]>
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>
31 int Manage_ro(char *devname
, int fd
, int readonly
)
33 /* switch to readonly or rw
36 * first check that array is runing
37 * use RESTART_ARRAY_RW or STOP_ARRAY_RO
43 /* If this is an externally-managed array, we need to modify the
44 * metadata_version so that mdmon doesn't undo our change.
46 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_VERSION
);
48 mdi
->array
.major_version
== -1 &&
49 is_subarray(mdi
->text_version
)) {
51 strcpy(vers
, "external:");
52 strcat(vers
, mdi
->text_version
);
55 /* We set readonly ourselves. */
57 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
60 rv
= sysfs_set_str(mdi
, NULL
, "array_state", "readonly");
63 pr_err("failed to set readonly for %s: %s\n",
64 devname
, strerror(errno
));
66 vers
[9] = mdi
->text_version
[0];
67 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
73 /* We cannot set read/write - must signal mdmon */
75 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
77 cp
= strchr(vers
+10, '/');
80 ping_monitor(vers
+10);
81 if (mdi
->array
.level
<= 0)
82 sysfs_set_str(mdi
, NULL
, "array_state", "active");
87 if (!md_array_active(fd
)) {
88 pr_err("%s does not appear to be active.\n", devname
);
94 if (ioctl(fd
, STOP_ARRAY_RO
, NULL
)) {
95 pr_err("failed to set readonly for %s: %s\n",
96 devname
, strerror(errno
));
100 } else if (readonly
< 0) {
101 if (ioctl(fd
, RESTART_ARRAY_RW
, NULL
)) {
102 pr_err("failed to set writable for %s: %s\n",
103 devname
, strerror(errno
));
113 static void remove_devices(char *devnm
, char *path
)
116 * Remove names at 'path' - possibly with
117 * partition suffixes - which link to the 'standard'
118 * name for devnm. These were probably created
119 * by mdadm when the array was assembled.
132 sprintf(base
, "/dev/%s", devnm
);
133 be
= base
+ strlen(base
);
135 path2
= xmalloc(strlen(path
)+20);
137 pe
= path2
+ strlen(path2
);
139 for (part
= 0; part
< 16; part
++) {
141 sprintf(be
, "p%d", part
);
144 sprintf(pe
, "p%d", part
);
146 sprintf(pe
, "%d", part
);
148 n
= readlink(path2
, link
, sizeof(link
));
149 if (n
> 0 && (int)strlen(base
) == n
&&
150 strncmp(link
, base
, n
) == 0)
156 int Manage_run(char *devname
, int fd
, struct context
*c
)
158 /* Run the array. Array must already be configured
165 pr_err("Cannot find %s in sysfs!!\n", devname
);
169 return IncrementalScan(c
, nm
);
172 int Manage_stop(char *devname
, int fd
, int verbose
, int will_retry
)
174 /* Stop the array. Array must already be configured
175 * 'will_retry' means that error messages are not wanted.
178 struct map_ent
*map
= NULL
;
181 char container
[MD_NAME_MAX
] = {0};
184 char buf
[SYSFS_MAX_BUF_SIZE
];
185 unsigned long long rd1
, rd2
;
187 if (will_retry
&& verbose
== 0)
190 strcpy(devnm
, fd2devnm(fd
));
191 /* Get EXCL access first. If this fails, then attempting
192 * to stop is probably a bad idea.
194 mdi
= sysfs_read(fd
, NULL
, GET_LEVEL
|GET_COMPONENT
|GET_VERSION
);
195 if (mdi
&& is_subarray(mdi
->text_version
))
196 sysfs_get_container_devnm(mdi
, container
);
200 while (((fd
= ((devname
[0] == '/')
201 ?open(devname
, O_RDONLY
|O_EXCL
)
202 :open_dev_flags(devnm
, O_RDONLY
|O_EXCL
))) < 0 ||
203 strcmp(fd2devnm(fd
), devnm
) != 0) && container
[0] &&
204 mdmon_running(container
) && count
) {
205 /* Can't open, so something might be wrong. However it
206 * is a container, so we might be racing with mdmon, so
211 flush_mdmon(container
);
214 if (fd
< 0 || strcmp(fd2devnm(fd
), devnm
) != 0) {
218 pr_err("Cannot get exclusive access to %s:Perhaps a running process, mounted filesystem or active volume group?\n",
223 /* If this is an mdmon managed array, just write 'inactive'
224 * to the array state and let mdmon clear up.
227 mdi
->array
.level
> 0 &&
228 is_subarray(mdi
->text_version
)) {
230 /* This is mdmon managed. */
233 /* As we had an O_EXCL open, any use of the device
234 * which blocks STOP_ARRAY is probably a transient use,
235 * so it is reasonable to retry for a while - 5 seconds.
239 (err
= sysfs_set_str(mdi
, NULL
,
243 sleep_for(0, MSEC_TO_NSEC(200), true);
248 pr_err("failed to stop array %s: %s\n",
249 devname
, strerror(errno
));
254 /* Give monitor a chance to act */
255 ping_monitor(mdi
->text_version
);
257 fd
= open_dev_excl(devnm
);
260 pr_err("failed to completely stop %s: Device is busy\n",
266 mdi
->array
.major_version
== -1 &&
267 mdi
->array
.minor_version
== -2 &&
268 !is_subarray(mdi
->text_version
)) {
269 struct mdstat_ent
*mds
, *m
;
270 /* container, possibly mdmon-managed.
271 * Make sure mdmon isn't opening it, which
272 * would interfere with the 'stop'
274 ping_monitor(mdi
->sys_name
);
276 /* now check that there are no existing arrays
277 * which are members of this array
279 mds
= mdstat_read(0, 0);
280 for (m
= mds
; m
; m
= m
->next
)
281 if (m
->metadata_version
&&
282 strncmp(m
->metadata_version
, "external:", 9)==0 &&
283 metadata_container_matches(m
->metadata_version
+9,
286 pr_err("Cannot stop container %s: member %s still active\n",
294 /* If the array is undergoing a reshape which changes the number
295 * of devices, then it would be nice to stop it at a point where
296 * it has completed a full number of stripes in both old and
297 * new layouts as this will allow the reshape to be reverted.
298 * So if 'sync_action' is "reshape" and 'raid_disks' shows two
299 * different numbers, then
301 * - set sync_max to next multiple of both data_disks and
302 * chunk sizes (or next but one)
304 * - wait on 'sync_completed' for that point to be reached.
306 if (mdi
&& is_level456(mdi
->array
.level
) &&
307 sysfs_attribute_available(mdi
, NULL
, "sync_action") &&
308 sysfs_attribute_available(mdi
, NULL
, "reshape_direction") &&
309 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
)) > 0 &&
310 strcmp(buf
, "reshape\n") == 0 &&
311 sysfs_get_two(mdi
, NULL
, "raid_disks", &rd1
, &rd2
) == 2) {
312 unsigned long long position
, curr
;
313 unsigned long long chunk1
, chunk2
;
314 unsigned long long rddiv
, chunkdiv
;
315 unsigned long long sectors
;
316 unsigned long long sync_max
, old_sync_max
;
317 unsigned long long completed
;
323 while (rd1
> rd2
&& delay
> 0 &&
324 sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) == 0) {
325 /* must be in the critical section - wait a bit */
327 sleep_for(0, MSEC_TO_NSEC(100), true);
330 if (sysfs_set_str(mdi
, NULL
, "sync_action", "frozen") != 0)
332 /* Array is frozen */
334 rd1
-= mdi
->array
.level
== 6 ? 2 : 1;
335 rd2
-= mdi
->array
.level
== 6 ? 2 : 1;
336 sysfs_get_str(mdi
, NULL
, "reshape_direction", buf
, sizeof(buf
));
337 if (strncmp(buf
, "back", 4) == 0)
339 if (sysfs_get_ll(mdi
, NULL
, "reshape_position", &position
) != 0) {
340 /* reshape must have finished now */
341 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
344 sysfs_get_two(mdi
, NULL
, "chunk_size", &chunk1
, &chunk2
);
347 rddiv
= GCD(rd1
, rd2
);
348 chunkdiv
= GCD(chunk1
, chunk2
);
349 sectors
= (chunk1
/chunkdiv
) * chunk2
* (rd1
/rddiv
) * rd2
;
352 /* Need to subtract 'reshape_position' from
353 * array size to get equivalent of sync_max.
354 * Size calculation based on raid5_size in kernel.
356 unsigned long long size
= mdi
->component_size
;
359 /* rd1 must be smaller */
360 /* Reshape may have progressed further backwards than
361 * recorded, so target even further back (hence "-1")
363 position
= (position
/ sectors
- 1) * sectors
;
364 /* rd1 is always the conversion factor between 'sync'
365 * position and 'reshape' position.
366 * We read 1 "new" stripe worth of data from where-ever,
367 * and when write out that full stripe.
369 sync_max
= size
- position
/rd1
;
371 /* Reshape will very likely be beyond position, and it may
372 * be too late to stop at '+1', so aim for '+2'
374 position
= (position
/ sectors
+ 2) * sectors
;
375 sync_max
= position
/rd1
;
377 if (sysfs_get_ll(mdi
, NULL
, "sync_max", &old_sync_max
) < 0)
378 old_sync_max
= mdi
->component_size
;
379 /* Must not advance sync_max as that could confuse
380 * the reshape monitor */
381 if (sync_max
< old_sync_max
)
382 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
383 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
385 /* That should have set things going again. Now we
386 * wait a little while (3 second max) for sync_completed
387 * to reach the target.
388 * The reshape process can block for 500msec if
389 * the sync speed limit is hit, so we need to wait
390 * a lot longer than that. 1 second is usually
394 scfd
= sysfs_open(mdi
->sys_name
, NULL
, "sync_completed");
395 while (scfd
>= 0 && delay
> 0 && old_sync_max
> 0) {
396 unsigned long long max_completed
;
397 sysfs_get_ll(mdi
, NULL
, "reshape_position", &curr
);
398 sysfs_fd_get_str(scfd
, buf
, sizeof(buf
));
399 if (str_is_none(buf
) == true) {
400 /* Either reshape has aborted, or hasn't
401 * quite started yet. Wait a bit and
402 * check 'sync_action' to see.
404 sleep_for(0, MSEC_TO_NSEC(10), true);
405 sysfs_get_str(mdi
, NULL
, "sync_action", buf
, sizeof(buf
));
406 if (strncmp(buf
, "reshape", 7) != 0)
410 if (sysfs_fd_get_two(scfd
, &completed
,
411 &max_completed
) == 2 &&
412 /* 'completed' sometimes reads as max-uulong */
413 completed
< max_completed
&&
414 (completed
> sync_max
||
415 (completed
== sync_max
&& curr
!= position
))) {
416 while (completed
> sync_max
) {
417 sync_max
+= sectors
/ rd1
;
423 if (sync_max
< old_sync_max
)
424 sysfs_set_num(mdi
, NULL
, "sync_max", sync_max
);
427 if (!backwards
&& curr
>= position
)
429 if (backwards
&& curr
<= position
)
431 sysfs_wait(scfd
, &delay
);
439 /* As we have an O_EXCL open, any use of the device
440 * which blocks STOP_ARRAY is probably a transient use,
441 * so it is reasonable to retry for a while - 5 seconds.
444 while (count
&& fd
>= 0 &&
445 (err
= ioctl(fd
, STOP_ARRAY
, NULL
)) < 0 && errno
== EBUSY
) {
446 sleep_for(0, MSEC_TO_NSEC(200), true);
449 if (fd
>= 0 && err
) {
451 pr_err("failed to stop array %s: %s\n",
452 devname
, strerror(errno
));
454 cont_err("Perhaps a running process, mounted filesystem or active volume group?\n");
460 if (devnm
[0] && udev_is_available()) {
461 struct map_ent
*mp
= map_by_devnm(&map
, devnm
);
462 remove_devices(devnm
, mp
? mp
->path
: NULL
);
466 pr_err("stopped %s\n", devname
);
468 map_remove(&map
, devnm
);
476 static struct mddev_dev
*add_one(struct mddev_dev
*dv
, char *name
, char disp
)
478 struct mddev_dev
*new;
479 new = xmalloc(sizeof(*new));
480 memset(new, 0, sizeof(*new));
481 new->devname
= xstrdup(name
);
482 new->disposition
= disp
;
483 new->next
= dv
->next
;
488 static void add_faulty(struct mddev_dev
*dv
, int fd
, char disp
)
490 mdu_array_info_t array
;
491 mdu_disk_info_t disk
;
495 if (md_get_array_info(fd
, &array
) != 0)
498 remaining_disks
= array
.nr_disks
;
499 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
502 if (md_get_disk_info(fd
, &disk
) != 0)
504 if (disk
.major
== 0 && disk
.minor
== 0)
507 if ((disk
.state
& 1) == 0) /* not faulty */
509 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
510 dv
= add_one(dv
, buf
, disp
);
514 static void add_detached(struct mddev_dev
*dv
, int fd
, char disp
)
516 mdu_array_info_t array
;
517 mdu_disk_info_t disk
;
521 if (md_get_array_info(fd
, &array
) != 0)
524 remaining_disks
= array
.nr_disks
;
525 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
529 if (md_get_disk_info(fd
, &disk
) != 0)
531 if (disk
.major
== 0 && disk
.minor
== 0)
534 if (disp
== 'f' && (disk
.state
& 1) != 0) /* already faulty */
536 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
537 sfd
= dev_open(buf
, O_RDONLY
);
544 /* Probably not detached */
546 dv
= add_one(dv
, buf
, disp
);
550 static void add_set(struct mddev_dev
*dv
, int fd
, char set_char
)
552 mdu_array_info_t array
;
553 mdu_disk_info_t disk
;
558 if (md_get_array_info(fd
, &array
) != 0)
560 if (array
.level
!= 10)
562 copies
= ((array
.layout
& 0xff) *
563 ((array
.layout
>> 8) & 0xff));
564 if (array
.raid_disks
% copies
)
567 remaining_disks
= array
.nr_disks
;
568 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
571 if (md_get_disk_info(fd
, &disk
) != 0)
573 if (disk
.major
== 0 && disk
.minor
== 0)
576 set
= disk
.raid_disk
% copies
;
577 if (set_char
!= set
+ 'A')
579 sprintf(buf
, "%d:%d", disk
.major
, disk
.minor
);
580 dv
= add_one(dv
, buf
, dv
->disposition
);
584 int attempt_re_add(int fd
, int tfd
, struct mddev_dev
*dv
,
585 struct supertype
*dev_st
, struct supertype
*tst
,
586 unsigned long rdev
, enum update_opt update
,
587 char *devname
, int verbose
, mdu_array_info_t
*array
)
593 dev_st
->ss
->getinfo_super(dev_st
, &mdi
, NULL
);
594 dev_st
->ss
->uuid_from_super(dev_st
, ouuid
);
596 tst
->ss
->uuid_from_super(tst
, duuid
);
598 /* Assume uuid matches: kernel will check */
599 memcpy(duuid
, ouuid
, sizeof(ouuid
));
600 if ((mdi
.disk
.state
& (1<<MD_DISK_ACTIVE
)) &&
601 !(mdi
.disk
.state
& (1<<MD_DISK_FAULTY
)) &&
602 memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
603 /* Looks like it is worth a
604 * try. Need to make sure
605 * kernel will accept it
608 mdu_disk_info_t disc
;
609 disc
.number
= mdi
.disk
.number
;
610 if (md_get_disk_info(fd
, &disc
) != 0 ||
611 disc
.major
!= 0 || disc
.minor
!= 0)
613 disc
.major
= major(rdev
);
614 disc
.minor
= minor(rdev
);
615 disc
.number
= mdi
.disk
.number
;
616 disc
.raid_disk
= mdi
.disk
.raid_disk
;
617 disc
.state
= mdi
.disk
.state
;
618 if (array
->state
& (1 << MD_SB_CLUSTERED
)) {
619 /* extra flags are needed when adding to a cluster as
620 * there are two cases to distinguish
622 if (dv
->disposition
== 'c')
623 disc
.state
|= (1 << MD_DISK_CANDIDATE
);
625 disc
.state
|= (1 << MD_DISK_CLUSTER_ADD
);
627 if (dv
->writemostly
== FlagSet
)
628 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
629 if (dv
->writemostly
== FlagClear
)
630 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
631 if (dv
->failfast
== FlagSet
)
632 disc
.state
|= 1 << MD_DISK_FAILFAST
;
633 if (dv
->failfast
== FlagClear
)
634 disc
.state
&= ~(1 << MD_DISK_FAILFAST
);
635 remove_partitions(tfd
);
636 if (update
|| dv
->writemostly
!= FlagDefault
||
637 dv
->failfast
!= FlagDefault
) {
639 tfd
= dev_open(dv
->devname
, O_RDWR
);
641 pr_err("failed to open %s for superblock update during re-add\n", dv
->devname
);
645 if (dv
->writemostly
== FlagSet
)
646 rv
= dev_st
->ss
->update_super(
647 dev_st
, NULL
, UOPT_SPEC_WRITEMOSTLY
,
648 devname
, verbose
, 0, NULL
);
649 if (dv
->writemostly
== FlagClear
)
650 rv
= dev_st
->ss
->update_super(
651 dev_st
, NULL
, UOPT_SPEC_READWRITE
,
652 devname
, verbose
, 0, NULL
);
653 if (dv
->failfast
== FlagSet
)
654 rv
= dev_st
->ss
->update_super(
655 dev_st
, NULL
, UOPT_SPEC_FAILFAST
,
656 devname
, verbose
, 0, NULL
);
657 if (dv
->failfast
== FlagClear
)
658 rv
= dev_st
->ss
->update_super(
659 dev_st
, NULL
, UOPT_SPEC_NOFAILFAST
,
660 devname
, verbose
, 0, NULL
);
662 rv
= dev_st
->ss
->update_super(
663 dev_st
, NULL
, update
,
664 devname
, verbose
, 0, NULL
);
666 rv
= dev_st
->ss
->store_super(dev_st
, tfd
);
669 pr_err("failed to update superblock during re-add\n");
673 /* don't even try if disk is marked as faulty */
675 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
677 pr_err("re-added %s\n", dv
->devname
);
680 if (errno
== ENOMEM
|| errno
== EROFS
) {
681 pr_err("add new device failed for %s: %s\n",
682 dv
->devname
, strerror(errno
));
683 if (dv
->disposition
== 'M')
693 * manage_add_external() - Add disk to external container.
694 * @st: external supertype pointer, must not be NULL, superblock is released here.
695 * @fd: container file descriptor, must not have O_EXCL mode.
696 * @disk_fd: device to add file descriptor.
697 * @disk_name: name of the device to add.
700 * Superblock is released here because any open fd with O_EXCL will block sysfs_add_disk().
702 mdadm_status_t
manage_add_external(struct supertype
*st
, int fd
, char *disk_name
,
703 mdu_disk_info_t
*disc
)
705 mdadm_status_t rv
= MDADM_STATUS_ERROR
;
706 char container_devpath
[MD_NAME_MAX
];
707 struct dev_policy
*pols
= NULL
;
708 struct mdinfo new_mdi
;
709 struct mdinfo
*sra
= NULL
;
713 snprintf(container_devpath
, MD_NAME_MAX
, "%s", fd2devnm(fd
));
715 container_fd
= open_dev_excl(container_devpath
);
716 if (!is_fd_valid(container_fd
)) {
717 pr_err("Failed to get exclusive access to container %s\n", container_devpath
);
718 return MDADM_STATUS_ERROR
;
721 /* Check if metadata handler is able to accept the drive */
722 if (!st
->ss
->validate_geometry(st
, LEVEL_CONTAINER
, 0, 1, NULL
, 0, 0, disk_name
, NULL
,
726 if (mddev_test_and_add_drive_policies(st
, &pols
, container_fd
, 1))
729 Kill(disk_name
, NULL
, 0, -1, 0);
731 disk_fd
= dev_open(disk_name
, O_RDWR
| O_EXCL
| O_DIRECT
);
732 if (!is_fd_valid(disk_fd
)) {
733 pr_err("Failed to exclusively open %s\n", disk_name
);
737 if (drive_test_and_add_policies(st
, &pols
, disk_fd
, 1))
740 if (st
->ss
->add_to_super(st
, disc
, disk_fd
, disk_name
, INVALID_SECTORS
))
743 if (!mdmon_running(st
->container_devnm
))
744 st
->ss
->sync_metadata(st
);
746 sra
= sysfs_read(container_fd
, NULL
, 0);
748 pr_err("Failed to read sysfs for %s\n", disk_name
);
752 sra
->array
.level
= LEVEL_CONTAINER
;
753 /* Need to set data_offset and component_size */
754 st
->ss
->getinfo_super(st
, &new_mdi
, NULL
);
755 new_mdi
.disk
.major
= disc
->major
;
756 new_mdi
.disk
.minor
= disc
->minor
;
757 new_mdi
.recovery_start
= 0;
759 st
->ss
->free_super(st
);
761 if (sysfs_add_disk(sra
, &new_mdi
, 0) != 0) {
762 pr_err("Failed to add %s to container %s\n", disk_name
, container_devpath
);
765 ping_monitor(container_devpath
);
766 rv
= MDADM_STATUS_SUCCESS
;
770 dev_policy_free(pols
);
775 if (rv
!= MDADM_STATUS_SUCCESS
&& is_fd_valid(disk_fd
))
776 /* Metadata handler records this descriptor, so release it only on failure. */
780 st
->ss
->free_super(st
);
785 int Manage_add(int fd
, int tfd
, struct mddev_dev
*dv
,
786 struct supertype
*tst
, mdu_array_info_t
*array
,
787 int force
, int verbose
, char *devname
,
788 enum update_opt update
, unsigned long rdev
,
789 unsigned long long array_size
, int raid_slot
)
791 unsigned long long ldsize
;
792 struct supertype
*dev_st
;
794 mdu_disk_info_t disc
;
795 struct map_ent
*map
= NULL
;
797 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
798 if (dv
->disposition
== 'M')
804 if (tst
->ss
== &super0
&& ldsize
> 4ULL*1024*1024*1024*1024) {
805 /* More than 4TB is wasted on v0.90 */
807 pr_err("%s is larger than %s can effectively use.\n"
808 " Add --force is you really want to add this device.\n",
809 dv
->devname
, devname
);
812 pr_err("%s is larger than %s can effectively use.\n"
813 " Adding anyway as --force was given.\n",
814 dv
->devname
, devname
);
817 if (array
->not_persistent
== 0 || tst
->ss
->external
) {
819 /* need to find a sample superblock to copy, and
820 * a spare slot to use.
821 * For 'external' array (well, container based),
822 * We can just load the metadata for the array->
826 /* already loaded */;
827 else if (tst
->ss
->external
) {
828 tst
->ss
->load_container(tst
, fd
, NULL
);
829 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
833 if (md_get_disk_info(fd
, &disc
))
835 if (disc
.major
==0 && disc
.minor
==0)
837 if ((disc
.state
& 4)==0) /* sync */
839 /* Looks like a good device to try */
840 dev
= map_dev(disc
.major
, disc
.minor
, 1);
843 dfd
= dev_open(dev
, O_RDONLY
);
846 if (tst
->ss
->load_super(tst
, dfd
,
854 /* FIXME this is a bad test to be using */
855 if (!tst
->sb
&& (dv
->disposition
!= 'a' &&
856 dv
->disposition
!= 'S')) {
857 /* we are re-adding a device to a
858 * completely dead array - have to depend
861 } else if (!tst
->sb
) {
862 pr_err("cannot load array metadata from %s\n", devname
);
866 /* Make sure device is large enough */
867 if (dv
->disposition
!= 'j' && /* skip size check for Journal */
869 tst
->ss
->avail_size(tst
, ldsize
/512, INVALID_SECTORS
) <
871 if (dv
->disposition
== 'M')
873 pr_err("%s not large enough to join array\n",
878 /* Possibly this device was recently part of
879 * the array and was temporarily removed, and
880 * is now being re-added. If so, we can
884 if (array
->not_persistent
== 0 && dv
->disposition
!= 'S') {
887 dev_st
= dup_super(tst
);
888 dev_st
->ss
->load_super(dev_st
, tfd
, NULL
);
891 rv
= attempt_re_add(fd
, tfd
, dv
, dev_st
, tst
, rdev
, update
,
892 devname
, verbose
, array
);
894 dev_st
->ss
->free_super(dev_st
);
902 if (dv
->disposition
== 'M') {
904 pr_err("--re-add for %s to %s is not possible\n",
905 dv
->devname
, devname
);
908 if (dv
->disposition
== 'A') {
909 pr_err("--re-add for %s to %s is not possible\n",
910 dv
->devname
, devname
);
913 if (array
->active_disks
< array
->raid_disks
) {
914 char *avail
= xcalloc(array
->raid_disks
, 1);
918 for (d
= 0; d
< MAX_DISKS
&& found
< array
->nr_disks
; d
++) {
920 if (md_get_disk_info(fd
, &disc
))
922 if (disc
.major
== 0 && disc
.minor
== 0)
924 if (!(disc
.state
& (1<<MD_DISK_SYNC
)))
926 avail
[disc
.raid_disk
] = 1;
929 array_failed
= !enough(array
->level
, array
->raid_disks
,
930 array
->layout
, 1, avail
);
935 pr_err("%s has failed so using --add cannot work and might destroy\n",
937 pr_err("data on %s. You should stop the array and re-assemble it.\n",
942 /* non-persistent. Must ensure that new drive
943 * is at least array->size big.
945 if (ldsize
/512 < array_size
) {
946 pr_err("%s not large enough to join array\n",
951 /* committed to really trying this device now*/
952 remove_partitions(tfd
);
954 /* in 2.6.17 and earlier, version-1 superblocks won't
955 * use the number we write, but will choose a free number.
956 * we must choose the same free number, which requires
957 * starting at 'raid_disks' and counting up
959 for (j
= array
->raid_disks
; j
< tst
->max_devs
; j
++) {
961 if (md_get_disk_info(fd
, &disc
))
963 if (disc
.major
==0 && disc
.minor
==0)
965 if (disc
.state
& 8) /* removed */
968 disc
.major
= major(rdev
);
969 disc
.minor
= minor(rdev
);
973 disc
.number
= raid_slot
;
976 /* only add journal to array that supports journaling */
977 if (dv
->disposition
== 'j') {
980 mdp
= sysfs_read(fd
, NULL
, GET_ARRAY_STATE
);
982 pr_err("%s unable to read array state.\n", devname
);
986 if (mdp
->array_state
!= ARRAY_READONLY
) {
988 pr_err("%s is not readonly, cannot add journal.\n", devname
);
998 pr_err("failed to get exclusive lock on mapfile when add disk\n");
1000 if (array
->not_persistent
==0) {
1002 if (dv
->disposition
== 'j')
1003 disc
.state
|= (1 << MD_DISK_JOURNAL
) | (1 << MD_DISK_SYNC
);
1004 if (dv
->writemostly
== FlagSet
)
1005 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
1006 if (dv
->failfast
== FlagSet
)
1007 disc
.state
|= 1 << MD_DISK_FAILFAST
;
1008 dfd
= dev_open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
1009 if (tst
->ss
->add_to_super(tst
, &disc
, dfd
,
1010 dv
->devname
, INVALID_SECTORS
))
1012 if (tst
->ss
->write_init_super(tst
))
1014 } else if (dv
->disposition
== 'A') {
1015 /* this had better be raid1.
1016 * As we are "--re-add"ing we must find a spare slot
1019 char *used
= xcalloc(array
->raid_disks
, 1);
1020 for (j
= 0; j
< tst
->max_devs
; j
++) {
1021 mdu_disk_info_t disc2
;
1023 if (md_get_disk_info(fd
, &disc2
))
1025 if (disc2
.major
==0 && disc2
.minor
==0)
1027 if (disc2
.state
& 8) /* removed */
1029 if (disc2
.raid_disk
< 0)
1031 if (disc2
.raid_disk
> array
->raid_disks
)
1033 used
[disc2
.raid_disk
] = 1;
1035 for (j
= 0 ; j
< array
->raid_disks
; j
++)
1038 disc
.state
|= (1<<MD_DISK_SYNC
);
1044 if (array
->state
& (1 << MD_SB_CLUSTERED
)) {
1045 if (dv
->disposition
== 'c')
1046 disc
.state
|= (1 << MD_DISK_CANDIDATE
);
1048 disc
.state
|= (1 << MD_DISK_CLUSTER_ADD
);
1051 if (dv
->writemostly
== FlagSet
)
1052 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1053 if (dv
->failfast
== FlagSet
)
1054 disc
.state
|= (1 << MD_DISK_FAILFAST
);
1055 if (tst
->ss
->external
) {
1056 if (manage_add_external(tst
, fd
, dv
->devname
, &disc
) != MDADM_STATUS_SUCCESS
)
1059 tst
->ss
->free_super(tst
);
1060 if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
1061 if (dv
->disposition
== 'j')
1062 pr_err("Failed to hot add %s as journal, "
1063 "please try restart %s.\n", dv
->devname
, devname
);
1065 pr_err("add new device failed for %s as %d: %s\n",
1066 dv
->devname
, j
, strerror(errno
));
1069 if (dv
->disposition
== 'j') {
1070 pr_err("Journal added successfully, making %s read-write\n", devname
);
1071 if (Manage_ro(devname
, fd
, -1))
1072 pr_err("Failed to make %s read-write\n", devname
);
1077 pr_err("added %s\n", dv
->devname
);
1085 int Manage_remove(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1086 int sysfd
, unsigned long rdev
, int force
, int verbose
, char *devname
)
1091 if (tst
->ss
->external
) {
1092 /* To remove a device from a container, we must
1093 * check that it isn't in use in an array.
1094 * This involves looking in the 'holders'
1095 * directory - there must be just one entry,
1097 * To ensure that it doesn't get used as a
1098 * hot spare while we are checking, we
1099 * get an O_EXCL open on the container
1103 strcpy(devnm
, fd2devnm(fd
));
1104 lfd
= open_dev_excl(devnm
);
1106 pr_err("Cannot get exclusive access to container - odd\n");
1109 /* We may not be able to check on holders in
1110 * sysfs, either because we don't have the dev num
1111 * (rdev == 0) or because the device has been detached
1112 * and the 'holders' directory no longer exists
1113 * (ret == -1). In that case, assume it is OK to
1120 * The drive has already been set to 'faulty', however
1121 * monitor might not have had time to process it and the
1122 * drive might still have an entry in the 'holders'
1123 * directory. Try a few times to avoid a false error
1128 ret
= sysfs_unique_holder(devnm
, rdev
);
1131 sleep_for(0, MSEC_TO_NSEC(100), true);
1132 } while (--count
> 0);
1135 pr_err("%s is not a member, cannot remove.\n",
1141 pr_err("%s is still in use, cannot remove.\n",
1148 /* FIXME check that it is a current member */
1150 /* device has been removed and we don't know
1151 * the major:minor number
1153 err
= sys_hot_remove_disk(sysfd
, force
);
1155 err
= hot_remove_disk(fd
, rdev
, force
);
1156 if (err
&& errno
== ENODEV
) {
1157 /* Old kernels rejected this if no personality
1159 struct mdinfo
*sra
= sysfs_read(fd
, NULL
, GET_DEVS
);
1160 struct mdinfo
*dv
= NULL
;
1163 for ( ; dv
; dv
=dv
->next
)
1164 if (dv
->disk
.major
== (int)major(rdev
) &&
1165 dv
->disk
.minor
== (int)minor(rdev
))
1168 err
= sysfs_set_str(sra
, dv
,
1176 pr_err("hot remove failed for %s: %s\n", dv
->devname
,
1182 if (tst
->ss
->external
) {
1184 * Before dropping our exclusive open we make an
1185 * attempt at preventing mdmon from seeing an
1186 * 'add' event before reconciling this 'remove'
1189 char *devnm
= fd2devnm(fd
);
1192 pr_err("unable to get container name\n");
1196 ping_manager(devnm
);
1201 pr_err("hot removed %s from %s\n",
1202 dv
->devname
, devname
);
1206 int Manage_replace(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1207 unsigned long rdev
, int verbose
, char *devname
)
1209 struct mdinfo
*mdi
, *di
;
1210 if (tst
->ss
->external
) {
1211 pr_err("--replace only supported for native metadata (0.90 or 1.x)\n");
1214 /* Need to find the device in sysfs and add 'want_replacement' to the
1217 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
);
1218 if (!mdi
|| !mdi
->devs
) {
1219 pr_err("Cannot find status of %s to enable replacement - strange\n",
1223 for (di
= mdi
->devs
; di
; di
= di
->next
)
1224 if (di
->disk
.major
== (int)major(rdev
) &&
1225 di
->disk
.minor
== (int)minor(rdev
))
1229 if (di
->disk
.raid_disk
< 0) {
1230 pr_err("%s is not active and so cannot be replaced.\n",
1235 rv
= sysfs_set_str(mdi
, di
,
1236 "state", "want_replacement");
1239 pr_err("Failed to request replacement for %s\n",
1244 pr_err("Marked %s (device %d in %s) for replacement\n",
1245 dv
->devname
, di
->disk
.raid_disk
, devname
);
1246 /* If there is a matching 'with', we need to tell it which
1249 while (dv
&& dv
->disposition
!= 'W')
1252 dv
->disposition
= 'w';
1253 dv
->used
= di
->disk
.raid_disk
;
1258 pr_err("%s not found in %s so cannot --replace it\n",
1259 dv
->devname
, devname
);
1263 int Manage_with(struct supertype
*tst
, int fd
, struct mddev_dev
*dv
,
1264 unsigned long rdev
, int verbose
, char *devname
)
1266 struct mdinfo
*mdi
, *di
;
1267 /* try to set 'slot' for 'rdev' in 'fd' to 'dv->used' */
1268 mdi
= sysfs_read(fd
, NULL
, GET_DEVS
|GET_STATE
);
1269 if (!mdi
|| !mdi
->devs
) {
1270 pr_err("Cannot find status of %s to enable replacement - strange\n",
1274 for (di
= mdi
->devs
; di
; di
= di
->next
)
1275 if (di
->disk
.major
== (int)major(rdev
) &&
1276 di
->disk
.minor
== (int)minor(rdev
))
1280 if (di
->disk
.state
& (1<<MD_DISK_FAULTY
)) {
1281 pr_err("%s is faulty and cannot be a replacement\n",
1286 if (di
->disk
.raid_disk
>= 0) {
1287 pr_err("%s is active and cannot be a replacement\n",
1292 rv
= sysfs_set_num(mdi
, di
,
1296 pr_err("Failed to set %s as preferred replacement.\n",
1301 pr_err("Marked %s in %s as replacement for device %d\n",
1302 dv
->devname
, devname
, dv
->used
);
1306 pr_err("%s not found in %s so cannot make it preferred replacement\n",
1307 dv
->devname
, devname
);
1312 * is_remove_safe() - Check if remove is safe.
1313 * @array: Array info.
1314 * @fd: Array file descriptor.
1315 * @devname: Name of device to remove.
1316 * @verbose: Verbose.
1318 * The function determines if array will be operational
1319 * after removing &devname.
1321 * Return: True if array will be operational, false otherwise.
1323 bool is_remove_safe(mdu_array_info_t
*array
, const int fd
, char *devname
, const int verbose
)
1325 dev_t devid
= devnm2devid(devname
+ 5);
1326 struct mdinfo
*mdi
= sysfs_read(fd
, NULL
, GET_DEVS
| GET_DISKS
| GET_STATE
);
1330 pr_err("Failed to read sysfs attributes for %s\n", devname
);
1334 char *avail
= xcalloc(array
->raid_disks
, sizeof(char));
1336 for (mdi
= mdi
->devs
; mdi
; mdi
= mdi
->next
) {
1337 if (mdi
->disk
.raid_disk
< 0)
1339 if (!(mdi
->disk
.state
& (1 << MD_DISK_SYNC
)))
1341 if (makedev(mdi
->disk
.major
, mdi
->disk
.minor
) == devid
)
1343 avail
[mdi
->disk
.raid_disk
] = 1;
1347 bool is_enough
= enough(array
->level
, array
->raid_disks
,
1348 array
->layout
, 1, avail
);
1355 * Manage_subdevs() - Execute operation depending on devmode.
1357 * @devname: name of the device.
1358 * @fd: file descriptor.
1359 * @devlist: list of sub-devices to manage.
1360 * @verbose: verbose level.
1362 * @update: type of update.
1363 * @force: force flag.
1365 * This function executes operation defined by devmode
1366 * for each dev from devlist.
1368 * 'a' - add the device
1369 * 'S' - add the device as a spare - don't try re-add
1370 * 'j' - add the device as a journal device
1371 * 'A' - re-add the device
1372 * 'r' - remove the device: HOT_REMOVE_DISK
1373 * device can be 'faulty' or 'detached' in which case all
1374 * matching devices are removed.
1375 * 'f' - set the device faulty SET_DISK_FAULTY
1376 * device can be 'detached' in which case any device that
1377 * is inaccessible will be marked faulty.
1378 * 'I' - remove device by using incremental fail
1379 * which is executed when device is removed surprisingly.
1380 * 'R' - mark this device as wanting replacement.
1381 * 'W' - this device is added if necessary and activated as
1382 * a replacement for a previous 'R' device.
1384 * 'w' - 'W' will be changed to 'w' when it is paired with
1385 * a 'R' device. If a 'W' is found while walking the list
1386 * it must be unpaired, and is an error.
1387 * 'M' - this is created by a 'missing' target. It is a slight
1389 * 'F' - Another variant of 'A', where the device was faulty
1390 * so must be removed from the array first.
1391 * 'c' - confirm the device as found (for clustered environments)
1393 * For 'f' and 'r', the device can also be a kernel-internal
1394 * name such as 'sdb'.
1396 * Return: 0 on success, otherwise 1 or 2.
1398 int Manage_subdevs(char *devname
, int fd
,
1399 struct mddev_dev
*devlist
, int verbose
, int test
,
1400 enum update_opt update
, int force
)
1402 mdu_array_info_t array
;
1403 unsigned long long array_size
;
1404 struct mddev_dev
*dv
;
1406 struct supertype
*tst
= NULL
;
1407 char *subarray
= NULL
;
1409 int count
= 0; /* number of actions taken */
1411 struct mdinfo devinfo
;
1416 if (sysfs_init(&info
, fd
, NULL
)) {
1417 pr_err("sysfs not availabile for %s\n", devname
);
1421 if (md_get_array_info(fd
, &array
)) {
1422 pr_err("Cannot get array info for %s\n", devname
);
1425 /* array.size is only 32 bits and may be truncated.
1426 * So read from sysfs if possible, and record number of sectors
1429 array_size
= get_component_size(fd
);
1430 if (array_size
<= 0)
1431 array_size
= array
.size
* 2;
1433 tst
= super_by_fd(fd
, &subarray
);
1435 pr_err("unsupport array - version %d.%d\n",
1436 array
.major_version
, array
.minor_version
);
1440 for (dv
= devlist
; dv
; dv
= dv
->next
) {
1441 dev_t rdev
= 0; /* device to add/remove etc */
1446 if (dv
->disposition
== 'c') {
1447 rv
= parse_cluster_confirm_arg(dv
->devname
,
1451 pr_err("Could not get the devname of cluster\n");
1456 if (strcmp(dv
->devname
, "failed") == 0 ||
1457 strcmp(dv
->devname
, "faulty") == 0) {
1458 if (dv
->disposition
!= 'A' && dv
->disposition
!= 'r') {
1459 pr_err("%s only meaningful with -r or --re-add, not -%c\n",
1460 dv
->devname
, dv
->disposition
);
1463 add_faulty(dv
, fd
, (dv
->disposition
== 'A'
1467 if (strcmp(dv
->devname
, "detached") == 0) {
1468 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
1469 pr_err("%s only meaningful with -r of -f, not -%c\n",
1470 dv
->devname
, dv
->disposition
);
1473 add_detached(dv
, fd
, dv
->disposition
);
1477 if (strcmp(dv
->devname
, "missing") == 0) {
1478 struct mddev_dev
*add_devlist
;
1479 struct mddev_dev
**dp
;
1480 if (dv
->disposition
== 'c') {
1481 rv
= ioctl(fd
, CLUSTERED_DISK_NACK
, NULL
);
1485 if (dv
->disposition
!= 'A') {
1486 pr_err("'missing' only meaningful with --re-add\n");
1489 add_devlist
= conf_get_devs();
1490 if (add_devlist
== NULL
) {
1491 pr_err("no devices to scan for missing members.\n");
1494 for (dp
= &add_devlist
; *dp
; dp
= & (*dp
)->next
)
1495 /* 'M' (for 'missing') is like 'A' without errors */
1496 (*dp
)->disposition
= 'M';
1498 dv
->next
= add_devlist
;
1502 if (strncmp(dv
->devname
, "set-", 4) == 0 &&
1503 strlen(dv
->devname
) == 5) {
1506 if (dv
->disposition
!= 'r' &&
1507 dv
->disposition
!= 'f') {
1508 pr_err("'%s' only meaningful with -r or -f\n",
1512 if (array
.level
!= 10) {
1513 pr_err("'%s' only meaningful with RAID10 arrays\n",
1517 copies
= ((array
.layout
& 0xff) *
1518 ((array
.layout
>> 8) & 0xff));
1519 if (array
.raid_disks
% copies
!= 0 ||
1520 dv
->devname
[4] < 'A' ||
1521 dv
->devname
[4] >= 'A' + copies
||
1523 pr_err("'%s' not meaningful with this array\n",
1527 add_set(dv
, fd
, dv
->devname
[4]);
1531 if (strchr(dv
->devname
, '/') == NULL
&&
1532 strchr(dv
->devname
, ':') == NULL
&&
1533 strlen(dv
->devname
) < 50) {
1534 /* Assume this is a kernel-internal name like 'sda1' */
1537 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f' &&
1538 dv
->disposition
!= 'I') {
1539 pr_err("%s only meaningful with -r, -f or -I, not -%c\n",
1540 dv
->devname
, dv
->disposition
);
1544 sprintf(dname
, "dev-%s", dv
->devname
);
1545 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "block/dev");
1547 char dn
[SYSFS_MAX_BUF_SIZE
];
1548 if (sysfs_fd_get_str(sysfd
, dn
, sizeof(dn
)) > 0 &&
1549 sscanf(dn
, "%d:%d", &mj
,&mn
) == 2) {
1550 rdev
= makedev(mj
,mn
);
1557 sysfd
= sysfs_open(fd2devnm(fd
), dname
, "state");
1559 pr_err("%s does not appear to be a component of %s\n",
1560 dv
->devname
, devname
);
1564 } else if ((dv
->disposition
== 'r' ||
1565 dv
->disposition
== 'f') &&
1566 get_maj_min(dv
->devname
, &mj
, &mn
)) {
1567 /* for 'fail' and 'remove', the device might
1570 rdev
= makedev(mj
, mn
);
1572 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1574 fstat_is_blkdev(tfd
, dv
->devname
, &rdev
);
1577 int open_err
= errno
;
1578 if (!stat_is_blkdev(dv
->devname
, &rdev
)) {
1579 if (dv
->disposition
== 'M')
1580 /* non-fatal. Also improbable */
1584 if (dv
->disposition
== 'r')
1585 /* Be happy, the stat worked, that is
1586 * enough for --remove
1590 if (dv
->disposition
== 'M')
1593 pr_err("Cannot open %s: %s\n",
1594 dv
->devname
, strerror(open_err
));
1599 switch(dv
->disposition
){
1601 pr_err("internal error - devmode[%s]=%d\n",
1602 dv
->devname
, dv
->disposition
);
1605 case 'S': /* --add-spare */
1606 case 'j': /* --add-journal */
1608 case 'M': /* --re-add missing */
1609 case 'F': /* --re-add faulty */
1610 case 'c': /* --cluster-confirm */
1611 /* add the device */
1613 pr_err("Cannot add disks to a \'member\' array, perform this operation on the parent container\n");
1617 /* Let's first try to write re-add to sysfs */
1619 (dv
->disposition
== 'A' || dv
->disposition
== 'F')) {
1620 sysfs_init_dev(&devinfo
, rdev
);
1621 if (sysfs_set_str(&info
, &devinfo
, "state", "re-add") == 0) {
1622 pr_err("re-add %s to %s succeed\n",
1623 dv
->devname
, info
.sys_name
);
1628 if (dv
->disposition
== 'F')
1629 /* Need to remove first */
1630 hot_remove_disk(fd
, rdev
, force
);
1631 /* Make sure it isn't in use (in 2.6 or later) */
1632 tfd
= dev_open(dv
->devname
, O_RDONLY
|O_EXCL
);
1634 /* We know no-one else is using it. We'll
1635 * need non-exclusive access to add it, so
1639 tfd
= dev_open(dv
->devname
, O_RDONLY
);
1642 if (dv
->disposition
== 'M')
1644 pr_err("Cannot open %s: %s\n",
1645 dv
->devname
, strerror(errno
));
1649 if (sysfs_freeze_array(&info
) == 1)
1654 rv
= Manage_add(fd
, tfd
, dv
, tst
, &array
,
1655 force
, verbose
, devname
, update
,
1656 rdev
, array_size
, raid_slot
);
1668 pr_err("Cannot remove disks from a \'member\' array, perform this operation on the parent container\n");
1671 rv
= Manage_remove(tst
, fd
, dv
, sysfd
,
1672 rdev
, verbose
, force
,
1683 case 'f': /* set faulty */
1684 if (!is_remove_safe(&array
, fd
, dv
->devname
, verbose
)) {
1685 pr_err("Cannot remove %s from %s, array will be failed.\n",
1686 dv
->devname
, devname
);
1691 case 'I': /* incremental fail */
1692 if ((sysfd
>= 0 && write(sysfd
, "faulty", 6) != 6) ||
1693 (sysfd
< 0 && ioctl(fd
, SET_DISK_FAULTY
,
1697 pr_err("set device faulty failed for %s: %s\n",
1698 dv
->devname
, strerror(errno
));
1708 pr_err("set %s faulty in %s\n",
1709 dv
->devname
, devname
);
1711 case 'R': /* Mark as replaceable */
1713 pr_err("Cannot replace disks in a \'member\' array, perform this operation on the parent container\n");
1717 if (sysfs_freeze_array(&info
) == 1)
1722 rv
= Manage_replace(tst
, fd
, dv
,
1731 case 'W': /* --with device that doesn't match */
1732 pr_err("No matching --replace device for --with %s\n",
1735 case 'w': /* --with device which was matched */
1736 rv
= Manage_with(tst
, fd
, dv
,
1737 rdev
, verbose
, devname
);
1745 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1746 if (test
&& count
== 0)
1753 sysfs_set_str(&info
, NULL
, "sync_action","idle");
1754 return !test
&& busy
? 2 : 1;
1757 int autodetect(void)
1759 /* Open any md device, and issue the RAID_AUTORUN ioctl */
1761 int fd
= dev_open("9:0", O_RDONLY
);
1763 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)
1770 int Update_subarray(char *dev
, char *subarray
, enum update_opt update
,
1771 struct mddev_ident
*ident
, int verbose
)
1773 struct supertype supertype
, *st
= &supertype
;
1775 struct mdinfo
*info
= NULL
;
1776 char *update_verb
= map_num(update_options
, update
);
1777 bool allow_active
= update
== UOPT_PPL
|| update
== UOPT_NO_PPL
;
1779 memset(st
, 0, sizeof(*st
));
1781 fd
= open_subarray(dev
, subarray
, st
, verbose
< 0);
1785 if (!st
->ss
->update_subarray
) {
1787 pr_err("Operation not supported for %s metadata\n",
1792 if (!allow_active
&& is_subarray_active(subarray
, st
->devnm
)) {
1794 pr_err("Subarray %s in %s is active, cannot update %s\n",
1795 subarray
, dev
, update_verb
);
1799 if (mdmon_running(st
->devnm
))
1800 st
->update_tail
= &st
->updates
;
1802 info
= st
->ss
->container_content(st
, subarray
);
1804 if (update
== UOPT_PPL
&& !is_level456(info
->array
.level
)) {
1805 pr_err("RWH policy ppl is supported only for raid4, raid5 and raid6.\n");
1809 rv
= st
->ss
->update_subarray(st
, subarray
, update
, ident
);
1813 pr_err("Failed to update %s of subarray-%s in %s\n",
1814 update_verb
, subarray
, dev
);
1815 } else if (st
->update_tail
)
1816 flush_metadata_updates(st
);
1818 st
->ss
->sync_metadata(st
);
1820 if (rv
== 0 && update
== UOPT_NAME
&& verbose
>= 0)
1821 pr_err("Updated subarray-%s name from %s, UUIDs may have changed\n",
1827 st
->ss
->free_super(st
);
1833 /* Move spare from one array to another If adding to destination array fails
1834 * add back to original array.
1835 * Returns 1 on success, 0 on failure */
1836 int move_spare(char *from_devname
, char *to_devname
, dev_t devid
)
1838 struct mddev_dev devlist
;
1841 /* try to remove and add */
1842 int fd1
= open(to_devname
, O_RDONLY
);
1843 int fd2
= open(from_devname
, O_RDONLY
);
1845 if (fd1
< 0 || fd2
< 0) {
1853 devlist
.next
= NULL
;
1855 devlist
.writemostly
= FlagDefault
;
1856 devlist
.failfast
= FlagDefault
;
1857 devlist
.devname
= devname
;
1858 sprintf(devname
, "%d:%d", major(devid
), minor(devid
));
1860 devlist
.disposition
= 'r';
1861 if (Manage_subdevs(from_devname
, fd2
, &devlist
, -1, 0, UOPT_UNDEFINED
, 0) == 0) {
1862 devlist
.disposition
= 'a';
1863 if (Manage_subdevs(to_devname
, fd1
, &devlist
, -1, 0,
1864 UOPT_UNDEFINED
, 0) == 0) {
1865 /* make sure manager is aware of changes */
1866 ping_manager(to_devname
);
1867 ping_manager(from_devname
);
1873 Manage_subdevs(from_devname
, fd2
, &devlist
,
1874 -1, 0, UOPT_UNDEFINED
, 0);