Archive for May, 2015

Appliances are Dangerous (because nobody patches them)

Tuesday, May 26th, 2015

Putting sensitive infrastructure on physical or virtual appliances, rather than running it as a traditional on-premise application or a newer software-as-a-service system is a security disaster just waiting to happen.

Why? Because unlike on-premise applications and also unlike the servers running SaaS applications, there is no guarantee that anyone will apply critical security patches to your appliances, either at all or on time. Systems with unpatched security vulnerabilities are an open door to your otherwise secure infrastructure. Tolerate them at your peril.

I just recently spoke with a customer of ours who had – a few years ago – deployed a privileged access management product from one of our competitors. That product includes one or more “jump servers” which mediate login sessions from the desktops of authorized users to logins on managed endpoint systems. Such a “jump server” architecture is common in the privileged access management product category.

The problem for this customer has been that these jump servers — which have access to the most sensitive passwords in the company — run on the original Windows 2008 Server OS (i.e., before the first service pack was released). Since the vendor has made custom changes to the OS to “harden” it, it has been impossible to patch the OS on these jump servers. As a result, today, these jump servers run an OS that was released on February 27, 2008. The OS was released 2,645 days ago. 7.25 years ago. Our customer is scrambling to rip out this product, which endangers their entire infrastructure (it also has performance problems, but that’s another story).

Just think about how many security exploits have been discovered for and patched on Windows 2008+IIS since this platform was released on 2008-02-27. This recently discovered vulnerability comes to mind:

HTTP.sys

Using this particularly dangerous vulnerability, an attacker can remotely gain full SYSTEM privileges on any Windows system running IIS. Yup – including Windows 2008. This exploit is being actively leveraged in the wild, so the risk is very real.

Imagine that your privileged access management system — or any other critical infrastucture — runs on an old, unpatched OS like this. How secure would your organization be?

Is it ever OK to use appliances — physical or virtual — instead of just installing software on a well managed OS image?

I can think of only two cases where appliances are acceptable:

  1. Physical appliances which incorporate specialized hardware, to perform some task very fast. There is simply no software alternative to custom ASICs.
  2. Appliances (physical or virtual) with an automatically managed patch system. i.e., they should run a stock OS and be subject to automatic and timely patches from all the software vendors that contributed components: OS, web server, app server, DB, etc.:
    • If human intervention is required to patch, you’re likely going to forget or at least be late, which will create windows of opportunity for attackers: no good.
    • If only some components get automatically patched (say just the OS), it follows that others aren’t being patched (say the app server) and again you’ll be vulnerable.
    • If the runtime platform has been significantly customized (i.e., “hardened”) then automatic patching will likely break and you’ll achieve insecurity by trying to be too clever.

What if you’ve already deployed appliances that aren’t automatically patched?

  1. Try to patch them manually. Right now.
  2. Talk to the vendor. They are putting you at risk and had better step up and correct the error of their ways, or else you’ll be obliged to rip out their products.
  3. Look for alternatives, since these things are ticking time bombs on your infrastructure.

Roles are for entitlements and user classes are for people, and never the twain shall meet

Thursday, May 21st, 2015

I keep running into bad terminology when talking to new and prospective customers about organization structure and roles. People frequently confound t
wo quite unrelated concepts, calling both “roles.” This leads to confusion and much wasted effort trying to design unworkable systems, as I’ll explain below.

First, what are the concepts? In identity and access management (IAM) systems, we’re mainly interested in managing the lifecycles of people — identities, and their entitlements — typically login accounts and group memberships on end systems. Confusingly, some systems use the term ‘role’ to mean ‘security group, assignable to users within this system or application.’ I can’t make vendors like Oracle change their terminology, but I’ll take it as a given that anything assignable to a user within a single system or application is either a login account or group — even if that system thinks its called a role.

  1. Role. Roles are named collections of entitlements, that the IAM system can assign to users. They might be assigned automatically, because of some policy, or because of an (approved) request.
  2. User class. A user class is a set of users (i.e., identities or people). Users might be included in the user class individually, but the more common scenario is to collect users into a set by some rule — say based on their department, location, business unit, etc.
  3. Organizational hierarchy. This just means that every user should, ideally have (at least one) manager. We like to know the manager/subordinate relationship for all users because this relationship feeds into many useful processes: change authorization, access certification and more.

Nesting is implied in both user classes and roles. When roles are nested, it means that parent roles also include the entitlements of their child roles. This is represented by attaching one or more roles as entitlements in a parent role. There should be no technical restriction on how many roles a role may contain, or how deeply nested roles can be. In practice, most implementations use this sparingly, but in theory, at least, nesting can be both broad and deep.

As for user classes, we can think of “all people sharing a given manager” as a user class. This specific type of user class represents a hierarchy, so can be thought of as being nested. We could ask the system to show us a list of people who report both directly and indirectly to a given person — as one way to exploit this hierarchy. The organizational hierarchy is just a (possibly visual) representation of this nesting of the manager/subordinate user class.

So what’s the problem?

Many people use the term ‘role’ to mean two totally unrelated things:

  • The set of people who fit into a particular part of the organizational hierarchy — say all reporting to some manager, or all working in some department. THIS IS NOT A ROLE!
  • A set of entitlements (i.e., really a role this time).
  • Some combination of these two incompatible ideas — i.e., both a set of people and a set of entitlements, mixed up together.

I think people do this because they haven’t thought about clear definitions for roles or user classes. Quite often, they haven’t thought about user classes at all, but instead have only a very ambiguous idea of “some hierarchy of people and entitlements.”

People then make it worse by talking about ‘role nesting’ but actually meaning the nesting of these imaginary, hybrid role/user class things (that should not exist in any well designed IAM system). What does nesting mean when we’re talking about users and entitlements at the same time?

My advice is for everyone to just stop doing that. Roles are one kind of thing — collections of entitlements. User classes are another kind of thing — collections of people. Each of them can have its own hierarchy. You cannot include a role as a member in a user class. You cannot include a user class (or even a single user) as a member of a role.

By keeping the language clear, we can design much simpler, cleaner systems. For example, automatically assign some role to all members of some user class. That’s a nice way to automate access in many cases. Or ask the members of a user class to approve the manual assignment of a role to users. Or ask the members of a user class to recertify the list of people who were (manually) assigned a role. All these things are simple, clear and useful. What someone would do with weird, hybrid, role/user class things is beyond me.