Ifindex nexthop

api/gr_net_types.h

@@ -23,12 +23,15 @@
 #endif
 
 typedef enum : uint8_t {
+	GR_AF_UNSPEC = AF_UNSPEC,
 	GR_AF_IP4 = AF_INET,
 	GR_AF_IP6 = AF_INET6,
 } addr_family_t;
 
 static inline const char *gr_af_name(addr_family_t af) {
 	switch (af) {
+	case GR_AF_UNSPEC:
+		return "Unspec";
 	case GR_AF_IP4:
 		return "IPv4";
 	case GR_AF_IP6:

frr/if_grout.c

@@ -174,6 +174,10 @@ void grout_interface_addr_dplane(struct gr_nexthop *gr_nh, bool new) {
 		p.family = AF_INET6;
 		memcpy(&p.u.prefix6, &gr_nh->ipv6, sizeof(p.u.prefix6));
 		break;
+	case GR_AF_UNSPEC:
+		gr_log_err("interface with addr family UNSPEC are not supported");
+		dplane_ctx_fini(&ctx);
+		return;
 	}
 	dplane_ctx_set_intf_addr(ctx, &p);
 	dplane_ctx_set_intf_metric(ctx, METRIC_MAX);

frr/rt_grout.c

@@ -227,6 +227,10 @@ static int grout_gr_nexthop_to_frr_nexthop(
 		*nh_family = AF_INET6;
 		memcpy(&nh->gate.ipv6, &gr_nh->ipv6, sz);
 		break;
+	case GR_AF_UNSPEC:
+		nh->type = NEXTHOP_TYPE_IFINDEX;
+		*nh_family = AF_UNSPEC;
+		break;
 	default:
 		gr_log_debug("inval nexthop family %u, nexthop not sync", gr_nh->af);
 		return -1;
@@ -294,7 +298,7 @@ static void grout_route_change(
 			return;
 		}
 
-		if (nh_family != family) {
+		if (nh_family != AF_UNSPEC && nh_family != family) {
 			gr_log_debug(
 				"nexthop family %u different that route family %u nexthop, "
 				"ignoring",
@@ -850,8 +854,11 @@ enum zebra_dplane_result grout_add_del_nexthop(struct zebra_dplane_ctx *ctx) {
 		gr_log_debug("add nexthop id %u gw %pI6", nh_id, &gr_nh->ipv6);
 		break;
 	case NEXTHOP_TYPE_IFINDEX:
-		gr_log_debug("add nexthop id %u ifindex %u", nh_id, gr_nh->iface_id);
-		return ZEBRA_DPLANE_REQUEST_FAILURE;
+		// dplane_ctx_get_nhe_afi(ctx) returns AFI_IP for a nexthop with no gateway
+		// force to UNSPEC for grout
+		gr_nh->af = GR_AF_UNSPEC;
+		gr_log_debug("add nexthop id %u with ifindex %u", nh_id, gr_nh->iface_id);
+		break;
 	default:
 		gr_log_err("impossible to add nexthop %u (type %u not supported)", nh_id, nh->type);
 		return ZEBRA_DPLANE_REQUEST_FAILURE;
@@ -890,6 +897,10 @@ void grout_nexthop_change(bool new, struct gr_nexthop *gr_nh) {
 		return;
 	}
 
+	// kernel set INET4 when no gateway, let's do the same
+	if (family == AF_UNSPEC)
+		family = AF_INET;
+
 	afi = family2afi(family);
 	type = origin2zebra(gr_nh->origin, family, false);
 	SET_FLAG(nh.flags, NEXTHOP_FLAG_ACTIVE);

frr/zebra_dplane_grout.c

@@ -315,6 +315,8 @@ static void dplane_read_notifications(struct event *event) {
 				gr_evt_to_str(gr_e->ev_type)
 			);
 			break;
+		case GR_AF_UNSPEC:
+			break;
 		}
 
 		grout_interface_addr_dplane(gr_nh, new);

main/gr_id_pool.h

@@ -0,0 +1,102 @@
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright (c) 2025 Maxime Leroy, Free Mobile
+// Simple ID allocator built on top of DPDK’s rte_bitmap
+// bit 1 ⇒ ID free
+// bit 0 ⇒ ID reserved
+// single writer, any number of readers
+
+#pragma once
+
+#include <gr_errno.h>
+
+#include <rte_bitmap.h>
+#include <rte_malloc.h>
+
+#include <stdint.h>
+
+struct gr_id_pool {
+	uint32_t max_ids;
+	uint32_t used;
+	// struct rte_bitmap must be aligned on a cache line
+	// it's required by rte_bitmap library.
+	struct rte_bitmap bmp __rte_cache_aligned;
+};
+
+static inline struct gr_id_pool *gr_id_pool_create(const char *id_pool_name, uint32_t max_ids) {
+	size_t p_size, b_size;
+	struct gr_id_pool *p;
+
+	b_size = rte_bitmap_get_memory_footprint(max_ids);
+	p_size = sizeof(struct gr_id_pool) - sizeof(struct rte_bitmap) + b_size;
+	p = rte_zmalloc(id_pool_name, p_size, RTE_CACHE_LINE_SIZE);
+	if (!p)
+		return NULL;
+
+	if (!rte_bitmap_init_with_all_set(max_ids, (uint8_t *)&p->bmp, b_size)) {
+		rte_free(p);
+		return NULL;
+	}
+
+	p->max_ids = max_ids;
+	p->used = 0;
+	return p;
+}
+
+static inline void gr_id_pool_destroy(struct gr_id_pool *p) {
+	if (!p)
+		return;
+
+	rte_free(p);
+}
+
+static inline uint32_t gr_id_pool_used(struct gr_id_pool *p) {
+	return p->used;
+}
+
+static inline uint32_t gr_id_pool_avail(struct gr_id_pool *p) {
+	return p->max_ids - p->used;
+}
+
+// Get the lowest‑numbered free ID; 0 if none
+static inline uint32_t gr_id_pool_get(struct gr_id_pool *p) {
+	uint64_t slab = 0;
+	uint32_t pos = 0;
+	uint32_t bit;
+
+	__rte_bitmap_scan_init(&p->bmp);
+	if (!rte_bitmap_scan(&p->bmp, &pos, &slab))
+		return 0; // pool is full
+
+	bit = pos + rte_ctz64(slab);
+	rte_bitmap_clear(&p->bmp, bit); // mark used
+
+	p->used++;
+	return bit + 1;
+}
+
+// Reserve a user‑chosen ID. Returns 0 on success, <0 on error
+static inline int gr_id_pool_book(struct gr_id_pool *p, uint32_t id) {
+	if (id == 0 || id > p->max_ids)
+		return errno_set(EINVAL);
+
+	// already used
+	if (!rte_bitmap_get(&p->bmp, id - 1))
+		return errno_set(EEXIST);
+
+	rte_bitmap_clear(&p->bmp, id - 1);
+	p->used++;
+	return 0;
+}
+
+// Put an ID back to the pool
+static inline int gr_id_pool_put(struct gr_id_pool *p, uint32_t id) {
+	if (id == 0 || id > p->max_ids)
+		return errno_set(EINVAL);
+
+	if (rte_bitmap_get(&p->bmp, id - 1))
+		return errno_set(EALREADY);
+
+	rte_bitmap_set(&p->bmp, id - 1);
+	p->used--;
+	return 0;
+}

modules/infra/api/nexthop.c

@@ -47,14 +47,25 @@ static struct api_out nh_add(const void *request, void ** /*response*/) {
 
 	int ret;
 
+	base.flags = 0;
 	switch (base.af) {
 	case GR_AF_IP4:
 		if (base.ipv4 == 0)
 			return api_out(EDESTADDRREQ, 0);
+
+		base.flags |= GR_NH_F_GATEWAY;
 		break;
 	case GR_AF_IP6:
 		if (rte_ipv6_addr_is_unspec(&base.ipv6))
 			return api_out(EDESTADDRREQ, 0);
+
+		base.flags |= GR_NH_F_GATEWAY;
+		break;
+	case GR_AF_UNSPEC:
+		if (base.ipv4 || !rte_ipv6_addr_is_unspec(&base.ipv6))
+			return api_out(EINVAL, 0);
+
+		base.flags |= GR_NH_F_LINK | GR_NH_F_STATIC;
 		break;
 	default:
 		return api_out(ENOPROTOOPT, 0);
@@ -67,10 +78,12 @@ static struct api_out nh_add(const void *request, void ** /*response*/) {
 
 	base.type = GR_NH_T_L3;
 	base.state = GR_NH_S_NEW;
-	base.flags = 0;
 	if (!rte_is_zero_ether_addr(&base.mac)) {
+		if (base.af == GR_AF_UNSPEC)
+			return api_out(EINVAL, 0);
+
 		base.state = GR_NH_S_REACHABLE;
-		base.flags = GR_NH_F_STATIC;
+		base.flags |= GR_NH_F_STATIC;
 	}
 
 	if (base.nh_id != GR_NH_ID_UNSET)
@@ -132,35 +145,11 @@ static struct api_out nh_del(const void *request, void ** /*response*/) {
 	const struct nexthop_af_ops *ops;
 	struct nexthop *nh;
 
-	if (req->nh.nh_id != GR_NH_ID_UNSET) {
-		nh = nexthop_lookup_by_id(req->nh.nh_id);
-		if (nh == NULL) {
-			if (req->missing_ok)
-				return api_out(0, 0);
-			return api_out(ENOENT, 0);
-		}
-	} else {
-		switch (req->nh.af) {
-		case GR_AF_IP4:
-			if (req->nh.ipv4 == 0)
-				return api_out(EDESTADDRREQ, 0);
-			break;
-		case GR_AF_IP6:
-			if (rte_ipv6_addr_is_unspec(&req->nh.ipv6))
-				return api_out(EDESTADDRREQ, 0);
-			break;
-		default:
-			return api_out(ENOPROTOOPT, 0);
-		}
-		if (req->nh.vrf_id >= MAX_VRFS)
-			return api_out(EOVERFLOW, 0);
-
-		nh = nexthop_lookup(req->nh.af, req->nh.vrf_id, req->nh.iface_id, &req->nh.addr);
-		if (nh == NULL) {
-			if (errno == ENOENT && req->missing_ok)
-				return api_out(0, 0);
-			return api_out(errno, 0);
-		}
+	nh = nexthop_lookup_by_id(req->nh.nh_id);
+	if (nh == NULL) {
+		if (req->missing_ok)
+			return api_out(0, 0);
+		return api_out(ENOENT, 0);
 	}
 
 	if ((nh->flags & (GR_NH_F_LOCAL | GR_NH_F_LINK | GR_NH_F_GATEWAY)) || nh->ref_count > 1)

modules/infra/cli/nexthop.c

@@ -62,6 +62,9 @@ static cmd_status_t nh_add(const struct gr_api_client *c, const struct ec_pnode
 	struct gr_nh_add_req req = {.exist_ok = true, .nh.origin = GR_NH_ORIGIN_USER};
 	struct gr_iface iface;
 
+	if (arg_u32(p, "ID", &req.nh.nh_id) < 0 && errno != ENOENT)
+		return CMD_ERROR;
+
 	switch (arg_ip4(p, "IP", &req.nh.ipv4)) {
 	case 0:
 		req.nh.af = GR_AF_IP4;
@@ -72,16 +75,15 @@ static cmd_status_t nh_add(const struct gr_api_client *c, const struct ec_pnode
 		req.nh.af = GR_AF_IP6;
 		break;
 	default:
-		return CMD_ERROR;
+		req.nh.af = GR_AF_UNSPEC;
+		break;
 	}
+
 	if (iface_from_name(c, arg_str(p, "IFACE"), &iface) < 0)
 		return CMD_ERROR;
 	req.nh.iface_id = iface.id;
 	req.nh.vrf_id = iface.vrf_id;
 
-	if (arg_u32(p, "ID", &req.nh.nh_id) < 0 && errno != ENOENT)
-		return CMD_ERROR;
-
 	if (arg_eth_addr(p, "MAC", &req.nh.mac) < 0 && errno != ENOENT)
 		return CMD_ERROR;
 
@@ -94,19 +96,9 @@ static cmd_status_t nh_add(const struct gr_api_client *c, const struct ec_pnode
 static cmd_status_t nh_del(const struct gr_api_client *c, const struct ec_pnode *p) {
 	struct gr_nh_del_req req = {.missing_ok = true};
 
-	switch (arg_ip4(p, "IP", &req.nh.ipv4)) {
-	case 0:
-		req.nh.af = GR_AF_IP4;
-		break;
-	case -EINVAL:
-		if (arg_ip6(p, "IP", &req.nh.ipv6) < 0)
-			return CMD_ERROR;
-		req.nh.af = GR_AF_IP6;
-		break;
-	default:
-		if (arg_u32(p, "ID", &req.nh.nh_id) < 0 && errno != ENOENT)
-			return CMD_ERROR;
-	}
+	if (arg_u32(p, "ID", &req.nh.nh_id) < 0)
+		return CMD_ERROR;
+
 	if (arg_u16(p, "VRF", &req.nh.vrf_id) < 0 && errno != ENOENT)
 		return CMD_ERROR;
 
@@ -164,7 +156,10 @@ static cmd_status_t nh_list(const struct gr_api_client *c, const struct ec_pnode
 			scols_line_set_data(line, 1, "");
 		scols_line_sprintf(line, 2, "%s", gr_nh_type_name(nh));
 		scols_line_sprintf(line, 3, "%s", gr_af_name(nh->af));
-		scols_line_sprintf(line, 4, ADDR_F, ADDR_W(nh->af), &nh->addr);
+		if (nh->af == GR_AF_UNSPEC)
+			scols_line_set_data(line, 4, "");
+		else
+			scols_line_sprintf(line, 4, ADDR_F, ADDR_W(nh->af), &nh->addr);
 
 		if (nh->state == GR_NH_S_REACHABLE)
 			scols_line_sprintf(line, 5, ETH_F, &nh->mac);
@@ -251,10 +246,10 @@ static int ctx_init(struct ec_node *root) {
 
 	ret = CLI_COMMAND(
 		CLI_CONTEXT(root, CTX_ADD),
-		"nexthop IP iface IFACE [id ID] [mac MAC]",
+		"nexthop [id ID] [address IP] iface IFACE [mac MAC]",
 		nh_add,
 		"Add a new next hop.",
-		with_help("IPv4 address.", ec_node_re("IP", IP_ANY_RE)),
+		with_help("IPv4/6 address.", ec_node_re("IP", IP_ANY_RE)),
 		with_help("Ethernet address.", ec_node_re("MAC", ETH_ADDR_RE)),
 		with_help("Nexthop ID.", ec_node_uint("ID", 1, UINT32_MAX - 1, 10)),
 		with_help("Output interface.", ec_node_dyn("IFACE", complete_iface_names, NULL))
@@ -263,10 +258,9 @@ static int ctx_init(struct ec_node *root) {
 		return ret;
 	ret = CLI_COMMAND(
 		CLI_CONTEXT(root, CTX_DEL),
-		"nexthop (id ID)|IP [vrf VRF]",
+		"nexthop ID [vrf VRF]",
 		nh_del,
 		"Delete a next hop.",
-		with_help("IPv4 address.", ec_node_re("IP", IP_ANY_RE)),
 		with_help("Nexthop ID.", ec_node_uint("ID", 1, UINT32_MAX - 1, 10)),
 		with_help("L3 routing domain ID.", ec_node_uint("VRF", 0, UINT16_MAX - 1, 10))
 	);