I finally found and squashed the elusive SD crash.

[bacula/bacula] / bacula / src / stored / stored.c

diff --git a/bacula/src/stored/stored.c b/bacula/src/stored/stored.c
index 2695fa70efc794018eba28a52bf1df51f32308b9..b91718ed486885119280e360242e83625754ac9d 100644 (file)
--- a/bacula/src/stored/stored.c
+++ b/bacula/src/stored/stored.c
@@ -14,7 +14,7 @@
 
    This program is free software; you can redistribute it and/or
    modify it under the terms of the GNU General Public License
-   version 2 as ammended with additional clauses defined in the
+   version 2 as amended with additional clauses defined in the
    file LICENSE in the main source directory.
 
    This program is distributed in the hope that it will be useful,
@@ -50,6 +50,7 @@ pthread_cond_t wait_device_release = PTHREAD_COND_INITIALIZER;
 static uint32_t VolSessionId = 0;
 uint32_t VolSessionTime;
 char *configfile = NULL;
+bool init_done = false;
 
 /* Global static variables */
 static int foreground = 0;
@@ -220,24 +221,14 @@ int main (int argc, char *argv[])
     /*
      * Start the device allocation thread
      */
+   create_volume_list();              /* do before device_init */
    if (pthread_create(&thid, NULL, device_initialization, NULL) != 0) {
       Emsg1(M_ABORT, 0, _("Unable to create thread. ERR=%s\n"), strerror(errno));
    }
 
    start_watchdog();                  /* start watchdog thread */
-
    init_jcr_subsystem();              /* start JCR watchdogs etc. */
 
-   /*
-    * Sleep a bit to give device thread a chance to lock the resource
-    * chain before we start the server.
-    */
-   bmicrosleep(1, 0);
-
-   /* Wait for device initialization to complete */
-   LockRes();
-   UnlockRes();
-
    /* Single server used for Director and File daemon */
    bnet_thread_server(me->sdaddrs, me->max_concurrent_jobs * 2 + 1,
                       &dird_workq, handle_connection_request);
@@ -262,7 +253,6 @@ static int check_resources()
    bool OK = true;
    AUTOCHANGER *changer;
 
-// LockRes();
 
    me = (STORES *)GetNextRes(R_STORAGE, NULL);
    if (!me) {
@@ -302,13 +292,18 @@ static int check_resources()
       OK = false;
    }
 
-#ifdef HAVE_TLS
    DIRRES *director;
    STORES *store;
    foreach_res(store, R_STORAGE) { 
       /* tls_require implies tls_enable */
       if (store->tls_require) {
-         store->tls_enable = true;
+         if (have_tls) {
+            store->tls_enable = true;
+         } else {
+            Jmsg(NULL, M_FATAL, 0, _("TLS required but not configured in Bacula.\n"));
+            OK = false;
+            continue;
+         }
       }
 
       if (!store->tls_certfile && store->tls_enable) {
@@ -394,7 +389,6 @@ static int check_resources()
          }
       }
    }
-#endif /* HAVE_TLS */
 
    /* Ensure that the media_type for each device is the same */
    foreach_res(changer, R_AUTOCHANGER) {
@@ -456,48 +450,57 @@ extern "C"
 void *device_initialization(void *arg)
 {
    DEVRES *device;
+   DCR *dcr;
+   JCR *jcr;
+   DEVICE *dev;
 
    LockRes();
+
    pthread_detach(pthread_self());
+   jcr = new_jcr(sizeof(JCR), stored_free_jcr);
+   jcr->JobType = JT_SYSTEM;
+   /* Initialize FD start condition variable */
+   int errstat = pthread_cond_init(&jcr->job_start_wait, NULL);
+   if (errstat != 0) {
+      Jmsg1(jcr, M_ABORT, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
+   }
 
    foreach_res(device, R_DEVICE) {
       Dmsg1(90, "calling init_dev %s\n", device->device_name);
-      device->dev = init_dev(NULL, NULL, device);
+      device->dev = dev = init_dev(NULL, device);
       Dmsg1(10, "SD init done %s\n", device->device_name);
-      if (!device->dev) {
+      if (!dev) {
          Jmsg1(NULL, M_ERROR, 0, _("Could not initialize %s\n"), device->device_name);
          continue;
       }
 
+      dcr = new_dcr(jcr, dev);
+
       if (device->cap_bits & CAP_ALWAYSOPEN) {
-         Dmsg1(20, "calling first_open_device %s\n", device->device_name);
-         if (!first_open_device(device->dev)) {
-            Jmsg1(NULL, M_ERROR, 0, _("Could not open device %s\n"), device->device_name);
+         Dmsg1(20, "calling first_open_device %s\n", dev->print_name());
+         if (!first_open_device(dcr)) {
+            Jmsg1(NULL, M_ERROR, 0, _("Could not open device %s\n"), dev->print_name());
+            Dmsg1(20, "Could not open device %s\n", dev->print_name());
+            term_dev(dev);
+            device->dev = NULL;
+            free_dcr(dcr);
+            continue;
          }
       }
-      if (device->cap_bits & CAP_AUTOMOUNT && device->dev &&
-          device->dev->is_open()) {
-         JCR *jcr;
-         DCR *dcr;
-         jcr = new_jcr(sizeof(JCR), stored_free_jcr);
-         jcr->JobType = JT_SYSTEM;
-         /* Initialize FD start condition variable */
-         int errstat = pthread_cond_init(&jcr->job_start_wait, NULL);
-         if (errstat != 0) {
-            Jmsg1(jcr, M_ABORT, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
-         }
-         dcr = new_dcr(jcr, device->dev);
+      if (device->cap_bits & CAP_AUTOMOUNT && dev->is_open()) {
          switch (read_dev_volume_label(dcr)) {
          case VOL_OK:
-            memcpy(&dcr->dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dcr->dev->VolCatInfo));
+            memcpy(&dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dev->VolCatInfo));
             break;
          default:
-            Jmsg1(NULL, M_WARNING, 0, _("Could not mount device %s\n"), device->device_name);
+            Jmsg1(NULL, M_WARNING, 0, _("Could not mount device %s\n"), dev->print_name());
             break;
          }
-         free_jcr(jcr);
       }
+      free_dcr(dcr);
    }
+   free_jcr(jcr); 
+   init_done = true;
    UnlockRes();
    return NULL;
 }
@@ -552,13 +555,19 @@ void terminate_stored(int sig)
    Dmsg1(200, "In terminate_stored() sig=%d\n", sig);
 
    foreach_res(device, R_DEVICE) {
+      Dmsg1(10, "Term device %s\n", device->device_name);
       if (device->dev) {
+         free_volume(device->dev);
          term_dev(device->dev);
+      } else {
+         Dmsg1(10, "No dev structure %s\n", device->device_name);
       }
    }
 
-   if (configfile)
-   free(configfile);
+   if (configfile) {
+      free(configfile);
+      configfile = NULL;
+   }
    free_config_resources();
 
    if (debug_level > 10) {
@@ -567,6 +576,7 @@ void terminate_stored(int sig)
    term_msg();
    stop_watchdog();
    cleanup_tls();
+   free_volume_list();
    close_memory_pool();
 
    sm_dump(false);                    /* dump orphaned buffers */