fs/btrfs/super.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/blkdev.h>
  20 #include <linux/module.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/fs.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/highmem.h>
  25 #include <linux/time.h>
  26 #include <linux/init.h>
  27 #include <linux/seq_file.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mount.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/parser.h>
  37 #include <linux/ctype.h>
  38 #include <linux/namei.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/magic.h>
  41 #include <linux/slab.h>
  42 #include <linux/cleancache.h>
  43 #include <linux/ratelimit.h>
  44 #include <linux/btrfs.h>
  45 #include "delayed-inode.h"
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "hash.h"
  52 #include "props.h"
  53 #include "xattr.h"
  54 #include "volumes.h"
  55 #include "export.h"
  56 #include "compression.h"
  57 #include "rcu-string.h"
  58 #include "dev-replace.h"
  59 #include "free-space-cache.h"
  60 #include "backref.h"
  61 #include "tests/btrfs-tests.h"
  62
  63 #include "qgroup.h"
  64 #define CREATE_TRACE_POINTS
  65 #include <trace/events/btrfs.h>
  66
  67 static const struct super_operations btrfs_super_ops;
  68 static struct file_system_type btrfs_fs_type;
  69
  70 static int btrfs_remount(struct super_block *sb, int *flags, char *data);
  71
  72 const char *btrfs_decode_error(int errno)
  73 {
  74         char *errstr = "unknown";
  75
  76         switch (errno) {
  77         case -EIO:
  78                 errstr = "IO failure";
  79                 break;
  80         case -ENOMEM:
  81                 errstr = "Out of memory";
  82                 break;
  83         case -EROFS:
  84                 errstr = "Readonly filesystem";
  85                 break;
  86         case -EEXIST:
  87                 errstr = "Object already exists";
  88                 break;
  89         case -ENOSPC:
  90                 errstr = "No space left";
  91                 break;
  92         case -ENOENT:
  93                 errstr = "No such entry";
  94                 break;
  95         }
  96
  97         return errstr;
  98 }
  99
 100 /* btrfs handle error by forcing the filesystem readonly */
 101 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 102 {
 103         struct super_block *sb = fs_info->sb;
 104
 105         if (sb->s_flags & MS_RDONLY)
 106                 return;
 107
 108         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 109                 sb->s_flags |= MS_RDONLY;
 110                 btrfs_info(fs_info, "forced readonly");
 111                 /*
 112                  * Note that a running device replace operation is not
 113                  * canceled here although there is no way to update
 114                  * the progress. It would add the risk of a deadlock,
 115                  * therefore the canceling is omitted. The only penalty
 116                  * is that some I/O remains active until the procedure
 117                  * completes. The next time when the filesystem is
 118                  * mounted writeable again, the device replace
 119                  * operation continues.
 120                  */
 121         }
 122 }
 123
 124 /*
 125  * __btrfs_handle_fs_error decodes expected errors from the caller and
 126  * invokes the approciate error response.
 127  */
 128 __cold
 129 void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function,
 130                        unsigned int line, int errno, const char *fmt, ...)
 131 {
 132         struct super_block *sb = fs_info->sb;
 133 #ifdef CONFIG_PRINTK
 134         const char *errstr;
 135 #endif
 136
 137         /*
 138          * Special case: if the error is EROFS, and we're already
 139          * under MS_RDONLY, then it is safe here.
 140          */
 141         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 142                 return;
 143
 144 #ifdef CONFIG_PRINTK
 145         errstr = btrfs_decode_error(errno);
 146         if (fmt) {
 147                 struct va_format vaf;
 148                 va_list args;
 149
 150                 va_start(args, fmt);
 151                 vaf.fmt = fmt;
 152                 vaf.va = &args;
 153
 154                 printk(KERN_CRIT
 155                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 156                         sb->s_id, function, line, errno, errstr, &vaf);
 157                 va_end(args);
 158         } else {
 159                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 160                         sb->s_id, function, line, errno, errstr);
 161         }
 162 #endif
 163
 164         /*
 165          * Today we only save the error info to memory.  Long term we'll
 166          * also send it down to the disk
 167          */
 168         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 169
 170         /* Don't go through full error handling during mount */
 171         if (sb->s_flags & MS_BORN)
 172                 btrfs_handle_error(fs_info);
 173 }
 174
 175 #ifdef CONFIG_PRINTK
 176 static const char * const logtypes[] = {
 177         "emergency",
 178         "alert",
 179         "critical",
 180         "error",
 181         "warning",
 182         "notice",
 183         "info",
 184         "debug",
 185 };
 186
 187
 188 /*
 189  * Use one ratelimit state per log level so that a flood of less important
 190  * messages doesn't cause more important ones to be dropped.
 191  */
 192 static struct ratelimit_state printk_limits[] = {
 193         RATELIMIT_STATE_INIT(printk_limits[0], DEFAULT_RATELIMIT_INTERVAL, 100),
 194         RATELIMIT_STATE_INIT(printk_limits[1], DEFAULT_RATELIMIT_INTERVAL, 100),
 195         RATELIMIT_STATE_INIT(printk_limits[2], DEFAULT_RATELIMIT_INTERVAL, 100),
 196         RATELIMIT_STATE_INIT(printk_limits[3], DEFAULT_RATELIMIT_INTERVAL, 100),
 197         RATELIMIT_STATE_INIT(printk_limits[4], DEFAULT_RATELIMIT_INTERVAL, 100),
 198         RATELIMIT_STATE_INIT(printk_limits[5], DEFAULT_RATELIMIT_INTERVAL, 100),
 199         RATELIMIT_STATE_INIT(printk_limits[6], DEFAULT_RATELIMIT_INTERVAL, 100),
 200         RATELIMIT_STATE_INIT(printk_limits[7], DEFAULT_RATELIMIT_INTERVAL, 100),
 201 };
 202
 203 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 204 {
 205         struct super_block *sb = fs_info->sb;
 206         char lvl[4];
 207         struct va_format vaf;
 208         va_list args;
 209         const char *type = logtypes[4];
 210         int kern_level;
 211         struct ratelimit_state *ratelimit;
 212
 213         va_start(args, fmt);
 214
 215         kern_level = printk_get_level(fmt);
 216         if (kern_level) {
 217                 size_t size = printk_skip_level(fmt) - fmt;
 218                 memcpy(lvl, fmt,  size);
 219                 lvl[size] = '\0';
 220                 fmt += size;
 221                 type = logtypes[kern_level - '0'];
 222                 ratelimit = &printk_limits[kern_level - '0'];
 223         } else {
 224                 *lvl = '\0';
 225                 /* Default to debug output */
 226                 ratelimit = &printk_limits[7];
 227         }
 228
 229         vaf.fmt = fmt;
 230         vaf.va = &args;
 231
 232         if (__ratelimit(ratelimit))
 233                 printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 234
 235         va_end(args);
 236 }
 237 #endif
 238
 239 /*
 240  * We only mark the transaction aborted and then set the file system read-only.
 241  * This will prevent new transactions from starting or trying to join this
 242  * one.
 243  *
 244  * This means that error recovery at the call site is limited to freeing
 245  * any local memory allocations and passing the error code up without
 246  * further cleanup. The transaction should complete as it normally would
 247  * in the call path but will return -EIO.
 248  *
 249  * We'll complete the cleanup in btrfs_end_transaction and
 250  * btrfs_commit_transaction.
 251  */
 252 __cold
 253 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 254                                struct btrfs_root *root, const char *function,
 255                                unsigned int line, int errno)
 256 {
 257         trans->aborted = errno;
 258         /* Nothing used. The other threads that have joined this
 259          * transaction may be able to continue. */
 260         if (!trans->dirty && list_empty(&trans->new_bgs)) {
 261                 const char *errstr;
 262
 263                 errstr = btrfs_decode_error(errno);
 264                 btrfs_warn(root->fs_info,
 265                            "%s:%d: Aborting unused transaction(%s).",
 266                            function, line, errstr);
 267                 return;
 268         }
 269         ACCESS_ONCE(trans->transaction->aborted) = errno;
 270         /* Wake up anybody who may be waiting on this transaction */
 271         wake_up(&root->fs_info->transaction_wait);
 272         wake_up(&root->fs_info->transaction_blocked_wait);
 273         __btrfs_handle_fs_error(root->fs_info, function, line, errno, NULL);
 274 }
 275 /*
 276  * __btrfs_panic decodes unexpected, fatal errors from the caller,
 277  * issues an alert, and either panics or BUGs, depending on mount options.
 278  */
 279 __cold
 280 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 281                    unsigned int line, int errno, const char *fmt, ...)
 282 {
 283         char *s_id = "<unknown>";
 284         const char *errstr;
 285         struct va_format vaf = { .fmt = fmt };
 286         va_list args;
 287
 288         if (fs_info)
 289                 s_id = fs_info->sb->s_id;
 290
 291         va_start(args, fmt);
 292         vaf.va = &args;
 293
 294         errstr = btrfs_decode_error(errno);
 295         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 296                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 297                         s_id, function, line, &vaf, errno, errstr);
 298
 299         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 300                    function, line, &vaf, errno, errstr);
 301         va_end(args);
 302         /* Caller calls BUG() */
 303 }
 304
 305 static void btrfs_put_super(struct super_block *sb)
 306 {
 307         close_ctree(btrfs_sb(sb)->tree_root);
 308 }
 309
 310 enum {
 311         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 312         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 313         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 314         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 315         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 316         Opt_space_cache, Opt_space_cache_version, Opt_clear_cache,
 317         Opt_user_subvol_rm_allowed, Opt_enospc_debug, Opt_subvolrootid,
 318         Opt_defrag, Opt_inode_cache, Opt_no_space_cache, Opt_recovery,
 319         Opt_skip_balance, Opt_check_integrity,
 320         Opt_check_integrity_including_extent_data,
 321         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 322         Opt_commit_interval, Opt_barrier, Opt_nodefrag, Opt_nodiscard,
 323         Opt_noenospc_debug, Opt_noflushoncommit, Opt_acl, Opt_datacow,
 324         Opt_datasum, Opt_treelog, Opt_noinode_cache, Opt_usebackuproot,
 325         Opt_nologreplay, Opt_norecovery,
 326 #ifdef CONFIG_BTRFS_DEBUG
 327         Opt_fragment_data, Opt_fragment_metadata, Opt_fragment_all,
 328 #endif
 329         Opt_err,
 330 };
 331
 332 static const match_table_t tokens = {
 333         {Opt_degraded, "degraded"},
 334         {Opt_subvol, "subvol=%s"},
 335         {Opt_subvolid, "subvolid=%s"},
 336         {Opt_device, "device=%s"},
 337         {Opt_nodatasum, "nodatasum"},
 338         {Opt_datasum, "datasum"},
 339         {Opt_nodatacow, "nodatacow"},
 340         {Opt_datacow, "datacow"},
 341         {Opt_nobarrier, "nobarrier"},
 342         {Opt_barrier, "barrier"},
 343         {Opt_max_inline, "max_inline=%s"},
 344         {Opt_alloc_start, "alloc_start=%s"},
 345         {Opt_thread_pool, "thread_pool=%d"},
 346         {Opt_compress, "compress"},
 347         {Opt_compress_type, "compress=%s"},
 348         {Opt_compress_force, "compress-force"},
 349         {Opt_compress_force_type, "compress-force=%s"},
 350         {Opt_ssd, "ssd"},
 351         {Opt_ssd_spread, "ssd_spread"},
 352         {Opt_nossd, "nossd"},
 353         {Opt_acl, "acl"},
 354         {Opt_noacl, "noacl"},
 355         {Opt_notreelog, "notreelog"},
 356         {Opt_treelog, "treelog"},
 357         {Opt_nologreplay, "nologreplay"},
 358         {Opt_norecovery, "norecovery"},
 359         {Opt_flushoncommit, "flushoncommit"},
 360         {Opt_noflushoncommit, "noflushoncommit"},
 361         {Opt_ratio, "metadata_ratio=%d"},
 362         {Opt_discard, "discard"},
 363         {Opt_nodiscard, "nodiscard"},
 364         {Opt_space_cache, "space_cache"},
 365         {Opt_space_cache_version, "space_cache=%s"},
 366         {Opt_clear_cache, "clear_cache"},
 367         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 368         {Opt_enospc_debug, "enospc_debug"},
 369         {Opt_noenospc_debug, "noenospc_debug"},
 370         {Opt_subvolrootid, "subvolrootid=%d"},
 371         {Opt_defrag, "autodefrag"},
 372         {Opt_nodefrag, "noautodefrag"},
 373         {Opt_inode_cache, "inode_cache"},
 374         {Opt_noinode_cache, "noinode_cache"},
 375         {Opt_no_space_cache, "nospace_cache"},
 376         {Opt_recovery, "recovery"}, /* deprecated */
 377         {Opt_usebackuproot, "usebackuproot"},
 378         {Opt_skip_balance, "skip_balance"},
 379         {Opt_check_integrity, "check_int"},
 380         {Opt_check_integrity_including_extent_data, "check_int_data"},
 381         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 382         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 383         {Opt_fatal_errors, "fatal_errors=%s"},
 384         {Opt_commit_interval, "commit=%d"},
 385 #ifdef CONFIG_BTRFS_DEBUG
 386         {Opt_fragment_data, "fragment=data"},
 387         {Opt_fragment_metadata, "fragment=metadata"},
 388         {Opt_fragment_all, "fragment=all"},
 389 #endif
 390         {Opt_err, NULL},
 391 };
 392
 393 /*
 394  * Regular mount options parser.  Everything that is needed only when
 395  * reading in a new superblock is parsed here.
 396  * XXX JDM: This needs to be cleaned up for remount.
 397  */
 398 int btrfs_parse_options(struct btrfs_root *root, char *options,
 399                         unsigned long new_flags)
 400 {
 401         struct btrfs_fs_info *info = root->fs_info;
 402         substring_t args[MAX_OPT_ARGS];
 403         char *p, *num, *orig = NULL;
 404         u64 cache_gen;
 405         int intarg;
 406         int ret = 0;
 407         char *compress_type;
 408         bool compress_force = false;
 409         enum btrfs_compression_type saved_compress_type;
 410         bool saved_compress_force;
 411         int no_compress = 0;
 412
 413         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 414         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE))
 415                 btrfs_set_opt(info->mount_opt, FREE_SPACE_TREE);
 416         else if (cache_gen)
 417                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 418
 419         /*
 420          * Even the options are empty, we still need to do extra check
 421          * against new flags
 422          */
 423         if (!options)
 424                 goto check;
 425
 426         /*
 427          * strsep changes the string, duplicate it because parse_options
 428          * gets called twice
 429          */
 430         options = kstrdup(options, GFP_NOFS);
 431         if (!options)
 432                 return -ENOMEM;
 433
 434         orig = options;
 435
 436         while ((p = strsep(&options, ",")) != NULL) {
 437                 int token;
 438                 if (!*p)
 439                         continue;
 440
 441                 token = match_token(p, tokens, args);
 442                 switch (token) {
 443                 case Opt_degraded:
 444                         btrfs_info(root->fs_info, "allowing degraded mounts");
 445                         btrfs_set_opt(info->mount_opt, DEGRADED);
 446                         break;
 447                 case Opt_subvol:
 448                 case Opt_subvolid:
 449                 case Opt_subvolrootid:
 450                 case Opt_device:
 451                         /*
 452                          * These are parsed by btrfs_parse_early_options
 453                          * and can be happily ignored here.
 454                          */
 455                         break;
 456                 case Opt_nodatasum:
 457                         btrfs_set_and_info(root, NODATASUM,
 458                                            "setting nodatasum");
 459                         break;
 460                 case Opt_datasum:
 461                         if (btrfs_test_opt(root, NODATASUM)) {
 462                                 if (btrfs_test_opt(root, NODATACOW))
 463                                         btrfs_info(root->fs_info, "setting datasum, datacow enabled");
 464                                 else
 465                                         btrfs_info(root->fs_info, "setting datasum");
 466                         }
 467                         btrfs_clear_opt(info->mount_opt, NODATACOW);
 468                         btrfs_clear_opt(info->mount_opt, NODATASUM);
 469                         break;
 470                 case Opt_nodatacow:
 471                         if (!btrfs_test_opt(root, NODATACOW)) {
 472                                 if (!btrfs_test_opt(root, COMPRESS) ||
 473                                     !btrfs_test_opt(root, FORCE_COMPRESS)) {
 474                                         btrfs_info(root->fs_info,
 475                                                    "setting nodatacow, compression disabled");
 476                                 } else {
 477                                         btrfs_info(root->fs_info, "setting nodatacow");
 478                                 }
 479                         }
 480                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 481                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 482                         btrfs_set_opt(info->mount_opt, NODATACOW);
 483                         btrfs_set_opt(info->mount_opt, NODATASUM);
 484                         break;
 485                 case Opt_datacow:
 486                         btrfs_clear_and_info(root, NODATACOW,
 487                                              "setting datacow");
 488                         break;
 489                 case Opt_compress_force:
 490                 case Opt_compress_force_type:
 491                         compress_force = true;
 492                         /* Fallthrough */
 493                 case Opt_compress:
 494                 case Opt_compress_type:
 495                         saved_compress_type = btrfs_test_opt(root, COMPRESS) ?
 496                                 info->compress_type : BTRFS_COMPRESS_NONE;
 497                         saved_compress_force =
 498                                 btrfs_test_opt(root, FORCE_COMPRESS);
 499                         if (token == Opt_compress ||
 500                             token == Opt_compress_force ||
 501                             strcmp(args[0].from, "zlib") == 0) {
 502                                 compress_type = "zlib";
 503                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 504                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 505                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 506                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 507                                 no_compress = 0;
 508                         } else if (strcmp(args[0].from, "lzo") == 0) {
 509                                 compress_type = "lzo";
 510                                 info->compress_type = BTRFS_COMPRESS_LZO;
 511                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 512                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 513                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 514                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 515                                 no_compress = 0;
 516                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 517                                 compress_type = "no";
 518                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 519                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 520                                 compress_force = false;
 521                                 no_compress++;
 522                         } else {
 523                                 ret = -EINVAL;
 524                                 goto out;
 525                         }
 526
 527                         if (compress_force) {
 528                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 529                         } else {
 530                                 /*
 531                                  * If we remount from compress-force=xxx to
 532                                  * compress=xxx, we need clear FORCE_COMPRESS
 533                                  * flag, otherwise, there is no way for users
 534                                  * to disable forcible compression separately.
 535                                  */
 536                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 537                         }
 538                         if ((btrfs_test_opt(root, COMPRESS) &&
 539                              (info->compress_type != saved_compress_type ||
 540                               compress_force != saved_compress_force)) ||
 541                             (!btrfs_test_opt(root, COMPRESS) &&
 542                              no_compress == 1)) {
 543                                 btrfs_info(root->fs_info,
 544                                            "%s %s compression",
 545                                            (compress_force) ? "force" : "use",
 546                                            compress_type);
 547                         }
 548                         compress_force = false;
 549                         break;
 550                 case Opt_ssd:
 551                         btrfs_set_and_info(root, SSD,
 552                                            "use ssd allocation scheme");
 553                         break;
 554                 case Opt_ssd_spread:
 555                         btrfs_set_and_info(root, SSD_SPREAD,
 556                                            "use spread ssd allocation scheme");
 557                         btrfs_set_opt(info->mount_opt, SSD);
 558                         break;
 559                 case Opt_nossd:
 560                         btrfs_set_and_info(root, NOSSD,
 561                                              "not using ssd allocation scheme");
 562                         btrfs_clear_opt(info->mount_opt, SSD);
 563                         break;
 564                 case Opt_barrier:
 565                         btrfs_clear_and_info(root, NOBARRIER,
 566                                              "turning on barriers");
 567                         break;
 568                 case Opt_nobarrier:
 569                         btrfs_set_and_info(root, NOBARRIER,
 570                                            "turning off barriers");
 571                         break;
 572                 case Opt_thread_pool:
 573                         ret = match_int(&args[0], &intarg);
 574                         if (ret) {
 575                                 goto out;
 576                         } else if (intarg > 0) {
 577                                 info->thread_pool_size = intarg;
 578                         } else {
 579                                 ret = -EINVAL;
 580                                 goto out;
 581                         }
 582                         break;
 583                 case Opt_max_inline:
 584                         num = match_strdup(&args[0]);
 585                         if (num) {
 586                                 info->max_inline = memparse(num, NULL);
 587                                 kfree(num);
 588
 589                                 if (info->max_inline) {
 590                                         info->max_inline = min_t(u64,
 591                                                 info->max_inline,
 592                                                 root->sectorsize);
 593                                 }
 594                                 btrfs_info(root->fs_info, "max_inline at %llu",
 595                                         info->max_inline);
 596                         } else {
 597                                 ret = -ENOMEM;
 598                                 goto out;
 599                         }
 600                         break;
 601                 case Opt_alloc_start:
 602                         num = match_strdup(&args[0]);
 603                         if (num) {
 604                                 mutex_lock(&info->chunk_mutex);
 605                                 info->alloc_start = memparse(num, NULL);
 606                                 mutex_unlock(&info->chunk_mutex);
 607                                 kfree(num);
 608                                 btrfs_info(root->fs_info, "allocations start at %llu",
 609                                         info->alloc_start);
 610                         } else {
 611                                 ret = -ENOMEM;
 612                                 goto out;
 613                         }
 614                         break;
 615                 case Opt_acl:
 616 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 617                         root->fs_info->sb->s_flags |= MS_POSIXACL;
 618                         break;
 619 #else
 620                         btrfs_err(root->fs_info,
 621                                 "support for ACL not compiled in!");
 622                         ret = -EINVAL;
 623                         goto out;
 624 #endif
 625                 case Opt_noacl:
 626                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 627                         break;
 628                 case Opt_notreelog:
 629                         btrfs_set_and_info(root, NOTREELOG,
 630                                            "disabling tree log");
 631                         break;
 632                 case Opt_treelog:
 633                         btrfs_clear_and_info(root, NOTREELOG,
 634                                              "enabling tree log");
 635                         break;
 636                 case Opt_norecovery:
 637                 case Opt_nologreplay:
 638                         btrfs_set_and_info(root, NOLOGREPLAY,
 639                                            "disabling log replay at mount time");
 640                         break;
 641                 case Opt_flushoncommit:
 642                         btrfs_set_and_info(root, FLUSHONCOMMIT,
 643                                            "turning on flush-on-commit");
 644                         break;
 645                 case Opt_noflushoncommit:
 646                         btrfs_clear_and_info(root, FLUSHONCOMMIT,
 647                                              "turning off flush-on-commit");
 648                         break;
 649                 case Opt_ratio:
 650                         ret = match_int(&args[0], &intarg);
 651                         if (ret) {
 652                                 goto out;
 653                         } else if (intarg >= 0) {
 654                                 info->metadata_ratio = intarg;
 655                                 btrfs_info(root->fs_info, "metadata ratio %d",
 656                                        info->metadata_ratio);
 657                         } else {
 658                                 ret = -EINVAL;
 659                                 goto out;
 660                         }
 661                         break;
 662                 case Opt_discard:
 663                         btrfs_set_and_info(root, DISCARD,
 664                                            "turning on discard");
 665                         break;
 666                 case Opt_nodiscard:
 667                         btrfs_clear_and_info(root, DISCARD,
 668                                              "turning off discard");
 669                         break;
 670                 case Opt_space_cache:
 671                 case Opt_space_cache_version:
 672                         if (token == Opt_space_cache ||
 673                             strcmp(args[0].from, "v1") == 0) {
 674                                 btrfs_clear_opt(root->fs_info->mount_opt,
 675                                                 FREE_SPACE_TREE);
 676                                 btrfs_set_and_info(root, SPACE_CACHE,
 677                                                    "enabling disk space caching");
 678                         } else if (strcmp(args[0].from, "v2") == 0) {
 679                                 btrfs_clear_opt(root->fs_info->mount_opt,
 680                                                 SPACE_CACHE);
 681                                 btrfs_set_and_info(root, FREE_SPACE_TREE,
 682                                                    "enabling free space tree");
 683                         } else {
 684                                 ret = -EINVAL;
 685                                 goto out;
 686                         }
 687                         break;
 688                 case Opt_rescan_uuid_tree:
 689                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 690                         break;
 691                 case Opt_no_space_cache:
 692                         if (btrfs_test_opt(root, SPACE_CACHE)) {
 693                                 btrfs_clear_and_info(root, SPACE_CACHE,
 694                                                      "disabling disk space caching");
 695                         }
 696                         if (btrfs_test_opt(root, FREE_SPACE_TREE)) {
 697                                 btrfs_clear_and_info(root, FREE_SPACE_TREE,
 698                                                      "disabling free space tree");
 699                         }
 700                         break;
 701                 case Opt_inode_cache:
 702                         btrfs_set_pending_and_info(info, INODE_MAP_CACHE,
 703                                            "enabling inode map caching");
 704                         break;
 705                 case Opt_noinode_cache:
 706                         btrfs_clear_pending_and_info(info, INODE_MAP_CACHE,
 707                                              "disabling inode map caching");
 708                         break;
 709                 case Opt_clear_cache:
 710                         btrfs_set_and_info(root, CLEAR_CACHE,
 711                                            "force clearing of disk cache");
 712                         break;
 713                 case Opt_user_subvol_rm_allowed:
 714                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 715                         break;
 716                 case Opt_enospc_debug:
 717                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 718                         break;
 719                 case Opt_noenospc_debug:
 720                         btrfs_clear_opt(info->mount_opt, ENOSPC_DEBUG);
 721                         break;
 722                 case Opt_defrag:
 723                         btrfs_set_and_info(root, AUTO_DEFRAG,
 724                                            "enabling auto defrag");
 725                         break;
 726                 case Opt_nodefrag:
 727                         btrfs_clear_and_info(root, AUTO_DEFRAG,
 728                                              "disabling auto defrag");
 729                         break;
 730                 case Opt_recovery:
 731                         btrfs_warn(root->fs_info,
 732                                    "'recovery' is deprecated, use 'usebackuproot' instead");
 733                 case Opt_usebackuproot:
 734                         btrfs_info(root->fs_info,
 735                                    "trying to use backup root at mount time");
 736                         btrfs_set_opt(info->mount_opt, USEBACKUPROOT);
 737                         break;
 738                 case Opt_skip_balance:
 739                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 740                         break;
 741 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 742                 case Opt_check_integrity_including_extent_data:
 743                         btrfs_info(root->fs_info,
 744                                    "enabling check integrity including extent data");
 745                         btrfs_set_opt(info->mount_opt,
 746                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 747                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 748                         break;
 749                 case Opt_check_integrity:
 750                         btrfs_info(root->fs_info, "enabling check integrity");
 751                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 752                         break;
 753                 case Opt_check_integrity_print_mask:
 754                         ret = match_int(&args[0], &intarg);
 755                         if (ret) {
 756                                 goto out;
 757                         } else if (intarg >= 0) {
 758                                 info->check_integrity_print_mask = intarg;
 759                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 760                                        info->check_integrity_print_mask);
 761                         } else {
 762                                 ret = -EINVAL;
 763                                 goto out;
 764                         }
 765                         break;
 766 #else
 767                 case Opt_check_integrity_including_extent_data:
 768                 case Opt_check_integrity:
 769                 case Opt_check_integrity_print_mask:
 770                         btrfs_err(root->fs_info,
 771                                 "support for check_integrity* not compiled in!");
 772                         ret = -EINVAL;
 773                         goto out;
 774 #endif
 775                 case Opt_fatal_errors:
 776                         if (strcmp(args[0].from, "panic") == 0)
 777                                 btrfs_set_opt(info->mount_opt,
 778                                               PANIC_ON_FATAL_ERROR);
 779                         else if (strcmp(args[0].from, "bug") == 0)
 780                                 btrfs_clear_opt(info->mount_opt,
 781                                               PANIC_ON_FATAL_ERROR);
 782                         else {
 783                                 ret = -EINVAL;
 784                                 goto out;
 785                         }
 786                         break;
 787                 case Opt_commit_interval:
 788                         intarg = 0;
 789                         ret = match_int(&args[0], &intarg);
 790                         if (ret < 0) {
 791                                 btrfs_err(root->fs_info, "invalid commit interval");
 792                                 ret = -EINVAL;
 793                                 goto out;
 794                         }
 795                         if (intarg > 0) {
 796                                 if (intarg > 300) {
 797                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 798                                                         intarg);
 799                                 }
 800                                 info->commit_interval = intarg;
 801                         } else {
 802                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 803                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 804                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 805                         }
 806                         break;
 807 #ifdef CONFIG_BTRFS_DEBUG
 808                 case Opt_fragment_all:
 809                         btrfs_info(root->fs_info, "fragmenting all space");
 810                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 811                         btrfs_set_opt(info->mount_opt, FRAGMENT_METADATA);
 812                         break;
 813                 case Opt_fragment_metadata:
 814                         btrfs_info(root->fs_info, "fragmenting metadata");
 815                         btrfs_set_opt(info->mount_opt,
 816                                       FRAGMENT_METADATA);
 817                         break;
 818                 case Opt_fragment_data:
 819                         btrfs_info(root->fs_info, "fragmenting data");
 820                         btrfs_set_opt(info->mount_opt, FRAGMENT_DATA);
 821                         break;
 822 #endif
 823                 case Opt_err:
 824                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 825                         ret = -EINVAL;
 826                         goto out;
 827                 default:
 828                         break;
 829                 }
 830         }
 831 check:
 832         /*
 833          * Extra check for current option against current flag
 834          */
 835         if (btrfs_test_opt(root, NOLOGREPLAY) && !(new_flags & MS_RDONLY)) {
 836                 btrfs_err(root->fs_info,
 837                           "nologreplay must be used with ro mount option");
 838                 ret = -EINVAL;
 839         }
 840 out:
 841         if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE) &&
 842             !btrfs_test_opt(root, FREE_SPACE_TREE) &&
 843             !btrfs_test_opt(root, CLEAR_CACHE)) {
 844                 btrfs_err(root->fs_info, "cannot disable free space tree");
 845                 ret = -EINVAL;
 846
 847         }
 848         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 849                 btrfs_info(root->fs_info, "disk space caching is enabled");
 850         if (!ret && btrfs_test_opt(root, FREE_SPACE_TREE))
 851                 btrfs_info(root->fs_info, "using free space tree");
 852         kfree(orig);
 853         return ret;
 854 }
 855
 856 /*
 857  * Parse mount options that are required early in the mount process.
 858  *
 859  * All other options will be parsed on much later in the mount process and
 860  * only when we need to allocate a new super block.
 861  */
 862 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 863                 void *holder, char **subvol_name, u64 *subvol_objectid,
 864                 struct btrfs_fs_devices **fs_devices)
 865 {
 866         substring_t args[MAX_OPT_ARGS];
 867         char *device_name, *opts, *orig, *p;
 868         char *num = NULL;
 869         int error = 0;
 870
 871         if (!options)
 872                 return 0;
 873
 874         /*
 875          * strsep changes the string, duplicate it because parse_options
 876          * gets called twice
 877          */
 878         opts = kstrdup(options, GFP_KERNEL);
 879         if (!opts)
 880                 return -ENOMEM;
 881         orig = opts;
 882
 883         while ((p = strsep(&opts, ",")) != NULL) {
 884                 int token;
 885                 if (!*p)
 886                         continue;
 887
 888                 token = match_token(p, tokens, args);
 889                 switch (token) {
 890                 case Opt_subvol:
 891                         kfree(*subvol_name);
 892                         *subvol_name = match_strdup(&args[0]);
 893                         if (!*subvol_name) {
 894                                 error = -ENOMEM;
 895                                 goto out;
 896                         }
 897                         break;
 898                 case Opt_subvolid:
 899                         num = match_strdup(&args[0]);
 900                         if (num) {
 901                                 *subvol_objectid = memparse(num, NULL);
 902                                 kfree(num);
 903                                 /* we want the original fs_tree */
 904                                 if (!*subvol_objectid)
 905                                         *subvol_objectid =
 906                                                 BTRFS_FS_TREE_OBJECTID;
 907                         } else {
 908                                 error = -EINVAL;
 909                                 goto out;
 910                         }
 911                         break;
 912                 case Opt_subvolrootid:
 913                         printk(KERN_WARNING
 914                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 915                                 "no effect\n");
 916                         break;
 917                 case Opt_device:
 918                         device_name = match_strdup(&args[0]);
 919                         if (!device_name) {
 920                                 error = -ENOMEM;
 921                                 goto out;
 922                         }
 923                         error = btrfs_scan_one_device(device_name,
 924                                         flags, holder, fs_devices);
 925                         kfree(device_name);
 926                         if (error)
 927                                 goto out;
 928                         break;
 929                 default:
 930                         break;
 931                 }
 932         }
 933
 934 out:
 935         kfree(orig);
 936         return error;
 937 }
 938
 939 static char *get_subvol_name_from_objectid(struct btrfs_fs_info *fs_info,
 940                                            u64 subvol_objectid)
 941 {
 942         struct btrfs_root *root = fs_info->tree_root;
 943         struct btrfs_root *fs_root;
 944         struct btrfs_root_ref *root_ref;
 945         struct btrfs_inode_ref *inode_ref;
 946         struct btrfs_key key;
 947         struct btrfs_path *path = NULL;
 948         char *name = NULL, *ptr;
 949         u64 dirid;
 950         int len;
 951         int ret;
 952
 953         path = btrfs_alloc_path();
 954         if (!path) {
 955                 ret = -ENOMEM;
 956                 goto err;
 957         }
 958         path->leave_spinning = 1;
 959
 960         name = kmalloc(PATH_MAX, GFP_NOFS);
 961         if (!name) {
 962                 ret = -ENOMEM;
 963                 goto err;
 964         }
 965         ptr = name + PATH_MAX - 1;
 966         ptr[0] = '\0';
 967
 968         /*
 969          * Walk up the subvolume trees in the tree of tree roots by root
 970          * backrefs until we hit the top-level subvolume.
 971          */
 972         while (subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
 973                 key.objectid = subvol_objectid;
 974                 key.type = BTRFS_ROOT_BACKREF_KEY;
 975                 key.offset = (u64)-1;
 976
 977                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 978                 if (ret < 0) {
 979                         goto err;
 980                 } else if (ret > 0) {
 981                         ret = btrfs_previous_item(root, path, subvol_objectid,
 982                                                   BTRFS_ROOT_BACKREF_KEY);
 983                         if (ret < 0) {
 984                                 goto err;
 985                         } else if (ret > 0) {
 986                                 ret = -ENOENT;
 987                                 goto err;
 988                         }
 989                 }
 990
 991                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 992                 subvol_objectid = key.offset;
 993
 994                 root_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
 995                                           struct btrfs_root_ref);
 996                 len = btrfs_root_ref_name_len(path->nodes[0], root_ref);
 997                 ptr -= len + 1;
 998                 if (ptr < name) {
 999                         ret = -ENAMETOOLONG;
1000                         goto err;
1001                 }
1002                 read_extent_buffer(path->nodes[0], ptr + 1,
1003                                    (unsigned long)(root_ref + 1), len);
1004                 ptr[0] = '/';
1005                 dirid = btrfs_root_ref_dirid(path->nodes[0], root_ref);
1006                 btrfs_release_path(path);
1007
1008                 key.objectid = subvol_objectid;
1009                 key.type = BTRFS_ROOT_ITEM_KEY;
1010                 key.offset = (u64)-1;
1011                 fs_root = btrfs_read_fs_root_no_name(fs_info, &key);
1012                 if (IS_ERR(fs_root)) {
1013                         ret = PTR_ERR(fs_root);
1014                         goto err;
1015                 }
1016
1017                 /*
1018                  * Walk up the filesystem tree by inode refs until we hit the
1019                  * root directory.
1020                  */
1021                 while (dirid != BTRFS_FIRST_FREE_OBJECTID) {
1022                         key.objectid = dirid;
1023                         key.type = BTRFS_INODE_REF_KEY;
1024                         key.offset = (u64)-1;
1025
1026                         ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1027                         if (ret < 0) {
1028                                 goto err;
1029                         } else if (ret > 0) {
1030                                 ret = btrfs_previous_item(fs_root, path, dirid,
1031                                                           BTRFS_INODE_REF_KEY);
1032                                 if (ret < 0) {
1033                                         goto err;
1034                                 } else if (ret > 0) {
1035                                         ret = -ENOENT;
1036                                         goto err;
1037                                 }
1038                         }
1039
1040                         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1041                         dirid = key.offset;
1042
1043                         inode_ref = btrfs_item_ptr(path->nodes[0],
1044                                                    path->slots[0],
1045                                                    struct btrfs_inode_ref);
1046                         len = btrfs_inode_ref_name_len(path->nodes[0],
1047                                                        inode_ref);
1048                         ptr -= len + 1;
1049                         if (ptr < name) {
1050                                 ret = -ENAMETOOLONG;
1051                                 goto err;
1052                         }
1053                         read_extent_buffer(path->nodes[0], ptr + 1,
1054                                            (unsigned long)(inode_ref + 1), len);
1055                         ptr[0] = '/';
1056                         btrfs_release_path(path);
1057                 }
1058         }
1059
1060         btrfs_free_path(path);
1061         if (ptr == name + PATH_MAX - 1) {
1062                 name[0] = '/';
1063                 name[1] = '\0';
1064         } else {
1065                 memmove(name, ptr, name + PATH_MAX - ptr);
1066         }
1067         return name;
1068
1069 err:
1070         btrfs_free_path(path);
1071         kfree(name);
1072         return ERR_PTR(ret);
1073 }
1074
1075 static int get_default_subvol_objectid(struct btrfs_fs_info *fs_info, u64 *objectid)
1076 {
1077         struct btrfs_root *root = fs_info->tree_root;
1078         struct btrfs_dir_item *di;
1079         struct btrfs_path *path;
1080         struct btrfs_key location;
1081         u64 dir_id;
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086         path->leave_spinning = 1;
1087
1088         /*
1089          * Find the "default" dir item which points to the root item that we
1090          * will mount by default if we haven't been given a specific subvolume
1091          * to mount.
1092          */
1093         dir_id = btrfs_super_root_dir(fs_info->super_copy);
1094         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
1095         if (IS_ERR(di)) {
1096                 btrfs_free_path(path);
1097                 return PTR_ERR(di);
1098         }
1099         if (!di) {
1100                 /*
1101                  * Ok the default dir item isn't there.  This is weird since
1102                  * it's always been there, but don't freak out, just try and
1103                  * mount the top-level subvolume.
1104                  */
1105                 btrfs_free_path(path);
1106                 *objectid = BTRFS_FS_TREE_OBJECTID;
1107                 return 0;
1108         }
1109
1110         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1111         btrfs_free_path(path);
1112         *objectid = location.objectid;
1113         return 0;
1114 }
1115
1116 static int btrfs_fill_super(struct super_block *sb,
1117                             struct btrfs_fs_devices *fs_devices,
1118                             void *data, int silent)
1119 {
1120         struct inode *inode;
1121         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1122         struct btrfs_key key;
1123         int err;
1124
1125         sb->s_maxbytes = MAX_LFS_FILESIZE;
1126         sb->s_magic = BTRFS_SUPER_MAGIC;
1127         sb->s_op = &btrfs_super_ops;
1128         sb->s_d_op = &btrfs_dentry_operations;
1129         sb->s_export_op = &btrfs_export_ops;
1130         sb->s_xattr = btrfs_xattr_handlers;
1131         sb->s_time_gran = 1;
1132 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
1133         sb->s_flags |= MS_POSIXACL;
1134 #endif
1135         sb->s_flags |= MS_I_VERSION;
1136         sb->s_iflags |= SB_I_CGROUPWB;
1137         err = open_ctree(sb, fs_devices, (char *)data);
1138         if (err) {
1139                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
1140                 return err;
1141         }
1142
1143         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
1144         key.type = BTRFS_INODE_ITEM_KEY;
1145         key.offset = 0;
1146         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
1147         if (IS_ERR(inode)) {
1148                 err = PTR_ERR(inode);
1149                 goto fail_close;
1150         }
1151
1152         sb->s_root = d_make_root(inode);
1153         if (!sb->s_root) {
1154                 err = -ENOMEM;
1155                 goto fail_close;
1156         }
1157
1158         save_mount_options(sb, data);
1159         cleancache_init_fs(sb);
1160         sb->s_flags |= MS_ACTIVE;
1161         return 0;
1162
1163 fail_close:
1164         close_ctree(fs_info->tree_root);
1165         return err;
1166 }
1167
1168 int btrfs_sync_fs(struct super_block *sb, int wait)
1169 {
1170         struct btrfs_trans_handle *trans;
1171         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1172         struct btrfs_root *root = fs_info->tree_root;
1173
1174         trace_btrfs_sync_fs(fs_info, wait);
1175
1176         if (!wait) {
1177                 filemap_flush(fs_info->btree_inode->i_mapping);
1178                 return 0;
1179         }
1180
1181         btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
1182
1183         trans = btrfs_attach_transaction_barrier(root);
1184         if (IS_ERR(trans)) {
1185                 /* no transaction, don't bother */
1186                 if (PTR_ERR(trans) == -ENOENT) {
1187                         /*
1188                          * Exit unless we have some pending changes
1189                          * that need to go through commit
1190                          */
1191                         if (fs_info->pending_changes == 0)
1192                                 return 0;
1193                         /*
1194                          * A non-blocking test if the fs is frozen. We must not
1195                          * start a new transaction here otherwise a deadlock
1196                          * happens. The pending operations are delayed to the
1197                          * next commit after thawing.
1198                          */
1199                         if (__sb_start_write(sb, SB_FREEZE_WRITE, false))
1200                                 __sb_end_write(sb, SB_FREEZE_WRITE);
1201                         else
1202                                 return 0;
1203                         trans = btrfs_start_transaction(root, 0);
1204                 }
1205                 if (IS_ERR(trans))
1206                         return PTR_ERR(trans);
1207         }
1208         return btrfs_commit_transaction(trans, root);
1209 }
1210
1211 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
1212 {
1213         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
1214         struct btrfs_root *root = info->tree_root;
1215         char *compress_type;
1216
1217         if (btrfs_test_opt(root, DEGRADED))
1218                 seq_puts(seq, ",degraded");
1219         if (btrfs_test_opt(root, NODATASUM))
1220                 seq_puts(seq, ",nodatasum");
1221         if (btrfs_test_opt(root, NODATACOW))
1222                 seq_puts(seq, ",nodatacow");
1223         if (btrfs_test_opt(root, NOBARRIER))
1224                 seq_puts(seq, ",nobarrier");
1225         if (info->max_inline != BTRFS_DEFAULT_MAX_INLINE)
1226                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
1227         if (info->alloc_start != 0)
1228                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
1229         if (info->thread_pool_size !=  min_t(unsigned long,
1230                                              num_online_cpus() + 2, 8))
1231                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
1232         if (btrfs_test_opt(root, COMPRESS)) {
1233                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
1234                         compress_type = "zlib";
1235                 else
1236                         compress_type = "lzo";
1237                 if (btrfs_test_opt(root, FORCE_COMPRESS))
1238                         seq_printf(seq, ",compress-force=%s", compress_type);
1239                 else
1240                         seq_printf(seq, ",compress=%s", compress_type);
1241         }
1242         if (btrfs_test_opt(root, NOSSD))
1243                 seq_puts(seq, ",nossd");
1244         if (btrfs_test_opt(root, SSD_SPREAD))
1245                 seq_puts(seq, ",ssd_spread");
1246         else if (btrfs_test_opt(root, SSD))
1247                 seq_puts(seq, ",ssd");
1248         if (btrfs_test_opt(root, NOTREELOG))
1249                 seq_puts(seq, ",notreelog");
1250         if (btrfs_test_opt(root, NOLOGREPLAY))
1251                 seq_puts(seq, ",nologreplay");
1252         if (btrfs_test_opt(root, FLUSHONCOMMIT))
1253                 seq_puts(seq, ",flushoncommit");
1254         if (btrfs_test_opt(root, DISCARD))
1255                 seq_puts(seq, ",discard");
1256         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
1257                 seq_puts(seq, ",noacl");
1258         if (btrfs_test_opt(root, SPACE_CACHE))
1259                 seq_puts(seq, ",space_cache");
1260         else if (btrfs_test_opt(root, FREE_SPACE_TREE))
1261                 seq_puts(seq, ",space_cache=v2");
1262         else
1263                 seq_puts(seq, ",nospace_cache");
1264         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
1265                 seq_puts(seq, ",rescan_uuid_tree");
1266         if (btrfs_test_opt(root, CLEAR_CACHE))
1267                 seq_puts(seq, ",clear_cache");
1268         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
1269                 seq_puts(seq, ",user_subvol_rm_allowed");
1270         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1271                 seq_puts(seq, ",enospc_debug");
1272         if (btrfs_test_opt(root, AUTO_DEFRAG))
1273                 seq_puts(seq, ",autodefrag");
1274         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1275                 seq_puts(seq, ",inode_cache");
1276         if (btrfs_test_opt(root, SKIP_BALANCE))
1277                 seq_puts(seq, ",skip_balance");
1278 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1279         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1280                 seq_puts(seq, ",check_int_data");
1281         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1282                 seq_puts(seq, ",check_int");
1283         if (info->check_integrity_print_mask)
1284                 seq_printf(seq, ",check_int_print_mask=%d",
1285                                 info->check_integrity_print_mask);
1286 #endif
1287         if (info->metadata_ratio)
1288                 seq_printf(seq, ",metadata_ratio=%d",
1289                                 info->metadata_ratio);
1290         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1291                 seq_puts(seq, ",fatal_errors=panic");
1292         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1293                 seq_printf(seq, ",commit=%d", info->commit_interval);
1294 #ifdef CONFIG_BTRFS_DEBUG
1295         if (btrfs_test_opt(root, FRAGMENT_DATA))
1296                 seq_puts(seq, ",fragment=data");
1297         if (btrfs_test_opt(root, FRAGMENT_METADATA))
1298                 seq_puts(seq, ",fragment=metadata");
1299 #endif
1300         seq_printf(seq, ",subvolid=%llu",
1301                   BTRFS_I(d_inode(dentry))->root->root_key.objectid);
1302         seq_puts(seq, ",subvol=");
1303         seq_dentry(seq, dentry, " \t\n\\");
1304         return 0;
1305 }
1306
1307 static int btrfs_test_super(struct super_block *s, void *data)
1308 {
1309         struct btrfs_fs_info *p = data;
1310         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1311
1312         return fs_info->fs_devices == p->fs_devices;
1313 }
1314
1315 static int btrfs_set_super(struct super_block *s, void *data)
1316 {
1317         int err = set_anon_super(s, data);
1318         if (!err)
1319                 s->s_fs_info = data;
1320         return err;
1321 }
1322
1323 /*
1324  * subvolumes are identified by ino 256
1325  */
1326 static inline int is_subvolume_inode(struct inode *inode)
1327 {
1328         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1329                 return 1;
1330         return 0;
1331 }
1332
1333 /*
1334  * This will add subvolid=0 to the argument string while removing any subvol=
1335  * and subvolid= arguments to make sure we get the top-level root for path
1336  * walking to the subvol we want.
1337  */
1338 static char *setup_root_args(char *args)
1339 {
1340         char *buf, *dst, *sep;
1341
1342         if (!args)
1343                 return kstrdup("subvolid=0", GFP_NOFS);
1344
1345         /* The worst case is that we add ",subvolid=0" to the end. */
1346         buf = dst = kmalloc(strlen(args) + strlen(",subvolid=0") + 1, GFP_NOFS);
1347         if (!buf)
1348                 return NULL;
1349
1350         while (1) {
1351                 sep = strchrnul(args, ',');
1352                 if (!strstarts(args, "subvol=") &&
1353                     !strstarts(args, "subvolid=")) {
1354                         memcpy(dst, args, sep - args);
1355                         dst += sep - args;
1356                         *dst++ = ',';
1357                 }
1358                 if (*sep)
1359                         args = sep + 1;
1360                 else
1361                         break;
1362         }
1363         strcpy(dst, "subvolid=0");
1364
1365         return buf;
1366 }
1367
1368 static struct dentry *mount_subvol(const char *subvol_name, u64 subvol_objectid,
1369                                    int flags, const char *device_name,
1370                                    char *data)
1371 {
1372         struct dentry *root;
1373         struct vfsmount *mnt = NULL;
1374         char *newargs;
1375         int ret;
1376
1377         newargs = setup_root_args(data);
1378         if (!newargs) {
1379                 root = ERR_PTR(-ENOMEM);
1380                 goto out;
1381         }
1382
1383         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name, newargs);
1384         if (PTR_ERR_OR_ZERO(mnt) == -EBUSY) {
1385                 if (flags & MS_RDONLY) {
1386                         mnt = vfs_kern_mount(&btrfs_fs_type, flags & ~MS_RDONLY,
1387                                              device_name, newargs);
1388                 } else {
1389                         mnt = vfs_kern_mount(&btrfs_fs_type, flags | MS_RDONLY,
1390                                              device_name, newargs);
1391                         if (IS_ERR(mnt)) {
1392                                 root = ERR_CAST(mnt);
1393                                 mnt = NULL;
1394                                 goto out;
1395                         }
1396
1397                         down_write(&mnt->mnt_sb->s_umount);
1398                         ret = btrfs_remount(mnt->mnt_sb, &flags, NULL);
1399                         up_write(&mnt->mnt_sb->s_umount);
1400                         if (ret < 0) {
1401                                 root = ERR_PTR(ret);
1402                                 goto out;
1403                         }
1404                 }
1405         }
1406         if (IS_ERR(mnt)) {
1407                 root = ERR_CAST(mnt);
1408                 mnt = NULL;
1409                 goto out;
1410         }
1411
1412         if (!subvol_name) {
1413                 if (!subvol_objectid) {
1414                         ret = get_default_subvol_objectid(btrfs_sb(mnt->mnt_sb),
1415                                                           &subvol_objectid);
1416                         if (ret) {
1417                                 root = ERR_PTR(ret);
1418                                 goto out;
1419                         }
1420                 }
1421                 subvol_name = get_subvol_name_from_objectid(btrfs_sb(mnt->mnt_sb),
1422                                                             subvol_objectid);
1423                 if (IS_ERR(subvol_name)) {
1424                         root = ERR_CAST(subvol_name);
1425                         subvol_name = NULL;
1426                         goto out;
1427                 }
1428
1429         }
1430
1431         root = mount_subtree(mnt, subvol_name);
1432         /* mount_subtree() drops our reference on the vfsmount. */
1433         mnt = NULL;
1434
1435         if (!IS_ERR(root)) {
1436                 struct super_block *s = root->d_sb;
1437                 struct inode *root_inode = d_inode(root);
1438                 u64 root_objectid = BTRFS_I(root_inode)->root->root_key.objectid;
1439
1440                 ret = 0;
1441                 if (!is_subvolume_inode(root_inode)) {
1442                         pr_err("BTRFS: '%s' is not a valid subvolume\n",
1443                                subvol_name);
1444                         ret = -EINVAL;
1445                 }
1446                 if (subvol_objectid && root_objectid != subvol_objectid) {
1447                         /*
1448                          * This will also catch a race condition where a
1449                          * subvolume which was passed by ID is renamed and
1450                          * another subvolume is renamed over the old location.
1451                          */
1452                         pr_err("BTRFS: subvol '%s' does not match subvolid %llu\n",
1453                                subvol_name, subvol_objectid);
1454                         ret = -EINVAL;
1455                 }
1456                 if (ret) {
1457                         dput(root);
1458                         root = ERR_PTR(ret);
1459                         deactivate_locked_super(s);
1460                 }
1461         }
1462
1463 out:
1464         mntput(mnt);
1465         kfree(newargs);
1466         kfree(subvol_name);
1467         return root;
1468 }
1469
1470 static int parse_security_options(char *orig_opts,
1471                                   struct security_mnt_opts *sec_opts)
1472 {
1473         char *secdata = NULL;
1474         int ret = 0;
1475
1476         secdata = alloc_secdata();
1477         if (!secdata)
1478                 return -ENOMEM;
1479         ret = security_sb_copy_data(orig_opts, secdata);
1480         if (ret) {
1481                 free_secdata(secdata);
1482                 return ret;
1483         }
1484         ret = security_sb_parse_opts_str(secdata, sec_opts);
1485         free_secdata(secdata);
1486         return ret;
1487 }
1488
1489 static int setup_security_options(struct btrfs_fs_info *fs_info,
1490                                   struct super_block *sb,
1491                                   struct security_mnt_opts *sec_opts)
1492 {
1493         int ret = 0;
1494
1495         /*
1496          * Call security_sb_set_mnt_opts() to check whether new sec_opts
1497          * is valid.
1498          */
1499         ret = security_sb_set_mnt_opts(sb, sec_opts, 0, NULL);
1500         if (ret)
1501                 return ret;
1502
1503 #ifdef CONFIG_SECURITY
1504         if (!fs_info->security_opts.num_mnt_opts) {
1505                 /* first time security setup, copy sec_opts to fs_info */
1506                 memcpy(&fs_info->security_opts, sec_opts, sizeof(*sec_opts));
1507         } else {
1508                 /*
1509                  * Since SELinux (the only one supporting security_mnt_opts)
1510                  * does NOT support changing context during remount/mount of
1511                  * the same sb, this must be the same or part of the same
1512                  * security options, just free it.
1513                  */
1514                 security_free_mnt_opts(sec_opts);
1515         }
1516 #endif
1517         return ret;
1518 }
1519
1520 /*
1521  * Find a superblock for the given device / mount point.
1522  *
1523  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1524  *        for multiple device setup.  Make sure to keep it in sync.
1525  */
1526 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1527                 const char *device_name, void *data)
1528 {
1529         struct block_device *bdev = NULL;
1530         struct super_block *s;
1531         struct btrfs_fs_devices *fs_devices = NULL;
1532         struct btrfs_fs_info *fs_info = NULL;
1533         struct security_mnt_opts new_sec_opts;
1534         fmode_t mode = FMODE_READ;
1535         char *subvol_name = NULL;
1536         u64 subvol_objectid = 0;
1537         int error = 0;
1538
1539         if (!(flags & MS_RDONLY))
1540                 mode |= FMODE_WRITE;
1541
1542         error = btrfs_parse_early_options(data, mode, fs_type,
1543                                           &subvol_name, &subvol_objectid,
1544                                           &fs_devices);
1545         if (error) {
1546                 kfree(subvol_name);
1547                 return ERR_PTR(error);
1548         }
1549
1550         if (subvol_name || subvol_objectid != BTRFS_FS_TREE_OBJECTID) {
1551                 /* mount_subvol() will free subvol_name. */
1552                 return mount_subvol(subvol_name, subvol_objectid, flags,
1553                                     device_name, data);
1554         }
1555
1556         security_init_mnt_opts(&new_sec_opts);
1557         if (data) {
1558                 error = parse_security_options(data, &new_sec_opts);
1559                 if (error)
1560                         return ERR_PTR(error);
1561         }
1562
1563         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1564         if (error)
1565                 goto error_sec_opts;
1566
1567         /*
1568          * Setup a dummy root and fs_info for test/set super.  This is because
1569          * we don't actually fill this stuff out until open_ctree, but we need
1570          * it for searching for existing supers, so this lets us do that and
1571          * then open_ctree will properly initialize everything later.
1572          */
1573         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1574         if (!fs_info) {
1575                 error = -ENOMEM;
1576                 goto error_sec_opts;
1577         }
1578
1579         fs_info->fs_devices = fs_devices;
1580
1581         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1582         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1583         security_init_mnt_opts(&fs_info->security_opts);
1584         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1585                 error = -ENOMEM;
1586                 goto error_fs_info;
1587         }
1588
1589         error = btrfs_open_devices(fs_devices, mode, fs_type);
1590         if (error)
1591                 goto error_fs_info;
1592
1593         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1594                 error = -EACCES;
1595                 goto error_close_devices;
1596         }
1597
1598         bdev = fs_devices->latest_bdev;
1599         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1600                  fs_info);
1601         if (IS_ERR(s)) {
1602                 error = PTR_ERR(s);
1603                 goto error_close_devices;
1604         }
1605
1606         if (s->s_root) {
1607                 btrfs_close_devices(fs_devices);
1608                 free_fs_info(fs_info);
1609                 if ((flags ^ s->s_flags) & MS_RDONLY)
1610                         error = -EBUSY;
1611         } else {
1612                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1613                 btrfs_sb(s)->bdev_holder = fs_type;
1614                 error = btrfs_fill_super(s, fs_devices, data,
1615                                          flags & MS_SILENT ? 1 : 0);
1616         }
1617         if (error) {
1618                 deactivate_locked_super(s);
1619                 goto error_sec_opts;
1620         }
1621
1622         fs_info = btrfs_sb(s);
1623         error = setup_security_options(fs_info, s, &new_sec_opts);
1624         if (error) {
1625                 deactivate_locked_super(s);
1626                 goto error_sec_opts;
1627         }
1628
1629         return dget(s->s_root);
1630
1631 error_close_devices:
1632         btrfs_close_devices(fs_devices);
1633 error_fs_info:
1634         free_fs_info(fs_info);
1635 error_sec_opts:
1636         security_free_mnt_opts(&new_sec_opts);
1637         return ERR_PTR(error);
1638 }
1639
1640 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1641                                      int new_pool_size, int old_pool_size)
1642 {
1643         if (new_pool_size == old_pool_size)
1644                 return;
1645
1646         fs_info->thread_pool_size = new_pool_size;
1647
1648         btrfs_info(fs_info, "resize thread pool %d -> %d",
1649                old_pool_size, new_pool_size);
1650
1651         btrfs_workqueue_set_max(fs_info->workers, new_pool_size);
1652         btrfs_workqueue_set_max(fs_info->delalloc_workers, new_pool_size);
1653         btrfs_workqueue_set_max(fs_info->submit_workers, new_pool_size);
1654         btrfs_workqueue_set_max(fs_info->caching_workers, new_pool_size);
1655         btrfs_workqueue_set_max(fs_info->endio_workers, new_pool_size);
1656         btrfs_workqueue_set_max(fs_info->endio_meta_workers, new_pool_size);
1657         btrfs_workqueue_set_max(fs_info->endio_meta_write_workers,
1658                                 new_pool_size);
1659         btrfs_workqueue_set_max(fs_info->endio_write_workers, new_pool_size);
1660         btrfs_workqueue_set_max(fs_info->endio_freespace_worker, new_pool_size);
1661         btrfs_workqueue_set_max(fs_info->delayed_workers, new_pool_size);
1662         btrfs_workqueue_set_max(fs_info->readahead_workers, new_pool_size);
1663         btrfs_workqueue_set_max(fs_info->scrub_wr_completion_workers,
1664                                 new_pool_size);
1665 }
1666
1667 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1668 {
1669         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1670 }
1671
1672 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1673                                        unsigned long old_opts, int flags)
1674 {
1675         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1676             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1677              (flags & MS_RDONLY))) {
1678                 /* wait for any defraggers to finish */
1679                 wait_event(fs_info->transaction_wait,
1680                            (atomic_read(&fs_info->defrag_running) == 0));
1681                 if (flags & MS_RDONLY)
1682                         sync_filesystem(fs_info->sb);
1683         }
1684 }
1685
1686 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1687                                          unsigned long old_opts)
1688 {
1689         /*
1690          * We need to cleanup all defragable inodes if the autodefragment is
1691          * close or the filesystem is read only.
1692          */
1693         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1694             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1695              (fs_info->sb->s_flags & MS_RDONLY))) {
1696                 btrfs_cleanup_defrag_inodes(fs_info);
1697         }
1698
1699         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1700 }
1701
1702 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1703 {
1704         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1705         struct btrfs_root *root = fs_info->tree_root;
1706         unsigned old_flags = sb->s_flags;
1707         unsigned long old_opts = fs_info->mount_opt;
1708         unsigned long old_compress_type = fs_info->compress_type;
1709         u64 old_max_inline = fs_info->max_inline;
1710         u64 old_alloc_start = fs_info->alloc_start;
1711         int old_thread_pool_size = fs_info->thread_pool_size;
1712         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1713         int ret;
1714
1715         sync_filesystem(sb);
1716         btrfs_remount_prepare(fs_info);
1717
1718         if (data) {
1719                 struct security_mnt_opts new_sec_opts;
1720
1721                 security_init_mnt_opts(&new_sec_opts);
1722                 ret = parse_security_options(data, &new_sec_opts);
1723                 if (ret)
1724                         goto restore;
1725                 ret = setup_security_options(fs_info, sb,
1726                                              &new_sec_opts);
1727                 if (ret) {
1728                         security_free_mnt_opts(&new_sec_opts);
1729                         goto restore;
1730                 }
1731         }
1732
1733         ret = btrfs_parse_options(root, data, *flags);
1734         if (ret) {
1735                 ret = -EINVAL;
1736                 goto restore;
1737         }
1738
1739         btrfs_remount_begin(fs_info, old_opts, *flags);
1740         btrfs_resize_thread_pool(fs_info,
1741                 fs_info->thread_pool_size, old_thread_pool_size);
1742
1743         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1744                 goto out;
1745
1746         if (*flags & MS_RDONLY) {
1747                 /*
1748                  * this also happens on 'umount -rf' or on shutdown, when
1749                  * the filesystem is busy.
1750                  */
1751                 cancel_work_sync(&fs_info->async_reclaim_work);
1752
1753                 /* wait for the uuid_scan task to finish */
1754                 down(&fs_info->uuid_tree_rescan_sem);
1755                 /* avoid complains from lockdep et al. */
1756                 up(&fs_info->uuid_tree_rescan_sem);
1757
1758                 sb->s_flags |= MS_RDONLY;
1759
1760                 /*
1761                  * Setting MS_RDONLY will put the cleaner thread to
1762                  * sleep at the next loop if it's already active.
1763                  * If it's already asleep, we'll leave unused block
1764                  * groups on disk until we're mounted read-write again
1765                  * unless we clean them up here.
1766                  */
1767                 btrfs_delete_unused_bgs(fs_info);
1768
1769                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1770                 btrfs_scrub_cancel(fs_info);
1771                 btrfs_pause_balance(fs_info);
1772
1773                 ret = btrfs_commit_super(root);
1774                 if (ret)
1775                         goto restore;
1776         } else {
1777                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1778                         btrfs_err(fs_info,
1779                                 "Remounting read-write after error is not allowed");
1780                         ret = -EINVAL;
1781                         goto restore;
1782                 }
1783                 if (fs_info->fs_devices->rw_devices == 0) {
1784                         ret = -EACCES;
1785                         goto restore;
1786                 }
1787
1788                 if (fs_info->fs_devices->missing_devices >
1789                      fs_info->num_tolerated_disk_barrier_failures &&
1790                     !(*flags & MS_RDONLY)) {
1791                         btrfs_warn(fs_info,
1792                                 "too many missing devices, writeable remount is not allowed");
1793                         ret = -EACCES;
1794                         goto restore;
1795                 }
1796
1797                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1798                         ret = -EINVAL;
1799                         goto restore;
1800                 }
1801
1802                 ret = btrfs_cleanup_fs_roots(fs_info);
1803                 if (ret)
1804                         goto restore;
1805
1806                 /* recover relocation */
1807                 mutex_lock(&fs_info->cleaner_mutex);
1808                 ret = btrfs_recover_relocation(root);
1809                 mutex_unlock(&fs_info->cleaner_mutex);
1810                 if (ret)
1811                         goto restore;
1812
1813                 ret = btrfs_resume_balance_async(fs_info);
1814                 if (ret)
1815                         goto restore;
1816
1817                 ret = btrfs_resume_dev_replace_async(fs_info);
1818                 if (ret) {
1819                         btrfs_warn(fs_info, "failed to resume dev_replace");
1820                         goto restore;
1821                 }
1822
1823                 if (!fs_info->uuid_root) {
1824                         btrfs_info(fs_info, "creating UUID tree");
1825                         ret = btrfs_create_uuid_tree(fs_info);
1826                         if (ret) {
1827                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1828                                 goto restore;
1829                         }
1830                 }
1831                 sb->s_flags &= ~MS_RDONLY;
1832
1833                 fs_info->open = 1;
1834         }
1835 out:
1836         wake_up_process(fs_info->transaction_kthread);
1837         btrfs_remount_cleanup(fs_info, old_opts);
1838         return 0;
1839
1840 restore:
1841         /* We've hit an error - don't reset MS_RDONLY */
1842         if (sb->s_flags & MS_RDONLY)
1843                 old_flags |= MS_RDONLY;
1844         sb->s_flags = old_flags;
1845         fs_info->mount_opt = old_opts;
1846         fs_info->compress_type = old_compress_type;
1847         fs_info->max_inline = old_max_inline;
1848         mutex_lock(&fs_info->chunk_mutex);
1849         fs_info->alloc_start = old_alloc_start;
1850         mutex_unlock(&fs_info->chunk_mutex);
1851         btrfs_resize_thread_pool(fs_info,
1852                 old_thread_pool_size, fs_info->thread_pool_size);
1853         fs_info->metadata_ratio = old_metadata_ratio;
1854         btrfs_remount_cleanup(fs_info, old_opts);
1855         return ret;
1856 }
1857
1858 /* Used to sort the devices by max_avail(descending sort) */
1859 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1860                                        const void *dev_info2)
1861 {
1862         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1863             ((struct btrfs_device_info *)dev_info2)->max_avail)
1864                 return -1;
1865         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1866                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1867                 return 1;
1868         else
1869         return 0;
1870 }
1871
1872 /*
1873  * sort the devices by max_avail, in which max free extent size of each device
1874  * is stored.(Descending Sort)
1875  */
1876 static inline void btrfs_descending_sort_devices(
1877                                         struct btrfs_device_info *devices,
1878                                         size_t nr_devices)
1879 {
1880         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1881              btrfs_cmp_device_free_bytes, NULL);
1882 }
1883
1884 /*
1885  * The helper to calc the free space on the devices that can be used to store
1886  * file data.
1887  */
1888 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1889 {
1890         struct btrfs_fs_info *fs_info = root->fs_info;
1891         struct btrfs_device_info *devices_info;
1892         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1893         struct btrfs_device *device;
1894         u64 skip_space;
1895         u64 type;
1896         u64 avail_space;
1897         u64 used_space;
1898         u64 min_stripe_size;
1899         int min_stripes = 1, num_stripes = 1;
1900         int i = 0, nr_devices;
1901         int ret;
1902
1903         /*
1904          * We aren't under the device list lock, so this is racy-ish, but good
1905          * enough for our purposes.
1906          */
1907         nr_devices = fs_info->fs_devices->open_devices;
1908         if (!nr_devices) {
1909                 smp_mb();
1910                 nr_devices = fs_info->fs_devices->open_devices;
1911                 ASSERT(nr_devices);
1912                 if (!nr_devices) {
1913                         *free_bytes = 0;
1914                         return 0;
1915                 }
1916         }
1917
1918         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1919                                GFP_NOFS);
1920         if (!devices_info)
1921                 return -ENOMEM;
1922
1923         /* calc min stripe number for data space allocation */
1924         type = btrfs_get_alloc_profile(root, 1);
1925         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1926                 min_stripes = 2;
1927                 num_stripes = nr_devices;
1928         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1929                 min_stripes = 2;
1930                 num_stripes = 2;
1931         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1932                 min_stripes = 4;
1933                 num_stripes = 4;
1934         }
1935
1936         if (type & BTRFS_BLOCK_GROUP_DUP)
1937                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1938         else
1939                 min_stripe_size = BTRFS_STRIPE_LEN;
1940
1941         if (fs_info->alloc_start)
1942                 mutex_lock(&fs_devices->device_list_mutex);
1943         rcu_read_lock();
1944         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
1945                 if (!device->in_fs_metadata || !device->bdev ||
1946                     device->is_tgtdev_for_dev_replace)
1947                         continue;
1948
1949                 if (i >= nr_devices)
1950                         break;
1951
1952                 avail_space = device->total_bytes - device->bytes_used;
1953
1954                 /* align with stripe_len */
1955                 avail_space = div_u64(avail_space, BTRFS_STRIPE_LEN);
1956                 avail_space *= BTRFS_STRIPE_LEN;
1957
1958                 /*
1959                  * In order to avoid overwriting the superblock on the drive,
1960                  * btrfs starts at an offset of at least 1MB when doing chunk
1961                  * allocation.
1962                  */
1963                 skip_space = SZ_1M;
1964
1965                 /* user can set the offset in fs_info->alloc_start. */
1966                 if (fs_info->alloc_start &&
1967                     fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1968                     device->total_bytes) {
1969                         rcu_read_unlock();
1970                         skip_space = max(fs_info->alloc_start, skip_space);
1971
1972                         /*
1973                          * btrfs can not use the free space in
1974                          * [0, skip_space - 1], we must subtract it from the
1975                          * total. In order to implement it, we account the used
1976                          * space in this range first.
1977                          */
1978                         ret = btrfs_account_dev_extents_size(device, 0,
1979                                                              skip_space - 1,
1980                                                              &used_space);
1981                         if (ret) {
1982                                 kfree(devices_info);
1983                                 mutex_unlock(&fs_devices->device_list_mutex);
1984                                 return ret;
1985                         }
1986
1987                         rcu_read_lock();
1988
1989                         /* calc the free space in [0, skip_space - 1] */
1990                         skip_space -= used_space;
1991                 }
1992
1993                 /*
1994                  * we can use the free space in [0, skip_space - 1], subtract
1995                  * it from the total.
1996                  */
1997                 if (avail_space && avail_space >= skip_space)
1998                         avail_space -= skip_space;
1999                 else
2000                         avail_space = 0;
2001
2002                 if (avail_space < min_stripe_size)
2003                         continue;
2004
2005                 devices_info[i].dev = device;
2006                 devices_info[i].max_avail = avail_space;
2007
2008                 i++;
2009         }
2010         rcu_read_unlock();
2011         if (fs_info->alloc_start)
2012                 mutex_unlock(&fs_devices->device_list_mutex);
2013
2014         nr_devices = i;
2015
2016         btrfs_descending_sort_devices(devices_info, nr_devices);
2017
2018         i = nr_devices - 1;
2019         avail_space = 0;
2020         while (nr_devices >= min_stripes) {
2021                 if (num_stripes > nr_devices)
2022                         num_stripes = nr_devices;
2023
2024                 if (devices_info[i].max_avail >= min_stripe_size) {
2025                         int j;
2026                         u64 alloc_size;
2027
2028                         avail_space += devices_info[i].max_avail * num_stripes;
2029                         alloc_size = devices_info[i].max_avail;
2030                         for (j = i + 1 - num_stripes; j <= i; j++)
2031                                 devices_info[j].max_avail -= alloc_size;
2032                 }
2033                 i--;
2034                 nr_devices--;
2035         }
2036
2037         kfree(devices_info);
2038         *free_bytes = avail_space;
2039         return 0;
2040 }
2041
2042 /*
2043  * Calculate numbers for 'df', pessimistic in case of mixed raid profiles.
2044  *
2045  * If there's a redundant raid level at DATA block groups, use the respective
2046  * multiplier to scale the sizes.
2047  *
2048  * Unused device space usage is based on simulating the chunk allocator
2049  * algorithm that respects the device sizes, order of allocations and the
2050  * 'alloc_start' value, this is a close approximation of the actual use but
2051  * there are other factors that may change the result (like a new metadata
2052  * chunk).
2053  *
2054  * If metadata is exhausted, f_bavail will be 0.
2055  */
2056 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2057 {
2058         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
2059         struct btrfs_super_block *disk_super = fs_info->super_copy;
2060         struct list_head *head = &fs_info->space_info;
2061         struct btrfs_space_info *found;
2062         u64 total_used = 0;
2063         u64 total_free_data = 0;
2064         u64 total_free_meta = 0;
2065         int bits = dentry->d_sb->s_blocksize_bits;
2066         __be32 *fsid = (__be32 *)fs_info->fsid;
2067         unsigned factor = 1;
2068         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
2069         int ret;
2070         u64 thresh = 0;
2071         int mixed = 0;
2072
2073         /*
2074          * holding chunk_mutex to avoid allocating new chunks, holding
2075          * device_list_mutex to avoid the device being removed
2076          */
2077         rcu_read_lock();
2078         list_for_each_entry_rcu(found, head, list) {
2079                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
2080                         int i;
2081
2082                         total_free_data += found->disk_total - found->disk_used;
2083                         total_free_data -=
2084                                 btrfs_account_ro_block_groups_free_space(found);
2085
2086                         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2087                                 if (!list_empty(&found->block_groups[i])) {
2088                                         switch (i) {
2089                                         case BTRFS_RAID_DUP:
2090                                         case BTRFS_RAID_RAID1:
2091                                         case BTRFS_RAID_RAID10:
2092                                                 factor = 2;
2093                                         }
2094                                 }
2095                         }
2096                 }
2097
2098                 /*
2099                  * Metadata in mixed block goup profiles are accounted in data
2100                  */
2101                 if (!mixed && found->flags & BTRFS_BLOCK_GROUP_METADATA) {
2102                         if (found->flags & BTRFS_BLOCK_GROUP_DATA)
2103                                 mixed = 1;
2104                         else
2105                                 total_free_meta += found->disk_total -
2106                                         found->disk_used;
2107                 }
2108
2109                 total_used += found->disk_used;
2110         }
2111
2112         rcu_read_unlock();
2113
2114         buf->f_blocks = div_u64(btrfs_super_total_bytes(disk_super), factor);
2115         buf->f_blocks >>= bits;
2116         buf->f_bfree = buf->f_blocks - (div_u64(total_used, factor) >> bits);
2117
2118         /* Account global block reserve as used, it's in logical size already */
2119         spin_lock(&block_rsv->lock);
2120         /* Mixed block groups accounting is not byte-accurate, avoid overflow */
2121         if (buf->f_bfree >= block_rsv->size >> bits)
2122                 buf->f_bfree -= block_rsv->size >> bits;
2123         else
2124                 buf->f_bfree = 0;
2125         spin_unlock(&block_rsv->lock);
2126
2127         buf->f_bavail = div_u64(total_free_data, factor);
2128         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
2129         if (ret)
2130                 return ret;
2131         buf->f_bavail += div_u64(total_free_data, factor);
2132         buf->f_bavail = buf->f_bavail >> bits;
2133
2134         /*
2135          * We calculate the remaining metadata space minus global reserve. If
2136          * this is (supposedly) smaller than zero, there's no space. But this
2137          * does not hold in practice, the exhausted state happens where's still
2138          * some positive delta. So we apply some guesswork and compare the
2139          * delta to a 4M threshold.  (Practically observed delta was ~2M.)
2140          *
2141          * We probably cannot calculate the exact threshold value because this
2142          * depends on the internal reservations requested by various
2143          * operations, so some operations that consume a few metadata will
2144          * succeed even if the Avail is zero. But this is better than the other
2145          * way around.
2146          */
2147         thresh = 4 * 1024 * 1024;
2148
2149         if (!mixed && total_free_meta - thresh < block_rsv->size)
2150                 buf->f_bavail = 0;
2151
2152         buf->f_type = BTRFS_SUPER_MAGIC;
2153         buf->f_bsize = dentry->d_sb->s_blocksize;
2154         buf->f_namelen = BTRFS_NAME_LEN;
2155
2156         /* We treat it as constant endianness (it doesn't matter _which_)
2157            because we want the fsid to come out the same whether mounted
2158            on a big-endian or little-endian host */
2159         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
2160         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
2161         /* Mask in the root object ID too, to disambiguate subvols */
2162         buf->f_fsid.val[0] ^= BTRFS_I(d_inode(dentry))->root->objectid >> 32;
2163         buf->f_fsid.val[1] ^= BTRFS_I(d_inode(dentry))->root->objectid;
2164
2165         return 0;
2166 }
2167
2168 static void btrfs_kill_super(struct super_block *sb)
2169 {
2170         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2171         kill_anon_super(sb);
2172         free_fs_info(fs_info);
2173 }
2174
2175 static struct file_system_type btrfs_fs_type = {
2176         .owner          = THIS_MODULE,
2177         .name           = "btrfs",
2178         .mount          = btrfs_mount,
2179         .kill_sb        = btrfs_kill_super,
2180         .fs_flags       = FS_REQUIRES_DEV | FS_BINARY_MOUNTDATA,
2181 };
2182 MODULE_ALIAS_FS("btrfs");
2183
2184 static int btrfs_control_open(struct inode *inode, struct file *file)
2185 {
2186         /*
2187          * The control file's private_data is used to hold the
2188          * transaction when it is started and is used to keep
2189          * track of whether a transaction is already in progress.
2190          */
2191         file->private_data = NULL;
2192         return 0;
2193 }
2194
2195 /*
2196  * used by btrfsctl to scan devices when no FS is mounted
2197  */
2198 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
2199                                 unsigned long arg)
2200 {
2201         struct btrfs_ioctl_vol_args *vol;
2202         struct btrfs_fs_devices *fs_devices;
2203         int ret = -ENOTTY;
2204
2205         if (!capable(CAP_SYS_ADMIN))
2206                 return -EPERM;
2207
2208         vol = memdup_user((void __user *)arg, sizeof(*vol));
2209         if (IS_ERR(vol))
2210                 return PTR_ERR(vol);
2211
2212         switch (cmd) {
2213         case BTRFS_IOC_SCAN_DEV:
2214                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2215                                             &btrfs_fs_type, &fs_devices);
2216                 break;
2217         case BTRFS_IOC_DEVICES_READY:
2218                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
2219                                             &btrfs_fs_type, &fs_devices);
2220                 if (ret)
2221                         break;
2222                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
2223                 break;
2224         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
2225                 ret = btrfs_ioctl_get_supported_features((void __user*)arg);
2226                 break;
2227         }
2228
2229         kfree(vol);
2230         return ret;
2231 }
2232
2233 static int btrfs_freeze(struct super_block *sb)
2234 {
2235         struct btrfs_trans_handle *trans;
2236         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
2237
2238         trans = btrfs_attach_transaction_barrier(root);
2239         if (IS_ERR(trans)) {
2240                 /* no transaction, don't bother */
2241                 if (PTR_ERR(trans) == -ENOENT)
2242                         return 0;
2243                 return PTR_ERR(trans);
2244         }
2245         return btrfs_commit_transaction(trans, root);
2246 }
2247
2248 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
2249 {
2250         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
2251         struct btrfs_fs_devices *cur_devices;
2252         struct btrfs_device *dev, *first_dev = NULL;
2253         struct list_head *head;
2254         struct rcu_string *name;
2255
2256         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2257         cur_devices = fs_info->fs_devices;
2258         while (cur_devices) {
2259                 head = &cur_devices->devices;
2260                 list_for_each_entry(dev, head, dev_list) {
2261                         if (dev->missing)
2262                                 continue;
2263                         if (!dev->name)
2264                                 continue;
2265                         if (!first_dev || dev->devid < first_dev->devid)
2266                                 first_dev = dev;
2267                 }
2268                 cur_devices = cur_devices->seed;
2269         }
2270
2271         if (first_dev) {
2272                 rcu_read_lock();
2273                 name = rcu_dereference(first_dev->name);
2274                 seq_escape(m, name->str, " \t\n\\");
2275                 rcu_read_unlock();
2276         } else {
2277                 WARN_ON(1);
2278         }
2279         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2280         return 0;
2281 }
2282
2283 static const struct super_operations btrfs_super_ops = {
2284         .drop_inode     = btrfs_drop_inode,
2285         .evict_inode    = btrfs_evict_inode,
2286         .put_super      = btrfs_put_super,
2287         .sync_fs        = btrfs_sync_fs,
2288         .show_options   = btrfs_show_options,
2289         .show_devname   = btrfs_show_devname,
2290         .write_inode    = btrfs_write_inode,
2291         .alloc_inode    = btrfs_alloc_inode,
2292         .destroy_inode  = btrfs_destroy_inode,
2293         .statfs         = btrfs_statfs,
2294         .remount_fs     = btrfs_remount,
2295         .freeze_fs      = btrfs_freeze,
2296 };
2297
2298 static const struct file_operations btrfs_ctl_fops = {
2299         .open = btrfs_control_open,
2300         .unlocked_ioctl  = btrfs_control_ioctl,
2301         .compat_ioctl = btrfs_control_ioctl,
2302         .owner   = THIS_MODULE,
2303         .llseek = noop_llseek,
2304 };
2305
2306 static struct miscdevice btrfs_misc = {
2307         .minor          = BTRFS_MINOR,
2308         .name           = "btrfs-control",
2309         .fops           = &btrfs_ctl_fops
2310 };
2311
2312 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
2313 MODULE_ALIAS("devname:btrfs-control");
2314
2315 static int btrfs_interface_init(void)
2316 {
2317         return misc_register(&btrfs_misc);
2318 }
2319
2320 static void btrfs_interface_exit(void)
2321 {
2322         misc_deregister(&btrfs_misc);
2323 }
2324
2325 static void btrfs_print_mod_info(void)
2326 {
2327         printk(KERN_INFO "Btrfs loaded, crc32c=%s"
2328 #ifdef CONFIG_BTRFS_DEBUG
2329                         ", debug=on"
2330 #endif
2331 #ifdef CONFIG_BTRFS_ASSERT
2332                         ", assert=on"
2333 #endif
2334 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2335                         ", integrity-checker=on"
2336 #endif
2337                         "\n",
2338                         btrfs_crc32c_impl());
2339 }
2340
2341 static int btrfs_run_sanity_tests(void)
2342 {
2343         int ret, i;
2344         u32 sectorsize, nodesize;
2345         u32 test_sectorsize[] = {
2346                 PAGE_SIZE,
2347         };
2348         ret = btrfs_init_test_fs();
2349         if (ret)
2350                 return ret;
2351         for (i = 0; i < ARRAY_SIZE(test_sectorsize); i++) {
2352                 sectorsize = test_sectorsize[i];
2353                 for (nodesize = sectorsize;
2354                      nodesize <= BTRFS_MAX_METADATA_BLOCKSIZE;
2355                      nodesize <<= 1) {
2356                         pr_info("BTRFS: selftest: sectorsize: %u  nodesize: %u\n",
2357                                 sectorsize, nodesize);
2358                         ret = btrfs_test_free_space_cache(sectorsize, nodesize);
2359                         if (ret)
2360                                 goto out;
2361                         ret = btrfs_test_extent_buffer_operations(sectorsize,
2362                                 nodesize);
2363                         if (ret)
2364                                 goto out;
2365                         ret = btrfs_test_extent_io(sectorsize, nodesize);
2366                         if (ret)
2367                                 goto out;
2368                         ret = btrfs_test_inodes(sectorsize, nodesize);
2369                         if (ret)
2370                                 goto out;
2371                         ret = btrfs_test_qgroups(sectorsize, nodesize);
2372                         if (ret)
2373                                 goto out;
2374                         ret = btrfs_test_free_space_tree(sectorsize, nodesize);
2375                         if (ret)
2376                                 goto out;
2377                 }
2378         }
2379 out:
2380         btrfs_destroy_test_fs();
2381         return ret;
2382 }
2383
2384 static int __init init_btrfs_fs(void)
2385 {
2386         int err;
2387
2388         err = btrfs_hash_init();
2389         if (err)
2390                 return err;
2391
2392         btrfs_props_init();
2393
2394         err = btrfs_init_sysfs();
2395         if (err)
2396                 goto free_hash;
2397
2398         btrfs_init_compress();
2399
2400         err = btrfs_init_cachep();
2401         if (err)
2402                 goto free_compress;
2403
2404         err = extent_io_init();
2405         if (err)
2406                 goto free_cachep;
2407
2408         err = extent_map_init();
2409         if (err)
2410                 goto free_extent_io;
2411
2412         err = ordered_data_init();
2413         if (err)
2414                 goto free_extent_map;
2415
2416         err = btrfs_delayed_inode_init();
2417         if (err)
2418                 goto free_ordered_data;
2419
2420         err = btrfs_auto_defrag_init();
2421         if (err)
2422                 goto free_delayed_inode;
2423
2424         err = btrfs_delayed_ref_init();
2425         if (err)
2426                 goto free_auto_defrag;
2427
2428         err = btrfs_prelim_ref_init();
2429         if (err)
2430                 goto free_delayed_ref;
2431
2432         err = btrfs_end_io_wq_init();
2433         if (err)
2434                 goto free_prelim_ref;
2435
2436         err = btrfs_interface_init();
2437         if (err)
2438                 goto free_end_io_wq;
2439
2440         btrfs_init_lockdep();
2441
2442         btrfs_print_mod_info();
2443
2444         err = btrfs_run_sanity_tests();
2445         if (err)
2446                 goto unregister_ioctl;
2447
2448         err = register_filesystem(&btrfs_fs_type);
2449         if (err)
2450                 goto unregister_ioctl;
2451
2452         return 0;
2453
2454 unregister_ioctl:
2455         btrfs_interface_exit();
2456 free_end_io_wq:
2457         btrfs_end_io_wq_exit();
2458 free_prelim_ref:
2459         btrfs_prelim_ref_exit();
2460 free_delayed_ref:
2461         btrfs_delayed_ref_exit();
2462 free_auto_defrag:
2463         btrfs_auto_defrag_exit();
2464 free_delayed_inode:
2465         btrfs_delayed_inode_exit();
2466 free_ordered_data:
2467         ordered_data_exit();
2468 free_extent_map:
2469         extent_map_exit();
2470 free_extent_io:
2471         extent_io_exit();
2472 free_cachep:
2473         btrfs_destroy_cachep();
2474 free_compress:
2475         btrfs_exit_compress();
2476         btrfs_exit_sysfs();
2477 free_hash:
2478         btrfs_hash_exit();
2479         return err;
2480 }
2481
2482 static void __exit exit_btrfs_fs(void)
2483 {
2484         btrfs_destroy_cachep();
2485         btrfs_delayed_ref_exit();
2486         btrfs_auto_defrag_exit();
2487         btrfs_delayed_inode_exit();
2488         btrfs_prelim_ref_exit();
2489         ordered_data_exit();
2490         extent_map_exit();
2491         extent_io_exit();
2492         btrfs_interface_exit();
2493         btrfs_end_io_wq_exit();
2494         unregister_filesystem(&btrfs_fs_type);
2495         btrfs_exit_sysfs();
2496         btrfs_cleanup_fs_uuids();
2497         btrfs_exit_compress();
2498         btrfs_hash_exit();
2499 }
2500
2501 late_initcall(init_btrfs_fs);
2502 module_exit(exit_btrfs_fs)
2503
2504 MODULE_LICENSE("GPL");