Highlights include:
* A fix for an RPCSEC_GSS hang, in which some servers would treat
the client as being crazy if it tried to destroy an RPCSEC_GSS
credential that had already been labelled as invalid by an
RPCSEC_GSS_CREDPROBLEM error in an RPC reply.
* I found a use-after-free case in call_decode() that was
introduced by commit 220bcc2afd7011b3e0569fc178331fa983c92c1b.
* I added code to ensure that we only break the connection _once_
if we have to resend multiple NFSv4 requests.
* I also tried to ensure that we don't invalidate a gss_cred which
may be in use by other requests (basically, by forcing
gss_refresh to always create a new credential if we find out
that our current one has been invalidated).
* I cleaned up xs_nospace() so that our use of the
transport->sock->flags conforms to the expectations of the rest
of the socket code.
* I've reintroduced the use of machine creds when calling
SETCLIENTID in NFSv4. If such a cred exists, then that will
ensure that we always authenticate to the same principal name
when rebooting, thus improving the ability to recover from
network partitions.
If there are no machine creds available, then we fall back to
the existing scheme where we take a user cred.
* NLM fixes to avoid creating 'phantom locks' on the client side.
* Fix error handling in the read/write code if rpc_run_task() fails.
Cheers
Trond