diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
index 4fb2bbc..cdea450 100644
--- a/fs/ceph/dir.c
+++ b/fs/ceph/dir.c
@@ -110,6 +110,50 @@
 	return 0;
 }
 
+
+static struct dentry *
+__dcache_find_get_entry(struct dentry *parent, u64 idx,
+			struct ceph_readdir_cache_control *cache_ctl)
+{
+	struct inode *dir = d_inode(parent);
+	struct dentry *dentry;
+	unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
+	loff_t ptr_pos = idx * sizeof(struct dentry *);
+	pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
+
+	if (ptr_pos >= i_size_read(dir))
+		return NULL;
+
+	if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
+		ceph_readdir_cache_release(cache_ctl);
+		cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
+		if (!cache_ctl->page) {
+			dout(" page %lu not found\n", ptr_pgoff);
+			return ERR_PTR(-EAGAIN);
+		}
+		/* reading/filling the cache are serialized by
+		   i_mutex, no need to use page lock */
+		unlock_page(cache_ctl->page);
+		cache_ctl->dentries = kmap(cache_ctl->page);
+	}
+
+	cache_ctl->index = idx & idx_mask;
+
+	rcu_read_lock();
+	spin_lock(&parent->d_lock);
+	/* check i_size again here, because empty directory can be
+	 * marked as complete while not holding the i_mutex. */
+	if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
+		dentry = cache_ctl->dentries[cache_ctl->index];
+	else
+		dentry = NULL;
+	spin_unlock(&parent->d_lock);
+	if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
+		dentry = NULL;
+	rcu_read_unlock();
+	return dentry ? : ERR_PTR(-EAGAIN);
+}
+
 /*
  * When possible, we try to satisfy a readdir by peeking at the
  * dcache.  We make this work by carefully ordering dentries on
@@ -129,62 +173,57 @@
 	struct inode *dir = d_inode(parent);
 	struct dentry *dentry, *last = NULL;
 	struct ceph_dentry_info *di;
-	unsigned nsize = PAGE_SIZE / sizeof(struct dentry *);
-	int err = 0;
-	loff_t ptr_pos = 0;
 	struct ceph_readdir_cache_control cache_ctl = {};
+	u64 idx = 0;
+	int err = 0;
 
 	dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
 
-	/* we can calculate cache index for the first dirfrag */
-	if (ceph_frag_is_leftmost(fpos_frag(ctx->pos))) {
-		cache_ctl.index = fpos_off(ctx->pos) - 2;
-		BUG_ON(cache_ctl.index < 0);
-		ptr_pos = cache_ctl.index * sizeof(struct dentry *);
+	/* search start position */
+	if (ctx->pos > 2) {
+		u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
+		while (count > 0) {
+			u64 step = count >> 1;
+			dentry = __dcache_find_get_entry(parent, idx + step,
+							 &cache_ctl);
+			if (!dentry) {
+				/* use linar search */
+				idx = 0;
+				break;
+			}
+			if (IS_ERR(dentry)) {
+				err = PTR_ERR(dentry);
+				goto out;
+			}
+			di = ceph_dentry(dentry);
+			spin_lock(&dentry->d_lock);
+			if (fpos_cmp(di->offset, ctx->pos) < 0) {
+				idx += step + 1;
+				count -= step + 1;
+			} else {
+				count = step;
+			}
+			spin_unlock(&dentry->d_lock);
+			dput(dentry);
+		}
+
+		dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
 	}
 
-	while (true) {
-		pgoff_t pgoff;
-		bool emit_dentry;
 
-		if (ptr_pos >= i_size_read(dir)) {
+	for (;;) {
+		bool emit_dentry = false;
+		dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
+		if (!dentry) {
 			fi->flags |= CEPH_F_ATEND;
 			err = 0;
 			break;
 		}
-
-		err = -EAGAIN;
-		pgoff = ptr_pos >> PAGE_SHIFT;
-		if (!cache_ctl.page || pgoff != page_index(cache_ctl.page)) {
-			ceph_readdir_cache_release(&cache_ctl);
-			cache_ctl.page = find_lock_page(&dir->i_data, pgoff);
-			if (!cache_ctl.page) {
-				dout(" page %lu not found\n", pgoff);
-				break;
-			}
-			/* reading/filling the cache are serialized by
-			 * i_mutex, no need to use page lock */
-			unlock_page(cache_ctl.page);
-			cache_ctl.dentries = kmap(cache_ctl.page);
+		if (IS_ERR(dentry)) {
+			err = PTR_ERR(dentry);
+			goto out;
 		}
 
-		rcu_read_lock();
-		spin_lock(&parent->d_lock);
-		/* check i_size again here, because empty directory can be
-		 * marked as complete while not holding the i_mutex. */
-		if (ceph_dir_is_complete_ordered(dir) &&
-		    ptr_pos < i_size_read(dir))
-			dentry = cache_ctl.dentries[cache_ctl.index % nsize];
-		else
-			dentry = NULL;
-		spin_unlock(&parent->d_lock);
-		if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
-			dentry = NULL;
-		rcu_read_unlock();
-		if (!dentry)
-			break;
-
-		emit_dentry = false;
 		di = ceph_dentry(dentry);
 		spin_lock(&dentry->d_lock);
 		if (di->lease_shared_gen == shared_gen &&
@@ -217,10 +256,8 @@
 		} else {
 			dput(dentry);
 		}
-
-		cache_ctl.index++;
-		ptr_pos += sizeof(struct dentry *);
 	}
+out:
 	ceph_readdir_cache_release(&cache_ctl);
 	if (last) {
 		int ret;