[CDRIVER-5604] Change updates to SRV rescan interval to be atomic (#1640)

Because the topology's rescan interval can be updated
in parallel with other threads reading its value, the
value needs to be atomic to prevent data races. This
does not necessarily fix race conditions, but does avoid
the UB of data races.

Author: vector-of-bool

src/libmongoc/src/mongoc/mongoc-topology-background-monitoring.c

@@ -53,7 +53,7 @@ static BSON_THREAD_FUN (srv_polling_run, topology_void)
       /* Unlock and sleep until next scan is due, or until shutdown signalled.
        */
       now_ms = bson_get_monotonic_time () / 1000;
-      scan_due_ms = topology->srv_polling_last_scan_ms + topology->srv_polling_rescan_interval_ms;
+      scan_due_ms = topology->srv_polling_last_scan_ms + _mongoc_topology_get_srv_polling_rescan_interval_ms (topology);
       sleep_duration_ms = scan_due_ms - now_ms;
 
       if (sleep_duration_ms > 0) {

src/libmongoc/src/mongoc/mongoc-topology-private.h

@@ -128,8 +128,10 @@ typedef struct _mongoc_topology_t {
    int64_t min_heartbeat_frequency_msec;
 
    /* Minimum of SRV record TTLs, but no lower than 60 seconds.
-    * May be zero for non-SRV/non-MongoS topology. */
-   int64_t srv_polling_rescan_interval_ms;
+    * May be zero for non-SRV/non-MongoS topology.
+    * DO NOT access directly: use the accessor methods to get/set the value.
+    */
+   int64_t _atomic_srv_polling_rescan_interval_ms;
    int64_t srv_polling_last_scan_ms;
    /* For multi-threaded, srv polling occurs in a separate thread. */
    bson_thread_t srv_polling_thread;
@@ -440,12 +442,23 @@ mongoc_topology_should_rescan_srv (mongoc_topology_t *topology);
 void
 _mongoc_topology_set_rr_resolver (mongoc_topology_t *topology, _mongoc_rr_resolver_fn rr_resolver);
 
-/* _mongoc_topology_set_srv_polling_rescan_interval_ms is called by tests to
- * shorten the rescan interval.
- * Callers should call this before monitoring starts.
+/**
+ * @brief Thread-safe update the SRV polling rescan interval on the given topology
  */
-void
-_mongoc_topology_set_srv_polling_rescan_interval_ms (mongoc_topology_t *topology, int64_t val);
+static inline void
+_mongoc_topology_set_srv_polling_rescan_interval_ms (mongoc_topology_t *topology, int64_t val)
+{
+   bson_atomic_int64_exchange (&topology->_atomic_srv_polling_rescan_interval_ms, val, bson_memory_order_seq_cst);
+}
+
+/**
+ * @brief Thread-safe get the SRV polling interval
+ */
+static inline int64_t
+_mongoc_topology_get_srv_polling_rescan_interval_ms (mongoc_topology_t const *topology)
+{
+   return bson_atomic_int64_fetch (&topology->_atomic_srv_polling_rescan_interval_ms, bson_memory_order_seq_cst);
+}
 
 /**
  * @brief Return the latest connection generation for the server_id and/or

src/libmongoc/src/mongoc/mongoc-topology.c

@@ -480,7 +480,7 @@ mongoc_topology_new (const mongoc_uri_t *uri, bool single_threaded)
        * lookup fails, SRV polling will still start when background monitoring
        * starts. */
       topology->srv_polling_last_scan_ms = bson_get_monotonic_time () / 1000;
-      topology->srv_polling_rescan_interval_ms = MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS;
+      _mongoc_topology_set_srv_polling_rescan_interval_ms (topology, MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS);
 
       /* a mongodb+srv URI. try SRV lookup, if no error then also try TXT */
       prefixed_hostname = bson_strdup_printf ("_%s._tcp.%s", mongoc_uri_get_srv_service_name (uri), srv_hostname);
@@ -518,8 +518,8 @@ mongoc_topology_new (const mongoc_uri_t *uri, bool single_threaded)
 
       topology->srv_polling_last_scan_ms = bson_get_monotonic_time () / 1000;
       /* TODO (CDRIVER-4047) use BSON_MIN */
-      topology->srv_polling_rescan_interval_ms =
-         BSON_MAX (rr_data.min_ttl * 1000, MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS);
+      int64_t new_iv = BSON_MAX (rr_data.min_ttl * 1000, MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS);
+      _mongoc_topology_set_srv_polling_rescan_interval_ms (topology, new_iv);
 
       topology->valid = true;
    srv_fail:
@@ -816,7 +816,7 @@ mongoc_topology_rescan_srv (mongoc_topology_t *topology)
    BSON_ASSERT (mongoc_topology_should_rescan_srv (topology));
 
    srv_hostname = mongoc_uri_get_srv_hostname (topology->uri);
-   scan_time_ms = topology->srv_polling_last_scan_ms + topology->srv_polling_rescan_interval_ms;
+   scan_time_ms = topology->srv_polling_last_scan_ms + _mongoc_topology_get_srv_polling_rescan_interval_ms (topology);
    if (bson_get_monotonic_time () / 1000 < scan_time_ms) {
       /* Query SRV no more frequently than srv_polling_rescan_interval_ms. */
       return;
@@ -839,14 +839,14 @@ mongoc_topology_rescan_srv (mongoc_topology_t *topology)
    topology->srv_polling_last_scan_ms = bson_get_monotonic_time () / 1000;
    if (!ret) {
       /* Failed querying, soldier on and try again next time. */
-      topology->srv_polling_rescan_interval_ms = td.ptr->heartbeat_msec;
+      _mongoc_topology_set_srv_polling_rescan_interval_ms (topology, td.ptr->heartbeat_msec);
       MONGOC_ERROR ("SRV polling error: %s", topology->scanner->error.message);
       GOTO (done);
    }
 
    /* TODO (CDRIVER-4047) use BSON_MIN */
-   topology->srv_polling_rescan_interval_ms =
-      BSON_MAX (rr_data.min_ttl * 1000, MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS);
+   const int64_t new_iv = BSON_MAX (rr_data.min_ttl * 1000, MONGOC_TOPOLOGY_MIN_RESCAN_SRV_INTERVAL_MS);
+   _mongoc_topology_set_srv_polling_rescan_interval_ms (topology, new_iv);
 
    tdmod = mc_tpld_modify_begin (topology);
    if (!mongoc_topology_apply_scanned_srv_hosts (
@@ -861,7 +861,7 @@ mongoc_topology_rescan_srv (mongoc_topology_t *topology)
        * to heartbeatFrequencyMS until at least one verified SRV record is
        * obtained."
        */
-      topology->srv_polling_rescan_interval_ms = td.ptr->heartbeat_msec;
+      _mongoc_topology_set_srv_polling_rescan_interval_ms (topology, td.ptr->heartbeat_msec);
    }
    mc_tpld_modify_commit (tdmod);
 
@@ -1889,12 +1889,6 @@ _mongoc_topology_set_rr_resolver (mongoc_topology_t *topology, _mongoc_rr_resolv
    topology->rr_resolver = rr_resolver;
 }
 
-void
-_mongoc_topology_set_srv_polling_rescan_interval_ms (mongoc_topology_t *topology, int64_t val)
-{
-   topology->srv_polling_rescan_interval_ms = val;
-}
-
 uint32_t
 _mongoc_topology_get_connection_pool_generation (const mongoc_topology_description_t *td,
                                                  uint32_t server_id,