[patch 23/26] ocfs2: fix write() performance regression

[Greg Kroah-Hartman <gregkh@suse.de>]

[-- Attachment #1: ocfs2-fix-write-performance-regression.patch --]
[-- Type: text/plain, Size: 2025 bytes --]


2.6.22-stable review patch.  If anyone has any objections, please let us
know.

------------------
From: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: fix write() performance regression

patch 4e9563fd55ff4479f2b118d0757d121dd0cfc39c in mainline.

On file systems which don't support sparse files, Ocfs2_map_page_blocks()
was reading blocks on appending writes. This caused write performance to
suffer dramatically. Fix this by detecting an appending write on a nonsparse
fs and skipping the read.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

---
 fs/ocfs2/aops.c |   24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -661,6 +661,27 @@ static void ocfs2_clear_page_regions(str
 }
 
 /*
+ * Nonsparse file systems fully allocate before we get to the write
+ * code. This prevents ocfs2_write() from tagging the write as an
+ * allocating one, which means ocfs2_map_page_blocks() might try to
+ * read-in the blocks at the tail of our file. Avoid reading them by
+ * testing i_size against each block offset.
+ */
+static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
+				 unsigned int block_start)
+{
+	u64 offset = page_offset(page) + block_start;
+
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 1;
+
+	if (i_size_read(inode) > offset)
+		return 1;
+
+	return 0;
+}
+
+/*
  * Some of this taken from block_prepare_write(). We already have our
  * mapping by now though, and the entire write will be allocating or
  * it won't, so not much need to use BH_New.
@@ -711,7 +732,8 @@ int ocfs2_map_page_blocks(struct page *p
 			if (!buffer_uptodate(bh))
 				set_buffer_uptodate(bh);
 		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
-		     (block_start < from || block_end > to)) {
+			   ocfs2_should_read_blk(inode, page, block_start) &&
+			   (block_start < from || block_end > to)) {
 			ll_rw_block(READ, 1, &bh);
 			*wait_bh++=bh;
 		}

--