HOWTO: Enterprise cloud encryption

Introduction

Most IPsec deployments fall into two types of deployment. The first type is the Remote Access, where roaming users (phones, laptops) connect to the corporate network. The second type of IPsec network is where two or more IPsec gateways connects different networks together. Both of these types of deployment use encryption only to encrypt the part of the network that is considered to be using the public internet. These type of deployments are called "static tunnels" or "roadwarrior tunnels". The configuration of these deployments is simple. Both endpoints are preconfigured to accept each other's credentials.

The goal of enterprise cloud encryption (also called mesh encryption) is to encrypt all traffic between all the nodes, without using dedicated IPsec gateways. This elliminates all plaintext traffic on a given network or cloud. These deployments are very dynamic. If the network contains thousands of hosts, where only some hosts communicate to some other hosts. In modern cloud deployments, these hosts also vary. Depending on the demand, servers are continiously added and removed from the network. Therefore, the classic configuration of each host having a configuration for all other hosts does not work. Even if one would use automated tools like puppet or ansible, adding a new host should not require all other hosts to have their IPsec configuration updated before they can communicate with the newly added host.

Traditionally, libreswan calls these type of dynamic configurations "opportunistic". This goes back to the FreeS/WAN days in the late nineties. Each node is configured to intercept all unencrypted traffic. Once an unencrypted packet is seen by the kernel, the kernel will notify the userland IKE daemon so it can attempt to setup an IPsec connection to the remote IP address. Once this tunnel policy is in place, the kernel will allow the packets to out encrypted. There are various tuning options to allow for different service levels. For example, it is possible to insist that some network ranges always have encryption, or never have encryption. Or encryption to some networks can be optional.

The basic configuration

Currently, libreswan supports a few different groups of connections:

• private - Traffic to these networks must be encrypted
• private-or-clear - Traffic should be encryped but may happen in the clear
• clear-or-private - Traffic per default will go in the clear, but if the remote endpoint negotiates encryption that is accepted
• clear - No encryption will ever be allowed and all traffic happens in the clear
• block - No traffic, clear or encrypted, is allowed to these destinations

To add a certain network range to one of these policies, simple add them in CIDR notation to the names files in /etc/ipsec.d/policies/. To enable mandatory encryption for all nodes within 10.0.0.0/8, simply add "10.0.0.0/8" on a line by itself to the file /etc/ipsec.d/policies/private

The above named policies must be created as "regular" IPsec connections within libreswan. This is done by placing a configuration file in /etc/ipsec.d/*.conf. You can find some examples in the source tree in the doc/examples directory. Specifically look at the oe-* prefixed files.

For example, when using X.509 certificates, the "private" connection would look something like:

# example for /etc/ipsec.d/private.conf
conn private
	leftid=%fromcert
	leftrsasigkey=%cert
	# nickname of your certificate imported to NSS
	leftcert=example-123.libreswan.org
	left=%defaultroute
	#
	right=%opportunisticgroup
	rightid=%fromcert
	rightrsasigkey=%cert
	#
	authby=rsasigkey
	negotiationshunt=hold
	failureshunt=drop
	type=tunnel
	ikev2=insist
	sendca=issuer
	auto=add

And the policy files in /etc/ipsec.d/policies/ would look like:

# /etc/ipsec.d/policies/private
10.0.0.0/8

# /etc/ipsec.d/policies/clear
0.0.0.0/0

Currently, libreswan orders these CIRD networks based on "longest prefix first", meaning the more specific and thus smaller the network range, the higher the priority.

When starting libreswan with this configuration, an IPsec policy will be installed in the kernel to "trap" any packets destined for IP addresses covered by entries in /etc/ipsec.d/policies/private, and libreswan will attempt to setup an IPsec tunnel to those IP addresses. If this fails, it will drop the traffic to prevent cleartext packet leaks. In the time that the IPsec connection is being negotiated, packets are held to prevent cleartext leaks.

An IP address is not a certified identifier

One issue is that any IPsec connection that is triggered by a packet needs to be authenticated using only the source and destination IP address. Even if certificates are exchanged, a server in the cloud cannot tell if that certificate belongs to that IP address. In the next release, libreswan will add an additional DNS lookup of the name specified by the received certificate to see if the DNS entry for that name has a proper address (A or AAAA) record to the IP that gave us the certificate. If this DNS domain is properly DNSSEC signed, we have some certainty that this host is really who they claim to be.

Authentication methods

The following authentication methods are planned to be supported with cloud encryption via Opportunistic IPsec:

• X.509 certificates (mutual authentication - the method described above)
• Single side X.509 certificates (eg LetsEncrypt, where client authenticates the server, but the server does not authenticate the client)
• GSSAPI / Kerberos (mutual authentication)
• DNSSEC based (mutual or single sided)
• DNSSEC via Reverse DNS (mutual or single sided)

Lets Encrypt

As of libreswan-3.119 there is support for single sided certificate authentication that can be used with LetsEncrypt or any other public CA. It also supports NAT-Traversal, so that this can be used by roaming clients (laptops, phones) that are often behind NAT. For details, see HOWTO: Opportunistic IPsec using LetsEncrypt