]>
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_run(char *devname
, int fd
, int verbose
)
175 /* Run the array. Array must already be configured
178 mdu_param_t param
; /* unused */
181 if (md_get_version(fd
) < 9000) {
182 pr_err("need md driver version 0.90.0 or later\n");
186 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
188 pr_err("failed to run array %s: %s\n",
189 devname
, strerror(errno
));
193 pr_err("started %s\n", devname
);
197 int Manage_stop(char *devname
, int fd
, int verbose
, int will_retry
)
199 /* Stop the array. Array must already be configured
200 * 'will_retry' means that error messages are not wanted.
203 struct map_ent
*map
= NULL
;
210 unsigned long long rd1
, rd2
;
212 if (will_retry
&& verbose
== 0)
215 if (md_get_version(fd
) < 9000) {
216 if (ioctl(fd
, STOP_MD
, 0) == 0)
218 pr_err("stopping device %s "
220 devname
, strerror(errno
));
224 /* If this is an mdmon managed array, just write 'inactive'
225 * to the array state and let mdmon clear up.
227 strcpy(devnm
, fd2devnm(fd
));
228 /* Get EXCL access first. If this fails, then attempting
229 * to stop is probably a bad idea.
231 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_COMPONENT
|GET_VERSION
);
232 if (mdi
&& is_subarray(mdi
->text_version
)) {
234 strncpy(container
, mdi
->text_version
+1, sizeof(container
));
235 container
[sizeof(container
)-1] = 0;
236 sl
= strchr(container
, '/');
243 while (((fd
= ((devnm
[0] == '/')
244 ?open(devname
, O_RDONLY
|O_EXCL
)
245 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0
246 || strcmp(fd2devnm(fd
), devnm
) != 0)
248 && mdmon_running(container
)
252 flush_mdmon(container
);
255 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
259 pr_err("Cannot get exclusive access to %s:"
261 "process, mounted filesystem "
262 "or active volume group?\n",
267 mdi
->array
.level
> 0 &&
268 is_subarray(mdi
->text_version
)) {
270 /* This is mdmon managed. */
273 /* As we have an O_EXCL open, any use of the device
274 * which blocks STOP_ARRAY is probably a transient use,
275 * so it is reasonable to retry for a while - 5 seconds.
279 (err
= sysfs_set_str(mdi
, NULL
,
288 pr_err("failed to stop array %s: %s\n",
289 devname
, strerror(errno
));
294 /* Give monitor a chance to act */
295 ping_monitor(mdi
->text_version
);
297 fd
= open_dev_excl(devnm
);
300 pr_err("failed to completely stop %s"
301 ": Device is busy\n",
307 mdi
->array
.major_version
== -1 &&
308 mdi
->array
.minor_version
== -2 &&
309 !is_subarray(mdi
->text_version
)) {
310 struct mdstat_ent
*mds
, *m
;
311 /* container, possibly mdmon-managed.
312 * Make sure mdmon isn't opening it, which
313 * would interfere with the 'stop'
315 ping_monitor(mdi
->sys_name
);
317 /* now check that there are no existing arrays
318 * which are members of this array
320 mds
= mdstat_read(0, 0);
321 for (m
= mds
; m
; m
= m
->next
)
322 if (m
->metadata_version
&&
323 strncmp(m
->metadata_version
, "external:", 9)==0 &&
324 metadata_container_matches(m
->metadata_version
+9,
327 pr_err("Cannot stop container %s: "
328 "member %s still active\n",
336 /* If the array is undergoing a reshape which changes the number
337 * of devices, then it would be nice to stop it at a point where
338 * it has completed a full number of stripes in both old and
339 * new layouts as this will allow the reshape to be reverted.
340 * So if 'sync_action' is "reshape" and 'raid_disks' shows two
341 * different numbers, then
343 * - set sync_max to next multiple of both data_disks and
344 * chunk sizes (or next but one)
346 * - wait on 'sync_completed' for that point to be reached.
348 if (mdi
&& (mdi
->array
.level
>= 4 && mdi
->array
.level
<= 6) &&
349 sysfs_attribute_available(mdi
, NULL
, "sync_action") &&
350 sysfs_attribute_available(mdi
, NULL
, "reshape_direction") &&
351 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, 20) > 0 &&
352 strcmp(buf
, "reshape\n") == 0 &&
353 sysfs_get_two(mdi
, NULL
, "raid_disks", &rd1
, &rd2
) == 2 &&
354 sysfs_set_str(mdi
, NULL
, "sync_action", "frozen") == 0) {
355 /* Array is frozen */
356 unsigned long long position
, curr
;
357 unsigned long long chunk1
, chunk2
;
358 unsigned long long rddiv
, chunkdiv
;
359 unsigned long long sectors
;
364 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
365 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
366 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
367 if (strncmp(buf
, "back", 4) == 0)
369 sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
);
370 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
373 rddiv
= GCD(rd1
, rd2
);
374 chunkdiv
= GCD(chunk1
, chunk2
);
375 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
378 /* Need to subtract 'reshape_position' from
379 * array size to get equivalent of sync_max.
380 * Size calculation based on raid5_size in kernel.
382 unsigned long long size
= mdi
->component_size
;
385 /* rd1 must be smaller */
387 position
= size
- position
;
388 position
= (position
/sectors
+ 2) * sectors
;
389 sysfs_set_num(mdi
, NULL
, "sync_max", position
/rd1
);
390 position
= size
- position
;
392 position
= (position
/sectors
+ 2) * sectors
;
393 sysfs_set_num(mdi
, NULL
, "sync_max", position
/rd1
);
395 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
397 /* That should have set things going again. Now we
398 * wait a little while (5 seconds) for sync_completed
399 * to reach the target.
402 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
403 while (scfd
>= 0 && delay
> 0) {
404 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
405 if (strncmp(buf
, "none", 4) == 0)
407 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
408 if (!backwards
&& curr
>= position
)
410 if (backwards
&& curr
<= position
)
412 sysfs_wait(scfd
, &delay
);
419 /* As we have an O_EXCL open, any use of the device
420 * which blocks STOP_ARRAY is probably a transient use,
421 * so it is reasonable to retry for a while - 5 seconds.
424 while (count
&& fd
>= 0
425 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
430 if (fd
>= 0 && err
) {
432 pr_err("failed to stop array %s: %s\n",
433 devname
, strerror(errno
));
435 cont_err("Perhaps a running "
436 "process, mounted filesystem "
437 "or active volume group?\n");
442 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
443 * was stopped, so We'll do it here just to be sure. Drop any
444 * partitions as well...
447 ioctl(fd
, BLKRRPART
, 0);
449 sysfs_uevent(mdi
, "change");
451 if (devnm
[0] && use_udev()) {
452 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
453 remove_devices(devnm
, mp
? mp
->path
: NULL
);
457 pr_err("stopped %s\n", devname
);
459 map_remove(&map
, devnm
);
468 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
470 struct mddev_dev
*new;
471 new = xmalloc(sizeof(*new));
472 memset(new, 0, sizeof(*new));
473 new->devname
= xstrdup(name
);
474 new->disposition
= disp
;
475 new->next
= dv
->next
;
480 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
482 mdu_array_info_t array
;
483 mdu_disk_info_t disk
;
487 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
490 remaining_disks
= array
.nr_disks
;
491 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
494 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
496 if (disk
.major
== 0 && disk
.minor
== 0)
499 if ((disk
.state
& 1) == 0) /* not faulty */
501 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
502 dv
= add_one(dv
, buf
, disp
);
506 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
508 mdu_array_info_t array
;
509 mdu_disk_info_t disk
;
513 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
516 remaining_disks
= array
.nr_disks
;
517 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
521 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
523 if (disk
.major
== 0 && disk
.minor
== 0)
526 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
528 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
529 sfd
= dev_open(buf
, O_RDONLY
);
536 /* Probably not detached */
538 dv
= add_one(dv
, buf
, disp
);
542 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
544 mdu_array_info_t array
;
545 mdu_disk_info_t disk
;
550 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
552 if (array
.level
!= 10)
554 copies
= ((array
.layout
& 0xff) *
555 ((array
.layout
>> 8) & 0xff));
556 if (array
.raid_disks
% copies
)
559 remaining_disks
= array
.nr_disks
;
560 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
563 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
565 if (disk
.major
== 0 && disk
.minor
== 0)
568 set
= disk
.raid_disk
% copies
;
569 if (set_char
!= set
+ 'A')
571 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
572 dv
= add_one(dv
, buf
, dv
->disposition
);
576 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
577 struct supertype
*dev_st
, struct supertype
*tst
,
579 char *update
, char *devname
, int verbose
,
580 mdu_array_info_t
*array
)
586 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
587 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
589 tst
->ss
->uuid_from_super(tst
, duuid
);
591 /* Assume uuid matches: kernel will check */
592 memcpy(duuid
, ouuid
, sizeof(ouuid
));
593 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
594 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
595 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
596 /* Looks like it is worth a
597 * try. Need to make sure
598 * kernel will accept it
601 mdu_disk_info_t disc
;
602 /* re-add doesn't work for version-1 superblocks
605 if (array
->major_version
== 1 &&
606 get_linux_version() <= 2006018)
608 disc
.number
= mdi
.disk
.number
;
609 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
610 || disc
.major
!= 0 || disc
.minor
!= 0
613 disc
.major
= major(rdev
);
614 disc
.minor
= minor(rdev
);
615 disc
.number
= mdi
.disk
.number
;
616 disc
.raid_disk
= mdi
.disk
.raid_disk
;
617 disc
.state
= mdi
.disk
.state
;
618 if (dv
->writemostly
== 1)
619 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
620 if (dv
->writemostly
== 2)
621 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
622 remove_partitions(tfd
);
623 if (update
|| dv
->writemostly
> 0) {
625 tfd
= dev_open(dv
->devname
, O_RDWR
);
627 pr_err("failed to open %s for"
628 " superblock update during re-add\n", dv
->devname
);
632 if (dv
->writemostly
== 1)
633 rv
= dev_st
->ss
->update_super(
634 dev_st
, NULL
, "writemostly",
635 devname
, verbose
, 0, NULL
);
636 if (dv
->writemostly
== 2)
637 rv
= dev_st
->ss
->update_super(
638 dev_st
, NULL
, "readwrite",
639 devname
, verbose
, 0, NULL
);
641 rv
= dev_st
->ss
->update_super(
642 dev_st
, NULL
, update
,
643 devname
, verbose
, 0, NULL
);
645 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
648 pr_err("failed to update"
649 " superblock during re-add\n");
653 /* don't even try if disk is marked as faulty */
655 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
657 pr_err("re-added %s\n", dv
->devname
);
660 if (errno
== ENOMEM
|| errno
== EROFS
) {
661 pr_err("add new device failed for %s: %s\n",
662 dv
->devname
, strerror(errno
));
663 if (dv
->disposition
== 'M')
672 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
673 struct supertype
*tst
, mdu_array_info_t
*array
,
674 int force
, int verbose
, char *devname
,
675 char *update
, unsigned long rdev
, unsigned long long array_size
)
677 unsigned long long ldsize
;
678 struct supertype
*dev_st
= NULL
;
680 mdu_disk_info_t disc
;
682 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
683 if (dv
->disposition
== 'M')
689 if (tst
->ss
->validate_geometry(
690 tst
, array
->level
, array
->layout
,
691 array
->raid_disks
, NULL
,
692 ldsize
>> 9, INVALID_SECTORS
, NULL
, NULL
, 0) == 0) {
694 pr_err("%s is larger than %s can "
696 " Add --force is you "
697 "really want to add this device.\n",
698 dv
->devname
, devname
);
701 pr_err("%s is larger than %s can "
703 " Adding anyway as --force "
705 dv
->devname
, devname
);
707 if (!tst
->ss
->external
&&
708 array
->major_version
== 0 &&
709 md_get_version(fd
)%100 < 2) {
710 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
712 pr_err("hot added %s\n",
717 pr_err("hot add failed for %s: %s\n",
718 dv
->devname
, strerror(errno
));
722 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
724 /* need to find a sample superblock to copy, and
725 * a spare slot to use.
726 * For 'external' array (well, container based),
727 * We can just load the metadata for the array->
731 /* already loaded */;
732 else if (tst
->ss
->external
) {
733 tst
->ss
->load_container(tst
, fd
, NULL
);
734 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
738 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
740 if (disc
.major
==0 && disc
.minor
==0)
742 if ((disc
.state
& 4)==0) /* sync */
744 /* Looks like a good device to try */
745 dev
= map_dev(disc
.major
, disc
.minor
, 1);
748 dfd
= dev_open(dev
, O_RDONLY
);
751 if (tst
->ss
->load_super(tst
, dfd
,
759 /* FIXME this is a bad test to be using */
760 if (!tst
->sb
&& dv
->disposition
!= 'a') {
761 /* we are re-adding a device to a
762 * completely dead array - have to depend
765 } else if (!tst
->sb
) {
766 pr_err("cannot load array metadata from %s\n", devname
);
770 /* Make sure device is large enough */
771 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
773 if (dv
->disposition
== 'M')
775 pr_err("%s not large enough to join array\n",
780 /* Possibly this device was recently part of
781 * the array and was temporarily removed, and
782 * is now being re-added. If so, we can
786 if (array
->not_persistent
==0) {
787 dev_st
= dup_super(tst
);
788 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
790 if (dev_st
&& dev_st
->sb
) {
791 int rv
= attempt_re_add(fd
, tfd
, dv
,
797 dev_st
->ss
->free_super(dev_st
);
801 if (dv
->disposition
== 'M') {
803 pr_err("--re-add for %s to %s is not possible\n",
804 dv
->devname
, devname
);
807 if (dv
->disposition
== 'A') {
808 pr_err("--re-add for %s to %s is not possible\n",
809 dv
->devname
, devname
);
812 if (array
->active_disks
< array
->raid_disks
) {
813 char *avail
= xcalloc(array
->raid_disks
, 1);
817 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
819 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
821 if (disc
.major
== 0 && disc
.minor
== 0)
823 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
825 avail
[disc
.raid_disk
] = 1;
828 array_failed
= !enough(array
->level
, array
->raid_disks
,
829 array
->layout
, 1, avail
);
833 pr_err("%s has failed so using --add cannot work and might destroy\n",
835 pr_err("data on %s. You should stop the array and re-assemble it.\n",
840 /* non-persistent. Must ensure that new drive
841 * is at least array->size big.
843 if (ldsize
/512 < array_size
) {
844 pr_err("%s not large enough to join array\n",
849 /* committed to really trying this device now*/
850 remove_partitions(tfd
);
852 /* in 2.6.17 and earlier, version-1 superblocks won't
853 * use the number we write, but will choose a free number.
854 * we must choose the same free number, which requires
855 * starting at 'raid_disks' and counting up
857 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
859 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
861 if (disc
.major
==0 && disc
.minor
==0)
863 if (disc
.state
& 8) /* removed */
866 disc
.major
= major(rdev
);
867 disc
.minor
= minor(rdev
);
870 if (array
->not_persistent
==0) {
872 if (dv
->writemostly
== 1)
873 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
874 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
875 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
876 dv
->devname
, INVALID_SECTORS
))
878 if (tst
->ss
->write_init_super(tst
))
880 } else if (dv
->disposition
== 'A') {
881 /* this had better be raid1.
882 * As we are "--re-add"ing we must find a spare slot
885 char *used
= xcalloc(array
->raid_disks
, 1);
886 for (j
= 0; j
< tst
->max_devs
; j
++) {
887 mdu_disk_info_t disc2
;
889 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
891 if (disc2
.major
==0 && disc2
.minor
==0)
893 if (disc2
.state
& 8) /* removed */
895 if (disc2
.raid_disk
< 0)
897 if (disc2
.raid_disk
> array
->raid_disks
)
899 used
[disc2
.raid_disk
] = 1;
901 for (j
= 0 ; j
< array
->raid_disks
; j
++)
904 disc
.state
|= (1<<MD_DISK_SYNC
);
909 if (dv
->writemostly
== 1)
910 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
911 if (tst
->ss
->external
) {
913 * to an external metadata container */
914 struct mdinfo new_mdi
;
920 strcpy(devnm
, fd2devnm(fd
));
922 container_fd
= open_dev_excl(devnm
);
923 if (container_fd
< 0) {
924 pr_err("add failed for %s:"
925 " could not get exclusive access to container\n",
927 tst
->ss
->free_super(tst
);
931 Kill(dv
->devname
, NULL
, 0, -1, 0);
932 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
933 if (mdmon_running(tst
->container_devnm
))
934 tst
->update_tail
= &tst
->updates
;
935 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
936 dv
->devname
, INVALID_SECTORS
)) {
941 if (tst
->update_tail
)
942 flush_metadata_updates(tst
);
944 tst
->ss
->sync_metadata(tst
);
946 sra
= sysfs_read(container_fd
, NULL
, 0);
948 pr_err("add failed for %s: sysfs_read failed\n",
951 tst
->ss
->free_super(tst
);
954 sra
->array
.level
= LEVEL_CONTAINER
;
955 /* Need to set data_offset and component_size */
956 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
957 new_mdi
.disk
.major
= disc
.major
;
958 new_mdi
.disk
.minor
= disc
.minor
;
959 new_mdi
.recovery_start
= 0;
960 /* Make sure fds are closed as they are O_EXCL which
961 * would block add_disk */
962 tst
->ss
->free_super(tst
);
963 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
964 pr_err("add new device to external metadata"
965 " failed for %s\n", dv
->devname
);
974 tst
->ss
->free_super(tst
);
975 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
976 pr_err("add new device failed for %s as %d: %s\n",
977 dv
->devname
, j
, strerror(errno
));
982 pr_err("added %s\n", dv
->devname
);
986 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
987 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
992 if (tst
->ss
->external
) {
993 /* To remove a device from a container, we must
994 * check that it isn't in use in an array.
995 * This involves looking in the 'holders'
996 * directory - there must be just one entry,
998 * To ensure that it doesn't get used as a
999 * hot spare while we are checking, we
1000 * get an O_EXCL open on the container
1004 strcpy(devnm
, fd2devnm(fd
));
1005 lfd
= open_dev_excl(devnm
);
1007 pr_err("Cannot get exclusive access "
1008 " to container - odd\n");
1011 /* We may not be able to check on holders in
1012 * sysfs, either because we don't have the dev num
1013 * (rdev == 0) or because the device has been detached
1014 * and the 'holders' directory no longer exists
1015 * (ret == -1). In that case, assume it is OK to
1021 ret
= sysfs_unique_holder(devnm
, rdev
);
1023 pr_err("%s is not a member, cannot remove.\n",
1029 pr_err("%s is still in use, cannot remove.\n",
1035 /* FIXME check that it is a current member */
1037 /* device has been removed and we don't know
1038 * the major:minor number
1040 int n
= write(sysfd
, "remove", 6);
1046 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1047 if (err
&& errno
== ENODEV
) {
1048 /* Old kernels rejected this if no personality
1050 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1051 struct mdinfo
*dv
= NULL
;
1054 for ( ; dv
; dv
=dv
->next
)
1055 if (dv
->disk
.major
== (int)major(rdev
) &&
1056 dv
->disk
.minor
== (int)minor(rdev
))
1059 err
= sysfs_set_str(sra
, dv
,
1068 pr_err("hot remove failed "
1069 "for %s: %s\n", dv
->devname
,
1075 if (tst
->ss
->external
) {
1077 * Before dropping our exclusive open we make an
1078 * attempt at preventing mdmon from seeing an
1079 * 'add' event before reconciling this 'remove'
1082 char *devnm
= fd2devnm(fd
);
1085 pr_err("unable to get container name\n");
1089 ping_manager(devnm
);
1094 pr_err("hot removed %s from %s\n",
1095 dv
->devname
, devname
);
1099 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1100 unsigned long rdev
, int verbose
, char *devname
)
1102 struct mdinfo
*mdi
, *di
;
1103 if (tst
->ss
->external
) {
1104 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1107 /* Need to find the device in sysfs and add 'want_replacement' to the
1110 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1111 if (!mdi
|| !mdi
->devs
) {
1112 pr_err("Cannot find status of %s to enable replacement - strange\n",
1116 for (di
= mdi
->devs
; di
; di
= di
->next
)
1117 if (di
->disk
.major
== (int)major(rdev
) &&
1118 di
->disk
.minor
== (int)minor(rdev
))
1122 if (di
->disk
.raid_disk
< 0) {
1123 pr_err("%s is not active and so cannot be replaced.\n",
1128 rv
= sysfs_set_str(mdi
, di
,
1129 "state", "want_replacement");
1132 pr_err("Failed to request replacement for %s\n",
1137 pr_err("Marked %s (device %d in %s) for replacement\n",
1138 dv
->devname
, di
->disk
.raid_disk
, devname
);
1139 /* If there is a matching 'with', we need to tell it which
1142 while (dv
&& dv
->disposition
!= 'W')
1145 dv
->disposition
= 'w';
1146 dv
->used
= di
->disk
.raid_disk
;
1151 pr_err("%s not found in %s so cannot --replace it\n",
1152 dv
->devname
, devname
);
1156 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1157 unsigned long rdev
, int verbose
, char *devname
)
1159 struct mdinfo
*mdi
, *di
;
1160 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1161 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1162 if (!mdi
|| !mdi
->devs
) {
1163 pr_err("Cannot find status of %s to enable replacement - strange\n",
1167 for (di
= mdi
->devs
; di
; di
= di
->next
)
1168 if (di
->disk
.major
== (int)major(rdev
) &&
1169 di
->disk
.minor
== (int)minor(rdev
))
1173 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1174 pr_err("%s is faulty and cannot be a replacement\n",
1179 if (di
->disk
.raid_disk
>= 0) {
1180 pr_err("%s is active and cannot be a replacement\n",
1185 rv
= sysfs_set_num(mdi
, di
,
1189 pr_err("Failed to %s as preferred replacement.\n",
1194 pr_err("Marked %s in %s as replacement for device %d\n",
1195 dv
->devname
, devname
, dv
->used
);
1199 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1200 dv
->devname
, devname
);
1204 int Manage_subdevs(char *devname
, int fd
,
1205 struct mddev_dev
*devlist
, int verbose
, int test
,
1206 char *update
, int force
)
1208 /* Do something to each dev.
1210 * 'a' - add the device
1212 * If that fails EINVAL, try ADD_NEW_DISK
1213 * 'A' - re-add the device
1214 * 'r' - remove the device: HOT_REMOVE_DISK
1215 * device can be 'faulty' or 'detached' in which case all
1216 * matching devices are removed.
1217 * 'f' - set the device faulty SET_DISK_FAULTY
1218 * device can be 'detached' in which case any device that
1219 * is inaccessible will be marked faulty.
1220 * 'R' - mark this device as wanting replacement.
1221 * 'W' - this device is added if necessary and activated as
1222 * a replacement for a previous 'R' device.
1224 * 'w' - 'W' will be changed to 'w' when it is paired with
1225 * a 'R' device. If a 'W' is found while walking the list
1226 * it must be unpaired, and is an error.
1227 * 'M' - this is created by a 'missing' target. It is a slight
1229 * 'F' - Another variant of 'A', where the device was faulty
1230 * so must be removed from the array first.
1232 * For 'f' and 'r', the device can also be a kernel-internal
1233 * name such as 'sdb'.
1235 mdu_array_info_t array
;
1236 unsigned long long array_size
;
1237 struct mddev_dev
*dv
;
1240 struct supertype
*tst
;
1241 char *subarray
= NULL
;
1243 int count
= 0; /* number of actions taken */
1248 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1249 pr_err("Cannot get array info for %s\n",
1253 sysfs_init(&info
, fd
, NULL
);
1255 /* array.size is only 32 bits and may be truncated.
1256 * So read from sysfs if possible, and record number of sectors
1259 array_size
= get_component_size(fd
);
1260 if (array_size
<= 0)
1261 array_size
= array
.size
* 2;
1263 tst
= super_by_fd(fd
, &subarray
);
1265 pr_err("unsupport array - version %d.%d\n",
1266 array
.major_version
, array
.minor_version
);
1271 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1274 if (strcmp(dv
->devname
, "failed") == 0 ||
1275 strcmp(dv
->devname
, "faulty") == 0) {
1276 if (dv
->disposition
!= 'A'
1277 && dv
->disposition
!= 'r') {
1278 pr_err("%s only meaningful "
1279 "with -r or --re-add, not -%c\n",
1280 dv
->devname
, dv
->disposition
);
1283 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1287 if (strcmp(dv
->devname
, "detached") == 0) {
1288 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1289 pr_err("%s only meaningful "
1290 "with -r of -f, not -%c\n",
1291 dv
->devname
, dv
->disposition
);
1294 add_detached(dv
, fd
, dv
->disposition
);
1298 if (strcmp(dv
->devname
, "missing") == 0) {
1299 struct mddev_dev
*add_devlist
= NULL
;
1300 struct mddev_dev
**dp
;
1301 if (dv
->disposition
!= 'A') {
1302 pr_err("'missing' only meaningful "
1306 add_devlist
= conf_get_devs();
1307 if (add_devlist
== NULL
) {
1308 pr_err("no devices to scan for missing members.");
1311 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1312 /* 'M' (for 'missing') is like 'A' without errors */
1313 (*dp
)->disposition
= 'M';
1315 dv
->next
= add_devlist
;
1319 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1320 strlen(dv
->devname
) == 5) {
1323 if (dv
->disposition
!= 'r' &&
1324 dv
->disposition
!= 'f') {
1325 pr_err("'%s' only meaningful with -r or -f\n",
1329 if (array
.level
!= 10) {
1330 pr_err("'%s' only meaningful with RAID10 arrays\n",
1334 copies
= ((array
.layout
& 0xff) *
1335 ((array
.layout
>> 8) & 0xff));
1336 if (array
.raid_disks
% copies
!= 0 ||
1337 dv
->devname
[4] < 'A' ||
1338 dv
->devname
[4] >= 'A' + copies
||
1340 pr_err("'%s' not meaningful with this array\n",
1344 add_set(dv
, fd
, dv
->devname
[4]);
1348 if (strchr(dv
->devname
, '/') == NULL
&&
1349 strchr(dv
->devname
, ':') == NULL
&&
1350 strlen(dv
->devname
) < 50) {
1351 /* Assume this is a kernel-internal name like 'sda1' */
1354 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1355 pr_err("%s only meaningful "
1356 "with -r or -f, not -%c\n",
1357 dv
->devname
, dv
->disposition
);
1361 sprintf(dname
, "dev-%s", dv
->devname
);
1362 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1366 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1367 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1368 stb
.st_rdev
= makedev(mj
,mn
);
1375 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1377 pr_err("%s does not appear "
1378 "to be a component of %s\n",
1379 dv
->devname
, devname
);
1384 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1388 int open_err
= errno
;
1389 if (stat(dv
->devname
, &stb
) != 0) {
1390 pr_err("Cannot find %s: %s\n",
1391 dv
->devname
, strerror(errno
));
1394 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1395 if (dv
->disposition
== 'M')
1396 /* non-fatal. Also improbable */
1398 pr_err("%s is not a block device.\n",
1402 if (dv
->disposition
== 'r')
1403 /* Be happy, the stat worked, that is
1404 * enough for --remove
1408 if (dv
->disposition
== 'M')
1411 pr_err("Cannot open %s: %s\n",
1412 dv
->devname
, strerror(open_err
));
1417 switch(dv
->disposition
){
1419 pr_err("internal error - devmode[%s]=%d\n",
1420 dv
->devname
, dv
->disposition
);
1424 case 'M': /* --re-add missing */
1425 case 'F': /* --re-add faulty */
1426 /* add the device */
1428 pr_err("Cannot add disks to a"
1429 " \'member\' array, perform this"
1430 " operation on the parent container\n");
1433 if (dv
->disposition
== 'F')
1434 /* Need to remove first */
1435 ioctl(fd
, HOT_REMOVE_DISK
,
1436 (unsigned long)stb
.st_rdev
);
1437 /* Make sure it isn't in use (in 2.6 or later) */
1438 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1440 /* We know no-one else is using it. We'll
1441 * need non-exclusive access to add it, so
1445 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1448 if (dv
->disposition
== 'M')
1450 pr_err("Cannot open %s: %s\n",
1451 dv
->devname
, strerror(errno
));
1455 if (sysfs_freeze_array(&info
) == 1)
1460 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1461 force
, verbose
, devname
, update
,
1462 stb
.st_rdev
, array_size
);
1474 pr_err("Cannot remove disks from a"
1475 " \'member\' array, perform this"
1476 " operation on the parent container\n");
1479 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1480 stb
.st_rdev
, verbose
,
1491 case 'f': /* set faulty */
1492 /* FIXME check current member */
1493 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1494 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1495 (unsigned long) stb
.st_rdev
))) {
1498 pr_err("set device faulty failed for %s: %s\n",
1499 dv
->devname
, strerror(errno
));
1509 pr_err("set %s faulty in %s\n",
1510 dv
->devname
, devname
);
1512 case 'R': /* Mark as replaceable */
1514 pr_err("Cannot replace disks in a"
1515 " \'member\' array, perform this"
1516 " operation on the parent container\n");
1520 if (sysfs_freeze_array(&info
) == 1)
1525 rv
= Manage_replace(tst
, fd
, dv
,
1526 stb
.st_rdev
, verbose
,
1534 case 'W': /* --with device that doesn't match */
1535 pr_err("No matching --replace device for --with %s\n",
1538 case 'w': /* --with device which was matched */
1539 rv
= Manage_with(tst
, fd
, dv
,
1540 stb
.st_rdev
, verbose
, devname
);
1547 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1548 if (test
&& count
== 0)
1554 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1555 return !test
&& busy
? 2 : 1;
1558 int autodetect(void)
1560 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1562 int fd
= dev_open("9:0", O_RDONLY
);
1564 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1571 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1573 struct supertype supertype
, *st
= &supertype
;
1576 memset(st
, 0, sizeof(*st
));
1578 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1582 if (!st
->ss
->update_subarray
) {
1584 pr_err("Operation not supported for %s metadata\n",
1589 if (mdmon_running(st
->devnm
))
1590 st
->update_tail
= &st
->updates
;
1592 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1596 pr_err("Failed to update %s of subarray-%s in %s\n",
1597 update
, subarray
, dev
);
1598 } else if (st
->update_tail
)
1599 flush_metadata_updates(st
);
1601 st
->ss
->sync_metadata(st
);
1603 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1604 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1608 st
->ss
->free_super(st
);
1614 /* Move spare from one array to another If adding to destination array fails
1615 * add back to original array.
1616 * Returns 1 on success, 0 on failure */
1617 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1619 struct mddev_dev devlist
;
1622 /* try to remove and add */
1623 int fd1
= open(to_devname
, O_RDONLY
);
1624 int fd2
= open(from_devname
, O_RDONLY
);
1626 if (fd1
< 0 || fd2
< 0) {
1627 if (fd1
>=0) close(fd1
);
1628 if (fd2
>=0) close(fd2
);
1632 devlist
.next
= NULL
;
1634 devlist
.writemostly
= 0;
1635 devlist
.devname
= devname
;
1636 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1638 devlist
.disposition
= 'r';
1639 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1640 devlist
.disposition
= 'a';
1641 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1642 /* make sure manager is aware of changes */
1643 ping_manager(to_devname
);
1644 ping_manager(from_devname
);
1649 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);