char *
opt_find (Opt o, char *key)
{
char *s;
NP_ASSERT (o->magic == OPT_MAGIC);
if (strchr (key, '='))
s = list_find_first (o->list, (ListFindF)_match_keyval, key);
else
s = list_find_first (o->list, (ListFindF)_match_key, key);
return s ? _optstr (s) : NULL;
}
int
cerebro_event_unregister(cerebro_t handle, int fd)
{
if (_cerebro_handle_check(handle) < 0)
return -1;
if (fd < 0)
{
handle->errnum = CEREBRO_ERR_PARAMETERS;
return -1;
}
if (!list_find_first(handle->event_fds, _event_fd_find, &fd))
{
handle->errnum = CEREBRO_ERR_PARAMETERS;
return -1;
}
if (!list_delete_all(handle->event_fds, _event_fd_find, &fd))
{
handle->errnum = CEREBRO_ERR_INTERNAL;
return -1;
}
/* ignore potential error, just return result */
close(fd);
handle->errnum = CEREBRO_ERR_SUCCESS;
return 0;
}
/*
* hostsfile_clusterlist_is_node_in_cluster
*
* hostsfile clusterlist module is_node_in_cluster function
*/
static int
hostsfile_clusterlist_is_node_in_cluster(nodeupdown_t handle, const char *node)
{
char nodebuf[NODEUPDOWN_MAXNODENAMELEN+1];
char *nodePtr = NULL;
void *ptr;
/* Have to assume it is */
if (!list_count(hosts))
return 1;
/* Shorten hostname if necessary */
if (strchr(node, '.'))
{
char *p;
memset(nodebuf, '\0', NODEUPDOWN_MAXNODENAMELEN+1);
strncpy(nodebuf, node, NODEUPDOWN_MAXNODENAMELEN);
p = strchr(nodebuf, '.');
*p = '\0';
nodePtr = nodebuf;
}
else
nodePtr = (char *)node;
ptr = list_find_first(hosts, _find_str, (void *)nodePtr);
if (ptr)
return 1;
else
return 0;
}
/*
* get_part_list - find record for named partition(s)
* IN name - partition name(s) in a comma separated list
* RET List of pointers to the partitions or NULL if not found
* NOTE: Caller must free the returned list
*/
extern List get_part_list(char *name)
{
struct part_record *part_ptr;
List job_part_list = NULL;
char *token, *last = NULL, *tmp_name;
if (name == NULL)
return job_part_list;
tmp_name = xstrdup(name);
token = strtok_r(tmp_name, ",", &last);
while (token) {
part_ptr = list_find_first(part_list, &list_find_part, token);
if (part_ptr) {
if (job_part_list == NULL) {
job_part_list = list_create(NULL);
}
list_append(job_part_list, part_ptr);
} else {
FREE_NULL_LIST(job_part_list);
break;
}
token = strtok_r(NULL, ",", &last);
}
xfree(tmp_name);
return job_part_list;
}
static void _remove_job_tres_time_from_cluster(List c_tres, List j_tres,
int seconds)
{
ListIterator c_itr;
local_tres_usage_t *loc_c_tres, *loc_j_tres;
uint64_t time;
if ((seconds <= 0) || !c_tres || !j_tres ||
!list_count(c_tres) || !list_count(j_tres))
return;
c_itr = list_iterator_create(c_tres);
while ((loc_c_tres = list_next(c_itr))) {
if (!(loc_j_tres = list_find_first(
j_tres, _find_loc_tres, &loc_c_tres->id)))
continue;
time = seconds * loc_j_tres->count;
if (time >= loc_c_tres->total_time)
loc_c_tres->total_time = 0;
else
loc_c_tres->total_time -= time;
}
list_iterator_destroy(c_itr);
}
/* Notify the gang scheduler that a job has been started */
extern int gs_job_start(struct job_record *job_ptr)
{
struct gs_part *p_ptr;
uint16_t job_state;
if (gs_debug_flags & DEBUG_FLAG_GANG)
info("gang: entering gs_job_start for job %u", job_ptr->job_id);
/* add job to partition */
pthread_mutex_lock(&data_mutex);
p_ptr = list_find_first(gs_part_list, _find_gs_part,
job_ptr->partition);
if (p_ptr) {
job_state = _add_job_to_part(p_ptr, job_ptr);
/* if this job is running then check for preemption */
if (job_state == GS_RESUME)
_update_all_active_rows();
}
pthread_mutex_unlock(&data_mutex);
if (!p_ptr) {
/* No partition was found for this job, so let it run
* uninterupted (what else can we do?)
*/
error("gang: could not find partition %s for job %u",
job_ptr->partition, job_ptr->job_id);
}
_preempt_job_dequeue(); /* MUST BE OUTSIDE OF data_mutex lock */
if (gs_debug_flags & DEBUG_FLAG_GANG)
info("gang: leaving gs_job_start");
return SLURM_SUCCESS;
}
/* Notify the gang scheduler that a job has completed */
extern int gs_job_fini(struct job_record *job_ptr)
{
struct gs_part *p_ptr;
char *part_name;
if (slurmctld_conf.debug_flags & DEBUG_FLAG_GANG)
info("gang: entering gs_job_fini for job %u", job_ptr->job_id);
if (job_ptr->part_ptr && job_ptr->part_ptr->name)
part_name = job_ptr->part_ptr->name;
else
part_name = job_ptr->partition;
pthread_mutex_lock(&data_mutex);
p_ptr = list_find_first(gs_part_list, _find_gs_part, part_name);
if (!p_ptr) {
pthread_mutex_unlock(&data_mutex);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_GANG)
info("gang: leaving gs_job_fini");
return SLURM_SUCCESS;
}
/* remove job from the partition */
_remove_job_from_part(job_ptr->job_id, p_ptr, true);
/* this job may have preempted other jobs, so
* check by updating all active rows */
_update_all_active_rows();
pthread_mutex_unlock(&data_mutex);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_GANG)
info("gang: leaving gs_job_fini");
return SLURM_SUCCESS;
}
/*
* license_job_get - Get the licenses required for a job
* IN job_ptr - job identification
* RET SLURM_SUCCESS or failure code
*/
extern int license_job_get(struct job_record *job_ptr)
{
ListIterator iter;
licenses_t *license_entry, *match;
int rc = SLURM_SUCCESS;
if (!job_ptr->license_list) /* no licenses needed */
return rc;
slurm_mutex_lock(&license_mutex);
iter = list_iterator_create(job_ptr->license_list);
if (iter == NULL)
fatal("malloc failure from list_iterator_create");
while ((license_entry = (licenses_t *) list_next(iter))) {
match = list_find_first(license_list, _license_find_rec,
license_entry->name);
if (match) {
match->used += license_entry->total;
license_entry->used += license_entry->total;
} else {
error("could not find license %s for job %u",
license_entry->name, job_ptr->job_id);
rc = SLURM_ERROR;
}
}
list_iterator_destroy(iter);
_licenses_print("acquire_license", license_list, job_ptr->job_id);
slurm_mutex_unlock(&license_mutex);
return rc;
}
/* Prepare cluster_list to be federation centric that will be passed to
* verify_clsuters_exists in federation_functions.c.
*/
static int _verify_fed_clusters(List cluster_list, const char *fed_name,
bool *existing_fed)
{
int rc = SLURM_SUCCESS;
char *tmp_name = NULL;
List tmp_list = list_create(slurmdb_destroy_cluster_rec);
ListIterator itr = list_iterator_create(cluster_list);
while ((tmp_name = list_next(itr))) {
slurmdb_cluster_rec_t *rec =
xmalloc(sizeof(slurmdb_cluster_rec_t));
slurmdb_init_cluster_rec(rec, 0);
rec->name = xstrdup(tmp_name);
list_append(tmp_list, rec);
}
if ((rc = verify_fed_clusters(tmp_list, fed_name, existing_fed)))
goto end_it;
/* have to reconcile lists now, clusters may have been removed from
* tmp_list */
list_iterator_reset(itr);
while ((tmp_name = list_next(itr))) {
if (!list_find_first(tmp_list, _find_cluster_rec_in_list,
tmp_name))
list_delete_item(itr);
}
end_it:
FREE_NULL_LIST(tmp_list);
list_iterator_destroy(itr);
return rc;
}
/*
* license_job_return - Return the licenses allocated to a job
* IN job_ptr - job identification
* RET SLURM_SUCCESS or failure code
*/
extern int license_job_return(struct job_record *job_ptr)
{
ListIterator iter;
licenses_t *license_entry, *match;
int rc = SLURM_SUCCESS;
if (!job_ptr->license_list) /* no licenses needed */
return rc;
slurm_mutex_lock(&license_mutex);
iter = list_iterator_create(job_ptr->license_list);
while ((license_entry = (licenses_t *) list_next(iter))) {
match = list_find_first(license_list, _license_find_rec,
license_entry->name);
if (match) {
if (match->used >= license_entry->total)
match->used -= license_entry->total;
else {
error("license use count underflow for %s",
match->name);
match->used = 0;
rc = SLURM_ERROR;
}
license_entry->used = 0;
} else {
/* This can happen after a reconfiguration */
error("job returning unknown license %s",
license_entry->name);
}
}
list_iterator_destroy(iter);
_licenses_print("return_license", license_list, job_ptr->job_id);
slurm_mutex_unlock(&license_mutex);
return rc;
}
void cproxy_disconnect(svc_client_handle_t *h)
{
struct _svc_cnx *cnx = (struct _svc_cnx*)
list_find_first(&_services, _cmp_handle, h);
if (cnx) {
cfw_close_service(h, cnx);
}
}
static void _build_tres_list(void)
{
ListIterator iter;
slurmdb_tres_rec_t *tres;
char *tres_tmp = NULL, *tres_tmp2 = NULL, *save_ptr = NULL, *tok;
if (!g_tres_list) {
slurmdb_tres_cond_t cond = {0};
g_tres_list = slurmdb_tres_get(db_conn, &cond);
if (!g_tres_list) {
fatal("Problem getting TRES data: %m");
exit(1);
}
}
FREE_NULL_LIST(tres_list);
tres_list = list_create(slurmdb_destroy_tres_rec);
if (!tres_str) {
int tres_cpu_id = TRES_CPU;
slurmdb_tres_rec_t *tres2;
if (!(tres = list_find_first(g_tres_list,
slurmdb_find_tres_in_list,
&tres_cpu_id)))
fatal("Failed to find CPU TRES!");
tres2 = slurmdb_copy_tres_rec(tres);
list_append(tres_list, tres2);
return;
}
tres_usage_str = "TRES";
iter = list_iterator_create(g_tres_list);
while ((tres = list_next(iter))) {
tres_tmp = xstrdup(tres_str);
xstrfmtcat(tres_tmp2, "%s%s%s",
tres->type,
tres->name ? "/" : "",
tres->name ? tres->name : "");
tok = strtok_r(tres_tmp, ",", &save_ptr);
while (tok) {
if (!xstrcasecmp(tres_tmp2, tok))
break;
tok = strtok_r(NULL, ",", &save_ptr);
}
if (tok) {
slurmdb_tres_rec_t *tres2 =
slurmdb_copy_tres_rec(tres);
list_append(tres_list, tres2);
}
xfree(tres_tmp2);
xfree(tres_tmp);
}
if (!list_count(tres_list))
fatal("No valid TRES given");
list_iterator_destroy(iter);
}
extern int fed_mgr_update_feds(slurmdb_update_object_t *update)
{
List feds;
slurmdb_federation_rec_t *fed = NULL;
slurmdb_cluster_rec_t *cluster = NULL;
slurmctld_lock_t fed_write_lock = {
NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, WRITE_LOCK };
if (!update->objects)
return SLURM_SUCCESS;
slurm_mutex_lock(&init_mutex);
if (!inited) {
slurm_mutex_unlock(&init_mutex);
return SLURM_SUCCESS; /* we haven't started the fed mgr and we
* can't start it from here, don't worry
* all will get set up later. */
}
slurm_mutex_unlock(&init_mutex);
/* we only want one update happening at a time. */
slurm_mutex_lock(&update_mutex);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("Got a federation update");
feds = update->objects;
/* find the federation that this cluster is in.
* if it's changed from last time then update stored information.
* grab other clusters in federation
* establish connections with each cluster in federation */
/* what if a remote cluster is removed from federation.
* have to detect that and close the connection to the remote */
while ((fed = list_pop(feds))) {
if (fed->cluster_list &&
(cluster = list_find_first(fed->cluster_list,
slurmdb_find_cluster_in_list,
slurmctld_cluster_name))) {
_join_federation(fed, cluster, true);
break;
}
slurmdb_destroy_federation_rec(fed);
}
if (!fed) {
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("Not part of any federation");
lock_slurmctld(fed_write_lock);
_leave_federation();
unlock_slurmctld(fed_write_lock);
}
slurm_mutex_unlock(&update_mutex);
return SLURM_SUCCESS;
}
/* Given a license string, return a list of license_t records */
static List _build_license_list(char *licenses, bool *valid)
{
int i;
char *end_num, *tmp_str, *token, *last;
licenses_t *license_entry;
List lic_list;
*valid = true;
if ((licenses == NULL) || (licenses[0] == '\0'))
return NULL;
lic_list = list_create(license_free_rec);
tmp_str = xstrdup(licenses);
token = strtok_r(tmp_str, ",;", &last);
while (token && *valid) {
uint32_t num = 1;
for (i = 0; token[i]; i++) {
if (isspace(token[i])) {
*valid = false;
break;
}
/* ':' is used as a separator in version 2.5 or later
* '*' is used as a separator in version 2.4 or earlier
*/
if ((token[i] == ':') || (token[i] == '*')) {
token[i++] = '\0';
num = (uint32_t)strtol(&token[i], &end_num,10);
}
}
if (num <= 0) {
*valid = false;
break;
}
license_entry = list_find_first(lic_list, _license_find_rec,
token);
if (license_entry) {
license_entry->total += num;
} else {
license_entry = xmalloc(sizeof(licenses_t));
license_entry->name = xstrdup(token);
license_entry->total = num;
list_push(lic_list, license_entry);
}
token = strtok_r(NULL, ",;", &last);
}
xfree(tmp_str);
if (*valid == false) {
list_destroy(lic_list);
lic_list = NULL;
}
return lic_list;
}
int optparse_parse_args (optparse_t p, int argc, char *argv[])
{
int c;
int li;
char *saved_argv0;
char *optstring = NULL;
struct option *optz = option_table_create (p, &optstring);
/* Always set optind = 0 here to force internal initialization of
* GNU options parser. See getopt_long(3) NOTES section.
*/
optind = 0;
saved_argv0 = argv[0];
argv[0] = p->program_name;
while ((c = getopt_long (argc, argv, optstring, optz, &li))) {
struct option_info *opt;
struct optparse_option *o;
if (c == -1)
break;
if (c == '?') {
fprintf (stderr, "Unknown option. Try --help\n");
optind = -1;
break;
}
opt = list_find_first (p->option_list, (ListFindF) by_val, &c);
if (opt == NULL) {
fprintf (stderr, "ugh, didn't find option associated with char %c\n", c);
continue;
}
opt->found++;
if (optarg) {
if (opt->optarg)
free (opt->optarg);
opt->optarg = strdup (optarg);
}
o = opt->p_opt;
if (o->cb && ((o->cb) (o, optarg, o->arg) < 0)) {
fprintf (stderr, "Option \"%s\" failed\n", o->name);
optind = -1;
break;
}
}
free (optz);
free (optstring);
argv[0] = saved_argv0;
return (optind);
}
static void _add_job_alloc_time_to_cluster(List c_tres_list, List j_tres)
{
ListIterator c_itr = list_iterator_create(c_tres_list);
local_tres_usage_t *loc_c_tres, *loc_j_tres;
while ((loc_c_tres = list_next(c_itr))) {
if (!(loc_j_tres = list_find_first(
j_tres, _find_loc_tres, &loc_c_tres->id)))
continue;
loc_c_tres->time_alloc += loc_j_tres->time_alloc;
}
list_iterator_destroy(c_itr);
}
static struct conn_helper_data *get_conn_helper(
_cfw_client_t *c,
struct conn_helper_data *
helper)
{
struct conn_helper_data *ret = NULL;
list_head_t *helper_list = &c->helper_list;
int flags = irq_lock();
ret = (struct conn_helper_data *)
list_find_first(helper_list, conn_helper_data_check, helper);
irq_unlock(flags);
return ret;
}
/*
* Must have FED unlocked prior to entering
*/
static void _fed_mgr_ptr_init(slurmdb_federation_rec_t *db_fed,
slurmdb_cluster_rec_t *cluster)
{
ListIterator c_itr;
slurmdb_cluster_rec_t *tmp_cluster, *db_cluster;
slurmctld_lock_t fed_write_lock = {
NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, WRITE_LOCK };
xassert(cluster);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("Joining federation %s", db_fed->name);
lock_slurmctld(fed_write_lock);
if (fed_mgr_fed_rec) {
/* we are already part of a federation, preserve existing
* conenctions */
c_itr = list_iterator_create(db_fed->cluster_list);
while ((db_cluster = list_next(c_itr))) {
if (!xstrcmp(db_cluster->name,
slurmctld_cluster_name)) {
fed_mgr_cluster_rec = db_cluster;
continue;
}
if (!(tmp_cluster =
list_find_first(fed_mgr_fed_rec->cluster_list,
slurmdb_find_cluster_in_list,
db_cluster->name))) {
/* don't worry about destroying the connection
* here. It will happen below when we free
* fed_mgr_fed_rec (automagically).
*/
continue;
}
slurm_mutex_lock(&tmp_cluster->lock);
/* transfer over the connections we already have */
db_cluster->fed.send = tmp_cluster->fed.send;
tmp_cluster->fed.send = NULL;
db_cluster->fed.recv = tmp_cluster->fed.recv;
tmp_cluster->fed.recv = NULL;
slurm_mutex_unlock(&tmp_cluster->lock);
}
list_iterator_destroy(c_itr);
slurmdb_destroy_federation_rec(fed_mgr_fed_rec);
} else
fed_mgr_cluster_rec = cluster;
fed_mgr_fed_rec = db_fed;
unlock_slurmctld(fed_write_lock);
}
char *
diod_get_exports (char *name, void *a)
{
List exports = diod_conf_get_exports ();
List seen = NULL;
ListIterator itr = NULL;
Export *x;
int len = 0;
char *s = NULL;
char *ret = NULL;
NP_ASSERT (exports != NULL);
if (!(seen = list_create (NULL))) {
np_uerror (ENOMEM);
goto done;
}
if (!(itr = list_iterator_create (exports))) {
np_uerror (ENOMEM);
goto done;
}
while ((x = list_next (itr))) {
if (list_find_first (seen, (ListFindF)_strmatch, x->path))
continue;
if (!list_append (seen, x->path)) {
np_uerror (ENOMEM);
goto done;
}
if (!(x->oflags & XFLAGS_SUPPRESS)) {
if (aspf (&s, &len, "%s %s %s %s\n",
x->path,
x->opts ? x->opts : "-",
x->users ? x->users : "-",
x->hosts ? x->hosts : "-") < 0) {
np_uerror (ENOMEM);
goto done;
}
}
}
if (diod_conf_get_exportall ())
if (!_get_mounts (&s, &len, seen))
goto done;
ret = s;
done:
if (itr)
list_iterator_destroy (itr);
if (seen)
list_destroy (seen);
return ret;
}
static void _setup_cluster_tres(List tres_list, uint32_t id,
uint64_t count, int seconds)
{
local_tres_usage_t *loc_tres =
list_find_first(tres_list, _find_loc_tres, &id);
if (!loc_tres) {
loc_tres = xmalloc(sizeof(local_tres_usage_t));
loc_tres->id = id;
list_append(tres_list, loc_tres);
}
loc_tres->count = count;
loc_tres->total_time += seconds * loc_tres->count;
}
/* Query the quota for uid and add it to qlist if successful.
*/
static void
add_quota(confent_t *cp, List qlist, uid_t uid, char *name)
{
quota_t q;
if (list_find_first(qlist, (ListFindF)quota_match_uid, &uid))
return;
q = quota_create(cp->cf_label, cp->cf_rhost, cp->cf_rpath, cp->cf_thresh);
if (quota_get(uid, q)) {
quota_destroy(q);
return;
}
if (name)
quota_adduser (q, name);
list_append(qlist, q);
}
extern char *fed_mgr_find_sibling_name_by_ip(char *ip)
{
char *name = NULL;
slurmdb_cluster_rec_t *sibling = NULL;
slurmctld_lock_t fed_read_lock = {
NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, READ_LOCK };
lock_slurmctld(fed_read_lock);
if (fed_mgr_fed_rec && fed_mgr_fed_rec->cluster_list &&
(sibling = list_find_first(fed_mgr_fed_rec->cluster_list,
_find_sibling_by_ip, ip)))
name = xstrdup(sibling->name);
unlock_slurmctld(fed_read_lock);
return name;
}
int usb_pm_unregister_callback(struct device *dev, void (*cb)(bool, void*))
{
struct usb_pm_info *usb_dev = (struct usb_pm_info*)dev->priv;
list_t *element = list_find_first(&usb_dev->cb_head, check_item_callback, (void*)cb);
if (element == NULL) {
// element not found
return -1;
}
// Remove list element
list_remove(&usb_dev->cb_head, element);
bfree(element);
return 0;
}
static void _handle_client_message(struct cfw_message *msg, void *data)
{
(void)data;
switch (CFW_MESSAGE_ID(msg)) {
case MSG_ID_CFW_OPEN_SERVICE:
{
/* We have passed the allocated cnx as an opaque data */
struct _svc_cnx *cnx = CFW_MESSAGE_PRIV(msg);
/* Get the service parameters from the message and store them locally */
cfw_open_conn_rsp_msg_t *con_msg = (cfw_open_conn_rsp_msg_t *)msg;
cnx->sh = (svc_client_handle_t *)(con_msg->client_handle);
cnx->src_port = con_msg->port;
cfw_msg_free(msg);
list_add(&_services, (list_t *)cnx);
break;
}
case MSG_ID_CFW_CLOSE_SERVICE:
{
struct _svc_cnx *cnx = CFW_MESSAGE_PRIV(msg);
list_remove(&_services, (list_t *)cnx);
bfree(cnx);
cfw_msg_free(msg);
break;
}
case MSG_ID_CFW_SVC_AVAIL_EVT:
{
struct _svc_cnx *cnx = CFW_MESSAGE_PRIV(msg);
if (((cfw_svc_available_evt_msg_t*)msg)->service_id == cnx->service_id) {
cfw_open_service(_proxy_handle, cnx->service_id, cnx);
}
cfw_msg_free(msg);
break;
}
default:
{
/* Find the service connection based on the message source port */
struct _svc_cnx *cnx = (struct _svc_cnx*)
list_find_first(&_services, _cmp_port, &(CFW_MESSAGE_SRC(msg)));
if (cnx) {
cnx->cb(msg, cnx->data);
} else {
cfw_msg_free(msg);
}
break;
}
}
}
static void _persist_callback_fini(void *arg)
{
slurm_persist_conn_t *persist_conn = arg;
slurmdb_cluster_rec_t *cluster;
slurmctld_lock_t fed_write_lock = {
NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, WRITE_LOCK };
/* If we are shutting down just return or you will get deadlock since
* all these locks are already locked.
*/
if (!persist_conn || *persist_conn->shutdown)
return;
lock_slurmctld(fed_write_lock);
/* shuting down */
if (!fed_mgr_fed_rec) {
unlock_slurmctld(fed_write_lock);
return;
}
if (!(cluster = list_find_first(fed_mgr_fed_rec->cluster_list,
slurmdb_find_cluster_in_list,
persist_conn->cluster_name))) {
info("Couldn't find cluster %s?",
persist_conn->cluster_name);
unlock_slurmctld(fed_write_lock);
return;
}
slurm_mutex_lock(&cluster->lock);
/* This will get handled at the end of the thread, don't free it here */
cluster->fed.recv = NULL;
// persist_conn = cluster->fed.recv;
// slurm_persist_conn_close(persist_conn);
persist_conn = cluster->fed.send;
if (persist_conn) {
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("Closing send to sibling cluster %s",
cluster->name);
slurm_persist_conn_close(persist_conn);
}
slurm_mutex_unlock(&cluster->lock);
unlock_slurmctld(fed_write_lock);
}
/* distribute unused reservation usage to associations that
could have run jobs */
static int
_process_resv_idle_time(List resv_usage_list, List assoc_usage_list)
{
ListIterator r_itr;
local_resv_usage_t *r_usage;
local_id_usage_t *a_usage = NULL;
int seconds;
int last_id = -1;
r_itr = list_iterator_create(resv_usage_list);
while((r_usage = list_next(r_itr))) {
char *assoc = NULL;
ListIterator tmp_itr = NULL;
int64_t idle = r_usage->total_time - r_usage->a_cpu;
if (idle <= 0)
continue;
/* now divide that time by the number of
associations in the reservation and add
them to each association */
seconds = idle / list_count(r_usage->local_assocs);
/* info("resv %d got %d for seconds for %d assocs", */
/* r_usage->id, seconds, */
/* list_count(r_usage->local_assocs)); */
tmp_itr = list_iterator_create(r_usage->local_assocs);
while((assoc = list_next(tmp_itr))) {
int associd = atoi(assoc);
if (last_id != associd) {
a_usage = list_find_first(assoc_usage_list,
_cmp_local_id,
&associd);
}
if (!a_usage) {
a_usage = xmalloc(sizeof(local_id_usage_t));
a_usage->id = associd;
list_append(assoc_usage_list, a_usage);
last_id = associd;
}
a_usage->a_cpu += seconds;
}
list_iterator_destroy(tmp_itr);
}
list_iterator_destroy(r_itr);
return SLURM_SUCCESS;
}
/*
* license_job_test - Test if the licenses required for a job are available
* IN job_ptr - job identification
* IN when - time to check
* RET: SLURM_SUCCESS, EAGAIN (not available now), SLURM_ERROR (never runnable)
*/
extern int license_job_test(struct job_record *job_ptr, time_t when)
{
ListIterator iter;
licenses_t *license_entry, *match;
int rc = SLURM_SUCCESS, resv_licenses;
if (!job_ptr->license_list) /* no licenses needed */
return rc;
slurm_mutex_lock(&license_mutex);
iter = list_iterator_create(job_ptr->license_list);
if (iter == NULL)
fatal("malloc failure from list_iterator_create");
while ((license_entry = (licenses_t *) list_next(iter))) {
match = list_find_first(license_list, _license_find_rec,
license_entry->name);
if (!match) {
error("could not find license %s for job %u",
license_entry->name, job_ptr->job_id);
rc = SLURM_ERROR;
break;
} else if (license_entry->total > match->total) {
info("job %u wants more %s licenses than configured",
job_ptr->job_id, match->name);
rc = SLURM_ERROR;
break;
} else if ((license_entry->total + match->used) >
match->total) {
rc = EAGAIN;
break;
} else {
resv_licenses = job_test_lic_resv(job_ptr,
license_entry->name,
when);
if ((license_entry->total + match->used +
resv_licenses) > match->total) {
rc = EAGAIN;
break;
}
}
}
list_iterator_destroy(iter);
slurm_mutex_unlock(&license_mutex);
return rc;
}
static local_tres_usage_t *_add_time_tres(List tres_list, int type, uint32_t id,
uint64_t time, bool times_count)
{
local_tres_usage_t *loc_tres;
if (!time)
return NULL;
loc_tres = list_find_first(tres_list, _find_loc_tres, &id);
if (!loc_tres) {
if (times_count)
return NULL;
loc_tres = xmalloc(sizeof(local_tres_usage_t));
loc_tres->id = id;
list_append(tres_list, loc_tres);
}
if (times_count) {
if (!loc_tres->count)
return NULL;
time *= loc_tres->count;
}
switch (type) {
case TIME_ALLOC:
loc_tres->time_alloc += time;
break;
case TIME_DOWN:
loc_tres->time_down += time;
break;
case TIME_PDOWN:
loc_tres->time_pd += time;
break;
case TIME_RESV:
loc_tres->time_resv += time;
break;
default:
error("_add_time_tres: unknown type %d given", type);
xassert(0);
break;
}
return loc_tres;
}
/* Update licenses on this system based upon slurm.conf.
* Preserve all previously allocated licenses */
extern int license_update(char *licenses)
{
ListIterator iter;
licenses_t *license_entry, *match;
List new_list;
bool valid;
new_list = _build_license_list(licenses, &valid);
if (!valid)
fatal("Invalid configured licenses: %s", licenses);
slurm_mutex_lock(&license_mutex);
if (!license_list) { /* no licenses before now */
license_list = new_list;
slurm_mutex_unlock(&license_mutex);
return SLURM_SUCCESS;
}
iter = list_iterator_create(license_list);
if (iter == NULL)
fatal("malloc failure from list_iterator_create");
while ((license_entry = (licenses_t *) list_next(iter))) {
match = list_find_first(new_list, _license_find_rec,
license_entry->name);
if (!match) {
info("license %s removed with %u in use",
license_entry->name, license_entry->used);
} else {
match->used = license_entry->used;
if (match->used > match->total) {
info("license %s count decreased",
match->name);
}
}
}
list_iterator_destroy(iter);
list_destroy(license_list);
license_list = new_list;
_licenses_print("update_license", license_list, 0);
slurm_mutex_unlock(&license_mutex);
return SLURM_SUCCESS;
}
/*
* license_validate - Test if the required licenses are valid
* IN licenses - required licenses
* OUT valid - true if required licenses are valid and a sufficient number
* are configured (though not necessarily available now)
* RET license_list, must be destroyed by caller
*/
extern List license_validate(char *licenses, bool *valid)
{
ListIterator iter;
licenses_t *license_entry, *match;
List job_license_list;
job_license_list = _build_license_list(licenses, valid);
if (!job_license_list)
return job_license_list;
slurm_mutex_lock(&license_mutex);
_licenses_print("request_license", job_license_list, 0);
iter = list_iterator_create(job_license_list);
if (iter == NULL)
fatal("malloc failure from list_iterator_create");
while ((license_entry = (licenses_t *) list_next(iter))) {
if (license_list) {
match = list_find_first(license_list,
_license_find_rec, license_entry->name);
} else
match = NULL;
if (!match) {
debug("could not find license %s for job",
license_entry->name);
*valid = false;
break;
} else if (license_entry->total > match->total) {
debug("job wants more %s licenses than configured",
match->name);
*valid = false;
break;
}
}
list_iterator_destroy(iter);
slurm_mutex_unlock(&license_mutex);
if (!(*valid)) {
list_destroy(job_license_list);
job_license_list = NULL;
}
return job_license_list;
}
