Hacker News Reader

Comment by gary_0 | Hacker News Reader

> what's so wrong with X11 so people need to replace it

1. Security - Any program using X11 can read keystrokes, passwords, or the contents of any other window. Fixing this would break all existing X11 applications.

2. Performance - X11's client-server model doesn't work with hardware accelerated graphics, requiring hacks to get around. X11 is basically stuck with this legacy.

The ground-up re-design of X11 to fix those two issues is Wayland.

View on HN

I 100% agree with the performance improvement goals, but I think the security claims are overblown, and overly cautious. I honestly don't understand the point of trying to implement the security boundary in the display manager. It solves one class of security issues, while breaking a lot of accessibility and automation. The display manager just shouldn't be enforcing rigid per processes security controls, that's better done further down the stack. Or at a minimum security controls should respect user freedom enough to let a user access normally restricted features, with out the all or nothing elevation to root. There's a middle ground here where we don't break the world, and they get their shiny security policies.

> that's better done further down the stack

If you do it further down the stack, you break accessibility and automation even more... this has been tried. Doesn't work.

The end goal is to have actually working Android-like sandboxing rather than some broken firejail crap.

So we don't get the security benefits or accessibility. I'm not sure what is being solved. I'm all for a modern display system, I'm just not convinced the security claims are in anyway justified.

How is preventing apps from spying on each other through the display manager not justified? That's the lowest hanging fruit for desktop sandboxing.

I think The fact that people e.g. run the ydotool service as root is an example of this. It's like making a safe that is so hard to open that people just drill a hole in the bottom; you end up with something less secure than a safe that was easier to open.

Where in the stack should it be enforced that my cute desktop clock doesn't pull a Copilot and takes a screenshot of the entire desktop every 15 seconds to send to a remote service?

If your system is already running malware, why wouldn't the malware use a privilege escalation exploit (which are relatively numerous on linux) to access your data rather than some X11 flaw which depends on their code getting started by the user?

Because it's not an x11 "flaw" or exploit, it's just how X works. I also just don't buy the whole "well other stuff has exploits too" mentality.

I mean, yeah, it does, maybe. So why bother creating a password to a service if their database is probably running Linux anyway and the rdbms is probably compromised and yadda yadda yadda. It's the kind of argument you can make for anything.

Also no - privilege escalation is not "numerous" on Linux. It's very difficult to do in practice. It's only really a problem on systems built on old kernels which refuse to update. But those will always be insecure, just like running Windows 7 will be insecure.

A quick search for "linux local privilege escalation" in the CVE database (https://www.cve.org/CVERecord/SearchResults?query=linux+loca...) shows 25 results just for this year, so clearly these are very common.

So basically we have two issues here:

1. either focus on security even though these changes don't really improve the threat model

2. or allow disabled users and anyone who uses accessibility features to use GUIs

A security in depth approach obviously. Run less, use vetted sources, when running suspect software execute in a properly sandbox context. Seriously what's the point of securing screenshot and key loggers if a malicious process has full access to the users home directory, auido stack, webcam and network?

If you can't trust the process don't run it. If you have to run it, isolate all of it.

Wayland gives you neither the freedom to safely tailor your security policy, nor the security guarantees to warrant its inflexibility.

In Xorg and the WM, using the XACE extension.

https://www.x.org/archive/X11R7.5/doc/security/XACE-Spec.htm...

Neither Copilot or Recall do or did this.

The fact that desktop Linux is all or nothing in terms of privilege escalation is a design issue, however arguably Wayland gives us the tools to be more granular. Android has a permission system that makes sense can it's display manager is definitely closer in design to Wayland than it is to x11.

Android is the antithesis of an open computing platform and if anything the Linux desktop should use it as an example of what not to do.

Android's sandboxing is doing work for the benefit of the user. This is similar to how your web browser sandboxes JavaScript. Not every app needs access to my location and providing it access shouldn't require root. The Linux ecosystem understands this and it's why there is a large push for sandboxing models in software such as flatpak. Even if you disagree with Android at some level it hard to argue that users benefit from being able to control what the software they run is capable of doing. Otherwise we wouldn't have filesystem permissions to begin with, in the name of "freedom".

But them it's a question about how trustworthy an app is. Wouldn't it be better for software installed from your own distro repository to be fully trusted and require few or no security popups? After all, they are vetted to a much, much higher standard than any app store. Meanwhile flatpak apps and a random binary you've donwload get the full security isolation, because you can't trust third party devs.

That's not a scalable solution as not every piece of software can pay the packaging cost for every Linux distro. Maybe it's fine for core system software, but it's too difficult to expect that model to work for all software. Imagine if every website you interacted with needed to ship new website updates by packaging it and getting it vetted.

I think you still need a centralized distribution model even for things like flatpak to ensure some level of centralized auditing and revocation for software that has access to sensitive capabilities. However this doesn't necessarily need to be as large of a barrier for shipping updates as trying to package your software for a distro (and playing the game of trying to get your shared library versions aligned).

> 1. Security - Any program using X11 can read keystrokes, passwords, or the contents of any other window. Fixing this would break all existing X11 applications.

This is a feature not an issue. And it's how every computer that hasn't tried to take away control from its users has worked. I WANT my programs to be empowered to act on my behalf. If you want a gimped platform to run untrusted apps go buy a phone.

> 2. Performance - X11's client-server model doesn't work with hardware accelerated graphics, requiring hacks to get around. X11 is basically stuck with this legacy.

It works very well in fact. What you call hacks is something you probably cannot comprehend: Actual long term backwards compatibility. All programs should strive to have more of that, but especially central ones like the display server.

> The ground-up re-design of X11 to fix those two issues is Wayland.

It does limit what programs can do and breaks backwards compatibility, yes. That's the problem.

While Wayland has kind of been a disaster I disagree with your premise.

> 1. Security

Linux (and Unix before it) has always had security mechanisms built in. File permissions, setuid bits, namespaces. Any old program shouldn't be able to access /etc/passwd, and likewise the `<input type="password">` in my browser should be protected. Wayland's problem isn't that it tried to add security, but that the design and development process went horrifically wrong so that 17 years after the first release, people still have trouble with screen sharing.

> 2. Performance

The client-server model is obsolete and unsupported by modern applications (and nowadays there are easier ways to do remote GUI access) and keeping it was just a big pile of tech debt.

The problem wasn't that Wayland tried to fix things, it's that the process took 17 years and still isn't finished or particularly successful. My uneducated guess at why Wayland failed to succeed is that it went for extreme modularity and refused to say (back in like 2009) "Here's the security mechanism everyone has to use to take screenshots, etc. If this breaks your spacebar heater, sucks to be you." Rather than just define a single API where 99% of graphical applications and desktop utilities can do the sorts of things they already do on Windows and MacOS, and call it 1.0, instead they built a sprawling monument to bikeshedding and over-engineering.

> My uneducated guess at why Wayland failed to succeed is that it went for extreme modularity and refused to say (back in like 2009) "Here's the security mechanism everyone has to use to take screenshots, etc. If this breaks your spacebar heater, sucks to be you."

I tend to agree. I have long attributed the mess that is wayland-protocols in large part to the fact that they didn’t define a security mechanism or permission model in place from the start.

They seemed to assume, at first, that it was reasonable to prevent all programs from doing what any program could abuse. Had they instead acknowledged that some programs need to be granted the ability to take actions that otherwise risk insecurity, they wouldn’t have needed to try to distort the protocols to fit the lacking security model of Wayland (or, in some cases, wouldn’t have needed to circumvent Wayland entirely to achieve their ends).

I generously assume they simply considered a mechanism to grant permissions to be out of scope of the original spec, a merely horrific error and disastrous design flaw. If they were instead completely ignorant to the existence of screen recording, password managers, screen readers, and so on... inconceivable idiocy. Either way, as long as something like Wayland can happen, Windows has nothing to fear from Linux.

On the other hand, #1 makes it extremely difficult (if not impossible at all) to have a decent UI automation on Wayland. Sure, you can still do it if you're not leaving the terminal or a web browser, but anything else (including Electron apps) is a no-go. All the existing tools are written for X11.

The last time I looked into it, I found out I would have to deal with each compositor separately. On top of that, the target apps would have to be written with the new API in mind.

> 1. Security - Any program using X11 can read keystrokes ...

IIRC: this isn't quite correct, there was an extension called XACE that can block this. Probably Xorg didn't implement it and desktops didn't have support for it though.

https://www.x.org/archive/X11R7.5/doc/security/XACE-Spec.htm...

> X11's client-server model doesn't work with hardware accelerated graphics

I really wish they would figure this out. It just feels like with Threadripper etc the time is perfect for a return of thin-clients/not-so-dummy terminals running X11-like applications over the network on a server. Especially for development where many of us are running under-powered laptops and could use the boost to compilation from a beefy machine.

It's not really something to figure out, it's just X protocol design was made for when thin clients running over the network was assumed to be the future, it wasn't. Unfortunately though unlike Mac or Windows the migration to a better protocol has been ugly.

When you start to consider things such as HDR, hardware planes (important if you want energy efficient video decoding) etc, the protocol just doesn't make that kind of thing easy, compared to Wayland which does by it's use of surfaces etc.

There is nothing to figure out, unfortunately. The design of X11 precludes hardware acceleration; the hardware acceleration you see on a X11 desktop works by using extensions to entirely route around the X11 client-server model. To make it work they'd need to rethink the design. And they did -- that's how we got Wayland.

> The design of X11 precludes hardware acceleration

I wasn't talking about X11, just that capability.

Never heard of Waypipe before, but it was mentioned in a sibling comment, and that might be exactly what I'm looking for.

https://gitlab.freedesktop.org/mstoeckl/waypipe/

If you want thin client like behavior, you should look at stuff like Waypipe or just outright using RDP directly, both of which are much better at the job of "display remote graphical application on my computer" than X11's client-server design ever managed to accomplish.

If anything, the variety of alternative solutions for that today -- everything from single-app to high-res full-desktop game streaming -- are much more robust and viable on modern networks than the X approach ever was, even if it was a neat-o "freebie" thing that fell out of its design. You get what you pay for, I guess.

It has already been figured out long ago, which is how we have hardware acceleration on X today. Wayland fanboys calling it a "workaround" does not change that it works.

https://www.x.org/releases/X11R7.6/doc/xextproto/security.ht...

Is the security extension from 1996, which has a section on keyboard security

and its crazy to me that this anyone can claim X11 can't be off loaded, which its been doing for decades. From all the crazy blt/pattern HW acceleration to GL/vulcan implementations to the fact that the entire server can be on the other side of a network pipe, meaning it could be anywhere, including entirely encapsulated on a graphics card/smart nic/etc.

And if your talking about the xlib serialization, that was largely fixed with XCB.

Thanks. I see both are very valid reasons to drop X11.

Reply on news.ycombinator.com