fs/btrfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/blkdev.h>
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/highmem.h>
  11 #include <linux/time.h>
  12 #include <linux/init.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/writeback.h>
  18 #include <linux/statfs.h>
  19 #include <linux/compat.h>
  20 #include <linux/parser.h>
  21 #include <linux/ctype.h>
  22 #include <linux/namei.h>
  23 #include <linux/miscdevice.h>
  24 #include <linux/magic.h>
  25 #include <linux/slab.h>
  26 #include <linux/cleancache.h>
  27 #include <linux/ratelimit.h>
  28 #include <linux/crc32c.h>
  29 #include <linux/btrfs.h>
  30 #include "delayed-inode.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "print-tree.h"
  36 #include "props.h"
  37 #include "xattr.h"
  38 #include "volumes.h"
  39 #include "export.h"
  40 #include "compression.h"
  41 #include "rcu-string.h"
  42 #include "dev-replace.h"
  43 #include "free-space-cache.h"
  44 #include "backref.h"
  45 #include "space-info.h"
  46 #include "sysfs.h"
  47 #include "zoned.h"
  48 #include "tests/btrfs-tests.h"
  49 #include "block-group.h"
  50 #include "discard.h"
  51 #include "qgroup.h"
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/btrfs.h>
  54
  55 static const struct super_operations btrfs_super_ops;
  56
  57 /*
  58  * Types for mounting the default subvolume and a subvolume explicitly
  59  * requested by subvol=/path. That way the callchain is straightforward and we
  60  * don't have to play tricks with the mount options and recursive calls to
  61  * btrfs_mount.
  62  *
  63  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
  64  */
  65 static struct file_system_type btrfs_fs_type;
  66 static struct file_system_type btrfs_root_fs_type;
  67
  68 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  69
  70 /*
  71  * Generally the error codes correspond to their respective errors, but there
  72  * are a few special cases.
  73  *
  74  * EUCLEAN: Any sort of corruption that we encounter.  The tree-checker for
  75  *          instance will return EUCLEAN if any of the blocks are corrupted in
  76  *          a way that is problematic.  We want to reserve EUCLEAN for these
  77  *          sort of corruptions.
  78  *
  79  * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
  80  *        need to use EROFS for this case.  We will have no idea of the
  81  *        original failure, that will have been reported at the time we tripped
  82  *        over the error.  Each subsequent error that doesn't have any context
  83  *        of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
  84  */
  85 const char * __attribute_const__ btrfs_decode_error(int errno)
  86 {
  87         char *errstr = "unknown";
  88
  89         switch (errno) {
  90         case -ENOENT:           /* -2 */
  91                 errstr = "No such entry";
  92                 break;
  93         case -EIO:              /* -5 */
  94                 errstr = "IO failure";
  95                 break;
  96         case -ENOMEM:           /* -12*/
  97                 errstr = "Out of memory";
  98                 break;
  99         case -EEXIST:           /* -17 */
 100                 errstr = "Object already exists";
 101                 break;
 102         case -ENOSPC:           /* -28 */
 103                 errstr = "No space left";
 104                 break;
 105         case -EROFS:            /* -30 */
 106                 errstr = "Readonly filesystem";
 107                 break;
 108         case -EOPNOTSUPP:       /* -95 */
 109                 errstr = "Operation not supported";
 110                 break;
 111         case -EUCLEAN:          /* -117 */
 112                 errstr = "Filesystem corrupted";
 113                 break;
 114         case -EDQUOT:           /* -122 */
 115                 errstr = "Quota exceeded";
 116                 break;
 117         }
 118
 119         return errstr;
 120 }
 121
 122 /*
 123  * __btrfs_handle_fs_error decodes expected errors from the caller and
 124  * invokes the appropriate error response.
 125  */
 126 __cold
 127 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 128                        unsigned int line, int errno, const char *fmt, ...)
 129 {
 130         struct super_block *sb = fs_info->sb;
 131 #ifdef CONFIG_PRINTK
 132         const char *errstr;
 133 #endif
 134
 135         /*
 136          * Special case: if the error is EROFS, and we're already
 137          * under SB_RDONLY, then it is safe here.
 138          */
 139         if (errno == -EROFS && sb_rdonly(sb))
 140                 return;
 141
 142 #ifdef CONFIG_PRINTK
 143         errstr = btrfs_decode_error(errno);
 144         if (fmt) {
 145                 struct va_format vaf;
 146                 va_list args;
 147
 148                 va_start(args, fmt);
 149                 vaf.fmt = fmt;
 150                 vaf.va = &args;
 151
 152                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 153                         sb->s_id, function, line, errno, errstr, &vaf);
 154                 va_end(args);
 155         } else {
 156                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 157                         sb->s_id, function, line, errno, errstr);
 158         }
 159 #endif
 160
 161         /*
 162          * Today we only save the error info to memory.  Long term we'll
 163          * also send it down to the disk
 164          */
 165         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 166
 167         /* Don't go through full error handling during mount */
 168         if (!(sb->s_flags & SB_BORN))
 169                 return;
 170
 171         if (sb_rdonly(sb))
 172                 return;
 173
 174         btrfs_discard_stop(fs_info);
 175
 176         /* btrfs handle error by forcing the filesystem readonly */
 177         btrfs_set_sb_rdonly(sb);
 178         btrfs_info(fs_info, "forced readonly");
 179         /*
 180          * Note that a running device replace operation is not canceled here
 181          * although there is no way to update the progress. It would add the
 182          * risk of a deadlock, therefore the canceling is omitted. The only
 183          * penalty is that some I/O remains active until the procedure
 184          * completes. The next time when the filesystem is mounted writable
 185          * again, the device replace operation continues.
 186          */
 187 }
 188
 189 #ifdef CONFIG_PRINTK
 190 static const char * const logtypes[] = {
 191         "emergency",
 192         "alert",
 193         "critical",
 194         "error",
 195         "warning",
 196         "notice",
 197         "info",
 198         "debug",
 199 };
 200
 201
 202 /*
 203  * Use one ratelimit state per log level so that a flood of less important
 204  * messages doesn't cause more important ones to be dropped.
 205  */
 206 static struct ratelimit_state printk_limits[] = {
 207         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 208         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 209         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 210         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 211         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 212         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 213         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 214         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 215 };
 216
 217 void __cold btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 218 {
 219         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 220         struct va_format vaf;
 221         va_list args;
 222         int kern_level;
 223         const char *type = logtypes[4];
 224         struct ratelimit_state *ratelimit = &printk_limits[4];
 225
 226         va_start(args, fmt);
 227
 228         while ((kern_level = printk_get_level(fmt)) != 0) {
 229                 size_t size = printk_skip_level(fmt) - fmt;
 230
 231                 if (kern_level >= '0' && kern_level <= '7') {
 232                         memcpy(lvl, fmt,  size);
 233                         lvl[size] = '\0';
 234                         type = logtypes[kern_level - '0'];
 235                         ratelimit = &printk_limits[kern_level - '0'];
 236                 }
 237                 fmt += size;
 238         }
 239
 240         vaf.fmt = fmt;
 241         vaf.va = &args;
 242
 243         if (__ratelimit(ratelimit)) {
 244                 if (fs_info)
 245                         printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
 246                                 fs_info->sb->s_id, &vaf);
 247                 else
 248                         printk("%sBTRFS %s: %pV\n", lvl, type, &vaf);
 249         }
 250
 251         va_end(args);
 252 }
 253 #endif
 254
 255 /*
 256  * We only mark the transaction aborted and then set the file system read-only.
 257  * This will prevent new transactions from starting or trying to join this
 258  * one.
 259  *
 260  * This means that error recovery at the call site is limited to freeing
 261  * any local memory allocations and passing the error code up without
 262  * further cleanup. The transaction should complete as it normally would
 263  * in the call path but will return -EIO.
 264  *
 265  * We'll complete the cleanup in btrfs_end_transaction and
 266  * btrfs_commit_transaction.
 267  */
 268 __cold
 269 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 270                                const char *function,
 271                                unsigned int line, int errno)
 272 {
 273         struct btrfs_fs_info *fs_info = trans->fs_info;
 274
 275         WRITE_ONCE(trans->aborted, errno);
 276         /* Nothing used. The other threads that have joined this
 277          * transaction may be able to continue. */
 278         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 279                 const char *errstr;
 280
 281                 errstr = btrfs_decode_error(errno);
 282                 btrfs_warn(fs_info,
 283                            "%s:%d: Aborting unused transaction(%s).",
 284                            function, line, errstr);
 285                 return;
 286         }
 287         WRITE_ONCE(trans->transaction->aborted, errno);
 288         /* Wake up anybody who may be waiting on this transaction */
 289         wake_up(&fs_info->transaction_wait);
 290         wake_up(&fs_info->transaction_blocked_wait);
 291         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 292 }
 293 /*
 294  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 295  * issues an alert, and either panics or BUGs, depending on mount options.
 296  */
 297 __cold
 298 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 299                    unsigned int line, int errno, const char *fmt, ...)
 300 {
 301         char *s_id = "<unknown>";
 302         const char *errstr;
 303         struct va_format vaf = { .fmt = fmt };
 304         va_list args;
 305
 306         if (fs_info)
 307                 s_id = fs_info->sb->s_id;
 308
 309         va_start(args, fmt);
 310         vaf.va = &args;
 311
 312         errstr = btrfs_decode_error(errno);
 313         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 314                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 315                         s_id, function, line, &vaf, errno, errstr);
 316
 317         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 318                    function, line, &vaf, errno, errstr);
 319         va_end(args);
 320         /* Caller calls BUG() */
 321 }
 322
 323 static void btrfs_put_super(struct super_block *sb)
 324 {
 325         close_ctree(btrfs_sb(sb));
 326 }
 327
 328 enum {
 329         Opt_acl, Opt_noacl,
 330         Opt_clear_cache,
 331         Opt_commit_interval,
 332         Opt_compress,
 333         Opt_compress_force,
 334         Opt_compress_force_type,
 335         Opt_compress_type,
 336         Opt_degraded,
 337         Opt_device,
 338         Opt_fatal_errors,
 339         Opt_flushoncommit, Opt_noflushoncommit,
 340         Opt_max_inline,
 341         Opt_barrier, Opt_nobarrier,
 342         Opt_datacow, Opt_nodatacow,
 343         Opt_datasum, Opt_nodatasum,
 344         Opt_defrag, Opt_nodefrag,
 345         Opt_discard, Opt_nodiscard,
 346         Opt_discard_mode,
 347         Opt_norecovery,
 348         Opt_ratio,
 349         Opt_rescan_uuid_tree,
 350         Opt_skip_balance,
 351         Opt_space_cache, Opt_no_space_cache,
 352         Opt_space_cache_version,
 353         Opt_ssd, Opt_nossd,
 354         Opt_ssd_spread, Opt_nossd_spread,
 355         Opt_subvol,
 356         Opt_subvol_empty,
 357         Opt_subvolid,
 358         Opt_thread_pool,
 359         Opt_treelog, Opt_notreelog,
 360         Opt_user_subvol_rm_allowed,
 361
 362         /* Rescue options */
 363         Opt_rescue,
 364         Opt_usebackuproot,
 365         Opt_nologreplay,
 366         Opt_ignorebadroots,
 367         Opt_ignoredatacsums,
 368         Opt_rescue_all,
 369
 370         /* Deprecated options */
 371         Opt_recovery,
 372         Opt_inode_cache, Opt_noinode_cache,
 373
 374         /* Debugging options */
 375         Opt_check_integrity,
 376         Opt_check_integrity_including_extent_data,
 377         Opt_check_integrity_print_mask,
 378         Opt_enospc_debug, Opt_noenospc_debug,
 379 #ifdef CONFIG_BTRFS_DEBUG
 380         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 381 #endif
 382 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 383         Opt_ref_verify,
 384 #endif
 385         Opt_err,
 386 };
 387
 388 static const match_table_t tokens = {
 389         {Opt_acl, "acl"},
 390         {Opt_noacl, "noacl"},
 391         {Opt_clear_cache, "clear_cache"},
 392         {Opt_commit_interval, "commit=%u"},
 393         {Opt_compress, "compress"},
 394         {Opt_compress_type, "compress=%s"},
 395         {Opt_compress_force, "compress-force"},
 396         {Opt_compress_force_type, "compress-force=%s"},
 397         {Opt_degraded, "degraded"},
 398         {Opt_device, "device=%s"},
 399         {Opt_fatal_errors, "fatal_errors=%s"},
 400         {Opt_flushoncommit, "flushoncommit"},
 401         {Opt_noflushoncommit, "noflushoncommit"},
 402         {Opt_inode_cache, "inode_cache"},
 403         {Opt_noinode_cache, "noinode_cache"},
 404         {Opt_max_inline, "max_inline=%s"},
 405         {Opt_barrier, "barrier"},
 406         {Opt_nobarrier, "nobarrier"},
 407         {Opt_datacow, "datacow"},
 408         {Opt_nodatacow, "nodatacow"},
 409         {Opt_datasum, "datasum"},
 410         {Opt_nodatasum, "nodatasum"},
 411         {Opt_defrag, "autodefrag"},
 412         {Opt_nodefrag, "noautodefrag"},
 413         {Opt_discard, "discard"},
 414         {Opt_discard_mode, "discard=%s"},
 415         {Opt_nodiscard, "nodiscard"},
 416         {Opt_norecovery, "norecovery"},
 417         {Opt_ratio, "metadata_ratio=%u"},
 418         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 419         {Opt_skip_balance, "skip_balance"},
 420         {Opt_space_cache, "space_cache"},
 421         {Opt_no_space_cache, "nospace_cache"},
 422         {Opt_space_cache_version, "space_cache=%s"},
 423         {Opt_ssd, "ssd"},
 424         {Opt_nossd, "nossd"},
 425         {Opt_ssd_spread, "ssd_spread"},
 426         {Opt_nossd_spread, "nossd_spread"},
 427         {Opt_subvol, "subvol=%s"},
 428         {Opt_subvol_empty, "subvol="},
 429         {Opt_subvolid, "subvolid=%s"},
 430         {Opt_thread_pool, "thread_pool=%u"},
 431         {Opt_treelog, "treelog"},
 432         {Opt_notreelog, "notreelog"},
 433         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 434
 435         /* Rescue options */
 436         {Opt_rescue, "rescue=%s"},
 437         /* Deprecated, with alias rescue=nologreplay */
 438         {Opt_nologreplay, "nologreplay"},
 439         /* Deprecated, with alias rescue=usebackuproot */
 440         {Opt_usebackuproot, "usebackuproot"},
 441
 442         /* Deprecated options */
 443         {Opt_recovery, "recovery"},
 444
 445         /* Debugging options */
 446         {Opt_check_integrity, "check_int"},
 447         {Opt_check_integrity_including_extent_data, "check_int_data"},
 448         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 449         {Opt_enospc_debug, "enospc_debug"},
 450         {Opt_noenospc_debug, "noenospc_debug"},
 451 #ifdef CONFIG_BTRFS_DEBUG
 452         {Opt_fragment_data, "fragment=data"},
 453         {Opt_fragment_metadata, "fragment=metadata"},
 454         {Opt_fragment_all, "fragment=all"},
 455 #endif
 456 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 457         {Opt_ref_verify, "ref_verify"},
 458 #endif
 459         {Opt_err, NULL},
 460 };
 461
 462 static const match_table_t rescue_tokens = {
 463         {Opt_usebackuproot, "usebackuproot"},
 464         {Opt_nologreplay, "nologreplay"},
 465         {Opt_ignorebadroots, "ignorebadroots"},
 466         {Opt_ignorebadroots, "ibadroots"},
 467         {Opt_ignoredatacsums, "ignoredatacsums"},
 468         {Opt_ignoredatacsums, "idatacsums"},
 469         {Opt_rescue_all, "all"},
 470         {Opt_err, NULL},
 471 };
 472
 473 static bool check_ro_option(struct btrfs_fs_info *fs_info, unsigned long opt,
 474                             const char *opt_name)
 475 {
 476         if (fs_info->mount_opt & opt) {
 477                 btrfs_err(fs_info, "%s must be used with ro mount option",
 478                           opt_name);
 479                 return true;
 480         }
 481         return false;
 482 }
 483
 484 static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
 485 {
 486         char *opts;
 487         char *orig;
 488         char *p;
 489         substring_t args[MAX_OPT_ARGS];
 490         int ret = 0;
 491
 492         opts = kstrdup(options, GFP_KERNEL);
 493         if (!opts)
 494                 return -ENOMEM;
 495         orig = opts;
 496
 497         while ((p = strsep(&opts, ":")) != NULL) {
 498                 int token;
 499
 500                 if (!*p)
 501                         continue;
 502                 token = match_token(p, rescue_tokens, args);
 503                 switch (token){
 504                 case Opt_usebackuproot:
 505                         btrfs_info(info,
 506                                    "trying to use backup root at mount time");
 507                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 508                         break;
 509                 case Opt_nologreplay:
 510                         btrfs_set_and_info(info, NOLOGREPLAY,
 511                                            "disabling log replay at mount time");
 512                         break;
 513                 case Opt_ignorebadroots:
 514                         btrfs_set_and_info(info, IGNOREBADROOTS,
 515                                            "ignoring bad roots");
 516                         break;
 517                 case Opt_ignoredatacsums:
 518                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 519                                            "ignoring data csums");
 520                         break;
 521                 case Opt_rescue_all:
 522                         btrfs_info(info, "enabling all of the rescue options");
 523                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 524                                            "ignoring data csums");
 525                         btrfs_set_and_info(info, IGNOREBADROOTS,
 526                                            "ignoring bad roots");
 527                         btrfs_set_and_info(info, NOLOGREPLAY,
 528                                            "disabling log replay at mount time");
 529                         break;
 530                 case Opt_err:
 531                         btrfs_info(info, "unrecognized rescue option '%s'", p);
 532                         ret = -EINVAL;
 533                         goto out;
 534                 default:
 535                         break;
 536                 }
 537
 538         }
 539 out:
 540         kfree(orig);
 541         return ret;
 542 }
 543
 544 /*
 545  * Regular mount options parser.  Everything that is needed only when
 546  * reading in a new superblock is parsed here.
 547  * XXX JDM: This needs to be cleaned up for remount.
 548  */
 549 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 550                         unsigned long new_flags)
 551 {
 552         substring_t args[MAX_OPT_ARGS];
 553         char *p, *num;
 554         int intarg;
 555         int ret = 0;
 556         char *compress_type;
 557         bool compress_force = false;
 558         enum btrfs_compression_type saved_compress_type;
 559         int saved_compress_level;
 560         bool saved_compress_force;
 561         int no_compress = 0;
 562
 563         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 564                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 565         else if (btrfs_free_space_cache_v1_active(info)) {
 566                 if (btrfs_is_zoned(info)) {
 567                         btrfs_info(info,
 568                         "zoned: clearing existing space cache");
 569                         btrfs_set_super_cache_generation(info->super_copy, 0);
 570                 } else {
 571                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 572                 }
 573         }
 574
 575         /*
 576          * Even the options are empty, we still need to do extra check
 577          * against new flags
 578          */
 579         if (!options)
 580                 goto check;
 581
 582         while ((p = strsep(&options, ",")) != NULL) {
 583                 int token;
 584                 if (!*p)
 585                         continue;
 586
 587                 token = match_token(p, tokens, args);
 588                 switch (token) {
 589                 case Opt_degraded:
 590                         btrfs_info(info, "allowing degraded mounts");
 591                         btrfs_set_opt(info->mount_opt, DEGRADED);
 592                         break;
 593                 case Opt_subvol:
 594                 case Opt_subvol_empty:
 595                 case Opt_subvolid:
 596                 case Opt_device:
 597                         /*
 598                          * These are parsed by btrfs_parse_subvol_options or
 599                          * btrfs_parse_device_options and can be ignored here.
 600                          */
 601                         break;
 602                 case Opt_nodatasum:
 603                         btrfs_set_and_info(info, NODATASUM,
 604                                            "setting nodatasum");
 605                         break;
 606                 case Opt_datasum:
 607                         if (btrfs_test_opt(info, NODATASUM)) {
 608                                 if (btrfs_test_opt(info, NODATACOW))
 609                                         btrfs_info(info,
 610                                                    "setting datasum, datacow enabled");
 611                                 else
 612                                         btrfs_info(info, "setting datasum");
 613                         }
 614                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 615                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 616                         break;
 617                 case Opt_nodatacow:
 618                         if (!btrfs_test_opt(info, NODATACOW)) {
 619                                 if (!btrfs_test_opt(info, COMPRESS) ||
 620                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 621                                         btrfs_info(info,
 622                                                    "setting nodatacow, compression disabled");
 623                                 } else {
 624                                         btrfs_info(info, "setting nodatacow");
 625                                 }
 626                         }
 627                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 628                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 629                         btrfs_set_opt(info->mount_opt, NODATACOW);
 630                         btrfs_set_opt(info->mount_opt, NODATASUM);
 631                         break;
 632                 case Opt_datacow:
 633                         btrfs_clear_and_info(info, NODATACOW,
 634                                              "setting datacow");
 635                         break;
 636                 case Opt_compress_force:
 637                 case Opt_compress_force_type:
 638                         compress_force = true;
 639                         fallthrough;
 640                 case Opt_compress:
 641                 case Opt_compress_type:
 642                         saved_compress_type = btrfs_test_opt(info,
 643                                                              COMPRESS) ?
 644                                 info->compress_type : BTRFS_COMPRESS_NONE;
 645                         saved_compress_force =
 646                                 btrfs_test_opt(info, FORCE_COMPRESS);
 647                         saved_compress_level = info->compress_level;
 648                         if (token == Opt_compress ||
 649                             token == Opt_compress_force ||
 650                             strncmp(args[0].from, "zlib", 4) == 0) {
 651                                 compress_type = "zlib";
 652
 653                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 654                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 655                                 /*
 656                                  * args[0] contains uninitialized data since
 657                                  * for these tokens we don't expect any
 658                                  * parameter.
 659                                  */
 660                                 if (token != Opt_compress &&
 661                                     token != Opt_compress_force)
 662                                         info->compress_level =
 663                                           btrfs_compress_str2level(
 664                                                         BTRFS_COMPRESS_ZLIB,
 665                                                         args[0].from + 4);
 666                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 667                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 668                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 669                                 no_compress = 0;
 670                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 671                                 compress_type = "lzo";
 672                                 info->compress_type = BTRFS_COMPRESS_LZO;
 673                                 info->compress_level = 0;
 674                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 675                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 676                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 677                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 678                                 no_compress = 0;
 679                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 680                                 compress_type = "zstd";
 681                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 682                                 info->compress_level =
 683                                         btrfs_compress_str2level(
 684                                                          BTRFS_COMPRESS_ZSTD,
 685                                                          args[0].from + 4);
 686                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 687                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 688                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 689                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 690                                 no_compress = 0;
 691                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 692                                 compress_type = "no";
 693                                 info->compress_level = 0;
 694                                 info->compress_type = 0;
 695                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 696                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 697                                 compress_force = false;
 698                                 no_compress++;
 699                         } else {
 700                                 ret = -EINVAL;
 701                                 goto out;
 702                         }
 703
 704                         if (compress_force) {
 705                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 706                         } else {
 707                                 /*
 708                                  * If we remount from compress-force=xxx to
 709                                  * compress=xxx, we need clear FORCE_COMPRESS
 710                                  * flag, otherwise, there is no way for users
 711                                  * to disable forcible compression separately.
 712                                  */
 713                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 714                         }
 715                         if (no_compress == 1) {
 716                                 btrfs_info(info, "use no compression");
 717                         } else if ((info->compress_type != saved_compress_type) ||
 718                                    (compress_force != saved_compress_force) ||
 719                                    (info->compress_level != saved_compress_level)) {
 720                                 btrfs_info(info, "%s %s compression, level %d",
 721                                            (compress_force) ? "force" : "use",
 722                                            compress_type, info->compress_level);
 723                         }
 724                         compress_force = false;
 725                         break;
 726                 case Opt_ssd:
 727                         btrfs_set_and_info(info, SSD,
 728                                            "enabling ssd optimizations");
 729                         btrfs_clear_opt(info->mount_opt, NOSSD);
 730                         break;
 731                 case Opt_ssd_spread:
 732                         btrfs_set_and_info(info, SSD,
 733                                            "enabling ssd optimizations");
 734                         btrfs_set_and_info(info, SSD_SPREAD,
 735                                            "using spread ssd allocation scheme");
 736                         btrfs_clear_opt(info->mount_opt, NOSSD);
 737                         break;
 738                 case Opt_nossd:
 739                         btrfs_set_opt(info->mount_opt, NOSSD);
 740                         btrfs_clear_and_info(info, SSD,
 741                                              "not using ssd optimizations");
 742                         fallthrough;
 743                 case Opt_nossd_spread:
 744                         btrfs_clear_and_info(info, SSD_SPREAD,
 745                                              "not using spread ssd allocation scheme");
 746                         break;
 747                 case Opt_barrier:
 748                         btrfs_clear_and_info(info, NOBARRIER,
 749                                              "turning on barriers");
 750                         break;
 751                 case Opt_nobarrier:
 752                         btrfs_set_and_info(info, NOBARRIER,
 753                                            "turning off barriers");
 754                         break;
 755                 case Opt_thread_pool:
 756                         ret = match_int(&args[0], &intarg);
 757                         if (ret) {
 758                                 goto out;
 759                         } else if (intarg == 0) {
 760                                 ret = -EINVAL;
 761                                 goto out;
 762                         }
 763                         info->thread_pool_size = intarg;
 764                         break;
 765                 case Opt_max_inline:
 766                         num = match_strdup(&args[0]);
 767                         if (num) {
 768                                 info->max_inline = memparse(num, NULL);
 769                                 kfree(num);
 770
 771                                 if (info->max_inline) {
 772                                         info->max_inline = min_t(u64,
 773                                                 info->max_inline,
 774                                                 info->sectorsize);
 775                                 }
 776                                 btrfs_info(info, "max_inline at %llu",
 777                                            info->max_inline);
 778                         } else {
 779                                 ret = -ENOMEM;
 780                                 goto out;
 781                         }
 782                         break;
 783                 case Opt_acl:
 784 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 785                         info->sb->s_flags |= SB_POSIXACL;
 786                         break;
 787 #else
 788                         btrfs_err(info, "support for ACL not compiled in!");
 789                         ret = -EINVAL;
 790                         goto out;
 791 #endif
 792                 case Opt_noacl:
 793                         info->sb->s_flags &= ~SB_POSIXACL;
 794                         break;
 795                 case Opt_notreelog:
 796                         btrfs_set_and_info(info, NOTREELOG,
 797                                            "disabling tree log");
 798                         break;
 799                 case Opt_treelog:
 800                         btrfs_clear_and_info(info, NOTREELOG,
 801                                              "enabling tree log");
 802                         break;
 803                 case Opt_norecovery:
 804                 case Opt_nologreplay:
 805                         btrfs_warn(info,
 806                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 807                         btrfs_set_and_info(info, NOLOGREPLAY,
 808                                            "disabling log replay at mount time");
 809                         break;
 810                 case Opt_flushoncommit:
 811                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 812                                            "turning on flush-on-commit");
 813                         break;
 814                 case Opt_noflushoncommit:
 815                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 816                                              "turning off flush-on-commit");
 817                         break;
 818                 case Opt_ratio:
 819                         ret = match_int(&args[0], &intarg);
 820                         if (ret)
 821                                 goto out;
 822                         info->metadata_ratio = intarg;
 823                         btrfs_info(info, "metadata ratio %u",
 824                                    info->metadata_ratio);
 825                         break;
 826                 case Opt_discard:
 827                 case Opt_discard_mode:
 828                         if (token == Opt_discard ||
 829                             strcmp(args[0].from, "sync") == 0) {
 830                                 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
 831                                 btrfs_set_and_info(info, DISCARD_SYNC,
 832                                                    "turning on sync discard");
 833                         } else if (strcmp(args[0].from, "async") == 0) {
 834                                 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
 835                                 btrfs_set_and_info(info, DISCARD_ASYNC,
 836                                                    "turning on async discard");
 837                         } else {
 838                                 ret = -EINVAL;
 839                                 goto out;
 840                         }
 841                         break;
 842                 case Opt_nodiscard:
 843                         btrfs_clear_and_info(info, DISCARD_SYNC,
 844                                              "turning off discard");
 845                         btrfs_clear_and_info(info, DISCARD_ASYNC,
 846                                              "turning off async discard");
 847                         break;
 848                 case Opt_space_cache:
 849                 case Opt_space_cache_version:
 850                         if (token == Opt_space_cache ||
 851                             strcmp(args[0].from, "v1") == 0) {
 852                                 btrfs_clear_opt(info->mount_opt,
 853                                                 FREE_SPACE_TREE);
 854                                 btrfs_set_and_info(info, SPACE_CACHE,
 855                                            "enabling disk space caching");
 856                         } else if (strcmp(args[0].from, "v2") == 0) {
 857                                 btrfs_clear_opt(info->mount_opt,
 858                                                 SPACE_CACHE);
 859                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 860                                                    "enabling free space tree");
 861                         } else {
 862                                 ret = -EINVAL;
 863                                 goto out;
 864                         }
 865                         break;
 866                 case Opt_rescan_uuid_tree:
 867                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 868                         break;
 869                 case Opt_no_space_cache:
 870                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 871                                 btrfs_clear_and_info(info, SPACE_CACHE,
 872                                              "disabling disk space caching");
 873                         }
 874                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 875                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 876                                              "disabling free space tree");
 877                         }
 878                         break;
 879                 case Opt_inode_cache:
 880                 case Opt_noinode_cache:
 881                         btrfs_warn(info,
 882         "the 'inode_cache' option is deprecated and has no effect since 5.11");
 883                         break;
 884                 case Opt_clear_cache:
 885                         btrfs_set_and_info(info, CLEAR_CACHE,
 886                                            "force clearing of disk cache");
 887                         break;
 888                 case Opt_user_subvol_rm_allowed:
 889                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 890                         break;
 891                 case Opt_enospc_debug:
 892                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 893                         break;
 894                 case Opt_noenospc_debug:
 895                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 896                         break;
 897                 case Opt_defrag:
 898                         btrfs_set_and_info(info, AUTO_DEFRAG,
 899                                            "enabling auto defrag");
 900                         break;
 901                 case Opt_nodefrag:
 902                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 903                                              "disabling auto defrag");
 904                         break;
 905                 case Opt_recovery:
 906                 case Opt_usebackuproot:
 907                         btrfs_warn(info,
 908                         "'%s' is deprecated, use 'rescue=usebackuproot' instead",
 909                                    token == Opt_recovery ? "recovery" :
 910                                    "usebackuproot");
 911                         btrfs_info(info,
 912                                    "trying to use backup root at mount time");
 913                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 914                         break;
 915                 case Opt_skip_balance:
 916                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 917                         break;
 918 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 919                 case Opt_check_integrity_including_extent_data:
 920                         btrfs_info(info,
 921                                    "enabling check integrity including extent data");
 922                         btrfs_set_opt(info->mount_opt,
 923                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 924                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 925                         break;
 926                 case Opt_check_integrity:
 927                         btrfs_info(info, "enabling check integrity");
 928                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 929                         break;
 930                 case Opt_check_integrity_print_mask:
 931                         ret = match_int(&args[0], &intarg);
 932                         if (ret)
 933                                 goto out;
 934                         info->check_integrity_print_mask = intarg;
 935                         btrfs_info(info, "check_integrity_print_mask 0x%x",
 936                                    info->check_integrity_print_mask);
 937                         break;
 938 #else
 939                 case Opt_check_integrity_including_extent_data:
 940                 case Opt_check_integrity:
 941                 case Opt_check_integrity_print_mask:
 942                         btrfs_err(info,
 943                                   "support for check_integrity* not compiled in!");
 944                         ret = -EINVAL;
 945                         goto out;
 946 #endif
 947                 case Opt_fatal_errors:
 948                         if (strcmp(args[0].from, "panic") == 0)
 949                                 btrfs_set_opt(info->mount_opt,
 950                                               PANIC_ON_FATAL_ERROR);
 951                         else if (strcmp(args[0].from, "bug") == 0)
 952                                 btrfs_clear_opt(info->mount_opt,
 953                                               PANIC_ON_FATAL_ERROR);
 954                         else {
 955                                 ret = -EINVAL;
 956                                 goto out;
 957                         }
 958                         break;
 959                 case Opt_commit_interval:
 960                         intarg = 0;
 961                         ret = match_int(&args[0], &intarg);
 962                         if (ret)
 963                                 goto out;
 964                         if (intarg == 0) {
 965                                 btrfs_info(info,
 966                                            "using default commit interval %us",
 967                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 968                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
 969                         } else if (intarg > 300) {
 970                                 btrfs_warn(info, "excessive commit interval %d",
 971                                            intarg);
 972                         }
 973                         info->commit_interval = intarg;
 974                         break;
 975                 case Opt_rescue:
 976                         ret = parse_rescue_options(info, args[0].from);
 977                         if (ret < 0)
 978                                 goto out;
 979                         break;
 980 #ifdef CONFIG_BTRFS_DEBUG
 981                 case Opt_fragment_all:
 982                         btrfs_info(info, "fragmenting all space");
 983                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 984                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 985                         break;
 986                 case Opt_fragment_metadata:
 987                         btrfs_info(info, "fragmenting metadata");
 988                         btrfs_set_opt(info->mount_opt,
 989                                       FRAGMENT_METADATA);
 990                         break;
 991                 case Opt_fragment_data:
 992                         btrfs_info(info, "fragmenting data");
 993                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 994                         break;
 995 #endif
 996 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 997                 case Opt_ref_verify:
 998                         btrfs_info(info, "doing ref verification");
 999                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
1000                         break;
1001 #endif
1002                 case Opt_err:
1003                         btrfs_err(info, "unrecognized mount option '%s'", p);
1004                         ret = -EINVAL;
1005                         goto out;
1006                 default:
1007                         break;
1008                 }
1009         }
1010 check:
1011         /* We're read-only, don't have to check. */
1012         if (new_flags & SB_RDONLY)
1013                 goto out;
1014
1015         if (check_ro_option(info, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
1016             check_ro_option(info, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
1017             check_ro_option(info, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums"))
1018                 ret = -EINVAL;
1019 out:
1020         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
1021             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
1022             !btrfs_test_opt(info, CLEAR_CACHE)) {
1023                 btrfs_err(info, "cannot disable free space tree");
1024                 ret = -EINVAL;
1025
1026         }
1027         if (!ret)
1028                 ret = btrfs_check_mountopts_zoned(info);
1029         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
1030                 btrfs_info(info, "disk space caching is enabled");
1031         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
1032                 btrfs_info(info, "using free space tree");
1033         return ret;
1034 }
1035
1036 /*
1037  * Parse mount options that are required early in the mount process.
1038  *
1039  * All other options will be parsed on much later in the mount process and
1040  * only when we need to allocate a new super block.
1041  */
1042 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1043                                       void *holder)
1044 {
1045         substring_t args[MAX_OPT_ARGS];
1046         char *device_name, *opts, *orig, *p;
1047         struct btrfs_device *device = NULL;
1048         int error = 0;
1049
1050         lockdep_assert_held(&uuid_mutex);
1051
1052         if (!options)
1053                 return 0;
1054
1055         /*
1056          * strsep changes the string, duplicate it because btrfs_parse_options
1057          * gets called later
1058          */
1059         opts = kstrdup(options, GFP_KERNEL);
1060         if (!opts)
1061                 return -ENOMEM;
1062         orig = opts;
1063
1064         while ((p = strsep(&opts, ",")) != NULL) {
1065                 int token;
1066
1067                 if (!*p)
1068                         continue;
1069
1070                 token = match_token(p, tokens, args);
1071                 if (token == Opt_device) {
1072                         device_name = match_strdup(&args[0]);
1073                         if (!device_name) {
1074                                 error = -ENOMEM;
1075                                 goto out;
1076                         }
1077                         device = btrfs_scan_one_device(device_name, flags,
1078                                         holder);
1079                         kfree(device_name);
1080                         if (IS_ERR(device)) {
1081                                 error = PTR_ERR(device);
1082                                 goto out;
1083                         }
1084                 }
1085         }
1086
1087 out:
1088         kfree(orig);
1089         return error;
1090 }
1091
1092 /*
1093  * Parse mount options that are related to subvolume id
1094  *
1095  * The value is later passed to mount_subvol()
1096  */
1097 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1098                 u64 *subvol_objectid)
1099 {
1100         substring_t args[MAX_OPT_ARGS];
1101         char *opts, *orig, *p;
1102         int error = 0;
1103         u64 subvolid;
1104
1105         if (!options)
1106                 return 0;
1107
1108         /*
1109          * strsep changes the string, duplicate it because
1110          * btrfs_parse_device_options gets called later
1111          */
1112         opts = kstrdup(options, GFP_KERNEL);
1113         if (!opts)
1114                 return -ENOMEM;
1115         orig = opts;
1116
1117         while ((p = strsep(&opts, ",")) != NULL) {
1118                 int token;
1119                 if (!*p)
1120                         continue;
1121
1122                 token = match_token(p, tokens, args);
1123                 switch (token) {
1124                 case Opt_subvol:
1125                         kfree(*subvol_name);
1126                         *subvol_name = match_strdup(&args[0]);
1127                         if (!*subvol_name) {
1128                                 error = -ENOMEM;
1129                                 goto out;
1130                         }
1131                         break;
1132                 case Opt_subvolid:
1133                         error = match_u64(&args[0], &subvolid);
1134                         if (error)
1135                                 goto out;
1136
1137                         /* we want the original fs_tree */
1138                         if (subvolid == 0)
1139                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1140
1141                         *subvol_objectid = subvolid;
1142                         break;
1143                 default:
1144                         break;
1145                 }
1146         }
1147
1148 out:
1149         kfree(orig);
1150         return error;
1151 }
1152
1153 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1154                                           u64 subvol_objectid)
1155 {
1156         struct btrfs_root *root = fs_info->tree_root;
1157         struct btrfs_root *fs_root = NULL;
1158         struct btrfs_root_ref *root_ref;
1159         struct btrfs_inode_ref *inode_ref;
1160         struct btrfs_key key;
1161         struct btrfs_path *path = NULL;
1162         char *name = NULL, *ptr;
1163         u64 dirid;
1164         int len;
1165         int ret;
1166
1167         path = btrfs_alloc_path();
1168         if (!path) {
1169                 ret = -ENOMEM;
1170                 goto err;
1171         }
1172
1173         name = kmalloc(PATH_MAX, GFP_KERNEL);
1174         if (!name) {
1175                 ret = -ENOMEM;
1176                 goto err;
1177         }
1178         ptr = name + PATH_MAX - 1;
1179         ptr[0] = '\0';
1180
1181         /*
1182          * Walk up the subvolume trees in the tree of tree roots by root
1183          * backrefs until we hit the top-level subvolume.
1184          */
1185         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1186                 key.objectid = subvol_objectid;
1187                 key.type = BTRFS_ROOT_BACKREF_KEY;
1188                 key.offset = (u64)-1;
1189
1190                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1191                 if (ret < 0) {
1192                         goto err;
1193                 } else if (ret > 0) {
1194                         ret = btrfs_previous_item(root, path, subvol_objectid,
1195                                                   BTRFS_ROOT_BACKREF_KEY);
1196                         if (ret < 0) {
1197                                 goto err;
1198                         } else if (ret > 0) {
1199                                 ret = -ENOENT;
1200                                 goto err;
1201                         }
1202                 }
1203
1204                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1205                 subvol_objectid = key.offset;
1206
1207                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1208                                           struct btrfs_root_ref);
1209                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1210                 ptr -= len + 1;
1211                 if (ptr < name) {
1212                         ret = -ENAMETOOLONG;
1213                         goto err;
1214                 }
1215                 read_extent_buffer(path->nodes[0], ptr + 1,
1216                                    (unsigned long)(root_ref + 1), len);
1217                 ptr[0] = '/';
1218                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1219                 btrfs_release_path(path);
1220
1221                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1222                 if (IS_ERR(fs_root)) {
1223                         ret = PTR_ERR(fs_root);
1224                         fs_root = NULL;
1225                         goto err;
1226                 }
1227
1228                 /*
1229                  * Walk up the filesystem tree by inode refs until we hit the
1230                  * root directory.
1231                  */
1232                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1233                         key.objectid = dirid;
1234                         key.type = BTRFS_INODE_REF_KEY;
1235                         key.offset = (u64)-1;
1236
1237                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1238                         if (ret < 0) {
1239                                 goto err;
1240                         } else if (ret > 0) {
1241                                 ret = btrfs_previous_item(fs_root, path, dirid,
1242                                                           BTRFS_INODE_REF_KEY);
1243                                 if (ret < 0) {
1244                                         goto err;
1245                                 } else if (ret > 0) {
1246                                         ret = -ENOENT;
1247                                         goto err;
1248                                 }
1249                         }
1250
1251                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1252                         dirid = key.offset;
1253
1254                         inode_ref = btrfs_item_ptr(path->nodes[0],
1255                                                    path->slots[0],
1256                                                    struct btrfs_inode_ref);
1257                         len = btrfs_inode_ref_name_len(path->nodes[0],
1258                                                        inode_ref);
1259                         ptr -= len + 1;
1260                         if (ptr < name) {
1261                                 ret = -ENAMETOOLONG;
1262                                 goto err;
1263                         }
1264                         read_extent_buffer(path->nodes[0], ptr + 1,
1265                                            (unsigned long)(inode_ref + 1), len);
1266                         ptr[0] = '/';
1267                         btrfs_release_path(path);
1268                 }
1269                 btrfs_put_root(fs_root);
1270                 fs_root = NULL;
1271         }
1272
1273         btrfs_free_path(path);
1274         if (ptr == name + PATH_MAX - 1) {
1275                 name[0] = '/';
1276                 name[1] = '\0';
1277         } else {
1278                 memmove(name, ptr, name + PATH_MAX - ptr);
1279         }
1280         return name;
1281
1282 err:
1283         btrfs_put_root(fs_root);
1284         btrfs_free_path(path);
1285         kfree(name);
1286         return ERR_PTR(ret);
1287 }
1288
1289 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1290 {
1291         struct btrfs_root *root = fs_info->tree_root;
1292         struct btrfs_dir_item *di;
1293         struct btrfs_path *path;
1294         struct btrfs_key location;
1295         u64 dir_id;
1296
1297         path = btrfs_alloc_path();
1298         if (!path)
1299                 return -ENOMEM;
1300
1301         /*
1302          * Find the "default" dir item which points to the root item that we
1303          * will mount by default if we haven't been given a specific subvolume
1304          * to mount.
1305          */
1306         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1307         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1308         if (IS_ERR(di)) {
1309                 btrfs_free_path(path);
1310                 return PTR_ERR(di);
1311         }
1312         if (!di) {
1313                 /*
1314                  * Ok the default dir item isn't there.  This is weird since
1315                  * it's always been there, but don't freak out, just try and
1316                  * mount the top-level subvolume.
1317                  */
1318                 btrfs_free_path(path);
1319                 *objectid = BTRFS_FS_TREE_OBJECTID;
1320                 return 0;
1321         }
1322
1323         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1324         btrfs_free_path(path);
1325         *objectid = location.objectid;
1326         return 0;
1327 }
1328
1329 static int btrfs_fill_super(struct super_block *sb,
1330                             struct btrfs_fs_devices *fs_devices,
1331                             void *data)
1332 {
1333         struct inode *inode;
1334         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1335         int err;
1336
1337         sb->s_maxbytes = MAX_LFS_FILESIZE;
1338         sb->s_magic = BTRFS_SUPER_MAGIC;
1339         sb->s_op = &btrfs_super_ops;
1340         sb->s_d_op = &btrfs_dentry_operations;
1341         sb->s_export_op = &btrfs_export_ops;
1342         sb->s_xattr = btrfs_xattr_handlers;
1343         sb->s_time_gran = 1;
1344 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1345         sb->s_flags |= SB_POSIXACL;
1346 #endif
1347         sb->s_flags |= SB_I_VERSION;
1348         sb->s_iflags |= SB_I_CGROUPWB;
1349
1350         err = super_setup_bdi(sb);
1351         if (err) {
1352                 btrfs_err(fs_info, "super_setup_bdi failed");
1353                 return err;
1354         }
1355
1356         err = open_ctree(sb, fs_devices, (char *)data);
1357         if (err) {
1358                 btrfs_err(fs_info, "open_ctree failed");
1359                 return err;
1360         }
1361
1362         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1363         if (IS_ERR(inode)) {
1364                 err = PTR_ERR(inode);
1365                 goto fail_close;
1366         }
1367
1368         sb->s_root = d_make_root(inode);
1369         if (!sb->s_root) {
1370                 err = -ENOMEM;
1371                 goto fail_close;
1372         }
1373
1374         cleancache_init_fs(sb);
1375         sb->s_flags |= SB_ACTIVE;
1376         return 0;
1377
1378 fail_close:
1379         close_ctree(fs_info);
1380         return err;
1381 }
1382
1383 int btrfs_sync_fs(struct super_block *sb, int wait)
1384 {
1385         struct btrfs_trans_handle *trans;
1386         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1387         struct btrfs_root *root = fs_info->tree_root;
1388
1389         trace_btrfs_sync_fs(fs_info, wait);
1390
1391         if (!wait) {
1392                 filemap_flush(fs_info->btree_inode->i_mapping);
1393                 return 0;
1394         }
1395
1396         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1397
1398         trans = btrfs_attach_transaction_barrier(root);
1399         if (IS_ERR(trans)) {
1400                 /* no transaction, don't bother */
1401                 if (PTR_ERR(trans) == -ENOENT) {
1402                         /*
1403                          * Exit unless we have some pending changes
1404                          * that need to go through commit
1405                          */
1406                         if (fs_info->pending_changes == 0)
1407                                 return 0;
1408                         /*
1409                          * A non-blocking test if the fs is frozen. We must not
1410                          * start a new transaction here otherwise a deadlock
1411                          * happens. The pending operations are delayed to the
1412                          * next commit after thawing.
1413                          */
1414                         if (sb_start_write_trylock(sb))
1415                                 sb_end_write(sb);
1416                         else
1417                                 return 0;
1418                         trans = btrfs_start_transaction(root, 0);
1419                 }
1420                 if (IS_ERR(trans))
1421                         return PTR_ERR(trans);
1422         }
1423         return btrfs_commit_transaction(trans);
1424 }
1425
1426 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1427 {
1428         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1429         *printed = true;
1430 }
1431
1432 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1433 {
1434         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1435         const char *compress_type;
1436         const char *subvol_name;
1437         bool printed = false;
1438
1439         if (btrfs_test_opt(info, DEGRADED))
1440                 seq_puts(seq, ",degraded");
1441         if (btrfs_test_opt(info, NODATASUM))
1442                 seq_puts(seq, ",nodatasum");
1443         if (btrfs_test_opt(info, NODATACOW))
1444                 seq_puts(seq, ",nodatacow");
1445         if (btrfs_test_opt(info, NOBARRIER))
1446                 seq_puts(seq, ",nobarrier");
1447         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1448                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1449         if (info->thread_pool_size !=  min_t(unsigned long,
1450                                              num_online_cpus() + 2, 8))
1451                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1452         if (btrfs_test_opt(info, COMPRESS)) {
1453                 compress_type = btrfs_compress_type2str(info->compress_type);
1454                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1455                         seq_printf(seq, ",compress-force=%s", compress_type);
1456                 else
1457                         seq_printf(seq, ",compress=%s", compress_type);
1458                 if (info->compress_level)
1459                         seq_printf(seq, ":%d", info->compress_level);
1460         }
1461         if (btrfs_test_opt(info, NOSSD))
1462                 seq_puts(seq, ",nossd");
1463         if (btrfs_test_opt(info, SSD_SPREAD))
1464                 seq_puts(seq, ",ssd_spread");
1465         else if (btrfs_test_opt(info, SSD))
1466                 seq_puts(seq, ",ssd");
1467         if (btrfs_test_opt(info, NOTREELOG))
1468                 seq_puts(seq, ",notreelog");
1469         if (btrfs_test_opt(info, NOLOGREPLAY))
1470                 print_rescue_option(seq, "nologreplay", &printed);
1471         if (btrfs_test_opt(info, USEBACKUPROOT))
1472                 print_rescue_option(seq, "usebackuproot", &printed);
1473         if (btrfs_test_opt(info, IGNOREBADROOTS))
1474                 print_rescue_option(seq, "ignorebadroots", &printed);
1475         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1476                 print_rescue_option(seq, "ignoredatacsums", &printed);
1477         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1478                 seq_puts(seq, ",flushoncommit");
1479         if (btrfs_test_opt(info, DISCARD_SYNC))
1480                 seq_puts(seq, ",discard");
1481         if (btrfs_test_opt(info, DISCARD_ASYNC))
1482                 seq_puts(seq, ",discard=async");
1483         if (!(info->sb->s_flags & SB_POSIXACL))
1484                 seq_puts(seq, ",noacl");
1485         if (btrfs_free_space_cache_v1_active(info))
1486                 seq_puts(seq, ",space_cache");
1487         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1488                 seq_puts(seq, ",space_cache=v2");
1489         else
1490                 seq_puts(seq, ",nospace_cache");
1491         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1492                 seq_puts(seq, ",rescan_uuid_tree");
1493         if (btrfs_test_opt(info, CLEAR_CACHE))
1494                 seq_puts(seq, ",clear_cache");
1495         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1496                 seq_puts(seq, ",user_subvol_rm_allowed");
1497         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1498                 seq_puts(seq, ",enospc_debug");
1499         if (btrfs_test_opt(info, AUTO_DEFRAG))
1500                 seq_puts(seq, ",autodefrag");
1501         if (btrfs_test_opt(info, SKIP_BALANCE))
1502                 seq_puts(seq, ",skip_balance");
1503 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1504         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1505                 seq_puts(seq, ",check_int_data");
1506         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1507                 seq_puts(seq, ",check_int");
1508         if (info->check_integrity_print_mask)
1509                 seq_printf(seq, ",check_int_print_mask=%d",
1510                                 info->check_integrity_print_mask);
1511 #endif
1512         if (info->metadata_ratio)
1513                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1514         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1515                 seq_puts(seq, ",fatal_errors=panic");
1516         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1517                 seq_printf(seq, ",commit=%u", info->commit_interval);
1518 #ifdef CONFIG_BTRFS_DEBUG
1519         if (btrfs_test_opt(info, FRAGMENT_DATA))
1520                 seq_puts(seq, ",fragment=data");
1521         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1522                 seq_puts(seq, ",fragment=metadata");
1523 #endif
1524         if (btrfs_test_opt(info, REF_VERIFY))
1525                 seq_puts(seq, ",ref_verify");
1526         seq_printf(seq, ",subvolid=%llu",
1527                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1528         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1529                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1530         if (!IS_ERR(subvol_name)) {
1531                 seq_puts(seq, ",subvol=");
1532                 seq_escape(seq, subvol_name, " \t\n\\");
1533                 kfree(subvol_name);
1534         }
1535         return 0;
1536 }
1537
1538 static int btrfs_test_super(struct super_block *s, void *data)
1539 {
1540         struct btrfs_fs_info *p = data;
1541         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1542
1543         return fs_info->fs_devices == p->fs_devices;
1544 }
1545
1546 static int btrfs_set_super(struct super_block *s, void *data)
1547 {
1548         int err = set_anon_super(s, data);
1549         if (!err)
1550                 s->s_fs_info = data;
1551         return err;
1552 }
1553
1554 /*
1555  * subvolumes are identified by ino 256
1556  */
1557 static inline int is_subvolume_inode(struct inode *inode)
1558 {
1559         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1560                 return 1;
1561         return 0;
1562 }
1563
1564 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1565                                    struct vfsmount *mnt)
1566 {
1567         struct dentry *root;
1568         int ret;
1569
1570         if (!subvol_name) {
1571                 if (!subvol_objectid) {
1572                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1573                                                           &subvol_objectid);
1574                         if (ret) {
1575                                 root = ERR_PTR(ret);
1576                                 goto out;
1577                         }
1578                 }
1579                 subvol_name = btrfs_get_subvol_name_from_objectid(
1580                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1581                 if (IS_ERR(subvol_name)) {
1582                         root = ERR_CAST(subvol_name);
1583                         subvol_name = NULL;
1584                         goto out;
1585                 }
1586
1587         }
1588
1589         root = mount_subtree(mnt, subvol_name);
1590         /* mount_subtree() drops our reference on the vfsmount. */
1591         mnt = NULL;
1592
1593         if (!IS_ERR(root)) {
1594                 struct super_block *s = root->d_sb;
1595                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1596                 struct inode *root_inode = d_inode(root);
1597                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1598
1599                 ret = 0;
1600                 if (!is_subvolume_inode(root_inode)) {
1601                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1602                                subvol_name);
1603                         ret = -EINVAL;
1604                 }
1605                 if (subvol_objectid && root_objectid != subvol_objectid) {
1606                         /*
1607                          * This will also catch a race condition where a
1608                          * subvolume which was passed by ID is renamed and
1609                          * another subvolume is renamed over the old location.
1610                          */
1611                         btrfs_err(fs_info,
1612                                   "subvol '%s' does not match subvolid %llu",
1613                                   subvol_name, subvol_objectid);
1614                         ret = -EINVAL;
1615                 }
1616                 if (ret) {
1617                         dput(root);
1618                         root = ERR_PTR(ret);
1619                         deactivate_locked_super(s);
1620                 }
1621         }
1622
1623 out:
1624         mntput(mnt);
1625         kfree(subvol_name);
1626         return root;
1627 }
1628
1629 /*
1630  * Find a superblock for the given device / mount point.
1631  *
1632  * Note: This is based on mount_bdev from fs/super.c with a few additions
1633  *       for multiple device setup.  Make sure to keep it in sync.
1634  */
1635 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1636                 int flags, const char *device_name, void *data)
1637 {
1638         struct block_device *bdev = NULL;
1639         struct super_block *s;
1640         struct btrfs_device *device = NULL;
1641         struct btrfs_fs_devices *fs_devices = NULL;
1642         struct btrfs_fs_info *fs_info = NULL;
1643         void *new_sec_opts = NULL;
1644         fmode_t mode = FMODE_READ;
1645         int error = 0;
1646
1647         if (!(flags & SB_RDONLY))
1648                 mode |= FMODE_WRITE;
1649
1650         if (data) {
1651                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1652                 if (error)
1653                         return ERR_PTR(error);
1654         }
1655
1656         /*
1657          * Setup a dummy root and fs_info for test/set super.  This is because
1658          * we don't actually fill this stuff out until open_ctree, but we need
1659          * then open_ctree will properly initialize the file system specific
1660          * settings later.  btrfs_init_fs_info initializes the static elements
1661          * of the fs_info (locks and such) to make cleanup easier if we find a
1662          * superblock with our given fs_devices later on at sget() time.
1663          */
1664         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1665         if (!fs_info) {
1666                 error = -ENOMEM;
1667                 goto error_sec_opts;
1668         }
1669         btrfs_init_fs_info(fs_info);
1670
1671         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1672         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1673         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1674                 error = -ENOMEM;
1675                 goto error_fs_info;
1676         }
1677
1678         mutex_lock(&uuid_mutex);
1679         error = btrfs_parse_device_options(data, mode, fs_type);
1680         if (error) {
1681                 mutex_unlock(&uuid_mutex);
1682                 goto error_fs_info;
1683         }
1684
1685         device = btrfs_scan_one_device(device_name, mode, fs_type);
1686         if (IS_ERR(device)) {
1687                 mutex_unlock(&uuid_mutex);
1688                 error = PTR_ERR(device);
1689                 goto error_fs_info;
1690         }
1691
1692         fs_devices = device->fs_devices;
1693         fs_info->fs_devices = fs_devices;
1694
1695         error = btrfs_open_devices(fs_devices, mode, fs_type);
1696         mutex_unlock(&uuid_mutex);
1697         if (error)
1698                 goto error_fs_info;
1699
1700         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1701                 error = -EACCES;
1702                 goto error_close_devices;
1703         }
1704
1705         bdev = fs_devices->latest_bdev;
1706         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1707                  fs_info);
1708         if (IS_ERR(s)) {
1709                 error = PTR_ERR(s);
1710                 goto error_close_devices;
1711         }
1712
1713         if (s->s_root) {
1714                 btrfs_close_devices(fs_devices);
1715                 btrfs_free_fs_info(fs_info);
1716                 if ((flags ^ s->s_flags) & SB_RDONLY)
1717                         error = -EBUSY;
1718         } else {
1719                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1720                 btrfs_sb(s)->bdev_holder = fs_type;
1721                 if (!strstr(crc32c_impl(), "generic"))
1722                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1723                 error = btrfs_fill_super(s, fs_devices, data);
1724         }
1725         if (!error)
1726                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1727         security_free_mnt_opts(&new_sec_opts);
1728         if (error) {
1729                 deactivate_locked_super(s);
1730                 return ERR_PTR(error);
1731         }
1732
1733         return dget(s->s_root);
1734
1735 error_close_devices:
1736         btrfs_close_devices(fs_devices);
1737 error_fs_info:
1738         btrfs_free_fs_info(fs_info);
1739 error_sec_opts:
1740         security_free_mnt_opts(&new_sec_opts);
1741         return ERR_PTR(error);
1742 }
1743
1744 /*
1745  * Mount function which is called by VFS layer.
1746  *
1747  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1748  * which needs vfsmount* of device's root (/).  This means device's root has to
1749  * be mounted internally in any case.
1750  *
1751  * Operation flow:
1752  *   1. Parse subvol id related options for later use in mount_subvol().
1753  *
1754  *   2. Mount device's root (/) by calling vfs_kern_mount().
1755  *
1756  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1757  *      first place. In order to avoid calling btrfs_mount() again, we use
1758  *      different file_system_type which is not registered to VFS by
1759  *      register_filesystem() (btrfs_root_fs_type). As a result,
1760  *      btrfs_mount_root() is called. The return value will be used by
1761  *      mount_subtree() in mount_subvol().
1762  *
1763  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1764  *      "btrfs subvolume set-default", mount_subvol() is called always.
1765  */
1766 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1767                 const char *device_name, void *data)
1768 {
1769         struct vfsmount *mnt_root;
1770         struct dentry *root;
1771         char *subvol_name = NULL;
1772         u64 subvol_objectid = 0;
1773         int error = 0;
1774
1775         error = btrfs_parse_subvol_options(data, &subvol_name,
1776                                         &subvol_objectid);
1777         if (error) {
1778                 kfree(subvol_name);
1779                 return ERR_PTR(error);
1780         }
1781
1782         /* mount device's root (/) */
1783         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1784         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1785                 if (flags & SB_RDONLY) {
1786                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1787                                 flags & ~SB_RDONLY, device_name, data);
1788                 } else {
1789                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1790                                 flags | SB_RDONLY, device_name, data);
1791                         if (IS_ERR(mnt_root)) {
1792                                 root = ERR_CAST(mnt_root);
1793                                 kfree(subvol_name);
1794                                 goto out;
1795                         }
1796
1797                         down_write(&mnt_root->mnt_sb->s_umount);
1798                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1799                         up_write(&mnt_root->mnt_sb->s_umount);
1800                         if (error < 0) {
1801                                 root = ERR_PTR(error);
1802                                 mntput(mnt_root);
1803                                 kfree(subvol_name);
1804                                 goto out;
1805                         }
1806                 }
1807         }
1808         if (IS_ERR(mnt_root)) {
1809                 root = ERR_CAST(mnt_root);
1810                 kfree(subvol_name);
1811                 goto out;
1812         }
1813
1814         /* mount_subvol() will free subvol_name and mnt_root */
1815         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1816
1817 out:
1818         return root;
1819 }
1820
1821 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1822                                      u32 new_pool_size, u32 old_pool_size)
1823 {
1824         if (new_pool_size == old_pool_size)
1825                 return;
1826
1827         fs_info->thread_pool_size = new_pool_size;
1828
1829         btrfs_info(fs_info, "resize thread pool %d -> %d",
1830                old_pool_size, new_pool_size);
1831
1832         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1833         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1834         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1835         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1836         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1837         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1838                                 new_pool_size);
1839         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1840         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1841         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1842         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1843         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1844                                 new_pool_size);
1845 }
1846
1847 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1848                                        unsigned long old_opts, int flags)
1849 {
1850         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1851             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1852              (flags & SB_RDONLY))) {
1853                 /* wait for any defraggers to finish */
1854                 wait_event(fs_info->transaction_wait,
1855                            (atomic_read(&fs_info->defrag_running) == 0));
1856                 if (flags & SB_RDONLY)
1857                         sync_filesystem(fs_info->sb);
1858         }
1859 }
1860
1861 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1862                                          unsigned long old_opts)
1863 {
1864         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1865
1866         /*
1867          * We need to cleanup all defragable inodes if the autodefragment is
1868          * close or the filesystem is read only.
1869          */
1870         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1871             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1872                 btrfs_cleanup_defrag_inodes(fs_info);
1873         }
1874
1875         /* If we toggled discard async */
1876         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1877             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1878                 btrfs_discard_resume(fs_info);
1879         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1880                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1881                 btrfs_discard_cleanup(fs_info);
1882
1883         /* If we toggled space cache */
1884         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1885                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1886 }
1887
1888 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1889 {
1890         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1891         unsigned old_flags = sb->s_flags;
1892         unsigned long old_opts = fs_info->mount_opt;
1893         unsigned long old_compress_type = fs_info->compress_type;
1894         u64 old_max_inline = fs_info->max_inline;
1895         u32 old_thread_pool_size = fs_info->thread_pool_size;
1896         u32 old_metadata_ratio = fs_info->metadata_ratio;
1897         int ret;
1898
1899         sync_filesystem(sb);
1900         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1901
1902         if (data) {
1903                 void *new_sec_opts = NULL;
1904
1905                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1906                 if (!ret)
1907                         ret = security_sb_remount(sb, new_sec_opts);
1908                 security_free_mnt_opts(&new_sec_opts);
1909                 if (ret)
1910                         goto restore;
1911         }
1912
1913         ret = btrfs_parse_options(fs_info, data, *flags);
1914         if (ret)
1915                 goto restore;
1916
1917         btrfs_remount_begin(fs_info, old_opts, *flags);
1918         btrfs_resize_thread_pool(fs_info,
1919                 fs_info->thread_pool_size, old_thread_pool_size);
1920
1921         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1922             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1923             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
1924                 btrfs_warn(fs_info,
1925                 "remount supports changing free space tree only from ro to rw");
1926                 /* Make sure free space cache options match the state on disk */
1927                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1928                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1929                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1930                 }
1931                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1932                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1933                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1934                 }
1935         }
1936
1937         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1938                 goto out;
1939
1940         if (*flags & SB_RDONLY) {
1941                 /*
1942                  * this also happens on 'umount -rf' or on shutdown, when
1943                  * the filesystem is busy.
1944                  */
1945                 cancel_work_sync(&fs_info->async_reclaim_work);
1946                 cancel_work_sync(&fs_info->async_data_reclaim_work);
1947
1948                 btrfs_discard_cleanup(fs_info);
1949
1950                 /* wait for the uuid_scan task to finish */
1951                 down(&fs_info->uuid_tree_rescan_sem);
1952                 /* avoid complains from lockdep et al. */
1953                 up(&fs_info->uuid_tree_rescan_sem);
1954
1955                 btrfs_set_sb_rdonly(sb);
1956
1957                 /*
1958                  * Setting SB_RDONLY will put the cleaner thread to
1959                  * sleep at the next loop if it's already active.
1960                  * If it's already asleep, we'll leave unused block
1961                  * groups on disk until we're mounted read-write again
1962                  * unless we clean them up here.
1963                  */
1964                 btrfs_delete_unused_bgs(fs_info);
1965
1966                 /*
1967                  * The cleaner task could be already running before we set the
1968                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
1969                  * We must make sure that after we finish the remount, i.e. after
1970                  * we call btrfs_commit_super(), the cleaner can no longer start
1971                  * a transaction - either because it was dropping a dead root,
1972                  * running delayed iputs or deleting an unused block group (the
1973                  * cleaner picked a block group from the list of unused block
1974                  * groups before we were able to in the previous call to
1975                  * btrfs_delete_unused_bgs()).
1976                  */
1977                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
1978                             TASK_UNINTERRUPTIBLE);
1979
1980                 /*
1981                  * We've set the superblock to RO mode, so we might have made
1982                  * the cleaner task sleep without running all pending delayed
1983                  * iputs. Go through all the delayed iputs here, so that if an
1984                  * unmount happens without remounting RW we don't end up at
1985                  * finishing close_ctree() with a non-empty list of delayed
1986                  * iputs.
1987                  */
1988                 btrfs_run_delayed_iputs(fs_info);
1989
1990                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1991                 btrfs_scrub_cancel(fs_info);
1992                 btrfs_pause_balance(fs_info);
1993
1994                 /*
1995                  * Pause the qgroup rescan worker if it is running. We don't want
1996                  * it to be still running after we are in RO mode, as after that,
1997                  * by the time we unmount, it might have left a transaction open,
1998                  * so we would leak the transaction and/or crash.
1999                  */
2000                 btrfs_qgroup_wait_for_completion(fs_info, false);
2001
2002                 ret = btrfs_commit_super(fs_info);
2003                 if (ret)
2004                         goto restore;
2005         } else {
2006                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2007                         btrfs_err(fs_info,
2008                                 "Remounting read-write after error is not allowed");
2009                         ret = -EINVAL;
2010                         goto restore;
2011                 }
2012                 if (fs_info->fs_devices->rw_devices == 0) {
2013                         ret = -EACCES;
2014                         goto restore;
2015                 }
2016
2017                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2018                         btrfs_warn(fs_info,
2019                 "too many missing devices, writable remount is not allowed");
2020                         ret = -EACCES;
2021                         goto restore;
2022                 }
2023
2024                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2025                         btrfs_warn(fs_info,
2026                 "mount required to replay tree-log, cannot remount read-write");
2027                         ret = -EINVAL;
2028                         goto restore;
2029                 }
2030                 if (fs_info->sectorsize < PAGE_SIZE) {
2031                         btrfs_warn(fs_info,
2032         "read-write mount is not yet allowed for sectorsize %u page size %lu",
2033                                    fs_info->sectorsize, PAGE_SIZE);
2034                         ret = -EINVAL;
2035                         goto restore;
2036                 }
2037
2038                 /*
2039                  * NOTE: when remounting with a change that does writes, don't
2040                  * put it anywhere above this point, as we are not sure to be
2041                  * safe to write until we pass the above checks.
2042                  */
2043                 ret = btrfs_start_pre_rw_mount(fs_info);
2044                 if (ret)
2045                         goto restore;
2046
2047                 btrfs_clear_sb_rdonly(sb);
2048
2049                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2050         }
2051 out:
2052         /*
2053          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2054          * since the absence of the flag means it can be toggled off by remount.
2055          */
2056         *flags |= SB_I_VERSION;
2057
2058         wake_up_process(fs_info->transaction_kthread);
2059         btrfs_remount_cleanup(fs_info, old_opts);
2060         btrfs_clear_oneshot_options(fs_info);
2061         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2062
2063         return 0;
2064
2065 restore:
2066         /* We've hit an error - don't reset SB_RDONLY */
2067         if (sb_rdonly(sb))
2068                 old_flags |= SB_RDONLY;
2069         if (!(old_flags & SB_RDONLY))
2070                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2071         sb->s_flags = old_flags;
2072         fs_info->mount_opt = old_opts;
2073         fs_info->compress_type = old_compress_type;
2074         fs_info->max_inline = old_max_inline;
2075         btrfs_resize_thread_pool(fs_info,
2076                 old_thread_pool_size, fs_info->thread_pool_size);
2077         fs_info->metadata_ratio = old_metadata_ratio;
2078         btrfs_remount_cleanup(fs_info, old_opts);
2079         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2080
2081         return ret;
2082 }
2083
2084 /* Used to sort the devices by max_avail(descending sort) */
2085 static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
2086                                        const void *dev_info2)
2087 {
2088         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2089             ((struct btrfs_device_info *)dev_info2)->max_avail)
2090                 return -1;
2091         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2092                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2093                 return 1;
2094         else
2095         return 0;
2096 }
2097
2098 /*
2099  * sort the devices by max_avail, in which max free extent size of each device
2100  * is stored.(Descending Sort)
2101  */
2102 static inline void btrfs_descending_sort_devices(
2103                                         struct btrfs_device_info *devices,
2104                                         size_t nr_devices)
2105 {
2106         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2107              btrfs_cmp_device_free_bytes, NULL);
2108 }
2109
2110 /*
2111  * The helper to calc the free space on the devices that can be used to store
2112  * file data.
2113  */
2114 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2115                                               u64 *free_bytes)
2116 {
2117         struct btrfs_device_info *devices_info;
2118         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2119         struct btrfs_device *device;
2120         u64 type;
2121         u64 avail_space;
2122         u64 min_stripe_size;
2123         int num_stripes = 1;
2124         int i = 0, nr_devices;
2125         const struct btrfs_raid_attr *rattr;
2126
2127         /*
2128          * We aren't under the device list lock, so this is racy-ish, but good
2129          * enough for our purposes.
2130          */
2131         nr_devices = fs_info->fs_devices->open_devices;
2132         if (!nr_devices) {
2133                 smp_mb();
2134                 nr_devices = fs_info->fs_devices->open_devices;
2135                 ASSERT(nr_devices);
2136                 if (!nr_devices) {
2137                         *free_bytes = 0;
2138                         return 0;
2139                 }
2140         }
2141
2142         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2143                                GFP_KERNEL);
2144         if (!devices_info)
2145                 return -ENOMEM;
2146
2147         /* calc min stripe number for data space allocation */
2148         type = btrfs_data_alloc_profile(fs_info);
2149         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2150
2151         if (type & BTRFS_BLOCK_GROUP_RAID0)
2152                 num_stripes = nr_devices;
2153         else if (type & BTRFS_BLOCK_GROUP_RAID1)
2154                 num_stripes = 2;
2155         else if (type & BTRFS_BLOCK_GROUP_RAID1C3)
2156                 num_stripes = 3;
2157         else if (type & BTRFS_BLOCK_GROUP_RAID1C4)
2158                 num_stripes = 4;
2159         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2160                 num_stripes = 4;
2161
2162         /* Adjust for more than 1 stripe per device */
2163         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2164
2165         rcu_read_lock();
2166         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2167                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2168                                                 &device->dev_state) ||
2169                     !device->bdev ||
2170                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2171                         continue;
2172
2173                 if (i >= nr_devices)
2174                         break;
2175
2176                 avail_space = device->total_bytes - device->bytes_used;
2177
2178                 /* align with stripe_len */
2179                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2180
2181                 /*
2182                  * In order to avoid overwriting the superblock on the drive,
2183                  * btrfs starts at an offset of at least 1MB when doing chunk
2184                  * allocation.
2185                  *
2186                  * This ensures we have at least min_stripe_size free space
2187                  * after excluding 1MB.
2188                  */
2189                 if (avail_space <= SZ_1M + min_stripe_size)
2190                         continue;
2191
2192                 avail_space -= SZ_1M;
2193
2194                 devices_info[i].dev = device;
2195                 devices_info[i].max_avail = avail_space;
2196
2197                 i++;
2198         }
2199         rcu_read_unlock();
2200
2201         nr_devices = i;
2202
2203         btrfs_descending_sort_devices(devices_info, nr_devices);
2204
2205         i = nr_devices - 1;
2206         avail_space = 0;
2207         while (nr_devices >= rattr->devs_min) {
2208                 num_stripes = min(num_stripes, nr_devices);
2209
2210                 if (devices_info[i].max_avail >= min_stripe_size) {
2211                         int j;
2212                         u64 alloc_size;
2213
2214                         avail_space += devices_info[i].max_avail * num_stripes;
2215                         alloc_size = devices_info[i].max_avail;
2216                         for (j = i + 1 - num_stripes; j <= i; j++)
2217                                 devices_info[j].max_avail -= alloc_size;
2218                 }
2219                 i--;
2220                 nr_devices--;
2221         }
2222
2223         kfree(devices_info);
2224         *free_bytes = avail_space;
2225         return 0;
2226 }
2227
2228 /*
2229  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2230  *
2231  * If there's a redundant raid level at DATA block groups, use the respective
2232  * multiplier to scale the sizes.
2233  *
2234  * Unused device space usage is based on simulating the chunk allocator
2235  * algorithm that respects the device sizes and order of allocations.  This is
2236  * a close approximation of the actual use but there are other factors that may
2237  * change the result (like a new metadata chunk).
2238  *
2239  * If metadata is exhausted, f_bavail will be 0.
2240  */
2241 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2242 {
2243         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2244         struct btrfs_super_block *disk_super = fs_info->super_copy;
2245         struct btrfs_space_info *found;
2246         u64 total_used = 0;
2247         u64 total_free_data = 0;
2248         u64 total_free_meta = 0;
2249         u32 bits = fs_info->sectorsize_bits;
2250         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2251         unsigned factor = 1;
2252         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2253         int ret;
2254         u64 thresh = 0;
2255         int mixed = 0;
2256
2257         list_for_each_entry(found, &fs_info->space_info, list) {
2258                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2259                         int i;
2260
2261                         total_free_data += found->disk_total - found->disk_used;
2262                         total_free_data -=
2263                                 btrfs_account_ro_block_groups_free_space(found);
2264
2265                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2266                                 if (!list_empty(&found->block_groups[i]))
2267                                         factor = btrfs_bg_type_to_factor(
2268                                                 btrfs_raid_array[i].bg_flag);
2269                         }
2270                 }
2271
2272                 /*
2273                  * Metadata in mixed block goup profiles are accounted in data
2274                  */
2275                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2276                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2277                                 mixed = 1;
2278                         else
2279                                 total_free_meta += found->disk_total -
2280                                         found->disk_used;
2281                 }
2282
2283                 total_used += found->disk_used;
2284         }
2285
2286         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2287         buf->f_blocks >>= bits;
2288         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2289
2290         /* Account global block reserve as used, it's in logical size already */
2291         spin_lock(&block_rsv->lock);
2292         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2293         if (buf->f_bfree >= block_rsv->size >> bits)
2294                 buf->f_bfree -= block_rsv->size >> bits;
2295         else
2296                 buf->f_bfree = 0;
2297         spin_unlock(&block_rsv->lock);
2298
2299         buf->f_bavail = div_u64(total_free_data, factor);
2300         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2301         if (ret)
2302                 return ret;
2303         buf->f_bavail += div_u64(total_free_data, factor);
2304         buf->f_bavail = buf->f_bavail >> bits;
2305
2306         /*
2307          * We calculate the remaining metadata space minus global reserve. If
2308          * this is (supposedly) smaller than zero, there's no space. But this
2309          * does not hold in practice, the exhausted state happens where's still
2310          * some positive delta. So we apply some guesswork and compare the
2311          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2312          *
2313          * We probably cannot calculate the exact threshold value because this
2314          * depends on the internal reservations requested by various
2315          * operations, so some operations that consume a few metadata will
2316          * succeed even if the Avail is zero. But this is better than the other
2317          * way around.
2318          */
2319         thresh = SZ_4M;
2320
2321         /*
2322          * We only want to claim there's no available space if we can no longer
2323          * allocate chunks for our metadata profile and our global reserve will
2324          * not fit in the free metadata space.  If we aren't ->full then we
2325          * still can allocate chunks and thus are fine using the currently
2326          * calculated f_bavail.
2327          */
2328         if (!mixed && block_rsv->space_info->full &&
2329             total_free_meta - thresh < block_rsv->size)
2330                 buf->f_bavail = 0;
2331
2332         buf->f_type = BTRFS_SUPER_MAGIC;
2333         buf->f_bsize = dentry->d_sb->s_blocksize;
2334         buf->f_namelen = BTRFS_NAME_LEN;
2335
2336         /* We treat it as constant endianness (it doesn't matter _which_)
2337            because we want the fsid to come out the same whether mounted
2338            on a big-endian or little-endian host */
2339         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2340         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2341         /* Mask in the root object ID too, to disambiguate subvols */
2342         buf->f_fsid.val[0] ^=
2343                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2344         buf->f_fsid.val[1] ^=
2345                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2346
2347         return 0;
2348 }
2349
2350 static void btrfs_kill_super(struct super_block *sb)
2351 {
2352         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2353         kill_anon_super(sb);
2354         btrfs_free_fs_info(fs_info);
2355 }
2356
2357 static struct file_system_type btrfs_fs_type = {
2358         .owner          = THIS_MODULE,
2359         .name           = "btrfs",
2360         .mount          = btrfs_mount,
2361         .kill_sb        = btrfs_kill_super,
2362         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2363 };
2364
2365 static struct file_system_type btrfs_root_fs_type = {
2366         .owner          = THIS_MODULE,
2367         .name           = "btrfs",
2368         .mount          = btrfs_mount_root,
2369         .kill_sb        = btrfs_kill_super,
2370         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2371 };
2372
2373 MODULE_ALIAS_FS("btrfs");
2374
2375 static int btrfs_control_open(struct inode *inode, struct file *file)
2376 {
2377         /*
2378          * The control file's private_data is used to hold the
2379          * transaction when it is started and is used to keep
2380          * track of whether a transaction is already in progress.
2381          */
2382         file->private_data = NULL;
2383         return 0;
2384 }
2385
2386 /*
2387  * Used by /dev/btrfs-control for devices ioctls.
2388  */
2389 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2390                                 unsigned long arg)
2391 {
2392         struct btrfs_ioctl_vol_args *vol;
2393         struct btrfs_device *device = NULL;
2394         int ret = -ENOTTY;
2395
2396         if (!capable(CAP_SYS_ADMIN))
2397                 return -EPERM;
2398
2399         vol = memdup_user((void __user *)arg, sizeof(*vol));
2400         if (IS_ERR(vol))
2401                 return PTR_ERR(vol);
2402         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2403
2404         switch (cmd) {
2405         case BTRFS_IOC_SCAN_DEV:
2406                 mutex_lock(&uuid_mutex);
2407                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2408                                                &btrfs_root_fs_type);
2409                 ret = PTR_ERR_OR_ZERO(device);
2410                 mutex_unlock(&uuid_mutex);
2411                 break;
2412         case BTRFS_IOC_FORGET_DEV:
2413                 ret = btrfs_forget_devices(vol->name);
2414                 break;
2415         case BTRFS_IOC_DEVICES_READY:
2416                 mutex_lock(&uuid_mutex);
2417                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2418                                                &btrfs_root_fs_type);
2419                 if (IS_ERR(device)) {
2420                         mutex_unlock(&uuid_mutex);
2421                         ret = PTR_ERR(device);
2422                         break;
2423                 }
2424                 ret = !(device->fs_devices->num_devices ==
2425                         device->fs_devices->total_devices);
2426                 mutex_unlock(&uuid_mutex);
2427                 break;
2428         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2429                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2430                 break;
2431         }
2432
2433         kfree(vol);
2434         return ret;
2435 }
2436
2437 static int btrfs_freeze(struct super_block *sb)
2438 {
2439         struct btrfs_trans_handle *trans;
2440         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2441         struct btrfs_root *root = fs_info->tree_root;
2442
2443         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2444         /*
2445          * We don't need a barrier here, we'll wait for any transaction that
2446          * could be in progress on other threads (and do delayed iputs that
2447          * we want to avoid on a frozen filesystem), or do the commit
2448          * ourselves.
2449          */
2450         trans = btrfs_attach_transaction_barrier(root);
2451         if (IS_ERR(trans)) {
2452                 /* no transaction, don't bother */
2453                 if (PTR_ERR(trans) == -ENOENT)
2454                         return 0;
2455                 return PTR_ERR(trans);
2456         }
2457         return btrfs_commit_transaction(trans);
2458 }
2459
2460 static int btrfs_unfreeze(struct super_block *sb)
2461 {
2462         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2463
2464         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2465         return 0;
2466 }
2467
2468 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2469 {
2470         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2471         struct btrfs_device *dev, *first_dev = NULL;
2472
2473         /*
2474          * Lightweight locking of the devices. We should not need
2475          * device_list_mutex here as we only read the device data and the list
2476          * is protected by RCU.  Even if a device is deleted during the list
2477          * traversals, we'll get valid data, the freeing callback will wait at
2478          * least until the rcu_read_unlock.
2479          */
2480         rcu_read_lock();
2481         list_for_each_entry_rcu(dev, &fs_info->fs_devices->devices, dev_list) {
2482                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2483                         continue;
2484                 if (!dev->name)
2485                         continue;
2486                 if (!first_dev || dev->devid < first_dev->devid)
2487                         first_dev = dev;
2488         }
2489
2490         if (first_dev)
2491                 seq_escape(m, rcu_str_deref(first_dev->name), " \t\n\\");
2492         else
2493                 WARN_ON(1);
2494         rcu_read_unlock();
2495         return 0;
2496 }
2497
2498 static const struct super_operations btrfs_super_ops = {
2499         .drop_inode     = btrfs_drop_inode,
2500         .evict_inode    = btrfs_evict_inode,
2501         .put_super      = btrfs_put_super,
2502         .sync_fs        = btrfs_sync_fs,
2503         .show_options   = btrfs_show_options,
2504         .show_devname   = btrfs_show_devname,
2505         .alloc_inode    = btrfs_alloc_inode,
2506         .destroy_inode  = btrfs_destroy_inode,
2507         .free_inode     = btrfs_free_inode,
2508         .statfs         = btrfs_statfs,
2509         .remount_fs     = btrfs_remount,
2510         .freeze_fs      = btrfs_freeze,
2511         .unfreeze_fs    = btrfs_unfreeze,
2512 };
2513
2514 static const struct file_operations btrfs_ctl_fops = {
2515         .open = btrfs_control_open,
2516         .unlocked_ioctl  = btrfs_control_ioctl,
2517         .compat_ioctl = compat_ptr_ioctl,
2518         .owner   = THIS_MODULE,
2519         .llseek = noop_llseek,
2520 };
2521
2522 static struct miscdevice btrfs_misc = {
2523         .minor          = BTRFS_MINOR,
2524         .name           = "btrfs-control",
2525         .fops           = &btrfs_ctl_fops
2526 };
2527
2528 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2529 MODULE_ALIAS("devname:btrfs-control");
2530
2531 static int __init btrfs_interface_init(void)
2532 {
2533         return misc_register(&btrfs_misc);
2534 }
2535
2536 static __cold void btrfs_interface_exit(void)
2537 {
2538         misc_deregister(&btrfs_misc);
2539 }
2540
2541 static void __init btrfs_print_mod_info(void)
2542 {
2543         static const char options[] = ""
2544 #ifdef CONFIG_BTRFS_DEBUG
2545                         ", debug=on"
2546 #endif
2547 #ifdef CONFIG_BTRFS_ASSERT
2548                         ", assert=on"
2549 #endif
2550 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2551                         ", integrity-checker=on"
2552 #endif
2553 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2554                         ", ref-verify=on"
2555 #endif
2556 #ifdef CONFIG_BLK_DEV_ZONED
2557                         ", zoned=yes"
2558 #else
2559                         ", zoned=no"
2560 #endif
2561                         ;
2562         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2563 }
2564
2565 static int __init init_btrfs_fs(void)
2566 {
2567         int err;
2568
2569         btrfs_props_init();
2570
2571         err = btrfs_init_sysfs();
2572         if (err)
2573                 return err;
2574
2575         btrfs_init_compress();
2576
2577         err = btrfs_init_cachep();
2578         if (err)
2579                 goto free_compress;
2580
2581         err = extent_io_init();
2582         if (err)
2583                 goto free_cachep;
2584
2585         err = extent_state_cache_init();
2586         if (err)
2587                 goto free_extent_io;
2588
2589         err = extent_map_init();
2590         if (err)
2591                 goto free_extent_state_cache;
2592
2593         err = ordered_data_init();
2594         if (err)
2595                 goto free_extent_map;
2596
2597         err = btrfs_delayed_inode_init();
2598         if (err)
2599                 goto free_ordered_data;
2600
2601         err = btrfs_auto_defrag_init();
2602         if (err)
2603                 goto free_delayed_inode;
2604
2605         err = btrfs_delayed_ref_init();
2606         if (err)
2607                 goto free_auto_defrag;
2608
2609         err = btrfs_prelim_ref_init();
2610         if (err)
2611                 goto free_delayed_ref;
2612
2613         err = btrfs_end_io_wq_init();
2614         if (err)
2615                 goto free_prelim_ref;
2616
2617         err = btrfs_interface_init();
2618         if (err)
2619                 goto free_end_io_wq;
2620
2621         btrfs_print_mod_info();
2622
2623         err = btrfs_run_sanity_tests();
2624         if (err)
2625                 goto unregister_ioctl;
2626
2627         err = register_filesystem(&btrfs_fs_type);
2628         if (err)
2629                 goto unregister_ioctl;
2630
2631         return 0;
2632
2633 unregister_ioctl:
2634         btrfs_interface_exit();
2635 free_end_io_wq:
2636         btrfs_end_io_wq_exit();
2637 free_prelim_ref:
2638         btrfs_prelim_ref_exit();
2639 free_delayed_ref:
2640         btrfs_delayed_ref_exit();
2641 free_auto_defrag:
2642         btrfs_auto_defrag_exit();
2643 free_delayed_inode:
2644         btrfs_delayed_inode_exit();
2645 free_ordered_data:
2646         ordered_data_exit();
2647 free_extent_map:
2648         extent_map_exit();
2649 free_extent_state_cache:
2650         extent_state_cache_exit();
2651 free_extent_io:
2652         extent_io_exit();
2653 free_cachep:
2654         btrfs_destroy_cachep();
2655 free_compress:
2656         btrfs_exit_compress();
2657         btrfs_exit_sysfs();
2658
2659         return err;
2660 }
2661
2662 static void __exit exit_btrfs_fs(void)
2663 {
2664         btrfs_destroy_cachep();
2665         btrfs_delayed_ref_exit();
2666         btrfs_auto_defrag_exit();
2667         btrfs_delayed_inode_exit();
2668         btrfs_prelim_ref_exit();
2669         ordered_data_exit();
2670         extent_map_exit();
2671         extent_state_cache_exit();
2672         extent_io_exit();
2673         btrfs_interface_exit();
2674         btrfs_end_io_wq_exit();
2675         unregister_filesystem(&btrfs_fs_type);
2676         btrfs_exit_sysfs();
2677         btrfs_cleanup_fs_uuids();
2678         btrfs_exit_compress();
2679 }
2680
2681 late_initcall(init_btrfs_fs);
2682 module_exit(exit_btrfs_fs)
2683
2684 MODULE_LICENSE("GPL");
2685 MODULE_SOFTDEP("pre: crc32c");
2686 MODULE_SOFTDEP("pre: xxhash64");
2687 MODULE_SOFTDEP("pre: sha256");
2688 MODULE_SOFTDEP("pre: blake2b-256");