Adjacency layout change and move to vnet/adj

Change-Id: I03195a86c69f84a301051c6b3ab64456bbf28645
Signed-off-by: Neale Ranns <nranns@cisco.com>

7 files changed, 824 insertions, 38 deletions

diff --git a/src/vnet/adj/adj.c b/src/vnet/adj/adj.c
index c1d036a0e8b..7cf9e9d081d 100644
--- a/src/vnet/adj/adj.c
+++ b/src/vnet/adj/adj.c
@@ -48,7 +48,7 @@ adj_alloc (fib_protocol_t proto)
 {
     ip_adjacency_t *adj;
 
-    pool_get(adj_pool, adj);
+    pool_get_aligned(adj_pool, adj, CLIB_CACHE_LINE_BYTES);
 
     adj_poison(adj);
 
@@ -58,7 +58,6 @@ adj_alloc (fib_protocol_t proto)
                                    adj_get_index(adj));
 
     adj->rewrite_header.sw_if_index = ~0;
-    adj->n_adj = 1;
     adj->lookup_next_index = 0;
 
     fib_node_init(&adj->ia_node,
diff --git a/src/vnet/adj/adj.h b/src/vnet/adj/adj.h
index 271fdbc6114..af7730f7086 100644
--- a/src/vnet/adj/adj.h
+++ b/src/vnet/adj/adj.h
@@ -24,7 +24,6 @@
  *            address in the ARP packet.
  *          UNSHARED. Only one per-interface.
  *   - midchain: a nighbour adj on a virtual/tunnel interface.
- *   - rewrite: an adj with no key, but with a rewrite string.
  *
  * The API to create and update the adjacency is very sub-type specific. This
  * is intentional as it encourages the user to carefully consider which adjacency
@@ -42,10 +41,227 @@
 #ifndef __ADJ_H__
 #define __ADJ_H__
 
-#include <vnet/ip/lookup.h>
 #include <vnet/adj/adj_types.h>
 #include <vnet/adj/adj_nbr.h>
 #include <vnet/adj/adj_glean.h>
+#include <vnet/adj/rewrite.h>
+
+/** @brief Common (IP4/IP6) next index stored in adjacency. */
+typedef enum
+{
+  /** Adjacency to drop this packet. */
+  IP_LOOKUP_NEXT_DROP,
+  /** Adjacency to punt this packet. */
+  IP_LOOKUP_NEXT_PUNT,
+
+  /** This packet is for one of our own IP addresses. */
+  IP_LOOKUP_NEXT_LOCAL,
+
+  /** This packet matches an "incomplete adjacency" and packets
+     need to be passed to ARP to find rewrite string for
+     this destination. */
+  IP_LOOKUP_NEXT_ARP,
+
+  /** This packet matches an "interface route" and packets
+     need to be passed to ARP to find rewrite string for
+     this destination. */
+  IP_LOOKUP_NEXT_GLEAN,
+
+  /** This packet is to be rewritten and forwarded to the next
+     processing node.  This is typically the output interface but
+     might be another node for further output processing. */
+  IP_LOOKUP_NEXT_REWRITE,
+
+  /** This packets follow a mid-chain adjacency */
+  IP_LOOKUP_NEXT_MIDCHAIN,
+
+  /** This packets needs to go to ICMP error */
+  IP_LOOKUP_NEXT_ICMP_ERROR,
+
+  /** Multicast Adjacency. */
+  IP_LOOKUP_NEXT_MCAST,
+
+  IP_LOOKUP_N_NEXT,
+} __attribute__ ((packed)) ip_lookup_next_t;
+
+typedef enum
+{
+  IP4_LOOKUP_N_NEXT = IP_LOOKUP_N_NEXT,
+} ip4_lookup_next_t;
+
+typedef enum
+{
+  /* Hop-by-hop header handling */
+  IP6_LOOKUP_NEXT_HOP_BY_HOP = IP_LOOKUP_N_NEXT,
+  IP6_LOOKUP_NEXT_ADD_HOP_BY_HOP,
+  IP6_LOOKUP_NEXT_POP_HOP_BY_HOP,
+  IP6_LOOKUP_N_NEXT,
+} ip6_lookup_next_t;
+
+#define IP4_LOOKUP_NEXT_NODES {					\
+    [IP_LOOKUP_NEXT_DROP] = "ip4-drop",				\
+    [IP_LOOKUP_NEXT_PUNT] = "ip4-punt",				\
+    [IP_LOOKUP_NEXT_LOCAL] = "ip4-local",			\
+    [IP_LOOKUP_NEXT_ARP] = "ip4-arp",				\
+    [IP_LOOKUP_NEXT_GLEAN] = "ip4-glean",			\
+    [IP_LOOKUP_NEXT_REWRITE] = "ip4-rewrite",    		\
+    [IP_LOOKUP_NEXT_MCAST] = "ip4-rewrite-mcast",	        \
+    [IP_LOOKUP_NEXT_MIDCHAIN] = "ip4-midchain",		        \
+    [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip4-icmp-error",		\
+}
+
+#define IP6_LOOKUP_NEXT_NODES {					\
+    [IP_LOOKUP_NEXT_DROP] = "ip6-drop",				\
+    [IP_LOOKUP_NEXT_PUNT] = "ip6-punt",				\
+    [IP_LOOKUP_NEXT_LOCAL] = "ip6-local",			\
+    [IP_LOOKUP_NEXT_ARP] = "ip6-discover-neighbor",		\
+    [IP_LOOKUP_NEXT_GLEAN] = "ip6-glean",			\
+    [IP_LOOKUP_NEXT_REWRITE] = "ip6-rewrite",			\
+    [IP_LOOKUP_NEXT_MCAST] = "ip6-rewrite-mcast",		\
+    [IP_LOOKUP_NEXT_MIDCHAIN] = "ip6-midchain",			\
+    [IP_LOOKUP_NEXT_ICMP_ERROR] = "ip6-icmp-error",		\
+    [IP6_LOOKUP_NEXT_HOP_BY_HOP] = "ip6-hop-by-hop",		\
+    [IP6_LOOKUP_NEXT_ADD_HOP_BY_HOP] = "ip6-add-hop-by-hop",	\
+    [IP6_LOOKUP_NEXT_POP_HOP_BY_HOP] = "ip6-pop-hop-by-hop",	\
+}
+
+/**
+ * Forward delcartion
+ */
+struct ip_adjacency_t_;
+
+/**
+ * @brief A function type for post-rewrite fixups on midchain adjacency
+ */
+typedef void (*adj_midchain_fixup_t) (vlib_main_t * vm,
+				      struct ip_adjacency_t_ * adj,
+				      vlib_buffer_t * b0);
+
+/**
+ * @brief Flags on an IP adjacency
+ */
+typedef enum ip_adjacency_flags_t_
+{
+    ADJ_FLAG_NONE = 0,
+
+    /**
+     * Currently a sync walk is active. Used to prevent re-entrant walking
+     */
+    ADJ_FLAG_SYNC_WALK_ACTIVE = (1 << 0),
+
+    /**
+     * Packets TX through the midchain do not increment the interface
+     * counters. This should be used when the adj is associated with an L2
+     * interface and that L2 interface is in a bridege domain. In that case
+     * the packet will have traversed the interface's TX node, and hence have
+     * been counted, before it traverses ths midchain
+     */
+    ADJ_FLAG_MIDCHAIN_NO_COUNT = (1 << 1),
+}  __attribute__ ((packed)) adj_flags_t;
+
+/**
+ * @brief IP unicast adjacency.
+ *  @note cache aligned.
+ *
+ * An adjacency is a represenation of a peer on a particular link.
+ */
+typedef struct ip_adjacency_t_
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+
+  /**
+   * Linkage into the FIB node grpah. First member since this type
+   * has 8 byte alignment requirements.
+   */
+  fib_node_t ia_node;
+
+  /**
+   * Next hop after ip4-lookup.
+   *  This is not accessed in the rewrite nodes.
+   * 1-bytes
+   */
+  ip_lookup_next_t lookup_next_index;
+
+  /**
+   * link/ether-type
+   * 1 bytes
+   */
+  vnet_link_t ia_link;
+
+  /**
+   * The protocol of the neighbor/peer. i.e. the protocol with
+   * which to interpret the 'next-hop' attirbutes of the sub-types.
+   * 1-btyes
+   */
+  fib_protocol_t ia_nh_proto;
+
+  /**
+   * Flags on the adjacency
+   * 1-bytes
+   */
+  adj_flags_t ia_flags;
+
+  union
+  {
+    /**
+     * IP_LOOKUP_NEXT_ARP/IP_LOOKUP_NEXT_REWRITE
+     *
+     * neighbour adjacency sub-type;
+     */
+    struct
+    {
+      ip46_address_t next_hop;
+    } nbr;
+      /**
+       * IP_LOOKUP_NEXT_MIDCHAIN
+       *
+       * A nbr adj that is also recursive. Think tunnels.
+       * A nbr adj can transition to be of type MDICHAIN
+       * so be sure to leave the two structs with the next_hop
+       * fields aligned.
+       */
+    struct
+    {
+      /**
+       * The recursive next-hop.
+       *  This field MUST be at the same memory location as
+       *   sub_type.nbr.next_hop
+       */
+      ip46_address_t next_hop;
+      /**
+       * The next DPO to use
+       */
+      dpo_id_t next_dpo;
+      /**
+       * A function to perform the post-rewrite fixup
+       */
+      adj_midchain_fixup_t fixup_func;
+    } midchain;
+    /**
+     * IP_LOOKUP_NEXT_GLEAN
+     *
+     * Glean the address to ARP for from the packet's destination.
+     * Technically these aren't adjacencies, i.e. they are not a
+     * representation of a peer. One day we might untangle this coupling
+     * and use a new Glean DPO.
+     */
+    struct
+    {
+      ip46_address_t receive_addr;
+    } glean;
+  } sub_type;
+
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline1);
+
+  /* Rewrite in second/third cache lines */
+  vnet_declare_rewrite (VLIB_BUFFER_PRE_DATA_SIZE);
+} ip_adjacency_t;
+
+STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, cacheline0) == 0),
+	       "IP adjacency cachline 0 is not offset");
+STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, cacheline1) ==
+		CLIB_CACHE_LINE_BYTES),
+	       "IP adjacency cachline 1 is more than one cachline size offset");
 
 /**
  * @brief
diff --git a/src/vnet/adj/adj_midchain.c b/src/vnet/adj/adj_midchain.c
index 55b5e44bc43..e8087f08d21 100644
--- a/src/vnet/adj/adj_midchain.c
+++ b/src/vnet/adj/adj_midchain.c
@@ -384,6 +384,14 @@ adj_midchain_get_feature_arc_index_for_link_type (const ip_adjacency_t *adj)
     return (arc);
 }
 
+static u32
+adj_nbr_midchain_get_tx_node (ip_adjacency_t *adj)
+{
+    return ((adj->ia_flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) ?
+            adj_midchain_tx_no_count_node.index :
+            adj_midchain_tx_node.index);
+}
+
 /**
  * adj_nbr_midchain_update_rewrite
  *
@@ -394,12 +402,12 @@ adj_midchain_get_feature_arc_index_for_link_type (const ip_adjacency_t *adj)
 void
 adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
 				 adj_midchain_fixup_t fixup,
-				 adj_midchain_flag_t flags,
+				 adj_flags_t flags,
 				 u8 *rewrite)
 {
+    u32 feature_index, tx_node;
     ip_adjacency_t *adj;
     u8 arc_index;
-    u32 feature_index;
 
     ASSERT(ADJ_INDEX_INVALID != adj_index);
 
@@ -416,15 +424,14 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
     ASSERT(NULL != rewrite);
 
     adj->sub_type.midchain.fixup_func = fixup;
+    adj->ia_flags |= flags;
 
     arc_index = adj_midchain_get_feature_arc_index_for_link_type (adj);
-    feature_index = (flags & ADJ_MIDCHAIN_FLAG_NO_COUNT) ?
+    feature_index = (flags & ADJ_FLAG_MIDCHAIN_NO_COUNT) ?
                     adj_midchain_tx_no_count_feature_node[adj->ia_link] :
                     adj_midchain_tx_feature_node[adj->ia_link];
 
-    adj->sub_type.midchain.tx_function_node = (flags & ADJ_MIDCHAIN_FLAG_NO_COUNT) ?
-                                               adj_midchain_tx_no_count_node.index :
-                                               adj_midchain_tx_node.index;
+    tx_node = adj_nbr_midchain_get_tx_node(adj);
 
     vnet_feature_enable_disable_with_index (arc_index, feature_index,
 					    adj->rewrite_header.sw_if_index,
@@ -437,7 +444,7 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
      * node are any output features, then the midchain-tx.  from there we
      * need to get to the stacked child's node.
      */
-    dpo_stack_from_node(adj->sub_type.midchain.tx_function_node,
+    dpo_stack_from_node(tx_node,
 			&adj->sub_type.midchain.next_dpo,
 			drop_dpo_get(vnet_link_to_dpo_proto(adj->ia_link)));
 
@@ -447,7 +454,7 @@ adj_nbr_midchain_update_rewrite (adj_index_t adj_index,
     adj_nbr_update_rewrite_internal(adj,
 				    IP_LOOKUP_NEXT_MIDCHAIN,
 				    adj_get_midchain_node(adj->ia_link),
-				    adj->sub_type.midchain.tx_function_node,
+				    tx_node,
 				    rewrite);
 }
 
@@ -491,7 +498,7 @@ adj_nbr_midchain_stack (adj_index_t adj_index,
 
     ASSERT(IP_LOOKUP_NEXT_MIDCHAIN == adj->lookup_next_index);
 
-    dpo_stack_from_node(adj->sub_type.midchain.tx_function_node,
+    dpo_stack_from_node(adj_nbr_midchain_get_tx_node(adj),
 			&adj->sub_type.midchain.next_dpo,
 			next);
 }
diff --git a/src/vnet/adj/adj_midchain.h b/src/vnet/adj/adj_midchain.h
index ae414aea6dc..27ca1d3398d 100644
--- a/src/vnet/adj/adj_midchain.h
+++ b/src/vnet/adj/adj_midchain.h
@@ -25,26 +25,6 @@
 #include <vnet/adj/adj.h>
 
 /**
- * @brief Flags controlling the midchain adjacency
- */
-typedef enum adj_midchain_flag_t_
-{
-    /**
-     * No flags
-     */
-    ADJ_MIDCHAIN_FLAG_NONE = 0,
-
-    /**
-     * Packets TX through the midchain do not increment the interface
-     * counters. This should be used when the adj is associated with an L2
-     * interface and that L2 interface is in a bridege domain. In that case
-     * the packet will have traversed the interface's TX node, and hence have
-     * been counted, before it traverses ths midchain
-     */
-    ADJ_MIDCHAIN_FLAG_NO_COUNT = (1 << 0),
-} adj_midchain_flag_t;
-
-/**
  * @brief
  *  Convert an existing neighbour adjacency into a midchain
  *
@@ -60,7 +40,7 @@ typedef enum adj_midchain_flag_t_
  */
 extern void adj_nbr_midchain_update_rewrite(adj_index_t adj_index,
 					    adj_midchain_fixup_t fixup,
-					    adj_midchain_flag_t flags,
+					    adj_flags_t flags,
 					    u8 *rewrite);
 
 /**
diff --git a/src/vnet/adj/adj_nbr.c b/src/vnet/adj/adj_nbr.c
index 072abd0300d..ddacb030f1d 100644
--- a/src/vnet/adj/adj_nbr.c
+++ b/src/vnet/adj/adj_nbr.c
@@ -333,7 +333,7 @@ adj_nbr_update_rewrite (adj_index_t adj_index,
  */
 void
 adj_nbr_update_rewrite_internal (ip_adjacency_t *adj,
-				 u32 adj_next_index,
+				 ip_lookup_next_t adj_next_index,
 				 u32 this_node,
 				 u32 next_node,
 				 u8 *rewrite)
@@ -367,7 +367,7 @@ adj_nbr_update_rewrite_internal (ip_adjacency_t *adj,
     if (ADJ_INDEX_INVALID != walk_ai)
     {
         walk_adj = adj_get(walk_ai);
-        if (IP_ADJ_SYNC_WALK_ACTIVE & walk_adj->ia_flags)
+        if (ADJ_FLAG_SYNC_WALK_ACTIVE & walk_adj->ia_flags)
         {
             do_walk = 0;
         }
@@ -376,7 +376,7 @@ adj_nbr_update_rewrite_internal (ip_adjacency_t *adj,
             /*
              * Prevent re-entrant walk of the same adj
              */
-            walk_adj->ia_flags |= IP_ADJ_SYNC_WALK_ACTIVE;
+            walk_adj->ia_flags |= ADJ_FLAG_SYNC_WALK_ACTIVE;
             do_walk = 1;
         }
     }
@@ -502,7 +502,7 @@ adj_nbr_update_rewrite_internal (ip_adjacency_t *adj,
      */
     if (do_walk)
     {
-        walk_adj->ia_flags &= ~IP_ADJ_SYNC_WALK_ACTIVE;
+        walk_adj->ia_flags &= ~ADJ_FLAG_SYNC_WALK_ACTIVE;
     }
 
     adj_unlock(adj_get_index(adj));
diff --git a/src/vnet/adj/rewrite.c b/src/vnet/adj/rewrite.c
new file mode 100644
index 00000000000..47fb74df01b
--- /dev/null
+++ b/src/vnet/adj/rewrite.c
@@ -0,0 +1,234 @@
+/*
+ * Copyright (c) 2015 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.
+ */
+/*
+ * rewrite.c: packet rewrite
+ *
+ * Copyright (c) 2008 Eliot Dresselhaus
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ *  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ *  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ *  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ *  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ *  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ *  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <vnet/vnet.h>
+#include <vnet/ip/lookup.h>
+
+void
+vnet_rewrite_copy_slow_path (vnet_rewrite_data_t * p0,
+			     vnet_rewrite_data_t * rw0,
+			     word n_left, uword most_likely_size)
+{
+  uword n_done =
+    round_pow2 (most_likely_size, sizeof (rw0[0])) / sizeof (rw0[0]);
+
+  p0 -= n_done;
+  rw0 -= n_done;
+
+  /* As we enter the cleanup loop, p0 and rw0 point to the last chunk written
+     by the fast path. Hence, the constant 1, which the
+     vnet_rewrite_copy_one macro renders as p0[-1] = rw0[-1]. */
+
+  while (n_left > 0)
+    {
+      vnet_rewrite_copy_one (p0, rw0, 1);
+      p0--;
+      rw0--;
+      n_left--;
+    }
+}
+
+u8 *
+format_vnet_rewrite (u8 * s, va_list * args)
+{
+  vnet_rewrite_header_t *rw = va_arg (*args, vnet_rewrite_header_t *);
+  u32 max_data_bytes = va_arg (*args, u32);
+  CLIB_UNUSED (uword indent) = va_arg (*args, u32);
+  vnet_main_t *vnm = vnet_get_main ();
+
+  if (rw->sw_if_index != ~0)
+    {
+      vnet_sw_interface_t *si;
+      si = vnet_get_sw_interface_safe (vnm, rw->sw_if_index);
+      if (NULL != si)
+	s = format (s, "%U: ", format_vnet_sw_interface_name, vnm, si);
+      else
+	s = format (s, "DELETED:%d", rw->sw_if_index);
+    }
+
+  /* Format rewrite string. */
+  if (rw->data_bytes > 0)
+
+    s = format (s, "%U",
+		format_hex_bytes,
+		rw->data + max_data_bytes - rw->data_bytes, rw->data_bytes);
+
+  return s;
+}
+
+u32
+vnet_tx_node_index_for_sw_interface (vnet_main_t * vnm, u32 sw_if_index)
+{
+  vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  return (hw->output_node_index);
+}
+
+void
+vnet_rewrite_init (vnet_main_t * vnm,
+		   u32 sw_if_index,
+		   u32 this_node, u32 next_node, vnet_rewrite_header_t * rw)
+{
+  rw->sw_if_index = sw_if_index;
+  rw->next_index = vlib_node_add_next (vnm->vlib_main, this_node, next_node);
+  rw->max_l3_packet_bytes =
+    vnet_sw_interface_get_mtu (vnm, sw_if_index, VLIB_TX);
+}
+
+void
+vnet_rewrite_for_sw_interface (vnet_main_t * vnm,
+			       vnet_link_t link_type,
+			       u32 sw_if_index,
+			       u32 node_index,
+			       void *dst_address,
+			       vnet_rewrite_header_t * rw,
+			       u32 max_rewrite_bytes)
+{
+
+  vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  vnet_hw_interface_class_t *hc =
+    vnet_get_hw_interface_class (vnm, hw->hw_class_index);
+  u8 *rewrite = NULL;
+
+  vnet_rewrite_init (vnm, sw_if_index, node_index,
+		     vnet_tx_node_index_for_sw_interface (vnm, sw_if_index),
+		     rw);
+
+  ASSERT (hc->build_rewrite);
+  rewrite = hc->build_rewrite (vnm, sw_if_index, link_type, dst_address);
+
+  ASSERT (vec_len (rewrite) < max_rewrite_bytes);
+  vnet_rewrite_set_data_internal (rw, max_rewrite_bytes, rewrite,
+				  vec_len (rewrite));
+  vec_free (rewrite);
+}
+
+void
+vnet_rewrite_for_tunnel (vnet_main_t * vnm,
+			 u32 tx_sw_if_index,
+			 u32 rewrite_node_index,
+			 u32 post_rewrite_node_index,
+			 vnet_rewrite_header_t * rw,
+			 u8 * rewrite_data, u32 rewrite_length)
+{
+  ip_adjacency_t *adj = 0;
+  /*
+   * Installed into vnet_buffer(b)->sw_if_index[VLIB_TX] e.g.
+   * by ip4_rewrite_inline. If the post-rewrite node injects into
+   * ipX-forward, this will be interpreted as a FIB number.
+   */
+  rw->sw_if_index = tx_sw_if_index;
+  rw->next_index = vlib_node_add_next (vnm->vlib_main, rewrite_node_index,
+				       post_rewrite_node_index);
+  rw->max_l3_packet_bytes = (u16) ~ 0;	/* we can't know at this point */
+
+  ASSERT (rewrite_length < sizeof (adj->rewrite_data));
+  /* Leave room for ethernet + VLAN tag */
+  vnet_rewrite_set_data_internal (rw, sizeof (adj->rewrite_data),
+				  rewrite_data, rewrite_length);
+}
+
+void
+serialize_vnet_rewrite (serialize_main_t * m, va_list * va)
+{
+  vnet_rewrite_header_t *rw = va_arg (*va, vnet_rewrite_header_t *);
+  u32 max_data_bytes = va_arg (*va, u32);
+  u8 *p;
+
+  serialize_integer (m, rw->sw_if_index, sizeof (rw->sw_if_index));
+  serialize_integer (m, rw->data_bytes, sizeof (rw->data_bytes));
+  serialize_integer (m, rw->max_l3_packet_bytes,
+		     sizeof (rw->max_l3_packet_bytes));
+  p = serialize_get (m, rw->data_bytes);
+  clib_memcpy (p, vnet_rewrite_get_data_internal (rw, max_data_bytes),
+	       rw->data_bytes);
+}
+
+void
+unserialize_vnet_rewrite (serialize_main_t * m, va_list * va)
+{
+  vnet_rewrite_header_t *rw = va_arg (*va, vnet_rewrite_header_t *);
+  u32 max_data_bytes = va_arg (*va, u32);
+  u8 *p;
+
+  /* It is up to user to fill these in. */
+  rw->next_index = ~0;
+
+  unserialize_integer (m, &rw->sw_if_index, sizeof (rw->sw_if_index));
+  unserialize_integer (m, &rw->data_bytes, sizeof (rw->data_bytes));
+  unserialize_integer (m, &rw->max_l3_packet_bytes,
+		       sizeof (rw->max_l3_packet_bytes));
+  p = unserialize_get (m, rw->data_bytes);
+  clib_memcpy (vnet_rewrite_get_data_internal (rw, max_data_bytes), p,
+	       rw->data_bytes);
+}
+
+u8 *
+vnet_build_rewrite_for_sw_interface (vnet_main_t * vnm,
+				     u32 sw_if_index,
+				     vnet_link_t link_type,
+				     const void *dst_address)
+{
+  vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  vnet_hw_interface_class_t *hc =
+    vnet_get_hw_interface_class (vnm, hw->hw_class_index);
+
+  ASSERT (hc->build_rewrite);
+  return (hc->build_rewrite (vnm, sw_if_index, link_type, dst_address));
+}
+
+
+void
+vnet_update_adjacency_for_sw_interface (vnet_main_t * vnm,
+					u32 sw_if_index, u32 ai)
+{
+  vnet_hw_interface_t *hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  vnet_hw_interface_class_t *hc =
+    vnet_get_hw_interface_class (vnm, hw->hw_class_index);
+
+  ASSERT (hc->update_adjacency);
+  hc->update_adjacency (vnm, sw_if_index, ai);
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/adj/rewrite.h b/src/vnet/adj/rewrite.h
new file mode 100644
index 00000000000..1dea72f5ed8
--- /dev/null
+++ b/src/vnet/adj/rewrite.h
@@ -0,0 +1,350 @@
+/*
+ * Copyright (c) 2015 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.
+ */
+/*
+ * rewrite.h: packet rewrite
+ *
+ * Copyright (c) 2008 Eliot Dresselhaus
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ *  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ *  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ *  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ *  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ *  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ *  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef included_vnet_rewrite_h
+#define included_vnet_rewrite_h
+
+#include <vlib/vlib.h>
+#include <vnet/l3_types.h>
+
+/* Consider using vector types for speed? */
+typedef uword vnet_rewrite_data_t;
+
+/**
+ * Flags associated with the rewrite/adjacency
+ */
+typedef enum vnet_rewrite_flags_t_
+{
+  /**
+   * This adjacency/interface has output features configured
+   */
+  VNET_REWRITE_HAS_FEATURES = (1 << 0),
+} __attribute__ ((packed)) vnet_rewrite_flags_t;
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED (struct {
+  /* Interface to mark re-written packets with. */
+  u32 sw_if_index;
+
+  /* Next node to feed after packet rewrite is done. */
+  u16 next_index;
+
+  /* Number of bytes in rewrite data. */
+  u16 data_bytes;
+
+  /* Max packet size layer 3 (MTU) for output interface.
+     Used for MTU check after packet rewrite. */
+  u16 max_l3_packet_bytes;
+
+  /* Data-plane flags on the adjacency/rewrite */
+  vnet_rewrite_flags_t flags;
+
+  /* When dynamically writing a multicast destination L2 addresss
+   * this is the offset from the IP address at which to write in the
+   * IP->MAC address translation.
+   */
+  u8 dst_mcast_offset;
+
+  /* The mask to apply to the lower 4 bytes of the IP address before ORing
+   * into the destinaiton MAC address */
+  u32 dst_mcast_mask;
+
+  /* Rewrite string starting at end and going backwards. */
+  u8 data[0];
+}) vnet_rewrite_header_t;
+/* *INDENT-ON* */
+
+/**
+ * At 16 bytes of rewrite herader we have enought space left for a IPv6
+ * (40 bytes) + LISP-GPE (8 bytes) in the cache line
+ */
+STATIC_ASSERT (sizeof (vnet_rewrite_header_t) <= 16,
+	       "Rewrite header too big");
+
+/*
+  Helper macro for declaring rewrite string w/ given max-size.
+
+  Typical usage:
+    typedef struct {
+      //
+      int a, b;
+
+      // Total adjacency is 64 bytes.
+      vnet_rewrite_declare(64 - 2*sizeof(int)) rw;
+    } my_adjacency_t;
+*/
+#define vnet_declare_rewrite(total_bytes)				\
+struct {								\
+  vnet_rewrite_header_t rewrite_header;  			        \
+									\
+  u8 rewrite_data[(total_bytes) - sizeof (vnet_rewrite_header_t)];	\
+}
+
+always_inline void
+vnet_rewrite_clear_data_internal (vnet_rewrite_header_t * rw, int max_size)
+{
+  /* Sanity check values carefully for this memset operation */
+  ASSERT ((max_size > 0) && (max_size < VLIB_BUFFER_PRE_DATA_SIZE));
+
+  rw->data_bytes = 0;
+  memset (rw->data, 0xfe, max_size);
+}
+
+always_inline void
+vnet_rewrite_set_data_internal (vnet_rewrite_header_t * rw,
+				int max_size, void *data, int data_bytes)
+{
+  /* Sanity check values carefully for this memset operation */
+  ASSERT ((max_size > 0) && (max_size < VLIB_BUFFER_PRE_DATA_SIZE));
+  ASSERT ((data_bytes >= 0) && (data_bytes < max_size));
+
+  rw->data_bytes = data_bytes;
+  clib_memcpy (rw->data + max_size - data_bytes, data, data_bytes);
+  memset (rw->data, 0xfe, max_size - data_bytes);
+}
+
+#define vnet_rewrite_set_data(rw,data,data_bytes)		\
+  vnet_rewrite_set_data_internal (&((rw).rewrite_header),	\
+				  sizeof ((rw).rewrite_data),	\
+				  (data),			\
+				  (data_bytes))
+
+always_inline void *
+vnet_rewrite_get_data_internal (vnet_rewrite_header_t * rw, int max_size)
+{
+  ASSERT (rw->data_bytes <= max_size);
+  return rw->data + max_size - rw->data_bytes;
+}
+
+#define vnet_rewrite_get_data(rw) \
+  vnet_rewrite_get_data_internal (&((rw).rewrite_header), sizeof ((rw).rewrite_data))
+
+always_inline void
+vnet_rewrite_copy_one (vnet_rewrite_data_t * p0, vnet_rewrite_data_t * rw0,
+		       int i)
+{
+  p0[-i] = rw0[-i];
+}
+
+void vnet_rewrite_copy_slow_path (vnet_rewrite_data_t * p0,
+				  vnet_rewrite_data_t * rw0,
+				  word n_left, uword most_likely_size);
+
+/* *INDENT-OFF* */
+typedef CLIB_PACKED (struct {
+  u64 a;
+  u32 b;
+  u16 c;
+}) eh_copy_t;
+/* *INDENT-ON* */
+
+always_inline void
+_vnet_rewrite_one_header (vnet_rewrite_header_t * h0,
+			  void *packet0, int max_size, int most_likely_size)
+{
+  vnet_rewrite_data_t *p0 = packet0;
+  vnet_rewrite_data_t *rw0 = (vnet_rewrite_data_t *) (h0->data + max_size);
+  word n_left0;
+
+  /* 0xfefe => poisoned adjacency => crash */
+  ASSERT (h0->data_bytes != 0xfefe);
+
+  if (PREDICT_TRUE (h0->data_bytes == sizeof (eh_copy_t)))
+    {
+      eh_copy_t *s, *d;
+      s = (eh_copy_t *) (h0->data + max_size - sizeof (eh_copy_t));
+      d = (eh_copy_t *) (((u8 *) packet0) - sizeof (eh_copy_t));
+      clib_memcpy (d, s, sizeof (eh_copy_t));
+      return;
+    }
+
+
+#define _(i)								\
+  do {									\
+    if (most_likely_size > ((i)-1)*sizeof (vnet_rewrite_data_t))	\
+      vnet_rewrite_copy_one (p0, rw0, (i));				\
+  } while (0)
+
+  _(4);
+  _(3);
+  _(2);
+  _(1);
+
+#undef _
+
+  n_left0 = (int)
+    (((int) h0->data_bytes - most_likely_size) + (sizeof (rw0[0]) - 1))
+    / (int) sizeof (rw0[0]);
+  if (PREDICT_FALSE (n_left0 > 0))
+    vnet_rewrite_copy_slow_path (p0, rw0, n_left0, most_likely_size);
+}
+
+always_inline void
+_vnet_rewrite_two_headers (vnet_rewrite_header_t * h0,
+			   vnet_rewrite_header_t * h1,
+			   void *packet0,
+			   void *packet1, int max_size, int most_likely_size)
+{
+  vnet_rewrite_data_t *p0 = packet0;
+  vnet_rewrite_data_t *p1 = packet1;
+  vnet_rewrite_data_t *rw0 = (vnet_rewrite_data_t *) (h0->data + max_size);
+  vnet_rewrite_data_t *rw1 = (vnet_rewrite_data_t *) (h1->data + max_size);
+  word n_left0, n_left1;
+  int slow_path;
+
+  /* 0xfefe => poisoned adjacency => crash */
+  ASSERT (h0->data_bytes != 0xfefe);
+  ASSERT (h1->data_bytes != 0xfefe);
+
+  /* Arithmetic calculation: bytes0 == bytes1 == 14 */
+  slow_path = h0->data_bytes ^ h1->data_bytes;
+  slow_path += h0->data_bytes ^ sizeof (eh_copy_t);
+
+  if (PREDICT_TRUE (slow_path == 0))
+    {
+      eh_copy_t *s0, *d0, *s1, *d1;
+      s0 = (eh_copy_t *) (h0->data + max_size - sizeof (eh_copy_t));
+      d0 = (eh_copy_t *) (((u8 *) packet0) - sizeof (eh_copy_t));
+      clib_memcpy (d0, s0, sizeof (eh_copy_t));
+      s1 = (eh_copy_t *) (h1->data + max_size - sizeof (eh_copy_t));
+      d1 = (eh_copy_t *) (((u8 *) packet1) - sizeof (eh_copy_t));
+      clib_memcpy (d1, s1, sizeof (eh_copy_t));
+      return;
+    }
+
+#define _(i)								\
+  do {									\
+    if (most_likely_size > ((i)-1)*sizeof (vnet_rewrite_data_t))	\
+      {									\
+	vnet_rewrite_copy_one (p0, rw0, (i));				\
+	vnet_rewrite_copy_one (p1, rw1, (i));				\
+      }									\
+  } while (0)
+
+  _(4);
+  _(3);
+  _(2);
+  _(1);
+
+#undef _
+
+  n_left0 = (int)
+    (((int) h0->data_bytes - most_likely_size) + (sizeof (rw0[0]) - 1))
+    / (int) sizeof (rw0[0]);
+  n_left1 = (int)
+    (((int) h1->data_bytes - most_likely_size) + (sizeof (rw1[0]) - 1))
+    / (int) sizeof (rw1[0]);
+
+  if (PREDICT_FALSE (n_left0 > 0 || n_left1 > 0))
+    {
+      vnet_rewrite_copy_slow_path (p0, rw0, n_left0, most_likely_size);
+      vnet_rewrite_copy_slow_path (p1, rw1, n_left1, most_likely_size);
+    }
+}
+
+#define vnet_rewrite_one_header(rw0,p0,most_likely_size)	\
+  _vnet_rewrite_one_header (&((rw0).rewrite_header), (p0),	\
+			    sizeof ((rw0).rewrite_data),	\
+			    (most_likely_size))
+
+#define vnet_rewrite_two_headers(rw0,rw1,p0,p1,most_likely_size)	\
+  _vnet_rewrite_two_headers (&((rw0).rewrite_header), &((rw1).rewrite_header), \
+			     (p0), (p1),				\
+			     sizeof ((rw0).rewrite_data),		\
+			     (most_likely_size))
+
+always_inline void
+_vnet_fixup_one_header (vnet_rewrite_header_t * h0,
+			u8 * addr, u32 addr_len, u8 * packet0)
+{
+  if (PREDICT_TRUE (h0->dst_mcast_mask))
+    {
+      /* location to write to in the packet */
+      u8 *p0 = packet0 - h0->dst_mcast_offset;
+      u32 *p1 = (u32 *) p0;
+      /* location to copy from in the L3 dest address */
+      u32 *a0 = (u32 *) (addr + addr_len - sizeof (h0->dst_mcast_mask));
+
+      *p1 |= (*a0 & h0->dst_mcast_mask);
+    }
+}
+
+#define vnet_fixup_one_header(rw0,addr,p0)              \
+  _vnet_fixup_one_header (&((rw0).rewrite_header),                      \
+                          (u8*)(addr), sizeof((*addr)),                 \
+                          (u8*)(p0))
+
+#define VNET_REWRITE_FOR_SW_INTERFACE_ADDRESS_BROADCAST ((void *) 0)
+/** Deprecated */
+void vnet_rewrite_for_sw_interface (struct vnet_main_t *vnm,
+				    vnet_link_t packet_type,
+				    u32 sw_if_index,
+				    u32 node_index,
+				    void *dst_address,
+				    vnet_rewrite_header_t * rw,
+				    u32 max_rewrite_bytes);
+
+u32 vnet_tx_node_index_for_sw_interface (struct vnet_main_t *vnm,
+					 u32 sw_if_index);
+
+void vnet_rewrite_init (struct vnet_main_t *vnm,
+			u32 sw_if_index,
+			u32 this_node,
+			u32 next_node, vnet_rewrite_header_t * rw);
+
+u8 *vnet_build_rewrite_for_sw_interface (struct vnet_main_t *vnm,
+					 u32 sw_if_index,
+					 vnet_link_t packet_type,
+					 const void *dst_address);
+void vnet_update_adjacency_for_sw_interface (struct vnet_main_t *vnm,
+					     u32 sw_if_index, u32 ai);
+
+format_function_t format_vnet_rewrite;
+
+serialize_function_t serialize_vnet_rewrite, unserialize_vnet_rewrite;
+
+#endif /* included_vnet_rewrite_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */