/* * OpenPBS (Portable Batch System) v2.3 Software License * * Copyright (c) 1999-2000 Veridian Information Solutions, Inc. * All rights reserved. * * --------------------------------------------------------------------------- * For a license to use or redistribute the OpenPBS software under conditions * other than those described below, or to purchase support for this software, * please contact Veridian Systems, PBS Products Department ("Licensor") at: * * www.OpenPBS.org +1 650 967-4675 sales@OpenPBS.org * 877 902-4PBS (US toll-free) * --------------------------------------------------------------------------- * * This license covers use of the OpenPBS v2.3 software (the "Software") at * your site or location, and, for certain users, redistribution of the * Software to other sites and locations. Use and redistribution of * OpenPBS v2.3 in source and binary forms, with or without modification, * are permitted provided that all of the following conditions are met. * After December 31, 2001, only conditions 3-6 must be met: * * 1. Commercial and/or non-commercial use of the Software is permitted * provided a current software registration is on file at www.OpenPBS.org. * If use of this software contributes to a publication, product, or * service, proper attribution must be given; see www.OpenPBS.org/credit.html * * 2. Redistribution in any form is only permitted for non-commercial, * non-profit purposes. There can be no charge for the Software or any * software incorporating the Software. Further, there can be no * expectation of revenue generated as a consequence of redistributing * the Software. * * 3. Any Redistribution of source code must retain the above copyright notice * and the acknowledgment contained in paragraph 6, this list of conditions * and the disclaimer contained in paragraph 7. * * 4. Any Redistribution in binary form must reproduce the above copyright * notice and the acknowledgment contained in paragraph 6, this list of * conditions and the disclaimer contained in paragraph 7 in the * documentation and/or other materials provided with the distribution. * * 5. Redistributions in any form must be accompanied by information on how to * obtain complete source code for the OpenPBS software and any * modifications and/or additions to the OpenPBS software. The source code * must either be included in the distribution or be available for no more * than the cost of distribution plus a nominal fee, and all modifications * and additions to the Software must be freely redistributable by any party * (including Licensor) without restriction. * * 6. All advertising materials mentioning features or use of the Software must * display the following acknowledgment: * * "This product includes software developed by NASA Ames Research Center, * Lawrence Livermore National Laboratory, and Veridian Information * Solutions, Inc. * Visit www.OpenPBS.org for OpenPBS software support, * products, and information." * * 7. DISCLAIMER OF WARRANTY * * THIS SOFTWARE IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND. ANY EXPRESS * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT * ARE EXPRESSLY DISCLAIMED. * * IN NO EVENT SHALL VERIDIAN CORPORATION, ITS AFFILIATED COMPANIES, OR THE * U.S. GOVERNMENT OR ANY OF ITS AGENCIES BE LIABLE FOR ANY DIRECT OR INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, * OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * This license will be governed by the laws of the Commonwealth of Virginia, * without reference to its choice of law rules. */ #include /* the master config generated by configure */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "pbs_error.h" #include "portability.h" #include "list_link.h" #include "server_limits.h" #include "attribute.h" #include "resource.h" #include "pbs_job.h" #include "log.h" #include "mom_mach.h" #include "resmon.h" #include "../rm_dep.h" /* ** System dependent code to gather information for the resource ** monitor for a Fujitsu VPP300 machine running UXP/V ** ** Resources known by this code: ** cput cpu time for a pid or job ** mem memory size for a pid or job in KB ** resi resident memory size for a pid or job in KB ** sessions list of sessions in the system ** pids list of pids in a job ** nsessions number of sessions in the system ** nusers number of users in the system ** ncpus number of cpus ** physmem physical memory size in KB ** availmem available memory size in KB ** totmem total memory size in KB ** size size of a file or filesystem in KB ** idletime seconds of idle time ** walltime wall clock time for a pid ** loadave current load average */ static char ident[] = "@(#) fujitsu/$RCSfile$ $Revision$"; #ifndef TRUE #define FALSE 0 #define TRUE 1 #endif /* TRUE */ #ifndef MAX #define MAX(a,b) (((a)>(b))?(a):(b)) #endif /* MAX */ /* ** external functions and data */ extern struct config *search(struct config *, char *); extern struct rm_attribute *momgetattr(char *); extern int rm_errno; extern unsigned int reqnum; extern double cputfactor; extern double wallfactor; extern long system_ncpus; extern int ignwalltime; extern int igncput; extern int ignvmem; extern int ignmem; /* ** local functions and data */ static char *resi (struct rm_attribute *attrib); static char *physmem (struct rm_attribute *attrib); static char *availmem (struct rm_attribute *attrib); static char *totmem (struct rm_attribute *attrib); static char *ncpus (struct rm_attribute *attrib); static char *walltime (struct rm_attribute *attrib); extern char *loadave (struct rm_attribute *attrib); extern char *nullproc (struct rm_attribute *attrib); char procfs[] = "/proc"; char procfmts[] = "/proc/%s"; char procfmtd[] = "/proc/%05.5d"; DIR *pdir; extern char *ret_string; int kd = -1; time_t wait_time = 10; int nproc = 0; int max_proc = 0; prstatus_t *proc_status = NULL; prpsinfo_t *proc_info = NULL; #define TBL_INC 20 extern time_t time_now; extern char extra_parm[]; extern char no_parm[]; char no_count[] = "count not found"; char nokernel[] = "kernel not available"; char nomemory[] = "no memory"; static long page_size; /* ** local resource array */ struct config dependent_config[] = { { "resi", {resi} }, { "totmem", {totmem} }, { "availmem", {availmem} }, { "physmem", {physmem} }, { "ncpus", {ncpus} }, { "loadave", {loadave} }, { "walltime", {walltime} }, { NULL, {nullproc} }, }; struct nlist nl[] = { { "anoninfo" }, /* 0 */ { "physmem" }, /* 1 */ { "avenrun" }, /* 2 */ { "" } }; #define KSYM_ANON 0 #define KSYM_PHYS 1 #define KSYM_LOAD 2 /* * This routine is called on each cycle of the main loop. */ void dep_main_loop_cycle(void) { /* Don't need any periodic processing. */ } void dep_initialize(void) { int i; static char id[] = "dep_initialize"; char kernel[] = "/unix"; char kmem[] = "/dev/kmem"; page_size = sysconf(_SC_PAGESIZE); if ((pdir = opendir(procfs)) == NULL) { log_err(errno, id, "opendir"); return; } if (kd == -1) { if ((kd = open(kmem, O_RDONLY)) == -1) { log_err(errno, id, kmem); return; } /* insure /dev/kmem closed on exec */ if ((i = fcntl(kd, F_GETFD)) == -1) { log_err(errno, id, "F_GETFD"); } i |= FD_CLOEXEC; if (fcntl(kd, F_SETFD, i) == -1) { log_err(errno, id, "F_SETFD"); } } if (nlist(kernel, nl) == -1) { log_err(errno, id, kernel); } } void dep_cleanup(void) { static char id[] = "dep_cleanup"; log_record(PBSEVENT_SYSTEM, 0, id, "dependent cleanup"); if (pdir) closedir(pdir); close(kd); kd = -1; } /* * Internal size decoding routine. * * Accepts a resource pointer and a pointer to the unsigned long integer * to receive the decoded value. It returns a PBS error code, and the * decoded value in the unsigned long integer. * * sizeof(word) = sizeof(int) */ static int mm_getsize(resource *pres, unsigned long *ret) { unsigned long value; if (pres->rs_value.at_type != ATR_TYPE_SIZE) return (PBSE_ATTRTYPE); value = pres->rs_value.at_val.at_size.atsv_num; if (pres->rs_value.at_val.at_size.atsv_units == ATR_SV_WORDSZ) { if (value > ULONG_MAX / sizeof(int)) return (PBSE_BADATVAL); value *= sizeof(int); } if (value > ULONG_MAX >> pres->rs_value.at_val.at_size.atsv_shift) return (PBSE_BADATVAL); *ret = value << pres->rs_value.at_val.at_size.atsv_shift; return (PBSE_NONE); } /* * Internal time decoding routine. * * Accepts a resource pointer and a pointer to the unsigned long integer * to receive the decoded value. It returns a PBS error code, and the * decoded value of time in seconds in the unsigned long integer. */ static int mm_gettime(resource *pres, unsigned long *ret) { if (pres->rs_value.at_type != ATR_TYPE_LONG) return (PBSE_ATTRTYPE); if (pres->rs_value.at_val.at_long < 0) return (PBSE_BADATVAL); *ret = pres->rs_value.at_val.at_long; return (PBSE_NONE); } /* * Time decoding macro. Accepts a timestruc_t pointer. Returns unsigned long * time in seconds. */ #define tv(val) (ulong)((val).tv_sec + ((val).tv_nsec + 500000000)/1000000000) static int injob(pjob, sid) job *pjob; pid_t sid; { task *ptask; for (ptask = (task *)GET_NEXT(pjob->ji_tasks); ptask; ptask = (task *)GET_NEXT(ptask->ti_jobtask)) { if (ptask->ti_qs.ti_sid <= 1) continue; if (ptask->ti_qs.ti_sid == sid) return TRUE; } return FALSE; } /* * Internal session cpu time decoding routine. * * Accepts a job pointer. Returns the sum of all cpu time * consumed for all tasks executed by the job, in seconds, adjusted * by cputfactor. */ static unsigned long cput_sum(job *pjob) { char *id = "cput_sum"; ulong cputime, addtime; int i; int nps = 0; prstatus_t *ps; prpsinfo_t *pi; cputime = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (!injob(pjob, pi->pr_sid)) continue; nps++; if (pi->pr_state == SZOMB) { cputime += tv(pi->pr_time); DBPRT(("%s: ses %d pid %d (zombie) cputime %d\n", id, pi->pr_sid, pi->pr_pid, cputime)) continue; } ps = &proc_status[i]; addtime = tv(ps->pr_utime) + tv(ps->pr_stime) + tv(ps->pr_cutime) + tv(ps->pr_cstime); cputime += addtime; DBPRT(("%s: ses %d pid %d cputime %d\n", id, pi->pr_sid, ps->pr_pid, cputime)) } if (nps == 0) pjob->ji_flags |= MOM_NO_PROC; else pjob->ji_flags &= ~MOM_NO_PROC; return ((unsigned long)((double)cputime * cputfactor)); } /* * Return TRUE if any process in the job is over limit for cputime usage. */ static int overcput_proc(job *pjob, unsigned long limit) { char *id = "overcput_proc"; ulong memsize; ulong cputime; int i; prstatus_t *ps; memsize = 0; for (i = 0; i < nproc; i++) { ps = &proc_status[i]; if (!injob(pjob, ps->pr_sid)) continue; cputime = (ulong)(cputfactor * (double)( tv(ps->pr_utime) + tv(ps->pr_stime) + tv(ps->pr_cutime) + tv(ps->pr_cstime))); if (cputime > limit) return (TRUE); } return (FALSE); } /* * Internal session memory usage function. * * Returns the total number of bytes of address * space consumed by all current tasks within the session. */ static unsigned long mem_sum(job *pjob) { char *id = "mem_sum"; ulong memsize; int i; prpsinfo_t *pi; memsize = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (!injob(pjob, pi->pr_sid)) continue; memsize += pi->pr_size * page_size; } return (memsize); } /* * Internal session mem (workingset) size function. */ static unsigned long resi_sum(job *pjob) { char *id = "resi_sum"; ulong resisize; int i; prpsinfo_t *pi; resisize = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (!injob(pjob, pi->pr_sid)) continue; resisize += pi->pr_rssize * page_size; } return (resisize); } extern char *msg_momsetlim; /* * Internal error routine */ int error(char *string, int value) { char *message; assert(string != NULL); assert(*string != '\0'); message = pbse_to_txt(value); assert(message != NULL); assert(*message != '\0'); (void)fprintf(stderr, msg_momsetlim, string, message); (void)fflush(stderr); return (value); } /* * Establish system-enforced limits for the job. * * Run through the resource list, checking the values for all items * we recognize. * * If set_mode is SET_LIMIT_SET, then also set hard limits for the * system enforced limits (not-polled). * If anything goes wrong with the process, return a PBS error code * and print a message on standard error. A zero-length resource list * is not an error. * * If set_mode is SET_LIMIT_SET the entry conditions are: * 1. MOM has already forked, and we are called from the child. * 2. The child is still running as root. * 3. Standard error is open to the user's file. * * If set_mode is SET_LIMIT_ALTER, we are beening called to modify * existing limits. Cannot alter those set by setrlimit (kernel) * because we are the wrong process. */ int mom_set_limits( job *pjob, int set_mode /* SET_LIMIT_SET or SET_LIMIT_ALTER */ ) { char *id = "mom_set_limits"; char *pname; int retval; unsigned long value; /* place in which to build resource value */ resource *pres; struct rlimit reslim; unsigned long mem_limit = 0; log_buffer[0] = '\0'; DBPRT(("%s: entered\n", id)) assert(pjob != NULL); assert(pjob->ji_wattr[(int)JOB_ATR_resource].at_type == ATR_TYPE_RESC); pres = (resource *) GET_NEXT(pjob->ji_wattr[(int)JOB_ATR_resource].at_val.at_list); /* * Cycle through all the resource specifications, * setting limits appropriately. */ while (pres != NULL) { assert(pres->rs_defin != NULL); pname = pres->rs_defin->rs_name; assert(pname != NULL); assert(*pname != '\0'); if (strcmp(pname, "cput") == 0) { if (igncput == FALSE) { /* cpu time - check, if less than pcput use it */ retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); } } else if (strcmp(pname, "pcput") == 0) { if (igncput == FALSE) { /* process cpu time - set */ retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); reslim.rlim_cur = reslim.rlim_max = (unsigned long)((double)value / cputfactor); if (setrlimit(RLIMIT_CPU, &reslim) < 0) return (error("RLIMIT_CPU", PBSE_SYSTEM)); } } else if (strcmp(pname, "file") == 0) /* set */ { if (set_mode == SET_LIMIT_SET) { retval = mm_getsize(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); if (value > INT_MAX) return (error(pname, PBSE_BADATVAL)); reslim.rlim_cur = reslim.rlim_max = value; if (setrlimit(RLIMIT_FSIZE, &reslim) < 0) return (error(pname, PBSE_SYSTEM)); } } else if (strcmp(pname, "mem") == 0) /* check */ { if (ignmem == FALSE) { retval = mm_getsize(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); } } else if (strcmp(pname, "vmem") == 0) /* check */ { if (ignvmem == FALSE) { retval = mm_getsize(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); if ((mem_limit == 0) || (value < mem_limit)) mem_limit = value; } } else if (strcmp(pname, "pvmem") == 0) /* set */ { if (ignvmem == FALSE) { if (set_mode == SET_LIMIT_SET) { retval = mm_getsize(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); if (value > INT_MAX) return (error(pname, PBSE_BADATVAL)); if ((mem_limit == 0) || (value < mem_limit)) mem_limit = value; } } } else if (strcmp(pname, "pmem") == 0) /* ignore */ { } else if (strcmp(pname, "walltime") == 0) /* Check */ { retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) return (error(pname, retval)); } else if (strcmp(pname, "nice") == 0) /* set nice */ { if (set_mode == SET_LIMIT_SET) { errno = 0; if ((nice((int)pres->rs_value.at_val.at_long) == -1) && (errno != 0)) return (error(pname, PBSE_BADATVAL)); } } else if ((pres->rs_defin->rs_flags & ATR_DFLAG_RMOMIG) == 0) /* don't recognize and not marked as ignore by mom */ return (error(pname, PBSE_UNKRESC)); pres = (resource *)GET_NEXT(pres->rs_link); } if (set_mode == SET_LIMIT_SET) { /* if either of vmem or pvmem was given, set sys limit to lesser */ if (mem_limit != 0) { reslim.rlim_cur = reslim.rlim_max = mem_limit; if ((ignvmem == 0) && (setrlimit(RLIMIT_VMEM, &reslim) < 0)) return (error("RLIMIT_VMEM", PBSE_SYSTEM)); } } return (PBSE_NONE); } /* * State whether MOM main loop has to poll this job to determine if some * limits are being exceeded. * * Sets flag TRUE if polling is necessary, FALSE otherwise. Actual * polling is done using the mom_over_limit machine-dependent function. */ int mom_do_poll(job *pjob) { char *id = "mom_do_poll"; char *pname; resource *pres; DBPRT(("%s: entered\n", id)) assert(pjob != NULL); assert(pjob->ji_wattr[(int)JOB_ATR_resource].at_type == ATR_TYPE_RESC); pres = (resource *) GET_NEXT(pjob->ji_wattr[(int)JOB_ATR_resource].at_val.at_list); while (pres != NULL) { assert(pres->rs_defin != NULL); pname = pres->rs_defin->rs_name; assert(pname != NULL); assert(*pname != '\0'); if (strcmp(pname, "walltime") == 0 || strcmp(pname, "cput") == 0 || strcmp(pname, "pcput") == 0 || strcmp(pname, "vmem") == 0) return (TRUE); pres = (resource *)GET_NEXT(pres->rs_link); } return (FALSE); } /* * Setup for polling. * * Open kernel device and get namelist info. */ int mom_open_poll(void) { char *id = "mom_open_poll"; DBPRT(("%s: entered\n", id)) proc_status = (prstatus_t *)calloc(sizeof(prstatus_t), TBL_INC); proc_info = (prpsinfo_t *)calloc(sizeof(prpsinfo_t), TBL_INC); if (proc_status == NULL || proc_info == NULL) { log_err(errno, id, "calloc"); return (PBSE_SYSTEM); } max_proc = TBL_INC; return (PBSE_NONE); } /* * Declare start of polling loop. */ int mom_get_sample(void) { static char id[] = "mom_get_sample"; int fd; struct dirent *dent; char procname[100]; prstatus_t *ps; prpsinfo_t *pi; DBPRT(("%s: entered\n", id)) rewinddir(pdir); ps = proc_status; pi = proc_info; nproc = 0; for (fd = -1; (dent = readdir(pdir)) != NULL; close(fd)) { if (!isdigit(dent->d_name[0])) continue; sprintf(procname, procfmts, dent->d_name); if ((fd = open(procname, O_RDONLY)) == -1) continue; if (ioctl(fd, PIOCPSINFO, pi) == -1) { sprintf(log_buffer, "%s: ioctl(PIOCPSINFO)", procname); log_err(errno, id, log_buffer); continue; } if (ioctl(fd, PIOCSTATUS, ps) == -1) { if (errno != ENOENT) { sprintf(log_buffer, "%s: ioctl(PIOCSTATUS)", procname); log_err(errno, id, log_buffer); } memset(ps, sizeof(prstatus_t), '\0'); } if (++nproc == max_proc) { void *hold; DBPRT(("%s: alloc more table space %d\n", id, nproc)) max_proc += TBL_INC; hold = realloc((void *)proc_status, max_proc * sizeof(prstatus_t)); assert(hold != NULL); proc_status = (prstatus_t *)hold; hold = realloc((void *)proc_info, max_proc * sizeof(prpsinfo_t)); assert(hold != NULL); proc_info = (prpsinfo_t *)hold; } ps = &proc_status[nproc]; pi = &proc_info[nproc]; } return (PBSE_NONE); } /* * Measure job resource usage and compare with its limits. * * If it has exceeded any well-formed polled limit return TRUE. * Otherwise, return FALSE. */ int mom_over_limit(job *pjob) { char *id = "mom_over_limit"; char *pname; int retval; unsigned long value, num; resource *pres; assert(pjob != NULL); assert(pjob->ji_wattr[(int)JOB_ATR_resource].at_type == ATR_TYPE_RESC); pres = (resource *) GET_NEXT(pjob->ji_wattr[(int)JOB_ATR_resource].at_val.at_list); DBPRT(("%s: entered\n", id)) for (; pres != NULL; pres = (resource *)GET_NEXT(pres->rs_link)) { assert(pres->rs_defin != NULL); pname = pres->rs_defin->rs_name; assert(pname != NULL); assert(*pname != '\0'); if ((igncput == FALSE) && (strcmp(pname, "cput") == 0)) { retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) continue; if ((num = cput_sum(pjob)) > value) { sprintf(log_buffer, "cput %lu exceeded limit %lu", num, value); return (TRUE); } } else if ((igncput == FALSE) && (strcmp(pname, "pcput") == 0)) { retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) continue; if (overcput_proc(pjob, value)) { sprintf(log_buffer, "pcput exceeded limit %lu", value); return (TRUE); } } else if (strcmp(pname, "vmem") == 0) { retval = mm_getsize(pres, &value); if (retval != PBSE_NONE) continue; if ((ignvmem == 0) && ((num = mem_sum(pjob)) > value)) { sprintf(log_buffer, "vmem %lu exceeded limit %lu", num, value); return (TRUE); } } else if (ignwalltime == 0 && strcmp(pname, "walltime") == 0) { if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_HERE) == 0) continue; retval = mm_gettime(pres, &value); if (retval != PBSE_NONE) continue; num = (unsigned long)((double)(time_now - pjob->ji_qs.ji_stime) * cputfactor); if ((time_now - pjob->ji_qs.ji_stime) > value) { sprintf(log_buffer, "walltime %d exceeded limit %d", num, value); return (TRUE); } } } return (FALSE); } /* * Update the job attribute for resources used. * * The first time this is called for a job, set up resource entries for * each resource that can be reported for this machine. Fill in the * correct values. Return an error code. * * Assumes that the session ID attribute has already been set. */ int mom_set_use(job *pjob) { char *id = "mom_set_use"; resource *pres; attribute *at; resource_def *rd; unsigned long *lp, lnum; assert(pjob != NULL); at = &pjob->ji_wattr[(int)JOB_ATR_resc_used]; assert(at->at_type == ATR_TYPE_RESC); at->at_flags |= ATR_VFLAG_MODIFY; if ((at->at_flags & ATR_VFLAG_SET) == 0) { at->at_flags |= ATR_VFLAG_SET; rd = find_resc_def(svr_resc_def, "cput", svr_resc_size); assert(rd != NULL); pres = add_resource_entry(at, rd); pres->rs_value.at_flags |= ATR_VFLAG_SET; pres->rs_value.at_type = ATR_TYPE_LONG; rd = find_resc_def(svr_resc_def, "vmem", svr_resc_size); assert(rd != NULL); pres = add_resource_entry(at, rd); pres->rs_value.at_flags |= ATR_VFLAG_SET; pres->rs_value.at_type = ATR_TYPE_SIZE; pres->rs_value.at_val.at_size.atsv_shift = 10; /* KB */ pres->rs_value.at_val.at_size.atsv_units = ATR_SV_BYTESZ; rd = find_resc_def(svr_resc_def, "walltime", svr_resc_size); assert(rd != NULL); pres = add_resource_entry(at, rd); pres->rs_value.at_flags |= ATR_VFLAG_SET; pres->rs_value.at_type = ATR_TYPE_LONG; } rd = find_resc_def(svr_resc_def, "cput", svr_resc_size); assert(rd != NULL); pres = find_resc_entry(at, rd); assert(pres != NULL); lp = (unsigned long *) & pres->rs_value.at_val.at_long; lnum = cput_sum(pjob); *lp = MAX(*lp, lnum); rd = find_resc_def(svr_resc_def, "vmem", svr_resc_size); assert(rd != NULL); pres = find_resc_entry(at, rd); assert(pres != NULL); lp = &pres->rs_value.at_val.at_size.atsv_num; lnum = (mem_sum(pjob) + 1023) >> 10; /* in KB */ *lp = MAX(*lp, lnum); rd = find_resc_def(svr_resc_def, "walltime", svr_resc_size); assert(rd != NULL); pres = find_resc_entry(at, rd); assert(pres != NULL); pres->rs_value.at_val.at_long = (long)((double)(time_now - pjob->ji_qs.ji_stime) * wallfactor); return (PBSE_NONE); } /* * Kill a task session. * Call with the task pointer and a signal number. */ int kill_task(task *ptask, int sig, int pg) { char *id = "kill_task"; int ct = 0; prpsinfo_t *pi; int i, sesid; sesid = ptask->ti_qs.ti_sid; if (sesid <= 1) return 0; (void)mom_get_sample(); for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (sesid == pi->pr_sid) { (void)kill(pi->pr_pid, sig); ++ct; } } return ct; } /* * Clean up everything related to polling. * * In the case of the sun, close the kernal if it is open. */ int mom_close_poll(void) { char *id = "mom_close_poll"; DBPRT(("%s: entered\n", id)) if (proc_status) free(proc_status); if (proc_info) free(proc_info); if (pdir) { if (closedir(pdir) != 0) { log_err(errno, id, "closedir"); return (PBSE_SYSTEM); } } return (PBSE_NONE); } /* * mom_does_checkpoint */ int mom_does_checkpoint(void) { return(CST_NONE); } /* * Checkpoint the job. * * If abort is true, kill it too. */ int mach_checkpoint(task *ptask, char *file, int abort) { return (-1); } /* * Restart the job from the checkpoint file. * * Return -1 on error or sid if okay. */ long mach_restart(task *ptask, char *file) { return (-1); } /* ** Return 1 if proc table can be read, 0 otherwise. */ int getprocs(void) { static unsigned int lastproc = 0; static char id[] = "getprocs"; if (lastproc == reqnum) /* don't need new proc table */ return 1; if (mom_get_sample() != PBSE_NONE) return 0; lastproc = reqnum; return 1; } double dsecs(t) timestruc_t *t; { DBPRT(("\tsecs: %d\tnsecs: %d\n", t->tv_sec, t->tv_nsec)) return ((double)t->tv_sec + ((double)t->tv_nsec * 1.0e-9)); } char * cput_job(jobid) pid_t jobid; { char *id = "cput_job"; int found = 0; int i; double cputime, addtime; prstatus_t *ps; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } cputime = 0.0; for (i = 0; i < nproc; i++) { ps = &proc_status[i]; if (jobid != ps->pr_sid) continue; found = 1; addtime = dsecs(&ps->pr_utime) + dsecs(&ps->pr_stime) + dsecs(&ps->pr_cutime) + dsecs(&ps->pr_cstime); cputime += addtime; DBPRT(("%s: total %.2f pid %d %.2f\n", id, cputime, ps->pr_pid, addtime)) } if (found) { sprintf(ret_string, "%.2f", cputime * cputfactor); return ret_string; } rm_errno = RM_ERR_EXIST; return NULL; } char * cput_proc(pid) pid_t pid; { char *id = "cput_pid"; double cputime; int i; prstatus_t *ps; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } for (i = 0; i < nproc; i++) { ps = &proc_status[i]; if (ps->pr_pid == pid) break; } if (i == nproc) { rm_errno = RM_ERR_EXIST; return NULL; } cputime = dsecs(&ps->pr_utime) + dsecs(&ps->pr_stime) + dsecs(&ps->pr_cutime) + dsecs(&ps->pr_cstime); sprintf(ret_string, "%.2f", cputime * cputfactor); return ret_string; } char * cput(struct rm_attribute *attrib) { char *id = "cput"; int value; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if ((value = atoi(attrib->a_value)) == 0) { sprintf(log_buffer, "bad param: %s", attrib->a_value); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (strcmp(attrib->a_qualifier, "session") == 0) return (cput_job((pid_t)value)); else if (strcmp(attrib->a_qualifier, "proc") == 0) return (cput_proc((pid_t)value)); else { rm_errno = RM_ERR_BADPARAM; return NULL; } } char * mem_job(jobid) pid_t jobid; { char *id = "mem_job"; int memsize; int i; int found = 0; prpsinfo_t *pi; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } memsize = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (jobid != pi->pr_sid) continue; found = 1; memsize += pi->pr_size; DBPRT(("%s: total %d pid %d %d\n", id, memsize*page_size, pi->pr_pid, pi->pr_size*page_size)) } if (found) { sprintf(ret_string, "%ukb", (memsize*page_size) >> 10); /* KB */ return ret_string; } rm_errno = RM_ERR_EXIST; return NULL; } char * mem_proc(pid) pid_t pid; { char *id = "mem_proc"; prpsinfo_t *pi; int i; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (pi->pr_pid == pid) break; } if (i == nproc) { rm_errno = RM_ERR_EXIST; return NULL; } sprintf(ret_string, "%ukb", (pi->pr_size * page_size) >> 10); /* KB */ return ret_string; } char * mem(struct rm_attribute *attrib) { char *id = "mem"; int value; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if ((value = atoi(attrib->a_value)) == 0) { sprintf(log_buffer, "bad param: %s", attrib->a_value); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (strcmp(attrib->a_qualifier, "session") == 0) return (mem_job((pid_t)value)); else if (strcmp(attrib->a_qualifier, "proc") == 0) return (mem_proc((pid_t)value)); else { rm_errno = RM_ERR_BADPARAM; return NULL; } } static char * resi_job(jobid) pid_t jobid; { char *id = "resi_job"; int resisize; int i; int found = 0; prpsinfo_t *pi; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } resisize = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (jobid != pi->pr_sid) continue; found = 1; resisize += pi->pr_rssize; } if (found) { /* in KB */ sprintf(ret_string, "%ukb", (resisize * page_size) >> 10); return ret_string; } rm_errno = RM_ERR_EXIST; return NULL; } static char * resi_proc(pid) pid_t pid; { char *id = "resi_proc"; prpsinfo_t *pi; int i; if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (pi->pr_pid == pid) break; } if (i == nproc) { rm_errno = RM_ERR_EXIST; return NULL; } /* in KB */ sprintf(ret_string, "%ukb", (pi->pr_rssize * page_size) >> 10); return ret_string; } static char * resi(struct rm_attribute *attrib) { char *id = "resi"; int value; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if ((value = atoi(attrib->a_value)) == 0) { sprintf(log_buffer, "bad param: %s", attrib->a_value); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (strcmp(attrib->a_qualifier, "session") == 0) return (resi_job((pid_t)value)); else if (strcmp(attrib->a_qualifier, "proc") == 0) return (resi_proc((pid_t)value)); else { rm_errno = RM_ERR_BADPARAM; return NULL; } } char * sessions(struct rm_attribute *attrib) { char *id = "sessions"; int i, j; prpsinfo_t *pi; char *fmt; int njids = 0; pid_t *jids, *hold; static int maxjid = 200; register pid_t jobid; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if ((jids = (pid_t *)calloc(maxjid, sizeof(pid_t))) == NULL) { log_err(errno, id, nomemory); rm_errno = RM_ERR_SYSTEM; return NULL; } if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } /* ** Search for members of job */ for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (pi->pr_uid == 0) continue; if ((jobid = pi->pr_sid) == 0) continue; DBPRT(("%s[%d]: pid %d sid %d\n", id, njids, pi->pr_pid, jobid)) for (j = 0; j < njids; j++) { if (jids[j] == jobid) break; } if (j == njids) /* not found */ { if (njids == maxjid) /* need more space */ { maxjid += 100; hold = (pid_t *)realloc(jids, maxjid); if (hold == NULL) { log_err(errno, id, "realloc"); rm_errno = RM_ERR_SYSTEM; free(jids); return NULL; } jids = hold; } jids[njids++] = jobid; /* add jobid to list */ } } fmt = ret_string; for (j = 0; j < njids; j++) { checkret(&fmt, 100); if (j == 0) sprintf(fmt, "%d", (int)jids[j]); else sprintf(fmt, " %d", (int)jids[j]); fmt += strlen(fmt); } free(jids); return ret_string; } char * nsessions(struct rm_attribute *attrib) { char *result, *ch; int num = 1; if ((result = sessions(attrib)) == NULL) return result; for (ch = result; *ch; ch++) { if (*ch == ' ') /* count blanks */ num++; } sprintf(ret_string, "%d", num); return ret_string; } char * pids(struct rm_attribute *attrib) { char *id = "pids"; pid_t jobid; int i, j; prpsinfo_t *pi; char *fmt; int num_pids; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if ((jobid = (pid_t)atoi(attrib->a_value)) == 0) { sprintf(log_buffer, "bad param: %s", attrib->a_value); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (strcmp(attrib->a_qualifier, "session") != 0) { rm_errno = RM_ERR_BADPARAM; return NULL; } if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } /* ** Search for members of session */ fmt = ret_string; num_pids = 0; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; DBPRT(("%s[%d]: pid: %d sid %d\n", id, num_pids, pi->pr_pid, pi->pr_sid)) if (jobid != pi->pr_sid) continue; sprintf(fmt, "%d ", pi->pr_pid); fmt += strlen(fmt); num_pids++; } if (num_pids == 0) { rm_errno = RM_ERR_EXIST; return NULL; } return ret_string; } char * nusers(struct rm_attribute *attrib) { char *id = "nusers"; int i, j; prpsinfo_t *pi; int nuids = 0; uid_t *uids, *hold; static int maxuid = 200; register uid_t uid; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if ((uids = (uid_t *)calloc(maxuid, sizeof(uid_t))) == NULL) { log_err(errno, id, nomemory); rm_errno = RM_ERR_SYSTEM; return NULL; } if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if ((uid = pi->pr_uid) == 0) continue; DBPRT(("%s[%d]: pid %d uid %d\n", id, nuids, pi->pr_pid, uid)) for (j = 0; j < nuids; j++) { if (uids[j] == uid) break; } if (j == nuids) /* not found */ { if (nuids == maxuid) /* need more space */ { maxuid += 100; hold = (uid_t *)realloc(uids, maxuid); if (hold == NULL) { log_err(errno, id, "realloc"); rm_errno = RM_ERR_SYSTEM; free(uids); return NULL; } uids = hold; } uids[nuids++] = uid; /* add uid to list */ } } sprintf(ret_string, "%d", nuids); free(uids); return ret_string; } /* * The following code provoded by Fujitsu and modified for this fuction. * * Return the number of "running" PEs * * Earl J. Dodd, FAI * 4-Oct-1997 */ static char * ncpus(struct rm_attribute *attrib) { char *id = "ncpus"; unsigned int pmem; int i, npes; int maxlistlen; size_t ppnsize; opr_peid_blk_t *peidlist; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } npes = 0; ppnsize = 0; /* get the PE table size */ ppnsize = syscall(SYS_sysopr, SKAIGETPPN); if (ppnsize <= 0) { log_err(errno, id, "SKAIGETPPN"); rm_errno = RM_ERR_SYSTEM; return NULL; } /* alloc the receiver PE table */ peidlist = (opr_peid_blk_t *)calloc(sizeof(opr_peid_blk_t), ppnsize); if (peidlist == NULL) { log_err(errno, id, nomemory); rm_errno = RM_ERR_SYSTEM; return NULL; } /* retrieve individual PE information */ maxlistlen = syscall(SYS_sysopr, SKAIGETPEIDL, ppnsize, peidlist); /* ** maxlistlen should = 16 (I think; really don't know what the ** return codes are) */ for (i = 0; i < ppnsize; i++) { if (peidlist[i].ppid >= 0 && peidlist[i].state > 0) npes++; #ifdef DEBUG fprintf(stderr, "peid: %d, ppid = %d\n", i, peidlist[i].ppid); fprintf(stderr, "peid: %d, peid = %d\n", i, peidlist[i].peid); fprintf(stderr, "peid: %d, pp_type = 0x%x\n", i, peidlist[i].pp_type); fprintf(stderr, "peid: %d, state = 0x%x\n", i, peidlist[i].state); #endif } free(peidlist); sprintf(ret_string, "%d", npes); system_ncpus = npes; return ret_string; } int get_la(double *rv) { char *id = "get_la"; long load; if (kd == -1) { log_err(-1, id, nokernel); return (rm_errno = RM_ERR_SYSTEM); } if (nl[KSYM_LOAD].n_type == 0) { log_err(-1, id, "loadaverage count not found"); return (rm_errno = RM_ERR_SYSTEM); } if (lseek(kd, nl[KSYM_LOAD].n_value, SEEK_SET) == -1) { log_err(errno, id, "lseek"); return (rm_errno = RM_ERR_SYSTEM); } if (read_ac_socket(kd, (char *)&load, sizeof(load)) != sizeof(load)) { log_err(errno, id, "read"); return (rm_errno = RM_ERR_SYSTEM); } *rv = (double)load / 256.0; return 0; } struct anoninfo ai; int getanon(char *id) { static unsigned int lastai = 0; if (lastai == reqnum) /* already have anoninfo */ return 0; if (kd == -1) { log_err(-1, id, nokernel); rm_errno = RM_ERR_SYSTEM; return 1; } if (nl[KSYM_ANON].n_type == 0) { log_err(-1, id, "anoninfo struct not found"); rm_errno = RM_ERR_SYSTEM; return 1; } if (lseek(kd, nl[KSYM_ANON].n_value, SEEK_SET) == -1) { log_err(errno, id, "lseek"); rm_errno = RM_ERR_SYSTEM; return NULL; } if (read_ac_socket(kd, (char *)&ai, sizeof(struct anoninfo)) != sizeof(struct anoninfo)) { log_err(errno, id, "read"); rm_errno = RM_ERR_SYSTEM; return 1; } lastai = reqnum; return 0; } static char * totmem(struct rm_attribute *attrib) { char *id = "totmem"; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (getanon(id)) return NULL; sprintf(ret_string, "%ukb", ctob(ai.ani_max) >> 10); /* KB */ return ret_string; } static char * availmem(struct rm_attribute *attrib) { char *id = "availmem"; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (getanon(id)) return NULL; /* in KB */ sprintf(ret_string, "%ukb", ctob(ai.ani_max - ai.ani_resv) >> 10); return ret_string; } static char * physmem(struct rm_attribute *attrib) { char *id = "physmem"; unsigned int pmem; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (kd == -1) { log_err(-1, id, nokernel); rm_errno = RM_ERR_SYSTEM; return NULL; } if (nl[KSYM_PHYS].n_type == 0) { log_err(-1, id, "physmem count not found"); rm_errno = RM_ERR_SYSTEM; return 0; } if (lseek(kd, nl[KSYM_PHYS].n_value, SEEK_SET) == -1) { log_err(errno, id, "lseek"); rm_errno = RM_ERR_SYSTEM; return NULL; } if (read_ac_socket(kd, (char *)&pmem, sizeof(pmem)) != sizeof(pmem)) { log_err(errno, id, "read"); rm_errno = RM_ERR_SYSTEM; return NULL; } pmem *= getpagesize(); sprintf(ret_string, "%ukb", pmem >> 10); /* in KB */ return ret_string; } char * size_fs(char *param) { char *id = "size_fs"; FILE *mf; struct mntent *mp; struct statvfs fsbuf; if (param[0] != '/') { sprintf(log_buffer, "%s: not full path filesystem name: %s\n", id, param); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (statvfs(param, &fsbuf) == -1) { log_err(errno, id, "statvfs"); rm_errno = RM_ERR_BADPARAM; return NULL; } /* in KB */ sprintf(ret_string, "%lukb", (unsigned long)(((double)fsbuf.f_bsize * (double)fsbuf.f_bfree) / 1024.0)); return ret_string; } char * size_file(char *param) { char *id = "size_file"; struct stat sbuf; if (param[0] != '/') { sprintf(log_buffer, "%s: not full path filesystem name: %s\n", id, param); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (stat(param, &sbuf) == -1) { log_err(errno, id, "stat"); rm_errno = RM_ERR_BADPARAM; return NULL; } sprintf(ret_string, "%ukb", sbuf.st_size >> 10); /* in KB */ return ret_string; } char * size(struct rm_attribute *attrib) { char *id = "size"; char *param; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } param = attrib->a_value; if (strcmp(attrib->a_qualifier, "file") == 0) return (size_file(param)); else if (strcmp(attrib->a_qualifier, "fs") == 0) return (size_fs(param)); else { rm_errno = RM_ERR_BADPARAM; return NULL; } } time_t maxtm; void setmax(char *dev) { struct stat sb; if (stat(dev, &sb) == -1) return; if (maxtm < sb.st_atime) maxtm = sb.st_atime; return; } char * idletime(struct rm_attribute *attrib) { char *id = "idletime"; DIR *dp; struct dirent *de; char ttyname[50]; time_t curtm; if (attrib) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if ((dp = opendir("/dev/pts")) == NULL) { log_err(errno, id, "opendir /dev"); rm_errno = RM_ERR_SYSTEM; return NULL; } maxtm = 0; curtm = time(NULL); setmax("/dev/mouse"); while ((de = readdir(dp)) != NULL) { char *name = de->d_name; if (maxtm >= curtm) break; if (*name == '.') continue; sprintf(ttyname, "/dev/pts/%s", name); setmax(ttyname); } closedir(dp); sprintf(ret_string, "%d", MAX(0, curtm - maxtm)); return ret_string; } static char * walltime(struct rm_attribute *attrib) { char *id = "walltime"; int value; int job; int found = 0; int i; time_t now, start; prpsinfo_t *pi; if (attrib == NULL) { log_err(-1, id, no_parm); rm_errno = RM_ERR_NOPARAM; return NULL; } if ((value = atoi(attrib->a_value)) == 0) { sprintf(log_buffer, "bad param: %s", attrib->a_value); log_err(-1, id, log_buffer); rm_errno = RM_ERR_BADPARAM; return NULL; } if (momgetattr(NULL)) { log_err(-1, id, extra_parm); rm_errno = RM_ERR_BADPARAM; return NULL; } if (strcmp(attrib->a_qualifier, "proc") == 0) job = 0; else if (strcmp(attrib->a_qualifier, "session") == 0) job = 1; else { rm_errno = RM_ERR_BADPARAM; return NULL; } if ((now = time(NULL)) <= 0) { log_err(errno, id, "time"); rm_errno = RM_ERR_SYSTEM; return NULL; } if (getprocs() == 0) { rm_errno = RM_ERR_SYSTEM; return NULL; } start = now; for (i = 0; i < nproc; i++) { pi = &proc_info[i]; if (job) { if (value != pi->pr_sid) continue; } else { if ((pid_t)value != pi->pr_pid) continue; } found = 1; start = MIN(start, pi->pr_start.tv_sec); } if (found) { sprintf(ret_string, "%ld", (long)((double)(now - start) * wallfactor));; return ret_string; } rm_errno = RM_ERR_EXIST; return NULL; } void scan_non_child_tasks(void) { /* NYI */ return; } /* END scan_non_child_tasks() */