Attempt to fix bat seg faults

[bacula/bacula] / bacula / src / lib / signal.c

diff --git a/bacula/src/lib/signal.c b/bacula/src/lib/signal.c
index 7840af690118c068216ca69788ca493b3886f9fa..16be3856e5ee9dbb15d8bfcaec5d0659119a1ecb 100644 (file)
--- a/bacula/src/lib/signal.c
+++ b/bacula/src/lib/signal.c
@@ -72,6 +72,45 @@ const char *get_signal_name(int sig)
    }
 }
 
+/* defined in jcr.c */
+extern void _dbg_print_jcr(FILE *fp);
+/* defined in plugin.c */
+extern void _dbg_print_plugin(FILE *fp);
+/* defined in lockmgr.c */
+extern void dbg_print_lock(FILE *fp);
+
+/*
+ * !!! WARNING !!! 
+ *
+ * This function should be used ONLY after a violent signal. We walk through the
+ * JCR chain without doing any lock, bacula should not be running.
+ */
+static void dbg_print_bacula()
+{
+   char buf[512];
+
+   snprintf(buf, sizeof(buf), "%s/%s.%d.bactrace", 
+            working_directory, my_name, getpid());
+   FILE *fp = fopen(buf, "ab") ;
+   if (!fp) {
+      fp = stderr;
+   }
+   
+   fprintf(stderr, "Dumping: %s\n", buf);
+
+   /* Print also B_DB and RWLOCK structure 
+    * Can add more info about JCR with dbg_jcr_add_hook()
+    */
+   _dbg_print_jcr(fp);
+
+   _dbg_print_plugin(fp);
+   dbg_print_lock(fp);
+
+   if (fp != stderr) {
+      fclose(fp);
+   }
+}
+
 /*
  * Handle signals here
  */
@@ -92,7 +131,10 @@ extern "C" void signal_handler(int sig)
    if (sig == SIGTERM) {
 //    Emsg1(M_TERM, -1, "Shutting down Bacula service: %s ...\n", my_name);
    } else {
-      Emsg2(M_FATAL, -1, _("Bacula interrupted by signal %d: %s\n"), sig, get_signal_name(sig));
+/* ***FIXME*** Display a message without taking any lock in the system
+ *    Emsg2(M_FATAL, -1, _("Bacula interrupted by signal %d: %s\n"), sig, get_signal_name(sig));
+ */
+      fprintf(stderr, _("Bacula interrupted by signal %d: %s\n"), sig, get_signal_name(sig));
    }
 
 #ifdef TRACEBACK
@@ -169,9 +211,8 @@ extern "C" void signal_handler(int sig)
          waitpid(pid, NULL, 0);       /* wait for child to produce dump */
          Dmsg0(500, "Done waitpid\n");
          fprintf(stderr, _("Traceback complete, attempting cleanup ...\n"));
-         /* print information about the current state into stderr */
-         print_lock_dbg();
-         print_jcr_dbg();
+         /* print information about the current state into working/<file>.bactrace */
+         dbg_print_bacula();
          exit_handler(sig);           /* clean up if possible */
          Dmsg0(500, "Done exit_handler\n");
       } else {
@@ -179,9 +220,9 @@ extern "C" void signal_handler(int sig)
          bmicrosleep(30, 0);
       }
       fprintf(stderr, _("It looks like the traceback worked ...\n"));
+      dbg_print_bacula();
    }
 #endif
-
    exit_handler(sig);
 }
 
@@ -192,6 +233,7 @@ extern "C" void signal_handler(int sig)
 void init_stack_dump(void)
 {
    main_pid = getpid();               /* save main thread's pid */
+   lmgr_init_thread();                /* initialize the lockmanager stack */
 }
 
 /*