]> git.sur5r.net Git - bacula/bacula/blobdiff - bacula/src/stored/job.c
This commit was manufactured by cvs2svn to create tag
[bacula/bacula] / bacula / src / stored / job.c
index c462f583485096f49a636b26b119b4f4ad326bb0..6c54e6d835bff105fada120bd7678bcd6df2089c 100644 (file)
@@ -5,7 +5,7 @@
  *
  */
 /*
-   Copyright (C) 2000, 2001, 2002 Kern Sibbald and John Walker
+   Copyright (C) 2000-2003 Kern Sibbald and John Walker
 
    This program is free software; you can redistribute it and/or
    modify it under the terms of the GNU General Public License as
@@ -38,15 +38,15 @@ static int use_device_cmd(JCR *jcr);
 
 /* Requests from the Director daemon */
 static char jobcmd[]     = "JobId=%d job=%127s job_name=%127s client_name=%127s \
-type=%d level=%d FileSet=%127s Allow=";
+type=%d level=%d FileSet=%127s NoAttr=%d SpoolAttr=%d FileSetMD5=%127s\n";
 static char use_device[] = "use device=%s media_type=%s pool_name=%s pool_type=%s\n";
 
 /* Responses sent to Director daemon */
 static char OKjob[]     = "3000 OK Job SDid=%u SDtime=%u Authorization=%s\n";
 static char OK_device[] = "3000 OK use device\n";
-static char NO_device[] = "3904 Device %s not available\n";
-static char BAD_use[]   = "3903 Bad use command: %s\n";
-static char BAD_job[]   = "3905 Bad Job command: %s\n";
+static char NO_device[] = "3914 Device \"%s\" not in SD Device resources.\n";
+static char BAD_use[]   = "3913 Bad use command: %s\n";
+static char BAD_job[]   = "3915 Bad Job command: %s\n";
 
 
 
@@ -65,56 +65,76 @@ int job_cmd(JCR *jcr)
    int JobId, errstat;
    char auth_key[100];
    BSOCK *dir = jcr->dir_bsock;
-   POOLMEM *job_name, *client_name, *job, *fileset_name;
-   int JobType, level;
+   POOLMEM *job_name, *client_name, *job, *fileset_name, *fileset_md5;
+   int JobType, level, spool_attributes, no_attributes;
    struct timeval tv;
    struct timezone tz;
    struct timespec timeout;
+   JCR *ojcr;
 
    /*
     * Get JobId and permissions from Director
     */
 
-   Dmsg1(30, "Job_cmd: %s\n", dir->msg);
+   Dmsg1(130, "Job_cmd: %s\n", dir->msg);
    job = get_memory(dir->msglen);
    job_name = get_memory(dir->msglen);
    client_name = get_memory(dir->msglen);
    fileset_name = get_memory(dir->msglen);
+   fileset_md5 = get_memory(dir->msglen);
    if (sscanf(dir->msg, jobcmd, &JobId, job, job_name, client_name,
-             &JobType, &level, fileset_name) != 7) {
-      bnet_fsend(dir, BAD_job, dir->msg);
-      Emsg1(M_FATAL, 0, _("Bad Job Command from Director: %s\n"), dir->msg);
+             &JobType, &level, fileset_name, &no_attributes,
+             &spool_attributes, fileset_md5) != 10) {
+      pm_strcpy(&jcr->errmsg, dir->msg);
+      bnet_fsend(dir, BAD_job, jcr->errmsg);
+      Emsg1(M_FATAL, 0, _("Bad Job Command from Director: %s\n"), jcr->errmsg);
       free_memory(job);
       free_memory(job_name);
       free_memory(client_name);
       free_memory(fileset_name);
-      jcr->JobStatus = JS_ErrorTerminated;
+      free_memory(fileset_md5);
+      set_jcr_job_status(jcr, JS_ErrorTerminated);
       return 0;
    }
+   /*        
+    * Since this job could be rescheduled, we
+    *  check to see if we have it already. If so
+    *  free the old jcr and use the new one.
+    */
+   ojcr = get_jcr_by_full_name(job);
+   if (ojcr && !ojcr->authenticated) {
+      Dmsg2(100, "Found ojcr=0x%x Job %s\n", (unsigned)ojcr, job);
+      free_jcr(ojcr);
+   }
    jcr->JobId = JobId;
    jcr->VolSessionId = newVolSessionId();
    jcr->VolSessionTime = VolSessionTime;
-   strcpy(jcr->Job, job);
+   bstrncpy(jcr->Job, job, sizeof(jcr->Job));
    unbash_spaces(job_name);
-   jcr->job_name = get_memory(strlen(job_name) + 1);
-   strcpy(jcr->job_name, job_name);
+   jcr->job_name = get_pool_memory(PM_NAME);
+   pm_strcpy(&jcr->job_name, job_name);
    unbash_spaces(client_name);
-   jcr->client_name = get_memory(strlen(client_name) + 1);
-   strcpy(jcr->client_name, client_name);
+   jcr->client_name = get_pool_memory(PM_NAME);
+   pm_strcpy(&jcr->client_name, client_name);
    unbash_spaces(fileset_name);
-   jcr->fileset_name = get_memory(strlen(fileset_name) + 1);
-   strcpy(jcr->fileset_name, fileset_name);
+   jcr->fileset_name = get_pool_memory(PM_NAME);
+   pm_strcpy(&jcr->fileset_name, fileset_name);
    jcr->JobType = JobType;
    jcr->JobLevel = level;
+   jcr->no_attributes = no_attributes;
+   jcr->spool_attributes = spool_attributes;
+   jcr->fileset_md5 = get_pool_memory(PM_NAME);
+   pm_strcpy(&jcr->fileset_md5, fileset_md5);
    free_memory(job);
    free_memory(job_name);
    free_memory(client_name);
    free_memory(fileset_name);
+   free_memory(fileset_md5);
 
    /* Initialize FD start condition variable */
    if ((errstat = pthread_cond_init(&jcr->job_start_wait, NULL)) != 0) {
-      Emsg1(M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
-      jcr->JobStatus = JS_ErrorTerminated;
+      Jmsg1(jcr, M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
+      set_jcr_job_status(jcr, JS_ErrorTerminated);
       return 0;
    }
    jcr->authenticated = FALSE;
@@ -122,22 +142,22 @@ int job_cmd(JCR *jcr)
    /*
     * Pass back an authorization key for the File daemon
     */
-   gettimeofday(&tv, &tz);
-   srandom(tv.tv_usec + tv.tv_sec);
-   sprintf(auth_key, "%ld", (long)random());
+   make_session_key(auth_key, NULL, 1);
    bnet_fsend(dir, OKjob, jcr->VolSessionId, jcr->VolSessionTime, auth_key);
-   Dmsg1(10, ">dird: %s", dir->msg);
+   Dmsg1(110, ">dird: %s", dir->msg);
    jcr->sd_auth_key = bstrdup(auth_key);
+   memset(auth_key, 0, sizeof(auth_key));    
 
    /*
     * Wait for the device, media, and pool information
     */
    if (!use_device_cmd(jcr)) {
-      jcr->JobStatus = JS_ErrorTerminated;
+      set_jcr_job_status(jcr, JS_ErrorTerminated);
+      memset(jcr->sd_auth_key, 0, strlen(jcr->sd_auth_key));
       return 0;
    }
 
-   jcr->JobStatus = JS_WaitFD;       /* wait for FD to connect */
+   set_jcr_job_status(jcr, JS_WaitFD);         /* wait for FD to connect */
    dir_send_job_status(jcr);
 
    gettimeofday(&tv, &tz);
@@ -145,9 +165,13 @@ int job_cmd(JCR *jcr)
    timeout.tv_sec = tv.tv_sec + 30 * 60;       /* wait 30 minutes */
 
 
-   Dmsg1(200, "%s waiting on job_start_wait\n", jcr->Job);
+   Dmsg1(100, "%s waiting on job_start_wait\n", jcr->Job);
+   /* Wait for the File daemon to contact us to start the Job,
+    *  when he does, we will be released, unless the 30 minutes
+    *  expires.
+    */
    P(jcr->mutex);
-   for ( ;!job_cancelled(jcr); ) {
+   for ( ;!job_canceled(jcr); ) {
       errstat = pthread_cond_timedwait(&jcr->job_start_wait, &jcr->mutex, &timeout);
       if (errstat == 0 || errstat == ETIMEDOUT) {
         break;
@@ -155,53 +179,36 @@ int job_cmd(JCR *jcr)
    }
    V(jcr->mutex);
 
-   if (jcr->authenticated && !job_cancelled(jcr)) {
+   memset(jcr->sd_auth_key, 0, strlen(jcr->sd_auth_key));
+
+   if (jcr->authenticated && !job_canceled(jcr)) {
+      Dmsg1(100, "Running job %s\n", jcr->Job);
       run_job(jcr);                  /* Run the job */
    }
    return 0;
 }
 
 /*
- * This entry point is only called if we have a separate
- *   Storage Daemon Data port. Otherwise, the connection
- *   is made to the main port, and if it is a File daemon
- *   calling, handl_filed_connection() is called directly.
- */
-void connection_from_filed(void *arg)
-{
-   BSOCK *fd = (BSOCK *)arg;
-   char job_name[MAX_NAME_LENGTH];
-
-   Dmsg0(200, "enter connection_from_filed\n");
-   if (bnet_recv(fd) <= 0) {
-      Emsg0(M_ERROR, 0, "Unable to authenticate Client.\n");
-      return;
-   }
-   Dmsg1(100, "got: %s\n", fd->msg);
-
-   if (sscanf(fd->msg, "Hello Start Job %127s\n", job_name) != 1) {
-      Emsg1(M_ERROR, 0, _("Authentication failure: %s"), fd->msg);
-      return;
-   }
-   handle_filed_connection(fd, job_name);
-   return;
-}
-
+ * After receiving a connection (in job.c) if it is
+ *   from the File daemon, this routine is called.
+ */  
 void handle_filed_connection(BSOCK *fd, char *job_name)
 {
    JCR *jcr;
 
+   bmicrosleep(0, 50000);            /* wait 50 millisecs */
    if (!(jcr=get_jcr_by_full_name(job_name))) {
-      Emsg1(M_ERROR, 0, "Job name not found: %s", job_name);
+      Jmsg1(NULL, M_FATAL, 0, _("Job name not found: %s\n"), job_name);
       return;
    }
 
    jcr->file_bsock = fd;
+   jcr->file_bsock->jcr = jcr;
 
-   Dmsg1(100, "Found Job %s\n", job_name);
+   Dmsg1(110, "Found Job %s\n", job_name);
 
    if (jcr->authenticated) {
-      Dmsg2(000, "Hey!!!! JobId %d Job %s already authenticated.\n", 
+      Pmsg2(000, "Hey!!!! JobId %u Job %s already authenticated.\n", 
         jcr->JobId, jcr->Job);
    }
   
@@ -213,12 +220,12 @@ void handle_filed_connection(BSOCK *fd, char *job_name)
       Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate File daemon\n"));
    } else {
       jcr->authenticated = TRUE;
-      Dmsg1(100, "OK Authentication Job %s\n", jcr->Job);
+      Dmsg1(110, "OK Authentication Job %s\n", jcr->Job);
    }
 
    P(jcr->mutex);
    if (!jcr->authenticated) {
-      jcr->JobStatus = JS_ErrorTerminated;
+      set_jcr_job_status(jcr, JS_ErrorTerminated);
    }
    pthread_cond_signal(&jcr->job_start_wait); /* wake waiting job */
    V(jcr->mutex);
@@ -242,11 +249,11 @@ static int use_device_cmd(JCR *jcr)
    DEVRES *device;
 
    if (bnet_recv(dir) <= 0) {
-      Emsg0(M_FATAL, 0, "No Device from Director\n");
+      Jmsg0(jcr, M_FATAL, 0, _("No Device from Director\n"));
       return 0;
    }
    
-   Dmsg1(20, "Use device: %s", dir->msg);
+   Dmsg1(120, "Use device: %s", dir->msg);
    dev_name = get_memory(dir->msglen);
    media_type = get_memory(dir->msglen);
    pool_name = get_memory(dir->msglen);
@@ -261,7 +268,7 @@ static int use_device_cmd(JCR *jcr)
       while ((device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device))) {
         /* Find resource, and make sure we were able to open it */
         if (strcmp(device->hdr.name, dev_name) == 0 && device->dev) {
-            Dmsg1(20, "Found device %s\n", device->hdr.name);
+            Dmsg1(120, "Found device %s\n", device->hdr.name);
            jcr->pool_name = get_memory(strlen(pool_name) + 1);
            strcpy(jcr->pool_name, pool_name);
            jcr->pool_type = get_memory(strlen(pool_type) + 1);
@@ -271,7 +278,7 @@ static int use_device_cmd(JCR *jcr)
            jcr->dev_name = get_memory(strlen(dev_name) + 1);
            strcpy(jcr->dev_name, dev_name);
            jcr->device = device;
-           Dmsg4(20, use_device, dev_name, media_type, pool_name, pool_type);
+           Dmsg4(120, use_device, dev_name, media_type, pool_name, pool_type);
            free_memory(dev_name);
            free_memory(media_type);
            free_memory(pool_name);
@@ -281,12 +288,23 @@ static int use_device_cmd(JCR *jcr)
         }
       }
       UnlockRes();
-      Jmsg(jcr, M_FATAL, 0, _("Requested device %s not found. Cannot continue.\n"),
+      if (verbose) {
+        unbash_spaces(dir->msg);
+        pm_strcpy(&jcr->errmsg, dir->msg);
+         Jmsg(jcr, M_INFO, 0, _("Failed command: %s\n"), jcr->errmsg);
+      }
+      Jmsg(jcr, M_FATAL, 0, _("\n"
+         "     Device \"%s\" requested by Dir not found in SD Device resources.\n"),
           dev_name);
       bnet_fsend(dir, NO_device, dev_name);
    } else {
-      Emsg1(M_FATAL, 0, _("store<dir: Bad Use Device command: %s\n"), dir->msg);
-      bnet_fsend(dir, BAD_use, dir->msg);
+      unbash_spaces(dir->msg);
+      pm_strcpy(&jcr->errmsg, dir->msg);
+      if (verbose) {
+         Jmsg(jcr, M_INFO, 0, _("Failed command: %s\n"), jcr->errmsg);
+      }
+      Jmsg(jcr, M_FATAL, 0, _("Bad Use Device command: %s\n"), jcr->errmsg);
+      bnet_fsend(dir, BAD_use, jcr->errmsg);
    }
 
    free_memory(dev_name);
@@ -304,6 +322,7 @@ void stored_free_jcr(JCR *jcr)
 {
    if (jcr->file_bsock) {
       bnet_close(jcr->file_bsock);
+      jcr->file_bsock = NULL;
    }
    if (jcr->pool_name) {
       free_memory(jcr->pool_name);
@@ -327,14 +346,21 @@ void stored_free_jcr(JCR *jcr)
    if (jcr->fileset_name) {
       free_memory(jcr->fileset_name);
    }
+   if (jcr->fileset_md5) {
+      free_memory(jcr->fileset_md5);
+   }
    if (jcr->bsr) {
       free_bsr(jcr->bsr);
+      jcr->bsr = NULL;
    }
    if (jcr->RestoreBootstrap) {
       unlink(jcr->RestoreBootstrap);
       free_pool_memory(jcr->RestoreBootstrap);
       jcr->RestoreBootstrap = NULL;
    }
-                     
+   if (jcr->next_dev && jcr->prev_dev) {
+      Emsg0(M_FATAL, 0, _("In free_jcr(), but still attached to device!!!!\n"));
+   }
+   pthread_cond_destroy(&jcr->job_start_wait);
    return;
 }