EROFS Switches To Buffered I/O For File-Backed Mounts To Speed-Up Container Start Times

[phoronix]

Phoronix: EROFS Switches To Buffered I/O For File-Backed Mounts To Speed-Up Container Start Times

Back for the Linux 6.12 kernel EROFS introduced support for file-backed mounts to help with container and sandboxing use-cases. As part of the EROFS "fixes" merged yesterday to the Linux 6.13 kernel, file-backed mounts are now using buffered I/O by default to speed-up container start times...

https://www.phoronix.com/news/EROFS-Buffered-File-Mounts

[Jumbotron]

I have just enough knowledge on this to ask a question. Could a knock on effect of this be to speed up load times of Snap packages and perhaps Flatpaks as well ?

[intelfx]

I have just enough knowledge on this to ask a question. Could a knock on effect of this be to speed up load times of Snap packages and perhaps Flatpaks as well ?

Snap packages do not use EROFS in any way, so no. Flatpak packages work entirely differently (thankfully!), so no as well.

[Jumbotron]

Snap packages do not use EROFS in any way, so no. Flatpak packages work entirely differently (thankfully!), so no as well.

Thanks ! Saw the mention of “sandboxed” apps in the article and it piqued my interest.

[NobodyXu]

I guess this is mostly for Android?

I don't think docker/podman/kubernate or even firejail supports this as backing fs in container

i do know docker/podman/kubernate support btrfs to some exteny, and google's gVisor uses a userspace overlayfs impl.

[intelfx]

I guess this is mostly for Android?

I don't think docker/podman/kubernate or even firejail supports this as backing fs in container

I believe that at least Podman (and various other things in the RH OCI ecosystem, like bootc) can make use of EROFS via the composefs graph driver backend.

[NobodyXu]

I believe that at least Podman (and various other things in the RH OCI ecosystem, like bootc) can make use of EROFS via the composefs graph driver backend.

Thanks, it's quite interesting.

[Britoid]

I have just enough knowledge on this to ask a question. Could a knock on effect of this be to speed up load times of Snap packages and perhaps Flatpaks as well ?

The only performance overhead of Flatpak is the time at startup to setup the namespacing, there's no I/O overhead outside of the OS not having the libraries in io cache which might be the case for a non-Flatpak application if said library on the host was already accessed. Creating namespaces is extremely quick (systemd is already doing it for everything).

There is composefs that Flatpak I guess could swap to at some point as it can cryptographically verify applications at runtime, that would benefit from this change though.