sharedfp_sm_file_component_query: add file open to ensure correct ope…

…rations

try to actually open the sharedfp/sm file during the query operation to ensure
that the component can actually run. This is based on some reports on the mailing list that
the sharedfp/sm operation causes problems in certain circumstances.

Fixes issue #9656

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

ompi/mca/sharedfp/sm/sharedfp_sm.c

@@ -94,6 +94,37 @@ struct mca_sharedfp_base_module_1_0_0_t * mca_sharedfp_sm_component_file_query(o
             return NULL;
         }
     }
+
+
+    /* Check that we can actually open the required file */
+    char *filename_basename = basename((char*)fh->f_filename);
+
+    /* format is "%s/%s_cid-%d-%d.sm", see below */
+    int sm_filename_length = strlen(ompi_process_info.job_session_dir) + 1 + strlen(filename_basename) + 5 + (3*sizeof(uint32_t)+1) + 4;
+    char *sm_filename = (char*) malloc( sizeof(char) * sm_filename_length);
+    if (NULL == sm_filename) {
+        opal_output(0, "mca_sharedfp_sm_component_file_query: Error, unable to malloc sm_filename\n");
+        return NULL;
+    }
+
+    int comm_cid = -1;
+    int pid = ompi_comm_rank (comm);
+
+    snprintf(sm_filename, sm_filename_length, "%s/%s_cid-%d-%d.sm", ompi_process_info.job_session_dir,
+             filename_basename, comm_cid, pid);
+
+    int sm_fd = open(sm_filename, O_RDWR | O_CREAT,
+                     S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
+    if ( sm_fd == -1){
+        /*error opening file*/
+        opal_output(0,"mca_sharedfp_sm_component_file_query: Error, unable to open file for mmap: %s\n",sm_filename);
+        free(sm_filename);
+        return NULL;
+    }
+    close (sm_fd);
+    unlink(sm_filename);
+    free (sm_filename);
+
     /* This module can run */
     *priority = mca_sharedfp_sm_priority;
     return &sm;