Will you be at KubeCon this year? Join us for CalicoCon 2019!

Security Policy Model

Calico applies security policy to endpoints. Calico policy is defined in terms of security profiles, which contain lists of rules to apply as well as sets of tags.

Endpoints

Endpoints are the TAPs, veths or other interfaces, which are attached to virtual machines or containers.

Calico applies one or more security profiles to each endpoint.

Calico always tries to fail safe: if the configuration for an endpoint is missing, or no profiles are configured, Calico will drop traffic to/from that endpoint. There is always an implicit default deny rule at the end of the list of profiles.

Security profiles: rules

Endpoints are configured to belong to one or more security profiles. Profiles encode the policy (i.e. which packets to allow or deny) to apply to an endpoint.

Policy is encoded as two lists of rules:

  • an inbound list, which is traversed for packets that are going to the endpoint
  • an outbound list, for packets coming from the endpoint

In the lists, each rule consists of a set of match criteria and an action. The match criteria include:

  • protocol,
  • source/dest CIDR,
  • source/dest tag (see below),
  • source/dest port,
  • ICMP type and code.

Calico supports actions, “allow” and “deny”, which immediately accept or reject the packet. Once a packet is accepted or rejected further rules are not processed.

If a packet does not match any of the rules in any of the profiles attached to an endpoint then the default is to deny traffic.

If a workload (such as a virtual machine) has multiple endpoints (for example, multiple vNICs) then each of those endpoints may belong to a different set of security profiles.

Security profiles: tags

Each profile also has a set of (opaque) tags attached to it. An endpoint is considered a member of a tag if one of its profiles contains that tag.

Profile rules may reference tags in the source and destination match criteria. Calico calculates the tag memberships dynamically, updating them as endpoints come and go and as profiles are updated. This allows for very fine-grained but also maintainable policy.

For example, an operator could add the “db-user” tag to all endpoints that are to use the database. Then, they can use a single “allow” rule in the database’s inbound chain to allow connections from all current members of the “db-user” tag.

Differences from OpenStack

Calico represents OpenStack security groups as profiles (with a single tag containing the name of the security group). While this is a simple 1-to-1 mapping at the rule level, there are some differences between Calico and OpenStack’s security models to consider:

Effective security in OpenStack is a product of the interaction between three kinds of objects: networks, routers and security groups. Calico, on the other hand, only uses security groups for security configuration; and networks and routers have no impact. The following subsections go into this in more detail, and discuss how these concepts map onto the Calico data model.

Networks and Routers

As discussed here, networks and routers are not used in Calico for connectivity purposes. Similarly, they serve no security purpose in a Calico environment.

Calico can provide equivalent functionality to networks and routers using security groups. To achieve it, rather than placing all ports that need to communicate into a single network, place them all in a security group that allows ingress from and egress to the same security group.

Architecture

At present, the flow of security information proceeds as follows:

[Configuration in OpenStack or other orchestrator] -(Plugin)-> [etcd] -(Felix)-> [Programmed iptables rules]

When a security group is configured, the Calico orchestrator plugin discovers the new configuration. This configuration is translated into the Calico data model and written to etcd. The Felix agent watches etcd for changes and applies the policy using the kernel’s iptables and ipsets.