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