Okay, even with the number of diagrams in my last post, there was still some confusion. So, I’m going to try to make it totally clear. Emphasis on “try”.
Regression Testing isn’t just about regression testing one part of a solution
Firstly, I got some feedback from SAP Support when I tried to raise this issue with them. They didn’t seem to understand my concern that I tested the whole recruit to hire to fire employee life cycle every time that SAP released new functionality for SuccessFactors in the preview environment before it got released to the productive environment. This was because:
“Since IAS productive and test instances are of the same version, there is from IdP perspective no difference at all.“
…
“As already mentioned above: IAS test- and productive instances are of the same release. There is no such thing as a preview environment for IAS.
Thus again one single IAS tenant will be fully sufficient to handle the described scenarios from IdP perspective“
I’m not sure how I can make this clearer. But, yes, I do want to test that during the preview release update period the changes that SuccessFactors make do not impact the provisioning and login processes that I have configured in IPS/IAS. I also want to ensure as well as ensure that these are working in the non-prod environment so that if I push up a change to fix a release impact, it doesn’t break stuff. I’m confused how the Identity services teams seem to think their solutions works in isolation! If it isn’t for the systems that use their services and provide the user details the solutions aren’t worth anything!
How does this all look if we try to connect it up?
Okay – here’s a relatively complex diagram. It shows how you’d have to wire up the provisioning and authentication trust when using more that one SuccessFactors system with a single IAS.
But there is the slim possibility it could work!
And it could work provided that you:
- Did some configuration in the IPS and IAS that SuccessFactors has not at all been documented for customers
- that populated userid into different custom attributes of the IAS user record per system
- then used that system specific field in the assertion sent to the different systems.
- AND – never maintained different employee attributes for the same email address in the two systems (if you don’t want the system to get hella confused.)
Unfortunately that’s about as likely as me winning the Powerball lottery. Whilst the first point is terribly technical and pedantic (therefore will be loved by half the people I know and hated by the other half) the second point pretty much means this is never going to happen. The reason for the second restriction is that you cannot have the same email address against multiple “user” records in IAS. Whilst through some technical wizardry I might get the same record to point to two different system ids, whether it is active or inactive is a non-system specific piece of user data. What name that user has is a non-system specific piece of user data.
So assume I do have the same email address assigned to an employee in both systems A and B. Terminating that employee in system A will cause a delta in the employee record that will get picked up by the IPS and deactivate the user in the IAS. Even though they should be active to be allowed to log in to system B, they won’t be able to as they are now “inactive”.
It is possible that customers will just decide that they will ensure consistency between user records in different systems, use same name, have both active, have both inactive, but I very much doubt it.
Likewise not using the same email address for employees in both systems is going to be hard (not to mention hard to track if it mistakenly happens).
Would be nice, but unlikely
In my previous post, I assumed that the previous scenario (especially as it involves undocumented configuration) would not be customers’ default. There I assumed that customers would use the default configuration that is deployed when a customer implements the SuccessFactors IAS “upgrade”. That then allows for all sorts of mischief!
How to abuse authentication when you control the data it is based upon.
Let’s have some fun. Assume that default IPS configuration is being used and the employee records from systems A and B are both trying to update the IAS user master record. Assume system B contains sensitive payroll data (for example it is a copy of productive system). Only Annie Admin has the roles in system B to see this data. I, Chris Creative, have access to system A where I’m doing some project work. I have a role where I can hire and fire employees (tends to happen in HR systems!)
- I firstly terminate an employee with Annie Admin’s email address in system A. If there isn’t one, I’ll just hire her then term her.
- This will trigger the IPS to update the IAS with the user with Annie’s email address as inactive so Annie can’t log into either system and stop my nefarious fun.
- Then I hire a new employee that has the same personnel number as Annie had in system B and put a newly generated email address (that I have access to) against the employee’s record
- I get emailed by the IAS that I have a new user record set up 🙂 what a cool system! I set/reset the password.
- Now I can use this email address to access system B and it thinks I’m logging in as Annie! Which is awesome as when they try to trace who downloaded all the payroll details from the system the audit logs are going to clearly point to Annie’s user! (Okay yes a bit of further checking and someone might see that it was my IP address, but I can hide that using a VPN, the audit logging in IAS is hard to get to (API access only) so going to be hard for them to find the random gmail address I made up and trace it.
- I access a bunch of data I’m not supposed to
- I go back to system A, change the email address of the fake Annie back to her email.
- She gets reactivated
- I just got away with accessing a system that I had no rights to access because I had rights to another system that was provisioning the same authentication system.
Are you worried yet?
Can you see now why I’m a little worried about this setup? This is why I also assumed that to prevent this sort of thing happening, most organisations if forced down the path of using just one IAS will choose the most secure SuccessFactors system to provision user data to the IAS. But that, as I wrote about in the previous post will cause a bunch of other problems.
Simple solution
In case it’s not obvious enough, there is a simple solution. Provide another IAS instance per SuccessFactors instance that a customer has.