On 2026-11-15 05:27, [email protected] wrote:
systemd answers openat2 with ENOSYS on purpose, expecting callers to fall back to openat: its RestrictSUIDSGID= seccomp filter cannot inspect struct open_how.
That's a severe deficiency in seccomp and should be fixed there, no? Do you have a bug report against seccomp and/or systemd that we can cite? I suppose that at configure time, we could see if it's possible that the bug is present, and work around it.
Suggested fix: keep the emulation fallback on ENOSYS even when libc provides openat2, for example by always wrapping it on Linux.
That would mean every program, not just Gnulib-using programs, must provide the functionality of openat2 themselves. Which is clearly a bad approach; it's not far from simply saying, "Don't use openat2."
