xfs: fix recursive splice read locking with DAX
authorDave Chinner <dchinner@redhat.com>
Mon, 4 Jan 2016 05:28:25 +0000 (16:28 +1100)
committerDave Chinner <david@fromorbit.com>
Mon, 4 Jan 2016 05:28:25 +0000 (16:28 +1100)
commita6d7636e8d0fd94fd1937db91d5b06a91fa85dde
tree41b326d5e1ee03a7948c692064002d18282d3fe2
parent3b0fe47805802216087259b07de691ef47ff6fbc
xfs: fix recursive splice read locking with DAX

Doing a splice read (generic/249) generates a lockdep splat because
we recursively lock the inode iolock in this path:

SyS_sendfile64
do_sendfile
do_splice_direct
splice_direct_to_actor
do_splice_to
xfs_file_splice_read <<<<<< lock here
default_file_splice_read
vfs_readv
do_readv_writev
do_iter_readv_writev
xfs_file_read_iter <<<<<< then here

The issue here is that for DAX inodes we need to avoid the page
cache path and hence simply push it into the normal read path.
Unfortunately, we can't tell down at xfs_file_read_iter() whether we
are being called from the splice path and hence we cannot avoid the
locking at this layer. Hence we simply have to drop the inode
locking at the higher splice layer for DAX.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Tested-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
fs/xfs/xfs_file.c