On Tue, Apr 11, 2023 at 1:03 AM Samuel Thibault <samuel.thiba...@gnu.org> wrote: > I tend to be very cautious with reply port reuse since it can confuse > servers a lot when e.g. interrupted, so it's generally safer not to try > to reuse them.
Huh? Confuse how? For one thing, the servers are unable to even tell if two send-once rights point to the same port (which is actually something that I'd very much like to change, since it would greatly benefit rpctrace, but that's a topic that deserves its own wall-of-text email). The servers can still notice if their send-once right dies though. If a call has been interrupted and we have not received a reply (EINTR or not), we surely must destroy the reply port used, if only for security reasons (so a malicious server won't be able to inject a fake reply to some other RPC we make later). But I don't see a reason to destroy the *current* reply port, the one used during signal handling, which we're longjumping out of. On the other hand it doesn't hurt much either, and signal handling is a slow path anyway, so we may just play it safe. Sergey
