IP Multicast FIB (mfib)

 - IPv[46] mfib tables with support for (*,G/m), (*,G) and (S,G) exact and longest prefix match
 - Replication represented via a new replicate DPO.
 - RPF configuration and data-plane checking
 - data-plane signals sent to listening control planes.

The functions of multicast forwarding entries differ from their unicast conterparts, so we introduce a new mfib_table_t and mfib_entry_t objects. However, we re-use the fib_path_list to resolve and build the entry's output list. the fib_path_list provides the service to construct a replicate DPO for multicast.

'make tests' is added to with two new suites; TEST=mfib, this is invocation of the CLI command 'test mfib' which deals with many path add/remove, flag set/unset scenarios, TEST=ip-mcast, data-plane forwarding tests.

Updated applications to use the new MIFB functions;
  - IPv6 NS/RA.
  - DHCPv6
 unit tests for these are undated accordingly.

Change-Id: I49ec37b01f1b170335a5697541c8fd30e6d3a961
Signed-off-by: Neale Ranns <nranns@cisco.com>

6 files changed, 931 insertions, 2 deletions

diff --git a/src/vnet/dpo/dpo.c b/src/vnet/dpo/dpo.c
index 688d2892412..cc2fa0eb91c 100644
--- a/src/vnet/dpo/dpo.c
+++ b/src/vnet/dpo/dpo.c
@@ -36,6 +36,7 @@
 #include <vnet/dpo/punt_dpo.h>
 #include <vnet/dpo/classify_dpo.h>
 #include <vnet/dpo/ip_null_dpo.h>
+#include <vnet/dpo/replicate_dpo.h>
 
 /**
  * Array of char* names for the DPO types and protos
@@ -449,6 +450,7 @@ dpo_module_init (vlib_main_t * vm)
     classify_dpo_module_init();
     lookup_dpo_module_init();
     ip_null_dpo_module_init();
+    replicate_module_init();
 
     return (NULL);
 }
diff --git a/src/vnet/dpo/dpo.h b/src/vnet/dpo/dpo.h
index 1efcbc8834b..aff4e1b82cc 100644
--- a/src/vnet/dpo/dpo.h
+++ b/src/vnet/dpo/dpo.h
@@ -100,15 +100,18 @@ typedef enum dpo_type_t_ {
      * @brief load-balancing over a choice of [un]equal cost paths
      */
     DPO_LOAD_BALANCE,
+    DPO_REPLICATE,
     DPO_ADJACENCY,
     DPO_ADJACENCY_INCOMPLETE,
     DPO_ADJACENCY_MIDCHAIN,
     DPO_ADJACENCY_GLEAN,
+    DPO_ADJACENCY_MCAST,
     DPO_RECEIVE,
     DPO_LOOKUP,
     DPO_LISP_CP,
     DPO_CLASSIFY,
     DPO_MPLS_LABEL,
+    DPO_MFIB_ENTRY,
     DPO_LAST,
 } __attribute__((packed)) dpo_type_t;
 
@@ -123,12 +126,15 @@ typedef enum dpo_type_t_ {
     [DPO_ADJACENCY_INCOMPLETE] = "dpo-adjacency-incomplete",	\
     [DPO_ADJACENCY_MIDCHAIN] = "dpo-adjacency-midcahin",	\
     [DPO_ADJACENCY_GLEAN] = "dpo-glean",	\
+    [DPO_ADJACENCY_MCAST] = "dpo-adj-mcast",	\
     [DPO_RECEIVE] = "dpo-receive",	\
     [DPO_LOOKUP] = "dpo-lookup",	\
     [DPO_LOAD_BALANCE] = "dpo-load-balance",	\
+    [DPO_REPLICATE] = "dpo-replicate",	\
     [DPO_LISP_CP] = "dpo-lisp-cp",	\
     [DPO_CLASSIFY] = "dpo-classify",	\
-    [DPO_MPLS_LABEL] = "dpo-mpls-label"	\
+    [DPO_MPLS_LABEL] = "dpo-mpls-label", \
+    [DPO_MFIB_ENTRY] = "dpo-mfib_entry"	\
 }
 
 /**
diff --git a/src/vnet/dpo/load_balance.c b/src/vnet/dpo/load_balance.c
index e70a7a306e1..f11b4e4de84 100644
--- a/src/vnet/dpo/load_balance.c
+++ b/src/vnet/dpo/load_balance.c
@@ -239,6 +239,17 @@ load_balance_is_drop (const dpo_id_t *dpo)
 }
 
 void
+load_balance_set_fib_entry_flags (index_t lbi,
+                                  fib_entry_flag_t flags)
+{
+    load_balance_t *lb;
+
+    lb = load_balance_get(lbi);
+    lb->lb_fib_entry_flags = flags;
+}
+
+
+void
 load_balance_set_urpf (index_t lbi,
 		       index_t urpf)
 {
@@ -683,7 +694,7 @@ load_balance_multipath_update (const dpo_id_t *dpo,
                                           buckets,
                                           n_buckets);
 
-                for (ii = old_n_buckets-n_buckets; ii < old_n_buckets; ii++)
+                for (ii = n_buckets; ii < old_n_buckets; ii++)
                 {
                     dpo_reset(&buckets[ii]);
                 }
diff --git a/src/vnet/dpo/load_balance.h b/src/vnet/dpo/load_balance.h
index 1799653628d..b901c5beb84 100644
--- a/src/vnet/dpo/load_balance.h
+++ b/src/vnet/dpo/load_balance.h
@@ -36,6 +36,7 @@
 #include <vnet/ip/lookup.h>
 #include <vnet/dpo/dpo.h>
 #include <vnet/fib/fib_types.h>
+#include <vnet/fib/fib_entry.h>
 
 /**
  * Load-balance main
@@ -99,6 +100,11 @@ typedef struct load_balance_t_ {
     dpo_proto_t lb_proto;
 
     /**
+     * Flags from the load-balance's associated fib_entry_t
+     */
+    fib_entry_flag_t lb_fib_entry_flags;
+
+    /**
      * The number of locks, which is approximately the number of users,
      * of this load-balance.
      * Load-balance objects of via-entries are heavily shared by recursives,
@@ -167,6 +173,8 @@ extern void load_balance_set_bucket(index_t lbi,
 				    const dpo_id_t *next);
 extern void load_balance_set_urpf(index_t lbi,
 				  index_t urpf);
+extern void load_balance_set_fib_entry_flags(index_t lbi,
+                                             fib_entry_flag_t flags);
 extern index_t load_balance_get_urpf(index_t lbi);
 
 extern u8* format_load_balance(u8 * s, va_list * args);
diff --git a/src/vnet/dpo/replicate_dpo.c b/src/vnet/dpo/replicate_dpo.c
new file mode 100644
index 00000000000..a2d5fdb68bd
--- /dev/null
+++ b/src/vnet/dpo/replicate_dpo.c
@@ -0,0 +1,759 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vnet/ip/lookup.h>
+#include <vnet/dpo/replicate_dpo.h>
+#include <vnet/dpo/drop_dpo.h>
+#include <vnet/adj/adj.h>
+
+#undef REP_DEBUG
+
+#ifdef REP_DEBUG
+#define REP_DBG(_rep, _fmt, _args...)                                   \
+{                                                                       \
+    u8* _tmp =NULL;                                                     \
+    clib_warning("rep:[%s]:" _fmt,                                      \
+                 replicate_format(replicate_get_index((_rep)),          \
+                                  0, _tmp),                             \
+                 ##_args);                                              \
+    vec_free(_tmp);                                                     \
+}
+#else
+#define REP_DBG(_p, _fmt, _args...)
+#endif
+
+
+/**
+ * Pool of all DPOs. It's not static so the DP can have fast access
+ */
+replicate_t *replicate_pool;
+
+/**
+ * The one instance of replicate main
+ */
+replicate_main_t replicate_main;
+
+static inline index_t
+replicate_get_index (const replicate_t *rep)
+{
+    return (rep - replicate_pool);
+}
+
+static inline dpo_id_t*
+replicate_get_buckets (replicate_t *rep)
+{
+    if (REP_HAS_INLINE_BUCKETS(rep))
+    {
+        return (rep->rep_buckets_inline);
+    }
+    else
+    {
+        return (rep->rep_buckets);
+    }
+}
+
+static replicate_t *
+replicate_alloc_i (void)
+{
+    replicate_t *rep;
+
+    pool_get_aligned(replicate_pool, rep, CLIB_CACHE_LINE_BYTES);
+    memset(rep, 0, sizeof(*rep));
+
+    vlib_validate_combined_counter(&(replicate_main.repm_counters),
+                                   replicate_get_index(rep));
+    vlib_zero_combined_counter(&(replicate_main.repm_counters),
+                               replicate_get_index(rep));
+
+    return (rep);
+}
+
+static u8*
+replicate_format (index_t repi,
+                  replicate_format_flags_t flags,
+                  u32 indent,
+                  u8 *s)
+{
+    vlib_counter_t to;
+    replicate_t *rep;
+    dpo_id_t *buckets;
+    u32 i;
+
+    rep = replicate_get(repi);
+    vlib_get_combined_counter(&(replicate_main.repm_counters), repi, &to);
+    buckets = replicate_get_buckets(rep);
+
+    s = format(s, "%U: ", format_dpo_type, DPO_REPLICATE);
+    s = format(s, "[index:%d buckets:%d ", repi, rep->rep_n_buckets);
+    s = format(s, "to:[%Ld:%Ld]]", to.packets, to.bytes);
+
+    for (i = 0; i < rep->rep_n_buckets; i++)
+    {
+        s = format(s, "\n%U", format_white_space, indent+2);
+        s = format(s, "[%d]", i);
+        s = format(s, " %U", format_dpo_id, &buckets[i], indent+6);
+    }
+    return (s);
+}
+
+u8*
+format_replicate (u8 * s, va_list * args)
+{
+    index_t repi = va_arg(*args, index_t);
+    replicate_format_flags_t flags = va_arg(*args, replicate_format_flags_t);
+
+    return (replicate_format(repi, flags, 0, s));
+}
+static u8*
+format_replicate_dpo (u8 * s, va_list * args)
+{
+    index_t repi = va_arg(*args, index_t);
+    u32 indent = va_arg(*args, u32);
+
+    return (replicate_format(repi, REPLICATE_FORMAT_DETAIL, indent, s));
+}
+
+
+static replicate_t *
+replicate_create_i (u32 num_buckets,
+                    dpo_proto_t rep_proto)
+{
+    replicate_t *rep;
+
+    rep = replicate_alloc_i();
+    rep->rep_n_buckets = num_buckets;
+    rep->rep_proto = rep_proto;
+
+    if (!REP_HAS_INLINE_BUCKETS(rep))
+    {
+        vec_validate_aligned(rep->rep_buckets,
+                             rep->rep_n_buckets - 1,
+                             CLIB_CACHE_LINE_BYTES);
+    }
+
+    REP_DBG(rep, "create");
+
+    return (rep);
+}
+
+index_t
+replicate_create (u32 n_buckets,
+                  dpo_proto_t rep_proto)
+{
+    return (replicate_get_index(replicate_create_i(n_buckets, rep_proto)));
+}
+
+static inline void
+replicate_set_bucket_i (replicate_t *rep,
+                        u32 bucket,
+                        dpo_id_t *buckets,
+                        const dpo_id_t *next)
+{
+    dpo_stack(DPO_REPLICATE, rep->rep_proto, &buckets[bucket], next);
+}
+
+void
+replicate_set_bucket (index_t repi,
+                      u32 bucket,
+                      const dpo_id_t *next)
+{
+    replicate_t *rep;
+    dpo_id_t *buckets;
+
+    rep = replicate_get(repi);
+    buckets = replicate_get_buckets(rep);
+
+    ASSERT(bucket < rep->rep_n_buckets);
+
+    replicate_set_bucket_i(rep, bucket, buckets, next);
+}
+
+int
+replicate_is_drop (const dpo_id_t *dpo)
+{
+    replicate_t *rep;
+
+    if (DPO_REPLICATE != dpo->dpoi_type)
+        return (0);
+
+    rep = replicate_get(dpo->dpoi_index);
+
+    if (1 == rep->rep_n_buckets)
+    {
+        return (dpo_is_drop(replicate_get_bucket_i(rep, 0)));
+    }
+    return (0);
+}
+
+const dpo_id_t *
+replicate_get_bucket (index_t repi,
+                      u32 bucket)
+{
+    replicate_t *rep;
+
+    rep = replicate_get(repi);
+
+    return (replicate_get_bucket_i(rep, bucket));
+}
+
+
+static load_balance_path_t *
+replicate_multipath_next_hop_fixup (load_balance_path_t *nhs,
+                                    dpo_proto_t drop_proto)
+{
+    if (0 == vec_len(nhs))
+    {
+        load_balance_path_t *nh;
+
+        /*
+         * we need something for the replicate. so use the drop
+         */
+        vec_add2(nhs, nh, 1);
+
+        nh->path_weight = 1;
+        dpo_copy(&nh->path_dpo, drop_dpo_get(drop_proto));
+    }
+
+    return (nhs);
+}
+
+/*
+ * Fill in adjacencies in block based on corresponding
+ * next hop adjacencies.
+ */
+static void
+replicate_fill_buckets (replicate_t *rep,
+                        load_balance_path_t *nhs,
+                        dpo_id_t *buckets,
+                        u32 n_buckets)
+{
+    load_balance_path_t * nh;
+    u16 ii, bucket;
+
+    bucket = 0;
+
+    /*
+     * the next-hops have normalised weights. that means their sum is the number
+     * of buckets we need to fill.
+     */
+    vec_foreach (nh, nhs)
+    {
+        for (ii = 0; ii < nh->path_weight; ii++)
+        {
+            ASSERT(bucket < n_buckets);
+            replicate_set_bucket_i(rep, bucket++, buckets, &nh->path_dpo);
+        }
+    }
+}
+
+static inline void
+replicate_set_n_buckets (replicate_t *rep,
+                         u32 n_buckets)
+{
+    rep->rep_n_buckets = n_buckets;
+}
+
+void
+replicate_multipath_update (const dpo_id_t *dpo,
+                            load_balance_path_t * next_hops)
+{
+    load_balance_path_t * nh, * nhs;
+    dpo_id_t *tmp_dpo;
+    u32 ii, n_buckets;
+    replicate_t *rep;
+
+    ASSERT(DPO_REPLICATE == dpo->dpoi_type);
+    rep = replicate_get(dpo->dpoi_index);
+    nhs = replicate_multipath_next_hop_fixup(next_hops,
+                                             rep->rep_proto);
+    n_buckets = vec_len(nhs);
+
+    if (0 == rep->rep_n_buckets)
+    {
+        /*
+         * first time initialisation. no packets inflight, so we can write
+         * at leisure.
+         */
+        replicate_set_n_buckets(rep, n_buckets);
+
+        if (!REP_HAS_INLINE_BUCKETS(rep))
+            vec_validate_aligned(rep->rep_buckets,
+                                 rep->rep_n_buckets - 1,
+                                 CLIB_CACHE_LINE_BYTES);
+
+        replicate_fill_buckets(rep, nhs,
+                               replicate_get_buckets(rep),
+                               n_buckets);
+    }
+    else
+    {
+        /*
+         * This is a modification of an existing replicate.
+         * We need to ensure that packets in flight see a consistent state, that
+         * is the number of reported buckets the REP has
+         * is not more than it actually has. So if the
+         * number of buckets is increasing, we must update the bucket array first,
+         * then the reported number. vice-versa if the number of buckets goes down.
+         */
+        if (n_buckets == rep->rep_n_buckets)
+        {
+            /*
+             * no change in the number of buckets. we can simply fill what
+             * is new over what is old.
+             */
+            replicate_fill_buckets(rep, nhs,
+                                   replicate_get_buckets(rep),
+                                   n_buckets);
+        }
+        else if (n_buckets > rep->rep_n_buckets)
+        {
+            /*
+             * we have more buckets. the old replicate map (if there is one)
+             * will remain valid, i.e. mapping to indices within range, so we
+             * update it last.
+             */
+            if (n_buckets > REP_NUM_INLINE_BUCKETS &&
+                rep->rep_n_buckets <= REP_NUM_INLINE_BUCKETS)
+            {
+                /*
+                 * the new increased number of buckets is crossing the threshold
+                 * from the inline storage to out-line. Alloc the outline buckets
+                 * first, then fixup the number. then reset the inlines.
+                 */
+                ASSERT(NULL == rep->rep_buckets);
+                vec_validate_aligned(rep->rep_buckets,
+                                     n_buckets - 1,
+                                     CLIB_CACHE_LINE_BYTES);
+
+                replicate_fill_buckets(rep, nhs,
+                                       rep->rep_buckets,
+                                       n_buckets);
+                CLIB_MEMORY_BARRIER();
+                replicate_set_n_buckets(rep, n_buckets);
+
+                CLIB_MEMORY_BARRIER();
+
+                for (ii = 0; ii < REP_NUM_INLINE_BUCKETS; ii++)
+                {
+                    dpo_reset(&rep->rep_buckets_inline[ii]);
+                }
+            }
+            else
+            {
+                if (n_buckets <= REP_NUM_INLINE_BUCKETS)
+                {
+                    /*
+                     * we are not crossing the threshold and it's still inline buckets.
+                     * we can write the new on the old..
+                     */
+                    replicate_fill_buckets(rep, nhs,
+                                           replicate_get_buckets(rep),
+                                           n_buckets);
+                    CLIB_MEMORY_BARRIER();
+                    replicate_set_n_buckets(rep, n_buckets);
+                }
+                else
+                {
+                    /*
+                     * we are not crossing the threshold. We need a new bucket array to
+                     * hold the increased number of choices.
+                     */
+                    dpo_id_t *new_buckets, *old_buckets, *tmp_dpo;
+
+                    new_buckets = NULL;
+                    old_buckets = replicate_get_buckets(rep);
+
+                    vec_validate_aligned(new_buckets,
+                                         n_buckets - 1,
+                                         CLIB_CACHE_LINE_BYTES);
+
+                    replicate_fill_buckets(rep, nhs, new_buckets, n_buckets);
+                    CLIB_MEMORY_BARRIER();
+                    rep->rep_buckets = new_buckets;
+                    CLIB_MEMORY_BARRIER();
+                    replicate_set_n_buckets(rep, n_buckets);
+
+                    vec_foreach(tmp_dpo, old_buckets)
+                    {
+                        dpo_reset(tmp_dpo);
+                    }
+                    vec_free(old_buckets);
+                }
+            }
+        }
+        else
+        {
+            /*
+             * bucket size shrinkage.
+             */
+            if (n_buckets <= REP_NUM_INLINE_BUCKETS &&
+                rep->rep_n_buckets > REP_NUM_INLINE_BUCKETS)
+            {
+                /*
+                 * the new decreased number of buckets is crossing the threshold
+                 * from out-line storage to inline:
+                 *   1 - Fill the inline buckets,
+                 *   2 - fixup the number (and this point the inline buckets are
+                 *       used).
+                 *   3 - free the outline buckets
+                 */
+                replicate_fill_buckets(rep, nhs,
+                                       rep->rep_buckets_inline,
+                                       n_buckets);
+                CLIB_MEMORY_BARRIER();
+                replicate_set_n_buckets(rep, n_buckets);
+                CLIB_MEMORY_BARRIER();
+
+                vec_foreach(tmp_dpo, rep->rep_buckets)
+                {
+                    dpo_reset(tmp_dpo);
+                }
+                vec_free(rep->rep_buckets);
+            }
+            else
+            {
+                /*
+                 * not crossing the threshold.
+                 *  1 - update the number to the smaller size
+                 *  2 - write the new buckets
+                 *  3 - reset those no longer used.
+                 */
+                dpo_id_t *buckets;
+                u32 old_n_buckets;
+
+                old_n_buckets = rep->rep_n_buckets;
+                buckets = replicate_get_buckets(rep);
+
+                replicate_set_n_buckets(rep, n_buckets);
+                CLIB_MEMORY_BARRIER();
+
+                replicate_fill_buckets(rep, nhs,
+                                       buckets,
+                                       n_buckets);
+
+                for (ii = n_buckets; ii < old_n_buckets; ii++)
+                {
+                    dpo_reset(&buckets[ii]);
+                }
+            }
+        }
+    }
+
+    vec_foreach (nh, nhs)
+    {
+        dpo_reset(&nh->path_dpo);
+    }
+    vec_free(nhs);
+}
+
+static void
+replicate_lock (dpo_id_t *dpo)
+{
+    replicate_t *rep;
+
+    rep = replicate_get(dpo->dpoi_index);
+
+    rep->rep_locks++;
+}
+
+static void
+replicate_destroy (replicate_t *rep)
+{
+    dpo_id_t *buckets;
+    int i;
+
+    buckets = replicate_get_buckets(rep);
+
+    for (i = 0; i < rep->rep_n_buckets; i++)
+    {
+        dpo_reset(&buckets[i]);
+    }
+
+    REP_DBG(rep, "destroy");
+    if (!REP_HAS_INLINE_BUCKETS(rep))
+    {
+        vec_free(rep->rep_buckets);
+    }
+
+    pool_put(replicate_pool, rep);
+}
+
+static void
+replicate_unlock (dpo_id_t *dpo)
+{
+    replicate_t *rep;
+
+    rep = replicate_get(dpo->dpoi_index);
+
+    rep->rep_locks--;
+
+    if (0 == rep->rep_locks)
+    {
+        replicate_destroy(rep);
+    }
+}
+
+static void
+replicate_mem_show (void)
+{
+    fib_show_memory_usage("replicate",
+			  pool_elts(replicate_pool),
+			  pool_len(replicate_pool),
+			  sizeof(replicate_t));
+}
+
+const static dpo_vft_t rep_vft = {
+    .dv_lock = replicate_lock,
+    .dv_unlock = replicate_unlock,
+    .dv_format = format_replicate_dpo,
+    .dv_mem_show = replicate_mem_show,
+};
+
+/**
+ * @brief The per-protocol VLIB graph nodes that are assigned to a replicate
+ *        object.
+ *
+ * this means that these graph nodes are ones from which a replicate is the
+ * parent object in the DPO-graph.
+ */
+const static char* const replicate_ip4_nodes[] =
+{
+    "ip4-replicate",
+    NULL,
+};
+const static char* const replicate_ip6_nodes[] =
+{
+    "ip6-replicate",
+    NULL,
+};
+const static char* const replicate_mpls_nodes[] =
+{
+    "mpls-replicate",
+    NULL,
+};
+
+const static char* const * const replicate_nodes[DPO_PROTO_NUM] =
+{
+    [DPO_PROTO_IP4]  = replicate_ip4_nodes,
+    [DPO_PROTO_IP6]  = replicate_ip6_nodes,
+    [DPO_PROTO_MPLS] = replicate_mpls_nodes,
+};
+
+void
+replicate_module_init (void)
+{
+    dpo_register(DPO_REPLICATE, &rep_vft, replicate_nodes);
+}
+
+static clib_error_t *
+replicate_show (vlib_main_t * vm,
+                unformat_input_t * input,
+                vlib_cli_command_t * cmd)
+{
+    index_t repi = INDEX_INVALID;
+
+    while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+    {
+        if (unformat (input, "%d", &repi))
+            ;
+        else
+            break;
+    }
+
+    if (INDEX_INVALID != repi)
+    {
+        vlib_cli_output (vm, "%U", format_replicate, repi,
+                         REPLICATE_FORMAT_DETAIL);
+    }
+    else
+    {
+        replicate_t *rep;
+
+        pool_foreach(rep, replicate_pool,
+        ({
+            vlib_cli_output (vm, "%U", format_replicate,
+                             replicate_get_index(rep),
+                             REPLICATE_FORMAT_NONE);
+        }));
+    }
+
+    return 0;
+}
+
+VLIB_CLI_COMMAND (replicate_show_command, static) = {
+    .path = "show replicate",
+    .short_help = "show replicate [<index>]",
+    .function = replicate_show,
+};
+
+typedef struct replicate_trace_t_
+{
+    index_t rep_index;
+    index_t dpo_index;
+    dpo_type_t dpo_type;
+} replicate_trace_t;
+
+static uword
+replicate_inline (vlib_main_t * vm,
+                  vlib_node_runtime_t * node,
+                  vlib_frame_t * frame)
+{
+    vlib_combined_counter_main_t * cm = &replicate_main.repm_counters;
+    u32 n_left_from, * from, * to_next, next_index;
+    u32 cpu_index = os_get_cpu_number();
+
+    from = vlib_frame_vector_args (frame);
+    n_left_from = frame->n_vectors;
+    next_index = node->cached_next_index;
+  
+    while (n_left_from > 0)
+    {
+        u32 n_left_to_next;
+
+        vlib_get_next_frame (vm, node, next_index,
+                             to_next, n_left_to_next);
+
+        while (n_left_from > 0 && n_left_to_next > 0)
+	{
+            u32 next0, ci0, bi0, bucket, repi0;
+            const replicate_t *rep0;
+            vlib_buffer_t * b0, *c0;
+            const dpo_id_t *dpo0;
+
+            bi0 = from[0];
+            to_next[0] = bi0;
+            from += 1;
+            to_next += 1;
+            n_left_from -= 1;
+            n_left_to_next -= 1;
+
+            b0 = vlib_get_buffer (vm, bi0);
+            repi0 = vnet_buffer (b0)->ip.adj_index[VLIB_TX];
+            rep0 = replicate_get(repi0);
+
+            vlib_increment_combined_counter(
+                cm, cpu_index, repi0, 1,
+                vlib_buffer_length_in_chain(vm, b0));
+
+            /* ship the original to the first bucket */
+            dpo0 = replicate_get_bucket_i(rep0, 0);
+            next0 = dpo0->dpoi_next_node;
+            vnet_buffer (b0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
+
+            if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+            {
+                replicate_trace_t *t = vlib_add_trace (vm, node, b0, sizeof (*t));
+                t->rep_index = repi0;
+                t->dpo_index = dpo0->dpoi_index;
+                t->dpo_type = dpo0->dpoi_type;
+            }
+            vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
+                                             to_next, n_left_to_next,
+                                             bi0, next0);
+
+            /* ship copies to the rest of the buckets */
+            for (bucket = 1; bucket < rep0->rep_n_buckets; bucket++)
+            {
+                /* Make a copy */
+                c0 = vlib_buffer_copy(vm, b0);
+                ci0 = vlib_get_buffer_index(vm, c0);
+
+                to_next[0] = ci0;
+                to_next += 1;
+                n_left_to_next -= 1;
+
+                dpo0 = replicate_get_bucket_i(rep0, bucket);
+                next0 = dpo0->dpoi_next_node;
+                vnet_buffer (c0)->ip.adj_index[VLIB_TX] = dpo0->dpoi_index;
+
+                if (PREDICT_FALSE(b0->flags & VLIB_BUFFER_IS_TRACED))
+                {
+                    replicate_trace_t *t = vlib_add_trace (vm, node, b0, sizeof (*t));
+                    t->rep_index = repi0;
+                    t->dpo_index = dpo0->dpoi_index;
+                    t->dpo_type = dpo0->dpoi_type;
+                }
+
+                vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
+                                                 to_next, n_left_to_next,
+                                                 ci0, next0);
+            }
+        }
+
+        vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+    }
+
+    return frame->n_vectors;
+}
+
+static u8 *
+format_replicate_trace (u8 * s, va_list * args)
+{
+  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *);
+  CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
+  replicate_trace_t *t = va_arg (*args, replicate_trace_t *);
+
+  s = format (s, "replicate: %d via %U:%d",
+              t->rep_index,
+              format_dpo_type, t->dpo_type,
+              t->dpo_index);
+  return s;
+}
+
+static uword
+ip4_replicate (vlib_main_t * vm,
+               vlib_node_runtime_t * node,
+               vlib_frame_t * frame)
+{
+    return (replicate_inline (vm, node, frame));
+}
+
+/**
+ * @brief
+ */
+VLIB_REGISTER_NODE (ip4_replicate_node) = {
+  .function = ip4_replicate,
+  .name = "ip4-replicate",
+  .vector_size = sizeof (u32),
+
+  .format_trace = format_replicate_trace,
+  .n_next_nodes = 1,
+  .next_nodes = {
+      [0] = "error-drop",
+  },
+};
+
+static uword
+ip6_replicate (vlib_main_t * vm,
+               vlib_node_runtime_t * node,
+               vlib_frame_t * frame)
+{
+    return (replicate_inline (vm, node, frame));
+}
+
+/**
+ * @brief
+ */
+VLIB_REGISTER_NODE (ip6_replicate_node) = {
+  .function = ip6_replicate,
+  .name = "ip6-replicate",
+  .vector_size = sizeof (u32),
+
+  .format_trace = format_replicate_trace,
+  .n_next_nodes = 1,
+  .next_nodes = {
+      [0] = "error-drop",
+  },
+};
diff --git a/src/vnet/dpo/replicate_dpo.h b/src/vnet/dpo/replicate_dpo.h
new file mode 100644
index 00000000000..a564739c9f2
--- /dev/null
+++ b/src/vnet/dpo/replicate_dpo.h
@@ -0,0 +1,143 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/**
+ * @brief
+ *
+ */
+
+#ifndef __REPLICATE_DPO_H__
+#define __REPLICATE_DPO_H__
+
+#include <vlib/vlib.h>
+#include <vnet/ip/lookup.h>
+#include <vnet/dpo/dpo.h>
+#include <vnet/dpo/load_balance.h>
+#include <vnet/fib/fib_types.h>
+
+/**
+ * replicate main
+ */
+typedef struct replicate_main_t_
+{
+    vlib_combined_counter_main_t repm_counters;
+} replicate_main_t;
+
+extern replicate_main_t replicate_main;
+
+/**
+ * The number of buckets that a load-balance object can have and still
+ * fit in one cache-line
+ */
+#define REP_NUM_INLINE_BUCKETS 4
+
+/**
+ * The FIB DPO provieds;
+ *  - load-balancing over the next DPOs in the chain/graph
+ *  - per-route counters
+ */
+typedef struct replicate_t_ {
+    /**
+     * number of buckets in the load-balance. always a power of 2.
+     */
+    u16 rep_n_buckets;
+
+   /**
+     * The protocol of packets that traverse this REP.
+     * need in combination with the flow hash config to determine how to hash.
+     * u8.
+     */
+    dpo_proto_t rep_proto;
+
+    /**
+     * The number of locks, which is approximately the number of users,
+     * of this load-balance.
+     * Load-balance objects of via-entries are heavily shared by recursives,
+     * so the lock count is a u32.
+     */
+    u32 rep_locks;
+
+    /**
+     * Vector of buckets containing the next DPOs, sized as repo_num
+     */
+    dpo_id_t *rep_buckets;
+
+    /**
+     * The rest of the cache line is used for buckets. In the common case
+     * where there there are less than 4 buckets, then the buckets are
+     * on the same cachlie and we save ourselves a pointer dereferance in 
+     * the data-path.
+     */
+    dpo_id_t rep_buckets_inline[REP_NUM_INLINE_BUCKETS];
+} replicate_t;
+
+STATIC_ASSERT(sizeof(replicate_t) <= CLIB_CACHE_LINE_BYTES,
+	      "A replicate object size exceeds one cachline");
+
+/**
+ * Flags controlling load-balance formatting/display
+ */
+typedef enum replicate_format_flags_t_ {
+    REPLICATE_FORMAT_NONE,
+    REPLICATE_FORMAT_DETAIL = (1 << 0),
+} replicate_format_flags_t;
+
+extern index_t replicate_create(u32 num_buckets,
+                                dpo_proto_t rep_proto);
+extern void replicate_multipath_update(
+    const dpo_id_t *dpo,
+    load_balance_path_t *next_hops);
+
+extern void replicate_set_bucket(index_t repi,
+				    u32 bucket,
+				    const dpo_id_t *next);
+
+extern u8* format_replicate(u8 * s, va_list * args);
+
+extern const dpo_id_t *replicate_get_bucket(index_t repi,
+					       u32 bucket);
+extern int replicate_is_drop(const dpo_id_t *dpo);
+
+/**
+ * The encapsulation breakages are for fast DP access
+ */
+extern replicate_t *replicate_pool;
+static inline replicate_t*
+replicate_get (index_t repi)
+{
+    return (pool_elt_at_index(replicate_pool, repi));
+}
+
+#define REP_HAS_INLINE_BUCKETS(_rep)		\
+    ((_rep)->rep_n_buckets <= REP_NUM_INLINE_BUCKETS)
+
+static inline const dpo_id_t *
+replicate_get_bucket_i (const replicate_t *rep,
+			   u32 bucket)
+{
+    ASSERT(bucket < rep->rep_n_buckets);
+
+    if (PREDICT_TRUE(REP_HAS_INLINE_BUCKETS(rep)))
+    {
+	return (&rep->rep_buckets_inline[bucket]);
+    }
+    else
+    {
+	return (&rep->rep_buckets[bucket]);
+    }
+}
+
+extern void replicate_module_init(void);
+
+#endif