605af45b5a8e2610e0e44ffec1c3b17b435d6d44
[platform/upstream/btrfs-progs.git] / cmds-scrub.c
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18
19 #include "kerncompat.h"
20
21 #include <sys/ioctl.h>
22 #include <sys/wait.h>
23 #include <sys/stat.h>
24 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/un.h>
27 #include <sys/syscall.h>
28 #include <poll.h>
29 #include <sys/file.h>
30 #include <uuid/uuid.h>
31 #include <fcntl.h>
32 #include <unistd.h>
33 #include <pthread.h>
34 #include <ctype.h>
35 #include <signal.h>
36 #include <stdarg.h>
37
38 #include "ctree.h"
39 #include "ioctl.h"
40 #include "utils.h"
41 #include "volumes.h"
42 #include "disk-io.h"
43
44 #include "commands.h"
45
46 static const char * const scrub_cmd_group_usage[] = {
47         "btrfs scrub <command> [options] <path>|<device>",
48         NULL
49 };
50
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"
55
56 struct scrub_stats {
57         time_t t_start;
58         time_t t_resumed;
59         u64 duration;
60         u64 finished;
61         u64 canceled;
62 };
63
64 /* TBD: replace with #include "linux/ioprio.h" in some years */
65 #if !defined (IOPRIO_H)
66 #define IOPRIO_WHO_PROCESS 1
67 #define IOPRIO_CLASS_SHIFT 13
68 #define IOPRIO_PRIO_VALUE(class, data) \
69                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
70 #define IOPRIO_CLASS_IDLE 3
71 #endif
72
73 struct scrub_progress {
74         struct btrfs_ioctl_scrub_args scrub_args;
75         int fd;
76         int ret;
77         int skip;
78         struct scrub_stats stats;
79         struct scrub_file_record *resumed;
80         int ioctl_errno;
81         pthread_mutex_t progress_mutex;
82         int ioprio_class;
83         int ioprio_classdata;
84 };
85
86 struct scrub_file_record {
87         u8 fsid[BTRFS_FSID_SIZE];
88         u64 devid;
89         struct scrub_stats stats;
90         struct btrfs_scrub_progress p;
91 };
92
93 struct scrub_progress_cycle {
94         int fdmnt;
95         int prg_fd;
96         int do_record;
97         struct btrfs_ioctl_fs_info_args *fi;
98         struct scrub_progress *progress;
99         struct scrub_progress *shared_progress;
100         pthread_mutex_t *write_mutex;
101 };
102
103 struct scrub_fs_stat {
104         struct btrfs_scrub_progress p;
105         struct scrub_stats s;
106         int i;
107 };
108
109 static void print_scrub_full(struct btrfs_scrub_progress *sp)
110 {
111         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
112         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
113         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
114         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
115         printf("\tread_errors: %lld\n", sp->read_errors);
116         printf("\tcsum_errors: %lld\n", sp->csum_errors);
117         printf("\tverify_errors: %lld\n", sp->verify_errors);
118         printf("\tno_csum: %lld\n", sp->no_csum);
119         printf("\tcsum_discards: %lld\n", sp->csum_discards);
120         printf("\tsuper_errors: %lld\n", sp->super_errors);
121         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
122         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
123         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
124         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
125         printf("\tlast_physical: %lld\n", sp->last_physical);
126 }
127
128 #define ERR(test, ...) do {                     \
129         if (test)                               \
130                 fprintf(stderr, __VA_ARGS__);   \
131 } while (0)
132
133 #define PRINT_SCRUB_ERROR(test, desc) do {      \
134         if (test)                               \
135                 printf(" %s=%llu", desc, test); \
136 } while (0)
137
138 static void print_scrub_summary(struct btrfs_scrub_progress *p)
139 {
140         u64 err_cnt;
141         u64 err_cnt2;
142
143         err_cnt = p->read_errors +
144                         p->csum_errors +
145                         p->verify_errors +
146                         p->super_errors;
147
148         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
149
150         if (p->malloc_errors)
151                 printf("*** WARNING: memory allocation failed while scrubbing. "
152                        "results may be inaccurate\n");
153
154         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
155                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
156                 max(err_cnt, err_cnt2));
157
158         if (err_cnt || err_cnt2) {
159                 printf("\terror details:");
160                 PRINT_SCRUB_ERROR(p->read_errors, "read");
161                 PRINT_SCRUB_ERROR(p->super_errors, "super");
162                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
163                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
164                 printf("\n");
165                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
166                         "unverified errors: %llu\n", p->corrected_errors,
167                         p->uncorrectable_errors, p->unverified_errors);
168         }
169 }
170
171 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
172         fs_stat->p.name += p->name;             \
173 } while (0)
174
175 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
176 do {                                            \
177         if (fs_stat->s.name > ss->name) {       \
178                 fs_stat->s.name = ss->name;     \
179         }                                       \
180 } while (0)
181
182 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
183 do {                                                            \
184         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
185                 fs_stat->s.name = ss->name;                     \
186         }                                                       \
187 } while (0)
188
189 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
190 do {                                                                    \
191         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
192                 (fs_stat)->s.name = (ss)->name;                         \
193         }                                                               \
194 } while (0)
195
196 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
197                                 struct scrub_stats *ss,
198                                 struct scrub_fs_stat *fs_stat)
199 {
200         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
201         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
202         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
203         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
204         _SCRUB_FS_STAT(p, read_errors, fs_stat);
205         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
206         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
207         _SCRUB_FS_STAT(p, no_csum, fs_stat);
208         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
209         _SCRUB_FS_STAT(p, super_errors, fs_stat);
210         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
211         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
212         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
213         _SCRUB_FS_STAT(p, last_physical, fs_stat);
214         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
215         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
216         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
217         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
218         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
219 }
220
221 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
222 {
223         memset(fs_stat, 0, sizeof(*fs_stat));
224         fs_stat->s.finished = 1;
225 }
226
227 static void _print_scrub_ss(struct scrub_stats *ss)
228 {
229         char t[4096];
230         struct tm tm;
231
232         if (!ss || !ss->t_start) {
233                 printf("\tno stats available\n");
234                 return;
235         }
236         if (ss->t_resumed) {
237                 localtime_r(&ss->t_resumed, &tm);
238                 strftime(t, sizeof(t), "%c", &tm);
239                 t[sizeof(t) - 1] = '\0';
240                 printf("\tscrub resumed at %s", t);
241         } else {
242                 localtime_r(&ss->t_start, &tm);
243                 strftime(t, sizeof(t), "%c", &tm);
244                 t[sizeof(t) - 1] = '\0';
245                 printf("\tscrub started at %s", t);
246         }
247         if (ss->finished && !ss->canceled) {
248                 printf(" and finished after %llu seconds\n",
249                        ss->duration);
250         } else if (ss->canceled) {
251                 printf(" and was aborted after %llu seconds\n",
252                        ss->duration);
253         } else {
254                 printf(", running for %llu seconds\n", ss->duration);
255         }
256 }
257
258 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
259                                 struct btrfs_scrub_progress *p, int raw,
260                                 const char *append, struct scrub_stats *ss)
261 {
262         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
263                append ? append : "");
264
265         _print_scrub_ss(ss);
266
267         if (p) {
268                 if (raw)
269                         print_scrub_full(p);
270                 else
271                         print_scrub_summary(p);
272         }
273 }
274
275 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
276 {
277         _print_scrub_ss(&fs_stat->s);
278
279         if (raw)
280                 print_scrub_full(&fs_stat->p);
281         else
282                 print_scrub_summary(&fs_stat->p);
283 }
284
285 static void free_history(struct scrub_file_record **last_scrubs)
286 {
287         struct scrub_file_record **l = last_scrubs;
288         if (!l)
289                 return;
290         while (*l)
291                 free(*l++);
292         free(last_scrubs);
293 }
294
295 /*
296  * cancels a running scrub and makes the master process record the current
297  * progress status before exiting.
298  */
299 static int cancel_fd = -1;
300 static void scrub_sigint_record_progress(int signal)
301 {
302         int ret;
303
304         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
305         if (ret < 0)
306                 perror("Scrub cancel failed");
307 }
308
309 static int scrub_handle_sigint_parent(void)
310 {
311         struct sigaction sa = {
312                 .sa_handler = SIG_IGN,
313                 .sa_flags = SA_RESTART,
314         };
315
316         return sigaction(SIGINT, &sa, NULL);
317 }
318
319 static int scrub_handle_sigint_child(int fd)
320 {
321         struct sigaction sa = {
322                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
323         };
324
325         cancel_fd = fd;
326         return sigaction(SIGINT, &sa, NULL);
327 }
328
329 static int scrub_datafile(const char *fn_base, const char *fn_local,
330                                 const char *fn_tmp, char *datafile, int size)
331 {
332         int ret;
333         int end = size - 2;
334
335         datafile[end + 1] = '\0';
336         strncpy(datafile, fn_base, end);
337         ret = strlen(datafile);
338
339         if (ret + 1 > end)
340                 return -EOVERFLOW;
341
342         datafile[ret] = '.';
343         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
344         ret = strlen(datafile);
345
346         if (ret + 1 > end)
347                 return -EOVERFLOW;
348
349         if (fn_tmp) {
350                 datafile[ret] = '_';
351                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
352                 ret = strlen(datafile);
353
354                 if (ret > end)
355                         return -EOVERFLOW;
356         }
357
358         return 0;
359 }
360
361 static int scrub_open_file(const char *datafile, int m)
362 {
363         int fd;
364         int ret;
365
366         fd = open(datafile, m, 0600);
367         if (fd < 0)
368                 return -errno;
369
370         ret = flock(fd, LOCK_EX|LOCK_NB);
371         if (ret) {
372                 ret = errno;
373                 close(fd);
374                 return -ret;
375         }
376
377         return fd;
378 }
379
380 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
381 {
382         int ret;
383         char datafile[BTRFS_PATH_NAME_MAX + 1];
384         ret = scrub_datafile(fn_base, fn_local, NULL,
385                                 datafile, sizeof(datafile));
386         if (ret < 0)
387                 return ret;
388         return scrub_open_file(datafile, O_RDONLY);
389 }
390
391 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
392                                 const char *tmp)
393 {
394         int ret;
395         char datafile[BTRFS_PATH_NAME_MAX + 1];
396         ret = scrub_datafile(fn_base, fn_local, tmp,
397                                 datafile, sizeof(datafile));
398         if (ret < 0)
399                 return ret;
400         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
401 }
402
403 static int scrub_rename_file(const char *fn_base, const char *fn_local,
404                                 const char *tmp)
405 {
406         int ret;
407         char datafile_old[BTRFS_PATH_NAME_MAX + 1];
408         char datafile_new[BTRFS_PATH_NAME_MAX + 1];
409         ret = scrub_datafile(fn_base, fn_local, tmp,
410                                 datafile_old, sizeof(datafile_old));
411         if (ret < 0)
412                 return ret;
413         ret = scrub_datafile(fn_base, fn_local, NULL,
414                                 datafile_new, sizeof(datafile_new));
415         if (ret < 0)
416                 return ret;
417         ret = rename(datafile_old, datafile_new);
418         return ret ? -errno : 0;
419 }
420
421 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
422         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
423 }
424
425 /*
426  * returns 0 if the key did not match (nothing was read)
427  *         1 if the key did match (success)
428  *        -1 if the key did match and an error occured
429  */
430 static int scrub_kvread(int *i, int len, int avail, const char *buf,
431                         const char *key, u64 *dest)
432 {
433         int j;
434
435         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
436                 *i += len - 1;
437                 if (buf[*i] != ':')
438                         return -1;
439                 *i += 1;
440                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
441                         ;
442                 if (*i + j >= avail)
443                         return -1;
444                 *dest = atoll(&buf[*i]);
445                 *i += j;
446                 return 1;
447         }
448
449         return 0;
450 }
451
452 #define _SCRUB_INVALID do {                                             \
453         if (report_errors)                                              \
454                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
455                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
456                         lineno, i, state, 20 > avail ? avail : 20,      \
457                         l + i,  __FILE__, __LINE__);                    \
458         goto skip;                                                      \
459 } while (0)
460
461 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
462 {
463         int avail = 0;
464         int old_avail = 0;
465         char l[16 * 1024];
466         int state = 0;
467         int curr = -1;
468         int i = 0;
469         int j;
470         int ret;
471         int eof = 0;
472         int lineno = 0;
473         u64 version;
474         char empty_uuid[BTRFS_FSID_SIZE] = {0};
475         struct scrub_file_record **p = NULL;
476
477         if (fd < 0)
478                 return ERR_PTR(-EINVAL);
479
480 again:
481         old_avail = avail - i;
482         BUG_ON(old_avail < 0);
483         if (old_avail)
484                 memmove(l, l + i, old_avail);
485         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
486         if (avail == 0)
487                 eof = 1;
488         if (avail == 0 && old_avail == 0) {
489                 if (curr >= 0 &&
490                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
491                         p[curr] = NULL;
492                 } else if (curr == -1) {
493                         p = ERR_PTR(-ENODATA);
494                 }
495                 return p;
496         }
497         if (avail == -1)
498                 return ERR_PTR(-errno);
499         avail += old_avail;
500
501         i = 0;
502         while (i < avail) {
503                 switch (state) {
504                 case 0: /* start of file */
505                         ret = scrub_kvread(&i,
506                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
507                                 SCRUB_FILE_VERSION_PREFIX, &version);
508                         if (ret != 1)
509                                 _SCRUB_INVALID;
510                         if (version != atoll(SCRUB_FILE_VERSION))
511                                 return ERR_PTR(-ENOTSUP);
512                         state = 6;
513                         continue;
514                 case 1: /* start of line, alloc */
515                         /*
516                          * this state makes sure we have a complete line in
517                          * further processing, so we don't need wrap-tracking
518                          * everywhere.
519                          */
520                         if (!eof && !memchr(l + i, '\n', avail - i))
521                                 goto again;
522                         ++lineno;
523                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
524                                                 BTRFS_FSID_SIZE) == 0) {
525                                 state = 2;
526                                 continue;
527                         }
528                         ++curr;
529                         p = realloc(p, (curr + 2) * sizeof(*p));
530                         if (p)
531                                 p[curr] = malloc(sizeof(**p));
532                         if (!p || !p[curr])
533                                 return ERR_PTR(-errno);
534                         memset(p[curr], 0, sizeof(**p));
535                         p[curr + 1] = NULL;
536                         ++state;
537                         /* fall through */
538                 case 2: /* start of line, skip space */
539                         while (isspace(l[i]) && i < avail) {
540                                 if (l[i] == '\n')
541                                         ++lineno;
542                                 ++i;
543                         }
544                         if (i >= avail ||
545                             (!eof && !memchr(l + i, '\n', avail - i)))
546                                 goto again;
547                         ++state;
548                         /* fall through */
549                 case 3: /* read fsid */
550                         if (i == avail)
551                                 continue;
552                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
553                                 ;
554                         if (i + j + 1 >= avail)
555                                 _SCRUB_INVALID;
556                         if (j != 36)
557                                 _SCRUB_INVALID;
558                         l[i + j] = '\0';
559                         ret = uuid_parse(l + i, p[curr]->fsid);
560                         if (ret)
561                                 _SCRUB_INVALID;
562                         i += j + 1;
563                         ++state;
564                         /* fall through */
565                 case 4: /* read dev id */
566                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
567                                 ;
568                         if (j == 0 || i + j + 1 >= avail)
569                                 _SCRUB_INVALID;
570                         p[curr]->devid = atoll(&l[i]);
571                         i += j + 1;
572                         ++state;
573                         /* fall through */
574                 case 5: /* read key/value pair */
575                         ret = 0;
576                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
577                                         &p[curr]->p);
578                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
579                                         &p[curr]->p);
580                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
581                                         &p[curr]->p);
582                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
583                                         &p[curr]->p);
584                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
585                                         &p[curr]->p);
586                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
587                                         &p[curr]->p);
588                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
589                                         &p[curr]->p);
590                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
591                                         &p[curr]->p);
592                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
593                                         &p[curr]->p);
594                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
595                                         &p[curr]->p);
596                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
597                                         &p[curr]->p);
598                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
599                                         &p[curr]->p);
600                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
601                                         &p[curr]->p);
602                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
603                                         &p[curr]->p);
604                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
605                                         &p[curr]->p);
606                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
607                                         &p[curr]->stats);
608                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
609                                         (u64 *)&p[curr]->stats);
610                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
611                                         (u64 *)&p[curr]->stats);
612                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
613                                         (u64 *)&p[curr]->stats);
614                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
615                                         &p[curr]->stats);
616                         if (ret != 1)
617                                 _SCRUB_INVALID;
618                         ++state;
619                         /* fall through */
620                 case 6: /* after number */
621                         if (l[i] == '|')
622                                 state = 5;
623                         else if (l[i] == '\n')
624                                 state = 1;
625                         else
626                                 _SCRUB_INVALID;
627                         ++i;
628                         continue;
629                 case 99: /* skip rest of line */
630 skip:
631                         state = 99;
632                         do {
633                                 ++i;
634                                 if (l[i - 1] == '\n') {
635                                         state = 1;
636                                         break;
637                                 }
638                         } while (i < avail);
639                         continue;
640                 }
641                 BUG();
642         }
643         goto again;
644 }
645
646 static int scrub_write_buf(int fd, const void *data, int len)
647 {
648         int ret;
649         ret = write(fd, data, len);
650         return ret - len;
651 }
652
653 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
654                                 __attribute__ ((format (printf, 4, 5)));
655 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
656 {
657         int ret;
658         va_list args;
659
660         va_start(args, fmt);
661         ret = vsnprintf(buf, max, fmt, args);
662         va_end(args);
663         if (ret >= max)
664                 return ret - max;
665         return scrub_write_buf(fd, buf, ret);
666 }
667
668 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
669                         data->resumed->p.name + data->scrub_args.progress.name
670
671 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
672                                                   struct scrub_progress *dest)
673 {
674         if (!data->resumed || data->skip)
675                 return data;
676
677         _SCRUB_SUM(dest, data, data_extents_scrubbed);
678         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
679         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
680         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
681         _SCRUB_SUM(dest, data, read_errors);
682         _SCRUB_SUM(dest, data, csum_errors);
683         _SCRUB_SUM(dest, data, verify_errors);
684         _SCRUB_SUM(dest, data, no_csum);
685         _SCRUB_SUM(dest, data, csum_discards);
686         _SCRUB_SUM(dest, data, super_errors);
687         _SCRUB_SUM(dest, data, malloc_errors);
688         _SCRUB_SUM(dest, data, uncorrectable_errors);
689         _SCRUB_SUM(dest, data, corrected_errors);
690         _SCRUB_SUM(dest, data, last_physical);
691         dest->stats.canceled = data->stats.canceled;
692         dest->stats.finished = data->stats.finished;
693         dest->stats.t_resumed = data->stats.t_start;
694         dest->stats.t_start = data->resumed->stats.t_start;
695         dest->stats.duration = data->resumed->stats.duration +
696                                                         data->stats.duration;
697         dest->scrub_args.devid = data->scrub_args.devid;
698         return dest;
699 }
700
701 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
702         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
703                         use->scrub_args.progress.name)
704
705 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
706         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
707                         use->stats.name)
708
709 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
710 {
711         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
712 }
713
714 static int scrub_write_file(int fd, const char *fsid,
715                                 struct scrub_progress *data, int n)
716 {
717         int ret = 0;
718         int i;
719         char buf[1024];
720         struct scrub_progress local;
721         struct scrub_progress *use;
722
723         if (n < 1)
724                 return -EINVAL;
725
726         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
727         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
728                                 SCRUB_FILE_VERSION "\n",
729                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
730                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
731         if (ret)
732                 return -EOVERFLOW;
733
734         for (i = 0; i < n; ++i) {
735                 use = scrub_resumed_stats(&data[i], &local);
736                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
737                     scrub_write_buf(fd, ":", 1) ||
738                     scrub_writev(fd, buf, sizeof(buf), "%lld",
739                                         use->scrub_args.devid) ||
740                     scrub_write_buf(fd, buf, ret) ||
741                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
742                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
743                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
744                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
745                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
746                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
747                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
748                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
749                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
750                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
751                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
752                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
753                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
754                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
755                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
756                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
757                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
758                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
759                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
760                     scrub_write_buf(fd, "\n", 1)) {
761                         return -EOVERFLOW;
762                 }
763         }
764
765         return 0;
766 }
767
768 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
769                                 struct scrub_progress *data, int n)
770 {
771         int ret;
772         int err;
773         int fd = -1;
774         int old;
775
776         ret = pthread_mutex_lock(m);
777         if (ret) {
778                 err = -ret;
779                 goto out;
780         }
781
782         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
783         if (ret) {
784                 err = -ret;
785                 goto out;
786         }
787
788         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
789         if (fd < 0) {
790                 err = fd;
791                 goto out;
792         }
793         err = scrub_write_file(fd, fsid, data, n);
794         if (err)
795                 goto out;
796         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
797         if (err)
798                 goto out;
799
800 out:
801         if (fd >= 0) {
802                 ret = close(fd);
803                 if (ret)
804                         err = -errno;
805         }
806
807         ret = pthread_mutex_unlock(m);
808         if (ret && !err)
809                 err = -ret;
810
811         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
812         if (ret && !err)
813                 err = -ret;
814
815         return err;
816 }
817
818 static void *scrub_one_dev(void *ctx)
819 {
820         struct scrub_progress *sp = ctx;
821         int ret;
822         struct timeval tv;
823
824         sp->stats.canceled = 0;
825         sp->stats.duration = 0;
826         sp->stats.finished = 0;
827
828         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
829                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
830                                         sp->ioprio_classdata));
831         if (ret)
832                 fprintf(stderr,
833                         "WARNING: setting ioprio failed: %s (ignored).\n",
834                         strerror(errno));
835
836         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
837         gettimeofday(&tv, NULL);
838         sp->ret = ret;
839         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
840         sp->stats.canceled = !!ret;
841         sp->ioctl_errno = errno;
842         ret = pthread_mutex_lock(&sp->progress_mutex);
843         if (ret)
844                 return ERR_PTR(-ret);
845         sp->stats.finished = 1;
846         ret = pthread_mutex_unlock(&sp->progress_mutex);
847         if (ret)
848                 return ERR_PTR(-ret);
849
850         return NULL;
851 }
852
853 static void *progress_one_dev(void *ctx)
854 {
855         struct scrub_progress *sp = ctx;
856
857         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
858         sp->ioctl_errno = errno;
859
860         return NULL;
861 }
862
863 /* nb: returns a negative errno via ERR_PTR */
864 static void *scrub_progress_cycle(void *ctx)
865 {
866         int ret;
867         int  perr = 0;  /* positive / pthread error returns */
868         int old;
869         int i;
870         char fsid[37];
871         struct scrub_progress *sp;
872         struct scrub_progress *sp_last;
873         struct scrub_progress *sp_shared;
874         struct timeval tv;
875         struct scrub_progress_cycle *spc = ctx;
876         int ndev = spc->fi->num_devices;
877         int this = 1;
878         int last = 0;
879         int peer_fd = -1;
880         struct pollfd accept_poll_fd = {
881                 .fd = spc->prg_fd,
882                 .events = POLLIN,
883                 .revents = 0,
884         };
885         struct pollfd write_poll_fd = {
886                 .events = POLLOUT,
887                 .revents = 0,
888         };
889         struct sockaddr_un peer;
890         socklen_t peer_size = sizeof(peer);
891
892         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
893         if (perr)
894                 goto out;
895
896         uuid_unparse(spc->fi->fsid, fsid);
897
898         for (i = 0; i < ndev; ++i) {
899                 sp = &spc->progress[i];
900                 sp_last = &spc->progress[i + ndev];
901                 sp_shared = &spc->shared_progress[i];
902                 sp->scrub_args.devid = sp_last->scrub_args.devid =
903                                                 sp_shared->scrub_args.devid;
904                 sp->fd = sp_last->fd = spc->fdmnt;
905                 sp->stats.t_start = sp_last->stats.t_start =
906                                                 sp_shared->stats.t_start;
907                 sp->resumed = sp_last->resumed = sp_shared->resumed;
908                 sp->skip = sp_last->skip = sp_shared->skip;
909                 sp->stats.finished = sp_last->stats.finished =
910                                                 sp_shared->stats.finished;
911         }
912
913         while (1) {
914                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
915                 if (ret == -1) {
916                         ret = -errno;
917                         goto out;
918                 }
919                 if (ret)
920                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
921                                          &peer_size);
922                 gettimeofday(&tv, NULL);
923                 this = (this + 1)%2;
924                 last = (last + 1)%2;
925                 for (i = 0; i < ndev; ++i) {
926                         sp = &spc->progress[this * ndev + i];
927                         sp_last = &spc->progress[last * ndev + i];
928                         sp_shared = &spc->shared_progress[i];
929                         if (sp->stats.finished)
930                                 continue;
931                         progress_one_dev(sp);
932                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
933                         if (!sp->ret)
934                                 continue;
935                         if (sp->ioctl_errno != ENOTCONN &&
936                             sp->ioctl_errno != ENODEV) {
937                                 ret = -sp->ioctl_errno;
938                                 goto out;
939                         }
940                         /*
941                          * scrub finished or device removed, check the
942                          * finished flag. if unset, just use the last
943                          * result we got for the current write and go
944                          * on. flag should be set on next cycle, then.
945                          */
946                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
947                         if (perr)
948                                 goto out;
949                         if (!sp_shared->stats.finished) {
950                                 perr = pthread_mutex_unlock(
951                                                 &sp_shared->progress_mutex);
952                                 if (perr)
953                                         goto out;
954                                 memcpy(sp, sp_last, sizeof(*sp));
955                                 continue;
956                         }
957                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
958                         if (perr)
959                                 goto out;
960                         memcpy(sp, sp_shared, sizeof(*sp));
961                         memcpy(sp_last, sp_shared, sizeof(*sp));
962                 }
963                 if (peer_fd != -1) {
964                         write_poll_fd.fd = peer_fd;
965                         ret = poll(&write_poll_fd, 1, 0);
966                         if (ret == -1) {
967                                 ret = -errno;
968                                 goto out;
969                         }
970                         if (ret) {
971                                 ret = scrub_write_file(
972                                         peer_fd, fsid,
973                                         &spc->progress[this * ndev], ndev);
974                                 if (ret)
975                                         goto out;
976                         }
977                         close(peer_fd);
978                         peer_fd = -1;
979                 }
980                 if (!spc->do_record)
981                         continue;
982                 ret = scrub_write_progress(spc->write_mutex, fsid,
983                                            &spc->progress[this * ndev], ndev);
984                 if (ret)
985                         goto out;
986         }
987 out:
988         if (peer_fd != -1)
989                 close(peer_fd);
990         if (perr)
991                 ret = -perr;
992         return ERR_PTR(ret);
993 }
994
995 static struct scrub_file_record *last_dev_scrub(
996                 struct scrub_file_record *const *const past_scrubs, u64 devid)
997 {
998         int i;
999
1000         if (!past_scrubs || IS_ERR(past_scrubs))
1001                 return NULL;
1002
1003         for (i = 0; past_scrubs[i]; ++i)
1004                 if (past_scrubs[i]->devid == devid)
1005                         return past_scrubs[i];
1006
1007         return NULL;
1008 }
1009
1010 static int mkdir_p(char *path)
1011 {
1012         int i;
1013         int ret;
1014
1015         for (i = 1; i < strlen(path); ++i) {
1016                 if (path[i] != '/')
1017                         continue;
1018                 path[i] = '\0';
1019                 ret = mkdir(path, 0777);
1020                 if (ret && errno != EEXIST)
1021                         return -errno;
1022                 path[i] = '/';
1023         }
1024
1025         return 0;
1026 }
1027
1028 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1029                                   struct btrfs_ioctl_dev_info_args *di_args,
1030                                   struct scrub_file_record **past_scrubs)
1031 {
1032         int i;
1033
1034         if (!fi_args || !di_args || !past_scrubs)
1035                 return 0;
1036
1037         for (i = 0; i < fi_args->num_devices; i++) {
1038                 struct scrub_file_record *sfr =
1039                         last_dev_scrub(past_scrubs, di_args[i].devid);
1040
1041                 if (!sfr)
1042                         continue;
1043                 if (!(sfr->stats.finished || sfr->stats.canceled))
1044                         return 1;
1045         }
1046         return 0;
1047 }
1048
1049 static const char * const cmd_scrub_start_usage[];
1050 static const char * const cmd_scrub_resume_usage[];
1051
1052 static int scrub_start(int argc, char **argv, int resume)
1053 {
1054         int fdmnt;
1055         int prg_fd = -1;
1056         int fdres = -1;
1057         int ret;
1058         pid_t pid;
1059         int c;
1060         int i;
1061         int err = 0;
1062         int e_uncorrectable = 0;
1063         int e_correctable = 0;
1064         int print_raw = 0;
1065         char *path;
1066         int do_background = 1;
1067         int do_wait = 0;
1068         int do_print = 0;
1069         int do_quiet = 0;
1070         int do_record = 1;
1071         int readonly = 0;
1072         int do_stats_per_dev = 0;
1073         int ioprio_class = IOPRIO_CLASS_IDLE;
1074         int ioprio_classdata = 0;
1075         int n_start = 0;
1076         int n_skip = 0;
1077         int n_resume = 0;
1078         struct btrfs_ioctl_fs_info_args fi_args;
1079         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1080         struct scrub_progress *sp = NULL;
1081         struct scrub_fs_stat fs_stat;
1082         struct timeval tv;
1083         struct sockaddr_un addr = {
1084                 .sun_family = AF_UNIX,
1085         };
1086         pthread_t *t_devs = NULL;
1087         pthread_t t_prog;
1088         pthread_attr_t t_attr;
1089         struct scrub_file_record **past_scrubs = NULL;
1090         struct scrub_file_record *last_scrub = NULL;
1091         char *datafile = strdup(SCRUB_DATA_FILE);
1092         char fsid[37];
1093         char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1094         struct scrub_progress_cycle spc;
1095         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1096         void *terr;
1097         u64 devid;
1098         DIR *dirstream = NULL;
1099
1100         optind = 1;
1101         while ((c = getopt(argc, argv, "BdqrRc:n:")) != -1) {
1102                 switch (c) {
1103                 case 'B':
1104                         do_background = 0;
1105                         do_wait = 1;
1106                         do_print = 1;
1107                         break;
1108                 case 'd':
1109                         do_stats_per_dev = 1;
1110                         break;
1111                 case 'q':
1112                         do_quiet = 1;
1113                         break;
1114                 case 'r':
1115                         readonly = 1;
1116                         break;
1117                 case 'R':
1118                         print_raw = 1;
1119                         break;
1120                 case 'c':
1121                         ioprio_class = (int)strtol(optarg, NULL, 10);
1122                         break;
1123                 case 'n':
1124                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1125                         break;
1126                 case '?':
1127                 default:
1128                         usage(resume ? cmd_scrub_resume_usage :
1129                                                 cmd_scrub_start_usage);
1130                 }
1131         }
1132
1133         /* try to catch most error cases before forking */
1134
1135         if (check_argc_exact(argc - optind, 1)) {
1136                 usage(resume ? cmd_scrub_resume_usage :
1137                                         cmd_scrub_start_usage);
1138         }
1139
1140         spc.progress = NULL;
1141         if (do_quiet && do_print)
1142                 do_print = 0;
1143
1144         if (mkdir_p(datafile)) {
1145                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1146                                "file, mkdir %s failed: %s. Status recording "
1147                                "disabled\n", datafile, strerror(errno));
1148                 do_record = 0;
1149         }
1150         free(datafile);
1151
1152         path = argv[optind];
1153
1154         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1155
1156         if (fdmnt < 0) {
1157                 ERR(!do_quiet, "ERROR: can't access '%s'\n", path);
1158                 return 1;
1159         }
1160
1161         ret = get_fs_info(path, &fi_args, &di_args);
1162         if (ret) {
1163                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1164                     "%s\n", strerror(-ret));
1165                 err = 1;
1166                 goto out;
1167         }
1168         if (!fi_args.num_devices) {
1169                 ERR(!do_quiet, "ERROR: no devices found\n");
1170                 err = 1;
1171                 goto out;
1172         }
1173
1174         uuid_unparse(fi_args.fsid, fsid);
1175         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1176         if (fdres < 0 && fdres != -ENOENT) {
1177                 ERR(!do_quiet, "WARNING: failed to open status file: "
1178                     "%s\n", strerror(-fdres));
1179         } else if (fdres >= 0) {
1180                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1181                 if (IS_ERR(past_scrubs))
1182                         ERR(!do_quiet, "WARNING: failed to read status file: "
1183                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1184                 close(fdres);
1185         }
1186
1187         /*
1188          * check whether any involved device is already busy running a
1189          * scrub. This would cause damaged status messages and the state
1190          * "aborted" without the explanation that a scrub was already
1191          * running. Therefore check it first, prevent it and give some
1192          * feedback to the user if scrub is already running.
1193          * Note that if scrub is started with a block device as the
1194          * parameter, only that particular block device is checked. It
1195          * is a normal mode of operation to start scrub on multiple
1196          * single devices, there is no reason to prevent this.
1197          */
1198         if (is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1199                 ERR(!do_quiet,
1200                     "ERROR: scrub is already running.\n"
1201                     "To cancel use 'btrfs scrub cancel %s'.\n"
1202                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1203                     path, path);
1204                 err = 1;
1205                 goto out;
1206         }
1207
1208         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1209         sp = calloc(fi_args.num_devices, sizeof(*sp));
1210         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1211
1212         if (!t_devs || !sp || !spc.progress) {
1213                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1214                 err = 1;
1215                 goto out;
1216         }
1217
1218         ret = pthread_attr_init(&t_attr);
1219         if (ret) {
1220                 ERR(!do_quiet, "ERROR: pthread_attr_init failed: %s\n",
1221                     strerror(ret));
1222                 err = 1;
1223                 goto out;
1224         }
1225
1226         for (i = 0; i < fi_args.num_devices; ++i) {
1227                 devid = di_args[i].devid;
1228                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1229                 if (ret) {
1230                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1231                             "%s\n", strerror(ret));
1232                         err = 1;
1233                         goto out;
1234                 }
1235                 last_scrub = last_dev_scrub(past_scrubs, devid);
1236                 sp[i].scrub_args.devid = devid;
1237                 sp[i].fd = fdmnt;
1238                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1239                                              !last_scrub->stats.finished)) {
1240                         ++n_resume;
1241                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1242                         sp[i].resumed = last_scrub;
1243                 } else if (resume) {
1244                         ++n_skip;
1245                         sp[i].skip = 1;
1246                         sp[i].resumed = last_scrub;
1247                         continue;
1248                 } else {
1249                         ++n_start;
1250                         sp[i].scrub_args.start = 0ll;
1251                         sp[i].resumed = NULL;
1252                 }
1253                 sp[i].skip = 0;
1254                 sp[i].scrub_args.end = (u64)-1ll;
1255                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1256                 sp[i].ioprio_class = ioprio_class;
1257                 sp[i].ioprio_classdata = ioprio_classdata;
1258         }
1259
1260         if (!n_start && !n_resume) {
1261                 if (!do_quiet)
1262                         printf("scrub: nothing to resume for %s, fsid %s\n",
1263                                path, fsid);
1264                 return 2;
1265         }
1266
1267         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1268         while (ret != -1) {
1269                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1270                                         sock_path, sizeof(sock_path));
1271                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1272                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1273                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1274                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1275                 if (ret != -1 || errno != EADDRINUSE)
1276                         break;
1277                 /*
1278                  * bind failed with EADDRINUSE. so let's see if anyone answers
1279                  * when we make a call to the socket ...
1280                  */
1281                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1282                 if (!ret || errno != ECONNREFUSED) {
1283                         /* ... yes, so scrub must be running. error out */
1284                         fprintf(stderr, "ERROR: scrub already running\n");
1285                         close(prg_fd);
1286                         prg_fd = -1;
1287                         goto out;
1288                 }
1289                 /*
1290                  * ... no, this means someone left us alone with an unused
1291                  * socket in the file system. remove it and try again.
1292                  */
1293                 ret = unlink(sock_path);
1294         }
1295         if (ret != -1)
1296                 ret = listen(prg_fd, 100);
1297         if (ret == -1) {
1298                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1299                     "socket at %s: %s. Progress cannot be queried\n",
1300                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1301                     strerror(errno));
1302                 if (prg_fd != -1) {
1303                         close(prg_fd);
1304                         prg_fd = -1;
1305                         if (sock_path[0])
1306                                 unlink(sock_path);
1307                 }
1308         }
1309
1310         if (do_record) {
1311                 /* write all-zero progress file for a start */
1312                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1313                                            fi_args.num_devices);
1314                 if (ret) {
1315                         ERR(!do_quiet, "WARNING: failed to write the progress "
1316                             "status file: %s. Status recording disabled\n",
1317                             strerror(-ret));
1318                         do_record = 0;
1319                 }
1320         }
1321
1322         if (do_background) {
1323                 pid = fork();
1324                 if (pid == -1) {
1325                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1326                                         "%s\n", strerror(errno));
1327                         err = 1;
1328                         goto out;
1329                 }
1330
1331                 if (pid) {
1332                         int stat;
1333                         scrub_handle_sigint_parent();
1334                         if (!do_quiet)
1335                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1336                                        n_start ? "started" : "resumed",
1337                                        path, fsid, pid);
1338                         if (!do_wait) {
1339                                 err = 0;
1340                                 goto out;
1341                         }
1342                         ret = wait(&stat);
1343                         if (ret != pid) {
1344                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1345                                     "%s\n", ret, strerror(errno));
1346                                 err = 1;
1347                                 goto out;
1348                         }
1349                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1350                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1351                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1352                                 goto out;
1353                         }
1354                         err = 0;
1355                         goto out;
1356                 }
1357         }
1358
1359         scrub_handle_sigint_child(fdmnt);
1360
1361         for (i = 0; i < fi_args.num_devices; ++i) {
1362                 if (sp[i].skip) {
1363                         sp[i].scrub_args.progress = sp[i].resumed->p;
1364                         sp[i].stats = sp[i].resumed->stats;
1365                         sp[i].ret = 0;
1366                         sp[i].stats.finished = 1;
1367                         continue;
1368                 }
1369                 devid = di_args[i].devid;
1370                 gettimeofday(&tv, NULL);
1371                 sp[i].stats.t_start = tv.tv_sec;
1372                 ret = pthread_create(&t_devs[i], &t_attr,
1373                                         scrub_one_dev, &sp[i]);
1374                 if (ret) {
1375                         if (do_print)
1376                                 fprintf(stderr, "ERROR: creating "
1377                                         "scrub_one_dev[%llu] thread failed: "
1378                                         "%s\n", devid, strerror(ret));
1379                         err = 1;
1380                         goto out;
1381                 }
1382         }
1383
1384         spc.fdmnt = fdmnt;
1385         spc.prg_fd = prg_fd;
1386         spc.do_record = do_record;
1387         spc.write_mutex = &spc_write_mutex;
1388         spc.shared_progress = sp;
1389         spc.fi = &fi_args;
1390         ret = pthread_create(&t_prog, &t_attr, scrub_progress_cycle, &spc);
1391         if (ret) {
1392                 if (do_print)
1393                         fprintf(stderr, "ERROR: creating progress thread "
1394                                 "failed: %s\n", strerror(ret));
1395                 err = 1;
1396                 goto out;
1397         }
1398
1399         err = 0;
1400         for (i = 0; i < fi_args.num_devices; ++i) {
1401                 if (sp[i].skip)
1402                         continue;
1403                 devid = di_args[i].devid;
1404                 ret = pthread_join(t_devs[i], NULL);
1405                 if (ret) {
1406                         if (do_print)
1407                                 fprintf(stderr, "ERROR: pthread_join failed "
1408                                         "for scrub_one_dev[%llu]: %s\n", devid,
1409                                         strerror(ret));
1410                         ++err;
1411                         continue;
1412                 }
1413                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1414                         if (do_print)
1415                                 fprintf(stderr, "WARNING: device %lld not "
1416                                         "present\n", devid);
1417                         continue;
1418                 }
1419                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1420                         ++err;
1421                 } else if (sp[i].ret) {
1422                         if (do_print)
1423                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1424                                         "for device id %lld (%s)\n", path,
1425                                         devid, strerror(sp[i].ioctl_errno));
1426                         ++err;
1427                         continue;
1428                 }
1429                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1430                         e_uncorrectable++;
1431                 if (sp[i].scrub_args.progress.corrected_errors > 0
1432                     || sp[i].scrub_args.progress.unverified_errors > 0)
1433                         e_correctable++;
1434         }
1435
1436         if (do_print) {
1437                 const char *append = "done";
1438                 if (!do_stats_per_dev)
1439                         init_fs_stat(&fs_stat);
1440                 for (i = 0; i < fi_args.num_devices; ++i) {
1441                         if (do_stats_per_dev) {
1442                                 print_scrub_dev(&di_args[i],
1443                                                 &sp[i].scrub_args.progress,
1444                                                 print_raw,
1445                                                 sp[i].ret ? "canceled" : "done",
1446                                                 &sp[i].stats);
1447                         } else {
1448                                 if (sp[i].ret)
1449                                         append = "canceled";
1450                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1451                                                 &sp[i].stats, &fs_stat);
1452                         }
1453                 }
1454                 if (!do_stats_per_dev) {
1455                         printf("scrub %s for %s\n", append, fsid);
1456                         print_fs_stat(&fs_stat, print_raw);
1457                 }
1458         }
1459
1460         ret = pthread_cancel(t_prog);
1461         if (!ret)
1462                 ret = pthread_join(t_prog, &terr);
1463
1464         /* check for errors from the handling of the progress thread */
1465         if (do_print && ret) {
1466                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1467                         strerror(ret));
1468         }
1469
1470         /* check for errors returned from the progress thread itself */
1471         if (do_print && terr && terr != PTHREAD_CANCELED) {
1472                 fprintf(stderr, "ERROR: recording progress "
1473                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1474         }
1475
1476         if (do_record) {
1477                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1478                                            fi_args.num_devices);
1479                 if (ret && do_print) {
1480                         fprintf(stderr, "ERROR: failed to record the result: "
1481                                 "%s\n", strerror(-ret));
1482                 }
1483         }
1484
1485         scrub_handle_sigint_child(-1);
1486
1487 out:
1488         free_history(past_scrubs);
1489         free(di_args);
1490         free(t_devs);
1491         free(sp);
1492         free(spc.progress);
1493         if (prg_fd > -1) {
1494                 close(prg_fd);
1495                 if (sock_path[0])
1496                         unlink(sock_path);
1497         }
1498         close_file_or_dir(fdmnt, dirstream);
1499
1500         if (err)
1501                 return 1;
1502         if (e_correctable)
1503                 return 3;
1504         if (e_uncorrectable)
1505                 return 4;
1506         return 0;
1507 }
1508
1509 static const char * const cmd_scrub_start_usage[] = {
1510         "btrfs scrub start [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1511         "Start a new scrub",
1512         "",
1513         "-B     do not background",
1514         "-d     stats per device (-B only)",
1515         "-q     be quiet",
1516         "-r     read only mode",
1517         "-R     raw print mode, print full data instead of summary"
1518         "-c     set ioprio class (see ionice(1) manpage)",
1519         "-n     set ioprio classdata (see ionice(1) manpage)",
1520         NULL
1521 };
1522
1523 static int cmd_scrub_start(int argc, char **argv)
1524 {
1525         return scrub_start(argc, argv, 0);
1526 }
1527
1528 static const char * const cmd_scrub_cancel_usage[] = {
1529         "btrfs scrub cancel <path>|<device>",
1530         "Cancel a running scrub",
1531         NULL
1532 };
1533
1534 static int cmd_scrub_cancel(int argc, char **argv)
1535 {
1536         char *path;
1537         int ret;
1538         int fdmnt = -1;
1539         DIR *dirstream = NULL;
1540
1541         if (check_argc_exact(argc, 2))
1542                 usage(cmd_scrub_cancel_usage);
1543
1544         path = argv[1];
1545
1546         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1547         if (fdmnt < 0) {
1548                 fprintf(stderr, "ERROR: could not open %s: %s\n",
1549                         path, strerror(errno));
1550                 ret = 1;
1551                 goto out;
1552         }
1553
1554         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1555
1556         if (ret < 0) {
1557                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1558                         errno == ENOTCONN ? "not running" : strerror(errno));
1559                 if (errno == ENOTCONN)
1560                         ret = 2;
1561                 else
1562                         ret = 1;
1563                 goto out;
1564         }
1565
1566         ret = 0;
1567         printf("scrub cancelled\n");
1568
1569 out:
1570         close_file_or_dir(fdmnt, dirstream);
1571         return ret;
1572 }
1573
1574 static const char * const cmd_scrub_resume_usage[] = {
1575         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1576         "Resume previously canceled or interrupted scrub",
1577         "",
1578         "-B     do not background",
1579         "-d     stats per device (-B only)",
1580         "-q     be quiet",
1581         "-r     read only mode",
1582         "-c     set ioprio class (see ionice(1) manpage)",
1583         "-n     set ioprio classdata (see ionice(1) manpage)",
1584         NULL
1585 };
1586
1587 static int cmd_scrub_resume(int argc, char **argv)
1588 {
1589         return scrub_start(argc, argv, 1);
1590 }
1591
1592 static const char * const cmd_scrub_status_usage[] = {
1593         "btrfs scrub status [-dR] <path>|<device>",
1594         "Show status of running or finished scrub",
1595         "",
1596         "-d     stats per device",
1597         "-R     print raw stats",
1598         NULL
1599 };
1600
1601 static int cmd_scrub_status(int argc, char **argv)
1602 {
1603         char *path;
1604         struct btrfs_ioctl_fs_info_args fi_args;
1605         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1606         struct scrub_file_record **past_scrubs = NULL;
1607         struct scrub_file_record *last_scrub;
1608         struct scrub_fs_stat fs_stat;
1609         struct sockaddr_un addr = {
1610                 .sun_family = AF_UNIX,
1611         };
1612         int ret;
1613         int i;
1614         int fdmnt;
1615         int print_raw = 0;
1616         int do_stats_per_dev = 0;
1617         int c;
1618         char fsid[37];
1619         int fdres = -1;
1620         int err = 0;
1621         DIR *dirstream = NULL;
1622
1623         optind = 1;
1624         while ((c = getopt(argc, argv, "dR")) != -1) {
1625                 switch (c) {
1626                 case 'd':
1627                         do_stats_per_dev = 1;
1628                         break;
1629                 case 'R':
1630                         print_raw = 1;
1631                         break;
1632                 case '?':
1633                 default:
1634                         usage(cmd_scrub_status_usage);
1635                 }
1636         }
1637
1638         if (check_argc_exact(argc - optind, 1))
1639                 usage(cmd_scrub_status_usage);
1640
1641         path = argv[optind];
1642
1643         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1644
1645         if (fdmnt < 0) {
1646                 fprintf(stderr, "ERROR: can't access to '%s'\n", path);
1647                 return 1;
1648         }
1649
1650         ret = get_fs_info(path, &fi_args, &di_args);
1651         if (ret) {
1652                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1653                                 "%s\n", strerror(-ret));
1654                 err = 1;
1655                 goto out;
1656         }
1657         if (!fi_args.num_devices) {
1658                 fprintf(stderr, "ERROR: no devices found\n");
1659                 err = 1;
1660                 goto out;
1661         }
1662
1663         uuid_unparse(fi_args.fsid, fsid);
1664
1665         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1666         if (fdres == -1) {
1667                 fprintf(stderr, "ERROR: failed to create socket to "
1668                         "receive progress information: %s\n",
1669                         strerror(errno));
1670                 err = 1;
1671                 goto out;
1672         }
1673         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1674                         NULL, addr.sun_path, sizeof(addr.sun_path));
1675         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1676         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1677         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1678         if (ret == -1) {
1679                 close(fdres);
1680                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1681                 if (fdres < 0 && fdres != -ENOENT) {
1682                         fprintf(stderr, "WARNING: failed to open status file: "
1683                                 "%s\n", strerror(-fdres));
1684                         err = 1;
1685                         goto out;
1686                 }
1687         }
1688
1689         if (fdres >= 0) {
1690                 past_scrubs = scrub_read_file(fdres, 1);
1691                 if (IS_ERR(past_scrubs))
1692                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1693                                 strerror(-PTR_ERR(past_scrubs)));
1694         }
1695
1696         printf("scrub status for %s\n", fsid);
1697
1698         if (do_stats_per_dev) {
1699                 for (i = 0; i < fi_args.num_devices; ++i) {
1700                         last_scrub = last_dev_scrub(past_scrubs,
1701                                                         di_args[i].devid);
1702                         if (!last_scrub) {
1703                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1704                                                 NULL, NULL);
1705                                 continue;
1706                         }
1707                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1708                                         last_scrub->stats.finished ?
1709                                                         "history" : "status",
1710                                         &last_scrub->stats);
1711                 }
1712         } else {
1713                 init_fs_stat(&fs_stat);
1714                 for (i = 0; i < fi_args.num_devices; ++i) {
1715                         last_scrub = last_dev_scrub(past_scrubs,
1716                                                         di_args[i].devid);
1717                         if (!last_scrub)
1718                                 continue;
1719                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1720                                         &fs_stat);
1721                 }
1722                 print_fs_stat(&fs_stat, print_raw);
1723         }
1724
1725 out:
1726         free_history(past_scrubs);
1727         free(di_args);
1728         if (fdres > -1)
1729                 close(fdres);
1730         close_file_or_dir(fdmnt, dirstream);
1731
1732         return !!err;
1733 }
1734
1735 const struct cmd_group scrub_cmd_group = {
1736         scrub_cmd_group_usage, NULL, {
1737                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1738                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1739                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1740                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1741                 NULL_CMD_STRUCT
1742         }
1743 };
1744
1745 int cmd_scrub(int argc, char **argv)
1746 {
1747         return handle_command_group(&scrub_cmd_group, argc, argv);
1748 }