Should we increase vm.max_map_count = 524240 → 1048576?
Your review all over the place is very much appreciated!
This is now in the testers repository.
But even if it doesn’t make sense for Chromium, would be useful to know generally for other applications from flatpak.
Hi folks, has anyone brainstormed this further or gotten flatpak working with hardened malloc in any capacity? I’m aiming to include this functionality in my security-focused image (github - /secureblue/secureblue) based on immutable fedora, and if someone’s already figured out how to do this I want to avoid reinventing the wheel. We have an ongoing thread there as well but I can’t link it here so I figured I’d ask.
No progress. And none should be expected. If there was progress, it would be posted here.
While experimenting on a Fedora Silverblue VM I’ve found:
You can override the /usr in the runtime with the hosts /usr with the following flag:
And then successfully preload the library:
flatpak run --usr-path=/usr --env=LD_PRELOAD=/usr/lib64/libhardened_malloc.so org.gnome.Calculator
Doesn’t this break sone flatpaks because they’re missing their specific /usr?
Over at secureblue we have since found a way to get this working, thanks to @34n0
Instead of using usr path, we have a short script that copies libhardened_malloc.so into ~/.local, and then override permissions for all flatpaks to grant them read only access specifically to that file only. They then use it when passed in the LD_PRELOAD environment variable, confirmed for several apps in /proc/$PID/maps.
A handful of apps still ignore it, notably steam, vscodium, and discord. But most apps I’ve tested so far don’t ignore it and do load it.
Interesting! Thanks for sharing!
(Upgraded your account just now. You can post links normally now.)
I think so yes. If the runtime’s /usr contains libraries which do not exist on the host. This method is flawed in other ways too… At least it proved a flatpak process can preload the library.
Thanks, here’s the PR to add this functionality:
If folks here have any additional input on the approach, it would be greatly appreciated.
Left some comments there.
Thanks, I’ve come up with a new solution I posted there. In short, we can directly reference
The only downside is that this requires
Which gives all flatpaks readonly access to all of /usr.
So now the question is whether this is a desirable tradeoff in favor of letting flatpaks use hardened_malloc.