]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - sysfs.c
2 * sysfs - extract md related information from sysfs. Part of:
3 * mdadm - manage Linux "md" devices aka RAID arrays.
5 * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Email: <neilb@suse.de>
30 int load_sys(char *path
, char *buf
)
32 int fd
= open(path
, O_RDONLY
);
36 n
= read(fd
, buf
, 1024);
38 if (n
<0 || n
>= 1024)
41 if (n
&& buf
[n
-1] == '\n')
46 void sysfs_free(struct mdinfo
*sra
)
49 struct mdinfo
*sra2
= sra
->next
;
51 struct mdinfo
*d
= sra
->devs
;
60 int sysfs_open(char *devnm
, char *devname
, char *attr
)
65 sprintf(fname
, "/sys/block/%s/md/", devnm
);
67 strcat(fname
, devname
);
71 fd
= open(fname
, O_RDWR
);
72 if (fd
< 0 && errno
== EACCES
)
73 fd
= open(fname
, O_RDONLY
);
77 void sysfs_init(struct mdinfo
*mdi
, int fd
, char *devnm
)
82 if (ioctl(fd
, RAID_VERSION
, &vers
) != 0)
88 strcpy(mdi
->sys_name
, devnm
);
91 struct mdinfo
*sysfs_read(int fd
, char *devnm
, unsigned long options
)
98 struct mdinfo
*dev
, **devp
;
102 sra
= xcalloc(1, sizeof(*sra
));
103 sysfs_init(sra
, fd
, devnm
);
104 if (sra
->sys_name
[0] == 0) {
109 sprintf(fname
, "/sys/block/%s/md/", sra
->sys_name
);
110 base
= fname
+ strlen(fname
);
113 if (options
& GET_VERSION
) {
114 strcpy(base
, "metadata_version");
115 if (load_sys(fname
, buf
))
117 if (strncmp(buf
, "none", 4) == 0) {
118 sra
->array
.major_version
=
119 sra
->array
.minor_version
= -1;
120 strcpy(sra
->text_version
, "");
121 } else if (strncmp(buf
, "external:", 9) == 0) {
122 sra
->array
.major_version
= -1;
123 sra
->array
.minor_version
= -2;
124 strcpy(sra
->text_version
, buf
+9);
127 &sra
->array
.major_version
,
128 &sra
->array
.minor_version
);
129 strcpy(sra
->text_version
, buf
);
132 if (options
& GET_LEVEL
) {
133 strcpy(base
, "level");
134 if (load_sys(fname
, buf
))
136 sra
->array
.level
= map_name(pers
, buf
);
138 if (options
& GET_LAYOUT
) {
139 strcpy(base
, "layout");
140 if (load_sys(fname
, buf
))
142 sra
->array
.layout
= strtoul(buf
, NULL
, 0);
144 if (options
& GET_DISKS
) {
145 strcpy(base
, "raid_disks");
146 if (load_sys(fname
, buf
))
148 sra
->array
.raid_disks
= strtoul(buf
, NULL
, 0);
150 if (options
& GET_DEGRADED
) {
151 strcpy(base
, "degraded");
152 if (load_sys(fname
, buf
))
154 sra
->array
.failed_disks
= strtoul(buf
, NULL
, 0);
156 if (options
& GET_COMPONENT
) {
157 strcpy(base
, "component_size");
158 if (load_sys(fname
, buf
))
160 sra
->component_size
= strtoull(buf
, NULL
, 0);
161 /* sysfs reports "K", but we want sectors */
162 sra
->component_size
*= 2;
164 if (options
& GET_CHUNK
) {
165 strcpy(base
, "chunk_size");
166 if (load_sys(fname
, buf
))
168 sra
->array
.chunk_size
= strtoul(buf
, NULL
, 0);
170 if (options
& GET_CACHE
) {
171 strcpy(base
, "stripe_cache_size");
172 if (load_sys(fname
, buf
))
173 /* Probably level doesn't support it */
176 sra
->cache_size
= strtoul(buf
, NULL
, 0);
178 if (options
& GET_MISMATCH
) {
179 strcpy(base
, "mismatch_cnt");
180 if (load_sys(fname
, buf
))
182 sra
->mismatch_cnt
= strtoul(buf
, NULL
, 0);
184 if (options
& GET_SAFEMODE
) {
191 strcpy(base
, "safe_mode_delay");
192 if (load_sys(fname
, buf
))
195 /* remove a period, and count digits after it */
197 for (i
= 0; i
< len
; i
++) {
199 if (isdigit(buf
[i
])) {
204 } else if (buf
[i
] == '.') {
209 msec
= strtoul(buf
, NULL
, 10);
210 msec
= (msec
* 1000) / scale
;
211 sra
->safe_mode_delay
= msec
;
213 if (options
& GET_BITMAP_LOCATION
) {
214 strcpy(base
, "bitmap/location");
215 if (load_sys(fname
, buf
))
217 if (strncmp(buf
, "file", 4) == 0)
218 sra
->bitmap_offset
= 1;
219 else if (strncmp(buf
, "none", 4) == 0)
220 sra
->bitmap_offset
= 0;
221 else if (buf
[0] == '+')
222 sra
->bitmap_offset
= strtol(buf
+1, NULL
, 10);
227 if (! (options
& GET_DEVS
))
230 /* Get all the devices as well */
232 dir
= opendir(fname
);
235 sra
->array
.spare_disks
= 0;
239 while ((de
= readdir(dir
)) != NULL
) {
241 if (de
->d_ino
== 0 ||
242 strncmp(de
->d_name
, "dev-", 4) != 0)
244 strcpy(base
, de
->d_name
);
245 dbase
= base
+ strlen(base
);
248 dev
= xmalloc(sizeof(*dev
));
250 /* Always get slot, major, minor */
251 strcpy(dbase
, "slot");
252 if (load_sys(fname
, buf
)) {
253 /* hmm... unable to read 'slot' maybe the device
256 strcpy(dbase
, "block");
257 if (readlink(fname
, buf
, sizeof(buf
)) < 0 &&
258 errno
!= ENAMETOOLONG
) {
259 /* ...yup device is gone */
263 /* slot is unreadable but 'block' link
264 * still intact... something bad is happening
272 strcpy(dev
->sys_name
, de
->d_name
);
273 dev
->disk
.raid_disk
= strtoul(buf
, &ep
, 10);
274 if (*ep
) dev
->disk
.raid_disk
= -1;
276 strcpy(dbase
, "block/dev");
277 if (load_sys(fname
, buf
)) {
278 /* assume this is a stale reference to a hot
284 sra
->array
.nr_disks
++;
285 sscanf(buf
, "%d:%d", &dev
->disk
.major
, &dev
->disk
.minor
);
287 /* special case check for block devices that can go 'offline' */
288 strcpy(dbase
, "block/device/state");
289 if (load_sys(fname
, buf
) == 0 &&
290 strncmp(buf
, "offline", 7) == 0) {
295 /* finally add this disk to the array */
300 if (options
& GET_OFFSET
) {
301 strcpy(dbase
, "offset");
302 if (load_sys(fname
, buf
))
304 dev
->data_offset
= strtoull(buf
, NULL
, 0);
305 strcpy(dbase
, "new_offset");
306 if (load_sys(fname
, buf
) == 0)
307 dev
->new_data_offset
= strtoull(buf
, NULL
, 0);
309 dev
->new_data_offset
= dev
->data_offset
;
311 if (options
& GET_SIZE
) {
312 strcpy(dbase
, "size");
313 if (load_sys(fname
, buf
))
315 dev
->component_size
= strtoull(buf
, NULL
, 0) * 2;
317 if (options
& GET_STATE
) {
319 strcpy(dbase
, "state");
320 if (load_sys(fname
, buf
))
322 if (strstr(buf
, "in_sync"))
323 dev
->disk
.state
|= (1<<MD_DISK_SYNC
);
324 if (strstr(buf
, "faulty"))
325 dev
->disk
.state
|= (1<<MD_DISK_FAULTY
);
326 if (dev
->disk
.state
== 0)
327 sra
->array
.spare_disks
++;
329 if (options
& GET_ERROR
) {
330 strcpy(buf
, "errors");
331 if (load_sys(fname
, buf
))
333 dev
->errors
= strtoul(buf
, NULL
, 0);
346 int sysfs_attr_match(const char *attr
, const char *str
)
348 /* See if attr, read from a sysfs file, matches
349 * str. They must either be the same, or attr can
350 * have a trailing newline or comma
352 while (*attr
&& *str
&& *attr
== *str
) {
357 if (*str
|| (*attr
&& *attr
!= ',' && *attr
!= '\n'))
362 int sysfs_match_word(const char *word
, char **list
)
365 for (n
=0; list
[n
]; n
++)
366 if (sysfs_attr_match(word
, list
[n
]))
371 unsigned long long get_component_size(int fd
)
373 /* Find out the component size of the array.
374 * We cannot trust GET_ARRAY_INFO ioctl as it's
375 * size field is only 32bits.
376 * So look in /sys/block/mdXXX/md/component_size
378 * This returns in units of sectors.
383 if (fstat(fd
, &stb
)) return 0;
384 if (major(stb
.st_rdev
) != (unsigned)get_mdp_major())
385 sprintf(fname
, "/sys/block/md%d/md/component_size",
386 (int)minor(stb
.st_rdev
));
388 sprintf(fname
, "/sys/block/md_d%d/md/component_size",
389 (int)minor(stb
.st_rdev
)>>MdpMinorShift
);
390 fd
= open(fname
, O_RDONLY
);
393 n
= read(fd
, fname
, sizeof(fname
));
395 if (n
< 0 || n
== sizeof(fname
))
398 return strtoull(fname
, NULL
, 10) * 2;
401 int sysfs_set_str(struct mdinfo
*sra
, struct mdinfo
*dev
,
402 char *name
, char *val
)
408 sprintf(fname
, "/sys/block/%s/md/%s/%s",
409 sra
->sys_name
, dev
?dev
->sys_name
:"", name
);
410 fd
= open(fname
, O_WRONLY
);
413 n
= write(fd
, val
, strlen(val
));
415 if (n
!= strlen(val
)) {
416 dprintf(Name
": failed to write '%s' to '%s' (%s)\n",
417 val
, fname
, strerror(errno
));
423 int sysfs_set_num(struct mdinfo
*sra
, struct mdinfo
*dev
,
424 char *name
, unsigned long long val
)
427 sprintf(valstr
, "%llu", val
);
428 return sysfs_set_str(sra
, dev
, name
, valstr
);
431 int sysfs_set_num_signed(struct mdinfo
*sra
, struct mdinfo
*dev
,
432 char *name
, long long val
)
435 sprintf(valstr
, "%lli", val
);
436 return sysfs_set_str(sra
, dev
, name
, valstr
);
439 int sysfs_uevent(struct mdinfo
*sra
, char *event
)
445 sprintf(fname
, "/sys/block/%s/uevent",
447 fd
= open(fname
, O_WRONLY
);
450 n
= write(fd
, event
, strlen(event
));
452 if (n
!= (int)strlen(event
)) {
453 dprintf(Name
": failed to write '%s' to '%s' (%s)\n",
454 event
, fname
, strerror(errno
));
460 int sysfs_attribute_available(struct mdinfo
*sra
, struct mdinfo
*dev
, char *name
)
465 sprintf(fname
, "/sys/block/%s/md/%s/%s",
466 sra
->sys_name
, dev
?dev
->sys_name
:"", name
);
468 return stat(fname
, &st
) == 0;
471 int sysfs_get_fd(struct mdinfo
*sra
, struct mdinfo
*dev
,
477 sprintf(fname
, "/sys/block/%s/md/%s/%s",
478 sra
->sys_name
, dev
?dev
->sys_name
:"", name
);
479 fd
= open(fname
, O_RDWR
);
481 fd
= open(fname
, O_RDONLY
);
485 int sysfs_fd_get_ll(int fd
, unsigned long long *val
)
492 n
= read(fd
, buf
, sizeof(buf
));
496 *val
= strtoull(buf
, &ep
, 0);
497 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
502 int sysfs_get_ll(struct mdinfo
*sra
, struct mdinfo
*dev
,
503 char *name
, unsigned long long *val
)
508 fd
= sysfs_get_fd(sra
, dev
, name
);
511 n
= sysfs_fd_get_ll(fd
, val
);
516 int sysfs_fd_get_two(int fd
, unsigned long long *v1
, unsigned long long *v2
)
518 /* two numbers in this sysfs file, either
528 n
= read(fd
, buf
, sizeof(buf
));
532 *v1
= strtoull(buf
, &ep
, 0);
533 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
535 while (*ep
== ' ' || *ep
== '/' || *ep
== '(')
537 *v2
= strtoull(ep
, &ep2
, 0);
538 if (ep2
== ep
|| (*ep2
!= 0 && *ep2
!= '\n' && *ep2
!= ' ' && *ep2
!= ')')) {
545 int sysfs_get_two(struct mdinfo
*sra
, struct mdinfo
*dev
,
546 char *name
, unsigned long long *v1
, unsigned long long *v2
)
551 fd
= sysfs_get_fd(sra
, dev
, name
);
554 n
= sysfs_fd_get_two(fd
, v1
, v2
);
559 int sysfs_fd_get_str(int fd
, char *val
, int size
)
564 n
= read(fd
, val
, size
);
571 int sysfs_get_str(struct mdinfo
*sra
, struct mdinfo
*dev
,
572 char *name
, char *val
, int size
)
577 fd
= sysfs_get_fd(sra
, dev
, name
);
580 n
= sysfs_fd_get_str(fd
, val
, size
);
585 int sysfs_set_safemode(struct mdinfo
*sra
, unsigned long ms
)
594 sprintf(delay
, "%ld.%03ld\n", sec
, msec
);
595 /* this '\n' ^ needed for kernels older than 2.6.28 */
596 return sysfs_set_str(sra
, NULL
, "safe_mode_delay", delay
);
599 int sysfs_set_array(struct mdinfo
*info
, int vers
)
603 int raid_disks
= info
->array
.raid_disks
;
606 if (info
->array
.major_version
== -1 &&
607 info
->array
.minor_version
== -2) {
610 strcat(strcpy(ver
, "external:"), info
->text_version
);
612 /* meta version might already be set if we are setting
613 * new geometry for a reshape. In that case we don't
614 * want to over-write the 'readonly' flag that is
615 * stored in the metadata version. So read the current
616 * version first, and preserve the flag
618 if (sysfs_get_str(info
, NULL
, "metadata_version",
620 if (strlen(buf
) >= 9 && buf
[9] == '-')
623 if ((vers
% 100) < 2 ||
624 sysfs_set_str(info
, NULL
, "metadata_version",
626 pr_err("This kernel does not "
627 "support external metadata.\n");
631 if (info
->array
.level
< 0)
632 return 0; /* FIXME */
633 rv
|= sysfs_set_str(info
, NULL
, "level",
634 map_num(pers
, info
->array
.level
));
635 if (info
->reshape_active
&& info
->delta_disks
!= UnSet
)
636 raid_disks
-= info
->delta_disks
;
637 rv
|= sysfs_set_num(info
, NULL
, "raid_disks", raid_disks
);
638 rv
|= sysfs_set_num(info
, NULL
, "chunk_size", info
->array
.chunk_size
);
639 rv
|= sysfs_set_num(info
, NULL
, "layout", info
->array
.layout
);
640 rv
|= sysfs_set_num(info
, NULL
, "component_size", info
->component_size
/2);
641 if (info
->custom_array_size
) {
644 rc
= sysfs_set_num(info
, NULL
, "array_size",
645 info
->custom_array_size
/2);
646 if (rc
&& errno
== ENOENT
) {
647 pr_err("This kernel does not "
648 "have the md/array_size attribute, "
649 "the array may be larger than expected\n");
655 if (info
->array
.level
> 0)
656 rv
|= sysfs_set_num(info
, NULL
, "resync_start", info
->resync_start
);
658 if (info
->reshape_active
) {
659 rv
|= sysfs_set_num(info
, NULL
, "reshape_position",
660 info
->reshape_progress
);
661 rv
|= sysfs_set_num(info
, NULL
, "chunk_size", info
->new_chunk
);
662 rv
|= sysfs_set_num(info
, NULL
, "layout", info
->new_layout
);
663 rv
|= sysfs_set_num(info
, NULL
, "raid_disks",
664 info
->array
.raid_disks
);
665 /* We don't set 'new_level' here. That can only happen
666 * once the reshape completes.
672 int sysfs_add_disk(struct mdinfo
*sra
, struct mdinfo
*sd
, int resume
)
679 sprintf(dv
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
680 rv
= sysfs_set_str(sra
, NULL
, "new_dev", dv
);
684 memset(nm
, 0, sizeof(nm
));
685 dname
= devid2kname(makedev(sd
->disk
.major
, sd
->disk
.minor
));
686 strcpy(sd
->sys_name
, "dev-");
687 strcpy(sd
->sys_name
+4, dname
);
689 /* test write to see if 'recovery_start' is available */
690 if (resume
&& sd
->recovery_start
< MaxSector
&&
691 sysfs_set_num(sra
, sd
, "recovery_start", 0)) {
692 sysfs_set_str(sra
, sd
, "state", "remove");
696 rv
= sysfs_set_num(sra
, sd
, "offset", sd
->data_offset
);
697 rv
|= sysfs_set_num(sra
, sd
, "size", (sd
->component_size
+1) / 2);
698 if (sra
->array
.level
!= LEVEL_CONTAINER
) {
699 if (sd
->recovery_start
== MaxSector
)
700 /* This can correctly fail if array isn't started,
701 * yet, so just ignore status for now.
703 sysfs_set_str(sra
, sd
, "state", "insync");
704 if (sd
->disk
.raid_disk
>= 0)
705 rv
|= sysfs_set_num(sra
, sd
, "slot", sd
->disk
.raid_disk
);
707 sysfs_set_num(sra
, sd
, "recovery_start", sd
->recovery_start
);
713 int sysfs_disk_to_sg(int fd
)
715 /* from an open block device, try find and open its corresponding
716 * scsi_generic interface
721 char sg_major_minor
[8];
725 int major
, minor
, rv
;
730 snprintf(path
, sizeof(path
), "/sys/dev/block/%d:%d/device",
731 major(st
.st_rdev
), minor(st
.st_rdev
));
739 if (strncmp("scsi_generic:", de
->d_name
,
740 strlen("scsi_generic:")) == 0)
749 snprintf(sg_path
, sizeof(sg_path
), "%s/%s/dev", path
, de
->d_name
);
750 fd
= open(sg_path
, O_RDONLY
);
754 rv
= read(fd
, sg_major_minor
, sizeof(sg_major_minor
));
759 sg_major_minor
[rv
- 1] = '\0';
761 c
= strchr(sg_major_minor
, ':');
764 major
= strtol(sg_major_minor
, NULL
, 10);
765 minor
= strtol(c
, NULL
, 10);
766 snprintf(path
, sizeof(path
), "/dev/.tmp.md.%d:%d:%d",
767 (int) getpid(), major
, minor
);
768 if (mknod(path
, S_IFCHR
|0600, makedev(major
, minor
))==0) {
769 fd
= open(path
, O_RDONLY
);
778 int sysfs_disk_to_scsi_id(int fd
, __u32
*id
)
780 /* from an open block device, try to retrieve it scsi_id */
785 int host
, bus
, target
, lun
;
790 snprintf(path
, sizeof(path
), "/sys/dev/block/%d:%d/device/scsi_device",
791 major(st
.st_rdev
), minor(st
.st_rdev
));
797 for (de
= readdir(dir
); de
; de
= readdir(dir
)) {
800 if (de
->d_type
!= DT_DIR
)
803 count
= sscanf(de
->d_name
, "%d:%d:%d:%d", &host
, &bus
, &target
, &lun
);
812 *id
= (host
<< 24) | (bus
<< 16) | (target
<< 8) | (lun
<< 0);
816 int sysfs_unique_holder(char *devnm
, long rdev
)
818 /* Check that devnm is a holder of rdev,
819 * and is the only holder.
820 * we should be locked against races by
823 * 0 - not unique, not even a holder
824 * 1 - unique, this is the only holder.
825 * 2/3 - not unique, there is another holder
826 * -1 - error, cannot find the holders
833 sprintf(dirname
, "/sys/dev/block/%d:%d/holders",
834 major(rdev
), minor(rdev
));
835 dir
= opendir(dirname
);
839 while ((de
= readdir(dir
)) != NULL
) {
846 if (de
->d_name
[0] == '.')
848 strcpy(dirname
+l
, "/");
849 strcat(dirname
+l
, de
->d_name
);
850 n
= readlink(dirname
, buf
, sizeof(buf
)-1);
854 sl
= strrchr(buf
, '/');
859 if (strcmp(devnm
, sl
) == 0)
868 int sysfs_freeze_array(struct mdinfo
*sra
)
870 /* Try to freeze resync/rebuild on this array/container.
871 * Return -1 if the array is busy,
872 * return 0 if this kernel doesn't support 'frozen'
873 * return 1 if it worked.
877 if (!sysfs_attribute_available(sra
, NULL
, "sync_action"))
878 return 1; /* no sync_action == frozen */
879 if (sysfs_get_str(sra
, NULL
, "sync_action", buf
, 20) <= 0)
881 if (strcmp(buf
, "frozen\n") == 0)
884 if (strcmp(buf
, "idle\n") != 0 && strcmp(buf
, "recover\n") != 0)
886 if (sysfs_set_str(sra
, NULL
, "sync_action", "frozen") < 0)
891 int sysfs_wait(int fd
, int *msec
)
893 /* Wait up to '*msec' for fd to have an exception condition.
894 * if msec == NULL, wait indefinitely.
901 n
= select(fd
+1, NULL
, NULL
, &fds
, NULL
);
905 struct timeval start
, end
, tv
;
906 gettimeofday(&start
, NULL
);
909 tv
.tv_usec
= (*msec
)*1000;
911 tv
.tv_sec
= (*msec
)/1000;
914 n
= select(fd
+1, NULL
, NULL
, &fds
, &tv
);
915 gettimeofday(&end
, NULL
);
916 end
.tv_sec
-= start
.tv_sec
;
917 *msec
-= (end
.tv_sec
* 1000 + end
.tv_usec
/1000
918 - start
.tv_usec
/1000) + 1;