1361 Add support for socket options TCP_KEEPCNT, TCP_KEEPIDLE, TCP_KEEPINTVL

Reviewed by: Pavan <pavan.tc@gmail.com>
Reviewed by: Dan McDonald <danmcd at nexenta.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Approved by: Garrett D'Amore <garrett@damore.org>

5 files changed, 133 insertions, 2 deletions

diff --git a/usr/src/uts/common/inet/tcp.h b/usr/src/uts/common/inet/tcp.h
index 460f183884..d95f2559c6 100644
--- a/usr/src/uts/common/inet/tcp.h
+++ b/usr/src/uts/common/inet/tcp.h
@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 /* Copyright (c) 1990 Mentat Inc. */
 
@@ -334,11 +335,26 @@ typedef struct tcp_s {
 	} tcp_conn;
 	uint32_t tcp_syn_rcvd_timeout;	/* How many SYN_RCVD timeout in q0 */
 
-	/* TCP Keepalive Timer members */
+	/*
+	 * TCP Keepalive Timer members.
+	 * All keepalive timer intervals are in milliseconds.
+	 */
 	int32_t	tcp_ka_last_intrvl;	/* Last probe interval */
 	timeout_id_t tcp_ka_tid;	/* Keepalive timer ID */
 	uint32_t tcp_ka_interval;	/* Keepalive interval */
+
+	/*
+	 * TCP connection is terminated if we don't hear back from the peer
+	 * for tcp_ka_abort_thres milliseconds after the first keepalive probe.
+	 * tcp_ka_rinterval is the interval in milliseconds between successive
+	 * keepalive probes. tcp_ka_cnt is the number of keepalive probes to
+	 * be sent before terminating the connection, if we don't hear back from
+	 * peer.
+	 * tcp_ka_abort_thres = tcp_ka_rinterval * tcp_ka_cnt
+	 */
+	uint32_t tcp_ka_rinterval;	/* keepalive retransmit interval */
 	uint32_t tcp_ka_abort_thres;	/* Keepalive abort threshold */
+	uint32_t tcp_ka_cnt;		/* count of keepalive probes */
 
 	int32_t	tcp_client_errno;	/* How the client screwed up */
 
diff --git a/usr/src/uts/common/inet/tcp/tcp.c b/usr/src/uts/common/inet/tcp/tcp.c
index 8d3dacf35b..0734468ea0 100644
--- a/usr/src/uts/common/inet/tcp/tcp.c
+++ b/usr/src/uts/common/inet/tcp/tcp.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, Joyent Inc. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 /* Copyright (c) 1990 Mentat Inc. */
 
@@ -2354,6 +2355,8 @@ tcp_init_values(tcp_t *tcp, tcp_t *parent)
 
 		tcp->tcp_ka_interval = tcps->tcps_keepalive_interval;
 		tcp->tcp_ka_abort_thres = tcps->tcps_keepalive_abort_interval;
+		tcp->tcp_ka_cnt = 0;
+		tcp->tcp_ka_rinterval = 0;
 
 		/*
 		 * Default value of tcp_init_cwnd is 0, so no need to set here
diff --git a/usr/src/uts/common/inet/tcp/tcp_opt_data.c b/usr/src/uts/common/inet/tcp/tcp_opt_data.c
index c1614463c2..960c3d8902 100644
--- a/usr/src/uts/common/inet/tcp/tcp_opt_data.c
+++ b/usr/src/uts/common/inet/tcp/tcp_opt_data.c
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -117,6 +118,12 @@ opdes_t	tcp_opt_arr[] = {
 { TCP_KEEPALIVE_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0,
 	sizeof (int), 0	},
 
+{ TCP_KEEPIDLE, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },
+
+{ TCP_KEEPCNT, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },
+
+{ TCP_KEEPINTVL, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },
+
 { TCP_KEEPALIVE_ABORT_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0,
 	sizeof (int), 0	},
 
@@ -403,6 +410,25 @@ tcp_opt_get(conn_t *connp, int level, int name, uchar_t *ptr)
 		case TCP_KEEPALIVE_THRESHOLD:
 			*i1 = tcp->tcp_ka_interval;
 			return (sizeof (int));
+
+		/*
+		 * TCP_KEEPIDLE expects value in seconds, but
+		 * tcp_ka_interval is in milliseconds.
+		 */
+		case TCP_KEEPIDLE:
+			*i1 = tcp->tcp_ka_interval / 1000;
+			return (sizeof (int));
+		case TCP_KEEPCNT:
+			*i1 = tcp->tcp_ka_cnt;
+			return (sizeof (int));
+
+		/*
+		 * TCP_KEEPINTVL expects value in seconds, but
+		 * tcp_ka_rinterval is in milliseconds.
+		 */
+		case TCP_KEEPINTVL:
+			*i1 = tcp->tcp_ka_rinterval / 1000;
+			return (sizeof (int));
 		case TCP_KEEPALIVE_ABORT_THRESHOLD:
 			*i1 = tcp->tcp_ka_abort_thres;
 			return (sizeof (int));
@@ -682,6 +708,18 @@ tcp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
 			}
 			tcp->tcp_init_cwnd = val;
 			break;
+
+		/*
+		 * TCP_KEEPIDLE is in seconds but TCP_KEEPALIVE_THRESHOLD
+		 * is in milliseconds. TCP_KEEPIDLE is introduced for
+		 * compatibility with other Unix flavors.
+		 * We can fall through TCP_KEEPALIVE_THRESHOLD logic after
+		 * converting the input to milliseconds.
+		 */
+		case TCP_KEEPIDLE:
+			*i1 *= 1000;
+			/* fall through */
+
 		case TCP_KEEPALIVE_THRESHOLD:
 			if (checkonly)
 				break;
@@ -708,6 +746,66 @@ tcp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
 				}
 			}
 			break;
+
+		/*
+		 * tcp_ka_abort_thres = tcp_ka_rinterval * tcp_ka_cnt.
+		 * So setting TCP_KEEPCNT or TCP_KEEPINTVL can affect all the
+		 * three members - tcp_ka_abort_thres, tcp_ka_rinterval and
+		 * tcp_ka_cnt.
+		 */
+		case TCP_KEEPCNT:
+			if (checkonly)
+				break;
+
+			if (*i1 == 0) {
+				return (EINVAL);
+			} else if (tcp->tcp_ka_rinterval == 0) {
+				if ((tcp->tcp_ka_abort_thres / *i1) <
+				    tcp->tcp_rto_min ||
+				    (tcp->tcp_ka_abort_thres / *i1) >
+				    tcp->tcp_rto_max)
+					return (EINVAL);
+
+				tcp->tcp_ka_rinterval =
+				    tcp->tcp_ka_abort_thres / *i1;
+			} else {
+				if ((*i1 * tcp->tcp_ka_rinterval) <
+				    tcps->tcps_keepalive_abort_interval_low ||
+				    (*i1 * tcp->tcp_ka_rinterval) >
+				    tcps->tcps_keepalive_abort_interval_high)
+					return (EINVAL);
+				tcp->tcp_ka_abort_thres =
+				    (*i1 * tcp->tcp_ka_rinterval);
+			}
+			tcp->tcp_ka_cnt = *i1;
+			break;
+		case TCP_KEEPINTVL:
+			/*
+			 * TCP_KEEPINTVL is specified in seconds, but
+			 * tcp_ka_rinterval is in milliseconds.
+			 */
+
+			if (checkonly)
+				break;
+
+			if ((*i1 * 1000) < tcp->tcp_rto_min ||
+			    (*i1 * 1000) > tcp->tcp_rto_max)
+				return (EINVAL);
+
+			if (tcp->tcp_ka_cnt == 0) {
+				tcp->tcp_ka_cnt =
+				    tcp->tcp_ka_abort_thres / (*i1 * 1000);
+			} else {
+				if ((*i1 * tcp->tcp_ka_cnt * 1000) <
+				    tcps->tcps_keepalive_abort_interval_low ||
+				    (*i1 * tcp->tcp_ka_cnt * 1000) >
+				    tcps->tcps_keepalive_abort_interval_high)
+					return (EINVAL);
+				tcp->tcp_ka_abort_thres =
+				    (*i1 * tcp->tcp_ka_cnt * 1000);
+			}
+			tcp->tcp_ka_rinterval = *i1 * 1000;
+			break;
 		case TCP_KEEPALIVE_ABORT_THRESHOLD:
 			if (!checkonly) {
 				if (*i1 <
@@ -718,6 +816,8 @@ tcp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,
 					return (EINVAL);
 				}
 				tcp->tcp_ka_abort_thres = *i1;
+				tcp->tcp_ka_cnt = 0;
+				tcp->tcp_ka_rinterval = 0;
 			}
 			break;
 		case TCP_CORK:
diff --git a/usr/src/uts/common/inet/tcp/tcp_timers.c b/usr/src/uts/common/inet/tcp/tcp_timers.c
index c883be8cfd..90e1c9178c 100644
--- a/usr/src/uts/common/inet/tcp/tcp_timers.c
+++ b/usr/src/uts/common/inet/tcp/tcp_timers.c
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <sys/types.h>
@@ -390,6 +391,11 @@ tcp_timers_stop(tcp_t *tcp)
  * (tcp_ka_interval + tcp_ka_abort_thres) we have not heard anything,
  * kill the connection unless the keepalive abort threshold is 0.  In
  * that case, we will probe "forever."
+ * If tcp_ka_cnt and tcp_ka_rinterval are non-zero, then we do not follow
+ * the exponential backoff, but send probes tcp_ka_cnt times in regular
+ * intervals of tcp_ka_rinterval milliseconds until we hear back from peer.
+ * Kill the connection if we don't hear back from peer after tcp_ka_cnt
+ * probes are sent.
  */
 void
 tcp_keepalive_timer(void *arg)
@@ -455,7 +461,9 @@ tcp_keepalive_timer(void *arg)
 			if (mp != NULL) {
 				tcp_send_data(tcp, mp);
 				TCPS_BUMP_MIB(tcps, tcpTimKeepaliveProbe);
-				if (tcp->tcp_ka_last_intrvl != 0) {
+				if (tcp->tcp_ka_rinterval) {
+					firetime = tcp->tcp_ka_rinterval;
+				} else if (tcp->tcp_ka_last_intrvl != 0) {
 					int max;
 					/*
 					 * We should probe again at least
diff --git a/usr/src/uts/common/netinet/tcp.h b/usr/src/uts/common/netinet/tcp.h
index 9a08545ab7..f6c2fc160b 100644
--- a/usr/src/uts/common/netinet/tcp.h
+++ b/usr/src/uts/common/netinet/tcp.h
@@ -21,6 +21,7 @@
 
 /*
  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
  */
 
 /*
@@ -125,6 +126,9 @@ struct tcphdr {
 /* gap for expansion of ``standard'' options */
 #define	TCP_ANONPRIVBIND		0x20	/* for internal use only  */
 #define	TCP_EXCLBIND			0x21	/* for internal use only  */
+#define	TCP_KEEPIDLE			0x22
+#define	TCP_KEEPCNT			0x23
+#define	TCP_KEEPINTVL			0x24
 
 #ifdef	__cplusplus
 }