]>
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
, -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)
176 int Manage_runstop(char *devname
, int fd
, int runstop
,
177 int verbose
, int will_retry
)
179 /* Run or stop the array. Array must already be configured
180 * 'Run' requires >= 0.90.0
181 * 'will_retry' is only relevant for 'stop', and means
182 * that error messages are not wanted.
184 mdu_param_t param
; /* unused */
187 if (will_retry
&& verbose
== 0)
190 if (runstop
== -1 && md_get_version(fd
) < 9000) {
191 if (ioctl(fd
, STOP_MD
, 0) == 0)
193 pr_err("stopping device %s "
195 devname
, strerror(errno
));
199 if (md_get_version(fd
) < 9000) {
200 pr_err("need md driver version 0.90.0 or later\n");
205 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
207 pr_err("failed to run array %s: %s\n",
208 devname
, strerror(errno
));
212 pr_err("started %s\n", devname
);
213 } else if (runstop
< 0){
214 struct map_ent
*map
= NULL
;
220 /* If this is an mdmon managed array, just write 'inactive'
221 * to the array state and let mdmon clear up.
223 devnum
= fd2devnum(fd
);
224 /* Get EXCL access first. If this fails, then attempting
225 * to stop is probably a bad idea.
228 fd
= open(devname
, O_RDONLY
|O_EXCL
);
229 if (fd
< 0 || fd2devnum(fd
) != devnum
) {
233 pr_err("Cannot get exclusive access to %s:"
235 "process, mounted filesystem "
236 "or active volume group?\n",
240 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
242 mdi
->array
.level
> 0 &&
243 is_subarray(mdi
->text_version
)) {
245 /* This is mdmon managed. */
248 /* As we have an O_EXCL open, any use of the device
249 * which blocks STOP_ARRAY is probably a transient use,
250 * so it is reasonable to retry for a while - 5 seconds.
254 (err
= sysfs_set_str(mdi
, NULL
,
263 pr_err("failed to stop array %s: %s\n",
264 devname
, strerror(errno
));
269 /* Give monitor a chance to act */
270 ping_monitor(mdi
->text_version
);
272 fd
= open_dev_excl(devnum
);
275 pr_err("failed to completely stop %s"
276 ": Device is busy\n",
282 mdi
->array
.major_version
== -1 &&
283 mdi
->array
.minor_version
== -2 &&
284 !is_subarray(mdi
->text_version
)) {
285 struct mdstat_ent
*mds
, *m
;
286 /* container, possibly mdmon-managed.
287 * Make sure mdmon isn't opening it, which
288 * would interfere with the 'stop'
290 ping_monitor(mdi
->sys_name
);
292 /* now check that there are no existing arrays
293 * which are members of this array
295 mds
= mdstat_read(0, 0);
296 for (m
= mds
; m
; m
= m
->next
)
297 if (m
->metadata_version
&&
298 strncmp(m
->metadata_version
, "external:", 9)==0 &&
299 is_subarray(m
->metadata_version
+9) &&
300 devname2devnum(m
->metadata_version
+10) == devnum
) {
302 pr_err("Cannot stop container %s: "
303 "member %s still active\n",
311 /* As we have an O_EXCL open, any use of the device
312 * which blocks STOP_ARRAY is probably a transient use,
313 * so it is reasonable to retry for a while - 5 seconds.
316 while (count
&& fd
>= 0
317 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
322 if (fd
>= 0 && err
) {
324 pr_err("failed to stop array %s: %s\n",
325 devname
, strerror(errno
));
327 fprintf(stderr
, "Perhaps a running "
328 "process, mounted filesystem "
329 "or active volume group?\n");
334 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
335 * was stopped, so We'll do it here just to be sure. Drop any
336 * partitions as well...
339 ioctl(fd
, BLKRRPART
, 0);
341 sysfs_uevent(mdi
, "change");
343 if (devnum
!= NoMdDev
&&
344 (stat("/dev/.udev", &stb
) != 0 ||
345 check_env("MDADM_NO_UDEV"))) {
346 struct map_ent
*mp
= map_by_devnum(&map
, devnum
);
347 remove_devices(devnum
, mp
? mp
->path
: NULL
);
351 pr_err("stopped %s\n", devname
);
353 map_remove(&map
, devnum
);
362 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
364 mdu_array_info_t array
;
365 mdu_disk_info_t disk
;
369 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
372 remaining_disks
= array
.nr_disks
;
373 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
374 struct mddev_dev
*new;
377 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
379 if (disk
.major
== 0 && disk
.minor
== 0)
382 if ((disk
.state
& 1) == 0) /* not faulty */
384 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
385 new = xmalloc(sizeof(*new));
386 new->devname
= xstrdup(buf
);
387 new->disposition
= disp
;
388 new->next
= dv
->next
;
394 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
396 mdu_array_info_t array
;
397 mdu_disk_info_t disk
;
401 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
404 remaining_disks
= array
.nr_disks
;
405 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
406 struct mddev_dev
*new;
410 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
412 if (disk
.major
== 0 && disk
.minor
== 0)
415 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
417 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
418 sfd
= dev_open(buf
, O_RDONLY
);
425 /* Probably not detached */
427 new = xmalloc(sizeof(*new));
428 new->devname
= xstrdup(buf
);
429 new->disposition
= disp
;
430 new->next
= dv
->next
;
436 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
437 struct supertype
*dev_st
, struct supertype
*tst
,
439 char *update
, char *devname
, int verbose
,
440 mdu_array_info_t
*array
)
446 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
447 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
449 tst
->ss
->uuid_from_super(tst
, duuid
);
451 /* Assume uuid matches: kernel will check */
452 memcpy(duuid
, ouuid
, sizeof(ouuid
));
453 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
454 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
455 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
456 /* Looks like it is worth a
457 * try. Need to make sure
458 * kernel will accept it
461 mdu_disk_info_t disc
;
462 /* re-add doesn't work for version-1 superblocks
465 if (array
->major_version
== 1 &&
466 get_linux_version() <= 2006018)
468 disc
.number
= mdi
.disk
.number
;
469 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
470 || disc
.major
!= 0 || disc
.minor
!= 0
473 disc
.major
= major(rdev
);
474 disc
.minor
= minor(rdev
);
475 disc
.number
= mdi
.disk
.number
;
476 disc
.raid_disk
= mdi
.disk
.raid_disk
;
477 disc
.state
= mdi
.disk
.state
;
478 if (dv
->writemostly
== 1)
479 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
480 if (dv
->writemostly
== 2)
481 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
482 remove_partitions(tfd
);
483 if (update
|| dv
->writemostly
> 0) {
485 tfd
= dev_open(dv
->devname
, O_RDWR
);
487 pr_err("failed to open %s for"
488 " superblock update during re-add\n", dv
->devname
);
492 if (dv
->writemostly
== 1)
493 rv
= dev_st
->ss
->update_super(
494 dev_st
, NULL
, "writemostly",
495 devname
, verbose
, 0, NULL
);
496 if (dv
->writemostly
== 2)
497 rv
= dev_st
->ss
->update_super(
498 dev_st
, NULL
, "readwrite",
499 devname
, verbose
, 0, NULL
);
501 rv
= dev_st
->ss
->update_super(
502 dev_st
, NULL
, update
,
503 devname
, verbose
, 0, NULL
);
505 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
508 pr_err("failed to update"
509 " superblock during re-add\n");
513 /* don't even try if disk is marked as faulty */
515 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
517 pr_err("re-added %s\n", dv
->devname
);
520 if (errno
== ENOMEM
|| errno
== EROFS
) {
521 pr_err("add new device failed for %s: %s\n",
522 dv
->devname
, strerror(errno
));
523 if (dv
->disposition
== 'M')
532 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
533 struct supertype
*tst
, mdu_array_info_t
*array
,
534 int force
, int verbose
, char *devname
,
535 char *update
, unsigned long rdev
, unsigned long long array_size
)
537 unsigned long long ldsize
;
538 struct supertype
*dev_st
= NULL
;
540 mdu_disk_info_t disc
;
542 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
543 if (dv
->disposition
== 'M')
549 if (tst
->ss
->validate_geometry(
550 tst
, array
->level
, array
->layout
,
551 array
->raid_disks
, NULL
,
552 ldsize
>> 9, INVALID_SECTORS
, NULL
, NULL
, 0) == 0) {
554 pr_err("%s is larger than %s can "
556 " Add --force is you "
557 "really want to add this device.\n",
558 dv
->devname
, devname
);
561 pr_err("%s is larger than %s can "
563 " Adding anyway as --force "
565 dv
->devname
, devname
);
567 if (!tst
->ss
->external
&&
568 array
->major_version
== 0 &&
569 md_get_version(fd
)%100 < 2) {
570 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
572 pr_err("hot added %s\n",
577 pr_err("hot add failed for %s: %s\n",
578 dv
->devname
, strerror(errno
));
582 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
584 /* need to find a sample superblock to copy, and
585 * a spare slot to use.
586 * For 'external' array (well, container based),
587 * We can just load the metadata for the array->
591 /* already loaded */;
592 else if (tst
->ss
->external
) {
593 tst
->ss
->load_container(tst
, fd
, NULL
);
594 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
598 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
600 if (disc
.major
==0 && disc
.minor
==0)
602 if ((disc
.state
& 4)==0) /* sync */
604 /* Looks like a good device to try */
605 dev
= map_dev(disc
.major
, disc
.minor
, 1);
608 dfd
= dev_open(dev
, O_RDONLY
);
611 if (tst
->ss
->load_super(tst
, dfd
,
619 /* FIXME this is a bad test to be using */
620 if (!tst
->sb
&& dv
->disposition
!= 'a') {
621 /* we are re-adding a device to a
622 * completely dead array - have to depend
625 } else if (!tst
->sb
) {
626 pr_err("cannot load array metadata from %s\n", devname
);
630 /* Make sure device is large enough */
631 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
633 if (dv
->disposition
== 'M')
635 pr_err("%s not large enough to join array\n",
640 /* Possibly this device was recently part of
641 * the array and was temporarily removed, and
642 * is now being re-added. If so, we can
646 if (array
->not_persistent
==0) {
647 dev_st
= dup_super(tst
);
648 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
650 if (dev_st
&& dev_st
->sb
) {
651 int rv
= attempt_re_add(fd
, tfd
, dv
,
657 dev_st
->ss
->free_super(dev_st
);
661 if (dv
->disposition
== 'M') {
663 pr_err("--re-add for %s to %s is not possible\n",
664 dv
->devname
, devname
);
667 if (dv
->disposition
== 'A') {
668 pr_err("--re-add for %s to %s is not possible\n",
669 dv
->devname
, devname
);
672 if (array
->active_disks
< array
->raid_disks
) {
673 char *avail
= xcalloc(array
->raid_disks
, 1);
677 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
679 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
681 if (disc
.major
== 0 && disc
.minor
== 0)
683 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
685 avail
[disc
.raid_disk
] = 1;
688 array_failed
= !enough(array
->level
, array
->raid_disks
,
689 array
->layout
, 1, avail
);
693 pr_err("%s has failed so using --add cannot work and might destroy\n",
695 pr_err("data on %s. You should stop the array and re-assemble it.\n",
700 /* non-persistent. Must ensure that new drive
701 * is at least array->size big.
703 if (ldsize
/512 < array_size
) {
704 pr_err("%s not large enough to join array\n",
709 /* committed to really trying this device now*/
710 remove_partitions(tfd
);
712 /* in 2.6.17 and earlier, version-1 superblocks won't
713 * use the number we write, but will choose a free number.
714 * we must choose the same free number, which requires
715 * starting at 'raid_disks' and counting up
717 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
719 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
721 if (disc
.major
==0 && disc
.minor
==0)
723 if (disc
.state
& 8) /* removed */
726 disc
.major
= major(rdev
);
727 disc
.minor
= minor(rdev
);
730 if (array
->not_persistent
==0) {
732 if (dv
->writemostly
== 1)
733 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
734 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
735 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
736 dv
->devname
, INVALID_SECTORS
))
738 if (tst
->ss
->write_init_super(tst
))
740 } else if (dv
->disposition
== 'A') {
741 /* this had better be raid1.
742 * As we are "--re-add"ing we must find a spare slot
745 char *used
= xcalloc(array
->raid_disks
, 1);
746 for (j
= 0; j
< tst
->max_devs
; j
++) {
747 mdu_disk_info_t disc2
;
749 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
751 if (disc2
.major
==0 && disc2
.minor
==0)
753 if (disc2
.state
& 8) /* removed */
755 if (disc2
.raid_disk
< 0)
757 if (disc2
.raid_disk
> array
->raid_disks
)
759 used
[disc2
.raid_disk
] = 1;
761 for (j
= 0 ; j
< array
->raid_disks
; j
++)
764 disc
.state
|= (1<<MD_DISK_SYNC
);
769 if (dv
->writemostly
== 1)
770 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
771 if (tst
->ss
->external
) {
773 * to an external metadata container */
774 struct mdinfo new_mdi
;
777 int devnum
= fd2devnum(fd
);
780 container_fd
= open_dev_excl(devnum
);
781 if (container_fd
< 0) {
782 pr_err("add failed for %s:"
783 " could not get exclusive access to container\n",
785 tst
->ss
->free_super(tst
);
789 Kill(dv
->devname
, NULL
, 0, -1, 0);
790 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
791 if (mdmon_running(tst
->container_dev
))
792 tst
->update_tail
= &tst
->updates
;
793 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
794 dv
->devname
, INVALID_SECTORS
)) {
799 if (tst
->update_tail
)
800 flush_metadata_updates(tst
);
802 tst
->ss
->sync_metadata(tst
);
804 sra
= sysfs_read(container_fd
, -1, 0);
806 pr_err("add failed for %s: sysfs_read failed\n",
809 tst
->ss
->free_super(tst
);
812 sra
->array
.level
= LEVEL_CONTAINER
;
813 /* Need to set data_offset and component_size */
814 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
815 new_mdi
.disk
.major
= disc
.major
;
816 new_mdi
.disk
.minor
= disc
.minor
;
817 new_mdi
.recovery_start
= 0;
818 /* Make sure fds are closed as they are O_EXCL which
819 * would block add_disk */
820 tst
->ss
->free_super(tst
);
821 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
822 pr_err("add new device to external metadata"
823 " failed for %s\n", dv
->devname
);
828 ping_monitor_by_id(devnum
);
832 tst
->ss
->free_super(tst
);
833 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
834 pr_err("add new device failed for %s as %d: %s\n",
835 dv
->devname
, j
, strerror(errno
));
840 pr_err("added %s\n", dv
->devname
);
844 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
845 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
850 if (tst
->ss
->external
) {
851 /* To remove a device from a container, we must
852 * check that it isn't in use in an array.
853 * This involves looking in the 'holders'
854 * directory - there must be just one entry,
856 * To ensure that it doesn't get used as a
857 * hot spare while we are checking, we
858 * get an O_EXCL open on the container
861 int dnum
= fd2devnum(fd
);
862 lfd
= open_dev_excl(dnum
);
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(dnum
, 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
, 0, 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 *name
= devnum2devname(fd2devnum(fd
));
942 pr_err("unable to get container name\n");
952 pr_err("hot removed %s from %s\n",
953 dv
->devname
, devname
);
957 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
958 unsigned long rdev
, int verbose
, char *devname
)
960 struct mdinfo
*mdi
, *di
;
961 if (tst
->ss
->external
) {
962 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
965 /* Need to find the device in sysfs and add 'want_replacement' to the
968 mdi
= sysfs_read(fd
, -1, GET_DEVS
);
969 if (!mdi
|| !mdi
->devs
) {
970 pr_err("Cannot find status of %s to enable replacement - strange\n",
974 for (di
= mdi
->devs
; di
; di
= di
->next
)
975 if (di
->disk
.major
== (int)major(rdev
) &&
976 di
->disk
.minor
== (int)minor(rdev
))
980 if (di
->disk
.raid_disk
< 0) {
981 pr_err("%s is not active and so cannot be replaced.\n",
986 rv
= sysfs_set_str(mdi
, di
,
987 "state", "want_replacement");
990 pr_err("Failed to request replacement for %s\n",
995 pr_err("Marked %s (device %d in %s) for replacement\n",
996 dv
->devname
, di
->disk
.raid_disk
, devname
);
997 /* If there is a matching 'with', we need to tell it which
1000 while (dv
&& dv
->disposition
!= 'W')
1003 dv
->disposition
= 'w';
1004 dv
->used
= di
->disk
.raid_disk
;
1009 pr_err("%s not found in %s so cannot --replace it\n",
1010 dv
->devname
, devname
);
1014 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1015 unsigned long rdev
, int verbose
, char *devname
)
1017 struct mdinfo
*mdi
, *di
;
1018 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1019 mdi
= sysfs_read(fd
, -1, GET_DEVS
|GET_STATE
);
1020 if (!mdi
|| !mdi
->devs
) {
1021 pr_err("Cannot find status of %s to enable replacement - strange\n",
1025 for (di
= mdi
->devs
; di
; di
= di
->next
)
1026 if (di
->disk
.major
== (int)major(rdev
) &&
1027 di
->disk
.minor
== (int)minor(rdev
))
1031 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1032 pr_err("%s is faulty and cannot be a replacement\n",
1037 if (di
->disk
.raid_disk
>= 0) {
1038 pr_err("%s is active and cannot be a replacement\n",
1043 rv
= sysfs_set_num(mdi
, di
,
1047 pr_err("Failed to %s as preferred replacement.\n",
1052 pr_err("Marked %s in %s as replacement for device %d\n",
1053 dv
->devname
, devname
, dv
->used
);
1057 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1058 dv
->devname
, devname
);
1062 int Manage_subdevs(char *devname
, int fd
,
1063 struct mddev_dev
*devlist
, int verbose
, int test
,
1064 char *update
, int force
)
1066 /* Do something to each dev.
1068 * 'a' - add the device
1070 * If that fails EINVAL, try ADD_NEW_DISK
1071 * 'A' - re-add the device
1072 * 'r' - remove the device: HOT_REMOVE_DISK
1073 * device can be 'faulty' or 'detached' in which case all
1074 * matching devices are removed.
1075 * 'f' - set the device faulty SET_DISK_FAULTY
1076 * device can be 'detached' in which case any device that
1077 * is inaccessible will be marked faulty.
1078 * 'R' - mark this device as wanting replacement.
1079 * 'W' - this device is added if necessary and activated as
1080 * a replacement for a previous 'R' device.
1082 * 'w' - 'W' will be changed to 'w' when it is paired with
1083 * a 'R' device. If a 'W' is found while walking the list
1084 * it must be unpaired, and is an error.
1085 * 'M' - this is created by a 'missing' target. It is a slight
1087 * 'F' - Another variant of 'A', where the device was faulty
1088 * so must be removed from the array first.
1090 * For 'f' and 'r', the device can also be a kernel-internal
1091 * name such as 'sdb'.
1093 mdu_array_info_t array
;
1094 unsigned long long array_size
;
1095 struct mddev_dev
*dv
;
1098 struct supertype
*tst
;
1099 char *subarray
= NULL
;
1101 int count
= 0; /* number of actions taken */
1105 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1106 pr_err("Cannot get array info for %s\n",
1110 sysfs_init(&info
, fd
, 0);
1112 /* array.size is only 32 bits and may be truncated.
1113 * So read from sysfs if possible, and record number of sectors
1116 array_size
= get_component_size(fd
);
1117 if (array_size
<= 0)
1118 array_size
= array
.size
* 2;
1120 tst
= super_by_fd(fd
, &subarray
);
1122 pr_err("unsupport array - version %d.%d\n",
1123 array
.major_version
, array
.minor_version
);
1128 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1131 if (strcmp(dv
->devname
, "failed") == 0 ||
1132 strcmp(dv
->devname
, "faulty") == 0) {
1133 if (dv
->disposition
!= 'A'
1134 && dv
->disposition
!= 'r') {
1135 pr_err("%s only meaningful "
1136 "with -r or --re-add, not -%c\n",
1137 dv
->devname
, dv
->disposition
);
1140 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1144 if (strcmp(dv
->devname
, "detached") == 0) {
1145 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1146 pr_err("%s only meaningful "
1147 "with -r of -f, not -%c\n",
1148 dv
->devname
, dv
->disposition
);
1151 add_detached(dv
, fd
, dv
->disposition
);
1155 if (strcmp(dv
->devname
, "missing") == 0) {
1156 struct mddev_dev
*add_devlist
= NULL
;
1157 struct mddev_dev
**dp
;
1158 if (dv
->disposition
!= 'A') {
1159 pr_err("'missing' only meaningful "
1163 add_devlist
= conf_get_devs();
1164 if (add_devlist
== NULL
) {
1165 pr_err("no devices to scan for missing members.");
1168 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1169 /* 'M' (for 'missing') is like 'A' without errors */
1170 (*dp
)->disposition
= 'M';
1172 dv
->next
= add_devlist
;
1176 if (strchr(dv
->devname
, '/') == NULL
&&
1177 strchr(dv
->devname
, ':') == NULL
&&
1178 strlen(dv
->devname
) < 50) {
1179 /* Assume this is a kernel-internal name like 'sda1' */
1182 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1183 pr_err("%s only meaningful "
1184 "with -r or -f, not -%c\n",
1185 dv
->devname
, dv
->disposition
);
1189 sprintf(dname
, "dev-%s", dv
->devname
);
1190 sysfd
= sysfs_open(fd2devnum(fd
), dname
, "block/dev");
1194 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1195 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1196 stb
.st_rdev
= makedev(mj
,mn
);
1203 sysfd
= sysfs_open(fd2devnum(fd
), dname
, "state");
1205 pr_err("%s does not appear "
1206 "to be a component of %s\n",
1207 dv
->devname
, devname
);
1212 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1216 int open_err
= errno
;
1217 if (stat(dv
->devname
, &stb
) != 0) {
1218 pr_err("Cannot find %s: %s\n",
1219 dv
->devname
, strerror(errno
));
1222 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1223 if (dv
->disposition
== 'M')
1224 /* non-fatal. Also improbable */
1226 pr_err("%s is not a block device.\n",
1230 if (dv
->disposition
== 'r')
1231 /* Be happy, the stat worked, that is
1232 * enough for --remove
1236 if (dv
->disposition
== 'M')
1239 pr_err("Cannot open %s: %s\n",
1240 dv
->devname
, strerror(open_err
));
1245 switch(dv
->disposition
){
1247 pr_err("internal error - devmode[%s]=%d\n",
1248 dv
->devname
, dv
->disposition
);
1252 case 'M': /* --re-add missing */
1253 case 'F': /* --re-add faulty */
1254 /* add the device */
1256 pr_err("Cannot add disks to a"
1257 " \'member\' array, perform this"
1258 " operation on the parent container\n");
1261 if (dv
->disposition
== 'F')
1262 /* Need to remove first */
1263 ioctl(fd
, HOT_REMOVE_DISK
,
1264 (unsigned long)stb
.st_rdev
);
1265 /* Make sure it isn't in use (in 2.6 or later) */
1266 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1268 /* We know no-one else is using it. We'll
1269 * need non-exclusive access to add it, so
1273 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1276 if (dv
->disposition
== 'M')
1278 pr_err("Cannot open %s: %s\n",
1279 dv
->devname
, strerror(errno
));
1283 if (sysfs_freeze_array(&info
) == 1)
1288 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1289 force
, verbose
, devname
, update
,
1290 stb
.st_rdev
, array_size
);
1302 pr_err("Cannot remove disks from a"
1303 " \'member\' array, perform this"
1304 " operation on the parent container\n");
1307 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1308 stb
.st_rdev
, verbose
,
1319 case 'f': /* set faulty */
1320 /* FIXME check current member */
1321 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1322 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1323 (unsigned long) stb
.st_rdev
))) {
1324 pr_err("set device faulty failed for %s: %s\n",
1325 dv
->devname
, strerror(errno
));
1335 pr_err("set %s faulty in %s\n",
1336 dv
->devname
, devname
);
1338 case 'R': /* Mark as replaceable */
1340 pr_err("Cannot replace disks in a"
1341 " \'member\' array, perform this"
1342 " operation on the parent container\n");
1346 if (sysfs_freeze_array(&info
) == 1)
1351 rv
= Manage_replace(tst
, fd
, dv
,
1352 stb
.st_rdev
, verbose
,
1360 case 'W': /* --with device that doesn't match */
1361 pr_err("No matching --replace device for --with %s\n",
1364 case 'w': /* --with device which was matched */
1365 rv
= Manage_with(tst
, fd
, dv
,
1366 stb
.st_rdev
, verbose
, devname
);
1373 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1374 if (test
&& count
== 0)
1380 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1384 int autodetect(void)
1386 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1388 int fd
= dev_open("9:0", O_RDONLY
);
1390 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1397 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1399 struct supertype supertype
, *st
= &supertype
;
1402 memset(st
, 0, sizeof(*st
));
1404 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1408 if (!st
->ss
->update_subarray
) {
1410 pr_err("Operation not supported for %s metadata\n",
1415 if (mdmon_running(st
->devnum
))
1416 st
->update_tail
= &st
->updates
;
1418 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1422 pr_err("Failed to update %s of subarray-%s in %s\n",
1423 update
, subarray
, dev
);
1424 } else if (st
->update_tail
)
1425 flush_metadata_updates(st
);
1427 st
->ss
->sync_metadata(st
);
1429 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1430 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1434 st
->ss
->free_super(st
);
1440 /* Move spare from one array to another If adding to destination array fails
1441 * add back to original array.
1442 * Returns 1 on success, 0 on failure */
1443 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1445 struct mddev_dev devlist
;
1448 /* try to remove and add */
1449 int fd1
= open(to_devname
, O_RDONLY
);
1450 int fd2
= open(from_devname
, O_RDONLY
);
1452 if (fd1
< 0 || fd2
< 0) {
1453 if (fd1
>=0) close(fd1
);
1454 if (fd2
>=0) close(fd2
);
1458 devlist
.next
= NULL
;
1460 devlist
.writemostly
= 0;
1461 devlist
.devname
= devname
;
1462 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1464 devlist
.disposition
= 'r';
1465 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1466 devlist
.disposition
= 'a';
1467 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1468 /* make sure manager is aware of changes */
1469 ping_manager(to_devname
);
1470 ping_manager(from_devname
);
1475 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);