From: Kern Sibbald <kern@sibbald.com>
Date: Tue, 19 May 2009 16:38:37 +0000 (+0000)
Subject: During jcr destruction hold jcr_chaing lock only for minimum
X-Git-Tag: Release-3.0.2~225
X-Git-Url: https://git.sur5r.net/?a=commitdiff_plain;h=ffcf8951cd956a49645ae2845f9b0397b84e6f02;p=bacula%2Fbacula

During jcr destruction hold jcr_chaing lock only for minimum
     time necessary.  This should fix the SD deadlock in bug #1287.

git-svn-id: https://bacula.svn.sourceforge.net/svnroot/bacula/trunk@8841 91ce42f0-d328-0410-95d8-f526ca767f89
---

diff --git a/bacula/patches/3.0.1-sd-hang.patch b/bacula/patches/3.0.1-sd-hang.patch
new file mode 100644
index 0000000000..74029c64ae
--- /dev/null
+++ b/bacula/patches/3.0.1-sd-hang.patch
@@ -0,0 +1,36 @@
+
+ This patch can be applied to version 3.0.1 and fixes a 
+ a race condition that causes the SD to hang. This should
+ fix bug #1287.
+
+ Apply it to version 3.0.1 with:
+
+ cd <bacula-source>
+ patch -p0 <3.0.1-sd-hang.patch
+ ./configure <your-options>
+ make
+ ...
+ make install
+
+
+
+Index: src/lib/jcr.c
+===================================================================
+--- src/lib/jcr.c	(revision 8840)
++++ src/lib/jcr.c	(working copy)
+@@ -518,6 +518,7 @@
+             jcr->JobId, jcr->use_count(), jcr->Job);
+    }
+    remove_jcr(jcr);                   /* remove Jcr from chain */
++   unlock_jcr_chain();
+ 
+    dequeue_messages(jcr);
+    job_end_pop(jcr);                  /* pop and call hooked routines */
+@@ -571,7 +572,6 @@
+       jcr->daemon_free_jcr(jcr);      /* call daemon free routine */
+    }
+ 
+-   unlock_jcr_chain();
+    free_common_jcr(jcr);
+    close_msg(NULL);                   /* flush any daemon messages */
+    garbage_collect_memory_pool();
diff --git a/bacula/src/lib/jcr.c b/bacula/src/lib/jcr.c
index 32b6aabd3b..4da6de29da 100644
--- a/bacula/src/lib/jcr.c
+++ b/bacula/src/lib/jcr.c
@@ -518,6 +518,7 @@ void free_jcr(JCR *jcr)
             jcr->JobId, jcr->use_count(), jcr->Job);
    }
    remove_jcr(jcr);                   /* remove Jcr from chain */
+   unlock_jcr_chain();
 
    dequeue_messages(jcr);
    job_end_pop(jcr);                  /* pop and call hooked routines */
@@ -571,7 +572,6 @@ void free_jcr(JCR *jcr)
       jcr->daemon_free_jcr(jcr);      /* call daemon free routine */
    }
 
-   unlock_jcr_chain();
    free_common_jcr(jcr);
    close_msg(NULL);                   /* flush any daemon messages */
    garbage_collect_memory_pool();
diff --git a/bacula/technotes b/bacula/technotes
index 5f793fc026..9d40a64e8b 100644
--- a/bacula/technotes
+++ b/bacula/technotes
@@ -2,6 +2,9 @@
           
 General:
 
+19May09
+kes  During jcr destruction hold jcr_chaing lock only for minimum
+     time necessary.  This should fix the SD deadlock in bug #1287.
 18May09
 kes  Simplify messages printed by SD when reserve fails. This
      should fix bug #1285.