Re: automatic creation of a chroot environment
Rozhuk Ivan <[email protected]> Sun, 19 Jul 2026 23:01:01 +0300
| Newsgroups | gmane.os.freebsd.devel.hackers |
|---|---|
| Message-ID | <[email protected]> |
On Fri, 17 Jul 2026 08:10:45 -0700 Adrian Chadd <[email protected]> wrote: > I would really, REALLY like to see sandbox / capability extensions > that let me do things such as "no you can't access the whole > filesystem, only binaries/libraries and a sandbox directory, not my > whole home directory" (which this kind of does?) as well as "why no > you can't talk to the network directly" - without needing root to > create a jail first. > > Eg, there's no reason why a little AI agent, some CI test code or a > shell process which is going to call a bunch of compiler stuff should > have access to the network. > > If we can figure out a way for users to be able to do this - whether > it is changes to the kernel, or what apple did (a sandbox service that > user processes use to request sandboxes be created) or something else? > That'd be huge. > > At least we can hook some libc syscalls like socket()/connect() with some custom lib via LD_PRELOAD. Probably MAC framework also can help with this, at least some custom module. Some basic filtering is added with patch - PF+anchor filters traffic by UID. In terms of this "framework" it is a bit hacky: many services run under "nobody" user, and there is no good way to hook process start - it is out of scope for now. I think I should add some "start" command, but it will not integrated with rc.d scripts, but will nice to manual run and some other non rc.d usage.