Share Rules Across Resources
Some applications have many resources that should have similar authorization rules applied to them. This is a common scenario in workflow driven applications that have different user types and a large number of resources.
A common set of rules will apply to many resources, with some exceptions. In this guide, we will cover various ways for modeling this scenario with Oso.
Setup
In this example, we will consider a hypothetical EMR (electronic medical records) application. We’ll discuss a few resources:
- Order: A record of an action that medical staff will perform on a patient
- Test: A diagnostic test that will be performed on a patient.
- Lab: A lab test that will be performed on a patient.
These resources are all examples of different types of patient data.
Basic Policy
Let’s start by considering a basic policy controlling access to these three resources:
allow(actor: User, "read", resource: Order) if
actor.role = "medical_staff" and
actor.treated(resource.patient);
allow(actor: User, "read", resource: Test) if
actor.role = "medical_staff" and
actor.treated(resource.patient);
allow(actor: User, "read", resource: Lab) if
actor.role = "medical_staff" and
actor.treated(resource.patient);
Let’s take a look at the first rule in the policy. This allow
rule permits an
actor to perform the "read"
action on an Order
if:
- The actor’s
role
property is equal to"medical_staff"
. - The actor has treated the patient associated with the
Order
in question, which is verified by calling the actor’streated()
method.
Note the head of the rule. Each argument uses a type specializer to ensure this
rule only applies to certain types of resources and actors. This rule indicates
that the actor
argument must be an instance of the User
class and the
resource
argument must be an instance of the Order
class.
This policy meets our goal above. We have expressed the same rule for the three types of patient data, but it is a bit repetitive. Let’s try to improve it.
Using a Rule to Express Common Behavior
Our policy doesn’t just need to contain allow
rules. We can write any rules
we’d like and compose them as needed to express our policy!
can_read_patient_data(actor, "read", resource) if
actor.role = "medical_staff" and
actor.treated(resource.patient);
allow(actor: User, "read", resource) if
can_read_patient_data(actor, "read", resource);
Now, we’ve taken the repeated logic and expressed it as the
can_read_patient_data
rule. When the allow
rule is evaluated, Oso will
check if the can_read_patient_data
is satisfied. The policy is much shorter!
Unfortunately, we’ve lost one property of our last policy: the specializers. This rule would be evaluated for any type of resource — not just our three examples of patient data above. That’s not what we want.
Bringing Back Specializers
We can combine this idea with our first policy to make sure only our three
patient data resources use the can_read_patient_data
rule.
allow(actor: User, "read", resource: Order) if
can_read_patient_data(actor, "read", resource);
allow(actor: User, "read", resource: Test) if
can_read_patient_data(actor, "read", resource);
allow(actor: User, "read", resource: Lab) if
can_read_patient_data(actor, "read", resource);
Now, we still have three rules, but the body isn’t repeated anymore.
One Rule to Rule Them All
We haven’t talked about the application side of this yet. So far, we’ve assumed
Order
, Lab
, and Test
are application classes.
Here’s how they might be implemented:
class PatientData:
def __init__(self, patient):
self.patient = patient
class Lab(PatientData):
pass
class Order(PatientData):
pass
class Test(PatientData):
pass
We used inheritance to capture some of the common functionality needed (storing the patient). In a real application these would probably be ORM models.
We can use the same idea to shorten our policy even further!
allow(actor: User, "read", resource: PatientData) if
actor.role = "medical_staff" and
actor.treated(resource.patient);
Now, this allow
rule will be evaluated for any instance that is a subclass of
PatientData
. Polar understands the class inheritance structure when selecting
rules to evaluate!
Summary
In this guide, we saw an example of an application policy that could result in significant repetition. We tried out a few strategies for representing common policy across many resource types. First, we wrote a custom rule that moved duplicated logic into one place. Then we used specializers and application types to condense our policy even further.
Connect with us on Slack
If you have any questions, or just want to talk something through, jump into Slack. An Oso engineer or one of the thousands of developers in the growing community will be happy to help.