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         int in_progress;
  63 };
  64
  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
  72 #endif
  73
  74 struct scrub_progress {
  75         struct btrfs_ioctl_scrub_args scrub_args;
  76         int fd;
  77         int ret;
  78         int skip;
  79         struct scrub_stats stats;
  80         struct scrub_file_record *resumed;
  81         int ioctl_errno;
  82         pthread_mutex_t progress_mutex;
  83         int ioprio_class;
  84         int ioprio_classdata;
  85 };
  86
  87 struct scrub_file_record {
  88         u8 fsid[BTRFS_FSID_SIZE];
  89         u64 devid;
  90         struct scrub_stats stats;
  91         struct btrfs_scrub_progress p;
  92 };
  93
  94 struct scrub_progress_cycle {
  95         int fdmnt;
  96         int prg_fd;
  97         int do_record;
  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;
 102 };
 103
 104 struct scrub_fs_stat {
 105         struct btrfs_scrub_progress p;
 106         struct scrub_stats s;
 107         int i;
 108 };
 109
 110 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 111 {
 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);
 127 }
 128
 129 #define ERR(test, ...) do {                     \
 130         if (test)                               \
 131                 fprintf(stderr, __VA_ARGS__);   \
 132 } while (0)
 133
 134 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 135         if (test)                               \
 136                 printf(" %s=%llu", desc, test); \
 137 } while (0)
 138
 139 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 140 {
 141         u64 err_cnt;
 142         u64 err_cnt2;
 143
 144         err_cnt = p->read_errors +
 145                         p->csum_errors +
 146                         p->verify_errors +
 147                         p->super_errors;
 148
 149         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 150
 151         if (p->malloc_errors)
 152                 printf("*** WARNING: memory allocation failed while scrubbing. "
 153                        "results may be inaccurate\n");
 154
 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));
 158
 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");
 165                 printf("\n");
 166                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 167                         "unverified errors: %llu\n", p->corrected_errors,
 168                         p->uncorrectable_errors, p->unverified_errors);
 169         }
 170 }
 171
 172 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 173         fs_stat->p.name += p->name;             \
 174 } while (0)
 175
 176 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 177 do {                                            \
 178         if (fs_stat->s.name > ss->name) {       \
 179                 fs_stat->s.name = ss->name;     \
 180         }                                       \
 181 } while (0)
 182
 183 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 184 do {                                                            \
 185         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 186                 fs_stat->s.name = ss->name;                     \
 187         }                                                       \
 188 } while (0)
 189
 190 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 191 do {                                                                    \
 192         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 193                 (fs_stat)->s.name = (ss)->name;                         \
 194         }                                                               \
 195 } while (0)
 196
 197 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 198                                 struct scrub_stats *ss,
 199                                 struct scrub_fs_stat *fs_stat)
 200 {
 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);
 220 }
 221
 222 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 223 {
 224         memset(fs_stat, 0, sizeof(*fs_stat));
 225         fs_stat->s.finished = 1;
 226 }
 227
 228 static void _print_scrub_ss(struct scrub_stats *ss)
 229 {
 230         char t[4096];
 231         struct tm tm;
 232         time_t seconds;
 233         unsigned hours;
 234
 235         if (!ss || !ss->t_start) {
 236                 printf("\tno stats available\n");
 237                 return;
 238         }
 239         if (ss->t_resumed) {
 240                 localtime_r(&ss->t_resumed, &tm);
 241                 strftime(t, sizeof(t), "%c", &tm);
 242                 t[sizeof(t) - 1] = '\0';
 243                 printf("\tscrub resumed at %s", t);
 244         } else {
 245                 localtime_r(&ss->t_start, &tm);
 246                 strftime(t, sizeof(t), "%c", &tm);
 247                 t[sizeof(t) - 1] = '\0';
 248                 printf("\tscrub started at %s", t);
 249         }
 250
 251         seconds = ss->duration;
 252         hours = ss->duration / (60 * 60);
 253         gmtime_r(&seconds, &tm);
 254         strftime(t, sizeof(t), "%M:%S", &tm);
 255         if (ss->finished && !ss->canceled) {
 256                 printf(" and finished after %02u:%s\n", hours, t);
 257         } else if (ss->canceled) {
 258                 printf(" and was aborted after %02u:%s\n", hours, t);
 259         } else {
 260                 if (ss->in_progress)
 261                         printf(", running for %02u:%s\n", hours, t);
 262                 else
 263                         printf(", interrupted after %02u:%s, not running\n",
 264                                         hours, t);
 265         }
 266 }
 267
 268 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 269                                 struct btrfs_scrub_progress *p, int raw,
 270                                 const char *append, struct scrub_stats *ss)
 271 {
 272         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 273                append ? append : "");
 274
 275         _print_scrub_ss(ss);
 276
 277         if (p) {
 278                 if (raw)
 279                         print_scrub_full(p);
 280                 else
 281                         print_scrub_summary(p);
 282         }
 283 }
 284
 285 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 286 {
 287         _print_scrub_ss(&fs_stat->s);
 288
 289         if (raw)
 290                 print_scrub_full(&fs_stat->p);
 291         else
 292                 print_scrub_summary(&fs_stat->p);
 293 }
 294
 295 static void free_history(struct scrub_file_record **last_scrubs)
 296 {
 297         struct scrub_file_record **l = last_scrubs;
 298         if (!l || IS_ERR(l))
 299                 return;
 300         while (*l)
 301                 free(*l++);
 302         free(last_scrubs);
 303 }
 304
 305 /*
 306  * cancels a running scrub and makes the master process record the current
 307  * progress status before exiting.
 308  */
 309 static int cancel_fd = -1;
 310 static void scrub_sigint_record_progress(int signal)
 311 {
 312         int ret;
 313
 314         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 315         if (ret < 0)
 316                 perror("Scrub cancel failed");
 317 }
 318
 319 static int scrub_handle_sigint_parent(void)
 320 {
 321         struct sigaction sa = {
 322                 .sa_handler = SIG_IGN,
 323                 .sa_flags = SA_RESTART,
 324         };
 325
 326         return sigaction(SIGINT, &sa, NULL);
 327 }
 328
 329 static int scrub_handle_sigint_child(int fd)
 330 {
 331         struct sigaction sa = {
 332                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 333         };
 334
 335         cancel_fd = fd;
 336         return sigaction(SIGINT, &sa, NULL);
 337 }
 338
 339 static int scrub_datafile(const char *fn_base, const char *fn_local,
 340                                 const char *fn_tmp, char *datafile, int size)
 341 {
 342         int ret;
 343         int end = size - 2;
 344
 345         datafile[end + 1] = '\0';
 346         strncpy(datafile, fn_base, end);
 347         ret = strlen(datafile);
 348
 349         if (ret + 1 > end)
 350                 return -EOVERFLOW;
 351
 352         datafile[ret] = '.';
 353         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 354         ret = strlen(datafile);
 355
 356         if (ret + 1 > end)
 357                 return -EOVERFLOW;
 358
 359         if (fn_tmp) {
 360                 datafile[ret] = '_';
 361                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 362                 ret = strlen(datafile);
 363
 364                 if (ret > end)
 365                         return -EOVERFLOW;
 366         }
 367
 368         return 0;
 369 }
 370
 371 static int scrub_open_file(const char *datafile, int m)
 372 {
 373         int fd;
 374         int ret;
 375
 376         fd = open(datafile, m, 0600);
 377         if (fd < 0)
 378                 return -errno;
 379
 380         ret = flock(fd, LOCK_EX|LOCK_NB);
 381         if (ret) {
 382                 ret = errno;
 383                 close(fd);
 384                 return -ret;
 385         }
 386
 387         return fd;
 388 }
 389
 390 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 391 {
 392         int ret;
 393         char datafile[PATH_MAX];
 394         ret = scrub_datafile(fn_base, fn_local, NULL,
 395                                 datafile, sizeof(datafile));
 396         if (ret < 0)
 397                 return ret;
 398         return scrub_open_file(datafile, O_RDONLY);
 399 }
 400
 401 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 402                                 const char *tmp)
 403 {
 404         int ret;
 405         char datafile[PATH_MAX];
 406         ret = scrub_datafile(fn_base, fn_local, tmp,
 407                                 datafile, sizeof(datafile));
 408         if (ret < 0)
 409                 return ret;
 410         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 411 }
 412
 413 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 414                                 const char *tmp)
 415 {
 416         int ret;
 417         char datafile_old[PATH_MAX];
 418         char datafile_new[PATH_MAX];
 419         ret = scrub_datafile(fn_base, fn_local, tmp,
 420                                 datafile_old, sizeof(datafile_old));
 421         if (ret < 0)
 422                 return ret;
 423         ret = scrub_datafile(fn_base, fn_local, NULL,
 424                                 datafile_new, sizeof(datafile_new));
 425         if (ret < 0)
 426                 return ret;
 427         ret = rename(datafile_old, datafile_new);
 428         return ret ? -errno : 0;
 429 }
 430
 431 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 432         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 433 }
 434
 435 /*
 436  * returns 0 if the key did not match (nothing was read)
 437  *         1 if the key did match (success)
 438  *        -1 if the key did match and an error occured
 439  */
 440 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 441                         const char *key, u64 *dest)
 442 {
 443         int j;
 444
 445         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 446                 *i += len - 1;
 447                 if (buf[*i] != ':')
 448                         return -1;
 449                 *i += 1;
 450                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 451                         ;
 452                 if (*i + j >= avail)
 453                         return -1;
 454                 *dest = atoll(&buf[*i]);
 455                 *i += j;
 456                 return 1;
 457         }
 458
 459         return 0;
 460 }
 461
 462 #define _SCRUB_INVALID do {                                             \
 463         if (report_errors)                                              \
 464                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 465                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 466                         lineno, i, state, 20 > avail ? avail : 20,      \
 467                         l + i,  __FILE__, __LINE__);                    \
 468         goto skip;                                                      \
 469 } while (0)
 470
 471 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 472 {
 473         int avail = 0;
 474         int old_avail = 0;
 475         char l[16 * 1024];
 476         int state = 0;
 477         int curr = -1;
 478         int i = 0;
 479         int j;
 480         int ret;
 481         int eof = 0;
 482         int lineno = 0;
 483         u64 version;
 484         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 485         struct scrub_file_record **p = NULL;
 486
 487 again:
 488         old_avail = avail - i;
 489         BUG_ON(old_avail < 0);
 490         if (old_avail)
 491                 memmove(l, l + i, old_avail);
 492         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 493         if (avail == 0)
 494                 eof = 1;
 495         if (avail == 0 && old_avail == 0) {
 496                 if (curr >= 0 &&
 497                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 498                         p[curr] = NULL;
 499                 } else if (curr == -1) {
 500                         p = ERR_PTR(-ENODATA);
 501                 }
 502                 return p;
 503         }
 504         if (avail == -1)
 505                 return ERR_PTR(-errno);
 506         avail += old_avail;
 507
 508         i = 0;
 509         while (i < avail) {
 510                 switch (state) {
 511                 case 0: /* start of file */
 512                         ret = scrub_kvread(&i,
 513                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 514                                 SCRUB_FILE_VERSION_PREFIX, &version);
 515                         if (ret != 1)
 516                                 _SCRUB_INVALID;
 517                         if (version != atoll(SCRUB_FILE_VERSION))
 518                                 return ERR_PTR(-ENOTSUP);
 519                         state = 6;
 520                         continue;
 521                 case 1: /* start of line, alloc */
 522                         /*
 523                          * this state makes sure we have a complete line in
 524                          * further processing, so we don't need wrap-tracking
 525                          * everywhere.
 526                          */
 527                         if (!eof && !memchr(l + i, '\n', avail - i))
 528                                 goto again;
 529                         ++lineno;
 530                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 531                                                 BTRFS_FSID_SIZE) == 0) {
 532                                 state = 2;
 533                                 continue;
 534                         }
 535                         ++curr;
 536                         p = realloc(p, (curr + 2) * sizeof(*p));
 537                         if (p)
 538                                 p[curr] = malloc(sizeof(**p));
 539                         if (!p || !p[curr])
 540                                 return ERR_PTR(-errno);
 541                         memset(p[curr], 0, sizeof(**p));
 542                         p[curr + 1] = NULL;
 543                         ++state;
 544                         /* fall through */
 545                 case 2: /* start of line, skip space */
 546                         while (isspace(l[i]) && i < avail) {
 547                                 if (l[i] == '\n')
 548                                         ++lineno;
 549                                 ++i;
 550                         }
 551                         if (i >= avail ||
 552                             (!eof && !memchr(l + i, '\n', avail - i)))
 553                                 goto again;
 554                         ++state;
 555                         /* fall through */
 556                 case 3: /* read fsid */
 557                         if (i == avail)
 558                                 continue;
 559                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 560                                 ;
 561                         if (i + j + 1 >= avail)
 562                                 _SCRUB_INVALID;
 563                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 564                                 _SCRUB_INVALID;
 565                         l[i + j] = '\0';
 566                         ret = uuid_parse(l + i, p[curr]->fsid);
 567                         if (ret)
 568                                 _SCRUB_INVALID;
 569                         i += j + 1;
 570                         ++state;
 571                         /* fall through */
 572                 case 4: /* read dev id */
 573                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 574                                 ;
 575                         if (j == 0 || i + j + 1 >= avail)
 576                                 _SCRUB_INVALID;
 577                         p[curr]->devid = atoll(&l[i]);
 578                         i += j + 1;
 579                         ++state;
 580                         /* fall through */
 581                 case 5: /* read key/value pair */
 582                         ret = 0;
 583                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 584                                         &p[curr]->p);
 585                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 586                                         &p[curr]->p);
 587                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 588                                         &p[curr]->p);
 589                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 590                                         &p[curr]->p);
 591                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 592                                         &p[curr]->p);
 593                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 594                                         &p[curr]->p);
 595                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 596                                         &p[curr]->p);
 597                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 598                                         &p[curr]->p);
 599                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 600                                         &p[curr]->p);
 601                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 602                                         &p[curr]->p);
 603                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 604                                         &p[curr]->p);
 605                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 606                                         &p[curr]->p);
 607                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 608                                         &p[curr]->p);
 609                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 610                                         &p[curr]->p);
 611                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 612                                         &p[curr]->p);
 613                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 614                                         &p[curr]->stats);
 615                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 616                                         (u64 *)&p[curr]->stats);
 617                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 618                                         (u64 *)&p[curr]->stats);
 619                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 620                                         (u64 *)&p[curr]->stats);
 621                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 622                                         &p[curr]->stats);
 623                         if (ret != 1)
 624                                 _SCRUB_INVALID;
 625                         ++state;
 626                         /* fall through */
 627                 case 6: /* after number */
 628                         if (l[i] == '|')
 629                                 state = 5;
 630                         else if (l[i] == '\n')
 631                                 state = 1;
 632                         else
 633                                 _SCRUB_INVALID;
 634                         ++i;
 635                         continue;
 636                 case 99: /* skip rest of line */
 637 skip:
 638                         state = 99;
 639                         do {
 640                                 ++i;
 641                                 if (l[i - 1] == '\n') {
 642                                         state = 1;
 643                                         break;
 644                                 }
 645                         } while (i < avail);
 646                         continue;
 647                 }
 648                 BUG();
 649         }
 650         goto again;
 651 }
 652
 653 static int scrub_write_buf(int fd, const void *data, int len)
 654 {
 655         int ret;
 656         ret = write(fd, data, len);
 657         return ret - len;
 658 }
 659
 660 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 661                                 __attribute__ ((format (printf, 4, 5)));
 662 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 663 {
 664         int ret;
 665         va_list args;
 666
 667         va_start(args, fmt);
 668         ret = vsnprintf(buf, max, fmt, args);
 669         va_end(args);
 670         if (ret >= max)
 671                 return ret - max;
 672         return scrub_write_buf(fd, buf, ret);
 673 }
 674
 675 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 676                         data->resumed->p.name + data->scrub_args.progress.name
 677
 678 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 679                                                   struct scrub_progress *dest)
 680 {
 681         if (!data->resumed || data->skip)
 682                 return data;
 683
 684         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 685         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 686         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 687         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 688         _SCRUB_SUM(dest, data, read_errors);
 689         _SCRUB_SUM(dest, data, csum_errors);
 690         _SCRUB_SUM(dest, data, verify_errors);
 691         _SCRUB_SUM(dest, data, no_csum);
 692         _SCRUB_SUM(dest, data, csum_discards);
 693         _SCRUB_SUM(dest, data, super_errors);
 694         _SCRUB_SUM(dest, data, malloc_errors);
 695         _SCRUB_SUM(dest, data, uncorrectable_errors);
 696         _SCRUB_SUM(dest, data, corrected_errors);
 697         _SCRUB_SUM(dest, data, last_physical);
 698         dest->stats.canceled = data->stats.canceled;
 699         dest->stats.finished = data->stats.finished;
 700         dest->stats.t_resumed = data->stats.t_start;
 701         dest->stats.t_start = data->resumed->stats.t_start;
 702         dest->stats.duration = data->resumed->stats.duration +
 703                                                         data->stats.duration;
 704         dest->scrub_args.devid = data->scrub_args.devid;
 705         return dest;
 706 }
 707
 708 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 709         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 710                         use->scrub_args.progress.name)
 711
 712 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 713         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 714                         use->stats.name)
 715
 716 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 717 {
 718         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 719 }
 720
 721 static int scrub_write_file(int fd, const char *fsid,
 722                                 struct scrub_progress *data, int n)
 723 {
 724         int ret = 0;
 725         int i;
 726         char buf[1024];
 727         struct scrub_progress local;
 728         struct scrub_progress *use;
 729
 730         if (n < 1)
 731                 return -EINVAL;
 732
 733         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 734         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 735                                 SCRUB_FILE_VERSION "\n",
 736                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 737                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 738         if (ret)
 739                 return -EOVERFLOW;
 740
 741         for (i = 0; i < n; ++i) {
 742                 use = scrub_resumed_stats(&data[i], &local);
 743                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 744                     scrub_write_buf(fd, ":", 1) ||
 745                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 746                                         use->scrub_args.devid) ||
 747                     scrub_write_buf(fd, buf, ret) ||
 748                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 749                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 750                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 751                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 752                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 753                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 754                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 755                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 756                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 757                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 758                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 759                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 760                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 761                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 762                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 763                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 764                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 765                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 766                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 767                     scrub_write_buf(fd, "\n", 1)) {
 768                         return -EOVERFLOW;
 769                 }
 770         }
 771
 772         return 0;
 773 }
 774
 775 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 776                                 struct scrub_progress *data, int n)
 777 {
 778         int ret;
 779         int err;
 780         int fd = -1;
 781         int old;
 782
 783         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 784         if (ret) {
 785                 err = -ret;
 786                 goto out3;
 787         }
 788
 789         ret = pthread_mutex_lock(m);
 790         if (ret) {
 791                 err = -ret;
 792                 goto out2;
 793         }
 794
 795         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 796         if (fd < 0) {
 797                 err = fd;
 798                 goto out1;
 799         }
 800         err = scrub_write_file(fd, fsid, data, n);
 801         if (err)
 802                 goto out1;
 803         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 804         if (err)
 805                 goto out1;
 806
 807 out1:
 808         if (fd >= 0) {
 809                 ret = close(fd);
 810                 if (ret)
 811                         err = -errno;
 812         }
 813
 814         ret = pthread_mutex_unlock(m);
 815         if (ret && !err)
 816                 err = -ret;
 817
 818 out2:
 819         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 820         if (ret && !err)
 821                 err = -ret;
 822
 823 out3:
 824         return err;
 825 }
 826
 827 static void *scrub_one_dev(void *ctx)
 828 {
 829         struct scrub_progress *sp = ctx;
 830         int ret;
 831         struct timeval tv;
 832
 833         sp->stats.canceled = 0;
 834         sp->stats.duration = 0;
 835         sp->stats.finished = 0;
 836
 837         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 838                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 839                                         sp->ioprio_classdata));
 840         if (ret)
 841                 fprintf(stderr,
 842                         "WARNING: setting ioprio failed: %s (ignored).\n",
 843                         strerror(errno));
 844
 845         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 846         gettimeofday(&tv, NULL);
 847         sp->ret = ret;
 848         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 849         sp->stats.canceled = !!ret;
 850         sp->ioctl_errno = errno;
 851         ret = pthread_mutex_lock(&sp->progress_mutex);
 852         if (ret)
 853                 return ERR_PTR(-ret);
 854         sp->stats.finished = 1;
 855         ret = pthread_mutex_unlock(&sp->progress_mutex);
 856         if (ret)
 857                 return ERR_PTR(-ret);
 858
 859         return NULL;
 860 }
 861
 862 static void *progress_one_dev(void *ctx)
 863 {
 864         struct scrub_progress *sp = ctx;
 865
 866         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 867         sp->ioctl_errno = errno;
 868
 869         return NULL;
 870 }
 871
 872 /* nb: returns a negative errno via ERR_PTR */
 873 static void *scrub_progress_cycle(void *ctx)
 874 {
 875         int ret = 0;
 876         int  perr = 0;  /* positive / pthread error returns */
 877         int old;
 878         int i;
 879         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 880         struct scrub_progress *sp;
 881         struct scrub_progress *sp_last;
 882         struct scrub_progress *sp_shared;
 883         struct timeval tv;
 884         struct scrub_progress_cycle *spc = ctx;
 885         int ndev = spc->fi->num_devices;
 886         int this = 1;
 887         int last = 0;
 888         int peer_fd = -1;
 889         struct pollfd accept_poll_fd = {
 890                 .fd = spc->prg_fd,
 891                 .events = POLLIN,
 892                 .revents = 0,
 893         };
 894         struct pollfd write_poll_fd = {
 895                 .events = POLLOUT,
 896                 .revents = 0,
 897         };
 898         struct sockaddr_un peer;
 899         socklen_t peer_size = sizeof(peer);
 900
 901         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 902         if (perr)
 903                 goto out;
 904
 905         uuid_unparse(spc->fi->fsid, fsid);
 906
 907         for (i = 0; i < ndev; ++i) {
 908                 sp = &spc->progress[i];
 909                 sp_last = &spc->progress[i + ndev];
 910                 sp_shared = &spc->shared_progress[i];
 911                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 912                                                 sp_shared->scrub_args.devid;
 913                 sp->fd = sp_last->fd = spc->fdmnt;
 914                 sp->stats.t_start = sp_last->stats.t_start =
 915                                                 sp_shared->stats.t_start;
 916                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 917                 sp->skip = sp_last->skip = sp_shared->skip;
 918                 sp->stats.finished = sp_last->stats.finished =
 919                                                 sp_shared->stats.finished;
 920         }
 921
 922         while (1) {
 923                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 924                 if (ret == -1) {
 925                         ret = -errno;
 926                         goto out;
 927                 }
 928                 if (ret)
 929                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 930                                          &peer_size);
 931                 gettimeofday(&tv, NULL);
 932                 this = (this + 1)%2;
 933                 last = (last + 1)%2;
 934                 for (i = 0; i < ndev; ++i) {
 935                         sp = &spc->progress[this * ndev + i];
 936                         sp_last = &spc->progress[last * ndev + i];
 937                         sp_shared = &spc->shared_progress[i];
 938                         if (sp->stats.finished)
 939                                 continue;
 940                         progress_one_dev(sp);
 941                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 942                         if (!sp->ret)
 943                                 continue;
 944                         if (sp->ioctl_errno != ENOTCONN &&
 945                             sp->ioctl_errno != ENODEV) {
 946                                 ret = -sp->ioctl_errno;
 947                                 goto out;
 948                         }
 949                         /*
 950                          * scrub finished or device removed, check the
 951                          * finished flag. if unset, just use the last
 952                          * result we got for the current write and go
 953                          * on. flag should be set on next cycle, then.
 954                          */
 955                         perr = pthread_setcancelstate(
 956                                         PTHREAD_CANCEL_DISABLE, &old);
 957                         if (perr)
 958                                 goto out;
 959                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 960                         if (perr)
 961                                 goto out;
 962                         if (!sp_shared->stats.finished) {
 963                                 perr = pthread_mutex_unlock(
 964                                                 &sp_shared->progress_mutex);
 965                                 if (perr)
 966                                         goto out;
 967                                 perr = pthread_setcancelstate(
 968                                                 PTHREAD_CANCEL_ENABLE, &old);
 969                                 if (perr)
 970                                         goto out;
 971                                 memcpy(sp, sp_last, sizeof(*sp));
 972                                 continue;
 973                         }
 974                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 975                         if (perr)
 976                                 goto out;
 977                         perr = pthread_setcancelstate(
 978                                         PTHREAD_CANCEL_ENABLE, &old);
 979                         if (perr)
 980                                 goto out;
 981                         memcpy(sp, sp_shared, sizeof(*sp));
 982                         memcpy(sp_last, sp_shared, sizeof(*sp));
 983                 }
 984                 if (peer_fd != -1) {
 985                         write_poll_fd.fd = peer_fd;
 986                         ret = poll(&write_poll_fd, 1, 0);
 987                         if (ret == -1) {
 988                                 ret = -errno;
 989                                 goto out;
 990                         }
 991                         if (ret) {
 992                                 ret = scrub_write_file(
 993                                         peer_fd, fsid,
 994                                         &spc->progress[this * ndev], ndev);
 995                                 if (ret)
 996                                         goto out;
 997                         }
 998                         close(peer_fd);
 999                         peer_fd = -1;
1000                 }
1001                 if (!spc->do_record)
1002                         continue;
1003                 ret = scrub_write_progress(spc->write_mutex, fsid,
1004                                            &spc->progress[this * ndev], ndev);
1005                 if (ret)
1006                         goto out;
1007         }
1008 out:
1009         if (peer_fd != -1)
1010                 close(peer_fd);
1011         if (perr)
1012                 ret = -perr;
1013         return ERR_PTR(ret);
1014 }
1015
1016 static struct scrub_file_record *last_dev_scrub(
1017                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1018 {
1019         int i;
1020
1021         if (!past_scrubs || IS_ERR(past_scrubs))
1022                 return NULL;
1023
1024         for (i = 0; past_scrubs[i]; ++i)
1025                 if (past_scrubs[i]->devid == devid)
1026                         return past_scrubs[i];
1027
1028         return NULL;
1029 }
1030
1031 static int mkdir_p(char *path)
1032 {
1033         int i;
1034         int ret;
1035
1036         for (i = 1; i < strlen(path); ++i) {
1037                 if (path[i] != '/')
1038                         continue;
1039                 path[i] = '\0';
1040                 ret = mkdir(path, 0777);
1041                 if (ret && errno != EEXIST)
1042                         return -errno;
1043                 path[i] = '/';
1044         }
1045
1046         return 0;
1047 }
1048
1049 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1050                                   struct btrfs_ioctl_dev_info_args *di_args,
1051                                   struct scrub_file_record **past_scrubs)
1052 {
1053         int i;
1054
1055         if (!fi_args || !di_args || !past_scrubs)
1056                 return 0;
1057
1058         for (i = 0; i < fi_args->num_devices; i++) {
1059                 struct scrub_file_record *sfr =
1060                         last_dev_scrub(past_scrubs, di_args[i].devid);
1061
1062                 if (!sfr)
1063                         continue;
1064                 if (!(sfr->stats.finished || sfr->stats.canceled))
1065                         return 1;
1066         }
1067         return 0;
1068 }
1069
1070 static int is_scrub_running_in_kernel(int fd,
1071                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1072 {
1073         struct scrub_progress sp;
1074         int i;
1075         int ret;
1076
1077         for (i = 0; i < max_devices; i++) {
1078                 memset(&sp, 0, sizeof(sp));
1079                 sp.scrub_args.devid = di_args[i].devid;
1080                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1081                 if (!ret)
1082                         return 1;
1083         }
1084
1085         return 0;
1086 }
1087
1088 static const char * const cmd_scrub_start_usage[];
1089 static const char * const cmd_scrub_resume_usage[];
1090
1091 static int scrub_start(int argc, char **argv, int resume)
1092 {
1093         int fdmnt;
1094         int prg_fd = -1;
1095         int fdres = -1;
1096         int ret;
1097         pid_t pid;
1098         int c;
1099         int i;
1100         int err = 0;
1101         int e_uncorrectable = 0;
1102         int e_correctable = 0;
1103         int print_raw = 0;
1104         char *path;
1105         int do_background = 1;
1106         int do_wait = 0;
1107         int do_print = 0;
1108         int do_quiet = 0;
1109         int do_record = 1;
1110         int readonly = 0;
1111         int do_stats_per_dev = 0;
1112         int ioprio_class = IOPRIO_CLASS_IDLE;
1113         int ioprio_classdata = 0;
1114         int n_start = 0;
1115         int n_skip = 0;
1116         int n_resume = 0;
1117         struct btrfs_ioctl_fs_info_args fi_args;
1118         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1119         struct scrub_progress *sp = NULL;
1120         struct scrub_fs_stat fs_stat;
1121         struct timeval tv;
1122         struct sockaddr_un addr = {
1123                 .sun_family = AF_UNIX,
1124         };
1125         pthread_t *t_devs = NULL;
1126         pthread_t t_prog;
1127         struct scrub_file_record **past_scrubs = NULL;
1128         struct scrub_file_record *last_scrub = NULL;
1129         char *datafile = strdup(SCRUB_DATA_FILE);
1130         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1131         char sock_path[PATH_MAX] = "";
1132         struct scrub_progress_cycle spc;
1133         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1134         void *terr;
1135         u64 devid;
1136         DIR *dirstream = NULL;
1137         int force = 0;
1138         int nothing_to_resume = 0;
1139
1140         optind = 1;
1141         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1142                 switch (c) {
1143                 case 'B':
1144                         do_background = 0;
1145                         do_wait = 1;
1146                         do_print = 1;
1147                         break;
1148                 case 'd':
1149                         do_stats_per_dev = 1;
1150                         break;
1151                 case 'q':
1152                         do_quiet = 1;
1153                         break;
1154                 case 'r':
1155                         readonly = 1;
1156                         break;
1157                 case 'R':
1158                         print_raw = 1;
1159                         break;
1160                 case 'c':
1161                         ioprio_class = (int)strtol(optarg, NULL, 10);
1162                         break;
1163                 case 'n':
1164                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1165                         break;
1166                 case 'f':
1167                         force = 1;
1168                         break;
1169                 case '?':
1170                 default:
1171                         usage(resume ? cmd_scrub_resume_usage :
1172                                                 cmd_scrub_start_usage);
1173                 }
1174         }
1175
1176         /* try to catch most error cases before forking */
1177
1178         if (check_argc_exact(argc - optind, 1)) {
1179                 usage(resume ? cmd_scrub_resume_usage :
1180                                         cmd_scrub_start_usage);
1181         }
1182
1183         spc.progress = NULL;
1184         if (do_quiet && do_print)
1185                 do_print = 0;
1186
1187         if (mkdir_p(datafile)) {
1188                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1189                                "file, mkdir %s failed: %s. Status recording "
1190                                "disabled\n", datafile, strerror(errno));
1191                 do_record = 0;
1192         }
1193         free(datafile);
1194
1195         path = argv[optind];
1196
1197         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1198
1199         if (fdmnt < 0) {
1200                 if (errno == EINVAL)
1201                         ERR(!do_quiet,
1202                             "ERROR: '%s' is not a mounted btrfs device\n",
1203                             path);
1204                 else
1205                         ERR(!do_quiet, "ERROR: can't access '%s': %s\n",
1206                             path, strerror(errno));
1207                 return 1;
1208         }
1209
1210         ret = get_fs_info(path, &fi_args, &di_args);
1211         if (ret) {
1212                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1213                     "%s\n", strerror(-ret));
1214                 err = 1;
1215                 goto out;
1216         }
1217         if (!fi_args.num_devices) {
1218                 ERR(!do_quiet, "ERROR: no devices found\n");
1219                 err = 1;
1220                 goto out;
1221         }
1222
1223         uuid_unparse(fi_args.fsid, fsid);
1224         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1225         if (fdres < 0 && fdres != -ENOENT) {
1226                 ERR(!do_quiet, "WARNING: failed to open status file: "
1227                     "%s\n", strerror(-fdres));
1228         } else if (fdres >= 0) {
1229                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1230                 if (IS_ERR(past_scrubs))
1231                         ERR(!do_quiet, "WARNING: failed to read status file: "
1232                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1233                 close(fdres);
1234         }
1235
1236         /*
1237          * Check for stale information in the status file, ie. if it's
1238          * canceled=0, finished=0 but no scrub is running.
1239          */
1240         if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1241                 force = 1;
1242
1243         /*
1244          * check whether any involved device is already busy running a
1245          * scrub. This would cause damaged status messages and the state
1246          * "aborted" without the explanation that a scrub was already
1247          * running. Therefore check it first, prevent it and give some
1248          * feedback to the user if scrub is already running.
1249          * Note that if scrub is started with a block device as the
1250          * parameter, only that particular block device is checked. It
1251          * is a normal mode of operation to start scrub on multiple
1252          * single devices, there is no reason to prevent this.
1253          */
1254         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1255                 ERR(!do_quiet,
1256                     "ERROR: scrub is already running.\n"
1257                     "To cancel use 'btrfs scrub cancel %s'.\n"
1258                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1259                     path, path);
1260                 err = 1;
1261                 goto out;
1262         }
1263
1264         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1265         sp = calloc(fi_args.num_devices, sizeof(*sp));
1266         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1267
1268         if (!t_devs || !sp || !spc.progress) {
1269                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1270                 err = 1;
1271                 goto out;
1272         }
1273
1274         for (i = 0; i < fi_args.num_devices; ++i) {
1275                 devid = di_args[i].devid;
1276                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1277                 if (ret) {
1278                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1279                             "%s\n", strerror(ret));
1280                         err = 1;
1281                         goto out;
1282                 }
1283                 last_scrub = last_dev_scrub(past_scrubs, devid);
1284                 sp[i].scrub_args.devid = devid;
1285                 sp[i].fd = fdmnt;
1286                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1287                                              !last_scrub->stats.finished)) {
1288                         ++n_resume;
1289                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1290                         sp[i].resumed = last_scrub;
1291                 } else if (resume) {
1292                         ++n_skip;
1293                         sp[i].skip = 1;
1294                         sp[i].resumed = last_scrub;
1295                         continue;
1296                 } else {
1297                         ++n_start;
1298                         sp[i].scrub_args.start = 0ll;
1299                         sp[i].resumed = NULL;
1300                 }
1301                 sp[i].skip = 0;
1302                 sp[i].scrub_args.end = (u64)-1ll;
1303                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1304                 sp[i].ioprio_class = ioprio_class;
1305                 sp[i].ioprio_classdata = ioprio_classdata;
1306         }
1307
1308         if (!n_start && !n_resume) {
1309                 if (!do_quiet)
1310                         printf("scrub: nothing to resume for %s, fsid %s\n",
1311                                path, fsid);
1312                 nothing_to_resume = 1;
1313                 goto out;
1314         }
1315
1316         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1317         while (ret != -1) {
1318                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1319                                         sock_path, sizeof(sock_path));
1320                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1321                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1322                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1323                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1324                 if (ret != -1 || errno != EADDRINUSE)
1325                         break;
1326                 /*
1327                  * bind failed with EADDRINUSE. so let's see if anyone answers
1328                  * when we make a call to the socket ...
1329                  */
1330                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1331                 if (!ret || errno != ECONNREFUSED) {
1332                         /* ... yes, so scrub must be running. error out */
1333                         fprintf(stderr, "ERROR: scrub already running\n");
1334                         close(prg_fd);
1335                         prg_fd = -1;
1336                         goto out;
1337                 }
1338                 /*
1339                  * ... no, this means someone left us alone with an unused
1340                  * socket in the file system. remove it and try again.
1341                  */
1342                 ret = unlink(sock_path);
1343         }
1344         if (ret != -1)
1345                 ret = listen(prg_fd, 100);
1346         if (ret == -1) {
1347                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1348                     "socket at %s: %s. Progress cannot be queried\n",
1349                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1350                     strerror(errno));
1351                 if (prg_fd != -1) {
1352                         close(prg_fd);
1353                         prg_fd = -1;
1354                         if (sock_path[0])
1355                                 unlink(sock_path);
1356                 }
1357         }
1358
1359         if (do_record) {
1360                 /* write all-zero progress file for a start */
1361                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1362                                            fi_args.num_devices);
1363                 if (ret) {
1364                         ERR(!do_quiet, "WARNING: failed to write the progress "
1365                             "status file: %s. Status recording disabled\n",
1366                             strerror(-ret));
1367                         do_record = 0;
1368                 }
1369         }
1370
1371         if (do_background) {
1372                 pid = fork();
1373                 if (pid == -1) {
1374                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1375                                         "%s\n", strerror(errno));
1376                         err = 1;
1377                         goto out;
1378                 }
1379
1380                 if (pid) {
1381                         int stat;
1382                         scrub_handle_sigint_parent();
1383                         if (!do_quiet)
1384                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1385                                        n_start ? "started" : "resumed",
1386                                        path, fsid, pid);
1387                         if (!do_wait) {
1388                                 err = 0;
1389                                 goto out;
1390                         }
1391                         ret = wait(&stat);
1392                         if (ret != pid) {
1393                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1394                                     "%s\n", ret, strerror(errno));
1395                                 err = 1;
1396                                 goto out;
1397                         }
1398                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1399                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1400                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1401                                 goto out;
1402                         }
1403                         err = 0;
1404                         goto out;
1405                 }
1406         }
1407
1408         scrub_handle_sigint_child(fdmnt);
1409
1410         for (i = 0; i < fi_args.num_devices; ++i) {
1411                 if (sp[i].skip) {
1412                         sp[i].scrub_args.progress = sp[i].resumed->p;
1413                         sp[i].stats = sp[i].resumed->stats;
1414                         sp[i].ret = 0;
1415                         sp[i].stats.finished = 1;
1416                         continue;
1417                 }
1418                 devid = di_args[i].devid;
1419                 gettimeofday(&tv, NULL);
1420                 sp[i].stats.t_start = tv.tv_sec;
1421                 ret = pthread_create(&t_devs[i], NULL,
1422                                         scrub_one_dev, &sp[i]);
1423                 if (ret) {
1424                         if (do_print)
1425                                 fprintf(stderr, "ERROR: creating "
1426                                         "scrub_one_dev[%llu] thread failed: "
1427                                         "%s\n", devid, strerror(ret));
1428                         err = 1;
1429                         goto out;
1430                 }
1431         }
1432
1433         spc.fdmnt = fdmnt;
1434         spc.prg_fd = prg_fd;
1435         spc.do_record = do_record;
1436         spc.write_mutex = &spc_write_mutex;
1437         spc.shared_progress = sp;
1438         spc.fi = &fi_args;
1439         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1440         if (ret) {
1441                 if (do_print)
1442                         fprintf(stderr, "ERROR: creating progress thread "
1443                                 "failed: %s\n", strerror(ret));
1444                 err = 1;
1445                 goto out;
1446         }
1447
1448         err = 0;
1449         for (i = 0; i < fi_args.num_devices; ++i) {
1450                 if (sp[i].skip)
1451                         continue;
1452                 devid = di_args[i].devid;
1453                 ret = pthread_join(t_devs[i], NULL);
1454                 if (ret) {
1455                         if (do_print)
1456                                 fprintf(stderr, "ERROR: pthread_join failed "
1457                                         "for scrub_one_dev[%llu]: %s\n", devid,
1458                                         strerror(ret));
1459                         ++err;
1460                         continue;
1461                 }
1462                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1463                         if (do_print)
1464                                 fprintf(stderr, "WARNING: device %lld not "
1465                                         "present\n", devid);
1466                         continue;
1467                 }
1468                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1469                         ++err;
1470                 } else if (sp[i].ret) {
1471                         if (do_print)
1472                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1473                                         "for device id %lld (%s)\n", path,
1474                                         devid, strerror(sp[i].ioctl_errno));
1475                         ++err;
1476                         continue;
1477                 }
1478                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1479                         e_uncorrectable++;
1480                 if (sp[i].scrub_args.progress.corrected_errors > 0
1481                     || sp[i].scrub_args.progress.unverified_errors > 0)
1482                         e_correctable++;
1483         }
1484
1485         if (do_print) {
1486                 const char *append = "done";
1487                 if (!do_stats_per_dev)
1488                         init_fs_stat(&fs_stat);
1489                 for (i = 0; i < fi_args.num_devices; ++i) {
1490                         if (do_stats_per_dev) {
1491                                 print_scrub_dev(&di_args[i],
1492                                                 &sp[i].scrub_args.progress,
1493                                                 print_raw,
1494                                                 sp[i].ret ? "canceled" : "done",
1495                                                 &sp[i].stats);
1496                         } else {
1497                                 if (sp[i].ret)
1498                                         append = "canceled";
1499                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1500                                                 &sp[i].stats, &fs_stat);
1501                         }
1502                 }
1503                 if (!do_stats_per_dev) {
1504                         printf("scrub %s for %s\n", append, fsid);
1505                         print_fs_stat(&fs_stat, print_raw);
1506                 }
1507         }
1508
1509         ret = pthread_cancel(t_prog);
1510         if (!ret)
1511                 ret = pthread_join(t_prog, &terr);
1512
1513         /* check for errors from the handling of the progress thread */
1514         if (do_print && ret) {
1515                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1516                         strerror(ret));
1517         }
1518
1519         /* check for errors returned from the progress thread itself */
1520         if (do_print && terr && terr != PTHREAD_CANCELED) {
1521                 fprintf(stderr, "ERROR: recording progress "
1522                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1523         }
1524
1525         if (do_record) {
1526                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1527                                            fi_args.num_devices);
1528                 if (ret && do_print) {
1529                         fprintf(stderr, "ERROR: failed to record the result: "
1530                                 "%s\n", strerror(-ret));
1531                 }
1532         }
1533
1534         scrub_handle_sigint_child(-1);
1535
1536 out:
1537         free_history(past_scrubs);
1538         free(di_args);
1539         free(t_devs);
1540         free(sp);
1541         free(spc.progress);
1542         if (prg_fd > -1) {
1543                 close(prg_fd);
1544                 if (sock_path[0])
1545                         unlink(sock_path);
1546         }
1547         close_file_or_dir(fdmnt, dirstream);
1548
1549         if (err)
1550                 return 1;
1551         if (nothing_to_resume)
1552                 return 2;
1553         if (e_uncorrectable) {
1554                 ERR(!do_quiet, "ERROR: There are uncorrectable errors.\n");
1555                 return 3;
1556         }
1557         if (e_correctable)
1558                 ERR(!do_quiet, "WARNING: errors detected during scrubbing, corrected.\n");
1559
1560         return 0;
1561 }
1562
1563 static const char * const cmd_scrub_start_usage[] = {
1564         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1565         "Start a new scrub. If a scrub is already running, the new one fails.",
1566         "",
1567         "-B     do not background",
1568         "-d     stats per device (-B only)",
1569         "-q     be quiet",
1570         "-r     read only mode",
1571         "-R     raw print mode, print full data instead of summary",
1572         "-c     set ioprio class (see ionice(1) manpage)",
1573         "-n     set ioprio classdata (see ionice(1) manpage)",
1574         "-f     force starting new scrub even if a scrub is already running",
1575         "       this is useful when scrub stats record file is damaged",
1576         NULL
1577 };
1578
1579 static int cmd_scrub_start(int argc, char **argv)
1580 {
1581         return scrub_start(argc, argv, 0);
1582 }
1583
1584 static const char * const cmd_scrub_cancel_usage[] = {
1585         "btrfs scrub cancel <path>|<device>",
1586         "Cancel a running scrub",
1587         NULL
1588 };
1589
1590 static int cmd_scrub_cancel(int argc, char **argv)
1591 {
1592         char *path;
1593         int ret;
1594         int fdmnt = -1;
1595         DIR *dirstream = NULL;
1596
1597         if (check_argc_exact(argc, 2))
1598                 usage(cmd_scrub_cancel_usage);
1599
1600         path = argv[1];
1601
1602         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1603         if (fdmnt < 0) {
1604                 if (errno == EINVAL)
1605                         fprintf(stderr,
1606                                 "ERROR: '%s' is not a mounted btrfs device\n",
1607                                 path);
1608                 else
1609                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1610                                 path, strerror(errno));
1611                 ret = 1;
1612                 goto out;
1613         }
1614
1615         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1616
1617         if (ret < 0) {
1618                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1619                         errno == ENOTCONN ? "not running" : strerror(errno));
1620                 if (errno == ENOTCONN)
1621                         ret = 2;
1622                 else
1623                         ret = 1;
1624                 goto out;
1625         }
1626
1627         ret = 0;
1628         printf("scrub cancelled\n");
1629
1630 out:
1631         close_file_or_dir(fdmnt, dirstream);
1632         return ret;
1633 }
1634
1635 static const char * const cmd_scrub_resume_usage[] = {
1636         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1637         "Resume previously canceled or interrupted scrub",
1638         "",
1639         "-B     do not background",
1640         "-d     stats per device (-B only)",
1641         "-q     be quiet",
1642         "-r     read only mode",
1643         "-R     raw print mode, print full data instead of summary",
1644         "-c     set ioprio class (see ionice(1) manpage)",
1645         "-n     set ioprio classdata (see ionice(1) manpage)",
1646         NULL
1647 };
1648
1649 static int cmd_scrub_resume(int argc, char **argv)
1650 {
1651         return scrub_start(argc, argv, 1);
1652 }
1653
1654 static const char * const cmd_scrub_status_usage[] = {
1655         "btrfs scrub status [-dR] <path>|<device>",
1656         "Show status of running or finished scrub",
1657         "",
1658         "-d     stats per device",
1659         "-R     print raw stats",
1660         NULL
1661 };
1662
1663 static int cmd_scrub_status(int argc, char **argv)
1664 {
1665         char *path;
1666         struct btrfs_ioctl_fs_info_args fi_args;
1667         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1668         struct scrub_file_record **past_scrubs = NULL;
1669         struct scrub_file_record *last_scrub;
1670         struct scrub_fs_stat fs_stat;
1671         struct sockaddr_un addr = {
1672                 .sun_family = AF_UNIX,
1673         };
1674         int in_progress;
1675         int ret;
1676         int i;
1677         int fdmnt;
1678         int print_raw = 0;
1679         int do_stats_per_dev = 0;
1680         int c;
1681         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1682         int fdres = -1;
1683         int err = 0;
1684         DIR *dirstream = NULL;
1685
1686         optind = 1;
1687         while ((c = getopt(argc, argv, "dR")) != -1) {
1688                 switch (c) {
1689                 case 'd':
1690                         do_stats_per_dev = 1;
1691                         break;
1692                 case 'R':
1693                         print_raw = 1;
1694                         break;
1695                 case '?':
1696                 default:
1697                         usage(cmd_scrub_status_usage);
1698                 }
1699         }
1700
1701         if (check_argc_exact(argc - optind, 1))
1702                 usage(cmd_scrub_status_usage);
1703
1704         path = argv[optind];
1705
1706         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1707
1708         if (fdmnt < 0) {
1709                 if (errno == EINVAL)
1710                         fprintf(stderr,
1711                                 "ERROR: '%s' is not a mounted btrfs device\n",
1712                                 path);
1713                 else
1714                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1715                                 path, strerror(errno));
1716                 return 1;
1717         }
1718
1719         ret = get_fs_info(path, &fi_args, &di_args);
1720         if (ret) {
1721                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1722                                 "%s\n", strerror(-ret));
1723                 err = 1;
1724                 goto out;
1725         }
1726         if (!fi_args.num_devices) {
1727                 fprintf(stderr, "ERROR: no devices found\n");
1728                 err = 1;
1729                 goto out;
1730         }
1731
1732         uuid_unparse(fi_args.fsid, fsid);
1733
1734         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1735         if (fdres == -1) {
1736                 fprintf(stderr, "ERROR: failed to create socket to "
1737                         "receive progress information: %s\n",
1738                         strerror(errno));
1739                 err = 1;
1740                 goto out;
1741         }
1742         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1743                         NULL, addr.sun_path, sizeof(addr.sun_path));
1744         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1745         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1746         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1747         if (ret == -1) {
1748                 close(fdres);
1749                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1750                 if (fdres < 0 && fdres != -ENOENT) {
1751                         fprintf(stderr, "WARNING: failed to open status file: "
1752                                 "%s\n", strerror(-fdres));
1753                         err = 1;
1754                         goto out;
1755                 }
1756         }
1757
1758         if (fdres >= 0) {
1759                 past_scrubs = scrub_read_file(fdres, 1);
1760                 if (IS_ERR(past_scrubs))
1761                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1762                                 strerror(-PTR_ERR(past_scrubs)));
1763         }
1764         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1765
1766         printf("scrub status for %s\n", fsid);
1767
1768         if (do_stats_per_dev) {
1769                 for (i = 0; i < fi_args.num_devices; ++i) {
1770                         last_scrub = last_dev_scrub(past_scrubs,
1771                                                         di_args[i].devid);
1772                         if (!last_scrub) {
1773                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1774                                                 NULL, NULL);
1775                                 continue;
1776                         }
1777                         last_scrub->stats.in_progress = in_progress;
1778                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1779                                         last_scrub->stats.finished ?
1780                                                         "history" : "status",
1781                                         &last_scrub->stats);
1782                 }
1783         } else {
1784                 init_fs_stat(&fs_stat);
1785                 fs_stat.s.in_progress = in_progress;
1786                 for (i = 0; i < fi_args.num_devices; ++i) {
1787                         last_scrub = last_dev_scrub(past_scrubs,
1788                                                         di_args[i].devid);
1789                         if (!last_scrub)
1790                                 continue;
1791                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1792                                         &fs_stat);
1793                 }
1794                 print_fs_stat(&fs_stat, print_raw);
1795         }
1796
1797 out:
1798         free_history(past_scrubs);
1799         free(di_args);
1800         if (fdres > -1)
1801                 close(fdres);
1802         close_file_or_dir(fdmnt, dirstream);
1803
1804         return !!err;
1805 }
1806
1807 static const char scrub_cmd_group_info[] =
1808 "verify checksums of data and metadata";
1809
1810 const struct cmd_group scrub_cmd_group = {
1811         scrub_cmd_group_usage, scrub_cmd_group_info, {
1812                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1813                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1814                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1815                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1816                 NULL_CMD_STRUCT
1817         }
1818 };
1819
1820 int cmd_scrub(int argc, char **argv)
1821 {
1822         return handle_command_group(&scrub_cmd_group, argc, argv);
1823 }