systemd version the issue has been seen with

9e12d5b

Used distribution

openSUSE Tumbleweed

Expected behaviour you didn't see

Starting a unit under frozen slice or stopping an indirectly frozen unit is rejected on the job level.

Unexpected behaviour you saw

Attempt to spawn starting/control process in a frozen cgroup leading to various timeouts and confusing reported states (new unit unfrozen, old unit failed (frozen)).

Steps to reproduce the problem

systemd-run --slice=frozen.slice cmd
systemctl stop unit-frozen-by-ancestor.slice

Cc: @msekletar