SELinux/Tutorials/How SELinux controls file and directory accesses

From Gentoo Wiki
Jump to:navigation Jump to:search
This page contains changes which are not marked for translation.

How SELinux controls file and directory accesses

In the previous tutorial, we learned that SELinux adds in another method for finding out what the privileges would be for a process: a security context. This security context, together with the run-time user that the process is in, would define what the process is allowed to do.

When most users hear about controlling access on a Linux system, they immediately think about allowing file and directory access. It is tangible, a real-life scenario and simple to imagine or even reproduce. So let's talk about how SELinux would control file and directory access - we'll talk about the various other resources that SELinux can control later.

File access on Linux, without SELinux

Let's rewind a bit, and consider file access on a Linux system, but without any additional access control methods.

Access to files and directories is governed through the process' run-time user, the file/directory owner information and permission bits. With ls -l, or the more verbose stat, you can get information about the file ownership and permission bits. Let's consider the following output (ignore the Context bit from stat for now).

user $stat /etc/tor/torrc
  File: 'torrc'
  Size: 227             Blocks: 8          IO Block: 4096   regular file
Device: 903h/2307d      Inode: 261536      Links: 1
Access: (0644/-rw-r--r--)  Uid: (    0/    root)   Gid: (    0/    root)
Context: system_u:object_r:tor_etc_t
Access: 2012-11-02 21:13:54.100965797 +0100
Modify: 2012-11-02 21:13:54.100965797 +0100
Change: 2012-11-25 16:00:44.244998238 +0100
 Birth: -
user $ls -l /etc/tor/torrc
-rw-r--r--. 1 root root 227 Nov  2 21:13 torrc

What this information tells us, is that the torrc file is owned by the root Linux user, part of the root Linux group, and that both the owner (root), group (root) and other users can read it (the r bit is set). Only the owner (root Linux user) has write access to the file (the w bit is set). So access-wise, a process that runs as a non-root user will be able to read the file, but not write to it.

So when something (a process, called the actor) tries to write (which is a permission) to the file /etc/tor/torrc (which is called the resource), Linux checks the access controls and denies the request. The permissions that are in scope for the standard Linux access controls are the well-known read/write/execute rights, and they are based on the process ownership and file ownership.

This file access control is very standard on Linux, and should be well known by administrators and users. When looking at the file (or directory) ownership, it should be immediately obvious for users what can and cannot happen against the file. Consider the /var/cache/gorg directory:

user $ls -ld /var/cache/gorg
drwxrwx---.  6 gorg    gorg    4096 Feb 19  2012 gorg

In the above example, you should immediately notice that - unless a process is running as the gorg user, or within the gorg group - no process has access to this directory. Well, none is wrong: root-owned processes can still get into the directory and manipulate it (as root is the all-powerful user). Or the /var/lib/slocate directory:

user $ls -ld /var/lib/slocate
drwxr-x---. 2 root locate 4096 Nov  2  2010 slocate

Here, root-running processes can write into the directory, and processes that are running within the "locate" group can read/access the directory. Other processes will not be able to, as shown by the next attempt (by a regular user process) to enter this directory:

user $cd /var/lib/slocate
bash: cd: slocate: Permission denied

What users like you and me do in order to know what would or wouldn't be allowed, is to check the actor information (in this case, the run-time user, which is a regular user account), the resource information (in this case, the ownership details for the file) and then match the permission we want to check (like entering a directory) against the allowed permissions (execute in this case).

SELinux-ifying the examples

Before going towards SELinux itself, let's first write the above examples in a policy-like language.

Let's start with the /etc/tor/torrrc file, which had the following information (meta-data):

user $ls -l /etc/tor/torrc
-rw-r--r--. 1 root root 227 Nov  2 21:13 /etc/tor/torrc

We could write the access controls against this file as follows:

  • Allow any process that runs as the root Linux user to read the torrc file
  • Allow any process that runs as the root Linux user to write to the torrc file
  • Allow any process that runs within the root group to read the torrc file
  • Allow any process to read the torrc file

Or with the slocate directory:

user $ls -ld /var/lib/slocate
drwxr-x---. 2 root locate 4096 Nov  2  2010 slocate

This could be written as:

  • Allow any process that runs as the root Linux user to read the slocate directory information
  • Allow any process that runs as the root Linux user to write into the slocate directory (create files and such)
  • Allow any process that runs as the root Linux user to enter the slocate directory (and thus search for files within)
  • Allow any process that runs within the locate group to read the slocate directory information
  • Allow any process that runs within the locate group to enter the slocate directory

Although quite verbose, this does tell you exactly what the permissions are on the torrc file or slocate directory.

In reality, it isn't so much the "slocate directory" and "torrc file" that is part of the rule, but the ownership and permission bits. So to be correct, the slocate access controls would be more like the following:
  • Allow any process that runs as the root Linux user to read any directory owned by root and with the 'owner read' flag set
  • Allow any process that runs as the root Linux user to write into any directory owned by root and with the 'owner write' flag set
  • etc...

It isn't so much the content of the rules that we want to make clear, but the way in which the rules are written. They all contain information such as

  1. the context of the process that is acting upon something
  2. the context of the resource on which the process is acting
  3. the type or class of the resource
  4. the permission or permissions that are allowed given the first three information blurbs

In other words, you can write it as follows:

Allow some process with a certain context (like run-time user or group information) to 
do something (the permissions) against a file or directory (the class) that 
matches another certain context (like ownership permission bits).

In SELinux vocabulary, we say that

  1. the context of the process that is acting upon something is called the domain
  2. the context of the resource on which the process is acting is called the type
  3. the object class of the resource (e.g. file or socket) is called the class
  4. the permission or permissions that are allowed given the domain, type and class are the permissions

With this vocabulary in mind, an SELinux allow statement is structured as follows:

allow <domain> <type>:<class> { <permissions> };

SELinux type enforcement

SELinux has several language constructs for its various features, but for now we'll stick with the type enforcement part. In the previous section, we already discussed that SELinux uses a construction with the following syntax:

 allow <domain> <type>:<class> { <permissions> };

The next thing we need to do is to substitute the place-holders with real values.

SELinux domains and types

We already discussed domains in the previous tutorial somewhat, but we didn't call it domains then just yet. Remember the security context that is assigned to a process? We had an example such as system_u:system_r:auditd_t. Well, the third field of the context (auditd_t) is the domain in which a process is running.

root #ps -eZ | grep auditd
system_u:system_r:auditd_t      24008 ?        00:00:00 auditd

The suffix _t is a naming convention (but not mandatory for SELinux) to tell administrators that the given label is a type (or domain when it is for a running process). Because of this, we call the feature of SELinux that governs these 'allow rules' the type enforcement part of SELinux. It is also the SELinux feature that you will get to know the most, since it is the part of SELinux that most development goes into.

Domains are nothing more than regular types, but with the added knowledge that they are assigned to running processes.

An example type enforcement rule

Let's fill in a few of the place-holders with real-life values.

allow auditd_t <type>:file { write };

In the above example, we used the domain for the audit daemons (auditd_t), the class file and the permission write. All we have left is the <type> of the file we want to allow writing to. Now, the Linux audit daemon has to write into the audit logs, right? So it makes sense that the type we put there is the type for the audit log file.

Well, given that types are, naming-convention-wise, suffixed with _t, and that we are talking about the audit daemon log files, then a name such as auditd_log_t would not be wrong, would it? So then our rule becomes:

allow auditd_t auditd_log_t:file { write };

Well, this is exactly what SELinux does:

  1. if the process runs within the auditd_t domain (and thus has a security context with auditd_t in its third position)
  2. and if the target has the type auditd_log_t set,
  3. and the target is a file
  4. then the permission write is granted

And behold, we can ask SELinux if this rule is enabled on our system, using sesearch.

root #sesearch --allow --source auditd_t --target auditd_log_t --class file --perm write
Found 1 semantic av rules:
   allow auditd_t auditd_log_t : file { ioctl read write create getattr setattr lock append unlink link rename open } ; 

As you can see from the output, processes running in the auditd_t domain can do much more than just write to files with the auditd_log_t type assigned to them. They can control I/O, read, write, create such files, get attributes, set attributes, lock the file, append, unlink or link the file, rename the file and open the file.

How does SELinux know the labels of files?

From the first tutorial, we know that all processes have a security context. And from the previous sections, we know this security context contains a part called the domain that is used by SELinux for its type enforcement policy rules. We also saw an example policy rule for the audit daemon, writing to the audit log files:

allow auditd_t auditd_log_t:file { write };

But how does SELinux know that a file is assigned the auditd_log_t type? Well, SELinux stores this information in an extended attribute of the file. Extended attributes are parameters (key and value pairs) that are assigned to a file. On a regular Linux system, extended attributes are optional and most often not used. If you start using file-level access controls (facl's) then extended attributes are used to store the additional access control information. SELinux too stores its information in an extended attribute. These extended attributes are always for one file only: files cannot "share" extended attributes, but multiple files can of course have the same values for their extended attributes.

The extended attribute key that we need is the security.selinux extended attribute. The security. prefix tells Linux that this is a security-related attribute and as such should not be simply controllable by regular users: you need specific permissions to change these extended attributes (and SELinux too can be used to govern who or what is able to change the security.selinux attribute).

Let's look at this attribute for the /var/log/audit/audit.log file.

root #getfattr -m security.selinux -d /var/log/audit/audit.log
getfattr: Removing leading '/' from absolute path names
# file: var/log/audit/audit.log

Surprise surprise, the extended attribute contains a security context identifier (system_u:object_r:auditd_log_t) with a third field that matches exactly what we would expect (auditd_log_t).

Now, you don't need to use getfattr (which is short for 'get file attribute') to get the SELinux context of a file. The ls command can be used as well (with the -Z option), and we noticed earlier already that stat also displays the SELinux context.

root #ls -lZ /var/log/audit/audit.log
-rw-------. 1 root root system_u:object_r:auditd_log_t 1297813 Mar 12 20:56 /var/log/audit/audit.log

Files are not the only resource that use extended attributes. Everything that you can find on a (modern) file system can have extended attributes: directories, symbolic links, socket files, block devices, character devices, etc. SELinux thus employs extended attributes to the fullest by storing the SELinux context (security context) as extended attributes of all these resources.

What you need to remember

From this tutorial, you should remember that SELinux

  • uses contexts for processes (domains) and contexts for files (types) as part of its internal language for allowing access
  • uses the allow <domain> <type>:<class> { <permissions> }; syntax for this access
  • stores the security context (or SELinux context) of a file or directory as an extended attribute of this file