Skip to contentSkip to conversation

mptcp: update window_clamp on subflows when SO_RCVBUF is set

PATCH In review Gang Yan 1 participant 2 messages Apr 16

Subsystems (2)
lore
Share & tools
lore

Gang Yan

author
#0 lore raw

PATCH 1/1 mptcp: update window_clamp on subflows when SO_RCVBUF is set

From: Gang Yan <[email protected]>

Add __mptcp_subflow_set_rcvbuf() helper that calls the subflow's ops->set_rcvbuf to update window_clamp, in addition to writing sk_rcvbuf. Use it in both mptcp_sol_socket_sync_intval() (setsockopt path) and sync_socket_options() (new subflow creation path).

Fixes a2cbb1603943 ("tcp: Update window clamping condition")

---
Notes:

Hi Matt:

Note that I do not implement set_rcvbuf in mptcp_stream_ops for the following reasons:

- The callback receives the MPTCP meta socket as 'sk', but tcp_set_rcvbuf() must be called on each TCP subflow to update window_clamp. A callback implementation would need to iterate all subflows, duplicating the work already done by mptcp_sol_socket_sync_intval().

- Subflows are grafted to the MPTCP socket via mptcp_sock_graft(), so ssk->sk_socket->ops points to mptcp_stream_ops, not inet_stream_ops. This means the TCP set_rcvbuf callback is not reachable through the subflow's ops either.

- The existing subflow sync functions cover both the setsockopt path (mptcp_sol_socket_sync_intval) and new subflow creation path (sync_socket_options), making the proto_ops callback unnecessary.

WDYT?

The packetdrill tests is also implemented:
-Link: https://github.com/multipath-tcp/packetdrill/pull/194

Thanks
Gang

---
 net/mptcp/sockopt.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
net/mptcp/sockopt.c +8 -2
index de90a2897d2d..abe63392826b 100644
@@ -67,6 +67,12 @@ static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
 	return 0;
 }
 
+static inline void __mptcp_subflow_set_rcvbuf(struct sock *ssk, int val)
+{
+	WRITE_ONCE(ssk->sk_rcvbuf, val);
+	tcp_set_rcvbuf(ssk, val);
+}
+
 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
 {
 	struct mptcp_subflow_context *subflow;
@@ -100,7 +106,7 @@ static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, in
 		case SO_RCVBUF:
 		case SO_RCVBUFFORCE:
 			ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
-			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
+			__mptcp_subflow_set_rcvbuf(ssk, sk->sk_rcvbuf);
 			break;
 		case SO_MARK:
 			if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
@@ -1556,7 +1562,7 @@ static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
 			mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
 		}
 		if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
-			WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
+			__mptcp_subflow_set_rcvbuf(ssk, sk->sk_rcvbuf);
 	}
 
 	if (sock_flag(sk, SOCK_LINGER)) {
Signature · 1 line
2.43.0
#1 lore raw
Quoted 8 lines · Gang Yan

From: Gang Yan <[email protected]>

Add __mptcp_subflow_set_rcvbuf() helper that calls the subflow's ops->set_rcvbuf to update window_clamp, in addition to writing sk_rcvbuf. Use it in both mptcp_sol_socket_sync_intval() (setsockopt path) and sync_socket_options() (new subflow creation path).

Sorry for the confusion in my commit message.

I need to clarify that we cannot call ops->set_rcvbuf() on the MPTCP subflows as mentioned in the commit message. I will fix this mistake and update the commit message properly in the next version.

Thank you for your understanding.

Thanks
Gang

Quoted 76 lines
Fixes a2cbb1603943 ("tcp: Update window clamping condition")

---
Notes:

Hi Matt:

Note that I do not implement set_rcvbuf in mptcp_stream_ops for the following reasons:

- The callback receives the MPTCP meta socket as 'sk', but tcp_set_rcvbuf() must be called on each TCP subflow to update window_clamp. A callback implementation would need to iterate all subflows, duplicating the work already done by mptcp_sol_socket_sync_intval().

- Subflows are grafted to the MPTCP socket via mptcp_sock_graft(), so ssk->sk_socket->ops points to mptcp_stream_ops, not inet_stream_ops. This means the TCP set_rcvbuf callback is not reachable through the subflow's ops either.

- The existing subflow sync functions cover both the setsockopt path (mptcp_sol_socket_sync_intval) and new subflow creation path (sync_socket_options), making the proto_ops callback unnecessary.

WDYT?

The packetdrill tests is also implemented:
-Link: https://github.com/multipath-tcp/packetdrill/pull/194

Thanks
Gang

---
 net/mptcp/sockopt.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
net/mptcp/sockopt.c +8 -2
index de90a2897d2d..abe63392826b 100644
@@ -67,6 +67,12 @@ static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval,
 return 0;
 }
 
+static inline void __mptcp_subflow_set_rcvbuf(struct sock *ssk, int val)
+{
+ WRITE_ONCE(ssk->sk_rcvbuf, val);
+ tcp_set_rcvbuf(ssk, val);
+}
+
 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val)
 {
 struct mptcp_subflow_context *subflow;
@@ -100,7 +106,7 @@ static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, in
 case SO_RCVBUF:
 case SO_RCVBUFFORCE:
 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK;
- WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
+ __mptcp_subflow_set_rcvbuf(ssk, sk->sk_rcvbuf);
 break;
 case SO_MARK:
 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) {
@@ -1556,7 +1562,7 @@ static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk)
 mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf;
 }
 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK)
- WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf);
+ __mptcp_subflow_set_rcvbuf(ssk, sk->sk_rcvbuf);
 }
 
 if (sock_flag(sk, SOCK_LINGER)) {
Signature · 1 line
2.43.0