View Source Ash.Policy.Check.Builtins (ash v3.4.44)

The global authorization checks built into ash

Summary

Functions

This check is true when the current action is being run "through" a relationship.

This check is true when the action name matches the provided action name or names.

This check is true when the action type matches the provided type

This check is false when there is an actor specified, and true when the actor is nil.

This check is true when the value of the specified attribute of the actor equals the specified value.

This check is true when there is an actor specified, and false when the actor is nil.

This check always passes.

This check is true when attribute changes correspond to the provided options.

This check is true when the specified relationship is changing

This check is true when the specified relationships are changing

This check is true when the value of the specified key or path in the changeset or query context equals the specified value.

This check is true when the field provided is being referenced anywhere in a filter statement.

This check is true when the field or relationship, or path to field, is being loaded and false when it is not.

This check is true when the specified function returns true

This check never passes.

This check passes if the data relates to the actor via the specified relationship or path of relationships.

This check is true when the specified relationship is being changed to the current actor.

This check is true when the resource name matches the provided resource name or names.

This check is true when the field is being selected and false when it is not.

Functions

accessing_from(resource, relationship)

@spec accessing_from(Ash.Resource.t(), atom()) :: Ash.Policy.Check.ref()

This check is true when the current action is being run "through" a relationship.

Cases where this happens:

  1. Loading related data
  2. Managing relationships
  3. Aggregating data
  4. Filtering on relationships

action(action)

@spec action(atom() | [atom()]) :: Ash.Policy.Check.ref()

This check is true when the action name matches the provided action name or names.

This is a very common pattern, allowing action-specific policies.

action_type(action_type)

This check is true when the action type matches the provided type

This is useful for writing policies that apply to all actions of a given type.

For example:

policy action_type(:read) do
  authorize_if relates_to_actor_via(:owner)
end

actor_absent()

@spec actor_absent() :: Ash.Policy.Check.ref()

This check is false when there is an actor specified, and true when the actor is nil.

actor_attribute_equals(attribute, value)

@spec actor_attribute_equals(atom(), any()) :: Ash.Policy.Check.ref()

This check is true when the value of the specified attribute of the actor equals the specified value.

This check will never pass if the actor does not have the specified key. For example, actor_attribute_equals(:missing_key, nil)

actor_present()

@spec actor_present() :: Ash.Policy.Check.ref()

This check is true when there is an actor specified, and false when the actor is nil.

always()

@spec always() :: Ash.Policy.Check.ref()

This check always passes.

Can be useful for "deny-list" style authorization. For example:

policy action_type(:read) do
  forbid_if actor_attribute_equals(:disabled, true)
  forbid_if actor_attribute_equals(:active, false)
  authorize_if always()
end

Without that last clause, the policy would never pass.

changing_attributes(opts)

This check is true when attribute changes correspond to the provided options.

Provide a keyword list of options or just an atom representing the attribute.

For example:

# if you are changing both first name and last name
changing_attributes([:first_name, :last_name])

# if you are changing first name to fred
changing_attributes(first_name: [to: "fred"])

# if you are changing last name from bob
changing_attributes(last_name: [from: "bob"])

# if you are changing :first_name at all, last_name from "bob" and middle name from "tom" to "george"
changing_attributes([:first_name, last_name: [from: "bob"], middle_name: [from: "tom", to: "george]])

changing_relationship(relationship)

This check is true when the specified relationship is changing

changing_relationships(relationships)

This check is true when the specified relationships are changing

context_equals(key, value)

This check is true when the value of the specified key or path in the changeset or query context equals the specified value.

Note that the context is not shared with other queries (e.g. loads).

For example:

# Given this check on Profile
authorize_if context_equals(:allow_this?, true)

# This load will not have the context and will not be authorized
Ash.load!(user, :profile, context: %{allow_this?: true})

# But this will have the context and will be authorized
Ash.load!(user, [profile: Ash.Query.set_context(Profile, %{allow_this?: true})])

filtering_on(path \\ [], field)

This function is deprecated. `filtering_on/2` check is deprecated. Instead, add arguments and add policies that said arguments are set. For complex queries, policies on what is being filtered on require multiple authorization passes of the same resource, leading to a large amount of typically unnecessary complexity. Additionally, they could yield false negatives in some scenarios, and more work would be needed to ensure that they don't. .
@spec filtering_on(atom() | [atom()], atom()) :: Ash.Policy.Check.ref()

This check is true when the field provided is being referenced anywhere in a filter statement.

This applies to related filters as well. For example:

policy actor_attribute_equals(:is_admin, false) do
  forbid_if filtering_on(:email)
  # a path can be provided as well
  forbid_if filtering_on([:owner], :email)
end

The first will return true in situations like:

Ash.Query.filter(User, email == "blah")
Ash.Query.filter(Tweet, author.email == "blah")

The second will return true on queries like:

Ash.Query.filter(Post, owner.email == "blah")
Ash.Query.filter(Comment, post.owner.email == "blah")

just_created_with_action(action_name)

@spec just_created_with_action(atom()) :: Ash.Policy.Check.ref()

loading(field)

@spec loading(atom()) :: Ash.Policy.Check.ref()

This check is true when the field or relationship, or path to field, is being loaded and false when it is not.

This is always false for create/update/destroy actions, because you cannot load fields on those action types.

matches(description, func)

(macro)

This check is true when the specified function returns true

never()

@spec never() :: Ash.Policy.Check.ref()

This check never passes.

There is, generally speaking, no reason to use this, but it exists for completeness sake.

relates_to_actor_via(relationship_path, opts \\ [])

@spec relates_to_actor_via(
  atom(),
  keyword()
) :: Ash.Policy.Check.ref()

This check passes if the data relates to the actor via the specified relationship or path of relationships.

For update & destroy actions, this check will apply to the original data before the changes are applied.

For create actions this check is very unlikely to pass. This is because relationships are modified after authorization happens, not before.

For example:

policy action_type(:read) do
  authorize_if relates_to_actor_via(:owner)

  # Path of relationships:
  authorize_if relates_to_actor_via([:account, :user])

  # When the resource relates to a field of the actor:
  authorize_if relates_to_actor_via(:roles, field: :role)
end

relating_to_actor(relationship)

This check is true when the specified relationship is being changed to the current actor.

This only supports belongs_to relationships at the moment, and will detect two cases:

  1. the source_attribute is being changed directly
  2. the relationship is being changed with on_lookup?: :relate, and a single input is being provided.

resource(resource)

@spec resource(atom() | [atom()]) :: Ash.Policy.Check.ref()

This check is true when the resource name matches the provided resource name or names.

selecting(attribute)

@spec selecting(atom()) :: Ash.Policy.Check.ref()

This check is true when the field is being selected and false when it is not.

This won't affect filters placed on this resource, so you may also want to either:

  • Mark the given field as filterable? false
  • Add another check for filtering_on(:field)

For example:

policy action_type(:read) do
  # The actor can read and filter on their own email
  authorize_if expr(id == ^actor(:id))

  # No one else can select or filter on their email
  forbid_if selecting(:email)
  forbid_if filtering_on(:email)

  # Otherwise, the policy passes
  authorize_if always()
end