Hi there, A linux-2.2(.16) kernel with NAT routing enabled can easily be "paniced". Assume that we have an ethernet interface eth0 with address 172.17.1.1/16, and that no default gateway exists. Now insert a NAT route: # ip route add nat 172.17.1.254 via 1.1.1.1 And add a default gateway: # ip route add default via 172.17.1.254 If there is some activity on the network, the kernel will probably soon panic (I know that this is a braindead combination of things, but it seems to me that panic is a bit drastic reaction). What happens is, that due to the NAT route, the default route will not be associated with a device. Sooner or later we will hit "out_dev = FIB_RES_DEV(res)->ip_ptr" in ip_route_input_slow(skb,...,dev) with a res that does not refer to a device (=>panic). I have not been able to pinpoint the exact origin of trouble, as there are a few things that are not clear to me: Should a NAT rule map to a device (the kernel does not accept e.g. "ip route add nat ... dev eth0")? Should a res e.g. after fib_lookup(&key,&res) always contain a valid device reference? Any suggestions/comments? Regards, Kim Woelders - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
