[OpenIndiana-discuss] nfsd - TLI error 17

[Marcel Telka]

On Mon, Jun 10, 2013 at 11:37:14PM +0200, Udo Grabowski (IMK) wrote:
> On 10/06/2013 23:23, Italo Santos wrote:
> >Hello all,
> >
> >I have a environment with a OpenIndiana storages and some days ago I faced a issue with the NFS daemon break intermittently. Looking in logs I seen some erros as you see below:
> >
> >Jun  8 00:59:01 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 791759 daemon.error] t_rcvrel(file descriptor 47/transport tcp) Resource temporarily unavailable
> >Jun  8 00:59:08 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 396295 daemon.error] t_rcvrel(file descriptor 36/transport tcp) TLI error 17
> >Jun 10 15:30:10 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 396295 daemon.error] t_rcvrel(file descriptor 68/transport tcp) TLI error 17
> >Jun 10 15:40:12 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 791759 daemon.error] t_rcvrel(file descriptor 15/transport tcp) Resource temporarily unavailable
> >Jun 10 15:45:48 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 396295 daemon.error] t_rcvrel(file descriptor 72/transport tcp) TLI error 17
> >Jun 10 15:54:08 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 791759 daemon.error] t_rcvrel(file descriptor 71/transport tcp) Resource temporarily unavailable
> >Jun 10 16:17:17 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 396295 daemon.error] t_rcvrel(file descriptor 83/transport tcp) TLI error 17
> >Jun 10 17:23:12 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 396295 daemon.error] t_rcvrel(file descriptor 86/transport tcp) TLI error 17
> >Jun 10 17:40:47 hm6525 /usr/lib/nfs/nfsd[12153]: [ID 791759 daemon.error] t_rcvrel(file descriptor 21/transport tcp) Resource temporarily unavailable
> >
> >Anyone knows about this error can help me?
> 
> These a calls to the XTI/TLI transport library, and if I recall
> right, error 17 says it got packages for a connection that was
> already closed.
> This is a very enigmatic error message that I've seen three
> times in my life, two times a broken network connection to
> a DNS server (where NFS asks for the NFS domain), one time
> a broken connection between server and client.
> You have to analyse your network in depth, in all three
> cases we had a hard time to find the culprit (partly broken
> hardware), in one case we didn't find the problem, it disappeared
> as suddenly as it came up.

IIRC, there is a bug in t_listen and/or t_rcvrel which
could cause this (or very similar) issue.

The problem could happen when _t_look_locked() returns T_ERROR.
The fix is to completely remove T_ERROR from TLI/XTI.

This issue is fixed in Solaris 11 (since June or July 2011).


Thanks.

-- 
+-------------------------------------------+
| Marcel Telka   e-mail:   marcel at telka.sk  |
|                homepage: http://telka.sk/ |
|                jabber:   marcel at jabber.sk |
+-------------------------------------------+