Btrfs: d_type optimization
[linux-2.6-block.git] / fs / btrfs / super.c
index a10e902d31025b29deace9200ba64f58bab4d2df..b58b4cf667667d1b8aba4b9583141509769ed28d 100644 (file)
 #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 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)
 {
        struct btrfs_path *path;
        struct btrfs_inode_item *inode_item;
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_key location;
+       struct btrfs_block_group_cache *alloc_group;
+       u64 alloc_group_block;
        int ret;
 
        path = btrfs_alloc_path();
@@ -82,6 +93,12 @@ 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);
+       ret = radix_tree_gang_lookup(&root->fs_info->block_group_radix,
+                                    (void **)&alloc_group,
+                                    alloc_group_block, 1);
+       BUG_ON(!ret);
+       BTRFS_I(inode)->block_group = alloc_group;
 
        btrfs_free_path(path);
        inode_item = NULL;
@@ -108,7 +125,8 @@ 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;
        }
        return;
@@ -136,9 +154,10 @@ static void fill_inode_item(struct btrfs_inode_item *item,
        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)
@@ -237,9 +256,11 @@ 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;
 }
 
@@ -262,6 +283,7 @@ 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;
@@ -312,6 +334,7 @@ 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;
@@ -338,6 +361,35 @@ static int btrfs_free_inode(struct btrfs_trans_handle *trans,
        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)
@@ -346,25 +398,24 @@ 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;
@@ -373,34 +424,81 @@ 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_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;
-               found_extent = 0;
-               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);
                        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 */
-                               inode->i_blocks -=
-                                       btrfs_file_extent_num_blocks(fi) << 3;
-                               found_extent = 1;
+                               num_dec = btrfs_file_extent_num_blocks(fi) << 3;
+                               if (extent_start != 0) {
+                                       found_extent = 1;
+                                       inode->i_blocks -= num_dec;
+                               }
                        }
                }
-               ret = btrfs_del_item(trans, root, path);
-               BUG_ON(ret);
+               if (del_item) {
+                       ret = btrfs_del_item(trans, root, path);
+                       BUG_ON(ret);
+               } else {
+                       break;
+               }
                btrfs_release_path(root, path);
                if (found_extent) {
                        ret = btrfs_free_extent(trans, root, extent_start,
@@ -416,6 +514,107 @@ error:
        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;
@@ -429,13 +628,13 @@ 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);
@@ -468,7 +667,7 @@ out:
        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)
 {
@@ -499,7 +698,7 @@ 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;
@@ -507,15 +706,15 @@ 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;
@@ -572,6 +771,36 @@ printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_r
        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;
@@ -585,7 +814,7 @@ 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;
@@ -606,12 +835,14 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
        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;
@@ -637,6 +868,7 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
                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),
@@ -731,6 +963,7 @@ static int btrfs_write_inode(struct inode *inode, int wait)
        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);
        }
@@ -744,25 +977,36 @@ static void btrfs_dirty_inode(struct inode *inode)
 
        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, int mode)
+                                    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;
 
        inode = new_inode(root->fs_info->sb);
        if (!inode)
                return ERR_PTR(-ENOMEM);
 
        BTRFS_I(inode)->root = root;
+       if (mode & S_IFDIR)
+               owner = 0;
+       else
+               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;
@@ -784,6 +1028,11 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
        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)
 {
@@ -798,7 +1047,7 @@ 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,
@@ -832,6 +1081,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 
        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) {
@@ -839,11 +1089,13 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                goto out_unlock;
        }
 
-       inode = btrfs_new_inode(trans, root, objectid, mode);
+       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;
@@ -853,6 +1105,8 @@ 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);
@@ -861,6 +1115,42 @@ out_unlock:
                inode_dec_link_count(inode);
                iput(inode);
        }
+       btrfs_btree_balance_dirty(root);
+       return err;
+}
+
+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;
 }
 
@@ -881,12 +1171,12 @@ static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
        btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 
        ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
-                                   &key, 1);
+                                   &key, BTRFS_FT_DIR);
        if (ret)
                goto error;
        key.objectid = dirid;
        ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
-                                   &key, 1);
+                                   &key, BTRFS_FT_DIR);
        if (ret)
                goto error;
 error:
@@ -904,6 +1194,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 
        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;
@@ -915,7 +1206,8 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
                goto out_unlock;
        }
 
-       inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
+       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;
@@ -923,6 +1215,7 @@ 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)
@@ -938,6 +1231,8 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
        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);
@@ -945,6 +1240,7 @@ out_unlock:
        mutex_unlock(&root->fs_info->fs_mutex);
        if (drop_on_err)
                iput(inode);
+       btrfs_btree_balance_dirty(root);
        return err;
 }
 
@@ -980,7 +1276,6 @@ 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);
@@ -991,6 +1286,10 @@ printk("btrfs sync_fs\n");
        return 0;
 }
 
+#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)
 {
@@ -1001,17 +1300,30 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
        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_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(NULL, root, path,
@@ -1025,7 +1337,7 @@ 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 out;
+                       goto not_found;
                }
                path->slots[0]--;
        }
@@ -1043,16 +1355,17 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
            found_type != BTRFS_EXTENT_DATA_KEY) {
                extent_end = 0;
                extent_start = 0;
-               btrfs_release_path(root, path);
-               goto out;
+               goto not_found;
        }
        found_type = btrfs_file_extent_type(item);
        extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
        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;
+               if (btrfs_file_extent_disk_blocknr(item) == 0)
+                       goto out;
                if (iblock >= extent_start && iblock < extent_end) {
-                       err = 0;
                        btrfs_map_bh_to_logical(root, result, blocknr +
                                                iblock - extent_start);
                        goto out;
@@ -1061,12 +1374,17 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
                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 out;
+                       goto not_found;
                }
                ptr = btrfs_file_extent_inline_start(item);
                map = kmap(result->b_page);
@@ -1078,8 +1396,24 @@ static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
                SetPageChecked(result->b_page);
                btrfs_map_bh_to_logical(root, result, 0);
        }
+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);
+       }
 out:
-       btrfs_release_path(root, path);
+       if (trans)
+               err = btrfs_end_transaction(trans, root);
        btrfs_free_path(path);
        return err;
 }
@@ -1095,10 +1429,25 @@ static int btrfs_get_block(struct inode *inode, sector_t iblock,
        return err;
 }
 
+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 generic_block_bmap(as, block, btrfs_get_block_bmap);
+}
+
 static int btrfs_prepare_write(struct file *file, struct page *page,
                               unsigned from, unsigned to)
 {
-       return nobh_prepare_write(page, from, to, btrfs_get_block);
+       return block_prepare_write(page, from, to, btrfs_get_block);
 }
 
 static void btrfs_write_super(struct super_block *sb)
@@ -1175,13 +1524,13 @@ static int __btrfs_write_full_page(struct inode *inode, struct page *page,
                } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
                        WARN_ON(bh->b_size != blocksize);
                        err = btrfs_get_block(inode, block, bh, 0);
-                       if (err)
+                       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);
-                               unmap_underlying_metadata(bh->b_bdev,
-                                                       bh->b_blocknr);
                        }
                }
                bh = bh->b_this_page;
@@ -1247,11 +1596,6 @@ done:
                if (uptodate)
                        SetPageUptodate(page);
                end_page_writeback(page);
-               /*
-                * The page and buffer_heads can be released at any time from
-                * here on.
-                */
-               wbc->pages_skipped++;   /* We didn't write this page */
        }
        return err;
 
@@ -1345,23 +1689,22 @@ 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);
 }
 
-/*
- * Make sure any changes to nobh_commit_write() are reflected in
- * nobh_truncate_page(), since it doesn't call commit_write().
- */
 static int btrfs_commit_write(struct file *file, struct page *page,
                              unsigned from, unsigned to)
 {
@@ -1371,6 +1714,7 @@ static int btrfs_commit_write(struct file *file, struct page *page,
 
        SetPageUptodate(page);
        bh = page_buffers(page);
+       set_buffer_uptodate(bh);
        if (buffer_mapped(bh) && bh->b_blocknr != 0) {
                set_page_dirty(page);
        }
@@ -1446,6 +1790,7 @@ 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_set_trans_block_group(trans, inode);
 
                bh = page_buffers(pages[i]);
                if (buffer_mapped(bh) && bh->b_blocknr == 0) {
@@ -1472,16 +1817,18 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
                        btrfs_set_file_extent_type(ei,
                                                   BTRFS_FILE_EXTENT_INLINE);
                        ptr = btrfs_file_extent_inline_start(ei);
-                       memcpy(ptr, bh->b_data, offset + write_bytes);
+                       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 {
+               } 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]);
                }
                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);
@@ -1503,7 +1850,7 @@ failed:
 static int drop_extents(struct btrfs_trans_handle *trans,
                          struct btrfs_root *root,
                          struct inode *inode,
-                         u64 start, u64 end)
+                         u64 start, u64 end, u64 *hint_block)
 {
        int ret;
        struct btrfs_key key;
@@ -1576,15 +1923,24 @@ static int drop_extents(struct btrfs_trans_handle *trans,
                        goto out;
                }
 
-               search_start = extent_end;
+               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));
-                               ret = btrfs_inc_extent_ref(trans, root,
-                                     btrfs_file_extent_disk_blocknr(&old),
-                                     btrfs_file_extent_disk_num_blocks(&old));
-                               BUG_ON(ret);
+                               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;
@@ -1600,17 +1956,17 @@ static int drop_extents(struct btrfs_trans_handle *trans,
                                new_num = (start - key.offset) >>
                                        inode->i_blkbits;
                                old_num = btrfs_file_extent_num_blocks(extent);
-                               inode->i_blocks -= (old_num - new_num) << 3;
+                               *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);
-                               /*
-                               ret = btrfs_truncate_item(trans, root, path,
-                                                         start - key.offset);
-                               BUG_ON(ret);
-                               */
                        }
                }
                if (!keep) {
@@ -1624,13 +1980,15 @@ static int drop_extents(struct btrfs_trans_handle *trans,
                                      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);
-                       if (found_extent) {
-                               inode->i_blocks -=
-                               btrfs_file_extent_num_blocks(extent) << 3;
+                       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);
@@ -1669,18 +2027,19 @@ static int drop_extents(struct btrfs_trans_handle *trans,
                                    btrfs_file_extent_offset(&old) +
                                    ((end - key.offset) >> inode->i_blkbits));
                        WARN_ON(btrfs_file_extent_num_blocks(&old) <
-                               (end - key.offset) >> inode->i_blkbits);
+                               (extent_end - end) >> inode->i_blkbits);
                        btrfs_set_file_extent_num_blocks(extent,
-                                   btrfs_file_extent_num_blocks(&old) -
-                                   ((end - key.offset) >> inode->i_blkbits));
+                                   (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]);
-                       inode->i_blocks +=
-                               btrfs_file_extent_num_blocks(extent) << 3;
+                       if (btrfs_file_extent_disk_blocknr(&old) != 0) {
+                               inode->i_blocks +=
+                                     btrfs_file_extent_num_blocks(extent) << 3;
+                       }
                        ret = 0;
                        goto out;
                }
@@ -1718,10 +2077,15 @@ static int prepare_pages(struct btrfs_root *root,
                        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);
-               create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
-                                    (1 << BH_Uptodate));
+               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 {
@@ -1761,15 +2125,17 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        struct inode *inode = file->f_path.dentry->d_inode;
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct page *pages[8];
-       struct page *pinned[2] = { NULL, NULL };
+       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 ins;
-
+       pinned[0] = NULL;
+       pinned[1] = NULL;
        if (file->f_flags & O_DIRECT)
                return -EINVAL;
        pos = *ppos;
@@ -1793,23 +2159,22 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        first_index = pos >> PAGE_CACHE_SHIFT;
        last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 
-       if ((first_index << PAGE_CACHE_SHIFT) < inode->i_size &&
-           (pos & (PAGE_CACHE_SIZE - 1))) {
+       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 (first_index != last_index &&
-           (last_index << PAGE_CACHE_SHIFT) < inode->i_size &&
-           (count & (PAGE_CACHE_SIZE - 1))) {
+       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);
+                       wait_on_page_locked(pinned[1]);
                } else {
                        unlock_page(pinned[1]);
                }
@@ -1822,23 +2187,42 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
                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));
+                                  ~((u64)root->blocksize - 1), &hint_block);
+               BUG_ON(ret);
+       }
+       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);
+               }
                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, 1, (u64)-1, &ins);
+                                        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);
+                                      start_pos, ins.objectid, ins.offset,
+                                      ins.offset);
                BUG_ON(ret);
        } else {
                ins.offset = 0;
@@ -1846,6 +2230,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        }
        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);
 
@@ -1879,6 +2264,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
                num_written += write_bytes;
 
                balance_dirty_pages_ratelimited(inode->i_mapping);
+               btrfs_btree_balance_dirty(root);
                cond_resched();
        }
 out_unlock:
@@ -1907,20 +2293,34 @@ 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;
+               struct buffer_head *bh;
+
+               if (page_has_buffers(page)) {
+                       bh = page_buffers(page);
+                       if (!buffer_mapped(bh)) {
+                               SetPageChecked(page);
+                               goto checked;
+                       }
+               }
 
-               int ret = btrfs_csum_verify_file_block(root,
+               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.
@@ -2018,6 +2418,7 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
        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;
@@ -2026,7 +2427,7 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
        trans = btrfs_start_transaction(root, 1);
        BUG_ON(!trans);
 
-       subvol = btrfs_alloc_free_block(trans, root);
+       subvol = btrfs_alloc_free_block(trans, root, 0);
        if (subvol == NULL)
                return -ENOSPC;
        leaf = btrfs_buffer_leaf(subvol);
@@ -2070,10 +2471,10 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
         * 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,
-                                   root->fs_info->sb->s_root->d_inode->i_ino,
-                                   &key, 0);
+                                   name, namelen, dir->i_ino, &key,
+                                   BTRFS_FT_DIR);
        BUG_ON(ret);
 
        ret = btrfs_commit_transaction(trans, root);
@@ -2085,7 +2486,8 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
        trans = btrfs_start_transaction(new_root, 1);
        BUG_ON(!trans);
 
-       inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
+       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;
 
@@ -2103,6 +2505,7 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
        iput(inode);
 
        mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
        return 0;
 }
 
@@ -2148,7 +2551,7 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
        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);
 
@@ -2158,6 +2561,7 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
        ret = btrfs_commit_transaction(trans, root);
        BUG_ON(ret);
        mutex_unlock(&root->fs_info->fs_mutex);
+       btrfs_btree_balance_dirty(root);
        return 0;
 }
 
@@ -2233,6 +2637,7 @@ out:
        mutex_unlock(&root->fs_info->fs_mutex);
 out_nolock:
        btrfs_free_path(path);
+       btrfs_btree_balance_dirty(root);
 
        return ret;
 }
@@ -2325,7 +2730,7 @@ 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);
        }
@@ -2381,7 +2786,6 @@ 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)
 {
@@ -2405,6 +2809,197 @@ static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
        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",
@@ -2430,8 +3025,12 @@ 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 = {
@@ -2451,11 +3050,18 @@ static struct address_space_operations btrfs_aops = {
        .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 = {
@@ -2469,6 +3075,12 @@ static struct file_operations btrfs_file_operations = {
        .fsync          = btrfs_sync_file,
 };
 
+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;
@@ -2476,12 +3088,7 @@ static int __init init_btrfs_fs(void)
        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;
 }
@@ -2490,7 +3097,6 @@ static void __exit exit_btrfs_fs(void)
 {
        destroy_inodecache();
        unregister_filesystem(&btrfs_fs_type);
-       subsystem_unregister(&btrfs_subsys);
        printk("btrfs unloaded\n");
 }