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/ratelimit.h>
  27 #include <linux/crc32c.h>
  28 #include <linux/btrfs.h>
  29 #include "delayed-inode.h"
  30 #include "ctree.h"
  31 #include "disk-io.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "props.h"
  36 #include "xattr.h"
  37 #include "volumes.h"
  38 #include "export.h"
  39 #include "compression.h"
  40 #include "rcu-string.h"
  41 #include "dev-replace.h"
  42 #include "free-space-cache.h"
  43 #include "backref.h"
  44 #include "space-info.h"
  45 #include "sysfs.h"
  46 #include "zoned.h"
  47 #include "tests/btrfs-tests.h"
  48 #include "block-group.h"
  49 #include "discard.h"
  50 #include "qgroup.h"
  51 #include "raid56.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 #ifdef CONFIG_PRINTK
  71
  72 #define STATE_STRING_PREFACE    ": state "
  73 #define STATE_STRING_BUF_LEN    (sizeof(STATE_STRING_PREFACE) + BTRFS_FS_STATE_COUNT)
  74
  75 /*
  76  * Characters to print to indicate error conditions or uncommon filesystem state.
  77  * RO is not an error.
  78  */
  79 static const char fs_state_chars[] = {
  80         [BTRFS_FS_STATE_ERROR]                  = 'E',
  81         [BTRFS_FS_STATE_REMOUNTING]             = 'M',
  82         [BTRFS_FS_STATE_RO]                     = 0,
  83         [BTRFS_FS_STATE_TRANS_ABORTED]          = 'A',
  84         [BTRFS_FS_STATE_DEV_REPLACING]          = 'R',
  85         [BTRFS_FS_STATE_DUMMY_FS_INFO]          = 0,
  86         [BTRFS_FS_STATE_NO_CSUMS]               = 'C',
  87         [BTRFS_FS_STATE_LOG_CLEANUP_ERROR]      = 'L',
  88 };
  89
  90 static void btrfs_state_to_string(const struct btrfs_fs_info *info, char *buf)
  91 {
  92         unsigned int bit;
  93         bool states_printed = false;
  94         unsigned long fs_state = READ_ONCE(info->fs_state);
  95         char *curr = buf;
  96
  97         memcpy(curr, STATE_STRING_PREFACE, sizeof(STATE_STRING_PREFACE));
  98         curr += sizeof(STATE_STRING_PREFACE) - 1;
  99
 100         for_each_set_bit(bit, &fs_state, sizeof(fs_state)) {
 101                 WARN_ON_ONCE(bit >= BTRFS_FS_STATE_COUNT);
 102                 if ((bit < BTRFS_FS_STATE_COUNT) && fs_state_chars[bit]) {
 103                         *curr++ = fs_state_chars[bit];
 104                         states_printed = true;
 105                 }
 106         }
 107
 108         /* If no states were printed, reset the buffer */
 109         if (!states_printed)
 110                 curr = buf;
 111
 112         *curr++ = 0;
 113 }
 114 #endif
 115
 116 /*
 117  * Generally the error codes correspond to their respective errors, but there
 118  * are a few special cases.
 119  *
 120  * EUCLEAN: Any sort of corruption that we encounter.  The tree-checker for
 121  *          instance will return EUCLEAN if any of the blocks are corrupted in
 122  *          a way that is problematic.  We want to reserve EUCLEAN for these
 123  *          sort of corruptions.
 124  *
 125  * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
 126  *        need to use EROFS for this case.  We will have no idea of the
 127  *        original failure, that will have been reported at the time we tripped
 128  *        over the error.  Each subsequent error that doesn't have any context
 129  *        of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
 130  */
 131 const char * __attribute_const__ btrfs_decode_error(int errno)
 132 {
 133         char *errstr = "unknown";
 134
 135         switch (errno) {
 136         case -ENOENT:           /* -2 */
 137                 errstr = "No such entry";
 138                 break;
 139         case -EIO:              /* -5 */
 140                 errstr = "IO failure";
 141                 break;
 142         case -ENOMEM:           /* -12*/
 143                 errstr = "Out of memory";
 144                 break;
 145         case -EEXIST:           /* -17 */
 146                 errstr = "Object already exists";
 147                 break;
 148         case -ENOSPC:           /* -28 */
 149                 errstr = "No space left";
 150                 break;
 151         case -EROFS:            /* -30 */
 152                 errstr = "Readonly filesystem";
 153                 break;
 154         case -EOPNOTSUPP:       /* -95 */
 155                 errstr = "Operation not supported";
 156                 break;
 157         case -EUCLEAN:          /* -117 */
 158                 errstr = "Filesystem corrupted";
 159                 break;
 160         case -EDQUOT:           /* -122 */
 161                 errstr = "Quota exceeded";
 162                 break;
 163         }
 164
 165         return errstr;
 166 }
 167
 168 /*
 169  * __btrfs_handle_fs_error decodes expected errors from the caller and
 170  * invokes the appropriate error response.
 171  */
 172 __cold
 173 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 174                        unsigned int line, int errno, const char *fmt, ...)
 175 {
 176         struct super_block *sb = fs_info->sb;
 177 #ifdef CONFIG_PRINTK
 178         char statestr[STATE_STRING_BUF_LEN];
 179         const char *errstr;
 180 #endif
 181
 182         /*
 183          * Special case: if the error is EROFS, and we're already
 184          * under SB_RDONLY, then it is safe here.
 185          */
 186         if (errno == -EROFS && sb_rdonly(sb))
 187                 return;
 188
 189 #ifdef CONFIG_PRINTK
 190         errstr = btrfs_decode_error(errno);
 191         btrfs_state_to_string(fs_info, statestr);
 192         if (fmt) {
 193                 struct va_format vaf;
 194                 va_list args;
 195
 196                 va_start(args, fmt);
 197                 vaf.fmt = fmt;
 198                 vaf.va = &args;
 199
 200                 pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s (%pV)\n",
 201                         sb->s_id, statestr, function, line, errno, errstr, &vaf);
 202                 va_end(args);
 203         } else {
 204                 pr_crit("BTRFS: error (device %s%s) in %s:%d: errno=%d %s\n",
 205                         sb->s_id, statestr, function, line, errno, errstr);
 206         }
 207 #endif
 208
 209         /*
 210          * Today we only save the error info to memory.  Long term we'll
 211          * also send it down to the disk
 212          */
 213         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 214
 215         /* Don't go through full error handling during mount */
 216         if (!(sb->s_flags & SB_BORN))
 217                 return;
 218
 219         if (sb_rdonly(sb))
 220                 return;
 221
 222         btrfs_discard_stop(fs_info);
 223
 224         /* btrfs handle error by forcing the filesystem readonly */
 225         btrfs_set_sb_rdonly(sb);
 226         btrfs_info(fs_info, "forced readonly");
 227         /*
 228          * Note that a running device replace operation is not canceled here
 229          * although there is no way to update the progress. It would add the
 230          * risk of a deadlock, therefore the canceling is omitted. The only
 231          * penalty is that some I/O remains active until the procedure
 232          * completes. The next time when the filesystem is mounted writable
 233          * again, the device replace operation continues.
 234          */
 235 }
 236
 237 #ifdef CONFIG_PRINTK
 238 static const char * const logtypes[] = {
 239         "emergency",
 240         "alert",
 241         "critical",
 242         "error",
 243         "warning",
 244         "notice",
 245         "info",
 246         "debug",
 247 };
 248
 249
 250 /*
 251  * Use one ratelimit state per log level so that a flood of less important
 252  * messages doesn't cause more important ones to be dropped.
 253  */
 254 static struct ratelimit_state printk_limits[] = {
 255         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 256         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 257         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 258         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 259         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 260         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 261         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 262         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 263 };
 264
 265 void __cold _btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 266 {
 267         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 268         struct va_format vaf;
 269         va_list args;
 270         int kern_level;
 271         const char *type = logtypes[4];
 272         struct ratelimit_state *ratelimit = &printk_limits[4];
 273
 274         va_start(args, fmt);
 275
 276         while ((kern_level = printk_get_level(fmt)) != 0) {
 277                 size_t size = printk_skip_level(fmt) - fmt;
 278
 279                 if (kern_level >= '0' && kern_level <= '7') {
 280                         memcpy(lvl, fmt,  size);
 281                         lvl[size] = '\0';
 282                         type = logtypes[kern_level - '0'];
 283                         ratelimit = &printk_limits[kern_level - '0'];
 284                 }
 285                 fmt += size;
 286         }
 287
 288         vaf.fmt = fmt;
 289         vaf.va = &args;
 290
 291         if (__ratelimit(ratelimit)) {
 292                 if (fs_info) {
 293                         char statestr[STATE_STRING_BUF_LEN];
 294
 295                         btrfs_state_to_string(fs_info, statestr);
 296                         _printk("%sBTRFS %s (device %s%s): %pV\n", lvl, type,
 297                                 fs_info->sb->s_id, statestr, &vaf);
 298                 } else {
 299                         _printk("%sBTRFS %s: %pV\n", lvl, type, &vaf);
 300                 }
 301         }
 302
 303         va_end(args);
 304 }
 305 #endif
 306
 307 #if BITS_PER_LONG == 32
 308 void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info)
 309 {
 310         if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) {
 311                 btrfs_warn(fs_info, "reaching 32bit limit for logical addresses");
 312                 btrfs_warn(fs_info,
 313 "due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT",
 314                            BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 315                 btrfs_warn(fs_info,
 316                            "please consider upgrading to 64bit kernel/hardware");
 317         }
 318 }
 319
 320 void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info)
 321 {
 322         if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) {
 323                 btrfs_err(fs_info, "reached 32bit limit for logical addresses");
 324                 btrfs_err(fs_info,
 325 "due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed",
 326                           BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 327                 btrfs_err(fs_info,
 328                            "please consider upgrading to 64bit kernel/hardware");
 329         }
 330 }
 331 #endif
 332
 333 /*
 334  * We only mark the transaction aborted and then set the file system read-only.
 335  * This will prevent new transactions from starting or trying to join this
 336  * one.
 337  *
 338  * This means that error recovery at the call site is limited to freeing
 339  * any local memory allocations and passing the error code up without
 340  * further cleanup. The transaction should complete as it normally would
 341  * in the call path but will return -EIO.
 342  *
 343  * We'll complete the cleanup in btrfs_end_transaction and
 344  * btrfs_commit_transaction.
 345  */
 346 __cold
 347 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 348                                const char *function,
 349                                unsigned int line, int errno, bool first_hit)
 350 {
 351         struct btrfs_fs_info *fs_info = trans->fs_info;
 352
 353         WRITE_ONCE(trans->aborted, errno);
 354         WRITE_ONCE(trans->transaction->aborted, errno);
 355         if (first_hit && errno == -ENOSPC)
 356                 btrfs_dump_space_info_for_trans_abort(fs_info);
 357         /* Wake up anybody who may be waiting on this transaction */
 358         wake_up(&fs_info->transaction_wait);
 359         wake_up(&fs_info->transaction_blocked_wait);
 360         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 361 }
 362 /*
 363  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 364  * issues an alert, and either panics or BUGs, depending on mount options.
 365  */
 366 __cold
 367 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 368                    unsigned int line, int errno, const char *fmt, ...)
 369 {
 370         char *s_id = "<unknown>";
 371         const char *errstr;
 372         struct va_format vaf = { .fmt = fmt };
 373         va_list args;
 374
 375         if (fs_info)
 376                 s_id = fs_info->sb->s_id;
 377
 378         va_start(args, fmt);
 379         vaf.va = &args;
 380
 381         errstr = btrfs_decode_error(errno);
 382         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 383                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 384                         s_id, function, line, &vaf, errno, errstr);
 385
 386         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 387                    function, line, &vaf, errno, errstr);
 388         va_end(args);
 389         /* Caller calls BUG() */
 390 }
 391
 392 static void btrfs_put_super(struct super_block *sb)
 393 {
 394         close_ctree(btrfs_sb(sb));
 395 }
 396
 397 enum {
 398         Opt_acl, Opt_noacl,
 399         Opt_clear_cache,
 400         Opt_commit_interval,
 401         Opt_compress,
 402         Opt_compress_force,
 403         Opt_compress_force_type,
 404         Opt_compress_type,
 405         Opt_degraded,
 406         Opt_device,
 407         Opt_fatal_errors,
 408         Opt_flushoncommit, Opt_noflushoncommit,
 409         Opt_max_inline,
 410         Opt_barrier, Opt_nobarrier,
 411         Opt_datacow, Opt_nodatacow,
 412         Opt_datasum, Opt_nodatasum,
 413         Opt_defrag, Opt_nodefrag,
 414         Opt_discard, Opt_nodiscard,
 415         Opt_discard_mode,
 416         Opt_norecovery,
 417         Opt_ratio,
 418         Opt_rescan_uuid_tree,
 419         Opt_skip_balance,
 420         Opt_space_cache, Opt_no_space_cache,
 421         Opt_space_cache_version,
 422         Opt_ssd, Opt_nossd,
 423         Opt_ssd_spread, Opt_nossd_spread,
 424         Opt_subvol,
 425         Opt_subvol_empty,
 426         Opt_subvolid,
 427         Opt_thread_pool,
 428         Opt_treelog, Opt_notreelog,
 429         Opt_user_subvol_rm_allowed,
 430
 431         /* Rescue options */
 432         Opt_rescue,
 433         Opt_usebackuproot,
 434         Opt_nologreplay,
 435         Opt_ignorebadroots,
 436         Opt_ignoredatacsums,
 437         Opt_rescue_all,
 438
 439         /* Deprecated options */
 440         Opt_recovery,
 441         Opt_inode_cache, Opt_noinode_cache,
 442
 443         /* Debugging options */
 444         Opt_check_integrity,
 445         Opt_check_integrity_including_extent_data,
 446         Opt_check_integrity_print_mask,
 447         Opt_enospc_debug, Opt_noenospc_debug,
 448 #ifdef CONFIG_BTRFS_DEBUG
 449         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 450 #endif
 451 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 452         Opt_ref_verify,
 453 #endif
 454         Opt_err,
 455 };
 456
 457 static const match_table_t tokens = {
 458         {Opt_acl, "acl"},
 459         {Opt_noacl, "noacl"},
 460         {Opt_clear_cache, "clear_cache"},
 461         {Opt_commit_interval, "commit=%u"},
 462         {Opt_compress, "compress"},
 463         {Opt_compress_type, "compress=%s"},
 464         {Opt_compress_force, "compress-force"},
 465         {Opt_compress_force_type, "compress-force=%s"},
 466         {Opt_degraded, "degraded"},
 467         {Opt_device, "device=%s"},
 468         {Opt_fatal_errors, "fatal_errors=%s"},
 469         {Opt_flushoncommit, "flushoncommit"},
 470         {Opt_noflushoncommit, "noflushoncommit"},
 471         {Opt_inode_cache, "inode_cache"},
 472         {Opt_noinode_cache, "noinode_cache"},
 473         {Opt_max_inline, "max_inline=%s"},
 474         {Opt_barrier, "barrier"},
 475         {Opt_nobarrier, "nobarrier"},
 476         {Opt_datacow, "datacow"},
 477         {Opt_nodatacow, "nodatacow"},
 478         {Opt_datasum, "datasum"},
 479         {Opt_nodatasum, "nodatasum"},
 480         {Opt_defrag, "autodefrag"},
 481         {Opt_nodefrag, "noautodefrag"},
 482         {Opt_discard, "discard"},
 483         {Opt_discard_mode, "discard=%s"},
 484         {Opt_nodiscard, "nodiscard"},
 485         {Opt_norecovery, "norecovery"},
 486         {Opt_ratio, "metadata_ratio=%u"},
 487         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 488         {Opt_skip_balance, "skip_balance"},
 489         {Opt_space_cache, "space_cache"},
 490         {Opt_no_space_cache, "nospace_cache"},
 491         {Opt_space_cache_version, "space_cache=%s"},
 492         {Opt_ssd, "ssd"},
 493         {Opt_nossd, "nossd"},
 494         {Opt_ssd_spread, "ssd_spread"},
 495         {Opt_nossd_spread, "nossd_spread"},
 496         {Opt_subvol, "subvol=%s"},
 497         {Opt_subvol_empty, "subvol="},
 498         {Opt_subvolid, "subvolid=%s"},
 499         {Opt_thread_pool, "thread_pool=%u"},
 500         {Opt_treelog, "treelog"},
 501         {Opt_notreelog, "notreelog"},
 502         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 503
 504         /* Rescue options */
 505         {Opt_rescue, "rescue=%s"},
 506         /* Deprecated, with alias rescue=nologreplay */
 507         {Opt_nologreplay, "nologreplay"},
 508         /* Deprecated, with alias rescue=usebackuproot */
 509         {Opt_usebackuproot, "usebackuproot"},
 510
 511         /* Deprecated options */
 512         {Opt_recovery, "recovery"},
 513
 514         /* Debugging options */
 515         {Opt_check_integrity, "check_int"},
 516         {Opt_check_integrity_including_extent_data, "check_int_data"},
 517         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 518         {Opt_enospc_debug, "enospc_debug"},
 519         {Opt_noenospc_debug, "noenospc_debug"},
 520 #ifdef CONFIG_BTRFS_DEBUG
 521         {Opt_fragment_data, "fragment=data"},
 522         {Opt_fragment_metadata, "fragment=metadata"},
 523         {Opt_fragment_all, "fragment=all"},
 524 #endif
 525 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 526         {Opt_ref_verify, "ref_verify"},
 527 #endif
 528         {Opt_err, NULL},
 529 };
 530
 531 static const match_table_t rescue_tokens = {
 532         {Opt_usebackuproot, "usebackuproot"},
 533         {Opt_nologreplay, "nologreplay"},
 534         {Opt_ignorebadroots, "ignorebadroots"},
 535         {Opt_ignorebadroots, "ibadroots"},
 536         {Opt_ignoredatacsums, "ignoredatacsums"},
 537         {Opt_ignoredatacsums, "idatacsums"},
 538         {Opt_rescue_all, "all"},
 539         {Opt_err, NULL},
 540 };
 541
 542 static bool check_ro_option(struct btrfs_fs_info *fs_info, unsigned long opt,
 543                             const char *opt_name)
 544 {
 545         if (fs_info->mount_opt & opt) {
 546                 btrfs_err(fs_info, "%s must be used with ro mount option",
 547                           opt_name);
 548                 return true;
 549         }
 550         return false;
 551 }
 552
 553 static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
 554 {
 555         char *opts;
 556         char *orig;
 557         char *p;
 558         substring_t args[MAX_OPT_ARGS];
 559         int ret = 0;
 560
 561         opts = kstrdup(options, GFP_KERNEL);
 562         if (!opts)
 563                 return -ENOMEM;
 564         orig = opts;
 565
 566         while ((p = strsep(&opts, ":")) != NULL) {
 567                 int token;
 568
 569                 if (!*p)
 570                         continue;
 571                 token = match_token(p, rescue_tokens, args);
 572                 switch (token){
 573                 case Opt_usebackuproot:
 574                         btrfs_info(info,
 575                                    "trying to use backup root at mount time");
 576                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 577                         break;
 578                 case Opt_nologreplay:
 579                         btrfs_set_and_info(info, NOLOGREPLAY,
 580                                            "disabling log replay at mount time");
 581                         break;
 582                 case Opt_ignorebadroots:
 583                         btrfs_set_and_info(info, IGNOREBADROOTS,
 584                                            "ignoring bad roots");
 585                         break;
 586                 case Opt_ignoredatacsums:
 587                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 588                                            "ignoring data csums");
 589                         break;
 590                 case Opt_rescue_all:
 591                         btrfs_info(info, "enabling all of the rescue options");
 592                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 593                                            "ignoring data csums");
 594                         btrfs_set_and_info(info, IGNOREBADROOTS,
 595                                            "ignoring bad roots");
 596                         btrfs_set_and_info(info, NOLOGREPLAY,
 597                                            "disabling log replay at mount time");
 598                         break;
 599                 case Opt_err:
 600                         btrfs_info(info, "unrecognized rescue option '%s'", p);
 601                         ret = -EINVAL;
 602                         goto out;
 603                 default:
 604                         break;
 605                 }
 606
 607         }
 608 out:
 609         kfree(orig);
 610         return ret;
 611 }
 612
 613 /*
 614  * Regular mount options parser.  Everything that is needed only when
 615  * reading in a new superblock is parsed here.
 616  * XXX JDM: This needs to be cleaned up for remount.
 617  */
 618 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 619                         unsigned long new_flags)
 620 {
 621         substring_t args[MAX_OPT_ARGS];
 622         char *p, *num;
 623         int intarg;
 624         int ret = 0;
 625         char *compress_type;
 626         bool compress_force = false;
 627         enum btrfs_compression_type saved_compress_type;
 628         int saved_compress_level;
 629         bool saved_compress_force;
 630         int no_compress = 0;
 631         const bool remounting = test_bit(BTRFS_FS_STATE_REMOUNTING, &info->fs_state);
 632
 633         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 634                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 635         else if (btrfs_free_space_cache_v1_active(info)) {
 636                 if (btrfs_is_zoned(info)) {
 637                         btrfs_info(info,
 638                         "zoned: clearing existing space cache");
 639                         btrfs_set_super_cache_generation(info->super_copy, 0);
 640                 } else {
 641                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 642                 }
 643         }
 644
 645         /*
 646          * Even the options are empty, we still need to do extra check
 647          * against new flags
 648          */
 649         if (!options)
 650                 goto check;
 651
 652         while ((p = strsep(&options, ",")) != NULL) {
 653                 int token;
 654                 if (!*p)
 655                         continue;
 656
 657                 token = match_token(p, tokens, args);
 658                 switch (token) {
 659                 case Opt_degraded:
 660                         btrfs_info(info, "allowing degraded mounts");
 661                         btrfs_set_opt(info->mount_opt, DEGRADED);
 662                         break;
 663                 case Opt_subvol:
 664                 case Opt_subvol_empty:
 665                 case Opt_subvolid:
 666                 case Opt_device:
 667                         /*
 668                          * These are parsed by btrfs_parse_subvol_options or
 669                          * btrfs_parse_device_options and can be ignored here.
 670                          */
 671                         break;
 672                 case Opt_nodatasum:
 673                         btrfs_set_and_info(info, NODATASUM,
 674                                            "setting nodatasum");
 675                         break;
 676                 case Opt_datasum:
 677                         if (btrfs_test_opt(info, NODATASUM)) {
 678                                 if (btrfs_test_opt(info, NODATACOW))
 679                                         btrfs_info(info,
 680                                                    "setting datasum, datacow enabled");
 681                                 else
 682                                         btrfs_info(info, "setting datasum");
 683                         }
 684                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 685                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 686                         break;
 687                 case Opt_nodatacow:
 688                         if (!btrfs_test_opt(info, NODATACOW)) {
 689                                 if (!btrfs_test_opt(info, COMPRESS) ||
 690                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 691                                         btrfs_info(info,
 692                                                    "setting nodatacow, compression disabled");
 693                                 } else {
 694                                         btrfs_info(info, "setting nodatacow");
 695                                 }
 696                         }
 697                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 698                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 699                         btrfs_set_opt(info->mount_opt, NODATACOW);
 700                         btrfs_set_opt(info->mount_opt, NODATASUM);
 701                         break;
 702                 case Opt_datacow:
 703                         btrfs_clear_and_info(info, NODATACOW,
 704                                              "setting datacow");
 705                         break;
 706                 case Opt_compress_force:
 707                 case Opt_compress_force_type:
 708                         compress_force = true;
 709                         fallthrough;
 710                 case Opt_compress:
 711                 case Opt_compress_type:
 712                         saved_compress_type = btrfs_test_opt(info,
 713                                                              COMPRESS) ?
 714                                 info->compress_type : BTRFS_COMPRESS_NONE;
 715                         saved_compress_force =
 716                                 btrfs_test_opt(info, FORCE_COMPRESS);
 717                         saved_compress_level = info->compress_level;
 718                         if (token == Opt_compress ||
 719                             token == Opt_compress_force ||
 720                             strncmp(args[0].from, "zlib", 4) == 0) {
 721                                 compress_type = "zlib";
 722
 723                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 724                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 725                                 /*
 726                                  * args[0] contains uninitialized data since
 727                                  * for these tokens we don't expect any
 728                                  * parameter.
 729                                  */
 730                                 if (token != Opt_compress &&
 731                                     token != Opt_compress_force)
 732                                         info->compress_level =
 733                                           btrfs_compress_str2level(
 734                                                         BTRFS_COMPRESS_ZLIB,
 735                                                         args[0].from + 4);
 736                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 737                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 738                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 739                                 no_compress = 0;
 740                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 741                                 compress_type = "lzo";
 742                                 info->compress_type = BTRFS_COMPRESS_LZO;
 743                                 info->compress_level = 0;
 744                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 745                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 746                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 747                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 748                                 no_compress = 0;
 749                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 750                                 compress_type = "zstd";
 751                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 752                                 info->compress_level =
 753                                         btrfs_compress_str2level(
 754                                                          BTRFS_COMPRESS_ZSTD,
 755                                                          args[0].from + 4);
 756                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 757                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 758                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 759                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 760                                 no_compress = 0;
 761                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 762                                 compress_type = "no";
 763                                 info->compress_level = 0;
 764                                 info->compress_type = 0;
 765                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 766                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 767                                 compress_force = false;
 768                                 no_compress++;
 769                         } else {
 770                                 btrfs_err(info, "unrecognized compression value %s",
 771                                           args[0].from);
 772                                 ret = -EINVAL;
 773                                 goto out;
 774                         }
 775
 776                         if (compress_force) {
 777                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 778                         } else {
 779                                 /*
 780                                  * If we remount from compress-force=xxx to
 781                                  * compress=xxx, we need clear FORCE_COMPRESS
 782                                  * flag, otherwise, there is no way for users
 783                                  * to disable forcible compression separately.
 784                                  */
 785                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 786                         }
 787                         if (no_compress == 1) {
 788                                 btrfs_info(info, "use no compression");
 789                         } else if ((info->compress_type != saved_compress_type) ||
 790                                    (compress_force != saved_compress_force) ||
 791                                    (info->compress_level != saved_compress_level)) {
 792                                 btrfs_info(info, "%s %s compression, level %d",
 793                                            (compress_force) ? "force" : "use",
 794                                            compress_type, info->compress_level);
 795                         }
 796                         compress_force = false;
 797                         break;
 798                 case Opt_ssd:
 799                         btrfs_set_and_info(info, SSD,
 800                                            "enabling ssd optimizations");
 801                         btrfs_clear_opt(info->mount_opt, NOSSD);
 802                         break;
 803                 case Opt_ssd_spread:
 804                         btrfs_set_and_info(info, SSD,
 805                                            "enabling ssd optimizations");
 806                         btrfs_set_and_info(info, SSD_SPREAD,
 807                                            "using spread ssd allocation scheme");
 808                         btrfs_clear_opt(info->mount_opt, NOSSD);
 809                         break;
 810                 case Opt_nossd:
 811                         btrfs_set_opt(info->mount_opt, NOSSD);
 812                         btrfs_clear_and_info(info, SSD,
 813                                              "not using ssd optimizations");
 814                         fallthrough;
 815                 case Opt_nossd_spread:
 816                         btrfs_clear_and_info(info, SSD_SPREAD,
 817                                              "not using spread ssd allocation scheme");
 818                         break;
 819                 case Opt_barrier:
 820                         btrfs_clear_and_info(info, NOBARRIER,
 821                                              "turning on barriers");
 822                         break;
 823                 case Opt_nobarrier:
 824                         btrfs_set_and_info(info, NOBARRIER,
 825                                            "turning off barriers");
 826                         break;
 827                 case Opt_thread_pool:
 828                         ret = match_int(&args[0], &intarg);
 829                         if (ret) {
 830                                 btrfs_err(info, "unrecognized thread_pool value %s",
 831                                           args[0].from);
 832                                 goto out;
 833                         } else if (intarg == 0) {
 834                                 btrfs_err(info, "invalid value 0 for thread_pool");
 835                                 ret = -EINVAL;
 836                                 goto out;
 837                         }
 838                         info->thread_pool_size = intarg;
 839                         break;
 840                 case Opt_max_inline:
 841                         num = match_strdup(&args[0]);
 842                         if (num) {
 843                                 info->max_inline = memparse(num, NULL);
 844                                 kfree(num);
 845
 846                                 if (info->max_inline) {
 847                                         info->max_inline = min_t(u64,
 848                                                 info->max_inline,
 849                                                 info->sectorsize);
 850                                 }
 851                                 btrfs_info(info, "max_inline at %llu",
 852                                            info->max_inline);
 853                         } else {
 854                                 ret = -ENOMEM;
 855                                 goto out;
 856                         }
 857                         break;
 858                 case Opt_acl:
 859 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 860                         info->sb->s_flags |= SB_POSIXACL;
 861                         break;
 862 #else
 863                         btrfs_err(info, "support for ACL not compiled in!");
 864                         ret = -EINVAL;
 865                         goto out;
 866 #endif
 867                 case Opt_noacl:
 868                         info->sb->s_flags &= ~SB_POSIXACL;
 869                         break;
 870                 case Opt_notreelog:
 871                         btrfs_set_and_info(info, NOTREELOG,
 872                                            "disabling tree log");
 873                         break;
 874                 case Opt_treelog:
 875                         btrfs_clear_and_info(info, NOTREELOG,
 876                                              "enabling tree log");
 877                         break;
 878                 case Opt_norecovery:
 879                 case Opt_nologreplay:
 880                         btrfs_warn(info,
 881                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 882                         btrfs_set_and_info(info, NOLOGREPLAY,
 883                                            "disabling log replay at mount time");
 884                         break;
 885                 case Opt_flushoncommit:
 886                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 887                                            "turning on flush-on-commit");
 888                         break;
 889                 case Opt_noflushoncommit:
 890                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 891                                              "turning off flush-on-commit");
 892                         break;
 893                 case Opt_ratio:
 894                         ret = match_int(&args[0], &intarg);
 895                         if (ret) {
 896                                 btrfs_err(info, "unrecognized metadata_ratio value %s",
 897                                           args[0].from);
 898                                 goto out;
 899                         }
 900                         info->metadata_ratio = intarg;
 901                         btrfs_info(info, "metadata ratio %u",
 902                                    info->metadata_ratio);
 903                         break;
 904                 case Opt_discard:
 905                 case Opt_discard_mode:
 906                         if (token == Opt_discard ||
 907                             strcmp(args[0].from, "sync") == 0) {
 908                                 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
 909                                 btrfs_set_and_info(info, DISCARD_SYNC,
 910                                                    "turning on sync discard");
 911                         } else if (strcmp(args[0].from, "async") == 0) {
 912                                 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
 913                                 btrfs_set_and_info(info, DISCARD_ASYNC,
 914                                                    "turning on async discard");
 915                         } else {
 916                                 btrfs_err(info, "unrecognized discard mode value %s",
 917                                           args[0].from);
 918                                 ret = -EINVAL;
 919                                 goto out;
 920                         }
 921                         break;
 922                 case Opt_nodiscard:
 923                         btrfs_clear_and_info(info, DISCARD_SYNC,
 924                                              "turning off discard");
 925                         btrfs_clear_and_info(info, DISCARD_ASYNC,
 926                                              "turning off async discard");
 927                         break;
 928                 case Opt_space_cache:
 929                 case Opt_space_cache_version:
 930                         /*
 931                          * We already set FREE_SPACE_TREE above because we have
 932                          * compat_ro(FREE_SPACE_TREE) set, and we aren't going
 933                          * to allow v1 to be set for extent tree v2, simply
 934                          * ignore this setting if we're extent tree v2.
 935                          */
 936                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 937                                 break;
 938                         if (token == Opt_space_cache ||
 939                             strcmp(args[0].from, "v1") == 0) {
 940                                 btrfs_clear_opt(info->mount_opt,
 941                                                 FREE_SPACE_TREE);
 942                                 btrfs_set_and_info(info, SPACE_CACHE,
 943                                            "enabling disk space caching");
 944                         } else if (strcmp(args[0].from, "v2") == 0) {
 945                                 btrfs_clear_opt(info->mount_opt,
 946                                                 SPACE_CACHE);
 947                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 948                                                    "enabling free space tree");
 949                         } else {
 950                                 btrfs_err(info, "unrecognized space_cache value %s",
 951                                           args[0].from);
 952                                 ret = -EINVAL;
 953                                 goto out;
 954                         }
 955                         break;
 956                 case Opt_rescan_uuid_tree:
 957                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 958                         break;
 959                 case Opt_no_space_cache:
 960                         /*
 961                          * We cannot operate without the free space tree with
 962                          * extent tree v2, ignore this option.
 963                          */
 964                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 965                                 break;
 966                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 967                                 btrfs_clear_and_info(info, SPACE_CACHE,
 968                                              "disabling disk space caching");
 969                         }
 970                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 971                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 972                                              "disabling free space tree");
 973                         }
 974                         break;
 975                 case Opt_inode_cache:
 976                 case Opt_noinode_cache:
 977                         btrfs_warn(info,
 978         "the 'inode_cache' option is deprecated and has no effect since 5.11");
 979                         break;
 980                 case Opt_clear_cache:
 981                         /*
 982                          * We cannot clear the free space tree with extent tree
 983                          * v2, ignore this option.
 984                          */
 985                         if (btrfs_fs_incompat(info, EXTENT_TREE_V2))
 986                                 break;
 987                         btrfs_set_and_info(info, CLEAR_CACHE,
 988                                            "force clearing of disk cache");
 989                         break;
 990                 case Opt_user_subvol_rm_allowed:
 991                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 992                         break;
 993                 case Opt_enospc_debug:
 994                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 995                         break;
 996                 case Opt_noenospc_debug:
 997                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 998                         break;
 999                 case Opt_defrag:
1000                         btrfs_set_and_info(info, AUTO_DEFRAG,
1001                                            "enabling auto defrag");
1002                         break;
1003                 case Opt_nodefrag:
1004                         btrfs_clear_and_info(info, AUTO_DEFRAG,
1005                                              "disabling auto defrag");
1006                         break;
1007                 case Opt_recovery:
1008                 case Opt_usebackuproot:
1009                         btrfs_warn(info,
1010                         "'%s' is deprecated, use 'rescue=usebackuproot' instead",
1011                                    token == Opt_recovery ? "recovery" :
1012                                    "usebackuproot");
1013                         btrfs_info(info,
1014                                    "trying to use backup root at mount time");
1015                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
1016                         break;
1017                 case Opt_skip_balance:
1018                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
1019                         break;
1020 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1021                 case Opt_check_integrity_including_extent_data:
1022                         btrfs_info(info,
1023                                    "enabling check integrity including extent data");
1024                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY_DATA);
1025                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
1026                         break;
1027                 case Opt_check_integrity:
1028                         btrfs_info(info, "enabling check integrity");
1029                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
1030                         break;
1031                 case Opt_check_integrity_print_mask:
1032                         ret = match_int(&args[0], &intarg);
1033                         if (ret) {
1034                                 btrfs_err(info,
1035                                 "unrecognized check_integrity_print_mask value %s",
1036                                         args[0].from);
1037                                 goto out;
1038                         }
1039                         info->check_integrity_print_mask = intarg;
1040                         btrfs_info(info, "check_integrity_print_mask 0x%x",
1041                                    info->check_integrity_print_mask);
1042                         break;
1043 #else
1044                 case Opt_check_integrity_including_extent_data:
1045                 case Opt_check_integrity:
1046                 case Opt_check_integrity_print_mask:
1047                         btrfs_err(info,
1048                                   "support for check_integrity* not compiled in!");
1049                         ret = -EINVAL;
1050                         goto out;
1051 #endif
1052                 case Opt_fatal_errors:
1053                         if (strcmp(args[0].from, "panic") == 0) {
1054                                 btrfs_set_opt(info->mount_opt,
1055                                               PANIC_ON_FATAL_ERROR);
1056                         } else if (strcmp(args[0].from, "bug") == 0) {
1057                                 btrfs_clear_opt(info->mount_opt,
1058                                               PANIC_ON_FATAL_ERROR);
1059                         } else {
1060                                 btrfs_err(info, "unrecognized fatal_errors value %s",
1061                                           args[0].from);
1062                                 ret = -EINVAL;
1063                                 goto out;
1064                         }
1065                         break;
1066                 case Opt_commit_interval:
1067                         intarg = 0;
1068                         ret = match_int(&args[0], &intarg);
1069                         if (ret) {
1070                                 btrfs_err(info, "unrecognized commit_interval value %s",
1071                                           args[0].from);
1072                                 ret = -EINVAL;
1073                                 goto out;
1074                         }
1075                         if (intarg == 0) {
1076                                 btrfs_info(info,
1077                                            "using default commit interval %us",
1078                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
1079                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
1080                         } else if (intarg > 300) {
1081                                 btrfs_warn(info, "excessive commit interval %d",
1082                                            intarg);
1083                         }
1084                         info->commit_interval = intarg;
1085                         break;
1086                 case Opt_rescue:
1087                         ret = parse_rescue_options(info, args[0].from);
1088                         if (ret < 0) {
1089                                 btrfs_err(info, "unrecognized rescue value %s",
1090                                           args[0].from);
1091                                 goto out;
1092                         }
1093                         break;
1094 #ifdef CONFIG_BTRFS_DEBUG
1095                 case Opt_fragment_all:
1096                         btrfs_info(info, "fragmenting all space");
1097                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1098                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
1099                         break;
1100                 case Opt_fragment_metadata:
1101                         btrfs_info(info, "fragmenting metadata");
1102                         btrfs_set_opt(info->mount_opt,
1103                                       FRAGMENT_METADATA);
1104                         break;
1105                 case Opt_fragment_data:
1106                         btrfs_info(info, "fragmenting data");
1107                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1108                         break;
1109 #endif
1110 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
1111                 case Opt_ref_verify:
1112                         btrfs_info(info, "doing ref verification");
1113                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
1114                         break;
1115 #endif
1116                 case Opt_err:
1117                         btrfs_err(info, "unrecognized mount option '%s'", p);
1118                         ret = -EINVAL;
1119                         goto out;
1120                 default:
1121                         break;
1122                 }
1123         }
1124 check:
1125         /* We're read-only, don't have to check. */
1126         if (new_flags & SB_RDONLY)
1127                 goto out;
1128
1129         if (check_ro_option(info, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
1130             check_ro_option(info, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
1131             check_ro_option(info, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums"))
1132                 ret = -EINVAL;
1133 out:
1134         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
1135             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
1136             !btrfs_test_opt(info, CLEAR_CACHE)) {
1137                 btrfs_err(info, "cannot disable free space tree");
1138                 ret = -EINVAL;
1139         }
1140         if (btrfs_fs_compat_ro(info, BLOCK_GROUP_TREE) &&
1141              !btrfs_test_opt(info, FREE_SPACE_TREE)) {
1142                 btrfs_err(info, "cannot disable free space tree with block-group-tree feature");
1143                 ret = -EINVAL;
1144         }
1145         if (!ret)
1146                 ret = btrfs_check_mountopts_zoned(info);
1147         if (!ret && !remounting) {
1148                 if (btrfs_test_opt(info, SPACE_CACHE))
1149                         btrfs_info(info, "disk space caching is enabled");
1150                 if (btrfs_test_opt(info, FREE_SPACE_TREE))
1151                         btrfs_info(info, "using free space tree");
1152         }
1153         return ret;
1154 }
1155
1156 /*
1157  * Parse mount options that are required early in the mount process.
1158  *
1159  * All other options will be parsed on much later in the mount process and
1160  * only when we need to allocate a new super block.
1161  */
1162 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1163                                       void *holder)
1164 {
1165         substring_t args[MAX_OPT_ARGS];
1166         char *device_name, *opts, *orig, *p;
1167         struct btrfs_device *device = NULL;
1168         int error = 0;
1169
1170         lockdep_assert_held(&uuid_mutex);
1171
1172         if (!options)
1173                 return 0;
1174
1175         /*
1176          * strsep changes the string, duplicate it because btrfs_parse_options
1177          * gets called later
1178          */
1179         opts = kstrdup(options, GFP_KERNEL);
1180         if (!opts)
1181                 return -ENOMEM;
1182         orig = opts;
1183
1184         while ((p = strsep(&opts, ",")) != NULL) {
1185                 int token;
1186
1187                 if (!*p)
1188                         continue;
1189
1190                 token = match_token(p, tokens, args);
1191                 if (token == Opt_device) {
1192                         device_name = match_strdup(&args[0]);
1193                         if (!device_name) {
1194                                 error = -ENOMEM;
1195                                 goto out;
1196                         }
1197                         device = btrfs_scan_one_device(device_name, flags,
1198                                         holder);
1199                         kfree(device_name);
1200                         if (IS_ERR(device)) {
1201                                 error = PTR_ERR(device);
1202                                 goto out;
1203                         }
1204                 }
1205         }
1206
1207 out:
1208         kfree(orig);
1209         return error;
1210 }
1211
1212 /*
1213  * Parse mount options that are related to subvolume id
1214  *
1215  * The value is later passed to mount_subvol()
1216  */
1217 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1218                 u64 *subvol_objectid)
1219 {
1220         substring_t args[MAX_OPT_ARGS];
1221         char *opts, *orig, *p;
1222         int error = 0;
1223         u64 subvolid;
1224
1225         if (!options)
1226                 return 0;
1227
1228         /*
1229          * strsep changes the string, duplicate it because
1230          * btrfs_parse_device_options gets called later
1231          */
1232         opts = kstrdup(options, GFP_KERNEL);
1233         if (!opts)
1234                 return -ENOMEM;
1235         orig = opts;
1236
1237         while ((p = strsep(&opts, ",")) != NULL) {
1238                 int token;
1239                 if (!*p)
1240                         continue;
1241
1242                 token = match_token(p, tokens, args);
1243                 switch (token) {
1244                 case Opt_subvol:
1245                         kfree(*subvol_name);
1246                         *subvol_name = match_strdup(&args[0]);
1247                         if (!*subvol_name) {
1248                                 error = -ENOMEM;
1249                                 goto out;
1250                         }
1251                         break;
1252                 case Opt_subvolid:
1253                         error = match_u64(&args[0], &subvolid);
1254                         if (error)
1255                                 goto out;
1256
1257                         /* we want the original fs_tree */
1258                         if (subvolid == 0)
1259                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1260
1261                         *subvol_objectid = subvolid;
1262                         break;
1263                 default:
1264                         break;
1265                 }
1266         }
1267
1268 out:
1269         kfree(orig);
1270         return error;
1271 }
1272
1273 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1274                                           u64 subvol_objectid)
1275 {
1276         struct btrfs_root *root = fs_info->tree_root;
1277         struct btrfs_root *fs_root = NULL;
1278         struct btrfs_root_ref *root_ref;
1279         struct btrfs_inode_ref *inode_ref;
1280         struct btrfs_key key;
1281         struct btrfs_path *path = NULL;
1282         char *name = NULL, *ptr;
1283         u64 dirid;
1284         int len;
1285         int ret;
1286
1287         path = btrfs_alloc_path();
1288         if (!path) {
1289                 ret = -ENOMEM;
1290                 goto err;
1291         }
1292
1293         name = kmalloc(PATH_MAX, GFP_KERNEL);
1294         if (!name) {
1295                 ret = -ENOMEM;
1296                 goto err;
1297         }
1298         ptr = name + PATH_MAX - 1;
1299         ptr[0] = '\0';
1300
1301         /*
1302          * Walk up the subvolume trees in the tree of tree roots by root
1303          * backrefs until we hit the top-level subvolume.
1304          */
1305         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1306                 key.objectid = subvol_objectid;
1307                 key.type = BTRFS_ROOT_BACKREF_KEY;
1308                 key.offset = (u64)-1;
1309
1310                 ret = btrfs_search_backwards(root, &key, path);
1311                 if (ret < 0) {
1312                         goto err;
1313                 } else if (ret > 0) {
1314                         ret = -ENOENT;
1315                         goto err;
1316                 }
1317
1318                 subvol_objectid = key.offset;
1319
1320                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1321                                           struct btrfs_root_ref);
1322                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1323                 ptr -= len + 1;
1324                 if (ptr < name) {
1325                         ret = -ENAMETOOLONG;
1326                         goto err;
1327                 }
1328                 read_extent_buffer(path->nodes[0], ptr + 1,
1329                                    (unsigned long)(root_ref + 1), len);
1330                 ptr[0] = '/';
1331                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1332                 btrfs_release_path(path);
1333
1334                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1335                 if (IS_ERR(fs_root)) {
1336                         ret = PTR_ERR(fs_root);
1337                         fs_root = NULL;
1338                         goto err;
1339                 }
1340
1341                 /*
1342                  * Walk up the filesystem tree by inode refs until we hit the
1343                  * root directory.
1344                  */
1345                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1346                         key.objectid = dirid;
1347                         key.type = BTRFS_INODE_REF_KEY;
1348                         key.offset = (u64)-1;
1349
1350                         ret = btrfs_search_backwards(fs_root, &key, path);
1351                         if (ret < 0) {
1352                                 goto err;
1353                         } else if (ret > 0) {
1354                                 ret = -ENOENT;
1355                                 goto err;
1356                         }
1357
1358                         dirid = key.offset;
1359
1360                         inode_ref = btrfs_item_ptr(path->nodes[0],
1361                                                    path->slots[0],
1362                                                    struct btrfs_inode_ref);
1363                         len = btrfs_inode_ref_name_len(path->nodes[0],
1364                                                        inode_ref);
1365                         ptr -= len + 1;
1366                         if (ptr < name) {
1367                                 ret = -ENAMETOOLONG;
1368                                 goto err;
1369                         }
1370                         read_extent_buffer(path->nodes[0], ptr + 1,
1371                                            (unsigned long)(inode_ref + 1), len);
1372                         ptr[0] = '/';
1373                         btrfs_release_path(path);
1374                 }
1375                 btrfs_put_root(fs_root);
1376                 fs_root = NULL;
1377         }
1378
1379         btrfs_free_path(path);
1380         if (ptr == name + PATH_MAX - 1) {
1381                 name[0] = '/';
1382                 name[1] = '\0';
1383         } else {
1384                 memmove(name, ptr, name + PATH_MAX - ptr);
1385         }
1386         return name;
1387
1388 err:
1389         btrfs_put_root(fs_root);
1390         btrfs_free_path(path);
1391         kfree(name);
1392         return ERR_PTR(ret);
1393 }
1394
1395 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1396 {
1397         struct btrfs_root *root = fs_info->tree_root;
1398         struct btrfs_dir_item *di;
1399         struct btrfs_path *path;
1400         struct btrfs_key location;
1401         u64 dir_id;
1402
1403         path = btrfs_alloc_path();
1404         if (!path)
1405                 return -ENOMEM;
1406
1407         /*
1408          * Find the "default" dir item which points to the root item that we
1409          * will mount by default if we haven't been given a specific subvolume
1410          * to mount.
1411          */
1412         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1413         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1414         if (IS_ERR(di)) {
1415                 btrfs_free_path(path);
1416                 return PTR_ERR(di);
1417         }
1418         if (!di) {
1419                 /*
1420                  * Ok the default dir item isn't there.  This is weird since
1421                  * it's always been there, but don't freak out, just try and
1422                  * mount the top-level subvolume.
1423                  */
1424                 btrfs_free_path(path);
1425                 *objectid = BTRFS_FS_TREE_OBJECTID;
1426                 return 0;
1427         }
1428
1429         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1430         btrfs_free_path(path);
1431         *objectid = location.objectid;
1432         return 0;
1433 }
1434
1435 static int btrfs_fill_super(struct super_block *sb,
1436                             struct btrfs_fs_devices *fs_devices,
1437                             void *data)
1438 {
1439         struct inode *inode;
1440         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1441         int err;
1442
1443         sb->s_maxbytes = MAX_LFS_FILESIZE;
1444         sb->s_magic = BTRFS_SUPER_MAGIC;
1445         sb->s_op = &btrfs_super_ops;
1446         sb->s_d_op = &btrfs_dentry_operations;
1447         sb->s_export_op = &btrfs_export_ops;
1448 #ifdef CONFIG_FS_VERITY
1449         sb->s_vop = &btrfs_verityops;
1450 #endif
1451         sb->s_xattr = btrfs_xattr_handlers;
1452         sb->s_time_gran = 1;
1453 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1454         sb->s_flags |= SB_POSIXACL;
1455 #endif
1456         sb->s_flags |= SB_I_VERSION;
1457         sb->s_iflags |= SB_I_CGROUPWB;
1458
1459         err = super_setup_bdi(sb);
1460         if (err) {
1461                 btrfs_err(fs_info, "super_setup_bdi failed");
1462                 return err;
1463         }
1464
1465         err = open_ctree(sb, fs_devices, (char *)data);
1466         if (err) {
1467                 btrfs_err(fs_info, "open_ctree failed");
1468                 return err;
1469         }
1470
1471         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1472         if (IS_ERR(inode)) {
1473                 err = PTR_ERR(inode);
1474                 goto fail_close;
1475         }
1476
1477         sb->s_root = d_make_root(inode);
1478         if (!sb->s_root) {
1479                 err = -ENOMEM;
1480                 goto fail_close;
1481         }
1482
1483         sb->s_flags |= SB_ACTIVE;
1484         return 0;
1485
1486 fail_close:
1487         close_ctree(fs_info);
1488         return err;
1489 }
1490
1491 int btrfs_sync_fs(struct super_block *sb, int wait)
1492 {
1493         struct btrfs_trans_handle *trans;
1494         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1495         struct btrfs_root *root = fs_info->tree_root;
1496
1497         trace_btrfs_sync_fs(fs_info, wait);
1498
1499         if (!wait) {
1500                 filemap_flush(fs_info->btree_inode->i_mapping);
1501                 return 0;
1502         }
1503
1504         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1505
1506         trans = btrfs_attach_transaction_barrier(root);
1507         if (IS_ERR(trans)) {
1508                 /* no transaction, don't bother */
1509                 if (PTR_ERR(trans) == -ENOENT) {
1510                         /*
1511                          * Exit unless we have some pending changes
1512                          * that need to go through commit
1513                          */
1514                         if (fs_info->pending_changes == 0)
1515                                 return 0;
1516                         /*
1517                          * A non-blocking test if the fs is frozen. We must not
1518                          * start a new transaction here otherwise a deadlock
1519                          * happens. The pending operations are delayed to the
1520                          * next commit after thawing.
1521                          */
1522                         if (sb_start_write_trylock(sb))
1523                                 sb_end_write(sb);
1524                         else
1525                                 return 0;
1526                         trans = btrfs_start_transaction(root, 0);
1527                 }
1528                 if (IS_ERR(trans))
1529                         return PTR_ERR(trans);
1530         }
1531         return btrfs_commit_transaction(trans);
1532 }
1533
1534 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1535 {
1536         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1537         *printed = true;
1538 }
1539
1540 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1541 {
1542         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1543         const char *compress_type;
1544         const char *subvol_name;
1545         bool printed = false;
1546
1547         if (btrfs_test_opt(info, DEGRADED))
1548                 seq_puts(seq, ",degraded");
1549         if (btrfs_test_opt(info, NODATASUM))
1550                 seq_puts(seq, ",nodatasum");
1551         if (btrfs_test_opt(info, NODATACOW))
1552                 seq_puts(seq, ",nodatacow");
1553         if (btrfs_test_opt(info, NOBARRIER))
1554                 seq_puts(seq, ",nobarrier");
1555         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1556                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1557         if (info->thread_pool_size !=  min_t(unsigned long,
1558                                              num_online_cpus() + 2, 8))
1559                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1560         if (btrfs_test_opt(info, COMPRESS)) {
1561                 compress_type = btrfs_compress_type2str(info->compress_type);
1562                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1563                         seq_printf(seq, ",compress-force=%s", compress_type);
1564                 else
1565                         seq_printf(seq, ",compress=%s", compress_type);
1566                 if (info->compress_level)
1567                         seq_printf(seq, ":%d", info->compress_level);
1568         }
1569         if (btrfs_test_opt(info, NOSSD))
1570                 seq_puts(seq, ",nossd");
1571         if (btrfs_test_opt(info, SSD_SPREAD))
1572                 seq_puts(seq, ",ssd_spread");
1573         else if (btrfs_test_opt(info, SSD))
1574                 seq_puts(seq, ",ssd");
1575         if (btrfs_test_opt(info, NOTREELOG))
1576                 seq_puts(seq, ",notreelog");
1577         if (btrfs_test_opt(info, NOLOGREPLAY))
1578                 print_rescue_option(seq, "nologreplay", &printed);
1579         if (btrfs_test_opt(info, USEBACKUPROOT))
1580                 print_rescue_option(seq, "usebackuproot", &printed);
1581         if (btrfs_test_opt(info, IGNOREBADROOTS))
1582                 print_rescue_option(seq, "ignorebadroots", &printed);
1583         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1584                 print_rescue_option(seq, "ignoredatacsums", &printed);
1585         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1586                 seq_puts(seq, ",flushoncommit");
1587         if (btrfs_test_opt(info, DISCARD_SYNC))
1588                 seq_puts(seq, ",discard");
1589         if (btrfs_test_opt(info, DISCARD_ASYNC))
1590                 seq_puts(seq, ",discard=async");
1591         if (!(info->sb->s_flags & SB_POSIXACL))
1592                 seq_puts(seq, ",noacl");
1593         if (btrfs_free_space_cache_v1_active(info))
1594                 seq_puts(seq, ",space_cache");
1595         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1596                 seq_puts(seq, ",space_cache=v2");
1597         else
1598                 seq_puts(seq, ",nospace_cache");
1599         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1600                 seq_puts(seq, ",rescan_uuid_tree");
1601         if (btrfs_test_opt(info, CLEAR_CACHE))
1602                 seq_puts(seq, ",clear_cache");
1603         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1604                 seq_puts(seq, ",user_subvol_rm_allowed");
1605         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1606                 seq_puts(seq, ",enospc_debug");
1607         if (btrfs_test_opt(info, AUTO_DEFRAG))
1608                 seq_puts(seq, ",autodefrag");
1609         if (btrfs_test_opt(info, SKIP_BALANCE))
1610                 seq_puts(seq, ",skip_balance");
1611 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1612         if (btrfs_test_opt(info, CHECK_INTEGRITY_DATA))
1613                 seq_puts(seq, ",check_int_data");
1614         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1615                 seq_puts(seq, ",check_int");
1616         if (info->check_integrity_print_mask)
1617                 seq_printf(seq, ",check_int_print_mask=%d",
1618                                 info->check_integrity_print_mask);
1619 #endif
1620         if (info->metadata_ratio)
1621                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1622         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1623                 seq_puts(seq, ",fatal_errors=panic");
1624         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1625                 seq_printf(seq, ",commit=%u", info->commit_interval);
1626 #ifdef CONFIG_BTRFS_DEBUG
1627         if (btrfs_test_opt(info, FRAGMENT_DATA))
1628                 seq_puts(seq, ",fragment=data");
1629         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1630                 seq_puts(seq, ",fragment=metadata");
1631 #endif
1632         if (btrfs_test_opt(info, REF_VERIFY))
1633                 seq_puts(seq, ",ref_verify");
1634         seq_printf(seq, ",subvolid=%llu",
1635                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1636         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1637                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1638         if (!IS_ERR(subvol_name)) {
1639                 seq_puts(seq, ",subvol=");
1640                 seq_escape(seq, subvol_name, " \t\n\\");
1641                 kfree(subvol_name);
1642         }
1643         return 0;
1644 }
1645
1646 static int btrfs_test_super(struct super_block *s, void *data)
1647 {
1648         struct btrfs_fs_info *p = data;
1649         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1650
1651         return fs_info->fs_devices == p->fs_devices;
1652 }
1653
1654 static int btrfs_set_super(struct super_block *s, void *data)
1655 {
1656         int err = set_anon_super(s, data);
1657         if (!err)
1658                 s->s_fs_info = data;
1659         return err;
1660 }
1661
1662 /*
1663  * subvolumes are identified by ino 256
1664  */
1665 static inline int is_subvolume_inode(struct inode *inode)
1666 {
1667         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1668                 return 1;
1669         return 0;
1670 }
1671
1672 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1673                                    struct vfsmount *mnt)
1674 {
1675         struct dentry *root;
1676         int ret;
1677
1678         if (!subvol_name) {
1679                 if (!subvol_objectid) {
1680                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1681                                                           &subvol_objectid);
1682                         if (ret) {
1683                                 root = ERR_PTR(ret);
1684                                 goto out;
1685                         }
1686                 }
1687                 subvol_name = btrfs_get_subvol_name_from_objectid(
1688                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1689                 if (IS_ERR(subvol_name)) {
1690                         root = ERR_CAST(subvol_name);
1691                         subvol_name = NULL;
1692                         goto out;
1693                 }
1694
1695         }
1696
1697         root = mount_subtree(mnt, subvol_name);
1698         /* mount_subtree() drops our reference on the vfsmount. */
1699         mnt = NULL;
1700
1701         if (!IS_ERR(root)) {
1702                 struct super_block *s = root->d_sb;
1703                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1704                 struct inode *root_inode = d_inode(root);
1705                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1706
1707                 ret = 0;
1708                 if (!is_subvolume_inode(root_inode)) {
1709                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1710                                subvol_name);
1711                         ret = -EINVAL;
1712                 }
1713                 if (subvol_objectid && root_objectid != subvol_objectid) {
1714                         /*
1715                          * This will also catch a race condition where a
1716                          * subvolume which was passed by ID is renamed and
1717                          * another subvolume is renamed over the old location.
1718                          */
1719                         btrfs_err(fs_info,
1720                                   "subvol '%s' does not match subvolid %llu",
1721                                   subvol_name, subvol_objectid);
1722                         ret = -EINVAL;
1723                 }
1724                 if (ret) {
1725                         dput(root);
1726                         root = ERR_PTR(ret);
1727                         deactivate_locked_super(s);
1728                 }
1729         }
1730
1731 out:
1732         mntput(mnt);
1733         kfree(subvol_name);
1734         return root;
1735 }
1736
1737 /*
1738  * Find a superblock for the given device / mount point.
1739  *
1740  * Note: This is based on mount_bdev from fs/super.c with a few additions
1741  *       for multiple device setup.  Make sure to keep it in sync.
1742  */
1743 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1744                 int flags, const char *device_name, void *data)
1745 {
1746         struct block_device *bdev = NULL;
1747         struct super_block *s;
1748         struct btrfs_device *device = NULL;
1749         struct btrfs_fs_devices *fs_devices = NULL;
1750         struct btrfs_fs_info *fs_info = NULL;
1751         void *new_sec_opts = NULL;
1752         fmode_t mode = FMODE_READ;
1753         int error = 0;
1754
1755         if (!(flags & SB_RDONLY))
1756                 mode |= FMODE_WRITE;
1757
1758         if (data) {
1759                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1760                 if (error)
1761                         return ERR_PTR(error);
1762         }
1763
1764         /*
1765          * Setup a dummy root and fs_info for test/set super.  This is because
1766          * we don't actually fill this stuff out until open_ctree, but we need
1767          * then open_ctree will properly initialize the file system specific
1768          * settings later.  btrfs_init_fs_info initializes the static elements
1769          * of the fs_info (locks and such) to make cleanup easier if we find a
1770          * superblock with our given fs_devices later on at sget() time.
1771          */
1772         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1773         if (!fs_info) {
1774                 error = -ENOMEM;
1775                 goto error_sec_opts;
1776         }
1777         btrfs_init_fs_info(fs_info);
1778
1779         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1780         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1781         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1782                 error = -ENOMEM;
1783                 goto error_fs_info;
1784         }
1785
1786         mutex_lock(&uuid_mutex);
1787         error = btrfs_parse_device_options(data, mode, fs_type);
1788         if (error) {
1789                 mutex_unlock(&uuid_mutex);
1790                 goto error_fs_info;
1791         }
1792
1793         device = btrfs_scan_one_device(device_name, mode, fs_type);
1794         if (IS_ERR(device)) {
1795                 mutex_unlock(&uuid_mutex);
1796                 error = PTR_ERR(device);
1797                 goto error_fs_info;
1798         }
1799
1800         fs_devices = device->fs_devices;
1801         fs_info->fs_devices = fs_devices;
1802
1803         error = btrfs_open_devices(fs_devices, mode, fs_type);
1804         mutex_unlock(&uuid_mutex);
1805         if (error)
1806                 goto error_fs_info;
1807
1808         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1809                 error = -EACCES;
1810                 goto error_close_devices;
1811         }
1812
1813         bdev = fs_devices->latest_dev->bdev;
1814         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1815                  fs_info);
1816         if (IS_ERR(s)) {
1817                 error = PTR_ERR(s);
1818                 goto error_close_devices;
1819         }
1820
1821         if (s->s_root) {
1822                 btrfs_close_devices(fs_devices);
1823                 btrfs_free_fs_info(fs_info);
1824                 if ((flags ^ s->s_flags) & SB_RDONLY)
1825                         error = -EBUSY;
1826         } else {
1827                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1828                 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s", fs_type->name,
1829                                         s->s_id);
1830                 btrfs_sb(s)->bdev_holder = fs_type;
1831                 error = btrfs_fill_super(s, fs_devices, data);
1832         }
1833         if (!error)
1834                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1835         security_free_mnt_opts(&new_sec_opts);
1836         if (error) {
1837                 deactivate_locked_super(s);
1838                 return ERR_PTR(error);
1839         }
1840
1841         return dget(s->s_root);
1842
1843 error_close_devices:
1844         btrfs_close_devices(fs_devices);
1845 error_fs_info:
1846         btrfs_free_fs_info(fs_info);
1847 error_sec_opts:
1848         security_free_mnt_opts(&new_sec_opts);
1849         return ERR_PTR(error);
1850 }
1851
1852 /*
1853  * Mount function which is called by VFS layer.
1854  *
1855  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1856  * which needs vfsmount* of device's root (/).  This means device's root has to
1857  * be mounted internally in any case.
1858  *
1859  * Operation flow:
1860  *   1. Parse subvol id related options for later use in mount_subvol().
1861  *
1862  *   2. Mount device's root (/) by calling vfs_kern_mount().
1863  *
1864  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1865  *      first place. In order to avoid calling btrfs_mount() again, we use
1866  *      different file_system_type which is not registered to VFS by
1867  *      register_filesystem() (btrfs_root_fs_type). As a result,
1868  *      btrfs_mount_root() is called. The return value will be used by
1869  *      mount_subtree() in mount_subvol().
1870  *
1871  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1872  *      "btrfs subvolume set-default", mount_subvol() is called always.
1873  */
1874 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1875                 const char *device_name, void *data)
1876 {
1877         struct vfsmount *mnt_root;
1878         struct dentry *root;
1879         char *subvol_name = NULL;
1880         u64 subvol_objectid = 0;
1881         int error = 0;
1882
1883         error = btrfs_parse_subvol_options(data, &subvol_name,
1884                                         &subvol_objectid);
1885         if (error) {
1886                 kfree(subvol_name);
1887                 return ERR_PTR(error);
1888         }
1889
1890         /* mount device's root (/) */
1891         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1892         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1893                 if (flags & SB_RDONLY) {
1894                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1895                                 flags & ~SB_RDONLY, device_name, data);
1896                 } else {
1897                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1898                                 flags | SB_RDONLY, device_name, data);
1899                         if (IS_ERR(mnt_root)) {
1900                                 root = ERR_CAST(mnt_root);
1901                                 kfree(subvol_name);
1902                                 goto out;
1903                         }
1904
1905                         down_write(&mnt_root->mnt_sb->s_umount);
1906                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1907                         up_write(&mnt_root->mnt_sb->s_umount);
1908                         if (error < 0) {
1909                                 root = ERR_PTR(error);
1910                                 mntput(mnt_root);
1911                                 kfree(subvol_name);
1912                                 goto out;
1913                         }
1914                 }
1915         }
1916         if (IS_ERR(mnt_root)) {
1917                 root = ERR_CAST(mnt_root);
1918                 kfree(subvol_name);
1919                 goto out;
1920         }
1921
1922         /* mount_subvol() will free subvol_name and mnt_root */
1923         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1924
1925 out:
1926         return root;
1927 }
1928
1929 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1930                                      u32 new_pool_size, u32 old_pool_size)
1931 {
1932         if (new_pool_size == old_pool_size)
1933                 return;
1934
1935         fs_info->thread_pool_size = new_pool_size;
1936
1937         btrfs_info(fs_info, "resize thread pool %d -> %d",
1938                old_pool_size, new_pool_size);
1939
1940         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1941         btrfs_workqueue_set_max(fs_info->hipri_workers, new_pool_size);
1942         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1943         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1944         workqueue_set_max_active(fs_info->endio_workers, new_pool_size);
1945         workqueue_set_max_active(fs_info->endio_meta_workers, new_pool_size);
1946         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1947         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1948         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1949 }
1950
1951 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1952                                        unsigned long old_opts, int flags)
1953 {
1954         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1955             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1956              (flags & SB_RDONLY))) {
1957                 /* wait for any defraggers to finish */
1958                 wait_event(fs_info->transaction_wait,
1959                            (atomic_read(&fs_info->defrag_running) == 0));
1960                 if (flags & SB_RDONLY)
1961                         sync_filesystem(fs_info->sb);
1962         }
1963 }
1964
1965 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1966                                          unsigned long old_opts)
1967 {
1968         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1969
1970         /*
1971          * We need to cleanup all defragable inodes if the autodefragment is
1972          * close or the filesystem is read only.
1973          */
1974         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1975             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1976                 btrfs_cleanup_defrag_inodes(fs_info);
1977         }
1978
1979         /* If we toggled discard async */
1980         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1981             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1982                 btrfs_discard_resume(fs_info);
1983         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1984                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1985                 btrfs_discard_cleanup(fs_info);
1986
1987         /* If we toggled space cache */
1988         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1989                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1990 }
1991
1992 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1993 {
1994         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1995         unsigned old_flags = sb->s_flags;
1996         unsigned long old_opts = fs_info->mount_opt;
1997         unsigned long old_compress_type = fs_info->compress_type;
1998         u64 old_max_inline = fs_info->max_inline;
1999         u32 old_thread_pool_size = fs_info->thread_pool_size;
2000         u32 old_metadata_ratio = fs_info->metadata_ratio;
2001         int ret;
2002
2003         sync_filesystem(sb);
2004         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2005
2006         if (data) {
2007                 void *new_sec_opts = NULL;
2008
2009                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
2010                 if (!ret)
2011                         ret = security_sb_remount(sb, new_sec_opts);
2012                 security_free_mnt_opts(&new_sec_opts);
2013                 if (ret)
2014                         goto restore;
2015         }
2016
2017         ret = btrfs_parse_options(fs_info, data, *flags);
2018         if (ret)
2019                 goto restore;
2020
2021         ret = btrfs_check_features(fs_info, !(*flags & SB_RDONLY));
2022         if (ret < 0)
2023                 goto restore;
2024
2025         btrfs_remount_begin(fs_info, old_opts, *flags);
2026         btrfs_resize_thread_pool(fs_info,
2027                 fs_info->thread_pool_size, old_thread_pool_size);
2028
2029         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
2030             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2031             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
2032                 btrfs_warn(fs_info,
2033                 "remount supports changing free space tree only from ro to rw");
2034                 /* Make sure free space cache options match the state on disk */
2035                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2036                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2037                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
2038                 }
2039                 if (btrfs_free_space_cache_v1_active(fs_info)) {
2040                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2041                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
2042                 }
2043         }
2044
2045         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
2046                 goto out;
2047
2048         if (*flags & SB_RDONLY) {
2049                 /*
2050                  * this also happens on 'umount -rf' or on shutdown, when
2051                  * the filesystem is busy.
2052                  */
2053                 cancel_work_sync(&fs_info->async_reclaim_work);
2054                 cancel_work_sync(&fs_info->async_data_reclaim_work);
2055
2056                 btrfs_discard_cleanup(fs_info);
2057
2058                 /* wait for the uuid_scan task to finish */
2059                 down(&fs_info->uuid_tree_rescan_sem);
2060                 /* avoid complains from lockdep et al. */
2061                 up(&fs_info->uuid_tree_rescan_sem);
2062
2063                 btrfs_set_sb_rdonly(sb);
2064
2065                 /*
2066                  * Setting SB_RDONLY will put the cleaner thread to
2067                  * sleep at the next loop if it's already active.
2068                  * If it's already asleep, we'll leave unused block
2069                  * groups on disk until we're mounted read-write again
2070                  * unless we clean them up here.
2071                  */
2072                 btrfs_delete_unused_bgs(fs_info);
2073
2074                 /*
2075                  * The cleaner task could be already running before we set the
2076                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
2077                  * We must make sure that after we finish the remount, i.e. after
2078                  * we call btrfs_commit_super(), the cleaner can no longer start
2079                  * a transaction - either because it was dropping a dead root,
2080                  * running delayed iputs or deleting an unused block group (the
2081                  * cleaner picked a block group from the list of unused block
2082                  * groups before we were able to in the previous call to
2083                  * btrfs_delete_unused_bgs()).
2084                  */
2085                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
2086                             TASK_UNINTERRUPTIBLE);
2087
2088                 /*
2089                  * We've set the superblock to RO mode, so we might have made
2090                  * the cleaner task sleep without running all pending delayed
2091                  * iputs. Go through all the delayed iputs here, so that if an
2092                  * unmount happens without remounting RW we don't end up at
2093                  * finishing close_ctree() with a non-empty list of delayed
2094                  * iputs.
2095                  */
2096                 btrfs_run_delayed_iputs(fs_info);
2097
2098                 btrfs_dev_replace_suspend_for_unmount(fs_info);
2099                 btrfs_scrub_cancel(fs_info);
2100                 btrfs_pause_balance(fs_info);
2101
2102                 /*
2103                  * Pause the qgroup rescan worker if it is running. We don't want
2104                  * it to be still running after we are in RO mode, as after that,
2105                  * by the time we unmount, it might have left a transaction open,
2106                  * so we would leak the transaction and/or crash.
2107                  */
2108                 btrfs_qgroup_wait_for_completion(fs_info, false);
2109
2110                 ret = btrfs_commit_super(fs_info);
2111                 if (ret)
2112                         goto restore;
2113         } else {
2114                 if (BTRFS_FS_ERROR(fs_info)) {
2115                         btrfs_err(fs_info,
2116                                 "Remounting read-write after error is not allowed");
2117                         ret = -EINVAL;
2118                         goto restore;
2119                 }
2120                 if (fs_info->fs_devices->rw_devices == 0) {
2121                         ret = -EACCES;
2122                         goto restore;
2123                 }
2124
2125                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2126                         btrfs_warn(fs_info,
2127                 "too many missing devices, writable remount is not allowed");
2128                         ret = -EACCES;
2129                         goto restore;
2130                 }
2131
2132                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2133                         btrfs_warn(fs_info,
2134                 "mount required to replay tree-log, cannot remount read-write");
2135                         ret = -EINVAL;
2136                         goto restore;
2137                 }
2138
2139                 /*
2140                  * NOTE: when remounting with a change that does writes, don't
2141                  * put it anywhere above this point, as we are not sure to be
2142                  * safe to write until we pass the above checks.
2143                  */
2144                 ret = btrfs_start_pre_rw_mount(fs_info);
2145                 if (ret)
2146                         goto restore;
2147
2148                 btrfs_clear_sb_rdonly(sb);
2149
2150                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2151         }
2152 out:
2153         /*
2154          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2155          * since the absence of the flag means it can be toggled off by remount.
2156          */
2157         *flags |= SB_I_VERSION;
2158
2159         wake_up_process(fs_info->transaction_kthread);
2160         btrfs_remount_cleanup(fs_info, old_opts);
2161         btrfs_clear_oneshot_options(fs_info);
2162         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2163
2164         return 0;
2165
2166 restore:
2167         /* We've hit an error - don't reset SB_RDONLY */
2168         if (sb_rdonly(sb))
2169                 old_flags |= SB_RDONLY;
2170         if (!(old_flags & SB_RDONLY))
2171                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2172         sb->s_flags = old_flags;
2173         fs_info->mount_opt = old_opts;
2174         fs_info->compress_type = old_compress_type;
2175         fs_info->max_inline = old_max_inline;
2176         btrfs_resize_thread_pool(fs_info,
2177                 old_thread_pool_size, fs_info->thread_pool_size);
2178         fs_info->metadata_ratio = old_metadata_ratio;
2179         btrfs_remount_cleanup(fs_info, old_opts);
2180         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2181
2182         return ret;
2183 }
2184
2185 /* Used to sort the devices by max_avail(descending sort) */
2186 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
2187 {
2188         const struct btrfs_device_info *dev_info1 = a;
2189         const struct btrfs_device_info *dev_info2 = b;
2190
2191         if (dev_info1->max_avail > dev_info2->max_avail)
2192                 return -1;
2193         else if (dev_info1->max_avail < dev_info2->max_avail)
2194                 return 1;
2195         return 0;
2196 }
2197
2198 /*
2199  * sort the devices by max_avail, in which max free extent size of each device
2200  * is stored.(Descending Sort)
2201  */
2202 static inline void btrfs_descending_sort_devices(
2203                                         struct btrfs_device_info *devices,
2204                                         size_t nr_devices)
2205 {
2206         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2207              btrfs_cmp_device_free_bytes, NULL);
2208 }
2209
2210 /*
2211  * The helper to calc the free space on the devices that can be used to store
2212  * file data.
2213  */
2214 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2215                                               u64 *free_bytes)
2216 {
2217         struct btrfs_device_info *devices_info;
2218         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2219         struct btrfs_device *device;
2220         u64 type;
2221         u64 avail_space;
2222         u64 min_stripe_size;
2223         int num_stripes = 1;
2224         int i = 0, nr_devices;
2225         const struct btrfs_raid_attr *rattr;
2226
2227         /*
2228          * We aren't under the device list lock, so this is racy-ish, but good
2229          * enough for our purposes.
2230          */
2231         nr_devices = fs_info->fs_devices->open_devices;
2232         if (!nr_devices) {
2233                 smp_mb();
2234                 nr_devices = fs_info->fs_devices->open_devices;
2235                 ASSERT(nr_devices);
2236                 if (!nr_devices) {
2237                         *free_bytes = 0;
2238                         return 0;
2239                 }
2240         }
2241
2242         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2243                                GFP_KERNEL);
2244         if (!devices_info)
2245                 return -ENOMEM;
2246
2247         /* calc min stripe number for data space allocation */
2248         type = btrfs_data_alloc_profile(fs_info);
2249         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2250
2251         if (type & BTRFS_BLOCK_GROUP_RAID0)
2252                 num_stripes = nr_devices;
2253         else if (type & BTRFS_BLOCK_GROUP_RAID1_MASK)
2254                 num_stripes = rattr->ncopies;
2255         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2256                 num_stripes = 4;
2257
2258         /* Adjust for more than 1 stripe per device */
2259         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2260
2261         rcu_read_lock();
2262         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2263                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2264                                                 &device->dev_state) ||
2265                     !device->bdev ||
2266                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2267                         continue;
2268
2269                 if (i >= nr_devices)
2270                         break;
2271
2272                 avail_space = device->total_bytes - device->bytes_used;
2273
2274                 /* align with stripe_len */
2275                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2276
2277                 /*
2278                  * Ensure we have at least min_stripe_size on top of the
2279                  * reserved space on the device.
2280                  */
2281                 if (avail_space <= BTRFS_DEVICE_RANGE_RESERVED + min_stripe_size)
2282                         continue;
2283
2284                 avail_space -= BTRFS_DEVICE_RANGE_RESERVED;
2285
2286                 devices_info[i].dev = device;
2287                 devices_info[i].max_avail = avail_space;
2288
2289                 i++;
2290         }
2291         rcu_read_unlock();
2292
2293         nr_devices = i;
2294
2295         btrfs_descending_sort_devices(devices_info, nr_devices);
2296
2297         i = nr_devices - 1;
2298         avail_space = 0;
2299         while (nr_devices >= rattr->devs_min) {
2300                 num_stripes = min(num_stripes, nr_devices);
2301
2302                 if (devices_info[i].max_avail >= min_stripe_size) {
2303                         int j;
2304                         u64 alloc_size;
2305
2306                         avail_space += devices_info[i].max_avail * num_stripes;
2307                         alloc_size = devices_info[i].max_avail;
2308                         for (j = i + 1 - num_stripes; j <= i; j++)
2309                                 devices_info[j].max_avail -= alloc_size;
2310                 }
2311                 i--;
2312                 nr_devices--;
2313         }
2314
2315         kfree(devices_info);
2316         *free_bytes = avail_space;
2317         return 0;
2318 }
2319
2320 /*
2321  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2322  *
2323  * If there's a redundant raid level at DATA block groups, use the respective
2324  * multiplier to scale the sizes.
2325  *
2326  * Unused device space usage is based on simulating the chunk allocator
2327  * algorithm that respects the device sizes and order of allocations.  This is
2328  * a close approximation of the actual use but there are other factors that may
2329  * change the result (like a new metadata chunk).
2330  *
2331  * If metadata is exhausted, f_bavail will be 0.
2332  */
2333 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2334 {
2335         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2336         struct btrfs_super_block *disk_super = fs_info->super_copy;
2337         struct btrfs_space_info *found;
2338         u64 total_used = 0;
2339         u64 total_free_data = 0;
2340         u64 total_free_meta = 0;
2341         u32 bits = fs_info->sectorsize_bits;
2342         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2343         unsigned factor = 1;
2344         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2345         int ret;
2346         u64 thresh = 0;
2347         int mixed = 0;
2348
2349         list_for_each_entry(found, &fs_info->space_info, list) {
2350                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2351                         int i;
2352
2353                         total_free_data += found->disk_total - found->disk_used;
2354                         total_free_data -=
2355                                 btrfs_account_ro_block_groups_free_space(found);
2356
2357                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2358                                 if (!list_empty(&found->block_groups[i]))
2359                                         factor = btrfs_bg_type_to_factor(
2360                                                 btrfs_raid_array[i].bg_flag);
2361                         }
2362                 }
2363
2364                 /*
2365                  * Metadata in mixed block goup profiles are accounted in data
2366                  */
2367                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2368                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2369                                 mixed = 1;
2370                         else
2371                                 total_free_meta += found->disk_total -
2372                                         found->disk_used;
2373                 }
2374
2375                 total_used += found->disk_used;
2376         }
2377
2378         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2379         buf->f_blocks >>= bits;
2380         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2381
2382         /* Account global block reserve as used, it's in logical size already */
2383         spin_lock(&block_rsv->lock);
2384         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2385         if (buf->f_bfree >= block_rsv->size >> bits)
2386                 buf->f_bfree -= block_rsv->size >> bits;
2387         else
2388                 buf->f_bfree = 0;
2389         spin_unlock(&block_rsv->lock);
2390
2391         buf->f_bavail = div_u64(total_free_data, factor);
2392         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2393         if (ret)
2394                 return ret;
2395         buf->f_bavail += div_u64(total_free_data, factor);
2396         buf->f_bavail = buf->f_bavail >> bits;
2397
2398         /*
2399          * We calculate the remaining metadata space minus global reserve. If
2400          * this is (supposedly) smaller than zero, there's no space. But this
2401          * does not hold in practice, the exhausted state happens where's still
2402          * some positive delta. So we apply some guesswork and compare the
2403          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2404          *
2405          * We probably cannot calculate the exact threshold value because this
2406          * depends on the internal reservations requested by various
2407          * operations, so some operations that consume a few metadata will
2408          * succeed even if the Avail is zero. But this is better than the other
2409          * way around.
2410          */
2411         thresh = SZ_4M;
2412
2413         /*
2414          * We only want to claim there's no available space if we can no longer
2415          * allocate chunks for our metadata profile and our global reserve will
2416          * not fit in the free metadata space.  If we aren't ->full then we
2417          * still can allocate chunks and thus are fine using the currently
2418          * calculated f_bavail.
2419          */
2420         if (!mixed && block_rsv->space_info->full &&
2421             total_free_meta - thresh < block_rsv->size)
2422                 buf->f_bavail = 0;
2423
2424         buf->f_type = BTRFS_SUPER_MAGIC;
2425         buf->f_bsize = dentry->d_sb->s_blocksize;
2426         buf->f_namelen = BTRFS_NAME_LEN;
2427
2428         /* We treat it as constant endianness (it doesn't matter _which_)
2429            because we want the fsid to come out the same whether mounted
2430            on a big-endian or little-endian host */
2431         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2432         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2433         /* Mask in the root object ID too, to disambiguate subvols */
2434         buf->f_fsid.val[0] ^=
2435                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2436         buf->f_fsid.val[1] ^=
2437                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2438
2439         return 0;
2440 }
2441
2442 static void btrfs_kill_super(struct super_block *sb)
2443 {
2444         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2445         kill_anon_super(sb);
2446         btrfs_free_fs_info(fs_info);
2447 }
2448
2449 static struct file_system_type btrfs_fs_type = {
2450         .owner          = THIS_MODULE,
2451         .name           = "btrfs",
2452         .mount          = btrfs_mount,
2453         .kill_sb        = btrfs_kill_super,
2454         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2455 };
2456
2457 static struct file_system_type btrfs_root_fs_type = {
2458         .owner          = THIS_MODULE,
2459         .name           = "btrfs",
2460         .mount          = btrfs_mount_root,
2461         .kill_sb        = btrfs_kill_super,
2462         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA | FS_ALLOW_IDMAP,
2463 };
2464
2465 MODULE_ALIAS_FS("btrfs");
2466
2467 static int btrfs_control_open(struct inode *inode, struct file *file)
2468 {
2469         /*
2470          * The control file's private_data is used to hold the
2471          * transaction when it is started and is used to keep
2472          * track of whether a transaction is already in progress.
2473          */
2474         file->private_data = NULL;
2475         return 0;
2476 }
2477
2478 /*
2479  * Used by /dev/btrfs-control for devices ioctls.
2480  */
2481 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2482                                 unsigned long arg)
2483 {
2484         struct btrfs_ioctl_vol_args *vol;
2485         struct btrfs_device *device = NULL;
2486         dev_t devt = 0;
2487         int ret = -ENOTTY;
2488
2489         if (!capable(CAP_SYS_ADMIN))
2490                 return -EPERM;
2491
2492         vol = memdup_user((void __user *)arg, sizeof(*vol));
2493         if (IS_ERR(vol))
2494                 return PTR_ERR(vol);
2495         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2496
2497         switch (cmd) {
2498         case BTRFS_IOC_SCAN_DEV:
2499                 mutex_lock(&uuid_mutex);
2500                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2501                                                &btrfs_root_fs_type);
2502                 ret = PTR_ERR_OR_ZERO(device);
2503                 mutex_unlock(&uuid_mutex);
2504                 break;
2505         case BTRFS_IOC_FORGET_DEV:
2506                 if (vol->name[0] != 0) {
2507                         ret = lookup_bdev(vol->name, &devt);
2508                         if (ret)
2509                                 break;
2510                 }
2511                 ret = btrfs_forget_devices(devt);
2512                 break;
2513         case BTRFS_IOC_DEVICES_READY:
2514                 mutex_lock(&uuid_mutex);
2515                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2516                                                &btrfs_root_fs_type);
2517                 if (IS_ERR(device)) {
2518                         mutex_unlock(&uuid_mutex);
2519                         ret = PTR_ERR(device);
2520                         break;
2521                 }
2522                 ret = !(device->fs_devices->num_devices ==
2523                         device->fs_devices->total_devices);
2524                 mutex_unlock(&uuid_mutex);
2525                 break;
2526         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2527                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2528                 break;
2529         }
2530
2531         kfree(vol);
2532         return ret;
2533 }
2534
2535 static int btrfs_freeze(struct super_block *sb)
2536 {
2537         struct btrfs_trans_handle *trans;
2538         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2539         struct btrfs_root *root = fs_info->tree_root;
2540
2541         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2542         /*
2543          * We don't need a barrier here, we'll wait for any transaction that
2544          * could be in progress on other threads (and do delayed iputs that
2545          * we want to avoid on a frozen filesystem), or do the commit
2546          * ourselves.
2547          */
2548         trans = btrfs_attach_transaction_barrier(root);
2549         if (IS_ERR(trans)) {
2550                 /* no transaction, don't bother */
2551                 if (PTR_ERR(trans) == -ENOENT)
2552                         return 0;
2553                 return PTR_ERR(trans);
2554         }
2555         return btrfs_commit_transaction(trans);
2556 }
2557
2558 static int check_dev_super(struct btrfs_device *dev)
2559 {
2560         struct btrfs_fs_info *fs_info = dev->fs_info;
2561         struct btrfs_super_block *sb;
2562         u16 csum_type;
2563         int ret = 0;
2564
2565         /* This should be called with fs still frozen. */
2566         ASSERT(test_bit(BTRFS_FS_FROZEN, &fs_info->flags));
2567
2568         /* Missing dev, no need to check. */
2569         if (!dev->bdev)
2570                 return 0;
2571
2572         /* Only need to check the primary super block. */
2573         sb = btrfs_read_dev_one_super(dev->bdev, 0, true);
2574         if (IS_ERR(sb))
2575                 return PTR_ERR(sb);
2576
2577         /* Verify the checksum. */
2578         csum_type = btrfs_super_csum_type(sb);
2579         if (csum_type != btrfs_super_csum_type(fs_info->super_copy)) {
2580                 btrfs_err(fs_info, "csum type changed, has %u expect %u",
2581                           csum_type, btrfs_super_csum_type(fs_info->super_copy));
2582                 ret = -EUCLEAN;
2583                 goto out;
2584         }
2585
2586         if (btrfs_check_super_csum(fs_info, sb)) {
2587                 btrfs_err(fs_info, "csum for on-disk super block no longer matches");
2588                 ret = -EUCLEAN;
2589                 goto out;
2590         }
2591
2592         /* Btrfs_validate_super() includes fsid check against super->fsid. */
2593         ret = btrfs_validate_super(fs_info, sb, 0);
2594         if (ret < 0)
2595                 goto out;
2596
2597         if (btrfs_super_generation(sb) != fs_info->last_trans_committed) {
2598                 btrfs_err(fs_info, "transid mismatch, has %llu expect %llu",
2599                         btrfs_super_generation(sb),
2600                         fs_info->last_trans_committed);
2601                 ret = -EUCLEAN;
2602                 goto out;
2603         }
2604 out:
2605         btrfs_release_disk_super(sb);
2606         return ret;
2607 }
2608
2609 static int btrfs_unfreeze(struct super_block *sb)
2610 {
2611         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2612         struct btrfs_device *device;
2613         int ret = 0;
2614
2615         /*
2616          * Make sure the fs is not changed by accident (like hibernation then
2617          * modified by other OS).
2618          * If we found anything wrong, we mark the fs error immediately.
2619          *
2620          * And since the fs is frozen, no one can modify the fs yet, thus
2621          * we don't need to hold device_list_mutex.
2622          */
2623         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
2624                 ret = check_dev_super(device);
2625                 if (ret < 0) {
2626                         btrfs_handle_fs_error(fs_info, ret,
2627                                 "super block on devid %llu got modified unexpectedly",
2628                                 device->devid);
2629                         break;
2630                 }
2631         }
2632         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2633
2634         /*
2635          * We still return 0, to allow VFS layer to unfreeze the fs even the
2636          * above checks failed. Since the fs is either fine or read-only, we're
2637          * safe to continue, without causing further damage.
2638          */
2639         return 0;
2640 }
2641
2642 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2643 {
2644         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2645
2646         /*
2647          * There should be always a valid pointer in latest_dev, it may be stale
2648          * for a short moment in case it's being deleted but still valid until
2649          * the end of RCU grace period.
2650          */
2651         rcu_read_lock();
2652         seq_escape(m, rcu_str_deref(fs_info->fs_devices->latest_dev->name), " \t\n\\");
2653         rcu_read_unlock();
2654
2655         return 0;
2656 }
2657
2658 static const struct super_operations btrfs_super_ops = {
2659         .drop_inode     = btrfs_drop_inode,
2660         .evict_inode    = btrfs_evict_inode,
2661         .put_super      = btrfs_put_super,
2662         .sync_fs        = btrfs_sync_fs,
2663         .show_options   = btrfs_show_options,
2664         .show_devname   = btrfs_show_devname,
2665         .alloc_inode    = btrfs_alloc_inode,
2666         .destroy_inode  = btrfs_destroy_inode,
2667         .free_inode     = btrfs_free_inode,
2668         .statfs         = btrfs_statfs,
2669         .remount_fs     = btrfs_remount,
2670         .freeze_fs      = btrfs_freeze,
2671         .unfreeze_fs    = btrfs_unfreeze,
2672 };
2673
2674 static const struct file_operations btrfs_ctl_fops = {
2675         .open = btrfs_control_open,
2676         .unlocked_ioctl  = btrfs_control_ioctl,
2677         .compat_ioctl = compat_ptr_ioctl,
2678         .owner   = THIS_MODULE,
2679         .llseek = noop_llseek,
2680 };
2681
2682 static struct miscdevice btrfs_misc = {
2683         .minor          = BTRFS_MINOR,
2684         .name           = "btrfs-control",
2685         .fops           = &btrfs_ctl_fops
2686 };
2687
2688 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2689 MODULE_ALIAS("devname:btrfs-control");
2690
2691 static int __init btrfs_interface_init(void)
2692 {
2693         return misc_register(&btrfs_misc);
2694 }
2695
2696 static __cold void btrfs_interface_exit(void)
2697 {
2698         misc_deregister(&btrfs_misc);
2699 }
2700
2701 static void __init btrfs_print_mod_info(void)
2702 {
2703         static const char options[] = ""
2704 #ifdef CONFIG_BTRFS_DEBUG
2705                         ", debug=on"
2706 #endif
2707 #ifdef CONFIG_BTRFS_ASSERT
2708                         ", assert=on"
2709 #endif
2710 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2711                         ", integrity-checker=on"
2712 #endif
2713 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2714                         ", ref-verify=on"
2715 #endif
2716 #ifdef CONFIG_BLK_DEV_ZONED
2717                         ", zoned=yes"
2718 #else
2719                         ", zoned=no"
2720 #endif
2721 #ifdef CONFIG_FS_VERITY
2722                         ", fsverity=yes"
2723 #else
2724                         ", fsverity=no"
2725 #endif
2726                         ;
2727         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2728 }
2729
2730 static int __init init_btrfs_fs(void)
2731 {
2732         int err;
2733
2734         btrfs_props_init();
2735
2736         err = btrfs_init_sysfs();
2737         if (err)
2738                 return err;
2739
2740         btrfs_init_compress();
2741
2742         err = btrfs_init_cachep();
2743         if (err)
2744                 goto free_compress;
2745
2746         err = extent_state_init_cachep();
2747         if (err)
2748                 goto free_cachep;
2749
2750         err = extent_buffer_init_cachep();
2751         if (err)
2752                 goto free_extent_cachep;
2753
2754         err = btrfs_bioset_init();
2755         if (err)
2756                 goto free_eb_cachep;
2757
2758         err = extent_map_init();
2759         if (err)
2760                 goto free_bioset;
2761
2762         err = ordered_data_init();
2763         if (err)
2764                 goto free_extent_map;
2765
2766         err = btrfs_delayed_inode_init();
2767         if (err)
2768                 goto free_ordered_data;
2769
2770         err = btrfs_auto_defrag_init();
2771         if (err)
2772                 goto free_delayed_inode;
2773
2774         err = btrfs_delayed_ref_init();
2775         if (err)
2776                 goto free_auto_defrag;
2777
2778         err = btrfs_prelim_ref_init();
2779         if (err)
2780                 goto free_delayed_ref;
2781
2782         err = btrfs_interface_init();
2783         if (err)
2784                 goto free_prelim_ref;
2785
2786         btrfs_print_mod_info();
2787
2788         err = btrfs_run_sanity_tests();
2789         if (err)
2790                 goto unregister_ioctl;
2791
2792         err = register_filesystem(&btrfs_fs_type);
2793         if (err)
2794                 goto unregister_ioctl;
2795
2796         return 0;
2797
2798 unregister_ioctl:
2799         btrfs_interface_exit();
2800 free_prelim_ref:
2801         btrfs_prelim_ref_exit();
2802 free_delayed_ref:
2803         btrfs_delayed_ref_exit();
2804 free_auto_defrag:
2805         btrfs_auto_defrag_exit();
2806 free_delayed_inode:
2807         btrfs_delayed_inode_exit();
2808 free_ordered_data:
2809         ordered_data_exit();
2810 free_extent_map:
2811         extent_map_exit();
2812 free_bioset:
2813         btrfs_bioset_exit();
2814 free_eb_cachep:
2815         extent_buffer_free_cachep();
2816 free_extent_cachep:
2817         extent_state_free_cachep();
2818 free_cachep:
2819         btrfs_destroy_cachep();
2820 free_compress:
2821         btrfs_exit_compress();
2822         btrfs_exit_sysfs();
2823
2824         return err;
2825 }
2826
2827 static void __exit exit_btrfs_fs(void)
2828 {
2829         btrfs_destroy_cachep();
2830         btrfs_delayed_ref_exit();
2831         btrfs_auto_defrag_exit();
2832         btrfs_delayed_inode_exit();
2833         btrfs_prelim_ref_exit();
2834         ordered_data_exit();
2835         extent_map_exit();
2836         btrfs_bioset_exit();
2837         extent_state_free_cachep();
2838         extent_buffer_free_cachep();
2839         btrfs_interface_exit();
2840         unregister_filesystem(&btrfs_fs_type);
2841         btrfs_exit_sysfs();
2842         btrfs_cleanup_fs_uuids();
2843         btrfs_exit_compress();
2844 }
2845
2846 late_initcall(init_btrfs_fs);
2847 module_exit(exit_btrfs_fs)
2848
2849 MODULE_LICENSE("GPL");
2850 MODULE_SOFTDEP("pre: crc32c");
2851 MODULE_SOFTDEP("pre: xxhash64");
2852 MODULE_SOFTDEP("pre: sha256");
2853 MODULE_SOFTDEP("pre: blake2b-256");