2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include "kerncompat.h"
21 #include <sys/ioctl.h>
24 #include <sys/types.h>
25 #include <sys/socket.h>
27 #include <sys/syscall.h>
30 #include <uuid/uuid.h>
46 static const char * const scrub_cmd_group_usage[] = {
47 "btrfs scrub <command> [options] <path>|<device>",
51 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
52 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
53 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
54 #define SCRUB_FILE_VERSION "1"
65 /* TBD: replace with #include "linux/ioprio.h" in some years */
66 #if !defined (IOPRIO_H)
67 #define IOPRIO_WHO_PROCESS 1
68 #define IOPRIO_CLASS_SHIFT 13
69 #define IOPRIO_PRIO_VALUE(class, data) \
70 (((class) << IOPRIO_CLASS_SHIFT) | (data))
71 #define IOPRIO_CLASS_IDLE 3
74 struct scrub_progress {
75 struct btrfs_ioctl_scrub_args scrub_args;
79 struct scrub_stats stats;
80 struct scrub_file_record *resumed;
82 pthread_mutex_t progress_mutex;
87 struct scrub_file_record {
88 u8 fsid[BTRFS_FSID_SIZE];
90 struct scrub_stats stats;
91 struct btrfs_scrub_progress p;
94 struct scrub_progress_cycle {
98 struct btrfs_ioctl_fs_info_args *fi;
99 struct scrub_progress *progress;
100 struct scrub_progress *shared_progress;
101 pthread_mutex_t *write_mutex;
104 struct scrub_fs_stat {
105 struct btrfs_scrub_progress p;
106 struct scrub_stats s;
110 static void print_scrub_full(struct btrfs_scrub_progress *sp)
112 printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
113 printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
114 printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
115 printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
116 printf("\tread_errors: %lld\n", sp->read_errors);
117 printf("\tcsum_errors: %lld\n", sp->csum_errors);
118 printf("\tverify_errors: %lld\n", sp->verify_errors);
119 printf("\tno_csum: %lld\n", sp->no_csum);
120 printf("\tcsum_discards: %lld\n", sp->csum_discards);
121 printf("\tsuper_errors: %lld\n", sp->super_errors);
122 printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
123 printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
124 printf("\tunverified_errors: %lld\n", sp->unverified_errors);
125 printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
126 printf("\tlast_physical: %lld\n", sp->last_physical);
129 #define ERR(test, ...) do { \
131 fprintf(stderr, __VA_ARGS__); \
134 #define PRINT_SCRUB_ERROR(test, desc) do { \
136 printf(" %s=%llu", desc, test); \
139 static void print_scrub_summary(struct btrfs_scrub_progress *p)
144 err_cnt = p->read_errors +
149 err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
151 if (p->malloc_errors)
152 printf("*** WARNING: memory allocation failed while scrubbing. "
153 "results may be inaccurate\n");
155 printf("\ttotal bytes scrubbed: %s with %llu errors\n",
156 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
157 max(err_cnt, err_cnt2));
159 if (err_cnt || err_cnt2) {
160 printf("\terror details:");
161 PRINT_SCRUB_ERROR(p->read_errors, "read");
162 PRINT_SCRUB_ERROR(p->super_errors, "super");
163 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
164 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
166 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
167 "unverified errors: %llu\n", p->corrected_errors,
168 p->uncorrectable_errors, p->unverified_errors);
172 #define _SCRUB_FS_STAT(p, name, fs_stat) do { \
173 fs_stat->p.name += p->name; \
176 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat) \
178 if (fs_stat->s.name > ss->name) { \
179 fs_stat->s.name = ss->name; \
183 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat) \
185 if (!fs_stat->s.name || fs_stat->s.name > ss->name) { \
186 fs_stat->s.name = ss->name; \
190 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat) \
192 if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) { \
193 (fs_stat)->s.name = (ss)->name; \
197 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
198 struct scrub_stats *ss,
199 struct scrub_fs_stat *fs_stat)
201 _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
202 _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
203 _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
204 _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
205 _SCRUB_FS_STAT(p, read_errors, fs_stat);
206 _SCRUB_FS_STAT(p, csum_errors, fs_stat);
207 _SCRUB_FS_STAT(p, verify_errors, fs_stat);
208 _SCRUB_FS_STAT(p, no_csum, fs_stat);
209 _SCRUB_FS_STAT(p, csum_discards, fs_stat);
210 _SCRUB_FS_STAT(p, super_errors, fs_stat);
211 _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
212 _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
213 _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
214 _SCRUB_FS_STAT(p, last_physical, fs_stat);
215 _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
216 _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
217 _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
218 _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
219 _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
222 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
224 memset(fs_stat, 0, sizeof(*fs_stat));
225 fs_stat->s.finished = 1;
228 static void _print_scrub_ss(struct scrub_stats *ss)
233 if (!ss || !ss->t_start) {
234 printf("\tno stats available\n");
238 localtime_r(&ss->t_resumed, &tm);
239 strftime(t, sizeof(t), "%c", &tm);
240 t[sizeof(t) - 1] = '\0';
241 printf("\tscrub resumed at %s", t);
243 localtime_r(&ss->t_start, &tm);
244 strftime(t, sizeof(t), "%c", &tm);
245 t[sizeof(t) - 1] = '\0';
246 printf("\tscrub started at %s", t);
248 if (ss->finished && !ss->canceled) {
249 printf(" and finished after %llu seconds\n",
251 } else if (ss->canceled) {
252 printf(" and was aborted after %llu seconds\n",
256 printf(", running for %llu seconds\n", ss->duration);
258 printf(", interrupted after %llu seconds, not running\n",
263 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
264 struct btrfs_scrub_progress *p, int raw,
265 const char *append, struct scrub_stats *ss)
267 printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
268 append ? append : "");
276 print_scrub_summary(p);
280 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
282 _print_scrub_ss(&fs_stat->s);
285 print_scrub_full(&fs_stat->p);
287 print_scrub_summary(&fs_stat->p);
290 static void free_history(struct scrub_file_record **last_scrubs)
292 struct scrub_file_record **l = last_scrubs;
301 * cancels a running scrub and makes the master process record the current
302 * progress status before exiting.
304 static int cancel_fd = -1;
305 static void scrub_sigint_record_progress(int signal)
309 ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
311 perror("Scrub cancel failed");
314 static int scrub_handle_sigint_parent(void)
316 struct sigaction sa = {
317 .sa_handler = SIG_IGN,
318 .sa_flags = SA_RESTART,
321 return sigaction(SIGINT, &sa, NULL);
324 static int scrub_handle_sigint_child(int fd)
326 struct sigaction sa = {
327 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
331 return sigaction(SIGINT, &sa, NULL);
334 static int scrub_datafile(const char *fn_base, const char *fn_local,
335 const char *fn_tmp, char *datafile, int size)
340 datafile[end + 1] = '\0';
341 strncpy(datafile, fn_base, end);
342 ret = strlen(datafile);
348 strncpy(datafile + ret + 1, fn_local, end - ret - 1);
349 ret = strlen(datafile);
356 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
357 ret = strlen(datafile);
366 static int scrub_open_file(const char *datafile, int m)
371 fd = open(datafile, m, 0600);
375 ret = flock(fd, LOCK_EX|LOCK_NB);
385 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
388 char datafile[BTRFS_PATH_NAME_MAX + 1];
389 ret = scrub_datafile(fn_base, fn_local, NULL,
390 datafile, sizeof(datafile));
393 return scrub_open_file(datafile, O_RDONLY);
396 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
400 char datafile[BTRFS_PATH_NAME_MAX + 1];
401 ret = scrub_datafile(fn_base, fn_local, tmp,
402 datafile, sizeof(datafile));
405 return scrub_open_file(datafile, O_WRONLY|O_CREAT);
408 static int scrub_rename_file(const char *fn_base, const char *fn_local,
412 char datafile_old[BTRFS_PATH_NAME_MAX + 1];
413 char datafile_new[BTRFS_PATH_NAME_MAX + 1];
414 ret = scrub_datafile(fn_base, fn_local, tmp,
415 datafile_old, sizeof(datafile_old));
418 ret = scrub_datafile(fn_base, fn_local, NULL,
419 datafile_new, sizeof(datafile_new));
422 ret = rename(datafile_old, datafile_new);
423 return ret ? -errno : 0;
426 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) { \
427 ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
431 * returns 0 if the key did not match (nothing was read)
432 * 1 if the key did match (success)
433 * -1 if the key did match and an error occured
435 static int scrub_kvread(int *i, int len, int avail, const char *buf,
436 const char *key, u64 *dest)
440 if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
445 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
449 *dest = atoll(&buf[*i]);
457 #define _SCRUB_INVALID do { \
459 fprintf(stderr, "WARNING: invalid data in line %d pos " \
460 "%d state %d (near \"%.*s\") at %s:%d\n", \
461 lineno, i, state, 20 > avail ? avail : 20, \
462 l + i, __FILE__, __LINE__); \
466 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
479 char empty_uuid[BTRFS_FSID_SIZE] = {0};
480 struct scrub_file_record **p = NULL;
483 old_avail = avail - i;
484 BUG_ON(old_avail < 0);
486 memmove(l, l + i, old_avail);
487 avail = read(fd, l + old_avail, sizeof(l) - old_avail);
490 if (avail == 0 && old_avail == 0) {
492 memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
494 } else if (curr == -1) {
495 p = ERR_PTR(-ENODATA);
500 return ERR_PTR(-errno);
506 case 0: /* start of file */
507 ret = scrub_kvread(&i,
508 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
509 SCRUB_FILE_VERSION_PREFIX, &version);
512 if (version != atoll(SCRUB_FILE_VERSION))
513 return ERR_PTR(-ENOTSUP);
516 case 1: /* start of line, alloc */
518 * this state makes sure we have a complete line in
519 * further processing, so we don't need wrap-tracking
522 if (!eof && !memchr(l + i, '\n', avail - i))
525 if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
526 BTRFS_FSID_SIZE) == 0) {
531 p = realloc(p, (curr + 2) * sizeof(*p));
533 p[curr] = malloc(sizeof(**p));
535 return ERR_PTR(-errno);
536 memset(p[curr], 0, sizeof(**p));
540 case 2: /* start of line, skip space */
541 while (isspace(l[i]) && i < avail) {
547 (!eof && !memchr(l + i, '\n', avail - i)))
551 case 3: /* read fsid */
554 for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
556 if (i + j + 1 >= avail)
558 if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
561 ret = uuid_parse(l + i, p[curr]->fsid);
567 case 4: /* read dev id */
568 for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
570 if (j == 0 || i + j + 1 >= avail)
572 p[curr]->devid = atoll(&l[i]);
576 case 5: /* read key/value pair */
578 _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
580 _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
582 _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
584 _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
586 _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
588 _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
590 _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
592 _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
594 _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
596 _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
598 _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
600 _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
602 _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
604 _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
606 _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
608 _SCRUB_KVREAD(ret, &i, finished, avail, l,
610 _SCRUB_KVREAD(ret, &i, t_start, avail, l,
611 (u64 *)&p[curr]->stats);
612 _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
613 (u64 *)&p[curr]->stats);
614 _SCRUB_KVREAD(ret, &i, duration, avail, l,
615 (u64 *)&p[curr]->stats);
616 _SCRUB_KVREAD(ret, &i, canceled, avail, l,
622 case 6: /* after number */
625 else if (l[i] == '\n')
631 case 99: /* skip rest of line */
636 if (l[i - 1] == '\n') {
648 static int scrub_write_buf(int fd, const void *data, int len)
651 ret = write(fd, data, len);
655 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
656 __attribute__ ((format (printf, 4, 5)));
657 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
663 ret = vsnprintf(buf, max, fmt, args);
667 return scrub_write_buf(fd, buf, ret);
670 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name = \
671 data->resumed->p.name + data->scrub_args.progress.name
673 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
674 struct scrub_progress *dest)
676 if (!data->resumed || data->skip)
679 _SCRUB_SUM(dest, data, data_extents_scrubbed);
680 _SCRUB_SUM(dest, data, tree_extents_scrubbed);
681 _SCRUB_SUM(dest, data, data_bytes_scrubbed);
682 _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
683 _SCRUB_SUM(dest, data, read_errors);
684 _SCRUB_SUM(dest, data, csum_errors);
685 _SCRUB_SUM(dest, data, verify_errors);
686 _SCRUB_SUM(dest, data, no_csum);
687 _SCRUB_SUM(dest, data, csum_discards);
688 _SCRUB_SUM(dest, data, super_errors);
689 _SCRUB_SUM(dest, data, malloc_errors);
690 _SCRUB_SUM(dest, data, uncorrectable_errors);
691 _SCRUB_SUM(dest, data, corrected_errors);
692 _SCRUB_SUM(dest, data, last_physical);
693 dest->stats.canceled = data->stats.canceled;
694 dest->stats.finished = data->stats.finished;
695 dest->stats.t_resumed = data->stats.t_start;
696 dest->stats.t_start = data->resumed->stats.t_start;
697 dest->stats.duration = data->resumed->stats.duration +
698 data->stats.duration;
699 dest->scrub_args.devid = data->scrub_args.devid;
703 #define _SCRUB_KVWRITE(fd, buf, name, use) \
704 scrub_kvwrite(fd, buf, sizeof(buf), #name, \
705 use->scrub_args.progress.name)
707 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use) \
708 scrub_kvwrite(fd, buf, sizeof(buf), #name, \
711 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
713 return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
716 static int scrub_write_file(int fd, const char *fsid,
717 struct scrub_progress *data, int n)
722 struct scrub_progress local;
723 struct scrub_progress *use;
728 /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
729 ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
730 SCRUB_FILE_VERSION "\n",
731 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
732 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
736 for (i = 0; i < n; ++i) {
737 use = scrub_resumed_stats(&data[i], &local);
738 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
739 scrub_write_buf(fd, ":", 1) ||
740 scrub_writev(fd, buf, sizeof(buf), "%lld",
741 use->scrub_args.devid) ||
742 scrub_write_buf(fd, buf, ret) ||
743 _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
744 _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
745 _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
746 _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
747 _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
748 _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
749 _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
750 _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
751 _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
752 _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
753 _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
754 _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
755 _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
756 _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
757 _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
758 _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
759 _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
760 _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
761 _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
762 scrub_write_buf(fd, "\n", 1)) {
770 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
771 struct scrub_progress *data, int n)
778 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
784 ret = pthread_mutex_lock(m);
790 fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
795 err = scrub_write_file(fd, fsid, data, n);
798 err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
809 ret = pthread_mutex_unlock(m);
814 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
822 static void *scrub_one_dev(void *ctx)
824 struct scrub_progress *sp = ctx;
828 sp->stats.canceled = 0;
829 sp->stats.duration = 0;
830 sp->stats.finished = 0;
832 ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
833 IOPRIO_PRIO_VALUE(sp->ioprio_class,
834 sp->ioprio_classdata));
837 "WARNING: setting ioprio failed: %s (ignored).\n",
840 ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
841 gettimeofday(&tv, NULL);
843 sp->stats.duration = tv.tv_sec - sp->stats.t_start;
844 sp->stats.canceled = !!ret;
845 sp->ioctl_errno = errno;
846 ret = pthread_mutex_lock(&sp->progress_mutex);
848 return ERR_PTR(-ret);
849 sp->stats.finished = 1;
850 ret = pthread_mutex_unlock(&sp->progress_mutex);
852 return ERR_PTR(-ret);
857 static void *progress_one_dev(void *ctx)
859 struct scrub_progress *sp = ctx;
861 sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
862 sp->ioctl_errno = errno;
867 /* nb: returns a negative errno via ERR_PTR */
868 static void *scrub_progress_cycle(void *ctx)
871 int perr = 0; /* positive / pthread error returns */
874 char fsid[BTRFS_UUID_UNPARSED_SIZE];
875 struct scrub_progress *sp;
876 struct scrub_progress *sp_last;
877 struct scrub_progress *sp_shared;
879 struct scrub_progress_cycle *spc = ctx;
880 int ndev = spc->fi->num_devices;
884 struct pollfd accept_poll_fd = {
889 struct pollfd write_poll_fd = {
893 struct sockaddr_un peer;
894 socklen_t peer_size = sizeof(peer);
896 perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
900 uuid_unparse(spc->fi->fsid, fsid);
902 for (i = 0; i < ndev; ++i) {
903 sp = &spc->progress[i];
904 sp_last = &spc->progress[i + ndev];
905 sp_shared = &spc->shared_progress[i];
906 sp->scrub_args.devid = sp_last->scrub_args.devid =
907 sp_shared->scrub_args.devid;
908 sp->fd = sp_last->fd = spc->fdmnt;
909 sp->stats.t_start = sp_last->stats.t_start =
910 sp_shared->stats.t_start;
911 sp->resumed = sp_last->resumed = sp_shared->resumed;
912 sp->skip = sp_last->skip = sp_shared->skip;
913 sp->stats.finished = sp_last->stats.finished =
914 sp_shared->stats.finished;
918 ret = poll(&accept_poll_fd, 1, 5 * 1000);
924 peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
926 gettimeofday(&tv, NULL);
929 for (i = 0; i < ndev; ++i) {
930 sp = &spc->progress[this * ndev + i];
931 sp_last = &spc->progress[last * ndev + i];
932 sp_shared = &spc->shared_progress[i];
933 if (sp->stats.finished)
935 progress_one_dev(sp);
936 sp->stats.duration = tv.tv_sec - sp->stats.t_start;
939 if (sp->ioctl_errno != ENOTCONN &&
940 sp->ioctl_errno != ENODEV) {
941 ret = -sp->ioctl_errno;
945 * scrub finished or device removed, check the
946 * finished flag. if unset, just use the last
947 * result we got for the current write and go
948 * on. flag should be set on next cycle, then.
950 perr = pthread_setcancelstate(
951 PTHREAD_CANCEL_DISABLE, &old);
954 perr = pthread_mutex_lock(&sp_shared->progress_mutex);
957 if (!sp_shared->stats.finished) {
958 perr = pthread_mutex_unlock(
959 &sp_shared->progress_mutex);
962 perr = pthread_setcancelstate(
963 PTHREAD_CANCEL_ENABLE, &old);
966 memcpy(sp, sp_last, sizeof(*sp));
969 perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
972 perr = pthread_setcancelstate(
973 PTHREAD_CANCEL_ENABLE, &old);
976 memcpy(sp, sp_shared, sizeof(*sp));
977 memcpy(sp_last, sp_shared, sizeof(*sp));
980 write_poll_fd.fd = peer_fd;
981 ret = poll(&write_poll_fd, 1, 0);
987 ret = scrub_write_file(
989 &spc->progress[this * ndev], ndev);
998 ret = scrub_write_progress(spc->write_mutex, fsid,
999 &spc->progress[this * ndev], ndev);
1008 return ERR_PTR(ret);
1011 static struct scrub_file_record *last_dev_scrub(
1012 struct scrub_file_record *const *const past_scrubs, u64 devid)
1016 if (!past_scrubs || IS_ERR(past_scrubs))
1019 for (i = 0; past_scrubs[i]; ++i)
1020 if (past_scrubs[i]->devid == devid)
1021 return past_scrubs[i];
1026 static int mkdir_p(char *path)
1031 for (i = 1; i < strlen(path); ++i) {
1035 ret = mkdir(path, 0777);
1036 if (ret && errno != EEXIST)
1044 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1045 struct btrfs_ioctl_dev_info_args *di_args,
1046 struct scrub_file_record **past_scrubs)
1050 if (!fi_args || !di_args || !past_scrubs)
1053 for (i = 0; i < fi_args->num_devices; i++) {
1054 struct scrub_file_record *sfr =
1055 last_dev_scrub(past_scrubs, di_args[i].devid);
1059 if (!(sfr->stats.finished || sfr->stats.canceled))
1065 static int is_scrub_running_in_kernel(int fd,
1066 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1068 struct scrub_progress sp;
1072 for (i = 0; i < max_devices; i++) {
1073 memset(&sp, 0, sizeof(sp));
1074 sp.scrub_args.devid = di_args[i].devid;
1075 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1076 if (ret < 0 && errno == ENODEV)
1078 if (ret < 0 && errno == ENOTCONN)
1087 static const char * const cmd_scrub_start_usage[];
1088 static const char * const cmd_scrub_resume_usage[];
1090 static int scrub_start(int argc, char **argv, int resume)
1100 int e_uncorrectable = 0;
1101 int e_correctable = 0;
1104 int do_background = 1;
1110 int do_stats_per_dev = 0;
1111 int ioprio_class = IOPRIO_CLASS_IDLE;
1112 int ioprio_classdata = 0;
1116 struct btrfs_ioctl_fs_info_args fi_args;
1117 struct btrfs_ioctl_dev_info_args *di_args = NULL;
1118 struct scrub_progress *sp = NULL;
1119 struct scrub_fs_stat fs_stat;
1121 struct sockaddr_un addr = {
1122 .sun_family = AF_UNIX,
1124 pthread_t *t_devs = NULL;
1126 struct scrub_file_record **past_scrubs = NULL;
1127 struct scrub_file_record *last_scrub = NULL;
1128 char *datafile = strdup(SCRUB_DATA_FILE);
1129 char fsid[BTRFS_UUID_UNPARSED_SIZE];
1130 char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1131 struct scrub_progress_cycle spc;
1132 pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1135 DIR *dirstream = NULL;
1137 int nothing_to_resume = 0;
1140 while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1148 do_stats_per_dev = 1;
1160 ioprio_class = (int)strtol(optarg, NULL, 10);
1163 ioprio_classdata = (int)strtol(optarg, NULL, 10);
1170 usage(resume ? cmd_scrub_resume_usage :
1171 cmd_scrub_start_usage);
1175 /* try to catch most error cases before forking */
1177 if (check_argc_exact(argc - optind, 1)) {
1178 usage(resume ? cmd_scrub_resume_usage :
1179 cmd_scrub_start_usage);
1182 spc.progress = NULL;
1183 if (do_quiet && do_print)
1186 if (mkdir_p(datafile)) {
1187 ERR(!do_quiet, "WARNING: cannot create scrub data "
1188 "file, mkdir %s failed: %s. Status recording "
1189 "disabled\n", datafile, strerror(errno));
1194 path = argv[optind];
1196 fdmnt = open_path_or_dev_mnt(path, &dirstream);
1199 if (errno == EINVAL)
1201 "ERROR: '%s' is not a mounted btrfs device\n",
1204 ERR(!do_quiet, "ERROR: can't access '%s': %s\n",
1205 path, strerror(errno));
1209 ret = get_fs_info(path, &fi_args, &di_args);
1211 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1212 "%s\n", strerror(-ret));
1216 if (!fi_args.num_devices) {
1217 ERR(!do_quiet, "ERROR: no devices found\n");
1222 uuid_unparse(fi_args.fsid, fsid);
1223 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1224 if (fdres < 0 && fdres != -ENOENT) {
1225 ERR(!do_quiet, "WARNING: failed to open status file: "
1226 "%s\n", strerror(-fdres));
1227 } else if (fdres >= 0) {
1228 past_scrubs = scrub_read_file(fdres, !do_quiet);
1229 if (IS_ERR(past_scrubs))
1230 ERR(!do_quiet, "WARNING: failed to read status file: "
1231 "%s\n", strerror(-PTR_ERR(past_scrubs)));
1236 * Check for stale information in the status file, ie. if it's
1237 * canceled=0, finished=0 but no scrub is running.
1239 if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1243 * check whether any involved device is already busy running a
1244 * scrub. This would cause damaged status messages and the state
1245 * "aborted" without the explanation that a scrub was already
1246 * running. Therefore check it first, prevent it and give some
1247 * feedback to the user if scrub is already running.
1248 * Note that if scrub is started with a block device as the
1249 * parameter, only that particular block device is checked. It
1250 * is a normal mode of operation to start scrub on multiple
1251 * single devices, there is no reason to prevent this.
1253 if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1255 "ERROR: scrub is already running.\n"
1256 "To cancel use 'btrfs scrub cancel %s'.\n"
1257 "To see the status use 'btrfs scrub status [-d] %s'.\n",
1263 t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1264 sp = calloc(fi_args.num_devices, sizeof(*sp));
1265 spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1267 if (!t_devs || !sp || !spc.progress) {
1268 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1273 for (i = 0; i < fi_args.num_devices; ++i) {
1274 devid = di_args[i].devid;
1275 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1277 ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1278 "%s\n", strerror(ret));
1282 last_scrub = last_dev_scrub(past_scrubs, devid);
1283 sp[i].scrub_args.devid = devid;
1285 if (resume && last_scrub && (last_scrub->stats.canceled ||
1286 !last_scrub->stats.finished)) {
1288 sp[i].scrub_args.start = last_scrub->p.last_physical;
1289 sp[i].resumed = last_scrub;
1290 } else if (resume) {
1293 sp[i].resumed = last_scrub;
1297 sp[i].scrub_args.start = 0ll;
1298 sp[i].resumed = NULL;
1301 sp[i].scrub_args.end = (u64)-1ll;
1302 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1303 sp[i].ioprio_class = ioprio_class;
1304 sp[i].ioprio_classdata = ioprio_classdata;
1307 if (!n_start && !n_resume) {
1309 printf("scrub: nothing to resume for %s, fsid %s\n",
1311 nothing_to_resume = 1;
1315 ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1317 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1318 sock_path, sizeof(sock_path));
1319 /* ignore EOVERFLOW, try using a shorter path for the socket */
1320 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1321 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1322 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1323 if (ret != -1 || errno != EADDRINUSE)
1326 * bind failed with EADDRINUSE. so let's see if anyone answers
1327 * when we make a call to the socket ...
1329 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1330 if (!ret || errno != ECONNREFUSED) {
1331 /* ... yes, so scrub must be running. error out */
1332 fprintf(stderr, "ERROR: scrub already running\n");
1338 * ... no, this means someone left us alone with an unused
1339 * socket in the file system. remove it and try again.
1341 ret = unlink(sock_path);
1344 ret = listen(prg_fd, 100);
1346 ERR(!do_quiet, "WARNING: failed to open the progress status "
1347 "socket at %s: %s. Progress cannot be queried\n",
1348 sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1359 /* write all-zero progress file for a start */
1360 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1361 fi_args.num_devices);
1363 ERR(!do_quiet, "WARNING: failed to write the progress "
1364 "status file: %s. Status recording disabled\n",
1370 if (do_background) {
1373 ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1374 "%s\n", strerror(errno));
1381 scrub_handle_sigint_parent();
1383 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1384 n_start ? "started" : "resumed",
1392 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1393 "%s\n", ret, strerror(errno));
1397 if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1398 ERR(!do_quiet, "ERROR: scrub process failed\n");
1399 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1407 scrub_handle_sigint_child(fdmnt);
1409 for (i = 0; i < fi_args.num_devices; ++i) {
1411 sp[i].scrub_args.progress = sp[i].resumed->p;
1412 sp[i].stats = sp[i].resumed->stats;
1414 sp[i].stats.finished = 1;
1417 devid = di_args[i].devid;
1418 gettimeofday(&tv, NULL);
1419 sp[i].stats.t_start = tv.tv_sec;
1420 ret = pthread_create(&t_devs[i], NULL,
1421 scrub_one_dev, &sp[i]);
1424 fprintf(stderr, "ERROR: creating "
1425 "scrub_one_dev[%llu] thread failed: "
1426 "%s\n", devid, strerror(ret));
1433 spc.prg_fd = prg_fd;
1434 spc.do_record = do_record;
1435 spc.write_mutex = &spc_write_mutex;
1436 spc.shared_progress = sp;
1438 ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1441 fprintf(stderr, "ERROR: creating progress thread "
1442 "failed: %s\n", strerror(ret));
1448 for (i = 0; i < fi_args.num_devices; ++i) {
1451 devid = di_args[i].devid;
1452 ret = pthread_join(t_devs[i], NULL);
1455 fprintf(stderr, "ERROR: pthread_join failed "
1456 "for scrub_one_dev[%llu]: %s\n", devid,
1461 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1463 fprintf(stderr, "WARNING: device %lld not "
1464 "present\n", devid);
1467 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1469 } else if (sp[i].ret) {
1471 fprintf(stderr, "ERROR: scrubbing %s failed "
1472 "for device id %lld (%s)\n", path,
1473 devid, strerror(sp[i].ioctl_errno));
1477 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1479 if (sp[i].scrub_args.progress.corrected_errors > 0
1480 || sp[i].scrub_args.progress.unverified_errors > 0)
1485 const char *append = "done";
1486 if (!do_stats_per_dev)
1487 init_fs_stat(&fs_stat);
1488 for (i = 0; i < fi_args.num_devices; ++i) {
1489 if (do_stats_per_dev) {
1490 print_scrub_dev(&di_args[i],
1491 &sp[i].scrub_args.progress,
1493 sp[i].ret ? "canceled" : "done",
1497 append = "canceled";
1498 add_to_fs_stat(&sp[i].scrub_args.progress,
1499 &sp[i].stats, &fs_stat);
1502 if (!do_stats_per_dev) {
1503 printf("scrub %s for %s\n", append, fsid);
1504 print_fs_stat(&fs_stat, print_raw);
1508 ret = pthread_cancel(t_prog);
1510 ret = pthread_join(t_prog, &terr);
1512 /* check for errors from the handling of the progress thread */
1513 if (do_print && ret) {
1514 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1518 /* check for errors returned from the progress thread itself */
1519 if (do_print && terr && terr != PTHREAD_CANCELED) {
1520 fprintf(stderr, "ERROR: recording progress "
1521 "failed: %s\n", strerror(-PTR_ERR(terr)));
1525 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1526 fi_args.num_devices);
1527 if (ret && do_print) {
1528 fprintf(stderr, "ERROR: failed to record the result: "
1529 "%s\n", strerror(-ret));
1533 scrub_handle_sigint_child(-1);
1536 free_history(past_scrubs);
1546 close_file_or_dir(fdmnt, dirstream);
1550 if (nothing_to_resume)
1552 if (e_uncorrectable) {
1553 ERR(!do_quiet, "ERROR: There are uncorrectable errors.\n");
1557 ERR(!do_quiet, "WARNING: errors detected during scrubbing, corrected.\n");
1562 static const char * const cmd_scrub_start_usage[] = {
1563 "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1564 "Start a new scrub. If a scrub is already running, the new one fails.",
1566 "-B do not background",
1567 "-d stats per device (-B only)",
1569 "-r read only mode",
1570 "-R raw print mode, print full data instead of summary",
1571 "-c set ioprio class (see ionice(1) manpage)",
1572 "-n set ioprio classdata (see ionice(1) manpage)",
1573 "-f force starting new scrub even if a scrub is already running",
1574 " this is useful when scrub stats record file is damaged",
1578 static int cmd_scrub_start(int argc, char **argv)
1580 return scrub_start(argc, argv, 0);
1583 static const char * const cmd_scrub_cancel_usage[] = {
1584 "btrfs scrub cancel <path>|<device>",
1585 "Cancel a running scrub",
1589 static int cmd_scrub_cancel(int argc, char **argv)
1594 DIR *dirstream = NULL;
1596 if (check_argc_exact(argc, 2))
1597 usage(cmd_scrub_cancel_usage);
1601 fdmnt = open_path_or_dev_mnt(path, &dirstream);
1603 if (errno == EINVAL)
1605 "ERROR: '%s' is not a mounted btrfs device\n",
1608 fprintf(stderr, "ERROR: can't access '%s': %s\n",
1609 path, strerror(errno));
1614 ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1617 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1618 errno == ENOTCONN ? "not running" : strerror(errno));
1619 if (errno == ENOTCONN)
1627 printf("scrub cancelled\n");
1630 close_file_or_dir(fdmnt, dirstream);
1634 static const char * const cmd_scrub_resume_usage[] = {
1635 "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1636 "Resume previously canceled or interrupted scrub",
1638 "-B do not background",
1639 "-d stats per device (-B only)",
1641 "-r read only mode",
1642 "-R raw print mode, print full data instead of summary",
1643 "-c set ioprio class (see ionice(1) manpage)",
1644 "-n set ioprio classdata (see ionice(1) manpage)",
1648 static int cmd_scrub_resume(int argc, char **argv)
1650 return scrub_start(argc, argv, 1);
1653 static const char * const cmd_scrub_status_usage[] = {
1654 "btrfs scrub status [-dR] <path>|<device>",
1655 "Show status of running or finished scrub",
1657 "-d stats per device",
1658 "-R print raw stats",
1662 static int cmd_scrub_status(int argc, char **argv)
1665 struct btrfs_ioctl_fs_info_args fi_args;
1666 struct btrfs_ioctl_dev_info_args *di_args = NULL;
1667 struct scrub_file_record **past_scrubs = NULL;
1668 struct scrub_file_record *last_scrub;
1669 struct scrub_fs_stat fs_stat;
1670 struct sockaddr_un addr = {
1671 .sun_family = AF_UNIX,
1678 int do_stats_per_dev = 0;
1680 char fsid[BTRFS_UUID_UNPARSED_SIZE];
1683 DIR *dirstream = NULL;
1686 while ((c = getopt(argc, argv, "dR")) != -1) {
1689 do_stats_per_dev = 1;
1696 usage(cmd_scrub_status_usage);
1700 if (check_argc_exact(argc - optind, 1))
1701 usage(cmd_scrub_status_usage);
1703 path = argv[optind];
1705 fdmnt = open_path_or_dev_mnt(path, &dirstream);
1708 if (errno == EINVAL)
1710 "ERROR: '%s' is not a mounted btrfs device\n",
1713 fprintf(stderr, "ERROR: can't access '%s': %s\n",
1714 path, strerror(errno));
1718 ret = get_fs_info(path, &fi_args, &di_args);
1720 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1721 "%s\n", strerror(-ret));
1725 if (!fi_args.num_devices) {
1726 fprintf(stderr, "ERROR: no devices found\n");
1731 uuid_unparse(fi_args.fsid, fsid);
1733 fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1735 fprintf(stderr, "ERROR: failed to create socket to "
1736 "receive progress information: %s\n",
1741 scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1742 NULL, addr.sun_path, sizeof(addr.sun_path));
1743 /* ignore EOVERFLOW, just use shorter name and hope for the best */
1744 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1745 ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1748 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1749 if (fdres < 0 && fdres != -ENOENT) {
1750 fprintf(stderr, "WARNING: failed to open status file: "
1751 "%s\n", strerror(-fdres));
1758 past_scrubs = scrub_read_file(fdres, 1);
1759 if (IS_ERR(past_scrubs))
1760 fprintf(stderr, "WARNING: failed to read status: %s\n",
1761 strerror(-PTR_ERR(past_scrubs)));
1763 in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1765 printf("scrub status for %s\n", fsid);
1767 if (do_stats_per_dev) {
1768 for (i = 0; i < fi_args.num_devices; ++i) {
1769 last_scrub = last_dev_scrub(past_scrubs,
1772 print_scrub_dev(&di_args[i], NULL, print_raw,
1776 last_scrub->stats.in_progress = in_progress;
1777 print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1778 last_scrub->stats.finished ?
1779 "history" : "status",
1780 &last_scrub->stats);
1783 init_fs_stat(&fs_stat);
1784 fs_stat.s.in_progress = in_progress;
1785 for (i = 0; i < fi_args.num_devices; ++i) {
1786 last_scrub = last_dev_scrub(past_scrubs,
1790 add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1793 print_fs_stat(&fs_stat, print_raw);
1797 free_history(past_scrubs);
1801 close_file_or_dir(fdmnt, dirstream);
1806 const struct cmd_group scrub_cmd_group = {
1807 scrub_cmd_group_usage, NULL, {
1808 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1809 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1810 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1811 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1816 int cmd_scrub(int argc, char **argv)
1818 return handle_command_group(&scrub_cmd_group, argc, argv);