Add _mbuf() variants of various inpcb-related interfaces, including lookup,

hash install, etc.  For now, these are arguments are unused, but as we add
RSS support, we will want to use hashes extracted from mbufs, rather than
manually calculated hashes of header fields, due to the expensive of the
software version of Toeplitz (and similar hashes).

Add notes that it would be nice to be able to pass mbufs into lookup
routines in pf(4), optimising firewall lookup in the same way, but the
code structure there doesn't facilitate that currently.

(In principle there is no reason this couldn't be MFCed -- the change
extends rather than modifies the KBI.  However, it won't be useful without
other previous possibly less MFCable changes.)

Reviewed by:    bz
Sponsored by:   Juniper Networks, Inc.

9 files changed, 96 insertions, 22 deletions

diff --git a/sys/contrib/pf/net/pf.c b/sys/contrib/pf/net/pf.c
index 2ce254f..135d734 100644
--- a/sys/contrib/pf/net/pf.c
+++ b/sys/contrib/pf/net/pf.c
@@ -3034,6 +3034,10 @@ pf_socket_lookup(int direction, struct pf_pdesc *pd)
 #ifdef INET
 	case AF_INET:
 #ifdef __FreeBSD__
+		/*
+		 * XXXRW: would be nice if we had an mbuf here so that we
+		 * could use in_pcblookup_mbuf().
+		 */
 		inp = in_pcblookup(pi, saddr->v4, sport, daddr->v4,
 			dport, INPLOOKUP_RLOCKPCB, NULL);
 		if (inp == NULL) {
@@ -3056,6 +3060,10 @@ pf_socket_lookup(int direction, struct pf_pdesc *pd)
 #ifdef INET6
 	case AF_INET6:
 #ifdef __FreeBSD__
+		/*
+		 * XXXRW: would be nice if we had an mbuf here so that we
+		 * could use in6_pcblookup_mbuf().
+		 */
 		inp = in6_pcblookup(pi, &saddr->v6, sport,
 			&daddr->v6, dport, INPLOOKUP_RLOCKPCB, NULL);
 		if (inp == NULL) {
diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index 4357b66..4aa998f 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -621,7 +621,8 @@ in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
  * then pick one.
  */
 int
-in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
+in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
+    struct ucred *cred, struct mbuf *m)
 {
 	u_short lport, fport;
 	in_addr_t laddr, faddr;
@@ -654,13 +655,20 @@ in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 	inp->inp_laddr.s_addr = laddr;
 	inp->inp_faddr.s_addr = faddr;
 	inp->inp_fport = fport;
-	in_pcbrehash(inp);
+	in_pcbrehash_mbuf(inp, m);
 
 	if (anonport)
 		inp->inp_flags |= INP_ANONPORT;
 	return (0);
 }
 
+int
+in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
+{
+
+	return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
+}
+
 /*
  * Do proper source address selection on an unbound socket in case
  * of connect. Take jails into account as well.
@@ -1626,7 +1634,8 @@ in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
 }
 
 /*
- * Public inpcb lookup routines, accepting a 4-tuple.
+ * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
+ * from which a pre-calculated hash value may be extracted.
  */
 struct inpcb *
 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
@@ -1641,6 +1650,21 @@ in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
 	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
 	    lookupflags, ifp));
 }
+
+struct inpcb *
+in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
+    u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
+    struct ifnet *ifp, struct mbuf *m)
+{
+
+	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
+	    ("%s: invalid lookup flags %d", __func__, lookupflags));
+	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
+	    ("%s: LOCKPCB not set", __func__));
+
+	return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
+	    lookupflags, ifp));
+}
 #endif /* INET */
 
 /*
@@ -1707,7 +1731,7 @@ in_pcbinshash(struct inpcb *inp)
  * not change after in_pcbinshash() has been called.
  */
 void
-in_pcbrehash(struct inpcb *inp)
+in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
 {
 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 	struct inpcbhead *head;
@@ -1733,6 +1757,13 @@ in_pcbrehash(struct inpcb *inp)
 	LIST_INSERT_HEAD(head, inp, inp_hash);
 }
 
+void
+in_pcbrehash(struct inpcb *inp)
+{
+
+	in_pcbrehash_mbuf(inp, NULL);
+}
+
 /*
  * Remove PCB from various lists.
  */
diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index bde9e8d..809bc05 100644
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -541,6 +541,8 @@ int	in_pcb_lport(struct inpcb *, struct in_addr *, u_short *,
 int	in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
 	    u_short *, struct ucred *);
 int	in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *);
+int	in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *,
+	    struct mbuf *);
 int	in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *,
 	    u_short *, in_addr_t *, u_short *, struct inpcb **,
 	    struct ucred *);
@@ -555,10 +557,14 @@ struct inpcb *
 struct inpcb *
 	in_pcblookup(struct inpcbinfo *, struct in_addr, u_int,
 	    struct in_addr, u_int, int, struct ifnet *);
+struct inpcb *
+	in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int,
+	    struct in_addr, u_int, int, struct ifnet *, struct mbuf *);
 void	in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr,
 	    int, struct inpcb *(*)(struct inpcb *, int));
 void	in_pcbref(struct inpcb *);
 void	in_pcbrehash(struct inpcb *);
+void	in_pcbrehash_mbuf(struct inpcb *, struct mbuf *);
 int	in_pcbrele(struct inpcb *);
 int	in_pcbrele_rlocked(struct inpcb *);
 int	in_pcbrele_wlocked(struct inpcb *);
diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c
index 080b4da..e3e9aa6 100644
--- a/sys/netinet/tcp_input.c
+++ b/sys/netinet/tcp_input.c
@@ -791,12 +791,14 @@ findpcb:
 		 * Transparently forwarded. Pretend to be the destination.
 		 * already got one like this?
 		 */
-		inp = in_pcblookup(&V_tcbinfo, ip->ip_src, th->th_sport,
+		inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
 		    ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
-		    m->m_pkthdr.rcvif);
+		    m->m_pkthdr.rcvif, m);
 		if (!inp) {
 			/*
 			 * It's new.  Try to find the ambushing socket.
+			 * Because we've rewritten the destination address,
+			 * any hardware-generated hash is ignored.
 			 */
 			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
 			    th->th_sport, next_hop->sin_addr,
@@ -812,19 +814,19 @@ findpcb:
 	{
 #ifdef INET6
 		if (isipv6)
-			inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
+			inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
 			    th->th_sport, &ip6->ip6_dst, th->th_dport,
 			    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
-			    m->m_pkthdr.rcvif);
+			    m->m_pkthdr.rcvif, m);
 #endif
 #if defined(INET) && defined(INET6)
 		else
 #endif
 #ifdef INET
-			inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
+			inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
 			    th->th_sport, ip->ip_dst, th->th_dport,
 			    INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
-			    m->m_pkthdr.rcvif);
+			    m->m_pkthdr.rcvif, m);
 #endif
 	}
 
diff --git a/sys/netinet/tcp_syncache.c b/sys/netinet/tcp_syncache.c
index 533de4b..5125134 100644
--- a/sys/netinet/tcp_syncache.c
+++ b/sys/netinet/tcp_syncache.c
@@ -730,8 +730,8 @@ syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
 		laddr6 = inp->in6p_laddr;
 		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
 			inp->in6p_laddr = sc->sc_inc.inc6_laddr;
-		if ((error = in6_pcbconnect(inp, (struct sockaddr *)&sin6,
-		    thread0.td_ucred)) != 0) {
+		if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6,
+		    thread0.td_ucred, m)) != 0) {
 			inp->in6p_laddr = laddr6;
 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
 				log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed "
@@ -770,8 +770,8 @@ syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m)
 		laddr = inp->inp_laddr;
 		if (inp->inp_laddr.s_addr == INADDR_ANY)
 			inp->inp_laddr = sc->sc_inc.inc_laddr;
-		if ((error = in_pcbconnect(inp, (struct sockaddr *)&sin,
-		    thread0.td_ucred)) != 0) {
+		if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin,
+		    thread0.td_ucred, m)) != 0) {
 			inp->inp_laddr = laddr;
 			if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) {
 				log(LOG_DEBUG, "%s; %s: in_pcbconnect failed "
diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index 5f2d7c3..fd864c0 100644
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -567,9 +567,9 @@ udp_input(struct mbuf *m, int off)
 	/*
 	 * Locate pcb for datagram.
 	 */
-	inp = in_pcblookup(&V_udbinfo, ip->ip_src, uh->uh_sport,
+	inp = in_pcblookup_mbuf(&V_udbinfo, ip->ip_src, uh->uh_sport,
 	    ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB,
-	    ifp);
+	    ifp, m);
 	if (inp == NULL) {
 		if (udp_log_in_vain) {
 			char buf[4*sizeof "123"];
diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index 17fb953..da73f21 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -364,8 +364,8 @@ in6_pcbladdr(register struct inpcb *inp, struct sockaddr *nam,
  * then pick one.
  */
 int
-in6_pcbconnect(register struct inpcb *inp, struct sockaddr *nam,
-    struct ucred *cred)
+in6_pcbconnect_mbuf(register struct inpcb *inp, struct sockaddr *nam,
+    struct ucred *cred, struct mbuf *m)
 {
 	struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 	register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
@@ -405,11 +405,18 @@ in6_pcbconnect(register struct inpcb *inp, struct sockaddr *nam,
 		inp->inp_flow |=
 		    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
 
-	in_pcbrehash(inp);
+	in_pcbrehash_mbuf(inp, m);
 
 	return (0);
 }
 
+int
+in6_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
+{
+
+	return (in6_pcbconnect_mbuf(inp, nam, cred, NULL));
+}
+
 void
 in6_pcbdisconnect(struct inpcb *inp)
 {
@@ -974,13 +981,27 @@ in6_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
 }
 
 /*
- * Public inpcb lookup routines, accepting a 4-tuple.
+ * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
+ * from which a pre-calculated hash value may be extracted.
  */
 struct inpcb *
 in6_pcblookup(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, u_int fport,
     struct in6_addr *laddr, u_int lport, int lookupflags, struct ifnet *ifp)
 {
+	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
+	    ("%s: invalid lookup flags %d", __func__, lookupflags));
+	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
+	    ("%s: LOCKPCB not set", __func__));
 
+	return (in6_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
+	    lookupflags, ifp));
+}
+
+struct inpcb *
+in6_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in6_addr *faddr,
+    u_int fport, struct in6_addr *laddr, u_int lport, int lookupflags,
+    struct ifnet *ifp, struct mbuf *m)
+{
 	KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
 	    ("%s: invalid lookup flags %d", __func__, lookupflags));
 	KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
diff --git a/sys/netinet6/in6_pcb.h b/sys/netinet6/in6_pcb.h
index bce5ee3..cf24704 100644
--- a/sys/netinet6/in6_pcb.h
+++ b/sys/netinet6/in6_pcb.h
@@ -73,6 +73,8 @@ void	in6_pcbpurgeif0 __P((struct inpcbinfo *, struct ifnet *));
 void	in6_losing __P((struct inpcb *));
 int	in6_pcbbind __P((struct inpcb *, struct sockaddr *, struct ucred *));
 int	in6_pcbconnect __P((struct inpcb *, struct sockaddr *, struct ucred *));
+int	in6_pcbconnect_mbuf __P((struct inpcb *, struct sockaddr *,
+	    struct ucred *, struct mbuf *));
 void	in6_pcbdisconnect __P((struct inpcb *));
 int	in6_pcbladdr(struct inpcb *, struct sockaddr *, struct in6_addr *);
 struct	inpcb *
@@ -87,6 +89,10 @@ struct	inpcb *
 	in6_pcblookup_hash_locked __P((struct inpcbinfo *, struct in6_addr *,
 			   u_int, struct in6_addr *, u_int, int,
 			   struct ifnet *));
+struct	inpcb *
+	in6_pcblookup_mbuf __P((struct inpcbinfo *, struct in6_addr *,
+			   u_int, struct in6_addr *, u_int, int,
+			   struct ifnet *ifp, struct mbuf *));
 void	in6_pcbnotify __P((struct inpcbinfo *, struct sockaddr *,
 			   u_int, const struct sockaddr *, u_int, int, void *,
 			   struct inpcb *(*)(struct inpcb *, int)));
diff --git a/sys/netinet6/udp6_usrreq.c b/sys/netinet6/udp6_usrreq.c
index 75145d7..6723007 100644
--- a/sys/netinet6/udp6_usrreq.c
+++ b/sys/netinet6/udp6_usrreq.c
@@ -377,9 +377,9 @@ udp6_input(struct mbuf **mp, int *offp, int proto)
 	/*
 	 * Locate pcb for datagram.
 	 */
-	inp = in6_pcblookup(&V_udbinfo, &ip6->ip6_src, uh->uh_sport,
+	inp = in6_pcblookup_mbuf(&V_udbinfo, &ip6->ip6_src, uh->uh_sport,
 	    &ip6->ip6_dst, uh->uh_dport, INPLOOKUP_WILDCARD |
-	    INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif);
+	    INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif, m);
 	if (inp == NULL) {
 		if (udp_log_in_vain) {
 			char ip6bufs[INET6_ADDRSTRLEN];