common/ompio and sharedfp/lockedfile: store full path

This fix is based on a bug report on the mailing list. If the user
opens a file in a directory, but executes a chdir later before closing the file,
ompio was unable to properly clean up the lock files generated by the lockedfile
component. After some investigation it was confirmed, that the same problem
arises if a user would perform the same sequence of operations using the MODE_DELETE_AFTER_CLOSE,
hence the fix also had to be applied to the general file handle, not just the sharedfp component.

Note, this fix should apply clearly to 5.0, but I am not entirely sure whether we'll have to generate
a separate patch for 4.1 and 4.0, since the common_ompio datastructure has changed and got an additional element.

Fixes issue open-mpi#9924

Signed-off-by: Edgar Gabriel <[email protected]>

incorporate changes requested during the review

Signed-off-by: Edgar Gabriel <[email protected]>

Author: edgargabriel

ompi/mca/common/ompio/common_ompio.h

@@ -158,6 +158,7 @@ struct ompio_file_t {
     int                    f_perm;
     ompi_communicator_t   *f_comm;
     const char            *f_filename;
+    char                  *f_fullfilename;
     char                  *f_datarep;
     opal_convertor_t      *f_mem_convertor;
     opal_convertor_t      *f_file_convertor;

ompi/mca/common/ompio/common_ompio_file_open.c

@@ -41,6 +41,9 @@
 #include <math.h>
 #include "common_ompio.h"
 #include "ompi/mca/topo/topo.h"
+#include "opal/util/opal_getcwd.h"
+#include "opal/util/path.h"
+#include "opal/util/os_path.h"
 
 static mca_common_ompio_generate_current_file_view_fn_t generate_current_file_view_fn;
 static mca_common_ompio_get_mca_parameter_value_fn_t get_mca_parameter_value_fn;
@@ -100,6 +103,22 @@ int mca_common_ompio_file_open (ompi_communicator_t *comm,
     ompio_fh->f_get_mca_parameter_value=get_mca_parameter_value_fn;
 
     ompio_fh->f_filename = filename;
+    if (opal_path_is_absolute(filename) ) {
+        ompio_fh->f_fullfilename = strdup(filename);
+    }
+    else {
+        char path[OPAL_PATH_MAX];
+        ret = opal_getcwd(path, OPAL_PATH_MAX);
+        if (OPAL_SUCCESS != ret) {
+            goto fn_fail;
+        }
+        ompio_fh->f_fullfilename = opal_os_path(0, path, filename, NULL);
+        if (NULL == ompio_fh->f_fullfilename){
+            ret = OMPI_ERROR;
+            goto fn_fail;
+        }
+    }
+    
     mca_common_ompio_set_file_defaults (ompio_fh);
 
     ompio_fh->f_split_coll_req    = NULL;
@@ -285,7 +304,7 @@ int mca_common_ompio_file_close (ompio_file_t *ompio_fh)
 	ret = ompio_fh->f_fs->fs_file_close (ompio_fh);
     }
     if ( delete_flag ) {
-        ret = mca_common_ompio_file_delete ( ompio_fh->f_filename, &(MPI_INFO_NULL->super) );
+        ret = mca_common_ompio_file_delete ( ompio_fh->f_fullfilename, &(MPI_INFO_NULL->super) );
     }
 
     if ( NULL != ompio_fh->f_fs ) {
@@ -350,7 +369,8 @@ int mca_common_ompio_file_close (ompio_file_t *ompio_fh)
         free ( ompio_fh->f_coll_write_time );
         ompio_fh->f_coll_write_time = NULL;
     }
-
+    free (ompio_fh->f_fullfilename);
+    
     if ( NULL != ompio_fh->f_coll_read_time ) {
         free ( ompio_fh->f_coll_read_time );
         ompio_fh->f_coll_read_time = NULL;
@@ -371,8 +391,7 @@ int mca_common_ompio_file_close (ompio_file_t *ompio_fh)
     if ( MPI_DATATYPE_NULL != ompio_fh->f_orig_filetype ){
 	ompi_datatype_destroy (&ompio_fh->f_orig_filetype);
     }
-
-
+    
     if (MPI_COMM_NULL != ompio_fh->f_comm && !(ompio_fh->f_flags & OMPIO_SHAREDFP_IS_SET) )  {
         ompi_comm_free (&ompio_fh->f_comm);
     }

ompi/mca/sharedfp/lockedfile/sharedfp_lockedfile_file_open.c

@@ -38,6 +38,9 @@
 #include <unistd.h>
 
 #include "opal/util/output.h"
+#include "opal/util/opal_getcwd.h"
+#include "opal/util/path.h"
+#include "opal/util/os_path.h"
 
 int mca_sharedfp_lockedfile_file_open (struct ompi_communicator_t *comm,
 				       const char* filename,
@@ -112,8 +115,26 @@ int mca_sharedfp_lockedfile_file_open (struct ompi_communicator_t *comm,
         return OMPI_ERR_OUT_OF_RESOURCE;
     }
     snprintf(lockedfilename, filenamelen, "%s-%u-%d%s",filename,masterjobid,int_pid,".lock");
-    module_data->filename = lockedfilename;
-    
+    if (opal_path_is_absolute(lockedfilename) ) {
+        module_data->filename = lockedfilename;
+    } else {
+        char path[OPAL_PATH_MAX];
+        err = opal_getcwd(path, OPAL_PATH_MAX);
+        if (OPAL_SUCCESS != err) {
+            free (sh);
+            free (module_data);
+            free (lockedfilename);
+            return err;
+        }
+        module_data->filename = opal_os_path(0, path, lockedfilename, NULL);
+        if (NULL == module_data->filename){
+            free (sh);
+            free (module_data);
+            free (lockedfilename);
+            return OMPI_ERROR;
+        }
+    }
+
     /*-------------------------------------------------*/
     /*Open the lockedfile without shared file pointer  */
     /*-------------------------------------------------*/
@@ -131,8 +152,8 @@ int mca_sharedfp_lockedfile_file_open (struct ompi_communicator_t *comm,
             free (lockedfilename);
             return OMPI_ERROR;
         }
-	err = opal_best_effort_write ( handle, &position, sizeof(OMPI_MPI_OFFSET_TYPE) );
-        if (OPAL_SUCCESS != err )  {
+	err = opal_best_effort_write (handle, &position, sizeof(OMPI_MPI_OFFSET_TYPE));
+        if (OPAL_SUCCESS != err)  {
             free (sh);
             free (module_data);
             free (lockedfilename);