From 51199405f967207de372d9b60989eb87d7ae8809 Mon Sep 17 00:00:00 2001 From: John Fastabend Date: Thu, 20 Dec 2018 11:35:32 -0800 Subject: [PATCH] bpf: skb_verdict, support SK_PASS on RX BPF path Add SK_PASS verdict support to SK_SKB_VERDICT programs. Now that support for redirects exists we can implement SK_PASS as a redirect to the same socket. This simplifies the BPF programs and avoids an extra map lookup on RX path for simple visibility cases. Further, reduces user (BPF programmer in this context) confusion when their program drops skb due to lack of support. Signed-off-by: John Fastabend Signed-off-by: Daniel Borkmann --- net/core/skmsg.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/net/core/skmsg.c b/net/core/skmsg.c index 56a99d0c9aa08..8a91a460de8f6 100644 --- a/net/core/skmsg.c +++ b/net/core/skmsg.c @@ -669,6 +669,22 @@ static void sk_psock_verdict_apply(struct sk_psock *psock, bool ingress; switch (verdict) { + case __SK_PASS: + sk_other = psock->sk; + if (sock_flag(sk_other, SOCK_DEAD) || + !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) { + goto out_free; + } + if (atomic_read(&sk_other->sk_rmem_alloc) <= + sk_other->sk_rcvbuf) { + struct tcp_skb_cb *tcp = TCP_SKB_CB(skb); + + tcp->bpf.flags |= BPF_F_INGRESS; + skb_queue_tail(&psock->ingress_skb, skb); + schedule_work(&psock->work); + break; + } + goto out_free; case __SK_REDIRECT: sk_other = tcp_skb_bpf_redirect_fetch(skb); if (unlikely(!sk_other))