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