RPC: Fix the TCP resend semantics for NFSv4
commit241c39b9ac4bf847013aa06cce6d4d61426a2006
authorTrond Myklebust <Trond.Myklebust@netapp.com>
Fri, 20 Apr 2007 20:12:55 +0000 (20 16:12 -0400)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Sat, 21 Apr 2007 05:56:30 +0000 (20 22:56 -0700)
tree6455884e65288e64576555a27c06a8ee2f1f6371
parent2b82f190c81bf1524447c021df4e9ce8ef379bd5
RPC: Fix the TCP resend semantics for NFSv4

Fix a regression due to the patch "NFS: disconnect before retrying NFSv4
requests over TCP"

The assumption made in xprt_transmit() that the condition
"req->rq_bytes_sent == 0 and request is on the receive list"
should imply that we're dealing with a retransmission is false.
Firstly, it may simply happen that the socket send queue was full
at the time the request was initially sent through xprt_transmit().
Secondly, doing this for each request that was retransmitted implies
that we disconnect and reconnect for _every_ request that happened to
be retransmitted irrespective of whether or not a disconnection has
already occurred.

Fix is to move this logic into the call_status request timeout handler.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
net/sunrpc/clnt.c
net/sunrpc/xprt.c