Hey Guys,
I'm just after a little advice on where the best place to get 
assistance/report a kernel bug, whats the best information to include, and 
any other etiquette to make it as easy for the developers to solve as 
possible.

The issue is we have a multihomed debian server, which I have upgraded to 
kernel 2.6.22.2 in an attempt to solve this problem (I don't think the 
latest kernels 2.6.22.3 and up have fixed the problem after going over the 
changelogs)

Our network setup is as follows:
eth0 - LAN
eth1 - Layer 3 ADSL (This is always on)
eth2 - link for Layer 3 ADSL
ppp0 - Layer 2 ADSL (this tends to go down a few times a day - but comes 
back up on next retry)
tun0 - l2tpns for Windows Roadwarriors.

We have ipsec with multiple peer to peer and roadwarrior setups using 
openswan 2.4.7 with the kernel's ipsec modules (not KLIPS).
I use shorewall for iptables, and traffic control (tc) although I don't do 
any rate limiting, just use tc for routing.
All ipsec communications are routed out eth1 using tc rules.
I have also removed all ipv6 modules from the kernel to make sure that is 
not the cause, with no success.

I can replicate the crash by:
1. bringing up the ppp0 interface.
2. restarting ipsec.
3. having one of our peers make a connection with us. (haven't tried with 
a Roadwarrior yet)
4, bring down ppp0 interface.

This gives the kernel error: kernel: unregister_netdevice: waiting for 
ppp0 to become free. Usage count = 1
It eventually crashes the server, but freezes pppd, ifconfig and ip 
commands with a stat of D (ps describes this as "Uninterruptible sleep").

I can stop the lock, hence freeing up all the commands, including pppd by 
running:
# killall _plutorun _plutoload pluto _pluto_adns

The only the pluto process isn't killed, but then running:
# killall _updown

Corrects the issue.

Any advice or assistance would be welcome.

TIA,

Scott
-- 
SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/
Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html

Reply via email to