]>
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 /* If this is an mdmon managed array, just write 'inactive'
220 * to the array state and let mdmon clear up.
222 strcpy(devnm
, fd2devnm(fd
));
223 /* Get EXCL access first. If this fails, then attempting
224 * to stop is probably a bad idea.
226 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_COMPONENT
|GET_VERSION
);
227 if (mdi
&& is_subarray(mdi
->text_version
)) {
229 strncpy(container
, mdi
->text_version
+1, sizeof(container
));
230 container
[sizeof(container
)-1] = 0;
231 sl
= strchr(container
, '/');
238 while (((fd
= ((devnm
[0] == '/')
239 ?open(devname
, O_RDONLY
|O_EXCL
)
240 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0
241 || strcmp(fd2devnm(fd
), devnm
) != 0)
243 && mdmon_running(container
)
247 flush_mdmon(container
);
250 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
254 pr_err("Cannot get exclusive access to %s:Perhaps a running process, mounted filesystem or active volume group?\n",
259 mdi
->array
.level
> 0 &&
260 is_subarray(mdi
->text_version
)) {
262 /* This is mdmon managed. */
265 /* As we have an O_EXCL open, any use of the device
266 * which blocks STOP_ARRAY is probably a transient use,
267 * so it is reasonable to retry for a while - 5 seconds.
271 (err
= sysfs_set_str(mdi
, NULL
,
280 pr_err("failed to stop array %s: %s\n",
281 devname
, strerror(errno
));
286 /* Give monitor a chance to act */
287 ping_monitor(mdi
->text_version
);
289 fd
= open_dev_excl(devnm
);
292 pr_err("failed to completely stop %s: Device is busy\n",
298 mdi
->array
.major_version
== -1 &&
299 mdi
->array
.minor_version
== -2 &&
300 !is_subarray(mdi
->text_version
)) {
301 struct mdstat_ent
*mds
, *m
;
302 /* container, possibly mdmon-managed.
303 * Make sure mdmon isn't opening it, which
304 * would interfere with the 'stop'
306 ping_monitor(mdi
->sys_name
);
308 /* now check that there are no existing arrays
309 * which are members of this array
311 mds
= mdstat_read(0, 0);
312 for (m
= mds
; m
; m
= m
->next
)
313 if (m
->metadata_version
&&
314 strncmp(m
->metadata_version
, "external:", 9)==0 &&
315 metadata_container_matches(m
->metadata_version
+9,
318 pr_err("Cannot stop container %s: member %s still active\n",
326 /* If the array is undergoing a reshape which changes the number
327 * of devices, then it would be nice to stop it at a point where
328 * it has completed a full number of stripes in both old and
329 * new layouts as this will allow the reshape to be reverted.
330 * So if 'sync_action' is "reshape" and 'raid_disks' shows two
331 * different numbers, then
333 * - set sync_max to next multiple of both data_disks and
334 * chunk sizes (or next but one)
336 * - wait on 'sync_completed' for that point to be reached.
338 if (mdi
&& (mdi
->array
.level
>= 4 && mdi
->array
.level
<= 6) &&
339 sysfs_attribute_available(mdi
, NULL
, "sync_action") &&
340 sysfs_attribute_available(mdi
, NULL
, "reshape_direction") &&
341 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, 20) > 0 &&
342 strcmp(buf
, "reshape\n") == 0 &&
343 sysfs_get_two(mdi
, NULL
, "raid_disks", &rd1
, &rd2
) == 2 &&
344 sysfs_set_str(mdi
, NULL
, "sync_action", "frozen") == 0) {
345 /* Array is frozen */
346 unsigned long long position
, curr
;
347 unsigned long long chunk1
, chunk2
;
348 unsigned long long rddiv
, chunkdiv
;
349 unsigned long long sectors
;
350 unsigned long long sync_max
, old_sync_max
;
351 unsigned long long completed
;
356 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
357 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
358 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
359 if (strncmp(buf
, "back", 4) == 0)
361 sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
);
362 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
365 rddiv
= GCD(rd1
, rd2
);
366 chunkdiv
= GCD(chunk1
, chunk2
);
367 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
370 /* Need to subtract 'reshape_position' from
371 * array size to get equivalent of sync_max.
372 * Size calculation based on raid5_size in kernel.
374 unsigned long long size
= mdi
->component_size
;
377 /* rd1 must be smaller */
378 position
= (position
/ sectors
- 1) * sectors
;
379 sync_max
= size
- position
/rd1
;
381 position
= (position
/ sectors
+ 2) * sectors
;
382 sync_max
= position
/rd1
;
384 if (sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) < 0)
385 old_sync_max
= mdi
->component_size
;
386 /* Must not advance sync_max as that could confuse
387 * the reshape monitor */
388 if (sync_max
< old_sync_max
)
389 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
390 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
392 /* That should have set things going again. Now we
393 * wait a little while (3 second max) for sync_completed
394 * to reach the target.
395 * The reshape process can block for 500msec if
396 * the sync speed limit is hit, so we need to wait
397 * a lot longer than that. 1 second is usually
401 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
402 while (scfd
>= 0 && delay
> 0 && old_sync_max
> 0) {
403 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
404 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
405 if (strncmp(buf
, "none", 4) == 0) {
406 /* Either reshape has aborted, or hasn't
407 * quite started yet. Wait a bit and
408 * check 'sync_action' to see.
411 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
));
412 if (strncmp(buf
, "reshape", 7) != 0)
416 if (sysfs_fd_get_ll(scfd
, &completed
) == 0 &&
417 (completed
> sync_max
||
418 (completed
== sync_max
&& curr
!= position
))) {
419 while (completed
> sync_max
) {
420 sync_max
+= sectors
/ rd1
;
426 if (sync_max
< old_sync_max
)
427 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
430 if (!backwards
&& curr
>= position
)
432 if (backwards
&& curr
<= position
)
434 sysfs_wait(scfd
, &delay
);
441 /* As we have an O_EXCL open, any use of the device
442 * which blocks STOP_ARRAY is probably a transient use,
443 * so it is reasonable to retry for a while - 5 seconds.
446 while (count
&& fd
>= 0
447 && (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0
452 if (fd
>= 0 && err
) {
454 pr_err("failed to stop array %s: %s\n",
455 devname
, strerror(errno
));
457 cont_err("Perhaps a running process, mounted filesystem or active volume group?\n");
462 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
463 * was stopped, so We'll do it here just to be sure. Drop any
464 * partitions as well...
467 ioctl(fd
, BLKRRPART
, 0);
469 sysfs_uevent(mdi
, "change");
471 if (devnm
[0] && use_udev()) {
472 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
473 remove_devices(devnm
, mp
? mp
->path
: NULL
);
477 pr_err("stopped %s\n", devname
);
479 map_remove(&map
, devnm
);
488 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
490 struct mddev_dev
*new;
491 new = xmalloc(sizeof(*new));
492 memset(new, 0, sizeof(*new));
493 new->devname
= xstrdup(name
);
494 new->disposition
= disp
;
495 new->next
= dv
->next
;
500 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
502 mdu_array_info_t array
;
503 mdu_disk_info_t disk
;
507 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
510 remaining_disks
= array
.nr_disks
;
511 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
514 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
516 if (disk
.major
== 0 && disk
.minor
== 0)
519 if ((disk
.state
& 1) == 0) /* not faulty */
521 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
522 dv
= add_one(dv
, buf
, disp
);
526 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
528 mdu_array_info_t array
;
529 mdu_disk_info_t disk
;
533 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
536 remaining_disks
= array
.nr_disks
;
537 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
541 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
543 if (disk
.major
== 0 && disk
.minor
== 0)
546 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
548 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
549 sfd
= dev_open(buf
, O_RDONLY
);
556 /* Probably not detached */
558 dv
= add_one(dv
, buf
, disp
);
562 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
564 mdu_array_info_t array
;
565 mdu_disk_info_t disk
;
570 if (ioctl(fd
, GET_ARRAY_INFO
, &array
) != 0)
572 if (array
.level
!= 10)
574 copies
= ((array
.layout
& 0xff) *
575 ((array
.layout
>> 8) & 0xff));
576 if (array
.raid_disks
% copies
)
579 remaining_disks
= array
.nr_disks
;
580 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
583 if (ioctl(fd
, GET_DISK_INFO
, &disk
) != 0)
585 if (disk
.major
== 0 && disk
.minor
== 0)
588 set
= disk
.raid_disk
% copies
;
589 if (set_char
!= set
+ 'A')
591 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
592 dv
= add_one(dv
, buf
, dv
->disposition
);
596 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
597 struct supertype
*dev_st
, struct supertype
*tst
,
599 char *update
, char *devname
, int verbose
,
600 mdu_array_info_t
*array
)
606 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
607 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
609 tst
->ss
->uuid_from_super(tst
, duuid
);
611 /* Assume uuid matches: kernel will check */
612 memcpy(duuid
, ouuid
, sizeof(ouuid
));
613 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
614 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
615 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
616 /* Looks like it is worth a
617 * try. Need to make sure
618 * kernel will accept it
621 mdu_disk_info_t disc
;
622 /* re-add doesn't work for version-1 superblocks
625 if (array
->major_version
== 1 &&
626 get_linux_version() <= 2006018)
628 disc
.number
= mdi
.disk
.number
;
629 if (ioctl(fd
, GET_DISK_INFO
, &disc
) != 0
630 || disc
.major
!= 0 || disc
.minor
!= 0
633 disc
.major
= major(rdev
);
634 disc
.minor
= minor(rdev
);
635 disc
.number
= mdi
.disk
.number
;
636 disc
.raid_disk
= mdi
.disk
.raid_disk
;
637 disc
.state
= mdi
.disk
.state
;
638 if (dv
->writemostly
== 1)
639 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
640 if (dv
->writemostly
== 2)
641 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
642 remove_partitions(tfd
);
643 if (update
|| dv
->writemostly
> 0) {
645 tfd
= dev_open(dv
->devname
, O_RDWR
);
647 pr_err("failed to open %s for superblock update during re-add\n", dv
->devname
);
651 if (dv
->writemostly
== 1)
652 rv
= dev_st
->ss
->update_super(
653 dev_st
, NULL
, "writemostly",
654 devname
, verbose
, 0, NULL
);
655 if (dv
->writemostly
== 2)
656 rv
= dev_st
->ss
->update_super(
657 dev_st
, NULL
, "readwrite",
658 devname
, verbose
, 0, NULL
);
660 rv
= dev_st
->ss
->update_super(
661 dev_st
, NULL
, update
,
662 devname
, verbose
, 0, NULL
);
664 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
667 pr_err("failed to update superblock during re-add\n");
671 /* don't even try if disk is marked as faulty */
673 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
675 pr_err("re-added %s\n", dv
->devname
);
678 if (errno
== ENOMEM
|| errno
== EROFS
) {
679 pr_err("add new device failed for %s: %s\n",
680 dv
->devname
, strerror(errno
));
681 if (dv
->disposition
== 'M')
690 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
691 struct supertype
*tst
, mdu_array_info_t
*array
,
692 int force
, int verbose
, char *devname
,
693 char *update
, unsigned long rdev
, unsigned long long array_size
,
696 unsigned long long ldsize
;
697 struct supertype
*dev_st
= NULL
;
699 mdu_disk_info_t disc
;
701 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
702 if (dv
->disposition
== 'M')
708 if (tst
->ss
== &super0
&& ldsize
> 4ULL*1024*1024*1024*1024) {
709 /* More than 4TB is wasted on v0.90 */
711 pr_err("%s is larger than %s can effectively use.\n"
712 " Add --force is you really want to add this device.\n",
713 dv
->devname
, devname
);
716 pr_err("%s is larger than %s can effectively use.\n"
717 " Adding anyway as --force was given.\n",
718 dv
->devname
, devname
);
720 if (!tst
->ss
->external
&&
721 array
->major_version
== 0 &&
722 md_get_version(fd
)%100 < 2) {
723 if (ioctl(fd
, HOT_ADD_DISK
, rdev
)==0) {
725 pr_err("hot added %s\n",
730 pr_err("hot add failed for %s: %s\n",
731 dv
->devname
, strerror(errno
));
735 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
737 /* need to find a sample superblock to copy, and
738 * a spare slot to use.
739 * For 'external' array (well, container based),
740 * We can just load the metadata for the array->
744 /* already loaded */;
745 else if (tst
->ss
->external
) {
746 tst
->ss
->load_container(tst
, fd
, NULL
);
747 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
751 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
753 if (disc
.major
==0 && disc
.minor
==0)
755 if ((disc
.state
& 4)==0) /* sync */
757 /* Looks like a good device to try */
758 dev
= map_dev(disc
.major
, disc
.minor
, 1);
761 dfd
= dev_open(dev
, O_RDONLY
);
764 if (tst
->ss
->load_super(tst
, dfd
,
772 /* FIXME this is a bad test to be using */
773 if (!tst
->sb
&& (dv
->disposition
!= 'a'
774 && dv
->disposition
!= 'S')) {
775 /* we are re-adding a device to a
776 * completely dead array - have to depend
779 } else if (!tst
->sb
) {
780 pr_err("cannot load array metadata from %s\n", devname
);
784 /* Make sure device is large enough */
786 tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
788 if (dv
->disposition
== 'M')
790 pr_err("%s not large enough to join array\n",
795 /* Possibly this device was recently part of
796 * the array and was temporarily removed, and
797 * is now being re-added. If so, we can
801 if (array
->not_persistent
==0) {
802 dev_st
= dup_super(tst
);
803 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
805 if (dev_st
&& dev_st
->sb
&& dv
->disposition
!= 'S') {
806 int rv
= attempt_re_add(fd
, tfd
, dv
,
812 dev_st
->ss
->free_super(dev_st
);
816 if (dv
->disposition
== 'M') {
818 pr_err("--re-add for %s to %s is not possible\n",
819 dv
->devname
, devname
);
822 if (dv
->disposition
== 'A') {
823 pr_err("--re-add for %s to %s is not possible\n",
824 dv
->devname
, devname
);
827 if (array
->active_disks
< array
->raid_disks
) {
828 char *avail
= xcalloc(array
->raid_disks
, 1);
832 for (d
= 0; d
< MAX_DISKS
&& found
< array
->nr_disks
; d
++) {
834 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
836 if (disc
.major
== 0 && disc
.minor
== 0)
839 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
841 avail
[disc
.raid_disk
] = 1;
843 array_failed
= !enough(array
->level
, array
->raid_disks
,
844 array
->layout
, 1, avail
);
849 pr_err("%s has failed so using --add cannot work and might destroy\n",
851 pr_err("data on %s. You should stop the array and re-assemble it.\n",
856 /* non-persistent. Must ensure that new drive
857 * is at least array->size big.
859 if (ldsize
/512 < array_size
) {
860 pr_err("%s not large enough to join array\n",
865 /* committed to really trying this device now*/
866 remove_partitions(tfd
);
868 /* in 2.6.17 and earlier, version-1 superblocks won't
869 * use the number we write, but will choose a free number.
870 * we must choose the same free number, which requires
871 * starting at 'raid_disks' and counting up
873 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
875 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
877 if (disc
.major
==0 && disc
.minor
==0)
879 if (disc
.state
& 8) /* removed */
882 disc
.major
= major(rdev
);
883 disc
.minor
= minor(rdev
);
887 disc
.number
= raid_slot
;
889 if (array
->not_persistent
==0) {
891 if (dv
->writemostly
== 1)
892 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
893 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
894 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
895 dv
->devname
, INVALID_SECTORS
))
897 if (tst
->ss
->write_init_super(tst
))
899 } else if (dv
->disposition
== 'A') {
900 /* this had better be raid1.
901 * As we are "--re-add"ing we must find a spare slot
904 char *used
= xcalloc(array
->raid_disks
, 1);
905 for (j
= 0; j
< tst
->max_devs
; j
++) {
906 mdu_disk_info_t disc2
;
908 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
910 if (disc2
.major
==0 && disc2
.minor
==0)
912 if (disc2
.state
& 8) /* removed */
914 if (disc2
.raid_disk
< 0)
916 if (disc2
.raid_disk
> array
->raid_disks
)
918 used
[disc2
.raid_disk
] = 1;
920 for (j
= 0 ; j
< array
->raid_disks
; j
++)
923 disc
.state
|= (1<<MD_DISK_SYNC
);
929 if (array
->state
& (1 << MD_SB_CLUSTERED
)) {
930 if (dv
->disposition
== 'c')
931 disc
.state
|= (1 << MD_DISK_CANDIDATE
);
933 disc
.state
|= (1 << MD_DISK_CLUSTER_ADD
);
936 if (dv
->writemostly
== 1)
937 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
938 if (tst
->ss
->external
) {
940 * to an external metadata container */
941 struct mdinfo new_mdi
;
947 strcpy(devnm
, fd2devnm(fd
));
949 container_fd
= open_dev_excl(devnm
);
950 if (container_fd
< 0) {
951 pr_err("add failed for %s: could not get exclusive access to container\n",
953 tst
->ss
->free_super(tst
);
957 Kill(dv
->devname
, NULL
, 0, -1, 0);
958 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
959 if (mdmon_running(tst
->container_devnm
))
960 tst
->update_tail
= &tst
->updates
;
961 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
962 dv
->devname
, INVALID_SECTORS
)) {
967 if (tst
->update_tail
)
968 flush_metadata_updates(tst
);
970 tst
->ss
->sync_metadata(tst
);
972 sra
= sysfs_read(container_fd
, NULL
, 0);
974 pr_err("add failed for %s: sysfs_read failed\n",
977 tst
->ss
->free_super(tst
);
980 sra
->array
.level
= LEVEL_CONTAINER
;
981 /* Need to set data_offset and component_size */
982 tst
->ss
->getinfo_super(tst
, &new_mdi
, NULL
);
983 new_mdi
.disk
.major
= disc
.major
;
984 new_mdi
.disk
.minor
= disc
.minor
;
985 new_mdi
.recovery_start
= 0;
986 /* Make sure fds are closed as they are O_EXCL which
987 * would block add_disk */
988 tst
->ss
->free_super(tst
);
989 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
990 pr_err("add new device to external metadata failed for %s\n", dv
->devname
);
999 tst
->ss
->free_super(tst
);
1000 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1001 pr_err("add new device failed for %s as %d: %s\n",
1002 dv
->devname
, j
, strerror(errno
));
1007 pr_err("added %s\n", dv
->devname
);
1011 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1012 int sysfd
, unsigned long rdev
, int verbose
, char *devname
)
1017 if (tst
->ss
->external
) {
1018 /* To remove a device from a container, we must
1019 * check that it isn't in use in an array.
1020 * This involves looking in the 'holders'
1021 * directory - there must be just one entry,
1023 * To ensure that it doesn't get used as a
1024 * hot spare while we are checking, we
1025 * get an O_EXCL open on the container
1029 strcpy(devnm
, fd2devnm(fd
));
1030 lfd
= open_dev_excl(devnm
);
1032 pr_err("Cannot get exclusive access to container - odd\n");
1035 /* We may not be able to check on holders in
1036 * sysfs, either because we don't have the dev num
1037 * (rdev == 0) or because the device has been detached
1038 * and the 'holders' directory no longer exists
1039 * (ret == -1). In that case, assume it is OK to
1045 ret
= sysfs_unique_holder(devnm
, rdev
);
1047 pr_err("%s is not a member, cannot remove.\n",
1053 pr_err("%s is still in use, cannot remove.\n",
1059 /* FIXME check that it is a current member */
1061 /* device has been removed and we don't know
1062 * the major:minor number
1064 int n
= write(sysfd
, "remove", 6);
1070 err
= ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1071 if (err
&& errno
== ENODEV
) {
1072 /* Old kernels rejected this if no personality
1074 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1075 struct mdinfo
*dv
= NULL
;
1078 for ( ; dv
; dv
=dv
->next
)
1079 if (dv
->disk
.major
== (int)major(rdev
) &&
1080 dv
->disk
.minor
== (int)minor(rdev
))
1083 err
= sysfs_set_str(sra
, dv
,
1092 pr_err("hot remove failed for %s: %s\n", dv
->devname
,
1098 if (tst
->ss
->external
) {
1100 * Before dropping our exclusive open we make an
1101 * attempt at preventing mdmon from seeing an
1102 * 'add' event before reconciling this 'remove'
1105 char *devnm
= fd2devnm(fd
);
1108 pr_err("unable to get container name\n");
1112 ping_manager(devnm
);
1117 pr_err("hot removed %s from %s\n",
1118 dv
->devname
, devname
);
1122 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1123 unsigned long rdev
, int verbose
, char *devname
)
1125 struct mdinfo
*mdi
, *di
;
1126 if (tst
->ss
->external
) {
1127 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1130 /* Need to find the device in sysfs and add 'want_replacement' to the
1133 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1134 if (!mdi
|| !mdi
->devs
) {
1135 pr_err("Cannot find status of %s to enable replacement - strange\n",
1139 for (di
= mdi
->devs
; di
; di
= di
->next
)
1140 if (di
->disk
.major
== (int)major(rdev
) &&
1141 di
->disk
.minor
== (int)minor(rdev
))
1145 if (di
->disk
.raid_disk
< 0) {
1146 pr_err("%s is not active and so cannot be replaced.\n",
1151 rv
= sysfs_set_str(mdi
, di
,
1152 "state", "want_replacement");
1155 pr_err("Failed to request replacement for %s\n",
1160 pr_err("Marked %s (device %d in %s) for replacement\n",
1161 dv
->devname
, di
->disk
.raid_disk
, devname
);
1162 /* If there is a matching 'with', we need to tell it which
1165 while (dv
&& dv
->disposition
!= 'W')
1168 dv
->disposition
= 'w';
1169 dv
->used
= di
->disk
.raid_disk
;
1174 pr_err("%s not found in %s so cannot --replace it\n",
1175 dv
->devname
, devname
);
1179 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1180 unsigned long rdev
, int verbose
, char *devname
)
1182 struct mdinfo
*mdi
, *di
;
1183 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1184 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1185 if (!mdi
|| !mdi
->devs
) {
1186 pr_err("Cannot find status of %s to enable replacement - strange\n",
1190 for (di
= mdi
->devs
; di
; di
= di
->next
)
1191 if (di
->disk
.major
== (int)major(rdev
) &&
1192 di
->disk
.minor
== (int)minor(rdev
))
1196 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1197 pr_err("%s is faulty and cannot be a replacement\n",
1202 if (di
->disk
.raid_disk
>= 0) {
1203 pr_err("%s is active and cannot be a replacement\n",
1208 rv
= sysfs_set_num(mdi
, di
,
1212 pr_err("Failed to set %s as preferred replacement.\n",
1217 pr_err("Marked %s in %s as replacement for device %d\n",
1218 dv
->devname
, devname
, dv
->used
);
1222 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1223 dv
->devname
, devname
);
1227 int Manage_subdevs(char *devname
, int fd
,
1228 struct mddev_dev
*devlist
, int verbose
, int test
,
1229 char *update
, int force
)
1231 /* Do something to each dev.
1233 * 'a' - add the device
1235 * If that fails EINVAL, try ADD_NEW_DISK
1236 * 'S' - add the device as a spare - don't try re-add
1237 * 'A' - re-add the device
1238 * 'r' - remove the device: HOT_REMOVE_DISK
1239 * device can be 'faulty' or 'detached' in which case all
1240 * matching devices are removed.
1241 * 'f' - set the device faulty SET_DISK_FAULTY
1242 * device can be 'detached' in which case any device that
1243 * is inaccessible will be marked faulty.
1244 * 'R' - mark this device as wanting replacement.
1245 * 'W' - this device is added if necessary and activated as
1246 * a replacement for a previous 'R' device.
1248 * 'w' - 'W' will be changed to 'w' when it is paired with
1249 * a 'R' device. If a 'W' is found while walking the list
1250 * it must be unpaired, and is an error.
1251 * 'M' - this is created by a 'missing' target. It is a slight
1253 * 'F' - Another variant of 'A', where the device was faulty
1254 * so must be removed from the array first.
1255 * 'c' - confirm the device as found (for clustered environments)
1257 * For 'f' and 'r', the device can also be a kernel-internal
1258 * name such as 'sdb'.
1260 mdu_array_info_t array
;
1261 unsigned long long array_size
;
1262 struct mddev_dev
*dv
;
1264 struct supertype
*tst
;
1265 char *subarray
= NULL
;
1267 int count
= 0; /* number of actions taken */
1273 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
1274 pr_err("Cannot get array info for %s\n",
1278 sysfs_init(&info
, fd
, NULL
);
1280 /* array.size is only 32 bits and may be truncated.
1281 * So read from sysfs if possible, and record number of sectors
1284 array_size
= get_component_size(fd
);
1285 if (array_size
<= 0)
1286 array_size
= array
.size
* 2;
1288 tst
= super_by_fd(fd
, &subarray
);
1290 pr_err("unsupport array - version %d.%d\n",
1291 array
.major_version
, array
.minor_version
);
1295 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1296 unsigned long rdev
= 0; /* device to add/remove etc */
1301 if (dv
->disposition
== 'c') {
1302 rv
= parse_cluster_confirm_arg(dv
->devname
,
1306 pr_err("Could not get the devname of cluster\n");
1311 if (strcmp(dv
->devname
, "failed") == 0 ||
1312 strcmp(dv
->devname
, "faulty") == 0) {
1313 if (dv
->disposition
!= 'A'
1314 && dv
->disposition
!= 'r') {
1315 pr_err("%s only meaningful with -r or --re-add, not -%c\n",
1316 dv
->devname
, dv
->disposition
);
1319 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1323 if (strcmp(dv
->devname
, "detached") == 0) {
1324 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1325 pr_err("%s only meaningful with -r of -f, not -%c\n",
1326 dv
->devname
, dv
->disposition
);
1329 add_detached(dv
, fd
, dv
->disposition
);
1333 if (strcmp(dv
->devname
, "missing") == 0) {
1334 struct mddev_dev
*add_devlist
= NULL
;
1335 struct mddev_dev
**dp
;
1336 if (dv
->disposition
== 'c') {
1337 rv
= ioctl(fd
, CLUSTERED_DISK_NACK
, NULL
);
1341 if (dv
->disposition
!= 'A') {
1342 pr_err("'missing' only meaningful with --re-add\n");
1345 add_devlist
= conf_get_devs();
1346 if (add_devlist
== NULL
) {
1347 pr_err("no devices to scan for missing members.");
1350 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1351 /* 'M' (for 'missing') is like 'A' without errors */
1352 (*dp
)->disposition
= 'M';
1354 dv
->next
= add_devlist
;
1358 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1359 strlen(dv
->devname
) == 5) {
1362 if (dv
->disposition
!= 'r' &&
1363 dv
->disposition
!= 'f') {
1364 pr_err("'%s' only meaningful with -r or -f\n",
1368 if (array
.level
!= 10) {
1369 pr_err("'%s' only meaningful with RAID10 arrays\n",
1373 copies
= ((array
.layout
& 0xff) *
1374 ((array
.layout
>> 8) & 0xff));
1375 if (array
.raid_disks
% copies
!= 0 ||
1376 dv
->devname
[4] < 'A' ||
1377 dv
->devname
[4] >= 'A' + copies
||
1379 pr_err("'%s' not meaningful with this array\n",
1383 add_set(dv
, fd
, dv
->devname
[4]);
1387 if (strchr(dv
->devname
, '/') == NULL
&&
1388 strchr(dv
->devname
, ':') == NULL
&&
1389 strlen(dv
->devname
) < 50) {
1390 /* Assume this is a kernel-internal name like 'sda1' */
1393 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1394 pr_err("%s only meaningful with -r or -f, not -%c\n",
1395 dv
->devname
, dv
->disposition
);
1399 sprintf(dname
, "dev-%s", dv
->devname
);
1400 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1403 if (sysfs_fd_get_str(sysfd
, dn
, 20) > 0 &&
1404 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1405 rdev
= makedev(mj
,mn
);
1412 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1414 pr_err("%s does not appear to be a component of %s\n",
1415 dv
->devname
, devname
);
1419 } else if ((dv
->disposition
== 'r' || dv
->disposition
== 'f')
1420 && get_maj_min(dv
->devname
, &mj
, &mn
)) {
1421 /* for 'fail' and 'remove', the device might
1424 rdev
= makedev(mj
, mn
);
1427 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1431 int open_err
= errno
;
1432 if (stat(dv
->devname
, &stb
) != 0) {
1433 pr_err("Cannot find %s: %s\n",
1434 dv
->devname
, strerror(errno
));
1437 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
1438 if (dv
->disposition
== 'M')
1439 /* non-fatal. Also improbable */
1441 pr_err("%s is not a block device.\n",
1445 if (dv
->disposition
== 'r')
1446 /* Be happy, the stat worked, that is
1447 * enough for --remove
1451 if (dv
->disposition
== 'M')
1454 pr_err("Cannot open %s: %s\n",
1455 dv
->devname
, strerror(open_err
));
1461 switch(dv
->disposition
){
1463 pr_err("internal error - devmode[%s]=%d\n",
1464 dv
->devname
, dv
->disposition
);
1467 case 'S': /* --add-spare */
1469 case 'M': /* --re-add missing */
1470 case 'F': /* --re-add faulty */
1471 case 'c': /* --cluster-confirm */
1472 /* add the device */
1474 pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n");
1477 if (dv
->disposition
== 'F')
1478 /* Need to remove first */
1479 ioctl(fd
, HOT_REMOVE_DISK
, rdev
);
1480 /* Make sure it isn't in use (in 2.6 or later) */
1481 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1483 /* We know no-one else is using it. We'll
1484 * need non-exclusive access to add it, so
1488 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1491 if (dv
->disposition
== 'M')
1493 pr_err("Cannot open %s: %s\n",
1494 dv
->devname
, strerror(errno
));
1498 if (sysfs_freeze_array(&info
) == 1)
1503 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1504 force
, verbose
, devname
, update
,
1505 rdev
, array_size
, raid_slot
);
1517 pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n");
1520 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1532 case 'f': /* set faulty */
1533 /* FIXME check current member */
1534 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1535 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1539 pr_err("set device faulty failed for %s: %s\n",
1540 dv
->devname
, strerror(errno
));
1550 pr_err("set %s faulty in %s\n",
1551 dv
->devname
, devname
);
1553 case 'R': /* Mark as replaceable */
1555 pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n");
1559 if (sysfs_freeze_array(&info
) == 1)
1564 rv
= Manage_replace(tst
, fd
, dv
,
1573 case 'W': /* --with device that doesn't match */
1574 pr_err("No matching --replace device for --with %s\n",
1577 case 'w': /* --with device which was matched */
1578 rv
= Manage_with(tst
, fd
, dv
,
1579 rdev
, verbose
, devname
);
1586 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1587 if (test
&& count
== 0)
1593 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1594 return !test
&& busy
? 2 : 1;
1597 int autodetect(void)
1599 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1601 int fd
= dev_open("9:0", O_RDONLY
);
1603 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1610 int Update_subarray(char *dev
, char *subarray
, char *update
, struct mddev_ident
*ident
, int verbose
)
1612 struct supertype supertype
, *st
= &supertype
;
1615 memset(st
, 0, sizeof(*st
));
1617 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1621 if (!st
->ss
->update_subarray
) {
1623 pr_err("Operation not supported for %s metadata\n",
1628 if (mdmon_running(st
->devnm
))
1629 st
->update_tail
= &st
->updates
;
1631 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1635 pr_err("Failed to update %s of subarray-%s in %s\n",
1636 update
, subarray
, dev
);
1637 } else if (st
->update_tail
)
1638 flush_metadata_updates(st
);
1640 st
->ss
->sync_metadata(st
);
1642 if (rv
== 0 && strcmp(update
, "name") == 0 && verbose
>= 0)
1643 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1647 st
->ss
->free_super(st
);
1653 /* Move spare from one array to another If adding to destination array fails
1654 * add back to original array.
1655 * Returns 1 on success, 0 on failure */
1656 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1658 struct mddev_dev devlist
;
1661 /* try to remove and add */
1662 int fd1
= open(to_devname
, O_RDONLY
);
1663 int fd2
= open(from_devname
, O_RDONLY
);
1665 if (fd1
< 0 || fd2
< 0) {
1666 if (fd1
>=0) close(fd1
);
1667 if (fd2
>=0) close(fd2
);
1671 devlist
.next
= NULL
;
1673 devlist
.writemostly
= 0;
1674 devlist
.devname
= devname
;
1675 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1677 devlist
.disposition
= 'r';
1678 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0) == 0) {
1679 devlist
.disposition
= 'a';
1680 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0, NULL
, 0) == 0) {
1681 /* make sure manager is aware of changes */
1682 ping_manager(to_devname
);
1683 ping_manager(from_devname
);
1688 else Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, NULL
, 0);