6.10 Are there any security risks in Emacs?

• The movemail incident. (No, this is not a risk.)

  In his book The Cuckoo’s Egg, Cliff Stoll describes this in chapter 4. The site at LBL had installed the /etc/movemail program setuid root. (As of version 19, movemail is in your architecture-specific directory; type C-h v exec-directory RET to see what it is.) Since movemail had not been designed for this situation, a security hole was created and users could get root privileges.

  movemail has since been changed so that this security hole will not exist, even if it is installed setuid root. However, movemail no longer needs to be installed setuid root, which should eliminate this particular risk.

  We have heard unverified reports that the 1988 Internet worm took advantage of this configuration problem.

• The file-local-variable feature. (Yes, a risk, but easy to change.)

  There is an Emacs feature that allows the setting of local values for variables when editing a file by including specially formatted text near the end of the file. This feature also includes the ability to have arbitrary Emacs Lisp code evaluated when the file is visited. Obviously, there is a potential for Trojan horses to exploit this feature.

  As of Emacs 22, Emacs has a list of local variables that are known to be safe to set. If a file tries to set any variable outside this list, it asks the user to confirm whether the variables should be set. You can also tell Emacs whether to allow the evaluation of Emacs Lisp code found at the bottom of files by setting the variable enable-local-eval.

  See File Variables in The GNU Emacs Manual.

• Synthetic X events. (Yes, a risk; use ‘MIT-MAGIC-COOKIE-1’ or better.)

  Emacs accepts synthetic X events generated by the SendEvent request as though they were regular events. As a result, if you are using the trivial host-based authentication, other users who can open X connections to your X workstation can make your Emacs process do anything, including run other processes with your privileges.

  The only fix for this is to prevent other users from being able to open X connections. The standard way to prevent this is to use a real authentication mechanism, such as ‘MIT-MAGIC-COOKIE-1’. If using the xauth program has any effect, then you are probably using ‘MIT-MAGIC-COOKIE-1’. Your site may be using a superior authentication method; ask your system administrator.

  If real authentication is not a possibility, you may be satisfied by just allowing hosts access for brief intervals while you start your X programs, then removing the access. This reduces the risk somewhat by narrowing the time window when hostile users would have access, but does not eliminate the risk.

  On most computers running Unix and X, you enable and disable access using the xhost command. To allow all hosts access to your X server, use

  xhost +
  

  at the shell prompt, which (on an HP machine, at least) produces the following message:

  access control disabled, clients can connect from any host
  

  To deny all hosts access to your X server (except those explicitly allowed by name), use

  xhost -
  

  On the test HP computer, this command generated the following message:

  access control enabled, only authorized clients can connect