Merge pull request #77 from openpmix/master

Fork Sync: Update from parent repository

Author: hppritcha

src/mca/ess/base/ess_base_bootstrap.c

@@ -223,6 +223,9 @@ int prte_ess_base_bootstrap(void)
     if (NULL != cluster) {
         free(cluster);
     }
+    if (NULL != ctrlhost) {
+        free(ctrlhost);
+    }
     if (NULL != dvmnodes) {
         free(dvmnodes);
     }

src/mca/ess/base/ess_base_std_prted.c

@@ -446,8 +446,9 @@ int prte_ess_base_prted_setup(void)
 error:
     pmix_show_help("help-prte-runtime.txt", "prte_init:startup:internal-failure", true,
                    error, PRTE_ERROR_NAME(ret), ret);
-    /* remove our use of the session directory tree */
-    PMIX_RELEASE(jdata);
+    if (NULL != jdata) {
+        PMIX_RELEASE(jdata);
+    }
     return PRTE_ERR_SILENT;
 }
 

src/mca/grpcomm/direct/grpcomm_direct_group.c

@@ -8,7 +8,7 @@
  * Copyright (c) 2014-2020 Intel, Inc.  All rights reserved.
  * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
- * Copyright (c) 2021-2024 Nanook Consulting  All rights reserved.
+ * Copyright (c) 2021-2025 Nanook Consulting  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -226,19 +226,43 @@ static void group(int sd, short args, void *cbdata)
         PMIx_Info_list_convert(grpinfo, &darray);
         info = (pmix_info_t*)darray.array;
         ninfo = darray.size;
-        PMIx_Data_pack(NULL, relay, &ninfo, 1, PMIX_SIZE);
+        rc = PMIx_Data_pack(NULL, relay, &ninfo, 1, PMIX_SIZE);
+        if (PMIX_SUCCESS != rc) {
+            PMIX_ERROR_LOG(rc);
+            PMIX_DATA_BUFFER_RELEASE(relay);
+            PMIX_DESTRUCT(&sig);
+            goto error;
+        }
         if (0 < ninfo) {
-            PMIx_Data_pack(NULL, relay, info, ninfo, PMIX_INFO);
+            rc = PMIx_Data_pack(NULL, relay, info, ninfo, PMIX_INFO);
+            if (PMIX_SUCCESS != rc) {
+                PMIX_ERROR_LOG(rc);
+                PMIX_DATA_BUFFER_RELEASE(relay);
+                PMIX_DESTRUCT(&sig);
+                goto error;
+            }
         }
         PMIX_DATA_ARRAY_DESTRUCT(&darray);
 
         // pack any endpts
         PMIx_Info_list_convert(endpts, &darray);
         info = (pmix_info_t*)darray.array;
         ninfo = darray.size;
-        PMIx_Data_pack(NULL, relay, &ninfo, 1, PMIX_SIZE);
+        rc = PMIx_Data_pack(NULL, relay, &ninfo, 1, PMIX_SIZE);
+        if (PMIX_SUCCESS != rc) {
+            PMIX_ERROR_LOG(rc);
+            PMIX_DATA_BUFFER_RELEASE(relay);
+            PMIX_DESTRUCT(&sig);
+            goto error;
+        }
         if (0 < ninfo) {
-            PMIx_Data_pack(NULL, relay, info, ninfo, PMIX_INFO);
+            rc = PMIx_Data_pack(NULL, relay, info, ninfo, PMIX_INFO);
+            if (PMIX_SUCCESS != rc) {
+                PMIX_ERROR_LOG(rc);
+                PMIX_DATA_BUFFER_RELEASE(relay);
+                PMIX_DESTRUCT(&sig);
+                goto error;
+            }
         }
         PMIX_DATA_ARRAY_DESTRUCT(&darray);
     }
@@ -298,7 +322,7 @@ void prte_grpcomm_direct_grp_recv(int status, pmix_proc_t *sender,
     prte_namelist_t *nm;
     pmix_data_array_t darray;
     pmix_status_t st;
-    pmix_info_t *info = NULL, *endpts, *grpinfo;
+    pmix_info_t *info = NULL, *endpts, *grpinfo = NULL;
     prte_grpcomm_direct_group_signature_t *sig = NULL;
     pmix_data_buffer_t *reply;
     prte_grpcomm_group_t *coll;
@@ -386,7 +410,9 @@ void prte_grpcomm_direct_grp_recv(int status, pmix_proc_t *sender,
         rc = PMIx_Data_unpack(NULL, buffer, &nendpts, &cnt, PMIX_SIZE);
         if (PMIX_SUCCESS != rc) {
             PMIX_ERROR_LOG(rc);
-            PMIX_INFO_FREE(grpinfo, ngrpinfo);
+            if (NULL != grpinfo) {
+                PMIX_INFO_FREE(grpinfo, ngrpinfo);
+            }
             PMIX_RELEASE(sig);
             return;
         }
@@ -396,7 +422,9 @@ void prte_grpcomm_direct_grp_recv(int status, pmix_proc_t *sender,
             rc = PMIx_Data_unpack(NULL, buffer, endpts, &cnt, PMIX_INFO);
             if (PMIX_SUCCESS != rc) {
                 PMIX_ERROR_LOG(rc);
-                PMIX_INFO_FREE(grpinfo, ngrpinfo);
+                if (NULL != grpinfo) {
+                    PMIX_INFO_FREE(grpinfo, ngrpinfo);
+                }
                 PMIX_INFO_FREE(endpts, nendpts);
                 PMIX_RELEASE(sig);
                 return;
@@ -619,19 +647,43 @@ void prte_grpcomm_direct_grp_recv(int status, pmix_proc_t *sender,
                 PMIx_Info_list_convert(coll->grpinfo, &darray);
                 info = (pmix_info_t*)darray.array;
                 ninfo = darray.size;
-                PMIx_Data_pack(NULL, reply, &ninfo, 1, PMIX_SIZE);
+                rc = PMIx_Data_pack(NULL, reply, &ninfo, 1, PMIX_SIZE);
+                if (PMIX_SUCCESS != rc) {
+                    PMIX_ERROR_LOG(rc);
+                    PMIX_DATA_BUFFER_RELEASE(reply);
+                    PMIX_RELEASE(sig);
+                    return;
+                }
                 if (0 < ninfo) {
-                    PMIx_Data_pack(NULL, reply, info, ninfo, PMIX_INFO);
+                    rc = PMIx_Data_pack(NULL, reply, info, ninfo, PMIX_INFO);
+                    if (PMIX_SUCCESS != rc) {
+                        PMIX_ERROR_LOG(rc);
+                        PMIX_DATA_BUFFER_RELEASE(reply);
+                        PMIX_RELEASE(sig);
+                        return;
+                    }
                 }
                 PMIX_DATA_ARRAY_DESTRUCT(&darray);
 
                 // pack any endpts
                 PMIx_Info_list_convert(coll->endpts, &darray);
                 info = (pmix_info_t*)darray.array;
                 ninfo = darray.size;
-                PMIx_Data_pack(NULL, reply, &ninfo, 1, PMIX_SIZE);
+                rc = PMIx_Data_pack(NULL, reply, &ninfo, 1, PMIX_SIZE);
+                if (PMIX_SUCCESS != rc) {
+                    PMIX_ERROR_LOG(rc);
+                    PMIX_DATA_BUFFER_RELEASE(reply);
+                    PMIX_RELEASE(sig);
+                    return;
+                }
                 if (0 < ninfo) {
-                    PMIx_Data_pack(NULL, reply, info, ninfo, PMIX_INFO);
+                    rc =PMIx_Data_pack(NULL, reply, info, ninfo, PMIX_INFO);
+                    if (PMIX_SUCCESS != rc) {
+                        PMIX_ERROR_LOG(rc);
+                        PMIX_DATA_BUFFER_RELEASE(reply);
+                        PMIX_RELEASE(sig);
+                        return;
+                    }
                 }
                 PMIX_DATA_ARRAY_DESTRUCT(&darray);
             }

src/mca/odls/base/base.h

@@ -12,7 +12,7 @@
  * Copyright (c) 2011-2020 Cisco Systems, Inc.  All rights reserved
  * Copyright (c) 2013      Los Alamos National Security, LLC.  All rights reserved.
  * Copyright (c) 2017-2019 Intel, Inc.  All rights reserved.
- * Copyright (c) 2021-2024 Nanook Consulting  All rights reserved.
+ * Copyright (c) 2021-2025 Nanook Consulting  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -56,7 +56,6 @@ typedef struct {
     char **ev_threads;            // event progress thread names
     int next_base;                // counter to load-level thread use
     bool signal_direct_children_only;
-    pmix_lock_t lock;
     char *exec_agent;
 } prte_odls_globals_t;
 

src/mca/odls/base/odls_base_bind.c

@@ -275,11 +275,10 @@ void prte_odls_base_set(prte_odls_spawn_caddy_t *cd, int write_fd)
         hwloc_bitmap_free(cpuset);
         /* if we got an error and this wasn't a default binding policy, then report it */
         if (rc < 0 && PRTE_BINDING_POLICY_IS_SET(jobdat->map->binding)) {
-            char *tmp = NULL;
             if (errno == ENOSYS) {
-                msg = "hwloc indicates cpu binding not supported";
+                msg = strdup("hwloc indicates cpu binding not supported");
             } else if (errno == EXDEV) {
-                msg = "hwloc indicates cpu binding cannot be enforced";
+                msg = strdup("hwloc indicates cpu binding cannot be enforced");
             } else {
                 pmix_asprintf(&msg, "hwloc_set_cpubind returned \"%s\" for bitmap \"%s\"",
                               prte_strerror(rc), child->cpuset);
@@ -291,19 +290,13 @@ void prte_odls_base_set(prte_odls_spawn_caddy_t *cd, int write_fd)
                                      "binding generic error",
                                      prte_process_info.nodename, context->app, msg,
                                      __FILE__, __LINE__);
+                free(msg);  // silence static analyzer warning
             } else {
                 send_warn_show_help(write_fd, "help-prte-odls-default.txt",
                                     "not bound", prte_process_info.nodename,
                                     context->app, msg, __FILE__, __LINE__);
-                if (NULL != tmp) {
-                    free(tmp);
-                    free(msg);
-                }
-                return;
-            }
-            if (NULL != tmp) {
-                free(tmp);
                 free(msg);
+                return;
             }
         }
 

src/mca/odls/base/odls_base_default_fns.c

@@ -1252,14 +1252,15 @@ void prte_odls_base_default_launch_local(int fd, short sd, void *cbdata)
     int j, idx;
     int total_num_local_procs = 0;
     prte_odls_launch_local_t *caddy = (prte_odls_launch_local_t *) cbdata;
-    prte_job_t *jobdat;
+    prte_job_t *jobdat, *parent;
     pmix_nspace_t job;
     prte_odls_base_fork_local_proc_fn_t fork_local = caddy->fork_local;
-    bool index_argv;
+    bool index_argv, inherit;
     char *msg, **xfer;
     prte_odls_spawn_caddy_t *cd;
     prte_event_base_t *evb;
     prte_schizo_base_module_t *schizo;
+    pmix_proc_t *nptr;
     PRTE_HIDE_UNUSED_PARAMS(fd, sd);
 
     PMIX_ACQUIRE_OBJECT(caddy);
@@ -1352,6 +1353,20 @@ void prte_odls_base_default_launch_local(int fd, short sd, void *cbdata)
         }
     }
 
+    // see if we have a parent in case of inheritance
+    nptr = NULL;
+    prte_get_attribute(&jobdat->attributes, PRTE_JOB_LAUNCH_PROXY, (void **) &nptr, PMIX_PROC);
+    if (NULL != nptr) {
+        parent = prte_get_job_data_object(nptr->nspace);
+        if (NULL != parent) {
+            inherit = prte_get_attribute(&parent->attributes, PRTE_JOB_INHERIT, NULL, PMIX_BOOL);
+        } else {
+            inherit = false;
+        }
+    } else {
+        inherit = false;
+    }
+
     for (j = 0; j < jobdat->apps->size; j++) {
         app = (prte_app_context_t *) pmix_pointer_array_get_item(jobdat->apps, j);
         if (NULL == app) {
@@ -1395,6 +1410,10 @@ void prte_odls_base_default_launch_local(int fd, short sd, void *cbdata)
         }
 
         // process any provided env directives
+        if (inherit) {
+            // start with the parent's directives
+            process_envars(parent, app);
+        }
         process_envars(jobdat, app);
 
 
@@ -2151,6 +2170,11 @@ int prte_odls_base_default_restart_proc(prte_proc_t *child,
         child->rml_uri = NULL;
     }
     app = (prte_app_context_t *) pmix_pointer_array_get_item(jobdat->apps, child->app_idx);
+    if (NULL == app) {
+        PRTE_ERROR_LOG(PRTE_ERR_NOT_FOUND);
+        rc = PRTE_ERR_NOT_FOUND;
+        goto CLEANUP;
+    }
 
     /* setup the path */
     if (PRTE_SUCCESS != (rc = setup_path(app, &wdir))) {

src/mca/odls/base/odls_base_frame.c

@@ -78,7 +78,6 @@ prte_odls_globals_t prte_odls_globals = {
     .ev_threads = NULL,
     .next_base = 0,
     .signal_direct_children_only = false,
-    .lock = PMIX_LOCK_STATIC_INIT,
     .exec_agent = NULL
 };
 
@@ -126,7 +125,6 @@ void prte_odls_base_harvest_threads(void)
 {
     int i;
 
-    PMIX_ACQUIRE_THREAD(&prte_odls_globals.lock);
     if (0 < prte_odls_globals.num_threads) {
         /* stop the progress threads */
         if (NULL != prte_odls_globals.ev_threads) {
@@ -144,18 +142,15 @@ void prte_odls_base_harvest_threads(void)
             prte_odls_globals.ev_threads = NULL;
         }
     }
-    PMIX_RELEASE_THREAD(&prte_odls_globals.lock);
 }
 
 void prte_odls_base_start_threads(prte_job_t *jdata)
 {
     int i;
     char *tmp;
 
-    PMIX_ACQUIRE_THREAD(&prte_odls_globals.lock);
     /* only do this once */
     if (NULL != prte_odls_globals.ev_threads) {
-        PMIX_RELEASE_THREAD(&prte_odls_globals.lock);
         return;
     }
 
@@ -205,7 +200,6 @@ void prte_odls_base_start_threads(prte_job_t *jdata)
             free(tmp);
         }
     }
-    PMIX_RELEASE_THREAD(&prte_odls_globals.lock);
 }
 
 static int prte_odls_base_close(void)
@@ -230,8 +224,6 @@ static int prte_odls_base_close(void)
 
     prte_odls_base_harvest_threads();
 
-    PMIX_DESTRUCT_LOCK(&prte_odls_globals.lock);
-
     return pmix_mca_base_framework_components_close(&prte_odls_base_framework, NULL);
 }
 
@@ -247,9 +239,6 @@ static int prte_odls_base_open(pmix_mca_base_open_flag_t flags)
     bool xterm_hold;
     sigset_t unblock;
 
-    PMIX_CONSTRUCT_LOCK(&prte_odls_globals.lock);
-    prte_odls_globals.lock.active = false; // start with nobody having the thread
-
     /* initialize the global array of local children */
     prte_local_children = PMIX_NEW(pmix_pointer_array_t);
     if (PRTE_SUCCESS
@@ -263,12 +252,9 @@ static int prte_odls_base_open(pmix_mca_base_open_flag_t flags)
     prte_odls_globals.xtermcmd = NULL;
 
     /* ensure that SIGCHLD is unblocked as we need to capture it */
-    if (0 != sigemptyset(&unblock)) {
-        return PRTE_ERROR;
-    }
-    if (0 != sigaddset(&unblock, SIGCHLD)) {
-        return PRTE_ERROR;
-    }
+    sigemptyset(&unblock);
+    sigaddset(&unblock, SIGCHLD);
+
     if (0 != sigprocmask(SIG_UNBLOCK, &unblock, NULL)) {
         return PRTE_ERR_NOT_SUPPORTED;
     }

src/mca/plm/base/plm_base_launch_support.c

@@ -191,9 +191,16 @@ void prte_plm_base_allocation_complete(int fd, short args, void *cbdata)
      * to map so we can see where the procs would have
      * gone - so skip to the mapping state */
     if (prte_get_attribute(&caddy->jdata->attributes, PRTE_JOB_DO_NOT_LAUNCH, NULL, PMIX_BOOL)) {
-        PRTE_ACTIVATE_JOB_STATE(caddy->jdata, PRTE_JOB_STATE_DAEMONS_REPORTED);
         node = (prte_node_t*)pmix_pointer_array_get_item(prte_node_pool, 0);
+        if (NULL == node) {
+            // should never happen
+            PRTE_ERROR_LOG(PRTE_ERR_NOT_FOUND);
+            PRTE_ACTIVATE_JOB_STATE(caddy->jdata, PRTE_JOB_STATE_FAILED_TO_START);
+            PMIX_RELEASE(caddy);
+            return;
+        }
         prte_rmaps_base.require_hwtcpus = !prte_hwloc_base_core_cpus(node->topology->topo);
+        PRTE_ACTIVATE_JOB_STATE(caddy->jdata, PRTE_JOB_STATE_DAEMONS_REPORTED);
     } else {
         /* move the state machine along */
         caddy->jdata->state = PRTE_JOB_STATE_ALLOCATION_COMPLETE;
@@ -1043,6 +1050,10 @@ void prte_plm_base_post_launch(int fd, short args, void *cbdata)
                 continue;
             }
             app = (prte_app_context_t*)pmix_pointer_array_get_item(jdata->apps, proc->app_idx);
+            if (NULL == app) {
+                // should never happen
+                continue;
+            }
             fprintf(fp, "(rank, host, exe, pid) = (%u, %s, %s, %d)\n",
                     proc->name.rank, proc->node->name, app->app, proc->pid);
         }

src/mca/ras/base/ras_base_allocate.c

@@ -306,7 +306,6 @@ void prte_ras_base_display_cpus(prte_job_t *jdata, char *nodelist)
             }
             if (0 == strcmp(nptr->name, nodes[j])) {
                 display_cpus(nptr->topology, jdata, nodes[j]);
-                moveon = true;
                 break;
             }
             if (NULL == nptr->aliases) {

src/mca/ras/pbs/ras_pbs_module.c

@@ -195,6 +195,8 @@ static int discover(pmix_list_t *nodelist, char *pbs_jobid)
                 if (prte_mca_ras_pbs_component.smp_mode) {
                     /* this cannot happen in smp mode */
                     pmix_show_help("help-ras-pbs.txt", "smp-multi", true);
+                    fclose(fp);
+                    free(hostname);
                     return PRTE_ERR_BAD_PARAM;
                 }
                 ++node->slots;