Add IP(V6)_VLAN_PCP to set 802.1 priority per-flow.

This adds a new IP_PROTO / IPV6_PROTO setsockopt (getsockopt)
option IP(V6)_VLAN_PCP, which can be set to -1 (interface
default), or explicitly to any priority between 0 and 7.

Note that for untagged traffic, explicitly adding a
priority will insert a special 801.1Q vlan header with
vlan ID = 0 to carry the priority setting

Reviewed by:	gallatin, rrs
MFC after:	2 weeks
Sponsored by:	NetApp, Inc.
Differential Revision:	https://reviews.freebsd.org/D26409

6 files changed, 109 insertions, 0 deletions

diff --git a/sys/net/if_ethersubr.c b/sys/net/if_ethersubr.c
index dbb4119d0ed9..35a2a85df457 100644
--- a/sys/net/if_ethersubr.c
+++ b/sys/net/if_ethersubr.c
@@ -1388,6 +1388,13 @@ ether_8021q_frame(struct mbuf **mp, struct ifnet *ife, struct ifnet *p,
 	}
 
 	/*
+	 * If PCP is set in mbuf, use it
+	 */
+	if ((*mp)->m_flags & M_VLANTAG) {
+		pcp = EVL_PRIOFTAG((*mp)->m_pkthdr.ether_vtag);
+	}
+
+	/*
 	 * If underlying interface can do VLAN tag insertion itself,
 	 * just pass the packet along. However, we need some way to
 	 * tell the interface where the packet came from so that it
diff --git a/sys/netinet/in.h b/sys/netinet/in.h
index 84a209eef779..28d6721edc53 100644
--- a/sys/netinet/in.h
+++ b/sys/netinet/in.h
@@ -483,6 +483,10 @@ __END_DECLS
 /* The following option is private; do not use it from user applications. */
 #define	IP_MSFILTER			74   /* set/get filter list */
 
+/* The following option deals with the 802.1Q Ethernet Priority Code Point */
+#define	IP_VLAN_PCP		75   /* int; set/get PCP used for packet, */
+				     /*      -1 use interface default */
+
 /* Protocol Independent Multicast API [RFC3678] */
 #define	MCAST_JOIN_GROUP		80   /* join an any-source group */
 #define	MCAST_LEAVE_GROUP		81   /* leave all sources for group */
diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index 9138ee186f6b..0a6ad1eaa9c4 100644
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -748,6 +748,13 @@ int	inp_so_options(const struct inpcb *inp);
 #define INP_SUPPORTS_MBUFQ	0x00004000 /* Supports the mbuf queue method of LRO */
 #define INP_MBUF_QUEUE_READY	0x00008000 /* The transport is pacing, inputs can be queued */
 #define INP_DONT_SACK_QUEUE	0x00010000 /* If a sack arrives do not wake me */
+#define INP_2PCP_SET		0x00020000 /* If the Eth PCP should be set explicitly */
+#define INP_2PCP_BIT0		0x00040000 /* Eth PCP Bit 0 */
+#define INP_2PCP_BIT1		0x00080000 /* Eth PCP Bit 1 */
+#define INP_2PCP_BIT2		0x00100000 /* Eth PCP Bit 2 */
+#define INP_2PCP_BASE	INP_2PCP_BIT0
+#define INP_2PCP_MASK	(INP_2PCP_BIT0 | INP_2PCP_BIT1 | INP_2PCP_BIT2)
+#define INP_2PCP_SHIFT		18         /* shift PCP field in/out of inp_flags2 */
 /*
  * Flags passed to in_pcblookup*() functions.
  */
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index a7e72f4ec407..2d45d583df15 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -62,7 +62,9 @@ __FBSDID("$FreeBSD$");
 
 #include <net/if.h>
 #include <net/if_var.h>
+#include <net/if_vlan_var.h>
 #include <net/if_llatbl.h>
+#include <net/ethernet.h>
 #include <net/netisr.h>
 #include <net/pfil.h>
 #include <net/route.h>
@@ -324,6 +326,7 @@ ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 	int hlen = sizeof (struct ip);
 	int mtu = 0;
 	int error = 0;
+	int vlan_pcp = -1;
 	struct sockaddr_in *dst, sin;
 	const struct sockaddr_in *gw;
 	struct in_ifaddr *ia = NULL;
@@ -345,6 +348,9 @@ ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 			m->m_pkthdr.flowid = inp->inp_flowid;
 			M_HASHTYPE_SET(m, inp->inp_flowtype);
 		}
+		if ((inp->inp_flags2 & INP_2PCP_SET) != 0)
+			vlan_pcp = (inp->inp_flags2 & INP_2PCP_MASK) >>
+			    INP_2PCP_SHIFT;
 #ifdef NUMA
 		m->m_pkthdr.numa_domain = inp->inp_numa_domain;
 #endif
@@ -717,6 +723,9 @@ sendit:
 		}
 	}
 
+	if (vlan_pcp > -1)
+		EVL_APPLY_PRI(m, vlan_pcp);
+
 	/* IN_LOOPBACK must not appear on the wire - RFC1122. */
 	if (IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) ||
 	    IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) {
@@ -1210,6 +1219,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 #ifdef	RSS
 		case IP_RECVRSSBUCKETID:
 #endif
+		case IP_VLAN_PCP:
 			error = sooptcopyin(sopt, &optval, sizeof optval,
 					    sizeof optval);
 			if (error)
@@ -1305,6 +1315,28 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 				OPTSET2(INP_RECVRSSBUCKETID, optval);
 				break;
 #endif
+			case IP_VLAN_PCP:
+				if ((optval >= -1) && (optval <=
+				    (INP_2PCP_MASK >> INP_2PCP_SHIFT))) {
+					if (optval == -1) {
+						INP_WLOCK(inp);
+						inp->inp_flags2 &=
+						    ~(INP_2PCP_SET |
+						      INP_2PCP_MASK);
+						INP_WUNLOCK(inp);
+					} else {
+						INP_WLOCK(inp);
+						inp->inp_flags2 |=
+						    INP_2PCP_SET;
+						inp->inp_flags2 &=
+						    ~INP_2PCP_MASK;
+						inp->inp_flags2 |=
+						    optval << INP_2PCP_SHIFT;
+						INP_WUNLOCK(inp);
+					}
+				} else
+					error = EINVAL;
+				break;
 			}
 			break;
 #undef OPTSET
@@ -1425,6 +1457,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 		case IP_RSSBUCKETID:
 		case IP_RECVRSSBUCKETID:
 #endif
+		case IP_VLAN_PCP:
 			switch (sopt->sopt_name) {
 			case IP_TOS:
 				optval = inp->inp_ip_tos;
@@ -1512,6 +1545,14 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 			case IP_BINDMULTI:
 				optval = OPTBIT2(INP_BINDMULTI);
 				break;
+			case IP_VLAN_PCP:
+				if (OPTBIT2(INP_2PCP_SET)) {
+					optval = (inp->inp_flags2 &
+					    INP_2PCP_MASK) >> INP_2PCP_SHIFT;
+				} else {
+					optval = -1;
+				}
+				break;
 			}
 			error = sooptcopyout(sopt, &optval, sizeof optval);
 			break;
diff --git a/sys/netinet6/in6.h b/sys/netinet6/in6.h
index 1092130919cb..ef4231098973 100644
--- a/sys/netinet6/in6.h
+++ b/sys/netinet6/in6.h
@@ -511,6 +511,10 @@ struct route_in6 {
 				    * set/get multicast source filter list.
 				    */
 
+/* The following option deals with the 802.1Q Ethernet Priority Code Point */
+#define	IPV6_VLAN_PCP		75  /* int; set/get PCP used for packet, */
+				    /*      -1 use interface default */
+
 /* to define items, should talk with KAME guys first, for *BSD compatibility */
 
 #define IPV6_RTHDR_LOOSE     0 /* this hop need not be a neighbor. XXX old spec */
diff --git a/sys/netinet6/ip6_output.c b/sys/netinet6/ip6_output.c
index 6be710085b87..59aabaa258f5 100644
--- a/sys/netinet6/ip6_output.c
+++ b/sys/netinet6/ip6_output.c
@@ -92,7 +92,9 @@ __FBSDID("$FreeBSD$");
 
 #include <net/if.h>
 #include <net/if_var.h>
+#include <net/if_vlan_var.h>
 #include <net/if_llatbl.h>
+#include <net/ethernet.h>
 #include <net/netisr.h>
 #include <net/route.h>
 #include <net/route/nhop.h>
@@ -436,6 +438,7 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts *opt,
 	u_char *nexthdrp;
 	int tlen, len;
 	int error = 0;
+	int vlan_pcp = -1;
 	struct in6_ifaddr *ia = NULL;
 	u_long mtu;
 	int alwaysfrag, dontfrag;
@@ -460,6 +463,9 @@ ip6_output(struct mbuf *m0, struct ip6_pktopts *opt,
 			m->m_pkthdr.flowid = inp->inp_flowid;
 			M_HASHTYPE_SET(m, inp->inp_flowtype);
 		}
+		if ((inp->inp_flags2 & INP_2PCP_SET) != 0)
+			vlan_pcp = (inp->inp_flags2 & INP_2PCP_MASK) >>
+			    INP_2PCP_SHIFT;
 #ifdef NUMA
 		m->m_pkthdr.numa_domain = inp->inp_numa_domain;
 #endif
@@ -1098,6 +1104,8 @@ nonh6lookup:
 	}
 
 passout:
+	if (vlan_pcp > -1)
+		EVL_APPLY_PRI(m, vlan_pcp);
 	/*
 	 * Send the packet to the outgoing interface.
 	 * If necessary, do IPv6 fragmentation before sending.
@@ -1265,6 +1273,8 @@ sendorfree:
 				counter_u64_add(ia->ia_ifa.ifa_obytes,
 				    m->m_pkthdr.len);
 			}
+			if (vlan_pcp > -1)
+				EVL_APPLY_PRI(m, vlan_pcp);
 			error = ip6_output_send(inp, ifp, origifp, m, dst, ro,
 			    true);
 		} else
@@ -1752,6 +1762,7 @@ ip6_ctloutput(struct socket *so, struct sockopt *sopt)
 #ifdef	RSS
 			case IPV6_RSS_LISTEN_BUCKET:
 #endif
+			case IPV6_VLAN_PCP:
 				if (optname == IPV6_BINDANY && td != NULL) {
 					error = priv_check(td,
 					    PRIV_NETINET_BINDANY);
@@ -1945,6 +1956,29 @@ do {									\
 					}
 					break;
 #endif
+				case IPV6_VLAN_PCP:
+					if ((optval >= -1) && (optval <=
+					    (INP_2PCP_MASK >> INP_2PCP_SHIFT))) {
+						if (optval == -1) {
+							INP_WLOCK(inp);
+							inp->inp_flags2 &=
+							    ~(INP_2PCP_SET |
+							    INP_2PCP_MASK);
+							INP_WUNLOCK(inp);
+						} else {
+							INP_WLOCK(inp);
+							inp->inp_flags2 |=
+							    INP_2PCP_SET;
+							inp->inp_flags2 &=
+							    ~INP_2PCP_MASK;
+							inp->inp_flags2 |=
+							    optval <<
+							    INP_2PCP_SHIFT;
+							INP_WUNLOCK(inp);
+						}
+					} else
+						error = EINVAL;
+					break;
 				}
 				break;
 
@@ -2168,6 +2202,7 @@ do {									\
 			case IPV6_RECVRSSBUCKETID:
 #endif
 			case IPV6_BINDMULTI:
+			case IPV6_VLAN_PCP:
 				switch (optname) {
 				case IPV6_RECVHOPOPTS:
 					optval = OPTBIT(IN6P_HOPOPTS);
@@ -2264,7 +2299,18 @@ do {									\
 				case IPV6_BINDMULTI:
 					optval = OPTBIT2(INP_BINDMULTI);
 					break;
+
+				case IPV6_VLAN_PCP:
+					if (OPTBIT2(INP_2PCP_SET)) {
+						optval = (inp->inp_flags2 &
+							    INP_2PCP_MASK) >>
+							    INP_2PCP_SHIFT;
+					} else {
+						optval = -1;
+					}
+					break;
 				}
+
 				if (error)
 					break;
 				error = sooptcopyout(sopt, &optval,