classifier-based ACL: refactor + add output ACL

For implementation of MACIP ACLs enhancement (VPP-1088), an outbound
classifier-based ACL would be needed. There was an existing incomplete
code for outbound ACLs, it looked almost exact copy of input ACLs, minus
the various enhancements, trying to sync that code seemed error-prone
and cumbersome to maintain in the longer run.

This change refactors the input+output ACLs processing into a unified
routine (thus any changes will have effect on both), and also adds
the API to set the output interface ACL, with the same format
and semantics as the existing input one (except working on output
ACL of course).

WARNING: IP outbound ACL in L3 mode clobbers the ip.* fields
in the vnet_buffer_opaque_t, since the code is using l2_classify.*
The net_buffer (p0)->ip.save_rewrite_length is rescued into
l2_classify.pad.l2_len, and used to rewind the header in case of
drop, so that ipX_drop prints something sensible.

Change-Id: I62f814f1e3650e504474a3a5359edb8a0a8836ed
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>

6 files changed, 192 insertions, 36 deletions

diff --git a/src/vnet/ip/ip4_error.h b/src/vnet/ip/ip4_error.h
index 00b1c6daad5..fa93e8673e5 100644
--- a/src/vnet/ip/ip4_error.h
+++ b/src/vnet/ip/ip4_error.h
@@ -76,6 +76,9 @@
   /* Errors singalled by ip4-inacl */                                   \
   _ (INACL_TABLE_MISS, "input ACL table-miss drops")                    \
   _ (INACL_SESSION_DENY, "input ACL session deny drops")                \
+  /* Errors singalled by ip4-outacl */                                  \
+  _ (OUTACL_TABLE_MISS, "output ACL table-miss drops")                  \
+  _ (OUTACL_SESSION_DENY, "output ACL session deny drops")              \
                                                                         \
   /* Erros from mfib-forward */                                         \
   _ (RPF_FAILURE, "Multicast RPF check failed")                         \
diff --git a/src/vnet/ip/ip4_forward.c b/src/vnet/ip/ip4_forward.c
index 2a21135bb3e..a0f3a4bfae0 100755
--- a/src/vnet/ip/ip4_forward.c
+++ b/src/vnet/ip/ip4_forward.c
@@ -1121,6 +1121,13 @@ VNET_FEATURE_INIT (ip4_source_and_port_range_check_tx, static) =
 {
   .arc_name = "ip4-output",
   .node_name = "ip4-source-and-port-range-check-tx",
+  .runs_before = VNET_FEATURES ("ip4-outacl"),
+};
+
+VNET_FEATURE_INIT (ip4_outacl, static) =
+{
+  .arc_name = "ip4-output",
+  .node_name = "ip4-outacl",
   .runs_before = VNET_FEATURES ("ipsec-output-ip4"),
 };
 
diff --git a/src/vnet/ip/ip6_error.h b/src/vnet/ip/ip6_error.h
index 19f20625a10..cc5afb93102 100644
--- a/src/vnet/ip/ip6_error.h
+++ b/src/vnet/ip/ip6_error.h
@@ -72,6 +72,9 @@
  /* Erros singalled by ip6-inacl */                                     \
   _ (INACL_TABLE_MISS, "input ACL table-miss drops")                    \
   _ (INACL_SESSION_DENY, "input ACL session deny drops")                \
+ /* Erros singalled by ip6-outacl */                                    \
+  _ (OUTACL_TABLE_MISS, "output ACL table-miss drops")                  \
+  _ (OUTACL_SESSION_DENY, "output ACL session deny drops")              \
                                                                         \
   /* Errors signalled by ip6-reassembly */                              \
   _ (REASS_MISSING_UPPER, "missing-upper layer drops")                  \
diff --git a/src/vnet/ip/ip6_forward.c b/src/vnet/ip/ip6_forward.c
index 5b06ca7609d..fc71c3d2d97 100644
--- a/src/vnet/ip/ip6_forward.c
+++ b/src/vnet/ip/ip6_forward.c
@@ -670,6 +670,12 @@ VNET_FEATURE_ARC_INIT (ip6_output, static) =
   .arc_index_ptr = &ip6_main.lookup_main.output_feature_arc_index,
 };
 
+VNET_FEATURE_INIT (ip6_outacl, static) = {
+  .arc_name = "ip6-output",
+  .node_name = "ip6-outacl",
+  .runs_before = VNET_FEATURES ("ipsec-output-ip6"),
+};
+
 VNET_FEATURE_INIT (ip6_ipsec_output, static) = {
   .arc_name = "ip6-output",
   .node_name = "ipsec-output-ip6",
diff --git a/src/vnet/ip/ip_input_acl.c b/src/vnet/ip/ip_in_out_acl.c
index 3c54fe8be2c..b50f5b1ff67 100644
--- a/src/vnet/ip/ip_input_acl.c
+++ b/src/vnet/ip/ip_in_out_acl.c
@@ -14,7 +14,7 @@
  */
 #include <vnet/ip/ip.h>
 #include <vnet/classify/vnet_classify.h>
-#include <vnet/classify/input_acl.h>
+#include <vnet/classify/in_out_acl.h>
 
 typedef struct
 {
@@ -22,29 +22,49 @@ typedef struct
   u32 next_index;
   u32 table_index;
   u32 offset;
-} ip_inacl_trace_t;
+} ip_in_out_acl_trace_t;
 
 /* packet trace format function */
 static u8 *
-format_ip_inacl_trace (u8 * s, va_list * args)
+format_ip_in_out_acl_trace (u8 * s, u32 is_output, 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 *);
-  ip_inacl_trace_t *t = va_arg (*args, ip_inacl_trace_t *);
+  ip_in_out_acl_trace_t *t = va_arg (*args, ip_in_out_acl_trace_t *);
 
-  s = format (s, "INACL: sw_if_index %d, next_index %d, table %d, offset %d",
+  s = format (s, "%s: sw_if_index %d, next_index %d, table %d, offset %d",
+	      is_output ? "OUTACL" : "INACL",
 	      t->sw_if_index, t->next_index, t->table_index, t->offset);
   return s;
 }
 
+static u8 *
+format_ip_inacl_trace (u8 * s, va_list * args)
+{
+  return format_ip_in_out_acl_trace (s, 0 /* is_output */ , args);
+}
+
+static u8 *
+format_ip_outacl_trace (u8 * s, va_list * args)
+{
+  return format_ip_in_out_acl_trace (s, 1 /* is_output */ , args);
+}
+
 vlib_node_registration_t ip4_inacl_node;
+vlib_node_registration_t ip4_outacl_node;
 vlib_node_registration_t ip6_inacl_node;
+vlib_node_registration_t ip6_outacl_node;
 
 #define foreach_ip_inacl_error                  \
 _(MISS, "input ACL misses")                     \
 _(HIT, "input ACL hits")                        \
 _(CHAIN_HIT, "input ACL hits after chain walk")
 
+#define foreach_ip_outacl_error                  \
+_(MISS, "output ACL misses")                     \
+_(HIT, "output ACL hits")                        \
+_(CHAIN_HIT, "output ACL hits after chain walk")
+
 typedef enum
 {
 #define _(sym,str) IP_INACL_ERROR_##sym,
@@ -59,19 +79,34 @@ static char *ip_inacl_error_strings[] = {
 #undef _
 };
 
+typedef enum
+{
+#define _(sym,str) IP_OUTACL_ERROR_##sym,
+  foreach_ip_outacl_error
+#undef _
+    IP_OUTACL_N_ERROR,
+} ip_outacl_error_t;
+
+static char *ip_outacl_error_strings[] = {
+#define _(sym,string) string,
+  foreach_ip_outacl_error
+#undef _
+};
+
 static inline uword
-ip_inacl_inline (vlib_main_t * vm,
-		 vlib_node_runtime_t * node, vlib_frame_t * frame, int is_ip4)
+ip_in_out_acl_inline (vlib_main_t * vm,
+		      vlib_node_runtime_t * node, vlib_frame_t * frame,
+		      int is_ip4, int is_output)
 {
   u32 n_left_from, *from, *to_next;
   acl_next_index_t next_index;
-  input_acl_main_t *am = &input_acl_main;
+  in_out_acl_main_t *am = &in_out_acl_main;
   vnet_classify_main_t *vcm = am->vnet_classify_main;
   f64 now = vlib_time_now (vm);
   u32 hits = 0;
   u32 misses = 0;
   u32 chain_hits = 0;
-  input_acl_table_id_t tid;
+  in_out_acl_table_id_t tid;
   vlib_node_runtime_t *error_node;
   u32 n_next_nodes;
 
@@ -79,12 +114,12 @@ ip_inacl_inline (vlib_main_t * vm,
 
   if (is_ip4)
     {
-      tid = INPUT_ACL_TABLE_IP4;
+      tid = IN_OUT_ACL_TABLE_IP4;
       error_node = vlib_node_get_runtime (vm, ip4_input_node.index);
     }
   else
     {
-      tid = INPUT_ACL_TABLE_IP6;
+      tid = IN_OUT_ACL_TABLE_IP6;
       error_node = vlib_node_get_runtime (vm, ip6_input_node.index);
     }
 
@@ -121,13 +156,15 @@ ip_inacl_inline (vlib_main_t * vm,
       bi1 = from[1];
       b1 = vlib_get_buffer (vm, bi1);
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      sw_if_index0 =
+	vnet_buffer (b0)->sw_if_index[is_output ? VLIB_TX : VLIB_RX];
       table_index0 =
-	am->classify_table_index_by_sw_if_index[tid][sw_if_index0];
+	am->classify_table_index_by_sw_if_index[is_output][tid][sw_if_index0];
 
-      sw_if_index1 = vnet_buffer (b1)->sw_if_index[VLIB_RX];
+      sw_if_index1 =
+	vnet_buffer (b1)->sw_if_index[is_output ? VLIB_TX : VLIB_RX];
       table_index1 =
-	am->classify_table_index_by_sw_if_index[tid][sw_if_index1];
+	am->classify_table_index_by_sw_if_index[is_output][tid][sw_if_index1];
 
       t0 = pool_elt_at_index (vcm->tables, table_index0);
 
@@ -138,6 +175,15 @@ ip_inacl_inline (vlib_main_t * vm,
       else
 	h0 = b0->data;
 
+      if (is_output)
+	{
+	  /* Save the rewrite length, since we are using the l2_classify struct */
+	  vnet_buffer (b0)->l2_classify.pad.l2_len =
+	    vnet_buffer (b0)->ip.save_rewrite_length;
+	  /* advance the match pointer so the matching happens on IP header */
+	  h0 += vnet_buffer (b0)->l2_classify.pad.l2_len;
+	}
+
       vnet_buffer (b0)->l2_classify.hash =
 	vnet_classify_hash_packet (t0, (u8 *) h0);
 
@@ -148,6 +194,15 @@ ip_inacl_inline (vlib_main_t * vm,
       else
 	h1 = b1->data;
 
+      if (is_output)
+	{
+	  /* Save the rewrite length, since we are using the l2_classify struct */
+	  vnet_buffer (b1)->l2_classify.pad.l2_len =
+	    vnet_buffer (b1)->ip.save_rewrite_length;
+	  /* advance the match pointer so the matching happens on IP header */
+	  h1 += vnet_buffer (b1)->l2_classify.pad.l2_len;
+	}
+
       vnet_buffer (b1)->l2_classify.hash =
 	vnet_classify_hash_packet (t1, (u8 *) h1);
 
@@ -173,9 +228,10 @@ ip_inacl_inline (vlib_main_t * vm,
       bi0 = from[0];
       b0 = vlib_get_buffer (vm, bi0);
 
-      sw_if_index0 = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+      sw_if_index0 =
+	vnet_buffer (b0)->sw_if_index[is_output ? VLIB_TX : VLIB_RX];
       table_index0 =
-	am->classify_table_index_by_sw_if_index[tid][sw_if_index0];
+	am->classify_table_index_by_sw_if_index[is_output][tid][sw_if_index0];
 
       t0 = pool_elt_at_index (vcm->tables, table_index0);
 
@@ -184,6 +240,15 @@ ip_inacl_inline (vlib_main_t * vm,
       else
 	h0 = b0->data;
 
+      if (is_output)
+	{
+	  /* Save the rewrite length, since we are using the l2_classify struct */
+	  vnet_buffer (b0)->l2_classify.pad.l2_len =
+	    vnet_buffer (b0)->ip.save_rewrite_length;
+	  /* advance the match pointer so the matching happens on IP header */
+	  h0 += vnet_buffer (b0)->l2_classify.pad.l2_len;
+	}
+
       vnet_buffer (b0)->l2_classify.hash =
 	vnet_classify_hash_packet (t0, (u8 *) h0);
 
@@ -235,6 +300,7 @@ ip_inacl_inline (vlib_main_t * vm,
 		}
 	    }
 
+
 	  /* speculatively enqueue b0 to the current next frame */
 	  bi0 = from[0];
 	  to_next[0] = bi0;
@@ -247,7 +313,7 @@ ip_inacl_inline (vlib_main_t * vm,
 	  table_index0 = vnet_buffer (b0)->l2_classify.table_index;
 	  e0 = 0;
 	  t0 = 0;
-	  vnet_get_config_data (am->vnet_config_main[tid],
+	  vnet_get_config_data (am->vnet_config_main[is_output][tid],
 				&b0->current_config_index, &next0,
 				/* # bytes of config data */ 0);
 
@@ -265,6 +331,10 @@ ip_inacl_inline (vlib_main_t * vm,
 	      else
 		h0 = b0->data;
 
+	      /* advance the match pointer so the matching happens on IP header */
+	      if (is_output)
+		h0 += vnet_buffer (b0)->l2_classify.pad.l2_len;
+
 	      e0 = vnet_classify_find_entry (t0, (u8 *) h0, hash0, now);
 	      if (e0)
 		{
@@ -279,17 +349,23 @@ ip_inacl_inline (vlib_main_t * vm,
 
 		  if (is_ip4)
 		    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-		      IP4_ERROR_INACL_SESSION_DENY : IP4_ERROR_NONE;
+		      (is_output ? IP4_ERROR_OUTACL_SESSION_DENY :
+		       IP4_ERROR_INACL_SESSION_DENY) : IP4_ERROR_NONE;
 		  else
 		    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-		      IP6_ERROR_INACL_SESSION_DENY : IP6_ERROR_NONE;
+		      (is_output ? IP6_ERROR_OUTACL_SESSION_DENY :
+		       IP6_ERROR_INACL_SESSION_DENY) : IP6_ERROR_NONE;
 		  b0->error = error_node->errors[error0];
 
-		  if (e0->action == CLASSIFY_ACTION_SET_IP4_FIB_INDEX ||
-		      e0->action == CLASSIFY_ACTION_SET_IP6_FIB_INDEX)
-		    vnet_buffer (b0)->sw_if_index[VLIB_TX] = e0->metadata;
-		  else if (e0->action == CLASSIFY_ACTION_SET_METADATA)
-		    vnet_buffer (b0)->ip.adj_index[VLIB_TX] = e0->metadata;
+		  if (!is_output)
+		    {
+		      if (e0->action == CLASSIFY_ACTION_SET_IP4_FIB_INDEX ||
+			  e0->action == CLASSIFY_ACTION_SET_IP6_FIB_INDEX)
+			vnet_buffer (b0)->sw_if_index[VLIB_TX] = e0->metadata;
+		      else if (e0->action == CLASSIFY_ACTION_SET_METADATA)
+			vnet_buffer (b0)->ip.adj_index[VLIB_TX] =
+			  e0->metadata;
+		    }
 		}
 	      else
 		{
@@ -307,10 +383,12 @@ ip_inacl_inline (vlib_main_t * vm,
 
 			  if (is_ip4)
 			    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-			      IP4_ERROR_INACL_TABLE_MISS : IP4_ERROR_NONE;
+			      (is_output ? IP4_ERROR_OUTACL_TABLE_MISS :
+			       IP4_ERROR_INACL_TABLE_MISS) : IP4_ERROR_NONE;
 			  else
 			    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-			      IP6_ERROR_INACL_TABLE_MISS : IP6_ERROR_NONE;
+			      (is_output ? IP6_ERROR_OUTACL_TABLE_MISS :
+			       IP6_ERROR_INACL_TABLE_MISS) : IP6_ERROR_NONE;
 			  b0->error = error_node->errors[error0];
 			  break;
 			}
@@ -338,10 +416,12 @@ ip_inacl_inline (vlib_main_t * vm,
 
 			  if (is_ip4)
 			    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-			      IP4_ERROR_INACL_SESSION_DENY : IP4_ERROR_NONE;
+			      (is_output ? IP4_ERROR_OUTACL_SESSION_DENY :
+			       IP4_ERROR_INACL_SESSION_DENY) : IP4_ERROR_NONE;
 			  else
 			    error0 = (next0 == ACL_NEXT_INDEX_DENY) ?
-			      IP6_ERROR_INACL_SESSION_DENY : IP6_ERROR_NONE;
+			      (is_output ? IP6_ERROR_OUTACL_SESSION_DENY :
+			       IP6_ERROR_INACL_SESSION_DENY) : IP6_ERROR_NONE;
 			  b0->error = error_node->errors[error0];
 
 			  if (e0->action == CLASSIFY_ACTION_SET_IP4_FIB_INDEX
@@ -358,14 +438,21 @@ ip_inacl_inline (vlib_main_t * vm,
 	  if (PREDICT_FALSE ((node->flags & VLIB_NODE_FLAG_TRACE)
 			     && (b0->flags & VLIB_BUFFER_IS_TRACED)))
 	    {
-	      ip_inacl_trace_t *t =
+	      ip_in_out_acl_trace_t *t =
 		vlib_add_trace (vm, node, b0, sizeof (*t));
-	      t->sw_if_index = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+	      t->sw_if_index =
+		vnet_buffer (b0)->sw_if_index[is_output ? VLIB_TX : VLIB_RX];
 	      t->next_index = next0;
 	      t->table_index = t0 ? t0 - vcm->tables : ~0;
 	      t->offset = (e0 && t0) ? vnet_classify_get_offset (t0, e0) : ~0;
 	    }
 
+	  if ((next0 == ACL_NEXT_INDEX_DENY) && is_output)
+	    {
+	      /* on output, for the drop node to work properly, go back to ip header */
+	      vlib_buffer_advance (b0, vnet_buffer (b0)->l2.l2_len);
+	    }
+
 	  /* verify speculative enqueue, maybe switch current next frame */
 	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
 					   to_next, n_left_to_next,
@@ -376,10 +463,13 @@ ip_inacl_inline (vlib_main_t * vm,
     }
 
   vlib_node_increment_counter (vm, node->node_index,
+			       is_output ? IP_OUTACL_ERROR_MISS :
 			       IP_INACL_ERROR_MISS, misses);
   vlib_node_increment_counter (vm, node->node_index,
+			       is_output ? IP_OUTACL_ERROR_HIT :
 			       IP_INACL_ERROR_HIT, hits);
   vlib_node_increment_counter (vm, node->node_index,
+			       is_output ? IP_OUTACL_ERROR_CHAIN_HIT :
 			       IP_INACL_ERROR_CHAIN_HIT, chain_hits);
   return frame->n_vectors;
 }
@@ -387,7 +477,16 @@ ip_inacl_inline (vlib_main_t * vm,
 static uword
 ip4_inacl (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
 {
-  return ip_inacl_inline (vm, node, frame, 1 /* is_ip4 */ );
+  return ip_in_out_acl_inline (vm, node, frame, 1 /* is_ip4 */ ,
+			       0 /* is_output */ );
+}
+
+static uword
+ip4_outacl (vlib_main_t * vm, vlib_node_runtime_t * node,
+	    vlib_frame_t * frame)
+{
+  return ip_in_out_acl_inline (vm, node, frame, 1 /* is_ip4 */ ,
+			       1 /* is_output */ );
 }
 
 
@@ -405,16 +504,39 @@ VLIB_REGISTER_NODE (ip4_inacl_node) = {
     [ACL_NEXT_INDEX_DENY] = "ip4-drop",
   },
 };
+
+VLIB_REGISTER_NODE (ip4_outacl_node) = {
+  .function = ip4_outacl,
+  .name = "ip4-outacl",
+  .vector_size = sizeof (u32),
+  .format_trace = format_ip_outacl_trace,
+  .n_errors = ARRAY_LEN(ip_outacl_error_strings),
+  .error_strings = ip_outacl_error_strings,
+
+  .n_next_nodes = ACL_NEXT_INDEX_N_NEXT,
+  .next_nodes = {
+    [ACL_NEXT_INDEX_DENY] = "ip4-drop",
+  },
+};
 /* *INDENT-ON* */
 
 VLIB_NODE_FUNCTION_MULTIARCH (ip4_inacl_node, ip4_inacl);
+VLIB_NODE_FUNCTION_MULTIARCH (ip4_outacl_node, ip4_outacl);
 
 static uword
 ip6_inacl (vlib_main_t * vm, vlib_node_runtime_t * node, vlib_frame_t * frame)
 {
-  return ip_inacl_inline (vm, node, frame, 0 /* is_ip4 */ );
+  return ip_in_out_acl_inline (vm, node, frame, 0 /* is_ip4 */ ,
+			       0 /* is_output */ );
 }
 
+static uword
+ip6_outacl (vlib_main_t * vm, vlib_node_runtime_t * node,
+	    vlib_frame_t * frame)
+{
+  return ip_in_out_acl_inline (vm, node, frame, 0 /* is_ip4 */ ,
+			       1 /* is_output */ );
+}
 
 /* *INDENT-OFF* */
 VLIB_REGISTER_NODE (ip6_inacl_node) = {
@@ -430,17 +552,32 @@ VLIB_REGISTER_NODE (ip6_inacl_node) = {
     [ACL_NEXT_INDEX_DENY] = "ip6-drop",
   },
 };
+
+VLIB_REGISTER_NODE (ip6_outacl_node) = {
+  .function = ip6_outacl,
+  .name = "ip6-outacl",
+  .vector_size = sizeof (u32),
+  .format_trace = format_ip_outacl_trace,
+  .n_errors = ARRAY_LEN(ip_outacl_error_strings),
+  .error_strings = ip_outacl_error_strings,
+
+  .n_next_nodes = ACL_NEXT_INDEX_N_NEXT,
+  .next_nodes = {
+    [ACL_NEXT_INDEX_DENY] = "ip6-drop",
+  },
+};
 /* *INDENT-ON* */
 
 VLIB_NODE_FUNCTION_MULTIARCH (ip6_inacl_node, ip6_inacl);
+VLIB_NODE_FUNCTION_MULTIARCH (ip6_outacl_node, ip6_outacl);
 
 static clib_error_t *
-ip_inacl_init (vlib_main_t * vm)
+ip_in_out_acl_init (vlib_main_t * vm)
 {
   return 0;
 }
 
-VLIB_INIT_FUNCTION (ip_inacl_init);
+VLIB_INIT_FUNCTION (ip_in_out_acl_init);
 
 
 /*
diff --git a/src/vnet/ip/ip_init.c b/src/vnet/ip/ip_init.c
index e181a4fdd9b..ddd66c2e817 100644
--- a/src/vnet/ip/ip_init.c
+++ b/src/vnet/ip/ip_init.c
@@ -128,7 +128,7 @@ do {						\
   if ((error = vlib_call_init_function (vm, ip_classify_init)))
     return error;
 
-  if ((error = vlib_call_init_function (vm, input_acl_init)))
+  if ((error = vlib_call_init_function (vm, in_out_acl_init)))
     return error;
 
   if ((error = vlib_call_init_function (vm, policer_classify_init)))