Get rid of le24 stuff, along with the bitfields use - all that stuff
can be done with standard stuff, in sparse-verifiable manner.  Moreover,
that way (shift-and-mask) often generates better code - gcc optimizer
sucks on bitfields...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NDave Kleikamp <dave.kleikamp@oracle.com>
----