On 26.09.23 05:00, D. Wythe wrote: > You are right. The key point is how to ensure the valid of smc sock during the life time of clc sock, If so, READ_ONCE is good > enough. Unfortunately, I found that there are no such guarantee, so it's still a life-time problem. Did you discover a scenario, where clc sock could live longer than smc sock? Wouldn't that be a dangerous scenario in itself? I still have some hope that the lifetime of an smc socket is by design longer than that of the corresponding tcp socket. Considering the const, maybe > we need to do : > > 1. hold a refcnt of smc_sock for syn_recv_sock to keep smc sock valid during life time of clc sock > 2. put the refcnt of smc_sock in sk_destruct in tcp_sock to release the very smc sock . > > In that way, we can always make sure the valid of smc sock during the life time of clc sock. Then we can use READ_ONCE rather > than lock. What do you think ? I am not sure I fully understand the details what you propose to do. And it is not only syn_recv_sock(), right? You need to consider all relations between smc socks and tcp socks; fallback to tcp, initial creation, children of listen sockets, variants of shutdown, ... Preferrably a single simple mechanism covers all situations. Maybe there is such a mechanism already today? (I don't think clcsock->sk->sk_user_data or sk_callback_lock provide this general coverage) If we really have a gap, a general refcnt'ing on smc sock could be a solution, but needs to be designed carefully. Many thanks to you and the team to help make smc more stable and robust.
