Fix: call_rcu: teardown default call_rcu worker on application exit

Teardown the default call_rcu worker thread if there are no queued
callbacks on process exit. This prevents leaking memory.

Here is how an application can ensure graceful teardown of this
worker thread:

- An application queuing call_rcu callbacks should invoke
  rcu_barrier() before it exits.
- When chaining call_rcu callbacks, the number of calls to
  rcu_barrier() on application exit must match at least the maximum
  number of chained callbacks.
- If an application chains callbacks endlessly, it would have to be
  modified to stop chaining callbacks when it detects an application
  exit (e.g. with a flag), and wait for quiescence with rcu_barrier()
  after setting that flag.
- The statements above apply to a library which queues call_rcu
  callbacks, only it needs to invoke rcu_barrier in its library
  destructor.

Fixes: #1317
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I40556bc872d3df58a22fb88a0dbb528ce5c9b4af

diff --git a/src/urcu-bp.c b/src/urcu-bp.c
index 0d6ff4b0071965a6897b5904c86ce1851ecb1e6b..5281d5e852b053eb2fc4ad19d562b84ae04ddb56 100644 (file)
--- a/src/urcu-bp.c
+++ b/src/urcu-bp.c
@@ -121,6 +121,7 @@ static
 void __attribute__((constructor)) _urcu_bp_init(void);
 static
 void __attribute__((destructor)) urcu_bp_exit(void);
+static void urcu_call_rcu_exit(void);
 
 #ifndef CONFIG_RCU_FORCE_SYS_MEMBARRIER
 int urcu_bp_has_sys_membarrier;
@@ -655,6 +656,8 @@ void _urcu_bp_init(void)
 static
 void urcu_bp_exit(void)
 {
+       urcu_call_rcu_exit();
+
        mutex_lock(&init_lock);
        if (!--urcu_bp_refcount) {
                struct registry_chunk *chunk, *tmp;

diff --git a/src/urcu-call-rcu-impl.h b/src/urcu-call-rcu-impl.h
index c01f02b2f5c7400edf8c9ed191c87a9cc5b9d52c..187727e4137083d975fb3268c7c9106f4bcb7417 100644 (file)
--- a/src/urcu-call-rcu-impl.h
+++ b/src/urcu-call-rcu-impl.h
@@ -449,8 +449,8 @@ static void call_rcu_data_init(struct call_rcu_data **crdpp,
        cds_list_add(&crdp->list, &call_rcu_data_list);
        crdp->cpu_affinity = cpu_affinity;
        crdp->gp_count = 0;
-       cmm_smp_mb();  /* Structure initialized before pointer is planted. */
-       *crdpp = crdp;
+       rcu_set_pointer(crdpp, crdp);
+
        ret = pthread_create(&crdp->tid, NULL, call_rcu_thread, crdp);
        if (ret)
                urcu_die(ret);
@@ -565,22 +565,27 @@ int set_cpu_call_rcu_data(int cpu, struct call_rcu_data *crdp)
 
 /*
  * Return a pointer to the default call_rcu_data structure, creating
- * one if need be.  Because we never free call_rcu_data structures,
- * we don't need to be in an RCU read-side critical section.
+ * one if need be.
+ *
+ * The call to this function with intent to use the returned
+ * call_rcu_data should be protected by RCU read-side lock.
  */
 
 struct call_rcu_data *get_default_call_rcu_data(void)
 {
-       if (default_call_rcu_data != NULL)
-               return rcu_dereference(default_call_rcu_data);
+       struct call_rcu_data *crdp;
+
+       crdp = rcu_dereference(default_call_rcu_data);
+       if (crdp != NULL)
+               return crdp;
+
        call_rcu_lock(&call_rcu_mutex);
-       if (default_call_rcu_data != NULL) {
-               call_rcu_unlock(&call_rcu_mutex);
-               return default_call_rcu_data;
-       }
-       call_rcu_data_init(&default_call_rcu_data, 0, -1);
+       if (default_call_rcu_data == NULL)
+               call_rcu_data_init(&default_call_rcu_data, 0, -1);
+       crdp = default_call_rcu_data;
        call_rcu_unlock(&call_rcu_mutex);
-       return default_call_rcu_data;
+
+       return crdp;
 }
 
 /*
@@ -1055,3 +1060,58 @@ void urcu_unregister_rculfhash_atfork(struct urcu_atfork *atfork __attribute__((
 {
        urcu_die(EPERM);
 }
+
+/*
+ * Teardown the default call_rcu worker thread if there are no queued
+ * callbacks on process exit. This prevents leaking memory.
+ *
+ * Here is how an application can ensure graceful teardown of this
+ * worker thread:
+ *
+ * - An application queuing call_rcu callbacks should invoke
+ *   rcu_barrier() before it exits.
+ * - When chaining call_rcu callbacks, the number of calls to
+ *   rcu_barrier() on application exit must match at least the maximum
+ *   number of chained callbacks.
+ * - If an application chains callbacks endlessly, it would have to be
+ *   modified to stop chaining callbacks when it detects an application
+ *   exit (e.g. with a flag), and wait for quiescence with rcu_barrier()
+ *   after setting that flag.
+ * - The statements above apply to a library which queues call_rcu
+ *   callbacks, only it needs to invoke rcu_barrier in its library
+ *   destructor.
+ *
+ * Note that this function does not presume it is being called when the
+ * application is single-threaded even though this is invoked from a
+ * destructor: this function synchronizes against concurrent calls to
+ * get_default_call_rcu_data().
+ */
+static void urcu_call_rcu_exit(void)
+{
+       struct call_rcu_data *crdp;
+       bool teardown = true;
+
+       if (default_call_rcu_data == NULL)
+               return;
+       call_rcu_lock(&call_rcu_mutex);
+       /*
+        * If the application leaves callbacks in the default call_rcu
+        * worker queue, keep the default worker in place.
+        */
+       crdp = default_call_rcu_data;
+       if (!crdp) {
+               teardown = false;
+               goto unlock;
+       }
+       if (!cds_wfcq_empty(&crdp->cbs_head, &crdp->cbs_tail)) {
+               teardown = false;
+               goto unlock;
+       }
+       rcu_set_pointer(&default_call_rcu_data, NULL);
+unlock:
+       call_rcu_unlock(&call_rcu_mutex);
+       if (teardown) {
+               synchronize_rcu();
+               call_rcu_data_free(crdp);
+       }
+}

diff --git a/src/urcu-qsbr.c b/src/urcu-qsbr.c
index 72d49c410199ee1499dcadd0b6db5e482188b8b2..8e5b3cd95a8b97c20a7f3539e063d663adb0ebc5 100644 (file)
--- a/src/urcu-qsbr.c
+++ b/src/urcu-qsbr.c
@@ -53,6 +53,7 @@
 #define _LGPL_SOURCE
 
 void __attribute__((destructor)) urcu_qsbr_exit(void);
+static void urcu_call_rcu_exit(void);
 
 /*
  * rcu_gp_lock ensures mutual exclusion between threads calling
@@ -509,6 +510,7 @@ void urcu_qsbr_exit(void)
         * readers, and left running at exit.
         * assert(cds_list_empty(&registry));
         */
+       urcu_call_rcu_exit();
 }
 
 DEFINE_RCU_FLAVOR(rcu_flavor);

diff --git a/src/urcu.c b/src/urcu.c
index f5a32317813a1290139e77c397d7d19db73b46c4..fd81a783c76c80ea551dec97aa3e1953c7279162 100644 (file)
--- a/src/urcu.c
+++ b/src/urcu.c
@@ -109,9 +109,11 @@ void rcu_init(void)
 static int init_done;
 
 void __attribute__((constructor)) rcu_init(void);
-void __attribute__((destructor)) rcu_exit(void);
 #endif
 
+void __attribute__((destructor)) rcu_exit(void);
+static void urcu_call_rcu_exit(void);
+
 /*
  * rcu_gp_lock ensures mutual exclusion between threads calling
  * synchronize_rcu().
@@ -648,20 +650,22 @@ void rcu_init(void)
                urcu_die(errno);
 }
 
+/*
+ * Don't unregister the SIGRCU signal handler anymore, because
+ * call_rcu threads could still be using it shortly before the
+ * application exits.
+ * Assertion disabled because call_rcu threads are now rcu
+ * readers, and left running at exit.
+ * assert(cds_list_empty(&registry));
+ */
+
+#endif /* #ifdef RCU_SIGNAL */
+
 void rcu_exit(void)
 {
-       /*
-        * Don't unregister the SIGRCU signal handler anymore, because
-        * call_rcu threads could still be using it shortly before the
-        * application exits.
-        * Assertion disabled because call_rcu threads are now rcu
-        * readers, and left running at exit.
-        * assert(cds_list_empty(&registry));
-        */
+       urcu_call_rcu_exit();
 }
 
-#endif /* #ifdef RCU_SIGNAL */
-
 DEFINE_RCU_FLAVOR(rcu_flavor);
 
 #include "urcu-call-rcu-impl.h"