Lock down a socket, milestone 1.

o Add a mutex (sb_mtx) to struct sockbuf. This protects the data in a
  socket buffer. The mutex in the receive buffer also protects the data
  in struct socket.

o Determine the lock strategy for each members in struct socket.

o Lock down the following members:

  - so_count
  - so_options
  - so_linger
  - so_state

o Remove *_locked() socket APIs.  Make the following socket APIs
  touching the members above now require a locked socket:

 - sodisconnect()
 - soisconnected()
 - soisconnecting()
 - soisdisconnected()
 - soisdisconnecting()
 - sofree()
 - soref()
 - sorele()
 - sorwakeup()
 - sotryfree()
 - sowakeup()
 - sowwakeup()

Reviewed by:	alfred

7 files changed, 151 insertions, 38 deletions

diff --git a/sys/netinet6/icmp6.c b/sys/netinet6/icmp6.c
index 38332b2..f9994c6 100644
--- a/sys/netinet6/icmp6.c
+++ b/sys/netinet6/icmp6.c
@@ -1941,8 +1941,11 @@ icmp6_rip6_input(mp, off)
 					if (opts) {
 						m_freem(opts);
 					}
-				} else
+				} else {
+					SOCK_LOCK(last->in6p_socket);
 					sorwakeup(last->in6p_socket);
+					SOCK_UNLOCK(last->in6p_socket);
+				}
 				opts = NULL;
 			}
 		}
@@ -1958,8 +1961,11 @@ icmp6_rip6_input(mp, off)
 			m_freem(m);
 			if (opts)
 				m_freem(opts);
-		} else
+		} else {
+			SOCK_LOCK(last->in6p_socket);
 			sorwakeup(last->in6p_socket);
+			SOCK_UNLOCK(last->in6p_socket);
+		}
 	} else {
 		m_freem(m);
 		ip6stat.ip6s_delivered--;
diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index 09a5c29..ab37249 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -124,14 +124,19 @@ in6_pcbbind(inp, nam, td)
 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)NULL;
 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 	u_short	lport = 0;
-	int wild = 0, reuseport = (so->so_options & SO_REUSEPORT);
+	int wild = 0, reuseport;
 
+	SOCK_LOCK(so);
+	reuseport = (so->so_options & SO_REUSEPORT);
+	SOCK_UNLOCK(so);
 	if (!in6_ifaddr) /* XXX broken! */
 		return (EADDRNOTAVAIL);
 	if (inp->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
 		return(EINVAL);
+	SOCK_LOCK(so);
 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
 		wild = 1;
+	SOCK_UNLOCK(so);
 	if (nam) {
 		sin6 = (struct sockaddr_in6 *)nam;
 		if (nam->sa_len != sizeof(*sin6))
@@ -157,8 +162,10 @@ in6_pcbbind(inp, nam, td)
 			 * and a multicast address is bound on both
 			 * new and duplicated sockets.
 			 */
+			SOCK_LOCK(so);
 			if (so->so_options & SO_REUSEADDR)
 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
+			SOCK_UNLOCK(so);
 		} else if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
 			struct ifaddr *ia = NULL;
 
@@ -190,14 +197,19 @@ in6_pcbbind(inp, nam, td)
 				t = in6_pcblookup_local(pcbinfo,
 				    &sin6->sin6_addr, lport,
 				    INPLOOKUP_WILDCARD);
-				if (t &&
-				    (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
-				     !IN6_IS_ADDR_UNSPECIFIED(&t->in6p_laddr) ||
-				     (t->inp_socket->so_options &
-				      SO_REUSEPORT) == 0) &&
-				    (so->so_cred->cr_uid !=
-				     t->inp_socket->so_cred->cr_uid))
-					return (EADDRINUSE);
+				if (t != NULL) {
+					SOCK_LOCK(t->inp_socket);
+					if ((!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
+					     !IN6_IS_ADDR_UNSPECIFIED(&t->in6p_laddr) ||
+					     (t->inp_socket->so_options &
+					      SO_REUSEPORT) == 0) &&
+					    (so->so_cred->cr_uid !=
+					     t->inp_socket->so_cred->cr_uid)) {
+						SOCK_UNLOCK(t->inp_socket);
+						return (EADDRINUSE);
+					}
+					SOCK_UNLOCK(t->inp_socket);
+				}
 				if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 &&
 				    IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
 					struct sockaddr_in sin;
@@ -218,8 +230,14 @@ in6_pcbbind(inp, nam, td)
 			}
 			t = in6_pcblookup_local(pcbinfo, &sin6->sin6_addr,
 						lport, wild);
-			if (t && (reuseport & t->inp_socket->so_options) == 0)
-				return(EADDRINUSE);
+			if (t != NULL) {
+				SOCK_LOCK(t->inp_socket);
+				if ((reuseport & t->inp_socket->so_options) == 0) {
+					SOCK_UNLOCK(t->inp_socket);
+					return(EADDRINUSE);
+				}
+				SOCK_UNLOCK(t->inp_socket);
+			}
 			if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 &&
 			    IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
 				struct sockaddr_in sin;
@@ -227,14 +245,19 @@ in6_pcbbind(inp, nam, td)
 				in6_sin6_2_sin(&sin, sin6);
 				t = in_pcblookup_local(pcbinfo, sin.sin_addr,
 						       lport, wild);
-				if (t &&
-				    (reuseport & t->inp_socket->so_options)
-				    == 0 &&
-				    (ntohl(t->inp_laddr.s_addr)
-				     != INADDR_ANY ||
-				     INP_SOCKAF(so) ==
-				     INP_SOCKAF(t->inp_socket)))
-					return (EADDRINUSE);
+				if (t != NULL) {
+					SOCK_LOCK(t->inp_socket);
+					if ((reuseport & t->inp_socket->so_options)
+					    == 0 &&
+					    (ntohl(t->inp_laddr.s_addr)
+					     != INADDR_ANY ||
+					     INP_SOCKAF(so) ==
+					     INP_SOCKAF(t->inp_socket))) {
+						SOCK_UNLOCK(t->inp_socket);
+						return (EADDRINUSE);
+					}
+					SOCK_UNLOCK(t->inp_socket);
+				}
 			}
 		}
 		inp->in6p_laddr = sin6->sin6_addr;
@@ -589,8 +612,12 @@ in6_pcbdisconnect(inp)
 	/* clear flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
 	inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK;
 	in_pcbrehash(inp);
-	if (inp->inp_socket->so_state & SS_NOFDREF)
+	SOCK_LOCK(inp->inp_socket);
+	if (inp->inp_socket->so_state & SS_NOFDREF) {
+		SOCK_UNLOCK(inp->inp_socket);
 		in6_pcbdetach(inp);
+	} else
+		SOCK_UNLOCK(inp->inp_socket);
 }
 
 void
@@ -607,6 +634,7 @@ in6_pcbdetach(inp)
 	inp->inp_gencnt = ++ipi->ipi_gencnt;
 	in_pcbremlists(inp);
 	sotoinpcb(so) = 0;
+	SOCK_LOCK(so);
 	sotryfree(so);
 
 	if (inp->in6p_options)
diff --git a/sys/netinet6/in6_src.c b/sys/netinet6/in6_src.c
index 3dd2212..910e66d 100644
--- a/sys/netinet6/in6_src.c
+++ b/sys/netinet6/in6_src.c
@@ -338,9 +338,11 @@ in6_pcbsetport(laddr, inp, td)
 	int count, error = 0, wild = 0;
 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 
+	SOCK_LOCK(so);
 	/* XXX: this is redundant when called from in6_pcbbind */
 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
 		wild = INPLOOKUP_WILDCARD;
+	SOCK_UNLOCK(so);
 
 	inp->inp_flags |= INP_ANONPORT;
 
diff --git a/sys/netinet6/ip6_input.c b/sys/netinet6/ip6_input.c
index 16df5bb..f237cdf 100644
--- a/sys/netinet6/ip6_input.c
+++ b/sys/netinet6/ip6_input.c
@@ -1147,16 +1147,19 @@ ip6_savecontrol(in6p, mp, ip6, m)
  		privileged++;
 
 #ifdef SO_TIMESTAMP
+	SOCK_LOCK(in6p->in6p_socket);
 	if ((in6p->in6p_socket->so_options & SO_TIMESTAMP) != 0) {
 		struct timeval tv;
 
+		SOCK_UNLOCK(in6p->in6p_socket);
 		microtime(&tv);
 		*mp = sbcreatecontrol((caddr_t) &tv, sizeof(tv),
 				      SCM_TIMESTAMP, SOL_SOCKET);
 		if (*mp) {
 			mp = &(*mp)->m_next;
 		}
-	}
+	} else
+		SOCK_UNLOCK(in6p->in6p_socket);
 #endif
 
 	/* RFC 2292 sec. 5 */
diff --git a/sys/netinet6/ip6_mroute.c b/sys/netinet6/ip6_mroute.c
index 9af7e81..513e307 100644
--- a/sys/netinet6/ip6_mroute.c
+++ b/sys/netinet6/ip6_mroute.c
@@ -892,7 +892,9 @@ socket_send(s, mm, src)
 		if (sbappendaddr(&s->so_rcv,
 				 (struct sockaddr *)src,
 				 mm, (struct mbuf *)0) != 0) {
+			SOCK_LOCK(s);
 			sorwakeup(s);
+			SOCK_UNLOCK(s);
 			return 0;
 		}
 	}
diff --git a/sys/netinet6/raw_ip6.c b/sys/netinet6/raw_ip6.c
index 2dcec0c..5fbd7c9 100644
--- a/sys/netinet6/raw_ip6.c
+++ b/sys/netinet6/raw_ip6.c
@@ -182,9 +182,16 @@ rip6_input(mp, offp, proto)
 			} else
 #endif /*IPSEC*/
 			if (n) {
-				if (last->in6p_flags & IN6P_CONTROLOPTS ||
-				    last->in6p_socket->so_options & SO_TIMESTAMP)
+				if (last->in6p_flags & IN6P_CONTROLOPTS)
 					ip6_savecontrol(last, &opts, ip6, n);
+				else {
+					SOCK_LOCK(last->in6p_socket);
+					if (last->in6p_socket->so_options & SO_TIMESTAMP) {
+						SOCK_UNLOCK(last->in6p_socket);
+						ip6_savecontrol(last, &opts, ip6, n);
+					} else
+						SOCK_UNLOCK(last->in6p_socket);
+				}
 				/* strip intermediate headers */
 				m_adj(n, *offp);
 				if (sbappendaddr(&last->in6p_socket->so_rcv,
@@ -194,8 +201,11 @@ rip6_input(mp, offp, proto)
 					if (opts)
 						m_freem(opts);
 					rip6stat.rip6s_fullsock++;
-				} else
+				} else {
+					SOCK_LOCK(last->in6p_socket);
 					sorwakeup(last->in6p_socket);
+					SOCK_UNLOCK(last->in6p_socket);
+				}
 				opts = NULL;
 			}
 		}
@@ -213,9 +223,16 @@ rip6_input(mp, offp, proto)
 	} else
 #endif /*IPSEC*/
 	if (last) {
-		if (last->in6p_flags & IN6P_CONTROLOPTS ||
-		    last->in6p_socket->so_options & SO_TIMESTAMP)
+		if (last->in6p_flags & IN6P_CONTROLOPTS)
 			ip6_savecontrol(last, &opts, ip6, m);
+		else {
+			SOCK_LOCK(last->in6p_socket);
+			if (last->in6p_socket->so_options & SO_TIMESTAMP) {
+				SOCK_UNLOCK(last->in6p_socket);
+				ip6_savecontrol(last, &opts, ip6, m);
+			} else
+				SOCK_UNLOCK(last->in6p_socket);
+		}
 		/* strip intermediate headers */
 		m_adj(m, *offp);
 		if (sbappendaddr(&last->in6p_socket->so_rcv,
@@ -224,8 +241,11 @@ rip6_input(mp, offp, proto)
 			if (opts)
 				m_freem(opts);
 			rip6stat.rip6s_fullsock++;
-		} else
+		} else {
+			SOCK_LOCK(last->in6p_socket);
 			sorwakeup(last->in6p_socket);
+			SOCK_UNLOCK(last->in6p_socket);
+		}
 	} else {
 		rip6stat.rip6s_nosock++;
 		if (m->m_flags & M_MCAST)
@@ -591,7 +611,9 @@ rip6_detach(struct socket *so)
 static int
 rip6_abort(struct socket *so)
 {
+	SOCK_LOCK(so);
 	soisdisconnected(so);
+	SOCK_UNLOCK(so);
 	return rip6_detach(so);
 }
 
@@ -600,8 +622,12 @@ rip6_disconnect(struct socket *so)
 {
 	struct inpcb *inp = sotoinpcb(so);
 
-	if ((so->so_state & SS_ISCONNECTED) == 0)
+	SOCK_LOCK(so);
+	if ((so->so_state & SS_ISCONNECTED) == 0) {
+		SOCK_UNLOCK(so);
 		return ENOTCONN;
+	}
+	SOCK_UNLOCK(so);
 	inp->in6p_faddr = in6addr_any;
 	return rip6_abort(so);
 }
@@ -669,7 +695,9 @@ rip6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 		return (error ? error : EADDRNOTAVAIL);
 	inp->in6p_laddr = *in6a;
 	inp->in6p_faddr = addr->sin6_addr;
+	SOCK_LOCK(so);
 	soisconnected(so);
+	SOCK_UNLOCK(so);
 	return 0;
 }
 
@@ -689,7 +717,9 @@ rip6_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 	struct sockaddr_in6 *dst;
 
 	/* always copy sockaddr to avoid overwrites */
+	SOCK_LOCK(so);
 	if (so->so_state & SS_ISCONNECTED) {
+		SOCK_UNLOCK(so);
 		if (nam) {
 			m_freem(m);
 			return EISCONN;
@@ -702,6 +732,7 @@ rip6_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
 		      sizeof(struct in6_addr));
 		dst = &tmp;
 	} else {
+		SOCK_UNLOCK(so);
 		if (nam == NULL) {
 			m_freem(m);
 			return ENOTCONN;
diff --git a/sys/netinet6/udp6_usrreq.c b/sys/netinet6/udp6_usrreq.c
index 20913e6..0e58be5 100644
--- a/sys/netinet6/udp6_usrreq.c
+++ b/sys/netinet6/udp6_usrreq.c
@@ -269,10 +269,18 @@ udp6_input(mp, offp, proto)
 					 * and m_copy() will copy M_PKTHDR
 					 * only if offset is 0.
 					 */
-					if (last->in6p_flags & IN6P_CONTROLOPTS
-					    || last->in6p_socket->so_options & SO_TIMESTAMP)
+					if (last->in6p_flags & IN6P_CONTROLOPTS)
 						ip6_savecontrol(last, &opts,
 								ip6, n);
+					else {
+						SOCK_LOCK(last->in6p_socket);
+						if (last->in6p_socket->so_options & SO_TIMESTAMP) {
+							SOCK_UNLOCK(last->in6p_socket);
+							ip6_savecontrol(last, &opts,
+									ip6, n);
+						} else
+							SOCK_UNLOCK(last->in6p_socket);
+					}
 								
 					m_adj(n, off + sizeof(struct udphdr));
 					if (sbappendaddr(&last->in6p_socket->so_rcv,
@@ -282,8 +290,11 @@ udp6_input(mp, offp, proto)
 						if (opts)
 							m_freem(opts);
 						udpstat.udps_fullsock++;
-					} else
+					} else {
+						SOCK_LOCK(last->in6p_socket);
 						sorwakeup(last->in6p_socket);
+						SOCK_UNLOCK(last->in6p_socket);
+					}
 					opts = NULL;
 				}
 			}
@@ -296,9 +307,13 @@ udp6_input(mp, offp, proto)
 			 * port.  It assumes that an application will never
 			 * clear these options after setting them.
 			 */
+			SOCK_LOCK(last->in6p_socket);
 			if ((last->in6p_socket->so_options &
-			     (SO_REUSEPORT|SO_REUSEADDR)) == 0)
+			     (SO_REUSEPORT|SO_REUSEADDR)) == 0) {
+				SOCK_UNLOCK(last->in6p_socket);
 				break;
+			} else
+				SOCK_UNLOCK(last->in6p_socket);
 		}
 
 		if (last == NULL) {
@@ -320,9 +335,16 @@ udp6_input(mp, offp, proto)
 			goto bad;
 		}
 #endif /* IPSEC */
-		if (last->in6p_flags & IN6P_CONTROLOPTS
-		    || last->in6p_socket->so_options & SO_TIMESTAMP)
+		if (last->in6p_flags & IN6P_CONTROLOPTS)
 			ip6_savecontrol(last, &opts, ip6, m);
+		else {
+			SOCK_LOCK(last->in6p_socket);
+			if (last->in6p_socket->so_options & SO_TIMESTAMP) {
+				SOCK_UNLOCK(last->in6p_socket);
+				ip6_savecontrol(last, &opts, ip6, m);
+			} else
+				SOCK_UNLOCK(last->in6p_socket);
+		}
 
 		m_adj(m, off + sizeof(struct udphdr));
 		if (sbappendaddr(&last->in6p_socket->so_rcv,
@@ -331,7 +353,9 @@ udp6_input(mp, offp, proto)
 			udpstat.udps_fullsock++;
 			goto bad;
 		}
+		SOCK_LOCK(last->in6p_socket);
 		sorwakeup(last->in6p_socket);
+		SOCK_UNLOCK(last->in6p_socket);
 		return IPPROTO_DONE;
 	}
 	/*
@@ -375,9 +399,16 @@ udp6_input(mp, offp, proto)
 	 */
 	init_sin6(&udp_in6, m); /* general init */
 	udp_in6.sin6_port = uh->uh_sport;
-	if (in6p->in6p_flags & IN6P_CONTROLOPTS
-	    || in6p->in6p_socket->so_options & SO_TIMESTAMP)
+	if (in6p->in6p_flags & IN6P_CONTROLOPTS)
 		ip6_savecontrol(in6p, &opts, ip6, m);
+	else {
+		SOCK_LOCK(in6p->in6p_socket);
+		if (in6p->in6p_socket->so_options & SO_TIMESTAMP) {
+			SOCK_UNLOCK(in6p->in6p_socket);
+			ip6_savecontrol(in6p, &opts, ip6, m);
+		} else
+			SOCK_UNLOCK(in6p->in6p_socket);
+	}
 	m_adj(m, off + sizeof(struct udphdr));
 	if (sbappendaddr(&in6p->in6p_socket->so_rcv,
 			(struct sockaddr *)&udp_in6,
@@ -385,7 +416,9 @@ udp6_input(mp, offp, proto)
 		udpstat.udps_fullsock++;
 		goto bad;
 	}
+	SOCK_LOCK(in6p->in6p_socket);
 	sorwakeup(in6p->in6p_socket);
+	SOCK_UNLOCK(in6p->in6p_socket);
 	return IPPROTO_DONE;
 bad:
 	if (m)
@@ -509,7 +542,9 @@ udp6_abort(struct socket *so)
 	inp = sotoinpcb(so);
 	if (inp == 0)
 		return EINVAL;	/* ??? possible? panic instead? */
+	SOCK_LOCK(so);
 	soisdisconnected(so);
+	SOCK_UNLOCK(so);
 	s = splnet();
 	in6_pcbdetach(inp);
 	splx(s);
@@ -614,7 +649,9 @@ udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 			if (error == 0) {
 				inp->inp_vflag |= INP_IPV4;
 				inp->inp_vflag &= ~INP_IPV6;
+				SOCK_LOCK(so);
 				soisconnected(so);
+				SOCK_UNLOCK(so);
 			}
 			return error;
 		}
@@ -629,7 +666,9 @@ udp6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 			inp->inp_vflag &= ~INP_IPV4;
 			inp->inp_vflag |= INP_IPV6;
 		}
+		SOCK_LOCK(so);
 		soisconnected(so);
+		SOCK_UNLOCK(so);
 	}
 	return error;
 }
@@ -673,7 +712,9 @@ udp6_disconnect(struct socket *so)
 	in6_pcbdisconnect(inp);
 	inp->in6p_laddr = in6addr_any;
 	splx(s);
+	SOCK_LOCK(so);
 	so->so_state &= ~SS_ISCONNECTED;		/* XXX */
+	SOCK_UNLOCK(so);
 	return 0;
 }