The RBAC feature enables restrictions of access to Argo CD resources. Argo CD does not have its own
user management system and has only one built-in user, admin
. The admin
user is a superuser and
it has unrestricted access to the system. RBAC requires SSO configuration or one or more local users setup.
Once SSO or local users are configured, additional RBAC roles can be defined, and SSO groups or local users can then be mapped to roles.
There are two main components where RBAC configuration can be defined:
- The global RBAC config map (see argo-rbac-cm.yaml)
- The AppProject's roles
Argo CD has two pre-defined roles but RBAC configuration allows defining roles and groups (see below).
role:readonly
: read-only access to all resourcesrole:admin
: unrestricted access to all resources
These default built-in role definitions can be seen in builtin-policy.csv
When a user is authenticated in Argo CD, it will be granted the role specified in policy.default
.
!!! warning "Restricting Default Permissions"
**All authenticated users get _at least_ the permissions granted by the default policies. This access cannot be blocked
by a `deny` rule.** It is recommended to create a new `role:authenticated` with the minimum set of permissions possible,
then grant permissions to individual roles as needed.
Enabling anonymous access to the Argo CD instance allows users to assume the default role permissions specified by policy.default
without being authenticated.
The anonymous access to Argo CD can be enabled using the users.anonymous.enabled
field in argocd-cm
(see argocd-cm.yaml).
!!! warning
When enabling anonymous access, consider creating a new default role and assigning it to the default policies
with `policy.default: role:unauthenticated`.
The model syntax is based on Casbin. There are two different types of syntax: one for assigning policies, and another one for assigning users to internal roles.
Group: Allows to assign authenticated users/groups to internal roles.
Syntax: g, <user/group>, <role>
<user/group>
: The entity to whom the role will be assigned. It can be a local user or a user authenticated with SSO. When SSO is used, theuser
will be based on thesub
claims, while the group is one of the values returned by thescopes
configuration.<role>
: The internal role to which the entity will be assigned.
Policy: Allows to assign permissions to an entity.
Syntax: p, <role/user/group>, <resource>, <action>, <object>, <effect>
<role/user/group>
: The entity to whom the policy will be assigned<resource>
: The type of resource on which the action is performed.<action>
: The operation that is being performed on the resource.<object>
: The object identifier representing the resource on which the action is performed. Depending on the resource, the object's format will vary.<effect>
: Whether this policy should grant or restrict the operation on the target object. One ofallow
ordeny
.
Below is a table that summarizes all possible resources and which actions are valid for each of them.
Resource\Action | get | create | update | delete | sync | action | override | invoke |
---|---|---|---|---|---|---|---|---|
applications | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ |
applicationsets | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
clusters | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
projects | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
repositories | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ |
accounts | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |
certificates | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
gpgkeys | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
logs | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
exec | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
extensions | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ |
Some policy only have meaning within an application. It is the case with the following resources:
applications
applicationsets
logs
exec
While they can be set in the global configuration, they can also be configured in AppProject's roles.
The expected <object>
value in the policy structure is replaced by <app-project>/<app-name>
.
For instance, these policies would grant example-user
access to get any applications,
but only be able to see logs in my-app
application part of the example-project
project.
p, example-user, applications, get, *, allow
p, example-user, logs, get, example-project/my-app, allow
When application in any namespace is enabled, the expected <object>
value in the policy structure is replaced by <app-project>/<app-ns>/<app-name>
.
Since multiple applications could have the same name in the same project, the policy below makes sure to restrict access only to app-namespace
.
p, example-user, applications, get, */app-namespace/*, allow
p, example-user, logs, get, example-project/app-namespace/my-app, allow
The applications
resource is an Application-Specific Policy.
The update
and delete
actions, when granted on an application, will allow the user to perform the operation on the application itself and all of its resources.
It can be desirable to only allow update
or delete
on specific resources within an application.
To do so, when the action if performed on an application's resource, the <action>
will have the <action>/<group>/<kind>/<ns>/<name>
format.
For instance, to grant access to example-user
to only delete Pods in the prod-app
Application, the policy could be:
p, example-user, applications, delete/*/Pod/*/*, default/prod-app, allow
!!!warning "Understand glob pattern behavior"
Argo CD RBAC does not use `/` as a separator when evaluating glob patterns. So the pattern `delete/*/kind/*`
will match `delete/<group>/kind/<namespace>/<name>` but also `delete/<group>/<kind>/kind/<name>`.
The fact that both of these match will generally not be a problem, because resource kinds generally contain capital
letters, and namespaces cannot contain capital letters. However, it is possible for a resource kind to be lowercase.
So it is better to just always include all the parts of the resource in the pattern (in other words, always use four
slashes).
If we want to grant access to the user to update all resources of an application, but not the application itself:
p, example-user, applications, update/*, default/prod-app, allow
If we want to explicitly deny delete of the application, but allow the user to delete Pods:
p, example-user, applications, delete, default/prod-app, deny
p, example-user, applications, delete/*/Pod/*/*, default/prod-app, allow
!!! note
It is not possible to deny fine-grained permissions for a sub-resource if the action was **explicitly allowed on the application**.
For instance, the following policies will **allow** a user to delete the Pod and any other resources in the application:
```csv
p, example-user, applications, delete, default/prod-app, allow
p, example-user, applications, delete/*/Pod/*/*, default/prod-app, deny
```
The action
action corresponds to either built-in resource customizations defined
in the Argo CD repository,
or to custom resource actions defined by you.
See the resource actions documentation for a list of built-in actions.
The <action>
has the action/<group>/<kind>/<action-name>
format.
For example, a resource customization path resource_customizations/extensions/DaemonSet/actions/restart/action.lua
corresponds to the action
path action/extensions/DaemonSet/restart
. If the resource is not under a group (for example, Pods or ConfigMaps),
then the path will be action//Pod/action-name
.
The following policies allows the user to perform any action on the DaemonSet resources, as well as the maintenance-off
action on a Pod:
p, example-user, applications, action//Pod/maintenance-off, default/*, allow
p, example-user, applications, action/extensions/DaemonSet/*, default/*, allow
To allow the user to perform any actions:
p, example-user, applications, action/*, default/*, allow
When granted along with the sync
action, the override action will allow a user to synchronize local manifests to the Application.
These manifests will be used instead of the configured source, until the next sync is performed.
The applicationsets
resource is an Application-Specific policy.
ApplicationSets provide a declarative way to automatically create/update/delete Applications.
Allowing the create
action on the resource effectively grants the ability to create Applications. While it doesn't allow the
user to create Applications directly, they can create Applications via an ApplicationSet.
!!! note
In v2.5, it is not possible to create an ApplicationSet with a templated Project field (e.g. `project: {{path.basename}}`)
via the API (or, by extension, the CLI). Disallowing templated projects makes project restrictions via RBAC safe:
With the resource being application-specific, the <object>
of the applicationsets policy will have the format <app-project>/<app-name>
.
However, since an ApplicationSet does belong to any project, the <app-project>
value represents the projects in which the ApplicationSet will be able to create Applications.
With the following policy, a dev-group
user will be unable to create an ApplicationSet capable of creating Applications
outside the dev-project
project.
p, dev-group, applicationsets, *, dev-project/*, allow
The logs
resource is an Application-Specific Policy.
When granted with the get
action, this policy allows a user to see Pod's logs of an application via
the Argo CD UI. The functionality is similar to kubectl logs
.
The exec
resource is an Application-Specific Policy.
When granted with the create
action, this policy allows a user to exec
into Pods of an application via
the Argo CD UI. The functionality is similar to kubectl exec
.
See Web-based Terminal for more info.
With the extensions
resource, it is possible to configure permissions to invoke proxy extensions.
The extensions
RBAC validation works in conjunction with the applications
resource.
A user needs to have read permission on the application where the request is originated from.
Consider the example below, it will allow the example-user
to invoke the httpbin
extensions in all
applications under the default
project.
p, example-user, applications, get, default/*, allow
p, example-user, extensions, invoke, httpbin, allow
When deny
is used as an effect in a policy, it will be effective if the policy matches.
Even if more specific policies with the allow
effect match as well, the deny
will have priority.
The order in which the policies appears in the policy file configuration has no impact, and the result is deterministic.
The evaluation of access is done in two parts: validating against the default policy configuration, then validating against the policies for the current user.
If an action is allowed or denied by the default policies, then this effect will be effective without further evaluation. When the effect is undefined, the evaluation will continue with subject-specific policies.
The access will be evaluated for the user, then for each configured group that the user is part of.
The matching engine, configured in policy.matchMode
, can use two different match modes to compare the values of tokens:
glob
: based on theglob
package.regex
: based on theregexp
package.
When all tokens match during the evaluation, the effect will be returned. The evaluation will continue until all matching policies are evaluated, or until a policy with the deny
effect matches.
After all policies are evaluated, if there was at least one allow
effect and no deny
, access will be granted.
When glob
is used, the policy tokens are treated as single terms, without separators.
Consider the following policy:
p, example-user, applications, action/extensions/*, default/*, allow
When the example-user
executes the extensions/DaemonSet/test
action, the following glob
matches will happen:
- The current user
example-user
matches the tokenexample-user
. - The value
applications
matches the tokenapplications
. - The value
action/extensions/DaemonSet/test
matchesaction/extensions/*
. Note that/
is not treated as a separator and the use of**
is not necessary. - The value
default/my-app
matchesdefault/*
.
The scopes
field controls which OIDC scopes to examine during RBAC enforcement (in addition to sub
scope).
If omitted, it defaults to '[groups]'
. The scope value can be a string, or a list of strings.
For more information on scopes
please review the User Management Documentation.
The following example shows targeting email
as well as groups
from your OIDC provider.
apiVersion: v1
kind: ConfigMap
metadata:
name: argocd-rbac-cm
namespace: argocd
labels:
app.kubernetes.io/name: argocd-rbac-cm
app.kubernetes.io/part-of: argocd
data:
policy.csv: |
p, my-org:team-alpha, applications, sync, my-project/*, allow
g, my-org:team-beta, role:admin
g, user@example.org, role:admin
policy.default: role:readonly
scopes: '[groups, email]'
This can be useful to associate users' emails and groups directly in AppProject.
apiVersion: argoproj.io/v1alpha1
kind: AppProject
metadata:
name: team-beta-project
namespace: argocd
spec:
roles:
- name: admin
description: Admin privileges to team-beta
policies:
- p, proj:team-beta-project:admin, applications, *, *, allow
groups:
- user@example.org # Value from the email scope
- my-org:team-beta # Value from the groups scope
Local users are assigned access by either grouping them with a role or by assigning policies directly to them.
The example below shows how to assign a policy directly to a local user.
p, my-local-user, applications, sync, my-project/*, allow
This example shows how to assign a role to a local user.
g, my-local-user, role:admin
!!! warning "Ambiguous Group Assignments"
If you have [enabled SSO](user-management/index.md#sso), any SSO user with a scope that matches a local user will be
added to the same roles as the local user. For example, if local user `sally` is assigned to `role:admin`, and if an
SSO user has a scope which happens to be named `sally`, that SSO user will also be assigned to `role:admin`.
An example of where this may be a problem is if your SSO provider is an SCM, and org members are automatically
granted scopes named after the orgs. If a user can create or add themselves to an org in the SCM, they can gain the
permissions of the local user with the same name.
To avoid ambiguity, if you are using local users and SSO, it is recommended to assign policies directly to local
users, and not to assign roles to local users. In other words, instead of using `g, my-local-user, role:admin`, you
should explicitly assign policies to `my-local-user`:
```yaml
p, my-local-user, *, *, *, allow
```
It is possible to provide additional entries in the argocd-rbac-cm
configmap to compose the final policy csv.
In this case, the key must follow the pattern policy.<any string>.csv
.
Argo CD will concatenate all additional policies it finds with this pattern below the main one ('policy.csv').
The order of additional provided policies are determined by the key string.
Example: if two additional policies are provided with keys policy.A.csv
and policy.B.csv
,
it will first concatenate policy.A.csv
and then policy.B.csv
.
This is useful to allow composing policies in config management tools like Kustomize, Helm, etc.
The example below shows how a Kustomize patch can be provided in an overlay to add additional configuration to an existing RBAC ConfigMap.
apiVersion: v1
kind: ConfigMap
metadata:
name: argocd-rbac-cm
namespace: argocd
data:
policy.tester-overlay.csv: |
p, role:tester, applications, *, */*, allow
p, role:tester, projects, *, *, allow
g, my-org:team-qa, role:tester
If you want to ensure that your RBAC policies are working as expected, you can
use the argocd admin settings rbac
command to validate them.
This tool allows you to test whether a certain role or subject can perform the requested action with a policy
that's not live yet in the system, i.e. from a local file or config map.
Additionally, it can be used against the live RBAC configuration in the cluster your Argo CD is running in.
To check whether your new policy configuration is valid and understood by Argo CD's RBAC implementation,
you can use the argocd admin settings rbac validate
command.
To test whether a role or subject (group or local user) has sufficient
permissions to execute certain actions on certain resources, you can
use the argocd admin settings rbac can
command.