NFS over TCP idle timeout

NFS over TCP idle timeout

[Paul B. Henson]

Does the Linux server implementation of NFS over TCP implement an idle
timeout that automatically disconnects clients after some amount of
inactivity?

I was testing NFSv4 from a Solaris 10 client to a Linux server (Gentoo,
2.6.20). The Solaris client does not tear down the TCP connection upon
system shutdown, which leaves an orphaned established TCP connection on the
server. The connection remained in the established state for days.

Sun support indicates their server automatically disconnects idle clients,
and don't consider not closing the client connection upon shutdown a bug.

I didn't find any documentation regarding the Linux implementation and idle
timeouts, although I did find the following code in net/sunrpc/svcsock.c:

-----
/* apparently the "standard" is that clients close
 * idle connections after 5 minutes, servers after
 * 6 minutes
 *   http://www.connectathon.org/talks96/nfstcp.pdf
 */
static int svc_conn_age_period = 6*60;
-----

that would seem to indicate it should close the connection after some
amount of idle time? However, empirically that doesn't seem to happen.

Should idle connections automatically be closed? Are there any tunables or
switches that need to be touched to enable that?

Thanks...


-- 
Paul B. Henson  |  (909) 979-6361  |  http://www.csupomona.edu/~henson/
Operating Systems and Network Analyst  |  henson@csupomona.edu
California State Polytechnic University  |  Pomona CA 91768

Re: NFS over TCP idle timeout

[J. Bruce Fields]

On Fri, Jan 25, 2008 at 04:41:00PM -0800, Paul B. Henson wrote:
> 
> Does the Linux server implementation of NFS over TCP implement an idle
> timeout that automatically disconnects clients after some amount of
> inactivity?
> 
> I was testing NFSv4 from a Solaris 10 client to a Linux server (Gentoo,
> 2.6.20). The Solaris client does not tear down the TCP connection upon
> system shutdown, which leaves an orphaned established TCP connection on the
> server. The connection remained in the established state for days.
> 
> Sun support indicates their server automatically disconnects idle clients,
> and don't consider not closing the client connection upon shutdown a bug.
> 
> I didn't find any documentation regarding the Linux implementation and idle
> timeouts, although I did find the following code in net/sunrpc/svcsock.c:
> 
> -----
> /* apparently the "standard" is that clients close
>  * idle connections after 5 minutes, servers after
>  * 6 minutes
>  *   http://www.connectathon.org/talks96/nfstcp.pdf
>  */
> static int svc_conn_age_period = 6*60;
> -----
> 
> that would seem to indicate it should close the connection after some
> amount of idle time? However, empirically that doesn't seem to happen.
> 
> Should idle connections automatically be closed? Are there any tunables or
> switches that need to be touched to enable that?

That sounds like a server bug to me.  Have you checked whether it still
behaves that way on more recent kernels?

--b.

Re: NFS over TCP idle timeout

[Paul B. Henson]

On Mon, 28 Jan 2008, J. Bruce Fields wrote:

> On Fri, Jan 25, 2008 at 04:41:00PM -0800, Paul B. Henson wrote:
> >
> > Does the Linux server implementation of NFS over TCP implement an idle
> > timeout that automatically disconnects clients after some amount of
> > inactivity?
>
> That sounds like a server bug to me.  Have you checked whether it still
> behaves that way on more recent kernels?

No. I'll probably be upgrading the server to 2.6.23 within the next few
weeks, but given the dearth of information I found in my searches, was
curious whether or not it was supposed to in the first place.

Thanks...


-- 
Paul B. Henson  |  (909) 979-6361  |  http://www.csupomona.edu/~henson/
Operating Systems and Network Analyst  |  henson@csupomona.edu
California State Polytechnic University  |  Pomona CA 91768

Re: NFS over TCP idle timeout

[Greg Banks]

On Fri, Jan 25, 2008 at 04:41:00PM -0800, Paul B. Henson wrote:
> 
> Does the Linux server implementation of NFS over TCP implement an idle
> timeout that automatically disconnects clients after some amount of
> inactivity?
>  [...]
> Should idle connections automatically be closed? Are there any tunables or
> switches that need to be touched to enable that?

Try applying this

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=7a1fa065a0264f6b3d3003ba5635289f6583c478

-- 
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
The cake is *not* a lie.
I don't speak for SGI.

Re: NFS over TCP idle timeout

[Paul B. Henson]

On Tue, 29 Jan 2008, Greg Banks wrote:

> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=7a1fa065a0264f6b3d3003ba5635289f6583c478

Ah, yes, the 2.6.20 kernel I am currently running does have the change from
aaf68cfbf2241d24d46583423f6bff5c47e088b3 introducing the bias to sk_inuse.
This is probably the problem I am running into. The 2.6.23 kernel I will be
upgrading to within a week or two appears to have this fix already applied,
so hopefully that upgrade will resolve my issue.

Thanks much for the information...


-- 
Paul B. Henson  |  (909) 979-6361  |  http://www.csupomona.edu/~henson/
Operating Systems and Network Analyst  |  henson@csupomona.edu
California State Polytechnic University  |  Pomona CA 91768