In theory the current code could cause two to be scheduled
   if something wakes up xprt->snd_task before keventd has
   had a chance to run xprt_sock_connect()

  The old code was wrong in that it assumed that we are out the grace
  period as soon as the client is finished doing lock recovery.
  Also ensure that we respect signals when waiting for the server grace
  period to end.

  portmap can fail due to the call to xprt_close() in xprt_connect():
    xprt_disconnect() wakes up xprt->snd_task, and sets -ENOTCONN,
    which again gets converted to EIO by xprt_connect_status()

  Fix is to remove call to xprt_disconnect(). We don't need it in the
  case when we are reconnecting. However we do need to ensure that we
  wake up xprt->snd_task if reconnection fails.

  Diagnosis & proposed solution by Olaf Kirch

requests for attributes that are not supported. In particular, we wish to
determine whether or not the server supports ACLs.

     length.
RPC: make the client receive xdr_buf return the actual length of the RPC
     length.
NFSv4/RPC: improved checks to prevent XDR reading beyond the actual end of
     the RPC reply.

READDIR RPC calls.

READLINK RPC calls.

sillyrename() completion.

STATFS RPC calls.

PATHCONF RPC calls.

CREATE RPC calls.

RENAME RPC calls.

REMOVE RPC calls.

mountpoint.

LOOKUP RPC calls.

new attribute revalidation scheme doesn't rely on them.

GETATTR RPC calls.

ACCESS RPC calls.

RPC: Clean up XDR encoding of opaque data.

from over-aggressive attribute cache revalidation.

starting offset when calculating the page length.

bytes than we requested due to resource limitations etc.

If a wait queue is defined as a "priority queue" then requests are dequeued
in blocks of 16 in order to work well with write gathering + readahead on the
server.
There are 3 levels of priority. The high priority tasks get scheduled 16 times
for each time the default level gets scheduled. The lowest level gets
scheduled once every 4 times the normal level gets scheduled.
Original patch contributed by Shantanu Goel.

RPC: remove unnecessary support for sk->sk_sleep on those sockets that are
  owned by the RPC client.

  writepages() method, throttle all writes to that mountpoint.

We default to "long" (which is what POSIX says), but since
a number of architectures have used "int" for historical
reasons, we need to allow overrides.

At least sparc64 needs this. Possibly others, but so far
architecture maintainers haven't spoken up. ppc64 and x86-64
are known to be ok with the default "long".

From Ivan Kokshaysky <ink@jurassic.park.msu.ru>.

into heffalump.twiddle.home:/home/rth/work/linux/axp-2.6