* Implement initial version of send buffer splitting.

* Make send/recv buffer splitting switchable via sysctl.
* While there: Fix some comments.

4 files changed, 36 insertions, 6 deletions

diff --git a/sys/netinet/sctp_constants.h b/sys/netinet/sctp_constants.h
index 078981d..212a2f9 100644
--- a/sys/netinet/sctp_constants.h
+++ b/sys/netinet/sctp_constants.h
@@ -948,6 +948,9 @@ __FBSDID("$FreeBSD$");
  */
 #define SCTP_TIME_WAIT 60
 
+#define SCTP_SEND_BUFFER_SPLITTING 0x00000001
+#define SCTP_RECV_BUFFER_SPLITTING 0x00000002
+
 /* The system retains a cache of free chunks such to
  * cut down on calls the memory allocation system. There
  * is a per association limit of free items and a overall
diff --git a/sys/netinet/sctp_output.c b/sys/netinet/sctp_output.c
index 4d0eb02..5623228 100644
--- a/sys/netinet/sctp_output.c
+++ b/sys/netinet/sctp_output.c
@@ -7407,7 +7407,7 @@ sctp_med_chunk_output(struct sctp_inpcb *inp,
 	/* temp arrays for unlinking */
 	struct sctp_tmit_chunk *data_list[SCTP_MAX_DATA_BUNDLING];
 	int no_fragmentflg, error;
-	unsigned int max_rwnd_per_dest;
+	unsigned int max_rwnd_per_dest, max_send_per_dest;
 	int one_chunk, hbflag, skip_data_for_this_net;
 	int asconf, cookie, no_out_cnt;
 	int bundle_at, ctl_cnt, no_data_chunks, eeor_mode;
@@ -7469,6 +7469,10 @@ sctp_med_chunk_output(struct sctp_inpcb *inp,
 		}
 	}
 	max_rwnd_per_dest = ((asoc->peers_rwnd + asoc->total_flight) / asoc->numnets);
+	if (stcb->sctp_socket)
+		max_send_per_dest = SCTP_SB_LIMIT_SND(stcb->sctp_socket) / asoc->numnets;
+	else
+		max_send_per_dest = 0;
 	if ((no_data_chunks == 0) && (!TAILQ_EMPTY(&asoc->out_wheel))) {
 		TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
 			/*
@@ -8039,9 +8043,22 @@ again_one_more_time:
 			goto no_data_fill;
 		}
 		if ((asoc->sctp_cmt_on_off == 1) &&
+		    (SCTP_BASE_SYSCTL(sctp_buffer_splitting) & SCTP_RECV_BUFFER_SPLITTING) &&
 		    (net->flight_size > max_rwnd_per_dest)) {
 			goto no_data_fill;
 		}
+		/*
+		 * We need a specific accounting for the usage of the send
+		 * buffer. We also need to check the number of messages per
+		 * net. For now, this is better than nothing and it disabled
+		 * by default...
+		 */
+		if ((asoc->sctp_cmt_on_off == 1) &&
+		    (SCTP_BASE_SYSCTL(sctp_buffer_splitting) & SCTP_SEND_BUFFER_SPLITTING) &&
+		    (max_send_per_dest > 0) &&
+		    (net->flight_size > max_send_per_dest)) {
+			goto no_data_fill;
+		}
 		/*********************/
 		/* Data transmission */
 		/*********************/
diff --git a/sys/netinet/sctp_sysctl.c b/sys/netinet/sctp_sysctl.c
index 50fbafd..ef1356c 100644
--- a/sys/netinet/sctp_sysctl.c
+++ b/sys/netinet/sctp_sysctl.c
@@ -109,6 +109,7 @@ sctp_init_sysctls()
 	SCTP_BASE_SYSCTL(sctp_mobility_base) = SCTPCTL_MOBILITY_BASE_DEFAULT;
 	SCTP_BASE_SYSCTL(sctp_mobility_fasthandoff) = SCTPCTL_MOBILITY_FASTHANDOFF_DEFAULT;
 	SCTP_BASE_SYSCTL(sctp_vtag_time_wait) = SCTPCTL_TIME_WAIT_DEFAULT;
+	SCTP_BASE_SYSCTL(sctp_buffer_splitting) = SCTPCTL_BUFFER_SPLITTING_DEFAULT;
 #if defined(SCTP_LOCAL_TRACE_BUF)
 	memset(&SCTP_BASE_SYSCTL(sctp_log), 0, sizeof(struct sctp_log));
 #endif
@@ -620,7 +621,7 @@ sysctl_sctp_check(SYSCTL_HANDLER_ARGS)
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_default_cc_module), SCTPCTL_DEFAULT_CC_MODULE_MIN, SCTPCTL_DEFAULT_CC_MODULE_MAX);
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_default_frag_interleave), SCTPCTL_DEFAULT_FRAG_INTERLEAVE_MIN, SCTPCTL_DEFAULT_FRAG_INTERLEAVE_MAX);
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_vtag_time_wait), SCTPCTL_TIME_WAIT_MIN, SCTPCTL_TIME_WAIT_MAX);
-
+		RANGECHK(SCTP_BASE_SYSCTL(sctp_buffer_splitting), SCTPCTL_BUFFER_SPLITTING_MIN, SCTPCTL_BUFFER_SPLITTING_MAX);
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_mobility_base), SCTPCTL_MOBILITY_BASE_MIN, SCTPCTL_MOBILITY_BASE_MAX);
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_mobility_fasthandoff), SCTPCTL_MOBILITY_FASTHANDOFF_MIN, SCTPCTL_MOBILITY_FASTHANDOFF_MAX);
 		RANGECHK(SCTP_BASE_SYSCTL(sctp_udp_tunneling_for_client_enable), SCTPCTL_UDP_TUNNELING_FOR_CLIENT_ENABLE_MIN, SCTPCTL_UDP_TUNNELING_FOR_CLIENT_ENABLE_MAX);
@@ -1067,6 +1068,10 @@ SYSCTL_PROC(_net_inet_sctp, OID_AUTO, vtag_time_wait, CTLTYPE_INT | CTLFLAG_RW,
     &SCTP_BASE_SYSCTL(sctp_vtag_time_wait), 0, sysctl_sctp_check, "IU",
     SCTPCTL_TIME_WAIT_DESC);
 
+SYSCTL_PROC(_net_inet_sctp, OID_AUTO, buffer_splitting, CTLTYPE_INT | CTLFLAG_RW,
+    &SCTP_BASE_SYSCTL(sctp_buffer_splitting), 0, sysctl_sctp_check, "IU",
+    SCTPCTL_BUFFER_SPLITTING_DESC);
+
 #ifdef SCTP_DEBUG
 SYSCTL_PROC(_net_inet_sctp, OID_AUTO, debug, CTLTYPE_INT | CTLFLAG_RW,
     &SCTP_BASE_SYSCTL(sctp_debug_on), 0, sysctl_sctp_check, "IU",
diff --git a/sys/netinet/sctp_sysctl.h b/sys/netinet/sctp_sysctl.h
index 1140b97..e52f40d 100644
--- a/sys/netinet/sctp_sysctl.h
+++ b/sys/netinet/sctp_sysctl.h
@@ -107,6 +107,7 @@ struct sctp_sysctl {
 	uint32_t sctp_udp_tunneling_port;
 	uint32_t sctp_enable_sack_immediately;
 	uint32_t sctp_vtag_time_wait;
+	uint32_t sctp_buffer_splitting;
 #if defined(SCTP_DEBUG)
 	uint32_t sctp_debug_on;
 #endif
@@ -478,19 +479,23 @@ struct sctp_sysctl {
 #define SCTPCTL_SACK_IMMEDIATELY_ENABLE_MAX	1
 #define SCTPCTL_SACK_IMMEDIATELY_ENABLE_DEFAULT	SCTPCTL_SACK_IMMEDIATELY_ENABLE_MIN
 
-/* Enable sending of the SACK-IMMEDIATELY bit */
+/* Enable sending of the NAT-FRIENDLY message */
 #define SCTPCTL_NAT_FRIENDLY_INITS_DESC	"Enable sending of the nat-friendly SCTP option on INITs."
 #define SCTPCTL_NAT_FRIENDLY_INITS_MIN	0
 #define SCTPCTL_NAT_FRIENDLY_INITS_MAX	1
 #define SCTPCTL_NAT_FRIENDLY_INITS_DEFAULT	SCTPCTL_NAT_FRIENDLY_INITS_MIN
 
-
-/* Vtag tiem wait bits */
-#define SCTPCTL_TIME_WAIT_DESC	"Vtag time wait time 0 disables."
+/* Vtag time wait in seconds */
+#define SCTPCTL_TIME_WAIT_DESC	"Vtag time wait time in seconds, 0 disables it."
 #define SCTPCTL_TIME_WAIT_MIN	0
 #define SCTPCTL_TIME_WAIT_MAX	0xffffffff
 #define SCTPCTL_TIME_WAIT_DEFAULT	SCTP_TIME_WAIT
 
+/* Enable Send/Receive buffer splitting */
+#define SCTPCTL_BUFFER_SPLITTING_DESC		"Enable send/receive buffer splitting."
+#define SCTPCTL_BUFFER_SPLITTING_MIN		0
+#define SCTPCTL_BUFFER_SPLITTING_MAX		0x3
+#define SCTPCTL_BUFFER_SPLITTING_DEFAULT	SCTPCTL_BUFFER_SPLITTING_MIN
 
 #if defined(SCTP_DEBUG)
 /* debug: Configure debug output */