bpf: tcp: Handle BPF SYN Cookie in skb_steal_sock().
authorKuniyuki Iwashima <kuniyu@amazon.com>
Mon, 15 Jan 2024 20:55:11 +0000 (12:55 -0800)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 23 Jan 2024 22:40:24 +0000 (14:40 -0800)
We will support arbitrary SYN Cookie with BPF.

If BPF prog validates ACK and kfunc allocates a reqsk, it will
be carried to TCP stack as skb->sk with req->syncookie 1.  Also,
the reqsk has its listener as req->rsk_listener with no refcnt
taken.

When the TCP stack looks up a socket from the skb, we steal
inet_reqsk(skb->sk)->rsk_listener in skb_steal_sock() so that
the skb will be processed in cookie_v[46]_check() with the
listener.

Note that we do not clear skb->sk and skb->destructor so that we
can carry the reqsk to cookie_v[46]_check().

Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240115205514.68364-4-kuniyu@amazon.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
include/net/request_sock.h

index 26c630c..8839133 100644 (file)
@@ -101,10 +101,21 @@ static inline struct sock *skb_steal_sock(struct sk_buff *skb,
        }
 
        *prefetched = skb_sk_is_prefetched(skb);
-       if (*prefetched)
+       if (*prefetched) {
+#if IS_ENABLED(CONFIG_SYN_COOKIES)
+               if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
+                       struct request_sock *req = inet_reqsk(sk);
+
+                       *refcounted = false;
+                       sk = req->rsk_listener;
+                       req->rsk_listener = NULL;
+                       return sk;
+               }
+#endif
                *refcounted = sk_is_refcounted(sk);
-       else
+       } else {
                *refcounted = true;
+       }
 
        skb->destructor = NULL;
        skb->sk = NULL;