2.41.0 release candidates

[trapexit]

https://github.com/trapexit/mergerfs/releases/tag/2.41.0-rc4

Looking for help testing the upcoming 2.41.0 release. It's over a year worth of random changes and it'd be good to get more testing. Just use it as normal but feel free to try some of the new features. The new docs include the details for them. You can check the changelog for details. The final release will get a summarized changelog.

If you find issues feel free to discuss here or if you're on Discord @ https://discord.gg/erGkcStHkV

[morice117]

So first of all, thank you very much for the release. I am testing it right now on openSUSE Tumbleweed. I noticed that you added the flag FUSE_ALLOW_IDMAP with the passthrough merge (#1472) to enable IDMAP support.

I tried to create a bind mount with an idmap like this:

mount -o bind,X-mount.idmap="u:0:1000:1 g:0:1000:1" /mnt/mergerfs /mnt/temp

but I get the following error:

mount: /mnt/temp: mount_setattr() failed: cannot set ID-mapping: Invalid argument.
dmesg(1) may have more information after failed mount system call.

Should this work yet, or is further work required to enable the feature? I assumed it would, because the enablement patch for virtiofsd looks similar:

[morice117]

Thank you, it works perfectly!

[trapexit]

Apparently, when using allow idmap the kernel returns -1 for uid and gid if not using idmapping which the code was not expecting leading to silent failures. I'm playing with ideas on how best to handle this but thinking of just returning an error when requests come in with -1.

[morice117]

okay, i'm not sure what you mean with that, i got my original patch still running on my server with both idmap and without, didn't had problems this far.

[trapexit]

It's "working" but isn't correct. The kernel gives mergerfs the identity of the thread making the request. This is vital to how mergerfs works. Without knowing who is sending the request it is impossible to change credentials. Given the way the code was structured it was not expecting it to return -1 as the ID and it just silently fails.

[trapexit]

https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/tree/include/uapi/linux/fuse.h?h=v6.17.7#n1013

The description of how it behaves is here. Many more calls besides those in mergerfs require uid and gid to function properly.

While I could attempt to lookup the uid/gid of the application making the request it would be expensive to do. It feels like this feature is incomplete. If you look at other filesystems some of them use code that isn't in the kernel or libfuse.

[Jiinja]

'posix-acl=true' as an argument is not working for me in this version (rc3). When using 'mount' with mergerfs defined in fstab, it says the parameter is invalid. When running mergerfs directly it says unknown. I tried using '-' and "_".

I have reverted to the version you provided me with ioprio a few weeks ago and it is working with posix_acl=true. I looked through issues, docs, and some commits for this rc and I didn't see anything involving posix_acl except for a linux-util bug that has been solved in the version of linux-util that I have installed.

Separately from that issue, in the version you provided me with ioprio, you mentioned I could try out passthrough. passthrough.io=true per the docs was not recognized. Do you recall what setting I need for passthrough in that build or is it not configurable like ioprio?

[trapexit]

Where do you see passthrough.io=true? It should be passthrough.io=rw

https://trapexit.github.io/mergerfs/latest/config/passthrough/

The posix-acl thing is already fixed.

[Jiinja]

By fixed do you mean there's going to be another rc with it fixed?

Disregard the passthrough part of my post, I can use it when my posix_acl issue is resolved.

[trapexit]

I mean it's fixed in the master branch and therefore in the next release in whatever form that is.