cmds-scrub.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include "kerncompat.h"
  20
  21 #include <sys/ioctl.h>
  22 #include <sys/wait.h>
  23 #include <sys/stat.h>
  24 #include <sys/types.h>
  25 #include <sys/socket.h>
  26 #include <sys/un.h>
  27 #include <poll.h>
  28 #include <sys/file.h>
  29 #include <uuid/uuid.h>
  30 #include <fcntl.h>
  31 #include <unistd.h>
  32 #include <pthread.h>
  33 #include <ctype.h>
  34 #include <signal.h>
  35 #include <stdarg.h>
  36
  37 #include "ctree.h"
  38 #include "ioctl.h"
  39 #include "utils.h"
  40 #include "volumes.h"
  41 #include "disk-io.h"
  42
  43 #include "commands.h"
  44
  45 static const char * const scrub_cmd_group_usage[] = {
  46         "btrfs scrub <command> [options] <path>|<device>",
  47         NULL
  48 };
  49
  50 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  51 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  52 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  53 #define SCRUB_FILE_VERSION "1"
  54
  55 struct scrub_stats {
  56         time_t t_start;
  57         time_t t_resumed;
  58         u64 duration;
  59         u64 finished;
  60         u64 canceled;
  61 };
  62
  63 struct scrub_progress {
  64         struct btrfs_ioctl_scrub_args scrub_args;
  65         int fd;
  66         int ret;
  67         int skip;
  68         struct scrub_stats stats;
  69         struct scrub_file_record *resumed;
  70         int ioctl_errno;
  71         pthread_mutex_t progress_mutex;
  72 };
  73
  74 struct scrub_file_record {
  75         u8 fsid[BTRFS_FSID_SIZE];
  76         u64 devid;
  77         struct scrub_stats stats;
  78         struct btrfs_scrub_progress p;
  79 };
  80
  81 struct scrub_progress_cycle {
  82         int fdmnt;
  83         int prg_fd;
  84         int do_record;
  85         struct btrfs_ioctl_fs_info_args *fi;
  86         struct scrub_progress *progress;
  87         struct scrub_progress *shared_progress;
  88         pthread_mutex_t *write_mutex;
  89 };
  90
  91 struct scrub_fs_stat {
  92         struct btrfs_scrub_progress p;
  93         struct scrub_stats s;
  94         int i;
  95 };
  96
  97 static void print_scrub_full(struct btrfs_scrub_progress *sp)
  98 {
  99         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 100         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 101         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 102         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 103         printf("\tread_errors: %lld\n", sp->read_errors);
 104         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 105         printf("\tverify_errors: %lld\n", sp->verify_errors);
 106         printf("\tno_csum: %lld\n", sp->no_csum);
 107         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 108         printf("\tsuper_errors: %lld\n", sp->super_errors);
 109         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 110         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 111         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 112         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 113         printf("\tlast_physical: %lld\n", sp->last_physical);
 114 }
 115
 116 #define ERR(test, ...) do {                     \
 117         if (test)                               \
 118                 fprintf(stderr, __VA_ARGS__);   \
 119 } while (0)
 120
 121 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 122         if (test)                               \
 123                 printf(" %s=%llu", desc, test); \
 124 } while (0)
 125
 126 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 127 {
 128         u64 err_cnt;
 129         u64 err_cnt2;
 130         char *bytes;
 131
 132         err_cnt = p->read_errors +
 133                         p->csum_errors +
 134                         p->verify_errors +
 135                         p->super_errors;
 136
 137         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 138
 139         if (p->malloc_errors)
 140                 printf("*** WARNING: memory allocation failed while scrubbing. "
 141                        "results may be inaccurate\n");
 142         bytes = pretty_sizes(p->data_bytes_scrubbed + p->tree_bytes_scrubbed);
 143         printf("\ttotal bytes scrubbed: %s with %llu errors\n", bytes,
 144                 max(err_cnt, err_cnt2));
 145         free(bytes);
 146         if (err_cnt || err_cnt2) {
 147                 printf("\terror details:");
 148                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 149                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 150                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 151                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 152                 printf("\n");
 153                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 154                         "unverified errors: %llu\n", p->corrected_errors,
 155                         p->uncorrectable_errors, p->unverified_errors);
 156         }
 157 }
 158
 159 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 160         fs_stat->p.name += p->name;             \
 161 } while (0)
 162
 163 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 164 do {                                            \
 165         if (fs_stat->s.name > ss->name) {       \
 166                 fs_stat->s.name = ss->name;     \
 167         }                                       \
 168 } while (0)
 169
 170 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 171 do {                                                            \
 172         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 173                 fs_stat->s.name = ss->name;                     \
 174         }                                                       \
 175 } while (0)
 176
 177 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 178 do {                                                                    \
 179         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 180                 (fs_stat)->s.name = (ss)->name;                         \
 181         }                                                               \
 182 } while (0)
 183
 184 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 185                                 struct scrub_stats *ss,
 186                                 struct scrub_fs_stat *fs_stat)
 187 {
 188         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 189         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 190         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 191         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 192         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 193         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 194         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 195         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 196         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 197         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 198         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 199         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 200         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 201         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 202         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 203         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 204         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 205         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 206         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 207 }
 208
 209 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 210 {
 211         memset(fs_stat, 0, sizeof(*fs_stat));
 212         fs_stat->s.finished = 1;
 213 }
 214
 215 static void _print_scrub_ss(struct scrub_stats *ss)
 216 {
 217         char t[4096];
 218         struct tm tm;
 219
 220         if (!ss || !ss->t_start) {
 221                 printf("\tno stats available\n");
 222                 return;
 223         }
 224         if (ss->t_resumed) {
 225                 localtime_r(&ss->t_resumed, &tm);
 226                 strftime(t, sizeof(t), "%c", &tm);
 227                 t[sizeof(t) - 1] = '\0';
 228                 printf("\tscrub resumed at %s", t);
 229         } else {
 230                 localtime_r(&ss->t_start, &tm);
 231                 strftime(t, sizeof(t), "%c", &tm);
 232                 t[sizeof(t) - 1] = '\0';
 233                 printf("\tscrub started at %s", t);
 234         }
 235         if (ss->finished && !ss->canceled) {
 236                 printf(" and finished after %llu seconds\n",
 237                        ss->duration);
 238         } else if (ss->canceled) {
 239                 printf(" and was aborted after %llu seconds\n",
 240                        ss->duration);
 241         } else {
 242                 printf(", running for %llu seconds\n", ss->duration);
 243         }
 244 }
 245
 246 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 247                                 struct btrfs_scrub_progress *p, int raw,
 248                                 const char *append, struct scrub_stats *ss)
 249 {
 250         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 251                append ? append : "");
 252
 253         _print_scrub_ss(ss);
 254
 255         if (p) {
 256                 if (raw)
 257                         print_scrub_full(p);
 258                 else
 259                         print_scrub_summary(p);
 260         }
 261 }
 262
 263 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 264 {
 265         _print_scrub_ss(&fs_stat->s);
 266
 267         if (raw)
 268                 print_scrub_full(&fs_stat->p);
 269         else
 270                 print_scrub_summary(&fs_stat->p);
 271 }
 272
 273 static void free_history(struct scrub_file_record **last_scrubs)
 274 {
 275         struct scrub_file_record **l = last_scrubs;
 276         if (!l)
 277                 return;
 278         while (*l)
 279                 free(*l++);
 280         free(last_scrubs);
 281 }
 282
 283 /*
 284  * cancels a running scrub and makes the master process record the current
 285  * progress status before exiting.
 286  */
 287 static int cancel_fd = -1;
 288 static void scrub_sigint_record_progress(int signal)
 289 {
 290         ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 291 }
 292
 293 static int scrub_handle_sigint_parent(void)
 294 {
 295         struct sigaction sa = {
 296                 .sa_handler = SIG_IGN,
 297                 .sa_flags = SA_RESTART,
 298         };
 299
 300         return sigaction(SIGINT, &sa, NULL);
 301 }
 302
 303 static int scrub_handle_sigint_child(int fd)
 304 {
 305         struct sigaction sa = {
 306                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 307         };
 308
 309         cancel_fd = fd;
 310         return sigaction(SIGINT, &sa, NULL);
 311 }
 312
 313 static int scrub_datafile(const char *fn_base, const char *fn_local,
 314                                 const char *fn_tmp, char *datafile, int size)
 315 {
 316         int ret;
 317         int end = size - 2;
 318
 319         datafile[end + 1] = '\0';
 320         strncpy(datafile, fn_base, end);
 321         ret = strlen(datafile);
 322
 323         if (ret + 1 > end)
 324                 return -EOVERFLOW;
 325
 326         datafile[ret] = '.';
 327         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 328         ret = strlen(datafile);
 329
 330         if (ret + 1 > end)
 331                 return -EOVERFLOW;
 332
 333         if (fn_tmp) {
 334                 datafile[ret] = '_';
 335                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 336                 ret = strlen(datafile);
 337
 338                 if (ret > end)
 339                         return -EOVERFLOW;
 340         }
 341
 342         return 0;
 343 }
 344
 345 static int scrub_open_file(const char *datafile, int m)
 346 {
 347         int fd;
 348         int ret;
 349
 350         fd = open(datafile, m, 0600);
 351         if (fd < 0)
 352                 return -errno;
 353
 354         ret = flock(fd, LOCK_EX|LOCK_NB);
 355         if (ret) {
 356                 ret = errno;
 357                 close(fd);
 358                 return -ret;
 359         }
 360
 361         return fd;
 362 }
 363
 364 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 365 {
 366         int ret;
 367         char datafile[BTRFS_PATH_NAME_MAX + 1];
 368         ret = scrub_datafile(fn_base, fn_local, NULL,
 369                                 datafile, sizeof(datafile));
 370         if (ret < 0)
 371                 return ret;
 372         return scrub_open_file(datafile, O_RDONLY);
 373 }
 374
 375 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 376                                 const char *tmp)
 377 {
 378         int ret;
 379         char datafile[BTRFS_PATH_NAME_MAX + 1];
 380         ret = scrub_datafile(fn_base, fn_local, tmp,
 381                                 datafile, sizeof(datafile));
 382         if (ret < 0)
 383                 return ret;
 384         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 385 }
 386
 387 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 388                                 const char *tmp)
 389 {
 390         int ret;
 391         char datafile_old[BTRFS_PATH_NAME_MAX + 1];
 392         char datafile_new[BTRFS_PATH_NAME_MAX + 1];
 393         ret = scrub_datafile(fn_base, fn_local, tmp,
 394                                 datafile_old, sizeof(datafile_old));
 395         if (ret < 0)
 396                 return ret;
 397         ret = scrub_datafile(fn_base, fn_local, NULL,
 398                                 datafile_new, sizeof(datafile_new));
 399         if (ret < 0)
 400                 return ret;
 401         ret = rename(datafile_old, datafile_new);
 402         return ret ? -errno : 0;
 403 }
 404
 405 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 406         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 407 }
 408
 409 /*
 410  * returns 0 if the key did not match (nothing was read)
 411  *         1 if the key did match (success)
 412  *        -1 if the key did match and an error occured
 413  */
 414 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 415                         const char *key, u64 *dest)
 416 {
 417         int j;
 418
 419         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 420                 *i += len - 1;
 421                 if (buf[*i] != ':')
 422                         return -1;
 423                 *i += 1;
 424                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 425                         ;
 426                 if (*i + j >= avail)
 427                         return -1;
 428                 *dest = atoll(&buf[*i]);
 429                 *i += j;
 430                 return 1;
 431         }
 432
 433         return 0;
 434 }
 435
 436 #define _SCRUB_INVALID do {                                             \
 437         if (report_errors)                                              \
 438                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 439                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 440                         lineno, i, state, 20 > avail ? avail : 20,      \
 441                         l + i,  __FILE__, __LINE__);                    \
 442         goto skip;                                                      \
 443 } while (0)
 444
 445 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 446 {
 447         int avail = 0;
 448         int old_avail = 0;
 449         char l[16 * 1024];
 450         int state = 0;
 451         int curr = -1;
 452         int i = 0;
 453         int j;
 454         int ret;
 455         int eof = 0;
 456         int lineno = 0;
 457         u64 version;
 458         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 459         struct scrub_file_record **p = NULL;
 460
 461         if (fd < 0)
 462                 return ERR_PTR(-EINVAL);
 463
 464 again:
 465         old_avail = avail - i;
 466         BUG_ON(old_avail < 0);
 467         if (old_avail)
 468                 memmove(l, l + i, old_avail);
 469         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 470         if (avail == 0)
 471                 eof = 1;
 472         if (avail == 0 && old_avail == 0) {
 473                 if (curr >= 0 &&
 474                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 475                         p[curr] = NULL;
 476                 } else if (curr == -1) {
 477                         p = ERR_PTR(-ENODATA);
 478                 }
 479                 return p;
 480         }
 481         if (avail == -1)
 482                 return ERR_PTR(-errno);
 483         avail += old_avail;
 484
 485         i = 0;
 486         while (i < avail) {
 487                 switch (state) {
 488                 case 0: /* start of file */
 489                         ret = scrub_kvread(&i,
 490                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 491                                 SCRUB_FILE_VERSION_PREFIX, &version);
 492                         if (ret != 1)
 493                                 _SCRUB_INVALID;
 494                         if (version != atoll(SCRUB_FILE_VERSION))
 495                                 return ERR_PTR(-ENOTSUP);
 496                         state = 6;
 497                         continue;
 498                 case 1: /* start of line, alloc */
 499                         /*
 500                          * this state makes sure we have a complete line in
 501                          * further processing, so we don't need wrap-tracking
 502                          * everywhere.
 503                          */
 504                         if (!eof && !memchr(l + i, '\n', avail - i))
 505                                 goto again;
 506                         ++lineno;
 507                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 508                                                 BTRFS_FSID_SIZE) == 0) {
 509                                 state = 2;
 510                                 continue;
 511                         }
 512                         ++curr;
 513                         p = realloc(p, (curr + 2) * sizeof(*p));
 514                         if (p)
 515                                 p[curr] = malloc(sizeof(**p));
 516                         if (!p || !p[curr])
 517                                 return ERR_PTR(-errno);
 518                         memset(p[curr], 0, sizeof(**p));
 519                         p[curr + 1] = NULL;
 520                         ++state;
 521                         /* fall through */
 522                 case 2: /* start of line, skip space */
 523                         while (isspace(l[i]) && i < avail) {
 524                                 if (l[i] == '\n')
 525                                         ++lineno;
 526                                 ++i;
 527                         }
 528                         if (i >= avail ||
 529                             (!eof && !memchr(l + i, '\n', avail - i)))
 530                                 goto again;
 531                         ++state;
 532                         /* fall through */
 533                 case 3: /* read fsid */
 534                         if (i == avail)
 535                                 continue;
 536                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 537                                 ;
 538                         if (i + j + 1 >= avail)
 539                                 _SCRUB_INVALID;
 540                         if (j != 36)
 541                                 _SCRUB_INVALID;
 542                         l[i + j] = '\0';
 543                         ret = uuid_parse(l + i, p[curr]->fsid);
 544                         if (ret)
 545                                 _SCRUB_INVALID;
 546                         i += j + 1;
 547                         ++state;
 548                         /* fall through */
 549                 case 4: /* read dev id */
 550                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 551                                 ;
 552                         if (j == 0 || i + j + 1 >= avail)
 553                                 _SCRUB_INVALID;
 554                         p[curr]->devid = atoll(&l[i]);
 555                         i += j + 1;
 556                         ++state;
 557                         /* fall through */
 558                 case 5: /* read key/value pair */
 559                         ret = 0;
 560                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 561                                         &p[curr]->p);
 562                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 563                                         &p[curr]->p);
 564                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 565                                         &p[curr]->p);
 566                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 567                                         &p[curr]->p);
 568                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 569                                         &p[curr]->p);
 570                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 571                                         &p[curr]->p);
 572                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 573                                         &p[curr]->p);
 574                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 575                                         &p[curr]->p);
 576                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 577                                         &p[curr]->p);
 578                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 579                                         &p[curr]->p);
 580                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 581                                         &p[curr]->p);
 582                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 583                                         &p[curr]->p);
 584                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 585                                         &p[curr]->p);
 586                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 587                                         &p[curr]->p);
 588                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 589                                         &p[curr]->p);
 590                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 591                                         &p[curr]->stats);
 592                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 593                                         (u64 *)&p[curr]->stats);
 594                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 595                                         (u64 *)&p[curr]->stats);
 596                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 597                                         (u64 *)&p[curr]->stats);
 598                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 599                                         &p[curr]->stats);
 600                         if (ret != 1)
 601                                 _SCRUB_INVALID;
 602                         ++state;
 603                         /* fall through */
 604                 case 6: /* after number */
 605                         if (l[i] == '|')
 606                                 state = 5;
 607                         else if (l[i] == '\n')
 608                                 state = 1;
 609                         else
 610                                 _SCRUB_INVALID;
 611                         ++i;
 612                         continue;
 613                 case 99: /* skip rest of line */
 614 skip:
 615                         state = 99;
 616                         do {
 617                                 ++i;
 618                                 if (l[i - 1] == '\n') {
 619                                         state = 1;
 620                                         break;
 621                                 }
 622                         } while (i < avail);
 623                         continue;
 624                 }
 625                 BUG();
 626         }
 627         goto again;
 628 }
 629
 630 static int scrub_write_buf(int fd, const void *data, int len)
 631 {
 632         int ret;
 633         ret = write(fd, data, len);
 634         return ret - len;
 635 }
 636
 637 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 638                                 __attribute__ ((format (printf, 4, 5)));
 639 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 640 {
 641         int ret;
 642         va_list args;
 643
 644         va_start(args, fmt);
 645         ret = vsnprintf(buf, max, fmt, args);
 646         va_end(args);
 647         if (ret >= max)
 648                 return ret - max;
 649         return scrub_write_buf(fd, buf, ret);
 650 }
 651
 652 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 653                         data->resumed->p.name + data->scrub_args.progress.name
 654
 655 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 656                                                   struct scrub_progress *dest)
 657 {
 658         if (!data->resumed || data->skip)
 659                 return data;
 660
 661         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 662         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 663         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 664         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 665         _SCRUB_SUM(dest, data, read_errors);
 666         _SCRUB_SUM(dest, data, csum_errors);
 667         _SCRUB_SUM(dest, data, verify_errors);
 668         _SCRUB_SUM(dest, data, no_csum);
 669         _SCRUB_SUM(dest, data, csum_discards);
 670         _SCRUB_SUM(dest, data, super_errors);
 671         _SCRUB_SUM(dest, data, malloc_errors);
 672         _SCRUB_SUM(dest, data, uncorrectable_errors);
 673         _SCRUB_SUM(dest, data, corrected_errors);
 674         _SCRUB_SUM(dest, data, last_physical);
 675         dest->stats.canceled = data->stats.canceled;
 676         dest->stats.finished = data->stats.finished;
 677         dest->stats.t_resumed = data->stats.t_start;
 678         dest->stats.t_start = data->resumed->stats.t_start;
 679         dest->stats.duration = data->resumed->stats.duration +
 680                                                         data->stats.duration;
 681         dest->scrub_args.devid = data->scrub_args.devid;
 682         return dest;
 683 }
 684
 685 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 686         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 687                         use->scrub_args.progress.name)
 688
 689 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 690         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 691                         use->stats.name)
 692
 693 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 694 {
 695         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 696 }
 697
 698 static int scrub_write_file(int fd, const char *fsid,
 699                                 struct scrub_progress *data, int n)
 700 {
 701         int ret = 0;
 702         int i;
 703         char buf[1024];
 704         struct scrub_progress local;
 705         struct scrub_progress *use;
 706
 707         if (n < 1)
 708                 return -EINVAL;
 709
 710         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 711         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 712                                 SCRUB_FILE_VERSION "\n",
 713                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 714                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 715         if (ret)
 716                 return -EOVERFLOW;
 717
 718         for (i = 0; i < n; ++i) {
 719                 use = scrub_resumed_stats(&data[i], &local);
 720                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 721                     scrub_write_buf(fd, ":", 1) ||
 722                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 723                                         use->scrub_args.devid) ||
 724                     scrub_write_buf(fd, buf, ret) ||
 725                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 726                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 727                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 728                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 729                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 730                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 731                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 732                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 733                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 734                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 735                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 736                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 737                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 738                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 739                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 740                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 741                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 742                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 743                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 744                     scrub_write_buf(fd, "\n", 1)) {
 745                         return -EOVERFLOW;
 746                 }
 747         }
 748
 749         return 0;
 750 }
 751
 752 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 753                                 struct scrub_progress *data, int n)
 754 {
 755         int ret;
 756         int err;
 757         int fd = -1;
 758         int old;
 759
 760         ret = pthread_mutex_lock(m);
 761         if (ret) {
 762                 err = -errno;
 763                 goto out;
 764         }
 765
 766         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 767         if (ret) {
 768                 err = -ret;
 769                 goto out;
 770         }
 771
 772         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 773         if (fd < 0) {
 774                 err = fd;
 775                 goto out;
 776         }
 777         err = scrub_write_file(fd, fsid, data, n);
 778         if (err)
 779                 goto out;
 780         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 781         if (err)
 782                 goto out;
 783
 784 out:
 785         if (fd >= 0) {
 786                 ret = close(fd);
 787                 if (ret)
 788                         err = -errno;
 789         }
 790
 791         ret = pthread_mutex_unlock(m);
 792         if (ret && !err)
 793                 err = -ret;
 794
 795         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 796         if (ret && !err)
 797                 err = -ret;
 798
 799         return err;
 800 }
 801
 802 static void *scrub_one_dev(void *ctx)
 803 {
 804         struct scrub_progress *sp = ctx;
 805         int ret;
 806         struct timeval tv;
 807
 808         sp->stats.canceled = 0;
 809         sp->stats.duration = 0;
 810         sp->stats.finished = 0;
 811
 812         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 813         gettimeofday(&tv, NULL);
 814         sp->ret = ret;
 815         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 816         sp->stats.canceled = !!ret;
 817         sp->ioctl_errno = errno;
 818         ret = pthread_mutex_lock(&sp->progress_mutex);
 819         if (ret)
 820                 return ERR_PTR(-ret);
 821         sp->stats.finished = 1;
 822         ret = pthread_mutex_unlock(&sp->progress_mutex);
 823         if (ret)
 824                 return ERR_PTR(-ret);
 825
 826         return NULL;
 827 }
 828
 829 static void *progress_one_dev(void *ctx)
 830 {
 831         struct scrub_progress *sp = ctx;
 832
 833         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 834         sp->ioctl_errno = errno;
 835
 836         return NULL;
 837 }
 838
 839 static void *scrub_progress_cycle(void *ctx)
 840 {
 841         int ret;
 842         int old;
 843         int i;
 844         char fsid[37];
 845         struct scrub_progress *sp;
 846         struct scrub_progress *sp_last;
 847         struct scrub_progress *sp_shared;
 848         struct timeval tv;
 849         struct scrub_progress_cycle *spc = ctx;
 850         int ndev = spc->fi->num_devices;
 851         int this = 1;
 852         int last = 0;
 853         int peer_fd = -1;
 854         struct pollfd accept_poll_fd = {
 855                 .fd = spc->prg_fd,
 856                 .events = POLLIN,
 857                 .revents = 0,
 858         };
 859         struct pollfd write_poll_fd = {
 860                 .events = POLLOUT,
 861                 .revents = 0,
 862         };
 863         struct sockaddr_un peer;
 864         socklen_t peer_size = sizeof(peer);
 865
 866         ret = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 867         if (ret)
 868                 return ERR_PTR(-ret);
 869
 870         uuid_unparse(spc->fi->fsid, fsid);
 871
 872         for (i = 0; i < ndev; ++i) {
 873                 sp = &spc->progress[i];
 874                 sp_last = &spc->progress[i + ndev];
 875                 sp_shared = &spc->shared_progress[i];
 876                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 877                                                 sp_shared->scrub_args.devid;
 878                 sp->fd = sp_last->fd = spc->fdmnt;
 879                 sp->stats.t_start = sp_last->stats.t_start =
 880                                                 sp_shared->stats.t_start;
 881                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 882                 sp->skip = sp_last->skip = sp_shared->skip;
 883                 sp->stats.finished = sp_last->stats.finished =
 884                                                 sp_shared->stats.finished;
 885         }
 886
 887         while (1) {
 888                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 889                 if (ret == -1)
 890                         return ERR_PTR(-errno);
 891                 if (ret)
 892                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 893                                          &peer_size);
 894                 gettimeofday(&tv, NULL);
 895                 this = (this + 1)%2;
 896                 last = (last + 1)%2;
 897                 for (i = 0; i < ndev; ++i) {
 898                         sp = &spc->progress[this * ndev + i];
 899                         sp_last = &spc->progress[last * ndev + i];
 900                         sp_shared = &spc->shared_progress[i];
 901                         if (sp->stats.finished)
 902                                 continue;
 903                         progress_one_dev(sp);
 904                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 905                         if (!sp->ret)
 906                                 continue;
 907                         if (sp->ioctl_errno != ENOTCONN &&
 908                             sp->ioctl_errno != ENODEV)
 909                                 return ERR_PTR(-sp->ioctl_errno);
 910                         /*
 911                          * scrub finished or device removed, check the
 912                          * finished flag. if unset, just use the last
 913                          * result we got for the current write and go
 914                          * on. flag should be set on next cycle, then.
 915                          */
 916                         ret = pthread_mutex_lock(&sp_shared->progress_mutex);
 917                         if (ret)
 918                                 return ERR_PTR(-ret);
 919                         if (!sp_shared->stats.finished) {
 920                                 ret = pthread_mutex_unlock(
 921                                                 &sp_shared->progress_mutex);
 922                                 if (ret)
 923                                         return ERR_PTR(-ret);
 924                                 memcpy(sp, sp_last, sizeof(*sp));
 925                                 continue;
 926                         }
 927                         ret = pthread_mutex_unlock(&sp_shared->progress_mutex);
 928                         if (ret)
 929                                 return ERR_PTR(-ret);
 930                         memcpy(sp, sp_shared, sizeof(*sp));
 931                         memcpy(sp_last, sp_shared, sizeof(*sp));
 932                 }
 933                 if (peer_fd != -1) {
 934                         write_poll_fd.fd = peer_fd;
 935                         ret = poll(&write_poll_fd, 1, 0);
 936                         if (ret == -1)
 937                                 return ERR_PTR(-errno);
 938                         if (ret) {
 939                                 ret = scrub_write_file(
 940                                         peer_fd, fsid,
 941                                         &spc->progress[this * ndev], ndev);
 942                                 if (ret)
 943                                         return ERR_PTR(ret);
 944                         }
 945                         close(peer_fd);
 946                         peer_fd = -1;
 947                 }
 948                 if (!spc->do_record)
 949                         continue;
 950                 ret = scrub_write_progress(spc->write_mutex, fsid,
 951                                            &spc->progress[this * ndev], ndev);
 952                 if (ret)
 953                         return ERR_PTR(ret);
 954         }
 955 }
 956
 957 static struct scrub_file_record *last_dev_scrub(
 958                 struct scrub_file_record *const *const past_scrubs, u64 devid)
 959 {
 960         int i;
 961
 962         if (!past_scrubs || IS_ERR(past_scrubs))
 963                 return NULL;
 964
 965         for (i = 0; past_scrubs[i]; ++i)
 966                 if (past_scrubs[i]->devid == devid)
 967                         return past_scrubs[i];
 968
 969         return NULL;
 970 }
 971
 972 int mkdir_p(char *path)
 973 {
 974         int i;
 975         int ret;
 976
 977         for (i = 1; i < strlen(path); ++i) {
 978                 if (path[i] != '/')
 979                         continue;
 980                 path[i] = '\0';
 981                 ret = mkdir(path, 0777);
 982                 if (ret && errno != EEXIST)
 983                         return 1;
 984                 path[i] = '/';
 985         }
 986
 987         return 0;
 988 }
 989
 990 static const char * const cmd_scrub_start_usage[];
 991 static const char * const cmd_scrub_resume_usage[];
 992
 993 static int scrub_start(int argc, char **argv, int resume)
 994 {
 995         int fdmnt;
 996         int prg_fd = -1;
 997         int fdres = -1;
 998         int ret;
 999         pid_t pid;
1000         int c;
1001         int i;
1002         int err = 0;
1003         int e_uncorrectable = 0;
1004         int e_correctable = 0;
1005         int print_raw = 0;
1006         char *path;
1007         int do_background = 1;
1008         int do_wait = 0;
1009         int do_print = 0;
1010         int do_quiet = 0;
1011         int do_record = 1;
1012         int readonly = 0;
1013         int do_stats_per_dev = 0;
1014         int n_start = 0;
1015         int n_skip = 0;
1016         int n_resume = 0;
1017         struct btrfs_ioctl_fs_info_args fi_args;
1018         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1019         struct scrub_progress *sp = NULL;
1020         struct scrub_fs_stat fs_stat;
1021         struct timeval tv;
1022         struct sockaddr_un addr = {
1023                 .sun_family = AF_UNIX,
1024         };
1025         pthread_t *t_devs = NULL;
1026         pthread_t t_prog;
1027         pthread_attr_t t_attr;
1028         struct scrub_file_record **past_scrubs = NULL;
1029         struct scrub_file_record *last_scrub = NULL;
1030         char *datafile = strdup(SCRUB_DATA_FILE);
1031         char fsid[37];
1032         char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1033         struct scrub_progress_cycle spc;
1034         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1035         void *terr;
1036         u64 devid;
1037
1038         optind = 1;
1039         while ((c = getopt(argc, argv, "BdqrR")) != -1) {
1040                 switch (c) {
1041                 case 'B':
1042                         do_background = 0;
1043                         do_wait = 1;
1044                         do_print = 1;
1045                         break;
1046                 case 'd':
1047                         do_stats_per_dev = 1;
1048                         break;
1049                 case 'q':
1050                         do_quiet = 1;
1051                         break;
1052                 case 'r':
1053                         readonly = 1;
1054                         break;
1055                 case 'R':
1056                         print_raw = 1;
1057                         break;
1058                 case '?':
1059                 default:
1060                         usage(resume ? cmd_scrub_resume_usage :
1061                                                 cmd_scrub_start_usage);
1062                 }
1063         }
1064
1065         /* try to catch most error cases before forking */
1066
1067         if (check_argc_exact(argc - optind, 1)) {
1068                 usage(resume ? cmd_scrub_resume_usage :
1069                                         cmd_scrub_start_usage);
1070         }
1071
1072         spc.progress = NULL;
1073         if (do_quiet && do_print)
1074                 do_print = 0;
1075
1076         if (mkdir_p(datafile)) {
1077                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1078                                "file, mkdir %s failed: %s. Status recording "
1079                                "disabled\n", datafile, strerror(errno));
1080                 do_record = 0;
1081         }
1082         free(datafile);
1083
1084         path = argv[optind];
1085
1086         fdmnt = open_file_or_dir(path);
1087         if (fdmnt < 0) {
1088                 ERR(!do_quiet, "ERROR: can't access '%s'\n", path);
1089                 return 12;
1090         }
1091
1092         ret = get_fs_info(fdmnt, path, &fi_args, &di_args);
1093         if (ret) {
1094                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1095                     "%s\n", strerror(-ret));
1096                 err = 1;
1097                 goto out;
1098         }
1099         if (!fi_args.num_devices) {
1100                 ERR(!do_quiet, "ERROR: no devices found\n");
1101                 err = 1;
1102                 goto out;
1103         }
1104
1105         uuid_unparse(fi_args.fsid, fsid);
1106         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1107         if (fdres < 0 && fdres != -ENOENT) {
1108                 ERR(!do_quiet, "WARNING: failed to open status file: "
1109                     "%s\n", strerror(-fdres));
1110         } else if (fdres >= 0) {
1111                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1112                 if (IS_ERR(past_scrubs))
1113                         ERR(!do_quiet, "WARNING: failed to read status file: "
1114                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1115                 close(fdres);
1116         }
1117
1118         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1119         sp = calloc(fi_args.num_devices, sizeof(*sp));
1120         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1121
1122         if (!t_devs || !sp || !spc.progress) {
1123                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1124                 err = 1;
1125                 goto out;
1126         }
1127
1128         ret = pthread_attr_init(&t_attr);
1129         if (ret) {
1130                 ERR(!do_quiet, "ERROR: pthread_attr_init failed: %s\n",
1131                     strerror(ret));
1132                 err = 1;
1133                 goto out;
1134         }
1135
1136         for (i = 0; i < fi_args.num_devices; ++i) {
1137                 devid = di_args[i].devid;
1138                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1139                 if (ret) {
1140                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1141                             "%s\n", strerror(ret));
1142                         err = 1;
1143                         goto out;
1144                 }
1145                 last_scrub = last_dev_scrub(past_scrubs, devid);
1146                 sp[i].scrub_args.devid = devid;
1147                 sp[i].fd = fdmnt;
1148                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1149                                              !last_scrub->stats.finished)) {
1150                         ++n_resume;
1151                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1152                         sp[i].resumed = last_scrub;
1153                 } else if (resume) {
1154                         ++n_skip;
1155                         sp[i].skip = 1;
1156                         sp[i].resumed = last_scrub;
1157                         continue;
1158                 } else {
1159                         ++n_start;
1160                         sp[i].scrub_args.start = 0ll;
1161                         sp[i].resumed = NULL;
1162                 }
1163                 sp[i].skip = 0;
1164                 sp[i].scrub_args.end = (u64)-1ll;
1165                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1166         }
1167
1168         if (!n_start && !n_resume) {
1169                 if (!do_quiet)
1170                         printf("scrub: nothing to resume for %s, fsid %s\n",
1171                                path, fsid);
1172                 err = 0;
1173                 goto out;
1174         }
1175
1176         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1177         while (ret != -1) {
1178                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1179                                         sock_path, sizeof(sock_path));
1180                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1181                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1182                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1183                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1184                 if (ret != -1 || errno != EADDRINUSE)
1185                         break;
1186                 /*
1187                  * bind failed with EADDRINUSE. so let's see if anyone answers
1188                  * when we make a call to the socket ...
1189                  */
1190                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1191                 if (!ret || errno != ECONNREFUSED) {
1192                         /* ... yes, so scrub must be running. error out */
1193                         fprintf(stderr, "ERROR: scrub already running\n");
1194                         close(prg_fd);
1195                         prg_fd = -1;
1196                         goto out;
1197                 }
1198                 /*
1199                  * ... no, this means someone left us alone with an unused
1200                  * socket in the file system. remove it and try again.
1201                  */
1202                 ret = unlink(sock_path);
1203         }
1204         if (ret != -1)
1205                 ret = listen(prg_fd, 100);
1206         if (ret == -1) {
1207                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1208                     "socket at %s: %s. Progress cannot be queried\n",
1209                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1210                     strerror(errno));
1211                 if (prg_fd != -1) {
1212                         close(prg_fd);
1213                         prg_fd = -1;
1214                         if (sock_path[0])
1215                                 unlink(sock_path);
1216                 }
1217         }
1218
1219         if (do_record) {
1220                 /* write all-zero progress file for a start */
1221                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1222                                            fi_args.num_devices);
1223                 if (ret) {
1224                         ERR(!do_quiet, "WARNING: failed to write the progress "
1225                             "status file: %s. Status recording disabled\n",
1226                             strerror(-ret));
1227                         do_record = 0;
1228                 }
1229         }
1230
1231         if (do_background) {
1232                 pid = fork();
1233                 if (pid == -1) {
1234                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1235                                         "%s\n", strerror(errno));
1236                         err = 1;
1237                         goto out;
1238                 }
1239
1240                 if (pid) {
1241                         int stat;
1242                         scrub_handle_sigint_parent();
1243                         if (!do_quiet)
1244                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1245                                        n_start ? "started" : "resumed",
1246                                        path, fsid, pid);
1247                         if (!do_wait) {
1248                                 err = 0;
1249                                 goto out;
1250                         }
1251                         ret = wait(&stat);
1252                         if (ret != pid) {
1253                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1254                                     "%s\n", ret, strerror(errno));
1255                                 err = 1;
1256                                 goto out;
1257                         }
1258                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1259                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1260                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1261                                 goto out;
1262                         }
1263                         err = 0;
1264                         goto out;
1265                 }
1266         }
1267
1268         scrub_handle_sigint_child(fdmnt);
1269
1270         for (i = 0; i < fi_args.num_devices; ++i) {
1271                 if (sp[i].skip) {
1272                         sp[i].scrub_args.progress = sp[i].resumed->p;
1273                         sp[i].stats = sp[i].resumed->stats;
1274                         sp[i].ret = 0;
1275                         sp[i].stats.finished = 1;
1276                         continue;
1277                 }
1278                 devid = di_args[i].devid;
1279                 gettimeofday(&tv, NULL);
1280                 sp[i].stats.t_start = tv.tv_sec;
1281                 ret = pthread_create(&t_devs[i], &t_attr,
1282                                         scrub_one_dev, &sp[i]);
1283                 if (ret) {
1284                         if (do_print)
1285                                 fprintf(stderr, "ERROR: creating "
1286                                         "scrub_one_dev[%llu] thread failed: "
1287                                         "%s\n", devid, strerror(ret));
1288                         err = 1;
1289                         goto out;
1290                 }
1291         }
1292
1293         spc.fdmnt = fdmnt;
1294         spc.prg_fd = prg_fd;
1295         spc.do_record = do_record;
1296         spc.write_mutex = &spc_write_mutex;
1297         spc.shared_progress = sp;
1298         spc.fi = &fi_args;
1299         ret = pthread_create(&t_prog, &t_attr, scrub_progress_cycle, &spc);
1300         if (ret) {
1301                 if (do_print)
1302                         fprintf(stderr, "ERROR: creating progress thread "
1303                                 "failed: %s\n", strerror(ret));
1304                 err = 1;
1305                 goto out;
1306         }
1307
1308         err = 0;
1309         for (i = 0; i < fi_args.num_devices; ++i) {
1310                 if (sp[i].skip)
1311                         continue;
1312                 devid = di_args[i].devid;
1313                 ret = pthread_join(t_devs[i], NULL);
1314                 if (ret) {
1315                         if (do_print)
1316                                 fprintf(stderr, "ERROR: pthread_join failed "
1317                                         "for scrub_one_dev[%llu]: %s\n", devid,
1318                                         strerror(ret));
1319                         ++err;
1320                         continue;
1321                 }
1322                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1323                         if (do_print)
1324                                 fprintf(stderr, "WARNING: device %lld not "
1325                                         "present\n", devid);
1326                         continue;
1327                 }
1328                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1329                         ++err;
1330                 } else if (sp[i].ret) {
1331                         if (do_print)
1332                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1333                                         "for device id %lld (%s)\n", path,
1334                                         devid, strerror(sp[i].ioctl_errno));
1335                         ++err;
1336                         continue;
1337                 }
1338                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1339                         e_uncorrectable++;
1340                 if (sp[i].scrub_args.progress.corrected_errors > 0
1341                     || sp[i].scrub_args.progress.unverified_errors > 0)
1342                         e_correctable++;
1343         }
1344
1345         if (do_print) {
1346                 const char *append = "done";
1347                 if (!do_stats_per_dev)
1348                         init_fs_stat(&fs_stat);
1349                 for (i = 0; i < fi_args.num_devices; ++i) {
1350                         if (do_stats_per_dev) {
1351                                 print_scrub_dev(&di_args[i],
1352                                                 &sp[i].scrub_args.progress,
1353                                                 print_raw,
1354                                                 sp[i].ret ? "canceled" : "done",
1355                                                 &sp[i].stats);
1356                         } else {
1357                                 if (sp[i].ret)
1358                                         append = "canceled";
1359                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1360                                                 &sp[i].stats, &fs_stat);
1361                         }
1362                 }
1363                 if (!do_stats_per_dev) {
1364                         printf("scrub %s for %s\n", append, fsid);
1365                         print_fs_stat(&fs_stat, print_raw);
1366                 }
1367         }
1368
1369         ret = pthread_cancel(t_prog);
1370         if (!ret)
1371                 ret = pthread_join(t_prog, &terr);
1372         if (do_print && ret) {
1373                 fprintf(stderr, "ERROR: progress thead handling failed: %s\n",
1374                         strerror(ret));
1375         }
1376
1377         if (do_print && terr && terr != PTHREAD_CANCELED) {
1378                 fprintf(stderr, "ERROR: recording progress "
1379                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1380         }
1381
1382         if (do_record) {
1383                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1384                                            fi_args.num_devices);
1385                 if (ret && do_print) {
1386                         fprintf(stderr, "ERROR: failed to record the result: "
1387                                 "%s\n", strerror(-ret));
1388                 }
1389         }
1390
1391         scrub_handle_sigint_child(-1);
1392
1393 out:
1394         free_history(past_scrubs);
1395         free(di_args);
1396         free(t_devs);
1397         free(sp);
1398         free(spc.progress);
1399         if (prg_fd > -1) {
1400                 close(prg_fd);
1401                 if (sock_path[0])
1402                         unlink(sock_path);
1403         }
1404         close(fdmnt);
1405
1406         if (err)
1407                 return 1;
1408         if (e_correctable)
1409                 return 7;
1410         if (e_uncorrectable)
1411                 return 8;
1412         return 0;
1413 }
1414
1415 static const char * const cmd_scrub_start_usage[] = {
1416         "btrfs scrub start [-Bdqr] <path>|<device>",
1417         "Start a new scrub",
1418         "",
1419         "-B     do not background",
1420         "-d     stats per device (-B only)",
1421         "-q     be quiet",
1422         "-r     read only mode",
1423         NULL
1424 };
1425
1426 static int cmd_scrub_start(int argc, char **argv)
1427 {
1428         return scrub_start(argc, argv, 0);
1429 }
1430
1431 static const char * const cmd_scrub_cancel_usage[] = {
1432         "btrfs scrub cancel <path>|<device>",
1433         "Cancel a running scrub",
1434         NULL
1435 };
1436
1437 static int cmd_scrub_cancel(int argc, char **argv)
1438 {
1439         char *path;
1440         int ret;
1441         int fdmnt;
1442         int err;
1443         char mp[BTRFS_PATH_NAME_MAX + 1];
1444         struct btrfs_fs_devices *fs_devices_mnt = NULL;
1445
1446         if (check_argc_exact(argc, 2))
1447                 usage(cmd_scrub_cancel_usage);
1448
1449         path = argv[1];
1450
1451         fdmnt = open_file_or_dir(path);
1452         if (fdmnt < 0) {
1453                 fprintf(stderr, "ERROR: scrub cancel failed\n");
1454                 return 12;
1455         }
1456
1457 again:
1458         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1459         err = errno;
1460
1461         if (ret && err == EINVAL) {
1462                 /* path is not a btrfs mount point.  See if it's a device. */
1463                 ret = check_mounted_where(fdmnt, path, mp, sizeof(mp),
1464                                           &fs_devices_mnt);
1465                 if (ret) {
1466                         /* It is a device; open the mountpoint. */
1467                         close(fdmnt);
1468                         fdmnt = open_file_or_dir(mp);
1469                         if (fdmnt >= 0) {
1470                                 path = mp;
1471                                 goto again;
1472                         }
1473                 }
1474         }
1475
1476         close(fdmnt);
1477
1478         if (ret) {
1479                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1480                         err == ENOTCONN ? "not running" : strerror(errno));
1481                 return 1;
1482         }
1483
1484         printf("scrub cancelled\n");
1485
1486         return 0;
1487 }
1488
1489 static const char * const cmd_scrub_resume_usage[] = {
1490         "btrfs scrub resume [-Bdqr] <path>|<device>",
1491         "Resume previously canceled or interrupted scrub",
1492         "",
1493         "-B     do not background",
1494         "-d     stats per device (-B only)",
1495         "-q     be quiet",
1496         "-r     read only mode",
1497         NULL
1498 };
1499
1500 static int cmd_scrub_resume(int argc, char **argv)
1501 {
1502         return scrub_start(argc, argv, 1);
1503 }
1504
1505 static const char * const cmd_scrub_status_usage[] = {
1506         "btrfs scrub status [-dR] <path>|<device>",
1507         "Show status of running or finished scrub",
1508         "",
1509         "-d     stats per device",
1510         "-R     print raw stats",
1511         NULL
1512 };
1513
1514 static int cmd_scrub_status(int argc, char **argv)
1515 {
1516         char *path;
1517         struct btrfs_ioctl_fs_info_args fi_args;
1518         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1519         struct scrub_file_record **past_scrubs = NULL;
1520         struct scrub_file_record *last_scrub;
1521         struct scrub_fs_stat fs_stat;
1522         struct sockaddr_un addr = {
1523                 .sun_family = AF_UNIX,
1524         };
1525         int ret;
1526         int i;
1527         int fdmnt;
1528         int print_raw = 0;
1529         int do_stats_per_dev = 0;
1530         int c;
1531         char fsid[37];
1532         int fdres = -1;
1533         int err = 0;
1534
1535         optind = 1;
1536         while ((c = getopt(argc, argv, "dR")) != -1) {
1537                 switch (c) {
1538                 case 'd':
1539                         do_stats_per_dev = 1;
1540                         break;
1541                 case 'R':
1542                         print_raw = 1;
1543                         break;
1544                 case '?':
1545                 default:
1546                         usage(cmd_scrub_status_usage);
1547                 }
1548         }
1549
1550         if (check_argc_exact(argc - optind, 1))
1551                 usage(cmd_scrub_status_usage);
1552
1553         path = argv[optind];
1554
1555         fdmnt = open_file_or_dir(path);
1556         if (fdmnt < 0) {
1557                 fprintf(stderr, "ERROR: can't access to '%s'\n", path);
1558                 return 12;
1559         }
1560
1561         ret = get_fs_info(fdmnt, path, &fi_args, &di_args);
1562         if (ret) {
1563                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1564                                 "%s\n", strerror(-ret));
1565                 err = 1;
1566                 goto out;
1567         }
1568         if (!fi_args.num_devices) {
1569                 fprintf(stderr, "ERROR: no devices found\n");
1570                 err = 1;
1571                 goto out;
1572         }
1573
1574         uuid_unparse(fi_args.fsid, fsid);
1575
1576         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1577         if (fdres == -1) {
1578                 fprintf(stderr, "ERROR: failed to create socket to "
1579                         "receive progress information: %s\n",
1580                         strerror(errno));
1581                 err = 1;
1582                 goto out;
1583         }
1584         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1585                         NULL, addr.sun_path, sizeof(addr.sun_path));
1586         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1587         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1588         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1589         if (ret == -1) {
1590                 close(fdres);
1591                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1592                 if (fdres < 0 && fdres != -ENOENT) {
1593                         fprintf(stderr, "WARNING: failed to open status file: "
1594                                 "%s\n", strerror(-fdres));
1595                         err = 1;
1596                         goto out;
1597                 }
1598         }
1599
1600         if (fdres >= 0) {
1601                 past_scrubs = scrub_read_file(fdres, 1);
1602                 if (IS_ERR(past_scrubs))
1603                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1604                                 strerror(-PTR_ERR(past_scrubs)));
1605         }
1606
1607         printf("scrub status for %s\n", fsid);
1608
1609         if (do_stats_per_dev) {
1610                 for (i = 0; i < fi_args.num_devices; ++i) {
1611                         last_scrub = last_dev_scrub(past_scrubs,
1612                                                         di_args[i].devid);
1613                         if (!last_scrub) {
1614                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1615                                                 NULL, NULL);
1616                                 continue;
1617                         }
1618                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1619                                         last_scrub->stats.finished ?
1620                                                         "history" : "status",
1621                                         &last_scrub->stats);
1622                 }
1623         } else {
1624                 init_fs_stat(&fs_stat);
1625                 for (i = 0; i < fi_args.num_devices; ++i) {
1626                         last_scrub = last_dev_scrub(past_scrubs,
1627                                                         di_args[i].devid);
1628                         if (!last_scrub)
1629                                 continue;
1630                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1631                                         &fs_stat);
1632                 }
1633                 print_fs_stat(&fs_stat, print_raw);
1634         }
1635
1636 out:
1637         free_history(past_scrubs);
1638         free(di_args);
1639         if (fdres > -1)
1640                 close(fdres);
1641
1642         return err;
1643 }
1644
1645 const struct cmd_group scrub_cmd_group = {
1646         scrub_cmd_group_usage, NULL, {
1647                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1648                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1649                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1650                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1651                 { 0, 0, 0, 0, 0 }
1652         }
1653 };
1654
1655 int cmd_scrub(int argc, char **argv)
1656 {
1657         return handle_command_group(&scrub_cmd_group, argc, argv);
1658 }