2 * top - a top users display for Unix
4 * SYNOPSIS: For FreeBSD-2.x and later
7 * Originally written for BSD4.4 system by Christos Zoulas.
8 * Ported to FreeBSD 2.x by Steven Wallace && Wolfram Schneider
9 * Order support hacked in from top-3.5beta6/machine/m_aix41.c
10 * by Monte Mitzelfelt (for latest top see http://www.groupsys.com/topinfo/)
12 * This is the machine-dependent module for FreeBSD 2.2
14 * FreeBSD 2.2.x, 3.x, 4.x, and probably FreeBSD 2.1.x
18 * AUTHOR: Christos Zoulas <christos@ee.cornell.edu>
19 * Steven Wallace <swallace@freebsd.org>
20 * Wolfram Schneider <wosch@FreeBSD.org>
22 * $FreeBSD: src/usr.bin/top/machine.c,v 1.29.2.2 2001/07/31 20:27:05 tmm Exp $
23 * $DragonFly: src/usr.bin/top/machine.c,v 1.10 2003/11/21 22:46:14 dillon Exp $
28 #include <sys/types.h>
29 #include <sys/signal.h>
30 #include <sys/param.h>
38 #include <sys/errno.h>
39 #include <sys/sysctl.h>
40 #include <sys/dkstat.h>
44 #include <sys/vmmeter.h>
45 #include <sys/resource.h>
46 #include <sys/rtprio.h>
52 #include <osreldate.h> /* for changes in kernel structures */
57 static int check_nlist(struct nlist *);
58 static int getkval(unsigned long, int *, int, char *);
59 extern char* printable(char *);
60 int swapmode(int *retavail, int *retfree);
62 static int namelength;
66 /* get_process_info passes back a handle. This is what it looks like: */
70 struct kinfo_proc **next_proc; /* points to next valid proc pointer */
71 int remaining; /* number of pointers remaining */
74 /* declarations for load_avg */
77 #define PP(pp, field) ((pp)->kp_proc . field)
78 #define EP(pp, field) ((pp)->kp_eproc . field)
79 #define TP(pp, field) ((pp)->kp_thread . field)
80 #define VP(pp, field) ((pp)->kp_eproc.e_vm . field)
82 /* define what weighted cpu is. */
83 #define weighted_cpu(pct, pp) (PP((pp), p_swtime) == 0 ? 0.0 : \
84 ((pct) / (1.0 - exp(PP((pp), p_swtime) * logcpu))))
86 /* what we consider to be process size: */
87 #define PROCSIZE(pp) (VP((pp), vm_map.size) / 1024)
89 /* definitions for indices in the nlist array */
91 static struct nlist nlst[] = {
100 { "_bufspace" }, /* K in buffer cache */
109 * These definitions control the format of the per-process area
112 static char smp_header[] =
113 " PID %-*.*s PRI NICE SIZE RES STATE C TIME WCPU CPU COMMAND";
115 #define smp_Proc_format \
116 "%5d %-*.*s %3d %3d%7s %6s %-6.6s %1x%7s %5.2f%% %5.2f%% %.*s"
118 static char up_header[] =
119 " PID %-*.*s PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND";
121 #define up_Proc_format \
122 "%5d %-*.*s %3d %3d%7s %6s %-6.6s%.0d%7s %5.2f%% %5.2f%% %.*s"
126 /* process state names for the "STATE" column of the display */
127 /* the extra nulls in the string "run" are for adding a slash and
128 the processor number when needed */
130 char *state_abbrev[] =
132 "", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB",
138 /* values that we stash away in _init and use in later routines */
140 static double logcpu;
142 /* these are retrieved from the kernel in _init */
144 static load_avg ccpu;
146 /* these are offsets obtained via nlist and used in the get_ functions */
148 static unsigned long cp_time_offset;
149 static unsigned long avenrun_offset;
150 static unsigned long lastpid_offset;
152 static unsigned long bufspace_offset;
155 /* these are for calculating cpu state percentages */
157 static long cp_time[CPUSTATES];
158 static long cp_old[CPUSTATES];
159 static long cp_diff[CPUSTATES];
161 /* these are for detailing the process states */
163 int process_states[6];
164 char *procstatenames[] = {
165 "", " starting, ", " running, ", " sleeping, ", " stopped, ",
170 /* these are for detailing the cpu states */
172 int cpu_states[CPUSTATES];
173 char *cpustatenames[] = {
174 "user", "nice", "system", "interrupt", "idle", NULL
177 /* these are for detailing the memory statistics */
180 char *memorynames[] = {
181 "K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ", "K Free",
186 char *swapnames[] = {
188 "K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
193 /* these are for keeping track of the proc array */
196 static int onproc = -1;
198 static struct kinfo_proc *pbase;
199 static struct kinfo_proc **pref;
201 /* these are for getting the memory statistics */
203 static int pageshift; /* log base 2 of the pagesize */
205 /* define pagetok in terms of pageshift */
207 #define pagetok(size) ((size) << pageshift)
209 /* useful externals */
213 /* sorting orders. first is default */
214 char *ordernames[] = {
215 "cpu", "size", "res", "time", "pri", NULL
220 machine_init(struct statics *statics)
223 register int pagesize;
227 modelen = sizeof(smpmode);
228 if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen, NULL, 0) < 0 &&
229 sysctlbyname("smp.smp_active", &smpmode, &modelen, NULL, 0) < 0) ||
230 modelen != sizeof(smpmode))
233 while ((pw = getpwent()) != NULL) {
234 if (strlen(pw->pw_name) > namelength)
235 namelength = strlen(pw->pw_name);
239 if (smpmode && namelength > 13)
241 else if (namelength > 15)
244 if ((kd = kvm_open(NULL, NULL, NULL, O_RDONLY, "kvm_open")) == NULL)
248 /* get the list of symbols we want to access in the kernel */
249 (void) kvm_nlist(kd, nlst);
250 if (nlst[0].n_type == 0)
252 fprintf(stderr, "top: nlist failed\n");
256 /* make sure they were all found */
257 if (i > 0 && check_nlist(nlst) > 0)
262 (void) getkval(nlst[X_CCPU].n_value, (int *)(&ccpu), sizeof(ccpu),
263 nlst[X_CCPU].n_name);
265 /* stash away certain offsets for later use */
266 cp_time_offset = nlst[X_CP_TIME].n_value;
267 avenrun_offset = nlst[X_AVENRUN].n_value;
268 lastpid_offset = nlst[X_LASTPID].n_value;
269 bufspace_offset = nlst[X_BUFSPACE].n_value;
271 /* this is used in calculating WCPU -- calculate it ahead of time */
272 logcpu = log(loaddouble(ccpu));
278 /* get the page size with "getpagesize" and calculate pageshift from it */
279 pagesize = getpagesize();
287 /* we only need the amount of log(2)1024 for our conversion */
288 pageshift -= LOG1024;
290 /* fill in the statics information */
291 statics->procstate_names = procstatenames;
292 statics->cpustate_names = cpustatenames;
293 statics->memory_names = memorynames;
294 statics->swap_names = swapnames;
296 statics->order_names = ordernames;
303 char *format_header(register char *uname_field)
306 static char Header[128];
308 snprintf(Header, sizeof(Header), smpmode ? smp_header : up_header,
309 namelength, namelength, uname_field);
311 cmdlength = 80 - strlen(Header) + 6;
316 static int swappgsin = -1;
317 static int swappgsout = -1;
318 extern struct timeval timeout;
321 get_system_info(struct system_info *si)
326 struct timeval boottime;
329 /* get the cp_time array */
330 (void) getkval(cp_time_offset, (int *)cp_time, sizeof(cp_time),
331 nlst[X_CP_TIME].n_name);
332 (void) getkval(avenrun_offset, (int *)avenrun, sizeof(avenrun),
333 nlst[X_AVENRUN].n_name);
335 (void) getkval(lastpid_offset, (int *)(&lastpid), sizeof(lastpid),
338 /* convert load averages to doubles */
341 register double *infoloadp;
345 struct loadavg sysload;
347 getkerninfo(KINFO_LOADAVG, &sysload, &size, 0);
350 infoloadp = si->load_avg;
352 for (i = 0; i < 3; i++)
355 *infoloadp++ = ((double) sysload.ldavg[i]) / sysload.fscale;
357 *infoloadp++ = loaddouble(*avenrunp++);
361 /* convert cp_time counts to percentages */
362 total = percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
364 /* sum memory & swap statistics */
368 int vms_size = sizeof(vms);
369 int vmm_size = sizeof(vmm);
370 static unsigned int swap_delay = 0;
371 static int swapavail = 0;
372 static int swapfree = 0;
373 static int bufspace = 0;
375 if (sysctlbyname("vm.vmstats", &vms, &vms_size, NULL, 0)) {
376 perror("sysctlbyname: vm.vmstats");
379 if (sysctlbyname("vm.vmmeter", &vmm, &vmm_size, NULL, 0)) {
380 perror("sysctlbyname: vm.vmstats");
383 (void) getkval(bufspace_offset, (int *)(&bufspace), sizeof(bufspace),
386 /* convert memory stats to Kbytes */
387 memory_stats[0] = pagetok(vms.v_active_count);
388 memory_stats[1] = pagetok(vms.v_inactive_count);
389 memory_stats[2] = pagetok(vms.v_wire_count);
390 memory_stats[3] = pagetok(vms.v_cache_count);
391 memory_stats[4] = bufspace / 1024;
392 memory_stats[5] = pagetok(vms.v_free_count);
393 memory_stats[6] = -1;
401 /* compute differences between old and new swap statistic */
403 swap_stats[4] = pagetok(((vmm.v_swappgsin - swappgsin)));
404 swap_stats[5] = pagetok(((vmm.v_swappgsout - swappgsout)));
407 swappgsin = vmm.v_swappgsin;
408 swappgsout = vmm.v_swappgsout;
410 /* call CPU heavy swapmode() only for changes */
411 if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
412 swap_stats[3] = swapmode(&swapavail, &swapfree);
413 swap_stats[0] = swapavail;
414 swap_stats[1] = swapavail - swapfree;
415 swap_stats[2] = swapfree;
421 /* set arrays and strings */
422 si->cpustates = cpu_states;
423 si->memory = memory_stats;
424 si->swap = swap_stats;
428 si->last_pid = lastpid;
434 * Print how long system has been up.
435 * (Found by looking getting "boottime" from the kernel)
438 mib[1] = KERN_BOOTTIME;
439 bt_size = sizeof(boottime);
440 if (sysctl(mib, 2, &boottime, &bt_size, NULL, 0) != -1 &&
441 boottime.tv_sec != 0) {
442 si->boottime = boottime;
444 si->boottime.tv_sec = -1;
448 static struct handle handle;
450 caddr_t get_process_info(struct system_info *si, struct process_select *sel,
454 register int total_procs;
455 register int active_procs;
456 register struct kinfo_proc **prefp;
457 register struct kinfo_proc *pp;
459 /* these are copied out of sel for speed */
467 pbase = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nproc);
469 pref = (struct kinfo_proc **) realloc(pref, sizeof(struct kinfo_proc *)
471 if (pref == NULL || pbase == NULL) {
472 (void) fprintf(stderr, "top: Out of memory.\n");
475 /* get a pointer to the states summary array */
476 si->procstates = process_states;
478 /* set up flags which define what we are going to select */
479 show_idle = sel->idle;
480 show_self = sel->self;
481 show_system = sel->system;
482 show_uid = sel->uid != -1;
483 show_command = sel->command != NULL;
485 /* count up process states and get pointers to interesting procs */
488 memset((char *)process_states, 0, sizeof(process_states));
490 for (pp = pbase, i = 0; i < nproc; pp++, i++)
493 * Place pointers to each valid proc structure in pref[].
494 * Process slots that are actually in use have a non-zero
495 * status field. Processes with P_SYSTEM set are system
496 * processes---these get ignored unless show_sysprocs is set.
498 if (PP(pp, p_stat) != 0 &&
499 (show_self != PP(pp, p_pid)) &&
500 (show_system || ((PP(pp, p_flag) & P_SYSTEM) == 0)))
503 process_states[(unsigned char) PP(pp, p_stat)]++;
504 if ((PP(pp, p_stat) != SZOMB) &&
505 (show_idle || (PP(pp, p_pctcpu) != 0) ||
506 (PP(pp, p_stat) == SRUN)) &&
507 (!show_uid || EP(pp, e_ucred.cr_ruid) == (uid_t)sel->uid))
515 /* if requested, sort the "interesting" processes */
518 qsort((char *)pref, active_procs, sizeof(struct kinfo_proc *), compare);
521 /* remember active and total counts */
522 si->p_total = total_procs;
523 si->p_active = pref_len = active_procs;
525 /* pass back a handle */
526 handle.next_proc = pref;
527 handle.remaining = active_procs;
528 return((caddr_t)&handle);
531 char fmt[128]; /* static area where result is built */
533 char *format_next_process(caddr_t handle, char *(*get_userid)())
535 struct kinfo_proc *pp;
543 /* find and remember the next proc structure */
544 hp = (struct handle *)handle;
545 pp = *(hp->next_proc++);
548 /* get the process's command name */
549 if ((PP(pp, p_flag) & P_INMEM) == 0) {
551 * Print swapped processes as <pname>
553 char *comm = TP(pp, td_comm);
554 #define COMSIZ sizeof(TP(pp, td_comm))
556 (void) strncpy(buf, comm, COMSIZ);
558 (void) strncpy(&comm[1], buf, COMSIZ - 2);
559 comm[COMSIZ - 2] = '\0';
560 (void) strncat(comm, ">", COMSIZ - 1);
561 comm[COMSIZ - 1] = '\0';
565 * Convert the process's runtime from microseconds to seconds. This
566 * time includes the interrupt time although that is not wanted here.
567 * ps(1) is similarly sloppy.
569 cputime = (EP(pp, e_uticks) + EP(pp, e_sticks)) / 1000000;
571 /* calculate the base for cpu percentages */
572 pct = pctdouble(PP(pp, p_pctcpu));
574 /* generate "STATE" field */
575 switch (state = PP(pp, p_stat)) {
577 if (smpmode && TP(pp, td_flags) & TDF_RUNNING)
578 sprintf(status, "CPU%d", EP(pp, e_cpuid));
580 strcpy(status, "RUN");
583 if (TP(pp, td_wmesg) != NULL) {
584 sprintf(status, "%.6s", EP(pp, e_wmesg));
591 state < sizeof(state_abbrev) / sizeof(*state_abbrev))
592 sprintf(status, "%.6s", state_abbrev[(unsigned char) state]);
594 sprintf(status, "?%5d", state);
599 * idle time 0 - 31 -> nice value +21 - +52
600 * normal time -> nice value -20 - +20
601 * real time 0 - 31 -> nice value -52 - -21
602 * thread 0 - 31 -> nice value -53 -
604 switch(PP(pp, p_rtprio.type)) {
605 case RTP_PRIO_REALTIME:
606 nice = PRIO_MIN - 1 - RTP_PRIO_MAX + PP(pp, p_rtprio.prio);
609 nice = PRIO_MAX + 1 + PP(pp, p_rtprio.prio);
611 case RTP_PRIO_THREAD:
612 nice = PRIO_MIN - 1 - RTP_PRIO_MAX - PP(pp, p_rtprio.prio);
615 nice = PP(pp, p_nice);
620 /* format this entry */
622 smpmode ? smp_Proc_format : up_Proc_format,
624 namelength, namelength,
625 (*get_userid)(EP(pp, e_ucred.cr_ruid)),
628 format_k2(PROCSIZE(pp)),
629 format_k2(pagetok(VP(pp, vm_rssize))),
631 smpmode ? EP(pp, e_cpuid) : 0,
632 format_time(cputime),
633 100.0 * weighted_cpu(pct, pp),
636 printable(TP(pp, td_comm)));
638 /* return the result */
644 * check_nlist(nlst) - checks the nlist to see if any symbols were not
645 * found. For every symbol that was not found, a one-line
646 * message is printed to stderr. The routine returns the
647 * number of symbols NOT found.
650 static int check_nlist(register struct nlist *nlst)
654 /* check to see if we got ALL the symbols we requested */
655 /* this will write one line to stderr for every symbol not found */
658 while (nlst->n_name != NULL)
660 if (nlst->n_type == 0)
662 /* this one wasn't found */
663 (void) fprintf(stderr, "kernel: no symbol named `%s'\n",
675 * getkval(offset, ptr, size, refstr) - get a value out of the kernel.
676 * "offset" is the byte offset into the kernel for the desired value,
677 * "ptr" points to a buffer into which the value is retrieved,
678 * "size" is the size of the buffer (and the object to retrieve),
679 * "refstr" is a reference string used when printing error meessages,
680 * if "refstr" starts with a '!', then a failure on read will not
681 * be fatal (this may seem like a silly way to do things, but I
682 * really didn't want the overhead of another argument).
686 static int getkval(unsigned long offset, int *ptr, int size, char *refstr)
688 if (kvm_read(kd, offset, (char *) ptr, size) != size)
696 fprintf(stderr, "top: kvm_read for %s: %s\n",
697 refstr, strerror(errno));
704 /* comparison routines for qsort */
707 * proc_compare - comparison function for "qsort"
708 * Compares the resource consumption of two processes using five
709 * distinct keys. The keys (in descending order of importance) are:
710 * percent cpu, cpu ticks, state, resident set size, total virtual
711 * memory usage. The process states are ordered as follows (from least
712 * to most important): WAIT, zombie, sleep, stop, start, run. The
713 * array declaration below maps a process state index into a number
714 * that reflects this ordering.
717 static unsigned char sorted_state[] =
721 1, /* ABANDONED (WAIT) */
729 #define ORDERKEY_PCTCPU \
730 if (lresult = (long) PP(p2, p_pctcpu) - (long) PP(p1, p_pctcpu), \
731 (result = lresult > 0 ? 1 : lresult < 0 ? -1 : 0) == 0)
733 #define CPTICKS(p) (EP(p, e_uticks) + EP(p, e_sticks))
735 #define ORDERKEY_CPTICKS \
736 if ((result = CPTICKS(p2) > CPTICKS(p1) ? 1 : \
737 CPTICKS(p2) < CPTICKS(p1) ? -1 : 0) == 0)
739 #define ORDERKEY_STATE \
740 if ((result = sorted_state[(unsigned char) PP(p2, p_stat)] - \
741 sorted_state[(unsigned char) PP(p1, p_stat)]) == 0)
743 #define ORDERKEY_PRIO \
744 if ((result = PP(p2, p_priority) - PP(p1, p_priority)) == 0)
746 #define ORDERKEY_RSSIZE \
747 if ((result = VP(p2, vm_rssize) - VP(p1, vm_rssize)) == 0)
749 #define ORDERKEY_MEM \
750 if ( (result = PROCSIZE(p2) - PROCSIZE(p1)) == 0 )
752 /* compare_cpu - the comparison function for sorting by cpu percentage */
756 compare_cpu(struct proc **pp1, struct proc **pp2)
758 proc_compare(struct proc **pp1, struct proc **pp2)
761 register struct kinfo_proc *p1;
762 register struct kinfo_proc *p2;
764 register pctcpu lresult;
766 /* remove one level of indirection */
767 p1 = *(struct kinfo_proc **) pp1;
768 p2 = *(struct kinfo_proc **) pp2;
782 /* compare routines */
783 int compare_size(), compare_res(), compare_time(), compare_prio();
785 int (*proc_compares[])() = {
794 /* compare_size - the comparison function for sorting by total memory usage */
797 compare_size(struct proc **pp1, struct proc **pp2)
799 register struct kinfo_proc *p1;
800 register struct kinfo_proc *p2;
802 register pctcpu lresult;
804 /* remove one level of indirection */
805 p1 = *(struct kinfo_proc **) pp1;
806 p2 = *(struct kinfo_proc **) pp2;
819 /* compare_res - the comparison function for sorting by resident set size */
822 compare_res(struct proc **pp1, struct proc **pp2)
824 register struct kinfo_proc *p1;
825 register struct kinfo_proc *p2;
827 register pctcpu lresult;
829 /* remove one level of indirection */
830 p1 = *(struct kinfo_proc **) pp1;
831 p2 = *(struct kinfo_proc **) pp2;
844 /* compare_time - the comparison function for sorting by total cpu time */
847 compare_time(struct proc **pp1, struct proc **pp2)
849 register struct kinfo_proc *p1;
850 register struct kinfo_proc *p2;
852 register pctcpu lresult;
854 /* remove one level of indirection */
855 p1 = *(struct kinfo_proc **) pp1;
856 p2 = *(struct kinfo_proc **) pp2;
869 /* compare_prio - the comparison function for sorting by cpu percentage */
872 compare_prio(struct proc **pp1, struct proc **pp2)
874 register struct kinfo_proc *p1;
875 register struct kinfo_proc *p2;
877 register pctcpu lresult;
879 /* remove one level of indirection */
880 p1 = *(struct kinfo_proc **) pp1;
881 p2 = *(struct kinfo_proc **) pp2;
896 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
897 * the process does not exist.
898 * It is EXTREMLY IMPORTANT that this function work correctly.
899 * If top runs setuid root (as in SVR4), then this function
900 * is the only thing that stands in the way of a serious
901 * security problem. It validates requests for the "kill"
902 * and "renice" commands.
905 int proc_owner(int pid)
908 register struct kinfo_proc **prefp;
909 register struct kinfo_proc *pp;
916 if (PP(pp, p_pid) == (pid_t)pid)
918 return((int)EP(pp, e_ucred.cr_ruid));
926 * swapmode is based on a program called swapinfo written
927 * by Kevin Lahey <kml@rokkaku.atl.ga.us>.
930 #define SVAR(var) __STRING(var) /* to force expansion */
931 #define KGET(idx, var) \
932 KGET1(idx, &var, sizeof(var), SVAR(var))
933 #define KGET1(idx, p, s, msg) \
934 KGET2(nlst[idx].n_value, p, s, msg)
935 #define KGET2(addr, p, s, msg) \
936 if (kvm_read(kd, (u_long)(addr), p, s) != s) { \
937 warnx("cannot read %s: %s", msg, kvm_geterr(kd)); \
940 #define KGETRET(addr, p, s, msg) \
941 if (kvm_read(kd, (u_long)(addr), p, s) != s) { \
942 warnx("cannot read %s: %s", msg, kvm_geterr(kd)); \
948 swapmode(int *retavail, int *retfree)
951 int pagesize = getpagesize();
952 struct kvm_swap swapary[1];
957 #define CONVERT(v) ((quad_t)(v) * pagesize / 1024)
959 n = kvm_getswapinfo(kd, swapary, 1, 0);
960 if (n < 0 || swapary[0].ksw_total == 0)
963 *retavail = CONVERT(swapary[0].ksw_total);
964 *retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
966 n = (int)((double)swapary[0].ksw_used * 100.0 /
967 (double)swapary[0].ksw_total);