'RELENG_4_2_0_RELEASE'.

Approved: jkh

Approved by: jkh

Fix a bogon..

    KASSERT if the original buffer for a background write was replaced,
    indicating a probable duplicate I/O or passing I/O resulting in
    filesystem corruption.  There are still probable bugs in the I/O
    handling and this will catch them rather then allow the FS to become
    corrupted.

    Remove the BX_BKGRDINPROG test in brelse() that was not working anyway,
    to restore prior NFS functionality (which we also believe does not work
    anyway, but this is safest until we figure out how to fix that whole
    section).

    Fixed bogus_page handling in brelse which could lead to invalidation
    of the wrong page and thus filesystem corruption.

    Fixed foff calculation in biodone().  I/O's with errors or I/O's
    brelse()'d rather then initiated could lead to an attempt to reprocess
    the same b_pages[] element twice, resulting in a 'busy < 0' panic.

Approved: jkh
Reviewed-by: mckusick, Ian Dowse <iedowse@maths.tcd.ie>
Submitted-by: dillon, mckusick, Ian Dowse <iedowse@maths.tcd.ie>

It should be `device stg0 at ...'.

Approved by:    jkh@FreeBSD.org
Pointed out by: Archie Cobbs <archie@packetdesign.com>

get better blocking semantics. This update does virtually nothing
except for providing better returns to ioctl() calls.

OK'ed by:	jkh

we could hose the lnc driver when we reject a chip that the lnc driver
wants. (Once in 32-bit mode, only a hardware reset can restore 16-bit
mode.)

Ok'ed by: jkh