]> pilppa.com Git - linux-2.6-omap-h63xx.git/commit
[GFS2] flush the glock completely in inode_go_sync
authorBenjamin Marzinski <bmarzins@redhat.com>
Wed, 2 May 2007 14:44:03 +0000 (09:44 -0500)
committerSteven Whitehouse <swhiteho@redhat.com>
Mon, 9 Jul 2007 07:22:07 +0000 (08:22 +0100)
commitb524fe646c9a226a847e30ca1221dc22e952f16b
tree873b37ab81cadbcc73b02c83cbfd85c1a5ccbee9
parent7dcca30a32aadb0520417521b0c44f42d09fe05c
[GFS2] flush the glock completely in inode_go_sync

Fix for bz #231910
When filemap_fdatawrite() is called on the inode mapping in data=ordered mode,
it will add the glock to the log. In inode_go_sync(), if you do the
gfs2_log_flush() before this, after the filemap_fdatawrite() call, the glock
and its associated data buffers will be on the log again. This means you can
demote a lock from exclusive, without having it flushed from the log. The
attached patch simply moves the gfs2_log_flush up to after the
filemap_fdatawrite() call.

Originally, I tried moving the gfs2_log_flush to after gfs2_meta_sync(), but
that caused me to trip the following assert.

GFS2: fsid=cypher-36:test.0: fatal: assertion "!buffer_busy(bh)" failed
GFS2: fsid=cypher-36:test.0:   function = gfs2_ail_empty_gl, file = fs/gfs2/glops.c, line = 61

It appears that gfs2_log_flush() puts some of the glocks buffers in the busy
state and the filemap_fdatawrite() call is necessary to flush them. This makes
me worry slightly that a related problem could happen because of moving the
gfs2_log_flush() after the initial filemap_fdatawrite(), but I assume that
gfs2_ail_empty_gl() would catch that case as well.

Signed-off-by: Benjamin E. Marzinski <bmarzins@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/gfs2/glops.c