I’m running 3 Raspberry Pi (running Stretch) VMs with VirtualBox and I have configured them all to be on one internal network with nothing else but these 3 VMs. They have the following IP addresses: 172.25.1.12, 172.25.1.13, 172.25.1.14. I’m trying to configure them so that when 172.25.1.12 sends packets to 172.25.1.14 the packets are routed through 172.25.1.13. And likewise when 172.25.1.14 sends packets to 172.25.1.12 the packets are also routed through 172.25.1.13.

First this is what I have done on 172.25.1.12:
sudo ip route add 172.25.1.14/32 via 172.25.1.13 dev eth0

Therefore doing an "ip route show" will show the following:
pi@raspberry:~ $ ip route show
172.25.1.0/24 dev eth0 proto kernel scope link src 172.25.1.12 metric 202
172.25.1.14 via 172.25.1.13 dev eth0

I did something similar on 172.25.1.14:
sudo ip route add 172.25.1.12/32 via 172.25.1.13 dev eth0

pi@raspberry:~ $ ip route show
172.25.1.0/24 dev eth0 proto kernel scope link src 172.25.1.14 metric 202
172.25.1.12 via 172.25.1.13 dev eth0

Also I needed to enable forwarding on the middle node i.e. 172.25.1.13:
In /etc/sysctl.conf ensure this line exists:
net.ipv4.ip_forward=1

And the run this:
sudo sysctl -p

I had a packet sniffer running on the middle node (i.e. 172.25.1.13) just to see if traffic was being passed through and sure enough pings and iperf traffic was routing through 172.25.1.13. However then it wouldn’t route through 172.25.1.13 any more and instead route it directly by skipping 172.25.1.13. And I don’t know why. I run traceroute and sometimes it will show the path through 172.25.1.13 and sometimes it will skip 172.25.1.13 only to go back through routing through 172.25.1.13. And I’m not really doing anything to the devices in that only time is passing.

Sometimes it’ll be good:
pi@raspberry:~ $ !trace
traceroute 172.25.1.12
traceroute to 172.25.1.12 (172.25.1.12), 30 hops max, 60 byte packets
1 172.25.1.13 (172.25.1.13) 0.340 ms 0.189 ms 0.185 ms
2 172.25.1.12 (172.25.1.12) 0.446 ms 0.447 ms 0.330 ms

pi@raspberry:~ $ !trace
traceroute 172.25.1.14
traceroute to 172.25.1.14 (172.25.1.14), 30 hops max, 60 byte packets
1 172.25.1.13 (172.25.1.13) 0.330 ms 0.193 ms 0.189 ms
2 172.25.1.14 (172.25.1.14) 0.514 ms 0.415 ms 0.292 ms

And sometimes it won’t be good seemingly arbitrarily:
pi@raspberry:~ $ traceroute 172.25.1.12
traceroute to 172.25.1.12 (172.25.1.12), 30 hops max, 60 byte packets
1 172.25.1.12 (172.25.1.12) 0.333 ms 0.190 ms 0.197 ms

pi@raspberry:~ $ traceroute 172.25.1.14
traceroute to 172.25.1.14 (172.25.1.14), 30 hops max, 60 byte packets
1 172.25.1.14 (172.25.1.14) 0.348 ms 0.198 ms 0.198 ms

Ultimately I would like the packets routed through 172.25.1.13 all the time so if you have any ideas as to why that is not happening consistently that would be helpful.

When a router receives a packet from host A that is to be delivered via gateway B, and it determines that host A and gateway B are in fact in the same IP network, it will inform the sender by means of an ICMP Redirect message that the packet could/should be sent via an alternate path.

The same happens if the recipient is in the same network as the incoming interface of the gateway, which is the case in your scenario. The ICMP Redirect message will tell the sender that the packet can be delivered directly to the recipient.

The standard does not dictate that an IP host has to process incoming ICMP Redirect messages, and likewise it's not mandatory for IP routers to generate them. These are the relevant files in /proc that control this behaviour:

1 members found this post helpful.

Hey thank you for your response! Following your suggestion I changed the /proc/sys/net/ipv4/conf/eth0/send_redirects file to "0" and it seems as if the packet path that I want to have used is consistently used now. Thanks for that!