I've run into the problem where Netfilter is not putting packets into
established state (where they should be).
Basically, I have a setup that consist of two testing machines acting as
VPN gateways. Machine A with IP addresses of 192.168.1.1 and
192.168.2.1, and machine B with IP address of 192.168.3.1 and
192.168.4.1. The 1.1 and 3.1 act as "external" IP addresses. 2.1 and
4.1 are on the "local networks" that I want to connect with a tunnel.
I've setup GRE tunnel between them like this (on machine A, machine B
has the same thing in opposite direction, the route is to
192.168.2.0/24, and the interface name is neta):
# modprobe ip_gre
# ip tunnel add netb mode gre remote 192.168.3.1 local 192.168.1.1 ttl 255
# ip link set netb up
# ip addr add 10.0.0.1 dev netb
# ip route add 192.168.4.0/24 dev netb src 192.168.2.1
Then I defined IPSec policy between 192.168.1.1 and 192.168.3.1,
transport mode, required, using racoon for automatic keying.
If I ping 192.168.4.1 from 192.168.2.1 everything works fine. Packets
go into the GRE tunnel, which than goes over IPSec.
If I ping 192.168.3.1 from 192.168.1.1, things do not work. After lots
of debugging (basically logging all traffic in all tables/chains, and
looking at the output of iptables -nvxL), it seems that returning
packets are not placed into established state.
On machine B, I have this set of rules:
# standard stuff to handle established connections
-A INPUT -m state --state ESTABLISHED,RELATED -j ACCEPT
-A OUTPUT -m state --state ESTABLISHED,RELATED -j ACCEPT
-A FORWARD -m state --state ESTABLISHED,RELATED -j ACCEPT
# gre/ipsec/ike stuff
-A INPUT -i eth0 -p esp -s 192.168.1.1 -d 192.168.3.1 -j ACCEPT
-A OUTPUT -o eth0 -p esp -s 192.168.3.1 -d 192.168.1.1 -j ACCEPT
-A INPUT -i eth0 -p gre -s 192.168.1.1 -d 192.168.3.1 -j ACCEPT
-A OUTPUT -o eth0 -p gre -s 192.168.3.1 -d 192.168.1.1 -j ACCEPT
-A INPUT -i eth0 -p udp --sport 500 --dport 500 -s 192.168.1.1 -d
192.168.3.1 -j ACCEPT
-A INPUT -i eth0 -p udp --sport 500 --dport 500 -s 192.168.3.1 -d
192.168.1.1 -j ACCEPT
# rules to allow ping from the other end (direct and through the tunnel)
-A INPUT -i eth0 -p icmp --icmp-type ping -s 192.168.1.1 -d 192.168.3.1
-m state --state NEW -j ACCEPT
-A INPUT -i neta -p icmp --icmp-type ping -s 192.168.2.1 -d 192.168.4.1
-m state --state NEW -j ACCEPT
# log everything before dropping it
-A INPUT -j LOG --log-prefix "INPUT "
-A OUTPUT -j LOG --log-prefix "OUTPUT "
-A FORWARD -j LOG --log-prefix "FORWARD "
As I wrote, when I do "ping -c 1 192.168.3.1" from machine A using
192.168.1.1 as source address, and looking at the output of iptables
-nvxL, I can see counter incremented for ESP packet, counter for ICMP
ping is incremented, but the ICMP pong is not cought by the second rule
(accept all established/related), and falls to the bottom of the chain
(gets logged and dropped).
On the other hand, if I'm doing "ping -c 192.168.4.1" from machine A
using 192.168.2.1 as source address, everything just works (and I see
counters incremented as expected).
I'm not using IPSec in tunneling mode to get around some other problems
I have. This approach also gives me interfaces that represent tunnel
endpoints that I can use for routing, something I don't have with native
Linux IPSec in tunneling mode. Plus it seems doing it this way might at
the end simplify my firewall rules (again, because there's an interface
representing tunnel endpoint).
