On 6/21/23 12:42 PM, Gilad Sever wrote:
When calling bpf_sk_lookup_tcp(), bpf_sk_lookup_udp() or bpf_skc_lookup_tcp() from tc/xdp ingress, VRF socket bindings aren't respoected, i.e. unbound sockets are returned, and bound sockets aren't found.
VRF binding is determined by the sdif argument to sk_lookup(), however when called from tc the IP SKB control block isn't initialized and thus inet{,6}_sdif() always returns 0.
Fix by calculating sdif for the tc/xdp flows by observing the device's l3 enslaved state.
The cg/sk_skb hooking points which are expected to support inet{,6}_sdif() pass sdif=-1 which makes __bpf_skc_lookup() use the existing logic.
Fixes: 6acc9b432e67 ("bpf: Add helper to retrieve socket in BPF") Acked-by: Stanislav Fomichev sdf@google.com Reviewed-by: Shmulik Ladkani shmulik.ladkani@gmail.com Reviewed-by: Eyal Birger eyal.birger@gmail.com Signed-off-by: Gilad Sever gilad9366@gmail.com
[...]
static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = { @@ -6776,12 +6788,17 @@ static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = { BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb, struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) {
- struct net *caller_net = dev_net(skb->dev);
- int ifindex = skb->dev->ifindex;
- struct net_device *dev = skb->dev;
- struct net *caller_net;
- int ifindex, sdif;
- caller_net = dev_net(dev);
- ifindex = dev->ifindex;
- sdif = dev_sdif(dev);
return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net, ifindex, IPPROTO_UDP, netns_id,
flags);
flags, sdif);
Applied, thanks! Btw, the above looks now way uglier than before, so I fixed this up wrt how it should have looked like for the submission:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=...
Thanks, Daniel