Rework IPV6 TCP path MTU discovery to match IPv4

- Re-write tcp_ctlinput6() to closely mimic the IPv4 tcp_ctlinput()

- Now that tcp_ctlinput6() updates t_maxseg, we can allow ip6_output()
  to send TCP packets without looking at the tcp host cache for every
  single transmit.

- Make the icmp6 code mimic the IPv4 code & avoid returning
  PRC_HOSTDEAD because it is so expensive.

Without these changes in place, every TCP6 pmtu discovery or host
unreachable ICMP resulted in a call to in6_pcbnotify() which walks the
tcbinfo table with the write lock held.  Because the tcbinfo table is
shared between IPv4 and IPv6, this causes huge scalabilty issues on
servers with lots of (~100K) TCP connections, to the point where even
a small percent of IPv6 traffic had a disproportionate impact on
overall throughput.

Reviewed by:	bz, rrs, ae (all earlier versions), lstewart (in Netflix's tree)
Sponsored by:		Netflix
Differential Revision:	https://reviews.freebsd.org/D7272

2 files changed, 14 insertions, 11 deletions

diff --git a/sys/netinet6/icmp6.c b/sys/netinet6/icmp6.c
index 69bb60fcfeaa..6b6c92e2539f 100644
--- a/sys/netinet6/icmp6.c
+++ b/sys/netinet6/icmp6.c
@@ -485,15 +485,13 @@ icmp6_input(struct mbuf **mp, int *offp, int proto)
 		icmp6_ifstat_inc(ifp, ifs6_in_dstunreach);
 		switch (code) {
 		case ICMP6_DST_UNREACH_NOROUTE:
+		case ICMP6_DST_UNREACH_ADDR:	/* PRC_HOSTDEAD is a DOS */
 			code = PRC_UNREACH_NET;
 			break;
 		case ICMP6_DST_UNREACH_ADMIN:
 			icmp6_ifstat_inc(ifp, ifs6_in_adminprohib);
 			code = PRC_UNREACH_PROTOCOL; /* is this a good code? */
 			break;
-		case ICMP6_DST_UNREACH_ADDR:
-			code = PRC_HOSTDEAD;
-			break;
 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
 			/* I mean "source address was incorrect." */
 			code = PRC_PARAMPROB;
diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
index a768f74d033e..a393cb51aaac 100644
--- a/sys/netinet6/ip6_output.c
+++ b/sys/netinet6/ip6_output.c
@@ -150,9 +150,10 @@ static int ip6_insertfraghdr(struct mbuf *, struct mbuf *, int,
 static int ip6_insert_jumboopt(struct ip6_exthdrs *, u_int32_t);
 static int ip6_splithdr(struct mbuf *, struct ip6_exthdrs *);
 static int ip6_getpmtu(struct route_in6 *, int,
-	struct ifnet *, const struct in6_addr *, u_long *, int *, u_int);
+	struct ifnet *, const struct in6_addr *, u_long *, int *, u_int,
+	u_int);
 static int ip6_calcmtu(struct ifnet *, const struct in6_addr *, u_long,
-	u_long *, int *);
+	u_long *, int *, u_int);
 static int ip6_getpmtu_ctl(u_int, const struct in6_addr *, u_long *);
 static int copypktopts(struct ip6_pktopts *, struct ip6_pktopts *, int);
 
@@ -718,7 +719,7 @@ again:
 
 	/* Determine path MTU. */
 	if ((error = ip6_getpmtu(ro_pmtu, ro != ro_pmtu, ifp, &ip6->ip6_dst,
-	    &mtu, &alwaysfrag, fibnum)) != 0)
+		    &mtu, &alwaysfrag, fibnum, *nexthdrp)) != 0)
 		goto bad;
 
 	/*
@@ -1250,7 +1251,7 @@ ip6_getpmtu_ctl(u_int fibnum, const struct in6_addr *dst, u_long *mtup)
 	ifp = nh6.nh_ifp;
 	mtu = nh6.nh_mtu;
 
-	error = ip6_calcmtu(ifp, dst, mtu, mtup, NULL);
+	error = ip6_calcmtu(ifp, dst, mtu, mtup, NULL, 0);
 	fib6_free_nh_ext(fibnum, &nh6);
 
 	return (error);
@@ -1269,7 +1270,7 @@ ip6_getpmtu_ctl(u_int fibnum, const struct in6_addr *dst, u_long *mtup)
 static int
 ip6_getpmtu(struct route_in6 *ro_pmtu, int do_lookup,
     struct ifnet *ifp, const struct in6_addr *dst, u_long *mtup,
-    int *alwaysfragp, u_int fibnum)
+    int *alwaysfragp, u_int fibnum, u_int proto)
 {
 	struct nhop6_basic nh6;
 	struct in6_addr kdst;
@@ -1307,7 +1308,7 @@ ip6_getpmtu(struct route_in6 *ro_pmtu, int do_lookup,
 	if (ro_pmtu->ro_rt)
 		mtu = ro_pmtu->ro_rt->rt_mtu;
 
-	return (ip6_calcmtu(ifp, dst, mtu, mtup, alwaysfragp));
+	return (ip6_calcmtu(ifp, dst, mtu, mtup, alwaysfragp, proto));
 }
 
 /*
@@ -1319,7 +1320,7 @@ ip6_getpmtu(struct route_in6 *ro_pmtu, int do_lookup,
  */
 static int
 ip6_calcmtu(struct ifnet *ifp, const struct in6_addr *dst, u_long rt_mtu,
-    u_long *mtup, int *alwaysfragp)
+    u_long *mtup, int *alwaysfragp, u_int proto)
 {
 	u_long mtu = 0;
 	int alwaysfrag = 0;
@@ -1334,7 +1335,11 @@ ip6_calcmtu(struct ifnet *ifp, const struct in6_addr *dst, u_long rt_mtu,
 		inc.inc6_faddr = *dst;
 
 		ifmtu = IN6_LINKMTU(ifp);
-		mtu = tcp_hc_getmtu(&inc);
+
+		/* TCP is known to react to pmtu changes so skip hc */
+		if (proto != IPPROTO_TCP)
+			mtu = tcp_hc_getmtu(&inc);
+
 		if (mtu)
 			mtu = min(mtu, rt_mtu);
 		else