Mdthbs

I am completely new to *NIX based OSes. One of the things that baffles me is that a process or program may execute setuid(0) and then perform privileged operations and revert back to its normal uid.

My question is what is the mechanism in *NIX to prevent any arbitrary process from possessing root ?

If I write a simple C program that calls setuid(0) under what conditions will that call succeed and under what conditions will it fail ?

The basic idea is that a process may only reduce its privileges. A process may not gain any privileges. There is one exception: a process that executes a program from a file that has a setuid or setgid flag set gains the privileges expressed by this flag.

Note how this mechanism does not allow a program to run arbitrary code with elevated privileges. The only code that can be run with elevated privileges is setuid/setgid executables.

The root user, i.e. the user with id 0, is more privileged than anything else. A process with user 0 is allowed to do anything. (Group 0 is not special.)

Most processes keep running with the same privileges. Programs that log a user in or start a daemon start as root, then drop all privileges and execute the desired program as the user (e.g. the user's login shell or session manager, or the daemon). Setuid (or setgid) programs can operate as the target user and group, but many switch between the caller's privileges and their own additional privileges depending on what they're doing, using the mechanisms I am going to describe now.

Every process has three user IDs: the real user ID (RUID), the effective user ID (EUID), and the saved user ID (SUID). The idea is that a process can temporarily gain privileges, then abandon them when it doesn't need them anymore, and gain them back when it needs them again. There's a similar mechanism for groups, with a real group ID (RGID), an effective group ID (EGID), a saved group ID (SGID) and supplementary groups. The way they work is:

• Most programs keep the same real UID and GID throughout. The main exception is login programs (and daemon launchers), which switch their RUID and RGID from root to the target user and group.


• File access, and operations that require root privileges, look at the effective UID and GID. Privileged programs often switch their effective IDs depending on whether they're executing a privileged operation.


• The saved IDs allow switching the effective IDs back and forth. A program may switch its effective ID between the saved ID and the real ID.

A program that needs to perform certain actions with root privileges normally runs with its EUID set to the RUID, but calls seteuid to set its EUID to 0 before running the action that requires privileges and calls seteuid again to the EUID change back to the RUID afterwards. In order to perform the call to seteuid(0) even though the EUID at the time is not 0, the SUID must be 0.

The same mechanism can be used to gain group privileges. A typical example is a game that saves high scores of local users. The game executable is setgid games. When the game starts, its EGID is set to games, but it changes back to the RGID so as not to risk performing any action that the user isn't normally allowed to do. When the game is about to save a high score, it changes its EGID temporarily to games. This way:

• Because the high score file requires privileges that ordinary users don't have, the only way to add an entry to the high score file is to play the game.


• If there's a security vulnerability in the game, the worst that it can do is grant a user permission to the games group, allowing them to cheat on high scores.


• If there's a bug in the game that doesn't result in the program calling the setegid function, e.g. a bug that only causes the game to write to an unintended file, then that bug doesn't allow cheating on high scores, because the game doesn't have the permission to write to the high score file without calling setegid.

What I wrote above is describes a basic traditional Unix system. Some modern systems have other features that complement the traditional Unix privilege model. These features come in addition to the basic user/group effective/real system and sometimes interact with it. I won't go into any detail about these additional features, but I'll just mention three features of the Linux security model.

• The permission to perform many actions is granted via a capability rather than to user ID 0. For example, changing user IDs requires the capability CAP_SETUID, rather than having user ID 0. Programs running as user ID 0 receive all capabilities unless they go out of their way, and programs running with CAP_SETUID can acquire root privileges, so in practice running as root and having CAP_SETUID are equivalent.


• Linux has several security frameworks that can restrict what a process can do, even if that process is running as user ID 0. With some security frameworks, unlike with the traditional Unix model and capabilities, a process may gain privileges upon execve due to the security framework's configuration rather than due to flags in the executable file's metadata.


• Linux has user namespaces. A process running as root in a namespace only has privileges inside that namespace.

In addition to the user id, a process is also associated with an effective user id. When a setuid root program like su is executed, the effective uid is set to 0, but the real uid stays the same. The effective uid is, as the name says, the uid that is used for permission checks, the real uid just tells "who we really are". A process with an effective uid equal to zero can successfully call setuid(0) to change the real uid to zero.

In order for setuid(some_uid) to work, the following conditions need to be met:

1. The executable's file needs to be owned by some_uid¹.


2. The executable's file needs to have the setuid permission.

The setuid permission can be granted or removed with chmod:

chmod u+s execuables_file_name

chmod u-s execuables_file_name

The setuid permission can be seen when displaying the file permissions as the execute permission will be replaced by an s if the setuid permission is granted.

> ll execuables_file_name

-rwsrwxr-x 1 root root 0 Sep 30 17:23 execuables_file_name* 

If for some reason the owner user doesn't have execute permissions, then it will be displayed as a capital S instead:

> ll execuables_file_name

-rwSrwxr-x 1 root root 0 Sep 30 17:23 execuables_file_name* 

Note that scripts are an entirely different beast.

A thorough guide to permissions.

^{1. setuid(some_uid) will also succeed if transitioning to the user id that started the process}