8.0-RELEASE network failure

B

bsdbigot

[font="Verdana"]Hi, all. I have a 8.0-RELEASE machine (x86) which is a gateway and file server. I use ipnat/ipfilter for firewall and samba-3 and nfs-3 for file serving.

Network interfaces are nfe0 (internal, on motherboard) and sk0 (PCI).

Recently, every 3 days or so, the internal interface will stop accepting new traffic - resulting in a scenario which to all other internal machines appears that this box has crashed.

It turns out, the external interface stays lit, for a while at least. Last night, it also stopped responding to new traffic.

In both scenarios, I'm able to log on to the console, stop and start interfaces (which has no effect), reload firewall/nat rules (which has no effect). I can start a ping and I can see the switch port light up with the ping (though I haven't tcpdump'd to verify) but all pings time out.

There are no errors on console or in /var/log/messages which indicate that anything is wrong. The box is quite happy thinking that everything is A-OK. So, I reboot, and we repeat in 3 days.

This most recent time I dumped sysctl -a to a file before reboot, but apparently that's too long for a forum post.[/font]
 
bsdbigot said:
I can start a ping and I can see the switch port light up with the ping (though I haven't tcpdump'd to verify) but all pings time out.
Click to expand...
Perhaps the problem is with the switch?
 
SirDice said:
Perhaps the problem is with the switch?
Click to expand...

I highly doubt it - there are 2 different switches, one to the nfe0 and one to the sk0. Early in the "failure," I'm still able to utilize the sk0 interface.

The likelihood of 2 switch ports from 2 different switches near-simultaneously failing, and then working after reboot of only the PC has got to be on the order of 0.
 
bsdbigot said:
This looks promising... I will report back if it works out. Thanks.
Click to expand...

I'm going on 5 days of uptime and my mbufs in use (first line of netstat -m) is stable. Before the patch, I was lucky to see 3 days of uptime and mbufs growing constantly.

For posterity, somebody did commit this patch to -STABLE.
 
You must log in or register to reply here.
Back
Top