!948 Dependency of Kmesh on Kernel Modification

Merge Pull Request from: @bitcoffee 
 
In the kernel, Kmesh forwards customer requests to actual backend service nodes through Layer 7 orchestration. This capability is per flow. When sending msg for the first time, Kmesh parses user Layer 7 packets and completes orchestration to complete link establishment. This requires that the pseudo link be established in the connect phase and the actual link be established in the sendmsg phase.
Therefore, the following modifications are involved:
1. The ULP framework needs to be supported in the connect phase. The l4 connect function needs to be replaced with the user-defined connect function.
2. After the L4 connect function is invoked, the L3 function can invoke the actual link establishment logic based on the error code and modify the return value of inet_stream_connect at the L3 layer.
3. In the sendmsg message, you can determine whether the delay in link setup is enabled based on the sock status.

Submission Instructions:
1. Add a writeable_tracepoint to modify the return value of __inet_stream_connect in inet_stream_connect.
2. The bpf_defer_connect flag is added to indicate whether the ebpf defer connect delay link establishment logic is enabled.
3. The ULP framework is added to support the ebpf program. The ULP framework can be used in the ebpf program.
4. A call type in sockops is added. This type is used to invoke the ebpf program in the kernel module and identify it when Kmesh delays link establishment. 
 
Link:https://gitee.com/openeuler/kernel/pulls/948 

Reviewed-by: Jialin Zhang <zhangjialin11@huawei.com> 
Signed-off-by: Jialin Zhang <zhangjialin11@huawei.com> 

include/net/inet_sock.h

@@ -240,10 +240,11 @@ struct inet_sock {
 				nodefrag:1;
 	__u8			bind_address_no_port:1,
 				recverr_rfc4884:1,
-				defer_connect:1; /* Indicates that fastopen_connect is set
+				defer_connect:1, /* Indicates that fastopen_connect is set
 						  * and cookie exists so we defer connect
 						  * until first data frame is written
 						  */
+				bpf_defer_connect:1;
 	__u8			rcv_tos;
 	__u8			convert_csum;
 	int			uc_index;

include/trace/events/sock.h

@@ -203,6 +203,20 @@ TRACE_EVENT(inet_sock_set_state,
 			show_tcp_state_name(__entry->newstate))
 );
 
+#undef NET_DECLARE_TRACE
+#ifdef DECLARE_TRACE_WRITABLE
+#define NET_DECLARE_TRACE(call, proto, args, size) \
+	DECLARE_TRACE_WRITABLE(call, PARAMS(proto), PARAMS(args), size)
+#else
+#define NET_DECLARE_TRACE(call, proto, args, size) \
+	DECLARE_TRACE(call, PARAMS(proto), PARAMS(args))
+#endif
+
+NET_DECLARE_TRACE(connect_ret,
+	TP_PROTO(int *err),
+	TP_ARGS(err),
+	sizeof(int));
+
 #endif /* _TRACE_SOCK_H */
 
 /* This part must be outside protection */

include/uapi/linux/bpf.h

@@ -4872,6 +4872,7 @@ enum {
 					 * by the kernel or the
 					 * earlier bpf-progs.
 					 */
+	BPF_SOCK_OPS_TCP_DEFER_CONNECT_CB,/* call ebpf to defer connect*/
 };
 
 /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect

net/core/filter.c

@@ -4837,6 +4837,13 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname,
 						    TCP_CA_NAME_MAX-1));
 			name[TCP_CA_NAME_MAX-1] = 0;
 			ret = tcp_set_congestion_control(sk, name, false, true);
+		} else if (optname == TCP_ULP) {
+			char name[TCP_ULP_NAME_MAX];
+
+			strncpy(name, optval, min_t(long, optlen,
+						    TCP_ULP_NAME_MAX - 1));
+			name[TCP_ULP_NAME_MAX - 1] = 0;
+			return tcp_set_ulp(sk, name);
 		} else {
 			struct inet_connection_sock *icsk = inet_csk(sk);
 			struct tcp_sock *tp = tcp_sk(sk);

net/ipv4/af_inet.c

@@ -729,6 +729,7 @@ int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
 	lock_sock(sock->sk);
 	err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
 	release_sock(sock->sk);
+	trace_connect_ret(&err);
 	return err;
 }
 EXPORT_SYMBOL(inet_stream_connect);

net/ipv4/tcp.c

@@ -590,7 +590,8 @@ __poll_t tcp_poll(struct file *file, struct socket *sock, poll_table *wait)
 
 		if (tp->urg_data & TCP_URG_VALID)
 			mask |= EPOLLPRI;
-	} else if (state == TCP_SYN_SENT && inet_sk(sk)->defer_connect) {
+	} else if (state == TCP_SYN_SENT &&
+			(inet_sk(sk)->defer_connect || inet_sk(sk)->bpf_defer_connect)) {
 		/* Active TCP fastopen socket with defer_connect
 		 * Return EPOLLOUT so application can call write()
 		 * in order for kernel to generate SYN+data

tools/include/uapi/linux/bpf.h

@@ -4872,6 +4872,7 @@ enum {
 					 * by the kernel or the
 					 * earlier bpf-progs.
 					 */
+	BPF_SOCK_OPS_TCP_DEFER_CONNECT_CB,/* call ebpf to defer connect*/
 };
 
 /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect