Authorization bypass due to cache misconfiguration

4 min readAug 21, 2024

This writeup is about one of my favorite findings as it was a very unexpected issue.

I was testing an ecommerce site.
It had 2 assets in scope.
target.com and admin.target.com

target.com was the user facing portal where users could go and buy items.
admin.target.com was basically the admin portal for sellers where they could list their items, track orders, customer info and so on.

I was testing for idors and access controls. I generally use Autorize for it.

If a lower privilege user is able to hit the admin endpoint Autorize will flag it as “bypassed” .

With the normal user cookie from target.com placed into Autorize , i was using the admin.target.com to check if a normal user can access the admin endpoints.

During my testing something unusual happened.

Every time i visited the endpoint:

https://admin.target.com/orders , following GraphQL request was being made.

POST /graphql
Host: admin.target.com

{"operationName":"GetOrders","variables":{"shop_id":"X"},"query":"query X"}

The response contained all the order information of my shop.
That is an expected behavior.

What was strange however is, Autorize flagged the endpoint as “bypassed” meaning that even a normal user is able to make this request and access order info of my shop.

But when i sent that request to repeater and tried to make request with user cookies it gave an error.

Hmmm🤔

autorize says bypassed , repeater says forbidden.

I assumed it to be a glitch within autorize and moved on.

For the entire week when i was testing the program, it kept happening.

Autorize kept showing the GetOrders endpoint as “bypassed” but when i used to send the request to repeater and test , it gave me the 403 forbidden error.

At this point, I was certain that it isn’t an issue with Autorize, and I am just missing something.

Then it clicked.

Only difference between Autorize and Repeater was the time interval.

While they both had same cookies/token.

Autorize was making immediate call to the admin endpoint.
Where as me making the request from repeater took some time.

To test my theory:

I made a request to GetOrders endpoint using admin token.

POST /graphql
Host: admin.target.com
Auth: Bearer admin

{"operationName":"GetOrders","variables":{"shop_id":"X"},"query":"query X"}

then immediately made the same request with user token.

POST /graphql
Host: admin.target.com
Auth: Bearer user

{"operationName":"GetOrders","variables":{"shop_id":"X"},"query":"query X"}

To my surprise i was able to get all the order info of that shop including customer details.

Issue

So what was happening is:
The server was caching the GetOrders response for a very brief period of 3/4 seconds.

So if an attacker makes the request at the same time when a normal shop admin is using his admin portal, attacker is able to fetch all the order/customer info belonging to any shop just using shop_id.

The shop_id is a publicly accessible id.

Exploit

Create a simple bash script that will make continuous request to the GetOrders endpoint throughout the day

When ever the admin visits their portal , the order/customer info gets cached for a window of 3/4 seconds allowing attacker to fetch them and bypassing all access control restrictions.

POC

I ran my intruder making request to GetOrders endpoint with user token.

It gave a 403 forbidden response initially due to access controls in place.