sudo in Dockerfile? about code-server HOT 21 CLOSED

This is how I see we're teaching people bad practices. They don't know linux, and they're learning via guides. If they decide to use Linux outside of Coder they won't know the concept behind sudo and permissions. We need to let them know "hey, you might need to run this specific command first before doing that", and this adds a certain familiarity of how the concept of permissions work.

We in turn, teach them how security usually works in Linux for beginners, and we give a sense of familiarity for other users. You don't see GCP giving you away a VM with a root account because you're expected to know about sudo and the concept of Linux security by permissions.

from code-server.

I'm willing to reimplement the Dockerfile in the repo by basing off from my already existing image, should I get enough consensus that we should do usermode in this Dockerfile.

from code-server.

Docker best-practices (“USER” section) describes how to do this.

As a side effect, running code-server as root will result in that the newly created files become owned by root, so soon enough user will end up with a project that cannot be modified from their regular user. To solve the permission issues, it is recommended to create a regular user and remap host UID & GID

from code-server.

I think @sr229 summed it up pretty well. I personally don't see running every command as sudo in this environment as a bad thing, but it doesn't instill good security practices & habits when this is transferred to other places.

from code-server.

I agree with @sr229 the container should not be launching commands as sudo all the time.

from code-server.

@incizzle appreciate the input but could you elaborate on your reasoning why?

from code-server.

I think this ^^ is the best idea at the moment. May want to wait for a code owner to give their opinions.

from code-server.

The design issue here is, that you should not make any modifications to the docker container after it started up.
Being able to install some dev tools etc. in the VS Code terminal while using it is very bad practice, and as mentioned in #86, there should be a seperate container for each terminal, which has the project directory mounted into it, and also the container image of that individually spawned container should have the required build/dev tools for each language or project type.

That said, adding a USER statement to the dockerfile would break the possibility to use the terminal for anything like package installation.

from code-server.

@PhilsLab I don't think it should be as locked down as you need to on #86. However, I did provide a variant that does exactly that which works for OpenShift Online, but its pain to compensate for common use cases.

from code-server.

@PhilsLab, In order to have a separate container for each dev environment, it would require a lot of recoding of the actual VS code online program. It sounds likes its gonna be very hard to implement but it is a cool idea.

from code-server.

Theia did support this kind of work (Remember Eclipse Che 7?), however, this would require substantial backend re-engineering which would increase the amount of difficulty per new VSCode release since it might not be interopable with each other

from code-server.

You could just mount the projects and data-dir into your docker container on each start. Then you start where you were in a completely new environment.

from code-server.

Implementing it as per #183.

from code-server.

https://docs.docker.com/develop/develop-images/dockerfile_best-practices/#user

Avoid installing or using sudo as it has unpredictable TTY and signal-forwarding behavior that can cause problems. If you absolutely need functionality similar to sudo, such as initializing the daemon as root but running it as non-root), consider using “gosu”.

This is interesting, though I'm not sure if its applicable to us or whether the advice is about RUN commands or the entrypoint.

@frol

As a side effect, running code-server as root will result in that the newly created files become owned by root, so soon enough user will end up with a project that cannot be modified from their regular user. To solve the permission issues, it is recommended to create a regular user and remap host UID & GID

This is definitely a big issue. I think this alone is enough to justify the use of a user account. We definitely do not want all files written as root but instead whatever user the dev is using.

from code-server.

Btw guys, container per terminal is being tracked at #190

from code-server.

@nhooyr From what I have done with user-mode containers, this might increase the image size, but I might be wrong.

from code-server.

@nhooyr From what I have done with user-mode containers, this might increase the image size, but I might be wrong.

Thats fine. Though, as I mentioned earlier, this issue has not yet been discussed by the team and we're not sure of its direction. Please keep in mind your changes may not be accepted.

Also, since we're using ENTRYPOINT we can omit the CMD stage IMHO. We should be fixing this in the program-side.

Already fixed. See #188

from code-server.

Do not like to run with root even inside the container, I think it important to have a user (ex: coder)
WIP in PR #192 by @sr229 (good job)

from code-server.

As an FYI: from what I've seen CentOS / Fedora based systems require Docker to be run as root.

https://www.projectatomic.io/blog/2015/08/why-we-dont-let-non-root-users-run-docker-in-centos-fedora-or-rhel/ -- is a good read

from code-server.

That is interesting but I believe orthogonal to our discussion. We're discussing whether the processes inside the container should run as root.

from code-server.

My bad, misunderstood.

from code-server.