Re: Fw: Deadlock regression in v2.6.31.6

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Thu, 2009-11-26 at 16:07 +0100, Stephen R. van den Berg wrote: 
> On Thu, Nov 26, 2009 at 16:01, Trond Myklebust
> <trond.myklebust@fys.uio.no> wrote:
> > On Thu, 2009-11-26 at 00:11 +0100, Stephen R. van den Berg wrote:
> >> 1.2.3.167 is the Linux client kernel which locks up, 1.2.3.151 is the
> >> unfs server.
> >> It looks like the client terminates the TCP connection.  The server
> >> confirms it, the client then sends a final acknowledge.  At that point
> >> the client kernel locks up in the infinite loop.
> 
> > OK. Have you tried running with RPCDBG_TRANS debugging enabled? I
> > suspect you might see a flood of 'sendmsg returned unrecognized error'
> > or 'connect returned unhandled error' messages if you do.
> 
> The pcap trace is not conclusive (enough)?
> I can run with RPCDBG_TRANS enabled, if it is needed to further
> pinpoint the problem.

The pcap trace shows what is happening: the socket is getting closed
correctly, and so the RPC client needs to initiate a reconnection before
it can transmit again. What I don't understand is why it is failing to
do so...

   Trond