MFC r272720, 273061, 273062, 273063, 273064

Implement PLPMTUD blackhole detection (RFC 4821), inspired by code
from xnu sources.  If we encounter a network where ICMP is blocked
the Needs Frag indicator may not propagate back to us.  Attempt to
downshift the mss once to a preconfigured value.

Note, this is turned off by default.

3 files changed, 182 insertions, 2 deletions

diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c
index 8c97c12..8abb16d 100644
--- a/sys/netinet/tcp_output.c
+++ b/sys/netinet/tcp_output.c
@@ -676,6 +676,12 @@ just_return:
 
 send:
 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
+	if (len > 0) {
+		if (len >= tp->t_maxseg)
+			tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT;
+		else
+			tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT;
+	}
 	/*
 	 * Before ESTABLISHED, force sending of initial options
 	 * unless TCP set not to do any options.
@@ -1184,6 +1190,11 @@ send:
 		 */
 		ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6));
 
+		if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
+			tp->t_flags2 |= TF2_PLPMTU_PMTUD;
+		else
+			tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
+
 		if (tp->t_state == TCPS_SYN_SENT)
 			TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th);
 
@@ -1220,8 +1231,12 @@ send:
 	 *
 	 * NB: Don't set DF on small MTU/MSS to have a safe fallback.
 	 */
-	if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss)
+	if (V_path_mtu_discovery && tp->t_maxopd > V_tcp_minmss) {
 		ip->ip_off |= htons(IP_DF);
+		tp->t_flags2 |= TF2_PLPMTU_PMTUD;
+	} else {
+		tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
+	}
 
 	if (tp->t_state == TCPS_SYN_SENT)
 		TCP_PROBE5(connect__request, NULL, tp, ip, tp, th);
diff --git a/sys/netinet/tcp_timer.c b/sys/netinet/tcp_timer.c
index bde7503..00fcb22 100644
--- a/sys/netinet/tcp_timer.c
+++ b/sys/netinet/tcp_timer.c
@@ -63,6 +63,9 @@ __FBSDID("$FreeBSD$");
 #include <netinet/tcp_fsm.h>
 #include <netinet/tcp_timer.h>
 #include <netinet/tcp_var.h>
+#ifdef INET6
+#include <netinet6/tcp6_var.h>
+#endif
 #include <netinet/tcpip.h>
 #ifdef TCPDEBUG
 #include <netinet/tcp_debug.h>
@@ -124,6 +127,54 @@ SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
     &tcp_rexmit_drop_options, 0,
     "Drop TCP options from 3rd and later retransmitted SYN");
 
+static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
+#define	V_tcp_pmtud_blackhole_detect	VNET(tcp_pmtud_blackhole_detect)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
+    CTLFLAG_RW,
+    &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
+    "Path MTU Discovery Black Hole Detection Enabled");
+
+static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
+#define	V_tcp_pmtud_blackhole_activated \
+    VNET(tcp_pmtud_blackhole_activated)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
+    CTLFLAG_RD,
+    &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
+    "Path MTU Discovery Black Hole Detection, Activation Count");
+
+static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
+#define	V_tcp_pmtud_blackhole_activated_min_mss \
+    VNET(tcp_pmtud_blackhole_activated_min_mss)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
+    CTLFLAG_RD,
+    &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
+    "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
+
+static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
+#define	V_tcp_pmtud_blackhole_failed	VNET(tcp_pmtud_blackhole_failed)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
+    CTLFLAG_RD,
+    &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
+    "Path MTU Discovery Black Hole Detection, Failure Count");
+
+#ifdef INET
+static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
+#define	V_tcp_pmtud_blackhole_mss	VNET(tcp_pmtud_blackhole_mss)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
+    CTLFLAG_RW,
+    &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
+    "Path MTU Discovery Black Hole Detection lowered MSS");
+#endif
+
+#ifdef INET6
+static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
+#define	V_tcp_v6pmtud_blackhole_mss	VNET(tcp_v6pmtud_blackhole_mss)
+SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
+    CTLFLAG_RW,
+    &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
+    "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
+#endif
+
 static int	per_cpu_timers = 0;
 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
     &per_cpu_timers , 0, "run tcp timers on all cpus");
@@ -492,6 +543,7 @@ tcp_timer_rexmt(void * xtp)
 
 	ostate = tp->t_state;
 #endif
+
 	INP_INFO_RLOCK(&V_tcbinfo);
 	inp = tp->t_inpcb;
 	/*
@@ -593,6 +645,110 @@ tcp_timer_rexmt(void * xtp)
 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
 		      tp->t_rttmin, TCPTV_REXMTMAX);
+
+	/*
+	 * We enter the path for PLMTUD if connection is established or, if
+	 * connection is FIN_WAIT_1 status, reason for the last is that if
+	 * amount of data we send is very small, we could send it in couple of
+	 * packets and process straight to FIN. In that case we won't catch
+	 * ESTABLISHED state.
+	 */
+	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
+	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
+		int optlen;
+#ifdef INET6
+		int isipv6;
+#endif
+
+		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
+		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
+		    (tp->t_rxtshift <= 2)) {
+			/*
+			 * Enter Path MTU Black-hole Detection mechanism:
+			 * - Disable Path MTU Discovery (IP "DF" bit).
+			 * - Reduce MTU to lower value than what we
+			 *   negotiated with peer.
+			 */
+			/* Record that we may have found a black hole. */
+			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
+
+			/* Keep track of previous MSS. */
+			optlen = tp->t_maxopd - tp->t_maxseg;
+			tp->t_pmtud_saved_maxopd = tp->t_maxopd;
+
+			/* 
+			 * Reduce the MSS to blackhole value or to the default
+			 * in an attempt to retransmit.
+			 */
+#ifdef INET6
+			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
+			if (isipv6 &&
+			    tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
+				/* Use the sysctl tuneable blackhole MSS. */
+				tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
+				V_tcp_pmtud_blackhole_activated++;
+			} else if (isipv6) {
+				/* Use the default MSS. */
+				tp->t_maxopd = V_tcp_v6mssdflt;
+				/*
+				 * Disable Path MTU Discovery when we switch to
+				 * minmss.
+				 */
+				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
+				V_tcp_pmtud_blackhole_activated_min_mss++;
+			}
+#endif
+#if defined(INET6) && defined(INET)
+			else
+#endif
+#ifdef INET
+			if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
+				/* Use the sysctl tuneable blackhole MSS. */
+				tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
+				V_tcp_pmtud_blackhole_activated++;
+			} else {
+				/* Use the default MSS. */
+				tp->t_maxopd = V_tcp_mssdflt;
+				/*
+				 * Disable Path MTU Discovery when we switch to
+				 * minmss.
+				 */
+				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
+				V_tcp_pmtud_blackhole_activated_min_mss++;
+			}
+#endif
+			tp->t_maxseg = tp->t_maxopd - optlen;
+			/*
+			 * Reset the slow-start flight size
+			 * as it may depend on the new MSS.
+			 */
+			if (CC_ALGO(tp)->conn_init != NULL)
+				CC_ALGO(tp)->conn_init(tp->ccv);
+		} else {
+			/*
+			 * If further retransmissions are still unsuccessful
+			 * with a lowered MTU, maybe this isn't a blackhole and
+			 * we restore the previous MSS and blackhole detection
+			 * flags.
+			 */
+			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
+			    (tp->t_rxtshift > 4)) {
+				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
+				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
+				optlen = tp->t_maxopd - tp->t_maxseg;
+				tp->t_maxopd = tp->t_pmtud_saved_maxopd;
+				tp->t_maxseg = tp->t_maxopd - optlen;
+				V_tcp_pmtud_blackhole_failed++;
+				/*
+				 * Reset the slow-start flight size as it
+				 * may depend on the new MSS.
+				 */
+				if (CC_ALGO(tp)->conn_init != NULL)
+					CC_ALGO(tp)->conn_init(tp->ccv);
+			}
+		}
+	}
+
 	/*
 	 * Disable RFC1323 and SACK if we haven't got any response to
 	 * our third SYN to work-around some broken terminal servers
diff --git a/sys/netinet/tcp_var.h b/sys/netinet/tcp_var.h
index 398df8b..10b1096 100644
--- a/sys/netinet/tcp_var.h
+++ b/sys/netinet/tcp_var.h
@@ -209,8 +209,10 @@ struct tcpcb {
 	u_int	t_keepcnt;		/* number of keepalives before close */
 
 	u_int	t_tsomax;		/* tso burst length limit */
+	u_int	t_pmtud_saved_maxopd;	/* pre-blackhole MSS */
+	u_int	t_flags2;		/* More tcpcb flags storage */
 
-	uint32_t t_ispare[8];		/* 5 UTO, 3 TBD */
+	uint32_t t_ispare[6];		/* 5 UTO, 1 TBD */
 	void	*t_pspare2[4];		/* 1 TCP_SIGNATURE, 3 TBD */
 	uint64_t _pad[6];		/* 6 TBD (1-2 CC/RTT?) */
 };
@@ -284,6 +286,13 @@ struct tcpcb {
 #endif /* TCP_SIGNATURE */
 
 /*
+ * Flags for PLPMTU handling, t_flags2
+ */
+#define	TF2_PLPMTU_BLACKHOLE	0x00000001 /* Possible PLPMTUD Black Hole. */
+#define	TF2_PLPMTU_PMTUD	0x00000002 /* Allowed to attempt PLPMTUD. */
+#define	TF2_PLPMTU_MAXSEGSNT	0x00000004 /* Last seg sent was full seg. */
+
+/*
  * Structure to hold TCP options that are only used during segment
  * processing (in tcp_input), but not held in the tcpcb.
  * It's basically used to reduce the number of parameters