int job_save(
job *pjob, /* pointer to job structure */
int updatetype) /* 0=quick, 1=full */
{
int fds;
int i;
char namebuf1[MAXPATHLEN];
char namebuf2[MAXPATHLEN];
int openflags;
int redo;
strcpy(namebuf1, path_jobs); /* job directory path */
strcat(namebuf1, pjob->ji_qs.ji_fileprefix);
strcpy(namebuf2, namebuf1); /* setup for later */
#ifdef PBS_MOM
strcat(namebuf1, JOB_FILE_SUFFIX);
#else
if (pjob->ji_isparent == TRUE)
{
strcat(namebuf1, JOB_FILE_TMP_SUFFIX);
}
else
{
strcat(namebuf1, JOB_FILE_SUFFIX);
}
#endif
/* if ji_modified is set, ie an attribute changed, then update mtime */
if (pjob->ji_modified)
{
pjob->ji_wattr[JOB_ATR_mtime].at_val.at_long = time_now;
}
if (updatetype == SAVEJOB_QUICK)
{
openflags = O_WRONLY | O_Sync;
/* NOTE: open, do not create */
fds = open(namebuf1, openflags, 0600);
if (fds < 0)
{
char tmpLine[1024];
snprintf(tmpLine, sizeof(tmpLine), "cannot open file '%s' for job %s in state %s (%s)",
namebuf1,
pjob->ji_qs.ji_jobid,
PJobSubState[MAX(0,pjob->ji_qs.ji_substate)],
(updatetype == 0) ? "quick" : "full");
log_err(errno, "job_save", tmpLine);
/* FAILURE */
return(-1);
}
/* just write the "critical" base structure to the file */
while ((i = write(fds, (char *) & pjob->ji_qs, quicksize)) != (ssize_t)quicksize)
{
if ((i < 0) && (errno == EINTR))
{
/* retry the write */
if (lseek(fds, (off_t)0, SEEK_SET) < 0)
{
log_err(errno, "job_save", "lseek");
fsync(fds);
close(fds);
return(-1);
}
continue;
}
else
{
log_err(errno, "job_save", "quickwrite");
fsync(fds);
close(fds);
/* FAILURE */
return(-1);
}
}
fsync(fds);
close(fds);
}
else /* SAVEJOB_FULL, SAVEJOB_NEW, SAVEJOB_ARY */
{
/*
* 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 structure,
* (2) the attribtes in "encoded" form,
* (3) the attributes in the "external" form, and last
* (4) the dependency list.
*/
strcat(namebuf2, JOB_FILE_COPY);
openflags = O_CREAT | O_WRONLY | O_Sync;
/* NOTE: create file if required */
if (updatetype == SAVEJOB_NEW)
fds = open(namebuf1, openflags, 0600);
else
fds = open(namebuf2, openflags, 0600);
if (fds < 0)
{
log_err(errno, "job_save", "open for full save");
return(-1);
}
for (i = 0;i < MAX_SAVE_TRIES;++i)
{
redo = 0; /* try to save twice */
save_setup(fds);
if (save_struct((char *)&pjob->ji_qs, (size_t)quicksize) != 0)
{
redo++;
}
else if (save_attr(job_attr_def, pjob->ji_wattr, (int)JOB_ATR_LAST) != 0)
{
redo++;
}
#ifdef PBS_MOM
else if (save_tmsock(pjob) != 0)
{
redo++;
}
else if (save_roottask(pjob) != 0)
{
redo++;
}
else if (save_jobflags(pjob) != 0)
{
redo++;
}
#endif /* PBS_MOM */
else if (save_flush() != 0)
{
redo++;
}
if (redo != 0)
{
if (lseek(fds, (off_t)0, SEEK_SET) < 0)
{
log_err(errno, "job_save", "full lseek");
}
}
else
{
break;
}
} /* END for (i) */
fsync(fds);
close(fds);
if (i >= MAX_SAVE_TRIES)
{
if (updatetype == SAVEJOB_FULL)
unlink(namebuf2);
return(-1);
}
if (updatetype == SAVEJOB_FULL)
{
unlink(namebuf1);
if (rename(namebuf2,namebuf1) != 0)
{
LOG_EVENT(
PBSEVENT_ERROR | PBSEVENT_SECURITY,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
"Link in job_save failed");
}
}
pjob->ji_modified = 0;
} /* END (updatetype == SAVEJOB_QUICK) */
return(0);
} /* END job_save() */
int job_save(
job *pjob, /* pointer to job structure */
int updatetype, /* 0=quick, 1=full, 2=new */
int mom_port) /* if 0 ignore otherwise append to end of job name. this is for multi-mom mode */
{
int fds;
int i;
char namebuf1[MAXPATHLEN];
char namebuf2[MAXPATHLEN];
char save_buf[SAVEJOB_BUF_SIZE];
const char *tmp_ptr = NULL;
size_t buf_remaining = sizeof(save_buf);
int openflags;
int redo;
time_t time_now = time(NULL);
#ifdef PBS_MOM
tmp_ptr = JOB_FILE_SUFFIX;
#else
if (pjob->ji_is_array_template == TRUE)
tmp_ptr = (char *)JOB_FILE_TMP_SUFFIX;
else
tmp_ptr = (char *)JOB_FILE_SUFFIX;
#endif
if (mom_port)
{
snprintf(namebuf1, MAXPATHLEN, "%s%s%d%s",
path_jobs, pjob->ji_qs.ji_fileprefix, mom_port, tmp_ptr);
snprintf(namebuf2, MAXPATHLEN, "%s%s%d%s",
path_jobs, pjob->ji_qs.ji_fileprefix, mom_port, JOB_FILE_COPY);
}
else
{
snprintf(namebuf1, MAXPATHLEN, "%s%s%s",
path_jobs, pjob->ji_qs.ji_fileprefix, tmp_ptr);
snprintf(namebuf2, MAXPATHLEN, "%s%s%s",
path_jobs, pjob->ji_qs.ji_fileprefix, JOB_FILE_COPY);
}
/* if ji_modified is set, ie an pbs_attribute changed, then update mtime */
if (pjob->ji_modified)
{
pjob->ji_wattr[JOB_ATR_mtime].at_val.at_long = time_now;
}
if (updatetype == SAVEJOB_QUICK)
{
openflags = O_WRONLY | O_Sync;
/* NOTE: open, do not create */
fds = open(namebuf1, openflags, 0600);
if (fds < 0)
{
char tmpLine[1024];
snprintf(tmpLine, sizeof(tmpLine), "cannot open file '%s' for job %s in state %s (%s)",
namebuf1,
pjob->ji_qs.ji_jobid,
PJobSubState[MAX(0,pjob->ji_qs.ji_substate)],
(updatetype == 0) ? "quick" : "full");
log_err(errno, "job_save", tmpLine);
/* FAILURE */
return(-1);
}
/* just write the "critical" base structure to the file */
while ((i = write_ac_socket(fds, (char *)&pjob->ji_qs, sizeof(pjob->ji_qs))) != sizeof(pjob->ji_qs))
{
if ((i < 0) && (errno == EINTR))
{
/* retry the write */
if (lseek(fds, (off_t)0, SEEK_SET) < 0)
{
log_err(errno, "job_save", (char *)"lseek");
close(fds);
return(-1);
}
continue;
}
else
{
log_err(errno, "job_save", (char *)"quickwrite");
close(fds);
/* FAILURE */
return(-1);
}
}
close(fds);
}
else /* SAVEJOB_FULL, SAVEJOB_NEW, SAVEJOB_ARY */
{
/*
* 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 structure,
* (2) the attribtes in "encoded" form,
* (3) the attributes in the "external" form, and last
* (4) the dependency list.
*/
openflags = O_CREAT | O_WRONLY | O_Sync;
/* NOTE: create file if required */
if (updatetype == SAVEJOB_NEW)
fds = open(namebuf1, openflags, 0600);
else
fds = open(namebuf2, openflags, 0600);
if (fds < 0)
{
log_err(errno, "job_save", (char *)"open for full save");
return(-1);
}
for (i = 0; i < MAX_SAVE_TRIES; i++)
{
redo = 0; /* try to save twice */
#ifndef PBS_MOM
lock_ss();
#endif
if (save_struct((char *)&pjob->ji_qs, sizeof(pjob->ji_qs), fds, save_buf, &buf_remaining, sizeof(save_buf)) != PBSE_NONE)
{
redo++;
}
else if (save_attr(job_attr_def,
pjob->ji_wattr,
JOB_ATR_LAST,
fds,
save_buf,
&buf_remaining,
sizeof(save_buf)) != PBSE_NONE)
{
redo++;
}
#ifdef PBS_MOM
else if (save_tmsock(pjob, fds, save_buf, &buf_remaining, sizeof(save_buf)) != PBSE_NONE)
{
redo++;
}
#endif /* PBS_MOM */
else if (write_buffer(save_buf, sizeof(save_buf) - buf_remaining, fds) != PBSE_NONE)
{
redo++;
}
#ifndef PBS_MOM
unlock_ss();
#endif
if (redo != 0)
{
if (lseek(fds, (off_t)0, SEEK_SET) < 0)
{
log_err(errno, "job_save", (char *)"full lseek");
}
}
else
{
break;
}
} /* END for (i) */
close(fds);
if (i >= MAX_SAVE_TRIES)
{
if (updatetype == SAVEJOB_FULL)
unlink(namebuf2);
return(-1);
}
if (updatetype == SAVEJOB_FULL)
{
unlink(namebuf1);
if (link(namebuf2, namebuf1) == -1)
{
log_event(
PBSEVENT_ERROR | PBSEVENT_SECURITY,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
(char *)"Link in job_save failed");
}
else
{
unlink(namebuf2);
}
}
pjob->ji_modified = 0;
} /* END (updatetype == SAVEJOB_QUICK) */
return(PBSE_NONE);
} /* END job_save() */
