🚧 The package is currently under active development, so it's not recommended for production
Row-Level Security (RLS) traditionally requires databases with native support and custom security policies for each table.
This library provides an alternative: a Prisma client extension that automatically adds "where" clauses to all model queries. This method works without database-side RLS support (e.g., in MySQL).
Note that this extension doesn't apply to raw queries. For those, you must handle them manually or choose database with built-in support.
Define shared types:
import { Prisma, User } from "@prisma/client";
import { PermissionsConfig } from "prisma-rls";
export type Role = "User" | "Guest";
export type PermissionsContext = { user: User | null };
export type RolePermissions = PermissionsConfig<Prisma.TypeMap, PermissionsContext>;
export type PermissionsRegistry = Record<Role, RolePermissions>;
Define permissions for each model in your schema, they also can be a function:
import { RolePermissions } from "./types";
export const User: RolePermissions = {
Post: {
read: { published: { equals: true } },
create: true,
update: (permissionContext) => ({
author: {
id: { equals: permissionContext.user.id },
},
}),
delete: (permissionContext) => ({
author: {
id: { equals: permissionContext.user.id },
},
}),
},
User: {
read: (permissionContext) => ({
id: { equals: permissionContext.user.id },
}),
create: false,
update: (permissionContext) => ({
id: { equals: permissionContext.user.id },
}),
delete: false,
},
}
In most cases, you will have multiple roles. To define them in a type-safe manner, follow this pattern:
import { guest } from "./guest";
import { user } from "./user";
import { PermissionsRegistry } from "./types";
export const permissionsRegistry = {
User: user,
Guest: guest,
} satisfies PermissionsRegistry;
Prisma extensions don't support dynamic contexts, so create an extension per context:
import { Prisma, PrismaClient } from "@prisma/client";
import Fastify, { FastifyRequest } from "fastify";
import { createRlsExtension } from "prisma-rls";
import { permissionsRegistry, PermissionsContext } from "./permissions";
(async () => {
const prisma = new PrismaClient();
const server = Fastify();
const resolveRequestConext = async (request: FastifyRequest) => {
const user = await resolveUser(request.headers.authorization);
const userRole = user ? user.role : "Guest";
const permissionsContext: PermissionsContext = { user };
const rlsExtension = createRlsExtension({
dmmf: Prisma.dmmf,
permissionsConfig: permissionsRegistry[userRole],
context: permissionsContext,
});
return { db: prisma.$extends(rlsExtension) };
};
server.get("/user/list", async function handler(request, reply) {
const { db } = resolveRequestConext(request);
return await db.user.findMany();
});
await server.listen({ port: 8080, host: "0.0.0.0" });
})();
After that, all non-raw queries will be executed according to the defined permissions.
At the moment there is only a known edge case.
An edge case affects all mandatory belongs-to relations on the owner side (the entity owns the foreign key).
In these cases, Prisma does not generate filters due to potential referential integrity violations. For performance reasons, no additional checks are applied by default. However, you can change this behavior by enabling the checkRequiredBelongsTo
flag:
const rlsExtension = createRlsExtension({
dmmf: Prisma.dmmf,
permissionsConfig: permissionsRegistry[userRole],
context: permissionsContext,
checkRequiredBelongsTo: true,
});
When checkRequiredBelongsTo
is set to true, the library performs an additional query for each required belongs-to relation (it makes one batched request per relation, not per record) to verify that all data complies with the current permissions.
If the policy restricts data access, an error will be thrown, this error should be handled at the application level:
try {
return await db.post.findMany({
select: { id: true, category: { select: { id: true } } }
});
} catch (error) {
if (error instanceof Error && error.message === "Referential integrity violation") {
return [];
}
throw error;
}
Alternatively, you can consider the following options:
- Make them optional - keep the foreign keys required but define the relationships as optional
- Handle at the policy level - apply consistent policy filters to restrict access to both sides of relation