Example #1
0
int opal_compress_bzip_compress_nb(char * fname, char **cname, char **postfix, pid_t *child_pid)
{
    char **argv = NULL;
    char * base_fname = NULL;
    char * dir_fname = NULL;
    int status;
    bool is_dir;

    is_dir = is_directory(fname);

    *child_pid = fork();
    if( *child_pid == 0 ) { /* Child */
        char * cmd;

        dir_fname  = opal_dirname(fname);
        base_fname = opal_basename(fname);

        chdir(dir_fname);

        if( is_dir ) {
#if 0
            opal_compress_base_tar_create(&base_fname);
            asprintf(cname, "%s.bz2", base_fname);
            asprintf(&cmd, "bzip2 %s", base_fname);
#else
            asprintf(cname, "%s.tar.bz2", base_fname);
            asprintf(&cmd, "tar -jcf %s %s", *cname, base_fname);
#endif
        } else {
            asprintf(cname, "%s.bz2", base_fname);
            asprintf(&cmd, "bzip2 %s", base_fname);
        }

        opal_output_verbose(10, mca_compress_bzip_component.super.output_handle,
                            "compress:bzip: compress_nb(%s -> [%s])",
                            fname, *cname);
        opal_output_verbose(10, mca_compress_bzip_component.super.output_handle,
                            "compress:bzip: compress_nb() command [%s]",
                            cmd);

        argv = opal_argv_split(cmd, ' ');
        status = execvp(argv[0], argv);

        opal_output(0, "compress:bzip: compress_nb: Failed to exec child [%s] status = %d\n", cmd, status);
        exit(OPAL_ERROR);
    }
    else if( *child_pid > 0 ) {
        if( is_dir ) {
            *postfix = strdup(".tar.bz2");
        } else {
            *postfix = strdup(".bz2");
        }
        asprintf(cname, "%s%s", fname, *postfix);
    }
    else {
        return OPAL_ERROR;
    }

    return OPAL_SUCCESS;
}
Example #2
0
static int parseable_print(orte_ps_mpirun_info_t *hnpinfo)
{
    orte_job_t **jobs;
    orte_node_t **nodes;
    orte_proc_t *proc;
    orte_app_context_t *app;
    char *appname;
    int i, j;
    char *nodename;

    /* don't include the daemon job in the number of jobs reported */
    printf("mpirun:%lu:num nodes:%d:num jobs:%d\n",
           (unsigned long)hnpinfo->hnp->pid, hnpinfo->num_nodes, hnpinfo->num_jobs-1);

    if (orte_ps_globals.nodes) {
        nodes = hnpinfo->nodes;
        for (i=0; i < hnpinfo->num_nodes; i++) {
            printf("node:%s:state:%s:slots:%d:in use:%d\n",
                   nodes[i]->name, pretty_node_state(nodes[i]->state),
                   nodes[i]->slots, nodes[i]->slots_inuse);
        }
    }

    jobs = hnpinfo->jobs;
    /* skip job=0 as that's the daemon job */
    for (i=1; i < hnpinfo->num_jobs; i++) {
        printf("jobid:%d:state:%s:slots:%d:num procs:%d\n",
               ORTE_LOCAL_JOBID(jobs[i]->jobid),
               orte_job_state_to_str(jobs[i]->state),
               jobs[i]->total_slots_alloc,
               jobs[i]->num_procs);
        /* print the proc info */
        for (j=0; j < jobs[i]->procs->size; j++) {
            if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jobs[i]->procs, j))) {
                continue;
            }
            app = (orte_app_context_t*)opal_pointer_array_get_item(jobs[i]->apps, proc->app_idx);
            if (NULL == app) {
                appname = strdup("NULL");
            } else {
                appname = opal_basename(app->app);
            }
            nodename = NULL;
            orte_get_attribute(&proc->attributes, ORTE_PROC_NODENAME, (void**)&nodename, OPAL_STRING);
            printf("process:%s:rank:%s:pid:%lu:node:%s:state:%s\n",
                   appname, ORTE_VPID_PRINT(proc->name.vpid),
                   (unsigned long)proc->pid,
                   (NULL == nodename) ? "unknown" : nodename,
                   orte_proc_state_to_str(proc->state));
            free(appname);
            if (NULL != nodename) {
                free(nodename);
            }
        }
    }

    return ORTE_SUCCESS;
}
Example #3
0
int orte_util_check_context_app(orte_app_context_t *context, char **env)
{
    char *tmp;

    /* If the app is a naked filename, we need to do a path search for
        it.  orterun will send in whatever the user specified (e.g.,
        "orterun -np 2 uptime"), so in some cases, we need to search
        the path to verify that we can find it.  Here's the
        possibilities:

        1. The user specified an absolute pathname for the executable.
        We simply need to verify that it exists and we can run it.

        2. The user specified a relative pathname for the executable.
        Ditto with #1 -- based on the cwd, we need to verify that it
        exists and we can run it.

        3. The user specified a naked filename.  We need to search the
        path, find a match, and verify that we can run it.

        Note that in some cases, we won't be doing this work here --
        bproc, for example, does not use the fork pls for launching, so
        it does this same work over there. */

    tmp = opal_basename(context->argv[0]);
    if (strlen(tmp) == strlen(context->argv[0])) {
        /* If this is a naked executable -- no relative or absolute
        pathname -- then search the PATH for it */
        free(tmp);
        tmp = opal_path_findv(context->argv[0], X_OK, env, context->cwd);
        if (NULL == tmp) {
            return ORTE_ERR_EXE_NOT_FOUND;
        }
        if (NULL != context->app) free(context->app);
        context->app = tmp;
    } else {
        if (0 != access(context->app, X_OK)) {
            return ORTE_ERR_EXE_NOT_ACCESSIBLE;
        }
    }

    /* All was good */
    return ORTE_SUCCESS;
}
Example #4
0
static int plm_alps_start_proc(int argc, char **argv, char **env,
                                char *prefix)
{
    int fd;
    pid_t alps_pid;
    char *exec_argv = opal_path_findv(argv[0], 0, env, NULL);

    if (NULL == exec_argv) {
        return ORTE_ERR_NOT_FOUND;
    }

    alps_pid = fork();
    if (-1 == alps_pid) {
        ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_CHILDREN);
        return ORTE_ERR_SYS_LIMITS_CHILDREN;
    }
    
    alpsrun = OBJ_NEW(orte_proc_t);
    alpsrun->pid = alps_pid;
    /* be sure to mark it as alive so we don't instantly fire */
    ORTE_FLAG_SET(alpsrun, ORTE_PROC_FLAG_ALIVE);
    /* setup the waitpid so we can find out if alps succeeds! */
    orte_wait_cb(alpsrun, alps_wait_cb, NULL);

    if (0 == alps_pid) {  /* child */
        char *bin_base = NULL, *lib_base = NULL;

        /* Figure out the basenames for the libdir and bindir.  There
           is a lengthy comment about this in plm_rsh_module.c
           explaining all the rationale for how / why we're doing
           this. */

        lib_base = opal_basename(opal_install_dirs.libdir);
        bin_base = opal_basename(opal_install_dirs.bindir);

        /* If we have a prefix, then modify the PATH and
           LD_LIBRARY_PATH environment variables.  */
        if (NULL != prefix) {
            char *oldenv, *newenv;

            /* Reset PATH */
            oldenv = getenv("PATH");
            if (NULL != oldenv) {
                asprintf(&newenv, "%s/%s:%s", prefix, bin_base, oldenv);
            } else {
                asprintf(&newenv, "%s/%s", prefix, bin_base);
            }
            opal_setenv("PATH", newenv, true, &env);
            if (mca_plm_alps_component.debug) {
                opal_output(0, "plm:alps: reset PATH: %s", newenv);
            }
            free(newenv);

            /* Reset LD_LIBRARY_PATH */
            oldenv = getenv("LD_LIBRARY_PATH");
            if (NULL != oldenv) {
                asprintf(&newenv, "%s/%s:%s", prefix, lib_base, oldenv);
            } else {
                asprintf(&newenv, "%s/%s", prefix, lib_base);
            }
            opal_setenv("LD_LIBRARY_PATH", newenv, true, &env);
            if (mca_plm_alps_component.debug) {
                opal_output(0, "plm:alps: reset LD_LIBRARY_PATH: %s",
                            newenv);
            }
            free(newenv);
        }

        fd = open("/dev/null", O_CREAT|O_WRONLY|O_TRUNC, 0666);
        if(fd > 0) {
            dup2(fd, 0);
        }

        /* When not in debug mode and --debug-daemons was not passed,
         * tie stdout/stderr to dev null so we don't see messages from orted */
        if (0 == mca_plm_alps_component.debug && !orte_debug_daemons_flag) {
            if (fd >= 0) {
                if (fd != 1) {
                    dup2(fd,1);
                }
                if (fd != 2) {
                    dup2(fd,2);
                }
            }
        }

        if (fd > 2) {
            close(fd);
        }

        /* get the alps process out of orterun's process group so that
           signals sent from the shell (like those resulting from
           cntl-c) don't get sent to alps */
        setpgid(0, 0);
         
        
        execve(exec_argv, argv, env);

        opal_output(0, "plm:alps:start_proc: exec failed");
        /* don't return - need to exit - returning would be bad -
           we're not in the calling process anymore */
        exit(1);
    } else {  /* parent */
        /* just in case, make sure that the alps process is not in our
        process group any more.  Stevens says always do this on both
        sides of the fork... */
        setpgid(alps_pid, alps_pid);
        
        free(exec_argv);
    }

    return ORTE_SUCCESS;
}
Example #5
0
static int plm_slurm_start_proc(int argc, char **argv, char **env,
                                char *prefix)
{
    int fd;
    int srun_pid;
    char *exec_argv = opal_path_findv(argv[0], 0, env, NULL);
    orte_proc_t *dummy;

    if (NULL == exec_argv) {
        return ORTE_ERR_NOT_FOUND;
    }

    srun_pid = fork();
    if (-1 == srun_pid) {
        ORTE_ERROR_LOG(ORTE_ERR_SYS_LIMITS_CHILDREN);
        free(exec_argv);
        return ORTE_ERR_SYS_LIMITS_CHILDREN;
    }
    
    /* setup a dummy proc object to track the srun */
    dummy = OBJ_NEW(orte_proc_t);
    dummy->pid = srun_pid;
    /* be sure to mark it as alive so we don't instantly fire */
    ORTE_FLAG_SET(dummy, ORTE_PROC_FLAG_ALIVE);
    /* setup the waitpid so we can find out if srun succeeds! */
    orte_wait_cb(dummy, srun_wait_cb, NULL);

    if (0 == srun_pid) {  /* child */
        char *bin_base = NULL, *lib_base = NULL;

        /* Figure out the basenames for the libdir and bindir.  There
           is a lengthy comment about this in plm_rsh_module.c
           explaining all the rationale for how / why we're doing
           this. */

        lib_base = opal_basename(opal_install_dirs.libdir);
        bin_base = opal_basename(opal_install_dirs.bindir);

        /* If we have a prefix, then modify the PATH and
           LD_LIBRARY_PATH environment variables.  */
        if (NULL != prefix) {
            char *oldenv, *newenv;

            /* Reset PATH */
            oldenv = getenv("PATH");
            if (NULL != oldenv) {
                asprintf(&newenv, "%s/%s:%s", prefix, bin_base, oldenv);
            } else {
                asprintf(&newenv, "%s/%s", prefix, bin_base);
            }
            opal_setenv("PATH", newenv, true, &env);
            OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                                 "%s plm:slurm: reset PATH: %s",
                                 ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                                 newenv));
            free(newenv);

            /* Reset LD_LIBRARY_PATH */
            oldenv = getenv("LD_LIBRARY_PATH");
            if (NULL != oldenv) {
                asprintf(&newenv, "%s/%s:%s", prefix, lib_base, oldenv);
            } else {
                asprintf(&newenv, "%s/%s", prefix, lib_base);
            }
            opal_setenv("LD_LIBRARY_PATH", newenv, true, &env);
            OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                                 "%s plm:slurm: reset LD_LIBRARY_PATH: %s",
                                 ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                                 newenv));
            free(newenv);
        }

        fd = open("/dev/null", O_CREAT|O_RDWR|O_TRUNC, 0666);
        if (fd >= 0) {
            dup2(fd, 0);
            /* When not in debug mode and --debug-daemons was not passed,
             * tie stdout/stderr to dev null so we don't see messages from orted
             * EXCEPT if the user has requested that we leave sessions attached
             */
            if (0 > opal_output_get_verbosity(orte_plm_base_framework.framework_output) &&
                !orte_debug_daemons_flag && !orte_leave_session_attached) {
                dup2(fd,1);
                dup2(fd,2);
            }

            /* Don't leave the extra fd to /dev/null open */
            if (fd > 2) {
                close(fd);
            }
        }

        /* get the srun process out of orterun's process group so that
           signals sent from the shell (like those resulting from
           cntl-c) don't get sent to srun */
        setpgid(0, 0);

        execve(exec_argv, argv, env);

        opal_output(0, "plm:slurm:start_proc: exec failed");
        /* don't return - need to exit - returning would be bad -
           we're not in the calling process anymore */
        exit(1);
    } else {  /* parent */
        /* just in case, make sure that the srun process is not in our
           process group any more.  Stevens says always do this on both
           sides of the fork... */
        setpgid(srun_pid, srun_pid);
        
        /* if this is the primary launch - i.e., not a comm_spawn of a
         * child job - then save the pid
         */
        if (!primary_pid_set) {
            primary_srun_pid = srun_pid;
            primary_pid_set = true;
        }
        
        free(exec_argv);
    }

    return ORTE_SUCCESS;
}
/*
 * Use $PWD instead of getcwd() a) if $PWD exists and b) is a valid
 * synonym for the results from getcwd(). If both of these conditions
 * are not met, just fall back and use the results of getcwd().
 */
int opal_getcwd(char *buf, size_t size)
{
    char cwd[OPAL_PATH_MAX];
    char *pwd = getenv("PWD");
    struct stat a, b;
    char *shortened;

    /* Bozo checks (e.g., if someone accidentally passed -1 to the
       unsigned "size" param) */
    if (NULL == buf || size > INT_MAX) {
        return OPAL_ERR_BAD_PARAM;
    }

    /* Call getcwd() to get a baseline result */
    if (NULL == getcwd(cwd, sizeof(cwd))) {
        return OPAL_ERR_IN_ERRNO;
    }

#if !defined(HAVE_SYS_STAT_H)
    /* If we don't have stat(), then we can't tell if the $PWD and cwd
       are the same, so just fall back to getcwd(). */
    pwd = cwd;
#else
    if (NULL == pwd) {
        pwd = cwd;
    } else {
        /* If the two are not the same value, figure out if they are
           pointing to the same place */
        if (0 != strcmp(pwd, cwd)) {
            /* If we can't stat() what getcwd() gave us, give up */
            if (0 != stat(cwd, &a)) {
                return OPAL_ERR_IN_ERRNO;
            } 
            /* If we can't stat() $PWD, then $PWD could just be stale
               -- so ignore it. */
            else if (0 != stat(pwd, &b)) {
                pwd = cwd;
            }
            /* Otherwise, we successfully stat()'ed them both, so
               compare.  If either the device or inode is not the
               same, then fallback to getcwd(). */
            else {
                if (a.st_dev != b.st_dev || a.st_ino != b.st_ino) {
                    pwd = cwd;
                }
            }
        }
    }
#endif

    /* If we got here, pwd is pointing to the result that we want to
       give.  Ensure the user's buffer is long enough.  If it is, copy
       in the value and be done. */
    if (strlen(pwd) > size) {
        /* if it isn't big enough, give them as much
         * of the basename as possible
         */
        shortened = opal_basename(pwd);
        strncpy(buf, shortened, size);
        free(shortened);
        /* ensure it is null terminated */
        buf[size-1] = '\0';
        /* indicate that it isn't the full path */
        return OPAL_ERR_TEMP_OUT_OF_RESOURCE;
    }
    strncpy(buf, pwd, size);
    return OPAL_SUCCESS;
}
Example #7
0
/* When working in this function, ALWAYS jump to "cleanup" if
 * you encounter an error so that orterun will be woken up and
 * the job can cleanly terminate
 */
static void launch_daemons(int fd, short args, void *cbdata)
{
    orte_job_map_t *map = NULL;
    orte_app_context_t *app;
    orte_node_t *node;
    int proc_vpid_index;
    char *param;
    char **env = NULL;
    char *var;
    char **argv = NULL;
    char **nodeargv;
    int argc = 0;
    int rc;
    orte_std_cntr_t i; 
    char *bin_base = NULL, *lib_base = NULL;
    tm_event_t *tm_events = NULL;
    tm_task_id *tm_task_ids = NULL;
    bool failed_launch = true;
    mode_t current_umask;
    char *nodelist;
    char* vpid_string;
    orte_job_t *daemons, *jdata;
    orte_state_caddy_t *state = (orte_state_caddy_t*)cbdata;
    int32_t launchid, *ldptr;
    char *prefix_dir = NULL;

    jdata = state->jdata;

    /* if we are launching debugger daemons, then just go
     * do it - no new daemons will be launched
     */
    if (ORTE_FLAG_TEST(state->jdata, ORTE_JOB_FLAG_DEBUGGER_DAEMON)) {
        jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
        OBJ_RELEASE(state);
        return;
    }

    /* setup the virtual machine */
    daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid);
    if (ORTE_SUCCESS != (rc = orte_plm_base_setup_virtual_machine(jdata))) {
        ORTE_ERROR_LOG(rc);
        goto cleanup;
    }

    /* if we don't want to launch, then don't attempt to
     * launch the daemons - the user really wants to just
     * look at the proposed process map
     */
    if (orte_do_not_launch) {
        /* set the state to indicate the daemons reported - this
         * will trigger the daemons_reported event and cause the
         * job to move to the following step
         */
        jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
        OBJ_RELEASE(state);
        return;
    }
    
    /* Get the map for this job */
    if (NULL == (map = daemons->map)) {
        ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
        rc = ORTE_ERR_NOT_FOUND;
        goto cleanup;
    }

    if (0 == map->num_new_daemons) {
        /* set the state to indicate the daemons reported - this
         * will trigger the daemons_reported event and cause the
         * job to move to the following step
         */
        jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
        ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_DAEMONS_REPORTED);
        OBJ_RELEASE(state);
        return;
    }
    
    OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                         "%s plm:tm: launching vm",
                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
    
    /* Allocate a bunch of TM events to use for tm_spawn()ing */
    tm_events = malloc(sizeof(tm_event_t) * map->num_new_daemons);
    if (NULL == tm_events) {
        rc = ORTE_ERR_OUT_OF_RESOURCE;
        ORTE_ERROR_LOG(rc);
        goto cleanup;
    }
    tm_task_ids = malloc(sizeof(tm_task_id) * map->num_new_daemons);
    if (NULL == tm_task_ids) {
        rc = ORTE_ERR_OUT_OF_RESOURCE;
        ORTE_ERROR_LOG(rc);
        goto cleanup;
    }

    /* add the daemon command (as specified by user) */
    orte_plm_base_setup_orted_cmd(&argc, &argv);

    /* create a list of nodes in this launch */
    nodeargv = NULL;
    for (i = 0; i < map->nodes->size; i++) {
        if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
            continue;
        }
        
        /* if this daemon already exists, don't launch it! */
        if (ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_DAEMON_LAUNCHED)) {
            continue;
        }
        
        /* add to list */
        opal_argv_append_nosize(&nodeargv, node->name);
    }
    nodelist = opal_argv_join(nodeargv, ',');
    opal_argv_free(nodeargv);
    
    /* Add basic orted command line options */
    orte_plm_base_orted_append_basic_args(&argc, &argv, "tm",
                                          &proc_vpid_index,
                                          nodelist);
    free(nodelist);
    
    if (0 < opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
        param = opal_argv_join(argv, ' ');
        OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                             "%s plm:tm: final top-level argv:\n\t%s",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                             (NULL == param) ? "NULL" : param));
        if (NULL != param) free(param);
    }

    if (!connected) {
        if (ORTE_SUCCESS != plm_tm_connect()) {
            goto cleanup;
        }
        connected = true;
    }

    /* Figure out the basenames for the libdir and bindir.  There is a
       lengthy comment about this in plm_rsh_module.c explaining all
       the rationale for how / why we're doing this. */
    lib_base = opal_basename(opal_install_dirs.libdir);
    bin_base = opal_basename(opal_install_dirs.bindir);

    /* setup environment */
    env = opal_argv_copy(orte_launch_environ);

    /* enable local launch by the orteds */
    (void) mca_base_var_env_name ("plm", &var);
    opal_setenv(var, "rsh", true, &env);
    free(var);
    
    /* add our umask -- see big note in orted.c */
    current_umask = umask(0);
    umask(current_umask);
    (void)asprintf(&var, "0%o", current_umask);
    opal_setenv("ORTE_DAEMON_UMASK_VALUE", var, true, &env);
    free(var);
    
    /* If we have a prefix, then modify the PATH and
       LD_LIBRARY_PATH environment variables. We only allow
       a single prefix to be specified. Since there will
       always be at least one app_context, we take it from
       there
    */
    app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, 0);
    orte_get_attribute(&app->attributes, ORTE_APP_PREFIX_DIR, (void**)&prefix_dir, OPAL_STRING);
    if (NULL != prefix_dir) {
        char *newenv;
        
        for (i = 0; NULL != env && NULL != env[i]; ++i) {
            /* Reset PATH */
            if (0 == strncmp("PATH=", env[i], 5)) {
                (void)asprintf(&newenv, "%s/%s:%s", 
                               prefix_dir, bin_base, env[i] + 5);
                OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                                     "%s plm:tm: resetting PATH: %s",
                                     ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                                     newenv));
                opal_setenv("PATH", newenv, true, &env);
                free(newenv);
            } 
            
            /* Reset LD_LIBRARY_PATH */
            else if (0 == strncmp("LD_LIBRARY_PATH=", env[i], 16)) {
                (void)asprintf(&newenv, "%s/%s:%s", 
                               prefix_dir, lib_base, env[i] + 16);
                OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                                     "%s plm:tm: resetting LD_LIBRARY_PATH: %s",
                                     ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                                     newenv));
                opal_setenv("LD_LIBRARY_PATH", newenv, true, &env);
                free(newenv);
            } 
        }
        free(prefix_dir);
    }
    
    /* Iterate through each of the nodes and spin
     * up a daemon.
     */
    ldptr = &launchid;
    for (i = 0; i < map->nodes->size; i++) {
        if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
            continue;
        }
        /* if this daemon already exists, don't launch it! */
        if (ORTE_FLAG_TEST(node, ORTE_NODE_FLAG_DAEMON_LAUNCHED)) {
            continue;
        }
 
        OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                             "%s plm:tm: launching on node %s",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                             node->name));
        
        /* setup process name */
        rc = orte_util_convert_vpid_to_string(&vpid_string, node->daemon->name.vpid);
        if (ORTE_SUCCESS != rc) {
            opal_output(0, "plm:tm: unable to get daemon vpid as string");
            exit(-1);
        }
        free(argv[proc_vpid_index]);
        argv[proc_vpid_index] = strdup(vpid_string);
        free(vpid_string);
        
        /* exec the daemon */
        if (0 < opal_output_get_verbosity(orte_plm_base_framework.framework_output)) {
            param = opal_argv_join(argv, ' ');
            OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                                 "%s plm:tm: executing:\n\t%s",
                                 ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                                 (NULL == param) ? "NULL" : param));
            if (NULL != param) free(param);
        }
        
        launchid = 0;
        if (!orte_get_attribute(&node->attributes, ORTE_NODE_LAUNCH_ID, (void**)&ldptr, OPAL_INT32)) {
            orte_show_help("help-plm-tm.txt", "tm-spawn-failed", true, argv[0], node->name, 0);
            rc = ORTE_ERROR;
            goto cleanup;
        }
        rc = tm_spawn(argc, argv, env, launchid, tm_task_ids + launched, tm_events + launched);
        if (TM_SUCCESS != rc) {
            orte_show_help("help-plm-tm.txt", "tm-spawn-failed", true, argv[0], node->name, launchid);
            rc = ORTE_ERROR;
            goto cleanup;
        }
        
        launched++;
    }

    /* indicate that the daemons for this job were launched */
    state->jdata->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;
    daemons->state = ORTE_JOB_STATE_DAEMONS_LAUNCHED;

    /* flag that launch was successful, so far as we currently know */
    failed_launch = false;

    OPAL_OUTPUT_VERBOSE((1, orte_plm_base_framework.framework_output,
                         "%s plm:tm:launch: finished spawning orteds",
                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));

 cleanup:
    /* cleanup */
    OBJ_RELEASE(state);

    /* check for failed launch - if so, force terminate */
    if (failed_launch) {
        ORTE_ACTIVATE_JOB_STATE(daemons, ORTE_JOB_STATE_FAILED_TO_START);
    }
}
Example #8
0
static int setup_fork(orte_job_t *jdata,
                      orte_app_context_t *app)
{
    int i;
    bool takeus = false;
    char *p, *t2;
    char dir[MAXPATHLEN];

    if (NULL != orte_schizo_base.personalities) {
        /* see if we are included */
        for (i=0; NULL != jdata->personality[i]; i++) {
            if (0 == strcmp(jdata->personality[i], "singularity")) {
                takeus = true;
                break;
            }
        }
    }
    if (!takeus) {
        /* even if they didn't specify, check to see if
         * this involves a singularity container */
        if (0 != strcmp(app->argv[0],"singularity") &&
            0 != strcmp(app->argv[0],"sapprun") &&
            NULL == strstr(app->argv[0], ".sapp")) {
            /* guess not! */
            return ORTE_ERR_TAKE_NEXT_OPTION;
        }
    }

    /* set the singularity cache dir, unless asked not to do so */
    if (!orte_get_attribute(&app->attributes, ORTE_APP_NO_CACHEDIR, NULL, OPAL_BOOL)) {
        opal_setenv("SINGULARITY_CACHEDIR", orte_process_info.job_session_dir, true, &app->env);
        opal_setenv("SINGULARITY_CACHEDIR", orte_process_info.job_session_dir, true, &environ);
    }

    /* save our current directory */
    getcwd(dir, sizeof(dir));

    /* change to the working directory for this context */
    chdir(app->cwd);

    /* if the app contains .sapp, then we need to strip that
     * extension so singularity doesn't bark at us */
    if (NULL != strstr(app->argv[0], ".sapp")) {
        /* ensure the app is installed */
        opal_output_verbose(1, orte_schizo_base_framework.framework_output,
                            "%s schizo:singularity: installing app %s",
                            ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), app->argv[0]);
        t2 = opal_basename(app->argv[0]);
        p = strstr(t2, ".sapp");
        *p = '\0'; // strip the extension
        if (0 < opal_output_get_verbosity(orte_schizo_base_framework.framework_output)) {
            (void)asprintf(&p, "singularity -vv install --runkey %s %s", t2, app->argv[0]);
        } else {
            (void)asprintf(&p, "singularity --quiet install --runkey %s %s", t2, app->argv[0]);
        }
        system(p);
        free(p);
        free(app->argv[0]);
        app->argv[0] = t2;
    }

    /* ensure that we use "singularity run" to execute this app */
    if (0 != strcmp(app->app, "singularity")) {
        opal_output_verbose(1, orte_schizo_base_framework.framework_output,
                            "%s schizo:singularity: adding singularity cmd",
                            ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
        /* change the app to the "singularity" command */
        free(app->app);
        app->app = strdup("singularity");
        opal_argv_prepend_nosize(&app->argv, "run");
        if (0 < opal_output_get_verbosity(orte_schizo_base_framework.framework_output)) {
            opal_argv_prepend_nosize(&app->argv, "-vv");
        } else {
            opal_argv_prepend_nosize(&app->argv, "--quiet");
        }
        opal_argv_prepend_nosize(&app->argv, "singularity");
    }

    /* return to the original directory */
    chdir(dir);

    return ORTE_SUCCESS;
}
Example #9
0
int
main(int argc, char *argv[])
{
    int exit_status = 0, ret, flags = 0, i;
    int exec_argc = 0, user_argc = 0;
    char **exec_argv = NULL, **user_argv = NULL;
    char *exec_command, *base_argv0 = NULL;
    bool disable_flags = true;
    bool real_flag = false;

    if (OPAL_SUCCESS != (ret = opal_init_util())) {
        return ret;
    }

    /****************************************************
     *
     * Setup compiler information
     *
     ****************************************************/

    base_argv0 = opal_basename(argv[0]);
#if defined(EXEEXT)
    if( 0 != strlen(EXEEXT) ) {
        char extension[] = EXEEXT;
        char* temp = strstr( base_argv0, extension );
        char* old_match = temp;
        while( NULL != temp ) {
            old_match = temp;
            temp = strstr( temp + 1, extension );
        }
        /* Only if there was a match of .exe, erase the last occurence of .exe */
        if ( NULL != old_match ) {
            *old_match = '\0';
        }
    }
#endif  /* defined(EXEEXT) */

    if (OPAL_SUCCESS != (ret = data_init(base_argv0))) {
        fprintf(stderr, "Error parsing data file %s: %s\n", base_argv0, opal_strerror(ret));
        return ret;
    }

    for (i = 1 ; i < argc && user_data_idx < 0 ; ++i) {
        user_data_idx = find_options_index(argv[i]);
    }
    /* if we didn't find a match, look for the NULL (base case) options */
    if (user_data_idx < 0) {
        user_data_idx = default_data_idx;
    }
    /* if we still didn't find a match, abort */
    if (user_data_idx < 0) {
        char *flat = opal_argv_join(argv, ' ');
        opal_show_help("help-opal-wrapper.txt", "no-options-support", true,
                       base_argv0, flat, NULL);
        free(flat);
        exit(1);
    }

    /* compiler */
    load_env_data(options_data[user_data_idx].project_short, options_data[user_data_idx].compiler_env, &options_data[user_data_idx].compiler);

    /* preprocessor flags */
    load_env_data_argv(options_data[user_data_idx].project_short, "CPPFLAGS", &options_data[user_data_idx].preproc_flags);

    /* compiler flags */
    load_env_data_argv(options_data[user_data_idx].project_short, options_data[user_data_idx].compiler_flags_env,
                       &options_data[user_data_idx].comp_flags);

    /* linker flags */
    load_env_data_argv(options_data[user_data_idx].project_short, "LDFLAGS", &options_data[user_data_idx].link_flags);

    /* libs */
    load_env_data_argv(options_data[user_data_idx].project_short, "LIBS", &options_data[user_data_idx].libs);


    /****************************************************
     *
     * Sanity Checks
     *
     ****************************************************/
    
    if (NULL != options_data[user_data_idx].req_file) {
        /* make sure the language is supported */
        if (0 == strcmp(options_data[user_data_idx].req_file, "not supported")) {
            opal_show_help("help-opal-wrapper.txt", "no-language-support", true,
                           options_data[user_data_idx].language, base_argv0, NULL);
            exit_status = 1;
            goto cleanup;
        }

        if (options_data[user_data_idx].req_file[0] != '\0') {
            char *filename;
            struct stat buf;
            filename = opal_os_path( false, options_data[user_data_idx].path_libdir, options_data[user_data_idx].req_file, NULL );
            if (0 != stat(filename, &buf)) {
                opal_show_help("help-opal-wrapper.txt", "file-not-found", true,
                               base_argv0, options_data[user_data_idx].req_file, options_data[user_data_idx].language, NULL);
            }
        }
    }

    /****************************************************
     *
     * Parse user flags
     *
     ****************************************************/
    flags = COMP_WANT_COMMAND|COMP_WANT_PREPROC|
        COMP_WANT_COMPILE|COMP_WANT_LINK;

    user_argv = opal_argv_copy(argv + 1);
    user_argc = opal_argv_count(user_argv);

    for (i = 0 ; i < user_argc ; ++i) {
        if (0 == strncmp(user_argv[i], "-showme", strlen("-showme")) ||
            0 == strncmp(user_argv[i], "--showme", strlen("--showme")) ||
            0 == strncmp(user_argv[i], "-show", strlen("-show")) ||
            0 == strncmp(user_argv[i], "--show", strlen("--show"))) {
            bool done_now = false;

            /* check for specific things we want to see.  First three
               still invoke all the building routines.  Last set want
               to parse out certain flags, so we don't go through the
               normal build routine - skip to cleanup. */
            if (0 == strncmp(user_argv[i], "-showme:command", strlen("-showme:command")) ||
                0 == strncmp(user_argv[i], "--showme:command", strlen("--showme:command"))) {
                flags = COMP_WANT_COMMAND;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:compile", strlen("-showme:compile")) ||
                0 == strncmp(user_argv[i], "--showme:compile", strlen("--showme:compile"))) {
                flags = COMP_WANT_PREPROC|COMP_WANT_COMPILE;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:link", strlen("-showme:link")) ||
                       0 == strncmp(user_argv[i], "--showme:link", strlen("--showme:link"))) {
                flags = COMP_WANT_COMPILE|COMP_WANT_LINK;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:incdirs", strlen("-showme:incdirs")) ||
                       0 == strncmp(user_argv[i], "--showme:incdirs", strlen("--showme:incdirs"))) {
                print_flags(options_data[user_data_idx].preproc_flags, OPAL_INCLUDE_FLAG);
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:libdirs", strlen("-showme:libdirs")) ||
                       0 == strncmp(user_argv[i], "--showme:libdirs", strlen("--showme:libdirs"))) {
                print_flags(options_data[user_data_idx].link_flags, OPAL_LIBDIR_FLAG);
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:libs", strlen("-showme:libs")) ||
                       0 == strncmp(user_argv[i], "--showme:libs", strlen("--showme:libs"))) {
                print_flags(options_data[user_data_idx].libs, "-l");
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:version", strlen("-showme:version")) ||
                       0 == strncmp(user_argv[i], "--showme:version", strlen("--showme:version"))) {
                opal_show_help("help-opal-wrapper.txt", "version", false,
                               argv[0], options_data[user_data_idx].project, options_data[user_data_idx].version, options_data[user_data_idx].language, NULL);
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:", strlen("-showme:")) ||
                       0 == strncmp(user_argv[i], "--showme:", strlen("--showme:"))) {
                opal_show_help("help-opal-wrapper.txt", "usage", true,
                               argv[0], options_data[user_data_idx].project, NULL);
                goto cleanup;
            }

            flags |= (COMP_DRY_RUN|COMP_SHOW_ERROR);
            /* remove element from user_argv */
            opal_argv_delete(&user_argc, &user_argv, i, 1);
            --i;

            if (done_now) {
                disable_flags = false;
                break;
            }

        } else if (0 == strcmp(user_argv[i], "-c")) {
            flags &= ~COMP_WANT_LINK;
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-E") || 
                   0 == strcmp(user_argv[i], "-M")) {
            flags &= ~(COMP_WANT_COMPILE | COMP_WANT_LINK);
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-S")) {
            flags &= ~COMP_WANT_LINK;
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-lpmpi")) {
            flags |= COMP_WANT_PMPI;

            /* remove element from user_argv */
            opal_argv_delete(&user_argc, &user_argv, i, 1);
            --i;
        } else if ('-' != user_argv[i][0]) {
            disable_flags = false;
            flags |= COMP_SHOW_ERROR;
            real_flag = true;
        } else { 
            /* if the option flag is one that we use to determine
               which set of compiler data to use, don't count it as a
               real option */
            if (find_options_index(user_argv[i]) < 0) {
                real_flag = true;
            }
        }
    }

    /* clear out the want_flags if we got no arguments not starting
       with a - (dash) and -showme wasn't given OR -showme was given
       and we had at least one more non-showme argument that started
       with a - (dash) and no other non-dash arguments.  Some examples:

       opal_wrapper                : clear our flags
       opal_wrapper -v             : clear our flags
       opal_wrapper -E a.c         : don't clear our flags
       opal_wrapper a.c            : don't clear our flags
       opal_wrapper -showme        : don't clear our flags
       opal_wrapper -showme -v     : clear our flags
       opal_wrapper -showme -E a.c : don't clear our flags
       opal_wrapper -showme a.c    : don't clear our flags
    */
    if (disable_flags && !((flags & COMP_DRY_RUN) && !real_flag)) {
        flags &= ~(COMP_WANT_PREPROC|COMP_WANT_COMPILE|COMP_WANT_LINK);
    }

#if !OMPI_ENABLE_MPI_PROFILING
    /* sanity check */
    if (flags & COMP_WANT_PMPI) {
	    opal_show_help("help-opal-wrapper.txt", "no-profiling-support", true,
		               argv[0], NULL);
    }
#endif


    /****************************************************
     *
     * Assemble the command line
     *
     ****************************************************/

    /* compiler (may be multiple arguments, so split) */
    if (flags & COMP_WANT_COMMAND) {
        exec_argv = opal_argv_split(options_data[user_data_idx].compiler, ' ');
        exec_argc = opal_argv_count(exec_argv);
    } else {
        exec_argv = (char **) malloc(sizeof(char*));
        exec_argv[0] = NULL;
        exec_argc = 0;
    }

    /* Per https://svn.open-mpi.org/trac/ompi/ticket/2201, add all the
       user arguments before anything else. */
    opal_argv_insert(&exec_argv, exec_argc, user_argv);
    exec_argc = opal_argv_count(exec_argv);

    /* preproc flags */
    if (flags & COMP_WANT_PREPROC) {
        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].preproc_flags);
        exec_argc = opal_argv_count(exec_argv);
    }

    /* compiler flags */
    if (flags & COMP_WANT_COMPILE) {
        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].comp_flags);
        /* Deal with languages like Fortran 90 that have special
           places and flags for modules or whatever */
        if (options_data[user_data_idx].module_option != NULL) {
            char *line;
            asprintf(&line, "%s%s", options_data[user_data_idx].module_option, options_data[user_data_idx].path_libdir);
            opal_argv_append_nosize(&exec_argv, line);
            free(line);
        }
        exec_argc = opal_argv_count(exec_argv);
    }

    /* link flags and libs */
    if (flags & COMP_WANT_LINK) {
        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].link_flags);
        exec_argc = opal_argv_count(exec_argv);

        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].libs);
        exec_argc = opal_argv_count(exec_argv);
    }


    /****************************************************
     *
     * Execute the command
     *
     ****************************************************/

    if (flags & COMP_DRY_RUN) {
        exec_command = opal_argv_join(exec_argv, ' ');
        printf("%s\n", exec_command);
    } else {
        char *tmp;

#if 0
        exec_command = opal_argv_join(exec_argv, ' ');
        printf("command: %s\n", exec_command);
#endif

        tmp = opal_path_findv(exec_argv[0], 0, environ, NULL);
        if (NULL == tmp) {
            opal_show_help("help-opal-wrapper.txt", "no-compiler-found", true,
                           exec_argv[0], NULL);
            errno = 0;
            exit_status = 1;
        }  else {
            int status;

            free(exec_argv[0]);
            exec_argv[0] = tmp;
            ret = opal_few(exec_argv, &status);
            exit_status = WIFEXITED(status) ? WEXITSTATUS(status) :
                              (WIFSIGNALED(status) ? WTERMSIG(status) :
                                  (WIFSTOPPED(status) ? WSTOPSIG(status) : 255));
            if( (OPAL_SUCCESS != ret) || ((0 != exit_status) && (flags & COMP_SHOW_ERROR)) ) {
                char* exec_command = opal_argv_join(exec_argv, ' ');
                if( OPAL_SUCCESS != ret ) {
                    opal_show_help("help-opal-wrapper.txt", "spawn-failed", true,
                                   exec_argv[0], strerror(status), exec_command, NULL);
                } else {
#if 0
                    opal_show_help("help-opal-wrapper.txt", "compiler-failed", true,
                                   exec_argv[0], exit_status, exec_command, NULL);
#endif
                }
                free(exec_command);
            }
        }
    }

    /****************************************************
     *
     * Cleanup
     *
     ****************************************************/
 cleanup:

    opal_argv_free(exec_argv);
    opal_argv_free(user_argv);
    if (NULL != base_argv0) free(base_argv0);

    if (OPAL_SUCCESS != (ret = data_finalize())) {
        return ret;
    }

    if (OPAL_SUCCESS != (ret = opal_finalize_util())) {
        return ret;
    }

    return exit_status;
}
Example #10
0
int
main(int argc, char *argv[])
{
    int ret, exit_status = ORTE_SUCCESS;
    pid_t child_pid = 0;
    orte_sstore_base_global_snapshot_info_t *snapshot = NULL;
    char *basedir = NULL;
    char *tmp_str = NULL;

    /***************
     * Initialize
     ***************/
    if (ORTE_SUCCESS != (ret = initialize(argc, argv))) {
        exit_status = ret;
        goto cleanup;
    }

    snapshot = OBJ_NEW(orte_sstore_base_global_snapshot_info_t);

    if( opal_path_is_absolute(orte_restart_globals.snapshot_ref) ) {
        basedir = opal_dirname(orte_restart_globals.snapshot_ref);
        tmp_str = opal_basename(orte_restart_globals.snapshot_ref);
        free(orte_restart_globals.snapshot_ref);
        orte_restart_globals.snapshot_ref = strdup(tmp_str);
        free(tmp_str);
        tmp_str = NULL;
    } else if( NULL != strchr(orte_restart_globals.snapshot_ref, '/') ) {
        basedir = opal_dirname(orte_restart_globals.snapshot_ref);
        tmp_str = opal_basename(orte_restart_globals.snapshot_ref);
        free(orte_restart_globals.snapshot_ref);
        orte_restart_globals.snapshot_ref = strdup(tmp_str);
        free(tmp_str);
        tmp_str = NULL;
    } else {
        basedir = NULL; /* Use MCA parameter */
    }

    /*
     * Note: If the seq # passed is -1, then the largest seq # is selected,
     *       ow the seq # requested is selected if available
     * 'basedir': Snapshot Base location to look in. If NULL then MCA parameter is used
     */
    if( ORTE_SUCCESS != (ret = orte_sstore.request_restart_handle(&(snapshot->ss_handle),
                                                                  basedir,
                                                                  orte_restart_globals.snapshot_ref,
                                                                  orte_restart_globals.seq_number,
                                                                  snapshot))) {
        opal_show_help("help-orte-restart.txt", "invalid_filename", true,
                       orte_restart_globals.snapshot_ref);
        exit_status = ret;
        goto cleanup;
    }
    orte_restart_globals.seq_number = snapshot->seq_num;

    if(orte_restart_globals.info_only ) {
        if (ORTE_SUCCESS != (ret = snapshot_info(snapshot))) {
            exit_status = ret;
            goto cleanup;
        }
        exit_status = ORTE_SUCCESS;
        goto cleanup;
    }

    /******************************
     * Create the app file to use with mpirun/orterun
     ******************************/
    if( ORTE_SUCCESS != (ret = create_appfile(snapshot) ) ) {
        exit_status = ret;
        goto cleanup;
    }

    if( orte_restart_globals.app_only ) {
        printf("Created Appfile:\n\t%s\n", orte_restart_globals.appfile);
        exit_status = ORTE_SUCCESS;
        goto cleanup;
    }

    /******************************
     * Restart in this process [mpirun/orterun]
     ******************************/
    if( orte_restart_globals.verbose ) {
        opal_output_verbose(10, orte_restart_globals.output,
                            "Restarting from file (%s)",
                            orte_restart_globals.snapshot_ref);
        
        if( orte_restart_globals.forked ) {
            opal_output_verbose(10, orte_restart_globals.output,
                                "\t Forking off a child");
        } else {
            opal_output_verbose(10, orte_restart_globals.output,
                                "\t Exec in self");
        }
    }

    if( ORTE_SUCCESS != (ret = spawn_children(snapshot, &child_pid)) ) {
        opal_show_help("help-orte-restart.txt", "restart_cmd_failure", true,
                       orte_restart_globals.snapshot_ref, ret);
        exit_status = ret;
        goto cleanup;
    }

    /***************
     * Cleanup
     ***************/
 cleanup:
    if( NULL != basedir ) {
        free(basedir);
        basedir = NULL;
    }
    if( NULL != tmp_str ) {
        free(tmp_str);
        tmp_str = NULL;
    }
    if( NULL != snapshot ) {
        OBJ_RELEASE(snapshot);
        snapshot = NULL;
    }

    if (OPAL_SUCCESS != (ret = finalize())) {
        return ret;
    }

    return exit_status;
}
Example #11
0
/*
 * This function makes a call to "ps" to find out the processes that
 * are running on this node.  It then attempts to kill off any orteds
 * and orteruns that are not related to this job.
 */
static
void kill_procs(void) {
    int ortedpid;
    char *fullprocname;
    char *procname;
    char *pidstr;
    char *user;
    int procpid;
    FILE *psfile;
    char *inputline;
    char *this_user;
    int uid;
    char *separator = " \t";  /* output can be delimited by space or tab */

    /*
     * This is the command that is used to get the information about
     * all the processes that are running.  The output looks like the
     * following:
     * COMMAND    PID     USER
     * tcsh     12556    rolfv
     * ps       14424    rolfv
     * etc.
     * Currently, we do not make use of the USER field, but we may later
     * on so we grab it also.
     */

    /*
     * The configure determines if there is a valid ps command for us to
     * use.  If it is set to unknown, then we skip this section.
     */
    char command[] = ORTE_CLEAN_PS_CMD;
    if (0 == strcmp("unknown", command)) {
        return;
    }

    if (orte_clean_globals.verbose) {
        fprintf(stderr, "orte-clean: killing any lingering procs\n");
    }

    /*
     * Get our parent pid which is the pid of the orted.
     */
    ortedpid = getppid();

    /* get the name of the user */
    uid = getuid();
    asprintf(&this_user, "%d", uid);

    /*
     * There is a race condition here.  The problem is that we are looking
     * for any processes named orted.  However, one may erroneously find more
     * orteds then there really are because the orted is doing a series of
     * fork/execs. If we run with more than one orte-clean on a node, then
     * one of the orte-cleans may catch the other one while it has forked,
     * but not exec'ed.  It will therefore kill an orte-clean.  Now one
     * can argue it is silly to run more than one orte-clean on a node, and
     * this is true.  We will have to figure out how to prevent this.  For
     * now, we use a big hammer and just sleep a second to decrease the
     * probability.
     */
    sleep(1);

    psfile = popen(command, "r");
    /*
     * Read the first line of the output.  We just throw it away
     * as it is the header consisting of the words COMMAND, PID and
     * USER.
     */
    if (NULL == (inputline = orte_getline(psfile))) {
        free(this_user);
        pclose(psfile);
        return;
    }
    free(inputline);  /* dump the header line */

    while (NULL != (inputline = orte_getline(psfile))) {

        /* The three fields are typically seperated by spaces */
        fullprocname = strtok(inputline, separator);
        pidstr = strtok(NULL, separator);
        user = strtok(NULL, separator);

        if (orte_clean_globals.debug) {
            fprintf(stdout, "\norte-clean: user(pid)=%s, me=%s\n",
                    user, this_user);
        }

        /* If the user is not us, and the user is not root, then skip
         * further checking.  If the user is root, then continue on as
         * we want root to kill off everybody. */
        if ((0 != strcmp(user, this_user)) && (0 != strcmp("root", this_user))) {
            /* not us */
            free(inputline);
            continue;
        }

        procpid = atoi(pidstr);
        procname = opal_basename(fullprocname);
        if (orte_clean_globals.debug) {
            fprintf(stdout, "orte-clean: fullname=%s, basename=%s, pid=%d\n",
                    fullprocname, procname, procpid);
        }

        /*
         * Look for any orteds that are not our parent and attempt to
         * kill them.  We currently do not worry whether we are the
         * owner or not.  If we are not, we will just fail to send
         * the signal and that is OK.  This also allows a root process
         * to kill all orteds.
         *
         * NOTE: need to also look for "(orted)" as a non-active
         * proc is sometimes reported that way
         */
        if (0 == strncmp("orted", procname, strlen("orted")) ||
            0 == strncmp("(orted)", procname, strlen("(orted)"))) {
            if (procpid != ortedpid) {
                if (orte_clean_globals.verbose) {
                    fprintf(stderr, "orte-clean: found potential rogue orted process"
                            " (pid=%d,user=%s), sending SIGKILL...\n",
                            procpid, user);
                }
                /*
                 * We ignore the return code here as we do not really
                 * care whether this worked or not.
                 */
                (void)kill(procpid, SIGKILL);
            }
        }

        /*
         * Now check for any orteruns.
         */
        if (0 == strncmp("orterun", procname, strlen("orterun")) ||
            0 == strncmp("mpirun", procname, strlen("mpirun"))) {
            /* if we are on the same node as the HNP, then the ortedpid
             * is the same as that of our orterun, so don't kill it
             */
            if (procpid != ortedpid) {
                if (orte_clean_globals.verbose) {
                    fprintf(stderr, "orte-clean: found potential rogue orterun process"
                            " (pid=%d,user=%s), sending SIGKILL...\n",
                            procpid, user);

                }
                /* if we are a singleton, check the hnp_pid as well */
                if (ORTE_PROC_IS_SINGLETON) {
                    if (procpid != orte_process_info.hnp_pid) {
                        (void)kill(procpid, SIGKILL);
                    }
                } else {
                    /* We ignore the return code here as we do not really
                     * care whether this worked or not.
                     */
                    (void)kill(procpid, SIGKILL);
                }
            }
        }
        free(inputline);
	free(procname);
    }
    free(this_user);
    pclose(psfile);
    return;
}
Example #12
0
int
main(int argc, char *argv[])
{
    int exit_status = 0, ret, flags = 0, i;
    int exec_argc = 0, user_argc = 0;
    char **exec_argv = NULL, **user_argv = NULL;
    char *exec_command, *base_argv0 = NULL;
    bool disable_flags = true;
    bool real_flag = false;

    if (OPAL_SUCCESS != (ret = opal_init_util(&argc, &argv))) {
        return ret;
    }

    /****************************************************
     *
     * Setup compiler information
     *
     ****************************************************/

    base_argv0 = opal_basename(argv[0]);
#if defined(EXEEXT)
    if( 0 != strlen(EXEEXT) ) {
        char extension[] = EXEEXT;
        char* temp = strstr( base_argv0, extension );
        char* old_match = temp;
        while( NULL != temp ) {
            old_match = temp;
            temp = strstr( temp + 1, extension );
        }
        /* Only if there was a match of .exe, erase the last occurence of .exe */
        if ( NULL != old_match ) {
            *old_match = '\0';
        }
    }
#endif  /* defined(EXEEXT) */

    if (OPAL_SUCCESS != (ret = data_init(base_argv0))) {
        fprintf(stderr, "Error parsing data file %s: %s\n", base_argv0, opal_strerror(ret));
        return ret;
    }

    for (i = 1 ; i < argc && user_data_idx < 0 ; ++i) {
        user_data_idx = find_options_index(argv[i]);
    }
    /* if we didn't find a match, look for the NULL (base case) options */
    if (user_data_idx < 0) {
        user_data_idx = default_data_idx;
    }
    /* if we still didn't find a match, abort */
    if (user_data_idx < 0) {
        char *flat = opal_argv_join(argv, ' ');
        opal_show_help("help-opal-wrapper.txt", "no-options-support", true,
                       base_argv0, flat, NULL);
        free(flat);
        exit(1);
    }

    /* compiler */
    load_env_data(options_data[user_data_idx].project_short, options_data[user_data_idx].compiler_env, &options_data[user_data_idx].compiler);

    /* preprocessor flags */
    load_env_data_argv(options_data[user_data_idx].project_short, "CPPFLAGS", &options_data[user_data_idx].preproc_flags);

    /* compiler flags */
    load_env_data_argv(options_data[user_data_idx].project_short, options_data[user_data_idx].compiler_flags_env,
                       &options_data[user_data_idx].comp_flags);

    /* linker flags */
    load_env_data_argv(options_data[user_data_idx].project_short, "LDFLAGS", &options_data[user_data_idx].link_flags);

    /* libs */
    load_env_data_argv(options_data[user_data_idx].project_short, "LIBS", &options_data[user_data_idx].libs);


    /****************************************************
     *
     * Sanity Checks
     *
     ****************************************************/
    
    if (NULL != options_data[user_data_idx].req_file) {
        /* make sure the language is supported */
        if (0 == strcmp(options_data[user_data_idx].req_file, "not supported")) {
            opal_show_help("help-opal-wrapper.txt", "no-language-support", true,
                           options_data[user_data_idx].language, base_argv0, NULL);
            exit_status = 1;
            goto cleanup;
        }

        if (options_data[user_data_idx].req_file[0] != '\0') {
            char *filename;
            struct stat buf;
            filename = opal_os_path( false, options_data[user_data_idx].path_libdir, options_data[user_data_idx].req_file, NULL );
            if (0 != stat(filename, &buf)) {
                opal_show_help("help-opal-wrapper.txt", "file-not-found", true,
                               base_argv0, options_data[user_data_idx].req_file, options_data[user_data_idx].language, NULL);
            }
        }
    }

    /****************************************************
     *
     * Parse user flags
     *
     ****************************************************/
    flags = COMP_WANT_COMMAND|COMP_WANT_PREPROC|
        COMP_WANT_COMPILE|COMP_WANT_LINK;

    user_argv = opal_argv_copy(argv + 1);
    user_argc = opal_argv_count(user_argv);

    for (i = 0 ; i < user_argc ; ++i) {
        if (0 == strncmp(user_argv[i], "-showme", strlen("-showme")) ||
            0 == strncmp(user_argv[i], "--showme", strlen("--showme")) ||
            0 == strncmp(user_argv[i], "-show", strlen("-show")) ||
            0 == strncmp(user_argv[i], "--show", strlen("--show"))) {
            bool done_now = false;

            /* check for specific things we want to see.  First three
               still invoke all the building routines.  Last set want
               to parse out certain flags, so we don't go through the
               normal build routine - skip to cleanup. */
            if (0 == strncmp(user_argv[i], "-showme:command", strlen("-showme:command")) ||
                0 == strncmp(user_argv[i], "--showme:command", strlen("--showme:command"))) {
                flags = COMP_WANT_COMMAND;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:compile", strlen("-showme:compile")) ||
                0 == strncmp(user_argv[i], "--showme:compile", strlen("--showme:compile"))) {
                flags = COMP_WANT_PREPROC|COMP_WANT_COMPILE;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:link", strlen("-showme:link")) ||
                       0 == strncmp(user_argv[i], "--showme:link", strlen("--showme:link"))) {
                flags = COMP_WANT_COMPILE|COMP_WANT_LINK;
                /* we know what we want, so don't process any more args */
                done_now = true;
            } else if (0 == strncmp(user_argv[i], "-showme:incdirs", strlen("-showme:incdirs")) ||
                       0 == strncmp(user_argv[i], "--showme:incdirs", strlen("--showme:incdirs"))) {
                print_flags(options_data[user_data_idx].preproc_flags, OPAL_INCLUDE_FLAG);
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:libdirs", strlen("-showme:libdirs")) ||
                       0 == strncmp(user_argv[i], "--showme:libdirs", strlen("--showme:libdirs"))) {
                print_flags(options_data[user_data_idx].link_flags, OPAL_LIBDIR_FLAG);
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:libs", strlen("-showme:libs")) ||
                       0 == strncmp(user_argv[i], "--showme:libs", strlen("--showme:libs"))) {
                print_flags(options_data[user_data_idx].libs, "-l");
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:version", strlen("-showme:version")) ||
                       0 == strncmp(user_argv[i], "--showme:version", strlen("--showme:version"))) {
                char * str;
                str = opal_show_help_string("help-opal-wrapper.txt",
                                            "version", false,
                                            argv[0], options_data[user_data_idx].project, options_data[user_data_idx].version, options_data[user_data_idx].language, NULL);
                if (NULL != str) {
                    printf("%s", str);
                    free(str);
                }
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:help", strlen("-showme:help")) ||
                       0 == strncmp(user_argv[i], "--showme:help", strlen("--showme:help"))) {
                char *str;
                str = opal_show_help_string("help-opal-wrapper.txt", "usage", 
                                            false, argv[0],
                                            options_data[user_data_idx].project, 
                                            NULL);
                if (NULL != str) {
                    printf("%s", str);
                    free(str);
                }

                exit_status = 0;
                goto cleanup;
            } else if (0 == strncmp(user_argv[i], "-showme:", strlen("-showme:")) ||
                       0 == strncmp(user_argv[i], "--showme:", strlen("--showme:"))) {
                fprintf(stderr, "%s: unrecognized option: %s\n", argv[0],
                        user_argv[i]);
                fprintf(stderr, "Type '%s --showme:help' for usage.\n",
                        argv[0]);
                exit_status = 1;
                goto cleanup;
            }

            flags |= (COMP_DRY_RUN|COMP_SHOW_ERROR);
            /* remove element from user_argv */
            opal_argv_delete(&user_argc, &user_argv, i, 1);
            --i;

            if (done_now) {
                disable_flags = false;
                break;
            }

        } else if (0 == strcmp(user_argv[i], "-c")) {
            flags &= ~COMP_WANT_LINK;
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-E") || 
                   0 == strcmp(user_argv[i], "-M")) {
            flags &= ~(COMP_WANT_COMPILE | COMP_WANT_LINK);
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-S")) {
            flags &= ~COMP_WANT_LINK;
            real_flag = true;
        } else if (0 == strcmp(user_argv[i], "-lpmpi")) {
            flags |= COMP_WANT_PMPI;

            /* remove element from user_argv */
            opal_argv_delete(&user_argc, &user_argv, i, 1);
            --i;
        } else if (0 == strcmp(user_argv[i], "-static") ||
                   0 == strcmp(user_argv[i], "--static") ||
                   0 == strcmp(user_argv[i], "-Bstatic") ||
                   0 == strcmp(user_argv[i], "-Wl,-static") ||
                   0 == strcmp(user_argv[i], "-Wl,--static") ||
                   0 == strcmp(user_argv[i], "-Wl,-Bstatic")) {
            flags |= COMP_WANT_STATIC;
        } else if (0 == strcmp(user_argv[i], "-dynamic") ||
                   0 == strcmp(user_argv[i], "--dynamic") ||
                   0 == strcmp(user_argv[i], "-Bdynamic") ||
                   0 == strcmp(user_argv[i], "-Wl,-dynamic") ||
                   0 == strcmp(user_argv[i], "-Wl,--dynamic") ||
                   0 == strcmp(user_argv[i], "-Wl,-Bdynamic")) {
            flags &= ~COMP_WANT_STATIC;
        } else if (0 == strcmp(user_argv[i], "--openmpi:linkall")) {
            /* This is an intentionally undocummented wrapper compiler
               switch.  It should only be used by Open MPI developers
               -- not end users.  It will cause mpicc to use the
               static library list, even if we're compiling
               dynamically (i.e., it'll specifically -lopen-rte and
               -lopen-pal (and all their dependent libs)).  We provide
               this flag for test MPI applications that also invoke
               ORTE and/or OPAL function calls.

               On some systems (e.g., OS X), if the top-level
               application calls ORTE/OPAL functions and you don't -l
               ORTE and OPAL, then the functions won't be resolved at
               link time (i.e., the implicit library dependencies of
               libmpi won't be pulled in at link time), and therefore
               the link will fail.  This flag will cause the wrapper
               to explicitly list the ORTE and OPAL libs on the
               underlying compiler command line, so the application
               will therefore link properly. */
            flags |= COMP_WANT_LINKALL;

            /* remove element from user_argv */
            opal_argv_delete(&user_argc, &user_argv, i, 1);
        } else if ('-' != user_argv[i][0]) {
            disable_flags = false;
            flags |= COMP_SHOW_ERROR;
            real_flag = true;
        } else { 
            /* if the option flag is one that we use to determine
               which set of compiler data to use, don't count it as a
               real option */
            if (find_options_index(user_argv[i]) < 0) {
                real_flag = true;
            }
        }
    }

    /* clear out the want_flags if we got no arguments not starting
       with a - (dash) and -showme wasn't given OR -showme was given
       and we had at least one more non-showme argument that started
       with a - (dash) and no other non-dash arguments.  Some examples:

       opal_wrapper                : clear our flags
       opal_wrapper -v             : clear our flags
       opal_wrapper -E a.c         : don't clear our flags
       opal_wrapper a.c            : don't clear our flags
       opal_wrapper -showme        : don't clear our flags
       opal_wrapper -showme -v     : clear our flags
       opal_wrapper -showme -E a.c : don't clear our flags
       opal_wrapper -showme a.c    : don't clear our flags
    */
    if (disable_flags && !((flags & COMP_DRY_RUN) && !real_flag)) {
        flags &= ~(COMP_WANT_PREPROC|COMP_WANT_COMPILE|COMP_WANT_LINK);
    }

    /****************************************************
     *
     * Assemble the command line
     *
     ****************************************************/

    /* compiler (may be multiple arguments, so split) */
    if (flags & COMP_WANT_COMMAND) {
        exec_argv = opal_argv_split(options_data[user_data_idx].compiler, ' ');
        exec_argc = opal_argv_count(exec_argv);
    } else {
        exec_argv = (char **) malloc(sizeof(char*));
        exec_argv[0] = NULL;
        exec_argc = 0;
    }

    /* This error would normally not happen unless the user edits the 
       wrapper data files manually */
    if (NULL == exec_argv) {
        opal_show_help("help-opal-wrapper.txt", "no-compiler-specified", true);
        return 1;
    }

    if (flags & COMP_WANT_COMPILE) {
        opal_argv_insert(&exec_argv, exec_argc,
                         options_data[user_data_idx].comp_flags_prefix);
        exec_argc = opal_argv_count(exec_argv);
    }

    /* Per https://svn.open-mpi.org/trac/ompi/ticket/2201, add all the
       user arguments before anything else. */
    opal_argv_insert(&exec_argv, exec_argc, user_argv);
    exec_argc = opal_argv_count(exec_argv);

    /* preproc flags */
    if (flags & COMP_WANT_PREPROC) {
        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].preproc_flags);
        exec_argc = opal_argv_count(exec_argv);
    }

    /* compiler flags */
    if (flags & COMP_WANT_COMPILE) {
        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].comp_flags);
        exec_argc = opal_argv_count(exec_argv);
    }

    /* link flags and libs */
    if (flags & COMP_WANT_LINK) {
        bool have_static_lib;
        bool have_dyn_lib;
        bool use_static_libs;
        char *filename1, *filename2;
        struct stat buf;

        opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].link_flags);
        exec_argc = opal_argv_count(exec_argv);

        /* Are we linking statically?  If so, decide what libraries to
           list.  It depends on two factors:

           1. Was --static (etc.) specified?
           2. Does OMPI have static, dynamic, or both libraries installed?

           Here's a matrix showing what we'll do in all 6 cases:

           What's installed    --static    no --static
           ----------------    ----------  -----------
           ompi .so libs       -lmpi       -lmpi
           ompi .a libs        all         all
           ompi both libs      all         -lmpi

        */

        filename1 = opal_os_path( false, options_data[user_data_idx].path_libdir, options_data[user_data_idx].static_lib_file, NULL );
        if (0 == stat(filename1, &buf)) {
            have_static_lib = true;
        } else {
            have_static_lib = false;
        }

        filename2 = opal_os_path( false, options_data[user_data_idx].path_libdir, options_data[user_data_idx].dyn_lib_file, NULL );
        if (0 == stat(filename2, &buf)) {
            have_dyn_lib = true;
        } else {
            have_dyn_lib = false;
        }

        /* Determine which set of libs to use: dynamic or static.  Be
           pedantic to make the code easy to read. */
        if (flags & COMP_WANT_LINKALL) {
            /* If --openmpi:linkall was specified, list all the libs
               (i.e., the static libs) if they're available, either in
               static or dynamic form. */
            if (have_static_lib || have_dyn_lib) {
                use_static_libs = true;
            } else {
                fprintf(stderr, "The linkall option has failed as we were unable to find either static or dynamic libs\n"
                        "Files looked for:\n  Static: %s\n  Dynamic: %s\n",
                        filename1, filename2);
                free(filename1);
                free(filename2);
                exit(1);
            }
        } else if (flags & COMP_WANT_STATIC) {
            /* If --static (or something like it) was specified, if we
               have the static libs, then use them.  Otherwise, use
               the dynamic libs. */
            if (have_static_lib) {
                use_static_libs = true;
            } else {
                use_static_libs = false;
            }
        } else {
            /* If --static (or something like it) was NOT specified
               (or if --dyanic, or something like it, was specified),
               if we have the dynamic libs, then use them.  Otherwise,
               use the static libs. */
            if (have_dyn_lib) {
                use_static_libs = false;
            } else {
                use_static_libs = true;
            }
        }
        free(filename1);
        free(filename2);

        if (use_static_libs) {
            opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].libs_static);
        } else {
            opal_argv_insert(&exec_argv, exec_argc, options_data[user_data_idx].libs);
        }
        exec_argc = opal_argv_count(exec_argv);
    }


    /****************************************************
     *
     * Execute the command
     *
     ****************************************************/

    if (flags & COMP_DRY_RUN) {
        exec_command = opal_argv_join(exec_argv, ' ');
        printf("%s\n", exec_command);
    } else {
        char *tmp;

#if 0
        exec_command = opal_argv_join(exec_argv, ' ');
        printf("command: %s\n", exec_command);
#endif

        tmp = opal_path_findv(exec_argv[0], 0, environ, NULL);
        if (NULL == tmp) {
            opal_show_help("help-opal-wrapper.txt", "no-compiler-found", true,
                           exec_argv[0], NULL);
            errno = 0;
            exit_status = 1;
        }  else {
            int status;

            free(exec_argv[0]);
            exec_argv[0] = tmp;
            ret = opal_few(exec_argv, &status);
            exit_status = WIFEXITED(status) ? WEXITSTATUS(status) :
                              (WIFSIGNALED(status) ? WTERMSIG(status) :
                                  (WIFSTOPPED(status) ? WSTOPSIG(status) : 255));
            if( (OPAL_SUCCESS != ret) || ((0 != exit_status) && (flags & COMP_SHOW_ERROR)) ) {
                char* exec_command = opal_argv_join(exec_argv, ' ');
                if( OPAL_SUCCESS != ret ) {
                    opal_show_help("help-opal-wrapper.txt", "spawn-failed", true,
                                   exec_argv[0], strerror(status), exec_command, NULL);
                } else {
#if 0
                    opal_show_help("help-opal-wrapper.txt", "compiler-failed", true,
                                   exec_argv[0], exit_status, exec_command, NULL);
#endif
                }
                free(exec_command);
            }
        }
    }

    /****************************************************
     *
     * Cleanup
     *
     ****************************************************/
 cleanup:

    opal_argv_free(exec_argv);
    opal_argv_free(user_argv);
    if (NULL != base_argv0) free(base_argv0);

    if (OPAL_SUCCESS != (ret = data_finalize())) {
        return ret;
    }

    if (OPAL_SUCCESS != (ret = opal_finalize_util())) {
        return ret;
    }

    return exit_status;
}
Example #13
0
int main(int argc, char *argv[])
{
    int rc, i, j;
    opal_cmd_line_t cmd_line;
    char *param, *value;
    orte_job_t *jdata=NULL;
    orte_app_context_t *app;
    char *uri, *ptr;

    /* Setup and parse the command line */
    memset(&myglobals, 0, sizeof(myglobals));
    /* find our basename (the name of the executable) so that we can
       use it in pretty-print error messages */
    myglobals.basename = opal_basename(argv[0]);

    opal_cmd_line_create(&cmd_line, cmd_line_init);
    mca_base_cmd_line_setup(&cmd_line);
    if (OPAL_SUCCESS != (rc = opal_cmd_line_parse(&cmd_line, true,
                                                  argc, argv)) ) {
        if (OPAL_ERR_SILENT != rc) {
            fprintf(stderr, "%s: command line error (%s)\n", argv[0],
                    opal_strerror(rc));
        }
        return rc;
    }

    /* print version if requested.  Do this before check for help so
       that --version --help works as one might expect. */
    if (myglobals.version) {
        char *str;
        str = opal_info_make_version_str("all",
                                         OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION,
                                         OPAL_RELEASE_VERSION,
                                         OPAL_GREEK_VERSION,
                                         OPAL_REPO_REV);
        if (NULL != str) {
            fprintf(stdout, "%s %s\n\nReport bugs to %s\n",
                    myglobals.basename, str, PACKAGE_BUGREPORT);
            free(str);
        }
        exit(0);
    }

    /* check if we are running as root - if we are, then only allow
     * us to proceed if the allow-run-as-root flag was given. Otherwise,
     * exit with a giant warning flag
     */
    if (0 == geteuid() && !myglobals.run_as_root) {
        fprintf(stderr, "--------------------------------------------------------------------------\n");
        if (myglobals.help) {
            fprintf(stderr, "%s cannot provide the help message when run as root\n", myglobals.basename);
        } else {
            /* show_help is not yet available, so print an error manually */
            fprintf(stderr, "%s has detected an attempt to run as root.\n", myglobals.basename);
        }
        fprintf(stderr, " This is *strongly* discouraged as any mistake (e.g., in defining TMPDIR) or bug can\n");
        fprintf(stderr, "result in catastrophic damage to the OS file system, leaving\n");
        fprintf(stderr, "your system in an unusable state.\n\n");
        fprintf(stderr, "You can override this protection by adding the --allow-run-as-root\n");
        fprintf(stderr, "option to your cmd line. However, we reiterate our strong advice\n");
        fprintf(stderr, "against doing so - please do so at your own risk.\n");
        fprintf(stderr, "--------------------------------------------------------------------------\n");
        exit(1);
    }

    /*
     * Since this process can now handle MCA/GMCA parameters, make sure to
     * process them.
     * NOTE: It is "safe" to call mca_base_cmd_line_process_args() before
     *  opal_init_util() since mca_base_cmd_line_process_args() does *not*
     *  depend upon opal_init_util() functionality.
     */
    if (OPAL_SUCCESS != mca_base_cmd_line_process_args(&cmd_line, &environ, &environ)) {
        exit(1);
    }

    /* Need to initialize OPAL so that install_dirs are filled in */
    if (OPAL_SUCCESS != opal_init(&argc, &argv)) {
        exit(1);
    }

    /* Check for help request */
    if (myglobals.help) {
        char *str, *args = NULL;
        char *project_name = NULL;
        if (0 == strcmp(myglobals.basename, "mpirun")) {
            project_name = "Open MPI";
        } else {
            project_name = "OpenRTE";
        }
        args = opal_cmd_line_get_usage_msg(&cmd_line);
        str = opal_show_help_string("help-orterun.txt", "orterun:usage", false,
                                    myglobals.basename, project_name, OPAL_VERSION,
                                    myglobals.basename, args,
                                    PACKAGE_BUGREPORT);
        if (NULL != str) {
            printf("%s", str);
            free(str);
        }
        free(args);

        /* If someone asks for help, that should be all we do */
        exit(0);
    }

    /* flag that I am the HNP */
    orte_process_info.proc_type = ORTE_PROC_HNP;

    /* Setup MCA params */
    orte_register_params();

    /* specify the DVM state machine */
    opal_setenv("OMPI_MCA_state", "dvm", true, &environ);

    /* Intialize our Open RTE environment */
    if (ORTE_SUCCESS != (rc = orte_init(&argc, &argv, ORTE_PROC_HNP))) {
        /* cannot call ORTE_ERROR_LOG as it could be the errmgr
         * never got loaded!
         */
        return rc;
    }
    /* finalize OPAL. As it was opened again from orte_init->opal_init
     * we continue to have a reference count on it. So we have to finalize it twice...
     */
    opal_finalize();

    /* check for request to report uri */
    uri = orte_rml.get_contact_info();
    if (NULL != myglobals.report_uri) {
        FILE *fp;
        if (0 == strcmp(myglobals.report_uri, "-")) {
            /* if '-', then output to stdout */
            printf("VMURI: %s\n", uri);
        } else if (0 == strcmp(myglobals.report_uri, "+")) {
            /* if '+', output to stderr */
            fprintf(stderr, "VMURI: %s\n", uri);
        } else if (0 == strncasecmp(myglobals.report_uri, "file:", strlen("file:"))) {
            ptr = strchr(myglobals.report_uri, ':');
            ++ptr;
            fp = fopen(ptr, "w");
            if (NULL == fp) {
                orte_show_help("help-orterun.txt", "orterun:write_file", false,
                               myglobals.basename, "pid", ptr);
                exit(0);
            }
            fprintf(fp, "%s\n", uri);
            fclose(fp);
        } else {
            fp = fopen(myglobals.report_uri, "w");
            if (NULL == fp) {
                orte_show_help("help-orterun.txt", "orterun:write_file", false,
                               myglobals.basename, "pid", myglobals.report_uri);
                exit(0);
            }
            fprintf(fp, "%s\n", uri);
            fclose(fp);
        }
        free(uri);
    } else {
        printf("VMURI: %s\n", uri);
    }

    /* get the daemon job object - was created by ess/hnp component */
    if (NULL == (jdata = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {
        orte_show_help("help-orterun.txt", "bad-job-object", true,
                       myglobals.basename);
        exit(0);
    }
    /* also should have created a daemon "app" */
    if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, 0))) {
        orte_show_help("help-orterun.txt", "bad-app-object", true,
                       myglobals.basename);
        exit(0);
    }

    /* Did the user specify a prefix, or want prefix by default? */
    if (opal_cmd_line_is_taken(&cmd_line, "prefix") || want_prefix_by_default) {
        size_t param_len;
        /* if both the prefix was given and we have a prefix
         * given above, check to see if they match
         */
        if (opal_cmd_line_is_taken(&cmd_line, "prefix") &&
            NULL != myglobals.prefix) {
            /* if they don't match, then that merits a warning */
            param = strdup(opal_cmd_line_get_param(&cmd_line, "prefix", 0, 0));
            /* ensure we strip any trailing '/' */
            if (0 == strcmp(OPAL_PATH_SEP, &(param[strlen(param)-1]))) {
                param[strlen(param)-1] = '\0';
            }
            value = strdup(myglobals.prefix);
            if (0 == strcmp(OPAL_PATH_SEP, &(value[strlen(value)-1]))) {
                value[strlen(value)-1] = '\0';
            }
            if (0 != strcmp(param, value)) {
                orte_show_help("help-orterun.txt", "orterun:app-prefix-conflict",
                               true, myglobals.basename, value, param);
                /* let the global-level prefix take precedence since we
                 * know that one is being used
                 */
                free(param);
                param = strdup(myglobals.prefix);
            }
            free(value);
        } else if (NULL != myglobals.prefix) {
            param = myglobals.prefix;
        } else if (opal_cmd_line_is_taken(&cmd_line, "prefix")){
            /* must be --prefix alone */
            param = strdup(opal_cmd_line_get_param(&cmd_line, "prefix", 0, 0));
        } else {
            /* --enable-orterun-prefix-default was given to orterun */
            param = strdup(opal_install_dirs.prefix);
        }

        if (NULL != param) {
            /* "Parse" the param, aka remove superfluous path_sep. */
            param_len = strlen(param);
            while (0 == strcmp (OPAL_PATH_SEP, &(param[param_len-1]))) {
                param[param_len-1] = '\0';
                param_len--;
                if (0 == param_len) {
                    orte_show_help("help-orterun.txt", "orterun:empty-prefix",
                                   true, myglobals.basename, myglobals.basename);
                    return ORTE_ERR_FATAL;
                }
            }
            orte_set_attribute(&app->attributes, ORTE_APP_PREFIX_DIR, ORTE_ATTR_GLOBAL, param, OPAL_STRING);
            free(param);
        }
    }

    /* Did the user specify a hostfile. Need to check for both
     * hostfile and machine file.
     * We can only deal with one hostfile per app context, otherwise give an error.
     */
    if (0 < (j = opal_cmd_line_get_ninsts(&cmd_line, "hostfile"))) {
        if(1 < j) {
            orte_show_help("help-orterun.txt", "orterun:multiple-hostfiles",
                           true, myglobals.basename, NULL);
            return ORTE_ERR_FATAL;
        } else {
            value = opal_cmd_line_get_param(&cmd_line, "hostfile", 0, 0);
            orte_set_attribute(&app->attributes, ORTE_APP_HOSTFILE, ORTE_ATTR_LOCAL, value, OPAL_STRING);
        }
    }
    if (0 < (j = opal_cmd_line_get_ninsts(&cmd_line, "machinefile"))) {
        if(1 < j || orte_get_attribute(&app->attributes, ORTE_APP_HOSTFILE, NULL, OPAL_STRING)) {
            orte_show_help("help-orterun.txt", "orterun:multiple-hostfiles",
                           true, myglobals.basename, NULL);
            return ORTE_ERR_FATAL;
        } else {
            value = opal_cmd_line_get_param(&cmd_line, "machinefile", 0, 0);
            orte_set_attribute(&app->attributes, ORTE_APP_HOSTFILE, ORTE_ATTR_LOCAL, value, OPAL_STRING);
        }
    }

    /* Did the user specify any hosts? */
    if (0 < (j = opal_cmd_line_get_ninsts(&cmd_line, "host"))) {
        char **targ=NULL, *tval;
        for (i = 0; i < j; ++i) {
            value = opal_cmd_line_get_param(&cmd_line, "host", i, 0);
            opal_argv_append_nosize(&targ, value);
        }
        tval = opal_argv_join(targ, ',');
        orte_set_attribute(&app->attributes, ORTE_APP_DASH_HOST, ORTE_ATTR_LOCAL, tval, OPAL_STRING);
        opal_argv_free(targ);
        free(tval);
    }
    OBJ_DESTRUCT(&cmd_line);

    /* setup to listen for commands sent specifically to me, even though I would probably
     * be the one sending them! Unfortunately, since I am a participating daemon,
     * there are times I need to send a command to "all daemons", and that means *I* have
     * to receive it too
     */
    orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_DAEMON,
                            ORTE_RML_PERSISTENT, orte_daemon_recv, NULL);

    /* override the notify_completed state so we can send a message
     * back to anyone who submits a job to us telling them the job
     * completed */
    if (ORTE_SUCCESS != (rc = orte_state.set_job_state_callback(ORTE_JOB_STATE_NOTIFY_COMPLETED, notify_requestor))) {
        ORTE_ERROR_LOG(rc);
        ORTE_UPDATE_EXIT_STATUS(rc);
        exit(orte_exit_status);
    }

    /* spawn the DVM - we skip the initial steps as this
     * isn't a user-level application */
    ORTE_ACTIVATE_JOB_STATE(jdata, ORTE_JOB_STATE_ALLOCATE);

    /* loop the event lib until an exit event is detected */
    while (orte_event_base_active) {
        opal_event_loop(orte_event_base, OPAL_EVLOOP_ONCE);
    }

    /* cleanup and leave */
    orte_finalize();

    if (orte_debug_flag) {
        fprintf(stderr, "exiting with status %d\n", orte_exit_status);
    }
    exit(orte_exit_status);
}
Example #14
0
static void ps_recv(int status,
                    orte_process_name_t *sender,
                    orcm_pnp_tag_t tag,
                    struct iovec *msg, int count,
                    opal_buffer_t *buf, void *cbdata)
{
    orte_jobid_t jobid;
    char *jname=NULL;
    orte_app_idx_t idx;
    char *app=NULL;
    int32_t n;
    int32_t max_restarts;
    orte_vpid_t vpid;
    pid_t pid;
    char *node=NULL;
    int32_t restarts;
    bool jfirst, afirst;
    int rc;

    /* output the title bars */
    opal_output(orte_clean_output, "JOB\t    JOB    \t  APP  \t        \t MAX");
    opal_output(orte_clean_output, "ID \t    NAME   \tCONTEXT\t  PATH  \tRESTARTS\tVPID\t  PID \t  NODE  \tRESTARTS");

    n=1;
    while (ORTE_SUCCESS == (rc = opal_dss.unpack(buf, &jobid, &n, ORTE_JOBID))) {
        /* get the job name */
        n=1;
        if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &jname, &n, OPAL_STRING))) {
            ORTE_ERROR_LOG(rc);
            goto release;
        }

        n=1;
        jfirst = true;
        afirst = true;
        while (ORTE_SUCCESS == (rc = opal_dss.unpack(buf, &idx, &n, ORTE_APP_IDX)) &&
               idx != ORTE_APP_IDX_MAX) {
            /* get the app data */
            n=1;
            if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &app, &n, OPAL_STRING))) {
                ORTE_ERROR_LOG(rc);
                goto release;
            }
            n=1;
            if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &max_restarts, &n, OPAL_INT32))) {
                ORTE_ERROR_LOG(rc);
                goto release;
            }

            /* unload the procs */
            n=1;
            while (ORTE_SUCCESS == (rc = opal_dss.unpack(buf, &vpid, &n, ORTE_VPID)) &&
                   vpid != ORTE_VPID_INVALID) {
                n=1;
                if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &pid, &n, OPAL_PID))) {
                    ORTE_ERROR_LOG(rc);
                    goto release;
                }
                n=1;
                if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &node, &n, OPAL_STRING))) {
                    ORTE_ERROR_LOG(rc);
                    goto release;
                }
                n=1;
                if (ORTE_SUCCESS != (rc = opal_dss.unpack(buf, &restarts, &n, OPAL_INT32))) {
                    ORTE_ERROR_LOG(rc);
                    goto release;
                }
                if (jfirst) {
                    /* output the job and app data with the proc */
                    opal_output(orte_clean_output, "%d\t%8s\t  %d\t%8s\t%8d\t%4s\t%6d\t%8s\t%d",
                                (int)ORTE_LOCAL_JOBID(jobid), jname, idx,
                                opal_basename(app), max_restarts,
                                ORTE_VPID_PRINT(vpid), (int)pid, node, restarts);
                    /* flag that we did the first one */
                    jfirst = false;
                } else if (afirst) {
                    opal_output(orte_clean_output, "\t\t\t  %d\t%8s\t%8d\t%4s\t%6d\t%8s\t%d",
                                idx, opal_basename(app), max_restarts,
                                ORTE_VPID_PRINT(vpid), (int)pid, node, restarts);
                    afirst = false;
                } else {
                    opal_output(orte_clean_output, "\t\t    \t    \t   \t%4s\t%6d\t%8s\t%d",
                                ORTE_VPID_PRINT(vpid), (int)pid, node, restarts);
                }
                free(node);
                node = NULL;
            }
            n=1;
            free(app);
            app = NULL;
        }
        n=1;
        free(jname);
        jname = NULL;
    }

 release:
    if (NULL != jname) {
        free(jname);
    }
    if (NULL != app) {
        free(app);
    }
    if (NULL != node) {
        free(node);
    }

    opal_fd_write(rel_pipe[1], sizeof(int), &rc);
}
Example #15
0
int orte_pls_rsh_launch(orte_jobid_t jobid)
{
    orte_job_map_t *map;
    opal_list_item_t *n_item;
    orte_mapped_node_t *rmaps_node;
    orte_std_cntr_t num_nodes;
    orte_vpid_t vpid;
    int node_name_index1;
    int node_name_index2;
    int proc_name_index;
    int local_exec_index, local_exec_index_end;
    char *jobid_string = NULL;
    char *uri, *param;
    char **argv = NULL, **tmp;
    char *prefix_dir;
    int argc;
    int rc;
    sigset_t sigs;
    struct passwd *p;
    bool remote_sh = false, remote_csh = false; 
    bool local_sh = false, local_csh = false;
    char *lib_base = NULL, *bin_base = NULL;
    orte_pls_daemon_info_t *dmn;
    orte_pls_rsh_shell_t shell;

    if (mca_pls_rsh_component.timing) {
        if (0 != gettimeofday(&joblaunchstart, NULL)) {
            opal_output(0, "pls_rsh: could not obtain start time");
            joblaunchstart.tv_sec = 0;
            joblaunchstart.tv_usec = 0;
        }        
    }
    
    /* setup a list that will contain the info for all the daemons
     * so we can store it on the registry when done and use it
     * locally to track their state
     */
    OBJ_CONSTRUCT(&active_daemons, opal_list_t);

    /* Get the map for this job
     * We need the entire mapping for a couple of reasons:
     *  - need the prefix to start with.
     *  - need to know the nodes we are launching on
     * All other mapping responsibilities fall to orted in the fork PLS
     */
    rc = orte_rmaps.get_job_map(&map, jobid);
    if (ORTE_SUCCESS != rc) {
        ORTE_ERROR_LOG(rc);
        OBJ_DESTRUCT(&active_daemons);
        return rc;
    }

    /* if the user requested that we re-use daemons,
     * launch the procs on any existing, re-usable daemons
     */
    if (orte_pls_base.reuse_daemons) {
        if (ORTE_SUCCESS != (rc = orte_pls_base_launch_on_existing_daemons(map))) {
            ORTE_ERROR_LOG(rc);
            OBJ_RELEASE(map);
            OBJ_DESTRUCT(&active_daemons);
            return rc;
        }
    }
    
    num_nodes = (orte_std_cntr_t)opal_list_get_size(&map->nodes);
    if (0 == num_nodes) {
        /* nothing left to do - just return */
        OBJ_RELEASE(map);
        OBJ_DESTRUCT(&active_daemons);
        return ORTE_SUCCESS;
    }

    if (mca_pls_rsh_component.debug_daemons &&
        mca_pls_rsh_component.num_concurrent < num_nodes) {
        /* we can't run in this situation, so pretty print the error
         * and exit
         */
        opal_show_help("help-pls-rsh.txt", "deadlock-params",
                       true, mca_pls_rsh_component.num_concurrent, num_nodes);
        OBJ_RELEASE(map);
        OBJ_DESTRUCT(&active_daemons);
        return ORTE_ERR_FATAL;
    }

    /*
     * After a discussion between Ralph & Jeff, we concluded that we
     * really are handling the prefix dir option incorrectly. It currently
     * is associated with an app_context, yet it really refers to the
     * location where OpenRTE/Open MPI is installed on a NODE. Fixing
     * this right now would involve significant change to orterun as well
     * as elsewhere, so we will intentionally leave this incorrect at this
     * point. The error, however, is identical to that seen in all prior
     * releases of OpenRTE/Open MPI, so our behavior is no worse than before.
     *
     * A note to fix this, along with ideas on how to do so, has been filed
     * on the project's Trac system under "feature enhancement".
     *
     * For now, default to the prefix_dir provided in the first app_context.
     * Since there always MUST be at least one app_context, we are safe in
     * doing this.
     */
    prefix_dir = map->apps[0]->prefix_dir;
    
    /*
     * Allocate a range of vpids for the daemons.
     */
    if (num_nodes == 0) {
        return ORTE_ERR_BAD_PARAM;
    }
    rc = orte_ns.reserve_range(0, num_nodes, &vpid);
    if (ORTE_SUCCESS != rc) {
        goto cleanup;
    }

    /* setup the orted triggers for passing their launch info */
    if (ORTE_SUCCESS != (rc = orte_smr.init_orted_stage_gates(jobid, num_nodes, NULL, NULL))) {
        ORTE_ERROR_LOG(rc);
        goto cleanup;
    }
    
    /* need integer value for command line parameter */
    if (ORTE_SUCCESS != (rc = orte_ns.convert_jobid_to_string(&jobid_string, jobid))) {
        ORTE_ERROR_LOG(rc);
        goto cleanup;
    }

    /* What is our local shell? */
    shell = ORTE_PLS_RSH_SHELL_UNKNOWN;
    p = getpwuid(getuid());
    if (NULL != p) {
        param = p->pw_shell;
        shell = find_shell(p->pw_shell);
    }
    /* If we didn't find it in getpwuid(), try looking at the $SHELL
       environment variable (see
       https://svn.open-mpi.org/trac/ompi/ticket/1060) */
    if (ORTE_PLS_RSH_SHELL_UNKNOWN == shell && 
        NULL != (param = getenv("SHELL"))) {
        shell = find_shell(param);
    }

    switch (shell) {
    case ORTE_PLS_RSH_SHELL_SH:  /* fall through */
    case ORTE_PLS_RSH_SHELL_KSH: /* fall through */
    case ORTE_PLS_RSH_SHELL_ZSH: /* fall through */
    case ORTE_PLS_RSH_SHELL_BASH: local_sh = true; break;
    case ORTE_PLS_RSH_SHELL_TCSH: /* fall through */
    case ORTE_PLS_RSH_SHELL_CSH:  local_csh = true; break;
    default:
        opal_output(0, "WARNING: local probe returned unhandled shell:%s assuming bash\n",
                    (NULL != param) ? param : "unknown");
        remote_sh = true;
        break;
    }

    if (mca_pls_rsh_component.debug) {
        opal_output(0, "pls:rsh: local csh: %d, local sh: %d\n",
                    local_csh, local_sh);
    }

    /* What is our remote shell? */
    if (mca_pls_rsh_component.assume_same_shell) {
        remote_sh = local_sh;
        remote_csh = local_csh;
        if (mca_pls_rsh_component.debug) {
            opal_output(0, "pls:rsh: assuming same remote shell as local shell");
        }
    } else {
        orte_pls_rsh_shell_t shell;
        rmaps_node = (orte_mapped_node_t*)opal_list_get_first(&map->nodes);
        rc = orte_pls_rsh_probe(rmaps_node, &shell);

        if (ORTE_SUCCESS != rc) {
            ORTE_ERROR_LOG(rc);
            return rc;
        }

        switch (shell) {
        case ORTE_PLS_RSH_SHELL_SH:  /* fall through */
        case ORTE_PLS_RSH_SHELL_KSH: /* fall through */
        case ORTE_PLS_RSH_SHELL_ZSH: /* fall through */
        case ORTE_PLS_RSH_SHELL_BASH: remote_sh = true; break;
        case ORTE_PLS_RSH_SHELL_TCSH: /* fall through */
        case ORTE_PLS_RSH_SHELL_CSH:  remote_csh = true; break;
        default:
            opal_output(0, "WARNING: rsh probe returned unhandled shell; assuming bash\n");
            remote_sh = true;
        }
    }
    if (mca_pls_rsh_component.debug) {
        opal_output(0, "pls:rsh: remote csh: %d, remote sh: %d\n",
                    remote_csh, remote_sh);
    }

    /*
     * Build argv array
     */
    argv = opal_argv_copy(mca_pls_rsh_component.agent_argv);
    argc = mca_pls_rsh_component.agent_argc;
    node_name_index1 = argc;
    opal_argv_append(&argc, &argv, "<template>");

    /* Do we need to source .profile on the remote side? */

    if (!(remote_csh || remote_sh)) {
        int i;
        tmp = opal_argv_split("( test ! -r ./.profile || . ./.profile;", ' ');
        if (NULL == tmp) {
            return ORTE_ERR_OUT_OF_RESOURCE;
        }
        for (i = 0; NULL != tmp[i]; ++i) {
            opal_argv_append(&argc, &argv, tmp[i]);
        }
        opal_argv_free(tmp);
    }

    /* add the daemon command (as specified by user) */
    local_exec_index = argc;
    opal_argv_append(&argc, &argv, mca_pls_rsh_component.orted);

    /* check for debug flags */
    orte_pls_base_mca_argv(&argc, &argv);

    opal_argv_append(&argc, &argv, "--bootproxy");
    opal_argv_append(&argc, &argv, jobid_string);
    opal_argv_append(&argc, &argv, "--name");
    proc_name_index = argc;
    opal_argv_append(&argc, &argv, "<template>");

    /* tell the daemon how many procs are in the daemon's job */
    opal_argv_append(&argc, &argv, "--num_procs");
    asprintf(&param, "%lu", (unsigned long)(vpid + num_nodes));
    opal_argv_append(&argc, &argv, param);
    free(param);
    /* tell the daemon the starting vpid of the daemon's job */
    opal_argv_append(&argc, &argv, "--vpid_start");
    opal_argv_append(&argc, &argv, "0");

    opal_argv_append(&argc, &argv, "--nodename");
    node_name_index2 = argc;
    opal_argv_append(&argc, &argv, "<template>");

    /* pass along the universe name and location info */
    opal_argv_append(&argc, &argv, "--universe");
    asprintf(&param, "%s@%s:%s", orte_universe_info.uid,
                orte_universe_info.host, orte_universe_info.name);
    opal_argv_append(&argc, &argv, param);
    free(param);

    /* setup ns contact info */
    opal_argv_append(&argc, &argv, "--nsreplica");
    if (NULL != orte_process_info.ns_replica_uri) {
        uri = strdup(orte_process_info.ns_replica_uri);
    } else {
        uri = orte_rml.get_uri();
    }
    asprintf(&param, "\"%s\"", uri);
    opal_argv_append(&argc, &argv, param);
    free(uri);
    free(param);

    /* setup gpr contact info */
    opal_argv_append(&argc, &argv, "--gprreplica");
    if (NULL != orte_process_info.gpr_replica_uri) {
        uri = strdup(orte_process_info.gpr_replica_uri);
    } else {
        uri = orte_rml.get_uri();
    }
    asprintf(&param, "\"%s\"", uri);
    opal_argv_append(&argc, &argv, param);
    free(uri);
    free(param);

    local_exec_index_end = argc;
    if (!(remote_csh || remote_sh)) {
        opal_argv_append(&argc, &argv, ")");
    }
    if (mca_pls_rsh_component.debug) {
        param = opal_argv_join(argv, ' ');
        if (NULL != param) {
            opal_output(0, "pls:rsh: final template argv:");
            opal_output(0, "pls:rsh:     %s", param);
            free(param);
        }
    }

    /* Figure out the basenames for the libdir and bindir.  This
       requires some explanation:

       - Use opal_install_dirs.libdir and opal_install_dirs.bindir instead of -D'ing some macros
         in this directory's Makefile.am because it makes all the
         dependencies work out correctly.  These are defined in
         opal/install_dirs.h.

       - After a discussion on the devel-core mailing list, the
         developers decided that we should use the local directory
         basenames as the basis for the prefix on the remote note.
         This does not handle a few notable cases (e.g., f the
         libdir/bindir is not simply a subdir under the prefix, if the
         libdir/bindir basename is not the same on the remote node as
         it is here in the local node, etc.), but we decided that
         --prefix was meant to handle "the common case".  If you need
         something more complex than this, a) edit your shell startup
         files to set PATH/LD_LIBRARY_PATH properly on the remove
         node, or b) use some new/to-be-defined options that
         explicitly allow setting the bindir/libdir on the remote
         node.  We decided to implement these options (e.g.,
         --remote-bindir and --remote-libdir) to orterun when it
         actually becomes a problem for someone (vs. a hypothetical
         situation).

       Hence, for now, we simply take the basename of this install's
       libdir and bindir and use it to append this install's prefix
       and use that on the remote node.
    */

    lib_base = opal_basename(opal_install_dirs.libdir);
    bin_base = opal_basename(opal_install_dirs.bindir);

    /*
     * Iterate through each of the nodes
     */
    if (mca_pls_rsh_component.timing) {
        /* allocate space to track the start times */
        launchstart = (struct timeval*)malloc((num_nodes+vpid) * sizeof(struct timeval));
    }
    
    for(n_item =  opal_list_get_first(&map->nodes);
        n_item != opal_list_get_end(&map->nodes);
        n_item =  opal_list_get_next(n_item)) {
        orte_process_name_t* name;
        pid_t pid;
        char *exec_path;
        char **exec_argv;
        
        rmaps_node = (orte_mapped_node_t*)n_item;
        
        if (mca_pls_rsh_component.timing) {
            if (0 != gettimeofday(&launchstart[vpid], NULL)) {
                opal_output(0, "pls_rsh: could not obtain start time");
            }
        }
        
        /* new daemon - setup to record its info */
        dmn = OBJ_NEW(orte_pls_daemon_info_t);
        dmn->active_job = jobid;
        opal_list_append(&active_daemons, &dmn->super);
        
        /* setup node name */
        free(argv[node_name_index1]);
        if (NULL != rmaps_node->username &&
            0 != strlen (rmaps_node->username)) {
            asprintf (&argv[node_name_index1], "%s@%s",
                      rmaps_node->username, rmaps_node->nodename);
        } else {
            argv[node_name_index1] = strdup(rmaps_node->nodename);
        }

        free(argv[node_name_index2]);
        argv[node_name_index2] = strdup(rmaps_node->nodename);
        
        /* save it in the daemon info */
        dmn->nodename = strdup(rmaps_node->nodename);

        /* initialize daemons process name */
        rc = orte_ns.create_process_name(&name, rmaps_node->cell, 0, vpid);
        if (ORTE_SUCCESS != rc) {
            ORTE_ERROR_LOG(rc);
            goto cleanup;
        }
        
        /* save it in the daemon info */
        dmn->cell = rmaps_node->cell;
        if (ORTE_SUCCESS != (rc = orte_dss.copy((void**)&(dmn->name), name, ORTE_NAME))) {
            ORTE_ERROR_LOG(rc);
            goto cleanup;
        }

        /* fork a child to exec the rsh/ssh session */
        
        /* set the process state to "launched" */
        if (ORTE_SUCCESS != (rc = orte_smr.set_proc_state(name, ORTE_PROC_STATE_LAUNCHED, 0))) {
            ORTE_ERROR_LOG(rc);
            goto cleanup;
        }

        pid = fork();
        if (pid < 0) {
            rc = ORTE_ERR_OUT_OF_RESOURCE;
            goto cleanup;
        }

        /* child */
        if (pid == 0) {
            char* name_string;
            char** env;
            char* var;
            long fd, fdmax = sysconf(_SC_OPEN_MAX);

            if (mca_pls_rsh_component.debug) {
                opal_output(0, "pls:rsh: launching on node %s\n",
                            rmaps_node->nodename);
            }

            /* We don't need to sense an oversubscribed condition and set the sched_yield
             * for the node as we are only launching the daemons at this time. The daemons
             * are now smart enough to set the oversubscribed condition themselves when
             * they launch the local procs.
             */

            /* Is this a local launch?
             *
             * Not all node names may be resolvable (if we found
             * localhost in the hostfile, for example).  So first
             * check trivial case of node_name being same as the
             * current nodename, which must be local.  If that doesn't
             * match, check using ifislocal().
             */
            if (!mca_pls_rsh_component.force_rsh &&
                (0 == strcmp(rmaps_node->nodename, orte_system_info.nodename) ||
                opal_ifislocal(rmaps_node->nodename))) {
                if (mca_pls_rsh_component.debug) {
                    opal_output(0, "pls:rsh: %s is a LOCAL node\n",
                                rmaps_node->nodename);
                }
                if (mca_pls_rsh_component.timing) {
                    /* since this is a local launch, the daemon will never reach
                     * the waitpid callback - so set the start value to
                     * something nonsensical
                     */
                    launchstart[vpid].tv_sec = 0;
                    launchstart[vpid].tv_usec = 0;
                }
                
                exec_path = opal_path_findv(argv[local_exec_index], 0, environ, NULL);

                if (NULL == exec_path && NULL == prefix_dir) {
                    rc = orte_pls_rsh_fill_exec_path (&exec_path);
                    if (ORTE_SUCCESS != rc) {
                        exit(-1);  /* the forked process MUST exit */
                    }
                } else {
                    if (NULL != prefix_dir) {
                        exec_path = opal_os_path( false, prefix_dir, bin_base, "orted", NULL );
                    }
                    /* If we yet did not fill up the execpath, do so now */
                    if (NULL == exec_path) {
                        rc = orte_pls_rsh_fill_exec_path (&exec_path);
                        if (ORTE_SUCCESS != rc) {
                            exit(-1);  /* the forked process MUST exit */
                        }
                    }
                }

                /* If we have a prefix, then modify the PATH and
                   LD_LIBRARY_PATH environment variables.  We're
                   already in the child process, so it's ok to modify
                   environ. */
                if (NULL != prefix_dir) {
                    char *oldenv, *newenv;

                    /* Reset PATH */
                    newenv = opal_os_path( false, prefix_dir, bin_base, NULL );
                    oldenv = getenv("PATH");
                    if (NULL != oldenv) {
                        char *temp;
                        asprintf(&temp, "%s:%s", newenv, oldenv );
                        free( newenv );
                        newenv = temp;
                    }
                    opal_setenv("PATH", newenv, true, &environ);
                    if (mca_pls_rsh_component.debug) {
                        opal_output(0, "pls:rsh: reset PATH: %s", newenv);
                    }
                    free(newenv);

                    /* Reset LD_LIBRARY_PATH */
                    newenv = opal_os_path( false, prefix_dir, lib_base, NULL );
                    oldenv = getenv("LD_LIBRARY_PATH");
                    if (NULL != oldenv) {
                        char* temp;
                        asprintf(&temp, "%s:%s", newenv, oldenv);
                        free(newenv);
                        newenv = temp;
                    }
                    opal_setenv("LD_LIBRARY_PATH", newenv, true, &environ);
                    if (mca_pls_rsh_component.debug) {
                        opal_output(0, "pls:rsh: reset LD_LIBRARY_PATH: %s",
                                    newenv);
                    }
                    free(newenv);
                }

                /* Since this is a local execution, we need to
                   potentially whack the final ")" in the argv (if
                   sh/csh conditionals, from above).  Note that we're
                   modifying the argv[] in the child process, so
                   there's no need to save this and restore it
                   afterward -- the parent's argv[] is unmodified. */
                if (NULL != argv[local_exec_index_end]) {
                    free(argv[local_exec_index_end]);
                    argv[local_exec_index_end] = NULL;
                }

                /* tell the daemon to setup its own process session/group */
                opal_argv_append(&argc, &argv, "--set-sid");
                exec_argv = &argv[local_exec_index];
                
                /* Finally, chdir($HOME) because we're making the
                   assumption that this is what will happen on
                   remote nodes (via rsh/ssh).  This allows a user
                   to specify a path that is relative to $HOME for
                   both the cwd and argv[0] and it will work on
                   all nodes -- including the local nost.
                   Otherwise, it would work on remote nodes and
                   not the local node.  If the user does not start
                   in $HOME on the remote nodes... well... let's
                   hope they start in $HOME.  :-) */
                var = getenv("HOME");
                if (NULL != var) {
                    if (mca_pls_rsh_component.debug) {
                        opal_output(0, "pls:rsh: changing to directory %s", var);
                    }
                    /* Ignore errors -- what are we going to do?
                       (and we ignore errors on the remote nodes
                       in the fork pls, so this is consistent) */
                    chdir(var);
                }
            } else {
                if (mca_pls_rsh_component.debug) {
                    opal_output(0, "pls:rsh: %s is a REMOTE node\n",
                                rmaps_node->nodename);
                }
                exec_argv = argv;
                exec_path = strdup(mca_pls_rsh_component.agent_path);

                if (NULL != prefix_dir) {
                    char *opal_prefix = getenv("OPAL_PREFIX");
                    if (remote_sh) {
                        asprintf (&argv[local_exec_index],
                                  "%s%s%s PATH=%s/%s:$PATH ; export PATH ; "
                                  "LD_LIBRARY_PATH=%s/%s:$LD_LIBRARY_PATH ; export LD_LIBRARY_PATH ; "
                                  "%s/%s/%s",
                                  (opal_prefix != NULL ? "OPAL_PREFIX=" : ""),
                                  (opal_prefix != NULL ? opal_prefix : ""),
                                  (opal_prefix != NULL ? " ;" : ""),
                                  prefix_dir, bin_base,
                                  prefix_dir, lib_base,
                                  prefix_dir, bin_base,
                                  mca_pls_rsh_component.orted);
                    }
                    if (remote_csh) {
                        /* [t]csh is a bit more challenging -- we
                           have to check whether LD_LIBRARY_PATH
                           is already set before we try to set it.
                           Must be very careful about obeying
                           [t]csh's order of evaluation and not
                           using a variable before it is defined.
                           See this thread for more details:
                           http://www.open-mpi.org/community/lists/users/2006/01/0517.php. */
                        asprintf (&argv[local_exec_index],
                                  "%s%s%s set path = ( %s/%s $path ) ; "
                                  "if ( $?LD_LIBRARY_PATH == 1 ) "
                                  "set OMPI_have_llp ; "
                                  "if ( $?LD_LIBRARY_PATH == 0 ) "
                                  "setenv LD_LIBRARY_PATH %s/%s ; "
                                  "if ( $?OMPI_have_llp == 1 ) "
                                  "setenv LD_LIBRARY_PATH %s/%s:$LD_LIBRARY_PATH ; "
                                  "%s/%s/%s",
                                  (opal_prefix != NULL ? "setenv OPAL_PREFIX " : ""),
                                  (opal_prefix != NULL ? opal_prefix : ""),
                                  (opal_prefix != NULL ? " ;" : ""),
                                  prefix_dir, bin_base,
                                  prefix_dir, lib_base,
                                  prefix_dir, lib_base,
                                  prefix_dir, bin_base,
                                  mca_pls_rsh_component.orted);
                    }
                }
            }

            /* setup process name */
            rc = orte_ns.get_proc_name_string(&name_string, name);
            if (ORTE_SUCCESS != rc) {
                opal_output(0, "orte_pls_rsh: unable to create process name");
                exit(-1);
            }
            free(argv[proc_name_index]);
            argv[proc_name_index] = strdup(name_string);

            if (!mca_pls_rsh_component.debug) {
                 /* setup stdin */
                int fd = open("/dev/null", O_RDWR);
                dup2(fd, 0);
                close(fd);
            }

            /* close all file descriptors w/ exception of stdin/stdout/stderr */
            for(fd=3; fd<fdmax; fd++)
                close(fd);

            /* Set signal handlers back to the default.  Do this close
                to the execve() because the event library may (and likely
                will) reset them.  If we don't do this, the event
                library may have left some set that, at least on some
                OS's, don't get reset via fork() or exec().  Hence, the
                orted could be unkillable (for example). */

            set_handler_default(SIGTERM);
            set_handler_default(SIGINT);
            set_handler_default(SIGHUP);
            set_handler_default(SIGPIPE);
            set_handler_default(SIGCHLD);
            
            /* Unblock all signals, for many of the same reasons that
                we set the default handlers, above.  This is noticable
                on Linux where the event library blocks SIGTERM, but we
                don't want that blocked by the orted (or, more
                specifically, we don't want it to be blocked by the
                orted and then inherited by the ORTE processes that it
                forks, making them unkillable by SIGTERM). */
            sigprocmask(0, 0, &sigs);
            sigprocmask(SIG_UNBLOCK, &sigs, 0);
            
            /* setup environment */
            env = opal_argv_copy(environ);
            var = mca_base_param_environ_variable("seed",NULL,NULL);
            opal_setenv(var, "0", true, &env);

            /* exec the daemon */
            if (mca_pls_rsh_component.debug) {
                param = opal_argv_join(exec_argv, ' ');
                if (NULL != param) {
                    char* env_array = opal_argv_join( env, ' ' );
                    opal_output(0, "pls:rsh: executing: (%s) %s [%s]",
                                exec_path, param, env_array);
                    free(param); free(env_array);
                }
            }
            execve(exec_path, exec_argv, env);
            opal_output(0, "pls:rsh: execv of %s failed with errno=%s(%d)\n",
                        exec_path, strerror(errno), errno);
            exit(-1);

        } else { /* father */
            OPAL_THREAD_LOCK(&mca_pls_rsh_component.lock);
            /* JJH Bug:
             * If we are in '--debug-daemons' we keep the ssh connection 
             * alive for the span of the run. If we use this option 
             * AND we launch on more than "num_concurrent" machines
             * then we will deadlock. No connections are terminated 
             * until the job is complete, no job is started
             * since all the orteds are waiting for all the others
             * to come online, and the others ore not launched because
             * we are waiting on those that have started to terminate
             * their ssh tunnels. :(
             */
            if (mca_pls_rsh_component.num_children++ >=
                mca_pls_rsh_component.num_concurrent) {
                opal_condition_wait(&mca_pls_rsh_component.cond, &mca_pls_rsh_component.lock);
            }
            OPAL_THREAD_UNLOCK(&mca_pls_rsh_component.lock);
            
            /* setup callback on sigchild - wait until setup above is complete
             * as the callback can occur in the call to orte_wait_cb
             */
            orte_wait_cb(pid, orte_pls_rsh_wait_daemon, dmn);

            /* if required - add delay to avoid problems w/ X11 authentication */
            if (mca_pls_rsh_component.debug && mca_pls_rsh_component.delay) {
                sleep(mca_pls_rsh_component.delay);
            }
            vpid++;
        }
        free(name);
    }
    
    /* all done, so store the daemon info on the registry */
    if (ORTE_SUCCESS != (rc = orte_pls_base_store_active_daemons(&active_daemons))) {
        ORTE_ERROR_LOG(rc);
    }

cleanup:
    OBJ_RELEASE(map);

    if (NULL != lib_base) {
        free(lib_base);
    }
    if (NULL != bin_base) {
        free(bin_base);
    }

    if (NULL != jobid_string) free(jobid_string);  /* done with this variable */
    if (NULL != argv) opal_argv_free(argv);

    return rc;
}