nat: fix counters increment for output feature

Type: fix

The NAT plugin stores packet counters and a counter represents a vector
indexed by interface index. When an interface is assigned a NAT role,
the counters are validated to be long enough for the given interface
index.

When a packet traverses NAT in2out and output feature is disabled, the
appropriate counters are updated by the RX interface index. In this
case, translation happens on the inside interface and its index was
ensured to be valid in all of the counters during NAT role assignment.

When a packet traverses NAT in2out and output feature is enabled, the
appropriate counters are updated by the RX interface index too. In this
case, translation happens on the outside interface and the packet could
be received on any interface, even with no NAT role assigned. If that's
the case and its index is greater than the greatest index validated in
the counters, a new counter value will be written to memory that does
not belong to the counter. As a result, a crash will occur at some
point.

With this change, use TX interface index to update the counters when
output feature is enabled. TX interface is an actual interface where
translation happens and its index is always valid in the counters.

Signed-off-by: Alexander Chernavin <achernavin@netgate.com>
Change-Id: I53a52af949fe96419e1b5fef4134ab4062198f51

3 files changed, 93 insertions, 68 deletions

diff --git a/src/plugins/nat/nat44-ed/nat44_ed_in2out.c b/src/plugins/nat/nat44-ed/nat44_ed_in2out.c
index 0065d7703b1..79c03bd87e5 100644
--- a/src/plugins/nat/nat44-ed/nat44_ed_in2out.c
+++ b/src/plugins/nat/nat44-ed/nat44_ed_in2out.c
@@ -988,7 +988,9 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t *vm,
   while (n_left_from > 0)
     {
       vlib_buffer_t *b0;
-      u32 sw_if_index0, rx_fib_index0, iph_offset0 = 0;
+      u32 rx_sw_if_index0, rx_fib_index0, iph_offset0 = 0;
+      u32 tx_sw_if_index0;
+      u32 cntr_sw_if_index0;
       nat_protocol_t proto0;
       ip4_header_t *ip0;
       snat_session_t *s0 = 0;
@@ -1023,9 +1025,12 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t *vm,
       ip0 =
 	(ip4_header_t *) ((u8 *) vlib_buffer_get_current (b0) + iph_offset0);
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
-      rx_fib_index0 =
-	fib_table_get_index_for_sw_if_index (FIB_PROTOCOL_IP4, sw_if_index0);
+      rx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      tx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+      cntr_sw_if_index0 =
+	is_output_feature ? tx_sw_if_index0 : rx_sw_if_index0;
+      rx_fib_index0 = fib_table_get_index_for_sw_if_index (FIB_PROTOCOL_IP4,
+							   rx_sw_if_index0);
       lookup.fib_index = rx_fib_index0;
 
       if (PREDICT_FALSE (!is_output_feature && ip0->ttl == 1))
@@ -1196,20 +1201,20 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t *vm,
 	{
 	case NAT_PROTOCOL_TCP:
 	  vlib_increment_simple_counter (&sm->counters.fastpath.in2out.tcp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  nat44_set_tcp_session_state_i2o (sm, now, s0, b0, thread_index);
 	  break;
 	case NAT_PROTOCOL_UDP:
 	  vlib_increment_simple_counter (&sm->counters.fastpath.in2out.udp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  break;
 	case NAT_PROTOCOL_ICMP:
 	  vlib_increment_simple_counter (&sm->counters.fastpath.in2out.icmp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  break;
 	case NAT_PROTOCOL_OTHER:
 	  vlib_increment_simple_counter (&sm->counters.fastpath.in2out.other,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  break;
 	}
 
@@ -1227,7 +1232,7 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t *vm,
 	{
 	  nat_in2out_ed_trace_t *t =
 	    vlib_add_trace (vm, node, b0, sizeof (*t));
-	  t->sw_if_index = sw_if_index0;
+	  t->sw_if_index = rx_sw_if_index0;
 	  t->next_index = next[0];
 	  t->is_slow_path = 0;
 	  t->translation_error = translation_error;
@@ -1250,7 +1255,7 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t *vm,
       if (next[0] == NAT_NEXT_DROP)
 	{
 	  vlib_increment_simple_counter (&sm->counters.fastpath.in2out.drops,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
 
       n_left_from--;
@@ -1285,7 +1290,9 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
   while (n_left_from > 0)
     {
       vlib_buffer_t *b0;
-      u32 sw_if_index0, rx_fib_index0, iph_offset0 = 0;
+      u32 rx_sw_if_index0, rx_fib_index0, iph_offset0 = 0;
+      u32 tx_sw_if_index0;
+      u32 cntr_sw_if_index0;
       nat_protocol_t proto0;
       ip4_header_t *ip0;
       udp_header_t *udp0;
@@ -1304,9 +1311,12 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
       ip0 = (ip4_header_t *) ((u8 *) vlib_buffer_get_current (b0) +
 			      iph_offset0);
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
-      rx_fib_index0 =
-	fib_table_get_index_for_sw_if_index (FIB_PROTOCOL_IP4, sw_if_index0);
+      rx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      tx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+      cntr_sw_if_index0 =
+	is_output_feature ? tx_sw_if_index0 : rx_sw_if_index0;
+      rx_fib_index0 = fib_table_get_index_for_sw_if_index (FIB_PROTOCOL_IP4,
+							   rx_sw_if_index0);
 
       if (PREDICT_FALSE (!is_output_feature && ip0->ttl == 1))
 	{
@@ -1342,14 +1352,14 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
 	    }
 
 	  vlib_increment_simple_counter (&sm->counters.slowpath.in2out.other,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  goto trace0;
 	}
 
       if (PREDICT_FALSE (proto0 == NAT_PROTOCOL_ICMP))
 	{
 	  next[0] = icmp_in2out_ed_slow_path (
-	    sm, b0, ip0, icmp0, sw_if_index0, rx_fib_index0, node, next[0],
+	    sm, b0, ip0, icmp0, rx_sw_if_index0, rx_fib_index0, node, next[0],
 	    now, thread_index, proto0, &s0, is_multi_worker);
 	  if (NAT_NEXT_DROP != next[0] && s0 &&
 	      NAT_ED_TRNSL_ERR_SUCCESS !=
@@ -1364,7 +1374,7 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
 	    }
 
 	  vlib_increment_simple_counter (&sm->counters.slowpath.in2out.icmp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  goto trace0;
 	}
 
@@ -1394,8 +1404,7 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
 	      if (PREDICT_FALSE (nat44_ed_not_translate_output_feature (
 		    sm, b0, ip0, vnet_buffer (b0)->ip.reass.l4_src_port,
 		    vnet_buffer (b0)->ip.reass.l4_dst_port, thread_index,
-		    sw_if_index0, vnet_buffer (b0)->sw_if_index[VLIB_TX], now,
-		    is_multi_worker)))
+		    rx_sw_if_index0, tx_sw_if_index0, now, is_multi_worker)))
 		goto trace0;
 
 	      /*
@@ -1412,8 +1421,8 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
 	  else
 	    {
 	      if (PREDICT_FALSE (nat44_ed_not_translate (
-		    vm, sm, node, sw_if_index0, b0, ip0, proto0, rx_fib_index0,
-		    thread_index)))
+		    vm, sm, node, rx_sw_if_index0, b0, ip0, proto0,
+		    rx_fib_index0, thread_index)))
 		goto trace0;
 	    }
 
@@ -1447,13 +1456,13 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
       if (PREDICT_TRUE (proto0 == NAT_PROTOCOL_TCP))
 	{
 	  vlib_increment_simple_counter (&sm->counters.slowpath.in2out.tcp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	  nat44_set_tcp_session_state_i2o (sm, now, s0, b0, thread_index);
 	}
       else
 	{
 	  vlib_increment_simple_counter (&sm->counters.slowpath.in2out.udp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
 
       /* Accounting */
@@ -1469,7 +1478,7 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
 	{
 	  nat_in2out_ed_trace_t *t =
 	    vlib_add_trace (vm, node, b0, sizeof (*t));
-	  t->sw_if_index = sw_if_index0;
+	  t->sw_if_index = rx_sw_if_index0;
 	  t->next_index = next[0];
 	  t->is_slow_path = 1;
 	  t->translation_error = translation_error;
@@ -1492,7 +1501,7 @@ nat44_ed_in2out_slow_path_node_fn_inline (vlib_main_t *vm,
       if (next[0] == NAT_NEXT_DROP)
 	{
 	  vlib_increment_simple_counter (&sm->counters.slowpath.in2out.drops,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
 
       n_left_from--;
diff --git a/src/plugins/nat/nat44-ei/nat44_ei_hairpinning.c b/src/plugins/nat/nat44-ei/nat44_ei_hairpinning.c
index a049e4659a7..c3d3cfba386 100644
--- a/src/plugins/nat/nat44-ei/nat44_ei_hairpinning.c
+++ b/src/plugins/nat/nat44-ei/nat44_ei_hairpinning.c
@@ -428,7 +428,8 @@ VLIB_NODE_FN (nat44_ei_hairpin_src_node)
 	  vlib_buffer_t *b0;
 	  u32 next0;
 	  nat44_ei_interface_t *i;
-	  u32 sw_if_index0;
+	  u32 rx_sw_if_index0;
+	  u32 tx_sw_if_index0;
 
 	  /* speculatively enqueue b0 to the current next frame */
 	  bi0 = from[0];
@@ -439,13 +440,14 @@ VLIB_NODE_FN (nat44_ei_hairpin_src_node)
 	  n_left_to_next -= 1;
 
 	  b0 = vlib_get_buffer (vm, bi0);
-	  sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+	  rx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+	  tx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
 
 	  pool_foreach (i, nm->output_feature_interfaces)
 	    {
 	      /* Only packets from NAT inside interface */
 	      if ((nat44_ei_interface_is_inside (i)) &&
-		  (sw_if_index0 == i->sw_if_index))
+		  (rx_sw_if_index0 == i->sw_if_index))
 		{
 		  if (PREDICT_FALSE ((vnet_buffer (b0)->snat.flags) &
 				     NAT44_EI_FLAG_HAIRPINNING))
@@ -470,8 +472,9 @@ VLIB_NODE_FN (nat44_ei_hairpin_src_node)
 
 	  if (next0 != NAT44_EI_HAIRPIN_SRC_NEXT_DROP)
 	    {
-	      vlib_increment_simple_counter (
-		&nm->counters.hairpinning, vm->thread_index, sw_if_index0, 1);
+	      vlib_increment_simple_counter (&nm->counters.hairpinning,
+					     vm->thread_index, tx_sw_if_index0,
+					     1);
 	    }
 
 	  /* verify speculative enqueue, maybe switch current next frame */
diff --git a/src/plugins/nat/nat44-ei/nat44_ei_in2out.c b/src/plugins/nat/nat44-ei/nat44_ei_in2out.c
index 7ac1a92a61b..1e28ed6b635 100644
--- a/src/plugins/nat/nat44-ei/nat44_ei_in2out.c
+++ b/src/plugins/nat/nat44-ei/nat44_ei_in2out.c
@@ -934,7 +934,9 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
     {
       vlib_buffer_t *b0, *b1;
       u32 next0, next1;
-      u32 sw_if_index0, sw_if_index1;
+      u32 rx_sw_if_index0, rx_sw_if_index1;
+      u32 tx_sw_if_index0, tx_sw_if_index1;
+      u32 cntr_sw_if_index0, cntr_sw_if_index1;
       ip4_header_t *ip0, *ip1;
       ip_csum_t sum0, sum1;
       u32 new_addr0, old_addr0, new_addr1, old_addr1;
@@ -978,9 +980,12 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
       tcp0 = (tcp_header_t *) udp0;
       icmp0 = (icmp46_header_t *) udp0;
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      rx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      tx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+      cntr_sw_if_index0 =
+	is_output_feature ? tx_sw_if_index0 : rx_sw_if_index0;
       rx_fib_index0 =
-	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, sw_if_index0);
+	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, rx_sw_if_index0);
 
       next0 = next1 = NAT44_EI_IN2OUT_NEXT_LOOKUP;
 
@@ -1010,19 +1015,19 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.other :
 			       &nm->counters.fastpath.in2out.other,
-		thread_index, sw_if_index0, 1);
+		thread_index, cntr_sw_if_index0, 1);
 	      goto trace00;
 	    }
 
 	  if (PREDICT_FALSE (proto0 == NAT_PROTOCOL_ICMP))
 	    {
 	      next0 = nat44_ei_icmp_in2out_slow_path (
-		nm, b0, ip0, icmp0, sw_if_index0, rx_fib_index0, node, next0,
-		now, thread_index, &s0);
+		nm, b0, ip0, icmp0, rx_sw_if_index0, rx_fib_index0, node,
+		next0, now, thread_index, &s0);
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.icmp :
 			       &nm->counters.fastpath.in2out.icmp,
-		thread_index, sw_if_index0, 1);
+		thread_index, cntr_sw_if_index0, 1);
 	      goto trace00;
 	    }
 	}
@@ -1055,7 +1060,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 			nm, ip0, proto0,
 			vnet_buffer (b0)->ip.reass.l4_src_port,
 			vnet_buffer (b0)->ip.reass.l4_dst_port, thread_index,
-			sw_if_index0)))
+			rx_sw_if_index0)))
 		    goto trace00;
 
 		  /*
@@ -1073,7 +1078,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      else
 		{
 		  if (PREDICT_FALSE (nat44_ei_not_translate (
-			nm, node, sw_if_index0, ip0, proto0, rx_fib_index0,
+			nm, node, rx_sw_if_index0, ip0, proto0, rx_fib_index0,
 			thread_index)))
 		    goto trace00;
 		}
@@ -1131,7 +1136,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.tcp :
 					   &nm->counters.fastpath.in2out.tcp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
       else
 	{
@@ -1155,7 +1160,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.udp :
 					   &nm->counters.fastpath.in2out.udp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
 
       /* Accounting */
@@ -1171,7 +1176,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  nat44_ei_in2out_trace_t *t =
 	    vlib_add_trace (vm, node, b0, sizeof (*t));
 	  t->is_slow_path = is_slow_path;
-	  t->sw_if_index = sw_if_index0;
+	  t->sw_if_index = rx_sw_if_index0;
 	  t->next_index = next0;
 	  t->session_index = ~0;
 	  if (s0)
@@ -1183,7 +1188,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (
 	    is_slow_path ? &nm->counters.slowpath.in2out.drops :
 			   &nm->counters.fastpath.in2out.drops,
-	    thread_index, sw_if_index0, 1);
+	    thread_index, cntr_sw_if_index0, 1);
 	}
 
       if (is_output_feature)
@@ -1196,9 +1201,12 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
       tcp1 = (tcp_header_t *) udp1;
       icmp1 = (icmp46_header_t *) udp1;
 
-      sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_RX];
+      rx_sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_RX];
+      tx_sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_TX];
+      cntr_sw_if_index1 =
+	is_output_feature ? tx_sw_if_index1 : rx_sw_if_index1;
       rx_fib_index1 =
-	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, sw_if_index1);
+	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, rx_sw_if_index1);
 
       if (PREDICT_FALSE (ip1->ttl == 1))
 	{
@@ -1226,19 +1234,19 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.other :
 			       &nm->counters.fastpath.in2out.other,
-		thread_index, sw_if_index1, 1);
+		thread_index, cntr_sw_if_index1, 1);
 	      goto trace01;
 	    }
 
 	  if (PREDICT_FALSE (proto1 == NAT_PROTOCOL_ICMP))
 	    {
 	      next1 = nat44_ei_icmp_in2out_slow_path (
-		nm, b1, ip1, icmp1, sw_if_index1, rx_fib_index1, node, next1,
-		now, thread_index, &s1);
+		nm, b1, ip1, icmp1, rx_sw_if_index1, rx_fib_index1, node,
+		next1, now, thread_index, &s1);
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.icmp :
 			       &nm->counters.fastpath.in2out.icmp,
-		thread_index, sw_if_index1, 1);
+		thread_index, cntr_sw_if_index1, 1);
 	      goto trace01;
 	    }
 	}
@@ -1271,7 +1279,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 			nm, ip1, proto1,
 			vnet_buffer (b1)->ip.reass.l4_src_port,
 			vnet_buffer (b1)->ip.reass.l4_dst_port, thread_index,
-			sw_if_index1)))
+			rx_sw_if_index1)))
 		    goto trace01;
 
 		  /*
@@ -1289,7 +1297,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      else
 		{
 		  if (PREDICT_FALSE (nat44_ei_not_translate (
-			nm, node, sw_if_index1, ip1, proto1, rx_fib_index1,
+			nm, node, rx_sw_if_index1, ip1, proto1, rx_fib_index1,
 			thread_index)))
 		    goto trace01;
 		}
@@ -1346,7 +1354,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.tcp :
 					   &nm->counters.fastpath.in2out.tcp,
-					 thread_index, sw_if_index1, 1);
+					 thread_index, cntr_sw_if_index1, 1);
 	}
       else
 	{
@@ -1370,7 +1378,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.udp :
 					   &nm->counters.fastpath.in2out.udp,
-					 thread_index, sw_if_index1, 1);
+					 thread_index, cntr_sw_if_index1, 1);
 	}
 
       /* Accounting */
@@ -1385,7 +1393,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	{
 	  nat44_ei_in2out_trace_t *t =
 	    vlib_add_trace (vm, node, b1, sizeof (*t));
-	  t->sw_if_index = sw_if_index1;
+	  t->sw_if_index = rx_sw_if_index1;
 	  t->next_index = next1;
 	  t->session_index = ~0;
 	  if (s1)
@@ -1397,7 +1405,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (
 	    is_slow_path ? &nm->counters.slowpath.in2out.drops :
 			   &nm->counters.fastpath.in2out.drops,
-	    thread_index, sw_if_index1, 1);
+	    thread_index, cntr_sw_if_index1, 1);
 	}
 
       n_left_from -= 2;
@@ -1410,7 +1418,9 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
     {
       vlib_buffer_t *b0;
       u32 next0;
-      u32 sw_if_index0;
+      u32 rx_sw_if_index0;
+      u32 tx_sw_if_index0;
+      u32 cntr_sw_if_index0;
       ip4_header_t *ip0;
       ip_csum_t sum0;
       u32 new_addr0, old_addr0;
@@ -1438,9 +1448,12 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
       tcp0 = (tcp_header_t *) udp0;
       icmp0 = (icmp46_header_t *) udp0;
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      rx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      tx_sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+      cntr_sw_if_index0 =
+	is_output_feature ? tx_sw_if_index0 : rx_sw_if_index0;
       rx_fib_index0 =
-	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, sw_if_index0);
+	vec_elt (nm->ip4_main->fib_index_by_sw_if_index, rx_sw_if_index0);
 
       if (PREDICT_FALSE (ip0->ttl == 1))
 	{
@@ -1468,19 +1481,19 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.other :
 			       &nm->counters.fastpath.in2out.other,
-		thread_index, sw_if_index0, 1);
+		thread_index, cntr_sw_if_index0, 1);
 	      goto trace0;
 	    }
 
 	  if (PREDICT_FALSE (proto0 == NAT_PROTOCOL_ICMP))
 	    {
 	      next0 = nat44_ei_icmp_in2out_slow_path (
-		nm, b0, ip0, icmp0, sw_if_index0, rx_fib_index0, node, next0,
-		now, thread_index, &s0);
+		nm, b0, ip0, icmp0, rx_sw_if_index0, rx_fib_index0, node,
+		next0, now, thread_index, &s0);
 	      vlib_increment_simple_counter (
 		is_slow_path ? &nm->counters.slowpath.in2out.icmp :
 			       &nm->counters.fastpath.in2out.icmp,
-		thread_index, sw_if_index0, 1);
+		thread_index, cntr_sw_if_index0, 1);
 	      goto trace0;
 	    }
 	}
@@ -1513,7 +1526,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 			nm, ip0, proto0,
 			vnet_buffer (b0)->ip.reass.l4_src_port,
 			vnet_buffer (b0)->ip.reass.l4_dst_port, thread_index,
-			sw_if_index0)))
+			rx_sw_if_index0)))
 		    goto trace0;
 
 		  /*
@@ -1531,7 +1544,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	      else
 		{
 		  if (PREDICT_FALSE (nat44_ei_not_translate (
-			nm, node, sw_if_index0, ip0, proto0, rx_fib_index0,
+			nm, node, rx_sw_if_index0, ip0, proto0, rx_fib_index0,
 			thread_index)))
 		    goto trace0;
 		}
@@ -1590,7 +1603,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.tcp :
 					   &nm->counters.fastpath.in2out.tcp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
       else
 	{
@@ -1615,7 +1628,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (is_slow_path ?
 					   &nm->counters.slowpath.in2out.udp :
 					   &nm->counters.fastpath.in2out.udp,
-					 thread_index, sw_if_index0, 1);
+					 thread_index, cntr_sw_if_index0, 1);
 	}
 
       /* Accounting */
@@ -1631,7 +1644,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  nat44_ei_in2out_trace_t *t =
 	    vlib_add_trace (vm, node, b0, sizeof (*t));
 	  t->is_slow_path = is_slow_path;
-	  t->sw_if_index = sw_if_index0;
+	  t->sw_if_index = rx_sw_if_index0;
 	  t->next_index = next0;
 	  t->session_index = ~0;
 	  if (s0)
@@ -1643,7 +1656,7 @@ nat44_ei_in2out_node_fn_inline (vlib_main_t *vm, vlib_node_runtime_t *node,
 	  vlib_increment_simple_counter (
 	    is_slow_path ? &nm->counters.slowpath.in2out.drops :
 			   &nm->counters.fastpath.in2out.drops,
-	    thread_index, sw_if_index0, 1);
+	    thread_index, cntr_sw_if_index0, 1);
 	}
 
       n_left_from--;