vlib: add buffer cloning support

Change-Id: I50070611af15b2b4cc29664a8bee4f821ac3c835
Signed-off-by: Damjan Marion <damarion@cisco.com>

4 files changed, 74 insertions, 57 deletions

diff --git a/src/vnet/devices/dpdk/buffer.c b/src/vnet/devices/dpdk/buffer.c
index 007093e493a..f95d4cb5c38 100644
--- a/src/vnet/devices/dpdk/buffer.c
+++ b/src/vnet/devices/dpdk/buffer.c
@@ -79,20 +79,46 @@
 STATIC_ASSERT (VLIB_BUFFER_PRE_DATA_SIZE == RTE_PKTMBUF_HEADROOM,
 	       "VLIB_BUFFER_PRE_DATA_SIZE must be equal to RTE_PKTMBUF_HEADROOM");
 
+static_always_inline void
+dpdk_rte_pktmbuf_free (vlib_main_t * vm, vlib_buffer_t * b)
+{
+  vlib_buffer_t *hb = b;
+  struct rte_mbuf *mb;
+  u32 next, flags;
+  mb = rte_mbuf_from_vlib_buffer (hb);
+
+next:
+  flags = b->flags;
+  next = b->next_buffer;
+  mb = rte_mbuf_from_vlib_buffer (b);
+
+  if (PREDICT_FALSE (b->n_add_refs))
+    {
+      rte_mbuf_refcnt_update (mb, b->n_add_refs);
+      b->n_add_refs = 0;
+    }
+
+  rte_pktmbuf_free_seg (mb);
+
+  if (flags & VLIB_BUFFER_NEXT_PRESENT)
+    {
+      b = vlib_get_buffer (vm, next);
+      goto next;
+    }
+}
+
 static void
 del_free_list (vlib_main_t * vm, vlib_buffer_free_list_t * f)
 {
   u32 i;
-  struct rte_mbuf *mb;
   vlib_buffer_t *b;
 
   for (i = 0; i < vec_len (f->buffers); i++)
     {
       b = vlib_get_buffer (vm, f->buffers[i]);
-      mb = rte_mbuf_from_vlib_buffer (b);
-      ASSERT (rte_mbuf_refcnt_read (mb) == 1);
-      rte_pktmbuf_free (mb);
+      dpdk_rte_pktmbuf_free (vm, b);
     }
+
   vec_free (f->name);
   vec_free (f->buffers);
 }
@@ -325,7 +351,6 @@ vlib_buffer_free_inline (vlib_main_t * vm,
   for (i = 0; i < n_buffers; i++)
     {
       vlib_buffer_t *b;
-      struct rte_mbuf *mb;
 
       b = vlib_get_buffer (vm, buffers[i]);
 
@@ -351,11 +376,7 @@ vlib_buffer_free_inline (vlib_main_t * vm,
       else
 	{
 	  if (PREDICT_TRUE ((b->flags & VLIB_BUFFER_RECYCLE) == 0))
-	    {
-	      mb = rte_mbuf_from_vlib_buffer (b);
-	      ASSERT (rte_mbuf_refcnt_read (mb) == 1);
-	      rte_pktmbuf_free (mb);
-	    }
+	    dpdk_rte_pktmbuf_free (vm, b);
 	}
     }
   if (vec_len (bm->announce_list))
diff --git a/src/vnet/devices/dpdk/device.c b/src/vnet/devices/dpdk/device.c
index c9d9a567b7a..17397900c59 100644
--- a/src/vnet/devices/dpdk/device.c
+++ b/src/vnet/devices/dpdk/device.c
@@ -168,13 +168,11 @@ dpdk_validate_rte_mbuf (vlib_main_t * vm, vlib_buffer_t * b,
 	{
 	  b2 = vlib_get_buffer (vm, b2->next_buffer);
 	  mb = rte_mbuf_from_vlib_buffer (b2);
-	  last_mb->next = mb;
-	  last_mb = mb;
 	  rte_pktmbuf_reset (mb);
 	}
     }
 
-  first_mb = mb = rte_mbuf_from_vlib_buffer (b);
+  last_mb = first_mb = mb = rte_mbuf_from_vlib_buffer (b);
   first_mb->nb_segs = 1;
   mb->data_len = b->current_length;
   mb->pkt_len = maybe_multiseg ? vlib_buffer_length_in_chain (vm, b) :
@@ -185,10 +183,17 @@ dpdk_validate_rte_mbuf (vlib_main_t * vm, vlib_buffer_t * b,
     {
       b = vlib_get_buffer (vm, b->next_buffer);
       mb = rte_mbuf_from_vlib_buffer (b);
+      last_mb->next = mb;
+      last_mb = mb;
       mb->data_len = b->current_length;
       mb->pkt_len = b->current_length;
       mb->data_off = VLIB_BUFFER_PRE_DATA_SIZE + b->current_data;
       first_mb->nb_segs++;
+      if (PREDICT_FALSE (b->n_add_refs))
+	{
+	  rte_mbuf_refcnt_update (mb, b->n_add_refs);
+	  b->n_add_refs = 0;
+	}
     }
 }
 
diff --git a/src/vnet/dpo/replicate_dpo.c b/src/vnet/dpo/replicate_dpo.c
index a67b19c893f..a9f334be7cf 100644
--- a/src/vnet/dpo/replicate_dpo.c
+++ b/src/vnet/dpo/replicate_dpo.c
@@ -625,6 +625,7 @@ replicate_inline (vlib_main_t * vm,
                   vlib_frame_t * frame)
 {
     vlib_combined_counter_main_t * cm = &replicate_main.repm_counters;
+    replicate_main_t * rm = &replicate_main;
     u32 n_left_from, * from, * to_next, next_index;
     u32 cpu_index = os_get_cpu_number();
 
@@ -645,13 +646,11 @@ replicate_inline (vlib_main_t * vm,
             const replicate_t *rep0;
             vlib_buffer_t * b0, *c0;
             const dpo_id_t *dpo0;
+	    u8 num_cloned;
 
             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];
@@ -661,50 +660,21 @@ replicate_inline (vlib_main_t * vm,
                 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;
+	    vec_validate (rm->clones[cpu_index], rep0->rep_n_buckets - 1);
 
-            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 = *dpo0;
-            }
-            vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
-                                             to_next, n_left_to_next,
-                                             bi0, next0);
+	    num_cloned = vlib_buffer_clone (vm, bi0, rm->clones[cpu_index], rep0->rep_n_buckets, 128);
 
-            /* ship copies to the rest of the buckets */
-            for (bucket = 1; bucket < rep0->rep_n_buckets; bucket++)
-            {
-                /*
-                 * After the enqueue of the first buffer, and of all subsequent
-                 * buffers in this loop, it is possible that we over-flow the
-                 * frame of the to-next node. When this happens we need to 'put'
-                 * that full frame to the node and get a fresh empty one.
-                 * Note that these are macros with side effects that change
-                 * to_next & n_left_to_next
-                 */
-                if (PREDICT_FALSE(0 == n_left_to_next))
-                {
-                    vlib_put_next_frame (vm, node, next_index, n_left_to_next);
-                    vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
-                }
+	    if (num_cloned != rep0->rep_n_buckets)
+	      {
+		vlib_node_increment_counter
+		  (vm, node->node_index,
+		   REPLICATE_DPO_ERROR_BUFFER_ALLOCATION_FAILURE, 1);
+	      }
 
-                /* Make a copy. This can fail, so deal with it. */
-                c0 = vlib_buffer_copy(vm, b0);
-                if (PREDICT_FALSE (c0 == 0))
-                  {
-                    vlib_node_increment_counter 
-                      (vm, node->node_index, 
-                       REPLICATE_DPO_ERROR_BUFFER_ALLOCATION_FAILURE,
-                       1);
-                    continue;
-                  }
-                
-                ci0 = vlib_get_buffer_index(vm, c0);
+            for (bucket = 0; bucket < num_cloned; bucket++)
+            {
+                ci0 = rm->clones[cpu_index][bucket];
+                c0 = vlib_get_buffer(vm, ci0);
 
                 to_next[0] = ci0;
                 to_next += 1;
@@ -724,7 +694,13 @@ replicate_inline (vlib_main_t * vm,
                 vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
                                                  to_next, n_left_to_next,
                                                  ci0, next0);
+		if (PREDICT_FALSE (n_left_to_next == 0))
+		  {
+		    vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+		    vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+		  }
             }
+	    vec_reset_length (rm->clones[cpu_index]);
         }
 
         vlib_put_next_frame (vm, node, next_index, n_left_to_next);
@@ -797,3 +773,15 @@ VLIB_REGISTER_NODE (ip6_replicate_node) = {
       [0] = "error-drop",
   },
 };
+
+clib_error_t *
+replicate_dpo_init (vlib_main_t * vm)
+{
+  replicate_main_t * rm = &replicate_main;
+
+  vec_validate (rm->clones, vlib_num_workers());
+
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (replicate_dpo_init);
diff --git a/src/vnet/dpo/replicate_dpo.h b/src/vnet/dpo/replicate_dpo.h
index a564739c9f2..77273015c9e 100644
--- a/src/vnet/dpo/replicate_dpo.h
+++ b/src/vnet/dpo/replicate_dpo.h
@@ -32,6 +32,9 @@
 typedef struct replicate_main_t_
 {
     vlib_combined_counter_main_t repm_counters;
+
+    /* per-cpu vector of cloned packets */
+    u32 **clones;
 } replicate_main_t;
 
 extern replicate_main_t replicate_main;