starting at 'raid_disks' and counting up
714 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
716 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
718 if (disc
.major
==0 && disc
.minor
==0)
720 if (disc
.state
& 8) /* removed */
723 disc
.major
= major(rdev
);
724 disc
.minor
= minor(rdev
);
727 if (array
->not_persistent
==0) {
729 if (dv
->writemostly
== 1)
730 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
731 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
732 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
733 dv
->devname
, INVALID_SECTORS
))
735 if (tst
->ss
->write_init_super(tst
))
737 } else if (dv
->disposition
== 'A') {
738 /* this had better be raid1.
739 * As we are "--re-add"ing we must find a spare slot
742 char *used
= xcalloc(array
->raid_disks
, 1);
743 for (j
= 0; j
< tst
->max_devs
; j
++) {
744 mdu_disk_info_t disc2
;
746 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
748 if (disc2
.major
==0 && disc2
.minor
==0)
750 if (disc2
.state
& 8) /* removed */
752 if (disc2
.raid_disk
< 0)
754 if (disc2
.raid_disk
> array
->raid_disks
)
756 used
[disc2
.raid_disk
] = 1;
758 for (j
= 0 ; j
< array
->raid_disks
; j
++)
761 disc
.state
|= (1<<MD_DISK_SYNC
);
766 if (dv
->writemostly
== 1)
767 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
768 if (tst
->ss
->external
) {
770 * to an external metadata container */
771 struct mdinfo new_mdi
;
777 strcpy(devnm
, fd2devnm(fd
));
779 container_fd
= open_dev_excl(devnm
);
780 if (container_fd
< 0) {
781 pr_err("add failed for %s:"
782 " could not get exclusive access to container\n",
784 tst
->ss
->free_super(tst
);
788 Kill(dv
->devname
, NULL
, 0, -1, 0);
789 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
790 if (mdmon_running(tst
->container_devnm
))
791 tst
->update_tail
= &tst
->updates
;
792 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
793 dv
->devname
, INVALID_SECTORS
)) {
798 if (tst
->update_tail
)
799 flush_metadata_updates(tst
);
801 tst
->ss
->sync_metadata(tst
);
803 sra
= sysfs_read(container_fd
, NULL
, 0);
805 pr_err("add failed for %s: sysfs_read failed\n",
808 tst
->ss
->free_super(tst
);
811 sra
->array
.level
= LEVEL_CONTAINER
;
812 /* Need to set data_offset and component_size */
813 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
814 new_mdi
.disk
.major
= disc
.major
;
815 new_mdi
.disk
.minor
= disc
.minor
;
816 new_mdi
.recovery_start
= 0;
817 /* Make sure fds are closed as they are O_EXCL which
818 * would block add_disk */
819 tst
->ss
->free_super(tst
);
820 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
821 pr_err("add new device to external metadata"
822 " failed for %s\n", dv
->devname
);
831 tst
->ss
->free_super(tst
);
832 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
833 pr_err("add new device failed for %s as %d: %s\n",
834 dv
->devname
, j
, strerror(errno
));
839 pr_err("added %s\n", dv
->devname
);
843 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
844 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
849 if (tst
->ss
->external
) {
850 /* To remove a device from a container, we must
851 * check that it isn't in use in an array.
852 * This involves looking in the 'holders'
853 * directory - there must be just one entry,
855 * To ensure that it doesn't get used as a
856 * hot spare while we are checking, we
857 * get an O_EXCL open on the container
861 strcpy(devnm
, fd2devnm(fd
));
862 lfd
= open_dev_excl(devnm
);
864 pr_err("Cannot get exclusive access "
865 " to container - odd\n");
868 /* We may not be able to check on holders in
869 * sysfs, either because we don't have the dev num
870 * (rdev == 0) or because the device has been detached
871 * and the 'holders' directory no longer exists
872 * (ret == -1). In that case, assume it is OK to
878 ret
= sysfs_unique_holder(devnm
, rdev
);
880 pr_err("%s is not a member, cannot remove.\n",
886 pr_err("%s is still in use, cannot remove.\n",
892 /* FIXME check that it is a current member */
894 /* device has been removed and we don't know
895 * the major:minor number
897 int n
= write(sysfd
, "remove", 6);
903 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
904 if (err
&& errno
== ENODEV
) {
905 /* Old kernels rejected this if no personality
907 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
908 struct mdinfo
*dv
= NULL
;
911 for ( ; dv
; dv
=dv
->next
)
912 if (dv
->disk
.major
== (int)major(rdev
) &&
913 dv
->disk
.minor
== (int)minor(rdev
))
916 err
= sysfs_set_str(sra
, dv
,
925 pr_err("hot remove failed "
926 "for %s: %s\n", dv
->devname
,
932 if (tst
->ss
->external
) {
934 * Before dropping our exclusive open we make an
935 * attempt at preventing mdmon from seeing an
936 * 'add' event before reconciling this 'remove'
939 char *devnm
= fd2devnm(fd
);
942 pr_err("unable to get container name\n");
951 pr_err("hot removed %s from %s\n",
952 dv
->devname
, devname
);
956 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
957 unsigned long rdev
, int verbose
, char *devname
)
959 struct mdinfo
*mdi
, *di
;
960 if (tst
->ss
->external
) {
961 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
964 /* Need to find the device in sysfs and add 'want_replacement' to the
967 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
968 if (!mdi
|| !mdi
->devs
) {
969 pr_err("Cannot find status of %s to enable replacement - strange\n",
973 for (di
= mdi
->devs
; di
; di
= di
->next
)
974 if (di
->disk
.major
== (int)major(rdev
) &&
975 di
->disk
.minor
== (int)minor(rdev
))
979 if (di
->disk
.raid_disk
< 0) {
980 pr_err("%s is not active and so cannot be replaced.\n",
985 rv
= sysfs_set_str(mdi
, di
,
986 "state", "want_replacement");
989 pr_err("Failed to request replacement for %s\n",
994 pr_err("Marked %s (device %d in %s) for replacement\n",
995 dv
->devname
, di
->disk
.raid_disk
, devname
);
996 /* If there is a matching 'with', we need to tell it which
999 while (dv
&& dv
->disposition
!= 'W')
1002 dv
->disposition
= 'w';
1003 dv
->used
= di
->disk
.raid_disk
;
1008 pr_err("%s not found in %s so cannot --replace it\n",
1009 dv
->devname
, devname
);
1013 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1014 unsigned long rdev
, int verbose
, char *devname
)
1016 struct mdinfo
*mdi
, *di
;
1017 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1018 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1019 if (!mdi
|| !mdi
->devs
) {
1020 pr_err("Cannot find status of %s to enable replacement - strange\n",
1024 for (di
= mdi
->devs
; di
; di
= di
->next
)
1025 if (di
->disk
.major
== (int)major(rdev
) &&
1026 di
->disk
.minor
== (int)minor(rdev
))
1030 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1031 pr_err("%s is faulty and cannot be a replacement\n",
1036 if (di
->disk
.raid_disk
>= 0) {
1037 pr_err("%s is active and cannot be a replacement\n",
1042 rv
= sysfs_set_num(mdi
, di
,
1046 pr_err("Failed to %s as preferred replacement.\n",
1051 pr_err("Marked %s in %s as replacement for device %d\n",
1052 dv
->devname
, devname
, dv
->used
);
1056 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1057 dv
->devname
, devname
);
1061 int Manage_subdevs(char *devname
, int fd
,
1062 struct mddev_dev
*devlist
, int verbose
, int test
,
1063 char *update
, int force
)
1065 /* Do something to each dev.
1067 * 'a' - add the device
1069 * If that fails EINVAL, try ADD_NEW_DISK
1070 * 'A' - re-add the device
1071 * 'r' - remove the device: HOT_REMOVE_DISK
1072 * device can be 'faulty' or 'detached' in which case all
1073 * matching devices are removed.
1074 * 'f' - set the device faulty SET_DISK_FAULTY
1075 * device can be 'detached' in which case any device that
1076 * is inaccessible will be marked faulty.
1077 * 'R' - mark this device as wanting replacement.
1078 * 'W' - this device is added if necessary and activated as
1079 * a replacement for a previous 'R' device.
1081 * 'w' - 'W' will be changed to 'w' when it is paired with
1082 * a 'R' device. If a 'W' is found while walking the list
1083 * it must be unpaired, and is an error.
1084 * 'M' - this is created by a 'missing' target. It is a slight
1086 * 'F' - Another variant of 'A', where the device was faulty
1087 * so must be removed from the array first.
1089 * For 'f' and 'r', the device can also be a kernel-internal
1090 * name such as 'sdb'.
1092 mdu_array_info_t array
;
1093 unsigned long long array_size
;
1094 struct mddev_dev
*dv
;
1097 struct supertype
*tst
;
1098 char *subarray
= NULL
;
1100 int count
= 0; /* number of actions taken */
1104 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1105 pr_err("Cannot get array info for %s\n",
1109 sysfs_init(&info
, fd
, NULL
);
1111 /* array.size is only 32 bits and may be truncated.
1112 * So read from sysfs if possible, and record number of sectors
1115 array_size
= get_component_size(fd
);
1116 if (array_size
<= 0)
1117 array_size
= array
.size
* 2;
1119 tst
= super_by_fd(fd
, &subarray
);
1121 pr_err("unsupport array - version %d.%d\n",
1122 array
.major_version
, array
.minor_version
);
1127 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1130 if (strcmp(dv
->devname
, "failed") == 0 ||
1131 strcmp(dv
->devname
, "faulty") == 0) {
1132 if (dv
->disposition
!= 'A'
1133 && dv
->disposition
!= 'r') {
1134 pr_err("%s only meaningful "
1135 "with -r or --re-add, not -%c\n",
1136 dv
->devname
, dv
->disposition
);
1139 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1143 if (strcmp(dv
->devname
, "detached") == 0) {
1144 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1145 pr_err("%s only meaningful "
1146 "with -r of -f, not -%c\n",
1147 dv
->devname
, dv
->disposition
);
1150 add_detached(dv
, fd
, dv
->disposition
);
1154 if (strcmp(dv
->devname
, "missing") == 0) {
1155 struct mddev_dev
*add_devlist
= NULL
;
1156 struct mddev_dev
**dp
;
1157 if (dv
->disposition
!= 'A') {
1158 pr_err("'missing' only meaningful "
1162 add_devlist
= conf_get_devs();
1163 if (add_devlist
== NULL
) {
1164 pr_err("no devices to scan for missing members.");
1167 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1168 /* 'M' (for 'missing') is like 'A' without errors */
1169 (*dp
)->disposition
= 'M';
1171 dv
->next
= add_devlist
;
1175 if (strchr(dv
->devname
, '/') == NULL
&&
1176 strchr(dv
->devname
, ':') == NULL
&&
1177 strlen(dv
->devname
) < 50) {
1178 /* Assume this is a kernel-internal name like 'sda1' */
1181 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1182 pr_err("%s only meaningful "
1183 "with -r or -f, not -%c\n",
1184 dv
->devname
, dv
->disposition
);
1188 sprintf(dname
, "dev-%s", dv
->devname
);
1189 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1193 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1194 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1195 stb
.st_rdev
= makedev(mj
,mn
);
1202 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1204 pr_err("%s does not appear "
1205 "to be a component of %s\n",
1206 dv
->devname
, devname
);
1211 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1215 int open_err
= errno
;
1216 if (stat(dv
->devname
, &stb
) != 0) {
1217 pr_err("Cannot find %s: %s\n",
1218 dv
->devname
, strerror(errno
));
1221 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1222 if (dv
->disposition
== 'M')
1223 /* non-fatal. Also improbable */
1225 pr_err("%s is not a block device.\n",
1229 if (dv
->disposition
== 'r')
1230 /* Be happy, the stat worked, that is
1231 * enough for --remove
1235 if (dv
->disposition
== 'M')
1238 pr_err("Cannot open %s: %s\n",
1239 dv
->devname
, strerror(open_err
));
1244 switch(dv
->disposition
){
1246 pr_err("internal error - devmode[%s]=%d\n",
1247 dv
->devname
, dv
->disposition
);
1251 case 'M': /* --re-add missing */
1252 case 'F': /* --re-add faulty */
1253 /* add the device */
1255 pr_err("Cannot add disks to a"
1256 " \'member\' array, perform this"
1257 " operation on the parent container\n");
1260 if (dv
->disposition
== 'F')
1261 /* Need to remove first */
1262 ioctl(fd
, HOT_REMOVE_DISK
,
1263 (unsigned long)stb
.st_rdev
);
1264 /* Make sure it isn't in use (in 2.6 or later) */
1265 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1267 /* We know no-one else is using it. We'll
1268 * need non-exclusive access to add it, so
1272 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1275 if (dv
->disposition
== 'M')
1277 pr_err("Cannot open %s: %s\n",
1278 dv
->devname
, strerror(errno
));
1282 if (sysfs_freeze_array(&info
) == 1)
1287 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1288 force
, verbose
, devname
, update
,
1289 stb
.st_rdev
, array_size
);
1301 pr_err("Cannot remove disks from a"
1302 " \'member\' array, perform this"
1303 " operation on the parent container\n");
1306 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1307 stb
.st_rdev
, verbose
,
1318 case 'f': /* set faulty */
1319 /* FIXME check current member */
1320 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1321 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1322 (unsigned long) stb
.st_rdev
))) {
1323 pr_err("set device faulty failed for %s: %s\n",
1324 dv
->devname
, strerror(errno
));
1334 pr_err("set %s faulty in %s\n",
1335 dv
->devname
, devname
);
1337 case 'R': /* Mark as replaceable */
1339 pr_err("Cannot replace disks in a"
1340 " \'member\' array, perform this"
1341 " operation on the parent container\n");
1345 if (sysfs_freeze_array(&info
) == 1)
1350 rv
= Manage_replace(tst
, fd
, dv
,
1351 stb
.st_rdev
, verbose
,
1359 case 'W': /* --with device that doesn't match */
1360 pr_err("No matching --replace device for --with %s\n",
1363 case 'w': /* --with device which was matched */
1364 rv
= Manage_with(tst
, fd
, dv
,
1365 stb
.st_rdev
, verbose
, devname
);
1372 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1373 if (test
&& count
== 0)
1379 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1383 int autodetect(void)
1385 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1387 int fd
= dev_open("9:0", O_RDONLY
);
1389 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1396 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1398 struct supertype supertype
, *st
= &supertype
;
1401 memset(st
, 0, sizeof(*st
));
1403 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1407 if (!st
->ss
->update_subarray
) {
1409 pr_err("Operation not supported for %s metadata\n",
1414 if (mdmon_running(st
->devnm
))
1415 st
->update_tail
= &st
->updates
;
1417 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1421 pr_err("Failed to update %s of subarray-%s in %s\n",
1422 update
, subarray
, dev
);
1423 } else if (st
->update_tail
)
1424 flush_metadata_updates(st
);
1426 st
->ss
->sync_metadata(st
);
1428 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1429 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1433 st
->ss
->free_super(st
);
1439 /* Move spare from one array to another If adding to destination array fails
1440 * add back to original array.
1441 * Returns 1 on success, 0 on failure */
1442 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1444 struct mddev_dev devlist
;
1447 /* try to remove and add */
1448 int fd1
= open(to_devname
, O_RDONLY
);
1449 int fd2
= open(from_devname
, O_RDONLY
);
1451 if (fd1
< 0 || fd2
< 0) {
1452 if (fd1
>=0) close(fd1
);
1453 if (fd2
>=0) close(fd2
);
1457 devlist
.next
= NULL
;
1459 devlist
.writemostly
= 0;
1460 devlist
.devname
= devname
;
1461 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1463 devlist
.disposition
= 'r';
1464 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1465 devlist
.disposition
= 'a';
1466 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1467 /* make sure manager is aware of changes */
1468 ping_manager(to_devname
);
1469 ping_manager(from_devname
);
1474 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);