/*
* Remove down nodes from hostlist wcoll using "nodeupdown_is_down_node"
* on each member of wcoll. Supposedly, it doesn't matter whether you
* pass in the canonical or altname.
*/
static void
remove_all_down_nodes(hostlist_t wcoll)
{
nodeupdown_t nh = NULL;
char * host = NULL;
hostlist_iterator_t i = NULL;
if ((nh = nodeupdown_handle_create()) == NULL)
errx("%p: Unable to create nodeupdown handle.\n");
#if HAVE_NODEUPDOWN_LOAD_DATA_6
if (nodeupdown_load_data(nh, NULL, NULL, NULL, 0, 0) < 0)
#else
if (nodeupdown_load_data(nh, NULL, 0, 0, NULL) < 0)
#endif
errx("%p: nodeupdown: %s\n", nodeupdown_errormsg(nh));
i = hostlist_iterator_create(wcoll);
while ((host = hostlist_next(i))) {
if (nodeupdown_is_node_down(nh, host) > 0)
hostlist_remove(i);
free(host);
}
hostlist_iterator_destroy(i);
if (nodeupdown_handle_destroy(nh) < 0)
err("%p: nodeupdown_handle_destroy: %s\n", nodeupdown_errormsg(nh));
return;
}
/*
* find_child_switches creates an array of indexes to the
* immediate descendants of switch sw.
*/
static void _find_child_switches (int sw)
{
int i;
int cldx; /* index into array of child switches */
hostlist_iterator_t hi;
hostlist_t swlist;
char *swname;
swlist = hostlist_create(switch_record_table[sw].switches);
switch_record_table[sw].num_switches = hostlist_count(swlist);
switch_record_table[sw].switch_index =
xmalloc(switch_record_table[sw].num_switches
* sizeof(uint16_t));
hi = hostlist_iterator_create(swlist);
cldx = 0;
while ((swname = hostlist_next(hi))) {
/* Find switch whose name is the name of this child.
* and add its index to child index array */
for (i=0; i<switch_record_cnt; i++) {
if (strcmp(swname, switch_record_table[i].name) == 0) {
switch_record_table[sw].switch_index[cldx] = i;
switch_record_table[i].parent = sw;
cldx++;
break;
}
}
free(swname);
}
hostlist_iterator_destroy(hi);
hostlist_destroy(swlist);
}
/*
* hostlist2bitmap - given a hostlist, build a bitmap representation
* IN hl - hostlist
* IN best_effort - if set don't return an error on invalid node name entries
* OUT bitmap - set to bitmap, may not have all bits set on error
* RET 0 if no error, otherwise EINVAL
*/
extern int hostlist2bitmap (hostlist_t hl, bool best_effort, bitstr_t **bitmap)
{
int rc = SLURM_SUCCESS;
bitstr_t *my_bitmap;
char *name;
hostlist_iterator_t hi;
FREE_NULL_BITMAP(*bitmap);
my_bitmap = (bitstr_t *) bit_alloc (node_record_count);
*bitmap = my_bitmap;
hi = hostlist_iterator_create(hl);
while ((name = hostlist_next(hi)) != NULL) {
struct node_record *node_ptr;
node_ptr = _find_node_record(name, best_effort, true);
if (node_ptr) {
bit_set (my_bitmap, (bitoff_t) (node_ptr -
node_record_table_ptr));
} else {
error ("hostlist2bitmap: invalid node specified %s",
name);
if (!best_effort)
rc = EINVAL;
}
free (name);
}
hostlist_iterator_destroy(hi);
return rc;
}
static void
_stat(hash_t hstatus, const char *nodes)
{
hostlist_iterator_t hlitr;
hostlist_t hlnodes;
char *node;
char *str;
assert(hstatus);
if (!(hlnodes = hostlist_create(nodes))) {
perror("hostlist_create");
exit(1);
}
if (!(hlitr = hostlist_iterator_create(hlnodes))) {
perror("hostlist_iterator_create");
exit(1);
}
while ((node = hostlist_next(hlitr))) {
if ((str = hash_find(hstatus, node)))
printf("%s: %s\n", node, str);
else
printf("%s: %s\n", node, "invalid hostname");
free(node);
}
hostlist_iterator_destroy(hlitr);
hostlist_destroy(hlnodes);
}
/* Send a signal RPC to a list of nodes */
static void _send_sig(uint32_t job_id, uint32_t step_id, uint16_t signal,
char *nodelist)
{
agent_arg_t *agent_args;
signal_tasks_msg_t *signal_tasks_msg;
hostlist_iterator_t hi;
char *host;
struct node_record *node_ptr;
signal_tasks_msg = xmalloc(sizeof(signal_tasks_msg_t));
signal_tasks_msg->job_id = job_id;
signal_tasks_msg->job_step_id = step_id;
signal_tasks_msg->signal = signal;
agent_args = xmalloc(sizeof(agent_arg_t));
agent_args->msg_type = REQUEST_SIGNAL_TASKS;
agent_args->retry = 1;
agent_args->msg_args = signal_tasks_msg;
agent_args->hostlist = hostlist_create(nodelist);
agent_args->node_count = hostlist_count(agent_args->hostlist);
agent_args->protocol_version = SLURM_PROTOCOL_VERSION;
hi = hostlist_iterator_create(agent_args->hostlist);
while ((host = hostlist_next(hi))) {
if ((node_ptr = find_node_record(host)) &&
(agent_args->protocol_version > node_ptr->protocol_version))
agent_args->protocol_version =
node_ptr->protocol_version;
free(host);
}
hostlist_iterator_destroy(hi);
agent_queue_request(agent_args);
}
hostlist_iterator_t
wrap_hostlist_iterator_create(WRAPPERS_ARGS, hostlist_t hl)
{
hostlist_iterator_t rv;
assert(file && function && hl);
if (!(rv = hostlist_iterator_create(hl)))
WRAPPERS_ERR_ERRNO("hostlist_iterator_create");
return rv;
}
/*
* Initialize an alpsc_ev_app_t
*/
static void _initialize_event(alpsc_ev_app_t *event,
struct job_record *job_ptr,
struct step_record *step_ptr,
alpsc_ev_app_state_e state)
{
hostlist_t hl;
hostlist_iterator_t hlit;
char *node;
int rv;
event->apid = SLURM_ID_HASH(job_ptr->job_id, step_ptr->step_id);
event->uid = job_ptr->user_id;
event->app_name = xstrdup(step_ptr->name);
event->batch_id = xmalloc(20); // More than enough to hold max uint32
snprintf(event->batch_id, 20, "%"PRIu32, job_ptr->job_id);
event->state = state;
event->nodes = NULL;
event->num_nodes = 0;
// Fill in nodes and num_nodes
if (step_ptr->step_layout) {
hl = hostlist_create(step_ptr->step_layout->node_list);
if (hl == NULL) {
return;
}
hlit = hostlist_iterator_create(hl);
if (hlit == NULL) {
hostlist_destroy(hl);
return;
}
event->nodes = xmalloc(step_ptr->step_layout->node_cnt
* sizeof(int32_t));
while ((node = hostlist_next(hlit)) != NULL) {
rv = sscanf(node, "nid%"SCNd32,
&event->nodes[event->num_nodes]);
if (rv) {
event->num_nodes++;
} else {
debug("%s: couldn't parse node %s, skipping",
__func__, node);
}
free(node);
}
hostlist_iterator_destroy(hlit);
hostlist_destroy(hl);
} else {
// TODO: do we have to worry about batch scripts?
}
return;
}
static int
_delete_all (hostlist_t hl, hostlist_t dl)
{
int rc = 0;
char * host = NULL;
hostlist_iterator_t i = hostlist_iterator_create (dl);
while ((host = hostlist_next (i))) {
rc += hostlist_delete_host (hl, host);
free (host);
}
hostlist_iterator_destroy (i);
return (rc);
}
int switch_p_build_jobinfo(switch_jobinfo_t *switch_job,
slurm_step_layout_t *step_layout, char *network)
{
sw_gen_step_info_t *gen_step_info = (sw_gen_step_info_t *) switch_job;
sw_gen_node_info_t *gen_node_info;
sw_gen_node_t *node_ptr;
hostlist_t hl = NULL;
hostlist_iterator_t hi;
char *host = NULL;
int i, j;
if (debug_flags & DEBUG_FLAG_SWITCH)
info("switch_p_build_jobinfo() starting");
xassert(gen_step_info);
xassert(gen_step_info->magic == SW_GEN_STEP_INFO_MAGIC);
hl = hostlist_create(step_layout->node_list);
if (!hl)
fatal("hostlist_create(%s): %m", step_layout->node_list);
gen_step_info->node_cnt = hostlist_count(hl);
gen_step_info->node_array = xmalloc(sizeof(sw_gen_node_t *) *
gen_step_info->node_cnt);
hi = hostlist_iterator_create(hl);
for (i = 0; (host = hostlist_next(hi)); i++) {
node_ptr = xmalloc(sizeof(sw_gen_node_t));
gen_step_info->node_array[i] = node_ptr;
node_ptr->node_name = xstrdup(host);
gen_node_info = _find_node(host);
if (gen_node_info) { /* Copy node info to this step */
node_ptr->ifa_cnt = gen_node_info->ifa_cnt;
node_ptr->ifa_array = xmalloc(sizeof(sw_gen_node_t *) *
node_ptr->ifa_cnt);
for (j = 0; j < node_ptr->ifa_cnt; j++) {
node_ptr->ifa_array[j] =
xmalloc(sizeof(sw_gen_node_t));
node_ptr->ifa_array[j]->ifa_addr = xstrdup(
gen_node_info->ifa_array[j]->ifa_addr);
node_ptr->ifa_array[j]->ifa_family = xstrdup(
gen_node_info->ifa_array[j]->ifa_family);
node_ptr->ifa_array[j]->ifa_name = xstrdup(
gen_node_info->ifa_array[j]->ifa_name);
}
}
free(host);
}
hostlist_iterator_destroy(hi);
hostlist_destroy(hl);
return SLURM_SUCCESS;
}
static void
_onoff(hash_t hstatus, const char *nodes, const char *state)
{
hostlist_iterator_t hlitr;
hostlist_t hlnodes;
char *node;
char *str;
assert(hstatus);
if (!(hlnodes = hostlist_create(nodes))) {
perror("hostlist_create");
exit(1);
}
if (!(hlitr = hostlist_iterator_create(hlnodes))) {
perror("hostlist_iterator_create");
exit(1);
}
while ((node = hostlist_next(hlitr))) {
if ((str = hash_find(hstatus, node))) {
printf("%s: %s\n", node, OK_STATUS);
hash_remove(hstatus, node);
if (!hash_insert(hstatus, (void *)node, (void *)state)) {
perror("hash_insert");
exit(1);
}
/* XXX: Don't free 'node' here, it needs to be alloc'd for
* the hash key. It's a mem-leak. Fix later.
*/
} else {
printf("%s: %s\n", node, "invalid hostname");
free(node);
}
}
hostlist_iterator_destroy(hlitr);
hostlist_destroy(hlnodes);
}
extern List setup_cluster_list_with_inx(mysql_conn_t *mysql_conn,
slurmdb_job_cond_t *job_cond,
void **curr_cluster)
{
List local_cluster_list = NULL;
time_t now = time(NULL);
MYSQL_RES *result = NULL;
MYSQL_ROW row;
hostlist_t temp_hl = NULL;
hostlist_iterator_t h_itr = NULL;
char *query = NULL;
int dims = 0;
if (!job_cond || !job_cond->used_nodes)
return NULL;
if (!job_cond->cluster_list
|| list_count(job_cond->cluster_list) != 1) {
error("If you are doing a query against nodes "
"you must only have 1 cluster "
"you are asking for.");
return NULL;
}
/* get the dimensions of this cluster so we know how to deal
with the hostlists */
query = xstrdup_printf("select dimensions, flags from %s where "
"name='%s'",
cluster_table,
(char *)list_peek(job_cond->cluster_list));
debug4("%d(%s:%d) query\n%s",
mysql_conn->conn, THIS_FILE, __LINE__, query);
if (!(result = mysql_db_query_ret(mysql_conn, query, 0))) {
xfree(query);
return NULL;
}
xfree(query);
if (!(row = mysql_fetch_row(result))) {
error("Couldn't get the dimensions of cluster '%s'.",
(char *)list_peek(job_cond->cluster_list));
mysql_free_result(result);
return NULL;
}
/* On a Cray System when dealing with hostlists as we are here
this always needs to be 1.
*/
if (slurm_atoul(row[1]) & CLUSTER_FLAG_CRAY_A)
dims = 1;
else
dims = atoi(row[0]);
mysql_free_result(result);
temp_hl = hostlist_create_dims(job_cond->used_nodes, dims);
if (hostlist_count(temp_hl) <= 0) {
error("we didn't get any real hosts to look for.");
goto no_hosts;
}
h_itr = hostlist_iterator_create(temp_hl);
query = xstrdup_printf("select cluster_nodes, time_start, "
"time_end from \"%s_%s\" where node_name='' "
"&& cluster_nodes !=''",
(char *)list_peek(job_cond->cluster_list),
event_table);
if (job_cond->usage_start) {
if (!job_cond->usage_end)
job_cond->usage_end = now;
xstrfmtcat(query,
" && ((time_start < %ld) "
"&& (time_end >= %ld || time_end = 0))",
job_cond->usage_end, job_cond->usage_start);
}
if (debug_flags & DEBUG_FLAG_DB_JOB)
DB_DEBUG(mysql_conn->conn, "query\n%s", query);
if (!(result = mysql_db_query_ret(mysql_conn, query, 0))) {
xfree(query);
goto no_hosts;
}
xfree(query);
local_cluster_list = list_create(_destroy_local_cluster);
while ((row = mysql_fetch_row(result))) {
char *host = NULL;
int loc = 0;
local_cluster_t *local_cluster =
xmalloc(sizeof(local_cluster_t));
local_cluster->hl = hostlist_create_dims(row[0], dims);
local_cluster->start = slurm_atoul(row[1]);
local_cluster->end = slurm_atoul(row[2]);
local_cluster->asked_bitmap =
bit_alloc(hostlist_count(local_cluster->hl));
while ((host = hostlist_next_dims(h_itr, dims))) {
if ((loc = hostlist_find(
local_cluster->hl, host)) != -1)
bit_set(local_cluster->asked_bitmap, loc);
free(host);
}
hostlist_iterator_reset(h_itr);
if (bit_ffs(local_cluster->asked_bitmap) != -1) {
list_append(local_cluster_list, local_cluster);
if (local_cluster->end == 0) {
local_cluster->end = now;
(*curr_cluster) = local_cluster;
} else if (!(*curr_cluster)
|| (((local_cluster_t *)(*curr_cluster))->end
< local_cluster->end)) {
(*curr_cluster) = local_cluster;
}
} else
_destroy_local_cluster(local_cluster);
}
mysql_free_result(result);
if (!list_count(local_cluster_list)) {
FREE_NULL_LIST(local_cluster_list);
local_cluster_list = NULL;
goto no_hosts;
}
no_hosts:
hostlist_iterator_destroy(h_itr);
hostlist_destroy(temp_hl);
return local_cluster_list;
}
/*
* setup_cluster_nodes - get cluster record list within requested
* time period with used nodes. Used for deciding whether a nodelist is
* overlapping with the required nodes.
*/
extern cluster_nodes_t *
setup_cluster_nodes(pgsql_conn_t *pg_conn, slurmdb_job_cond_t *job_cond)
{
DEF_VARS;
cluster_nodes_t *cnodes = NULL;
time_t now = time(NULL);
hostlist_t temp_hl = NULL;
hostlist_iterator_t h_itr = NULL;
if (!job_cond || !job_cond->used_nodes)
return NULL;
if (!job_cond->cluster_list || list_count(job_cond->cluster_list) != 1) {
error("If you are doing a query against nodes "
"you must only have 1 cluster "
"you are asking for.");
return NULL;
}
temp_hl = hostlist_create(job_cond->used_nodes);
if (!hostlist_count(temp_hl)) {
error("we didn't get any real hosts to look for.");
hostlist_destroy(temp_hl);
return NULL;
}
query = xstrdup_printf("SELECT cluster_nodes, time_start, "
"time_end FROM %s.%s WHERE node_name='' "
"AND cluster_nodes !=''",
(char *)list_peek(job_cond->cluster_list),
event_table);
if (job_cond->usage_start) {
if (!job_cond->usage_end)
job_cond->usage_end = now;
xstrfmtcat(query, " AND ((time_start<%ld) "
"AND (time_end>=%ld OR time_end=0))",
job_cond->usage_end, job_cond->usage_start);
}
result = DEF_QUERY_RET;
if (!result) {
hostlist_destroy(temp_hl);
return NULL;
}
h_itr = hostlist_iterator_create(temp_hl);
cnodes = xmalloc(sizeof(cluster_nodes_t));
cnodes->cluster_list = list_create(_destroy_local_cluster);
FOR_EACH_ROW {
char *host = NULL;
int loc = 0;
local_cluster_t *local_cluster =
xmalloc(sizeof(local_cluster_t));
local_cluster->hl = hostlist_create(ROW(0));
local_cluster->start = atoi(ROW(1));
local_cluster->end = atoi(ROW(2));
local_cluster->asked_bitmap =
bit_alloc(hostlist_count(local_cluster->hl));
while((host = hostlist_next(h_itr))) {
if ((loc = hostlist_find(
local_cluster->hl, host)) != -1)
bit_set(local_cluster->asked_bitmap, loc);
free(host);
}
hostlist_iterator_reset(h_itr);
if (bit_ffs(local_cluster->asked_bitmap) != -1) {
list_append(cnodes->cluster_list, local_cluster);
if (local_cluster->end == 0) {
local_cluster->end = now;
cnodes->curr_cluster = local_cluster;
}
} else
_destroy_local_cluster(local_cluster);
} END_EACH_ROW;
PQclear(result);
hostlist_iterator_destroy(h_itr);
if (!list_count(cnodes->cluster_list)) {
destroy_cluster_nodes(cnodes);
cnodes = NULL;
}
hostlist_destroy(temp_hl);
return cnodes;
}
static void
_nodes_setup (void)
{
hostlist_iterator_t itr = NULL;
char *host = NULL;
int i = 0;
assert (fds);
assert (fds_count);
assert (!nodes);
assert (nodes_count);
assert (!nodes_index);
if (!(nodes = list_create ((ListDelF)free)))
err_exit ("list_create: %s", strerror (errno));
if (!(nodes_index = hash_create (nodes_count,
(hash_key_f)hash_key_string,
(hash_cmp_f)strcmp,
NULL)))
err_exit ("hash_create: %s", strerror (errno));
if (!(itr = hostlist_iterator_create (conf.hosts)))
err_exit ("hostlist_iterator_create: %s", strerror (errno));
while ((host = hostlist_next (itr)))
{
struct ipmidetectd_info *info = NULL;
struct hostent *h;
char *tmpstr;
char *ip;
int len;
char *host_copy = NULL;
char *host_ptr;
uint16_t port = RMCP_PRIMARY_RMCP_PORT;
if (!(info = (struct ipmidetectd_info *)malloc (sizeof (struct ipmidetectd_info))))
err_exit ("malloc: %s", strerror (errno));
memset (info, '\0', sizeof (struct ipmidetectd_info));
if (strchr (host, ':'))
{
char *ptr;
if (!(host_copy = strdup (host)))
err_exit ("strdup: %s", strerror (errno));
if ((ptr = strchr (host_copy, ':')))
{
char *endptr;
int tmp;
*ptr = '\0';
ptr++;
errno = 0;
tmp = strtol (ptr, &endptr, 0);
if (errno
|| endptr[0] != '\0'
|| tmp <= 0
|| tmp > USHRT_MAX)
err_exit ("invalid port specified: %s", host);
port = tmp;
}
host_ptr = host_copy;
}
else
host_ptr = host;
if (!(info->hostname = strdup (host_ptr)))
err_exit ("strdup: %s", strerror (errno));
/* Use random number for starting sequence number to avoid probability of
* duplicates and "hanging" BMC issue.
*/
if ((len = ipmi_get_random (&(info->sequence_number),
sizeof (info->sequence_number))) < 0)
err_exit ("ipmi_get_random: %s", strerror (errno));
if (len != sizeof (info->sequence_number))
err_exit ("ipmi_get_random: invalid len returned");
info->fd = fds[i/IPMIDETECTD_NODES_PER_SOCKET];
if (!(h = gethostbyname (host_ptr)))
{
#if HAVE_HSTRERROR
err_exit ("gethostbyname: %s", hstrerror (h_errno));
#else /* !HAVE_HSTRERROR */
err_exit ("gethostbyname: h_errno = %d", h_errno);
#endif /* !HAVE_HSTRERROR */
}
info->destaddr.sin_family = AF_INET;
info->destaddr.sin_addr = *((struct in_addr *)h->h_addr);
info->destaddr.sin_port = htons (port);
free (host_copy);
free (host);
if (!list_append (nodes, info))
err_exit ("list_append: %s", strerror (errno));
if (!(tmpstr = inet_ntoa (info->destaddr.sin_addr)))
err_exit ("inet_ntoa: %s", strerror (errno)); /* strerror? */
if (!(ip = strdup (tmpstr)))
err_exit ("strdup: %s", strerror (errno));
if (hash_find (nodes_index, ip))
err_exit ("Duplicate host ip: %s", ip);
if (!hash_insert (nodes_index, ip, info))
err_exit ("hash_insert: %s", strerror (errno));
i++;
}
hostlist_iterator_destroy (itr);
}
static void
_prompt_loop(void)
{
char buf[128];
char bufnode[128];
hash_t hstatus = NULL;
hostlist_t hl = NULL;
hostlist_iterator_t hlitr = NULL;
char *node;
assert(hostname);
if (!(hstatus = hash_create(HASH_SIZE,
(hash_key_f)hash_key_string,
(hash_cmp_f)strcmp,
(hash_del_f)NULL))) {
perror("hash_create");
exit(1);
}
if (!(hl = hostlist_create(hostname))) {
perror("hostlist_create");
exit(1);
}
if (!(hlitr = hostlist_iterator_create(hl))) {
perror("hostlist_iterator");
exit(1);
}
/* all nodes begin as off */
while ((node = hostlist_next(hlitr))) {
if (!hash_insert(hstatus, (void *)node, OFF_STATUS)) {
perror("hash_insert");
exit(1);
}
/* XXX: Don't free 'node' here, it needs to be alloc'd for
* the hash key. It's a mem-leak. Fix later.
*/
}
hostlist_iterator_destroy(hlitr);
hostlist_destroy(hl);
while (1) {
if (xreadline(CMD_PROMPT, buf, sizeof(buf)) == NULL) {
break;
} else if (strlen(buf) == 0) {
continue;
} else if (!strcmp(buf, "quit")) {
break;
} else if (!strcmp(buf, "stat")) {
_stat(hstatus, hostname);
} else if (sscanf(buf, "stat %s", bufnode) == 1) {
_stat(hstatus, bufnode);
} else if (!strcmp(buf, "on")) {
_onoff(hstatus, hostname, ON_STATUS);
} else if (sscanf(buf, "on %s", bufnode) == 1) {
_onoff(hstatus, bufnode, ON_STATUS);
} else if (!strcmp(buf, "off")) {
_onoff(hstatus, hostname, OFF_STATUS);
} else if (sscanf(buf, "off %s", bufnode) == 1) {
_onoff(hstatus, bufnode, OFF_STATUS);
} else
printf("unknown command - type \"help\"\n");
}
hash_destroy(hstatus);
}
int
main (int argc, char *argv[])
{
char *dir = NULL;
char *spec, *host;
char *nspec = NULL;
int c, i;
int nopt = 0;
int vopt = 0;
int fopt = 0;
int aopt = 0;
int dopt = 0;
int rfd = -1, wfd = -1;
Opt o;
diod_log_init (argv[0]);
o = opt_create ();
opterr = 0;
while ((c = GETOPT (argc, argv, OPTIONS, longopts)) != -1) {
switch (c) {
case 'f': /* --fake-mount */
fopt = 1;
break;
case 'n': /* --no-mtab */
nopt = 1;
break;
case 'v': /* --verbose */
vopt++;
break;
case 'o': /* --options OPT[,OPT]... */
opt_addf (o, "%s", optarg);
break;
case 'a': /* --9nbd-attach */
aopt++;
break;
case 'd': /* --9nbd-detach */
dopt++;
break;
default:
usage ();
}
}
/* Take care of 9nbd operations and exit.
*/
if (aopt) {
_nbd_attach (o, argc - optind, argv + optind, nopt, vopt);
exit (0);
}
if (dopt) {
_nbd_detach (o, argc - optind, argv + optind, nopt, vopt);
exit (0);
}
if (optind != argc - 2)
usage ();
if (geteuid () != 0)
msg_exit ("you must be root");
spec = argv[optind++];
dir = argv[optind++];
host = _parse_spec (spec, o);
_verify_mountpoint (dir);
/* Remount - only pass mount flags into the VFS for an existing mount.
* Take care of it here and exit.
*/
if (opt_find (o, "remount")) {
if (opt_check_allowed_csv (o, "ro,rw,aname,remount"))
msg_exit ("-oremount can only be used with ro,rw");
_diod_remount (o, spec, dir, vopt, fopt);
goto done;
}
/* Ensure uname and access are set, and to diod-compatible values.
* The uname user becomes the euid which will be used by munge auth.
*/
_parse_uname_access (o);
if (seteuid (_uname2uid (opt_find (o, "uname"))) < 0)
err_exit ("seteuid");
/* We require -otrans=fd because auth occurs in user space, then live fd
* is passed to the kernel via -orfdno,wfdno.
*/
if (!opt_find (o, "trans"))
opt_addf (o, "trans=%s", "fd");
else if (!opt_find (o, "trans=fd"))
msg_exit ("only -otrans=fd transport is supported");
/* Set msize if not already set. Validate it later.
*/
if (!opt_find (o, "msize"))
opt_addf (o, "msize=%d", DIOD_DEFAULT_MSIZE);
/* Only .L version is supported.
*/
if (!opt_find (o, "version"))
opt_addf (o, "version=%s", "9p2000.L");
else if (!opt_find (o, "version=9p2000.L"))
msg_exit ("only -oversion=9p2000.L is supported (little p, big L)");
/* Set debug level.
*/
if (!opt_find (o, "debug"))
opt_addf (o, "debug=%d", 0x1); /* send errors to dmesg */
/* Set rwdepth (number of concurrent reads with buffer > msize).
* N.B. this option is not upstream yet but unknown options are ignored.
*/
if (!opt_find (o, "rwdepth"))
opt_addf (o, "rwdepth=%d", 1);
/* Server is on an inherited file descriptor.
* For testing, we start server on a socketpair duped to fd 0.
*/
if (opt_find (o, "rfdno") || opt_find (o, "wfdno")) {
if (!opt_scanf (o, "rfdno=%d", &rfd) || !opt_scanf (o, "wfdno=%d",&wfd))
msg_exit ("-orfdno,wfdno must be used together");
nopt = 1; /* force no mtab */
/* Connect to server on UNIX domain socket
*/
} else if (host[0] == '/') {
if (opt_find (o, "port"))
msg_exit ("-oport won't work with UNIX domain socket");
if ((rfd = diod_sock_connect_unix (host, 0)) < 0)
exit (1);
wfd = rfd;
opt_addf (o, "rfdno=%d", rfd);
opt_addf (o, "wfdno=%d", wfd);
/* Connect to server on IANA port (or user-specified) and host.
*/
} else {
char *port = opt_find (o, "port");
hostlist_iterator_t hi;
hostlist_t hl;
char *h;
if (!port)
port = "564";
if (!(hl = hostlist_create (host)))
msg_exit ("error parsing host string: %s", host);
if (!(hi = hostlist_iterator_create (hl)))
msg_exit ("out of memory");
while ((h = hostlist_next (hi))) {
if (vopt)
msg ("trying to connect to %s:%s", h, port);
if ((rfd = diod_sock_connect_inet (h, port, DIOD_SOCK_QUIET)) >= 0)
break;
}
if (h) { /* create new 'spec' string identifying successful host */
char *p = strchr (spec , ':');
int len = strlen (h) + (p ? strlen (p) : 0) + 1;
if (!(nspec = malloc (len)))
msg_exit ("out of memory");
snprintf (nspec, len, "%s%s", h, p ? p : "");
}
hostlist_destroy (hl);
if (rfd < 0)
msg_exit ("could not connect to server(s), giving up");
wfd = rfd;
opt_delete (o, "port");
opt_addf (o, "rfdno=%d", rfd);
opt_addf (o, "wfdno=%d", wfd);
}
NP_ASSERT (opt_find (o, "trans=fd"));
NP_ASSERT (opt_scanf (o, "msize=%d", &i));
NP_ASSERT (opt_find (o, "version=9p2000.L"));
NP_ASSERT (opt_scanf (o, "debug=%d", &i) || opt_scanf (o, "debug=%x", &i));
NP_ASSERT (opt_scanf (o, "wfdno=%d", &i) && opt_scanf (o, "rfdno=%d", &i));
NP_ASSERT ((opt_find (o, "access=user") && opt_find(o, "uname=root"))
|| (opt_scanf (o, "access=%d", &i) && opt_find(o, "uname")));
NP_ASSERT (!opt_find (o, "port"));
_diod_mount (o, rfd, wfd, nspec ? nspec : spec, dir, vopt, fopt, nopt);
done:
opt_destroy (o);
exit (0);
}
int
pstdout_launch(const char *hostnames, Pstdout_Thread pstdout_func, void *arg)
{
struct pstdout_thread_data **tdata = NULL;
struct pstdout_state pstate;
unsigned int pstate_init = 0;
hostlist_iterator_t hitr = NULL;
hostlist_t h = NULL;
int h_count = 0;
char *host = NULL;
int exit_code = -1;
sighandler_t sighandler_save = NULL;
int sighandler_set = 0;
int rc;
int i;
if (!pstdout_initialized)
{
pstdout_errnum = PSTDOUT_ERR_UNINITIALIZED;
return -1;
}
if (!pstdout_func)
{
pstdout_errnum = PSTDOUT_ERR_PARAMETERS;
return -1;
}
if ((rc = pthread_mutex_lock(&pstdout_launch_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_mutex_lock: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
/* Special case */
if (!hostnames)
{
if (_pstdout_state_init(&pstate, NULL) < 0)
goto cleanup;
pstate_init++;
exit_code = pstdout_func(&pstate, NULL, arg);
pstdout_errnum = PSTDOUT_ERR_SUCCESS;
goto cleanup;
}
if (!(h = hostlist_create(hostnames)))
{
pstdout_errnum = PSTDOUT_ERR_OUTMEM;
goto cleanup;
}
h_count = hostlist_count(h);
/* Sanity check */
if (h_count <= 0)
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "h_count = %d\n", h_count);
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
/* Special case */
if (h_count == 1)
{
if (_pstdout_state_init(&pstate, hostnames) < 0)
goto cleanup;
pstate_init++;
exit_code = pstdout_func(&pstate, hostnames, arg);
pstdout_errnum = PSTDOUT_ERR_SUCCESS;
goto cleanup;
}
if ((sighandler_save = signal(SIGINT, _pstdout_sigint)) == SIG_ERR)
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "signal\n");
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
sighandler_set++;
if (!(hitr = hostlist_iterator_create(h)))
{
pstdout_errnum = PSTDOUT_ERR_OUTMEM;
goto cleanup;
}
if (!(tdata = (struct pstdout_thread_data **)malloc(sizeof(struct pstdout_thread_data *) * h_count)))
{
pstdout_errnum = PSTDOUT_ERR_OUTMEM;
goto cleanup;
}
memset(tdata, '\0', sizeof(struct pstdout_thread_data *) * h_count);
i = 0;
while ((host = hostlist_next(hitr)))
{
if (!(tdata[i] = (struct pstdout_thread_data *)malloc(sizeof(struct pstdout_thread_data))))
{
pstdout_errnum = PSTDOUT_ERR_OUTMEM;
goto cleanup;
}
memset(tdata[i], '\0', sizeof(struct pstdout_thread_data));
if (!(tdata[i]->hostname = strdup(host)))
{
pstdout_errnum = PSTDOUT_ERR_OUTMEM;
goto cleanup;
}
tdata[i]->pstdout_func = pstdout_func;
tdata[i]->arg = arg;
if ((rc = pthread_attr_init(&(tdata[i]->attr))))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_attr_init: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
if ((rc = pthread_attr_setdetachstate(&(tdata[i]->attr), PTHREAD_CREATE_DETACHED)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_attr_setdetachstate: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
free(host);
i++;
}
host = NULL;
hostlist_iterator_destroy(hitr);
hitr = NULL;
hostlist_destroy(h);
h = NULL;
/* Launch threads up to fanout */
for (i = 0; i < h_count; i++)
{
if ((rc = pthread_mutex_lock(&pstdout_threadcount_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_mutex_lock: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
if (pstdout_threadcount == pstdout_fanout)
{
if ((rc = pthread_cond_wait(&pstdout_threadcount_cond, &pstdout_threadcount_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_cond_wait: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
}
if ((rc = pthread_create(&(tdata[i]->tid),
&(tdata[i]->attr),
_pstdout_func_entry,
(void *) tdata[i])))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_create: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
pstdout_threadcount++;
if ((rc = pthread_mutex_unlock(&pstdout_threadcount_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_mutex_unlock: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
}
/* Wait for Threads to finish */
if ((rc = pthread_mutex_lock(&pstdout_threadcount_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_mutex_lock: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
while (pstdout_threadcount > 0)
{
if ((rc = pthread_cond_wait(&pstdout_threadcount_cond, &pstdout_threadcount_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_cond_wait: %s\n", strerror(rc));
pstdout_errnum = PSTDOUT_ERR_INTERNAL;
goto cleanup;
}
}
if (_pstdout_output_consolidated_finish() < 0)
goto cleanup;
/* Determine exit code */
exit_code = 0;
for (i = 0; i < h_count; i++)
{
if (tdata[i]->exit_code > exit_code)
exit_code = tdata[i]->exit_code;
}
cleanup:
/* Cannot pass NULL for key, so just pass dummy key */
list_delete_all(pstdout_consolidated_stdout, _pstdout_consolidated_data_delete_all, "");
list_delete_all(pstdout_consolidated_stderr, _pstdout_consolidated_data_delete_all, "");
if (pstate_init)
_pstdout_state_cleanup(&pstate);
if (tdata)
{
for (i = 0; i < h_count; i++)
{
if (tdata[i])
{
free(tdata[i]->hostname);
pthread_attr_destroy(&(tdata[i]->attr));
free(tdata[i]);
}
}
free(tdata);
}
if (hitr)
hostlist_iterator_destroy(hitr);
if (h)
hostlist_destroy(h);
free(host);
if ((rc = pthread_mutex_unlock(&pstdout_launch_mutex)))
{
if (pstdout_debug_flags & PSTDOUT_DEBUG_STANDARD)
fprintf(stderr, "pthread_mutex_unlock: %s\n", strerror(rc));
/* Don't change error code, just move on */
}
if (sighandler_set)
signal(SIGINT, sighandler_save);
return exit_code;
}
int main(int argc, char *argv[])
{
extern int optind;
extern char *optarg;
int c;
PlugList pl = NULL;
char *hwplugs = NULL;
char *nodelist = NULL;
char *pluglist = NULL;
char *findplug = NULL;
err_init(basename(argv[0]));
while ((c = getopt(argc, argv, "p:f:")) != EOF) {
switch (c) {
case 'p':
hwplugs = optarg;
break;
case 'f':
findplug = optarg;
break;
default:
usage();
}
}
if (argc - optind == 0)
usage();
nodelist = argv[optind++];
if (argc - optind == 1)
pluglist = argv[optind++];
if (argc - optind != 0)
usage();
if (hwplugs) {
hostlist_t hl = hostlist_create(hwplugs);
hostlist_iterator_t itr = hostlist_iterator_create(hl);
List l = list_create((ListDelF)xfree);
char *plug;
while ((plug = hostlist_next(itr)))
list_append(l, xstrdup(plug));
hostlist_iterator_destroy(itr);
hostlist_destroy(hl);
pl = pluglist_create(l);
list_destroy(l);
} else
pl = pluglist_create(NULL);
switch (pluglist_map(pl, nodelist, pluglist)) {
case EPL_DUPNODE:
fprintf(stderr, "duplicate node\n");
break;
case EPL_UNKPLUG:
fprintf(stderr, "unknown plug\n");
break;
case EPL_DUPPLUG:
fprintf(stderr, "duplicate plug\n");
break;
case EPL_NOPLUGS:
fprintf(stderr, "more nodes than plugs\n");
break;
case EPL_NONODES:
fprintf(stderr, "more plugs than nodes\n");
break;
case EPL_SUCCESS:
break;
}
if (findplug) {
Plug *plug = pluglist_find(pl, findplug);
if (plug)
printf("plug=%s node=%s\n", plug->name,
plug->node ? plug->node : "NULL");
else
printf("plug %s: not found\n", findplug);
} else {
PlugListIterator itr = pluglist_iterator_create(pl);
Plug *plug;
while ((plug = pluglist_next(itr))) {
printf("plug=%s node=%s\n", plug->name,
plug->node ? plug->node : "NULL");
}
pluglist_iterator_destroy(itr);
}
exit(0);
}
struct ipmipower_connection *
ipmipower_connection_array_create(const char *hostname, unsigned int *len)
{
char *str = NULL;
int index = 0;
hostlist_t hl = NULL;
hostlist_iterator_t itr = NULL;
struct ipmipower_connection *ics;
int size = sizeof(struct ipmipower_connection);
int hl_count;
int errcount = 0;
int emfilecount = 0;
assert(hostname && len);
*len = 0;
if (!(hl = hostlist_create(hostname)))
{
ipmipower_output(MSG_TYPE_HOSTNAME_INVALID, hostname);
return NULL;
}
if (!(itr = hostlist_iterator_create(hl)))
ierr_exit("hostlist_iterator_create() error");
hostlist_uniq(hl);
hl_count = hostlist_count(hl);
ics = (struct ipmipower_connection *)Malloc(size * hl_count);
memset(ics, '\0', (size * hl_count));
while ((str = hostlist_next(itr)))
{
ics[index].ipmi_fd = -1;
ics[index].ping_fd = -1;
/* cleanup only at the end, gather all error outputs for
* later
*/
if (_connection_setup(&ics[index], str) < 0)
{
if (errno == EMFILE && !emfilecount)
{
cbuf_printf(ttyout, "file descriptor limit reached\n");
emfilecount++;
}
errcount++;
}
free(str);
index++;
}
hostlist_iterator_destroy(itr);
hostlist_destroy(hl);
if (errcount)
{
int i;
for (i = 0; i < hl_count; i++)
{
close(ics[i].ipmi_fd);
close(ics[i].ping_fd);
if (ics[i].ipmi_in)
cbuf_destroy(ics[i].ipmi_in);
if (ics[i].ipmi_out)
cbuf_destroy(ics[i].ipmi_out);
if (ics[i].ping_in)
cbuf_destroy(ics[i].ping_in);
if (ics[i].ping_out)
cbuf_destroy(ics[i].ping_out);
}
Free(ics);
return NULL;
}
*len = hl_count;
return ics;
}
int
main (int argc, char *argv[])
{
char *server = NULL;
int msize = 65536;
uid_t uid = geteuid ();
int topt = 0;
Npcfsys *fs = NULL;
Npcfid *fid, *afid, *root;
int c, fd;
char buf[80], *host, *p;
hostlist_t hl;
hostlist_iterator_t itr;
int lopt = 0;
diod_log_init (argv[0]);
opterr = 0;
while ((c = GETOPT (argc, argv, OPTIONS, longopts)) != -1) {
switch (c) {
case 's': /* --server HOST[:PORT] or /path/to/socket */
server = optarg;
break;
case 'm': /* --msize SIZE */
msize = strtoul (optarg, NULL, 10);
break;
case 'u': /* --uid UID */
uid = strtoul (optarg, NULL, 10);
break;
case 't': /* --timeout SECS */
topt = strtoul (optarg, NULL, 10);
break;
case 'l': /* --long */
lopt = 1;
break;
default:
usage ();
}
}
if (signal (SIGPIPE, SIG_IGN) == SIG_ERR)
err_exit ("signal");
if (signal (SIGALRM, sigalarm) == SIG_ERR)
err_exit ("signal");
if (topt > 0)
alarm (topt);
if ((fd = diod_sock_connect (server, 0)) < 0)
exit (1);
if (!(fs = npc_start (fd, fd, msize, 0)))
errn_exit (np_rerror (), "error negotiating protocol with server");
if (!(afid = npc_auth (fs, "ctl", uid, diod_auth)) && np_rerror () != 0)
errn_exit (np_rerror (), "error authenticating to server");
if (!(root = npc_attach (fs, afid, "ctl", uid)))
errn_exit (np_rerror (), "error attaching to aname=ctl");
if (!(fid = npc_open_bypath (root, "connections", O_RDONLY)))
errn_exit (np_rerror (), "open connections");
if (!(hl = hostlist_create (NULL)))
err_exit ("hostlist_create");
while (npc_gets (fid, buf, sizeof(buf))) {
if ((p = strchr (buf, ' ')))
*p = '\0';
if (!lopt && (p = strchr (buf, '.')))
*p = '\0';
if (!hostlist_push_host (hl, buf))
err_exit ("hostlist_push_host");
}
hostlist_uniq (hl);
if (lopt) {
if (!(itr = hostlist_iterator_create (hl)))
err_exit ("hostlist_iterator_create");
while ((host = hostlist_next (itr)))
printf ("%s\n", host);
hostlist_iterator_destroy (itr);
} else {
char s[1024];
if (hostlist_ranged_string (hl, sizeof (s), s) < 0)
msg_exit ("hostlist output would be too long (use -l)");
printf ("%s\n", s);
}
hostlist_destroy (hl);
if (npc_clunk (fid) < 0)
errn_exit (np_rerror (), "clunk connections");
if (npc_clunk (root) < 0)
errn_exit (np_rerror (), "error clunking ctl");
if (npc_clunk (afid) < 0)
errn_exit (np_rerror (), "error clunking afid");
npc_finish (fs);
exit(0);
}
/* use specific set run tasks on each host listed in hostfile
* XXX: Need to handle over-subscribe.
*/
static int _task_layout_hostfile(slurm_step_layout_t *step_layout,
const char *arbitrary_nodes)
{
int i=0, j, taskid = 0, task_cnt=0;
hostlist_iterator_t itr = NULL, itr_task = NULL;
char *host = NULL;
hostlist_t job_alloc_hosts = NULL;
hostlist_t step_alloc_hosts = NULL;
int step_inx = 0, step_hosts_cnt = 0;
struct node_record **step_hosts_ptrs = NULL;
struct node_record *host_ptr = NULL;
debug2("job list is %s", step_layout->node_list);
if (!arbitrary_nodes) {
error("no hostlist given for arbitrary dist");
return SLURM_ERROR;
}
debug2("list is %s", arbitrary_nodes);
step_alloc_hosts = hostlist_create(arbitrary_nodes);
if (hostlist_count(step_alloc_hosts) != step_layout->task_cnt) {
error("Asked for %u tasks have %d in the nodelist. "
"Check your nodelist, or set the -n option to be %d",
step_layout->task_cnt,
hostlist_count(step_alloc_hosts),
hostlist_count(step_alloc_hosts));
hostlist_destroy(step_alloc_hosts);
return SLURM_ERROR;
}
job_alloc_hosts = hostlist_create(step_layout->node_list);
itr = hostlist_iterator_create(job_alloc_hosts);
itr_task = hostlist_iterator_create(step_alloc_hosts);
/*
* Build array of pointers so that we can do pointer comparisons rather
* than strcmp's on nodes.
*/
step_hosts_cnt = hostlist_count(step_alloc_hosts);
step_hosts_ptrs = xmalloc(sizeof(struct node_record *) *
step_hosts_cnt);
step_inx = 0;
while((host = hostlist_next(itr_task))) {
step_hosts_ptrs[step_inx++] = find_node_record_no_alias(host);
free(host);
}
while((host = hostlist_next(itr))) {
host_ptr = find_node_record(host);
step_layout->tasks[i] = 0;
for (step_inx = 0; step_inx < step_hosts_cnt; step_inx++) {
if (host_ptr == step_hosts_ptrs[step_inx]) {
step_layout->tasks[i]++;
task_cnt++;
}
if (task_cnt >= step_layout->task_cnt)
break;
}
debug3("%s got %u tasks", host, step_layout->tasks[i]);
if (step_layout->tasks[i] == 0)
goto reset_hosts;
step_layout->tids[i] = xmalloc(sizeof(uint32_t)
* step_layout->tasks[i]);
taskid = 0;
j = 0;
for (step_inx = 0; step_inx < step_hosts_cnt; step_inx++) {
if (host_ptr == step_hosts_ptrs[step_inx]) {
step_layout->tids[i][j] = taskid;
j++;
}
taskid++;
if (j >= step_layout->tasks[i])
break;
}
i++;
reset_hosts:
free(host);
if (i > step_layout->task_cnt)
break;
}
hostlist_iterator_destroy(itr);
hostlist_iterator_destroy(itr_task);
hostlist_destroy(job_alloc_hosts);
hostlist_destroy(step_alloc_hosts);
xfree(step_hosts_ptrs);
if (task_cnt != step_layout->task_cnt) {
error("Asked for %u tasks but placed %d. Check your nodelist",
step_layout->task_cnt, task_cnt);
return SLURM_ERROR;
}
return SLURM_SUCCESS;
}
/* use specific set run tasks on each host listed in hostfile
* XXX: Need to handle over-subscribe.
*/
static int _task_layout_hostfile(slurm_step_layout_t *step_layout,
const char *arbitrary_nodes)
{
int i=0, j, taskid = 0, task_cnt=0;
hostlist_iterator_t itr = NULL, itr_task = NULL;
char *host = NULL;
char *host_task = NULL;
hostlist_t job_alloc_hosts = NULL;
hostlist_t step_alloc_hosts = NULL;
debug2("job list is %s", step_layout->node_list);
job_alloc_hosts = hostlist_create(step_layout->node_list);
itr = hostlist_iterator_create(job_alloc_hosts);
if (!arbitrary_nodes) {
error("no hostlist given for arbitrary dist");
return SLURM_ERROR;
}
debug2("list is %s", arbitrary_nodes);
step_alloc_hosts = hostlist_create(arbitrary_nodes);
if (hostlist_count(step_alloc_hosts) != step_layout->task_cnt) {
error("Asked for %u tasks have %d in the nodelist. "
"Check your nodelist, or set the -n option to be %d",
step_layout->task_cnt,
hostlist_count(step_alloc_hosts),
hostlist_count(step_alloc_hosts));
return SLURM_ERROR;
}
itr_task = hostlist_iterator_create(step_alloc_hosts);
while((host = hostlist_next(itr))) {
step_layout->tasks[i] = 0;
while((host_task = hostlist_next(itr_task))) {
if (!strcmp(host, host_task)) {
step_layout->tasks[i]++;
task_cnt++;
}
free(host_task);
if (task_cnt >= step_layout->task_cnt)
break;
}
debug3("%s got %u tasks", host, step_layout->tasks[i]);
if (step_layout->tasks[i] == 0)
goto reset_hosts;
step_layout->tids[i] = xmalloc(sizeof(uint32_t)
* step_layout->tasks[i]);
taskid = 0;
j = 0;
hostlist_iterator_reset(itr_task);
while((host_task = hostlist_next(itr_task))) {
if (!strcmp(host, host_task)) {
step_layout->tids[i][j] = taskid;
j++;
}
taskid++;
free(host_task);
if (j >= step_layout->tasks[i])
break;
}
i++;
reset_hosts:
hostlist_iterator_reset(itr_task);
free(host);
if (i > step_layout->task_cnt)
break;
}
hostlist_iterator_destroy(itr);
hostlist_iterator_destroy(itr_task);
hostlist_destroy(job_alloc_hosts);
hostlist_destroy(step_alloc_hosts);
if (task_cnt != step_layout->task_cnt) {
error("Asked for %u tasks but placed %d. Check your nodelist",
step_layout->task_cnt, task_cnt);
return SLURM_ERROR;
}
return SLURM_SUCCESS;
}
void *_forward_thread(void *arg)
{
forward_msg_t *fwd_msg = (forward_msg_t *)arg;
forward_struct_t *fwd_struct = fwd_msg->fwd_struct;
Buf buffer = init_buf(BUF_SIZE); /* probably enough for header */
List ret_list = NULL;
int fd = -1;
ret_data_info_t *ret_data_info = NULL;
char *name = NULL;
hostlist_t hl = hostlist_create(fwd_msg->header.forward.nodelist);
slurm_addr_t addr;
char *buf = NULL;
int steps = 0;
int start_timeout = fwd_msg->timeout;
/* repeat until we are sure the message was sent */
while ((name = hostlist_shift(hl))) {
if (slurm_conf_get_addr(name, &addr) == SLURM_ERROR) {
error("forward_thread: can't find address for host "
"%s, check slurm.conf", name);
slurm_mutex_lock(&fwd_struct->forward_mutex);
mark_as_failed_forward(&fwd_struct->ret_list, name,
SLURM_UNKNOWN_FORWARD_ADDR);
free(name);
if (hostlist_count(hl) > 0) {
slurm_mutex_unlock(&fwd_struct->forward_mutex);
continue;
}
goto cleanup;
}
if ((fd = slurm_open_msg_conn(&addr)) < 0) {
error("forward_thread to %s: %m", name);
slurm_mutex_lock(&fwd_struct->forward_mutex);
mark_as_failed_forward(
&fwd_struct->ret_list, name,
SLURM_COMMUNICATIONS_CONNECTION_ERROR);
free(name);
if (hostlist_count(hl) > 0) {
slurm_mutex_unlock(&fwd_struct->forward_mutex);
/* Abandon tree. This way if all the
* nodes in the branch are down we
* don't have to time out for each
* node serially.
*/
_forward_msg_internal(hl, NULL, fwd_struct,
&fwd_msg->header, 0,
hostlist_count(hl));
continue;
}
goto cleanup;
}
buf = hostlist_ranged_string_xmalloc(hl);
xfree(fwd_msg->header.forward.nodelist);
fwd_msg->header.forward.nodelist = buf;
fwd_msg->header.forward.cnt = hostlist_count(hl);
#if 0
info("sending %d forwards (%s) to %s",
fwd_msg->header.forward.cnt,
fwd_msg->header.forward.nodelist, name);
#endif
if (fwd_msg->header.forward.nodelist[0]) {
debug3("forward: send to %s along with %s",
name, fwd_msg->header.forward.nodelist);
} else
debug3("forward: send to %s ", name);
pack_header(&fwd_msg->header, buffer);
/* add forward data to buffer */
if (remaining_buf(buffer) < fwd_struct->buf_len) {
int new_size = buffer->processed + fwd_struct->buf_len;
new_size += 1024; /* padded for paranoia */
xrealloc_nz(buffer->head, new_size);
buffer->size = new_size;
}
if (fwd_struct->buf_len) {
memcpy(&buffer->head[buffer->processed],
fwd_struct->buf, fwd_struct->buf_len);
buffer->processed += fwd_struct->buf_len;
}
/*
* forward message
*/
if (slurm_msg_sendto(fd,
get_buf_data(buffer),
get_buf_offset(buffer),
SLURM_PROTOCOL_NO_SEND_RECV_FLAGS ) < 0) {
error("forward_thread: slurm_msg_sendto: %m");
slurm_mutex_lock(&fwd_struct->forward_mutex);
mark_as_failed_forward(&fwd_struct->ret_list, name,
errno);
free(name);
if (hostlist_count(hl) > 0) {
free_buf(buffer);
buffer = init_buf(fwd_struct->buf_len);
slurm_mutex_unlock(&fwd_struct->forward_mutex);
slurm_close(fd);
fd = -1;
/* Abandon tree. This way if all the
* nodes in the branch are down we
* don't have to time out for each
* node serially.
*/
_forward_msg_internal(hl, NULL, fwd_struct,
&fwd_msg->header, 0,
hostlist_count(hl));
continue;
}
goto cleanup;
}
/* These messages don't have a return message, but if
* we got here things worked out so make note of the
* list of nodes as success.
*/
if ((fwd_msg->header.msg_type == REQUEST_SHUTDOWN) ||
(fwd_msg->header.msg_type == REQUEST_RECONFIGURE) ||
(fwd_msg->header.msg_type == REQUEST_REBOOT_NODES)) {
slurm_mutex_lock(&fwd_struct->forward_mutex);
ret_data_info = xmalloc(sizeof(ret_data_info_t));
list_push(fwd_struct->ret_list, ret_data_info);
ret_data_info->node_name = xstrdup(name);
free(name);
while ((name = hostlist_shift(hl))) {
ret_data_info =
xmalloc(sizeof(ret_data_info_t));
list_push(fwd_struct->ret_list, ret_data_info);
ret_data_info->node_name = xstrdup(name);
free(name);
}
goto cleanup;
}
if (fwd_msg->header.forward.cnt > 0) {
static int message_timeout = -1;
if (message_timeout < 0)
message_timeout =
slurm_get_msg_timeout() * 1000;
if (!fwd_msg->header.forward.tree_width)
fwd_msg->header.forward.tree_width =
slurm_get_tree_width();
steps = (fwd_msg->header.forward.cnt+1) /
fwd_msg->header.forward.tree_width;
fwd_msg->timeout = (message_timeout*steps);
/* info("got %d * %d = %d", message_timeout, */
/* steps, fwd_msg->timeout); */
steps++;
fwd_msg->timeout += (start_timeout*steps);
/* info("now + %d*%d = %d", start_timeout, */
/* steps, fwd_msg->timeout); */
}
ret_list = slurm_receive_msgs(fd, steps, fwd_msg->timeout);
/* info("sent %d forwards got %d back", */
/* fwd_msg->header.forward.cnt, list_count(ret_list)); */
if (!ret_list || (fwd_msg->header.forward.cnt != 0
&& list_count(ret_list) <= 1)) {
slurm_mutex_lock(&fwd_struct->forward_mutex);
mark_as_failed_forward(&fwd_struct->ret_list, name,
errno);
free(name);
FREE_NULL_LIST(ret_list);
if (hostlist_count(hl) > 0) {
free_buf(buffer);
buffer = init_buf(fwd_struct->buf_len);
slurm_mutex_unlock(&fwd_struct->forward_mutex);
slurm_close(fd);
fd = -1;
continue;
}
goto cleanup;
} else if ((fwd_msg->header.forward.cnt+1)
!= list_count(ret_list)) {
/* this should never be called since the above
should catch the failed forwards and pipe
them back down, but this is here so we
never have to worry about a locked
mutex */
ListIterator itr = NULL;
char *tmp = NULL;
int first_node_found = 0;
hostlist_iterator_t host_itr
= hostlist_iterator_create(hl);
error("We shouldn't be here. We forwarded to %d "
"but only got %d back",
(fwd_msg->header.forward.cnt+1),
list_count(ret_list));
while ((tmp = hostlist_next(host_itr))) {
int node_found = 0;
itr = list_iterator_create(ret_list);
while ((ret_data_info = list_next(itr))) {
if (!ret_data_info->node_name) {
first_node_found = 1;
ret_data_info->node_name =
xstrdup(name);
}
if (!xstrcmp(tmp,
ret_data_info->node_name)) {
node_found = 1;
break;
}
}
list_iterator_destroy(itr);
if (!node_found) {
mark_as_failed_forward(
&fwd_struct->ret_list,
tmp,
SLURM_COMMUNICATIONS_CONNECTION_ERROR);
}
free(tmp);
}
hostlist_iterator_destroy(host_itr);
if (!first_node_found) {
mark_as_failed_forward(
&fwd_struct->ret_list,
name,
SLURM_COMMUNICATIONS_CONNECTION_ERROR);
}
}
break;
}
slurm_mutex_lock(&fwd_struct->forward_mutex);
if (ret_list) {
while ((ret_data_info = list_pop(ret_list)) != NULL) {
if (!ret_data_info->node_name) {
ret_data_info->node_name = xstrdup(name);
}
list_push(fwd_struct->ret_list, ret_data_info);
debug3("got response from %s",
ret_data_info->node_name);
}
FREE_NULL_LIST(ret_list);
}
free(name);
cleanup:
if ((fd >= 0) && slurm_close(fd) < 0)
error ("close(%d): %m", fd);
hostlist_destroy(hl);
destroy_forward(&fwd_msg->header.forward);
free_buf(buffer);
slurm_cond_signal(&fwd_struct->notify);
slurm_mutex_unlock(&fwd_struct->forward_mutex);
xfree(fwd_msg);
return (NULL);
}
static int
eliminate_nodes (char **hosts)
{
hostlist_t hl = NULL;
hostlist_t hlnew = NULL;
hostlist_iterator_t hitr = NULL;
ipmidetect_t id = NULL;
char *host = NULL;
char hostbuf[HOSTLIST_BUFLEN + 1];
int rv = -1;
assert (hosts);
assert (*hosts);
if (!(id = ipmidetect_handle_create ()))
{
fprintf (stderr,
"ipmidetect_handle_create\n");
goto cleanup;
}
if (ipmidetect_load_data (id,
NULL,
0,
0) < 0)
{
if (ipmidetect_errnum (id) == IPMIDETECT_ERR_CONNECT
|| ipmidetect_errnum (id) == IPMIDETECT_ERR_CONNECT_TIMEOUT)
fprintf (stderr,
"Error connecting to ipmidetect daemon\n");
else
fprintf (stderr,
"ipmidetect_load_data: %s\n", ipmidetect_errormsg (id));
goto cleanup;
}
if (!(hl = hostlist_create (*hosts)))
{
fprintf (stderr,
"hostlist_create: %s\n",
strerror (errno));
goto cleanup;
}
if (!(hlnew = hostlist_create (*hosts)))
{
fprintf (stderr,
"hostlist_create: %s\n",
strerror (errno));
goto cleanup;
}
if (!(hitr = hostlist_iterator_create (hl)))
{
fprintf (stderr,
"hostlist_iterator_create: %s\n",
strerror (errno));
goto cleanup;
}
while ((host = hostlist_next (hitr)))
{
int ret;
if ((ret = ipmidetect_is_node_detected (id, host)) < 0)
{
if (ipmidetect_errnum (id) == IPMIDETECT_ERR_NOTFOUND)
fprintf (stderr,
"Node '%s' unrecognized by ipmidetect\n", host);
else
fprintf (stderr,
"ipmidetect_is_node_detected: %s\n", ipmidetect_errormsg (id));
goto cleanup;
}
if (!ret)
hostlist_delete (hlnew, host);
free (host);
}
host = NULL;
if (!hostlist_count (hlnew))
{
rv = 0;
goto cleanup;
}
memset (hostbuf, '\0', HOSTLIST_BUFLEN + 1);
if (hostlist_ranged_string (hlnew, HOSTLIST_BUFLEN, hostbuf) < 0)
{
fprintf (stderr,
"hostlist_ranged_string: truncation\n");
goto cleanup;
}
free (*hosts);
if (!(*hosts = strdup (hostbuf)))
{
fprintf (stderr, "strdup: %s\n", strerror (errno));
goto cleanup;
}
rv = hostlist_count (hlnew);
cleanup:
if (id)
ipmidetect_handle_destroy (id);
if (hitr)
hostlist_iterator_destroy (hitr);
if (hl)
hostlist_destroy (hl);
if (hlnew)
hostlist_destroy (hlnew);
free (host);
return (rv);
}