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 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/wait.h>
  24 #include <sys/stat.h>
  25 #include <sys/types.h>
  26 #include <sys/socket.h>
  27 #include <sys/un.h>
  28 #include <sys/syscall.h>
  29 #include <poll.h>
  30 #include <sys/file.h>
  31 #include <uuid/uuid.h>
  32 #include <fcntl.h>
  33 #include <unistd.h>
  34 #include <pthread.h>
  35 #include <ctype.h>
  36 #include <signal.h>
  37 #include <stdarg.h>
  38 #include <limits.h>
  39
  40 #include "ctree.h"
  41 #include "ioctl.h"
  42 #include "utils.h"
  43 #include "volumes.h"
  44 #include "disk-io.h"
  45
  46 #include "commands.h"
  47
  48 static const char * const scrub_cmd_group_usage[] = {
  49         "btrfs scrub <command> [options] <path>|<device>",
  50         NULL
  51 };
  52
  53 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  54 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  55 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  56 #define SCRUB_FILE_VERSION "1"
  57
  58 struct scrub_stats {
  59         time_t t_start;
  60         time_t t_resumed;
  61         u64 duration;
  62         u64 finished;
  63         u64 canceled;
  64         int in_progress;
  65 };
  66
  67 /* TBD: replace with #include "linux/ioprio.h" in some years */
  68 #if !defined (IOPRIO_H)
  69 #define IOPRIO_WHO_PROCESS 1
  70 #define IOPRIO_CLASS_SHIFT 13
  71 #define IOPRIO_PRIO_VALUE(class, data) \
  72                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
  73 #define IOPRIO_CLASS_IDLE 3
  74 #endif
  75
  76 struct scrub_progress {
  77         struct btrfs_ioctl_scrub_args scrub_args;
  78         int fd;
  79         int ret;
  80         int skip;
  81         struct scrub_stats stats;
  82         struct scrub_file_record *resumed;
  83         int ioctl_errno;
  84         pthread_mutex_t progress_mutex;
  85         int ioprio_class;
  86         int ioprio_classdata;
  87 };
  88
  89 struct scrub_file_record {
  90         u8 fsid[BTRFS_FSID_SIZE];
  91         u64 devid;
  92         struct scrub_stats stats;
  93         struct btrfs_scrub_progress p;
  94 };
  95
  96 struct scrub_progress_cycle {
  97         int fdmnt;
  98         int prg_fd;
  99         int do_record;
 100         struct btrfs_ioctl_fs_info_args *fi;
 101         struct scrub_progress *progress;
 102         struct scrub_progress *shared_progress;
 103         pthread_mutex_t *write_mutex;
 104 };
 105
 106 struct scrub_fs_stat {
 107         struct btrfs_scrub_progress p;
 108         struct scrub_stats s;
 109         int i;
 110 };
 111
 112 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 113 {
 114         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 115         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 116         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 117         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 118         printf("\tread_errors: %lld\n", sp->read_errors);
 119         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 120         printf("\tverify_errors: %lld\n", sp->verify_errors);
 121         printf("\tno_csum: %lld\n", sp->no_csum);
 122         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 123         printf("\tsuper_errors: %lld\n", sp->super_errors);
 124         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 125         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 126         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 127         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 128         printf("\tlast_physical: %lld\n", sp->last_physical);
 129 }
 130
 131 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 132         if (test)                               \
 133                 printf(" %s=%llu", desc, test); \
 134 } while (0)
 135
 136 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 137 {
 138         u64 err_cnt;
 139         u64 err_cnt2;
 140
 141         err_cnt = p->read_errors +
 142                         p->csum_errors +
 143                         p->verify_errors +
 144                         p->super_errors;
 145
 146         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 147
 148         if (p->malloc_errors)
 149                 printf("*** WARNING: memory allocation failed while scrubbing. "
 150                        "results may be inaccurate\n");
 151
 152         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
 153                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
 154                 max(err_cnt, err_cnt2));
 155
 156         if (err_cnt || err_cnt2) {
 157                 printf("\terror details:");
 158                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 159                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 160                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 161                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 162                 printf("\n");
 163                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 164                         "unverified errors: %llu\n", p->corrected_errors,
 165                         p->uncorrectable_errors, p->unverified_errors);
 166         }
 167 }
 168
 169 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 170         fs_stat->p.name += p->name;             \
 171 } while (0)
 172
 173 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 174 do {                                            \
 175         if (fs_stat->s.name > ss->name) {       \
 176                 fs_stat->s.name = ss->name;     \
 177         }                                       \
 178 } while (0)
 179
 180 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 181 do {                                                            \
 182         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 183                 fs_stat->s.name = ss->name;                     \
 184         }                                                       \
 185 } while (0)
 186
 187 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 188 do {                                                                    \
 189         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 190                 (fs_stat)->s.name = (ss)->name;                         \
 191         }                                                               \
 192 } while (0)
 193
 194 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 195                                 struct scrub_stats *ss,
 196                                 struct scrub_fs_stat *fs_stat)
 197 {
 198         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 199         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 200         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 201         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 202         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 203         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 204         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 205         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 206         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 207         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 208         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 209         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 210         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 211         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 212         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 213         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 214         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 215         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 216         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 217 }
 218
 219 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 220 {
 221         memset(fs_stat, 0, sizeof(*fs_stat));
 222         fs_stat->s.finished = 1;
 223 }
 224
 225 static void _print_scrub_ss(struct scrub_stats *ss)
 226 {
 227         char t[4096];
 228         struct tm tm;
 229         time_t seconds;
 230         unsigned hours;
 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
 248         seconds = ss->duration;
 249         hours = ss->duration / (60 * 60);
 250         gmtime_r(&seconds, &tm);
 251         strftime(t, sizeof(t), "%M:%S", &tm);
 252         if (ss->in_progress)
 253                 printf(", running for %02u:%s\n", hours, t);
 254         else if (ss->canceled)
 255                 printf(" and was aborted after %02u:%s\n", hours, t);
 256         else if (ss->finished)
 257                 printf(" and finished after %02u:%s\n", hours, t);
 258         else
 259                 printf(", interrupted after %02u:%s, not running\n",
 260                        hours, t);
 261 }
 262
 263 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 264                                 struct btrfs_scrub_progress *p, int raw,
 265                                 const char *append, struct scrub_stats *ss)
 266 {
 267         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 268                append ? append : "");
 269
 270         _print_scrub_ss(ss);
 271
 272         if (p) {
 273                 if (raw)
 274                         print_scrub_full(p);
 275                 else
 276                         print_scrub_summary(p);
 277         }
 278 }
 279
 280 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 281 {
 282         _print_scrub_ss(&fs_stat->s);
 283
 284         if (raw)
 285                 print_scrub_full(&fs_stat->p);
 286         else
 287                 print_scrub_summary(&fs_stat->p);
 288 }
 289
 290 static void free_history(struct scrub_file_record **last_scrubs)
 291 {
 292         struct scrub_file_record **l = last_scrubs;
 293         if (!l || IS_ERR(l))
 294                 return;
 295         while (*l)
 296                 free(*l++);
 297         free(last_scrubs);
 298 }
 299
 300 /*
 301  * cancels a running scrub and makes the master process record the current
 302  * progress status before exiting.
 303  */
 304 static int cancel_fd = -1;
 305 static void scrub_sigint_record_progress(int signal)
 306 {
 307         int ret;
 308
 309         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 310         if (ret < 0)
 311                 perror("Scrub cancel failed");
 312 }
 313
 314 static int scrub_handle_sigint_parent(void)
 315 {
 316         struct sigaction sa = {
 317                 .sa_handler = SIG_IGN,
 318                 .sa_flags = SA_RESTART,
 319         };
 320
 321         return sigaction(SIGINT, &sa, NULL);
 322 }
 323
 324 static int scrub_handle_sigint_child(int fd)
 325 {
 326         struct sigaction sa = {
 327                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 328         };
 329
 330         cancel_fd = fd;
 331         return sigaction(SIGINT, &sa, NULL);
 332 }
 333
 334 static int scrub_datafile(const char *fn_base, const char *fn_local,
 335                                 const char *fn_tmp, char *datafile, int size)
 336 {
 337         int ret;
 338         int end = size - 2;
 339
 340         datafile[end + 1] = '\0';
 341         strncpy(datafile, fn_base, end);
 342         ret = strlen(datafile);
 343
 344         if (ret + 1 > end)
 345                 return -EOVERFLOW;
 346
 347         datafile[ret] = '.';
 348         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 349         ret = strlen(datafile);
 350
 351         if (ret + 1 > end)
 352                 return -EOVERFLOW;
 353
 354         if (fn_tmp) {
 355                 datafile[ret] = '_';
 356                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 357                 ret = strlen(datafile);
 358
 359                 if (ret > end)
 360                         return -EOVERFLOW;
 361         }
 362
 363         return 0;
 364 }
 365
 366 static int scrub_open_file(const char *datafile, int m)
 367 {
 368         int fd;
 369         int ret;
 370
 371         fd = open(datafile, m, 0600);
 372         if (fd < 0)
 373                 return -errno;
 374
 375         ret = flock(fd, LOCK_EX|LOCK_NB);
 376         if (ret) {
 377                 ret = errno;
 378                 close(fd);
 379                 return -ret;
 380         }
 381
 382         return fd;
 383 }
 384
 385 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 386 {
 387         int ret;
 388         char datafile[PATH_MAX];
 389         ret = scrub_datafile(fn_base, fn_local, NULL,
 390                                 datafile, sizeof(datafile));
 391         if (ret < 0)
 392                 return ret;
 393         return scrub_open_file(datafile, O_RDONLY);
 394 }
 395
 396 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 397                                 const char *tmp)
 398 {
 399         int ret;
 400         char datafile[PATH_MAX];
 401         ret = scrub_datafile(fn_base, fn_local, tmp,
 402                                 datafile, sizeof(datafile));
 403         if (ret < 0)
 404                 return ret;
 405         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 406 }
 407
 408 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 409                                 const char *tmp)
 410 {
 411         int ret;
 412         char datafile_old[PATH_MAX];
 413         char datafile_new[PATH_MAX];
 414         ret = scrub_datafile(fn_base, fn_local, tmp,
 415                                 datafile_old, sizeof(datafile_old));
 416         if (ret < 0)
 417                 return ret;
 418         ret = scrub_datafile(fn_base, fn_local, NULL,
 419                                 datafile_new, sizeof(datafile_new));
 420         if (ret < 0)
 421                 return ret;
 422         ret = rename(datafile_old, datafile_new);
 423         return ret ? -errno : 0;
 424 }
 425
 426 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 427         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 428 }
 429
 430 /*
 431  * returns 0 if the key did not match (nothing was read)
 432  *         1 if the key did match (success)
 433  *        -1 if the key did match and an error occurred
 434  */
 435 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 436                         const char *key, u64 *dest)
 437 {
 438         int j;
 439
 440         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 441                 *i += len - 1;
 442                 if (buf[*i] != ':')
 443                         return -1;
 444                 *i += 1;
 445                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 446                         ;
 447                 if (*i + j >= avail)
 448                         return -1;
 449                 *dest = atoll(&buf[*i]);
 450                 *i += j;
 451                 return 1;
 452         }
 453
 454         return 0;
 455 }
 456
 457 #define _SCRUB_INVALID do {                                             \
 458         if (report_errors)                                              \
 459                 warning("invalid data on line %d pos "                  \
 460                         "%d state %d (near \"%.*s\") at %s:%d",         \
 461                         lineno, i, state, 20 > avail ? avail : 20,      \
 462                         l + i,  __FILE__, __LINE__);                    \
 463         goto skip;                                                      \
 464 } while (0)
 465
 466 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 467 {
 468         int avail = 0;
 469         int old_avail = 0;
 470         char l[16 * 1024];
 471         int state = 0;
 472         int curr = -1;
 473         int i = 0;
 474         int j;
 475         int ret;
 476         int eof = 0;
 477         int lineno = 0;
 478         u64 version;
 479         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 480         struct scrub_file_record **p = NULL;
 481
 482 again:
 483         old_avail = avail - i;
 484         BUG_ON(old_avail < 0);
 485         if (old_avail)
 486                 memmove(l, l + i, old_avail);
 487         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 488         if (avail == 0)
 489                 eof = 1;
 490         if (avail == 0 && old_avail == 0) {
 491                 if (curr >= 0 &&
 492                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 493                         p[curr] = NULL;
 494                 } else if (curr == -1) {
 495                         p = ERR_PTR(-ENODATA);
 496                 }
 497                 return p;
 498         }
 499         if (avail == -1) {
 500                 free_history(p);
 501                 return ERR_PTR(-errno);
 502         }
 503         avail += old_avail;
 504
 505         i = 0;
 506         while (i < avail) {
 507                 void *tmp;
 508
 509                 switch (state) {
 510                 case 0: /* start of file */
 511                         ret = scrub_kvread(&i,
 512                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 513                                 SCRUB_FILE_VERSION_PREFIX, &version);
 514                         if (ret != 1)
 515                                 _SCRUB_INVALID;
 516                         if (version != atoll(SCRUB_FILE_VERSION))
 517                                 return ERR_PTR(-ENOTSUP);
 518                         state = 6;
 519                         continue;
 520                 case 1: /* start of line, alloc */
 521                         /*
 522                          * this state makes sure we have a complete line in
 523                          * further processing, so we don't need wrap-tracking
 524                          * everywhere.
 525                          */
 526                         if (!eof && !memchr(l + i, '\n', avail - i))
 527                                 goto again;
 528                         ++lineno;
 529                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 530                                                 BTRFS_FSID_SIZE) == 0) {
 531                                 state = 2;
 532                                 continue;
 533                         }
 534                         ++curr;
 535                         tmp = p;
 536                         p = realloc(p, (curr + 2) * sizeof(*p));
 537                         if (!p) {
 538                                 free_history(tmp);
 539                                 return ERR_PTR(-errno);
 540                         }
 541                         p[curr] = malloc(sizeof(**p));
 542                         if (!p[curr]) {
 543                                 free_history(p);
 544                                 return ERR_PTR(-errno);
 545                         }
 546                         memset(p[curr], 0, sizeof(**p));
 547                         p[curr + 1] = NULL;
 548                         ++state;
 549                         /* fall through */
 550                 case 2: /* start of line, skip space */
 551                         while (isspace(l[i]) && i < avail) {
 552                                 if (l[i] == '\n')
 553                                         ++lineno;
 554                                 ++i;
 555                         }
 556                         if (i >= avail ||
 557                             (!eof && !memchr(l + i, '\n', avail - i)))
 558                                 goto again;
 559                         ++state;
 560                         /* fall through */
 561                 case 3: /* read fsid */
 562                         if (i == avail)
 563                                 continue;
 564                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 565                                 ;
 566                         if (i + j + 1 >= avail)
 567                                 _SCRUB_INVALID;
 568                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 569                                 _SCRUB_INVALID;
 570                         l[i + j] = '\0';
 571                         ret = uuid_parse(l + i, p[curr]->fsid);
 572                         if (ret)
 573                                 _SCRUB_INVALID;
 574                         i += j + 1;
 575                         ++state;
 576                         /* fall through */
 577                 case 4: /* read dev id */
 578                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 579                                 ;
 580                         if (j == 0 || i + j + 1 >= avail)
 581                                 _SCRUB_INVALID;
 582                         p[curr]->devid = atoll(&l[i]);
 583                         i += j + 1;
 584                         ++state;
 585                         /* fall through */
 586                 case 5: /* read key/value pair */
 587                         ret = 0;
 588                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 589                                         &p[curr]->p);
 590                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 591                                         &p[curr]->p);
 592                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 593                                         &p[curr]->p);
 594                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 595                                         &p[curr]->p);
 596                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 597                                         &p[curr]->p);
 598                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 599                                         &p[curr]->p);
 600                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 601                                         &p[curr]->p);
 602                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 603                                         &p[curr]->p);
 604                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 605                                         &p[curr]->p);
 606                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 607                                         &p[curr]->p);
 608                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 609                                         &p[curr]->p);
 610                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 611                                         &p[curr]->p);
 612                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 613                                         &p[curr]->p);
 614                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 615                                         &p[curr]->p);
 616                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 617                                         &p[curr]->p);
 618                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 619                                         &p[curr]->stats);
 620                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 621                                         (u64 *)&p[curr]->stats);
 622                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 623                                         (u64 *)&p[curr]->stats);
 624                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 625                                         (u64 *)&p[curr]->stats);
 626                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 627                                         &p[curr]->stats);
 628                         if (ret != 1)
 629                                 _SCRUB_INVALID;
 630                         ++state;
 631                         /* fall through */
 632                 case 6: /* after number */
 633                         if (l[i] == '|')
 634                                 state = 5;
 635                         else if (l[i] == '\n')
 636                                 state = 1;
 637                         else
 638                                 _SCRUB_INVALID;
 639                         ++i;
 640                         continue;
 641                 case 99: /* skip rest of line */
 642 skip:
 643                         state = 99;
 644                         do {
 645                                 ++i;
 646                                 if (l[i - 1] == '\n') {
 647                                         state = 1;
 648                                         break;
 649                                 }
 650                         } while (i < avail);
 651                         continue;
 652                 }
 653                 BUG();
 654         }
 655         goto again;
 656 }
 657
 658 static int scrub_write_buf(int fd, const void *data, int len)
 659 {
 660         int ret;
 661         ret = write(fd, data, len);
 662         return ret - len;
 663 }
 664
 665 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 666                                 __attribute__ ((format (printf, 4, 5)));
 667 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 668 {
 669         int ret;
 670         va_list args;
 671
 672         va_start(args, fmt);
 673         ret = vsnprintf(buf, max, fmt, args);
 674         va_end(args);
 675         if (ret >= max)
 676                 return ret - max;
 677         return scrub_write_buf(fd, buf, ret);
 678 }
 679
 680 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 681                         data->resumed->p.name + data->scrub_args.progress.name
 682
 683 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 684                                                   struct scrub_progress *dest)
 685 {
 686         if (!data->resumed || data->skip)
 687                 return data;
 688
 689         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 690         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 691         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 692         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 693         _SCRUB_SUM(dest, data, read_errors);
 694         _SCRUB_SUM(dest, data, csum_errors);
 695         _SCRUB_SUM(dest, data, verify_errors);
 696         _SCRUB_SUM(dest, data, no_csum);
 697         _SCRUB_SUM(dest, data, csum_discards);
 698         _SCRUB_SUM(dest, data, super_errors);
 699         _SCRUB_SUM(dest, data, malloc_errors);
 700         _SCRUB_SUM(dest, data, uncorrectable_errors);
 701         _SCRUB_SUM(dest, data, corrected_errors);
 702         _SCRUB_SUM(dest, data, last_physical);
 703         dest->stats.canceled = data->stats.canceled;
 704         dest->stats.finished = data->stats.finished;
 705         dest->stats.t_resumed = data->stats.t_start;
 706         dest->stats.t_start = data->resumed->stats.t_start;
 707         dest->stats.duration = data->resumed->stats.duration +
 708                                                         data->stats.duration;
 709         dest->scrub_args.devid = data->scrub_args.devid;
 710         return dest;
 711 }
 712
 713 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 714         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 715                         use->scrub_args.progress.name)
 716
 717 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 718         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 719                         use->stats.name)
 720
 721 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 722 {
 723         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 724 }
 725
 726 static int scrub_write_file(int fd, const char *fsid,
 727                                 struct scrub_progress *data, int n)
 728 {
 729         int ret = 0;
 730         int i;
 731         char buf[1024];
 732         struct scrub_progress local;
 733         struct scrub_progress *use;
 734
 735         if (n < 1)
 736                 return -EINVAL;
 737
 738         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 739         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 740                                 SCRUB_FILE_VERSION "\n",
 741                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 742                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 743         if (ret)
 744                 return -EOVERFLOW;
 745
 746         for (i = 0; i < n; ++i) {
 747                 use = scrub_resumed_stats(&data[i], &local);
 748                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 749                     scrub_write_buf(fd, ":", 1) ||
 750                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 751                                         use->scrub_args.devid) ||
 752                     scrub_write_buf(fd, buf, ret) ||
 753                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 754                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 755                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 756                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 757                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 758                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 759                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 760                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 761                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 762                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 763                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 764                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 765                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 766                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 767                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 768                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 769                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 770                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 771                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 772                     scrub_write_buf(fd, "\n", 1)) {
 773                         return -EOVERFLOW;
 774                 }
 775         }
 776
 777         return 0;
 778 }
 779
 780 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 781                                 struct scrub_progress *data, int n)
 782 {
 783         int ret;
 784         int err;
 785         int fd = -1;
 786         int old;
 787
 788         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 789         if (ret) {
 790                 err = -ret;
 791                 goto out3;
 792         }
 793
 794         ret = pthread_mutex_lock(m);
 795         if (ret) {
 796                 err = -ret;
 797                 goto out2;
 798         }
 799
 800         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 801         if (fd < 0) {
 802                 err = fd;
 803                 goto out1;
 804         }
 805         err = scrub_write_file(fd, fsid, data, n);
 806         if (err)
 807                 goto out1;
 808         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 809         if (err)
 810                 goto out1;
 811
 812 out1:
 813         if (fd >= 0) {
 814                 ret = close(fd);
 815                 if (ret)
 816                         err = -errno;
 817         }
 818
 819         ret = pthread_mutex_unlock(m);
 820         if (ret && !err)
 821                 err = -ret;
 822
 823 out2:
 824         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 825         if (ret && !err)
 826                 err = -ret;
 827
 828 out3:
 829         return err;
 830 }
 831
 832 static void *scrub_one_dev(void *ctx)
 833 {
 834         struct scrub_progress *sp = ctx;
 835         int ret;
 836         struct timeval tv;
 837
 838         sp->stats.canceled = 0;
 839         sp->stats.duration = 0;
 840         sp->stats.finished = 0;
 841
 842         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 843                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 844                                         sp->ioprio_classdata));
 845         if (ret)
 846                 warning("setting ioprio failed: %s (ignored)",
 847                         strerror(errno));
 848
 849         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 850         gettimeofday(&tv, NULL);
 851         sp->ret = ret;
 852         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 853         sp->stats.canceled = !!ret;
 854         sp->ioctl_errno = errno;
 855         ret = pthread_mutex_lock(&sp->progress_mutex);
 856         if (ret)
 857                 return ERR_PTR(-ret);
 858         sp->stats.finished = 1;
 859         ret = pthread_mutex_unlock(&sp->progress_mutex);
 860         if (ret)
 861                 return ERR_PTR(-ret);
 862
 863         return NULL;
 864 }
 865
 866 static void *progress_one_dev(void *ctx)
 867 {
 868         struct scrub_progress *sp = ctx;
 869
 870         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 871         sp->ioctl_errno = errno;
 872
 873         return NULL;
 874 }
 875
 876 /* nb: returns a negative errno via ERR_PTR */
 877 static void *scrub_progress_cycle(void *ctx)
 878 {
 879         int ret = 0;
 880         int  perr = 0;  /* positive / pthread error returns */
 881         int old;
 882         int i;
 883         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 884         struct scrub_progress *sp;
 885         struct scrub_progress *sp_last;
 886         struct scrub_progress *sp_shared;
 887         struct timeval tv;
 888         struct scrub_progress_cycle *spc = ctx;
 889         int ndev = spc->fi->num_devices;
 890         int this = 1;
 891         int last = 0;
 892         int peer_fd = -1;
 893         struct pollfd accept_poll_fd = {
 894                 .fd = spc->prg_fd,
 895                 .events = POLLIN,
 896                 .revents = 0,
 897         };
 898         struct pollfd write_poll_fd = {
 899                 .events = POLLOUT,
 900                 .revents = 0,
 901         };
 902         struct sockaddr_un peer;
 903         socklen_t peer_size = sizeof(peer);
 904
 905         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 906         if (perr)
 907                 goto out;
 908
 909         uuid_unparse(spc->fi->fsid, fsid);
 910
 911         for (i = 0; i < ndev; ++i) {
 912                 sp = &spc->progress[i];
 913                 sp_last = &spc->progress[i + ndev];
 914                 sp_shared = &spc->shared_progress[i];
 915                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 916                                                 sp_shared->scrub_args.devid;
 917                 sp->fd = sp_last->fd = spc->fdmnt;
 918                 sp->stats.t_start = sp_last->stats.t_start =
 919                                                 sp_shared->stats.t_start;
 920                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 921                 sp->skip = sp_last->skip = sp_shared->skip;
 922                 sp->stats.finished = sp_last->stats.finished =
 923                                                 sp_shared->stats.finished;
 924         }
 925
 926         while (1) {
 927                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 928                 if (ret == -1) {
 929                         ret = -errno;
 930                         goto out;
 931                 }
 932                 if (ret)
 933                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 934                                          &peer_size);
 935                 gettimeofday(&tv, NULL);
 936                 this = (this + 1)%2;
 937                 last = (last + 1)%2;
 938                 for (i = 0; i < ndev; ++i) {
 939                         sp = &spc->progress[this * ndev + i];
 940                         sp_last = &spc->progress[last * ndev + i];
 941                         sp_shared = &spc->shared_progress[i];
 942                         if (sp->stats.finished)
 943                                 continue;
 944                         progress_one_dev(sp);
 945                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 946                         if (!sp->ret)
 947                                 continue;
 948                         if (sp->ioctl_errno != ENOTCONN &&
 949                             sp->ioctl_errno != ENODEV) {
 950                                 ret = -sp->ioctl_errno;
 951                                 goto out;
 952                         }
 953                         /*
 954                          * scrub finished or device removed, check the
 955                          * finished flag. if unset, just use the last
 956                          * result we got for the current write and go
 957                          * on. flag should be set on next cycle, then.
 958                          */
 959                         perr = pthread_setcancelstate(
 960                                         PTHREAD_CANCEL_DISABLE, &old);
 961                         if (perr)
 962                                 goto out;
 963                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 964                         if (perr)
 965                                 goto out;
 966                         if (!sp_shared->stats.finished) {
 967                                 perr = pthread_mutex_unlock(
 968                                                 &sp_shared->progress_mutex);
 969                                 if (perr)
 970                                         goto out;
 971                                 perr = pthread_setcancelstate(
 972                                                 PTHREAD_CANCEL_ENABLE, &old);
 973                                 if (perr)
 974                                         goto out;
 975                                 memcpy(sp, sp_last, sizeof(*sp));
 976                                 continue;
 977                         }
 978                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 979                         if (perr)
 980                                 goto out;
 981                         perr = pthread_setcancelstate(
 982                                         PTHREAD_CANCEL_ENABLE, &old);
 983                         if (perr)
 984                                 goto out;
 985                         memcpy(sp, sp_shared, sizeof(*sp));
 986                         memcpy(sp_last, sp_shared, sizeof(*sp));
 987                 }
 988                 if (peer_fd != -1) {
 989                         write_poll_fd.fd = peer_fd;
 990                         ret = poll(&write_poll_fd, 1, 0);
 991                         if (ret == -1) {
 992                                 ret = -errno;
 993                                 goto out;
 994                         }
 995                         if (ret) {
 996                                 ret = scrub_write_file(
 997                                         peer_fd, fsid,
 998                                         &spc->progress[this * ndev], ndev);
 999                                 if (ret)
1000                                         goto out;
1001                         }
1002                         close(peer_fd);
1003                         peer_fd = -1;
1004                 }
1005                 if (!spc->do_record)
1006                         continue;
1007                 ret = scrub_write_progress(spc->write_mutex, fsid,
1008                                            &spc->progress[this * ndev], ndev);
1009                 if (ret)
1010                         goto out;
1011         }
1012 out:
1013         if (peer_fd != -1)
1014                 close(peer_fd);
1015         if (perr)
1016                 ret = -perr;
1017         return ERR_PTR(ret);
1018 }
1019
1020 static struct scrub_file_record *last_dev_scrub(
1021                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1022 {
1023         int i;
1024
1025         if (!past_scrubs || IS_ERR(past_scrubs))
1026                 return NULL;
1027
1028         for (i = 0; past_scrubs[i]; ++i)
1029                 if (past_scrubs[i]->devid == devid)
1030                         return past_scrubs[i];
1031
1032         return NULL;
1033 }
1034
1035 static int mkdir_p(char *path)
1036 {
1037         int i;
1038         int ret;
1039
1040         for (i = 1; i < strlen(path); ++i) {
1041                 if (path[i] != '/')
1042                         continue;
1043                 path[i] = '\0';
1044                 ret = mkdir(path, 0777);
1045                 if (ret && errno != EEXIST)
1046                         return -errno;
1047                 path[i] = '/';
1048         }
1049
1050         return 0;
1051 }
1052
1053 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1054                                   struct btrfs_ioctl_dev_info_args *di_args,
1055                                   struct scrub_file_record **past_scrubs)
1056 {
1057         int i;
1058
1059         if (!fi_args || !di_args || !past_scrubs)
1060                 return 0;
1061
1062         for (i = 0; i < fi_args->num_devices; i++) {
1063                 struct scrub_file_record *sfr =
1064                         last_dev_scrub(past_scrubs, di_args[i].devid);
1065
1066                 if (!sfr)
1067                         continue;
1068                 if (!(sfr->stats.finished || sfr->stats.canceled))
1069                         return 1;
1070         }
1071         return 0;
1072 }
1073
1074 static int is_scrub_running_in_kernel(int fd,
1075                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1076 {
1077         struct scrub_progress sp;
1078         int i;
1079         int ret;
1080
1081         for (i = 0; i < max_devices; i++) {
1082                 memset(&sp, 0, sizeof(sp));
1083                 sp.scrub_args.devid = di_args[i].devid;
1084                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1085                 if (!ret)
1086                         return 1;
1087         }
1088
1089         return 0;
1090 }
1091
1092 static const char * const cmd_scrub_start_usage[];
1093 static const char * const cmd_scrub_resume_usage[];
1094
1095 static int scrub_start(int argc, char **argv, int resume)
1096 {
1097         int fdmnt;
1098         int prg_fd = -1;
1099         int fdres = -1;
1100         int ret;
1101         pid_t pid;
1102         int c;
1103         int i;
1104         int err = 0;
1105         int e_uncorrectable = 0;
1106         int e_correctable = 0;
1107         int print_raw = 0;
1108         char *path;
1109         int do_background = 1;
1110         int do_wait = 0;
1111         int do_print = 0;
1112         int do_quiet = 0;
1113         int do_record = 1;
1114         int readonly = 0;
1115         int do_stats_per_dev = 0;
1116         int ioprio_class = IOPRIO_CLASS_IDLE;
1117         int ioprio_classdata = 0;
1118         int n_start = 0;
1119         int n_skip = 0;
1120         int n_resume = 0;
1121         struct btrfs_ioctl_fs_info_args fi_args;
1122         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1123         struct scrub_progress *sp = NULL;
1124         struct scrub_fs_stat fs_stat;
1125         struct timeval tv;
1126         struct sockaddr_un addr = {
1127                 .sun_family = AF_UNIX,
1128         };
1129         pthread_t *t_devs = NULL;
1130         pthread_t t_prog;
1131         struct scrub_file_record **past_scrubs = NULL;
1132         struct scrub_file_record *last_scrub = NULL;
1133         char *datafile = strdup(SCRUB_DATA_FILE);
1134         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1135         char sock_path[PATH_MAX] = "";
1136         struct scrub_progress_cycle spc;
1137         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1138         void *terr;
1139         u64 devid;
1140         DIR *dirstream = NULL;
1141         int force = 0;
1142         int nothing_to_resume = 0;
1143
1144         optind = 1;
1145         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1146                 switch (c) {
1147                 case 'B':
1148                         do_background = 0;
1149                         do_wait = 1;
1150                         do_print = 1;
1151                         break;
1152                 case 'd':
1153                         do_stats_per_dev = 1;
1154                         break;
1155                 case 'q':
1156                         do_quiet = 1;
1157                         break;
1158                 case 'r':
1159                         readonly = 1;
1160                         break;
1161                 case 'R':
1162                         print_raw = 1;
1163                         break;
1164                 case 'c':
1165                         ioprio_class = (int)strtol(optarg, NULL, 10);
1166                         break;
1167                 case 'n':
1168                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1169                         break;
1170                 case 'f':
1171                         force = 1;
1172                         break;
1173                 case '?':
1174                 default:
1175                         usage(resume ? cmd_scrub_resume_usage :
1176                                                 cmd_scrub_start_usage);
1177                 }
1178         }
1179
1180         /* try to catch most error cases before forking */
1181
1182         if (check_argc_exact(argc - optind, 1)) {
1183                 usage(resume ? cmd_scrub_resume_usage :
1184                                         cmd_scrub_start_usage);
1185         }
1186
1187         spc.progress = NULL;
1188         if (do_quiet && do_print)
1189                 do_print = 0;
1190
1191         if (mkdir_p(datafile)) {
1192                 warning_on(!do_quiet,
1193     "cannot create scrub data file, mkdir %s failed: %s. Status recording disabled",
1194                         datafile, strerror(errno));
1195                 do_record = 0;
1196         }
1197         free(datafile);
1198
1199         path = argv[optind];
1200
1201         fdmnt = open_path_or_dev_mnt(path, &dirstream, !do_quiet);
1202         if (fdmnt < 0)
1203                 return 1;
1204
1205         ret = get_fs_info(path, &fi_args, &di_args);
1206         if (ret) {
1207                 error_on(!do_quiet,
1208                         "getting dev info for scrub failed: %s",
1209                          strerror(-ret));
1210                 err = 1;
1211                 goto out;
1212         }
1213         if (!fi_args.num_devices) {
1214                 error_on(!do_quiet, "no devices found");
1215                 err = 1;
1216                 goto out;
1217         }
1218
1219         uuid_unparse(fi_args.fsid, fsid);
1220         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1221         if (fdres < 0 && fdres != -ENOENT) {
1222                 warning_on(!do_quiet, "failed to open status file: %s",
1223                         strerror(-fdres));
1224         } else if (fdres >= 0) {
1225                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1226                 if (IS_ERR(past_scrubs))
1227                         warning_on(!do_quiet, "failed to read status file: %s",
1228                                 strerror(-PTR_ERR(past_scrubs)));
1229                 close(fdres);
1230         }
1231
1232         /*
1233          * Check for stale information in the status file, ie. if it's
1234          * canceled=0, finished=0 but no scrub is running.
1235          */
1236         if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1237                 force = 1;
1238
1239         /*
1240          * check whether any involved device is already busy running a
1241          * scrub. This would cause damaged status messages and the state
1242          * "aborted" without the explanation that a scrub was already
1243          * running. Therefore check it first, prevent it and give some
1244          * feedback to the user if scrub is already running.
1245          * Note that if scrub is started with a block device as the
1246          * parameter, only that particular block device is checked. It
1247          * is a normal mode of operation to start scrub on multiple
1248          * single devices, there is no reason to prevent this.
1249          */
1250         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1251                 error_on(!do_quiet,
1252                         "Scrub is already running.\n"
1253                         "To cancel use 'btrfs scrub cancel %s'.\n"
1254                         "To see the status use 'btrfs scrub status [-d] %s'",
1255                         path, path);
1256                 err = 1;
1257                 goto out;
1258         }
1259
1260         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1261         sp = calloc(fi_args.num_devices, sizeof(*sp));
1262         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1263
1264         if (!t_devs || !sp || !spc.progress) {
1265                 error_on(!do_quiet, "scrub failed: %s", strerror(errno));
1266                 err = 1;
1267                 goto out;
1268         }
1269
1270         for (i = 0; i < fi_args.num_devices; ++i) {
1271                 devid = di_args[i].devid;
1272                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1273                 if (ret) {
1274                         error_on(!do_quiet, "pthread_mutex_init failed: %s",
1275                                 strerror(ret));
1276                         err = 1;
1277                         goto out;
1278                 }
1279                 last_scrub = last_dev_scrub(past_scrubs, devid);
1280                 sp[i].scrub_args.devid = devid;
1281                 sp[i].fd = fdmnt;
1282                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1283                                              !last_scrub->stats.finished)) {
1284                         ++n_resume;
1285                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1286                         sp[i].resumed = last_scrub;
1287                 } else if (resume) {
1288                         ++n_skip;
1289                         sp[i].skip = 1;
1290                         sp[i].resumed = last_scrub;
1291                         continue;
1292                 } else {
1293                         ++n_start;
1294                         sp[i].scrub_args.start = 0ll;
1295                         sp[i].resumed = NULL;
1296                 }
1297                 sp[i].skip = 0;
1298                 sp[i].scrub_args.end = (u64)-1ll;
1299                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1300                 sp[i].ioprio_class = ioprio_class;
1301                 sp[i].ioprio_classdata = ioprio_classdata;
1302         }
1303
1304         if (!n_start && !n_resume) {
1305                 if (!do_quiet)
1306                         printf("scrub: nothing to resume for %s, fsid %s\n",
1307                                path, fsid);
1308                 nothing_to_resume = 1;
1309                 goto out;
1310         }
1311
1312         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1313         while (ret != -1) {
1314                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1315                                         sock_path, sizeof(sock_path));
1316                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1317                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1318                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1319                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1320                 if (ret != -1 || errno != EADDRINUSE)
1321                         break;
1322                 /*
1323                  * bind failed with EADDRINUSE. so let's see if anyone answers
1324                  * when we make a call to the socket ...
1325                  */
1326                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1327                 if (!ret || errno != ECONNREFUSED) {
1328                         /* ... yes, so scrub must be running. error out */
1329                         error("scrub already running");
1330                         close(prg_fd);
1331                         prg_fd = -1;
1332                         goto out;
1333                 }
1334                 /*
1335                  * ... no, this means someone left us alone with an unused
1336                  * socket in the file system. remove it and try again.
1337                  */
1338                 ret = unlink(sock_path);
1339         }
1340         if (ret != -1)
1341                 ret = listen(prg_fd, 100);
1342         if (ret == -1) {
1343                 warning_on(!do_quiet,
1344    "failed to open the progress status socket at %s: %s. Progress cannot be queried",
1345                         sock_path[0] ? sock_path :
1346                         SCRUB_PROGRESS_SOCKET_PATH, strerror(errno));
1347                 if (prg_fd != -1) {
1348                         close(prg_fd);
1349                         prg_fd = -1;
1350                         if (sock_path[0])
1351                                 unlink(sock_path);
1352                 }
1353         }
1354
1355         if (do_record) {
1356                 /* write all-zero progress file for a start */
1357                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1358                                            fi_args.num_devices);
1359                 if (ret) {
1360                         warning_on(!do_quiet,
1361    "failed to write the progress status file: %s. Status recording disabled",
1362                                 strerror(-ret));
1363                         do_record = 0;
1364                 }
1365         }
1366
1367         if (do_background) {
1368                 pid = fork();
1369                 if (pid == -1) {
1370                         error_on(!do_quiet, "cannot scrub, fork failed: %s",
1371                                 strerror(errno));
1372                         err = 1;
1373                         goto out;
1374                 }
1375
1376                 if (pid) {
1377                         int stat;
1378                         scrub_handle_sigint_parent();
1379                         if (!do_quiet)
1380                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1381                                        n_start ? "started" : "resumed",
1382                                        path, fsid, pid);
1383                         if (!do_wait) {
1384                                 err = 0;
1385                                 goto out;
1386                         }
1387                         ret = wait(&stat);
1388                         if (ret != pid) {
1389                                 error_on(!do_quiet, "wait failed (ret=%d): %s",
1390                                         ret, strerror(errno));
1391                                 err = 1;
1392                                 goto out;
1393                         }
1394                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1395                                 error_on(!do_quiet, "scrub process failed");
1396                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1397                                 goto out;
1398                         }
1399                         err = 0;
1400                         goto out;
1401                 }
1402         }
1403
1404         scrub_handle_sigint_child(fdmnt);
1405
1406         for (i = 0; i < fi_args.num_devices; ++i) {
1407                 if (sp[i].skip) {
1408                         sp[i].scrub_args.progress = sp[i].resumed->p;
1409                         sp[i].stats = sp[i].resumed->stats;
1410                         sp[i].ret = 0;
1411                         sp[i].stats.finished = 1;
1412                         continue;
1413                 }
1414                 devid = di_args[i].devid;
1415                 gettimeofday(&tv, NULL);
1416                 sp[i].stats.t_start = tv.tv_sec;
1417                 ret = pthread_create(&t_devs[i], NULL,
1418                                         scrub_one_dev, &sp[i]);
1419                 if (ret) {
1420                         if (do_print)
1421                         error("creating scrub_one_dev[%llu] thread failed: %s",
1422                                 devid, strerror(ret));
1423                         err = 1;
1424                         goto out;
1425                 }
1426         }
1427
1428         spc.fdmnt = fdmnt;
1429         spc.prg_fd = prg_fd;
1430         spc.do_record = do_record;
1431         spc.write_mutex = &spc_write_mutex;
1432         spc.shared_progress = sp;
1433         spc.fi = &fi_args;
1434         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1435         if (ret) {
1436                 if (do_print)
1437                         error("creating progress thread failed: %s",
1438                                 strerror(ret));
1439                 err = 1;
1440                 goto out;
1441         }
1442
1443         err = 0;
1444         for (i = 0; i < fi_args.num_devices; ++i) {
1445                 if (sp[i].skip)
1446                         continue;
1447                 devid = di_args[i].devid;
1448                 ret = pthread_join(t_devs[i], NULL);
1449                 if (ret) {
1450                         if (do_print)
1451                           error("pthread_join failed for scrub_one_dev[%llu]: %s",
1452                                 devid, strerror(ret));
1453                         ++err;
1454                         continue;
1455                 }
1456                 if (sp[i].ret) {
1457                         switch (sp[i].ioctl_errno) {
1458                         case ENODEV:
1459                                 if (do_print)
1460                                         warning("device %lld not present",
1461                                                 devid);
1462                                 continue;
1463                         case ECANCELED:
1464                                 ++err;
1465                                 break;
1466                         default:
1467                                 if (do_print)
1468                 error("scrubbing %s failed for device id %lld: ret=%d, errno=%d (%s)",
1469                                         path, devid,
1470                                         sp[i].ret, sp[i].ioctl_errno,
1471                                         strerror(sp[i].ioctl_errno));
1472                                 ++err;
1473                                 continue;
1474                         }
1475                 }
1476                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1477                         e_uncorrectable++;
1478                 if (sp[i].scrub_args.progress.corrected_errors > 0
1479                     || sp[i].scrub_args.progress.unverified_errors > 0)
1480                         e_correctable++;
1481         }
1482
1483         if (do_print) {
1484                 const char *append = "done";
1485                 if (!do_stats_per_dev)
1486                         init_fs_stat(&fs_stat);
1487                 for (i = 0; i < fi_args.num_devices; ++i) {
1488                         if (do_stats_per_dev) {
1489                                 print_scrub_dev(&di_args[i],
1490                                                 &sp[i].scrub_args.progress,
1491                                                 print_raw,
1492                                                 sp[i].ret ? "canceled" : "done",
1493                                                 &sp[i].stats);
1494                         } else {
1495                                 if (sp[i].ret)
1496                                         append = "canceled";
1497                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1498                                                 &sp[i].stats, &fs_stat);
1499                         }
1500                 }
1501                 if (!do_stats_per_dev) {
1502                         printf("scrub %s for %s\n", append, fsid);
1503                         print_fs_stat(&fs_stat, print_raw);
1504                 }
1505         }
1506
1507         ret = pthread_cancel(t_prog);
1508         if (!ret)
1509                 ret = pthread_join(t_prog, &terr);
1510
1511         /* check for errors from the handling of the progress thread */
1512         if (do_print && ret) {
1513                 error("progress thread handling failed: %s",
1514                         strerror(ret));
1515         }
1516
1517         /* check for errors returned from the progress thread itself */
1518         if (do_print && terr && terr != PTHREAD_CANCELED)
1519                 error("recording progress failed: %s",
1520                         strerror(-PTR_ERR(terr)));
1521
1522         if (do_record) {
1523                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1524                                            fi_args.num_devices);
1525                 if (ret && do_print)
1526                         error("failed to record the result: %s",
1527                                 strerror(-ret));
1528         }
1529
1530         scrub_handle_sigint_child(-1);
1531
1532 out:
1533         free_history(past_scrubs);
1534         free(di_args);
1535         free(t_devs);
1536         free(sp);
1537         free(spc.progress);
1538         if (prg_fd > -1) {
1539                 close(prg_fd);
1540                 if (sock_path[0])
1541                         unlink(sock_path);
1542         }
1543         close_file_or_dir(fdmnt, dirstream);
1544
1545         if (err)
1546                 return 1;
1547         if (nothing_to_resume)
1548                 return 2;
1549         if (e_uncorrectable) {
1550                 error_on(!do_quiet, "there are uncorrectable errors");
1551                 return 3;
1552         }
1553         if (e_correctable)
1554                 warning_on(!do_quiet,
1555                         "errors detected during scrubbing, corrected");
1556
1557         return 0;
1558 }
1559
1560 static const char * const cmd_scrub_start_usage[] = {
1561         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1562         "Start a new scrub. If a scrub is already running, the new one fails.",
1563         "",
1564         "-B     do not background",
1565         "-d     stats per device (-B only)",
1566         "-q     be quiet",
1567         "-r     read only mode",
1568         "-R     raw print mode, print full data instead of summary",
1569         "-c     set ioprio class (see ionice(1) manpage)",
1570         "-n     set ioprio classdata (see ionice(1) manpage)",
1571         "-f     force starting new scrub even if a scrub is already running",
1572         "       this is useful when scrub stats record file is damaged",
1573         NULL
1574 };
1575
1576 static int cmd_scrub_start(int argc, char **argv)
1577 {
1578         return scrub_start(argc, argv, 0);
1579 }
1580
1581 static const char * const cmd_scrub_cancel_usage[] = {
1582         "btrfs scrub cancel <path>|<device>",
1583         "Cancel a running scrub",
1584         NULL
1585 };
1586
1587 static int cmd_scrub_cancel(int argc, char **argv)
1588 {
1589         char *path;
1590         int ret;
1591         int fdmnt = -1;
1592         DIR *dirstream = NULL;
1593
1594         clean_args_no_options(argc, argv, cmd_scrub_cancel_usage);
1595
1596         if (check_argc_exact(argc - optind, 1))
1597                 usage(cmd_scrub_cancel_usage);
1598
1599         path = argv[optind];
1600
1601         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1602         if (fdmnt < 0) {
1603                 ret = 1;
1604                 goto out;
1605         }
1606
1607         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1608
1609         if (ret < 0) {
1610                 error("scrub cancel failed on %s: %s", path,
1611                         errno == ENOTCONN ? "not running" : strerror(errno));
1612                 if (errno == ENOTCONN)
1613                         ret = 2;
1614                 else
1615                         ret = 1;
1616                 goto out;
1617         }
1618
1619         ret = 0;
1620         printf("scrub cancelled\n");
1621
1622 out:
1623         close_file_or_dir(fdmnt, dirstream);
1624         return ret;
1625 }
1626
1627 static const char * const cmd_scrub_resume_usage[] = {
1628         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1629         "Resume previously canceled or interrupted scrub",
1630         "",
1631         "-B     do not background",
1632         "-d     stats per device (-B only)",
1633         "-q     be quiet",
1634         "-r     read only mode",
1635         "-R     raw print mode, print full data instead of summary",
1636         "-c     set ioprio class (see ionice(1) manpage)",
1637         "-n     set ioprio classdata (see ionice(1) manpage)",
1638         NULL
1639 };
1640
1641 static int cmd_scrub_resume(int argc, char **argv)
1642 {
1643         return scrub_start(argc, argv, 1);
1644 }
1645
1646 static const char * const cmd_scrub_status_usage[] = {
1647         "btrfs scrub status [-dR] <path>|<device>",
1648         "Show status of running or finished scrub",
1649         "",
1650         "-d     stats per device",
1651         "-R     print raw stats",
1652         NULL
1653 };
1654
1655 static int cmd_scrub_status(int argc, char **argv)
1656 {
1657         char *path;
1658         struct btrfs_ioctl_fs_info_args fi_args;
1659         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1660         struct scrub_file_record **past_scrubs = NULL;
1661         struct scrub_file_record *last_scrub;
1662         struct scrub_fs_stat fs_stat;
1663         struct sockaddr_un addr = {
1664                 .sun_family = AF_UNIX,
1665         };
1666         int in_progress;
1667         int ret;
1668         int i;
1669         int fdmnt;
1670         int print_raw = 0;
1671         int do_stats_per_dev = 0;
1672         int c;
1673         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1674         int fdres = -1;
1675         int err = 0;
1676         DIR *dirstream = NULL;
1677
1678         optind = 1;
1679         while ((c = getopt(argc, argv, "dR")) != -1) {
1680                 switch (c) {
1681                 case 'd':
1682                         do_stats_per_dev = 1;
1683                         break;
1684                 case 'R':
1685                         print_raw = 1;
1686                         break;
1687                 case '?':
1688                 default:
1689                         usage(cmd_scrub_status_usage);
1690                 }
1691         }
1692
1693         if (check_argc_exact(argc - optind, 1))
1694                 usage(cmd_scrub_status_usage);
1695
1696         path = argv[optind];
1697
1698         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1699         if (fdmnt < 0)
1700                 return 1;
1701
1702         ret = get_fs_info(path, &fi_args, &di_args);
1703         if (ret) {
1704                 error("getting dev info for scrub failed: %s",
1705                         strerror(-ret));
1706                 err = 1;
1707                 goto out;
1708         }
1709         if (!fi_args.num_devices) {
1710                 error("no devices found");
1711                 err = 1;
1712                 goto out;
1713         }
1714
1715         uuid_unparse(fi_args.fsid, fsid);
1716
1717         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1718         if (fdres == -1) {
1719                 error("failed to create socket to receive progress information: %s",
1720                         strerror(errno));
1721                 err = 1;
1722                 goto out;
1723         }
1724         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1725                         NULL, addr.sun_path, sizeof(addr.sun_path));
1726         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1727         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1728         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1729         if (ret == -1) {
1730                 close(fdres);
1731                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1732                 if (fdres < 0 && fdres != -ENOENT) {
1733                         warning("failed to open status file: %s",
1734                                 strerror(-fdres));
1735                         err = 1;
1736                         goto out;
1737                 }
1738         }
1739
1740         if (fdres >= 0) {
1741                 past_scrubs = scrub_read_file(fdres, 1);
1742                 if (IS_ERR(past_scrubs))
1743                         warning("failed to read status: %s",
1744                                 strerror(-PTR_ERR(past_scrubs)));
1745         }
1746         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1747
1748         printf("scrub status for %s\n", fsid);
1749
1750         if (do_stats_per_dev) {
1751                 for (i = 0; i < fi_args.num_devices; ++i) {
1752                         last_scrub = last_dev_scrub(past_scrubs,
1753                                                         di_args[i].devid);
1754                         if (!last_scrub) {
1755                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1756                                                 NULL, NULL);
1757                                 continue;
1758                         }
1759                         last_scrub->stats.in_progress = in_progress;
1760                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1761                                         last_scrub->stats.finished ?
1762                                                         "history" : "status",
1763                                         &last_scrub->stats);
1764                 }
1765         } else {
1766                 init_fs_stat(&fs_stat);
1767                 fs_stat.s.in_progress = in_progress;
1768                 for (i = 0; i < fi_args.num_devices; ++i) {
1769                         last_scrub = last_dev_scrub(past_scrubs,
1770                                                         di_args[i].devid);
1771                         if (!last_scrub)
1772                                 continue;
1773                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1774                                         &fs_stat);
1775                 }
1776                 print_fs_stat(&fs_stat, print_raw);
1777         }
1778
1779 out:
1780         free_history(past_scrubs);
1781         free(di_args);
1782         if (fdres > -1)
1783                 close(fdres);
1784         close_file_or_dir(fdmnt, dirstream);
1785
1786         return !!err;
1787 }
1788
1789 static const char scrub_cmd_group_info[] =
1790 "verify checksums of data and metadata";
1791
1792 const struct cmd_group scrub_cmd_group = {
1793         scrub_cmd_group_usage, scrub_cmd_group_info, {
1794                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1795                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1796                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1797                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1798                 NULL_CMD_STRUCT
1799         }
1800 };
1801
1802 int cmd_scrub(int argc, char **argv)
1803 {
1804         return handle_command_group(&scrub_cmd_group, argc, argv);
1805 }