tcp: mptcp: use mptcp receive buffer space to select rcv window

In MPTCP, the receive window is shared across all subflows, because it
refers to the mptcp-level sequence space.

MPTCP receivers already place incoming packets on the mptcp socket
receive queue and will charge it to the mptcp socket rcvbuf until
userspace consumes the data.

Update __tcp_select_window to use the occupancy of the parent/mptcp
socket instead of the subflow socket in case the tcp socket is part
of a logical mptcp connection.

This commit doesn't change choice of initial window for passive or active
connections.
While it would be possible to change those as well, this adds complexity
(especially when handling MP_JOIN requests).  Furthermore, the MPTCP RFC
specifically says that a MPTCP sender 'MUST NOT use the RCV.WND field
of a TCP segment at the connection level if it does not also carry a DSS
option with a Data ACK field.'

SYN/SYNACK packets do not carry a DSS option with a Data ACK field.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/net/mptcp.h b/include/net/mptcp.h
index 0e7c5471010bda14c7a17f096036cdf425b63fc3..5288fba56e553fdd858c636e98e117f85d8d3a44 100644 (file)
--- a/include/net/mptcp.h
+++ b/include/net/mptcp.h
@@ -68,6 +68,8 @@ static inline bool rsk_is_mptcp(const struct request_sock *req)
        return tcp_rsk(req)->is_mptcp;
 }
 
+void mptcp_space(const struct sock *ssk, int *space, int *full_space);
+
 void mptcp_parse_option(const struct sk_buff *skb, const unsigned char *ptr,
                        int opsize, struct tcp_options_received *opt_rx);
 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
@@ -197,6 +199,7 @@ static inline bool mptcp_sk_is_subflow(const struct sock *sk)
        return false;
 }
 
+static inline void mptcp_space(const struct sock *ssk, int *s, int *fs) { }
 static inline void mptcp_seq_show(struct seq_file *seq) { }
 #endif /* CONFIG_MPTCP */
 

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 2f45cde168c4abca8248fb835f79149260c7f418..ba4482130f08ea386913bfa5de3253e9bbfa3030 100644 (file)
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -2772,8 +2772,12 @@ u32 __tcp_select_window(struct sock *sk)
        int mss = icsk->icsk_ack.rcv_mss;
        int free_space = tcp_space(sk);
        int allowed_space = tcp_full_space(sk);
-       int full_space = min_t(int, tp->window_clamp, allowed_space);
-       int window;
+       int full_space, window;
+
+       if (sk_is_mptcp(sk))
+               mptcp_space(sk, &free_space, &allowed_space);
+
+       full_space = min_t(int, tp->window_clamp, allowed_space);
 
        if (unlikely(mss > full_space)) {
                mss = full_space;

diff --git a/net/mptcp/subflow.c b/net/mptcp/subflow.c
index fabd06f2ff455e76cf281c4e6fc2d41b144ac2d9..87c094702d6377a37c2a69fe733ecbe214954a54 100644 (file)
--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -821,6 +821,24 @@ bool mptcp_subflow_data_available(struct sock *sk)
        return subflow->data_avail;
 }
 
+/* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
+ * not the ssk one.
+ *
+ * In mptcp, rwin is about the mptcp-level connection data.
+ *
+ * Data that is still on the ssk rx queue can thus be ignored,
+ * as far as mptcp peer is concerened that data is still inflight.
+ * DSS ACK is updated when skb is moved to the mptcp rx queue.
+ */
+void mptcp_space(const struct sock *ssk, int *space, int *full_space)
+{
+       const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
+       const struct sock *sk = subflow->conn;
+
+       *space = tcp_space(sk);
+       *full_space = tcp_full_space(sk);
+}
+
 static void subflow_data_ready(struct sock *sk)
 {
        struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);