Btrfs: fix oops after block group lookup
[linux-2.6-block.git] / fs / btrfs / super.c
index bbe5cabcb42e875394d1ea8d6eacf6e9c1e8eff8..9749090cfbb8ac1385eccab3e4e44fd0c0778b7d 100644 (file)
 #include <linux/mpage.h>
 #include <linux/swap.h>
 #include <linux/writeback.h>
+#include <linux/statfs.h>
+#include <linux/compat.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
 #include "btrfs_inode.h"
 #include "ioctl.h"
-
-void btrfs_fsinfo_release(struct kobject *obj)
-{
-       struct btrfs_fs_info *fsinfo = container_of(obj,
-                                           struct btrfs_fs_info, kobj);
-       kfree(fsinfo);
-}
-
-struct kobj_type btrfs_fsinfo_ktype = {
-       .release = btrfs_fsinfo_release,
-};
+#include "print-tree.h"
 
 struct btrfs_iget_args {
        u64 ino;
        struct btrfs_root *root;
 };
 
-decl_subsys(btrfs, &btrfs_fsinfo_ktype, NULL);
-
 #define BTRFS_SUPER_MAGIC 0x9123682E
 
 static struct inode_operations btrfs_dir_inode_operations;
+static struct inode_operations btrfs_symlink_inode_operations;
 static struct inode_operations btrfs_dir_ro_inode_operations;
 static struct super_operations btrfs_super_ops;
 static struct file_operations btrfs_dir_file_operations;
 static struct inode_operations btrfs_file_inode_operations;
 static struct address_space_operations btrfs_aops;
+static struct address_space_operations btrfs_symlink_aops;
 static struct file_operations btrfs_file_operations;
 
-static int check_inode(struct inode *inode)
-{
-       struct btrfs_inode *ei = BTRFS_I(inode);
-       WARN_ON(ei->magic != 0xDEADBEEF);
-       WARN_ON(ei->magic2 != 0xDEADBEAF);
-       return 0;
-}
+static int drop_extents(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root,
+                         struct inode *inode,
+                         u64 start, u64 end, u64 *hint_block);
+static int btrfs_get_block(struct inode *inode, sector_t iblock,
+                          struct buffer_head *result, int create);
+
+
+#define S_SHIFT 12
+static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
+       [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
+       [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
+       [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
+       [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
+       [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
+       [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
+       [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
+};
 
 static void btrfs_read_locked_inode(struct inode *inode)
 {
@@ -59,6 +62,7 @@ static void btrfs_read_locked_inode(struct inode *inode)
        struct btrfs_inode_item *inode_item;
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_key location;
+       u64 alloc_group_block;
        int ret;
 
        path = btrfs_alloc_path();
@@ -66,15 +70,12 @@ static void btrfs_read_locked_inode(struct inode *inode)
        btrfs_init_path(path);
        mutex_lock(&root->fs_info->fs_mutex);
 
-       check_inode(inode);
-
        memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
        ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
        if (ret) {
                btrfs_free_path(path);
                goto make_bad;
        }
-       check_inode(inode);
        inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
                                  path->slots[0],
                                  struct btrfs_inode_item);
@@ -92,12 +93,15 @@ static void btrfs_read_locked_inode(struct inode *inode)
        inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
        inode->i_blocks = btrfs_inode_nblocks(inode_item);
        inode->i_generation = btrfs_inode_generation(inode_item);
+       alloc_group_block = btrfs_inode_block_group(inode_item);
+       BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
+                                                      alloc_group_block);
 
        btrfs_free_path(path);
        inode_item = NULL;
 
        mutex_unlock(&root->fs_info->fs_mutex);
-       check_inode(inode);
+
        switch (inode->i_mode & S_IFMT) {
 #if 0
        default:
@@ -118,10 +122,10 @@ static void btrfs_read_locked_inode(struct inode *inode)
                        inode->i_op = &btrfs_dir_inode_operations;
                break;
        case S_IFLNK:
-               // inode->i_op = &page_symlink_inode_operations;
+               inode->i_op = &btrfs_symlink_inode_operations;
+               inode->i_mapping->a_ops = &btrfs_symlink_aops;
                break;
        }
-       check_inode(inode);
        return;
 
 make_bad:
@@ -131,6 +135,59 @@ make_bad:
        make_bad_inode(inode);
 }
 
+static void fill_inode_item(struct btrfs_inode_item *item,
+                           struct inode *inode)
+{
+       btrfs_set_inode_uid(item, inode->i_uid);
+       btrfs_set_inode_gid(item, inode->i_gid);
+       btrfs_set_inode_size(item, inode->i_size);
+       btrfs_set_inode_mode(item, inode->i_mode);
+       btrfs_set_inode_nlink(item, inode->i_nlink);
+       btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
+       btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
+       btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
+       btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
+       btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
+       btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
+       btrfs_set_inode_nblocks(item, inode->i_blocks);
+       btrfs_set_inode_generation(item, inode->i_generation);
+       btrfs_set_inode_block_group(item,
+                                   BTRFS_I(inode)->block_group->key.objectid);
+}
+
+static int btrfs_update_inode(struct btrfs_trans_handle *trans,
+                             struct btrfs_root *root,
+                             struct inode *inode)
+{
+       struct btrfs_inode_item *inode_item;
+       struct btrfs_path *path;
+       int ret;
+
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+       btrfs_init_path(path);
+       ret = btrfs_lookup_inode(trans, root, path,
+                                &BTRFS_I(inode)->location, 1);
+       if (ret) {
+               if (ret > 0)
+                       ret = -ENOENT;
+               goto failed;
+       }
+
+       inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+                                 path->slots[0],
+                                 struct btrfs_inode_item);
+
+       fill_inode_item(inode_item, inode);
+       btrfs_mark_buffer_dirty(path->nodes[0]);
+       ret = 0;
+failed:
+       btrfs_release_path(root, path);
+       btrfs_free_path(path);
+       return ret;
+}
+
+
 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
                              struct btrfs_root *root,
                              struct inode *dir,
@@ -139,42 +196,50 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
        struct btrfs_path *path;
        const char *name = dentry->d_name.name;
        int name_len = dentry->d_name.len;
-       int ret;
+       int ret = 0;
        u64 objectid;
        struct btrfs_dir_item *di;
 
        path = btrfs_alloc_path();
        BUG_ON(!path);
        btrfs_init_path(path);
-       ret = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
+       di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
                                    name, name_len, -1);
-       if (ret < 0)
+       if (IS_ERR(di)) {
+               ret = PTR_ERR(di);
                goto err;
-       if (ret > 0) {
+       }
+       if (!di) {
                ret = -ENOENT;
                goto err;
        }
-       di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
-                           struct btrfs_dir_item);
        objectid = btrfs_disk_key_objectid(&di->location);
-
-       ret = btrfs_del_item(trans, root, path);
+       ret = btrfs_delete_one_dir_name(trans, root, path, di);
        BUG_ON(ret);
-
        btrfs_release_path(root, path);
-       ret = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
-                                         objectid, -1);
-       BUG_ON(ret);
-       ret = btrfs_del_item(trans, root, path);
+
+       di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
+                                        objectid, name, name_len, -1);
+       if (IS_ERR(di)) {
+               ret = PTR_ERR(di);
+               goto err;
+       }
+       if (!di) {
+               ret = -ENOENT;
+               goto err;
+       }
+       ret = btrfs_delete_one_dir_name(trans, root, path, di);
        BUG_ON(ret);
+
        dentry->d_inode->i_ctime = dir->i_ctime;
 err:
-       btrfs_release_path(root, path);
        btrfs_free_path(path);
-       if (ret == 0) {
-               inode_dec_link_count(dentry->d_inode);
+       if (!ret) {
                dir->i_size -= name_len * 2;
-               mark_inode_dirty(dir);
+               btrfs_update_inode(trans, root, dir);
+               drop_nlink(dentry->d_inode);
+               btrfs_update_inode(trans, root, dentry->d_inode);
+               dir->i_sb->s_dirt = 1;
        }
        return ret;
 }
@@ -188,9 +253,11 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
        root = BTRFS_I(dir)->root;
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, dir);
        ret = btrfs_unlink_trans(trans, root, dir, dentry);
        btrfs_end_transaction(trans, root);
        mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
        return ret;
 }
 
@@ -213,6 +280,7 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
        btrfs_init_path(path);
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, dir);
        key.objectid = inode->i_ino;
        key.offset = (u64)-1;
        key.flags = (u32)-1;
@@ -263,6 +331,7 @@ out:
        btrfs_free_path(path);
        mutex_unlock(&root->fs_info->fs_mutex);
        ret = btrfs_end_transaction(trans, root);
+       btrfs_btree_balance_dirty(root);
        if (ret && !err)
                err = ret;
        return err;
@@ -272,10 +341,7 @@ static int btrfs_free_inode(struct btrfs_trans_handle *trans,
                            struct btrfs_root *root,
                            struct inode *inode)
 {
-       u64 objectid = inode->i_ino;
        struct btrfs_path *path;
-       struct btrfs_inode_map_item *map;
-       struct btrfs_key stat_data_key;
        int ret;
 
        clear_inode(inode);
@@ -283,30 +349,44 @@ static int btrfs_free_inode(struct btrfs_trans_handle *trans,
        path = btrfs_alloc_path();
        BUG_ON(!path);
        btrfs_init_path(path);
-       ret = btrfs_lookup_inode_map(trans, root, path, objectid, -1);
-       if (ret) {
-               if (ret > 0)
-                       ret = -ENOENT;
-               goto error;
-       }
-       map = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
-                           struct btrfs_inode_map_item);
-       btrfs_disk_key_to_cpu(&stat_data_key, &map->key);
-       ret = btrfs_del_item(trans, root->fs_info->inode_root, path);
-       BUG_ON(ret);
-       btrfs_release_path(root, path);
-
        ret = btrfs_lookup_inode(trans, root, path,
                                 &BTRFS_I(inode)->location, -1);
        BUG_ON(ret);
        ret = btrfs_del_item(trans, root, path);
        BUG_ON(ret);
-error:
-       btrfs_release_path(root, path);
        btrfs_free_path(path);
        return ret;
 }
 
+static void reada_truncate(struct btrfs_root *root, struct btrfs_path *path,
+                          u64 objectid)
+{
+       struct btrfs_node *node;
+       int i;
+       int nritems;
+       u64 item_objectid;
+       u64 blocknr;
+       int slot;
+       int ret;
+
+       if (!path->nodes[1])
+               return;
+       node = btrfs_buffer_node(path->nodes[1]);
+       slot = path->slots[1];
+       if (slot == 0)
+               return;
+       nritems = btrfs_header_nritems(&node->header);
+       for (i = slot - 1; i >= 0; i--) {
+               item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
+               if (item_objectid != objectid)
+                       break;
+               blocknr = btrfs_node_blockptr(node, i);
+               ret = readahead_tree_block(root, blocknr);
+               if (ret)
+                       break;
+       }
+}
+
 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
                                   struct btrfs_root *root,
                                   struct inode *inode)
@@ -315,25 +395,24 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
        struct btrfs_path *path;
        struct btrfs_key key;
        struct btrfs_disk_key *found_key;
+       u32 found_type;
        struct btrfs_leaf *leaf;
-       struct btrfs_file_extent_item *fi = NULL;
+       struct btrfs_file_extent_item *fi;
        u64 extent_start = 0;
        u64 extent_num_blocks = 0;
+       u64 item_end = 0;
        int found_extent;
+       int del_item;
 
        path = btrfs_alloc_path();
        BUG_ON(!path);
        /* FIXME, add redo link to tree so we don't leak on crash */
        key.objectid = inode->i_ino;
        key.offset = (u64)-1;
-       key.flags = 0;
-       /*
-        * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
-        * or extent data
-        */
-       btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
+       key.flags = (u32)-1;
        while(1) {
                btrfs_init_path(path);
+               fi = NULL;
                ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
                if (ret < 0) {
                        goto error;
@@ -342,31 +421,81 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
                        BUG_ON(path->slots[0] == 0);
                        path->slots[0]--;
                }
+               reada_truncate(root, path, inode->i_ino);
                leaf = btrfs_buffer_leaf(path->nodes[0]);
                found_key = &leaf->items[path->slots[0]].key;
+               found_type = btrfs_disk_key_type(found_key);
                if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
                        break;
-               if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
-                   btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
-                   btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
+               if (found_type != BTRFS_CSUM_ITEM_KEY &&
+                   found_type != BTRFS_DIR_ITEM_KEY &&
+                   found_type != BTRFS_DIR_INDEX_KEY &&
+                   found_type != BTRFS_EXTENT_DATA_KEY)
                        break;
-               if (btrfs_disk_key_offset(found_key) < inode->i_size)
-                       break;
-               if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
+               item_end = btrfs_disk_key_offset(found_key);
+               if (found_type == BTRFS_EXTENT_DATA_KEY) {
                        fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
                                            path->slots[0],
                                            struct btrfs_file_extent_item);
-                       extent_start = btrfs_file_extent_disk_blocknr(fi);
-                       extent_num_blocks =
-                               btrfs_file_extent_disk_num_blocks(fi);
-                       inode->i_blocks -=
-                               btrfs_file_extent_num_blocks(fi) >> 9;
-                       found_extent = 1;
+                       if (btrfs_file_extent_type(fi) !=
+                           BTRFS_FILE_EXTENT_INLINE) {
+                               item_end += btrfs_file_extent_num_blocks(fi) <<
+                                               inode->i_blkbits;
+                       }
+               }
+               if (found_type == BTRFS_CSUM_ITEM_KEY) {
+                       ret = btrfs_csum_truncate(trans, root, path,
+                                                 inode->i_size);
+                       BUG_ON(ret);
+               }
+               if (item_end < inode->i_size) {
+                       if (found_type) {
+                               btrfs_set_key_type(&key, found_type - 1);
+                               continue;
+                       }
+                       break;
+               }
+               if (btrfs_disk_key_offset(found_key) >= inode->i_size)
+                       del_item = 1;
+               else
+                       del_item = 0;
+               found_extent = 0;
+
+               if (found_type == BTRFS_EXTENT_DATA_KEY &&
+                          btrfs_file_extent_type(fi) !=
+                          BTRFS_FILE_EXTENT_INLINE) {
+                       u64 num_dec;
+                       if (!del_item) {
+                               u64 orig_num_blocks =
+                                       btrfs_file_extent_num_blocks(fi);
+                               extent_num_blocks = inode->i_size -
+                                       btrfs_disk_key_offset(found_key) +
+                                       root->blocksize - 1;
+                               extent_num_blocks >>= inode->i_blkbits;
+                               btrfs_set_file_extent_num_blocks(fi,
+                                                        extent_num_blocks);
+                               inode->i_blocks -= (orig_num_blocks -
+                                       extent_num_blocks) << 3;
+                               mark_buffer_dirty(path->nodes[0]);
+                       } else {
+                               extent_start =
+                                       btrfs_file_extent_disk_blocknr(fi);
+                               extent_num_blocks =
+                                       btrfs_file_extent_disk_num_blocks(fi);
+                               /* FIXME blocksize != 4096 */
+                               num_dec = btrfs_file_extent_num_blocks(fi) << 3;
+                               if (extent_start != 0) {
+                                       found_extent = 1;
+                                       inode->i_blocks -= num_dec;
+                               }
+                       }
+               }
+               if (del_item) {
+                       ret = btrfs_del_item(trans, root, path);
+                       BUG_ON(ret);
                } else {
-                       found_extent = 0;
+                       break;
                }
-               ret = btrfs_del_item(trans, root, path);
-               BUG_ON(ret);
                btrfs_release_path(root, path);
                if (found_extent) {
                        ret = btrfs_free_extent(trans, root, extent_start,
@@ -378,9 +507,111 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 error:
        btrfs_release_path(root, path);
        btrfs_free_path(path);
+       inode->i_sb->s_dirt = 1;
        return ret;
 }
 
+static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
+{
+       struct inode *inode = mapping->host;
+       unsigned blocksize = 1 << inode->i_blkbits;
+       pgoff_t index = from >> PAGE_CACHE_SHIFT;
+       unsigned offset = from & (PAGE_CACHE_SIZE-1);
+       struct page *page;
+       char *kaddr;
+       int ret = 0;
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       u64 alloc_hint;
+       struct btrfs_key ins;
+       struct btrfs_trans_handle *trans;
+
+       if ((offset & (blocksize - 1)) == 0)
+               goto out;
+
+       ret = -ENOMEM;
+       page = grab_cache_page(mapping, index);
+       if (!page)
+               goto out;
+
+       if (!PageUptodate(page)) {
+               ret = mpage_readpage(page, btrfs_get_block);
+               lock_page(page);
+               if (!PageUptodate(page)) {
+                       ret = -EIO;
+                       goto out;
+               }
+       }
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, inode);
+
+       ret = drop_extents(trans, root, inode, page->index << PAGE_CACHE_SHIFT,
+                          (page->index + 1) << PAGE_CACHE_SHIFT, &alloc_hint);
+       BUG_ON(ret);
+       ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1,
+                                alloc_hint, (u64)-1, &ins, 1);
+       BUG_ON(ret);
+       ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
+                                      page->index << PAGE_CACHE_SHIFT,
+                                      ins.objectid, 1, 1);
+       BUG_ON(ret);
+       SetPageChecked(page);
+       kaddr = kmap(page);
+       memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
+       flush_dcache_page(page);
+       btrfs_csum_file_block(trans, root, inode->i_ino,
+                             page->index << PAGE_CACHE_SHIFT,
+                             kaddr, PAGE_CACHE_SIZE);
+       kunmap(page);
+       btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+
+       set_page_dirty(page);
+       unlock_page(page);
+       page_cache_release(page);
+out:
+       return ret;
+}
+
+static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
+{
+       struct inode *inode = dentry->d_inode;
+       int err;
+
+       err = inode_change_ok(inode, attr);
+       if (err)
+               return err;
+
+       if (S_ISREG(inode->i_mode) &&
+           attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
+               struct btrfs_trans_handle *trans;
+               struct btrfs_root *root = BTRFS_I(inode)->root;
+               u64 mask = root->blocksize - 1;
+               u64 pos = (inode->i_size + mask) & ~mask;
+               u64 hole_size;
+
+               if (attr->ia_size <= pos)
+                       goto out;
+
+               btrfs_truncate_page(inode->i_mapping, inode->i_size);
+
+               hole_size = (attr->ia_size - pos + mask) & ~mask;
+               hole_size >>= inode->i_blkbits;
+
+               mutex_lock(&root->fs_info->fs_mutex);
+               trans = btrfs_start_transaction(root, 1);
+               btrfs_set_trans_block_group(trans, inode);
+               err = btrfs_insert_file_extent(trans, root, inode->i_ino,
+                                              pos, 0, 0, hole_size);
+               BUG_ON(err);
+               btrfs_end_transaction(trans, root);
+               mutex_unlock(&root->fs_info->fs_mutex);
+       }
+out:
+       err = inode_setattr(inode, attr);
+
+       return err;
+}
 static void btrfs_delete_inode(struct inode *inode)
 {
        struct btrfs_trans_handle *trans;
@@ -394,13 +625,13 @@ static void btrfs_delete_inode(struct inode *inode)
        inode->i_size = 0;
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
-       if (S_ISREG(inode->i_mode)) {
-               ret = btrfs_truncate_in_trans(trans, root, inode);
-               BUG_ON(ret);
-       }
+       btrfs_set_trans_block_group(trans, inode);
+       ret = btrfs_truncate_in_trans(trans, root, inode);
+       BUG_ON(ret);
        btrfs_free_inode(trans, root, inode);
        btrfs_end_transaction(trans, root);
        mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
        return;
 no_delete:
        clear_inode(inode);
@@ -419,24 +650,21 @@ static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
        path = btrfs_alloc_path();
        BUG_ON(!path);
        btrfs_init_path(path);
-       ret = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
+       di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
                                    namelen, 0);
-       if (ret || !btrfs_match_dir_item_name(root, path, name, namelen)) {
+       if (!di || IS_ERR(di)) {
                location->objectid = 0;
                ret = 0;
                goto out;
        }
-       di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
-                           struct btrfs_dir_item);
        btrfs_disk_key_to_cpu(location, &di->location);
 out:
        btrfs_release_path(root, path);
        btrfs_free_path(path);
-       check_inode(dir);
        return ret;
 }
 
-int fixup_tree_root_location(struct btrfs_root *root,
+static int fixup_tree_root_location(struct btrfs_root *root,
                             struct btrfs_key *location,
                             struct btrfs_root **sub_root)
 {
@@ -467,7 +695,7 @@ int fixup_tree_root_location(struct btrfs_root *root,
        return 0;
 }
 
-int btrfs_init_locked_inode(struct inode *inode, void *p)
+static int btrfs_init_locked_inode(struct inode *inode, void *p)
 {
        struct btrfs_iget_args *args = p;
        inode->i_ino = args->ino;
@@ -475,15 +703,15 @@ int btrfs_init_locked_inode(struct inode *inode, void *p)
        return 0;
 }
 
-int btrfs_find_actor(struct inode *inode, void *opaque)
+static int btrfs_find_actor(struct inode *inode, void *opaque)
 {
        struct btrfs_iget_args *args = opaque;
        return (args->ino == inode->i_ino &&
                args->root == BTRFS_I(inode)->root);
 }
 
-struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
-                               struct btrfs_root *root)
+static struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
+                                      struct btrfs_root *root)
 {
        struct inode *inode;
        struct btrfs_iget_args args;
@@ -526,10 +754,6 @@ static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
                        return ERR_PTR(-EACCES);
                if (inode->i_state & I_NEW) {
                        if (sub_root != root) {
-                               ret = radix_tree_insert(
-                                               &root->fs_info->fs_roots_radix,
-                                               (unsigned long)sub_root,
-                                               sub_root);
 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
                                igrab(inode);
                                sub_root->inode = inode;
@@ -540,12 +764,40 @@ printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_r
                        btrfs_read_locked_inode(inode);
                        unlock_new_inode(inode);
                }
-               check_inode(inode);
        }
-       check_inode(dir);
        return d_splice_alias(inode, dentry);
 }
 
+static void reada_leaves(struct btrfs_root *root, struct btrfs_path *path,
+                        u64 objectid)
+{
+       struct btrfs_node *node;
+       int i;
+       u32 nritems;
+       u64 item_objectid;
+       u64 blocknr;
+       int slot;
+       int ret;
+
+       if (!path->nodes[1])
+               return;
+       node = btrfs_buffer_node(path->nodes[1]);
+       slot = path->slots[1];
+       nritems = btrfs_header_nritems(&node->header);
+       for (i = slot + 1; i < nritems; i++) {
+               item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
+               if (item_objectid != objectid)
+                       break;
+               blocknr = btrfs_node_blockptr(node, i);
+               ret = readahead_tree_block(root, blocknr);
+               if (ret)
+                       break;
+       }
+}
+static unsigned char btrfs_filetype_table[] = {
+       DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+};
+
 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 {
        struct inode *inode = filp->f_path.dentry->d_inode;
@@ -559,14 +811,16 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
        struct btrfs_leaf *leaf;
        int slot;
        int advance;
-       unsigned char d_type = DT_UNKNOWN;
+       unsigned char d_type;
        int over = 0;
+       u32 di_cur;
+       u32 di_total;
+       u32 di_len;
        int key_type = BTRFS_DIR_INDEX_KEY;
 
        /* FIXME, use a real flag for deciding about the key type */
        if (root->fs_info->tree_root == root)
                key_type = BTRFS_DIR_ITEM_KEY;
-
        mutex_lock(&root->fs_info->fs_mutex);
        key.objectid = inode->i_ino;
        key.flags = 0;
@@ -575,16 +829,17 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
        path = btrfs_alloc_path();
        btrfs_init_path(path);
        ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
-       if (ret < 0) {
+       if (ret < 0)
                goto err;
-       }
        advance = 0;
+       reada_leaves(root, path, inode->i_ino);
        while(1) {
                leaf = btrfs_buffer_leaf(path->nodes[0]);
                nritems = btrfs_header_nritems(&leaf->header);
                slot = path->slots[0];
                if (advance || slot >= nritems) {
                        if (slot >= nritems -1) {
+                               reada_leaves(root, path, inode->i_ino);
                                ret = btrfs_next_leaf(root, path);
                                if (ret)
                                        break;
@@ -600,23 +855,28 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
                item = leaf->items + slot;
                if (btrfs_disk_key_objectid(&item->key) != key.objectid)
                        break;
-               if (key_type == BTRFS_DIR_INDEX_KEY &&
-                   btrfs_disk_key_offset(&item->key) >
-                   root->fs_info->highest_inode)
-                       break;
                if (btrfs_disk_key_type(&item->key) != key_type)
-                       continue;
+                       break;
                if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
                        continue;
                filp->f_pos = btrfs_disk_key_offset(&item->key);
                advance = 1;
                di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
-               over = filldir(dirent, (const char *)(di + 1),
-                              btrfs_dir_name_len(di),
-                              btrfs_disk_key_offset(&item->key),
-                              btrfs_disk_key_objectid(&di->location), d_type);
-               if (over)
-                       goto nopos;
+               di_cur = 0;
+               di_total = btrfs_item_size(leaf->items + slot);
+               while(di_cur < di_total) {
+                       d_type = btrfs_filetype_table[btrfs_dir_type(di)];
+                       over = filldir(dirent, (const char *)(di + 1),
+                                      btrfs_dir_name_len(di),
+                                      btrfs_disk_key_offset(&item->key),
+                                      btrfs_disk_key_objectid(&di->location),
+                                      d_type);
+                       if (over)
+                               goto nopos;
+                       di_len = btrfs_dir_name_len(di) + sizeof(*di);
+                       di_cur += di_len;
+                       di = (struct btrfs_dir_item *)((char *)di + di_len);
+               }
        }
        filp->f_pos++;
 nopos:
@@ -687,99 +947,63 @@ static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
                return -ENOMEM;
        }
        sb->s_root = root_dentry;
-
+       btrfs_transaction_queue_work(tree_root, HZ * 30);
        return 0;
 }
 
-static void fill_inode_item(struct btrfs_inode_item *item,
-                           struct inode *inode)
+static int btrfs_write_inode(struct inode *inode, int wait)
 {
-       btrfs_set_inode_uid(item, inode->i_uid);
-       btrfs_set_inode_gid(item, inode->i_gid);
-       btrfs_set_inode_size(item, inode->i_size);
-       btrfs_set_inode_mode(item, inode->i_mode);
-       btrfs_set_inode_nlink(item, inode->i_nlink);
-       btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
-       btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
-       btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
-       btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
-       btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
-       btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
-       btrfs_set_inode_nblocks(item, inode->i_blocks);
-       btrfs_set_inode_generation(item, inode->i_generation);
-       check_inode(inode);
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct btrfs_trans_handle *trans;
+       int ret = 0;
+
+       if (wait) {
+               mutex_lock(&root->fs_info->fs_mutex);
+               trans = btrfs_start_transaction(root, 1);
+               btrfs_set_trans_block_group(trans, inode);
+               ret = btrfs_commit_transaction(trans, root);
+               mutex_unlock(&root->fs_info->fs_mutex);
+       }
+       return ret;
 }
 
-static int btrfs_update_inode(struct btrfs_trans_handle *trans,
-                             struct btrfs_root *root,
-                             struct inode *inode)
+static void btrfs_dirty_inode(struct inode *inode)
 {
-       struct btrfs_inode_item *inode_item;
-       struct btrfs_path *path;
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct btrfs_trans_handle *trans;
+
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, inode);
+       btrfs_update_inode(trans, root, inode);
+       btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
+}
+
+static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
+                                    struct btrfs_root *root,
+                                    u64 objectid,
+                                    struct btrfs_block_group_cache *group,
+                                    int mode)
+{
+       struct inode *inode;
+       struct btrfs_inode_item inode_item;
+       struct btrfs_key *location;
        int ret;
+       int owner;
 
-       path = btrfs_alloc_path();
-       BUG_ON(!path);
-       btrfs_init_path(path);
+       inode = new_inode(root->fs_info->sb);
+       if (!inode)
+               return ERR_PTR(-ENOMEM);
 
-       ret = btrfs_lookup_inode(trans, root, path,
-                                &BTRFS_I(inode)->location, 1);
-       if (ret) {
-               if (ret > 0)
-                       ret = -ENOENT;
-               goto failed;
-       }
-
-       inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
-                                 path->slots[0],
-                                 struct btrfs_inode_item);
-
-       fill_inode_item(inode_item, inode);
-       btrfs_mark_buffer_dirty(path->nodes[0]);
-failed:
-       btrfs_release_path(root, path);
-       btrfs_free_path(path);
-       check_inode(inode);
-       return 0;
-}
-
-static int btrfs_write_inode(struct inode *inode, int wait)
-{
-       struct btrfs_root *root = BTRFS_I(inode)->root;
-       struct btrfs_trans_handle *trans;
-       int ret;
-
-       mutex_lock(&root->fs_info->fs_mutex);
-       trans = btrfs_start_transaction(root, 1);
-       ret = btrfs_update_inode(trans, root, inode);
-       if (wait)
-               btrfs_commit_transaction(trans, root);
+       BTRFS_I(inode)->root = root;
+       if (mode & S_IFDIR)
+               owner = 0;
        else
-               btrfs_end_transaction(trans, root);
-       mutex_unlock(&root->fs_info->fs_mutex);
-       check_inode(inode);
-       return ret;
-}
-
-static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
-                                    struct inode *dir, int mode)
-{
-       struct inode *inode;
-       struct btrfs_inode_item inode_item;
-       struct btrfs_root *root = BTRFS_I(dir)->root;
-       struct btrfs_key *key;
-       int ret;
-       u64 objectid;
-
-       inode = new_inode(dir->i_sb);
-       if (!inode)
-               return ERR_PTR(-ENOMEM);
-
-       BTRFS_I(inode)->root = BTRFS_I(dir)->root;
-       key = &BTRFS_I(inode)->location;
-       check_inode(inode);
-       ret = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
-       BUG_ON(ret);
+               owner = 1;
+       group = btrfs_find_block_group(root, group, 0, 0, owner);
+       BTRFS_I(inode)->block_group = group;
 
        inode->i_uid = current->fsuid;
        inode->i_gid = current->fsgid;
@@ -788,23 +1012,24 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
        inode->i_blocks = 0;
        inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
        fill_inode_item(&inode_item, inode);
-
-       key->objectid = objectid;
-       key->flags = 0;
-       key->offset = 0;
-       btrfs_set_key_type(key, BTRFS_INODE_ITEM_KEY);
-       ret = btrfs_insert_inode_map(trans, root, objectid, key);
-       BUG_ON(ret);
+       location = &BTRFS_I(inode)->location;
+       location->objectid = objectid;
+       location->flags = 0;
+       location->offset = 0;
+       btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 
        ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
        BUG_ON(ret);
 
        insert_inode_hash(inode);
-       check_inode(inode);
-       check_inode(dir);
        return inode;
 }
 
+static inline u8 btrfs_inode_type(struct inode *inode)
+{
+       return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
+}
+
 static int btrfs_add_link(struct btrfs_trans_handle *trans,
                            struct dentry *dentry, struct inode *inode)
 {
@@ -819,14 +1044,12 @@ static int btrfs_add_link(struct btrfs_trans_handle *trans,
        ret = btrfs_insert_dir_item(trans, root,
                                    dentry->d_name.name, dentry->d_name.len,
                                    dentry->d_parent->d_inode->i_ino,
-                                   &key, 0);
+                                   &key, btrfs_inode_type(inode));
        if (ret == 0) {
                dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
                ret = btrfs_update_inode(trans, root,
                                         dentry->d_parent->d_inode);
        }
-       check_inode(inode);
-       check_inode(dentry->d_parent->d_inode);
        return ret;
 }
 
@@ -840,7 +1063,6 @@ static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
        }
        if (err > 0)
                err = -EEXIST;
-       check_inode(inode);
        return err;
 }
 
@@ -852,14 +1074,25 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
        struct inode *inode;
        int err;
        int drop_inode = 0;
+       u64 objectid;
 
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
-       inode = btrfs_new_inode(trans, dir, mode);
+       btrfs_set_trans_block_group(trans, dir);
+
+       err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+       if (err) {
+               err = -ENOSPC;
+               goto out_unlock;
+       }
+
+       inode = btrfs_new_inode(trans, root, objectid,
+                               BTRFS_I(dir)->block_group, mode);
        err = PTR_ERR(inode);
        if (IS_ERR(inode))
                goto out_unlock;
-       // FIXME mark the inode dirty
+
+       btrfs_set_trans_block_group(trans, inode);
        err = btrfs_add_nondir(trans, dentry, inode);
        if (err)
                drop_inode = 1;
@@ -869,23 +1102,59 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                inode->i_op = &btrfs_file_inode_operations;
        }
        dir->i_sb->s_dirt = 1;
+       btrfs_update_inode_block_group(trans, inode);
+       btrfs_update_inode_block_group(trans, dir);
 out_unlock:
        btrfs_end_transaction(trans, root);
        mutex_unlock(&root->fs_info->fs_mutex);
-       check_inode(inode);
-       check_inode(dir);
 
        if (drop_inode) {
                inode_dec_link_count(inode);
                iput(inode);
        }
+       btrfs_btree_balance_dirty(root);
        return err;
 }
 
-static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
-                               struct inode *inode, struct inode *dir)
+static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
+                     struct dentry *dentry)
 {
+       struct btrfs_trans_handle *trans;
        struct btrfs_root *root = BTRFS_I(dir)->root;
+       struct inode *inode = old_dentry->d_inode;
+       int err;
+       int drop_inode = 0;
+
+       if (inode->i_nlink == 0)
+               return -ENOENT;
+
+       inc_nlink(inode);
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, dir);
+       atomic_inc(&inode->i_count);
+       err = btrfs_add_nondir(trans, dentry, inode);
+       if (err)
+               drop_inode = 1;
+       dir->i_sb->s_dirt = 1;
+       btrfs_update_inode_block_group(trans, dir);
+       btrfs_update_inode(trans, root, inode);
+
+       btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+
+       if (drop_inode) {
+               inode_dec_link_count(inode);
+               iput(inode);
+       }
+       btrfs_btree_balance_dirty(root);
+       return err;
+}
+
+static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root,
+                               u64 objectid, u64 dirid)
+{
        int ret;
        char buf[2];
        struct btrfs_key key;
@@ -893,22 +1162,20 @@ static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
        buf[0] = '.';
        buf[1] = '.';
 
-       key.objectid = inode->i_ino;
+       key.objectid = objectid;
        key.offset = 0;
        key.flags = 0;
        btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 
-       ret = btrfs_insert_dir_item(trans, root, buf, 1, inode->i_ino,
-                                   &key, 1);
+       ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
+                                   &key, BTRFS_FT_DIR);
        if (ret)
                goto error;
-       key.objectid = dir->i_ino;
-       ret = btrfs_insert_dir_item(trans, root, buf, 2, inode->i_ino,
-                                   &key, 1);
+       key.objectid = dirid;
+       ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
+                                   &key, BTRFS_FT_DIR);
        if (ret)
                goto error;
-       inode->i_size = 6;
-       ret = btrfs_update_inode(trans, root, inode);
 error:
        return ret;
 }
@@ -920,14 +1187,24 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
        struct btrfs_root *root = BTRFS_I(dir)->root;
        int err = 0;
        int drop_on_err = 0;
+       u64 objectid;
 
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, dir);
        if (IS_ERR(trans)) {
                err = PTR_ERR(trans);
                goto out_unlock;
        }
-       inode = btrfs_new_inode(trans, dir, S_IFDIR | mode);
+
+       err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+       if (err) {
+               err = -ENOSPC;
+               goto out_unlock;
+       }
+
+       inode = btrfs_new_inode(trans, root, objectid,
+                               BTRFS_I(dir)->block_group, S_IFDIR | mode);
        if (IS_ERR(inode)) {
                err = PTR_ERR(inode);
                goto out_fail;
@@ -935,8 +1212,14 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
        drop_on_err = 1;
        inode->i_op = &btrfs_dir_inode_operations;
        inode->i_fop = &btrfs_dir_file_operations;
+       btrfs_set_trans_block_group(trans, inode);
+
+       err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
+       if (err)
+               goto out_fail;
 
-       err = btrfs_make_empty_dir(trans, inode, dir);
+       inode->i_size = 6;
+       err = btrfs_update_inode(trans, root, inode);
        if (err)
                goto out_fail;
        err = btrfs_add_link(trans, dentry, inode);
@@ -944,6 +1227,9 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
                goto out_fail;
        d_instantiate(dentry, inode);
        drop_on_err = 0;
+       dir->i_sb->s_dirt = 1;
+       btrfs_update_inode_block_group(trans, inode);
+       btrfs_update_inode_block_group(trans, dir);
 
 out_fail:
        btrfs_end_transaction(trans, root);
@@ -951,9 +1237,30 @@ out_unlock:
        mutex_unlock(&root->fs_info->fs_mutex);
        if (drop_on_err)
                iput(inode);
+       btrfs_btree_balance_dirty(root);
        return err;
 }
 
+static int btrfs_sync_file(struct file *file,
+                          struct dentry *dentry, int datasync)
+{
+       struct inode *inode = dentry->d_inode;
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       int ret;
+       struct btrfs_trans_handle *trans;
+
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       if (!trans) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       ret = btrfs_commit_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+out:
+       return ret > 0 ? EIO : ret;
+}
+
 static int btrfs_sync_fs(struct super_block *sb, int wait)
 {
        struct btrfs_trans_handle *trans;
@@ -966,7 +1273,6 @@ static int btrfs_sync_fs(struct super_block *sb, int wait)
                filemap_flush(root->fs_info->btree_inode->i_mapping);
                return 0;
        }
-       filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
        ret = btrfs_commit_transaction(trans, root);
@@ -977,74 +1283,9 @@ printk("btrfs sync_fs\n");
        return 0;
 }
 
-#if 0
-static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
-                          struct buffer_head *result, int create)
-{
-       struct btrfs_root *root = btrfs_sb(inode->i_sb);
-       struct btrfs_path *path;
-       struct btrfs_key key;
-       struct btrfs_leaf *leaf;
-       int num_bytes = result->b_size;
-       int item_size;
-       int ret;
-       u64 pos;
-       char *ptr;
-       int copy_size;
-       int err = 0;
-       char *safe_ptr;
-       char *data_ptr;
-
-       path = btrfs_alloc_path();
-       BUG_ON(!path);
-
-       WARN_ON(create);
-       if (create) {
-               return 0;
-       }
-       pos = iblock << inode->i_blkbits;
-       key.objectid = inode->i_ino;
-       key.flags = 0;
-       btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
-       ptr = kmap(result->b_page);
-       safe_ptr = ptr;
-       ptr += (pos & (PAGE_CACHE_SIZE -1));
-again:
-       key.offset = pos;
-       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
-       if (ret) {
-               if (ret < 0)
-                       err = ret;
-               else
-                       err = 0;
-               goto out;
-       }
-       leaf = btrfs_buffer_leaf(path->nodes[0]);
-       item_size = btrfs_item_size(leaf->items + path->slots[0]);
-       copy_size = min(num_bytes, item_size);
-       data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
-       WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
-       memcpy(ptr, data_ptr, copy_size);
-       pos += copy_size;
-       num_bytes -= copy_size;
-       WARN_ON(num_bytes < 0);
-       ptr += copy_size;
-       btrfs_release_path(root, path);
-       if (num_bytes != 0) {
-               if (pos >= i_size_read(inode))
-                       memset(ptr, 0, num_bytes);
-               else
-                       goto again;
-       }
-       set_buffer_uptodate(result);
-       map_bh(result, inode->i_sb, 0);
-       err = 0;
-out:
-       btrfs_free_path(path);
-       kunmap(result->b_page);
-       return err;
-}
-#endif
+#define BTRFS_GET_BLOCK_NO_CREATE 0
+#define BTRFS_GET_BLOCK_CREATE 1
+#define BTRFS_GET_BLOCK_NO_DIRECT 2
 
 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
                           struct buffer_head *result, int create)
@@ -1055,20 +1296,34 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
        u64 extent_start = 0;
        u64 extent_end = 0;
        u64 objectid = inode->i_ino;
+       u32 found_type;
+       u64 alloc_hint = 0;
        struct btrfs_path *path;
        struct btrfs_root *root = BTRFS_I(inode)->root;
-       struct btrfs_trans_handle *trans = NULL;
        struct btrfs_file_extent_item *item;
        struct btrfs_leaf *leaf;
        struct btrfs_disk_key *found_key;
+       struct btrfs_trans_handle *trans = NULL;
 
        path = btrfs_alloc_path();
        BUG_ON(!path);
        btrfs_init_path(path);
-       if (create)
+       if (create & BTRFS_GET_BLOCK_CREATE) {
+               WARN_ON(1);
+               /* this almost but not quite works */
                trans = btrfs_start_transaction(root, 1);
+               if (!trans) {
+                       err = -ENOMEM;
+                       goto out;
+               }
+               ret = drop_extents(trans, root, inode,
+                                  iblock << inode->i_blkbits,
+                                  (iblock + 1) << inode->i_blkbits,
+                                  &alloc_hint);
+               BUG_ON(ret);
+       }
 
-       ret = btrfs_lookup_file_extent(trans, root, path,
+       ret = btrfs_lookup_file_extent(NULL, root, path,
                                       inode->i_ino,
                                       iblock << inode->i_blkbits, 0);
        if (ret < 0) {
@@ -1079,7 +1334,7 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
        if (ret != 0) {
                if (path->slots[0] == 0) {
                        btrfs_release_path(root, path);
-                       goto allocate;
+                       goto not_found;
                }
                path->slots[0]--;
        }
@@ -1090,54 +1345,73 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
        blocknr = btrfs_file_extent_disk_blocknr(item);
        blocknr += btrfs_file_extent_offset(item);
 
-       /* exact match found, use it */
-       if (ret == 0) {
-               err = 0;
-               map_bh(result, inode->i_sb, blocknr);
-               goto out;
-       }
-
        /* are we inside the extent that was found? */
        found_key = &leaf->items[path->slots[0]].key;
+       found_type = btrfs_disk_key_type(found_key);
        if (btrfs_disk_key_objectid(found_key) != objectid ||
-           btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
+           found_type != BTRFS_EXTENT_DATA_KEY) {
                extent_end = 0;
                extent_start = 0;
-               btrfs_release_path(root, path);
-               goto allocate;
+               goto not_found;
        }
-
+       found_type = btrfs_file_extent_type(item);
        extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
-       extent_start = extent_start >> inode->i_blkbits;
-       extent_start += btrfs_file_extent_offset(item);
-       extent_end = extent_start + btrfs_file_extent_num_blocks(item);
-       if (iblock >= extent_start && iblock < extent_end) {
+       if (found_type == BTRFS_FILE_EXTENT_REG) {
+               extent_start = extent_start >> inode->i_blkbits;
+               extent_end = extent_start + btrfs_file_extent_num_blocks(item);
                err = 0;
-               map_bh(result, inode->i_sb, blocknr + iblock - extent_start);
-               goto out;
-       }
-allocate:
-       /* ok, create a new extent */
-       if (!create) {
-               err = 0;
-               goto out;
+               if (btrfs_file_extent_disk_blocknr(item) == 0)
+                       goto out;
+               if (iblock >= extent_start && iblock < extent_end) {
+                       btrfs_map_bh_to_logical(root, result, blocknr +
+                                               iblock - extent_start);
+                       goto out;
+               }
+       } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+               char *ptr;
+               char *map;
+               u32 size;
+
+               if (create & BTRFS_GET_BLOCK_NO_DIRECT) {
+                       err = -EINVAL;
+                       goto out;
+               }
+               size = btrfs_file_extent_inline_len(leaf->items +
+                                                   path->slots[0]);
+               extent_end = (extent_start + size) >> inode->i_blkbits;
+               extent_start >>= inode->i_blkbits;
+               if (iblock < extent_start || iblock > extent_end) {
+                       goto not_found;
+               }
+               ptr = btrfs_file_extent_inline_start(item);
+               map = kmap(result->b_page);
+               memcpy(map, ptr, size);
+               memset(map + size, 0, PAGE_CACHE_SIZE - size);
+               flush_dcache_page(result->b_page);
+               kunmap(result->b_page);
+               set_buffer_uptodate(result);
+               SetPageChecked(result->b_page);
+               btrfs_map_bh_to_logical(root, result, 0);
        }
-       ret = btrfs_alloc_file_extent(trans, root, objectid,
-                                     iblock << inode->i_blkbits,
-                                     1, extent_end, &blocknr);
-       if (ret) {
-               err = ret;
-               goto out;
+not_found:
+       if (create & BTRFS_GET_BLOCK_CREATE) {
+               struct btrfs_key ins;
+               ret = btrfs_alloc_extent(trans, root, inode->i_ino,
+                                        1, alloc_hint, (u64)-1,
+                                        &ins, 1);
+               BUG_ON(ret);
+               ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
+                                              iblock << inode->i_blkbits,
+                                              ins.objectid, ins.offset,
+                                              ins.offset);
+               BUG_ON(ret);
+               SetPageChecked(result->b_page);
+               btrfs_map_bh_to_logical(root, result, ins.objectid);
        }
-       inode->i_blocks += inode->i_sb->s_blocksize >> 9;
-       set_buffer_new(result);
-       map_bh(result, inode->i_sb, blocknr);
-
 out:
-       btrfs_release_path(root, path);
-       btrfs_free_path(path);
        if (trans)
-               btrfs_end_transaction(trans, root);
+               err = btrfs_end_transaction(trans, root);
+       btrfs_free_path(path);
        return err;
 }
 
@@ -1148,25 +1422,34 @@ static int btrfs_get_block(struct inode *inode, sector_t iblock,
        struct btrfs_root *root = BTRFS_I(inode)->root;
        mutex_lock(&root->fs_info->fs_mutex);
        err = btrfs_get_block_lock(inode, iblock, result, create);
-       // err = btrfs_get_block_inline(inode, iblock, result, create);
        mutex_unlock(&root->fs_info->fs_mutex);
        return err;
 }
 
-static int btrfs_prepare_write(struct file *file, struct page *page,
-                              unsigned from, unsigned to)
+static int btrfs_get_block_bmap(struct inode *inode, sector_t iblock,
+                          struct buffer_head *result, int create)
+{
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       mutex_lock(&root->fs_info->fs_mutex);
+       btrfs_get_block_lock(inode, iblock, result, BTRFS_GET_BLOCK_NO_DIRECT);
+       mutex_unlock(&root->fs_info->fs_mutex);
+       return 0;
+}
+
+static sector_t btrfs_bmap(struct address_space *as, sector_t block)
 {
-       return nobh_prepare_write(page, from, to, btrfs_get_block);
+       return generic_block_bmap(as, block, btrfs_get_block_bmap);
 }
-static int btrfs_commit_write(struct file *file, struct page *page,
+
+static int btrfs_prepare_write(struct file *file, struct page *page,
                               unsigned from, unsigned to)
 {
-       return nobh_commit_write(file, page, from, to);
+       return block_prepare_write(page, from, to, btrfs_get_block);
 }
 
 static void btrfs_write_super(struct super_block *sb)
 {
-       btrfs_sync_fs(sb, 1);
+       sb->s_dirt = 0;
 }
 
 static int btrfs_readpage(struct file *file, struct page *page)
@@ -1174,15 +1457,222 @@ static int btrfs_readpage(struct file *file, struct page *page)
        return mpage_readpage(page, btrfs_get_block);
 }
 
-static int btrfs_readpages(struct file *file, struct address_space *mapping,
-                          struct list_head *pages, unsigned nr_pages)
+/*
+ * While block_write_full_page is writing back the dirty buffers under
+ * the page lock, whoever dirtied the buffers may decide to clean them
+ * again at any time.  We handle that by only looking at the buffer
+ * state inside lock_buffer().
+ *
+ * If block_write_full_page() is called for regular writeback
+ * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
+ * locked buffer.   This only can happen if someone has written the buffer
+ * directly, with submit_bh().  At the address_space level PageWriteback
+ * prevents this contention from occurring.
+ */
+static int __btrfs_write_full_page(struct inode *inode, struct page *page,
+                                  struct writeback_control *wbc)
 {
-       return mpage_readpages(mapping, pages, nr_pages, btrfs_get_block);
+       int err;
+       sector_t block;
+       sector_t last_block;
+       struct buffer_head *bh, *head;
+       const unsigned blocksize = 1 << inode->i_blkbits;
+       int nr_underway = 0;
+
+       BUG_ON(!PageLocked(page));
+
+       last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
+
+       if (!page_has_buffers(page)) {
+               create_empty_buffers(page, blocksize,
+                                       (1 << BH_Dirty)|(1 << BH_Uptodate));
+       }
+
+       /*
+        * Be very careful.  We have no exclusion from __set_page_dirty_buffers
+        * here, and the (potentially unmapped) buffers may become dirty at
+        * any time.  If a buffer becomes dirty here after we've inspected it
+        * then we just miss that fact, and the page stays dirty.
+        *
+        * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
+        * handle that here by just cleaning them.
+        */
+
+       block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
+       head = page_buffers(page);
+       bh = head;
+
+       /*
+        * Get all the dirty buffers mapped to disk addresses and
+        * handle any aliases from the underlying blockdev's mapping.
+        */
+       do {
+               if (block > last_block) {
+                       /*
+                        * mapped buffers outside i_size will occur, because
+                        * this page can be outside i_size when there is a
+                        * truncate in progress.
+                        */
+                       /*
+                        * The buffer was zeroed by block_write_full_page()
+                        */
+                       clear_buffer_dirty(bh);
+                       set_buffer_uptodate(bh);
+               } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
+                       WARN_ON(bh->b_size != blocksize);
+                       err = btrfs_get_block(inode, block, bh, 0);
+                       if (err) {
+printk("writepage going to recovery err %d\n", err);
+                               goto recover;
+                       }
+                       if (buffer_new(bh)) {
+                               /* blockdev mappings never come here */
+                               clear_buffer_new(bh);
+                       }
+               }
+               bh = bh->b_this_page;
+               block++;
+       } while (bh != head);
+
+       do {
+               if (!buffer_mapped(bh))
+                       continue;
+               /*
+                * If it's a fully non-blocking write attempt and we cannot
+                * lock the buffer then redirty the page.  Note that this can
+                * potentially cause a busy-wait loop from pdflush and kswapd
+                * activity, but those code paths have their own higher-level
+                * throttling.
+                */
+               if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
+                       lock_buffer(bh);
+               } else if (test_set_buffer_locked(bh)) {
+                       redirty_page_for_writepage(wbc, page);
+                       continue;
+               }
+               if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
+                       mark_buffer_async_write(bh);
+               } else {
+                       unlock_buffer(bh);
+               }
+       } while ((bh = bh->b_this_page) != head);
+
+       /*
+        * The page and its buffers are protected by PageWriteback(), so we can
+        * drop the bh refcounts early.
+        */
+       BUG_ON(PageWriteback(page));
+       set_page_writeback(page);
+
+       do {
+               struct buffer_head *next = bh->b_this_page;
+               if (buffer_async_write(bh)) {
+                       submit_bh(WRITE, bh);
+                       nr_underway++;
+               }
+               bh = next;
+       } while (bh != head);
+       unlock_page(page);
+
+       err = 0;
+done:
+       if (nr_underway == 0) {
+               /*
+                * The page was marked dirty, but the buffers were
+                * clean.  Someone wrote them back by hand with
+                * ll_rw_block/submit_bh.  A rare case.
+                */
+               int uptodate = 1;
+               do {
+                       if (!buffer_uptodate(bh)) {
+                               uptodate = 0;
+                               break;
+                       }
+                       bh = bh->b_this_page;
+               } while (bh != head);
+               if (uptodate)
+                       SetPageUptodate(page);
+               end_page_writeback(page);
+       }
+       return err;
+
+recover:
+       /*
+        * ENOSPC, or some other error.  We may already have added some
+        * blocks to the file, so we need to write these out to avoid
+        * exposing stale data.
+        * The page is currently locked and not marked for writeback
+        */
+       bh = head;
+       /* Recovery: lock and submit the mapped buffers */
+       do {
+               if (buffer_mapped(bh) && buffer_dirty(bh)) {
+                       lock_buffer(bh);
+                       mark_buffer_async_write(bh);
+               } else {
+                       /*
+                        * The buffer may have been set dirty during
+                        * attachment to a dirty page.
+                        */
+                       clear_buffer_dirty(bh);
+               }
+       } while ((bh = bh->b_this_page) != head);
+       SetPageError(page);
+       BUG_ON(PageWriteback(page));
+       set_page_writeback(page);
+       do {
+               struct buffer_head *next = bh->b_this_page;
+               if (buffer_async_write(bh)) {
+                       clear_buffer_dirty(bh);
+                       submit_bh(WRITE, bh);
+                       nr_underway++;
+               }
+               bh = next;
+       } while (bh != head);
+       unlock_page(page);
+       goto done;
 }
 
+/*
+ * The generic ->writepage function for buffer-backed address_spaces
+ */
 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
 {
-       return nobh_writepage(page, btrfs_get_block, wbc);
+       struct inode * const inode = page->mapping->host;
+       loff_t i_size = i_size_read(inode);
+       const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
+       unsigned offset;
+       void *kaddr;
+
+       /* Is the page fully inside i_size? */
+       if (page->index < end_index)
+               return __btrfs_write_full_page(inode, page, wbc);
+
+       /* Is the page fully outside i_size? (truncate in progress) */
+       offset = i_size & (PAGE_CACHE_SIZE-1);
+       if (page->index >= end_index+1 || !offset) {
+               /*
+                * The page may have dirty, unmapped buffers.  For example,
+                * they may have been added in ext3_writepage().  Make them
+                * freeable here, so the page does not leak.
+                */
+               block_invalidatepage(page, 0);
+               unlock_page(page);
+               return 0; /* don't care */
+       }
+
+       /*
+        * The page straddles i_size.  It must be zeroed out on each and every
+        * writepage invokation because it may be mmapped.  "A file is mapped
+        * in multiples of the page size.  For a file that is not a multiple of
+        * the  page size, the remaining memory is zeroed when mapped, and
+        * writes to that region are not written out to the file."
+        */
+       kaddr = kmap_atomic(page, KM_USER0);
+       memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
+       flush_dcache_page(page);
+       kunmap_atomic(kaddr, KM_USER0);
+       return __btrfs_write_full_page(inode, page, wbc);
 }
 
 static void btrfs_truncate(struct inode *inode)
@@ -1196,17 +1686,40 @@ static void btrfs_truncate(struct inode *inode)
        if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
                return;
 
-       nobh_truncate_page(inode->i_mapping, inode->i_size);
+       btrfs_truncate_page(inode->i_mapping, inode->i_size);
 
-       /* FIXME, add redo link to tree so we don't leak on crash */
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, inode);
+
+       /* FIXME, add redo link to tree so we don't leak on crash */
        ret = btrfs_truncate_in_trans(trans, root, inode);
        BUG_ON(ret);
+       btrfs_update_inode(trans, root, inode);
        ret = btrfs_end_transaction(trans, root);
        BUG_ON(ret);
        mutex_unlock(&root->fs_info->fs_mutex);
-       mark_inode_dirty(inode);
+       btrfs_btree_balance_dirty(root);
+}
+
+static int btrfs_commit_write(struct file *file, struct page *page,
+                             unsigned from, unsigned to)
+{
+       struct inode *inode = page->mapping->host;
+       struct buffer_head *bh;
+       loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
+
+       SetPageUptodate(page);
+       bh = page_buffers(page);
+       set_buffer_uptodate(bh);
+       if (buffer_mapped(bh) && bh->b_blocknr != 0) {
+               set_page_dirty(page);
+       }
+       if (pos > inode->i_size) {
+               i_size_write(inode, pos);
+               mark_inode_dirty(inode);
+       }
+       return 0;
 }
 
 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
@@ -1264,6 +1777,8 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
        int ret;
        int this_write;
        struct inode *inode = file->f_path.dentry->d_inode;
+       struct buffer_head *bh;
+       struct btrfs_file_extent_item *ei;
 
        for (i = 0; i < num_pages; i++) {
                offset = pos & (PAGE_CACHE_SIZE -1);
@@ -1272,16 +1787,50 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 
                mutex_lock(&root->fs_info->fs_mutex);
                trans = btrfs_start_transaction(root, 1);
-               btrfs_csum_file_block(trans, root, inode->i_ino,
+               btrfs_set_trans_block_group(trans, inode);
+
+               bh = page_buffers(pages[i]);
+               if (buffer_mapped(bh) && bh->b_blocknr == 0) {
+                       struct btrfs_key key;
+                       struct btrfs_path *path;
+                       char *ptr;
+                       u32 datasize;
+
+                       path = btrfs_alloc_path();
+                       BUG_ON(!path);
+                       key.objectid = inode->i_ino;
+                       key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
+                       key.flags = 0;
+                       btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+                       BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
+                       datasize = offset +
+                               btrfs_file_extent_calc_inline_size(write_bytes);
+                       ret = btrfs_insert_empty_item(trans, root, path, &key,
+                                                     datasize);
+                       BUG_ON(ret);
+                       ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+                              path->slots[0], struct btrfs_file_extent_item);
+                       btrfs_set_file_extent_generation(ei, trans->transid);
+                       btrfs_set_file_extent_type(ei,
+                                                  BTRFS_FILE_EXTENT_INLINE);
+                       ptr = btrfs_file_extent_inline_start(ei);
+                       btrfs_memcpy(root, path->nodes[0]->b_data,
+                                    ptr, bh->b_data, offset + write_bytes);
+                       mark_buffer_dirty(path->nodes[0]);
+                       btrfs_free_path(path);
+               } else if (buffer_mapped(bh)) {
+                       btrfs_csum_file_block(trans, root, inode->i_ino,
                                      pages[i]->index << PAGE_CACHE_SHIFT,
                                      kmap(pages[i]), PAGE_CACHE_SIZE);
-               kunmap(pages[i]);
+                       kunmap(pages[i]);
+               }
                SetPageChecked(pages[i]);
+               // btrfs_update_inode_block_group(trans, inode);
                ret = btrfs_end_transaction(trans, root);
                BUG_ON(ret);
                mutex_unlock(&root->fs_info->fs_mutex);
 
-               ret = nobh_commit_write(file, pages[i], offset,
+               ret = btrfs_commit_write(file, pages[i], offset,
                                         offset + this_write);
                pos += this_write;
                if (ret) {
@@ -1295,21 +1844,226 @@ failed:
        return err;
 }
 
-static int prepare_pages(struct btrfs_trans_handle *trans,
-                        struct btrfs_root *root,
+static int drop_extents(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root,
+                         struct inode *inode,
+                         u64 start, u64 end, u64 *hint_block)
+{
+       int ret;
+       struct btrfs_key key;
+       struct btrfs_leaf *leaf;
+       int slot;
+       struct btrfs_file_extent_item *extent;
+       u64 extent_end = 0;
+       int keep;
+       struct btrfs_file_extent_item old;
+       struct btrfs_path *path;
+       u64 search_start = start;
+       int bookend;
+       int found_type;
+       int found_extent;
+       int found_inline;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+       while(1) {
+               btrfs_release_path(root, path);
+               ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
+                                              search_start, -1);
+               if (ret < 0)
+                       goto out;
+               if (ret > 0) {
+                       if (path->slots[0] == 0) {
+                               ret = 0;
+                               goto out;
+                       }
+                       path->slots[0]--;
+               }
+               keep = 0;
+               bookend = 0;
+               found_extent = 0;
+               found_inline = 0;
+               extent = NULL;
+               leaf = btrfs_buffer_leaf(path->nodes[0]);
+               slot = path->slots[0];
+               btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
+               if (key.offset >= end || key.objectid != inode->i_ino) {
+                       ret = 0;
+                       goto out;
+               }
+               if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
+                       ret = 0;
+                       goto out;
+               }
+               extent = btrfs_item_ptr(leaf, slot,
+                                       struct btrfs_file_extent_item);
+               found_type = btrfs_file_extent_type(extent);
+               if (found_type == BTRFS_FILE_EXTENT_REG) {
+                       extent_end = key.offset +
+                               (btrfs_file_extent_num_blocks(extent) <<
+                                inode->i_blkbits);
+                       found_extent = 1;
+               } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+                       found_inline = 1;
+                       extent_end = key.offset +
+                            btrfs_file_extent_inline_len(leaf->items + slot);
+               }
+
+               if (!found_extent && !found_inline) {
+                       ret = 0;
+                       goto out;
+               }
+
+               if (search_start >= extent_end) {
+                       ret = 0;
+                       goto out;
+               }
+
+               if (found_inline) {
+                       u64 mask = root->blocksize - 1;
+                       search_start = (extent_end + mask) & ~mask;
+               } else
+                       search_start = extent_end;
+
+               if (end < extent_end && end >= key.offset) {
+                       if (found_extent) {
+                               u64 disk_blocknr =
+                                       btrfs_file_extent_disk_blocknr(extent);
+                               u64 disk_num_blocks =
+                                     btrfs_file_extent_disk_num_blocks(extent);
+                               memcpy(&old, extent, sizeof(old));
+                               if (disk_blocknr != 0) {
+                                       ret = btrfs_inc_extent_ref(trans, root,
+                                                disk_blocknr, disk_num_blocks);
+                                       BUG_ON(ret);
+                               }
+                       }
+                       WARN_ON(found_inline);
+                       bookend = 1;
+               }
+
+               if (start > key.offset) {
+                       u64 new_num;
+                       u64 old_num;
+                       /* truncate existing extent */
+                       keep = 1;
+                       WARN_ON(start & (root->blocksize - 1));
+                       if (found_extent) {
+                               new_num = (start - key.offset) >>
+                                       inode->i_blkbits;
+                               old_num = btrfs_file_extent_num_blocks(extent);
+                               *hint_block =
+                                       btrfs_file_extent_disk_blocknr(extent);
+                               if (btrfs_file_extent_disk_blocknr(extent)) {
+                                       inode->i_blocks -=
+                                               (old_num - new_num) << 3;
+                               }
+                               btrfs_set_file_extent_num_blocks(extent,
+                                                                new_num);
+                               mark_buffer_dirty(path->nodes[0]);
+                       } else {
+                               WARN_ON(1);
+                       }
+               }
+               if (!keep) {
+                       u64 disk_blocknr = 0;
+                       u64 disk_num_blocks = 0;
+                       u64 extent_num_blocks = 0;
+                       if (found_extent) {
+                               disk_blocknr =
+                                     btrfs_file_extent_disk_blocknr(extent);
+                               disk_num_blocks =
+                                     btrfs_file_extent_disk_num_blocks(extent);
+                               extent_num_blocks =
+                                     btrfs_file_extent_num_blocks(extent);
+                               *hint_block =
+                                       btrfs_file_extent_disk_blocknr(extent);
+                       }
+                       ret = btrfs_del_item(trans, root, path);
+                       BUG_ON(ret);
+                       btrfs_release_path(root, path);
+                       extent = NULL;
+                       if (found_extent && disk_blocknr != 0) {
+                               inode->i_blocks -= extent_num_blocks << 3;
+                               ret = btrfs_free_extent(trans, root,
+                                                       disk_blocknr,
+                                                       disk_num_blocks, 0);
+                       }
+
+                       BUG_ON(ret);
+                       if (!bookend && search_start >= end) {
+                               ret = 0;
+                               goto out;
+                       }
+                       if (!bookend)
+                               continue;
+               }
+               if (bookend && found_extent) {
+                       /* create bookend */
+                       struct btrfs_key ins;
+                       ins.objectid = inode->i_ino;
+                       ins.offset = end;
+                       ins.flags = 0;
+                       btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
+
+                       btrfs_release_path(root, path);
+                       ret = btrfs_insert_empty_item(trans, root, path, &ins,
+                                                     sizeof(*extent));
+                       BUG_ON(ret);
+                       extent = btrfs_item_ptr(
+                                   btrfs_buffer_leaf(path->nodes[0]),
+                                   path->slots[0],
+                                   struct btrfs_file_extent_item);
+                       btrfs_set_file_extent_disk_blocknr(extent,
+                                   btrfs_file_extent_disk_blocknr(&old));
+                       btrfs_set_file_extent_disk_num_blocks(extent,
+                                   btrfs_file_extent_disk_num_blocks(&old));
+
+                       btrfs_set_file_extent_offset(extent,
+                                   btrfs_file_extent_offset(&old) +
+                                   ((end - key.offset) >> inode->i_blkbits));
+                       WARN_ON(btrfs_file_extent_num_blocks(&old) <
+                               (extent_end - end) >> inode->i_blkbits);
+                       btrfs_set_file_extent_num_blocks(extent,
+                                   (extent_end - end) >> inode->i_blkbits);
+
+                       btrfs_set_file_extent_type(extent,
+                                                  BTRFS_FILE_EXTENT_REG);
+                       btrfs_set_file_extent_generation(extent,
+                                   btrfs_file_extent_generation(&old));
+                       btrfs_mark_buffer_dirty(path->nodes[0]);
+                       if (btrfs_file_extent_disk_blocknr(&old) != 0) {
+                               inode->i_blocks +=
+                                     btrfs_file_extent_num_blocks(extent) << 3;
+                       }
+                       ret = 0;
+                       goto out;
+               }
+       }
+out:
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int prepare_pages(struct btrfs_root *root,
                         struct file *file,
                         struct page **pages,
                         size_t num_pages,
                         loff_t pos,
-                        size_t write_bytes)
+                        unsigned long first_index,
+                        unsigned long last_index,
+                        size_t write_bytes,
+                        u64 alloc_extent_start)
 {
        int i;
        unsigned long index = pos >> PAGE_CACHE_SHIFT;
        struct inode *inode = file->f_path.dentry->d_inode;
        int offset;
        int err = 0;
-       int ret;
        int this_write;
+       struct buffer_head *bh;
+       struct buffer_head *head;
        loff_t isize = i_size_read(inode);
 
        memset(pages, 0, num_pages * sizeof(struct page *));
@@ -1320,16 +2074,28 @@ static int prepare_pages(struct btrfs_trans_handle *trans,
                        err = -ENOMEM;
                        goto failed_release;
                }
+               cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
+               wait_on_page_writeback(pages[i]);
                offset = pos & (PAGE_CACHE_SIZE -1);
                this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
-               ret = nobh_prepare_write(pages[i], offset,
-                                        offset + this_write,
-                                        btrfs_get_block);
-               pos += this_write;
-               if (ret) {
-                       err = ret;
-                       goto failed_truncate;
+               if (!page_has_buffers(pages[i])) {
+                       create_empty_buffers(pages[i],
+                                            root->fs_info->sb->s_blocksize,
+                                            (1 << BH_Uptodate));
                }
+               head = page_buffers(pages[i]);
+               bh = head;
+               do {
+                       err = btrfs_map_bh_to_logical(root, bh,
+                                                     alloc_extent_start);
+                       BUG_ON(err);
+                       if (err)
+                               goto failed_truncate;
+                       bh = bh->b_this_page;
+                       if (alloc_extent_start)
+                               alloc_extent_start++;
+               } while (bh != head);
+               pos += this_write;
                WARN_ON(this_write > write_bytes);
                write_bytes -= this_write;
        }
@@ -1355,169 +2121,21 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        int ret = 0;
        struct inode *inode = file->f_path.dentry->d_inode;
        struct btrfs_root *root = BTRFS_I(inode)->root;
-       struct page *pages[1];
-
-       if (file->f_flags & O_DIRECT)
-               return -EINVAL;
-       pos = *ppos;
-
-       vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
-       current->backing_dev_info = inode->i_mapping->backing_dev_info;
-       err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
-       if (err)
-               goto out;
-       if (count == 0)
-               goto out;
-       err = remove_suid(file->f_path.dentry);
-       if (err)
-               goto out;
-       file_update_time(file);
-       mutex_lock(&inode->i_mutex);
-       while(count > 0) {
-               size_t offset = pos & (PAGE_CACHE_SIZE - 1);
-               size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
-               size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
-                                       PAGE_CACHE_SHIFT;
-               ret = prepare_pages(NULL, root, file, pages, num_pages,
-                                   pos, write_bytes);
-               BUG_ON(ret);
-               ret = btrfs_copy_from_user(pos, num_pages,
-                                          write_bytes, pages, buf);
-               BUG_ON(ret);
-
-               ret = dirty_and_release_pages(NULL, root, file, pages,
-                                             num_pages, pos, write_bytes);
-               BUG_ON(ret);
-               btrfs_drop_pages(pages, num_pages);
-
-               buf += write_bytes;
-               count -= write_bytes;
-               pos += write_bytes;
-               num_written += write_bytes;
-
-               balance_dirty_pages_ratelimited(inode->i_mapping);
-               cond_resched();
-       }
-       mutex_unlock(&inode->i_mutex);
-out:
-       *ppos = pos;
-       current->backing_dev_info = NULL;
-       return num_written ? num_written : err;
-}
-
-#if 0
-static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
-                          struct page *page, loff_t pos,
-                          size_t offset, size_t write_bytes)
-{
-       struct btrfs_path *path;
+       struct page *pages[8];
+       struct page *pinned[2];
+       unsigned long first_index;
+       unsigned long last_index;
+       u64 start_pos;
+       u64 num_blocks;
+       u64 alloc_extent_start;
+       u64 hint_block;
        struct btrfs_trans_handle *trans;
-       struct btrfs_key key;
-       struct btrfs_leaf *leaf;
-       struct btrfs_key found_key;
-       int ret;
-       size_t copy_size = 0;
-       char *dst = NULL;
-       int err = 0;
-       size_t num_written = 0;
-
-       path = btrfs_alloc_path();
-       BUG_ON(!path);
-       mutex_lock(&root->fs_info->fs_mutex);
-       trans = btrfs_start_transaction(root, 1);
-       key.objectid = inode->i_ino;
-       key.flags = 0;
-       btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
-
-again:
-       key.offset = pos;
-       ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
-       if (ret < 0) {
-               err = ret;
-               goto out;
-       }
-       if (ret == 0) {
-               leaf = btrfs_buffer_leaf(path->nodes[0]);
-               btrfs_disk_key_to_cpu(&found_key,
-                                     &leaf->items[path->slots[0]].key);
-               copy_size = btrfs_item_size(leaf->items + path->slots[0]);
-               dst = btrfs_item_ptr(leaf, path->slots[0], char);
-               copy_size = min(write_bytes, copy_size);
-               goto copyit;
-       } else {
-               int slot = path->slots[0];
-               if (slot > 0) {
-                       slot--;
-               }
-               // FIXME find max key
-               leaf = btrfs_buffer_leaf(path->nodes[0]);
-               btrfs_disk_key_to_cpu(&found_key,
-                                     &leaf->items[slot].key);
-               if (found_key.objectid != inode->i_ino)
-                       goto insert;
-               if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
-                       goto insert;
-               copy_size = btrfs_item_size(leaf->items + slot);
-               if (found_key.offset + copy_size <= pos)
-                       goto insert;
-               dst = btrfs_item_ptr(leaf, path->slots[0], char);
-               dst += pos - found_key.offset;
-               copy_size = copy_size - (pos - found_key.offset);
-               BUG_ON(copy_size < 0);
-               copy_size = min(write_bytes, copy_size);
-               WARN_ON(copy_size == 0);
-               goto copyit;
-       }
-insert:
-       btrfs_release_path(root, path);
-       copy_size = min(write_bytes,
-                       (size_t)BTRFS_LEAF_DATA_SIZE(root) -
-                       sizeof(struct btrfs_item) * 4);
-       ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
-       BUG_ON(ret);
-       dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
-                            path->slots[0], char);
-copyit:
-       WARN_ON(copy_size == 0);
-       WARN_ON(dst + copy_size >
-               btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
-                                                path->slots[0], char) +
-               btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
-                                                 path->slots[0]));
-       btrfs_memcpy(root, path->nodes[0]->b_data, dst,
-                    page_address(page) + offset, copy_size);
-       mark_buffer_dirty(path->nodes[0]);
-       btrfs_release_path(root, path);
-       pos += copy_size;
-       offset += copy_size;
-       num_written += copy_size;
-       write_bytes -= copy_size;
-       if (write_bytes)
-               goto again;
-out:
-       btrfs_free_path(path);
-       ret = btrfs_end_transaction(trans, root);
-       BUG_ON(ret);
-       mutex_unlock(&root->fs_info->fs_mutex);
-       return num_written ? num_written : err;
-}
-
-static ssize_t btrfs_file_inline_write(struct file *file,
-                                      const char __user *buf,
-                                      size_t count, loff_t *ppos)
-{
-       loff_t pos;
-       size_t num_written = 0;
-       int err = 0;
-       int ret = 0;
-       struct inode *inode = file->f_path.dentry->d_inode;
-       struct btrfs_root *root = BTRFS_I(inode)->root;
-       unsigned long page_index;
-
+       struct btrfs_key ins;
+       pinned[0] = NULL;
+       pinned[1] = NULL;
        if (file->f_flags & O_DIRECT)
                return -EINVAL;
        pos = *ppos;
-
        vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
        current->backing_dev_info = inode->i_mapping->backing_dev_info;
        err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
@@ -1529,49 +2147,135 @@ static ssize_t btrfs_file_inline_write(struct file *file,
        if (err)
                goto out;
        file_update_time(file);
+
+       start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
+       num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
+                       inode->i_blkbits;
+
        mutex_lock(&inode->i_mutex);
-       while(count > 0) {
-               size_t offset = pos & (PAGE_CACHE_SIZE - 1);
-               size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
-               struct page *page;
+       first_index = pos >> PAGE_CACHE_SHIFT;
+       last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 
-               page_index = pos >> PAGE_CACHE_SHIFT;
-               page = grab_cache_page(inode->i_mapping, page_index);
-               if (!PageUptodate(page)) {
-                       ret = mpage_readpage(page, btrfs_get_block);
+       if ((pos & (PAGE_CACHE_SIZE - 1))) {
+               pinned[0] = grab_cache_page(inode->i_mapping, first_index);
+               if (!PageUptodate(pinned[0])) {
+                       ret = mpage_readpage(pinned[0], btrfs_get_block);
+                       BUG_ON(ret);
+                       wait_on_page_locked(pinned[0]);
+               } else {
+                       unlock_page(pinned[0]);
+               }
+       }
+       if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
+               pinned[1] = grab_cache_page(inode->i_mapping, last_index);
+               if (!PageUptodate(pinned[1])) {
+                       ret = mpage_readpage(pinned[1], btrfs_get_block);
                        BUG_ON(ret);
-                       lock_page(page);
+                       wait_on_page_locked(pinned[1]);
+               } else {
+                       unlock_page(pinned[1]);
                }
-               ret = btrfs_copy_from_user(pos, 1,
-                                          write_bytes, &page, buf);
+       }
+
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       if (!trans) {
+               err = -ENOMEM;
+               mutex_unlock(&root->fs_info->fs_mutex);
+               goto out_unlock;
+       }
+       btrfs_set_trans_block_group(trans, inode);
+       /* FIXME blocksize != 4096 */
+       inode->i_blocks += num_blocks << 3;
+       hint_block = 0;
+       if (start_pos < inode->i_size) {
+               /* FIXME blocksize != pagesize */
+               ret = drop_extents(trans, root, inode,
+                                  start_pos,
+                                  (pos + count + root->blocksize -1) &
+                                  ~((u64)root->blocksize - 1), &hint_block);
                BUG_ON(ret);
-               write_bytes = inline_one_page(root, inode, page, pos,
-                                     offset, write_bytes);
-               SetPageUptodate(page);
-               if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
-                       i_size_write(inode, pos + write_bytes);
-                       mark_inode_dirty(inode);
+       }
+       if (inode->i_size < start_pos) {
+               u64 last_pos_in_file;
+               u64 hole_size;
+               u64 mask = root->blocksize - 1;
+               last_pos_in_file = (inode->i_size + mask) & ~mask;
+               hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
+               hole_size >>= inode->i_blkbits;
+               if (last_pos_in_file < start_pos) {
+                       ret = btrfs_insert_file_extent(trans, root,
+                                                      inode->i_ino,
+                                                      last_pos_in_file,
+                                                      0, 0, hole_size);
                }
-               page_cache_release(page);
-               unlock_page(page);
-               if (write_bytes < 0)
-                       goto out_unlock;
+               BUG_ON(ret);
+       }
+       if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
+           pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
+               ret = btrfs_alloc_extent(trans, root, inode->i_ino,
+                                        num_blocks, hint_block, (u64)-1,
+                                        &ins, 1);
+               BUG_ON(ret);
+               ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
+                                      start_pos, ins.objectid, ins.offset,
+                                      ins.offset);
+               BUG_ON(ret);
+       } else {
+               ins.offset = 0;
+               ins.objectid = 0;
+       }
+       BUG_ON(ret);
+       alloc_extent_start = ins.objectid;
+       // btrfs_update_inode_block_group(trans, inode);
+       ret = btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+
+       while(count > 0) {
+               size_t offset = pos & (PAGE_CACHE_SIZE - 1);
+               size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
+               size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
+                                       PAGE_CACHE_SHIFT;
+
+               memset(pages, 0, sizeof(pages));
+               ret = prepare_pages(root, file, pages, num_pages,
+                                   pos, first_index, last_index,
+                                   write_bytes, alloc_extent_start);
+               BUG_ON(ret);
+
+               /* FIXME blocks != pagesize */
+               if (alloc_extent_start)
+                       alloc_extent_start += num_pages;
+               ret = btrfs_copy_from_user(pos, num_pages,
+                                          write_bytes, pages, buf);
+               BUG_ON(ret);
+
+               ret = dirty_and_release_pages(NULL, root, file, pages,
+                                             num_pages, pos, write_bytes);
+               BUG_ON(ret);
+               btrfs_drop_pages(pages, num_pages);
+
                buf += write_bytes;
                count -= write_bytes;
                pos += write_bytes;
                num_written += write_bytes;
 
                balance_dirty_pages_ratelimited(inode->i_mapping);
+               btrfs_btree_balance_dirty(root);
                cond_resched();
        }
 out_unlock:
        mutex_unlock(&inode->i_mutex);
 out:
+       if (pinned[0])
+               page_cache_release(pinned[0]);
+       if (pinned[1])
+               page_cache_release(pinned[1]);
        *ppos = pos;
        current->backing_dev_info = NULL;
+       mark_inode_dirty(inode);
        return num_written ? num_written : err;
 }
-#endif
 
 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
                        unsigned long offset, unsigned long size)
@@ -1586,19 +2290,34 @@ static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
        if (!PageChecked(page)) {
                /* FIXME, do it per block */
                struct btrfs_root *root = BTRFS_I(inode)->root;
-               int ret = btrfs_csum_verify_file_block(root,
-                                         page->mapping->host->i_ino,
-                                         page->index << PAGE_CACHE_SHIFT,
-                                         kmap(page), PAGE_CACHE_SIZE);
+               int ret;
+               struct buffer_head *bh;
+
+               if (page_has_buffers(page)) {
+                       bh = page_buffers(page);
+                       if (!buffer_mapped(bh)) {
+                               SetPageChecked(page);
+                               goto checked;
+                       }
+               }
+
+               ret = btrfs_csum_verify_file_block(root,
+                                 page->mapping->host->i_ino,
+                                 page->index << PAGE_CACHE_SHIFT,
+                                 kmap(page), PAGE_CACHE_SIZE);
                if (ret) {
-                       printk("failed to verify ino %lu page %lu\n",
-                              page->mapping->host->i_ino,
-                              page->index);
-                       memset(page_address(page), 0, PAGE_CACHE_SIZE);
+                       if (ret != -ENOENT) {
+                               printk("failed to verify ino %lu page %lu ret %d\n",
+                                      page->mapping->host->i_ino,
+                                      page->index, ret);
+                               memset(page_address(page), 1, PAGE_CACHE_SIZE);
+                               flush_dcache_page(page);
+                       }
                }
                SetPageChecked(page);
                kunmap(page);
        }
+checked:
        /*
         * Faults on the destination of a read are common, so do it before
         * taking the kmap.
@@ -1686,6 +2405,107 @@ static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
        return retval;
 }
 
+static int create_subvol(struct btrfs_root *root, char *name, int namelen)
+{
+       struct btrfs_trans_handle *trans;
+       struct btrfs_key key;
+       struct btrfs_root_item root_item;
+       struct btrfs_inode_item *inode_item;
+       struct buffer_head *subvol;
+       struct btrfs_leaf *leaf;
+       struct btrfs_root *new_root;
+       struct inode *inode;
+       struct inode *dir;
+       int ret;
+       u64 objectid;
+       u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
+
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       BUG_ON(!trans);
+
+       subvol = btrfs_alloc_free_block(trans, root, 0);
+       if (subvol == NULL)
+               return -ENOSPC;
+       leaf = btrfs_buffer_leaf(subvol);
+       btrfs_set_header_nritems(&leaf->header, 0);
+       btrfs_set_header_level(&leaf->header, 0);
+       btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
+       btrfs_set_header_generation(&leaf->header, trans->transid);
+       btrfs_set_header_owner(&leaf->header, root->root_key.objectid);
+       memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
+              sizeof(leaf->header.fsid));
+       mark_buffer_dirty(subvol);
+
+       inode_item = &root_item.inode;
+       memset(inode_item, 0, sizeof(*inode_item));
+       btrfs_set_inode_generation(inode_item, 1);
+       btrfs_set_inode_size(inode_item, 3);
+       btrfs_set_inode_nlink(inode_item, 1);
+       btrfs_set_inode_nblocks(inode_item, 1);
+       btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
+
+       btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
+       btrfs_set_root_refs(&root_item, 1);
+       brelse(subvol);
+       subvol = NULL;
+
+       ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
+                                      0, &objectid);
+       BUG_ON(ret);
+
+       btrfs_set_root_dirid(&root_item, new_dirid);
+
+       key.objectid = objectid;
+       key.offset = 1;
+       key.flags = 0;
+       btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+       ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
+                               &root_item);
+       BUG_ON(ret);
+
+       /*
+        * insert the directory item
+        */
+       key.offset = (u64)-1;
+       dir = root->fs_info->sb->s_root->d_inode;
+       ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
+                                   name, namelen, dir->i_ino, &key,
+                                   BTRFS_FT_DIR);
+       BUG_ON(ret);
+
+       ret = btrfs_commit_transaction(trans, root);
+       BUG_ON(ret);
+
+       new_root = btrfs_read_fs_root(root->fs_info, &key);
+       BUG_ON(!new_root);
+
+       trans = btrfs_start_transaction(new_root, 1);
+       BUG_ON(!trans);
+
+       inode = btrfs_new_inode(trans, new_root, new_dirid,
+                               BTRFS_I(dir)->block_group, S_IFDIR | 0700);
+       inode->i_op = &btrfs_dir_inode_operations;
+       inode->i_fop = &btrfs_dir_file_operations;
+
+       ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
+       BUG_ON(ret);
+
+       inode->i_nlink = 1;
+       inode->i_size = 6;
+       ret = btrfs_update_inode(trans, new_root, inode);
+       BUG_ON(ret);
+
+       ret = btrfs_commit_transaction(trans, new_root);
+       BUG_ON(ret);
+
+       iput(inode);
+
+       mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
+       return 0;
+}
+
 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 {
        struct btrfs_trans_handle *trans;
@@ -1694,6 +2514,9 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
        int ret;
        u64 objectid;
 
+       if (!root->ref_cows)
+               return -EINVAL;
+
        mutex_lock(&root->fs_info->fs_mutex);
        trans = btrfs_start_transaction(root, 1);
        BUG_ON(!trans);
@@ -1701,32 +2524,23 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
        ret = btrfs_update_inode(trans, root, root->inode);
        BUG_ON(ret);
 
-       ret = btrfs_find_free_objectid(trans, root, 0, &objectid);
+       ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
+                                      0, &objectid);
        BUG_ON(ret);
 
-       memset(&new_root_item, 0, sizeof(new_root_item));
        memcpy(&new_root_item, &root->root_item,
               sizeof(new_root_item));
 
-       key.objectid = objectid;
-       key.flags = 0;
-       key.offset = 0;
-       btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
-       ret = btrfs_insert_inode_map(trans, root, objectid, &key);
-       BUG_ON(ret);
-
        key.objectid = objectid;
        key.offset = 1;
        key.flags = 0;
        btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
-       btrfs_set_root_blocknr(&new_root_item, root->node->b_blocknr);
+       btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
 
        ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
                                &new_root_item);
        BUG_ON(ret);
 
-printk("adding snapshot name %.*s root %Lu %Lu %u\n", namelen, name, key.objectid, key.offset, key.flags);
-
        /*
         * insert the directory item
         */
@@ -1734,7 +2548,7 @@ printk("adding snapshot name %.*s root %Lu %Lu %u\n", namelen, name, key.objecti
        ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
                                    name, namelen,
                                    root->fs_info->sb->s_root->d_inode->i_ino,
-                                   &key, 0);
+                                   &key, BTRFS_FT_DIR);
 
        BUG_ON(ret);
 
@@ -1744,6 +2558,7 @@ printk("adding snapshot name %.*s root %Lu %Lu %u\n", namelen, name, key.objecti
        ret = btrfs_commit_transaction(trans, root);
        BUG_ON(ret);
        mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
        return 0;
 }
 
@@ -1752,11 +2567,12 @@ static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
 {
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_ioctl_vol_args vol_args;
-       int ret;
+       int ret = 0;
+       struct btrfs_dir_item *di;
        int namelen;
+       struct btrfs_path *path;
+       u64 root_dirid;
 
-       if (!root->ref_cows)
-               return -EINVAL;
        switch (cmd) {
        case BTRFS_IOC_SNAP_CREATE:
                if (copy_from_user(&vol_args,
@@ -1766,14 +2582,44 @@ static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
                namelen = strlen(vol_args.name);
                if (namelen > BTRFS_VOL_NAME_MAX)
                        return -EINVAL;
-               ret = create_snapshot(root, vol_args.name, namelen);
+               path = btrfs_alloc_path();
+               if (!path)
+                       return -ENOMEM;
+               root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
+               mutex_lock(&root->fs_info->fs_mutex);
+               di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
+                                   path, root_dirid,
+                                   vol_args.name, namelen, 0);
+               mutex_unlock(&root->fs_info->fs_mutex);
+               btrfs_free_path(path);
+               if (di && !IS_ERR(di))
+                       return -EEXIST;
+
+               if (root == root->fs_info->tree_root)
+                       ret = create_subvol(root, vol_args.name, namelen);
+               else
+                       ret = create_snapshot(root, vol_args.name, namelen);
                WARN_ON(ret);
                break;
        default:
                return -ENOTTY;
        }
-       return 0;
+       return ret;
+}
+
+#ifdef CONFIG_COMPAT
+static long btrfs_compat_ioctl(struct file *file, unsigned int cmd,
+                              unsigned long arg)
+{
+       struct inode *inode = file->f_path.dentry->d_inode;
+       int ret;
+       lock_kernel();
+       ret = btrfs_ioctl(inode, file, cmd, (unsigned long) compat_ptr(arg));
+       unlock_kernel();
+       return ret;
+
 }
+#endif
 
 static struct kmem_cache *btrfs_inode_cachep;
 struct kmem_cache *btrfs_trans_handle_cachep;
@@ -1791,21 +2637,14 @@ static struct inode *btrfs_alloc_inode(struct super_block *sb)
        ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
        if (!ei)
                return NULL;
-       ei->magic = 0xDEADBEEF;
-       ei->magic2 = 0xDEADBEAF;
        return &ei->vfs_inode;
 }
 
 static void btrfs_destroy_inode(struct inode *inode)
 {
-       struct btrfs_inode *ei = BTRFS_I(inode);
-       WARN_ON(ei->magic != 0xDEADBEEF);
-       WARN_ON(ei->magic2 != 0xDEADBEAF);
        WARN_ON(!list_empty(&inode->i_dentry));
        WARN_ON(inode->i_data.nrpages);
 
-       ei->magic = 0;
-       ei->magic2 = 0;
        kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
 }
 
@@ -1814,7 +2653,7 @@ static void init_once(void * foo, struct kmem_cache * cachep,
 {
        struct btrfs_inode *ei = (struct btrfs_inode *) foo;
 
-       if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
+       if ((flags & (SLAB_CTOR_CONSTRUCTOR)) ==
            SLAB_CTOR_CONSTRUCTOR) {
                inode_init_once(&ei->vfs_inode);
        }
@@ -1870,6 +2709,220 @@ static int btrfs_get_sb(struct file_system_type *fs_type,
                           btrfs_fill_super, mnt);
 }
 
+static int btrfs_getattr(struct vfsmount *mnt,
+                        struct dentry *dentry, struct kstat *stat)
+{
+       struct inode *inode = dentry->d_inode;
+       generic_fillattr(inode, stat);
+       stat->blksize = 256 * 1024;
+       return 0;
+}
+
+static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+       struct btrfs_root *root = btrfs_sb(dentry->d_sb);
+       struct btrfs_super_block *disk_super = root->fs_info->disk_super;
+
+       buf->f_namelen = BTRFS_NAME_LEN;
+       buf->f_blocks = btrfs_super_total_blocks(disk_super);
+       buf->f_bfree = buf->f_blocks - btrfs_super_blocks_used(disk_super);
+       buf->f_bavail = buf->f_bfree;
+       buf->f_bsize = dentry->d_sb->s_blocksize;
+       buf->f_type = BTRFS_SUPER_MAGIC;
+       return 0;
+}
+
+static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
+                          struct inode * new_dir,struct dentry *new_dentry)
+{
+       struct btrfs_trans_handle *trans;
+       struct btrfs_root *root = BTRFS_I(old_dir)->root;
+       struct inode *new_inode = new_dentry->d_inode;
+       struct inode *old_inode = old_dentry->d_inode;
+       struct timespec ctime = CURRENT_TIME;
+       struct btrfs_path *path;
+       struct btrfs_dir_item *di;
+       int ret;
+
+       if (S_ISDIR(old_inode->i_mode) && new_inode &&
+           new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
+               return -ENOTEMPTY;
+       }
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, new_dir);
+       path = btrfs_alloc_path();
+       if (!path) {
+               ret = -ENOMEM;
+               goto out_fail;
+       }
+
+       old_dentry->d_inode->i_nlink++;
+       old_dir->i_ctime = old_dir->i_mtime = ctime;
+       new_dir->i_ctime = new_dir->i_mtime = ctime;
+       old_inode->i_ctime = ctime;
+       if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
+               struct btrfs_key *location = &BTRFS_I(new_dir)->location;
+               u64 old_parent_oid;
+               di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
+                                          "..", 2, -1);
+               if (IS_ERR(di)) {
+                       ret = PTR_ERR(di);
+                       goto out_fail;
+               }
+               if (!di) {
+                       ret = -ENOENT;
+                       goto out_fail;
+               }
+               old_parent_oid = btrfs_disk_key_objectid(&di->location);
+               ret = btrfs_del_item(trans, root, path);
+               if (ret) {
+                       ret = -EIO;
+                       goto out_fail;
+               }
+               btrfs_release_path(root, path);
+
+               di = btrfs_lookup_dir_index_item(trans, root, path,
+                                                old_inode->i_ino,
+                                                old_parent_oid,
+                                                "..", 2, -1);
+               if (IS_ERR(di)) {
+                       ret = PTR_ERR(di);
+                       goto out_fail;
+               }
+               if (!di) {
+                       ret = -ENOENT;
+                       goto out_fail;
+               }
+               ret = btrfs_del_item(trans, root, path);
+               if (ret) {
+                       ret = -EIO;
+                       goto out_fail;
+               }
+               btrfs_release_path(root, path);
+
+               ret = btrfs_insert_dir_item(trans, root, "..", 2,
+                                           old_inode->i_ino, location,
+                                           BTRFS_FT_DIR);
+               if (ret)
+                       goto out_fail;
+       }
+
+
+       ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
+       if (ret)
+               goto out_fail;
+
+       if (new_inode) {
+               new_inode->i_ctime = CURRENT_TIME;
+               ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
+               if (ret)
+                       goto out_fail;
+               if (S_ISDIR(new_inode->i_mode))
+                       clear_nlink(new_inode);
+               else
+                       drop_nlink(new_inode);
+               btrfs_update_inode(trans, root, new_inode);
+       }
+       ret = btrfs_add_link(trans, new_dentry, old_inode);
+       if (ret)
+               goto out_fail;
+
+out_fail:
+       btrfs_free_path(path);
+       btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+       return ret;
+}
+
+static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
+                        const char *symname)
+{
+       struct btrfs_trans_handle *trans;
+       struct btrfs_root *root = BTRFS_I(dir)->root;
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       struct inode *inode;
+       int err;
+       int drop_inode = 0;
+       u64 objectid;
+       int name_len;
+       int datasize;
+       char *ptr;
+       struct btrfs_file_extent_item *ei;
+
+       name_len = strlen(symname) + 1;
+       if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
+               return -ENAMETOOLONG;
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, dir);
+
+       err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
+       if (err) {
+               err = -ENOSPC;
+               goto out_unlock;
+       }
+
+       inode = btrfs_new_inode(trans, root, objectid,
+                               BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
+       err = PTR_ERR(inode);
+       if (IS_ERR(inode))
+               goto out_unlock;
+
+       btrfs_set_trans_block_group(trans, inode);
+       err = btrfs_add_nondir(trans, dentry, inode);
+       if (err)
+               drop_inode = 1;
+       else {
+               inode->i_mapping->a_ops = &btrfs_aops;
+               inode->i_fop = &btrfs_file_operations;
+               inode->i_op = &btrfs_file_inode_operations;
+       }
+       dir->i_sb->s_dirt = 1;
+       btrfs_update_inode_block_group(trans, inode);
+       btrfs_update_inode_block_group(trans, dir);
+       if (drop_inode)
+               goto out_unlock;
+
+       path = btrfs_alloc_path();
+       BUG_ON(!path);
+       key.objectid = inode->i_ino;
+       key.offset = 0;
+       key.flags = 0;
+       btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+       datasize = btrfs_file_extent_calc_inline_size(name_len);
+       err = btrfs_insert_empty_item(trans, root, path, &key,
+                                     datasize);
+       BUG_ON(err);
+       ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+              path->slots[0], struct btrfs_file_extent_item);
+       btrfs_set_file_extent_generation(ei, trans->transid);
+       btrfs_set_file_extent_type(ei,
+                                  BTRFS_FILE_EXTENT_INLINE);
+       ptr = btrfs_file_extent_inline_start(ei);
+       btrfs_memcpy(root, path->nodes[0]->b_data,
+                    ptr, symname, name_len);
+       mark_buffer_dirty(path->nodes[0]);
+       btrfs_free_path(path);
+       inode->i_op = &btrfs_symlink_inode_operations;
+       inode->i_mapping->a_ops = &btrfs_symlink_aops;
+       inode->i_size = name_len - 1;
+       btrfs_update_inode(trans, root, inode);
+       err = 0;
+
+out_unlock:
+       btrfs_end_transaction(trans, root);
+       mutex_unlock(&root->fs_info->fs_mutex);
+
+       if (drop_inode) {
+               inode_dec_link_count(inode);
+               iput(inode);
+       }
+       btrfs_btree_balance_dirty(root);
+       return err;
+}
+
 static struct file_system_type btrfs_fs_type = {
        .owner          = THIS_MODULE,
        .name           = "btrfs",
@@ -1879,23 +2932,28 @@ static struct file_system_type btrfs_fs_type = {
 };
 
 static struct super_operations btrfs_super_ops = {
-       .statfs         = simple_statfs,
        .delete_inode   = btrfs_delete_inode,
        .put_super      = btrfs_put_super,
        .read_inode     = btrfs_read_locked_inode,
        .write_super    = btrfs_write_super,
        .sync_fs        = btrfs_sync_fs,
        .write_inode    = btrfs_write_inode,
+       .dirty_inode    = btrfs_dirty_inode,
        .alloc_inode    = btrfs_alloc_inode,
        .destroy_inode  = btrfs_destroy_inode,
+       .statfs         = btrfs_statfs,
 };
 
 static struct inode_operations btrfs_dir_inode_operations = {
        .lookup         = btrfs_lookup,
        .create         = btrfs_create,
        .unlink         = btrfs_unlink,
+       .link           = btrfs_link,
        .mkdir          = btrfs_mkdir,
        .rmdir          = btrfs_rmdir,
+       .rename         = btrfs_rename,
+       .symlink        = btrfs_symlink,
+       .setattr        = btrfs_setattr,
 };
 
 static struct inode_operations btrfs_dir_ro_inode_operations = {
@@ -1907,19 +2965,29 @@ static struct file_operations btrfs_dir_file_operations = {
        .read           = generic_read_dir,
        .readdir        = btrfs_readdir,
        .ioctl          = btrfs_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = btrfs_compat_ioctl,
+#endif
 };
 
 static struct address_space_operations btrfs_aops = {
        .readpage       = btrfs_readpage,
-       .readpages      = btrfs_readpages,
        .writepage      = btrfs_writepage,
        .sync_page      = block_sync_page,
        .prepare_write  = btrfs_prepare_write,
        .commit_write   = btrfs_commit_write,
+       .bmap           = btrfs_bmap,
+};
+
+static struct address_space_operations btrfs_symlink_aops = {
+       .readpage       = btrfs_readpage,
+       .writepage      = btrfs_writepage,
 };
 
 static struct inode_operations btrfs_file_inode_operations = {
        .truncate       = btrfs_truncate,
+       .getattr        = btrfs_getattr,
+       .setattr        = btrfs_setattr,
 };
 
 static struct file_operations btrfs_file_operations = {
@@ -1930,30 +2998,36 @@ static struct file_operations btrfs_file_operations = {
        .mmap           = generic_file_mmap,
        .open           = generic_file_open,
        .ioctl          = btrfs_ioctl,
+       .fsync          = btrfs_sync_file,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = btrfs_compat_ioctl,
+#endif
+};
+
+static struct inode_operations btrfs_symlink_inode_operations = {
+       .readlink       = generic_readlink,
+       .follow_link    = page_follow_link_light,
+       .put_link       = page_put_link,
 };
 
 static int __init init_btrfs_fs(void)
 {
        int err;
        printk("btrfs loaded!\n");
+       btrfs_init_transaction_sys();
        err = init_inodecache();
        if (err)
                return err;
-       kset_set_kset_s(&btrfs_subsys, fs_subsys);
-       err = subsystem_register(&btrfs_subsys);
-       if (err)
-               goto out;
        return register_filesystem(&btrfs_fs_type);
-out:
        destroy_inodecache();
        return err;
 }
 
 static void __exit exit_btrfs_fs(void)
 {
+       btrfs_exit_transaction_sys();
        destroy_inodecache();
        unregister_filesystem(&btrfs_fs_type);
-       subsystem_unregister(&btrfs_subsys);
        printk("btrfs unloaded\n");
 }