Merge pull request #12985 from FRRouting/mergify/bp/stable/8.3/pr-11913

bgpd: release rcu lock in bgp keepalive pthread (backport #11913)

1 files changed, 9 insertions, 0 deletions

diff --git a/bgpd/bgp_keepalives.c b/bgpd/bgp_keepalives.c
index 86202a0e3d..d0eb249d38 100644
--- a/bgpd/bgp_keepalives.c
+++ b/bgpd/bgp_keepalives.c
@@ -175,6 +175,15 @@ void *bgp_keepalives_start(void *arg)
 	struct timeval next_update = {0, 0};
 	struct timespec next_update_ts = {0, 0};
 
+	/*
+	 * The RCU mechanism for each pthread is initialized in a "locked"
+	 * state. That's ok for pthreads using the frr_pthread,
+	 * thread_fetch event loop, because that event loop unlocks regularly.
+	 * For foreign pthreads, the lock needs to be unlocked so that the
+	 * background rcu pthread can run.
+	 */
+	rcu_read_unlock();
+
 	peerhash_mtx = XCALLOC(MTYPE_TMP, sizeof(pthread_mutex_t));
 	peerhash_cond = XCALLOC(MTYPE_TMP, sizeof(pthread_cond_t));