Re: [PATCH RFC 02/10] nfsd41: sunrpc: Added rpc server-side backchannel handling

["J. Bruce Fields" <bfields@fieldses.org>]

On Mon, Aug 24, 2009 at 07:42:27PM -0400, Trond Myklebust wrote:
> On Mon, 2009-08-24 at 19:34 -0400, J. Bruce Fields wrote:
> > On Thu, Aug 20, 2009 at 03:34:23AM +0300, Benny Halevy wrote:
> > > From: Rahul Iyer <iyer@netapp.com>
> > > 
> > > Signed-off-by: Rahul Iyer <iyer@netapp.com>
> > > Signed-off-by: Mike Sager <sager@netapp.com>
> > > Signed-off-by: Marc Eshel <eshel@almaden.ibm.com>
> > > Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> > > Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
> > 
> > This patch needs an ACK from Trond.
> > 
> > > 
> > > When the call direction is a reply, copy the xid and call direction into the
> > > req->rq_private_buf.head[0].iov_base otherwise rpc_verify_header returns
> > > rpc_garbage.
> > 
> > Looks mostly OK, though blocking the client rpciod on the
> > bc_send_request method may be a problem--rpciod normally tries not to
> > sleep, and the other send_request methods look like they avoid it.
> 
> Agreed. Blocking on sending is unacceptable inside rpciod. Please either
> use non-blocking I/O, or use a different thread context for this.

We did some work to avoid having to spawn a thread on the server for
every recall, and I'd still prefer to avoid that.

But I'm not sure what's required to make the server send routine
non-blocking.

--b.