From AppArmor
Jump to: navigation, search

Policy authoring introduction

In AppArmor terms policy is the sum of all profiles defined for the system. Creating good policy is hard. It involves a lot more than just creating a quick profile around an application. It requires knowledge of the system, the interaction of applications and the types of restrictions that the policy is designed to enforce. Even when just creating a single profile for an application, there needs to be some understanding of the resources the application is using to determine what is safe to give an application access to.

This is a general guide on policy authoring, providing tricks and pit falls for writing a good policy.

Protection goals

One of the decisions that needs to be made when authoring policy is what are the security goals. What needs to be protected, does data need to not leak out, or is it sufficient to prevent execution of malicious content. ??? something about users being the week point.

Policy Types

There are several general types of policies that require different levels of attention to detail.

  • Targeted - Targeted is the standard way in which AppArmor is deployed. It trades off the security of a totally system look down for easy of use and policy authoring, by recognizing that there are set of high value applications that can be more easily confined than the entire system while still providing value. In a targeted policy only a selection of the task on a system are confined in any meaningful way. Usually the targeted applications are network facing apps.

??? something about it being hard to confine the desktop ???

  • Total system - Total system confinement involves confining every application on the system.

??? what is missing from current apparmor to do this well ??? link to doc explaining how to load policy in the initrd, so that you can confine init confining all applications

  • RBAC - Role based access control can be done as a total system confinement or a per Role confinement but all tasks within a role must be completely confined.

AppArmor rules can make authoring profiles easy to understand but hide policy interactions (type vs file location). Good abstractions can be used to achieve much of this

Hrrmmm have some general rules and some policy specific guides

Exec rules

flexible making it easy to shoot self in foot

When to use ix, px, cx

Discussions on - creating profiles,

 when to use children, hats, ...

- creating abstractions

Creating Abstractions discussion - where to put this???

Abstracting profiles can make them easier to read, understand, and update.

Abstracting done through 2 main mechanisms - includes - variables

Includes provide - rule sharing - single editing point - method of expanding profile without editing file via directory includes

 - drop profile fragment in a directory

Variables - provide means of dynamically modifying what a rule means

- How to identify what can be shared/should be shared?

Includes for Abstractions - used when rule and permissions should be the same for all consumers - can split permission across several includes, and allow include to accumulate perms

 eg.  foo-ro  contains read only perms, while foo-rw includes foo and expands it to include write perms

Problem with Include abstractions is the assumption that profiles want same permissions for a given set of files - Get around this with variables

 - define common variables in global/common include
 - use the variable and specify permission in profile or include
 - can do a form of typing by defining a variable, and sharing it (every profile then gets to define access perms for the type)
    - eg.  @{PRIVATE_FILES} = @{HOME}/.ssh  ...
           then in profile
           @{PRIVATE_FILES}  r,
           @{PRIVATE_FILES}  rw,  ..
   - limit currently variables can contain are only expanded at the rule level
     so the @{TYPE1} variable couldn't contain partial rules for both network and file access.