Use rwlock per inpcb table to protect notify list. The notify

function may sleep, so holding a mutex is not possible.  The same
list entry and rwlock is used for UDP multicast and raw IP delivery.
By adding a write lock, exclusive netlock is no longer necessary
for PCB notify and UDP and raw IP input.
OK mvs@

2 files changed, 11 insertions, 5 deletions

diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index 43a6e739e28..471a9614fea 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in6_pcb.c,v 1.119 2022/08/08 12:06:31 bluhm Exp $	*/
+/*	$OpenBSD: in6_pcb.c,v 1.120 2022/08/22 10:37:27 bluhm Exp $	*/
 
 /*
  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -387,8 +387,6 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
 	u_int32_t flowinfo;
 	u_int rdomain;
 
-	NET_ASSERT_LOCKED_EXCLUSIVE();
-
 	if ((unsigned)cmd >= PRC_NCMDS)
 		return;
 
@@ -430,6 +428,7 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
 
 	SIMPLEQ_INIT(&inpcblist);
 	rdomain = rtable_l2(rtable);
+	rw_enter_write(&table->inpt_notify);
 	mtx_enter(&table->inpt_mtx);
 	TAILQ_FOREACH(inp, &table->inpt_queue, inp_queue) {
 		if ((inp->inp_flags & INP_IPV6) == 0)
@@ -513,6 +512,7 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
 		(*notify)(inp, errno);
 		in_pcbunref(inp);
 	}
+	rw_exit_write(&table->inpt_notify);
 }
 
 struct inpcb *
diff --git a/sys/netinet6/raw_ip6.c b/sys/netinet6/raw_ip6.c
index 3809f776044..29af83ad5e4 100644
--- a/sys/netinet6/raw_ip6.c
+++ b/sys/netinet6/raw_ip6.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: raw_ip6.c,v 1.154 2022/08/22 08:08:46 mvs Exp $	*/
+/*	$OpenBSD: raw_ip6.c,v 1.155 2022/08/22 10:37:27 bluhm Exp $	*/
 /*	$KAME: raw_ip6.c,v 1.69 2001/03/04 15:55:44 itojun Exp $	*/
 
 /*
@@ -172,8 +172,8 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
 		}
 	}
 #endif
-	NET_ASSERT_LOCKED_EXCLUSIVE();
 	SIMPLEQ_INIT(&inpcblist);
+	rw_enter_write(&rawin6pcbtable.inpt_notify);
 	mtx_enter(&rawin6pcbtable.inpt_mtx);
 	TAILQ_FOREACH(in6p, &rawin6pcbtable.inpt_queue, inp_queue) {
 		if (in6p->inp_socket->so_state & SS_CANTRCVMORE)
@@ -224,6 +224,8 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
 		struct counters_ref ref;
 		uint64_t *counters;
 
+		rw_exit_write(&rawin6pcbtable.inpt_notify);
+
 		if (proto != IPPROTO_ICMPV6) {
 			rip6stat_inc(rip6s_nosock);
 			if (m->m_flags & M_MCAST)
@@ -240,6 +242,8 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
 		counters = counters_enter(&ref, ip6counters);
 		counters[ip6s_delivered]--;
 		counters_leave(&ref, ip6counters);
+
+		return IPPROTO_DONE;
 	}
 
 	while ((in6p = SIMPLEQ_FIRST(&inpcblist)) != NULL) {
@@ -267,6 +271,8 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
 		}
 		in_pcbunref(in6p);
 	}
+	rw_exit_write(&rawin6pcbtable.inpt_notify);
+
 	return IPPROTO_DONE;
 }