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 <sys/syscall.h>
  28 #include <poll.h>
  29 #include <sys/file.h>
  30 #include <uuid/uuid.h>
  31 #include <fcntl.h>
  32 #include <unistd.h>
  33 #include <pthread.h>
  34 #include <ctype.h>
  35 #include <signal.h>
  36 #include <stdarg.h>
  37
  38 #include "ctree.h"
  39 #include "ioctl.h"
  40 #include "utils.h"
  41 #include "volumes.h"
  42 #include "disk-io.h"
  43
  44 #include "commands.h"
  45
  46 static const char * const scrub_cmd_group_usage[] = {
  47         "btrfs scrub <command> [options] <path>|<device>",
  48         NULL
  49 };
  50
  51 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  52 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  53 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  54 #define SCRUB_FILE_VERSION "1"
  55
  56 struct scrub_stats {
  57         time_t t_start;
  58         time_t t_resumed;
  59         u64 duration;
  60         u64 finished;
  61         u64 canceled;
  62 };
  63
  64 /* TBD: replace with #include "linux/ioprio.h" in some years */
  65 #if !defined (IOPRIO_H)
  66 #define IOPRIO_WHO_PROCESS 1
  67 #define IOPRIO_CLASS_SHIFT 13
  68 #define IOPRIO_PRIO_VALUE(class, data) \
  69                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
  70 #define IOPRIO_CLASS_IDLE 3
  71 #endif
  72
  73 struct scrub_progress {
  74         struct btrfs_ioctl_scrub_args scrub_args;
  75         int fd;
  76         int ret;
  77         int skip;
  78         struct scrub_stats stats;
  79         struct scrub_file_record *resumed;
  80         int ioctl_errno;
  81         pthread_mutex_t progress_mutex;
  82         int ioprio_class;
  83         int ioprio_classdata;
  84 };
  85
  86 struct scrub_file_record {
  87         u8 fsid[BTRFS_FSID_SIZE];
  88         u64 devid;
  89         struct scrub_stats stats;
  90         struct btrfs_scrub_progress p;
  91 };
  92
  93 struct scrub_progress_cycle {
  94         int fdmnt;
  95         int prg_fd;
  96         int do_record;
  97         struct btrfs_ioctl_fs_info_args *fi;
  98         struct scrub_progress *progress;
  99         struct scrub_progress *shared_progress;
 100         pthread_mutex_t *write_mutex;
 101 };
 102
 103 struct scrub_fs_stat {
 104         struct btrfs_scrub_progress p;
 105         struct scrub_stats s;
 106         int i;
 107 };
 108
 109 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 110 {
 111         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 112         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 113         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 114         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 115         printf("\tread_errors: %lld\n", sp->read_errors);
 116         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 117         printf("\tverify_errors: %lld\n", sp->verify_errors);
 118         printf("\tno_csum: %lld\n", sp->no_csum);
 119         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 120         printf("\tsuper_errors: %lld\n", sp->super_errors);
 121         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 122         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 123         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 124         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 125         printf("\tlast_physical: %lld\n", sp->last_physical);
 126 }
 127
 128 #define ERR(test, ...) do {                     \
 129         if (test)                               \
 130                 fprintf(stderr, __VA_ARGS__);   \
 131 } while (0)
 132
 133 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 134         if (test)                               \
 135                 printf(" %s=%llu", desc, test); \
 136 } while (0)
 137
 138 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 139 {
 140         u64 err_cnt;
 141         u64 err_cnt2;
 142
 143         err_cnt = p->read_errors +
 144                         p->csum_errors +
 145                         p->verify_errors +
 146                         p->super_errors;
 147
 148         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 149
 150         if (p->malloc_errors)
 151                 printf("*** WARNING: memory allocation failed while scrubbing. "
 152                        "results may be inaccurate\n");
 153
 154         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
 155                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
 156                 max(err_cnt, err_cnt2));
 157
 158         if (err_cnt || err_cnt2) {
 159                 printf("\terror details:");
 160                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 161                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 162                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 163                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 164                 printf("\n");
 165                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 166                         "unverified errors: %llu\n", p->corrected_errors,
 167                         p->uncorrectable_errors, p->unverified_errors);
 168         }
 169 }
 170
 171 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 172         fs_stat->p.name += p->name;             \
 173 } while (0)
 174
 175 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 176 do {                                            \
 177         if (fs_stat->s.name > ss->name) {       \
 178                 fs_stat->s.name = ss->name;     \
 179         }                                       \
 180 } while (0)
 181
 182 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 183 do {                                                            \
 184         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 185                 fs_stat->s.name = ss->name;                     \
 186         }                                                       \
 187 } while (0)
 188
 189 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 190 do {                                                                    \
 191         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 192                 (fs_stat)->s.name = (ss)->name;                         \
 193         }                                                               \
 194 } while (0)
 195
 196 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 197                                 struct scrub_stats *ss,
 198                                 struct scrub_fs_stat *fs_stat)
 199 {
 200         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 201         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 202         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 203         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 204         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 205         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 206         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 207         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 208         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 209         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 210         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 211         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 212         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 213         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 214         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 215         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 216         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 217         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 218         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 219 }
 220
 221 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 222 {
 223         memset(fs_stat, 0, sizeof(*fs_stat));
 224         fs_stat->s.finished = 1;
 225 }
 226
 227 static void _print_scrub_ss(struct scrub_stats *ss)
 228 {
 229         char t[4096];
 230         struct tm tm;
 231
 232         if (!ss || !ss->t_start) {
 233                 printf("\tno stats available\n");
 234                 return;
 235         }
 236         if (ss->t_resumed) {
 237                 localtime_r(&ss->t_resumed, &tm);
 238                 strftime(t, sizeof(t), "%c", &tm);
 239                 t[sizeof(t) - 1] = '\0';
 240                 printf("\tscrub resumed at %s", t);
 241         } else {
 242                 localtime_r(&ss->t_start, &tm);
 243                 strftime(t, sizeof(t), "%c", &tm);
 244                 t[sizeof(t) - 1] = '\0';
 245                 printf("\tscrub started at %s", t);
 246         }
 247         if (ss->finished && !ss->canceled) {
 248                 printf(" and finished after %llu seconds\n",
 249                        ss->duration);
 250         } else if (ss->canceled) {
 251                 printf(" and was aborted after %llu seconds\n",
 252                        ss->duration);
 253         } else {
 254                 printf(", running for %llu seconds\n", ss->duration);
 255         }
 256 }
 257
 258 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 259                                 struct btrfs_scrub_progress *p, int raw,
 260                                 const char *append, struct scrub_stats *ss)
 261 {
 262         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 263                append ? append : "");
 264
 265         _print_scrub_ss(ss);
 266
 267         if (p) {
 268                 if (raw)
 269                         print_scrub_full(p);
 270                 else
 271                         print_scrub_summary(p);
 272         }
 273 }
 274
 275 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 276 {
 277         _print_scrub_ss(&fs_stat->s);
 278
 279         if (raw)
 280                 print_scrub_full(&fs_stat->p);
 281         else
 282                 print_scrub_summary(&fs_stat->p);
 283 }
 284
 285 static void free_history(struct scrub_file_record **last_scrubs)
 286 {
 287         struct scrub_file_record **l = last_scrubs;
 288         if (!l || IS_ERR(l))
 289                 return;
 290         while (*l)
 291                 free(*l++);
 292         free(last_scrubs);
 293 }
 294
 295 /*
 296  * cancels a running scrub and makes the master process record the current
 297  * progress status before exiting.
 298  */
 299 static int cancel_fd = -1;
 300 static void scrub_sigint_record_progress(int signal)
 301 {
 302         int ret;
 303
 304         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 305         if (ret < 0)
 306                 perror("Scrub cancel failed");
 307 }
 308
 309 static int scrub_handle_sigint_parent(void)
 310 {
 311         struct sigaction sa = {
 312                 .sa_handler = SIG_IGN,
 313                 .sa_flags = SA_RESTART,
 314         };
 315
 316         return sigaction(SIGINT, &sa, NULL);
 317 }
 318
 319 static int scrub_handle_sigint_child(int fd)
 320 {
 321         struct sigaction sa = {
 322                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 323         };
 324
 325         cancel_fd = fd;
 326         return sigaction(SIGINT, &sa, NULL);
 327 }
 328
 329 static int scrub_datafile(const char *fn_base, const char *fn_local,
 330                                 const char *fn_tmp, char *datafile, int size)
 331 {
 332         int ret;
 333         int end = size - 2;
 334
 335         datafile[end + 1] = '\0';
 336         strncpy(datafile, fn_base, end);
 337         ret = strlen(datafile);
 338
 339         if (ret + 1 > end)
 340                 return -EOVERFLOW;
 341
 342         datafile[ret] = '.';
 343         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 344         ret = strlen(datafile);
 345
 346         if (ret + 1 > end)
 347                 return -EOVERFLOW;
 348
 349         if (fn_tmp) {
 350                 datafile[ret] = '_';
 351                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 352                 ret = strlen(datafile);
 353
 354                 if (ret > end)
 355                         return -EOVERFLOW;
 356         }
 357
 358         return 0;
 359 }
 360
 361 static int scrub_open_file(const char *datafile, int m)
 362 {
 363         int fd;
 364         int ret;
 365
 366         fd = open(datafile, m, 0600);
 367         if (fd < 0)
 368                 return -errno;
 369
 370         ret = flock(fd, LOCK_EX|LOCK_NB);
 371         if (ret) {
 372                 ret = errno;
 373                 close(fd);
 374                 return -ret;
 375         }
 376
 377         return fd;
 378 }
 379
 380 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 381 {
 382         int ret;
 383         char datafile[BTRFS_PATH_NAME_MAX + 1];
 384         ret = scrub_datafile(fn_base, fn_local, NULL,
 385                                 datafile, sizeof(datafile));
 386         if (ret < 0)
 387                 return ret;
 388         return scrub_open_file(datafile, O_RDONLY);
 389 }
 390
 391 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 392                                 const char *tmp)
 393 {
 394         int ret;
 395         char datafile[BTRFS_PATH_NAME_MAX + 1];
 396         ret = scrub_datafile(fn_base, fn_local, tmp,
 397                                 datafile, sizeof(datafile));
 398         if (ret < 0)
 399                 return ret;
 400         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 401 }
 402
 403 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 404                                 const char *tmp)
 405 {
 406         int ret;
 407         char datafile_old[BTRFS_PATH_NAME_MAX + 1];
 408         char datafile_new[BTRFS_PATH_NAME_MAX + 1];
 409         ret = scrub_datafile(fn_base, fn_local, tmp,
 410                                 datafile_old, sizeof(datafile_old));
 411         if (ret < 0)
 412                 return ret;
 413         ret = scrub_datafile(fn_base, fn_local, NULL,
 414                                 datafile_new, sizeof(datafile_new));
 415         if (ret < 0)
 416                 return ret;
 417         ret = rename(datafile_old, datafile_new);
 418         return ret ? -errno : 0;
 419 }
 420
 421 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 422         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 423 }
 424
 425 /*
 426  * returns 0 if the key did not match (nothing was read)
 427  *         1 if the key did match (success)
 428  *        -1 if the key did match and an error occured
 429  */
 430 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 431                         const char *key, u64 *dest)
 432 {
 433         int j;
 434
 435         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 436                 *i += len - 1;
 437                 if (buf[*i] != ':')
 438                         return -1;
 439                 *i += 1;
 440                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 441                         ;
 442                 if (*i + j >= avail)
 443                         return -1;
 444                 *dest = atoll(&buf[*i]);
 445                 *i += j;
 446                 return 1;
 447         }
 448
 449         return 0;
 450 }
 451
 452 #define _SCRUB_INVALID do {                                             \
 453         if (report_errors)                                              \
 454                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 455                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 456                         lineno, i, state, 20 > avail ? avail : 20,      \
 457                         l + i,  __FILE__, __LINE__);                    \
 458         goto skip;                                                      \
 459 } while (0)
 460
 461 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 462 {
 463         int avail = 0;
 464         int old_avail = 0;
 465         char l[16 * 1024];
 466         int state = 0;
 467         int curr = -1;
 468         int i = 0;
 469         int j;
 470         int ret;
 471         int eof = 0;
 472         int lineno = 0;
 473         u64 version;
 474         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 475         struct scrub_file_record **p = NULL;
 476
 477         if (fd < 0)
 478                 return ERR_PTR(-EINVAL);
 479
 480 again:
 481         old_avail = avail - i;
 482         BUG_ON(old_avail < 0);
 483         if (old_avail)
 484                 memmove(l, l + i, old_avail);
 485         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 486         if (avail == 0)
 487                 eof = 1;
 488         if (avail == 0 && old_avail == 0) {
 489                 if (curr >= 0 &&
 490                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 491                         p[curr] = NULL;
 492                 } else if (curr == -1) {
 493                         p = ERR_PTR(-ENODATA);
 494                 }
 495                 return p;
 496         }
 497         if (avail == -1)
 498                 return ERR_PTR(-errno);
 499         avail += old_avail;
 500
 501         i = 0;
 502         while (i < avail) {
 503                 switch (state) {
 504                 case 0: /* start of file */
 505                         ret = scrub_kvread(&i,
 506                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 507                                 SCRUB_FILE_VERSION_PREFIX, &version);
 508                         if (ret != 1)
 509                                 _SCRUB_INVALID;
 510                         if (version != atoll(SCRUB_FILE_VERSION))
 511                                 return ERR_PTR(-ENOTSUP);
 512                         state = 6;
 513                         continue;
 514                 case 1: /* start of line, alloc */
 515                         /*
 516                          * this state makes sure we have a complete line in
 517                          * further processing, so we don't need wrap-tracking
 518                          * everywhere.
 519                          */
 520                         if (!eof && !memchr(l + i, '\n', avail - i))
 521                                 goto again;
 522                         ++lineno;
 523                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 524                                                 BTRFS_FSID_SIZE) == 0) {
 525                                 state = 2;
 526                                 continue;
 527                         }
 528                         ++curr;
 529                         p = realloc(p, (curr + 2) * sizeof(*p));
 530                         if (p)
 531                                 p[curr] = malloc(sizeof(**p));
 532                         if (!p || !p[curr])
 533                                 return ERR_PTR(-errno);
 534                         memset(p[curr], 0, sizeof(**p));
 535                         p[curr + 1] = NULL;
 536                         ++state;
 537                         /* fall through */
 538                 case 2: /* start of line, skip space */
 539                         while (isspace(l[i]) && i < avail) {
 540                                 if (l[i] == '\n')
 541                                         ++lineno;
 542                                 ++i;
 543                         }
 544                         if (i >= avail ||
 545                             (!eof && !memchr(l + i, '\n', avail - i)))
 546                                 goto again;
 547                         ++state;
 548                         /* fall through */
 549                 case 3: /* read fsid */
 550                         if (i == avail)
 551                                 continue;
 552                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 553                                 ;
 554                         if (i + j + 1 >= avail)
 555                                 _SCRUB_INVALID;
 556                         if (j != 36)
 557                                 _SCRUB_INVALID;
 558                         l[i + j] = '\0';
 559                         ret = uuid_parse(l + i, p[curr]->fsid);
 560                         if (ret)
 561                                 _SCRUB_INVALID;
 562                         i += j + 1;
 563                         ++state;
 564                         /* fall through */
 565                 case 4: /* read dev id */
 566                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 567                                 ;
 568                         if (j == 0 || i + j + 1 >= avail)
 569                                 _SCRUB_INVALID;
 570                         p[curr]->devid = atoll(&l[i]);
 571                         i += j + 1;
 572                         ++state;
 573                         /* fall through */
 574                 case 5: /* read key/value pair */
 575                         ret = 0;
 576                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 577                                         &p[curr]->p);
 578                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 579                                         &p[curr]->p);
 580                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 581                                         &p[curr]->p);
 582                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 583                                         &p[curr]->p);
 584                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 585                                         &p[curr]->p);
 586                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 587                                         &p[curr]->p);
 588                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 589                                         &p[curr]->p);
 590                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 591                                         &p[curr]->p);
 592                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 593                                         &p[curr]->p);
 594                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 595                                         &p[curr]->p);
 596                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 597                                         &p[curr]->p);
 598                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 599                                         &p[curr]->p);
 600                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 601                                         &p[curr]->p);
 602                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 603                                         &p[curr]->p);
 604                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 605                                         &p[curr]->p);
 606                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 607                                         &p[curr]->stats);
 608                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 609                                         (u64 *)&p[curr]->stats);
 610                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 611                                         (u64 *)&p[curr]->stats);
 612                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 613                                         (u64 *)&p[curr]->stats);
 614                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 615                                         &p[curr]->stats);
 616                         if (ret != 1)
 617                                 _SCRUB_INVALID;
 618                         ++state;
 619                         /* fall through */
 620                 case 6: /* after number */
 621                         if (l[i] == '|')
 622                                 state = 5;
 623                         else if (l[i] == '\n')
 624                                 state = 1;
 625                         else
 626                                 _SCRUB_INVALID;
 627                         ++i;
 628                         continue;
 629                 case 99: /* skip rest of line */
 630 skip:
 631                         state = 99;
 632                         do {
 633                                 ++i;
 634                                 if (l[i - 1] == '\n') {
 635                                         state = 1;
 636                                         break;
 637                                 }
 638                         } while (i < avail);
 639                         continue;
 640                 }
 641                 BUG();
 642         }
 643         goto again;
 644 }
 645
 646 static int scrub_write_buf(int fd, const void *data, int len)
 647 {
 648         int ret;
 649         ret = write(fd, data, len);
 650         return ret - len;
 651 }
 652
 653 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 654                                 __attribute__ ((format (printf, 4, 5)));
 655 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 656 {
 657         int ret;
 658         va_list args;
 659
 660         va_start(args, fmt);
 661         ret = vsnprintf(buf, max, fmt, args);
 662         va_end(args);
 663         if (ret >= max)
 664                 return ret - max;
 665         return scrub_write_buf(fd, buf, ret);
 666 }
 667
 668 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 669                         data->resumed->p.name + data->scrub_args.progress.name
 670
 671 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 672                                                   struct scrub_progress *dest)
 673 {
 674         if (!data->resumed || data->skip)
 675                 return data;
 676
 677         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 678         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 679         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 680         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 681         _SCRUB_SUM(dest, data, read_errors);
 682         _SCRUB_SUM(dest, data, csum_errors);
 683         _SCRUB_SUM(dest, data, verify_errors);
 684         _SCRUB_SUM(dest, data, no_csum);
 685         _SCRUB_SUM(dest, data, csum_discards);
 686         _SCRUB_SUM(dest, data, super_errors);
 687         _SCRUB_SUM(dest, data, malloc_errors);
 688         _SCRUB_SUM(dest, data, uncorrectable_errors);
 689         _SCRUB_SUM(dest, data, corrected_errors);
 690         _SCRUB_SUM(dest, data, last_physical);
 691         dest->stats.canceled = data->stats.canceled;
 692         dest->stats.finished = data->stats.finished;
 693         dest->stats.t_resumed = data->stats.t_start;
 694         dest->stats.t_start = data->resumed->stats.t_start;
 695         dest->stats.duration = data->resumed->stats.duration +
 696                                                         data->stats.duration;
 697         dest->scrub_args.devid = data->scrub_args.devid;
 698         return dest;
 699 }
 700
 701 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 702         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 703                         use->scrub_args.progress.name)
 704
 705 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 706         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 707                         use->stats.name)
 708
 709 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 710 {
 711         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 712 }
 713
 714 static int scrub_write_file(int fd, const char *fsid,
 715                                 struct scrub_progress *data, int n)
 716 {
 717         int ret = 0;
 718         int i;
 719         char buf[1024];
 720         struct scrub_progress local;
 721         struct scrub_progress *use;
 722
 723         if (n < 1)
 724                 return -EINVAL;
 725
 726         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 727         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 728                                 SCRUB_FILE_VERSION "\n",
 729                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 730                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 731         if (ret)
 732                 return -EOVERFLOW;
 733
 734         for (i = 0; i < n; ++i) {
 735                 use = scrub_resumed_stats(&data[i], &local);
 736                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 737                     scrub_write_buf(fd, ":", 1) ||
 738                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 739                                         use->scrub_args.devid) ||
 740                     scrub_write_buf(fd, buf, ret) ||
 741                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 742                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 743                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 744                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 745                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 746                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 747                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 748                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 749                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 750                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 751                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 752                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 753                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 754                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 755                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 756                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 757                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 758                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 759                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 760                     scrub_write_buf(fd, "\n", 1)) {
 761                         return -EOVERFLOW;
 762                 }
 763         }
 764
 765         return 0;
 766 }
 767
 768 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 769                                 struct scrub_progress *data, int n)
 770 {
 771         int ret;
 772         int err;
 773         int fd = -1;
 774         int old;
 775
 776         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 777         if (ret) {
 778                 err = -ret;
 779                 goto out3;
 780         }
 781
 782         ret = pthread_mutex_lock(m);
 783         if (ret) {
 784                 err = -ret;
 785                 goto out2;
 786         }
 787
 788         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 789         if (fd < 0) {
 790                 err = fd;
 791                 goto out1;
 792         }
 793         err = scrub_write_file(fd, fsid, data, n);
 794         if (err)
 795                 goto out1;
 796         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 797         if (err)
 798                 goto out1;
 799
 800 out1:
 801         if (fd >= 0) {
 802                 ret = close(fd);
 803                 if (ret)
 804                         err = -errno;
 805         }
 806
 807         ret = pthread_mutex_unlock(m);
 808         if (ret && !err)
 809                 err = -ret;
 810
 811 out2:
 812         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 813         if (ret && !err)
 814                 err = -ret;
 815
 816 out3:
 817         return err;
 818 }
 819
 820 static void *scrub_one_dev(void *ctx)
 821 {
 822         struct scrub_progress *sp = ctx;
 823         int ret;
 824         struct timeval tv;
 825
 826         sp->stats.canceled = 0;
 827         sp->stats.duration = 0;
 828         sp->stats.finished = 0;
 829
 830         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 831                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 832                                         sp->ioprio_classdata));
 833         if (ret)
 834                 fprintf(stderr,
 835                         "WARNING: setting ioprio failed: %s (ignored).\n",
 836                         strerror(errno));
 837
 838         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 839         gettimeofday(&tv, NULL);
 840         sp->ret = ret;
 841         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 842         sp->stats.canceled = !!ret;
 843         sp->ioctl_errno = errno;
 844         ret = pthread_mutex_lock(&sp->progress_mutex);
 845         if (ret)
 846                 return ERR_PTR(-ret);
 847         sp->stats.finished = 1;
 848         ret = pthread_mutex_unlock(&sp->progress_mutex);
 849         if (ret)
 850                 return ERR_PTR(-ret);
 851
 852         return NULL;
 853 }
 854
 855 static void *progress_one_dev(void *ctx)
 856 {
 857         struct scrub_progress *sp = ctx;
 858
 859         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 860         sp->ioctl_errno = errno;
 861
 862         return NULL;
 863 }
 864
 865 /* nb: returns a negative errno via ERR_PTR */
 866 static void *scrub_progress_cycle(void *ctx)
 867 {
 868         int ret = 0;
 869         int  perr = 0;  /* positive / pthread error returns */
 870         int old;
 871         int i;
 872         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 873         struct scrub_progress *sp;
 874         struct scrub_progress *sp_last;
 875         struct scrub_progress *sp_shared;
 876         struct timeval tv;
 877         struct scrub_progress_cycle *spc = ctx;
 878         int ndev = spc->fi->num_devices;
 879         int this = 1;
 880         int last = 0;
 881         int peer_fd = -1;
 882         struct pollfd accept_poll_fd = {
 883                 .fd = spc->prg_fd,
 884                 .events = POLLIN,
 885                 .revents = 0,
 886         };
 887         struct pollfd write_poll_fd = {
 888                 .events = POLLOUT,
 889                 .revents = 0,
 890         };
 891         struct sockaddr_un peer;
 892         socklen_t peer_size = sizeof(peer);
 893
 894         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 895         if (perr)
 896                 goto out;
 897
 898         uuid_unparse(spc->fi->fsid, fsid);
 899
 900         for (i = 0; i < ndev; ++i) {
 901                 sp = &spc->progress[i];
 902                 sp_last = &spc->progress[i + ndev];
 903                 sp_shared = &spc->shared_progress[i];
 904                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 905                                                 sp_shared->scrub_args.devid;
 906                 sp->fd = sp_last->fd = spc->fdmnt;
 907                 sp->stats.t_start = sp_last->stats.t_start =
 908                                                 sp_shared->stats.t_start;
 909                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 910                 sp->skip = sp_last->skip = sp_shared->skip;
 911                 sp->stats.finished = sp_last->stats.finished =
 912                                                 sp_shared->stats.finished;
 913         }
 914
 915         while (1) {
 916                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 917                 if (ret == -1) {
 918                         ret = -errno;
 919                         goto out;
 920                 }
 921                 if (ret)
 922                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 923                                          &peer_size);
 924                 gettimeofday(&tv, NULL);
 925                 this = (this + 1)%2;
 926                 last = (last + 1)%2;
 927                 for (i = 0; i < ndev; ++i) {
 928                         sp = &spc->progress[this * ndev + i];
 929                         sp_last = &spc->progress[last * ndev + i];
 930                         sp_shared = &spc->shared_progress[i];
 931                         if (sp->stats.finished)
 932                                 continue;
 933                         progress_one_dev(sp);
 934                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 935                         if (!sp->ret)
 936                                 continue;
 937                         if (sp->ioctl_errno != ENOTCONN &&
 938                             sp->ioctl_errno != ENODEV) {
 939                                 ret = -sp->ioctl_errno;
 940                                 goto out;
 941                         }
 942                         /*
 943                          * scrub finished or device removed, check the
 944                          * finished flag. if unset, just use the last
 945                          * result we got for the current write and go
 946                          * on. flag should be set on next cycle, then.
 947                          */
 948                         perr = pthread_setcancelstate(
 949                                         PTHREAD_CANCEL_DISABLE, &old);
 950                         if (perr)
 951                                 goto out;
 952                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 953                         if (perr)
 954                                 goto out;
 955                         if (!sp_shared->stats.finished) {
 956                                 perr = pthread_mutex_unlock(
 957                                                 &sp_shared->progress_mutex);
 958                                 if (perr)
 959                                         goto out;
 960                                 perr = pthread_setcancelstate(
 961                                                 PTHREAD_CANCEL_ENABLE, &old);
 962                                 if (perr)
 963                                         goto out;
 964                                 memcpy(sp, sp_last, sizeof(*sp));
 965                                 continue;
 966                         }
 967                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 968                         if (perr)
 969                                 goto out;
 970                         perr = pthread_setcancelstate(
 971                                         PTHREAD_CANCEL_ENABLE, &old);
 972                         if (perr)
 973                                 goto out;
 974                         memcpy(sp, sp_shared, sizeof(*sp));
 975                         memcpy(sp_last, sp_shared, sizeof(*sp));
 976                 }
 977                 if (peer_fd != -1) {
 978                         write_poll_fd.fd = peer_fd;
 979                         ret = poll(&write_poll_fd, 1, 0);
 980                         if (ret == -1) {
 981                                 ret = -errno;
 982                                 goto out;
 983                         }
 984                         if (ret) {
 985                                 ret = scrub_write_file(
 986                                         peer_fd, fsid,
 987                                         &spc->progress[this * ndev], ndev);
 988                                 if (ret)
 989                                         goto out;
 990                         }
 991                         close(peer_fd);
 992                         peer_fd = -1;
 993                 }
 994                 if (!spc->do_record)
 995                         continue;
 996                 ret = scrub_write_progress(spc->write_mutex, fsid,
 997                                            &spc->progress[this * ndev], ndev);
 998                 if (ret)
 999                         goto out;
1000         }
1001 out:
1002         if (peer_fd != -1)
1003                 close(peer_fd);
1004         if (perr)
1005                 ret = -perr;
1006         return ERR_PTR(ret);
1007 }
1008
1009 static struct scrub_file_record *last_dev_scrub(
1010                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1011 {
1012         int i;
1013
1014         if (!past_scrubs || IS_ERR(past_scrubs))
1015                 return NULL;
1016
1017         for (i = 0; past_scrubs[i]; ++i)
1018                 if (past_scrubs[i]->devid == devid)
1019                         return past_scrubs[i];
1020
1021         return NULL;
1022 }
1023
1024 static int mkdir_p(char *path)
1025 {
1026         int i;
1027         int ret;
1028
1029         for (i = 1; i < strlen(path); ++i) {
1030                 if (path[i] != '/')
1031                         continue;
1032                 path[i] = '\0';
1033                 ret = mkdir(path, 0777);
1034                 if (ret && errno != EEXIST)
1035                         return -errno;
1036                 path[i] = '/';
1037         }
1038
1039         return 0;
1040 }
1041
1042 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1043                                   struct btrfs_ioctl_dev_info_args *di_args,
1044                                   struct scrub_file_record **past_scrubs)
1045 {
1046         int i;
1047
1048         if (!fi_args || !di_args || !past_scrubs)
1049                 return 0;
1050
1051         for (i = 0; i < fi_args->num_devices; i++) {
1052                 struct scrub_file_record *sfr =
1053                         last_dev_scrub(past_scrubs, di_args[i].devid);
1054
1055                 if (!sfr)
1056                         continue;
1057                 if (!(sfr->stats.finished || sfr->stats.canceled))
1058                         return 1;
1059         }
1060         return 0;
1061 }
1062
1063 static const char * const cmd_scrub_start_usage[];
1064 static const char * const cmd_scrub_resume_usage[];
1065
1066 static int scrub_start(int argc, char **argv, int resume)
1067 {
1068         int fdmnt;
1069         int prg_fd = -1;
1070         int fdres = -1;
1071         int ret;
1072         pid_t pid;
1073         int c;
1074         int i;
1075         int err = 0;
1076         int e_uncorrectable = 0;
1077         int e_correctable = 0;
1078         int print_raw = 0;
1079         char *path;
1080         int do_background = 1;
1081         int do_wait = 0;
1082         int do_print = 0;
1083         int do_quiet = 0;
1084         int do_record = 1;
1085         int readonly = 0;
1086         int do_stats_per_dev = 0;
1087         int ioprio_class = IOPRIO_CLASS_IDLE;
1088         int ioprio_classdata = 0;
1089         int n_start = 0;
1090         int n_skip = 0;
1091         int n_resume = 0;
1092         struct btrfs_ioctl_fs_info_args fi_args;
1093         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1094         struct scrub_progress *sp = NULL;
1095         struct scrub_fs_stat fs_stat;
1096         struct timeval tv;
1097         struct sockaddr_un addr = {
1098                 .sun_family = AF_UNIX,
1099         };
1100         pthread_t *t_devs = NULL;
1101         pthread_t t_prog;
1102         struct scrub_file_record **past_scrubs = NULL;
1103         struct scrub_file_record *last_scrub = NULL;
1104         char *datafile = strdup(SCRUB_DATA_FILE);
1105         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1106         char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1107         struct scrub_progress_cycle spc;
1108         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1109         void *terr;
1110         u64 devid;
1111         DIR *dirstream = NULL;
1112         int force = 0;
1113         int nothing_to_resume = 0;
1114
1115         optind = 1;
1116         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1117                 switch (c) {
1118                 case 'B':
1119                         do_background = 0;
1120                         do_wait = 1;
1121                         do_print = 1;
1122                         break;
1123                 case 'd':
1124                         do_stats_per_dev = 1;
1125                         break;
1126                 case 'q':
1127                         do_quiet = 1;
1128                         break;
1129                 case 'r':
1130                         readonly = 1;
1131                         break;
1132                 case 'R':
1133                         print_raw = 1;
1134                         break;
1135                 case 'c':
1136                         ioprio_class = (int)strtol(optarg, NULL, 10);
1137                         break;
1138                 case 'n':
1139                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1140                         break;
1141                 case 'f':
1142                         force = 1;
1143                         break;
1144                 case '?':
1145                 default:
1146                         usage(resume ? cmd_scrub_resume_usage :
1147                                                 cmd_scrub_start_usage);
1148                 }
1149         }
1150
1151         /* try to catch most error cases before forking */
1152
1153         if (check_argc_exact(argc - optind, 1)) {
1154                 usage(resume ? cmd_scrub_resume_usage :
1155                                         cmd_scrub_start_usage);
1156         }
1157
1158         spc.progress = NULL;
1159         if (do_quiet && do_print)
1160                 do_print = 0;
1161
1162         if (mkdir_p(datafile)) {
1163                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1164                                "file, mkdir %s failed: %s. Status recording "
1165                                "disabled\n", datafile, strerror(errno));
1166                 do_record = 0;
1167         }
1168         free(datafile);
1169
1170         path = argv[optind];
1171
1172         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1173
1174         if (fdmnt < 0) {
1175                 if (errno == EINVAL)
1176                         ERR(!do_quiet,
1177                             "ERROR: '%s' is not a mounted btrfs device\n",
1178                             path);
1179                 else
1180                         ERR(!do_quiet, "ERROR: can't access '%s': %s\n",
1181                             path, strerror(errno));
1182                 return 1;
1183         }
1184
1185         ret = get_fs_info(path, &fi_args, &di_args);
1186         if (ret) {
1187                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1188                     "%s\n", strerror(-ret));
1189                 err = 1;
1190                 goto out;
1191         }
1192         if (!fi_args.num_devices) {
1193                 ERR(!do_quiet, "ERROR: no devices found\n");
1194                 err = 1;
1195                 goto out;
1196         }
1197
1198         uuid_unparse(fi_args.fsid, fsid);
1199         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1200         if (fdres < 0 && fdres != -ENOENT) {
1201                 ERR(!do_quiet, "WARNING: failed to open status file: "
1202                     "%s\n", strerror(-fdres));
1203         } else if (fdres >= 0) {
1204                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1205                 if (IS_ERR(past_scrubs))
1206                         ERR(!do_quiet, "WARNING: failed to read status file: "
1207                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1208                 close(fdres);
1209         }
1210
1211         /*
1212          * check whether any involved device is already busy running a
1213          * scrub. This would cause damaged status messages and the state
1214          * "aborted" without the explanation that a scrub was already
1215          * running. Therefore check it first, prevent it and give some
1216          * feedback to the user if scrub is already running.
1217          * Note that if scrub is started with a block device as the
1218          * parameter, only that particular block device is checked. It
1219          * is a normal mode of operation to start scrub on multiple
1220          * single devices, there is no reason to prevent this.
1221          */
1222         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1223                 ERR(!do_quiet,
1224                     "ERROR: scrub is already running.\n"
1225                     "To cancel use 'btrfs scrub cancel %s'.\n"
1226                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1227                     path, path);
1228                 err = 1;
1229                 goto out;
1230         }
1231
1232         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1233         sp = calloc(fi_args.num_devices, sizeof(*sp));
1234         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1235
1236         if (!t_devs || !sp || !spc.progress) {
1237                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1238                 err = 1;
1239                 goto out;
1240         }
1241
1242         for (i = 0; i < fi_args.num_devices; ++i) {
1243                 devid = di_args[i].devid;
1244                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1245                 if (ret) {
1246                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1247                             "%s\n", strerror(ret));
1248                         err = 1;
1249                         goto out;
1250                 }
1251                 last_scrub = last_dev_scrub(past_scrubs, devid);
1252                 sp[i].scrub_args.devid = devid;
1253                 sp[i].fd = fdmnt;
1254                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1255                                              !last_scrub->stats.finished)) {
1256                         ++n_resume;
1257                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1258                         sp[i].resumed = last_scrub;
1259                 } else if (resume) {
1260                         ++n_skip;
1261                         sp[i].skip = 1;
1262                         sp[i].resumed = last_scrub;
1263                         continue;
1264                 } else {
1265                         ++n_start;
1266                         sp[i].scrub_args.start = 0ll;
1267                         sp[i].resumed = NULL;
1268                 }
1269                 sp[i].skip = 0;
1270                 sp[i].scrub_args.end = (u64)-1ll;
1271                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1272                 sp[i].ioprio_class = ioprio_class;
1273                 sp[i].ioprio_classdata = ioprio_classdata;
1274         }
1275
1276         if (!n_start && !n_resume) {
1277                 if (!do_quiet)
1278                         printf("scrub: nothing to resume for %s, fsid %s\n",
1279                                path, fsid);
1280                 nothing_to_resume = 1;
1281                 goto out;
1282         }
1283
1284         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1285         while (ret != -1) {
1286                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1287                                         sock_path, sizeof(sock_path));
1288                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1289                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1290                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1291                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1292                 if (ret != -1 || errno != EADDRINUSE)
1293                         break;
1294                 /*
1295                  * bind failed with EADDRINUSE. so let's see if anyone answers
1296                  * when we make a call to the socket ...
1297                  */
1298                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1299                 if (!ret || errno != ECONNREFUSED) {
1300                         /* ... yes, so scrub must be running. error out */
1301                         fprintf(stderr, "ERROR: scrub already running\n");
1302                         close(prg_fd);
1303                         prg_fd = -1;
1304                         goto out;
1305                 }
1306                 /*
1307                  * ... no, this means someone left us alone with an unused
1308                  * socket in the file system. remove it and try again.
1309                  */
1310                 ret = unlink(sock_path);
1311         }
1312         if (ret != -1)
1313                 ret = listen(prg_fd, 100);
1314         if (ret == -1) {
1315                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1316                     "socket at %s: %s. Progress cannot be queried\n",
1317                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1318                     strerror(errno));
1319                 if (prg_fd != -1) {
1320                         close(prg_fd);
1321                         prg_fd = -1;
1322                         if (sock_path[0])
1323                                 unlink(sock_path);
1324                 }
1325         }
1326
1327         if (do_record) {
1328                 /* write all-zero progress file for a start */
1329                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1330                                            fi_args.num_devices);
1331                 if (ret) {
1332                         ERR(!do_quiet, "WARNING: failed to write the progress "
1333                             "status file: %s. Status recording disabled\n",
1334                             strerror(-ret));
1335                         do_record = 0;
1336                 }
1337         }
1338
1339         if (do_background) {
1340                 pid = fork();
1341                 if (pid == -1) {
1342                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1343                                         "%s\n", strerror(errno));
1344                         err = 1;
1345                         goto out;
1346                 }
1347
1348                 if (pid) {
1349                         int stat;
1350                         scrub_handle_sigint_parent();
1351                         if (!do_quiet)
1352                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1353                                        n_start ? "started" : "resumed",
1354                                        path, fsid, pid);
1355                         if (!do_wait) {
1356                                 err = 0;
1357                                 goto out;
1358                         }
1359                         ret = wait(&stat);
1360                         if (ret != pid) {
1361                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1362                                     "%s\n", ret, strerror(errno));
1363                                 err = 1;
1364                                 goto out;
1365                         }
1366                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1367                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1368                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1369                                 goto out;
1370                         }
1371                         err = 0;
1372                         goto out;
1373                 }
1374         }
1375
1376         scrub_handle_sigint_child(fdmnt);
1377
1378         for (i = 0; i < fi_args.num_devices; ++i) {
1379                 if (sp[i].skip) {
1380                         sp[i].scrub_args.progress = sp[i].resumed->p;
1381                         sp[i].stats = sp[i].resumed->stats;
1382                         sp[i].ret = 0;
1383                         sp[i].stats.finished = 1;
1384                         continue;
1385                 }
1386                 devid = di_args[i].devid;
1387                 gettimeofday(&tv, NULL);
1388                 sp[i].stats.t_start = tv.tv_sec;
1389                 ret = pthread_create(&t_devs[i], NULL,
1390                                         scrub_one_dev, &sp[i]);
1391                 if (ret) {
1392                         if (do_print)
1393                                 fprintf(stderr, "ERROR: creating "
1394                                         "scrub_one_dev[%llu] thread failed: "
1395                                         "%s\n", devid, strerror(ret));
1396                         err = 1;
1397                         goto out;
1398                 }
1399         }
1400
1401         spc.fdmnt = fdmnt;
1402         spc.prg_fd = prg_fd;
1403         spc.do_record = do_record;
1404         spc.write_mutex = &spc_write_mutex;
1405         spc.shared_progress = sp;
1406         spc.fi = &fi_args;
1407         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1408         if (ret) {
1409                 if (do_print)
1410                         fprintf(stderr, "ERROR: creating progress thread "
1411                                 "failed: %s\n", strerror(ret));
1412                 err = 1;
1413                 goto out;
1414         }
1415
1416         err = 0;
1417         for (i = 0; i < fi_args.num_devices; ++i) {
1418                 if (sp[i].skip)
1419                         continue;
1420                 devid = di_args[i].devid;
1421                 ret = pthread_join(t_devs[i], NULL);
1422                 if (ret) {
1423                         if (do_print)
1424                                 fprintf(stderr, "ERROR: pthread_join failed "
1425                                         "for scrub_one_dev[%llu]: %s\n", devid,
1426                                         strerror(ret));
1427                         ++err;
1428                         continue;
1429                 }
1430                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1431                         if (do_print)
1432                                 fprintf(stderr, "WARNING: device %lld not "
1433                                         "present\n", devid);
1434                         continue;
1435                 }
1436                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1437                         ++err;
1438                 } else if (sp[i].ret) {
1439                         if (do_print)
1440                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1441                                         "for device id %lld (%s)\n", path,
1442                                         devid, strerror(sp[i].ioctl_errno));
1443                         ++err;
1444                         continue;
1445                 }
1446                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1447                         e_uncorrectable++;
1448                 if (sp[i].scrub_args.progress.corrected_errors > 0
1449                     || sp[i].scrub_args.progress.unverified_errors > 0)
1450                         e_correctable++;
1451         }
1452
1453         if (do_print) {
1454                 const char *append = "done";
1455                 if (!do_stats_per_dev)
1456                         init_fs_stat(&fs_stat);
1457                 for (i = 0; i < fi_args.num_devices; ++i) {
1458                         if (do_stats_per_dev) {
1459                                 print_scrub_dev(&di_args[i],
1460                                                 &sp[i].scrub_args.progress,
1461                                                 print_raw,
1462                                                 sp[i].ret ? "canceled" : "done",
1463                                                 &sp[i].stats);
1464                         } else {
1465                                 if (sp[i].ret)
1466                                         append = "canceled";
1467                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1468                                                 &sp[i].stats, &fs_stat);
1469                         }
1470                 }
1471                 if (!do_stats_per_dev) {
1472                         printf("scrub %s for %s\n", append, fsid);
1473                         print_fs_stat(&fs_stat, print_raw);
1474                 }
1475         }
1476
1477         ret = pthread_cancel(t_prog);
1478         if (!ret)
1479                 ret = pthread_join(t_prog, &terr);
1480
1481         /* check for errors from the handling of the progress thread */
1482         if (do_print && ret) {
1483                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1484                         strerror(ret));
1485         }
1486
1487         /* check for errors returned from the progress thread itself */
1488         if (do_print && terr && terr != PTHREAD_CANCELED) {
1489                 fprintf(stderr, "ERROR: recording progress "
1490                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1491         }
1492
1493         if (do_record) {
1494                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1495                                            fi_args.num_devices);
1496                 if (ret && do_print) {
1497                         fprintf(stderr, "ERROR: failed to record the result: "
1498                                 "%s\n", strerror(-ret));
1499                 }
1500         }
1501
1502         scrub_handle_sigint_child(-1);
1503
1504 out:
1505         free_history(past_scrubs);
1506         free(di_args);
1507         free(t_devs);
1508         free(sp);
1509         free(spc.progress);
1510         if (prg_fd > -1) {
1511                 close(prg_fd);
1512                 if (sock_path[0])
1513                         unlink(sock_path);
1514         }
1515         close_file_or_dir(fdmnt, dirstream);
1516
1517         if (nothing_to_resume)
1518                 return 2;
1519         if (err)
1520                 return 1;
1521         if (e_correctable)
1522                 return 3;
1523         if (e_uncorrectable)
1524                 return 4;
1525         return 0;
1526 }
1527
1528 static const char * const cmd_scrub_start_usage[] = {
1529         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1530         "Start a new scrub",
1531         "",
1532         "-B     do not background",
1533         "-d     stats per device (-B only)",
1534         "-q     be quiet",
1535         "-r     read only mode",
1536         "-R     raw print mode, print full data instead of summary"
1537         "-c     set ioprio class (see ionice(1) manpage)",
1538         "-n     set ioprio classdata (see ionice(1) manpage)",
1539         "-f     force to skip checking whether scrub has started/resumed in userspace ",
1540         "       this is useful when scrub stats record file is damaged",
1541         NULL
1542 };
1543
1544 static int cmd_scrub_start(int argc, char **argv)
1545 {
1546         return scrub_start(argc, argv, 0);
1547 }
1548
1549 static const char * const cmd_scrub_cancel_usage[] = {
1550         "btrfs scrub cancel <path>|<device>",
1551         "Cancel a running scrub",
1552         NULL
1553 };
1554
1555 static int cmd_scrub_cancel(int argc, char **argv)
1556 {
1557         char *path;
1558         int ret;
1559         int fdmnt = -1;
1560         DIR *dirstream = NULL;
1561
1562         if (check_argc_exact(argc, 2))
1563                 usage(cmd_scrub_cancel_usage);
1564
1565         path = argv[1];
1566
1567         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1568         if (fdmnt < 0) {
1569                 if (errno == EINVAL)
1570                         fprintf(stderr,
1571                                 "ERROR: '%s' is not a mounted btrfs device\n",
1572                                 path);
1573                 else
1574                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1575                                 path, strerror(errno));
1576                 ret = 1;
1577                 goto out;
1578         }
1579
1580         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1581
1582         if (ret < 0) {
1583                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1584                         errno == ENOTCONN ? "not running" : strerror(errno));
1585                 if (errno == ENOTCONN)
1586                         ret = 2;
1587                 else
1588                         ret = 1;
1589                 goto out;
1590         }
1591
1592         ret = 0;
1593         printf("scrub cancelled\n");
1594
1595 out:
1596         close_file_or_dir(fdmnt, dirstream);
1597         return ret;
1598 }
1599
1600 static const char * const cmd_scrub_resume_usage[] = {
1601         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1602         "Resume previously canceled or interrupted scrub",
1603         "",
1604         "-B     do not background",
1605         "-d     stats per device (-B only)",
1606         "-q     be quiet",
1607         "-r     read only mode",
1608         "-c     set ioprio class (see ionice(1) manpage)",
1609         "-n     set ioprio classdata (see ionice(1) manpage)",
1610         NULL
1611 };
1612
1613 static int cmd_scrub_resume(int argc, char **argv)
1614 {
1615         return scrub_start(argc, argv, 1);
1616 }
1617
1618 static const char * const cmd_scrub_status_usage[] = {
1619         "btrfs scrub status [-dR] <path>|<device>",
1620         "Show status of running or finished scrub",
1621         "",
1622         "-d     stats per device",
1623         "-R     print raw stats",
1624         NULL
1625 };
1626
1627 static int cmd_scrub_status(int argc, char **argv)
1628 {
1629         char *path;
1630         struct btrfs_ioctl_fs_info_args fi_args;
1631         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1632         struct scrub_file_record **past_scrubs = NULL;
1633         struct scrub_file_record *last_scrub;
1634         struct scrub_fs_stat fs_stat;
1635         struct sockaddr_un addr = {
1636                 .sun_family = AF_UNIX,
1637         };
1638         int ret;
1639         int i;
1640         int fdmnt;
1641         int print_raw = 0;
1642         int do_stats_per_dev = 0;
1643         int c;
1644         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1645         int fdres = -1;
1646         int err = 0;
1647         DIR *dirstream = NULL;
1648
1649         optind = 1;
1650         while ((c = getopt(argc, argv, "dR")) != -1) {
1651                 switch (c) {
1652                 case 'd':
1653                         do_stats_per_dev = 1;
1654                         break;
1655                 case 'R':
1656                         print_raw = 1;
1657                         break;
1658                 case '?':
1659                 default:
1660                         usage(cmd_scrub_status_usage);
1661                 }
1662         }
1663
1664         if (check_argc_exact(argc - optind, 1))
1665                 usage(cmd_scrub_status_usage);
1666
1667         path = argv[optind];
1668
1669         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1670
1671         if (fdmnt < 0) {
1672                 if (errno == EINVAL)
1673                         fprintf(stderr,
1674                                 "ERROR: '%s' is not a mounted btrfs device\n",
1675                                 path);
1676                 else
1677                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1678                                 path, strerror(errno));
1679                 return 1;
1680         }
1681
1682         ret = get_fs_info(path, &fi_args, &di_args);
1683         if (ret) {
1684                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1685                                 "%s\n", strerror(-ret));
1686                 err = 1;
1687                 goto out;
1688         }
1689         if (!fi_args.num_devices) {
1690                 fprintf(stderr, "ERROR: no devices found\n");
1691                 err = 1;
1692                 goto out;
1693         }
1694
1695         uuid_unparse(fi_args.fsid, fsid);
1696
1697         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1698         if (fdres == -1) {
1699                 fprintf(stderr, "ERROR: failed to create socket to "
1700                         "receive progress information: %s\n",
1701                         strerror(errno));
1702                 err = 1;
1703                 goto out;
1704         }
1705         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1706                         NULL, addr.sun_path, sizeof(addr.sun_path));
1707         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1708         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1709         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1710         if (ret == -1) {
1711                 close(fdres);
1712                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1713                 if (fdres < 0 && fdres != -ENOENT) {
1714                         fprintf(stderr, "WARNING: failed to open status file: "
1715                                 "%s\n", strerror(-fdres));
1716                         err = 1;
1717                         goto out;
1718                 }
1719         }
1720
1721         if (fdres >= 0) {
1722                 past_scrubs = scrub_read_file(fdres, 1);
1723                 if (IS_ERR(past_scrubs))
1724                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1725                                 strerror(-PTR_ERR(past_scrubs)));
1726         }
1727
1728         printf("scrub status for %s\n", fsid);
1729
1730         if (do_stats_per_dev) {
1731                 for (i = 0; i < fi_args.num_devices; ++i) {
1732                         last_scrub = last_dev_scrub(past_scrubs,
1733                                                         di_args[i].devid);
1734                         if (!last_scrub) {
1735                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1736                                                 NULL, NULL);
1737                                 continue;
1738                         }
1739                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1740                                         last_scrub->stats.finished ?
1741                                                         "history" : "status",
1742                                         &last_scrub->stats);
1743                 }
1744         } else {
1745                 init_fs_stat(&fs_stat);
1746                 for (i = 0; i < fi_args.num_devices; ++i) {
1747                         last_scrub = last_dev_scrub(past_scrubs,
1748                                                         di_args[i].devid);
1749                         if (!last_scrub)
1750                                 continue;
1751                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1752                                         &fs_stat);
1753                 }
1754                 print_fs_stat(&fs_stat, print_raw);
1755         }
1756
1757 out:
1758         free_history(past_scrubs);
1759         free(di_args);
1760         if (fdres > -1)
1761                 close(fdres);
1762         close_file_or_dir(fdmnt, dirstream);
1763
1764         return !!err;
1765 }
1766
1767 const struct cmd_group scrub_cmd_group = {
1768         scrub_cmd_group_usage, NULL, {
1769                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1770                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1771                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1772                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1773                 NULL_CMD_STRUCT
1774         }
1775 };
1776
1777 int cmd_scrub(int argc, char **argv)
1778 {
1779         return handle_command_group(&scrub_cmd_group, argc, argv);
1780 }