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         if (old_avail < 0) {
 485                 error("scrub record file corrupted near byte %d", i);
 486                 return ERR_PTR(-EINVAL);
 487         }
 488         if (old_avail)
 489                 memmove(l, l + i, old_avail);
 490         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 491         if (avail == 0)
 492                 eof = 1;
 493         if (avail == 0 && old_avail == 0) {
 494                 if (curr >= 0 &&
 495                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 496                         p[curr] = NULL;
 497                 } else if (curr == -1) {
 498                         p = ERR_PTR(-ENODATA);
 499                 }
 500                 return p;
 501         }
 502         if (avail == -1) {
 503                 free_history(p);
 504                 return ERR_PTR(-errno);
 505         }
 506         avail += old_avail;
 507
 508         i = 0;
 509         while (i < avail) {
 510                 void *tmp;
 511
 512                 switch (state) {
 513                 case 0: /* start of file */
 514                         ret = scrub_kvread(&i,
 515                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 516                                 SCRUB_FILE_VERSION_PREFIX, &version);
 517                         if (ret != 1)
 518                                 _SCRUB_INVALID;
 519                         if (version != atoll(SCRUB_FILE_VERSION))
 520                                 return ERR_PTR(-ENOTSUP);
 521                         state = 6;
 522                         continue;
 523                 case 1: /* start of line, alloc */
 524                         /*
 525                          * this state makes sure we have a complete line in
 526                          * further processing, so we don't need wrap-tracking
 527                          * everywhere.
 528                          */
 529                         if (!eof && !memchr(l + i, '\n', avail - i))
 530                                 goto again;
 531                         ++lineno;
 532                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 533                                                 BTRFS_FSID_SIZE) == 0) {
 534                                 state = 2;
 535                                 continue;
 536                         }
 537                         ++curr;
 538                         tmp = p;
 539                         p = realloc(p, (curr + 2) * sizeof(*p));
 540                         if (!p) {
 541                                 free_history(tmp);
 542                                 return ERR_PTR(-errno);
 543                         }
 544                         p[curr] = malloc(sizeof(**p));
 545                         if (!p[curr]) {
 546                                 free_history(p);
 547                                 return ERR_PTR(-errno);
 548                         }
 549                         memset(p[curr], 0, sizeof(**p));
 550                         p[curr + 1] = NULL;
 551                         ++state;
 552                         /* fall through */
 553                 case 2: /* start of line, skip space */
 554                         while (isspace(l[i]) && i < avail) {
 555                                 if (l[i] == '\n')
 556                                         ++lineno;
 557                                 ++i;
 558                         }
 559                         if (i >= avail ||
 560                             (!eof && !memchr(l + i, '\n', avail - i)))
 561                                 goto again;
 562                         ++state;
 563                         /* fall through */
 564                 case 3: /* read fsid */
 565                         if (i == avail)
 566                                 continue;
 567                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 568                                 ;
 569                         if (i + j + 1 >= avail)
 570                                 _SCRUB_INVALID;
 571                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 572                                 _SCRUB_INVALID;
 573                         l[i + j] = '\0';
 574                         ret = uuid_parse(l + i, p[curr]->fsid);
 575                         if (ret)
 576                                 _SCRUB_INVALID;
 577                         i += j + 1;
 578                         ++state;
 579                         /* fall through */
 580                 case 4: /* read dev id */
 581                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 582                                 ;
 583                         if (j == 0 || i + j + 1 >= avail)
 584                                 _SCRUB_INVALID;
 585                         p[curr]->devid = atoll(&l[i]);
 586                         i += j + 1;
 587                         ++state;
 588                         /* fall through */
 589                 case 5: /* read key/value pair */
 590                         ret = 0;
 591                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 592                                         &p[curr]->p);
 593                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 594                                         &p[curr]->p);
 595                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 596                                         &p[curr]->p);
 597                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 598                                         &p[curr]->p);
 599                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 600                                         &p[curr]->p);
 601                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 602                                         &p[curr]->p);
 603                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 604                                         &p[curr]->p);
 605                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 606                                         &p[curr]->p);
 607                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 608                                         &p[curr]->p);
 609                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 610                                         &p[curr]->p);
 611                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 612                                         &p[curr]->p);
 613                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 614                                         &p[curr]->p);
 615                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 616                                         &p[curr]->p);
 617                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 618                                         &p[curr]->p);
 619                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 620                                         &p[curr]->p);
 621                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 622                                         &p[curr]->stats);
 623                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 624                                         (u64 *)&p[curr]->stats);
 625                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 626                                         (u64 *)&p[curr]->stats);
 627                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 628                                         (u64 *)&p[curr]->stats);
 629                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 630                                         &p[curr]->stats);
 631                         if (ret != 1)
 632                                 _SCRUB_INVALID;
 633                         ++state;
 634                         /* fall through */
 635                 case 6: /* after number */
 636                         if (l[i] == '|')
 637                                 state = 5;
 638                         else if (l[i] == '\n')
 639                                 state = 1;
 640                         else
 641                                 _SCRUB_INVALID;
 642                         ++i;
 643                         continue;
 644                 case 99: /* skip rest of line */
 645 skip:
 646                         state = 99;
 647                         do {
 648                                 ++i;
 649                                 if (l[i - 1] == '\n') {
 650                                         state = 1;
 651                                         break;
 652                                 }
 653                         } while (i < avail);
 654                         continue;
 655                 }
 656                 error("internal error: unknown parser state %d near byte %d",
 657                                 state, i);
 658                 return ERR_PTR(-EINVAL);
 659         }
 660         goto again;
 661 }
 662
 663 static int scrub_write_buf(int fd, const void *data, int len)
 664 {
 665         int ret;
 666         ret = write(fd, data, len);
 667         return ret - len;
 668 }
 669
 670 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 671                                 __attribute__ ((format (printf, 4, 5)));
 672 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 673 {
 674         int ret;
 675         va_list args;
 676
 677         va_start(args, fmt);
 678         ret = vsnprintf(buf, max, fmt, args);
 679         va_end(args);
 680         if (ret >= max)
 681                 return ret - max;
 682         return scrub_write_buf(fd, buf, ret);
 683 }
 684
 685 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 686                         data->resumed->p.name + data->scrub_args.progress.name
 687
 688 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 689                                                   struct scrub_progress *dest)
 690 {
 691         if (!data->resumed || data->skip)
 692                 return data;
 693
 694         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 695         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 696         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 697         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 698         _SCRUB_SUM(dest, data, read_errors);
 699         _SCRUB_SUM(dest, data, csum_errors);
 700         _SCRUB_SUM(dest, data, verify_errors);
 701         _SCRUB_SUM(dest, data, no_csum);
 702         _SCRUB_SUM(dest, data, csum_discards);
 703         _SCRUB_SUM(dest, data, super_errors);
 704         _SCRUB_SUM(dest, data, malloc_errors);
 705         _SCRUB_SUM(dest, data, uncorrectable_errors);
 706         _SCRUB_SUM(dest, data, corrected_errors);
 707         _SCRUB_SUM(dest, data, last_physical);
 708         dest->stats.canceled = data->stats.canceled;
 709         dest->stats.finished = data->stats.finished;
 710         dest->stats.t_resumed = data->stats.t_start;
 711         dest->stats.t_start = data->resumed->stats.t_start;
 712         dest->stats.duration = data->resumed->stats.duration +
 713                                                         data->stats.duration;
 714         dest->scrub_args.devid = data->scrub_args.devid;
 715         return dest;
 716 }
 717
 718 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 719         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 720                         use->scrub_args.progress.name)
 721
 722 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 723         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 724                         use->stats.name)
 725
 726 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 727 {
 728         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 729 }
 730
 731 static int scrub_write_file(int fd, const char *fsid,
 732                                 struct scrub_progress *data, int n)
 733 {
 734         int ret = 0;
 735         int i;
 736         char buf[1024];
 737         struct scrub_progress local;
 738         struct scrub_progress *use;
 739
 740         if (n < 1)
 741                 return -EINVAL;
 742
 743         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 744         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 745                                 SCRUB_FILE_VERSION "\n",
 746                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 747                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 748         if (ret)
 749                 return -EOVERFLOW;
 750
 751         for (i = 0; i < n; ++i) {
 752                 use = scrub_resumed_stats(&data[i], &local);
 753                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 754                     scrub_write_buf(fd, ":", 1) ||
 755                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 756                                         use->scrub_args.devid) ||
 757                     scrub_write_buf(fd, buf, ret) ||
 758                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 759                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 760                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 761                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 762                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 763                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 764                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 765                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 766                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 767                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 768                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 769                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 770                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 771                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 772                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 773                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 774                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 775                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 776                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 777                     scrub_write_buf(fd, "\n", 1)) {
 778                         return -EOVERFLOW;
 779                 }
 780         }
 781
 782         return 0;
 783 }
 784
 785 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 786                                 struct scrub_progress *data, int n)
 787 {
 788         int ret;
 789         int err;
 790         int fd = -1;
 791         int old;
 792
 793         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 794         if (ret) {
 795                 err = -ret;
 796                 goto out3;
 797         }
 798
 799         ret = pthread_mutex_lock(m);
 800         if (ret) {
 801                 err = -ret;
 802                 goto out2;
 803         }
 804
 805         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 806         if (fd < 0) {
 807                 err = fd;
 808                 goto out1;
 809         }
 810         err = scrub_write_file(fd, fsid, data, n);
 811         if (err)
 812                 goto out1;
 813         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 814         if (err)
 815                 goto out1;
 816
 817 out1:
 818         if (fd >= 0) {
 819                 ret = close(fd);
 820                 if (ret)
 821                         err = -errno;
 822         }
 823
 824         ret = pthread_mutex_unlock(m);
 825         if (ret && !err)
 826                 err = -ret;
 827
 828 out2:
 829         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 830         if (ret && !err)
 831                 err = -ret;
 832
 833 out3:
 834         return err;
 835 }
 836
 837 static void *scrub_one_dev(void *ctx)
 838 {
 839         struct scrub_progress *sp = ctx;
 840         int ret;
 841         struct timeval tv;
 842
 843         sp->stats.canceled = 0;
 844         sp->stats.duration = 0;
 845         sp->stats.finished = 0;
 846
 847         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 848                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 849                                         sp->ioprio_classdata));
 850         if (ret)
 851                 warning("setting ioprio failed: %s (ignored)",
 852                         strerror(errno));
 853
 854         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 855         gettimeofday(&tv, NULL);
 856         sp->ret = ret;
 857         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 858         sp->stats.canceled = !!ret;
 859         sp->ioctl_errno = errno;
 860         ret = pthread_mutex_lock(&sp->progress_mutex);
 861         if (ret)
 862                 return ERR_PTR(-ret);
 863         sp->stats.finished = 1;
 864         ret = pthread_mutex_unlock(&sp->progress_mutex);
 865         if (ret)
 866                 return ERR_PTR(-ret);
 867
 868         return NULL;
 869 }
 870
 871 static void *progress_one_dev(void *ctx)
 872 {
 873         struct scrub_progress *sp = ctx;
 874
 875         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 876         sp->ioctl_errno = errno;
 877
 878         return NULL;
 879 }
 880
 881 /* nb: returns a negative errno via ERR_PTR */
 882 static void *scrub_progress_cycle(void *ctx)
 883 {
 884         int ret = 0;
 885         int  perr = 0;  /* positive / pthread error returns */
 886         int old;
 887         int i;
 888         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 889         struct scrub_progress *sp;
 890         struct scrub_progress *sp_last;
 891         struct scrub_progress *sp_shared;
 892         struct timeval tv;
 893         struct scrub_progress_cycle *spc = ctx;
 894         int ndev = spc->fi->num_devices;
 895         int this = 1;
 896         int last = 0;
 897         int peer_fd = -1;
 898         struct pollfd accept_poll_fd = {
 899                 .fd = spc->prg_fd,
 900                 .events = POLLIN,
 901                 .revents = 0,
 902         };
 903         struct pollfd write_poll_fd = {
 904                 .events = POLLOUT,
 905                 .revents = 0,
 906         };
 907         struct sockaddr_un peer;
 908         socklen_t peer_size = sizeof(peer);
 909
 910         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 911         if (perr)
 912                 goto out;
 913
 914         uuid_unparse(spc->fi->fsid, fsid);
 915
 916         for (i = 0; i < ndev; ++i) {
 917                 sp = &spc->progress[i];
 918                 sp_last = &spc->progress[i + ndev];
 919                 sp_shared = &spc->shared_progress[i];
 920                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 921                                                 sp_shared->scrub_args.devid;
 922                 sp->fd = sp_last->fd = spc->fdmnt;
 923                 sp->stats.t_start = sp_last->stats.t_start =
 924                                                 sp_shared->stats.t_start;
 925                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 926                 sp->skip = sp_last->skip = sp_shared->skip;
 927                 sp->stats.finished = sp_last->stats.finished =
 928                                                 sp_shared->stats.finished;
 929         }
 930
 931         while (1) {
 932                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 933                 if (ret == -1) {
 934                         ret = -errno;
 935                         goto out;
 936                 }
 937                 if (ret)
 938                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 939                                          &peer_size);
 940                 gettimeofday(&tv, NULL);
 941                 this = (this + 1)%2;
 942                 last = (last + 1)%2;
 943                 for (i = 0; i < ndev; ++i) {
 944                         sp = &spc->progress[this * ndev + i];
 945                         sp_last = &spc->progress[last * ndev + i];
 946                         sp_shared = &spc->shared_progress[i];
 947                         if (sp->stats.finished)
 948                                 continue;
 949                         progress_one_dev(sp);
 950                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 951                         if (!sp->ret)
 952                                 continue;
 953                         if (sp->ioctl_errno != ENOTCONN &&
 954                             sp->ioctl_errno != ENODEV) {
 955                                 ret = -sp->ioctl_errno;
 956                                 goto out;
 957                         }
 958                         /*
 959                          * scrub finished or device removed, check the
 960                          * finished flag. if unset, just use the last
 961                          * result we got for the current write and go
 962                          * on. flag should be set on next cycle, then.
 963                          */
 964                         perr = pthread_setcancelstate(
 965                                         PTHREAD_CANCEL_DISABLE, &old);
 966                         if (perr)
 967                                 goto out;
 968                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 969                         if (perr)
 970                                 goto out;
 971                         if (!sp_shared->stats.finished) {
 972                                 perr = pthread_mutex_unlock(
 973                                                 &sp_shared->progress_mutex);
 974                                 if (perr)
 975                                         goto out;
 976                                 perr = pthread_setcancelstate(
 977                                                 PTHREAD_CANCEL_ENABLE, &old);
 978                                 if (perr)
 979                                         goto out;
 980                                 memcpy(sp, sp_last, sizeof(*sp));
 981                                 continue;
 982                         }
 983                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 984                         if (perr)
 985                                 goto out;
 986                         perr = pthread_setcancelstate(
 987                                         PTHREAD_CANCEL_ENABLE, &old);
 988                         if (perr)
 989                                 goto out;
 990                         memcpy(sp, sp_shared, sizeof(*sp));
 991                         memcpy(sp_last, sp_shared, sizeof(*sp));
 992                 }
 993                 if (peer_fd != -1) {
 994                         write_poll_fd.fd = peer_fd;
 995                         ret = poll(&write_poll_fd, 1, 0);
 996                         if (ret == -1) {
 997                                 ret = -errno;
 998                                 goto out;
 999                         }
1000                         if (ret) {
1001                                 ret = scrub_write_file(
1002                                         peer_fd, fsid,
1003                                         &spc->progress[this * ndev], ndev);
1004                                 if (ret)
1005                                         goto out;
1006                         }
1007                         close(peer_fd);
1008                         peer_fd = -1;
1009                 }
1010                 if (!spc->do_record)
1011                         continue;
1012                 ret = scrub_write_progress(spc->write_mutex, fsid,
1013                                            &spc->progress[this * ndev], ndev);
1014                 if (ret)
1015                         goto out;
1016         }
1017 out:
1018         if (peer_fd != -1)
1019                 close(peer_fd);
1020         if (perr)
1021                 ret = -perr;
1022         return ERR_PTR(ret);
1023 }
1024
1025 static struct scrub_file_record *last_dev_scrub(
1026                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1027 {
1028         int i;
1029
1030         if (!past_scrubs || IS_ERR(past_scrubs))
1031                 return NULL;
1032
1033         for (i = 0; past_scrubs[i]; ++i)
1034                 if (past_scrubs[i]->devid == devid)
1035                         return past_scrubs[i];
1036
1037         return NULL;
1038 }
1039
1040 static int mkdir_p(char *path)
1041 {
1042         int i;
1043         int ret;
1044
1045         for (i = 1; i < strlen(path); ++i) {
1046                 if (path[i] != '/')
1047                         continue;
1048                 path[i] = '\0';
1049                 ret = mkdir(path, 0777);
1050                 if (ret && errno != EEXIST)
1051                         return -errno;
1052                 path[i] = '/';
1053         }
1054
1055         return 0;
1056 }
1057
1058 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1059                                   struct btrfs_ioctl_dev_info_args *di_args,
1060                                   struct scrub_file_record **past_scrubs)
1061 {
1062         int i;
1063
1064         if (!fi_args || !di_args || !past_scrubs)
1065                 return 0;
1066
1067         for (i = 0; i < fi_args->num_devices; i++) {
1068                 struct scrub_file_record *sfr =
1069                         last_dev_scrub(past_scrubs, di_args[i].devid);
1070
1071                 if (!sfr)
1072                         continue;
1073                 if (!(sfr->stats.finished || sfr->stats.canceled))
1074                         return 1;
1075         }
1076         return 0;
1077 }
1078
1079 static int is_scrub_running_in_kernel(int fd,
1080                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1081 {
1082         struct scrub_progress sp;
1083         int i;
1084         int ret;
1085
1086         for (i = 0; i < max_devices; i++) {
1087                 memset(&sp, 0, sizeof(sp));
1088                 sp.scrub_args.devid = di_args[i].devid;
1089                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1090                 if (!ret)
1091                         return 1;
1092         }
1093
1094         return 0;
1095 }
1096
1097 static const char * const cmd_scrub_start_usage[];
1098 static const char * const cmd_scrub_resume_usage[];
1099
1100 static int scrub_start(int argc, char **argv, int resume)
1101 {
1102         int fdmnt;
1103         int prg_fd = -1;
1104         int fdres = -1;
1105         int ret;
1106         pid_t pid;
1107         int c;
1108         int i;
1109         int err = 0;
1110         int e_uncorrectable = 0;
1111         int e_correctable = 0;
1112         int print_raw = 0;
1113         char *path;
1114         int do_background = 1;
1115         int do_wait = 0;
1116         int do_print = 0;
1117         int do_quiet = 0;
1118         int do_record = 1;
1119         int readonly = 0;
1120         int do_stats_per_dev = 0;
1121         int ioprio_class = IOPRIO_CLASS_IDLE;
1122         int ioprio_classdata = 0;
1123         int n_start = 0;
1124         int n_skip = 0;
1125         int n_resume = 0;
1126         struct btrfs_ioctl_fs_info_args fi_args;
1127         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1128         struct scrub_progress *sp = NULL;
1129         struct scrub_fs_stat fs_stat;
1130         struct timeval tv;
1131         struct sockaddr_un addr = {
1132                 .sun_family = AF_UNIX,
1133         };
1134         pthread_t *t_devs = NULL;
1135         pthread_t t_prog;
1136         struct scrub_file_record **past_scrubs = NULL;
1137         struct scrub_file_record *last_scrub = NULL;
1138         char *datafile = strdup(SCRUB_DATA_FILE);
1139         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1140         char sock_path[PATH_MAX] = "";
1141         struct scrub_progress_cycle spc;
1142         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1143         void *terr;
1144         u64 devid;
1145         DIR *dirstream = NULL;
1146         int force = 0;
1147         int nothing_to_resume = 0;
1148
1149         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1150                 switch (c) {
1151                 case 'B':
1152                         do_background = 0;
1153                         do_wait = 1;
1154                         do_print = 1;
1155                         break;
1156                 case 'd':
1157                         do_stats_per_dev = 1;
1158                         break;
1159                 case 'q':
1160                         do_quiet = 1;
1161                         break;
1162                 case 'r':
1163                         readonly = 1;
1164                         break;
1165                 case 'R':
1166                         print_raw = 1;
1167                         break;
1168                 case 'c':
1169                         ioprio_class = (int)strtol(optarg, NULL, 10);
1170                         break;
1171                 case 'n':
1172                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1173                         break;
1174                 case 'f':
1175                         force = 1;
1176                         break;
1177                 case '?':
1178                 default:
1179                         usage(resume ? cmd_scrub_resume_usage :
1180                                                 cmd_scrub_start_usage);
1181                 }
1182         }
1183
1184         /* try to catch most error cases before forking */
1185
1186         if (check_argc_exact(argc - optind, 1)) {
1187                 usage(resume ? cmd_scrub_resume_usage :
1188                                         cmd_scrub_start_usage);
1189         }
1190
1191         spc.progress = NULL;
1192         if (do_quiet && do_print)
1193                 do_print = 0;
1194
1195         if (mkdir_p(datafile)) {
1196                 warning_on(!do_quiet,
1197     "cannot create scrub data file, mkdir %s failed: %s. Status recording disabled",
1198                         datafile, strerror(errno));
1199                 do_record = 0;
1200         }
1201         free(datafile);
1202
1203         path = argv[optind];
1204
1205         fdmnt = open_path_or_dev_mnt(path, &dirstream, !do_quiet);
1206         if (fdmnt < 0)
1207                 return 1;
1208
1209         ret = get_fs_info(path, &fi_args, &di_args);
1210         if (ret) {
1211                 error_on(!do_quiet,
1212                         "getting dev info for scrub failed: %s",
1213                          strerror(-ret));
1214                 err = 1;
1215                 goto out;
1216         }
1217         if (!fi_args.num_devices) {
1218                 error_on(!do_quiet, "no devices found");
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                 warning_on(!do_quiet, "failed to open status file: %s",
1227                         strerror(-fdres));
1228         } else if (fdres >= 0) {
1229                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1230                 if (IS_ERR(past_scrubs))
1231                         warning_on(!do_quiet, "failed to read status file: %s",
1232                                 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                 error_on(!do_quiet,
1256                         "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'",
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                 error_on(!do_quiet, "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                         error_on(!do_quiet, "pthread_mutex_init failed: %s",
1279                                 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                         error("scrub already running");
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                 warning_on(!do_quiet,
1348    "failed to open the progress status socket at %s: %s. Progress cannot be queried",
1349                         sock_path[0] ? sock_path :
1350                         SCRUB_PROGRESS_SOCKET_PATH, 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                         warning_on(!do_quiet,
1365    "failed to write the progress status file: %s. Status recording disabled",
1366                                 strerror(-ret));
1367                         do_record = 0;
1368                 }
1369         }
1370
1371         if (do_background) {
1372                 pid = fork();
1373                 if (pid == -1) {
1374                         error_on(!do_quiet, "cannot scrub, fork failed: %s",
1375                                 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                                 error_on(!do_quiet, "wait failed (ret=%d): %s",
1394                                         ret, strerror(errno));
1395                                 err = 1;
1396                                 goto out;
1397                         }
1398                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1399                                 error_on(!do_quiet, "scrub process failed");
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                         error("creating scrub_one_dev[%llu] thread failed: %s",
1426                                 devid, strerror(ret));
1427                         err = 1;
1428                         goto out;
1429                 }
1430         }
1431
1432         spc.fdmnt = fdmnt;
1433         spc.prg_fd = prg_fd;
1434         spc.do_record = do_record;
1435         spc.write_mutex = &spc_write_mutex;
1436         spc.shared_progress = sp;
1437         spc.fi = &fi_args;
1438         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1439         if (ret) {
1440                 if (do_print)
1441                         error("creating progress thread failed: %s",
1442                                 strerror(ret));
1443                 err = 1;
1444                 goto out;
1445         }
1446
1447         err = 0;
1448         for (i = 0; i < fi_args.num_devices; ++i) {
1449                 if (sp[i].skip)
1450                         continue;
1451                 devid = di_args[i].devid;
1452                 ret = pthread_join(t_devs[i], NULL);
1453                 if (ret) {
1454                         if (do_print)
1455                           error("pthread_join failed for scrub_one_dev[%llu]: %s",
1456                                 devid, strerror(ret));
1457                         ++err;
1458                         continue;
1459                 }
1460                 if (sp[i].ret) {
1461                         switch (sp[i].ioctl_errno) {
1462                         case ENODEV:
1463                                 if (do_print)
1464                                         warning("device %lld not present",
1465                                                 devid);
1466                                 continue;
1467                         case ECANCELED:
1468                                 ++err;
1469                                 break;
1470                         default:
1471                                 if (do_print)
1472                 error("scrubbing %s failed for device id %lld: ret=%d, errno=%d (%s)",
1473                                         path, devid,
1474                                         sp[i].ret, sp[i].ioctl_errno,
1475                                         strerror(sp[i].ioctl_errno));
1476                                 ++err;
1477                                 continue;
1478                         }
1479                 }
1480                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1481                         e_uncorrectable++;
1482                 if (sp[i].scrub_args.progress.corrected_errors > 0
1483                     || sp[i].scrub_args.progress.unverified_errors > 0)
1484                         e_correctable++;
1485         }
1486
1487         if (do_print) {
1488                 const char *append = "done";
1489                 if (!do_stats_per_dev)
1490                         init_fs_stat(&fs_stat);
1491                 for (i = 0; i < fi_args.num_devices; ++i) {
1492                         if (do_stats_per_dev) {
1493                                 print_scrub_dev(&di_args[i],
1494                                                 &sp[i].scrub_args.progress,
1495                                                 print_raw,
1496                                                 sp[i].ret ? "canceled" : "done",
1497                                                 &sp[i].stats);
1498                         } else {
1499                                 if (sp[i].ret)
1500                                         append = "canceled";
1501                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1502                                                 &sp[i].stats, &fs_stat);
1503                         }
1504                 }
1505                 if (!do_stats_per_dev) {
1506                         printf("scrub %s for %s\n", append, fsid);
1507                         print_fs_stat(&fs_stat, print_raw);
1508                 }
1509         }
1510
1511         ret = pthread_cancel(t_prog);
1512         if (!ret)
1513                 ret = pthread_join(t_prog, &terr);
1514
1515         /* check for errors from the handling of the progress thread */
1516         if (do_print && ret) {
1517                 error("progress thread handling failed: %s",
1518                         strerror(ret));
1519         }
1520
1521         /* check for errors returned from the progress thread itself */
1522         if (do_print && terr && terr != PTHREAD_CANCELED)
1523                 error("recording progress failed: %s",
1524                         strerror(-PTR_ERR(terr)));
1525
1526         if (do_record) {
1527                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1528                                            fi_args.num_devices);
1529                 if (ret && do_print)
1530                         error("failed to record the result: %s",
1531                                 strerror(-ret));
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                 error_on(!do_quiet, "there are uncorrectable errors");
1555                 return 3;
1556         }
1557         if (e_correctable)
1558                 warning_on(!do_quiet,
1559                         "errors detected during scrubbing, corrected");
1560
1561         return 0;
1562 }
1563
1564 static const char * const cmd_scrub_start_usage[] = {
1565         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1566         "Start a new scrub. If a scrub is already running, the new one fails.",
1567         "",
1568         "-B     do not background",
1569         "-d     stats per device (-B only)",
1570         "-q     be quiet",
1571         "-r     read only mode",
1572         "-R     raw print mode, print full data instead of summary",
1573         "-c     set ioprio class (see ionice(1) manpage)",
1574         "-n     set ioprio classdata (see ionice(1) manpage)",
1575         "-f     force starting new scrub even if a scrub is already running",
1576         "       this is useful when scrub stats record file is damaged",
1577         NULL
1578 };
1579
1580 static int cmd_scrub_start(int argc, char **argv)
1581 {
1582         return scrub_start(argc, argv, 0);
1583 }
1584
1585 static const char * const cmd_scrub_cancel_usage[] = {
1586         "btrfs scrub cancel <path>|<device>",
1587         "Cancel a running scrub",
1588         NULL
1589 };
1590
1591 static int cmd_scrub_cancel(int argc, char **argv)
1592 {
1593         char *path;
1594         int ret;
1595         int fdmnt = -1;
1596         DIR *dirstream = NULL;
1597
1598         clean_args_no_options(argc, argv, cmd_scrub_cancel_usage);
1599
1600         if (check_argc_exact(argc - optind, 1))
1601                 usage(cmd_scrub_cancel_usage);
1602
1603         path = argv[optind];
1604
1605         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1606         if (fdmnt < 0) {
1607                 ret = 1;
1608                 goto out;
1609         }
1610
1611         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1612
1613         if (ret < 0) {
1614                 error("scrub cancel failed on %s: %s", path,
1615                         errno == ENOTCONN ? "not running" : strerror(errno));
1616                 if (errno == ENOTCONN)
1617                         ret = 2;
1618                 else
1619                         ret = 1;
1620                 goto out;
1621         }
1622
1623         ret = 0;
1624         printf("scrub cancelled\n");
1625
1626 out:
1627         close_file_or_dir(fdmnt, dirstream);
1628         return ret;
1629 }
1630
1631 static const char * const cmd_scrub_resume_usage[] = {
1632         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1633         "Resume previously canceled or interrupted scrub",
1634         "",
1635         "-B     do not background",
1636         "-d     stats per device (-B only)",
1637         "-q     be quiet",
1638         "-r     read only mode",
1639         "-R     raw print mode, print full data instead of summary",
1640         "-c     set ioprio class (see ionice(1) manpage)",
1641         "-n     set ioprio classdata (see ionice(1) manpage)",
1642         NULL
1643 };
1644
1645 static int cmd_scrub_resume(int argc, char **argv)
1646 {
1647         return scrub_start(argc, argv, 1);
1648 }
1649
1650 static const char * const cmd_scrub_status_usage[] = {
1651         "btrfs scrub status [-dR] <path>|<device>",
1652         "Show status of running or finished scrub",
1653         "",
1654         "-d     stats per device",
1655         "-R     print raw stats",
1656         NULL
1657 };
1658
1659 static int cmd_scrub_status(int argc, char **argv)
1660 {
1661         char *path;
1662         struct btrfs_ioctl_fs_info_args fi_args;
1663         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1664         struct scrub_file_record **past_scrubs = NULL;
1665         struct scrub_file_record *last_scrub;
1666         struct scrub_fs_stat fs_stat;
1667         struct sockaddr_un addr = {
1668                 .sun_family = AF_UNIX,
1669         };
1670         int in_progress;
1671         int ret;
1672         int i;
1673         int fdmnt;
1674         int print_raw = 0;
1675         int do_stats_per_dev = 0;
1676         int c;
1677         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1678         int fdres = -1;
1679         int err = 0;
1680         DIR *dirstream = NULL;
1681
1682         while ((c = getopt(argc, argv, "dR")) != -1) {
1683                 switch (c) {
1684                 case 'd':
1685                         do_stats_per_dev = 1;
1686                         break;
1687                 case 'R':
1688                         print_raw = 1;
1689                         break;
1690                 case '?':
1691                 default:
1692                         usage(cmd_scrub_status_usage);
1693                 }
1694         }
1695
1696         if (check_argc_exact(argc - optind, 1))
1697                 usage(cmd_scrub_status_usage);
1698
1699         path = argv[optind];
1700
1701         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1702         if (fdmnt < 0)
1703                 return 1;
1704
1705         ret = get_fs_info(path, &fi_args, &di_args);
1706         if (ret) {
1707                 error("getting dev info for scrub failed: %s",
1708                         strerror(-ret));
1709                 err = 1;
1710                 goto out;
1711         }
1712         if (!fi_args.num_devices) {
1713                 error("no devices found");
1714                 err = 1;
1715                 goto out;
1716         }
1717
1718         uuid_unparse(fi_args.fsid, fsid);
1719
1720         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1721         if (fdres == -1) {
1722                 error("failed to create socket to receive progress information: %s",
1723                         strerror(errno));
1724                 err = 1;
1725                 goto out;
1726         }
1727         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1728                         NULL, addr.sun_path, sizeof(addr.sun_path));
1729         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1730         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1731         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1732         if (ret == -1) {
1733                 close(fdres);
1734                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1735                 if (fdres < 0 && fdres != -ENOENT) {
1736                         warning("failed to open status file: %s",
1737                                 strerror(-fdres));
1738                         err = 1;
1739                         goto out;
1740                 }
1741         }
1742
1743         if (fdres >= 0) {
1744                 past_scrubs = scrub_read_file(fdres, 1);
1745                 if (IS_ERR(past_scrubs))
1746                         warning("failed to read status: %s",
1747                                 strerror(-PTR_ERR(past_scrubs)));
1748         }
1749         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1750
1751         printf("scrub status for %s\n", fsid);
1752
1753         if (do_stats_per_dev) {
1754                 for (i = 0; i < fi_args.num_devices; ++i) {
1755                         last_scrub = last_dev_scrub(past_scrubs,
1756                                                         di_args[i].devid);
1757                         if (!last_scrub) {
1758                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1759                                                 NULL, NULL);
1760                                 continue;
1761                         }
1762                         last_scrub->stats.in_progress = in_progress;
1763                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1764                                         last_scrub->stats.finished ?
1765                                                         "history" : "status",
1766                                         &last_scrub->stats);
1767                 }
1768         } else {
1769                 init_fs_stat(&fs_stat);
1770                 fs_stat.s.in_progress = in_progress;
1771                 for (i = 0; i < fi_args.num_devices; ++i) {
1772                         last_scrub = last_dev_scrub(past_scrubs,
1773                                                         di_args[i].devid);
1774                         if (!last_scrub)
1775                                 continue;
1776                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1777                                         &fs_stat);
1778                 }
1779                 print_fs_stat(&fs_stat, print_raw);
1780         }
1781
1782 out:
1783         free_history(past_scrubs);
1784         free(di_args);
1785         if (fdres > -1)
1786                 close(fdres);
1787         close_file_or_dir(fdmnt, dirstream);
1788
1789         return !!err;
1790 }
1791
1792 static const char scrub_cmd_group_info[] =
1793 "verify checksums of data and metadata";
1794
1795 const struct cmd_group scrub_cmd_group = {
1796         scrub_cmd_group_usage, scrub_cmd_group_info, {
1797                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1798                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1799                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1800                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1801                 NULL_CMD_STRUCT
1802         }
1803 };
1804
1805 int cmd_scrub(int argc, char **argv)
1806 {
1807         return handle_command_group(&scrub_cmd_group, argc, argv);
1808 }