pim6d: pim6d socket changes

socket changes to support IPv6 PIM

Signed-off-by: Balaji Gurudoss <G_Balaji1@dell.com>
[DL: cleaned up & refactored a whole bunch more.]
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>

1 files changed, 194 insertions, 224 deletions

diff --git a/pimd/pim_sock.c b/pimd/pim_sock.c
index 8d3fbabd27..92e2d18451 100644
--- a/pimd/pim_sock.c
+++ b/pimd/pim_sock.c
@@ -34,22 +34,27 @@
 #include "vrf.h"
 #include "sockopt.h"
 #include "lib_errors.h"
+#include "network.h"
 
 #include "pimd.h"
 #include "pim_mroute.h"
 #include "pim_sock.h"
 #include "pim_str.h"
 
-/* GLOBAL VARS */
+#if PIM_IPV == 4
+#define setsockopt_iptos setsockopt_ipv4_tos
+#define setsockopt_multicast_loop setsockopt_ipv4_multicast_loop
+#else
+#define setsockopt_iptos setsockopt_ipv6_tclass
+#define setsockopt_multicast_loop setsockopt_ipv6_multicast_loop
+#endif
 
 int pim_socket_raw(int protocol)
 {
 	int fd;
 
 	frr_with_privs(&pimd_privs) {
-
-		fd = socket(AF_INET, SOCK_RAW, protocol);
-
+		fd = socket(PIM_AF, SOCK_RAW, protocol);
 	}
 
 	if (fd < 0) {
@@ -66,10 +71,16 @@ void pim_socket_ip_hdr(int fd)
 	const int on = 1;
 
 	frr_with_privs(&pimd_privs) {
-
+#if PIM_IPV == 4
 		if (setsockopt(fd, IPPROTO_IP, IP_HDRINCL, &on, sizeof(on)))
-			zlog_err("%s: Could not turn on IP_HDRINCL option: %s",
-				 __func__, safe_strerror(errno));
+			zlog_err("%s: Could not turn on IP_HDRINCL option: %m",
+				 __func__);
+#else
+		if (setsockopt(fd, IPPROTO_IPV6, IPV6_HDRINCL, &on, sizeof(on)))
+			zlog_err(
+				"%s: Could not turn on IPV6_HDRINCL option: %m",
+				__func__);
+#endif
 	}
 }
 
@@ -80,248 +91,252 @@ void pim_socket_ip_hdr(int fd)
 int pim_socket_bind(int fd, struct interface *ifp)
 {
 	int ret = 0;
-#ifdef SO_BINDTODEVICE
 
+#ifdef SO_BINDTODEVICE
 	frr_with_privs(&pimd_privs) {
-
 		ret = setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, ifp->name,
 				 strlen(ifp->name));
-
 	}
-
 #endif
 	return ret;
 }
 
-int pim_socket_mcast(int protocol, struct in_addr ifaddr, struct interface *ifp,
-		     uint8_t loop)
+#if PIM_IPV == 4
+static inline int pim_setsockopt(int protocol, int fd, struct interface *ifp)
 {
-	int rcvbuf = 1024 * 1024 * 8;
-#ifdef HAVE_STRUCT_IP_MREQN_IMR_IFINDEX
-	struct ip_mreqn mreq;
-#else
-	struct ip_mreq mreq;
-#endif
-	int fd;
+	int one = 1;
+	int ttl = 1;
 
-	fd = pim_socket_raw(protocol);
-	if (fd < 0) {
-		zlog_warn("Could not create multicast socket: errno=%d: %s",
-			  errno, safe_strerror(errno));
-		return PIM_SOCK_ERR_SOCKET;
-	}
-
-#ifdef SO_BINDTODEVICE
-	int ret;
-
-	ret = pim_socket_bind(fd, ifp);
-	if (ret) {
-		close(fd);
-		zlog_warn(
-			"Could not set fd: %d for interface: %s to device",
-			fd, ifp->name);
-		return PIM_SOCK_ERR_BIND;
-	}
-#else
-/* XXX: use IP_PKTINFO / IP_RECVIF to emulate behaviour?  Or change to
- * only use 1 socket for all interfaces? */
-#endif
-
-	/* Needed to obtain destination address from recvmsg() */
-	{
 #if defined(HAVE_IP_PKTINFO)
-		/* Linux and Solaris IP_PKTINFO */
-		int opt = 1;
-		if (setsockopt(fd, IPPROTO_IP, IP_PKTINFO, &opt, sizeof(opt))) {
-			zlog_warn(
-				"Could not set IP_PKTINFO on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
-		}
+	/* Linux and Solaris IP_PKTINFO */
+	if (setsockopt(fd, IPPROTO_IP, IP_PKTINFO, &one, sizeof(one)))
+		zlog_warn("Could not set PKTINFO on socket fd=%d: %m", fd);
 #elif defined(HAVE_IP_RECVDSTADDR)
-		/* BSD IP_RECVDSTADDR */
-		int opt = 1;
-		if (setsockopt(fd, IPPROTO_IP, IP_RECVDSTADDR, &opt,
-			       sizeof(opt))) {
-			zlog_warn(
-				"Could not set IP_RECVDSTADDR on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
-		}
+	/* BSD IP_RECVDSTADDR */
+	if (setsockopt(fd, IPPROTO_IP, IP_RECVDSTADDR, &one, sizeof(one)))
+		zlog_warn("Could not set IP_RECVDSTADDR on socket fd=%d: %m",
+			  fd);
 #else
-		flog_err(
-			EC_LIB_DEVELOPMENT,
-			"%s %s: Missing IP_PKTINFO and IP_RECVDSTADDR: unable to get dst addr from recvmsg()",
-			__FILE__, __func__);
-		close(fd);
-		return PIM_SOCK_ERR_DSTADDR;
+	flog_err(
+		EC_LIB_DEVELOPMENT,
+		"Missing IP_PKTINFO and IP_RECVDSTADDR: unable to get dst addr from recvmsg()");
+	close(fd);
+	return PIM_SOCK_ERR_DSTADDR;
 #endif
-	}
-
 
-	/* Set router alert (RFC 2113) for all IGMP messages (RFC 3376 4.
-	 * Message Formats)*/
+	/* Set router alert (RFC 2113) for all IGMP messages (RFC
+	 * 3376 4. Message Formats)*/
 	if (protocol == IPPROTO_IGMP) {
 		uint8_t ra[4];
+
 		ra[0] = 148;
 		ra[1] = 4;
 		ra[2] = 0;
 		ra[3] = 0;
 		if (setsockopt(fd, IPPROTO_IP, IP_OPTIONS, ra, 4)) {
 			zlog_warn(
-				"Could not set Router Alert Option on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
+				"Could not set Router Alert Option on socket fd=%d: %m",
+				fd);
 			close(fd);
 			return PIM_SOCK_ERR_RA;
 		}
 	}
 
-	{
-		int reuse = 1;
-		if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (void *)&reuse,
-			       sizeof(reuse))) {
-			zlog_warn(
-				"Could not set Reuse Address Option on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
-			close(fd);
-			return PIM_SOCK_ERR_REUSE;
-		}
-	}
-
-	{
-		const int MTTL = 1;
-		int ttl = MTTL;
-		if (setsockopt(fd, IPPROTO_IP, IP_MULTICAST_TTL, (void *)&ttl,
-			       sizeof(ttl))) {
-			zlog_warn(
-				"Could not set multicast TTL=%d on socket fd=%d: errno=%d: %s",
-				MTTL, fd, errno, safe_strerror(errno));
-			close(fd);
-			return PIM_SOCK_ERR_TTL;
-		}
+	if (setsockopt(fd, IPPROTO_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl))) {
+		zlog_warn("Could not set multicast TTL=%d on socket fd=%d: %m",
+			  ttl, fd);
+		close(fd);
+		return PIM_SOCK_ERR_TTL;
 	}
 
-	if (setsockopt_ipv4_multicast_loop(fd, loop)) {
+	if (setsockopt_ipv4_multicast_if(fd, PIMADDR_ANY, ifp->ifindex)) {
 		zlog_warn(
-			"Could not %s Multicast Loopback Option on socket fd=%d: errno=%d: %s",
-			loop ? "enable" : "disable", fd, errno,
-			safe_strerror(errno));
+			"Could not set Outgoing Interface Option on socket fd=%d: %m",
+			fd);
 		close(fd);
-		return PIM_SOCK_ERR_LOOP;
+		return PIM_SOCK_ERR_IFACE;
 	}
 
-	memset(&mreq, 0, sizeof(mreq));
-#ifdef HAVE_STRUCT_IP_MREQN_IMR_IFINDEX
-	mreq.imr_ifindex = ifp->ifindex;
-#else
-/*
- * I am not sure what to do here yet for *BSD
- */
-// mreq.imr_interface = ifindex;
-#endif
+	return 0;
+}
+#else /* PIM_IPV != 4 */
+static inline int pim_setsockopt(int protocol, int fd, struct interface *ifp)
+{
+	int ttl = 1;
+	struct ipv6_mreq mreq = {};
+
+	setsockopt_ipv6_pktinfo(fd, 1);
+	setsockopt_ipv6_multicast_hops(fd, ttl);
 
-	if (setsockopt(fd, IPPROTO_IP, IP_MULTICAST_IF, (void *)&mreq,
+	mreq.ipv6mr_interface = ifp->ifindex;
+	if (setsockopt(fd, IPPROTO_IPV6, IPV6_MULTICAST_IF, &mreq,
 		       sizeof(mreq))) {
 		zlog_warn(
-			"Could not set Outgoing Interface Option on socket fd=%d: errno=%d: %s",
-			fd, errno, safe_strerror(errno));
+			"Could not set Outgoing Interface Option on socket fd=%d: %m",
+			fd);
 		close(fd);
 		return PIM_SOCK_ERR_IFACE;
 	}
 
-	if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, sizeof(rcvbuf)))
-		zlog_warn("%s: Failure to set buffer size to %d", __func__,
-			  rcvbuf);
+	return 0;
+}
+#endif
 
-	{
-		long flags;
+int pim_socket_mcast(int protocol, pim_addr ifaddr, struct interface *ifp,
+		     uint8_t loop)
+{
+	int fd;
+	int ret;
 
-		flags = fcntl(fd, F_GETFL, 0);
-		if (flags < 0) {
-			zlog_warn(
-				"Could not get fcntl(F_GETFL,O_NONBLOCK) on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
-			close(fd);
-			return PIM_SOCK_ERR_NONBLOCK_GETFL;
-		}
+	fd = pim_socket_raw(protocol);
+	if (fd < 0) {
+		zlog_warn("Could not create multicast socket: errno=%d: %s",
+			  errno, safe_strerror(errno));
+		return PIM_SOCK_ERR_SOCKET;
+	}
 
-		if (fcntl(fd, F_SETFL, flags | O_NONBLOCK)) {
-			zlog_warn(
-				"Could not set fcntl(F_SETFL,O_NONBLOCK) on socket fd=%d: errno=%d: %s",
-				fd, errno, safe_strerror(errno));
-			close(fd);
-			return PIM_SOCK_ERR_NONBLOCK_SETFL;
-		}
+	/* XXX: if SO_BINDTODEVICE isn't available, use IP_PKTINFO / IP_RECVIF
+	 * to emulate behaviour?  Or change to only use 1 socket for all
+	 * interfaces? */
+	ret = pim_socket_bind(fd, ifp);
+	if (ret) {
+		close(fd);
+		zlog_warn("Could not set fd: %d for interface: %s to device",
+			  fd, ifp->name);
+		return PIM_SOCK_ERR_BIND;
 	}
 
-	/* Set Tx socket DSCP byte */
-	if (setsockopt_ipv4_tos(fd, IPTOS_PREC_INTERNETCONTROL)) {
-		zlog_warn("can't set sockopt IP_TOS to PIM/IGMP socket %d: %s",
-			  fd, safe_strerror(errno));
+	set_nonblocking(fd);
+	sockopt_reuseaddr(fd);
+	setsockopt_so_recvbuf(fd, 8 * 1024 * 1024);
+
+	ret = pim_setsockopt(protocol, fd, ifp);
+	if (ret) {
+		zlog_warn("pim_setsockopt failed for interface: %s to device ",
+			  ifp->name);
+		return ret;
 	}
 
+	/* leftover common sockopts */
+	if (setsockopt_multicast_loop(fd, loop)) {
+		zlog_warn(
+			"Could not %s Multicast Loopback Option on socket fd=%d: %m",
+			loop ? "enable" : "disable", fd);
+		close(fd);
+		return PIM_SOCK_ERR_LOOP;
+	}
+
+	/* Set Tx socket DSCP byte */
+	if (setsockopt_iptos(fd, IPTOS_PREC_INTERNETCONTROL))
+		zlog_warn("can't set sockopt IP[V6]_TOS to socket %d: %m", fd);
+
 	return fd;
 }
 
-int pim_socket_join(int fd, struct in_addr group, struct in_addr ifaddr,
-		    ifindex_t ifindex)
+int pim_socket_join(int fd, pim_addr group, pim_addr ifaddr, ifindex_t ifindex)
 {
 	int ret;
 
-#ifdef HAVE_STRUCT_IP_MREQN_IMR_IFINDEX
-	struct ip_mreqn opt;
+#if PIM_IPV == 4
+	ret = setsockopt_ipv4_multicast(fd, IP_ADD_MEMBERSHIP, ifaddr,
+					group.s_addr, ifindex);
 #else
-	struct ip_mreq opt;
-#endif
-
-	opt.imr_multiaddr = group;
+	struct ipv6_mreq opt;
 
-#ifdef HAVE_STRUCT_IP_MREQN_IMR_IFINDEX
-	opt.imr_address = ifaddr;
-	opt.imr_ifindex = ifindex;
-#else
-	opt.imr_interface = ifaddr;
+	memcpy(&opt.ipv6mr_multiaddr, &group, 16);
+	opt.ipv6mr_interface = ifindex;
+	ret = setsockopt(fd, IPPROTO_IPV6, IPV6_JOIN_GROUP, &opt, sizeof(opt));
 #endif
 
-	ret = setsockopt(fd, IPPROTO_IP, IP_ADD_MEMBERSHIP, &opt, sizeof(opt));
 	if (ret) {
-		char group_str[INET_ADDRSTRLEN];
-		char ifaddr_str[INET_ADDRSTRLEN];
-		if (!inet_ntop(AF_INET, &group, group_str, sizeof(group_str)))
-			snprintf(group_str, sizeof(group_str), "<group?>");
-		if (!inet_ntop(AF_INET, &ifaddr, ifaddr_str,
-			       sizeof(ifaddr_str)))
-			snprintf(ifaddr_str, sizeof(ifaddr_str), "<ifaddr?>");
-
 		flog_err(
 			EC_LIB_SOCKET,
-			"Failure socket joining fd=%d group %s on interface address %s: errno=%d: %s",
-			fd, group_str, ifaddr_str, errno, safe_strerror(errno));
+			"Failure socket joining fd=%d group %pPAs on interface address %pPAs: %m",
+			fd, &group, &ifaddr);
 		return ret;
 	}
 
-	if (PIM_DEBUG_TRACE) {
-		char group_str[INET_ADDRSTRLEN];
-		char ifaddr_str[INET_ADDRSTRLEN];
-		if (!inet_ntop(AF_INET, &group, group_str, sizeof(group_str)))
-			snprintf(group_str, sizeof(group_str), "<group?>");
-		if (!inet_ntop(AF_INET, &ifaddr, ifaddr_str,
-			       sizeof(ifaddr_str)))
-			snprintf(ifaddr_str, sizeof(ifaddr_str), "<ifaddr?>");
-
+	if (PIM_DEBUG_TRACE)
 		zlog_debug(
-			"Socket fd=%d joined group %s on interface address %s",
-			fd, group_str, ifaddr_str);
+			"Socket fd=%d joined group %pPAs on interface address %pPAs",
+			fd, &group, &ifaddr);
+	return ret;
+}
+
+#if PIM_IPV == 4
+static void cmsg_getdstaddr(struct msghdr *mh, struct sockaddr_storage *dst,
+			    ifindex_t *ifindex)
+{
+	struct cmsghdr *cmsg;
+	struct sockaddr_in *dst4 = (struct sockaddr_in *)dst;
+
+	for (cmsg = CMSG_FIRSTHDR(mh); cmsg != NULL;
+	     cmsg = CMSG_NXTHDR(mh, cmsg)) {
+#ifdef HAVE_IP_PKTINFO
+		if ((cmsg->cmsg_level == IPPROTO_IP) &&
+		    (cmsg->cmsg_type == IP_PKTINFO)) {
+			struct in_pktinfo *i;
+
+			i = (struct in_pktinfo *)CMSG_DATA(cmsg);
+			if (dst4)
+				dst4->sin_addr = i->ipi_addr;
+			if (ifindex)
+				*ifindex = i->ipi_ifindex;
+
+			break;
+		}
+#endif
+
+#ifdef HAVE_IP_RECVDSTADDR
+		if ((cmsg->cmsg_level == IPPROTO_IP) &&
+		    (cmsg->cmsg_type == IP_RECVDSTADDR)) {
+			struct in_addr *i = (struct in_addr *)CMSG_DATA(cmsg);
+
+			if (dst4)
+				dst4->sin_addr = *i;
+
+			break;
+		}
+#endif
+
+#if defined(HAVE_IP_RECVIF) && defined(CMSG_IFINDEX)
+		if (cmsg->cmsg_type == IP_RECVIF)
+			if (ifindex)
+				*ifindex = CMSG_IFINDEX(cmsg);
+#endif
 	}
+}
+#else  /* PIM_IPV != 4 */
+static void cmsg_getdstaddr(struct msghdr *mh, struct sockaddr_storage *dst,
+			    ifindex_t *ifindex)
+{
+	struct cmsghdr *cmsg;
+	struct sockaddr_in6 *dst6 = (struct sockaddr_in6 *)dst;
 
-	return ret;
+	for (cmsg = CMSG_FIRSTHDR(mh); cmsg != NULL;
+	     cmsg = CMSG_NXTHDR(mh, cmsg)) {
+		if ((cmsg->cmsg_level == IPPROTO_IPV6) &&
+		    (cmsg->cmsg_type == IPV6_PKTINFO)) {
+			struct in6_pktinfo *i;
+
+			i = (struct in6_pktinfo *)CMSG_DATA(cmsg);
+
+			if (dst6)
+				dst6->sin6_addr = i->ipi6_addr;
+			if (ifindex)
+				*ifindex = i->ipi6_ifindex;
+			break;
+		}
+	}
 }
+#endif /* PIM_IPV != 4 */
 
 int pim_socket_recvfromto(int fd, uint8_t *buf, size_t len,
-			  struct sockaddr_in *from, socklen_t *fromlen,
-			  struct sockaddr_in *to, socklen_t *tolen,
+			  struct sockaddr_storage *from, socklen_t *fromlen,
+			  struct sockaddr_storage *to, socklen_t *tolen,
 			  ifindex_t *ifindex)
 {
 	struct msghdr msgh;
-	struct cmsghdr *cmsg;
 	struct iovec iov;
 	char cbuf[1000];
 	int err;
@@ -331,19 +346,12 @@ int pim_socket_recvfromto(int fd, uint8_t *buf, size_t len,
 	 * Use getsockname() to get sin_port.
 	 */
 	if (to) {
-		struct sockaddr_in si;
-		socklen_t si_len = sizeof(si);
-
-		memset(&si, 0, sizeof(si));
-		to->sin_family = AF_INET;
-
-		pim_socket_getsockname(fd, (struct sockaddr *)&si, &si_len);
+		socklen_t to_len = sizeof(*to);
 
-		to->sin_port = si.sin_port;
-		to->sin_addr = si.sin_addr;
+		pim_socket_getsockname(fd, (struct sockaddr *)to, &to_len);
 
 		if (tolen)
-			*tolen = sizeof(si);
+			*tolen = sizeof(*to);
 	}
 
 	memset(&msgh, 0, sizeof(struct msghdr));
@@ -364,45 +372,7 @@ int pim_socket_recvfromto(int fd, uint8_t *buf, size_t len,
 	if (fromlen)
 		*fromlen = msgh.msg_namelen;
 
-	for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
-	     cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
-
-#ifdef HAVE_IP_PKTINFO
-		if ((cmsg->cmsg_level == IPPROTO_IP)
-		    && (cmsg->cmsg_type == IP_PKTINFO)) {
-			struct in_pktinfo *i =
-				(struct in_pktinfo *)CMSG_DATA(cmsg);
-			if (to)
-				to->sin_addr = i->ipi_addr;
-			if (tolen)
-				*tolen = sizeof(struct sockaddr_in);
-			if (ifindex)
-				*ifindex = i->ipi_ifindex;
-
-			break;
-		}
-#endif
-
-#ifdef HAVE_IP_RECVDSTADDR
-		if ((cmsg->cmsg_level == IPPROTO_IP)
-		    && (cmsg->cmsg_type == IP_RECVDSTADDR)) {
-			struct in_addr *i = (struct in_addr *)CMSG_DATA(cmsg);
-			if (to)
-				to->sin_addr = *i;
-			if (tolen)
-				*tolen = sizeof(struct sockaddr_in);
-
-			break;
-		}
-#endif
-
-#if defined(HAVE_IP_RECVIF) && defined(CMSG_IFINDEX)
-		if (cmsg->cmsg_type == IP_RECVIF)
-			if (ifindex)
-				*ifindex = CMSG_IFINDEX(cmsg);
-#endif
-
-	} /* for (cmsg) */
+	cmsg_getdstaddr(&msgh, to, ifindex);
 
 	return err; /* len */
 }