metrics: Add new relay metrics to MetricsPort

This commit adds the total number of DROP cell seen, the total number of
DESTROY cell received and the total number of protocol violation that lead to a
circuit close.

Closes #40816

Signed-off-by: David Goulet <dgoulet@torproject.org>

changes/ticket40816

@@ -0,0 +1,4 @@
+  o Minor feature (metrics port, relay):
+    - Add new metrics for relays on the MetricsPort namely the count of drop
+      cell, destroy cell and the number of circuit protocol violation seen that
+      lead to a circuit close. Closes ticket 40816.

src/core/or/circuitlist.c

@@ -160,6 +160,10 @@ double cc_stats_circ_close_ss_cwnd_ma = 0;
 
 uint64_t cc_stats_circs_closed = 0;
 
+/** Total number of circuit protocol violation. This is incremented when the
+ * END_CIRC_REASON_TORPROTOCOL is used to close a circuit. */
+uint64_t circ_n_proto_violation = 0;
+
 /********* END VARIABLES ************/
 
 /* Implement circuit handle helpers. */
@@ -2197,6 +2201,10 @@ circuit_mark_for_close_, (circuit_t *circ, int reason, int line,
   tor_assert(line);
   tor_assert(file);
 
+  if (reason == END_CIRC_REASON_TORPROTOCOL) {
+    circ_n_proto_violation++;
+  }
+
   /* Check whether the circuitpadding subsystem wants to block this close */
   if (circpad_marked_circuit_for_padding(circ, reason)) {
     return;

src/core/or/circuitlist.h

@@ -172,6 +172,7 @@
 extern double cc_stats_circ_close_cwnd_ma;
 extern double cc_stats_circ_close_ss_cwnd_ma;
 extern uint64_t cc_stats_circs_closed;
+extern uint64_t circ_n_proto_violation;
 
 /** Convert a circuit_t* to a pointer to the enclosing or_circuit_t.  Assert
  * if the cast is impossible. */

src/feature/relay/relay_metrics.c

@@ -13,6 +13,7 @@
 #include "core/or/or.h"
 #include "core/mainloop/connection.h"
 #include "core/mainloop/mainloop.h"
+#include "core/or/command.h"
 #include "core/or/congestion_control_common.h"
 #include "core/or/congestion_control_vegas.h"
 #include "core/or/congestion_control_flow.h"
@@ -54,6 +55,9 @@ static void fill_socket_values(void);
 static void fill_onionskins_values(void);
 static void fill_oom_values(void);
 static void fill_streams_values(void);
+static void fill_relay_circ_proto_violation(void);
+static void fill_relay_destroy_cell(void);
+static void fill_relay_drop_cell(void);
 static void fill_relay_flags(void);
 static void fill_tcp_exhaustion_values(void);
 static void fill_traffic_values(void);
@@ -217,6 +221,27 @@ static const relay_metrics_entry_t base_metrics[] =
     .help = "Total number of REND1 cells we received",
     .fill_fn = fill_rend1_cells,
   },
+  {
+    .key = RELAY_METRICS_CIRC_DESTROY_CELL,
+    .type = METRICS_TYPE_COUNTER,
+    .name = METRICS_NAME(relay_destroy_cell_total),
+    .help = "Total number of DESTROY cell we received",
+    .fill_fn = fill_relay_destroy_cell,
+  },
+  {
+    .key = RELAY_METRICS_CIRC_PROTO_VIOLATION,
+    .type = METRICS_TYPE_COUNTER,
+    .name = METRICS_NAME(relay_circ_proto_violation_total),
+    .help = "Total number of circuit protocol violation",
+    .fill_fn = fill_relay_circ_proto_violation,
+  },
+  {
+    .key = RELAY_METRICS_CIRC_DROP_CELL,
+    .type = METRICS_TYPE_COUNTER,
+    .name = METRICS_NAME(relay_drop_cell_total),
+    .help = "Total number of DROP cell we received",
+    .fill_fn = fill_relay_drop_cell,
+  },
 };
 static const size_t num_base_metrics = ARRAY_LENGTH(base_metrics);
 
@@ -1206,6 +1231,46 @@ fill_rend1_cells(void)
   }
 }
 
+/** Fill the metrics store for the RELAY_METRICS_CIRC_DESTROY_CELL counter. */
+static void
+fill_relay_destroy_cell(void)
+{
+  metrics_store_entry_t *sentry;
+  const relay_metrics_entry_t *rentry =
+    &base_metrics[RELAY_METRICS_CIRC_DESTROY_CELL];
+
+  sentry = metrics_store_add(the_store, rentry->type, rentry->name,
+                             rentry->help, 0, NULL);
+  metrics_store_entry_update(sentry,
+                             (int64_t) stats_n_destroy_cells_processed);
+}
+
+/** Fill the metrics store for the RELAY_METRICS_CIRC_DROP_CELL counter. */
+static void
+fill_relay_drop_cell(void)
+{
+  metrics_store_entry_t *sentry;
+  const relay_metrics_entry_t *rentry =
+    &base_metrics[RELAY_METRICS_CIRC_DROP_CELL];
+
+  sentry = metrics_store_add(the_store, rentry->type, rentry->name,
+                             rentry->help, 0, NULL);
+  metrics_store_entry_update(sentry, rep_hist_get_drop_cell_received_count());
+}
+
+/** Fill the metrics store for the RELAY_METRICS_CIRC_PROTO_VIOLATION. */
+static void
+fill_relay_circ_proto_violation(void)
+{
+  metrics_store_entry_t *sentry;
+  const relay_metrics_entry_t *rentry =
+    &base_metrics[RELAY_METRICS_CIRC_PROTO_VIOLATION];
+
+  sentry = metrics_store_add(the_store, rentry->type, rentry->name,
+                             rentry->help, 0, NULL);
+  metrics_store_entry_update(sentry, circ_n_proto_violation);
+}
+
 /** Reset the global store and fill it with all the metrics from base_metrics
  * and their associated values.
  *

src/feature/relay/relay_metrics.h

@@ -57,6 +57,12 @@ typedef enum {
   RELAY_METRICS_NUM_INTRO1_CELLS,
   /** Number of times we received a REND1 cell */
   RELAY_METRICS_NUM_REND1_CELLS,
+  /** Number of circuit closed by receiving a DESTROY cell. */
+  RELAY_METRICS_CIRC_DESTROY_CELL,
+  /** Number of circuits closed due to protocol violation. */
+  RELAY_METRICS_CIRC_PROTO_VIOLATION,
+  /** Number of drop cell seen. */
+  RELAY_METRICS_CIRC_DROP_CELL,
 } relay_metrics_key_t;
 
 /** The metadata of a relay metric. */

src/feature/stats/rephist.c

@@ -280,6 +280,9 @@ static dns_stats_t dns_AAAA_stats;
 /** DNS query statistics store. It covers all type of queries. */
 static dns_stats_t dns_all_stats;
 
+/** Counter of the total number of DROP cell received. */
+static uint64_t relay_circ_n_drop_cell_received = 0;
+
 /** Return the point to the DNS statistics store. Ignore the type for now
  * because of a libevent problem. */
 static inline dns_stats_t *
@@ -2815,6 +2818,8 @@ rep_hist_padding_count_write(padding_type_t type)
   switch (type) {
     case PADDING_TYPE_DROP:
       padding_current.write_drop_cell_count++;
+      /* Padding stats get reset thus why we have two counters. */
+      relay_circ_n_drop_cell_received++;
       break;
     case PADDING_TYPE_CELL:
       padding_current.write_pad_cell_count++;
@@ -3022,6 +3027,13 @@ rep_hist_consensus_has_changed(const networkstatus_t *ns)
                             OVERLOAD_ONIONSKIN_NTOR_PERIOD_SECS_MAX);
 }
 
+/** Relay Only: return the total number of DROP cell received. */
+uint64_t
+rep_hist_get_drop_cell_received_count(void)
+{
+  return relay_circ_n_drop_cell_received;
+}
+
 #ifdef TOR_UNIT_TESTS
 /* only exists for unit tests: get HSv2 stats object */
 const hs_v2_stats_t *

src/feature/stats/rephist.h

@@ -192,6 +192,8 @@ uint64_t rep_hist_get_n_tcp_exhaustion(void);
 uint64_t rep_hist_get_n_read_limit_reached(void);
 uint64_t rep_hist_get_n_write_limit_reached(void);
 
+uint64_t rep_hist_get_drop_cell_received_count(void);
+
 #ifdef TOR_UNIT_TESTS
 struct hs_v2_stats_t;
 const struct hs_v2_stats_t *rep_hist_get_hs_v2_stats(void);