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         }
1141         if (!ret)
1142                 ret = btrfs_check_mountopts_zoned(info);
1143         if (!ret && !remounting) {
1144                 if (btrfs_test_opt(info, SPACE_CACHE))
1145                         btrfs_info(info, "disk space caching is enabled");
1146                 if (btrfs_test_opt(info, FREE_SPACE_TREE))
1147                         btrfs_info(info, "using free space tree");
1148         }
1149         return ret;
1150 }
1151
1152 /*
1153  * Parse mount options that are required early in the mount process.
1154  *
1155  * All other options will be parsed on much later in the mount process and
1156  * only when we need to allocate a new super block.
1157  */
1158 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1159                                       void *holder)
1160 {
1161         substring_t args[MAX_OPT_ARGS];
1162         char *device_name, *opts, *orig, *p;
1163         struct btrfs_device *device = NULL;
1164         int error = 0;
1165
1166         lockdep_assert_held(&uuid_mutex);
1167
1168         if (!options)
1169                 return 0;
1170
1171         /*
1172          * strsep changes the string, duplicate it because btrfs_parse_options
1173          * gets called later
1174          */
1175         opts = kstrdup(options, GFP_KERNEL);
1176         if (!opts)
1177                 return -ENOMEM;
1178         orig = opts;
1179
1180         while ((p = strsep(&opts, ",")) != NULL) {
1181                 int token;
1182
1183                 if (!*p)
1184                         continue;
1185
1186                 token = match_token(p, tokens, args);
1187                 if (token == Opt_device) {
1188                         device_name = match_strdup(&args[0]);
1189                         if (!device_name) {
1190                                 error = -ENOMEM;
1191                                 goto out;
1192                         }
1193                         device = btrfs_scan_one_device(device_name, flags,
1194                                         holder);
1195                         kfree(device_name);
1196                         if (IS_ERR(device)) {
1197                                 error = PTR_ERR(device);
1198                                 goto out;
1199                         }
1200                 }
1201         }
1202
1203 out:
1204         kfree(orig);
1205         return error;
1206 }
1207
1208 /*
1209  * Parse mount options that are related to subvolume id
1210  *
1211  * The value is later passed to mount_subvol()
1212  */
1213 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1214                 u64 *subvol_objectid)
1215 {
1216         substring_t args[MAX_OPT_ARGS];
1217         char *opts, *orig, *p;
1218         int error = 0;
1219         u64 subvolid;
1220
1221         if (!options)
1222                 return 0;
1223
1224         /*
1225          * strsep changes the string, duplicate it because
1226          * btrfs_parse_device_options gets called later
1227          */
1228         opts = kstrdup(options, GFP_KERNEL);
1229         if (!opts)
1230                 return -ENOMEM;
1231         orig = opts;
1232
1233         while ((p = strsep(&opts, ",")) != NULL) {
1234                 int token;
1235                 if (!*p)
1236                         continue;
1237
1238                 token = match_token(p, tokens, args);
1239                 switch (token) {
1240                 case Opt_subvol:
1241                         kfree(*subvol_name);
1242                         *subvol_name = match_strdup(&args[0]);
1243                         if (!*subvol_name) {
1244                                 error = -ENOMEM;
1245                                 goto out;
1246                         }
1247                         break;
1248                 case Opt_subvolid:
1249                         error = match_u64(&args[0], &subvolid);
1250                         if (error)
1251                                 goto out;
1252
1253                         /* we want the original fs_tree */
1254                         if (subvolid == 0)
1255                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1256
1257                         *subvol_objectid = subvolid;
1258                         break;
1259                 default:
1260                         break;
1261                 }
1262         }
1263
1264 out:
1265         kfree(orig);
1266         return error;
1267 }
1268
1269 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1270                                           u64 subvol_objectid)
1271 {
1272         struct btrfs_root *root = fs_info->tree_root;
1273         struct btrfs_root *fs_root = NULL;
1274         struct btrfs_root_ref *root_ref;
1275         struct btrfs_inode_ref *inode_ref;
1276         struct btrfs_key key;
1277         struct btrfs_path *path = NULL;
1278         char *name = NULL, *ptr;
1279         u64 dirid;
1280         int len;
1281         int ret;
1282
1283         path = btrfs_alloc_path();
1284         if (!path) {
1285                 ret = -ENOMEM;
1286                 goto err;
1287         }
1288
1289         name = kmalloc(PATH_MAX, GFP_KERNEL);
1290         if (!name) {
1291                 ret = -ENOMEM;
1292                 goto err;
1293         }
1294         ptr = name + PATH_MAX - 1;
1295         ptr[0] = '\0';
1296
1297         /*
1298          * Walk up the subvolume trees in the tree of tree roots by root
1299          * backrefs until we hit the top-level subvolume.
1300          */
1301         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1302                 key.objectid = subvol_objectid;
1303                 key.type = BTRFS_ROOT_BACKREF_KEY;
1304                 key.offset = (u64)-1;
1305
1306                 ret = btrfs_search_backwards(root, &key, path);
1307                 if (ret < 0) {
1308                         goto err;
1309                 } else if (ret > 0) {
1310                         ret = -ENOENT;
1311                         goto err;
1312                 }
1313
1314                 subvol_objectid = key.offset;
1315
1316                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1317                                           struct btrfs_root_ref);
1318                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1319                 ptr -= len + 1;
1320                 if (ptr < name) {
1321                         ret = -ENAMETOOLONG;
1322                         goto err;
1323                 }
1324                 read_extent_buffer(path->nodes[0], ptr + 1,
1325                                    (unsigned long)(root_ref + 1), len);
1326                 ptr[0] = '/';
1327                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1328                 btrfs_release_path(path);
1329
1330                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1331                 if (IS_ERR(fs_root)) {
1332                         ret = PTR_ERR(fs_root);
1333                         fs_root = NULL;
1334                         goto err;
1335                 }
1336
1337                 /*
1338                  * Walk up the filesystem tree by inode refs until we hit the
1339                  * root directory.
1340                  */
1341                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1342                         key.objectid = dirid;
1343                         key.type = BTRFS_INODE_REF_KEY;
1344                         key.offset = (u64)-1;
1345
1346                         ret = btrfs_search_backwards(fs_root, &key, path);
1347                         if (ret < 0) {
1348                                 goto err;
1349                         } else if (ret > 0) {
1350                                 ret = -ENOENT;
1351                                 goto err;
1352                         }
1353
1354                         dirid = key.offset;
1355
1356                         inode_ref = btrfs_item_ptr(path->nodes[0],
1357                                                    path->slots[0],
1358                                                    struct btrfs_inode_ref);
1359                         len = btrfs_inode_ref_name_len(path->nodes[0],
1360                                                        inode_ref);
1361                         ptr -= len + 1;
1362                         if (ptr < name) {
1363                                 ret = -ENAMETOOLONG;
1364                                 goto err;
1365                         }
1366                         read_extent_buffer(path->nodes[0], ptr + 1,
1367                                            (unsigned long)(inode_ref + 1), len);
1368                         ptr[0] = '/';
1369                         btrfs_release_path(path);
1370                 }
1371                 btrfs_put_root(fs_root);
1372                 fs_root = NULL;
1373         }
1374
1375         btrfs_free_path(path);
1376         if (ptr == name + PATH_MAX - 1) {
1377                 name[0] = '/';
1378                 name[1] = '\0';
1379         } else {
1380                 memmove(name, ptr, name + PATH_MAX - ptr);
1381         }
1382         return name;
1383
1384 err:
1385         btrfs_put_root(fs_root);
1386         btrfs_free_path(path);
1387         kfree(name);
1388         return ERR_PTR(ret);
1389 }
1390
1391 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1392 {
1393         struct btrfs_root *root = fs_info->tree_root;
1394         struct btrfs_dir_item *di;
1395         struct btrfs_path *path;
1396         struct btrfs_key location;
1397         u64 dir_id;
1398
1399         path = btrfs_alloc_path();
1400         if (!path)
1401                 return -ENOMEM;
1402
1403         /*
1404          * Find the "default" dir item which points to the root item that we
1405          * will mount by default if we haven't been given a specific subvolume
1406          * to mount.
1407          */
1408         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1409         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1410         if (IS_ERR(di)) {
1411                 btrfs_free_path(path);
1412                 return PTR_ERR(di);
1413         }
1414         if (!di) {
1415                 /*
1416                  * Ok the default dir item isn't there.  This is weird since
1417                  * it's always been there, but don't freak out, just try and
1418                  * mount the top-level subvolume.
1419                  */
1420                 btrfs_free_path(path);
1421                 *objectid = BTRFS_FS_TREE_OBJECTID;
1422                 return 0;
1423         }
1424
1425         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1426         btrfs_free_path(path);
1427         *objectid = location.objectid;
1428         return 0;
1429 }
1430
1431 static int btrfs_fill_super(struct super_block *sb,
1432                             struct btrfs_fs_devices *fs_devices,
1433                             void *data)
1434 {
1435         struct inode *inode;
1436         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1437         int err;
1438
1439         sb->s_maxbytes = MAX_LFS_FILESIZE;
1440         sb->s_magic = BTRFS_SUPER_MAGIC;
1441         sb->s_op = &btrfs_super_ops;
1442         sb->s_d_op = &btrfs_dentry_operations;
1443         sb->s_export_op = &btrfs_export_ops;
1444 #ifdef CONFIG_FS_VERITY
1445         sb->s_vop = &btrfs_verityops;
1446 #endif
1447         sb->s_xattr = btrfs_xattr_handlers;
1448         sb->s_time_gran = 1;
1449 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1450         sb->s_flags |= SB_POSIXACL;
1451 #endif
1452         sb->s_flags |= SB_I_VERSION;
1453         sb->s_iflags |= SB_I_CGROUPWB;
1454
1455         err = super_setup_bdi(sb);
1456         if (err) {
1457                 btrfs_err(fs_info, "super_setup_bdi failed");
1458                 return err;
1459         }
1460
1461         err = open_ctree(sb, fs_devices, (char *)data);
1462         if (err) {
1463                 btrfs_err(fs_info, "open_ctree failed");
1464                 return err;
1465         }
1466
1467         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1468         if (IS_ERR(inode)) {
1469                 err = PTR_ERR(inode);
1470                 goto fail_close;
1471         }
1472
1473         sb->s_root = d_make_root(inode);
1474         if (!sb->s_root) {
1475                 err = -ENOMEM;
1476                 goto fail_close;
1477         }
1478
1479         sb->s_flags |= SB_ACTIVE;
1480         return 0;
1481
1482 fail_close:
1483         close_ctree(fs_info);
1484         return err;
1485 }
1486
1487 int btrfs_sync_fs(struct super_block *sb, int wait)
1488 {
1489         struct btrfs_trans_handle *trans;
1490         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1491         struct btrfs_root *root = fs_info->tree_root;
1492
1493         trace_btrfs_sync_fs(fs_info, wait);
1494
1495         if (!wait) {
1496                 filemap_flush(fs_info->btree_inode->i_mapping);
1497                 return 0;
1498         }
1499
1500         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1501
1502         trans = btrfs_attach_transaction_barrier(root);
1503         if (IS_ERR(trans)) {
1504                 /* no transaction, don't bother */
1505                 if (PTR_ERR(trans) == -ENOENT) {
1506                         /*
1507                          * Exit unless we have some pending changes
1508                          * that need to go through commit
1509                          */
1510                         if (fs_info->pending_changes == 0)
1511                                 return 0;
1512                         /*
1513                          * A non-blocking test if the fs is frozen. We must not
1514                          * start a new transaction here otherwise a deadlock
1515                          * happens. The pending operations are delayed to the
1516                          * next commit after thawing.
1517                          */
1518                         if (sb_start_write_trylock(sb))
1519                                 sb_end_write(sb);
1520                         else
1521                                 return 0;
1522                         trans = btrfs_start_transaction(root, 0);
1523                 }
1524                 if (IS_ERR(trans))
1525                         return PTR_ERR(trans);
1526         }
1527         return btrfs_commit_transaction(trans);
1528 }
1529
1530 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1531 {
1532         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1533         *printed = true;
1534 }
1535
1536 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1537 {
1538         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1539         const char *compress_type;
1540         const char *subvol_name;
1541         bool printed = false;
1542
1543         if (btrfs_test_opt(info, DEGRADED))
1544                 seq_puts(seq, ",degraded");
1545         if (btrfs_test_opt(info, NODATASUM))
1546                 seq_puts(seq, ",nodatasum");
1547         if (btrfs_test_opt(info, NODATACOW))
1548                 seq_puts(seq, ",nodatacow");
1549         if (btrfs_test_opt(info, NOBARRIER))
1550                 seq_puts(seq, ",nobarrier");
1551         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1552                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1553         if (info->thread_pool_size !=  min_t(unsigned long,
1554                                              num_online_cpus() + 2, 8))
1555                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1556         if (btrfs_test_opt(info, COMPRESS)) {
1557                 compress_type = btrfs_compress_type2str(info->compress_type);
1558                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1559                         seq_printf(seq, ",compress-force=%s", compress_type);
1560                 else
1561                         seq_printf(seq, ",compress=%s", compress_type);
1562                 if (info->compress_level)
1563                         seq_printf(seq, ":%d", info->compress_level);
1564         }
1565         if (btrfs_test_opt(info, NOSSD))
1566                 seq_puts(seq, ",nossd");
1567         if (btrfs_test_opt(info, SSD_SPREAD))
1568                 seq_puts(seq, ",ssd_spread");
1569         else if (btrfs_test_opt(info, SSD))
1570                 seq_puts(seq, ",ssd");
1571         if (btrfs_test_opt(info, NOTREELOG))
1572                 seq_puts(seq, ",notreelog");
1573         if (btrfs_test_opt(info, NOLOGREPLAY))
1574                 print_rescue_option(seq, "nologreplay", &printed);
1575         if (btrfs_test_opt(info, USEBACKUPROOT))
1576                 print_rescue_option(seq, "usebackuproot", &printed);
1577         if (btrfs_test_opt(info, IGNOREBADROOTS))
1578                 print_rescue_option(seq, "ignorebadroots", &printed);
1579         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1580                 print_rescue_option(seq, "ignoredatacsums", &printed);
1581         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1582                 seq_puts(seq, ",flushoncommit");
1583         if (btrfs_test_opt(info, DISCARD_SYNC))
1584                 seq_puts(seq, ",discard");
1585         if (btrfs_test_opt(info, DISCARD_ASYNC))
1586                 seq_puts(seq, ",discard=async");
1587         if (!(info->sb->s_flags & SB_POSIXACL))
1588                 seq_puts(seq, ",noacl");
1589         if (btrfs_free_space_cache_v1_active(info))
1590                 seq_puts(seq, ",space_cache");
1591         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1592                 seq_puts(seq, ",space_cache=v2");
1593         else
1594                 seq_puts(seq, ",nospace_cache");
1595         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1596                 seq_puts(seq, ",rescan_uuid_tree");
1597         if (btrfs_test_opt(info, CLEAR_CACHE))
1598                 seq_puts(seq, ",clear_cache");
1599         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1600                 seq_puts(seq, ",user_subvol_rm_allowed");
1601         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1602                 seq_puts(seq, ",enospc_debug");
1603         if (btrfs_test_opt(info, AUTO_DEFRAG))
1604                 seq_puts(seq, ",autodefrag");
1605         if (btrfs_test_opt(info, SKIP_BALANCE))
1606                 seq_puts(seq, ",skip_balance");
1607 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1608         if (btrfs_test_opt(info, CHECK_INTEGRITY_DATA))
1609                 seq_puts(seq, ",check_int_data");
1610         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1611                 seq_puts(seq, ",check_int");
1612         if (info->check_integrity_print_mask)
1613                 seq_printf(seq, ",check_int_print_mask=%d",
1614                                 info->check_integrity_print_mask);
1615 #endif
1616         if (info->metadata_ratio)
1617                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1618         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1619                 seq_puts(seq, ",fatal_errors=panic");
1620         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1621                 seq_printf(seq, ",commit=%u", info->commit_interval);
1622 #ifdef CONFIG_BTRFS_DEBUG
1623         if (btrfs_test_opt(info, FRAGMENT_DATA))
1624                 seq_puts(seq, ",fragment=data");
1625         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1626                 seq_puts(seq, ",fragment=metadata");
1627 #endif
1628         if (btrfs_test_opt(info, REF_VERIFY))
1629                 seq_puts(seq, ",ref_verify");
1630         seq_printf(seq, ",subvolid=%llu",
1631                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1632         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1633                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1634         if (!IS_ERR(subvol_name)) {
1635                 seq_puts(seq, ",subvol=");
1636                 seq_escape(seq, subvol_name, " \t\n\\");
1637                 kfree(subvol_name);
1638         }
1639         return 0;
1640 }
1641
1642 static int btrfs_test_super(struct super_block *s, void *data)
1643 {
1644         struct btrfs_fs_info *p = data;
1645         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1646
1647         return fs_info->fs_devices == p->fs_devices;
1648 }
1649
1650 static int btrfs_set_super(struct super_block *s, void *data)
1651 {
1652         int err = set_anon_super(s, data);
1653         if (!err)
1654                 s->s_fs_info = data;
1655         return err;
1656 }
1657
1658 /*
1659  * subvolumes are identified by ino 256
1660  */
1661 static inline int is_subvolume_inode(struct inode *inode)
1662 {
1663         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1664                 return 1;
1665         return 0;
1666 }
1667
1668 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1669                                    struct vfsmount *mnt)
1670 {
1671         struct dentry *root;
1672         int ret;
1673
1674         if (!subvol_name) {
1675                 if (!subvol_objectid) {
1676                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1677                                                           &subvol_objectid);
1678                         if (ret) {
1679                                 root = ERR_PTR(ret);
1680                                 goto out;
1681                         }
1682                 }
1683                 subvol_name = btrfs_get_subvol_name_from_objectid(
1684                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1685                 if (IS_ERR(subvol_name)) {
1686                         root = ERR_CAST(subvol_name);
1687                         subvol_name = NULL;
1688                         goto out;
1689                 }
1690
1691         }
1692
1693         root = mount_subtree(mnt, subvol_name);
1694         /* mount_subtree() drops our reference on the vfsmount. */
1695         mnt = NULL;
1696
1697         if (!IS_ERR(root)) {
1698                 struct super_block *s = root->d_sb;
1699                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1700                 struct inode *root_inode = d_inode(root);
1701                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1702
1703                 ret = 0;
1704                 if (!is_subvolume_inode(root_inode)) {
1705                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1706                                subvol_name);
1707                         ret = -EINVAL;
1708                 }
1709                 if (subvol_objectid && root_objectid != subvol_objectid) {
1710                         /*
1711                          * This will also catch a race condition where a
1712                          * subvolume which was passed by ID is renamed and
1713                          * another subvolume is renamed over the old location.
1714                          */
1715                         btrfs_err(fs_info,
1716                                   "subvol '%s' does not match subvolid %llu",
1717                                   subvol_name, subvol_objectid);
1718                         ret = -EINVAL;
1719                 }
1720                 if (ret) {
1721                         dput(root);
1722                         root = ERR_PTR(ret);
1723                         deactivate_locked_super(s);
1724                 }
1725         }
1726
1727 out:
1728         mntput(mnt);
1729         kfree(subvol_name);
1730         return root;
1731 }
1732
1733 /*
1734  * Find a superblock for the given device / mount point.
1735  *
1736  * Note: This is based on mount_bdev from fs/super.c with a few additions
1737  *       for multiple device setup.  Make sure to keep it in sync.
1738  */
1739 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1740                 int flags, const char *device_name, void *data)
1741 {
1742         struct block_device *bdev = NULL;
1743         struct super_block *s;
1744         struct btrfs_device *device = NULL;
1745         struct btrfs_fs_devices *fs_devices = NULL;
1746         struct btrfs_fs_info *fs_info = NULL;
1747         void *new_sec_opts = NULL;
1748         fmode_t mode = FMODE_READ;
1749         int error = 0;
1750
1751         if (!(flags & SB_RDONLY))
1752                 mode |= FMODE_WRITE;
1753
1754         if (data) {
1755                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1756                 if (error)
1757                         return ERR_PTR(error);
1758         }
1759
1760         /*
1761          * Setup a dummy root and fs_info for test/set super.  This is because
1762          * we don't actually fill this stuff out until open_ctree, but we need
1763          * then open_ctree will properly initialize the file system specific
1764          * settings later.  btrfs_init_fs_info initializes the static elements
1765          * of the fs_info (locks and such) to make cleanup easier if we find a
1766          * superblock with our given fs_devices later on at sget() time.
1767          */
1768         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1769         if (!fs_info) {
1770                 error = -ENOMEM;
1771                 goto error_sec_opts;
1772         }
1773         btrfs_init_fs_info(fs_info);
1774
1775         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1776         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1777         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1778                 error = -ENOMEM;
1779                 goto error_fs_info;
1780         }
1781
1782         mutex_lock(&uuid_mutex);
1783         error = btrfs_parse_device_options(data, mode, fs_type);
1784         if (error) {
1785                 mutex_unlock(&uuid_mutex);
1786                 goto error_fs_info;
1787         }
1788
1789         device = btrfs_scan_one_device(device_name, mode, fs_type);
1790         if (IS_ERR(device)) {
1791                 mutex_unlock(&uuid_mutex);
1792                 error = PTR_ERR(device);
1793                 goto error_fs_info;
1794         }
1795
1796         fs_devices = device->fs_devices;
1797         fs_info->fs_devices = fs_devices;
1798
1799         error = btrfs_open_devices(fs_devices, mode, fs_type);
1800         mutex_unlock(&uuid_mutex);
1801         if (error)
1802                 goto error_fs_info;
1803
1804         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1805                 error = -EACCES;
1806                 goto error_close_devices;
1807         }
1808
1809         bdev = fs_devices->latest_dev->bdev;
1810         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1811                  fs_info);
1812         if (IS_ERR(s)) {
1813                 error = PTR_ERR(s);
1814                 goto error_close_devices;
1815         }
1816
1817         if (s->s_root) {
1818                 btrfs_close_devices(fs_devices);
1819                 btrfs_free_fs_info(fs_info);
1820                 if ((flags ^ s->s_flags) & SB_RDONLY)
1821                         error = -EBUSY;
1822         } else {
1823                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1824                 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s", fs_type->name,
1825                                         s->s_id);
1826                 btrfs_sb(s)->bdev_holder = fs_type;
1827                 if (!strstr(crc32c_impl(), "generic"))
1828                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1829                 error = btrfs_fill_super(s, fs_devices, data);
1830         }
1831         if (!error)
1832                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1833         security_free_mnt_opts(&new_sec_opts);
1834         if (error) {
1835                 deactivate_locked_super(s);
1836                 return ERR_PTR(error);
1837         }
1838
1839         return dget(s->s_root);
1840
1841 error_close_devices:
1842         btrfs_close_devices(fs_devices);
1843 error_fs_info:
1844         btrfs_free_fs_info(fs_info);
1845 error_sec_opts:
1846         security_free_mnt_opts(&new_sec_opts);
1847         return ERR_PTR(error);
1848 }
1849
1850 /*
1851  * Mount function which is called by VFS layer.
1852  *
1853  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1854  * which needs vfsmount* of device's root (/).  This means device's root has to
1855  * be mounted internally in any case.
1856  *
1857  * Operation flow:
1858  *   1. Parse subvol id related options for later use in mount_subvol().
1859  *
1860  *   2. Mount device's root (/) by calling vfs_kern_mount().
1861  *
1862  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1863  *      first place. In order to avoid calling btrfs_mount() again, we use
1864  *      different file_system_type which is not registered to VFS by
1865  *      register_filesystem() (btrfs_root_fs_type). As a result,
1866  *      btrfs_mount_root() is called. The return value will be used by
1867  *      mount_subtree() in mount_subvol().
1868  *
1869  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1870  *      "btrfs subvolume set-default", mount_subvol() is called always.
1871  */
1872 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1873                 const char *device_name, void *data)
1874 {
1875         struct vfsmount *mnt_root;
1876         struct dentry *root;
1877         char *subvol_name = NULL;
1878         u64 subvol_objectid = 0;
1879         int error = 0;
1880
1881         error = btrfs_parse_subvol_options(data, &subvol_name,
1882                                         &subvol_objectid);
1883         if (error) {
1884                 kfree(subvol_name);
1885                 return ERR_PTR(error);
1886         }
1887
1888         /* mount device's root (/) */
1889         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1890         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1891                 if (flags & SB_RDONLY) {
1892                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1893                                 flags & ~SB_RDONLY, device_name, data);
1894                 } else {
1895                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1896                                 flags | SB_RDONLY, device_name, data);
1897                         if (IS_ERR(mnt_root)) {
1898                                 root = ERR_CAST(mnt_root);
1899                                 kfree(subvol_name);
1900                                 goto out;
1901                         }
1902
1903                         down_write(&mnt_root->mnt_sb->s_umount);
1904                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1905                         up_write(&mnt_root->mnt_sb->s_umount);
1906                         if (error < 0) {
1907                                 root = ERR_PTR(error);
1908                                 mntput(mnt_root);
1909                                 kfree(subvol_name);
1910                                 goto out;
1911                         }
1912                 }
1913         }
1914         if (IS_ERR(mnt_root)) {
1915                 root = ERR_CAST(mnt_root);
1916                 kfree(subvol_name);
1917                 goto out;
1918         }
1919
1920         /* mount_subvol() will free subvol_name and mnt_root */
1921         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1922
1923 out:
1924         return root;
1925 }
1926
1927 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1928                                      u32 new_pool_size, u32 old_pool_size)
1929 {
1930         if (new_pool_size == old_pool_size)
1931                 return;
1932
1933         fs_info->thread_pool_size = new_pool_size;
1934
1935         btrfs_info(fs_info, "resize thread pool %d -> %d",
1936                old_pool_size, new_pool_size);
1937
1938         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1939         btrfs_workqueue_set_max(fs_info->hipri_workers, new_pool_size);
1940         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1941         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1942         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1943         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1944         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1945 }
1946
1947 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1948                                        unsigned long old_opts, int flags)
1949 {
1950         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1951             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1952              (flags & SB_RDONLY))) {
1953                 /* wait for any defraggers to finish */
1954                 wait_event(fs_info->transaction_wait,
1955                            (atomic_read(&fs_info->defrag_running) == 0));
1956                 if (flags & SB_RDONLY)
1957                         sync_filesystem(fs_info->sb);
1958         }
1959 }
1960
1961 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1962                                          unsigned long old_opts)
1963 {
1964         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1965
1966         /*
1967          * We need to cleanup all defragable inodes if the autodefragment is
1968          * close or the filesystem is read only.
1969          */
1970         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1971             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1972                 btrfs_cleanup_defrag_inodes(fs_info);
1973         }
1974
1975         /* If we toggled discard async */
1976         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1977             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1978                 btrfs_discard_resume(fs_info);
1979         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1980                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1981                 btrfs_discard_cleanup(fs_info);
1982
1983         /* If we toggled space cache */
1984         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1985                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1986 }
1987
1988 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1989 {
1990         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1991         unsigned old_flags = sb->s_flags;
1992         unsigned long old_opts = fs_info->mount_opt;
1993         unsigned long old_compress_type = fs_info->compress_type;
1994         u64 old_max_inline = fs_info->max_inline;
1995         u32 old_thread_pool_size = fs_info->thread_pool_size;
1996         u32 old_metadata_ratio = fs_info->metadata_ratio;
1997         int ret;
1998
1999         sync_filesystem(sb);
2000         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2001
2002         if (data) {
2003                 void *new_sec_opts = NULL;
2004
2005                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
2006                 if (!ret)
2007                         ret = security_sb_remount(sb, new_sec_opts);
2008                 security_free_mnt_opts(&new_sec_opts);
2009                 if (ret)
2010                         goto restore;
2011         }
2012
2013         ret = btrfs_parse_options(fs_info, data, *flags);
2014         if (ret)
2015                 goto restore;
2016
2017         ret = btrfs_check_features(fs_info, sb);
2018         if (ret < 0)
2019                 goto restore;
2020
2021         btrfs_remount_begin(fs_info, old_opts, *flags);
2022         btrfs_resize_thread_pool(fs_info,
2023                 fs_info->thread_pool_size, old_thread_pool_size);
2024
2025         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
2026             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2027             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
2028                 btrfs_warn(fs_info,
2029                 "remount supports changing free space tree only from ro to rw");
2030                 /* Make sure free space cache options match the state on disk */
2031                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2032                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2033                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
2034                 }
2035                 if (btrfs_free_space_cache_v1_active(fs_info)) {
2036                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
2037                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
2038                 }
2039         }
2040
2041         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
2042                 goto out;
2043
2044         if (*flags & SB_RDONLY) {
2045                 /*
2046                  * this also happens on 'umount -rf' or on shutdown, when
2047                  * the filesystem is busy.
2048                  */
2049                 cancel_work_sync(&fs_info->async_reclaim_work);
2050                 cancel_work_sync(&fs_info->async_data_reclaim_work);
2051
2052                 btrfs_discard_cleanup(fs_info);
2053
2054                 /* wait for the uuid_scan task to finish */
2055                 down(&fs_info->uuid_tree_rescan_sem);
2056                 /* avoid complains from lockdep et al. */
2057                 up(&fs_info->uuid_tree_rescan_sem);
2058
2059                 btrfs_set_sb_rdonly(sb);
2060
2061                 /*
2062                  * Setting SB_RDONLY will put the cleaner thread to
2063                  * sleep at the next loop if it's already active.
2064                  * If it's already asleep, we'll leave unused block
2065                  * groups on disk until we're mounted read-write again
2066                  * unless we clean them up here.
2067                  */
2068                 btrfs_delete_unused_bgs(fs_info);
2069
2070                 /*
2071                  * The cleaner task could be already running before we set the
2072                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
2073                  * We must make sure that after we finish the remount, i.e. after
2074                  * we call btrfs_commit_super(), the cleaner can no longer start
2075                  * a transaction - either because it was dropping a dead root,
2076                  * running delayed iputs or deleting an unused block group (the
2077                  * cleaner picked a block group from the list of unused block
2078                  * groups before we were able to in the previous call to
2079                  * btrfs_delete_unused_bgs()).
2080                  */
2081                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
2082                             TASK_UNINTERRUPTIBLE);
2083
2084                 /*
2085                  * We've set the superblock to RO mode, so we might have made
2086                  * the cleaner task sleep without running all pending delayed
2087                  * iputs. Go through all the delayed iputs here, so that if an
2088                  * unmount happens without remounting RW we don't end up at
2089                  * finishing close_ctree() with a non-empty list of delayed
2090                  * iputs.
2091                  */
2092                 btrfs_run_delayed_iputs(fs_info);
2093
2094                 btrfs_dev_replace_suspend_for_unmount(fs_info);
2095                 btrfs_scrub_cancel(fs_info);
2096                 btrfs_pause_balance(fs_info);
2097
2098                 /*
2099                  * Pause the qgroup rescan worker if it is running. We don't want
2100                  * it to be still running after we are in RO mode, as after that,
2101                  * by the time we unmount, it might have left a transaction open,
2102                  * so we would leak the transaction and/or crash.
2103                  */
2104                 btrfs_qgroup_wait_for_completion(fs_info, false);
2105
2106                 ret = btrfs_commit_super(fs_info);
2107                 if (ret)
2108                         goto restore;
2109         } else {
2110                 if (BTRFS_FS_ERROR(fs_info)) {
2111                         btrfs_err(fs_info,
2112                                 "Remounting read-write after error is not allowed");
2113                         ret = -EINVAL;
2114                         goto restore;
2115                 }
2116                 if (fs_info->fs_devices->rw_devices == 0) {
2117                         ret = -EACCES;
2118                         goto restore;
2119                 }
2120
2121                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2122                         btrfs_warn(fs_info,
2123                 "too many missing devices, writable remount is not allowed");
2124                         ret = -EACCES;
2125                         goto restore;
2126                 }
2127
2128                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2129                         btrfs_warn(fs_info,
2130                 "mount required to replay tree-log, cannot remount read-write");
2131                         ret = -EINVAL;
2132                         goto restore;
2133                 }
2134
2135                 /*
2136                  * NOTE: when remounting with a change that does writes, don't
2137                  * put it anywhere above this point, as we are not sure to be
2138                  * safe to write until we pass the above checks.
2139                  */
2140                 ret = btrfs_start_pre_rw_mount(fs_info);
2141                 if (ret)
2142                         goto restore;
2143
2144                 btrfs_clear_sb_rdonly(sb);
2145
2146                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2147         }
2148 out:
2149         /*
2150          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2151          * since the absence of the flag means it can be toggled off by remount.
2152          */
2153         *flags |= SB_I_VERSION;
2154
2155         wake_up_process(fs_info->transaction_kthread);
2156         btrfs_remount_cleanup(fs_info, old_opts);
2157         btrfs_clear_oneshot_options(fs_info);
2158         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2159
2160         return 0;
2161
2162 restore:
2163         /* We've hit an error - don't reset SB_RDONLY */
2164         if (sb_rdonly(sb))
2165                 old_flags |= SB_RDONLY;
2166         if (!(old_flags & SB_RDONLY))
2167                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2168         sb->s_flags = old_flags;
2169         fs_info->mount_opt = old_opts;
2170         fs_info->compress_type = old_compress_type;
2171         fs_info->max_inline = old_max_inline;
2172         btrfs_resize_thread_pool(fs_info,
2173                 old_thread_pool_size, fs_info->thread_pool_size);
2174         fs_info->metadata_ratio = old_metadata_ratio;
2175         btrfs_remount_cleanup(fs_info, old_opts);
2176         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2177
2178         return ret;
2179 }
2180
2181 /* Used to sort the devices by max_avail(descending sort) */
2182 static int btrfs_cmp_device_free_bytes(const void *a, const void *b)
2183 {
2184         const struct btrfs_device_info *dev_info1 = a;
2185         const struct btrfs_device_info *dev_info2 = b;
2186
2187         if (dev_info1->max_avail > dev_info2->max_avail)
2188                 return -1;
2189         else if (dev_info1->max_avail < dev_info2->max_avail)
2190                 return 1;
2191         return 0;
2192 }
2193
2194 /*
2195  * sort the devices by max_avail, in which max free extent size of each device
2196  * is stored.(Descending Sort)
2197  */
2198 static inline void btrfs_descending_sort_devices(
2199                                         struct btrfs_device_info *devices,
2200                                         size_t nr_devices)
2201 {
2202         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2203              btrfs_cmp_device_free_bytes, NULL);
2204 }
2205
2206 /*
2207  * The helper to calc the free space on the devices that can be used to store
2208  * file data.
2209  */
2210 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2211                                               u64 *free_bytes)
2212 {
2213         struct btrfs_device_info *devices_info;
2214         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2215         struct btrfs_device *device;
2216         u64 type;
2217         u64 avail_space;
2218         u64 min_stripe_size;
2219         int num_stripes = 1;
2220         int i = 0, nr_devices;
2221         const struct btrfs_raid_attr *rattr;
2222
2223         /*
2224          * We aren't under the device list lock, so this is racy-ish, but good
2225          * enough for our purposes.
2226          */
2227         nr_devices = fs_info->fs_devices->open_devices;
2228         if (!nr_devices) {
2229                 smp_mb();
2230                 nr_devices = fs_info->fs_devices->open_devices;
2231                 ASSERT(nr_devices);
2232                 if (!nr_devices) {
2233                         *free_bytes = 0;
2234                         return 0;
2235                 }
2236         }
2237
2238         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2239                                GFP_KERNEL);
2240         if (!devices_info)
2241                 return -ENOMEM;
2242
2243         /* calc min stripe number for data space allocation */
2244         type = btrfs_data_alloc_profile(fs_info);
2245         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2246
2247         if (type & BTRFS_BLOCK_GROUP_RAID0)
2248                 num_stripes = nr_devices;
2249         else if (type & BTRFS_BLOCK_GROUP_RAID1_MASK)
2250                 num_stripes = rattr->ncopies;
2251         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2252                 num_stripes = 4;
2253
2254         /* Adjust for more than 1 stripe per device */
2255         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2256
2257         rcu_read_lock();
2258         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2259                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2260                                                 &device->dev_state) ||
2261                     !device->bdev ||
2262                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2263                         continue;
2264
2265                 if (i >= nr_devices)
2266                         break;
2267
2268                 avail_space = device->total_bytes - device->bytes_used;
2269
2270                 /* align with stripe_len */
2271                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2272
2273                 /*
2274                  * Ensure we have at least min_stripe_size on top of the
2275                  * reserved space on the device.
2276                  */
2277                 if (avail_space <= BTRFS_DEVICE_RANGE_RESERVED + min_stripe_size)
2278                         continue;
2279
2280                 avail_space -= BTRFS_DEVICE_RANGE_RESERVED;
2281
2282                 devices_info[i].dev = device;
2283                 devices_info[i].max_avail = avail_space;
2284
2285                 i++;
2286         }
2287         rcu_read_unlock();
2288
2289         nr_devices = i;
2290
2291         btrfs_descending_sort_devices(devices_info, nr_devices);
2292
2293         i = nr_devices - 1;
2294         avail_space = 0;
2295         while (nr_devices >= rattr->devs_min) {
2296                 num_stripes = min(num_stripes, nr_devices);
2297
2298                 if (devices_info[i].max_avail >= min_stripe_size) {
2299                         int j;
2300                         u64 alloc_size;
2301
2302                         avail_space += devices_info[i].max_avail * num_stripes;
2303                         alloc_size = devices_info[i].max_avail;
2304                         for (j = i + 1 - num_stripes; j <= i; j++)
2305                                 devices_info[j].max_avail -= alloc_size;
2306                 }
2307                 i--;
2308                 nr_devices--;
2309         }
2310
2311         kfree(devices_info);
2312         *free_bytes = avail_space;
2313         return 0;
2314 }
2315
2316 /*
2317  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2318  *
2319  * If there's a redundant raid level at DATA block groups, use the respective
2320  * multiplier to scale the sizes.
2321  *
2322  * Unused device space usage is based on simulating the chunk allocator
2323  * algorithm that respects the device sizes and order of allocations.  This is
2324  * a close approximation of the actual use but there are other factors that may
2325  * change the result (like a new metadata chunk).
2326  *
2327  * If metadata is exhausted, f_bavail will be 0.
2328  */
2329 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2330 {
2331         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2332         struct btrfs_super_block *disk_super = fs_info->super_copy;
2333         struct btrfs_space_info *found;
2334         u64 total_used = 0;
2335         u64 total_free_data = 0;
2336         u64 total_free_meta = 0;
2337         u32 bits = fs_info->sectorsize_bits;
2338         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2339         unsigned factor = 1;
2340         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2341         int ret;
2342         u64 thresh = 0;
2343         int mixed = 0;
2344
2345         list_for_each_entry(found, &fs_info->space_info, list) {
2346                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2347                         int i;
2348
2349                         total_free_data += found->disk_total - found->disk_used;
2350                         total_free_data -=
2351                                 btrfs_account_ro_block_groups_free_space(found);
2352
2353                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2354                                 if (!list_empty(&found->block_groups[i]))
2355                                         factor = btrfs_bg_type_to_factor(
2356                                                 btrfs_raid_array[i].bg_flag);
2357                         }
2358                 }
2359
2360                 /*
2361                  * Metadata in mixed block goup profiles are accounted in data
2362                  */
2363                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2364                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2365                                 mixed = 1;
2366                         else
2367                                 total_free_meta += found->disk_total -
2368                                         found->disk_used;
2369                 }
2370
2371                 total_used += found->disk_used;
2372         }
2373
2374         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2375         buf->f_blocks >>= bits;
2376         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2377
2378         /* Account global block reserve as used, it's in logical size already */
2379         spin_lock(&block_rsv->lock);
2380         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2381         if (buf->f_bfree >= block_rsv->size >> bits)
2382                 buf->f_bfree -= block_rsv->size >> bits;
2383         else
2384                 buf->f_bfree = 0;
2385         spin_unlock(&block_rsv->lock);
2386
2387         buf->f_bavail = div_u64(total_free_data, factor);
2388         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2389         if (ret)
2390                 return ret;
2391         buf->f_bavail += div_u64(total_free_data, factor);
2392         buf->f_bavail = buf->f_bavail >> bits;
2393
2394         /*
2395          * We calculate the remaining metadata space minus global reserve. If
2396          * this is (supposedly) smaller than zero, there's no space. But this
2397          * does not hold in practice, the exhausted state happens where's still
2398          * some positive delta. So we apply some guesswork and compare the
2399          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2400          *
2401          * We probably cannot calculate the exact threshold value because this
2402          * depends on the internal reservations requested by various
2403          * operations, so some operations that consume a few metadata will
2404          * succeed even if the Avail is zero. But this is better than the other
2405          * way around.
2406          */
2407         thresh = SZ_4M;
2408
2409         /*
2410          * We only want to claim there's no available space if we can no longer
2411          * allocate chunks for our metadata profile and our global reserve will
2412          * not fit in the free metadata space.  If we aren't ->full then we
2413          * still can allocate chunks and thus are fine using the currently
2414          * calculated f_bavail.
2415          */
2416         if (!mixed && block_rsv->space_info->full &&
2417             total_free_meta - thresh < block_rsv->size)
2418                 buf->f_bavail = 0;
2419
2420         buf->f_type = BTRFS_SUPER_MAGIC;
2421         buf->f_bsize = dentry->d_sb->s_blocksize;
2422         buf->f_namelen = BTRFS_NAME_LEN;
2423
2424         /* We treat it as constant endianness (it doesn't matter _which_)
2425            because we want the fsid to come out the same whether mounted
2426            on a big-endian or little-endian host */
2427         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2428         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2429         /* Mask in the root object ID too, to disambiguate subvols */
2430         buf->f_fsid.val[0] ^=
2431                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2432         buf->f_fsid.val[1] ^=
2433                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2434
2435         return 0;
2436 }
2437
2438 static void btrfs_kill_super(struct super_block *sb)
2439 {
2440         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2441         kill_anon_super(sb);
2442         btrfs_free_fs_info(fs_info);
2443 }
2444
2445 static struct file_system_type btrfs_fs_type = {
2446         .owner          = THIS_MODULE,
2447         .name           = "btrfs",
2448         .mount          = btrfs_mount,
2449         .kill_sb        = btrfs_kill_super,
2450         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2451 };
2452
2453 static struct file_system_type btrfs_root_fs_type = {
2454         .owner          = THIS_MODULE,
2455         .name           = "btrfs",
2456         .mount          = btrfs_mount_root,
2457         .kill_sb        = btrfs_kill_super,
2458         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA | FS_ALLOW_IDMAP,
2459 };
2460
2461 MODULE_ALIAS_FS("btrfs");
2462
2463 static int btrfs_control_open(struct inode *inode, struct file *file)
2464 {
2465         /*
2466          * The control file's private_data is used to hold the
2467          * transaction when it is started and is used to keep
2468          * track of whether a transaction is already in progress.
2469          */
2470         file->private_data = NULL;
2471         return 0;
2472 }
2473
2474 /*
2475  * Used by /dev/btrfs-control for devices ioctls.
2476  */
2477 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2478                                 unsigned long arg)
2479 {
2480         struct btrfs_ioctl_vol_args *vol;
2481         struct btrfs_device *device = NULL;
2482         dev_t devt = 0;
2483         int ret = -ENOTTY;
2484
2485         if (!capable(CAP_SYS_ADMIN))
2486                 return -EPERM;
2487
2488         vol = memdup_user((void __user *)arg, sizeof(*vol));
2489         if (IS_ERR(vol))
2490                 return PTR_ERR(vol);
2491         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2492
2493         switch (cmd) {
2494         case BTRFS_IOC_SCAN_DEV:
2495                 mutex_lock(&uuid_mutex);
2496                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2497                                                &btrfs_root_fs_type);
2498                 ret = PTR_ERR_OR_ZERO(device);
2499                 mutex_unlock(&uuid_mutex);
2500                 break;
2501         case BTRFS_IOC_FORGET_DEV:
2502                 if (vol->name[0] != 0) {
2503                         ret = lookup_bdev(vol->name, &devt);
2504                         if (ret)
2505                                 break;
2506                 }
2507                 ret = btrfs_forget_devices(devt);
2508                 break;
2509         case BTRFS_IOC_DEVICES_READY:
2510                 mutex_lock(&uuid_mutex);
2511                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2512                                                &btrfs_root_fs_type);
2513                 if (IS_ERR(device)) {
2514                         mutex_unlock(&uuid_mutex);
2515                         ret = PTR_ERR(device);
2516                         break;
2517                 }
2518                 ret = !(device->fs_devices->num_devices ==
2519                         device->fs_devices->total_devices);
2520                 mutex_unlock(&uuid_mutex);
2521                 break;
2522         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2523                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2524                 break;
2525         }
2526
2527         kfree(vol);
2528         return ret;
2529 }
2530
2531 static int btrfs_freeze(struct super_block *sb)
2532 {
2533         struct btrfs_trans_handle *trans;
2534         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2535         struct btrfs_root *root = fs_info->tree_root;
2536
2537         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2538         /*
2539          * We don't need a barrier here, we'll wait for any transaction that
2540          * could be in progress on other threads (and do delayed iputs that
2541          * we want to avoid on a frozen filesystem), or do the commit
2542          * ourselves.
2543          */
2544         trans = btrfs_attach_transaction_barrier(root);
2545         if (IS_ERR(trans)) {
2546                 /* no transaction, don't bother */
2547                 if (PTR_ERR(trans) == -ENOENT)
2548                         return 0;
2549                 return PTR_ERR(trans);
2550         }
2551         return btrfs_commit_transaction(trans);
2552 }
2553
2554 static int check_dev_super(struct btrfs_device *dev)
2555 {
2556         struct btrfs_fs_info *fs_info = dev->fs_info;
2557         struct btrfs_super_block *sb;
2558         u16 csum_type;
2559         int ret = 0;
2560
2561         /* This should be called with fs still frozen. */
2562         ASSERT(test_bit(BTRFS_FS_FROZEN, &fs_info->flags));
2563
2564         /* Missing dev, no need to check. */
2565         if (!dev->bdev)
2566                 return 0;
2567
2568         /* Only need to check the primary super block. */
2569         sb = btrfs_read_dev_one_super(dev->bdev, 0, true);
2570         if (IS_ERR(sb))
2571                 return PTR_ERR(sb);
2572
2573         /* Verify the checksum. */
2574         csum_type = btrfs_super_csum_type(sb);
2575         if (csum_type != btrfs_super_csum_type(fs_info->super_copy)) {
2576                 btrfs_err(fs_info, "csum type changed, has %u expect %u",
2577                           csum_type, btrfs_super_csum_type(fs_info->super_copy));
2578                 ret = -EUCLEAN;
2579                 goto out;
2580         }
2581
2582         if (btrfs_check_super_csum(fs_info, sb)) {
2583                 btrfs_err(fs_info, "csum for on-disk super block no longer matches");
2584                 ret = -EUCLEAN;
2585                 goto out;
2586         }
2587
2588         /* Btrfs_validate_super() includes fsid check against super->fsid. */
2589         ret = btrfs_validate_super(fs_info, sb, 0);
2590         if (ret < 0)
2591                 goto out;
2592
2593         if (btrfs_super_generation(sb) != fs_info->last_trans_committed) {
2594                 btrfs_err(fs_info, "transid mismatch, has %llu expect %llu",
2595                         btrfs_super_generation(sb),
2596                         fs_info->last_trans_committed);
2597                 ret = -EUCLEAN;
2598                 goto out;
2599         }
2600 out:
2601         btrfs_release_disk_super(sb);
2602         return ret;
2603 }
2604
2605 static int btrfs_unfreeze(struct super_block *sb)
2606 {
2607         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2608         struct btrfs_device *device;
2609         int ret = 0;
2610
2611         /*
2612          * Make sure the fs is not changed by accident (like hibernation then
2613          * modified by other OS).
2614          * If we found anything wrong, we mark the fs error immediately.
2615          *
2616          * And since the fs is frozen, no one can modify the fs yet, thus
2617          * we don't need to hold device_list_mutex.
2618          */
2619         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
2620                 ret = check_dev_super(device);
2621                 if (ret < 0) {
2622                         btrfs_handle_fs_error(fs_info, ret,
2623                                 "super block on devid %llu got modified unexpectedly",
2624                                 device->devid);
2625                         break;
2626                 }
2627         }
2628         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2629
2630         /*
2631          * We still return 0, to allow VFS layer to unfreeze the fs even the
2632          * above checks failed. Since the fs is either fine or read-only, we're
2633          * safe to continue, without causing further damage.
2634          */
2635         return 0;
2636 }
2637
2638 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2639 {
2640         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2641
2642         /*
2643          * There should be always a valid pointer in latest_dev, it may be stale
2644          * for a short moment in case it's being deleted but still valid until
2645          * the end of RCU grace period.
2646          */
2647         rcu_read_lock();
2648         seq_escape(m, rcu_str_deref(fs_info->fs_devices->latest_dev->name), " \t\n\\");
2649         rcu_read_unlock();
2650
2651         return 0;
2652 }
2653
2654 static const struct super_operations btrfs_super_ops = {
2655         .drop_inode     = btrfs_drop_inode,
2656         .evict_inode    = btrfs_evict_inode,
2657         .put_super      = btrfs_put_super,
2658         .sync_fs        = btrfs_sync_fs,
2659         .show_options   = btrfs_show_options,
2660         .show_devname   = btrfs_show_devname,
2661         .alloc_inode    = btrfs_alloc_inode,
2662         .destroy_inode  = btrfs_destroy_inode,
2663         .free_inode     = btrfs_free_inode,
2664         .statfs         = btrfs_statfs,
2665         .remount_fs     = btrfs_remount,
2666         .freeze_fs      = btrfs_freeze,
2667         .unfreeze_fs    = btrfs_unfreeze,
2668 };
2669
2670 static const struct file_operations btrfs_ctl_fops = {
2671         .open = btrfs_control_open,
2672         .unlocked_ioctl  = btrfs_control_ioctl,
2673         .compat_ioctl = compat_ptr_ioctl,
2674         .owner   = THIS_MODULE,
2675         .llseek = noop_llseek,
2676 };
2677
2678 static struct miscdevice btrfs_misc = {
2679         .minor          = BTRFS_MINOR,
2680         .name           = "btrfs-control",
2681         .fops           = &btrfs_ctl_fops
2682 };
2683
2684 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2685 MODULE_ALIAS("devname:btrfs-control");
2686
2687 static int __init btrfs_interface_init(void)
2688 {
2689         return misc_register(&btrfs_misc);
2690 }
2691
2692 static __cold void btrfs_interface_exit(void)
2693 {
2694         misc_deregister(&btrfs_misc);
2695 }
2696
2697 static void __init btrfs_print_mod_info(void)
2698 {
2699         static const char options[] = ""
2700 #ifdef CONFIG_BTRFS_DEBUG
2701                         ", debug=on"
2702 #endif
2703 #ifdef CONFIG_BTRFS_ASSERT
2704                         ", assert=on"
2705 #endif
2706 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2707                         ", integrity-checker=on"
2708 #endif
2709 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2710                         ", ref-verify=on"
2711 #endif
2712 #ifdef CONFIG_BLK_DEV_ZONED
2713                         ", zoned=yes"
2714 #else
2715                         ", zoned=no"
2716 #endif
2717 #ifdef CONFIG_FS_VERITY
2718                         ", fsverity=yes"
2719 #else
2720                         ", fsverity=no"
2721 #endif
2722                         ;
2723         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2724 }
2725
2726 static int __init init_btrfs_fs(void)
2727 {
2728         int err;
2729
2730         btrfs_props_init();
2731
2732         err = btrfs_init_sysfs();
2733         if (err)
2734                 return err;
2735
2736         btrfs_init_compress();
2737
2738         err = btrfs_init_cachep();
2739         if (err)
2740                 goto free_compress;
2741
2742         err = extent_state_init_cachep();
2743         if (err)
2744                 goto free_cachep;
2745
2746         err = extent_buffer_init_cachep();
2747         if (err)
2748                 goto free_extent_cachep;
2749
2750         err = btrfs_bioset_init();
2751         if (err)
2752                 goto free_eb_cachep;
2753
2754         err = extent_map_init();
2755         if (err)
2756                 goto free_bioset;
2757
2758         err = ordered_data_init();
2759         if (err)
2760                 goto free_extent_map;
2761
2762         err = btrfs_delayed_inode_init();
2763         if (err)
2764                 goto free_ordered_data;
2765
2766         err = btrfs_auto_defrag_init();
2767         if (err)
2768                 goto free_delayed_inode;
2769
2770         err = btrfs_delayed_ref_init();
2771         if (err)
2772                 goto free_auto_defrag;
2773
2774         err = btrfs_prelim_ref_init();
2775         if (err)
2776                 goto free_delayed_ref;
2777
2778         err = btrfs_interface_init();
2779         if (err)
2780                 goto free_prelim_ref;
2781
2782         btrfs_print_mod_info();
2783
2784         err = btrfs_run_sanity_tests();
2785         if (err)
2786                 goto unregister_ioctl;
2787
2788         err = register_filesystem(&btrfs_fs_type);
2789         if (err)
2790                 goto unregister_ioctl;
2791
2792         return 0;
2793
2794 unregister_ioctl:
2795         btrfs_interface_exit();
2796 free_prelim_ref:
2797         btrfs_prelim_ref_exit();
2798 free_delayed_ref:
2799         btrfs_delayed_ref_exit();
2800 free_auto_defrag:
2801         btrfs_auto_defrag_exit();
2802 free_delayed_inode:
2803         btrfs_delayed_inode_exit();
2804 free_ordered_data:
2805         ordered_data_exit();
2806 free_extent_map:
2807         extent_map_exit();
2808 free_bioset:
2809         btrfs_bioset_exit();
2810 free_eb_cachep:
2811         extent_buffer_free_cachep();
2812 free_extent_cachep:
2813         extent_state_free_cachep();
2814 free_cachep:
2815         btrfs_destroy_cachep();
2816 free_compress:
2817         btrfs_exit_compress();
2818         btrfs_exit_sysfs();
2819
2820         return err;
2821 }
2822
2823 static void __exit exit_btrfs_fs(void)
2824 {
2825         btrfs_destroy_cachep();
2826         btrfs_delayed_ref_exit();
2827         btrfs_auto_defrag_exit();
2828         btrfs_delayed_inode_exit();
2829         btrfs_prelim_ref_exit();
2830         ordered_data_exit();
2831         extent_map_exit();
2832         btrfs_bioset_exit();
2833         extent_state_free_cachep();
2834         extent_buffer_free_cachep();
2835         btrfs_interface_exit();
2836         unregister_filesystem(&btrfs_fs_type);
2837         btrfs_exit_sysfs();
2838         btrfs_cleanup_fs_uuids();
2839         btrfs_exit_compress();
2840 }
2841
2842 late_initcall(init_btrfs_fs);
2843 module_exit(exit_btrfs_fs)
2844
2845 MODULE_LICENSE("GPL");
2846 MODULE_SOFTDEP("pre: crc32c");
2847 MODULE_SOFTDEP("pre: xxhash64");
2848 MODULE_SOFTDEP("pre: sha256");
2849 MODULE_SOFTDEP("pre: blake2b-256");