2000-11-07 22:28:24

by Sasi Peter

[permalink] [raw]
Subject: Stange NFS messages - 2.2.18pre19

Hi!

I'm getting this under moderate NFS load:
Nov 6 17:39:56 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 6 17:40:08 iq kernel: svc: unknown program 100227 (me 100003)
Nov 6 19:06:11 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 6 19:38:48 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 6 19:54:51 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 6 20:08:53 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 6 20:53:23 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 09:03:28 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 09:15:18 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 11:12:51 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 18:01:17 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 19:34:59 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 19:40:59 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)
Nov 7 21:43:48 iq kernel: svc: server socket destroy delayed (sk_inuse:
1)

What do these means? Is this a kernel bug?

-- SaPE

Peter, Sasi <[email protected]>


2000-11-09 02:16:39

by Scott McDermott

[permalink] [raw]
Subject: Re: Stange NFS messages - 2.2.18pre19

Sasi Peter on Tue 7/11 23:28 +0100:
> I'm getting this under moderate NFS load:
> Nov 6 17:39:56 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
> Nov 6 17:40:08 iq kernel: svc: unknown program 100227 (me 100003)
> Nov 6 19:06:11 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
> Nov 6 19:38:48 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
>
> What do these means? Is this a kernel bug?

Your Suns are using TCP mounts, this got introduced into 2.2.18
somewhere and is a bit broken, do a patch -R with
ftp://oss.sgi.com/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif and
these go away. Suns try TCP mounts first. Be careful to unmount them
first or they will hang waiting for the TCP server to come back up.

2000-11-09 06:57:08

by Ion Badulescu

[permalink] [raw]
Subject: Re: Stange NFS messages - 2.2.18pre19

On Wed, 8 Nov 2000 21:15:38 -0500, Scott McDermott <[email protected]> wrote:
> Sasi Peter on Tue 7/11 23:28 +0100:
>> I'm getting this under moderate NFS load:
>> Nov 6 17:39:56 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
>> Nov 6 17:40:08 iq kernel: svc: unknown program 100227 (me 100003)
>> Nov 6 19:06:11 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
>> Nov 6 19:38:48 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
>>
>> What do these means? Is this a kernel bug?
>
> Your Suns are using TCP mounts, this got introduced into 2.2.18
> somewhere and is a bit broken, do a patch -R with
> ftp://oss.sgi.com/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif and
> these go away. Suns try TCP mounts first. Be careful to unmount them
> first or they will hang waiting for the TCP server to come back up.

I really really think this should be backed out -- or at the very least
disabled. The code wasn't part of the dhiggen merge, it wasn't tested,
and it doesn't work well. Heck, it's still experimental and not recommended
in 2.4.0-test.

What's worse, it will burn everybody out there who is using am-utils or
an automounter that tries TCP mounts first. NFS/UDP server support in 2.2.18
is so much better than in previous versions, it would a shame to ruin
it with this ill-fated patch.


Ion

--
It is better to keep your mouth shut and be thought a fool,
than to open it and remove all doubt.

2000-11-09 07:33:46

by Sasi Peter

[permalink] [raw]
Subject: Re: Stange NFS messages - 2.2.18pre19

On Wed, 8 Nov 2000, Scott McDermott wrote:

> Sasi Peter on Tue 7/11 23:28 +0100:
> > I'm getting this under moderate NFS load:
> > Nov 6 17:39:56 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
> > Nov 6 17:40:08 iq kernel: svc: unknown program 100227 (me 100003)
> > Nov 6 19:06:11 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
> > Nov 6 19:38:48 iq kernel: svc: server socket destroy delayed (sk_inuse: 1)
> > What do these means? Is this a kernel bug?
> Your Suns are using TCP mounts, this got introduced into 2.2.18
> somewhere and is a bit broken, do a patch -R with
> ftp://oss.sgi.com/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif and
> these go away. Suns try TCP mounts first. Be careful to unmount them
> first or they will hang waiting for the TCP server to come back up.

Broken link:
[[email protected] patches]# wget
ftp://oss.sgi.com/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif
--08:31:28--
ftp://oss.sgi.com:21/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif
=> `nfs_tcp-2.2.17.dif'
Connecting to oss.sgi.com:21... connected!
Logging in as anonymous ... Logged in!
==> TYPE I ... done. ==> CWD www.projects/nfs3/download ...
No such directory `www.projects/nfs3/download'.

-- SaPE

Peter, Sasi <[email protected]>

2000-11-09 11:58:26

by Klaus Naumann

[permalink] [raw]
Subject: Re: Stange NFS messages - 2.2.18pre19

Sasi Peter wrote:
> Broken link:
> [[email protected] patches]# wget
> ftp://oss.sgi.com/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif
> --08:31:28--
> ftp://oss.sgi.com:21/http://www.projects/nfs3/download/nfs_tcp-2.2.17.dif
> => `nfs_tcp-2.2.17.dif'
> Connecting to oss.sgi.com:21... connected!
> Logging in as anonymous ... Logged in!
> ==> TYPE I ... done. ==> CWD www.projects/nfs3/download ...
> No such directory `www.projects/nfs3/download'.


http://oss.sgi.com/projects/nfs3/download/nfs_tcp-2.2.17.dif


KN

--
Klaus Naumann (mailto:[email protected] for personal contacts)
http://www.mgnet.de/
Phone: ++49/8761727852

2000-11-09 13:19:53

by Alan Cox

[permalink] [raw]
Subject: Re: Stange NFS messages - 2.2.18pre19

> I really really think this should be backed out -- or at the very least
> disabled. The code wasn't part of the dhiggen merge, it wasn't tested,
> and it doesn't work well. Heck, it's still experimental and not recommended
> in 2.4.0-test.

Its now an experimental option in my pre21 build tree. So people can play with
it and debug it if they wish