]>
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(int devnum
, char *devname
, char *attr
)
64 char *mdname
= devnum2devname(devnum
);
69 sprintf(fname
, "/sys/block/%s/md/", mdname
);
71 strcat(fname
, devname
);
75 fd
= open(fname
, O_RDWR
);
76 if (fd
< 0 && errno
== EACCES
)
77 fd
= open(fname
, O_RDONLY
);
82 void sysfs_init(struct mdinfo
*mdi
, int fd
, int devnum
)
87 if (ioctl(fd
, RAID_VERSION
, &vers
) != 0)
89 devnum
= fd2devnum(fd
);
91 if (devnum
== NoMdDev
)
94 sprintf(mdi
->sys_name
, "md%d", devnum
);
96 sprintf(mdi
->sys_name
, "md_d%d",
101 struct mdinfo
*sysfs_read(int fd
, int devnum
, unsigned long options
)
103 char fname
[PATH_MAX
];
112 sra
= malloc(sizeof(*sra
));
115 memset(sra
, 0, sizeof(*sra
));
116 sysfs_init(sra
, fd
, devnum
);
117 if (sra
->sys_name
[0] == 0) {
122 sprintf(fname
, "/sys/block/%s/md/", sra
->sys_name
);
123 base
= fname
+ strlen(fname
);
126 if (options
& GET_VERSION
) {
127 strcpy(base
, "metadata_version");
128 if (load_sys(fname
, buf
))
130 if (strncmp(buf
, "none", 4) == 0) {
131 sra
->array
.major_version
=
132 sra
->array
.minor_version
= -1;
133 strcpy(sra
->text_version
, "");
134 } else if (strncmp(buf
, "external:", 9) == 0) {
135 sra
->array
.major_version
= -1;
136 sra
->array
.minor_version
= -2;
137 strcpy(sra
->text_version
, buf
+9);
140 &sra
->array
.major_version
,
141 &sra
->array
.minor_version
);
142 strcpy(sra
->text_version
, buf
);
145 if (options
& GET_LEVEL
) {
146 strcpy(base
, "level");
147 if (load_sys(fname
, buf
))
149 sra
->array
.level
= map_name(pers
, buf
);
151 if (options
& GET_LAYOUT
) {
152 strcpy(base
, "layout");
153 if (load_sys(fname
, buf
))
155 sra
->array
.layout
= strtoul(buf
, NULL
, 0);
157 if (options
& GET_DISKS
) {
158 strcpy(base
, "raid_disks");
159 if (load_sys(fname
, buf
))
161 sra
->array
.raid_disks
= strtoul(buf
, NULL
, 0);
163 if (options
& GET_DEGRADED
) {
164 strcpy(base
, "degraded");
165 if (load_sys(fname
, buf
))
167 sra
->array
.failed_disks
= strtoul(buf
, NULL
, 0);
169 if (options
& GET_COMPONENT
) {
170 strcpy(base
, "component_size");
171 if (load_sys(fname
, buf
))
173 sra
->component_size
= strtoull(buf
, NULL
, 0);
174 /* sysfs reports "K", but we want sectors */
175 sra
->component_size
*= 2;
177 if (options
& GET_CHUNK
) {
178 strcpy(base
, "chunk_size");
179 if (load_sys(fname
, buf
))
181 sra
->array
.chunk_size
= strtoul(buf
, NULL
, 0);
183 if (options
& GET_CACHE
) {
184 strcpy(base
, "stripe_cache_size");
185 if (load_sys(fname
, buf
))
187 sra
->cache_size
= strtoul(buf
, NULL
, 0);
189 if (options
& GET_MISMATCH
) {
190 strcpy(base
, "mismatch_cnt");
191 if (load_sys(fname
, buf
))
193 sra
->mismatch_cnt
= strtoul(buf
, NULL
, 0);
195 if (options
& GET_SAFEMODE
) {
202 strcpy(base
, "safe_mode_delay");
203 if (load_sys(fname
, buf
))
206 /* remove a period, and count digits after it */
208 for (i
= 0; i
< len
; i
++) {
210 if (isdigit(buf
[i
])) {
215 } else if (buf
[i
] == '.') {
220 msec
= strtoul(buf
, NULL
, 10);
221 msec
= (msec
* 1000) / scale
;
222 sra
->safe_mode_delay
= msec
;
225 if (! (options
& GET_DEVS
))
228 /* Get all the devices as well */
230 dir
= opendir(fname
);
233 sra
->array
.spare_disks
= 0;
235 while ((de
= readdir(dir
)) != NULL
) {
237 if (de
->d_ino
== 0 ||
238 strncmp(de
->d_name
, "dev-", 4) != 0)
240 strcpy(base
, de
->d_name
);
241 dbase
= base
+ strlen(base
);
244 dev
= malloc(sizeof(*dev
));
248 /* Always get slot, major, minor */
249 strcpy(dbase
, "slot");
250 if (load_sys(fname
, buf
)) {
251 /* hmm... unable to read 'slot' maybe the device
254 strcpy(dbase
, "block");
255 if (readlink(fname
, buf
, sizeof(buf
)) < 0 &&
256 errno
!= ENAMETOOLONG
) {
257 /* ...yup device is gone */
261 /* slot is unreadable but 'block' link
262 * still intact... something bad is happening
270 strcpy(dev
->sys_name
, de
->d_name
);
271 dev
->disk
.raid_disk
= strtoul(buf
, &ep
, 10);
272 if (*ep
) dev
->disk
.raid_disk
= -1;
274 strcpy(dbase
, "block/dev");
275 if (load_sys(fname
, buf
)) {
277 if (options
& SKIP_GONE_DEVS
)
282 sscanf(buf
, "%d:%d", &dev
->disk
.major
, &dev
->disk
.minor
);
284 /* special case check for block devices that can go 'offline' */
285 if (options
& SKIP_GONE_DEVS
) {
286 strcpy(dbase
, "block/device/state");
287 if (load_sys(fname
, buf
) == 0 &&
288 strncmp(buf
, "offline", 7) == 0) {
294 /* finally add this disk to the array */
295 dev
->next
= sra
->devs
;
298 if (options
& GET_OFFSET
) {
299 strcpy(dbase
, "offset");
300 if (load_sys(fname
, buf
))
302 dev
->data_offset
= strtoull(buf
, NULL
, 0);
304 if (options
& GET_SIZE
) {
305 strcpy(dbase
, "size");
306 if (load_sys(fname
, buf
))
308 dev
->component_size
= strtoull(buf
, NULL
, 0) * 2;
310 if (options
& GET_STATE
) {
312 strcpy(dbase
, "state");
313 if (load_sys(fname
, buf
))
315 if (strstr(buf
, "in_sync"))
316 dev
->disk
.state
|= (1<<MD_DISK_SYNC
);
317 if (strstr(buf
, "faulty"))
318 dev
->disk
.state
|= (1<<MD_DISK_FAULTY
);
319 if (dev
->disk
.state
== 0)
320 sra
->array
.spare_disks
++;
322 if (options
& GET_ERROR
) {
323 strcpy(buf
, "errors");
324 if (load_sys(fname
, buf
))
326 dev
->errors
= strtoul(buf
, NULL
, 0);
339 int sysfs_attr_match(const char *attr
, const char *str
)
341 /* See if attr, read from a sysfs file, matches
342 * str. They must either be the same, or attr can
343 * have a trailing newline or comma
345 while (*attr
&& *str
&& *attr
== *str
) {
350 if (*str
|| (*attr
&& *attr
!= ',' && *attr
!= '\n'))
355 int sysfs_match_word(const char *word
, char **list
)
358 for (n
=0; list
[n
]; n
++)
359 if (sysfs_attr_match(word
, list
[n
]))
364 unsigned long long get_component_size(int fd
)
366 /* Find out the component size of the array.
367 * We cannot trust GET_ARRAY_INFO ioctl as it's
368 * size field is only 32bits.
369 * So look in /sys/block/mdXXX/md/component_size
371 * This returns in units of sectors.
376 if (fstat(fd
, &stb
)) return 0;
377 if (major(stb
.st_rdev
) != get_mdp_major())
378 sprintf(fname
, "/sys/block/md%d/md/component_size",
379 (int)minor(stb
.st_rdev
));
381 sprintf(fname
, "/sys/block/md_d%d/md/component_size",
382 (int)minor(stb
.st_rdev
)>>MdpMinorShift
);
383 fd
= open(fname
, O_RDONLY
);
386 n
= read(fd
, fname
, sizeof(fname
));
388 if (n
== sizeof(fname
))
391 return strtoull(fname
, NULL
, 10) * 2;
394 int sysfs_set_str(struct mdinfo
*sra
, struct mdinfo
*dev
,
395 char *name
, char *val
)
401 sprintf(fname
, "/sys/block/%s/md/%s/%s",
402 sra
->sys_name
, dev
?dev
->sys_name
:"", name
);
403 fd
= open(fname
, O_WRONLY
);
406 n
= write(fd
, val
, strlen(val
));
408 if (n
!= strlen(val
)) {
409 dprintf(Name
": failed to write '%s' to '%s' (%s)\n",
410 val
, fname
, strerror(errno
));
416 int sysfs_set_num(struct mdinfo
*sra
, struct mdinfo
*dev
,
417 char *name
, unsigned long long val
)
420 sprintf(valstr
, "%llu", val
);
421 return sysfs_set_str(sra
, dev
, name
, valstr
);
424 int sysfs_uevent(struct mdinfo
*sra
, char *event
)
430 sprintf(fname
, "/sys/block/%s/uevent",
432 fd
= open(fname
, O_WRONLY
);
435 n
= write(fd
, event
, strlen(event
));
440 int sysfs_get_fd(struct mdinfo
*sra
, struct mdinfo
*dev
,
446 sprintf(fname
, "/sys/block/%s/md/%s/%s",
447 sra
->sys_name
, dev
?dev
->sys_name
:"", name
);
448 fd
= open(fname
, O_RDWR
);
450 fd
= open(fname
, O_RDONLY
);
454 int sysfs_fd_get_ll(int fd
, unsigned long long *val
)
461 n
= read(fd
, buf
, sizeof(buf
));
465 *val
= strtoull(buf
, &ep
, 0);
466 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n' && *ep
!= ' '))
471 int sysfs_get_ll(struct mdinfo
*sra
, struct mdinfo
*dev
,
472 char *name
, unsigned long long *val
)
477 fd
= sysfs_get_fd(sra
, dev
, name
);
480 n
= sysfs_fd_get_ll(fd
, val
);
485 int sysfs_fd_get_str(int fd
, char *val
, int size
)
490 n
= read(fd
, val
, size
);
497 int sysfs_get_str(struct mdinfo
*sra
, struct mdinfo
*dev
,
498 char *name
, char *val
, int size
)
503 fd
= sysfs_get_fd(sra
, dev
, name
);
506 n
= sysfs_fd_get_str(fd
, val
, size
);
511 int sysfs_set_safemode(struct mdinfo
*sra
, unsigned long ms
)
520 sprintf(delay
, "%ld.%03ld\n", sec
, msec
);
521 /* this '\n' ^ needed for kernels older than 2.6.28 */
522 return sysfs_set_str(sra
, NULL
, "safe_mode_delay", delay
);
525 int sysfs_set_array(struct mdinfo
*info
, int vers
)
531 if (info
->array
.major_version
== -1 &&
532 info
->array
.minor_version
== -2) {
533 strcat(strcpy(ver
, "external:"), info
->text_version
);
535 if ((vers
% 100) < 2 ||
536 sysfs_set_str(info
, NULL
, "metadata_version",
538 fprintf(stderr
, Name
": This kernel does not "
539 "support external metadata.\n");
543 if (info
->array
.level
< 0)
544 return 0; /* FIXME */
545 rv
|= sysfs_set_str(info
, NULL
, "level",
546 map_num(pers
, info
->array
.level
));
547 rv
|= sysfs_set_num(info
, NULL
, "raid_disks", info
->array
.raid_disks
);
548 rv
|= sysfs_set_num(info
, NULL
, "chunk_size", info
->array
.chunk_size
);
549 rv
|= sysfs_set_num(info
, NULL
, "layout", info
->array
.layout
);
550 rv
|= sysfs_set_num(info
, NULL
, "component_size", info
->component_size
/2);
551 if (info
->custom_array_size
) {
554 rc
= sysfs_set_num(info
, NULL
, "array_size",
555 info
->custom_array_size
/2);
556 if (rc
&& errno
== ENOENT
) {
557 fprintf(stderr
, Name
": This kernel does not "
558 "have the md/array_size attribute, "
559 "the array may be larger than expected\n");
565 if (info
->array
.level
> 0)
566 rv
|= sysfs_set_num(info
, NULL
, "resync_start", info
->resync_start
);
570 int sysfs_add_disk(struct mdinfo
*sra
, struct mdinfo
*sd
, int resume
)
577 sprintf(dv
, "%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
578 rv
= sysfs_set_str(sra
, NULL
, "new_dev", dv
);
582 memset(nm
, 0, sizeof(nm
));
583 sprintf(dv
, "/sys/dev/block/%d:%d", sd
->disk
.major
, sd
->disk
.minor
);
584 rv
= readlink(dv
, nm
, sizeof(nm
));
588 dname
= strrchr(nm
, '/');
590 strcpy(sd
->sys_name
, "dev-");
591 strcpy(sd
->sys_name
+4, dname
);
593 /* test write to see if 'recovery_start' is available */
594 if (resume
&& sd
->recovery_start
< MaxSector
&&
595 sysfs_set_num(sra
, sd
, "recovery_start", 0)) {
596 sysfs_set_str(sra
, sd
, "state", "remove");
600 rv
= sysfs_set_num(sra
, sd
, "offset", sd
->data_offset
);
601 rv
|= sysfs_set_num(sra
, sd
, "size", (sd
->component_size
+1) / 2);
602 if (sra
->array
.level
!= LEVEL_CONTAINER
) {
603 if (sd
->recovery_start
== MaxSector
)
604 /* This can correctly fail if array isn't started,
605 * yet, so just ignore status for now.
607 sysfs_set_str(sra
, sd
, "state", "insync");
608 rv
|= sysfs_set_num(sra
, sd
, "slot", sd
->disk
.raid_disk
);
610 sysfs_set_num(sra
, sd
, "recovery_start", sd
->recovery_start
);
616 int sysfs_disk_to_sg(int fd
)
618 /* from an open block device, try find and open its corresponding
619 * scsi_generic interface
624 char sg_major_minor
[8];
628 int major
, minor
, rv
;
633 snprintf(path
, sizeof(path
), "/sys/dev/block/%d:%d/device",
634 major(st
.st_rdev
), minor(st
.st_rdev
));
642 if (strncmp("scsi_generic:", de
->d_name
,
643 strlen("scsi_generic:")) == 0)
652 snprintf(sg_path
, sizeof(sg_path
), "%s/%s/dev", path
, de
->d_name
);
653 fd
= open(sg_path
, O_RDONLY
);
657 rv
= read(fd
, sg_major_minor
, sizeof(sg_major_minor
));
662 sg_major_minor
[rv
- 1] = '\0';
664 c
= strchr(sg_major_minor
, ':');
667 major
= strtol(sg_major_minor
, NULL
, 10);
668 minor
= strtol(c
, NULL
, 10);
669 snprintf(path
, sizeof(path
), "/dev/.tmp.md.%d:%d:%d",
670 (int) getpid(), major
, minor
);
671 if (mknod(path
, S_IFCHR
|0600, makedev(major
, minor
))==0) {
672 fd
= open(path
, O_RDONLY
);
681 int sysfs_disk_to_scsi_id(int fd
, __u32
*id
)
683 /* from an open block device, try to retrieve it scsi_id */
693 snprintf(path
, sizeof(path
), "/sys/dev/block/%d:%d/device",
694 major(st
.st_rdev
), minor(st
.st_rdev
));
702 if (strncmp("scsi_disk:", de
->d_name
,
703 strlen("scsi_disk:")) == 0)
712 c1
= strchr(de
->d_name
, ':');
714 c2
= strchr(c1
, ':');
716 *id
= strtol(c1
, NULL
, 10) << 24; /* host */
718 c2
= strchr(c1
, ':');
720 *id
|= strtol(c1
, NULL
, 10) << 16; /* channel */
722 c2
= strchr(c1
, ':');
724 *id
|= strtol(c1
, NULL
, 10) << 8; /* lun */
726 *id
|= strtol(c1
, NULL
, 10); /* id */
732 int sysfs_unique_holder(int devnum
, long rdev
)
734 /* Check that devnum is a holder of rdev,
735 * and is the only holder.
736 * we should be locked against races by
737 * an O_EXCL on devnum
744 sprintf(dirname
, "/sys/dev/block/%d:%d/holders",
745 major(rdev
), minor(rdev
));
746 dir
= opendir(dirname
);
751 while ((de
= readdir(dir
)) != NULL
) {
760 if (de
->d_name
[0] == '.')
762 strcpy(dirname
+l
, "/");
763 strcat(dirname
+l
, de
->d_name
);
764 strcat(dirname
+l
, "/dev");
765 fd
= open(dirname
, O_RDONLY
);
770 n
= read(fd
, buf
, sizeof(buf
)-1);
773 if (sscanf(buf
, "%d:%d%c", &mj
, &mn
, &c
) != 3 ||
796 static char *clean_states
[] = {
797 "clear", "inactive", "readonly", "read-auto", "clean", NULL
};
799 int WaitClean(char *dev
, int sock
, int verbose
)
806 fd
= open(dev
, O_RDONLY
);
809 fprintf(stderr
, Name
": Couldn't open %s: %s\n", dev
, strerror(errno
));
813 devnum
= fd2devnum(fd
);
814 mdi
= sysfs_read(fd
, devnum
, GET_VERSION
|GET_LEVEL
|GET_SAFEMODE
);
817 fprintf(stderr
, Name
": Failed to read sysfs attributes for "
823 switch(mdi
->array
.level
) {
825 case LEVEL_MULTIPATH
:
827 /* safemode delay is irrelevant for these levels */
832 /* for internal metadata the kernel handles the final clean
833 * transition, containers can never be dirty
835 if (!is_subarray(mdi
->text_version
))
838 /* safemode disabled ? */
839 if (mdi
->safe_mode_delay
== 0)
843 int state_fd
= sysfs_open(fd2devnum(fd
), NULL
, "array_state");
848 /* minimize the safe_mode_delay and prepare to wait up to 5s
849 * for writes to quiesce
851 sysfs_set_safemode(mdi
, 1);
855 /* give mdmon a chance to checkpoint resync */
856 sysfs_set_str(mdi
, NULL
, "sync_action", "idle");
860 /* wait for array_state to be clean */
862 rv
= read(state_fd
, buf
, sizeof(buf
));
865 if (sysfs_match_word(buf
, clean_states
) <= 4)
867 FD_SET(state_fd
, &fds
);
868 rv
= select(state_fd
+ 1, NULL
, NULL
, &fds
, &tm
);
869 if (rv
< 0 && errno
!= EINTR
)
871 lseek(state_fd
, 0, SEEK_SET
);
875 else if (fping_monitor(sock
) == 0 ||
876 ping_monitor(mdi
->text_version
) == 0) {
877 /* we need to ping to close the window between array
878 * state transitioning to clean and the metadata being
885 fprintf(stderr
, Name
": Error waiting for %s to be clean\n",
888 /* restore the original safe_mode_delay */
889 sysfs_set_safemode(mdi
, mdi
->safe_mode_delay
);
898 #endif /* MDASSEMBLE */