- Somehow the disable fragment option got lost. We could

   set/clear it but would not do it. Now we will.
-  Moved to latest socket api for extended sndrcv info struct.
-  Moved to support all new levels of fragment interleave.

12 files changed, 150 insertions, 57 deletions

diff --git a/sys/netinet/sctp.h b/sys/netinet/sctp.h
index ab29820..d89560d 100644
--- a/sys/netinet/sctp.h
+++ b/sys/netinet/sctp.h
@@ -223,6 +223,15 @@ struct sctp_paramhdr {
 /* Debug things that need to be purged */
 #define SCTP_SET_INITIAL_DBG_SEQ	0x00009f00
 
+
+/* fragment interleave constants
+ * setting must be one of these or
+ * EINVAL returned.
+ */
+#define SCTP_FRAG_LEVEL_0    0x00000000
+#define SCTP_FRAG_LEVEL_1    0x00000001
+#define SCTP_FRAG_LEVEL_2    0x00000002
+
 /*
  * user state values
  */
diff --git a/sys/netinet/sctp_indata.c b/sys/netinet/sctp_indata.c
index 81d6780..2010d2a 100644
--- a/sys/netinet/sctp_indata.c
+++ b/sys/netinet/sctp_indata.c
@@ -211,6 +211,7 @@ sctp_build_readq_entry(struct sctp_tcb *stcb,
 	read_queue_e->port_from = stcb->rport;
 	read_queue_e->do_not_ref_stcb = 0;
 	read_queue_e->end_added = 0;
+	read_queue_e->some_taken = 0;
 	read_queue_e->pdapi_aborted = 0;
 failed_build:
 	return (read_queue_e);
@@ -249,6 +250,7 @@ sctp_build_readq_entry_chk(struct sctp_tcb *stcb,
 	read_queue_e->spec_flags = 0;
 	read_queue_e->do_not_ref_stcb = 0;
 	read_queue_e->end_added = 0;
+	read_queue_e->some_taken = 0;
 	read_queue_e->pdapi_aborted = 0;
 failed_build:
 	return (read_queue_e);
@@ -5810,6 +5812,8 @@ sctp_handle_forward_tsn(struct sctp_tcb *stcb,
 				 */
 				if ((asoc->fragmented_delivery_inprogress) &&
 				    (chk->rec.data.rcv_flags & SCTP_DATA_FIRST_FRAG)) {
+					uint32_t str_seq;
+
 					/*
 					 * Special case PD-API is up and
 					 * what we fwd-tsn' over includes
@@ -5817,8 +5821,10 @@ sctp_handle_forward_tsn(struct sctp_tcb *stcb,
 					 * longer need to do the PD-API.
 					 */
 					asoc->fragmented_delivery_inprogress = 0;
+
+					str_seq = (asoc->str_of_pdapi << 16) | asoc->ssn_of_pdapi;
 					sctp_ulp_notify(SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION,
-					    stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)NULL);
+					    stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)&str_seq);
 
 				}
 				break;
@@ -5831,8 +5837,11 @@ sctp_handle_forward_tsn(struct sctp_tcb *stcb,
 		 * Ok we removed cnt_gone chunks in the PD-API queue that
 		 * were being delivered. So now we must turn off the flag.
 		 */
+		uint32_t str_seq;
+
+		str_seq = (asoc->str_of_pdapi << 16) | asoc->ssn_of_pdapi;
 		sctp_ulp_notify(SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION,
-		    stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)NULL);
+		    stcb, SCTP_PARTIAL_DELIVERY_ABORTED, (void *)&str_seq);
 		asoc->fragmented_delivery_inprogress = 0;
 	}
 	/*************************************************************/
diff --git a/sys/netinet/sctp_indata.h b/sys/netinet/sctp_indata.h
index 394e4c9..f1a9d41 100644
--- a/sys/netinet/sctp_indata.h
+++ b/sys/netinet/sctp_indata.h
@@ -70,7 +70,7 @@ sctp_build_readq_entry(struct sctp_tcb *stcb,
 		(_ctl)->do_not_ref_stcb = 0; \
 		(_ctl)->end_added = 0; \
 		(_ctl)->pdapi_aborted = 0; \
-		(_ctl)->resv = 0; \
+		(_ctl)->some_taken = 0; \
 	} \
 } while (0)
 
diff --git a/sys/netinet/sctp_input.c b/sys/netinet/sctp_input.c
index 6dee5f4..90de716 100644
--- a/sys/netinet/sctp_input.c
+++ b/sys/netinet/sctp_input.c
@@ -287,14 +287,8 @@ sctp_process_init(struct sctp_init_chunk *cp, struct sctp_tcb *stcb,
 		 * supported) when the INIT-ACK arrives.
 		 */
 		TAILQ_INIT(&asoc->strmin[i].inqueue);
-		/*
-		 * we are not on any wheel, pr-sctp streams will go on the
-		 * wheel when they have data waiting for reorder.
-		 */
-		asoc->strmin[i].next_spoke.tqe_next = 0;
-		asoc->strmin[i].next_spoke.tqe_prev = 0;
+		asoc->strmin[i].delivery_started = 0;
 	}
-
 	/*
 	 * load_address_from_init will put the addresses into the
 	 * association when the COOKIE is processed or the INIT-ACK is
diff --git a/sys/netinet/sctp_output.c b/sys/netinet/sctp_output.c
index 304d299..71ef568 100644
--- a/sys/netinet/sctp_output.c
+++ b/sys/netinet/sctp_output.c
@@ -10879,6 +10879,13 @@ sctp_lower_sosend(struct socket *so,
 		non_blocking = 1;
 	}
 	asoc = &stcb->asoc;
+
+	if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_NO_FRAGMENT)) {
+		if (sndlen > asoc->smallest_mtu) {
+			error = EMSGSIZE;
+			goto out_unlocked;
+		}
+	}
 	/* would we block? */
 	if (non_blocking) {
 		if ((SCTP_SB_LIMIT_SND(so) <
diff --git a/sys/netinet/sctp_pcb.c b/sys/netinet/sctp_pcb.c
index 24c04b3..376d511 100644
--- a/sys/netinet/sctp_pcb.c
+++ b/sys/netinet/sctp_pcb.c
@@ -3606,9 +3606,12 @@ sctp_free_assoc(struct sctp_inpcb *inp, struct sctp_tcb *stcb, int from_inpcbfre
 						 * added right after this
 						 * msg.
 						 */
+						uint32_t strseq;
+
 						stcb->asoc.control_pdapi = sq;
+						strseq = (sq->sinfo_stream << 16) | sq->sinfo_ssn;
 						sctp_notify_partial_delivery_indication(stcb,
-						    SCTP_PARTIAL_DELIVERY_ABORTED, 1);
+						    SCTP_PARTIAL_DELIVERY_ABORTED, 1, strseq);
 						stcb->asoc.control_pdapi = NULL;
 					}
 				}
diff --git a/sys/netinet/sctp_pcb.h b/sys/netinet/sctp_pcb.h
index 022a359..1c6b523 100644
--- a/sys/netinet/sctp_pcb.h
+++ b/sys/netinet/sctp_pcb.h
@@ -84,9 +84,10 @@ TAILQ_HEAD(sctp_streamhead, sctp_stream_queue_pending);
 /*
  * PCB Features (in sctp_features bitmask)
  */
-#define SCTP_PCB_FLAGS_EXT_RCVINFO      0x00000004
-#define SCTP_PCB_FLAGS_DONOT_HEARTBEAT  0x00000008
-#define SCTP_PCB_FLAGS_FRAG_INTERLEAVE  0x00000010
+#define SCTP_PCB_FLAGS_EXT_RCVINFO      0x00000002
+#define SCTP_PCB_FLAGS_DONOT_HEARTBEAT  0x00000004
+#define SCTP_PCB_FLAGS_FRAG_INTERLEAVE  0x00000008
+#define SCTP_PCB_FLAGS_INTERLEAVE_STRMS	0x00000010
 #define SCTP_PCB_FLAGS_DO_ASCONF	0x00000020
 #define SCTP_PCB_FLAGS_AUTO_ASCONF	0x00000040
 /* socket options */
@@ -103,7 +104,7 @@ TAILQ_HEAD(sctp_streamhead, sctp_stream_queue_pending);
 #define SCTP_PCB_FLAGS_AUTHEVNT		0x00040000
 #define SCTP_PCB_FLAGS_STREAM_RESETEVNT 0x00080000
 #define SCTP_PCB_FLAGS_NO_FRAGMENT	0x00100000
-#define SCTP_PCB_FLAGS_EXPLICIT_EOR     0x00200000
+#define SCTP_PCB_FLAGS_EXPLICIT_EOR     0x00400000
 
 
 #define SCTP_PCBHASH_ALLADDR(port, mask) (port & mask)
diff --git a/sys/netinet/sctp_structs.h b/sys/netinet/sctp_structs.h
index 4e88c1f..fc6588b 100644
--- a/sys/netinet/sctp_structs.h
+++ b/sys/netinet/sctp_structs.h
@@ -375,7 +375,7 @@ struct sctp_queued_to_read {	/* sinfo structure Pluse more */
 	uint8_t do_not_ref_stcb;
 	uint8_t end_added;
 	uint8_t pdapi_aborted;
-	uint8_t resv;
+	uint8_t some_taken;
 };
 
 /* This data structure will be on the outbound
@@ -425,9 +425,9 @@ struct sctp_stream_queue_pending {
 TAILQ_HEAD(sctpwheelunrel_listhead, sctp_stream_in);
 struct sctp_stream_in {
 	struct sctp_readhead inqueue;
-	              TAILQ_ENTRY(sctp_stream_in) next_spoke;
 	uint16_t stream_no;
 	uint16_t last_sequence_delivered;	/* used for re-order */
+	uint8_t delivery_started;
 };
 
 /* This struct is used to track the traffic on outbound streams */
diff --git a/sys/netinet/sctp_uio.h b/sys/netinet/sctp_uio.h
index 263f417..1e54e9c 100644
--- a/sys/netinet/sctp_uio.h
+++ b/sys/netinet/sctp_uio.h
@@ -105,20 +105,12 @@ struct sctp_sndrcvinfo {
 };
 
 struct sctp_extrcvinfo {
-	uint16_t sinfo_stream;
-	uint16_t sinfo_ssn;
-	uint16_t sinfo_flags;
-	uint32_t sinfo_ppid;
-	uint32_t sinfo_context;
-	uint32_t sinfo_timetolive;
-	uint32_t sinfo_tsn;
-	uint32_t sinfo_cumtsn;
-	sctp_assoc_t sinfo_assoc_id;
-	uint16_t next_flags;
-	uint16_t next_stream;
-	uint32_t next_asocid;
-	uint32_t next_length;
-	uint32_t next_ppid;
+	struct sctp_sndrcvinfo sreinfo_sinfo;
+	uint16_t sreinfo_next_flags;
+	uint16_t sreinfo_next_stream;
+	uint32_t sreinfo_next_aid;
+	uint32_t sreinfo_next_length;
+	uint32_t sreinfo_next_ppid;
 	uint8_t __reserve_pad[SCTP_ALIGN_RESV_PAD_SHORT];
 };
 
@@ -318,6 +310,8 @@ struct sctp_pdapi_event {
 	uint16_t pdapi_flags;
 	uint32_t pdapi_length;
 	uint32_t pdapi_indication;
+	uint16_t pdapi_stream;
+	uint16_t pdapi_seq;
 	sctp_assoc_t pdapi_assoc_id;
 };
 
diff --git a/sys/netinet/sctp_usrreq.c b/sys/netinet/sctp_usrreq.c
index 7f9bdf3..b9797a2 100644
--- a/sys/netinet/sctp_usrreq.c
+++ b/sys/netinet/sctp_usrreq.c
@@ -1516,7 +1516,15 @@ sctp_getopt(struct socket *so, int optname, void *optval, size_t *optsize,
 			uint32_t *value;
 
 			SCTP_CHECK_AND_CAST(value, optval, uint32_t, *optsize);
-			*value = sctp_is_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE);
+			if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE)) {
+				if (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS)) {
+					*value = SCTP_FRAG_LEVEL_2;
+				} else {
+					*value = SCTP_FRAG_LEVEL_1;
+				}
+			} else {
+				*value = SCTP_FRAG_LEVEL_0;
+			}
 			*optsize = sizeof(uint32_t);
 		}
 		break;
@@ -2457,13 +2465,21 @@ sctp_setopt(struct socket *so, int optname, void *optval, size_t optsize,
 	case SCTP_FRAGMENT_INTERLEAVE:
 		/* not yet until we re-write sctp_recvmsg() */
 		{
-			uint32_t *on_off;
+			uint32_t *level;
 
-			SCTP_CHECK_AND_CAST(on_off, optval, uint32_t, optsize);
-			if (*on_off) {
+			SCTP_CHECK_AND_CAST(level, optval, uint32_t, optsize);
+			if (*level == SCTP_FRAG_LEVEL_2) {
 				sctp_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE);
+				sctp_feature_on(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS);
+			} else if (*level == SCTP_FRAG_LEVEL_1) {
+				sctp_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE);
+				sctp_feature_off(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS);
+			} else if (*level == SCTP_FRAG_LEVEL_0) {
+				sctp_feature_on(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE);
+				sctp_feature_off(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS);
+
 			} else {
-				sctp_feature_off(inp, SCTP_PCB_FLAGS_FRAG_INTERLEAVE);
+				error = EINVAL;
 			}
 		}
 		break;
diff --git a/sys/netinet/sctputil.c b/sys/netinet/sctputil.c
index c94b2d4..9d03b7c 100644
--- a/sys/netinet/sctputil.c
+++ b/sys/netinet/sctputil.c
@@ -3002,7 +3002,7 @@ sctp_notify_adaptation_layer(struct sctp_tcb *stcb,
 /* This always must be called with the read-queue LOCKED in the INP */
 void
 sctp_notify_partial_delivery_indication(struct sctp_tcb *stcb,
-    uint32_t error, int nolock)
+    uint32_t error, int nolock, uint32_t val)
 {
 	struct mbuf *m_notify;
 	struct sctp_pdapi_event *pdapi;
@@ -3023,6 +3023,8 @@ sctp_notify_partial_delivery_indication(struct sctp_tcb *stcb,
 	pdapi->pdapi_flags = 0;
 	pdapi->pdapi_length = sizeof(struct sctp_pdapi_event);
 	pdapi->pdapi_indication = error;
+	pdapi->pdapi_stream = (val >> 16);
+	pdapi->pdapi_seq = (val & 0x0000ffff);
 	pdapi->pdapi_assoc_id = sctp_get_associd(stcb);
 
 	SCTP_BUF_LEN(m_notify) = sizeof(struct sctp_pdapi_event);
@@ -3265,7 +3267,13 @@ sctp_ulp_notify(uint32_t notification, struct sctp_tcb *stcb,
 		sctp_notify_adaptation_layer(stcb, error);
 		break;
 	case SCTP_NOTIFY_PARTIAL_DELVIERY_INDICATION:
-		sctp_notify_partial_delivery_indication(stcb, error, 0);
+		{
+			uint32_t val;
+
+			val = *((uint32_t *) data);
+
+			sctp_notify_partial_delivery_indication(stcb, error, 0, val);
+		}
 		break;
 	case SCTP_NOTIFY_STRDATA_ERR:
 		break;
@@ -4803,8 +4811,31 @@ restart_nosblocks:
 			/* find a more suitable one then this */
 			ctl = TAILQ_NEXT(control, next);
 			while (ctl) {
-				if ((ctl->stcb != control->stcb) && (ctl->length)) {
-					/* found one */
+				if ((ctl->stcb != control->stcb) && (ctl->length) &&
+				    (ctl->some_taken ||
+				    ((ctl->do_not_ref_stcb == 0) &&
+				    (ctl->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started == 0)))
+				    ) {
+					/*-
+					 * If we have a different TCB next, and there is data
+					 * present. If we have already taken some (pdapi), OR we can
+					 * ref the tcb and no delivery as started on this stream, we
+					 * take it.
+					 */
+					control = ctl;
+					goto found_one;
+				} else if ((sctp_is_feature_on(inp, SCTP_PCB_FLAGS_INTERLEAVE_STRMS)) &&
+					    (ctl->length) &&
+					    ((ctl->some_taken) ||
+					    ((ctl->do_not_ref_stcb == 0) &&
+					    (ctl->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started == 0)))
+				    ) {
+					/*-
+					 * If we have the same tcb, and there is data present, and we
+					 * have the strm interleave feature present. Then if we have
+					 * taken some (pdapi) or we can refer to tht tcb AND we have
+					 * not started a delivery for this stream, we can take it.
+					 */
 					control = ctl;
 					goto found_one;
 				}
@@ -4831,6 +4862,10 @@ found_one:
 	 * If we reach here, control has a some data for us to read off.
 	 * Note that stcb COULD be NULL.
 	 */
+	if (control->do_not_ref_stcb == 0) {
+		control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 1;
+	}
+	control->some_taken = 1;
 	if (hold_sblock) {
 		SOCKBUF_UNLOCK(&so->so_rcv);
 		hold_sblock = 0;
@@ -4874,21 +4909,22 @@ found_one:
 			struct sctp_extrcvinfo *s_extra;
 
 			s_extra = (struct sctp_extrcvinfo *)sinfo;
-			if (nxt) {
-				s_extra->next_flags = SCTP_NEXT_MSG_AVAIL;
+			if ((nxt) &&
+			    (nxt->length)) {
+				s_extra->sreinfo_next_flags = SCTP_NEXT_MSG_AVAIL;
 				if (nxt->sinfo_flags & SCTP_UNORDERED) {
-					s_extra->next_flags |= SCTP_NEXT_MSG_IS_UNORDERED;
+					s_extra->sreinfo_next_flags |= SCTP_NEXT_MSG_IS_UNORDERED;
 				}
 				if (nxt->spec_flags & M_NOTIFICATION) {
-					s_extra->next_flags |= SCTP_NEXT_MSG_IS_NOTIFICATION;
+					s_extra->sreinfo_next_flags |= SCTP_NEXT_MSG_IS_NOTIFICATION;
 				}
-				s_extra->next_asocid = nxt->sinfo_assoc_id;
-				s_extra->next_length = nxt->length;
-				s_extra->next_ppid = nxt->sinfo_ppid;
-				s_extra->next_stream = nxt->sinfo_stream;
+				s_extra->sreinfo_next_aid = nxt->sinfo_assoc_id;
+				s_extra->sreinfo_next_length = nxt->length;
+				s_extra->sreinfo_next_ppid = nxt->sinfo_ppid;
+				s_extra->sreinfo_next_stream = nxt->sinfo_stream;
 				if (nxt->tail_mbuf != NULL) {
 					if (nxt->end_added) {
-						s_extra->next_flags |= SCTP_NEXT_MSG_ISCOMPLETE;
+						s_extra->sreinfo_next_flags |= SCTP_NEXT_MSG_ISCOMPLETE;
 					}
 				}
 			} else {
@@ -4898,11 +4934,12 @@ found_one:
 				 * sinfo_ that is on the control's structure
 				 * :-D
 				 */
-				s_extra->next_flags = SCTP_NO_NEXT_MSG;
-				s_extra->next_asocid = 0;
-				s_extra->next_length = 0;
-				s_extra->next_ppid = 0;
-				s_extra->next_stream = 0;
+				nxt = NULL;
+				s_extra->sreinfo_next_flags = SCTP_NO_NEXT_MSG;
+				s_extra->sreinfo_next_aid = 0;
+				s_extra->sreinfo_next_length = 0;
+				s_extra->sreinfo_next_ppid = 0;
+				s_extra->sreinfo_next_stream = 0;
 			}
 		}
 		/*
@@ -5023,6 +5060,8 @@ get_more_data:
 				if ((SCTP_BUF_NEXT(m) == NULL) &&
 				    (control->end_added)) {
 					out_flags |= MSG_EOR;
+					if (control->do_not_ref_stcb == 0)
+						control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 0;
 				}
 				if (control->spec_flags & M_NOTIFICATION) {
 					out_flags |= MSG_NOTIFICATION;
@@ -5293,8 +5332,15 @@ wait_some_more:
 			if (control->end_added == 1) {
 				/* he aborted, or is done i.e.did a shutdown */
 				out_flags |= MSG_EOR;
-				if (control->pdapi_aborted)
+				if (control->pdapi_aborted) {
+					if (control->do_not_ref_stcb == 0)
+						control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 0;
+
 					out_flags |= MSG_TRUNC;
+				} else {
+					if (control->do_not_ref_stcb == 0)
+						control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 0;
+				}
 				goto done_with_control;
 			}
 			if (so->so_rcv.sb_cc > held_length) {
@@ -5343,6 +5389,8 @@ get_more_data2:
 			}
 			if (control->end_added) {
 				out_flags |= MSG_EOR;
+				if (control->do_not_ref_stcb == 0)
+					control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 0;
 			}
 			if (control->spec_flags & M_NOTIFICATION) {
 				out_flags |= MSG_NOTIFICATION;
@@ -5405,8 +5453,11 @@ get_more_data2:
 					 * shutdown
 					 */
 					out_flags |= MSG_EOR;
-					if (control->pdapi_aborted)
+					if (control->pdapi_aborted) {
 						out_flags |= MSG_TRUNC;
+						if (control->do_not_ref_stcb == 0)
+							control->stcb->asoc.strmin[ctl->sinfo_stream].delivery_started = 0;
+					}
 					goto done_with_control;
 				}
 				if (so->so_rcv.sb_cc > held_length) {
@@ -5535,6 +5586,15 @@ release_unlocked:
 	if (msg_flags)
 		*msg_flags |= out_flags;
 out:
+	if (((out_flags & MSG_EOR) == 0) &&
+	    ((in_flags & MSG_PEEK) == 0) &&
+	    (sinfo) &&
+	    (sctp_is_feature_on(inp, SCTP_PCB_FLAGS_EXT_RCVINFO))) {
+		struct sctp_extrcvinfo *s_extra;
+
+		s_extra = (struct sctp_extrcvinfo *)sinfo;
+		s_extra->sreinfo_next_flags = SCTP_NO_NEXT_MSG;
+	}
 	if (hold_rlock == 1) {
 		SCTP_INP_READ_UNLOCK(inp);
 		hold_rlock = 0;
diff --git a/sys/netinet/sctputil.h b/sys/netinet/sctputil.h
index 8496f42..e0ae668 100644
--- a/sys/netinet/sctputil.h
+++ b/sys/netinet/sctputil.h
@@ -242,7 +242,7 @@ void sctp_print_address_pkt(struct ip *, struct sctphdr *);
 
 void
 sctp_notify_partial_delivery_indication(struct sctp_tcb *stcb,
-    uint32_t error, int no_lock);
+    uint32_t error, int no_lock, uint32_t strseq);
 
 int
 sctp_release_pr_sctp_chunk(struct sctp_tcb *, struct sctp_tmit_chunk *,