Podman Linkding Issues
1d 8h ago by sopuli.xyz/u/theorangeninja in selfhosted
Hello everyone,
I am currently trying to transition from docker-compose to podman-compose before trying out podman quadlets eventually. The first couple containers work great but today I tried Linkding and I run into a weird error.
Linkding can't access the data directory because the permission gets denied. After inspecting the container all the directories inside belong to root. But podman runs rootless so that must be the issues. I tried to change the owner of the data directory on the host to root but then the data directory in the container belongs to nobody and nogroup (?). After checking the environment variable documentation of Linkding it seems like there is no environment variable for a UID and GID.
I think I have a fundamental misunderstanding how rootfull and rootless containers work so I would be very grateful if anybody could point me in the right direction on where to get a solution for this problem or anybody had success running Linkding rootless.
Thanks a lot in advance!
Edit:
I used named volumes because that's what the dev used in the example compose file. Now I tried to use named volumes instead and now everything seems to work fine. No error in the logs and the web ui is accessible.
I am currently trying to transition from docker-compose to podman-compose before trying out podman quadlets eventually.
Just FYI and not related to your problem, you can run docker-compose with podman engine. You don't need docker engine installed for this. If podman-compose is set up properly, this is what it does for you anyway. If not, it falls back to an incomplete Python hack. Might as well cut out the middle-man.
systemctl --user enable --now podman
DOCKER_HOST=unix://${XDG_RUNTIME_DIR}/podman/podman.sock docker-compose up
I was unsure if I installed docker on this machine so I ran docker-compose and the help page showed up (another one than for podman-compose). Then I queried my installed packages and grepped them for docker and nothing shows up. Only podman-compose has docker in the description. So I accidentaly used that compatibility layer already without knowing.
But one reason I consider to switch is because compose files are not really standardized I heard and quadlets are structured like systemd files so I seems more applicable. But that is still a long way.
I usually set :Z at the end of volume mounts and it fixes the permission issues. Now that I think about it, all my Quadlets are using this option.
This solved it when using a bind mount! The other option is using named volumes which also works without errors.
On a different note, can I ask you where you learned about Quadlets. It seems the tutorials are still very sparse.
Glad to hear that!
A bit of Arch Wiki and Podman's own documentation.
Which system are you using? SELinux/AppArmot active? Can you share your compose? There are manyavariables at play here.
Other than that: Setting UID/GID via environmental variable is usually wrong, mostly from a design perspective of the container. There is a user directive during build as well as during deploy to use for that.
From a quick look at the docker file it does look like the user you use to run linkding needs to be in the root group.
BUT rootless podman maps the root user (usually to your user ID) to so the root user inside the container has not the same ID as the one outside. So I would suggest setting the permissions of the volume to your user for now.
Another way to figure out which user to use: just start a new/clean instance of the service and look at the new volumes.
I tried to use named volumes and now everything works fine, weird.
I think Mora is on the ball but we'd need their questions answered to know.
One possibility is that you have SELinux enabled. Check by sudo getenforce. The podman manpage explains a bit about labels and shares for mounts. Read up on :z and :Z and see if appending either to the volumes in your compose file unlocks it.
If running rootless, your host user also obviously needs be able to access it.
getenforce gives me Enforcing. And I think I have SELinux.
I had a look at this tutorial https://www.tutorialworks.com/podman-rootless-volumes/ suggested by another commenter and after running podman unshare ls -al in the folder with the bind mount it returns root root as the owner of the directory. So as far as I understand this means for the podman namespace this folder belongs to root? Like I said in my edit using named volumes solved the issue in on way.
I just tried the :Z label too and it seems to work too. So it was probably a SELinux issue?
With the getenfore command from kumi below I get Enforcing so I guess I use SELinux.
I now have two working options. a) Using named volumes (I'm still unsure if this is the way to go or not, generally speaking) and b) using the private label :Z for the bind mount.
Without :Z podman unshare yields root:root for the data directory. After setting the label it is a different user alltogether.
@theorangeninja Rootless podman container and owner of created files - always a mystery.
Maybe, the part belonging to "Using volumes" could help:
https://github.com/containers/podman/blob/main/docs/tutorials/rootless/_tutorial.md
If the container process is running with another UID than 0 (root), created files on the host belongs to another UID, calculated based on settings from /etc/suduid.
You should have a look into --userns for mapping of UIDs between container and host:
https://docs.podman.io/en/latest/markdown/podman-run.1.html
For PostgreSQL I'm using keep-id:uid=999,gid=999.
I went on and tried something else and when using named volumes (the dev uses bind mounts in his compose file) everything works without errors.
@theorangeninja Did you have tried a ls -al on the used volume?
The podman volume path can be found here:
`podman info --format '{{.Store.VolumePath}}'``
When you use $HOME/linkding as volume mount and the linkding container process is running with <> UID 0, then the created files are belonging to another UID than your UID.
Maybe this tutorial explains it better:
https://www.tutorialworks.com/podman-rootless-volumes/
I checked the tutorial and setting the private label :Z worked when using $HOME/... as bind mount. For named volumes from podman itself that was not necessary, it worked out of the box.
Like I said the dev used bind mounts so I sticked with that but he was probably using docker so he didn't have this problem.
@theorangeninja I would suggest again, that you try to access the content of the podman volume as host user, which is running the podman container.
I think, that it would looks like this and that you can only access it using podman unshare:
drwxr-xr-x 1 166446 166446 66 28. Jul 20:43 \_data
When I run the compose file with bind mount ./data and no :Z label I get these results:
ls -la yields the currently logged in user as owner of the data directory.
podman unshare ls -la yields root as owner of the data directory.
So do you think this is the basis of the problem?
question, why would you leave docker for podman?
Podman doesn't need a daemon running as root. If a container gets compromised and gets control of the container process, they won't have root access outside of the container.
makes sense but wasn't that already fixed with dockers rootless patches?
Yes, but it is not rootless by default. Most people don't even know this is something they could do.