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/mnt_namespace.h>
  44 #include <linux/ratelimit.h>
  45 #include "compat.h"
  46 #include "delayed-inode.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "ioctl.h"
  52 #include "print-tree.h"
  53 #include "xattr.h"
  54 #include "volumes.h"
  55 #include "version.h"
  56 #include "export.h"
  57 #include "compression.h"
  58
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/btrfs.h>
  61
  62 static const struct super_operations btrfs_super_ops;
  63 static struct file_system_type btrfs_fs_type;
  64
  65 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  66                                       char nbuf[16])
  67 {
  68         char *errstr = NULL;
  69
  70         switch (errno) {
  71         case -EIO:
  72                 errstr = "IO failure";
  73                 break;
  74         case -ENOMEM:
  75                 errstr = "Out of memory";
  76                 break;
  77         case -EROFS:
  78                 errstr = "Readonly filesystem";
  79                 break;
  80         default:
  81                 if (nbuf) {
  82                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  83                                 errstr = nbuf;
  84                 }
  85                 break;
  86         }
  87
  88         return errstr;
  89 }
  90
  91 static void __save_error_info(struct btrfs_fs_info *fs_info)
  92 {
  93         /*
  94          * today we only save the error info into ram.  Long term we'll
  95          * also send it down to the disk
  96          */
  97         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
  98 }
  99
 100 /* NOTE:
 101  *      We move write_super stuff at umount in order to avoid deadlock
 102  *      for umount hold all lock.
 103  */
 104 static void save_error_info(struct btrfs_fs_info *fs_info)
 105 {
 106         __save_error_info(fs_info);
 107 }
 108
 109 /* btrfs handle error by forcing the filesystem readonly */
 110 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 111 {
 112         struct super_block *sb = fs_info->sb;
 113
 114         if (sb->s_flags & MS_RDONLY)
 115                 return;
 116
 117         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 118                 sb->s_flags |= MS_RDONLY;
 119                 printk(KERN_INFO "btrfs is forced readonly\n");
 120         }
 121 }
 122
 123 /*
 124  * __btrfs_std_error decodes expected errors from the caller and
 125  * invokes the approciate error response.
 126  */
 127 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 128                      unsigned int line, int errno)
 129 {
 130         struct super_block *sb = fs_info->sb;
 131         char nbuf[16];
 132         const char *errstr;
 133
 134         /*
 135          * Special case: if the error is EROFS, and we're already
 136          * under MS_RDONLY, then it is safe here.
 137          */
 138         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 139                 return;
 140
 141         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 142         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 143                 sb->s_id, function, line, errstr);
 144         save_error_info(fs_info);
 145
 146         btrfs_handle_error(fs_info);
 147 }
 148
 149 static void btrfs_put_super(struct super_block *sb)
 150 {
 151         struct btrfs_root *root = btrfs_sb(sb);
 152         int ret;
 153
 154         ret = close_ctree(root);
 155         sb->s_fs_info = NULL;
 156
 157         (void)ret; /* FIXME: need to fix VFS to return error? */
 158 }
 159
 160 enum {
 161         Opt_degraded, Opt_subvol, Opt_subvolid, Opt_device, Opt_nodatasum,
 162         Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
 163         Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
 164         Opt_compress_type, Opt_compress_force, Opt_compress_force_type,
 165         Opt_notreelog, Opt_ratio, Opt_flushoncommit, Opt_discard,
 166         Opt_space_cache, Opt_clear_cache, Opt_user_subvol_rm_allowed,
 167         Opt_enospc_debug, Opt_subvolrootid, Opt_defrag,
 168         Opt_inode_cache, Opt_no_space_cache, Opt_recovery, Opt_err,
 169 };
 170
 171 static match_table_t tokens = {
 172         {Opt_degraded, "degraded"},
 173         {Opt_subvol, "subvol=%s"},
 174         {Opt_subvolid, "subvolid=%d"},
 175         {Opt_device, "device=%s"},
 176         {Opt_nodatasum, "nodatasum"},
 177         {Opt_nodatacow, "nodatacow"},
 178         {Opt_nobarrier, "nobarrier"},
 179         {Opt_max_inline, "max_inline=%s"},
 180         {Opt_alloc_start, "alloc_start=%s"},
 181         {Opt_thread_pool, "thread_pool=%d"},
 182         {Opt_compress, "compress"},
 183         {Opt_compress_type, "compress=%s"},
 184         {Opt_compress_force, "compress-force"},
 185         {Opt_compress_force_type, "compress-force=%s"},
 186         {Opt_ssd, "ssd"},
 187         {Opt_ssd_spread, "ssd_spread"},
 188         {Opt_nossd, "nossd"},
 189         {Opt_noacl, "noacl"},
 190         {Opt_notreelog, "notreelog"},
 191         {Opt_flushoncommit, "flushoncommit"},
 192         {Opt_ratio, "metadata_ratio=%d"},
 193         {Opt_discard, "discard"},
 194         {Opt_space_cache, "space_cache"},
 195         {Opt_clear_cache, "clear_cache"},
 196         {Opt_user_subvol_rm_allowed, "user_subvol_rm_allowed"},
 197         {Opt_enospc_debug, "enospc_debug"},
 198         {Opt_subvolrootid, "subvolrootid=%d"},
 199         {Opt_defrag, "autodefrag"},
 200         {Opt_inode_cache, "inode_cache"},
 201         {Opt_no_space_cache, "nospace_cache"},
 202         {Opt_recovery, "recovery"},
 203         {Opt_err, NULL},
 204 };
 205
 206 /*
 207  * Regular mount options parser.  Everything that is needed only when
 208  * reading in a new superblock is parsed here.
 209  */
 210 int btrfs_parse_options(struct btrfs_root *root, char *options)
 211 {
 212         struct btrfs_fs_info *info = root->fs_info;
 213         substring_t args[MAX_OPT_ARGS];
 214         char *p, *num, *orig = NULL;
 215         u64 cache_gen;
 216         int intarg;
 217         int ret = 0;
 218         char *compress_type;
 219         bool compress_force = false;
 220
 221         cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
 222         if (cache_gen)
 223                 btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 224
 225         if (!options)
 226                 goto out;
 227
 228         /*
 229          * strsep changes the string, duplicate it because parse_options
 230          * gets called twice
 231          */
 232         options = kstrdup(options, GFP_NOFS);
 233         if (!options)
 234                 return -ENOMEM;
 235
 236         orig = options;
 237
 238         while ((p = strsep(&options, ",")) != NULL) {
 239                 int token;
 240                 if (!*p)
 241                         continue;
 242
 243                 token = match_token(p, tokens, args);
 244                 switch (token) {
 245                 case Opt_degraded:
 246                         printk(KERN_INFO "btrfs: allowing degraded mounts\n");
 247                         btrfs_set_opt(info->mount_opt, DEGRADED);
 248                         break;
 249                 case Opt_subvol:
 250                 case Opt_subvolid:
 251                 case Opt_subvolrootid:
 252                 case Opt_device:
 253                         /*
 254                          * These are parsed by btrfs_parse_early_options
 255                          * and can be happily ignored here.
 256                          */
 257                         break;
 258                 case Opt_nodatasum:
 259                         printk(KERN_INFO "btrfs: setting nodatasum\n");
 260                         btrfs_set_opt(info->mount_opt, NODATASUM);
 261                         break;
 262                 case Opt_nodatacow:
 263                         printk(KERN_INFO "btrfs: setting nodatacow\n");
 264                         btrfs_set_opt(info->mount_opt, NODATACOW);
 265                         btrfs_set_opt(info->mount_opt, NODATASUM);
 266                         break;
 267                 case Opt_compress_force:
 268                 case Opt_compress_force_type:
 269                         compress_force = true;
 270                 case Opt_compress:
 271                 case Opt_compress_type:
 272                         if (token == Opt_compress ||
 273                             token == Opt_compress_force ||
 274                             strcmp(args[0].from, "zlib") == 0) {
 275                                 compress_type = "zlib";
 276                                 info->compress_type = BTRFS_COMPRESS_ZLIB;
 277                         } else if (strcmp(args[0].from, "lzo") == 0) {
 278                                 compress_type = "lzo";
 279                                 info->compress_type = BTRFS_COMPRESS_LZO;
 280                         } else {
 281                                 ret = -EINVAL;
 282                                 goto out;
 283                         }
 284
 285                         btrfs_set_opt(info->mount_opt, COMPRESS);
 286                         if (compress_force) {
 287                                 btrfs_set_opt(info->mount_opt, FORCE_COMPRESS);
 288                                 pr_info("btrfs: force %s compression\n",
 289                                         compress_type);
 290                         } else
 291                                 pr_info("btrfs: use %s compression\n",
 292                                         compress_type);
 293                         break;
 294                 case Opt_ssd:
 295                         printk(KERN_INFO "btrfs: use ssd allocation scheme\n");
 296                         btrfs_set_opt(info->mount_opt, SSD);
 297                         break;
 298                 case Opt_ssd_spread:
 299                         printk(KERN_INFO "btrfs: use spread ssd "
 300                                "allocation scheme\n");
 301                         btrfs_set_opt(info->mount_opt, SSD);
 302                         btrfs_set_opt(info->mount_opt, SSD_SPREAD);
 303                         break;
 304                 case Opt_nossd:
 305                         printk(KERN_INFO "btrfs: not using ssd allocation "
 306                                "scheme\n");
 307                         btrfs_set_opt(info->mount_opt, NOSSD);
 308                         btrfs_clear_opt(info->mount_opt, SSD);
 309                         btrfs_clear_opt(info->mount_opt, SSD_SPREAD);
 310                         break;
 311                 case Opt_nobarrier:
 312                         printk(KERN_INFO "btrfs: turning off barriers\n");
 313                         btrfs_set_opt(info->mount_opt, NOBARRIER);
 314                         break;
 315                 case Opt_thread_pool:
 316                         intarg = 0;
 317                         match_int(&args[0], &intarg);
 318                         if (intarg) {
 319                                 info->thread_pool_size = intarg;
 320                                 printk(KERN_INFO "btrfs: thread pool %d\n",
 321                                        info->thread_pool_size);
 322                         }
 323                         break;
 324                 case Opt_max_inline:
 325                         num = match_strdup(&args[0]);
 326                         if (num) {
 327                                 info->max_inline = memparse(num, NULL);
 328                                 kfree(num);
 329
 330                                 if (info->max_inline) {
 331                                         info->max_inline = max_t(u64,
 332                                                 info->max_inline,
 333                                                 root->sectorsize);
 334                                 }
 335                                 printk(KERN_INFO "btrfs: max_inline at %llu\n",
 336                                         (unsigned long long)info->max_inline);
 337                         }
 338                         break;
 339                 case Opt_alloc_start:
 340                         num = match_strdup(&args[0]);
 341                         if (num) {
 342                                 info->alloc_start = memparse(num, NULL);
 343                                 kfree(num);
 344                                 printk(KERN_INFO
 345                                         "btrfs: allocations start at %llu\n",
 346                                         (unsigned long long)info->alloc_start);
 347                         }
 348                         break;
 349                 case Opt_noacl:
 350                         root->fs_info->sb->s_flags &= ~MS_POSIXACL;
 351                         break;
 352                 case Opt_notreelog:
 353                         printk(KERN_INFO "btrfs: disabling tree log\n");
 354                         btrfs_set_opt(info->mount_opt, NOTREELOG);
 355                         break;
 356                 case Opt_flushoncommit:
 357                         printk(KERN_INFO "btrfs: turning on flush-on-commit\n");
 358                         btrfs_set_opt(info->mount_opt, FLUSHONCOMMIT);
 359                         break;
 360                 case Opt_ratio:
 361                         intarg = 0;
 362                         match_int(&args[0], &intarg);
 363                         if (intarg) {
 364                                 info->metadata_ratio = intarg;
 365                                 printk(KERN_INFO "btrfs: metadata ratio %d\n",
 366                                        info->metadata_ratio);
 367                         }
 368                         break;
 369                 case Opt_discard:
 370                         btrfs_set_opt(info->mount_opt, DISCARD);
 371                         break;
 372                 case Opt_space_cache:
 373                         btrfs_set_opt(info->mount_opt, SPACE_CACHE);
 374                         break;
 375                 case Opt_no_space_cache:
 376                         printk(KERN_INFO "btrfs: disabling disk space caching\n");
 377                         btrfs_clear_opt(info->mount_opt, SPACE_CACHE);
 378                         break;
 379                 case Opt_inode_cache:
 380                         printk(KERN_INFO "btrfs: enabling inode map caching\n");
 381                         btrfs_set_opt(info->mount_opt, INODE_MAP_CACHE);
 382                         break;
 383                 case Opt_clear_cache:
 384                         printk(KERN_INFO "btrfs: force clearing of disk cache\n");
 385                         btrfs_set_opt(info->mount_opt, CLEAR_CACHE);
 386                         break;
 387                 case Opt_user_subvol_rm_allowed:
 388                         btrfs_set_opt(info->mount_opt, USER_SUBVOL_RM_ALLOWED);
 389                         break;
 390                 case Opt_enospc_debug:
 391                         btrfs_set_opt(info->mount_opt, ENOSPC_DEBUG);
 392                         break;
 393                 case Opt_defrag:
 394                         printk(KERN_INFO "btrfs: enabling auto defrag");
 395                         btrfs_set_opt(info->mount_opt, AUTO_DEFRAG);
 396                         break;
 397                 case Opt_recovery:
 398                         printk(KERN_INFO "btrfs: enabling auto recovery");
 399                         btrfs_set_opt(info->mount_opt, RECOVERY);
 400                         break;
 401                 case Opt_err:
 402                         printk(KERN_INFO "btrfs: unrecognized mount option "
 403                                "'%s'\n", p);
 404                         ret = -EINVAL;
 405                         goto out;
 406                 default:
 407                         break;
 408                 }
 409         }
 410 out:
 411         if (!ret && btrfs_test_opt(root, SPACE_CACHE))
 412                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
 413         kfree(orig);
 414         return ret;
 415 }
 416
 417 /*
 418  * Parse mount options that are required early in the mount process.
 419  *
 420  * All other options will be parsed on much later in the mount process and
 421  * only when we need to allocate a new super block.
 422  */
 423 static int btrfs_parse_early_options(const char *options, fmode_t flags,
 424                 void *holder, char **subvol_name, u64 *subvol_objectid,
 425                 u64 *subvol_rootid, struct btrfs_fs_devices **fs_devices)
 426 {
 427         substring_t args[MAX_OPT_ARGS];
 428         char *device_name, *opts, *orig, *p;
 429         int error = 0;
 430         int intarg;
 431
 432         if (!options)
 433                 return 0;
 434
 435         /*
 436          * strsep changes the string, duplicate it because parse_options
 437          * gets called twice
 438          */
 439         opts = kstrdup(options, GFP_KERNEL);
 440         if (!opts)
 441                 return -ENOMEM;
 442         orig = opts;
 443
 444         while ((p = strsep(&opts, ",")) != NULL) {
 445                 int token;
 446                 if (!*p)
 447                         continue;
 448
 449                 token = match_token(p, tokens, args);
 450                 switch (token) {
 451                 case Opt_subvol:
 452                         kfree(*subvol_name);
 453                         *subvol_name = match_strdup(&args[0]);
 454                         break;
 455                 case Opt_subvolid:
 456                         intarg = 0;
 457                         error = match_int(&args[0], &intarg);
 458                         if (!error) {
 459                                 /* we want the original fs_tree */
 460                                 if (!intarg)
 461                                         *subvol_objectid =
 462                                                 BTRFS_FS_TREE_OBJECTID;
 463                                 else
 464                                         *subvol_objectid = intarg;
 465                         }
 466                         break;
 467                 case Opt_subvolrootid:
 468                         intarg = 0;
 469                         error = match_int(&args[0], &intarg);
 470                         if (!error) {
 471                                 /* we want the original fs_tree */
 472                                 if (!intarg)
 473                                         *subvol_rootid =
 474                                                 BTRFS_FS_TREE_OBJECTID;
 475                                 else
 476                                         *subvol_rootid = intarg;
 477                         }
 478                         break;
 479                 case Opt_device:
 480                         device_name = match_strdup(&args[0]);
 481                         if (!device_name) {
 482                                 error = -ENOMEM;
 483                                 goto out;
 484                         }
 485                         error = btrfs_scan_one_device(device_name,
 486                                         flags, holder, fs_devices);
 487                         kfree(device_name);
 488                         if (error)
 489                                 goto out;
 490                         break;
 491                 default:
 492                         break;
 493                 }
 494         }
 495
 496 out:
 497         kfree(orig);
 498         return error;
 499 }
 500
 501 static struct dentry *get_default_root(struct super_block *sb,
 502                                        u64 subvol_objectid)
 503 {
 504         struct btrfs_root *root = sb->s_fs_info;
 505         struct btrfs_root *new_root;
 506         struct btrfs_dir_item *di;
 507         struct btrfs_path *path;
 508         struct btrfs_key location;
 509         struct inode *inode;
 510         u64 dir_id;
 511         int new = 0;
 512
 513         /*
 514          * We have a specific subvol we want to mount, just setup location and
 515          * go look up the root.
 516          */
 517         if (subvol_objectid) {
 518                 location.objectid = subvol_objectid;
 519                 location.type = BTRFS_ROOT_ITEM_KEY;
 520                 location.offset = (u64)-1;
 521                 goto find_root;
 522         }
 523
 524         path = btrfs_alloc_path();
 525         if (!path)
 526                 return ERR_PTR(-ENOMEM);
 527         path->leave_spinning = 1;
 528
 529         /*
 530          * Find the "default" dir item which points to the root item that we
 531          * will mount by default if we haven't been given a specific subvolume
 532          * to mount.
 533          */
 534         dir_id = btrfs_super_root_dir(root->fs_info->super_copy);
 535         di = btrfs_lookup_dir_item(NULL, root, path, dir_id, "default", 7, 0);
 536         if (IS_ERR(di)) {
 537                 btrfs_free_path(path);
 538                 return ERR_CAST(di);
 539         }
 540         if (!di) {
 541                 /*
 542                  * Ok the default dir item isn't there.  This is weird since
 543                  * it's always been there, but don't freak out, just try and
 544                  * mount to root most subvolume.
 545                  */
 546                 btrfs_free_path(path);
 547                 dir_id = BTRFS_FIRST_FREE_OBJECTID;
 548                 new_root = root->fs_info->fs_root;
 549                 goto setup_root;
 550         }
 551
 552         btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 553         btrfs_free_path(path);
 554
 555 find_root:
 556         new_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
 557         if (IS_ERR(new_root))
 558                 return ERR_CAST(new_root);
 559
 560         if (btrfs_root_refs(&new_root->root_item) == 0)
 561                 return ERR_PTR(-ENOENT);
 562
 563         dir_id = btrfs_root_dirid(&new_root->root_item);
 564 setup_root:
 565         location.objectid = dir_id;
 566         location.type = BTRFS_INODE_ITEM_KEY;
 567         location.offset = 0;
 568
 569         inode = btrfs_iget(sb, &location, new_root, &new);
 570         if (IS_ERR(inode))
 571                 return ERR_CAST(inode);
 572
 573         /*
 574          * If we're just mounting the root most subvol put the inode and return
 575          * a reference to the dentry.  We will have already gotten a reference
 576          * to the inode in btrfs_fill_super so we're good to go.
 577          */
 578         if (!new && sb->s_root->d_inode == inode) {
 579                 iput(inode);
 580                 return dget(sb->s_root);
 581         }
 582
 583         return d_obtain_alias(inode);
 584 }
 585
 586 static int btrfs_fill_super(struct super_block *sb,
 587                             struct btrfs_fs_devices *fs_devices,
 588                             void *data, int silent)
 589 {
 590         struct inode *inode;
 591         struct dentry *root_dentry;
 592         struct btrfs_root *tree_root;
 593         struct btrfs_key key;
 594         int err;
 595
 596         sb->s_maxbytes = MAX_LFS_FILESIZE;
 597         sb->s_magic = BTRFS_SUPER_MAGIC;
 598         sb->s_op = &btrfs_super_ops;
 599         sb->s_d_op = &btrfs_dentry_operations;
 600         sb->s_export_op = &btrfs_export_ops;
 601         sb->s_xattr = btrfs_xattr_handlers;
 602         sb->s_time_gran = 1;
 603 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 604         sb->s_flags |= MS_POSIXACL;
 605 #endif
 606
 607         tree_root = open_ctree(sb, fs_devices, (char *)data);
 608
 609         if (IS_ERR(tree_root)) {
 610                 printk("btrfs: open_ctree failed\n");
 611                 return PTR_ERR(tree_root);
 612         }
 613         sb->s_fs_info = tree_root;
 614
 615         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 616         key.type = BTRFS_INODE_ITEM_KEY;
 617         key.offset = 0;
 618         inode = btrfs_iget(sb, &key, tree_root->fs_info->fs_root, NULL);
 619         if (IS_ERR(inode)) {
 620                 err = PTR_ERR(inode);
 621                 goto fail_close;
 622         }
 623
 624         root_dentry = d_alloc_root(inode);
 625         if (!root_dentry) {
 626                 iput(inode);
 627                 err = -ENOMEM;
 628                 goto fail_close;
 629         }
 630
 631         sb->s_root = root_dentry;
 632
 633         save_mount_options(sb, data);
 634         cleancache_init_fs(sb);
 635         return 0;
 636
 637 fail_close:
 638         close_ctree(tree_root);
 639         return err;
 640 }
 641
 642 int btrfs_sync_fs(struct super_block *sb, int wait)
 643 {
 644         struct btrfs_trans_handle *trans;
 645         struct btrfs_root *root = btrfs_sb(sb);
 646         int ret;
 647
 648         trace_btrfs_sync_fs(wait);
 649
 650         if (!wait) {
 651                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 652                 return 0;
 653         }
 654
 655         btrfs_start_delalloc_inodes(root, 0);
 656         btrfs_wait_ordered_extents(root, 0, 0);
 657
 658         trans = btrfs_start_transaction(root, 0);
 659         if (IS_ERR(trans))
 660                 return PTR_ERR(trans);
 661         ret = btrfs_commit_transaction(trans, root);
 662         return ret;
 663 }
 664
 665 static int btrfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
 666 {
 667         struct btrfs_root *root = btrfs_sb(vfs->mnt_sb);
 668         struct btrfs_fs_info *info = root->fs_info;
 669         char *compress_type;
 670
 671         if (btrfs_test_opt(root, DEGRADED))
 672                 seq_puts(seq, ",degraded");
 673         if (btrfs_test_opt(root, NODATASUM))
 674                 seq_puts(seq, ",nodatasum");
 675         if (btrfs_test_opt(root, NODATACOW))
 676                 seq_puts(seq, ",nodatacow");
 677         if (btrfs_test_opt(root, NOBARRIER))
 678                 seq_puts(seq, ",nobarrier");
 679         if (info->max_inline != 8192 * 1024)
 680                 seq_printf(seq, ",max_inline=%llu",
 681                            (unsigned long long)info->max_inline);
 682         if (info->alloc_start != 0)
 683                 seq_printf(seq, ",alloc_start=%llu",
 684                            (unsigned long long)info->alloc_start);
 685         if (info->thread_pool_size !=  min_t(unsigned long,
 686                                              num_online_cpus() + 2, 8))
 687                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 688         if (btrfs_test_opt(root, COMPRESS)) {
 689                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 690                         compress_type = "zlib";
 691                 else
 692                         compress_type = "lzo";
 693                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 694                         seq_printf(seq, ",compress-force=%s", compress_type);
 695                 else
 696                         seq_printf(seq, ",compress=%s", compress_type);
 697         }
 698         if (btrfs_test_opt(root, NOSSD))
 699                 seq_puts(seq, ",nossd");
 700         if (btrfs_test_opt(root, SSD_SPREAD))
 701                 seq_puts(seq, ",ssd_spread");
 702         else if (btrfs_test_opt(root, SSD))
 703                 seq_puts(seq, ",ssd");
 704         if (btrfs_test_opt(root, NOTREELOG))
 705                 seq_puts(seq, ",notreelog");
 706         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 707                 seq_puts(seq, ",flushoncommit");
 708         if (btrfs_test_opt(root, DISCARD))
 709                 seq_puts(seq, ",discard");
 710         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 711                 seq_puts(seq, ",noacl");
 712         if (btrfs_test_opt(root, SPACE_CACHE))
 713                 seq_puts(seq, ",space_cache");
 714         else
 715                 seq_puts(seq, ",nospace_cache");
 716         if (btrfs_test_opt(root, CLEAR_CACHE))
 717                 seq_puts(seq, ",clear_cache");
 718         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 719                 seq_puts(seq, ",user_subvol_rm_allowed");
 720         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 721                 seq_puts(seq, ",enospc_debug");
 722         if (btrfs_test_opt(root, AUTO_DEFRAG))
 723                 seq_puts(seq, ",autodefrag");
 724         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 725                 seq_puts(seq, ",inode_cache");
 726         return 0;
 727 }
 728
 729 static int btrfs_test_super(struct super_block *s, void *data)
 730 {
 731         struct btrfs_root *test_root = data;
 732         struct btrfs_root *root = btrfs_sb(s);
 733
 734         /*
 735          * If this super block is going away, return false as it
 736          * can't match as an existing super block.
 737          */
 738         if (!atomic_read(&s->s_active))
 739                 return 0;
 740         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 741 }
 742
 743 static int btrfs_set_super(struct super_block *s, void *data)
 744 {
 745         s->s_fs_info = data;
 746
 747         return set_anon_super(s, data);
 748 }
 749
 750 /*
 751  * subvolumes are identified by ino 256
 752  */
 753 static inline int is_subvolume_inode(struct inode *inode)
 754 {
 755         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 756                 return 1;
 757         return 0;
 758 }
 759
 760 /*
 761  * This will strip out the subvol=%s argument for an argument string and add
 762  * subvolid=0 to make sure we get the actual tree root for path walking to the
 763  * subvol we want.
 764  */
 765 static char *setup_root_args(char *args)
 766 {
 767         unsigned copied = 0;
 768         unsigned len = strlen(args) + 2;
 769         char *pos;
 770         char *ret;
 771
 772         /*
 773          * We need the same args as before, but minus
 774          *
 775          * subvol=a
 776          *
 777          * and add
 778          *
 779          * subvolid=0
 780          *
 781          * which is a difference of 2 characters, so we allocate strlen(args) +
 782          * 2 characters.
 783          */
 784         ret = kzalloc(len * sizeof(char), GFP_NOFS);
 785         if (!ret)
 786                 return NULL;
 787         pos = strstr(args, "subvol=");
 788
 789         /* This shouldn't happen, but just in case.. */
 790         if (!pos) {
 791                 kfree(ret);
 792                 return NULL;
 793         }
 794
 795         /*
 796          * The subvol=<> arg is not at the front of the string, copy everybody
 797          * up to that into ret.
 798          */
 799         if (pos != args) {
 800                 *pos = '\0';
 801                 strcpy(ret, args);
 802                 copied += strlen(args);
 803                 pos++;
 804         }
 805
 806         strncpy(ret + copied, "subvolid=0", len - copied);
 807
 808         /* Length of subvolid=0 */
 809         copied += 10;
 810
 811         /*
 812          * If there is no , after the subvol= option then we know there's no
 813          * other options and we can just return.
 814          */
 815         pos = strchr(pos, ',');
 816         if (!pos)
 817                 return ret;
 818
 819         /* Copy the rest of the arguments into our buffer */
 820         strncpy(ret + copied, pos, len - copied);
 821         copied += strlen(pos);
 822
 823         return ret;
 824 }
 825
 826 static struct dentry *mount_subvol(const char *subvol_name, int flags,
 827                                    const char *device_name, char *data)
 828 {
 829         struct dentry *root;
 830         struct vfsmount *mnt;
 831         char *newargs;
 832
 833         newargs = setup_root_args(data);
 834         if (!newargs)
 835                 return ERR_PTR(-ENOMEM);
 836         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
 837                              newargs);
 838         kfree(newargs);
 839         if (IS_ERR(mnt))
 840                 return ERR_CAST(mnt);
 841
 842         root = mount_subtree(mnt, subvol_name);
 843
 844         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
 845                 struct super_block *s = root->d_sb;
 846                 dput(root);
 847                 root = ERR_PTR(-EINVAL);
 848                 deactivate_locked_super(s);
 849                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
 850                                 subvol_name);
 851         }
 852
 853         return root;
 854 }
 855
 856 /*
 857  * Find a superblock for the given device / mount point.
 858  *
 859  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 860  *        for multiple device setup.  Make sure to keep it in sync.
 861  */
 862 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
 863                 const char *device_name, void *data)
 864 {
 865         struct block_device *bdev = NULL;
 866         struct super_block *s;
 867         struct dentry *root;
 868         struct btrfs_fs_devices *fs_devices = NULL;
 869         struct btrfs_fs_info *fs_info = NULL;
 870         fmode_t mode = FMODE_READ;
 871         char *subvol_name = NULL;
 872         u64 subvol_objectid = 0;
 873         u64 subvol_rootid = 0;
 874         int error = 0;
 875
 876         if (!(flags & MS_RDONLY))
 877                 mode |= FMODE_WRITE;
 878
 879         error = btrfs_parse_early_options(data, mode, fs_type,
 880                                           &subvol_name, &subvol_objectid,
 881                                           &subvol_rootid, &fs_devices);
 882         if (error) {
 883                 kfree(subvol_name);
 884                 return ERR_PTR(error);
 885         }
 886
 887         if (subvol_name) {
 888                 root = mount_subvol(subvol_name, flags, device_name, data);
 889                 kfree(subvol_name);
 890                 return root;
 891         }
 892
 893         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
 894         if (error)
 895                 return ERR_PTR(error);
 896
 897         /*
 898          * Setup a dummy root and fs_info for test/set super.  This is because
 899          * we don't actually fill this stuff out until open_ctree, but we need
 900          * it for searching for existing supers, so this lets us do that and
 901          * then open_ctree will properly initialize everything later.
 902          */
 903         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 904         if (!fs_info)
 905                 return ERR_PTR(-ENOMEM);
 906
 907         fs_info->tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 908         if (!fs_info->tree_root) {
 909                 error = -ENOMEM;
 910                 goto error_fs_info;
 911         }
 912         fs_info->tree_root->fs_info = fs_info;
 913         fs_info->fs_devices = fs_devices;
 914
 915         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
 916         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
 917         if (!fs_info->super_copy || !fs_info->super_for_commit) {
 918                 error = -ENOMEM;
 919                 goto error_fs_info;
 920         }
 921
 922         error = btrfs_open_devices(fs_devices, mode, fs_type);
 923         if (error)
 924                 goto error_fs_info;
 925
 926         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 927                 error = -EACCES;
 928                 goto error_close_devices;
 929         }
 930
 931         bdev = fs_devices->latest_bdev;
 932         s = sget(fs_type, btrfs_test_super, btrfs_set_super,
 933                  fs_info->tree_root);
 934         if (IS_ERR(s)) {
 935                 error = PTR_ERR(s);
 936                 goto error_close_devices;
 937         }
 938
 939         if (s->s_root) {
 940                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 941                         deactivate_locked_super(s);
 942                         error = -EBUSY;
 943                         goto error_close_devices;
 944                 }
 945
 946                 btrfs_close_devices(fs_devices);
 947                 free_fs_info(fs_info);
 948         } else {
 949                 char b[BDEVNAME_SIZE];
 950
 951                 s->s_flags = flags | MS_NOSEC;
 952                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 953                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 954                 error = btrfs_fill_super(s, fs_devices, data,
 955                                          flags & MS_SILENT ? 1 : 0);
 956                 if (error) {
 957                         deactivate_locked_super(s);
 958                         return ERR_PTR(error);
 959                 }
 960
 961                 s->s_flags |= MS_ACTIVE;
 962         }
 963
 964         root = get_default_root(s, subvol_objectid);
 965         if (IS_ERR(root)) {
 966                 deactivate_locked_super(s);
 967                 return root;
 968         }
 969
 970         return root;
 971
 972 error_close_devices:
 973         btrfs_close_devices(fs_devices);
 974 error_fs_info:
 975         free_fs_info(fs_info);
 976         return ERR_PTR(error);
 977 }
 978
 979 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 980 {
 981         struct btrfs_root *root = btrfs_sb(sb);
 982         int ret;
 983
 984         ret = btrfs_parse_options(root, data);
 985         if (ret)
 986                 return -EINVAL;
 987
 988         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 989                 return 0;
 990
 991         if (*flags & MS_RDONLY) {
 992                 sb->s_flags |= MS_RDONLY;
 993
 994                 ret =  btrfs_commit_super(root);
 995                 WARN_ON(ret);
 996         } else {
 997                 if (root->fs_info->fs_devices->rw_devices == 0)
 998                         return -EACCES;
 999
1000                 if (btrfs_super_log_root(root->fs_info->super_copy) != 0)
1001                         return -EINVAL;
1002
1003                 ret = btrfs_cleanup_fs_roots(root->fs_info);
1004                 WARN_ON(ret);
1005
1006                 /* recover relocation */
1007                 ret = btrfs_recover_relocation(root);
1008                 WARN_ON(ret);
1009
1010                 sb->s_flags &= ~MS_RDONLY;
1011         }
1012
1013         return 0;
1014 }
1015
1016 /* Used to sort the devices by max_avail(descending sort) */
1017 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1018                                        const void *dev_info2)
1019 {
1020         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1021             ((struct btrfs_device_info *)dev_info2)->max_avail)
1022                 return -1;
1023         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1024                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1025                 return 1;
1026         else
1027         return 0;
1028 }
1029
1030 /*
1031  * sort the devices by max_avail, in which max free extent size of each device
1032  * is stored.(Descending Sort)
1033  */
1034 static inline void btrfs_descending_sort_devices(
1035                                         struct btrfs_device_info *devices,
1036                                         size_t nr_devices)
1037 {
1038         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1039              btrfs_cmp_device_free_bytes, NULL);
1040 }
1041
1042 /*
1043  * The helper to calc the free space on the devices that can be used to store
1044  * file data.
1045  */
1046 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1047 {
1048         struct btrfs_fs_info *fs_info = root->fs_info;
1049         struct btrfs_device_info *devices_info;
1050         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1051         struct btrfs_device *device;
1052         u64 skip_space;
1053         u64 type;
1054         u64 avail_space;
1055         u64 used_space;
1056         u64 min_stripe_size;
1057         int min_stripes = 1, num_stripes = 1;
1058         int i = 0, nr_devices;
1059         int ret;
1060
1061         nr_devices = fs_info->fs_devices->open_devices;
1062         BUG_ON(!nr_devices);
1063
1064         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1065                                GFP_NOFS);
1066         if (!devices_info)
1067                 return -ENOMEM;
1068
1069         /* calc min stripe number for data space alloction */
1070         type = btrfs_get_alloc_profile(root, 1);
1071         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1072                 min_stripes = 2;
1073                 num_stripes = nr_devices;
1074         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1075                 min_stripes = 2;
1076                 num_stripes = 2;
1077         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1078                 min_stripes = 4;
1079                 num_stripes = 4;
1080         }
1081
1082         if (type & BTRFS_BLOCK_GROUP_DUP)
1083                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1084         else
1085                 min_stripe_size = BTRFS_STRIPE_LEN;
1086
1087         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1088                 if (!device->in_fs_metadata || !device->bdev)
1089                         continue;
1090
1091                 avail_space = device->total_bytes - device->bytes_used;
1092
1093                 /* align with stripe_len */
1094                 do_div(avail_space, BTRFS_STRIPE_LEN);
1095                 avail_space *= BTRFS_STRIPE_LEN;
1096
1097                 /*
1098                  * In order to avoid overwritting the superblock on the drive,
1099                  * btrfs starts at an offset of at least 1MB when doing chunk
1100                  * allocation.
1101                  */
1102                 skip_space = 1024 * 1024;
1103
1104                 /* user can set the offset in fs_info->alloc_start. */
1105                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1106                     device->total_bytes)
1107                         skip_space = max(fs_info->alloc_start, skip_space);
1108
1109                 /*
1110                  * btrfs can not use the free space in [0, skip_space - 1],
1111                  * we must subtract it from the total. In order to implement
1112                  * it, we account the used space in this range first.
1113                  */
1114                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1115                                                      &used_space);
1116                 if (ret) {
1117                         kfree(devices_info);
1118                         return ret;
1119                 }
1120
1121                 /* calc the free space in [0, skip_space - 1] */
1122                 skip_space -= used_space;
1123
1124                 /*
1125                  * we can use the free space in [0, skip_space - 1], subtract
1126                  * it from the total.
1127                  */
1128                 if (avail_space && avail_space >= skip_space)
1129                         avail_space -= skip_space;
1130                 else
1131                         avail_space = 0;
1132
1133                 if (avail_space < min_stripe_size)
1134                         continue;
1135
1136                 devices_info[i].dev = device;
1137                 devices_info[i].max_avail = avail_space;
1138
1139                 i++;
1140         }
1141
1142         nr_devices = i;
1143
1144         btrfs_descending_sort_devices(devices_info, nr_devices);
1145
1146         i = nr_devices - 1;
1147         avail_space = 0;
1148         while (nr_devices >= min_stripes) {
1149                 if (num_stripes > nr_devices)
1150                         num_stripes = nr_devices;
1151
1152                 if (devices_info[i].max_avail >= min_stripe_size) {
1153                         int j;
1154                         u64 alloc_size;
1155
1156                         avail_space += devices_info[i].max_avail * num_stripes;
1157                         alloc_size = devices_info[i].max_avail;
1158                         for (j = i + 1 - num_stripes; j <= i; j++)
1159                                 devices_info[j].max_avail -= alloc_size;
1160                 }
1161                 i--;
1162                 nr_devices--;
1163         }
1164
1165         kfree(devices_info);
1166         *free_bytes = avail_space;
1167         return 0;
1168 }
1169
1170 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1171 {
1172         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1173         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1174         struct list_head *head = &root->fs_info->space_info;
1175         struct btrfs_space_info *found;
1176         u64 total_used = 0;
1177         u64 total_free_data = 0;
1178         int bits = dentry->d_sb->s_blocksize_bits;
1179         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1180         int ret;
1181
1182         /* holding chunk_muext to avoid allocating new chunks */
1183         mutex_lock(&root->fs_info->chunk_mutex);
1184         rcu_read_lock();
1185         list_for_each_entry_rcu(found, head, list) {
1186                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1187                         total_free_data += found->disk_total - found->disk_used;
1188                         total_free_data -=
1189                                 btrfs_account_ro_block_groups_free_space(found);
1190                 }
1191
1192                 total_used += found->disk_used;
1193         }
1194         rcu_read_unlock();
1195
1196         buf->f_namelen = BTRFS_NAME_LEN;
1197         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1198         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1199         buf->f_bsize = dentry->d_sb->s_blocksize;
1200         buf->f_type = BTRFS_SUPER_MAGIC;
1201         buf->f_bavail = total_free_data;
1202         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1203         if (ret) {
1204                 mutex_unlock(&root->fs_info->chunk_mutex);
1205                 return ret;
1206         }
1207         buf->f_bavail += total_free_data;
1208         buf->f_bavail = buf->f_bavail >> bits;
1209         mutex_unlock(&root->fs_info->chunk_mutex);
1210
1211         /* We treat it as constant endianness (it doesn't matter _which_)
1212            because we want the fsid to come out the same whether mounted
1213            on a big-endian or little-endian host */
1214         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1215         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1216         /* Mask in the root object ID too, to disambiguate subvols */
1217         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1218         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1219
1220         return 0;
1221 }
1222
1223 static struct file_system_type btrfs_fs_type = {
1224         .owner          = THIS_MODULE,
1225         .name           = "btrfs",
1226         .mount          = btrfs_mount,
1227         .kill_sb        = kill_anon_super,
1228         .fs_flags       = FS_REQUIRES_DEV,
1229 };
1230
1231 /*
1232  * used by btrfsctl to scan devices when no FS is mounted
1233  */
1234 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1235                                 unsigned long arg)
1236 {
1237         struct btrfs_ioctl_vol_args *vol;
1238         struct btrfs_fs_devices *fs_devices;
1239         int ret = -ENOTTY;
1240
1241         if (!capable(CAP_SYS_ADMIN))
1242                 return -EPERM;
1243
1244         vol = memdup_user((void __user *)arg, sizeof(*vol));
1245         if (IS_ERR(vol))
1246                 return PTR_ERR(vol);
1247
1248         switch (cmd) {
1249         case BTRFS_IOC_SCAN_DEV:
1250                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1251                                             &btrfs_fs_type, &fs_devices);
1252                 break;
1253         }
1254
1255         kfree(vol);
1256         return ret;
1257 }
1258
1259 static int btrfs_freeze(struct super_block *sb)
1260 {
1261         struct btrfs_root *root = btrfs_sb(sb);
1262         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1263         mutex_lock(&root->fs_info->cleaner_mutex);
1264         return 0;
1265 }
1266
1267 static int btrfs_unfreeze(struct super_block *sb)
1268 {
1269         struct btrfs_root *root = btrfs_sb(sb);
1270         mutex_unlock(&root->fs_info->cleaner_mutex);
1271         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1272         return 0;
1273 }
1274
1275 static void btrfs_fs_dirty_inode(struct inode *inode, int flags)
1276 {
1277         int ret;
1278
1279         ret = btrfs_dirty_inode(inode);
1280         if (ret)
1281                 printk_ratelimited(KERN_ERR "btrfs: fail to dirty inode %Lu "
1282                                    "error %d\n", btrfs_ino(inode), ret);
1283 }
1284
1285 static const struct super_operations btrfs_super_ops = {
1286         .drop_inode     = btrfs_drop_inode,
1287         .evict_inode    = btrfs_evict_inode,
1288         .put_super      = btrfs_put_super,
1289         .sync_fs        = btrfs_sync_fs,
1290         .show_options   = btrfs_show_options,
1291         .write_inode    = btrfs_write_inode,
1292         .dirty_inode    = btrfs_fs_dirty_inode,
1293         .alloc_inode    = btrfs_alloc_inode,
1294         .destroy_inode  = btrfs_destroy_inode,
1295         .statfs         = btrfs_statfs,
1296         .remount_fs     = btrfs_remount,
1297         .freeze_fs      = btrfs_freeze,
1298         .unfreeze_fs    = btrfs_unfreeze,
1299 };
1300
1301 static const struct file_operations btrfs_ctl_fops = {
1302         .unlocked_ioctl  = btrfs_control_ioctl,
1303         .compat_ioctl = btrfs_control_ioctl,
1304         .owner   = THIS_MODULE,
1305         .llseek = noop_llseek,
1306 };
1307
1308 static struct miscdevice btrfs_misc = {
1309         .minor          = BTRFS_MINOR,
1310         .name           = "btrfs-control",
1311         .fops           = &btrfs_ctl_fops
1312 };
1313
1314 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1315 MODULE_ALIAS("devname:btrfs-control");
1316
1317 static int btrfs_interface_init(void)
1318 {
1319         return misc_register(&btrfs_misc);
1320 }
1321
1322 static void btrfs_interface_exit(void)
1323 {
1324         if (misc_deregister(&btrfs_misc) < 0)
1325                 printk(KERN_INFO "misc_deregister failed for control device");
1326 }
1327
1328 static int __init init_btrfs_fs(void)
1329 {
1330         int err;
1331
1332         err = btrfs_init_sysfs();
1333         if (err)
1334                 return err;
1335
1336         err = btrfs_init_compress();
1337         if (err)
1338                 goto free_sysfs;
1339
1340         err = btrfs_init_cachep();
1341         if (err)
1342                 goto free_compress;
1343
1344         err = extent_io_init();
1345         if (err)
1346                 goto free_cachep;
1347
1348         err = extent_map_init();
1349         if (err)
1350                 goto free_extent_io;
1351
1352         err = btrfs_delayed_inode_init();
1353         if (err)
1354                 goto free_extent_map;
1355
1356         err = btrfs_interface_init();
1357         if (err)
1358                 goto free_delayed_inode;
1359
1360         err = register_filesystem(&btrfs_fs_type);
1361         if (err)
1362                 goto unregister_ioctl;
1363
1364         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1365         return 0;
1366
1367 unregister_ioctl:
1368         btrfs_interface_exit();
1369 free_delayed_inode:
1370         btrfs_delayed_inode_exit();
1371 free_extent_map:
1372         extent_map_exit();
1373 free_extent_io:
1374         extent_io_exit();
1375 free_cachep:
1376         btrfs_destroy_cachep();
1377 free_compress:
1378         btrfs_exit_compress();
1379 free_sysfs:
1380         btrfs_exit_sysfs();
1381         return err;
1382 }
1383
1384 static void __exit exit_btrfs_fs(void)
1385 {
1386         btrfs_destroy_cachep();
1387         btrfs_delayed_inode_exit();
1388         extent_map_exit();
1389         extent_io_exit();
1390         btrfs_interface_exit();
1391         unregister_filesystem(&btrfs_fs_type);
1392         btrfs_exit_sysfs();
1393         btrfs_cleanup_fs_uuids();
1394         btrfs_exit_compress();
1395 }
1396
1397 module_init(init_btrfs_fs)
1398 module_exit(exit_btrfs_fs)
1399
1400 MODULE_LICENSE("GPL");