mlxsw: spectrum_router: Add support for IPv6 routes addition / deletion
authorIdo Schimmel <idosch@mellanox.com>
Thu, 3 Aug 2017 11:28:28 +0000 (13:28 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 3 Aug 2017 22:36:00 +0000 (15:36 -0700)
Allow directly connected and remote unicast IPv6 routes to be programmed
to the device's tables.

As with IPv4, identical routes - sharing the same destination prefix -
are ordered in a FIB node according to their table ID and then the
metric. While the kernel doesn't share the same trie for the local and
main table, this does happen in the device, so ordering according to
table ID is needed.

Since individual nexthops can be added and deleted in IPv6, each FIB
entry stores a linked list of the rt6_info structs it represents. Upon
the addition or deletion of a nexthop, a new nexthop group is allocated
according to the new configuration and the old one is destroyed.
Identical groups aren't currently consolidated, but will be in a
follow-up patchset.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlxsw/Kconfig
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c

index 695adff89d717d8249fcae7ab0a56dfff32981ca..d56eea3105090051aad3138a837f7ef67c850b45 100644 (file)
@@ -75,6 +75,7 @@ config MLXSW_SPECTRUM
        depends on MLXSW_CORE && MLXSW_PCI && NET_SWITCHDEV && VLAN_8021Q
        depends on PSAMPLE || PSAMPLE=n
        depends on BRIDGE || BRIDGE=n
+       depends on IPV6 || IPV6=n
        select PARMAN
        select MLXFW
        default m
index 6c7fc6a66aca118f0ee8cd8dd607dd94c5d11ac3..2345c00ef7062e2d7c08cc5d76d59d7d92963226 100644 (file)
@@ -44,6 +44,7 @@
 #include <linux/netdevice.h>
 #include <linux/if_bridge.h>
 #include <linux/socket.h>
+#include <linux/route.h>
 #include <net/netevent.h>
 #include <net/neighbour.h>
 #include <net/arp.h>
@@ -406,6 +407,17 @@ struct mlxsw_sp_fib4_entry {
        u8 type;
 };
 
+struct mlxsw_sp_fib6_entry {
+       struct mlxsw_sp_fib_entry common;
+       struct list_head rt6_list;
+       unsigned int nrt6;
+};
+
+struct mlxsw_sp_rt6 {
+       struct list_head list;
+       struct rt6_info *rt;
+};
+
 enum mlxsw_sp_l3proto {
        MLXSW_SP_L3_PROTO_IPV4,
        MLXSW_SP_L3_PROTO_IPV6,
@@ -2126,6 +2138,26 @@ mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
        }
 }
 
+static struct mlxsw_sp_nexthop *
+mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
+                    const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
+{
+       int i;
+
+       for (i = 0; i < nh_grp->count; i++) {
+               struct mlxsw_sp_nexthop *nh = &nh_grp->nexthops[i];
+               struct rt6_info *rt = mlxsw_sp_rt6->rt;
+
+               if (nh->rif && nh->rif->dev == rt->dst.dev &&
+                   ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
+                                   &rt->rt6i_gateway))
+                       return nh;
+               continue;
+       }
+
+       return NULL;
+}
+
 static void
 mlxsw_sp_fib4_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
 {
@@ -2160,6 +2192,48 @@ mlxsw_sp_fib4_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
        }
 }
 
+static void
+mlxsw_sp_fib6_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+
+       fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
+                                 common);
+
+       if (fib_entry->type == MLXSW_SP_FIB_ENTRY_TYPE_LOCAL) {
+               list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
+                                list)->rt->rt6i_flags |= RTF_OFFLOAD;
+               return;
+       }
+
+       list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
+               struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
+               struct mlxsw_sp_nexthop *nh;
+
+               nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
+               if (nh && nh->offloaded)
+                       mlxsw_sp_rt6->rt->rt6i_flags |= RTF_OFFLOAD;
+               else
+                       mlxsw_sp_rt6->rt->rt6i_flags &= ~RTF_OFFLOAD;
+       }
+}
+
+static void
+mlxsw_sp_fib6_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+
+       fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
+                                 common);
+       list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
+               struct rt6_info *rt = mlxsw_sp_rt6->rt;
+
+               rt->rt6i_flags &= ~RTF_OFFLOAD;
+       }
+}
+
 static void mlxsw_sp_fib_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
 {
        switch (fib_entry->fib_node->fib->proto) {
@@ -2167,7 +2241,8 @@ static void mlxsw_sp_fib_entry_offload_set(struct mlxsw_sp_fib_entry *fib_entry)
                mlxsw_sp_fib4_entry_offload_set(fib_entry);
                break;
        case MLXSW_SP_L3_PROTO_IPV6:
-               WARN_ON_ONCE(1);
+               mlxsw_sp_fib6_entry_offload_set(fib_entry);
+               break;
        }
 }
 
@@ -2179,7 +2254,8 @@ mlxsw_sp_fib_entry_offload_unset(struct mlxsw_sp_fib_entry *fib_entry)
                mlxsw_sp_fib4_entry_offload_unset(fib_entry);
                break;
        case MLXSW_SP_L3_PROTO_IPV6:
-               WARN_ON_ONCE(1);
+               mlxsw_sp_fib6_entry_offload_unset(fib_entry);
+               break;
        }
 }
 
@@ -2887,6 +2963,615 @@ static void mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
        mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
 }
 
+static bool mlxsw_sp_fib6_rt_should_ignore(const struct rt6_info *rt)
+{
+       /* Packets with link-local destination IP arriving to the router
+        * are trapped to the CPU, so no need to program specific routes
+        * for them.
+        */
+       if (ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LINKLOCAL)
+               return true;
+
+       /* Multicast routes aren't supported, so ignore them. Neighbour
+        * Discovery packets are specifically trapped.
+        */
+       if (ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_MULTICAST)
+               return true;
+
+       /* Cloned routes are irrelevant in the forwarding path. */
+       if (rt->rt6i_flags & RTF_CACHE)
+               return true;
+
+       return false;
+}
+
+static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct rt6_info *rt)
+{
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+
+       mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
+       if (!mlxsw_sp_rt6)
+               return ERR_PTR(-ENOMEM);
+
+       /* In case of route replace, replaced route is deleted with
+        * no notification. Take reference to prevent accessing freed
+        * memory.
+        */
+       mlxsw_sp_rt6->rt = rt;
+       rt6_hold(rt);
+
+       return mlxsw_sp_rt6;
+}
+
+#if IS_ENABLED(CONFIG_IPV6)
+static void mlxsw_sp_rt6_release(struct rt6_info *rt)
+{
+       rt6_release(rt);
+}
+#else
+static void mlxsw_sp_rt6_release(struct rt6_info *rt)
+{
+}
+#endif
+
+static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
+{
+       mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
+       kfree(mlxsw_sp_rt6);
+}
+
+static bool mlxsw_sp_fib6_rt_can_mp(const struct rt6_info *rt)
+{
+       /* RTF_CACHE routes are ignored */
+       return (rt->rt6i_flags & (RTF_GATEWAY | RTF_ADDRCONF)) == RTF_GATEWAY;
+}
+
+static struct rt6_info *
+mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
+                               list)->rt;
+}
+
+static struct mlxsw_sp_fib6_entry *
+mlxsw_sp_fib6_node_mp_entry_find(const struct mlxsw_sp_fib_node *fib_node,
+                                const struct rt6_info *nrt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+
+       if (!mlxsw_sp_fib6_rt_can_mp(nrt))
+               return NULL;
+
+       list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
+               struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
+
+               /* RT6_TABLE_LOCAL and RT6_TABLE_MAIN share the same
+                * virtual router.
+                */
+               if (rt->rt6i_table->tb6_id > nrt->rt6i_table->tb6_id)
+                       continue;
+               if (rt->rt6i_table->tb6_id != nrt->rt6i_table->tb6_id)
+                       break;
+               if (rt->rt6i_metric < nrt->rt6i_metric)
+                       continue;
+               if (rt->rt6i_metric == nrt->rt6i_metric &&
+                   mlxsw_sp_fib6_rt_can_mp(rt))
+                       return fib6_entry;
+               if (rt->rt6i_metric > nrt->rt6i_metric)
+                       break;
+       }
+
+       return NULL;
+}
+
+static struct mlxsw_sp_rt6 *
+mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
+                           const struct rt6_info *rt)
+{
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+
+       list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
+               if (mlxsw_sp_rt6->rt == rt)
+                       return mlxsw_sp_rt6;
+       }
+
+       return NULL;
+}
+
+static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
+                                 struct mlxsw_sp_nexthop_group *nh_grp,
+                                 struct mlxsw_sp_nexthop *nh,
+                                 const struct rt6_info *rt)
+{
+       struct net_device *dev = rt->dst.dev;
+       struct mlxsw_sp_rif *rif;
+       int err;
+
+       nh->nh_grp = nh_grp;
+       memcpy(&nh->gw_addr, &rt->rt6i_gateway, sizeof(nh->gw_addr));
+
+       if (!dev)
+               return 0;
+
+       rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
+       if (!rif)
+               return 0;
+       mlxsw_sp_nexthop_rif_init(nh, rif);
+
+       err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
+       if (err)
+               goto err_nexthop_neigh_init;
+
+       return 0;
+
+err_nexthop_neigh_init:
+       mlxsw_sp_nexthop_rif_fini(nh);
+       return err;
+}
+
+static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
+                                  struct mlxsw_sp_nexthop *nh)
+{
+       mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
+       mlxsw_sp_nexthop_rif_fini(nh);
+}
+
+static struct mlxsw_sp_nexthop_group *
+mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
+                              struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       struct mlxsw_sp_nexthop_group *nh_grp;
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+       struct mlxsw_sp_nexthop *nh;
+       size_t alloc_size;
+       int i = 0;
+       int err;
+
+       alloc_size = sizeof(*nh_grp) +
+                    fib6_entry->nrt6 * sizeof(struct mlxsw_sp_nexthop);
+       nh_grp = kzalloc(alloc_size, GFP_KERNEL);
+       if (!nh_grp)
+               return ERR_PTR(-ENOMEM);
+       INIT_LIST_HEAD(&nh_grp->fib_list);
+#if IS_ENABLED(CONFIG_IPV6)
+       nh_grp->neigh_tbl = &nd_tbl;
+#endif
+       mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
+                                       struct mlxsw_sp_rt6, list);
+       nh_grp->gateway = !!(mlxsw_sp_rt6->rt->rt6i_flags & RTF_GATEWAY);
+       nh_grp->count = fib6_entry->nrt6;
+       for (i = 0; i < nh_grp->count; i++) {
+               struct rt6_info *rt = mlxsw_sp_rt6->rt;
+
+               nh = &nh_grp->nexthops[i];
+               err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
+               if (err)
+                       goto err_nexthop6_init;
+               mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
+       }
+       mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
+       return nh_grp;
+
+err_nexthop6_init:
+       for (i--; i >= 0; i--) {
+               nh = &nh_grp->nexthops[i];
+               mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
+       }
+       kfree(nh_grp);
+       return ERR_PTR(err);
+}
+
+static void
+mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
+                               struct mlxsw_sp_nexthop_group *nh_grp)
+{
+       struct mlxsw_sp_nexthop *nh;
+       int i = nh_grp->count;
+
+       for (i--; i >= 0; i--) {
+               nh = &nh_grp->nexthops[i];
+               mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
+       }
+       mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
+       WARN_ON(nh_grp->adj_index_valid);
+       kfree(nh_grp);
+}
+
+static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
+                                      struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       struct mlxsw_sp_nexthop_group *nh_grp;
+
+       /* For now, don't consolidate nexthop groups */
+       nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
+       if (IS_ERR(nh_grp))
+               return PTR_ERR(nh_grp);
+
+       list_add_tail(&fib6_entry->common.nexthop_group_node,
+                     &nh_grp->fib_list);
+       fib6_entry->common.nh_group = nh_grp;
+
+       return 0;
+}
+
+static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
+                                       struct mlxsw_sp_fib_entry *fib_entry)
+{
+       struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
+
+       list_del(&fib_entry->nexthop_group_node);
+       if (!list_empty(&nh_grp->fib_list))
+               return;
+       mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
+}
+
+static int
+mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
+                              struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
+       int err;
+
+       fib6_entry->common.nh_group = NULL;
+       list_del(&fib6_entry->common.nexthop_group_node);
+
+       err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
+       if (err)
+               goto err_nexthop6_group_get;
+
+       /* In case this entry is offloaded, then the adjacency index
+        * currently associated with it in the device's table is that
+        * of the old group. Start using the new one instead.
+        */
+       err = mlxsw_sp_fib_node_entry_add(mlxsw_sp, &fib6_entry->common);
+       if (err)
+               goto err_fib_node_entry_add;
+
+       if (list_empty(&old_nh_grp->fib_list))
+               mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
+
+       return 0;
+
+err_fib_node_entry_add:
+       mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
+err_nexthop6_group_get:
+       list_add_tail(&fib6_entry->common.nexthop_group_node,
+                     &old_nh_grp->fib_list);
+       fib6_entry->common.nh_group = old_nh_grp;
+       return err;
+}
+
+static int
+mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
+                               struct mlxsw_sp_fib6_entry *fib6_entry,
+                               struct rt6_info *rt)
+{
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+       int err;
+
+       mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt);
+       if (IS_ERR(mlxsw_sp_rt6))
+               return PTR_ERR(mlxsw_sp_rt6);
+
+       list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
+       fib6_entry->nrt6++;
+
+       err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
+       if (err)
+               goto err_nexthop6_group_update;
+
+       return 0;
+
+err_nexthop6_group_update:
+       fib6_entry->nrt6--;
+       list_del(&mlxsw_sp_rt6->list);
+       mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
+       return err;
+}
+
+static void
+mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
+                               struct mlxsw_sp_fib6_entry *fib6_entry,
+                               struct rt6_info *rt)
+{
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+
+       mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt);
+       if (WARN_ON(!mlxsw_sp_rt6))
+               return;
+
+       fib6_entry->nrt6--;
+       list_del(&mlxsw_sp_rt6->list);
+       mlxsw_sp_nexthop6_group_update(mlxsw_sp, fib6_entry);
+       mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
+}
+
+static void mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp_fib_entry *fib_entry,
+                                        const struct rt6_info *rt)
+{
+       /* Packets hitting RTF_REJECT routes need to be discarded by the
+        * stack. We can rely on their destination device not having a
+        * RIF (it's the loopback device) and can thus use action type
+        * local, which will cause them to be trapped with a lower
+        * priority than packets that need to be locally received.
+        */
+       if (rt->rt6i_flags & RTF_LOCAL)
+               fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
+       else if (rt->rt6i_flags & RTF_REJECT)
+               fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
+       else if (rt->rt6i_flags & RTF_GATEWAY)
+               fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
+       else
+               fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
+}
+
+static void
+mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
+
+       list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
+                                list) {
+               fib6_entry->nrt6--;
+               list_del(&mlxsw_sp_rt6->list);
+               mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
+       }
+}
+
+static struct mlxsw_sp_fib6_entry *
+mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
+                          struct mlxsw_sp_fib_node *fib_node,
+                          struct rt6_info *rt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_fib_entry *fib_entry;
+       struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
+       int err;
+
+       fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
+       if (!fib6_entry)
+               return ERR_PTR(-ENOMEM);
+       fib_entry = &fib6_entry->common;
+
+       mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt);
+       if (IS_ERR(mlxsw_sp_rt6)) {
+               err = PTR_ERR(mlxsw_sp_rt6);
+               goto err_rt6_create;
+       }
+
+       mlxsw_sp_fib6_entry_type_set(fib_entry, mlxsw_sp_rt6->rt);
+
+       INIT_LIST_HEAD(&fib6_entry->rt6_list);
+       list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
+       fib6_entry->nrt6 = 1;
+       err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
+       if (err)
+               goto err_nexthop6_group_get;
+
+       fib_entry->fib_node = fib_node;
+
+       return fib6_entry;
+
+err_nexthop6_group_get:
+       list_del(&mlxsw_sp_rt6->list);
+       mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
+err_rt6_create:
+       kfree(fib6_entry);
+       return ERR_PTR(err);
+}
+
+static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
+                                       struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
+       mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
+       WARN_ON(fib6_entry->nrt6);
+       kfree(fib6_entry);
+}
+
+static struct mlxsw_sp_fib6_entry *
+mlxsw_sp_fib6_node_entry_find(const struct mlxsw_sp_fib_node *fib_node,
+                             const struct rt6_info *nrt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+
+       list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
+               struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
+
+               if (rt->rt6i_table->tb6_id > nrt->rt6i_table->tb6_id)
+                       continue;
+               if (rt->rt6i_table->tb6_id != nrt->rt6i_table->tb6_id)
+                       break;
+               if (rt->rt6i_metric > nrt->rt6i_metric)
+                       return fib6_entry;
+       }
+
+       return NULL;
+}
+
+static int
+mlxsw_sp_fib6_node_list_insert(struct mlxsw_sp_fib6_entry *new6_entry)
+{
+       struct mlxsw_sp_fib_node *fib_node = new6_entry->common.fib_node;
+       struct rt6_info *nrt = mlxsw_sp_fib6_entry_rt(new6_entry);
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+
+       fib6_entry = mlxsw_sp_fib6_node_entry_find(fib_node, nrt);
+
+       if (fib6_entry) {
+               list_add_tail(&new6_entry->common.list,
+                             &fib6_entry->common.list);
+       } else {
+               struct mlxsw_sp_fib6_entry *last;
+
+               list_for_each_entry(last, &fib_node->entry_list, common.list) {
+                       struct rt6_info *rt = mlxsw_sp_fib6_entry_rt(last);
+
+                       if (nrt->rt6i_table->tb6_id > rt->rt6i_table->tb6_id)
+                               break;
+                       fib6_entry = last;
+               }
+
+               if (fib6_entry)
+                       list_add(&new6_entry->common.list,
+                                &fib6_entry->common.list);
+               else
+                       list_add(&new6_entry->common.list,
+                                &fib_node->entry_list);
+       }
+
+       return 0;
+}
+
+static void
+mlxsw_sp_fib6_node_list_remove(struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       list_del(&fib6_entry->common.list);
+}
+
+static int mlxsw_sp_fib6_node_entry_link(struct mlxsw_sp *mlxsw_sp,
+                                        struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       int err;
+
+       err = mlxsw_sp_fib6_node_list_insert(fib6_entry);
+       if (err)
+               return err;
+
+       err = mlxsw_sp_fib_node_entry_add(mlxsw_sp, &fib6_entry->common);
+       if (err)
+               goto err_fib_node_entry_add;
+
+       return 0;
+
+err_fib_node_entry_add:
+       mlxsw_sp_fib6_node_list_remove(fib6_entry);
+       return err;
+}
+
+static void
+mlxsw_sp_fib6_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
+                               struct mlxsw_sp_fib6_entry *fib6_entry)
+{
+       mlxsw_sp_fib_node_entry_del(mlxsw_sp, &fib6_entry->common);
+       mlxsw_sp_fib6_node_list_remove(fib6_entry);
+}
+
+static struct mlxsw_sp_fib6_entry *
+mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
+                          const struct rt6_info *rt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_fib_node *fib_node;
+       struct mlxsw_sp_fib *fib;
+       struct mlxsw_sp_vr *vr;
+
+       vr = mlxsw_sp_vr_find(mlxsw_sp, rt->rt6i_table->tb6_id);
+       if (!vr)
+               return NULL;
+       fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
+
+       fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->rt6i_dst.addr,
+                                           sizeof(rt->rt6i_dst.addr),
+                                           rt->rt6i_dst.plen);
+       if (!fib_node)
+               return NULL;
+
+       list_for_each_entry(fib6_entry, &fib_node->entry_list, common.list) {
+               struct rt6_info *iter_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
+
+               if (rt->rt6i_table->tb6_id == iter_rt->rt6i_table->tb6_id &&
+                   rt->rt6i_metric == iter_rt->rt6i_metric &&
+                   mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
+                       return fib6_entry;
+       }
+
+       return NULL;
+}
+
+static int mlxsw_sp_router_fib6_add(struct mlxsw_sp *mlxsw_sp,
+                                   struct rt6_info *rt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_fib_node *fib_node;
+       int err;
+
+       if (mlxsw_sp->router->aborted)
+               return 0;
+
+       if (mlxsw_sp_fib6_rt_should_ignore(rt))
+               return 0;
+
+       fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->rt6i_table->tb6_id,
+                                        &rt->rt6i_dst.addr,
+                                        sizeof(rt->rt6i_dst.addr),
+                                        rt->rt6i_dst.plen,
+                                        MLXSW_SP_L3_PROTO_IPV6);
+       if (IS_ERR(fib_node))
+               return PTR_ERR(fib_node);
+
+       /* Before creating a new entry, try to append route to an existing
+        * multipath entry.
+        */
+       fib6_entry = mlxsw_sp_fib6_node_mp_entry_find(fib_node, rt);
+       if (fib6_entry) {
+               err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, fib6_entry, rt);
+               if (err)
+                       goto err_fib6_entry_nexthop_add;
+               return 0;
+       }
+
+       fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt);
+       if (IS_ERR(fib6_entry)) {
+               err = PTR_ERR(fib6_entry);
+               goto err_fib6_entry_create;
+       }
+
+       err = mlxsw_sp_fib6_node_entry_link(mlxsw_sp, fib6_entry);
+       if (err)
+               goto err_fib6_node_entry_link;
+
+       return 0;
+
+err_fib6_node_entry_link:
+       mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
+err_fib6_entry_create:
+err_fib6_entry_nexthop_add:
+       mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
+       return err;
+}
+
+static void mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
+                                    struct rt6_info *rt)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry;
+       struct mlxsw_sp_fib_node *fib_node;
+
+       if (mlxsw_sp->router->aborted)
+               return;
+
+       if (mlxsw_sp_fib6_rt_should_ignore(rt))
+               return;
+
+       fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
+       if (WARN_ON(!fib6_entry))
+               return;
+
+       /* If route is part of a multipath entry, but not the last one
+        * removed, then only reduce its nexthop group.
+        */
+       if (!list_is_singular(&fib6_entry->rt6_list)) {
+               mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, fib6_entry, rt);
+               return;
+       }
+
+       fib_node = fib6_entry->common.fib_node;
+
+       mlxsw_sp_fib6_node_entry_unlink(mlxsw_sp, fib6_entry);
+       mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
+       mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
+}
+
 static int __mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp,
                                            enum mlxsw_reg_ralxx_protocol proto,
                                            u8 tree_id)
@@ -2967,6 +3652,23 @@ static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
        }
 }
 
+static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
+                                    struct mlxsw_sp_fib_node *fib_node)
+{
+       struct mlxsw_sp_fib6_entry *fib6_entry, *tmp;
+
+       list_for_each_entry_safe(fib6_entry, tmp, &fib_node->entry_list,
+                                common.list) {
+               bool do_break = &tmp->common.list == &fib_node->entry_list;
+
+               mlxsw_sp_fib6_node_entry_unlink(mlxsw_sp, fib6_entry);
+               mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
+               mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
+               if (do_break)
+                       break;
+       }
+}
+
 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
                                    struct mlxsw_sp_fib_node *fib_node)
 {
@@ -2975,7 +3677,7 @@ static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
                mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
                break;
        case MLXSW_SP_L3_PROTO_IPV6:
-               WARN_ON_ONCE(1);
+               mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
                break;
        }
 }
@@ -3033,6 +3735,7 @@ static void mlxsw_sp_router_fib_abort(struct mlxsw_sp *mlxsw_sp)
 struct mlxsw_sp_fib_event_work {
        struct work_struct work;
        union {
+               struct fib6_entry_notifier_info fen6_info;
                struct fib_entry_notifier_info fen_info;
                struct fib_rule_notifier_info fr_info;
                struct fib_nh_notifier_info fnh_info;
@@ -3092,9 +3795,21 @@ static void mlxsw_sp_router_fib6_event_work(struct work_struct *work)
                container_of(work, struct mlxsw_sp_fib_event_work, work);
        struct mlxsw_sp *mlxsw_sp = fib_work->mlxsw_sp;
        struct fib_rule *rule;
+       int err;
 
        rtnl_lock();
        switch (fib_work->event) {
+       case FIB_EVENT_ENTRY_ADD:
+               err = mlxsw_sp_router_fib6_add(mlxsw_sp,
+                                              fib_work->fen6_info.rt);
+               if (err)
+                       mlxsw_sp_router_fib_abort(mlxsw_sp);
+               mlxsw_sp_rt6_release(fib_work->fen6_info.rt);
+               break;
+       case FIB_EVENT_ENTRY_DEL:
+               mlxsw_sp_router_fib6_del(mlxsw_sp, fib_work->fen6_info.rt);
+               mlxsw_sp_rt6_release(fib_work->fen6_info.rt);
+               break;
        case FIB_EVENT_RULE_ADD: /* fall through */
        case FIB_EVENT_RULE_DEL:
                rule = fib_work->fr_info.rule;
@@ -3138,6 +3853,11 @@ static void mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event_work *fib_work,
                                       struct fib_notifier_info *info)
 {
        switch (fib_work->event) {
+       case FIB_EVENT_ENTRY_ADD: /* fall through */
+       case FIB_EVENT_ENTRY_DEL:
+               memcpy(&fib_work->fen6_info, info, sizeof(fib_work->fen6_info));
+               rt6_hold(fib_work->fen6_info.rt);
+               break;
        case FIB_EVENT_RULE_ADD: /* fall through */
        case FIB_EVENT_RULE_DEL:
                memcpy(&fib_work->fr_info, info, sizeof(fib_work->fr_info));