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 "xattr.h"
  52 #include "volumes.h"
  53 #include "export.h"
  54 #include "compression.h"
  55 #include "rcu-string.h"
  56 #include "dev-replace.h"
  57 #include "free-space-cache.h"
  58 #include "backref.h"
  59 #include "tests/btrfs-tests.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/btrfs.h>
  63
  64 static const struct super_operations btrfs_super_ops;
  65 static struct file_system_type btrfs_fs_type;
  66
  67 static const char *btrfs_decode_error(int errno)
  68 {
  69         char *errstr = "unknown";
  70
  71         switch (errno) {
  72         case -EIO:
  73                 errstr = "IO failure";
  74                 break;
  75         case -ENOMEM:
  76                 errstr = "Out of memory";
  77                 break;
  78         case -EROFS:
  79                 errstr = "Readonly filesystem";
  80                 break;
  81         case -EEXIST:
  82                 errstr = "Object already exists";
  83                 break;
  84         case -ENOSPC:
  85                 errstr = "No space left";
  86                 break;
  87         case -ENOENT:
  88                 errstr = "No such entry";
  89                 break;
  90         }
  91
  92         return errstr;
  93 }
  94
  95 static void save_error_info(struct btrfs_fs_info *fs_info)
  96 {
  97         /*
  98          * today we only save the error info into ram.  Long term we'll
  99          * also send it down to the disk
 100          */
 101         set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
 102 }
 103
 104 /* btrfs handle error by forcing the filesystem readonly */
 105 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 106 {
 107         struct super_block *sb = fs_info->sb;
 108
 109         if (sb->s_flags & MS_RDONLY)
 110                 return;
 111
 112         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
 113                 sb->s_flags |= MS_RDONLY;
 114                 btrfs_info(fs_info, "forced readonly");
 115                 /*
 116                  * Note that a running device replace operation is not
 117                  * canceled here although there is no way to update
 118                  * the progress. It would add the risk of a deadlock,
 119                  * therefore the canceling is ommited. The only penalty
 120                  * is that some I/O remains active until the procedure
 121                  * completes. The next time when the filesystem is
 122                  * mounted writeable again, the device replace
 123                  * operation continues.
 124                  */
 125         }
 126 }
 127
 128 #ifdef CONFIG_PRINTK
 129 /*
 130  * __btrfs_std_error decodes expected errors from the caller and
 131  * invokes the approciate error response.
 132  */
 133 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 134                        unsigned int line, int errno, const char *fmt, ...)
 135 {
 136         struct super_block *sb = fs_info->sb;
 137         const char *errstr;
 138
 139         /*
 140          * Special case: if the error is EROFS, and we're already
 141          * under MS_RDONLY, then it is safe here.
 142          */
 143         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 144                 return;
 145
 146         errstr = btrfs_decode_error(errno);
 147         if (fmt) {
 148                 struct va_format vaf;
 149                 va_list args;
 150
 151                 va_start(args, fmt);
 152                 vaf.fmt = fmt;
 153                 vaf.va = &args;
 154
 155                 printk(KERN_CRIT
 156                         "BTRFS: error (device %s) in %s:%d: errno=%d %s (%pV)\n",
 157                         sb->s_id, function, line, errno, errstr, &vaf);
 158                 va_end(args);
 159         } else {
 160                 printk(KERN_CRIT "BTRFS: error (device %s) in %s:%d: errno=%d %s\n",
 161                         sb->s_id, function, line, errno, errstr);
 162         }
 163
 164         /* Don't go through full error handling during mount */
 165         save_error_info(fs_info);
 166         if (sb->s_flags & MS_BORN)
 167                 btrfs_handle_error(fs_info);
 168 }
 169
 170 static const char * const logtypes[] = {
 171         "emergency",
 172         "alert",
 173         "critical",
 174         "error",
 175         "warning",
 176         "notice",
 177         "info",
 178         "debug",
 179 };
 180
 181 void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
 182 {
 183         struct super_block *sb = fs_info->sb;
 184         char lvl[4];
 185         struct va_format vaf;
 186         va_list args;
 187         const char *type = logtypes[4];
 188         int kern_level;
 189
 190         va_start(args, fmt);
 191
 192         kern_level = printk_get_level(fmt);
 193         if (kern_level) {
 194                 size_t size = printk_skip_level(fmt) - fmt;
 195                 memcpy(lvl, fmt,  size);
 196                 lvl[size] = '\0';
 197                 fmt += size;
 198                 type = logtypes[kern_level - '0'];
 199         } else
 200                 *lvl = '\0';
 201
 202         vaf.fmt = fmt;
 203         vaf.va = &args;
 204
 205         printk("%sBTRFS %s (device %s): %pV\n", lvl, type, sb->s_id, &vaf);
 206
 207         va_end(args);
 208 }
 209
 210 #else
 211
 212 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 213                        unsigned int line, int errno, const char *fmt, ...)
 214 {
 215         struct super_block *sb = fs_info->sb;
 216
 217         /*
 218          * Special case: if the error is EROFS, and we're already
 219          * under MS_RDONLY, then it is safe here.
 220          */
 221         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 222                 return;
 223
 224         /* Don't go through full error handling during mount */
 225         if (sb->s_flags & MS_BORN) {
 226                 save_error_info(fs_info);
 227                 btrfs_handle_error(fs_info);
 228         }
 229 }
 230 #endif
 231
 232 /*
 233  * We only mark the transaction aborted and then set the file system read-only.
 234  * This will prevent new transactions from starting or trying to join this
 235  * one.
 236  *
 237  * This means that error recovery at the call site is limited to freeing
 238  * any local memory allocations and passing the error code up without
 239  * further cleanup. The transaction should complete as it normally would
 240  * in the call path but will return -EIO.
 241  *
 242  * We'll complete the cleanup in btrfs_end_transaction and
 243  * btrfs_commit_transaction.
 244  */
 245 void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
 246                                struct btrfs_root *root, const char *function,
 247                                unsigned int line, int errno)
 248 {
 249         /*
 250          * Report first abort since mount
 251          */
 252         if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,
 253                                 &root->fs_info->fs_state)) {
 254                 WARN(1, KERN_DEBUG "BTRFS: Transaction aborted (error %d)\n",
 255                                 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->blocks_used) {
 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_std_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 void __btrfs_panic(struct btrfs_fs_info *fs_info, const char *function,
 280                    unsigned int line, int errno, const char *fmt, ...)
 281 {
 282         char *s_id = "<unknown>";
 283         const char *errstr;
 284         struct va_format vaf = { .fmt = fmt };
 285         va_list args;
 286
 287         if (fs_info)
 288                 s_id = fs_info->sb->s_id;
 289
 290         va_start(args, fmt);
 291         vaf.va = &args;
 292
 293         errstr = btrfs_decode_error(errno);
 294         if (fs_info && (fs_info->mount_opt & BTRFS_MOUNT_PANIC_ON_FATAL_ERROR))
 295                 panic(KERN_CRIT "BTRFS panic (device %s) in %s:%d: %pV (errno=%d %s)\n",
 296                         s_id, function, line, &vaf, errno, errstr);
 297
 298         btrfs_crit(fs_info, "panic in %s:%d: %pV (errno=%d %s)",
 299                    function, line, &vaf, errno, errstr);
 300         va_end(args);
 301         /* Caller calls BUG() */
 302 }
 303
 304 static void btrfs_put_super(struct super_block *sb)
 305 {
 306         (void)close_ctree(btrfs_sb(sb)->tree_root);
 307         /* FIXME: need to fix VFS to return error? */
 308         /* AV: return it _where_?  ->put_super() can be triggered by any number
 309          * of async events, up to and including delivery of SIGKILL to the
 310          * last process that kept it busy.  Or segfault in the aforementioned
 311          * process...  Whom would you report that to?
 312          */
 313 }
 314
 315 enum {
 316         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 317         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 318         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 319         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 320         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 321         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 322         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag, Opt_inode_cache,
 323         Opt_no_space_cache, Opt_recovery, Opt_skip_balance,
 324         Opt_check_integrity, Opt_check_integrity_including_extent_data,
 325         Opt_check_integrity_print_mask, Opt_fatal_errors, Opt_rescan_uuid_tree,
 326         Opt_commit_interval, Opt_barrier,
 327         Opt_err,
 328 };
 329
 330 static match_table_t tokens = {
 331         {Opt_degraded, "degraded"},
 332         {Opt_subvol, "subvol=%s"},
 333         {Opt_subvolid, "subvolid=%s"},
 334         {Opt_device, "device=%s"},
 335         {Opt_nodatasum, "nodatasum"},
 336         {Opt_nodatacow, "nodatacow"},
 337         {Opt_nobarrier, "nobarrier"},
 338         {Opt_barrier, "barrier"},
 339         {Opt_max_inline, "max_inline=%s"},
 340         {Opt_alloc_start, "alloc_start=%s"},
 341         {Opt_thread_pool, "thread_pool=%d"},
 342         {Opt_compress, "compress"},
 343         {Opt_compress_type, "compress=%s"},
 344         {Opt_compress_force, "compress-force"},
 345         {Opt_compress_force_type, "compress-force=%s"},
 346         {Opt_ssd, "ssd"},
 347         {Opt_ssd_spread, "ssd_spread"},
 348         {Opt_nossd, "nossd"},
 349         {Opt_noacl, "noacl"},
 350         {Opt_notreelog, "notreelog"},
 351         {Opt_flushoncommit, "flushoncommit"},
 352         {Opt_ratio, "metadata_ratio=%d"},
 353         {Opt_discard, "discard"},
 354         {Opt_space_cache, "space_cache"},
 355         {Opt_clear_cache, "clear_cache"},
 356         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 357         {Opt_enospc_debug, "enospc_debug"},
 358         {Opt_subvolrootid, "subvolrootid=%d"},
 359         {Opt_defrag, "autodefrag"},
 360         {Opt_inode_cache, "inode_cache"},
 361         {Opt_no_space_cache, "nospace_cache"},
 362         {Opt_recovery, "recovery"},
 363         {Opt_skip_balance, "skip_balance"},
 364         {Opt_check_integrity, "check_int"},
 365         {Opt_check_integrity_including_extent_data, "check_int_data"},
 366         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 367         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 368         {Opt_fatal_errors, "fatal_errors=%s"},
 369         {Opt_commit_interval, "commit=%d"},
 370         {Opt_err, NULL},
 371 };
 372
 373 /*
 374  * Regular mount options parser.  Everything that is needed only when
 375  * reading in a new superblock is parsed here.
 376  * XXX JDM: This needs to be cleaned up for remount.
 377  */
 378 int btrfs_parse_options(struct btrfs_root *root, char *options)
 379 {
 380         struct btrfs_fs_info *info = root->fs_info;
 381         substring_t args[MAX_OPT_ARGS];
 382         char *p, *num, *orig = NULL;
 383         u64 cache_gen;
 384         int intarg;
 385         int ret = 0;
 386         char *compress_type;
 387         bool compress_force = false;
 388
 389         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 390         if (cache_gen)
 391                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 392
 393         if (!options)
 394                 goto out;
 395
 396         /*
 397          * strsep changes the string, duplicate it because parse_options
 398          * gets called twice
 399          */
 400         options = kstrdup(options, GFP_NOFS);
 401         if (!options)
 402                 return -ENOMEM;
 403
 404         orig = options;
 405
 406         while ((p = strsep(&options, ",")) != NULL) {
 407                 int token;
 408                 if (!*p)
 409                         continue;
 410
 411                 token = match_token(p, tokens, args);
 412                 switch (token) {
 413                 case Opt_degraded:
 414                         btrfs_info(root->fs_info, "allowing degraded mounts");
 415                         btrfs_set_opt(info->mount_opt, DEGRADED);
 416                         break;
 417                 case Opt_subvol:
 418                 case Opt_subvolid:
 419                 case Opt_subvolrootid:
 420                 case Opt_device:
 421                         /*
 422                          * These are parsed by btrfs_parse_early_options
 423                          * and can be happily ignored here.
 424                          */
 425                         break;
 426                 case Opt_nodatasum:
 427                         btrfs_info(root->fs_info, "setting nodatasum");
 428                         btrfs_set_opt(info->mount_opt, NODATASUM);
 429                         break;
 430                 case Opt_nodatacow:
 431                         if (!btrfs_test_opt(root, COMPRESS) ||
 432                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 433                                         btrfs_info(root->fs_info,
 434                                                 "setting nodatacow, compression disabled");
 435                         } else {
 436                                 btrfs_info(root->fs_info, "setting nodatacow");
 437                         }
 438                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 439                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 440                         btrfs_set_opt(info->mount_opt, NODATACOW);
 441                         btrfs_set_opt(info->mount_opt, NODATASUM);
 442                         break;
 443                 case Opt_compress_force:
 444                 case Opt_compress_force_type:
 445                         compress_force = true;
 446                         /* Fallthrough */
 447                 case Opt_compress:
 448                 case Opt_compress_type:
 449                         if (token == Opt_compress ||
 450                             token == Opt_compress_force ||
 451                             strcmp(args[0].from, "zlib") == 0) {
 452                                 compress_type = "zlib";
 453                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 454                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 455                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 456                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 457                         } else if (strcmp(args[0].from, "lzo") == 0) {
 458                                 compress_type = "lzo";
 459                                 info->compress_type = BTRFS_COMPRESS_LZO;
 460                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 461                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 462                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 463                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 464                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 465                                 compress_type = "no";
 466                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 467                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 468                                 compress_force = false;
 469                         } else {
 470                                 ret = -EINVAL;
 471                                 goto out;
 472                         }
 473
 474                         if (compress_force) {
 475                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 476                                 btrfs_info(root->fs_info, "force %s compression",
 477                                         compress_type);
 478                         } else if (btrfs_test_opt(root, COMPRESS)) {
 479                                 pr_info("btrfs: use %s compression\n",
 480                                         compress_type);
 481                         }
 482                         break;
 483                 case Opt_ssd:
 484                         btrfs_info(root->fs_info, "use ssd allocation scheme");
 485                         btrfs_set_opt(info->mount_opt, SSD);
 486                         break;
 487                 case Opt_ssd_spread:
 488                         btrfs_info(root->fs_info, "use spread ssd allocation scheme");
 489                         btrfs_set_opt(info->mount_opt, SSD);
 490                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 491                         break;
 492                 case Opt_nossd:
 493                         btrfs_info(root->fs_info, "not using ssd allocation scheme");
 494                         btrfs_set_opt(info->mount_opt, NOSSD);
 495                         btrfs_clear_opt(info->mount_opt, SSD);
 496                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 497                         break;
 498                 case Opt_barrier:
 499                         if (btrfs_test_opt(root, NOBARRIER))
 500                                 btrfs_info(root->fs_info, "turning on barriers");
 501                         btrfs_clear_opt(info->mount_opt, NOBARRIER);
 502                         break;
 503                 case Opt_nobarrier:
 504                         btrfs_info(root->fs_info, "turning off barriers");
 505                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 506                         break;
 507                 case Opt_thread_pool:
 508                         ret = match_int(&args[0], &intarg);
 509                         if (ret) {
 510                                 goto out;
 511                         } else if (intarg > 0) {
 512                                 info->thread_pool_size = intarg;
 513                         } else {
 514                                 ret = -EINVAL;
 515                                 goto out;
 516                         }
 517                         break;
 518                 case Opt_max_inline:
 519                         num = match_strdup(&args[0]);
 520                         if (num) {
 521                                 info->max_inline = memparse(num, NULL);
 522                                 kfree(num);
 523
 524                                 if (info->max_inline) {
 525                                         info->max_inline = max_t(u64,
 526                                                 info->max_inline,
 527                                                 root->sectorsize);
 528                                 }
 529                                 btrfs_info(root->fs_info, "max_inline at %llu",
 530                                         info->max_inline);
 531                         } else {
 532                                 ret = -ENOMEM;
 533                                 goto out;
 534                         }
 535                         break;
 536                 case Opt_alloc_start:
 537                         num = match_strdup(&args[0]);
 538                         if (num) {
 539                                 mutex_lock(&info->chunk_mutex);
 540                                 info->alloc_start = memparse(num, NULL);
 541                                 mutex_unlock(&info->chunk_mutex);
 542                                 kfree(num);
 543                                 btrfs_info(root->fs_info, "allocations start at %llu",
 544                                         info->alloc_start);
 545                         } else {
 546                                 ret = -ENOMEM;
 547                                 goto out;
 548                         }
 549                         break;
 550                 case Opt_noacl:
 551                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 552                         break;
 553                 case Opt_notreelog:
 554                         btrfs_info(root->fs_info, "disabling tree log");
 555                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 556                         break;
 557                 case Opt_flushoncommit:
 558                         btrfs_info(root->fs_info, "turning on flush-on-commit");
 559                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 560                         break;
 561                 case Opt_ratio:
 562                         ret = match_int(&args[0], &intarg);
 563                         if (ret) {
 564                                 goto out;
 565                         } else if (intarg >= 0) {
 566                                 info->metadata_ratio = intarg;
 567                                 btrfs_info(root->fs_info, "metadata ratio %d",
 568                                        info->metadata_ratio);
 569                         } else {
 570                                 ret = -EINVAL;
 571                                 goto out;
 572                         }
 573                         break;
 574                 case Opt_discard:
 575                         btrfs_set_opt(info->mount_opt, DISCARD);
 576                         break;
 577                 case Opt_space_cache:
 578                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 579                         break;
 580                 case Opt_rescan_uuid_tree:
 581                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 582                         break;
 583                 case Opt_no_space_cache:
 584                         btrfs_info(root->fs_info, "disabling disk space caching");
 585                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 586                         break;
 587                 case Opt_inode_cache:
 588                         btrfs_info(root->fs_info, "enabling inode map caching");
 589                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 590                         break;
 591                 case Opt_clear_cache:
 592                         btrfs_info(root->fs_info, "force clearing of disk cache");
 593                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 594                         break;
 595                 case Opt_user_subvol_rm_allowed:
 596                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 597                         break;
 598                 case Opt_enospc_debug:
 599                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 600                         break;
 601                 case Opt_defrag:
 602                         btrfs_info(root->fs_info, "enabling auto defrag");
 603                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 604                         break;
 605                 case Opt_recovery:
 606                         btrfs_info(root->fs_info, "enabling auto recovery");
 607                         btrfs_set_opt(info->mount_opt, RECOVERY);
 608                         break;
 609                 case Opt_skip_balance:
 610                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 611                         break;
 612 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 613                 case Opt_check_integrity_including_extent_data:
 614                         btrfs_info(root->fs_info,
 615                                    "enabling check integrity including extent data");
 616                         btrfs_set_opt(info->mount_opt,
 617                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 618                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 619                         break;
 620                 case Opt_check_integrity:
 621                         btrfs_info(root->fs_info, "enabling check integrity");
 622                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 623                         break;
 624                 case Opt_check_integrity_print_mask:
 625                         ret = match_int(&args[0], &intarg);
 626                         if (ret) {
 627                                 goto out;
 628                         } else if (intarg >= 0) {
 629                                 info->check_integrity_print_mask = intarg;
 630                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 631                                        info->check_integrity_print_mask);
 632                         } else {
 633                                 ret = -EINVAL;
 634                                 goto out;
 635                         }
 636                         break;
 637 #else
 638                 case Opt_check_integrity_including_extent_data:
 639                 case Opt_check_integrity:
 640                 case Opt_check_integrity_print_mask:
 641                         btrfs_err(root->fs_info,
 642                                 "support for check_integrity* not compiled in!");
 643                         ret = -EINVAL;
 644                         goto out;
 645 #endif
 646                 case Opt_fatal_errors:
 647                         if (strcmp(args[0].from, "panic") == 0)
 648                                 btrfs_set_opt(info->mount_opt,
 649                                               PANIC_ON_FATAL_ERROR);
 650                         else if (strcmp(args[0].from, "bug") == 0)
 651                                 btrfs_clear_opt(info->mount_opt,
 652                                               PANIC_ON_FATAL_ERROR);
 653                         else {
 654                                 ret = -EINVAL;
 655                                 goto out;
 656                         }
 657                         break;
 658                 case Opt_commit_interval:
 659                         intarg = 0;
 660                         ret = match_int(&args[0], &intarg);
 661                         if (ret < 0) {
 662                                 btrfs_err(root->fs_info, "invalid commit interval");
 663                                 ret = -EINVAL;
 664                                 goto out;
 665                         }
 666                         if (intarg > 0) {
 667                                 if (intarg > 300) {
 668                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 669                                                         intarg);
 670                                 }
 671                                 info->commit_interval = intarg;
 672                         } else {
 673                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 674                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 675                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 676                         }
 677                         break;
 678                 case Opt_err:
 679                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 680                         ret = -EINVAL;
 681                         goto out;
 682                 default:
 683                         break;
 684                 }
 685         }
 686 out:
 687         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 688                 btrfs_info(root->fs_info, "disk space caching is enabled");
 689         kfree(orig);
 690         return ret;
 691 }
 692
 693 /*
 694  * Parse mount options that are required early in the mount process.
 695  *
 696  * All other options will be parsed on much later in the mount process and
 697  * only when we need to allocate a new super block.
 698  */
 699 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 700                 void *holder, char **subvol_name, u64 *subvol_objectid,
 701                 struct btrfs_fs_devices **fs_devices)
 702 {
 703         substring_t args[MAX_OPT_ARGS];
 704         char *device_name, *opts, *orig, *p;
 705         char *num = NULL;
 706         int error = 0;
 707
 708         if (!options)
 709                 return 0;
 710
 711         /*
 712          * strsep changes the string, duplicate it because parse_options
 713          * gets called twice
 714          */
 715         opts = kstrdup(options, GFP_KERNEL);
 716         if (!opts)
 717                 return -ENOMEM;
 718         orig = opts;
 719
 720         while ((p = strsep(&opts, ",")) != NULL) {
 721                 int token;
 722                 if (!*p)
 723                         continue;
 724
 725                 token = match_token(p, tokens, args);
 726                 switch (token) {
 727                 case Opt_subvol:
 728                         kfree(*subvol_name);
 729                         *subvol_name = match_strdup(&args[0]);
 730                         if (!*subvol_name) {
 731                                 error = -ENOMEM;
 732                                 goto out;
 733                         }
 734                         break;
 735                 case Opt_subvolid:
 736                         num = match_strdup(&args[0]);
 737                         if (num) {
 738                                 *subvol_objectid = memparse(num, NULL);
 739                                 kfree(num);
 740                                 /* we want the original fs_tree */
 741                                 if (!*subvol_objectid)
 742                                         *subvol_objectid =
 743                                                 BTRFS_FS_TREE_OBJECTID;
 744                         } else {
 745                                 error = -EINVAL;
 746                                 goto out;
 747                         }
 748                         break;
 749                 case Opt_subvolrootid:
 750                         printk(KERN_WARNING
 751                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 752                                 "no effect\n");
 753                         break;
 754                 case Opt_device:
 755                         device_name = match_strdup(&args[0]);
 756                         if (!device_name) {
 757                                 error = -ENOMEM;
 758                                 goto out;
 759                         }
 760                         error = btrfs_scan_one_device(device_name,
 761                                         flags, holder, fs_devices);
 762                         kfree(device_name);
 763                         if (error)
 764                                 goto out;
 765                         break;
 766                 default:
 767                         break;
 768                 }
 769         }
 770
 771 out:
 772         kfree(orig);
 773         return error;
 774 }
 775
 776 static struct dentry *get_default_root(struct super_block *sb,
 777                                        u64 subvol_objectid)
 778 {
 779         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 780         struct btrfs_root *root = fs_info->tree_root;
 781         struct btrfs_root *new_root;
 782         struct btrfs_dir_item *di;
 783         struct btrfs_path *path;
 784         struct btrfs_key location;
 785         struct inode *inode;
 786         u64 dir_id;
 787         int new = 0;
 788
 789         /*
 790          * We have a specific subvol we want to mount, just setup location and
 791          * go look up the root.
 792          */
 793         if (subvol_objectid) {
 794                 location.objectid = subvol_objectid;
 795                 location.type = BTRFS_ROOT_ITEM_KEY;
 796                 location.offset = (u64)-1;
 797                 goto find_root;
 798         }
 799
 800         path = btrfs_alloc_path();
 801         if (!path)
 802                 return ERR_PTR(-ENOMEM);
 803         path->leave_spinning = 1;
 804
 805         /*
 806          * Find the "default" dir item which points to the root item that we
 807          * will mount by default if we haven't been given a specific subvolume
 808          * to mount.
 809          */
 810         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 811         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 812         if (IS_ERR(di)) {
 813                 btrfs_free_path(path);
 814                 return ERR_CAST(di);
 815         }
 816         if (!di) {
 817                 /*
 818                  * Ok the default dir item isn't there.  This is weird since
 819                  * it's always been there, but don't freak out, just try and
 820                  * mount to root most subvolume.
 821                  */
 822                 btrfs_free_path(path);
 823                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 824                 new_root = fs_info->fs_root;
 825                 goto setup_root;
 826         }
 827
 828         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 829         btrfs_free_path(path);
 830
 831 find_root:
 832         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 833         if (IS_ERR(new_root))
 834                 return ERR_CAST(new_root);
 835
 836         dir_id = btrfs_root_dirid(&new_root->root_item);
 837 setup_root:
 838         location.objectid = dir_id;
 839         location.type = BTRFS_INODE_ITEM_KEY;
 840         location.offset = 0;
 841
 842         inode = btrfs_iget(sb, &location, new_root, &new);
 843         if (IS_ERR(inode))
 844                 return ERR_CAST(inode);
 845
 846         /*
 847          * If we're just mounting the root most subvol put the inode and return
 848          * a reference to the dentry.  We will have already gotten a reference
 849          * to the inode in btrfs_fill_super so we're good to go.
 850          */
 851         if (!new && sb->s_root->d_inode == inode) {
 852                 iput(inode);
 853                 return dget(sb->s_root);
 854         }
 855
 856         return d_obtain_alias(inode);
 857 }
 858
 859 static int btrfs_fill_super(struct super_block *sb,
 860                             struct btrfs_fs_devices *fs_devices,
 861                             void *data, int silent)
 862 {
 863         struct inode *inode;
 864         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 865         struct btrfs_key key;
 866         int err;
 867
 868         sb->s_maxbytes = MAX_LFS_FILESIZE;
 869         sb->s_magic = BTRFS_SUPER_MAGIC;
 870         sb->s_op = &btrfs_super_ops;
 871         sb->s_d_op = &btrfs_dentry_operations;
 872         sb->s_export_op = &btrfs_export_ops;
 873         sb->s_xattr = btrfs_xattr_handlers;
 874         sb->s_time_gran = 1;
 875 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 876         sb->s_flags |= MS_POSIXACL;
 877 #endif
 878         sb->s_flags |= MS_I_VERSION;
 879         err = open_ctree(sb, fs_devices, (char *)data);
 880         if (err) {
 881                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
 882                 return err;
 883         }
 884
 885         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 886         key.type = BTRFS_INODE_ITEM_KEY;
 887         key.offset = 0;
 888         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 889         if (IS_ERR(inode)) {
 890                 err = PTR_ERR(inode);
 891                 goto fail_close;
 892         }
 893
 894         sb->s_root = d_make_root(inode);
 895         if (!sb->s_root) {
 896                 err = -ENOMEM;
 897                 goto fail_close;
 898         }
 899
 900         save_mount_options(sb, data);
 901         cleancache_init_fs(sb);
 902         sb->s_flags |= MS_ACTIVE;
 903         return 0;
 904
 905 fail_close:
 906         close_ctree(fs_info->tree_root);
 907         return err;
 908 }
 909
 910 int btrfs_sync_fs(struct super_block *sb, int wait)
 911 {
 912         struct btrfs_trans_handle *trans;
 913         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 914         struct btrfs_root *root = fs_info->tree_root;
 915
 916         trace_btrfs_sync_fs(wait);
 917
 918         if (!wait) {
 919                 filemap_flush(fs_info->btree_inode->i_mapping);
 920                 return 0;
 921         }
 922
 923         btrfs_wait_ordered_roots(fs_info, -1);
 924
 925         trans = btrfs_attach_transaction_barrier(root);
 926         if (IS_ERR(trans)) {
 927                 /* no transaction, don't bother */
 928                 if (PTR_ERR(trans) == -ENOENT)
 929                         return 0;
 930                 return PTR_ERR(trans);
 931         }
 932         return btrfs_commit_transaction(trans, root);
 933 }
 934
 935 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 936 {
 937         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 938         struct btrfs_root *root = info->tree_root;
 939         char *compress_type;
 940
 941         if (btrfs_test_opt(root, DEGRADED))
 942                 seq_puts(seq, ",degraded");
 943         if (btrfs_test_opt(root, NODATASUM))
 944                 seq_puts(seq, ",nodatasum");
 945         if (btrfs_test_opt(root, NODATACOW))
 946                 seq_puts(seq, ",nodatacow");
 947         if (btrfs_test_opt(root, NOBARRIER))
 948                 seq_puts(seq, ",nobarrier");
 949         if (info->max_inline != 8192 * 1024)
 950                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
 951         if (info->alloc_start != 0)
 952                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
 953         if (info->thread_pool_size !=  min_t(unsigned long,
 954                                              num_online_cpus() + 2, 8))
 955                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 956         if (btrfs_test_opt(root, COMPRESS)) {
 957                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 958                         compress_type = "zlib";
 959                 else
 960                         compress_type = "lzo";
 961                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 962                         seq_printf(seq, ",compress-force=%s", compress_type);
 963                 else
 964                         seq_printf(seq, ",compress=%s", compress_type);
 965         }
 966         if (btrfs_test_opt(root, NOSSD))
 967                 seq_puts(seq, ",nossd");
 968         if (btrfs_test_opt(root, SSD_SPREAD))
 969                 seq_puts(seq, ",ssd_spread");
 970         else if (btrfs_test_opt(root, SSD))
 971                 seq_puts(seq, ",ssd");
 972         if (btrfs_test_opt(root, NOTREELOG))
 973                 seq_puts(seq, ",notreelog");
 974         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 975                 seq_puts(seq, ",flushoncommit");
 976         if (btrfs_test_opt(root, DISCARD))
 977                 seq_puts(seq, ",discard");
 978         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 979                 seq_puts(seq, ",noacl");
 980         if (btrfs_test_opt(root, SPACE_CACHE))
 981                 seq_puts(seq, ",space_cache");
 982         else
 983                 seq_puts(seq, ",nospace_cache");
 984         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
 985                 seq_puts(seq, ",rescan_uuid_tree");
 986         if (btrfs_test_opt(root, CLEAR_CACHE))
 987                 seq_puts(seq, ",clear_cache");
 988         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 989                 seq_puts(seq, ",user_subvol_rm_allowed");
 990         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 991                 seq_puts(seq, ",enospc_debug");
 992         if (btrfs_test_opt(root, AUTO_DEFRAG))
 993                 seq_puts(seq, ",autodefrag");
 994         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 995                 seq_puts(seq, ",inode_cache");
 996         if (btrfs_test_opt(root, SKIP_BALANCE))
 997                 seq_puts(seq, ",skip_balance");
 998         if (btrfs_test_opt(root, RECOVERY))
 999                 seq_puts(seq, ",recovery");
1000 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1001         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1002                 seq_puts(seq, ",check_int_data");
1003         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1004                 seq_puts(seq, ",check_int");
1005         if (info->check_integrity_print_mask)
1006                 seq_printf(seq, ",check_int_print_mask=%d",
1007                                 info->check_integrity_print_mask);
1008 #endif
1009         if (info->metadata_ratio)
1010                 seq_printf(seq, ",metadata_ratio=%d",
1011                                 info->metadata_ratio);
1012         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1013                 seq_puts(seq, ",fatal_errors=panic");
1014         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1015                 seq_printf(seq, ",commit=%d", info->commit_interval);
1016         return 0;
1017 }
1018
1019 static int btrfs_test_super(struct super_block *s, void *data)
1020 {
1021         struct btrfs_fs_info *p = data;
1022         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1023
1024         return fs_info->fs_devices == p->fs_devices;
1025 }
1026
1027 static int btrfs_set_super(struct super_block *s, void *data)
1028 {
1029         int err = set_anon_super(s, data);
1030         if (!err)
1031                 s->s_fs_info = data;
1032         return err;
1033 }
1034
1035 /*
1036  * subvolumes are identified by ino 256
1037  */
1038 static inline int is_subvolume_inode(struct inode *inode)
1039 {
1040         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1041                 return 1;
1042         return 0;
1043 }
1044
1045 /*
1046  * This will strip out the subvol=%s argument for an argument string and add
1047  * subvolid=0 to make sure we get the actual tree root for path walking to the
1048  * subvol we want.
1049  */
1050 static char *setup_root_args(char *args)
1051 {
1052         unsigned len = strlen(args) + 2 + 1;
1053         char *src, *dst, *buf;
1054
1055         /*
1056          * We need the same args as before, but with this substitution:
1057          * s!subvol=[^,]+!subvolid=0!
1058          *
1059          * Since the replacement string is up to 2 bytes longer than the
1060          * original, allocate strlen(args) + 2 + 1 bytes.
1061          */
1062
1063         src = strstr(args, "subvol=");
1064         /* This shouldn't happen, but just in case.. */
1065         if (!src)
1066                 return NULL;
1067
1068         buf = dst = kmalloc(len, GFP_NOFS);
1069         if (!buf)
1070                 return NULL;
1071
1072         /*
1073          * If the subvol= arg is not at the start of the string,
1074          * copy whatever precedes it into buf.
1075          */
1076         if (src != args) {
1077                 *src++ = '\0';
1078                 strcpy(buf, args);
1079                 dst += strlen(args);
1080         }
1081
1082         strcpy(dst, "subvolid=0");
1083         dst += strlen("subvolid=0");
1084
1085         /*
1086          * If there is a "," after the original subvol=... string,
1087          * copy that suffix into our buffer.  Otherwise, we're done.
1088          */
1089         src = strchr(src, ',');
1090         if (src)
1091                 strcpy(dst, src);
1092
1093         return buf;
1094 }
1095
1096 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1097                                    const char *device_name, char *data)
1098 {
1099         struct dentry *root;
1100         struct vfsmount *mnt;
1101         char *newargs;
1102
1103         newargs = setup_root_args(data);
1104         if (!newargs)
1105                 return ERR_PTR(-ENOMEM);
1106         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1107                              newargs);
1108         kfree(newargs);
1109         if (IS_ERR(mnt))
1110                 return ERR_CAST(mnt);
1111
1112         root = mount_subtree(mnt, subvol_name);
1113
1114         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1115                 struct super_block *s = root->d_sb;
1116                 dput(root);
1117                 root = ERR_PTR(-EINVAL);
1118                 deactivate_locked_super(s);
1119                 printk(KERN_ERR "BTRFS: '%s' is not a valid subvolume\n",
1120                                 subvol_name);
1121         }
1122
1123         return root;
1124 }
1125
1126 /*
1127  * Find a superblock for the given device / mount point.
1128  *
1129  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1130  *        for multiple device setup.  Make sure to keep it in sync.
1131  */
1132 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1133                 const char *device_name, void *data)
1134 {
1135         struct block_device *bdev = NULL;
1136         struct super_block *s;
1137         struct dentry *root;
1138         struct btrfs_fs_devices *fs_devices = NULL;
1139         struct btrfs_fs_info *fs_info = NULL;
1140         fmode_t mode = FMODE_READ;
1141         char *subvol_name = NULL;
1142         u64 subvol_objectid = 0;
1143         int error = 0;
1144
1145         if (!(flags & MS_RDONLY))
1146                 mode |= FMODE_WRITE;
1147
1148         error = btrfs_parse_early_options(data, mode, fs_type,
1149                                           &subvol_name, &subvol_objectid,
1150                                           &fs_devices);
1151         if (error) {
1152                 kfree(subvol_name);
1153                 return ERR_PTR(error);
1154         }
1155
1156         if (subvol_name) {
1157                 root = mount_subvol(subvol_name, flags, device_name, data);
1158                 kfree(subvol_name);
1159                 return root;
1160         }
1161
1162         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1163         if (error)
1164                 return ERR_PTR(error);
1165
1166         /*
1167          * Setup a dummy root and fs_info for test/set super.  This is because
1168          * we don't actually fill this stuff out until open_ctree, but we need
1169          * it for searching for existing supers, so this lets us do that and
1170          * then open_ctree will properly initialize everything later.
1171          */
1172         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1173         if (!fs_info)
1174                 return ERR_PTR(-ENOMEM);
1175
1176         fs_info->fs_devices = fs_devices;
1177
1178         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1179         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1180         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1181                 error = -ENOMEM;
1182                 goto error_fs_info;
1183         }
1184
1185         error = btrfs_open_devices(fs_devices, mode, fs_type);
1186         if (error)
1187                 goto error_fs_info;
1188
1189         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1190                 error = -EACCES;
1191                 goto error_close_devices;
1192         }
1193
1194         bdev = fs_devices->latest_bdev;
1195         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1196                  fs_info);
1197         if (IS_ERR(s)) {
1198                 error = PTR_ERR(s);
1199                 goto error_close_devices;
1200         }
1201
1202         if (s->s_root) {
1203                 btrfs_close_devices(fs_devices);
1204                 free_fs_info(fs_info);
1205                 if ((flags ^ s->s_flags) & MS_RDONLY)
1206                         error = -EBUSY;
1207         } else {
1208                 char b[BDEVNAME_SIZE];
1209
1210                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1211                 btrfs_sb(s)->bdev_holder = fs_type;
1212                 error = btrfs_fill_super(s, fs_devices, data,
1213                                          flags & MS_SILENT ? 1 : 0);
1214         }
1215
1216         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1217         if (IS_ERR(root))
1218                 deactivate_locked_super(s);
1219
1220         return root;
1221
1222 error_close_devices:
1223         btrfs_close_devices(fs_devices);
1224 error_fs_info:
1225         free_fs_info(fs_info);
1226         return ERR_PTR(error);
1227 }
1228
1229 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1230 {
1231         spin_lock_irq(&workers->lock);
1232         workers->max_workers = new_limit;
1233         spin_unlock_irq(&workers->lock);
1234 }
1235
1236 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1237                                      int new_pool_size, int old_pool_size)
1238 {
1239         if (new_pool_size == old_pool_size)
1240                 return;
1241
1242         fs_info->thread_pool_size = new_pool_size;
1243
1244         btrfs_info(fs_info, "resize thread pool %d -> %d",
1245                old_pool_size, new_pool_size);
1246
1247         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1248         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1249         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1250         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1251         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1252         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1253         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1254         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1255         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1256         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1257         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1258         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1259         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1260         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1261                               new_pool_size);
1262 }
1263
1264 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1265 {
1266         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1267 }
1268
1269 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1270                                        unsigned long old_opts, int flags)
1271 {
1272         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1273             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1274              (flags & MS_RDONLY))) {
1275                 /* wait for any defraggers to finish */
1276                 wait_event(fs_info->transaction_wait,
1277                            (atomic_read(&fs_info->defrag_running) == 0));
1278                 if (flags & MS_RDONLY)
1279                         sync_filesystem(fs_info->sb);
1280         }
1281 }
1282
1283 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1284                                          unsigned long old_opts)
1285 {
1286         /*
1287          * We need cleanup all defragable inodes if the autodefragment is
1288          * close or the fs is R/O.
1289          */
1290         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1291             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1292              (fs_info->sb->s_flags & MS_RDONLY))) {
1293                 btrfs_cleanup_defrag_inodes(fs_info);
1294         }
1295
1296         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1297 }
1298
1299 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1300 {
1301         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1302         struct btrfs_root *root = fs_info->tree_root;
1303         unsigned old_flags = sb->s_flags;
1304         unsigned long old_opts = fs_info->mount_opt;
1305         unsigned long old_compress_type = fs_info->compress_type;
1306         u64 old_max_inline = fs_info->max_inline;
1307         u64 old_alloc_start = fs_info->alloc_start;
1308         int old_thread_pool_size = fs_info->thread_pool_size;
1309         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1310         int ret;
1311
1312         btrfs_remount_prepare(fs_info);
1313
1314         ret = btrfs_parse_options(root, data);
1315         if (ret) {
1316                 ret = -EINVAL;
1317                 goto restore;
1318         }
1319
1320         btrfs_remount_begin(fs_info, old_opts, *flags);
1321         btrfs_resize_thread_pool(fs_info,
1322                 fs_info->thread_pool_size, old_thread_pool_size);
1323
1324         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1325                 goto out;
1326
1327         if (*flags & MS_RDONLY) {
1328                 /*
1329                  * this also happens on 'umount -rf' or on shutdown, when
1330                  * the filesystem is busy.
1331                  */
1332
1333                 /* wait for the uuid_scan task to finish */
1334                 down(&fs_info->uuid_tree_rescan_sem);
1335                 /* avoid complains from lockdep et al. */
1336                 up(&fs_info->uuid_tree_rescan_sem);
1337
1338                 sb->s_flags |= MS_RDONLY;
1339
1340                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1341                 btrfs_scrub_cancel(fs_info);
1342                 btrfs_pause_balance(fs_info);
1343
1344                 ret = btrfs_commit_super(root);
1345                 if (ret)
1346                         goto restore;
1347         } else {
1348                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1349                         btrfs_err(fs_info,
1350                                 "Remounting read-write after error is not allowed");
1351                         ret = -EINVAL;
1352                         goto restore;
1353                 }
1354                 if (fs_info->fs_devices->rw_devices == 0) {
1355                         ret = -EACCES;
1356                         goto restore;
1357                 }
1358
1359                 if (fs_info->fs_devices->missing_devices >
1360                      fs_info->num_tolerated_disk_barrier_failures &&
1361                     !(*flags & MS_RDONLY)) {
1362                         btrfs_warn(fs_info,
1363                                 "too many missing devices, writeable remount is not allowed");
1364                         ret = -EACCES;
1365                         goto restore;
1366                 }
1367
1368                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1369                         ret = -EINVAL;
1370                         goto restore;
1371                 }
1372
1373                 ret = btrfs_cleanup_fs_roots(fs_info);
1374                 if (ret)
1375                         goto restore;
1376
1377                 /* recover relocation */
1378                 ret = btrfs_recover_relocation(root);
1379                 if (ret)
1380                         goto restore;
1381
1382                 ret = btrfs_resume_balance_async(fs_info);
1383                 if (ret)
1384                         goto restore;
1385
1386                 ret = btrfs_resume_dev_replace_async(fs_info);
1387                 if (ret) {
1388                         btrfs_warn(fs_info, "failed to resume dev_replace");
1389                         goto restore;
1390                 }
1391
1392                 if (!fs_info->uuid_root) {
1393                         btrfs_info(fs_info, "creating UUID tree");
1394                         ret = btrfs_create_uuid_tree(fs_info);
1395                         if (ret) {
1396                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1397                                 goto restore;
1398                         }
1399                 }
1400                 sb->s_flags &= ~MS_RDONLY;
1401         }
1402 out:
1403         btrfs_remount_cleanup(fs_info, old_opts);
1404         return 0;
1405
1406 restore:
1407         /* We've hit an error - don't reset MS_RDONLY */
1408         if (sb->s_flags & MS_RDONLY)
1409                 old_flags |= MS_RDONLY;
1410         sb->s_flags = old_flags;
1411         fs_info->mount_opt = old_opts;
1412         fs_info->compress_type = old_compress_type;
1413         fs_info->max_inline = old_max_inline;
1414         mutex_lock(&fs_info->chunk_mutex);
1415         fs_info->alloc_start = old_alloc_start;
1416         mutex_unlock(&fs_info->chunk_mutex);
1417         btrfs_resize_thread_pool(fs_info,
1418                 old_thread_pool_size, fs_info->thread_pool_size);
1419         fs_info->metadata_ratio = old_metadata_ratio;
1420         btrfs_remount_cleanup(fs_info, old_opts);
1421         return ret;
1422 }
1423
1424 /* Used to sort the devices by max_avail(descending sort) */
1425 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1426                                        const void *dev_info2)
1427 {
1428         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1429             ((struct btrfs_device_info *)dev_info2)->max_avail)
1430                 return -1;
1431         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1432                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1433                 return 1;
1434         else
1435         return 0;
1436 }
1437
1438 /*
1439  * sort the devices by max_avail, in which max free extent size of each device
1440  * is stored.(Descending Sort)
1441  */
1442 static inline void btrfs_descending_sort_devices(
1443                                         struct btrfs_device_info *devices,
1444                                         size_t nr_devices)
1445 {
1446         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1447              btrfs_cmp_device_free_bytes, NULL);
1448 }
1449
1450 /*
1451  * The helper to calc the free space on the devices that can be used to store
1452  * file data.
1453  */
1454 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1455 {
1456         struct btrfs_fs_info *fs_info = root->fs_info;
1457         struct btrfs_device_info *devices_info;
1458         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1459         struct btrfs_device *device;
1460         u64 skip_space;
1461         u64 type;
1462         u64 avail_space;
1463         u64 used_space;
1464         u64 min_stripe_size;
1465         int min_stripes = 1, num_stripes = 1;
1466         int i = 0, nr_devices;
1467         int ret;
1468
1469         nr_devices = fs_info->fs_devices->open_devices;
1470         BUG_ON(!nr_devices);
1471
1472         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1473                                GFP_NOFS);
1474         if (!devices_info)
1475                 return -ENOMEM;
1476
1477         /* calc min stripe number for data space alloction */
1478         type = btrfs_get_alloc_profile(root, 1);
1479         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1480                 min_stripes = 2;
1481                 num_stripes = nr_devices;
1482         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1483                 min_stripes = 2;
1484                 num_stripes = 2;
1485         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1486                 min_stripes = 4;
1487                 num_stripes = 4;
1488         }
1489
1490         if (type & BTRFS_BLOCK_GROUP_DUP)
1491                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1492         else
1493                 min_stripe_size = BTRFS_STRIPE_LEN;
1494
1495         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1496                 if (!device->in_fs_metadata || !device->bdev ||
1497                     device->is_tgtdev_for_dev_replace)
1498                         continue;
1499
1500                 avail_space = device->total_bytes - device->bytes_used;
1501
1502                 /* align with stripe_len */
1503                 do_div(avail_space, BTRFS_STRIPE_LEN);
1504                 avail_space *= BTRFS_STRIPE_LEN;
1505
1506                 /*
1507                  * In order to avoid overwritting the superblock on the drive,
1508                  * btrfs starts at an offset of at least 1MB when doing chunk
1509                  * allocation.
1510                  */
1511                 skip_space = 1024 * 1024;
1512
1513                 /* user can set the offset in fs_info->alloc_start. */
1514                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1515                     device->total_bytes)
1516                         skip_space = max(fs_info->alloc_start, skip_space);
1517
1518                 /*
1519                  * btrfs can not use the free space in [0, skip_space - 1],
1520                  * we must subtract it from the total. In order to implement
1521                  * it, we account the used space in this range first.
1522                  */
1523                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1524                                                      &used_space);
1525                 if (ret) {
1526                         kfree(devices_info);
1527                         return ret;
1528                 }
1529
1530                 /* calc the free space in [0, skip_space - 1] */
1531                 skip_space -= used_space;
1532
1533                 /*
1534                  * we can use the free space in [0, skip_space - 1], subtract
1535                  * it from the total.
1536                  */
1537                 if (avail_space && avail_space >= skip_space)
1538                         avail_space -= skip_space;
1539                 else
1540                         avail_space = 0;
1541
1542                 if (avail_space < min_stripe_size)
1543                         continue;
1544
1545                 devices_info[i].dev = device;
1546                 devices_info[i].max_avail = avail_space;
1547
1548                 i++;
1549         }
1550
1551         nr_devices = i;
1552
1553         btrfs_descending_sort_devices(devices_info, nr_devices);
1554
1555         i = nr_devices - 1;
1556         avail_space = 0;
1557         while (nr_devices >= min_stripes) {
1558                 if (num_stripes > nr_devices)
1559                         num_stripes = nr_devices;
1560
1561                 if (devices_info[i].max_avail >= min_stripe_size) {
1562                         int j;
1563                         u64 alloc_size;
1564
1565                         avail_space += devices_info[i].max_avail * num_stripes;
1566                         alloc_size = devices_info[i].max_avail;
1567                         for (j = i + 1 - num_stripes; j <= i; j++)
1568                                 devices_info[j].max_avail -= alloc_size;
1569                 }
1570                 i--;
1571                 nr_devices--;
1572         }
1573
1574         kfree(devices_info);
1575         *free_bytes = avail_space;
1576         return 0;
1577 }
1578
1579 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1580 {
1581         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1582         struct btrfs_super_block *disk_super = fs_info->super_copy;
1583         struct list_head *head = &fs_info->space_info;
1584         struct btrfs_space_info *found;
1585         u64 total_used = 0;
1586         u64 total_free_data = 0;
1587         int bits = dentry->d_sb->s_blocksize_bits;
1588         __be32 *fsid = (__be32 *)fs_info->fsid;
1589         int ret;
1590
1591         /* holding chunk_muext to avoid allocating new chunks */
1592         mutex_lock(&fs_info->chunk_mutex);
1593         rcu_read_lock();
1594         list_for_each_entry_rcu(found, head, list) {
1595                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1596                         total_free_data += found->disk_total - found->disk_used;
1597                         total_free_data -=
1598                                 btrfs_account_ro_block_groups_free_space(found);
1599                 }
1600
1601                 total_used += found->disk_used;
1602         }
1603         rcu_read_unlock();
1604
1605         buf->f_namelen = BTRFS_NAME_LEN;
1606         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1607         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1608         buf->f_bsize = dentry->d_sb->s_blocksize;
1609         buf->f_type = BTRFS_SUPER_MAGIC;
1610         buf->f_bavail = total_free_data;
1611         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1612         if (ret) {
1613                 mutex_unlock(&fs_info->chunk_mutex);
1614                 return ret;
1615         }
1616         buf->f_bavail += total_free_data;
1617         buf->f_bavail = buf->f_bavail >> bits;
1618         mutex_unlock(&fs_info->chunk_mutex);
1619
1620         /* We treat it as constant endianness (it doesn't matter _which_)
1621            because we want the fsid to come out the same whether mounted
1622            on a big-endian or little-endian host */
1623         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1624         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1625         /* Mask in the root object ID too, to disambiguate subvols */
1626         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1627         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1628
1629         return 0;
1630 }
1631
1632 static void btrfs_kill_super(struct super_block *sb)
1633 {
1634         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1635         kill_anon_super(sb);
1636         free_fs_info(fs_info);
1637 }
1638
1639 static struct file_system_type btrfs_fs_type = {
1640         .owner          = THIS_MODULE,
1641         .name           = "btrfs",
1642         .mount          = btrfs_mount,
1643         .kill_sb        = btrfs_kill_super,
1644         .fs_flags       = FS_REQUIRES_DEV,
1645 };
1646 MODULE_ALIAS_FS("btrfs");
1647
1648 /*
1649  * used by btrfsctl to scan devices when no FS is mounted
1650  */
1651 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1652                                 unsigned long arg)
1653 {
1654         struct btrfs_ioctl_vol_args *vol;
1655         struct btrfs_fs_devices *fs_devices;
1656         int ret = -ENOTTY;
1657
1658         if (!capable(CAP_SYS_ADMIN))
1659                 return -EPERM;
1660
1661         vol = memdup_user((void __user *)arg, sizeof(*vol));
1662         if (IS_ERR(vol))
1663                 return PTR_ERR(vol);
1664
1665         switch (cmd) {
1666         case BTRFS_IOC_SCAN_DEV:
1667                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1668                                             &btrfs_fs_type, &fs_devices);
1669                 break;
1670         case BTRFS_IOC_DEVICES_READY:
1671                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1672                                             &btrfs_fs_type, &fs_devices);
1673                 if (ret)
1674                         break;
1675                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1676                 break;
1677         }
1678
1679         kfree(vol);
1680         return ret;
1681 }
1682
1683 static int btrfs_freeze(struct super_block *sb)
1684 {
1685         struct btrfs_trans_handle *trans;
1686         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1687
1688         trans = btrfs_attach_transaction_barrier(root);
1689         if (IS_ERR(trans)) {
1690                 /* no transaction, don't bother */
1691                 if (PTR_ERR(trans) == -ENOENT)
1692                         return 0;
1693                 return PTR_ERR(trans);
1694         }
1695         return btrfs_commit_transaction(trans, root);
1696 }
1697
1698 static int btrfs_unfreeze(struct super_block *sb)
1699 {
1700         return 0;
1701 }
1702
1703 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1704 {
1705         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1706         struct btrfs_fs_devices *cur_devices;
1707         struct btrfs_device *dev, *first_dev = NULL;
1708         struct list_head *head;
1709         struct rcu_string *name;
1710
1711         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1712         cur_devices = fs_info->fs_devices;
1713         while (cur_devices) {
1714                 head = &cur_devices->devices;
1715                 list_for_each_entry(dev, head, dev_list) {
1716                         if (dev->missing)
1717                                 continue;
1718                         if (!first_dev || dev->devid < first_dev->devid)
1719                                 first_dev = dev;
1720                 }
1721                 cur_devices = cur_devices->seed;
1722         }
1723
1724         if (first_dev) {
1725                 rcu_read_lock();
1726                 name = rcu_dereference(first_dev->name);
1727                 seq_escape(m, name->str, " \t\n\\");
1728                 rcu_read_unlock();
1729         } else {
1730                 WARN_ON(1);
1731         }
1732         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1733         return 0;
1734 }
1735
1736 static const struct super_operations btrfs_super_ops = {
1737         .drop_inode     = btrfs_drop_inode,
1738         .evict_inode    = btrfs_evict_inode,
1739         .put_super      = btrfs_put_super,
1740         .sync_fs        = btrfs_sync_fs,
1741         .show_options   = btrfs_show_options,
1742         .show_devname   = btrfs_show_devname,
1743         .write_inode    = btrfs_write_inode,
1744         .alloc_inode    = btrfs_alloc_inode,
1745         .destroy_inode  = btrfs_destroy_inode,
1746         .statfs         = btrfs_statfs,
1747         .remount_fs     = btrfs_remount,
1748         .freeze_fs      = btrfs_freeze,
1749         .unfreeze_fs    = btrfs_unfreeze,
1750 };
1751
1752 static const struct file_operations btrfs_ctl_fops = {
1753         .unlocked_ioctl  = btrfs_control_ioctl,
1754         .compat_ioctl = btrfs_control_ioctl,
1755         .owner   = THIS_MODULE,
1756         .llseek = noop_llseek,
1757 };
1758
1759 static struct miscdevice btrfs_misc = {
1760         .minor          = BTRFS_MINOR,
1761         .name           = "btrfs-control",
1762         .fops           = &btrfs_ctl_fops
1763 };
1764
1765 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1766 MODULE_ALIAS("devname:btrfs-control");
1767
1768 static int btrfs_interface_init(void)
1769 {
1770         return misc_register(&btrfs_misc);
1771 }
1772
1773 static void btrfs_interface_exit(void)
1774 {
1775         if (misc_deregister(&btrfs_misc) < 0)
1776                 printk(KERN_INFO "BTRFS: misc_deregister failed for control device\n");
1777 }
1778
1779 static void btrfs_print_info(void)
1780 {
1781         printk(KERN_INFO "Btrfs loaded"
1782 #ifdef CONFIG_BTRFS_DEBUG
1783                         ", debug=on"
1784 #endif
1785 #ifdef CONFIG_BTRFS_ASSERT
1786                         ", assert=on"
1787 #endif
1788 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1789                         ", integrity-checker=on"
1790 #endif
1791                         "\n");
1792 }
1793
1794 static int btrfs_run_sanity_tests(void)
1795 {
1796         int ret;
1797
1798         ret = btrfs_init_test_fs();
1799         if (ret)
1800                 return ret;
1801
1802         ret = btrfs_test_free_space_cache();
1803         if (ret)
1804                 goto out;
1805         ret = btrfs_test_extent_buffer_operations();
1806         if (ret)
1807                 goto out;
1808         ret = btrfs_test_extent_io();
1809         if (ret)
1810                 goto out;
1811         ret = btrfs_test_inodes();
1812 out:
1813         btrfs_destroy_test_fs();
1814         return ret;
1815 }
1816
1817 static int __init init_btrfs_fs(void)
1818 {
1819         int err;
1820
1821         err = btrfs_init_sysfs();
1822         if (err)
1823                 return err;
1824
1825         btrfs_init_compress();
1826
1827         err = btrfs_init_cachep();
1828         if (err)
1829                 goto free_compress;
1830
1831         err = extent_io_init();
1832         if (err)
1833                 goto free_cachep;
1834
1835         err = extent_map_init();
1836         if (err)
1837                 goto free_extent_io;
1838
1839         err = ordered_data_init();
1840         if (err)
1841                 goto free_extent_map;
1842
1843         err = btrfs_delayed_inode_init();
1844         if (err)
1845                 goto free_ordered_data;
1846
1847         err = btrfs_auto_defrag_init();
1848         if (err)
1849                 goto free_delayed_inode;
1850
1851         err = btrfs_delayed_ref_init();
1852         if (err)
1853                 goto free_auto_defrag;
1854
1855         err = btrfs_prelim_ref_init();
1856         if (err)
1857                 goto free_prelim_ref;
1858
1859         err = btrfs_interface_init();
1860         if (err)
1861                 goto free_delayed_ref;
1862
1863         btrfs_init_lockdep();
1864
1865         btrfs_print_info();
1866
1867         err = btrfs_run_sanity_tests();
1868         if (err)
1869                 goto unregister_ioctl;
1870
1871         err = register_filesystem(&btrfs_fs_type);
1872         if (err)
1873                 goto unregister_ioctl;
1874
1875         return 0;
1876
1877 unregister_ioctl:
1878         btrfs_interface_exit();
1879 free_prelim_ref:
1880         btrfs_prelim_ref_exit();
1881 free_delayed_ref:
1882         btrfs_delayed_ref_exit();
1883 free_auto_defrag:
1884         btrfs_auto_defrag_exit();
1885 free_delayed_inode:
1886         btrfs_delayed_inode_exit();
1887 free_ordered_data:
1888         ordered_data_exit();
1889 free_extent_map:
1890         extent_map_exit();
1891 free_extent_io:
1892         extent_io_exit();
1893 free_cachep:
1894         btrfs_destroy_cachep();
1895 free_compress:
1896         btrfs_exit_compress();
1897         btrfs_exit_sysfs();
1898         return err;
1899 }
1900
1901 static void __exit exit_btrfs_fs(void)
1902 {
1903         btrfs_destroy_cachep();
1904         btrfs_delayed_ref_exit();
1905         btrfs_auto_defrag_exit();
1906         btrfs_delayed_inode_exit();
1907         btrfs_prelim_ref_exit();
1908         ordered_data_exit();
1909         extent_map_exit();
1910         extent_io_exit();
1911         btrfs_interface_exit();
1912         unregister_filesystem(&btrfs_fs_type);
1913         btrfs_exit_sysfs();
1914         btrfs_cleanup_fs_uuids();
1915         btrfs_exit_compress();
1916 }
1917
1918 module_init(init_btrfs_fs)
1919 module_exit(exit_btrfs_fs)
1920
1921 MODULE_LICENSE("GPL");