Secure Data Connect with authorization and attestation

Firebase Data Connect provides robust client-side security with:

• Mobile and web client authorization
• Individual query- and mutation-level authorization controls
• App attestation with Firebase App Check.

Data Connect extends this security with:

• Server-side authorization
• Firebase project and Cloud SQL user security with IAM.

Authorize client queries and mutations

Data Connect is fully integrated with Firebase Authentication, so you can use rich data about users who are accessing your data (authentication) in your design for what data those users can access (authorization).

Data Connect provides an @auth directive for queries and mutations that lets you set the level of authentication required to authorize the operation. This guide introduces the @auth directive, with examples.

In addition, Data Connect supports execution of queries embedded in mutations, so you can retrieve additional authorization criteria you've stored in your database, and use those criteria in @check directives to decide if enclosing mutations are authorized. For this authorization case, the @redact directive allows you control whether query results are returned to clients in the wire protocol and the embedded query omitted in generated SDKs. Find the introduction to these directives, with examples.

Understand the @auth directive

You can parameterize the @auth directive to follow one of several preset access levels that cover many common access scenarios. These levels range from PUBLIC (which allows queries and mutations from all clients without authentication of any kind) to NO_ACCESS (which disallows queries and mutations outside of privileged server environments using the Firebase Admin SDK). Each of these levels is correlated with authentication flows provided by Firebase Authentication.

Using these preset access levels as a starting point, you can define complex and robust authorization checks in the @auth directive using where filters and Common Expression Language (CEL) expressions evaluated on the server.

Use the @auth directive to implement common authorization scenarios

The preset access levels are the starting point for authorization.

The USER access level is the most widely useful basic level to start with.

Fully secure access will build on the USER level plus filters and expressions that check user attributes, resource attributes, roles and other checks. The USER_ANON and USER_EMAIL_VERIFIED levels are variations on the USER case.

Expression syntax lets you evaluate data using an auth object representing authentication data passed with operations, both standard data in auth tokens and custom data in tokens. For the list of fields available in the auth object, see the reference section.

There are of course use cases where PUBLIC is the correct access level to start with. Again, an access level is always a starting point, and additional filters and expressions are needed for robust security.

This guide now gives examples of how to build on USER and PUBLIC.

A motivating example

The following best practice examples refer to the following schema for a blogging platform with certain content locked behind a payment plan.

Such a platform would likely model Users andPosts.

type User @table(key: "uid") {
  uid: String!
  name: String
  birthday: Date
  createdAt: Timestamp! @default(expr: "request.time")
}

type Post @table {
  author: User!
  text: String!
  # "one of 'draft', 'public', or 'pro'"
  visibility: String! @default(value: "draft")
  # "the time at which the post should be considered published. defaults to
  # immediately"
  publishedAt: Timestamp! @default(expr: "request.time")
  createdAt: Timestamp! @default(expr: "request.time")
  updatedAt: Timestamp! @default(expr: "request.time")
}

User-owned resources

Firebase recommends that you write filters and expressions that test user ownership of a resource, in the following cases, ownership of Posts.

In the following examples, data from auth tokens is read and compared using expressions. The typical pattern is to use expressions like where: {authorUid: {eq_expr: "auth.uid"}} to compare a stored authorUid to the auth.uid (user ID) passed in the authentication token.

Create

This authorization practice starts by adding the auth.uid from the auth token to each new Post as an authorUid field to allow comparison in subsequence authorization tests.

# Create a new post as the current user
mutation CreatePost($text: String!, $visibility: String) @auth(level: USER) {
  post_insert(data: {
    # set the author's uid to the current user uid
    authorUid_expr: "auth.uid"
    text: $text
    visibility: $visibility
  })
}

Update

When a client attempts to update a Post, you can test the passed auth.uid against the stored authorUid.

# Update one of the current user's posts
mutation UpdatePost($id: UUID!, $text: String, $visibility: String) @auth(level:USER) {
  post_update(
    # only update posts whose author is the current user
    first: { where: {
      id: {eq: $id}
      authorUid: {eq_expr: "auth.uid"}
    }}
    data: {
      text: $text
      visibility: $visibility
      # insert the current server time for updatedAt
      updatedAt_expr: "request.time"
    }
  )
}

Delete

The same technique is used to authorize delete operations.

# Delete one of the current user's posts
mutation DeletePost($id: UUID!) @auth(level: USER) {
  post_delete(
    # only delete posts whose author is the current user
    first: { where: {
      id: {eq: $id}
      authorUid: {eq_expr: "auth.uid"}
    }}
  )
}

# Common display information for a post
fragment DisplayPost on Post {
  id, text, createdAt, updatedAt
  author { uid, name }
}

List

# List all posts belonging to the current user
query ListMyPosts @auth(level: USER) {
  posts(where: {
    userUid: {eq_expr: "auth.uid"}
  }) {
    # See the fragment above
    ...DisplayPost
    # also show visibility since it is user-controlled
    visibility
  }
}

Get

# Get a post only if it belongs to the current user
query GetMyPost($id: UUID!) @auth(level: USER) {
  post(key: {id: $id},
    first: {where: {
      id: {eq: $id}
      authorUid: {eq_expr: "auth.uid"}}
      }}, {
      # See the fragment above
      ...DisplayPost
      # also show visibility since it is user-controlled
      visibility
  }
}

Filter Data

Data Connect's authorization system lets you write sophisticated filters combined with preset access levels like PUBLIC as well as by using data from auth tokens.

The authorization system also lets you use expressions only, without a base access level, as shown in some of the following examples.

Filter by resource attributes

Here, authorization is not based on auth tokens since the base security level is set to PUBLIC. But, we can explicitly set records in our database as suitable for public access; assume we have Post records in our database with visibility set to "public".

# List all posts marked as 'public' visibility
query ListPublicPosts @auth(level: PUBLIC) {
  posts(where: {
    # Test that visibility is "public"
    visibility: {eq: "public"}
    # Only display articles that are already published
    publishedAt: {lt_expr: "request.time"}
  }) {
    # see the fragment above
    ...DisplayPost
  }
}

Filter by user claims

Here, assume you've set up custom user claims that pass in auth tokens to identify users in a "pro" plan for your app, flagged with an auth.token.plan field in the auth token. Your expressions can test against this field.

# List all public or pro posts, only permitted if user has "pro" plan claim
query ProListPosts @auth(expr: "auth.token.plan == 'pro'") {
  posts(where: {
    # display both public posts and "pro" posts
    visibility: {in: ['public', 'pro']},
    # only display articles that are already published
    publishedAt: {lt_expr: "request.time"},
  }) {
    # see the fragment above
    ...DisplayPost
    # show visibility so pro users can see which posts are pro\
    visibility
  }
}

Filter by order + limit

Or again, you may have set visibility in Post records to identify they are content available for "pro" users, but for a preview or teaser listing of data, further limit the number of records returned.

# Show 2 oldest Pro post as a preview
query ProTeaser @auth(level: USER) {
  posts(
    where: {
      # show only pro posts
      visibility: {eq: "pro"}
      # that have already been published more than 30 days ago
      publishedAt: {lt_time: {now: true, sub: {days: 30}}}
    },
    # order by publish time
    orderBy: [{publishedAt: DESC}],
    # only return two posts
    limit: 2
  ) {
    # See the fragment above
    ...DisplayPost
  }
}

Filter by role

If your custom claim defines an admin role, you can test and authorize operations accordingly.

# List all posts unconditionally iff the current user has an admin claim
query AdminListPosts @auth(expr: "auth.token.admin == true") {
  posts { ...DisplayPost }
}

Understand the @check and @redact directives

The @check directive verifies that specified fields are present in query results. A Common Expression Language (CEL) expression is used to test field values. The default behavior of the directive is to check for and reject null-valued nodes.

The @redact directive redacts a part of the response from the client. Redacted fields are still evaluated for side effects (including data changes and @check) and the results are still available to later steps in CEL expressions.

In Data Connect, the @check and @redact directives are most often used in the context of authorization checks; refer to the discussion of authorization data lookup.

Add the @check and @redact directives to look up authorization data

A common authorization use case involves storing custom authorization roles in your database, for example in a special permissions table, and using those roles to authorize mutations to create, update, or delete data.

Using authorization data lookups, you can query for roles based on a userID and use CEL expressions to decide if the mutation is authorized. For example, you might want to write an UpdateMovieTitle mutation that lets an authorized client update movie titles.

For the rest of this discussion, assume the movie review app database stores an authorization role in a MoviePermission table.

# MoviePermission
# Suppose a user has an authorization role with respect to records in the Movie table
type MoviePermission @table(key: ["doc", "userId"]) {
  movie: Movie! # implies another field: movieId: UUID!
  userId: String! # Can also be a reference to a User table, doesn't matter
  role: String!
}

In the following example implementation, the UpdateMovieTitle mutation includes a query field to retrieve data from MoviePermission, and the following directives to ensure the operation is secure and robust:

• A @transaction directive to ensure all authorization queries and checks are completed or fail atomically.
• The @redact directive to omit query results from the response, meaning our authorization check is performed on the Data Connect server but sensitive data is not exposed to the client.

• A pair of @check directives to evaluate authorization logic on query results, such as testing that a given userID has an appropriate role to make modifications.

mutation UpdateMovieTitle($movieId: UUID!, $newTitle: String!) @auth(level: USER) @transaction {
  # Step 1: Query and check
  query @redact {
    moviePermission( # Look up a join table called MoviePermission with a compound key.
      key: {movieId: $movieId, userId_expr: "auth.uid"}
    # Step 1a: Use @check to test if the user has any role associated with the movie
    # Here the `this` binding refers the lookup result, i.e. a MoviePermission object or null
    # The `this != null` expression could be omitted since rejecting on null is default behavior
    ) @check(expr: "this != null", message: "You do not have access to this movie") {
      # Step 1b: Check if the user has the editor role for the movie
      # Next we execute another @check; now `this` refers to the contents of the `role` field
      role @check(expr: "this == 'editor'", message: "You must be an editor of this movie to update title")
    }
  }
  # Step 2: Act
  movie_update(id: $movieId, data: {
    title: $newTitle
  })
}

Antipatterns to avoid in authorization

The previous section covers patterns to follow when using the @auth directive.

You should also be aware of important antipatterns to avoid.

Avoid passing user attributes IDs and auth token parameters in query and mutation arguments

Firebase Authentication is a powerful tool for presenting authentication flows and securely capturing authentication data such as registered user IDs and numerous fields stored in auth tokens.

It's not a recommended practice to pass user IDs and auth token data in query and mutation arguments.

# Antipattern!
# This incorrectly allows any user to view any other user's posts
query AllMyPosts($userId: String!) @auth(level: USER) {
  posts(where: {authorUid: {eq: $userId}}) {
    id, text, createdAt
  }
}

Avoid using the USER access level without any filters

As discussed several times in the guide, the core access levels like USER, USER_ANON, USER_EMAIL_VERIFIED are baselines and starting points for authorization checks, to be enhanced with filters and expressions. Using these levels without a corresponding filter or expression that checks which user is performing the request is essentially equivalent to using the PUBLIC level.

# Antipattern!
# This incorrectly allows any user to view all documents
query ListDocuments @auth(level: USER) {
  documents {
    id
    title
    text
  }
}

Avoid using PUBLIC or USER access level for prototyping

To speed up development, it can be tempting to set all operations to the PUBLIC access level or to USER access level without further enhancements to authorize all operations and let you quickly test your code.

When you've done very initial prototyping this way, begin to switch from NO_ACCESS to production-ready authorization with PUBLIC and USER levels. However, don't deploy them as PUBLIC or USER without adding additional logic as shown in this guide.

# Antipattern!
# This incorrectly allows anyone to delete any post
mutation DeletePost($id: UUID!) @auth(level: PUBLIC) {
  post: post_delete(
    id: $id,
  )
}

Use Firebase App Check for app attestation

Authentication and authorization are critical components of Data Connect security. Authentication and authorization combined with app attestation makes for a very robust security solution.

With attestation through Firebase App Check, devices running your app will use an app or device attestation provider that attests that Data Connect operations originate from your authentic app and requests originate from an authentic, untampered device. This attestation is attached to every request your app makes to Data Connect.

To learn how to enable App Check for Data Connect and include its client SDK in your app, have a look at the App Check overview.

Authentication levels for the @auth(level) directive

The following table lists all standard access levels and their CEL equivalents. Authentication levels are listed from broad to narrow -- each level encompasses all users who match following levels.

CEL Reference for @auth(expr) and @check(expr)

As shown in examples elsewhere in this guide, you can and should use expressions defined in Common Expression Language (CEL) to control authorization for Data Connect using @auth(expr:) and @check directives.

This section covers CEL syntax relevant to creating expressions for these directives.

Complete reference information for CEL is provided in the CEL specification.

Test variables passed in queries and mutations

@auth(expr) syntax allows you access and test variables from queries and mutations.

For example, you can include an operation variable, such as $status, using vars.status.

mutation Update($id: UUID!, $status: Any) @auth(expr: "has(vars.status)")

Data available to expressions

Both @auth(expr:) and @check(expr:) CEL expressions can evaluate the following:

• request.operationName
• vars (alias for request.variables)
• auth (alias for request.auth)

Additionally, @check(expr:) expressions can evaluate:

• this (the value of the current field)

The request.operationName object

The request.operarationName object stores the type of operation, either query or mutation.

The vars object

The vars object allows your expressions to access all variables passed in your query or mutation.

You can use vars.<variablename> in an expression as an alias for the fully-qualified request.variables.<variablename>:

# The following are equivalent
mutation StringType($v: String!) @auth(expr: "vars.v == 'hello'")
mutation StringType($v: String!) @auth(expr: "request.variables.v == 'hello'")

The auth object

Authentication identifies users requesting access to your data and provides that information as an object you can build on in your expressions.

In your filters and expressions, you can use auth as an alias for request.auth.

The auth object contains the following information:

• uid: A unique user ID, assigned to the requesting user.
• token: A map of values collected by Authentication.

For more details on the contenst of auth.token see Data in auth tokens

The this binding

The binding this evaluates to the field that the @check directive is attached to. In a basic case, you might evaluate single-valued query results.

mutation UpdateMovieTitle($movieId: UUID!, $newTitle: String!) @auth(level: USER) @transaction {
  # Step 1: Query and check
  query @redact {
    moviePermission( # Look up a join table called MoviePermission with a compound key.
      key: {movieId: $movieId, userId_expr: "auth.uid"}
    ) {
      # Check if the user has the editor role for the movie. `this` is the string value of `role`.
      # If the parent moviePermission is null, the @check will also fail automatically.
      role @check(expr: "this == 'editor'", message: "You must be an editor of this movie to update title")
    }
  }
  # Step 2: Act
  movie_update(id: $movieId, data: {
    title: $newTitle
  })
}

If the returned field occurs multiple times because any ancestor is a list, each occurrence is tested with this bound to each value.

For any given path, if an ancestor is null or [], the field won't be reached and the CEL evaluation will be skipped for that path. In other words, evaluation only takes place when this is null or non-null, but never undefined.

When the field itself is a list or object, this follows the same structure (including all decendents selected in case of objects), as illustrated in the following example.

mutation UpdateMovieTitle2($movieId: UUID!, $newTitle: String!) @auth(level: USER) @transaction {
  # Step 1: Query and check
  query {
    moviePermissions( # Now we query for a list of all matching MoviePermissions.
      where: {movieId: {eq: $movieId}, userId: {eq_expr: "auth.uid"}}
    # This time we execute the @check on the list, so `this` is the list of objects.
    # We can use the `.exists` macro to check if there is at least one matching entry.
    ) @check(expr: "this.exists(p, p.role == 'editor')", message: "You must be an editor of this movie to update title") {
      role
    }
  }
  # Step 2: Act
  movie_update(id: $movieId, data: {
    title: $newTitle
  })
}

Complex expression syntax

You can write more complex expressions by combining with the && and || operators.

mutation UpsertUser($username: String!) @auth(expr: "(auth != null) && (vars.username == 'joe')")

The following section describes all available operators.

Operators and operator precedence

Use the following table as a reference for operators and their corresponding precedence.

Given arbitrary expressions a and b, a field f, and an index i.

Data in auth tokens

The auth.token object may contain the following values:

Additional fields in JWT ID tokens

You can also access the following auth.token fields:

What's next?

• Firebase Data Connect provides an Admin SDK to let you perform queries and mutations from privileged environments.
• Learn about IAM security in the guide for managing services and databases.