One of the big topics of discourse in the Linux gaming sphere recently has been Tim Sweeney's statement on porting Fortnite to the Steam Deck, where Sweeney argues that Linux would be too difficult of a target and the market not big enough to warrant the amount of resources it would take to bring all of Fortnite on the platform.
The central crux of the issue, from Sweeney's point of view, is that making Easy Anti-Cheat, with all of its capabilities, run on Steam Deck (and thus on Linux) would be extremely difficult. He argues, that for a game of Fortnite's size this would open the flood-gates to significant influx of cheaters.
There have been some responses to this from the Linux side, with some accusing Sweeney of exaggerating the difficulty of such a port or that his statements are conflicting, because he simultaneously believes the Linux market is too small to be worthwhile but also would provide a way for too many cheaters. I will address some of these aspects a bit later, but for now let's focus on the main technical blocker, which is Easy Anti-Cheat.
Easy Anti-Cheat, or EAC, is an anti-cheat solution which apparently comes in a few configurations. We know that it can be run in a configuration where it is compatible with Linux/Proton apparently with just a relatively simple toggle. However, this mode of operation is seemingly a comparatively high-trust configuration, where only part of the anti-tampering protections of EAC are active. This may prevent some cheats but fail to detect others, which can be perfectly reasonable for games, where the number of cheaters and potential cheaters are fairly low or other systems complement the anti-cheat solution. There are plenty of games, even some popular free-to-play titles, which at best have this level of anti-tamper protection and they don't seem to have a major cheating epidemic, so clearly in many cases this should be enough. We also don't know the scope of cheats that are detected by EAC in this configuration, so this system by itself may already be fairly comprehensive.
EAC also contains a kernel-level component, which on Windows is installed as a kernel driver. This allows EAC code to run at a very privileged level and inspect essentially any and all parts of the system in order to detect tampering. This provides a very broad level of monitoring, which is also harder to bypass. Based on Sweeney's comments, this is the mode of operation used by Fortnite. It is also a mode of operation that is technically incompatible with the Linux way of doing things.
In Linux, the standard way of delivering drivers is by submitting the driver into the kernel source code tree, which naturally requires that the driver be open source. Most drivers are delivered this way, where the driver gets tightly integrated into the kernel and the drivers are updated when the kernel is updated. There are of course some notable exceptions to this rule, the largest of which is the Nvidia driver. The Nvidia driver is instead loaded as a separate kernel module, which allows Nvidia to keep its source code hidden, but also allows the driver to be updated separately from the kernel. So, EAC could surely use this approach as well, right?
The separate kernel module approach comes with some gotchas. Firstly, the kernel is licensed under GPLv2 and many of the parts in the kernel require the calling code to also be GPLv2 due to the "viral" quality of GPL. This means that, legally speaking, if Epic were to turn EAC into a kernel module and started poking around the kernel APIs, they'd have to open source EAC or they'd be in a legal grey area. The first approach is obviously not possible due to their business model and the second is at least not a great look.
Another problem with separate kernel modules is that the Linux kernel only guarantees a stable user-facing interface. This means that almost anything is allowed to change inside the kernel as long as user-level programs continue functioning. This is also the reason why sometimes the Nvidia driver stops working when you upgrade from one kernel to the next without installing an up-to-date Nvidia driver as well. So, when Sweeney is complaining about the multitude of kernel configurations, he's not wrong. EAC would need to maintain a compatibility shim similar to that of the Nvidia driver, which ensures that the EAC kernel module functions with each kernel version out there. Every time the kernel updates, an EAC engineer would need to go over the changes and update the compatibility shim every time there's a breaking change while still maintaining the compatibility with older kernel versions.
Theoretically you could overcome this problem somewhat by only targeting the Steam Deck and its SteamOS. This would give you a single kernel version to target, although Epic would need to negotiate with Valve in order to ensure their driver is somehow shipped with SteamOS.
But the problems don't end there. Since Linux is a fully open platform, there is technically nothing that would prevent a determined cheater from cracking open the Linux source code and making some tactical changes to how the kernel behaves, building the kernel and then making the EAC kernel module blind. On Windows the EAC developers can assume that the black box that is the NT kernel is at least somewhat difficult to modify by users. This means that in kernel-space they can assume some level of security through obscurity. On Linux this assumption does not hold. The only way for Epic to overcome that problem would be to negotiate with Valve to lock down the Steam Deck, which Valve has already decided not to do.
So, from EAC's point of view the Linux platform can never be quite fully trusted, which is entirely fair, because from the user's point of view EAC can never be quite fully trusted.
But surely Epic could still somehow bring Fortnite to the Steam Deck, right? Surely they could ship a version of Fortnite without the kernel-level component, right?
That they could, which brings us to the points about market share and the viability of cheating. Sweeney argues that the Linux market is too small, which initially sounds a bit odd because he then goes on to worry about the large numbers of cheaters. The kicker is here that the small Linux market doesn't necessarily guarantee a low number of cheaters. If it turns out that certain cheats are possible via a Linux version of Fortnite, this will attract some cheaters to use the platform in order to bypass EAC. It won't be all of the cheaters, many casual cheaters would likely not bother to learn Linux in order to cheat in a video game, but there is no doubt a group of cheaters that would take the opportunity. So, Fortnite would see some increase in cheating, but without good data it is hard to determine how big that effect would be. However, considering the popularity and free-to-play nature of Fortnite, it could very well be that it would be an attractive enough target for cheaters to attack even if there is a slightly higher initial investment. Cheat makers on the other hand would probably eventually find ways to package their offerings in an accessible enough format, like boot-to-cheat USBs or pre-configured VM images.
Some solutions to this problem have been proposed. For example, they could silo Steam Deck/Linux users in such a way that they will never come into contact with the rest of the playerbase. This would contain cheating, but it's also a hard-handed measure that would likely be unpopular. It would also require some amount of work to accomplish and I think it's fair for Epic to discount options that would cause extra work on them.
So, what's the solution to the problem? Here's the thing: I don't think there is one. My personal opinion is that client-side anti-cheat is fundamentally limited and taking it into the kernel is a bandaid that comes with excessive cost and is simply incompatible with the Linux platform. So, as long as Epic insists on maintaining its current anti-cheat approach with Fortnite, I just don't think there's going to be Fortnite on Linux.
And that doesn't mean Tim Sweeney is wrong or lying about the difficulties of adapting that approach to Linux. It just means that a new or different approach is needed in the future.
Apart from "power users" most people will just run whatever kernel Valve delivers.
For normal Linux support, i guess he has a point, though..
Dreadfully off topic but any chance you and Liam will be available for more episodes of the Gaming on Linux podcast?
QuoteSo, what's the solution to the problem? Here's the thing: I don't think there is one. My personal opinion is that client-side anti-cheat is fundamentally limited and taking it into the kernel is a bandaid that comes with excessive cost and is simply incompatible with the Linux platform
That's the point - client side anticheat is destined to fail if the client isn't under your control. And this even includes Windows. It's no secret that EAC isn't useful to prevent cheats, it's just making it hard enough so the cheat-devs can charge money for their working cheats, thus limiting the adoption.
But of course server side anticheat would require a certain amount of "i care about my userbase", not something Epic (or any of the big publishers for that matter) are known for ^^
So I'm not sure that a kernel level component is actually required. Mind you that measured boot does not imply that the platform is locked down. It only implies that user programs can check that the system was booted in a known good configuration. You are still free to modify it, however a cheat detection program such as EAC could then refuse to run.
Last edited by Kimyrielle on 9 February 2022 at 7:38 pm UTC
He's right about the fact that EAC would never stop cheaters running Linux, especially considering it can't even stop them on Windows.
Quoting: marcusI still think that if Valve provided a measured boot facility where a user-program could verify that it is running on an un-modified kernel then EAC could assume that there are also no kernel-level cheats present and that all kernel-level introspection APIs present correct and unmodified results.All fun and games until someone comes up with a way to spoof the measured boot report. If you add more layers, someone will eventually figure out how to hack the layer above.
Anyone who cares about security of their systems shouldn't come anywhere close to stuff like that.
Last edited by Shmerl on 9 February 2022 at 7:47 pm UTC
See more from me