int
job_or_resv_save_fs(void *pobj, int updatetype, int objtype)
{
int fds;
int openflags;
int redo;
char *filename;
char namebuf1[MAXPATHLEN+1];
char namebuf2[MAXPATHLEN+1];
char *path = (char*)0;
char *err_msg = (char*)0;
char *err_msgl = (char*)0;
char *prefix = (char*)0;
char *suffix = (char*)0;
char *cpsuffix = (char*)0;
char *p_oid = (char*)0;
long *p_mtime = (long*)0;
int *p_modified = (int*)0;
void *pfixed = (void*)0;
ssize_t i;
size_t fixed_size;
attribute_def *p_attr_def = (attribute_def*)0;
attribute *wattr = (attribute*)0;
int final_attr;
int eventclass;
int pmode;
#ifdef WIN32
pmode = _S_IWRITE | _S_IREAD;
#else
pmode = 0600;
#endif
if (objtype == RESC_RESV_OBJECT || objtype == RESV_JOB_OBJECT) {
#ifndef PBS_MOM /* MOM knows not of resc_resv structs */
resc_resv *presv;
presv = (resc_resv *)pobj;
err_msg = "reservation_save";
err_msgl = "Link in reservation_save failed";
path = path_resvs;
prefix = presv->ri_qs.ri_fileprefix;
suffix = RESV_FILE_SUFFIX;
cpsuffix = RESV_FILE_COPY;
p_modified = &presv->ri_modified;
pfixed = (void *)&presv->ri_qs;
fixed_size = sizeof(struct resvfix);
p_attr_def = resv_attr_def;
wattr = presv->ri_wattr;
final_attr = RESV_ATR_LAST;
p_mtime = &presv->ri_wattr[RESV_ATR_mtime].at_val.at_long;
p_oid = presv->ri_qs.ri_resvID;
eventclass = PBS_EVENTCLASS_RESV;
#else /* PBS_MOM only: Execution will never come here for MOM */
return (-1);
#endif
}
else if (objtype == JOB_OBJECT) {
job *pj = (job *)pobj;
#ifndef PBS_MOM /*MOM knows not of resc_resv structs*/
if (pj->ji_resvp) {
int rc = 0;
if (updatetype == SAVEJOB_QUICK)
rc = job_or_resv_save((void *)pj->ji_resvp,
SAVERESV_QUICK,
RESC_RESV_OBJECT);
else if ((updatetype == SAVEJOB_FULL) ||
(updatetype == SAVEJOB_FULLFORCE) ||
(updatetype == SAVEJOB_NEW))
rc = job_or_resv_save((void *)pj->ji_resvp,
SAVERESV_FULL,
RESC_RESV_OBJECT);
if (rc)
return (rc);
}
#endif
err_msg = "job_save";
err_msgl = "Link in job_save failed";
path = path_jobs;
if (*pj->ji_qs.ji_fileprefix != '\0')
prefix = pj->ji_qs.ji_fileprefix;
else
prefix = pj->ji_qs.ji_jobid;
suffix = JOB_FILE_SUFFIX;
cpsuffix = JOB_FILE_COPY;
p_modified = &pj->ji_modified;
pfixed = (void *)&pj->ji_qs;
fixed_size = sizeof(struct jobfix);
p_attr_def = job_attr_def;
wattr = pj->ji_wattr;
final_attr = JOB_ATR_LAST;
p_mtime = &pj->ji_wattr[JOB_ATR_mtime].at_val.at_long;
p_oid = pj->ji_qs.ji_jobid;
eventclass = PBS_EVENTCLASS_JOB;
return (job_save_fs(pj, updatetype));
} else {
/*Don't expect to get here; incorrect object type*/
return (-1);
}
(void)strcpy(namebuf1, path); /* directory path */
(void)strcat(namebuf1, prefix);
(void)strcpy(namebuf2, namebuf1); /* setup for later */
(void)strcat(namebuf1, suffix);
/*if an attribute changed (modified==1) update mtime*/
if (*p_modified) {
*p_mtime = time_now;
}
if (updatetype == SAVEJOB_QUICK || updatetype == SAVERESV_QUICK) {
openflags = O_WRONLY | O_Sync;
fds = open(namebuf1, openflags, pmode);
if (fds < 0) {
log_err(errno, err_msg, "error on open");
return (-1);
}
#ifdef WIN32
secure_file(namebuf1, "Administrators",
READS_MASK|WRITES_MASK|STANDARD_RIGHTS_REQUIRED);
setmode(fds, O_BINARY);
#endif
/* just write the "critical" base structure to the file */
while ((i = write(fds, (char *)pfixed, fixed_size)) !=
fixed_size) {
if ((i < 0) && (errno == EINTR)) {
/* retry the write */
if (lseek(fds, (off_t)0, SEEK_SET) < 0) {
log_err(errno, err_msg, "lseek");
(void)close(fds);
return (-1);
}
continue;
} else {
log_err(errno, err_msg, "quickwrite");
(void)close(fds);
return (-1);
}
}
#ifdef WIN32
if (_commit(fds) != 0) {
log_err(errno, err_msg,
"flushing job file to disk failed!");
(void)close(fds);
return (-1);
}
#endif
(void)close(fds);
} else {
/*
* write the whole structure to the file.
* For a update, this is done to a new file to protect the
* old against crashs.
* The file is written in four parts:
* (1) the job (resc_resv) structure,
* (2) the attributes in "encoded" form,
* (3) the attributes in the "external" form, and last
* (4) the dependency list.
*/
(void)strcat(namebuf2, cpsuffix);
openflags = O_CREAT | O_WRONLY | O_Sync;
#ifdef WIN32
fix_perms2(namebuf2, namebuf1);
#endif
if (updatetype == SAVEJOB_NEW || updatetype == SAVERESV_NEW)
filename = namebuf1;
else
filename = namebuf2;
fds = open(filename, openflags, pmode);
if (fds < 0) {
log_err(errno, err_msg, "open for full save");
return (-1);
}
#ifdef WIN32
secure_file(filename, "Administrators",
READS_MASK|WRITES_MASK|STANDARD_RIGHTS_REQUIRED);
setmode(fds, O_BINARY);
#endif
redo = 0; /* try to save twice */
do {
save_setup(fds);
if (save_struct((char *)pfixed, fixed_size) != 0) {
redo++;
} else if (save_attr_fs(p_attr_def, wattr, final_attr) != 0) {
redo++;
} else if (save_flush() != 0) {
redo++;
}
if (redo != 0) {
if (lseek(fds, (off_t)0, SEEK_SET) < 0) {
log_err(errno, err_msg, "full lseek");
redo++;
}
}
} while (redo == 1);
#ifdef WIN32
if (_commit(fds) != 0) {
log_err(errno, err_msg,
"flush job file to disk failed!");
close(fds);
return (-1);
}
#endif
(void)close(fds);
if (redo > 1) {
if (updatetype == SAVEJOB_FULL ||
updatetype == SAVEJOB_FULLFORCE ||
updatetype == SAVERESV_FULL)
(void)unlink(namebuf2);
return (-1);
}
if (updatetype == SAVEJOB_FULL ||
updatetype == SAVEJOB_FULLFORCE ||
updatetype == SAVERESV_FULL) {
#ifdef WIN32
if (MoveFileEx(namebuf2, namebuf1,
MOVEFILE_REPLACE_EXISTING|MOVEFILE_WRITE_THROUGH) ==
0) {
errno = GetLastError();
sprintf(log_buffer, "MoveFileEx(%s,%s) failed!",
namebuf2, namebuf1);
log_err(errno, err_msg, log_buffer);
}
secure_file(namebuf1, "Administrators",
READS_MASK|WRITES_MASK|STANDARD_RIGHTS_REQUIRED);
#else
if (rename(namebuf2, namebuf1) == -1) {
log_event(PBSEVENT_ERROR|PBSEVENT_SECURITY,
eventclass, LOG_ERR, p_oid, err_msgl);
}
#endif
}
*p_modified = 0;
}
return (0);
}
void
req_confirmresv(struct batch_request *preq)
{
char buf[PBS_MAXQRESVNAME+PBS_MAXHOSTNAME+256] = {0}; /* FQDN resvID+text */
time_t newstart = 0;
attribute *petime = NULL;
resc_resv *presv = NULL;
int rc = 0;
int state = 0;
int sub = 0;
int resv_count = 0;
int is_degraded = 0;
long next_retry_time = 0;
char *execvnodes = NULL;
char *next_execvnode = NULL;
char **short_xc = NULL;
char **tofree = NULL;
char *str_time = NULL;
extern char server_host[];
int is_being_altered = 0;
char *tmp_buf = NULL;
size_t tmp_buf_size = 0;
if ((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0) {
req_reject(PBSE_PERM, 0, preq);
return;
}
presv = find_resv(preq->rq_ind.rq_run.rq_jid);
if (presv == NULL) {
req_reject(PBSE_UNKRESVID, 0, preq);
return;
}
is_degraded = presv->ri_qs.ri_substate == RESV_DEGRADED ? 1 : 0;
is_being_altered = presv->ri_alter_flags;
if (preq->rq_extend == NULL) {
req_reject(PBSE_resvFail, 0, preq);
return;
}
/* If the reservation was degraded and it could not be reconfirmed by the
* scheduler, then the retry time for that reservation is reset to the half-
* time between now and the time to reservation start or, if the retry time
* is invalid, set it to some time after the soonest occurrence is to start
*/
if (strcmp(preq->rq_extend, PBS_RESV_CONFIRM_FAIL) == 0) {
if (is_degraded && !is_being_altered) {
long degraded_time = presv->ri_degraded_time;
DBPRT(("degraded_time of %s is %s", presv->ri_qs.ri_resvID, ctime(°raded_time)));
next_retry_time = time_now + ((degraded_time - time_now)/2);
/* If reservation is still degraded, and time of degraded resv to start
* is over cutoff from now, then set a time to try again.
*/
if (next_retry_time <= (degraded_time - reserve_retry_cutoff)) {
set_resv_retry(presv, next_retry_time);
str_time = ctime(&(presv->ri_wattr[RESV_ATR_retry].at_val.at_long));
if (str_time != NULL) {
str_time[strlen(str_time)-1] = '\0';
(void)snprintf(log_buffer, sizeof(log_buffer), "Next attempt to reconfirm reservation will be made on %s", str_time);
log_event(PBSEVENT_DEBUG2, PBS_EVENTCLASS_RESV, LOG_NOTICE, presv->ri_qs.ri_resvID, log_buffer);
}
}
else {
/* reached a retry attempt that falls within the cutoff
* When processing an advance reservation, unset retry attribute
*/
if (presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long == 0) {
unset_resv_retry(presv);
}
else {
/* When processing a standing reservation, set a retry time
* past the end time of the soonest occurrence.
*/
set_resv_retry(presv, presv->ri_wattr[RESV_ATR_end].at_val.at_long + RESV_RETRY_DELAY);
}
}
}
else {
if (!is_being_altered)
log_event(PBS_EVENTCLASS_RESV, PBS_EVENTCLASS_RESV,
LOG_INFO, presv->ri_qs.ri_resvID,
"Reservation denied");
/* Clients waiting on an interactive request must be
* notified of the failure to confirm
*/
if ((presv->ri_brp != NULL) &&
(presv->ri_wattr[RESV_ATR_interactive].at_flags &
ATR_VFLAG_SET)) {
presv->ri_wattr[RESV_ATR_interactive].at_flags &= ~ATR_VFLAG_SET;
snprintf(buf, sizeof(buf), "%s DENIED",
presv->ri_qs.ri_resvID);
(void)reply_text(presv->ri_brp,
PBSE_NONE, buf);
presv->ri_brp = NULL;
}
if (!is_being_altered) {
(void)snprintf(log_buffer, sizeof(log_buffer),
"requestor=%s@%s", msg_daemonname, server_host);
account_recordResv(PBS_ACCT_DRss, presv, log_buffer);
log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_RESV,
LOG_NOTICE, presv->ri_qs.ri_resvID,
"reservation deleted");
resv_purge(presv);
}
}
if (presv->ri_qs.ri_state == RESV_BEING_ALTERED) {
resv_revert_alter_times(presv);
log_event(PBSEVENT_RESV, PBS_EVENTCLASS_RESV, LOG_INFO,
presv->ri_qs.ri_resvID, "Reservation alter denied");
}
reply_ack(preq);
return;
}
#ifdef NAS /* localmod 122 */
/* If an advance reservation has already been confirmed there's no
* work to be done.
*/
if (presv->ri_qs.ri_state == RESV_CONFIRMED &&
!presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long) {
reply_ack(preq);
return;
}
#endif /* localmod 122 */
/* Do not alter a reservation that started running when the reconfirmation
* message was received. If a standing reservation, then set a retry time
* past the end of this occurrence.
*/
if (presv->ri_qs.ri_state == RESV_RUNNING) {
if (presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long)
set_resv_retry(presv, presv->ri_wattr[RESV_ATR_end].at_val.at_long + 10);
req_reject(PBSE_TOOLATE, 0, preq);
return;
}
petime = &presv->ri_wattr[RESV_ATR_end];
/* if passed in the confirmation, set a new start time */
if ((newstart = (time_t)preq->rq_ind.rq_run.rq_resch) != 0) {
presv->ri_qs.ri_stime = newstart;
presv->ri_wattr[RESV_ATR_start].at_val.at_long = newstart;
presv->ri_wattr[RESV_ATR_start].at_flags
|= ATR_VFLAG_SET | ATR_VFLAG_MODIFY | ATR_VFLAG_MODCACHE;
presv->ri_qs.ri_etime = newstart + presv->ri_qs.ri_duration;
petime->at_val.at_long = presv->ri_qs.ri_etime;
petime->at_flags |= ATR_VFLAG_SET | ATR_VFLAG_MODIFY | ATR_VFLAG_MODCACHE;
}
/* The main difference between an advance reservation and a standing
* reservation is the format of the execvnodes returned by "rq_destin":
* An advance reservation has a single execvnode while a standing reservation
* has a sting with the particular format:
* <num_resv>#<execvnode1>[<range>]<exevnode2>[...
* describing the execvnodes associated to each occurrence.
*/
if (presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long) {
/* The number of occurrences in the standing reservation and index are parsed
* from the execvnode string which is of the form:
* <num_occurrences>#<vnode1>[range1]<vnode2>[range2]...
*/
resv_count = get_execvnodes_count(preq->rq_ind.rq_run.rq_destin);
if (resv_count == 0) {
req_reject(PBSE_INTERNAL, 0, preq);
return;
}
execvnodes = strdup(preq->rq_ind.rq_run.rq_destin);
if (execvnodes == NULL) {
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
DBPRT(("stdg_resv conf: execvnodes_seq is %s\n", execvnodes));
/* execvnodes is of the form:
* <num_resv>#<(execvnode1)>[<range>]<(exevnode2)>[...
* this "condensed" string is unrolled into a pointer array of
* execvnodes per occurrence, e.g. short_xc[0] are the execvnodes
* for 1st occurrence, short_xc[1] for the 2nd etc...
* If something goes wrong during unrolling then NULL is returned.
* which causes the confirmation message to be rejected
*/
short_xc = unroll_execvnode_seq(execvnodes, &tofree);
if (short_xc == NULL) {
free(execvnodes);
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
/* The execvnode of the soonest (i.e., next) occurrence */
next_execvnode = strdup(short_xc[0]);
if (next_execvnode == NULL) {
free(short_xc);
free_execvnode_seq(tofree);
free(execvnodes);
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
/* Release the now obsolete allocations used to manipulate the
* unrolled string */
free(short_xc);
free_execvnode_seq(tofree);
free(execvnodes);
/* When confirming for the first time, set the index and count */
if (!is_degraded) {
/* Add first occurrence's end date on timed task list */
if (presv->ri_wattr[RESV_ATR_start].at_val.at_long
!= PBS_RESV_FUTURE_SCH) {
if (gen_task_EndResvWindow(presv)) {
free(next_execvnode);
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
}
if (!is_being_altered) {
presv->ri_wattr[RESV_ATR_resv_count].at_val.at_long = resv_count;
presv->ri_wattr[RESV_ATR_resv_count].at_flags |= ATR_VFLAG_SET
| ATR_VFLAG_MODIFY | ATR_VFLAG_MODCACHE;
}
/* Set first occurrence to index 1
* (rather than 0 because it gets displayed in pbs_rstat -f) */
presv->ri_wattr[RESV_ATR_resv_idx].at_val.at_long = 1;
presv->ri_wattr[RESV_ATR_resv_idx].at_flags |= ATR_VFLAG_SET
| ATR_VFLAG_MODIFY | ATR_VFLAG_MODCACHE;
}
/* Skip setting the execvnodes sequence when reconfirming the last
* occurrence or when altering a reservation.
*/
if (!is_being_altered) {
if (presv->ri_wattr[RESV_ATR_resv_idx].at_val.at_long
< presv->ri_wattr[RESV_ATR_resv_count].at_val.at_long) {
/* now assign the execvnodes sequence attribute */
(void) resv_attr_def[(int)RESV_ATR_resv_execvnodes].at_free(
&presv->ri_wattr[(int)RESV_ATR_resv_execvnodes]);
(void) resv_attr_def[(int)RESV_ATR_resv_execvnodes].at_decode(
&presv->ri_wattr[(int)RESV_ATR_resv_execvnodes],
NULL,
NULL,
preq->rq_ind.rq_run.rq_destin);
}
}
}
else { /* Advance reservation */
next_execvnode = strdup(preq->rq_ind.rq_run.rq_destin);
if (next_execvnode == NULL) {
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
}
/* Is reservation still a viable reservation? */
if ((rc = chk_resvReq_viable(presv)) != 0) {
free(next_execvnode);
req_reject(PBSE_BADTSPEC, 0, preq);
return;
}
/* When reconfirming a degraded reservation, first free the nodes linked
* to the reservation and unset all attributes relating to retry attempts
*/
if (is_degraded) {
free_resvNodes(presv);
/* Reset retry time */
unset_resv_retry(presv);
/* reset vnodes_down counter to 0 */
presv->ri_vnodes_down = 0;
}
if (is_being_altered & RESV_END_TIME_MODIFIED) {
if (gen_task_EndResvWindow(presv)) {
free(next_execvnode);
req_reject(PBSE_SYSTEM, 0, preq);
return;
}
}
/*
* Assign the allocated resources to the reservation
* and the reservation to the associated vnodes.
*/
if (is_being_altered)
free_resvNodes(presv);
rc = assign_resv_resc(presv, next_execvnode);
if (rc != PBSE_NONE) {
free(next_execvnode);
req_reject(rc, 0, preq);
return;
}
/* place "Time4resv" task on "task_list_timed" only if this is a
* confirmation but not the reconfirmation of a degraded reservation as
* in this case, the reservation had already been confirmed and added to
* the task list before
*/
if (!is_degraded && (is_being_altered != RESV_END_TIME_MODIFIED) &&
(rc = gen_task_Time4resv(presv)) != 0) {
free(next_execvnode);
req_reject(rc, 0, preq);
return;
}
/*
* compute new values for state and substate
* and update the resc_resv object with these
* newly computed values
*/
eval_resvState(presv, RESVSTATE_gen_task_Time4resv, 0, &state, &sub);
(void)resv_setResvState(presv, state, sub);
cmp_resvStateRelated_attrs((void *)presv,
presv->ri_qs.ri_type);
Update_Resvstate_if_resv(presv->ri_jbp);
if (presv->ri_modified)
(void)job_or_resv_save((void *)presv, SAVERESV_FULL, RESC_RESV_OBJECT);
log_buffer[0] = '\0';
/*
* Notify all interested parties that the reservation
* is moving from state UNCONFIRMED to CONFIRMED
*/
if (presv->ri_brp) {
presv = find_resv(presv->ri_qs.ri_resvID);
if (presv->ri_wattr[(int)RESV_ATR_convert].at_val.at_str != NULL) {
rc = cnvrt_qmove(presv);
if (rc != 0) {
snprintf(buf, sizeof(buf), "%.240s FAILED", presv->ri_qs.ri_resvID);
} else {
snprintf(buf, sizeof(buf), "%.240s CONFIRMED", presv->ri_qs.ri_resvID);
}
} else {
snprintf(buf, sizeof(buf), "%.240s CONFIRMED", presv->ri_qs.ri_resvID);
}
rc = reply_text(presv->ri_brp, PBSE_NONE, buf);
presv->ri_brp = NULL;
}
svr_mailownerResv(presv, MAIL_CONFIRM, MAIL_NORMAL, log_buffer);
presv->ri_wattr[RESV_ATR_interactive].at_flags &= ~ATR_VFLAG_SET;
if (is_being_altered) {
/*
* If the reservation is currently running and its start time is being
* altered after the current time, It is going back to the confirmed state.
* We need to stop the reservation queue as it would have been started at
* the original start time.
* This will prevent any jobs - that are submitted after the
* reservation's start time is changed - from running.
* The reservation went to CO from RN while being altered, that means the reservation
* had resources assigned. We should decrement their usages until it starts running
* again, where the resources will be accounted again.
*/
if (presv->ri_qs.ri_state == RESV_CONFIRMED && presv->ri_alter_state == RESV_RUNNING) {
change_enableORstart(presv, Q_CHNG_START, "FALSE");
if (presv->ri_giveback) {
set_resc_assigned((void *)presv, 1, DECR);
presv->ri_giveback = 0;
}
}
/*
* Reset only the flags and end time backup here, as we will need
* the start time backup in Time4occurrenceFinish for a standing
* reservation. Reset it for an advanced reservation.
*/
if (!(presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long)) {
presv->ri_alter_stime = 0;
}
presv->ri_alter_etime = 0;
presv->ri_alter_flags = 0;
log_event(PBSEVENT_RESV, PBS_EVENTCLASS_RESV, LOG_INFO,
presv->ri_qs.ri_resvID, "Reservation alter confirmed");
} else {
log_event(PBSEVENT_RESV, PBS_EVENTCLASS_RESV, LOG_INFO,
presv->ri_qs.ri_resvID, "Reservation confirmed");
}
if (!is_degraded) {
/* 100 extra bytes for field names, times, and count */
tmp_buf_size = 100 + strlen(preq->rq_user) + strlen(preq->rq_host) + strlen(next_execvnode);
if (tmp_buf_size > sizeof(buf)) {
tmp_buf = malloc(tmp_buf_size);
if (tmp_buf == NULL) {
snprintf(log_buffer, LOG_BUF_SIZE-1, "malloc failure (errno %d)", errno);
log_err(PBSE_SYSTEM, __func__, log_buffer);
free(next_execvnode);
reply_ack(preq);
return;
}
} else {
tmp_buf = buf;
tmp_buf_size = sizeof(buf);
}
if (presv->ri_wattr[RESV_ATR_resv_standing].at_val.at_long) {
(void)snprintf(tmp_buf, tmp_buf_size, "requestor=%s@%s start=%ld end=%ld nodes=%s count=%ld",
preq->rq_user, preq->rq_host,
presv->ri_qs.ri_stime, presv->ri_qs.ri_etime,
next_execvnode,
presv->ri_wattr[RESV_ATR_resv_count].at_val.at_long);
} else {
(void)snprintf(tmp_buf, tmp_buf_size, "requestor=%s@%s start=%ld end=%ld nodes=%s",
preq->rq_user, preq->rq_host,
presv->ri_qs.ri_stime, presv->ri_qs.ri_etime,
next_execvnode);
}
account_recordResv(PBS_ACCT_CR, presv, tmp_buf);
if (tmp_buf != buf) {
free(tmp_buf);
tmp_buf_size = 0;
}
}
free(next_execvnode);
reply_ack(preq);
return;
}
