]>
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
;
360 unsigned long long sync_max
, old_sync_max
;
361 unsigned long long completed
;
366 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
367 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
368 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
369 if (strncmp(buf
, "back", 4) == 0)
371 sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
);
372 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
375 rddiv
= GCD(rd1
, rd2
);
376 chunkdiv
= GCD(chunk1
, chunk2
);
377 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
380 /* Need to subtract 'reshape_position' from
381 * array size to get equivalent of sync_max.
382 * Size calculation based on raid5_size in kernel.
384 unsigned long long size
= mdi
->component_size
;
387 /* rd1 must be smaller */
388 position
= (position
/ sectors
- 1) * sectors
;
389 sync_max
= size
- position
/rd1
;
391 position
= (position
/ sectors
+ 2) * sectors
;
392 sync_max
= position
/rd1
;
394 if (sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) < 0)
395 old_sync_max
= mdi
->component_size
;
396 /* Must not advance sync_max as that could confuse
397 * the reshape monitor */
398 if (sync_max
< old_sync_max
)
399 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
400 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
402 /* That should have set things going again. Now we
403 * wait a little while (3 second max) for sync_completed
404 * to reach the target.
405 * The reshape process can block for 500msec if
406 * the sync speed limit is hit, so we need to wait
407 * a lot longer than that. 1 second is usually
411 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
412 while (scfd
>= 0 && delay
> 0 && old_sync_max
> 0) {
413 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
414 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
415 if (strncmp(buf
, "none", 4) == 0) {
416 /* Either reshape has aborted, or hasn't
417 * quite started yet. Wait a bit and
418 * check 'sync_action' to see.
421 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
));
422 if (strncmp(buf
, "reshape", 7) != 0)
426 if (sysfs_fd_get_ll(scfd
, &completed
) == 0 &&
427 (completed
> sync_max
||
428 (completed
== sync_max
&& curr
!= position
))) {
429 while (completed
> sync_max
) {
430 sync_max
+= sectors
/ rd1
;
436 if (sync_max
< old_sync_max
)
437 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
440 if (!backwards
&& curr
>= position
)
442 if (backwards
&& curr
<= position
)
444 sysfs_wait(scfd
, &delay
);
451 /* As we have an O_EXCL open, any use of the device
452 * which blocks STOP_ARRAY is probably a transient use,
453 * so it is reasonable to retry for a while - 5 seconds.
456 while (count
&& fd
>= 0
457 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
462 if (fd
>= 0 && err
) {
464 pr_err("failed to stop array %s: %s\n",
465 devname
, strerror(errno
));
467 cont_err("Perhaps a running "
468 "process, mounted filesystem "
469 "or active volume group?\n");
474 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
475 * was stopped, so We'll do it here just to be sure. Drop any
476 * partitions as well...
479 ioctl(fd
, BLKRRPART
, 0);
481 sysfs_uevent(mdi
, "change");
483 if (devnm
[0] && use_udev()) {
484 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
485 remove_devices(devnm
, mp
? mp
->path
: NULL
);
489 pr_err("stopped %s\n", devname
);
491 map_remove(&map
, devnm
);
500 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
502 struct mddev_dev
*new;
503 new = xmalloc(sizeof(*new));
504 memset(new, 0, sizeof(*new));
505 new->devname
= xstrdup(name
);
506 new->disposition
= disp
;
507 new->next
= dv
->next
;
512 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
514 mdu_array_info_t array
;
515 mdu_disk_info_t disk
;
519 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
522 remaining_disks
= array
.nr_disks
;
523 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
526 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
528 if (disk
.major
== 0 && disk
.minor
== 0)
531 if ((disk
.state
& 1) == 0) /* not faulty */
533 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
534 dv
= add_one(dv
, buf
, disp
);
538 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
540 mdu_array_info_t array
;
541 mdu_disk_info_t disk
;
545 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
548 remaining_disks
= array
.nr_disks
;
549 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
553 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
555 if (disk
.major
== 0 && disk
.minor
== 0)
558 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
560 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
561 sfd
= dev_open(buf
, O_RDONLY
);
568 /* Probably not detached */
570 dv
= add_one(dv
, buf
, disp
);
574 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
576 mdu_array_info_t array
;
577 mdu_disk_info_t disk
;
582 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
584 if (array
.level
!= 10)
586 copies
= ((array
.layout
& 0xff) *
587 ((array
.layout
>> 8) & 0xff));
588 if (array
.raid_disks
% copies
)
591 remaining_disks
= array
.nr_disks
;
592 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
595 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
597 if (disk
.major
== 0 && disk
.minor
== 0)
600 set
= disk
.raid_disk
% copies
;
601 if (set_char
!= set
+ 'A')
603 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
604 dv
= add_one(dv
, buf
, dv
->disposition
);
608 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
609 struct supertype
*dev_st
, struct supertype
*tst
,
611 char *update
, char *devname
, int verbose
,
612 mdu_array_info_t
*array
)
618 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
619 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
621 tst
->ss
->uuid_from_super(tst
, duuid
);
623 /* Assume uuid matches: kernel will check */
624 memcpy(duuid
, ouuid
, sizeof(ouuid
));
625 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
626 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
627 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
628 /* Looks like it is worth a
629 * try. Need to make sure
630 * kernel will accept it
633 mdu_disk_info_t disc
;
634 /* re-add doesn't work for version-1 superblocks
637 if (array
->major_version
== 1 &&
638 get_linux_version() <= 2006018)
640 disc
.number
= mdi
.disk
.number
;
641 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
642 || disc
.major
!= 0 || disc
.minor
!= 0
645 disc
.major
= major(rdev
);
646 disc
.minor
= minor(rdev
);
647 disc
.number
= mdi
.disk
.number
;
648 disc
.raid_disk
= mdi
.disk
.raid_disk
;
649 disc
.state
= mdi
.disk
.state
;
650 if (dv
->writemostly
== 1)
651 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
652 if (dv
->writemostly
== 2)
653 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
654 remove_partitions(tfd
);
655 if (update
|| dv
->writemostly
> 0) {
657 tfd
= dev_open(dv
->devname
, O_RDWR
);
659 pr_err("failed to open %s for"
660 " superblock update during re-add\n", dv
->devname
);
664 if (dv
->writemostly
== 1)
665 rv
= dev_st
->ss
->update_super(
666 dev_st
, NULL
, "writemostly",
667 devname
, verbose
, 0, NULL
);
668 if (dv
->writemostly
== 2)
669 rv
= dev_st
->ss
->update_super(
670 dev_st
, NULL
, "readwrite",
671 devname
, verbose
, 0, NULL
);
673 rv
= dev_st
->ss
->update_super(
674 dev_st
, NULL
, update
,
675 devname
, verbose
, 0, NULL
);
677 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
680 pr_err("failed to update"
681 " superblock during re-add\n");
685 /* don't even try if disk is marked as faulty */
687 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
689 pr_err("re-added %s\n", dv
->devname
);
692 if (errno
== ENOMEM
|| errno
== EROFS
) {
693 pr_err("add new device failed for %s: %s\n",
694 dv
->devname
, strerror(errno
));
695 if (dv
->disposition
== 'M')
704 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
705 struct supertype
*tst
, mdu_array_info_t
*array
,
706 int force
, int verbose
, char *devname
,
707 char *update
, unsigned long rdev
, unsigned long long array_size
)
709 unsigned long long ldsize
;
710 struct supertype
*dev_st
= NULL
;
712 mdu_disk_info_t disc
;
714 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
715 if (dv
->disposition
== 'M')
721 if (tst
->ss
->validate_geometry(
722 tst
, array
->level
, array
->layout
,
723 array
->raid_disks
, NULL
,
724 ldsize
>> 9, INVALID_SECTORS
, NULL
, NULL
, 0) == 0) {
726 pr_err("%s is larger than %s can "
728 " Add --force is you "
729 "really want to add this device.\n",
730 dv
->devname
, devname
);
733 pr_err("%s is larger than %s can "
735 " Adding anyway as --force "
737 dv
->devname
, devname
);
739 if (!tst
->ss
->external
&&
740 array
->major_version
== 0 &&
741 md_get_version(fd
)%100 < 2) {
742 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
744 pr_err("hot added %s\n",
749 pr_err("hot add failed for %s: %s\n",
750 dv
->devname
, strerror(errno
));
754 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
756 /* need to find a sample superblock to copy, and
757 * a spare slot to use.
758 * For 'external' array (well, container based),
759 * We can just load the metadata for the array->
763 /* already loaded */;
764 else if (tst
->ss
->external
) {
765 tst
->ss
->load_container(tst
, fd
, NULL
);
766 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
770 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
772 if (disc
.major
==0 && disc
.minor
==0)
774 if ((disc
.state
& 4)==0) /* sync */
776 /* Looks like a good device to try */
777 dev
= map_dev(disc
.major
, disc
.minor
, 1);
780 dfd
= dev_open(dev
, O_RDONLY
);
783 if (tst
->ss
->load_super(tst
, dfd
,
791 /* FIXME this is a bad test to be using */
792 if (!tst
->sb
&& dv
->disposition
!= 'a') {
793 /* we are re-adding a device to a
794 * completely dead array - have to depend
797 } else if (!tst
->sb
) {
798 pr_err("cannot load array metadata from %s\n", devname
);
802 /* Make sure device is large enough */
803 if (tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
805 if (dv
->disposition
== 'M')
807 pr_err("%s not large enough to join array\n",
812 /* Possibly this device was recently part of
813 * the array and was temporarily removed, and
814 * is now being re-added. If so, we can
818 if (array
->not_persistent
==0) {
819 dev_st
= dup_super(tst
);
820 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
822 if (dev_st
&& dev_st
->sb
) {
823 int rv
= attempt_re_add(fd
, tfd
, dv
,
829 dev_st
->ss
->free_super(dev_st
);
833 if (dv
->disposition
== 'M') {
835 pr_err("--re-add for %s to %s is not possible\n",
836 dv
->devname
, devname
);
839 if (dv
->disposition
== 'A') {
840 pr_err("--re-add for %s to %s is not possible\n",
841 dv
->devname
, devname
);
844 if (array
->active_disks
< array
->raid_disks
) {
845 char *avail
= xcalloc(array
->raid_disks
, 1);
849 for (d
= 0; d
< MAX_DISKS
&& found
< array
->active_disks
; d
++) {
851 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
853 if (disc
.major
== 0 && disc
.minor
== 0)
855 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
857 avail
[disc
.raid_disk
] = 1;
860 array_failed
= !enough(array
->level
, array
->raid_disks
,
861 array
->layout
, 1, avail
);
865 pr_err("%s has failed so using --add cannot work and might destroy\n",
867 pr_err("data on %s. You should stop the array and re-assemble it.\n",
872 /* non-persistent. Must ensure that new drive
873 * is at least array->size big.
875 if (ldsize
/512 < array_size
) {
876 pr_err("%s not large enough to join array\n",
881 /* committed to really trying this device now*/
882 remove_partitions(tfd
);
884 /* in 2.6.17 and earlier, version-1 superblocks won't
885 * use the number we write, but will choose a free number.
886 * we must choose the same free number, which requires
887 * starting at 'raid_disks' and counting up
889 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
891 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
893 if (disc
.major
==0 && disc
.minor
==0)
895 if (disc
.state
& 8) /* removed */
898 disc
.major
= major(rdev
);
899 disc
.minor
= minor(rdev
);
902 if (array
->not_persistent
==0) {
904 if (dv
->writemostly
== 1)
905 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
906 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
907 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
908 dv
->devname
, INVALID_SECTORS
))
910 if (tst
->ss
->write_init_super(tst
))
912 } else if (dv
->disposition
== 'A') {
913 /* this had better be raid1.
914 * As we are "--re-add"ing we must find a spare slot
917 char *used
= xcalloc(array
->raid_disks
, 1);
918 for (j
= 0; j
< tst
->max_devs
; j
++) {
919 mdu_disk_info_t disc2
;
921 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
923 if (disc2
.major
==0 && disc2
.minor
==0)
925 if (disc2
.state
& 8) /* removed */
927 if (disc2
.raid_disk
< 0)
929 if (disc2
.raid_disk
> array
->raid_disks
)
931 used
[disc2
.raid_disk
] = 1;
933 for (j
= 0 ; j
< array
->raid_disks
; j
++)
936 disc
.state
|= (1<<MD_DISK_SYNC
);
941 if (dv
->writemostly
== 1)
942 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
943 if (tst
->ss
->external
) {
945 * to an external metadata container */
946 struct mdinfo new_mdi
;
952 strcpy(devnm
, fd2devnm(fd
));
954 container_fd
= open_dev_excl(devnm
);
955 if (container_fd
< 0) {
956 pr_err("add failed for %s:"
957 " could not get exclusive access to container\n",
959 tst
->ss
->free_super(tst
);
963 Kill(dv
->devname
, NULL
, 0, -1, 0);
964 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
965 if (mdmon_running(tst
->container_devnm
))
966 tst
->update_tail
= &tst
->updates
;
967 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
968 dv
->devname
, INVALID_SECTORS
)) {
973 if (tst
->update_tail
)
974 flush_metadata_updates(tst
);
976 tst
->ss
->sync_metadata(tst
);
978 sra
= sysfs_read(container_fd
, NULL
, 0);
980 pr_err("add failed for %s: sysfs_read failed\n",
983 tst
->ss
->free_super(tst
);
986 sra
->array
.level
= LEVEL_CONTAINER
;
987 /* Need to set data_offset and component_size */
988 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
989 new_mdi
.disk
.major
= disc
.major
;
990 new_mdi
.disk
.minor
= disc
.minor
;
991 new_mdi
.recovery_start
= 0;
992 /* Make sure fds are closed as they are O_EXCL which
993 * would block add_disk */
994 tst
->ss
->free_super(tst
);
995 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
996 pr_err("add new device to external metadata"
997 " failed for %s\n", dv
->devname
);
1002 ping_monitor(devnm
);
1004 close(container_fd
);
1006 tst
->ss
->free_super(tst
);
1007 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1008 pr_err("add new device failed for %s as %d: %s\n",
1009 dv
->devname
, j
, strerror(errno
));
1014 pr_err("added %s\n", dv
->devname
);
1018 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1019 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
1024 if (tst
->ss
->external
) {
1025 /* To remove a device from a container, we must
1026 * check that it isn't in use in an array.
1027 * This involves looking in the 'holders'
1028 * directory - there must be just one entry,
1030 * To ensure that it doesn't get used as a
1031 * hot spare while we are checking, we
1032 * get an O_EXCL open on the container
1036 strcpy(devnm
, fd2devnm(fd
));
1037 lfd
= open_dev_excl(devnm
);
1039 pr_err("Cannot get exclusive access "
1040 " to container - odd\n");
1043 /* We may not be able to check on holders in
1044 * sysfs, either because we don't have the dev num
1045 * (rdev == 0) or because the device has been detached
1046 * and the 'holders' directory no longer exists
1047 * (ret == -1). In that case, assume it is OK to
1053 ret
= sysfs_unique_holder(devnm
, rdev
);
1055 pr_err("%s is not a member, cannot remove.\n",
1061 pr_err("%s is still in use, cannot remove.\n",
1067 /* FIXME check that it is a current member */
1069 /* device has been removed and we don't know
1070 * the major:minor number
1072 int n
= write(sysfd
, "remove", 6);
1078 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1079 if (err
&& errno
== ENODEV
) {
1080 /* Old kernels rejected this if no personality
1082 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1083 struct mdinfo
*dv
= NULL
;
1086 for ( ; dv
; dv
=dv
->next
)
1087 if (dv
->disk
.major
== (int)major(rdev
) &&
1088 dv
->disk
.minor
== (int)minor(rdev
))
1091 err
= sysfs_set_str(sra
, dv
,
1100 pr_err("hot remove failed "
1101 "for %s: %s\n", dv
->devname
,
1107 if (tst
->ss
->external
) {
1109 * Before dropping our exclusive open we make an
1110 * attempt at preventing mdmon from seeing an
1111 * 'add' event before reconciling this 'remove'
1114 char *devnm
= fd2devnm(fd
);
1117 pr_err("unable to get container name\n");
1121 ping_manager(devnm
);
1126 pr_err("hot removed %s from %s\n",
1127 dv
->devname
, devname
);
1131 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1132 unsigned long rdev
, int verbose
, char *devname
)
1134 struct mdinfo
*mdi
, *di
;
1135 if (tst
->ss
->external
) {
1136 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1139 /* Need to find the device in sysfs and add 'want_replacement' to the
1142 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1143 if (!mdi
|| !mdi
->devs
) {
1144 pr_err("Cannot find status of %s to enable replacement - strange\n",
1148 for (di
= mdi
->devs
; di
; di
= di
->next
)
1149 if (di
->disk
.major
== (int)major(rdev
) &&
1150 di
->disk
.minor
== (int)minor(rdev
))
1154 if (di
->disk
.raid_disk
< 0) {
1155 pr_err("%s is not active and so cannot be replaced.\n",
1160 rv
= sysfs_set_str(mdi
, di
,
1161 "state", "want_replacement");
1164 pr_err("Failed to request replacement for %s\n",
1169 pr_err("Marked %s (device %d in %s) for replacement\n",
1170 dv
->devname
, di
->disk
.raid_disk
, devname
);
1171 /* If there is a matching 'with', we need to tell it which
1174 while (dv
&& dv
->disposition
!= 'W')
1177 dv
->disposition
= 'w';
1178 dv
->used
= di
->disk
.raid_disk
;
1183 pr_err("%s not found in %s so cannot --replace it\n",
1184 dv
->devname
, devname
);
1188 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1189 unsigned long rdev
, int verbose
, char *devname
)
1191 struct mdinfo
*mdi
, *di
;
1192 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1193 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1194 if (!mdi
|| !mdi
->devs
) {
1195 pr_err("Cannot find status of %s to enable replacement - strange\n",
1199 for (di
= mdi
->devs
; di
; di
= di
->next
)
1200 if (di
->disk
.major
== (int)major(rdev
) &&
1201 di
->disk
.minor
== (int)minor(rdev
))
1205 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1206 pr_err("%s is faulty and cannot be a replacement\n",
1211 if (di
->disk
.raid_disk
>= 0) {
1212 pr_err("%s is active and cannot be a replacement\n",
1217 rv
= sysfs_set_num(mdi
, di
,
1221 pr_err("Failed to %s as preferred replacement.\n",
1226 pr_err("Marked %s in %s as replacement for device %d\n",
1227 dv
->devname
, devname
, dv
->used
);
1231 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1232 dv
->devname
, devname
);
1236 int Manage_subdevs(char *devname
, int fd
,
1237 struct mddev_dev
*devlist
, int verbose
, int test
,
1238 char *update
, int force
)
1240 /* Do something to each dev.
1242 * 'a' - add the device
1244 * If that fails EINVAL, try ADD_NEW_DISK
1245 * 'A' - re-add the device
1246 * 'r' - remove the device: HOT_REMOVE_DISK
1247 * device can be 'faulty' or 'detached' in which case all
1248 * matching devices are removed.
1249 * 'f' - set the device faulty SET_DISK_FAULTY
1250 * device can be 'detached' in which case any device that
1251 * is inaccessible will be marked faulty.
1252 * 'R' - mark this device as wanting replacement.
1253 * 'W' - this device is added if necessary and activated as
1254 * a replacement for a previous 'R' device.
1256 * 'w' - 'W' will be changed to 'w' when it is paired with
1257 * a 'R' device. If a 'W' is found while walking the list
1258 * it must be unpaired, and is an error.
1259 * 'M' - this is created by a 'missing' target. It is a slight
1261 * 'F' - Another variant of 'A', where the device was faulty
1262 * so must be removed from the array first.
1264 * For 'f' and 'r', the device can also be a kernel-internal
1265 * name such as 'sdb'.
1267 mdu_array_info_t array
;
1268 unsigned long long array_size
;
1269 struct mddev_dev
*dv
;
1272 struct supertype
*tst
;
1273 char *subarray
= NULL
;
1275 int count
= 0; /* number of actions taken */
1280 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1281 pr_err("Cannot get array info for %s\n",
1285 sysfs_init(&info
, fd
, NULL
);
1287 /* array.size is only 32 bits and may be truncated.
1288 * So read from sysfs if possible, and record number of sectors
1291 array_size
= get_component_size(fd
);
1292 if (array_size
<= 0)
1293 array_size
= array
.size
* 2;
1295 tst
= super_by_fd(fd
, &subarray
);
1297 pr_err("unsupport array - version %d.%d\n",
1298 array
.major_version
, array
.minor_version
);
1303 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1306 if (strcmp(dv
->devname
, "failed") == 0 ||
1307 strcmp(dv
->devname
, "faulty") == 0) {
1308 if (dv
->disposition
!= 'A'
1309 && dv
->disposition
!= 'r') {
1310 pr_err("%s only meaningful "
1311 "with -r or --re-add, not -%c\n",
1312 dv
->devname
, dv
->disposition
);
1315 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1319 if (strcmp(dv
->devname
, "detached") == 0) {
1320 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1321 pr_err("%s only meaningful "
1322 "with -r of -f, not -%c\n",
1323 dv
->devname
, dv
->disposition
);
1326 add_detached(dv
, fd
, dv
->disposition
);
1330 if (strcmp(dv
->devname
, "missing") == 0) {
1331 struct mddev_dev
*add_devlist
= NULL
;
1332 struct mddev_dev
**dp
;
1333 if (dv
->disposition
!= 'A') {
1334 pr_err("'missing' only meaningful "
1338 add_devlist
= conf_get_devs();
1339 if (add_devlist
== NULL
) {
1340 pr_err("no devices to scan for missing members.");
1343 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1344 /* 'M' (for 'missing') is like 'A' without errors */
1345 (*dp
)->disposition
= 'M';
1347 dv
->next
= add_devlist
;
1351 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1352 strlen(dv
->devname
) == 5) {
1355 if (dv
->disposition
!= 'r' &&
1356 dv
->disposition
!= 'f') {
1357 pr_err("'%s' only meaningful with -r or -f\n",
1361 if (array
.level
!= 10) {
1362 pr_err("'%s' only meaningful with RAID10 arrays\n",
1366 copies
= ((array
.layout
& 0xff) *
1367 ((array
.layout
>> 8) & 0xff));
1368 if (array
.raid_disks
% copies
!= 0 ||
1369 dv
->devname
[4] < 'A' ||
1370 dv
->devname
[4] >= 'A' + copies
||
1372 pr_err("'%s' not meaningful with this array\n",
1376 add_set(dv
, fd
, dv
->devname
[4]);
1380 if (strchr(dv
->devname
, '/') == NULL
&&
1381 strchr(dv
->devname
, ':') == NULL
&&
1382 strlen(dv
->devname
) < 50) {
1383 /* Assume this is a kernel-internal name like 'sda1' */
1386 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1387 pr_err("%s only meaningful "
1388 "with -r or -f, not -%c\n",
1389 dv
->devname
, dv
->disposition
);
1393 sprintf(dname
, "dev-%s", dv
->devname
);
1394 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1398 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1399 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1400 stb
.st_rdev
= makedev(mj
,mn
);
1407 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1409 pr_err("%s does not appear "
1410 "to be a component of %s\n",
1411 dv
->devname
, devname
);
1416 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1420 int open_err
= errno
;
1421 if (stat(dv
->devname
, &stb
) != 0) {
1422 pr_err("Cannot find %s: %s\n",
1423 dv
->devname
, strerror(errno
));
1426 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1427 if (dv
->disposition
== 'M')
1428 /* non-fatal. Also improbable */
1430 pr_err("%s is not a block device.\n",
1434 if (dv
->disposition
== 'r')
1435 /* Be happy, the stat worked, that is
1436 * enough for --remove
1440 if (dv
->disposition
== 'M')
1443 pr_err("Cannot open %s: %s\n",
1444 dv
->devname
, strerror(open_err
));
1449 switch(dv
->disposition
){
1451 pr_err("internal error - devmode[%s]=%d\n",
1452 dv
->devname
, dv
->disposition
);
1456 case 'M': /* --re-add missing */
1457 case 'F': /* --re-add faulty */
1458 /* add the device */
1460 pr_err("Cannot add disks to a"
1461 " \'member\' array, perform this"
1462 " operation on the parent container\n");
1465 if (dv
->disposition
== 'F')
1466 /* Need to remove first */
1467 ioctl(fd
, HOT_REMOVE_DISK
,
1468 (unsigned long)stb
.st_rdev
);
1469 /* Make sure it isn't in use (in 2.6 or later) */
1470 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1472 /* We know no-one else is using it. We'll
1473 * need non-exclusive access to add it, so
1477 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1480 if (dv
->disposition
== 'M')
1482 pr_err("Cannot open %s: %s\n",
1483 dv
->devname
, strerror(errno
));
1487 if (sysfs_freeze_array(&info
) == 1)
1492 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1493 force
, verbose
, devname
, update
,
1494 stb
.st_rdev
, array_size
);
1506 pr_err("Cannot remove disks from a"
1507 " \'member\' array, perform this"
1508 " operation on the parent container\n");
1511 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1512 stb
.st_rdev
, verbose
,
1523 case 'f': /* set faulty */
1524 /* FIXME check current member */
1525 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1526 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1527 (unsigned long) stb
.st_rdev
))) {
1530 pr_err("set device faulty failed for %s: %s\n",
1531 dv
->devname
, strerror(errno
));
1541 pr_err("set %s faulty in %s\n",
1542 dv
->devname
, devname
);
1544 case 'R': /* Mark as replaceable */
1546 pr_err("Cannot replace disks in a"
1547 " \'member\' array, perform this"
1548 " operation on the parent container\n");
1552 if (sysfs_freeze_array(&info
) == 1)
1557 rv
= Manage_replace(tst
, fd
, dv
,
1558 stb
.st_rdev
, verbose
,
1566 case 'W': /* --with device that doesn't match */
1567 pr_err("No matching --replace device for --with %s\n",
1570 case 'w': /* --with device which was matched */
1571 rv
= Manage_with(tst
, fd
, dv
,
1572 stb
.st_rdev
, verbose
, devname
);
1579 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1580 if (test
&& count
== 0)
1586 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1587 return !test
&& busy
? 2 : 1;
1590 int autodetect(void)
1592 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1594 int fd
= dev_open("9:0", O_RDONLY
);
1596 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1603 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1605 struct supertype supertype
, *st
= &supertype
;
1608 memset(st
, 0, sizeof(*st
));
1610 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1614 if (!st
->ss
->update_subarray
) {
1616 pr_err("Operation not supported for %s metadata\n",
1621 if (mdmon_running(st
->devnm
))
1622 st
->update_tail
= &st
->updates
;
1624 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1628 pr_err("Failed to update %s of subarray-%s in %s\n",
1629 update
, subarray
, dev
);
1630 } else if (st
->update_tail
)
1631 flush_metadata_updates(st
);
1633 st
->ss
->sync_metadata(st
);
1635 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1636 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1640 st
->ss
->free_super(st
);
1646 /* Move spare from one array to another If adding to destination array fails
1647 * add back to original array.
1648 * Returns 1 on success, 0 on failure */
1649 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1651 struct mddev_dev devlist
;
1654 /* try to remove and add */
1655 int fd1
= open(to_devname
, O_RDONLY
);
1656 int fd2
= open(from_devname
, O_RDONLY
);
1658 if (fd1
< 0 || fd2
< 0) {
1659 if (fd1
>=0) close(fd1
);
1660 if (fd2
>=0) close(fd2
);
1664 devlist
.next
= NULL
;
1666 devlist
.writemostly
= 0;
1667 devlist
.devname
= devname
;
1668 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1670 devlist
.disposition
= 'r';
1671 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1672 devlist
.disposition
= 'a';
1673 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1674 /* make sure manager is aware of changes */
1675 ping_manager(to_devname
);
1676 ping_manager(from_devname
);
1681 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);