Re: problems with ip_conntrack on Linux 2.6.x (x>5)

I just tested this and it seems to work fine.

  Both client and server are Debian boxes.  Both machines are running
  the CVS version of SFS checked out on 30 June 2004.

  Client is running a custom 2.6.6 kernel (but it only slightly differs
  from the stock Debian kernel config in that the APM options are
  tweaked and MTRR is turned off).

  Server is running a stock Debian 2.6.7 kernel.

I did a 'modprobe ip_conntrack' on both the client and the server;
started sfssd on the server and sfscd on the client; ran my agent;
changed to an SFS-exported directory on the server.

Nothing abnormal to report.  No relevant SFS or kernel debug
messages (except ip_conntrack module loaded successfully/SFS started
successfully).  I don't seem to have any problems traversing, reading or
writing the file system.  /proc/net/ip_conntrack also seems reasonable.

Does it look like I've done anything different in my setup from what you
have?  What version of SFS are you running?  There have been a number of
changes recently, particularly if you're running 0.7.2 still.

Michael

Clint Adams [[email protected]] wrote:
> With Linux 2.6.6 or later, neither the SFS client nor the server is
> usable when the ip_conntrack module is loaded.
> 
> The message is
> ip_conntrack_in: Frag of proto 17 (hook=0)
> 
> and the NFS mount keeps hanging (with occasional short bursts of data
> transfer).
> 
> If one uses Linux 2.6.5 with ip_conntrack, or later kernels without
> ip_conntrack, SFS behaves fine.
> 
> Jaako Niemi has pointed out that a localhost NFS mount produces the same
> effect.
> 
> I have tried to use 2.6.7's raw table and NOTRACK target, but I am
> either usinng it incorrectly or it doesn't make a difference.
> 
> This threatens to make SFS nonviable for me.  Can anyone help?