On Fri, Feb 19, 2021 at 10:25 AM Jakub Sitnicki [off-list ref] wrote:

On Tue, Feb 16, 2021 at 07:42 AM CET, Cong Wang wrote:

quoted

From: Cong Wang <redacted>

As suggested by John, clean up sockmap related Kconfigs:

Reduce the scope of CONFIG_BPF_STREAM_PARSER down to TCP stream
parser, to reflect its name.

Make the rest sockmap code simply depend on CONFIG_BPF_SYSCALL.
And leave CONFIG_NET_SOCK_MSG untouched, as it is used by
non-sockmap cases.

Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Reviewed-by: Lorenz Bauer <redacted>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Cong Wang <redacted>
---

Sorry for the delay. There's a lot happening here. Took me a while to
dig through it.

I have a couple of nit-picks, which easily can be addressed as
follow-ups, and one comment.

No problem, it is not merged, so V5 is definitely not a problem.

sock_map_prog_update and sk_psock_done_strp are only used in
net/core/sock_map.c and can be static.

1. This seems to be unrelated to this patch? But I am still happy to
address it.

2. sk_psock_done_strp() is in skmsg.c, hence why it is non-static.
And I believe it fits in skmsg.c better than in sock_map.c, because
it operates on psock rather than sock_map itself.

So, I can make sock_map_prog_update() static in a separate patch
and carry it in V5.

[...]

quoted

diff --git a/net/ipv4/tcp_bpf.c b/net/ipv4/tcp_bpf.c
index bc7d2a586e18..b2c4865eb39b 100644
--- a/net/ipv4/tcp_bpf.c
+++ b/net/ipv4/tcp_bpf.c

@@ -229,7 +229,6 @@ int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg,
 }
 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);

-#ifdef CONFIG_BPF_STREAM_PARSER
 static bool tcp_bpf_stream_read(const struct sock *sk)
 {
      struct sk_psock *psock;

@@ -561,8 +560,10 @@ static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
                                 struct proto *base)
 {
      prot[TCP_BPF_BASE]                      = *base;
+#if defined(CONFIG_BPF_SYSCALL)
      prot[TCP_BPF_BASE].unhash               = sock_map_unhash;
      prot[TCP_BPF_BASE].close                = sock_map_close;
+#endif
      prot[TCP_BPF_BASE].recvmsg              = tcp_bpf_recvmsg;
      prot[TCP_BPF_BASE].stream_memory_read   = tcp_bpf_stream_read;

@@ -629,4 +630,3 @@ void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
      if (prot == &tcp_bpf_prots[family][TCP_BPF_BASE])
              newsk->sk_prot = sk->sk_prot_creator;
 }
-#endif /* CONFIG_BPF_STREAM_PARSER */

net/core/sock_map.o now is built only when CONFIG_BPF_SYSCALL is set.
While tcp_bpf_get_proto is only called from net/core/sock_map.o.

Seems there is no sense in compiling tcp_bpf_get_proto, and everything
it depends on which was enclosed by CONFIG_BPF_STREAM_PARSER check, when
CONFIG_BPF_SYSCALL is unset.

I can try but I am definitely not sure whether kTLS is happy about
it, clearly kTLS at least uses __tcp_bpf_recvmsg() and
tcp_bpf_sendmsg_redir().

quoted

diff --git a/net/ipv4/udp_bpf.c b/net/ipv4/udp_bpf.c
index 7a94791efc1a..e635ccc175ca 100644
--- a/net/ipv4/udp_bpf.c
+++ b/net/ipv4/udp_bpf.c

@@ -18,8 +18,10 @@ static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];
 static void udp_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
 {
      *prot        = *base;
+#if defined(CONFIG_BPF_SYSCALL)
      prot->unhash = sock_map_unhash;
      prot->close  = sock_map_close;
+#endif
 }

 static void udp_bpf_check_v6_needs_rebuild(struct proto *ops)

Same situation here but for udp_bpf_get_proto.

UDP is different, as kTLS certainly doesn't and won't use it. I think
udp_bpf.c can be just put under CONFIG_BPF_SYSCALL.

Thanks.