You can restrict casual directory access if you’re careful, and well-informed, and diligent about the (often overly generous) permissions chosen by each Flatpak’s maintainers, and those chosen by runtimes and base applications, and about changes over time.
However, files and directories are only a small fraction of what hostile code can exploit. There are also d-bus services, abstract sockets, shared memory areas, display server protocols, and device nodes, for example. In the fingerprinting and tracking department, we have things like hardware peculiarities and installed font lists. If you’re concerned not only about privilege escalation and data exfiltration, but also about collection of your biometrics, then you should consider keystroke timing, mouse movement patterns, voice samples, and the like. These are just examples, not exhaustive lists.
More fundamentally, Flatpak sandboxes are made of Linux namespaces, which have a long history of vulnerabilities and exploits. They’re pretty good at containing buggy software, but software that wants to circumvent them can find a way, and if its developers are sufficiently skilled, will find a way.
The only thing I really use them for is to disable internet for open source applications that I trust aren’t malicious, but might still have telemetry in them.
However don’t let perfect be the enemy of good. Flatpak permissions are a step forward, at the very least defining a permission framework. Initially, permissions were defined but every app had all permissions, since app devs were lazy. Now, app devs are being better about defining what permissions are needed. In the future I imagine that Linux and Flatpak will lock down the security, making the permissions more concrete.
They’re useful like a lock on a door is useful. They protect from some circumstances, but not all, and won’t stop a skilled attacker.
If you’re determined to let hostile code execute on your hardware, you should get yourself some dedicated hardware for that purpose, isolate it from your network, and never use it for anything else. A game console, for example.
You can restrict casual directory access if you’re careful, and well-informed, and diligent about the (often overly generous) permissions chosen by each Flatpak’s maintainers, and those chosen by runtimes and base applications, and about changes over time.
However, files and directories are only a small fraction of what hostile code can exploit. There are also d-bus services, abstract sockets, shared memory areas, display server protocols, and device nodes, for example. In the fingerprinting and tracking department, we have things like hardware peculiarities and installed font lists. If you’re concerned not only about privilege escalation and data exfiltration, but also about collection of your biometrics, then you should consider keystroke timing, mouse movement patterns, voice samples, and the like. These are just examples, not exhaustive lists.
More fundamentally, Flatpak sandboxes are made of Linux namespaces, which have a long history of vulnerabilities and exploits. They’re pretty good at containing buggy software, but software that wants to circumvent them can find a way, and if its developers are sufficiently skilled, will find a way.
So Flatpak’s “security measures” are basically completely useless, is what I am to understand?
The only thing I really use them for is to disable internet for open source applications that I trust aren’t malicious, but might still have telemetry in them.
However don’t let perfect be the enemy of good. Flatpak permissions are a step forward, at the very least defining a permission framework. Initially, permissions were defined but every app had all permissions, since app devs were lazy. Now, app devs are being better about defining what permissions are needed. In the future I imagine that Linux and Flatpak will lock down the security, making the permissions more concrete.
It’s a step by step process.
Then I’ll keep installing software I need but don’t necessarily trust using Flatpak 👌
They’re useful like a lock on a door is useful. They protect from some circumstances, but not all, and won’t stop a skilled attacker.
If you’re determined to let hostile code execute on your hardware, you should get yourself some dedicated hardware for that purpose, isolate it from your network, and never use it for anything else. A game console, for example.
Would a container be enough? A VM?
From weakest to strongest protection:
VFIO gaming falls somewhere between 1 and 2.
You’ll have to decide for yourself how to balance risk, reward, and cost with your threat model.
Makes sense.
I’m running Heroic launcher anyway but thanks for all the info! You settled my curiosity. 😊