v4.0.x: Applying fixes from #8217

These are not my changes, but I'm trying to incorporate them
from the v4.0.x PR #8217.

v4.0.x version of master PR #8747

Not a cherry-pick (includes orte and openib changes)

Signed-off-by: Geoffrey Paulsen [email protected]

ompi/mca/common/monitoring/common_monitoring.c

@@ -228,9 +228,19 @@ int mca_common_monitoring_init( void )
 
 void mca_common_monitoring_finalize( void )
 {
-    if( ! mca_common_monitoring_enabled || /* Don't release if not last */
-        0 < opal_atomic_sub_fetch_32(&mca_common_monitoring_hold, 1) ) return;
-
+    /* Even if not enabled, this string is allocated at l.310 */
+    if (!mca_common_monitoring_enabled) {
+        if (NULL != mca_common_monitoring_current_filename) {
+            free(mca_common_monitoring_current_filename);
+            mca_common_monitoring_current_filename = NULL;
+        }
+        return;
+    }
+
+    /* Don't release if not last */
+    if (0 < opal_atomic_sub_fetch_32(&mca_common_monitoring_hold, 1))
+        return;
+
     OPAL_MONITORING_PRINT_INFO("common_component_finish");
     /* Dump monitoring informations */
     mca_common_monitoring_flush(mca_common_monitoring_output_enabled,
@@ -241,11 +251,11 @@ void mca_common_monitoring_finalize( void )
     opal_output_close(mca_common_monitoring_output_stream_id);
     free(mca_common_monitoring_output_stream_obj.lds_prefix);
     /* Free internal data structure */
-    free(pml_data);  /* a single allocation */
-    opal_hash_table_remove_all( common_monitoring_translation_ht );
+    free(pml_data); /* a single allocation */
+    opal_hash_table_remove_all(common_monitoring_translation_ht);
     OBJ_RELEASE(common_monitoring_translation_ht);
     mca_common_monitoring_coll_finalize();
-    if( NULL != mca_common_monitoring_current_filename ) {
+    if (NULL != mca_common_monitoring_current_filename) {
         free(mca_common_monitoring_current_filename);
         mca_common_monitoring_current_filename = NULL;
     }

ompi/mca/mtl/ofi/mtl_ofi.c

@@ -104,6 +104,7 @@ ompi_mtl_ofi_add_procs(struct mca_mtl_base_module_t *mtl,
             goto bail;
         }
         memcpy(&ep_names[i*namelen], ep_name, namelen);
+        free(ep_name); // malloc'd inside OFI_COMPAT_MODEX_RECV in pmix3x.c:1097
     }
 
     /**

ompi/mca/mtl/ofi/mtl_ofi_component.c

@@ -821,6 +821,12 @@ ompi_mtl_ofi_finalize(struct mca_mtl_base_module_t *mtl)
         goto finalize_err;
     }
 
+    /* This was strdup()ed at L.714 */
+    if (NULL != ompi_mtl_ofi.provider_name) {
+        free(ompi_mtl_ofi.provider_name);
+        ompi_mtl_ofi.provider_name = NULL;
+    }
+
     return OMPI_SUCCESS;
 
 finalize_err:

opal/mca/btl/openib/btl_openib_component.c

@@ -211,6 +211,11 @@ static int btl_openib_component_close(void)
     OBJ_DESTRUCT(&mca_btl_openib_component.srq_manager.lock);
     OBJ_DESTRUCT(&mca_btl_openib_component.srq_manager.srq_addr_table);
 
+    /* destroy the tx/rx queues */
+    OBJ_DESTRUCT(&mca_btl_openib_component.send_free_coalesced);
+    OBJ_DESTRUCT(&mca_btl_openib_component.send_user_free);
+    OBJ_DESTRUCT(&mca_btl_openib_component.recv_user_free);
+
     opal_btl_openib_connect_base_finalize();
     opal_btl_openib_ini_finalize();
 
@@ -912,6 +917,7 @@ static void device_construct(mca_btl_openib_device_t *device)
     device->ib_channel = NULL;
 #endif
     device->btls = 0;
+    device->allowed_btls = 0;
     device->endpoints = NULL;
     device->device_btls = NULL;
     device->ib_cq[BTL_OPENIB_HP_CQ] = NULL;
@@ -1876,6 +1882,8 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
     device->rcache =
         mca_rcache_base_module_create (mca_btl_openib_component.ib_rcache_name,
                                        device, &rcache_resources);
+    free(rcache_resources.cache_name);
+
     if (NULL == device->rcache) {
         /* Don't print an error message here -- we'll get one from
            mpool_create anyway */
@@ -2277,6 +2285,7 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
     } else if (device->btls > 0) {
         /* no port is allowed to be used by btl/openib,
          * so release the device right away */
+        OBJ_RELEASE(device->device_btls);
         OBJ_RELEASE(device);
         return OPAL_SUCCESS;
     }
@@ -2292,6 +2301,7 @@ static int init_one_device(opal_list_t *btl_list, struct ibv_device* ib_dev)
     if (NULL != allowed_ports) {
         free(allowed_ports);
     }
+    OBJ_RELEASE(device->device_btls);
     OBJ_RELEASE(device);
     return ret;
 }
@@ -2704,6 +2714,7 @@ btl_openib_component_init(int *num_btl_modules,
         }
     }
 
+    // These are used nowhere else, ergo...
     OBJ_CONSTRUCT(&mca_btl_openib_component.send_free_coalesced, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_openib_component.send_user_free, opal_free_list_t);
     OBJ_CONSTRUCT(&mca_btl_openib_component.recv_user_free, opal_free_list_t);
@@ -2733,6 +2744,7 @@ btl_openib_component_init(int *num_btl_modules,
         goto no_btls;
     }
 
+    free(init_data);
     init_data = (mca_btl_openib_frag_init_data_t *) malloc(sizeof(mca_btl_openib_frag_init_data_t));
     if (NULL == init_data) {
         BTL_ERROR(("Failed malloc: %s:%d", __FILE__, __LINE__));
@@ -2754,6 +2766,7 @@ btl_openib_component_init(int *num_btl_modules,
         goto no_btls;
     }
 
+    free(init_data);
     init_data = (mca_btl_openib_frag_init_data_t *) malloc(sizeof(mca_btl_openib_frag_init_data_t));
     if (NULL == init_data) {
         BTL_ERROR(("Failed malloc: %s:%d", __FILE__, __LINE__));
@@ -2773,6 +2786,8 @@ btl_openib_component_init(int *num_btl_modules,
         goto no_btls;
     }
 
+    free(init_data);
+
     /* If fork support is requested, try to enable it */
     if (OPAL_SUCCESS != (ret = opal_common_verbs_fork_test())) {
         goto no_btls;
@@ -3058,6 +3073,8 @@ btl_openib_component_init(int *num_btl_modules,
     if (NULL != btls) {
         free(btls);
     }
+
+    opal_ibv_free_device_list(ib_devs);
     return NULL;
 }
 

opal/mca/btl/openib/btl_openib_ini.c

@@ -358,13 +358,15 @@ static int parse_line(parsed_section_values_t *sv)
     if (0 == strcasecmp(key_buffer, "vendor_id")) {
         if (OPAL_SUCCESS != (ret = opal_btl_openib_ini_intify_list(value, &sv->vendor_ids,
                                                &sv->vendor_ids_len))) {
+            free(value);
             return ret;
         }
     }
 
     else if (0 == strcasecmp(key_buffer, "vendor_part_id")) {
         if (OPAL_SUCCESS != (ret = opal_btl_openib_ini_intify_list(value, &sv->vendor_part_ids,
                                                &sv->vendor_part_ids_len))) {
+            free(value);
             return ret;
         }
     }
@@ -468,6 +470,11 @@ static void reset_section(bool had_previous_value, parsed_section_values_t *s)
         if (NULL != s->vendor_part_ids) {
             free(s->vendor_part_ids);
         }
+        // This could be freed here, or had_previous_value forwarded to
+        // reset_values...
+        if (NULL != s->values.receive_queues) {
+            free(s->values.receive_queues);
+        }
     }
 
     s->name = NULL;

opal/mca/mpool/hugepage/mpool_hugepage_component.c

@@ -274,7 +274,8 @@ static void mca_mpool_hugepage_find_hugepages (void) {
             opal_output_verbose (MCA_BASE_VERBOSE_INFO, opal_mpool_base_framework.framework_output,
                                  "found huge page with size = %lu, path = %s, mmap flags = 0x%x, with invalid " 
                                  "permissions, skipping", hp->page_size, hp->path, hp->mmap_flags);
-        }        
+        }
+        OBJ_RELEASE(hp);
     }
 
     opal_list_sort (&mca_mpool_hugepage_component.huge_pages, page_compare);

orte/mca/ess/base/ess_base_fns.c

@@ -344,6 +344,9 @@ int orte_ess_base_proc_binding(void)
                     }
                 }
             }
+        } else {
+            /* Prevent cpus from being unfreed if we fall through */
+            hwloc_bitmap_free(cpus);
         }
     } else {
         OPAL_OUTPUT_VERBOSE((5, orte_ess_base_framework.framework_output,

orte/util/show_help.c

@@ -721,6 +721,7 @@ int orte_show_help_norender(const char *filename, const char *topic,
                     nanosleep(&tp, NULL);
                 }
                 OBJ_RELEASE(buf);
+                free(kv->data.bo.bytes);
                 kv->data.bo.bytes = NULL;
                 OPAL_LIST_DESTRUCT(&info);
                 rc = ORTE_SUCCESS;