Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec/ocfs2: (85 commits)
  ocfs2: Use buffer IO if we are appending a file.
  ocfs2: add spinlock protection when dealing with lockres->purge.
  dlmglue.c: add missed mlog lines
  ocfs2: __ocfs2_abort() should not enable panic for local mounts
  ocfs2: Add ioctl for reflink.
  ocfs2: Enable refcount tree support.
  ocfs2: Implement ocfs2_reflink.
  ocfs2: Add preserve to reflink.
  ocfs2: Create reflinked file in orphan dir.
  ocfs2: Use proper parameter for some inode operation.
  ocfs2: Make transaction extend more efficient.
  ocfs2: Don't merge in 1st refcount ops of reflink.
  ocfs2: Modify removing xattr process for refcount.
  ocfs2: Add reflink support for xattr.
  ocfs2: Create an xattr indexed block if needed.
  ocfs2: Call refcount tree remove process properly.
  ocfs2: Attach xattr clusters to refcount tree.
  ocfs2: Abstract ocfs2 xattr tree extend rec iteration process.
  ocfs2: Abstract the creation of xattr block.
  ocfs2: Remove inode from ocfs2_xattr_bucket_get_name_value.
  ...

fs/ocfs2/Makefile

@@ -28,6 +28,7 @@ ocfs2-objs := \
 	locks.o			\
 	mmap.o 			\
 	namei.o 		\
+	refcounttree.o		\
 	resize.o		\
 	slot_map.o 		\
 	suballoc.o 		\

fs/ocfs2/alloc.h

@@ -45,7 +45,8 @@
  *
  * ocfs2_extent_tree contains info for the root of the b-tree, it must have a
  * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree
- * functions.  With metadata ecc, we now call different journal_access
+ * functions.  It needs the ocfs2_caching_info structure associated with
+ * I/O on the tree.  With metadata ecc, we now call different journal_access
  * functions for each type of metadata, so it must have the
  * root_journal_access function.
  * ocfs2_extent_tree_operations abstract the normal operations we do for
@@ -56,6 +57,7 @@ struct ocfs2_extent_tree {
 	struct ocfs2_extent_tree_operations	*et_ops;
 	struct buffer_head			*et_root_bh;
 	struct ocfs2_extent_list		*et_root_el;
+	struct ocfs2_caching_info		*et_ci;
 	ocfs2_journal_access_func		et_root_journal_access;
 	void					*et_object;
 	unsigned int				et_max_leaf_clusters;
@@ -66,31 +68,32 @@ struct ocfs2_extent_tree {
  * specified object buffer.
  */
 void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
-				   struct inode *inode,
+				   struct ocfs2_caching_info *ci,
 				   struct buffer_head *bh);
 void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
-				       struct inode *inode,
+				       struct ocfs2_caching_info *ci,
 				       struct buffer_head *bh);
 struct ocfs2_xattr_value_buf;
 void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
-					struct inode *inode,
+					struct ocfs2_caching_info *ci,
 					struct ocfs2_xattr_value_buf *vb);
 void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
-				    struct inode *inode,
+				    struct ocfs2_caching_info *ci,
 				    struct buffer_head *bh);
+void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *bh);
 
 /*
  * Read an extent block into *bh.  If *bh is NULL, a bh will be
  * allocated.  This is a cached read.  The extent block will be validated
  * with ocfs2_validate_extent_block().
  */
-int ocfs2_read_extent_block(struct inode *inode, u64 eb_blkno,
+int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
 			    struct buffer_head **bh);
 
 struct ocfs2_alloc_context;
-int ocfs2_insert_extent(struct ocfs2_super *osb,
-			handle_t *handle,
-			struct inode *inode,
+int ocfs2_insert_extent(handle_t *handle,
 			struct ocfs2_extent_tree *et,
 			u32 cpos,
 			u64 start_blk,
@@ -103,25 +106,36 @@ enum ocfs2_alloc_restarted {
 	RESTART_TRANS,
 	RESTART_META
 };
-int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb,
-				struct inode *inode,
+int ocfs2_add_clusters_in_btree(handle_t *handle,
+				struct ocfs2_extent_tree *et,
 				u32 *logical_offset,
 				u32 clusters_to_add,
 				int mark_unwritten,
-				struct ocfs2_extent_tree *et,
-				handle_t *handle,
 				struct ocfs2_alloc_context *data_ac,
 				struct ocfs2_alloc_context *meta_ac,
 				enum ocfs2_alloc_restarted *reason_ret);
 struct ocfs2_cached_dealloc_ctxt;
+struct ocfs2_path;
+int ocfs2_split_extent(handle_t *handle,
+		       struct ocfs2_extent_tree *et,
+		       struct ocfs2_path *path,
+		       int split_index,
+		       struct ocfs2_extent_rec *split_rec,
+		       struct ocfs2_alloc_context *meta_ac,
+		       struct ocfs2_cached_dealloc_ctxt *dealloc);
 int ocfs2_mark_extent_written(struct inode *inode,
 			      struct ocfs2_extent_tree *et,
 			      handle_t *handle, u32 cpos, u32 len, u32 phys,
 			      struct ocfs2_alloc_context *meta_ac,
 			      struct ocfs2_cached_dealloc_ctxt *dealloc);
-int ocfs2_remove_extent(struct inode *inode,
-			struct ocfs2_extent_tree *et,
-			u32 cpos, u32 len, handle_t *handle,
+int ocfs2_change_extent_flag(handle_t *handle,
+			     struct ocfs2_extent_tree *et,
+			     u32 cpos, u32 len, u32 phys,
+			     struct ocfs2_alloc_context *meta_ac,
+			     struct ocfs2_cached_dealloc_ctxt *dealloc,
+			     int new_flags, int clear_flags);
+int ocfs2_remove_extent(handle_t *handle, struct ocfs2_extent_tree *et,
+			u32 cpos, u32 len,
 			struct ocfs2_alloc_context *meta_ac,
 			struct ocfs2_cached_dealloc_ctxt *dealloc);
 int ocfs2_remove_btree_range(struct inode *inode,
@@ -130,7 +144,6 @@ int ocfs2_remove_btree_range(struct inode *inode,
 			     struct ocfs2_cached_dealloc_ctxt *dealloc);
 
 int ocfs2_num_free_extents(struct ocfs2_super *osb,
-			   struct inode *inode,
 			   struct ocfs2_extent_tree *et);
 
 /*
@@ -195,6 +208,9 @@ static inline void ocfs2_init_dealloc_ctxt(struct ocfs2_cached_dealloc_ctxt *c)
 }
 int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
 				u64 blkno, unsigned int bit);
+int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+			      int type, int slot, u64 blkno,
+			      unsigned int bit);
 static inline int ocfs2_dealloc_has_cluster(struct ocfs2_cached_dealloc_ctxt *c)
 {
 	return c->c_global_allocator != NULL;
@@ -222,8 +238,9 @@ int ocfs2_commit_truncate(struct ocfs2_super *osb,
 int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
 			  unsigned int start, unsigned int end, int trunc);
 
-int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
-		    u32 cpos, struct buffer_head **leaf_bh);
+int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
+		    struct ocfs2_extent_list *root_el, u32 cpos,
+		    struct buffer_head **leaf_bh);
 int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster);
 
 /*
@@ -254,4 +271,50 @@ static inline int ocfs2_is_empty_extent(struct ocfs2_extent_rec *rec)
 	return !rec->e_leaf_clusters;
 }
 
+int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
+		     struct page **pages, int *num);
+void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
+			      unsigned int from, unsigned int to,
+			      struct page *page, int zero, u64 *phys);
+/*
+ * Structures which describe a path through a btree, and functions to
+ * manipulate them.
+ *
+ * The idea here is to be as generic as possible with the tree
+ * manipulation code.
+ */
+struct ocfs2_path_item {
+	struct buffer_head		*bh;
+	struct ocfs2_extent_list	*el;
+};
+
+#define OCFS2_MAX_PATH_DEPTH	5
+
+struct ocfs2_path {
+	int				p_tree_depth;
+	ocfs2_journal_access_func	p_root_access;
+	struct ocfs2_path_item		p_node[OCFS2_MAX_PATH_DEPTH];
+};
+
+#define path_root_bh(_path) ((_path)->p_node[0].bh)
+#define path_root_el(_path) ((_path)->p_node[0].el)
+#define path_root_access(_path)((_path)->p_root_access)
+#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
+#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
+#define path_num_items(_path) ((_path)->p_tree_depth + 1)
+
+void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root);
+void ocfs2_free_path(struct ocfs2_path *path);
+int ocfs2_find_path(struct ocfs2_caching_info *ci,
+		    struct ocfs2_path *path,
+		    u32 cpos);
+struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path);
+struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et);
+int ocfs2_path_bh_journal_access(handle_t *handle,
+				 struct ocfs2_caching_info *ci,
+				 struct ocfs2_path *path,
+				 int idx);
+int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
+			      handle_t *handle,
+			      struct ocfs2_path *path);
 #endif /* OCFS2_ALLOC_H */

fs/ocfs2/aops.c

@@ -44,6 +44,7 @@
 #include "suballoc.h"
 #include "super.h"
 #include "symlink.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -126,8 +127,8 @@ bail:
 	return err;
 }
 
-static int ocfs2_get_block(struct inode *inode, sector_t iblock,
-			   struct buffer_head *bh_result, int create)
+int ocfs2_get_block(struct inode *inode, sector_t iblock,
+		    struct buffer_head *bh_result, int create)
 {
 	int err = 0;
 	unsigned int ext_flags;
@@ -590,6 +591,8 @@ static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
 		goto bail;
 	}
 
+	/* We should already CoW the refcounted extent. */
+	BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
 	/*
 	 * get_more_blocks() expects us to describe a hole by clearing
 	 * the mapped bit on bh_result().
@@ -687,6 +690,10 @@ static ssize_t ocfs2_direct_IO(int rw,
 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 		return 0;
 
+	/* Fallback to buffered I/O if we are appending. */
+	if (i_size_read(inode) <= offset)
+		return 0;
+
 	ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
 					    inode->i_sb->s_bdev, iov, offset,
 					    nr_segs, 
@@ -1259,7 +1266,8 @@ static int ocfs2_write_cluster(struct address_space *mapping,
 			goto out;
 		}
 	} else if (unwritten) {
-		ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
+					      wc->w_di_bh);
 		ret = ocfs2_mark_extent_written(inode, &et,
 						wc->w_handle, cpos, 1, phys,
 						meta_ac, &wc->w_dealloc);
@@ -1448,6 +1456,9 @@ static int ocfs2_populate_write_desc(struct inode *inode,
 				goto out;
 			}
 
+			/* We should already CoW the refcountd extent. */
+			BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
+
 			/*
 			 * Assume worst case - that we're writing in
 			 * the middle of the extent.
@@ -1528,7 +1539,7 @@ static int ocfs2_write_begin_inline(struct address_space *mapping,
 		goto out;
 	}
 
-	ret = ocfs2_journal_access_di(handle, inode, wc->w_di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		ocfs2_commit_trans(osb, handle);
@@ -1699,6 +1710,19 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 		goto out;
 	}
 
+	ret = ocfs2_check_range_for_refcount(inode, pos, len);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	} else if (ret == 1) {
+		ret = ocfs2_refcount_cow(inode, di_bh,
+					 wc->w_cpos, wc->w_clen, UINT_MAX);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
 	ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
 					&extents_to_split);
 	if (ret) {
@@ -1726,7 +1750,8 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 		     (long long)i_size_read(inode), le32_to_cpu(di->i_clusters),
 		     clusters_to_alloc, extents_to_split);
 
-		ocfs2_init_dinode_extent_tree(&et, inode, wc->w_di_bh);
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
+					      wc->w_di_bh);
 		ret = ocfs2_lock_allocators(inode, &et,
 					    clusters_to_alloc, extents_to_split,
 					    &data_ac, &meta_ac);
@@ -1773,7 +1798,7 @@ int ocfs2_write_begin_nolock(struct address_space *mapping,
 	 * We don't want this to fail in ocfs2_write_end(), so do it
 	 * here.
 	 */
-	ret = ocfs2_journal_access_di(handle, inode, wc->w_di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);

fs/ocfs2/aops.h

@@ -57,6 +57,8 @@ int ocfs2_read_inline_data(struct inode *inode, struct page *page,
 			   struct buffer_head *di_bh);
 int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size);
 
+int ocfs2_get_block(struct inode *inode, sector_t iblock,
+		    struct buffer_head *bh_result, int create);
 /* all ocfs2_dio_end_io()'s fault */
 #define ocfs2_iocb_is_rw_locked(iocb) \
 	test_bit(0, (unsigned long *)&iocb->private)

fs/ocfs2/buffer_head_io.c

@@ -52,12 +52,12 @@ enum ocfs2_state_bits {
 BUFFER_FNS(NeedsValidate, needs_validate);
 
 int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
-		      struct inode *inode)
+		      struct ocfs2_caching_info *ci)
 {
 	int ret = 0;
 
-	mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n",
-		   (unsigned long long)bh->b_blocknr, inode);
+	mlog_entry("(bh->b_blocknr = %llu, ci=%p)\n",
+		   (unsigned long long)bh->b_blocknr, ci);
 
 	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
 	BUG_ON(buffer_jbd(bh));
@@ -70,7 +70,7 @@ int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
 		goto out;
 	}
 
-	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_lock(ci);
 
 	lock_buffer(bh);
 	set_buffer_uptodate(bh);
@@ -85,7 +85,7 @@ int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
 	wait_on_buffer(bh);
 
 	if (buffer_uptodate(bh)) {
-		ocfs2_set_buffer_uptodate(inode, bh);
+		ocfs2_set_buffer_uptodate(ci, bh);
 	} else {
 		/* We don't need to remove the clustered uptodate
 		 * information for this bh as it's not marked locally
@@ -94,7 +94,7 @@ int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
 		put_bh(bh);
 	}
 
-	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_unlock(ci);
 out:
 	mlog_exit(ret);
 	return ret;
@@ -177,7 +177,7 @@ bail:
 	return status;
 }
 
-int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
+int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
 		      struct buffer_head *bhs[], int flags,
 		      int (*validate)(struct super_block *sb,
 				      struct buffer_head *bh))
@@ -185,11 +185,12 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 	int status = 0;
 	int i, ignore_cache = 0;
 	struct buffer_head *bh;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
 
-	mlog_entry("(inode=%p, block=(%llu), nr=(%d), flags=%d)\n",
-		   inode, (unsigned long long)block, nr, flags);
+	mlog_entry("(ci=%p, block=(%llu), nr=(%d), flags=%d)\n",
+		   ci, (unsigned long long)block, nr, flags);
 
-	BUG_ON(!inode);
+	BUG_ON(!ci);
 	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
 	       (flags & OCFS2_BH_IGNORE_CACHE));
 
@@ -212,12 +213,12 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 		goto bail;
 	}
 
-	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_lock(ci);
 	for (i = 0 ; i < nr ; i++) {
 		if (bhs[i] == NULL) {
-			bhs[i] = sb_getblk(inode->i_sb, block++);
+			bhs[i] = sb_getblk(sb, block++);
 			if (bhs[i] == NULL) {
-				mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
+				ocfs2_metadata_cache_io_unlock(ci);
 				status = -EIO;
 				mlog_errno(status);
 				goto bail;
@@ -250,11 +251,11 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 		 *    before our is-it-in-flight check.
 		 */
 
-		if (!ignore_cache && !ocfs2_buffer_uptodate(inode, bh)) {
+		if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
 			mlog(ML_UPTODATE,
-			     "bh (%llu), inode %llu not uptodate\n",
+			     "bh (%llu), owner %llu not uptodate\n",
 			     (unsigned long long)bh->b_blocknr,
-			     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+			     (unsigned long long)ocfs2_metadata_cache_owner(ci));
 			/* We're using ignore_cache here to say
 			 * "go to disk" */
 			ignore_cache = 1;
@@ -283,7 +284,7 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 			 * previously submitted request than we are
 			 * done here. */
 			if ((flags & OCFS2_BH_READAHEAD)
-			    && ocfs2_buffer_read_ahead(inode, bh))
+			    && ocfs2_buffer_read_ahead(ci, bh))
 				continue;
 
 			lock_buffer(bh);
@@ -305,7 +306,7 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 			 * buffer lock. */
 			if (!(flags & OCFS2_BH_IGNORE_CACHE)
 			    && !(flags & OCFS2_BH_READAHEAD)
-			    && ocfs2_buffer_uptodate(inode, bh)) {
+			    && ocfs2_buffer_uptodate(ci, bh)) {
 				unlock_buffer(bh);
 				continue;
 			}
@@ -327,7 +328,7 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 
 		if (!(flags & OCFS2_BH_READAHEAD)) {
 			/* We know this can't have changed as we hold the
-			 * inode sem. Avoid doing any work on the bh if the
+			 * owner sem. Avoid doing any work on the bh if the
 			 * journal has it. */
 			if (!buffer_jbd(bh))
 				wait_on_buffer(bh);
@@ -351,7 +352,7 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 				 * that better not have changed */
 				BUG_ON(buffer_jbd(bh));
 				clear_buffer_needs_validate(bh);
-				status = validate(inode->i_sb, bh);
+				status = validate(sb, bh);
 				if (status) {
 					put_bh(bh);
 					bhs[i] = NULL;
@@ -363,9 +364,9 @@ int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
 		/* Always set the buffer in the cache, even if it was
 		 * a forced read, or read-ahead which hasn't yet
 		 * completed. */
-		ocfs2_set_buffer_uptodate(inode, bh);
+		ocfs2_set_buffer_uptodate(ci, bh);
 	}
-	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_unlock(ci);
 
 	mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n", 
 	     (unsigned long long)block, nr,
@@ -399,7 +400,7 @@ static void ocfs2_check_super_or_backup(struct super_block *sb,
 
 /*
  * Write super block and backups doesn't need to collaborate with journal,
- * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed
+ * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
  * into this function.
  */
 int ocfs2_write_super_or_backup(struct ocfs2_super *osb,

fs/ocfs2/buffer_head_io.h

@@ -33,7 +33,7 @@ void ocfs2_end_buffer_io_sync(struct buffer_head *bh,
 
 int ocfs2_write_block(struct ocfs2_super          *osb,
 		      struct buffer_head  *bh,
-		      struct inode        *inode);
+		      struct ocfs2_caching_info   *ci);
 int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
 			   unsigned int nr, struct buffer_head *bhs[]);
 
@@ -44,7 +44,7 @@ int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
  * be set even for a READAHEAD call, as it marks the buffer for later
  * validation.
  */
-int ocfs2_read_blocks(struct inode *inode, u64 block, int nr,
+int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
 		      struct buffer_head *bhs[], int flags,
 		      int (*validate)(struct super_block *sb,
 				      struct buffer_head *bh));
@@ -55,7 +55,7 @@ int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
 #define OCFS2_BH_IGNORE_CACHE      1
 #define OCFS2_BH_READAHEAD         8
 
-static inline int ocfs2_read_block(struct inode *inode, u64 off,
+static inline int ocfs2_read_block(struct ocfs2_caching_info *ci, u64 off,
 				   struct buffer_head **bh,
 				   int (*validate)(struct super_block *sb,
 						   struct buffer_head *bh))
@@ -68,7 +68,7 @@ static inline int ocfs2_read_block(struct inode *inode, u64 off,
 		goto bail;
 	}
 
-	status = ocfs2_read_blocks(inode, off, 1, bh, 0, validate);
+	status = ocfs2_read_blocks(ci, off, 1, bh, 0, validate);
 
 bail:
 	return status;

fs/ocfs2/cluster/masklog.c

@@ -111,6 +111,7 @@ static struct mlog_attribute mlog_attrs[MLOG_MAX_BITS] = {
 	define_mask(EXPORT),
 	define_mask(XATTR),
 	define_mask(QUOTA),
+	define_mask(REFCOUNT),
 	define_mask(ERROR),
 	define_mask(NOTICE),
 	define_mask(KTHREAD),

fs/ocfs2/cluster/masklog.h

@@ -113,6 +113,7 @@
 #define ML_EXPORT	0x0000000010000000ULL /* ocfs2 export operations */
 #define ML_XATTR	0x0000000020000000ULL /* ocfs2 extended attributes */
 #define ML_QUOTA	0x0000000040000000ULL /* ocfs2 quota operations */
+#define ML_REFCOUNT	0x0000000080000000ULL /* refcount tree operations */
 /* bits that are infrequently given and frequently matched in the high word */
 #define ML_ERROR	0x0000000100000000ULL /* sent to KERN_ERR */
 #define ML_NOTICE	0x0000000200000000ULL /* setn to KERN_NOTICE */

fs/ocfs2/dir.c

@@ -176,7 +176,7 @@ static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
 	struct ocfs2_dx_root_block *dx_root;
 	struct ocfs2_dir_block_trailer *trailer;
 
-	ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -564,7 +564,8 @@ static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
 	int ret;
 	struct buffer_head *tmp = *bh;
 
-	ret = ocfs2_read_block(dir, phys, &tmp, ocfs2_validate_dir_block);
+	ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
+			       ocfs2_validate_dir_block);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
@@ -622,7 +623,8 @@ static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
 	u64 blkno = le64_to_cpu(di->i_dx_root);
 	struct buffer_head *tmp = *dx_root_bh;
 
-	ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_root);
+	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
+			       ocfs2_validate_dx_root);
 
 	/* If ocfs2_read_block() got us a new bh, pass it up. */
 	if (!ret && !*dx_root_bh)
@@ -662,7 +664,8 @@ static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
 	int ret;
 	struct buffer_head *tmp = *dx_leaf_bh;
 
-	ret = ocfs2_read_block(dir, blkno, &tmp, ocfs2_validate_dx_leaf);
+	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
+			       ocfs2_validate_dx_leaf);
 
 	/* If ocfs2_read_block() got us a new bh, pass it up. */
 	if (!ret && !*dx_leaf_bh)
@@ -680,7 +683,7 @@ static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
 {
 	int ret;
 
-	ret = ocfs2_read_blocks(dir, start, num, dx_leaf_bhs, 0,
+	ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
 				ocfs2_validate_dx_leaf);
 	if (ret)
 		mlog_errno(ret);
@@ -802,7 +805,8 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
 	struct ocfs2_extent_rec *rec = NULL;
 
 	if (el->l_tree_depth) {
-		ret = ocfs2_find_leaf(inode, el, major_hash, &eb_bh);
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
+				      &eb_bh);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -1133,7 +1137,8 @@ int ocfs2_update_entry(struct inode *dir, handle_t *handle,
 	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 		access = ocfs2_journal_access_di;
 
-	ret = access(handle, dir, de_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	ret = access(handle, INODE_CACHE(dir), de_bh,
+		     OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
@@ -1176,7 +1181,7 @@ static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
 			goto bail;
 		}
 		if (de == de_del)  {
-			status = access(handle, dir, bh,
+			status = access(handle, INODE_CACHE(dir), bh,
 					OCFS2_JOURNAL_ACCESS_WRITE);
 			if (status < 0) {
 				status = -EIO;
@@ -1326,7 +1331,7 @@ static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
 	 * the entry count needs to be updated. Also, we might be
 	 * adding to the start of the free list.
 	 */
-	ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -1334,7 +1339,7 @@ static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
 	}
 
 	if (!ocfs2_dx_root_inline(dx_root)) {
-		ret = ocfs2_journal_access_dl(handle, dir,
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
 					      lookup->dl_dx_leaf_bh,
 					      OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret) {
@@ -1493,7 +1498,7 @@ static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
 	int ret;
 	struct ocfs2_dx_leaf *dx_leaf;
 
-	ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh,
+	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -1523,7 +1528,7 @@ static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
 	struct ocfs2_dx_root_block *dx_root;
 	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
 
-	ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -1645,11 +1650,13 @@ int __ocfs2_add_entry(handle_t *handle,
 		 */
 		if (ocfs2_free_list_at_root(lookup)) {
 			bh = lookup->dl_dx_root_bh;
-			retval = ocfs2_journal_access_dr(handle, dir, bh,
+			retval = ocfs2_journal_access_dr(handle,
+						 INODE_CACHE(dir), bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		} else {
 			bh = lookup->dl_prev_leaf_bh;
-			retval = ocfs2_journal_access_db(handle, dir, bh,
+			retval = ocfs2_journal_access_db(handle,
+						 INODE_CACHE(dir), bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		}
 		if (retval) {
@@ -1700,11 +1707,13 @@ int __ocfs2_add_entry(handle_t *handle,
 			}
 
 			if (insert_bh == parent_fe_bh)
-				status = ocfs2_journal_access_di(handle, dir,
+				status = ocfs2_journal_access_di(handle,
+								 INODE_CACHE(dir),
 								 insert_bh,
 								 OCFS2_JOURNAL_ACCESS_WRITE);
 			else {
-				status = ocfs2_journal_access_db(handle, dir,
+				status = ocfs2_journal_access_db(handle,
+								 INODE_CACHE(dir),
 								 insert_bh,
 					      OCFS2_JOURNAL_ACCESS_WRITE);
 
@@ -2280,7 +2289,7 @@ static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
 	struct ocfs2_inline_data *data = &di->id2.i_data;
 	unsigned int size = le16_to_cpu(data->id_count);
 
-	ret = ocfs2_journal_access_di(handle, inode, di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -2332,9 +2341,9 @@ static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	ocfs2_set_new_buffer_uptodate(inode, new_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
 
-	status = ocfs2_journal_access_db(handle, inode, new_bh,
+	status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -2418,9 +2427,9 @@ static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
 		ret = -EIO;
 		goto out;
 	}
-	ocfs2_set_new_buffer_uptodate(dir, dx_root_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
 
-	ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 				      OCFS2_JOURNAL_ACCESS_CREATE);
 	if (ret < 0) {
 		mlog_errno(ret);
@@ -2454,7 +2463,7 @@ static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
 	if (ret)
 		mlog_errno(ret);
 
-	ret = ocfs2_journal_access_di(handle, dir, di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
 				      OCFS2_JOURNAL_ACCESS_CREATE);
 	if (ret) {
 		mlog_errno(ret);
@@ -2495,9 +2504,9 @@ static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
 		}
 		dx_leaves[i] = bh;
 
-		ocfs2_set_new_buffer_uptodate(dir, bh);
+		ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
 
-		ret = ocfs2_journal_access_dl(handle, dir, bh,
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
 					      OCFS2_JOURNAL_ACCESS_CREATE);
 		if (ret < 0) {
 			mlog_errno(ret);
@@ -2582,7 +2591,6 @@ static int ocfs2_dx_dir_new_cluster(struct inode *dir,
 {
 	int ret;
 	u64 phys_blkno;
-	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 
 	ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
 					 num_dx_leaves, &phys_blkno);
@@ -2591,7 +2599,7 @@ static int ocfs2_dx_dir_new_cluster(struct inode *dir,
 		goto out;
 	}
 
-	ret = ocfs2_insert_extent(osb, handle, dir, et, cpos, phys_blkno, 1, 0,
+	ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
 				  meta_ac);
 	if (ret)
 		mlog_errno(ret);
@@ -2895,7 +2903,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 	struct ocfs2_extent_tree dx_et;
 	int did_quota = 0, bytes_allocated = 0;
 
-	ocfs2_init_dinode_extent_tree(&et, dir, di_bh);
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
 
 	alloc = ocfs2_clusters_for_bytes(sb, bytes);
 	dx_alloc = 0;
@@ -3005,9 +3013,9 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 		goto out_commit;
 	}
 
-	ocfs2_set_new_buffer_uptodate(dir, dirdata_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
 
-	ret = ocfs2_journal_access_db(handle, dir, dirdata_bh,
+	ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
 				      OCFS2_JOURNAL_ACCESS_CREATE);
 	if (ret) {
 		mlog_errno(ret);
@@ -3060,7 +3068,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 	 * We let the later dirent insert modify c/mtime - to the user
 	 * the data hasn't changed.
 	 */
-	ret = ocfs2_journal_access_di(handle, dir, di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
 				      OCFS2_JOURNAL_ACCESS_CREATE);
 	if (ret) {
 		mlog_errno(ret);
@@ -3085,7 +3093,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 	 * This should never fail as our extent list is empty and all
 	 * related blocks have been journaled already.
 	 */
-	ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len,
+	ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
 				  0, NULL);
 	if (ret) {
 		mlog_errno(ret);
@@ -3117,8 +3125,10 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 			ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
 						      dirdata_bh);
 		} else {
-			ocfs2_init_dx_root_extent_tree(&dx_et, dir, dx_root_bh);
-			ret = ocfs2_insert_extent(osb, handle, dir, &dx_et, 0,
+			ocfs2_init_dx_root_extent_tree(&dx_et,
+						       INODE_CACHE(dir),
+						       dx_root_bh);
+			ret = ocfs2_insert_extent(handle, &dx_et, 0,
 						  dx_insert_blkno, 1, 0, NULL);
 			if (ret)
 				mlog_errno(ret);
@@ -3138,7 +3148,7 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
 		}
 		blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
 
-		ret = ocfs2_insert_extent(osb, handle, dir, &et, 1,
+		ret = ocfs2_insert_extent(handle, &et, 1,
 					  blkno, len, 0, NULL);
 		if (ret) {
 			mlog_errno(ret);
@@ -3337,8 +3347,9 @@ static int ocfs2_extend_dir(struct ocfs2_super *osb,
 	spin_lock(&OCFS2_I(dir)->ip_lock);
 	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 		spin_unlock(&OCFS2_I(dir)->ip_lock);
-		ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh);
-		num_free_extents = ocfs2_num_free_extents(osb, dir, &et);
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
+					      parent_fe_bh);
+		num_free_extents = ocfs2_num_free_extents(osb, &et);
 		if (num_free_extents < 0) {
 			status = num_free_extents;
 			mlog_errno(status);
@@ -3387,9 +3398,9 @@ do_extend:
 		goto bail;
 	}
 
-	ocfs2_set_new_buffer_uptodate(dir, new_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
 
-	status = ocfs2_journal_access_db(handle, dir, new_bh,
+	status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -3829,7 +3840,7 @@ static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
 	     (unsigned long long)OCFS2_I(dir)->ip_blkno,
 	     (unsigned long long)leaf_blkno, insert_hash);
 
-	ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
 
 	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
 	/*
@@ -3885,7 +3896,7 @@ static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
 	}
 	did_quota = 1;
 
-	ret = ocfs2_journal_access_dl(handle, dir, dx_leaf_bh,
+	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -3949,7 +3960,8 @@ static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
 	}
 
 	for (i = 0; i < num_dx_leaves; i++) {
-		ret = ocfs2_journal_access_dl(handle, dir, orig_dx_leaves[i],
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
+					      orig_dx_leaves[i],
 					      OCFS2_JOURNAL_ACCESS_WRITE);
 		if (ret) {
 			mlog_errno(ret);
@@ -4165,7 +4177,7 @@ static int ocfs2_expand_inline_dx_root(struct inode *dir,
 	 * failure to add the dx_root_bh to the journal won't result
 	 * us losing clusters.
 	 */
-	ret = ocfs2_journal_access_dr(handle, dir, dx_root_bh,
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -4207,9 +4219,8 @@ static int ocfs2_expand_inline_dx_root(struct inode *dir,
 
 	/* This should never fail considering we start with an empty
 	 * dx_root. */
-	ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
-	ret = ocfs2_insert_extent(osb, handle, dir, &et, 0,
-				  insert_blkno, 1, 0, NULL);
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
+	ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
 	if (ret)
 		mlog_errno(ret);
 	did_quota = 0;
@@ -4469,7 +4480,7 @@ static int ocfs2_dx_dir_remove_index(struct inode *dir,
 		goto out_unlock;
 	}
 
-	ret = ocfs2_journal_access_di(handle, dir, di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -4532,7 +4543,7 @@ int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
 	if (ocfs2_dx_root_inline(dx_root))
 		goto remove_index;
 
-	ocfs2_init_dx_root_extent_tree(&et, dir, dx_root_bh);
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
 
 	/* XXX: What if dr_clusters is too large? */
 	while (le32_to_cpu(dx_root->dr_clusters)) {
@@ -4565,7 +4576,7 @@ remove_index:
 		goto out;
 	}
 
-	ocfs2_remove_from_cache(dir, dx_root_bh);
+	ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
 out:
 	ocfs2_schedule_truncate_log_flush(osb, 1);
 	ocfs2_run_deallocs(osb, &dealloc);

fs/ocfs2/dlm/dlmthread.c

@@ -212,14 +212,18 @@ static int dlm_purge_lockres(struct dlm_ctxt *dlm,
 		spin_lock(&dlm->spinlock);
 	}
 
+	spin_lock(&res->spinlock);
 	if (!list_empty(&res->purge)) {
 		mlog(0, "removing lockres %.*s:%p from purgelist, "
 		     "master = %d\n", res->lockname.len, res->lockname.name,
 		     res, master);
 		list_del_init(&res->purge);
+		spin_unlock(&res->spinlock);
 		dlm_lockres_put(res);
 		dlm->purge_count--;
-	}
+	} else
+		spin_unlock(&res->spinlock);
+
 	__dlm_unhash_lockres(res);
 
 	/* lockres is not in the hash now.  drop the flag and wake up

fs/ocfs2/dlmglue.c

@@ -53,6 +53,7 @@
 #include "super.h"
 #include "uptodate.h"
 #include "quota.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -110,6 +111,11 @@ static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
 
 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
 
+static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
+					    int new_level);
+static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
+					 int blocking);
+
 #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
 
 /* This aids in debugging situations where a bad LVB might be involved. */
@@ -278,6 +284,12 @@ static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
 	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
 };
 
+static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
+	.check_downconvert = ocfs2_check_refcount_downconvert,
+	.downconvert_worker = ocfs2_refcount_convert_worker,
+	.flags		= 0,
+};
+
 static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
 {
 	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
@@ -306,6 +318,12 @@ static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_re
 	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
 }
 
+static inline struct ocfs2_refcount_tree *
+ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
+{
+	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
+}
+
 static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
 {
 	if (lockres->l_ops->get_osb)
@@ -693,6 +711,17 @@ void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
 				   info);
 }
 
+void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
+				  struct ocfs2_super *osb, u64 ref_blkno,
+				  unsigned int generation)
+{
+	ocfs2_lock_res_init_once(lockres);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
+			      generation, lockres->l_name);
+	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
+				   &ocfs2_refcount_block_lops, osb);
+}
+
 void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
 {
 	mlog_entry_void();
@@ -1548,8 +1577,10 @@ int ocfs2_rw_lock(struct inode *inode, int write)
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
 	     write ? "EXMODE" : "PRMODE");
 
-	if (ocfs2_mount_local(osb))
+	if (ocfs2_mount_local(osb)) {
+		mlog_exit(0);
 		return 0;
+	}
 
 	lockres = &OCFS2_I(inode)->ip_rw_lockres;
 
@@ -2127,7 +2158,7 @@ static int ocfs2_inode_lock_update(struct inode *inode,
 
 	/* This will discard any caching information we might have had
 	 * for the inode metadata. */
-	ocfs2_metadata_cache_purge(inode);
+	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
 
 	ocfs2_extent_map_trunc(inode, 0);
 
@@ -3009,6 +3040,7 @@ static void ocfs2_unlock_ast(void *opaque, int error)
 		     "unlock_action %d\n", error, lockres->l_name,
 		     lockres->l_unlock_action);
 		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		mlog_exit_void();
 		return;
 	}
 
@@ -3495,11 +3527,11 @@ out:
 	return UNBLOCK_CONTINUE;
 }
 
-static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
-					int new_level)
+static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
+				 struct ocfs2_lock_res *lockres,
+				 int new_level)
 {
-	struct inode *inode = ocfs2_lock_res_inode(lockres);
-	int checkpointed = ocfs2_inode_fully_checkpointed(inode);
+	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
 
 	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
 	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
@@ -3507,10 +3539,18 @@ static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
 	if (checkpointed)
 		return 1;
 
-	ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb));
+	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
 	return 0;
 }
 
+static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
+					int new_level)
+{
+	struct inode *inode = ocfs2_lock_res_inode(lockres);
+
+	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
+}
+
 static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
 {
 	struct inode *inode = ocfs2_lock_res_inode(lockres);
@@ -3640,6 +3680,26 @@ static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
 	return UNBLOCK_CONTINUE_POST;
 }
 
+static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
+					    int new_level)
+{
+	struct ocfs2_refcount_tree *tree =
+				ocfs2_lock_res_refcount_tree(lockres);
+
+	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
+}
+
+static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
+					 int blocking)
+{
+	struct ocfs2_refcount_tree *tree =
+				ocfs2_lock_res_refcount_tree(lockres);
+
+	ocfs2_metadata_cache_purge(&tree->rf_ci);
+
+	return UNBLOCK_CONTINUE;
+}
+
 static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
 {
 	struct ocfs2_qinfo_lvb *lvb;
@@ -3752,6 +3812,37 @@ bail:
 	return status;
 }
 
+int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
+{
+	int status;
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
+	struct ocfs2_super *osb = lockres->l_priv;
+
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
+{
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
+	struct ocfs2_super *osb = lockres->l_priv;
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres, level);
+}
+
 /*
  * This is the filesystem locking protocol.  It provides the lock handling
  * hooks for the underlying DLM.  It has a maximum version number.

fs/ocfs2/dlmglue.h

@@ -101,6 +101,9 @@ void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
 struct ocfs2_mem_dqinfo;
 void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
                                struct ocfs2_mem_dqinfo *info);
+void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
+				  struct ocfs2_super *osb, u64 ref_blkno,
+				  unsigned int generation);
 void ocfs2_lock_res_free(struct ocfs2_lock_res *res);
 int ocfs2_create_new_inode_locks(struct inode *inode);
 int ocfs2_drop_inode_locks(struct inode *inode);
@@ -148,6 +151,9 @@ int ocfs2_file_lock(struct file *file, int ex, int trylock);
 void ocfs2_file_unlock(struct file *file);
 int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex);
 void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex);
+struct ocfs2_refcount_tree;
+int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex);
+void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex);
 
 
 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres);

fs/ocfs2/extent_map.c

@@ -293,7 +293,7 @@ static int ocfs2_last_eb_is_empty(struct inode *inode,
 	struct ocfs2_extent_block *eb;
 	struct ocfs2_extent_list *el;
 
-	ret = ocfs2_read_extent_block(inode, last_eb_blk, &eb_bh);
+	ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
 	if (ret) {
 		mlog_errno(ret);
 		goto out;
@@ -353,11 +353,11 @@ static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
  * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
  * containing el.
  */
-static int ocfs2_figure_hole_clusters(struct inode *inode,
-				      struct ocfs2_extent_list *el,
-				      struct buffer_head *eb_bh,
-				      u32 v_cluster,
-				      u32 *num_clusters)
+int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
+			       struct ocfs2_extent_list *el,
+			       struct buffer_head *eb_bh,
+			       u32 v_cluster,
+			       u32 *num_clusters)
 {
 	int ret, i;
 	struct buffer_head *next_eb_bh = NULL;
@@ -375,7 +375,7 @@ static int ocfs2_figure_hole_clusters(struct inode *inode,
 		if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
 			goto no_more_extents;
 
-		ret = ocfs2_read_extent_block(inode,
+		ret = ocfs2_read_extent_block(ci,
 					      le64_to_cpu(eb->h_next_leaf_blk),
 					      &next_eb_bh);
 		if (ret) {
@@ -428,7 +428,8 @@ static int ocfs2_get_clusters_nocache(struct inode *inode,
 	tree_height = le16_to_cpu(el->l_tree_depth);
 
 	if (tree_height > 0) {
-		ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
+				      &eb_bh);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -455,7 +456,8 @@ static int ocfs2_get_clusters_nocache(struct inode *inode,
 		 * field.
 		 */
 		if (hole_len) {
-			ret = ocfs2_figure_hole_clusters(inode, el, eb_bh,
+			ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
+							 el, eb_bh,
 							 v_cluster, &len);
 			if (ret) {
 				mlog_errno(ret);
@@ -539,7 +541,8 @@ static void ocfs2_relative_extent_offsets(struct super_block *sb,
 
 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 			     u32 *p_cluster, u32 *num_clusters,
-			     struct ocfs2_extent_list *el)
+			     struct ocfs2_extent_list *el,
+			     unsigned int *extent_flags)
 {
 	int ret = 0, i;
 	struct buffer_head *eb_bh = NULL;
@@ -548,7 +551,8 @@ int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 	u32 coff;
 
 	if (el->l_tree_depth) {
-		ret = ocfs2_find_leaf(inode, el, v_cluster, &eb_bh);
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
+				      &eb_bh);
 		if (ret) {
 			mlog_errno(ret);
 			goto out;
@@ -590,6 +594,9 @@ int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 		*p_cluster = *p_cluster + coff;
 		if (num_clusters)
 			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
+
+		if (extent_flags)
+			*extent_flags = rec->e_flags;
 	}
 out:
 	if (eb_bh)
@@ -862,8 +869,8 @@ int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 			BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
 		}
 
-		rc = ocfs2_read_blocks(inode, p_block, count, bhs + done,
-				       flags, validate);
+		rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
+				       bhs + done, flags, validate);
 		if (rc) {
 			mlog_errno(rc);
 			break;

fs/ocfs2/extent_map.h

@@ -55,12 +55,18 @@ int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 
 int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
 			     u32 *p_cluster, u32 *num_clusters,
-			     struct ocfs2_extent_list *el);
+			     struct ocfs2_extent_list *el,
+			     unsigned int *extent_flags);
 
 int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
 			   struct buffer_head *bhs[], int flags,
 			   int (*validate)(struct super_block *sb,
 					   struct buffer_head *bh));
+int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
+			       struct ocfs2_extent_list *el,
+			       struct buffer_head *eb_bh,
+			       u32 v_cluster,
+			       u32 *num_clusters);
 static inline int ocfs2_read_virt_block(struct inode *inode, u64 v_block,
 					struct buffer_head **bh,
 					int (*validate)(struct super_block *sb,

fs/ocfs2/file.c

@@ -59,6 +59,7 @@
 #include "xattr.h"
 #include "acl.h"
 #include "quota.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -259,7 +260,7 @@ int ocfs2_update_inode_atime(struct inode *inode,
 		goto out;
 	}
 
-	ret = ocfs2_journal_access_di(handle, inode, bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret) {
 		mlog_errno(ret);
@@ -334,6 +335,39 @@ out:
 	return ret;
 }
 
+static int ocfs2_cow_file_pos(struct inode *inode,
+			      struct buffer_head *fe_bh,
+			      u64 offset)
+{
+	int status;
+	u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+
+	/*
+	 * If the new offset is aligned to the range of the cluster, there is
+	 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
+	 * CoW either.
+	 */
+	if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
+		return 0;
+
+	status = ocfs2_get_clusters(inode, cpos, &phys,
+				    &num_clusters, &ext_flags);
+	if (status) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
+		goto out;
+
+	return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
+
+out:
+	return status;
+}
+
 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 				     struct inode *inode,
 				     struct buffer_head *fe_bh,
@@ -346,6 +380,17 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 
 	mlog_entry_void();
 
+	/*
+	 * We need to CoW the cluster contains the offset if it is reflinked
+	 * since we will call ocfs2_zero_range_for_truncate later which will
+	 * write "0" from offset to the end of the cluster.
+	 */
+	status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
+	if (status) {
+		mlog_errno(status);
+		return status;
+	}
+
 	/* TODO: This needs to actually orphan the inode in this
 	 * transaction. */
 
@@ -356,7 +401,7 @@ static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
 		goto out;
 	}
 
-	status = ocfs2_journal_access_di(handle, inode, fe_bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -486,6 +531,8 @@ bail_unlock_sem:
 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
 
 bail:
+	if (!status && OCFS2_I(inode)->ip_clusters == 0)
+		status = ocfs2_try_remove_refcount_tree(inode, di_bh);
 
 	mlog_exit(status);
 	return status;
@@ -515,11 +562,10 @@ int ocfs2_add_inode_data(struct ocfs2_super *osb,
 	int ret;
 	struct ocfs2_extent_tree et;
 
-	ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
-	ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
-					   clusters_to_add, mark_unwritten,
-					   &et, handle,
-					   data_ac, meta_ac, reason_ret);
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
+	ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
+					  clusters_to_add, mark_unwritten,
+					  data_ac, meta_ac, reason_ret);
 
 	return ret;
 }
@@ -564,7 +610,7 @@ restart_all:
 	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
 	     (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
 	     clusters_to_add);
-	ocfs2_init_dinode_extent_tree(&et, inode, bh);
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
 	status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 				       &data_ac, &meta_ac);
 	if (status) {
@@ -593,7 +639,7 @@ restarted_transaction:
 	/* reserve a write to the file entry early on - that we if we
 	 * run out of credits in the allocation path, we can still
 	 * update i_size. */
-	status = ocfs2_journal_access_di(handle, inode, bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1131,7 +1177,7 @@ static int __ocfs2_write_remove_suid(struct inode *inode,
 		goto out;
 	}
 
-	ret = ocfs2_journal_access_di(handle, inode, bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
@@ -1395,7 +1441,7 @@ static int ocfs2_remove_inode_range(struct inode *inode,
 	struct address_space *mapping = inode->i_mapping;
 	struct ocfs2_extent_tree et;
 
-	ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
 	ocfs2_init_dealloc_ctxt(&dealloc);
 
 	if (byte_len == 0)
@@ -1657,6 +1703,70 @@ static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
 					 OCFS2_IOC_RESVSP64, &sr, change_size);
 }
 
+int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
+				   size_t count)
+{
+	int ret = 0;
+	unsigned int extent_flags;
+	u32 cpos, clusters, extent_len, phys_cpos;
+	struct super_block *sb = inode->i_sb;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
+	    !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
+		return 0;
+
+	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
+					 &extent_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
+			ret = 1;
+			break;
+		}
+
+		if (extent_len > clusters)
+			extent_len = clusters;
+
+		clusters -= extent_len;
+		cpos += extent_len;
+	}
+out:
+	return ret;
+}
+
+static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
+					    loff_t pos, size_t count,
+					    int *meta_level)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	u32 clusters =
+		ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*meta_level = 1;
+
+	ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(di_bh);
+	return ret;
+}
+
 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 loff_t *ppos,
 					 size_t count,
@@ -1713,6 +1823,22 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 
 		end = saved_pos + count;
 
+		ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
+		if (ret == 1) {
+			ocfs2_inode_unlock(inode, meta_level);
+			meta_level = -1;
+
+			ret = ocfs2_prepare_inode_for_refcount(inode,
+							       saved_pos,
+							       count,
+							       &meta_level);
+		}
+
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+
 		/*
 		 * Skip the O_DIRECT checks if we don't need
 		 * them.
@@ -1759,7 +1885,8 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 		*ppos = saved_pos;
 
 out_unlock:
-	ocfs2_inode_unlock(inode, meta_level);
+	if (meta_level >= 0)
+		ocfs2_inode_unlock(inode, meta_level);
 
 out:
 	return ret;

fs/ocfs2/file.h

@@ -69,4 +69,6 @@ int ocfs2_update_inode_atime(struct inode *inode,
 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
 			    struct ocfs2_space_resv *sr);
 
+int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
+				   size_t count);
 #endif /* OCFS2_FILE_H */

fs/ocfs2/inode.c

@@ -53,6 +53,7 @@
 #include "sysfile.h"
 #include "uptodate.h"
 #include "xattr.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -562,7 +563,8 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
 			goto out;
 		}
 
-		status = ocfs2_journal_access_di(handle, inode, fe_bh,
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+						 fe_bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		if (status < 0) {
 			mlog_errno(status);
@@ -646,7 +648,7 @@ static int ocfs2_remove_inode(struct inode *inode,
 	}
 
 	/* set the inodes dtime */
-	status = ocfs2_journal_access_di(handle, inode, di_bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -662,7 +664,7 @@ static int ocfs2_remove_inode(struct inode *inode,
 		goto bail_commit;
 	}
 
-	ocfs2_remove_from_cache(inode, di_bh);
+	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
 	vfs_dq_free_inode(inode);
 
 	status = ocfs2_free_dinode(handle, inode_alloc_inode,
@@ -781,6 +783,12 @@ static int ocfs2_wipe_inode(struct inode *inode,
 		goto bail_unlock_dir;
 	}
 
+	status = ocfs2_remove_refcount_tree(inode, di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_dir;
+	}
+
 	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
 				    orphan_dir_bh);
 	if (status < 0)
@@ -1112,13 +1120,14 @@ void ocfs2_clear_inode(struct inode *inode)
 	ocfs2_lock_res_free(&oi->ip_inode_lockres);
 	ocfs2_lock_res_free(&oi->ip_open_lockres);
 
-	ocfs2_metadata_cache_purge(inode);
+	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
 
-	mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
+	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
 			"Clear inode of %llu, inode has %u cache items\n",
-			(unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
+			(unsigned long long)oi->ip_blkno,
+			INODE_CACHE(inode)->ci_num_cached);
 
-	mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
+	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
 			"Clear inode of %llu, inode has a bad flag\n",
 			(unsigned long long)oi->ip_blkno);
 
@@ -1145,9 +1154,7 @@ void ocfs2_clear_inode(struct inode *inode)
 			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
 
 	/* Clear all other flags. */
-	oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
-	oi->ip_created_trans = 0;
-	oi->ip_last_trans = 0;
+	oi->ip_flags = 0;
 	oi->ip_dir_start_lookup = 0;
 	oi->ip_blkno = 0ULL;
 
@@ -1239,7 +1246,7 @@ int ocfs2_mark_inode_dirty(handle_t *handle,
 	mlog_entry("(inode %llu)\n",
 		   (unsigned long long)OCFS2_I(inode)->ip_blkno);
 
-	status = ocfs2_journal_access_di(handle, inode, bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1380,8 +1387,8 @@ int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
 	int rc;
 	struct buffer_head *tmp = *bh;
 
-	rc = ocfs2_read_blocks(inode, OCFS2_I(inode)->ip_blkno, 1, &tmp,
-			       flags, ocfs2_validate_inode_block);
+	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
+			       1, &tmp, flags, ocfs2_validate_inode_block);
 
 	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
 	if (!rc && !*bh)
@@ -1394,3 +1401,56 @@ int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
 {
 	return ocfs2_read_inode_block_full(inode, bh, 0);
 }
+
+
+static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	return oi->ip_blkno;
+}
+
+static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	return oi->vfs_inode.i_sb;
+}
+
+static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	spin_lock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	spin_unlock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	mutex_lock(&oi->ip_io_mutex);
+}
+
+static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	mutex_unlock(&oi->ip_io_mutex);
+}
+
+const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
+	.co_owner		= ocfs2_inode_cache_owner,
+	.co_get_super		= ocfs2_inode_cache_get_super,
+	.co_cache_lock		= ocfs2_inode_cache_lock,
+	.co_cache_unlock	= ocfs2_inode_cache_unlock,
+	.co_io_lock		= ocfs2_inode_cache_io_lock,
+	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
+};
+

fs/ocfs2/inode.h

@@ -60,12 +60,6 @@ struct ocfs2_inode_info
 
 	u32				ip_dir_start_lookup;
 
-	/* next two are protected by trans_inc_lock */
-	/* which transaction were we created on? Zero if none. */
-	unsigned long			ip_created_trans;
-	/* last transaction we were a part of. */
-	unsigned long			ip_last_trans;
-
 	struct ocfs2_caching_info	ip_metadata_cache;
 
 	struct ocfs2_extent_map		ip_extent_map;
@@ -106,8 +100,6 @@ struct ocfs2_inode_info
 #define OCFS2_INODE_MAYBE_ORPHANED	0x00000020
 /* Does someone have the file open O_DIRECT */
 #define OCFS2_INODE_OPEN_DIRECT		0x00000040
-/* Indicates that the metadata cache should be used as an array. */
-#define OCFS2_INODE_CACHE_INLINE	0x00000080
 
 static inline struct ocfs2_inode_info *OCFS2_I(struct inode *inode)
 {
@@ -120,6 +112,12 @@ static inline struct ocfs2_inode_info *OCFS2_I(struct inode *inode)
 extern struct kmem_cache *ocfs2_inode_cache;
 
 extern const struct address_space_operations ocfs2_aops;
+extern const struct ocfs2_caching_operations ocfs2_inode_caching_ops;
+
+static inline struct ocfs2_caching_info *INODE_CACHE(struct inode *inode)
+{
+	return &OCFS2_I(inode)->ip_metadata_cache;
+}
 
 void ocfs2_clear_inode(struct inode *inode);
 void ocfs2_delete_inode(struct inode *inode);
@@ -172,4 +170,10 @@ int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh);
 /* The same, but can be passed OCFS2_BH_* flags */
 int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
 				int flags);
+
+static inline struct ocfs2_inode_info *cache_info_to_inode(struct ocfs2_caching_info *ci)
+{
+	return container_of(ci, struct ocfs2_inode_info, ip_metadata_cache);
+}
+
 #endif /* OCFS2_INODE_H */

fs/ocfs2/ioctl.c

@@ -21,6 +21,7 @@
 #include "ocfs2_fs.h"
 #include "ioctl.h"
 #include "resize.h"
+#include "refcounttree.h"
 
 #include <linux/ext2_fs.h>
 
@@ -115,6 +116,9 @@ long ocfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 	int status;
 	struct ocfs2_space_resv sr;
 	struct ocfs2_new_group_input input;
+	struct reflink_arguments args;
+	const char *old_path, *new_path;
+	bool preserve;
 
 	switch (cmd) {
 	case OCFS2_IOC_GETFLAGS:
@@ -160,6 +164,15 @@ long ocfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 			return -EFAULT;
 
 		return ocfs2_group_add(inode, &input);
+	case OCFS2_IOC_REFLINK:
+		if (copy_from_user(&args, (struct reflink_arguments *)arg,
+				   sizeof(args)))
+			return -EFAULT;
+		old_path = (const char *)(unsigned long)args.old_path;
+		new_path = (const char *)(unsigned long)args.new_path;
+		preserve = (args.preserve != 0);
+
+		return ocfs2_reflink_ioctl(inode, old_path, new_path, preserve);
 	default:
 		return -ENOTTY;
 	}
@@ -182,6 +195,7 @@ long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 	case OCFS2_IOC_GROUP_EXTEND:
 	case OCFS2_IOC_GROUP_ADD:
 	case OCFS2_IOC_GROUP_ADD64:
+	case OCFS2_IOC_REFLINK:
 		break;
 	default:
 		return -ENOIOCTLCMD;

fs/ocfs2/journal.c

@@ -48,6 +48,7 @@
 #include "slot_map.h"
 #include "super.h"
 #include "sysfile.h"
+#include "uptodate.h"
 #include "quota.h"
 
 #include "buffer_head_io.h"
@@ -554,6 +555,14 @@ static struct ocfs2_triggers eb_triggers = {
 	.ot_offset	= offsetof(struct ocfs2_extent_block, h_check),
 };
 
+static struct ocfs2_triggers rb_triggers = {
+	.ot_triggers = {
+		.t_commit = ocfs2_commit_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_refcount_block, rf_check),
+};
+
 static struct ocfs2_triggers gd_triggers = {
 	.ot_triggers = {
 		.t_commit = ocfs2_commit_trigger,
@@ -601,14 +610,16 @@ static struct ocfs2_triggers dl_triggers = {
 };
 
 static int __ocfs2_journal_access(handle_t *handle,
-				  struct inode *inode,
+				  struct ocfs2_caching_info *ci,
 				  struct buffer_head *bh,
 				  struct ocfs2_triggers *triggers,
 				  int type)
 {
 	int status;
+	struct ocfs2_super *osb =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
 
-	BUG_ON(!inode);
+	BUG_ON(!ci || !ci->ci_ops);
 	BUG_ON(!handle);
 	BUG_ON(!bh);
 
@@ -627,15 +638,15 @@ static int __ocfs2_journal_access(handle_t *handle,
 		BUG();
 	}
 
-	/* Set the current transaction information on the inode so
+	/* Set the current transaction information on the ci so
 	 * that the locking code knows whether it can drop it's locks
-	 * on this inode or not. We're protected from the commit
+	 * on this ci or not. We're protected from the commit
 	 * thread updating the current transaction id until
 	 * ocfs2_commit_trans() because ocfs2_start_trans() took
 	 * j_trans_barrier for us. */
-	ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode);
+	ocfs2_set_ci_lock_trans(osb->journal, ci);
 
-	mutex_lock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_lock(ci);
 	switch (type) {
 	case OCFS2_JOURNAL_ACCESS_CREATE:
 	case OCFS2_JOURNAL_ACCESS_WRITE:
@@ -650,9 +661,9 @@ static int __ocfs2_journal_access(handle_t *handle,
 		status = -EINVAL;
 		mlog(ML_ERROR, "Uknown access type!\n");
 	}
-	if (!status && ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)) && triggers)
+	if (!status && ocfs2_meta_ecc(osb) && triggers)
 		jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
-	mutex_unlock(&OCFS2_I(inode)->ip_io_mutex);
+	ocfs2_metadata_cache_io_unlock(ci);
 
 	if (status < 0)
 		mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
@@ -662,66 +673,65 @@ static int __ocfs2_journal_access(handle_t *handle,
 	return status;
 }
 
-int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
-			       struct buffer_head *bh, int type)
+int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &di_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
 }
 
-int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &eb_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
 }
 
-int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &gd_triggers,
+	return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
 				      type);
 }
 
-int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &db_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
 }
 
-int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &xb_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
 }
 
-int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &dq_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
 }
 
-int ocfs2_journal_access_dr(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &dr_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
 }
 
-int ocfs2_journal_access_dl(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, &dl_triggers,
-				      type);
+	return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
+}
+
+int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
 }
 
-int ocfs2_journal_access(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
 			 struct buffer_head *bh, int type)
 {
-	return __ocfs2_journal_access(handle, inode, bh, NULL, type);
+	return __ocfs2_journal_access(handle, ci, bh, NULL, type);
 }
 
 int ocfs2_journal_dirty(handle_t *handle,
@@ -898,7 +908,7 @@ static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
 		ocfs2_bump_recovery_generation(fe);
 
 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
-	status = ocfs2_write_block(osb, bh, journal->j_inode);
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
 	if (status < 0)
 		mlog_errno(status);
 
@@ -1642,7 +1652,7 @@ static int ocfs2_replay_journal(struct ocfs2_super *osb,
 					ocfs2_get_recovery_generation(fe);
 
 	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
-	status = ocfs2_write_block(osb, bh, inode);
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
 	if (status < 0)
 		mlog_errno(status);
 

fs/ocfs2/journal.h

@@ -90,56 +90,66 @@ static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
 	return old_id;
 }
 
-static inline void ocfs2_set_inode_lock_trans(struct ocfs2_journal *journal,
-					      struct inode *inode)
+static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
+					   struct ocfs2_caching_info *ci)
 {
 	spin_lock(&trans_inc_lock);
-	OCFS2_I(inode)->ip_last_trans = journal->j_trans_id;
+	ci->ci_last_trans = journal->j_trans_id;
 	spin_unlock(&trans_inc_lock);
 }
 
 /* Used to figure out whether it's safe to drop a metadata lock on an
- * inode. Returns true if all the inodes changes have been
+ * cached object. Returns true if all the object's changes have been
  * checkpointed to disk. You should be holding the spinlock on the
  * metadata lock while calling this to be sure that nobody can take
  * the lock and put it on another transaction. */
-static inline int ocfs2_inode_fully_checkpointed(struct inode *inode)
+static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
 {
 	int ret;
-	struct ocfs2_journal *journal = OCFS2_SB(inode->i_sb)->journal;
+	struct ocfs2_journal *journal =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
 
 	spin_lock(&trans_inc_lock);
-	ret = time_after(journal->j_trans_id, OCFS2_I(inode)->ip_last_trans);
+	ret = time_after(journal->j_trans_id, ci->ci_last_trans);
 	spin_unlock(&trans_inc_lock);
 	return ret;
 }
 
-/* convenience function to check if an inode is still new (has never
- * hit disk) Will do you a favor and set created_trans = 0 when you've
- * been checkpointed.  returns '1' if the inode is still new. */
-static inline int ocfs2_inode_is_new(struct inode *inode)
+/* convenience function to check if an object backed by struct
+ * ocfs2_caching_info  is still new (has never hit disk) Will do you a
+ * favor and set created_trans = 0 when you've
+ * been checkpointed.  returns '1' if the ci is still new. */
+static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
 {
 	int ret;
+	struct ocfs2_journal *journal =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
 
+	spin_lock(&trans_inc_lock);
+	ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
+	if (!ret)
+		ci->ci_created_trans = 0;
+	spin_unlock(&trans_inc_lock);
+	return ret;
+}
+
+/* Wrapper for inodes so we can check system files */
+static inline int ocfs2_inode_is_new(struct inode *inode)
+{
 	/* System files are never "new" as they're written out by
 	 * mkfs. This helps us early during mount, before we have the
 	 * journal open and j_trans_id could be junk. */
 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
 		return 0;
-	spin_lock(&trans_inc_lock);
-	ret = !(time_after(OCFS2_SB(inode->i_sb)->journal->j_trans_id,
-			   OCFS2_I(inode)->ip_created_trans));
-	if (!ret)
-		OCFS2_I(inode)->ip_created_trans = 0;
-	spin_unlock(&trans_inc_lock);
-	return ret;
+
+	return ocfs2_ci_is_new(INODE_CACHE(inode));
 }
 
-static inline void ocfs2_inode_set_new(struct ocfs2_super *osb,
-				       struct inode *inode)
+static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
+				    struct ocfs2_caching_info *ci)
 {
 	spin_lock(&trans_inc_lock);
-	OCFS2_I(inode)->ip_created_trans = osb->journal->j_trans_id;
+	ci->ci_created_trans = osb->journal->j_trans_id;
 	spin_unlock(&trans_inc_lock);
 }
 
@@ -200,7 +210,7 @@ static inline void ocfs2_checkpoint_inode(struct inode *inode)
 	if (ocfs2_mount_local(osb))
 		return;
 
-	if (!ocfs2_inode_fully_checkpointed(inode)) {
+	if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
 		/* WARNING: This only kicks off a single
 		 * checkpoint. If someone races you and adds more
 		 * metadata to the journal, you won't know, and will
@@ -210,7 +220,7 @@ static inline void ocfs2_checkpoint_inode(struct inode *inode)
 		ocfs2_start_checkpoint(osb);
 
 		wait_event(osb->journal->j_checkpointed,
-			   ocfs2_inode_fully_checkpointed(inode));
+			   ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
 	}
 }
 
@@ -266,31 +276,34 @@ int			     ocfs2_extend_trans(handle_t *handle, int nblocks);
 
 
 /* ocfs2_inode */
-int ocfs2_journal_access_di(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* ocfs2_extent_block */
-int ocfs2_journal_access_eb(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_refcount_block */
+int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* ocfs2_group_desc */
-int ocfs2_journal_access_gd(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* ocfs2_xattr_block */
-int ocfs2_journal_access_xb(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* quota blocks */
-int ocfs2_journal_access_dq(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* dirblock */
-int ocfs2_journal_access_db(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* ocfs2_dx_root_block */
-int ocfs2_journal_access_dr(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* ocfs2_dx_leaf */
-int ocfs2_journal_access_dl(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh, int type);
 /* Anything that has no ecc */
-int ocfs2_journal_access(handle_t *handle, struct inode *inode,
+int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
 			 struct buffer_head *bh, int type);
 
 /*
@@ -477,6 +490,23 @@ static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
 	return credits;
 }
 
+/* inode update, new refcount block and its allocation credits. */
+#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
+					    + OCFS2_SUBALLOC_ALLOC)
+
+/* inode and the refcount block update. */
+#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/*
+ * inode and the refcount block update.
+ * It doesn't include the credits for sub alloc change.
+ * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
+ */
+#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/* 2 metadata alloc, 2 new blocks and root refcount block */
+#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
+
 /*
  * Please note that the caller must make sure that root_el is the root
  * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise

fs/ocfs2/localalloc.c

@@ -297,8 +297,8 @@ void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
 	}
 	memcpy(alloc_copy, alloc, bh->b_size);
 
-	status = ocfs2_journal_access_di(handle, local_alloc_inode, bh,
-					 OCFS2_JOURNAL_ACCESS_WRITE);
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode),
+					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
 		goto out_commit;
@@ -392,7 +392,7 @@ int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
 	ocfs2_clear_local_alloc(alloc);
 
 	ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
-	status = ocfs2_write_block(osb, alloc_bh, inode);
+	status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
 	if (status < 0)
 		mlog_errno(status);
 
@@ -678,7 +678,8 @@ int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
 	 * delete bits from it! */
 	*num_bits = bits_wanted;
 
-	status = ocfs2_journal_access_di(handle, local_alloc_inode,
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(local_alloc_inode),
 					 osb->local_alloc_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
@@ -1156,7 +1157,8 @@ static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 	}
 	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
 
-	status = ocfs2_journal_access_di(handle, local_alloc_inode,
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(local_alloc_inode),
 					 osb->local_alloc_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {

fs/ocfs2/namei.c

@@ -69,7 +69,6 @@
 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 			      struct inode *dir,
 			      struct inode *inode,
-			      struct dentry *dentry,
 			      dev_t dev,
 			      struct buffer_head **new_fe_bh,
 			      struct buffer_head *parent_fe_bh,
@@ -78,7 +77,7 @@ static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 
 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 				    struct inode **ret_orphan_dir,
-				    struct inode *inode,
+				    u64 blkno,
 				    char *name,
 				    struct ocfs2_dir_lookup_result *lookup);
 
@@ -358,8 +357,12 @@ static int ocfs2_mknod(struct inode *dir,
 	}
 	did_quota_inode = 1;
 
+	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
+		   inode->i_mode, (unsigned long)dev, dentry->d_name.len,
+		   dentry->d_name.name);
+
 	/* do the real work now. */
-	status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
+	status = ocfs2_mknod_locked(osb, dir, inode, dev,
 				    &new_fe_bh, parent_fe_bh, handle,
 				    inode_ac);
 	if (status < 0) {
@@ -375,7 +378,8 @@ static int ocfs2_mknod(struct inode *dir,
 			goto leave;
 		}
 
-		status = ocfs2_journal_access_di(handle, dir, parent_fe_bh,
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
+						 parent_fe_bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		if (status < 0) {
 			mlog_errno(status);
@@ -465,7 +469,6 @@ leave:
 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 			      struct inode *dir,
 			      struct inode *inode,
-			      struct dentry *dentry,
 			      dev_t dev,
 			      struct buffer_head **new_fe_bh,
 			      struct buffer_head *parent_fe_bh,
@@ -479,10 +482,6 @@ static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 	u16 suballoc_bit;
 	u16 feat;
 
-	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
-		   inode->i_mode, (unsigned long)dev, dentry->d_name.len,
-		   dentry->d_name.name);
-
 	*new_fe_bh = NULL;
 
 	status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
@@ -507,9 +506,10 @@ static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 		mlog_errno(status);
 		goto leave;
 	}
-	ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 
-	status = ocfs2_journal_access_di(handle, inode, *new_fe_bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+					 *new_fe_bh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -565,7 +565,7 @@ static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 	}
 
 	ocfs2_populate_inode(inode, fe, 1);
-	ocfs2_inode_set_new(osb, inode);
+	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 	if (!ocfs2_mount_local(osb)) {
 		status = ocfs2_create_new_inode_locks(inode);
 		if (status < 0)
@@ -682,7 +682,7 @@ static int ocfs2_link(struct dentry *old_dentry,
 		goto out_unlock_inode;
 	}
 
-	err = ocfs2_journal_access_di(handle, inode, fe_bh,
+	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (err < 0) {
 		mlog_errno(err);
@@ -850,7 +850,8 @@ static int ocfs2_unlink(struct inode *dir,
 	}
 
 	if (inode_is_unlinkable(inode)) {
-		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
+		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
+						  OCFS2_I(inode)->ip_blkno,
 						  orphan_name, &orphan_insert);
 		if (status < 0) {
 			mlog_errno(status);
@@ -866,7 +867,7 @@ static int ocfs2_unlink(struct inode *dir,
 		goto leave;
 	}
 
-	status = ocfs2_journal_access_di(handle, inode, fe_bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1241,9 +1242,8 @@ static int ocfs2_rename(struct inode *old_dir,
 
 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
-							  new_inode,
-							  orphan_name,
-							  &orphan_insert);
+						OCFS2_I(new_inode)->ip_blkno,
+						orphan_name, &orphan_insert);
 			if (status < 0) {
 				mlog_errno(status);
 				goto bail;
@@ -1284,7 +1284,8 @@ static int ocfs2_rename(struct inode *old_dir,
 				goto bail;
 			}
 		}
-		status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
+						 newfe_bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		if (status < 0) {
 			mlog_errno(status);
@@ -1331,7 +1332,8 @@ static int ocfs2_rename(struct inode *old_dir,
 	old_inode->i_ctime = CURRENT_TIME;
 	mark_inode_dirty(old_inode);
 
-	status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
+					 old_inode_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status >= 0) {
 		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
@@ -1407,9 +1409,10 @@ static int ocfs2_rename(struct inode *old_dir,
 			     (int)old_dir_nlink, old_dir->i_nlink);
 		} else {
 			struct ocfs2_dinode *fe;
-			status = ocfs2_journal_access_di(handle, old_dir,
-						      old_dir_bh,
-						      OCFS2_JOURNAL_ACCESS_WRITE);
+			status = ocfs2_journal_access_di(handle,
+							 INODE_CACHE(old_dir),
+							 old_dir_bh,
+							 OCFS2_JOURNAL_ACCESS_WRITE);
 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
 			ocfs2_set_links_count(fe, old_dir->i_nlink);
 			status = ocfs2_journal_dirty(handle, old_dir_bh);
@@ -1527,9 +1530,11 @@ static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 			mlog_errno(status);
 			goto bail;
 		}
-		ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
+		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
+					      bhs[virtual]);
 
-		status = ocfs2_journal_access(handle, inode, bhs[virtual],
+		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
+					      bhs[virtual],
 					      OCFS2_JOURNAL_ACCESS_CREATE);
 		if (status < 0) {
 			mlog_errno(status);
@@ -1692,7 +1697,11 @@ static int ocfs2_symlink(struct inode *dir,
 	}
 	did_quota_inode = 1;
 
-	status = ocfs2_mknod_locked(osb, dir, inode, dentry,
+	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
+		   inode->i_mode, dentry->d_name.len,
+		   dentry->d_name.name);
+
+	status = ocfs2_mknod_locked(osb, dir, inode,
 				    0, &new_fe_bh, parent_fe_bh, handle,
 				    inode_ac);
 	if (status < 0) {
@@ -1842,7 +1851,7 @@ bail:
 
 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 				    struct inode **ret_orphan_dir,
-				    struct inode *inode,
+				    u64 blkno,
 				    char *name,
 				    struct ocfs2_dir_lookup_result *lookup)
 {
@@ -1850,7 +1859,7 @@ static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 	struct buffer_head *orphan_dir_bh = NULL;
 	int status = 0;
 
-	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
+	status = ocfs2_blkno_stringify(blkno, name);
 	if (status < 0) {
 		mlog_errno(status);
 		return status;
@@ -1917,7 +1926,9 @@ static int ocfs2_orphan_add(struct ocfs2_super *osb,
 		goto leave;
 	}
 
-	status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(orphan_dir_inode),
+					 orphan_dir_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -2002,7 +2013,9 @@ int ocfs2_orphan_del(struct ocfs2_super *osb,
 		goto leave;
 	}
 
-	status = ocfs2_journal_access_di(handle,orphan_dir_inode,  orphan_dir_bh,
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(orphan_dir_inode),
+					 orphan_dir_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -2028,6 +2041,274 @@ leave:
 	return status;
 }
 
+int ocfs2_create_inode_in_orphan(struct inode *dir,
+				 int mode,
+				 struct inode **new_inode)
+{
+	int status, did_quota_inode = 0;
+	struct inode *inode = NULL;
+	struct inode *orphan_dir = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *di = NULL;
+	handle_t *handle = NULL;
+	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
+	struct buffer_head *parent_di_bh = NULL;
+	struct buffer_head *new_di_bh = NULL;
+	struct ocfs2_alloc_context *inode_ac = NULL;
+	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
+
+	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	/*
+	 * We give the orphan dir the root blkno to fake an orphan name,
+	 * and allocate enough space for our insertion.
+	 */
+	status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
+					  osb->root_blkno,
+					  orphan_name, &orphan_insert);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* reserve an inode spot */
+	status = ocfs2_reserve_new_inode(osb, &inode_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	inode = ocfs2_get_init_inode(dir, mode);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* We don't use standard VFS wrapper because we don't want vfs_dq_init
+	 * to be called. */
+	if (sb_any_quota_active(osb->sb) &&
+	    osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
+		status = -EDQUOT;
+		goto leave;
+	}
+	did_quota_inode = 1;
+
+	/* do the real work now. */
+	status = ocfs2_mknod_locked(osb, dir, inode,
+				    0, &new_di_bh, parent_di_bh, handle,
+				    inode_ac);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	di = (struct ocfs2_dinode *)new_di_bh->b_data;
+	status = ocfs2_orphan_add(osb, handle, inode, di, orphan_name,
+				  &orphan_insert, orphan_dir);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* get open lock so that only nodes can't remove it from orphan dir. */
+	status = ocfs2_open_lock(inode);
+	if (status < 0)
+		mlog_errno(status);
+
+leave:
+	if (status < 0 && did_quota_inode)
+		vfs_dq_free_inode(inode);
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	if (orphan_dir) {
+		/* This was locked for us in ocfs2_prepare_orphan_dir() */
+		ocfs2_inode_unlock(orphan_dir, 1);
+		mutex_unlock(&orphan_dir->i_mutex);
+		iput(orphan_dir);
+	}
+
+	if (status == -ENOSPC)
+		mlog(0, "Disk is full\n");
+
+	if ((status < 0) && inode) {
+		clear_nlink(inode);
+		iput(inode);
+	}
+
+	if (inode_ac)
+		ocfs2_free_alloc_context(inode_ac);
+
+	brelse(new_di_bh);
+
+	if (!status)
+		*new_inode = inode;
+
+	ocfs2_free_dir_lookup_result(&orphan_insert);
+
+	ocfs2_inode_unlock(dir, 1);
+	brelse(parent_di_bh);
+	return status;
+}
+
+int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
+				   struct inode *inode,
+				   struct dentry *dentry)
+{
+	int status = 0;
+	struct buffer_head *parent_di_bh = NULL;
+	handle_t *handle = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *dir_di, *di;
+	struct inode *orphan_dir_inode = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+
+	mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
+		   dentry->d_name.len, dentry->d_name.name);
+
+	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
+	if (!dir_di->i_links_count) {
+		/* can't make a file in a deleted directory. */
+		status = -ENOENT;
+		goto leave;
+	}
+
+	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
+					   dentry->d_name.len);
+	if (status)
+		goto leave;
+
+	/* get a spot inside the dir. */
+	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
+					      dentry->d_name.name,
+					      dentry->d_name.len, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+						       ORPHAN_DIR_SYSTEM_INODE,
+						       osb->slot_num);
+	if (!orphan_dir_inode) {
+		status = -EEXIST;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	mutex_lock(&orphan_dir_inode->i_mutex);
+
+	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		mutex_unlock(&orphan_dir_inode->i_mutex);
+		iput(orphan_dir_inode);
+		goto leave;
+	}
+
+	status = ocfs2_read_inode_block(inode, &di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto orphan_unlock;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto orphan_unlock;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
+				  orphan_dir_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
+	di->i_orphaned_slot = 0;
+	ocfs2_journal_dirty(handle, di_bh);
+
+	status = ocfs2_add_entry(handle, dentry, inode,
+				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
+				 &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	status = ocfs2_dentry_attach_lock(dentry, inode,
+					  OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	insert_inode_hash(inode);
+	dentry->d_op = &ocfs2_dentry_ops;
+	d_instantiate(dentry, inode);
+	status = 0;
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+orphan_unlock:
+	ocfs2_inode_unlock(orphan_dir_inode, 1);
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	iput(orphan_dir_inode);
+leave:
+
+	ocfs2_inode_unlock(dir, 1);
+
+	brelse(di_bh);
+	brelse(parent_di_bh);
+	brelse(orphan_dir_bh);
+
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	mlog_exit(status);
+
+	return status;
+}
+
 const struct inode_operations ocfs2_dir_iops = {
 	.create		= ocfs2_create,
 	.lookup		= ocfs2_lookup,

fs/ocfs2/namei.h

@@ -35,5 +35,11 @@ int ocfs2_orphan_del(struct ocfs2_super *osb,
 		     struct inode *orphan_dir_inode,
 		     struct inode *inode,
 		     struct buffer_head *orphan_dir_bh);
+int ocfs2_create_inode_in_orphan(struct inode *dir,
+				 int mode,
+				 struct inode **new_inode);
+int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
+				   struct inode *new_inode,
+				   struct dentry *new_dentry);
 
 #endif /* OCFS2_NAMEI_H */

fs/ocfs2/ocfs2.h

@@ -51,20 +51,51 @@
 /* For struct ocfs2_blockcheck_stats */
 #include "blockcheck.h"
 
+
+/* Caching of metadata buffers */
+
 /* Most user visible OCFS2 inodes will have very few pieces of
  * metadata, but larger files (including bitmaps, etc) must be taken
  * into account when designing an access scheme. We allow a small
  * amount of inlined blocks to be stored on an array and grow the
  * structure into a rb tree when necessary. */
-#define OCFS2_INODE_MAX_CACHE_ARRAY 2
+#define OCFS2_CACHE_INFO_MAX_ARRAY 2
+
+/* Flags for ocfs2_caching_info */
+
+enum ocfs2_caching_info_flags {
+	/* Indicates that the metadata cache is using the inline array */
+	OCFS2_CACHE_FL_INLINE	= 1<<1,
+};
 
+struct ocfs2_caching_operations;
 struct ocfs2_caching_info {
+	/*
+	 * The parent structure provides the locks, but because the
+	 * parent structure can differ, it provides locking operations
+	 * to struct ocfs2_caching_info.
+	 */
+	const struct ocfs2_caching_operations *ci_ops;
+
+	/* next two are protected by trans_inc_lock */
+	/* which transaction were we created on? Zero if none. */
+	unsigned long		ci_created_trans;
+	/* last transaction we were a part of. */
+	unsigned long		ci_last_trans;
+
+	/* Cache structures */
+	unsigned int		ci_flags;
 	unsigned int		ci_num_cached;
 	union {
-		sector_t	ci_array[OCFS2_INODE_MAX_CACHE_ARRAY];
+	sector_t	ci_array[OCFS2_CACHE_INFO_MAX_ARRAY];
 		struct rb_root	ci_tree;
 	} ci_cache;
 };
+/*
+ * Need this prototype here instead of in uptodate.h because journal.h
+ * uses it.
+ */
+struct super_block *ocfs2_metadata_cache_get_super(struct ocfs2_caching_info *ci);
 
 /* this limits us to 256 nodes
  * if we need more, we can do a kmalloc for the map */
@@ -377,12 +408,17 @@ struct ocfs2_super
 
 	/* the group we used to allocate inodes. */
 	u64				osb_inode_alloc_group;
+
+	/* rb tree root for refcount lock. */
+	struct rb_root	osb_rf_lock_tree;
+	struct ocfs2_refcount_tree *osb_ref_tree_lru;
 };
 
 #define OCFS2_SB(sb)	    ((struct ocfs2_super *)(sb)->s_fs_info)
 
 /* Useful typedef for passing around journal access functions */
-typedef int (*ocfs2_journal_access_func)(handle_t *handle, struct inode *inode,
+typedef int (*ocfs2_journal_access_func)(handle_t *handle,
+					 struct ocfs2_caching_info *ci,
 					 struct buffer_head *bh, int type);
 
 static inline int ocfs2_should_order_data(struct inode *inode)
@@ -480,6 +516,13 @@ static inline void ocfs2_add_links_count(struct ocfs2_dinode *di, int n)
 	ocfs2_set_links_count(di, links);
 }
 
+static inline int ocfs2_refcount_tree(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE)
+		return 1;
+	return 0;
+}
+
 /* set / clear functions because cluster events can make these happen
  * in parallel so we want the transitions to be atomic. this also
  * means that any future flags osb_flags must be protected by spinlock
@@ -578,6 +621,9 @@ static inline int ocfs2_uses_extended_slot_map(struct ocfs2_super *osb)
 #define OCFS2_IS_VALID_DX_LEAF(ptr)					\
 	(!strcmp((ptr)->dl_signature, OCFS2_DX_LEAF_SIGNATURE))
 
+#define OCFS2_IS_VALID_REFCOUNT_BLOCK(ptr)				\
+	(!strcmp((ptr)->rf_signature, OCFS2_REFCOUNT_BLOCK_SIGNATURE))
+
 static inline unsigned long ino_from_blkno(struct super_block *sb,
 					   u64 blkno)
 {

fs/ocfs2/ocfs2_fs.h

@@ -68,6 +68,7 @@
 #define OCFS2_DIR_TRAILER_SIGNATURE	"DIRTRL1"
 #define OCFS2_DX_ROOT_SIGNATURE		"DXDIR01"
 #define OCFS2_DX_LEAF_SIGNATURE		"DXLEAF1"
+#define OCFS2_REFCOUNT_BLOCK_SIGNATURE	"REFCNT1"
 
 /* Compatibility flags */
 #define OCFS2_HAS_COMPAT_FEATURE(sb,mask)			\
@@ -98,7 +99,8 @@
 					 | OCFS2_FEATURE_INCOMPAT_USERSPACE_STACK \
 					 | OCFS2_FEATURE_INCOMPAT_XATTR \
 					 | OCFS2_FEATURE_INCOMPAT_META_ECC \
-					 | OCFS2_FEATURE_INCOMPAT_INDEXED_DIRS)
+					 | OCFS2_FEATURE_INCOMPAT_INDEXED_DIRS \
+					 | OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE)
 #define OCFS2_FEATURE_RO_COMPAT_SUPP	(OCFS2_FEATURE_RO_COMPAT_UNWRITTEN \
 					 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA \
 					 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
@@ -160,6 +162,9 @@
 /* Metadata checksum and error correction */
 #define OCFS2_FEATURE_INCOMPAT_META_ECC		0x0800
 
+/* Refcount tree support */
+#define OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE	0x1000
+
 /*
  * backup superblock flag is used to indicate that this volume
  * has backup superblocks.
@@ -223,6 +228,7 @@
 #define OCFS2_HAS_XATTR_FL	(0x0002)
 #define OCFS2_INLINE_XATTR_FL	(0x0004)
 #define OCFS2_INDEXED_DIR_FL	(0x0008)
+#define OCFS2_HAS_REFCOUNT_FL   (0x0010)
 
 /* Inode attributes, keep in sync with EXT2 */
 #define OCFS2_SECRM_FL		(0x00000001)	/* Secure deletion */
@@ -241,8 +247,11 @@
 /*
  * Extent record flags (e_node.leaf.flags)
  */
-#define OCFS2_EXT_UNWRITTEN	(0x01)	/* Extent is allocated but
-					 * unwritten */
+#define OCFS2_EXT_UNWRITTEN		(0x01)	/* Extent is allocated but
+						 * unwritten */
+#define OCFS2_EXT_REFCOUNTED		(0x02)  /* Extent is reference
+						 * counted in an associated
+						 * refcount tree */
 
 /*
  * ioctl commands
@@ -292,6 +301,15 @@ struct ocfs2_new_group_input {
 #define OCFS2_IOC_GROUP_ADD	_IOW('o', 2,struct ocfs2_new_group_input)
 #define OCFS2_IOC_GROUP_ADD64	_IOW('o', 3,struct ocfs2_new_group_input)
 
+/* Used to pass 2 file names to reflink. */
+struct reflink_arguments {
+	__u64 old_path;
+	__u64 new_path;
+	__u64 preserve;
+};
+#define OCFS2_IOC_REFLINK	_IOW('o', 4, struct reflink_arguments)
+
+
 /*
  * Journal Flags (ocfs2_dinode.id1.journal1.i_flags)
  */
@@ -717,7 +735,8 @@ struct ocfs2_dinode {
 	__le64 i_xattr_loc;
 /*80*/	struct ocfs2_block_check i_check;	/* Error checking */
 /*88*/	__le64 i_dx_root;		/* Pointer to dir index root block */
-	__le64 i_reserved2[5];
+/*90*/	__le64 i_refcount_loc;
+	__le64 i_reserved2[4];
 /*B8*/	union {
 		__le64 i_pad1;		/* Generic way to refer to this
 					   64bit union */
@@ -901,6 +920,60 @@ struct ocfs2_group_desc
 /*40*/	__u8    bg_bitmap[0];
 };
 
+struct ocfs2_refcount_rec {
+/*00*/	__le64 r_cpos;		/* Physical offset, in clusters */
+	__le32 r_clusters;	/* Clusters covered by this extent */
+	__le32 r_refcount;	/* Reference count of this extent */
+/*10*/
+};
+#define OCFS2_32BIT_POS_MASK		(0xffffffffULL)
+
+#define OCFS2_REFCOUNT_LEAF_FL          (0x00000001)
+#define OCFS2_REFCOUNT_TREE_FL          (0x00000002)
+
+struct ocfs2_refcount_list {
+/*00*/	__le16 rl_count;	/* Maximum number of entries possible
+				   in rl_records */
+	__le16 rl_used;		/* Current number of used records */
+	__le32 rl_reserved2;
+	__le64 rl_reserved1;	/* Pad to sizeof(ocfs2_refcount_record) */
+/*10*/	struct ocfs2_refcount_rec rl_recs[0];	/* Refcount records */
+};
+
+
+struct ocfs2_refcount_block {
+/*00*/	__u8 rf_signature[8];		/* Signature for verification */
+	__le16 rf_suballoc_slot;	/* Slot suballocator this block
+					   belongs to */
+	__le16 rf_suballoc_bit;		/* Bit offset in suballocator
+					   block group */
+	__le32 rf_fs_generation;	/* Must match superblock */
+/*10*/	__le64 rf_blkno;		/* Offset on disk, in blocks */
+	__le64 rf_parent;		/* Parent block, only valid if
+					   OCFS2_REFCOUNT_LEAF_FL is set in
+					   rf_flags */
+/*20*/	struct ocfs2_block_check rf_check;	/* Error checking */
+	__le64 rf_last_eb_blk;		/* Pointer to last extent block */
+/*30*/	__le32 rf_count;		/* Number of inodes sharing this
+					   refcount tree */
+	__le32 rf_flags;		/* See the flags above */
+	__le32 rf_clusters;		/* clusters covered by refcount tree. */
+	__le32 rf_cpos;			/* cluster offset in refcount tree.*/
+/*40*/	__le32 rf_generation;		/* generation number. all be the same
+					 * for the same refcount tree. */
+	__le32 rf_reserved0;
+	__le64 rf_reserved1[7];
+/*80*/	union {
+		struct ocfs2_refcount_list rf_records;  /* List of refcount
+							  records */
+		struct ocfs2_extent_list rf_list;	/* Extent record list,
+							only valid if
+							OCFS2_REFCOUNT_TREE_FL
+							is set in rf_flags */
+	};
+/* Actual on-disk size is one block */
+};
+
 /*
  * On disk extended attribute structure for OCFS2.
  */
@@ -1312,6 +1385,32 @@ static inline u16 ocfs2_xattr_recs_per_xb(struct super_block *sb)
 
 	return size / sizeof(struct ocfs2_extent_rec);
 }
+
+static inline u16 ocfs2_extent_recs_per_rb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_refcount_block, rf_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline u16 ocfs2_refcount_recs_per_rb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_refcount_block, rf_records.rl_recs);
+
+	return size / sizeof(struct ocfs2_refcount_rec);
+}
+
+static inline u32
+ocfs2_get_ref_rec_low_cpos(const struct ocfs2_refcount_rec *rec)
+{
+	return le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
+}
 #else
 static inline int ocfs2_fast_symlink_chars(int blocksize)
 {

fs/ocfs2/ocfs2_lockid.h

@@ -49,6 +49,7 @@ enum ocfs2_lock_type {
 	OCFS2_LOCK_TYPE_QINFO,
 	OCFS2_LOCK_TYPE_NFS_SYNC,
 	OCFS2_LOCK_TYPE_ORPHAN_SCAN,
+	OCFS2_LOCK_TYPE_REFCOUNT,
 	OCFS2_NUM_LOCK_TYPES
 };
 
@@ -89,6 +90,9 @@ static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)
 		case OCFS2_LOCK_TYPE_ORPHAN_SCAN:
 			c = 'P';
 			break;
+		case OCFS2_LOCK_TYPE_REFCOUNT:
+			c = 'T';
+			break;
 		default:
 			c = '\0';
 	}
@@ -110,6 +114,7 @@ static char *ocfs2_lock_type_strings[] = {
 	[OCFS2_LOCK_TYPE_QINFO] = "Quota",
 	[OCFS2_LOCK_TYPE_NFS_SYNC] = "NFSSync",
 	[OCFS2_LOCK_TYPE_ORPHAN_SCAN] = "OrphanScan",
+	[OCFS2_LOCK_TYPE_REFCOUNT] = "Refcount",
 };
 
 static inline const char *ocfs2_lock_type_string(enum ocfs2_lock_type type)

fs/ocfs2/quota_global.c

@@ -253,8 +253,9 @@ ssize_t ocfs2_quota_write(struct super_block *sb, int type,
 	flush_dcache_page(bh->b_page);
 	set_buffer_uptodate(bh);
 	unlock_buffer(bh);
-	ocfs2_set_buffer_uptodate(gqinode, bh);
-	err = ocfs2_journal_access_dq(handle, gqinode, bh, ja_type);
+	ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
+	err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
+				      ja_type);
 	if (err < 0) {
 		brelse(bh);
 		goto out;

fs/ocfs2/quota_local.c

@@ -108,7 +108,7 @@ static int ocfs2_modify_bh(struct inode *inode, struct buffer_head *bh,
 		mlog_errno(status);
 		return status;
 	}
-	status = ocfs2_journal_access_dq(handle, inode, bh,
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(inode), bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -510,7 +510,8 @@ static int ocfs2_recover_local_quota_file(struct inode *lqinode,
 				goto out_commit;
 			}
 			/* Release local quota file entry */
-			status = ocfs2_journal_access_dq(handle, lqinode,
+			status = ocfs2_journal_access_dq(handle,
+					INODE_CACHE(lqinode),
 					qbh, OCFS2_JOURNAL_ACCESS_WRITE);
 			if (status < 0) {
 				mlog_errno(status);
@@ -619,7 +620,8 @@ int ocfs2_finish_quota_recovery(struct ocfs2_super *osb,
 			mlog_errno(status);
 			goto out_bh;
 		}
-		status = ocfs2_journal_access_dq(handle, lqinode, bh,
+		status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode),
+						 bh,
 						 OCFS2_JOURNAL_ACCESS_WRITE);
 		if (status < 0) {
 			mlog_errno(status);
@@ -993,8 +995,8 @@ static struct ocfs2_quota_chunk *ocfs2_local_quota_add_chunk(
 		goto out_trans;
 	}
 	dchunk = (struct ocfs2_local_disk_chunk *)bh->b_data;
-	ocfs2_set_new_buffer_uptodate(lqinode, bh);
-	status = ocfs2_journal_access_dq(handle, lqinode, bh,
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), bh);
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), bh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1027,8 +1029,8 @@ static struct ocfs2_quota_chunk *ocfs2_local_quota_add_chunk(
 		mlog_errno(status);
 		goto out_trans;
 	}
-	ocfs2_set_new_buffer_uptodate(lqinode, dbh);
-	status = ocfs2_journal_access_dq(handle, lqinode, dbh,
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), dbh);
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), dbh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1131,7 +1133,7 @@ static struct ocfs2_quota_chunk *ocfs2_extend_local_quota_file(
 		mlog_errno(status);
 		goto out;
 	}
-	ocfs2_set_new_buffer_uptodate(lqinode, bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), bh);
 
 	/* Local quota info, chunk header and the new block we initialize */
 	handle = ocfs2_start_trans(OCFS2_SB(sb),
@@ -1143,7 +1145,7 @@ static struct ocfs2_quota_chunk *ocfs2_extend_local_quota_file(
 		goto out;
 	}
 	/* Zero created block */
-	status = ocfs2_journal_access_dq(handle, lqinode, bh,
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), bh,
 				 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1158,7 +1160,8 @@ static struct ocfs2_quota_chunk *ocfs2_extend_local_quota_file(
 		goto out_trans;
 	}
 	/* Update chunk header */
-	status = ocfs2_journal_access_dq(handle, lqinode, chunk->qc_headerbh,
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode),
+					 chunk->qc_headerbh,
 				 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1292,7 +1295,8 @@ static int ocfs2_local_release_dquot(struct dquot *dquot)
 		goto out;
 	}
 
-	status = ocfs2_journal_access_dq(handle, sb_dqopt(sb)->files[type],
+	status = ocfs2_journal_access_dq(handle,
+			INODE_CACHE(sb_dqopt(sb)->files[type]),
 			od->dq_chunk->qc_headerbh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);

fs/ocfs2/refcounttree.c

@@ -0,0 +1,4313 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * refcounttree.c
+ *
+ * Copyright (C) 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/sort.h>
+#define MLOG_MASK_PREFIX ML_REFCOUNT
+#include <cluster/masklog.h>
+#include "ocfs2.h"
+#include "inode.h"
+#include "alloc.h"
+#include "suballoc.h"
+#include "journal.h"
+#include "uptodate.h"
+#include "super.h"
+#include "buffer_head_io.h"
+#include "blockcheck.h"
+#include "refcounttree.h"
+#include "sysfile.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "aops.h"
+#include "xattr.h"
+#include "namei.h"
+
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/gfp.h>
+#include <linux/slab.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#include <linux/swap.h>
+#include <linux/security.h>
+#include <linux/fsnotify.h>
+#include <linux/quotaops.h>
+#include <linux/namei.h>
+#include <linux/mount.h>
+
+struct ocfs2_cow_context {
+	struct inode *inode;
+	u32 cow_start;
+	u32 cow_len;
+	struct ocfs2_extent_tree data_et;
+	struct ocfs2_refcount_tree *ref_tree;
+	struct buffer_head *ref_root_bh;
+	struct ocfs2_alloc_context *meta_ac;
+	struct ocfs2_alloc_context *data_ac;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	void *cow_object;
+	struct ocfs2_post_refcount *post_refcount;
+	int extra_credits;
+	int (*get_clusters)(struct ocfs2_cow_context *context,
+			    u32 v_cluster, u32 *p_cluster,
+			    u32 *num_clusters,
+			    unsigned int *extent_flags);
+	int (*cow_duplicate_clusters)(handle_t *handle,
+				      struct ocfs2_cow_context *context,
+				      u32 cpos, u32 old_cluster,
+				      u32 new_cluster, u32 new_len);
+};
+
+static inline struct ocfs2_refcount_tree *
+cache_info_to_refcount(struct ocfs2_caching_info *ci)
+{
+	return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
+}
+
+static int ocfs2_validate_refcount_block(struct super_block *sb,
+					 struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)bh->b_data;
+
+	mlog(0, "Validating refcount block %llu\n",
+	     (unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
+	if (rc) {
+		mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		return rc;
+	}
+
+
+	if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has bad signature %.*s",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    rb->rf_signature);
+		return -EINVAL;
+	}
+
+	if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has an invalid rf_blkno "
+			    "of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(rb->rf_blkno));
+		return -EINVAL;
+	}
+
+	if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has an invalid "
+			    "rf_fs_generation of #%u",
+			    (unsigned long long)bh->b_blocknr,
+			    le32_to_cpu(rb->rf_fs_generation));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
+				     u64 rb_blkno,
+				     struct buffer_head **bh)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_block(ci, rb_blkno, &tmp,
+			      ocfs2_validate_refcount_block);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	return rf->rf_blkno;
+}
+
+static struct super_block *
+ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	return rf->rf_sb;
+}
+
+static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	spin_lock(&rf->rf_lock);
+}
+
+static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	spin_unlock(&rf->rf_lock);
+}
+
+static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	mutex_lock(&rf->rf_io_mutex);
+}
+
+static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	mutex_unlock(&rf->rf_io_mutex);
+}
+
+static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
+	.co_owner		= ocfs2_refcount_cache_owner,
+	.co_get_super		= ocfs2_refcount_cache_get_super,
+	.co_cache_lock		= ocfs2_refcount_cache_lock,
+	.co_cache_unlock	= ocfs2_refcount_cache_unlock,
+	.co_io_lock		= ocfs2_refcount_cache_io_lock,
+	.co_io_unlock		= ocfs2_refcount_cache_io_unlock,
+};
+
+static struct ocfs2_refcount_tree *
+ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
+{
+	struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
+	struct ocfs2_refcount_tree *tree = NULL;
+
+	while (n) {
+		tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
+
+		if (blkno < tree->rf_blkno)
+			n = n->rb_left;
+		else if (blkno > tree->rf_blkno)
+			n = n->rb_right;
+		else
+			return tree;
+	}
+
+	return NULL;
+}
+
+/* osb_lock is already locked. */
+static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
+				       struct ocfs2_refcount_tree *new)
+{
+	u64 rf_blkno = new->rf_blkno;
+	struct rb_node *parent = NULL;
+	struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
+	struct ocfs2_refcount_tree *tmp;
+
+	while (*p) {
+		parent = *p;
+
+		tmp = rb_entry(parent, struct ocfs2_refcount_tree,
+			       rf_node);
+
+		if (rf_blkno < tmp->rf_blkno)
+			p = &(*p)->rb_left;
+		else if (rf_blkno > tmp->rf_blkno)
+			p = &(*p)->rb_right;
+		else {
+			/* This should never happen! */
+			mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
+			     (unsigned long long)rf_blkno);
+			BUG();
+		}
+	}
+
+	rb_link_node(&new->rf_node, parent, p);
+	rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
+}
+
+static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
+{
+	ocfs2_metadata_cache_exit(&tree->rf_ci);
+	ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
+	ocfs2_lock_res_free(&tree->rf_lockres);
+	kfree(tree);
+}
+
+static inline void
+ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *tree)
+{
+	rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
+	if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
+		osb->osb_ref_tree_lru = NULL;
+}
+
+static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *tree)
+{
+	spin_lock(&osb->osb_lock);
+	ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
+	spin_unlock(&osb->osb_lock);
+}
+
+void ocfs2_kref_remove_refcount_tree(struct kref *kref)
+{
+	struct ocfs2_refcount_tree *tree =
+		container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
+
+	ocfs2_free_refcount_tree(tree);
+}
+
+static inline void
+ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
+{
+	kref_get(&tree->rf_getcnt);
+}
+
+static inline void
+ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
+{
+	kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
+}
+
+static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
+					       struct super_block *sb)
+{
+	ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
+	mutex_init(&new->rf_io_mutex);
+	new->rf_sb = sb;
+	spin_lock_init(&new->rf_lock);
+}
+
+static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *new,
+					u64 rf_blkno, u32 generation)
+{
+	init_rwsem(&new->rf_sem);
+	ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
+				     rf_blkno, generation);
+}
+
+static struct ocfs2_refcount_tree*
+ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
+{
+	struct ocfs2_refcount_tree *new;
+
+	new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
+	if (!new)
+		return NULL;
+
+	new->rf_blkno = rf_blkno;
+	kref_init(&new->rf_getcnt);
+	ocfs2_init_refcount_tree_ci(new, osb->sb);
+
+	return new;
+}
+
+static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
+				   struct ocfs2_refcount_tree **ret_tree)
+{
+	int ret = 0;
+	struct ocfs2_refcount_tree *tree, *new = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *ref_rb;
+
+	spin_lock(&osb->osb_lock);
+	if (osb->osb_ref_tree_lru &&
+	    osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
+		tree = osb->osb_ref_tree_lru;
+	else
+		tree = ocfs2_find_refcount_tree(osb, rf_blkno);
+	if (tree)
+		goto out;
+
+	spin_unlock(&osb->osb_lock);
+
+	new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
+	if (!new) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		return ret;
+	}
+	/*
+	 * We need the generation to create the refcount tree lock and since
+	 * it isn't changed during the tree modification, we are safe here to
+	 * read without protection.
+	 * We also have to purge the cache after we create the lock since the
+	 * refcount block may have the stale data. It can only be trusted when
+	 * we hold the refcount lock.
+	 */
+	ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_metadata_cache_exit(&new->rf_ci);
+		kfree(new);
+		return ret;
+	}
+
+	ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
+	ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
+				      new->rf_generation);
+	ocfs2_metadata_cache_purge(&new->rf_ci);
+
+	spin_lock(&osb->osb_lock);
+	tree = ocfs2_find_refcount_tree(osb, rf_blkno);
+	if (tree)
+		goto out;
+
+	ocfs2_insert_refcount_tree(osb, new);
+
+	tree = new;
+	new = NULL;
+
+out:
+	*ret_tree = tree;
+
+	osb->osb_ref_tree_lru = tree;
+
+	spin_unlock(&osb->osb_lock);
+
+	if (new)
+		ocfs2_free_refcount_tree(new);
+
+	brelse(ref_root_bh);
+	return ret;
+}
+
+static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	*ref_blkno = le64_to_cpu(di->i_refcount_loc);
+	brelse(di_bh);
+out:
+	return ret;
+}
+
+static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
+				      struct ocfs2_refcount_tree *tree, int rw)
+{
+	int ret;
+
+	ret = ocfs2_refcount_lock(tree, rw);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (rw)
+		down_write(&tree->rf_sem);
+	else
+		down_read(&tree->rf_sem);
+
+out:
+	return ret;
+}
+
+/*
+ * Lock the refcount tree pointed by ref_blkno and return the tree.
+ * In most case, we lock the tree and read the refcount block.
+ * So read it here if the caller really needs it.
+ *
+ * If the tree has been re-created by other node, it will free the
+ * old one and re-create it.
+ */
+int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
+			     u64 ref_blkno, int rw,
+			     struct ocfs2_refcount_tree **ret_tree,
+			     struct buffer_head **ref_bh)
+{
+	int ret, delete_tree = 0;
+	struct ocfs2_refcount_tree *tree = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+
+again:
+	ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ocfs2_refcount_tree_get(tree);
+
+	ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_refcount_tree_put(tree);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_unlock_refcount_tree(osb, tree, rw);
+		ocfs2_refcount_tree_put(tree);
+		goto out;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	/*
+	 * If the refcount block has been freed and re-created, we may need
+	 * to recreate the refcount tree also.
+	 *
+	 * Here we just remove the tree from the rb-tree, and the last
+	 * kref holder will unlock and delete this refcount_tree.
+	 * Then we goto "again" and ocfs2_get_refcount_tree will create
+	 * the new refcount tree for us.
+	 */
+	if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
+		if (!tree->rf_removed) {
+			ocfs2_erase_refcount_tree_from_list(osb, tree);
+			tree->rf_removed = 1;
+			delete_tree = 1;
+		}
+
+		ocfs2_unlock_refcount_tree(osb, tree, rw);
+		/*
+		 * We get an extra reference when we create the refcount
+		 * tree, so another put will destroy it.
+		 */
+		if (delete_tree)
+			ocfs2_refcount_tree_put(tree);
+		brelse(ref_root_bh);
+		ref_root_bh = NULL;
+		goto again;
+	}
+
+	*ret_tree = tree;
+	if (ref_bh) {
+		*ref_bh = ref_root_bh;
+		ref_root_bh = NULL;
+	}
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+int ocfs2_lock_refcount_tree_by_inode(struct inode *inode, int rw,
+				      struct ocfs2_refcount_tree **ret_tree,
+				      struct buffer_head **ref_bh)
+{
+	int ret;
+	u64 ref_blkno;
+
+	ret = ocfs2_get_refcount_block(inode, &ref_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	return ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno,
+					rw, ret_tree, ref_bh);
+}
+
+void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
+				struct ocfs2_refcount_tree *tree, int rw)
+{
+	if (rw)
+		up_write(&tree->rf_sem);
+	else
+		up_read(&tree->rf_sem);
+
+	ocfs2_refcount_unlock(tree, rw);
+	ocfs2_refcount_tree_put(tree);
+}
+
+void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
+{
+	struct rb_node *node;
+	struct ocfs2_refcount_tree *tree;
+	struct rb_root *root = &osb->osb_rf_lock_tree;
+
+	while ((node = rb_last(root)) != NULL) {
+		tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
+
+		mlog(0, "Purge tree %llu\n",
+		     (unsigned long long) tree->rf_blkno);
+
+		rb_erase(&tree->rf_node, root);
+		ocfs2_free_refcount_tree(tree);
+	}
+}
+
+/*
+ * Create a refcount tree for an inode.
+ * We take for granted that the inode is already locked.
+ */
+static int ocfs2_create_refcount_tree(struct inode *inode,
+				      struct buffer_head *di_bh)
+{
+	int ret;
+	handle_t *handle = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 first_blkno;
+
+	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
+
+	mlog(0, "create tree for inode %lu\n", inode->i_ino);
+
+	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
+				   &suballoc_bit_start, &num_got,
+				   &first_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
+	if (!new_tree) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	new_bh = sb_getblk(inode->i_sb, first_blkno);
+	ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/* Initialize ocfs2_refcount_block. */
+	rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	memset(rb, 0, inode->i_sb->s_blocksize);
+	strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
+	rb->rf_suballoc_slot = cpu_to_le16(osb->slot_num);
+	rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
+	rb->rf_blkno = cpu_to_le64(first_blkno);
+	rb->rf_count = cpu_to_le32(1);
+	rb->rf_records.rl_count =
+			cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
+	spin_lock(&osb->osb_lock);
+	rb->rf_generation = osb->s_next_generation++;
+	spin_unlock(&osb->osb_lock);
+
+	ocfs2_journal_dirty(handle, new_bh);
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = cpu_to_le64(first_blkno);
+	spin_unlock(&oi->ip_lock);
+
+	mlog(0, "created tree for inode %lu, refblock %llu\n",
+	     inode->i_ino, (unsigned long long)first_blkno);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	/*
+	 * We have to init the tree lock here since it will use
+	 * the generation number to create it.
+	 */
+	new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
+	ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
+				      new_tree->rf_generation);
+
+	spin_lock(&osb->osb_lock);
+	tree = ocfs2_find_refcount_tree(osb, first_blkno);
+
+	/*
+	 * We've just created a new refcount tree in this block.  If
+	 * we found a refcount tree on the ocfs2_super, it must be
+	 * one we just deleted.  We free the old tree before
+	 * inserting the new tree.
+	 */
+	BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
+	if (tree)
+		ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
+	ocfs2_insert_refcount_tree(osb, new_tree);
+	spin_unlock(&osb->osb_lock);
+	new_tree = NULL;
+	if (tree)
+		ocfs2_refcount_tree_put(tree);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (new_tree) {
+		ocfs2_metadata_cache_exit(&new_tree->rf_ci);
+		kfree(new_tree);
+	}
+
+	brelse(new_bh);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	return ret;
+}
+
+static int ocfs2_set_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh,
+				   u64 refcount_loc)
+{
+	int ret;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_tree *ref_tree;
+
+	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
+
+	ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
+				       &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	le32_add_cpu(&rb->rf_count, 1);
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = cpu_to_le64(refcount_loc);
+	spin_unlock(&oi->ip_lock);
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+
+	return ret;
+}
+
+int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
+{
+	int ret, delete_tree = 0;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_refcount_block *rb;
+	struct inode *alloc_inode = NULL;
+	struct buffer_head *alloc_bh = NULL;
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_refcount_tree *ref_tree;
+	int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
+	u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
+	u16 bit = 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
+		return 0;
+
+	BUG_ON(!ref_blkno);
+	ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
+
+	/*
+	 * If we are the last user, we need to free the block.
+	 * So lock the allocator ahead.
+	 */
+	if (le32_to_cpu(rb->rf_count) == 1) {
+		blk = le64_to_cpu(rb->rf_blkno);
+		bit = le16_to_cpu(rb->rf_suballoc_bit);
+		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
+
+		alloc_inode = ocfs2_get_system_file_inode(osb,
+					EXTENT_ALLOC_SYSTEM_INODE,
+					le16_to_cpu(rb->rf_suballoc_slot));
+		if (!alloc_inode) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+		mutex_lock(&alloc_inode->i_mutex);
+
+		ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_mutex;
+		}
+
+		credits += OCFS2_SUBALLOC_FREE;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = 0;
+	spin_unlock(&oi->ip_lock);
+	ocfs2_journal_dirty(handle, di_bh);
+
+	le32_add_cpu(&rb->rf_count , -1);
+	ocfs2_journal_dirty(handle, blk_bh);
+
+	if (!rb->rf_count) {
+		delete_tree = 1;
+		ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
+		ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
+					       alloc_bh, bit, bg_blkno, 1);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out_unlock:
+	if (alloc_inode) {
+		ocfs2_inode_unlock(alloc_inode, 1);
+		brelse(alloc_bh);
+	}
+out_mutex:
+	if (alloc_inode) {
+		mutex_unlock(&alloc_inode->i_mutex);
+		iput(alloc_inode);
+	}
+out:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	if (delete_tree)
+		ocfs2_refcount_tree_put(ref_tree);
+	brelse(blk_bh);
+
+	return ret;
+}
+
+static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
+					  struct buffer_head *ref_leaf_bh,
+					  u64 cpos, unsigned int len,
+					  struct ocfs2_refcount_rec *ret_rec,
+					  int *index)
+{
+	int i = 0;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_rec *rec = NULL;
+
+	for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
+		rec = &rb->rf_records.rl_recs[i];
+
+		if (le64_to_cpu(rec->r_cpos) +
+		    le32_to_cpu(rec->r_clusters) <= cpos)
+			continue;
+		else if (le64_to_cpu(rec->r_cpos) > cpos)
+			break;
+
+		/* ok, cpos fail in this rec. Just return. */
+		if (ret_rec)
+			*ret_rec = *rec;
+		goto out;
+	}
+
+	if (ret_rec) {
+		/* We meet with a hole here, so fake the rec. */
+		ret_rec->r_cpos = cpu_to_le64(cpos);
+		ret_rec->r_refcount = 0;
+		if (i < le16_to_cpu(rb->rf_records.rl_used) &&
+		    le64_to_cpu(rec->r_cpos) < cpos + len)
+			ret_rec->r_clusters =
+				cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
+		else
+			ret_rec->r_clusters = cpu_to_le32(len);
+	}
+
+out:
+	*index = i;
+}
+
+/*
+ * Try to remove refcount tree. The mechanism is:
+ * 1) Check whether i_clusters == 0, if no, exit.
+ * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
+ * 3) Check whether we have inline xattr stored outside, if yes, exit.
+ * 4) Remove the tree.
+ */
+int ocfs2_try_remove_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh)
+{
+	int ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	down_write(&oi->ip_xattr_sem);
+	down_write(&oi->ip_alloc_sem);
+
+	if (oi->ip_clusters)
+		goto out;
+
+	if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
+		goto out;
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
+	    ocfs2_has_inline_xattr_value_outside(inode, di))
+		goto out;
+
+	ret = ocfs2_remove_refcount_tree(inode, di_bh);
+	if (ret)
+		mlog_errno(ret);
+out:
+	up_write(&oi->ip_alloc_sem);
+	up_write(&oi->ip_xattr_sem);
+	return 0;
+}
+
+/*
+ * Given a cpos and len, try to find the refcount record which contains cpos.
+ * 1. If cpos can be found in one refcount record, return the record.
+ * 2. If cpos can't be found, return a fake record which start from cpos
+ *    and end at a small value between cpos+len and start of the next record.
+ *    This fake record has r_refcount = 0.
+ */
+static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
+				  struct buffer_head *ref_root_bh,
+				  u64 cpos, unsigned int len,
+				  struct ocfs2_refcount_rec *ret_rec,
+				  int *index,
+				  struct buffer_head **ret_bh)
+{
+	int ret = 0, i, found;
+	u32 low_cpos;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *tmp, *rec = NULL;
+	struct ocfs2_extent_block *eb;
+	struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
+		ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
+					      ret_rec, index);
+		*ret_bh = ref_root_bh;
+		get_bh(ref_root_bh);
+		return 0;
+	}
+
+	el = &rb->rf_list;
+	low_cpos = cpos & OCFS2_32BIT_POS_MASK;
+
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(sb,
+			"refcount tree %llu has non zero tree "
+			"depth in leaf btree tree block %llu\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci),
+			(unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	found = 0;
+	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
+		rec = &el->l_recs[i];
+
+		if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
+			found = 1;
+			break;
+		}
+	}
+
+	/* adjust len when we have ocfs2_extent_rec after it. */
+	if (found && i < le16_to_cpu(el->l_next_free_rec) - 1) {
+		tmp = &el->l_recs[i+1];
+
+		if (le32_to_cpu(tmp->e_cpos) < cpos + len)
+			len = le32_to_cpu(tmp->e_cpos) - cpos;
+	}
+
+	ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
+					&ref_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
+				      ret_rec, index);
+	*ret_bh = ref_leaf_bh;
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+enum ocfs2_ref_rec_contig {
+	REF_CONTIG_NONE = 0,
+	REF_CONTIG_LEFT,
+	REF_CONTIG_RIGHT,
+	REF_CONTIG_LEFTRIGHT,
+};
+
+static enum ocfs2_ref_rec_contig
+	ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
+				    int index)
+{
+	if ((rb->rf_records.rl_recs[index].r_refcount ==
+	    rb->rf_records.rl_recs[index + 1].r_refcount) &&
+	    (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
+	    le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
+	    le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
+		return REF_CONTIG_RIGHT;
+
+	return REF_CONTIG_NONE;
+}
+
+static enum ocfs2_ref_rec_contig
+	ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
+				  int index)
+{
+	enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
+
+	if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
+		ret = ocfs2_refcount_rec_adjacent(rb, index);
+
+	if (index > 0) {
+		enum ocfs2_ref_rec_contig tmp;
+
+		tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
+
+		if (tmp == REF_CONTIG_RIGHT) {
+			if (ret == REF_CONTIG_RIGHT)
+				ret = REF_CONTIG_LEFTRIGHT;
+			else
+				ret = REF_CONTIG_LEFT;
+		}
+	}
+
+	return ret;
+}
+
+static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
+					   int index)
+{
+	BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
+	       rb->rf_records.rl_recs[index+1].r_refcount);
+
+	le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
+		     le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
+
+	if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
+		memmove(&rb->rf_records.rl_recs[index + 1],
+			&rb->rf_records.rl_recs[index + 2],
+			sizeof(struct ocfs2_refcount_rec) *
+			(le16_to_cpu(rb->rf_records.rl_used) - index - 2));
+
+	memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
+	       0, sizeof(struct ocfs2_refcount_rec));
+	le16_add_cpu(&rb->rf_records.rl_used, -1);
+}
+
+/*
+ * Merge the refcount rec if we are contiguous with the adjacent recs.
+ */
+static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
+				     int index)
+{
+	enum ocfs2_ref_rec_contig contig =
+				ocfs2_refcount_rec_contig(rb, index);
+
+	if (contig == REF_CONTIG_NONE)
+		return;
+
+	if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
+		BUG_ON(index == 0);
+		index--;
+	}
+
+	ocfs2_rotate_refcount_rec_left(rb, index);
+
+	if (contig == REF_CONTIG_LEFTRIGHT)
+		ocfs2_rotate_refcount_rec_left(rb, index);
+}
+
+/*
+ * Change the refcount indexed by "index" in ref_bh.
+ * If refcount reaches 0, remove it.
+ */
+static int ocfs2_change_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_leaf_bh,
+				     int index, int merge, int change)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rl = &rb->rf_records;
+	struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "change index %d, old count %u, change %d\n", index,
+	     le32_to_cpu(rec->r_refcount), change);
+	le32_add_cpu(&rec->r_refcount, change);
+
+	if (!rec->r_refcount) {
+		if (index != le16_to_cpu(rl->rl_used) - 1) {
+			memmove(rec, rec + 1,
+				(le16_to_cpu(rl->rl_used) - index - 1) *
+				sizeof(struct ocfs2_refcount_rec));
+			memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
+			       0, sizeof(struct ocfs2_refcount_rec));
+		}
+
+		le16_add_cpu(&rl->rl_used, -1);
+	} else if (merge)
+		ocfs2_refcount_rec_merge(rb, index);
+
+	ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
+	if (ret)
+		mlog_errno(ret);
+out:
+	return ret;
+}
+
+static int ocfs2_expand_inline_ref_root(handle_t *handle,
+					struct ocfs2_caching_info *ci,
+					struct buffer_head *ref_root_bh,
+					struct buffer_head **ref_leaf_bh,
+					struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 blkno;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *new_rb;
+	struct ocfs2_refcount_block *root_rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_claim_metadata(OCFS2_SB(sb), handle, meta_ac, 1,
+				   &suballoc_bit_start, &num_got,
+				   &blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	new_bh = sb_getblk(sb, blkno);
+	if (new_bh == NULL) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Initialize ocfs2_refcount_block.
+	 * It should contain the same information as the old root.
+	 * so just memcpy it and change the corresponding field.
+	 */
+	memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
+
+	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	new_rb->rf_suballoc_slot = cpu_to_le16(OCFS2_SB(sb)->slot_num);
+	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	new_rb->rf_blkno = cpu_to_le64(blkno);
+	new_rb->rf_cpos = cpu_to_le32(0);
+	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
+	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
+	ocfs2_journal_dirty(handle, new_bh);
+
+	/* Now change the root. */
+	memset(&root_rb->rf_list, 0, sb->s_blocksize -
+	       offsetof(struct ocfs2_refcount_block, rf_list));
+	root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
+	root_rb->rf_clusters = cpu_to_le32(1);
+	root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
+	root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
+	root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
+	root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+	mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno,
+	     le16_to_cpu(new_rb->rf_records.rl_used));
+
+	*ref_leaf_bh = new_bh;
+	new_bh = NULL;
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
+					   struct ocfs2_refcount_rec *next)
+{
+	if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
+		ocfs2_get_ref_rec_low_cpos(next))
+		return 1;
+
+	return 0;
+}
+
+static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
+{
+	const struct ocfs2_refcount_rec *l = a, *r = b;
+	u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
+	u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
+
+	if (l_cpos > r_cpos)
+		return 1;
+	if (l_cpos < r_cpos)
+		return -1;
+	return 0;
+}
+
+static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
+{
+	const struct ocfs2_refcount_rec *l = a, *r = b;
+	u64 l_cpos = le64_to_cpu(l->r_cpos);
+	u64 r_cpos = le64_to_cpu(r->r_cpos);
+
+	if (l_cpos > r_cpos)
+		return 1;
+	if (l_cpos < r_cpos)
+		return -1;
+	return 0;
+}
+
+static void swap_refcount_rec(void *a, void *b, int size)
+{
+	struct ocfs2_refcount_rec *l = a, *r = b, tmp;
+
+	tmp = *(struct ocfs2_refcount_rec *)l;
+	*(struct ocfs2_refcount_rec *)l =
+			*(struct ocfs2_refcount_rec *)r;
+	*(struct ocfs2_refcount_rec *)r = tmp;
+}
+
+/*
+ * The refcount cpos are ordered by their 64bit cpos,
+ * But we will use the low 32 bit to be the e_cpos in the b-tree.
+ * So we need to make sure that this pos isn't intersected with others.
+ *
+ * Note: The refcount block is already sorted by their low 32 bit cpos,
+ *       So just try the middle pos first, and we will exit when we find
+ *       the good position.
+ */
+static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
+					 u32 *split_pos, int *split_index)
+{
+	int num_used = le16_to_cpu(rl->rl_used);
+	int delta, middle = num_used / 2;
+
+	for (delta = 0; delta < middle; delta++) {
+		/* Let's check delta earlier than middle */
+		if (ocfs2_refcount_rec_no_intersect(
+					&rl->rl_recs[middle - delta - 1],
+					&rl->rl_recs[middle - delta])) {
+			*split_index = middle - delta;
+			break;
+		}
+
+		/* For even counts, don't walk off the end */
+		if ((middle + delta + 1) == num_used)
+			continue;
+
+		/* Now try delta past middle */
+		if (ocfs2_refcount_rec_no_intersect(
+					&rl->rl_recs[middle + delta],
+					&rl->rl_recs[middle + delta + 1])) {
+			*split_index = middle + delta + 1;
+			break;
+		}
+	}
+
+	if (delta >= middle)
+		return -ENOSPC;
+
+	*split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
+	return 0;
+}
+
+static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
+					    struct buffer_head *new_bh,
+					    u32 *split_cpos)
+{
+	int split_index = 0, num_moved, ret;
+	u32 cpos = 0;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rl = &rb->rf_records;
+	struct ocfs2_refcount_block *new_rb =
+			(struct ocfs2_refcount_block *)new_bh->b_data;
+	struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
+
+	mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
+	     (unsigned long long)ref_leaf_bh->b_blocknr,
+	     le32_to_cpu(rl->rl_count), le32_to_cpu(rl->rl_used));
+
+	/*
+	 * XXX: Improvement later.
+	 * If we know all the high 32 bit cpos is the same, no need to sort.
+	 *
+	 * In order to make the whole process safe, we do:
+	 * 1. sort the entries by their low 32 bit cpos first so that we can
+	 *    find the split cpos easily.
+	 * 2. call ocfs2_insert_extent to insert the new refcount block.
+	 * 3. move the refcount rec to the new block.
+	 * 4. sort the entries by their 64 bit cpos.
+	 * 5. dirty the new_rb and rb.
+	 */
+	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
+
+	ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	new_rb->rf_cpos = cpu_to_le32(cpos);
+
+	/* move refcount records starting from split_index to the new block. */
+	num_moved = le16_to_cpu(rl->rl_used) - split_index;
+	memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
+	       num_moved * sizeof(struct ocfs2_refcount_rec));
+
+	/*ok, remove the entries we just moved over to the other block. */
+	memset(&rl->rl_recs[split_index], 0,
+	       num_moved * sizeof(struct ocfs2_refcount_rec));
+
+	/* change old and new rl_used accordingly. */
+	le16_add_cpu(&rl->rl_used, -num_moved);
+	new_rl->rl_used = cpu_to_le32(num_moved);
+
+	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
+
+	sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
+
+	*split_cpos = cpos;
+	return 0;
+}
+
+static int ocfs2_new_leaf_refcount_block(handle_t *handle,
+					 struct ocfs2_caching_info *ci,
+					 struct buffer_head *ref_root_bh,
+					 struct buffer_head *ref_leaf_bh,
+					 struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	u16 suballoc_bit_start;
+	u32 num_got, new_cpos;
+	u64 blkno;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *root_rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *new_rb;
+	struct ocfs2_extent_tree ref_et;
+
+	BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_claim_metadata(OCFS2_SB(sb), handle, meta_ac, 1,
+				   &suballoc_bit_start, &num_got,
+				   &blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	new_bh = sb_getblk(sb, blkno);
+	if (new_bh == NULL) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Initialize ocfs2_refcount_block. */
+	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	memset(new_rb, 0, sb->s_blocksize);
+	strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
+	new_rb->rf_suballoc_slot = cpu_to_le16(OCFS2_SB(sb)->slot_num);
+	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
+	new_rb->rf_blkno = cpu_to_le64(blkno);
+	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
+	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
+	new_rb->rf_records.rl_count =
+				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
+	new_rb->rf_generation = root_rb->rf_generation;
+
+	ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+	ocfs2_journal_dirty(handle, new_bh);
+
+	ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
+
+	mlog(0, "insert new leaf block %llu at %u\n",
+	     (unsigned long long)new_bh->b_blocknr, new_cpos);
+
+	/* Insert the new leaf block with the specific offset cpos. */
+	ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
+				  1, 0, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int ocfs2_expand_refcount_tree(handle_t *handle,
+				      struct ocfs2_caching_info *ci,
+				      struct buffer_head *ref_root_bh,
+				      struct buffer_head *ref_leaf_bh,
+				      struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	struct buffer_head *expand_bh = NULL;
+
+	if (ref_root_bh == ref_leaf_bh) {
+		/*
+		 * the old root bh hasn't been expanded to a b-tree,
+		 * so expand it first.
+		 */
+		ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
+						   &expand_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else {
+		expand_bh = ref_leaf_bh;
+		get_bh(expand_bh);
+	}
+
+
+	/* Now add a new refcount block into the tree.*/
+	ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
+					    expand_bh, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(expand_bh);
+	return ret;
+}
+
+/*
+ * Adjust the extent rec in b-tree representing ref_leaf_bh.
+ *
+ * Only called when we have inserted a new refcount rec at index 0
+ * which means ocfs2_extent_rec.e_cpos may need some change.
+ */
+static int ocfs2_adjust_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     struct buffer_head *ref_leaf_bh,
+				     struct ocfs2_refcount_rec *rec)
+{
+	int ret = 0, i;
+	u32 new_cpos, old_cpos;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	struct ocfs2_extent_list *el;
+
+	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
+		goto out;
+
+	rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	old_cpos = le32_to_cpu(rb->rf_cpos);
+	new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
+	if (old_cpos <= new_cpos)
+		goto out;
+
+	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+
+	path = ocfs2_new_path_from_et(&et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(ci, path, old_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * 2 more credits, one for the leaf refcount block, one for
+	 * the extent block contains the extent rec.
+	 */
+	ret = ocfs2_extend_trans(handle, handle->h_buffer_credits + 2);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* change the leaf extent block first. */
+	el = path_leaf_el(path);
+
+	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
+		if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
+			break;
+
+	BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
+
+	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
+
+	/* change the r_cpos in the leaf block. */
+	rb->rf_cpos = cpu_to_le32(new_cpos);
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(path));
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+static int ocfs2_insert_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     struct buffer_head *ref_leaf_bh,
+				     struct ocfs2_refcount_rec *rec,
+				     int index, int merge,
+				     struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
+	struct buffer_head *new_bh = NULL;
+
+	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
+
+	if (rf_list->rl_used == rf_list->rl_count) {
+		u64 cpos = le64_to_cpu(rec->r_cpos);
+		u32 len = le32_to_cpu(rec->r_clusters);
+
+		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
+						 ref_leaf_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, NULL, &index,
+					     &new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ref_leaf_bh = new_bh;
+		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+		rf_list = &rb->rf_records;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (index < le16_to_cpu(rf_list->rl_used))
+		memmove(&rf_list->rl_recs[index + 1],
+			&rf_list->rl_recs[index],
+			(le16_to_cpu(rf_list->rl_used) - index) *
+			 sizeof(struct ocfs2_refcount_rec));
+
+	mlog(0, "insert refcount record start %llu, len %u, count %u "
+	     "to leaf block %llu at index %d\n",
+	     (unsigned long long)le64_to_cpu(rec->r_cpos),
+	     le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
+	     (unsigned long long)ref_leaf_bh->b_blocknr, index);
+
+	rf_list->rl_recs[index] = *rec;
+
+	le16_add_cpu(&rf_list->rl_used, 1);
+
+	if (merge)
+		ocfs2_refcount_rec_merge(rb, index);
+
+	ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (index == 0) {
+		ret = ocfs2_adjust_refcount_rec(handle, ci,
+						ref_root_bh,
+						ref_leaf_bh, rec);
+		if (ret)
+			mlog_errno(ret);
+	}
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+/*
+ * Split the refcount_rec indexed by "index" in ref_leaf_bh.
+ * This is much simple than our b-tree code.
+ * split_rec is the new refcount rec we want to insert.
+ * If split_rec->r_refcount > 0, we are changing the refcount(in case we
+ * increase refcount or decrease a refcount to non-zero).
+ * If split_rec->r_refcount == 0, we are punching a hole in current refcount
+ * rec( in case we decrease a refcount to zero).
+ */
+static int ocfs2_split_refcount_rec(handle_t *handle,
+				    struct ocfs2_caching_info *ci,
+				    struct buffer_head *ref_root_bh,
+				    struct buffer_head *ref_leaf_bh,
+				    struct ocfs2_refcount_rec *split_rec,
+				    int index, int merge,
+				    struct ocfs2_alloc_context *meta_ac,
+				    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, recs_need;
+	u32 len;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
+	struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
+	struct ocfs2_refcount_rec *tail_rec = NULL;
+	struct buffer_head *new_bh = NULL;
+
+	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
+
+	mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
+	     le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
+	     le64_to_cpu(split_rec->r_cpos),
+	     le32_to_cpu(split_rec->r_clusters));
+
+	/*
+	 * If we just need to split the header or tail clusters,
+	 * no more recs are needed, just split is OK.
+	 * Otherwise we at least need one new recs.
+	 */
+	if (!split_rec->r_refcount &&
+	    (split_rec->r_cpos == orig_rec->r_cpos ||
+	     le64_to_cpu(split_rec->r_cpos) +
+	     le32_to_cpu(split_rec->r_clusters) ==
+	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
+		recs_need = 0;
+	else
+		recs_need = 1;
+
+	/*
+	 * We need one more rec if we split in the middle and the new rec have
+	 * some refcount in it.
+	 */
+	if (split_rec->r_refcount &&
+	    (split_rec->r_cpos != orig_rec->r_cpos &&
+	     le64_to_cpu(split_rec->r_cpos) +
+	     le32_to_cpu(split_rec->r_clusters) !=
+	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
+		recs_need++;
+
+	/* If the leaf block don't have enough record, expand it. */
+	if (le16_to_cpu(rf_list->rl_used) + recs_need > rf_list->rl_count) {
+		struct ocfs2_refcount_rec tmp_rec;
+		u64 cpos = le64_to_cpu(orig_rec->r_cpos);
+		len = le32_to_cpu(orig_rec->r_clusters);
+		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
+						 ref_leaf_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * We have to re-get it since now cpos may be moved to
+		 * another leaf block.
+		 */
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &tmp_rec, &index,
+					     &new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ref_leaf_bh = new_bh;
+		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+		rf_list = &rb->rf_records;
+		orig_rec = &rf_list->rl_recs[index];
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We have calculated out how many new records we need and store
+	 * in recs_need, so spare enough space first by moving the records
+	 * after "index" to the end.
+	 */
+	if (index != le16_to_cpu(rf_list->rl_used) - 1)
+		memmove(&rf_list->rl_recs[index + 1 + recs_need],
+			&rf_list->rl_recs[index + 1],
+			(le16_to_cpu(rf_list->rl_used) - index - 1) *
+			 sizeof(struct ocfs2_refcount_rec));
+
+	len = (le64_to_cpu(orig_rec->r_cpos) +
+	      le32_to_cpu(orig_rec->r_clusters)) -
+	      (le64_to_cpu(split_rec->r_cpos) +
+	      le32_to_cpu(split_rec->r_clusters));
+
+	/*
+	 * If we have "len", the we will split in the tail and move it
+	 * to the end of the space we have just spared.
+	 */
+	if (len) {
+		tail_rec = &rf_list->rl_recs[index + recs_need];
+
+		memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
+		le64_add_cpu(&tail_rec->r_cpos,
+			     le32_to_cpu(tail_rec->r_clusters) - len);
+		tail_rec->r_clusters = le32_to_cpu(len);
+	}
+
+	/*
+	 * If the split pos isn't the same as the original one, we need to
+	 * split in the head.
+	 *
+	 * Note: We have the chance that split_rec.r_refcount = 0,
+	 * recs_need = 0 and len > 0, which means we just cut the head from
+	 * the orig_rec and in that case we have done some modification in
+	 * orig_rec above, so the check for r_cpos is faked.
+	 */
+	if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
+		len = le64_to_cpu(split_rec->r_cpos) -
+		      le64_to_cpu(orig_rec->r_cpos);
+		orig_rec->r_clusters = cpu_to_le32(len);
+		index++;
+	}
+
+	le16_add_cpu(&rf_list->rl_used, recs_need);
+
+	if (split_rec->r_refcount) {
+		rf_list->rl_recs[index] = *split_rec;
+		mlog(0, "insert refcount record start %llu, len %u, count %u "
+		     "to leaf block %llu at index %d\n",
+		     (unsigned long long)le64_to_cpu(split_rec->r_cpos),
+		     le32_to_cpu(split_rec->r_clusters),
+		     le32_to_cpu(split_rec->r_refcount),
+		     (unsigned long long)ref_leaf_bh->b_blocknr, index);
+
+		if (merge)
+			ocfs2_refcount_rec_merge(rb, index);
+	}
+
+	ret = ocfs2_journal_dirty(handle, ref_leaf_bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int __ocfs2_increase_refcount(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     u64 cpos, u32 len, int merge,
+				     struct ocfs2_alloc_context *meta_ac,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0, index;
+	struct buffer_head *ref_leaf_bh = NULL;
+	struct ocfs2_refcount_rec rec;
+	unsigned int set_len = 0;
+
+	mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+	     (unsigned long long)cpos, len);
+
+	while (len) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &rec, &index,
+					     &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		set_len = le32_to_cpu(rec.r_clusters);
+
+		/*
+		 * Here we may meet with 3 situations:
+		 *
+		 * 1. If we find an already existing record, and the length
+		 *    is the same, cool, we just need to increase the r_refcount
+		 *    and it is OK.
+		 * 2. If we find a hole, just insert it with r_refcount = 1.
+		 * 3. If we are in the middle of one extent record, split
+		 *    it.
+		 */
+		if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
+		    set_len <= len) {
+			mlog(0, "increase refcount rec, start %llu, len %u, "
+			     "count %u\n", (unsigned long long)cpos, set_len,
+			     le32_to_cpu(rec.r_refcount));
+			ret = ocfs2_change_refcount_rec(handle, ci,
+							ref_leaf_bh, index,
+							merge, 1);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else if (!rec.r_refcount) {
+			rec.r_refcount = cpu_to_le32(1);
+
+			mlog(0, "insert refcount rec, start %llu, len %u\n",
+			     (unsigned long long)le64_to_cpu(rec.r_cpos),
+			     set_len);
+			ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
+							ref_leaf_bh,
+							&rec, index,
+							merge, meta_ac);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else  {
+			set_len = min((u64)(cpos + len),
+				      le64_to_cpu(rec.r_cpos) + set_len) - cpos;
+			rec.r_cpos = cpu_to_le64(cpos);
+			rec.r_clusters = cpu_to_le32(set_len);
+			le32_add_cpu(&rec.r_refcount, 1);
+
+			mlog(0, "split refcount rec, start %llu, "
+			     "len %u, count %u\n",
+			     (unsigned long long)le64_to_cpu(rec.r_cpos),
+			     set_len, le32_to_cpu(rec.r_refcount));
+			ret = ocfs2_split_refcount_rec(handle, ci,
+						       ref_root_bh, ref_leaf_bh,
+						       &rec, index, merge,
+						       meta_ac, dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += set_len;
+		len -= set_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+static int ocfs2_remove_refcount_extent(handle_t *handle,
+				struct ocfs2_caching_info *ci,
+				struct buffer_head *ref_root_bh,
+				struct buffer_head *ref_leaf_bh,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_extent_tree et;
+
+	BUG_ON(rb->rf_records.rl_used);
+
+	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+	ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
+				  1, meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_remove_from_cache(ci, ref_leaf_bh);
+
+	/*
+	 * add the freed block to the dealloc so that it will be freed
+	 * when we run dealloc.
+	 */
+	ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
+					le16_to_cpu(rb->rf_suballoc_slot),
+					le64_to_cpu(rb->rf_blkno),
+					le16_to_cpu(rb->rf_suballoc_bit));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	le32_add_cpu(&rb->rf_clusters, -1);
+
+	/*
+	 * check whether we need to restore the root refcount block if
+	 * there is no leaf extent block at atll.
+	 */
+	if (!rb->rf_list.l_next_free_rec) {
+		BUG_ON(rb->rf_clusters);
+
+		mlog(0, "reset refcount tree root %llu to be a record block.\n",
+		     (unsigned long long)ref_root_bh->b_blocknr);
+
+		rb->rf_flags = 0;
+		rb->rf_parent = 0;
+		rb->rf_cpos = 0;
+		memset(&rb->rf_records, 0, sb->s_blocksize -
+		       offsetof(struct ocfs2_refcount_block, rf_records));
+		rb->rf_records.rl_count =
+				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
+	}
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+out:
+	return ret;
+}
+
+int ocfs2_increase_refcount(handle_t *handle,
+			    struct ocfs2_caching_info *ci,
+			    struct buffer_head *ref_root_bh,
+			    u64 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
+					 cpos, len, 1,
+					 meta_ac, dealloc);
+}
+
+static int ocfs2_decrease_refcount_rec(handle_t *handle,
+				struct ocfs2_caching_info *ci,
+				struct buffer_head *ref_root_bh,
+				struct buffer_head *ref_leaf_bh,
+				int index, u64 cpos, unsigned int len,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
+
+	BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
+	BUG_ON(cpos + len >
+	       le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
+
+	if (cpos == le64_to_cpu(rec->r_cpos) &&
+	    len == le32_to_cpu(rec->r_clusters))
+		ret = ocfs2_change_refcount_rec(handle, ci,
+						ref_leaf_bh, index, 1, -1);
+	else {
+		struct ocfs2_refcount_rec split = *rec;
+		split.r_cpos = cpu_to_le64(cpos);
+		split.r_clusters = cpu_to_le32(len);
+
+		le32_add_cpu(&split.r_refcount, -1);
+
+		mlog(0, "split refcount rec, start %llu, "
+		     "len %u, count %u, original start %llu, len %u\n",
+		     (unsigned long long)le64_to_cpu(split.r_cpos),
+		     len, le32_to_cpu(split.r_refcount),
+		     (unsigned long long)le64_to_cpu(rec->r_cpos),
+		     le32_to_cpu(rec->r_clusters));
+		ret = ocfs2_split_refcount_rec(handle, ci,
+					       ref_root_bh, ref_leaf_bh,
+					       &split, index, 1,
+					       meta_ac, dealloc);
+	}
+
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Remove the leaf refcount block if it contains no refcount record. */
+	if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
+		ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
+						   ref_leaf_bh, meta_ac,
+						   dealloc);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	return ret;
+}
+
+static int __ocfs2_decrease_refcount(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     u64 cpos, u32 len,
+				     struct ocfs2_alloc_context *meta_ac,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     int delete)
+{
+	int ret = 0, index = 0;
+	struct ocfs2_refcount_rec rec;
+	unsigned int r_count = 0, r_len;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct buffer_head *ref_leaf_bh = NULL;
+
+	mlog(0, "Tree owner %llu, decrease refcount start %llu, "
+	     "len %u, delete %u\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+	     (unsigned long long)cpos, len, delete);
+
+	while (len) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &rec, &index,
+					     &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		r_count = le32_to_cpu(rec.r_refcount);
+		BUG_ON(r_count == 0);
+		if (!delete)
+			BUG_ON(r_count > 1);
+
+		r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
+			      le32_to_cpu(rec.r_clusters)) - cpos;
+
+		ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
+						  ref_leaf_bh, index,
+						  cpos, r_len,
+						  meta_ac, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
+			ret = ocfs2_cache_cluster_dealloc(dealloc,
+					  ocfs2_clusters_to_blocks(sb, cpos),
+							  r_len);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += r_len;
+		len -= r_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+/* Caller must hold refcount tree lock. */
+int ocfs2_decrease_refcount(struct inode *inode,
+			    handle_t *handle, u32 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    int delete)
+{
+	int ret;
+	u64 ref_blkno;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *tree;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_get_refcount_block(inode, &ref_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
+					cpos, len, meta_ac, dealloc, delete);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+/*
+ * Mark the already-existing extent at cpos as refcounted for len clusters.
+ * This adds the refcount extent flag.
+ *
+ * If the existing extent is larger than the request, initiate a
+ * split. An attempt will be made at merging with adjacent extents.
+ *
+ * The caller is responsible for passing down meta_ac if we'll need it.
+ */
+static int ocfs2_mark_extent_refcounted(struct inode *inode,
+				struct ocfs2_extent_tree *et,
+				handle_t *handle, u32 cpos,
+				u32 len, u32 phys,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+
+	mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
+	     inode->i_ino, cpos, len, phys);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		ret = -EROFS;
+		goto out;
+	}
+
+	ret = ocfs2_change_extent_flag(handle, et, cpos,
+				       len, phys, meta_ac, dealloc,
+				       OCFS2_EXT_REFCOUNTED, 0);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+/*
+ * Given some contiguous physical clusters, calculate what we need
+ * for modifying their refcount.
+ */
+static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
+					    struct ocfs2_caching_info *ci,
+					    struct buffer_head *ref_root_bh,
+					    u64 start_cpos,
+					    u32 clusters,
+					    int *meta_add,
+					    int *credits)
+{
+	int ret = 0, index, ref_blocks = 0, recs_add = 0;
+	u64 cpos = start_cpos;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_rec rec;
+	struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
+	u32 len;
+
+	mlog(0, "start_cpos %llu, clusters %u\n",
+	     (unsigned long long)start_cpos, clusters);
+	while (clusters) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, clusters, &rec,
+					     &index, &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (ref_leaf_bh != prev_bh) {
+			/*
+			 * Now we encounter a new leaf block, so calculate
+			 * whether we need to extend the old leaf.
+			 */
+			if (prev_bh) {
+				rb = (struct ocfs2_refcount_block *)
+							prev_bh->b_data;
+
+				if (le64_to_cpu(rb->rf_records.rl_used) +
+				    recs_add >
+				    le16_to_cpu(rb->rf_records.rl_count))
+					ref_blocks++;
+			}
+
+			recs_add = 0;
+			*credits += 1;
+			brelse(prev_bh);
+			prev_bh = ref_leaf_bh;
+			get_bh(prev_bh);
+		}
+
+		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+
+		mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
+		     "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
+		     recs_add, (unsigned long long)cpos, clusters,
+		     (unsigned long long)le64_to_cpu(rec.r_cpos),
+		     le32_to_cpu(rec.r_clusters),
+		     le32_to_cpu(rec.r_refcount), index);
+
+		len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
+			  le32_to_cpu(rec.r_clusters)) - cpos;
+		/*
+		 * If the refcount rec already exist, cool. We just need
+		 * to check whether there is a split. Otherwise we just need
+		 * to increase the refcount.
+		 * If we will insert one, increases recs_add.
+		 *
+		 * We record all the records which will be inserted to the
+		 * same refcount block, so that we can tell exactly whether
+		 * we need a new refcount block or not.
+		 */
+		if (rec.r_refcount) {
+			/* Check whether we need a split at the beginning. */
+			if (cpos == start_cpos &&
+			    cpos != le64_to_cpu(rec.r_cpos))
+				recs_add++;
+
+			/* Check whether we need a split in the end. */
+			if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
+			    le32_to_cpu(rec.r_clusters))
+				recs_add++;
+		} else
+			recs_add++;
+
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+		clusters -= len;
+		cpos += len;
+	}
+
+	if (prev_bh) {
+		rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
+
+		if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
+		    le16_to_cpu(rb->rf_records.rl_count))
+			ref_blocks++;
+
+		*credits += 1;
+	}
+
+	if (!ref_blocks)
+		goto out;
+
+	mlog(0, "we need ref_blocks %d\n", ref_blocks);
+	*meta_add += ref_blocks;
+	*credits += ref_blocks;
+
+	/*
+	 * So we may need ref_blocks to insert into the tree.
+	 * That also means we need to change the b-tree and add that number
+	 * of records since we never merge them.
+	 * We need one more block for expansion since the new created leaf
+	 * block is also full and needs split.
+	 */
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
+		struct ocfs2_extent_tree et;
+
+		ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+		*meta_add += ocfs2_extend_meta_needed(et.et_root_el);
+		*credits += ocfs2_calc_extend_credits(sb,
+						      et.et_root_el,
+						      ref_blocks);
+	} else {
+		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+		*meta_add += 1;
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	brelse(prev_bh);
+	return ret;
+}
+
+/*
+ * For refcount tree, we will decrease some contiguous clusters
+ * refcount count, so just go through it to see how many blocks
+ * we gonna touch and whether we need to create new blocks.
+ *
+ * Normally the refcount blocks store these refcount should be
+ * continguous also, so that we can get the number easily.
+ * As for meta_ac, we will at most add split 2 refcount record and
+ * 2 more refcount block, so just check it in a rough way.
+ *
+ * Caller must hold refcount tree lock.
+ */
+int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
+					  struct buffer_head *di_bh,
+					  u64 phys_blkno,
+					  u32 clusters,
+					  int *credits,
+					  struct ocfs2_alloc_context **meta_ac)
+{
+	int ret, ref_blocks = 0;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *tree;
+	u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		ret = -EROFS;
+		goto out;
+	}
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
+				      le64_to_cpu(di->i_refcount_loc), &tree);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci,
+					le64_to_cpu(di->i_refcount_loc),
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
+					       &tree->rf_ci,
+					       ref_root_bh,
+					       start_cpos, clusters,
+					       &ref_blocks, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "reserve new metadata %d, credits = %d\n",
+	     ref_blocks, *credits);
+
+	if (ref_blocks) {
+		ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
+							ref_blocks, meta_ac);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+#define	MAX_CONTIG_BYTES	1048576
+
+static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
+{
+	return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
+}
+
+static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
+{
+	return ~(ocfs2_cow_contig_clusters(sb) - 1);
+}
+
+/*
+ * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
+ * find an offset (start + (n * contig_clusters)) that is closest to cpos
+ * while still being less than or equal to it.
+ *
+ * The goal is to break the extent at a multiple of contig_clusters.
+ */
+static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
+						 unsigned int start,
+						 unsigned int cpos)
+{
+	BUG_ON(start > cpos);
+
+	return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
+}
+
+/*
+ * Given a cluster count of len, pad it out so that it is a multiple
+ * of contig_clusters.
+ */
+static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
+						  unsigned int len)
+{
+	unsigned int padded =
+		(len + (ocfs2_cow_contig_clusters(sb) - 1)) &
+		ocfs2_cow_contig_mask(sb);
+
+	/* Did we wrap? */
+	if (padded < len)
+		padded = UINT_MAX;
+
+	return padded;
+}
+
+/*
+ * Calculate out the start and number of virtual clusters we need to to CoW.
+ *
+ * cpos is vitual start cluster position we want to do CoW in a
+ * file and write_len is the cluster length.
+ * max_cpos is the place where we want to stop CoW intentionally.
+ *
+ * Normal we will start CoW from the beginning of extent record cotaining cpos.
+ * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
+ * get good I/O from the resulting extent tree.
+ */
+static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
+					   struct ocfs2_extent_list *el,
+					   u32 cpos,
+					   u32 write_len,
+					   u32 max_cpos,
+					   u32 *cow_start,
+					   u32 *cow_len)
+{
+	int ret = 0;
+	int tree_height = le16_to_cpu(el->l_tree_depth), i;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb = NULL;
+	struct ocfs2_extent_rec *rec;
+	unsigned int want_clusters, rec_end = 0;
+	int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
+	int leaf_clusters;
+
+	BUG_ON(cpos + write_len > max_cpos);
+
+	if (tree_height > 0) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "leaf block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	*cow_len = 0;
+	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+
+		if (ocfs2_is_empty_extent(rec)) {
+			mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
+					"index %d\n", inode->i_ino, i);
+			continue;
+		}
+
+		if (le32_to_cpu(rec->e_cpos) +
+		    le16_to_cpu(rec->e_leaf_clusters) <= cpos)
+			continue;
+
+		if (*cow_len == 0) {
+			/*
+			 * We should find a refcounted record in the
+			 * first pass.
+			 */
+			BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
+			*cow_start = le32_to_cpu(rec->e_cpos);
+		}
+
+		/*
+		 * If we encounter a hole, a non-refcounted record or
+		 * pass the max_cpos, stop the search.
+		 */
+		if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
+		    (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
+		    (max_cpos <= le32_to_cpu(rec->e_cpos)))
+			break;
+
+		leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
+		rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
+		if (rec_end > max_cpos) {
+			rec_end = max_cpos;
+			leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
+		}
+
+		/*
+		 * How many clusters do we actually need from
+		 * this extent?  First we see how many we actually
+		 * need to complete the write.  If that's smaller
+		 * than contig_clusters, we try for contig_clusters.
+		 */
+		if (!*cow_len)
+			want_clusters = write_len;
+		else
+			want_clusters = (cpos + write_len) -
+				(*cow_start + *cow_len);
+		if (want_clusters < contig_clusters)
+			want_clusters = contig_clusters;
+
+		/*
+		 * If the write does not cover the whole extent, we
+		 * need to calculate how we're going to split the extent.
+		 * We try to do it on contig_clusters boundaries.
+		 *
+		 * Any extent smaller than contig_clusters will be
+		 * CoWed in its entirety.
+		 */
+		if (leaf_clusters <= contig_clusters)
+			*cow_len += leaf_clusters;
+		else if (*cow_len || (*cow_start == cpos)) {
+			/*
+			 * This extent needs to be CoW'd from its
+			 * beginning, so all we have to do is compute
+			 * how many clusters to grab.  We align
+			 * want_clusters to the edge of contig_clusters
+			 * to get better I/O.
+			 */
+			want_clusters = ocfs2_cow_align_length(inode->i_sb,
+							       want_clusters);
+
+			if (leaf_clusters < want_clusters)
+				*cow_len += leaf_clusters;
+			else
+				*cow_len += want_clusters;
+		} else if ((*cow_start + contig_clusters) >=
+			   (cpos + write_len)) {
+			/*
+			 * Breaking off contig_clusters at the front
+			 * of the extent will cover our write.  That's
+			 * easy.
+			 */
+			*cow_len = contig_clusters;
+		} else if ((rec_end - cpos) <= contig_clusters) {
+			/*
+			 * Breaking off contig_clusters at the tail of
+			 * this extent will cover cpos.
+			 */
+			*cow_start = rec_end - contig_clusters;
+			*cow_len = contig_clusters;
+		} else if ((rec_end - cpos) <= want_clusters) {
+			/*
+			 * While we can't fit the entire write in this
+			 * extent, we know that the write goes from cpos
+			 * to the end of the extent.  Break that off.
+			 * We try to break it at some multiple of
+			 * contig_clusters from the front of the extent.
+			 * Failing that (ie, cpos is within
+			 * contig_clusters of the front), we'll CoW the
+			 * entire extent.
+			 */
+			*cow_start = ocfs2_cow_align_start(inode->i_sb,
+							   *cow_start, cpos);
+			*cow_len = rec_end - *cow_start;
+		} else {
+			/*
+			 * Ok, the entire write lives in the middle of
+			 * this extent.  Let's try to slice the extent up
+			 * nicely.  Optimally, our CoW region starts at
+			 * m*contig_clusters from the beginning of the
+			 * extent and goes for n*contig_clusters,
+			 * covering the entire write.
+			 */
+			*cow_start = ocfs2_cow_align_start(inode->i_sb,
+							   *cow_start, cpos);
+
+			want_clusters = (cpos + write_len) - *cow_start;
+			want_clusters = ocfs2_cow_align_length(inode->i_sb,
+							       want_clusters);
+			if (*cow_start + want_clusters <= rec_end)
+				*cow_len = want_clusters;
+			else
+				*cow_len = rec_end - *cow_start;
+		}
+
+		/* Have we covered our entire write yet? */
+		if ((*cow_start + *cow_len) >= (cpos + write_len))
+			break;
+
+		/*
+		 * If we reach the end of the extent block and don't get enough
+		 * clusters, continue with the next extent block if possible.
+		 */
+		if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
+		    eb && eb->h_next_leaf_blk) {
+			brelse(eb_bh);
+			eb_bh = NULL;
+
+			ret = ocfs2_read_extent_block(INODE_CACHE(inode),
+					       le64_to_cpu(eb->h_next_leaf_blk),
+					       &eb_bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+			el = &eb->h_list;
+			i = -1;
+		}
+	}
+
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+/*
+ * Prepare meta_ac, data_ac and calculate credits when we want to add some
+ * num_clusters in data_tree "et" and change the refcount for the old
+ * clusters(starting form p_cluster) in the refcount tree.
+ *
+ * Note:
+ * 1. since we may split the old tree, so we at most will need num_clusters + 2
+ *    more new leaf records.
+ * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
+ *    just give data_ac = NULL.
+ */
+static int ocfs2_lock_refcount_allocators(struct super_block *sb,
+					u32 p_cluster, u32 num_clusters,
+					struct ocfs2_extent_tree *et,
+					struct ocfs2_caching_info *ref_ci,
+					struct buffer_head *ref_root_bh,
+					struct ocfs2_alloc_context **meta_ac,
+					struct ocfs2_alloc_context **data_ac,
+					int *credits)
+{
+	int ret = 0, meta_add = 0;
+	int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
+
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (num_free_extents < num_clusters + 2)
+		meta_add =
+			ocfs2_extend_meta_needed(et->et_root_el);
+
+	*credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
+					      num_clusters + 2);
+
+	ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
+					       p_cluster, num_clusters,
+					       &meta_add, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
+	     meta_add, num_clusters, *credits);
+	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
+						meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (data_ac) {
+		ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
+					     data_ac);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+	}
+
+	return ret;
+}
+
+static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
+{
+	BUG_ON(buffer_dirty(bh));
+
+	clear_buffer_mapped(bh);
+
+	return 0;
+}
+
+static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
+					    struct ocfs2_cow_context *context,
+					    u32 cpos, u32 old_cluster,
+					    u32 new_cluster, u32 new_len)
+{
+	int ret = 0, partial;
+	struct ocfs2_caching_info *ci = context->data_et.et_ci;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
+	struct page *page;
+	pgoff_t page_index;
+	unsigned int from, to;
+	loff_t offset, end, map_end;
+	struct address_space *mapping = context->inode->i_mapping;
+
+	mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
+	     new_cluster, new_len, cpos);
+
+	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
+	end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
+
+	while (offset < end) {
+		page_index = offset >> PAGE_CACHE_SHIFT;
+		map_end = (page_index + 1) << PAGE_CACHE_SHIFT;
+		if (map_end > end)
+			map_end = end;
+
+		/* from, to is the offset within the page. */
+		from = offset & (PAGE_CACHE_SIZE - 1);
+		to = PAGE_CACHE_SIZE;
+		if (map_end & (PAGE_CACHE_SIZE - 1))
+			to = map_end & (PAGE_CACHE_SIZE - 1);
+
+		page = grab_cache_page(mapping, page_index);
+
+		/* This page can't be dirtied before we CoW it out. */
+		BUG_ON(PageDirty(page));
+
+		if (!PageUptodate(page)) {
+			ret = block_read_full_page(page, ocfs2_get_block);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+			lock_page(page);
+		}
+
+		if (page_has_buffers(page)) {
+			ret = walk_page_buffers(handle, page_buffers(page),
+						from, to, &partial,
+						ocfs2_clear_cow_buffer);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+		}
+
+		ocfs2_map_and_dirty_page(context->inode,
+					 handle, from, to,
+					 page, 0, &new_block);
+		mark_page_accessed(page);
+unlock:
+		unlock_page(page);
+		page_cache_release(page);
+		page = NULL;
+		offset = map_end;
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
+					   struct ocfs2_cow_context *context,
+					   u32 cpos, u32 old_cluster,
+					   u32 new_cluster, u32 new_len)
+{
+	int ret = 0;
+	struct super_block *sb = context->inode->i_sb;
+	struct ocfs2_caching_info *ci = context->data_et.et_ci;
+	int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
+	u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
+	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	struct buffer_head *old_bh = NULL;
+	struct buffer_head *new_bh = NULL;
+
+	mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
+	     new_cluster, new_len);
+
+	for (i = 0; i < blocks; i++, old_block++, new_block++) {
+		new_bh = sb_getblk(osb->sb, new_block);
+		if (new_bh == NULL) {
+			ret = -EIO;
+			mlog_errno(ret);
+			break;
+		}
+
+		ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+		ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_journal_access(handle, ci, new_bh,
+					   OCFS2_JOURNAL_ACCESS_CREATE);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
+		ret = ocfs2_journal_dirty(handle, new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		brelse(new_bh);
+		brelse(old_bh);
+		new_bh = NULL;
+		old_bh = NULL;
+	}
+
+	brelse(new_bh);
+	brelse(old_bh);
+	return ret;
+}
+
+static int ocfs2_clear_ext_refcount(handle_t *handle,
+				    struct ocfs2_extent_tree *et,
+				    u32 cpos, u32 p_cluster, u32 len,
+				    unsigned int ext_flags,
+				    struct ocfs2_alloc_context *meta_ac,
+				    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, index;
+	struct ocfs2_extent_rec replace_rec;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_list *el;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+	u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
+
+	mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
+	     (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
+
+	memset(&replace_rec, 0, sizeof(replace_rec));
+	replace_rec.e_cpos = cpu_to_le32(cpos);
+	replace_rec.e_leaf_clusters = cpu_to_le16(len);
+	replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
+								   p_cluster));
+	replace_rec.e_flags = ext_flags;
+	replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
+
+	path = ocfs2_new_path_from_et(et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
+		ocfs2_error(sb,
+			    "Inode %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)ino, cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	ret = ocfs2_split_extent(handle, et, path, index,
+				 &replace_rec, meta_ac, dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+static int ocfs2_replace_clusters(handle_t *handle,
+				  struct ocfs2_cow_context *context,
+				  u32 cpos, u32 old,
+				  u32 new, u32 len,
+				  unsigned int ext_flags)
+{
+	int ret;
+	struct ocfs2_caching_info *ci = context->data_et.et_ci;
+	u64 ino = ocfs2_metadata_cache_owner(ci);
+
+	mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
+	     (unsigned long long)ino, cpos, old, new, len, ext_flags);
+
+	/*If the old clusters is unwritten, no need to duplicate. */
+	if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
+		ret = context->cow_duplicate_clusters(handle, context, cpos,
+						      old, new, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
+				       cpos, new, len, ext_flags,
+				       context->meta_ac, &context->dealloc);
+	if (ret)
+		mlog_errno(ret);
+out:
+	return ret;
+}
+
+static int ocfs2_cow_sync_writeback(struct super_block *sb,
+				    struct ocfs2_cow_context *context,
+				    u32 cpos, u32 num_clusters)
+{
+	int ret = 0;
+	loff_t offset, end, map_end;
+	pgoff_t page_index;
+	struct page *page;
+
+	if (ocfs2_should_order_data(context->inode))
+		return 0;
+
+	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
+	end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
+
+	ret = filemap_fdatawrite_range(context->inode->i_mapping,
+				       offset, end - 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	while (offset < end) {
+		page_index = offset >> PAGE_CACHE_SHIFT;
+		map_end = (page_index + 1) << PAGE_CACHE_SHIFT;
+		if (map_end > end)
+			map_end = end;
+
+		page = grab_cache_page(context->inode->i_mapping, page_index);
+		BUG_ON(!page);
+
+		wait_on_page_writeback(page);
+		if (PageError(page)) {
+			ret = -EIO;
+			mlog_errno(ret);
+		} else
+			mark_page_accessed(page);
+
+		unlock_page(page);
+		page_cache_release(page);
+		page = NULL;
+		offset = map_end;
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
+				 u32 v_cluster, u32 *p_cluster,
+				 u32 *num_clusters,
+				 unsigned int *extent_flags)
+{
+	return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
+				  num_clusters, extent_flags);
+}
+
+static int ocfs2_make_clusters_writable(struct super_block *sb,
+					struct ocfs2_cow_context *context,
+					u32 cpos, u32 p_cluster,
+					u32 num_clusters, unsigned int e_flags)
+{
+	int ret, delete, index, credits =  0;
+	u32 new_bit, new_len;
+	unsigned int set_len;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	handle_t *handle;
+	struct buffer_head *ref_leaf_bh = NULL;
+	struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
+	struct ocfs2_refcount_rec rec;
+
+	mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
+	     cpos, p_cluster, num_clusters, e_flags);
+
+	ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
+					     &context->data_et,
+					     ref_ci,
+					     context->ref_root_bh,
+					     &context->meta_ac,
+					     &context->data_ac, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	if (context->post_refcount)
+		credits += context->post_refcount->credits;
+
+	credits += context->extra_credits;
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	while (num_clusters) {
+		ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
+					     p_cluster, num_clusters,
+					     &rec, &index, &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		BUG_ON(!rec.r_refcount);
+		set_len = min((u64)p_cluster + num_clusters,
+			      le64_to_cpu(rec.r_cpos) +
+			      le32_to_cpu(rec.r_clusters)) - p_cluster;
+
+		/*
+		 * There are many different situation here.
+		 * 1. If refcount == 1, remove the flag and don't COW.
+		 * 2. If refcount > 1, allocate clusters.
+		 *    Here we may not allocate r_len once at a time, so continue
+		 *    until we reach num_clusters.
+		 */
+		if (le32_to_cpu(rec.r_refcount) == 1) {
+			delete = 0;
+			ret = ocfs2_clear_ext_refcount(handle,
+						       &context->data_et,
+						       cpos, p_cluster,
+						       set_len, e_flags,
+						       context->meta_ac,
+						       &context->dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+		} else {
+			delete = 1;
+
+			ret = __ocfs2_claim_clusters(osb, handle,
+						     context->data_ac,
+						     1, set_len,
+						     &new_bit, &new_len);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+
+			ret = ocfs2_replace_clusters(handle, context,
+						     cpos, p_cluster, new_bit,
+						     new_len, e_flags);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+			set_len = new_len;
+		}
+
+		ret = __ocfs2_decrease_refcount(handle, ref_ci,
+						context->ref_root_bh,
+						p_cluster, set_len,
+						context->meta_ac,
+						&context->dealloc, delete);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		cpos += set_len;
+		p_cluster += set_len;
+		num_clusters -= set_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+	/* handle any post_cow action. */
+	if (context->post_refcount && context->post_refcount->func) {
+		ret = context->post_refcount->func(context->inode, handle,
+						context->post_refcount->para);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+	}
+
+	/*
+	 * Here we should write the new page out first if we are
+	 * in write-back mode.
+	 */
+	if (context->get_clusters == ocfs2_di_get_clusters) {
+		ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (context->data_ac) {
+		ocfs2_free_alloc_context(context->data_ac);
+		context->data_ac = NULL;
+	}
+	if (context->meta_ac) {
+		ocfs2_free_alloc_context(context->meta_ac);
+		context->meta_ac = NULL;
+	}
+	brelse(ref_leaf_bh);
+
+	return ret;
+}
+
+static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
+{
+	int ret = 0;
+	struct inode *inode = context->inode;
+	u32 cow_start = context->cow_start, cow_len = context->cow_len;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		return -EROFS;
+	}
+
+	ocfs2_init_dealloc_ctxt(&context->dealloc);
+
+	while (cow_len) {
+		ret = context->get_clusters(context, cow_start, &p_cluster,
+					    &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
+
+		if (cow_len < num_clusters)
+			num_clusters = cow_len;
+
+		ret = ocfs2_make_clusters_writable(inode->i_sb, context,
+						   cow_start, p_cluster,
+						   num_clusters, ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		cow_len -= num_clusters;
+		cow_start += num_clusters;
+	}
+
+	if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &context->dealloc);
+	}
+
+	return ret;
+}
+
+/*
+ * Starting at cpos, try to CoW write_len clusters.  Don't CoW
+ * past max_cpos.  This will stop when it runs into a hole or an
+ * unrefcounted extent.
+ */
+static int ocfs2_refcount_cow_hunk(struct inode *inode,
+				   struct buffer_head *di_bh,
+				   u32 cpos, u32 write_len, u32 max_cpos)
+{
+	int ret;
+	u32 cow_start = 0, cow_len = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *ref_tree;
+	struct ocfs2_cow_context *context = NULL;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
+					      cpos, write_len, max_cpos,
+					      &cow_start, &cow_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
+	     "cow_len %u\n", inode->i_ino,
+	     cpos, write_len, cow_start, cow_len);
+
+	BUG_ON(cow_len == 0);
+
+	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
+	if (!context) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
+				       1, &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	context->inode = inode;
+	context->cow_start = cow_start;
+	context->cow_len = cow_len;
+	context->ref_tree = ref_tree;
+	context->ref_root_bh = ref_root_bh;
+	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
+	context->get_clusters = ocfs2_di_get_clusters;
+
+	ocfs2_init_dinode_extent_tree(&context->data_et,
+				      INODE_CACHE(inode), di_bh);
+
+	ret = ocfs2_replace_cow(context);
+	if (ret)
+		mlog_errno(ret);
+
+	/*
+	 * truncate the extent map here since no matter whether we meet with
+	 * any error during the action, we shouldn't trust cached extent map
+	 * any more.
+	 */
+	ocfs2_extent_map_trunc(inode, cow_start);
+
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+out:
+	kfree(context);
+	return ret;
+}
+
+/*
+ * CoW any and all clusters between cpos and cpos+write_len.
+ * Don't CoW past max_cpos.  If this returns successfully, all
+ * clusters between cpos and cpos+write_len are safe to modify.
+ */
+int ocfs2_refcount_cow(struct inode *inode,
+		       struct buffer_head *di_bh,
+		       u32 cpos, u32 write_len, u32 max_cpos)
+{
+	int ret = 0;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+
+	while (write_len) {
+		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		if (write_len < num_clusters)
+			num_clusters = write_len;
+
+		if (ext_flags & OCFS2_EXT_REFCOUNTED) {
+			ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
+						      num_clusters, max_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				break;
+			}
+		}
+
+		write_len -= num_clusters;
+		cpos += num_clusters;
+	}
+
+	return ret;
+}
+
+static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
+					  u32 v_cluster, u32 *p_cluster,
+					  u32 *num_clusters,
+					  unsigned int *extent_flags)
+{
+	struct inode *inode = context->inode;
+	struct ocfs2_xattr_value_root *xv = context->cow_object;
+
+	return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
+					num_clusters, &xv->xr_list,
+					extent_flags);
+}
+
+/*
+ * Given a xattr value root, calculate the most meta/credits we need for
+ * refcount tree change if we truncate it to 0.
+ */
+int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
+				       struct ocfs2_caching_info *ref_ci,
+				       struct buffer_head *ref_root_bh,
+				       struct ocfs2_xattr_value_root *xv,
+				       int *meta_add, int *credits)
+{
+	int ret = 0, index, ref_blocks = 0;
+	u32 p_cluster, num_clusters;
+	u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_rec rec;
+	struct buffer_head *ref_leaf_bh = NULL;
+
+	while (cpos < clusters) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
+					       &num_clusters, &xv->xr_list,
+					       NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		cpos += num_clusters;
+
+		while (num_clusters) {
+			ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
+						     p_cluster, num_clusters,
+						     &rec, &index,
+						     &ref_leaf_bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			BUG_ON(!rec.r_refcount);
+
+			rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+
+			/*
+			 * We really don't know whether the other clusters is in
+			 * this refcount block or not, so just take the worst
+			 * case that all the clusters are in this block and each
+			 * one will split a refcount rec, so totally we need
+			 * clusters * 2 new refcount rec.
+			 */
+			if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
+			    le16_to_cpu(rb->rf_records.rl_count))
+				ref_blocks++;
+
+			*credits += 1;
+			brelse(ref_leaf_bh);
+			ref_leaf_bh = NULL;
+
+			if (num_clusters <= le32_to_cpu(rec.r_clusters))
+				break;
+			else
+				num_clusters -= le32_to_cpu(rec.r_clusters);
+			p_cluster += num_clusters;
+		}
+	}
+
+	*meta_add += ref_blocks;
+	if (!ref_blocks)
+		goto out;
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
+		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+	else {
+		struct ocfs2_extent_tree et;
+
+		ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
+		*credits += ocfs2_calc_extend_credits(inode->i_sb,
+						      et.et_root_el,
+						      ref_blocks);
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+/*
+ * Do CoW for xattr.
+ */
+int ocfs2_refcount_cow_xattr(struct inode *inode,
+			     struct ocfs2_dinode *di,
+			     struct ocfs2_xattr_value_buf *vb,
+			     struct ocfs2_refcount_tree *ref_tree,
+			     struct buffer_head *ref_root_bh,
+			     u32 cpos, u32 write_len,
+			     struct ocfs2_post_refcount *post)
+{
+	int ret;
+	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_cow_context *context = NULL;
+	u32 cow_start, cow_len;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
+					      cpos, write_len, UINT_MAX,
+					      &cow_start, &cow_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(cow_len == 0);
+
+	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
+	if (!context) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	context->inode = inode;
+	context->cow_start = cow_start;
+	context->cow_len = cow_len;
+	context->ref_tree = ref_tree;
+	context->ref_root_bh = ref_root_bh;;
+	context->cow_object = xv;
+
+	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
+	/* We need the extra credits for duplicate_clusters by jbd. */
+	context->extra_credits =
+		ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
+	context->get_clusters = ocfs2_xattr_value_get_clusters;
+	context->post_refcount = post;
+
+	ocfs2_init_xattr_value_extent_tree(&context->data_et,
+					   INODE_CACHE(inode), vb);
+
+	ret = ocfs2_replace_cow(context);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	kfree(context);
+	return ret;
+}
+
+/*
+ * Insert a new extent into refcount tree and mark a extent rec
+ * as refcounted in the dinode tree.
+ */
+int ocfs2_add_refcount_flag(struct inode *inode,
+			    struct ocfs2_extent_tree *data_et,
+			    struct ocfs2_caching_info *ref_ci,
+			    struct buffer_head *ref_root_bh,
+			    u32 cpos, u32 p_cluster, u32 num_clusters,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    struct ocfs2_post_refcount *post)
+{
+	int ret;
+	handle_t *handle;
+	int credits = 1, ref_blocks = 0;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+
+	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
+					       ref_ci, ref_root_bh,
+					       p_cluster, num_clusters,
+					       &ref_blocks, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mlog(0, "reserve new metadata %d, credits = %d\n",
+	     ref_blocks, credits);
+
+	if (ref_blocks) {
+		ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
+							ref_blocks, &meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (post)
+		credits += post->credits;
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
+					   cpos, num_clusters, p_cluster,
+					   meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
+					p_cluster, num_clusters, 0,
+					meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	if (post && post->func) {
+		ret = post->func(inode, handle, post->para);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+static int ocfs2_change_ctime(struct inode *inode,
+			      struct buffer_head *di_bh)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
+				   OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	inode->i_ctime = CURRENT_TIME;
+	di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out:
+	return ret;
+}
+
+static int ocfs2_attach_refcount_tree(struct inode *inode,
+				      struct buffer_head *di_bh)
+{
+	int ret, data_changed = 0;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_refcount_tree *ref_tree;
+	unsigned int ext_flags;
+	loff_t size;
+	u32 cpos, num_clusters, clusters, p_cluster;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_extent_tree di_et;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
+		ret = ocfs2_create_refcount_tree(inode, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	BUG_ON(!di->i_refcount_loc);
+	ret = ocfs2_lock_refcount_tree(osb,
+				       le64_to_cpu(di->i_refcount_loc), 1,
+				       &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
+
+	size = i_size_read(inode);
+	clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+
+		if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
+			ret = ocfs2_add_refcount_flag(inode, &di_et,
+						      &ref_tree->rf_ci,
+						      ref_root_bh, cpos,
+						      p_cluster, num_clusters,
+						      &dealloc, NULL);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+
+			data_changed = 1;
+		}
+		cpos += num_clusters;
+	}
+
+	if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+		ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
+						       &ref_tree->rf_ci,
+						       ref_root_bh,
+						       &dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto unlock;
+		}
+	}
+
+	if (data_changed) {
+		ret = ocfs2_change_ctime(inode, di_bh);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+unlock:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+
+	if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &dealloc);
+	}
+out:
+	/*
+	 * Empty the extent map so that we may get the right extent
+	 * record from the disk.
+	 */
+	ocfs2_extent_map_trunc(inode, 0);
+
+	return ret;
+}
+
+static int ocfs2_add_refcounted_extent(struct inode *inode,
+				   struct ocfs2_extent_tree *et,
+				   struct ocfs2_caching_info *ref_ci,
+				   struct buffer_head *ref_root_bh,
+				   u32 cpos, u32 p_cluster, u32 num_clusters,
+				   unsigned int ext_flags,
+				   struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	handle_t *handle;
+	int credits = 0;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+
+	ret = ocfs2_lock_refcount_allocators(inode->i_sb,
+					     p_cluster, num_clusters,
+					     et, ref_ci,
+					     ref_root_bh, &meta_ac,
+					     NULL, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_insert_extent(handle, et, cpos,
+			cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb,
+							     p_cluster)),
+			num_clusters, ext_flags, meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
+				      p_cluster, num_clusters,
+				      meta_ac, dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+static int ocfs2_duplicate_extent_list(struct inode *s_inode,
+				struct inode *t_inode,
+				struct buffer_head *t_bh,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	u32 p_cluster, num_clusters, clusters, cpos;
+	loff_t size;
+	unsigned int ext_flags;
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
+
+	size = i_size_read(s_inode);
+	clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+
+		if (p_cluster) {
+			ret = ocfs2_add_refcounted_extent(t_inode, &et,
+							  ref_ci, ref_root_bh,
+							  cpos, p_cluster,
+							  num_clusters,
+							  ext_flags,
+							  dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += num_clusters;
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * change the new file's attributes to the src.
+ *
+ * reflink creates a snapshot of a file, that means the attributes
+ * must be identical except for three exceptions - nlink, ino, and ctime.
+ */
+static int ocfs2_complete_reflink(struct inode *s_inode,
+				  struct buffer_head *s_bh,
+				  struct inode *t_inode,
+				  struct buffer_head *t_bh,
+				  bool preserve)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
+	loff_t size = i_size_read(s_inode);
+
+	handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
+				   OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&OCFS2_I(t_inode)->ip_lock);
+	OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
+	OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
+	OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
+	spin_unlock(&OCFS2_I(t_inode)->ip_lock);
+	i_size_write(t_inode, size);
+
+	di->i_xattr_inline_size = s_di->i_xattr_inline_size;
+	di->i_clusters = s_di->i_clusters;
+	di->i_size = s_di->i_size;
+	di->i_dyn_features = s_di->i_dyn_features;
+	di->i_attr = s_di->i_attr;
+
+	if (preserve) {
+		di->i_uid = s_di->i_uid;
+		di->i_gid = s_di->i_gid;
+		di->i_mode = s_di->i_mode;
+
+		/*
+		 * update time.
+		 * we want mtime to appear identical to the source and
+		 * update ctime.
+		 */
+		t_inode->i_ctime = CURRENT_TIME;
+
+		di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
+		di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
+
+		t_inode->i_mtime = s_inode->i_mtime;
+		di->i_mtime = s_di->i_mtime;
+		di->i_mtime_nsec = s_di->i_mtime_nsec;
+	}
+
+	ocfs2_journal_dirty(handle, t_bh);
+
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
+	return ret;
+}
+
+static int ocfs2_create_reflink_node(struct inode *s_inode,
+				     struct buffer_head *s_bh,
+				     struct inode *t_inode,
+				     struct buffer_head *t_bh,
+				     bool preserve)
+{
+	int ret;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
+	struct ocfs2_refcount_tree *ref_tree;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	ret = ocfs2_set_refcount_tree(t_inode, t_bh,
+				      le64_to_cpu(di->i_refcount_loc));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
+				       1, &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
+					  &ref_tree->rf_ci, ref_root_bh,
+					  &dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_refcount;
+	}
+
+	ret = ocfs2_complete_reflink(s_inode, s_bh, t_inode, t_bh, preserve);
+	if (ret)
+		mlog_errno(ret);
+
+out_unlock_refcount:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+out:
+	if (ocfs2_dealloc_has_cluster(&dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &dealloc);
+	}
+
+	return ret;
+}
+
+static int __ocfs2_reflink(struct dentry *old_dentry,
+			   struct buffer_head *old_bh,
+			   struct inode *new_inode,
+			   bool preserve)
+{
+	int ret;
+	struct inode *inode = old_dentry->d_inode;
+	struct buffer_head *new_bh = NULL;
+
+	ret = filemap_fdatawrite(inode->i_mapping);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_attach_refcount_tree(inode, old_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&new_inode->i_mutex);
+	ret = ocfs2_inode_lock(new_inode, &new_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_create_reflink_node(inode, old_bh,
+					new_inode, new_bh, preserve);
+	if (ret) {
+		mlog_errno(ret);
+		goto inode_unlock;
+	}
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+		ret = ocfs2_reflink_xattrs(inode, old_bh,
+					   new_inode, new_bh,
+					   preserve);
+		if (ret)
+			mlog_errno(ret);
+	}
+inode_unlock:
+	ocfs2_inode_unlock(new_inode, 1);
+	brelse(new_bh);
+out_unlock:
+	mutex_unlock(&new_inode->i_mutex);
+out:
+	if (!ret) {
+		ret = filemap_fdatawait(inode->i_mapping);
+		if (ret)
+			mlog_errno(ret);
+	}
+	return ret;
+}
+
+static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
+			 struct dentry *new_dentry, bool preserve)
+{
+	int error;
+	struct inode *inode = old_dentry->d_inode;
+	struct buffer_head *old_bh = NULL;
+	struct inode *new_orphan_inode = NULL;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
+					     &new_orphan_inode);
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	error = ocfs2_inode_lock(inode, &old_bh, 1);
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	down_write(&OCFS2_I(inode)->ip_xattr_sem);
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+	error = __ocfs2_reflink(old_dentry, old_bh,
+				new_orphan_inode, preserve);
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	up_write(&OCFS2_I(inode)->ip_xattr_sem);
+
+	ocfs2_inode_unlock(inode, 1);
+	brelse(old_bh);
+
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	/* If the security isn't preserved, we need to re-initialize them. */
+	if (!preserve) {
+		error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
+		if (error)
+			mlog_errno(error);
+	}
+out:
+	if (!error) {
+		error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
+						       new_dentry);
+		if (error)
+			mlog_errno(error);
+	}
+
+	if (new_orphan_inode) {
+		/*
+		 * We need to open_unlock the inode no matter whether we
+		 * succeed or not, so that other nodes can delete it later.
+		 */
+		ocfs2_open_unlock(new_orphan_inode);
+		if (error)
+			iput(new_orphan_inode);
+	}
+
+	return error;
+}
+
+/*
+ * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
+ * sys_reflink().  This will go away when vfs_reflink() exists in
+ * fs/namei.c.
+ */
+
+/* copied from may_create in VFS. */
+static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
+{
+	if (child->d_inode)
+		return -EEXIST;
+	if (IS_DEADDIR(dir))
+		return -ENOENT;
+	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
+}
+
+/* copied from user_path_parent. */
+static int ocfs2_user_path_parent(const char __user *path,
+				  struct nameidata *nd, char **name)
+{
+	char *s = getname(path);
+	int error;
+
+	if (IS_ERR(s))
+		return PTR_ERR(s);
+
+	error = path_lookup(s, LOOKUP_PARENT, nd);
+	if (error)
+		putname(s);
+	else
+		*name = s;
+
+	return error;
+}
+
+/**
+ * ocfs2_vfs_reflink - Create a reference-counted link
+ *
+ * @old_dentry:        source dentry + inode
+ * @dir:       directory to create the target
+ * @new_dentry:        target dentry
+ * @preserve:  if true, preserve all file attributes
+ */
+int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
+		      struct dentry *new_dentry, bool preserve)
+{
+	struct inode *inode = old_dentry->d_inode;
+	int error;
+
+	if (!inode)
+		return -ENOENT;
+
+	error = ocfs2_may_create(dir, new_dentry);
+	if (error)
+		return error;
+
+	if (dir->i_sb != inode->i_sb)
+		return -EXDEV;
+
+	/*
+	 * A reflink to an append-only or immutable file cannot be created.
+	 */
+	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+		return -EPERM;
+
+	/* Only regular files can be reflinked. */
+	if (!S_ISREG(inode->i_mode))
+		return -EPERM;
+
+	/*
+	 * If the caller wants to preserve ownership, they require the
+	 * rights to do so.
+	 */
+	if (preserve) {
+		if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
+			return -EPERM;
+		if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
+			return -EPERM;
+	}
+
+	/*
+	 * If the caller is modifying any aspect of the attributes, they
+	 * are not creating a snapshot.  They need read permission on the
+	 * file.
+	 */
+	if (!preserve) {
+		error = inode_permission(inode, MAY_READ);
+		if (error)
+			return error;
+	}
+
+	mutex_lock(&inode->i_mutex);
+	vfs_dq_init(dir);
+	error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
+	mutex_unlock(&inode->i_mutex);
+	if (!error)
+		fsnotify_create(dir, new_dentry);
+	return error;
+}
+/*
+ * Most codes are copied from sys_linkat.
+ */
+int ocfs2_reflink_ioctl(struct inode *inode,
+			const char __user *oldname,
+			const char __user *newname,
+			bool preserve)
+{
+	struct dentry *new_dentry;
+	struct nameidata nd;
+	struct path old_path;
+	int error;
+	char *to = NULL;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
+	if (error) {
+		mlog_errno(error);
+		return error;
+	}
+
+	error = ocfs2_user_path_parent(newname, &nd, &to);
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	error = -EXDEV;
+	if (old_path.mnt != nd.path.mnt)
+		goto out_release;
+	new_dentry = lookup_create(&nd, 0);
+	error = PTR_ERR(new_dentry);
+	if (IS_ERR(new_dentry)) {
+		mlog_errno(error);
+		goto out_unlock;
+	}
+
+	error = mnt_want_write(nd.path.mnt);
+	if (error) {
+		mlog_errno(error);
+		goto out_dput;
+	}
+
+	error = ocfs2_vfs_reflink(old_path.dentry,
+				  nd.path.dentry->d_inode,
+				  new_dentry, preserve);
+	mnt_drop_write(nd.path.mnt);
+out_dput:
+	dput(new_dentry);
+out_unlock:
+	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+out_release:
+	path_put(&nd.path);
+	putname(to);
+out:
+	path_put(&old_path);
+
+	return error;
+}

fs/ocfs2/refcounttree.h

@@ -0,0 +1,106 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * refcounttree.h
+ *
+ * Copyright (C) 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#ifndef OCFS2_REFCOUNTTREE_H
+#define OCFS2_REFCOUNTTREE_H
+
+struct ocfs2_refcount_tree {
+	struct rb_node rf_node;
+	u64 rf_blkno;
+	u32 rf_generation;
+	struct rw_semaphore rf_sem;
+	struct ocfs2_lock_res rf_lockres;
+	struct kref rf_getcnt;
+	int rf_removed;
+
+	/* the following 4 fields are used by caching_info. */
+	struct ocfs2_caching_info rf_ci;
+	spinlock_t rf_lock;
+	struct mutex rf_io_mutex;
+	struct super_block *rf_sb;
+};
+
+void ocfs2_purge_refcount_trees(struct ocfs2_super *osb);
+int ocfs2_lock_refcount_tree(struct ocfs2_super *osb, u64 ref_blkno, int rw,
+			     struct ocfs2_refcount_tree **tree,
+			     struct buffer_head **ref_bh);
+void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
+				struct ocfs2_refcount_tree *tree,
+				int rw);
+
+int ocfs2_decrease_refcount(struct inode *inode,
+			    handle_t *handle, u32 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    int delete);
+int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
+					  struct buffer_head *di_bh,
+					  u64 phys_blkno,
+					  u32 clusters,
+					  int *credits,
+					  struct ocfs2_alloc_context **meta_ac);
+int ocfs2_refcount_cow(struct inode *inode, struct buffer_head *di_bh,
+		       u32 cpos, u32 write_len, u32 max_cpos);
+
+typedef int (ocfs2_post_refcount_func)(struct inode *inode,
+				       handle_t *handle,
+				       void *para);
+/*
+ * Some refcount caller need to do more work after we modify the data b-tree
+ * during refcount operation(including CoW and add refcount flag), and make the
+ * transaction complete. So it must give us this structure so that we can do it
+ * within our transaction.
+ *
+ */
+struct ocfs2_post_refcount {
+	int credits;			/* credits it need for journal. */
+	ocfs2_post_refcount_func *func;	/* real function. */
+	void *para;
+};
+
+int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
+				       struct ocfs2_caching_info *ref_ci,
+				       struct buffer_head *ref_root_bh,
+				       struct ocfs2_xattr_value_root *xv,
+				       int *meta_add, int *credits);
+int ocfs2_refcount_cow_xattr(struct inode *inode,
+			     struct ocfs2_dinode *di,
+			     struct ocfs2_xattr_value_buf *vb,
+			     struct ocfs2_refcount_tree *ref_tree,
+			     struct buffer_head *ref_root_bh,
+			     u32 cpos, u32 write_len,
+			     struct ocfs2_post_refcount *post);
+int ocfs2_add_refcount_flag(struct inode *inode,
+			    struct ocfs2_extent_tree *data_et,
+			    struct ocfs2_caching_info *ref_ci,
+			    struct buffer_head *ref_root_bh,
+			    u32 cpos, u32 p_cluster, u32 num_clusters,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    struct ocfs2_post_refcount *post);
+int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh);
+int ocfs2_try_remove_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh);
+int ocfs2_increase_refcount(handle_t *handle,
+			    struct ocfs2_caching_info *ci,
+			    struct buffer_head *ref_root_bh,
+			    u64 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_reflink_ioctl(struct inode *inode,
+			const char __user *oldname,
+			const char __user *newname,
+			bool preserve);
+#endif /* OCFS2_REFCOUNTTREE_H */

fs/ocfs2/resize.c

@@ -106,8 +106,8 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
 	mlog_entry("(new_clusters=%d, first_new_cluster = %u)\n",
 		   new_clusters, first_new_cluster);
 
-	ret = ocfs2_journal_access_gd(handle, bm_inode, group_bh,
-				      OCFS2_JOURNAL_ACCESS_WRITE);
+	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
+				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out;
@@ -141,7 +141,7 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
 	}
 
 	/* update the inode accordingly. */
-	ret = ocfs2_journal_access_di(handle, bm_inode, bm_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
@@ -514,7 +514,7 @@ int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
 		goto out_unlock;
 	}
 
-	ocfs2_set_new_buffer_uptodate(inode, group_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);
 
 	ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
 	if (ret) {
@@ -536,8 +536,8 @@ int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
 	cl = &fe->id2.i_chain;
 	cr = &cl->cl_recs[input->chain];
 
-	ret = ocfs2_journal_access_gd(handle, main_bm_inode, group_bh,
-				      OCFS2_JOURNAL_ACCESS_WRITE);
+	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
+				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out_commit;
@@ -552,8 +552,8 @@ int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
 		goto out_commit;
 	}
 
-	ret = ocfs2_journal_access_di(handle, main_bm_inode, main_bm_bh,
-				      OCFS2_JOURNAL_ACCESS_WRITE);
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
+				      main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
 		goto out_commit;

fs/ocfs2/slot_map.c

@@ -150,8 +150,8 @@ int ocfs2_refresh_slot_info(struct ocfs2_super *osb)
 	 * be !NULL.  Thus, ocfs2_read_blocks() will ignore blocknr.  If
 	 * this is not true, the read of -1 (UINT64_MAX) will fail.
 	 */
-	ret = ocfs2_read_blocks(si->si_inode, -1, si->si_blocks, si->si_bh,
-				OCFS2_BH_IGNORE_CACHE, NULL);
+	ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks,
+				si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL);
 	if (ret == 0) {
 		spin_lock(&osb->osb_lock);
 		ocfs2_update_slot_info(si);
@@ -213,7 +213,7 @@ static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
 		ocfs2_update_disk_slot_old(si, slot_num, &bh);
 	spin_unlock(&osb->osb_lock);
 
-	status = ocfs2_write_block(osb, bh, si->si_inode);
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
 	if (status < 0)
 		mlog_errno(status);
 
@@ -404,8 +404,8 @@ static int ocfs2_map_slot_buffers(struct ocfs2_super *osb,
 		     (unsigned long long)blkno);
 
 		bh = NULL;  /* Acquire a fresh bh */
-		status = ocfs2_read_blocks(si->si_inode, blkno, 1, &bh,
-					   OCFS2_BH_IGNORE_CACHE, NULL);
+		status = ocfs2_read_blocks(INODE_CACHE(si->si_inode), blkno,
+					   1, &bh, OCFS2_BH_IGNORE_CACHE, NULL);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;

fs/ocfs2/suballoc.c

@@ -310,7 +310,7 @@ int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 	int rc;
 	struct buffer_head *tmp = *bh;
 
-	rc = ocfs2_read_block(inode, gd_blkno, &tmp,
+	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
 			      ocfs2_validate_group_descriptor);
 	if (rc)
 		goto out;
@@ -352,7 +352,7 @@ static int ocfs2_block_group_fill(handle_t *handle,
 	}
 
 	status = ocfs2_journal_access_gd(handle,
-					 alloc_inode,
+					 INODE_CACHE(alloc_inode),
 					 bg_bh,
 					 OCFS2_JOURNAL_ACCESS_CREATE);
 	if (status < 0) {
@@ -476,7 +476,7 @@ static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 		mlog_errno(status);
 		goto bail;
 	}
-	ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 
 	status = ocfs2_block_group_fill(handle,
 					alloc_inode,
@@ -491,7 +491,7 @@ static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 
 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 
-	status = ocfs2_journal_access_di(handle, alloc_inode,
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1033,7 +1033,7 @@ static inline int ocfs2_block_group_set_bits(handle_t *handle,
 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
 
 	status = ocfs2_journal_access_gd(handle,
-					 alloc_inode,
+					 INODE_CACHE(alloc_inode),
 					 group_bh,
 					 journal_type);
 	if (status < 0) {
@@ -1106,7 +1106,8 @@ static int ocfs2_relink_block_group(handle_t *handle,
 	bg_ptr = le64_to_cpu(bg->bg_next_group);
 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
 
-	status = ocfs2_journal_access_gd(handle, alloc_inode, prev_bg_bh,
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 prev_bg_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
@@ -1121,8 +1122,8 @@ static int ocfs2_relink_block_group(handle_t *handle,
 		goto out_rollback;
 	}
 
-	status = ocfs2_journal_access_gd(handle, alloc_inode, bg_bh,
-					 OCFS2_JOURNAL_ACCESS_WRITE);
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
 		goto out_rollback;
@@ -1136,8 +1137,8 @@ static int ocfs2_relink_block_group(handle_t *handle,
 		goto out_rollback;
 	}
 
-	status = ocfs2_journal_access_di(handle, alloc_inode, fe_bh,
-					 OCFS2_JOURNAL_ACCESS_WRITE);
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
+					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
 		goto out_rollback;
@@ -1288,7 +1289,7 @@ static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
 
-	ret = ocfs2_journal_access_di(handle, inode, di_bh,
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
 				      OCFS2_JOURNAL_ACCESS_WRITE);
 	if (ret < 0) {
 		mlog_errno(ret);
@@ -1461,7 +1462,7 @@ static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
 	/* Ok, claim our bits now: set the info on dinode, chainlist
 	 * and then the group */
 	status = ocfs2_journal_access_di(handle,
-					 alloc_inode,
+					 INODE_CACHE(alloc_inode),
 					 ac->ac_bh,
 					 OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
@@ -1907,8 +1908,8 @@ static inline int ocfs2_block_group_clear_bits(handle_t *handle,
 	if (ocfs2_is_cluster_bitmap(alloc_inode))
 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
 
-	status = ocfs2_journal_access_gd(handle, alloc_inode, group_bh,
-					 journal_type);
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 group_bh, journal_type);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
@@ -1993,8 +1994,8 @@ int ocfs2_free_suballoc_bits(handle_t *handle,
 		goto bail;
 	}
 
-	status = ocfs2_journal_access_di(handle, alloc_inode, alloc_bh,
-					 OCFS2_JOURNAL_ACCESS_WRITE);
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
+					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
@@ -2151,7 +2152,7 @@ int ocfs2_lock_allocators(struct inode *inode,
 
 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
 
-	num_free_extents = ocfs2_num_free_extents(osb, inode, et);
+	num_free_extents = ocfs2_num_free_extents(osb, et);
 	if (num_free_extents < 0) {
 		ret = num_free_extents;
 		mlog_errno(ret);

fs/ocfs2/super.c

@@ -69,6 +69,7 @@
 #include "ver.h"
 #include "xattr.h"
 #include "quota.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -1668,8 +1669,6 @@ static void ocfs2_inode_init_once(void *data)
 	spin_lock_init(&oi->ip_lock);
 	ocfs2_extent_map_init(&oi->vfs_inode);
 	INIT_LIST_HEAD(&oi->ip_io_markers);
-	oi->ip_created_trans = 0;
-	oi->ip_last_trans = 0;
 	oi->ip_dir_start_lookup = 0;
 
 	init_rwsem(&oi->ip_alloc_sem);
@@ -1683,7 +1682,8 @@ static void ocfs2_inode_init_once(void *data)
 	ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
 	ocfs2_lock_res_init_once(&oi->ip_open_lockres);
 
-	ocfs2_metadata_cache_init(&oi->vfs_inode);
+	ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
+				  &ocfs2_inode_caching_ops);
 
 	inode_init_once(&oi->vfs_inode);
 }
@@ -1859,6 +1859,8 @@ static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
 
 	ocfs2_sync_blockdev(sb);
 
+	ocfs2_purge_refcount_trees(osb);
+
 	/* No cluster connection means we've failed during mount, so skip
 	 * all the steps which depended on that to complete. */
 	if (osb->cconn) {
@@ -2065,6 +2067,8 @@ static int ocfs2_initialize_super(struct super_block *sb,
 		goto bail;
 	}
 
+	osb->osb_rf_lock_tree = RB_ROOT;
+
 	osb->s_feature_compat =
 		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
 	osb->s_feature_ro_compat =
@@ -2490,7 +2494,8 @@ void __ocfs2_abort(struct super_block* sb,
 	/* Force a panic(). This stinks, but it's better than letting
 	 * things continue without having a proper hard readonly
 	 * here. */
-	OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
+	if (!ocfs2_mount_local(OCFS2_SB(sb)))
+		OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 	ocfs2_handle_error(sb);
 }
 

fs/ocfs2/uptodate.c

@@ -75,15 +75,77 @@ struct ocfs2_meta_cache_item {
 
 static struct kmem_cache *ocfs2_uptodate_cachep = NULL;
 
-void ocfs2_metadata_cache_init(struct inode *inode)
+u64 ocfs2_metadata_cache_owner(struct ocfs2_caching_info *ci)
 {
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
+	BUG_ON(!ci || !ci->ci_ops);
 
-	oi->ip_flags |= OCFS2_INODE_CACHE_INLINE;
+	return ci->ci_ops->co_owner(ci);
+}
+
+struct super_block *ocfs2_metadata_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	return ci->ci_ops->co_get_super(ci);
+}
+
+static void ocfs2_metadata_cache_lock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_cache_lock(ci);
+}
+
+static void ocfs2_metadata_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_cache_unlock(ci);
+}
+
+void ocfs2_metadata_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_io_lock(ci);
+}
+
+void ocfs2_metadata_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_io_unlock(ci);
+}
+
+
+static void ocfs2_metadata_cache_reset(struct ocfs2_caching_info *ci,
+				       int clear)
+{
+	ci->ci_flags |= OCFS2_CACHE_FL_INLINE;
 	ci->ci_num_cached = 0;
+
+	if (clear) {
+		ci->ci_created_trans = 0;
+		ci->ci_last_trans = 0;
+	}
+}
+
+void ocfs2_metadata_cache_init(struct ocfs2_caching_info *ci,
+			       const struct ocfs2_caching_operations *ops)
+{
+	BUG_ON(!ops);
+
+	ci->ci_ops = ops;
+	ocfs2_metadata_cache_reset(ci, 1);
 }
 
+void ocfs2_metadata_cache_exit(struct ocfs2_caching_info *ci)
+{
+	ocfs2_metadata_cache_purge(ci);
+	ocfs2_metadata_cache_reset(ci, 1);
+}
+
+
 /* No lock taken here as 'root' is not expected to be visible to other
  * processes. */
 static unsigned int ocfs2_purge_copied_metadata_tree(struct rb_root *root)
@@ -112,19 +174,20 @@ static unsigned int ocfs2_purge_copied_metadata_tree(struct rb_root *root)
  * This function is a few more lines longer than necessary due to some
  * accounting done here, but I think it's worth tracking down those
  * bugs sooner -- Mark */
-void ocfs2_metadata_cache_purge(struct inode *inode)
+void ocfs2_metadata_cache_purge(struct ocfs2_caching_info *ci)
 {
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
 	unsigned int tree, to_purge, purged;
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
 	struct rb_root root = RB_ROOT;
 
-	spin_lock(&oi->ip_lock);
-	tree = !(oi->ip_flags & OCFS2_INODE_CACHE_INLINE);
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ocfs2_metadata_cache_lock(ci);
+	tree = !(ci->ci_flags & OCFS2_CACHE_FL_INLINE);
 	to_purge = ci->ci_num_cached;
 
-	mlog(0, "Purge %u %s items from Inode %llu\n", to_purge,
-	     tree ? "array" : "tree", (unsigned long long)oi->ip_blkno);
+	mlog(0, "Purge %u %s items from Owner %llu\n", to_purge,
+	     tree ? "array" : "tree",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci));
 
 	/* If we're a tree, save off the root so that we can safely
 	 * initialize the cache. We do the work to free tree members
@@ -132,16 +195,17 @@ void ocfs2_metadata_cache_purge(struct inode *inode)
 	if (tree)
 		root = ci->ci_cache.ci_tree;
 
-	ocfs2_metadata_cache_init(inode);
-	spin_unlock(&oi->ip_lock);
+	ocfs2_metadata_cache_reset(ci, 0);
+	ocfs2_metadata_cache_unlock(ci);
 
 	purged = ocfs2_purge_copied_metadata_tree(&root);
 	/* If possible, track the number wiped so that we can more
 	 * easily detect counting errors. Unfortunately, this is only
 	 * meaningful for trees. */
 	if (tree && purged != to_purge)
-		mlog(ML_ERROR, "Inode %llu, count = %u, purged = %u\n",
-		     (unsigned long long)oi->ip_blkno, to_purge, purged);
+		mlog(ML_ERROR, "Owner %llu, count = %u, purged = %u\n",
+		     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+		     to_purge, purged);
 }
 
 /* Returns the index in the cache array, -1 if not found.
@@ -182,27 +246,25 @@ ocfs2_search_cache_tree(struct ocfs2_caching_info *ci,
 	return NULL;
 }
 
-static int ocfs2_buffer_cached(struct ocfs2_inode_info *oi,
+static int ocfs2_buffer_cached(struct ocfs2_caching_info *ci,
 			       struct buffer_head *bh)
 {
 	int index = -1;
 	struct ocfs2_meta_cache_item *item = NULL;
 
-	spin_lock(&oi->ip_lock);
+	ocfs2_metadata_cache_lock(ci);
 
-	mlog(0, "Inode %llu, query block %llu (inline = %u)\n",
-	     (unsigned long long)oi->ip_blkno,
+	mlog(0, "Owner %llu, query block %llu (inline = %u)\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
 	     (unsigned long long) bh->b_blocknr,
-	     !!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE));
+	     !!(ci->ci_flags & OCFS2_CACHE_FL_INLINE));
 
-	if (oi->ip_flags & OCFS2_INODE_CACHE_INLINE)
-		index = ocfs2_search_cache_array(&oi->ip_metadata_cache,
-						 bh->b_blocknr);
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE)
+		index = ocfs2_search_cache_array(ci, bh->b_blocknr);
 	else
-		item = ocfs2_search_cache_tree(&oi->ip_metadata_cache,
-					       bh->b_blocknr);
+		item = ocfs2_search_cache_tree(ci, bh->b_blocknr);
 
-	spin_unlock(&oi->ip_lock);
+	ocfs2_metadata_cache_unlock(ci);
 
 	mlog(0, "index = %d, item = %p\n", index, item);
 
@@ -214,7 +276,7 @@ static int ocfs2_buffer_cached(struct ocfs2_inode_info *oi,
  * 
  * This can be called under lock_buffer()
  */
-int ocfs2_buffer_uptodate(struct inode *inode,
+int ocfs2_buffer_uptodate(struct ocfs2_caching_info *ci,
 			  struct buffer_head *bh)
 {
 	/* Doesn't matter if the bh is in our cache or not -- if it's
@@ -230,24 +292,24 @@ int ocfs2_buffer_uptodate(struct inode *inode,
 
 	/* Ok, locally the buffer is marked as up to date, now search
 	 * our cache to see if we can trust that. */
-	return ocfs2_buffer_cached(OCFS2_I(inode), bh);
+	return ocfs2_buffer_cached(ci, bh);
 }
 
-/* 
+/*
  * Determine whether a buffer is currently out on a read-ahead request.
- * ip_io_sem should be held to serialize submitters with the logic here.
+ * ci_io_sem should be held to serialize submitters with the logic here.
  */
-int ocfs2_buffer_read_ahead(struct inode *inode,
+int ocfs2_buffer_read_ahead(struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh)
 {
-	return buffer_locked(bh) && ocfs2_buffer_cached(OCFS2_I(inode), bh);
+	return buffer_locked(bh) && ocfs2_buffer_cached(ci, bh);
 }
 
 /* Requires ip_lock */
 static void ocfs2_append_cache_array(struct ocfs2_caching_info *ci,
 				     sector_t block)
 {
-	BUG_ON(ci->ci_num_cached >= OCFS2_INODE_MAX_CACHE_ARRAY);
+	BUG_ON(ci->ci_num_cached >= OCFS2_CACHE_INFO_MAX_ARRAY);
 
 	mlog(0, "block %llu takes position %u\n", (unsigned long long) block,
 	     ci->ci_num_cached);
@@ -292,66 +354,64 @@ static void __ocfs2_insert_cache_tree(struct ocfs2_caching_info *ci,
 	ci->ci_num_cached++;
 }
 
-static inline int ocfs2_insert_can_use_array(struct ocfs2_inode_info *oi,
-					     struct ocfs2_caching_info *ci)
+/* co_cache_lock() must be held */
+static inline int ocfs2_insert_can_use_array(struct ocfs2_caching_info *ci)
 {
-	assert_spin_locked(&oi->ip_lock);
-
-	return (oi->ip_flags & OCFS2_INODE_CACHE_INLINE) &&
-		(ci->ci_num_cached < OCFS2_INODE_MAX_CACHE_ARRAY);
+	return (ci->ci_flags & OCFS2_CACHE_FL_INLINE) &&
+		(ci->ci_num_cached < OCFS2_CACHE_INFO_MAX_ARRAY);
 }
 
-/* tree should be exactly OCFS2_INODE_MAX_CACHE_ARRAY wide. NULL the
+/* tree should be exactly OCFS2_CACHE_INFO_MAX_ARRAY wide. NULL the
  * pointers in tree after we use them - this allows caller to detect
- * when to free in case of error. */
-static void ocfs2_expand_cache(struct ocfs2_inode_info *oi,
+ * when to free in case of error.
+ *
+ * The co_cache_lock() must be held. */
+static void ocfs2_expand_cache(struct ocfs2_caching_info *ci,
 			       struct ocfs2_meta_cache_item **tree)
 {
 	int i;
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
 
-	mlog_bug_on_msg(ci->ci_num_cached != OCFS2_INODE_MAX_CACHE_ARRAY,
-			"Inode %llu, num cached = %u, should be %u\n",
-			(unsigned long long)oi->ip_blkno, ci->ci_num_cached,
-			OCFS2_INODE_MAX_CACHE_ARRAY);
-	mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
-			"Inode %llu not marked as inline anymore!\n",
-			(unsigned long long)oi->ip_blkno);
-	assert_spin_locked(&oi->ip_lock);
+	mlog_bug_on_msg(ci->ci_num_cached != OCFS2_CACHE_INFO_MAX_ARRAY,
+			"Owner %llu, num cached = %u, should be %u\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci),
+			ci->ci_num_cached, OCFS2_CACHE_INFO_MAX_ARRAY);
+	mlog_bug_on_msg(!(ci->ci_flags & OCFS2_CACHE_FL_INLINE),
+			"Owner %llu not marked as inline anymore!\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci));
 
 	/* Be careful to initialize the tree members *first* because
 	 * once the ci_tree is used, the array is junk... */
-	for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++)
+	for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++)
 		tree[i]->c_block = ci->ci_cache.ci_array[i];
 
-	oi->ip_flags &= ~OCFS2_INODE_CACHE_INLINE;
+	ci->ci_flags &= ~OCFS2_CACHE_FL_INLINE;
 	ci->ci_cache.ci_tree = RB_ROOT;
 	/* this will be set again by __ocfs2_insert_cache_tree */
 	ci->ci_num_cached = 0;
 
-	for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++) {
+	for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++) {
 		__ocfs2_insert_cache_tree(ci, tree[i]);
 		tree[i] = NULL;
 	}
 
 	mlog(0, "Expanded %llu to a tree cache: flags 0x%x, num = %u\n",
-	     (unsigned long long)oi->ip_blkno, oi->ip_flags, ci->ci_num_cached);
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+	     ci->ci_flags, ci->ci_num_cached);
 }
 
 /* Slow path function - memory allocation is necessary. See the
  * comment above ocfs2_set_buffer_uptodate for more information. */
-static void __ocfs2_set_buffer_uptodate(struct ocfs2_inode_info *oi,
+static void __ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
 					sector_t block,
 					int expand_tree)
 {
 	int i;
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
 	struct ocfs2_meta_cache_item *new = NULL;
-	struct ocfs2_meta_cache_item *tree[OCFS2_INODE_MAX_CACHE_ARRAY] =
+	struct ocfs2_meta_cache_item *tree[OCFS2_CACHE_INFO_MAX_ARRAY] =
 		{ NULL, };
 
-	mlog(0, "Inode %llu, block %llu, expand = %d\n",
-	     (unsigned long long)oi->ip_blkno,
+	mlog(0, "Owner %llu, block %llu, expand = %d\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
 	     (unsigned long long)block, expand_tree);
 
 	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_NOFS);
@@ -364,7 +424,7 @@ static void __ocfs2_set_buffer_uptodate(struct ocfs2_inode_info *oi,
 	if (expand_tree) {
 		/* Do *not* allocate an array here - the removal code
 		 * has no way of tracking that. */
-		for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++) {
+		for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++) {
 			tree[i] = kmem_cache_alloc(ocfs2_uptodate_cachep,
 						   GFP_NOFS);
 			if (!tree[i]) {
@@ -376,21 +436,21 @@ static void __ocfs2_set_buffer_uptodate(struct ocfs2_inode_info *oi,
 		}
 	}
 
-	spin_lock(&oi->ip_lock);
-	if (ocfs2_insert_can_use_array(oi, ci)) {
+	ocfs2_metadata_cache_lock(ci);
+	if (ocfs2_insert_can_use_array(ci)) {
 		mlog(0, "Someone cleared the tree underneath us\n");
 		/* Ok, items were removed from the cache in between
 		 * locks. Detect this and revert back to the fast path */
 		ocfs2_append_cache_array(ci, block);
-		spin_unlock(&oi->ip_lock);
+		ocfs2_metadata_cache_unlock(ci);
 		goto out_free;
 	}
 
 	if (expand_tree)
-		ocfs2_expand_cache(oi, tree);
+		ocfs2_expand_cache(ci, tree);
 
 	__ocfs2_insert_cache_tree(ci, new);
-	spin_unlock(&oi->ip_lock);
+	ocfs2_metadata_cache_unlock(ci);
 
 	new = NULL;
 out_free:
@@ -400,14 +460,14 @@ out_free:
 	/* If these were used, then ocfs2_expand_cache re-set them to
 	 * NULL for us. */
 	if (tree[0]) {
-		for(i = 0; i < OCFS2_INODE_MAX_CACHE_ARRAY; i++)
+		for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++)
 			if (tree[i])
 				kmem_cache_free(ocfs2_uptodate_cachep,
 						tree[i]);
 	}
 }
 
-/* Item insertion is guarded by ip_io_mutex, so the insertion path takes
+/* Item insertion is guarded by co_io_lock(), so the insertion path takes
  * advantage of this by not rechecking for a duplicate insert during
  * the slow case. Additionally, if the cache needs to be bumped up to
  * a tree, the code will not recheck after acquiring the lock --
@@ -425,59 +485,55 @@ out_free:
  * Readahead buffers can be passed in here before the I/O request is
  * completed.
  */
-void ocfs2_set_buffer_uptodate(struct inode *inode,
+void ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
 			       struct buffer_head *bh)
 {
 	int expand;
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
 
 	/* The block may very well exist in our cache already, so avoid
 	 * doing any more work in that case. */
-	if (ocfs2_buffer_cached(oi, bh))
+	if (ocfs2_buffer_cached(ci, bh))
 		return;
 
-	mlog(0, "Inode %llu, inserting block %llu\n",
-	     (unsigned long long)oi->ip_blkno,
+	mlog(0, "Owner %llu, inserting block %llu\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
 	     (unsigned long long)bh->b_blocknr);
 
 	/* No need to recheck under spinlock - insertion is guarded by
-	 * ip_io_mutex */
-	spin_lock(&oi->ip_lock);
-	if (ocfs2_insert_can_use_array(oi, ci)) {
+	 * co_io_lock() */
+	ocfs2_metadata_cache_lock(ci);
+	if (ocfs2_insert_can_use_array(ci)) {
 		/* Fast case - it's an array and there's a free
 		 * spot. */
 		ocfs2_append_cache_array(ci, bh->b_blocknr);
-		spin_unlock(&oi->ip_lock);
+		ocfs2_metadata_cache_unlock(ci);
 		return;
 	}
 
 	expand = 0;
-	if (oi->ip_flags & OCFS2_INODE_CACHE_INLINE) {
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE) {
 		/* We need to bump things up to a tree. */
 		expand = 1;
 	}
-	spin_unlock(&oi->ip_lock);
+	ocfs2_metadata_cache_unlock(ci);
 
-	__ocfs2_set_buffer_uptodate(oi, bh->b_blocknr, expand);
+	__ocfs2_set_buffer_uptodate(ci, bh->b_blocknr, expand);
 }
 
 /* Called against a newly allocated buffer. Most likely nobody should
  * be able to read this sort of metadata while it's still being
- * allocated, but this is careful to take ip_io_mutex anyway. */
-void ocfs2_set_new_buffer_uptodate(struct inode *inode,
+ * allocated, but this is careful to take co_io_lock() anyway. */
+void ocfs2_set_new_buffer_uptodate(struct ocfs2_caching_info *ci,
 				   struct buffer_head *bh)
 {
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-
 	/* This should definitely *not* exist in our cache */
-	BUG_ON(ocfs2_buffer_cached(oi, bh));
+	BUG_ON(ocfs2_buffer_cached(ci, bh));
 
 	set_buffer_uptodate(bh);
 
-	mutex_lock(&oi->ip_io_mutex);
-	ocfs2_set_buffer_uptodate(inode, bh);
-	mutex_unlock(&oi->ip_io_mutex);
+	ocfs2_metadata_cache_io_lock(ci);
+	ocfs2_set_buffer_uptodate(ci, bh);
+	ocfs2_metadata_cache_io_unlock(ci);
 }
 
 /* Requires ip_lock. */
@@ -487,7 +543,7 @@ static void ocfs2_remove_metadata_array(struct ocfs2_caching_info *ci,
 	sector_t *array = ci->ci_cache.ci_array;
 	int bytes;
 
-	BUG_ON(index < 0 || index >= OCFS2_INODE_MAX_CACHE_ARRAY);
+	BUG_ON(index < 0 || index >= OCFS2_CACHE_INFO_MAX_ARRAY);
 	BUG_ON(index >= ci->ci_num_cached);
 	BUG_ON(!ci->ci_num_cached);
 
@@ -515,21 +571,19 @@ static void ocfs2_remove_metadata_tree(struct ocfs2_caching_info *ci,
 	ci->ci_num_cached--;
 }
 
-static void ocfs2_remove_block_from_cache(struct inode *inode,
+static void ocfs2_remove_block_from_cache(struct ocfs2_caching_info *ci,
 					  sector_t block)
 {
 	int index;
 	struct ocfs2_meta_cache_item *item = NULL;
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-	struct ocfs2_caching_info *ci = &oi->ip_metadata_cache;
 
-	spin_lock(&oi->ip_lock);
-	mlog(0, "Inode %llu, remove %llu, items = %u, array = %u\n",
-	     (unsigned long long)oi->ip_blkno,
+	ocfs2_metadata_cache_lock(ci);
+	mlog(0, "Owner %llu, remove %llu, items = %u, array = %u\n",
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
 	     (unsigned long long) block, ci->ci_num_cached,
-	     oi->ip_flags & OCFS2_INODE_CACHE_INLINE);
+	     ci->ci_flags & OCFS2_CACHE_FL_INLINE);
 
-	if (oi->ip_flags & OCFS2_INODE_CACHE_INLINE) {
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE) {
 		index = ocfs2_search_cache_array(ci, block);
 		if (index != -1)
 			ocfs2_remove_metadata_array(ci, index);
@@ -538,7 +592,7 @@ static void ocfs2_remove_block_from_cache(struct inode *inode,
 		if (item)
 			ocfs2_remove_metadata_tree(ci, item);
 	}
-	spin_unlock(&oi->ip_lock);
+	ocfs2_metadata_cache_unlock(ci);
 
 	if (item)
 		kmem_cache_free(ocfs2_uptodate_cachep, item);
@@ -549,23 +603,24 @@ static void ocfs2_remove_block_from_cache(struct inode *inode,
  * bother reverting things to an inlined array in the case of a remove
  * which moves us back under the limit.
  */
-void ocfs2_remove_from_cache(struct inode *inode,
+void ocfs2_remove_from_cache(struct ocfs2_caching_info *ci,
 			     struct buffer_head *bh)
 {
 	sector_t block = bh->b_blocknr;
 
-	ocfs2_remove_block_from_cache(inode, block);
+	ocfs2_remove_block_from_cache(ci, block);
 }
 
 /* Called when we remove xattr clusters from an inode. */
-void ocfs2_remove_xattr_clusters_from_cache(struct inode *inode,
+void ocfs2_remove_xattr_clusters_from_cache(struct ocfs2_caching_info *ci,
 					    sector_t block,
 					    u32 c_len)
 {
-	unsigned int i, b_len = ocfs2_clusters_to_blocks(inode->i_sb, 1) * c_len;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	unsigned int i, b_len = ocfs2_clusters_to_blocks(sb, 1) * c_len;
 
 	for (i = 0; i < b_len; i++, block++)
-		ocfs2_remove_block_from_cache(inode, block);
+		ocfs2_remove_block_from_cache(ci, block);
 }
 
 int __init init_ocfs2_uptodate_cache(void)
@@ -577,7 +632,7 @@ int __init init_ocfs2_uptodate_cache(void)
 		return -ENOMEM;
 
 	mlog(0, "%u inlined cache items per inode.\n",
-	     OCFS2_INODE_MAX_CACHE_ARRAY);
+	     OCFS2_CACHE_INFO_MAX_ARRAY);
 
 	return 0;
 }

fs/ocfs2/uptodate.h

@@ -26,24 +26,59 @@
 #ifndef OCFS2_UPTODATE_H
 #define OCFS2_UPTODATE_H
 
+/*
+ * The caching code relies on locking provided by the user of
+ * struct ocfs2_caching_info.  These operations connect that up.
+ */
+struct ocfs2_caching_operations {
+	/*
+	 * A u64 representing the owning structure.  Usually this
+	 * is the block number (i_blkno or whatnot).  This is used so
+	 * that caching log messages can identify the owning structure.
+	 */
+	u64	(*co_owner)(struct ocfs2_caching_info *ci);
+
+	/* The superblock is needed during I/O. */
+	struct super_block *(*co_get_super)(struct ocfs2_caching_info *ci);
+	/*
+	 * Lock and unlock the caching data.  These will not sleep, and
+	 * should probably be spinlocks.
+	 */
+	void	(*co_cache_lock)(struct ocfs2_caching_info *ci);
+	void	(*co_cache_unlock)(struct ocfs2_caching_info *ci);
+
+	/*
+	 * Lock and unlock for disk I/O.  These will sleep, and should
+	 * be mutexes.
+	 */
+	void	(*co_io_lock)(struct ocfs2_caching_info *ci);
+	void	(*co_io_unlock)(struct ocfs2_caching_info *ci);
+};
+
 int __init init_ocfs2_uptodate_cache(void);
 void exit_ocfs2_uptodate_cache(void);
 
-void ocfs2_metadata_cache_init(struct inode *inode);
-void ocfs2_metadata_cache_purge(struct inode *inode);
+void ocfs2_metadata_cache_init(struct ocfs2_caching_info *ci,
+			       const struct ocfs2_caching_operations *ops);
+void ocfs2_metadata_cache_purge(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_exit(struct ocfs2_caching_info *ci);
+
+u64 ocfs2_metadata_cache_owner(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_io_lock(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_io_unlock(struct ocfs2_caching_info *ci);
 
-int ocfs2_buffer_uptodate(struct inode *inode,
+int ocfs2_buffer_uptodate(struct ocfs2_caching_info *ci,
 			  struct buffer_head *bh);
-void ocfs2_set_buffer_uptodate(struct inode *inode,
+void ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
 			       struct buffer_head *bh);
-void ocfs2_set_new_buffer_uptodate(struct inode *inode,
+void ocfs2_set_new_buffer_uptodate(struct ocfs2_caching_info *ci,
 				   struct buffer_head *bh);
-void ocfs2_remove_from_cache(struct inode *inode,
+void ocfs2_remove_from_cache(struct ocfs2_caching_info *ci,
 			     struct buffer_head *bh);
-void ocfs2_remove_xattr_clusters_from_cache(struct inode *inode,
+void ocfs2_remove_xattr_clusters_from_cache(struct ocfs2_caching_info *ci,
 					    sector_t block,
 					    u32 c_len);
-int ocfs2_buffer_read_ahead(struct inode *inode,
+int ocfs2_buffer_read_ahead(struct ocfs2_caching_info *ci,
 			    struct buffer_head *bh);
 
 #endif /* OCFS2_UPTODATE_H */

fs/ocfs2/xattr.h

@@ -55,6 +55,8 @@ int ocfs2_xattr_set_handle(handle_t *, struct inode *, struct buffer_head *,
 			   int, const char *, const void *, size_t, int,
 			   struct ocfs2_alloc_context *,
 			   struct ocfs2_alloc_context *);
+int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
+					 struct ocfs2_dinode *di);
 int ocfs2_xattr_remove(struct inode *, struct buffer_head *);
 int ocfs2_init_security_get(struct inode *, struct inode *,
 			    struct ocfs2_security_xattr_info *);
@@ -83,5 +85,16 @@ struct ocfs2_xattr_value_buf {
 	struct ocfs2_xattr_value_root	*vb_xv;
 };
 
-
+int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
+				     struct buffer_head *fe_bh,
+				     struct ocfs2_caching_info *ref_ci,
+				     struct buffer_head *ref_root_bh,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_reflink_xattrs(struct inode *old_inode,
+			 struct buffer_head *old_bh,
+			 struct inode *new_inode,
+			 struct buffer_head *new_bh,
+			 bool preserve_security);
+int ocfs2_init_security_and_acl(struct inode *dir,
+				struct inode *inode);
 #endif /* OCFS2_XATTR_H */