Synopsis: link() call can be raced resulting in false success return

State-Changed-From-To: closed->open
State-Changed-By: tls@netbsd.org
State-Changed-When: Tue, 30 Nov 2004 00:34:25 +0000
State-Changed-Why:
The problem in fact still exists.  Brian Marcotte at Panix points out that,
in fact, the problem is that the code -- still present in 2.0 -- does not
even attempt to keep track of retransmissions; thus a retransmission can
succeed even after another link syscall completed (successfully) in the
middle.

Brian _thinks_ this may be fixed in post-2.0 -current.  I'm hoping he will
respond pointing out the specific change that may actually be a fix.

Do not close this PR again without moving it to feedback state and
confirming that it cannot be reproduced *at Panix*.  Panix has a NFS
environment that has repeatedly exhibited this problem when others'
tests have failed to reproduce it.