relay: Use the configured number of threads for worker work calculation

We cap our number of CPU worker threads to at least 2 even if we have a
single core. But also, before we used to always add one extra thread
regardless of the number of core.

This meant that we were off when re-using the get_num_cpus() function
when calculating our onionskin work overhead because we were always off
by one.

This commit makes it that we always use the number of thread our actual
thread pool was configured with.

Fixes #40719

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

6 files changed, 31 insertions, 1 deletions

diff --git a/changes/ticket40719 b/changes/ticket40719
new file mode 100644
index 0000000000..eec84dce0f
--- /dev/null
+++ b/changes/ticket40719
@@ -0,0 +1,3 @@
+  o Minor bugfixes (cpuworker, relay):
+    - Fix an off by one overload calculation on the number of CPUs being used by
+      our thread pool. Fixes bug 40719; bugfix on 0.3.5.1-alpha.
diff --git a/src/core/mainloop/cpuworker.c b/src/core/mainloop/cpuworker.c
index 3ad556a184..39d4899075 100644
--- a/src/core/mainloop/cpuworker.c
+++ b/src/core/mainloop/cpuworker.c
@@ -144,6 +144,16 @@ cpu_init(void)
   set_max_pending_tasks(NULL);
 }
 
+/** Return the number of threads configured for our CPU worker. */
+unsigned int
+cpuworker_get_n_threads(void)
+{
+  if (!threadpool) {
+    return 0;
+  }
+  return threadpool_get_n_threads(threadpool);
+}
+
 /** Magic numbers to make sure our cpuworker_requests don't grow any
  * mis-framing bugs. */
 #define CPUWORKER_REQUEST_MAGIC 0xda4afeed
diff --git a/src/core/mainloop/cpuworker.h b/src/core/mainloop/cpuworker.h
index 94545af586..9eee287c1f 100644
--- a/src/core/mainloop/cpuworker.h
+++ b/src/core/mainloop/cpuworker.h
@@ -38,5 +38,7 @@ void cpuworker_log_onionskin_overhead(int severity, int onionskin_type,
                                       const char *onionskin_type_name);
 void cpuworker_cancel_circ_handshake(or_circuit_t *circ);
 
+unsigned int cpuworker_get_n_threads(void);
+
 #endif /* !defined(TOR_CPUWORKER_H) */
 
diff --git a/src/feature/relay/onion_queue.c b/src/feature/relay/onion_queue.c
index 6d1a6de15c..b844aefcd1 100644
--- a/src/feature/relay/onion_queue.c
+++ b/src/feature/relay/onion_queue.c
@@ -152,7 +152,13 @@ have_room_for_onionskin(uint16_t type)
   /* If we've got fewer than 50 entries, we always have room for one more. */
   if (ol_entries[type] < 50)
     return 1;
-  num_cpus = get_num_cpus(options);
+
+  /* If zero, this means our thread pool was never initialized meaning we can't
+   * really get here but make sure we don't have such value because we are
+   * using as a divisor. */
+  num_cpus = cpuworker_get_n_threads();
+  tor_assert(num_cpus > 0);
+
   max_onion_queue_delay = get_onion_queue_max_delay(options);
 
   /* Compute how many microseconds we'd expect to need to clear all
diff --git a/src/lib/evloop/workqueue.c b/src/lib/evloop/workqueue.c
index 603dddd5a3..bc929148eb 100644
--- a/src/lib/evloop/workqueue.c
+++ b/src/lib/evloop/workqueue.c
@@ -672,3 +672,11 @@ replyqueue_process(replyqueue_t *queue)
 
   tor_mutex_release(&queue->lock);
 }
+
+/** Return the number of threads configured for the given pool. */
+unsigned int
+threadpool_get_n_threads(threadpool_t *tp)
+{
+  tor_assert(tp);
+  return tp->n_threads;
+}
diff --git a/src/lib/evloop/workqueue.h b/src/lib/evloop/workqueue.h
index 50391759bf..134fe7434f 100644
--- a/src/lib/evloop/workqueue.h
+++ b/src/lib/evloop/workqueue.h
@@ -65,5 +65,6 @@ void replyqueue_process(replyqueue_t *queue);
 
 int threadpool_register_reply_event(threadpool_t *tp,
                                     void (*cb)(threadpool_t *tp));
+unsigned int threadpool_get_n_threads(threadpool_t *tp);
 
 #endif /* !defined(TOR_WORKQUEUE_H) */