bgpd: implement SendHoldTimer

As described by
https://www.ietf.org/archive/id/draft-spaghetti-idr-bgp-sendholdtimer-04.html

Since this replicates the HoldTime check on the receiver that is already
part of the protocol, I do not believe it necessary to wait for IETF
progress on this draft.  It's just replicating an existing element of
the protocol at the other side of the session.

Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

1 files changed, 37 insertions, 0 deletions

diff --git a/bgpd/bgp_packet.c b/bgpd/bgp_packet.c
index cacc4995fc..e4697fdb2b 100644
--- a/bgpd/bgp_packet.c
+++ b/bgpd/bgp_packet.c
@@ -122,8 +122,45 @@ void bgp_packet_set_size(struct stream *s)
  */
 static void bgp_packet_add(struct peer *peer, struct stream *s)
 {
+	intmax_t delta;
+	uint32_t holdtime;
+
 	frr_with_mutex(&peer->io_mtx) {
+		/* if the queue is empty, reset the "last OK" timestamp to
+		 * now, otherwise if we write another packet immediately
+		 * after it'll get confused
+		 */
+		if (!stream_fifo_count_safe(peer->obuf))
+			peer->last_sendq_ok = bgp_clock();
+
 		stream_fifo_push(peer->obuf, s);
+
+		delta = bgp_clock() - peer->last_sendq_ok;
+		holdtime = atomic_load_explicit(&peer->holdtime,
+						memory_order_relaxed);
+
+		/* Note that when we're here, we're adding some packet to the
+		 * OutQ.  That includes keepalives when there is nothing to
+		 * do, so there's a guarantee we pass by here once in a while.
+		 *
+		 * That implies there is no need to go set up another separate
+		 * timer that ticks down SendHoldTime, as we'll be here sooner
+		 * or later anyway and will see the checks below failing.
+		 */
+		if (delta > 2 * (intmax_t)holdtime) {
+			flog_err(
+				EC_BGP_SENDQ_STUCK_PROPER,
+				"%s has not made any SendQ progress for 2 holdtimes, terminating session",
+				peer->host);
+			BGP_EVENT_ADD(peer, TCP_fatal_error);
+		} else if (delta > (intmax_t)holdtime &&
+			   bgp_clock() - peer->last_sendq_warn > 5) {
+			flog_warn(
+				EC_BGP_SENDQ_STUCK_WARN,
+				"%s has not made any SendQ progress for 1 holdtime, peer overloaded?",
+				peer->host);
+			peer->last_sendq_warn = bgp_clock();
+		}
 	}
 }