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 #include "androidcompat.h"
  21
  22 #include <sys/ioctl.h>
  23 #include <sys/wait.h>
  24 #include <sys/stat.h>
  25 #include <sys/types.h>
  26 #include <sys/socket.h>
  27 #include <sys/un.h>
  28 #include <sys/syscall.h>
  29 #include <poll.h>
  30 #include <sys/file.h>
  31 #include <uuid/uuid.h>
  32 #include <fcntl.h>
  33 #include <unistd.h>
  34 #include <pthread.h>
  35 #include <ctype.h>
  36 #include <signal.h>
  37 #include <stdarg.h>
  38 #include <limits.h>
  39
  40 #include "ctree.h"
  41 #include "ioctl.h"
  42 #include "utils.h"
  43 #include "volumes.h"
  44 #include "disk-io.h"
  45
  46 #include "commands.h"
  47 #include "help.h"
  48
  49 static const char * const scrub_cmd_group_usage[] = {
  50         "btrfs scrub <command> [options] <path>|<device>",
  51         NULL
  52 };
  53
  54 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  55 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  56 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  57 #define SCRUB_FILE_VERSION "1"
  58
  59 struct scrub_stats {
  60         time_t t_start;
  61         time_t t_resumed;
  62         u64 duration;
  63         u64 finished;
  64         u64 canceled;
  65         int in_progress;
  66 };
  67
  68 /* TBD: replace with #include "linux/ioprio.h" in some years */
  69 #if !defined (IOPRIO_H)
  70 #define IOPRIO_WHO_PROCESS 1
  71 #define IOPRIO_CLASS_SHIFT 13
  72 #define IOPRIO_PRIO_VALUE(class, data) \
  73                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
  74 #define IOPRIO_CLASS_IDLE 3
  75 #endif
  76
  77 struct scrub_progress {
  78         struct btrfs_ioctl_scrub_args scrub_args;
  79         int fd;
  80         int ret;
  81         int skip;
  82         struct scrub_stats stats;
  83         struct scrub_file_record *resumed;
  84         int ioctl_errno;
  85         pthread_mutex_t progress_mutex;
  86         int ioprio_class;
  87         int ioprio_classdata;
  88 };
  89
  90 struct scrub_file_record {
  91         u8 fsid[BTRFS_FSID_SIZE];
  92         u64 devid;
  93         struct scrub_stats stats;
  94         struct btrfs_scrub_progress p;
  95 };
  96
  97 struct scrub_progress_cycle {
  98         int fdmnt;
  99         int prg_fd;
 100         int do_record;
 101         struct btrfs_ioctl_fs_info_args *fi;
 102         struct scrub_progress *progress;
 103         struct scrub_progress *shared_progress;
 104         pthread_mutex_t *write_mutex;
 105 };
 106
 107 struct scrub_fs_stat {
 108         struct btrfs_scrub_progress p;
 109         struct scrub_stats s;
 110         int i;
 111 };
 112
 113 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 114 {
 115         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 116         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 117         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 118         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 119         printf("\tread_errors: %lld\n", sp->read_errors);
 120         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 121         printf("\tverify_errors: %lld\n", sp->verify_errors);
 122         printf("\tno_csum: %lld\n", sp->no_csum);
 123         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 124         printf("\tsuper_errors: %lld\n", sp->super_errors);
 125         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 126         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 127         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 128         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 129         printf("\tlast_physical: %lld\n", sp->last_physical);
 130 }
 131
 132 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 133         if (test)                               \
 134                 printf(" %s=%llu", desc, test); \
 135 } while (0)
 136
 137 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 138 {
 139         u64 err_cnt;
 140         u64 err_cnt2;
 141
 142         err_cnt = p->read_errors +
 143                         p->csum_errors +
 144                         p->verify_errors +
 145                         p->super_errors;
 146
 147         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 148
 149         if (p->malloc_errors)
 150                 printf("*** WARNING: memory allocation failed while scrubbing. "
 151                        "results may be inaccurate\n");
 152
 153         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
 154                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
 155                 max(err_cnt, err_cnt2));
 156
 157         if (err_cnt || err_cnt2) {
 158                 printf("\terror details:");
 159                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 160                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 161                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 162                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 163                 printf("\n");
 164                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 165                         "unverified errors: %llu\n", p->corrected_errors,
 166                         p->uncorrectable_errors, p->unverified_errors);
 167         }
 168 }
 169
 170 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 171         fs_stat->p.name += p->name;             \
 172 } while (0)
 173
 174 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 175 do {                                            \
 176         if (fs_stat->s.name > ss->name) {       \
 177                 fs_stat->s.name = ss->name;     \
 178         }                                       \
 179 } while (0)
 180
 181 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 182 do {                                                            \
 183         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 184                 fs_stat->s.name = ss->name;                     \
 185         }                                                       \
 186 } while (0)
 187
 188 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 189 do {                                                                    \
 190         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 191                 (fs_stat)->s.name = (ss)->name;                         \
 192         }                                                               \
 193 } while (0)
 194
 195 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 196                                 struct scrub_stats *ss,
 197                                 struct scrub_fs_stat *fs_stat)
 198 {
 199         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 200         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 201         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 202         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 203         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 204         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 205         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 206         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 207         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 208         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 209         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 210         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 211         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 212         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 213         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 214         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 215         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 216         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 217         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 218 }
 219
 220 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 221 {
 222         memset(fs_stat, 0, sizeof(*fs_stat));
 223         fs_stat->s.finished = 1;
 224 }
 225
 226 static void _print_scrub_ss(struct scrub_stats *ss)
 227 {
 228         char t[4096];
 229         struct tm tm;
 230         time_t seconds;
 231         unsigned hours;
 232
 233         if (!ss || !ss->t_start) {
 234                 printf("\tno stats available\n");
 235                 return;
 236         }
 237         if (ss->t_resumed) {
 238                 localtime_r(&ss->t_resumed, &tm);
 239                 strftime(t, sizeof(t), "%c", &tm);
 240                 t[sizeof(t) - 1] = '\0';
 241                 printf("\tscrub resumed at %s", t);
 242         } else {
 243                 localtime_r(&ss->t_start, &tm);
 244                 strftime(t, sizeof(t), "%c", &tm);
 245                 t[sizeof(t) - 1] = '\0';
 246                 printf("\tscrub started at %s", t);
 247         }
 248
 249         seconds = ss->duration;
 250         hours = ss->duration / (60 * 60);
 251         gmtime_r(&seconds, &tm);
 252         strftime(t, sizeof(t), "%M:%S", &tm);
 253         if (ss->in_progress)
 254                 printf(", running for %02u:%s\n", hours, t);
 255         else if (ss->canceled)
 256                 printf(" and was aborted after %02u:%s\n", hours, t);
 257         else if (ss->finished)
 258                 printf(" and finished after %02u:%s\n", hours, t);
 259         else
 260                 printf(", interrupted after %02u:%s, not running\n",
 261                        hours, t);
 262 }
 263
 264 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 265                                 struct btrfs_scrub_progress *p, int raw,
 266                                 const char *append, struct scrub_stats *ss)
 267 {
 268         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 269                append ? append : "");
 270
 271         _print_scrub_ss(ss);
 272
 273         if (p) {
 274                 if (raw)
 275                         print_scrub_full(p);
 276                 else
 277                         print_scrub_summary(p);
 278         }
 279 }
 280
 281 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 282 {
 283         _print_scrub_ss(&fs_stat->s);
 284
 285         if (raw)
 286                 print_scrub_full(&fs_stat->p);
 287         else
 288                 print_scrub_summary(&fs_stat->p);
 289 }
 290
 291 static void free_history(struct scrub_file_record **last_scrubs)
 292 {
 293         struct scrub_file_record **l = last_scrubs;
 294         if (!l || IS_ERR(l))
 295                 return;
 296         while (*l)
 297                 free(*l++);
 298         free(last_scrubs);
 299 }
 300
 301 /*
 302  * cancels a running scrub and makes the master process record the current
 303  * progress status before exiting.
 304  */
 305 static int cancel_fd = -1;
 306 static void scrub_sigint_record_progress(int signal)
 307 {
 308         int ret;
 309
 310         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 311         if (ret < 0)
 312                 perror("Scrub cancel failed");
 313 }
 314
 315 static int scrub_handle_sigint_parent(void)
 316 {
 317         struct sigaction sa = {
 318                 .sa_handler = SIG_IGN,
 319                 .sa_flags = SA_RESTART,
 320         };
 321
 322         return sigaction(SIGINT, &sa, NULL);
 323 }
 324
 325 static int scrub_handle_sigint_child(int fd)
 326 {
 327         struct sigaction sa = {
 328                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 329         };
 330
 331         cancel_fd = fd;
 332         return sigaction(SIGINT, &sa, NULL);
 333 }
 334
 335 static int scrub_datafile(const char *fn_base, const char *fn_local,
 336                                 const char *fn_tmp, char *datafile, int size)
 337 {
 338         int ret;
 339         int end = size - 2;
 340
 341         datafile[end + 1] = '\0';
 342         strncpy(datafile, fn_base, end);
 343         ret = strlen(datafile);
 344
 345         if (ret + 1 > end)
 346                 return -EOVERFLOW;
 347
 348         datafile[ret] = '.';
 349         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 350         ret = strlen(datafile);
 351
 352         if (ret + 1 > end)
 353                 return -EOVERFLOW;
 354
 355         if (fn_tmp) {
 356                 datafile[ret] = '_';
 357                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 358                 ret = strlen(datafile);
 359
 360                 if (ret > end)
 361                         return -EOVERFLOW;
 362         }
 363
 364         return 0;
 365 }
 366
 367 static int scrub_open_file(const char *datafile, int m)
 368 {
 369         int fd;
 370         int ret;
 371
 372         fd = open(datafile, m, 0600);
 373         if (fd < 0)
 374                 return -errno;
 375
 376         ret = flock(fd, LOCK_EX|LOCK_NB);
 377         if (ret) {
 378                 ret = errno;
 379                 close(fd);
 380                 return -ret;
 381         }
 382
 383         return fd;
 384 }
 385
 386 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 387 {
 388         int ret;
 389         char datafile[PATH_MAX];
 390         ret = scrub_datafile(fn_base, fn_local, NULL,
 391                                 datafile, sizeof(datafile));
 392         if (ret < 0)
 393                 return ret;
 394         return scrub_open_file(datafile, O_RDONLY);
 395 }
 396
 397 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 398                                 const char *tmp)
 399 {
 400         int ret;
 401         char datafile[PATH_MAX];
 402         ret = scrub_datafile(fn_base, fn_local, tmp,
 403                                 datafile, sizeof(datafile));
 404         if (ret < 0)
 405                 return ret;
 406         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 407 }
 408
 409 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 410                                 const char *tmp)
 411 {
 412         int ret;
 413         char datafile_old[PATH_MAX];
 414         char datafile_new[PATH_MAX];
 415         ret = scrub_datafile(fn_base, fn_local, tmp,
 416                                 datafile_old, sizeof(datafile_old));
 417         if (ret < 0)
 418                 return ret;
 419         ret = scrub_datafile(fn_base, fn_local, NULL,
 420                                 datafile_new, sizeof(datafile_new));
 421         if (ret < 0)
 422                 return ret;
 423         ret = rename(datafile_old, datafile_new);
 424         return ret ? -errno : 0;
 425 }
 426
 427 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 428         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 429 }
 430
 431 /*
 432  * returns 0 if the key did not match (nothing was read)
 433  *         1 if the key did match (success)
 434  *        -1 if the key did match and an error occurred
 435  */
 436 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 437                         const char *key, u64 *dest)
 438 {
 439         int j;
 440
 441         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 442                 *i += len - 1;
 443                 if (buf[*i] != ':')
 444                         return -1;
 445                 *i += 1;
 446                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 447                         ;
 448                 if (*i + j >= avail)
 449                         return -1;
 450                 *dest = atoll(&buf[*i]);
 451                 *i += j;
 452                 return 1;
 453         }
 454
 455         return 0;
 456 }
 457
 458 #define _SCRUB_INVALID do {                                             \
 459         if (report_errors)                                              \
 460                 warning("invalid data on line %d pos "                  \
 461                         "%d state %d (near \"%.*s\") at %s:%d",         \
 462                         lineno, i, state, 20 > avail ? avail : 20,      \
 463                         l + i,  __FILE__, __LINE__);                    \
 464         goto skip;                                                      \
 465 } while (0)
 466
 467 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 468 {
 469         int avail = 0;
 470         int old_avail = 0;
 471         char l[SZ_16K];
 472         int state = 0;
 473         int curr = -1;
 474         int i = 0;
 475         int j;
 476         int ret;
 477         int eof = 0;
 478         int lineno = 0;
 479         u64 version;
 480         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 481         struct scrub_file_record **p = NULL;
 482
 483 again:
 484         old_avail = avail - i;
 485         if (old_avail < 0) {
 486                 error("scrub record file corrupted near byte %d", i);
 487                 return ERR_PTR(-EINVAL);
 488         }
 489         if (old_avail)
 490                 memmove(l, l + i, old_avail);
 491         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 492         if (avail == 0)
 493                 eof = 1;
 494         if (avail == 0 && old_avail == 0) {
 495                 if (curr >= 0 &&
 496                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 497                         p[curr] = NULL;
 498                 } else if (curr == -1) {
 499                         p = ERR_PTR(-ENODATA);
 500                 }
 501                 return p;
 502         }
 503         if (avail == -1) {
 504                 free_history(p);
 505                 return ERR_PTR(-errno);
 506         }
 507         avail += old_avail;
 508
 509         i = 0;
 510         while (i < avail) {
 511                 void *tmp;
 512
 513                 switch (state) {
 514                 case 0: /* start of file */
 515                         ret = scrub_kvread(&i,
 516                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 517                                 SCRUB_FILE_VERSION_PREFIX, &version);
 518                         if (ret != 1)
 519                                 _SCRUB_INVALID;
 520                         if (version != atoll(SCRUB_FILE_VERSION))
 521                                 return ERR_PTR(-ENOTSUP);
 522                         state = 6;
 523                         continue;
 524                 case 1: /* start of line, alloc */
 525                         /*
 526                          * this state makes sure we have a complete line in
 527                          * further processing, so we don't need wrap-tracking
 528                          * everywhere.
 529                          */
 530                         if (!eof && !memchr(l + i, '\n', avail - i))
 531                                 goto again;
 532                         ++lineno;
 533                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 534                                                 BTRFS_FSID_SIZE) == 0) {
 535                                 state = 2;
 536                                 continue;
 537                         }
 538                         ++curr;
 539                         tmp = p;
 540                         p = realloc(p, (curr + 2) * sizeof(*p));
 541                         if (!p) {
 542                                 free_history(tmp);
 543                                 return ERR_PTR(-errno);
 544                         }
 545                         p[curr] = malloc(sizeof(**p));
 546                         if (!p[curr]) {
 547                                 free_history(p);
 548                                 return ERR_PTR(-errno);
 549                         }
 550                         memset(p[curr], 0, sizeof(**p));
 551                         p[curr + 1] = NULL;
 552                         ++state;
 553                         /* fall through */
 554                 case 2: /* start of line, skip space */
 555                         while (isspace(l[i]) && i < avail) {
 556                                 if (l[i] == '\n')
 557                                         ++lineno;
 558                                 ++i;
 559                         }
 560                         if (i >= avail ||
 561                             (!eof && !memchr(l + i, '\n', avail - i)))
 562                                 goto again;
 563                         ++state;
 564                         /* fall through */
 565                 case 3: /* read fsid */
 566                         if (i == avail)
 567                                 continue;
 568                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 569                                 ;
 570                         if (i + j + 1 >= avail)
 571                                 _SCRUB_INVALID;
 572                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 573                                 _SCRUB_INVALID;
 574                         l[i + j] = '\0';
 575                         ret = uuid_parse(l + i, p[curr]->fsid);
 576                         if (ret)
 577                                 _SCRUB_INVALID;
 578                         i += j + 1;
 579                         ++state;
 580                         /* fall through */
 581                 case 4: /* read dev id */
 582                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 583                                 ;
 584                         if (j == 0 || i + j + 1 >= avail)
 585                                 _SCRUB_INVALID;
 586                         p[curr]->devid = atoll(&l[i]);
 587                         i += j + 1;
 588                         ++state;
 589                         /* fall through */
 590                 case 5: /* read key/value pair */
 591                         ret = 0;
 592                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 593                                         &p[curr]->p);
 594                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 595                                         &p[curr]->p);
 596                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 597                                         &p[curr]->p);
 598                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 599                                         &p[curr]->p);
 600                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 601                                         &p[curr]->p);
 602                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 603                                         &p[curr]->p);
 604                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 605                                         &p[curr]->p);
 606                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 607                                         &p[curr]->p);
 608                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 609                                         &p[curr]->p);
 610                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 611                                         &p[curr]->p);
 612                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 613                                         &p[curr]->p);
 614                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 615                                         &p[curr]->p);
 616                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 617                                         &p[curr]->p);
 618                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 619                                         &p[curr]->p);
 620                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 621                                         &p[curr]->p);
 622                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 623                                         &p[curr]->stats);
 624                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 625                                         (u64 *)&p[curr]->stats);
 626                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 627                                         (u64 *)&p[curr]->stats);
 628                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 629                                         (u64 *)&p[curr]->stats);
 630                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 631                                         &p[curr]->stats);
 632                         if (ret != 1)
 633                                 _SCRUB_INVALID;
 634                         ++state;
 635                         /* fall through */
 636                 case 6: /* after number */
 637                         if (l[i] == '|')
 638                                 state = 5;
 639                         else if (l[i] == '\n')
 640                                 state = 1;
 641                         else
 642                                 _SCRUB_INVALID;
 643                         ++i;
 644                         continue;
 645                 case 99: /* skip rest of line */
 646 skip:
 647                         state = 99;
 648                         do {
 649                                 ++i;
 650                                 if (l[i - 1] == '\n') {
 651                                         state = 1;
 652                                         break;
 653                                 }
 654                         } while (i < avail);
 655                         continue;
 656                 }
 657                 error("internal error: unknown parser state %d near byte %d",
 658                                 state, i);
 659                 return ERR_PTR(-EINVAL);
 660         }
 661         goto again;
 662 }
 663
 664 static int scrub_write_buf(int fd, const void *data, int len)
 665 {
 666         int ret;
 667         ret = write(fd, data, len);
 668         return ret - len;
 669 }
 670
 671 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 672                                 __attribute__ ((format (printf, 4, 5)));
 673 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 674 {
 675         int ret;
 676         va_list args;
 677
 678         va_start(args, fmt);
 679         ret = vsnprintf(buf, max, fmt, args);
 680         va_end(args);
 681         if (ret >= max)
 682                 return ret - max;
 683         return scrub_write_buf(fd, buf, ret);
 684 }
 685
 686 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 687                         data->resumed->p.name + data->scrub_args.progress.name
 688
 689 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 690                                                   struct scrub_progress *dest)
 691 {
 692         if (!data->resumed || data->skip)
 693                 return data;
 694
 695         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 696         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 697         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 698         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 699         _SCRUB_SUM(dest, data, read_errors);
 700         _SCRUB_SUM(dest, data, csum_errors);
 701         _SCRUB_SUM(dest, data, verify_errors);
 702         _SCRUB_SUM(dest, data, no_csum);
 703         _SCRUB_SUM(dest, data, csum_discards);
 704         _SCRUB_SUM(dest, data, super_errors);
 705         _SCRUB_SUM(dest, data, malloc_errors);
 706         _SCRUB_SUM(dest, data, uncorrectable_errors);
 707         _SCRUB_SUM(dest, data, corrected_errors);
 708         _SCRUB_SUM(dest, data, last_physical);
 709         dest->stats.canceled = data->stats.canceled;
 710         dest->stats.finished = data->stats.finished;
 711         dest->stats.t_resumed = data->stats.t_start;
 712         dest->stats.t_start = data->resumed->stats.t_start;
 713         dest->stats.duration = data->resumed->stats.duration +
 714                                                         data->stats.duration;
 715         dest->scrub_args.devid = data->scrub_args.devid;
 716         return dest;
 717 }
 718
 719 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 720         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 721                         use->scrub_args.progress.name)
 722
 723 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 724         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 725                         use->stats.name)
 726
 727 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 728 {
 729         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 730 }
 731
 732 static int scrub_write_file(int fd, const char *fsid,
 733                                 struct scrub_progress *data, int n)
 734 {
 735         int ret = 0;
 736         int i;
 737         char buf[1024];
 738         struct scrub_progress local;
 739         struct scrub_progress *use;
 740
 741         if (n < 1)
 742                 return -EINVAL;
 743
 744         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 745         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 746                                 SCRUB_FILE_VERSION "\n",
 747                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 748                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 749         if (ret)
 750                 return -EOVERFLOW;
 751
 752         for (i = 0; i < n; ++i) {
 753                 use = scrub_resumed_stats(&data[i], &local);
 754                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 755                     scrub_write_buf(fd, ":", 1) ||
 756                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 757                                         use->scrub_args.devid) ||
 758                     scrub_write_buf(fd, buf, ret) ||
 759                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 760                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 761                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 762                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 763                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 764                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 765                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 766                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 767                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 768                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 769                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 770                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 771                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 772                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 773                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 774                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 775                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 776                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 777                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 778                     scrub_write_buf(fd, "\n", 1)) {
 779                         return -EOVERFLOW;
 780                 }
 781         }
 782
 783         return 0;
 784 }
 785
 786 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 787                                 struct scrub_progress *data, int n)
 788 {
 789         int ret;
 790         int err;
 791         int fd = -1;
 792         int old;
 793
 794         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 795         if (ret) {
 796                 err = -ret;
 797                 goto out3;
 798         }
 799
 800         ret = pthread_mutex_lock(m);
 801         if (ret) {
 802                 err = -ret;
 803                 goto out2;
 804         }
 805
 806         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 807         if (fd < 0) {
 808                 err = fd;
 809                 goto out1;
 810         }
 811         err = scrub_write_file(fd, fsid, data, n);
 812         if (err)
 813                 goto out1;
 814         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 815         if (err)
 816                 goto out1;
 817
 818 out1:
 819         if (fd >= 0) {
 820                 ret = close(fd);
 821                 if (ret)
 822                         err = -errno;
 823         }
 824
 825         ret = pthread_mutex_unlock(m);
 826         if (ret && !err)
 827                 err = -ret;
 828
 829 out2:
 830         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 831         if (ret && !err)
 832                 err = -ret;
 833
 834 out3:
 835         return err;
 836 }
 837
 838 static void *scrub_one_dev(void *ctx)
 839 {
 840         struct scrub_progress *sp = ctx;
 841         int ret;
 842         struct timeval tv;
 843
 844         sp->stats.canceled = 0;
 845         sp->stats.duration = 0;
 846         sp->stats.finished = 0;
 847
 848         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 849                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 850                                         sp->ioprio_classdata));
 851         if (ret)
 852                 warning("setting ioprio failed: %s (ignored)",
 853                         strerror(errno));
 854
 855         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 856         gettimeofday(&tv, NULL);
 857         sp->ret = ret;
 858         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 859         sp->stats.canceled = !!ret;
 860         sp->ioctl_errno = errno;
 861         ret = pthread_mutex_lock(&sp->progress_mutex);
 862         if (ret)
 863                 return ERR_PTR(-ret);
 864         sp->stats.finished = 1;
 865         ret = pthread_mutex_unlock(&sp->progress_mutex);
 866         if (ret)
 867                 return ERR_PTR(-ret);
 868
 869         return NULL;
 870 }
 871
 872 static void *progress_one_dev(void *ctx)
 873 {
 874         struct scrub_progress *sp = ctx;
 875
 876         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 877         sp->ioctl_errno = errno;
 878
 879         return NULL;
 880 }
 881
 882 /* nb: returns a negative errno via ERR_PTR */
 883 static void *scrub_progress_cycle(void *ctx)
 884 {
 885         int ret = 0;
 886         int  perr = 0;  /* positive / pthread error returns */
 887         int old;
 888         int i;
 889         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 890         struct scrub_progress *sp;
 891         struct scrub_progress *sp_last;
 892         struct scrub_progress *sp_shared;
 893         struct timeval tv;
 894         struct scrub_progress_cycle *spc = ctx;
 895         int ndev = spc->fi->num_devices;
 896         int this = 1;
 897         int last = 0;
 898         int peer_fd = -1;
 899         struct pollfd accept_poll_fd = {
 900                 .fd = spc->prg_fd,
 901                 .events = POLLIN,
 902                 .revents = 0,
 903         };
 904         struct pollfd write_poll_fd = {
 905                 .events = POLLOUT,
 906                 .revents = 0,
 907         };
 908         struct sockaddr_un peer;
 909         socklen_t peer_size = sizeof(peer);
 910
 911         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 912         if (perr)
 913                 goto out;
 914
 915         uuid_unparse(spc->fi->fsid, fsid);
 916
 917         for (i = 0; i < ndev; ++i) {
 918                 sp = &spc->progress[i];
 919                 sp_last = &spc->progress[i + ndev];
 920                 sp_shared = &spc->shared_progress[i];
 921                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 922                                                 sp_shared->scrub_args.devid;
 923                 sp->fd = sp_last->fd = spc->fdmnt;
 924                 sp->stats.t_start = sp_last->stats.t_start =
 925                                                 sp_shared->stats.t_start;
 926                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 927                 sp->skip = sp_last->skip = sp_shared->skip;
 928                 sp->stats.finished = sp_last->stats.finished =
 929                                                 sp_shared->stats.finished;
 930         }
 931
 932         while (1) {
 933                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 934                 if (ret == -1) {
 935                         ret = -errno;
 936                         goto out;
 937                 }
 938                 if (ret)
 939                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 940                                          &peer_size);
 941                 gettimeofday(&tv, NULL);
 942                 this = (this + 1)%2;
 943                 last = (last + 1)%2;
 944                 for (i = 0; i < ndev; ++i) {
 945                         sp = &spc->progress[this * ndev + i];
 946                         sp_last = &spc->progress[last * ndev + i];
 947                         sp_shared = &spc->shared_progress[i];
 948                         if (sp->stats.finished)
 949                                 continue;
 950                         progress_one_dev(sp);
 951                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 952                         if (!sp->ret)
 953                                 continue;
 954                         if (sp->ioctl_errno != ENOTCONN &&
 955                             sp->ioctl_errno != ENODEV) {
 956                                 ret = -sp->ioctl_errno;
 957                                 goto out;
 958                         }
 959                         /*
 960                          * scrub finished or device removed, check the
 961                          * finished flag. if unset, just use the last
 962                          * result we got for the current write and go
 963                          * on. flag should be set on next cycle, then.
 964                          */
 965                         perr = pthread_setcancelstate(
 966                                         PTHREAD_CANCEL_DISABLE, &old);
 967                         if (perr)
 968                                 goto out;
 969                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 970                         if (perr)
 971                                 goto out;
 972                         if (!sp_shared->stats.finished) {
 973                                 perr = pthread_mutex_unlock(
 974                                                 &sp_shared->progress_mutex);
 975                                 if (perr)
 976                                         goto out;
 977                                 perr = pthread_setcancelstate(
 978                                                 PTHREAD_CANCEL_ENABLE, &old);
 979                                 if (perr)
 980                                         goto out;
 981                                 memcpy(sp, sp_last, sizeof(*sp));
 982                                 continue;
 983                         }
 984                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 985                         if (perr)
 986                                 goto out;
 987                         perr = pthread_setcancelstate(
 988                                         PTHREAD_CANCEL_ENABLE, &old);
 989                         if (perr)
 990                                 goto out;
 991                         memcpy(sp, sp_shared, sizeof(*sp));
 992                         memcpy(sp_last, sp_shared, sizeof(*sp));
 993                 }
 994                 if (peer_fd != -1) {
 995                         write_poll_fd.fd = peer_fd;
 996                         ret = poll(&write_poll_fd, 1, 0);
 997                         if (ret == -1) {
 998                                 ret = -errno;
 999                                 goto out;
1000                         }
1001                         if (ret) {
1002                                 ret = scrub_write_file(
1003                                         peer_fd, fsid,
1004                                         &spc->progress[this * ndev], ndev);
1005                                 if (ret)
1006                                         goto out;
1007                         }
1008                         close(peer_fd);
1009                         peer_fd = -1;
1010                 }
1011                 if (!spc->do_record)
1012                         continue;
1013                 ret = scrub_write_progress(spc->write_mutex, fsid,
1014                                            &spc->progress[this * ndev], ndev);
1015                 if (ret)
1016                         goto out;
1017         }
1018 out:
1019         if (peer_fd != -1)
1020                 close(peer_fd);
1021         if (perr)
1022                 ret = -perr;
1023         return ERR_PTR(ret);
1024 }
1025
1026 static struct scrub_file_record *last_dev_scrub(
1027                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1028 {
1029         int i;
1030
1031         if (!past_scrubs || IS_ERR(past_scrubs))
1032                 return NULL;
1033
1034         for (i = 0; past_scrubs[i]; ++i)
1035                 if (past_scrubs[i]->devid == devid)
1036                         return past_scrubs[i];
1037
1038         return NULL;
1039 }
1040
1041 static int mkdir_p(char *path)
1042 {
1043         int i;
1044         int ret;
1045
1046         for (i = 1; i < strlen(path); ++i) {
1047                 if (path[i] != '/')
1048                         continue;
1049                 path[i] = '\0';
1050                 ret = mkdir(path, 0777);
1051                 if (ret && errno != EEXIST)
1052                         return -errno;
1053                 path[i] = '/';
1054         }
1055
1056         return 0;
1057 }
1058
1059 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1060                                   struct btrfs_ioctl_dev_info_args *di_args,
1061                                   struct scrub_file_record **past_scrubs)
1062 {
1063         int i;
1064
1065         if (!fi_args || !di_args || !past_scrubs)
1066                 return 0;
1067
1068         for (i = 0; i < fi_args->num_devices; i++) {
1069                 struct scrub_file_record *sfr =
1070                         last_dev_scrub(past_scrubs, di_args[i].devid);
1071
1072                 if (!sfr)
1073                         continue;
1074                 if (!(sfr->stats.finished || sfr->stats.canceled))
1075                         return 1;
1076         }
1077         return 0;
1078 }
1079
1080 static int is_scrub_running_in_kernel(int fd,
1081                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1082 {
1083         struct scrub_progress sp;
1084         int i;
1085         int ret;
1086
1087         for (i = 0; i < max_devices; i++) {
1088                 memset(&sp, 0, sizeof(sp));
1089                 sp.scrub_args.devid = di_args[i].devid;
1090                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1091                 if (!ret)
1092                         return 1;
1093         }
1094
1095         return 0;
1096 }
1097
1098 static const char * const cmd_scrub_start_usage[];
1099 static const char * const cmd_scrub_resume_usage[];
1100
1101 static int scrub_start(int argc, char **argv, int resume)
1102 {
1103         int fdmnt;
1104         int prg_fd = -1;
1105         int fdres = -1;
1106         int ret;
1107         pid_t pid;
1108         int c;
1109         int i;
1110         int err = 0;
1111         int e_uncorrectable = 0;
1112         int e_correctable = 0;
1113         int print_raw = 0;
1114         char *path;
1115         int do_background = 1;
1116         int do_wait = 0;
1117         int do_print = 0;
1118         int do_quiet = 0;
1119         int do_record = 1;
1120         int readonly = 0;
1121         int do_stats_per_dev = 0;
1122         int ioprio_class = IOPRIO_CLASS_IDLE;
1123         int ioprio_classdata = 0;
1124         int n_start = 0;
1125         int n_skip = 0;
1126         int n_resume = 0;
1127         struct btrfs_ioctl_fs_info_args fi_args;
1128         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1129         struct scrub_progress *sp = NULL;
1130         struct scrub_fs_stat fs_stat;
1131         struct timeval tv;
1132         struct sockaddr_un addr = {
1133                 .sun_family = AF_UNIX,
1134         };
1135         pthread_t *t_devs = NULL;
1136         pthread_t t_prog;
1137         struct scrub_file_record **past_scrubs = NULL;
1138         struct scrub_file_record *last_scrub = NULL;
1139         char *datafile = strdup(SCRUB_DATA_FILE);
1140         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1141         char sock_path[PATH_MAX] = "";
1142         struct scrub_progress_cycle spc;
1143         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1144         void *terr;
1145         u64 devid;
1146         DIR *dirstream = NULL;
1147         int force = 0;
1148         int nothing_to_resume = 0;
1149
1150         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1151                 switch (c) {
1152                 case 'B':
1153                         do_background = 0;
1154                         do_wait = 1;
1155                         do_print = 1;
1156                         break;
1157                 case 'd':
1158                         do_stats_per_dev = 1;
1159                         break;
1160                 case 'q':
1161                         do_quiet = 1;
1162                         break;
1163                 case 'r':
1164                         readonly = 1;
1165                         break;
1166                 case 'R':
1167                         print_raw = 1;
1168                         break;
1169                 case 'c':
1170                         ioprio_class = (int)strtol(optarg, NULL, 10);
1171                         break;
1172                 case 'n':
1173                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1174                         break;
1175                 case 'f':
1176                         force = 1;
1177                         break;
1178                 case '?':
1179                 default:
1180                         usage(resume ? cmd_scrub_resume_usage :
1181                                                 cmd_scrub_start_usage);
1182                 }
1183         }
1184
1185         /* try to catch most error cases before forking */
1186
1187         if (check_argc_exact(argc - optind, 1)) {
1188                 usage(resume ? cmd_scrub_resume_usage :
1189                                         cmd_scrub_start_usage);
1190         }
1191
1192         spc.progress = NULL;
1193         if (do_quiet && do_print)
1194                 do_print = 0;
1195
1196         if (mkdir_p(datafile)) {
1197                 warning_on(!do_quiet,
1198     "cannot create scrub data file, mkdir %s failed: %s. Status recording disabled",
1199                         datafile, strerror(errno));
1200                 do_record = 0;
1201         }
1202         free(datafile);
1203
1204         path = argv[optind];
1205
1206         fdmnt = open_path_or_dev_mnt(path, &dirstream, !do_quiet);
1207         if (fdmnt < 0)
1208                 return 1;
1209
1210         ret = get_fs_info(path, &fi_args, &di_args);
1211         if (ret) {
1212                 error_on(!do_quiet,
1213                         "getting dev info for scrub failed: %s",
1214                          strerror(-ret));
1215                 err = 1;
1216                 goto out;
1217         }
1218         if (!fi_args.num_devices) {
1219                 error_on(!do_quiet, "no devices found");
1220                 err = 1;
1221                 goto out;
1222         }
1223
1224         uuid_unparse(fi_args.fsid, fsid);
1225         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1226         if (fdres < 0 && fdres != -ENOENT) {
1227                 warning_on(!do_quiet, "failed to open status file: %s",
1228                         strerror(-fdres));
1229         } else if (fdres >= 0) {
1230                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1231                 if (IS_ERR(past_scrubs))
1232                         warning_on(!do_quiet, "failed to read status file: %s",
1233                                 strerror(-PTR_ERR(past_scrubs)));
1234                 close(fdres);
1235         }
1236
1237         /*
1238          * Check for stale information in the status file, ie. if it's
1239          * canceled=0, finished=0 but no scrub is running.
1240          */
1241         if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1242                 force = 1;
1243
1244         /*
1245          * check whether any involved device is already busy running a
1246          * scrub. This would cause damaged status messages and the state
1247          * "aborted" without the explanation that a scrub was already
1248          * running. Therefore check it first, prevent it and give some
1249          * feedback to the user if scrub is already running.
1250          * Note that if scrub is started with a block device as the
1251          * parameter, only that particular block device is checked. It
1252          * is a normal mode of operation to start scrub on multiple
1253          * single devices, there is no reason to prevent this.
1254          */
1255         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1256                 error_on(!do_quiet,
1257                         "Scrub is already running.\n"
1258                         "To cancel use 'btrfs scrub cancel %s'.\n"
1259                         "To see the status use 'btrfs scrub status [-d] %s'",
1260                         path, path);
1261                 err = 1;
1262                 goto out;
1263         }
1264
1265         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1266         sp = calloc(fi_args.num_devices, sizeof(*sp));
1267         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1268
1269         if (!t_devs || !sp || !spc.progress) {
1270                 error_on(!do_quiet, "scrub failed: %s", strerror(errno));
1271                 err = 1;
1272                 goto out;
1273         }
1274
1275         for (i = 0; i < fi_args.num_devices; ++i) {
1276                 devid = di_args[i].devid;
1277                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1278                 if (ret) {
1279                         error_on(!do_quiet, "pthread_mutex_init failed: %s",
1280                                 strerror(ret));
1281                         err = 1;
1282                         goto out;
1283                 }
1284                 last_scrub = last_dev_scrub(past_scrubs, devid);
1285                 sp[i].scrub_args.devid = devid;
1286                 sp[i].fd = fdmnt;
1287                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1288                                              !last_scrub->stats.finished)) {
1289                         ++n_resume;
1290                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1291                         sp[i].resumed = last_scrub;
1292                 } else if (resume) {
1293                         ++n_skip;
1294                         sp[i].skip = 1;
1295                         sp[i].resumed = last_scrub;
1296                         continue;
1297                 } else {
1298                         ++n_start;
1299                         sp[i].scrub_args.start = 0ll;
1300                         sp[i].resumed = NULL;
1301                 }
1302                 sp[i].skip = 0;
1303                 sp[i].scrub_args.end = (u64)-1ll;
1304                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1305                 sp[i].ioprio_class = ioprio_class;
1306                 sp[i].ioprio_classdata = ioprio_classdata;
1307         }
1308
1309         if (!n_start && !n_resume) {
1310                 if (!do_quiet)
1311                         printf("scrub: nothing to resume for %s, fsid %s\n",
1312                                path, fsid);
1313                 nothing_to_resume = 1;
1314                 goto out;
1315         }
1316
1317         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1318         while (ret != -1) {
1319                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1320                                         sock_path, sizeof(sock_path));
1321                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1322                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1323                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1324                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1325                 if (ret != -1 || errno != EADDRINUSE)
1326                         break;
1327                 /*
1328                  * bind failed with EADDRINUSE. so let's see if anyone answers
1329                  * when we make a call to the socket ...
1330                  */
1331                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1332                 if (!ret || errno != ECONNREFUSED) {
1333                         /* ... yes, so scrub must be running. error out */
1334                         error("scrub already running");
1335                         close(prg_fd);
1336                         prg_fd = -1;
1337                         goto out;
1338                 }
1339                 /*
1340                  * ... no, this means someone left us alone with an unused
1341                  * socket in the file system. remove it and try again.
1342                  */
1343                 ret = unlink(sock_path);
1344         }
1345         if (ret != -1)
1346                 ret = listen(prg_fd, 100);
1347         if (ret == -1) {
1348                 warning_on(!do_quiet,
1349    "failed to open the progress status socket at %s: %s. Progress cannot be queried",
1350                         sock_path[0] ? sock_path :
1351                         SCRUB_PROGRESS_SOCKET_PATH, strerror(errno));
1352                 if (prg_fd != -1) {
1353                         close(prg_fd);
1354                         prg_fd = -1;
1355                         if (sock_path[0])
1356                                 unlink(sock_path);
1357                 }
1358         }
1359
1360         if (do_record) {
1361                 /* write all-zero progress file for a start */
1362                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1363                                            fi_args.num_devices);
1364                 if (ret) {
1365                         warning_on(!do_quiet,
1366    "failed to write the progress status file: %s. Status recording disabled",
1367                                 strerror(-ret));
1368                         do_record = 0;
1369                 }
1370         }
1371
1372         if (do_background) {
1373                 pid = fork();
1374                 if (pid == -1) {
1375                         error_on(!do_quiet, "cannot scrub, fork failed: %s",
1376                                 strerror(errno));
1377                         err = 1;
1378                         goto out;
1379                 }
1380
1381                 if (pid) {
1382                         int stat;
1383                         scrub_handle_sigint_parent();
1384                         if (!do_quiet)
1385                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1386                                        n_start ? "started" : "resumed",
1387                                        path, fsid, pid);
1388                         if (!do_wait) {
1389                                 err = 0;
1390                                 goto out;
1391                         }
1392                         ret = wait(&stat);
1393                         if (ret != pid) {
1394                                 error_on(!do_quiet, "wait failed (ret=%d): %s",
1395                                         ret, strerror(errno));
1396                                 err = 1;
1397                                 goto out;
1398                         }
1399                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1400                                 error_on(!do_quiet, "scrub process failed");
1401                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1402                                 goto out;
1403                         }
1404                         err = 0;
1405                         goto out;
1406                 }
1407         }
1408
1409         scrub_handle_sigint_child(fdmnt);
1410
1411         for (i = 0; i < fi_args.num_devices; ++i) {
1412                 if (sp[i].skip) {
1413                         sp[i].scrub_args.progress = sp[i].resumed->p;
1414                         sp[i].stats = sp[i].resumed->stats;
1415                         sp[i].ret = 0;
1416                         sp[i].stats.finished = 1;
1417                         continue;
1418                 }
1419                 devid = di_args[i].devid;
1420                 gettimeofday(&tv, NULL);
1421                 sp[i].stats.t_start = tv.tv_sec;
1422                 ret = pthread_create(&t_devs[i], NULL,
1423                                         scrub_one_dev, &sp[i]);
1424                 if (ret) {
1425                         if (do_print)
1426                         error("creating scrub_one_dev[%llu] thread failed: %s",
1427                                 devid, strerror(ret));
1428                         err = 1;
1429                         goto out;
1430                 }
1431         }
1432
1433         spc.fdmnt = fdmnt;
1434         spc.prg_fd = prg_fd;
1435         spc.do_record = do_record;
1436         spc.write_mutex = &spc_write_mutex;
1437         spc.shared_progress = sp;
1438         spc.fi = &fi_args;
1439         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1440         if (ret) {
1441                 if (do_print)
1442                         error("creating progress thread failed: %s",
1443                                 strerror(ret));
1444                 err = 1;
1445                 goto out;
1446         }
1447
1448         err = 0;
1449         for (i = 0; i < fi_args.num_devices; ++i) {
1450                 if (sp[i].skip)
1451                         continue;
1452                 devid = di_args[i].devid;
1453                 ret = pthread_join(t_devs[i], NULL);
1454                 if (ret) {
1455                         if (do_print)
1456                           error("pthread_join failed for scrub_one_dev[%llu]: %s",
1457                                 devid, strerror(ret));
1458                         ++err;
1459                         continue;
1460                 }
1461                 if (sp[i].ret) {
1462                         switch (sp[i].ioctl_errno) {
1463                         case ENODEV:
1464                                 if (do_print)
1465                                         warning("device %lld not present",
1466                                                 devid);
1467                                 continue;
1468                         case ECANCELED:
1469                                 ++err;
1470                                 break;
1471                         default:
1472                                 if (do_print)
1473                 error("scrubbing %s failed for device id %lld: ret=%d, errno=%d (%s)",
1474                                         path, devid,
1475                                         sp[i].ret, sp[i].ioctl_errno,
1476                                         strerror(sp[i].ioctl_errno));
1477                                 ++err;
1478                                 continue;
1479                         }
1480                 }
1481                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1482                         e_uncorrectable++;
1483                 if (sp[i].scrub_args.progress.corrected_errors > 0
1484                     || sp[i].scrub_args.progress.unverified_errors > 0)
1485                         e_correctable++;
1486         }
1487
1488         if (do_print) {
1489                 const char *append = "done";
1490                 if (!do_stats_per_dev)
1491                         init_fs_stat(&fs_stat);
1492                 for (i = 0; i < fi_args.num_devices; ++i) {
1493                         if (do_stats_per_dev) {
1494                                 print_scrub_dev(&di_args[i],
1495                                                 &sp[i].scrub_args.progress,
1496                                                 print_raw,
1497                                                 sp[i].ret ? "canceled" : "done",
1498                                                 &sp[i].stats);
1499                         } else {
1500                                 if (sp[i].ret)
1501                                         append = "canceled";
1502                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1503                                                 &sp[i].stats, &fs_stat);
1504                         }
1505                 }
1506                 if (!do_stats_per_dev) {
1507                         printf("scrub %s for %s\n", append, fsid);
1508                         print_fs_stat(&fs_stat, print_raw);
1509                 }
1510         }
1511
1512         ret = pthread_cancel(t_prog);
1513         if (!ret)
1514                 ret = pthread_join(t_prog, &terr);
1515
1516         /* check for errors from the handling of the progress thread */
1517         if (do_print && ret) {
1518                 error("progress thread handling failed: %s",
1519                         strerror(ret));
1520         }
1521
1522         /* check for errors returned from the progress thread itself */
1523         if (do_print && terr && terr != PTHREAD_CANCELED)
1524                 error("recording progress failed: %s",
1525                         strerror(-PTR_ERR(terr)));
1526
1527         if (do_record) {
1528                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1529                                            fi_args.num_devices);
1530                 if (ret && do_print)
1531                         error("failed to record the result: %s",
1532                                 strerror(-ret));
1533         }
1534
1535         scrub_handle_sigint_child(-1);
1536
1537 out:
1538         free_history(past_scrubs);
1539         free(di_args);
1540         free(t_devs);
1541         free(sp);
1542         free(spc.progress);
1543         if (prg_fd > -1) {
1544                 close(prg_fd);
1545                 if (sock_path[0])
1546                         unlink(sock_path);
1547         }
1548         close_file_or_dir(fdmnt, dirstream);
1549
1550         if (err)
1551                 return 1;
1552         if (nothing_to_resume)
1553                 return 2;
1554         if (e_uncorrectable) {
1555                 error_on(!do_quiet, "there are uncorrectable errors");
1556                 return 3;
1557         }
1558         if (e_correctable)
1559                 warning_on(!do_quiet,
1560                         "errors detected during scrubbing, corrected");
1561
1562         return 0;
1563 }
1564
1565 static const char * const cmd_scrub_start_usage[] = {
1566         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1567         "Start a new scrub. If a scrub is already running, the new one fails.",
1568         "",
1569         "-B     do not background",
1570         "-d     stats per device (-B only)",
1571         "-q     be quiet",
1572         "-r     read only mode",
1573         "-R     raw print mode, print full data instead of summary",
1574         "-c     set ioprio class (see ionice(1) manpage)",
1575         "-n     set ioprio classdata (see ionice(1) manpage)",
1576         "-f     force starting new scrub even if a scrub is already running",
1577         "       this is useful when scrub stats record file is damaged",
1578         NULL
1579 };
1580
1581 static int cmd_scrub_start(int argc, char **argv)
1582 {
1583         return scrub_start(argc, argv, 0);
1584 }
1585
1586 static const char * const cmd_scrub_cancel_usage[] = {
1587         "btrfs scrub cancel <path>|<device>",
1588         "Cancel a running scrub",
1589         NULL
1590 };
1591
1592 static int cmd_scrub_cancel(int argc, char **argv)
1593 {
1594         char *path;
1595         int ret;
1596         int fdmnt = -1;
1597         DIR *dirstream = NULL;
1598
1599         clean_args_no_options(argc, argv, cmd_scrub_cancel_usage);
1600
1601         if (check_argc_exact(argc - optind, 1))
1602                 usage(cmd_scrub_cancel_usage);
1603
1604         path = argv[optind];
1605
1606         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1607         if (fdmnt < 0) {
1608                 ret = 1;
1609                 goto out;
1610         }
1611
1612         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1613
1614         if (ret < 0) {
1615                 error("scrub cancel failed on %s: %s", path,
1616                         errno == ENOTCONN ? "not running" : strerror(errno));
1617                 if (errno == ENOTCONN)
1618                         ret = 2;
1619                 else
1620                         ret = 1;
1621                 goto out;
1622         }
1623
1624         ret = 0;
1625         printf("scrub cancelled\n");
1626
1627 out:
1628         close_file_or_dir(fdmnt, dirstream);
1629         return ret;
1630 }
1631
1632 static const char * const cmd_scrub_resume_usage[] = {
1633         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1634         "Resume previously canceled or interrupted scrub",
1635         "",
1636         "-B     do not background",
1637         "-d     stats per device (-B only)",
1638         "-q     be quiet",
1639         "-r     read only mode",
1640         "-R     raw print mode, print full data instead of summary",
1641         "-c     set ioprio class (see ionice(1) manpage)",
1642         "-n     set ioprio classdata (see ionice(1) manpage)",
1643         NULL
1644 };
1645
1646 static int cmd_scrub_resume(int argc, char **argv)
1647 {
1648         return scrub_start(argc, argv, 1);
1649 }
1650
1651 static const char * const cmd_scrub_status_usage[] = {
1652         "btrfs scrub status [-dR] <path>|<device>",
1653         "Show status of running or finished scrub",
1654         "",
1655         "-d     stats per device",
1656         "-R     print raw stats",
1657         NULL
1658 };
1659
1660 static int cmd_scrub_status(int argc, char **argv)
1661 {
1662         char *path;
1663         struct btrfs_ioctl_fs_info_args fi_args;
1664         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1665         struct scrub_file_record **past_scrubs = NULL;
1666         struct scrub_file_record *last_scrub;
1667         struct scrub_fs_stat fs_stat;
1668         struct sockaddr_un addr = {
1669                 .sun_family = AF_UNIX,
1670         };
1671         int in_progress;
1672         int ret;
1673         int i;
1674         int fdmnt;
1675         int print_raw = 0;
1676         int do_stats_per_dev = 0;
1677         int c;
1678         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1679         int fdres = -1;
1680         int err = 0;
1681         DIR *dirstream = NULL;
1682
1683         while ((c = getopt(argc, argv, "dR")) != -1) {
1684                 switch (c) {
1685                 case 'd':
1686                         do_stats_per_dev = 1;
1687                         break;
1688                 case 'R':
1689                         print_raw = 1;
1690                         break;
1691                 case '?':
1692                 default:
1693                         usage(cmd_scrub_status_usage);
1694                 }
1695         }
1696
1697         if (check_argc_exact(argc - optind, 1))
1698                 usage(cmd_scrub_status_usage);
1699
1700         path = argv[optind];
1701
1702         fdmnt = open_path_or_dev_mnt(path, &dirstream, 1);
1703         if (fdmnt < 0)
1704                 return 1;
1705
1706         ret = get_fs_info(path, &fi_args, &di_args);
1707         if (ret) {
1708                 error("getting dev info for scrub failed: %s",
1709                         strerror(-ret));
1710                 err = 1;
1711                 goto out;
1712         }
1713         if (!fi_args.num_devices) {
1714                 error("no devices found");
1715                 err = 1;
1716                 goto out;
1717         }
1718
1719         uuid_unparse(fi_args.fsid, fsid);
1720
1721         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1722         if (fdres == -1) {
1723                 error("failed to create socket to receive progress information: %s",
1724                         strerror(errno));
1725                 err = 1;
1726                 goto out;
1727         }
1728         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1729                         NULL, addr.sun_path, sizeof(addr.sun_path));
1730         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1731         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1732         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1733         if (ret == -1) {
1734                 close(fdres);
1735                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1736                 if (fdres < 0 && fdres != -ENOENT) {
1737                         warning("failed to open status file: %s",
1738                                 strerror(-fdres));
1739                         err = 1;
1740                         goto out;
1741                 }
1742         }
1743
1744         if (fdres >= 0) {
1745                 past_scrubs = scrub_read_file(fdres, 1);
1746                 if (IS_ERR(past_scrubs))
1747                         warning("failed to read status: %s",
1748                                 strerror(-PTR_ERR(past_scrubs)));
1749         }
1750         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1751
1752         printf("scrub status for %s\n", fsid);
1753
1754         if (do_stats_per_dev) {
1755                 for (i = 0; i < fi_args.num_devices; ++i) {
1756                         last_scrub = last_dev_scrub(past_scrubs,
1757                                                         di_args[i].devid);
1758                         if (!last_scrub) {
1759                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1760                                                 NULL, NULL);
1761                                 continue;
1762                         }
1763                         last_scrub->stats.in_progress = in_progress;
1764                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1765                                         last_scrub->stats.finished ?
1766                                                         "history" : "status",
1767                                         &last_scrub->stats);
1768                 }
1769         } else {
1770                 init_fs_stat(&fs_stat);
1771                 fs_stat.s.in_progress = in_progress;
1772                 for (i = 0; i < fi_args.num_devices; ++i) {
1773                         last_scrub = last_dev_scrub(past_scrubs,
1774                                                         di_args[i].devid);
1775                         if (!last_scrub)
1776                                 continue;
1777                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1778                                         &fs_stat);
1779                 }
1780                 print_fs_stat(&fs_stat, print_raw);
1781         }
1782
1783 out:
1784         free_history(past_scrubs);
1785         free(di_args);
1786         if (fdres > -1)
1787                 close(fdres);
1788         close_file_or_dir(fdmnt, dirstream);
1789
1790         return !!err;
1791 }
1792
1793 static const char scrub_cmd_group_info[] =
1794 "verify checksums of data and metadata";
1795
1796 const struct cmd_group scrub_cmd_group = {
1797         scrub_cmd_group_usage, scrub_cmd_group_info, {
1798                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1799                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1800                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1801                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1802                 NULL_CMD_STRUCT
1803         }
1804 };
1805
1806 int cmd_scrub(int argc, char **argv)
1807 {
1808         return handle_command_group(&scrub_cmd_group, argc, argv);
1809 }