Just a note, this is still racy.

I originally had just (p = tp->t_session->s_leader) != NULL but then copied the example from tty_signal_sessleader. To make sure I understand correctly there's not a unique problem with my use but rather this pattern in general?

In D44086#1005754, @emaste wrote:

Just a note, this is still racy.

I originally had just (p = tp->t_session->s_leader) != NULL but then copied the example from tty_signal_sessleader. To make sure I understand correctly there's not a unique problem with my use but rather this pattern in general?

The session is not locked, so leader might be changed under us. It should not cause kernel memory corruption, since struct proc is type-stable, but in principle results in accessing tp after the test result is potentially invalid outright or invalidated after the check.