Privacy

The main advantage of the privacy layer is that, you write the privacy policy once (in the schema), and it is always evaluated. No matter where queries and mutations are performed in your codebase, it will always go through the privacy layer.

In this tutorial, we will start by going over the basic terms we use in the framework, continue with a section for configuring the policy feature to your project, and finish with a few examples.

The ent.Policy interface contains two methods: EvalQuery and EvalMutation. The first defines the read-policy, and the second defines the write-policy. A policy contains zero or more privacy rules (see below). These rules are evaluated in the same order they are declared in the schema.

If all rules are evaluated without returning an error, the evaluation finishes successfully, and the executed operation gets access to the target nodes.

However, if one of the evaluated rules returns an error or a privacy.Deny decision (see below), the executed operation returns an error, and it is cancelled.

Privacy Rules

Each policy (mutation or query) includes one or more privacy rules. The function signature for these rules is as follows:

There are three types of decision that can help you control the privacy rules evaluation.

• privacy.Allow - If returned from a privacy rule, the evaluation stops (next rules will be skipped), and the executed operation (query or mutation) gets access to the target nodes.
• privacy.Deny - If returned from a privacy rule, the evaluation stops (next rules will be skipped), and the executed operation is cancelled. This equivalent to returning any error.
• privacy.Skip - Skip the current rule, and jump to the next privacy rule. This equivalent to returning a nil error.

Now that we’ve covered the basic terms, let’s start writing some code.

In order to enable the privacy option in your code generation, enable the privacy feature with one of two options:

1. If you are using the default go generate config, add --feature privacy option to the ent/generate.go file as follows:

package ent
//go:generate go run -mod=mod entgo.io/ent/cmd/ent generate --feature privacy ./schema

It is recommended to add the feature-flag along with the privacy to enhance the development experience (e.g. --feature privacy,schema/snapshot)

1. If you are using the configuration from the GraphQL documentation, add the feature flag as follows:

// Copyright 2019-present Facebook Inc. All rights reserved.
// This source code is licensed under the Apache 2.0 license found
// in the LICENSE file in the root directory of this source tree.
// +build ignore
package main
import (
    "log"
    "entgo.io/ent/entc"
    "entgo.io/ent/entc/gen"
    "entgo.io/contrib/entgql"
)
func main() {
    opts := []entc.Option{
        entc.FeatureNames("privacy"),
    }
    err := entc.Generate("./schema", &gen.Config{
        Templates: entgql.AllTemplates,
    }, opts...)
    if err != nil {
        log.Fatalf("running ent codegen: %v", err)
    }
}

important" class="reference-link">important

You should notice that, similar to , if you use the Policy option in your schema, you MUST add the following import in the main package, because a circular import is possible between the schema package, and the generated ent package:

import _ "<project>/ent/runtime"

Admin Only

We start with a simple example of an application that lets any user read any data, and accepts mutations only from users with admin role. We will create 2 additional packages for the purpose of the examples:

• rule - for holding the different privacy rules in our schema.
• viewer - for getting and setting the user/viewer who’s executing the operation. In this simple example, it can be either a normal user or an admin.

After running the code-generation (with the feature-flag for privacy), we add the Policy method with 2 generated policy rules.

examples/privacyadmin/ent/schema/user.go

package schema
import (
    "entgo.io/ent"
    "entgo.io/ent/examples/privacyadmin/ent/privacy"
)
// User holds the schema definition for the User entity.
type User struct {
    ent.Schema
}
// Policy defines the privacy policy of the User.
func (User) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            // Deny if not set otherwise. 
            privacy.AlwaysDenyRule(),
        },
        Query: privacy.QueryPolicy{
            // Allow any viewer to read anything.
            privacy.AlwaysAllowRule(),
        },
    }
}

We defined a policy that rejects any mutation and accepts any query. However, as mentioned above, in this example, we accept mutations only from viewers with admin role. Let’s create 2 privacy rules to enforce this:

package rule
import (
    "context"
    "entgo.io/ent/examples/privacyadmin/ent/privacy"
    "entgo.io/ent/examples/privacyadmin/viewer"
)
// DenyIfNoViewer is a rule that returns Deny decision if the viewer is
// missing in the context.
func DenyIfNoViewer() privacy.QueryMutationRule {
    return privacy.ContextQueryMutationRule(func(ctx context.Context) error {
        view := viewer.FromContext(ctx)
        if view == nil {
            return privacy.Denyf("viewer-context is missing")
        }
        // Skip to the next privacy rule (equivalent to returning nil).
        return privacy.Skip
    })
}
// AllowIfAdmin is a rule that returns Allow decision if the viewer is admin.
func AllowIfAdmin() privacy.QueryMutationRule {
    return privacy.ContextQueryMutationRule(func(ctx context.Context) error {
        view := viewer.FromContext(ctx)
        if view.Admin() {
            return privacy.Allow
        }
        return privacy.Skip
    })
}

As you can see, the first rule DenyIfNoViewer, makes sure every operation has a viewer in its context, otherwise, the operation rejected. The second rule AllowIfAdmin, accepts any operation from viewer with admin role. Let’s add them to the schema, and run the code-generation:

examples/privacyadmin/ent/schema/user.go

// Policy defines the privacy policy of the User.
func (User) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            rule.DenyIfNoViewer(),
            privacy.AlwaysDenyRule(),
        },
        Query: privacy.QueryPolicy{
            privacy.AlwaysAllowRule(),
        },
    }
}

Since we define the DenyIfNoViewer first, it will be executed before all other rules, and accessing the viewer.Viewer object is safe in the AllowIfAdmin rule.

After adding the rules above and running the code-generation, we expect the privacy-layer logic to be applied on ent.Client operations.

examples/privacyadmin/example_test.go

Sometimes, we want to bind a specific privacy decision to the context.Context. In cases like this, we can use the privacy.DecisionContext function to create a new context with a privacy decision attached to it.

examples/privacyadmin/example_test.go

func Do(ctx context.Context, client *ent.Client) error {
    // Bind a privacy decision to the context (bypass all other rules).
    allow := privacy.DecisionContext(ctx, privacy.Allow)
    if err := client.User.Create().Exec(allow); err != nil {
        return fmt.Errorf("expect operation to pass, but got %w", err)
    }
    return nil
}

The full example exists in .

Multi Tenancy

In this example, we’re going to create a schema with 3 entity types - Tenant, User and Group. The helper packages viewer and rule (as mentioned above) also exist in this example to help us structure the application.

Let’s start building this application piece by piece. We begin by creating 3 different schemas (see the full code ), and since we want to share some logic between them, we create another mixed-in schema and add it to all other schemas as follows:

examples/privacytenant/ent/schema/mixin.go

// BaseMixin for all schemas in the graph.
type BaseMixin struct {
    mixin.Schema
}
// Policy defines the privacy policy of the BaseMixin.
func (BaseMixin) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            rule.DenyIfNoViewer(),
        },
        Query: privacy.QueryPolicy{
            rule.DenyIfNoViewer(),
        },
    }
}

examples/privacytenant/ent/schema/tenant.go

// Mixin of the Tenant schema.
func (Tenant) Mixin() []ent.Mixin {
    return []ent.Mixin{
        BaseMixin{},
    }
}

As explained in the first example, the DenyIfNoViewer privacy rule, denies the operation if the context.Context does not contain the viewer.Viewer information.

Similar to the previous example, we want add a constraint that only admin users can create tenants (and deny otherwise). We do it by copying the AllowIfAdmin rule from above, and adding it to the Policy of the Tenant schema:

examples/privacytenant/ent/schema/tenant.go

// Policy defines the privacy policy of the User.
func (Tenant) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            // For Tenant type, we only allow admin users to mutate
            // the tenant information and deny otherwise.
            rule.AllowIfAdmin(),
            privacy.AlwaysDenyRule(),
        },
    }
}

Then, we expect the following code to run successfully:

examples/privacytenant/example_test.go

func Do(ctx context.Context, client *ent.Client) error {
    // Expect operation to fail, because viewer-context
    // is missing (first mutation rule check).
    if err := client.Tenant.Create().Exec(ctx); !errors.Is(err, privacy.Deny) {
        return fmt.Errorf("expect operation to fail, but got %w", err)
    }
    // Deny tenant creation if the viewer is not admin.
    viewCtx := viewer.NewContext(ctx, viewer.UserViewer{Role: viewer.View})
    if err := client.Tenant.Create().Exec(viewCtx); !errors.Is(err, privacy.Deny) {
        return fmt.Errorf("expect operation to fail, but got %w", err)
    }
    // Apply the same operation with "Admin" role, expect it to pass.
    adminCtx := viewer.NewContext(ctx, viewer.UserViewer{Role: viewer.Admin})
    hub, err := client.Tenant.Create().SetName("GitHub").Save(adminCtx)
    if err != nil {
        return fmt.Errorf("expect operation to pass, but got %w", err)
    }
    fmt.Println(hub)
    lab, err := client.Tenant.Create().SetName("GitLab").Save(adminCtx)
    if err != nil {
        return fmt.Errorf("expect operation to pass, but got %w", err)
    }
    fmt.Println(lab)
    return nil
}

We continue by adding the rest of the edges in our data-model (see image above), and since both User and Group have an edge to the Tenant schema, we create a shared named TenantMixin for this:

// TenantMixin for embedding the tenant info in different schemas.
type TenantMixin struct {
    mixin.Schema
}
// Edges for all schemas that embed TenantMixin.
func (TenantMixin) Edges() []ent.Edge {
    return []ent.Edge{
        edge.To("tenant", Tenant.Type).
            Unique().
            Required(),
    }
}

Next, we may want to enforce a rule that will limit viewers to only query groups and users that are connected to the tenant they belong to. For use cases like this, Ent has an additional type of privacy rule named Filter. We can use Filter rules to filter out entities based on the identity of the viewer. Unlike the rules we previously discussed, Filter rules can limit the scope of the queries a viewer can make, in addition to returning privacy decisions.

examples/privacytenant/rule/rule.go

After creating the FilterTenantRule privacy rule, we add it to the TenantMixin to make sure all schemas that use this mixin, will also have this privacy rule.

examples/privacytenant/ent/schema/mixin.go

// Policy for all schemas that embed TenantMixin.
func (TenantMixin) Policy() ent.Policy {
        Query: privacy.QueryPolicy{
            rule.AllowIfAdmin(),
            // Filter out entities that are not connected to the tenant.
            // If the viewer is admin, this policy rule is skipped above.
            rule.FilterTenantRule(),
    }
}

Then, after running the code-generation, we expect the privacy-rules to take effect on the client operations.

examples/privacytenant/example_test.go

func Do(ctx context.Context, client *ent.Client) error {
    // A continuation of the code-block above.
    // Create 2 users connected to the 2 tenants we created above
    hubUser := client.User.Create().SetName("a8m").SetTenant(hub).SaveX(adminCtx)
    labUser := client.User.Create().SetName("nati").SetTenant(lab).SaveX(adminCtx)
    hubView := viewer.NewContext(ctx, viewer.UserViewer{T: hub})
    out := client.User.Query().OnlyX(hubView)
    // Expect that "GitHub" tenant to read only its users (i.e. a8m).
    if out.ID != hubUser.ID {
        return fmt.Errorf("expect result for user query, got %v", out)
    }
    fmt.Println(out)
    labView := viewer.NewContext(ctx, viewer.UserViewer{T: lab})
    out = client.User.Query().OnlyX(labView)
    // Expect that "GitLab" tenant to read only its users (i.e. nati).
    if out.ID != labUser.ID {
        return fmt.Errorf("expect result for user query, got %v", out)
    }
    fmt.Println(out)
    return nil
}

We finish our example with another privacy-rule named DenyMismatchedTenants on the Group schema. The DenyMismatchedTenants rule rejects group creation if the associated users don’t belong to the same tenant as the group.

examples/privacytenant/rule/rule.go

// DenyMismatchedTenants is a rule that runs only on create operations and returns a deny
// decision if the operation tries to add users to groups that are not in the same tenant.
func DenyMismatchedTenants() privacy.MutationRule {
    return privacy.GroupMutationRuleFunc(func(ctx context.Context, m *ent.GroupMutation) error {
        tid, exists := m.TenantID()
        if !exists {
            return privacy.Denyf("missing tenant information in mutation")
        }
        users := m.UsersIDs()
        // If there are no users in the mutation, skip this rule-check.
        if len(users) == 0 {
            return privacy.Skip
        }
        // Query the tenant-id of all users. Expect to have exact 1 result,
        // and it matches the tenant-id of the group above.
        id, err := m.Client().User.Query().Where(user.IDIn(users...)).QueryTenant().OnlyID(ctx)
        if err != nil {
            return privacy.Denyf("querying the tenant-id %v", err)
        }
        if id != tid {
            return privacy.Denyf("mismatch tenant-ids for group/users %d != %d", tid, id)
        }
        // Skip to the next privacy rule (equivalent to return nil).
        return privacy.Skip
    })
}

We add this rule to the Group schema and run code-generation.

examples/privacytenant/ent/schema/group.go

// Policy defines the privacy policy of the Group.
func (Group) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            // Limit DenyMismatchedTenants only for
            // Create operation
            privacy.OnMutationOperation(
                rule.DenyMismatchedTenants(),
                ent.OpCreate,
            ),
        },
    }
}

Again, we expect the privacy-rules to take effect on the client operations.

examples/privacytenant/example_test.go

func Do(ctx context.Context, client *ent.Client) error {
    // A continuation of the code-block above.
    // Expect operation to fail because the DenyMismatchedTenants rule
    // makes sure the group and the users are connected to the same tenant.
    err = client.Group.Create().SetName("entgo.io").SetTenant(hub).AddUsers(labUser).Exec(adminCtx)
    if !errors.Is(err, privacy.Deny) {
        return fmt.Errorf("expect operation to fail, since user (nati) is not connected to the same tenant")
    }
    err = client.Group.Create().SetName("entgo.io").SetTenant(hub).AddUsers(labUser, hubUser).Exec(adminCtx)
    if !errors.Is(err, privacy.Deny) {
        return fmt.Errorf("expect operation to fail, since some users (nati) are not connected to the same tenant")
    }
    entgo, err := client.Group.Create().SetName("entgo.io").SetTenant(hub).AddUsers(hubUser).Save(adminCtx)
    if err != nil {
        return fmt.Errorf("expect operation to pass, but got %w", err)
    }
    fmt.Println(entgo)
    return nil
}

In some cases, we want to reject user operations on entities that don’t belong to their tenant without loading these entities from the database (unlike the DenyMismatchedTenants example above). To achieve this, we can use the FilterTenantRule rule for mutations as well, but limit it to specific operations as follows:

examples/privacytenant/ent/schema/group.go

// Policy defines the privacy policy of the Group.
func (Group) Policy() ent.Policy {
    return privacy.Policy{
        Mutation: privacy.MutationPolicy{
            // Limit DenyMismatchedTenants only for
            // Create operations
            privacy.OnMutationOperation(
                rule.DenyMismatchedTenants(),
                ent.OpCreate,
            ),
            // Limit the FilterTenantRule only for
            // UpdateOne and DeleteOne operations.
            privacy.OnMutationOperation(
                rule.FilterTenantRule(),
                ent.OpUpdateOne|ent.OpDeleteOne,
            ),
        },

Then, we expect the privacy-rules to take effect on the client operations.

examples/privacytenant/example_test.go

The full example exists in GitHub.

Please note that this documentation is under active development.