xref: /freebsd/usr.bin/top/machine.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*
2  * top - a top users display for Unix
3  *
4  * SYNOPSIS:  For FreeBSD-2.x and later
5  *
6  * DESCRIPTION:
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/)
11  *
12  * This is the machine-dependent module for FreeBSD 2.2
13  * Works for:
14  *	FreeBSD 2.2.x, 3.x, 4.x, and probably FreeBSD 2.1.x
15  *
16  * LIBS: -lkvm
17  *
18  * AUTHOR:  Christos Zoulas <christos@ee.cornell.edu>
19  *          Steven Wallace  <swallace@freebsd.org>
20  *          Wolfram Schneider <wosch@FreeBSD.org>
21  *          Thomas Moestl <tmoestl@gmx.net>
22  *
23  * $FreeBSD$
24  */
25 
26 #include <sys/param.h>
27 #include <sys/errno.h>
28 #include <sys/file.h>
29 #include <sys/proc.h>
30 #include <sys/resource.h>
31 #include <sys/rtprio.h>
32 #include <sys/signal.h>
33 #include <sys/sysctl.h>
34 #include <sys/time.h>
35 #include <sys/user.h>
36 #include <sys/vmmeter.h>
37 
38 #include <err.h>
39 #include <kvm.h>
40 #include <math.h>
41 #include <nlist.h>
42 #include <paths.h>
43 #include <pwd.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <string.h>
47 #include <strings.h>
48 #include <unistd.h>
49 #include <vis.h>
50 
51 #include "top.h"
52 #include "machine.h"
53 #include "screen.h"
54 #include "utils.h"
55 #include "layout.h"
56 
57 #define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
58 #define	SMPUNAMELEN	13
59 #define	UPUNAMELEN	15
60 
61 extern struct process_select ps;
62 extern char* printable(char *);
63 static int smpmode;
64 enum displaymodes displaymode;
65 #ifdef TOP_USERNAME_LEN
66 static int namelength = TOP_USERNAME_LEN;
67 #else
68 static int namelength = 8;
69 #endif
70 static int cmdlengthdelta;
71 
72 /* Prototypes for top internals */
73 void quit(int);
74 
75 /* get_process_info passes back a handle.  This is what it looks like: */
76 
77 struct handle {
78 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
79 	int remaining;			/* number of pointers remaining */
80 };
81 
82 /* declarations for load_avg */
83 #include "loadavg.h"
84 
85 /* define what weighted cpu is.  */
86 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
87 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
88 
89 /* what we consider to be process size: */
90 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
91 
92 #define RU(pp)	(&(pp)->ki_rusage)
93 #define RUTOT(pp) \
94 	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
95 
96 
97 /* definitions for indices in the nlist array */
98 
99 /*
100  *  These definitions control the format of the per-process area
101  */
102 
103 static char io_header[] =
104     "  PID%s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
105 
106 #define io_Proc_format \
107     "%5d%s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s"
108 
109 static char smp_header_thr[] =
110     "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE   C   TIME %6s COMMAND";
111 static char smp_header[] =
112     "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE   C   TIME %6s COMMAND";
113 
114 #define smp_Proc_format \
115     "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s %2d%7s %5.2f%% %.*s"
116 
117 static char up_header_thr[] =
118     "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
119 static char up_header[] =
120     "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
121 
122 #define up_Proc_format \
123     "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s%.0d%7s %5.2f%% %.*s"
124 
125 
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 */
129 
130 char *state_abbrev[] = {
131 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
132 };
133 
134 
135 static kvm_t *kd;
136 
137 /* values that we stash away in _init and use in later routines */
138 
139 static double logcpu;
140 
141 /* these are retrieved from the kernel in _init */
142 
143 static load_avg  ccpu;
144 
145 /* these are used in the get_ functions */
146 
147 static int lastpid;
148 
149 /* these are for calculating cpu state percentages */
150 
151 static long cp_time[CPUSTATES];
152 static long cp_old[CPUSTATES];
153 static long cp_diff[CPUSTATES];
154 
155 /* these are for detailing the process states */
156 
157 int process_states[8];
158 char *procstatenames[] = {
159 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
160 	" zombie, ", " waiting, ", " lock, ",
161 	NULL
162 };
163 
164 /* these are for detailing the cpu states */
165 
166 int cpu_states[CPUSTATES];
167 char *cpustatenames[] = {
168 	"user", "nice", "system", "interrupt", "idle", NULL
169 };
170 
171 /* these are for detailing the memory statistics */
172 
173 int memory_stats[7];
174 char *memorynames[] = {
175 	"K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ",
176 	"K Free", NULL
177 };
178 
179 int swap_stats[7];
180 char *swapnames[] = {
181 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
182 	NULL
183 };
184 
185 
186 /* these are for keeping track of the proc array */
187 
188 static int nproc;
189 static int onproc = -1;
190 static int pref_len;
191 static struct kinfo_proc *pbase;
192 static struct kinfo_proc **pref;
193 static struct kinfo_proc *previous_procs;
194 static struct kinfo_proc **previous_pref;
195 static int previous_proc_count = 0;
196 static int previous_proc_count_max = 0;
197 
198 /* total number of io operations */
199 static long total_inblock;
200 static long total_oublock;
201 static long total_majflt;
202 
203 /* these are for getting the memory statistics */
204 
205 static int pageshift;		/* log base 2 of the pagesize */
206 
207 /* define pagetok in terms of pageshift */
208 
209 #define pagetok(size) ((size) << pageshift)
210 
211 /* useful externals */
212 long percentages();
213 
214 #ifdef ORDER
215 /*
216  * Sorting orders.  The first element is the default.
217  */
218 char *ordernames[] = {
219 	"cpu", "size", "res", "time", "pri", "threads",
220 	"total", "read", "write", "fault", "vcsw", "ivcsw",
221 	"jid", NULL
222 };
223 #endif
224 
225 /* Per-cpu time states */
226 static int maxcpu;
227 static int maxid;
228 static int ncpus;
229 static u_long cpumask;
230 static long *times;
231 static long *pcpu_cp_time;
232 static long *pcpu_cp_old;
233 static long *pcpu_cp_diff;
234 static int *pcpu_cpu_states;
235 
236 static int compare_jid(const void *a, const void *b);
237 static int compare_pid(const void *a, const void *b);
238 static const char *format_nice(const struct kinfo_proc *pp);
239 static void getsysctl(const char *name, void *ptr, size_t len);
240 static int swapmode(int *retavail, int *retfree);
241 
242 int
243 machine_init(struct statics *statics, char do_unames)
244 {
245 	int pagesize;
246 	size_t modelen;
247 	struct passwd *pw;
248 
249 	modelen = sizeof(smpmode);
250 	if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen,
251 	    NULL, 0) != 0 &&
252 	    sysctlbyname("kern.smp.active", &smpmode, &modelen,
253 	    NULL, 0) != 0) ||
254 	    modelen != sizeof(smpmode))
255 		smpmode = 0;
256 
257 	if (do_unames) {
258 	    while ((pw = getpwent()) != NULL) {
259 		if (strlen(pw->pw_name) > namelength)
260 			namelength = strlen(pw->pw_name);
261 	    }
262 	}
263 	if (smpmode && namelength > SMPUNAMELEN)
264 		namelength = SMPUNAMELEN;
265 	else if (namelength > UPUNAMELEN)
266 		namelength = UPUNAMELEN;
267 
268 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
269 	if (kd == NULL)
270 		return (-1);
271 
272 	GETSYSCTL("kern.ccpu", ccpu);
273 
274 	/* this is used in calculating WCPU -- calculate it ahead of time */
275 	logcpu = log(loaddouble(ccpu));
276 
277 	pbase = NULL;
278 	pref = NULL;
279 	nproc = 0;
280 	onproc = -1;
281 
282 	/* get the page size and calculate pageshift from it */
283 	pagesize = getpagesize();
284 	pageshift = 0;
285 	while (pagesize > 1) {
286 		pageshift++;
287 		pagesize >>= 1;
288 	}
289 
290 	/* we only need the amount of log(2)1024 for our conversion */
291 	pageshift -= LOG1024;
292 
293 	/* fill in the statics information */
294 	statics->procstate_names = procstatenames;
295 	statics->cpustate_names = cpustatenames;
296 	statics->memory_names = memorynames;
297 	statics->swap_names = swapnames;
298 #ifdef ORDER
299 	statics->order_names = ordernames;
300 #endif
301 
302 	/* Adjust display based on ncpus */
303 	if (pcpu_stats) {
304 		int i, j, empty;
305 		size_t size;
306 
307 		cpumask = 0;
308 		ncpus = 0;
309 		GETSYSCTL("kern.smp.maxcpus", maxcpu);
310 		size = sizeof(long) * maxcpu * CPUSTATES;
311 		times = malloc(size);
312 		if (times == NULL)
313 			err(1, "malloc %zd bytes", size);
314 		if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
315 			err(1, "sysctlbyname kern.cp_times");
316 		pcpu_cp_time = calloc(1, size);
317 		maxid = (size / CPUSTATES / sizeof(long)) - 1;
318 		for (i = 0; i <= maxid; i++) {
319 			empty = 1;
320 			for (j = 0; empty && j < CPUSTATES; j++) {
321 				if (times[i * CPUSTATES + j] != 0)
322 					empty = 0;
323 			}
324 			if (!empty) {
325 				cpumask |= (1ul << i);
326 				ncpus++;
327 			}
328 		}
329 
330 		if (ncpus > 1) {
331 			y_mem += ncpus - 1;	/* 3 */
332 			y_swap += ncpus - 1;	/* 4 */
333 			y_idlecursor += ncpus - 1; /* 5 */
334 			y_message += ncpus - 1;	/* 5 */
335 			y_header += ncpus - 1;	/* 6 */
336 			y_procs += ncpus - 1;	/* 7 */
337 			Header_lines += ncpus - 1; /* 7 */
338 		}
339 		size = sizeof(long) * ncpus * CPUSTATES;
340 		pcpu_cp_old = calloc(1, size);
341 		pcpu_cp_diff = calloc(1, size);
342 		pcpu_cpu_states = calloc(1, size);
343 		statics->ncpus = ncpus;
344 	} else {
345 		statics->ncpus = 1;
346 	}
347 
348 	/* all done! */
349 	return (0);
350 }
351 
352 char *
353 format_header(char *uname_field)
354 {
355 	static char Header[128];
356 	const char *prehead;
357 
358 	switch (displaymode) {
359 	case DISP_CPU:
360 		/*
361 		 * The logic of picking the right header format seems reverse
362 		 * here because we only want to display a THR column when
363 		 * "thread mode" is off (and threads are not listed as
364 		 * separate lines).
365 		 */
366 		prehead = smpmode ?
367 		    (ps.thread ? smp_header : smp_header_thr) :
368 		    (ps.thread ? up_header : up_header_thr);
369 		snprintf(Header, sizeof(Header), prehead,
370 		    ps.jail ? " JID" : "",
371 		    namelength, namelength, uname_field,
372 		    ps.wcpu ? "WCPU" : "CPU");
373 		break;
374 	case DISP_IO:
375 		prehead = io_header;
376 		snprintf(Header, sizeof(Header), prehead,
377 		    ps.jail ? " JID" : "",
378 		    namelength, namelength, uname_field);
379 		break;
380 	}
381 	cmdlengthdelta = strlen(Header) - 7;
382 	return (Header);
383 }
384 
385 static int swappgsin = -1;
386 static int swappgsout = -1;
387 extern struct timeval timeout;
388 
389 
390 void
391 get_system_info(struct system_info *si)
392 {
393 	long total;
394 	struct loadavg sysload;
395 	int mib[2];
396 	struct timeval boottime;
397 	size_t bt_size;
398 	int i, j;
399 	size_t size;
400 
401 	/* get the cp_time array */
402 	if (pcpu_stats) {
403 		size = (maxid + 1) * CPUSTATES * sizeof(long);
404 		if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
405 			err(1, "sysctlbyname kern.cp_times");
406 	} else {
407 		GETSYSCTL("kern.cp_time", cp_time);
408 	}
409 	GETSYSCTL("vm.loadavg", sysload);
410 	GETSYSCTL("kern.lastpid", lastpid);
411 
412 	/* convert load averages to doubles */
413 	for (i = 0; i < 3; i++)
414 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
415 
416 	if (pcpu_stats) {
417 		for (i = j = 0; i <= maxid; i++) {
418 			if ((cpumask & (1ul << i)) == 0)
419 				continue;
420 			/* convert cp_time counts to percentages */
421 			percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
422 			    &pcpu_cp_time[j * CPUSTATES],
423 			    &pcpu_cp_old[j * CPUSTATES],
424 			    &pcpu_cp_diff[j * CPUSTATES]);
425 			j++;
426 		}
427 	} else {
428 		/* convert cp_time counts to percentages */
429 		percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
430 	}
431 
432 	/* sum memory & swap statistics */
433 	{
434 		static unsigned int swap_delay = 0;
435 		static int swapavail = 0;
436 		static int swapfree = 0;
437 		static long bufspace = 0;
438 		static int nspgsin, nspgsout;
439 
440 		GETSYSCTL("vfs.bufspace", bufspace);
441 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
442 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
443 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[2]);
444 		GETSYSCTL("vm.stats.vm.v_cache_count", memory_stats[3]);
445 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
446 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
447 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
448 		/* convert memory stats to Kbytes */
449 		memory_stats[0] = pagetok(memory_stats[0]);
450 		memory_stats[1] = pagetok(memory_stats[1]);
451 		memory_stats[2] = pagetok(memory_stats[2]);
452 		memory_stats[3] = pagetok(memory_stats[3]);
453 		memory_stats[4] = bufspace / 1024;
454 		memory_stats[5] = pagetok(memory_stats[5]);
455 		memory_stats[6] = -1;
456 
457 		/* first interval */
458 		if (swappgsin < 0) {
459 			swap_stats[4] = 0;
460 			swap_stats[5] = 0;
461 		}
462 
463 		/* compute differences between old and new swap statistic */
464 		else {
465 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
466 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
467 		}
468 
469 		swappgsin = nspgsin;
470 		swappgsout = nspgsout;
471 
472 		/* call CPU heavy swapmode() only for changes */
473 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
474 			swap_stats[3] = swapmode(&swapavail, &swapfree);
475 			swap_stats[0] = swapavail;
476 			swap_stats[1] = swapavail - swapfree;
477 			swap_stats[2] = swapfree;
478 		}
479 		swap_delay = 1;
480 		swap_stats[6] = -1;
481 	}
482 
483 	/* set arrays and strings */
484 	if (pcpu_stats) {
485 		si->cpustates = pcpu_cpu_states;
486 		si->ncpus = ncpus;
487 	} else {
488 		si->cpustates = cpu_states;
489 		si->ncpus = 1;
490 	}
491 	si->memory = memory_stats;
492 	si->swap = swap_stats;
493 
494 
495 	if (lastpid > 0) {
496 		si->last_pid = lastpid;
497 	} else {
498 		si->last_pid = -1;
499 	}
500 
501 	/*
502 	 * Print how long system has been up.
503 	 * (Found by looking getting "boottime" from the kernel)
504 	 */
505 	mib[0] = CTL_KERN;
506 	mib[1] = KERN_BOOTTIME;
507 	bt_size = sizeof(boottime);
508 	if (sysctl(mib, 2, &boottime, &bt_size, NULL, 0) != -1 &&
509 	    boottime.tv_sec != 0) {
510 		si->boottime = boottime;
511 	} else {
512 		si->boottime.tv_sec = -1;
513 	}
514 }
515 
516 #define NOPROC	((void *)-1)
517 
518 /*
519  * We need to compare data from the old process entry with the new
520  * process entry.
521  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
522  * structure to cache the mapping.  We also use a negative cache pointer
523  * of NOPROC to avoid duplicate lookups.
524  * XXX: this could be done when the actual processes are fetched, we do
525  * it here out of laziness.
526  */
527 const struct kinfo_proc *
528 get_old_proc(struct kinfo_proc *pp)
529 {
530 	struct kinfo_proc **oldpp, *oldp;
531 
532 	/*
533 	 * If this is the first fetch of the kinfo_procs then we don't have
534 	 * any previous entries.
535 	 */
536 	if (previous_proc_count == 0)
537 		return (NULL);
538 	/* negative cache? */
539 	if (pp->ki_udata == NOPROC)
540 		return (NULL);
541 	/* cached? */
542 	if (pp->ki_udata != NULL)
543 		return (pp->ki_udata);
544 	/*
545 	 * Not cached,
546 	 * 1) look up based on pid.
547 	 * 2) compare process start.
548 	 * If we fail here, then setup a negative cache entry, otherwise
549 	 * cache it.
550 	 */
551 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
552 	    sizeof(*previous_pref), compare_pid);
553 	if (oldpp == NULL) {
554 		pp->ki_udata = NOPROC;
555 		return (NULL);
556 	}
557 	oldp = *oldpp;
558 	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
559 		pp->ki_udata = NOPROC;
560 		return (NULL);
561 	}
562 	pp->ki_udata = oldp;
563 	return (oldp);
564 }
565 
566 /*
567  * Return the total amount of IO done in blocks in/out and faults.
568  * store the values individually in the pointers passed in.
569  */
570 long
571 get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
572     long *vcsw, long *ivcsw)
573 {
574 	const struct kinfo_proc *oldp;
575 	static struct kinfo_proc dummy;
576 	long ret;
577 
578 	oldp = get_old_proc(pp);
579 	if (oldp == NULL) {
580 		bzero(&dummy, sizeof(dummy));
581 		oldp = &dummy;
582 	}
583 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
584 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
585 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
586 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
587 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
588 	ret =
589 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
590 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
591 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
592 	return (ret);
593 }
594 
595 /*
596  * Return the total number of block in/out and faults by a process.
597  */
598 long
599 get_io_total(struct kinfo_proc *pp)
600 {
601 	long dummy;
602 
603 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
604 }
605 
606 static struct handle handle;
607 
608 caddr_t
609 get_process_info(struct system_info *si, struct process_select *sel,
610     int (*compare)(const void *, const void *))
611 {
612 	int i;
613 	int total_procs;
614 	long p_io;
615 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
616 	int active_procs;
617 	struct kinfo_proc **prefp;
618 	struct kinfo_proc *pp;
619 	struct kinfo_proc *prev_pp = NULL;
620 
621 	/* these are copied out of sel for speed */
622 	int show_idle;
623 	int show_self;
624 	int show_system;
625 	int show_uid;
626 	int show_command;
627 	int show_kidle;
628 
629 	/*
630 	 * Save the previous process info.
631 	 */
632 	if (previous_proc_count_max < nproc) {
633 		free(previous_procs);
634 		previous_procs = malloc(nproc * sizeof(*previous_procs));
635 		free(previous_pref);
636 		previous_pref = malloc(nproc * sizeof(*previous_pref));
637 		if (previous_procs == NULL || previous_pref == NULL) {
638 			(void) fprintf(stderr, "top: Out of memory.\n");
639 			quit(23);
640 		}
641 		previous_proc_count_max = nproc;
642 	}
643 	if (nproc) {
644 		for (i = 0; i < nproc; i++)
645 			previous_pref[i] = &previous_procs[i];
646 		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
647 		qsort(previous_pref, nproc, sizeof(*previous_pref),
648 		    compare_pid);
649 	}
650 	previous_proc_count = nproc;
651 
652 	pbase = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nproc);
653 	if (nproc > onproc)
654 		pref = realloc(pref, sizeof(*pref) * (onproc = nproc));
655 	if (pref == NULL || pbase == NULL) {
656 		(void) fprintf(stderr, "top: Out of memory.\n");
657 		quit(23);
658 	}
659 	/* get a pointer to the states summary array */
660 	si->procstates = process_states;
661 
662 	/* set up flags which define what we are going to select */
663 	show_idle = sel->idle;
664 	show_self = sel->self == -1;
665 	show_system = sel->system;
666 	show_uid = sel->uid != -1;
667 	show_command = sel->command != NULL;
668 	show_kidle = sel->kidle;
669 
670 	/* count up process states and get pointers to interesting procs */
671 	total_procs = 0;
672 	active_procs = 0;
673 	total_inblock = 0;
674 	total_oublock = 0;
675 	total_majflt = 0;
676 	memset((char *)process_states, 0, sizeof(process_states));
677 	prefp = pref;
678 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
679 
680 		if (pp->ki_stat == 0)
681 			/* not in use */
682 			continue;
683 
684 		if (!show_self && pp->ki_pid == sel->self)
685 			/* skip self */
686 			continue;
687 
688 		if (!show_system && (pp->ki_flag & P_SYSTEM))
689 			/* skip system process */
690 			continue;
691 
692 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
693 		    &p_vcsw, &p_ivcsw);
694 		total_inblock += p_inblock;
695 		total_oublock += p_oublock;
696 		total_majflt += p_majflt;
697 		total_procs++;
698 		process_states[pp->ki_stat]++;
699 
700 		if (pp->ki_stat == SZOMB)
701 			/* skip zombies */
702 			continue;
703 
704 		if (displaymode == DISP_CPU && !show_idle &&
705 		    (pp->ki_pctcpu == 0 ||
706 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
707 			/* skip idle or non-running processes */
708 			continue;
709 
710 		if (displaymode == DISP_CPU && !show_kidle &&
711 		    pp->ki_tdflags & TDF_IDLETD)
712 			/* skip kernel idle process */
713 			continue;
714 
715 		if (displaymode == DISP_IO && !show_idle && p_io == 0)
716 			/* skip processes that aren't doing I/O */
717 			continue;
718 
719 		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
720 			/* skip proc. that don't belong to the selected UID */
721 			continue;
722 
723 		/*
724 		 * When not showing threads, take the first thread
725 		 * for output and add the fields that we can from
726 		 * the rest of the process's threads rather than
727 		 * using the system's mostly-broken KERN_PROC_PROC.
728 		 */
729 		if (sel->thread || prev_pp == NULL ||
730 		    prev_pp->ki_pid != pp->ki_pid) {
731 			*prefp++ = pp;
732 			active_procs++;
733 			prev_pp = pp;
734 		} else {
735 			prev_pp->ki_pctcpu += pp->ki_pctcpu;
736 			prev_pp->ki_runtime += pp->ki_runtime;
737 		}
738 	}
739 
740 	/* if requested, sort the "interesting" processes */
741 	if (compare != NULL)
742 		qsort(pref, active_procs, sizeof(*pref), compare);
743 
744 	/* remember active and total counts */
745 	si->p_total = total_procs;
746 	si->p_active = pref_len = active_procs;
747 
748 	/* pass back a handle */
749 	handle.next_proc = pref;
750 	handle.remaining = active_procs;
751 	return ((caddr_t)&handle);
752 }
753 
754 static char fmt[128];	/* static area where result is built */
755 
756 char *
757 format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
758 {
759 	struct kinfo_proc *pp;
760 	const struct kinfo_proc *oldp;
761 	long cputime;
762 	double pct;
763 	struct handle *hp;
764 	char status[16];
765 	int state;
766 	struct rusage ru, *rup;
767 	long p_tot, s_tot;
768 	char *proc_fmt, thr_buf[6], jid_buf[6];
769 	char *cmdbuf = NULL;
770 	char **args;
771 
772 	/* find and remember the next proc structure */
773 	hp = (struct handle *)handle;
774 	pp = *(hp->next_proc++);
775 	hp->remaining--;
776 
777 	/* get the process's command name */
778 	if ((pp->ki_flag & P_INMEM) == 0) {
779 		/*
780 		 * Print swapped processes as <pname>
781 		 */
782 		size_t len;
783 
784 		len = strlen(pp->ki_comm);
785 		if (len > sizeof(pp->ki_comm) - 3)
786 			len = sizeof(pp->ki_comm) - 3;
787 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
788 		pp->ki_comm[0] = '<';
789 		pp->ki_comm[len + 1] = '>';
790 		pp->ki_comm[len + 2] = '\0';
791 	}
792 
793 	/*
794 	 * Convert the process's runtime from microseconds to seconds.  This
795 	 * time includes the interrupt time although that is not wanted here.
796 	 * ps(1) is similarly sloppy.
797 	 */
798 	cputime = (pp->ki_runtime + 500000) / 1000000;
799 
800 	/* calculate the base for cpu percentages */
801 	pct = pctdouble(pp->ki_pctcpu);
802 
803 	/* generate "STATE" field */
804 	switch (state = pp->ki_stat) {
805 	case SRUN:
806 		if (smpmode && pp->ki_oncpu != 0xff)
807 			sprintf(status, "CPU%d", pp->ki_oncpu);
808 		else
809 			strcpy(status, "RUN");
810 		break;
811 	case SLOCK:
812 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
813 			sprintf(status, "*%.6s", pp->ki_lockname);
814 			break;
815 		}
816 		/* fall through */
817 	case SSLEEP:
818 		if (pp->ki_wmesg != NULL) {
819 			sprintf(status, "%.6s", pp->ki_wmesg);
820 			break;
821 		}
822 		/* FALLTHROUGH */
823 	default:
824 
825 		if (state >= 0 &&
826 		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
827 			sprintf(status, "%.6s", state_abbrev[state]);
828 		else
829 			sprintf(status, "?%5d", state);
830 		break;
831 	}
832 
833 	cmdbuf = (char *)malloc(cmdlengthdelta + 1);
834 	if (cmdbuf == NULL) {
835 		warn("malloc(%d)", cmdlengthdelta + 1);
836 		return NULL;
837 	}
838 
839 	if (!(flags & FMT_SHOWARGS)) {
840 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
841 		    pp->ki_ocomm[0]) {
842 			snprintf(cmdbuf, cmdlengthdelta, "{%s}", pp->ki_ocomm);
843 		} else {
844 			snprintf(cmdbuf, cmdlengthdelta, "%s", pp->ki_comm);
845 		}
846 	} else {
847 		if (pp->ki_flag & P_SYSTEM ||
848 		    pp->ki_args == NULL ||
849 		    (args = kvm_getargv(kd, pp, cmdlengthdelta)) == NULL ||
850 		    !(*args)) {
851 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
852 		    	pp->ki_ocomm[0]) {
853 				snprintf(cmdbuf, cmdlengthdelta,
854 				    "{%s}", pp->ki_ocomm);
855 			} else {
856 				snprintf(cmdbuf, cmdlengthdelta,
857 				    "[%s]", pp->ki_comm);
858 			}
859 		} else {
860 			char *src, *dst, *argbuf;
861 			char *cmd;
862 			size_t argbuflen;
863 			size_t len;
864 
865 			argbuflen = cmdlengthdelta * 4;
866 			argbuf = (char *)malloc(argbuflen + 1);
867 			if (argbuf == NULL) {
868 				warn("malloc(%d)", argbuflen + 1);
869 				free(cmdbuf);
870 				return NULL;
871 			}
872 
873 			dst = argbuf;
874 
875 			/* Extract cmd name from argv */
876 			cmd = strrchr(*args, '/');
877 			if (cmd == NULL)
878 				cmd = *args;
879 			else
880 				cmd++;
881 
882 			for (; (src = *args++) != NULL; ) {
883 				if (*src == '\0')
884 					continue;
885 				len = (argbuflen - (dst - argbuf) - 1) / 4;
886 				strvisx(dst, src,
887 				    strlen(src) < len ? strlen(src) : len,
888 				    VIS_NL | VIS_CSTYLE);
889 				while (*dst != '\0')
890 					dst++;
891 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
892 					*dst++ = ' '; /* add delimiting space */
893 			}
894 			if (dst != argbuf && dst[-1] == ' ')
895 				dst--;
896 			*dst = '\0';
897 
898 			if (strcmp(cmd, pp->ki_comm) != 0 )
899 				snprintf(cmdbuf, cmdlengthdelta,
900 				    "%s (%s)",argbuf,  pp->ki_comm);
901 			else
902 				strlcpy(cmdbuf, argbuf, cmdlengthdelta);
903 
904 			free(argbuf);
905 		}
906 	}
907 
908 	if (ps.jail == 0)
909 		jid_buf[0] = '\0';
910 	else
911 		snprintf(jid_buf, sizeof(jid_buf), " %*d",
912 		    sizeof(jid_buf) - 3, pp->ki_jid);
913 
914 	if (displaymode == DISP_IO) {
915 		oldp = get_old_proc(pp);
916 		if (oldp != NULL) {
917 			ru.ru_inblock = RU(pp)->ru_inblock -
918 			    RU(oldp)->ru_inblock;
919 			ru.ru_oublock = RU(pp)->ru_oublock -
920 			    RU(oldp)->ru_oublock;
921 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
922 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
923 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
924 			rup = &ru;
925 		} else {
926 			rup = RU(pp);
927 		}
928 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
929 		s_tot = total_inblock + total_oublock + total_majflt;
930 
931 		sprintf(fmt, io_Proc_format,
932 		    pp->ki_pid,
933 		    jid_buf,
934 		    namelength, namelength, (*get_userid)(pp->ki_ruid),
935 		    rup->ru_nvcsw,
936 		    rup->ru_nivcsw,
937 		    rup->ru_inblock,
938 		    rup->ru_oublock,
939 		    rup->ru_majflt,
940 		    p_tot,
941 		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
942 		    screen_width > cmdlengthdelta ?
943 		    screen_width - cmdlengthdelta : 0,
944 		    printable(cmdbuf));
945 
946 		free(cmdbuf);
947 
948 		return (fmt);
949 	}
950 
951 	/* format this entry */
952 	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
953 	if (ps.thread != 0)
954 		thr_buf[0] = '\0';
955 	else
956 		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
957 		    sizeof(thr_buf) - 2, pp->ki_numthreads);
958 
959 	sprintf(fmt, proc_fmt,
960 	    pp->ki_pid,
961 	    jid_buf,
962 	    namelength, namelength, (*get_userid)(pp->ki_ruid),
963 	    thr_buf,
964 	    pp->ki_pri.pri_level - PZERO,
965 	    format_nice(pp),
966 	    format_k2(PROCSIZE(pp)),
967 	    format_k2(pagetok(pp->ki_rssize)),
968 	    status,
969 	    smpmode ? pp->ki_lastcpu : 0,
970 	    format_time(cputime),
971 	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
972 	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
973 	    printable(cmdbuf));
974 
975 	free(cmdbuf);
976 
977 	/* return the result */
978 	return (fmt);
979 }
980 
981 static void
982 getsysctl(const char *name, void *ptr, size_t len)
983 {
984 	size_t nlen = len;
985 
986 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
987 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
988 		    strerror(errno));
989 		quit(23);
990 	}
991 	if (nlen != len) {
992 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
993 		    name, (unsigned long)len, (unsigned long)nlen);
994 		quit(23);
995 	}
996 }
997 
998 static const char *
999 format_nice(const struct kinfo_proc *pp)
1000 {
1001 	const char *fifo, *kthread;
1002 	int rtpri;
1003 	static char nicebuf[4 + 1];
1004 
1005 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1006 	kthread = (pp->ki_flag & P_KTHREAD) ? "k" : "";
1007 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1008 	case PRI_ITHD:
1009 		return ("-");
1010 	case PRI_REALTIME:
1011 		/*
1012 		 * XXX: the kernel doesn't tell us the original rtprio and
1013 		 * doesn't really know what it was, so to recover it we
1014 		 * must be more chummy with the implementation than the
1015 		 * implementation is with itself.  pri_user gives a
1016 		 * constant "base" priority, but is only initialized
1017 		 * properly for user threads.  pri_native gives what the
1018 		 * kernel calls the "base" priority, but it isn't constant
1019 		 * since it is changed by priority propagation.  pri_native
1020 		 * also isn't properly initialized for all threads, but it
1021 		 * is properly initialized for kernel realtime and idletime
1022 		 * threads.  Thus we use pri_user for the base priority of
1023 		 * user threads (it is always correct) and pri_native for
1024 		 * the base priority of kernel realtime and idletime threads
1025 		 * (there is nothing better, and it is usually correct).
1026 		 *
1027 		 * The field width and thus the buffer are too small for
1028 		 * values like "kr31F", but such values shouldn't occur,
1029 		 * and if they do then the tailing "F" is not displayed.
1030 		 */
1031 		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1032 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1033 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1034 		    kthread, rtpri, fifo);
1035 		break;
1036 	case PRI_TIMESHARE:
1037 		if (pp->ki_flag & P_KTHREAD)
1038 			return ("-");
1039 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1040 		break;
1041 	case PRI_IDLE:
1042 		/* XXX: as above. */
1043 		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1044 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1045 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1046 		    kthread, rtpri, fifo);
1047 		break;
1048 	default:
1049 		return ("?");
1050 	}
1051 	return (nicebuf);
1052 }
1053 
1054 /* comparison routines for qsort */
1055 
1056 static int
1057 compare_pid(const void *p1, const void *p2)
1058 {
1059 	const struct kinfo_proc * const *pp1 = p1;
1060 	const struct kinfo_proc * const *pp2 = p2;
1061 
1062 	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1063 		abort();
1064 
1065 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1066 }
1067 
1068 /*
1069  *  proc_compare - comparison function for "qsort"
1070  *	Compares the resource consumption of two processes using five
1071  *	distinct keys.  The keys (in descending order of importance) are:
1072  *	percent cpu, cpu ticks, state, resident set size, total virtual
1073  *	memory usage.  The process states are ordered as follows (from least
1074  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1075  *	array declaration below maps a process state index into a number
1076  *	that reflects this ordering.
1077  */
1078 
1079 static int sorted_state[] = {
1080 	0,	/* not used		*/
1081 	3,	/* sleep		*/
1082 	1,	/* ABANDONED (WAIT)	*/
1083 	6,	/* run			*/
1084 	5,	/* start		*/
1085 	2,	/* zombie		*/
1086 	4	/* stop			*/
1087 };
1088 
1089 
1090 #define ORDERKEY_PCTCPU(a, b) do { \
1091 	long diff; \
1092 	if (ps.wcpu) \
1093 		diff = floor(1.0E6 * weighted_cpu(pctdouble((b)->ki_pctcpu), \
1094 		    (b))) - \
1095 		    floor(1.0E6 * weighted_cpu(pctdouble((a)->ki_pctcpu), \
1096 		    (a))); \
1097 	else \
1098 		diff = (long)(b)->ki_pctcpu - (long)(a)->ki_pctcpu; \
1099 	if (diff != 0) \
1100 		return (diff > 0 ? 1 : -1); \
1101 } while (0)
1102 
1103 #define ORDERKEY_CPTICKS(a, b) do { \
1104 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1105 	if (diff != 0) \
1106 		return (diff > 0 ? 1 : -1); \
1107 } while (0)
1108 
1109 #define ORDERKEY_STATE(a, b) do { \
1110 	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1111 	if (diff != 0) \
1112 		return (diff > 0 ? 1 : -1); \
1113 } while (0)
1114 
1115 #define ORDERKEY_PRIO(a, b) do { \
1116 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1117 	if (diff != 0) \
1118 		return (diff > 0 ? 1 : -1); \
1119 } while (0)
1120 
1121 #define	ORDERKEY_THREADS(a, b) do { \
1122 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1123 	if (diff != 0) \
1124 		return (diff > 0 ? 1 : -1); \
1125 } while (0)
1126 
1127 #define ORDERKEY_RSSIZE(a, b) do { \
1128 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1129 	if (diff != 0) \
1130 		return (diff > 0 ? 1 : -1); \
1131 } while (0)
1132 
1133 #define ORDERKEY_MEM(a, b) do { \
1134 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1135 	if (diff != 0) \
1136 		return (diff > 0 ? 1 : -1); \
1137 } while (0)
1138 
1139 #define ORDERKEY_JID(a, b) do { \
1140 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1141 	if (diff != 0) \
1142 		return (diff > 0 ? 1 : -1); \
1143 } while (0)
1144 
1145 /* compare_cpu - the comparison function for sorting by cpu percentage */
1146 
1147 int
1148 #ifdef ORDER
1149 compare_cpu(void *arg1, void *arg2)
1150 #else
1151 proc_compare(void *arg1, void *arg2)
1152 #endif
1153 {
1154 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1155 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1156 
1157 	ORDERKEY_PCTCPU(p1, p2);
1158 	ORDERKEY_CPTICKS(p1, p2);
1159 	ORDERKEY_STATE(p1, p2);
1160 	ORDERKEY_PRIO(p1, p2);
1161 	ORDERKEY_RSSIZE(p1, p2);
1162 	ORDERKEY_MEM(p1, p2);
1163 
1164 	return (0);
1165 }
1166 
1167 #ifdef ORDER
1168 /* "cpu" compare routines */
1169 int compare_size(), compare_res(), compare_time(), compare_prio(),
1170     compare_threads();
1171 
1172 /*
1173  * "io" compare routines.  Context switches aren't i/o, but are displayed
1174  * on the "io" display.
1175  */
1176 int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1177     compare_vcsw(), compare_ivcsw();
1178 
1179 int (*compares[])() = {
1180 	compare_cpu,
1181 	compare_size,
1182 	compare_res,
1183 	compare_time,
1184 	compare_prio,
1185 	compare_threads,
1186 	compare_iototal,
1187 	compare_ioread,
1188 	compare_iowrite,
1189 	compare_iofault,
1190 	compare_vcsw,
1191 	compare_ivcsw,
1192 	compare_jid,
1193 	NULL
1194 };
1195 
1196 /* compare_size - the comparison function for sorting by total memory usage */
1197 
1198 int
1199 compare_size(void *arg1, void *arg2)
1200 {
1201 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1202 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1203 
1204 	ORDERKEY_MEM(p1, p2);
1205 	ORDERKEY_RSSIZE(p1, p2);
1206 	ORDERKEY_PCTCPU(p1, p2);
1207 	ORDERKEY_CPTICKS(p1, p2);
1208 	ORDERKEY_STATE(p1, p2);
1209 	ORDERKEY_PRIO(p1, p2);
1210 
1211 	return (0);
1212 }
1213 
1214 /* compare_res - the comparison function for sorting by resident set size */
1215 
1216 int
1217 compare_res(void *arg1, void *arg2)
1218 {
1219 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1220 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1221 
1222 	ORDERKEY_RSSIZE(p1, p2);
1223 	ORDERKEY_MEM(p1, p2);
1224 	ORDERKEY_PCTCPU(p1, p2);
1225 	ORDERKEY_CPTICKS(p1, p2);
1226 	ORDERKEY_STATE(p1, p2);
1227 	ORDERKEY_PRIO(p1, p2);
1228 
1229 	return (0);
1230 }
1231 
1232 /* compare_time - the comparison function for sorting by total cpu time */
1233 
1234 int
1235 compare_time(void *arg1, void *arg2)
1236 {
1237 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1238 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1239 
1240 	ORDERKEY_CPTICKS(p1, p2);
1241 	ORDERKEY_PCTCPU(p1, p2);
1242 	ORDERKEY_STATE(p1, p2);
1243 	ORDERKEY_PRIO(p1, p2);
1244 	ORDERKEY_RSSIZE(p1, p2);
1245 	ORDERKEY_MEM(p1, p2);
1246 
1247 	return (0);
1248 }
1249 
1250 /* compare_prio - the comparison function for sorting by priority */
1251 
1252 int
1253 compare_prio(void *arg1, void *arg2)
1254 {
1255 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1256 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1257 
1258 	ORDERKEY_PRIO(p1, p2);
1259 	ORDERKEY_CPTICKS(p1, p2);
1260 	ORDERKEY_PCTCPU(p1, p2);
1261 	ORDERKEY_STATE(p1, p2);
1262 	ORDERKEY_RSSIZE(p1, p2);
1263 	ORDERKEY_MEM(p1, p2);
1264 
1265 	return (0);
1266 }
1267 
1268 /* compare_threads - the comparison function for sorting by threads */
1269 int
1270 compare_threads(void *arg1, void *arg2)
1271 {
1272 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1273 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1274 
1275 	ORDERKEY_THREADS(p1, p2);
1276 	ORDERKEY_PCTCPU(p1, p2);
1277 	ORDERKEY_CPTICKS(p1, p2);
1278 	ORDERKEY_STATE(p1, p2);
1279 	ORDERKEY_PRIO(p1, p2);
1280 	ORDERKEY_RSSIZE(p1, p2);
1281 	ORDERKEY_MEM(p1, p2);
1282 
1283 	return (0);
1284 }
1285 
1286 /* compare_jid - the comparison function for sorting by jid */
1287 static int
1288 compare_jid(const void *arg1, const void *arg2)
1289 {
1290 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1291 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1292 
1293 	ORDERKEY_JID(p1, p2);
1294 	ORDERKEY_PCTCPU(p1, p2);
1295 	ORDERKEY_CPTICKS(p1, p2);
1296 	ORDERKEY_STATE(p1, p2);
1297 	ORDERKEY_PRIO(p1, p2);
1298 	ORDERKEY_RSSIZE(p1, p2);
1299 	ORDERKEY_MEM(p1, p2);
1300 
1301 	return (0);
1302 }
1303 #endif /* ORDER */
1304 
1305 /* assorted comparison functions for sorting by i/o */
1306 
1307 int
1308 #ifdef ORDER
1309 compare_iototal(void *arg1, void *arg2)
1310 #else
1311 io_compare(void *arg1, void *arg2)
1312 #endif
1313 {
1314 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1315 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1316 
1317 	return (get_io_total(p2) - get_io_total(p1));
1318 }
1319 
1320 #ifdef ORDER
1321 int
1322 compare_ioread(void *arg1, void *arg2)
1323 {
1324 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1325 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1326 	long dummy, inp1, inp2;
1327 
1328 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1329 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1330 
1331 	return (inp2 - inp1);
1332 }
1333 
1334 int
1335 compare_iowrite(void *arg1, void *arg2)
1336 {
1337 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1338 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1339 	long dummy, oup1, oup2;
1340 
1341 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1342 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1343 
1344 	return (oup2 - oup1);
1345 }
1346 
1347 int
1348 compare_iofault(void *arg1, void *arg2)
1349 {
1350 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1351 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1352 	long dummy, flp1, flp2;
1353 
1354 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1355 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1356 
1357 	return (flp2 - flp1);
1358 }
1359 
1360 int
1361 compare_vcsw(void *arg1, void *arg2)
1362 {
1363 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1364 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1365 	long dummy, flp1, flp2;
1366 
1367 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1368 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1369 
1370 	return (flp2 - flp1);
1371 }
1372 
1373 int
1374 compare_ivcsw(void *arg1, void *arg2)
1375 {
1376 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1377 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1378 	long dummy, flp1, flp2;
1379 
1380 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1381 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1382 
1383 	return (flp2 - flp1);
1384 }
1385 #endif /* ORDER */
1386 
1387 /*
1388  * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1389  *		the process does not exist.
1390  *		It is EXTREMLY IMPORTANT that this function work correctly.
1391  *		If top runs setuid root (as in SVR4), then this function
1392  *		is the only thing that stands in the way of a serious
1393  *		security problem.  It validates requests for the "kill"
1394  *		and "renice" commands.
1395  */
1396 
1397 int
1398 proc_owner(int pid)
1399 {
1400 	int cnt;
1401 	struct kinfo_proc **prefp;
1402 	struct kinfo_proc *pp;
1403 
1404 	prefp = pref;
1405 	cnt = pref_len;
1406 	while (--cnt >= 0) {
1407 		pp = *prefp++;
1408 		if (pp->ki_pid == (pid_t)pid)
1409 			return ((int)pp->ki_ruid);
1410 	}
1411 	return (-1);
1412 }
1413 
1414 static int
1415 swapmode(int *retavail, int *retfree)
1416 {
1417 	int n;
1418 	int pagesize = getpagesize();
1419 	struct kvm_swap swapary[1];
1420 
1421 	*retavail = 0;
1422 	*retfree = 0;
1423 
1424 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1425 
1426 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1427 	if (n < 0 || swapary[0].ksw_total == 0)
1428 		return (0);
1429 
1430 	*retavail = CONVERT(swapary[0].ksw_total);
1431 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1432 
1433 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1434 	return (n);
1435 }
1436