]>
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 failed: %s\n",
215 devname
, strerror(errno
));
219 strcpy(devnm
, fd2devnm(fd
));
220 /* Get EXCL access first. If this fails, then attempting
221 * to stop is probably a bad idea.
223 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_COMPONENT
|GET_VERSION
);
224 if (mdi
&& is_subarray(mdi
->text_version
)) {
226 strncpy(container
, mdi
->text_version
+1, sizeof(container
));
227 container
[sizeof(container
)-1] = 0;
228 sl
= strchr(container
, '/');
235 while (((fd
= ((devname
[0] == '/')
236 ?open(devname
, O_RDONLY
|O_EXCL
)
237 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0
238 || strcmp(fd2devnm(fd
), devnm
) != 0)
240 && mdmon_running(container
)
242 /* Can't open, so something might be wrong. However it
243 * is a container, so we might be racing with mdmon, so
248 flush_mdmon(container
);
251 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
255 pr_err("Cannot get exclusive access to %s:Perhaps a running process, mounted filesystem or active volume group?\n",
259 /* If this is an mdmon managed array, just write 'inactive'
260 * to the array state and let mdmon clear up.
263 mdi
->array
.level
> 0 &&
264 is_subarray(mdi
->text_version
)) {
266 /* This is mdmon managed. */
269 /* As we had 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: Device is busy\n",
302 mdi
->array
.major_version
== -1 &&
303 mdi
->array
.minor_version
== -2 &&
304 !is_subarray(mdi
->text_version
)) {
305 struct mdstat_ent
*mds
, *m
;
306 /* container, possibly mdmon-managed.
307 * Make sure mdmon isn't opening it, which
308 * would interfere with the 'stop'
310 ping_monitor(mdi
->sys_name
);
312 /* now check that there are no existing arrays
313 * which are members of this array
315 mds
= mdstat_read(0, 0);
316 for (m
= mds
; m
; m
= m
->next
)
317 if (m
->metadata_version
&&
318 strncmp(m
->metadata_version
, "external:", 9)==0 &&
319 metadata_container_matches(m
->metadata_version
+9,
322 pr_err("Cannot stop container %s: member %s still active\n",
330 /* If the array is undergoing a reshape which changes the number
331 * of devices, then it would be nice to stop it at a point where
332 * it has completed a full number of stripes in both old and
333 * new layouts as this will allow the reshape to be reverted.
334 * So if 'sync_action' is "reshape" and 'raid_disks' shows two
335 * different numbers, then
337 * - set sync_max to next multiple of both data_disks and
338 * chunk sizes (or next but one)
340 * - wait on 'sync_completed' for that point to be reached.
342 if (mdi
&& (mdi
->array
.level
>= 4 && mdi
->array
.level
<= 6) &&
343 sysfs_attribute_available(mdi
, NULL
, "sync_action") &&
344 sysfs_attribute_available(mdi
, NULL
, "reshape_direction") &&
345 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, 20) > 0 &&
346 strcmp(buf
, "reshape\n") == 0 &&
347 sysfs_get_two(mdi
, NULL
, "raid_disks", &rd1
, &rd2
) == 2) {
348 unsigned long long position
, curr
;
349 unsigned long long chunk1
, chunk2
;
350 unsigned long long rddiv
, chunkdiv
;
351 unsigned long long sectors
;
352 unsigned long long sync_max
, old_sync_max
;
353 unsigned long long completed
;
359 while (rd1
> rd2
&& delay
> 0 &&
360 sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) == 0) {
361 /* must be in the critical section - wait a bit */
366 if (sysfs_set_str(mdi
, NULL
, "sync_action", "frozen") != 0)
368 /* Array is frozen */
370 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
371 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
372 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
373 if (strncmp(buf
, "back", 4) == 0)
375 if (sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
) != 0) {
376 /* reshape must have finished now */
377 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
380 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
383 rddiv
= GCD(rd1
, rd2
);
384 chunkdiv
= GCD(chunk1
, chunk2
);
385 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
388 /* Need to subtract 'reshape_position' from
389 * array size to get equivalent of sync_max.
390 * Size calculation based on raid5_size in kernel.
392 unsigned long long size
= mdi
->component_size
;
395 /* rd1 must be smaller */
396 /* Reshape may have progressed further backwards than
397 * recorded, so target even further back (hence "-1")
399 position
= (position
/ sectors
- 1) * sectors
;
400 /* rd1 is always the conversion factor between 'sync'
401 * position and 'reshape' position.
402 * We read 1 "new" stripe worth of data from where-ever,
403 * and when write out that full stripe.
405 sync_max
= size
- position
/rd1
;
407 /* Reshape will very likely be beyond position, and it may
408 * be too late to stop at '+1', so aim for '+2'
410 position
= (position
/ sectors
+ 2) * sectors
;
411 sync_max
= position
/rd1
;
413 if (sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) < 0)
414 old_sync_max
= mdi
->component_size
;
415 /* Must not advance sync_max as that could confuse
416 * the reshape monitor */
417 if (sync_max
< old_sync_max
)
418 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
419 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
421 /* That should have set things going again. Now we
422 * wait a little while (3 second max) for sync_completed
423 * to reach the target.
424 * The reshape process can block for 500msec if
425 * the sync speed limit is hit, so we need to wait
426 * a lot longer than that. 1 second is usually
430 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
431 while (scfd
>= 0 && delay
> 0 && old_sync_max
> 0) {
432 unsigned long long max_completed
;
433 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
434 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
435 if (strncmp(buf
, "none", 4) == 0) {
436 /* Either reshape has aborted, or hasn't
437 * quite started yet. Wait a bit and
438 * check 'sync_action' to see.
441 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
));
442 if (strncmp(buf
, "reshape", 7) != 0)
446 if (sysfs_fd_get_two(scfd
, &completed
,
447 &max_completed
) == 2 &&
448 /* 'completed' sometimes reads as max-uulong */
449 completed
< max_completed
&&
450 (completed
> sync_max
||
451 (completed
== sync_max
&& curr
!= position
))) {
452 while (completed
> sync_max
) {
453 sync_max
+= sectors
/ rd1
;
459 if (sync_max
< old_sync_max
)
460 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
463 if (!backwards
&& curr
>= position
)
465 if (backwards
&& curr
<= position
)
467 sysfs_wait(scfd
, &delay
);
475 /* As we have an O_EXCL open, any use of the device
476 * which blocks STOP_ARRAY is probably a transient use,
477 * so it is reasonable to retry for a while - 5 seconds.
480 while (count
&& fd
>= 0
481 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
486 if (fd
>= 0 && err
) {
488 pr_err("failed to stop array %s: %s\n",
489 devname
, strerror(errno
));
491 cont_err("Perhaps a running process, mounted filesystem or active volume group?\n");
496 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
497 * was stopped, so We'll do it here just to be sure. Drop any
498 * partitions as well...
501 ioctl(fd
, BLKRRPART
, 0);
503 sysfs_uevent(mdi
, "change");
505 if (devnm
[0] && use_udev()) {
506 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
507 remove_devices(devnm
, mp
? mp
->path
: NULL
);
511 pr_err("stopped %s\n", devname
);
513 map_remove(&map
, devnm
);
522 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
524 struct mddev_dev
*new;
525 new = xmalloc(sizeof(*new));
526 memset(new, 0, sizeof(*new));
527 new->devname
= xstrdup(name
);
528 new->disposition
= disp
;
529 new->next
= dv
->next
;
534 static void add_faulty(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
++) {
548 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
550 if (disk
.major
== 0 && disk
.minor
== 0)
553 if ((disk
.state
& 1) == 0) /* not faulty */
555 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
556 dv
= add_one(dv
, buf
, disp
);
560 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
562 mdu_array_info_t array
;
563 mdu_disk_info_t disk
;
567 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
570 remaining_disks
= array
.nr_disks
;
571 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
575 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
577 if (disk
.major
== 0 && disk
.minor
== 0)
580 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
582 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
583 sfd
= dev_open(buf
, O_RDONLY
);
590 /* Probably not detached */
592 dv
= add_one(dv
, buf
, disp
);
596 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
598 mdu_array_info_t array
;
599 mdu_disk_info_t disk
;
604 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
606 if (array
.level
!= 10)
608 copies
= ((array
.layout
& 0xff) *
609 ((array
.layout
>> 8) & 0xff));
610 if (array
.raid_disks
% copies
)
613 remaining_disks
= array
.nr_disks
;
614 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
617 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
619 if (disk
.major
== 0 && disk
.minor
== 0)
622 set
= disk
.raid_disk
% copies
;
623 if (set_char
!= set
+ 'A')
625 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
626 dv
= add_one(dv
, buf
, dv
->disposition
);
630 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
631 struct supertype
*dev_st
, struct supertype
*tst
,
633 char *update
, char *devname
, int verbose
,
634 mdu_array_info_t
*array
)
640 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
641 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
643 tst
->ss
->uuid_from_super(tst
, duuid
);
645 /* Assume uuid matches: kernel will check */
646 memcpy(duuid
, ouuid
, sizeof(ouuid
));
647 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
648 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
649 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
650 /* Looks like it is worth a
651 * try. Need to make sure
652 * kernel will accept it
655 mdu_disk_info_t disc
;
656 /* re-add doesn't work for version-1 superblocks
659 if (array
->major_version
== 1 &&
660 get_linux_version() <= 2006018)
662 disc
.number
= mdi
.disk
.number
;
663 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
664 || disc
.major
!= 0 || disc
.minor
!= 0
667 disc
.major
= major(rdev
);
668 disc
.minor
= minor(rdev
);
669 disc
.number
= mdi
.disk
.number
;
670 disc
.raid_disk
= mdi
.disk
.raid_disk
;
671 disc
.state
= mdi
.disk
.state
;
672 if (array
->state
& (1 << MD_SB_CLUSTERED
)) {
673 /* extra flags are needed when adding to a cluster as
674 * there are two cases to distinguish
676 if (dv
->disposition
== 'c')
677 disc
.state
|= (1 << MD_DISK_CANDIDATE
);
679 disc
.state
|= (1 << MD_DISK_CLUSTER_ADD
);
681 if (dv
->writemostly
== 1)
682 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
683 if (dv
->writemostly
== 2)
684 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
685 remove_partitions(tfd
);
686 if (update
|| dv
->writemostly
> 0) {
688 tfd
= dev_open(dv
->devname
, O_RDWR
);
690 pr_err("failed to open %s for superblock update during re-add\n", dv
->devname
);
694 if (dv
->writemostly
== 1)
695 rv
= dev_st
->ss
->update_super(
696 dev_st
, NULL
, "writemostly",
697 devname
, verbose
, 0, NULL
);
698 if (dv
->writemostly
== 2)
699 rv
= dev_st
->ss
->update_super(
700 dev_st
, NULL
, "readwrite",
701 devname
, verbose
, 0, NULL
);
703 rv
= dev_st
->ss
->update_super(
704 dev_st
, NULL
, update
,
705 devname
, verbose
, 0, NULL
);
707 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
710 pr_err("failed to update superblock during re-add\n");
714 /* don't even try if disk is marked as faulty */
716 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
718 pr_err("re-added %s\n", dv
->devname
);
721 if (errno
== ENOMEM
|| errno
== EROFS
) {
722 pr_err("add new device failed for %s: %s\n",
723 dv
->devname
, strerror(errno
));
724 if (dv
->disposition
== 'M')
733 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
734 struct supertype
*tst
, mdu_array_info_t
*array
,
735 int force
, int verbose
, char *devname
,
736 char *update
, unsigned long rdev
, unsigned long long array_size
,
739 unsigned long long ldsize
;
740 struct supertype
*dev_st
= NULL
;
742 mdu_disk_info_t disc
;
744 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
745 if (dv
->disposition
== 'M')
751 if (tst
->ss
== &super0
&& ldsize
> 4ULL*1024*1024*1024*1024) {
752 /* More than 4TB is wasted on v0.90 */
754 pr_err("%s is larger than %s can effectively use.\n"
755 " Add --force is you really want to add this device.\n",
756 dv
->devname
, devname
);
759 pr_err("%s is larger than %s can effectively use.\n"
760 " Adding anyway as --force was given.\n",
761 dv
->devname
, devname
);
763 if (!tst
->ss
->external
&&
764 array
->major_version
== 0 &&
765 md_get_version(fd
)%100 < 2) {
766 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
768 pr_err("hot added %s\n",
773 pr_err("hot add failed for %s: %s\n",
774 dv
->devname
, strerror(errno
));
778 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
780 /* need to find a sample superblock to copy, and
781 * a spare slot to use.
782 * For 'external' array (well, container based),
783 * We can just load the metadata for the array->
787 /* already loaded */;
788 else if (tst
->ss
->external
) {
789 tst
->ss
->load_container(tst
, fd
, NULL
);
790 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
794 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
796 if (disc
.major
==0 && disc
.minor
==0)
798 if ((disc
.state
& 4)==0) /* sync */
800 /* Looks like a good device to try */
801 dev
= map_dev(disc
.major
, disc
.minor
, 1);
804 dfd
= dev_open(dev
, O_RDONLY
);
807 if (tst
->ss
->load_super(tst
, dfd
,
815 /* FIXME this is a bad test to be using */
816 if (!tst
->sb
&& (dv
->disposition
!= 'a'
817 && dv
->disposition
!= 'S')) {
818 /* we are re-adding a device to a
819 * completely dead array - have to depend
822 } else if (!tst
->sb
) {
823 pr_err("cannot load array metadata from %s\n", devname
);
827 /* Make sure device is large enough */
828 if (dv
->disposition
!= 'j' && /* skip size check for Journal */
830 tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
832 if (dv
->disposition
== 'M')
834 pr_err("%s not large enough to join array\n",
839 /* Possibly this device was recently part of
840 * the array and was temporarily removed, and
841 * is now being re-added. If so, we can
845 if (array
->not_persistent
==0) {
846 dev_st
= dup_super(tst
);
847 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
849 if (dev_st
&& dev_st
->sb
&& dv
->disposition
!= 'S') {
850 int rv
= attempt_re_add(fd
, tfd
, dv
,
856 dev_st
->ss
->free_super(dev_st
);
860 if (dv
->disposition
== 'M') {
862 pr_err("--re-add for %s to %s is not possible\n",
863 dv
->devname
, devname
);
866 if (dv
->disposition
== 'A') {
867 pr_err("--re-add for %s to %s is not possible\n",
868 dv
->devname
, devname
);
871 if (array
->active_disks
< array
->raid_disks
) {
872 char *avail
= xcalloc(array
->raid_disks
, 1);
876 for (d
= 0; d
< MAX_DISKS
&& found
< array
->nr_disks
; d
++) {
878 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
880 if (disc
.major
== 0 && disc
.minor
== 0)
883 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
885 avail
[disc
.raid_disk
] = 1;
887 array_failed
= !enough(array
->level
, array
->raid_disks
,
888 array
->layout
, 1, avail
);
893 pr_err("%s has failed so using --add cannot work and might destroy\n",
895 pr_err("data on %s. You should stop the array and re-assemble it.\n",
900 /* non-persistent. Must ensure that new drive
901 * is at least array->size big.
903 if (ldsize
/512 < array_size
) {
904 pr_err("%s not large enough to join array\n",
909 /* committed to really trying this device now*/
910 remove_partitions(tfd
);
912 /* in 2.6.17 and earlier, version-1 superblocks won't
913 * use the number we write, but will choose a free number.
914 * we must choose the same free number, which requires
915 * starting at 'raid_disks' and counting up
917 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
919 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
921 if (disc
.major
==0 && disc
.minor
==0)
923 if (disc
.state
& 8) /* removed */
926 disc
.major
= major(rdev
);
927 disc
.minor
= minor(rdev
);
931 disc
.number
= raid_slot
;
934 /* only add journal to array that supports journaling */
935 if (dv
->disposition
== 'j') {
939 mdp
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
941 if (strncmp(mdp
->sysfs_array_state
, "readonly", 8) != 0) {
942 pr_err("%s is not readonly, cannot add journal.\n", devname
);
946 tst
->ss
->getinfo_super(tst
, &mdi
, NULL
);
947 if (mdi
.journal_device_required
== 0) {
948 pr_err("%s does not support journal device.\n", devname
);
951 disc
.raid_disk
= array
->raid_disks
;
954 if (array
->not_persistent
==0) {
956 if (dv
->disposition
== 'j')
957 disc
.state
|= (1 << MD_DISK_JOURNAL
) | (1 << MD_DISK_SYNC
);
958 if (dv
->writemostly
== 1)
959 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
960 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
961 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
962 dv
->devname
, INVALID_SECTORS
))
964 if (tst
->ss
->write_init_super(tst
))
966 } else if (dv
->disposition
== 'A') {
967 /* this had better be raid1.
968 * As we are "--re-add"ing we must find a spare slot
971 char *used
= xcalloc(array
->raid_disks
, 1);
972 for (j
= 0; j
< tst
->max_devs
; j
++) {
973 mdu_disk_info_t disc2
;
975 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
977 if (disc2
.major
==0 && disc2
.minor
==0)
979 if (disc2
.state
& 8) /* removed */
981 if (disc2
.raid_disk
< 0)
983 if (disc2
.raid_disk
> array
->raid_disks
)
985 used
[disc2
.raid_disk
] = 1;
987 for (j
= 0 ; j
< array
->raid_disks
; j
++)
990 disc
.state
|= (1<<MD_DISK_SYNC
);
996 if (array
->state
& (1 << MD_SB_CLUSTERED
)) {
997 if (dv
->disposition
== 'c')
998 disc
.state
|= (1 << MD_DISK_CANDIDATE
);
1000 disc
.state
|= (1 << MD_DISK_CLUSTER_ADD
);
1003 if (dv
->writemostly
== 1)
1004 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1005 if (tst
->ss
->external
) {
1007 * to an external metadata container */
1008 struct mdinfo new_mdi
;
1014 strcpy(devnm
, fd2devnm(fd
));
1016 container_fd
= open_dev_excl(devnm
);
1017 if (container_fd
< 0) {
1018 pr_err("add failed for %s: could not get exclusive access to container\n",
1020 tst
->ss
->free_super(tst
);
1024 Kill(dv
->devname
, NULL
, 0, -1, 0);
1025 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
1026 if (mdmon_running(tst
->container_devnm
))
1027 tst
->update_tail
= &tst
->updates
;
1028 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
1029 dv
->devname
, INVALID_SECTORS
)) {
1031 close(container_fd
);
1034 if (tst
->update_tail
)
1035 flush_metadata_updates(tst
);
1037 tst
->ss
->sync_metadata(tst
);
1039 sra
= sysfs_read(container_fd
, NULL
, 0);
1041 pr_err("add failed for %s: sysfs_read failed\n",
1043 close(container_fd
);
1044 tst
->ss
->free_super(tst
);
1047 sra
->array
.level
= LEVEL_CONTAINER
;
1048 /* Need to set data_offset and component_size */
1049 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
1050 new_mdi
.disk
.major
= disc
.major
;
1051 new_mdi
.disk
.minor
= disc
.minor
;
1052 new_mdi
.recovery_start
= 0;
1053 /* Make sure fds are closed as they are O_EXCL which
1054 * would block add_disk */
1055 tst
->ss
->free_super(tst
);
1056 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
1057 pr_err("add new device to external metadata failed for %s\n", dv
->devname
);
1058 close(container_fd
);
1062 ping_monitor(devnm
);
1064 close(container_fd
);
1066 tst
->ss
->free_super(tst
);
1067 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1068 if (dv
->disposition
== 'j')
1069 pr_err("Failed to hot add %s as journal, "
1070 "please try restart %s.\n", dv
->devname
, devname
);
1072 pr_err("add new device failed for %s as %d: %s\n",
1073 dv
->devname
, j
, strerror(errno
));
1076 if (dv
->disposition
== 'j') {
1077 pr_err("Journal added successfully, making %s read-write\n", devname
);
1078 if (Manage_ro(devname
, fd
, -1))
1079 pr_err("Failed to make %s read-write\n", devname
);
1084 pr_err("added %s\n", dv
->devname
);
1088 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1089 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
1094 if (tst
->ss
->external
) {
1095 /* To remove a device from a container, we must
1096 * check that it isn't in use in an array.
1097 * This involves looking in the 'holders'
1098 * directory - there must be just one entry,
1100 * To ensure that it doesn't get used as a
1101 * hot spare while we are checking, we
1102 * get an O_EXCL open on the container
1106 strcpy(devnm
, fd2devnm(fd
));
1107 lfd
= open_dev_excl(devnm
);
1109 pr_err("Cannot get exclusive access to container - odd\n");
1112 /* We may not be able to check on holders in
1113 * sysfs, either because we don't have the dev num
1114 * (rdev == 0) or because the device has been detached
1115 * and the 'holders' directory no longer exists
1116 * (ret == -1). In that case, assume it is OK to
1122 ret
= sysfs_unique_holder(devnm
, rdev
);
1124 pr_err("%s is not a member, cannot remove.\n",
1130 pr_err("%s is still in use, cannot remove.\n",
1136 /* FIXME check that it is a current member */
1138 /* device has been removed and we don't know
1139 * the major:minor number
1141 int n
= write(sysfd
, "remove", 6);
1147 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1148 if (err
&& errno
== ENODEV
) {
1149 /* Old kernels rejected this if no personality
1151 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1152 struct mdinfo
*dv
= NULL
;
1155 for ( ; dv
; dv
=dv
->next
)
1156 if (dv
->disk
.major
== (int)major(rdev
) &&
1157 dv
->disk
.minor
== (int)minor(rdev
))
1160 err
= sysfs_set_str(sra
, dv
,
1169 pr_err("hot remove failed for %s: %s\n", dv
->devname
,
1175 if (tst
->ss
->external
) {
1177 * Before dropping our exclusive open we make an
1178 * attempt at preventing mdmon from seeing an
1179 * 'add' event before reconciling this 'remove'
1182 char *devnm
= fd2devnm(fd
);
1185 pr_err("unable to get container name\n");
1189 ping_manager(devnm
);
1194 pr_err("hot removed %s from %s\n",
1195 dv
->devname
, devname
);
1199 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1200 unsigned long rdev
, int verbose
, char *devname
)
1202 struct mdinfo
*mdi
, *di
;
1203 if (tst
->ss
->external
) {
1204 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1207 /* Need to find the device in sysfs and add 'want_replacement' to the
1210 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1211 if (!mdi
|| !mdi
->devs
) {
1212 pr_err("Cannot find status of %s to enable replacement - strange\n",
1216 for (di
= mdi
->devs
; di
; di
= di
->next
)
1217 if (di
->disk
.major
== (int)major(rdev
) &&
1218 di
->disk
.minor
== (int)minor(rdev
))
1222 if (di
->disk
.raid_disk
< 0) {
1223 pr_err("%s is not active and so cannot be replaced.\n",
1228 rv
= sysfs_set_str(mdi
, di
,
1229 "state", "want_replacement");
1232 pr_err("Failed to request replacement for %s\n",
1237 pr_err("Marked %s (device %d in %s) for replacement\n",
1238 dv
->devname
, di
->disk
.raid_disk
, devname
);
1239 /* If there is a matching 'with', we need to tell it which
1242 while (dv
&& dv
->disposition
!= 'W')
1245 dv
->disposition
= 'w';
1246 dv
->used
= di
->disk
.raid_disk
;
1251 pr_err("%s not found in %s so cannot --replace it\n",
1252 dv
->devname
, devname
);
1256 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1257 unsigned long rdev
, int verbose
, char *devname
)
1259 struct mdinfo
*mdi
, *di
;
1260 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1261 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1262 if (!mdi
|| !mdi
->devs
) {
1263 pr_err("Cannot find status of %s to enable replacement - strange\n",
1267 for (di
= mdi
->devs
; di
; di
= di
->next
)
1268 if (di
->disk
.major
== (int)major(rdev
) &&
1269 di
->disk
.minor
== (int)minor(rdev
))
1273 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1274 pr_err("%s is faulty and cannot be a replacement\n",
1279 if (di
->disk
.raid_disk
>= 0) {
1280 pr_err("%s is active and cannot be a replacement\n",
1285 rv
= sysfs_set_num(mdi
, di
,
1289 pr_err("Failed to set %s as preferred replacement.\n",
1294 pr_err("Marked %s in %s as replacement for device %d\n",
1295 dv
->devname
, devname
, dv
->used
);
1299 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1300 dv
->devname
, devname
);
1304 int Manage_subdevs(char *devname
, int fd
,
1305 struct mddev_dev
*devlist
, int verbose
, int test
,
1306 char *update
, int force
)
1308 /* Do something to each dev.
1310 * 'a' - add the device
1312 * If that fails EINVAL, try ADD_NEW_DISK
1313 * 'S' - add the device as a spare - don't try re-add
1314 * 'j' - add the device as a journal device
1315 * 'A' - re-add the device
1316 * 'r' - remove the device: HOT_REMOVE_DISK
1317 * device can be 'faulty' or 'detached' in which case all
1318 * matching devices are removed.
1319 * 'f' - set the device faulty SET_DISK_FAULTY
1320 * device can be 'detached' in which case any device that
1321 * is inaccessible will be marked faulty.
1322 * 'R' - mark this device as wanting replacement.
1323 * 'W' - this device is added if necessary and activated as
1324 * a replacement for a previous 'R' device.
1326 * 'w' - 'W' will be changed to 'w' when it is paired with
1327 * a 'R' device. If a 'W' is found while walking the list
1328 * it must be unpaired, and is an error.
1329 * 'M' - this is created by a 'missing' target. It is a slight
1331 * 'F' - Another variant of 'A', where the device was faulty
1332 * so must be removed from the array first.
1333 * 'c' - confirm the device as found (for clustered environments)
1335 * For 'f' and 'r', the device can also be a kernel-internal
1336 * name such as 'sdb'.
1338 mdu_array_info_t array
;
1339 unsigned long long array_size
;
1340 struct mddev_dev
*dv
;
1342 struct supertype
*tst
;
1343 char *subarray
= NULL
;
1345 int count
= 0; /* number of actions taken */
1347 struct mdinfo devinfo
;
1352 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1353 pr_err("Cannot get array info for %s\n",
1357 sysfs_init(&info
, fd
, NULL
);
1359 /* array.size is only 32 bits and may be truncated.
1360 * So read from sysfs if possible, and record number of sectors
1363 array_size
= get_component_size(fd
);
1364 if (array_size
<= 0)
1365 array_size
= array
.size
* 2;
1367 tst
= super_by_fd(fd
, &subarray
);
1369 pr_err("unsupport array - version %d.%d\n",
1370 array
.major_version
, array
.minor_version
);
1374 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1375 unsigned long rdev
= 0; /* device to add/remove etc */
1380 if (dv
->disposition
== 'c') {
1381 rv
= parse_cluster_confirm_arg(dv
->devname
,
1385 pr_err("Could not get the devname of cluster\n");
1390 if (strcmp(dv
->devname
, "failed") == 0 ||
1391 strcmp(dv
->devname
, "faulty") == 0) {
1392 if (dv
->disposition
!= 'A'
1393 && dv
->disposition
!= 'r') {
1394 pr_err("%s only meaningful with -r or --re-add, not -%c\n",
1395 dv
->devname
, dv
->disposition
);
1398 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1402 if (strcmp(dv
->devname
, "detached") == 0) {
1403 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1404 pr_err("%s only meaningful with -r of -f, not -%c\n",
1405 dv
->devname
, dv
->disposition
);
1408 add_detached(dv
, fd
, dv
->disposition
);
1412 if (strcmp(dv
->devname
, "missing") == 0) {
1413 struct mddev_dev
*add_devlist
= NULL
;
1414 struct mddev_dev
**dp
;
1415 if (dv
->disposition
== 'c') {
1416 rv
= ioctl(fd
, CLUSTERED_DISK_NACK
, NULL
);
1420 if (dv
->disposition
!= 'A') {
1421 pr_err("'missing' only meaningful with --re-add\n");
1424 add_devlist
= conf_get_devs();
1425 if (add_devlist
== NULL
) {
1426 pr_err("no devices to scan for missing members.");
1429 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1430 /* 'M' (for 'missing') is like 'A' without errors */
1431 (*dp
)->disposition
= 'M';
1433 dv
->next
= add_devlist
;
1437 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1438 strlen(dv
->devname
) == 5) {
1441 if (dv
->disposition
!= 'r' &&
1442 dv
->disposition
!= 'f') {
1443 pr_err("'%s' only meaningful with -r or -f\n",
1447 if (array
.level
!= 10) {
1448 pr_err("'%s' only meaningful with RAID10 arrays\n",
1452 copies
= ((array
.layout
& 0xff) *
1453 ((array
.layout
>> 8) & 0xff));
1454 if (array
.raid_disks
% copies
!= 0 ||
1455 dv
->devname
[4] < 'A' ||
1456 dv
->devname
[4] >= 'A' + copies
||
1458 pr_err("'%s' not meaningful with this array\n",
1462 add_set(dv
, fd
, dv
->devname
[4]);
1466 if (strchr(dv
->devname
, '/') == NULL
&&
1467 strchr(dv
->devname
, ':') == NULL
&&
1468 strlen(dv
->devname
) < 50) {
1469 /* Assume this is a kernel-internal name like 'sda1' */
1472 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1473 pr_err("%s only meaningful with -r or -f, not -%c\n",
1474 dv
->devname
, dv
->disposition
);
1478 sprintf(dname
, "dev-%s", dv
->devname
);
1479 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1482 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1483 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1484 rdev
= makedev(mj
,mn
);
1491 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1493 pr_err("%s does not appear to be a component of %s\n",
1494 dv
->devname
, devname
);
1498 } else if ((dv
->disposition
== 'r' || dv
->disposition
== 'f')
1499 && get_maj_min(dv
->devname
, &mj
, &mn
)) {
1500 /* for 'fail' and 'remove', the device might
1503 rdev
= makedev(mj
, mn
);
1506 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1510 int open_err
= errno
;
1511 if (stat(dv
->devname
, &stb
) != 0) {
1512 pr_err("Cannot find %s: %s\n",
1513 dv
->devname
, strerror(errno
));
1516 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1517 if (dv
->disposition
== 'M')
1518 /* non-fatal. Also improbable */
1520 pr_err("%s is not a block device.\n",
1524 if (dv
->disposition
== 'r')
1525 /* Be happy, the stat worked, that is
1526 * enough for --remove
1530 if (dv
->disposition
== 'M')
1533 pr_err("Cannot open %s: %s\n",
1534 dv
->devname
, strerror(open_err
));
1540 switch(dv
->disposition
){
1542 pr_err("internal error - devmode[%s]=%d\n",
1543 dv
->devname
, dv
->disposition
);
1546 case 'S': /* --add-spare */
1547 case 'j': /* --add-journal */
1549 case 'M': /* --re-add missing */
1550 case 'F': /* --re-add faulty */
1551 case 'c': /* --cluster-confirm */
1552 /* add the device */
1554 pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n");
1558 /* Let's first try to write re-add to sysfs */
1560 (dv
->disposition
== 'A' || dv
->disposition
== 'F')) {
1561 sysfs_init_dev(&devinfo
, rdev
);
1562 if (sysfs_set_str(&info
, &devinfo
, "state", "re-add") == 0) {
1563 pr_err("re-add %s to %s succeed\n",
1564 dv
->devname
, info
.sys_name
);
1569 if (dv
->disposition
== 'F')
1570 /* Need to remove first */
1571 ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1572 /* Make sure it isn't in use (in 2.6 or later) */
1573 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1575 /* We know no-one else is using it. We'll
1576 * need non-exclusive access to add it, so
1580 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1583 if (dv
->disposition
== 'M')
1585 pr_err("Cannot open %s: %s\n",
1586 dv
->devname
, strerror(errno
));
1590 if (sysfs_freeze_array(&info
) == 1)
1595 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1596 force
, verbose
, devname
, update
,
1597 rdev
, array_size
, raid_slot
);
1609 pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n");
1612 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1624 case 'f': /* set faulty */
1625 /* FIXME check current member */
1626 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1627 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1631 pr_err("set device faulty failed for %s: %s\n",
1632 dv
->devname
, strerror(errno
));
1642 pr_err("set %s faulty in %s\n",
1643 dv
->devname
, devname
);
1645 case 'R': /* Mark as replaceable */
1647 pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n");
1651 if (sysfs_freeze_array(&info
) == 1)
1656 rv
= Manage_replace(tst
, fd
, dv
,
1665 case 'W': /* --with device that doesn't match */
1666 pr_err("No matching --replace device for --with %s\n",
1669 case 'w': /* --with device which was matched */
1670 rv
= Manage_with(tst
, fd
, dv
,
1671 rdev
, verbose
, devname
);
1678 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1679 if (test
&& count
== 0)
1685 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1686 return !test
&& busy
? 2 : 1;
1689 int autodetect(void)
1691 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1693 int fd
= dev_open("9:0", O_RDONLY
);
1695 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1702 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1704 struct supertype supertype
, *st
= &supertype
;
1707 memset(st
, 0, sizeof(*st
));
1709 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1713 if (!st
->ss
->update_subarray
) {
1715 pr_err("Operation not supported for %s metadata\n",
1720 if (mdmon_running(st
->devnm
))
1721 st
->update_tail
= &st
->updates
;
1723 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1727 pr_err("Failed to update %s of subarray-%s in %s\n",
1728 update
, subarray
, dev
);
1729 } else if (st
->update_tail
)
1730 flush_metadata_updates(st
);
1732 st
->ss
->sync_metadata(st
);
1734 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1735 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1739 st
->ss
->free_super(st
);
1745 /* Move spare from one array to another If adding to destination array fails
1746 * add back to original array.
1747 * Returns 1 on success, 0 on failure */
1748 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1750 struct mddev_dev devlist
;
1753 /* try to remove and add */
1754 int fd1
= open(to_devname
, O_RDONLY
);
1755 int fd2
= open(from_devname
, O_RDONLY
);
1757 if (fd1
< 0 || fd2
< 0) {
1758 if (fd1
>=0) close(fd1
);
1759 if (fd2
>=0) close(fd2
);
1763 devlist
.next
= NULL
;
1765 devlist
.writemostly
= 0;
1766 devlist
.devname
= devname
;
1767 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1769 devlist
.disposition
= 'r';
1770 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1771 devlist
.disposition
= 'a';
1772 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1773 /* make sure manager is aware of changes */
1774 ping_manager(to_devname
);
1775 ping_manager(from_devname
);
1780 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);