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, int len)
32 int fd = open(path, O_RDONLY);
36 n = read(fd, buf, len);
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_dev(struct mdinfo *mdi, unsigned long devid)
79 snprintf(mdi->sys_name,
80 sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
83 void sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
88 if (ioctl(fd, RAID_VERSION, &vers) != 0)
94 strcpy(mdi->sys_name, devnm);
97 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
104 struct mdinfo *dev, **devp;
108 sra = xcalloc(1, sizeof(*sra));
109 sysfs_init(sra, fd, devnm);
110 if (sra->sys_name[0] == 0) {
115 sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
116 base = fname + strlen(fname);
119 if (options & GET_VERSION) {
120 strcpy(base, "metadata_version");
121 if (load_sys(fname, buf, sizeof(buf)))
123 if (strncmp(buf, "none", 4) == 0) {
124 sra->array.major_version =
125 sra->array.minor_version = -1;
126 strcpy(sra->text_version, "");
127 } else if (strncmp(buf, "external:", 9) == 0) {
128 sra->array.major_version = -1;
129 sra->array.minor_version = -2;
130 strcpy(sra->text_version, buf+9);
133 &sra->array.major_version,
134 &sra->array.minor_version);
135 strcpy(sra->text_version, buf);
138 if (options & GET_LEVEL) {
139 strcpy(base, "level");
140 if (load_sys(fname, buf, sizeof(buf)))
142 sra->array.level = map_name(pers, buf);
144 if (options & GET_LAYOUT) {
145 strcpy(base, "layout");
146 if (load_sys(fname, buf, sizeof(buf)))
148 sra->array.layout = strtoul(buf, NULL, 0);
150 if (options & GET_DISKS) {
151 strcpy(base, "raid_disks");
152 if (load_sys(fname, buf, sizeof(buf)))
154 sra->array.raid_disks = strtoul(buf, NULL, 0);
156 if (options & GET_DEGRADED) {
157 strcpy(base, "degraded");
158 if (load_sys(fname, buf, sizeof(buf)))
160 sra->array.failed_disks = strtoul(buf, NULL, 0);
162 if (options & GET_COMPONENT) {
163 strcpy(base, "component_size");
164 if (load_sys(fname, buf, sizeof(buf)))
166 sra->component_size = strtoull(buf, NULL, 0);
167 /* sysfs reports "K", but we want sectors */
168 sra->component_size *= 2;
170 if (options & GET_CHUNK) {
171 strcpy(base, "chunk_size");
172 if (load_sys(fname, buf, sizeof(buf)))
174 sra->array.chunk_size = strtoul(buf, NULL, 0);
176 if (options & GET_CACHE) {
177 strcpy(base, "stripe_cache_size");
178 if (load_sys(fname, buf, sizeof(buf)))
179 /* Probably level doesn't support it */
182 sra->cache_size = strtoul(buf, NULL, 0);
184 if (options & GET_MISMATCH) {
185 strcpy(base, "mismatch_cnt");
186 if (load_sys(fname, buf, sizeof(buf)))
188 sra->mismatch_cnt = strtoul(buf, NULL, 0);
190 if (options & GET_SAFEMODE) {
197 strcpy(base, "safe_mode_delay");
198 if (load_sys(fname, buf, sizeof(buf)))
201 /* remove a period, and count digits after it */
203 for (i = 0; i < len; i++) {
205 if (isdigit(buf[i])) {
210 } else if (buf[i] == '.') {
215 msec = strtoul(buf, NULL, 10);
216 msec = (msec * 1000) / scale;
217 sra->safe_mode_delay = msec;
219 if (options & GET_BITMAP_LOCATION) {
220 strcpy(base, "bitmap/location");
221 if (load_sys(fname, buf, sizeof(buf)))
223 if (strncmp(buf, "file", 4) == 0)
224 sra->bitmap_offset = 1;
225 else if (strncmp(buf, "none", 4) == 0)
226 sra->bitmap_offset = 0;
227 else if (buf[0] == '+')
228 sra->bitmap_offset = strtol(buf+1, NULL, 10);
233 if (options & GET_ARRAY_STATE) {
234 strcpy(base, "array_state");
235 if (load_sys(fname, sra->sysfs_array_state,
236 sizeof(sra->sysfs_array_state)))
239 sra->sysfs_array_state[0] = 0;
241 if (! (options & GET_DEVS))
244 /* Get all the devices as well */
246 dir = opendir(fname);
249 sra->array.spare_disks = 0;
253 while ((de = readdir(dir)) != NULL) {
255 if (de->d_ino == 0 ||
256 strncmp(de->d_name, "dev-", 4) != 0)
258 strcpy(base, de->d_name);
259 dbase = base + strlen(base);
262 dev = xmalloc(sizeof(*dev));
264 /* Always get slot, major, minor */
265 strcpy(dbase, "slot");
266 if (load_sys(fname, buf, sizeof(buf))) {
267 /* hmm... unable to read 'slot' maybe the device
270 strcpy(dbase, "block");
271 if (readlink(fname, buf, sizeof(buf)) < 0 &&
272 errno != ENAMETOOLONG) {
273 /* ...yup device is gone */
277 /* slot is unreadable but 'block' link
278 * still intact... something bad is happening
286 strcpy(dev->sys_name, de->d_name);
287 dev->disk.raid_disk = strtoul(buf, &ep, 10);
288 if (*ep) dev->disk.raid_disk = -1;
290 strcpy(dbase, "block/dev");
291 if (load_sys(fname, buf, sizeof(buf))) {
292 /* assume this is a stale reference to a hot
298 sra->array.nr_disks++;
299 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
301 /* special case check for block devices that can go 'offline' */
302 strcpy(dbase, "block/device/state");
303 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
304 strncmp(buf, "offline", 7) == 0) {
309 /* finally add this disk to the array */
314 if (options & GET_OFFSET) {
315 strcpy(dbase, "offset");
316 if (load_sys(fname, buf, sizeof(buf)))
318 dev->data_offset = strtoull(buf, NULL, 0);
319 strcpy(dbase, "new_offset");
320 if (load_sys(fname, buf, sizeof(buf)) == 0)
321 dev->new_data_offset = strtoull(buf, NULL, 0);
323 dev->new_data_offset = dev->data_offset;
325 if (options & GET_SIZE) {
326 strcpy(dbase, "size");
327 if (load_sys(fname, buf, sizeof(buf)))
329 dev->component_size = strtoull(buf, NULL, 0) * 2;
331 if (options & GET_STATE) {
333 strcpy(dbase, "state");
334 if (load_sys(fname, buf, sizeof(buf)))
336 if (strstr(buf, "in_sync"))
337 dev->disk.state |= (1<<MD_DISK_SYNC);
338 if (strstr(buf, "faulty"))
339 dev->disk.state |= (1<<MD_DISK_FAULTY);
340 if (dev->disk.state == 0)
341 sra->array.spare_disks++;
343 if (options & GET_ERROR) {
344 strcpy(buf, "errors");
345 if (load_sys(fname, buf, sizeof(buf)))
347 dev->errors = strtoul(buf, NULL, 0);
360 int sysfs_attr_match(const char *attr, const char *str)
362 /* See if attr, read from a sysfs file, matches
363 * str. They must either be the same, or attr can
364 * have a trailing newline or comma
366 while (*attr && *str && *attr == *str) {
371 if (*str || (*attr && *attr != ',' && *attr != '\n'))
376 int sysfs_match_word(const char *word, char **list)
379 for (n=0; list[n]; n++)
380 if (sysfs_attr_match(word, list[n]))
385 unsigned long long get_component_size(int fd)
387 /* Find out the component size of the array.
388 * We cannot trust GET_ARRAY_INFO ioctl as it's
389 * size field is only 32bits.
390 * So look in /sys/block/mdXXX/md/component_size
392 * This returns in units of sectors.
397 if (fstat(fd, &stb)) return 0;
398 if (major(stb.st_rdev) != (unsigned)get_mdp_major())
399 sprintf(fname, "/sys/block/md%d/md/component_size",
400 (int)minor(stb.st_rdev));
402 sprintf(fname, "/sys/block/md_d%d/md/component_size",
403 (int)minor(stb.st_rdev)>>MdpMinorShift);
404 fd = open(fname, O_RDONLY);
407 n = read(fd, fname, sizeof(fname));
409 if (n < 0 || n == sizeof(fname))
412 return strtoull(fname, NULL, 10) * 2;
415 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
416 char *name, char *val)
422 sprintf(fname, "/sys/block/%s/md/%s/%s",
423 sra->sys_name, dev?dev->sys_name:"", name);
424 fd = open(fname, O_WRONLY);
427 n = write(fd, val, strlen(val));
429 if (n != strlen(val)) {
430 dprintf("failed to write '%s' to '%s' (%s)\n",
431 val, fname, strerror(errno));
437 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
438 char *name, unsigned long long val)
441 sprintf(valstr, "%llu", val);
442 return sysfs_set_str(sra, dev, name, valstr);
445 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
446 char *name, long long val)
449 sprintf(valstr, "%lli", val);
450 return sysfs_set_str(sra, dev, name, valstr);
453 int sysfs_uevent(struct mdinfo *sra, char *event)
459 sprintf(fname, "/sys/block/%s/uevent",
461 fd = open(fname, O_WRONLY);
464 n = write(fd, event, strlen(event));
466 if (n != (int)strlen(event)) {
467 dprintf("failed to write '%s' to '%s' (%s)\n",
468 event, fname, strerror(errno));
474 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
479 sprintf(fname, "/sys/block/%s/md/%s/%s",
480 sra->sys_name, dev?dev->sys_name:"", name);
482 return stat(fname, &st) == 0;
485 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
491 sprintf(fname, "/sys/block/%s/md/%s/%s",
492 sra->sys_name, dev?dev->sys_name:"", name);
493 fd = open(fname, O_RDWR);
495 fd = open(fname, O_RDONLY);
499 int sysfs_fd_get_ll(int fd, unsigned long long *val)
506 n = read(fd, buf, sizeof(buf));
507 if (n <= 0 || n == sizeof(buf))
510 *val = strtoull(buf, &ep, 0);
511 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
516 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
517 char *name, unsigned long long *val)
522 fd = sysfs_get_fd(sra, dev, name);
525 n = sysfs_fd_get_ll(fd, val);
530 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
532 /* two numbers in this sysfs file, either
542 n = read(fd, buf, sizeof(buf));
543 if (n <= 0 || n == sizeof(buf))
546 *v1 = strtoull(buf, &ep, 0);
547 if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
549 while (*ep == ' ' || *ep == '/' || *ep == '(')
551 *v2 = strtoull(ep, &ep2, 0);
552 if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
559 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
560 char *name, unsigned long long *v1, unsigned long long *v2)
565 fd = sysfs_get_fd(sra, dev, name);
568 n = sysfs_fd_get_two(fd, v1, v2);
573 int sysfs_fd_get_str(int fd, char *val, int size)
578 n = read(fd, val, size);
579 if (n <= 0 || n == size)
585 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
586 char *name, char *val, int size)
591 fd = sysfs_get_fd(sra, dev, name);
594 n = sysfs_fd_get_str(fd, val, size);
599 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
608 sprintf(delay, "%ld.%03ld\n", sec, msec);
609 /* this '\n' ^ needed for kernels older than 2.6.28 */
610 return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
613 int sysfs_set_array(struct mdinfo *info, int vers)
617 int raid_disks = info->array.raid_disks;
620 if (info->array.major_version == -1 &&
621 info->array.minor_version == -2) {
624 strcat(strcpy(ver, "external:"), info->text_version);
626 /* meta version might already be set if we are setting
627 * new geometry for a reshape. In that case we don't
628 * want to over-write the 'readonly' flag that is
629 * stored in the metadata version. So read the current
630 * version first, and preserve the flag
632 if (sysfs_get_str(info, NULL, "metadata_version",
634 if (strlen(buf) >= 9 && buf[9] == '-')
637 if ((vers % 100) < 2 ||
638 sysfs_set_str(info, NULL, "metadata_version",
640 pr_err("This kernel does not support external metadata.\n");
644 if (info->array.level < 0)
645 return 0; /* FIXME */
646 rv |= sysfs_set_str(info, NULL, "level",
647 map_num(pers, info->array.level));
648 if (info->reshape_active && info->delta_disks != UnSet)
649 raid_disks -= info->delta_disks;
650 rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
651 rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
652 rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
653 rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
654 if (info->custom_array_size) {
657 rc = sysfs_set_num(info, NULL, "array_size",
658 info->custom_array_size/2);
659 if (rc && errno == ENOENT) {
660 pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
666 if (info->array.level > 0)
667 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
669 if (info->reshape_active) {
670 rv |= sysfs_set_num(info, NULL, "reshape_position",
671 info->reshape_progress);
672 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
673 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
674 rv |= sysfs_set_num(info, NULL, "raid_disks",
675 info->array.raid_disks);
676 /* We don't set 'new_level' here. That can only happen
677 * once the reshape completes.
683 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
690 sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
691 rv = sysfs_set_str(sra, NULL, "new_dev", dv);
695 memset(nm, 0, sizeof(nm));
696 dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
697 strcpy(sd->sys_name, "dev-");
698 strcpy(sd->sys_name+4, dname);
700 /* test write to see if 'recovery_start' is available */
701 if (resume && sd->recovery_start < MaxSector &&
702 sysfs_set_num(sra, sd, "recovery_start", 0)) {
703 sysfs_set_str(sra, sd, "state", "remove");
707 rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
708 rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
709 if (sra->array.level != LEVEL_CONTAINER) {
710 if (sd->recovery_start == MaxSector)
711 /* This can correctly fail if array isn't started,
712 * yet, so just ignore status for now.
714 sysfs_set_str(sra, sd, "state", "insync");
715 if (sd->disk.raid_disk >= 0)
716 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
718 sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
724 int sysfs_disk_to_sg(int fd)
726 /* from an open block device, try find and open its corresponding
727 * scsi_generic interface
732 char sg_major_minor[10];
736 int major, minor, rv;
741 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
742 major(st.st_rdev), minor(st.st_rdev));
750 if (strncmp("scsi_generic:", de->d_name,
751 strlen("scsi_generic:")) == 0)
760 snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
761 fd = open(sg_path, O_RDONLY);
765 rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
767 if (rv < 0 || rv == sizeof(sg_major_minor))
770 sg_major_minor[rv - 1] = '\0';
772 c = strchr(sg_major_minor, ':');
775 major = strtol(sg_major_minor, NULL, 10);
776 minor = strtol(c, NULL, 10);
777 snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
778 (int) getpid(), major, minor);
779 if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
780 fd = open(path, O_RDONLY);
789 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
791 /* from an open block device, try to retrieve it scsi_id */
796 int host, bus, target, lun;
801 snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
802 major(st.st_rdev), minor(st.st_rdev));
808 for (de = readdir(dir); de; de = readdir(dir)) {
811 if (de->d_type != DT_DIR)
814 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
823 *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
827 int sysfs_unique_holder(char *devnm, long rdev)
829 /* Check that devnm is a holder of rdev,
830 * and is the only holder.
831 * we should be locked against races by
834 * 0 - not unique, not even a holder
835 * 1 - unique, this is the only holder.
836 * 2/3 - not unique, there is another holder
837 * -1 - error, cannot find the holders
844 sprintf(dirname, "/sys/dev/block/%d:%d/holders",
845 major(rdev), minor(rdev));
846 dir = opendir(dirname);
850 while ((de = readdir(dir)) != NULL) {
857 if (de->d_name[0] == '.')
859 strcpy(dirname+l, "/");
860 strcat(dirname+l, de->d_name);
861 n = readlink(dirname, buf, sizeof(buf)-1);
865 sl = strrchr(buf, '/');
870 if (strcmp(devnm, sl) == 0)
879 int sysfs_freeze_array(struct mdinfo *sra)
881 /* Try to freeze resync/rebuild on this array/container.
882 * Return -1 if the array is busy,
883 * return 0 if this kernel doesn't support 'frozen'
884 * return 1 if it worked.
888 if (!sysfs_attribute_available(sra, NULL, "sync_action"))
889 return 1; /* no sync_action == frozen */
890 if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
892 if (strcmp(buf, "frozen\n") == 0)
895 if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
897 if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
902 int sysfs_wait(int fd, int *msec)
904 /* Wait up to '*msec' for fd to have an exception condition.
905 * if msec == NULL, wait indefinitely.
912 n = select(fd+1, NULL, NULL, &fds, NULL);
916 struct timeval start, end, tv;
917 gettimeofday(&start, NULL);
920 tv.tv_usec = (*msec)*1000;
922 tv.tv_sec = (*msec)/1000;
925 n = select(fd+1, NULL, NULL, &fds, &tv);
926 gettimeofday(&end, NULL);
927 end.tv_sec -= start.tv_sec;
928 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
929 - start.tv_usec/1000) + 1;