]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Manage.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
34 #define REGISTER_DEV _IO (MD_MAJOR, 1)
35 #define START_MD _IO (MD_MAJOR, 2)
36 #define STOP_MD _IO (MD_MAJOR, 3)
38 int Manage_ro(char *devname
, int fd
, int readonly
)
40 /* switch to readonly or rw
43 * first check that array is runing
44 * use RESTART_ARRAY_RW or STOP_ARRAY_RO
47 mdu_array_info_t array
;
52 if (md_get_version(fd
) < 9000) {
53 fprintf(stderr
, Name
": need md driver version 0.90.0 or later\n");
57 /* If this is an externally-manage array, we need to modify the
58 * metadata_version so that mdmon doesn't undo our change.
60 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
62 mdi
->array
.major_version
== -1 &&
63 mdi
->array
.level
> 0 &&
64 is_subarray(mdi
->text_version
)) {
66 strcpy(vers
, "external:");
67 strcat(vers
, mdi
->text_version
);
70 /* We set readonly ourselves. */
72 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
75 rv
= sysfs_set_str(mdi
, NULL
, "array_state", "readonly");
78 fprintf(stderr
, Name
": failed to set readonly for %s: %s\n",
79 devname
, strerror(errno
));
81 vers
[9] = mdi
->text_version
[0];
82 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
87 /* We cannot set read/write - must signal mdmon */
89 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
91 cp
= strchr(vers
+10, '/');
94 ping_monitor(vers
+10);
99 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
100 fprintf(stderr
, Name
": %s does not appear to be active.\n",
106 if (ioctl(fd
, STOP_ARRAY_RO
, NULL
)) {
107 fprintf(stderr
, Name
": failed to set readonly for %s: %s\n",
108 devname
, strerror(errno
));
111 } else if (readonly
< 0) {
112 if (ioctl(fd
, RESTART_ARRAY_RW
, NULL
)) {
113 fprintf(stderr
, Name
": failed to set writable for %s: %s\n",
114 devname
, strerror(errno
));
123 int Manage_runstop(char *devname
, int fd
, int runstop
, int quiet
)
125 /* Run or stop the array. array must already be configured
127 * Only print failure messages if quiet == 0;
128 * quiet > 0 means really be quiet
129 * quiet < 0 means we will try again if it fails.
131 mdu_param_t param
; /* unused */
133 if (runstop
== -1 && md_get_version(fd
) < 9000) {
134 if (ioctl(fd
, STOP_MD
, 0)) {
135 if (quiet
== 0) fprintf(stderr
,
136 Name
": stopping device %s "
138 devname
, strerror(errno
));
143 if (md_get_version(fd
) < 9000) {
144 fprintf(stderr
, Name
": need md driver version 0.90.0 or later\n");
148 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
149 fprintf(stderr, Name ": %s does not appear to be active.\n",
155 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
156 fprintf(stderr
, Name
": failed to run array %s: %s\n",
157 devname
, strerror(errno
));
161 fprintf(stderr
, Name
": started %s\n", devname
);
162 } else if (runstop
< 0){
163 struct map_ent
*map
= NULL
;
166 /* If this is an mdmon managed array, just write 'inactive'
167 * to the array state and let mdmon clear up.
169 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
171 mdi
->array
.level
> 0 &&
172 is_subarray(mdi
->text_version
)) {
173 /* This is mdmon managed. */
175 if (sysfs_set_str(mdi
, NULL
,
176 "array_state", "inactive") < 0) {
179 ": failed to stop array %s: %s\n",
180 devname
, strerror(errno
));
184 /* Give monitor a chance to act */
185 ping_monitor(mdi
->text_version
);
187 fd
= open(devname
, O_RDONLY
);
189 mdi
->array
.major_version
== -1 &&
190 mdi
->array
.minor_version
== -2 &&
191 !is_subarray(mdi
->text_version
)) {
192 /* container, possibly mdmon-managed.
193 * Make sure mdmon isn't opening it, which
194 * would interfere with the 'stop'
196 ping_monitor(mdi
->sys_name
);
201 if (fd
>= 0 && ioctl(fd
, STOP_ARRAY
, NULL
)) {
204 ": failed to stop array %s: %s\n",
205 devname
, strerror(errno
));
207 fprintf(stderr
, "Perhaps a running "
208 "process, mounted filesystem "
209 "or active volume group?\n");
215 fprintf(stderr
, Name
": stopped %s\n", devname
);
216 if (fd
>= 0 && fstat(fd
, &stb
) == 0) {
218 if (major(stb
.st_rdev
) == MD_MAJOR
)
219 devnum
= minor(stb
.st_rdev
);
221 devnum
= -1-(minor(stb
.st_rdev
)>>6);
222 map_delete(&map
, devnum
);
230 int Manage_resize(char *devname
, int fd
, long long size
, int raid_disks
)
232 mdu_array_info_t info
;
233 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
234 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
235 devname
, strerror(errno
));
241 info
.raid_disks
= raid_disks
;
242 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
243 fprintf(stderr
, Name
": Cannot set device size/shape for %s: %s\n",
244 devname
, strerror(errno
));
250 int Manage_reconfig(char *devname
, int fd
, int layout
)
252 mdu_array_info_t info
;
253 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
254 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
255 devname
, strerror(errno
));
258 info
.layout
= layout
;
259 printf("layout set to %d\n", info
.layout
);
260 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
261 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
262 devname
, strerror(errno
));
268 int Manage_subdevs(char *devname
, int fd
,
269 mddev_dev_t devlist
, int verbose
)
271 /* do something to each dev.
273 * 'a' - add the device
275 * If that fails EINVAL, try ADD_NEW_DISK
276 * 'r' - remove the device HOT_REMOVE_DISK
277 * device can be 'faulty' or 'detached' in which case all
278 * matching devices are removed.
279 * 'f' - set the device faulty SET_DISK_FAULTY
280 * device can be 'detached' in which case any device that
281 * is inaccessible will be marked faulty.
283 mdu_array_info_t array
;
284 mdu_disk_info_t disc
;
285 unsigned long long array_size
;
286 mddev_dev_t dv
, next
= NULL
;
290 struct supertype
*st
, *tst
;
295 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
296 fprintf(stderr
, Name
": cannot get array info for %s\n",
301 /* array.size is only 32 bit and may be truncated.
302 * So read from sysfs if possible, and record number of sectors
305 array_size
= get_component_size(fd
);
307 array_size
= array
.size
* 2;
309 tst
= super_by_fd(fd
);
311 fprintf(stderr
, Name
": unsupport array - version %d.%d\n",
312 array
.major_version
, array
.minor_version
);
316 for (dv
= devlist
, j
=0 ; dv
; dv
= next
, j
= jnext
) {
317 unsigned long long ldsize
;
319 char *dnprintable
= dv
->devname
;
325 if (strcmp(dv
->devname
, "failed")==0 ||
326 strcmp(dv
->devname
, "faulty")==0) {
327 if (dv
->disposition
!= 'r') {
328 fprintf(stderr
, Name
": %s only meaningful "
329 "with -r, not -%c\n",
330 dv
->devname
, dv
->disposition
);
333 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
335 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
337 if (disc
.major
== 0 && disc
.minor
== 0)
339 if ((disc
.state
& 1) == 0) /* faulty */
341 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
344 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
345 dnprintable
= dvname
;
350 } else if (strcmp(dv
->devname
, "detached") == 0) {
351 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
352 fprintf(stderr
, Name
": %s only meaningful "
353 "with -r of -f, not -%c\n",
354 dv
->devname
, dv
->disposition
);
357 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
360 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
362 if (disc
.major
== 0 && disc
.minor
== 0)
364 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
365 sfd
= dev_open(dvname
, O_RDONLY
);
370 if (dv
->disposition
== 'f' &&
371 (disc
.state
& 1) == 1) /* already faulty */
375 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
378 dnprintable
= dvname
;
386 if (stat(dv
->devname
, &stb
)) {
387 fprintf(stderr
, Name
": cannot find %s: %s\n",
388 dv
->devname
, strerror(errno
));
391 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
392 fprintf(stderr
, Name
": %s is not a "
398 switch(dv
->disposition
){
400 fprintf(stderr
, Name
": internal error - devmode[%s]=%d\n",
401 dv
->devname
, dv
->disposition
);
405 if (tst
->subarray
[0]) {
406 fprintf(stderr
, Name
": Cannot add disks to a"
407 " \'member\' array, perform this"
408 " operation on the parent container\n");
411 /* Make sure it isn't in use (in 2.6 or later) */
412 tfd
= open(dv
->devname
, O_RDONLY
|O_EXCL
|O_DIRECT
);
414 fprintf(stderr
, Name
": Cannot open %s: %s\n",
415 dv
->devname
, strerror(errno
));
418 remove_partitions(tfd
);
422 if (array
.not_persistent
==0)
423 st
->ss
->load_super(st
, tfd
, NULL
);
425 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
432 if (!tst
->ss
->external
&&
433 array
.major_version
== 0 &&
434 md_get_version(fd
)%100 < 2) {
435 if (ioctl(fd
, HOT_ADD_DISK
,
436 (unsigned long)stb
.st_rdev
)==0) {
438 fprintf(stderr
, Name
": hot added %s\n",
443 fprintf(stderr
, Name
": hot add failed for %s: %s\n",
444 dv
->devname
, strerror(errno
));
448 if (array
.not_persistent
== 0 || tst
->ss
->external
) {
450 /* need to find a sample superblock to copy, and
451 * a spare slot to use.
452 * For 'external' array (well, container based),
453 * We can just load the metadata for the array.
455 if (tst
->ss
->external
) {
456 tst
->ss
->load_super(tst
, fd
, NULL
);
457 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
461 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
463 if (disc
.major
==0 && disc
.minor
==0)
465 if ((disc
.state
& 4)==0) continue; /* sync */
466 /* Looks like a good device to try */
467 dev
= map_dev(disc
.major
, disc
.minor
, 1);
469 dfd
= dev_open(dev
, O_RDONLY
);
470 if (dfd
< 0) continue;
471 if (tst
->ss
->load_super(tst
, dfd
,
479 /* FIXME this is a bad test to be using */
481 fprintf(stderr
, Name
": cannot find valid superblock in this array - HELP\n");
485 /* Make sure device is large enough */
486 if (tst
->ss
->avail_size(tst
, ldsize
/512) <
488 fprintf(stderr
, Name
": %s not large enough to join array\n",
493 /* Possibly this device was recently part of the array
494 * and was temporarily removed, and is now being re-added.
495 * If so, we can simply re-add it.
497 tst
->ss
->uuid_from_super(tst
, duuid
);
499 /* re-add doesn't work for version-1 superblocks
502 if (array
.major_version
== 1 &&
503 get_linux_version() <= 2006018)
506 st
->ss
->uuid_from_super(st
, ouuid
);
507 if (memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
508 /* looks close enough for now. Kernel
509 * will worry about whether a bitmap
510 * based reconstruction is possible.
513 st
->ss
->getinfo_super(st
, &mdi
);
514 disc
.major
= major(stb
.st_rdev
);
515 disc
.minor
= minor(stb
.st_rdev
);
516 disc
.number
= mdi
.disk
.number
;
517 disc
.raid_disk
= mdi
.disk
.raid_disk
;
518 disc
.state
= mdi
.disk
.state
;
519 if (dv
->writemostly
== 1)
520 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
521 if (dv
->writemostly
== 2)
522 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
523 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
525 fprintf(stderr
, Name
": re-added %s\n", dv
->devname
);
528 /* fall back on normal-add */
532 /* non-persistent. Must ensure that new drive
533 * is at least array.size big.
535 if (ldsize
/512 < array_size
) {
536 fprintf(stderr
, Name
": %s not large enough to join array\n",
541 /* in 2.6.17 and earlier, version-1 superblocks won't
542 * use the number we write, but will choose a free number.
543 * we must choose the same free number, which requires
544 * starting at 'raid_disks' and counting up
546 for (j
= array
.raid_disks
; j
< tst
->max_devs
; j
++) {
548 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
550 if (disc
.major
==0 && disc
.minor
==0)
552 if (disc
.state
& 8) /* removed */
555 disc
.major
= major(stb
.st_rdev
);
556 disc
.minor
= minor(stb
.st_rdev
);
559 if (array
.not_persistent
==0 || tst
->ss
->external
) {
561 if (dv
->writemostly
== 1)
562 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
563 dfd
= open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
564 tst
->ss
->add_to_super(tst
, &disc
, dfd
,
566 /* write_init_super will close 'dfd' */
567 if (tst
->ss
->external
)
568 /* mdmon will write the metadata */
570 else if (tst
->ss
->write_init_super(tst
))
572 } else if (dv
->re_add
) {
573 /* this had better be raid1.
574 * As we are "--re-add"ing we must find a spare slot
577 char *used
= malloc(array
.raid_disks
);
578 memset(used
, 0, array
.raid_disks
);
579 for (j
=0; j
< tst
->max_devs
; j
++) {
580 mdu_disk_info_t disc2
;
582 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
584 if (disc2
.major
==0 && disc2
.minor
==0)
586 if (disc2
.state
& 8) /* removed */
588 if (disc2
.raid_disk
< 0)
590 if (disc2
.raid_disk
> array
.raid_disks
)
592 used
[disc2
.raid_disk
] = 1;
594 for (j
=0 ; j
<array
.raid_disks
; j
++)
597 disc
.state
|= (1<<MD_DISK_SYNC
);
601 if (dv
->writemostly
== 1)
602 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
603 if (tst
->ss
->external
) {
604 /* add a disk to an external metadata container
605 * only if mdmon is around to see it
607 struct mdinfo new_mdi
;
610 int devnum
= fd2devnum(fd
);
612 container_fd
= open_dev_excl(devnum
);
613 if (container_fd
< 0) {
614 fprintf(stderr
, Name
": add failed for %s:"
615 " could not get exclusive access to container\n",
620 if (!mdmon_running(devnum
)) {
621 fprintf(stderr
, Name
": add failed for %s: mdmon not running\n",
627 sra
= sysfs_read(container_fd
, -1, 0);
629 fprintf(stderr
, Name
": add failed for %s: sysfs_read failed\n",
634 sra
->array
.level
= LEVEL_CONTAINER
;
635 /* Need to set data_offset and component_size */
636 tst
->ss
->getinfo_super(tst
, &new_mdi
);
637 new_mdi
.disk
.major
= disc
.major
;
638 new_mdi
.disk
.minor
= disc
.minor
;
639 if (sysfs_add_disk(sra
, &new_mdi
) != 0) {
640 fprintf(stderr
, Name
": add new device to external metadata"
641 " failed for %s\n", dv
->devname
);
645 ping_monitor(devnum2devname(devnum
));
648 } else if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
649 fprintf(stderr
, Name
": add new device failed for %s as %d: %s\n",
650 dv
->devname
, j
, strerror(errno
));
654 fprintf(stderr
, Name
": added %s\n", dv
->devname
);
659 if (tst
->subarray
[0]) {
660 fprintf(stderr
, Name
": Cannot remove disks from a"
661 " \'member\' array, perform this"
662 " operation on the parent container\n");
665 if (tst
->ss
->external
) {
666 /* To remove a device from a container, we must
667 * check that it isn't in use in an array.
668 * This involves looking in the 'holders'
669 * directory - there must be just one entry,
671 * To ensure that it doesn't get used as a
672 * hold spare while we are checking, we
673 * get an O_EXCL open on the container
675 int dnum
= fd2devnum(fd
);
676 lfd
= open_dev_excl(dnum
);
679 ": Cannot get exclusive access "
680 " to container - odd\n");
683 if (!sysfs_unique_holder(dnum
, stb
.st_rdev
)) {
685 ": %s is %s, cannot remove.\n",
687 errno
== EEXIST
? "still in use":
693 /* FIXME check that it is a current member */
694 err
= ioctl(fd
, HOT_REMOVE_DISK
, (unsigned long)stb
.st_rdev
);
695 if (err
&& errno
== ENODEV
) {
696 /* Old kernels rejected this if no personality
698 struct mdinfo
*sra
= sysfs_read(fd
, 0, GET_DEVS
);
699 struct mdinfo
*dv
= NULL
;
702 for ( ; dv
; dv
=dv
->next
)
703 if (dv
->disk
.major
== major(stb
.st_rdev
) &&
704 dv
->disk
.minor
== minor(stb
.st_rdev
))
707 err
= sysfs_set_str(sra
, dv
,
715 fprintf(stderr
, Name
": hot remove failed "
716 "for %s: %s\n", dnprintable
,
722 if (tst
->ss
->external
) {
724 * Before dropping our exclusive open we make an
725 * attempt at preventing mdmon from seeing an
726 * 'add' event before reconciling this 'remove'
729 char *name
= devnum2devname(fd2devnum(fd
));
732 fprintf(stderr
, Name
": unable to get container name\n");
741 fprintf(stderr
, Name
": hot removed %s\n",
745 case 'f': /* set faulty */
746 /* FIXME check current member */
747 if (ioctl(fd
, SET_DISK_FAULTY
, (unsigned long) stb
.st_rdev
)) {
748 fprintf(stderr
, Name
": set device faulty failed for %s: %s\n",
749 dnprintable
, strerror(errno
));
753 fprintf(stderr
, Name
": set %s faulty in %s\n",
754 dnprintable
, devname
);
764 /* Open any md device, and issue the RAID_AUTORUN ioctl */
766 int fd
= dev_open("9:0", O_RDONLY
);
768 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)