fa94735b84
Resolves: RHEL-73774 Signed-off-by: Andrew Cagney <cagney@gnu.org> Signed-off-by: Daiki Ueno <dueno@redhat.com>
98 lines
3.4 KiB
Diff
98 lines
3.4 KiB
Diff
From eb31dda36d082e33749bc294d56ec493a094336d Mon Sep 17 00:00:00 2001
|
|
From: Andrew Cagney <cagney@gnu.org>
|
|
Date: Sun, 16 Nov 2025 10:33:53 -0500
|
|
Subject: [PATCH] helpers: work around NSS by joining helper threads
|
|
|
|
Much of the analysis and testing by Ondrej Moris
|
|
|
|
NSS attaches stuff to the threads onexit queue that must
|
|
be run before the main thread exits (if it doesn't things
|
|
explode during shutdown).
|
|
|
|
See: Race condition in helper_thread_stopped_callback() #2461
|
|
See: PR_Cleanup() doesn't wait for pthread_create() threads
|
|
https://bugzilla.mozilla.org/show_bug.cgi?id=1992272
|
|
---
|
|
programs/pluto/server_pool.c | 42 +++++++++++++++++++++++++++++++-----
|
|
1 file changed, 37 insertions(+), 5 deletions(-)
|
|
|
|
diff --git a/programs/pluto/server_pool.c b/programs/pluto/server_pool.c
|
|
index 056a007301..5c7dc5e18d 100644
|
|
--- a/programs/pluto/server_pool.c
|
|
+++ b/programs/pluto/server_pool.c
|
|
@@ -275,6 +275,22 @@ static void *helper_thread(void *arg)
|
|
dbg("helper %u: telling main thread that it is exiting", w->helper_id);
|
|
schedule_callback("helper stopped", deltatime(0), SOS_NOBODY,
|
|
helper_thread_stopped_callback, NULL);
|
|
+ /*
|
|
+ * Danger. This isn't the end.
|
|
+ *
|
|
+ * NSS still has stuff in thread-exit handlers to execute and
|
|
+ * there's no clean way of forcing its execution (and if it
|
|
+ * isn't allowed to run NSS crashes!). Hence, the main thread
|
|
+ * will need to wait for this thread to exit.
|
|
+ *
|
|
+ * But wait, there's more. The main thread also needs to keep
|
|
+ * the event loop running while these threads are exiting so
|
|
+ * ptread_join() needs to be called with care.
|
|
+ *
|
|
+ * See: Race condition in helper_thread_stopped_callback() #2461
|
|
+ * See: PR_Cleanup() doesn't wait for pthread_create() threads
|
|
+ * https://bugzilla.mozilla.org/show_bug.cgi?id=1992272
|
|
+ */
|
|
return NULL;
|
|
}
|
|
|
|
@@ -589,10 +605,6 @@ void start_server_helpers(uintmax_t nhelpers, struct logger *logger)
|
|
|
|
/*
|
|
* Repeatedly nudge the helper threads until they all exit.
|
|
- *
|
|
- * Note that pthread_join() doesn't work here: an any-thread join may
|
|
- * end up joining an unrelated thread (for instance the CRL helper);
|
|
- * and a specific thread join may block waiting for the wrong thread.
|
|
*/
|
|
|
|
static void (*server_helpers_stopped_callback)(void);
|
|
@@ -605,6 +617,17 @@ static void helper_thread_stopped_callback(const char *story UNUSED,
|
|
dbg("one helper thread exited, %u remaining",
|
|
helper_threads_started-helper_threads_stopped);
|
|
|
|
+ /*
|
|
+ * Danger:
|
|
+ *
|
|
+ * Delay joining W.pid until all helper threads have exited.
|
|
+ * This way the event-loop is kept running.
|
|
+ *
|
|
+ * Even though W is on the exit path it still needs to execute
|
|
+ * NSS's thread exit code - who knows what that is doing and
|
|
+ * how long it will take -
|
|
+ */
|
|
+
|
|
/* wait for more? */
|
|
if (helper_threads_started > helper_threads_stopped) {
|
|
/* poke threads waiting for work */
|
|
@@ -612,9 +635,18 @@ static void helper_thread_stopped_callback(const char *story UNUSED,
|
|
return;
|
|
}
|
|
|
|
- /* all done; cleanup */
|
|
+ /*
|
|
+ * All done; cleanup
|
|
+ *
|
|
+ * All helper threads are on the exit war-path so, hopefully,
|
|
+ * this join will not block (but no telling what NSS did).
|
|
+ */
|
|
for (unsigned h = 0; h < helper_threads_started; h++) {
|
|
struct helper_thread *w = &helper_threads[h];
|
|
+ int e = pthread_join(w->pid, NULL);
|
|
+ if (e != 0) {
|
|
+ llog_errno(RC_LOG, w->logger, e, "WARNING: pthread_join() failed, ");
|
|
+ }
|
|
free_logger(&w->logger, HERE);
|
|
}
|
|
|
|
--
|
|
2.52.0
|
|
|