How to get Query/Mutation operation name

随声附和 提交于 2019-12-06 15:50:43

I think you're thinking of authorization in REST terms here, and it doesn't map well to GraphQL. Instead of a single decision at the top level based on the operation name (or based on the URL in REST), you need a more granular approach. You need to know who's allowed to see/do what at the field level, as the client is allowed make ad-hoc selections.

There's multiple ways to do this, but since you mentioned Spring, you can simply use Spring Security at the service level. If each protected field is backed by a service method (and it should be), you can protect those methods using Spring Security as usual.

Even better, you should also provide a custom GraphqlFieldVisibility implementation, so that unauthorized clients can't even know about the the existence of fields they're not allowed to see in the schema. You can use e.g. Spring's SpelExpressionParser to make decisions on what parts of the schema are visible dynamically, for each user, based on Spring Security rules.

If Spring Security is not an option, you can implement a custom Instrumentation (e.g. by extending SimpleInstrumentation). There you can implement the callbacks like beginExecuteOperation, that will give you access to the parsed query (enough if you really just want to do REST-style top-level auth only), or begin(Deferred)Field (which gives you access to the FieldDefinition) or beginFieldFetch/instrumentDataFetcher (which gives you access to the entire DataFetchingEnvironment) to perform auth per-field.

If you go this way, you can keep the auth information (e.g. the required roles) in the field definition itself as directives. And keep the currently logged in user in the shared context. This way you always have everything you need to do authentication at each level.

In all cases, it's advisable to provide GraphqlFieldVisibility to completely hide the existence of the protected fields contextually.

Here's an abstract example showing the major points using the Instrumentation approach (as you need nothing special for the Spring Security approach, just use Spring Security as usual):

//Checks if the current user has the needed roles for each field
public class AuthInstrumentation extends SimpleInstrumentation {
    @Override
    public DataFetcher<?> instrumentDataFetcher(DataFetcher<?> dataFetcher, InstrumentationFieldFetchParameters parameters) {
        GraphQLFieldDefinition fieldDefinition = parameters.getEnvironment().getFieldDefinition();
        //Each protected field is expected to have a directive called "auth" with an argument called "rolesRequired" that is a list of strings representing the roles
        Optional<GraphQLArgument> rolesRequired = DirectivesUtil.directiveWithArg(fieldDefinition.getDirectives(), "auth", "rolesRequired");
        if (rolesRequired.isPresent()) {
            List<String> roles = (List<String>) rolesRequired.get().getValue();
            User currentUser = parameters.getEnvironment().getContext(); //get the user from context
            if (!currentUser.getRoles().containsAll(roles)) {
                //Replace the normal resolution logic with the one that always returns null (or throws an exception) when the user doesn't have access
                return env -> null;
            }
        }
        return super.instrumentDataFetcher(dataFetcher, parameters);
    }
}

You don't have to store the required roles in the directives, it's just a convenient place. You can get the same info from an external source if it's appropriate.

Then register this instrumentation:

GraphQL graphQL = GraphQL.newGraphQL(schema)
    .instrumentation(new AuthInstrumentation())
    .build();

And when executing a query, put the current user into the context:

//Get the current user's roles however you normally do
User user = loadUser(userName);
ExecutionInput input = ExecutionInput.newExecutionInput()
    .query(operation)
    .context(user) //put the user into context so the instrumentation can get it
    .build()

This way you have everything neatly separated (no auth logic in resolvers, no external context needed) and contextual per field, even without using Spring Security.

Let's go further and make a custom GraphqlFieldVisibility:

public class RoleBasedVisibility implements GraphqlFieldVisibility {

    private final User currentUser;

    public RoleBasedVisibility(User currentUser) {
        this.currentUser = currentUser;
    }

    @Override
    public List<GraphQLFieldDefinition> getFieldDefinitions(GraphQLFieldsContainer fieldsContainer) {
        return fieldsContainer.getFieldDefinitions().stream()
                .filter(field -> isFieldAllowed(field, currentUser))
                .collect(Collectors.toList());
    }

    @Override
    public GraphQLFieldDefinition getFieldDefinition(GraphQLFieldsContainer fieldsContainer, String fieldName) {
        GraphQLFieldDefinition fieldDefinition = fieldsContainer.getFieldDefinition(fieldName);
        return fieldDefinition == null || !isFieldAllowed(fieldDefinition, currentUser) ? null : fieldDefinition;
    }

    private boolean isFieldAllowed(GraphQLDirectiveContainer field, User user) {
        //Same as above, extract this into a common function
        Optional<GraphQLArgument> rolesRequired = DirectivesUtil.directiveWithArg(field.getDirectives(), "auth", "rolesRequired");
        List<String> roles = (List<String>) rolesRequired.get().getValue();
        return currentUser.getRoles().containsAll(roles);
    }
}

As you see, visibility depends on the user, which this time you can not get from the context, so you have to instantiate it per request. This means you need to transform the schema and instantiate GraphQL per request as well. The rest is the same.

GraphQLSchema schema = baseSchema.transform(
    schemaBuilder -> schemaBuilder.fieldVisibility(new RoleBasedVisibility(currentUser)));
GraphQL graphQL = GraphQL.newGraphQL(schema)
        .instrumentation(new AuthInstrumentation())
        .build();

With that, you have a full security setup. Unauthorized users won't even know a field exists if they're not allowed to. If they're allowed to see it in general, but they can only fetch it conditionally, the AuthInstrumentation covers it.

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!