Subject: X Security - A Crash Course by runeb@stud.cs.uit.no Crash Course in X Windows Security 1. Motivation / introduction 2. How open X displays are found 3. The local-host problem 4. Snooping techniques - dumping windows 5. Snooping techniques - reading keyboard 6. Xterm - secure keyboard option 7. Trojan X programs [xlock and xdm] 8. X Security tools - xauth MIT-MAGIC-COOKIE 9. Concluding remarks 1. Motivation / introduction X windows pose a security risk. Through a network, anyone can connect to an open X display, read the keyboard, dump the screen and windows and start applications on the unprotected display. Even if this is a known fact throughout the computer security world, few attempts on informing the user community of the security risks involved have been made. This article deals with some of the aspects of X windows security. It is in no sense a complete guide to the subject, but rather an introduction to a not-so-known field of computer security. Knowledge of the basics of the X windows system is necessary, I haven't bothered including an introductory section to explain the fundamentals. I wrote some code during the research for this article, but none of it is included herein. If the lingual flow of English seem mayhap strange and erroneous from byte to byte, this is due to the fact that I'm Scandinavian. Bare with it. :) 2. How open X displays are found An open X display is in formal terms an X server that has its access control disabled. Disabling access control is normally done with the xhost command. $ xhost + allows connections from any host. A single host can be allowed connection with the command $ xhost + ZZZ.ZZZ.ZZZ.ZZZ where Z is the IP address or host-name. Access control can be enabled by issuing an $ xhost - command. In this case no host but the local-host can connect to the display. Period. It is as simple as that - if the display runs in 'xhost -' state, you are safe from programs that scans and attaches to unprotected X displays. You can check the access control of your display by simply typing xhost from a shell. Sadly enough, most sites run their X displays with access control disabled as default. They are therefore easy prey for the various scanner programs circulating on the net. Anyone with a bit of knowledge about Xlib and sockets programming can write an X scanner in a couple of hours. The task is normally accomplished by probing the port that is reserved for X windows, number 6000. If anything is alive at that port, the scanner calls XOpenDisplay("IP-ADDRESS:0.0") that will return a pointer to the display structure, if and only if the target display has its access control disabled. If access control is enabled, XOpenDisplay returns 0 and reports that the display could not be opened. E.g: Xlib: connection to "display:0.0" refused by server Xlib: Client is not authorized to connect to Server The probing of port 6000 is necessary because of the fact that calling XOpenDisplay() on a host that runs no X server will simply hang the calling process. So much for unix programming conventions. :) I wrote a program called xscan that could scan an entire subnet or scan the entries in /etc/hosts for open X displays. My remark about most sites running X displays with access control disabled, originates from running xscan towards several sites on the internet. 3. The localhost problem Running your display with access control enabled by using 'xhost -' will guard you from XOpenDisplay attempts through port number 6000. But there is one way an eavesdropper can bypass this protection. If he can log into your host, he can connect to the display of the localhost. The trick is fairly simple. By issuing these few lines, dumping the screen of the host 'target' is accomplished: $ rlogin target $ xwd -root -display localhost:0.0 > ~/snarfed.xwd $ exit $ xwud -in ~/snarfed.xwd And voila, we have a screendump of the root window of the X server target. Of course, an intruder must have an account on your system and be able to log into the host where the specific X server runs. On sites with a lot of X terminals, this means that no X display is safe from those with access. If you can run a process on a host, you can connect to (any of) its X displays. Every Xlib routine has the Display structure as it's first argument. By successfully opening a display, you can manipulate it with every Xlib call available. For an intruder, the most 'important' ways of manipulating is grabbing windows and keystrokes. 4. Snooping techniques - dumping windows The most natural way of snarfing a window from an X server is by using the X11R5 utility xwd or X Window System dumping utility. To get a grip of the program, here's a small excerpt from the man page DESCRIPTION Xwd is an X Window System window dumping utility. Xwd allows Xusers to store window images in a specially formatted dump file. This file can then be read by various other X utilities for redisplay, printing, editing, formatting, archiving, image processing, etc. The target window is selected by clicking the pointer in the desired window. The keyboard bell is rung once at the beginning of the dump and twice when the dump is completed. Shortly, xwd is a tool for dumping X windows into a format readable by another program, xwud. To keep the trend, here's an excerpt from the man page of xwud: DESCRIPTION Xwud is an X Window System image undumping utility. Xwud allows X users to display in a window an image saved in a specially formatted dump file, such as produced by xwd(1). I will not go in detail of how to use these programs, as they are both self-explanatory and easy to use. Both the entire root window, a specified window (by name) can be dumped, or a specified screen. As a 'security measure' xwd will beep the terminal it is dumping from, once when xwd is started, and once when it is finished (regardless of the xset b off command). But with the source code available, it is a matter of small modification to compile a version of xwd that doesn't beep or otherwise identifies itself - on the process list e.g. If we wanted to dump the root window or any other window from a host, we could simply pick a window from the process list, which often gives away the name of the window through the -name flag. As before mentioned, to dump the entire screen from a host: $ xwd -root localhost:0.0 > file the output can be directed to a file, and read with $ xwud -in file or just piped straight to the xwud command. Xterm windows are a different thing. You can not specify the name of an xterm and then dump it. They are somehow blocked towards the X_Getimage primitive used by xwd, so the following $ xwd -name xterm will result in an error. However, the entire root window (with Xterms and all) can still be dumped and watched by xwud. Some protection. 5. Snooping techniques - reading keyboard If you can connect to a display, you can also log and store every keystroke that passes through the X server. A program circulating the net, called xkey, does this trick. A kind of higher-level version of the infamous ttysnoop.c. I wrote my own, who could read the keystrokes of a specific window ID (not just every keystroke, as my version of xkey). The window ID's of a specific root-window, can be acquired with a call to XQueryTree(), that will return the XWindowAttributes of every window present. The window manager must be able to control every window-ID and what keys are pressed down at what time. By use of the window-manager functions of Xlib, KeyPress events can be captured, and KeySyms can be turned into characters by continuous calls to XLookupString. You can even send KeySym's to a Window. An intruder may therefore not only snoop on your activity, he can also send keyboard events to processes, like they were typed on the keyboard. Reading/writing keyboard events to an xterm window opens new horizons in process manipulation from remote. Luckily, xterm has good protection techniques for prohibiting access to the keyboard events. 6. Xterm - Secure keyboard option A lot of passwords is typed in an xterm window. It is therefore crucial that the user has full control over which processes can read and write to an xterm. The permission for the X server to send events to an Xterm window, is set at compile time. The default is false, meaning that all SendEvent requests from the X server to an xterm window is discarded. You can overwrite the compile-time setting with a standard resource definition in the .Xdefaults file: xterm*allowSendEvents True or by selecting Allow Sendevents on the Xterm Main Options menu. (Accessed by pressing CTRL and the left mouse button But this is _not_ recommended. Neither by me, nor the man page. ;) Read access is a different thing. Xterms mechanism for hindering other X clients to read the keyboard during entering of sensitive data, passwords etc. is by using the XGrabKeyboard() call. Only one process can grab the keyboard at any one time. To activate the Secure Keyboard option, choose the Main Options menu in your Xterm window (CTRL+Left mouse button) and select Secure Keyboard. If the colors of your xterm window inverts, the keyboard is now Grabbed, and no other X client can read the KeySyms. The versions of Xterm X11R5 without patch26 also contain a rather nasty and very well known security hole that enables any user to become root through clever use of symbolic links to the password file. The Xterm process need to be setuid for this hole to be exploitable. Refer to the Cert Advisory: CA-93:17.xterm.logging.vulnerability. 7. Trojan X clients - xlock and X based logins Can you think of a more suitable program for installing a password-grabbing trojan horse than xlock? I myself cannot. With a few lines added to the getPassword routine in xlock.c, the password of every user using the trojan version of xlock can be stashed away in a file for later use by an intruder. The changes are so minimal, only a couple of bytes will tell the real version from the trojan version. If a user has a writable homedir and a ./ in her PATH environment variable, she is vulnerable to this kind of attack. Getting the password is achieved by placing a trojan version of Xlock in the users homedir and waiting for an invocation. The functionality of the original Xlock is contained in the trojan version. The trojan version can even tidy up and destroy itself after one succesfull attempt, and the user will not know that his password has been captured. Xlock, like every password-prompting program, should be regarded with suspicion if it shows up in places it should not be, like in your own homedir. Spoofed X based logins however are a bit more tricky for the intruder to accomplish. He must simulate the login screen of the login program ran by XDM. The only way to ensure that you get the proper XDM login program (if you want to be really paranoid) is to restart the X-terminal, whatever key combination that will be for the terminal in question. 8. X Security tools - xauth MIT-MAGIC-COOKIE To avoid unathorized connections to your X display, the command xauth for encrypted X connections is widely used. When you login, xdm creates a file .Xauthority in your homedir. This file is binary, and readable only through the xauth command. If you issue the command $ xauth list you will get an output of: your.display.ip:0 MIT-MAGIC-COOKIE-1 73773549724b76682f726d42544a684a display name authorization type key The .Xauthority file sometimes contains information from older sessions, but this is not important, as a new key is created at every login session. To access a display with xauth active - you must have the current access key. If you want to open your display for connections from a particular user, you must inform him of your key. He must then issue the command $ xauth add your.display.ip:0 MIT-MAGIC-COOKIE-1 73773549724b7668etc. Now, only that user (including yourself) can connect to your display. Xauthority is simple and powerful, and eliminates many of the security problems with X.