mptcp: BLocking ESTimation-based (BLEST) Scheduler

This scheduler works much like the default MPTCP scheduler. It always
prefers the subflow with the smallest round-trip-time that is available.
If the network paths are very heterogeneous, then BLEST can be more
beneficial than the default scheduler because it estimates if sending on
a "slower" subflow would cause head-of-line blocking and decides to wait
until the "faster" subflow becomes available again.

Co-authored-by: Simone Ferlin <ferlin@simula.no>
Signed-off-by: Daniel Weber <weberd@cs.uni-bonn.de>

net/ipv4/tcp_output.c

@@ -1821,6 +1821,7 @@ unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
 	halfcwnd = max(cwnd >> 1, 1U);
 	return min(halfcwnd, cwnd - in_flight);
 }
+EXPORT_SYMBOL(tcp_cwnd_test);
 
 /* Initialize TSO state of a skb.
  * This must be invoked the first time we consider transmitting
@@ -1875,6 +1876,7 @@ bool tcp_snd_wnd_test(const struct tcp_sock *tp, const struct sk_buff *skb,
 
 	return !after(end_seq, tcp_wnd_end(tp));
 }
+EXPORT_SYMBOL(tcp_snd_wnd_test);
 
 /* Trim TSO SKB to LEN bytes, put the remaining data into a new packet
  * which is put after SKB on the list.  It is very much like

net/mptcp/Kconfig

@@ -89,6 +89,12 @@ menuconfig MPTCP_SCHED_ADVANCED
 
 if MPTCP_SCHED_ADVANCED
 
+config MPTCP_BLEST
+	tristate "MPTCP BLEST"
+	depends on MPTCP=y
+	---help---
+	  This is an experimental BLocking ESTimation-based (BLEST) scheduler.
+
 config MPTCP_ROUNDROBIN
 	tristate "MPTCP Round-Robin"
 	depends on (MPTCP=y)

net/mptcp/Makefile

@@ -19,5 +19,6 @@ obj-$(CONFIG_MPTCP_BINDER) += mptcp_binder.o
 obj-$(CONFIG_MPTCP_NETLINK) += mptcp_netlink.o
 obj-$(CONFIG_MPTCP_ROUNDROBIN) += mptcp_rr.o
 obj-$(CONFIG_MPTCP_REDUNDANT) += mptcp_redundant.o
+obj-$(CONFIG_MPTCP_BLEST) += mptcp_blest.o
 
 mptcp-$(subst m,y,$(CONFIG_IPV6)) += mptcp_ipv6.o