I'm not sure if systemd understands the crypttab format. At least just a single 'timeout=0' as fourth field causes the crypt branch to wait forever. It allows three tries, then returns. Now the fstab branch, which I think was already started, waits forever. Since there will be no more events the system became broken. To fix this bug there must be a separation between "providing block devices" and "find block devices listed in fstab". The former can already handle the infinte timeout by using 'timeout=0'. Not sure if the thing would allow other values for timeout, or number of retries. But having either eternity or 90 seconds as timeout is fine with me. The latter can also handle timeouts. Just the bug is that both run in parallel. If they are separated in steps the regression is fixed and the case can be closed.