]>
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
);
199 if (fd
>= 0 && ioctl(fd
, STOP_ARRAY
, NULL
)) {
202 ": failed to stop array %s: %s\n",
203 devname
, strerror(errno
));
205 fprintf(stderr
, "Perhaps a running "
206 "process, mounted filesystem "
207 "or active volume group?\n");
213 /* prior to 2.6.28, KOBJ_CHANGE was not sent when an md array
214 * was stopped, so We'll do it here just to be sure.
217 sysfs_uevent(mdi
, "change");
220 fprintf(stderr
, Name
": stopped %s\n", devname
);
221 if (fd
>= 0 && fstat(fd
, &stb
) == 0) {
223 if (major(stb
.st_rdev
) == MD_MAJOR
)
224 devnum
= minor(stb
.st_rdev
);
226 devnum
= -1-(minor(stb
.st_rdev
)>>6);
227 map_delete(&map
, devnum
);
235 int Manage_resize(char *devname
, int fd
, long long size
, int raid_disks
)
237 mdu_array_info_t info
;
238 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
239 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
240 devname
, strerror(errno
));
246 info
.raid_disks
= raid_disks
;
247 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
248 fprintf(stderr
, Name
": Cannot set device size/shape for %s: %s\n",
249 devname
, strerror(errno
));
255 int Manage_reconfig(char *devname
, int fd
, int layout
)
257 mdu_array_info_t info
;
258 if (ioctl(fd
, GET_ARRAY_INFO
, &info
) != 0) {
259 fprintf(stderr
, Name
": Cannot get array information for %s: %s\n",
260 devname
, strerror(errno
));
263 info
.layout
= layout
;
264 printf("layout set to %d\n", info
.layout
);
265 if (ioctl(fd
, SET_ARRAY_INFO
, &info
) != 0) {
266 fprintf(stderr
, Name
": Cannot set layout for %s: %s\n",
267 devname
, strerror(errno
));
273 int Manage_subdevs(char *devname
, int fd
,
274 mddev_dev_t devlist
, int verbose
)
276 /* do something to each dev.
278 * 'a' - add the device
280 * If that fails EINVAL, try ADD_NEW_DISK
281 * 'r' - remove the device HOT_REMOVE_DISK
282 * device can be 'faulty' or 'detached' in which case all
283 * matching devices are removed.
284 * 'f' - set the device faulty SET_DISK_FAULTY
285 * device can be 'detached' in which case any device that
286 * is inaccessible will be marked faulty.
288 mdu_array_info_t array
;
289 mdu_disk_info_t disc
;
290 unsigned long long array_size
;
291 mddev_dev_t dv
, next
= NULL
;
295 struct supertype
*st
, *tst
;
300 if (ioctl(fd
, GET_ARRAY_INFO
, &array
)) {
301 fprintf(stderr
, Name
": cannot get array info for %s\n",
306 /* array.size is only 32 bit and may be truncated.
307 * So read from sysfs if possible, and record number of sectors
310 array_size
= get_component_size(fd
);
312 array_size
= array
.size
* 2;
314 tst
= super_by_fd(fd
);
316 fprintf(stderr
, Name
": unsupport array - version %d.%d\n",
317 array
.major_version
, array
.minor_version
);
321 for (dv
= devlist
, j
=0 ; dv
; dv
= next
, j
= jnext
) {
322 unsigned long long ldsize
;
324 char *dnprintable
= dv
->devname
;
330 if (strcmp(dv
->devname
, "failed")==0 ||
331 strcmp(dv
->devname
, "faulty")==0) {
332 if (dv
->disposition
!= 'r') {
333 fprintf(stderr
, Name
": %s only meaningful "
334 "with -r, not -%c\n",
335 dv
->devname
, dv
->disposition
);
338 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
340 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
342 if (disc
.major
== 0 && disc
.minor
== 0)
344 if ((disc
.state
& 1) == 0) /* faulty */
346 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
349 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
350 dnprintable
= dvname
;
355 } else if (strcmp(dv
->devname
, "detached") == 0) {
356 if (dv
->disposition
!= 'r' && dv
->disposition
!= 'f') {
357 fprintf(stderr
, Name
": %s only meaningful "
358 "with -r of -f, not -%c\n",
359 dv
->devname
, dv
->disposition
);
362 for (; j
< array
.raid_disks
+ array
.nr_disks
; j
++) {
365 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
367 if (disc
.major
== 0 && disc
.minor
== 0)
369 sprintf(dvname
,"%d:%d", disc
.major
, disc
.minor
);
370 sfd
= dev_open(dvname
, O_RDONLY
);
375 if (dv
->disposition
== 'f' &&
376 (disc
.state
& 1) == 1) /* already faulty */
380 stb
.st_rdev
= makedev(disc
.major
, disc
.minor
);
383 dnprintable
= dvname
;
391 if (stat(dv
->devname
, &stb
)) {
392 fprintf(stderr
, Name
": cannot find %s: %s\n",
393 dv
->devname
, strerror(errno
));
396 if ((stb
.st_mode
& S_IFMT
) != S_IFBLK
) {
397 fprintf(stderr
, Name
": %s is not a "
403 switch(dv
->disposition
){
405 fprintf(stderr
, Name
": internal error - devmode[%s]=%d\n",
406 dv
->devname
, dv
->disposition
);
410 if (tst
->subarray
[0]) {
411 fprintf(stderr
, Name
": Cannot add disks to a"
412 " \'member\' array, perform this"
413 " operation on the parent container\n");
416 /* Make sure it isn't in use (in 2.6 or later) */
417 tfd
= open(dv
->devname
, O_RDONLY
|O_EXCL
|O_DIRECT
);
419 fprintf(stderr
, Name
": Cannot open %s: %s\n",
420 dv
->devname
, strerror(errno
));
423 remove_partitions(tfd
);
427 if (array
.not_persistent
==0)
428 st
->ss
->load_super(st
, tfd
, NULL
);
430 if (!get_dev_size(tfd
, dv
->devname
, &ldsize
)) {
437 if (!tst
->ss
->external
&&
438 array
.major_version
== 0 &&
439 md_get_version(fd
)%100 < 2) {
440 if (ioctl(fd
, HOT_ADD_DISK
,
441 (unsigned long)stb
.st_rdev
)==0) {
443 fprintf(stderr
, Name
": hot added %s\n",
448 fprintf(stderr
, Name
": hot add failed for %s: %s\n",
449 dv
->devname
, strerror(errno
));
453 if (array
.not_persistent
== 0 || tst
->ss
->external
) {
455 /* need to find a sample superblock to copy, and
456 * a spare slot to use.
457 * For 'external' array (well, container based),
458 * We can just load the metadata for the array.
460 if (tst
->ss
->external
) {
461 tst
->ss
->load_super(tst
, fd
, NULL
);
462 } else for (j
= 0; j
< tst
->max_devs
; j
++) {
466 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
468 if (disc
.major
==0 && disc
.minor
==0)
470 if ((disc
.state
& 4)==0) continue; /* sync */
471 /* Looks like a good device to try */
472 dev
= map_dev(disc
.major
, disc
.minor
, 1);
474 dfd
= dev_open(dev
, O_RDONLY
);
475 if (dfd
< 0) continue;
476 if (tst
->ss
->load_super(tst
, dfd
,
484 /* FIXME this is a bad test to be using */
486 fprintf(stderr
, Name
": cannot find valid superblock in this array - HELP\n");
490 /* Make sure device is large enough */
491 if (tst
->ss
->avail_size(tst
, ldsize
/512) <
493 fprintf(stderr
, Name
": %s not large enough to join array\n",
498 /* Possibly this device was recently part of the array
499 * and was temporarily removed, and is now being re-added.
500 * If so, we can simply re-add it.
502 tst
->ss
->uuid_from_super(tst
, duuid
);
504 /* re-add doesn't work for version-1 superblocks
507 if (array
.major_version
== 1 &&
508 get_linux_version() <= 2006018)
511 st
->ss
->uuid_from_super(st
, ouuid
);
512 if (memcmp(duuid
, ouuid
, sizeof(ouuid
))==0) {
513 /* looks close enough for now. Kernel
514 * will worry about whether a bitmap
515 * based reconstruction is possible.
518 st
->ss
->getinfo_super(st
, &mdi
);
519 disc
.major
= major(stb
.st_rdev
);
520 disc
.minor
= minor(stb
.st_rdev
);
521 disc
.number
= mdi
.disk
.number
;
522 disc
.raid_disk
= mdi
.disk
.raid_disk
;
523 disc
.state
= mdi
.disk
.state
;
524 if (dv
->writemostly
== 1)
525 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
526 if (dv
->writemostly
== 2)
527 disc
.state
&= ~(1 << MD_DISK_WRITEMOSTLY
);
528 if (ioctl(fd
, ADD_NEW_DISK
, &disc
) == 0) {
530 fprintf(stderr
, Name
": re-added %s\n", dv
->devname
);
533 /* fall back on normal-add */
537 /* non-persistent. Must ensure that new drive
538 * is at least array.size big.
540 if (ldsize
/512 < array_size
) {
541 fprintf(stderr
, Name
": %s not large enough to join array\n",
546 /* in 2.6.17 and earlier, version-1 superblocks won't
547 * use the number we write, but will choose a free number.
548 * we must choose the same free number, which requires
549 * starting at 'raid_disks' and counting up
551 for (j
= array
.raid_disks
; j
< tst
->max_devs
; j
++) {
553 if (ioctl(fd
, GET_DISK_INFO
, &disc
))
555 if (disc
.major
==0 && disc
.minor
==0)
557 if (disc
.state
& 8) /* removed */
560 disc
.major
= major(stb
.st_rdev
);
561 disc
.minor
= minor(stb
.st_rdev
);
564 if (array
.not_persistent
==0 || tst
->ss
->external
) {
566 if (dv
->writemostly
== 1)
567 disc
.state
|= 1 << MD_DISK_WRITEMOSTLY
;
568 dfd
= open(dv
->devname
, O_RDWR
| O_EXCL
|O_DIRECT
);
569 tst
->ss
->add_to_super(tst
, &disc
, dfd
,
571 /* write_init_super will close 'dfd' */
572 if (tst
->ss
->external
)
573 /* mdmon will write the metadata */
575 else if (tst
->ss
->write_init_super(tst
))
577 } else if (dv
->re_add
) {
578 /* this had better be raid1.
579 * As we are "--re-add"ing we must find a spare slot
582 char *used
= malloc(array
.raid_disks
);
583 memset(used
, 0, array
.raid_disks
);
584 for (j
=0; j
< tst
->max_devs
; j
++) {
585 mdu_disk_info_t disc2
;
587 if (ioctl(fd
, GET_DISK_INFO
, &disc2
))
589 if (disc2
.major
==0 && disc2
.minor
==0)
591 if (disc2
.state
& 8) /* removed */
593 if (disc2
.raid_disk
< 0)
595 if (disc2
.raid_disk
> array
.raid_disks
)
597 used
[disc2
.raid_disk
] = 1;
599 for (j
=0 ; j
<array
.raid_disks
; j
++)
602 disc
.state
|= (1<<MD_DISK_SYNC
);
606 if (dv
->writemostly
== 1)
607 disc
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
608 if (tst
->ss
->external
) {
609 /* add a disk to an external metadata container
610 * only if mdmon is around to see it
612 struct mdinfo new_mdi
;
615 int devnum
= fd2devnum(fd
);
617 container_fd
= open_dev_excl(devnum
);
618 if (container_fd
< 0) {
619 fprintf(stderr
, Name
": add failed for %s:"
620 " could not get exclusive access to container\n",
625 if (!mdmon_running(devnum
)) {
626 fprintf(stderr
, Name
": add failed for %s: mdmon not running\n",
632 sra
= sysfs_read(container_fd
, -1, 0);
634 fprintf(stderr
, Name
": add failed for %s: sysfs_read failed\n",
639 sra
->array
.level
= LEVEL_CONTAINER
;
640 /* Need to set data_offset and component_size */
641 tst
->ss
->getinfo_super(tst
, &new_mdi
);
642 new_mdi
.disk
.major
= disc
.major
;
643 new_mdi
.disk
.minor
= disc
.minor
;
644 if (sysfs_add_disk(sra
, &new_mdi
) != 0) {
645 fprintf(stderr
, Name
": add new device to external metadata"
646 " failed for %s\n", dv
->devname
);
650 ping_monitor(devnum2devname(devnum
));
653 } else if (ioctl(fd
, ADD_NEW_DISK
, &disc
)) {
654 fprintf(stderr
, Name
": add new device failed for %s as %d: %s\n",
655 dv
->devname
, j
, strerror(errno
));
659 fprintf(stderr
, Name
": added %s\n", dv
->devname
);
664 if (tst
->subarray
[0]) {
665 fprintf(stderr
, Name
": Cannot remove disks from a"
666 " \'member\' array, perform this"
667 " operation on the parent container\n");
670 if (tst
->ss
->external
) {
671 /* To remove a device from a container, we must
672 * check that it isn't in use in an array.
673 * This involves looking in the 'holders'
674 * directory - there must be just one entry,
676 * To ensure that it doesn't get used as a
677 * hold spare while we are checking, we
678 * get an O_EXCL open on the container
680 int dnum
= fd2devnum(fd
);
681 lfd
= open_dev_excl(dnum
);
684 ": Cannot get exclusive access "
685 " to container - odd\n");
688 if (!sysfs_unique_holder(dnum
, stb
.st_rdev
)) {
690 ": %s is %s, cannot remove.\n",
692 errno
== EEXIST
? "still in use":
698 /* FIXME check that it is a current member */
699 err
= ioctl(fd
, HOT_REMOVE_DISK
, (unsigned long)stb
.st_rdev
);
700 if (err
&& errno
== ENODEV
) {
701 /* Old kernels rejected this if no personality
703 struct mdinfo
*sra
= sysfs_read(fd
, 0, GET_DEVS
);
704 struct mdinfo
*dv
= NULL
;
707 for ( ; dv
; dv
=dv
->next
)
708 if (dv
->disk
.major
== major(stb
.st_rdev
) &&
709 dv
->disk
.minor
== minor(stb
.st_rdev
))
712 err
= sysfs_set_str(sra
, dv
,
720 fprintf(stderr
, Name
": hot remove failed "
721 "for %s: %s\n", dnprintable
,
727 if (tst
->ss
->external
) {
729 * Before dropping our exclusive open we make an
730 * attempt at preventing mdmon from seeing an
731 * 'add' event before reconciling this 'remove'
734 char *name
= devnum2devname(fd2devnum(fd
));
737 fprintf(stderr
, Name
": unable to get container name\n");
746 fprintf(stderr
, Name
": hot removed %s\n",
750 case 'f': /* set faulty */
751 /* FIXME check current member */
752 if (ioctl(fd
, SET_DISK_FAULTY
, (unsigned long) stb
.st_rdev
)) {
753 fprintf(stderr
, Name
": set device faulty failed for %s: %s\n",
754 dnprintable
, strerror(errno
));
758 fprintf(stderr
, Name
": set %s faulty in %s\n",
759 dnprintable
, devname
);
769 /* Open any md device, and issue the RAID_AUTORUN ioctl */
771 int fd
= dev_open("9:0", O_RDONLY
);
773 if (ioctl(fd
, RAID_AUTORUN
, 0) == 0)