MFC r264321, r264342, r264351, r264356, r273850, r274629:

Currently, the TCP slow timer can starve TCP input processing while it
walks the list of connections in TIME_WAIT closing expired connections
due to contention on the global TCP pcbinfo lock.

To remediate, introduce a new global lock to protect the list of
connections in TIME_WAIT.  Only acquire the TCP pcbinfo lock when
closing an expired connection.  This limits the window of time when
TCP input processing is stopped to the amount of time needed to close
a single connection.

Approved by:    jhb (mentor)

1 files changed, 15 insertions, 0 deletions

diff --git a/sys/netinet/tcp_usrreq.c b/sys/netinet/tcp_usrreq.c
index 8435791..8d92803 100644
--- a/sys/netinet/tcp_usrreq.c
+++ b/sys/netinet/tcp_usrreq.c
@@ -182,6 +182,21 @@ tcp_detach(struct socket *so, struct inpcb *inp)
 		 * present until timewait ends.
 		 *
 		 * XXXRW: Would it be cleaner to free the tcptw here?
+		 *
+		 * Astute question indeed, from twtcp perspective there are
+		 * three cases to consider:
+		 *
+		 * #1 tcp_detach is called at tcptw creation time by
+		 *  tcp_twstart, then do not discard the newly created tcptw
+		 *  and leave inpcb present until timewait ends
+		 * #2 tcp_detach is called at timewait end (or reuse) by
+		 *  tcp_twclose, then the tcptw has already been discarded
+		 *  and inpcb is freed here
+		 * #3 tcp_detach is called() after timewait ends (or reuse)
+		 *  (e.g. by soclose), then tcptw has already been discarded
+		 *  and inpcb is freed here
+		 *
+		 *  In all three cases the tcptw should not be freed here.
 		 */
 		if (inp->inp_flags & INP_DROPPED) {
 			KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "