Run Docker container as unprivileged user, allow PUID/PGID selection

[lkubb]

Currently, the container process runs with root privileges. This cannot be changed, even by specifying PUID/PGID (as suggested in docker-compose.yml).

This PR migrates to running as an unprivileged user and makes it possible to specify PUID/PGID environment variables to choose UID/GID. Migration of existing data owned by root is handled transparently.

A side-effect is that the /app directory is read-only to the python process, I'm not sure if that is a problem since I'm not familiar with the architecture of changedetection.

I also took the liberty to clean up the Dockerfile (and apt cache) a bit, hope that's not a problem. :) Please feel free to suggest changes or deny the PR, if it does not fit your vision.

Fixes #565.

[dgtlmoon]

hmmm what is the problem you're trying to solve here?

[lkubb]

Assuming you're referring to the line below: If the container is e.g. run with docker-compose run changedetection -c, this would run python ./changedetection.py -d /datastore -c inside the container. It's mostly a convention, nothing that's necessary, just nice to have. If you prefer it removed, no worries. :)

[dgtlmoon]

see PR comments - mainly about being sure it's actually /datastore ?

[dgtlmoon]

why 911 ? :) Is it better to choose a UID > 1000 ?

[lkubb]

It's the standard one used by linuxserver.io and makes it a system user, but a higher default one like 3000 is fine as well imho. Would you prefer the default changed? It can be chosen freely anyways (docker-entrypoint.sh:L16-20).

[dgtlmoon]

master now tests the application inside the fully built docker container

also need to test this on an existing install which used root.root

[dgtlmoon]

unsure

[dgtlmoon]

@lkubb biggest one so far, gosu not found

[dgtlmoon]

hmm when it's running as test, it needs to use /datastore but only when run in github/automated, because often people test locally when developing too

[dgtlmoon]

changedetection.io/changedetectionio/tests/conftest.py

Line 33 in e99f07a

datastore_path = "./test-datastore"

somehow needs a var passed to pytest like pytest -d /datastore or.. unsure

[lkubb]

a) Introduce a new env var DATASTORE_TESTING, which defaults to test-datastore, but override it in the github workflow
b) Just mkdir the test directory in the docker build script. Not very clean but it shouldn't hurt.

[dgtlmoon]

b) Just mkdir the test directory in the docker build script. Not very clean but it shouldn't hurt.

This is fine if theres a comment above like
# creating test-dir for pytest to run in

[lkubb]

I added mkdir earlier, but you will need to approve the workflow.

[dgtlmoon]

needs /app/changedetectionio/.pytest_cache

but why not just chown all the /app to the same UID/GUID that we run the container as?

[lkubb]

Because usually, there is no good reason for an application to be able to modify itself. We could of course disable all that with a build arg specifically for the workflow, but I would be very hesitant to chown the app folder for production.

[dgtlmoon]

haha I love tests, what looked like a simple PR always shows up something

FAILED tests/test_backup.py::test_backup - PermissionError: [Errno 13] Permis...

so /datastore directory isnt writable (it creates the backup ZIP there)

[lkubb]

so /datastore directory isnt writable (it creates the backup ZIP there)

From what I can tell, it creates it as /app/changedetectionio/download.zip, right?

============================= test session starts ==============================
platform linux -- Python 3.10.9, pytest-6.2.5, py-1.11.0, pluggy-1.0.0
rootdir: /app/changedetectionio, configfile: pytest.ini
plugins: flask-1.2.0
collected 1 item

tests/test_backup.py::test_backup 
[...]
>       with open("download.zip", 'wb') as f:
E       PermissionError: [Errno 13] Permission denied: 'download.zip'

[dgtlmoon]

Oops right you are, that test needs fixing so its using the test-datastore directory

[dgtlmoon]

Or better is that it doesnt even write the file to disk and just stores it memory

[dgtlmoon]

to be honest the whole memory test thing is not needed.. you can remove it, or on another PR, it didnt really add anything interesting :)

[lkubb]

Alright, I chose to just remove writing the file since the rest of the code does not interfere.

[dgtlmoon]

agreed, no problems, some things seem like a nice idea - but just get in the way

[dgtlmoon]

sometimes there are "random" fails, i'll keep an eye on it

[lkubb]

sometimes there are "random" fails, i'll keep an eye on it

👍 It's still complaining that /app/changedetectionio/.pytest_cache is not writable, might need to account for that somehow, maybe create the directory in the entrypoint if it detects we're running pytest.

[dgtlmoon]

hmm https://stackoverflow.com/questions/33310615/is-there-a-way-to-change-the-location-of-pytests-cache-directory could help

[lkubb]

The SMTP tests were failing because aiosmtp could not be installed since the changedetection user does not have a home directory. I added one, the SMTP container should start fine now.

[lkubb]

Alright, all tests are passing now. For most setups, the entrypoint script should automatically chown the datastore to the correct user. The one exception is in case it is a network share mount, where it might fail. In this case, admins must ensure the correct PUID/PGID are set in the container so the script does not attempt to chown any files.

In case this logic causes any problems, it can be skipped completely by setting the KEEP_PRIVILEGES environment variable to a non-empty value. This results in the previous behavior.

I'd really like to wrap this PR up soon since it has been going for nearly two and a half years. :)

[dgtlmoon]

amazing thanks, will have a play with it - yeah its one of those "might break something at some time for somewhere" commits but it NEEDS to be in the codebase

[dgtlmoon]

can you think of some test cases to try? like i already have a datastore owned by some user and...

[lkubb]

Unless otherwise noted, everything assumes 0755/0644 dir/file permissions (or 0 for others, just not 7/6).

First, I would verify everything works as before when KEEP_PRIVILEGES is set to provide an immediate bandaid to any problems.

Then, with a root-owned (populated) local datastore:

1. Starting the container changes the ownership to 911:911 and everything works fine.
2. Starting the container with PUID=1000 changes the ownership to 1000:911
3. Starting the container with PGID=1000 changes the ownership to 911:1000
4. Starting the container with PGID=1000 PGID=1000 changes the ownership to 1000:1000

Then chown -R the datastore to 1234:5678 and do 1 again.
Then chown -R the datastore to 911:0, start the container and verify nothing was changed.
Then chown -R the datastore to 0:911 and do 1 again.
Then chown -R the datastore to 0:911, chmod -R g+w, start the container and verify nothing was changed.

You should check that running the container rootless (still?) works with 1 (/4) (so root here is actually the local user account, PUID/PGID are relative to the user namespace).

I'd probably check that correctly setup NFS/SAMBA mounts work.

Tbh I'm not familiar with the application architecture, so I'm not sure if there are any other paths that should be tested.

Edit: I misremembered what the entrypoint script does. It chowns the user only, not the group. This could be a problem in case someone has mounted a network share intending to use group write permissions (the amount of setups like this is probably very low).

I'm currently considering if it would be a better idea to just check if a specific uid:gid has write permissions for the data dir and if not, chown -R it once.

Edit2: The entrypoint now only checks if the datastore is writable and tries to chown -R it if it's not.

[dgtlmoon]

I'm currently considering if it would be a better idea to just check if a specific uid:gid has write permissions for the data dir and if not, chown -R it once.

better than doing it every time, some datasets are pretty big

[lkubb]

Alright, the entrypoint now checks writability of the datastore by the designated PUID:PGID combination and only intervenes if that fails. I'll adjust the proposed test cases above.

[dgtlmoon]

shouldnt this stay as if ?

[lkubb]

No, it would then not respect the DATASTORE_PATH defined by the user (or the default one set during container build time).

[leleogere]

Thanks for this work! It seems that all tests pass, there are no conflicts with the base branch, and still this PR hasn't been updated for the past 5 months. Is there anything preventing it to be merged?

[kimpenhaus]

@dgtlmoon is there any chance to get this pr merged? it looks like it is ready to go - and while the current container is running as root and writing files as root I am not able to backup the files running borg backup as unprivileged user:

/srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/29a24202d0194baec6c01affcfd8c350.txt.br: open: [Errno 13] Permission denied: '29a24202d0194baec6c01affcfd8c350.txt.br'
E /srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/29a24202d0194baec6c01affcfd8c350.txt.br
/srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/90f37966f5feb05c818dac3873001005.txt.br: open: [Errno 13] Permission denied: '90f37966f5feb05c818dac3873001005.txt.br'
E /srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/90f37966f5feb05c818dac3873001005.txt.br
/srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/a2c1f9154902cd93b7d9cbdc1857bd91.txt.br: open: [Errno 13] Permission denied: 'a2c1f9154902cd93b7d9cbdc1857bd91.txt.br'
E /srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/a2c1f9154902cd93b7d9cbdc1857bd91.txt.br
/srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/6356681e7375d1b9a075783be56afdf4.txt.br: open: [Errno 13] Permission denied: '6356681e7375d1b9a075783be56afdf4.txt.br'
E /srv/nfs/docker/changedetection/16ddaa48-20e3-4e01-afd6-7cbace31a96e/6356681e7375d1b9a075783be56afdf4.txt.br

I have high hopes that this PR could solve my issue.

thanks

[dgtlmoon]

Thanks all, its a been a while, so I re-rolled it into #3900 , please, I would like your feedback :)