Fix misleading use of bpf_ntohl

`local_port` in `struct bpf_sock_ops` and `struct sk_msg_md` is stored in host byte order. The use of `bpf_ntohl` is kind of misleading. Signed-off-by: Wang Zhen <lazybetrayer@gmail.com>
cilium · Mar 22, 2023 · 522004b · 522004b
1 parent 29b29fe
commit 522004b
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/bpf/sockops/bpf_redir.c b/bpf/sockops/bpf_redir.c
@@ -29,7 +29,7 @@ static __always_inline void sk_msg_extract4_key(const struct sk_msg_md *msg,
 	key->sip4 = msg->local_ip4;
 	key->family = ENDPOINT_KEY_IPV4;
 
-	key->sport = (bpf_ntohl(msg->local_port) >> 16);
+	key->sport = (bpf_htonl(msg->local_port) >> 16);
 	/* clang-7.1 or higher seems to think it can do a 16-bit read here
 	 * which unfortunately most kernels (as of October 2019) do not
 	 * support, which leads to verifier failures. Insert a READ_ONCE

diff --git a/bpf/sockops/bpf_sockops.c b/bpf/sockops/bpf_sockops.c
@@ -30,7 +30,7 @@ static __always_inline void sk_extract4_key(const struct bpf_sock_ops *ops,
 	key->sip4 = ops->local_ip4;
 	key->family = ENDPOINT_KEY_IPV4;
 
-	key->sport = (bpf_ntohl(ops->local_port) >> 16);
+	key->sport = (bpf_htonl(ops->local_port) >> 16);
 	/* clang-7.1 or higher seems to think it can do a 16-bit read here
 	 * which unfortunately most kernels (as of October 2019) do not
 	 * support, which leads to verifier failures. Insert a READ_ONCE