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