filedaemon: solve a race condition within the heartbeat thread shutdown

bareos · Aug 27, 2020 · a1c51d2 · a1c51d2
1 parent 2536984
commit a1c51d2
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 3 deletions.
diff --git a/core/src/filed/heartbeat.cc b/core/src/filed/heartbeat.cc
@@ -72,6 +72,8 @@ extern "C" void* sd_heartbeat_thread(void* arg)
   jcr->impl->hb_dir_bsock = dir;
   dir->suppress_error_msgs_ = true;
   sd->suppress_error_msgs_ = true;
+  jcr->impl->hb_initialized_once =
+      true;  // initialize last to avoid race condition
 
   /* Hang reading the socket to the SD, and every time we get
    * a heartbeat or we get a wait timeout (5 seconds), we
@@ -122,6 +124,7 @@ void StartHeartbeatMonitor(JobControlRecord* jcr)
   if (!no_signals) {
     jcr->impl->hb_bsock = NULL;
     jcr->impl->hb_started = false;
+    jcr->impl->hb_initialized_once = false;
     jcr->impl->hb_dir_bsock = NULL;
     pthread_create(&jcr->impl->heartbeat_id, NULL, sd_heartbeat_thread,
                    (void*)jcr);
@@ -133,8 +136,9 @@ void StopHeartbeatMonitor(JobControlRecord* jcr)
 {
   int cnt = 0;
   if (no_signals) { return; }
+
   /* Wait max 10 secs for heartbeat thread to start */
-  while (!jcr->impl->hb_started && cnt++ < 200) {
+  while (!jcr->impl->hb_initialized_once && cnt++ < 200) {
     Bmicrosleep(0, 50000); /* wait for start */
   }
 
@@ -175,6 +179,8 @@ void StopHeartbeatMonitor(JobControlRecord* jcr)
     // delete jcr->impl_->hb_dir_bsock;
     jcr->impl->hb_dir_bsock.reset();
   }
+
+  jcr->impl->hb_initialized_once = false;
 }
 
 /**
@@ -198,6 +204,8 @@ extern "C" void* dir_heartbeat_thread(void* arg)
   jcr->impl->hb_bsock.reset(dir);
   jcr->impl->hb_started = true;
   dir->suppress_error_msgs_ = true;
+  jcr->impl->hb_initialized_once =
+      true;  // initialize last to avoid race condition
 
   while (!dir->IsStop()) {
     time_t now, next;

diff --git a/core/src/filed/jcr_private.h b/core/src/filed/jcr_private.h
@@ -3,7 +3,7 @@
 
    Copyright (C) 2000-2012 Free Software Foundation Europe e.V.
    Copyright (C) 2011-2012 Planets Communications B.V.
-   Copyright (C) 2013-2019 Bareos GmbH & Co. KG
+   Copyright (C) 2013-2020 Bareos GmbH & Co. KG
 
    This program is Free Software; you can redistribute it and/or
    modify it under the terms of version three of the GNU Affero General Public
@@ -26,6 +26,8 @@
 
 #include "include/bareos.h"
 
+#include <atomic>
+
 struct AclData;
 struct XattrData;
 
@@ -68,7 +70,8 @@ struct JobControlRecordPrivate {
   uint32_t StartBlock{};
   uint32_t EndBlock{};
   pthread_t heartbeat_id{};       /**< Id of heartbeat thread */
-  volatile bool hb_started{};     /**< Heartbeat running */
+  std::atomic<bool> hb_initialized_once{};    /**< Heartbeat initialized */
+  std::atomic<bool> hb_started{};             /**< Heartbeat running */
   std::shared_ptr<BareosSocket> hb_bsock;     /**< Duped SD socket */
   std::shared_ptr<BareosSocket> hb_dir_bsock; /**< Duped DIR socket */
   alist* RunScripts{};            /**< Commands to run before and after job */