Use *at functions for directory creation

[Vogtinator]

Perform path traversal manually and do not follow symlinks.

Draft because some TODOs still open. Also absolutely untested.

[jsegitz]

Thanks for the changes. Since the path sources come from trusted system data that seems good enough. I added two suggestions

[Vogtinator]

I wonder whether there is a bug lingering here with umask and the mode parameter.

[jsegitz]

I don't see a problem with this. Why is this concerning to you?

[Vogtinator]

If the desired mode of the directory has bits set that are umasked away, they'd be missing.

[jsegitz]

since this is running in a special context I assumed it has no umask set. But yes, if there's on this will be a problem. So setting the umask explicitly is probably a good idea

[Vogtinator]

Should be functionally complete now, but still untested. I converted some leftover uses of node_basename to use the directory's FD instead as well.

I don't like the goto driven cleanup, I'll try converting that to __attribute__((cleanup(...)).

[jsegitz]

Should be functionally complete now, but still untested. I converted some leftover uses of node_basename to use the directory's FD instead as well.

I agree. Do we have tests for this in openQA that we can use to test it?

I don't like the goto driven cleanup, I'll try converting that to __attribute__((cleanup(...)).

Agreed. But from my POV that's not in scope for this, but feel free to add it