]>
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.
224 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
);
225 if (mdi
&& is_subarray(mdi
->text_version
)) {
227 strncpy(container
, mdi
->text_version
+1, sizeof(container
));
228 container
[sizeof(container
)-1] = 0;
229 sl
= strchr(container
, '/');
236 while (((fd
= ((devnm
[0] == '/')
237 ?open(devname
, O_RDONLY
|O_EXCL
)
238 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0
239 || strcmp(fd2devnm(fd
), devnm
) != 0)
241 && mdmon_running(container
)
245 flush_mdmon(container
);
248 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
252 pr_err("Cannot get exclusive access to %s:"
254 "process, mounted filesystem "
255 "or active volume group?\n",
260 mdi
->array
.level
> 0 &&
261 is_subarray(mdi
->text_version
)) {
263 /* This is mdmon managed. */
266 /* As we have an O_EXCL open, any use of the device
267 * which blocks STOP_ARRAY is probably a transient use,
268 * so it is reasonable to retry for a while - 5 seconds.
272 (err
= sysfs_set_str(mdi
, NULL
,
281 pr_err("failed to stop array %s: %s\n",
282 devname
, strerror(errno
));
287 /* Give monitor a chance to act */
288 ping_monitor(mdi
->text_version
);
290 fd
= open_dev_excl(devnm
);
293 pr_err("failed to completely stop %s"
294 ": Device is busy\n",
300 mdi
->array
.major_version
== -1 &&
301 mdi
->array
.minor_version
== -2 &&
302 !is_subarray(mdi
->text_version
)) {
303 struct mdstat_ent
*mds
, *m
;
304 /* container, possibly mdmon-managed.
305 * Make sure mdmon isn't opening it, which
306 * would interfere with the 'stop'
308 ping_monitor(mdi
->sys_name
);
310 /* now check that there are no existing arrays
311 * which are members of this array
313 mds
= mdstat_read(0, 0);
314 for (m
= mds
; m
; m
= m
->next
)
315 if (m
->metadata_version
&&
316 strncmp(m
->metadata_version
, "external:", 9)==0 &&
317 metadata_container_matches(m
->metadata_version
+9,
320 pr_err("Cannot stop container %s: "
321 "member %s still active\n",
329 /* As we have an O_EXCL open, any use of the device
330 * which blocks STOP_ARRAY is probably a transient use,
331 * so it is reasonable to retry for a while - 5 seconds.
334 while (count
&& fd
>= 0
335 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
340 if (fd
>= 0 && err
) {
342 pr_err("failed to stop array %s: %s\n",
343 devname
, strerror(errno
));
345 fprintf(stderr
, "Perhaps a running "
346 "process, mounted filesystem "
347 "or active volume group?\n");
352 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
353 * was stopped, so We'll do it here just to be sure. Drop any
354 * partitions as well...
357 ioctl(fd
, BLKRRPART
, 0);
359 sysfs_uevent(mdi
, "change");
361 if (devnm
[0] && use_udev()) {
362 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
363 remove_devices(devnm
, mp
? mp
->path
: NULL
);
367 pr_err("stopped %s\n", devname
);
369 map_remove(&map
, devnm
);
378 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
380 struct mddev_dev
*new;
381 new = xmalloc(sizeof(*new));
382 memset(new, 0, sizeof(*new));
383 new->devname
= xstrdup(name
);
384 new->disposition
= disp
;
385 new->next
= dv
->next
;
390 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
392 mdu_array_info_t array
;
393 mdu_disk_info_t disk
;
397 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
400 remaining_disks
= array
.nr_disks
;
401 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
404 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
406 if (disk
.major
== 0 && disk
.minor
== 0)
409 if ((disk
.state
& 1) == 0) /* not faulty */
411 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
412 dv
= add_one(dv
, buf
, disp
);
416 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
418 mdu_array_info_t array
;
419 mdu_disk_info_t disk
;
423 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
426 remaining_disks
= array
.nr_disks
;
427 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
431 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
433 if (disk
.major
== 0 && disk
.minor
== 0)
436 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
438 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
439 sfd
= dev_open(buf
, O_RDONLY
);
446 /* Probably not detached */
448 dv
= add_one(dv
, buf
, disp
);
452 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
454 mdu_array_info_t array
;
455 mdu_disk_info_t disk
;
460 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
462 if (array
.level
!= 10)
464 copies
= ((array
.layout
& 0xff) *
465 ((array
.layout
>> 8) & 0xff));
466 if (array
.raid_disks
% copies
)
469 remaining_disks
= array
.nr_disks
;
470 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
473 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
475 if (disk
.major
== 0 && disk
.minor
== 0)
478 set
= disk
.raid_disk
% copies
;
479 if (set_char
!= set
+ 'A')
481 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
482 dv
= add_one(dv
, buf
, dv
->disposition
);
486 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
487 struct supertype
*dev_st
, struct supertype
*tst
,
489 char *update
, char *devname
, int verbose
,
490 mdu_array_info_t
*array
)
496 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
497 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
499 tst
->ss
->uuid_from_super(tst
, duuid
);
501 /* Assume uuid matches: kernel will check */
502 memcpy(duuid
, ouuid
, sizeof(ouuid
));
503 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
504 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
505 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
506 /* Looks like it is worth a
507 * try. Need to make sure
508 * kernel will accept it
511 mdu_disk_info_t disc
;
512 /* re-add doesn't work for version-1 superblocks
515 if (array
->major_version
== 1 &&
516 get_linux_version() <= 2006018)
518 disc
.number
= mdi
.disk
.number
;
519 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
520 || disc
.major
!= 0 || disc
.minor
!= 0
523 disc
.major
= major(rdev
);
524 disc
.minor
= minor(rdev
);
525 disc
.number
= mdi
.disk
.number
;
526 disc
.raid_disk
= mdi
.disk
.raid_disk
;
527 disc
.state
= mdi
.disk
.state
;
528 if (dv
->writemostly
== 1)
529 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
530 if (dv
->writemostly
== 2)
531 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
532 remove_partitions(tfd
);
533 if (update
|| dv
->writemostly
> 0) {
535 tfd
= dev_open(dv
->devname
, O_RDWR
);
537 pr_err("failed to open %s for"
538 " superblock update during re-add\n", dv
->devname
);
542 if (dv
->writemostly
== 1)
543 rv
= dev_st
->ss
->update_super(
544 dev_st
, NULL
, "writemostly",
545 devname
, verbose
, 0, NULL
);
546 if (dv
->writemostly
== 2)
547 rv
= dev_st
->ss
->update_super(
548 dev_st
, NULL
, "readwrite",
549 devname
, verbose
, 0, NULL
);
551 rv
= dev_st
->ss
->update_super(
552 dev_st
, NULL
, update
,
553 devname
, verbose
, 0, NULL
);
555 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
558 pr_err("failed to update"
559 " superblock during re-add\n");
563 /* don't even try if disk is marked as faulty */
565 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
567 pr_err("re-added %s\n", dv
->devname
);
570 if (errno
== ENOMEM
|| errno
== EROFS
) {
571 pr_err("add new device failed for %s: %s\n",
572 dv
->devname
, strerror(errno
));
573 if (dv
->disposition
== 'M')
582 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
583 struct supertype
*tst
, mdu_array_info_t
*array
,
584 int force
, int verbose
, char *devname
,
585 char *update
, unsigned long rdev
, unsigned long long array_size
)
587 unsigned long long ldsize
;
588 struct supertype
*dev_st
= NULL
;
590 mdu_disk_info_t disc
;
592 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
593 if (dv
->disposition
== 'M')
599 if (tst
->ss
->validate_geometry(
600 tst
, array
->level
, array
->layout
,
601 array
->raid_disks
, NULL
,
602 ldsize
>> 9, INVALID_SECTORS
, NULL
, NULL
, 0) == 0) {
604 pr_err("%s is larger than %s can "
606 " Add --force is you "
607 "really want to add this device.\n",
608 dv
->devname
, devname
);
611 pr_err("%s is larger than %s can "
613 " Adding anyway as --force "
615 dv
->devname
, devname
);
617 if (!tst
->ss
->external
&&
618 array
->major_version
== 0 &&
619 md_get_version(fd
)%100 < 2) {
620 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
622 pr_err("hot added %s\n",
627 pr_err("hot add failed for %s: %s\n",
628 dv
->devname
, strerror(errno
));
632 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
634 /* need to find a sample superblock to copy, and
635 * a spare slot to use.
636 * For 'external' array (well, container based),
637 * We can just load the metadata for the array->
641 /* already loaded */;
642 else if (tst
->ss
->external
) {
643 tst
->ss
->load_container(tst
, fd
, NULL
);
644 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
648 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
650 if (disc
.major
==0 && disc
.minor
==0)
652 if ((disc
.state
& 4)==0) /* sync */
654 /* Looks like a good device to try */
655 dev
= map_dev(disc
.major
, disc
.minor
, 1);
658 dfd
= dev_open(dev
, O_RDONLY
);
661 if (tst
->ss
->load_super(tst
, dfd
,
669 /* FIXME this is a bad test to be using */
670 if (!tst
->sb
&& dv
->disposition
!= 'a') {
671 /* we are re-adding a device to a
672 * completely dead array - have to depend
675 } else if (!tst
->sb
) {
676 pr_err("cannot load array metadata from %s\n", devname
);
680 /* Make sure device is large enough */
681 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
683 if (dv
->disposition
== 'M')
685 pr_err("%s not large enough to join array\n",
690 /* Possibly this device was recently part of
691 * the array and was temporarily removed, and
692 * is now being re-added. If so, we can
696 if (array
->not_persistent
==0) {
697 dev_st
= dup_super(tst
);
698 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
700 if (dev_st
&& dev_st
->sb
) {
701 int rv
= attempt_re_add(fd
, tfd
, dv
,
707 dev_st
->ss
->free_super(dev_st
);
711 if (dv
->disposition
== 'M') {
713 pr_err("--re-add for %s to %s is not possible\n",
714 dv
->devname
, devname
);
717 if (dv
->disposition
== 'A') {
718 pr_err("--re-add for %s to %s is not possible\n",
719 dv
->devname
, devname
);
722 if (array
->active_disks
< array
->raid_disks
) {
723 char *avail
= xcalloc(array
->raid_disks
, 1);
727 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
729 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
731 if (disc
.major
== 0 && disc
.minor
== 0)
733 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
735 avail
[disc
.raid_disk
] = 1;
738 array_failed
= !enough(array
->level
, array
->raid_disks
,
739 array
->layout
, 1, avail
);
743 pr_err("%s has failed so using --add cannot work and might destroy\n",
745 pr_err("data on %s. You should stop the array and re-assemble it.\n",
750 /* non-persistent. Must ensure that new drive
751 * is at least array->size big.
753 if (ldsize
/512 < array_size
) {
754 pr_err("%s not large enough to join array\n",
759 /* committed to really trying this device now*/
760 remove_partitions(tfd
);
762 /* in 2.6.17 and earlier, version-1 superblocks won't
763 * use the number we write, but will choose a free number.
764 * we must choose the same free number, which requires
765 * starting at 'raid_disks' and counting up
767 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
769 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
771 if (disc
.major
==0 && disc
.minor
==0)
773 if (disc
.state
& 8) /* removed */
776 disc
.major
= major(rdev
);
777 disc
.minor
= minor(rdev
);
780 if (array
->not_persistent
==0) {
782 if (dv
->writemostly
== 1)
783 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
784 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
785 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
786 dv
->devname
, INVALID_SECTORS
))
788 if (tst
->ss
->write_init_super(tst
))
790 } else if (dv
->disposition
== 'A') {
791 /* this had better be raid1.
792 * As we are "--re-add"ing we must find a spare slot
795 char *used
= xcalloc(array
->raid_disks
, 1);
796 for (j
= 0; j
< tst
->max_devs
; j
++) {
797 mdu_disk_info_t disc2
;
799 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
801 if (disc2
.major
==0 && disc2
.minor
==0)
803 if (disc2
.state
& 8) /* removed */
805 if (disc2
.raid_disk
< 0)
807 if (disc2
.raid_disk
> array
->raid_disks
)
809 used
[disc2
.raid_disk
] = 1;
811 for (j
= 0 ; j
< array
->raid_disks
; j
++)
814 disc
.state
|= (1<<MD_DISK_SYNC
);
819 if (dv
->writemostly
== 1)
820 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
821 if (tst
->ss
->external
) {
823 * to an external metadata container */
824 struct mdinfo new_mdi
;
830 strcpy(devnm
, fd2devnm(fd
));
832 container_fd
= open_dev_excl(devnm
);
833 if (container_fd
< 0) {
834 pr_err("add failed for %s:"
835 " could not get exclusive access to container\n",
837 tst
->ss
->free_super(tst
);
841 Kill(dv
->devname
, NULL
, 0, -1, 0);
842 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
843 if (mdmon_running(tst
->container_devnm
))
844 tst
->update_tail
= &tst
->updates
;
845 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
846 dv
->devname
, INVALID_SECTORS
)) {
851 if (tst
->update_tail
)
852 flush_metadata_updates(tst
);
854 tst
->ss
->sync_metadata(tst
);
856 sra
= sysfs_read(container_fd
, NULL
, 0);
858 pr_err("add failed for %s: sysfs_read failed\n",
861 tst
->ss
->free_super(tst
);
864 sra
->array
.level
= LEVEL_CONTAINER
;
865 /* Need to set data_offset and component_size */
866 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
867 new_mdi
.disk
.major
= disc
.major
;
868 new_mdi
.disk
.minor
= disc
.minor
;
869 new_mdi
.recovery_start
= 0;
870 /* Make sure fds are closed as they are O_EXCL which
871 * would block add_disk */
872 tst
->ss
->free_super(tst
);
873 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
874 pr_err("add new device to external metadata"
875 " failed for %s\n", dv
->devname
);
884 tst
->ss
->free_super(tst
);
885 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
886 pr_err("add new device failed for %s as %d: %s\n",
887 dv
->devname
, j
, strerror(errno
));
892 pr_err("added %s\n", dv
->devname
);
896 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
897 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
902 if (tst
->ss
->external
) {
903 /* To remove a device from a container, we must
904 * check that it isn't in use in an array.
905 * This involves looking in the 'holders'
906 * directory - there must be just one entry,
908 * To ensure that it doesn't get used as a
909 * hot spare while we are checking, we
910 * get an O_EXCL open on the container
914 strcpy(devnm
, fd2devnm(fd
));
915 lfd
= open_dev_excl(devnm
);
917 pr_err("Cannot get exclusive access "
918 " to container - odd\n");
921 /* We may not be able to check on holders in
922 * sysfs, either because we don't have the dev num
923 * (rdev == 0) or because the device has been detached
924 * and the 'holders' directory no longer exists
925 * (ret == -1). In that case, assume it is OK to
931 ret
= sysfs_unique_holder(devnm
, rdev
);
933 pr_err("%s is not a member, cannot remove.\n",
939 pr_err("%s is still in use, cannot remove.\n",
945 /* FIXME check that it is a current member */
947 /* device has been removed and we don't know
948 * the major:minor number
950 int n
= write(sysfd
, "remove", 6);
956 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
957 if (err
&& errno
== ENODEV
) {
958 /* Old kernels rejected this if no personality
960 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
961 struct mdinfo
*dv
= NULL
;
964 for ( ; dv
; dv
=dv
->next
)
965 if (dv
->disk
.major
== (int)major(rdev
) &&
966 dv
->disk
.minor
== (int)minor(rdev
))
969 err
= sysfs_set_str(sra
, dv
,
978 pr_err("hot remove failed "
979 "for %s: %s\n", dv
->devname
,
985 if (tst
->ss
->external
) {
987 * Before dropping our exclusive open we make an
988 * attempt at preventing mdmon from seeing an
989 * 'add' event before reconciling this 'remove'
992 char *devnm
= fd2devnm(fd
);
995 pr_err("unable to get container name\n");
1004 pr_err("hot removed %s from %s\n",
1005 dv
->devname
, devname
);
1009 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1010 unsigned long rdev
, int verbose
, char *devname
)
1012 struct mdinfo
*mdi
, *di
;
1013 if (tst
->ss
->external
) {
1014 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1017 /* Need to find the device in sysfs and add 'want_replacement' to the
1020 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1021 if (!mdi
|| !mdi
->devs
) {
1022 pr_err("Cannot find status of %s to enable replacement - strange\n",
1026 for (di
= mdi
->devs
; di
; di
= di
->next
)
1027 if (di
->disk
.major
== (int)major(rdev
) &&
1028 di
->disk
.minor
== (int)minor(rdev
))
1032 if (di
->disk
.raid_disk
< 0) {
1033 pr_err("%s is not active and so cannot be replaced.\n",
1038 rv
= sysfs_set_str(mdi
, di
,
1039 "state", "want_replacement");
1042 pr_err("Failed to request replacement for %s\n",
1047 pr_err("Marked %s (device %d in %s) for replacement\n",
1048 dv
->devname
, di
->disk
.raid_disk
, devname
);
1049 /* If there is a matching 'with', we need to tell it which
1052 while (dv
&& dv
->disposition
!= 'W')
1055 dv
->disposition
= 'w';
1056 dv
->used
= di
->disk
.raid_disk
;
1061 pr_err("%s not found in %s so cannot --replace it\n",
1062 dv
->devname
, devname
);
1066 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1067 unsigned long rdev
, int verbose
, char *devname
)
1069 struct mdinfo
*mdi
, *di
;
1070 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1071 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1072 if (!mdi
|| !mdi
->devs
) {
1073 pr_err("Cannot find status of %s to enable replacement - strange\n",
1077 for (di
= mdi
->devs
; di
; di
= di
->next
)
1078 if (di
->disk
.major
== (int)major(rdev
) &&
1079 di
->disk
.minor
== (int)minor(rdev
))
1083 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1084 pr_err("%s is faulty and cannot be a replacement\n",
1089 if (di
->disk
.raid_disk
>= 0) {
1090 pr_err("%s is active and cannot be a replacement\n",
1095 rv
= sysfs_set_num(mdi
, di
,
1099 pr_err("Failed to %s as preferred replacement.\n",
1104 pr_err("Marked %s in %s as replacement for device %d\n",
1105 dv
->devname
, devname
, dv
->used
);
1109 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1110 dv
->devname
, devname
);
1114 int Manage_subdevs(char *devname
, int fd
,
1115 struct mddev_dev
*devlist
, int verbose
, int test
,
1116 char *update
, int force
)
1118 /* Do something to each dev.
1120 * 'a' - add the device
1122 * If that fails EINVAL, try ADD_NEW_DISK
1123 * 'A' - re-add the device
1124 * 'r' - remove the device: HOT_REMOVE_DISK
1125 * device can be 'faulty' or 'detached' in which case all
1126 * matching devices are removed.
1127 * 'f' - set the device faulty SET_DISK_FAULTY
1128 * device can be 'detached' in which case any device that
1129 * is inaccessible will be marked faulty.
1130 * 'R' - mark this device as wanting replacement.
1131 * 'W' - this device is added if necessary and activated as
1132 * a replacement for a previous 'R' device.
1134 * 'w' - 'W' will be changed to 'w' when it is paired with
1135 * a 'R' device. If a 'W' is found while walking the list
1136 * it must be unpaired, and is an error.
1137 * 'M' - this is created by a 'missing' target. It is a slight
1139 * 'F' - Another variant of 'A', where the device was faulty
1140 * so must be removed from the array first.
1142 * For 'f' and 'r', the device can also be a kernel-internal
1143 * name such as 'sdb'.
1145 mdu_array_info_t array
;
1146 unsigned long long array_size
;
1147 struct mddev_dev
*dv
;
1150 struct supertype
*tst
;
1151 char *subarray
= NULL
;
1153 int count
= 0; /* number of actions taken */
1158 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1159 pr_err("Cannot get array info for %s\n",
1163 sysfs_init(&info
, fd
, NULL
);
1165 /* array.size is only 32 bits and may be truncated.
1166 * So read from sysfs if possible, and record number of sectors
1169 array_size
= get_component_size(fd
);
1170 if (array_size
<= 0)
1171 array_size
= array
.size
* 2;
1173 tst
= super_by_fd(fd
, &subarray
);
1175 pr_err("unsupport array - version %d.%d\n",
1176 array
.major_version
, array
.minor_version
);
1181 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1184 if (strcmp(dv
->devname
, "failed") == 0 ||
1185 strcmp(dv
->devname
, "faulty") == 0) {
1186 if (dv
->disposition
!= 'A'
1187 && dv
->disposition
!= 'r') {
1188 pr_err("%s only meaningful "
1189 "with -r or --re-add, not -%c\n",
1190 dv
->devname
, dv
->disposition
);
1193 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1197 if (strcmp(dv
->devname
, "detached") == 0) {
1198 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1199 pr_err("%s only meaningful "
1200 "with -r of -f, not -%c\n",
1201 dv
->devname
, dv
->disposition
);
1204 add_detached(dv
, fd
, dv
->disposition
);
1208 if (strcmp(dv
->devname
, "missing") == 0) {
1209 struct mddev_dev
*add_devlist
= NULL
;
1210 struct mddev_dev
**dp
;
1211 if (dv
->disposition
!= 'A') {
1212 pr_err("'missing' only meaningful "
1216 add_devlist
= conf_get_devs();
1217 if (add_devlist
== NULL
) {
1218 pr_err("no devices to scan for missing members.");
1221 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1222 /* 'M' (for 'missing') is like 'A' without errors */
1223 (*dp
)->disposition
= 'M';
1225 dv
->next
= add_devlist
;
1229 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1230 strlen(dv
->devname
) == 5) {
1233 if (dv
->disposition
!= 'r' &&
1234 dv
->disposition
!= 'f') {
1235 pr_err("'%s' only meaningful with -r or -f\n",
1239 if (array
.level
!= 10) {
1240 pr_err("'%s' only meaningful with RAID10 arrays\n",
1244 copies
= ((array
.layout
& 0xff) *
1245 ((array
.layout
>> 8) & 0xff));
1246 if (array
.raid_disks
% copies
!= 0 ||
1247 dv
->devname
[4] < 'A' ||
1248 dv
->devname
[4] >= 'A' + copies
||
1250 pr_err("'%s' not meaningful with this array\n",
1254 add_set(dv
, fd
, dv
->devname
[4]);
1258 if (strchr(dv
->devname
, '/') == NULL
&&
1259 strchr(dv
->devname
, ':') == NULL
&&
1260 strlen(dv
->devname
) < 50) {
1261 /* Assume this is a kernel-internal name like 'sda1' */
1264 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1265 pr_err("%s only meaningful "
1266 "with -r or -f, not -%c\n",
1267 dv
->devname
, dv
->disposition
);
1271 sprintf(dname
, "dev-%s", dv
->devname
);
1272 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1276 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1277 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1278 stb
.st_rdev
= makedev(mj
,mn
);
1285 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1287 pr_err("%s does not appear "
1288 "to be a component of %s\n",
1289 dv
->devname
, devname
);
1294 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1298 int open_err
= errno
;
1299 if (stat(dv
->devname
, &stb
) != 0) {
1300 pr_err("Cannot find %s: %s\n",
1301 dv
->devname
, strerror(errno
));
1304 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1305 if (dv
->disposition
== 'M')
1306 /* non-fatal. Also improbable */
1308 pr_err("%s is not a block device.\n",
1312 if (dv
->disposition
== 'r')
1313 /* Be happy, the stat worked, that is
1314 * enough for --remove
1318 if (dv
->disposition
== 'M')
1321 pr_err("Cannot open %s: %s\n",
1322 dv
->devname
, strerror(open_err
));
1327 switch(dv
->disposition
){
1329 pr_err("internal error - devmode[%s]=%d\n",
1330 dv
->devname
, dv
->disposition
);
1334 case 'M': /* --re-add missing */
1335 case 'F': /* --re-add faulty */
1336 /* add the device */
1338 pr_err("Cannot add disks to a"
1339 " \'member\' array, perform this"
1340 " operation on the parent container\n");
1343 if (dv
->disposition
== 'F')
1344 /* Need to remove first */
1345 ioctl(fd
, HOT_REMOVE_DISK
,
1346 (unsigned long)stb
.st_rdev
);
1347 /* Make sure it isn't in use (in 2.6 or later) */
1348 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1350 /* We know no-one else is using it. We'll
1351 * need non-exclusive access to add it, so
1355 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1358 if (dv
->disposition
== 'M')
1360 pr_err("Cannot open %s: %s\n",
1361 dv
->devname
, strerror(errno
));
1365 if (sysfs_freeze_array(&info
) == 1)
1370 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1371 force
, verbose
, devname
, update
,
1372 stb
.st_rdev
, array_size
);
1384 pr_err("Cannot remove disks from a"
1385 " \'member\' array, perform this"
1386 " operation on the parent container\n");
1389 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1390 stb
.st_rdev
, verbose
,
1401 case 'f': /* set faulty */
1402 /* FIXME check current member */
1403 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1404 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1405 (unsigned long) stb
.st_rdev
))) {
1408 pr_err("set device faulty failed for %s: %s\n",
1409 dv
->devname
, strerror(errno
));
1419 pr_err("set %s faulty in %s\n",
1420 dv
->devname
, devname
);
1422 case 'R': /* Mark as replaceable */
1424 pr_err("Cannot replace disks in a"
1425 " \'member\' array, perform this"
1426 " operation on the parent container\n");
1430 if (sysfs_freeze_array(&info
) == 1)
1435 rv
= Manage_replace(tst
, fd
, dv
,
1436 stb
.st_rdev
, verbose
,
1444 case 'W': /* --with device that doesn't match */
1445 pr_err("No matching --replace device for --with %s\n",
1448 case 'w': /* --with device which was matched */
1449 rv
= Manage_with(tst
, fd
, dv
,
1450 stb
.st_rdev
, verbose
, devname
);
1457 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1458 if (test
&& count
== 0)
1464 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1465 return !test
&& busy
? 2 : 1;
1468 int autodetect(void)
1470 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1472 int fd
= dev_open("9:0", O_RDONLY
);
1474 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1481 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1483 struct supertype supertype
, *st
= &supertype
;
1486 memset(st
, 0, sizeof(*st
));
1488 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1492 if (!st
->ss
->update_subarray
) {
1494 pr_err("Operation not supported for %s metadata\n",
1499 if (mdmon_running(st
->devnm
))
1500 st
->update_tail
= &st
->updates
;
1502 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1506 pr_err("Failed to update %s of subarray-%s in %s\n",
1507 update
, subarray
, dev
);
1508 } else if (st
->update_tail
)
1509 flush_metadata_updates(st
);
1511 st
->ss
->sync_metadata(st
);
1513 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1514 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1518 st
->ss
->free_super(st
);
1524 /* Move spare from one array to another If adding to destination array fails
1525 * add back to original array.
1526 * Returns 1 on success, 0 on failure */
1527 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1529 struct mddev_dev devlist
;
1532 /* try to remove and add */
1533 int fd1
= open(to_devname
, O_RDONLY
);
1534 int fd2
= open(from_devname
, O_RDONLY
);
1536 if (fd1
< 0 || fd2
< 0) {
1537 if (fd1
>=0) close(fd1
);
1538 if (fd2
>=0) close(fd2
);
1542 devlist
.next
= NULL
;
1544 devlist
.writemostly
= 0;
1545 devlist
.devname
= devname
;
1546 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1548 devlist
.disposition
= 'r';
1549 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1550 devlist
.disposition
= 'a';
1551 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1552 /* make sure manager is aware of changes */
1553 ping_manager(to_devname
);
1554 ping_manager(from_devname
);
1559 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);