Improve the scalability by using the local and remote port when

putting inps in the tcpephash.

MFC after: 3 days.

1 files changed, 33 insertions, 30 deletions

diff --git a/sys/netinet/sctp_pcb.c b/sys/netinet/sctp_pcb.c
index c5967ef..7e305c4 100644
--- a/sys/netinet/sctp_pcb.c
+++ b/sys/netinet/sctp_pcb.c
@@ -567,8 +567,8 @@ sctp_add_addr_to_vrf(uint32_t vrf_id, void *ifn, uint32_t ifn_index,
 		} else {
 			if (sctp_ifap->ifn_p) {
 				/*
-				 * The last IFN gets the address, remove
-				 * the old one
+				 * The last IFN gets the address, remove the
+				 * old one
 				 */
 				SCTPDBG(SCTP_DEBUG_PCB4, "Moving ifa %p from %s (0x%x) to %s (0x%x)\n",
 				    sctp_ifap, sctp_ifap->ifn_p->ifn_name,
@@ -849,7 +849,7 @@ sctp_tcb_special_locate(struct sctp_inpcb **inp_p, struct sockaddr *from,
 	} else {
 		return NULL;
 	}
-	ephead = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport), SCTP_BASE_INFO(hashtcpmark))];
+	ephead = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport | rport), SCTP_BASE_INFO(hashtcpmark))];
 	/*
 	 * Ok now for each of the guys in this bucket we must look and see:
 	 * - Does the remote port match. - Does there single association's
@@ -1771,6 +1771,7 @@ sctp_pcb_findep(struct sockaddr *nam, int find_tcp_pool, int have_lock,
 	struct sockaddr_in *sin;
 	struct sockaddr_in6 *sin6;
 	int lport;
+	unsigned int i;
 
 	if (nam->sa_family == AF_INET) {
 		sin = (struct sockaddr_in *)nam;
@@ -1797,17 +1798,22 @@ sctp_pcb_findep(struct sockaddr *nam, int find_tcp_pool, int have_lock,
 
 	/*
 	 * If the TCP model exists it could be that the main listening
-	 * endpoint is gone but there exists a connected socket for this guy
-	 * yet. If so we can return the first one that we find. This may NOT
-	 * be the correct one so the caller should be wary on the return
-	 * INP. Currently the only caller that sets this flag is in bindx
-	 * where we are verifying that a user CAN bind the address. He
-	 * either has bound it already, or someone else has, or its open to
-	 * bind, so this is good enough.
+	 * endpoint is gone but there still exists a connected socket for
+	 * this guy. If so we can return the first one that we find. This
+	 * may NOT be the correct one so the caller should be wary on the
+	 * returned INP. Currently the only caller that sets find_tcp_pool
+	 * is in bindx where we are verifying that a user CAN bind the
+	 * address. He either has bound it already, or someone else has, or
+	 * its open to bind, so this is good enough.
 	 */
 	if (inp == NULL && find_tcp_pool) {
-		head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashtcpmark))];
-		inp = sctp_endpoint_probe(nam, head, lport, vrf_id);
+		for (i = 0; i < SCTP_BASE_INFO(hashtcpmark) + 1; i++) {
+			head = &SCTP_BASE_INFO(sctp_tcpephash)[i];
+			inp = sctp_endpoint_probe(nam, head, lport, vrf_id);
+			if (inp) {
+				break;
+			}
+		}
 	}
 	if (inp) {
 		SCTP_INP_INCR_REF(inp);
@@ -2603,8 +2609,7 @@ sctp_move_pcb_and_assoc(struct sctp_inpcb *old_inp, struct sctp_inpcb *new_inp,
 		LIST_REMOVE(stcb, sctp_tcbasocidhash);
 	}
 	/* Now insert the new_inp into the TCP connected hash */
-	head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport),
-	    SCTP_BASE_INFO(hashtcpmark))];
+	head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport | rport), SCTP_BASE_INFO(hashtcpmark))];
 
 	LIST_INSERT_HEAD(head, new_inp, sctp_hash);
 	/* Its safe to access */
@@ -3089,12 +3094,10 @@ continue_anyway:
 	/* find the bucket */
 	if (port_reuse_active) {
 		/* Put it into tcp 1-2-1 hash */
-		head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR((lport),
-		    SCTP_BASE_INFO(hashtcpmark))];
+		head = &SCTP_BASE_INFO(sctp_tcpephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashtcpmark))];
 		inp->sctp_flags |= SCTP_PCB_FLAGS_IN_TCPPOOL;
 	} else {
-		head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(lport,
-		    SCTP_BASE_INFO(hashmark))];
+		head = &SCTP_BASE_INFO(sctp_ephash)[SCTP_PCBHASH_ALLADDR(lport, SCTP_BASE_INFO(hashmark))];
 	}
 	/* put it in the bucket */
 	LIST_INSERT_HEAD(head, inp, sctp_hash);
@@ -3716,8 +3719,8 @@ sctp_add_remote_addr(struct sctp_tcb *stcb, struct sockaddr *newaddr,
 {
 	/*
 	 * The following is redundant to the same lines in the
-	 * sctp_aloc_assoc() but is needed since others call the add
-	 * address function
+	 * sctp_aloc_assoc() but is needed since others call the add address
+	 * function
 	 */
 	struct sctp_nets *net, *netfirst;
 	int addr_inscope;
@@ -5159,10 +5162,10 @@ sctp_free_assoc(struct sctp_inpcb *inp, struct sctp_tcb *stcb, int from_inpcbfre
 			SCTP_INP_RUNLOCK(inp);
 			/*
 			 * This will start the kill timer (if we are the
-			 * last one) since we hold an increment yet. But this
-			 * is the only safe way to do this since otherwise
-			 * if the socket closes at the same time we are here
-			 * we might collide in the cleanup.
+			 * last one) since we hold an increment yet. But
+			 * this is the only safe way to do this since
+			 * otherwise if the socket closes at the same time
+			 * we are here we might collide in the cleanup.
 			 */
 			sctp_inpcb_free(inp,
 			    SCTP_FREE_SHOULD_USE_GRACEFUL_CLOSE,
@@ -6676,12 +6679,12 @@ sctp_drain_mbufs(struct sctp_inpcb *inp, struct sctp_tcb *stcb)
 	}
 	/*
 	 * Another issue, in un-setting the TSN's in the mapping array we
-	 * DID NOT adjust the highest_tsn marker.  This will cause one of two
-	 * things to occur. It may cause us to do extra work in checking for
-	 * our mapping array movement. More importantly it may cause us to
-	 * SACK every datagram. This may not be a bad thing though since we
-	 * will recover once we get our cum-ack above and all this stuff we
-	 * dumped recovered.
+	 * DID NOT adjust the highest_tsn marker.  This will cause one of
+	 * two things to occur. It may cause us to do extra work in checking
+	 * for our mapping array movement. More importantly it may cause us
+	 * to SACK every datagram. This may not be a bad thing though since
+	 * we will recover once we get our cum-ack above and all this stuff
+	 * we dumped recovered.
 	 */
 }