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 "compat.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 "ioctl.h"
  51 #include "print-tree.h"
  52 #include "xattr.h"
  53 #include "volumes.h"
  54 #include "version.h"
  55 #include "export.h"
  56 #include "compression.h"
  57
  58 #define CREATE_TRACE_POINTS
  59 #include <trace/events/btrfs.h>
  60
  61 static const struct super_operations btrfs_super_ops;
  62 static struct file_system_type btrfs_fs_type;
  63
  64 static const char *btrfs_decode_error(struct btrfs_fs_info *fs_info, int errno,
  65                                       char nbuf[16])
  66 {
  67         char *errstr = NULL;
  68
  69         switch (errno) {
  70         case -EIO:
  71                 errstr = "IO failure";
  72                 break;
  73         case -ENOMEM:
  74                 errstr = "Out of memory";
  75                 break;
  76         case -EROFS:
  77                 errstr = "Readonly filesystem";
  78                 break;
  79         default:
  80                 if (nbuf) {
  81                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
  82                                 errstr = nbuf;
  83                 }
  84                 break;
  85         }
  86
  87         return errstr;
  88 }
  89
  90 static void __save_error_info(struct btrfs_fs_info *fs_info)
  91 {
  92         /*
  93          * today we only save the error info into ram.  Long term we'll
  94          * also send it down to the disk
  95          */
  96         fs_info->fs_state = BTRFS_SUPER_FLAG_ERROR;
  97 }
  98
  99 /* NOTE:
 100  *      We move write_super stuff at umount in order to avoid deadlock
 101  *      for umount hold all lock.
 102  */
 103 static void save_error_info(struct btrfs_fs_info *fs_info)
 104 {
 105         __save_error_info(fs_info);
 106 }
 107
 108 /* btrfs handle error by forcing the filesystem readonly */
 109 static void btrfs_handle_error(struct btrfs_fs_info *fs_info)
 110 {
 111         struct super_block *sb = fs_info->sb;
 112
 113         if (sb->s_flags & MS_RDONLY)
 114                 return;
 115
 116         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
 117                 sb->s_flags |= MS_RDONLY;
 118                 printk(KERN_INFO "btrfs is forced readonly\n");
 119         }
 120 }
 121
 122 /*
 123  * __btrfs_std_error decodes expected errors from the caller and
 124  * invokes the approciate error response.
 125  */
 126 void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 127                      unsigned int line, int errno)
 128 {
 129         struct super_block *sb = fs_info->sb;
 130         char nbuf[16];
 131         const char *errstr;
 132
 133         /*
 134          * Special case: if the error is EROFS, and we're already
 135          * under MS_RDONLY, then it is safe here.
 136          */
 137         if (errno == -EROFS && (sb->s_flags & MS_RDONLY))
 138                 return;
 139
 140         errstr = btrfs_decode_error(fs_info, errno, nbuf);
 141         printk(KERN_CRIT "BTRFS error (device %s) in %s:%d: %s\n",
 142                 sb->s_id, function, line, errstr);
 143         save_error_info(fs_info);
 144
 145         btrfs_handle_error(fs_info);
 146 }
 147
 148 static void btrfs_put_super(struct super_block *sb)
 149 {
 150         struct btrfs_root *root = btrfs_sb(sb);
 151         int ret;
 152
 153         ret = close_ctree(root);
 154         free_fs_info(root->fs_info);
 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_fs_info *fs_info;
 594         struct btrfs_key key;
 595         int err;
 596
 597         sb->s_maxbytes = MAX_LFS_FILESIZE;
 598         sb->s_magic = BTRFS_SUPER_MAGIC;
 599         sb->s_op = &btrfs_super_ops;
 600         sb->s_d_op = &btrfs_dentry_operations;
 601         sb->s_export_op = &btrfs_export_ops;
 602         sb->s_xattr = btrfs_xattr_handlers;
 603         sb->s_time_gran = 1;
 604 #ifdef CONFIG_BTRFS_FS_POSIX_ACL
 605         sb->s_flags |= MS_POSIXACL;
 606 #endif
 607
 608         tree_root = open_ctree(sb, fs_devices, (char *)data);
 609
 610         if (IS_ERR(tree_root)) {
 611                 printk("btrfs: open_ctree failed\n");
 612                 return PTR_ERR(tree_root);
 613         }
 614         fs_info = tree_root->fs_info;
 615         sb->s_fs_info = tree_root;
 616
 617         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 618         key.type = BTRFS_INODE_ITEM_KEY;
 619         key.offset = 0;
 620         inode = btrfs_iget(sb, &key, fs_info->fs_root, NULL);
 621         if (IS_ERR(inode)) {
 622                 err = PTR_ERR(inode);
 623                 goto fail_close;
 624         }
 625
 626         root_dentry = d_alloc_root(inode);
 627         if (!root_dentry) {
 628                 iput(inode);
 629                 err = -ENOMEM;
 630                 goto fail_close;
 631         }
 632
 633         sb->s_root = root_dentry;
 634
 635         save_mount_options(sb, data);
 636         cleancache_init_fs(sb);
 637         return 0;
 638
 639 fail_close:
 640         close_ctree(tree_root);
 641         free_fs_info(fs_info);
 642         return err;
 643 }
 644
 645 int btrfs_sync_fs(struct super_block *sb, int wait)
 646 {
 647         struct btrfs_trans_handle *trans;
 648         struct btrfs_root *root = btrfs_sb(sb);
 649         int ret;
 650
 651         trace_btrfs_sync_fs(wait);
 652
 653         if (!wait) {
 654                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 655                 return 0;
 656         }
 657
 658         btrfs_start_delalloc_inodes(root, 0);
 659         btrfs_wait_ordered_extents(root, 0, 0);
 660
 661         trans = btrfs_start_transaction(root, 0);
 662         if (IS_ERR(trans))
 663                 return PTR_ERR(trans);
 664         ret = btrfs_commit_transaction(trans, root);
 665         return ret;
 666 }
 667
 668 static int btrfs_show_options(struct seq_file *seq, struct dentry *dentry)
 669 {
 670         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
 671         struct btrfs_fs_info *info = root->fs_info;
 672         char *compress_type;
 673
 674         if (btrfs_test_opt(root, DEGRADED))
 675                 seq_puts(seq, ",degraded");
 676         if (btrfs_test_opt(root, NODATASUM))
 677                 seq_puts(seq, ",nodatasum");
 678         if (btrfs_test_opt(root, NODATACOW))
 679                 seq_puts(seq, ",nodatacow");
 680         if (btrfs_test_opt(root, NOBARRIER))
 681                 seq_puts(seq, ",nobarrier");
 682         if (info->max_inline != 8192 * 1024)
 683                 seq_printf(seq, ",max_inline=%llu",
 684                            (unsigned long long)info->max_inline);
 685         if (info->alloc_start != 0)
 686                 seq_printf(seq, ",alloc_start=%llu",
 687                            (unsigned long long)info->alloc_start);
 688         if (info->thread_pool_size !=  min_t(unsigned long,
 689                                              num_online_cpus() + 2, 8))
 690                 seq_printf(seq, ",thread_pool=%d", info->thread_pool_size);
 691         if (btrfs_test_opt(root, COMPRESS)) {
 692                 if (info->compress_type == BTRFS_COMPRESS_ZLIB)
 693                         compress_type = "zlib";
 694                 else
 695                         compress_type = "lzo";
 696                 if (btrfs_test_opt(root, FORCE_COMPRESS))
 697                         seq_printf(seq, ",compress-force=%s", compress_type);
 698                 else
 699                         seq_printf(seq, ",compress=%s", compress_type);
 700         }
 701         if (btrfs_test_opt(root, NOSSD))
 702                 seq_puts(seq, ",nossd");
 703         if (btrfs_test_opt(root, SSD_SPREAD))
 704                 seq_puts(seq, ",ssd_spread");
 705         else if (btrfs_test_opt(root, SSD))
 706                 seq_puts(seq, ",ssd");
 707         if (btrfs_test_opt(root, NOTREELOG))
 708                 seq_puts(seq, ",notreelog");
 709         if (btrfs_test_opt(root, FLUSHONCOMMIT))
 710                 seq_puts(seq, ",flushoncommit");
 711         if (btrfs_test_opt(root, DISCARD))
 712                 seq_puts(seq, ",discard");
 713         if (!(root->fs_info->sb->s_flags & MS_POSIXACL))
 714                 seq_puts(seq, ",noacl");
 715         if (btrfs_test_opt(root, SPACE_CACHE))
 716                 seq_puts(seq, ",space_cache");
 717         else
 718                 seq_puts(seq, ",nospace_cache");
 719         if (btrfs_test_opt(root, CLEAR_CACHE))
 720                 seq_puts(seq, ",clear_cache");
 721         if (btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED))
 722                 seq_puts(seq, ",user_subvol_rm_allowed");
 723         if (btrfs_test_opt(root, ENOSPC_DEBUG))
 724                 seq_puts(seq, ",enospc_debug");
 725         if (btrfs_test_opt(root, AUTO_DEFRAG))
 726                 seq_puts(seq, ",autodefrag");
 727         if (btrfs_test_opt(root, INODE_MAP_CACHE))
 728                 seq_puts(seq, ",inode_cache");
 729         return 0;
 730 }
 731
 732 static int btrfs_test_super(struct super_block *s, void *data)
 733 {
 734         struct btrfs_root *test_root = data;
 735         struct btrfs_root *root = btrfs_sb(s);
 736
 737         /*
 738          * If this super block is going away, return false as it
 739          * can't match as an existing super block.
 740          */
 741         if (!atomic_read(&s->s_active))
 742                 return 0;
 743         return root->fs_info->fs_devices == test_root->fs_info->fs_devices;
 744 }
 745
 746 static int btrfs_set_super(struct super_block *s, void *data)
 747 {
 748         s->s_fs_info = data;
 749
 750         return set_anon_super(s, data);
 751 }
 752
 753 /*
 754  * subvolumes are identified by ino 256
 755  */
 756 static inline int is_subvolume_inode(struct inode *inode)
 757 {
 758         if (inode && inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
 759                 return 1;
 760         return 0;
 761 }
 762
 763 /*
 764  * This will strip out the subvol=%s argument for an argument string and add
 765  * subvolid=0 to make sure we get the actual tree root for path walking to the
 766  * subvol we want.
 767  */
 768 static char *setup_root_args(char *args)
 769 {
 770         unsigned copied = 0;
 771         unsigned len = strlen(args) + 2;
 772         char *pos;
 773         char *ret;
 774
 775         /*
 776          * We need the same args as before, but minus
 777          *
 778          * subvol=a
 779          *
 780          * and add
 781          *
 782          * subvolid=0
 783          *
 784          * which is a difference of 2 characters, so we allocate strlen(args) +
 785          * 2 characters.
 786          */
 787         ret = kzalloc(len * sizeof(char), GFP_NOFS);
 788         if (!ret)
 789                 return NULL;
 790         pos = strstr(args, "subvol=");
 791
 792         /* This shouldn't happen, but just in case.. */
 793         if (!pos) {
 794                 kfree(ret);
 795                 return NULL;
 796         }
 797
 798         /*
 799          * The subvol=<> arg is not at the front of the string, copy everybody
 800          * up to that into ret.
 801          */
 802         if (pos != args) {
 803                 *pos = '\0';
 804                 strcpy(ret, args);
 805                 copied += strlen(args);
 806                 pos++;
 807         }
 808
 809         strncpy(ret + copied, "subvolid=0", len - copied);
 810
 811         /* Length of subvolid=0 */
 812         copied += 10;
 813
 814         /*
 815          * If there is no , after the subvol= option then we know there's no
 816          * other options and we can just return.
 817          */
 818         pos = strchr(pos, ',');
 819         if (!pos)
 820                 return ret;
 821
 822         /* Copy the rest of the arguments into our buffer */
 823         strncpy(ret + copied, pos, len - copied);
 824         copied += strlen(pos);
 825
 826         return ret;
 827 }
 828
 829 static struct dentry *mount_subvol(const char *subvol_name, int flags,
 830                                    const char *device_name, char *data)
 831 {
 832         struct dentry *root;
 833         struct vfsmount *mnt;
 834         char *newargs;
 835
 836         newargs = setup_root_args(data);
 837         if (!newargs)
 838                 return ERR_PTR(-ENOMEM);
 839         mnt = vfs_kern_mount(&btrfs_fs_type, flags, device_name,
 840                              newargs);
 841         kfree(newargs);
 842         if (IS_ERR(mnt))
 843                 return ERR_CAST(mnt);
 844
 845         root = mount_subtree(mnt, subvol_name);
 846
 847         if (!IS_ERR(root) && !is_subvolume_inode(root->d_inode)) {
 848                 struct super_block *s = root->d_sb;
 849                 dput(root);
 850                 root = ERR_PTR(-EINVAL);
 851                 deactivate_locked_super(s);
 852                 printk(KERN_ERR "btrfs: '%s' is not a valid subvolume\n",
 853                                 subvol_name);
 854         }
 855
 856         return root;
 857 }
 858
 859 /*
 860  * Find a superblock for the given device / mount point.
 861  *
 862  * Note:  This is based on get_sb_bdev from fs/super.c with a few additions
 863  *        for multiple device setup.  Make sure to keep it in sync.
 864  */
 865 static struct dentry *btrfs_mount(struct file_system_type *fs_type, int flags,
 866                 const char *device_name, void *data)
 867 {
 868         struct block_device *bdev = NULL;
 869         struct super_block *s;
 870         struct dentry *root;
 871         struct btrfs_fs_devices *fs_devices = NULL;
 872         struct btrfs_fs_info *fs_info = NULL;
 873         fmode_t mode = FMODE_READ;
 874         char *subvol_name = NULL;
 875         u64 subvol_objectid = 0;
 876         u64 subvol_rootid = 0;
 877         int error = 0;
 878
 879         if (!(flags & MS_RDONLY))
 880                 mode |= FMODE_WRITE;
 881
 882         error = btrfs_parse_early_options(data, mode, fs_type,
 883                                           &subvol_name, &subvol_objectid,
 884                                           &subvol_rootid, &fs_devices);
 885         if (error) {
 886                 kfree(subvol_name);
 887                 return ERR_PTR(error);
 888         }
 889
 890         if (subvol_name) {
 891                 root = mount_subvol(subvol_name, flags, device_name, data);
 892                 kfree(subvol_name);
 893                 return root;
 894         }
 895
 896         error = btrfs_scan_one_device(device_name, mode, fs_type, &fs_devices);
 897         if (error)
 898                 return ERR_PTR(error);
 899
 900         /*
 901          * Setup a dummy root and fs_info for test/set super.  This is because
 902          * we don't actually fill this stuff out until open_ctree, but we need
 903          * it for searching for existing supers, so this lets us do that and
 904          * then open_ctree will properly initialize everything later.
 905          */
 906         fs_info = kzalloc(sizeof(struct btrfs_fs_info), GFP_NOFS);
 907         if (!fs_info)
 908                 return ERR_PTR(-ENOMEM);
 909
 910         fs_info->tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
 911         if (!fs_info->tree_root) {
 912                 error = -ENOMEM;
 913                 goto error_fs_info;
 914         }
 915         fs_info->tree_root->fs_info = fs_info;
 916         fs_info->fs_devices = fs_devices;
 917
 918         fs_info->super_copy = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
 919         fs_info->super_for_commit = kzalloc(BTRFS_SUPER_INFO_SIZE, GFP_NOFS);
 920         if (!fs_info->super_copy || !fs_info->super_for_commit) {
 921                 error = -ENOMEM;
 922                 goto error_fs_info;
 923         }
 924
 925         error = btrfs_open_devices(fs_devices, mode, fs_type);
 926         if (error)
 927                 goto error_fs_info;
 928
 929         if (!(flags & MS_RDONLY) && fs_devices->rw_devices == 0) {
 930                 error = -EACCES;
 931                 goto error_close_devices;
 932         }
 933
 934         bdev = fs_devices->latest_bdev;
 935         s = sget(fs_type, btrfs_test_super, btrfs_set_super,
 936                  fs_info->tree_root);
 937         if (IS_ERR(s)) {
 938                 error = PTR_ERR(s);
 939                 goto error_close_devices;
 940         }
 941
 942         if (s->s_root) {
 943                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 944                         deactivate_locked_super(s);
 945                         error = -EBUSY;
 946                         goto error_close_devices;
 947                 }
 948
 949                 btrfs_close_devices(fs_devices);
 950                 free_fs_info(fs_info);
 951         } else {
 952                 char b[BDEVNAME_SIZE];
 953
 954                 s->s_flags = flags | MS_NOSEC;
 955                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 956                 btrfs_sb(s)->fs_info->bdev_holder = fs_type;
 957                 error = btrfs_fill_super(s, fs_devices, data,
 958                                          flags & MS_SILENT ? 1 : 0);
 959                 if (error) {
 960                         deactivate_locked_super(s);
 961                         return ERR_PTR(error);
 962                 }
 963
 964                 s->s_flags |= MS_ACTIVE;
 965         }
 966
 967         root = get_default_root(s, subvol_objectid);
 968         if (IS_ERR(root)) {
 969                 deactivate_locked_super(s);
 970                 return root;
 971         }
 972
 973         return root;
 974
 975 error_close_devices:
 976         btrfs_close_devices(fs_devices);
 977 error_fs_info:
 978         free_fs_info(fs_info);
 979         return ERR_PTR(error);
 980 }
 981
 982 static int btrfs_remount(struct super_block *sb, int *flags, char *data)
 983 {
 984         struct btrfs_root *root = btrfs_sb(sb);
 985         int ret;
 986
 987         ret = btrfs_parse_options(root, data);
 988         if (ret)
 989                 return -EINVAL;
 990
 991         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
 992                 return 0;
 993
 994         if (*flags & MS_RDONLY) {
 995                 sb->s_flags |= MS_RDONLY;
 996
 997                 ret =  btrfs_commit_super(root);
 998                 WARN_ON(ret);
 999         } else {
1000                 if (root->fs_info->fs_devices->rw_devices == 0)
1001                         return -EACCES;
1002
1003                 if (btrfs_super_log_root(root->fs_info->super_copy) != 0)
1004                         return -EINVAL;
1005
1006                 ret = btrfs_cleanup_fs_roots(root->fs_info);
1007                 WARN_ON(ret);
1008
1009                 /* recover relocation */
1010                 ret = btrfs_recover_relocation(root);
1011                 WARN_ON(ret);
1012
1013                 sb->s_flags &= ~MS_RDONLY;
1014         }
1015
1016         return 0;
1017 }
1018
1019 /* Used to sort the devices by max_avail(descending sort) */
1020 static int btrfs_cmp_device_free_bytes(const void *dev_info1,
1021                                        const void *dev_info2)
1022 {
1023         if (((struct btrfs_device_info *)dev_info1)->max_avail >
1024             ((struct btrfs_device_info *)dev_info2)->max_avail)
1025                 return -1;
1026         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
1027                  ((struct btrfs_device_info *)dev_info2)->max_avail)
1028                 return 1;
1029         else
1030         return 0;
1031 }
1032
1033 /*
1034  * sort the devices by max_avail, in which max free extent size of each device
1035  * is stored.(Descending Sort)
1036  */
1037 static inline void btrfs_descending_sort_devices(
1038                                         struct btrfs_device_info *devices,
1039                                         size_t nr_devices)
1040 {
1041         sort(devices, nr_devices, sizeof(struct btrfs_device_info),
1042              btrfs_cmp_device_free_bytes, NULL);
1043 }
1044
1045 /*
1046  * The helper to calc the free space on the devices that can be used to store
1047  * file data.
1048  */
1049 static int btrfs_calc_avail_data_space(struct btrfs_root *root, u64 *free_bytes)
1050 {
1051         struct btrfs_fs_info *fs_info = root->fs_info;
1052         struct btrfs_device_info *devices_info;
1053         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1054         struct btrfs_device *device;
1055         u64 skip_space;
1056         u64 type;
1057         u64 avail_space;
1058         u64 used_space;
1059         u64 min_stripe_size;
1060         int min_stripes = 1, num_stripes = 1;
1061         int i = 0, nr_devices;
1062         int ret;
1063
1064         nr_devices = fs_info->fs_devices->open_devices;
1065         BUG_ON(!nr_devices);
1066
1067         devices_info = kmalloc(sizeof(*devices_info) * nr_devices,
1068                                GFP_NOFS);
1069         if (!devices_info)
1070                 return -ENOMEM;
1071
1072         /* calc min stripe number for data space alloction */
1073         type = btrfs_get_alloc_profile(root, 1);
1074         if (type & BTRFS_BLOCK_GROUP_RAID0) {
1075                 min_stripes = 2;
1076                 num_stripes = nr_devices;
1077         } else if (type & BTRFS_BLOCK_GROUP_RAID1) {
1078                 min_stripes = 2;
1079                 num_stripes = 2;
1080         } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
1081                 min_stripes = 4;
1082                 num_stripes = 4;
1083         }
1084
1085         if (type & BTRFS_BLOCK_GROUP_DUP)
1086                 min_stripe_size = 2 * BTRFS_STRIPE_LEN;
1087         else
1088                 min_stripe_size = BTRFS_STRIPE_LEN;
1089
1090         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1091                 if (!device->in_fs_metadata || !device->bdev)
1092                         continue;
1093
1094                 avail_space = device->total_bytes - device->bytes_used;
1095
1096                 /* align with stripe_len */
1097                 do_div(avail_space, BTRFS_STRIPE_LEN);
1098                 avail_space *= BTRFS_STRIPE_LEN;
1099
1100                 /*
1101                  * In order to avoid overwritting the superblock on the drive,
1102                  * btrfs starts at an offset of at least 1MB when doing chunk
1103                  * allocation.
1104                  */
1105                 skip_space = 1024 * 1024;
1106
1107                 /* user can set the offset in fs_info->alloc_start. */
1108                 if (fs_info->alloc_start + BTRFS_STRIPE_LEN <=
1109                     device->total_bytes)
1110                         skip_space = max(fs_info->alloc_start, skip_space);
1111
1112                 /*
1113                  * btrfs can not use the free space in [0, skip_space - 1],
1114                  * we must subtract it from the total. In order to implement
1115                  * it, we account the used space in this range first.
1116                  */
1117                 ret = btrfs_account_dev_extents_size(device, 0, skip_space - 1,
1118                                                      &used_space);
1119                 if (ret) {
1120                         kfree(devices_info);
1121                         return ret;
1122                 }
1123
1124                 /* calc the free space in [0, skip_space - 1] */
1125                 skip_space -= used_space;
1126
1127                 /*
1128                  * we can use the free space in [0, skip_space - 1], subtract
1129                  * it from the total.
1130                  */
1131                 if (avail_space && avail_space >= skip_space)
1132                         avail_space -= skip_space;
1133                 else
1134                         avail_space = 0;
1135
1136                 if (avail_space < min_stripe_size)
1137                         continue;
1138
1139                 devices_info[i].dev = device;
1140                 devices_info[i].max_avail = avail_space;
1141
1142                 i++;
1143         }
1144
1145         nr_devices = i;
1146
1147         btrfs_descending_sort_devices(devices_info, nr_devices);
1148
1149         i = nr_devices - 1;
1150         avail_space = 0;
1151         while (nr_devices >= min_stripes) {
1152                 if (num_stripes > nr_devices)
1153                         num_stripes = nr_devices;
1154
1155                 if (devices_info[i].max_avail >= min_stripe_size) {
1156                         int j;
1157                         u64 alloc_size;
1158
1159                         avail_space += devices_info[i].max_avail * num_stripes;
1160                         alloc_size = devices_info[i].max_avail;
1161                         for (j = i + 1 - num_stripes; j <= i; j++)
1162                                 devices_info[j].max_avail -= alloc_size;
1163                 }
1164                 i--;
1165                 nr_devices--;
1166         }
1167
1168         kfree(devices_info);
1169         *free_bytes = avail_space;
1170         return 0;
1171 }
1172
1173 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1174 {
1175         struct btrfs_root *root = btrfs_sb(dentry->d_sb);
1176         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1177         struct list_head *head = &root->fs_info->space_info;
1178         struct btrfs_space_info *found;
1179         u64 total_used = 0;
1180         u64 total_free_data = 0;
1181         int bits = dentry->d_sb->s_blocksize_bits;
1182         __be32 *fsid = (__be32 *)root->fs_info->fsid;
1183         int ret;
1184
1185         /* holding chunk_muext to avoid allocating new chunks */
1186         mutex_lock(&root->fs_info->chunk_mutex);
1187         rcu_read_lock();
1188         list_for_each_entry_rcu(found, head, list) {
1189                 if (found->flags & BTRFS_BLOCK_GROUP_DATA) {
1190                         total_free_data += found->disk_total - found->disk_used;
1191                         total_free_data -=
1192                                 btrfs_account_ro_block_groups_free_space(found);
1193                 }
1194
1195                 total_used += found->disk_used;
1196         }
1197         rcu_read_unlock();
1198
1199         buf->f_namelen = BTRFS_NAME_LEN;
1200         buf->f_blocks = btrfs_super_total_bytes(disk_super) >> bits;
1201         buf->f_bfree = buf->f_blocks - (total_used >> bits);
1202         buf->f_bsize = dentry->d_sb->s_blocksize;
1203         buf->f_type = BTRFS_SUPER_MAGIC;
1204         buf->f_bavail = total_free_data;
1205         ret = btrfs_calc_avail_data_space(root, &total_free_data);
1206         if (ret) {
1207                 mutex_unlock(&root->fs_info->chunk_mutex);
1208                 return ret;
1209         }
1210         buf->f_bavail += total_free_data;
1211         buf->f_bavail = buf->f_bavail >> bits;
1212         mutex_unlock(&root->fs_info->chunk_mutex);
1213
1214         /* We treat it as constant endianness (it doesn't matter _which_)
1215            because we want the fsid to come out the same whether mounted
1216            on a big-endian or little-endian host */
1217         buf->f_fsid.val[0] = be32_to_cpu(fsid[0]) ^ be32_to_cpu(fsid[2]);
1218         buf->f_fsid.val[1] = be32_to_cpu(fsid[1]) ^ be32_to_cpu(fsid[3]);
1219         /* Mask in the root object ID too, to disambiguate subvols */
1220         buf->f_fsid.val[0] ^= BTRFS_I(dentry->d_inode)->root->objectid >> 32;
1221         buf->f_fsid.val[1] ^= BTRFS_I(dentry->d_inode)->root->objectid;
1222
1223         return 0;
1224 }
1225
1226 static struct file_system_type btrfs_fs_type = {
1227         .owner          = THIS_MODULE,
1228         .name           = "btrfs",
1229         .mount          = btrfs_mount,
1230         .kill_sb        = kill_anon_super,
1231         .fs_flags       = FS_REQUIRES_DEV,
1232 };
1233
1234 /*
1235  * used by btrfsctl to scan devices when no FS is mounted
1236  */
1237 static long btrfs_control_ioctl(struct file *file, unsigned int cmd,
1238                                 unsigned long arg)
1239 {
1240         struct btrfs_ioctl_vol_args *vol;
1241         struct btrfs_fs_devices *fs_devices;
1242         int ret = -ENOTTY;
1243
1244         if (!capable(CAP_SYS_ADMIN))
1245                 return -EPERM;
1246
1247         vol = memdup_user((void __user *)arg, sizeof(*vol));
1248         if (IS_ERR(vol))
1249                 return PTR_ERR(vol);
1250
1251         switch (cmd) {
1252         case BTRFS_IOC_SCAN_DEV:
1253                 ret = btrfs_scan_one_device(vol->name, FMODE_READ,
1254                                             &btrfs_fs_type, &fs_devices);
1255                 break;
1256         }
1257
1258         kfree(vol);
1259         return ret;
1260 }
1261
1262 static int btrfs_freeze(struct super_block *sb)
1263 {
1264         struct btrfs_root *root = btrfs_sb(sb);
1265         mutex_lock(&root->fs_info->transaction_kthread_mutex);
1266         mutex_lock(&root->fs_info->cleaner_mutex);
1267         return 0;
1268 }
1269
1270 static int btrfs_unfreeze(struct super_block *sb)
1271 {
1272         struct btrfs_root *root = btrfs_sb(sb);
1273         mutex_unlock(&root->fs_info->cleaner_mutex);
1274         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1275         return 0;
1276 }
1277
1278 static void btrfs_fs_dirty_inode(struct inode *inode, int flags)
1279 {
1280         int ret;
1281
1282         ret = btrfs_dirty_inode(inode);
1283         if (ret)
1284                 printk_ratelimited(KERN_ERR "btrfs: fail to dirty inode %Lu "
1285                                    "error %d\n", btrfs_ino(inode), ret);
1286 }
1287
1288 static const struct super_operations btrfs_super_ops = {
1289         .drop_inode     = btrfs_drop_inode,
1290         .evict_inode    = btrfs_evict_inode,
1291         .put_super      = btrfs_put_super,
1292         .sync_fs        = btrfs_sync_fs,
1293         .show_options   = btrfs_show_options,
1294         .write_inode    = btrfs_write_inode,
1295         .dirty_inode    = btrfs_fs_dirty_inode,
1296         .alloc_inode    = btrfs_alloc_inode,
1297         .destroy_inode  = btrfs_destroy_inode,
1298         .statfs         = btrfs_statfs,
1299         .remount_fs     = btrfs_remount,
1300         .freeze_fs      = btrfs_freeze,
1301         .unfreeze_fs    = btrfs_unfreeze,
1302 };
1303
1304 static const struct file_operations btrfs_ctl_fops = {
1305         .unlocked_ioctl  = btrfs_control_ioctl,
1306         .compat_ioctl = btrfs_control_ioctl,
1307         .owner   = THIS_MODULE,
1308         .llseek = noop_llseek,
1309 };
1310
1311 static struct miscdevice btrfs_misc = {
1312         .minor          = BTRFS_MINOR,
1313         .name           = "btrfs-control",
1314         .fops           = &btrfs_ctl_fops
1315 };
1316
1317 MODULE_ALIAS_MISCDEV(BTRFS_MINOR);
1318 MODULE_ALIAS("devname:btrfs-control");
1319
1320 static int btrfs_interface_init(void)
1321 {
1322         return misc_register(&btrfs_misc);
1323 }
1324
1325 static void btrfs_interface_exit(void)
1326 {
1327         if (misc_deregister(&btrfs_misc) < 0)
1328                 printk(KERN_INFO "misc_deregister failed for control device");
1329 }
1330
1331 static int __init init_btrfs_fs(void)
1332 {
1333         int err;
1334
1335         err = btrfs_init_sysfs();
1336         if (err)
1337                 return err;
1338
1339         err = btrfs_init_compress();
1340         if (err)
1341                 goto free_sysfs;
1342
1343         err = btrfs_init_cachep();
1344         if (err)
1345                 goto free_compress;
1346
1347         err = extent_io_init();
1348         if (err)
1349                 goto free_cachep;
1350
1351         err = extent_map_init();
1352         if (err)
1353                 goto free_extent_io;
1354
1355         err = btrfs_delayed_inode_init();
1356         if (err)
1357                 goto free_extent_map;
1358
1359         err = btrfs_interface_init();
1360         if (err)
1361                 goto free_delayed_inode;
1362
1363         err = register_filesystem(&btrfs_fs_type);
1364         if (err)
1365                 goto unregister_ioctl;
1366
1367         printk(KERN_INFO "%s loaded\n", BTRFS_BUILD_VERSION);
1368         return 0;
1369
1370 unregister_ioctl:
1371         btrfs_interface_exit();
1372 free_delayed_inode:
1373         btrfs_delayed_inode_exit();
1374 free_extent_map:
1375         extent_map_exit();
1376 free_extent_io:
1377         extent_io_exit();
1378 free_cachep:
1379         btrfs_destroy_cachep();
1380 free_compress:
1381         btrfs_exit_compress();
1382 free_sysfs:
1383         btrfs_exit_sysfs();
1384         return err;
1385 }
1386
1387 static void __exit exit_btrfs_fs(void)
1388 {
1389         btrfs_destroy_cachep();
1390         btrfs_delayed_inode_exit();
1391         extent_map_exit();
1392         extent_io_exit();
1393         btrfs_interface_exit();
1394         unregister_filesystem(&btrfs_fs_type);
1395         btrfs_exit_sysfs();
1396         btrfs_cleanup_fs_uuids();
1397         btrfs_exit_compress();
1398 }
1399
1400 module_init(init_btrfs_fs)
1401 module_exit(exit_btrfs_fs)
1402
1403 MODULE_LICENSE("GPL");