Split in6_selectsrc() into in6_selectsrc_addr() and in6_selectsrc_socket().

in6_selectsrc() has 2 class of users: socket-based one (raw/udp/pcb/etc) and
  socket-less (ND code). The main reason for that change is inability to
  specify non-default FIB for callers w/o socket since (internally) inpcb
  is used to determine fib.

As as result, add 2 wrappers for in6_selectsrc() (making in6_selectsrc()
  static):
1) in6_selectsrc_socket() for the former class. Embed scope_ambiguous check
  along with returning hop limit when needed.
2) in6_selectsrc_addr() for the latter case. Add 'fibnum' argument and
  pass IPv6 address  w/ explicitly specified scope as separate argument.

Reviewed by:	ae (previous version)

7 files changed, 121 insertions, 88 deletions

diff --git a/sys/netinet6/icmp6.c b/sys/netinet6/icmp6.c
index 2007a18..dd77527 100644
--- a/sys/netinet6/icmp6.c
+++ b/sys/netinet6/icmp6.c
@@ -2194,26 +2194,25 @@ icmp6_reflect(struct mbuf *m, size_t off)
 	}
 
 	if (srcp == NULL) {
-		int e;
-		struct sockaddr_in6 sin6;
+		int error;
+		struct in6_addr dst6;
+		uint32_t scopeid;
 
 		/*
 		 * This case matches to multicasts, our anycast, or unicasts
 		 * that we do not own.  Select a source address based on the
 		 * source address of the erroneous packet.
 		 */
-		bzero(&sin6, sizeof(sin6));
-		sin6.sin6_family = AF_INET6;
-		sin6.sin6_len = sizeof(sin6);
-		sin6.sin6_addr = ip6->ip6_dst; /* zone ID should be embedded */
+		in6_splitscope(&ip6->ip6_dst, &dst6, &scopeid);
+		error = in6_selectsrc_addr(RT_DEFAULT_FIB, &dst6,
+		    scopeid, NULL, &src6, &hlim);
 
-		e = in6_selectsrc(&sin6, NULL, NULL, NULL, &outif, &src6);
-		if (e) {
+		if (error) {
 			char ip6buf[INET6_ADDRSTRLEN];
 			nd6log((LOG_DEBUG,
 			    "icmp6_reflect: source can't be determined: "
 			    "dst=%s, error=%d\n",
-			    ip6_sprintf(ip6buf, &ip6->ip6_dst), e));
+			    ip6_sprintf(ip6buf, &ip6->ip6_dst), error));
 			goto bad;
 		}
 		srcp = &src6;
@@ -2228,10 +2227,7 @@ icmp6_reflect(struct mbuf *m, size_t off)
 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
 	ip6->ip6_vfc |= IPV6_VERSION;
 	ip6->ip6_nxt = IPPROTO_ICMPV6;
-	if (outif)
-		ip6->ip6_hlim = ND_IFINFO(outif)->chlim;
-	else
-		ip6->ip6_hlim = hlim;
+	ip6->ip6_hlim = hlim;
 
 	icmp6->icmp6_cksum = 0;
 	icmp6->icmp6_cksum = in6_cksum(m, IPPROTO_ICMPV6,
diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index 67ecb8c..a779b8d 100644
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -328,7 +328,6 @@ in6_pcbladdr(register struct inpcb *inp, struct sockaddr *nam,
 {
 	register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
 	int error = 0;
-	struct ifnet *ifp = NULL;
 	int scope_ambiguous = 0;
 	struct in6_addr in6a;
 
@@ -358,20 +357,15 @@ in6_pcbladdr(register struct inpcb *inp, struct sockaddr *nam,
 	if ((error = prison_remote_ip6(inp->inp_cred, &sin6->sin6_addr)) != 0)
 		return (error);
 
-	error = in6_selectsrc(sin6, inp->in6p_outputopts,
-	    inp, inp->inp_cred, &ifp, &in6a);
+	error = in6_selectsrc_socket(sin6, inp->in6p_outputopts,
+	    inp, inp->inp_cred, scope_ambiguous, &in6a, NULL);
 	if (error)
 		return (error);
 
-	if (ifp && scope_ambiguous &&
-	    (error = in6_setscope(&sin6->sin6_addr, ifp, NULL)) != 0) {
-		return(error);
-	}
-
 	/*
 	 * Do not update this earlier, in case we return with an error.
 	 *
-	 * XXX: this in6_selectsrc result might replace the bound local
+	 * XXX: this in6_selectsrc_socket result might replace the bound local
 	 * address with the address specified by setsockopt(IPV6_PKTINFO).
 	 * Is it the intended behavior?
 	 */
diff --git a/sys/netinet6/in6_src.c b/sys/netinet6/in6_src.c
index fb362e2..4029402 100644
--- a/sys/netinet6/in6_src.c
+++ b/sys/netinet6/in6_src.c
@@ -136,6 +136,9 @@ static int selectroute(struct sockaddr_in6 *, struct ip6_pktopts *,
 static int in6_selectif(struct sockaddr_in6 *, struct ip6_pktopts *,
 	struct ip6_moptions *, struct ifnet **,
 	struct ifnet *, u_int);
+static int in6_selectsrc(uint32_t, struct sockaddr_in6 *,
+	struct ip6_pktopts *, struct inpcb *, struct ucred *,
+	struct ifnet **, struct in6_addr *);
 
 static struct in6_addrpolicy *lookup_addrsel_policy(struct sockaddr_in6 *);
 
@@ -175,9 +178,9 @@ static struct in6_addrpolicy *match_addrsel_policy(struct sockaddr_in6 *);
 	goto out;		/* XXX: we can't use 'break' here */ \
 } while(0)
 
-int
-in6_selectsrc(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
-    struct inpcb *inp, struct ucred *cred,
+static int
+in6_selectsrc(uint32_t fibnum, struct sockaddr_in6 *dstsock,
+    struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred,
     struct ifnet **ifpp, struct in6_addr *srcp)
 {
 	struct rm_priotracker in6_ifa_tracker;
@@ -228,7 +231,7 @@ in6_selectsrc(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
 
 		/* get the outgoing interface */
 		if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp,
-		    (inp != NULL) ? inp->inp_inc.inc_fibnum : RT_DEFAULT_FIB))
+		    fibnum))
 		    != 0)
 			return (error);
 
@@ -545,6 +548,79 @@ in6_selectsrc(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
 }
 
 /*
+ * Select source address based on @inp, @dstsock and @opts.
+ * Stores selected address to @srcp. If @scope_ambiguous is set,
+ * embed scope from selected outgoing interface. If @hlim pointer
+ * is provided, stores calculated hop limit there.
+ * Returns 0 on success.
+ */
+int
+in6_selectsrc_socket(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts,
+    struct inpcb *inp, struct ucred *cred, int scope_ambiguous,
+    struct in6_addr *srcp, int *hlim)
+{
+	struct ifnet *retifp;
+	uint32_t fibnum;
+	int error;
+
+	fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
+	retifp = NULL;
+
+	error = in6_selectsrc(fibnum, dstsock, opts, inp, cred, &retifp, srcp);
+	if (error != 0)
+		return (error);
+
+	if (hlim != NULL)
+		*hlim = in6_selecthlim(inp, retifp);
+
+	if (retifp == NULL || scope_ambiguous == 0)
+		return (0);
+
+	/*
+	 * Application should provide a proper zone ID or the use of
+	 * default zone IDs should be enabled.  Unfortunately, some
+	 * applications do not behave as it should, so we need a
+	 * workaround.  Even if an appropriate ID is not determined
+	 * (when it's required), if we can determine the outgoing
+	 * interface. determine the zone ID based on the interface.
+	 */
+	error = in6_setscope(&dstsock->sin6_addr, retifp, NULL);
+
+	return (error);
+}
+
+/*
+ * Select source address based on @fibnum, @dst and @scopeid.
+ * Stores selected address to @srcp.
+ * Returns 0 on success.
+ *
+ * Used by non-socket based consumers (ND code mostly)
+ */
+int
+in6_selectsrc_addr(uint32_t fibnum, const struct in6_addr *dst,
+    uint32_t scopeid, struct ifnet *ifp, struct in6_addr *srcp,
+    int *hlim)
+{
+	struct ifnet *retifp;
+	struct sockaddr_in6 dst_sa;
+	int error;
+
+	retifp = ifp;
+	bzero(&dst_sa, sizeof(dst_sa));
+	dst_sa.sin6_family = AF_INET6;
+	dst_sa.sin6_len = sizeof(dst_sa);
+	dst_sa.sin6_addr = *dst;
+	dst_sa.sin6_scope_id = scopeid;
+	sa6_embedscope(&dst_sa, 0);
+
+	error = in6_selectsrc(fibnum, &dst_sa, NULL, NULL, NULL, &retifp, srcp);
+	if (hlim != NULL)
+		*hlim = in6_selecthlim(NULL, retifp);
+
+	return (error);
+}
+
+/*
  * clone - meaningful only for bsdi and freebsd
  */
 static int
diff --git a/sys/netinet6/ip6_var.h b/sys/netinet6/ip6_var.h
index f43a28c..ef86eca 100644
--- a/sys/netinet6/ip6_var.h
+++ b/sys/netinet6/ip6_var.h
@@ -418,9 +418,10 @@ int	rip6_usrreq(struct socket *,
 int	dest6_input(struct mbuf **, int *, int);
 int	none_input(struct mbuf **, int *, int);
 
-int	in6_selectsrc(struct sockaddr_in6 *, struct ip6_pktopts *,
-	struct inpcb *inp, struct ucred *cred,
-	struct ifnet **, struct in6_addr *);
+int	in6_selectsrc_socket(struct sockaddr_in6 *, struct ip6_pktopts *,
+    struct inpcb *, struct ucred *, int, struct in6_addr *, int *);
+int	in6_selectsrc_addr(uint32_t, const struct in6_addr *,
+    uint32_t, struct ifnet *, struct in6_addr *, int *);
 int in6_selectroute(struct sockaddr_in6 *, struct ip6_pktopts *,
 	struct ip6_moptions *, struct route_in6 *, struct ifnet **,
 	struct rtentry **);
diff --git a/sys/netinet6/nd6_nbr.c b/sys/netinet6/nd6_nbr.c
index b7ba0b2..3a4e8a0 100644
--- a/sys/netinet6/nd6_nbr.c
+++ b/sys/netinet6/nd6_nbr.c
@@ -481,27 +481,21 @@ nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *saddr6,
 			ifa_free(ifa);
 		} else {
 			int error;
-			struct sockaddr_in6 dst_sa;
-			struct in6_addr src_in;
-			struct ifnet *oifp;
-
-			bzero(&dst_sa, sizeof(dst_sa));
-			dst_sa.sin6_family = AF_INET6;
-			dst_sa.sin6_len = sizeof(dst_sa);
-			dst_sa.sin6_addr = ip6->ip6_dst;
-
-			oifp = ifp;
-			error = in6_selectsrc(&dst_sa, NULL,
-			    NULL, NULL, &oifp, &src_in);
+			struct in6_addr dst6, src6;
+			uint32_t scopeid;
+
+			in6_splitscope(&ip6->ip6_dst, &dst6, &scopeid);
+			error = in6_selectsrc_addr(RT_DEFAULT_FIB, &dst6,
+			    scopeid, ifp, &src6, NULL);
 			if (error) {
 				char ip6buf[INET6_ADDRSTRLEN];
 				nd6log((LOG_DEBUG, "%s: source can't be "
 				    "determined: dst=%s, error=%d\n", __func__,
-				    ip6_sprintf(ip6buf, &dst_sa.sin6_addr),
+				    ip6_sprintf(ip6buf, &dst6),
 				    error));
 				goto bad;
 			}
-			ip6->ip6_src = src_in;
+			ip6->ip6_src = src6;
 		}
 	} else {
 		/*
@@ -941,12 +935,12 @@ nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0,
 {
 	struct mbuf *m;
 	struct m_tag *mtag;
-	struct ifnet *oifp;
 	struct ip6_hdr *ip6;
 	struct nd_neighbor_advert *nd_na;
 	struct ip6_moptions im6o;
-	struct in6_addr src, daddr6;
-	struct sockaddr_in6 dst_sa;
+	struct in6_addr daddr6, dst6, src6;
+	uint32_t scopeid;
+
 	int icmp6len, maxlen, error;
 	caddr_t mac = NULL;
 
@@ -998,24 +992,21 @@ nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0,
 		flags &= ~ND_NA_FLAG_SOLICITED;
 	}
 	ip6->ip6_dst = daddr6;
-	bzero(&dst_sa, sizeof(struct sockaddr_in6));
-	dst_sa.sin6_family = AF_INET6;
-	dst_sa.sin6_len = sizeof(struct sockaddr_in6);
-	dst_sa.sin6_addr = daddr6;
 
 	/*
 	 * Select a source whose scope is the same as that of the dest.
 	 */
-	oifp = ifp;
-	error = in6_selectsrc(&dst_sa, NULL, NULL, NULL, &oifp, &src);
+	in6_splitscope(&daddr6, &dst6, &scopeid);
+	error = in6_selectsrc_addr(RT_DEFAULT_FIB, &dst6,
+	    scopeid, ifp, &src6, NULL);
 	if (error) {
 		char ip6buf[INET6_ADDRSTRLEN];
 		nd6log((LOG_DEBUG, "nd6_na_output: source can't be "
 		    "determined: dst=%s, error=%d\n",
-		    ip6_sprintf(ip6buf, &dst_sa.sin6_addr), error));
+		    ip6_sprintf(ip6buf, &daddr6), error));
 		goto bad;
 	}
-	ip6->ip6_src = src;
+	ip6->ip6_src = src6;
 	nd_na = (struct nd_neighbor_advert *)(ip6 + 1);
 	nd_na->nd_na_type = ND_NEIGHBOR_ADVERT;
 	nd_na->nd_na_code = 0;
diff --git a/sys/netinet6/raw_ip6.c b/sys/netinet6/raw_ip6.c
index 0725842..972eb9f 100644
--- a/sys/netinet6/raw_ip6.c
+++ b/sys/netinet6/raw_ip6.c
@@ -397,6 +397,7 @@ rip6_output(struct mbuf *m, struct socket *so, ...)
 	int type = 0, code = 0;		/* for ICMPv6 output statistics only */
 	int scope_ambiguous = 0;
 	int use_defzone = 0;
+	int hlim = 0;
 	struct in6_addr in6a;
 	va_list ap;
 
@@ -460,8 +461,9 @@ rip6_output(struct mbuf *m, struct socket *so, ...)
 	/*
 	 * Source address selection.
 	 */
-	error = in6_selectsrc(dstsock, optp, in6p, so->so_cred,
-	    &oifp, &in6a);
+	error = in6_selectsrc_socket(dstsock, optp, in6p, so->so_cred,
+	    scope_ambiguous, &in6a, &hlim);
+
 	if (error)
 		goto bad;
 	error = prison_check_ip6(in6p->inp_cred, &in6a);
@@ -469,19 +471,6 @@ rip6_output(struct mbuf *m, struct socket *so, ...)
 		goto bad;
 	ip6->ip6_src = in6a;
 
-	if (oifp && scope_ambiguous) {
-		/*
-		 * Application should provide a proper zone ID or the use of
-		 * default zone IDs should be enabled.  Unfortunately, some
-		 * applications do not behave as it should, so we need a
-		 * workaround.  Even if an appropriate ID is not determined
-		 * (when it's required), if we can determine the outgoing
-		 * interface. determine the zone ID based on the interface.
-		 */
-		error = in6_setscope(&dstsock->sin6_addr, oifp, NULL);
-		if (error != 0)
-			goto bad;
-	}
 	ip6->ip6_dst = dstsock->sin6_addr;
 
 	/*
@@ -496,7 +485,7 @@ rip6_output(struct mbuf *m, struct socket *so, ...)
 	 * ip6_plen will be filled in ip6_output, so not fill it here.
 	 */
 	ip6->ip6_nxt = in6p->inp_ip_p;
-	ip6->ip6_hlim = in6_selecthlim(in6p, oifp);
+	ip6->ip6_hlim = hlim;
 
 	if (so->so_proto->pr_protocol == IPPROTO_ICMPV6 ||
 	    in6p->in6p_cksum != -1) {
@@ -784,7 +773,6 @@ rip6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 	struct inpcb *inp;
 	struct sockaddr_in6 *addr = (struct sockaddr_in6 *)nam;
 	struct in6_addr in6a;
-	struct ifnet *ifp = NULL;
 	int error = 0, scope_ambiguous = 0;
 
 	inp = sotoinpcb(so);
@@ -813,21 +801,14 @@ rip6_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 	INP_INFO_WLOCK(&V_ripcbinfo);
 	INP_WLOCK(inp);
 	/* Source address selection. XXX: need pcblookup? */
-	error = in6_selectsrc(addr, inp->in6p_outputopts,
-	    inp, so->so_cred, &ifp, &in6a);
+	error = in6_selectsrc_socket(addr, inp->in6p_outputopts,
+	    inp, so->so_cred, scope_ambiguous, &in6a, NULL);
 	if (error) {
 		INP_WUNLOCK(inp);
 		INP_INFO_WUNLOCK(&V_ripcbinfo);
 		return (error);
 	}
 
-	/* XXX: see above */
-	if (ifp && scope_ambiguous &&
-	    (error = in6_setscope(&addr->sin6_addr, ifp, NULL)) != 0) {
-		INP_WUNLOCK(inp);
-		INP_INFO_WUNLOCK(&V_ripcbinfo);
-		return (error);
-	}
 	inp->in6p_faddr = addr->sin6_addr;
 	inp->in6p_laddr = in6a;
 	soisconnected(so);
diff --git a/sys/netinet6/udp6_usrreq.c b/sys/netinet6/udp6_usrreq.c
index 353b260..9768df5 100644
--- a/sys/netinet6/udp6_usrreq.c
+++ b/sys/netinet6/udp6_usrreq.c
@@ -631,7 +631,6 @@ udp6_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr6,
 	struct udphdr *udp6;
 	struct in6_addr *laddr, *faddr, in6a;
 	struct sockaddr_in6 *sin6 = NULL;
-	struct ifnet *oifp = NULL;
 	int cscov_partial = 0;
 	int scope_ambiguous = 0;
 	u_short fport;
@@ -731,15 +730,10 @@ udp6_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr6,
 		}
 
 		if (!IN6_IS_ADDR_V4MAPPED(faddr)) {
-			error = in6_selectsrc(sin6, optp, inp,
-			    td->td_ucred, &oifp, &in6a);
+			error = in6_selectsrc_socket(sin6, optp, inp,
+			    td->td_ucred, scope_ambiguous, &in6a, NULL);
 			if (error)
 				goto release;
-			if (oifp && scope_ambiguous &&
-			    (error = in6_setscope(&sin6->sin6_addr,
-			    oifp, NULL))) {
-				goto release;
-			}
 			laddr = &in6a;
 		} else
 			laddr = &inp->in6p_laddr;	/* XXX */