From: Eric Sandeen <sandeen@sgi.com>
Subject: TAKE - Add buffer to inode's dirty list after truncate()
To: undisclosed-recipients: ;
Date: Tue, 24 Sep 2002 16:39:35 -0500
Lines: 29

block_truncate_page() does a __mark_buffer_dirty(bh) at the end, but it
does not file the buffer on the inode's dirty data queue, so only
bdflush can ever get to it, and other sync mechanisms which call
fsync_inode_data_buffers() do not see it.

This was the root cause of the mapcheck program failing on fsr'd files,
it found data past EOF as a result of the failed sync after truncate.

This one seems to have the Kernel Gods' blessings, and will hopefully
be in 2.4.20.

Index: linux/fs/buffer.c
===================================================================
RCS file: /cvs/linux-2.4-xfs/linux/fs/buffer.c,v
retrieving revision 1.109
retrieving revision 1.110
diff -u -r1.109 -r1.110
--- linux/fs/buffer.c	2002/07/25 15:31:36	1.109
+++ linux/fs/buffer.c	2002/09/24 21:36:23	1.110
@@ -2028,7 +2028,12 @@
 	flush_dcache_page(page);
 	kunmap(page);
 
-	__mark_buffer_dirty(bh);
+	if (!atomic_set_buffer_dirty(bh)) {
+		__mark_dirty(bh);
+		buffer_insert_inode_data_queue(bh, inode);
+		balance_dirty();
+	}
+
 	err = 0;
 
 unlock: