apr_thread_pool: stop the threads before destroying the thread_pool's pool.

With APR <= 1.7 the threads' pools are children of the thread_pool's pool and
APR_POOL_DEBUG would deadlock if thread_pool_cleanup() is called while the
thread_pool's pool is destroyed (because of parent locking).

With APR > 1.7 the threads' pools are unmanaged so there is no such issue,
yet it does not hurt to stop the threads before.
 
* misc/apr_thread_pool.c(apr_thread_pool_destroy):
  Explicitely run the cleanup (which stops the threads) before destroying
  the thread_pool's pool.

Note: this is APR-util material only since APR trunk does not have the issue
      (per above).


git-svn-id: https://svn.apache.org/repos/asf/apr/apr-util/branches/1.7.x@1904743 13f79535-47bb-0310-9956-ffa450edef68

1 files changed, 8 insertions, 0 deletions

diff --git a/misc/apr_thread_pool.c b/misc/apr_thread_pool.c
index 6e36ea78..5aa3b65a 100644
--- a/misc/apr_thread_pool.c
+++ b/misc/apr_thread_pool.c
@@ -452,6 +452,14 @@ APU_DECLARE(apr_status_t) apr_thread_pool_create(apr_thread_pool_t ** me,
 
 APU_DECLARE(apr_status_t) apr_thread_pool_destroy(apr_thread_pool_t * me)
 {
+    /* Stop the threads before destroying me->pool: with APR <= 1.7 the
+     * threads' pools are children of me->pool and APR_POOL_DEBUG would
+     * deadlock if thread_pool_cleanup() is called while me->pool is
+     * destroyed (because of parent locking).
+     * With APR > 1.7 the threads' pools are unmanaged so there is no
+     * such issue, yet it does not hurt to stop the threads before.
+     */
+    apr_pool_cleanup_run(me->pool, me, thread_pool_cleanup);
     apr_pool_destroy(me->pool);
     return APR_SUCCESS;
 }