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, Opt_nodefrag, Opt_nodiscard,
 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_nodiscard, "nodiscard"},
 355         {Opt_space_cache, "space_cache"},
 356         {Opt_clear_cache, "clear_cache"},
 357         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 358         {Opt_enospc_debug, "enospc_debug"},
 359         {Opt_subvolrootid, "subvolrootid=%d"},
 360         {Opt_defrag, "autodefrag"},
 361         {Opt_nodefrag, "noautodefrag"},
 362         {Opt_inode_cache, "inode_cache"},
 363         {Opt_no_space_cache, "nospace_cache"},
 364         {Opt_recovery, "recovery"},
 365         {Opt_skip_balance, "skip_balance"},
 366         {Opt_check_integrity, "check_int"},
 367         {Opt_check_integrity_including_extent_data, "check_int_data"},
 368         {Opt_check_integrity_print_mask, "check_int_print_mask=%d"},
 369         {Opt_rescan_uuid_tree, "rescan_uuid_tree"},
 370         {Opt_fatal_errors, "fatal_errors=%s"},
 371         {Opt_commit_interval, "commit=%d"},
 372         {Opt_err, NULL},
 373 };
 374
 375 /*
 376  * Regular mount options parser.  Everything that is needed only when
 377  * reading in a new superblock is parsed here.
 378  * XXX JDM: This needs to be cleaned up for remount.
 379  */
 380 int btrfs_parse_options(struct btrfs_root *root, char *options)
 381 {
 382         struct btrfs_fs_info *info = root->fs_info;
 383         substring_t args[MAX_OPT_ARGS];
 384         char *p, *num, *orig = NULL;
 385         u64 cache_gen;
 386         int intarg;
 387         int ret = 0;
 388         char *compress_type;
 389         bool compress_force = false;
 390
 391         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 392         if (cache_gen)
 393                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 394
 395         if (!options)
 396                 goto out;
 397
 398         /*
 399          * strsep changes the string, duplicate it because parse_options
 400          * gets called twice
 401          */
 402         options = kstrdup(options, GFP_NOFS);
 403         if (!options)
 404                 return -ENOMEM;
 405
 406         orig = options;
 407
 408         while ((p = strsep(&options, ",")) != NULL) {
 409                 int token;
 410                 if (!*p)
 411                         continue;
 412
 413                 token = match_token(p, tokens, args);
 414                 switch (token) {
 415                 case Opt_degraded:
 416                         btrfs_info(root->fs_info, "allowing degraded mounts");
 417                         btrfs_set_opt(info->mount_opt, DEGRADED);
 418                         break;
 419                 case Opt_subvol:
 420                 case Opt_subvolid:
 421                 case Opt_subvolrootid:
 422                 case Opt_device:
 423                         /*
 424                          * These are parsed by btrfs_parse_early_options
 425                          * and can be happily ignored here.
 426                          */
 427                         break;
 428                 case Opt_nodatasum:
 429                         btrfs_info(root->fs_info, "setting nodatasum");
 430                         btrfs_set_opt(info->mount_opt, NODATASUM);
 431                         break;
 432                 case Opt_nodatacow:
 433                         if (!btrfs_test_opt(root, COMPRESS) ||
 434                                 !btrfs_test_opt(root, FORCE_COMPRESS)) {
 435                                         btrfs_info(root->fs_info,
 436                                                 "setting nodatacow, compression disabled");
 437                         } else {
 438                                 btrfs_info(root->fs_info, "setting nodatacow");
 439                         }
 440                         btrfs_clear_opt(info->mount_opt, COMPRESS);
 441                         btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 442                         btrfs_set_opt(info->mount_opt, NODATACOW);
 443                         btrfs_set_opt(info->mount_opt, NODATASUM);
 444                         break;
 445                 case Opt_compress_force:
 446                 case Opt_compress_force_type:
 447                         compress_force = true;
 448                         /* Fallthrough */
 449                 case Opt_compress:
 450                 case Opt_compress_type:
 451                         if (token == Opt_compress ||
 452                             token == Opt_compress_force ||
 453                             strcmp(args[0].from, "zlib") == 0) {
 454                                 compress_type = "zlib";
 455                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 456                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 457                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 458                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 459                         } else if (strcmp(args[0].from, "lzo") == 0) {
 460                                 compress_type = "lzo";
 461                                 info->compress_type = BTRFS_COMPRESS_LZO;
 462                                 btrfs_set_opt(info->mount_opt, COMPRESS);
 463                                 btrfs_clear_opt(info->mount_opt, NODATACOW);
 464                                 btrfs_clear_opt(info->mount_opt, NODATASUM);
 465                                 btrfs_set_fs_incompat(info, COMPRESS_LZO);
 466                         } else if (strncmp(args[0].from, "no", 2) == 0) {
 467                                 compress_type = "no";
 468                                 btrfs_clear_opt(info->mount_opt, COMPRESS);
 469                                 btrfs_clear_opt(info->mount_opt, FORCE_COMPRESS);
 470                                 compress_force = false;
 471                         } else {
 472                                 ret = -EINVAL;
 473                                 goto out;
 474                         }
 475
 476                         if (compress_force) {
 477                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 478                                 btrfs_info(root->fs_info, "force %s compression",
 479                                         compress_type);
 480                         } else if (btrfs_test_opt(root, COMPRESS)) {
 481                                 pr_info("btrfs: use %s compression\n",
 482                                         compress_type);
 483                         }
 484                         break;
 485                 case Opt_ssd:
 486                         btrfs_info(root->fs_info, "use ssd allocation scheme");
 487                         btrfs_set_opt(info->mount_opt, SSD);
 488                         break;
 489                 case Opt_ssd_spread:
 490                         btrfs_info(root->fs_info, "use spread ssd allocation scheme");
 491                         btrfs_set_opt(info->mount_opt, SSD);
 492                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 493                         break;
 494                 case Opt_nossd:
 495                         btrfs_info(root->fs_info, "not using ssd allocation scheme");
 496                         btrfs_set_opt(info->mount_opt, NOSSD);
 497                         btrfs_clear_opt(info->mount_opt, SSD);
 498                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 499                         break;
 500                 case Opt_barrier:
 501                         if (btrfs_test_opt(root, NOBARRIER))
 502                                 btrfs_info(root->fs_info, "turning on barriers");
 503                         btrfs_clear_opt(info->mount_opt, NOBARRIER);
 504                         break;
 505                 case Opt_nobarrier:
 506                         btrfs_info(root->fs_info, "turning off barriers");
 507                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 508                         break;
 509                 case Opt_thread_pool:
 510                         ret = match_int(&args[0], &intarg);
 511                         if (ret) {
 512                                 goto out;
 513                         } else if (intarg > 0) {
 514                                 info->thread_pool_size = intarg;
 515                         } else {
 516                                 ret = -EINVAL;
 517                                 goto out;
 518                         }
 519                         break;
 520                 case Opt_max_inline:
 521                         num = match_strdup(&args[0]);
 522                         if (num) {
 523                                 info->max_inline = memparse(num, NULL);
 524                                 kfree(num);
 525
 526                                 if (info->max_inline) {
 527                                         info->max_inline = max_t(u64,
 528                                                 info->max_inline,
 529                                                 root->sectorsize);
 530                                 }
 531                                 btrfs_info(root->fs_info, "max_inline at %llu",
 532                                         info->max_inline);
 533                         } else {
 534                                 ret = -ENOMEM;
 535                                 goto out;
 536                         }
 537                         break;
 538                 case Opt_alloc_start:
 539                         num = match_strdup(&args[0]);
 540                         if (num) {
 541                                 mutex_lock(&info->chunk_mutex);
 542                                 info->alloc_start = memparse(num, NULL);
 543                                 mutex_unlock(&info->chunk_mutex);
 544                                 kfree(num);
 545                                 btrfs_info(root->fs_info, "allocations start at %llu",
 546                                         info->alloc_start);
 547                         } else {
 548                                 ret = -ENOMEM;
 549                                 goto out;
 550                         }
 551                         break;
 552                 case Opt_noacl:
 553                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 554                         break;
 555                 case Opt_notreelog:
 556                         btrfs_info(root->fs_info, "disabling tree log");
 557                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 558                         break;
 559                 case Opt_flushoncommit:
 560                         btrfs_info(root->fs_info, "turning on flush-on-commit");
 561                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 562                         break;
 563                 case Opt_ratio:
 564                         ret = match_int(&args[0], &intarg);
 565                         if (ret) {
 566                                 goto out;
 567                         } else if (intarg >= 0) {
 568                                 info->metadata_ratio = intarg;
 569                                 btrfs_info(root->fs_info, "metadata ratio %d",
 570                                        info->metadata_ratio);
 571                         } else {
 572                                 ret = -EINVAL;
 573                                 goto out;
 574                         }
 575                         break;
 576                 case Opt_discard:
 577                         btrfs_set_opt(info->mount_opt, DISCARD);
 578                         break;
 579                 case Opt_nodiscard:
 580                         btrfs_clear_opt(info->mount_opt, DISCARD);
 581                         break;
 582                 case Opt_space_cache:
 583                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 584                         break;
 585                 case Opt_rescan_uuid_tree:
 586                         btrfs_set_opt(info->mount_opt, RESCAN_UUID_TREE);
 587                         break;
 588                 case Opt_no_space_cache:
 589                         btrfs_info(root->fs_info, "disabling disk space caching");
 590                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 591                         break;
 592                 case Opt_inode_cache:
 593                         btrfs_info(root->fs_info, "enabling inode map caching");
 594                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 595                         break;
 596                 case Opt_clear_cache:
 597                         btrfs_info(root->fs_info, "force clearing of disk cache");
 598                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 599                         break;
 600                 case Opt_user_subvol_rm_allowed:
 601                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 602                         break;
 603                 case Opt_enospc_debug:
 604                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 605                         break;
 606                 case Opt_defrag:
 607                         btrfs_info(root->fs_info, "enabling auto defrag");
 608                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 609                         break;
 610                 case Opt_nodefrag:
 611                         if (btrfs_test_opt(root, AUTO_DEFRAG))
 612                                 btrfs_info(root->fs_info, "disabling auto defrag");
 613                         btrfs_clear_opt(info->mount_opt, AUTO_DEFRAG);
 614                         break;
 615                 case Opt_recovery:
 616                         btrfs_info(root->fs_info, "enabling auto recovery");
 617                         btrfs_set_opt(info->mount_opt, RECOVERY);
 618                         break;
 619                 case Opt_skip_balance:
 620                         btrfs_set_opt(info->mount_opt, SKIP_BALANCE);
 621                         break;
 622 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 623                 case Opt_check_integrity_including_extent_data:
 624                         btrfs_info(root->fs_info,
 625                                    "enabling check integrity including extent data");
 626                         btrfs_set_opt(info->mount_opt,
 627                                       CHECK_INTEGRITY_INCLUDING_EXTENT_DATA);
 628                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 629                         break;
 630                 case Opt_check_integrity:
 631                         btrfs_info(root->fs_info, "enabling check integrity");
 632                         btrfs_set_opt(info->mount_opt, CHECK_INTEGRITY);
 633                         break;
 634                 case Opt_check_integrity_print_mask:
 635                         ret = match_int(&args[0], &intarg);
 636                         if (ret) {
 637                                 goto out;
 638                         } else if (intarg >= 0) {
 639                                 info->check_integrity_print_mask = intarg;
 640                                 btrfs_info(root->fs_info, "check_integrity_print_mask 0x%x",
 641                                        info->check_integrity_print_mask);
 642                         } else {
 643                                 ret = -EINVAL;
 644                                 goto out;
 645                         }
 646                         break;
 647 #else
 648                 case Opt_check_integrity_including_extent_data:
 649                 case Opt_check_integrity:
 650                 case Opt_check_integrity_print_mask:
 651                         btrfs_err(root->fs_info,
 652                                 "support for check_integrity* not compiled in!");
 653                         ret = -EINVAL;
 654                         goto out;
 655 #endif
 656                 case Opt_fatal_errors:
 657                         if (strcmp(args[0].from, "panic") == 0)
 658                                 btrfs_set_opt(info->mount_opt,
 659                                               PANIC_ON_FATAL_ERROR);
 660                         else if (strcmp(args[0].from, "bug") == 0)
 661                                 btrfs_clear_opt(info->mount_opt,
 662                                               PANIC_ON_FATAL_ERROR);
 663                         else {
 664                                 ret = -EINVAL;
 665                                 goto out;
 666                         }
 667                         break;
 668                 case Opt_commit_interval:
 669                         intarg = 0;
 670                         ret = match_int(&args[0], &intarg);
 671                         if (ret < 0) {
 672                                 btrfs_err(root->fs_info, "invalid commit interval");
 673                                 ret = -EINVAL;
 674                                 goto out;
 675                         }
 676                         if (intarg > 0) {
 677                                 if (intarg > 300) {
 678                                         btrfs_warn(root->fs_info, "excessive commit interval %d",
 679                                                         intarg);
 680                                 }
 681                                 info->commit_interval = intarg;
 682                         } else {
 683                                 btrfs_info(root->fs_info, "using default commit interval %ds",
 684                                     BTRFS_DEFAULT_COMMIT_INTERVAL);
 685                                 info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
 686                         }
 687                         break;
 688                 case Opt_err:
 689                         btrfs_info(root->fs_info, "unrecognized mount option '%s'", p);
 690                         ret = -EINVAL;
 691                         goto out;
 692                 default:
 693                         break;
 694                 }
 695         }
 696 out:
 697         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 698                 btrfs_info(root->fs_info, "disk space caching is enabled");
 699         kfree(orig);
 700         return ret;
 701 }
 702
 703 /*
 704  * Parse mount options that are required early in the mount process.
 705  *
 706  * All other options will be parsed on much later in the mount process and
 707  * only when we need to allocate a new super block.
 708  */
 709 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 710                 void *holder, char **subvol_name, u64 *subvol_objectid,
 711                 struct btrfs_fs_devices **fs_devices)
 712 {
 713         substring_t args[MAX_OPT_ARGS];
 714         char *device_name, *opts, *orig, *p;
 715         char *num = NULL;
 716         int error = 0;
 717
 718         if (!options)
 719                 return 0;
 720
 721         /*
 722          * strsep changes the string, duplicate it because parse_options
 723          * gets called twice
 724          */
 725         opts = kstrdup(options, GFP_KERNEL);
 726         if (!opts)
 727                 return -ENOMEM;
 728         orig = opts;
 729
 730         while ((p = strsep(&opts, ",")) != NULL) {
 731                 int token;
 732                 if (!*p)
 733                         continue;
 734
 735                 token = match_token(p, tokens, args);
 736                 switch (token) {
 737                 case Opt_subvol:
 738                         kfree(*subvol_name);
 739                         *subvol_name = match_strdup(&args[0]);
 740                         if (!*subvol_name) {
 741                                 error = -ENOMEM;
 742                                 goto out;
 743                         }
 744                         break;
 745                 case Opt_subvolid:
 746                         num = match_strdup(&args[0]);
 747                         if (num) {
 748                                 *subvol_objectid = memparse(num, NULL);
 749                                 kfree(num);
 750                                 /* we want the original fs_tree */
 751                                 if (!*subvol_objectid)
 752                                         *subvol_objectid =
 753                                                 BTRFS_FS_TREE_OBJECTID;
 754                         } else {
 755                                 error = -EINVAL;
 756                                 goto out;
 757                         }
 758                         break;
 759                 case Opt_subvolrootid:
 760                         printk(KERN_WARNING
 761                                 "BTRFS: 'subvolrootid' mount option is deprecated and has "
 762                                 "no effect\n");
 763                         break;
 764                 case Opt_device:
 765                         device_name = match_strdup(&args[0]);
 766                         if (!device_name) {
 767                                 error = -ENOMEM;
 768                                 goto out;
 769                         }
 770                         error = btrfs_scan_one_device(device_name,
 771                                         flags, holder, fs_devices);
 772                         kfree(device_name);
 773                         if (error)
 774                                 goto out;
 775                         break;
 776                 default:
 777                         break;
 778                 }
 779         }
 780
 781 out:
 782         kfree(orig);
 783         return error;
 784 }
 785
 786 static struct dentry *get_default_root(struct super_block *sb,
 787                                        u64 subvol_objectid)
 788 {
 789         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 790         struct btrfs_root *root = fs_info->tree_root;
 791         struct btrfs_root *new_root;
 792         struct btrfs_dir_item *di;
 793         struct btrfs_path *path;
 794         struct btrfs_key location;
 795         struct inode *inode;
 796         u64 dir_id;
 797         int new = 0;
 798
 799         /*
 800          * We have a specific subvol we want to mount, just setup location and
 801          * go look up the root.
 802          */
 803         if (subvol_objectid) {
 804                 location.objectid = subvol_objectid;
 805                 location.type = BTRFS_ROOT_ITEM_KEY;
 806                 location.offset = (u64)-1;
 807                 goto find_root;
 808         }
 809
 810         path = btrfs_alloc_path();
 811         if (!path)
 812                 return ERR_PTR(-ENOMEM);
 813         path->leave_spinning = 1;
 814
 815         /*
 816          * Find the "default" dir item which points to the root item that we
 817          * will mount by default if we haven't been given a specific subvolume
 818          * to mount.
 819          */
 820         dir_id = btrfs_super_root_dir(fs_info->super_copy);
 821         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 822         if (IS_ERR(di)) {
 823                 btrfs_free_path(path);
 824                 return ERR_CAST(di);
 825         }
 826         if (!di) {
 827                 /*
 828                  * Ok the default dir item isn't there.  This is weird since
 829                  * it's always been there, but don't freak out, just try and
 830                  * mount to root most subvolume.
 831                  */
 832                 btrfs_free_path(path);
 833                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 834                 new_root = fs_info->fs_root;
 835                 goto setup_root;
 836         }
 837
 838         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 839         btrfs_free_path(path);
 840
 841 find_root:
 842         new_root = btrfs_read_fs_root_no_name(fs_info, &location);
 843         if (IS_ERR(new_root))
 844                 return ERR_CAST(new_root);
 845
 846         dir_id = btrfs_root_dirid(&new_root->root_item);
 847 setup_root:
 848         location.objectid = dir_id;
 849         location.type = BTRFS_INODE_ITEM_KEY;
 850         location.offset = 0;
 851
 852         inode = btrfs_iget(sb, &location, new_root, &new);
 853         if (IS_ERR(inode))
 854                 return ERR_CAST(inode);
 855
 856         /*
 857          * If we're just mounting the root most subvol put the inode and return
 858          * a reference to the dentry.  We will have already gotten a reference
 859          * to the inode in btrfs_fill_super so we're good to go.
 860          */
 861         if (!new && sb->s_root->d_inode == inode) {
 862                 iput(inode);
 863                 return dget(sb->s_root);
 864         }
 865
 866         return d_obtain_alias(inode);
 867 }
 868
 869 static int btrfs_fill_super(struct super_block *sb,
 870                             struct btrfs_fs_devices *fs_devices,
 871                             void *data, int silent)
 872 {
 873         struct inode *inode;
 874         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 875         struct btrfs_key key;
 876         int err;
 877
 878         sb->s_maxbytes = MAX_LFS_FILESIZE;
 879         sb->s_magic = BTRFS_SUPER_MAGIC;
 880         sb->s_op = &btrfs_super_ops;
 881         sb->s_d_op = &btrfs_dentry_operations;
 882         sb->s_export_op = &btrfs_export_ops;
 883         sb->s_xattr = btrfs_xattr_handlers;
 884         sb->s_time_gran = 1;
 885 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 886         sb->s_flags |= MS_POSIXACL;
 887 #endif
 888         sb->s_flags |= MS_I_VERSION;
 889         err = open_ctree(sb, fs_devices, (char *)data);
 890         if (err) {
 891                 printk(KERN_ERR "BTRFS: open_ctree failed\n");
 892                 return err;
 893         }
 894
 895         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 896         key.type = BTRFS_INODE_ITEM_KEY;
 897         key.offset = 0;
 898         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 899         if (IS_ERR(inode)) {
 900                 err = PTR_ERR(inode);
 901                 goto fail_close;
 902         }
 903
 904         sb->s_root = d_make_root(inode);
 905         if (!sb->s_root) {
 906                 err = -ENOMEM;
 907                 goto fail_close;
 908         }
 909
 910         save_mount_options(sb, data);
 911         cleancache_init_fs(sb);
 912         sb->s_flags |= MS_ACTIVE;
 913         return 0;
 914
 915 fail_close:
 916         close_ctree(fs_info->tree_root);
 917         return err;
 918 }
 919
 920 int btrfs_sync_fs(struct super_block *sb, int wait)
 921 {
 922         struct btrfs_trans_handle *trans;
 923         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
 924         struct btrfs_root *root = fs_info->tree_root;
 925
 926         trace_btrfs_sync_fs(wait);
 927
 928         if (!wait) {
 929                 filemap_flush(fs_info->btree_inode->i_mapping);
 930                 return 0;
 931         }
 932
 933         btrfs_wait_ordered_roots(fs_info, -1);
 934
 935         trans = btrfs_attach_transaction_barrier(root);
 936         if (IS_ERR(trans)) {
 937                 /* no transaction, don't bother */
 938                 if (PTR_ERR(trans) == -ENOENT)
 939                         return 0;
 940                 return PTR_ERR(trans);
 941         }
 942         return btrfs_commit_transaction(trans, root);
 943 }
 944
 945 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 946 {
 947         struct btrfs_fs_info *info = btrfs_sb(dentry->d_sb);
 948         struct btrfs_root *root = info->tree_root;
 949         char *compress_type;
 950
 951         if (btrfs_test_opt(root, DEGRADED))
 952                 seq_puts(seq, ",degraded");
 953         if (btrfs_test_opt(root, NODATASUM))
 954                 seq_puts(seq, ",nodatasum");
 955         if (btrfs_test_opt(root, NODATACOW))
 956                 seq_puts(seq, ",nodatacow");
 957         if (btrfs_test_opt(root, NOBARRIER))
 958                 seq_puts(seq, ",nobarrier");
 959         if (info->max_inline != 8192 * 1024)
 960                 seq_printf(seq, ",max_inline=%llu", info->max_inline);
 961         if (info->alloc_start != 0)
 962                 seq_printf(seq, ",alloc_start=%llu", info->alloc_start);
 963         if (info->thread_pool_size !=  min_t(unsigned long,
 964                                              num_online_cpus() + 2, 8))
 965                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 966         if (btrfs_test_opt(root, COMPRESS)) {
 967                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 968                         compress_type = "zlib";
 969                 else
 970                         compress_type = "lzo";
 971                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 972                         seq_printf(seq, ",compress-force=%s", compress_type);
 973                 else
 974                         seq_printf(seq, ",compress=%s", compress_type);
 975         }
 976         if (btrfs_test_opt(root, NOSSD))
 977                 seq_puts(seq, ",nossd");
 978         if (btrfs_test_opt(root, SSD_SPREAD))
 979                 seq_puts(seq, ",ssd_spread");
 980         else if (btrfs_test_opt(root, SSD))
 981                 seq_puts(seq, ",ssd");
 982         if (btrfs_test_opt(root, NOTREELOG))
 983                 seq_puts(seq, ",notreelog");
 984         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 985                 seq_puts(seq, ",flushoncommit");
 986         if (btrfs_test_opt(root, DISCARD))
 987                 seq_puts(seq, ",discard");
 988         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 989                 seq_puts(seq, ",noacl");
 990         if (btrfs_test_opt(root, SPACE_CACHE))
 991                 seq_puts(seq, ",space_cache");
 992         else
 993                 seq_puts(seq, ",nospace_cache");
 994         if (btrfs_test_opt(root, RESCAN_UUID_TREE))
 995                 seq_puts(seq, ",rescan_uuid_tree");
 996         if (btrfs_test_opt(root, CLEAR_CACHE))
 997                 seq_puts(seq, ",clear_cache");
 998         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 999                 seq_puts(seq, ",user_subvol_rm_allowed");
1000         if (btrfs_test_opt(root, ENOSPC_DEBUG))
1001                 seq_puts(seq, ",enospc_debug");
1002         if (btrfs_test_opt(root, AUTO_DEFRAG))
1003                 seq_puts(seq, ",autodefrag");
1004         if (btrfs_test_opt(root, INODE_MAP_CACHE))
1005                 seq_puts(seq, ",inode_cache");
1006         if (btrfs_test_opt(root, SKIP_BALANCE))
1007                 seq_puts(seq, ",skip_balance");
1008         if (btrfs_test_opt(root, RECOVERY))
1009                 seq_puts(seq, ",recovery");
1010 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1011         if (btrfs_test_opt(root, CHECK_INTEGRITY_INCLUDING_EXTENT_DATA))
1012                 seq_puts(seq, ",check_int_data");
1013         else if (btrfs_test_opt(root, CHECK_INTEGRITY))
1014                 seq_puts(seq, ",check_int");
1015         if (info->check_integrity_print_mask)
1016                 seq_printf(seq, ",check_int_print_mask=%d",
1017                                 info->check_integrity_print_mask);
1018 #endif
1019         if (info->metadata_ratio)
1020                 seq_printf(seq, ",metadata_ratio=%d",
1021                                 info->metadata_ratio);
1022         if (btrfs_test_opt(root, PANIC_ON_FATAL_ERROR))
1023                 seq_puts(seq, ",fatal_errors=panic");
1024         if (info->commit_interval != BTRFS_DEFAULT_COMMIT_INTERVAL)
1025                 seq_printf(seq, ",commit=%d", info->commit_interval);
1026         return 0;
1027 }
1028
1029 static int btrfs_test_super(struct super_block *s, void *data)
1030 {
1031         struct btrfs_fs_info *p = data;
1032         struct btrfs_fs_info *fs_info = btrfs_sb(s);
1033
1034         return fs_info->fs_devices == p->fs_devices;
1035 }
1036
1037 static int btrfs_set_super(struct super_block *s, void *data)
1038 {
1039         int err = set_anon_super(s, data);
1040         if (!err)
1041                 s->s_fs_info = data;
1042         return err;
1043 }
1044
1045 /*
1046  * subvolumes are identified by ino 256
1047  */
1048 static inline int is_subvolume_inode(struct inode *inode)
1049 {
1050         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
1051                 return 1;
1052         return 0;
1053 }
1054
1055 /*
1056  * This will strip out the subvol=%s argument for an argument string and add
1057  * subvolid=0 to make sure we get the actual tree root for path walking to the
1058  * subvol we want.
1059  */
1060 static char *setup_root_args(char *args)
1061 {
1062         unsigned len = strlen(args) + 2 + 1;
1063         char *src, *dst, *buf;
1064
1065         /*
1066          * We need the same args as before, but with this substitution:
1067          * s!subvol=[^,]+!subvolid=0!
1068          *
1069          * Since the replacement string is up to 2 bytes longer than the
1070          * original, allocate strlen(args) + 2 + 1 bytes.
1071          */
1072
1073         src = strstr(args, "subvol=");
1074         /* This shouldn't happen, but just in case.. */
1075         if (!src)
1076                 return NULL;
1077
1078         buf = dst = kmalloc(len, GFP_NOFS);
1079         if (!buf)
1080                 return NULL;
1081
1082         /*
1083          * If the subvol= arg is not at the start of the string,
1084          * copy whatever precedes it into buf.
1085          */
1086         if (src != args) {
1087                 *src++ = '\0';
1088                 strcpy(buf, args);
1089                 dst += strlen(args);
1090         }
1091
1092         strcpy(dst, "subvolid=0");
1093         dst += strlen("subvolid=0");
1094
1095         /*
1096          * If there is a "," after the original subvol=... string,
1097          * copy that suffix into our buffer.  Otherwise, we're done.
1098          */
1099         src = strchr(src, ',');
1100         if (src)
1101                 strcpy(dst, src);
1102
1103         return buf;
1104 }
1105
1106 static struct dentry *mount_subvol(const char *subvol_name, int flags,
1107                                    const char *device_name, char *data)
1108 {
1109         struct dentry *root;
1110         struct vfsmount *mnt;
1111         char *newargs;
1112
1113         newargs = setup_root_args(data);
1114         if (!newargs)
1115                 return ERR_PTR(-ENOMEM);
1116         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
1117                              newargs);
1118         kfree(newargs);
1119         if (IS_ERR(mnt))
1120                 return ERR_CAST(mnt);
1121
1122         root = mount_subtree(mnt, subvol_name);
1123
1124         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
1125                 struct super_block *s = root->d_sb;
1126                 dput(root);
1127                 root = ERR_PTR(-EINVAL);
1128                 deactivate_locked_super(s);
1129                 printk(KERN_ERR "BTRFS: '%s' is not a valid subvolume\n",
1130                                 subvol_name);
1131         }
1132
1133         return root;
1134 }
1135
1136 /*
1137  * Find a superblock for the given device / mount point.
1138  *
1139  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
1140  *        for multiple device setup.  Make sure to keep it in sync.
1141  */
1142 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
1143                 const char *device_name, void *data)
1144 {
1145         struct block_device *bdev = NULL;
1146         struct super_block *s;
1147         struct dentry *root;
1148         struct btrfs_fs_devices *fs_devices = NULL;
1149         struct btrfs_fs_info *fs_info = NULL;
1150         fmode_t mode = FMODE_READ;
1151         char *subvol_name = NULL;
1152         u64 subvol_objectid = 0;
1153         int error = 0;
1154
1155         if (!(flags & MS_RDONLY))
1156                 mode |= FMODE_WRITE;
1157
1158         error = btrfs_parse_early_options(data, mode, fs_type,
1159                                           &subvol_name, &subvol_objectid,
1160                                           &fs_devices);
1161         if (error) {
1162                 kfree(subvol_name);
1163                 return ERR_PTR(error);
1164         }
1165
1166         if (subvol_name) {
1167                 root = mount_subvol(subvol_name, flags, device_name, data);
1168                 kfree(subvol_name);
1169                 return root;
1170         }
1171
1172         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
1173         if (error)
1174                 return ERR_PTR(error);
1175
1176         /*
1177          * Setup a dummy root and fs_info for test/set super.  This is because
1178          * we don't actually fill this stuff out until open_ctree, but we need
1179          * it for searching for existing supers, so this lets us do that and
1180          * then open_ctree will properly initialize everything later.
1181          */
1182         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
1183         if (!fs_info)
1184                 return ERR_PTR(-ENOMEM);
1185
1186         fs_info->fs_devices = fs_devices;
1187
1188         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1189         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
1190         if (!fs_info->super_copy || !fs_info->super_for_commit) {
1191                 error = -ENOMEM;
1192                 goto error_fs_info;
1193         }
1194
1195         error = btrfs_open_devices(fs_devices, mode, fs_type);
1196         if (error)
1197                 goto error_fs_info;
1198
1199         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
1200                 error = -EACCES;
1201                 goto error_close_devices;
1202         }
1203
1204         bdev = fs_devices->latest_bdev;
1205         s = sget(fs_type, btrfs_test_super, btrfs_set_super, flags | MS_NOSEC,
1206                  fs_info);
1207         if (IS_ERR(s)) {
1208                 error = PTR_ERR(s);
1209                 goto error_close_devices;
1210         }
1211
1212         if (s->s_root) {
1213                 btrfs_close_devices(fs_devices);
1214                 free_fs_info(fs_info);
1215                 if ((flags ^ s->s_flags) & MS_RDONLY)
1216                         error = -EBUSY;
1217         } else {
1218                 char b[BDEVNAME_SIZE];
1219
1220                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1221                 btrfs_sb(s)->bdev_holder = fs_type;
1222                 error = btrfs_fill_super(s, fs_devices, data,
1223                                          flags & MS_SILENT ? 1 : 0);
1224         }
1225
1226         root = !error ? get_default_root(s, subvol_objectid) : ERR_PTR(error);
1227         if (IS_ERR(root))
1228                 deactivate_locked_super(s);
1229
1230         return root;
1231
1232 error_close_devices:
1233         btrfs_close_devices(fs_devices);
1234 error_fs_info:
1235         free_fs_info(fs_info);
1236         return ERR_PTR(error);
1237 }
1238
1239 static void btrfs_set_max_workers(struct btrfs_workers *workers, int new_limit)
1240 {
1241         spin_lock_irq(&workers->lock);
1242         workers->max_workers = new_limit;
1243         spin_unlock_irq(&workers->lock);
1244 }
1245
1246 static void btrfs_resize_thread_pool(struct btrfs_fs_info *fs_info,
1247                                      int new_pool_size, int old_pool_size)
1248 {
1249         if (new_pool_size == old_pool_size)
1250                 return;
1251
1252         fs_info->thread_pool_size = new_pool_size;
1253
1254         btrfs_info(fs_info, "resize thread pool %d -> %d",
1255                old_pool_size, new_pool_size);
1256
1257         btrfs_set_max_workers(&fs_info->generic_worker, new_pool_size);
1258         btrfs_set_max_workers(&fs_info->workers, new_pool_size);
1259         btrfs_set_max_workers(&fs_info->delalloc_workers, new_pool_size);
1260         btrfs_set_max_workers(&fs_info->submit_workers, new_pool_size);
1261         btrfs_set_max_workers(&fs_info->caching_workers, new_pool_size);
1262         btrfs_set_max_workers(&fs_info->fixup_workers, new_pool_size);
1263         btrfs_set_max_workers(&fs_info->endio_workers, new_pool_size);
1264         btrfs_set_max_workers(&fs_info->endio_meta_workers, new_pool_size);
1265         btrfs_set_max_workers(&fs_info->endio_meta_write_workers, new_pool_size);
1266         btrfs_set_max_workers(&fs_info->endio_write_workers, new_pool_size);
1267         btrfs_set_max_workers(&fs_info->endio_freespace_worker, new_pool_size);
1268         btrfs_set_max_workers(&fs_info->delayed_workers, new_pool_size);
1269         btrfs_set_max_workers(&fs_info->readahead_workers, new_pool_size);
1270         btrfs_set_max_workers(&fs_info->scrub_wr_completion_workers,
1271                               new_pool_size);
1272 }
1273
1274 static inline void btrfs_remount_prepare(struct btrfs_fs_info *fs_info)
1275 {
1276         set_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1277 }
1278
1279 static inline void btrfs_remount_begin(struct btrfs_fs_info *fs_info,
1280                                        unsigned long old_opts, int flags)
1281 {
1282         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1283             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1284              (flags & MS_RDONLY))) {
1285                 /* wait for any defraggers to finish */
1286                 wait_event(fs_info->transaction_wait,
1287                            (atomic_read(&fs_info->defrag_running) == 0));
1288                 if (flags & MS_RDONLY)
1289                         sync_filesystem(fs_info->sb);
1290         }
1291 }
1292
1293 static inline void btrfs_remount_cleanup(struct btrfs_fs_info *fs_info,
1294                                          unsigned long old_opts)
1295 {
1296         /*
1297          * We need cleanup all defragable inodes if the autodefragment is
1298          * close or the fs is R/O.
1299          */
1300         if (btrfs_raw_test_opt(old_opts, AUTO_DEFRAG) &&
1301             (!btrfs_raw_test_opt(fs_info->mount_opt, AUTO_DEFRAG) ||
1302              (fs_info->sb->s_flags & MS_RDONLY))) {
1303                 btrfs_cleanup_defrag_inodes(fs_info);
1304         }
1305
1306         clear_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
1307 }
1308
1309 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
1310 {
1311         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1312         struct btrfs_root *root = fs_info->tree_root;
1313         unsigned old_flags = sb->s_flags;
1314         unsigned long old_opts = fs_info->mount_opt;
1315         unsigned long old_compress_type = fs_info->compress_type;
1316         u64 old_max_inline = fs_info->max_inline;
1317         u64 old_alloc_start = fs_info->alloc_start;
1318         int old_thread_pool_size = fs_info->thread_pool_size;
1319         unsigned int old_metadata_ratio = fs_info->metadata_ratio;
1320         int ret;
1321
1322         btrfs_remount_prepare(fs_info);
1323
1324         ret = btrfs_parse_options(root, data);
1325         if (ret) {
1326                 ret = -EINVAL;
1327                 goto restore;
1328         }
1329
1330         btrfs_remount_begin(fs_info, old_opts, *flags);
1331         btrfs_resize_thread_pool(fs_info,
1332                 fs_info->thread_pool_size, old_thread_pool_size);
1333
1334         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1335                 goto out;
1336
1337         if (*flags & MS_RDONLY) {
1338                 /*
1339                  * this also happens on 'umount -rf' or on shutdown, when
1340                  * the filesystem is busy.
1341                  */
1342
1343                 /* wait for the uuid_scan task to finish */
1344                 down(&fs_info->uuid_tree_rescan_sem);
1345                 /* avoid complains from lockdep et al. */
1346                 up(&fs_info->uuid_tree_rescan_sem);
1347
1348                 sb->s_flags |= MS_RDONLY;
1349
1350                 btrfs_dev_replace_suspend_for_unmount(fs_info);
1351                 btrfs_scrub_cancel(fs_info);
1352                 btrfs_pause_balance(fs_info);
1353
1354                 ret = btrfs_commit_super(root);
1355                 if (ret)
1356                         goto restore;
1357         } else {
1358                 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1359                         btrfs_err(fs_info,
1360                                 "Remounting read-write after error is not allowed");
1361                         ret = -EINVAL;
1362                         goto restore;
1363                 }
1364                 if (fs_info->fs_devices->rw_devices == 0) {
1365                         ret = -EACCES;
1366                         goto restore;
1367                 }
1368
1369                 if (fs_info->fs_devices->missing_devices >
1370                      fs_info->num_tolerated_disk_barrier_failures &&
1371                     !(*flags & MS_RDONLY)) {
1372                         btrfs_warn(fs_info,
1373                                 "too many missing devices, writeable remount is not allowed");
1374                         ret = -EACCES;
1375                         goto restore;
1376                 }
1377
1378                 if (btrfs_super_log_root(fs_info->super_copy) != 0) {
1379                         ret = -EINVAL;
1380                         goto restore;
1381                 }
1382
1383                 ret = btrfs_cleanup_fs_roots(fs_info);
1384                 if (ret)
1385                         goto restore;
1386
1387                 /* recover relocation */
1388                 ret = btrfs_recover_relocation(root);
1389                 if (ret)
1390                         goto restore;
1391
1392                 ret = btrfs_resume_balance_async(fs_info);
1393                 if (ret)
1394                         goto restore;
1395
1396                 ret = btrfs_resume_dev_replace_async(fs_info);
1397                 if (ret) {
1398                         btrfs_warn(fs_info, "failed to resume dev_replace");
1399                         goto restore;
1400                 }
1401
1402                 if (!fs_info->uuid_root) {
1403                         btrfs_info(fs_info, "creating UUID tree");
1404                         ret = btrfs_create_uuid_tree(fs_info);
1405                         if (ret) {
1406                                 btrfs_warn(fs_info, "failed to create the UUID tree %d", ret);
1407                                 goto restore;
1408                         }
1409                 }
1410                 sb->s_flags &= ~MS_RDONLY;
1411         }
1412 out:
1413         btrfs_remount_cleanup(fs_info, old_opts);
1414         return 0;
1415
1416 restore:
1417         /* We've hit an error - don't reset MS_RDONLY */
1418         if (sb->s_flags & MS_RDONLY)
1419                 old_flags |= MS_RDONLY;
1420         sb->s_flags = old_flags;
1421         fs_info->mount_opt = old_opts;
1422         fs_info->compress_type = old_compress_type;
1423         fs_info->max_inline = old_max_inline;
1424         mutex_lock(&fs_info->chunk_mutex);
1425         fs_info->alloc_start = old_alloc_start;
1426         mutex_unlock(&fs_info->chunk_mutex);
1427         btrfs_resize_thread_pool(fs_info,
1428                 old_thread_pool_size, fs_info->thread_pool_size);
1429         fs_info->metadata_ratio = old_metadata_ratio;
1430         btrfs_remount_cleanup(fs_info, old_opts);
1431         return ret;
1432 }
1433
1434 /* Used to sort the devices by max_avail(descending sort) */
1435 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1436                                        const void *dev_info2)
1437 {
1438         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1439             ((struct btrfs_device_info *)dev_info2)->max_avail)
1440                 return -1;
1441         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1442                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1443                 return 1;
1444         else
1445         return 0;
1446 }
1447
1448 /*
1449  * sort the devices by max_avail, in which max free extent size of each device
1450  * is stored.(Descending Sort)
1451  */
1452 static inline void btrfs_descending_sort_devices(
1453                                         struct btrfs_device_info *devices,
1454                                         size_t nr_devices)
1455 {
1456         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1457              btrfs_cmp_device_free_bytes, NULL);
1458 }
1459
1460 /*
1461  * The helper to calc the free space on the devices that can be used to store
1462  * file data.
1463  */
1464 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1465 {
1466         struct btrfs_fs_info *fs_info = root->fs_info;
1467         struct btrfs_device_info *devices_info;
1468         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1469         struct btrfs_device *device;
1470         u64 skip_space;
1471         u64 type;
1472         u64 avail_space;
1473         u64 used_space;
1474         u64 min_stripe_size;
1475         int min_stripes = 1, num_stripes = 1;
1476         int i = 0, nr_devices;
1477         int ret;
1478
1479         nr_devices = fs_info->fs_devices->open_devices;
1480         BUG_ON(!nr_devices);
1481
1482         devices_info = kmalloc_array(nr_devices, sizeof(*devices_info),
1483                                GFP_NOFS);
1484         if (!devices_info)
1485                 return -ENOMEM;
1486
1487         /* calc min stripe number for data space alloction */
1488         type = btrfs_get_alloc_profile(root, 1);
1489         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1490                 min_stripes = 2;
1491                 num_stripes = nr_devices;
1492         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1493                 min_stripes = 2;
1494                 num_stripes = 2;
1495         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1496                 min_stripes = 4;
1497                 num_stripes = 4;
1498         }
1499
1500         if (type & BTRFS_BLOCK_GROUP_DUP)
1501                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1502         else
1503                 min_stripe_size = BTRFS_STRIPE_LEN;
1504
1505         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1506                 if (!device->in_fs_metadata || !device->bdev ||
1507                     device->is_tgtdev_for_dev_replace)
1508                         continue;
1509
1510                 avail_space = device->total_bytes - device->bytes_used;
1511
1512                 /* align with stripe_len */
1513                 do_div(avail_space, BTRFS_STRIPE_LEN);
1514                 avail_space *= BTRFS_STRIPE_LEN;
1515
1516                 /*
1517                  * In order to avoid overwritting the superblock on the drive,
1518                  * btrfs starts at an offset of at least 1MB when doing chunk
1519                  * allocation.
1520                  */
1521                 skip_space = 1024 * 1024;
1522
1523                 /* user can set the offset in fs_info->alloc_start. */
1524                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1525                     device->total_bytes)
1526                         skip_space = max(fs_info->alloc_start, skip_space);
1527
1528                 /*
1529                  * btrfs can not use the free space in [0, skip_space - 1],
1530                  * we must subtract it from the total. In order to implement
1531                  * it, we account the used space in this range first.
1532                  */
1533                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1534                                                      &used_space);
1535                 if (ret) {
1536                         kfree(devices_info);
1537                         return ret;
1538                 }
1539
1540                 /* calc the free space in [0, skip_space - 1] */
1541                 skip_space -= used_space;
1542
1543                 /*
1544                  * we can use the free space in [0, skip_space - 1], subtract
1545                  * it from the total.
1546                  */
1547                 if (avail_space && avail_space >= skip_space)
1548                         avail_space -= skip_space;
1549                 else
1550                         avail_space = 0;
1551
1552                 if (avail_space < min_stripe_size)
1553                         continue;
1554
1555                 devices_info[i].dev = device;
1556                 devices_info[i].max_avail = avail_space;
1557
1558                 i++;
1559         }
1560
1561         nr_devices = i;
1562
1563         btrfs_descending_sort_devices(devices_info, nr_devices);
1564
1565         i = nr_devices - 1;
1566         avail_space = 0;
1567         while (nr_devices >= min_stripes) {
1568                 if (num_stripes > nr_devices)
1569                         num_stripes = nr_devices;
1570
1571                 if (devices_info[i].max_avail >= min_stripe_size) {
1572                         int j;
1573                         u64 alloc_size;
1574
1575                         avail_space += devices_info[i].max_avail * num_stripes;
1576                         alloc_size = devices_info[i].max_avail;
1577                         for (j = i + 1 - num_stripes; j <= i; j++)
1578                                 devices_info[j].max_avail -= alloc_size;
1579                 }
1580                 i--;
1581                 nr_devices--;
1582         }
1583
1584         kfree(devices_info);
1585         *free_bytes = avail_space;
1586         return 0;
1587 }
1588
1589 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1590 {
1591         struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
1592         struct btrfs_super_block *disk_super = fs_info->super_copy;
1593         struct list_head *head = &fs_info->space_info;
1594         struct btrfs_space_info *found;
1595         u64 total_used = 0;
1596         u64 total_free_data = 0;
1597         int bits = dentry->d_sb->s_blocksize_bits;
1598         __be32 *fsid = (__be32 *)fs_info->fsid;
1599         int ret;
1600
1601         /* holding chunk_muext to avoid allocating new chunks */
1602         mutex_lock(&fs_info->chunk_mutex);
1603         rcu_read_lock();
1604         list_for_each_entry_rcu(found, head, list) {
1605                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1606                         total_free_data += found->disk_total - found->disk_used;
1607                         total_free_data -=
1608                                 btrfs_account_ro_block_groups_free_space(found);
1609                 }
1610
1611                 total_used += found->disk_used;
1612         }
1613         rcu_read_unlock();
1614
1615         buf->f_namelen = BTRFS_NAME_LEN;
1616         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1617         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1618         buf->f_bsize = dentry->d_sb->s_blocksize;
1619         buf->f_type = BTRFS_SUPER_MAGIC;
1620         buf->f_bavail = total_free_data;
1621         ret = btrfs_calc_avail_data_space(fs_info->tree_root, &total_free_data);
1622         if (ret) {
1623                 mutex_unlock(&fs_info->chunk_mutex);
1624                 return ret;
1625         }
1626         buf->f_bavail += total_free_data;
1627         buf->f_bavail = buf->f_bavail >> bits;
1628         mutex_unlock(&fs_info->chunk_mutex);
1629
1630         /* We treat it as constant endianness (it doesn't matter _which_)
1631            because we want the fsid to come out the same whether mounted
1632            on a big-endian or little-endian host */
1633         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1634         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1635         /* Mask in the root object ID too, to disambiguate subvols */
1636         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1637         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1638
1639         return 0;
1640 }
1641
1642 static void btrfs_kill_super(struct super_block *sb)
1643 {
1644         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1645         kill_anon_super(sb);
1646         free_fs_info(fs_info);
1647 }
1648
1649 static struct file_system_type btrfs_fs_type = {
1650         .owner          = THIS_MODULE,
1651         .name           = "btrfs",
1652         .mount          = btrfs_mount,
1653         .kill_sb        = btrfs_kill_super,
1654         .fs_flags       = FS_REQUIRES_DEV,
1655 };
1656 MODULE_ALIAS_FS("btrfs");
1657
1658 /*
1659  * used by btrfsctl to scan devices when no FS is mounted
1660  */
1661 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1662                                 unsigned long arg)
1663 {
1664         struct btrfs_ioctl_vol_args *vol;
1665         struct btrfs_fs_devices *fs_devices;
1666         int ret = -ENOTTY;
1667
1668         if (!capable(CAP_SYS_ADMIN))
1669                 return -EPERM;
1670
1671         vol = memdup_user((void __user *)arg, sizeof(*vol));
1672         if (IS_ERR(vol))
1673                 return PTR_ERR(vol);
1674
1675         switch (cmd) {
1676         case BTRFS_IOC_SCAN_DEV:
1677                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1678                                             &btrfs_fs_type, &fs_devices);
1679                 break;
1680         case BTRFS_IOC_DEVICES_READY:
1681                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1682                                             &btrfs_fs_type, &fs_devices);
1683                 if (ret)
1684                         break;
1685                 ret = !(fs_devices->num_devices == fs_devices->total_devices);
1686                 break;
1687         }
1688
1689         kfree(vol);
1690         return ret;
1691 }
1692
1693 static int btrfs_freeze(struct super_block *sb)
1694 {
1695         struct btrfs_trans_handle *trans;
1696         struct btrfs_root *root = btrfs_sb(sb)->tree_root;
1697
1698         trans = btrfs_attach_transaction_barrier(root);
1699         if (IS_ERR(trans)) {
1700                 /* no transaction, don't bother */
1701                 if (PTR_ERR(trans) == -ENOENT)
1702                         return 0;
1703                 return PTR_ERR(trans);
1704         }
1705         return btrfs_commit_transaction(trans, root);
1706 }
1707
1708 static int btrfs_unfreeze(struct super_block *sb)
1709 {
1710         return 0;
1711 }
1712
1713 static int btrfs_show_devname(struct seq_file *m, struct dentry *root)
1714 {
1715         struct btrfs_fs_info *fs_info = btrfs_sb(root->d_sb);
1716         struct btrfs_fs_devices *cur_devices;
1717         struct btrfs_device *dev, *first_dev = NULL;
1718         struct list_head *head;
1719         struct rcu_string *name;
1720
1721         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1722         cur_devices = fs_info->fs_devices;
1723         while (cur_devices) {
1724                 head = &cur_devices->devices;
1725                 list_for_each_entry(dev, head, dev_list) {
1726                         if (dev->missing)
1727                                 continue;
1728                         if (!first_dev || dev->devid < first_dev->devid)
1729                                 first_dev = dev;
1730                 }
1731                 cur_devices = cur_devices->seed;
1732         }
1733
1734         if (first_dev) {
1735                 rcu_read_lock();
1736                 name = rcu_dereference(first_dev->name);
1737                 seq_escape(m, name->str, " \t\n\\");
1738                 rcu_read_unlock();
1739         } else {
1740                 WARN_ON(1);
1741         }
1742         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1743         return 0;
1744 }
1745
1746 static const struct super_operations btrfs_super_ops = {
1747         .drop_inode     = btrfs_drop_inode,
1748         .evict_inode    = btrfs_evict_inode,
1749         .put_super      = btrfs_put_super,
1750         .sync_fs        = btrfs_sync_fs,
1751         .show_options   = btrfs_show_options,
1752         .show_devname   = btrfs_show_devname,
1753         .write_inode    = btrfs_write_inode,
1754         .alloc_inode    = btrfs_alloc_inode,
1755         .destroy_inode  = btrfs_destroy_inode,
1756         .statfs         = btrfs_statfs,
1757         .remount_fs     = btrfs_remount,
1758         .freeze_fs      = btrfs_freeze,
1759         .unfreeze_fs    = btrfs_unfreeze,
1760 };
1761
1762 static const struct file_operations btrfs_ctl_fops = {
1763         .unlocked_ioctl  = btrfs_control_ioctl,
1764         .compat_ioctl = btrfs_control_ioctl,
1765         .owner   = THIS_MODULE,
1766         .llseek = noop_llseek,
1767 };
1768
1769 static struct miscdevice btrfs_misc = {
1770         .minor          = BTRFS_MINOR,
1771         .name           = "btrfs-control",
1772         .fops           = &btrfs_ctl_fops
1773 };
1774
1775 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1776 MODULE_ALIAS("devname:btrfs-control");
1777
1778 static int btrfs_interface_init(void)
1779 {
1780         return misc_register(&btrfs_misc);
1781 }
1782
1783 static void btrfs_interface_exit(void)
1784 {
1785         if (misc_deregister(&btrfs_misc) < 0)
1786                 printk(KERN_INFO "BTRFS: misc_deregister failed for control device\n");
1787 }
1788
1789 static void btrfs_print_info(void)
1790 {
1791         printk(KERN_INFO "Btrfs loaded"
1792 #ifdef CONFIG_BTRFS_DEBUG
1793                         ", debug=on"
1794 #endif
1795 #ifdef CONFIG_BTRFS_ASSERT
1796                         ", assert=on"
1797 #endif
1798 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1799                         ", integrity-checker=on"
1800 #endif
1801                         "\n");
1802 }
1803
1804 static int btrfs_run_sanity_tests(void)
1805 {
1806         int ret;
1807
1808         ret = btrfs_init_test_fs();
1809         if (ret)
1810                 return ret;
1811
1812         ret = btrfs_test_free_space_cache();
1813         if (ret)
1814                 goto out;
1815         ret = btrfs_test_extent_buffer_operations();
1816         if (ret)
1817                 goto out;
1818         ret = btrfs_test_extent_io();
1819         if (ret)
1820                 goto out;
1821         ret = btrfs_test_inodes();
1822 out:
1823         btrfs_destroy_test_fs();
1824         return ret;
1825 }
1826
1827 static int __init init_btrfs_fs(void)
1828 {
1829         int err;
1830
1831         err = btrfs_init_sysfs();
1832         if (err)
1833                 return err;
1834
1835         btrfs_init_compress();
1836
1837         err = btrfs_init_cachep();
1838         if (err)
1839                 goto free_compress;
1840
1841         err = extent_io_init();
1842         if (err)
1843                 goto free_cachep;
1844
1845         err = extent_map_init();
1846         if (err)
1847                 goto free_extent_io;
1848
1849         err = ordered_data_init();
1850         if (err)
1851                 goto free_extent_map;
1852
1853         err = btrfs_delayed_inode_init();
1854         if (err)
1855                 goto free_ordered_data;
1856
1857         err = btrfs_auto_defrag_init();
1858         if (err)
1859                 goto free_delayed_inode;
1860
1861         err = btrfs_delayed_ref_init();
1862         if (err)
1863                 goto free_auto_defrag;
1864
1865         err = btrfs_prelim_ref_init();
1866         if (err)
1867                 goto free_prelim_ref;
1868
1869         err = btrfs_interface_init();
1870         if (err)
1871                 goto free_delayed_ref;
1872
1873         btrfs_init_lockdep();
1874
1875         btrfs_print_info();
1876
1877         err = btrfs_run_sanity_tests();
1878         if (err)
1879                 goto unregister_ioctl;
1880
1881         err = register_filesystem(&btrfs_fs_type);
1882         if (err)
1883                 goto unregister_ioctl;
1884
1885         return 0;
1886
1887 unregister_ioctl:
1888         btrfs_interface_exit();
1889 free_prelim_ref:
1890         btrfs_prelim_ref_exit();
1891 free_delayed_ref:
1892         btrfs_delayed_ref_exit();
1893 free_auto_defrag:
1894         btrfs_auto_defrag_exit();
1895 free_delayed_inode:
1896         btrfs_delayed_inode_exit();
1897 free_ordered_data:
1898         ordered_data_exit();
1899 free_extent_map:
1900         extent_map_exit();
1901 free_extent_io:
1902         extent_io_exit();
1903 free_cachep:
1904         btrfs_destroy_cachep();
1905 free_compress:
1906         btrfs_exit_compress();
1907         btrfs_exit_sysfs();
1908         return err;
1909 }
1910
1911 static void __exit exit_btrfs_fs(void)
1912 {
1913         btrfs_destroy_cachep();
1914         btrfs_delayed_ref_exit();
1915         btrfs_auto_defrag_exit();
1916         btrfs_delayed_inode_exit();
1917         btrfs_prelim_ref_exit();
1918         ordered_data_exit();
1919         extent_map_exit();
1920         extent_io_exit();
1921         btrfs_interface_exit();
1922         unregister_filesystem(&btrfs_fs_type);
1923         btrfs_exit_sysfs();
1924         btrfs_cleanup_fs_uuids();
1925         btrfs_exit_compress();
1926 }
1927
1928 module_init(init_btrfs_fs)
1929 module_exit(exit_btrfs_fs)
1930
1931 MODULE_LICENSE("GPL");