Mult looks interesting, but there's not a lot of material I could find online, other than on archive.org.
It's also rather more radical than what I was thinking of - which is something more akin to how FreeBSD jail works - I was thinking of starting by adding a syscall (which must be called by uid 0) that creates a new "jail ID" for the calling process, after checking it's not already in a jail (i.e, it's in jail ID 0). Then any other syscalls that interact with other processes check if the jail ID matches - if not, it acts as if the target PID does not exist. With this a simple userland utility could do a chroot and then call the new SYS_JAIL syscall and have process table isolation. Once that's working, it should be possible to integrate a facility for isolation of network interfaces and so on - i'd also want to isolate devices too (for obvious reasons - imagine if a jailed process can just access the block device for the root filesystem). But that's my basic idea - instead of a whole new process table, just adding a single field and checking it, and slowly integrating checks on other subsystems. Anyone got thoughts? On Mon, 28 Sept 2026 at 20:59, Janne Johansson <[email protected]> wrote: > Den mån 28 sep. 2026 kl 21:35 skrev Gwen Nelson <[email protected]>: > >> Hey all >> I'm considering implementing a jail-like subsystem for OpenBSD as part of >> restarting an old pubnix project. >> I don't mean porting or reproducing FreeBSD jails. I have some ideas for >> an implementation designed around OpenBSD's existing architecture and >> conventions. >> Before spending significant time developing it, I'd like to establish >> whether such a facility would be considered desirable in principle, >> assuming an implementation met the project's technical and code-quality >> standards. >> If there's no interest in such a facility upstream, that's fine; I'll >> design it as an external project instead. >> >> I'm particularly interested in whether this has been discussed >> previously, and whether there are architectural or policy reasons the >> project has deliberately avoided this kind of abstraction. >> > > Around the time for OpenBSD 4.4, Kristaps Dzonsons made "mult" work on > OpenBSD, enough for allowing sshd to run in such a "jail". > > Interview about the idea: > > https://bsdtalk.blogspot.com/2008/02/bsdtalk140-mult-project-with-kristaps.html > > In the end, sshd on OpenBSD probably fared better with pledge, unveil and > privsep, but I am sure someone would like to have jail-a-likes for network > services on OpenBSD given how popular the namespacing thing on linux and > FreeBSD jails have become. > > If it would ever go into base? No idea. > > -- > May the most significant bit of your life be positive. >
