strongSwan

h1. Setting Up a VPN into Amazon's Public Cloud VPC

"Amazon Web Services' VPC":http://aws.amazon.com/vpc/ (Virtual Private Cloud) is somewhat inconvenient for developers. The standard way to access it is through an IPsec "hardware VPN". In practice this means having to deal not just with IPsec, but also with BGP. This document simplifies things by using strongSwan to access the VPC instances. Neither hardware nor BGP are required.

h2. Scenario

We assume a single VPC subnet with Internet access (i.e., located behind an Internet Gateway). We have a small number of clients accessing the VPC remotely, all running on Linux machines. I believe the solution can be tweaked to allow for larger deployments. For example, you will want to replace preshared key authentication by certificate-based authentication to support a large number of clients.

h2. Solution Overview

We create a new, dedicated instance serving as a VPN gateway for the whole VPC. The solution uses tunnel-mode IPsec with IKEv2 and a virtual IP pool. For simplicity, we use preshared keys rather than certificates. strongSwan is deployed on both client and gateway.

h2. General Warnings

* Debugging IPsec is hard. Debugging networking on public cloud virtual machines is hard. Please *follow these instructions carefully*.

* Amazon's cloud is constantly changing, mostly for the better. This document has been validated with the current feature set, as of today (May 2012). It may not be valid tomorrow.

h2. Solution Steps

# Create a new VPC instance (the minimal instance type in VPC is @m1.small@). This will become your VPN Gateway. We have used standard Ubuntu images, (Oneirc @ami-a562a9cc@ and Precise @ami-a29943cb@). Instead of launching a new, dedicated instance, you can reuse an existing instance but that would be much less secure. Note that the VPN Gateway instance can be stopped when not in use, and later restarted.

# Disable source/destination check on the VPN Gateway instance.

# Assign an Elastic IP for the instance. This will be the gateway's address, but first we will use it to access the gateway and install it.

# Install strongSwan on the gateway (and on your client, too). We have used the version available in the repository, 4.5.2. Modify the configuration files per the next section.

# Enable IP forwarding on the gateway (you need to do _both_ of the following):

#* For the current session, run @echo 1 > /proc/sys/net/ipv4/ip_forward@

#* To ensure this setting persists after a reboot: edit @/etc/sysctl.conf@ and uncomment the line @net.ipv4.ip_forward=1@.

# Define the gateway's security group(s) to allow incoming TCP/22, UDP/500 and UDP/4500.

#* Once the gateway is fully set up, you will be able to disable TCP/22 in the security group and tunnel SSH through IPsec instead of directly.

# Define a subnet for the virtual IP pool. It doesn't need to be inside the VPC. In our example the VPC encompasses 10.10.0.0/16, and the virtual IP pool will be drawn from 10.100.0.0/16.

# The strongSwan Gateway will assign addresses for IPsec clients from the virtual address pool. But it needs a bit of routing help: First, note the instance ID of the VPN gateway. Then locate the routing table associated with the subnet of protected instances (this may or may not be the main routing table), and add a routing rule that routes all traffic destined to the pool's subnet (10.100.0.0/16) through the gateway.

# Allow any incoming traffic from the pool's subnet into all VPC instances. For example, by adding an "all traffic" rule to the @default@ security group of your VPC.

# For each instance that is in the VPC, disable the instance's Source/Dest. check (from the EC2 Instances page). *Edit:* this is only required for step #3 in the Bonus section below. Do not make this change unless you implement the routing changes.

# Finally, @sudo ipsec restart@ on the gateway and the client, and you are good to go!

h2. Bonus

The only job of Amazon's NAT instance is to run a single iptables rule. You can deploy that rule on your VPN gateway and save the price of a dedicated NAT instance.

# Add the "masquerade" NAT rule for the entire VPC: @sudo iptables --table nat --append POSTROUTING --source 10.10.0.0/16 -j MASQUERADE@

# Add the same line into @/etc/rc.local@ so that it will survive a reboot.

# Change the EC2 routing table so that the VPN gateway (rather than the NAT instance) becomes the default route of your private subnet.

# As an alternative to the previous step, update the /etc/network/interfaces file on your non-accessible VPC instances to use the VPN gateway as their default route:

<pre>

# static routes

up route del -net 0.0.0.0/0 gw 10.0.10.1 dev eth0

up route add -net 0.0.0.0/0 gw 10.0.10.10 dev eth0 

</pre>

*Edit:* changing the routing table on a guest is "discouraged by Amazon":https://forums.aws.amazon.com/thread.jspa?threadID=89866. Therefore the preceding method (modification of the EC2 routing table) is preferred.

h2. Debugging

These instructions didn't work for me out of the box, because I set my network up a little differently. To debug, I found it useful to do the following:

* Ping the IP address of the next closest interface.

* If you don't get replies, you need to find out what's wrong - if there is a wrong route, IP Tables rule, or something else. Use tshark to capture and display ICMP packets (replace 192.168.254.10 with the IP address of the host you are interested in):

** To capture:

<pre>

sudo tshark -f "host 192.168.254.10" -i eth0 -w /tmp/capture.cap

</pre>

** To display:

<pre>

sudo tshark -R "icmp and host 192.168.254.10" -r /tmp/capture.cap

</pre>

* You may need to install tshark on the client, VPN gateway, and VPC box to get a good view of what is happening.

h2. Configuration Files

h3. /etc/ipsec.conf on the Client

<pre>

# ipsec.conf - strongSwan IPsec configuration file

# basic configuration

config setup

	# nat_traversal=yes

	charonstart=yes

	plutostart=no

	# charondebug="ike 2, knl 2, cfg 2, mgr 2, chd 2, net 2"

# Connections into AWS VPC

conn %default

	ikelifetime=60m

	keylife=20m

	rekeymargin=3m

	keyingtries=1

	keyexchange=ikev2

	authby=secret

conn us-east-1-vpc

	left=%any

	leftsourceip=%config

	leftid=<my-email-address>

	leftfirewall=yes

	right=<gateway's elastic IP>

	rightsubnet=10.10.0.0/16

	rightid=@us-east-gw.example.com

	auto=start

# Add connections here.

# include /var/lib/strongswan/ipsec.conf.inc

</pre>

h3. /etc/ipsec.secrets on the Client

<pre>

us-east-gw.example.com : PSK "aa82c7a776e2175114213acc02dda9951a6bc25deb433e6d5d6ef7058626c589"

</pre>

h3. /etc/ipsec.conf on the Gateway

<pre>

# ipsec.conf - strongSwan IPsec configuration file

# basic configuration

config setup

	# nat_traversal=yes

	charonstart=yes

	plutostart=no

	# charondebug="ike 2, knl 2, cfg 2, mgr 3, chd 2, net 2"

# /etc/ipsec.conf - strongSwan IPsec configuration file

conn %default

	ikelifetime=60m

	keylife=20m

	rekeymargin=3m

	keyingtries=1

	keyexchange=ikev2

	authby=secret

conn client

	# The leftid parameter is not a real DNS name

	leftid=us-east-gw.example.com

	# The "left" parameter is the gateway's private IP 

	left=10.10.0.10

	# We are protecting the entire VPC, not just this subnet

	leftsubnet=10.10.0.0/16

	leftfirewall=yes

	right=%any

	# The virtual IP pool is outside the VPC!

	rightsourceip=10.100.255.0/28

	auto=add

# Add connections here.

# include /var/lib/strongswan/ipsec.conf.inc

</pre>

h3. /etc/ipsec.secrets on the Gateway

<pre>

<my-email-address> : PSK "aa82c7a776e2175114213acc02dda9951a6bc25deb433e6d5d6ef7058626c589"

</pre>