NIC dropping inbound traffic after some time - CentOS 5 - BroadCom BCM5721
 

Hey guys,

Here is my issue. I'm setting up a new server and came accros an odd issue. For some reason, after a few minutes, my NIC will stop anwsering to INBOUND traffic, that is, anything coming from the outside world to it. If I ping the server, it will tell me that all is good, but that is ICMP. If I start to sniff the packets on the interface, I realize that NOTHING is sniffed execpt for "spanning-tree". Ssh wont go through, DNS queries, etc, all dropped even before I can sniff them with WireShark/Ethereal.

Now, if I go on the server, everything is working correctly. I can navigate the internet, ssh to any servers I want. Then, the fun part, if I do that, and use my other computer to ping the server, I can sniff the ICMP packets and i can SSH to it. Then, my dns receives the query once again.

Looks like there should be somekind of "keep-alive" flag on the NIC I missed. Or it could be driver issue.

Can someone point me to troubleshooting tips ? Anyone seen this ? I must say my troubleshooting skills on linux are very good, so any help will be appreciated.

* edit *
Iptables is NOT running and I tried with SELinux enabled/disabled

Which kernel are you using?

I'm using the 2.6.18 Kernel, straight from the CentOS 5 install with all the updates automatically provieded by yum.

How is the server's IP address specified? DHCP? Lease time?

The server has 2 NIC, one has a static ip adress, the other one is not used right now but will be statically assigned too.

I tried assigning the static ip to the other NIC, same problem has before.
I'm also configuring a PXE-BOOT and getting strange "PXE-E11: ARP timeout". I have the same dhcpd.conf running on another server and the same boxes will boot just fine with pxe...

Any ideas ?

Well, i'll give an update since I cant find a solution to my problem.

I deceided to install CentOS on another server box, just to see. I installed CentOS 5 on a Dell 1750, same dhcp configuration, SELinux and iptables activated and I have no problems at all. Looks like the NIC or the driver for it is the issue. When i'll have time to test more, i'll switch my two NIC and make sure it's not the card that is defect.

Also note that on the 1750, i dont get the "ARP Timeout" message on a box that uses it has a PXE server. At first I tought my configuration was the issue, but now, I have the certainty that the issue is the coming from the BCM5721 or the TG3 drivers for it. Is it CentOS 5 ? Is it the driver, I cant tell.