On Thu, 2008-10-16 at 10:33 -0700, Pranav Desai wrote:
On Thu, Oct 16, 2008 at 5:37 AM, Balazs Scheidler <bazsi@balabit.hu> wrote:
On Wed, 2008-10-15 at 23:22 -0700, Pranav Desai wrote:
On Wed, Oct 15, 2008 at 5:29 PM, Pranav Desai <pranavadesai@gmail.com> wrote:
Hello,
We have a http proxy server which is in full transparent mode using tproxy 2.0.6 + kernel 2.6.20.15.
The iptables rule to redirect port 80 traffic from clients is:
574K 34M REDIRECT tcp -- * * 0.0.0.0/0 0.0.0.0/0 tcp dpts:80:82 redir ports 8001
8001 is our proxy port. We use tproxy to send connections to web servers using the clients IP. Using setsockopt(TPROXY_ASSIGN) ...
The server is at around 240req/s. Most of the time its working fine i.e. I see a lot of traffic on port 80 going out using the clients IP, but I see quite a few pkts going out using src port 8001 (trace below). The rate is not as high port 80 traffic, but still high enough to be a concern.
The interesting thing is I don't see any pkts destined to port 8001. I have been monitoring this server for an entire day. Hence, I am guessing that at times for some reason tproxy is not able to assign the foreign address and just uses the local address:port ...
Any ideas or suggestions on how I should go about debugging this. Or whether I should be looking somewhere other than the tproxy module.
Its a production server, so I can't get the client side traces, but I can get more info from the server. Let me know if you need any other information.
I would really appreciate any help I can get.
Thanks -- Pranav
The 10.10.224.6 is the server IP running tproxy. The 10.1.x.x addresses are the client address.
So the return traffic or the response seems to be having the problem where sometimes the src is 10.10.224.6:8001 instead of the origin server IP:PORT.
Are you using bridging as well? Do you have CONFIG_NETFILTER_BRIDGE enabled in your kernel?
There is no bridging, only bonding.
because in that case tcpdump will see the unnated traffic, bridging plays some nasty games with netfilter.
we thought that tcpdump might be screwing up somewhere, but even on external network elements (load balancer) we don't see anything going to the tproxy server on port 8001.
On the client side of the proxy, the TPROXY_ASSIGN stuff does not really matter, only when going to the server.
Hmm. so does the netfilter nat rules control the client-side, and could that be screwing up somehow. Can the conntrack table be of some assistance here ?
Yes, it'd be useful to see the conntrack entry of the invalid packets. -- Bazsi