zebra: add a runtime flag to enable vrf with netns

The netns backend is chosen by VRF if a runtime flag named vrfwnetns is
selected when running zebra.
In the case the NETNS backend is chosen, in some case the VRFID value is
being assigned the value of the NSID. Within the perimeter of that work,
this is why the vrf_lookup_by_table function is extended with a new
parameter.

Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>

7 files changed, 65 insertions, 20 deletions

diff --git a/lib/ns.c b/lib/ns.c
index 0b2a3bec7..e3a1d9d0d 100644
--- a/lib/ns.c
+++ b/lib/ns.c
@@ -378,7 +378,9 @@ static int ns_config_write(struct vty *vty)
 	struct ns *ns;
 	int write = 0;
 
-	RB_FOREACH (ns, ns_head, &ns_tree) {
+	if (vrf_is_backend_netns())
+		return 0;
+	RB_FOREACH(ns, ns_head, &ns_tree) {
 		if (ns->ns_id == NS_DEFAULT || ns->name == NULL)
 			continue;
 
@@ -411,7 +413,7 @@ void ns_init(void)
 		exit(1);
 	}
 
-	if (have_netns()) {
+	if (have_netns() && !vrf_is_backend_netns()) {
 		/* Install NS commands. */
 		install_node(&ns_node, ns_config_write);
 		install_element(CONFIG_NODE, &ns_netns_cmd);
diff --git a/lib/vrf.c b/lib/vrf.c
index 02946df2b..c300a87a3 100644
--- a/lib/vrf.c
+++ b/lib/vrf.c
@@ -44,6 +44,8 @@ RB_GENERATE(vrf_name_head, vrf, name_entry, vrf_name_compare);
 struct vrf_id_head vrfs_by_id = RB_INITIALIZER(&vrfs_by_id);
 struct vrf_name_head vrfs_by_name = RB_INITIALIZER(&vrfs_by_name);
 
+static int vrf_backend;
+
 /*
  * Turn on/off debug code
  * for vrf.
@@ -446,6 +448,21 @@ int vrf_socket(int domain, int type, int protocol, vrf_id_t vrf_id)
 	return ret;
 }
 
+int vrf_is_backend_netns(void)
+{
+	return (vrf_backend == VRF_BACKEND_NETNS);
+}
+
+int vrf_get_backend(void)
+{
+	return vrf_backend;
+}
+
+void vrf_configure_backend(int vrf_backend_netns)
+{
+	vrf_backend = vrf_backend_netns;
+}
+
 /* vrf CLI commands */
 DEFUN_NOSH (vrf,
        vrf_cmd,
diff --git a/lib/vrf.h b/lib/vrf.h
index 99c048c70..f1dc45019 100644
--- a/lib/vrf.h
+++ b/lib/vrf.h
@@ -96,6 +96,9 @@ RB_HEAD(vrf_name_head, vrf);
 RB_PROTOTYPE(vrf_name_head, vrf, name_entry, vrf_name_compare)
 DECLARE_QOBJ_TYPE(vrf)
 
+/* Allow VRF with netns as backend */
+#define VRF_BACKEND_VRF_LITE 0
+#define VRF_BACKEND_NETNS    1
 
 extern struct vrf_id_head vrfs_by_id;
 extern struct vrf_name_head vrfs_by_name;
@@ -203,6 +206,9 @@ extern void vrf_cmd_init(int (*writefunc)(struct vty *vty));
 
 /* Create a socket serving for the given VRF */
 extern int vrf_socket(int, int, int, vrf_id_t);
+extern void vrf_configure_backend(int vrf_backend_netns);
+extern int vrf_get_backend(void);
+extern int vrf_is_backend_netns(void);
 
 /*
  * VRF Debugging
diff --git a/zebra/if_netlink.c b/zebra/if_netlink.c
index 14905b738..ef30c7830 100644
--- a/zebra/if_netlink.c
+++ b/zebra/if_netlink.c
@@ -615,13 +615,14 @@ static int netlink_interface(struct sockaddr_nl *snl, struct nlmsghdr *h,
 	}
 
 	/* If VRF, create the VRF structure itself. */
-	if (zif_type == ZEBRA_IF_VRF) {
+	if (zif_type == ZEBRA_IF_VRF && !vrf_is_backend_netns()) {
 		netlink_vrf_change(h, tb[IFLA_LINKINFO], name);
 		vrf_id = (vrf_id_t)ifi->ifi_index;
 	}
 
 	if (tb[IFLA_MASTER]) {
-		if (slave_kind && (strcmp(slave_kind, "vrf") == 0)) {
+		if (slave_kind && (strcmp(slave_kind, "vrf") == 0)
+		    && !vrf_is_backend_netns()) {
 			zif_slave_type = ZEBRA_IF_SLAVE_VRF;
 			vrf_id = *(u_int32_t *)RTA_DATA(tb[IFLA_MASTER]);
 		} else if (slave_kind && (strcmp(slave_kind, "bridge") == 0)) {
@@ -631,6 +632,8 @@ static int netlink_interface(struct sockaddr_nl *snl, struct nlmsghdr *h,
 		} else
 			zif_slave_type = ZEBRA_IF_SLAVE_OTHER;
 	}
+	if (vrf_is_backend_netns())
+		vrf_id = (vrf_id_t)ns_id;
 
 	/* If linking to another interface, note it. */
 	if (tb[IFLA_LINK])
@@ -1074,7 +1077,7 @@ int netlink_link_change(struct sockaddr_nl *snl, struct nlmsghdr *h,
 	}
 
 	/* If VRF, create or update the VRF structure itself. */
-	if (zif_type == ZEBRA_IF_VRF) {
+	if (zif_type == ZEBRA_IF_VRF && !vrf_is_backend_netns()) {
 		netlink_vrf_change(h, tb[IFLA_LINKINFO], name);
 		vrf_id = (vrf_id_t)ifi->ifi_index;
 	}
@@ -1091,7 +1094,8 @@ int netlink_link_change(struct sockaddr_nl *snl, struct nlmsghdr *h,
 
 	if (h->nlmsg_type == RTM_NEWLINK) {
 		if (tb[IFLA_MASTER]) {
-			if (slave_kind && (strcmp(slave_kind, "vrf") == 0)) {
+			if (slave_kind && (strcmp(slave_kind, "vrf") == 0)
+			    && !vrf_is_backend_netns()) {
 				zif_slave_type = ZEBRA_IF_SLAVE_VRF;
 				vrf_id =
 					*(u_int32_t *)RTA_DATA(tb[IFLA_MASTER]);
diff --git a/zebra/main.c b/zebra/main.c
index 19b16936d..a881fcb9c 100644
--- a/zebra/main.c
+++ b/zebra/main.c
@@ -85,6 +85,7 @@ struct option longopts[] = {{"batch", no_argument, NULL, 'b'},
 			    {"label_socket", no_argument, NULL, 'l'},
 			    {"retain", no_argument, NULL, 'r'},
 #ifdef HAVE_NETLINK
+			    {"vrfwnetns", no_argument, NULL, 'n'},
 			    {"nl-bufsize", required_argument, NULL, 's'},
 #endif /* HAVE_NETLINK */
 			    {0}};
@@ -205,12 +206,14 @@ int main(int argc, char **argv)
 	char *fuzzing = NULL;
 #endif
 
+	vrf_configure_backend(VRF_BACKEND_VRF_LITE);
+
 	frr_preinit(&zebra_di, argc, argv);
 
 	frr_opt_add(
 		"bakz:e:l:r"
 #ifdef HAVE_NETLINK
-		"s:"
+		"s:n"
 #endif
 #if defined(HANDLE_ZAPI_FUZZING)
 		"c:"
@@ -225,6 +228,7 @@ int main(int argc, char **argv)
 		"  -k, --keep_kernel  Don't delete old routes which installed by zebra.\n"
 		"  -r, --retain       When program terminates, retain added route by zebra.\n"
 #ifdef HAVE_NETLINK
+		"  -n, --vrfwnetns    Set VRF with NetNS\n"
 		"  -s, --nl-bufsize   Set netlink receive buffer size\n"
 #endif /* HAVE_NETLINK */
 #if defined(HANDLE_ZAPI_FUZZING)
@@ -279,6 +283,9 @@ int main(int argc, char **argv)
 		case 's':
 			nl_rcvbufsize = atoi(optarg);
 			break;
+		case 'n':
+			vrf_configure_backend(VRF_BACKEND_NETNS);
+			break;
 #endif /* HAVE_NETLINK */
 #if defined(HANDLE_ZAPI_FUZZING)
 		case 'c':
diff --git a/zebra/rt_netlink.c b/zebra/rt_netlink.c
index a80ab9d83..20abd7697 100644
--- a/zebra/rt_netlink.c
+++ b/zebra/rt_netlink.c
@@ -194,16 +194,25 @@ static inline int proto2zebra(int proto, int family)
 /*
 Pending: create an efficient table_id (in a tree/hash) based lookup)
  */
-static vrf_id_t vrf_lookup_by_table(u_int32_t table_id)
+static vrf_id_t vrf_lookup_by_table(u_int32_t table_id, ns_id_t ns_id)
 {
 	struct vrf *vrf;
 	struct zebra_vrf *zvrf;
 
 	RB_FOREACH (vrf, vrf_id_head, &vrfs_by_id) {
-		if ((zvrf = vrf->info) == NULL || (zvrf->table_id != table_id))
+		zvrf = vrf->info;
+		if (zvrf == NULL)
 			continue;
-
-		return zvrf_id(zvrf);
+		/* case vrf with netns : match the netnsid */
+		if (vrf_is_backend_netns()) {
+			if (ns_id == zvrf_id(zvrf))
+				return zvrf_id(zvrf);
+		} else {
+			/* VRF is VRF_BACKEND_VRF_LITE */
+			if (zvrf->table_id != table_id)
+				continue;
+			return zvrf_id(zvrf);
+		}
 	}
 
 	return VRF_DEFAULT;
@@ -220,7 +229,7 @@ static int netlink_route_change_read_unicast(struct sockaddr_nl *snl,
 	u_char flags = 0;
 	struct prefix p;
 	struct prefix_ipv6 src_p = {};
-	vrf_id_t vrf_id = VRF_DEFAULT;
+	vrf_id_t vrf_id;
 
 	char anyaddr[16] = {0};
 
@@ -288,7 +297,7 @@ static int netlink_route_change_read_unicast(struct sockaddr_nl *snl,
 		table = rtm->rtm_table;
 
 	/* Map to VRF */
-	vrf_id = vrf_lookup_by_table(table);
+	vrf_id = vrf_lookup_by_table(table, ns_id);
 	if (vrf_id == VRF_DEFAULT) {
 		if (!is_zebra_valid_kernel_table(table)
 		    && !is_zebra_main_routing_table(table))
@@ -609,7 +618,7 @@ static int netlink_route_change_read_multicast(struct sockaddr_nl *snl,
 	char sbuf[40];
 	char gbuf[40];
 	char oif_list[256] = "\0";
-	vrf_id_t vrf = ns_id;
+	vrf_id_t vrf;
 	int table;
 
 	if (mroute)
@@ -631,7 +640,7 @@ static int netlink_route_change_read_multicast(struct sockaddr_nl *snl,
 	else
 		table = rtm->rtm_table;
 
-	vrf = vrf_lookup_by_table(table);
+	vrf = vrf_lookup_by_table(table, ns_id);
 
 	if (tb[RTA_IIF])
 		iif = *(int *)RTA_DATA(tb[RTA_IIF]);
@@ -687,24 +696,23 @@ int netlink_route_change(struct sockaddr_nl *snl, struct nlmsghdr *h,
 			 ns_id_t ns_id, int startup)
 {
 	int len;
-	vrf_id_t vrf_id = ns_id;
 	struct rtmsg *rtm;
 
 	rtm = NLMSG_DATA(h);
 
 	if (!(h->nlmsg_type == RTM_NEWROUTE || h->nlmsg_type == RTM_DELROUTE)) {
 		/* If this is not route add/delete message print warning. */
-		zlog_warn("Kernel message: %d vrf %u\n", h->nlmsg_type, vrf_id);
+		zlog_warn("Kernel message: %d NS %u\n", h->nlmsg_type, ns_id);
 		return 0;
 	}
 
 	/* Connected route. */
 	if (IS_ZEBRA_DEBUG_KERNEL)
-		zlog_debug("%s %s %s proto %s vrf %u",
+		zlog_debug("%s %s %s proto %s NS %u",
 			   nl_msg_type_to_str(h->nlmsg_type),
 			   nl_family_to_str(rtm->rtm_family),
 			   nl_rttype_to_str(rtm->rtm_type),
-			   nl_rtproto_to_str(rtm->rtm_protocol), vrf_id);
+			   nl_rtproto_to_str(rtm->rtm_protocol), ns_id);
 
 	/* We don't care about change notifications for the MPLS table. */
 	/* TODO: Revisit this. */
diff --git a/zebra/zebra_vrf.c b/zebra/zebra_vrf.c
index cd47f2127..6eec2c18c 100644
--- a/zebra/zebra_vrf.c
+++ b/zebra/zebra_vrf.c
@@ -25,8 +25,9 @@
 #include "command.h"
 #include "memory.h"
 #include "srcdest_table.h"
-
+#include "vrf.h"
 #include "vty.h"
+
 #include "zebra/debug.h"
 #include "zebra/zserv.h"
 #include "zebra/rib.h"