/*
* Given a buffer containing a network byte order 16-bit integer,
* and an arbitrary data string, copy the data string into the location
* specified by valp. Also return the sizes of 'valp' in bytes.
* Adjust buffer counters.
* NOTE: valp is set to point into a newly created buffer,
* the caller is responsible for calling free() on *valp
* if non-NULL (set to NULL on zero size buffer value)
*/
int unpackmem_malloc(char **valp, uint32_t * size_valp, Buf buffer)
{
uint32_t ns;
if (remaining_buf(buffer) < sizeof(ns))
return SLURM_ERROR;
memcpy(&ns, &buffer->head[buffer->processed], sizeof(ns));
*size_valp = ntohl(ns);
buffer->processed += sizeof(ns);
if (*size_valp > MAX_PACK_MEM_LEN) {
error("%s: Buffer to be unpacked is too large (%u > %u)",
__func__, *size_valp, MAX_PACK_MEM_LEN);
return SLURM_ERROR;
}
else if (*size_valp > 0) {
if (remaining_buf(buffer) < *size_valp)
return SLURM_ERROR;
*valp = malloc(*size_valp);
if (*valp == NULL) {
log_oom(__FILE__, __LINE__, __func__);
abort();
}
memcpy(*valp, &buffer->head[buffer->processed],
*size_valp);
buffer->processed += *size_valp;
} else
*valp = NULL;
return SLURM_SUCCESS;
}
/*
* Given a buffer containing a network byte order 16-bit integer,
* and an arbitrary data string, copy the data string into the location
* specified by valp. Also return the sizes of 'valp' in bytes.
* Adjust buffer counters.
* NOTE: The caller is responsible for the management of valp and
* insuring it has sufficient size
*/
int unpackmem(char *valp, uint32_t * size_valp, Buf buffer)
{
uint32_t ns;
if (remaining_buf(buffer) < sizeof(ns))
return SLURM_ERROR;
memcpy(&ns, &buffer->head[buffer->processed], sizeof(ns));
*size_valp = ntohl(ns);
buffer->processed += sizeof(ns);
if (*size_valp > MAX_PACK_MEM_LEN) {
error("%s: Buffer to be unpacked is too large (%u > %u)",
__func__, *size_valp, MAX_PACK_MEM_LEN);
return SLURM_ERROR;
}
else if (*size_valp > 0) {
if (remaining_buf(buffer) < *size_valp)
return SLURM_ERROR;
memcpy(valp, &buffer->head[buffer->processed], *size_valp);
buffer->processed += *size_valp;
} else
*valp = 0;
return SLURM_SUCCESS;
}
/*
* Given a pointer to memory (valp) and a size (size_val), convert
* size_val to network byte order and store at buffer followed by
* the data at valp. Adjust buffer counters.
*/
void packmem(char *valp, uint32_t size_val, Buf buffer)
{
uint32_t ns = htonl(size_val);
if (size_val > MAX_PACK_MEM_LEN) {
error("%s: Buffer to be packed is too large (%u > %u)",
__func__, size_val, MAX_PACK_MEM_LEN);
return;
}
if (remaining_buf(buffer) < (sizeof(ns) + size_val)) {
if ((buffer->size + size_val + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + size_val + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += (size_val + BUF_SIZE);
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
if (size_val) {
memcpy(&buffer->head[buffer->processed], valp, size_val);
buffer->processed += size_val;
}
}
/*
* Given 'buffer' pointing to a network byte order 16-bit integer
* (size) and a array of strings store the number of strings in
* 'size_valp' and the array of strings in valp
* NOTE: valp is set to point into a newly created buffer,
* the caller is responsible for calling xfree on *valp
* if non-NULL (set to NULL on zero size buffer value)
*/
int unpackstr_array(char ***valp, uint32_t * size_valp, Buf buffer)
{
int i;
uint32_t ns;
uint32_t uint32_tmp;
if (remaining_buf(buffer) < sizeof(ns))
return SLURM_ERROR;
memcpy(&ns, &buffer->head[buffer->processed], sizeof(ns));
*size_valp = ntohl(ns);
buffer->processed += sizeof(ns);
if (*size_valp > MAX_PACK_ARRAY_LEN) {
error("%s: Buffer to be unpacked is too large (%u > %u)",
__func__, *size_valp, MAX_PACK_ARRAY_LEN);
return SLURM_ERROR;
}
else if (*size_valp > 0) {
*valp = xmalloc_nz(sizeof(char *) * (*size_valp + 1));
for (i = 0; i < *size_valp; i++) {
if (unpackmem_xmalloc(&(*valp)[i], &uint32_tmp, buffer))
return SLURM_ERROR;
}
(*valp)[i] = NULL; /* NULL terminated array so that execle */
/* can detect end of array */
} else
*valp = NULL;
return SLURM_SUCCESS;
}
/*
* Given a double, multiple by FLOAT_MULT and then
* typecast to a uint64_t in host byte order, convert to network byte order
* store in buffer, and adjust buffer counters.
* NOTE: There is an IEEE standard format for double.
*/
void packdouble(double val, Buf buffer)
{
uint64_t nl;
union {
double d;
uint64_t u;
} uval;
/* The 0.5 is here to round off. We have found on systems going out
* more than 15 decimals will mess things up, but this corrects it. */
uval.d = (val * FLOAT_MULT);
nl = HTON_uint64(uval.u);
if (remaining_buf(buffer) < sizeof(nl)) {
if ((buffer->size + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += BUF_SIZE;
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &nl, sizeof(nl));
buffer->processed += sizeof(nl);
}
/*
* Given a buffer containing a network byte order 8-bit integer,
* store a host integer at 'valp', and adjust buffer counters.
*/
int unpack8(uint8_t * valp, Buf buffer)
{
if (remaining_buf(buffer) < sizeof(uint8_t))
return SLURM_ERROR;
memcpy(valp, &buffer->head[buffer->processed], sizeof(uint8_t));
buffer->processed += sizeof(uint8_t);
return SLURM_SUCCESS;
}
/*
* Given a buffer containing a network byte order 32-bit integer,
* store a host integer at 'valp', and adjust buffer counters.
*/
int unpack32(uint32_t * valp, Buf buffer)
{
uint32_t nl;
if (remaining_buf(buffer) < sizeof(nl))
return SLURM_ERROR;
memcpy(&nl, &buffer->head[buffer->processed], sizeof(nl));
*valp = ntohl(nl);
buffer->processed += sizeof(nl);
return SLURM_SUCCESS;
}
/*
* Given a pointer to memory (valp), size (size_val), and buffer,
* store the buffer contents into memory
*/
int unpackmem_array(char *valp, uint32_t size_valp, Buf buffer)
{
if (remaining_buf(buffer) >= size_valp) {
memcpy(valp, &buffer->head[buffer->processed], size_valp);
buffer->processed += size_valp;
return SLURM_SUCCESS;
} else {
*valp = 0;
return SLURM_ERROR;
}
}
/*
* Given a buffer containing a network byte order 16-bit integer,
* store a host integer at 'valp', and adjust buffer counters.
*/
int unpack16(uint16_t * valp, Buf buffer)
{
uint16_t ns;
if (remaining_buf(buffer) < sizeof(ns))
return SLURM_ERROR;
memcpy(&ns, &buffer->head[buffer->processed], sizeof(ns));
*valp = ntohs(ns);
buffer->processed += sizeof(ns);
return SLURM_SUCCESS;
}
int unpack_time(time_t * valp, Buf buffer)
{
int64_t n64;
if (remaining_buf(buffer) < sizeof(n64))
return SLURM_ERROR;
memcpy(&n64, &buffer->head[buffer->processed], sizeof(n64));
buffer->processed += sizeof(n64);
*valp = (time_t) NTOH_int64(n64);
return SLURM_SUCCESS;
}
/*
* Given a buffer containing a network byte order 16-bit integer,
* and an arbitrary data string, copy the data string into the location
* specified by valp. Also return the sizes of 'valp' in bytes.
* Adjust buffer counters.
* NOTE: The caller is responsible for the management of valp and
* insuring it has sufficient size
*/
int unpackmem(char *valp, uint32_t * size_valp, Buf buffer)
{
uint32_t ns;
if (remaining_buf(buffer) < sizeof(ns))
return SLURM_ERROR;
memcpy(&ns, &buffer->head[buffer->processed], sizeof(ns));
*size_valp = ntohl(ns);
buffer->processed += sizeof(ns);
if (*size_valp > MAX_PACK_MEM_LEN)
return SLURM_ERROR;
else if (*size_valp > 0) {
if (remaining_buf(buffer) < *size_valp)
return SLURM_ERROR;
memcpy(valp, &buffer->head[buffer->processed], *size_valp);
buffer->processed += *size_valp;
} else
*valp = 0;
return SLURM_SUCCESS;
}
/*
* Given a 8-bit integer in host byte order, convert to network byte order
* store in buffer, and adjust buffer counters.
*/
void pack8(uint8_t val, Buf buffer)
{
if (remaining_buf(buffer) < sizeof(uint8_t)) {
if (buffer->size > (MAX_BUF_SIZE - BUF_SIZE)) {
error("pack8: buffer size too large");
return;
}
buffer->size += BUF_SIZE;
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &val, sizeof(uint8_t));
buffer->processed += sizeof(uint8_t);
}
/*
* Given a pointer to memory (valp), size (size_val), and buffer,
* store the memory contents into the buffer
*/
void packmem_array(char *valp, uint32_t size_val, Buf buffer)
{
if (remaining_buf(buffer) < size_val) {
if (buffer->size > (MAX_BUF_SIZE - size_val - BUF_SIZE)) {
error("packmem_array: buffer size too large");
return;
}
buffer->size += (size_val + BUF_SIZE);
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], valp, size_val);
buffer->processed += size_val;
}
/*
* Given a 8-bit integer in host byte order, convert to network byte order
* store in buffer, and adjust buffer counters.
*/
void pack8(uint8_t val, Buf buffer)
{
if (remaining_buf(buffer) < sizeof(uint8_t)) {
if ((buffer->size + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += BUF_SIZE;
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &val, sizeof(uint8_t));
buffer->processed += sizeof(uint8_t);
}
/*
* Given a time_t in host byte order, promote it to int64_t, convert to
* network byte order, store in buffer and adjust buffer acc'd'ngly
*/
void pack_time(time_t val, Buf buffer)
{
int64_t n64 = HTON_int64((int64_t) val);
if (remaining_buf(buffer) < sizeof(n64)) {
if (buffer->size > (MAX_BUF_SIZE - BUF_SIZE)) {
error("pack_time: buffer size too large");
return;
}
buffer->size += BUF_SIZE;
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &n64, sizeof(n64));
buffer->processed += sizeof(n64);
}
/*
* Given a 32-bit integer in host byte order, convert to network byte order
* store in buffer, and adjust buffer counters.
*/
void pack32(uint32_t val, Buf buffer)
{
uint32_t nl = htonl(val);
if (remaining_buf(buffer) < sizeof(nl)) {
if (buffer->size > (MAX_BUF_SIZE - BUF_SIZE)) {
error("pack32: buffer size too large");
return;
}
buffer->size += BUF_SIZE;
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &nl, sizeof(nl));
buffer->processed += sizeof(nl);
}
/*
* Given a 16-bit integer in host byte order, convert to network byte order,
* store in buffer and adjust buffer counters.
*/
void pack16(uint16_t val, Buf buffer)
{
uint16_t ns = htons(val);
if (remaining_buf(buffer) < sizeof(ns)) {
if (buffer->size > (MAX_BUF_SIZE - BUF_SIZE)) {
error("pack16: buffer size too large");
return;
}
buffer->size += BUF_SIZE;
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
}
/*
* Given a pointer to memory (valp), size (size_val), and buffer,
* store the memory contents into the buffer
*/
void packmem_array(char *valp, uint32_t size_val, Buf buffer)
{
if (remaining_buf(buffer) < size_val) {
if ((buffer->size + size_val + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + size_val + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += (size_val + BUF_SIZE);
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], valp, size_val);
buffer->processed += size_val;
}
/*
* Given a 16-bit integer in host byte order, convert to network byte order,
* store in buffer and adjust buffer counters.
*/
void pack16(uint16_t val, Buf buffer)
{
uint16_t ns = htons(val);
if (remaining_buf(buffer) < sizeof(ns)) {
if ((buffer->size + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%d > %d)",
__func__, (buffer->size + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += BUF_SIZE;
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
}
/*
* Given a time_t in host byte order, promote it to int64_t, convert to
* network byte order, store in buffer and adjust buffer acc'd'ngly
*/
void pack_time(time_t val, Buf buffer)
{
int64_t n64 = HTON_int64((int64_t) val);
if (remaining_buf(buffer) < sizeof(n64)) {
if ((buffer->size + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += BUF_SIZE;
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &n64, sizeof(n64));
buffer->processed += sizeof(n64);
}
/*
* Given a buffer containing a network byte order 64-bit integer,
* typecast as double, and divide by FLOAT_MULT
* store a host double at 'valp', and adjust buffer counters.
* NOTE: There is an IEEE standard format for double.
*/
int unpackdouble(double *valp, Buf buffer)
{
uint64_t nl;
union {
double d;
uint64_t u;
} uval;
if (remaining_buf(buffer) < sizeof(nl))
return SLURM_ERROR;
memcpy(&nl, &buffer->head[buffer->processed], sizeof(nl));
buffer->processed += sizeof(nl);
uval.u = NTOH_uint64(nl);
*valp = uval.d / FLOAT_MULT;
return SLURM_SUCCESS;
}
/*
* Given a pointer to memory (valp) and a size (size_val), convert
* size_val to network byte order and store at buffer followed by
* the data at valp. Adjust buffer counters.
*/
void packmem(char *valp, uint32_t size_val, Buf buffer)
{
uint32_t ns = htonl(size_val);
if (remaining_buf(buffer) < (sizeof(ns) + size_val)) {
if (buffer->size > (MAX_BUF_SIZE - size_val - BUF_SIZE)) {
error("packmem: buffer size too large");
return;
}
buffer->size += (size_val + BUF_SIZE);
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
if (size_val) {
memcpy(&buffer->head[buffer->processed], valp, size_val);
buffer->processed += size_val;
}
}
extern void msg_aggr_add_comp(Buf buffer, void *auth_cred, header_t *header)
{
slurm_msg_t *msg;
if (!msg_collection.running)
return;
msg = xmalloc_nz(sizeof(slurm_msg_t));
slurm_msg_t_init(msg);
msg->protocol_version = header->version;
msg->msg_type = header->msg_type;
msg->flags = header->flags;
msg->auth_cred = auth_cred;
msg->data = buffer;
msg->data_size = remaining_buf(buffer);
msg_aggr_add_msg(msg, 0, NULL);
}
inline static int _pmixp_coll_contrib(pmixp_coll_ring_ctx_t *coll_ctx,
int contrib_id,
uint32_t hop, char *data, size_t size)
{
pmixp_coll_t *coll = _ctx_get_coll(coll_ctx);
char *data_ptr = NULL;
int ret;
/* change the state */
coll->ts = time(NULL);
/* save contribution */
if (!size_buf(coll_ctx->ring_buf)) {
grow_buf(coll_ctx->ring_buf, size * coll->peers_cnt);
} else if(remaining_buf(coll_ctx->ring_buf) < size) {
uint32_t new_size = size_buf(coll_ctx->ring_buf) + size *
_ring_remain_contrib(coll_ctx);
grow_buf(coll_ctx->ring_buf, new_size);
}
grow_buf(coll_ctx->ring_buf, size);
data_ptr = get_buf_data(coll_ctx->ring_buf) +
get_buf_offset(coll_ctx->ring_buf);
memcpy(data_ptr, data, size);
set_buf_offset(coll_ctx->ring_buf,
get_buf_offset(coll_ctx->ring_buf) + size);
/* check for ring is complete */
if (contrib_id != _ring_next_id(coll)) {
/* forward data to the next node */
ret = _ring_forward_data(coll_ctx, contrib_id, hop,
data_ptr, size);
if (ret) {
PMIXP_ERROR("Cannot forward ring data");
return SLURM_ERROR;
}
}
return SLURM_SUCCESS;
}
/*
* Given a pointer to array of char * (char ** or char *[] ) and a size
* (size_val), convert size_val to network byte order and store in the
* buffer followed by the data at valp. Adjust buffer counters.
*/
void packstr_array(char **valp, uint32_t size_val, Buf buffer)
{
int i;
uint32_t ns = htonl(size_val);
if (remaining_buf(buffer) < sizeof(ns)) {
if (buffer->size > (MAX_BUF_SIZE - BUF_SIZE)) {
error("packstr_array: buffer size too large");
return;
}
buffer->size += BUF_SIZE;
xrealloc(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
for (i = 0; i < size_val; i++) {
packstr(valp[i], buffer);
}
}
extern int
temp_kvs_merge(Buf buf)
{
char *data;
uint32_t offset, size;
size = remaining_buf(buf);
if (size == 0) {
return SLURM_SUCCESS;
}
data = get_buf_data(buf);
offset = get_buf_offset(buf);
if (temp_kvs_cnt + size > temp_kvs_size) {
temp_kvs_size += size;
xrealloc(temp_kvs_buf, temp_kvs_size);
}
memcpy(&temp_kvs_buf[temp_kvs_cnt], &data[offset], size);
temp_kvs_cnt += size;
return SLURM_SUCCESS;
}
void _progres_fan_out(pmixp_coll_t *coll, Buf buf)
{
PMIXP_DEBUG("%s:%d: start", pmixp_info_namespace(), pmixp_info_nodeid());
pmixp_coll_sanity_check(coll);
xassert(PMIXP_COLL_FAN_OUT == coll->state || PMIXP_COLL_FAN_OUT_IN == coll->state);
/* update the database */
if (NULL != coll->cbfunc) {
void *data = get_buf_data(buf) + get_buf_offset(buf);
size_t size = remaining_buf(buf);
PMIXP_DEBUG("%s:%d: use the callback", pmixp_info_namespace(),
pmixp_info_nodeid());
coll->cbfunc(PMIX_SUCCESS, data, size, coll->cbdata,
pmixp_free_Buf, (void *)buf);
}
/* Prepare for the next collective operation */
_fan_out_finished(coll);
PMIXP_DEBUG("%s:%d: collective is prepared for the next use",
pmixp_info_namespace(), pmixp_info_nodeid());
}
/*
* Given a pointer to array of char * (char ** or char *[] ) and a size
* (size_val), convert size_val to network byte order and store in the
* buffer followed by the data at valp. Adjust buffer counters.
*/
void packstr_array(char **valp, uint32_t size_val, Buf buffer)
{
int i;
uint32_t ns = htonl(size_val);
if (remaining_buf(buffer) < sizeof(ns)) {
if ((buffer->size + BUF_SIZE) > MAX_BUF_SIZE) {
error("%s: Buffer size limit exceeded (%u > %u)",
__func__, (buffer->size + BUF_SIZE),
MAX_BUF_SIZE);
return;
}
buffer->size += BUF_SIZE;
xrealloc_nz(buffer->head, buffer->size);
}
memcpy(&buffer->head[buffer->processed], &ns, sizeof(ns));
buffer->processed += sizeof(ns);
for (i = 0; i < size_val; i++) {
packstr(valp[i], buffer);
}
}
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);
}
/*
* load_all_part_state - load the partition state from file, recover on
* slurmctld restart. execute this after loading the configuration
* file data.
* NOTE: READ lock_slurmctld config before entry
*/
int load_all_part_state(void)
{
char *part_name = NULL, *allow_groups = NULL, *nodes = NULL;
char *state_file, *data = NULL;
uint32_t max_time, default_time, max_nodes, min_nodes;
uint32_t max_cpus_per_node = INFINITE, grace_time = 0;
time_t time;
uint16_t flags;
uint16_t max_share, preempt_mode, priority, state_up, cr_type;
struct part_record *part_ptr;
uint32_t data_size = 0, name_len;
int data_allocated, data_read = 0, error_code = 0, part_cnt = 0;
int state_fd;
Buf buffer;
char *ver_str = NULL;
char* allow_alloc_nodes = NULL;
uint16_t protocol_version = (uint16_t)NO_VAL;
char* alternate = NULL;
/* read the file */
lock_state_files();
state_fd = _open_part_state_file(&state_file);
if (state_fd < 0) {
info("No partition state file (%s) to recover",
state_file);
error_code = ENOENT;
} else {
data_allocated = BUF_SIZE;
data = xmalloc(data_allocated);
while (1) {
data_read = read(state_fd, &data[data_size],
BUF_SIZE);
if (data_read < 0) {
if (errno == EINTR)
continue;
else {
error("Read error on %s: %m",
state_file);
break;
}
} else if (data_read == 0) /* eof */
break;
data_size += data_read;
data_allocated += data_read;
xrealloc(data, data_allocated);
}
close(state_fd);
}
xfree(state_file);
unlock_state_files();
buffer = create_buf(data, data_size);
safe_unpackstr_xmalloc( &ver_str, &name_len, buffer);
debug3("Version string in part_state header is %s", ver_str);
if (ver_str) {
if (!strcmp(ver_str, PART_STATE_VERSION)) {
protocol_version = SLURM_PROTOCOL_VERSION;
} else if (!strcmp(ver_str, PART_2_5_STATE_VERSION)) {
protocol_version = SLURM_2_5_PROTOCOL_VERSION;
}
}
if (protocol_version == (uint16_t)NO_VAL) {
error("**********************************************************");
error("Can not recover partition state, data version incompatible");
error("**********************************************************");
xfree(ver_str);
free_buf(buffer);
return EFAULT;
}
xfree(ver_str);
safe_unpack_time(&time, buffer);
while (remaining_buf(buffer) > 0) {
if (protocol_version >= SLURM_2_6_PROTOCOL_VERSION) {
safe_unpackstr_xmalloc(&part_name, &name_len, buffer);
safe_unpack32(&grace_time, buffer);
safe_unpack32(&max_time, buffer);
safe_unpack32(&default_time, buffer);
safe_unpack32(&max_cpus_per_node, buffer);
safe_unpack32(&max_nodes, buffer);
safe_unpack32(&min_nodes, buffer);
safe_unpack16(&flags, buffer);
safe_unpack16(&max_share, buffer);
safe_unpack16(&preempt_mode, buffer);
safe_unpack16(&priority, buffer);
if (priority > part_max_priority)
part_max_priority = priority;
safe_unpack16(&state_up, buffer);
safe_unpack16(&cr_type, buffer);
safe_unpackstr_xmalloc(&allow_groups,
&name_len, buffer);
safe_unpackstr_xmalloc(&allow_alloc_nodes,
&name_len, buffer);
safe_unpackstr_xmalloc(&alternate, &name_len, buffer);
safe_unpackstr_xmalloc(&nodes, &name_len, buffer);
if ((flags & PART_FLAG_DEFAULT_CLR) ||
(flags & PART_FLAG_HIDDEN_CLR) ||
(flags & PART_FLAG_NO_ROOT_CLR) ||
(flags & PART_FLAG_ROOT_ONLY_CLR) ||
(flags & PART_FLAG_REQ_RESV_CLR)) {
error("Invalid data for partition %s: flags=%u",
part_name, flags);
error_code = EINVAL;
}
} else if (protocol_version >= SLURM_2_4_PROTOCOL_VERSION) {
safe_unpackstr_xmalloc(&part_name, &name_len, buffer);
safe_unpack32(&grace_time, buffer);
safe_unpack32(&max_time, buffer);
safe_unpack32(&default_time, buffer);
safe_unpack32(&max_nodes, buffer);
safe_unpack32(&min_nodes, buffer);
safe_unpack16(&flags, buffer);
safe_unpack16(&max_share, buffer);
safe_unpack16(&preempt_mode, buffer);
safe_unpack16(&priority, buffer);
if (priority > part_max_priority)
part_max_priority = priority;
cr_type = 0; /* Default value */
safe_unpack16(&state_up, buffer);
safe_unpackstr_xmalloc(&allow_groups,
&name_len, buffer);
safe_unpackstr_xmalloc(&allow_alloc_nodes,
&name_len, buffer);
safe_unpackstr_xmalloc(&alternate, &name_len, buffer);
safe_unpackstr_xmalloc(&nodes, &name_len, buffer);
if ((flags & PART_FLAG_DEFAULT_CLR) ||
(flags & PART_FLAG_HIDDEN_CLR) ||
(flags & PART_FLAG_NO_ROOT_CLR) ||
(flags & PART_FLAG_ROOT_ONLY_CLR) ||
(flags & PART_FLAG_REQ_RESV_CLR)) {
error("Invalid data for partition %s: flags=%u",
part_name, flags);
error_code = EINVAL;
}
} else {
error("load_all_part_state: protocol_version "
"%hu not supported", protocol_version);
goto unpack_error;
}
/* validity test as possible */
if (state_up > PARTITION_UP) {
error("Invalid data for partition %s: state_up=%u",
part_name, state_up);
error_code = EINVAL;
}
if (error_code) {
error("No more partition data will be processed from "
"the checkpoint file");
xfree(allow_groups);
xfree(allow_alloc_nodes);
xfree(alternate);
xfree(part_name);
xfree(nodes);
error_code = EINVAL;
break;
}
/* find record and perform update */
part_ptr = list_find_first(part_list, &list_find_part,
part_name);
part_cnt++;
if (part_ptr == NULL) {
info("load_all_part_state: partition %s missing from "
"configuration file", part_name);
part_ptr = create_part_record();
xfree(part_ptr->name);
part_ptr->name = xstrdup(part_name);
}
part_ptr->flags = flags;
if (part_ptr->flags & PART_FLAG_DEFAULT) {
xfree(default_part_name);
default_part_name = xstrdup(part_name);
default_part_loc = part_ptr;
}
part_ptr->max_time = max_time;
part_ptr->default_time = default_time;
part_ptr->max_cpus_per_node = max_cpus_per_node;
part_ptr->max_nodes = max_nodes;
part_ptr->max_nodes_orig = max_nodes;
part_ptr->min_nodes = min_nodes;
part_ptr->min_nodes_orig = min_nodes;
part_ptr->max_share = max_share;
part_ptr->grace_time = grace_time;
if (preempt_mode != (uint16_t) NO_VAL)
part_ptr->preempt_mode = preempt_mode;
part_ptr->priority = priority;
part_ptr->state_up = state_up;
part_ptr->cr_type = cr_type;
xfree(part_ptr->allow_groups);
part_ptr->allow_groups = allow_groups;
xfree(part_ptr->allow_alloc_nodes);
part_ptr->allow_alloc_nodes = allow_alloc_nodes;
xfree(part_ptr->alternate);
part_ptr->alternate = alternate;
xfree(part_ptr->nodes);
part_ptr->nodes = nodes;
xfree(part_name);
}
info("Recovered state of %d partitions", part_cnt);
free_buf(buffer);
return error_code;
unpack_error:
error("Incomplete partition data checkpoint file");
info("Recovered state of %d partitions", part_cnt);
free_buf(buffer);
return EFAULT;
}
