acl: implement counters

implement per-acl-number counters in the stats segment.
They are created during the ACL creation,
the counters are incremented in the dataplane using
the new inline function with the extra parameter being
the packet size. Counting in shared segment adds
a noticeable overhead, so add also an API to
turn the counters on.

Type: feature

Change-Id: I8af7b0c31a3d986b68089eb52452aed45df66c7b
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>

5 files changed, 183 insertions, 9 deletions

diff --git a/src/plugins/acl/acl.api b/src/plugins/acl/acl.api
index cde46c444dd..0c230c5fd54 100644
--- a/src/plugins/acl/acl.api
+++ b/src/plugins/acl/acl.api
@@ -547,3 +547,15 @@ define acl_interface_etype_whitelist_details
   u16 whitelist[count];
 };
 
+/** \brief Enable or disable incrementing ACL counters in stats segment by interface processing
+    @param client_index - opaque cookie to identify the sender
+    @param context - sender context, to match reply w/ request
+    @param enable - whether to enable or disable incrementing the counters
+*/
+
+autoreply define acl_stats_intf_counters_enable
+{
+  u32 client_index;
+  u32 context;
+  bool enable;
+};
diff --git a/src/plugins/acl/acl.c b/src/plugins/acl/acl.c
index 9ebb349d1b1..24dd53bd749 100644
--- a/src/plugins/acl/acl.c
+++ b/src/plugins/acl/acl.c
@@ -86,7 +86,8 @@ _(MACIP_ACL_INTERFACE_GET, macip_acl_interface_get) \
 _(MACIP_ACL_INTERFACE_LIST_DUMP, macip_acl_interface_list_dump) \
 _(ACL_INTERFACE_SET_ETYPE_WHITELIST, acl_interface_set_etype_whitelist) \
 _(ACL_INTERFACE_ETYPE_WHITELIST_DUMP, acl_interface_etype_whitelist_dump) \
-_(ACL_PLUGIN_GET_CONN_TABLE_MAX_ENTRIES,acl_plugin_get_conn_table_max_entries)
+_(ACL_PLUGIN_GET_CONN_TABLE_MAX_ENTRIES,acl_plugin_get_conn_table_max_entries) \
+_(ACL_STATS_INTF_COUNTERS_ENABLE, acl_stats_intf_counters_enable)
 
 
 /* *INDENT-OFF* */
@@ -373,6 +374,33 @@ policy_notify_acl_change (acl_main_t * am, u32 acl_num)
 }
 
 
+static void
+validate_and_reset_acl_counters (acl_main_t * am, u32 acl_index)
+{
+  int i;
+  /* counters are set as vectors [acl#] pointing to vectors of [acl rule] */
+  acl_plugin_counter_lock (am);
+
+  int old_len = vec_len (am->combined_acl_counters);
+
+  vec_validate (am->combined_acl_counters, acl_index);
+
+  for (i = old_len; i < vec_len (am->combined_acl_counters); i++)
+    {
+      am->combined_acl_counters[i].name = 0;
+      /* filled in once only */
+      am->combined_acl_counters[i].stat_segment_name = (void *)
+	format (0, "/acl/%d/matches%c", i, 0);
+      clib_warning ("add stats segment: %s",
+		    am->combined_acl_counters[i].stat_segment_name);
+      i32 rule_count = vec_len (am->acls[acl_index].rules);
+      /* Validate one extra so we always have at least one counter for an ACL */
+      vlib_validate_combined_counter (&am->combined_acl_counters[i],
+				      rule_count);
+      vlib_zero_combined_counter (&am->combined_acl_counters[i], rule_count);
+    }
+  acl_plugin_counter_unlock (am);
+}
 
 static int
 acl_add_list (u32 count, vl_api_acl_rule_t rules[],
@@ -465,6 +493,11 @@ acl_add_list (u32 count, vl_api_acl_rule_t rules[],
       policy_notify_acl_change (am, *acl_list_index);
     }
 
+  /* stats segment expects global heap, so restore it temporarily */
+  clib_mem_set_heap (oldheap);
+  validate_and_reset_acl_counters (am, *acl_list_index);
+  oldheap = acl_set_heap (am);
+
   /* notify the lookup contexts about the ACL changes */
   acl_plugin_lookup_context_notify_acl_change (*acl_list_index);
   clib_mem_set_heap (oldheap);
@@ -662,6 +695,16 @@ acl_interface_out_enable_disable (acl_main_t * am, u32 sw_if_index,
 }
 
 static int
+acl_stats_intf_counters_enable_disable (acl_main_t * am, int enable_disable)
+{
+  int rv = 0;
+
+  am->interface_acl_counters_enabled = enable_disable;
+
+  return rv;
+}
+
+static int
 acl_interface_inout_enable_disable (acl_main_t * am, u32 sw_if_index,
 				    int is_input, int enable_disable)
 {
@@ -1893,6 +1936,21 @@ vl_api_acl_del_t_handler (vl_api_acl_del_t * mp)
   REPLY_MACRO (VL_API_ACL_DEL_REPLY);
 }
 
+
+static void
+  vl_api_acl_stats_intf_counters_enable_t_handler
+  (vl_api_acl_stats_intf_counters_enable_t * mp)
+{
+  acl_main_t *am = &acl_main;
+  vl_api_acl_stats_intf_counters_enable_reply_t *rmp;
+  int rv;
+
+  rv = acl_stats_intf_counters_enable_disable (am, ntohl (mp->enable));
+
+  REPLY_MACRO (VL_API_ACL_DEL_REPLY);
+}
+
+
 static void
 vl_api_acl_interface_add_del_t_handler (vl_api_acl_interface_add_del_t * mp)
 {
@@ -3390,6 +3448,8 @@ acl_show_aclplugin_tables_fn (vlib_main_t * vm,
       show_applied_info = 1;
       show_bihash = 1;
     }
+  vlib_cli_output (vm, "Stats counters enabled for interface ACLs: %d",
+		   acl_main.interface_acl_counters_enabled);
   if (show_mask_type)
     acl_plugin_show_tables_mask_type ();
   if (show_acl_hash_info)
@@ -3659,6 +3719,10 @@ acl_init (vlib_main_t * vm)
     acl_plugin.register_user_module ("interface ACL", "sw_if_index",
 				     "is_input");
 
+  am->acl_counter_lock = clib_mem_alloc_aligned (CLIB_CACHE_LINE_BYTES,
+						 CLIB_CACHE_LINE_BYTES);
+  am->acl_counter_lock[0] = 0;	/* should be no need */
+
   return error;
 }
 
diff --git a/src/plugins/acl/acl.h b/src/plugins/acl/acl.h
index 53650916423..229ee057725 100644
--- a/src/plugins/acl/acl.h
+++ b/src/plugins/acl/acl.h
@@ -27,6 +27,7 @@
 #include <vppinfra/bihash_48_8.h>
 #include <vppinfra/bihash_40_8.h>
 #include <vppinfra/bihash_16_8.h>
+#include <vlib/counter.h>
 
 #include "types.h"
 #include "fa_node.h"
@@ -34,7 +35,7 @@
 #include "lookup_context.h"
 
 #define  ACL_PLUGIN_VERSION_MAJOR 1
-#define  ACL_PLUGIN_VERSION_MINOR 3
+#define  ACL_PLUGIN_VERSION_MINOR 4
 
 #define UDP_SESSION_IDLE_TIMEOUT_SEC 600
 #define TCP_SESSION_IDLE_TIMEOUT_SEC (3600*24)
@@ -314,6 +315,11 @@ typedef struct {
   vnet_main_t * vnet_main;
   /* logging */
   vlib_log_class_t log_default;
+  /* acl counters exposed via stats segment */
+  volatile u32 *acl_counter_lock;
+  vlib_combined_counter_main_t *combined_acl_counters;
+  /* enable/disable ACL counters for interface processing */
+  u32 interface_acl_counters_enabled;
 } acl_main_t;
 
 #define acl_log_err(...) \
@@ -326,6 +332,21 @@ typedef struct {
   vlib_log(VLIB_LOG_LEVEL_INFO, acl_main.log_default, __VA_ARGS__)
 
 
+static inline void
+acl_plugin_counter_lock (acl_main_t * am)
+{
+  if (am->acl_counter_lock)
+    while (clib_atomic_test_and_set (am->acl_counter_lock))
+      /* zzzz */ ;
+}
+
+static inline void
+acl_plugin_counter_unlock (acl_main_t * am)
+{
+  if (am->acl_counter_lock)
+    clib_atomic_release (am->acl_counter_lock);
+}
+
 
 #define foreach_acl_eh                                          \
    _(HOPBYHOP , 0  , "IPv6ExtHdrHopByHop")                      \
diff --git a/src/plugins/acl/dataplane_node.c b/src/plugins/acl/dataplane_node.c
index 0bdcc850054..c738f664571 100644
--- a/src/plugins/acl/dataplane_node.c
+++ b/src/plugins/acl/dataplane_node.c
@@ -565,6 +565,11 @@ acl_fa_inner_node_fn (vlib_main_t * vm,
   u32 *sw_if_index;
   fa_5tuple_t *fa_5tuple;
   u64 *hash;
+  /* for the delayed counters */
+  u32 saved_matched_acl_index = 0;
+  u32 saved_matched_ace_index = 0;
+  u32 saved_packet_count = 0;
+  u32 saved_byte_count = 0;
 
   from = vlib_frame_vector_args (frame);
   error_node = vlib_node_get_runtime (vm, node->node_index);
@@ -690,13 +695,34 @@ acl_fa_inner_node_fn (vlib_main_t * vm,
 		  am->output_lc_index_by_sw_if_index[sw_if_index[0]];
 
 	      action = 0;	/* deny by default */
-	      acl_plugin_match_5tuple_inline (am, lc_index0,
-					      (fa_5tuple_opaque_t *) &
-					      fa_5tuple[0], is_ip6, &action,
-					      &match_acl_pos,
-					      &match_acl_in_index,
-					      &match_rule_index,
-					      &trace_bitmap);
+	      int is_match = acl_plugin_match_5tuple_inline (am, lc_index0,
+							     (fa_5tuple_opaque_t *) & fa_5tuple[0], is_ip6,
+							     &action,
+							     &match_acl_pos,
+							     &match_acl_in_index,
+							     &match_rule_index,
+							     &trace_bitmap);
+	      if (PREDICT_FALSE
+		  (is_match && am->interface_acl_counters_enabled))
+		{
+		  u32 buf_len = vlib_buffer_length_in_chain (vm, b[0]);
+		  vlib_increment_combined_counter (am->combined_acl_counters +
+						   saved_matched_acl_index,
+						   thread_index,
+						   saved_matched_ace_index,
+						   saved_packet_count,
+						   saved_byte_count);
+		  saved_matched_acl_index = match_acl_in_index;
+		  saved_matched_ace_index = match_rule_index;
+		  saved_packet_count = 1;
+		  saved_byte_count = buf_len;
+		  /* prefetch the counter that we are going to increment */
+		  vlib_prefetch_combined_counter (am->combined_acl_counters +
+						  saved_matched_acl_index,
+						  thread_index,
+						  saved_matched_ace_index);
+		}
+
 	      b[0]->error = error_node->errors[action];
 
 	      if (1 == action)
@@ -778,6 +804,16 @@ acl_fa_inner_node_fn (vlib_main_t * vm,
 
   vlib_buffer_enqueue_to_next (vm, node, from, pw->nexts, frame->n_vectors);
 
+  /*
+   * if we were had an acl match then we have a counter to increment.
+   * else it is all zeroes, so this will be harmless.
+   */
+  vlib_increment_combined_counter (am->combined_acl_counters +
+				   saved_matched_acl_index,
+				   thread_index,
+				   saved_matched_ace_index,
+				   saved_packet_count, saved_byte_count);
+
   vlib_node_increment_counter (vm, node->node_index,
 			       ACL_FA_ERROR_ACL_CHECK, frame->n_vectors);
   vlib_node_increment_counter (vm, node->node_index,
diff --git a/src/plugins/acl/public_inlines.h b/src/plugins/acl/public_inlines.h
index 03b64012a74..6b69bcef61e 100644
--- a/src/plugins/acl/public_inlines.h
+++ b/src/plugins/acl/public_inlines.h
@@ -682,5 +682,46 @@ acl_plugin_match_5tuple_inline (void *p_acl_main, u32 lc_index,
 }
 
 
+always_inline int
+acl_plugin_match_5tuple_inline_and_count (void *p_acl_main, u32 lc_index,
+                                           fa_5tuple_opaque_t * pkt_5tuple,
+                                           int is_ip6, u8 * r_action,
+                                           u32 * r_acl_pos_p,
+                                           u32 * r_acl_match_p,
+                                           u32 * r_rule_match_p,
+                                           u32 * trace_bitmap,
+					   u32 packet_size)
+{
+  acl_main_t *am = p_acl_main;
+  int ret = 0;
+  fa_5tuple_t * pkt_5tuple_internal = (fa_5tuple_t *)pkt_5tuple;
+  pkt_5tuple_internal->pkt.lc_index = lc_index;
+  if (PREDICT_TRUE(am->use_hash_acl_matching)) {
+    if (PREDICT_FALSE(pkt_5tuple_internal->pkt.is_nonfirst_fragment)) {
+      /*
+       * tuplemerge does not take fragments into account,
+       * and in general making fragments first class citizens has
+       * proved more overhead than it's worth - so just fall back to linear
+       * matching in that case.
+       */
+      ret = linear_multi_acl_match_5tuple(p_acl_main, lc_index, pkt_5tuple_internal, is_ip6, r_action,
+                                 r_acl_pos_p, r_acl_match_p, r_rule_match_p, trace_bitmap);
+    } else {
+      ret = hash_multi_acl_match_5tuple(p_acl_main, lc_index, pkt_5tuple_internal, is_ip6, r_action,
+                                 r_acl_pos_p, r_acl_match_p, r_rule_match_p, trace_bitmap);
+    }
+  } else {
+    ret = linear_multi_acl_match_5tuple(p_acl_main, lc_index, pkt_5tuple_internal, is_ip6, r_action,
+                                 r_acl_pos_p, r_acl_match_p, r_rule_match_p, trace_bitmap);
+  }
+  if (PREDICT_TRUE(ret)) {
+	  u16 thread_index = os_get_thread_index ();
+	  vlib_increment_combined_counter(am->combined_acl_counters + *r_acl_match_p, thread_index, *r_rule_match_p, 1, packet_size);
+  }
+  return ret;
+}
+
+
+
 
 #endif