4.1.4 Release

CHANGELOG

@@ -73,6 +73,8 @@ c - crash     b - bug fix    e - enhancement    f - new feature  n - note
       set to the login nodes. TRQ-1482.
   b - Fix a bug that using -V got rid of -v. TRQ-1457.
   b - Make qsub -I -x work again. TRQ-1483.
+  c - Fix a potential crash when getting the status of a login node in cray mode.
+      TRQ-1491.
 
 4.1.3
   b - fix a security loophole that potentially allowed an interactive job to run 

Makefile.am

@@ -8,6 +8,7 @@ EXTRA_DIST = acinclude.m4 \
   CHANGELOG \
   configure.ac \
   cov_file_results.pl \
+	current_hash \
   Doxyfile \
   INSTALL \
 	INSTALL.GNU \

configure.ac

@@ -69,13 +69,16 @@ dnl  be created in the include directory.
 AM_CONFIG_HEADER([src/include/pbs_config.h])
 
 dnl
-dnl Find our svn revision number for program outputs
+dnl Find our git revision hash for program outputs
 dnl
-AC_PATH_PROG(svnversioncommand, svnversion)
-if test "X$svnversioncommand" != "X" && test `$svnversioncommand -n '.'` != "exported"; then
-  AC_DEFINE_UNQUOTED(SVN_VERSION, ["`svnversion -n`"], [repository svn version])
+githash="`git rev-parse HEAD`"
+success=$?
+if test "X$success" != "X0" ; then
+  AC_DEFINE_UNQUOTED(GIT_HASH, ["$githash"], [repository svn version])
 else
-  AC_DEFINE_UNQUOTED(SVN_VERSION, ["unknown"], [repository svn version])
+  dnl current hash is a file checked in to the git repo in the base directory
+  dnl containing the most recent hash
+  AC_DEFINE_UNQUOTED(GIT_HASH, ["`cat current_hash`"], [repository svn version])
 fi
 
 

current_hash

@@ -0,0 +1 @@
+a2498b31a71e656959f38ce8a88024fbc869eaad

src/cmds/pbsnodes.c

@@ -712,7 +712,7 @@ int main(
         if ((optarg != NULL) && !strcmp(optarg, "version"))
           {
           fprintf(stderr, "Version: %s\nRevision: %s\n",
-            PACKAGE_VERSION, SVN_VERSION);
+            PACKAGE_VERSION, GIT_HASH);
 
           exit(0);
           }

src/cmds/qstat.c

@@ -2207,7 +2207,7 @@ int main(
         if ((optarg != NULL) && !strcmp(optarg, "version"))
           {
           fprintf(stderr, "Version: %s\nRevision: %s\n",
-            PACKAGE_VERSION, SVN_VERSION);
+            PACKAGE_VERSION, GIT_HASH);
 
           exit(0);
           }

src/include/pbs_job.h

@@ -1087,7 +1087,7 @@ extern char *prefix_std_file(job *, dynamic_string *, int);
 extern char *add_std_filename(job *, char *, int, dynamic_string *);
 extern int   set_jobexid(job *, pbs_attribute *, char *);
 extern int   site_check_user_map(job *, char *, char *, int);
-int  svr_dequejob(char *, int);
+int  svr_dequejob(job *, int);
 extern int   svr_enquejob(job *, int, int);
 extern void  svr_evaljobstate(job *, int *, int *, int);
 extern void  svr_mailowner(job *, int, int, char *);

src/include/pbs_nodes.h

@@ -332,6 +332,8 @@ struct pbsnode
   struct AvlNode       *node_boards;         /* private tree of numa nodes */
   char                 *numa_str;            /* comma-delimited string of processor values */
   char                 *gpu_str;             /* comma-delimited string of the number of gpus for each nodeboard */
+
+  unsigned char         nd_mom_reported_down;/* notes that the mom reported its own shutdown */
 
   unsigned char         nd_is_alps_reporter;
   unsigned char         nd_is_alps_login;

src/lib/Libifl/tm.c

@@ -170,8 +170,10 @@ static event_info *event_hash[EVENT_HASH];
 /*
 ** Find an event number or return a NULL.
 */
-static event_info *
-find_event(tm_event_t x)
+event_info *find_event(
+
+  tm_event_t x)
+
   {
   event_info *ep;
 
@@ -187,10 +189,11 @@ find_event(tm_event_t x)
 /*
 ** Delete an event.
 */
-static void
-del_event(event_info *ep)
-  {
+void del_event(
+    
+  event_info *ep)
 
+  {
   /* unlink event from hash list */
   if (ep->e_prev)
     ep->e_prev->e_next = ep->e_next;
@@ -245,8 +248,8 @@ del_event(event_info *ep)
 /*
 ** Create a new event number.
 */
-static tm_event_t
-new_event(void)
+tm_event_t new_event(void)
+
   {
   static tm_event_t next_event = TM_NULL_EVENT + 1;
   event_info  *ep;
@@ -275,8 +278,12 @@ new_event(void)
 /*
 ** Link new event number into the above hash table.
 */
-static void
-add_event(tm_event_t event, tm_node_id node, int type, void *info)
+void add_event(
+
+  tm_event_t  event,
+  tm_node_id  node,
+  int         type,
+  void       *info)
   {
   event_info  *ep, **head;
 
@@ -322,8 +329,10 @@ static task_info *task_hash[TASK_HASH];
 /*
 ** Find a task table entry for a given task number or return a NULL.
 */
-static task_info *
-find_task(tm_task_id x)
+task_info *find_task(
+
+  tm_task_id x)
+
   {
   task_info *tp;
 
@@ -340,8 +349,12 @@ find_task(tm_task_id x)
 ** Create a new task entry and link it into the above hash
 ** table.
 */
-static tm_task_id
-new_task(char *jobid, tm_node_id node, tm_task_id task)
+tm_task_id new_task(
+
+  char       *jobid,
+  tm_node_id  node,
+  tm_task_id  task)
+
   {
   task_info  *tp, **head;
 
@@ -559,8 +572,8 @@ static int localmom(void)
 
 static int startcom(
 
-  int        com,
-  tm_event_t event,
+  int               com,
+  tm_event_t        event,
   struct tcp_chan **pchan)
 
   {
@@ -943,6 +956,7 @@ int tm_obit(
   tm_task_id  tid,  /* in  */
   int        *obitval, /* out */
   tm_event_t *event)  /* out */
+
   {
   int rc = TM_SUCCESS;
   task_info *tp;
@@ -1002,14 +1016,14 @@ struct taskhold
 ** is a valid node number, it returns the event that the list of
 ** tasks on <node> is available.
 */
-int
-tm_taskinfo(
-  tm_node_id node,  /* in  */
+int tm_taskinfo(
+
+  tm_node_id  node,  /* in  */
   tm_task_id *tid_list, /* out */
-  int list_size, /* in  */
-  int *ntasks, /* out */
-  tm_event_t *event  /* out */
-)
+  int         list_size, /* in  */
+  int        *ntasks, /* out */
+  tm_event_t *event)  /* out */
+
   {
   struct taskhold *thold;
   struct tcp_chan *chan = NULL;
@@ -1053,11 +1067,11 @@ tm_taskinfo(
 ** Returns the job-relative node number that holds or held <tid>.  In
 ** case of an error, it returns TM_ERROR_NODE.
 */
-int
-tm_atnode(
-  tm_task_id tid,  /* in  */
-  tm_node_id *node  /* out */
-)
+int tm_atnode(
+
+  tm_task_id  tid,  /* in  */
+  tm_node_id *node)  /* out */
+
   {
   task_info *tp;
 
@@ -1085,12 +1099,13 @@ struct reschold
 ** <node> is available.  It returns ERROR_EVENT otherwise.
 */
 int tm_rescinfo(
-  tm_node_id node,  /* in  */
-  char *resource, /* out */
-  int len,  /* in  */
+
+  tm_node_id  node,  /* in  */
+  char       *resource, /* out */
+  int         len,  /* in  */
   tm_event_t *event)  /* out */
-  {
 
+  {
   struct reschold *rhold;
   struct tcp_chan *chan = NULL;
 
@@ -1125,21 +1140,24 @@ int tm_rescinfo(
   add_event(*event, node, TM_RESOURCES, (void *)rhold);
 
   return TM_SUCCESS;
-  }
+  } /* END tm_rescinfo() */
+
+
+
 
 /*
 ** Posts the first <nbytes> of a copy of *<info> within MOM on
 ** this node, and associated with this task.  If <info> is
 ** non-NULL, it returns the event that the effort to post *<info>
 ** is complete.  It returns ERROR_EVENT otherwise.
 */
-int
-tm_publish(
-  char *name,  /* in  */
-  void *info,  /* in  */
-  int len,  /* in  */
-  tm_event_t *event  /* out */
-)
+int tm_publish(
+  
+  char       *name,  /* in  */
+  void       *info,  /* in  */
+  int         len,  /* in  */
+  tm_event_t *event)  /* out */
+
   {
   int rc = TM_SUCCESS;
   struct tcp_chan *chan = NULL;
@@ -1173,7 +1191,7 @@ tm_publish(
     DIS_tcp_cleanup(chan);
 
   return rc;
-  }
+  } /* tm_publish() */
 
 struct infohold
   {
@@ -1187,18 +1205,18 @@ struct infohold
 ** <tid> is a valid task, it returns the event that the
 ** string specifying the info posted by <tid> is available.
 */
-int
-tm_subscribe(
-  tm_task_id tid,  /* in  */
-  char *name,  /* in  */
-  void *info,  /* out */
-  int len,  /* in  */
-  int *info_len, /* out */
-  tm_event_t *event  /* out */
-)
+int tm_subscribe(
+  
+  tm_task_id  tid,     /* in  */
+  char       *name,    /* in  */
+  void       *info,    /* out */
+  int         len,     /* in  */
+  int        *info_len,/* out */
+  tm_event_t *event)    /* out */
+  
   {
-  int rc = TM_SUCCESS;
-  task_info  *tp;
+  int              rc = TM_SUCCESS;
+  task_info       *tp;
   struct tcp_chan *chan = NULL;
 
   struct infohold *ihold;

src/resmom/linux/mom_mach.c

@@ -2690,6 +2690,9 @@ int mom_close_poll(void)
   if (proc_array != NULL)
     {
     free(proc_array);
+    proc_array = NULL;
+    nproc = 0;
+    max_proc = TBL_INC;
     }
 
   return(PBSE_NONE);