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 #if BITS_PER_LONG == 32
 256 void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info)
 257 {
 258         if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) {
 259                 btrfs_warn(fs_info, "reaching 32bit limit for logical addresses");
 260                 btrfs_warn(fs_info,
 261 "due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT",
 262                            BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 263                 btrfs_warn(fs_info,
 264                            "please consider upgrading to 64bit kernel/hardware");
 265         }
 266 }
 267
 268 void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info)
 269 {
 270         if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) {
 271                 btrfs_err(fs_info, "reached 32bit limit for logical addresses");
 272                 btrfs_err(fs_info,
 273 "due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed",
 274                           BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 275                 btrfs_err(fs_info,
 276                            "please consider upgrading to 64bit kernel/hardware");
 277         }
 278 }
 279 #endif
 280
 281 /*
 282  * We only mark the transaction aborted and then set the file system read-only.
 283  * This will prevent new transactions from starting or trying to join this
 284  * one.
 285  *
 286  * This means that error recovery at the call site is limited to freeing
 287  * any local memory allocations and passing the error code up without
 288  * further cleanup. The transaction should complete as it normally would
 289  * in the call path but will return -EIO.
 290  *
 291  * We'll complete the cleanup in btrfs_end_transaction and
 292  * btrfs_commit_transaction.
 293  */
 294 __cold
 295 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 296                                const char *function,
 297                                unsigned int line, int errno)
 298 {
 299         struct btrfs_fs_info *fs_info = trans->fs_info;
 300
 301         WRITE_ONCE(trans->aborted, errno);
 302         /* Nothing used. The other threads that have joined this
 303          * transaction may be able to continue. */
 304         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 305                 const char *errstr;
 306
 307                 errstr = btrfs_decode_error(errno);
 308                 btrfs_warn(fs_info,
 309                            "%s:%d: Aborting unused transaction(%s).",
 310                            function, line, errstr);
 311                 return;
 312         }
 313         WRITE_ONCE(trans->transaction->aborted, errno);
 314         /* Wake up anybody who may be waiting on this transaction */
 315         wake_up(&fs_info->transaction_wait);
 316         wake_up(&fs_info->transaction_blocked_wait);
 317         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 318 }
 319 /*
 320  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 321  * issues an alert, and either panics or BUGs, depending on mount options.
 322  */
 323 __cold
 324 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 325                    unsigned int line, int errno, const char *fmt, ...)
 326 {
 327         char *s_id = "<unknown>";
 328         const char *errstr;
 329         struct va_format vaf = { .fmt = fmt };
 330         va_list args;
 331
 332         if (fs_info)
 333                 s_id = fs_info->sb->s_id;
 334
 335         va_start(args, fmt);
 336         vaf.va = &args;
 337
 338         errstr = btrfs_decode_error(errno);
 339         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 340                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 341                         s_id, function, line, &vaf, errno, errstr);
 342
 343         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 344                    function, line, &vaf, errno, errstr);
 345         va_end(args);
 346         /* Caller calls BUG() */
 347 }
 348
 349 static void btrfs_put_super(struct super_block *sb)
 350 {
 351         close_ctree(btrfs_sb(sb));
 352 }
 353
 354 enum {
 355         Opt_acl, Opt_noacl,
 356         Opt_clear_cache,
 357         Opt_commit_interval,
 358         Opt_compress,
 359         Opt_compress_force,
 360         Opt_compress_force_type,
 361         Opt_compress_type,
 362         Opt_degraded,
 363         Opt_device,
 364         Opt_fatal_errors,
 365         Opt_flushoncommit, Opt_noflushoncommit,
 366         Opt_max_inline,
 367         Opt_barrier, Opt_nobarrier,
 368         Opt_datacow, Opt_nodatacow,
 369         Opt_datasum, Opt_nodatasum,
 370         Opt_defrag, Opt_nodefrag,
 371         Opt_discard, Opt_nodiscard,
 372         Opt_discard_mode,
 373         Opt_norecovery,
 374         Opt_ratio,
 375         Opt_rescan_uuid_tree,
 376         Opt_skip_balance,
 377         Opt_space_cache, Opt_no_space_cache,
 378         Opt_space_cache_version,
 379         Opt_ssd, Opt_nossd,
 380         Opt_ssd_spread, Opt_nossd_spread,
 381         Opt_subvol,
 382         Opt_subvol_empty,
 383         Opt_subvolid,
 384         Opt_thread_pool,
 385         Opt_treelog, Opt_notreelog,
 386         Opt_user_subvol_rm_allowed,
 387
 388         /* Rescue options */
 389         Opt_rescue,
 390         Opt_usebackuproot,
 391         Opt_nologreplay,
 392         Opt_ignorebadroots,
 393         Opt_ignoredatacsums,
 394         Opt_rescue_all,
 395
 396         /* Deprecated options */
 397         Opt_recovery,
 398         Opt_inode_cache, Opt_noinode_cache,
 399
 400         /* Debugging options */
 401         Opt_check_integrity,
 402         Opt_check_integrity_including_extent_data,
 403         Opt_check_integrity_print_mask,
 404         Opt_enospc_debug, Opt_noenospc_debug,
 405 #ifdef CONFIG_BTRFS_DEBUG
 406         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 407 #endif
 408 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 409         Opt_ref_verify,
 410 #endif
 411         Opt_err,
 412 };
 413
 414 static const match_table_t tokens = {
 415         {Opt_acl, "acl"},
 416         {Opt_noacl, "noacl"},
 417         {Opt_clear_cache, "clear_cache"},
 418         {Opt_commit_interval, "commit=%u"},
 419         {Opt_compress, "compress"},
 420         {Opt_compress_type, "compress=%s"},
 421         {Opt_compress_force, "compress-force"},
 422         {Opt_compress_force_type, "compress-force=%s"},
 423         {Opt_degraded, "degraded"},
 424         {Opt_device, "device=%s"},
 425         {Opt_fatal_errors, "fatal_errors=%s"},
 426         {Opt_flushoncommit, "flushoncommit"},
 427         {Opt_noflushoncommit, "noflushoncommit"},
 428         {Opt_inode_cache, "inode_cache"},
 429         {Opt_noinode_cache, "noinode_cache"},
 430         {Opt_max_inline, "max_inline=%s"},
 431         {Opt_barrier, "barrier"},
 432         {Opt_nobarrier, "nobarrier"},
 433         {Opt_datacow, "datacow"},
 434         {Opt_nodatacow, "nodatacow"},
 435         {Opt_datasum, "datasum"},
 436         {Opt_nodatasum, "nodatasum"},
 437         {Opt_defrag, "autodefrag"},
 438         {Opt_nodefrag, "noautodefrag"},
 439         {Opt_discard, "discard"},
 440         {Opt_discard_mode, "discard=%s"},
 441         {Opt_nodiscard, "nodiscard"},
 442         {Opt_norecovery, "norecovery"},
 443         {Opt_ratio, "metadata_ratio=%u"},
 444         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 445         {Opt_skip_balance, "skip_balance"},
 446         {Opt_space_cache, "space_cache"},
 447         {Opt_no_space_cache, "nospace_cache"},
 448         {Opt_space_cache_version, "space_cache=%s"},
 449         {Opt_ssd, "ssd"},
 450         {Opt_nossd, "nossd"},
 451         {Opt_ssd_spread, "ssd_spread"},
 452         {Opt_nossd_spread, "nossd_spread"},
 453         {Opt_subvol, "subvol=%s"},
 454         {Opt_subvol_empty, "subvol="},
 455         {Opt_subvolid, "subvolid=%s"},
 456         {Opt_thread_pool, "thread_pool=%u"},
 457         {Opt_treelog, "treelog"},
 458         {Opt_notreelog, "notreelog"},
 459         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 460
 461         /* Rescue options */
 462         {Opt_rescue, "rescue=%s"},
 463         /* Deprecated, with alias rescue=nologreplay */
 464         {Opt_nologreplay, "nologreplay"},
 465         /* Deprecated, with alias rescue=usebackuproot */
 466         {Opt_usebackuproot, "usebackuproot"},
 467
 468         /* Deprecated options */
 469         {Opt_recovery, "recovery"},
 470
 471         /* Debugging options */
 472         {Opt_check_integrity, "check_int"},
 473         {Opt_check_integrity_including_extent_data, "check_int_data"},
 474         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 475         {Opt_enospc_debug, "enospc_debug"},
 476         {Opt_noenospc_debug, "noenospc_debug"},
 477 #ifdef CONFIG_BTRFS_DEBUG
 478         {Opt_fragment_data, "fragment=data"},
 479         {Opt_fragment_metadata, "fragment=metadata"},
 480         {Opt_fragment_all, "fragment=all"},
 481 #endif
 482 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 483         {Opt_ref_verify, "ref_verify"},
 484 #endif
 485         {Opt_err, NULL},
 486 };
 487
 488 static const match_table_t rescue_tokens = {
 489         {Opt_usebackuproot, "usebackuproot"},
 490         {Opt_nologreplay, "nologreplay"},
 491         {Opt_ignorebadroots, "ignorebadroots"},
 492         {Opt_ignorebadroots, "ibadroots"},
 493         {Opt_ignoredatacsums, "ignoredatacsums"},
 494         {Opt_ignoredatacsums, "idatacsums"},
 495         {Opt_rescue_all, "all"},
 496         {Opt_err, NULL},
 497 };
 498
 499 static bool check_ro_option(struct btrfs_fs_info *fs_info, unsigned long opt,
 500                             const char *opt_name)
 501 {
 502         if (fs_info->mount_opt & opt) {
 503                 btrfs_err(fs_info, "%s must be used with ro mount option",
 504                           opt_name);
 505                 return true;
 506         }
 507         return false;
 508 }
 509
 510 static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
 511 {
 512         char *opts;
 513         char *orig;
 514         char *p;
 515         substring_t args[MAX_OPT_ARGS];
 516         int ret = 0;
 517
 518         opts = kstrdup(options, GFP_KERNEL);
 519         if (!opts)
 520                 return -ENOMEM;
 521         orig = opts;
 522
 523         while ((p = strsep(&opts, ":")) != NULL) {
 524                 int token;
 525
 526                 if (!*p)
 527                         continue;
 528                 token = match_token(p, rescue_tokens, args);
 529                 switch (token){
 530                 case Opt_usebackuproot:
 531                         btrfs_info(info,
 532                                    "trying to use backup root at mount time");
 533                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 534                         break;
 535                 case Opt_nologreplay:
 536                         btrfs_set_and_info(info, NOLOGREPLAY,
 537                                            "disabling log replay at mount time");
 538                         break;
 539                 case Opt_ignorebadroots:
 540                         btrfs_set_and_info(info, IGNOREBADROOTS,
 541                                            "ignoring bad roots");
 542                         break;
 543                 case Opt_ignoredatacsums:
 544                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 545                                            "ignoring data csums");
 546                         break;
 547                 case Opt_rescue_all:
 548                         btrfs_info(info, "enabling all of the rescue options");
 549                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 550                                            "ignoring data csums");
 551                         btrfs_set_and_info(info, IGNOREBADROOTS,
 552                                            "ignoring bad roots");
 553                         btrfs_set_and_info(info, NOLOGREPLAY,
 554                                            "disabling log replay at mount time");
 555                         break;
 556                 case Opt_err:
 557                         btrfs_info(info, "unrecognized rescue option '%s'", p);
 558                         ret = -EINVAL;
 559                         goto out;
 560                 default:
 561                         break;
 562                 }
 563
 564         }
 565 out:
 566         kfree(orig);
 567         return ret;
 568 }
 569
 570 /*
 571  * Regular mount options parser.  Everything that is needed only when
 572  * reading in a new superblock is parsed here.
 573  * XXX JDM: This needs to be cleaned up for remount.
 574  */
 575 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 576                         unsigned long new_flags)
 577 {
 578         substring_t args[MAX_OPT_ARGS];
 579         char *p, *num;
 580         int intarg;
 581         int ret = 0;
 582         char *compress_type;
 583         bool compress_force = false;
 584         enum btrfs_compression_type saved_compress_type;
 585         int saved_compress_level;
 586         bool saved_compress_force;
 587         int no_compress = 0;
 588
 589         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 590                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 591         else if (btrfs_free_space_cache_v1_active(info)) {
 592                 if (btrfs_is_zoned(info)) {
 593                         btrfs_info(info,
 594                         "zoned: clearing existing space cache");
 595                         btrfs_set_super_cache_generation(info->super_copy, 0);
 596                 } else {
 597                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 598                 }
 599         }
 600
 601         /*
 602          * Even the options are empty, we still need to do extra check
 603          * against new flags
 604          */
 605         if (!options)
 606                 goto check;
 607
 608         while ((p = strsep(&options, ",")) != NULL) {
 609                 int token;
 610                 if (!*p)
 611                         continue;
 612
 613                 token = match_token(p, tokens, args);
 614                 switch (token) {
 615                 case Opt_degraded:
 616                         btrfs_info(info, "allowing degraded mounts");
 617                         btrfs_set_opt(info->mount_opt, DEGRADED);
 618                         break;
 619                 case Opt_subvol:
 620                 case Opt_subvol_empty:
 621                 case Opt_subvolid:
 622                 case Opt_device:
 623                         /*
 624                          * These are parsed by btrfs_parse_subvol_options or
 625                          * btrfs_parse_device_options and can be ignored here.
 626                          */
 627                         break;
 628                 case Opt_nodatasum:
 629                         btrfs_set_and_info(info, NODATASUM,
 630                                            "setting nodatasum");
 631                         break;
 632                 case Opt_datasum:
 633                         if (btrfs_test_opt(info, NODATASUM)) {
 634                                 if (btrfs_test_opt(info, NODATACOW))
 635                                         btrfs_info(info,
 636                                                    "setting datasum, datacow enabled");
 637                                 else
 638                                         btrfs_info(info, "setting datasum");
 639                         }
 640                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 641                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 642                         break;
 643                 case Opt_nodatacow:
 644                         if (!btrfs_test_opt(info, NODATACOW)) {
 645                                 if (!btrfs_test_opt(info, COMPRESS) ||
 646                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 647                                         btrfs_info(info,
 648                                                    "setting nodatacow, compression disabled");
 649                                 } else {
 650                                         btrfs_info(info, "setting nodatacow");
 651                                 }
 652                         }
 653                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 654                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 655                         btrfs_set_opt(info->mount_opt, NODATACOW);
 656                         btrfs_set_opt(info->mount_opt, NODATASUM);
 657                         break;
 658                 case Opt_datacow:
 659                         btrfs_clear_and_info(info, NODATACOW,
 660                                              "setting datacow");
 661                         break;
 662                 case Opt_compress_force:
 663                 case Opt_compress_force_type:
 664                         compress_force = true;
 665                         fallthrough;
 666                 case Opt_compress:
 667                 case Opt_compress_type:
 668                         saved_compress_type = btrfs_test_opt(info,
 669                                                              COMPRESS) ?
 670                                 info->compress_type : BTRFS_COMPRESS_NONE;
 671                         saved_compress_force =
 672                                 btrfs_test_opt(info, FORCE_COMPRESS);
 673                         saved_compress_level = info->compress_level;
 674                         if (token == Opt_compress ||
 675                             token == Opt_compress_force ||
 676                             strncmp(args[0].from, "zlib", 4) == 0) {
 677                                 compress_type = "zlib";
 678
 679                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 680                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 681                                 /*
 682                                  * args[0] contains uninitialized data since
 683                                  * for these tokens we don't expect any
 684                                  * parameter.
 685                                  */
 686                                 if (token != Opt_compress &&
 687                                     token != Opt_compress_force)
 688                                         info->compress_level =
 689                                           btrfs_compress_str2level(
 690                                                         BTRFS_COMPRESS_ZLIB,
 691                                                         args[0].from + 4);
 692                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 693                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 694                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 695                                 no_compress = 0;
 696                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 697                                 compress_type = "lzo";
 698                                 info->compress_type = BTRFS_COMPRESS_LZO;
 699                                 info->compress_level = 0;
 700                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 701                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 702                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 703                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 704                                 no_compress = 0;
 705                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 706                                 compress_type = "zstd";
 707                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 708                                 info->compress_level =
 709                                         btrfs_compress_str2level(
 710                                                          BTRFS_COMPRESS_ZSTD,
 711                                                          args[0].from + 4);
 712                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 713                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 714                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 715                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 716                                 no_compress = 0;
 717                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 718                                 compress_type = "no";
 719                                 info->compress_level = 0;
 720                                 info->compress_type = 0;
 721                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 722                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 723                                 compress_force = false;
 724                                 no_compress++;
 725                         } else {
 726                                 ret = -EINVAL;
 727                                 goto out;
 728                         }
 729
 730                         if (compress_force) {
 731                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 732                         } else {
 733                                 /*
 734                                  * If we remount from compress-force=xxx to
 735                                  * compress=xxx, we need clear FORCE_COMPRESS
 736                                  * flag, otherwise, there is no way for users
 737                                  * to disable forcible compression separately.
 738                                  */
 739                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 740                         }
 741                         if (no_compress == 1) {
 742                                 btrfs_info(info, "use no compression");
 743                         } else if ((info->compress_type != saved_compress_type) ||
 744                                    (compress_force != saved_compress_force) ||
 745                                    (info->compress_level != saved_compress_level)) {
 746                                 btrfs_info(info, "%s %s compression, level %d",
 747                                            (compress_force) ? "force" : "use",
 748                                            compress_type, info->compress_level);
 749                         }
 750                         compress_force = false;
 751                         break;
 752                 case Opt_ssd:
 753                         btrfs_set_and_info(info, SSD,
 754                                            "enabling ssd optimizations");
 755                         btrfs_clear_opt(info->mount_opt, NOSSD);
 756                         break;
 757                 case Opt_ssd_spread:
 758                         btrfs_set_and_info(info, SSD,
 759                                            "enabling ssd optimizations");
 760                         btrfs_set_and_info(info, SSD_SPREAD,
 761                                            "using spread ssd allocation scheme");
 762                         btrfs_clear_opt(info->mount_opt, NOSSD);
 763                         break;
 764                 case Opt_nossd:
 765                         btrfs_set_opt(info->mount_opt, NOSSD);
 766                         btrfs_clear_and_info(info, SSD,
 767                                              "not using ssd optimizations");
 768                         fallthrough;
 769                 case Opt_nossd_spread:
 770                         btrfs_clear_and_info(info, SSD_SPREAD,
 771                                              "not using spread ssd allocation scheme");
 772                         break;
 773                 case Opt_barrier:
 774                         btrfs_clear_and_info(info, NOBARRIER,
 775                                              "turning on barriers");
 776                         break;
 777                 case Opt_nobarrier:
 778                         btrfs_set_and_info(info, NOBARRIER,
 779                                            "turning off barriers");
 780                         break;
 781                 case Opt_thread_pool:
 782                         ret = match_int(&args[0], &intarg);
 783                         if (ret) {
 784                                 goto out;
 785                         } else if (intarg == 0) {
 786                                 ret = -EINVAL;
 787                                 goto out;
 788                         }
 789                         info->thread_pool_size = intarg;
 790                         break;
 791                 case Opt_max_inline:
 792                         num = match_strdup(&args[0]);
 793                         if (num) {
 794                                 info->max_inline = memparse(num, NULL);
 795                                 kfree(num);
 796
 797                                 if (info->max_inline) {
 798                                         info->max_inline = min_t(u64,
 799                                                 info->max_inline,
 800                                                 info->sectorsize);
 801                                 }
 802                                 btrfs_info(info, "max_inline at %llu",
 803                                            info->max_inline);
 804                         } else {
 805                                 ret = -ENOMEM;
 806                                 goto out;
 807                         }
 808                         break;
 809                 case Opt_acl:
 810 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 811                         info->sb->s_flags |= SB_POSIXACL;
 812                         break;
 813 #else
 814                         btrfs_err(info, "support for ACL not compiled in!");
 815                         ret = -EINVAL;
 816                         goto out;
 817 #endif
 818                 case Opt_noacl:
 819                         info->sb->s_flags &= ~SB_POSIXACL;
 820                         break;
 821                 case Opt_notreelog:
 822                         btrfs_set_and_info(info, NOTREELOG,
 823                                            "disabling tree log");
 824                         break;
 825                 case Opt_treelog:
 826                         btrfs_clear_and_info(info, NOTREELOG,
 827                                              "enabling tree log");
 828                         break;
 829                 case Opt_norecovery:
 830                 case Opt_nologreplay:
 831                         btrfs_warn(info,
 832                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 833                         btrfs_set_and_info(info, NOLOGREPLAY,
 834                                            "disabling log replay at mount time");
 835                         break;
 836                 case Opt_flushoncommit:
 837                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 838                                            "turning on flush-on-commit");
 839                         break;
 840                 case Opt_noflushoncommit:
 841                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 842                                              "turning off flush-on-commit");
 843                         break;
 844                 case Opt_ratio:
 845                         ret = match_int(&args[0], &intarg);
 846                         if (ret)
 847                                 goto out;
 848                         info->metadata_ratio = intarg;
 849                         btrfs_info(info, "metadata ratio %u",
 850                                    info->metadata_ratio);
 851                         break;
 852                 case Opt_discard:
 853                 case Opt_discard_mode:
 854                         if (token == Opt_discard ||
 855                             strcmp(args[0].from, "sync") == 0) {
 856                                 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
 857                                 btrfs_set_and_info(info, DISCARD_SYNC,
 858                                                    "turning on sync discard");
 859                         } else if (strcmp(args[0].from, "async") == 0) {
 860                                 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
 861                                 btrfs_set_and_info(info, DISCARD_ASYNC,
 862                                                    "turning on async discard");
 863                         } else {
 864                                 ret = -EINVAL;
 865                                 goto out;
 866                         }
 867                         break;
 868                 case Opt_nodiscard:
 869                         btrfs_clear_and_info(info, DISCARD_SYNC,
 870                                              "turning off discard");
 871                         btrfs_clear_and_info(info, DISCARD_ASYNC,
 872                                              "turning off async discard");
 873                         break;
 874                 case Opt_space_cache:
 875                 case Opt_space_cache_version:
 876                         if (token == Opt_space_cache ||
 877                             strcmp(args[0].from, "v1") == 0) {
 878                                 btrfs_clear_opt(info->mount_opt,
 879                                                 FREE_SPACE_TREE);
 880                                 btrfs_set_and_info(info, SPACE_CACHE,
 881                                            "enabling disk space caching");
 882                         } else if (strcmp(args[0].from, "v2") == 0) {
 883                                 btrfs_clear_opt(info->mount_opt,
 884                                                 SPACE_CACHE);
 885                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 886                                                    "enabling free space tree");
 887                         } else {
 888                                 ret = -EINVAL;
 889                                 goto out;
 890                         }
 891                         break;
 892                 case Opt_rescan_uuid_tree:
 893                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 894                         break;
 895                 case Opt_no_space_cache:
 896                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 897                                 btrfs_clear_and_info(info, SPACE_CACHE,
 898                                              "disabling disk space caching");
 899                         }
 900                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 901                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 902                                              "disabling free space tree");
 903                         }
 904                         break;
 905                 case Opt_inode_cache:
 906                 case Opt_noinode_cache:
 907                         btrfs_warn(info,
 908         "the 'inode_cache' option is deprecated and has no effect since 5.11");
 909                         break;
 910                 case Opt_clear_cache:
 911                         btrfs_set_and_info(info, CLEAR_CACHE,
 912                                            "force clearing of disk cache");
 913                         break;
 914                 case Opt_user_subvol_rm_allowed:
 915                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 916                         break;
 917                 case Opt_enospc_debug:
 918                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 919                         break;
 920                 case Opt_noenospc_debug:
 921                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 922                         break;
 923                 case Opt_defrag:
 924                         btrfs_set_and_info(info, AUTO_DEFRAG,
 925                                            "enabling auto defrag");
 926                         break;
 927                 case Opt_nodefrag:
 928                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 929                                              "disabling auto defrag");
 930                         break;
 931                 case Opt_recovery:
 932                 case Opt_usebackuproot:
 933                         btrfs_warn(info,
 934                         "'%s' is deprecated, use 'rescue=usebackuproot' instead",
 935                                    token == Opt_recovery ? "recovery" :
 936                                    "usebackuproot");
 937                         btrfs_info(info,
 938                                    "trying to use backup root at mount time");
 939                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 940                         break;
 941                 case Opt_skip_balance:
 942                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 943                         break;
 944 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 945                 case Opt_check_integrity_including_extent_data:
 946                         btrfs_info(info,
 947                                    "enabling check integrity including extent data");
 948                         btrfs_set_opt(info->mount_opt,
 949                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 950                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 951                         break;
 952                 case Opt_check_integrity:
 953                         btrfs_info(info, "enabling check integrity");
 954                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 955                         break;
 956                 case Opt_check_integrity_print_mask:
 957                         ret = match_int(&args[0], &intarg);
 958                         if (ret)
 959                                 goto out;
 960                         info->check_integrity_print_mask = intarg;
 961                         btrfs_info(info, "check_integrity_print_mask 0x%x",
 962                                    info->check_integrity_print_mask);
 963                         break;
 964 #else
 965                 case Opt_check_integrity_including_extent_data:
 966                 case Opt_check_integrity:
 967                 case Opt_check_integrity_print_mask:
 968                         btrfs_err(info,
 969                                   "support for check_integrity* not compiled in!");
 970                         ret = -EINVAL;
 971                         goto out;
 972 #endif
 973                 case Opt_fatal_errors:
 974                         if (strcmp(args[0].from, "panic") == 0)
 975                                 btrfs_set_opt(info->mount_opt,
 976                                               PANIC_ON_FATAL_ERROR);
 977                         else if (strcmp(args[0].from, "bug") == 0)
 978                                 btrfs_clear_opt(info->mount_opt,
 979                                               PANIC_ON_FATAL_ERROR);
 980                         else {
 981                                 ret = -EINVAL;
 982                                 goto out;
 983                         }
 984                         break;
 985                 case Opt_commit_interval:
 986                         intarg = 0;
 987                         ret = match_int(&args[0], &intarg);
 988                         if (ret)
 989                                 goto out;
 990                         if (intarg == 0) {
 991                                 btrfs_info(info,
 992                                            "using default commit interval %us",
 993                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 994                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
 995                         } else if (intarg > 300) {
 996                                 btrfs_warn(info, "excessive commit interval %d",
 997                                            intarg);
 998                         }
 999                         info->commit_interval = intarg;
1000                         break;
1001                 case Opt_rescue:
1002                         ret = parse_rescue_options(info, args[0].from);
1003                         if (ret < 0)
1004                                 goto out;
1005                         break;
1006 #ifdef CONFIG_BTRFS_DEBUG
1007                 case Opt_fragment_all:
1008                         btrfs_info(info, "fragmenting all space");
1009                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1010                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
1011                         break;
1012                 case Opt_fragment_metadata:
1013                         btrfs_info(info, "fragmenting metadata");
1014                         btrfs_set_opt(info->mount_opt,
1015                                       FRAGMENT_METADATA);
1016                         break;
1017                 case Opt_fragment_data:
1018                         btrfs_info(info, "fragmenting data");
1019                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1020                         break;
1021 #endif
1022 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
1023                 case Opt_ref_verify:
1024                         btrfs_info(info, "doing ref verification");
1025                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
1026                         break;
1027 #endif
1028                 case Opt_err:
1029                         btrfs_err(info, "unrecognized mount option '%s'", p);
1030                         ret = -EINVAL;
1031                         goto out;
1032                 default:
1033                         break;
1034                 }
1035         }
1036 check:
1037         /* We're read-only, don't have to check. */
1038         if (new_flags & SB_RDONLY)
1039                 goto out;
1040
1041         if (check_ro_option(info, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
1042             check_ro_option(info, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
1043             check_ro_option(info, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums"))
1044                 ret = -EINVAL;
1045 out:
1046         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
1047             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
1048             !btrfs_test_opt(info, CLEAR_CACHE)) {
1049                 btrfs_err(info, "cannot disable free space tree");
1050                 ret = -EINVAL;
1051
1052         }
1053         if (!ret)
1054                 ret = btrfs_check_mountopts_zoned(info);
1055         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
1056                 btrfs_info(info, "disk space caching is enabled");
1057         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
1058                 btrfs_info(info, "using free space tree");
1059         return ret;
1060 }
1061
1062 /*
1063  * Parse mount options that are required early in the mount process.
1064  *
1065  * All other options will be parsed on much later in the mount process and
1066  * only when we need to allocate a new super block.
1067  */
1068 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1069                                       void *holder)
1070 {
1071         substring_t args[MAX_OPT_ARGS];
1072         char *device_name, *opts, *orig, *p;
1073         struct btrfs_device *device = NULL;
1074         int error = 0;
1075
1076         lockdep_assert_held(&uuid_mutex);
1077
1078         if (!options)
1079                 return 0;
1080
1081         /*
1082          * strsep changes the string, duplicate it because btrfs_parse_options
1083          * gets called later
1084          */
1085         opts = kstrdup(options, GFP_KERNEL);
1086         if (!opts)
1087                 return -ENOMEM;
1088         orig = opts;
1089
1090         while ((p = strsep(&opts, ",")) != NULL) {
1091                 int token;
1092
1093                 if (!*p)
1094                         continue;
1095
1096                 token = match_token(p, tokens, args);
1097                 if (token == Opt_device) {
1098                         device_name = match_strdup(&args[0]);
1099                         if (!device_name) {
1100                                 error = -ENOMEM;
1101                                 goto out;
1102                         }
1103                         device = btrfs_scan_one_device(device_name, flags,
1104                                         holder);
1105                         kfree(device_name);
1106                         if (IS_ERR(device)) {
1107                                 error = PTR_ERR(device);
1108                                 goto out;
1109                         }
1110                 }
1111         }
1112
1113 out:
1114         kfree(orig);
1115         return error;
1116 }
1117
1118 /*
1119  * Parse mount options that are related to subvolume id
1120  *
1121  * The value is later passed to mount_subvol()
1122  */
1123 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1124                 u64 *subvol_objectid)
1125 {
1126         substring_t args[MAX_OPT_ARGS];
1127         char *opts, *orig, *p;
1128         int error = 0;
1129         u64 subvolid;
1130
1131         if (!options)
1132                 return 0;
1133
1134         /*
1135          * strsep changes the string, duplicate it because
1136          * btrfs_parse_device_options gets called later
1137          */
1138         opts = kstrdup(options, GFP_KERNEL);
1139         if (!opts)
1140                 return -ENOMEM;
1141         orig = opts;
1142
1143         while ((p = strsep(&opts, ",")) != NULL) {
1144                 int token;
1145                 if (!*p)
1146                         continue;
1147
1148                 token = match_token(p, tokens, args);
1149                 switch (token) {
1150                 case Opt_subvol:
1151                         kfree(*subvol_name);
1152                         *subvol_name = match_strdup(&args[0]);
1153                         if (!*subvol_name) {
1154                                 error = -ENOMEM;
1155                                 goto out;
1156                         }
1157                         break;
1158                 case Opt_subvolid:
1159                         error = match_u64(&args[0], &subvolid);
1160                         if (error)
1161                                 goto out;
1162
1163                         /* we want the original fs_tree */
1164                         if (subvolid == 0)
1165                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1166
1167                         *subvol_objectid = subvolid;
1168                         break;
1169                 default:
1170                         break;
1171                 }
1172         }
1173
1174 out:
1175         kfree(orig);
1176         return error;
1177 }
1178
1179 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1180                                           u64 subvol_objectid)
1181 {
1182         struct btrfs_root *root = fs_info->tree_root;
1183         struct btrfs_root *fs_root = NULL;
1184         struct btrfs_root_ref *root_ref;
1185         struct btrfs_inode_ref *inode_ref;
1186         struct btrfs_key key;
1187         struct btrfs_path *path = NULL;
1188         char *name = NULL, *ptr;
1189         u64 dirid;
1190         int len;
1191         int ret;
1192
1193         path = btrfs_alloc_path();
1194         if (!path) {
1195                 ret = -ENOMEM;
1196                 goto err;
1197         }
1198
1199         name = kmalloc(PATH_MAX, GFP_KERNEL);
1200         if (!name) {
1201                 ret = -ENOMEM;
1202                 goto err;
1203         }
1204         ptr = name + PATH_MAX - 1;
1205         ptr[0] = '\0';
1206
1207         /*
1208          * Walk up the subvolume trees in the tree of tree roots by root
1209          * backrefs until we hit the top-level subvolume.
1210          */
1211         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1212                 key.objectid = subvol_objectid;
1213                 key.type = BTRFS_ROOT_BACKREF_KEY;
1214                 key.offset = (u64)-1;
1215
1216                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1217                 if (ret < 0) {
1218                         goto err;
1219                 } else if (ret > 0) {
1220                         ret = btrfs_previous_item(root, path, subvol_objectid,
1221                                                   BTRFS_ROOT_BACKREF_KEY);
1222                         if (ret < 0) {
1223                                 goto err;
1224                         } else if (ret > 0) {
1225                                 ret = -ENOENT;
1226                                 goto err;
1227                         }
1228                 }
1229
1230                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1231                 subvol_objectid = key.offset;
1232
1233                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1234                                           struct btrfs_root_ref);
1235                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1236                 ptr -= len + 1;
1237                 if (ptr < name) {
1238                         ret = -ENAMETOOLONG;
1239                         goto err;
1240                 }
1241                 read_extent_buffer(path->nodes[0], ptr + 1,
1242                                    (unsigned long)(root_ref + 1), len);
1243                 ptr[0] = '/';
1244                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1245                 btrfs_release_path(path);
1246
1247                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1248                 if (IS_ERR(fs_root)) {
1249                         ret = PTR_ERR(fs_root);
1250                         fs_root = NULL;
1251                         goto err;
1252                 }
1253
1254                 /*
1255                  * Walk up the filesystem tree by inode refs until we hit the
1256                  * root directory.
1257                  */
1258                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1259                         key.objectid = dirid;
1260                         key.type = BTRFS_INODE_REF_KEY;
1261                         key.offset = (u64)-1;
1262
1263                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1264                         if (ret < 0) {
1265                                 goto err;
1266                         } else if (ret > 0) {
1267                                 ret = btrfs_previous_item(fs_root, path, dirid,
1268                                                           BTRFS_INODE_REF_KEY);
1269                                 if (ret < 0) {
1270                                         goto err;
1271                                 } else if (ret > 0) {
1272                                         ret = -ENOENT;
1273                                         goto err;
1274                                 }
1275                         }
1276
1277                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1278                         dirid = key.offset;
1279
1280                         inode_ref = btrfs_item_ptr(path->nodes[0],
1281                                                    path->slots[0],
1282                                                    struct btrfs_inode_ref);
1283                         len = btrfs_inode_ref_name_len(path->nodes[0],
1284                                                        inode_ref);
1285                         ptr -= len + 1;
1286                         if (ptr < name) {
1287                                 ret = -ENAMETOOLONG;
1288                                 goto err;
1289                         }
1290                         read_extent_buffer(path->nodes[0], ptr + 1,
1291                                            (unsigned long)(inode_ref + 1), len);
1292                         ptr[0] = '/';
1293                         btrfs_release_path(path);
1294                 }
1295                 btrfs_put_root(fs_root);
1296                 fs_root = NULL;
1297         }
1298
1299         btrfs_free_path(path);
1300         if (ptr == name + PATH_MAX - 1) {
1301                 name[0] = '/';
1302                 name[1] = '\0';
1303         } else {
1304                 memmove(name, ptr, name + PATH_MAX - ptr);
1305         }
1306         return name;
1307
1308 err:
1309         btrfs_put_root(fs_root);
1310         btrfs_free_path(path);
1311         kfree(name);
1312         return ERR_PTR(ret);
1313 }
1314
1315 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1316 {
1317         struct btrfs_root *root = fs_info->tree_root;
1318         struct btrfs_dir_item *di;
1319         struct btrfs_path *path;
1320         struct btrfs_key location;
1321         u64 dir_id;
1322
1323         path = btrfs_alloc_path();
1324         if (!path)
1325                 return -ENOMEM;
1326
1327         /*
1328          * Find the "default" dir item which points to the root item that we
1329          * will mount by default if we haven't been given a specific subvolume
1330          * to mount.
1331          */
1332         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1333         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1334         if (IS_ERR(di)) {
1335                 btrfs_free_path(path);
1336                 return PTR_ERR(di);
1337         }
1338         if (!di) {
1339                 /*
1340                  * Ok the default dir item isn't there.  This is weird since
1341                  * it's always been there, but don't freak out, just try and
1342                  * mount the top-level subvolume.
1343                  */
1344                 btrfs_free_path(path);
1345                 *objectid = BTRFS_FS_TREE_OBJECTID;
1346                 return 0;
1347         }
1348
1349         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1350         btrfs_free_path(path);
1351         *objectid = location.objectid;
1352         return 0;
1353 }
1354
1355 static int btrfs_fill_super(struct super_block *sb,
1356                             struct btrfs_fs_devices *fs_devices,
1357                             void *data)
1358 {
1359         struct inode *inode;
1360         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1361         int err;
1362
1363         sb->s_maxbytes = MAX_LFS_FILESIZE;
1364         sb->s_magic = BTRFS_SUPER_MAGIC;
1365         sb->s_op = &btrfs_super_ops;
1366         sb->s_d_op = &btrfs_dentry_operations;
1367         sb->s_export_op = &btrfs_export_ops;
1368         sb->s_xattr = btrfs_xattr_handlers;
1369         sb->s_time_gran = 1;
1370 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1371         sb->s_flags |= SB_POSIXACL;
1372 #endif
1373         sb->s_flags |= SB_I_VERSION;
1374         sb->s_iflags |= SB_I_CGROUPWB;
1375
1376         err = super_setup_bdi(sb);
1377         if (err) {
1378                 btrfs_err(fs_info, "super_setup_bdi failed");
1379                 return err;
1380         }
1381
1382         err = open_ctree(sb, fs_devices, (char *)data);
1383         if (err) {
1384                 btrfs_err(fs_info, "open_ctree failed");
1385                 return err;
1386         }
1387
1388         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1389         if (IS_ERR(inode)) {
1390                 err = PTR_ERR(inode);
1391                 goto fail_close;
1392         }
1393
1394         sb->s_root = d_make_root(inode);
1395         if (!sb->s_root) {
1396                 err = -ENOMEM;
1397                 goto fail_close;
1398         }
1399
1400         cleancache_init_fs(sb);
1401         sb->s_flags |= SB_ACTIVE;
1402         return 0;
1403
1404 fail_close:
1405         close_ctree(fs_info);
1406         return err;
1407 }
1408
1409 int btrfs_sync_fs(struct super_block *sb, int wait)
1410 {
1411         struct btrfs_trans_handle *trans;
1412         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1413         struct btrfs_root *root = fs_info->tree_root;
1414
1415         trace_btrfs_sync_fs(fs_info, wait);
1416
1417         if (!wait) {
1418                 filemap_flush(fs_info->btree_inode->i_mapping);
1419                 return 0;
1420         }
1421
1422         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1423
1424         trans = btrfs_attach_transaction_barrier(root);
1425         if (IS_ERR(trans)) {
1426                 /* no transaction, don't bother */
1427                 if (PTR_ERR(trans) == -ENOENT) {
1428                         /*
1429                          * Exit unless we have some pending changes
1430                          * that need to go through commit
1431                          */
1432                         if (fs_info->pending_changes == 0)
1433                                 return 0;
1434                         /*
1435                          * A non-blocking test if the fs is frozen. We must not
1436                          * start a new transaction here otherwise a deadlock
1437                          * happens. The pending operations are delayed to the
1438                          * next commit after thawing.
1439                          */
1440                         if (sb_start_write_trylock(sb))
1441                                 sb_end_write(sb);
1442                         else
1443                                 return 0;
1444                         trans = btrfs_start_transaction(root, 0);
1445                 }
1446                 if (IS_ERR(trans))
1447                         return PTR_ERR(trans);
1448         }
1449         return btrfs_commit_transaction(trans);
1450 }
1451
1452 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1453 {
1454         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1455         *printed = true;
1456 }
1457
1458 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1459 {
1460         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1461         const char *compress_type;
1462         const char *subvol_name;
1463         bool printed = false;
1464
1465         if (btrfs_test_opt(info, DEGRADED))
1466                 seq_puts(seq, ",degraded");
1467         if (btrfs_test_opt(info, NODATASUM))
1468                 seq_puts(seq, ",nodatasum");
1469         if (btrfs_test_opt(info, NODATACOW))
1470                 seq_puts(seq, ",nodatacow");
1471         if (btrfs_test_opt(info, NOBARRIER))
1472                 seq_puts(seq, ",nobarrier");
1473         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1474                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1475         if (info->thread_pool_size !=  min_t(unsigned long,
1476                                              num_online_cpus() + 2, 8))
1477                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1478         if (btrfs_test_opt(info, COMPRESS)) {
1479                 compress_type = btrfs_compress_type2str(info->compress_type);
1480                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1481                         seq_printf(seq, ",compress-force=%s", compress_type);
1482                 else
1483                         seq_printf(seq, ",compress=%s", compress_type);
1484                 if (info->compress_level)
1485                         seq_printf(seq, ":%d", info->compress_level);
1486         }
1487         if (btrfs_test_opt(info, NOSSD))
1488                 seq_puts(seq, ",nossd");
1489         if (btrfs_test_opt(info, SSD_SPREAD))
1490                 seq_puts(seq, ",ssd_spread");
1491         else if (btrfs_test_opt(info, SSD))
1492                 seq_puts(seq, ",ssd");
1493         if (btrfs_test_opt(info, NOTREELOG))
1494                 seq_puts(seq, ",notreelog");
1495         if (btrfs_test_opt(info, NOLOGREPLAY))
1496                 print_rescue_option(seq, "nologreplay", &printed);
1497         if (btrfs_test_opt(info, USEBACKUPROOT))
1498                 print_rescue_option(seq, "usebackuproot", &printed);
1499         if (btrfs_test_opt(info, IGNOREBADROOTS))
1500                 print_rescue_option(seq, "ignorebadroots", &printed);
1501         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1502                 print_rescue_option(seq, "ignoredatacsums", &printed);
1503         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1504                 seq_puts(seq, ",flushoncommit");
1505         if (btrfs_test_opt(info, DISCARD_SYNC))
1506                 seq_puts(seq, ",discard");
1507         if (btrfs_test_opt(info, DISCARD_ASYNC))
1508                 seq_puts(seq, ",discard=async");
1509         if (!(info->sb->s_flags & SB_POSIXACL))
1510                 seq_puts(seq, ",noacl");
1511         if (btrfs_free_space_cache_v1_active(info))
1512                 seq_puts(seq, ",space_cache");
1513         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1514                 seq_puts(seq, ",space_cache=v2");
1515         else
1516                 seq_puts(seq, ",nospace_cache");
1517         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1518                 seq_puts(seq, ",rescan_uuid_tree");
1519         if (btrfs_test_opt(info, CLEAR_CACHE))
1520                 seq_puts(seq, ",clear_cache");
1521         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1522                 seq_puts(seq, ",user_subvol_rm_allowed");
1523         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1524                 seq_puts(seq, ",enospc_debug");
1525         if (btrfs_test_opt(info, AUTO_DEFRAG))
1526                 seq_puts(seq, ",autodefrag");
1527         if (btrfs_test_opt(info, SKIP_BALANCE))
1528                 seq_puts(seq, ",skip_balance");
1529 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1530         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1531                 seq_puts(seq, ",check_int_data");
1532         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1533                 seq_puts(seq, ",check_int");
1534         if (info->check_integrity_print_mask)
1535                 seq_printf(seq, ",check_int_print_mask=%d",
1536                                 info->check_integrity_print_mask);
1537 #endif
1538         if (info->metadata_ratio)
1539                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1540         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1541                 seq_puts(seq, ",fatal_errors=panic");
1542         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1543                 seq_printf(seq, ",commit=%u", info->commit_interval);
1544 #ifdef CONFIG_BTRFS_DEBUG
1545         if (btrfs_test_opt(info, FRAGMENT_DATA))
1546                 seq_puts(seq, ",fragment=data");
1547         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1548                 seq_puts(seq, ",fragment=metadata");
1549 #endif
1550         if (btrfs_test_opt(info, REF_VERIFY))
1551                 seq_puts(seq, ",ref_verify");
1552         seq_printf(seq, ",subvolid=%llu",
1553                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1554         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1555                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1556         if (!IS_ERR(subvol_name)) {
1557                 seq_puts(seq, ",subvol=");
1558                 seq_escape(seq, subvol_name, " \t\n\\");
1559                 kfree(subvol_name);
1560         }
1561         return 0;
1562 }
1563
1564 static int btrfs_test_super(struct super_block *s, void *data)
1565 {
1566         struct btrfs_fs_info *p = data;
1567         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1568
1569         return fs_info->fs_devices == p->fs_devices;
1570 }
1571
1572 static int btrfs_set_super(struct super_block *s, void *data)
1573 {
1574         int err = set_anon_super(s, data);
1575         if (!err)
1576                 s->s_fs_info = data;
1577         return err;
1578 }
1579
1580 /*
1581  * subvolumes are identified by ino 256
1582  */
1583 static inline int is_subvolume_inode(struct inode *inode)
1584 {
1585         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1586                 return 1;
1587         return 0;
1588 }
1589
1590 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1591                                    struct vfsmount *mnt)
1592 {
1593         struct dentry *root;
1594         int ret;
1595
1596         if (!subvol_name) {
1597                 if (!subvol_objectid) {
1598                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1599                                                           &subvol_objectid);
1600                         if (ret) {
1601                                 root = ERR_PTR(ret);
1602                                 goto out;
1603                         }
1604                 }
1605                 subvol_name = btrfs_get_subvol_name_from_objectid(
1606                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1607                 if (IS_ERR(subvol_name)) {
1608                         root = ERR_CAST(subvol_name);
1609                         subvol_name = NULL;
1610                         goto out;
1611                 }
1612
1613         }
1614
1615         root = mount_subtree(mnt, subvol_name);
1616         /* mount_subtree() drops our reference on the vfsmount. */
1617         mnt = NULL;
1618
1619         if (!IS_ERR(root)) {
1620                 struct super_block *s = root->d_sb;
1621                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1622                 struct inode *root_inode = d_inode(root);
1623                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1624
1625                 ret = 0;
1626                 if (!is_subvolume_inode(root_inode)) {
1627                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1628                                subvol_name);
1629                         ret = -EINVAL;
1630                 }
1631                 if (subvol_objectid && root_objectid != subvol_objectid) {
1632                         /*
1633                          * This will also catch a race condition where a
1634                          * subvolume which was passed by ID is renamed and
1635                          * another subvolume is renamed over the old location.
1636                          */
1637                         btrfs_err(fs_info,
1638                                   "subvol '%s' does not match subvolid %llu",
1639                                   subvol_name, subvol_objectid);
1640                         ret = -EINVAL;
1641                 }
1642                 if (ret) {
1643                         dput(root);
1644                         root = ERR_PTR(ret);
1645                         deactivate_locked_super(s);
1646                 }
1647         }
1648
1649 out:
1650         mntput(mnt);
1651         kfree(subvol_name);
1652         return root;
1653 }
1654
1655 /*
1656  * Find a superblock for the given device / mount point.
1657  *
1658  * Note: This is based on mount_bdev from fs/super.c with a few additions
1659  *       for multiple device setup.  Make sure to keep it in sync.
1660  */
1661 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1662                 int flags, const char *device_name, void *data)
1663 {
1664         struct block_device *bdev = NULL;
1665         struct super_block *s;
1666         struct btrfs_device *device = NULL;
1667         struct btrfs_fs_devices *fs_devices = NULL;
1668         struct btrfs_fs_info *fs_info = NULL;
1669         void *new_sec_opts = NULL;
1670         fmode_t mode = FMODE_READ;
1671         int error = 0;
1672
1673         if (!(flags & SB_RDONLY))
1674                 mode |= FMODE_WRITE;
1675
1676         if (data) {
1677                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1678                 if (error)
1679                         return ERR_PTR(error);
1680         }
1681
1682         /*
1683          * Setup a dummy root and fs_info for test/set super.  This is because
1684          * we don't actually fill this stuff out until open_ctree, but we need
1685          * then open_ctree will properly initialize the file system specific
1686          * settings later.  btrfs_init_fs_info initializes the static elements
1687          * of the fs_info (locks and such) to make cleanup easier if we find a
1688          * superblock with our given fs_devices later on at sget() time.
1689          */
1690         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1691         if (!fs_info) {
1692                 error = -ENOMEM;
1693                 goto error_sec_opts;
1694         }
1695         btrfs_init_fs_info(fs_info);
1696
1697         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1698         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1699         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1700                 error = -ENOMEM;
1701                 goto error_fs_info;
1702         }
1703
1704         mutex_lock(&uuid_mutex);
1705         error = btrfs_parse_device_options(data, mode, fs_type);
1706         if (error) {
1707                 mutex_unlock(&uuid_mutex);
1708                 goto error_fs_info;
1709         }
1710
1711         device = btrfs_scan_one_device(device_name, mode, fs_type);
1712         if (IS_ERR(device)) {
1713                 mutex_unlock(&uuid_mutex);
1714                 error = PTR_ERR(device);
1715                 goto error_fs_info;
1716         }
1717
1718         fs_devices = device->fs_devices;
1719         fs_info->fs_devices = fs_devices;
1720
1721         error = btrfs_open_devices(fs_devices, mode, fs_type);
1722         mutex_unlock(&uuid_mutex);
1723         if (error)
1724                 goto error_fs_info;
1725
1726         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1727                 error = -EACCES;
1728                 goto error_close_devices;
1729         }
1730
1731         bdev = fs_devices->latest_bdev;
1732         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1733                  fs_info);
1734         if (IS_ERR(s)) {
1735                 error = PTR_ERR(s);
1736                 goto error_close_devices;
1737         }
1738
1739         if (s->s_root) {
1740                 btrfs_close_devices(fs_devices);
1741                 btrfs_free_fs_info(fs_info);
1742                 if ((flags ^ s->s_flags) & SB_RDONLY)
1743                         error = -EBUSY;
1744         } else {
1745                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1746                 btrfs_sb(s)->bdev_holder = fs_type;
1747                 if (!strstr(crc32c_impl(), "generic"))
1748                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1749                 error = btrfs_fill_super(s, fs_devices, data);
1750         }
1751         if (!error)
1752                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1753         security_free_mnt_opts(&new_sec_opts);
1754         if (error) {
1755                 deactivate_locked_super(s);
1756                 return ERR_PTR(error);
1757         }
1758
1759         return dget(s->s_root);
1760
1761 error_close_devices:
1762         btrfs_close_devices(fs_devices);
1763 error_fs_info:
1764         btrfs_free_fs_info(fs_info);
1765 error_sec_opts:
1766         security_free_mnt_opts(&new_sec_opts);
1767         return ERR_PTR(error);
1768 }
1769
1770 /*
1771  * Mount function which is called by VFS layer.
1772  *
1773  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1774  * which needs vfsmount* of device's root (/).  This means device's root has to
1775  * be mounted internally in any case.
1776  *
1777  * Operation flow:
1778  *   1. Parse subvol id related options for later use in mount_subvol().
1779  *
1780  *   2. Mount device's root (/) by calling vfs_kern_mount().
1781  *
1782  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1783  *      first place. In order to avoid calling btrfs_mount() again, we use
1784  *      different file_system_type which is not registered to VFS by
1785  *      register_filesystem() (btrfs_root_fs_type). As a result,
1786  *      btrfs_mount_root() is called. The return value will be used by
1787  *      mount_subtree() in mount_subvol().
1788  *
1789  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1790  *      "btrfs subvolume set-default", mount_subvol() is called always.
1791  */
1792 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1793                 const char *device_name, void *data)
1794 {
1795         struct vfsmount *mnt_root;
1796         struct dentry *root;
1797         char *subvol_name = NULL;
1798         u64 subvol_objectid = 0;
1799         int error = 0;
1800
1801         error = btrfs_parse_subvol_options(data, &subvol_name,
1802                                         &subvol_objectid);
1803         if (error) {
1804                 kfree(subvol_name);
1805                 return ERR_PTR(error);
1806         }
1807
1808         /* mount device's root (/) */
1809         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1810         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1811                 if (flags & SB_RDONLY) {
1812                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1813                                 flags & ~SB_RDONLY, device_name, data);
1814                 } else {
1815                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1816                                 flags | SB_RDONLY, device_name, data);
1817                         if (IS_ERR(mnt_root)) {
1818                                 root = ERR_CAST(mnt_root);
1819                                 kfree(subvol_name);
1820                                 goto out;
1821                         }
1822
1823                         down_write(&mnt_root->mnt_sb->s_umount);
1824                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1825                         up_write(&mnt_root->mnt_sb->s_umount);
1826                         if (error < 0) {
1827                                 root = ERR_PTR(error);
1828                                 mntput(mnt_root);
1829                                 kfree(subvol_name);
1830                                 goto out;
1831                         }
1832                 }
1833         }
1834         if (IS_ERR(mnt_root)) {
1835                 root = ERR_CAST(mnt_root);
1836                 kfree(subvol_name);
1837                 goto out;
1838         }
1839
1840         /* mount_subvol() will free subvol_name and mnt_root */
1841         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1842
1843 out:
1844         return root;
1845 }
1846
1847 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1848                                      u32 new_pool_size, u32 old_pool_size)
1849 {
1850         if (new_pool_size == old_pool_size)
1851                 return;
1852
1853         fs_info->thread_pool_size = new_pool_size;
1854
1855         btrfs_info(fs_info, "resize thread pool %d -> %d",
1856                old_pool_size, new_pool_size);
1857
1858         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1859         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1860         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1861         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1862         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1863         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1864                                 new_pool_size);
1865         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1866         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1867         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1868         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1869         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1870                                 new_pool_size);
1871 }
1872
1873 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1874                                        unsigned long old_opts, int flags)
1875 {
1876         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1877             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1878              (flags & SB_RDONLY))) {
1879                 /* wait for any defraggers to finish */
1880                 wait_event(fs_info->transaction_wait,
1881                            (atomic_read(&fs_info->defrag_running) == 0));
1882                 if (flags & SB_RDONLY)
1883                         sync_filesystem(fs_info->sb);
1884         }
1885 }
1886
1887 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1888                                          unsigned long old_opts)
1889 {
1890         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1891
1892         /*
1893          * We need to cleanup all defragable inodes if the autodefragment is
1894          * close or the filesystem is read only.
1895          */
1896         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1897             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1898                 btrfs_cleanup_defrag_inodes(fs_info);
1899         }
1900
1901         /* If we toggled discard async */
1902         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1903             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1904                 btrfs_discard_resume(fs_info);
1905         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1906                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1907                 btrfs_discard_cleanup(fs_info);
1908
1909         /* If we toggled space cache */
1910         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1911                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1912 }
1913
1914 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1915 {
1916         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1917         unsigned old_flags = sb->s_flags;
1918         unsigned long old_opts = fs_info->mount_opt;
1919         unsigned long old_compress_type = fs_info->compress_type;
1920         u64 old_max_inline = fs_info->max_inline;
1921         u32 old_thread_pool_size = fs_info->thread_pool_size;
1922         u32 old_metadata_ratio = fs_info->metadata_ratio;
1923         int ret;
1924
1925         sync_filesystem(sb);
1926         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1927
1928         if (data) {
1929                 void *new_sec_opts = NULL;
1930
1931                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1932                 if (!ret)
1933                         ret = security_sb_remount(sb, new_sec_opts);
1934                 security_free_mnt_opts(&new_sec_opts);
1935                 if (ret)
1936                         goto restore;
1937         }
1938
1939         ret = btrfs_parse_options(fs_info, data, *flags);
1940         if (ret)
1941                 goto restore;
1942
1943         btrfs_remount_begin(fs_info, old_opts, *flags);
1944         btrfs_resize_thread_pool(fs_info,
1945                 fs_info->thread_pool_size, old_thread_pool_size);
1946
1947         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1948             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1949             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
1950                 btrfs_warn(fs_info,
1951                 "remount supports changing free space tree only from ro to rw");
1952                 /* Make sure free space cache options match the state on disk */
1953                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1954                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1955                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1956                 }
1957                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1958                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1959                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1960                 }
1961         }
1962
1963         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1964                 goto out;
1965
1966         if (*flags & SB_RDONLY) {
1967                 /*
1968                  * this also happens on 'umount -rf' or on shutdown, when
1969                  * the filesystem is busy.
1970                  */
1971                 cancel_work_sync(&fs_info->async_reclaim_work);
1972                 cancel_work_sync(&fs_info->async_data_reclaim_work);
1973
1974                 btrfs_discard_cleanup(fs_info);
1975
1976                 /* wait for the uuid_scan task to finish */
1977                 down(&fs_info->uuid_tree_rescan_sem);
1978                 /* avoid complains from lockdep et al. */
1979                 up(&fs_info->uuid_tree_rescan_sem);
1980
1981                 btrfs_set_sb_rdonly(sb);
1982
1983                 /*
1984                  * Setting SB_RDONLY will put the cleaner thread to
1985                  * sleep at the next loop if it's already active.
1986                  * If it's already asleep, we'll leave unused block
1987                  * groups on disk until we're mounted read-write again
1988                  * unless we clean them up here.
1989                  */
1990                 btrfs_delete_unused_bgs(fs_info);
1991
1992                 /*
1993                  * The cleaner task could be already running before we set the
1994                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
1995                  * We must make sure that after we finish the remount, i.e. after
1996                  * we call btrfs_commit_super(), the cleaner can no longer start
1997                  * a transaction - either because it was dropping a dead root,
1998                  * running delayed iputs or deleting an unused block group (the
1999                  * cleaner picked a block group from the list of unused block
2000                  * groups before we were able to in the previous call to
2001                  * btrfs_delete_unused_bgs()).
2002                  */
2003                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
2004                             TASK_UNINTERRUPTIBLE);
2005
2006                 /*
2007                  * We've set the superblock to RO mode, so we might have made
2008                  * the cleaner task sleep without running all pending delayed
2009                  * iputs. Go through all the delayed iputs here, so that if an
2010                  * unmount happens without remounting RW we don't end up at
2011                  * finishing close_ctree() with a non-empty list of delayed
2012                  * iputs.
2013                  */
2014                 btrfs_run_delayed_iputs(fs_info);
2015
2016                 btrfs_dev_replace_suspend_for_unmount(fs_info);
2017                 btrfs_scrub_cancel(fs_info);
2018                 btrfs_pause_balance(fs_info);
2019
2020                 /*
2021                  * Pause the qgroup rescan worker if it is running. We don't want
2022                  * it to be still running after we are in RO mode, as after that,
2023                  * by the time we unmount, it might have left a transaction open,
2024                  * so we would leak the transaction and/or crash.
2025                  */
2026                 btrfs_qgroup_wait_for_completion(fs_info, false);
2027
2028                 ret = btrfs_commit_super(fs_info);
2029                 if (ret)
2030                         goto restore;
2031         } else {
2032                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2033                         btrfs_err(fs_info,
2034                                 "Remounting read-write after error is not allowed");
2035                         ret = -EINVAL;
2036                         goto restore;
2037                 }
2038                 if (fs_info->fs_devices->rw_devices == 0) {
2039                         ret = -EACCES;
2040                         goto restore;
2041                 }
2042
2043                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2044                         btrfs_warn(fs_info,
2045                 "too many missing devices, writable remount is not allowed");
2046                         ret = -EACCES;
2047                         goto restore;
2048                 }
2049
2050                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2051                         btrfs_warn(fs_info,
2052                 "mount required to replay tree-log, cannot remount read-write");
2053                         ret = -EINVAL;
2054                         goto restore;
2055                 }
2056                 if (fs_info->sectorsize < PAGE_SIZE) {
2057                         btrfs_warn(fs_info,
2058         "read-write mount is not yet allowed for sectorsize %u page size %lu",
2059                                    fs_info->sectorsize, PAGE_SIZE);
2060                         ret = -EINVAL;
2061                         goto restore;
2062                 }
2063
2064                 /*
2065                  * NOTE: when remounting with a change that does writes, don't
2066                  * put it anywhere above this point, as we are not sure to be
2067                  * safe to write until we pass the above checks.
2068                  */
2069                 ret = btrfs_start_pre_rw_mount(fs_info);
2070                 if (ret)
2071                         goto restore;
2072
2073                 btrfs_clear_sb_rdonly(sb);
2074
2075                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2076         }
2077 out:
2078         /*
2079          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2080          * since the absence of the flag means it can be toggled off by remount.
2081          */
2082         *flags |= SB_I_VERSION;
2083
2084         wake_up_process(fs_info->transaction_kthread);
2085         btrfs_remount_cleanup(fs_info, old_opts);
2086         btrfs_clear_oneshot_options(fs_info);
2087         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2088
2089         return 0;
2090
2091 restore:
2092         /* We've hit an error - don't reset SB_RDONLY */
2093         if (sb_rdonly(sb))
2094                 old_flags |= SB_RDONLY;
2095         if (!(old_flags & SB_RDONLY))
2096                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2097         sb->s_flags = old_flags;
2098         fs_info->mount_opt = old_opts;
2099         fs_info->compress_type = old_compress_type;
2100         fs_info->max_inline = old_max_inline;
2101         btrfs_resize_thread_pool(fs_info,
2102                 old_thread_pool_size, fs_info->thread_pool_size);
2103         fs_info->metadata_ratio = old_metadata_ratio;
2104         btrfs_remount_cleanup(fs_info, old_opts);
2105         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2106
2107         return ret;
2108 }
2109
2110 /* Used to sort the devices by max_avail(descending sort) */
2111 static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
2112                                        const void *dev_info2)
2113 {
2114         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2115             ((struct btrfs_device_info *)dev_info2)->max_avail)
2116                 return -1;
2117         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2118                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2119                 return 1;
2120         else
2121         return 0;
2122 }
2123
2124 /*
2125  * sort the devices by max_avail, in which max free extent size of each device
2126  * is stored.(Descending Sort)
2127  */
2128 static inline void btrfs_descending_sort_devices(
2129                                         struct btrfs_device_info *devices,
2130                                         size_t nr_devices)
2131 {
2132         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2133              btrfs_cmp_device_free_bytes, NULL);
2134 }
2135
2136 /*
2137  * The helper to calc the free space on the devices that can be used to store
2138  * file data.
2139  */
2140 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2141                                               u64 *free_bytes)
2142 {
2143         struct btrfs_device_info *devices_info;
2144         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2145         struct btrfs_device *device;
2146         u64 type;
2147         u64 avail_space;
2148         u64 min_stripe_size;
2149         int num_stripes = 1;
2150         int i = 0, nr_devices;
2151         const struct btrfs_raid_attr *rattr;
2152
2153         /*
2154          * We aren't under the device list lock, so this is racy-ish, but good
2155          * enough for our purposes.
2156          */
2157         nr_devices = fs_info->fs_devices->open_devices;
2158         if (!nr_devices) {
2159                 smp_mb();
2160                 nr_devices = fs_info->fs_devices->open_devices;
2161                 ASSERT(nr_devices);
2162                 if (!nr_devices) {
2163                         *free_bytes = 0;
2164                         return 0;
2165                 }
2166         }
2167
2168         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2169                                GFP_KERNEL);
2170         if (!devices_info)
2171                 return -ENOMEM;
2172
2173         /* calc min stripe number for data space allocation */
2174         type = btrfs_data_alloc_profile(fs_info);
2175         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2176
2177         if (type & BTRFS_BLOCK_GROUP_RAID0)
2178                 num_stripes = nr_devices;
2179         else if (type & BTRFS_BLOCK_GROUP_RAID1)
2180                 num_stripes = 2;
2181         else if (type & BTRFS_BLOCK_GROUP_RAID1C3)
2182                 num_stripes = 3;
2183         else if (type & BTRFS_BLOCK_GROUP_RAID1C4)
2184                 num_stripes = 4;
2185         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2186                 num_stripes = 4;
2187
2188         /* Adjust for more than 1 stripe per device */
2189         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2190
2191         rcu_read_lock();
2192         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2193                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2194                                                 &device->dev_state) ||
2195                     !device->bdev ||
2196                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2197                         continue;
2198
2199                 if (i >= nr_devices)
2200                         break;
2201
2202                 avail_space = device->total_bytes - device->bytes_used;
2203
2204                 /* align with stripe_len */
2205                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2206
2207                 /*
2208                  * In order to avoid overwriting the superblock on the drive,
2209                  * btrfs starts at an offset of at least 1MB when doing chunk
2210                  * allocation.
2211                  *
2212                  * This ensures we have at least min_stripe_size free space
2213                  * after excluding 1MB.
2214                  */
2215                 if (avail_space <= SZ_1M + min_stripe_size)
2216                         continue;
2217
2218                 avail_space -= SZ_1M;
2219
2220                 devices_info[i].dev = device;
2221                 devices_info[i].max_avail = avail_space;
2222
2223                 i++;
2224         }
2225         rcu_read_unlock();
2226
2227         nr_devices = i;
2228
2229         btrfs_descending_sort_devices(devices_info, nr_devices);
2230
2231         i = nr_devices - 1;
2232         avail_space = 0;
2233         while (nr_devices >= rattr->devs_min) {
2234                 num_stripes = min(num_stripes, nr_devices);
2235
2236                 if (devices_info[i].max_avail >= min_stripe_size) {
2237                         int j;
2238                         u64 alloc_size;
2239
2240                         avail_space += devices_info[i].max_avail * num_stripes;
2241                         alloc_size = devices_info[i].max_avail;
2242                         for (j = i + 1 - num_stripes; j <= i; j++)
2243                                 devices_info[j].max_avail -= alloc_size;
2244                 }
2245                 i--;
2246                 nr_devices--;
2247         }
2248
2249         kfree(devices_info);
2250         *free_bytes = avail_space;
2251         return 0;
2252 }
2253
2254 /*
2255  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2256  *
2257  * If there's a redundant raid level at DATA block groups, use the respective
2258  * multiplier to scale the sizes.
2259  *
2260  * Unused device space usage is based on simulating the chunk allocator
2261  * algorithm that respects the device sizes and order of allocations.  This is
2262  * a close approximation of the actual use but there are other factors that may
2263  * change the result (like a new metadata chunk).
2264  *
2265  * If metadata is exhausted, f_bavail will be 0.
2266  */
2267 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2268 {
2269         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2270         struct btrfs_super_block *disk_super = fs_info->super_copy;
2271         struct btrfs_space_info *found;
2272         u64 total_used = 0;
2273         u64 total_free_data = 0;
2274         u64 total_free_meta = 0;
2275         u32 bits = fs_info->sectorsize_bits;
2276         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2277         unsigned factor = 1;
2278         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2279         int ret;
2280         u64 thresh = 0;
2281         int mixed = 0;
2282
2283         list_for_each_entry(found, &fs_info->space_info, list) {
2284                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2285                         int i;
2286
2287                         total_free_data += found->disk_total - found->disk_used;
2288                         total_free_data -=
2289                                 btrfs_account_ro_block_groups_free_space(found);
2290
2291                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2292                                 if (!list_empty(&found->block_groups[i]))
2293                                         factor = btrfs_bg_type_to_factor(
2294                                                 btrfs_raid_array[i].bg_flag);
2295                         }
2296                 }
2297
2298                 /*
2299                  * Metadata in mixed block goup profiles are accounted in data
2300                  */
2301                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2302                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2303                                 mixed = 1;
2304                         else
2305                                 total_free_meta += found->disk_total -
2306                                         found->disk_used;
2307                 }
2308
2309                 total_used += found->disk_used;
2310         }
2311
2312         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2313         buf->f_blocks >>= bits;
2314         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2315
2316         /* Account global block reserve as used, it's in logical size already */
2317         spin_lock(&block_rsv->lock);
2318         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2319         if (buf->f_bfree >= block_rsv->size >> bits)
2320                 buf->f_bfree -= block_rsv->size >> bits;
2321         else
2322                 buf->f_bfree = 0;
2323         spin_unlock(&block_rsv->lock);
2324
2325         buf->f_bavail = div_u64(total_free_data, factor);
2326         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2327         if (ret)
2328                 return ret;
2329         buf->f_bavail += div_u64(total_free_data, factor);
2330         buf->f_bavail = buf->f_bavail >> bits;
2331
2332         /*
2333          * We calculate the remaining metadata space minus global reserve. If
2334          * this is (supposedly) smaller than zero, there's no space. But this
2335          * does not hold in practice, the exhausted state happens where's still
2336          * some positive delta. So we apply some guesswork and compare the
2337          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2338          *
2339          * We probably cannot calculate the exact threshold value because this
2340          * depends on the internal reservations requested by various
2341          * operations, so some operations that consume a few metadata will
2342          * succeed even if the Avail is zero. But this is better than the other
2343          * way around.
2344          */
2345         thresh = SZ_4M;
2346
2347         /*
2348          * We only want to claim there's no available space if we can no longer
2349          * allocate chunks for our metadata profile and our global reserve will
2350          * not fit in the free metadata space.  If we aren't ->full then we
2351          * still can allocate chunks and thus are fine using the currently
2352          * calculated f_bavail.
2353          */
2354         if (!mixed && block_rsv->space_info->full &&
2355             total_free_meta - thresh < block_rsv->size)
2356                 buf->f_bavail = 0;
2357
2358         buf->f_type = BTRFS_SUPER_MAGIC;
2359         buf->f_bsize = dentry->d_sb->s_blocksize;
2360         buf->f_namelen = BTRFS_NAME_LEN;
2361
2362         /* We treat it as constant endianness (it doesn't matter _which_)
2363            because we want the fsid to come out the same whether mounted
2364            on a big-endian or little-endian host */
2365         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2366         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2367         /* Mask in the root object ID too, to disambiguate subvols */
2368         buf->f_fsid.val[0] ^=
2369                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2370         buf->f_fsid.val[1] ^=
2371                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2372
2373         return 0;
2374 }
2375
2376 static void btrfs_kill_super(struct super_block *sb)
2377 {
2378         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2379         kill_anon_super(sb);
2380         btrfs_free_fs_info(fs_info);
2381 }
2382
2383 static struct file_system_type btrfs_fs_type = {
2384         .owner          = THIS_MODULE,
2385         .name           = "btrfs",
2386         .mount          = btrfs_mount,
2387         .kill_sb        = btrfs_kill_super,
2388         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2389 };
2390
2391 static struct file_system_type btrfs_root_fs_type = {
2392         .owner          = THIS_MODULE,
2393         .name           = "btrfs",
2394         .mount          = btrfs_mount_root,
2395         .kill_sb        = btrfs_kill_super,
2396         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2397 };
2398
2399 MODULE_ALIAS_FS("btrfs");
2400
2401 static int btrfs_control_open(struct inode *inode, struct file *file)
2402 {
2403         /*
2404          * The control file's private_data is used to hold the
2405          * transaction when it is started and is used to keep
2406          * track of whether a transaction is already in progress.
2407          */
2408         file->private_data = NULL;
2409         return 0;
2410 }
2411
2412 /*
2413  * Used by /dev/btrfs-control for devices ioctls.
2414  */
2415 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2416                                 unsigned long arg)
2417 {
2418         struct btrfs_ioctl_vol_args *vol;
2419         struct btrfs_device *device = NULL;
2420         int ret = -ENOTTY;
2421
2422         if (!capable(CAP_SYS_ADMIN))
2423                 return -EPERM;
2424
2425         vol = memdup_user((void __user *)arg, sizeof(*vol));
2426         if (IS_ERR(vol))
2427                 return PTR_ERR(vol);
2428         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2429
2430         switch (cmd) {
2431         case BTRFS_IOC_SCAN_DEV:
2432                 mutex_lock(&uuid_mutex);
2433                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2434                                                &btrfs_root_fs_type);
2435                 ret = PTR_ERR_OR_ZERO(device);
2436                 mutex_unlock(&uuid_mutex);
2437                 break;
2438         case BTRFS_IOC_FORGET_DEV:
2439                 ret = btrfs_forget_devices(vol->name);
2440                 break;
2441         case BTRFS_IOC_DEVICES_READY:
2442                 mutex_lock(&uuid_mutex);
2443                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2444                                                &btrfs_root_fs_type);
2445                 if (IS_ERR(device)) {
2446                         mutex_unlock(&uuid_mutex);
2447                         ret = PTR_ERR(device);
2448                         break;
2449                 }
2450                 ret = !(device->fs_devices->num_devices ==
2451                         device->fs_devices->total_devices);
2452                 mutex_unlock(&uuid_mutex);
2453                 break;
2454         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2455                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2456                 break;
2457         }
2458
2459         kfree(vol);
2460         return ret;
2461 }
2462
2463 static int btrfs_freeze(struct super_block *sb)
2464 {
2465         struct btrfs_trans_handle *trans;
2466         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2467         struct btrfs_root *root = fs_info->tree_root;
2468
2469         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2470         /*
2471          * We don't need a barrier here, we'll wait for any transaction that
2472          * could be in progress on other threads (and do delayed iputs that
2473          * we want to avoid on a frozen filesystem), or do the commit
2474          * ourselves.
2475          */
2476         trans = btrfs_attach_transaction_barrier(root);
2477         if (IS_ERR(trans)) {
2478                 /* no transaction, don't bother */
2479                 if (PTR_ERR(trans) == -ENOENT)
2480                         return 0;
2481                 return PTR_ERR(trans);
2482         }
2483         return btrfs_commit_transaction(trans);
2484 }
2485
2486 static int btrfs_unfreeze(struct super_block *sb)
2487 {
2488         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2489
2490         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2491         return 0;
2492 }
2493
2494 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2495 {
2496         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2497         struct btrfs_device *dev, *first_dev = NULL;
2498
2499         /*
2500          * Lightweight locking of the devices. We should not need
2501          * device_list_mutex here as we only read the device data and the list
2502          * is protected by RCU.  Even if a device is deleted during the list
2503          * traversals, we'll get valid data, the freeing callback will wait at
2504          * least until the rcu_read_unlock.
2505          */
2506         rcu_read_lock();
2507         list_for_each_entry_rcu(dev, &fs_info->fs_devices->devices, dev_list) {
2508                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2509                         continue;
2510                 if (!dev->name)
2511                         continue;
2512                 if (!first_dev || dev->devid < first_dev->devid)
2513                         first_dev = dev;
2514         }
2515
2516         if (first_dev)
2517                 seq_escape(m, rcu_str_deref(first_dev->name), " \t\n\\");
2518         else
2519                 WARN_ON(1);
2520         rcu_read_unlock();
2521         return 0;
2522 }
2523
2524 static const struct super_operations btrfs_super_ops = {
2525         .drop_inode     = btrfs_drop_inode,
2526         .evict_inode    = btrfs_evict_inode,
2527         .put_super      = btrfs_put_super,
2528         .sync_fs        = btrfs_sync_fs,
2529         .show_options   = btrfs_show_options,
2530         .show_devname   = btrfs_show_devname,
2531         .alloc_inode    = btrfs_alloc_inode,
2532         .destroy_inode  = btrfs_destroy_inode,
2533         .free_inode     = btrfs_free_inode,
2534         .statfs         = btrfs_statfs,
2535         .remount_fs     = btrfs_remount,
2536         .freeze_fs      = btrfs_freeze,
2537         .unfreeze_fs    = btrfs_unfreeze,
2538 };
2539
2540 static const struct file_operations btrfs_ctl_fops = {
2541         .open = btrfs_control_open,
2542         .unlocked_ioctl  = btrfs_control_ioctl,
2543         .compat_ioctl = compat_ptr_ioctl,
2544         .owner   = THIS_MODULE,
2545         .llseek = noop_llseek,
2546 };
2547
2548 static struct miscdevice btrfs_misc = {
2549         .minor          = BTRFS_MINOR,
2550         .name           = "btrfs-control",
2551         .fops           = &btrfs_ctl_fops
2552 };
2553
2554 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2555 MODULE_ALIAS("devname:btrfs-control");
2556
2557 static int __init btrfs_interface_init(void)
2558 {
2559         return misc_register(&btrfs_misc);
2560 }
2561
2562 static __cold void btrfs_interface_exit(void)
2563 {
2564         misc_deregister(&btrfs_misc);
2565 }
2566
2567 static void __init btrfs_print_mod_info(void)
2568 {
2569         static const char options[] = ""
2570 #ifdef CONFIG_BTRFS_DEBUG
2571                         ", debug=on"
2572 #endif
2573 #ifdef CONFIG_BTRFS_ASSERT
2574                         ", assert=on"
2575 #endif
2576 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2577                         ", integrity-checker=on"
2578 #endif
2579 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2580                         ", ref-verify=on"
2581 #endif
2582 #ifdef CONFIG_BLK_DEV_ZONED
2583                         ", zoned=yes"
2584 #else
2585                         ", zoned=no"
2586 #endif
2587                         ;
2588         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2589 }
2590
2591 static int __init init_btrfs_fs(void)
2592 {
2593         int err;
2594
2595         btrfs_props_init();
2596
2597         err = btrfs_init_sysfs();
2598         if (err)
2599                 return err;
2600
2601         btrfs_init_compress();
2602
2603         err = btrfs_init_cachep();
2604         if (err)
2605                 goto free_compress;
2606
2607         err = extent_io_init();
2608         if (err)
2609                 goto free_cachep;
2610
2611         err = extent_state_cache_init();
2612         if (err)
2613                 goto free_extent_io;
2614
2615         err = extent_map_init();
2616         if (err)
2617                 goto free_extent_state_cache;
2618
2619         err = ordered_data_init();
2620         if (err)
2621                 goto free_extent_map;
2622
2623         err = btrfs_delayed_inode_init();
2624         if (err)
2625                 goto free_ordered_data;
2626
2627         err = btrfs_auto_defrag_init();
2628         if (err)
2629                 goto free_delayed_inode;
2630
2631         err = btrfs_delayed_ref_init();
2632         if (err)
2633                 goto free_auto_defrag;
2634
2635         err = btrfs_prelim_ref_init();
2636         if (err)
2637                 goto free_delayed_ref;
2638
2639         err = btrfs_end_io_wq_init();
2640         if (err)
2641                 goto free_prelim_ref;
2642
2643         err = btrfs_interface_init();
2644         if (err)
2645                 goto free_end_io_wq;
2646
2647         btrfs_print_mod_info();
2648
2649         err = btrfs_run_sanity_tests();
2650         if (err)
2651                 goto unregister_ioctl;
2652
2653         err = register_filesystem(&btrfs_fs_type);
2654         if (err)
2655                 goto unregister_ioctl;
2656
2657         return 0;
2658
2659 unregister_ioctl:
2660         btrfs_interface_exit();
2661 free_end_io_wq:
2662         btrfs_end_io_wq_exit();
2663 free_prelim_ref:
2664         btrfs_prelim_ref_exit();
2665 free_delayed_ref:
2666         btrfs_delayed_ref_exit();
2667 free_auto_defrag:
2668         btrfs_auto_defrag_exit();
2669 free_delayed_inode:
2670         btrfs_delayed_inode_exit();
2671 free_ordered_data:
2672         ordered_data_exit();
2673 free_extent_map:
2674         extent_map_exit();
2675 free_extent_state_cache:
2676         extent_state_cache_exit();
2677 free_extent_io:
2678         extent_io_exit();
2679 free_cachep:
2680         btrfs_destroy_cachep();
2681 free_compress:
2682         btrfs_exit_compress();
2683         btrfs_exit_sysfs();
2684
2685         return err;
2686 }
2687
2688 static void __exit exit_btrfs_fs(void)
2689 {
2690         btrfs_destroy_cachep();
2691         btrfs_delayed_ref_exit();
2692         btrfs_auto_defrag_exit();
2693         btrfs_delayed_inode_exit();
2694         btrfs_prelim_ref_exit();
2695         ordered_data_exit();
2696         extent_map_exit();
2697         extent_state_cache_exit();
2698         extent_io_exit();
2699         btrfs_interface_exit();
2700         btrfs_end_io_wq_exit();
2701         unregister_filesystem(&btrfs_fs_type);
2702         btrfs_exit_sysfs();
2703         btrfs_cleanup_fs_uuids();
2704         btrfs_exit_compress();
2705 }
2706
2707 late_initcall(init_btrfs_fs);
2708 module_exit(exit_btrfs_fs)
2709
2710 MODULE_LICENSE("GPL");
2711 MODULE_SOFTDEP("pre: crc32c");
2712 MODULE_SOFTDEP("pre: xxhash64");
2713 MODULE_SOFTDEP("pre: sha256");
2714 MODULE_SOFTDEP("pre: blake2b-256");