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_mutex_lock(m);
 777         if (ret) {
 778                 err = -ret;
 779                 goto fail;
 780         }
 781
 782         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 783         if (ret) {
 784                 err = -ret;
 785                 goto out;
 786         }
 787
 788         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 789         if (fd < 0) {
 790                 err = fd;
 791                 goto out;
 792         }
 793         err = scrub_write_file(fd, fsid, data, n);
 794         if (err)
 795                 goto out;
 796         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 797         if (err)
 798                 goto out;
 799
 800 out:
 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 fail:
 812         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 813         if (ret && !err)
 814                 err = -ret;
 815
 816         return err;
 817 }
 818
 819 static void *scrub_one_dev(void *ctx)
 820 {
 821         struct scrub_progress *sp = ctx;
 822         int ret;
 823         struct timeval tv;
 824
 825         sp->stats.canceled = 0;
 826         sp->stats.duration = 0;
 827         sp->stats.finished = 0;
 828
 829         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 830                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 831                                         sp->ioprio_classdata));
 832         if (ret)
 833                 fprintf(stderr,
 834                         "WARNING: setting ioprio failed: %s (ignored).\n",
 835                         strerror(errno));
 836
 837         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 838         gettimeofday(&tv, NULL);
 839         sp->ret = ret;
 840         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 841         sp->stats.canceled = !!ret;
 842         sp->ioctl_errno = errno;
 843         ret = pthread_mutex_lock(&sp->progress_mutex);
 844         if (ret)
 845                 return ERR_PTR(-ret);
 846         sp->stats.finished = 1;
 847         ret = pthread_mutex_unlock(&sp->progress_mutex);
 848         if (ret)
 849                 return ERR_PTR(-ret);
 850
 851         return NULL;
 852 }
 853
 854 static void *progress_one_dev(void *ctx)
 855 {
 856         struct scrub_progress *sp = ctx;
 857
 858         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 859         sp->ioctl_errno = errno;
 860
 861         return NULL;
 862 }
 863
 864 /* nb: returns a negative errno via ERR_PTR */
 865 static void *scrub_progress_cycle(void *ctx)
 866 {
 867         int ret;
 868         int  perr = 0;  /* positive / pthread error returns */
 869         int old;
 870         int i;
 871         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 872         struct scrub_progress *sp;
 873         struct scrub_progress *sp_last;
 874         struct scrub_progress *sp_shared;
 875         struct timeval tv;
 876         struct scrub_progress_cycle *spc = ctx;
 877         int ndev = spc->fi->num_devices;
 878         int this = 1;
 879         int last = 0;
 880         int peer_fd = -1;
 881         struct pollfd accept_poll_fd = {
 882                 .fd = spc->prg_fd,
 883                 .events = POLLIN,
 884                 .revents = 0,
 885         };
 886         struct pollfd write_poll_fd = {
 887                 .events = POLLOUT,
 888                 .revents = 0,
 889         };
 890         struct sockaddr_un peer;
 891         socklen_t peer_size = sizeof(peer);
 892
 893         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 894         if (perr)
 895                 goto out;
 896
 897         uuid_unparse(spc->fi->fsid, fsid);
 898
 899         for (i = 0; i < ndev; ++i) {
 900                 sp = &spc->progress[i];
 901                 sp_last = &spc->progress[i + ndev];
 902                 sp_shared = &spc->shared_progress[i];
 903                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 904                                                 sp_shared->scrub_args.devid;
 905                 sp->fd = sp_last->fd = spc->fdmnt;
 906                 sp->stats.t_start = sp_last->stats.t_start =
 907                                                 sp_shared->stats.t_start;
 908                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 909                 sp->skip = sp_last->skip = sp_shared->skip;
 910                 sp->stats.finished = sp_last->stats.finished =
 911                                                 sp_shared->stats.finished;
 912         }
 913
 914         while (1) {
 915                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 916                 if (ret == -1) {
 917                         ret = -errno;
 918                         goto out;
 919                 }
 920                 if (ret)
 921                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 922                                          &peer_size);
 923                 gettimeofday(&tv, NULL);
 924                 this = (this + 1)%2;
 925                 last = (last + 1)%2;
 926                 for (i = 0; i < ndev; ++i) {
 927                         sp = &spc->progress[this * ndev + i];
 928                         sp_last = &spc->progress[last * ndev + i];
 929                         sp_shared = &spc->shared_progress[i];
 930                         if (sp->stats.finished)
 931                                 continue;
 932                         progress_one_dev(sp);
 933                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 934                         if (!sp->ret)
 935                                 continue;
 936                         if (sp->ioctl_errno != ENOTCONN &&
 937                             sp->ioctl_errno != ENODEV) {
 938                                 ret = -sp->ioctl_errno;
 939                                 goto out;
 940                         }
 941                         /*
 942                          * scrub finished or device removed, check the
 943                          * finished flag. if unset, just use the last
 944                          * result we got for the current write and go
 945                          * on. flag should be set on next cycle, then.
 946                          */
 947                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 948                         if (perr)
 949                                 goto out;
 950                         if (!sp_shared->stats.finished) {
 951                                 perr = pthread_mutex_unlock(
 952                                                 &sp_shared->progress_mutex);
 953                                 if (perr)
 954                                         goto out;
 955                                 memcpy(sp, sp_last, sizeof(*sp));
 956                                 continue;
 957                         }
 958                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 959                         if (perr)
 960                                 goto out;
 961                         memcpy(sp, sp_shared, sizeof(*sp));
 962                         memcpy(sp_last, sp_shared, sizeof(*sp));
 963                 }
 964                 if (peer_fd != -1) {
 965                         write_poll_fd.fd = peer_fd;
 966                         ret = poll(&write_poll_fd, 1, 0);
 967                         if (ret == -1) {
 968                                 ret = -errno;
 969                                 goto out;
 970                         }
 971                         if (ret) {
 972                                 ret = scrub_write_file(
 973                                         peer_fd, fsid,
 974                                         &spc->progress[this * ndev], ndev);
 975                                 if (ret)
 976                                         goto out;
 977                         }
 978                         close(peer_fd);
 979                         peer_fd = -1;
 980                 }
 981                 if (!spc->do_record)
 982                         continue;
 983                 ret = scrub_write_progress(spc->write_mutex, fsid,
 984                                            &spc->progress[this * ndev], ndev);
 985                 if (ret)
 986                         goto out;
 987         }
 988 out:
 989         if (peer_fd != -1)
 990                 close(peer_fd);
 991         if (perr)
 992                 ret = -perr;
 993         return ERR_PTR(ret);
 994 }
 995
 996 static struct scrub_file_record *last_dev_scrub(
 997                 struct scrub_file_record *const *const past_scrubs, u64 devid)
 998 {
 999         int i;
1000
1001         if (!past_scrubs || IS_ERR(past_scrubs))
1002                 return NULL;
1003
1004         for (i = 0; past_scrubs[i]; ++i)
1005                 if (past_scrubs[i]->devid == devid)
1006                         return past_scrubs[i];
1007
1008         return NULL;
1009 }
1010
1011 static int mkdir_p(char *path)
1012 {
1013         int i;
1014         int ret;
1015
1016         for (i = 1; i < strlen(path); ++i) {
1017                 if (path[i] != '/')
1018                         continue;
1019                 path[i] = '\0';
1020                 ret = mkdir(path, 0777);
1021                 if (ret && errno != EEXIST)
1022                         return -errno;
1023                 path[i] = '/';
1024         }
1025
1026         return 0;
1027 }
1028
1029 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1030                                   struct btrfs_ioctl_dev_info_args *di_args,
1031                                   struct scrub_file_record **past_scrubs)
1032 {
1033         int i;
1034
1035         if (!fi_args || !di_args || !past_scrubs)
1036                 return 0;
1037
1038         for (i = 0; i < fi_args->num_devices; i++) {
1039                 struct scrub_file_record *sfr =
1040                         last_dev_scrub(past_scrubs, di_args[i].devid);
1041
1042                 if (!sfr)
1043                         continue;
1044                 if (!(sfr->stats.finished || sfr->stats.canceled))
1045                         return 1;
1046         }
1047         return 0;
1048 }
1049
1050 static const char * const cmd_scrub_start_usage[];
1051 static const char * const cmd_scrub_resume_usage[];
1052
1053 static int scrub_start(int argc, char **argv, int resume)
1054 {
1055         int fdmnt;
1056         int prg_fd = -1;
1057         int fdres = -1;
1058         int ret;
1059         pid_t pid;
1060         int c;
1061         int i;
1062         int err = 0;
1063         int e_uncorrectable = 0;
1064         int e_correctable = 0;
1065         int print_raw = 0;
1066         char *path;
1067         int do_background = 1;
1068         int do_wait = 0;
1069         int do_print = 0;
1070         int do_quiet = 0;
1071         int do_record = 1;
1072         int readonly = 0;
1073         int do_stats_per_dev = 0;
1074         int ioprio_class = IOPRIO_CLASS_IDLE;
1075         int ioprio_classdata = 0;
1076         int n_start = 0;
1077         int n_skip = 0;
1078         int n_resume = 0;
1079         struct btrfs_ioctl_fs_info_args fi_args;
1080         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1081         struct scrub_progress *sp = NULL;
1082         struct scrub_fs_stat fs_stat;
1083         struct timeval tv;
1084         struct sockaddr_un addr = {
1085                 .sun_family = AF_UNIX,
1086         };
1087         pthread_t *t_devs = NULL;
1088         pthread_t t_prog;
1089         struct scrub_file_record **past_scrubs = NULL;
1090         struct scrub_file_record *last_scrub = NULL;
1091         char *datafile = strdup(SCRUB_DATA_FILE);
1092         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1093         char sock_path[BTRFS_PATH_NAME_MAX + 1] = "";
1094         struct scrub_progress_cycle spc;
1095         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1096         void *terr;
1097         u64 devid;
1098         DIR *dirstream = NULL;
1099         int force = 0;
1100         int nothing_to_resume = 0;
1101
1102         optind = 1;
1103         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1104                 switch (c) {
1105                 case 'B':
1106                         do_background = 0;
1107                         do_wait = 1;
1108                         do_print = 1;
1109                         break;
1110                 case 'd':
1111                         do_stats_per_dev = 1;
1112                         break;
1113                 case 'q':
1114                         do_quiet = 1;
1115                         break;
1116                 case 'r':
1117                         readonly = 1;
1118                         break;
1119                 case 'R':
1120                         print_raw = 1;
1121                         break;
1122                 case 'c':
1123                         ioprio_class = (int)strtol(optarg, NULL, 10);
1124                         break;
1125                 case 'n':
1126                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1127                         break;
1128                 case 'f':
1129                         force = 1;
1130                         break;
1131                 case '?':
1132                 default:
1133                         usage(resume ? cmd_scrub_resume_usage :
1134                                                 cmd_scrub_start_usage);
1135                 }
1136         }
1137
1138         /* try to catch most error cases before forking */
1139
1140         if (check_argc_exact(argc - optind, 1)) {
1141                 usage(resume ? cmd_scrub_resume_usage :
1142                                         cmd_scrub_start_usage);
1143         }
1144
1145         spc.progress = NULL;
1146         if (do_quiet && do_print)
1147                 do_print = 0;
1148
1149         if (mkdir_p(datafile)) {
1150                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1151                                "file, mkdir %s failed: %s. Status recording "
1152                                "disabled\n", datafile, strerror(errno));
1153                 do_record = 0;
1154         }
1155         free(datafile);
1156
1157         path = argv[optind];
1158
1159         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1160
1161         if (fdmnt < 0) {
1162                 ERR(!do_quiet, "ERROR: can't access '%s'\n", path);
1163                 return 1;
1164         }
1165
1166         ret = get_fs_info(path, &fi_args, &di_args);
1167         if (ret) {
1168                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1169                     "%s\n", strerror(-ret));
1170                 err = 1;
1171                 goto out;
1172         }
1173         if (!fi_args.num_devices) {
1174                 ERR(!do_quiet, "ERROR: no devices found\n");
1175                 err = 1;
1176                 goto out;
1177         }
1178
1179         uuid_unparse(fi_args.fsid, fsid);
1180         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1181         if (fdres < 0 && fdres != -ENOENT) {
1182                 ERR(!do_quiet, "WARNING: failed to open status file: "
1183                     "%s\n", strerror(-fdres));
1184         } else if (fdres >= 0) {
1185                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1186                 if (IS_ERR(past_scrubs))
1187                         ERR(!do_quiet, "WARNING: failed to read status file: "
1188                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1189                 close(fdres);
1190         }
1191
1192         /*
1193          * check whether any involved device is already busy running a
1194          * scrub. This would cause damaged status messages and the state
1195          * "aborted" without the explanation that a scrub was already
1196          * running. Therefore check it first, prevent it and give some
1197          * feedback to the user if scrub is already running.
1198          * Note that if scrub is started with a block device as the
1199          * parameter, only that particular block device is checked. It
1200          * is a normal mode of operation to start scrub on multiple
1201          * single devices, there is no reason to prevent this.
1202          */
1203         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1204                 ERR(!do_quiet,
1205                     "ERROR: scrub is already running.\n"
1206                     "To cancel use 'btrfs scrub cancel %s'.\n"
1207                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1208                     path, path);
1209                 err = 1;
1210                 goto out;
1211         }
1212
1213         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1214         sp = calloc(fi_args.num_devices, sizeof(*sp));
1215         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1216
1217         if (!t_devs || !sp || !spc.progress) {
1218                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1219                 err = 1;
1220                 goto out;
1221         }
1222
1223         for (i = 0; i < fi_args.num_devices; ++i) {
1224                 devid = di_args[i].devid;
1225                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1226                 if (ret) {
1227                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1228                             "%s\n", strerror(ret));
1229                         err = 1;
1230                         goto out;
1231                 }
1232                 last_scrub = last_dev_scrub(past_scrubs, devid);
1233                 sp[i].scrub_args.devid = devid;
1234                 sp[i].fd = fdmnt;
1235                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1236                                              !last_scrub->stats.finished)) {
1237                         ++n_resume;
1238                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1239                         sp[i].resumed = last_scrub;
1240                 } else if (resume) {
1241                         ++n_skip;
1242                         sp[i].skip = 1;
1243                         sp[i].resumed = last_scrub;
1244                         continue;
1245                 } else {
1246                         ++n_start;
1247                         sp[i].scrub_args.start = 0ll;
1248                         sp[i].resumed = NULL;
1249                 }
1250                 sp[i].skip = 0;
1251                 sp[i].scrub_args.end = (u64)-1ll;
1252                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1253                 sp[i].ioprio_class = ioprio_class;
1254                 sp[i].ioprio_classdata = ioprio_classdata;
1255         }
1256
1257         if (!n_start && !n_resume) {
1258                 if (!do_quiet)
1259                         printf("scrub: nothing to resume for %s, fsid %s\n",
1260                                path, fsid);
1261                 nothing_to_resume = 1;
1262                 goto out;
1263         }
1264
1265         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1266         while (ret != -1) {
1267                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1268                                         sock_path, sizeof(sock_path));
1269                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1270                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1271                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1272                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1273                 if (ret != -1 || errno != EADDRINUSE)
1274                         break;
1275                 /*
1276                  * bind failed with EADDRINUSE. so let's see if anyone answers
1277                  * when we make a call to the socket ...
1278                  */
1279                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1280                 if (!ret || errno != ECONNREFUSED) {
1281                         /* ... yes, so scrub must be running. error out */
1282                         fprintf(stderr, "ERROR: scrub already running\n");
1283                         close(prg_fd);
1284                         prg_fd = -1;
1285                         goto out;
1286                 }
1287                 /*
1288                  * ... no, this means someone left us alone with an unused
1289                  * socket in the file system. remove it and try again.
1290                  */
1291                 ret = unlink(sock_path);
1292         }
1293         if (ret != -1)
1294                 ret = listen(prg_fd, 100);
1295         if (ret == -1) {
1296                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1297                     "socket at %s: %s. Progress cannot be queried\n",
1298                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1299                     strerror(errno));
1300                 if (prg_fd != -1) {
1301                         close(prg_fd);
1302                         prg_fd = -1;
1303                         if (sock_path[0])
1304                                 unlink(sock_path);
1305                 }
1306         }
1307
1308         if (do_record) {
1309                 /* write all-zero progress file for a start */
1310                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1311                                            fi_args.num_devices);
1312                 if (ret) {
1313                         ERR(!do_quiet, "WARNING: failed to write the progress "
1314                             "status file: %s. Status recording disabled\n",
1315                             strerror(-ret));
1316                         do_record = 0;
1317                 }
1318         }
1319
1320         if (do_background) {
1321                 pid = fork();
1322                 if (pid == -1) {
1323                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1324                                         "%s\n", strerror(errno));
1325                         err = 1;
1326                         goto out;
1327                 }
1328
1329                 if (pid) {
1330                         int stat;
1331                         scrub_handle_sigint_parent();
1332                         if (!do_quiet)
1333                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1334                                        n_start ? "started" : "resumed",
1335                                        path, fsid, pid);
1336                         if (!do_wait) {
1337                                 err = 0;
1338                                 goto out;
1339                         }
1340                         ret = wait(&stat);
1341                         if (ret != pid) {
1342                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1343                                     "%s\n", ret, strerror(errno));
1344                                 err = 1;
1345                                 goto out;
1346                         }
1347                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1348                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1349                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1350                                 goto out;
1351                         }
1352                         err = 0;
1353                         goto out;
1354                 }
1355         }
1356
1357         scrub_handle_sigint_child(fdmnt);
1358
1359         for (i = 0; i < fi_args.num_devices; ++i) {
1360                 if (sp[i].skip) {
1361                         sp[i].scrub_args.progress = sp[i].resumed->p;
1362                         sp[i].stats = sp[i].resumed->stats;
1363                         sp[i].ret = 0;
1364                         sp[i].stats.finished = 1;
1365                         continue;
1366                 }
1367                 devid = di_args[i].devid;
1368                 gettimeofday(&tv, NULL);
1369                 sp[i].stats.t_start = tv.tv_sec;
1370                 ret = pthread_create(&t_devs[i], NULL,
1371                                         scrub_one_dev, &sp[i]);
1372                 if (ret) {
1373                         if (do_print)
1374                                 fprintf(stderr, "ERROR: creating "
1375                                         "scrub_one_dev[%llu] thread failed: "
1376                                         "%s\n", devid, strerror(ret));
1377                         err = 1;
1378                         goto out;
1379                 }
1380         }
1381
1382         spc.fdmnt = fdmnt;
1383         spc.prg_fd = prg_fd;
1384         spc.do_record = do_record;
1385         spc.write_mutex = &spc_write_mutex;
1386         spc.shared_progress = sp;
1387         spc.fi = &fi_args;
1388         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1389         if (ret) {
1390                 if (do_print)
1391                         fprintf(stderr, "ERROR: creating progress thread "
1392                                 "failed: %s\n", strerror(ret));
1393                 err = 1;
1394                 goto out;
1395         }
1396
1397         err = 0;
1398         for (i = 0; i < fi_args.num_devices; ++i) {
1399                 if (sp[i].skip)
1400                         continue;
1401                 devid = di_args[i].devid;
1402                 ret = pthread_join(t_devs[i], NULL);
1403                 if (ret) {
1404                         if (do_print)
1405                                 fprintf(stderr, "ERROR: pthread_join failed "
1406                                         "for scrub_one_dev[%llu]: %s\n", devid,
1407                                         strerror(ret));
1408                         ++err;
1409                         continue;
1410                 }
1411                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1412                         if (do_print)
1413                                 fprintf(stderr, "WARNING: device %lld not "
1414                                         "present\n", devid);
1415                         continue;
1416                 }
1417                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1418                         ++err;
1419                 } else if (sp[i].ret) {
1420                         if (do_print)
1421                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1422                                         "for device id %lld (%s)\n", path,
1423                                         devid, strerror(sp[i].ioctl_errno));
1424                         ++err;
1425                         continue;
1426                 }
1427                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1428                         e_uncorrectable++;
1429                 if (sp[i].scrub_args.progress.corrected_errors > 0
1430                     || sp[i].scrub_args.progress.unverified_errors > 0)
1431                         e_correctable++;
1432         }
1433
1434         if (do_print) {
1435                 const char *append = "done";
1436                 if (!do_stats_per_dev)
1437                         init_fs_stat(&fs_stat);
1438                 for (i = 0; i < fi_args.num_devices; ++i) {
1439                         if (do_stats_per_dev) {
1440                                 print_scrub_dev(&di_args[i],
1441                                                 &sp[i].scrub_args.progress,
1442                                                 print_raw,
1443                                                 sp[i].ret ? "canceled" : "done",
1444                                                 &sp[i].stats);
1445                         } else {
1446                                 if (sp[i].ret)
1447                                         append = "canceled";
1448                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1449                                                 &sp[i].stats, &fs_stat);
1450                         }
1451                 }
1452                 if (!do_stats_per_dev) {
1453                         printf("scrub %s for %s\n", append, fsid);
1454                         print_fs_stat(&fs_stat, print_raw);
1455                 }
1456         }
1457
1458         ret = pthread_cancel(t_prog);
1459         if (!ret)
1460                 ret = pthread_join(t_prog, &terr);
1461
1462         /* check for errors from the handling of the progress thread */
1463         if (do_print && ret) {
1464                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1465                         strerror(ret));
1466         }
1467
1468         /* check for errors returned from the progress thread itself */
1469         if (do_print && terr && terr != PTHREAD_CANCELED) {
1470                 fprintf(stderr, "ERROR: recording progress "
1471                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1472         }
1473
1474         if (do_record) {
1475                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1476                                            fi_args.num_devices);
1477                 if (ret && do_print) {
1478                         fprintf(stderr, "ERROR: failed to record the result: "
1479                                 "%s\n", strerror(-ret));
1480                 }
1481         }
1482
1483         scrub_handle_sigint_child(-1);
1484
1485 out:
1486         free_history(past_scrubs);
1487         free(di_args);
1488         free(t_devs);
1489         free(sp);
1490         free(spc.progress);
1491         if (prg_fd > -1) {
1492                 close(prg_fd);
1493                 if (sock_path[0])
1494                         unlink(sock_path);
1495         }
1496         close_file_or_dir(fdmnt, dirstream);
1497
1498         if (nothing_to_resume)
1499                 return 2;
1500         if (err)
1501                 return 1;
1502         if (e_correctable)
1503                 return 3;
1504         if (e_uncorrectable)
1505                 return 4;
1506         return 0;
1507 }
1508
1509 static const char * const cmd_scrub_start_usage[] = {
1510         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1511         "Start a new scrub",
1512         "",
1513         "-B     do not background",
1514         "-d     stats per device (-B only)",
1515         "-q     be quiet",
1516         "-r     read only mode",
1517         "-R     raw print mode, print full data instead of summary"
1518         "-c     set ioprio class (see ionice(1) manpage)",
1519         "-n     set ioprio classdata (see ionice(1) manpage)",
1520         "-f     force to skip checking whether scrub has started/resumed in userspace ",
1521         "       this is useful when scrub stats record file is damaged",
1522         NULL
1523 };
1524
1525 static int cmd_scrub_start(int argc, char **argv)
1526 {
1527         return scrub_start(argc, argv, 0);
1528 }
1529
1530 static const char * const cmd_scrub_cancel_usage[] = {
1531         "btrfs scrub cancel <path>|<device>",
1532         "Cancel a running scrub",
1533         NULL
1534 };
1535
1536 static int cmd_scrub_cancel(int argc, char **argv)
1537 {
1538         char *path;
1539         int ret;
1540         int fdmnt = -1;
1541         DIR *dirstream = NULL;
1542
1543         if (check_argc_exact(argc, 2))
1544                 usage(cmd_scrub_cancel_usage);
1545
1546         path = argv[1];
1547
1548         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1549         if (fdmnt < 0) {
1550                 fprintf(stderr, "ERROR: could not open %s: %s\n",
1551                         path, strerror(errno));
1552                 ret = 1;
1553                 goto out;
1554         }
1555
1556         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1557
1558         if (ret < 0) {
1559                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1560                         errno == ENOTCONN ? "not running" : strerror(errno));
1561                 if (errno == ENOTCONN)
1562                         ret = 2;
1563                 else
1564                         ret = 1;
1565                 goto out;
1566         }
1567
1568         ret = 0;
1569         printf("scrub cancelled\n");
1570
1571 out:
1572         close_file_or_dir(fdmnt, dirstream);
1573         return ret;
1574 }
1575
1576 static const char * const cmd_scrub_resume_usage[] = {
1577         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1578         "Resume previously canceled or interrupted scrub",
1579         "",
1580         "-B     do not background",
1581         "-d     stats per device (-B only)",
1582         "-q     be quiet",
1583         "-r     read only mode",
1584         "-c     set ioprio class (see ionice(1) manpage)",
1585         "-n     set ioprio classdata (see ionice(1) manpage)",
1586         NULL
1587 };
1588
1589 static int cmd_scrub_resume(int argc, char **argv)
1590 {
1591         return scrub_start(argc, argv, 1);
1592 }
1593
1594 static const char * const cmd_scrub_status_usage[] = {
1595         "btrfs scrub status [-dR] <path>|<device>",
1596         "Show status of running or finished scrub",
1597         "",
1598         "-d     stats per device",
1599         "-R     print raw stats",
1600         NULL
1601 };
1602
1603 static int cmd_scrub_status(int argc, char **argv)
1604 {
1605         char *path;
1606         struct btrfs_ioctl_fs_info_args fi_args;
1607         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1608         struct scrub_file_record **past_scrubs = NULL;
1609         struct scrub_file_record *last_scrub;
1610         struct scrub_fs_stat fs_stat;
1611         struct sockaddr_un addr = {
1612                 .sun_family = AF_UNIX,
1613         };
1614         int ret;
1615         int i;
1616         int fdmnt;
1617         int print_raw = 0;
1618         int do_stats_per_dev = 0;
1619         int c;
1620         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1621         int fdres = -1;
1622         int err = 0;
1623         DIR *dirstream = NULL;
1624
1625         optind = 1;
1626         while ((c = getopt(argc, argv, "dR")) != -1) {
1627                 switch (c) {
1628                 case 'd':
1629                         do_stats_per_dev = 1;
1630                         break;
1631                 case 'R':
1632                         print_raw = 1;
1633                         break;
1634                 case '?':
1635                 default:
1636                         usage(cmd_scrub_status_usage);
1637                 }
1638         }
1639
1640         if (check_argc_exact(argc - optind, 1))
1641                 usage(cmd_scrub_status_usage);
1642
1643         path = argv[optind];
1644
1645         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1646
1647         if (fdmnt < 0) {
1648                 fprintf(stderr, "ERROR: can't access '%s'\n", path);
1649                 return 1;
1650         }
1651
1652         ret = get_fs_info(path, &fi_args, &di_args);
1653         if (ret) {
1654                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1655                                 "%s\n", strerror(-ret));
1656                 err = 1;
1657                 goto out;
1658         }
1659         if (!fi_args.num_devices) {
1660                 fprintf(stderr, "ERROR: no devices found\n");
1661                 err = 1;
1662                 goto out;
1663         }
1664
1665         uuid_unparse(fi_args.fsid, fsid);
1666
1667         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1668         if (fdres == -1) {
1669                 fprintf(stderr, "ERROR: failed to create socket to "
1670                         "receive progress information: %s\n",
1671                         strerror(errno));
1672                 err = 1;
1673                 goto out;
1674         }
1675         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1676                         NULL, addr.sun_path, sizeof(addr.sun_path));
1677         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1678         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1679         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1680         if (ret == -1) {
1681                 close(fdres);
1682                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1683                 if (fdres < 0 && fdres != -ENOENT) {
1684                         fprintf(stderr, "WARNING: failed to open status file: "
1685                                 "%s\n", strerror(-fdres));
1686                         err = 1;
1687                         goto out;
1688                 }
1689         }
1690
1691         if (fdres >= 0) {
1692                 past_scrubs = scrub_read_file(fdres, 1);
1693                 if (IS_ERR(past_scrubs))
1694                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1695                                 strerror(-PTR_ERR(past_scrubs)));
1696         }
1697
1698         printf("scrub status for %s\n", fsid);
1699
1700         if (do_stats_per_dev) {
1701                 for (i = 0; i < fi_args.num_devices; ++i) {
1702                         last_scrub = last_dev_scrub(past_scrubs,
1703                                                         di_args[i].devid);
1704                         if (!last_scrub) {
1705                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1706                                                 NULL, NULL);
1707                                 continue;
1708                         }
1709                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1710                                         last_scrub->stats.finished ?
1711                                                         "history" : "status",
1712                                         &last_scrub->stats);
1713                 }
1714         } else {
1715                 init_fs_stat(&fs_stat);
1716                 for (i = 0; i < fi_args.num_devices; ++i) {
1717                         last_scrub = last_dev_scrub(past_scrubs,
1718                                                         di_args[i].devid);
1719                         if (!last_scrub)
1720                                 continue;
1721                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1722                                         &fs_stat);
1723                 }
1724                 print_fs_stat(&fs_stat, print_raw);
1725         }
1726
1727 out:
1728         free_history(past_scrubs);
1729         free(di_args);
1730         if (fdres > -1)
1731                 close(fdres);
1732         close_file_or_dir(fdmnt, dirstream);
1733
1734         return !!err;
1735 }
1736
1737 const struct cmd_group scrub_cmd_group = {
1738         scrub_cmd_group_usage, NULL, {
1739                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1740                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1741                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1742                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1743                 NULL_CMD_STRUCT
1744         }
1745 };
1746
1747 int cmd_scrub(int argc, char **argv)
1748 {
1749         return handle_command_group(&scrub_cmd_group, argc, argv);
1750 }