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
  48 static const char * const scrub_cmd_group_usage[] = {
  49         "btrfs scrub <command> [options] <path>|<device>",
  50         NULL
  51 };
  52
  53 #define SCRUB_DATA_FILE "/var/lib/btrfs/scrub.status"
  54 #define SCRUB_PROGRESS_SOCKET_PATH "/var/lib/btrfs/scrub.progress"
  55 #define SCRUB_FILE_VERSION_PREFIX "scrub status"
  56 #define SCRUB_FILE_VERSION "1"
  57
  58 struct scrub_stats {
  59         time_t t_start;
  60         time_t t_resumed;
  61         u64 duration;
  62         u64 finished;
  63         u64 canceled;
  64         int in_progress;
  65 };
  66
  67 /* TBD: replace with #include "linux/ioprio.h" in some years */
  68 #if !defined (IOPRIO_H)
  69 #define IOPRIO_WHO_PROCESS 1
  70 #define IOPRIO_CLASS_SHIFT 13
  71 #define IOPRIO_PRIO_VALUE(class, data) \
  72                 (((class) << IOPRIO_CLASS_SHIFT) | (data))
  73 #define IOPRIO_CLASS_IDLE 3
  74 #endif
  75
  76 struct scrub_progress {
  77         struct btrfs_ioctl_scrub_args scrub_args;
  78         int fd;
  79         int ret;
  80         int skip;
  81         struct scrub_stats stats;
  82         struct scrub_file_record *resumed;
  83         int ioctl_errno;
  84         pthread_mutex_t progress_mutex;
  85         int ioprio_class;
  86         int ioprio_classdata;
  87 };
  88
  89 struct scrub_file_record {
  90         u8 fsid[BTRFS_FSID_SIZE];
  91         u64 devid;
  92         struct scrub_stats stats;
  93         struct btrfs_scrub_progress p;
  94 };
  95
  96 struct scrub_progress_cycle {
  97         int fdmnt;
  98         int prg_fd;
  99         int do_record;
 100         struct btrfs_ioctl_fs_info_args *fi;
 101         struct scrub_progress *progress;
 102         struct scrub_progress *shared_progress;
 103         pthread_mutex_t *write_mutex;
 104 };
 105
 106 struct scrub_fs_stat {
 107         struct btrfs_scrub_progress p;
 108         struct scrub_stats s;
 109         int i;
 110 };
 111
 112 static void print_scrub_full(struct btrfs_scrub_progress *sp)
 113 {
 114         printf("\tdata_extents_scrubbed: %lld\n", sp->data_extents_scrubbed);
 115         printf("\ttree_extents_scrubbed: %lld\n", sp->tree_extents_scrubbed);
 116         printf("\tdata_bytes_scrubbed: %lld\n", sp->data_bytes_scrubbed);
 117         printf("\ttree_bytes_scrubbed: %lld\n", sp->tree_bytes_scrubbed);
 118         printf("\tread_errors: %lld\n", sp->read_errors);
 119         printf("\tcsum_errors: %lld\n", sp->csum_errors);
 120         printf("\tverify_errors: %lld\n", sp->verify_errors);
 121         printf("\tno_csum: %lld\n", sp->no_csum);
 122         printf("\tcsum_discards: %lld\n", sp->csum_discards);
 123         printf("\tsuper_errors: %lld\n", sp->super_errors);
 124         printf("\tmalloc_errors: %lld\n", sp->malloc_errors);
 125         printf("\tuncorrectable_errors: %lld\n", sp->uncorrectable_errors);
 126         printf("\tunverified_errors: %lld\n", sp->unverified_errors);
 127         printf("\tcorrected_errors: %lld\n", sp->corrected_errors);
 128         printf("\tlast_physical: %lld\n", sp->last_physical);
 129 }
 130
 131 #define ERR(test, ...) do {                     \
 132         if (test)                               \
 133                 fprintf(stderr, __VA_ARGS__);   \
 134 } while (0)
 135
 136 #define PRINT_SCRUB_ERROR(test, desc) do {      \
 137         if (test)                               \
 138                 printf(" %s=%llu", desc, test); \
 139 } while (0)
 140
 141 static void print_scrub_summary(struct btrfs_scrub_progress *p)
 142 {
 143         u64 err_cnt;
 144         u64 err_cnt2;
 145
 146         err_cnt = p->read_errors +
 147                         p->csum_errors +
 148                         p->verify_errors +
 149                         p->super_errors;
 150
 151         err_cnt2 = p->corrected_errors + p->uncorrectable_errors;
 152
 153         if (p->malloc_errors)
 154                 printf("*** WARNING: memory allocation failed while scrubbing. "
 155                        "results may be inaccurate\n");
 156
 157         printf("\ttotal bytes scrubbed: %s with %llu errors\n",
 158                 pretty_size(p->data_bytes_scrubbed + p->tree_bytes_scrubbed),
 159                 max(err_cnt, err_cnt2));
 160
 161         if (err_cnt || err_cnt2) {
 162                 printf("\terror details:");
 163                 PRINT_SCRUB_ERROR(p->read_errors, "read");
 164                 PRINT_SCRUB_ERROR(p->super_errors, "super");
 165                 PRINT_SCRUB_ERROR(p->verify_errors, "verify");
 166                 PRINT_SCRUB_ERROR(p->csum_errors, "csum");
 167                 printf("\n");
 168                 printf("\tcorrected errors: %llu, uncorrectable errors: %llu, "
 169                         "unverified errors: %llu\n", p->corrected_errors,
 170                         p->uncorrectable_errors, p->unverified_errors);
 171         }
 172 }
 173
 174 #define _SCRUB_FS_STAT(p, name, fs_stat) do {   \
 175         fs_stat->p.name += p->name;             \
 176 } while (0)
 177
 178 #define _SCRUB_FS_STAT_MIN(ss, name, fs_stat)   \
 179 do {                                            \
 180         if (fs_stat->s.name > ss->name) {       \
 181                 fs_stat->s.name = ss->name;     \
 182         }                                       \
 183 } while (0)
 184
 185 #define _SCRUB_FS_STAT_ZMIN(ss, name, fs_stat)                  \
 186 do {                                                            \
 187         if (!fs_stat->s.name || fs_stat->s.name > ss->name) {   \
 188                 fs_stat->s.name = ss->name;                     \
 189         }                                                       \
 190 } while (0)
 191
 192 #define _SCRUB_FS_STAT_ZMAX(ss, name, fs_stat)                          \
 193 do {                                                                    \
 194         if (!(fs_stat)->s.name || (fs_stat)->s.name < (ss)->name) {     \
 195                 (fs_stat)->s.name = (ss)->name;                         \
 196         }                                                               \
 197 } while (0)
 198
 199 static void add_to_fs_stat(struct btrfs_scrub_progress *p,
 200                                 struct scrub_stats *ss,
 201                                 struct scrub_fs_stat *fs_stat)
 202 {
 203         _SCRUB_FS_STAT(p, data_extents_scrubbed, fs_stat);
 204         _SCRUB_FS_STAT(p, tree_extents_scrubbed, fs_stat);
 205         _SCRUB_FS_STAT(p, data_bytes_scrubbed, fs_stat);
 206         _SCRUB_FS_STAT(p, tree_bytes_scrubbed, fs_stat);
 207         _SCRUB_FS_STAT(p, read_errors, fs_stat);
 208         _SCRUB_FS_STAT(p, csum_errors, fs_stat);
 209         _SCRUB_FS_STAT(p, verify_errors, fs_stat);
 210         _SCRUB_FS_STAT(p, no_csum, fs_stat);
 211         _SCRUB_FS_STAT(p, csum_discards, fs_stat);
 212         _SCRUB_FS_STAT(p, super_errors, fs_stat);
 213         _SCRUB_FS_STAT(p, malloc_errors, fs_stat);
 214         _SCRUB_FS_STAT(p, uncorrectable_errors, fs_stat);
 215         _SCRUB_FS_STAT(p, corrected_errors, fs_stat);
 216         _SCRUB_FS_STAT(p, last_physical, fs_stat);
 217         _SCRUB_FS_STAT_ZMIN(ss, t_start, fs_stat);
 218         _SCRUB_FS_STAT_ZMIN(ss, t_resumed, fs_stat);
 219         _SCRUB_FS_STAT_ZMAX(ss, duration, fs_stat);
 220         _SCRUB_FS_STAT_ZMAX(ss, canceled, fs_stat);
 221         _SCRUB_FS_STAT_MIN(ss, finished, fs_stat);
 222 }
 223
 224 static void init_fs_stat(struct scrub_fs_stat *fs_stat)
 225 {
 226         memset(fs_stat, 0, sizeof(*fs_stat));
 227         fs_stat->s.finished = 1;
 228 }
 229
 230 static void _print_scrub_ss(struct scrub_stats *ss)
 231 {
 232         char t[4096];
 233         struct tm tm;
 234         time_t seconds;
 235         unsigned hours;
 236
 237         if (!ss || !ss->t_start) {
 238                 printf("\tno stats available\n");
 239                 return;
 240         }
 241         if (ss->t_resumed) {
 242                 localtime_r(&ss->t_resumed, &tm);
 243                 strftime(t, sizeof(t), "%c", &tm);
 244                 t[sizeof(t) - 1] = '\0';
 245                 printf("\tscrub resumed at %s", t);
 246         } else {
 247                 localtime_r(&ss->t_start, &tm);
 248                 strftime(t, sizeof(t), "%c", &tm);
 249                 t[sizeof(t) - 1] = '\0';
 250                 printf("\tscrub started at %s", t);
 251         }
 252
 253         seconds = ss->duration;
 254         hours = ss->duration / (60 * 60);
 255         gmtime_r(&seconds, &tm);
 256         strftime(t, sizeof(t), "%M:%S", &tm);
 257         if (ss->finished && !ss->canceled) {
 258                 printf(" and finished after %02u:%s\n", hours, t);
 259         } else if (ss->canceled) {
 260                 printf(" and was aborted after %02u:%s\n", hours, t);
 261         } else {
 262                 if (ss->in_progress)
 263                         printf(", running for %02u:%s\n", hours, t);
 264                 else
 265                         printf(", interrupted after %02u:%s, not running\n",
 266                                         hours, t);
 267         }
 268 }
 269
 270 static void print_scrub_dev(struct btrfs_ioctl_dev_info_args *di,
 271                                 struct btrfs_scrub_progress *p, int raw,
 272                                 const char *append, struct scrub_stats *ss)
 273 {
 274         printf("scrub device %s (id %llu) %s\n", di->path, di->devid,
 275                append ? append : "");
 276
 277         _print_scrub_ss(ss);
 278
 279         if (p) {
 280                 if (raw)
 281                         print_scrub_full(p);
 282                 else
 283                         print_scrub_summary(p);
 284         }
 285 }
 286
 287 static void print_fs_stat(struct scrub_fs_stat *fs_stat, int raw)
 288 {
 289         _print_scrub_ss(&fs_stat->s);
 290
 291         if (raw)
 292                 print_scrub_full(&fs_stat->p);
 293         else
 294                 print_scrub_summary(&fs_stat->p);
 295 }
 296
 297 static void free_history(struct scrub_file_record **last_scrubs)
 298 {
 299         struct scrub_file_record **l = last_scrubs;
 300         if (!l || IS_ERR(l))
 301                 return;
 302         while (*l)
 303                 free(*l++);
 304         free(last_scrubs);
 305 }
 306
 307 /*
 308  * cancels a running scrub and makes the master process record the current
 309  * progress status before exiting.
 310  */
 311 static int cancel_fd = -1;
 312 static void scrub_sigint_record_progress(int signal)
 313 {
 314         int ret;
 315
 316         ret = ioctl(cancel_fd, BTRFS_IOC_SCRUB_CANCEL, NULL);
 317         if (ret < 0)
 318                 perror("Scrub cancel failed");
 319 }
 320
 321 static int scrub_handle_sigint_parent(void)
 322 {
 323         struct sigaction sa = {
 324                 .sa_handler = SIG_IGN,
 325                 .sa_flags = SA_RESTART,
 326         };
 327
 328         return sigaction(SIGINT, &sa, NULL);
 329 }
 330
 331 static int scrub_handle_sigint_child(int fd)
 332 {
 333         struct sigaction sa = {
 334                 .sa_handler = fd == -1 ? SIG_DFL : scrub_sigint_record_progress,
 335         };
 336
 337         cancel_fd = fd;
 338         return sigaction(SIGINT, &sa, NULL);
 339 }
 340
 341 static int scrub_datafile(const char *fn_base, const char *fn_local,
 342                                 const char *fn_tmp, char *datafile, int size)
 343 {
 344         int ret;
 345         int end = size - 2;
 346
 347         datafile[end + 1] = '\0';
 348         strncpy(datafile, fn_base, end);
 349         ret = strlen(datafile);
 350
 351         if (ret + 1 > end)
 352                 return -EOVERFLOW;
 353
 354         datafile[ret] = '.';
 355         strncpy(datafile + ret + 1, fn_local, end - ret - 1);
 356         ret = strlen(datafile);
 357
 358         if (ret + 1 > end)
 359                 return -EOVERFLOW;
 360
 361         if (fn_tmp) {
 362                 datafile[ret] = '_';
 363                 strncpy(datafile + ret + 1, fn_tmp, end - ret - 1);
 364                 ret = strlen(datafile);
 365
 366                 if (ret > end)
 367                         return -EOVERFLOW;
 368         }
 369
 370         return 0;
 371 }
 372
 373 static int scrub_open_file(const char *datafile, int m)
 374 {
 375         int fd;
 376         int ret;
 377
 378         fd = open(datafile, m, 0600);
 379         if (fd < 0)
 380                 return -errno;
 381
 382         ret = flock(fd, LOCK_EX|LOCK_NB);
 383         if (ret) {
 384                 ret = errno;
 385                 close(fd);
 386                 return -ret;
 387         }
 388
 389         return fd;
 390 }
 391
 392 static int scrub_open_file_r(const char *fn_base, const char *fn_local)
 393 {
 394         int ret;
 395         char datafile[PATH_MAX];
 396         ret = scrub_datafile(fn_base, fn_local, NULL,
 397                                 datafile, sizeof(datafile));
 398         if (ret < 0)
 399                 return ret;
 400         return scrub_open_file(datafile, O_RDONLY);
 401 }
 402
 403 static int scrub_open_file_w(const char *fn_base, const char *fn_local,
 404                                 const char *tmp)
 405 {
 406         int ret;
 407         char datafile[PATH_MAX];
 408         ret = scrub_datafile(fn_base, fn_local, tmp,
 409                                 datafile, sizeof(datafile));
 410         if (ret < 0)
 411                 return ret;
 412         return scrub_open_file(datafile, O_WRONLY|O_CREAT);
 413 }
 414
 415 static int scrub_rename_file(const char *fn_base, const char *fn_local,
 416                                 const char *tmp)
 417 {
 418         int ret;
 419         char datafile_old[PATH_MAX];
 420         char datafile_new[PATH_MAX];
 421         ret = scrub_datafile(fn_base, fn_local, tmp,
 422                                 datafile_old, sizeof(datafile_old));
 423         if (ret < 0)
 424                 return ret;
 425         ret = scrub_datafile(fn_base, fn_local, NULL,
 426                                 datafile_new, sizeof(datafile_new));
 427         if (ret < 0)
 428                 return ret;
 429         ret = rename(datafile_old, datafile_new);
 430         return ret ? -errno : 0;
 431 }
 432
 433 #define _SCRUB_KVREAD(ret, i, name, avail, l, dest) if (ret == 0) {       \
 434         ret = scrub_kvread(i, sizeof(#name), avail, l, #name, dest.name); \
 435 }
 436
 437 /*
 438  * returns 0 if the key did not match (nothing was read)
 439  *         1 if the key did match (success)
 440  *        -1 if the key did match and an error occured
 441  */
 442 static int scrub_kvread(int *i, int len, int avail, const char *buf,
 443                         const char *key, u64 *dest)
 444 {
 445         int j;
 446
 447         if (*i + len + 1 < avail && strncmp(&buf[*i], key, len - 1) == 0) {
 448                 *i += len - 1;
 449                 if (buf[*i] != ':')
 450                         return -1;
 451                 *i += 1;
 452                 for (j = 0; isdigit(buf[*i + j]) && *i + j < avail; ++j)
 453                         ;
 454                 if (*i + j >= avail)
 455                         return -1;
 456                 *dest = atoll(&buf[*i]);
 457                 *i += j;
 458                 return 1;
 459         }
 460
 461         return 0;
 462 }
 463
 464 #define _SCRUB_INVALID do {                                             \
 465         if (report_errors)                                              \
 466                 fprintf(stderr, "WARNING: invalid data in line %d pos " \
 467                         "%d state %d (near \"%.*s\") at %s:%d\n",       \
 468                         lineno, i, state, 20 > avail ? avail : 20,      \
 469                         l + i,  __FILE__, __LINE__);                    \
 470         goto skip;                                                      \
 471 } while (0)
 472
 473 static struct scrub_file_record **scrub_read_file(int fd, int report_errors)
 474 {
 475         int avail = 0;
 476         int old_avail = 0;
 477         char l[16 * 1024];
 478         int state = 0;
 479         int curr = -1;
 480         int i = 0;
 481         int j;
 482         int ret;
 483         int eof = 0;
 484         int lineno = 0;
 485         u64 version;
 486         char empty_uuid[BTRFS_FSID_SIZE] = {0};
 487         struct scrub_file_record **p = NULL;
 488
 489 again:
 490         old_avail = avail - i;
 491         BUG_ON(old_avail < 0);
 492         if (old_avail)
 493                 memmove(l, l + i, old_avail);
 494         avail = read(fd, l + old_avail, sizeof(l) - old_avail);
 495         if (avail == 0)
 496                 eof = 1;
 497         if (avail == 0 && old_avail == 0) {
 498                 if (curr >= 0 &&
 499                     memcmp(p[curr]->fsid, empty_uuid, BTRFS_FSID_SIZE) == 0) {
 500                         p[curr] = NULL;
 501                 } else if (curr == -1) {
 502                         p = ERR_PTR(-ENODATA);
 503                 }
 504                 return p;
 505         }
 506         if (avail == -1) {
 507                 free_history(p);
 508                 return ERR_PTR(-errno);
 509         }
 510         avail += old_avail;
 511
 512         i = 0;
 513         while (i < avail) {
 514                 void *tmp;
 515
 516                 switch (state) {
 517                 case 0: /* start of file */
 518                         ret = scrub_kvread(&i,
 519                                 sizeof(SCRUB_FILE_VERSION_PREFIX), avail, l,
 520                                 SCRUB_FILE_VERSION_PREFIX, &version);
 521                         if (ret != 1)
 522                                 _SCRUB_INVALID;
 523                         if (version != atoll(SCRUB_FILE_VERSION))
 524                                 return ERR_PTR(-ENOTSUP);
 525                         state = 6;
 526                         continue;
 527                 case 1: /* start of line, alloc */
 528                         /*
 529                          * this state makes sure we have a complete line in
 530                          * further processing, so we don't need wrap-tracking
 531                          * everywhere.
 532                          */
 533                         if (!eof && !memchr(l + i, '\n', avail - i))
 534                                 goto again;
 535                         ++lineno;
 536                         if (curr > -1 && memcmp(p[curr]->fsid, empty_uuid,
 537                                                 BTRFS_FSID_SIZE) == 0) {
 538                                 state = 2;
 539                                 continue;
 540                         }
 541                         ++curr;
 542                         tmp = p;
 543                         p = realloc(p, (curr + 2) * sizeof(*p));
 544                         if (!p) {
 545                                 free_history(tmp);
 546                                 return ERR_PTR(-errno);
 547                         }
 548                         p[curr] = malloc(sizeof(**p));
 549                         if (!p[curr]) {
 550                                 free_history(p);
 551                                 return ERR_PTR(-errno);
 552                         }
 553                         memset(p[curr], 0, sizeof(**p));
 554                         p[curr + 1] = NULL;
 555                         ++state;
 556                         /* fall through */
 557                 case 2: /* start of line, skip space */
 558                         while (isspace(l[i]) && i < avail) {
 559                                 if (l[i] == '\n')
 560                                         ++lineno;
 561                                 ++i;
 562                         }
 563                         if (i >= avail ||
 564                             (!eof && !memchr(l + i, '\n', avail - i)))
 565                                 goto again;
 566                         ++state;
 567                         /* fall through */
 568                 case 3: /* read fsid */
 569                         if (i == avail)
 570                                 continue;
 571                         for (j = 0; l[i + j] != ':' && i + j < avail; ++j)
 572                                 ;
 573                         if (i + j + 1 >= avail)
 574                                 _SCRUB_INVALID;
 575                         if (j != BTRFS_UUID_UNPARSED_SIZE - 1)
 576                                 _SCRUB_INVALID;
 577                         l[i + j] = '\0';
 578                         ret = uuid_parse(l + i, p[curr]->fsid);
 579                         if (ret)
 580                                 _SCRUB_INVALID;
 581                         i += j + 1;
 582                         ++state;
 583                         /* fall through */
 584                 case 4: /* read dev id */
 585                         for (j = 0; isdigit(l[i + j]) && i+j < avail; ++j)
 586                                 ;
 587                         if (j == 0 || i + j + 1 >= avail)
 588                                 _SCRUB_INVALID;
 589                         p[curr]->devid = atoll(&l[i]);
 590                         i += j + 1;
 591                         ++state;
 592                         /* fall through */
 593                 case 5: /* read key/value pair */
 594                         ret = 0;
 595                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 596                                         &p[curr]->p);
 597                         _SCRUB_KVREAD(ret, &i, data_extents_scrubbed, avail, l,
 598                                         &p[curr]->p);
 599                         _SCRUB_KVREAD(ret, &i, tree_extents_scrubbed, avail, l,
 600                                         &p[curr]->p);
 601                         _SCRUB_KVREAD(ret, &i, data_bytes_scrubbed, avail, l,
 602                                         &p[curr]->p);
 603                         _SCRUB_KVREAD(ret, &i, tree_bytes_scrubbed, avail, l,
 604                                         &p[curr]->p);
 605                         _SCRUB_KVREAD(ret, &i, read_errors, avail, l,
 606                                         &p[curr]->p);
 607                         _SCRUB_KVREAD(ret, &i, csum_errors, avail, l,
 608                                         &p[curr]->p);
 609                         _SCRUB_KVREAD(ret, &i, verify_errors, avail, l,
 610                                         &p[curr]->p);
 611                         _SCRUB_KVREAD(ret, &i, no_csum, avail, l,
 612                                         &p[curr]->p);
 613                         _SCRUB_KVREAD(ret, &i, csum_discards, avail, l,
 614                                         &p[curr]->p);
 615                         _SCRUB_KVREAD(ret, &i, super_errors, avail, l,
 616                                         &p[curr]->p);
 617                         _SCRUB_KVREAD(ret, &i, malloc_errors, avail, l,
 618                                         &p[curr]->p);
 619                         _SCRUB_KVREAD(ret, &i, uncorrectable_errors, avail, l,
 620                                         &p[curr]->p);
 621                         _SCRUB_KVREAD(ret, &i, corrected_errors, avail, l,
 622                                         &p[curr]->p);
 623                         _SCRUB_KVREAD(ret, &i, last_physical, avail, l,
 624                                         &p[curr]->p);
 625                         _SCRUB_KVREAD(ret, &i, finished, avail, l,
 626                                         &p[curr]->stats);
 627                         _SCRUB_KVREAD(ret, &i, t_start, avail, l,
 628                                         (u64 *)&p[curr]->stats);
 629                         _SCRUB_KVREAD(ret, &i, t_resumed, avail, l,
 630                                         (u64 *)&p[curr]->stats);
 631                         _SCRUB_KVREAD(ret, &i, duration, avail, l,
 632                                         (u64 *)&p[curr]->stats);
 633                         _SCRUB_KVREAD(ret, &i, canceled, avail, l,
 634                                         &p[curr]->stats);
 635                         if (ret != 1)
 636                                 _SCRUB_INVALID;
 637                         ++state;
 638                         /* fall through */
 639                 case 6: /* after number */
 640                         if (l[i] == '|')
 641                                 state = 5;
 642                         else if (l[i] == '\n')
 643                                 state = 1;
 644                         else
 645                                 _SCRUB_INVALID;
 646                         ++i;
 647                         continue;
 648                 case 99: /* skip rest of line */
 649 skip:
 650                         state = 99;
 651                         do {
 652                                 ++i;
 653                                 if (l[i - 1] == '\n') {
 654                                         state = 1;
 655                                         break;
 656                                 }
 657                         } while (i < avail);
 658                         continue;
 659                 }
 660                 BUG();
 661         }
 662         goto again;
 663 }
 664
 665 static int scrub_write_buf(int fd, const void *data, int len)
 666 {
 667         int ret;
 668         ret = write(fd, data, len);
 669         return ret - len;
 670 }
 671
 672 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 673                                 __attribute__ ((format (printf, 4, 5)));
 674 static int scrub_writev(int fd, char *buf, int max, const char *fmt, ...)
 675 {
 676         int ret;
 677         va_list args;
 678
 679         va_start(args, fmt);
 680         ret = vsnprintf(buf, max, fmt, args);
 681         va_end(args);
 682         if (ret >= max)
 683                 return ret - max;
 684         return scrub_write_buf(fd, buf, ret);
 685 }
 686
 687 #define _SCRUB_SUM(dest, data, name) dest->scrub_args.progress.name =   \
 688                         data->resumed->p.name + data->scrub_args.progress.name
 689
 690 static struct scrub_progress *scrub_resumed_stats(struct scrub_progress *data,
 691                                                   struct scrub_progress *dest)
 692 {
 693         if (!data->resumed || data->skip)
 694                 return data;
 695
 696         _SCRUB_SUM(dest, data, data_extents_scrubbed);
 697         _SCRUB_SUM(dest, data, tree_extents_scrubbed);
 698         _SCRUB_SUM(dest, data, data_bytes_scrubbed);
 699         _SCRUB_SUM(dest, data, tree_bytes_scrubbed);
 700         _SCRUB_SUM(dest, data, read_errors);
 701         _SCRUB_SUM(dest, data, csum_errors);
 702         _SCRUB_SUM(dest, data, verify_errors);
 703         _SCRUB_SUM(dest, data, no_csum);
 704         _SCRUB_SUM(dest, data, csum_discards);
 705         _SCRUB_SUM(dest, data, super_errors);
 706         _SCRUB_SUM(dest, data, malloc_errors);
 707         _SCRUB_SUM(dest, data, uncorrectable_errors);
 708         _SCRUB_SUM(dest, data, corrected_errors);
 709         _SCRUB_SUM(dest, data, last_physical);
 710         dest->stats.canceled = data->stats.canceled;
 711         dest->stats.finished = data->stats.finished;
 712         dest->stats.t_resumed = data->stats.t_start;
 713         dest->stats.t_start = data->resumed->stats.t_start;
 714         dest->stats.duration = data->resumed->stats.duration +
 715                                                         data->stats.duration;
 716         dest->scrub_args.devid = data->scrub_args.devid;
 717         return dest;
 718 }
 719
 720 #define _SCRUB_KVWRITE(fd, buf, name, use)              \
 721         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 722                         use->scrub_args.progress.name)
 723
 724 #define _SCRUB_KVWRITE_STATS(fd, buf, name, use)        \
 725         scrub_kvwrite(fd, buf, sizeof(buf), #name,      \
 726                         use->stats.name)
 727
 728 static int scrub_kvwrite(int fd, char *buf, int max, const char *key, u64 val)
 729 {
 730         return scrub_writev(fd, buf, max, "|%s:%lld", key, val);
 731 }
 732
 733 static int scrub_write_file(int fd, const char *fsid,
 734                                 struct scrub_progress *data, int n)
 735 {
 736         int ret = 0;
 737         int i;
 738         char buf[1024];
 739         struct scrub_progress local;
 740         struct scrub_progress *use;
 741
 742         if (n < 1)
 743                 return -EINVAL;
 744
 745         /* each -1 is to subtract one \0 byte, the + 2 is for ':' and '\n' */
 746         ret = scrub_write_buf(fd, SCRUB_FILE_VERSION_PREFIX ":"
 747                                 SCRUB_FILE_VERSION "\n",
 748                                 (sizeof(SCRUB_FILE_VERSION_PREFIX) - 1) +
 749                                 (sizeof(SCRUB_FILE_VERSION) - 1) + 2);
 750         if (ret)
 751                 return -EOVERFLOW;
 752
 753         for (i = 0; i < n; ++i) {
 754                 use = scrub_resumed_stats(&data[i], &local);
 755                 if (scrub_write_buf(fd, fsid, strlen(fsid)) ||
 756                     scrub_write_buf(fd, ":", 1) ||
 757                     scrub_writev(fd, buf, sizeof(buf), "%lld",
 758                                         use->scrub_args.devid) ||
 759                     scrub_write_buf(fd, buf, ret) ||
 760                     _SCRUB_KVWRITE(fd, buf, data_extents_scrubbed, use) ||
 761                     _SCRUB_KVWRITE(fd, buf, tree_extents_scrubbed, use) ||
 762                     _SCRUB_KVWRITE(fd, buf, data_bytes_scrubbed, use) ||
 763                     _SCRUB_KVWRITE(fd, buf, tree_bytes_scrubbed, use) ||
 764                     _SCRUB_KVWRITE(fd, buf, read_errors, use) ||
 765                     _SCRUB_KVWRITE(fd, buf, csum_errors, use) ||
 766                     _SCRUB_KVWRITE(fd, buf, verify_errors, use) ||
 767                     _SCRUB_KVWRITE(fd, buf, no_csum, use) ||
 768                     _SCRUB_KVWRITE(fd, buf, csum_discards, use) ||
 769                     _SCRUB_KVWRITE(fd, buf, super_errors, use) ||
 770                     _SCRUB_KVWRITE(fd, buf, malloc_errors, use) ||
 771                     _SCRUB_KVWRITE(fd, buf, uncorrectable_errors, use) ||
 772                     _SCRUB_KVWRITE(fd, buf, corrected_errors, use) ||
 773                     _SCRUB_KVWRITE(fd, buf, last_physical, use) ||
 774                     _SCRUB_KVWRITE_STATS(fd, buf, t_start, use) ||
 775                     _SCRUB_KVWRITE_STATS(fd, buf, t_resumed, use) ||
 776                     _SCRUB_KVWRITE_STATS(fd, buf, duration, use) ||
 777                     _SCRUB_KVWRITE_STATS(fd, buf, canceled, use) ||
 778                     _SCRUB_KVWRITE_STATS(fd, buf, finished, use) ||
 779                     scrub_write_buf(fd, "\n", 1)) {
 780                         return -EOVERFLOW;
 781                 }
 782         }
 783
 784         return 0;
 785 }
 786
 787 static int scrub_write_progress(pthread_mutex_t *m, const char *fsid,
 788                                 struct scrub_progress *data, int n)
 789 {
 790         int ret;
 791         int err;
 792         int fd = -1;
 793         int old;
 794
 795         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &old);
 796         if (ret) {
 797                 err = -ret;
 798                 goto out3;
 799         }
 800
 801         ret = pthread_mutex_lock(m);
 802         if (ret) {
 803                 err = -ret;
 804                 goto out2;
 805         }
 806
 807         fd = scrub_open_file_w(SCRUB_DATA_FILE, fsid, "tmp");
 808         if (fd < 0) {
 809                 err = fd;
 810                 goto out1;
 811         }
 812         err = scrub_write_file(fd, fsid, data, n);
 813         if (err)
 814                 goto out1;
 815         err = scrub_rename_file(SCRUB_DATA_FILE, fsid, "tmp");
 816         if (err)
 817                 goto out1;
 818
 819 out1:
 820         if (fd >= 0) {
 821                 ret = close(fd);
 822                 if (ret)
 823                         err = -errno;
 824         }
 825
 826         ret = pthread_mutex_unlock(m);
 827         if (ret && !err)
 828                 err = -ret;
 829
 830 out2:
 831         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &old);
 832         if (ret && !err)
 833                 err = -ret;
 834
 835 out3:
 836         return err;
 837 }
 838
 839 static void *scrub_one_dev(void *ctx)
 840 {
 841         struct scrub_progress *sp = ctx;
 842         int ret;
 843         struct timeval tv;
 844
 845         sp->stats.canceled = 0;
 846         sp->stats.duration = 0;
 847         sp->stats.finished = 0;
 848
 849         ret = syscall(SYS_ioprio_set, IOPRIO_WHO_PROCESS, 0,
 850                       IOPRIO_PRIO_VALUE(sp->ioprio_class,
 851                                         sp->ioprio_classdata));
 852         if (ret)
 853                 fprintf(stderr,
 854                         "WARNING: setting ioprio failed: %s (ignored).\n",
 855                         strerror(errno));
 856
 857         ret = ioctl(sp->fd, BTRFS_IOC_SCRUB, &sp->scrub_args);
 858         gettimeofday(&tv, NULL);
 859         sp->ret = ret;
 860         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 861         sp->stats.canceled = !!ret;
 862         sp->ioctl_errno = errno;
 863         ret = pthread_mutex_lock(&sp->progress_mutex);
 864         if (ret)
 865                 return ERR_PTR(-ret);
 866         sp->stats.finished = 1;
 867         ret = pthread_mutex_unlock(&sp->progress_mutex);
 868         if (ret)
 869                 return ERR_PTR(-ret);
 870
 871         return NULL;
 872 }
 873
 874 static void *progress_one_dev(void *ctx)
 875 {
 876         struct scrub_progress *sp = ctx;
 877
 878         sp->ret = ioctl(sp->fd, BTRFS_IOC_SCRUB_PROGRESS, &sp->scrub_args);
 879         sp->ioctl_errno = errno;
 880
 881         return NULL;
 882 }
 883
 884 /* nb: returns a negative errno via ERR_PTR */
 885 static void *scrub_progress_cycle(void *ctx)
 886 {
 887         int ret = 0;
 888         int  perr = 0;  /* positive / pthread error returns */
 889         int old;
 890         int i;
 891         char fsid[BTRFS_UUID_UNPARSED_SIZE];
 892         struct scrub_progress *sp;
 893         struct scrub_progress *sp_last;
 894         struct scrub_progress *sp_shared;
 895         struct timeval tv;
 896         struct scrub_progress_cycle *spc = ctx;
 897         int ndev = spc->fi->num_devices;
 898         int this = 1;
 899         int last = 0;
 900         int peer_fd = -1;
 901         struct pollfd accept_poll_fd = {
 902                 .fd = spc->prg_fd,
 903                 .events = POLLIN,
 904                 .revents = 0,
 905         };
 906         struct pollfd write_poll_fd = {
 907                 .events = POLLOUT,
 908                 .revents = 0,
 909         };
 910         struct sockaddr_un peer;
 911         socklen_t peer_size = sizeof(peer);
 912
 913         perr = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &old);
 914         if (perr)
 915                 goto out;
 916
 917         uuid_unparse(spc->fi->fsid, fsid);
 918
 919         for (i = 0; i < ndev; ++i) {
 920                 sp = &spc->progress[i];
 921                 sp_last = &spc->progress[i + ndev];
 922                 sp_shared = &spc->shared_progress[i];
 923                 sp->scrub_args.devid = sp_last->scrub_args.devid =
 924                                                 sp_shared->scrub_args.devid;
 925                 sp->fd = sp_last->fd = spc->fdmnt;
 926                 sp->stats.t_start = sp_last->stats.t_start =
 927                                                 sp_shared->stats.t_start;
 928                 sp->resumed = sp_last->resumed = sp_shared->resumed;
 929                 sp->skip = sp_last->skip = sp_shared->skip;
 930                 sp->stats.finished = sp_last->stats.finished =
 931                                                 sp_shared->stats.finished;
 932         }
 933
 934         while (1) {
 935                 ret = poll(&accept_poll_fd, 1, 5 * 1000);
 936                 if (ret == -1) {
 937                         ret = -errno;
 938                         goto out;
 939                 }
 940                 if (ret)
 941                         peer_fd = accept(spc->prg_fd, (struct sockaddr *)&peer,
 942                                          &peer_size);
 943                 gettimeofday(&tv, NULL);
 944                 this = (this + 1)%2;
 945                 last = (last + 1)%2;
 946                 for (i = 0; i < ndev; ++i) {
 947                         sp = &spc->progress[this * ndev + i];
 948                         sp_last = &spc->progress[last * ndev + i];
 949                         sp_shared = &spc->shared_progress[i];
 950                         if (sp->stats.finished)
 951                                 continue;
 952                         progress_one_dev(sp);
 953                         sp->stats.duration = tv.tv_sec - sp->stats.t_start;
 954                         if (!sp->ret)
 955                                 continue;
 956                         if (sp->ioctl_errno != ENOTCONN &&
 957                             sp->ioctl_errno != ENODEV) {
 958                                 ret = -sp->ioctl_errno;
 959                                 goto out;
 960                         }
 961                         /*
 962                          * scrub finished or device removed, check the
 963                          * finished flag. if unset, just use the last
 964                          * result we got for the current write and go
 965                          * on. flag should be set on next cycle, then.
 966                          */
 967                         perr = pthread_setcancelstate(
 968                                         PTHREAD_CANCEL_DISABLE, &old);
 969                         if (perr)
 970                                 goto out;
 971                         perr = pthread_mutex_lock(&sp_shared->progress_mutex);
 972                         if (perr)
 973                                 goto out;
 974                         if (!sp_shared->stats.finished) {
 975                                 perr = pthread_mutex_unlock(
 976                                                 &sp_shared->progress_mutex);
 977                                 if (perr)
 978                                         goto out;
 979                                 perr = pthread_setcancelstate(
 980                                                 PTHREAD_CANCEL_ENABLE, &old);
 981                                 if (perr)
 982                                         goto out;
 983                                 memcpy(sp, sp_last, sizeof(*sp));
 984                                 continue;
 985                         }
 986                         perr = pthread_mutex_unlock(&sp_shared->progress_mutex);
 987                         if (perr)
 988                                 goto out;
 989                         perr = pthread_setcancelstate(
 990                                         PTHREAD_CANCEL_ENABLE, &old);
 991                         if (perr)
 992                                 goto out;
 993                         memcpy(sp, sp_shared, sizeof(*sp));
 994                         memcpy(sp_last, sp_shared, sizeof(*sp));
 995                 }
 996                 if (peer_fd != -1) {
 997                         write_poll_fd.fd = peer_fd;
 998                         ret = poll(&write_poll_fd, 1, 0);
 999                         if (ret == -1) {
1000                                 ret = -errno;
1001                                 goto out;
1002                         }
1003                         if (ret) {
1004                                 ret = scrub_write_file(
1005                                         peer_fd, fsid,
1006                                         &spc->progress[this * ndev], ndev);
1007                                 if (ret)
1008                                         goto out;
1009                         }
1010                         close(peer_fd);
1011                         peer_fd = -1;
1012                 }
1013                 if (!spc->do_record)
1014                         continue;
1015                 ret = scrub_write_progress(spc->write_mutex, fsid,
1016                                            &spc->progress[this * ndev], ndev);
1017                 if (ret)
1018                         goto out;
1019         }
1020 out:
1021         if (peer_fd != -1)
1022                 close(peer_fd);
1023         if (perr)
1024                 ret = -perr;
1025         return ERR_PTR(ret);
1026 }
1027
1028 static struct scrub_file_record *last_dev_scrub(
1029                 struct scrub_file_record *const *const past_scrubs, u64 devid)
1030 {
1031         int i;
1032
1033         if (!past_scrubs || IS_ERR(past_scrubs))
1034                 return NULL;
1035
1036         for (i = 0; past_scrubs[i]; ++i)
1037                 if (past_scrubs[i]->devid == devid)
1038                         return past_scrubs[i];
1039
1040         return NULL;
1041 }
1042
1043 static int mkdir_p(char *path)
1044 {
1045         int i;
1046         int ret;
1047
1048         for (i = 1; i < strlen(path); ++i) {
1049                 if (path[i] != '/')
1050                         continue;
1051                 path[i] = '\0';
1052                 ret = mkdir(path, 0777);
1053                 if (ret && errno != EEXIST)
1054                         return -errno;
1055                 path[i] = '/';
1056         }
1057
1058         return 0;
1059 }
1060
1061 static int is_scrub_running_on_fs(struct btrfs_ioctl_fs_info_args *fi_args,
1062                                   struct btrfs_ioctl_dev_info_args *di_args,
1063                                   struct scrub_file_record **past_scrubs)
1064 {
1065         int i;
1066
1067         if (!fi_args || !di_args || !past_scrubs)
1068                 return 0;
1069
1070         for (i = 0; i < fi_args->num_devices; i++) {
1071                 struct scrub_file_record *sfr =
1072                         last_dev_scrub(past_scrubs, di_args[i].devid);
1073
1074                 if (!sfr)
1075                         continue;
1076                 if (!(sfr->stats.finished || sfr->stats.canceled))
1077                         return 1;
1078         }
1079         return 0;
1080 }
1081
1082 static int is_scrub_running_in_kernel(int fd,
1083                 struct btrfs_ioctl_dev_info_args *di_args, u64 max_devices)
1084 {
1085         struct scrub_progress sp;
1086         int i;
1087         int ret;
1088
1089         for (i = 0; i < max_devices; i++) {
1090                 memset(&sp, 0, sizeof(sp));
1091                 sp.scrub_args.devid = di_args[i].devid;
1092                 ret = ioctl(fd, BTRFS_IOC_SCRUB_PROGRESS, &sp.scrub_args);
1093                 if (!ret)
1094                         return 1;
1095         }
1096
1097         return 0;
1098 }
1099
1100 static const char * const cmd_scrub_start_usage[];
1101 static const char * const cmd_scrub_resume_usage[];
1102
1103 static int scrub_start(int argc, char **argv, int resume)
1104 {
1105         int fdmnt;
1106         int prg_fd = -1;
1107         int fdres = -1;
1108         int ret;
1109         pid_t pid;
1110         int c;
1111         int i;
1112         int err = 0;
1113         int e_uncorrectable = 0;
1114         int e_correctable = 0;
1115         int print_raw = 0;
1116         char *path;
1117         int do_background = 1;
1118         int do_wait = 0;
1119         int do_print = 0;
1120         int do_quiet = 0;
1121         int do_record = 1;
1122         int readonly = 0;
1123         int do_stats_per_dev = 0;
1124         int ioprio_class = IOPRIO_CLASS_IDLE;
1125         int ioprio_classdata = 0;
1126         int n_start = 0;
1127         int n_skip = 0;
1128         int n_resume = 0;
1129         struct btrfs_ioctl_fs_info_args fi_args;
1130         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1131         struct scrub_progress *sp = NULL;
1132         struct scrub_fs_stat fs_stat;
1133         struct timeval tv;
1134         struct sockaddr_un addr = {
1135                 .sun_family = AF_UNIX,
1136         };
1137         pthread_t *t_devs = NULL;
1138         pthread_t t_prog;
1139         struct scrub_file_record **past_scrubs = NULL;
1140         struct scrub_file_record *last_scrub = NULL;
1141         char *datafile = strdup(SCRUB_DATA_FILE);
1142         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1143         char sock_path[PATH_MAX] = "";
1144         struct scrub_progress_cycle spc;
1145         pthread_mutex_t spc_write_mutex = PTHREAD_MUTEX_INITIALIZER;
1146         void *terr;
1147         u64 devid;
1148         DIR *dirstream = NULL;
1149         int force = 0;
1150         int nothing_to_resume = 0;
1151
1152         optind = 1;
1153         while ((c = getopt(argc, argv, "BdqrRc:n:f")) != -1) {
1154                 switch (c) {
1155                 case 'B':
1156                         do_background = 0;
1157                         do_wait = 1;
1158                         do_print = 1;
1159                         break;
1160                 case 'd':
1161                         do_stats_per_dev = 1;
1162                         break;
1163                 case 'q':
1164                         do_quiet = 1;
1165                         break;
1166                 case 'r':
1167                         readonly = 1;
1168                         break;
1169                 case 'R':
1170                         print_raw = 1;
1171                         break;
1172                 case 'c':
1173                         ioprio_class = (int)strtol(optarg, NULL, 10);
1174                         break;
1175                 case 'n':
1176                         ioprio_classdata = (int)strtol(optarg, NULL, 10);
1177                         break;
1178                 case 'f':
1179                         force = 1;
1180                         break;
1181                 case '?':
1182                 default:
1183                         usage(resume ? cmd_scrub_resume_usage :
1184                                                 cmd_scrub_start_usage);
1185                 }
1186         }
1187
1188         /* try to catch most error cases before forking */
1189
1190         if (check_argc_exact(argc - optind, 1)) {
1191                 usage(resume ? cmd_scrub_resume_usage :
1192                                         cmd_scrub_start_usage);
1193         }
1194
1195         spc.progress = NULL;
1196         if (do_quiet && do_print)
1197                 do_print = 0;
1198
1199         if (mkdir_p(datafile)) {
1200                 ERR(!do_quiet, "WARNING: cannot create scrub data "
1201                                "file, mkdir %s failed: %s. Status recording "
1202                                "disabled\n", datafile, strerror(errno));
1203                 do_record = 0;
1204         }
1205         free(datafile);
1206
1207         path = argv[optind];
1208
1209         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1210
1211         if (fdmnt < 0) {
1212                 if (errno == EINVAL)
1213                         ERR(!do_quiet,
1214                             "ERROR: '%s' is not a mounted btrfs device\n",
1215                             path);
1216                 else
1217                         ERR(!do_quiet, "ERROR: can't access '%s': %s\n",
1218                             path, strerror(errno));
1219                 return 1;
1220         }
1221
1222         ret = get_fs_info(path, &fi_args, &di_args);
1223         if (ret) {
1224                 ERR(!do_quiet, "ERROR: getting dev info for scrub failed: "
1225                     "%s\n", strerror(-ret));
1226                 err = 1;
1227                 goto out;
1228         }
1229         if (!fi_args.num_devices) {
1230                 ERR(!do_quiet, "ERROR: no devices found\n");
1231                 err = 1;
1232                 goto out;
1233         }
1234
1235         uuid_unparse(fi_args.fsid, fsid);
1236         fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1237         if (fdres < 0 && fdres != -ENOENT) {
1238                 ERR(!do_quiet, "WARNING: failed to open status file: "
1239                     "%s\n", strerror(-fdres));
1240         } else if (fdres >= 0) {
1241                 past_scrubs = scrub_read_file(fdres, !do_quiet);
1242                 if (IS_ERR(past_scrubs))
1243                         ERR(!do_quiet, "WARNING: failed to read status file: "
1244                             "%s\n", strerror(-PTR_ERR(past_scrubs)));
1245                 close(fdres);
1246         }
1247
1248         /*
1249          * Check for stale information in the status file, ie. if it's
1250          * canceled=0, finished=0 but no scrub is running.
1251          */
1252         if (!is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices))
1253                 force = 1;
1254
1255         /*
1256          * check whether any involved device is already busy running a
1257          * scrub. This would cause damaged status messages and the state
1258          * "aborted" without the explanation that a scrub was already
1259          * running. Therefore check it first, prevent it and give some
1260          * feedback to the user if scrub is already running.
1261          * Note that if scrub is started with a block device as the
1262          * parameter, only that particular block device is checked. It
1263          * is a normal mode of operation to start scrub on multiple
1264          * single devices, there is no reason to prevent this.
1265          */
1266         if (!force && is_scrub_running_on_fs(&fi_args, di_args, past_scrubs)) {
1267                 ERR(!do_quiet,
1268                     "ERROR: scrub is already running.\n"
1269                     "To cancel use 'btrfs scrub cancel %s'.\n"
1270                     "To see the status use 'btrfs scrub status [-d] %s'.\n",
1271                     path, path);
1272                 err = 1;
1273                 goto out;
1274         }
1275
1276         t_devs = malloc(fi_args.num_devices * sizeof(*t_devs));
1277         sp = calloc(fi_args.num_devices, sizeof(*sp));
1278         spc.progress = calloc(fi_args.num_devices * 2, sizeof(*spc.progress));
1279
1280         if (!t_devs || !sp || !spc.progress) {
1281                 ERR(!do_quiet, "ERROR: scrub failed: %s", strerror(errno));
1282                 err = 1;
1283                 goto out;
1284         }
1285
1286         for (i = 0; i < fi_args.num_devices; ++i) {
1287                 devid = di_args[i].devid;
1288                 ret = pthread_mutex_init(&sp[i].progress_mutex, NULL);
1289                 if (ret) {
1290                         ERR(!do_quiet, "ERROR: pthread_mutex_init failed: "
1291                             "%s\n", strerror(ret));
1292                         err = 1;
1293                         goto out;
1294                 }
1295                 last_scrub = last_dev_scrub(past_scrubs, devid);
1296                 sp[i].scrub_args.devid = devid;
1297                 sp[i].fd = fdmnt;
1298                 if (resume && last_scrub && (last_scrub->stats.canceled ||
1299                                              !last_scrub->stats.finished)) {
1300                         ++n_resume;
1301                         sp[i].scrub_args.start = last_scrub->p.last_physical;
1302                         sp[i].resumed = last_scrub;
1303                 } else if (resume) {
1304                         ++n_skip;
1305                         sp[i].skip = 1;
1306                         sp[i].resumed = last_scrub;
1307                         continue;
1308                 } else {
1309                         ++n_start;
1310                         sp[i].scrub_args.start = 0ll;
1311                         sp[i].resumed = NULL;
1312                 }
1313                 sp[i].skip = 0;
1314                 sp[i].scrub_args.end = (u64)-1ll;
1315                 sp[i].scrub_args.flags = readonly ? BTRFS_SCRUB_READONLY : 0;
1316                 sp[i].ioprio_class = ioprio_class;
1317                 sp[i].ioprio_classdata = ioprio_classdata;
1318         }
1319
1320         if (!n_start && !n_resume) {
1321                 if (!do_quiet)
1322                         printf("scrub: nothing to resume for %s, fsid %s\n",
1323                                path, fsid);
1324                 nothing_to_resume = 1;
1325                 goto out;
1326         }
1327
1328         ret = prg_fd = socket(AF_UNIX, SOCK_STREAM, 0);
1329         while (ret != -1) {
1330                 ret = scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid, NULL,
1331                                         sock_path, sizeof(sock_path));
1332                 /* ignore EOVERFLOW, try using a shorter path for the socket */
1333                 addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1334                 strncpy(addr.sun_path, sock_path, sizeof(addr.sun_path) - 1);
1335                 ret = bind(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1336                 if (ret != -1 || errno != EADDRINUSE)
1337                         break;
1338                 /*
1339                  * bind failed with EADDRINUSE. so let's see if anyone answers
1340                  * when we make a call to the socket ...
1341                  */
1342                 ret = connect(prg_fd, (struct sockaddr *)&addr, sizeof(addr));
1343                 if (!ret || errno != ECONNREFUSED) {
1344                         /* ... yes, so scrub must be running. error out */
1345                         fprintf(stderr, "ERROR: scrub already running\n");
1346                         close(prg_fd);
1347                         prg_fd = -1;
1348                         goto out;
1349                 }
1350                 /*
1351                  * ... no, this means someone left us alone with an unused
1352                  * socket in the file system. remove it and try again.
1353                  */
1354                 ret = unlink(sock_path);
1355         }
1356         if (ret != -1)
1357                 ret = listen(prg_fd, 100);
1358         if (ret == -1) {
1359                 ERR(!do_quiet, "WARNING: failed to open the progress status "
1360                     "socket at %s: %s. Progress cannot be queried\n",
1361                     sock_path[0] ? sock_path : SCRUB_PROGRESS_SOCKET_PATH,
1362                     strerror(errno));
1363                 if (prg_fd != -1) {
1364                         close(prg_fd);
1365                         prg_fd = -1;
1366                         if (sock_path[0])
1367                                 unlink(sock_path);
1368                 }
1369         }
1370
1371         if (do_record) {
1372                 /* write all-zero progress file for a start */
1373                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1374                                            fi_args.num_devices);
1375                 if (ret) {
1376                         ERR(!do_quiet, "WARNING: failed to write the progress "
1377                             "status file: %s. Status recording disabled\n",
1378                             strerror(-ret));
1379                         do_record = 0;
1380                 }
1381         }
1382
1383         if (do_background) {
1384                 pid = fork();
1385                 if (pid == -1) {
1386                         ERR(!do_quiet, "ERROR: cannot scrub, fork failed: "
1387                                         "%s\n", strerror(errno));
1388                         err = 1;
1389                         goto out;
1390                 }
1391
1392                 if (pid) {
1393                         int stat;
1394                         scrub_handle_sigint_parent();
1395                         if (!do_quiet)
1396                                 printf("scrub %s on %s, fsid %s (pid=%d)\n",
1397                                        n_start ? "started" : "resumed",
1398                                        path, fsid, pid);
1399                         if (!do_wait) {
1400                                 err = 0;
1401                                 goto out;
1402                         }
1403                         ret = wait(&stat);
1404                         if (ret != pid) {
1405                                 ERR(!do_quiet, "ERROR: wait failed: (ret=%d) "
1406                                     "%s\n", ret, strerror(errno));
1407                                 err = 1;
1408                                 goto out;
1409                         }
1410                         if (!WIFEXITED(stat) || WEXITSTATUS(stat)) {
1411                                 ERR(!do_quiet, "ERROR: scrub process failed\n");
1412                                 err = WIFEXITED(stat) ? WEXITSTATUS(stat) : -1;
1413                                 goto out;
1414                         }
1415                         err = 0;
1416                         goto out;
1417                 }
1418         }
1419
1420         scrub_handle_sigint_child(fdmnt);
1421
1422         for (i = 0; i < fi_args.num_devices; ++i) {
1423                 if (sp[i].skip) {
1424                         sp[i].scrub_args.progress = sp[i].resumed->p;
1425                         sp[i].stats = sp[i].resumed->stats;
1426                         sp[i].ret = 0;
1427                         sp[i].stats.finished = 1;
1428                         continue;
1429                 }
1430                 devid = di_args[i].devid;
1431                 gettimeofday(&tv, NULL);
1432                 sp[i].stats.t_start = tv.tv_sec;
1433                 ret = pthread_create(&t_devs[i], NULL,
1434                                         scrub_one_dev, &sp[i]);
1435                 if (ret) {
1436                         if (do_print)
1437                                 fprintf(stderr, "ERROR: creating "
1438                                         "scrub_one_dev[%llu] thread failed: "
1439                                         "%s\n", devid, strerror(ret));
1440                         err = 1;
1441                         goto out;
1442                 }
1443         }
1444
1445         spc.fdmnt = fdmnt;
1446         spc.prg_fd = prg_fd;
1447         spc.do_record = do_record;
1448         spc.write_mutex = &spc_write_mutex;
1449         spc.shared_progress = sp;
1450         spc.fi = &fi_args;
1451         ret = pthread_create(&t_prog, NULL, scrub_progress_cycle, &spc);
1452         if (ret) {
1453                 if (do_print)
1454                         fprintf(stderr, "ERROR: creating progress thread "
1455                                 "failed: %s\n", strerror(ret));
1456                 err = 1;
1457                 goto out;
1458         }
1459
1460         err = 0;
1461         for (i = 0; i < fi_args.num_devices; ++i) {
1462                 if (sp[i].skip)
1463                         continue;
1464                 devid = di_args[i].devid;
1465                 ret = pthread_join(t_devs[i], NULL);
1466                 if (ret) {
1467                         if (do_print)
1468                                 fprintf(stderr, "ERROR: pthread_join failed "
1469                                         "for scrub_one_dev[%llu]: %s\n", devid,
1470                                         strerror(ret));
1471                         ++err;
1472                         continue;
1473                 }
1474                 if (sp[i].ret && sp[i].ioctl_errno == ENODEV) {
1475                         if (do_print)
1476                                 fprintf(stderr, "WARNING: device %lld not "
1477                                         "present\n", devid);
1478                         continue;
1479                 }
1480                 if (sp[i].ret && sp[i].ioctl_errno == ECANCELED) {
1481                         ++err;
1482                 } else if (sp[i].ret) {
1483                         if (do_print)
1484                                 fprintf(stderr, "ERROR: scrubbing %s failed "
1485                                         "for device id %lld (%s)\n", path,
1486                                         devid, strerror(sp[i].ioctl_errno));
1487                         ++err;
1488                         continue;
1489                 }
1490                 if (sp[i].scrub_args.progress.uncorrectable_errors > 0)
1491                         e_uncorrectable++;
1492                 if (sp[i].scrub_args.progress.corrected_errors > 0
1493                     || sp[i].scrub_args.progress.unverified_errors > 0)
1494                         e_correctable++;
1495         }
1496
1497         if (do_print) {
1498                 const char *append = "done";
1499                 if (!do_stats_per_dev)
1500                         init_fs_stat(&fs_stat);
1501                 for (i = 0; i < fi_args.num_devices; ++i) {
1502                         if (do_stats_per_dev) {
1503                                 print_scrub_dev(&di_args[i],
1504                                                 &sp[i].scrub_args.progress,
1505                                                 print_raw,
1506                                                 sp[i].ret ? "canceled" : "done",
1507                                                 &sp[i].stats);
1508                         } else {
1509                                 if (sp[i].ret)
1510                                         append = "canceled";
1511                                 add_to_fs_stat(&sp[i].scrub_args.progress,
1512                                                 &sp[i].stats, &fs_stat);
1513                         }
1514                 }
1515                 if (!do_stats_per_dev) {
1516                         printf("scrub %s for %s\n", append, fsid);
1517                         print_fs_stat(&fs_stat, print_raw);
1518                 }
1519         }
1520
1521         ret = pthread_cancel(t_prog);
1522         if (!ret)
1523                 ret = pthread_join(t_prog, &terr);
1524
1525         /* check for errors from the handling of the progress thread */
1526         if (do_print && ret) {
1527                 fprintf(stderr, "ERROR: progress thread handling failed: %s\n",
1528                         strerror(ret));
1529         }
1530
1531         /* check for errors returned from the progress thread itself */
1532         if (do_print && terr && terr != PTHREAD_CANCELED) {
1533                 fprintf(stderr, "ERROR: recording progress "
1534                         "failed: %s\n", strerror(-PTR_ERR(terr)));
1535         }
1536
1537         if (do_record) {
1538                 ret = scrub_write_progress(&spc_write_mutex, fsid, sp,
1539                                            fi_args.num_devices);
1540                 if (ret && do_print) {
1541                         fprintf(stderr, "ERROR: failed to record the result: "
1542                                 "%s\n", strerror(-ret));
1543                 }
1544         }
1545
1546         scrub_handle_sigint_child(-1);
1547
1548 out:
1549         free_history(past_scrubs);
1550         free(di_args);
1551         free(t_devs);
1552         free(sp);
1553         free(spc.progress);
1554         if (prg_fd > -1) {
1555                 close(prg_fd);
1556                 if (sock_path[0])
1557                         unlink(sock_path);
1558         }
1559         close_file_or_dir(fdmnt, dirstream);
1560
1561         if (err)
1562                 return 1;
1563         if (nothing_to_resume)
1564                 return 2;
1565         if (e_uncorrectable) {
1566                 ERR(!do_quiet, "ERROR: There are uncorrectable errors.\n");
1567                 return 3;
1568         }
1569         if (e_correctable)
1570                 ERR(!do_quiet, "WARNING: errors detected during scrubbing, corrected.\n");
1571
1572         return 0;
1573 }
1574
1575 static const char * const cmd_scrub_start_usage[] = {
1576         "btrfs scrub start [-BdqrRf] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1577         "Start a new scrub. If a scrub is already running, the new one fails.",
1578         "",
1579         "-B     do not background",
1580         "-d     stats per device (-B only)",
1581         "-q     be quiet",
1582         "-r     read only mode",
1583         "-R     raw print mode, print full data instead of summary",
1584         "-c     set ioprio class (see ionice(1) manpage)",
1585         "-n     set ioprio classdata (see ionice(1) manpage)",
1586         "-f     force starting new scrub even if a scrub is already running",
1587         "       this is useful when scrub stats record file is damaged",
1588         NULL
1589 };
1590
1591 static int cmd_scrub_start(int argc, char **argv)
1592 {
1593         return scrub_start(argc, argv, 0);
1594 }
1595
1596 static const char * const cmd_scrub_cancel_usage[] = {
1597         "btrfs scrub cancel <path>|<device>",
1598         "Cancel a running scrub",
1599         NULL
1600 };
1601
1602 static int cmd_scrub_cancel(int argc, char **argv)
1603 {
1604         char *path;
1605         int ret;
1606         int fdmnt = -1;
1607         DIR *dirstream = NULL;
1608
1609         if (check_argc_exact(argc, 2))
1610                 usage(cmd_scrub_cancel_usage);
1611
1612         path = argv[1];
1613
1614         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1615         if (fdmnt < 0) {
1616                 if (errno == EINVAL)
1617                         fprintf(stderr,
1618                                 "ERROR: '%s' is not a mounted btrfs device\n",
1619                                 path);
1620                 else
1621                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1622                                 path, strerror(errno));
1623                 ret = 1;
1624                 goto out;
1625         }
1626
1627         ret = ioctl(fdmnt, BTRFS_IOC_SCRUB_CANCEL, NULL);
1628
1629         if (ret < 0) {
1630                 fprintf(stderr, "ERROR: scrub cancel failed on %s: %s\n", path,
1631                         errno == ENOTCONN ? "not running" : strerror(errno));
1632                 if (errno == ENOTCONN)
1633                         ret = 2;
1634                 else
1635                         ret = 1;
1636                 goto out;
1637         }
1638
1639         ret = 0;
1640         printf("scrub cancelled\n");
1641
1642 out:
1643         close_file_or_dir(fdmnt, dirstream);
1644         return ret;
1645 }
1646
1647 static const char * const cmd_scrub_resume_usage[] = {
1648         "btrfs scrub resume [-BdqrR] [-c ioprio_class -n ioprio_classdata] <path>|<device>",
1649         "Resume previously canceled or interrupted scrub",
1650         "",
1651         "-B     do not background",
1652         "-d     stats per device (-B only)",
1653         "-q     be quiet",
1654         "-r     read only mode",
1655         "-R     raw print mode, print full data instead of summary",
1656         "-c     set ioprio class (see ionice(1) manpage)",
1657         "-n     set ioprio classdata (see ionice(1) manpage)",
1658         NULL
1659 };
1660
1661 static int cmd_scrub_resume(int argc, char **argv)
1662 {
1663         return scrub_start(argc, argv, 1);
1664 }
1665
1666 static const char * const cmd_scrub_status_usage[] = {
1667         "btrfs scrub status [-dR] <path>|<device>",
1668         "Show status of running or finished scrub",
1669         "",
1670         "-d     stats per device",
1671         "-R     print raw stats",
1672         NULL
1673 };
1674
1675 static int cmd_scrub_status(int argc, char **argv)
1676 {
1677         char *path;
1678         struct btrfs_ioctl_fs_info_args fi_args;
1679         struct btrfs_ioctl_dev_info_args *di_args = NULL;
1680         struct scrub_file_record **past_scrubs = NULL;
1681         struct scrub_file_record *last_scrub;
1682         struct scrub_fs_stat fs_stat;
1683         struct sockaddr_un addr = {
1684                 .sun_family = AF_UNIX,
1685         };
1686         int in_progress;
1687         int ret;
1688         int i;
1689         int fdmnt;
1690         int print_raw = 0;
1691         int do_stats_per_dev = 0;
1692         int c;
1693         char fsid[BTRFS_UUID_UNPARSED_SIZE];
1694         int fdres = -1;
1695         int err = 0;
1696         DIR *dirstream = NULL;
1697
1698         optind = 1;
1699         while ((c = getopt(argc, argv, "dR")) != -1) {
1700                 switch (c) {
1701                 case 'd':
1702                         do_stats_per_dev = 1;
1703                         break;
1704                 case 'R':
1705                         print_raw = 1;
1706                         break;
1707                 case '?':
1708                 default:
1709                         usage(cmd_scrub_status_usage);
1710                 }
1711         }
1712
1713         if (check_argc_exact(argc - optind, 1))
1714                 usage(cmd_scrub_status_usage);
1715
1716         path = argv[optind];
1717
1718         fdmnt = open_path_or_dev_mnt(path, &dirstream);
1719
1720         if (fdmnt < 0) {
1721                 if (errno == EINVAL)
1722                         fprintf(stderr,
1723                                 "ERROR: '%s' is not a mounted btrfs device\n",
1724                                 path);
1725                 else
1726                         fprintf(stderr, "ERROR: can't access '%s': %s\n",
1727                                 path, strerror(errno));
1728                 return 1;
1729         }
1730
1731         ret = get_fs_info(path, &fi_args, &di_args);
1732         if (ret) {
1733                 fprintf(stderr, "ERROR: getting dev info for scrub failed: "
1734                                 "%s\n", strerror(-ret));
1735                 err = 1;
1736                 goto out;
1737         }
1738         if (!fi_args.num_devices) {
1739                 fprintf(stderr, "ERROR: no devices found\n");
1740                 err = 1;
1741                 goto out;
1742         }
1743
1744         uuid_unparse(fi_args.fsid, fsid);
1745
1746         fdres = socket(AF_UNIX, SOCK_STREAM, 0);
1747         if (fdres == -1) {
1748                 fprintf(stderr, "ERROR: failed to create socket to "
1749                         "receive progress information: %s\n",
1750                         strerror(errno));
1751                 err = 1;
1752                 goto out;
1753         }
1754         scrub_datafile(SCRUB_PROGRESS_SOCKET_PATH, fsid,
1755                         NULL, addr.sun_path, sizeof(addr.sun_path));
1756         /* ignore EOVERFLOW, just use shorter name and hope for the best */
1757         addr.sun_path[sizeof(addr.sun_path) - 1] = '\0';
1758         ret = connect(fdres, (struct sockaddr *)&addr, sizeof(addr));
1759         if (ret == -1) {
1760                 close(fdres);
1761                 fdres = scrub_open_file_r(SCRUB_DATA_FILE, fsid);
1762                 if (fdres < 0 && fdres != -ENOENT) {
1763                         fprintf(stderr, "WARNING: failed to open status file: "
1764                                 "%s\n", strerror(-fdres));
1765                         err = 1;
1766                         goto out;
1767                 }
1768         }
1769
1770         if (fdres >= 0) {
1771                 past_scrubs = scrub_read_file(fdres, 1);
1772                 if (IS_ERR(past_scrubs))
1773                         fprintf(stderr, "WARNING: failed to read status: %s\n",
1774                                 strerror(-PTR_ERR(past_scrubs)));
1775         }
1776         in_progress = is_scrub_running_in_kernel(fdmnt, di_args, fi_args.num_devices);
1777
1778         printf("scrub status for %s\n", fsid);
1779
1780         if (do_stats_per_dev) {
1781                 for (i = 0; i < fi_args.num_devices; ++i) {
1782                         last_scrub = last_dev_scrub(past_scrubs,
1783                                                         di_args[i].devid);
1784                         if (!last_scrub) {
1785                                 print_scrub_dev(&di_args[i], NULL, print_raw,
1786                                                 NULL, NULL);
1787                                 continue;
1788                         }
1789                         last_scrub->stats.in_progress = in_progress;
1790                         print_scrub_dev(&di_args[i], &last_scrub->p, print_raw,
1791                                         last_scrub->stats.finished ?
1792                                                         "history" : "status",
1793                                         &last_scrub->stats);
1794                 }
1795         } else {
1796                 init_fs_stat(&fs_stat);
1797                 fs_stat.s.in_progress = in_progress;
1798                 for (i = 0; i < fi_args.num_devices; ++i) {
1799                         last_scrub = last_dev_scrub(past_scrubs,
1800                                                         di_args[i].devid);
1801                         if (!last_scrub)
1802                                 continue;
1803                         add_to_fs_stat(&last_scrub->p, &last_scrub->stats,
1804                                         &fs_stat);
1805                 }
1806                 print_fs_stat(&fs_stat, print_raw);
1807         }
1808
1809 out:
1810         free_history(past_scrubs);
1811         free(di_args);
1812         if (fdres > -1)
1813                 close(fdres);
1814         close_file_or_dir(fdmnt, dirstream);
1815
1816         return !!err;
1817 }
1818
1819 static const char scrub_cmd_group_info[] =
1820 "verify checksums of data and metadata";
1821
1822 const struct cmd_group scrub_cmd_group = {
1823         scrub_cmd_group_usage, scrub_cmd_group_info, {
1824                 { "start", cmd_scrub_start, cmd_scrub_start_usage, NULL, 0 },
1825                 { "cancel", cmd_scrub_cancel, cmd_scrub_cancel_usage, NULL, 0 },
1826                 { "resume", cmd_scrub_resume, cmd_scrub_resume_usage, NULL, 0 },
1827                 { "status", cmd_scrub_status, cmd_scrub_status_usage, NULL, 0 },
1828                 NULL_CMD_STRUCT
1829         }
1830 };
1831
1832 int cmd_scrub(int argc, char **argv)
1833 {
1834         return handle_command_group(&scrub_cmd_group, argc, argv);
1835 }