]>
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
35 #define REGISTER_DEV _IO (MD_MAJOR, 1)
36 #define START_MD _IO (MD_MAJOR, 2)
37 #define STOP_MD _IO (MD_MAJOR, 3)
39 int Manage_ro(char *devname
, int fd
, int readonly
)
41 /* switch to readonly or rw
44 * first check that array is runing
45 * use RESTART_ARRAY_RW or STOP_ARRAY_RO
48 mdu_array_info_t array
;
53 if (md_get_version(fd
) < 9000) {
54 fprintf(stderr
, Name
": need md driver version 0.90.0 or later\n");
58 /* If this is an externally-manage array, we need to modify the
59 * metadata_version so that mdmon doesn't undo our change.
61 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
63 mdi
->array
.major_version
== -1 &&
64 mdi
->array
.level
> 0 &&
65 is_subarray(mdi
->text_version
)) {
67 strcpy(vers
, "external:");
68 strcat(vers
, mdi
->text_version
);
71 /* We set readonly ourselves. */
73 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
76 rv
= sysfs_set_str(mdi
, NULL
, "array_state", "readonly");
79 fprintf(stderr
, Name
": failed to set readonly for %s: %s\n",
80 devname
, strerror(errno
));
82 vers
[9] = mdi
->text_version
[0];
83 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
88 /* We cannot set read/write - must signal mdmon */
90 sysfs_set_str(mdi
, NULL
, "metadata_version", vers
);
92 cp
= strchr(vers
+10, '/');
95 ping_monitor(vers
+10);
100 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
101 fprintf(stderr
, Name
": %s does not appear to be active.\n",
107 if (ioctl(fd
, STOP_ARRAY_RO
, NULL
)) {
108 fprintf(stderr
, Name
": failed to set readonly for %s: %s\n",
109 devname
, strerror(errno
));
112 } else if (readonly
< 0) {
113 if (ioctl(fd
, RESTART_ARRAY_RW
, NULL
)) {
114 fprintf(stderr
, Name
": failed to set writable for %s: %s\n",
115 devname
, strerror(errno
));
124 static void remove_devices(int devnum
, char *path
)
126 /* Remove all 'standard' devices for 'devnum', including
127 * partitions. Also remove names at 'path' - possibly with
128 * partition suffixes - which link to those names.
139 sprintf(base
, "/dev/md%d", devnum
);
141 sprintf(base
, "/dev/md_d%d", -1-devnum
);
142 be
= base
+ strlen(base
);
144 path2
= malloc(strlen(path
)+20);
146 pe
= path2
+ strlen(path2
);
150 for (part
= 0; part
< 16; part
++) {
152 sprintf(be
, "p%d", part
);
154 sprintf(pe
, "p%d", part
);
156 sprintf(pe
, "%d", part
);
158 /* FIXME test if really is md device ?? */
161 n
= readlink(path2
, link
, sizeof(link
));
162 if (n
&& strlen(base
) == n
&&
163 strncmp(link
, base
, n
) == 0)
170 int Manage_runstop(char *devname
, int fd
, int runstop
, int quiet
)
172 /* Run or stop the array. array must already be configured
174 * Only print failure messages if quiet == 0;
175 * quiet > 0 means really be quiet
176 * quiet < 0 means we will try again if it fails.
178 mdu_param_t param
; /* unused */
180 if (runstop
== -1 && md_get_version(fd
) < 9000) {
181 if (ioctl(fd
, STOP_MD
, 0)) {
182 if (quiet
== 0) fprintf(stderr
,
183 Name
": stopping device %s "
185 devname
, strerror(errno
));
190 if (md_get_version(fd
) < 9000) {
191 fprintf(stderr
, Name
": need md driver version 0.90.0 or later\n");
195 if (ioctl(fd, GET_ARRAY_INFO, &array)) {
196 fprintf(stderr, Name ": %s does not appear to be active.\n",
202 if (ioctl(fd
, RUN_ARRAY
, ¶m
)) {
203 fprintf(stderr
, Name
": failed to run array %s: %s\n",
204 devname
, strerror(errno
));
208 fprintf(stderr
, Name
": started %s\n", devname
);
209 } else if (runstop
< 0){
210 struct map_ent
*map
= NULL
;
214 /* If this is an mdmon managed array, just write 'inactive'
215 * to the array state and let mdmon clear up.
217 devnum
= fd2devnum(fd
);
218 mdi
= sysfs_read(fd
, -1, GET_LEVEL
|GET_VERSION
);
220 mdi
->array
.level
> 0 &&
221 is_subarray(mdi
->text_version
)) {
222 /* This is mdmon managed. */
224 if (sysfs_set_str(mdi
, NULL
,
225 "array_state", "inactive") < 0) {
228 ": failed to stop array %s: %s\n",
229 devname
, strerror(errno
));
233 /* Give monitor a chance to act */
234 ping_monitor(mdi
->text_version
);
236 fd
= open(devname
, O_RDONLY
);
238 mdi
->array
.major_version
== -1 &&
239 mdi
->array
.minor_version
== -2 &&
240 !is_subarray(mdi
->text_version
)) {
241 /* container, possibly mdmon-managed.
242 * Make sure mdmon isn't opening it, which
243 * would interfere with the 'stop'
245 ping_monitor(mdi
->sys_name
);
248 if (fd
>= 0 && ioctl(fd
, STOP_ARRAY
, NULL
)) {
251 ": failed to stop array %s: %s\n",
252 devname
, strerror(errno
));
254 fprintf(stderr
, "Perhaps a running "
255 "process, mounted filesystem "
256 "or active volume group?\n");
262 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
263 * was stopped, so We'll do it here just to be sure.
266 sysfs_uevent(mdi
, "change");
269 if (devnum
!= NoMdDev
&&
270 (stat("/dev/.udev", &stb
) != 0 ||
271 check_env("MDADM_NO_UDEV"))) {
272 struct map_ent
*mp
= map_by_devnum(&map
, devnum
);
273 remove_devices(devnum
, mp
? mp
->path
: NULL
);
278 fprintf(stderr
, Name
": stopped %s\n", devname
);
279 if (devnum
!= NoMdDev
) {
280 map_delete(&map
, devnum
);
288 int Manage_resize(char *devname
, int fd
, long long size
, int raid_disks
)
290 mdu_array_info_t info
;
291 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
292 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
293 devname
, strerror(errno
));
299 info
.raid_disks
= raid_disks
;
300 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
301 fprintf(stderr
, Name
": Cannot set device size/shape for %s: %s\n",
302 devname
, strerror(errno
));
308 int Manage_reconfig(char *devname
, int fd
, int layout
)
310 mdu_array_info_t info
;
311 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
312 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
313 devname
, strerror(errno
));
316 info
.layout
= layout
;
317 printf("layout set to %d\n", info
.layout
);
318 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
319 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
320 devname
, strerror(errno
));
326 int Manage_subdevs(char *devname
, int fd
,
327 mddev_dev_t devlist
, int verbose
)
329 /* do something to each dev.
331 * 'a' - add the device
333 * If that fails EINVAL, try ADD_NEW_DISK
334 * 'r' - remove the device HOT_REMOVE_DISK
335 * device can be 'faulty' or 'detached' in which case all
336 * matching devices are removed.
337 * 'f' - set the device faulty SET_DISK_FAULTY
338 * device can be 'detached' in which case any device that
339 * is inaccessible will be marked faulty.
341 mdu_array_info_t array
;
342 mdu_disk_info_t disc
;
343 unsigned long long array_size
;
344 mddev_dev_t dv
, next
= NULL
;
348 struct supertype
*st
, *tst
;
353 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
354 fprintf(stderr
, Name
": cannot get array info for %s\n",
359 /* array.size is only 32 bit and may be truncated.
360 * So read from sysfs if possible, and record number of sectors
363 array_size
= get_component_size(fd
);
365 array_size
= array
.size
* 2;
367 tst
= super_by_fd(fd
);
369 fprintf(stderr
, Name
": unsupport array - version %d.%d\n",
370 array
.major_version
, array
.minor_version
);
374 for (dv
= devlist
, j
=0 ; dv
; dv
= next
, j
= jnext
) {
375 unsigned long long ldsize
;
377 char *dnprintable
= dv
->devname
;
383 if (strcmp(dv
->devname
, "failed")==0 ||
384 strcmp(dv
->devname
, "faulty")==0) {
385 if (dv
->disposition
!= 'r') {
386 fprintf(stderr
, Name
": %s only meaningful "
387 "with -r, not -%c\n",
388 dv
->devname
, dv
->disposition
);
391 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
393 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
395 if (disc
.major
== 0 && disc
.minor
== 0)
397 if ((disc
.state
& 1) == 0) /* faulty */
399 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
402 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
403 dnprintable
= dvname
;
408 } else if (strcmp(dv
->devname
, "detached") == 0) {
409 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
410 fprintf(stderr
, Name
": %s only meaningful "
411 "with -r of -f, not -%c\n",
412 dv
->devname
, dv
->disposition
);
415 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
418 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
420 if (disc
.major
== 0 && disc
.minor
== 0)
422 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
423 sfd
= dev_open(dvname
, O_RDONLY
);
428 if (dv
->disposition
== 'f' &&
429 (disc
.state
& 1) == 1) /* already faulty */
433 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
436 dnprintable
= dvname
;
444 if (stat(dv
->devname
, &stb
)) {
445 fprintf(stderr
, Name
": cannot find %s: %s\n",
446 dv
->devname
, strerror(errno
));
449 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
450 fprintf(stderr
, Name
": %s is not a "
456 switch(dv
->disposition
){
458 fprintf(stderr
, Name
": internal error - devmode[%s]=%d\n",
459 dv
->devname
, dv
->disposition
);
463 if (tst
->subarray
[0]) {
464 fprintf(stderr
, Name
": Cannot add disks to a"
465 " \'member\' array, perform this"
466 " operation on the parent container\n");
469 /* Make sure it isn't in use (in 2.6 or later) */
470 tfd
= open(dv
->devname
, O_RDONLY
|O_EXCL
|O_DIRECT
);
472 fprintf(stderr
, Name
": Cannot open %s: %s\n",
473 dv
->devname
, strerror(errno
));
476 remove_partitions(tfd
);
480 if (array
.not_persistent
==0)
481 st
->ss
->load_super(st
, tfd
, NULL
);
483 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
490 if (!tst
->ss
->external
&&
491 array
.major_version
== 0 &&
492 md_get_version(fd
)%100 < 2) {
493 if (ioctl(fd
, HOT_ADD_DISK
,
494 (unsigned long)stb
.st_rdev
)==0) {
496 fprintf(stderr
, Name
": hot added %s\n",
501 fprintf(stderr
, Name
": hot add failed for %s: %s\n",
502 dv
->devname
, strerror(errno
));
506 if (array
.not_persistent
== 0 || tst
->ss
->external
) {
508 /* need to find a sample superblock to copy, and
509 * a spare slot to use.
510 * For 'external' array (well, container based),
511 * We can just load the metadata for the array.
513 if (tst
->ss
->external
) {
514 tst
->ss
->load_super(tst
, fd
, NULL
);
515 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
519 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
521 if (disc
.major
==0 && disc
.minor
==0)
523 if ((disc
.state
& 4)==0) continue; /* sync */
524 /* Looks like a good device to try */
525 dev
= map_dev(disc
.major
, disc
.minor
, 1);
527 dfd
= dev_open(dev
, O_RDONLY
);
528 if (dfd
< 0) continue;
529 if (tst
->ss
->load_super(tst
, dfd
,
537 /* FIXME this is a bad test to be using */
539 fprintf(stderr
, Name
": cannot find valid superblock in this array - HELP\n");
543 /* Make sure device is large enough */
544 if (tst
->ss
->avail_size(tst
, ldsize
/512) <
546 fprintf(stderr
, Name
": %s not large enough to join array\n",
551 /* Possibly this device was recently part of the array
552 * and was temporarily removed, and is now being re-added.
553 * If so, we can simply re-add it.
555 tst
->ss
->uuid_from_super(tst
, duuid
);
557 /* re-add doesn't work for version-1 superblocks
560 if (array
.major_version
== 1 &&
561 get_linux_version() <= 2006018)
564 st
->ss
->uuid_from_super(st
, ouuid
);
565 if (memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
566 /* looks close enough for now. Kernel
567 * will worry about whether a bitmap
568 * based reconstruction is possible.
571 st
->ss
->getinfo_super(st
, &mdi
);
572 disc
.major
= major(stb
.st_rdev
);
573 disc
.minor
= minor(stb
.st_rdev
);
574 disc
.number
= mdi
.disk
.number
;
575 disc
.raid_disk
= mdi
.disk
.raid_disk
;
576 disc
.state
= mdi
.disk
.state
;
577 if (dv
->writemostly
== 1)
578 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
579 if (dv
->writemostly
== 2)
580 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
581 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
583 fprintf(stderr
, Name
": re-added %s\n", dv
->devname
);
586 /* fall back on normal-add */
590 /* non-persistent. Must ensure that new drive
591 * is at least array.size big.
593 if (ldsize
/512 < array_size
) {
594 fprintf(stderr
, Name
": %s not large enough to join array\n",
599 /* in 2.6.17 and earlier, version-1 superblocks won't
600 * use the number we write, but will choose a free number.
601 * we must choose the same free number, which requires
602 * starting at 'raid_disks' and counting up
604 for (j
= array
.raid_disks
; j
< tst
->max_devs
; j
++) {
606 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
608 if (disc
.major
==0 && disc
.minor
==0)
610 if (disc
.state
& 8) /* removed */
613 disc
.major
= major(stb
.st_rdev
);
614 disc
.minor
= minor(stb
.st_rdev
);
617 if (array
.not_persistent
==0 || tst
->ss
->external
) {
619 if (dv
->writemostly
== 1)
620 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
621 dfd
= open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
622 tst
->ss
->add_to_super(tst
, &disc
, dfd
,
624 /* write_init_super will close 'dfd' */
625 if (tst
->ss
->external
)
626 /* mdmon will write the metadata */
628 else if (tst
->ss
->write_init_super(tst
))
630 } else if (dv
->re_add
) {
631 /* this had better be raid1.
632 * As we are "--re-add"ing we must find a spare slot
635 char *used
= malloc(array
.raid_disks
);
636 memset(used
, 0, array
.raid_disks
);
637 for (j
=0; j
< tst
->max_devs
; j
++) {
638 mdu_disk_info_t disc2
;
640 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
642 if (disc2
.major
==0 && disc2
.minor
==0)
644 if (disc2
.state
& 8) /* removed */
646 if (disc2
.raid_disk
< 0)
648 if (disc2
.raid_disk
> array
.raid_disks
)
650 used
[disc2
.raid_disk
] = 1;
652 for (j
=0 ; j
<array
.raid_disks
; j
++)
655 disc
.state
|= (1<<MD_DISK_SYNC
);
659 if (dv
->writemostly
== 1)
660 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
661 if (tst
->ss
->external
) {
662 /* add a disk to an external metadata container
663 * only if mdmon is around to see it
665 struct mdinfo new_mdi
;
668 int devnum
= fd2devnum(fd
);
670 container_fd
= open_dev_excl(devnum
);
671 if (container_fd
< 0) {
672 fprintf(stderr
, Name
": add failed for %s:"
673 " could not get exclusive access to container\n",
678 if (!mdmon_running(devnum
)) {
679 fprintf(stderr
, Name
": add failed for %s: mdmon not running\n",
685 sra
= sysfs_read(container_fd
, -1, 0);
687 fprintf(stderr
, Name
": add failed for %s: sysfs_read failed\n",
692 sra
->array
.level
= LEVEL_CONTAINER
;
693 /* Need to set data_offset and component_size */
694 tst
->ss
->getinfo_super(tst
, &new_mdi
);
695 new_mdi
.disk
.major
= disc
.major
;
696 new_mdi
.disk
.minor
= disc
.minor
;
697 if (sysfs_add_disk(sra
, &new_mdi
) != 0) {
698 fprintf(stderr
, Name
": add new device to external metadata"
699 " failed for %s\n", dv
->devname
);
703 ping_monitor(devnum2devname(devnum
));
706 } else if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
707 fprintf(stderr
, Name
": add new device failed for %s as %d: %s\n",
708 dv
->devname
, j
, strerror(errno
));
712 fprintf(stderr
, Name
": added %s\n", dv
->devname
);
717 if (tst
->subarray
[0]) {
718 fprintf(stderr
, Name
": Cannot remove disks from a"
719 " \'member\' array, perform this"
720 " operation on the parent container\n");
723 if (tst
->ss
->external
) {
724 /* To remove a device from a container, we must
725 * check that it isn't in use in an array.
726 * This involves looking in the 'holders'
727 * directory - there must be just one entry,
729 * To ensure that it doesn't get used as a
730 * hold spare while we are checking, we
731 * get an O_EXCL open on the container
733 int dnum
= fd2devnum(fd
);
734 lfd
= open_dev_excl(dnum
);
737 ": Cannot get exclusive access "
738 " to container - odd\n");
741 if (!sysfs_unique_holder(dnum
, stb
.st_rdev
)) {
743 ": %s is %s, cannot remove.\n",
745 errno
== EEXIST
? "still in use":
751 /* FIXME check that it is a current member */
752 err
= ioctl(fd
, HOT_REMOVE_DISK
, (unsigned long)stb
.st_rdev
);
753 if (err
&& errno
== ENODEV
) {
754 /* Old kernels rejected this if no personality
756 struct mdinfo
*sra
= sysfs_read(fd
, 0, GET_DEVS
);
757 struct mdinfo
*dv
= NULL
;
760 for ( ; dv
; dv
=dv
->next
)
761 if (dv
->disk
.major
== major(stb
.st_rdev
) &&
762 dv
->disk
.minor
== minor(stb
.st_rdev
))
765 err
= sysfs_set_str(sra
, dv
,
773 fprintf(stderr
, Name
": hot remove failed "
774 "for %s: %s\n", dnprintable
,
780 if (tst
->ss
->external
) {
782 * Before dropping our exclusive open we make an
783 * attempt at preventing mdmon from seeing an
784 * 'add' event before reconciling this 'remove'
787 char *name
= devnum2devname(fd2devnum(fd
));
790 fprintf(stderr
, Name
": unable to get container name\n");
799 fprintf(stderr
, Name
": hot removed %s\n",
803 case 'f': /* set faulty */
804 /* FIXME check current member */
805 if (ioctl(fd
, SET_DISK_FAULTY
, (unsigned long) stb
.st_rdev
)) {
806 fprintf(stderr
, Name
": set device faulty failed for %s: %s\n",
807 dnprintable
, strerror(errno
));
811 fprintf(stderr
, Name
": set %s faulty in %s\n",
812 dnprintable
, devname
);
822 /* Open any md device, and issue the RAID_AUTORUN ioctl */
824 int fd
= dev_open("9:0", O_RDONLY
);
826 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)