📄 mom_start.c
字号:
struct startjob_rtn *sjr;{ uid_t uid = pjob->ji_qs.ji_un.ji_momt.ji_exuid; if ((sjr->sj_session = setjob(uid, WJSIGNAL)) == -1) return (-1); if (sesscntl(sjr->sj_session, S_SETFL, S_BATCH) == -1) return (-1); sjr->sj_pid = getpid(); return (sjr->sj_session);}/*** set_globid - set the global id for a machine type.*/voidset_globid(pjob, sjr) job *pjob; struct startjob_rtn *sjr;{ char cvtbuf[20]; (void)sprintf(cvtbuf, "%ld", (long)sjr->sj_pid); if (pjob->ji_globid) free(pjob->ji_globid); pjob->ji_globid = strdup(cvtbuf); (void)decode_str(&pjob->ji_wattr[JOB_ATR_altid], ATTR_altid, NULL, cvtbuf);}/* * set_mach_vars - setup machine dependent environment variables */intset_mach_vars(pjob, vtab) job *pjob; struct var_table *vtab;{ int ret; /* Add names of temp directories to environment */ ret = mktmpdir(pjob, vtab); switch (ret) { case 0: return 0; case 1: /* old dirs exist, doing remove */ rmtmpdir(pjob->ji_qs.ji_jobid); return JOB_EXEC_RETRY; default: return JOB_EXEC_FAIL1; }}char *set_shell(pjob, pwdp) job *pjob; struct passwd *pwdp;{ char *cp; int i; char *shell; struct array_strings *vstrs; /* * find which shell to use, one specified or the login shell */ shell = pwdp->pw_shell; if ((pjob->ji_wattr[(int)JOB_ATR_shell].at_flags & ATR_VFLAG_SET) && (vstrs = pjob->ji_wattr[(int)JOB_ATR_shell].at_val.at_arst)) { for (i = 0; i < vstrs->as_usedptr; ++i) { cp = strchr(vstrs->as_string[i], '@'); if (cp) { if (!strncmp(mom_host, cp+1, strlen(cp+1))) { *cp = '\0'; /* host name matches */ shell = vstrs->as_string[i]; break; } } else { shell = vstrs->as_string[i]; /* wildcard */ } } } return (shell);}/* * scan_for_terminated - scan the list of runnings jobs for one whose * session id matched that of a terminated child pid. Mark that * job as Exiting. */void scan_for_terminated(){ static char id[] = "scan_for_terminated"; int exiteval; pid_t pid; int sid; job *pjob; task *ptask; int statloc; /* update the latest intelligence about the running jobs; */ /* must be done before we reap the zombies, else we lose the info */ termin_child = 0; if (mom_get_sample() == PBSE_NONE) { pjob = (job *)GET_NEXT(svr_alljobs); while (pjob) { mom_set_use(pjob); pjob = (job *)GET_NEXT(pjob->ji_alljobs); } } while ((pid = waitpid(-1, &statloc, WNOHANG)) > 0) {#ifdef DEBUG sprintf(log_buffer, "Waitpid = %d", pid); LOG_EVENT(PBSEVENT_DEBUG, PBS_EVENTCLASS_SERVER, "PBS_MOM", log_buffer);#endif /* ** see if process was a child doing a special function for MOM */ pjob = (job *)GET_NEXT(svr_alljobs); while (pjob) { if (pid == pjob->ji_momsubt) { if (WIFEXITED(statloc)) statloc = WEXITSTATUS(statloc); else statloc = -1; if (pjob->ji_mompost) { pjob->ji_mompost(pjob, statloc); pjob->ji_mompost = 0; } pjob->ji_momsubt = 0; (void)job_save(pjob, SAVEJOB_QUICK); break; } pjob = (job *)GET_NEXT(pjob->ji_alljobs); } } /* Now figure out which task(s) have terminated (are zombies) */ while ( (sid = waitjob(NULL)) > 0 ) { sprintf(log_buffer, "Waitjob = %d", sid); LOG_EVENT(PBSEVENT_DEBUG, PBS_EVENTCLASS_SERVER, "PBS_MOM", log_buffer); pjob = (job *)GET_NEXT(svr_alljobs); while (pjob) { ptask = (task *)GET_NEXT(pjob->ji_tasks); while (ptask) { if (ptask->ti_qs.ti_sid == sid) break; ptask = (task *)GET_NEXT(ptask->ti_jobtask); } if (ptask != NULL) break; pjob = (job *)GET_NEXT(pjob->ji_alljobs); } if (pjob == NULL) { DBPRT(("%s: sid %d not a task\n", id, sid)) continue; } /* ** We found task within the job which has exited. */ if (WIFEXITED(statloc)) exiteval = WEXITSTATUS(statloc); else if (WIFSIGNALED(statloc)) exiteval = WTERMSIG(statloc) + 10000; else exiteval = 1; DBPRT(("%s: task %d sid %d exit value %d\n", id, ptask->ti_qs.ti_task, sid, exiteval)) kill_task(ptask, SIGKILL); ptask->ti_qs.ti_exitstat = exiteval; ptask->ti_qs.ti_status = TI_STATE_EXITED; task_save(ptask); sprintf(log_buffer, "task %d terminated", ptask->ti_qs.ti_task); LOG_EVENT(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buffer); exiting_tasks = 1; }}#define PTY_SIZE 12/* * creat the master pty, this particular * piece of code does it the hard way, it searches */int open_master(rtn_name) char **rtn_name; /* RETURN name of slave pts */{ int i = 0; int ptc; /* master file descriptor */ static char pty_name[PTY_SIZE+1]; /* "/dev/[pt]tyXY" */ while (1) { (void)sprintf(pty_name, "/dev/pty/%03d", i++); if ((ptc = open(pty_name, O_RDWR | O_NOCTTY, 0)) >= 0) { /* Got a master, fix name to matching slave */ pty_name[5] = 't'; pty_name[6] = 't'; pty_name[7] = 'y'; pty_name[8] = 'p'; *rtn_name = pty_name; return (ptc); } else if (errno == ENOENT) return (-1); /* tried all entries, give up */ }}/* * struct sig_tbl = map of signal names to numbers, * see req_signal() in ../requests.c */struct sig_tbl sig_tbl[] = { { "NULL", 0 }, { "HUP", SIGHUP }, { "INT", SIGINT }, { "QUIT", SIGQUIT }, { "ILL", SIGILL }, { "TRAP", SIGTRAP }, { "ABRT", SIGABRT }, { "IOT", SIGIOT }, { "HWE", SIGHWE }, { "ERR", SIGERR }, { "EMT", SIGEMT }, { "FPE", SIGFPE }, { "KILL", SIGKILL }, { "PRE", SIGPRE }, { "BUS", SIGBUS }, { "ORE", SIGORE }, { "SEGV", SIGSEGV }, { "SYS", SIGSYS }, { "PIPE", SIGPIPE }, { "ALRM", SIGALRM }, { "TERM", SIGTERM }, { "IO", SIGIO }, { "URG", SIGURG }, { "CHLD", SIGCHLD }, { "PWR", SIGPWR }, { "MT", SIGMT }, { "MTKILL", SIGMTKILL }, { "BUFIO", SIGBUFIO }, { "RECOVERY", SIGRECOVERY }, { "UME", SIGUME }, { "DLK", SIGDLK }, { "CPULIM", SIGCPULIM }, { "SHUTDN", SIGSHUTDN }, { "STOP", SIGSTOP }, { "TSTP", SIGTSTP }, { "CONT", SIGCONT }, { "TTIN", SIGTTIN }, { "TTOU", SIGTTOU }, { "WINCH", SIGWINCH }, { "RPE", SIGRPE }, { "WRBKPT", SIGWRBKPT }, { "NOBDM", SIGNOBDM },/* On older Unicoses (before 10.x) SIGAPTEAM was CRI reserved SIGCRAY6 */#if defined(SIGAPTEAM) { "APTEAM", SIGAPTEAM },#else { "CRAY6", SIGCRAY6 },#endif /* SIGAPTEAM *//* On older Unicoses (before 10.x) SIGCANCEL was CRI reserved SIGCRAY7 */#if defined(SIGCANCEL) { "CANCEL", SIGCANCEL },#else { "CRAY7", SIGCRAY7 },#endif /* SIGCANCEL */ { "CRAY8", SIGCRAY8 }, { "CRAY9", SIGCRAY9 }, { "CRAY10", SIGCRAY10 }, { "CRAY11", SIGCRAY11 }, { "CRAY12", SIGCRAY12 }, { "CRAY13", SIGCRAY13 }, { "CRAY14", SIGCRAY14 }, { "INFO", SIGINFO }, { "USR1", SIGUSR1 }, { "USR2", SIGUSR2 }, { (char *)0, -1 }};/* * post_suspend - post exit of child for suspending a job */void post_suspend(pjob, err) job *pjob; int err;{ if (err == 0) { pjob->ji_qs.ji_substate = JOB_SUBSTATE_SUSPEND; pjob->ji_qs.ji_svrflags |= JOB_SVFLG_Suspend; }}/* * post_resume - post exit of child for suspending a job */void post_resume(pjob, err) job *pjob; int err;{ if (err == 0) { /* * adjust walltime for time suspended, ji_momstat contains * time when suspended, ji_stime when job started; adjust * stime to current time minus (back to) amount of time * used before suspended. */ if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_Suspend) { pjob->ji_qs.ji_stime = time_now - (pjob->ji_momstat - pjob->ji_qs.ji_stime); } pjob->ji_qs.ji_substate = JOB_SUBSTATE_RUNNING; pjob->ji_qs.ji_svrflags &= ~JOB_SVFLG_Suspend; }}
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -