1

We had good discussions about it being safe or not five years ago, and it was defaulted to yes on the kernel way over 3 years ago (with a note calling anyone security conscious who disable it as paranoid)

So, was it exploited after all in those 3 years? I couldn't find anything and this is not something I tracked during this time.

The 5+yr old issues were mostly kernel code checking privilege without the namespace checks (so root inside namespace could call uid 0 code on the host), which I hope they ironed out with some confidence before the default to Yes. Which then I would further assume it's mostly-harmless now unless you are injecting weird modules. And by mostly-harmless I mean no worse than any local exploit in the kernel would be. To validate all those assumptions I was trying to search for exploits using this but found none. Is this really or did I fail to find them?

Improve this question
2
  • 1
    Situation seems to have been normalized in the last year (see: this) but of course enabling unprivileged user ns may be a risk. If you don't need containers, you can disable it. But if you need containers, having rootless container managers is probably worth it.
    – Margaret Bloom
    Commented Oct 9 at 12:21
  • @MargaretBloom thanks for the link! But I will add everyone have a need for "containers" today, as it is the basis for any sensible linux hardening (systemd service isolation, bublewrap/firejail/apparmour/etc) I don't think we can keep the cat in the box anymore, even if its just for bind-mounts.
    – gcb
    Commented Oct 9 at 14:25

0

Reset to default

You must log in to answer this question.

Browse other questions tagged .