Skip to main content
Version: v2.x

Permissions Examples

Introduction

This is a guide to help you set up a basic authorization architecture for your GraphQL fields. It is recommended that you first check out Roles & Session variables and other pages in configuring permission rules which will be referred to throughout this guide.

Here are some examples of common use cases.

Unauthorized users

Unauthorized users are those which are unauthenticated or not logged in and thus default to the defined unauthorized role. Requests from these users will typically have no officially identifiable session variables. Follow these steps in order to create permission rules for an anonymous user.

  • Create a role called anonymous (this name is up to you, you could call the role public or anything else).
  • We'll be adding select permissions. Generally, for security, you wouldn't be adding insert, update, or delete permissions.
  • For the select permission condition, create a valid condition depending on your data model. For example, is_published: {_eq: true} or if you don't have a condition, then just set the permission to Without any checks, which is represented by an empty set of braces {}.
  • Choose the right set of columns that will get exposed in the GraphQL schema as fields. Ensure that sensitive columns are not exposed.

Example

Here's an example of allowing the defined unauthorized role, named in this case: anonymous, to query the products table for only rows where the is_published column is set to true.

Access control for an anonymous role

We are also limiting the columns which can be returned by the query. If we wanted to allow the anonymous role to be able to query the products table for all rows, we would simply set the "Row select permissions" to "Without any checks".

See Unauthenticated / Public access for steps to configure the unauthorized user role in Hasura.

Logged-in users

  • Create a role called user.
  • Permissions in this case make use of a user_id or a owner_id column.
  • Set up a permission for an insert/select/update/delete operation that uses said column. E.g.: user_id: {_eq: "X-Hasura-User-Id"} for a profile or shopping carts table.
  • Note that the X-Hasura-User-Id is a dynamic session variable that is set and returned from your auth service, or as a request header.

Example

Here's an example of creating a permission which allows only users to access their own shopping carts. The user_id column is a column in the carts table which stores the id of the user who created it. We are restricting access to rows of this table to requests which contain the same user id in the X-Hasura-User-Id session variable.

Access control for a logged-in user

Now we can test out a query to see that when posing as that user, we can access only their carts and not those of other users.

API testing for a logged-in user
Posing as a user in testing

In development, if you're testing your logged-in users' access and aren't utilizing authenticated tokens, you must include the X-Hasura-Admin-Secret header. You can learn more about this here.

Same table ownership information

Suppose you have a multi-tenant application where a manager of a particular organization should be able to see all the data that belongs to that particular organization. In this case, the particular table might have an id column which denotes the organization.

Let's say we have an online store where each vendor on the store has its own products. We want to allow the manager of a vendor to be able to see all the products that belong to that vendor and the indentifying vendor_id is saved on the products table itself.

  • Create a role called manager.
  • Create a permission for select in the products table, which has the condition: vendor_id: {_eq: "X-Hasura-Vendor-Id"}.
  • X-Hasura-Vendor-Id is a session variable which is set and returned by your auth service for an incoming request.
Access control for a manager of an vendor
GraphQL query for a manager of an vendor

Let's say the ownership or visibility information for a data model (table) is not present as a column in the table, but in a different related table.

For example, suppose that in a products table we have a added_by_user_id column which stores the id of the user who added the product but no vendor_id column on the table itself.

However, we want to allow other members of the vendor to be able to access the product too. We have another related table users_in_vendors which associates users with vendors. The relationship from the product to the users_in_vendors table is named userInVendorByUserId and is configured as: products.added_by_user_id → users_in_vendors.user_id.

We can use this relationship to check that the X-Hasura-Vendor-Id on the incoming request session variable matches the vendor which the user that added the product is a member of and therefore allow other members of the vendor organization to access the product.

  • Create a relationship called userInVendorByUserId from the product table.
    • Object relationship (product has only a single user which added it): products.added_by_user_id → users_in_vendors.user_id.
  • Create a role called manager.
  • Create a select permission on the products table, which has the condition: {"userInVendorByUserId":{"vendor_id":{"_eq":"X-Hasura-Vendor-Id"}}}.
    • This reads as: Allow the role manager to select if users_in_vendors.vendor_id has a vendor_id equal to that of X-Hasura-Vendor-Id session variable on the incoming request.
Different table ownership permissions
Different table ownership permissions

Multiple roles per user

Sometimes your data model requires that:

  • Users can have multiple roles.
  • Each role has access to different parts of your database schema.

If you have the information about roles and how they map to your data in the same database as the one configured with the Hasura Engine, you can leverage relationships to define permissions that effectively control access to data and the operations each role is allowed to perform.

To understand how this works, let's model the roles and corresponding permissions in the context of a blog app with the following roles:

  • author: Users with this role can submit their own articles.
  • reviewer: Users with this role can review articles assigned to them and add a review comment to each article. A mapping of articles to reviewers is maintained in the reviewers table.
  • editor: Users with this role can edit and publish any article. They can also leave a private rating for each article. However, they cannot overwrite a reviewer's notes. A list of editors is maintained in the editors table.

Database Schema

The following is a reference database schema for our example:

Database schema example for multiple roles per user

Based on the above schema, we'll create the following tables:

-- user information from your auth system

users (
id INT PRIMARY KEY,
name TEXT,
profile JSONB, -- some profile information like display_name, etc.
registered_at TIMESTAMP -- the time when this user registered
)

-- information about articles

articles (
id INTEGER PRIMARY KEY,
title TEXT,
author_id INT REFERENCES users(id), -- Foreign key to users :: id
is_reviewed BOOLEAN DEFAULT FALSE,
review_comment TEXT,
is_published BOOLEAN DEFAULT FALSE,
editor_rating INTEGER
)

-- mapping of reviewers to articles

reviewers (
id INTEGER PRIMARY KEY,
article_id INTEGER REFERENCES articles(id), -- Foreign key to articles :: id
reviewer_id INTEGER REFERENCES users(id) -- Foreign key to users :: id
)

-- a list of editors

editors (
editor_id INTEGER PRIMARY KEY REFERENCES users(id) -- Foreign key to users :: id
)

Relationships

Create an array relationship named reviewers based on the foreign key constraint reviewers :: article_idarticles :: id:

Create an array relationship

Permissions

The following is an example summary of the access control requirements for the articles table based on the above schema:

Client Nameauthorreviewereditor
insertselectupdateselectupdateselect
id
title
author_id
is_reviewed
review_comment
is_published
editor_rating

Additional restrictions are required to ensure that a user with the role author can submit only their own article i.e. author_id should be the same as the user's id.

We'll create permission rules for the roles and Actions listed above (you can easily extend them for the Actions not documented here) .

Permissions for role author

  • Allow users with the role author to insert only their own articles

    For this permission rule, we'll make use of two features of the GraphQL Engine's permissions system:

    • Column-level permissions: Restrict access to certain columns only.
    • Column presets: Session-variable-based column preset for the author_id column to automatically insert the user's ID i.e. the X-Hasura-User-Id session-variable's value. It also helps us avoid explicitly passing the user's ID in the insert mutation.
Permissions for the role author

Notice how we don't need to have an explicit row-level permission (a custom check) as only authenticated users with the role author can perform this action. As we have a column preset for the author_id column that automatically takes the author's ID (and the id column is an auto-increment integer field), we only need to allow access to the title column.

  • Allow users with the role author to select certain columns only

Again, we'll use column-level permissions to restrict access to certain columns. Additionally, we need to define row-level permissions (a custom check) to restrict access to only those articles authored by the current user:

Column access for the role author

The row-level permission rule shown here translates to "if the value in the author_id column of this row is equal to the user's ID i.e. the* X-Hasura-User-Id _session-variable's value, allow access to it".

Permissions for role reviewer

  • Allow users with the role reviewer to update articles assigned to them for reviews

    For this use-case, we'll use relationship or nested-object permissions based on the array relationship reviewers to restrict access to assigned articles only.

Permissions for the role reviewer

The array-relationship based permission rule in the above image reads as "if the ID of any reviewer assigned to this article is equal to the user's ID i.e. the X-Hasura-User-Id session-variable's value, allow access to it". The columns' access is restricted using the column-level permissions highlighted above.

  • Allow users with the role reviewer to select articles assigned to them for reviews

    This permission rule is pretty much the same as the one for update, the only difference being the column-level permissions.

Column access for the role reviewer

Permissions for role editor

  • Allow editors to select any article's data

    This is a straightforward rule - there's no need for any row-level permissions since editors have access to all rows and they can read all columns.

Permissions for the role editor
  • Allow editors to update an article

    There's no need for row-level permissions in this case either but we need to restrict access to certain columns only:

Column access for the role editor

Multiple Permissions for the Same Role

In some cases we might want to allow access to certain columns for a role only if a condition is met, while allowing access to other columns based on a different condition.

Currently, it is not possible to define multiple column & row permission combinations for the same role. However, we can work around this limitation by using views.

Example

Let's say for privacy reasons we only want users to be able to access their own email, phone and address but allow all users to access each others name and city information.

We have a table called user_profile with columns (id, name, city, email, phone, address) and we want the role user to be able to access:

  • all columns only if the id column is the requesting user's id, i.e. the current user is the "owner" of the row.
  • only the id, name and city columns for all other users.

We can achieve this via the following steps:

Step 1: Create a view

Create a view called user_profile_private with columns (user_id, email, phone, address):

CREATE VIEW user_profile_private AS
SELECT id AS user_id, email, phone, address
FROM user_profile;
Create a view

Step 2: Create a relationship

For the table user_profile, create a manual object relationship called user_profile_private using user_profile.id -> user_profile_private.user_id:

Create a manual object relationship

Step 3: Define permissions

On the user_profile table for the role user, create the following permissions for select:

  • allow access to id, name and city without any row conditions.
Column access for the role user
  • On the user_profile_private view allow access to id, phone, email and address if the user-id passed in the session variable is equal to the row's user_id.
Column access for the role user based on row level permissions

Step 4: Query with appropriate access control

Now we can fetch the required data with the appropriate access control by using the relationship.

If the X-Hasura-Role and the X-Hasura-User-Id session variables are set to user and 2 respectively (and the X-Hasura-Admin-Secret value is set to allow us to pose as the user), we'll get the following result:

Column access for the role user based on row level permissions

Observe that the user_profile_private field is returned as null for all rows without the appropriate access.