Network policy

A network policy resource (NetworkPolicy) represents an ordered set of rules which are applied to a collection of endpoints that match a label selector.

NetworkPolicy is a namespaced resource. NetworkPolicy in a specific namespace only applies to workload endpoint resources in that namespace. Two resources are in the same namespace if the namespace value is set the same on both. See global network policy resource for non-namespaced network policy.

NetworkPolicy resources can be used to define network connectivity rules between groups of Calico endpoints and host endpoints, and take precedence over profile resources if any are defined.

Sample YAML

This sample policy allows TCP traffic from frontend endpoints to port 6379 on database endpoints.

apiVersion: projectcalico.org/v3
kind: NetworkPolicy
metadata:
  name: allow-tcp-6379
  namespace: production
spec:
  selector: role == 'database'
  types:
    - Ingress
    - Egress
  ingress:
    - action: Allow
      metadata:
        annotations:
          from: frontend
          to: database
      protocol: TCP
      source:
        selector: role == 'frontend'
      destination:
        ports:
          - 6379
  egress:
    - action: Allow

Definition

Metadata

Spec

* If types has no value, Calico defaults as follows.

Ingress Rules Present	Egress Rules Present	Types value
No	No	Ingress
Yes	No	Ingress
No	Yes	Egress
Yes	Yes	Ingress, Egress

Rule

A single rule matches a set of packets and applies some action to them. When multiple rules are specified, they are executed in order.

After a Log action, processing continues with the next rule; Allow and Deny are immediate and final and no further rules are processed.

An action of Pass in a NetworkPolicy or GlobalNetworkPolicy will skip over the remaining policies and jump to the first profile assigned to the endpoint, applying the policy configured in the profile; if there are no Profiles configured for the endpoint the default applied action is Deny.

RuleMetadata

Metadata associated with a specific rule (rather than the policy as a whole). The contents of the metadata does not affect how a rule is interpreted or enforced; it is simply a way to store additional information for use by operators or applications that interact with Calico.

Example:

metadata:
  annotations:
    app: database
    owner: devops

Annotations follow the same rules as Kubernetes for valid syntax and character set.

On Linux with the iptables dataplane, rule annotations are rendered as comments in the form -m comment --comment "<key>=<value>" on the iptables rule(s) that correspond to the Calico rule.

ICMP

EntityRule

Entity rules specify the attributes of the source or destination of a packet that must match for the rule as a whole to match. Packets can be matched on combinations of:

• Identity of the source/destination, by using Selectors or by specifying a particular Kubernetes Service. Selectors can match workload endpoints, host endpoints and (namespaced or global) network sets.
• Source/destination IP address, protocol and port.

If the rule contains multiple match criteria (for example, an IP and a port) then all match criteria must match for the rule as a whole to match a packet.

When using selectors in network policy, remember that selectors only match (known) resources, but rules match packets. A rule with a selector all() won’t match “all packets”, it will match “packets from all in-scope endpoints and network sets”. To match all packets, do not include a selector in the rule at all.

Understanding notSelector

notSelector is somewhat subtle because the not in notSelector negates the packet match rather than the selector:

• selector: !has(foo) matches packets from/to endpoints and network sets that do not have the label “foo”.
• notSelector: has(foo) matches packets from/to anywhere (including outside the cluster), except traffic from/to endpoints and network sets that have the label “foo”.

Selector performance in EntityRules

When rendering policy into the dataplane, Calico must identify the endpoints that match the selectors in all active rules. This calculation is optimized for certain common selector types. Using the optimized selector types reduces CPU usage (and policy rendering time) by orders of magnitude. This becomes important at high scale (hundreds of active rules, hundreds of thousands of endpoints).

The optimized operators are as follows:

• label == "value"
• label in { 'v1', 'v2' }
• has(label)
• <expression 1> && <expression 2> is optimized if either <expression 1> or <expression 2> is optimized.

The following perform like has(label). All endpoints with the label will be scanned to find matches:

• label contains 's'
• label starts with 's'
• label ends with 's'

The other operators, and in particular, all(), !, || and != are not optimized.

Examples:

• a == 'b' - optimized
• a == 'b' && has(c) - optimized
• a == 'b' || has(c) - not optimized due to use of ||
• c != 'd' - not optimized due to use of !=
• !has(a) - not optimized due to use of !
• a == 'b' && c != 'd' - optimized, a =='b' is optimized so a == 'b' && <anything> is optimized.
• c != 'd' && a == 'b' - optimized, a =='b' is optimized so <anything> && a == 'b' is optimized.

Selectors

A label selector is an expression which either matches or does not match a resource based on its labels.

Calico label selectors support a number of operators, which can be combined into larger expressions using the boolean operators and parentheses.

Operators have the following precedence:

• Highest: all the match operators
• Parentheses ( ... )
• Negation with !
• Conjunction with &&
• Lowest: Disjunction with ||

For example, the expression

! has(my-label) || my-label starts with 'prod' && role in {'frontend','business'}

Would be “bracketed” like this:

((!(has(my-label)) || ((my-label starts with 'prod') && (role in {'frontend','business'}))

It would match:

• Any resource that did not have label “my-label”.
• Any resource that both:
  • Has a value for my-label that starts with “prod”, and,
  • Has a role label with value either “frontend”, or “business”.

Understanding scopes and the all() and global() operators: selectors have a scope of resources that they are matched against, which depends on the context in which they are used. For example:

• The nodeSelector in an IPPool selects over Node resources.

• The top-level selector in a NetworkPolicy selects over the workloads in the same namespace as the NetworkPolicy.

• The top-level selector in a GlobalNetworkPolicy doesn’t have the same restriction, it selects over all endpoints including namespaced WorkloadEndpoints and non-namespaced HostEndpoints.

• The namespaceSelector in a NetworkPolicy (or GlobalNetworkPolicy) rule selects over the labels on namespaces rather than workloads.

• The namespaceSelector determines the scope of the accompanying selector in the entity rule. If no namespaceSelector is present then the rule’s selector matches the default scope for that type of policy. (This is the same namespace for NetworkPolicy and all endpoints/network sets for GlobalNetworkPolicy)

• The global() operator can be used (only) in a namespaceSelector to change the scope of the main selector to include non-namespaced resources such as GlobalNetworkSet. This allows namespaced NetworkPolicy resources to refer to global non-namespaced resources, which would otherwise be impossible.

Ports

Calico supports the following syntaxes for expressing ports.

An individual numeric port may be specified as a YAML/JSON integer. A port range or named port must be represented as as a string. For example, this would be a valid list of ports:

ports: [8080, '1234:5678', 'named-port']

Named ports

Using a named port in an EntityRule, instead of a numeric port, gives a layer of indirection, allowing for the named port to map to different numeric values for each endpoint.

For example, suppose you have multiple HTTP servers running as workloads; some exposing their HTTP port on port 80 and others on port 8080. In each workload, you could create a named port called http-port that maps to the correct local port. Then, in a rule, you could refer to the name http-port instead of writing a different rule for each type of server.

note

Since each named port may refer to many endpoints (and Calico has to expand a named port into a set of endpoint/port combinations), using a named port is considerably more expensive in terms of CPU than using a simple numeric port. We recommend that they are used sparingly, only where the extra indirection is required.

ServiceAccountMatch

A ServiceAccountMatch matches service accounts in an EntityRule.

ServiceMatch

A ServiceMatch matches a service in an EntityRule.

Service matches are only supported when using the Kubernetes datastore driver. They are not supported by the etcd datastore driver.

Performance Hints

Performance hints provide a way to tell Calico about the intended use of the policy so that it may process it more efficiently. Currently only one hint is defined:

• AssumeNeededOnEveryNode: normally, Calico only calculates a policy’s rules and selectors on nodes where the policy is actually in use (i.e. its selector matches a local endpoint). This saves work in most cases. The AssumeNeededOnEveryNode hint tells Calico to treat the policy as “in use” on every node. This is useful for large policy sets that are known to apply to all (or nearly all) endpoints. It effectively “preloads” the policy on every node so that there is less work to do when the first endpoint matching the policy shows up. It also prevents work from being done to tear down the policy when the last endpoint is drained.

Application layer policy

Application layer policy is an optional feature of Calico and must be enabled to use the following match criteria.

note

Application layer policy match criteria are supported with the following restrictions.

• Only ingress policy is supported. Egress policy must not contain any application layer policy match clauses.
• Rules must have the action Allow if they contain application layer policy match clauses.

HTTPMatch

An HTTPMatch matches attributes of an HTTP request. The presence of an HTTPMatch clause on a Rule will cause that rule to only match HTTP traffic. Other application layer protocols will not match the rule.

Example:

http:
  methods: ['GET', 'PUT']
  paths:
    - exact: '/projects/calico'
    - prefix: '/users'

HTTPPathMatch

Supported operations