Add flowtable support to IPv6

Tested by: qingli@

Reviewed by:	qingli@
MFC after:	3 days

4 files changed, 60 insertions, 6 deletions

diff --git a/sys/netinet6/in6_proto.c b/sys/netinet6/in6_proto.c
index 51f5187..b27893c 100644
--- a/sys/netinet6/in6_proto.c
+++ b/sys/netinet6/in6_proto.c
@@ -70,6 +70,7 @@ __FBSDID("$FreeBSD$");
 #include "opt_carp.h"
 #include "opt_sctp.h"
 #include "opt_mpath.h"
+#include "opt_route.h"
 
 #include <sys/param.h>
 #include <sys/socket.h>
@@ -130,6 +131,10 @@ __FBSDID("$FreeBSD$");
 
 #include <netinet6/ip6protosw.h>
 
+#ifdef FLOWTABLE
+#include <net/flowtable.h>
+#endif
+
 /*
  * TCP/IP protocol family: IP6, ICMP6, UDP, TCP.
  */
@@ -569,6 +574,16 @@ SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_STEALTH, stealth, CTLFLAG_RW,
 	&VNET_NAME(ip6stealth), 0, "");
 #endif
 
+#ifdef FLOWTABLE
+VNET_DEFINE(int, ip6_output_flowtable_size) = 2048;
+VNET_DEFINE(struct flowtable *, ip6_ft);
+#define	V_ip6_output_flowtable_size	VNET(ip6_output_flowtable_size)
+
+SYSCTL_VNET_INT(_net_inet6_ip6, OID_AUTO, output_flowtable_size, CTLFLAG_RDTUN,
+    &VNET_NAME(ip6_output_flowtable_size), 2048,
+    "number of entries in the per-cpu output flow caches");
+#endif
+
 /* net.inet6.icmp6 */
 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_REDIRACCEPT, rediraccept,
 	CTLFLAG_RW, &VNET_NAME(icmp6_rediraccept), 0, "");
diff --git a/sys/netinet6/ip6_input.c b/sys/netinet6/ip6_input.c
index 42f9351..088e6fa 100644
--- a/sys/netinet6/ip6_input.c
+++ b/sys/netinet6/ip6_input.c
@@ -66,6 +66,7 @@ __FBSDID("$FreeBSD$");
 #include "opt_inet.h"
 #include "opt_inet6.h"
 #include "opt_ipsec.h"
+#include "opt_route.h"
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -113,6 +114,12 @@ __FBSDID("$FreeBSD$");
 
 #include <netinet6/ip6protosw.h>
 
+#ifdef FLOWTABLE
+#include <net/flowtable.h>
+extern VNET_DEFINE(int, ip6_output_flowtable_size);
+#define	V_ip6_output_flowtable_size	VNET(ip6_output_flowtable_size)
+#endif
+
 extern struct domain inet6domain;
 
 u_char ip6_protox[IPPROTO_MAX];
@@ -169,6 +176,12 @@ ip6_init(void)
 	nd6_init();
 	frag6_init();
 
+#ifdef FLOWTABLE
+ 	TUNABLE_INT_FETCH("net.inet6.ip6.output_flowtable_size",
+ 	    &V_ip6_output_flowtable_size);
+ 	V_ip6_ft = flowtable_alloc("ipv6", V_ip6_output_flowtable_size, FL_PCPU);
+#endif	
+	
 	V_ip6_desync_factor = arc4random() % MAX_TEMP_DESYNC_FACTOR;
 
 	/* Skip global initialization stuff for non-default instances. */
diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
index a878aac..30ce838 100644
--- a/sys/netinet6/ip6_output.c
+++ b/sys/netinet6/ip6_output.c
@@ -67,6 +67,7 @@ __FBSDID("$FreeBSD$");
 #include "opt_inet6.h"
 #include "opt_ipsec.h"
 #include "opt_sctp.h"
+#include "opt_route.h"
 
 #include <sys/param.h>
 #include <sys/kernel.h>
@@ -111,6 +112,10 @@ __FBSDID("$FreeBSD$");
 #include <netinet6/ip6protosw.h>
 #include <netinet6/scope6_var.h>
 
+#ifdef FLOWTABLE
+#include <net/flowtable.h>
+#endif
+
 extern int in6_mcast_loop;
 
 struct ip6_exthdrs {
@@ -211,6 +216,7 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts *opt,
 	struct in6_addr finaldst, src0, dst0;
 	u_int32_t zone;
 	struct route_in6 *ro_pmtu = NULL;
+	int flevalid = 0;
 	int hdrsplit = 0;
 	int needipsec = 0;
 #ifdef SCTP
@@ -231,9 +237,7 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts *opt,
 	}
 
 	finaldst = ip6->ip6_dst;
-
 	bzero(&exthdrs, sizeof(exthdrs));
-
 	if (opt) {
 		/* Hop-by-Hop options header */
 		MAKE_EXTHDR(opt->ip6po_hbh, &exthdrs.ip6e_hbh);
@@ -470,7 +474,22 @@ skip_ipsec2:;
 	if (opt && opt->ip6po_rthdr)
 		ro = &opt->ip6po_route;
 	dst = (struct sockaddr_in6 *)&ro->ro_dst;
-
+#ifdef FLOWTABLE
+	if (ro == &ip6route) {
+		struct flentry *fle;
+		
+		/*
+		 * The flow table returns route entries valid for up to 30
+		 * seconds; we rely on the remainder of ip_output() taking no
+		 * longer than that long for the stability of ro_rt.  The
+		 * flow ID assignment must have happened before this point.
+		 */
+		if ((fle = flowtable_lookup_mbuf(V_ip6_ft, m, AF_INET6)) != NULL) {
+			flow_to_route_in6(fle, ro);
+			flevalid = 1;
+		}
+	}
+#endif	
 again:
 	/*
 	 * if specified, try to fill in the traffic class field.
@@ -576,7 +595,10 @@ again:
 	dst_sa.sin6_family = AF_INET6;
 	dst_sa.sin6_len = sizeof(dst_sa);
 	dst_sa.sin6_addr = ip6->ip6_dst;
-	if ((error = in6_selectroute(&dst_sa, opt, im6o, ro,
+	if (flevalid) {
+		rt = ro->ro_rt;
+		ifp = ro->ro_rt->rt_ifp;
+	} else if ((error = in6_selectroute(&dst_sa, opt, im6o, ro,
 	    &ifp, &rt)) != 0) {
 		switch (error) {
 		case EHOSTUNREACH:
@@ -1071,9 +1093,11 @@ sendorfree:
 		V_ip6stat.ip6s_fragmented++;
 
 done:
-	if (ro == &ip6route && ro->ro_rt) { /* brace necessary for RTFREE */
+	if (ro == &ip6route && ro->ro_rt && flevalid == 0) {
+                /* brace necessary for RTFREE */
 		RTFREE(ro->ro_rt);
-	} else if (ro_pmtu == &ip6route && ro_pmtu->ro_rt) {
+	} else if (ro_pmtu == &ip6route && ro_pmtu->ro_rt &&
+	    ((flevalid == 0) || (ro_pmtu != ro))) {
 		RTFREE(ro_pmtu->ro_rt);
 	}
 #ifdef IPSEC
diff --git a/sys/netinet6/udp6_usrreq.c b/sys/netinet6/udp6_usrreq.c
index 552e488..65e6f56 100644
--- a/sys/netinet6/udp6_usrreq.c
+++ b/sys/netinet6/udp6_usrreq.c
@@ -1079,7 +1079,9 @@ udp6_send(struct socket *so, int flags, struct mbuf *m,
 	mac_inpcb_create_mbuf(inp, m);
 #endif
 	error = udp6_output(inp, m, addr, control, td);
+#ifdef INET
 out:
+#endif	
 	INP_WUNLOCK(inp);
 	INP_INFO_WUNLOCK(&V_udbinfo);
 	return (error);