MFC r325370:

Allow the setting of the MTU for future paths using an SCTP socket option.
This functionality was missing.

7 files changed, 78 insertions, 21 deletions

diff --git a/sys/netinet/sctp_output.c b/sys/netinet/sctp_output.c
index 12075fe..2a667c1 100644
--- a/sys/netinet/sctp_output.c
+++ b/sys/netinet/sctp_output.c
@@ -4273,12 +4273,8 @@ sctp_lowlevel_chunk_output(struct sctp_inpcb *inp,
 				/* free tempy routes */
 				RO_RTFREE(ro);
 			} else {
-				/*
-				 * PMTU check versus smallest asoc MTU goes
-				 * here
-				 */
-				if ((ro->ro_rt != NULL) &&
-				    (net->ro._s_addr)) {
+				if ((ro->ro_rt != NULL) && (net->ro._s_addr) &&
+				    ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0)) {
 					uint32_t mtu;
 
 					mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_rt);
@@ -4635,8 +4631,8 @@ sctp_lowlevel_chunk_output(struct sctp_inpcb *inp,
 					}
 					net->src_addr_selected = 0;
 				}
-				if ((ro->ro_rt != NULL) &&
-				    (net->ro._s_addr)) {
+				if ((ro->ro_rt != NULL) && (net->ro._s_addr) &&
+				    ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0)) {
 					uint32_t mtu;
 
 					mtu = SCTP_GATHER_MTU_FROM_ROUTE(net->ro._s_addr, &net->ro._l_addr.sa, ro->ro_rt);
diff --git a/sys/netinet/sctp_pcb.c b/sys/netinet/sctp_pcb.c
index a60a619..6669d1b 100644
--- a/sys/netinet/sctp_pcb.c
+++ b/sys/netinet/sctp_pcb.c
@@ -2577,6 +2577,7 @@ sctp_inpcb_alloc(struct socket *so, uint32_t vrf_id)
 	/* number of streams to pre-open on a association */
 	m->pre_open_stream_count = SCTP_BASE_SYSCTL(sctp_nr_outgoing_streams_default);
 
+	m->default_mtu = 0;
 	/* Add adaptation cookie */
 	m->adaptation_layer_indicator = 0;
 	m->adaptation_layer_indicator_provided = 0;
@@ -3942,7 +3943,28 @@ sctp_add_remote_addr(struct sctp_tcb *stcb, struct sockaddr *newaddr,
 		    net,
 		    0,
 		    stcb->asoc.vrf_id);
-		if (net->ro._s_addr != NULL) {
+		if (stcb->asoc.default_mtu > 0) {
+			net->mtu = stcb->asoc.default_mtu;
+			switch (net->ro._l_addr.sa.sa_family) {
+#ifdef INET
+			case AF_INET:
+				net->mtu += SCTP_MIN_V4_OVERHEAD;
+				break;
+#endif
+#ifdef INET6
+			case AF_INET6:
+				net->mtu += SCTP_MIN_OVERHEAD;
+				break;
+#endif
+			default:
+				break;
+			}
+#if defined(INET) || defined(INET6)
+			if (net->port) {
+				net->mtu += (uint32_t)sizeof(struct udphdr);
+			}
+#endif
+		} else if (net->ro._s_addr != NULL) {
 			uint32_t imtu, rmtu, hcmtu;
 
 			net->src_addr_selected = 1;
@@ -3966,19 +3988,42 @@ sctp_add_remote_addr(struct sctp_tcb *stcb, struct sockaddr *newaddr,
 		}
 	}
 	if (net->mtu == 0) {
-		switch (newaddr->sa_family) {
+		if (stcb->asoc.default_mtu > 0) {
+			net->mtu = stcb->asoc.default_mtu;
+			switch (net->ro._l_addr.sa.sa_family) {
 #ifdef INET
-		case AF_INET:
-			net->mtu = SCTP_DEFAULT_MTU;
-			break;
+			case AF_INET:
+				net->mtu += SCTP_MIN_V4_OVERHEAD;
+				break;
 #endif
 #ifdef INET6
-		case AF_INET6:
-			net->mtu = 1280;
-			break;
+			case AF_INET6:
+				net->mtu += SCTP_MIN_OVERHEAD;
+				break;
 #endif
-		default:
-			break;
+			default:
+				break;
+			}
+#if defined(INET) || defined(INET6)
+			if (net->port) {
+				net->mtu += (uint32_t)sizeof(struct udphdr);
+			}
+#endif
+		} else {
+			switch (newaddr->sa_family) {
+#ifdef INET
+			case AF_INET:
+				net->mtu = SCTP_DEFAULT_MTU;
+				break;
+#endif
+#ifdef INET6
+			case AF_INET6:
+				net->mtu = 1280;
+				break;
+#endif
+			default:
+				break;
+			}
 		}
 	}
 #if defined(INET) || defined(INET6)
diff --git a/sys/netinet/sctp_pcb.h b/sys/netinet/sctp_pcb.h
index acc6831..8d40b69 100644
--- a/sys/netinet/sctp_pcb.h
+++ b/sys/netinet/sctp_pcb.h
@@ -287,6 +287,7 @@ struct sctp_pcb {
 	sctp_auth_chklist_t *local_auth_chunks;
 	sctp_hmaclist_t *local_hmacs;
 	uint16_t default_keyid;
+	uint32_t default_mtu;
 
 	/* various thresholds */
 	/* Max times I will init at a guy */
diff --git a/sys/netinet/sctp_structs.h b/sys/netinet/sctp_structs.h
index f737174..ed907f1 100644
--- a/sys/netinet/sctp_structs.h
+++ b/sys/netinet/sctp_structs.h
@@ -1098,6 +1098,7 @@ struct sctp_association {
 	uint32_t chunks_on_out_queue;	/* total chunks floating around,
 					 * locked by send socket buffer */
 	uint32_t peers_adaptation;
+	uint32_t default_mtu;
 	uint16_t peer_hmac_id;	/* peer HMAC id to send */
 
 	/*
diff --git a/sys/netinet/sctp_usrreq.c b/sys/netinet/sctp_usrreq.c
index 54e2632..f96816d 100644
--- a/sys/netinet/sctp_usrreq.c
+++ b/sys/netinet/sctp_usrreq.c
@@ -202,6 +202,10 @@ sctp_notify(struct sctp_inpcb *inp,
 #endif
 		/* no need to unlock here, since the TCB is gone */
 	} else if (icmp_code == ICMP_UNREACH_NEEDFRAG) {
+		if ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0) {
+			SCTP_TCB_UNLOCK(stcb);
+			return;
+		}
 		/* Find the next (smaller) MTU */
 		if (next_mtu == 0) {
 			/*
@@ -2447,7 +2451,7 @@ flags_out:
 					 * value
 					 */
 					paddrp->spp_pathmaxrxt = stcb->asoc.def_net_failure;
-					paddrp->spp_pathmtu = 0;
+					paddrp->spp_pathmtu = stcb->asoc.default_mtu;
 					if (stcb->asoc.default_dscp & 0x01) {
 						paddrp->spp_dscp = stcb->asoc.default_dscp & 0xfc;
 						paddrp->spp_flags |= SPP_DSCP;
@@ -2494,8 +2498,7 @@ flags_out:
 						paddrp->spp_flags |= SPP_IPV6_FLOWLABEL;
 					}
 #endif
-					/* can't return this */
-					paddrp->spp_pathmtu = 0;
+					paddrp->spp_pathmtu = inp->sctp_ep.default_mtu;
 
 					if (sctp_is_feature_off(inp, SCTP_PCB_FLAGS_DONOT_HEARTBEAT)) {
 						paddrp->spp_flags |= SPP_HB_ENABLE;
@@ -5479,6 +5482,7 @@ sctp_setopt(struct socket *so, int optname, void *optval, size_t optsize,
 								sctp_pathmtu_adjustment(stcb, net->mtu);
 							}
 						}
+						stcb->asoc.default_mtu = paddrp->spp_pathmtu;
 						sctp_stcb_feature_on(inp, stcb, SCTP_PCB_FLAGS_DO_NOT_PMTUD);
 					}
 					if (paddrp->spp_flags & SPP_PMTUD_ENABLE) {
@@ -5488,6 +5492,7 @@ sctp_setopt(struct socket *so, int optname, void *optval, size_t optsize,
 							}
 							net->dest_state &= ~SCTP_ADDR_NO_PMTUD;
 						}
+						stcb->asoc.default_mtu = 0;
 						sctp_stcb_feature_off(inp, stcb, SCTP_PCB_FLAGS_DO_NOT_PMTUD);
 					}
 					if (paddrp->spp_flags & SPP_DSCP) {
@@ -5544,8 +5549,12 @@ sctp_setopt(struct socket *so, int optname, void *optval, size_t optsize,
 						sctp_feature_on(inp, SCTP_PCB_FLAGS_DONOT_HEARTBEAT);
 					}
 					if (paddrp->spp_flags & SPP_PMTUD_ENABLE) {
+						inp->sctp_ep.default_mtu = 0;
 						sctp_feature_off(inp, SCTP_PCB_FLAGS_DO_NOT_PMTUD);
 					} else if (paddrp->spp_flags & SPP_PMTUD_DISABLE) {
+						if (paddrp->spp_pathmtu >= SCTP_SMALLEST_PMTU) {
+							inp->sctp_ep.default_mtu = paddrp->spp_pathmtu;
+						}
 						sctp_feature_on(inp, SCTP_PCB_FLAGS_DO_NOT_PMTUD);
 					}
 					if (paddrp->spp_flags & SPP_DSCP) {
diff --git a/sys/netinet/sctputil.c b/sys/netinet/sctputil.c
index 97ed5b1..8950ba4 100644
--- a/sys/netinet/sctputil.c
+++ b/sys/netinet/sctputil.c
@@ -1042,6 +1042,7 @@ sctp_init_asoc(struct sctp_inpcb *inp, struct sctp_tcb *stcb,
 	asoc->initial_init_rto_max = inp->sctp_ep.initial_init_rto_max;
 	asoc->initial_rto = inp->sctp_ep.initial_rto;
 
+	asoc->default_mtu = inp->sctp_ep.default_mtu;
 	asoc->max_init_times = inp->sctp_ep.max_init_times;
 	asoc->max_send_times = inp->sctp_ep.max_send_times;
 	asoc->def_net_failure = inp->sctp_ep.def_net_failure;
diff --git a/sys/netinet6/sctp6_usrreq.c b/sys/netinet6/sctp6_usrreq.c
index c88ca96..a61eae1 100644
--- a/sys/netinet6/sctp6_usrreq.c
+++ b/sys/netinet6/sctp6_usrreq.c
@@ -229,6 +229,10 @@ sctp6_notify(struct sctp_inpcb *inp,
 		}
 		break;
 	case ICMP6_PACKET_TOO_BIG:
+		if ((net->dest_state & SCTP_ADDR_NO_PMTUD) == 0) {
+			SCTP_TCB_UNLOCK(stcb);
+			break;
+		}
 		if (SCTP_OS_TIMER_PENDING(&net->pmtu_timer.timer)) {
 			timer_stopped = 1;
 			sctp_timer_stop(SCTP_TIMER_TYPE_PATHMTURAISE, inp, stcb, net,