bpf, sk_msg: Don't clear saved sock proto on restore
authorJakub Sitnicki <jakub@cloudflare.com>
Mon, 17 Feb 2020 12:15:29 +0000 (12:15 +0000)
committerDaniel Borkmann <daniel@iogearbox.net>
Wed, 19 Feb 2020 15:54:05 +0000 (16:54 +0100)
There is no need to clear psock->sk_proto when restoring socket protocol
callbacks in sk->sk_prot. The psock is about to get detached from the sock
and eventually destroyed. At worst we will restore the protocol callbacks
and the write callback twice.

This makes reasoning about psock state easier. Once psock is initialized,
we can count on psock->sk_proto always being set.

Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200217121530.754315-3-jakub@cloudflare.com
include/linux/skmsg.h

index 8605947d6c08c9ca979c9dd1dcfabd6aea4ccf5c..d90ef61712a160f9a2e3ececf1af1464f73be94b 100644 (file)
@@ -359,13 +359,7 @@ static inline void sk_psock_restore_proto(struct sock *sk,
                                          struct sk_psock *psock)
 {
        sk->sk_prot->unhash = psock->saved_unhash;
-
-       if (psock->sk_proto) {
-               tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
-               psock->sk_proto = NULL;
-       } else {
-               sk->sk_write_space = psock->saved_write_space;
-       }
+       tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
 }
 
 static inline void sk_psock_set_state(struct sk_psock *psock,