Slashdot Mirror
Moblin Will Run X Server As Logged-In User, Not Root
nerdyH writes "An architect of the Moblin Project has announced that Moblin 2.0 for netbooks and nettops is the first Linux distribution to run the X server as the logged-in user, rather than SUID'd to root. The fix to this decades-old security liability comes thanks to 'NRX' (No-root X) technology reportedly developed by Intel, Red Hat, and others in the X community, and the Moblin-sponsored 'Secure X' project. Besides making Linux netbooks a lot more snoop-proof, it seems like this could lead to an X-hosting renaissance of sorts, since you wouldn't be risking the whole system just to open up a specific user's account to remote X servers."
Linux's SUID X server problem has been kind of a "dirty little secret" for many years. Most modern distributions include a few crude workarounds, such as dimming the display and then freezing X whenever the user is asked to type in a root password. Getting rid of the SUID bit altogether ought to make netbooks powered by Moblin technology much more difficult to snoop on over the network.
This does not make sense. Graphical sudo wrappers have nothing to do with X being suid, and neither does anything to do with network traffic.
It seems likely that with NRX technology, you could run X apps over a network with much less risk to the app server (the system that runs the "X client" component, in the backwards terminology of X).
This is actually backwards, the only place there's less risk is for the system that the X server is running on.
I'm not sure I grasp the concept of X Hosting, and how this non-SUID server would help that.
X is not required to be running on the remote system for X11 forwarding over SSH. Even running an Xvnc server doesn't require it to be SUID. This seems to be entirely a local security gain for users who will be interacting with local graphics hardware.
But running apps remotely and having them display on a local X server _NEVER_ required root access of any kind on the remote server....
I don't know how they've done it, but I know this is a good thing.
They've done it by removing the responsibility of X talking directly to the graphics hardware by implementing Kernel Mode Switching for graphics drivers (among other much needed overhauls to the Linux graphics stack). Thus X can now access what it needs at the logged-in users' level and doesn't need root.
The article repeats the common misunderstanding: "in the backwards terminology of X"
What exactly is backwards about this? X is the server, and the apps are clients.
Think about it: The client initiates the conversation with the server. The client tells the server what to do.
How is this backwards?
Just got fixed by this. To be honest, I don't know how they've done it, but I know this is a good thing. This will make X and linux more secure and I can only applaud that.
I think what is basically boils down to, is that instead of X talking to the hardware directly it now talks to a file under /dev/ just like everything else.
It doesn't?
> it seems like this could lead to an X-hosting renaissance of sorts,
> since you wouldn't be risking the whole system just to open up a
> specific user's account to remote X servers.
What a clueless statement. Somebody doesn't understand how X works. The server part that runs SUID root has never ran on the app server.
What this does do is stop a random remote app getting to root on your workstation but any local exploit of the X server gets them your user account and that can cause a lot of mischief and only needs a different local root exploit to get the rest of the way to 0wn1ng your local desktop machine.
Democrat delenda est
I am a bit confused by the submitter's comment about remote X servers. I understand the appeal of remote X clients: I can, e.g., log into a big fast machine and run MATLAB (the X client) there while interacting with the window on my less-powerful laptop (which runs the X server). But what's the point of a remote X server? Why would anyone want to run an X server (software sense of 'server') on a server (hardware sense of 'server')?
> Can someone spare me reading the article and let me know if DRI is still possible without root?
Yup, this whole thing rests on the new kernel modesetting. That was the last thing that required root to be able to directly frob bits on the video card. DRI also goes into the kernel as it should. The kernel is supposed to own all of the hardware and expose safe APIs for user apps to access it. For historical reasons video has been the exception to that rule. No longer.
Democrat delenda est
Er, the same way USB was for years? Actually, DRI, too. The driver exposes a pseudo-device in /dev/, which actually is a socket-like, high-throughput mmap wrapper and the X server opens it. Given appropriate file permissions and group membership, this can be done from a user account.
This is Slashdot. Common sense is futile. You will be modded down.
If graphics drivers were implemented in the kernel instead of the X server, this problem wouldn't have existed in the first place.
1. Does this mean you can't login at a graphical interface? I.e. will you have to login at a terminal and then wait for X server to come up?
No. There should be a login X server (running as root or nobody or whatever) to display GDM, then during login this server will exit and launch a new server under your uid. Or something like that.
2. If multiple users login, will each user get their own instance of X server? This seems like overkill...
I think fast user switching already works that way. We don't consider it overkill that each user gets their own instance of Firefox; why is X any different?
I am not sure that this is the right solution. Not running it as root is good, but running it as me - I don't know. I'd rather that the user that runneth the X server is some sort of 'xserver' user - to whose process I connect. That 'xserver' user then has the right to push my screen into VGA mode and all that. Also, this doesn't fix all those other services (that gnome has, for example) that allow my X programs to mount stuff etc. Which is, again, a security risk by itself.
Religion is what happens when nature strikes and groupthink goes wrong.
How big of a security problem was this? I haven't seen a linux system with open ports for X in 10 years. Anyone who wants to use remote X just uses ssh -X, it's easier to set up than xhost anyway.
Give me Classic Slashdot or give me death!
Sounds like Windows NT 3.5, wonder if it will get moved back into kernel space for performance reasons just like NT4 moved video back into kernel space.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Yes, DRI access is done through /dev/dri* and works correctly.
I just loaded it on my Eee PC and it turns the machine into a kiosk. Very unappealing for anyone who actually wants to use their netbook. Its very flashy and friendly if all you do is check your email and browse the web though.
Mr. Green
Well, if the flaw is in an X *library*, it doesn't matter, as only the clients (running as the regular user) use those. The X server doesn't need or use libX11, libXrandr, libXext, etc. at all.
But yes, true -- any exploitable flaw in the X server itself (or any of the extensions compiled into or loaded dynamically by the server) could result in root access.
Xfce: Lighter than some, heavier than others. Just right.
> Sounds like Windows NT 3.5, wonder if it will get moved back into kernel
> space for performance reasons just like NT4 moved video back into kernel space.
Not the same thing. The video hardware belongs in the kernel in exactly the same way as sound, mass storage and the keyboard/mouse do. *NIX and Windows are now alike in that and it is good.
What Windows did was bring most of the next layer up the chain into kernel space. This would be more like putting the whole X server and bits of GTK and/or Qt into the kernel, not just running it as root. Yes it improved performance some, but the security implications are horrific.
Democrat delenda est
Not exactly. There is an average of one bug per 1,000 lines of new code. X.org has been in constant development since 1984. A lot of those 2,000 will have already been fixed. Note that X.org is part of the OpenBSD base system and so undergoes the same kind of rigourous code review. X.org, XFree86, and then X.org is probably the most reviewed and tested piece of software in widespread use.
I am TheRaven on Soylent News
No, it doesn't. It runs most everything as the "Administrator" user, which is a lot like a root account, but without even the level of security that logging into Linux/Unix as root provides ;-)
Guns don't kill people; Physics kills people! - John Lithgow as Dick Solomon on Third Rock From The Sun
Certainly this isn't worse than the current situation. But if my data is still at risk I have a hard time caring much about any "security" advantage.
Without my data the machine is worse than worthless.
Platform advocacy is like choosing a favorite severely developmentally disabled child.
Since there was never any reason for the X server and the clients to need to use the same uid, why move the X server from root to the logged in user? It could as well be moved from root to a uid dedicated to the X server. Then you would get another level of separation, at essentially no price. (There is of course a caveat in case you have multiple X servers running at the same time, but that could be solved by allocating a uid per X server).
Does graphics mode switching inside the kernel mean that we can soon expect switching between VTs to work even if the X server is locked up? Or is the keyboard handling still going to prevent that? (Doing the switching from a remote login would work around the keyboard issue).
Do you care about the security of your wireless mouse?
I was a video driver developer for Sun for many years. The window system *always* ran as the logged-in user. When I started developing for Linux, I was appalled when I realized that Linux ran the windows server as root.
Here's how we did it at Sun: For every supported video card, there is a device driver. The driver provides basic services such as cursor and color-table management (there are advantages to doing this in the kernel), and additionally allows the user logged in at the console to map in the device registers. This means that the window system doesn't need any special privileges to run.
There are other advantages to having a device driver manage user-level hardware mapping. Not the least of which is that it allowed us to implement full-bore context switching at the device level. The advantages of this are enormous.