Blob Blame Raw
From 666fdac51b94450391e8fec8d16db34db09502ae Mon Sep 17 00:00:00 2001
From: Ludwig Krispenz <lkrispen@redhat.com>
Date: Wed, 13 Jan 2016 13:15:53 +0100
Subject: [PATCH] Ticket: 48341 - deadlock on connection mutex

If  thread is blocked in connection_read_operation() it holds the connection mutex
and the main thread iterating through the connection table is also blocked.

But if the main thread would get the mutex it would just detect that the connection has still the
C_gettingber flag set and immediately release the lock.

The check if c_gettingber == 0 can be done without holding the mutex and so the deadlock
can be avoided

Reviewed by Rich, Thanks

(cherry picked from commit a1635fc45f681ed9066f6beed9be7e1672490f9f)
---
 ldap/servers/slapd/daemon.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/ldap/servers/slapd/daemon.c b/ldap/servers/slapd/daemon.c
index 355f0fc..d25c44d 100644
--- a/ldap/servers/slapd/daemon.c
+++ b/ldap/servers/slapd/daemon.c
@@ -1740,6 +1740,9 @@ handle_pr_read_ready(Connection_Table *ct, PRIntn num_poll)
 	{
 		if ( c->c_mutex != NULL )
 		{
+			/* this check can be done without acquiring the mutex */
+			if (c->c_gettingber) continue;
+
 			PR_EnterMonitor(c->c_mutex);
 			if ( connection_is_active_nolock (c) && c->c_gettingber == 0 )
 			{
-- 
2.4.3