]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Manage.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2013 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
, struct context
*c
)
175 /* Run the array. Array must already be configured
180 if (md_get_version(fd
) < 9000) {
181 pr_err("need md driver version 0.90.0 or later\n");
186 pr_err("Cannot find %s in sysfs!!\n", devname
);
190 return IncrementalScan(c
, nm
);
193 int Manage_stop(char *devname
, int fd
, int verbose
, int will_retry
)
195 /* Stop the array. Array must already be configured
196 * 'will_retry' means that error messages are not wanted.
199 struct map_ent
*map
= NULL
;
206 unsigned long long rd1
, rd2
;
208 if (will_retry
&& verbose
== 0)
211 if (md_get_version(fd
) < 9000) {
212 if (ioctl(fd
, STOP_MD
, 0) == 0)
214 pr_err("stopping device %s "
216 devname
, strerror(errno
));
220 /* If this is an mdmon managed array, just write 'inactive'
221 * to the array state and let mdmon clear up.
223 strcpy(devnm
, fd2devnm(fd
));
224 /* Get EXCL access first. If this fails, then attempting
225 * to stop is probably a bad idea.
227 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_COMPONENT
|GET_VERSION
);
228 if (mdi
&& is_subarray(mdi
->text_version
)) {
230 strncpy(container
, mdi
->text_version
+1, sizeof(container
));
231 container
[sizeof(container
)-1] = 0;
232 sl
= strchr(container
, '/');
239 while (((fd
= ((devnm
[0] == '/')
240 ?open(devname
, O_RDONLY
|O_EXCL
)
241 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0
242 || strcmp(fd2devnm(fd
), devnm
) != 0)
244 && mdmon_running(container
)
248 flush_mdmon(container
);
251 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
255 pr_err("Cannot get exclusive access to %s:"
257 "process, mounted filesystem "
258 "or active volume group?\n",
263 mdi
->array
.level
> 0 &&
264 is_subarray(mdi
->text_version
)) {
266 /* This is mdmon managed. */
269 /* As we have an O_EXCL open, any use of the device
270 * which blocks STOP_ARRAY is probably a transient use,
271 * so it is reasonable to retry for a while - 5 seconds.
275 (err
= sysfs_set_str(mdi
, NULL
,
284 pr_err("failed to stop array %s: %s\n",
285 devname
, strerror(errno
));
290 /* Give monitor a chance to act */
291 ping_monitor(mdi
->text_version
);
293 fd
= open_dev_excl(devnm
);
296 pr_err("failed to completely stop %s"
297 ": Device is busy\n",
303 mdi
->array
.major_version
== -1 &&
304 mdi
->array
.minor_version
== -2 &&
305 !is_subarray(mdi
->text_version
)) {
306 struct mdstat_ent
*mds
, *m
;
307 /* container, possibly mdmon-managed.
308 * Make sure mdmon isn't opening it, which
309 * would interfere with the 'stop'
311 ping_monitor(mdi
->sys_name
);
313 /* now check that there are no existing arrays
314 * which are members of this array
316 mds
= mdstat_read(0, 0);
317 for (m
= mds
; m
; m
= m
->next
)
318 if (m
->metadata_version
&&
319 strncmp(m
->metadata_version
, "external:", 9)==0 &&
320 metadata_container_matches(m
->metadata_version
+9,
323 pr_err("Cannot stop container %s: "
324 "member %s still active\n",
332 /* If the array is undergoing a reshape which changes the number
333 * of devices, then it would be nice to stop it at a point where
334 * it has completed a full number of stripes in both old and
335 * new layouts as this will allow the reshape to be reverted.
336 * So if 'sync_action' is "reshape" and 'raid_disks' shows two
337 * different numbers, then
339 * - set sync_max to next multiple of both data_disks and
340 * chunk sizes (or next but one)
342 * - wait on 'sync_completed' for that point to be reached.
344 if (mdi
&& (mdi
->array
.level
>= 4 && mdi
->array
.level
<= 6) &&
345 sysfs_attribute_available(mdi
, NULL
, "sync_action") &&
346 sysfs_attribute_available(mdi
, NULL
, "reshape_direction") &&
347 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, 20) > 0 &&
348 strcmp(buf
, "reshape\n") == 0 &&
349 sysfs_get_two(mdi
, NULL
, "raid_disks", &rd1
, &rd2
) == 2 &&
350 sysfs_set_str(mdi
, NULL
, "sync_action", "frozen") == 0) {
351 /* Array is frozen */
352 unsigned long long position
, curr
;
353 unsigned long long chunk1
, chunk2
;
354 unsigned long long rddiv
, chunkdiv
;
355 unsigned long long sectors
;
356 unsigned long long sync_max
, old_sync_max
;
357 unsigned long long completed
;
362 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
363 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
364 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
365 if (strncmp(buf
, "back", 4) == 0)
367 sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
);
368 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
371 rddiv
= GCD(rd1
, rd2
);
372 chunkdiv
= GCD(chunk1
, chunk2
);
373 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
376 /* Need to subtract 'reshape_position' from
377 * array size to get equivalent of sync_max.
378 * Size calculation based on raid5_size in kernel.
380 unsigned long long size
= mdi
->component_size
;
383 /* rd1 must be smaller */
384 position
= (position
/ sectors
- 1) * sectors
;
385 sync_max
= size
- position
/rd1
;
387 position
= (position
/ sectors
+ 2) * sectors
;
388 sync_max
= position
/rd1
;
390 if (sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) < 0)
391 old_sync_max
= mdi
->component_size
;
392 /* Must not advance sync_max as that could confuse
393 * the reshape monitor */
394 if (sync_max
< old_sync_max
)
395 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
396 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
398 /* That should have set things going again. Now we
399 * wait a little while (3 second max) for sync_completed
400 * to reach the target.
401 * The reshape process can block for 500msec if
402 * the sync speed limit is hit, so we need to wait
403 * a lot longer than that. 1 second is usually
407 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
408 while (scfd
>= 0 && delay
> 0 && old_sync_max
> 0) {
409 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
410 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
411 if (strncmp(buf
, "none", 4) == 0) {
412 /* Either reshape has aborted, or hasn't
413 * quite started yet. Wait a bit and
414 * check 'sync_action' to see.
417 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
));
418 if (strncmp(buf
, "reshape", 7) != 0)
422 if (sysfs_fd_get_ll(scfd
, &completed
) == 0 &&
423 (completed
> sync_max
||
424 (completed
== sync_max
&& curr
!= position
))) {
425 while (completed
> sync_max
) {
426 sync_max
+= sectors
/ rd1
;
432 if (sync_max
< old_sync_max
)
433 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
436 if (!backwards
&& curr
>= position
)
438 if (backwards
&& curr
<= position
)
440 sysfs_wait(scfd
, &delay
);
447 /* As we have an O_EXCL open, any use of the device
448 * which blocks STOP_ARRAY is probably a transient use,
449 * so it is reasonable to retry for a while - 5 seconds.
452 while (count
&& fd
>= 0
453 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
458 if (fd
>= 0 && err
) {
460 pr_err("failed to stop array %s: %s\n",
461 devname
, strerror(errno
));
463 cont_err("Perhaps a running "
464 "process, mounted filesystem "
465 "or active volume group?\n");
470 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
471 * was stopped, so We'll do it here just to be sure. Drop any
472 * partitions as well...
475 ioctl(fd
, BLKRRPART
, 0);
477 sysfs_uevent(mdi
, "change");
479 if (devnm
[0] && use_udev()) {
480 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
481 remove_devices(devnm
, mp
? mp
->path
: NULL
);
485 pr_err("stopped %s\n", devname
);
487 map_remove(&map
, devnm
);
496 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
498 struct mddev_dev
*new;
499 new = xmalloc(sizeof(*new));
500 memset(new, 0, sizeof(*new));
501 new->devname
= xstrdup(name
);
502 new->disposition
= disp
;
503 new->next
= dv
->next
;
508 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
510 mdu_array_info_t array
;
511 mdu_disk_info_t disk
;
515 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
518 remaining_disks
= array
.nr_disks
;
519 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
522 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
524 if (disk
.major
== 0 && disk
.minor
== 0)
527 if ((disk
.state
& 1) == 0) /* not faulty */
529 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
530 dv
= add_one(dv
, buf
, disp
);
534 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
536 mdu_array_info_t array
;
537 mdu_disk_info_t disk
;
541 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
544 remaining_disks
= array
.nr_disks
;
545 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
549 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
551 if (disk
.major
== 0 && disk
.minor
== 0)
554 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
556 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
557 sfd
= dev_open(buf
, O_RDONLY
);
564 /* Probably not detached */
566 dv
= add_one(dv
, buf
, disp
);
570 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
572 mdu_array_info_t array
;
573 mdu_disk_info_t disk
;
578 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
580 if (array
.level
!= 10)
582 copies
= ((array
.layout
& 0xff) *
583 ((array
.layout
>> 8) & 0xff));
584 if (array
.raid_disks
% copies
)
587 remaining_disks
= array
.nr_disks
;
588 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
591 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
593 if (disk
.major
== 0 && disk
.minor
== 0)
596 set
= disk
.raid_disk
% copies
;
597 if (set_char
!= set
+ 'A')
599 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
600 dv
= add_one(dv
, buf
, dv
->disposition
);
604 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
605 struct supertype
*dev_st
, struct supertype
*tst
,
607 char *update
, char *devname
, int verbose
,
608 mdu_array_info_t
*array
)
614 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
615 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
617 tst
->ss
->uuid_from_super(tst
, duuid
);
619 /* Assume uuid matches: kernel will check */
620 memcpy(duuid
, ouuid
, sizeof(ouuid
));
621 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
622 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
623 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
624 /* Looks like it is worth a
625 * try. Need to make sure
626 * kernel will accept it
629 mdu_disk_info_t disc
;
630 /* re-add doesn't work for version-1 superblocks
633 if (array
->major_version
== 1 &&
634 get_linux_version() <= 2006018)
636 disc
.number
= mdi
.disk
.number
;
637 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
638 || disc
.major
!= 0 || disc
.minor
!= 0
641 disc
.major
= major(rdev
);
642 disc
.minor
= minor(rdev
);
643 disc
.number
= mdi
.disk
.number
;
644 disc
.raid_disk
= mdi
.disk
.raid_disk
;
645 disc
.state
= mdi
.disk
.state
;
646 if (dv
->writemostly
== 1)
647 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
648 if (dv
->writemostly
== 2)
649 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
650 remove_partitions(tfd
);
651 if (update
|| dv
->writemostly
> 0) {
653 tfd
= dev_open(dv
->devname
, O_RDWR
);
655 pr_err("failed to open %s for"
656 " superblock update during re-add\n", dv
->devname
);
660 if (dv
->writemostly
== 1)
661 rv
= dev_st
->ss
->update_super(
662 dev_st
, NULL
, "writemostly",
663 devname
, verbose
, 0, NULL
);
664 if (dv
->writemostly
== 2)
665 rv
= dev_st
->ss
->update_super(
666 dev_st
, NULL
, "readwrite",
667 devname
, verbose
, 0, NULL
);
669 rv
= dev_st
->ss
->update_super(
670 dev_st
, NULL
, update
,
671 devname
, verbose
, 0, NULL
);
673 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
676 pr_err("failed to update"
677 " superblock during re-add\n");
681 /* don't even try if disk is marked as faulty */
683 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
685 pr_err("re-added %s\n", dv
->devname
);
688 if (errno
== ENOMEM
|| errno
== EROFS
) {
689 pr_err("add new device failed for %s: %s\n",
690 dv
->devname
, strerror(errno
));
691 if (dv
->disposition
== 'M')
700 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
701 struct supertype
*tst
, mdu_array_info_t
*array
,
702 int force
, int verbose
, char *devname
,
703 char *update
, unsigned long rdev
, unsigned long long array_size
)
705 unsigned long long ldsize
;
706 struct supertype
*dev_st
= NULL
;
708 mdu_disk_info_t disc
;
710 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
711 if (dv
->disposition
== 'M')
717 if (tst
->ss
== &super0
&& ldsize
> 4ULL*1024*1024*1024*1024) {
718 /* More than 4TB is wasted on v0.90 */
720 pr_err("%s is larger than %s can "
722 " Add --force is you "
723 "really want to add this device.\n",
724 dv
->devname
, devname
);
727 pr_err("%s is larger than %s can "
729 " Adding anyway as --force "
731 dv
->devname
, devname
);
733 if (!tst
->ss
->external
&&
734 array
->major_version
== 0 &&
735 md_get_version(fd
)%100 < 2) {
736 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
738 pr_err("hot added %s\n",
743 pr_err("hot add failed for %s: %s\n",
744 dv
->devname
, strerror(errno
));
748 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
750 /* need to find a sample superblock to copy, and
751 * a spare slot to use.
752 * For 'external' array (well, container based),
753 * We can just load the metadata for the array->
757 /* already loaded */;
758 else if (tst
->ss
->external
) {
759 tst
->ss
->load_container(tst
, fd
, NULL
);
760 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
764 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
766 if (disc
.major
==0 && disc
.minor
==0)
768 if ((disc
.state
& 4)==0) /* sync */
770 /* Looks like a good device to try */
771 dev
= map_dev(disc
.major
, disc
.minor
, 1);
774 dfd
= dev_open(dev
, O_RDONLY
);
777 if (tst
->ss
->load_super(tst
, dfd
,
785 /* FIXME this is a bad test to be using */
786 if (!tst
->sb
&& (dv
->disposition
!= 'a'
787 && dv
->disposition
!= 'S')) {
788 /* we are re-adding a device to a
789 * completely dead array - have to depend
792 } else if (!tst
->sb
) {
793 pr_err("cannot load array metadata from %s\n", devname
);
797 /* Make sure device is large enough */
798 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
800 if (dv
->disposition
== 'M')
802 pr_err("%s not large enough to join array\n",
807 /* Possibly this device was recently part of
808 * the array and was temporarily removed, and
809 * is now being re-added. If so, we can
813 if (array
->not_persistent
==0) {
814 dev_st
= dup_super(tst
);
815 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
817 if (dev_st
&& dev_st
->sb
&& dv
->disposition
!= 'S') {
818 int rv
= attempt_re_add(fd
, tfd
, dv
,
824 dev_st
->ss
->free_super(dev_st
);
828 if (dv
->disposition
== 'M') {
830 pr_err("--re-add for %s to %s is not possible\n",
831 dv
->devname
, devname
);
834 if (dv
->disposition
== 'A') {
835 pr_err("--re-add for %s to %s is not possible\n",
836 dv
->devname
, devname
);
839 if (array
->active_disks
< array
->raid_disks
) {
840 char *avail
= xcalloc(array
->raid_disks
, 1);
844 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
846 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
848 if (disc
.major
== 0 && disc
.minor
== 0)
851 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
853 avail
[disc
.raid_disk
] = 1;
855 array_failed
= !enough(array
->level
, array
->raid_disks
,
856 array
->layout
, 1, avail
);
861 pr_err("%s has failed so using --add cannot work and might destroy\n",
863 pr_err("data on %s. You should stop the array and re-assemble it.\n",
868 /* non-persistent. Must ensure that new drive
869 * is at least array->size big.
871 if (ldsize
/512 < array_size
) {
872 pr_err("%s not large enough to join array\n",
877 /* committed to really trying this device now*/
878 remove_partitions(tfd
);
880 /* in 2.6.17 and earlier, version-1 superblocks won't
881 * use the number we write, but will choose a free number.
882 * we must choose the same free number, which requires
883 * starting at 'raid_disks' and counting up
885 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
887 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
889 if (disc
.major
==0 && disc
.minor
==0)
891 if (disc
.state
& 8) /* removed */
894 disc
.major
= major(rdev
);
895 disc
.minor
= minor(rdev
);
898 if (array
->not_persistent
==0) {
900 if (dv
->writemostly
== 1)
901 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
902 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
903 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
904 dv
->devname
, INVALID_SECTORS
))
906 if (tst
->ss
->write_init_super(tst
))
908 } else if (dv
->disposition
== 'A') {
909 /* this had better be raid1.
910 * As we are "--re-add"ing we must find a spare slot
913 char *used
= xcalloc(array
->raid_disks
, 1);
914 for (j
= 0; j
< tst
->max_devs
; j
++) {
915 mdu_disk_info_t disc2
;
917 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
919 if (disc2
.major
==0 && disc2
.minor
==0)
921 if (disc2
.state
& 8) /* removed */
923 if (disc2
.raid_disk
< 0)
925 if (disc2
.raid_disk
> array
->raid_disks
)
927 used
[disc2
.raid_disk
] = 1;
929 for (j
= 0 ; j
< array
->raid_disks
; j
++)
932 disc
.state
|= (1<<MD_DISK_SYNC
);
937 if (dv
->writemostly
== 1)
938 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
939 if (tst
->ss
->external
) {
941 * to an external metadata container */
942 struct mdinfo new_mdi
;
948 strcpy(devnm
, fd2devnm(fd
));
950 container_fd
= open_dev_excl(devnm
);
951 if (container_fd
< 0) {
952 pr_err("add failed for %s:"
953 " could not get exclusive access to container\n",
955 tst
->ss
->free_super(tst
);
959 Kill(dv
->devname
, NULL
, 0, -1, 0);
960 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
961 if (mdmon_running(tst
->container_devnm
))
962 tst
->update_tail
= &tst
->updates
;
963 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
964 dv
->devname
, INVALID_SECTORS
)) {
969 if (tst
->update_tail
)
970 flush_metadata_updates(tst
);
972 tst
->ss
->sync_metadata(tst
);
974 sra
= sysfs_read(container_fd
, NULL
, 0);
976 pr_err("add failed for %s: sysfs_read failed\n",
979 tst
->ss
->free_super(tst
);
982 sra
->array
.level
= LEVEL_CONTAINER
;
983 /* Need to set data_offset and component_size */
984 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
985 new_mdi
.disk
.major
= disc
.major
;
986 new_mdi
.disk
.minor
= disc
.minor
;
987 new_mdi
.recovery_start
= 0;
988 /* Make sure fds are closed as they are O_EXCL which
989 * would block add_disk */
990 tst
->ss
->free_super(tst
);
991 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
992 pr_err("add new device to external metadata"
993 " failed for %s\n", dv
->devname
);
1000 close(container_fd
);
1002 tst
->ss
->free_super(tst
);
1003 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1004 pr_err("add new device failed for %s as %d: %s\n",
1005 dv
->devname
, j
, strerror(errno
));
1010 pr_err("added %s\n", dv
->devname
);
1014 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1015 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
1020 if (tst
->ss
->external
) {
1021 /* To remove a device from a container, we must
1022 * check that it isn't in use in an array.
1023 * This involves looking in the 'holders'
1024 * directory - there must be just one entry,
1026 * To ensure that it doesn't get used as a
1027 * hot spare while we are checking, we
1028 * get an O_EXCL open on the container
1032 strcpy(devnm
, fd2devnm(fd
));
1033 lfd
= open_dev_excl(devnm
);
1035 pr_err("Cannot get exclusive access "
1036 " to container - odd\n");
1039 /* We may not be able to check on holders in
1040 * sysfs, either because we don't have the dev num
1041 * (rdev == 0) or because the device has been detached
1042 * and the 'holders' directory no longer exists
1043 * (ret == -1). In that case, assume it is OK to
1049 ret
= sysfs_unique_holder(devnm
, rdev
);
1051 pr_err("%s is not a member, cannot remove.\n",
1057 pr_err("%s is still in use, cannot remove.\n",
1063 /* FIXME check that it is a current member */
1065 /* device has been removed and we don't know
1066 * the major:minor number
1068 int n
= write(sysfd
, "remove", 6);
1074 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1075 if (err
&& errno
== ENODEV
) {
1076 /* Old kernels rejected this if no personality
1078 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1079 struct mdinfo
*dv
= NULL
;
1082 for ( ; dv
; dv
=dv
->next
)
1083 if (dv
->disk
.major
== (int)major(rdev
) &&
1084 dv
->disk
.minor
== (int)minor(rdev
))
1087 err
= sysfs_set_str(sra
, dv
,
1096 pr_err("hot remove failed "
1097 "for %s: %s\n", dv
->devname
,
1103 if (tst
->ss
->external
) {
1105 * Before dropping our exclusive open we make an
1106 * attempt at preventing mdmon from seeing an
1107 * 'add' event before reconciling this 'remove'
1110 char *devnm
= fd2devnm(fd
);
1113 pr_err("unable to get container name\n");
1117 ping_manager(devnm
);
1122 pr_err("hot removed %s from %s\n",
1123 dv
->devname
, devname
);
1127 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1128 unsigned long rdev
, int verbose
, char *devname
)
1130 struct mdinfo
*mdi
, *di
;
1131 if (tst
->ss
->external
) {
1132 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1135 /* Need to find the device in sysfs and add 'want_replacement' to the
1138 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1139 if (!mdi
|| !mdi
->devs
) {
1140 pr_err("Cannot find status of %s to enable replacement - strange\n",
1144 for (di
= mdi
->devs
; di
; di
= di
->next
)
1145 if (di
->disk
.major
== (int)major(rdev
) &&
1146 di
->disk
.minor
== (int)minor(rdev
))
1150 if (di
->disk
.raid_disk
< 0) {
1151 pr_err("%s is not active and so cannot be replaced.\n",
1156 rv
= sysfs_set_str(mdi
, di
,
1157 "state", "want_replacement");
1160 pr_err("Failed to request replacement for %s\n",
1165 pr_err("Marked %s (device %d in %s) for replacement\n",
1166 dv
->devname
, di
->disk
.raid_disk
, devname
);
1167 /* If there is a matching 'with', we need to tell it which
1170 while (dv
&& dv
->disposition
!= 'W')
1173 dv
->disposition
= 'w';
1174 dv
->used
= di
->disk
.raid_disk
;
1179 pr_err("%s not found in %s so cannot --replace it\n",
1180 dv
->devname
, devname
);
1184 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1185 unsigned long rdev
, int verbose
, char *devname
)
1187 struct mdinfo
*mdi
, *di
;
1188 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1189 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1190 if (!mdi
|| !mdi
->devs
) {
1191 pr_err("Cannot find status of %s to enable replacement - strange\n",
1195 for (di
= mdi
->devs
; di
; di
= di
->next
)
1196 if (di
->disk
.major
== (int)major(rdev
) &&
1197 di
->disk
.minor
== (int)minor(rdev
))
1201 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1202 pr_err("%s is faulty and cannot be a replacement\n",
1207 if (di
->disk
.raid_disk
>= 0) {
1208 pr_err("%s is active and cannot be a replacement\n",
1213 rv
= sysfs_set_num(mdi
, di
,
1217 pr_err("Failed to set %s as preferred replacement.\n",
1222 pr_err("Marked %s in %s as replacement for device %d\n",
1223 dv
->devname
, devname
, dv
->used
);
1227 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1228 dv
->devname
, devname
);
1232 int Manage_subdevs(char *devname
, int fd
,
1233 struct mddev_dev
*devlist
, int verbose
, int test
,
1234 char *update
, int force
)
1236 /* Do something to each dev.
1238 * 'a' - add the device
1240 * If that fails EINVAL, try ADD_NEW_DISK
1241 * 'S' - add the device as a spare - don't try re-add
1242 * 'A' - re-add the device
1243 * 'r' - remove the device: HOT_REMOVE_DISK
1244 * device can be 'faulty' or 'detached' in which case all
1245 * matching devices are removed.
1246 * 'f' - set the device faulty SET_DISK_FAULTY
1247 * device can be 'detached' in which case any device that
1248 * is inaccessible will be marked faulty.
1249 * 'R' - mark this device as wanting replacement.
1250 * 'W' - this device is added if necessary and activated as
1251 * a replacement for a previous 'R' device.
1253 * 'w' - 'W' will be changed to 'w' when it is paired with
1254 * a 'R' device. If a 'W' is found while walking the list
1255 * it must be unpaired, and is an error.
1256 * 'M' - this is created by a 'missing' target. It is a slight
1258 * 'F' - Another variant of 'A', where the device was faulty
1259 * so must be removed from the array first.
1261 * For 'f' and 'r', the device can also be a kernel-internal
1262 * name such as 'sdb'.
1264 mdu_array_info_t array
;
1265 unsigned long long array_size
;
1266 struct mddev_dev
*dv
;
1268 struct supertype
*tst
;
1269 char *subarray
= NULL
;
1271 int count
= 0; /* number of actions taken */
1276 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1277 pr_err("Cannot get array info for %s\n",
1281 sysfs_init(&info
, fd
, NULL
);
1283 /* array.size is only 32 bits and may be truncated.
1284 * So read from sysfs if possible, and record number of sectors
1287 array_size
= get_component_size(fd
);
1288 if (array_size
<= 0)
1289 array_size
= array
.size
* 2;
1291 tst
= super_by_fd(fd
, &subarray
);
1293 pr_err("unsupport array - version %d.%d\n",
1294 array
.major_version
, array
.minor_version
);
1298 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1299 unsigned long rdev
= 0; /* device to add/remove etc */
1303 if (strcmp(dv
->devname
, "failed") == 0 ||
1304 strcmp(dv
->devname
, "faulty") == 0) {
1305 if (dv
->disposition
!= 'A'
1306 && dv
->disposition
!= 'r') {
1307 pr_err("%s only meaningful "
1308 "with -r or --re-add, not -%c\n",
1309 dv
->devname
, dv
->disposition
);
1312 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1316 if (strcmp(dv
->devname
, "detached") == 0) {
1317 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1318 pr_err("%s only meaningful "
1319 "with -r of -f, not -%c\n",
1320 dv
->devname
, dv
->disposition
);
1323 add_detached(dv
, fd
, dv
->disposition
);
1327 if (strcmp(dv
->devname
, "missing") == 0) {
1328 struct mddev_dev
*add_devlist
= NULL
;
1329 struct mddev_dev
**dp
;
1330 if (dv
->disposition
!= 'A') {
1331 pr_err("'missing' only meaningful "
1335 add_devlist
= conf_get_devs();
1336 if (add_devlist
== NULL
) {
1337 pr_err("no devices to scan for missing members.");
1340 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1341 /* 'M' (for 'missing') is like 'A' without errors */
1342 (*dp
)->disposition
= 'M';
1344 dv
->next
= add_devlist
;
1348 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1349 strlen(dv
->devname
) == 5) {
1352 if (dv
->disposition
!= 'r' &&
1353 dv
->disposition
!= 'f') {
1354 pr_err("'%s' only meaningful with -r or -f\n",
1358 if (array
.level
!= 10) {
1359 pr_err("'%s' only meaningful with RAID10 arrays\n",
1363 copies
= ((array
.layout
& 0xff) *
1364 ((array
.layout
>> 8) & 0xff));
1365 if (array
.raid_disks
% copies
!= 0 ||
1366 dv
->devname
[4] < 'A' ||
1367 dv
->devname
[4] >= 'A' + copies
||
1369 pr_err("'%s' not meaningful with this array\n",
1373 add_set(dv
, fd
, dv
->devname
[4]);
1377 if (strchr(dv
->devname
, '/') == NULL
&&
1378 strchr(dv
->devname
, ':') == NULL
&&
1379 strlen(dv
->devname
) < 50) {
1380 /* Assume this is a kernel-internal name like 'sda1' */
1383 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1384 pr_err("%s only meaningful "
1385 "with -r or -f, not -%c\n",
1386 dv
->devname
, dv
->disposition
);
1390 sprintf(dname
, "dev-%s", dv
->devname
);
1391 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1394 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1395 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1396 rdev
= makedev(mj
,mn
);
1403 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1405 pr_err("%s does not appear "
1406 "to be a component of %s\n",
1407 dv
->devname
, devname
);
1411 } else if ((dv
->disposition
== 'r' || dv
->disposition
== 'f')
1412 && get_maj_min(dv
->devname
, &mj
, &mn
)) {
1413 /* for 'fail' and 'remove', the device might
1416 rdev
= makedev(mj
, mn
);
1419 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1423 int open_err
= errno
;
1424 if (stat(dv
->devname
, &stb
) != 0) {
1425 pr_err("Cannot find %s: %s\n",
1426 dv
->devname
, strerror(errno
));
1429 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1430 if (dv
->disposition
== 'M')
1431 /* non-fatal. Also improbable */
1433 pr_err("%s is not a block device.\n",
1437 if (dv
->disposition
== 'r')
1438 /* Be happy, the stat worked, that is
1439 * enough for --remove
1443 if (dv
->disposition
== 'M')
1446 pr_err("Cannot open %s: %s\n",
1447 dv
->devname
, strerror(open_err
));
1453 switch(dv
->disposition
){
1455 pr_err("internal error - devmode[%s]=%d\n",
1456 dv
->devname
, dv
->disposition
);
1459 case 'S': /* --add-spare */
1461 case 'M': /* --re-add missing */
1462 case 'F': /* --re-add faulty */
1463 /* add the device */
1465 pr_err("Cannot add disks to a"
1466 " \'member\' array, perform this"
1467 " operation on the parent container\n");
1470 if (dv
->disposition
== 'F')
1471 /* Need to remove first */
1472 ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1473 /* Make sure it isn't in use (in 2.6 or later) */
1474 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1476 /* We know no-one else is using it. We'll
1477 * need non-exclusive access to add it, so
1481 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1484 if (dv
->disposition
== 'M')
1486 pr_err("Cannot open %s: %s\n",
1487 dv
->devname
, strerror(errno
));
1491 if (sysfs_freeze_array(&info
) == 1)
1496 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1497 force
, verbose
, devname
, update
,
1510 pr_err("Cannot remove disks from a"
1511 " \'member\' array, perform this"
1512 " operation on the parent container\n");
1515 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1527 case 'f': /* set faulty */
1528 /* FIXME check current member */
1529 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1530 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1534 pr_err("set device faulty failed for %s: %s\n",
1535 dv
->devname
, strerror(errno
));
1545 pr_err("set %s faulty in %s\n",
1546 dv
->devname
, devname
);
1548 case 'R': /* Mark as replaceable */
1550 pr_err("Cannot replace disks in a"
1551 " \'member\' array, perform this"
1552 " operation on the parent container\n");
1556 if (sysfs_freeze_array(&info
) == 1)
1561 rv
= Manage_replace(tst
, fd
, dv
,
1570 case 'W': /* --with device that doesn't match */
1571 pr_err("No matching --replace device for --with %s\n",
1574 case 'w': /* --with device which was matched */
1575 rv
= Manage_with(tst
, fd
, dv
,
1576 rdev
, verbose
, devname
);
1583 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1584 if (test
&& count
== 0)
1590 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1591 return !test
&& busy
? 2 : 1;
1594 int autodetect(void)
1596 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1598 int fd
= dev_open("9:0", O_RDONLY
);
1600 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1607 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1609 struct supertype supertype
, *st
= &supertype
;
1612 memset(st
, 0, sizeof(*st
));
1614 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1618 if (!st
->ss
->update_subarray
) {
1620 pr_err("Operation not supported for %s metadata\n",
1625 if (mdmon_running(st
->devnm
))
1626 st
->update_tail
= &st
->updates
;
1628 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1632 pr_err("Failed to update %s of subarray-%s in %s\n",
1633 update
, subarray
, dev
);
1634 } else if (st
->update_tail
)
1635 flush_metadata_updates(st
);
1637 st
->ss
->sync_metadata(st
);
1639 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1640 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1644 st
->ss
->free_super(st
);
1650 /* Move spare from one array to another If adding to destination array fails
1651 * add back to original array.
1652 * Returns 1 on success, 0 on failure */
1653 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1655 struct mddev_dev devlist
;
1658 /* try to remove and add */
1659 int fd1
= open(to_devname
, O_RDONLY
);
1660 int fd2
= open(from_devname
, O_RDONLY
);
1662 if (fd1
< 0 || fd2
< 0) {
1663 if (fd1
>=0) close(fd1
);
1664 if (fd2
>=0) close(fd2
);
1668 devlist
.next
= NULL
;
1670 devlist
.writemostly
= 0;
1671 devlist
.devname
= devname
;
1672 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1674 devlist
.disposition
= 'r';
1675 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1676 devlist
.disposition
= 'a';
1677 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1678 /* make sure manager is aware of changes */
1679 ping_manager(to_devname
);
1680 ping_manager(from_devname
);
1685 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);