ospfd: support write socket per interface

Add support for a write socket per interface, enabled by
default at the ospf instance level. An ospf instance-level
config allows this to be disabled, reverting to the older
behavior where a single per-instance socket is used for
sending and receiving packets.

Signed-off-by: Mark Stapp <mjs@labn.net>

8 files changed, 179 insertions, 34 deletions

diff --git a/ospfd/ospf_interface.c b/ospfd/ospf_interface.c
index 649ba70e02..5742ece1f7 100644
--- a/ospfd/ospf_interface.c
+++ b/ospfd/ospf_interface.c
@@ -651,6 +651,8 @@ int ospf_if_new_hook(struct interface *ifp)
 
 	ifp->info = XCALLOC(MTYPE_OSPF_IF_INFO, sizeof(struct ospf_if_info));
 
+	IF_OSPF_IF_INFO(ifp)->oii_fd = -1;
+
 	IF_OIFS(ifp) = route_table_init();
 	IF_OIFS_PARAMS(ifp) = route_table_init();
 
@@ -691,6 +693,8 @@ static int ospf_if_delete_hook(struct interface *ifp)
 {
 	int rc = 0;
 	struct route_node *rn;
+	struct ospf_if_info *oii;
+
 	rc = ospf_opaque_del_if(ifp);
 
 	/*
@@ -707,6 +711,13 @@ static int ospf_if_delete_hook(struct interface *ifp)
 	route_table_finish(IF_OIFS(ifp));
 	route_table_finish(IF_OIFS_PARAMS(ifp));
 
+	/* Close per-interface socket */
+	oii = ifp->info;
+	if (oii && oii->oii_fd > 0) {
+		close(oii->oii_fd);
+		oii->oii_fd = -1;
+	}
+
 	XFREE(MTYPE_OSPF_IF_INFO, ifp->info);
 
 	return rc;
@@ -1367,6 +1378,16 @@ static int ospf_ifp_up(struct interface *ifp)
 	struct ospf_interface *oi;
 	struct route_node *rn;
 	struct ospf_if_info *oii = ifp->info;
+	struct ospf *ospf;
+
+	if (IS_DEBUG_OSPF(zebra, ZEBRA_INTERFACE))
+		zlog_debug("Zebra: Interface[%s] state change to up.",
+			   ifp->name);
+
+	/* Open per-intf write socket if configured */
+	ospf = ifp->vrf->info;
+	if (ospf && ospf->intf_socket_enabled)
+		ospf_ifp_sock_init(ifp);
 
 	ospf_if_recalculate_output_cost(ifp);
 
@@ -1384,10 +1405,6 @@ static int ospf_ifp_up(struct interface *ifp)
 		return 0;
 	}
 
-	if (IS_DEBUG_OSPF(zebra, ZEBRA_INTERFACE))
-		zlog_debug("Zebra: Interface[%s] state change to up.",
-			   ifp->name);
-
 	for (rn = route_top(IF_OIFS(ifp)); rn; rn = route_next(rn)) {
 		if ((oi = rn->info) == NULL)
 			continue;
@@ -1416,6 +1433,9 @@ static int ospf_ifp_down(struct interface *ifp)
 		ospf_if_down(oi);
 	}
 
+	/* Close per-interface write socket if configured */
+	ospf_ifp_sock_close(ifp);
+
 	return 0;
 }
 
diff --git a/ospfd/ospf_interface.h b/ospfd/ospf_interface.h
index 8625a72ac1..649df437a4 100644
--- a/ospfd/ospf_interface.h
+++ b/ospfd/ospf_interface.h
@@ -121,6 +121,9 @@ struct ospf_if_info {
 		membership_counts[MEMBER_MAX]; /* multicast group refcnts */
 
 	uint32_t curr_mtu;
+
+	/* Per-interface write socket, configured via 'ospf' object */
+	int oii_fd;
 };
 
 struct ospf_interface;
diff --git a/ospfd/ospf_network.c b/ospfd/ospf_network.c
index bd5cd7682a..aff8ed05c7 100644
--- a/ospfd/ospf_network.c
+++ b/ospfd/ospf_network.c
@@ -15,6 +15,7 @@
 #include "sockopt.h"
 #include "privs.h"
 #include "lib_errors.h"
+#include "lib/table.h"
 
 #include "ospfd/ospfd.h"
 #include "ospfd/ospf_network.h"
@@ -119,61 +120,60 @@ int ospf_if_drop_alldrouters(struct ospf *top, struct prefix *p,
 	return ret;
 }
 
-int ospf_if_ipmulticast(struct ospf *top, struct prefix *p, ifindex_t ifindex)
+int ospf_if_ipmulticast(int fd, struct prefix *p, ifindex_t ifindex)
 {
 	uint8_t val;
 	int ret, len;
 
 	/* Prevent receiving self-origined multicast packets. */
-	ret = setsockopt_ipv4_multicast_loop(top->fd, 0);
+	ret = setsockopt_ipv4_multicast_loop(fd, 0);
 	if (ret < 0)
 		flog_err(EC_LIB_SOCKET,
 			 "can't setsockopt IP_MULTICAST_LOOP(0) for fd %d: %s",
-			 top->fd, safe_strerror(errno));
+			 fd, safe_strerror(errno));
 
 	/* Explicitly set multicast ttl to 1 -- endo. */
 	val = 1;
 	len = sizeof(val);
-	ret = setsockopt(top->fd, IPPROTO_IP, IP_MULTICAST_TTL, (void *)&val,
-			 len);
+	ret = setsockopt(fd, IPPROTO_IP, IP_MULTICAST_TTL, (void *)&val, len);
 	if (ret < 0)
 		flog_err(EC_LIB_SOCKET,
 			 "can't setsockopt IP_MULTICAST_TTL(1) for fd %d: %s",
-			 top->fd, safe_strerror(errno));
+			 fd, safe_strerror(errno));
 #ifndef GNU_LINUX
 	/* For GNU LINUX ospf_write uses IP_PKTINFO, in_pktinfo to send
 	 * packet out of ifindex. Below would be used Non Linux system.
 	 */
-	ret = setsockopt_ipv4_multicast_if(top->fd, p->u.prefix4, ifindex);
+	ret = setsockopt_ipv4_multicast_if(fd, p->u.prefix4, ifindex);
 	if (ret < 0)
 		flog_err(EC_LIB_SOCKET,
 			 "can't setsockopt IP_MULTICAST_IF(fd %d, addr %pI4, ifindex %u): %s",
-			 top->fd, &p->u.prefix4, ifindex,
+			 fd, &p->u.prefix4, ifindex,
 			 safe_strerror(errno));
 #endif
 
 	return ret;
 }
 
-int ospf_sock_init(struct ospf *ospf)
+/*
+ * Helper to open and set up a socket; returns the new fd on success,
+ * -1 on error.
+ */
+static int sock_init_common(vrf_id_t vrf_id, const char *name, int *pfd)
 {
 	int ospf_sock;
 	int ret, hincl = 1;
 
-	/* silently ignore. already done */
-	if (ospf->fd > 0)
-		return -1;
-
-	if (ospf->vrf_id == VRF_UNKNOWN) {
+	if (vrf_id == VRF_UNKNOWN) {
 		/* silently return since VRF is not ready */
 		return -1;
 	}
+
 	frr_with_privs(&ospfd_privs) {
 		ospf_sock = vrf_socket(AF_INET, SOCK_RAW, IPPROTO_OSPFIGP,
-				       ospf->vrf_id, ospf->name);
+				       vrf_id, name);
 		if (ospf_sock < 0) {
-			flog_err(EC_LIB_SOCKET,
-				 "ospf_read_sock_init: socket: %s",
+			flog_err(EC_LIB_SOCKET, "%s: socket: %s", __func__,
 				 safe_strerror(errno));
 			return -1;
 		}
@@ -212,10 +212,8 @@ int ospf_sock_init(struct ospf *ospf)
 				 ospf_sock);
 	}
 
-	/* Update socket buffer sizes */
-	ospf_sock_bufsize_update(ospf, ospf_sock, OSPF_SOCK_BOTH);
+	*pfd = ospf_sock;
 
-	ospf->fd = ospf_sock;
 	return ret;
 }
 
@@ -237,3 +235,79 @@ void ospf_sock_bufsize_update(const struct ospf *ospf, int sock,
 		setsockopt_so_sendbuf(sock, bufsize);
 	}
 }
+
+int ospf_sock_init(struct ospf *ospf)
+{
+	int ret;
+
+	/* silently ignore. already done */
+	if (ospf->fd > 0)
+		return -1;
+
+	ret = sock_init_common(ospf->vrf_id, ospf->name, &(ospf->fd));
+
+	if (ret >= 0) /* Update socket buffer sizes */
+		ospf_sock_bufsize_update(ospf, ospf->fd, OSPF_SOCK_BOTH);
+
+	return ret;
+}
+
+/*
+ * Open per-interface write socket
+ */
+int ospf_ifp_sock_init(struct interface *ifp)
+{
+	struct ospf_if_info *oii;
+	struct ospf_interface *oi;
+	struct ospf *ospf;
+	struct route_node *rn;
+	int ret;
+
+	oii = IF_OSPF_IF_INFO(ifp);
+	if (oii == NULL)
+		return -1;
+
+	if (oii->oii_fd > 0)
+		return 0;
+
+	rn = route_top(IF_OIFS(ifp));
+	if (rn && rn->info) {
+		oi = rn->info;
+		ospf = oi->ospf;
+	} else
+		return -1;
+
+	ret = sock_init_common(ifp->vrf->vrf_id, ifp->name, &oii->oii_fd);
+
+	if (ret >= 0) /* Update socket buffer sizes */
+		ospf_sock_bufsize_update(ospf, oii->oii_fd, OSPF_SOCK_BOTH);
+
+	if (IS_DEBUG_OSPF_EVENT)
+		zlog_debug("%s: ifp %s, oii %p, fd %d", __func__, ifp->name,
+			   oii, oii->oii_fd);
+
+	return ret;
+}
+
+/*
+ * Close per-interface write socket
+ */
+int ospf_ifp_sock_close(struct interface *ifp)
+{
+	struct ospf_if_info *oii;
+
+	oii = IF_OSPF_IF_INFO(ifp);
+	if (oii == NULL)
+		return 0;
+
+	if (oii->oii_fd > 0) {
+		if (IS_DEBUG_OSPF_EVENT)
+			zlog_debug("%s: ifp %s, oii %p, fd %d", __func__,
+				   ifp->name, oii, oii->oii_fd);
+
+		close(oii->oii_fd);
+		oii->oii_fd = -1;
+	}
+
+	return 0;
+}
diff --git a/ospfd/ospf_network.h b/ospfd/ospf_network.h
index d9b579c04a..b810bad50b 100644
--- a/ospfd/ospf_network.h
+++ b/ospfd/ospf_network.h
@@ -13,8 +13,11 @@ extern int ospf_if_drop_allspfrouters(struct ospf *, struct prefix *,
 				      ifindex_t);
 extern int ospf_if_add_alldrouters(struct ospf *, struct prefix *, ifindex_t);
 extern int ospf_if_drop_alldrouters(struct ospf *, struct prefix *, ifindex_t);
-extern int ospf_if_ipmulticast(struct ospf *, struct prefix *, ifindex_t);
+extern int ospf_if_ipmulticast(int fd, struct prefix *, ifindex_t);
 extern int ospf_sock_init(struct ospf *ospf);
+/* Open, close per-interface write socket */
+int ospf_ifp_sock_init(struct interface *ifp);
+int ospf_ifp_sock_close(struct interface *ifp);
 
 enum ospf_sock_type_e {
 	OSPF_SOCK_NONE = 0,
diff --git a/ospfd/ospf_packet.c b/ospfd/ospf_packet.c
index 5f7d49e0bb..552acfd6d3 100644
--- a/ospfd/ospf_packet.c
+++ b/ospfd/ospf_packet.c
@@ -618,7 +618,7 @@ static void ospf_write(struct event *thread)
 	struct msghdr msg;
 	struct iovec iov[2];
 	uint8_t type;
-	int ret;
+	int ret, fd;
 	int flags = 0;
 	struct listnode *node;
 #ifdef WANT_OSPF_WRITE_FRAGMENT
@@ -633,11 +633,12 @@ static void ospf_write(struct event *thread)
 	struct cmsghdr *cm = (struct cmsghdr *)cmsgbuf;
 	struct in_pktinfo *pi;
 #endif
+	fd = ospf->fd;
 
-	if (ospf->fd < 0 || ospf->oi_running == 0) {
+	if (fd < 0 || ospf->oi_running == 0) {
 		if (IS_DEBUG_OSPF_EVENT)
 			zlog_debug("%s failed to send, fd %d, instance %u",
-				   __func__, ospf->fd, ospf->oi_running);
+				   __func__, fd, ospf->oi_running);
 		return;
 	}
 
@@ -657,6 +658,15 @@ static void ospf_write(struct event *thread)
 		/* convenience - max OSPF data per packet */
 		maxdatasize = oi->ifp->mtu - sizeof(struct ip);
 #endif /* WANT_OSPF_WRITE_FRAGMENT */
+
+		/* Reset socket fd to use. */
+		fd = ospf->fd;
+
+		/* Check for per-interface socket */
+		if (ospf->intf_socket_enabled &&
+		    (IF_OSPF_IF_INFO(oi->ifp))->oii_fd > 0)
+			fd = (IF_OSPF_IF_INFO(oi->ifp))->oii_fd;
+
 		/* Get one packet from queue. */
 		op = ospf_fifo_head(oi->obuf);
 		assert(op);
@@ -664,8 +674,7 @@ static void ospf_write(struct event *thread)
 
 		if (op->dst.s_addr == htonl(OSPF_ALLSPFROUTERS)
 		    || op->dst.s_addr == htonl(OSPF_ALLDROUTERS))
-			ospf_if_ipmulticast(ospf, oi->address,
-					    oi->ifp->ifindex);
+			ospf_if_ipmulticast(fd, oi->address, oi->ifp->ifindex);
 
 		/* Rewrite the md5 signature & update the seq */
 		ospf_make_md5_digest(oi, op);
@@ -760,13 +769,13 @@ static void ospf_write(struct event *thread)
 
 #ifdef WANT_OSPF_WRITE_FRAGMENT
 		if (op->length > maxdatasize)
-			ospf_write_frags(ospf->fd, op, &iph, &msg, maxdatasize,
+			ospf_write_frags(fd, op, &iph, &msg, maxdatasize,
 					 oi->ifp->mtu, flags, type);
 #endif /* WANT_OSPF_WRITE_FRAGMENT */
 
 		/* send final fragment (could be first) */
 		sockopt_iphdrincl_swab_htosys(&iph);
-		ret = sendmsg(ospf->fd, &msg, flags);
+		ret = sendmsg(fd, &msg, flags);
 		sockopt_iphdrincl_swab_systoh(&iph);
 		if (IS_DEBUG_OSPF_EVENT)
 			zlog_debug(
diff --git a/ospfd/ospf_vty.c b/ospfd/ospf_vty.c
index 9b918798c0..3c0e0fcb63 100644
--- a/ospfd/ospf_vty.c
+++ b/ospfd/ospf_vty.c
@@ -35,12 +35,11 @@
 #include "ospfd/ospf_spf.h"
 #include "ospfd/ospf_route.h"
 #include "ospfd/ospf_zebra.h"
-/*#include "ospfd/ospf_routemap.h" */
 #include "ospfd/ospf_vty.h"
 #include "ospfd/ospf_dump.h"
 #include "ospfd/ospf_bfd.h"
 #include "ospfd/ospf_ldp_sync.h"
-
+#include "ospfd/ospf_network.h"
 
 FRR_CFG_DEFAULT_BOOL(OSPF_LOG_ADJACENCY_CHANGES,
 	{ .val_bool = true, .match_profile = "datacenter", },
@@ -12511,6 +12510,9 @@ static int ospf_config_write_one(struct vty *vty, struct ospf *ospf)
 	if (ospf->fr_configured)
 		vty_out(vty, " flood-reduction\n");
 
+	if (!ospf->intf_socket_enabled)
+		vty_out(vty, " no socket-per-interface\n");
+
 	/* Redistribute information print. */
 	config_write_ospf_redistribute(vty, ospf);
 
@@ -13075,6 +13077,35 @@ DEFPY(ospf_socket_bufsizes,
 	return CMD_SUCCESS;
 }
 
+DEFPY (per_intf_socket,
+       per_intf_socket_cmd,
+       "[no] socket-per-interface",
+       NO_STR
+       "Use write socket per interface\n")
+{
+	VTY_DECLVAR_INSTANCE_CONTEXT(ospf, ospf);
+	struct listnode *node;
+	struct ospf_interface *oi;
+
+	if (no) {
+		if (ospf->intf_socket_enabled) {
+			ospf->intf_socket_enabled = false;
+
+			/* Iterate and close any sockets */
+			for (ALL_LIST_ELEMENTS_RO(ospf->oiflist, node, oi))
+				ospf_ifp_sock_close(oi->ifp);
+		}
+	} else if (!ospf->intf_socket_enabled) {
+		ospf->intf_socket_enabled = true;
+
+		/* Iterate and open sockets */
+		for (ALL_LIST_ELEMENTS_RO(ospf->oiflist, node, oi))
+			ospf_ifp_sock_init(oi->ifp);
+	}
+
+	return CMD_SUCCESS;
+}
+
 void ospf_vty_clear_init(void)
 {
 	install_element(ENABLE_NODE, &clear_ip_ospf_interface_cmd);
@@ -13239,6 +13270,7 @@ void ospf_vty_init(void)
 	install_element(OSPF_NODE, &no_flood_reduction_area_cmd);
 
 	install_element(OSPF_NODE, &ospf_socket_bufsizes_cmd);
+	install_element(OSPF_NODE, &per_intf_socket_cmd);
 
 	/* Init interface related vty commands. */
 	ospf_vty_if_init();
diff --git a/ospfd/ospfd.c b/ospfd/ospfd.c
index 15ce1c48a4..7e83714c0a 100644
--- a/ospfd/ospfd.c
+++ b/ospfd/ospfd.c
@@ -419,6 +419,7 @@ struct ospf *ospf_new_alloc(unsigned short instance, const char *name)
 	QOBJ_REG(new, ospf);
 
 	new->fd = -1;
+	new->intf_socket_enabled = true;
 
 	new->recv_sock_bufsize = OSPF_DEFAULT_SOCK_BUFSIZE;
 	new->send_sock_bufsize = OSPF_DEFAULT_SOCK_BUFSIZE;
diff --git a/ospfd/ospfd.h b/ospfd/ospfd.h
index 1b66d4d930..1f8d1a32e6 100644
--- a/ospfd/ospfd.h
+++ b/ospfd/ospfd.h
@@ -431,6 +431,9 @@ struct ospf {
 	uint32_t recv_sock_bufsize;
 	uint32_t send_sock_bufsize;
 
+	/* Per-interface write socket */
+	bool intf_socket_enabled;
+
 	QOBJ_FIELDS;
 };
 DECLARE_QOBJ_TYPE(ospf);