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 #define MAX_SYSFS_PATH_LEN 120
32 int load_sys(char *path, char *buf, int len)
34 int fd = open(path, O_RDONLY);
38 n = read(fd, buf, len);
43 if (n && buf[n-1] == '\n')
48 void sysfs_free(struct mdinfo *sra)
51 struct mdinfo *sra2 = sra->next;
53 struct mdinfo *d = sra->devs;
58 free(sra->bb.entries);
64 int sysfs_open(char *devnm, char *devname, char *attr)
66 char fname[MAX_SYSFS_PATH_LEN];
69 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/", devnm);
71 strncat(fname, devname, MAX_SYSFS_PATH_LEN - strlen(fname));
72 strncat(fname, "/", MAX_SYSFS_PATH_LEN - strlen(fname));
74 strncat(fname, attr, MAX_SYSFS_PATH_LEN - strlen(fname));
75 fd = open(fname, O_RDWR);
76 if (fd < 0 && errno == EACCES)
77 fd = open(fname, O_RDONLY);
81 void sysfs_init_dev(struct mdinfo *mdi, unsigned long devid)
83 snprintf(mdi->sys_name,
84 sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
87 void sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
92 if (ioctl(fd, RAID_VERSION, &vers) != 0)
98 strcpy(mdi->sys_name, devnm);
101 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
103 char fname[PATH_MAX];
108 struct mdinfo *dev, **devp;
112 sra = xcalloc(1, sizeof(*sra));
113 sysfs_init(sra, fd, devnm);
114 if (sra->sys_name[0] == 0) {
119 sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
120 base = fname + strlen(fname);
123 if (options & GET_VERSION) {
124 strcpy(base, "metadata_version");
125 if (load_sys(fname, buf, sizeof(buf)))
127 if (strncmp(buf, "none", 4) == 0) {
128 sra->array.major_version =
129 sra->array.minor_version = -1;
130 strcpy(sra->text_version, "");
131 } else if (strncmp(buf, "external:", 9) == 0) {
132 sra->array.major_version = -1;
133 sra->array.minor_version = -2;
134 strcpy(sra->text_version, buf+9);
137 &sra->array.major_version,
138 &sra->array.minor_version);
139 strcpy(sra->text_version, buf);
142 if (options & GET_LEVEL) {
143 strcpy(base, "level");
144 if (load_sys(fname, buf, sizeof(buf)))
146 sra->array.level = map_name(pers, buf);
148 if (options & GET_LAYOUT) {
149 strcpy(base, "layout");
150 if (load_sys(fname, buf, sizeof(buf)))
152 sra->array.layout = strtoul(buf, NULL, 0);
154 if (options & GET_DISKS) {
155 strcpy(base, "raid_disks");
156 if (load_sys(fname, buf, sizeof(buf)))
158 sra->array.raid_disks = strtoul(buf, NULL, 0);
160 if (options & GET_DEGRADED) {
161 strcpy(base, "degraded");
162 if (load_sys(fname, buf, sizeof(buf)))
164 sra->array.failed_disks = strtoul(buf, NULL, 0);
166 if (options & GET_COMPONENT) {
167 strcpy(base, "component_size");
168 if (load_sys(fname, buf, sizeof(buf)))
170 sra->component_size = strtoull(buf, NULL, 0);
171 /* sysfs reports "K", but we want sectors */
172 sra->component_size *= 2;
174 if (options & GET_CHUNK) {
175 strcpy(base, "chunk_size");
176 if (load_sys(fname, buf, sizeof(buf)))
178 sra->array.chunk_size = strtoul(buf, NULL, 0);
180 if (options & GET_CACHE) {
181 strcpy(base, "stripe_cache_size");
182 if (load_sys(fname, buf, sizeof(buf)))
183 /* Probably level doesn't support it */
186 sra->cache_size = strtoul(buf, NULL, 0);
188 if (options & GET_MISMATCH) {
189 strcpy(base, "mismatch_cnt");
190 if (load_sys(fname, buf, sizeof(buf)))
192 sra->mismatch_cnt = strtoul(buf, NULL, 0);
194 if (options & GET_SAFEMODE) {
201 strcpy(base, "safe_mode_delay");
202 if (load_sys(fname, buf, sizeof(buf)))
205 /* remove a period, and count digits after it */
207 for (i = 0; i < len; i++) {
209 if (isdigit(buf[i])) {
214 } else if (buf[i] == '.') {
219 msec = strtoul(buf, NULL, 10);
220 msec = (msec * 1000) / scale;
221 sra->safe_mode_delay = msec;
223 if (options & GET_BITMAP_LOCATION) {
224 strcpy(base, "bitmap/location");
225 if (load_sys(fname, buf, sizeof(buf)))
227 if (strncmp(buf, "file", 4) == 0)
228 sra->bitmap_offset = 1;
229 else if (strncmp(buf, "none", 4) == 0)
230 sra->bitmap_offset = 0;
231 else if (buf[0] == '+')
232 sra->bitmap_offset = strtol(buf+1, NULL, 10);
237 if (options & GET_ARRAY_STATE) {
238 strcpy(base, "array_state");
239 if (load_sys(fname, sra->sysfs_array_state,
240 sizeof(sra->sysfs_array_state)))
243 sra->sysfs_array_state[0] = 0;
245 if (! (options & GET_DEVS))
248 /* Get all the devices as well */
250 dir = opendir(fname);
253 sra->array.spare_disks = 0;
257 while ((de = readdir(dir)) != NULL) {
259 if (de->d_ino == 0 ||
260 strncmp(de->d_name, "dev-", 4) != 0)
262 strcpy(base, de->d_name);
263 dbase = base + strlen(base);
266 dev = xcalloc(1, sizeof(*dev));
268 /* Always get slot, major, minor */
269 strcpy(dbase, "slot");
270 if (load_sys(fname, buf, sizeof(buf))) {
271 /* hmm... unable to read 'slot' maybe the device
274 strcpy(dbase, "block");
275 if (readlink(fname, buf, sizeof(buf)) < 0 &&
276 errno != ENAMETOOLONG) {
277 /* ...yup device is gone */
281 /* slot is unreadable but 'block' link
282 * still intact... something bad is happening
290 strcpy(dev->sys_name, de->d_name);
291 dev->disk.raid_disk = strtoul(buf, &ep, 10);
292 if (*ep) dev->disk.raid_disk = -1;
294 strcpy(dbase, "block/dev");
295 if (load_sys(fname, buf, sizeof(buf))) {
296 /* assume this is a stale reference to a hot
302 sra->array.nr_disks++;
303 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
305 /* special case check for block devices that can go 'offline' */
306 strcpy(dbase, "block/device/state");
307 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
308 strncmp(buf, "offline", 7) == 0) {
313 /* finally add this disk to the array */
318 if (options & GET_OFFSET) {
319 strcpy(dbase, "offset");
320 if (load_sys(fname, buf, sizeof(buf)))
322 dev->data_offset = strtoull(buf, NULL, 0);
323 strcpy(dbase, "new_offset");
324 if (load_sys(fname, buf, sizeof(buf)) == 0)
325 dev->new_data_offset = strtoull(buf, NULL, 0);
327 dev->new_data_offset = dev->data_offset;
329 if (options & GET_SIZE) {
330 strcpy(dbase, "size");
331 if (load_sys(fname, buf, sizeof(buf)))
333 dev->component_size = strtoull(buf, NULL, 0) * 2;
335 if (options & GET_STATE) {
337 strcpy(dbase, "state");
338 if (load_sys(fname, buf, sizeof(buf)))
340 if (strstr(buf, "in_sync"))
341 dev->disk.state |= (1<<MD_DISK_SYNC);
342 if (strstr(buf, "faulty"))
343 dev->disk.state |= (1<<MD_DISK_FAULTY);
344 if (dev->disk.state == 0)
345 sra->array.spare_disks++;
347 if (options & GET_ERROR) {
348 strcpy(buf, "errors");
349 if (load_sys(fname, buf, sizeof(buf)))
351 dev->errors = strtoul(buf, NULL, 0);
364 int sysfs_attr_match(const char *attr, const char *str)
366 /* See if attr, read from a sysfs file, matches
367 * str. They must either be the same, or attr can
368 * have a trailing newline or comma
370 while (*attr && *str && *attr == *str) {
375 if (*str || (*attr && *attr != ',' && *attr != '\n'))
380 int sysfs_match_word(const char *word, char **list)
383 for (n=0; list[n]; n++)
384 if (sysfs_attr_match(word, list[n]))
389 unsigned long long get_component_size(int fd)
391 /* Find out the component size of the array.
392 * We cannot trust GET_ARRAY_INFO ioctl as it's
393 * size field is only 32bits.
394 * So look in /sys/block/mdXXX/md/component_size
396 * This returns in units of sectors.
399 char fname[MAX_SYSFS_PATH_LEN];
403 snprintf(fname, MAX_SYSFS_PATH_LEN,
404 "/sys/block/%s/md/component_size", stat2devnm(&stb));
405 fd = open(fname, O_RDONLY);
408 n = read(fd, fname, sizeof(fname));
410 if (n < 0 || n == sizeof(fname))
413 return strtoull(fname, NULL, 10) * 2;
416 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
417 char *name, char *val)
419 char fname[MAX_SYSFS_PATH_LEN];
423 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
424 sra->sys_name, dev?dev->sys_name:"", name);
425 fd = open(fname, O_WRONLY);
428 n = write(fd, val, strlen(val));
430 if (n != strlen(val)) {
431 dprintf("failed to write '%s' to '%s' (%s)\n",
432 val, fname, strerror(errno));
438 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
439 char *name, unsigned long long val)
442 sprintf(valstr, "%llu", val);
443 return sysfs_set_str(sra, dev, name, valstr);
446 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
447 char *name, long long val)
450 sprintf(valstr, "%lli", val);
451 return sysfs_set_str(sra, dev, name, valstr);
454 int sysfs_uevent(struct mdinfo *sra, char *event)
456 char fname[MAX_SYSFS_PATH_LEN];
460 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
462 fd = open(fname, O_WRONLY);
465 n = write(fd, event, strlen(event));
467 if (n != (int)strlen(event)) {
468 dprintf("failed to write '%s' to '%s' (%s)\n",
469 event, fname, strerror(errno));
475 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
477 char fname[MAX_SYSFS_PATH_LEN];
480 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
481 sra->sys_name, dev?dev->sys_name:"", name);
483 return stat(fname, &st) == 0;
486 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
489 char fname[MAX_SYSFS_PATH_LEN];
492 snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
493 sra->sys_name, dev?dev->sys_name:"", name);
494 fd = open(fname, O_RDWR);
496 fd = open(fname, O_RDONLY);
500 int sysfs_fd_get_ll(int fd, unsigned long long *val)
507 n = read(fd, buf, sizeof(buf));
508 if (n <= 0 || n == sizeof(buf))
511 *val = strtoull(buf, &ep, 0);
512 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
517 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
518 char *name, unsigned long long *val)
523 fd = sysfs_get_fd(sra, dev, name);
526 n = sysfs_fd_get_ll(fd, val);
531 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
533 /* two numbers in this sysfs file, either
543 n = read(fd, buf, sizeof(buf));
544 if (n <= 0 || n == sizeof(buf))
547 *v1 = strtoull(buf, &ep, 0);
548 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
550 while (*ep == ' ' || *ep == '/' || *ep == '(')
552 *v2 = strtoull(ep, &ep2, 0);
553 if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
560 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
561 char *name, unsigned long long *v1, unsigned long long *v2)
566 fd = sysfs_get_fd(sra, dev, name);
569 n = sysfs_fd_get_two(fd, v1, v2);
574 int sysfs_fd_get_str(int fd, char *val, int size)
579 n = read(fd, val, size);
580 if (n <= 0 || n == size)
586 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
587 char *name, char *val, int size)
592 fd = sysfs_get_fd(sra, dev, name);
595 n = sysfs_fd_get_str(fd, val, size);
600 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
609 sprintf(delay, "%ld.%03ld\n", sec, msec);
610 /* this '\n' ^ needed for kernels older than 2.6.28 */
611 return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
614 int sysfs_set_array(struct mdinfo *info, int vers)
618 int raid_disks = info->array.raid_disks;
621 if (info->array.major_version == -1 &&
622 info->array.minor_version == -2) {
625 strcat(strcpy(ver, "external:"), info->text_version);
627 /* meta version might already be set if we are setting
628 * new geometry for a reshape. In that case we don't
629 * want to over-write the 'readonly' flag that is
630 * stored in the metadata version. So read the current
631 * version first, and preserve the flag
633 if (sysfs_get_str(info, NULL, "metadata_version",
635 if (strlen(buf) >= 9 && buf[9] == '-')
638 if ((vers % 100) < 2 ||
639 sysfs_set_str(info, NULL, "metadata_version",
641 pr_err("This kernel does not support external metadata.\n");
645 if (info->array.level < 0)
646 return 0; /* FIXME */
647 rv |= sysfs_set_str(info, NULL, "level",
648 map_num(pers, info->array.level));
649 if (info->reshape_active && info->delta_disks != UnSet)
650 raid_disks -= info->delta_disks;
651 rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
652 rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
653 rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
654 rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
655 if (info->custom_array_size) {
658 rc = sysfs_set_num(info, NULL, "array_size",
659 info->custom_array_size/2);
660 if (rc && errno == ENOENT) {
661 pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
667 if (info->array.level > 0)
668 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
670 if (info->reshape_active) {
671 rv |= sysfs_set_num(info, NULL, "reshape_position",
672 info->reshape_progress);
673 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
674 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
675 rv |= sysfs_set_num(info, NULL, "raid_disks",
676 info->array.raid_disks);
677 /* We don't set 'new_level' here. That can only happen
678 * once the reshape completes.
684 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
692 sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
693 rv = sysfs_set_str(sra, NULL, "new_dev", dv);
697 memset(nm, 0, sizeof(nm));
698 dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
699 strcpy(sd->sys_name, "dev-");
700 strcpy(sd->sys_name+4, dname);
702 /* test write to see if 'recovery_start' is available */
703 if (resume && sd->recovery_start < MaxSector &&
704 sysfs_set_num(sra, sd, "recovery_start", 0)) {
705 sysfs_set_str(sra, sd, "state", "remove");
709 rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
710 rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
711 if (sra->array.level != LEVEL_CONTAINER) {
712 if (sd->recovery_start == MaxSector)
713 /* This can correctly fail if array isn't started,
714 * yet, so just ignore status for now.
716 sysfs_set_str(sra, sd, "state", "insync");
717 if (sd->disk.raid_disk >= 0)
718 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
720 sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
722 if (sd->bb.supported) {
723 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
725 * backward compatibility - if kernel doesn't support
726 * bad blocks for external metadata, let it continue
727 * as long as there are none known so far
730 pr_err("The kernel has no support for bad blocks in external metadata\n");
735 for (i = 0; i < sd->bb.count; i++) {
737 const struct md_bb_entry *entry = &sd->bb.entries[i];
739 snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
741 rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
748 int sysfs_disk_to_sg(int fd)
750 /* from an open block device, try find and open its corresponding
751 * scsi_generic interface
756 char sg_major_minor[10];
760 int major, minor, rv;
765 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
766 major(st.st_rdev), minor(st.st_rdev));
774 if (strncmp("scsi_generic:", de->d_name,
775 strlen("scsi_generic:")) == 0)
784 snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
785 fd = open(sg_path, O_RDONLY);
789 rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
791 if (rv < 0 || rv == sizeof(sg_major_minor))
794 sg_major_minor[rv - 1] = '\0';
796 c = strchr(sg_major_minor, ':');
799 major = strtol(sg_major_minor, NULL, 10);
800 minor = strtol(c, NULL, 10);
801 snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
802 (int) getpid(), major, minor);
803 if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
804 fd = open(path, O_RDONLY);
813 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
815 /* from an open block device, try to retrieve it scsi_id */
820 int host, bus, target, lun;
825 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
826 major(st.st_rdev), minor(st.st_rdev));
832 for (de = readdir(dir); de; de = readdir(dir)) {
835 if (de->d_type != DT_DIR)
838 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
847 *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
851 int sysfs_unique_holder(char *devnm, long rdev)
853 /* Check that devnm is a holder of rdev,
854 * and is the only holder.
855 * we should be locked against races by
858 * 0 - not unique, not even a holder
859 * 1 - unique, this is the only holder.
860 * 2/3 - not unique, there is another holder
861 * -1 - error, cannot find the holders
868 sprintf(dirname, "/sys/dev/block/%d:%d/holders",
869 major(rdev), minor(rdev));
870 dir = opendir(dirname);
874 while ((de = readdir(dir)) != NULL) {
881 if (de->d_name[0] == '.')
883 strcpy(dirname+l, "/");
884 strcat(dirname+l, de->d_name);
885 n = readlink(dirname, buf, sizeof(buf)-1);
889 sl = strrchr(buf, '/');
894 if (strcmp(devnm, sl) == 0)
903 int sysfs_freeze_array(struct mdinfo *sra)
905 /* Try to freeze resync/rebuild on this array/container.
906 * Return -1 if the array is busy,
907 * return 0 if this kernel doesn't support 'frozen'
908 * return 1 if it worked.
912 if (!sysfs_attribute_available(sra, NULL, "sync_action"))
913 return 1; /* no sync_action == frozen */
914 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
916 if (strcmp(buf, "frozen\n") == 0)
919 if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
921 if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
926 int sysfs_wait(int fd, int *msec)
928 /* Wait up to '*msec' for fd to have an exception condition.
929 * if msec == NULL, wait indefinitely.
936 n = select(fd+1, NULL, NULL, &fds, NULL);
940 struct timeval start, end, tv;
941 gettimeofday(&start, NULL);
944 tv.tv_usec = (*msec)*1000;
946 tv.tv_sec = (*msec)/1000;
949 n = select(fd+1, NULL, NULL, &fds, &tv);
950 gettimeofday(&end, NULL);
951 end.tv_sec -= start.tv_sec;
952 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
953 - start.tv_usec/1000) + 1;