fs/btrfs/super.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/blkdev.h>
   7 #include <linux/module.h>
   8 #include <linux/fs.h>
   9 #include <linux/pagemap.h>
  10 #include <linux/highmem.h>
  11 #include <linux/time.h>
  12 #include <linux/init.h>
  13 #include <linux/seq_file.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/writeback.h>
  18 #include <linux/statfs.h>
  19 #include <linux/compat.h>
  20 #include <linux/parser.h>
  21 #include <linux/ctype.h>
  22 #include <linux/namei.h>
  23 #include <linux/miscdevice.h>
  24 #include <linux/magic.h>
  25 #include <linux/slab.h>
  26 #include <linux/cleancache.h>
  27 #include <linux/ratelimit.h>
  28 #include <linux/crc32c.h>
  29 #include <linux/btrfs.h>
  30 #include "delayed-inode.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "print-tree.h"
  36 #include "props.h"
  37 #include "xattr.h"
  38 #include "volumes.h"
  39 #include "export.h"
  40 #include "compression.h"
  41 #include "rcu-string.h"
  42 #include "dev-replace.h"
  43 #include "free-space-cache.h"
  44 #include "backref.h"
  45 #include "space-info.h"
  46 #include "sysfs.h"
  47 #include "zoned.h"
  48 #include "tests/btrfs-tests.h"
  49 #include "block-group.h"
  50 #include "discard.h"
  51 #include "qgroup.h"
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/btrfs.h>
  54
  55 static const struct super_operations btrfs_super_ops;
  56
  57 /*
  58  * Types for mounting the default subvolume and a subvolume explicitly
  59  * requested by subvol=/path. That way the callchain is straightforward and we
  60  * don't have to play tricks with the mount options and recursive calls to
  61  * btrfs_mount.
  62  *
  63  * The new btrfs_root_fs_type also servers as a tag for the bdev_holder.
  64  */
  65 static struct file_system_type btrfs_fs_type;
  66 static struct file_system_type btrfs_root_fs_type;
  67
  68 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  69
  70 /*
  71  * Generally the error codes correspond to their respective errors, but there
  72  * are a few special cases.
  73  *
  74  * EUCLEAN: Any sort of corruption that we encounter.  The tree-checker for
  75  *          instance will return EUCLEAN if any of the blocks are corrupted in
  76  *          a way that is problematic.  We want to reserve EUCLEAN for these
  77  *          sort of corruptions.
  78  *
  79  * EROFS: If we check BTRFS_FS_STATE_ERROR and fail out with a return error, we
  80  *        need to use EROFS for this case.  We will have no idea of the
  81  *        original failure, that will have been reported at the time we tripped
  82  *        over the error.  Each subsequent error that doesn't have any context
  83  *        of the original error should use EROFS when handling BTRFS_FS_STATE_ERROR.
  84  */
  85 const char * __attribute_const__ btrfs_decode_error(int errno)
  86 {
  87         char *errstr = "unknown";
  88
  89         switch (errno) {
  90         case -ENOENT:           /* -2 */
  91                 errstr = "No such entry";
  92                 break;
  93         case -EIO:              /* -5 */
  94                 errstr = "IO failure";
  95                 break;
  96         case -ENOMEM:           /* -12*/
  97                 errstr = "Out of memory";
  98                 break;
  99         case -EEXIST:           /* -17 */
 100                 errstr = "Object already exists";
 101                 break;
 102         case -ENOSPC:           /* -28 */
 103                 errstr = "No space left";
 104                 break;
 105         case -EROFS:            /* -30 */
 106                 errstr = "Readonly filesystem";
 107                 break;
 108         case -EOPNOTSUPP:       /* -95 */
 109                 errstr = "Operation not supported";
 110                 break;
 111         case -EUCLEAN:          /* -117 */
 112                 errstr = "Filesystem corrupted";
 113                 break;
 114         case -EDQUOT:           /* -122 */
 115                 errstr = "Quota exceeded";
 116                 break;
 117         }
 118
 119         return errstr;
 120 }
 121
 122 /*
 123  * __btrfs_handle_fs_error decodes expected errors from the caller and
 124  * invokes the appropriate error response.
 125  */
 126 __cold
 127 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 128                        unsigned int line, int errno, const char *fmt, ...)
 129 {
 130         struct super_block *sb = fs_info->sb;
 131 #ifdef CONFIG_PRINTK
 132         const char *errstr;
 133 #endif
 134
 135         /*
 136          * Special case: if the error is EROFS, and we're already
 137          * under SB_RDONLY, then it is safe here.
 138          */
 139         if (errno == -EROFS && sb_rdonly(sb))
 140                 return;
 141
 142 #ifdef CONFIG_PRINTK
 143         errstr = btrfs_decode_error(errno);
 144         if (fmt) {
 145                 struct va_format vaf;
 146                 va_list args;
 147
 148                 va_start(args, fmt);
 149                 vaf.fmt = fmt;
 150                 vaf.va = &args;
 151
 152                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 153                         sb->s_id, function, line, errno, errstr, &vaf);
 154                 va_end(args);
 155         } else {
 156                 pr_crit("BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 157                         sb->s_id, function, line, errno, errstr);
 158         }
 159 #endif
 160
 161         /*
 162          * Today we only save the error info to memory.  Long term we'll
 163          * also send it down to the disk
 164          */
 165         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 166
 167         /* Don't go through full error handling during mount */
 168         if (!(sb->s_flags & SB_BORN))
 169                 return;
 170
 171         if (sb_rdonly(sb))
 172                 return;
 173
 174         btrfs_discard_stop(fs_info);
 175
 176         /* btrfs handle error by forcing the filesystem readonly */
 177         btrfs_set_sb_rdonly(sb);
 178         btrfs_info(fs_info, "forced readonly");
 179         /*
 180          * Note that a running device replace operation is not canceled here
 181          * although there is no way to update the progress. It would add the
 182          * risk of a deadlock, therefore the canceling is omitted. The only
 183          * penalty is that some I/O remains active until the procedure
 184          * completes. The next time when the filesystem is mounted writable
 185          * again, the device replace operation continues.
 186          */
 187 }
 188
 189 #ifdef CONFIG_PRINTK
 190 static const char * const logtypes[] = {
 191         "emergency",
 192         "alert",
 193         "critical",
 194         "error",
 195         "warning",
 196         "notice",
 197         "info",
 198         "debug",
 199 };
 200
 201
 202 /*
 203  * Use one ratelimit state per log level so that a flood of less important
 204  * messages doesn't cause more important ones to be dropped.
 205  */
 206 static struct ratelimit_state printk_limits[] = {
 207         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 208         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 209         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 210         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 211         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 212         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 213         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 214         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 215 };
 216
 217 void __cold btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 218 {
 219         char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1] = "\0";
 220         struct va_format vaf;
 221         va_list args;
 222         int kern_level;
 223         const char *type = logtypes[4];
 224         struct ratelimit_state *ratelimit = &printk_limits[4];
 225
 226         va_start(args, fmt);
 227
 228         while ((kern_level = printk_get_level(fmt)) != 0) {
 229                 size_t size = printk_skip_level(fmt) - fmt;
 230
 231                 if (kern_level >= '0' && kern_level <= '7') {
 232                         memcpy(lvl, fmt,  size);
 233                         lvl[size] = '\0';
 234                         type = logtypes[kern_level - '0'];
 235                         ratelimit = &printk_limits[kern_level - '0'];
 236                 }
 237                 fmt += size;
 238         }
 239
 240         vaf.fmt = fmt;
 241         vaf.va = &args;
 242
 243         if (__ratelimit(ratelimit)) {
 244                 if (fs_info)
 245                         printk("%sBTRFS %s (device %s): %pV\n", lvl, type,
 246                                 fs_info->sb->s_id, &vaf);
 247                 else
 248                         printk("%sBTRFS %s: %pV\n", lvl, type, &vaf);
 249         }
 250
 251         va_end(args);
 252 }
 253 #endif
 254
 255 #if BITS_PER_LONG == 32
 256 void __cold btrfs_warn_32bit_limit(struct btrfs_fs_info *fs_info)
 257 {
 258         if (!test_and_set_bit(BTRFS_FS_32BIT_WARN, &fs_info->flags)) {
 259                 btrfs_warn(fs_info, "reaching 32bit limit for logical addresses");
 260                 btrfs_warn(fs_info,
 261 "due to page cache limit on 32bit systems, btrfs can't access metadata at or beyond %lluT",
 262                            BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 263                 btrfs_warn(fs_info,
 264                            "please consider upgrading to 64bit kernel/hardware");
 265         }
 266 }
 267
 268 void __cold btrfs_err_32bit_limit(struct btrfs_fs_info *fs_info)
 269 {
 270         if (!test_and_set_bit(BTRFS_FS_32BIT_ERROR, &fs_info->flags)) {
 271                 btrfs_err(fs_info, "reached 32bit limit for logical addresses");
 272                 btrfs_err(fs_info,
 273 "due to page cache limit on 32bit systems, metadata beyond %lluT can't be accessed",
 274                           BTRFS_32BIT_MAX_FILE_SIZE >> 40);
 275                 btrfs_err(fs_info,
 276                            "please consider upgrading to 64bit kernel/hardware");
 277         }
 278 }
 279 #endif
 280
 281 /*
 282  * We only mark the transaction aborted and then set the file system read-only.
 283  * This will prevent new transactions from starting or trying to join this
 284  * one.
 285  *
 286  * This means that error recovery at the call site is limited to freeing
 287  * any local memory allocations and passing the error code up without
 288  * further cleanup. The transaction should complete as it normally would
 289  * in the call path but will return -EIO.
 290  *
 291  * We'll complete the cleanup in btrfs_end_transaction and
 292  * btrfs_commit_transaction.
 293  */
 294 __cold
 295 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 296                                const char *function,
 297                                unsigned int line, int errno)
 298 {
 299         struct btrfs_fs_info *fs_info = trans->fs_info;
 300
 301         WRITE_ONCE(trans->aborted, errno);
 302         WRITE_ONCE(trans->transaction->aborted, errno);
 303         /* Wake up anybody who may be waiting on this transaction */
 304         wake_up(&fs_info->transaction_wait);
 305         wake_up(&fs_info->transaction_blocked_wait);
 306         __btrfs_handle_fs_error(fs_info, function, line, errno, NULL);
 307 }
 308 /*
 309  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 310  * issues an alert, and either panics or BUGs, depending on mount options.
 311  */
 312 __cold
 313 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 314                    unsigned int line, int errno, const char *fmt, ...)
 315 {
 316         char *s_id = "<unknown>";
 317         const char *errstr;
 318         struct va_format vaf = { .fmt = fmt };
 319         va_list args;
 320
 321         if (fs_info)
 322                 s_id = fs_info->sb->s_id;
 323
 324         va_start(args, fmt);
 325         vaf.va = &args;
 326
 327         errstr = btrfs_decode_error(errno);
 328         if (fs_info && (btrfs_test_opt(fs_info, PANIC_ON_FATAL_ERROR)))
 329                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 330                         s_id, function, line, &vaf, errno, errstr);
 331
 332         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 333                    function, line, &vaf, errno, errstr);
 334         va_end(args);
 335         /* Caller calls BUG() */
 336 }
 337
 338 static void btrfs_put_super(struct super_block *sb)
 339 {
 340         close_ctree(btrfs_sb(sb));
 341 }
 342
 343 enum {
 344         Opt_acl, Opt_noacl,
 345         Opt_clear_cache,
 346         Opt_commit_interval,
 347         Opt_compress,
 348         Opt_compress_force,
 349         Opt_compress_force_type,
 350         Opt_compress_type,
 351         Opt_degraded,
 352         Opt_device,
 353         Opt_fatal_errors,
 354         Opt_flushoncommit, Opt_noflushoncommit,
 355         Opt_max_inline,
 356         Opt_barrier, Opt_nobarrier,
 357         Opt_datacow, Opt_nodatacow,
 358         Opt_datasum, Opt_nodatasum,
 359         Opt_defrag, Opt_nodefrag,
 360         Opt_discard, Opt_nodiscard,
 361         Opt_discard_mode,
 362         Opt_norecovery,
 363         Opt_ratio,
 364         Opt_rescan_uuid_tree,
 365         Opt_skip_balance,
 366         Opt_space_cache, Opt_no_space_cache,
 367         Opt_space_cache_version,
 368         Opt_ssd, Opt_nossd,
 369         Opt_ssd_spread, Opt_nossd_spread,
 370         Opt_subvol,
 371         Opt_subvol_empty,
 372         Opt_subvolid,
 373         Opt_thread_pool,
 374         Opt_treelog, Opt_notreelog,
 375         Opt_user_subvol_rm_allowed,
 376
 377         /* Rescue options */
 378         Opt_rescue,
 379         Opt_usebackuproot,
 380         Opt_nologreplay,
 381         Opt_ignorebadroots,
 382         Opt_ignoredatacsums,
 383         Opt_rescue_all,
 384
 385         /* Deprecated options */
 386         Opt_recovery,
 387         Opt_inode_cache, Opt_noinode_cache,
 388
 389         /* Debugging options */
 390         Opt_check_integrity,
 391         Opt_check_integrity_including_extent_data,
 392         Opt_check_integrity_print_mask,
 393         Opt_enospc_debug, Opt_noenospc_debug,
 394 #ifdef CONFIG_BTRFS_DEBUG
 395         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 396 #endif
 397 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 398         Opt_ref_verify,
 399 #endif
 400         Opt_err,
 401 };
 402
 403 static const match_table_t tokens = {
 404         {Opt_acl, "acl"},
 405         {Opt_noacl, "noacl"},
 406         {Opt_clear_cache, "clear_cache"},
 407         {Opt_commit_interval, "commit=%u"},
 408         {Opt_compress, "compress"},
 409         {Opt_compress_type, "compress=%s"},
 410         {Opt_compress_force, "compress-force"},
 411         {Opt_compress_force_type, "compress-force=%s"},
 412         {Opt_degraded, "degraded"},
 413         {Opt_device, "device=%s"},
 414         {Opt_fatal_errors, "fatal_errors=%s"},
 415         {Opt_flushoncommit, "flushoncommit"},
 416         {Opt_noflushoncommit, "noflushoncommit"},
 417         {Opt_inode_cache, "inode_cache"},
 418         {Opt_noinode_cache, "noinode_cache"},
 419         {Opt_max_inline, "max_inline=%s"},
 420         {Opt_barrier, "barrier"},
 421         {Opt_nobarrier, "nobarrier"},
 422         {Opt_datacow, "datacow"},
 423         {Opt_nodatacow, "nodatacow"},
 424         {Opt_datasum, "datasum"},
 425         {Opt_nodatasum, "nodatasum"},
 426         {Opt_defrag, "autodefrag"},
 427         {Opt_nodefrag, "noautodefrag"},
 428         {Opt_discard, "discard"},
 429         {Opt_discard_mode, "discard=%s"},
 430         {Opt_nodiscard, "nodiscard"},
 431         {Opt_norecovery, "norecovery"},
 432         {Opt_ratio, "metadata_ratio=%u"},
 433         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 434         {Opt_skip_balance, "skip_balance"},
 435         {Opt_space_cache, "space_cache"},
 436         {Opt_no_space_cache, "nospace_cache"},
 437         {Opt_space_cache_version, "space_cache=%s"},
 438         {Opt_ssd, "ssd"},
 439         {Opt_nossd, "nossd"},
 440         {Opt_ssd_spread, "ssd_spread"},
 441         {Opt_nossd_spread, "nossd_spread"},
 442         {Opt_subvol, "subvol=%s"},
 443         {Opt_subvol_empty, "subvol="},
 444         {Opt_subvolid, "subvolid=%s"},
 445         {Opt_thread_pool, "thread_pool=%u"},
 446         {Opt_treelog, "treelog"},
 447         {Opt_notreelog, "notreelog"},
 448         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 449
 450         /* Rescue options */
 451         {Opt_rescue, "rescue=%s"},
 452         /* Deprecated, with alias rescue=nologreplay */
 453         {Opt_nologreplay, "nologreplay"},
 454         /* Deprecated, with alias rescue=usebackuproot */
 455         {Opt_usebackuproot, "usebackuproot"},
 456
 457         /* Deprecated options */
 458         {Opt_recovery, "recovery"},
 459
 460         /* Debugging options */
 461         {Opt_check_integrity, "check_int"},
 462         {Opt_check_integrity_including_extent_data, "check_int_data"},
 463         {Opt_check_integrity_print_mask, "check_int_print_mask=%u"},
 464         {Opt_enospc_debug, "enospc_debug"},
 465         {Opt_noenospc_debug, "noenospc_debug"},
 466 #ifdef CONFIG_BTRFS_DEBUG
 467         {Opt_fragment_data, "fragment=data"},
 468         {Opt_fragment_metadata, "fragment=metadata"},
 469         {Opt_fragment_all, "fragment=all"},
 470 #endif
 471 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
 472         {Opt_ref_verify, "ref_verify"},
 473 #endif
 474         {Opt_err, NULL},
 475 };
 476
 477 static const match_table_t rescue_tokens = {
 478         {Opt_usebackuproot, "usebackuproot"},
 479         {Opt_nologreplay, "nologreplay"},
 480         {Opt_ignorebadroots, "ignorebadroots"},
 481         {Opt_ignorebadroots, "ibadroots"},
 482         {Opt_ignoredatacsums, "ignoredatacsums"},
 483         {Opt_ignoredatacsums, "idatacsums"},
 484         {Opt_rescue_all, "all"},
 485         {Opt_err, NULL},
 486 };
 487
 488 static bool check_ro_option(struct btrfs_fs_info *fs_info, unsigned long opt,
 489                             const char *opt_name)
 490 {
 491         if (fs_info->mount_opt & opt) {
 492                 btrfs_err(fs_info, "%s must be used with ro mount option",
 493                           opt_name);
 494                 return true;
 495         }
 496         return false;
 497 }
 498
 499 static int parse_rescue_options(struct btrfs_fs_info *info, const char *options)
 500 {
 501         char *opts;
 502         char *orig;
 503         char *p;
 504         substring_t args[MAX_OPT_ARGS];
 505         int ret = 0;
 506
 507         opts = kstrdup(options, GFP_KERNEL);
 508         if (!opts)
 509                 return -ENOMEM;
 510         orig = opts;
 511
 512         while ((p = strsep(&opts, ":")) != NULL) {
 513                 int token;
 514
 515                 if (!*p)
 516                         continue;
 517                 token = match_token(p, rescue_tokens, args);
 518                 switch (token){
 519                 case Opt_usebackuproot:
 520                         btrfs_info(info,
 521                                    "trying to use backup root at mount time");
 522                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 523                         break;
 524                 case Opt_nologreplay:
 525                         btrfs_set_and_info(info, NOLOGREPLAY,
 526                                            "disabling log replay at mount time");
 527                         break;
 528                 case Opt_ignorebadroots:
 529                         btrfs_set_and_info(info, IGNOREBADROOTS,
 530                                            "ignoring bad roots");
 531                         break;
 532                 case Opt_ignoredatacsums:
 533                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 534                                            "ignoring data csums");
 535                         break;
 536                 case Opt_rescue_all:
 537                         btrfs_info(info, "enabling all of the rescue options");
 538                         btrfs_set_and_info(info, IGNOREDATACSUMS,
 539                                            "ignoring data csums");
 540                         btrfs_set_and_info(info, IGNOREBADROOTS,
 541                                            "ignoring bad roots");
 542                         btrfs_set_and_info(info, NOLOGREPLAY,
 543                                            "disabling log replay at mount time");
 544                         break;
 545                 case Opt_err:
 546                         btrfs_info(info, "unrecognized rescue option '%s'", p);
 547                         ret = -EINVAL;
 548                         goto out;
 549                 default:
 550                         break;
 551                 }
 552
 553         }
 554 out:
 555         kfree(orig);
 556         return ret;
 557 }
 558
 559 /*
 560  * Regular mount options parser.  Everything that is needed only when
 561  * reading in a new superblock is parsed here.
 562  * XXX JDM: This needs to be cleaned up for remount.
 563  */
 564 int btrfs_parse_options(struct btrfs_fs_info *info, char *options,
 565                         unsigned long new_flags)
 566 {
 567         substring_t args[MAX_OPT_ARGS];
 568         char *p, *num;
 569         int intarg;
 570         int ret = 0;
 571         char *compress_type;
 572         bool compress_force = false;
 573         enum btrfs_compression_type saved_compress_type;
 574         int saved_compress_level;
 575         bool saved_compress_force;
 576         int no_compress = 0;
 577
 578         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
 579                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 580         else if (btrfs_free_space_cache_v1_active(info)) {
 581                 if (btrfs_is_zoned(info)) {
 582                         btrfs_info(info,
 583                         "zoned: clearing existing space cache");
 584                         btrfs_set_super_cache_generation(info->super_copy, 0);
 585                 } else {
 586                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 587                 }
 588         }
 589
 590         /*
 591          * Even the options are empty, we still need to do extra check
 592          * against new flags
 593          */
 594         if (!options)
 595                 goto check;
 596
 597         while ((p = strsep(&options, ",")) != NULL) {
 598                 int token;
 599                 if (!*p)
 600                         continue;
 601
 602                 token = match_token(p, tokens, args);
 603                 switch (token) {
 604                 case Opt_degraded:
 605                         btrfs_info(info, "allowing degraded mounts");
 606                         btrfs_set_opt(info->mount_opt, DEGRADED);
 607                         break;
 608                 case Opt_subvol:
 609                 case Opt_subvol_empty:
 610                 case Opt_subvolid:
 611                 case Opt_device:
 612                         /*
 613                          * These are parsed by btrfs_parse_subvol_options or
 614                          * btrfs_parse_device_options and can be ignored here.
 615                          */
 616                         break;
 617                 case Opt_nodatasum:
 618                         btrfs_set_and_info(info, NODATASUM,
 619                                            "setting nodatasum");
 620                         break;
 621                 case Opt_datasum:
 622                         if (btrfs_test_opt(info, NODATASUM)) {
 623                                 if (btrfs_test_opt(info, NODATACOW))
 624                                         btrfs_info(info,
 625                                                    "setting datasum, datacow enabled");
 626                                 else
 627                                         btrfs_info(info, "setting datasum");
 628                         }
 629                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 630                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 631                         break;
 632                 case Opt_nodatacow:
 633                         if (!btrfs_test_opt(info, NODATACOW)) {
 634                                 if (!btrfs_test_opt(info, COMPRESS) ||
 635                                     !btrfs_test_opt(info, FORCE_COMPRESS)) {
 636                                         btrfs_info(info,
 637                                                    "setting nodatacow, compression disabled");
 638                                 } else {
 639                                         btrfs_info(info, "setting nodatacow");
 640                                 }
 641                         }
 642                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 643                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 644                         btrfs_set_opt(info->mount_opt, NODATACOW);
 645                         btrfs_set_opt(info->mount_opt, NODATASUM);
 646                         break;
 647                 case Opt_datacow:
 648                         btrfs_clear_and_info(info, NODATACOW,
 649                                              "setting datacow");
 650                         break;
 651                 case Opt_compress_force:
 652                 case Opt_compress_force_type:
 653                         compress_force = true;
 654                         fallthrough;
 655                 case Opt_compress:
 656                 case Opt_compress_type:
 657                         saved_compress_type = btrfs_test_opt(info,
 658                                                              COMPRESS) ?
 659                                 info->compress_type : BTRFS_COMPRESS_NONE;
 660                         saved_compress_force =
 661                                 btrfs_test_opt(info, FORCE_COMPRESS);
 662                         saved_compress_level = info->compress_level;
 663                         if (token == Opt_compress ||
 664                             token == Opt_compress_force ||
 665                             strncmp(args[0].from, "zlib", 4) == 0) {
 666                                 compress_type = "zlib";
 667
 668                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 669                                 info->compress_level = BTRFS_ZLIB_DEFAULT_LEVEL;
 670                                 /*
 671                                  * args[0] contains uninitialized data since
 672                                  * for these tokens we don't expect any
 673                                  * parameter.
 674                                  */
 675                                 if (token != Opt_compress &&
 676                                     token != Opt_compress_force)
 677                                         info->compress_level =
 678                                           btrfs_compress_str2level(
 679                                                         BTRFS_COMPRESS_ZLIB,
 680                                                         args[0].from + 4);
 681                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 682                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 683                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 684                                 no_compress = 0;
 685                         } else if (strncmp(args[0].from, "lzo", 3) == 0) {
 686                                 compress_type = "lzo";
 687                                 info->compress_type = BTRFS_COMPRESS_LZO;
 688                                 info->compress_level = 0;
 689                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 690                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 691                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 692                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 693                                 no_compress = 0;
 694                         } else if (strncmp(args[0].from, "zstd", 4) == 0) {
 695                                 compress_type = "zstd";
 696                                 info->compress_type = BTRFS_COMPRESS_ZSTD;
 697                                 info->compress_level =
 698                                         btrfs_compress_str2level(
 699                                                          BTRFS_COMPRESS_ZSTD,
 700                                                          args[0].from + 4);
 701                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 702                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 703                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 704                                 btrfs_set_fs_incompat(info, COMPRESS_ZSTD);
 705                                 no_compress = 0;
 706                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 707                                 compress_type = "no";
 708                                 info->compress_level = 0;
 709                                 info->compress_type = 0;
 710                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 711                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 712                                 compress_force = false;
 713                                 no_compress++;
 714                         } else {
 715                                 ret = -EINVAL;
 716                                 goto out;
 717                         }
 718
 719                         if (compress_force) {
 720                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 721                         } else {
 722                                 /*
 723                                  * If we remount from compress-force=xxx to
 724                                  * compress=xxx, we need clear FORCE_COMPRESS
 725                                  * flag, otherwise, there is no way for users
 726                                  * to disable forcible compression separately.
 727                                  */
 728                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 729                         }
 730                         if (no_compress == 1) {
 731                                 btrfs_info(info, "use no compression");
 732                         } else if ((info->compress_type != saved_compress_type) ||
 733                                    (compress_force != saved_compress_force) ||
 734                                    (info->compress_level != saved_compress_level)) {
 735                                 btrfs_info(info, "%s %s compression, level %d",
 736                                            (compress_force) ? "force" : "use",
 737                                            compress_type, info->compress_level);
 738                         }
 739                         compress_force = false;
 740                         break;
 741                 case Opt_ssd:
 742                         btrfs_set_and_info(info, SSD,
 743                                            "enabling ssd optimizations");
 744                         btrfs_clear_opt(info->mount_opt, NOSSD);
 745                         break;
 746                 case Opt_ssd_spread:
 747                         btrfs_set_and_info(info, SSD,
 748                                            "enabling ssd optimizations");
 749                         btrfs_set_and_info(info, SSD_SPREAD,
 750                                            "using spread ssd allocation scheme");
 751                         btrfs_clear_opt(info->mount_opt, NOSSD);
 752                         break;
 753                 case Opt_nossd:
 754                         btrfs_set_opt(info->mount_opt, NOSSD);
 755                         btrfs_clear_and_info(info, SSD,
 756                                              "not using ssd optimizations");
 757                         fallthrough;
 758                 case Opt_nossd_spread:
 759                         btrfs_clear_and_info(info, SSD_SPREAD,
 760                                              "not using spread ssd allocation scheme");
 761                         break;
 762                 case Opt_barrier:
 763                         btrfs_clear_and_info(info, NOBARRIER,
 764                                              "turning on barriers");
 765                         break;
 766                 case Opt_nobarrier:
 767                         btrfs_set_and_info(info, NOBARRIER,
 768                                            "turning off barriers");
 769                         break;
 770                 case Opt_thread_pool:
 771                         ret = match_int(&args[0], &intarg);
 772                         if (ret) {
 773                                 goto out;
 774                         } else if (intarg == 0) {
 775                                 ret = -EINVAL;
 776                                 goto out;
 777                         }
 778                         info->thread_pool_size = intarg;
 779                         break;
 780                 case Opt_max_inline:
 781                         num = match_strdup(&args[0]);
 782                         if (num) {
 783                                 info->max_inline = memparse(num, NULL);
 784                                 kfree(num);
 785
 786                                 if (info->max_inline) {
 787                                         info->max_inline = min_t(u64,
 788                                                 info->max_inline,
 789                                                 info->sectorsize);
 790                                 }
 791                                 btrfs_info(info, "max_inline at %llu",
 792                                            info->max_inline);
 793                         } else {
 794                                 ret = -ENOMEM;
 795                                 goto out;
 796                         }
 797                         break;
 798                 case Opt_acl:
 799 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 800                         info->sb->s_flags |= SB_POSIXACL;
 801                         break;
 802 #else
 803                         btrfs_err(info, "support for ACL not compiled in!");
 804                         ret = -EINVAL;
 805                         goto out;
 806 #endif
 807                 case Opt_noacl:
 808                         info->sb->s_flags &= ~SB_POSIXACL;
 809                         break;
 810                 case Opt_notreelog:
 811                         btrfs_set_and_info(info, NOTREELOG,
 812                                            "disabling tree log");
 813                         break;
 814                 case Opt_treelog:
 815                         btrfs_clear_and_info(info, NOTREELOG,
 816                                              "enabling tree log");
 817                         break;
 818                 case Opt_norecovery:
 819                 case Opt_nologreplay:
 820                         btrfs_warn(info,
 821                 "'nologreplay' is deprecated, use 'rescue=nologreplay' instead");
 822                         btrfs_set_and_info(info, NOLOGREPLAY,
 823                                            "disabling log replay at mount time");
 824                         break;
 825                 case Opt_flushoncommit:
 826                         btrfs_set_and_info(info, FLUSHONCOMMIT,
 827                                            "turning on flush-on-commit");
 828                         break;
 829                 case Opt_noflushoncommit:
 830                         btrfs_clear_and_info(info, FLUSHONCOMMIT,
 831                                              "turning off flush-on-commit");
 832                         break;
 833                 case Opt_ratio:
 834                         ret = match_int(&args[0], &intarg);
 835                         if (ret)
 836                                 goto out;
 837                         info->metadata_ratio = intarg;
 838                         btrfs_info(info, "metadata ratio %u",
 839                                    info->metadata_ratio);
 840                         break;
 841                 case Opt_discard:
 842                 case Opt_discard_mode:
 843                         if (token == Opt_discard ||
 844                             strcmp(args[0].from, "sync") == 0) {
 845                                 btrfs_clear_opt(info->mount_opt, DISCARD_ASYNC);
 846                                 btrfs_set_and_info(info, DISCARD_SYNC,
 847                                                    "turning on sync discard");
 848                         } else if (strcmp(args[0].from, "async") == 0) {
 849                                 btrfs_clear_opt(info->mount_opt, DISCARD_SYNC);
 850                                 btrfs_set_and_info(info, DISCARD_ASYNC,
 851                                                    "turning on async discard");
 852                         } else {
 853                                 ret = -EINVAL;
 854                                 goto out;
 855                         }
 856                         break;
 857                 case Opt_nodiscard:
 858                         btrfs_clear_and_info(info, DISCARD_SYNC,
 859                                              "turning off discard");
 860                         btrfs_clear_and_info(info, DISCARD_ASYNC,
 861                                              "turning off async discard");
 862                         break;
 863                 case Opt_space_cache:
 864                 case Opt_space_cache_version:
 865                         if (token == Opt_space_cache ||
 866                             strcmp(args[0].from, "v1") == 0) {
 867                                 btrfs_clear_opt(info->mount_opt,
 868                                                 FREE_SPACE_TREE);
 869                                 btrfs_set_and_info(info, SPACE_CACHE,
 870                                            "enabling disk space caching");
 871                         } else if (strcmp(args[0].from, "v2") == 0) {
 872                                 btrfs_clear_opt(info->mount_opt,
 873                                                 SPACE_CACHE);
 874                                 btrfs_set_and_info(info, FREE_SPACE_TREE,
 875                                                    "enabling free space tree");
 876                         } else {
 877                                 ret = -EINVAL;
 878                                 goto out;
 879                         }
 880                         break;
 881                 case Opt_rescan_uuid_tree:
 882                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 883                         break;
 884                 case Opt_no_space_cache:
 885                         if (btrfs_test_opt(info, SPACE_CACHE)) {
 886                                 btrfs_clear_and_info(info, SPACE_CACHE,
 887                                              "disabling disk space caching");
 888                         }
 889                         if (btrfs_test_opt(info, FREE_SPACE_TREE)) {
 890                                 btrfs_clear_and_info(info, FREE_SPACE_TREE,
 891                                              "disabling free space tree");
 892                         }
 893                         break;
 894                 case Opt_inode_cache:
 895                 case Opt_noinode_cache:
 896                         btrfs_warn(info,
 897         "the 'inode_cache' option is deprecated and has no effect since 5.11");
 898                         break;
 899                 case Opt_clear_cache:
 900                         btrfs_set_and_info(info, CLEAR_CACHE,
 901                                            "force clearing of disk cache");
 902                         break;
 903                 case Opt_user_subvol_rm_allowed:
 904                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 905                         break;
 906                 case Opt_enospc_debug:
 907                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 908                         break;
 909                 case Opt_noenospc_debug:
 910                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 911                         break;
 912                 case Opt_defrag:
 913                         btrfs_set_and_info(info, AUTO_DEFRAG,
 914                                            "enabling auto defrag");
 915                         break;
 916                 case Opt_nodefrag:
 917                         btrfs_clear_and_info(info, AUTO_DEFRAG,
 918                                              "disabling auto defrag");
 919                         break;
 920                 case Opt_recovery:
 921                 case Opt_usebackuproot:
 922                         btrfs_warn(info,
 923                         "'%s' is deprecated, use 'rescue=usebackuproot' instead",
 924                                    token == Opt_recovery ? "recovery" :
 925                                    "usebackuproot");
 926                         btrfs_info(info,
 927                                    "trying to use backup root at mount time");
 928                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 929                         break;
 930                 case Opt_skip_balance:
 931                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 932                         break;
 933 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 934                 case Opt_check_integrity_including_extent_data:
 935                         btrfs_info(info,
 936                                    "enabling check integrity including extent data");
 937                         btrfs_set_opt(info->mount_opt,
 938                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 939                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 940                         break;
 941                 case Opt_check_integrity:
 942                         btrfs_info(info, "enabling check integrity");
 943                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 944                         break;
 945                 case Opt_check_integrity_print_mask:
 946                         ret = match_int(&args[0], &intarg);
 947                         if (ret)
 948                                 goto out;
 949                         info->check_integrity_print_mask = intarg;
 950                         btrfs_info(info, "check_integrity_print_mask 0x%x",
 951                                    info->check_integrity_print_mask);
 952                         break;
 953 #else
 954                 case Opt_check_integrity_including_extent_data:
 955                 case Opt_check_integrity:
 956                 case Opt_check_integrity_print_mask:
 957                         btrfs_err(info,
 958                                   "support for check_integrity* not compiled in!");
 959                         ret = -EINVAL;
 960                         goto out;
 961 #endif
 962                 case Opt_fatal_errors:
 963                         if (strcmp(args[0].from, "panic") == 0)
 964                                 btrfs_set_opt(info->mount_opt,
 965                                               PANIC_ON_FATAL_ERROR);
 966                         else if (strcmp(args[0].from, "bug") == 0)
 967                                 btrfs_clear_opt(info->mount_opt,
 968                                               PANIC_ON_FATAL_ERROR);
 969                         else {
 970                                 ret = -EINVAL;
 971                                 goto out;
 972                         }
 973                         break;
 974                 case Opt_commit_interval:
 975                         intarg = 0;
 976                         ret = match_int(&args[0], &intarg);
 977                         if (ret)
 978                                 goto out;
 979                         if (intarg == 0) {
 980                                 btrfs_info(info,
 981                                            "using default commit interval %us",
 982                                            BTRFS_DEFAULT_COMMIT_INTERVAL);
 983                                 intarg = BTRFS_DEFAULT_COMMIT_INTERVAL;
 984                         } else if (intarg > 300) {
 985                                 btrfs_warn(info, "excessive commit interval %d",
 986                                            intarg);
 987                         }
 988                         info->commit_interval = intarg;
 989                         break;
 990                 case Opt_rescue:
 991                         ret = parse_rescue_options(info, args[0].from);
 992                         if (ret < 0)
 993                                 goto out;
 994                         break;
 995 #ifdef CONFIG_BTRFS_DEBUG
 996                 case Opt_fragment_all:
 997                         btrfs_info(info, "fragmenting all space");
 998                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 999                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
1000                         break;
1001                 case Opt_fragment_metadata:
1002                         btrfs_info(info, "fragmenting metadata");
1003                         btrfs_set_opt(info->mount_opt,
1004                                       FRAGMENT_METADATA);
1005                         break;
1006                 case Opt_fragment_data:
1007                         btrfs_info(info, "fragmenting data");
1008                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
1009                         break;
1010 #endif
1011 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
1012                 case Opt_ref_verify:
1013                         btrfs_info(info, "doing ref verification");
1014                         btrfs_set_opt(info->mount_opt, REF_VERIFY);
1015                         break;
1016 #endif
1017                 case Opt_err:
1018                         btrfs_err(info, "unrecognized mount option '%s'", p);
1019                         ret = -EINVAL;
1020                         goto out;
1021                 default:
1022                         break;
1023                 }
1024         }
1025 check:
1026         /* We're read-only, don't have to check. */
1027         if (new_flags & SB_RDONLY)
1028                 goto out;
1029
1030         if (check_ro_option(info, BTRFS_MOUNT_NOLOGREPLAY, "nologreplay") ||
1031             check_ro_option(info, BTRFS_MOUNT_IGNOREBADROOTS, "ignorebadroots") ||
1032             check_ro_option(info, BTRFS_MOUNT_IGNOREDATACSUMS, "ignoredatacsums"))
1033                 ret = -EINVAL;
1034 out:
1035         if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE) &&
1036             !btrfs_test_opt(info, FREE_SPACE_TREE) &&
1037             !btrfs_test_opt(info, CLEAR_CACHE)) {
1038                 btrfs_err(info, "cannot disable free space tree");
1039                 ret = -EINVAL;
1040
1041         }
1042         if (!ret)
1043                 ret = btrfs_check_mountopts_zoned(info);
1044         if (!ret && btrfs_test_opt(info, SPACE_CACHE))
1045                 btrfs_info(info, "disk space caching is enabled");
1046         if (!ret && btrfs_test_opt(info, FREE_SPACE_TREE))
1047                 btrfs_info(info, "using free space tree");
1048         return ret;
1049 }
1050
1051 /*
1052  * Parse mount options that are required early in the mount process.
1053  *
1054  * All other options will be parsed on much later in the mount process and
1055  * only when we need to allocate a new super block.
1056  */
1057 static int btrfs_parse_device_options(const char *options, fmode_t flags,
1058                                       void *holder)
1059 {
1060         substring_t args[MAX_OPT_ARGS];
1061         char *device_name, *opts, *orig, *p;
1062         struct btrfs_device *device = NULL;
1063         int error = 0;
1064
1065         lockdep_assert_held(&uuid_mutex);
1066
1067         if (!options)
1068                 return 0;
1069
1070         /*
1071          * strsep changes the string, duplicate it because btrfs_parse_options
1072          * gets called later
1073          */
1074         opts = kstrdup(options, GFP_KERNEL);
1075         if (!opts)
1076                 return -ENOMEM;
1077         orig = opts;
1078
1079         while ((p = strsep(&opts, ",")) != NULL) {
1080                 int token;
1081
1082                 if (!*p)
1083                         continue;
1084
1085                 token = match_token(p, tokens, args);
1086                 if (token == Opt_device) {
1087                         device_name = match_strdup(&args[0]);
1088                         if (!device_name) {
1089                                 error = -ENOMEM;
1090                                 goto out;
1091                         }
1092                         device = btrfs_scan_one_device(device_name, flags,
1093                                         holder);
1094                         kfree(device_name);
1095                         if (IS_ERR(device)) {
1096                                 error = PTR_ERR(device);
1097                                 goto out;
1098                         }
1099                 }
1100         }
1101
1102 out:
1103         kfree(orig);
1104         return error;
1105 }
1106
1107 /*
1108  * Parse mount options that are related to subvolume id
1109  *
1110  * The value is later passed to mount_subvol()
1111  */
1112 static int btrfs_parse_subvol_options(const char *options, char **subvol_name,
1113                 u64 *subvol_objectid)
1114 {
1115         substring_t args[MAX_OPT_ARGS];
1116         char *opts, *orig, *p;
1117         int error = 0;
1118         u64 subvolid;
1119
1120         if (!options)
1121                 return 0;
1122
1123         /*
1124          * strsep changes the string, duplicate it because
1125          * btrfs_parse_device_options gets called later
1126          */
1127         opts = kstrdup(options, GFP_KERNEL);
1128         if (!opts)
1129                 return -ENOMEM;
1130         orig = opts;
1131
1132         while ((p = strsep(&opts, ",")) != NULL) {
1133                 int token;
1134                 if (!*p)
1135                         continue;
1136
1137                 token = match_token(p, tokens, args);
1138                 switch (token) {
1139                 case Opt_subvol:
1140                         kfree(*subvol_name);
1141                         *subvol_name = match_strdup(&args[0]);
1142                         if (!*subvol_name) {
1143                                 error = -ENOMEM;
1144                                 goto out;
1145                         }
1146                         break;
1147                 case Opt_subvolid:
1148                         error = match_u64(&args[0], &subvolid);
1149                         if (error)
1150                                 goto out;
1151
1152                         /* we want the original fs_tree */
1153                         if (subvolid == 0)
1154                                 subvolid = BTRFS_FS_TREE_OBJECTID;
1155
1156                         *subvol_objectid = subvolid;
1157                         break;
1158                 default:
1159                         break;
1160                 }
1161         }
1162
1163 out:
1164         kfree(orig);
1165         return error;
1166 }
1167
1168 char *btrfs_get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
1169                                           u64 subvol_objectid)
1170 {
1171         struct btrfs_root *root = fs_info->tree_root;
1172         struct btrfs_root *fs_root = NULL;
1173         struct btrfs_root_ref *root_ref;
1174         struct btrfs_inode_ref *inode_ref;
1175         struct btrfs_key key;
1176         struct btrfs_path *path = NULL;
1177         char *name = NULL, *ptr;
1178         u64 dirid;
1179         int len;
1180         int ret;
1181
1182         path = btrfs_alloc_path();
1183         if (!path) {
1184                 ret = -ENOMEM;
1185                 goto err;
1186         }
1187
1188         name = kmalloc(PATH_MAX, GFP_KERNEL);
1189         if (!name) {
1190                 ret = -ENOMEM;
1191                 goto err;
1192         }
1193         ptr = name + PATH_MAX - 1;
1194         ptr[0] = '\0';
1195
1196         /*
1197          * Walk up the subvolume trees in the tree of tree roots by root
1198          * backrefs until we hit the top-level subvolume.
1199          */
1200         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1201                 key.objectid = subvol_objectid;
1202                 key.type = BTRFS_ROOT_BACKREF_KEY;
1203                 key.offset = (u64)-1;
1204
1205                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1206                 if (ret < 0) {
1207                         goto err;
1208                 } else if (ret > 0) {
1209                         ret = btrfs_previous_item(root, path, subvol_objectid,
1210                                                   BTRFS_ROOT_BACKREF_KEY);
1211                         if (ret < 0) {
1212                                 goto err;
1213                         } else if (ret > 0) {
1214                                 ret = -ENOENT;
1215                                 goto err;
1216                         }
1217                 }
1218
1219                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1220                 subvol_objectid = key.offset;
1221
1222                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1223                                           struct btrfs_root_ref);
1224                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
1225                 ptr -= len + 1;
1226                 if (ptr < name) {
1227                         ret = -ENAMETOOLONG;
1228                         goto err;
1229                 }
1230                 read_extent_buffer(path->nodes[0], ptr + 1,
1231                                    (unsigned long)(root_ref + 1), len);
1232                 ptr[0] = '/';
1233                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1234                 btrfs_release_path(path);
1235
1236                 fs_root = btrfs_get_fs_root(fs_info, subvol_objectid, true);
1237                 if (IS_ERR(fs_root)) {
1238                         ret = PTR_ERR(fs_root);
1239                         fs_root = NULL;
1240                         goto err;
1241                 }
1242
1243                 /*
1244                  * Walk up the filesystem tree by inode refs until we hit the
1245                  * root directory.
1246                  */
1247                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1248                         key.objectid = dirid;
1249                         key.type = BTRFS_INODE_REF_KEY;
1250                         key.offset = (u64)-1;
1251
1252                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1253                         if (ret < 0) {
1254                                 goto err;
1255                         } else if (ret > 0) {
1256                                 ret = btrfs_previous_item(fs_root, path, dirid,
1257                                                           BTRFS_INODE_REF_KEY);
1258                                 if (ret < 0) {
1259                                         goto err;
1260                                 } else if (ret > 0) {
1261                                         ret = -ENOENT;
1262                                         goto err;
1263                                 }
1264                         }
1265
1266                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1267                         dirid = key.offset;
1268
1269                         inode_ref = btrfs_item_ptr(path->nodes[0],
1270                                                    path->slots[0],
1271                                                    struct btrfs_inode_ref);
1272                         len = btrfs_inode_ref_name_len(path->nodes[0],
1273                                                        inode_ref);
1274                         ptr -= len + 1;
1275                         if (ptr < name) {
1276                                 ret = -ENAMETOOLONG;
1277                                 goto err;
1278                         }
1279                         read_extent_buffer(path->nodes[0], ptr + 1,
1280                                            (unsigned long)(inode_ref + 1), len);
1281                         ptr[0] = '/';
1282                         btrfs_release_path(path);
1283                 }
1284                 btrfs_put_root(fs_root);
1285                 fs_root = NULL;
1286         }
1287
1288         btrfs_free_path(path);
1289         if (ptr == name + PATH_MAX - 1) {
1290                 name[0] = '/';
1291                 name[1] = '\0';
1292         } else {
1293                 memmove(name, ptr, name + PATH_MAX - ptr);
1294         }
1295         return name;
1296
1297 err:
1298         btrfs_put_root(fs_root);
1299         btrfs_free_path(path);
1300         kfree(name);
1301         return ERR_PTR(ret);
1302 }
1303
1304 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1305 {
1306         struct btrfs_root *root = fs_info->tree_root;
1307         struct btrfs_dir_item *di;
1308         struct btrfs_path *path;
1309         struct btrfs_key location;
1310         u64 dir_id;
1311
1312         path = btrfs_alloc_path();
1313         if (!path)
1314                 return -ENOMEM;
1315
1316         /*
1317          * Find the "default" dir item which points to the root item that we
1318          * will mount by default if we haven't been given a specific subvolume
1319          * to mount.
1320          */
1321         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1322         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1323         if (IS_ERR(di)) {
1324                 btrfs_free_path(path);
1325                 return PTR_ERR(di);
1326         }
1327         if (!di) {
1328                 /*
1329                  * Ok the default dir item isn't there.  This is weird since
1330                  * it's always been there, but don't freak out, just try and
1331                  * mount the top-level subvolume.
1332                  */
1333                 btrfs_free_path(path);
1334                 *objectid = BTRFS_FS_TREE_OBJECTID;
1335                 return 0;
1336         }
1337
1338         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1339         btrfs_free_path(path);
1340         *objectid = location.objectid;
1341         return 0;
1342 }
1343
1344 static int btrfs_fill_super(struct super_block *sb,
1345                             struct btrfs_fs_devices *fs_devices,
1346                             void *data)
1347 {
1348         struct inode *inode;
1349         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1350         int err;
1351
1352         sb->s_maxbytes = MAX_LFS_FILESIZE;
1353         sb->s_magic = BTRFS_SUPER_MAGIC;
1354         sb->s_op = &btrfs_super_ops;
1355         sb->s_d_op = &btrfs_dentry_operations;
1356         sb->s_export_op = &btrfs_export_ops;
1357         sb->s_xattr = btrfs_xattr_handlers;
1358         sb->s_time_gran = 1;
1359 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1360         sb->s_flags |= SB_POSIXACL;
1361 #endif
1362         sb->s_flags |= SB_I_VERSION;
1363         sb->s_iflags |= SB_I_CGROUPWB;
1364
1365         err = super_setup_bdi(sb);
1366         if (err) {
1367                 btrfs_err(fs_info, "super_setup_bdi failed");
1368                 return err;
1369         }
1370
1371         err = open_ctree(sb, fs_devices, (char *)data);
1372         if (err) {
1373                 btrfs_err(fs_info, "open_ctree failed");
1374                 return err;
1375         }
1376
1377         inode = btrfs_iget(sb, BTRFS_FIRST_FREE_OBJECTID, fs_info->fs_root);
1378         if (IS_ERR(inode)) {
1379                 err = PTR_ERR(inode);
1380                 goto fail_close;
1381         }
1382
1383         sb->s_root = d_make_root(inode);
1384         if (!sb->s_root) {
1385                 err = -ENOMEM;
1386                 goto fail_close;
1387         }
1388
1389         cleancache_init_fs(sb);
1390         sb->s_flags |= SB_ACTIVE;
1391         return 0;
1392
1393 fail_close:
1394         close_ctree(fs_info);
1395         return err;
1396 }
1397
1398 int btrfs_sync_fs(struct super_block *sb, int wait)
1399 {
1400         struct btrfs_trans_handle *trans;
1401         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1402         struct btrfs_root *root = fs_info->tree_root;
1403
1404         trace_btrfs_sync_fs(fs_info, wait);
1405
1406         if (!wait) {
1407                 filemap_flush(fs_info->btree_inode->i_mapping);
1408                 return 0;
1409         }
1410
1411         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1412
1413         trans = btrfs_attach_transaction_barrier(root);
1414         if (IS_ERR(trans)) {
1415                 /* no transaction, don't bother */
1416                 if (PTR_ERR(trans) == -ENOENT) {
1417                         /*
1418                          * Exit unless we have some pending changes
1419                          * that need to go through commit
1420                          */
1421                         if (fs_info->pending_changes == 0)
1422                                 return 0;
1423                         /*
1424                          * A non-blocking test if the fs is frozen. We must not
1425                          * start a new transaction here otherwise a deadlock
1426                          * happens. The pending operations are delayed to the
1427                          * next commit after thawing.
1428                          */
1429                         if (sb_start_write_trylock(sb))
1430                                 sb_end_write(sb);
1431                         else
1432                                 return 0;
1433                         trans = btrfs_start_transaction(root, 0);
1434                 }
1435                 if (IS_ERR(trans))
1436                         return PTR_ERR(trans);
1437         }
1438         return btrfs_commit_transaction(trans);
1439 }
1440
1441 static void print_rescue_option(struct seq_file *seq, const char *s, bool *printed)
1442 {
1443         seq_printf(seq, "%s%s", (*printed) ? ":" : ",rescue=", s);
1444         *printed = true;
1445 }
1446
1447 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1448 {
1449         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1450         const char *compress_type;
1451         const char *subvol_name;
1452         bool printed = false;
1453
1454         if (btrfs_test_opt(info, DEGRADED))
1455                 seq_puts(seq, ",degraded");
1456         if (btrfs_test_opt(info, NODATASUM))
1457                 seq_puts(seq, ",nodatasum");
1458         if (btrfs_test_opt(info, NODATACOW))
1459                 seq_puts(seq, ",nodatacow");
1460         if (btrfs_test_opt(info, NOBARRIER))
1461                 seq_puts(seq, ",nobarrier");
1462         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1463                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1464         if (info->thread_pool_size !=  min_t(unsigned long,
1465                                              num_online_cpus() + 2, 8))
1466                 seq_printf(seq, ",thread_pool=%u", info->thread_pool_size);
1467         if (btrfs_test_opt(info, COMPRESS)) {
1468                 compress_type = btrfs_compress_type2str(info->compress_type);
1469                 if (btrfs_test_opt(info, FORCE_COMPRESS))
1470                         seq_printf(seq, ",compress-force=%s", compress_type);
1471                 else
1472                         seq_printf(seq, ",compress=%s", compress_type);
1473                 if (info->compress_level)
1474                         seq_printf(seq, ":%d", info->compress_level);
1475         }
1476         if (btrfs_test_opt(info, NOSSD))
1477                 seq_puts(seq, ",nossd");
1478         if (btrfs_test_opt(info, SSD_SPREAD))
1479                 seq_puts(seq, ",ssd_spread");
1480         else if (btrfs_test_opt(info, SSD))
1481                 seq_puts(seq, ",ssd");
1482         if (btrfs_test_opt(info, NOTREELOG))
1483                 seq_puts(seq, ",notreelog");
1484         if (btrfs_test_opt(info, NOLOGREPLAY))
1485                 print_rescue_option(seq, "nologreplay", &printed);
1486         if (btrfs_test_opt(info, USEBACKUPROOT))
1487                 print_rescue_option(seq, "usebackuproot", &printed);
1488         if (btrfs_test_opt(info, IGNOREBADROOTS))
1489                 print_rescue_option(seq, "ignorebadroots", &printed);
1490         if (btrfs_test_opt(info, IGNOREDATACSUMS))
1491                 print_rescue_option(seq, "ignoredatacsums", &printed);
1492         if (btrfs_test_opt(info, FLUSHONCOMMIT))
1493                 seq_puts(seq, ",flushoncommit");
1494         if (btrfs_test_opt(info, DISCARD_SYNC))
1495                 seq_puts(seq, ",discard");
1496         if (btrfs_test_opt(info, DISCARD_ASYNC))
1497                 seq_puts(seq, ",discard=async");
1498         if (!(info->sb->s_flags & SB_POSIXACL))
1499                 seq_puts(seq, ",noacl");
1500         if (btrfs_free_space_cache_v1_active(info))
1501                 seq_puts(seq, ",space_cache");
1502         else if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
1503                 seq_puts(seq, ",space_cache=v2");
1504         else
1505                 seq_puts(seq, ",nospace_cache");
1506         if (btrfs_test_opt(info, RESCAN_UUID_TREE))
1507                 seq_puts(seq, ",rescan_uuid_tree");
1508         if (btrfs_test_opt(info, CLEAR_CACHE))
1509                 seq_puts(seq, ",clear_cache");
1510         if (btrfs_test_opt(info, USER_SUBVOL_RM_ALLOWED))
1511                 seq_puts(seq, ",user_subvol_rm_allowed");
1512         if (btrfs_test_opt(info, ENOSPC_DEBUG))
1513                 seq_puts(seq, ",enospc_debug");
1514         if (btrfs_test_opt(info, AUTO_DEFRAG))
1515                 seq_puts(seq, ",autodefrag");
1516         if (btrfs_test_opt(info, SKIP_BALANCE))
1517                 seq_puts(seq, ",skip_balance");
1518 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1519         if (btrfs_test_opt(info, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1520                 seq_puts(seq, ",check_int_data");
1521         else if (btrfs_test_opt(info, CHECK_INTEGRITY))
1522                 seq_puts(seq, ",check_int");
1523         if (info->check_integrity_print_mask)
1524                 seq_printf(seq, ",check_int_print_mask=%d",
1525                                 info->check_integrity_print_mask);
1526 #endif
1527         if (info->metadata_ratio)
1528                 seq_printf(seq, ",metadata_ratio=%u", info->metadata_ratio);
1529         if (btrfs_test_opt(info, PANIC_ON_FATAL_ERROR))
1530                 seq_puts(seq, ",fatal_errors=panic");
1531         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1532                 seq_printf(seq, ",commit=%u", info->commit_interval);
1533 #ifdef CONFIG_BTRFS_DEBUG
1534         if (btrfs_test_opt(info, FRAGMENT_DATA))
1535                 seq_puts(seq, ",fragment=data");
1536         if (btrfs_test_opt(info, FRAGMENT_METADATA))
1537                 seq_puts(seq, ",fragment=metadata");
1538 #endif
1539         if (btrfs_test_opt(info, REF_VERIFY))
1540                 seq_puts(seq, ",ref_verify");
1541         seq_printf(seq, ",subvolid=%llu",
1542                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1543         subvol_name = btrfs_get_subvol_name_from_objectid(info,
1544                         BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1545         if (!IS_ERR(subvol_name)) {
1546                 seq_puts(seq, ",subvol=");
1547                 seq_escape(seq, subvol_name, " \t\n\\");
1548                 kfree(subvol_name);
1549         }
1550         return 0;
1551 }
1552
1553 static int btrfs_test_super(struct super_block *s, void *data)
1554 {
1555         struct btrfs_fs_info *p = data;
1556         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1557
1558         return fs_info->fs_devices == p->fs_devices;
1559 }
1560
1561 static int btrfs_set_super(struct super_block *s, void *data)
1562 {
1563         int err = set_anon_super(s, data);
1564         if (!err)
1565                 s->s_fs_info = data;
1566         return err;
1567 }
1568
1569 /*
1570  * subvolumes are identified by ino 256
1571  */
1572 static inline int is_subvolume_inode(struct inode *inode)
1573 {
1574         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1575                 return 1;
1576         return 0;
1577 }
1578
1579 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1580                                    struct vfsmount *mnt)
1581 {
1582         struct dentry *root;
1583         int ret;
1584
1585         if (!subvol_name) {
1586                 if (!subvol_objectid) {
1587                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1588                                                           &subvol_objectid);
1589                         if (ret) {
1590                                 root = ERR_PTR(ret);
1591                                 goto out;
1592                         }
1593                 }
1594                 subvol_name = btrfs_get_subvol_name_from_objectid(
1595                                         btrfs_sb(mnt->mnt_sb), subvol_objectid);
1596                 if (IS_ERR(subvol_name)) {
1597                         root = ERR_CAST(subvol_name);
1598                         subvol_name = NULL;
1599                         goto out;
1600                 }
1601
1602         }
1603
1604         root = mount_subtree(mnt, subvol_name);
1605         /* mount_subtree() drops our reference on the vfsmount. */
1606         mnt = NULL;
1607
1608         if (!IS_ERR(root)) {
1609                 struct super_block *s = root->d_sb;
1610                 struct btrfs_fs_info *fs_info = btrfs_sb(s);
1611                 struct inode *root_inode = d_inode(root);
1612                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1613
1614                 ret = 0;
1615                 if (!is_subvolume_inode(root_inode)) {
1616                         btrfs_err(fs_info, "'%s' is not a valid subvolume",
1617                                subvol_name);
1618                         ret = -EINVAL;
1619                 }
1620                 if (subvol_objectid && root_objectid != subvol_objectid) {
1621                         /*
1622                          * This will also catch a race condition where a
1623                          * subvolume which was passed by ID is renamed and
1624                          * another subvolume is renamed over the old location.
1625                          */
1626                         btrfs_err(fs_info,
1627                                   "subvol '%s' does not match subvolid %llu",
1628                                   subvol_name, subvol_objectid);
1629                         ret = -EINVAL;
1630                 }
1631                 if (ret) {
1632                         dput(root);
1633                         root = ERR_PTR(ret);
1634                         deactivate_locked_super(s);
1635                 }
1636         }
1637
1638 out:
1639         mntput(mnt);
1640         kfree(subvol_name);
1641         return root;
1642 }
1643
1644 /*
1645  * Find a superblock for the given device / mount point.
1646  *
1647  * Note: This is based on mount_bdev from fs/super.c with a few additions
1648  *       for multiple device setup.  Make sure to keep it in sync.
1649  */
1650 static struct dentry *btrfs_mount_root(struct file_system_type *fs_type,
1651                 int flags, const char *device_name, void *data)
1652 {
1653         struct block_device *bdev = NULL;
1654         struct super_block *s;
1655         struct btrfs_device *device = NULL;
1656         struct btrfs_fs_devices *fs_devices = NULL;
1657         struct btrfs_fs_info *fs_info = NULL;
1658         void *new_sec_opts = NULL;
1659         fmode_t mode = FMODE_READ;
1660         int error = 0;
1661
1662         if (!(flags & SB_RDONLY))
1663                 mode |= FMODE_WRITE;
1664
1665         if (data) {
1666                 error = security_sb_eat_lsm_opts(data, &new_sec_opts);
1667                 if (error)
1668                         return ERR_PTR(error);
1669         }
1670
1671         /*
1672          * Setup a dummy root and fs_info for test/set super.  This is because
1673          * we don't actually fill this stuff out until open_ctree, but we need
1674          * then open_ctree will properly initialize the file system specific
1675          * settings later.  btrfs_init_fs_info initializes the static elements
1676          * of the fs_info (locks and such) to make cleanup easier if we find a
1677          * superblock with our given fs_devices later on at sget() time.
1678          */
1679         fs_info = kvzalloc(sizeof(struct btrfs_fs_info), GFP_KERNEL);
1680         if (!fs_info) {
1681                 error = -ENOMEM;
1682                 goto error_sec_opts;
1683         }
1684         btrfs_init_fs_info(fs_info);
1685
1686         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1687         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_KERNEL);
1688         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1689                 error = -ENOMEM;
1690                 goto error_fs_info;
1691         }
1692
1693         mutex_lock(&uuid_mutex);
1694         error = btrfs_parse_device_options(data, mode, fs_type);
1695         if (error) {
1696                 mutex_unlock(&uuid_mutex);
1697                 goto error_fs_info;
1698         }
1699
1700         device = btrfs_scan_one_device(device_name, mode, fs_type);
1701         if (IS_ERR(device)) {
1702                 mutex_unlock(&uuid_mutex);
1703                 error = PTR_ERR(device);
1704                 goto error_fs_info;
1705         }
1706
1707         fs_devices = device->fs_devices;
1708         fs_info->fs_devices = fs_devices;
1709
1710         error = btrfs_open_devices(fs_devices, mode, fs_type);
1711         mutex_unlock(&uuid_mutex);
1712         if (error)
1713                 goto error_fs_info;
1714
1715         if (!(flags & SB_RDONLY) && fs_devices->rw_devices == 0) {
1716                 error = -EACCES;
1717                 goto error_close_devices;
1718         }
1719
1720         bdev = fs_devices->latest_bdev;
1721         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | SB_NOSEC,
1722                  fs_info);
1723         if (IS_ERR(s)) {
1724                 error = PTR_ERR(s);
1725                 goto error_close_devices;
1726         }
1727
1728         if (s->s_root) {
1729                 btrfs_close_devices(fs_devices);
1730                 btrfs_free_fs_info(fs_info);
1731                 if ((flags ^ s->s_flags) & SB_RDONLY)
1732                         error = -EBUSY;
1733         } else {
1734                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1735                 btrfs_sb(s)->bdev_holder = fs_type;
1736                 if (!strstr(crc32c_impl(), "generic"))
1737                         set_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags);
1738                 error = btrfs_fill_super(s, fs_devices, data);
1739         }
1740         if (!error)
1741                 error = security_sb_set_mnt_opts(s, new_sec_opts, 0, NULL);
1742         security_free_mnt_opts(&new_sec_opts);
1743         if (error) {
1744                 deactivate_locked_super(s);
1745                 return ERR_PTR(error);
1746         }
1747
1748         return dget(s->s_root);
1749
1750 error_close_devices:
1751         btrfs_close_devices(fs_devices);
1752 error_fs_info:
1753         btrfs_free_fs_info(fs_info);
1754 error_sec_opts:
1755         security_free_mnt_opts(&new_sec_opts);
1756         return ERR_PTR(error);
1757 }
1758
1759 /*
1760  * Mount function which is called by VFS layer.
1761  *
1762  * In order to allow mounting a subvolume directly, btrfs uses mount_subtree()
1763  * which needs vfsmount* of device's root (/).  This means device's root has to
1764  * be mounted internally in any case.
1765  *
1766  * Operation flow:
1767  *   1. Parse subvol id related options for later use in mount_subvol().
1768  *
1769  *   2. Mount device's root (/) by calling vfs_kern_mount().
1770  *
1771  *      NOTE: vfs_kern_mount() is used by VFS to call btrfs_mount() in the
1772  *      first place. In order to avoid calling btrfs_mount() again, we use
1773  *      different file_system_type which is not registered to VFS by
1774  *      register_filesystem() (btrfs_root_fs_type). As a result,
1775  *      btrfs_mount_root() is called. The return value will be used by
1776  *      mount_subtree() in mount_subvol().
1777  *
1778  *   3. Call mount_subvol() to get the dentry of subvolume. Since there is
1779  *      "btrfs subvolume set-default", mount_subvol() is called always.
1780  */
1781 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1782                 const char *device_name, void *data)
1783 {
1784         struct vfsmount *mnt_root;
1785         struct dentry *root;
1786         char *subvol_name = NULL;
1787         u64 subvol_objectid = 0;
1788         int error = 0;
1789
1790         error = btrfs_parse_subvol_options(data, &subvol_name,
1791                                         &subvol_objectid);
1792         if (error) {
1793                 kfree(subvol_name);
1794                 return ERR_PTR(error);
1795         }
1796
1797         /* mount device's root (/) */
1798         mnt_root = vfs_kern_mount(&btrfs_root_fs_type, flags, device_name, data);
1799         if (PTR_ERR_OR_ZERO(mnt_root) == -EBUSY) {
1800                 if (flags & SB_RDONLY) {
1801                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1802                                 flags & ~SB_RDONLY, device_name, data);
1803                 } else {
1804                         mnt_root = vfs_kern_mount(&btrfs_root_fs_type,
1805                                 flags | SB_RDONLY, device_name, data);
1806                         if (IS_ERR(mnt_root)) {
1807                                 root = ERR_CAST(mnt_root);
1808                                 kfree(subvol_name);
1809                                 goto out;
1810                         }
1811
1812                         down_write(&mnt_root->mnt_sb->s_umount);
1813                         error = btrfs_remount(mnt_root->mnt_sb, &flags, NULL);
1814                         up_write(&mnt_root->mnt_sb->s_umount);
1815                         if (error < 0) {
1816                                 root = ERR_PTR(error);
1817                                 mntput(mnt_root);
1818                                 kfree(subvol_name);
1819                                 goto out;
1820                         }
1821                 }
1822         }
1823         if (IS_ERR(mnt_root)) {
1824                 root = ERR_CAST(mnt_root);
1825                 kfree(subvol_name);
1826                 goto out;
1827         }
1828
1829         /* mount_subvol() will free subvol_name and mnt_root */
1830         root = mount_subvol(subvol_name, subvol_objectid, mnt_root);
1831
1832 out:
1833         return root;
1834 }
1835
1836 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1837                                      u32 new_pool_size, u32 old_pool_size)
1838 {
1839         if (new_pool_size == old_pool_size)
1840                 return;
1841
1842         fs_info->thread_pool_size = new_pool_size;
1843
1844         btrfs_info(fs_info, "resize thread pool %d -> %d",
1845                old_pool_size, new_pool_size);
1846
1847         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1848         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1849         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1850         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1851         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1852         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1853                                 new_pool_size);
1854         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1855         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1856         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1857         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1858         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1859                                 new_pool_size);
1860 }
1861
1862 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1863                                        unsigned long old_opts, int flags)
1864 {
1865         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1866             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1867              (flags & SB_RDONLY))) {
1868                 /* wait for any defraggers to finish */
1869                 wait_event(fs_info->transaction_wait,
1870                            (atomic_read(&fs_info->defrag_running) == 0));
1871                 if (flags & SB_RDONLY)
1872                         sync_filesystem(fs_info->sb);
1873         }
1874 }
1875
1876 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1877                                          unsigned long old_opts)
1878 {
1879         const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE);
1880
1881         /*
1882          * We need to cleanup all defragable inodes if the autodefragment is
1883          * close or the filesystem is read only.
1884          */
1885         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1886             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) || sb_rdonly(fs_info->sb))) {
1887                 btrfs_cleanup_defrag_inodes(fs_info);
1888         }
1889
1890         /* If we toggled discard async */
1891         if (!btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1892             btrfs_test_opt(fs_info, DISCARD_ASYNC))
1893                 btrfs_discard_resume(fs_info);
1894         else if (btrfs_raw_test_opt(old_opts, DISCARD_ASYNC) &&
1895                  !btrfs_test_opt(fs_info, DISCARD_ASYNC))
1896                 btrfs_discard_cleanup(fs_info);
1897
1898         /* If we toggled space cache */
1899         if (cache_opt != btrfs_free_space_cache_v1_active(fs_info))
1900                 btrfs_set_free_space_cache_v1_active(fs_info, cache_opt);
1901 }
1902
1903 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1904 {
1905         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1906         unsigned old_flags = sb->s_flags;
1907         unsigned long old_opts = fs_info->mount_opt;
1908         unsigned long old_compress_type = fs_info->compress_type;
1909         u64 old_max_inline = fs_info->max_inline;
1910         u32 old_thread_pool_size = fs_info->thread_pool_size;
1911         u32 old_metadata_ratio = fs_info->metadata_ratio;
1912         int ret;
1913
1914         sync_filesystem(sb);
1915         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1916
1917         if (data) {
1918                 void *new_sec_opts = NULL;
1919
1920                 ret = security_sb_eat_lsm_opts(data, &new_sec_opts);
1921                 if (!ret)
1922                         ret = security_sb_remount(sb, new_sec_opts);
1923                 security_free_mnt_opts(&new_sec_opts);
1924                 if (ret)
1925                         goto restore;
1926         }
1927
1928         ret = btrfs_parse_options(fs_info, data, *flags);
1929         if (ret)
1930                 goto restore;
1931
1932         btrfs_remount_begin(fs_info, old_opts, *flags);
1933         btrfs_resize_thread_pool(fs_info,
1934                 fs_info->thread_pool_size, old_thread_pool_size);
1935
1936         if ((bool)btrfs_test_opt(fs_info, FREE_SPACE_TREE) !=
1937             (bool)btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
1938             (!sb_rdonly(sb) || (*flags & SB_RDONLY))) {
1939                 btrfs_warn(fs_info,
1940                 "remount supports changing free space tree only from ro to rw");
1941                 /* Make sure free space cache options match the state on disk */
1942                 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
1943                         btrfs_set_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1944                         btrfs_clear_opt(fs_info->mount_opt, SPACE_CACHE);
1945                 }
1946                 if (btrfs_free_space_cache_v1_active(fs_info)) {
1947                         btrfs_clear_opt(fs_info->mount_opt, FREE_SPACE_TREE);
1948                         btrfs_set_opt(fs_info->mount_opt, SPACE_CACHE);
1949                 }
1950         }
1951
1952         if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1953                 goto out;
1954
1955         if (*flags & SB_RDONLY) {
1956                 /*
1957                  * this also happens on 'umount -rf' or on shutdown, when
1958                  * the filesystem is busy.
1959                  */
1960                 cancel_work_sync(&fs_info->async_reclaim_work);
1961                 cancel_work_sync(&fs_info->async_data_reclaim_work);
1962
1963                 btrfs_discard_cleanup(fs_info);
1964
1965                 /* wait for the uuid_scan task to finish */
1966                 down(&fs_info->uuid_tree_rescan_sem);
1967                 /* avoid complains from lockdep et al. */
1968                 up(&fs_info->uuid_tree_rescan_sem);
1969
1970                 btrfs_set_sb_rdonly(sb);
1971
1972                 /*
1973                  * Setting SB_RDONLY will put the cleaner thread to
1974                  * sleep at the next loop if it's already active.
1975                  * If it's already asleep, we'll leave unused block
1976                  * groups on disk until we're mounted read-write again
1977                  * unless we clean them up here.
1978                  */
1979                 btrfs_delete_unused_bgs(fs_info);
1980
1981                 /*
1982                  * The cleaner task could be already running before we set the
1983                  * flag BTRFS_FS_STATE_RO (and SB_RDONLY in the superblock).
1984                  * We must make sure that after we finish the remount, i.e. after
1985                  * we call btrfs_commit_super(), the cleaner can no longer start
1986                  * a transaction - either because it was dropping a dead root,
1987                  * running delayed iputs or deleting an unused block group (the
1988                  * cleaner picked a block group from the list of unused block
1989                  * groups before we were able to in the previous call to
1990                  * btrfs_delete_unused_bgs()).
1991                  */
1992                 wait_on_bit(&fs_info->flags, BTRFS_FS_CLEANER_RUNNING,
1993                             TASK_UNINTERRUPTIBLE);
1994
1995                 /*
1996                  * We've set the superblock to RO mode, so we might have made
1997                  * the cleaner task sleep without running all pending delayed
1998                  * iputs. Go through all the delayed iputs here, so that if an
1999                  * unmount happens without remounting RW we don't end up at
2000                  * finishing close_ctree() with a non-empty list of delayed
2001                  * iputs.
2002                  */
2003                 btrfs_run_delayed_iputs(fs_info);
2004
2005                 btrfs_dev_replace_suspend_for_unmount(fs_info);
2006                 btrfs_scrub_cancel(fs_info);
2007                 btrfs_pause_balance(fs_info);
2008
2009                 /*
2010                  * Pause the qgroup rescan worker if it is running. We don't want
2011                  * it to be still running after we are in RO mode, as after that,
2012                  * by the time we unmount, it might have left a transaction open,
2013                  * so we would leak the transaction and/or crash.
2014                  */
2015                 btrfs_qgroup_wait_for_completion(fs_info, false);
2016
2017                 ret = btrfs_commit_super(fs_info);
2018                 if (ret)
2019                         goto restore;
2020         } else {
2021                 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2022                         btrfs_err(fs_info,
2023                                 "Remounting read-write after error is not allowed");
2024                         ret = -EINVAL;
2025                         goto restore;
2026                 }
2027                 if (fs_info->fs_devices->rw_devices == 0) {
2028                         ret = -EACCES;
2029                         goto restore;
2030                 }
2031
2032                 if (!btrfs_check_rw_degradable(fs_info, NULL)) {
2033                         btrfs_warn(fs_info,
2034                 "too many missing devices, writable remount is not allowed");
2035                         ret = -EACCES;
2036                         goto restore;
2037                 }
2038
2039                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
2040                         btrfs_warn(fs_info,
2041                 "mount required to replay tree-log, cannot remount read-write");
2042                         ret = -EINVAL;
2043                         goto restore;
2044                 }
2045                 if (fs_info->sectorsize < PAGE_SIZE) {
2046                         btrfs_warn(fs_info,
2047         "read-write mount is not yet allowed for sectorsize %u page size %lu",
2048                                    fs_info->sectorsize, PAGE_SIZE);
2049                         ret = -EINVAL;
2050                         goto restore;
2051                 }
2052
2053                 /*
2054                  * NOTE: when remounting with a change that does writes, don't
2055                  * put it anywhere above this point, as we are not sure to be
2056                  * safe to write until we pass the above checks.
2057                  */
2058                 ret = btrfs_start_pre_rw_mount(fs_info);
2059                 if (ret)
2060                         goto restore;
2061
2062                 btrfs_clear_sb_rdonly(sb);
2063
2064                 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
2065         }
2066 out:
2067         /*
2068          * We need to set SB_I_VERSION here otherwise it'll get cleared by VFS,
2069          * since the absence of the flag means it can be toggled off by remount.
2070          */
2071         *flags |= SB_I_VERSION;
2072
2073         wake_up_process(fs_info->transaction_kthread);
2074         btrfs_remount_cleanup(fs_info, old_opts);
2075         btrfs_clear_oneshot_options(fs_info);
2076         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2077
2078         return 0;
2079
2080 restore:
2081         /* We've hit an error - don't reset SB_RDONLY */
2082         if (sb_rdonly(sb))
2083                 old_flags |= SB_RDONLY;
2084         if (!(old_flags & SB_RDONLY))
2085                 clear_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state);
2086         sb->s_flags = old_flags;
2087         fs_info->mount_opt = old_opts;
2088         fs_info->compress_type = old_compress_type;
2089         fs_info->max_inline = old_max_inline;
2090         btrfs_resize_thread_pool(fs_info,
2091                 old_thread_pool_size, fs_info->thread_pool_size);
2092         fs_info->metadata_ratio = old_metadata_ratio;
2093         btrfs_remount_cleanup(fs_info, old_opts);
2094         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2095
2096         return ret;
2097 }
2098
2099 /* Used to sort the devices by max_avail(descending sort) */
2100 static inline int btrfs_cmp_device_free_bytes(const void *dev_info1,
2101                                        const void *dev_info2)
2102 {
2103         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2104             ((struct btrfs_device_info *)dev_info2)->max_avail)
2105                 return -1;
2106         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2107                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2108                 return 1;
2109         else
2110         return 0;
2111 }
2112
2113 /*
2114  * sort the devices by max_avail, in which max free extent size of each device
2115  * is stored.(Descending Sort)
2116  */
2117 static inline void btrfs_descending_sort_devices(
2118                                         struct btrfs_device_info *devices,
2119                                         size_t nr_devices)
2120 {
2121         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
2122              btrfs_cmp_device_free_bytes, NULL);
2123 }
2124
2125 /*
2126  * The helper to calc the free space on the devices that can be used to store
2127  * file data.
2128  */
2129 static inline int btrfs_calc_avail_data_space(struct btrfs_fs_info *fs_info,
2130                                               u64 *free_bytes)
2131 {
2132         struct btrfs_device_info *devices_info;
2133         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2134         struct btrfs_device *device;
2135         u64 type;
2136         u64 avail_space;
2137         u64 min_stripe_size;
2138         int num_stripes = 1;
2139         int i = 0, nr_devices;
2140         const struct btrfs_raid_attr *rattr;
2141
2142         /*
2143          * We aren't under the device list lock, so this is racy-ish, but good
2144          * enough for our purposes.
2145          */
2146         nr_devices = fs_info->fs_devices->open_devices;
2147         if (!nr_devices) {
2148                 smp_mb();
2149                 nr_devices = fs_info->fs_devices->open_devices;
2150                 ASSERT(nr_devices);
2151                 if (!nr_devices) {
2152                         *free_bytes = 0;
2153                         return 0;
2154                 }
2155         }
2156
2157         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
2158                                GFP_KERNEL);
2159         if (!devices_info)
2160                 return -ENOMEM;
2161
2162         /* calc min stripe number for data space allocation */
2163         type = btrfs_data_alloc_profile(fs_info);
2164         rattr = &btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)];
2165
2166         if (type & BTRFS_BLOCK_GROUP_RAID0)
2167                 num_stripes = nr_devices;
2168         else if (type & BTRFS_BLOCK_GROUP_RAID1)
2169                 num_stripes = 2;
2170         else if (type & BTRFS_BLOCK_GROUP_RAID1C3)
2171                 num_stripes = 3;
2172         else if (type & BTRFS_BLOCK_GROUP_RAID1C4)
2173                 num_stripes = 4;
2174         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2175                 num_stripes = 4;
2176
2177         /* Adjust for more than 1 stripe per device */
2178         min_stripe_size = rattr->dev_stripes * BTRFS_STRIPE_LEN;
2179
2180         rcu_read_lock();
2181         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2182                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2183                                                 &device->dev_state) ||
2184                     !device->bdev ||
2185                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2186                         continue;
2187
2188                 if (i >= nr_devices)
2189                         break;
2190
2191                 avail_space = device->total_bytes - device->bytes_used;
2192
2193                 /* align with stripe_len */
2194                 avail_space = rounddown(avail_space, BTRFS_STRIPE_LEN);
2195
2196                 /*
2197                  * In order to avoid overwriting the superblock on the drive,
2198                  * btrfs starts at an offset of at least 1MB when doing chunk
2199                  * allocation.
2200                  *
2201                  * This ensures we have at least min_stripe_size free space
2202                  * after excluding 1MB.
2203                  */
2204                 if (avail_space <= SZ_1M + min_stripe_size)
2205                         continue;
2206
2207                 avail_space -= SZ_1M;
2208
2209                 devices_info[i].dev = device;
2210                 devices_info[i].max_avail = avail_space;
2211
2212                 i++;
2213         }
2214         rcu_read_unlock();
2215
2216         nr_devices = i;
2217
2218         btrfs_descending_sort_devices(devices_info, nr_devices);
2219
2220         i = nr_devices - 1;
2221         avail_space = 0;
2222         while (nr_devices >= rattr->devs_min) {
2223                 num_stripes = min(num_stripes, nr_devices);
2224
2225                 if (devices_info[i].max_avail >= min_stripe_size) {
2226                         int j;
2227                         u64 alloc_size;
2228
2229                         avail_space += devices_info[i].max_avail * num_stripes;
2230                         alloc_size = devices_info[i].max_avail;
2231                         for (j = i + 1 - num_stripes; j <= i; j++)
2232                                 devices_info[j].max_avail -= alloc_size;
2233                 }
2234                 i--;
2235                 nr_devices--;
2236         }
2237
2238         kfree(devices_info);
2239         *free_bytes = avail_space;
2240         return 0;
2241 }
2242
2243 /*
2244  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2245  *
2246  * If there's a redundant raid level at DATA block groups, use the respective
2247  * multiplier to scale the sizes.
2248  *
2249  * Unused device space usage is based on simulating the chunk allocator
2250  * algorithm that respects the device sizes and order of allocations.  This is
2251  * a close approximation of the actual use but there are other factors that may
2252  * change the result (like a new metadata chunk).
2253  *
2254  * If metadata is exhausted, f_bavail will be 0.
2255  */
2256 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2257 {
2258         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2259         struct btrfs_super_block *disk_super = fs_info->super_copy;
2260         struct btrfs_space_info *found;
2261         u64 total_used = 0;
2262         u64 total_free_data = 0;
2263         u64 total_free_meta = 0;
2264         u32 bits = fs_info->sectorsize_bits;
2265         __be32 *fsid = (__be32 *)fs_info->fs_devices->fsid;
2266         unsigned factor = 1;
2267         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2268         int ret;
2269         u64 thresh = 0;
2270         int mixed = 0;
2271
2272         list_for_each_entry(found, &fs_info->space_info, list) {
2273                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2274                         int i;
2275
2276                         total_free_data += found->disk_total - found->disk_used;
2277                         total_free_data -=
2278                                 btrfs_account_ro_block_groups_free_space(found);
2279
2280                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2281                                 if (!list_empty(&found->block_groups[i]))
2282                                         factor = btrfs_bg_type_to_factor(
2283                                                 btrfs_raid_array[i].bg_flag);
2284                         }
2285                 }
2286
2287                 /*
2288                  * Metadata in mixed block goup profiles are accounted in data
2289                  */
2290                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2291                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2292                                 mixed = 1;
2293                         else
2294                                 total_free_meta += found->disk_total -
2295                                         found->disk_used;
2296                 }
2297
2298                 total_used += found->disk_used;
2299         }
2300
2301         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2302         buf->f_blocks >>= bits;
2303         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2304
2305         /* Account global block reserve as used, it's in logical size already */
2306         spin_lock(&block_rsv->lock);
2307         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2308         if (buf->f_bfree >= block_rsv->size >> bits)
2309                 buf->f_bfree -= block_rsv->size >> bits;
2310         else
2311                 buf->f_bfree = 0;
2312         spin_unlock(&block_rsv->lock);
2313
2314         buf->f_bavail = div_u64(total_free_data, factor);
2315         ret = btrfs_calc_avail_data_space(fs_info, &total_free_data);
2316         if (ret)
2317                 return ret;
2318         buf->f_bavail += div_u64(total_free_data, factor);
2319         buf->f_bavail = buf->f_bavail >> bits;
2320
2321         /*
2322          * We calculate the remaining metadata space minus global reserve. If
2323          * this is (supposedly) smaller than zero, there's no space. But this
2324          * does not hold in practice, the exhausted state happens where's still
2325          * some positive delta. So we apply some guesswork and compare the
2326          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2327          *
2328          * We probably cannot calculate the exact threshold value because this
2329          * depends on the internal reservations requested by various
2330          * operations, so some operations that consume a few metadata will
2331          * succeed even if the Avail is zero. But this is better than the other
2332          * way around.
2333          */
2334         thresh = SZ_4M;
2335
2336         /*
2337          * We only want to claim there's no available space if we can no longer
2338          * allocate chunks for our metadata profile and our global reserve will
2339          * not fit in the free metadata space.  If we aren't ->full then we
2340          * still can allocate chunks and thus are fine using the currently
2341          * calculated f_bavail.
2342          */
2343         if (!mixed && block_rsv->space_info->full &&
2344             total_free_meta - thresh < block_rsv->size)
2345                 buf->f_bavail = 0;
2346
2347         buf->f_type = BTRFS_SUPER_MAGIC;
2348         buf->f_bsize = dentry->d_sb->s_blocksize;
2349         buf->f_namelen = BTRFS_NAME_LEN;
2350
2351         /* We treat it as constant endianness (it doesn't matter _which_)
2352            because we want the fsid to come out the same whether mounted
2353            on a big-endian or little-endian host */
2354         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2355         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2356         /* Mask in the root object ID too, to disambiguate subvols */
2357         buf->f_fsid.val[0] ^=
2358                 BTRFS_I(d_inode(dentry))->root->root_key.objectid >> 32;
2359         buf->f_fsid.val[1] ^=
2360                 BTRFS_I(d_inode(dentry))->root->root_key.objectid;
2361
2362         return 0;
2363 }
2364
2365 static void btrfs_kill_super(struct super_block *sb)
2366 {
2367         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2368         kill_anon_super(sb);
2369         btrfs_free_fs_info(fs_info);
2370 }
2371
2372 static struct file_system_type btrfs_fs_type = {
2373         .owner          = THIS_MODULE,
2374         .name           = "btrfs",
2375         .mount          = btrfs_mount,
2376         .kill_sb        = btrfs_kill_super,
2377         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2378 };
2379
2380 static struct file_system_type btrfs_root_fs_type = {
2381         .owner          = THIS_MODULE,
2382         .name           = "btrfs",
2383         .mount          = btrfs_mount_root,
2384         .kill_sb        = btrfs_kill_super,
2385         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2386 };
2387
2388 MODULE_ALIAS_FS("btrfs");
2389
2390 static int btrfs_control_open(struct inode *inode, struct file *file)
2391 {
2392         /*
2393          * The control file's private_data is used to hold the
2394          * transaction when it is started and is used to keep
2395          * track of whether a transaction is already in progress.
2396          */
2397         file->private_data = NULL;
2398         return 0;
2399 }
2400
2401 /*
2402  * Used by /dev/btrfs-control for devices ioctls.
2403  */
2404 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2405                                 unsigned long arg)
2406 {
2407         struct btrfs_ioctl_vol_args *vol;
2408         struct btrfs_device *device = NULL;
2409         int ret = -ENOTTY;
2410
2411         if (!capable(CAP_SYS_ADMIN))
2412                 return -EPERM;
2413
2414         vol = memdup_user((void __user *)arg, sizeof(*vol));
2415         if (IS_ERR(vol))
2416                 return PTR_ERR(vol);
2417         vol->name[BTRFS_PATH_NAME_MAX] = '\0';
2418
2419         switch (cmd) {
2420         case BTRFS_IOC_SCAN_DEV:
2421                 mutex_lock(&uuid_mutex);
2422                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2423                                                &btrfs_root_fs_type);
2424                 ret = PTR_ERR_OR_ZERO(device);
2425                 mutex_unlock(&uuid_mutex);
2426                 break;
2427         case BTRFS_IOC_FORGET_DEV:
2428                 ret = btrfs_forget_devices(vol->name);
2429                 break;
2430         case BTRFS_IOC_DEVICES_READY:
2431                 mutex_lock(&uuid_mutex);
2432                 device = btrfs_scan_one_device(vol->name, FMODE_READ,
2433                                                &btrfs_root_fs_type);
2434                 if (IS_ERR(device)) {
2435                         mutex_unlock(&uuid_mutex);
2436                         ret = PTR_ERR(device);
2437                         break;
2438                 }
2439                 ret = !(device->fs_devices->num_devices ==
2440                         device->fs_devices->total_devices);
2441                 mutex_unlock(&uuid_mutex);
2442                 break;
2443         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2444                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2445                 break;
2446         }
2447
2448         kfree(vol);
2449         return ret;
2450 }
2451
2452 static int btrfs_freeze(struct super_block *sb)
2453 {
2454         struct btrfs_trans_handle *trans;
2455         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2456         struct btrfs_root *root = fs_info->tree_root;
2457
2458         set_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2459         /*
2460          * We don't need a barrier here, we'll wait for any transaction that
2461          * could be in progress on other threads (and do delayed iputs that
2462          * we want to avoid on a frozen filesystem), or do the commit
2463          * ourselves.
2464          */
2465         trans = btrfs_attach_transaction_barrier(root);
2466         if (IS_ERR(trans)) {
2467                 /* no transaction, don't bother */
2468                 if (PTR_ERR(trans) == -ENOENT)
2469                         return 0;
2470                 return PTR_ERR(trans);
2471         }
2472         return btrfs_commit_transaction(trans);
2473 }
2474
2475 static int btrfs_unfreeze(struct super_block *sb)
2476 {
2477         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2478
2479         clear_bit(BTRFS_FS_FROZEN, &fs_info->flags);
2480         return 0;
2481 }
2482
2483 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2484 {
2485         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2486         struct btrfs_device *dev, *first_dev = NULL;
2487
2488         /*
2489          * Lightweight locking of the devices. We should not need
2490          * device_list_mutex here as we only read the device data and the list
2491          * is protected by RCU.  Even if a device is deleted during the list
2492          * traversals, we'll get valid data, the freeing callback will wait at
2493          * least until the rcu_read_unlock.
2494          */
2495         rcu_read_lock();
2496         list_for_each_entry_rcu(dev, &fs_info->fs_devices->devices, dev_list) {
2497                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
2498                         continue;
2499                 if (!dev->name)
2500                         continue;
2501                 if (!first_dev || dev->devid < first_dev->devid)
2502                         first_dev = dev;
2503         }
2504
2505         if (first_dev)
2506                 seq_escape(m, rcu_str_deref(first_dev->name), " \t\n\\");
2507         else
2508                 WARN_ON(1);
2509         rcu_read_unlock();
2510         return 0;
2511 }
2512
2513 static const struct super_operations btrfs_super_ops = {
2514         .drop_inode     = btrfs_drop_inode,
2515         .evict_inode    = btrfs_evict_inode,
2516         .put_super      = btrfs_put_super,
2517         .sync_fs        = btrfs_sync_fs,
2518         .show_options   = btrfs_show_options,
2519         .show_devname   = btrfs_show_devname,
2520         .alloc_inode    = btrfs_alloc_inode,
2521         .destroy_inode  = btrfs_destroy_inode,
2522         .free_inode     = btrfs_free_inode,
2523         .statfs         = btrfs_statfs,
2524         .remount_fs     = btrfs_remount,
2525         .freeze_fs      = btrfs_freeze,
2526         .unfreeze_fs    = btrfs_unfreeze,
2527 };
2528
2529 static const struct file_operations btrfs_ctl_fops = {
2530         .open = btrfs_control_open,
2531         .unlocked_ioctl  = btrfs_control_ioctl,
2532         .compat_ioctl = compat_ptr_ioctl,
2533         .owner   = THIS_MODULE,
2534         .llseek = noop_llseek,
2535 };
2536
2537 static struct miscdevice btrfs_misc = {
2538         .minor          = BTRFS_MINOR,
2539         .name           = "btrfs-control",
2540         .fops           = &btrfs_ctl_fops
2541 };
2542
2543 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2544 MODULE_ALIAS("devname:btrfs-control");
2545
2546 static int __init btrfs_interface_init(void)
2547 {
2548         return misc_register(&btrfs_misc);
2549 }
2550
2551 static __cold void btrfs_interface_exit(void)
2552 {
2553         misc_deregister(&btrfs_misc);
2554 }
2555
2556 static void __init btrfs_print_mod_info(void)
2557 {
2558         static const char options[] = ""
2559 #ifdef CONFIG_BTRFS_DEBUG
2560                         ", debug=on"
2561 #endif
2562 #ifdef CONFIG_BTRFS_ASSERT
2563                         ", assert=on"
2564 #endif
2565 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2566                         ", integrity-checker=on"
2567 #endif
2568 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
2569                         ", ref-verify=on"
2570 #endif
2571 #ifdef CONFIG_BLK_DEV_ZONED
2572                         ", zoned=yes"
2573 #else
2574                         ", zoned=no"
2575 #endif
2576                         ;
2577         pr_info("Btrfs loaded, crc32c=%s%s\n", crc32c_impl(), options);
2578 }
2579
2580 static int __init init_btrfs_fs(void)
2581 {
2582         int err;
2583
2584         btrfs_props_init();
2585
2586         err = btrfs_init_sysfs();
2587         if (err)
2588                 return err;
2589
2590         btrfs_init_compress();
2591
2592         err = btrfs_init_cachep();
2593         if (err)
2594                 goto free_compress;
2595
2596         err = extent_io_init();
2597         if (err)
2598                 goto free_cachep;
2599
2600         err = extent_state_cache_init();
2601         if (err)
2602                 goto free_extent_io;
2603
2604         err = extent_map_init();
2605         if (err)
2606                 goto free_extent_state_cache;
2607
2608         err = ordered_data_init();
2609         if (err)
2610                 goto free_extent_map;
2611
2612         err = btrfs_delayed_inode_init();
2613         if (err)
2614                 goto free_ordered_data;
2615
2616         err = btrfs_auto_defrag_init();
2617         if (err)
2618                 goto free_delayed_inode;
2619
2620         err = btrfs_delayed_ref_init();
2621         if (err)
2622                 goto free_auto_defrag;
2623
2624         err = btrfs_prelim_ref_init();
2625         if (err)
2626                 goto free_delayed_ref;
2627
2628         err = btrfs_end_io_wq_init();
2629         if (err)
2630                 goto free_prelim_ref;
2631
2632         err = btrfs_interface_init();
2633         if (err)
2634                 goto free_end_io_wq;
2635
2636         btrfs_print_mod_info();
2637
2638         err = btrfs_run_sanity_tests();
2639         if (err)
2640                 goto unregister_ioctl;
2641
2642         err = register_filesystem(&btrfs_fs_type);
2643         if (err)
2644                 goto unregister_ioctl;
2645
2646         return 0;
2647
2648 unregister_ioctl:
2649         btrfs_interface_exit();
2650 free_end_io_wq:
2651         btrfs_end_io_wq_exit();
2652 free_prelim_ref:
2653         btrfs_prelim_ref_exit();
2654 free_delayed_ref:
2655         btrfs_delayed_ref_exit();
2656 free_auto_defrag:
2657         btrfs_auto_defrag_exit();
2658 free_delayed_inode:
2659         btrfs_delayed_inode_exit();
2660 free_ordered_data:
2661         ordered_data_exit();
2662 free_extent_map:
2663         extent_map_exit();
2664 free_extent_state_cache:
2665         extent_state_cache_exit();
2666 free_extent_io:
2667         extent_io_exit();
2668 free_cachep:
2669         btrfs_destroy_cachep();
2670 free_compress:
2671         btrfs_exit_compress();
2672         btrfs_exit_sysfs();
2673
2674         return err;
2675 }
2676
2677 static void __exit exit_btrfs_fs(void)
2678 {
2679         btrfs_destroy_cachep();
2680         btrfs_delayed_ref_exit();
2681         btrfs_auto_defrag_exit();
2682         btrfs_delayed_inode_exit();
2683         btrfs_prelim_ref_exit();
2684         ordered_data_exit();
2685         extent_map_exit();
2686         extent_state_cache_exit();
2687         extent_io_exit();
2688         btrfs_interface_exit();
2689         btrfs_end_io_wq_exit();
2690         unregister_filesystem(&btrfs_fs_type);
2691         btrfs_exit_sysfs();
2692         btrfs_cleanup_fs_uuids();
2693         btrfs_exit_compress();
2694 }
2695
2696 late_initcall(init_btrfs_fs);
2697 module_exit(exit_btrfs_fs)
2698
2699 MODULE_LICENSE("GPL");
2700 MODULE_SOFTDEP("pre: crc32c");
2701 MODULE_SOFTDEP("pre: xxhash64");
2702 MODULE_SOFTDEP("pre: sha256");
2703 MODULE_SOFTDEP("pre: blake2b-256");