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 <sys/ioctl.h>
  20 #include <sys/wait.h>
  21 #include <sys/stat.h>
  22 #include <sys/types.h>
  23 #include <sys/socket.h>
  24 #include <sys/un.h>
  25 #include <poll.h>
  26 #include <sys/file.h>
  27 #include <uuid/uuid.h>
  28 #include <fcntl.h>
  29 #include <unistd.h>
  30 #include <pthread.h>
  31 #include <ctype.h>
  32 #include <signal.h>
  33 #include <stdarg.h>
  34
  35 #include "ctree.h"
  36 #include "ioctl.h"
  37 #include "utils.h"
  38 #include "volumes.h"
  39 #include "disk-io.h"
  40
  41 #include "commands.h"
  42
  43 static const char * const scrub_cmd_group_usage[] = {
  44         "btrfs scrub <command> [options] <path>|<device>",
  45         NULL
  46 };
  47
  48 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  49 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  50 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  51 #define SCRUB_FILE_VERSION "1"
  52
  53 struct scrub_stats {
  54         time_t t_start;
  55         time_t t_resumed;
  56         u64 duration;
  57         u64 finished;
  58         u64 canceled;
  59 };
  60
  61 struct scrub_progress {
  62         struct btrfs_ioctl_scrub_args scrub_args;
  63         int fd;
  64         int ret;
  65         int skip;
  66         struct scrub_stats stats;
  67         struct scrub_file_record *resumed;
  68         int ioctl_errno;
  69         pthread_mutex_t progress_mutex;
  70 };
  71
  72 struct scrub_file_record {
  73         u8 fsid[BTRFS_FSID_SIZE];
  74         u64 devid;
  75         struct scrub_stats stats;
  76         struct btrfs_scrub_progress p;
  77 };
  78
  79 struct scrub_progress_cycle {
  80         int fdmnt;
  81         int prg_fd;
  82         int do_record;
  83         struct btrfs_ioctl_fs_info_args *fi;
  84         struct scrub_progress *progress;
  85         struct scrub_progress *shared_progress;
  86         pthread_mutex_t *write_mutex;
  87 };
  88
  89 struct scrub_fs_stat {
  90         struct btrfs_scrub_progress p;
  91         struct scrub_stats s;
  92         int i;
  93 };
  94
  95 static void print_scrub_full(struct btrfs_scrub_progress *sp)
  96 {
  97         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
  98         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
  99         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 100         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 101         printf("\tread_errors: %lld\n", sp->read_errors);
 102         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 103         printf("\tverify_errors: %lld\n", sp->verify_errors);
 104         printf("\tno_csum: %lld\n", sp->no_csum);
 105         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 106         printf("\tsuper_errors: %lld\n", sp->super_errors);
 107         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 108         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 109         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 110         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 111         printf("\tlast_physical: %lld\n", sp->last_physical);
 112 }
 113
 114 #define ERR(test, ...) do {                     \
 115         if (test)                               \
 116                 fprintf(stderr, __VA_ARGS__);   \
 117 } while (0)
 118
 119 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 120         if (test)                               \
 121                 printf(" %s=%llu", desc, test); \
 122 } while (0)
 123
 124 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 125 {
 126         u64 err_cnt;
 127         u64 err_cnt2;
 128         char *bytes;
 129
 130         err_cnt = p->read_errors +
 131                         p->csum_errors +
 132                         p->verify_errors +
 133                         p->super_errors;
 134
 135         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 136
 137         if (p->malloc_errors)
 138                 printf("*** WARNING: memory allocation failed while scrubbing. "
 139                        "results may be inaccurate\n");
 140         bytes = pretty_sizes(p->data_bytes_scrubbed + p->tree_bytes_scrubbed);
 141         printf("\ttotal bytes scrubbed: %s with %llu errors\n", bytes,
 142                 max(err_cnt, err_cnt2));
 143         free(bytes);
 144         if (err_cnt || err_cnt2) {
 145                 printf("\terror details:");
 146                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 147                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 148                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 149                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 150                 printf("\n");
 151                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 152                         "unverified errors: %llu\n", p->corrected_errors,
 153                         p->uncorrectable_errors, p->unverified_errors);
 154         }
 155 }
 156
 157 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 158         fs_stat->p.name += p->name;             \
 159 } while (0)
 160
 161 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 162 do {                                            \
 163         if (fs_stat->s.name > ss->name) {       \
 164                 fs_stat->s.name = ss->name;     \
 165         }                                       \
 166 } while (0)
 167
 168 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 169 do {                                                            \
 170         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 171                 fs_stat->s.name = ss->name;                     \
 172         }                                                       \
 173 } while (0)
 174
 175 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 176 do {                                                                    \
 177         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 178                 (fs_stat)->s.name = (ss)->name;                         \
 179         }                                                               \
 180 } while (0)
 181
 182 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 183                                 struct scrub_stats *ss,
 184                                 struct scrub_fs_stat *fs_stat)
 185 {
 186         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 187         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 188         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 189         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 190         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 191         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 192         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 193         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 194         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 195         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 196         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 197         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 198         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 199         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 200         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 201         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 202         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 203         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 204         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 205 }
 206
 207 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 208 {
 209         memset(fs_stat, 0, sizeof(*fs_stat));
 210         fs_stat->s.finished = 1;
 211 }
 212
 213 static void _print_scrub_ss(struct scrub_stats *ss)
 214 {
 215         char t[4096];
 216         struct tm tm;
 217
 218         if (!ss || !ss->t_start) {
 219                 printf("\tno stats available\n");
 220                 return;
 221         }
 222         if (ss->t_resumed) {
 223                 localtime_r(&ss->t_resumed, &tm);
 224                 strftime(t, sizeof(t), "%c", &tm);
 225                 t[sizeof(t) - 1] = '\0';
 226                 printf("\tscrub resumed at %s", t);
 227         } else {
 228                 localtime_r(&ss->t_start, &tm);
 229                 strftime(t, sizeof(t), "%c", &tm);
 230                 t[sizeof(t) - 1] = '\0';
 231                 printf("\tscrub started at %s", t);
 232         }
 233         if (ss->finished && !ss->canceled) {
 234                 printf(" and finished after %llu seconds\n",
 235                        ss->duration);
 236         } else if (ss->canceled) {
 237                 printf(" and was aborted after %llu seconds\n",
 238                        ss->duration);
 239         } else {
 240                 printf(", running for %llu seconds\n", ss->duration);
 241         }
 242 }
 243
 244 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 245                                 struct btrfs_scrub_progress *p, int raw,
 246                                 const char *append, struct scrub_stats *ss)
 247 {
 248         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 249                append ? append : "");
 250
 251         _print_scrub_ss(ss);
 252
 253         if (p) {
 254                 if (raw)
 255                         print_scrub_full(p);
 256                 else
 257                         print_scrub_summary(p);
 258         }
 259 }
 260
 261 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 262 {
 263         _print_scrub_ss(&fs_stat->s);
 264
 265         if (raw)
 266                 print_scrub_full(&fs_stat->p);
 267         else
 268                 print_scrub_summary(&fs_stat->p);
 269 }
 270
 271 static void free_history(struct scrub_file_record **last_scrubs)
 272 {
 273         struct scrub_file_record **l = last_scrubs;
 274         if (!l)
 275                 return;
 276         while (*l)
 277                 free(*l++);
 278         free(last_scrubs);
 279 }
 280
 281 /*
 282  * cancels a running scrub and makes the master process record the current
 283  * progress status before exiting.
 284  */
 285 static int cancel_fd = -1;
 286 static void scrub_sigint_record_progress(int signal)
 287 {
 288         ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 289 }
 290
 291 static int scrub_handle_sigint_parent(void)
 292 {
 293         struct sigaction sa = {
 294                 .sa_handler = SIG_IGN,
 295                 .sa_flags = SA_RESTART,
 296         };
 297
 298         return sigaction(SIGINT, &sa, NULL);
 299 }
 300
 301 static int scrub_handle_sigint_child(int fd)
 302 {
 303         struct sigaction sa = {
 304                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 305         };
 306
 307         cancel_fd = fd;
 308         return sigaction(SIGINT, &sa, NULL);
 309 }
 310
 311 static int scrub_datafile(const char *fn_base, const char *fn_local,
 312                                 const char *fn_tmp, char *datafile, int size)
 313 {
 314         int ret;
 315         int end = size - 2;
 316
 317         datafile[end + 1] = '\0';
 318         strncpy(datafile, fn_base, end);
 319         ret = strlen(datafile);
 320
 321         if (ret + 1 > end)
 322                 return -EOVERFLOW;
 323
 324         datafile[ret] = '.';
 325         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 326         ret = strlen(datafile);
 327
 328         if (ret + 1 > end)
 329                 return -EOVERFLOW;
 330
 331         if (fn_tmp) {
 332                 datafile[ret] = '_';
 333                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 334                 ret = strlen(datafile);
 335
 336                 if (ret > end)
 337                         return -EOVERFLOW;
 338         }
 339
 340         return 0;
 341 }
 342
 343 static int scrub_open_file(const char *datafile, int m)
 344 {
 345         int fd;
 346         int ret;
 347
 348         fd = open(datafile, m, 0600);
 349         if (fd < 0)
 350                 return -errno;
 351
 352         ret = flock(fd, LOCK_EX|LOCK_NB);
 353         if (ret) {
 354                 ret = errno;
 355                 close(fd);
 356                 return -ret;
 357         }
 358
 359         return fd;
 360 }
 361
 362 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 363 {
 364         int ret;
 365         char datafile[BTRFS_PATH_NAME_MAX + 1];
 366         ret = scrub_datafile(fn_base, fn_local, NULL,
 367                                 datafile, sizeof(datafile));
 368         if (ret < 0)
 369                 return ret;
 370         return scrub_open_file(datafile, O_RDONLY);
 371 }
 372
 373 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 374                                 const char *tmp)
 375 {
 376         int ret;
 377         char datafile[BTRFS_PATH_NAME_MAX + 1];
 378         ret = scrub_datafile(fn_base, fn_local, tmp,
 379                                 datafile, sizeof(datafile));
 380         if (ret < 0)
 381                 return ret;
 382         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 383 }
 384
 385 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 386                                 const char *tmp)
 387 {
 388         int ret;
 389         char datafile_old[BTRFS_PATH_NAME_MAX + 1];
 390         char datafile_new[BTRFS_PATH_NAME_MAX + 1];
 391         ret = scrub_datafile(fn_base, fn_local, tmp,
 392                                 datafile_old, sizeof(datafile_old));
 393         if (ret < 0)
 394                 return ret;
 395         ret = scrub_datafile(fn_base, fn_local, NULL,
 396                                 datafile_new, sizeof(datafile_new));
 397         if (ret < 0)
 398                 return ret;
 399         ret = rename(datafile_old, datafile_new);
 400         return ret ? -errno : 0;
 401 }
 402
 403 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 404         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 405 }
 406
 407 /*
 408  * returns 0 if the key did not match (nothing was read)
 409  *         1 if the key did match (success)
 410  *        -1 if the key did match and an error occured
 411  */
 412 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 413                         const char *key, u64 *dest)
 414 {
 415         int j;
 416
 417         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 418                 *i += len - 1;
 419                 if (buf[*i] != ':')
 420                         return -1;
 421                 *i += 1;
 422                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 423                         ;
 424                 if (*i + j >= avail)
 425                         return -1;
 426                 *dest = atoll(&buf[*i]);
 427                 *i += j;
 428                 return 1;
 429         }
 430
 431         return 0;
 432 }
 433
 434 #define _SCRUB_INVALID do {                                             \
 435         if (report_errors)                                              \
 436                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 437                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 438                         lineno, i, state, 20 > avail ? avail : 20,      \
 439                         l + i,  __FILE__, __LINE__);                    \
 440         goto skip;                                                      \
 441 } while (0)
 442
 443 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 444 {
 445         int avail = 0;
 446         int old_avail = 0;
 447         char l[16 * 1024];
 448         int state = 0;
 449         int curr = -1;
 450         int i = 0;
 451         int j;
 452         int ret;
 453         int eof = 0;
 454         int lineno = 0;
 455         u64 version;
 456         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 457         struct scrub_file_record **p = NULL;
 458
 459         if (fd < 0)
 460                 return ERR_PTR(-EINVAL);
 461
 462 again:
 463         old_avail = avail - i;
 464         BUG_ON(old_avail < 0);
 465         if (old_avail)
 466                 memmove(l, l + i, old_avail);
 467         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 468         if (avail == 0)
 469                 eof = 1;
 470         if (avail == 0 && old_avail == 0) {
 471                 if (curr >= 0 &&
 472                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 473                         p[curr] = NULL;
 474                 } else if (curr == -1) {
 475                         p = ERR_PTR(-ENODATA);
 476                 }
 477                 return p;
 478         }
 479         if (avail == -1)
 480                 return ERR_PTR(-errno);
 481         avail += old_avail;
 482
 483         i = 0;
 484         while (i < avail) {
 485                 switch (state) {
 486                 case 0: /* start of file */
 487                         ret = scrub_kvread(&i,
 488                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 489                                 SCRUB_FILE_VERSION_PREFIX, &version);
 490                         if (ret != 1)
 491                                 _SCRUB_INVALID;
 492                         if (version != atoll(SCRUB_FILE_VERSION))
 493                                 return ERR_PTR(-ENOTSUP);
 494                         state = 6;
 495                         continue;
 496                 case 1: /* start of line, alloc */
 497                         /*
 498                          * this state makes sure we have a complete line in
 499                          * further processing, so we don't need wrap-tracking
 500                          * everywhere.
 501                          */
 502                         if (!eof && !memchr(l + i, '\n', avail - i))
 503                                 goto again;
 504                         ++lineno;
 505                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 506                                                 BTRFS_FSID_SIZE) == 0) {
 507                                 state = 2;
 508                                 continue;
 509                         }
 510                         ++curr;
 511                         p = realloc(p, (curr + 2) * sizeof(*p));
 512                         if (p)
 513                                 p[curr] = malloc(sizeof(**p));
 514                         if (!p || !p[curr])
 515                                 return ERR_PTR(-errno);
 516                         memset(p[curr], 0, sizeof(**p));
 517                         p[curr + 1] = NULL;
 518                         ++state;
 519                         /* fall through */
 520                 case 2: /* start of line, skip space */
 521                         while (isspace(l[i]) && i < avail) {
 522                                 if (l[i] == '\n')
 523                                         ++lineno;
 524                                 ++i;
 525                         }
 526                         if (i >= avail ||
 527                             (!eof && !memchr(l + i, '\n', avail - i)))
 528                                 goto again;
 529                         ++state;
 530                         /* fall through */
 531                 case 3: /* read fsid */
 532                         if (i == avail)
 533                                 continue;
 534                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 535                                 ;
 536                         if (i + j + 1 >= avail)
 537                                 _SCRUB_INVALID;
 538                         if (j != 36)
 539                                 _SCRUB_INVALID;
 540                         l[i + j] = '\0';
 541                         ret = uuid_parse(l + i, p[curr]->fsid);
 542                         if (ret)
 543                                 _SCRUB_INVALID;
 544                         i += j + 1;
 545                         ++state;
 546                         /* fall through */
 547                 case 4: /* read dev id */
 548                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 549                                 ;
 550                         if (j == 0 || i + j + 1 >= avail)
 551                                 _SCRUB_INVALID;
 552                         p[curr]->devid = atoll(&l[i]);
 553                         i += j + 1;
 554                         ++state;
 555                         /* fall through */
 556                 case 5: /* read key/value pair */
 557                         ret = 0;
 558                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 559                                         &p[curr]->p);
 560                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 561                                         &p[curr]->p);
 562                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 563                                         &p[curr]->p);
 564                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 565                                         &p[curr]->p);
 566                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 567                                         &p[curr]->p);
 568                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 569                                         &p[curr]->p);
 570                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 571                                         &p[curr]->p);
 572                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 573                                         &p[curr]->p);
 574                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 575                                         &p[curr]->p);
 576                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 577                                         &p[curr]->p);
 578                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 579                                         &p[curr]->p);
 580                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 581                                         &p[curr]->p);
 582                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 583                                         &p[curr]->p);
 584                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 585                                         &p[curr]->p);
 586                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 587                                         &p[curr]->p);
 588                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 589                                         &p[curr]->stats);
 590                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 591                                         (u64 *)&p[curr]->stats);
 592                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 593                                         (u64 *)&p[curr]->stats);
 594                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 595                                         (u64 *)&p[curr]->stats);
 596                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 597                                         &p[curr]->stats);
 598                         if (ret != 1)
 599                                 _SCRUB_INVALID;
 600                         ++state;
 601                         /* fall through */
 602                 case 6: /* after number */
 603                         if (l[i] == '|')
 604                                 state = 5;
 605                         else if (l[i] == '\n')
 606                                 state = 1;
 607                         else
 608                                 _SCRUB_INVALID;
 609                         ++i;
 610                         continue;
 611                 case 99: /* skip rest of line */
 612 skip:
 613                         state = 99;
 614                         do {
 615                                 ++i;
 616                                 if (l[i - 1] == '\n') {
 617                                         state = 1;
 618                                         break;
 619                                 }
 620                         } while (i < avail);
 621                         continue;
 622                 }
 623                 BUG();
 624         }
 625         goto again;
 626 }
 627
 628 static int scrub_write_buf(int fd, const void *data, int len)
 629 {
 630         int ret;
 631         ret = write(fd, data, len);
 632         return ret - len;
 633 }
 634
 635 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 636                                 __attribute__ ((format (printf, 4, 5)));
 637 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 638 {
 639         int ret;
 640         va_list args;
 641
 642         va_start(args, fmt);
 643         ret = vsnprintf(buf, max, fmt, args);
 644         va_end(args);
 645         if (ret >= max)
 646                 return ret - max;
 647         return scrub_write_buf(fd, buf, ret);
 648 }
 649
 650 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 651                         data->resumed->p.name + data->scrub_args.progress.name
 652
 653 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 654                                                   struct scrub_progress *dest)
 655 {
 656         if (!data->resumed || data->skip)
 657                 return data;
 658
 659         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 660         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 661         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 662         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 663         _SCRUB_SUM(dest, data, read_errors);
 664         _SCRUB_SUM(dest, data, csum_errors);
 665         _SCRUB_SUM(dest, data, verify_errors);
 666         _SCRUB_SUM(dest, data, no_csum);
 667         _SCRUB_SUM(dest, data, csum_discards);
 668         _SCRUB_SUM(dest, data, super_errors);
 669         _SCRUB_SUM(dest, data, malloc_errors);
 670         _SCRUB_SUM(dest, data, uncorrectable_errors);
 671         _SCRUB_SUM(dest, data, corrected_errors);
 672         _SCRUB_SUM(dest, data, last_physical);
 673         dest->stats.canceled = data->stats.canceled;
 674         dest->stats.finished = data->stats.finished;
 675         dest->stats.t_resumed = data->stats.t_start;
 676         dest->stats.t_start = data->resumed->stats.t_start;
 677         dest->stats.duration = data->resumed->stats.duration +
 678                                                         data->stats.duration;
 679         dest->scrub_args.devid = data->scrub_args.devid;
 680         return dest;
 681 }
 682
 683 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 684         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 685                         use->scrub_args.progress.name)
 686
 687 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 688         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 689                         use->stats.name)
 690
 691 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 692 {
 693         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 694 }
 695
 696 static int scrub_write_file(int fd, const char *fsid,
 697                                 struct scrub_progress *data, int n)
 698 {
 699         int ret = 0;
 700         int i;
 701         char buf[1024];
 702         struct scrub_progress local;
 703         struct scrub_progress *use;
 704
 705         if (n < 1)
 706                 return -EINVAL;
 707
 708         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 709         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 710                                 SCRUB_FILE_VERSION "\n",
 711                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 712                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 713         if (ret)
 714                 return -EOVERFLOW;
 715
 716         for (i = 0; i < n; ++i) {
 717                 use = scrub_resumed_stats(&data[i], &local);
 718                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 719                     scrub_write_buf(fd, ":", 1) ||
 720                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 721                                         use->scrub_args.devid) ||
 722                     scrub_write_buf(fd, buf, ret) ||
 723                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 724                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 725                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 726                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 727                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 728                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 729                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 730                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 731                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 732                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 733                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 734                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 735                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 736                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 737                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 738                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 739                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 740                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 741                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 742                     scrub_write_buf(fd, "\n", 1)) {
 743                         return -EOVERFLOW;
 744                 }
 745         }
 746
 747         return 0;
 748 }
 749
 750 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 751                                 struct scrub_progress *data, int n)
 752 {
 753         int ret;
 754         int err;
 755         int fd = 0;
 756         int old;
 757
 758         ret = pthread_mutex_lock(m);
 759         if (ret) {
 760                 err = -errno;
 761                 goto out;
 762         }
 763
 764         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 765         if (ret) {
 766                 err = -ret;
 767                 goto out;
 768         }
 769
 770         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 771         if (fd < 0) {
 772                 err = fd;
 773                 goto out;
 774         }
 775         err = scrub_write_file(fd, fsid, data, n);
 776         if (err)
 777                 goto out;
 778         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 779         if (err)
 780                 goto out;
 781
 782 out:
 783         if (fd > 0) {
 784                 ret = close(fd);
 785                 if (ret)
 786                         err = -errno;
 787         }
 788
 789         ret = pthread_mutex_unlock(m);
 790         if (ret && !err)
 791                 err = -ret;
 792
 793         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 794         if (ret && !err)
 795                 err = -ret;
 796
 797         return err;
 798 }
 799
 800 static void *scrub_one_dev(void *ctx)
 801 {
 802         struct scrub_progress *sp = ctx;
 803         int ret;
 804         struct timeval tv;
 805
 806         sp->stats.canceled = 0;
 807         sp->stats.duration = 0;
 808         sp->stats.finished = 0;
 809
 810         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 811         gettimeofday(&tv, NULL);
 812         sp->ret = ret;
 813         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 814         sp->stats.canceled = !!ret;
 815         sp->ioctl_errno = errno;
 816         ret = pthread_mutex_lock(&sp->progress_mutex);
 817         if (ret)
 818                 return ERR_PTR(-ret);
 819         sp->stats.finished = 1;
 820         ret = pthread_mutex_unlock(&sp->progress_mutex);
 821         if (ret)
 822                 return ERR_PTR(-ret);
 823
 824         return NULL;
 825 }
 826
 827 static void *progress_one_dev(void *ctx)
 828 {
 829         struct scrub_progress *sp = ctx;
 830
 831         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 832         sp->ioctl_errno = errno;
 833
 834         return NULL;
 835 }
 836
 837 static void *scrub_progress_cycle(void *ctx)
 838 {
 839         int ret;
 840         int old;
 841         int i;
 842         char fsid[37];
 843         struct scrub_progress *sp;
 844         struct scrub_progress *sp_last;
 845         struct scrub_progress *sp_shared;
 846         struct timeval tv;
 847         struct scrub_progress_cycle *spc = ctx;
 848         int ndev = spc->fi->num_devices;
 849         int this = 1;
 850         int last = 0;
 851         int peer_fd = -1;
 852         struct pollfd accept_poll_fd = {
 853                 .fd = spc->prg_fd,
 854                 .events = POLLIN,
 855                 .revents = 0,
 856         };
 857         struct pollfd write_poll_fd = {
 858                 .events = POLLOUT,
 859                 .revents = 0,
 860         };
 861         struct sockaddr_un peer;
 862         socklen_t peer_size = sizeof(peer);
 863
 864         ret = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 865         if (ret)
 866                 return ERR_PTR(-ret);
 867
 868         uuid_unparse(spc->fi->fsid, fsid);
 869
 870         for (i = 0; i < ndev; ++i) {
 871                 sp = &spc->progress[i];
 872                 sp_last = &spc->progress[i + ndev];
 873                 sp_shared = &spc->shared_progress[i];
 874                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 875                                                 sp_shared->scrub_args.devid;
 876                 sp->fd = sp_last->fd = spc->fdmnt;
 877                 sp->stats.t_start = sp_last->stats.t_start =
 878                                                 sp_shared->stats.t_start;
 879                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 880                 sp->skip = sp_last->skip = sp_shared->skip;
 881                 sp->stats.finished = sp_last->stats.finished =
 882                                                 sp_shared->stats.finished;
 883         }
 884
 885         while (1) {
 886                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 887                 if (ret == -1)
 888                         return ERR_PTR(-errno);
 889                 if (ret)
 890                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 891                                          &peer_size);
 892                 gettimeofday(&tv, NULL);
 893                 this = (this + 1)%2;
 894                 last = (last + 1)%2;
 895                 for (i = 0; i < ndev; ++i) {
 896                         sp = &spc->progress[this * ndev + i];
 897                         sp_last = &spc->progress[last * ndev + i];
 898                         sp_shared = &spc->shared_progress[i];
 899                         if (sp->stats.finished)
 900                                 continue;
 901                         progress_one_dev(sp);
 902                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 903                         if (!sp->ret)
 904                                 continue;
 905                         if (sp->ioctl_errno != ENOTCONN &&
 906                             sp->ioctl_errno != ENODEV)
 907                                 return ERR_PTR(-sp->ioctl_errno);
 908                         /*
 909                          * scrub finished or device removed, check the
 910                          * finished flag. if unset, just use the last
 911                          * result we got for the current write and go
 912                          * on. flag should be set on next cycle, then.
 913                          */
 914                         ret = pthread_mutex_lock(&sp_shared->progress_mutex);
 915                         if (ret)
 916                                 return ERR_PTR(-ret);
 917                         if (!sp_shared->stats.finished) {
 918                                 ret = pthread_mutex_unlock(
 919                                                 &sp_shared->progress_mutex);
 920                                 if (ret)
 921                                         return ERR_PTR(-ret);
 922                                 memcpy(sp, sp_last, sizeof(*sp));
 923                                 continue;
 924                         }
 925                         ret = pthread_mutex_unlock(&sp_shared->progress_mutex);
 926                         if (ret)
 927                                 return ERR_PTR(-ret);
 928                         memcpy(sp, sp_shared, sizeof(*sp));
 929                         memcpy(sp_last, sp_shared, sizeof(*sp));
 930                 }
 931                 if (peer_fd != -1) {
 932                         write_poll_fd.fd = peer_fd;
 933                         ret = poll(&write_poll_fd, 1, 0);
 934                         if (ret == -1)
 935                                 return ERR_PTR(-errno);
 936                         if (ret) {
 937                                 ret = scrub_write_file(
 938                                         peer_fd, fsid,
 939                                         &spc->progress[this * ndev], ndev);
 940                                 if (ret)
 941                                         return ERR_PTR(ret);
 942                         }
 943                         close(peer_fd);
 944                         peer_fd = -1;
 945                 }
 946                 if (!spc->do_record)
 947                         continue;
 948                 ret = scrub_write_progress(spc->write_mutex, fsid,
 949                                            &spc->progress[this * ndev], ndev);
 950                 if (ret)
 951                         return ERR_PTR(ret);
 952         }
 953 }
 954
 955 static struct scrub_file_record *last_dev_scrub(
 956                 struct scrub_file_record *const *const past_scrubs, u64 devid)
 957 {
 958         int i;
 959
 960         if (!past_scrubs || IS_ERR(past_scrubs))
 961                 return NULL;
 962
 963         for (i = 0; past_scrubs[i]; ++i)
 964                 if (past_scrubs[i]->devid == devid)
 965                         return past_scrubs[i];
 966
 967         return NULL;
 968 }
 969
 970 static int scrub_device_info(int fd, u64 devid,
 971                              struct btrfs_ioctl_dev_info_args *di_args)
 972 {
 973         int ret;
 974
 975         di_args->devid = devid;
 976         memset(&di_args->uuid, '\0', sizeof(di_args->uuid));
 977
 978         ret = ioctl(fd, BTRFS_IOC_DEV_INFO, di_args);
 979         return ret ? -errno : 0;
 980 }
 981
 982 static int scrub_fs_info(int fd, char *path,
 983                                 struct btrfs_ioctl_fs_info_args *fi_args,
 984                                 struct btrfs_ioctl_dev_info_args **di_ret)
 985 {
 986         int ret = 0;
 987         int ndevs = 0;
 988         int i = 1;
 989         struct btrfs_fs_devices *fs_devices_mnt = NULL;
 990         struct btrfs_ioctl_dev_info_args *di_args;
 991         char mp[BTRFS_PATH_NAME_MAX + 1];
 992
 993         memset(fi_args, 0, sizeof(*fi_args));
 994
 995         ret = ioctl(fd, BTRFS_IOC_FS_INFO, fi_args);
 996         if (ret && errno == EINVAL) {
 997                 /* path is no mounted btrfs. try if it's a device */
 998                 ret = check_mounted_where(fd, path, mp, sizeof(mp),
 999                                                 &fs_devices_mnt);
1000                 if (!ret)
1001                         return -EINVAL;
1002                 if (ret < 0)
1003                         return ret;
1004                 fi_args->num_devices = 1;
1005                 fi_args->max_id = fs_devices_mnt->latest_devid;
1006                 i = fs_devices_mnt->latest_devid;
1007                 memcpy(fi_args->fsid, fs_devices_mnt->fsid, BTRFS_FSID_SIZE);
1008                 close(fd);
1009                 fd = open_file_or_dir(mp);
1010                 if (fd < 0)
1011                         return -errno;
1012         } else if (ret) {
1013                 return -errno;
1014         }
1015
1016         if (!fi_args->num_devices)
1017                 return 0;
1018
1019         di_args = *di_ret = malloc(fi_args->num_devices * sizeof(*di_args));
1020         if (!di_args)
1021                 return -errno;
1022
1023         for (; i <= fi_args->max_id; ++i) {
1024                 BUG_ON(ndevs >= fi_args->num_devices);
1025                 ret = scrub_device_info(fd, i, &di_args[ndevs]);
1026                 if (ret == -ENODEV)
1027                         continue;
1028                 if (ret)
1029                         return ret;
1030                 ++ndevs;
1031         }
1032
1033         BUG_ON(ndevs == 0);
1034
1035         return 0;
1036 }
1037
1038 int mkdir_p(char *path)
1039 {
1040         int i;
1041         int ret;
1042
1043         for (i = 1; i < strlen(path); ++i) {
1044                 if (path[i] != '/')
1045                         continue;
1046                 path[i] = '\0';
1047                 ret = mkdir(path, 0777);
1048                 if (ret && errno != EEXIST)
1049                         return 1;
1050                 path[i] = '/';
1051         }
1052
1053         return 0;
1054 }
1055
1056 static const char * const cmd_scrub_start_usage[];
1057 static const char * const cmd_scrub_resume_usage[];
1058
1059 static int scrub_start(int argc, char **argv, int resume)
1060 {
1061         int fdmnt;
1062         int prg_fd = -1;
1063         int fdres = -1;
1064         int ret;
1065         pid_t pid;
1066         int c;
1067         int i;
1068         int err = 0;
1069         int e_uncorrectable = 0;
1070         int e_correctable = 0;
1071         int print_raw = 0;
1072         char *path;
1073         int do_background = 1;
1074         int do_wait = 0;
1075         int do_print = 0;
1076         int do_quiet = 0;
1077         int do_record = 1;
1078         int readonly = 0;
1079         int do_stats_per_dev = 0;
1080         int n_start = 0;
1081         int n_skip = 0;
1082         int n_resume = 0;
1083         struct btrfs_ioctl_fs_info_args fi_args;
1084         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1085         struct scrub_progress *sp = NULL;
1086         struct scrub_fs_stat fs_stat;
1087         struct timeval tv;
1088         struct sockaddr_un addr = {
1089                 .sun_family = AF_UNIX,
1090         };
1091         pthread_t *t_devs = NULL;
1092         pthread_t t_prog;
1093         pthread_attr_t t_attr;
1094         struct scrub_file_record **past_scrubs = NULL;
1095         struct scrub_file_record *last_scrub = NULL;
1096         char *datafile = strdup(SCRUB_DATA_FILE);
1097         char fsid[37];
1098         char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1099         struct scrub_progress_cycle spc;
1100         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1101         void *terr;
1102         u64 devid;
1103
1104         optind = 1;
1105         while ((c = getopt(argc, argv, "BdqrR")) != -1) {
1106                 switch (c) {
1107                 case 'B':
1108                         do_background = 0;
1109                         do_wait = 1;
1110                         do_print = 1;
1111                         break;
1112                 case 'd':
1113                         do_stats_per_dev = 1;
1114                         break;
1115                 case 'q':
1116                         do_quiet = 1;
1117                         break;
1118                 case 'r':
1119                         readonly = 1;
1120                         break;
1121                 case 'R':
1122                         print_raw = 1;
1123                         break;
1124                 case '?':
1125                 default:
1126                         usage(resume ? cmd_scrub_resume_usage :
1127                                                 cmd_scrub_start_usage);
1128                 }
1129         }
1130
1131         /* try to catch most error cases before forking */
1132
1133         if (check_argc_exact(argc - optind, 1)) {
1134                 usage(resume ? cmd_scrub_resume_usage :
1135                                         cmd_scrub_start_usage);
1136         }
1137
1138         spc.progress = NULL;
1139         if (do_quiet && do_print)
1140                 do_print = 0;
1141
1142         if (mkdir_p(datafile)) {
1143                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1144                                "file, mkdir %s failed: %s. Status recording "
1145                                "disabled\n", datafile, strerror(errno));
1146                 do_record = 0;
1147         }
1148         free(datafile);
1149
1150         path = argv[optind];
1151
1152         fdmnt = open_file_or_dir(path);
1153         if (fdmnt < 0) {
1154                 ERR(!do_quiet, "ERROR: can't access '%s'\n", path);
1155                 return 12;
1156         }
1157
1158         ret = scrub_fs_info(fdmnt, path, &fi_args, &di_args);
1159         if (ret) {
1160                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1161                     "%s\n", strerror(-ret));
1162                 err = 1;
1163                 goto out;
1164         }
1165         if (!fi_args.num_devices) {
1166                 ERR(!do_quiet, "ERROR: no devices found\n");
1167                 err = 1;
1168                 goto out;
1169         }
1170
1171         uuid_unparse(fi_args.fsid, fsid);
1172         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1173         if (fdres < 0 && fdres != -ENOENT) {
1174                 ERR(!do_quiet, "WARNING: failed to open status file: "
1175                     "%s\n", strerror(-fdres));
1176         } else if (fdres >= 0) {
1177                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1178                 if (IS_ERR(past_scrubs))
1179                         ERR(!do_quiet, "WARNING: failed to read status file: "
1180                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1181                 close(fdres);
1182         }
1183
1184         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1185         sp = calloc(fi_args.num_devices, sizeof(*sp));
1186         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1187
1188         if (!t_devs || !sp || !spc.progress) {
1189                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1190                 err = 1;
1191                 goto out;
1192         }
1193
1194         ret = pthread_attr_init(&t_attr);
1195         if (ret) {
1196                 ERR(!do_quiet, "ERROR: pthread_attr_init failed: %s\n",
1197                     strerror(ret));
1198                 err = 1;
1199                 goto out;
1200         }
1201
1202         for (i = 0; i < fi_args.num_devices; ++i) {
1203                 devid = di_args[i].devid;
1204                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1205                 if (ret) {
1206                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1207                             "%s\n", strerror(ret));
1208                         err = 1;
1209                         goto out;
1210                 }
1211                 last_scrub = last_dev_scrub(past_scrubs, devid);
1212                 sp[i].scrub_args.devid = devid;
1213                 sp[i].fd = fdmnt;
1214                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1215                                              !last_scrub->stats.finished)) {
1216                         ++n_resume;
1217                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1218                         sp[i].resumed = last_scrub;
1219                 } else if (resume) {
1220                         ++n_skip;
1221                         sp[i].skip = 1;
1222                         sp[i].resumed = last_scrub;
1223                         continue;
1224                 } else {
1225                         ++n_start;
1226                         sp[i].scrub_args.start = 0ll;
1227                         sp[i].resumed = NULL;
1228                 }
1229                 sp[i].skip = 0;
1230                 sp[i].scrub_args.end = (u64)-1ll;
1231                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1232         }
1233
1234         if (!n_start && !n_resume) {
1235                 if (!do_quiet)
1236                         printf("scrub: nothing to resume for %s, fsid %s\n",
1237                                path, fsid);
1238                 err = 0;
1239                 goto out;
1240         }
1241
1242         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1243         while (ret != -1) {
1244                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1245                                         sock_path, sizeof(sock_path));
1246                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1247                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1248                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1249                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1250                 if (ret != -1 || errno != EADDRINUSE)
1251                         break;
1252                 /*
1253                  * bind failed with EADDRINUSE. so let's see if anyone answers
1254                  * when we make a call to the socket ...
1255                  */
1256                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1257                 if (!ret || errno != ECONNREFUSED) {
1258                         /* ... yes, so scrub must be running. error out */
1259                         fprintf(stderr, "ERROR: scrub already running\n");
1260                         close(prg_fd);
1261                         goto out;
1262                 }
1263                 /*
1264                  * ... no, this means someone left us alone with an unused
1265                  * socket in the file system. remove it and try again.
1266                  */
1267                 ret = unlink(sock_path);
1268         }
1269         if (ret != -1)
1270                 ret = listen(prg_fd, 100);
1271         if (ret == -1) {
1272                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1273                     "socket at %s: %s. Progress cannot be queried\n",
1274                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1275                     strerror(errno));
1276                 if (prg_fd != -1) {
1277                         close(prg_fd);
1278                         prg_fd = -1;
1279                         if (sock_path[0])
1280                                 unlink(sock_path);
1281                 }
1282         }
1283
1284         if (do_record) {
1285                 /* write all-zero progress file for a start */
1286                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1287                                            fi_args.num_devices);
1288                 if (ret) {
1289                         ERR(!do_quiet, "WARNING: failed to write the progress "
1290                             "status file: %s. Status recording disabled\n",
1291                             strerror(-ret));
1292                         do_record = 0;
1293                 }
1294         }
1295
1296         if (do_background) {
1297                 pid = fork();
1298                 if (pid == -1) {
1299                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1300                                         "%s\n", strerror(errno));
1301                         err = 1;
1302                         goto out;
1303                 }
1304
1305                 if (pid) {
1306                         int stat;
1307                         scrub_handle_sigint_parent();
1308                         if (!do_quiet)
1309                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1310                                        n_start ? "started" : "resumed",
1311                                        path, fsid, pid);
1312                         if (!do_wait) {
1313                                 err = 0;
1314                                 goto out;
1315                         }
1316                         ret = wait(&stat);
1317                         if (ret != pid) {
1318                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1319                                     "%s\n", ret, strerror(errno));
1320                                 err = 1;
1321                                 goto out;
1322                         }
1323                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1324                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1325                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1326                                 goto out;
1327                         }
1328                         err = 0;
1329                         goto out;
1330                 }
1331         }
1332
1333         scrub_handle_sigint_child(fdmnt);
1334
1335         for (i = 0; i < fi_args.num_devices; ++i) {
1336                 if (sp[i].skip) {
1337                         sp[i].scrub_args.progress = sp[i].resumed->p;
1338                         sp[i].stats = sp[i].resumed->stats;
1339                         sp[i].ret = 0;
1340                         sp[i].stats.finished = 1;
1341                         continue;
1342                 }
1343                 devid = di_args[i].devid;
1344                 gettimeofday(&tv, NULL);
1345                 sp[i].stats.t_start = tv.tv_sec;
1346                 ret = pthread_create(&t_devs[i], &t_attr,
1347                                         scrub_one_dev, &sp[i]);
1348                 if (ret) {
1349                         if (do_print)
1350                                 fprintf(stderr, "ERROR: creating "
1351                                         "scrub_one_dev[%llu] thread failed: "
1352                                         "%s\n", devid, strerror(ret));
1353                         err = 1;
1354                         goto out;
1355                 }
1356         }
1357
1358         spc.fdmnt = fdmnt;
1359         spc.prg_fd = prg_fd;
1360         spc.do_record = do_record;
1361         spc.write_mutex = &spc_write_mutex;
1362         spc.shared_progress = sp;
1363         spc.fi = &fi_args;
1364         ret = pthread_create(&t_prog, &t_attr, scrub_progress_cycle, &spc);
1365         if (ret) {
1366                 if (do_print)
1367                         fprintf(stderr, "ERROR: creating progress thread "
1368                                 "failed: %s\n", strerror(ret));
1369                 err = 1;
1370                 goto out;
1371         }
1372
1373         err = 0;
1374         for (i = 0; i < fi_args.num_devices; ++i) {
1375                 if (sp[i].skip)
1376                         continue;
1377                 devid = di_args[i].devid;
1378                 ret = pthread_join(t_devs[i], NULL);
1379                 if (ret) {
1380                         if (do_print)
1381                                 fprintf(stderr, "ERROR: pthread_join failed "
1382                                         "for scrub_one_dev[%llu]: %s\n", devid,
1383                                         strerror(ret));
1384                         ++err;
1385                         continue;
1386                 }
1387                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1388                         if (do_print)
1389                                 fprintf(stderr, "WARNING: device %lld not "
1390                                         "present\n", devid);
1391                         continue;
1392                 }
1393                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1394                         ++err;
1395                 } else if (sp[i].ret) {
1396                         if (do_print)
1397                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1398                                         "for device id %lld (%s)\n", path,
1399                                         devid, strerror(sp[i].ioctl_errno));
1400                         ++err;
1401                         continue;
1402                 }
1403                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1404                         e_uncorrectable++;
1405                 if (sp[i].scrub_args.progress.corrected_errors > 0
1406                     || sp[i].scrub_args.progress.unverified_errors > 0)
1407                         e_correctable++;
1408         }
1409
1410         if (do_print) {
1411                 const char *append = "done";
1412                 if (!do_stats_per_dev)
1413                         init_fs_stat(&fs_stat);
1414                 for (i = 0; i < fi_args.num_devices; ++i) {
1415                         if (do_stats_per_dev) {
1416                                 print_scrub_dev(&di_args[i],
1417                                                 &sp[i].scrub_args.progress,
1418                                                 print_raw,
1419                                                 sp[i].ret ? "canceled" : "done",
1420                                                 &sp[i].stats);
1421                         } else {
1422                                 if (sp[i].ret)
1423                                         append = "canceled";
1424                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1425                                                 &sp[i].stats, &fs_stat);
1426                         }
1427                 }
1428                 if (!do_stats_per_dev) {
1429                         printf("scrub %s for %s\n", append, fsid);
1430                         print_fs_stat(&fs_stat, print_raw);
1431                 }
1432         }
1433
1434         ret = pthread_cancel(t_prog);
1435         if (!ret)
1436                 ret = pthread_join(t_prog, &terr);
1437         if (do_print && ret) {
1438                 fprintf(stderr, "ERROR: progress thead handling failed: %s\n",
1439                         strerror(ret));
1440         }
1441
1442         if (do_print && terr && terr != PTHREAD_CANCELED) {
1443                 fprintf(stderr, "ERROR: recording progress "
1444                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1445         }
1446
1447         if (do_record) {
1448                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1449                                            fi_args.num_devices);
1450                 if (ret && do_print) {
1451                         fprintf(stderr, "ERROR: failed to record the result: "
1452                                 "%s\n", strerror(-ret));
1453                 }
1454         }
1455
1456         scrub_handle_sigint_child(-1);
1457
1458 out:
1459         free_history(past_scrubs);
1460         free(di_args);
1461         free(t_devs);
1462         free(sp);
1463         free(spc.progress);
1464         if (prg_fd > -1) {
1465                 close(prg_fd);
1466                 if (sock_path[0])
1467                         unlink(sock_path);
1468         }
1469         close(fdmnt);
1470
1471         if (err)
1472                 return 1;
1473         if (e_correctable)
1474                 return 7;
1475         if (e_uncorrectable)
1476                 return 8;
1477         return 0;
1478 }
1479
1480 static const char * const cmd_scrub_start_usage[] = {
1481         "btrfs scrub start [-Bdqr] <path>|<device>",
1482         "Start a new scrub",
1483         "",
1484         "-B     do not background",
1485         "-d     stats per device (-B only)",
1486         "-q     be quiet",
1487         "-r     read only mode",
1488         NULL
1489 };
1490
1491 static int cmd_scrub_start(int argc, char **argv)
1492 {
1493         return scrub_start(argc, argv, 0);
1494 }
1495
1496 static const char * const cmd_scrub_cancel_usage[] = {
1497         "btrfs scrub cancel <path>|<device>",
1498         "Cancel a running scrub",
1499         NULL
1500 };
1501
1502 static int cmd_scrub_cancel(int argc, char **argv)
1503 {
1504         char *path;
1505         int ret;
1506         int fdmnt;
1507         int err;
1508         char mp[BTRFS_PATH_NAME_MAX + 1];
1509         struct btrfs_fs_devices *fs_devices_mnt = NULL;
1510
1511         if (check_argc_exact(argc, 2))
1512                 usage(cmd_scrub_cancel_usage);
1513
1514         path = argv[1];
1515
1516         fdmnt = open_file_or_dir(path);
1517         if (fdmnt < 0) {
1518                 fprintf(stderr, "ERROR: scrub cancel failed\n");
1519                 return 12;
1520         }
1521
1522 again:
1523         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1524         err = errno;
1525         close(fdmnt);
1526
1527         if (ret && err == EINVAL) {
1528                 /* path is no mounted btrfs. try if it's a device */
1529                 ret = check_mounted_where(fdmnt, path, mp, sizeof(mp),
1530                                           &fs_devices_mnt);
1531                 close(fdmnt);
1532                 if (ret) {
1533                         fdmnt = open_file_or_dir(mp);
1534                         if (fdmnt >= 0) {
1535                                 path = mp;
1536                                 goto again;
1537                         }
1538                 }
1539         }
1540
1541         if (ret) {
1542                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1543                         err == ENOTCONN ? "not running" : strerror(errno));
1544                 return 1;
1545         }
1546
1547         printf("scrub cancelled\n");
1548
1549         return 0;
1550 }
1551
1552 static const char * const cmd_scrub_resume_usage[] = {
1553         "btrfs scrub resume [-Bdqr] <path>|<device>",
1554         "Resume previously canceled or interrupted scrub",
1555         "",
1556         "-B     do not background",
1557         "-d     stats per device (-B only)",
1558         "-q     be quiet",
1559         "-r     read only mode",
1560         NULL
1561 };
1562
1563 static int cmd_scrub_resume(int argc, char **argv)
1564 {
1565         return scrub_start(argc, argv, 1);
1566 }
1567
1568 static const char * const cmd_scrub_status_usage[] = {
1569         "btrfs scrub status [-dR] <path>|<device>",
1570         "Show status of running or finished scrub",
1571         "",
1572         "-d     stats per device",
1573         "-R     print raw stats",
1574         NULL
1575 };
1576
1577 static int cmd_scrub_status(int argc, char **argv)
1578 {
1579         char *path;
1580         struct btrfs_ioctl_fs_info_args fi_args;
1581         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1582         struct scrub_file_record **past_scrubs = NULL;
1583         struct scrub_file_record *last_scrub;
1584         struct scrub_fs_stat fs_stat;
1585         struct sockaddr_un addr = {
1586                 .sun_family = AF_UNIX,
1587         };
1588         int ret;
1589         int fdmnt;
1590         int i;
1591         int print_raw = 0;
1592         int do_stats_per_dev = 0;
1593         char c;
1594         char fsid[37];
1595         int fdres = -1;
1596         int err = 0;
1597
1598         optind = 1;
1599         while ((c = getopt(argc, argv, "dR")) != -1) {
1600                 switch (c) {
1601                 case 'd':
1602                         do_stats_per_dev = 1;
1603                         break;
1604                 case 'R':
1605                         print_raw = 1;
1606                         break;
1607                 case '?':
1608                 default:
1609                         usage(cmd_scrub_status_usage);
1610                 }
1611         }
1612
1613         if (check_argc_exact(argc - optind, 1))
1614                 usage(cmd_scrub_status_usage);
1615
1616         path = argv[optind];
1617
1618         fdmnt = open_file_or_dir(path);
1619         if (fdmnt < 0) {
1620                 fprintf(stderr, "ERROR: can't access to '%s'\n", path);
1621                 return 12;
1622         }
1623
1624         ret = scrub_fs_info(fdmnt, path, &fi_args, &di_args);
1625         if (ret) {
1626                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1627                                 "%s\n", strerror(-ret));
1628                 err = 1;
1629                 goto out;
1630         }
1631         if (!fi_args.num_devices) {
1632                 fprintf(stderr, "ERROR: no devices found\n");
1633                 err = 1;
1634                 goto out;
1635         }
1636
1637         uuid_unparse(fi_args.fsid, fsid);
1638
1639         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1640         if (fdres == -1) {
1641                 fprintf(stderr, "ERROR: failed to create socket to "
1642                         "receive progress information: %s\n",
1643                         strerror(errno));
1644                 err = 1;
1645                 goto out;
1646         }
1647         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1648                         NULL, addr.sun_path, sizeof(addr.sun_path));
1649         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1650         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1651         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1652         if (ret == -1) {
1653                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1654                 if (fdres < 0 && fdres != -ENOENT) {
1655                         fprintf(stderr, "WARNING: failed to open status file: "
1656                                 "%s\n", strerror(-fdres));
1657                         err = 1;
1658                         goto out;
1659                 }
1660         }
1661
1662         if (fdres >= 0) {
1663                 past_scrubs = scrub_read_file(fdres, 1);
1664                 if (IS_ERR(past_scrubs))
1665                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1666                                 strerror(-PTR_ERR(past_scrubs)));
1667         }
1668
1669         printf("scrub status for %s\n", fsid);
1670
1671         if (do_stats_per_dev) {
1672                 for (i = 0; i < fi_args.num_devices; ++i) {
1673                         last_scrub = last_dev_scrub(past_scrubs,
1674                                                         di_args[i].devid);
1675                         if (!last_scrub) {
1676                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1677                                                 NULL, NULL);
1678                                 continue;
1679                         }
1680                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1681                                         last_scrub->stats.finished ?
1682                                                         "history" : "status",
1683                                         &last_scrub->stats);
1684                 }
1685         } else {
1686                 init_fs_stat(&fs_stat);
1687                 for (i = 0; i < fi_args.num_devices; ++i) {
1688                         last_scrub = last_dev_scrub(past_scrubs,
1689                                                         di_args[i].devid);
1690                         if (!last_scrub)
1691                                 continue;
1692                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1693                                         &fs_stat);
1694                 }
1695                 print_fs_stat(&fs_stat, print_raw);
1696         }
1697
1698 out:
1699         free_history(past_scrubs);
1700         free(di_args);
1701         close(fdmnt);
1702         if (fdres > -1)
1703                 close(fdres);
1704
1705         return err;
1706 }
1707
1708 const struct cmd_group scrub_cmd_group = {
1709         scrub_cmd_group_usage, NULL, {
1710                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1711                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1712                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1713                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1714                 { 0, 0, 0, 0, 0 }
1715         }
1716 };
1717
1718 int cmd_scrub(int argc, char **argv)
1719 {
1720         return handle_command_group(&scrub_cmd_group, argc, argv);
1721 }