ext4: disable merging of uninitialized extents
authorDmitry Monakhov <dmonakhov@openvz.org>
Mon, 4 Mar 2013 05:36:06 +0000 (00:36 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 4 Mar 2013 05:36:06 +0000 (00:36 -0500)
Derived from Jan's patch:http://permalink.gmane.org/gmane.comp.file-systems.ext4/36470

Merging of uninitialized extents creates all sorts of interesting race
possibilities when writeback / DIO races with fallocate. Thus
ext4_convert_unwritten_extents_endio() has to deal with a case where
extent to be converted needs to be split out first. That isn't nice
for two reasons:

1) It may need allocation of extent tree block so ENOSPC is possible.
2) It complicates end_io handling code

So we disable merging of uninitialized extents which allows us to simplify
the code. Extents will get merged after they are converted to initialized
ones.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
fs/ext4/extents.c

index bef194a144377527d06d6d22f6bea5f5c83f2e2e..60818ed1f6a9b13a9da2dc375a84741d5f533650 100644 (file)
@@ -1584,10 +1584,12 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
        unsigned short ext1_ee_len, ext2_ee_len, max_len;
 
        /*
-        * Make sure that either both extents are uninitialized, or
-        * both are _not_.
+        * Make sure that both extents are initialized. We don't merge
+        * uninitialized extents so that we can be sure that end_io code has
+        * the extent that was written properly split out and conversion to
+        * initialized is trivial.
         */
-       if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
+       if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2))
                return 0;
 
        if (ext4_ext_is_uninitialized(ex1))