Saturday, February 6, 2016

Generic setuid/setgid wrapper for scripts

The setuid or setgid bit on a binary executable allows the binary to run with the effective uid or gid set to the owner of the executable, rather than the user that runs it. It is a rudimentary way for anyone to impersonate as the executable owner, so the owner can provide limited services using the owner's identity, including using files only accessible by the owner.

Historically, system binaries such as /sbin/ping are setuid root since ping requires privileged network access, but capabilities have reduced those needs. Binaries that manipulate privileged files still need to be setuid root, such as /bin/passwd. To setuid non-root user is less common but is still useful to control filesystem access.

Sometimes it is useful to write a shell script and making it setuid or setgid for the sake of controlling filesystem access. Many experts generally warn against setuid shell scripts. Here are the common security issues specific to scripts.
  • Anyone could create a symlink that resembles shell options and obtain an interactive shell with elevated privileges.
  • Anyone could create a symlink to a genuine setuid script, then after the OS executes the interpreter, perform a symlink switcheroo to a malicious script.
  • Command line interpretation can be changed by PATH or IFS.
Some people advocate (warning: bad advice) creating a binary specifically for executing the shell, but it can render other safety guards ineffective. The dynamic loader is more careful with the LD_PRELOAD and LD_LIBRARY_PATH directives with a setuid executable. But in this case, only the wrapper binary is setuid, but the shell run under is not, so it will pull in the whole slew of dynamic loading hooks unless we sanitize the environment.

One technique that an OS overcomes the switcheroo problem is to open the script file and pass /dev/fd/N to the interpreter, which ensures the same file is seen by the OS and the interpreter. This must be explicitly enabled on the OS and is still ignored by Linux. At least on Mac OS X, the interpreter's own setuid bit is also ignored.

When it comes to my own need, it basically boils down to:
  1. Don't bother with setuid/setgid, but use sudo which sanitizes environments. My use case is most similar to the line "ALL ALL = (script-owner) NOPASSWD: /path/to/script" in /etc/sudoers, to be run as "sudo -u script-owner /path/to/script". But sudoers manual begins with a "Quick guide to EBNF" which scares me.
  2. Rewrite my shell script as a full blown C program. I'm not feeling overly excited about that.
  3. Write a setuid/setgid wrapper binary in C, and the rest as shell script.
Needless to say, I took the last approach.

ps. In doing this, I found out about the shortest open source license called "Fair License" which is appropriate for such a short program.
Post a Comment