From 3cca425b2916f7d7b2ed9e757c2ab48b99843044 Mon Sep 17 00:00:00 2001
From: Michael Tuexen <tuexen@FreeBSD.org>
Date: Tue, 12 Jun 2012 14:02:38 +0000
Subject: [PATCH] Add a IP_RECVTOS socket option to receive for received
 UDP/IPv4 packets a cmsg of type IP_RECVTOS which contains the TOS byte. Much
 like IP_RECVTTL does for TTL. This allows to implement a protocol on top of
 UDP and implementing ECN.

MFC after: 3 days
---
 share/man/man4/ip.4     | 25 ++++++++++++++++++++++++-
 sys/netinet/in.h        |  1 +
 sys/netinet/in_pcb.c    |  4 ++++
 sys/netinet/in_pcb.h    |  3 ++-
 sys/netinet/ip_input.c  |  6 ++++++
 sys/netinet/ip_output.c |  8 ++++++++
 6 files changed, 45 insertions(+), 2 deletions(-)

diff --git a/share/man/man4/ip.4 b/share/man/man4/ip.4
index 96b17c70473c..d89b54d93c04 100644
--- a/share/man/man4/ip.4
+++ b/share/man/man4/ip.4
@@ -32,7 +32,7 @@
 .\"     @(#)ip.4	8.2 (Berkeley) 11/30/93
 .\" $FreeBSD$
 .\"
-.Dd November 14, 2011
+.Dd June 12, 2012
 .Dt IP 4
 .Os
 .Sh NAME
@@ -286,6 +286,29 @@ cmsg_type = IP_RECVTTL
 .\"
 .Pp
 If the
+.Dv IP_RECVTOS
+option is enabled on a
+.Dv SOCK_DGRAM
+socket, the
+.Xr recvmsg 2
+call will return the
+.Tn IP
+.Tn TOS
+(type of service) field for a
+.Tn UDP
+datagram.
+The msg_control field in the msghdr structure points to a buffer
+that contains a cmsghdr structure followed by the
+.Tn TOS .
+The cmsghdr fields have the following values:
+.Bd -literal
+cmsg_len = CMSG_LEN(sizeof(u_char))
+cmsg_level = IPPROTO_IP
+cmsg_type = IP_RECVTOS
+.Ed
+.\"
+.Pp
+If the
 .Dv IP_RECVIF
 option is enabled on a
 .Dv SOCK_DGRAM
diff --git a/sys/netinet/in.h b/sys/netinet/in.h
index 56150dad49cc..40f595295860 100644
--- a/sys/netinet/in.h
+++ b/sys/netinet/in.h
@@ -462,6 +462,7 @@ __END_DECLS
 #define	IP_RECVTTL		65   /* bool; receive IP TTL w/dgram */
 #define	IP_MINTTL		66   /* minimum TTL for packet or drop */
 #define	IP_DONTFRAG		67   /* don't fragment packet */
+#define	IP_RECVTOS		68   /* bool; receive IP TOS w/dgram */
 
 /* IPv4 Source Filter Multicast API [RFC3678] */
 #define	IP_ADD_SOURCE_MEMBERSHIP	70   /* join a source-specific group */
diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index e509610470cd..dd1460f585b7 100644
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -2295,6 +2295,10 @@ db_print_inpflags(int inp_flags)
 		db_printf("%sINP_DONTFRAG", comma ? ", " : "");
 		comma = 1;
 	}
+	if (inp_flags & INP_RECVTOS) {
+		db_printf("%sINP_RECVTOS", comma ? ", " : "");
+		comma = 1;
+	}
 	if (inp_flags & IN6P_IPV6_V6ONLY) {
 		db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
 		comma = 1;
diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index c17213b11068..4945b0f17942 100644
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -509,6 +509,7 @@ void 	inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
 #define	INP_DONTFRAG		0x00000800 /* don't fragment packet */
 #define	INP_BINDANY		0x00001000 /* allow bind to any address */
 #define	INP_INHASHLIST		0x00002000 /* in_pcbinshash() has been called */
+#define	INP_RECVTOS		0x00004000 /* receive incoming IP TOS */
 #define	IN6P_IPV6_V6ONLY	0x00008000 /* restrict AF_INET6 socket for v6 */
 #define	IN6P_PKTINFO		0x00010000 /* receive IP6 dst and I/F */
 #define	IN6P_HOPLIMIT		0x00020000 /* receive hoplimit */
@@ -528,7 +529,7 @@ void 	inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
 #define	IN6P_MTU		0x80000000 /* receive path MTU */
 
 #define	INP_CONTROLOPTS		(INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\
-				 INP_RECVIF|INP_RECVTTL|\
+				 INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\
 				 IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\
 				 IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\
 				 IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\
diff --git a/sys/netinet/ip_input.c b/sys/netinet/ip_input.c
index 728b50b558a8..ab78fa235c97 100644
--- a/sys/netinet/ip_input.c
+++ b/sys/netinet/ip_input.c
@@ -1684,6 +1684,12 @@ ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip,
 		if (*mp)
 			mp = &(*mp)->m_next;
 	}
+	if (inp->inp_flags & INP_RECVTOS) {
+		*mp = sbcreatecontrol((caddr_t) &ip->ip_tos,
+		    sizeof(u_char), IP_RECVTOS, IPPROTO_IP);
+		if (*mp)
+			mp = &(*mp)->m_next;
+	}
 }
 
 /*
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index a9008c525afa..d13c39719ce7 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -984,6 +984,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 		case IP_FAITH:
 		case IP_ONESBCAST:
 		case IP_DONTFRAG:
+		case IP_RECVTOS:
 			error = sooptcopyin(sopt, &optval, sizeof optval,
 					    sizeof optval);
 			if (error)
@@ -1047,6 +1048,9 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 			case IP_BINDANY:
 				OPTSET(INP_BINDANY);
 				break;
+			case IP_RECVTOS:
+				OPTSET(INP_RECVTOS);
+				break;
 			}
 			break;
 #undef OPTSET
@@ -1156,6 +1160,7 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 		case IP_ONESBCAST:
 		case IP_DONTFRAG:
 		case IP_BINDANY:
+		case IP_RECVTOS:
 			switch (sopt->sopt_name) {
 
 			case IP_TOS:
@@ -1214,6 +1219,9 @@ ip_ctloutput(struct socket *so, struct sockopt *sopt)
 			case IP_BINDANY:
 				optval = OPTBIT(INP_BINDANY);
 				break;
+			case IP_RECVTOS:
+				optval = OPTBIT(INP_RECVTOS);
+				break;
 			}
 			error = sooptcopyout(sopt, &optval, sizeof optval);
 			break;