[GFS2] Data corruption fix
authorWendy Cheng <wcheng@redhat.com>
Tue, 18 Sep 2007 13:19:13 +0000 (09:19 -0400)
committerSteven Whitehouse <swhiteho@redhat.com>
Wed, 10 Oct 2007 07:56:26 +0000 (08:56 +0100)
commitde986e859a29097fb9211b052d86a9a2c868f6cd
tree85c97fe22e0eafb3c7440c8ff21fe73d8c456960
parent16615be18cadf53ee6f8a4f0bdd647f0753421b1
[GFS2] Data corruption fix

* GFS2 has been using i_cache array to store its indirect meta blocks.
Its flush routine doesn't correctly clean up all the entries. The
problem would show while multiple nodes do simultaneous writes to the
same file. Upon glock exclusive lock transfer, if the file is a sparse
file with large file size where the indirect meta blocks span multiple
array entries with "zero" entries in between. The flush routine
prematurely stops the flushing that leaves old (stale) entries around.
This leads to several nasty issues, including data corruption.
* Fix gfs2_get_block_noalloc checking to correctly return EIO upon
unmapped buffer.

Signed-off-by: Wendy Cheng <wcheng@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/gfs2/meta_io.c
fs/gfs2/ops_address.c