HELP: kernel IP fragmentation reassembly problem?

Hi,

We've got a UDP-based network loader that is used for analysis of radio
communications systems. In short is sends scheduled bursts of UDP
datagrams with tx timestamp in the payload and the UDP servers receive
the datagrams and record the rx timestamp. The UDP datagrams range from
100 bytes up to (65535 - 28).

We're using the latest Red Hat 9 kernel 2.4.20-20.9 and we first noticed
that the IP header had the Don't Fragment (DF) bit set all the time.
Also the More Fragments (MF) bit was set for packets larger than the
MTU...which is normal. So we turned on the "ip_no_pmtu_disc" (set to 1)
and the DF flag is no longer set for all packets.

Now the problem is that the my UDP server process running on another RHL
9 box does not receive any packets that are over the MTU. I see the IP
packets get fragmented but my UDP server doesn't get anything back from
the recvfrom().

Any ideas of other flags to turn on/off in /proc?

Thanks,
Tuan

Tuan <(E-Mail Removed)> wrote:

[large UDP packets and PMTU discovery disabled]

> Now the problem is that the my UDP server process running on another RHL
> 9 box does not receive any packets that are over the MTU. I see the IP
> packets get fragmented but my UDP server doesn't get anything back from
> the recvfrom().
>
> Any ideas of other flags to turn on/off in /proc?

First check what the problem exactly is. Use tcpdump or
ethereal to check whether all fragments of the packets
are hitting your box. Also pay attention whether there
are ICMP error messages indicating that Time-to-Live
is exceeded sent.

Make sure that no strange packet filter rule is blocking
the fragmented UDP packets.

You might also tweak the /proc/sys/net/ipv4/ipfrag*
files in try-and-error manner. ;-)


Ciao, Horst
--
»When pings go wrong (It hurts me too)« E.Clapton/E.James/P.Tscharn