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 #include <limits.h>
  38
  39 #include "ctree.h"
  40 #include "ioctl.h"
  41 #include "utils.h"
  42 #include "volumes.h"
  43 #include "disk-io.h"
  44
  45 #include "commands.h"
  46
  47 static const char * const scrub_cmd_group_usage[] = {
  48         "btrfs scrub <command> [options] <path>|<device>",
  49         NULL
  50 };
  51
  52 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  53 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  54 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  55 #define SCRUB_FILE_VERSION "1"
  56
  57 struct scrub_stats {
  58         time_t t_start;
  59         time_t t_resumed;
  60         u64 duration;
  61         u64 finished;
  62         u64 canceled;
  63         int in_progress;
  64 };
  65
  66 /* TBD: replace with #include "linux/ioprio.h" in some years */
  67 #if !defined (IOPRIO_H)
  68 #define IOPRIO_WHO_PROCESS 1
  69 #define IOPRIO_CLASS_SHIFT 13
  70 #define IOPRIO_PRIO_VALUE(class, data) \
  71                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
  72 #define IOPRIO_CLASS_IDLE 3
  73 #endif
  74
  75 struct scrub_progress {
  76         struct btrfs_ioctl_scrub_args scrub_args;
  77         int fd;
  78         int ret;
  79         int skip;
  80         struct scrub_stats stats;
  81         struct scrub_file_record *resumed;
  82         int ioctl_errno;
  83         pthread_mutex_t progress_mutex;
  84         int ioprio_class;
  85         int ioprio_classdata;
  86 };
  87
  88 struct scrub_file_record {
  89         u8 fsid[BTRFS_FSID_SIZE];
  90         u64 devid;
  91         struct scrub_stats stats;
  92         struct btrfs_scrub_progress p;
  93 };
  94
  95 struct scrub_progress_cycle {
  96         int fdmnt;
  97         int prg_fd;
  98         int do_record;
  99         struct btrfs_ioctl_fs_info_args *fi;
 100         struct scrub_progress *progress;
 101         struct scrub_progress *shared_progress;
 102         pthread_mutex_t *write_mutex;
 103 };
 104
 105 struct scrub_fs_stat {
 106         struct btrfs_scrub_progress p;
 107         struct scrub_stats s;
 108         int i;
 109 };
 110
 111 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 112 {
 113         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 114         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 115         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 116         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 117         printf("\tread_errors: %lld\n", sp->read_errors);
 118         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 119         printf("\tverify_errors: %lld\n", sp->verify_errors);
 120         printf("\tno_csum: %lld\n", sp->no_csum);
 121         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 122         printf("\tsuper_errors: %lld\n", sp->super_errors);
 123         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 124         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 125         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 126         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 127         printf("\tlast_physical: %lld\n", sp->last_physical);
 128 }
 129
 130 #define ERR(test, ...) do {                     \
 131         if (test)                               \
 132                 fprintf(stderr, __VA_ARGS__);   \
 133 } while (0)
 134
 135 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 136         if (test)                               \
 137                 printf(" %s=%llu", desc, test); \
 138 } while (0)
 139
 140 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 141 {
 142         u64 err_cnt;
 143         u64 err_cnt2;
 144
 145         err_cnt = p->read_errors +
 146                         p->csum_errors +
 147                         p->verify_errors +
 148                         p->super_errors;
 149
 150         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 151
 152         if (p->malloc_errors)
 153                 printf("*** WARNING: memory allocation failed while scrubbing. "
 154                        "results may be inaccurate\n");
 155
 156         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
 157                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
 158                 max(err_cnt, err_cnt2));
 159
 160         if (err_cnt || err_cnt2) {
 161                 printf("\terror details:");
 162                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 163                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 164                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 165                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 166                 printf("\n");
 167                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 168                         "unverified errors: %llu\n", p->corrected_errors,
 169                         p->uncorrectable_errors, p->unverified_errors);
 170         }
 171 }
 172
 173 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 174         fs_stat->p.name += p->name;             \
 175 } while (0)
 176
 177 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 178 do {                                            \
 179         if (fs_stat->s.name > ss->name) {       \
 180                 fs_stat->s.name = ss->name;     \
 181         }                                       \
 182 } while (0)
 183
 184 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 185 do {                                                            \
 186         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 187                 fs_stat->s.name = ss->name;                     \
 188         }                                                       \
 189 } while (0)
 190
 191 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 192 do {                                                                    \
 193         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 194                 (fs_stat)->s.name = (ss)->name;                         \
 195         }                                                               \
 196 } while (0)
 197
 198 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 199                                 struct scrub_stats *ss,
 200                                 struct scrub_fs_stat *fs_stat)
 201 {
 202         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 203         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 204         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 205         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 206         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 207         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 208         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 209         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 210         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 211         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 212         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 213         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 214         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 215         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 216         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 217         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 218         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 219         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 220         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 221 }
 222
 223 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 224 {
 225         memset(fs_stat, 0, sizeof(*fs_stat));
 226         fs_stat->s.finished = 1;
 227 }
 228
 229 static void _print_scrub_ss(struct scrub_stats *ss)
 230 {
 231         char t[4096];
 232         struct tm tm;
 233         time_t seconds;
 234         unsigned hours;
 235
 236         if (!ss || !ss->t_start) {
 237                 printf("\tno stats available\n");
 238                 return;
 239         }
 240         if (ss->t_resumed) {
 241                 localtime_r(&ss->t_resumed, &tm);
 242                 strftime(t, sizeof(t), "%c", &tm);
 243                 t[sizeof(t) - 1] = '\0';
 244                 printf("\tscrub resumed at %s", t);
 245         } else {
 246                 localtime_r(&ss->t_start, &tm);
 247                 strftime(t, sizeof(t), "%c", &tm);
 248                 t[sizeof(t) - 1] = '\0';
 249                 printf("\tscrub started at %s", t);
 250         }
 251
 252         seconds = ss->duration;
 253         hours = ss->duration / (60 * 60);
 254         gmtime_r(&seconds, &tm);
 255         strftime(t, sizeof(t), "%M:%S", &tm);
 256         if (ss->finished && !ss->canceled) {
 257                 printf(" and finished after %02u:%s\n", hours, t);
 258         } else if (ss->canceled) {
 259                 printf(" and was aborted after %02u:%s\n", hours, t);
 260         } else {
 261                 if (ss->in_progress)
 262                         printf(", running for %02u:%s\n", hours, t);
 263                 else
 264                         printf(", interrupted after %02u:%s, not running\n",
 265                                         hours, t);
 266         }
 267 }
 268
 269 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 270                                 struct btrfs_scrub_progress *p, int raw,
 271                                 const char *append, struct scrub_stats *ss)
 272 {
 273         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 274                append ? append : "");
 275
 276         _print_scrub_ss(ss);
 277
 278         if (p) {
 279                 if (raw)
 280                         print_scrub_full(p);
 281                 else
 282                         print_scrub_summary(p);
 283         }
 284 }
 285
 286 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 287 {
 288         _print_scrub_ss(&fs_stat->s);
 289
 290         if (raw)
 291                 print_scrub_full(&fs_stat->p);
 292         else
 293                 print_scrub_summary(&fs_stat->p);
 294 }
 295
 296 static void free_history(struct scrub_file_record **last_scrubs)
 297 {
 298         struct scrub_file_record **l = last_scrubs;
 299         if (!l || IS_ERR(l))
 300                 return;
 301         while (*l)
 302                 free(*l++);
 303         free(last_scrubs);
 304 }
 305
 306 /*
 307  * cancels a running scrub and makes the master process record the current
 308  * progress status before exiting.
 309  */
 310 static int cancel_fd = -1;
 311 static void scrub_sigint_record_progress(int signal)
 312 {
 313         int ret;
 314
 315         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 316         if (ret < 0)
 317                 perror("Scrub cancel failed");
 318 }
 319
 320 static int scrub_handle_sigint_parent(void)
 321 {
 322         struct sigaction sa = {
 323                 .sa_handler = SIG_IGN,
 324                 .sa_flags = SA_RESTART,
 325         };
 326
 327         return sigaction(SIGINT, &sa, NULL);
 328 }
 329
 330 static int scrub_handle_sigint_child(int fd)
 331 {
 332         struct sigaction sa = {
 333                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 334         };
 335
 336         cancel_fd = fd;
 337         return sigaction(SIGINT, &sa, NULL);
 338 }
 339
 340 static int scrub_datafile(const char *fn_base, const char *fn_local,
 341                                 const char *fn_tmp, char *datafile, int size)
 342 {
 343         int ret;
 344         int end = size - 2;
 345
 346         datafile[end + 1] = '\0';
 347         strncpy(datafile, fn_base, end);
 348         ret = strlen(datafile);
 349
 350         if (ret + 1 > end)
 351                 return -EOVERFLOW;
 352
 353         datafile[ret] = '.';
 354         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 355         ret = strlen(datafile);
 356
 357         if (ret + 1 > end)
 358                 return -EOVERFLOW;
 359
 360         if (fn_tmp) {
 361                 datafile[ret] = '_';
 362                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 363                 ret = strlen(datafile);
 364
 365                 if (ret > end)
 366                         return -EOVERFLOW;
 367         }
 368
 369         return 0;
 370 }
 371
 372 static int scrub_open_file(const char *datafile, int m)
 373 {
 374         int fd;
 375         int ret;
 376
 377         fd = open(datafile, m, 0600);
 378         if (fd < 0)
 379                 return -errno;
 380
 381         ret = flock(fd, LOCK_EX|LOCK_NB);
 382         if (ret) {
 383                 ret = errno;
 384                 close(fd);
 385                 return -ret;
 386         }
 387
 388         return fd;
 389 }
 390
 391 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 392 {
 393         int ret;
 394         char datafile[PATH_MAX];
 395         ret = scrub_datafile(fn_base, fn_local, NULL,
 396                                 datafile, sizeof(datafile));
 397         if (ret < 0)
 398                 return ret;
 399         return scrub_open_file(datafile, O_RDONLY);
 400 }
 401
 402 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 403                                 const char *tmp)
 404 {
 405         int ret;
 406         char datafile[PATH_MAX];
 407         ret = scrub_datafile(fn_base, fn_local, tmp,
 408                                 datafile, sizeof(datafile));
 409         if (ret < 0)
 410                 return ret;
 411         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 412 }
 413
 414 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 415                                 const char *tmp)
 416 {
 417         int ret;
 418         char datafile_old[PATH_MAX];
 419         char datafile_new[PATH_MAX];
 420         ret = scrub_datafile(fn_base, fn_local, tmp,
 421                                 datafile_old, sizeof(datafile_old));
 422         if (ret < 0)
 423                 return ret;
 424         ret = scrub_datafile(fn_base, fn_local, NULL,
 425                                 datafile_new, sizeof(datafile_new));
 426         if (ret < 0)
 427                 return ret;
 428         ret = rename(datafile_old, datafile_new);
 429         return ret ? -errno : 0;
 430 }
 431
 432 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 433         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 434 }
 435
 436 /*
 437  * returns 0 if the key did not match (nothing was read)
 438  *         1 if the key did match (success)
 439  *        -1 if the key did match and an error occured
 440  */
 441 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 442                         const char *key, u64 *dest)
 443 {
 444         int j;
 445
 446         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 447                 *i += len - 1;
 448                 if (buf[*i] != ':')
 449                         return -1;
 450                 *i += 1;
 451                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 452                         ;
 453                 if (*i + j >= avail)
 454                         return -1;
 455                 *dest = atoll(&buf[*i]);
 456                 *i += j;
 457                 return 1;
 458         }
 459
 460         return 0;
 461 }
 462
 463 #define _SCRUB_INVALID do {                                             \
 464         if (report_errors)                                              \
 465                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 466                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 467                         lineno, i, state, 20 > avail ? avail : 20,      \
 468                         l + i,  __FILE__, __LINE__);                    \
 469         goto skip;                                                      \
 470 } while (0)
 471
 472 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 473 {
 474         int avail = 0;
 475         int old_avail = 0;
 476         char l[16 * 1024];
 477         int state = 0;
 478         int curr = -1;
 479         int i = 0;
 480         int j;
 481         int ret;
 482         int eof = 0;
 483         int lineno = 0;
 484         u64 version;
 485         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 486         struct scrub_file_record **p = NULL;
 487
 488 again:
 489         old_avail = avail - i;
 490         BUG_ON(old_avail < 0);
 491         if (old_avail)
 492                 memmove(l, l + i, old_avail);
 493         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 494         if (avail == 0)
 495                 eof = 1;
 496         if (avail == 0 && old_avail == 0) {
 497                 if (curr >= 0 &&
 498                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 499                         p[curr] = NULL;
 500                 } else if (curr == -1) {
 501                         p = ERR_PTR(-ENODATA);
 502                 }
 503                 return p;
 504         }
 505         if (avail == -1) {
 506                 free_history(p);
 507                 return ERR_PTR(-errno);
 508         }
 509         avail += old_avail;
 510
 511         i = 0;
 512         while (i < avail) {
 513                 void *tmp;
 514
 515                 switch (state) {
 516                 case 0: /* start of file */
 517                         ret = scrub_kvread(&i,
 518                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 519                                 SCRUB_FILE_VERSION_PREFIX, &version);
 520                         if (ret != 1)
 521                                 _SCRUB_INVALID;
 522                         if (version != atoll(SCRUB_FILE_VERSION))
 523                                 return ERR_PTR(-ENOTSUP);
 524                         state = 6;
 525                         continue;
 526                 case 1: /* start of line, alloc */
 527                         /*
 528                          * this state makes sure we have a complete line in
 529                          * further processing, so we don't need wrap-tracking
 530                          * everywhere.
 531                          */
 532                         if (!eof && !memchr(l + i, '\n', avail - i))
 533                                 goto again;
 534                         ++lineno;
 535                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 536                                                 BTRFS_FSID_SIZE) == 0) {
 537                                 state = 2;
 538                                 continue;
 539                         }
 540                         ++curr;
 541                         tmp = p;
 542                         p = realloc(p, (curr + 2) * sizeof(*p));
 543                         if (!p) {
 544                                 free_history(tmp);
 545                                 return ERR_PTR(-errno);
 546                         }
 547                         p[curr] = malloc(sizeof(**p));
 548                         if (!p[curr]) {
 549                                 free_history(p);
 550                                 return ERR_PTR(-errno);
 551                         }
 552                         memset(p[curr], 0, sizeof(**p));
 553                         p[curr + 1] = NULL;
 554                         ++state;
 555                         /* fall through */
 556                 case 2: /* start of line, skip space */
 557                         while (isspace(l[i]) && i < avail) {
 558                                 if (l[i] == '\n')
 559                                         ++lineno;
 560                                 ++i;
 561                         }
 562                         if (i >= avail ||
 563                             (!eof && !memchr(l + i, '\n', avail - i)))
 564                                 goto again;
 565                         ++state;
 566                         /* fall through */
 567                 case 3: /* read fsid */
 568                         if (i == avail)
 569                                 continue;
 570                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 571                                 ;
 572                         if (i + j + 1 >= avail)
 573                                 _SCRUB_INVALID;
 574                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 575                                 _SCRUB_INVALID;
 576                         l[i + j] = '\0';
 577                         ret = uuid_parse(l + i, p[curr]->fsid);
 578                         if (ret)
 579                                 _SCRUB_INVALID;
 580                         i += j + 1;
 581                         ++state;
 582                         /* fall through */
 583                 case 4: /* read dev id */
 584                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 585                                 ;
 586                         if (j == 0 || i + j + 1 >= avail)
 587                                 _SCRUB_INVALID;
 588                         p[curr]->devid = atoll(&l[i]);
 589                         i += j + 1;
 590                         ++state;
 591                         /* fall through */
 592                 case 5: /* read key/value pair */
 593                         ret = 0;
 594                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 595                                         &p[curr]->p);
 596                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 597                                         &p[curr]->p);
 598                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 599                                         &p[curr]->p);
 600                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 601                                         &p[curr]->p);
 602                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 603                                         &p[curr]->p);
 604                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 605                                         &p[curr]->p);
 606                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 607                                         &p[curr]->p);
 608                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 609                                         &p[curr]->p);
 610                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 611                                         &p[curr]->p);
 612                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 613                                         &p[curr]->p);
 614                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 615                                         &p[curr]->p);
 616                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 617                                         &p[curr]->p);
 618                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 619                                         &p[curr]->p);
 620                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 621                                         &p[curr]->p);
 622                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 623                                         &p[curr]->p);
 624                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 625                                         &p[curr]->stats);
 626                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 627                                         (u64 *)&p[curr]->stats);
 628                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 629                                         (u64 *)&p[curr]->stats);
 630                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 631                                         (u64 *)&p[curr]->stats);
 632                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 633                                         &p[curr]->stats);
 634                         if (ret != 1)
 635                                 _SCRUB_INVALID;
 636                         ++state;
 637                         /* fall through */
 638                 case 6: /* after number */
 639                         if (l[i] == '|')
 640                                 state = 5;
 641                         else if (l[i] == '\n')
 642                                 state = 1;
 643                         else
 644                                 _SCRUB_INVALID;
 645                         ++i;
 646                         continue;
 647                 case 99: /* skip rest of line */
 648 skip:
 649                         state = 99;
 650                         do {
 651                                 ++i;
 652                                 if (l[i - 1] == '\n') {
 653                                         state = 1;
 654                                         break;
 655                                 }
 656                         } while (i < avail);
 657                         continue;
 658                 }
 659                 BUG();
 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                 fprintf(stderr,
 853                         "WARNING: setting ioprio failed: %s (ignored).\n",
 854                         strerror(errno));
 855
 856         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 857         gettimeofday(&tv, NULL);
 858         sp->ret = ret;
 859         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 860         sp->stats.canceled = !!ret;
 861         sp->ioctl_errno = errno;
 862         ret = pthread_mutex_lock(&sp->progress_mutex);
 863         if (ret)
 864                 return ERR_PTR(-ret);
 865         sp->stats.finished = 1;
 866         ret = pthread_mutex_unlock(&sp->progress_mutex);
 867         if (ret)
 868                 return ERR_PTR(-ret);
 869
 870         return NULL;
 871 }
 872
 873 static void *progress_one_dev(void *ctx)
 874 {
 875         struct scrub_progress *sp = ctx;
 876
 877         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 878         sp->ioctl_errno = errno;
 879
 880         return NULL;
 881 }
 882
 883 /* nb: returns a negative errno via ERR_PTR */
 884 static void *scrub_progress_cycle(void *ctx)
 885 {
 886         int ret = 0;
 887         int  perr = 0;  /* positive / pthread error returns */
 888         int old;
 889         int i;
 890         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 891         struct scrub_progress *sp;
 892         struct scrub_progress *sp_last;
 893         struct scrub_progress *sp_shared;
 894         struct timeval tv;
 895         struct scrub_progress_cycle *spc = ctx;
 896         int ndev = spc->fi->num_devices;
 897         int this = 1;
 898         int last = 0;
 899         int peer_fd = -1;
 900         struct pollfd accept_poll_fd = {
 901                 .fd = spc->prg_fd,
 902                 .events = POLLIN,
 903                 .revents = 0,
 904         };
 905         struct pollfd write_poll_fd = {
 906                 .events = POLLOUT,
 907                 .revents = 0,
 908         };
 909         struct sockaddr_un peer;
 910         socklen_t peer_size = sizeof(peer);
 911
 912         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 913         if (perr)
 914                 goto out;
 915
 916         uuid_unparse(spc->fi->fsid, fsid);
 917
 918         for (i = 0; i < ndev; ++i) {
 919                 sp = &spc->progress[i];
 920                 sp_last = &spc->progress[i + ndev];
 921                 sp_shared = &spc->shared_progress[i];
 922                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 923                                                 sp_shared->scrub_args.devid;
 924                 sp->fd = sp_last->fd = spc->fdmnt;
 925                 sp->stats.t_start = sp_last->stats.t_start =
 926                                                 sp_shared->stats.t_start;
 927                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 928                 sp->skip = sp_last->skip = sp_shared->skip;
 929                 sp->stats.finished = sp_last->stats.finished =
 930                                                 sp_shared->stats.finished;
 931         }
 932
 933         while (1) {
 934                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 935                 if (ret == -1) {
 936                         ret = -errno;
 937                         goto out;
 938                 }
 939                 if (ret)
 940                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 941                                          &peer_size);
 942                 gettimeofday(&tv, NULL);
 943                 this = (this + 1)%2;
 944                 last = (last + 1)%2;
 945                 for (i = 0; i < ndev; ++i) {
 946                         sp = &spc->progress[this * ndev + i];
 947                         sp_last = &spc->progress[last * ndev + i];
 948                         sp_shared = &spc->shared_progress[i];
 949                         if (sp->stats.finished)
 950                                 continue;
 951                         progress_one_dev(sp);
 952                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 953                         if (!sp->ret)
 954                                 continue;
 955                         if (sp->ioctl_errno != ENOTCONN &&
 956                             sp->ioctl_errno != ENODEV) {
 957                                 ret = -sp->ioctl_errno;
 958                                 goto out;
 959                         }
 960                         /*
 961                          * scrub finished or device removed, check the
 962                          * finished flag. if unset, just use the last
 963                          * result we got for the current write and go
 964                          * on. flag should be set on next cycle, then.
 965                          */
 966                         perr = pthread_setcancelstate(
 967                                         PTHREAD_CANCEL_DISABLE, &old);
 968                         if (perr)
 969                                 goto out;
 970                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 971                         if (perr)
 972                                 goto out;
 973                         if (!sp_shared->stats.finished) {
 974                                 perr = pthread_mutex_unlock(
 975                                                 &sp_shared->progress_mutex);
 976                                 if (perr)
 977                                         goto out;
 978                                 perr = pthread_setcancelstate(
 979                                                 PTHREAD_CANCEL_ENABLE, &old);
 980                                 if (perr)
 981                                         goto out;
 982                                 memcpy(sp, sp_last, sizeof(*sp));
 983                                 continue;
 984                         }
 985                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 986                         if (perr)
 987                                 goto out;
 988                         perr = pthread_setcancelstate(
 989                                         PTHREAD_CANCEL_ENABLE, &old);
 990                         if (perr)
 991                                 goto out;
 992                         memcpy(sp, sp_shared, sizeof(*sp));
 993                         memcpy(sp_last, sp_shared, sizeof(*sp));
 994                 }
 995                 if (peer_fd != -1) {
 996                         write_poll_fd.fd = peer_fd;
 997                         ret = poll(&write_poll_fd, 1, 0);
 998                         if (ret == -1) {
 999                                 ret = -errno;
1000                                 goto out;
1001                         }
1002                         if (ret) {
1003                                 ret = scrub_write_file(
1004                                         peer_fd, fsid,
1005                                         &spc->progress[this * ndev], ndev);
1006                                 if (ret)
1007                                         goto out;
1008                         }
1009                         close(peer_fd);
1010                         peer_fd = -1;
1011                 }
1012                 if (!spc->do_record)
1013                         continue;
1014                 ret = scrub_write_progress(spc->write_mutex, fsid,
1015                                            &spc->progress[this * ndev], ndev);
1016                 if (ret)
1017                         goto out;
1018         }
1019 out:
1020         if (peer_fd != -1)
1021                 close(peer_fd);
1022         if (perr)
1023                 ret = -perr;
1024         return ERR_PTR(ret);
1025 }
1026
1027 static struct scrub_file_record *last_dev_scrub(
1028                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1029 {
1030         int i;
1031
1032         if (!past_scrubs || IS_ERR(past_scrubs))
1033                 return NULL;
1034
1035         for (i = 0; past_scrubs[i]; ++i)
1036                 if (past_scrubs[i]->devid == devid)
1037                         return past_scrubs[i];
1038
1039         return NULL;
1040 }
1041
1042 static int mkdir_p(char *path)
1043 {
1044         int i;
1045         int ret;
1046
1047         for (i = 1; i < strlen(path); ++i) {
1048                 if (path[i] != '/')
1049                         continue;
1050                 path[i] = '\0';
1051                 ret = mkdir(path, 0777);
1052                 if (ret && errno != EEXIST)
1053                         return -errno;
1054                 path[i] = '/';
1055         }
1056
1057         return 0;
1058 }
1059
1060 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1061                                   struct btrfs_ioctl_dev_info_args *di_args,
1062                                   struct scrub_file_record **past_scrubs)
1063 {
1064         int i;
1065
1066         if (!fi_args || !di_args || !past_scrubs)
1067                 return 0;
1068
1069         for (i = 0; i < fi_args->num_devices; i++) {
1070                 struct scrub_file_record *sfr =
1071                         last_dev_scrub(past_scrubs, di_args[i].devid);
1072
1073                 if (!sfr)
1074                         continue;
1075                 if (!(sfr->stats.finished || sfr->stats.canceled))
1076                         return 1;
1077         }
1078         return 0;
1079 }
1080
1081 static int is_scrub_running_in_kernel(int fd,
1082                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1083 {
1084         struct scrub_progress sp;
1085         int i;
1086         int ret;
1087
1088         for (i = 0; i < max_devices; i++) {
1089                 memset(&sp, 0, sizeof(sp));
1090                 sp.scrub_args.devid = di_args[i].devid;
1091                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1092                 if (!ret)
1093                         return 1;
1094         }
1095
1096         return 0;
1097 }
1098
1099 static const char * const cmd_scrub_start_usage[];
1100 static const char * const cmd_scrub_resume_usage[];
1101
1102 static int scrub_start(int argc, char **argv, int resume)
1103 {
1104         int fdmnt;
1105         int prg_fd = -1;
1106         int fdres = -1;
1107         int ret;
1108         pid_t pid;
1109         int c;
1110         int i;
1111         int err = 0;
1112         int e_uncorrectable = 0;
1113         int e_correctable = 0;
1114         int print_raw = 0;
1115         char *path;
1116         int do_background = 1;
1117         int do_wait = 0;
1118         int do_print = 0;
1119         int do_quiet = 0;
1120         int do_record = 1;
1121         int readonly = 0;
1122         int do_stats_per_dev = 0;
1123         int ioprio_class = IOPRIO_CLASS_IDLE;
1124         int ioprio_classdata = 0;
1125         int n_start = 0;
1126         int n_skip = 0;
1127         int n_resume = 0;
1128         struct btrfs_ioctl_fs_info_args fi_args;
1129         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1130         struct scrub_progress *sp = NULL;
1131         struct scrub_fs_stat fs_stat;
1132         struct timeval tv;
1133         struct sockaddr_un addr = {
1134                 .sun_family = AF_UNIX,
1135         };
1136         pthread_t *t_devs = NULL;
1137         pthread_t t_prog;
1138         struct scrub_file_record **past_scrubs = NULL;
1139         struct scrub_file_record *last_scrub = NULL;
1140         char *datafile = strdup(SCRUB_DATA_FILE);
1141         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1142         char sock_path[PATH_MAX] = "";
1143         struct scrub_progress_cycle spc;
1144         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1145         void *terr;
1146         u64 devid;
1147         DIR *dirstream = NULL;
1148         int force = 0;
1149         int nothing_to_resume = 0;
1150
1151         optind = 1;
1152         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1153                 switch (c) {
1154                 case 'B':
1155                         do_background = 0;
1156                         do_wait = 1;
1157                         do_print = 1;
1158                         break;
1159                 case 'd':
1160                         do_stats_per_dev = 1;
1161                         break;
1162                 case 'q':
1163                         do_quiet = 1;
1164                         break;
1165                 case 'r':
1166                         readonly = 1;
1167                         break;
1168                 case 'R':
1169                         print_raw = 1;
1170                         break;
1171                 case 'c':
1172                         ioprio_class = (int)strtol(optarg, NULL, 10);
1173                         break;
1174                 case 'n':
1175                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1176                         break;
1177                 case 'f':
1178                         force = 1;
1179                         break;
1180                 case '?':
1181                 default:
1182                         usage(resume ? cmd_scrub_resume_usage :
1183                                                 cmd_scrub_start_usage);
1184                 }
1185         }
1186
1187         /* try to catch most error cases before forking */
1188
1189         if (check_argc_exact(argc - optind, 1)) {
1190                 usage(resume ? cmd_scrub_resume_usage :
1191                                         cmd_scrub_start_usage);
1192         }
1193
1194         spc.progress = NULL;
1195         if (do_quiet && do_print)
1196                 do_print = 0;
1197
1198         if (mkdir_p(datafile)) {
1199                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1200                                "file, mkdir %s failed: %s. Status recording "
1201                                "disabled\n", datafile, strerror(errno));
1202                 do_record = 0;
1203         }
1204         free(datafile);
1205
1206         path = argv[optind];
1207
1208         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1209
1210         if (fdmnt < 0) {
1211                 if (errno == EINVAL)
1212                         ERR(!do_quiet,
1213                             "ERROR: '%s' is not a mounted btrfs device\n",
1214                             path);
1215                 else
1216                         ERR(!do_quiet, "ERROR: can't access '%s': %s\n",
1217                             path, strerror(errno));
1218                 return 1;
1219         }
1220
1221         ret = get_fs_info(path, &fi_args, &di_args);
1222         if (ret) {
1223                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1224                     "%s\n", strerror(-ret));
1225                 err = 1;
1226                 goto out;
1227         }
1228         if (!fi_args.num_devices) {
1229                 ERR(!do_quiet, "ERROR: no devices found\n");
1230                 err = 1;
1231                 goto out;
1232         }
1233
1234         uuid_unparse(fi_args.fsid, fsid);
1235         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1236         if (fdres < 0 && fdres != -ENOENT) {
1237                 ERR(!do_quiet, "WARNING: failed to open status file: "
1238                     "%s\n", strerror(-fdres));
1239         } else if (fdres >= 0) {
1240                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1241                 if (IS_ERR(past_scrubs))
1242                         ERR(!do_quiet, "WARNING: failed to read status file: "
1243                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1244                 close(fdres);
1245         }
1246
1247         /*
1248          * Check for stale information in the status file, ie. if it's
1249          * canceled=0, finished=0 but no scrub is running.
1250          */
1251         if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1252                 force = 1;
1253
1254         /*
1255          * check whether any involved device is already busy running a
1256          * scrub. This would cause damaged status messages and the state
1257          * "aborted" without the explanation that a scrub was already
1258          * running. Therefore check it first, prevent it and give some
1259          * feedback to the user if scrub is already running.
1260          * Note that if scrub is started with a block device as the
1261          * parameter, only that particular block device is checked. It
1262          * is a normal mode of operation to start scrub on multiple
1263          * single devices, there is no reason to prevent this.
1264          */
1265         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1266                 ERR(!do_quiet,
1267                     "ERROR: scrub is already running.\n"
1268                     "To cancel use 'btrfs scrub cancel %s'.\n"
1269                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1270                     path, path);
1271                 err = 1;
1272                 goto out;
1273         }
1274
1275         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1276         sp = calloc(fi_args.num_devices, sizeof(*sp));
1277         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1278
1279         if (!t_devs || !sp || !spc.progress) {
1280                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1281                 err = 1;
1282                 goto out;
1283         }
1284
1285         for (i = 0; i < fi_args.num_devices; ++i) {
1286                 devid = di_args[i].devid;
1287                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1288                 if (ret) {
1289                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1290                             "%s\n", strerror(ret));
1291                         err = 1;
1292                         goto out;
1293                 }
1294                 last_scrub = last_dev_scrub(past_scrubs, devid);
1295                 sp[i].scrub_args.devid = devid;
1296                 sp[i].fd = fdmnt;
1297                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1298                                              !last_scrub->stats.finished)) {
1299                         ++n_resume;
1300                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1301                         sp[i].resumed = last_scrub;
1302                 } else if (resume) {
1303                         ++n_skip;
1304                         sp[i].skip = 1;
1305                         sp[i].resumed = last_scrub;
1306                         continue;
1307                 } else {
1308                         ++n_start;
1309                         sp[i].scrub_args.start = 0ll;
1310                         sp[i].resumed = NULL;
1311                 }
1312                 sp[i].skip = 0;
1313                 sp[i].scrub_args.end = (u64)-1ll;
1314                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1315                 sp[i].ioprio_class = ioprio_class;
1316                 sp[i].ioprio_classdata = ioprio_classdata;
1317         }
1318
1319         if (!n_start && !n_resume) {
1320                 if (!do_quiet)
1321                         printf("scrub: nothing to resume for %s, fsid %s\n",
1322                                path, fsid);
1323                 nothing_to_resume = 1;
1324                 goto out;
1325         }
1326
1327         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1328         while (ret != -1) {
1329                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1330                                         sock_path, sizeof(sock_path));
1331                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1332                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1333                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1334                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1335                 if (ret != -1 || errno != EADDRINUSE)
1336                         break;
1337                 /*
1338                  * bind failed with EADDRINUSE. so let's see if anyone answers
1339                  * when we make a call to the socket ...
1340                  */
1341                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1342                 if (!ret || errno != ECONNREFUSED) {
1343                         /* ... yes, so scrub must be running. error out */
1344                         fprintf(stderr, "ERROR: scrub already running\n");
1345                         close(prg_fd);
1346                         prg_fd = -1;
1347                         goto out;
1348                 }
1349                 /*
1350                  * ... no, this means someone left us alone with an unused
1351                  * socket in the file system. remove it and try again.
1352                  */
1353                 ret = unlink(sock_path);
1354         }
1355         if (ret != -1)
1356                 ret = listen(prg_fd, 100);
1357         if (ret == -1) {
1358                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1359                     "socket at %s: %s. Progress cannot be queried\n",
1360                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1361                     strerror(errno));
1362                 if (prg_fd != -1) {
1363                         close(prg_fd);
1364                         prg_fd = -1;
1365                         if (sock_path[0])
1366                                 unlink(sock_path);
1367                 }
1368         }
1369
1370         if (do_record) {
1371                 /* write all-zero progress file for a start */
1372                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1373                                            fi_args.num_devices);
1374                 if (ret) {
1375                         ERR(!do_quiet, "WARNING: failed to write the progress "
1376                             "status file: %s. Status recording disabled\n",
1377                             strerror(-ret));
1378                         do_record = 0;
1379                 }
1380         }
1381
1382         if (do_background) {
1383                 pid = fork();
1384                 if (pid == -1) {
1385                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1386                                         "%s\n", strerror(errno));
1387                         err = 1;
1388                         goto out;
1389                 }
1390
1391                 if (pid) {
1392                         int stat;
1393                         scrub_handle_sigint_parent();
1394                         if (!do_quiet)
1395                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1396                                        n_start ? "started" : "resumed",
1397                                        path, fsid, pid);
1398                         if (!do_wait) {
1399                                 err = 0;
1400                                 goto out;
1401                         }
1402                         ret = wait(&stat);
1403                         if (ret != pid) {
1404                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1405                                     "%s\n", ret, strerror(errno));
1406                                 err = 1;
1407                                 goto out;
1408                         }
1409                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1410                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1411                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1412                                 goto out;
1413                         }
1414                         err = 0;
1415                         goto out;
1416                 }
1417         }
1418
1419         scrub_handle_sigint_child(fdmnt);
1420
1421         for (i = 0; i < fi_args.num_devices; ++i) {
1422                 if (sp[i].skip) {
1423                         sp[i].scrub_args.progress = sp[i].resumed->p;
1424                         sp[i].stats = sp[i].resumed->stats;
1425                         sp[i].ret = 0;
1426                         sp[i].stats.finished = 1;
1427                         continue;
1428                 }
1429                 devid = di_args[i].devid;
1430                 gettimeofday(&tv, NULL);
1431                 sp[i].stats.t_start = tv.tv_sec;
1432                 ret = pthread_create(&t_devs[i], NULL,
1433                                         scrub_one_dev, &sp[i]);
1434                 if (ret) {
1435                         if (do_print)
1436                                 fprintf(stderr, "ERROR: creating "
1437                                         "scrub_one_dev[%llu] thread failed: "
1438                                         "%s\n", devid, strerror(ret));
1439                         err = 1;
1440                         goto out;
1441                 }
1442         }
1443
1444         spc.fdmnt = fdmnt;
1445         spc.prg_fd = prg_fd;
1446         spc.do_record = do_record;
1447         spc.write_mutex = &spc_write_mutex;
1448         spc.shared_progress = sp;
1449         spc.fi = &fi_args;
1450         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1451         if (ret) {
1452                 if (do_print)
1453                         fprintf(stderr, "ERROR: creating progress thread "
1454                                 "failed: %s\n", strerror(ret));
1455                 err = 1;
1456                 goto out;
1457         }
1458
1459         err = 0;
1460         for (i = 0; i < fi_args.num_devices; ++i) {
1461                 if (sp[i].skip)
1462                         continue;
1463                 devid = di_args[i].devid;
1464                 ret = pthread_join(t_devs[i], NULL);
1465                 if (ret) {
1466                         if (do_print)
1467                                 fprintf(stderr, "ERROR: pthread_join failed "
1468                                         "for scrub_one_dev[%llu]: %s\n", devid,
1469                                         strerror(ret));
1470                         ++err;
1471                         continue;
1472                 }
1473                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1474                         if (do_print)
1475                                 fprintf(stderr, "WARNING: device %lld not "
1476                                         "present\n", devid);
1477                         continue;
1478                 }
1479                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1480                         ++err;
1481                 } else if (sp[i].ret) {
1482                         if (do_print)
1483                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1484                                         "for device id %lld (%s)\n", path,
1485                                         devid, strerror(sp[i].ioctl_errno));
1486                         ++err;
1487                         continue;
1488                 }
1489                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1490                         e_uncorrectable++;
1491                 if (sp[i].scrub_args.progress.corrected_errors > 0
1492                     || sp[i].scrub_args.progress.unverified_errors > 0)
1493                         e_correctable++;
1494         }
1495
1496         if (do_print) {
1497                 const char *append = "done";
1498                 if (!do_stats_per_dev)
1499                         init_fs_stat(&fs_stat);
1500                 for (i = 0; i < fi_args.num_devices; ++i) {
1501                         if (do_stats_per_dev) {
1502                                 print_scrub_dev(&di_args[i],
1503                                                 &sp[i].scrub_args.progress,
1504                                                 print_raw,
1505                                                 sp[i].ret ? "canceled" : "done",
1506                                                 &sp[i].stats);
1507                         } else {
1508                                 if (sp[i].ret)
1509                                         append = "canceled";
1510                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1511                                                 &sp[i].stats, &fs_stat);
1512                         }
1513                 }
1514                 if (!do_stats_per_dev) {
1515                         printf("scrub %s for %s\n", append, fsid);
1516                         print_fs_stat(&fs_stat, print_raw);
1517                 }
1518         }
1519
1520         ret = pthread_cancel(t_prog);
1521         if (!ret)
1522                 ret = pthread_join(t_prog, &terr);
1523
1524         /* check for errors from the handling of the progress thread */
1525         if (do_print && ret) {
1526                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1527                         strerror(ret));
1528         }
1529
1530         /* check for errors returned from the progress thread itself */
1531         if (do_print && terr && terr != PTHREAD_CANCELED) {
1532                 fprintf(stderr, "ERROR: recording progress "
1533                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1534         }
1535
1536         if (do_record) {
1537                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1538                                            fi_args.num_devices);
1539                 if (ret && do_print) {
1540                         fprintf(stderr, "ERROR: failed to record the result: "
1541                                 "%s\n", strerror(-ret));
1542                 }
1543         }
1544
1545         scrub_handle_sigint_child(-1);
1546
1547 out:
1548         free_history(past_scrubs);
1549         free(di_args);
1550         free(t_devs);
1551         free(sp);
1552         free(spc.progress);
1553         if (prg_fd > -1) {
1554                 close(prg_fd);
1555                 if (sock_path[0])
1556                         unlink(sock_path);
1557         }
1558         close_file_or_dir(fdmnt, dirstream);
1559
1560         if (err)
1561                 return 1;
1562         if (nothing_to_resume)
1563                 return 2;
1564         if (e_uncorrectable) {
1565                 ERR(!do_quiet, "ERROR: There are uncorrectable errors.\n");
1566                 return 3;
1567         }
1568         if (e_correctable)
1569                 ERR(!do_quiet, "WARNING: errors detected during scrubbing, corrected.\n");
1570
1571         return 0;
1572 }
1573
1574 static const char * const cmd_scrub_start_usage[] = {
1575         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1576         "Start a new scrub. If a scrub is already running, the new one fails.",
1577         "",
1578         "-B     do not background",
1579         "-d     stats per device (-B only)",
1580         "-q     be quiet",
1581         "-r     read only mode",
1582         "-R     raw print mode, print full data instead of summary",
1583         "-c     set ioprio class (see ionice(1) manpage)",
1584         "-n     set ioprio classdata (see ionice(1) manpage)",
1585         "-f     force starting new scrub even if a scrub is already running",
1586         "       this is useful when scrub stats record file is damaged",
1587         NULL
1588 };
1589
1590 static int cmd_scrub_start(int argc, char **argv)
1591 {
1592         return scrub_start(argc, argv, 0);
1593 }
1594
1595 static const char * const cmd_scrub_cancel_usage[] = {
1596         "btrfs scrub cancel <path>|<device>",
1597         "Cancel a running scrub",
1598         NULL
1599 };
1600
1601 static int cmd_scrub_cancel(int argc, char **argv)
1602 {
1603         char *path;
1604         int ret;
1605         int fdmnt = -1;
1606         DIR *dirstream = NULL;
1607
1608         if (check_argc_exact(argc, 2))
1609                 usage(cmd_scrub_cancel_usage);
1610
1611         path = argv[1];
1612
1613         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1614         if (fdmnt < 0) {
1615                 if (errno == EINVAL)
1616                         fprintf(stderr,
1617                                 "ERROR: '%s' is not a mounted btrfs device\n",
1618                                 path);
1619                 else
1620                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1621                                 path, strerror(errno));
1622                 ret = 1;
1623                 goto out;
1624         }
1625
1626         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1627
1628         if (ret < 0) {
1629                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1630                         errno == ENOTCONN ? "not running" : strerror(errno));
1631                 if (errno == ENOTCONN)
1632                         ret = 2;
1633                 else
1634                         ret = 1;
1635                 goto out;
1636         }
1637
1638         ret = 0;
1639         printf("scrub cancelled\n");
1640
1641 out:
1642         close_file_or_dir(fdmnt, dirstream);
1643         return ret;
1644 }
1645
1646 static const char * const cmd_scrub_resume_usage[] = {
1647         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1648         "Resume previously canceled or interrupted scrub",
1649         "",
1650         "-B     do not background",
1651         "-d     stats per device (-B only)",
1652         "-q     be quiet",
1653         "-r     read only mode",
1654         "-R     raw print mode, print full data instead of summary",
1655         "-c     set ioprio class (see ionice(1) manpage)",
1656         "-n     set ioprio classdata (see ionice(1) manpage)",
1657         NULL
1658 };
1659
1660 static int cmd_scrub_resume(int argc, char **argv)
1661 {
1662         return scrub_start(argc, argv, 1);
1663 }
1664
1665 static const char * const cmd_scrub_status_usage[] = {
1666         "btrfs scrub status [-dR] <path>|<device>",
1667         "Show status of running or finished scrub",
1668         "",
1669         "-d     stats per device",
1670         "-R     print raw stats",
1671         NULL
1672 };
1673
1674 static int cmd_scrub_status(int argc, char **argv)
1675 {
1676         char *path;
1677         struct btrfs_ioctl_fs_info_args fi_args;
1678         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1679         struct scrub_file_record **past_scrubs = NULL;
1680         struct scrub_file_record *last_scrub;
1681         struct scrub_fs_stat fs_stat;
1682         struct sockaddr_un addr = {
1683                 .sun_family = AF_UNIX,
1684         };
1685         int in_progress;
1686         int ret;
1687         int i;
1688         int fdmnt;
1689         int print_raw = 0;
1690         int do_stats_per_dev = 0;
1691         int c;
1692         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1693         int fdres = -1;
1694         int err = 0;
1695         DIR *dirstream = NULL;
1696
1697         optind = 1;
1698         while ((c = getopt(argc, argv, "dR")) != -1) {
1699                 switch (c) {
1700                 case 'd':
1701                         do_stats_per_dev = 1;
1702                         break;
1703                 case 'R':
1704                         print_raw = 1;
1705                         break;
1706                 case '?':
1707                 default:
1708                         usage(cmd_scrub_status_usage);
1709                 }
1710         }
1711
1712         if (check_argc_exact(argc - optind, 1))
1713                 usage(cmd_scrub_status_usage);
1714
1715         path = argv[optind];
1716
1717         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1718
1719         if (fdmnt < 0) {
1720                 if (errno == EINVAL)
1721                         fprintf(stderr,
1722                                 "ERROR: '%s' is not a mounted btrfs device\n",
1723                                 path);
1724                 else
1725                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1726                                 path, strerror(errno));
1727                 return 1;
1728         }
1729
1730         ret = get_fs_info(path, &fi_args, &di_args);
1731         if (ret) {
1732                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1733                                 "%s\n", strerror(-ret));
1734                 err = 1;
1735                 goto out;
1736         }
1737         if (!fi_args.num_devices) {
1738                 fprintf(stderr, "ERROR: no devices found\n");
1739                 err = 1;
1740                 goto out;
1741         }
1742
1743         uuid_unparse(fi_args.fsid, fsid);
1744
1745         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1746         if (fdres == -1) {
1747                 fprintf(stderr, "ERROR: failed to create socket to "
1748                         "receive progress information: %s\n",
1749                         strerror(errno));
1750                 err = 1;
1751                 goto out;
1752         }
1753         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1754                         NULL, addr.sun_path, sizeof(addr.sun_path));
1755         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1756         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1757         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1758         if (ret == -1) {
1759                 close(fdres);
1760                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1761                 if (fdres < 0 && fdres != -ENOENT) {
1762                         fprintf(stderr, "WARNING: failed to open status file: "
1763                                 "%s\n", strerror(-fdres));
1764                         err = 1;
1765                         goto out;
1766                 }
1767         }
1768
1769         if (fdres >= 0) {
1770                 past_scrubs = scrub_read_file(fdres, 1);
1771                 if (IS_ERR(past_scrubs))
1772                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1773                                 strerror(-PTR_ERR(past_scrubs)));
1774         }
1775         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1776
1777         printf("scrub status for %s\n", fsid);
1778
1779         if (do_stats_per_dev) {
1780                 for (i = 0; i < fi_args.num_devices; ++i) {
1781                         last_scrub = last_dev_scrub(past_scrubs,
1782                                                         di_args[i].devid);
1783                         if (!last_scrub) {
1784                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1785                                                 NULL, NULL);
1786                                 continue;
1787                         }
1788                         last_scrub->stats.in_progress = in_progress;
1789                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1790                                         last_scrub->stats.finished ?
1791                                                         "history" : "status",
1792                                         &last_scrub->stats);
1793                 }
1794         } else {
1795                 init_fs_stat(&fs_stat);
1796                 fs_stat.s.in_progress = in_progress;
1797                 for (i = 0; i < fi_args.num_devices; ++i) {
1798                         last_scrub = last_dev_scrub(past_scrubs,
1799                                                         di_args[i].devid);
1800                         if (!last_scrub)
1801                                 continue;
1802                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1803                                         &fs_stat);
1804                 }
1805                 print_fs_stat(&fs_stat, print_raw);
1806         }
1807
1808 out:
1809         free_history(past_scrubs);
1810         free(di_args);
1811         if (fdres > -1)
1812                 close(fdres);
1813         close_file_or_dir(fdmnt, dirstream);
1814
1815         return !!err;
1816 }
1817
1818 static const char scrub_cmd_group_info[] =
1819 "verify checksums of data and metadata";
1820
1821 const struct cmd_group scrub_cmd_group = {
1822         scrub_cmd_group_usage, scrub_cmd_group_info, {
1823                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1824                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1825                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1826                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1827                 NULL_CMD_STRUCT
1828         }
1829 };
1830
1831 int cmd_scrub(int argc, char **argv)
1832 {
1833         return handle_command_group(&scrub_cmd_group, argc, argv);
1834 }