xref: /freebsd/usr.bin/top/machine.c (revision 955c8cbb4960e6cf3602de144b1b9154a5092968)
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 arc_stats[7];
180 char *arcnames[] = {
181 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
182 	NULL
183 };
184 
185 int swap_stats[7];
186 char *swapnames[] = {
187 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
188 	NULL
189 };
190 
191 
192 /* these are for keeping track of the proc array */
193 
194 static int nproc;
195 static int onproc = -1;
196 static int pref_len;
197 static struct kinfo_proc *pbase;
198 static struct kinfo_proc **pref;
199 static struct kinfo_proc *previous_procs;
200 static struct kinfo_proc **previous_pref;
201 static int previous_proc_count = 0;
202 static int previous_proc_count_max = 0;
203 static int arc_enabled;
204 
205 /* total number of io operations */
206 static long total_inblock;
207 static long total_oublock;
208 static long total_majflt;
209 
210 /* these are for getting the memory statistics */
211 
212 static int pageshift;		/* log base 2 of the pagesize */
213 
214 /* define pagetok in terms of pageshift */
215 
216 #define pagetok(size) ((size) << pageshift)
217 
218 /* useful externals */
219 long percentages();
220 
221 #ifdef ORDER
222 /*
223  * Sorting orders.  The first element is the default.
224  */
225 char *ordernames[] = {
226 	"cpu", "size", "res", "time", "pri", "threads",
227 	"total", "read", "write", "fault", "vcsw", "ivcsw",
228 	"jid", "pid", NULL
229 };
230 #endif
231 
232 /* Per-cpu time states */
233 static int maxcpu;
234 static int maxid;
235 static int ncpus;
236 static u_long cpumask;
237 static long *times;
238 static long *pcpu_cp_time;
239 static long *pcpu_cp_old;
240 static long *pcpu_cp_diff;
241 static int *pcpu_cpu_states;
242 
243 static int compare_jid(const void *a, const void *b);
244 static int compare_pid(const void *a, const void *b);
245 static int compare_tid(const void *a, const void *b);
246 static const char *format_nice(const struct kinfo_proc *pp);
247 static void getsysctl(const char *name, void *ptr, size_t len);
248 static int swapmode(int *retavail, int *retfree);
249 static void update_layout(void);
250 
251 void
252 toggle_pcpustats(void)
253 {
254 
255 	if (ncpus == 1)
256 		return;
257 	update_layout();
258 }
259 
260 /* Adjust display based on ncpus and the ARC state. */
261 static void
262 update_layout(void)
263 {
264 
265 	y_mem = 3;
266 	y_arc = 4;
267 	y_swap = 4 + arc_enabled;
268 	y_idlecursor = 5 + arc_enabled;
269 	y_message = 5 + arc_enabled;
270 	y_header = 6 + arc_enabled;
271 	y_procs = 7 + arc_enabled;
272 	Header_lines = 7 + arc_enabled;
273 
274 	if (pcpu_stats) {
275 		y_mem += ncpus - 1;
276 		y_arc += ncpus - 1;
277 		y_swap += ncpus - 1;
278 		y_idlecursor += ncpus - 1;
279 		y_message += ncpus - 1;
280 		y_header += ncpus - 1;
281 		y_procs += ncpus - 1;
282 		Header_lines += ncpus - 1;
283 	}
284 }
285 
286 int
287 machine_init(struct statics *statics, char do_unames)
288 {
289 	int i, j, empty, pagesize;
290 	uint64_t arc_size;
291 	size_t size;
292 	struct passwd *pw;
293 
294 	size = sizeof(smpmode);
295 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
296 	    NULL, 0) != 0 &&
297 	    sysctlbyname("kern.smp.active", &smpmode, &size,
298 	    NULL, 0) != 0) ||
299 	    size != sizeof(smpmode))
300 		smpmode = 0;
301 
302 	size = sizeof(arc_size);
303 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
304 	    NULL, 0) == 0 && arc_size != 0)
305 		arc_enabled = 1;
306 
307 	if (do_unames) {
308 	    while ((pw = getpwent()) != NULL) {
309 		if (strlen(pw->pw_name) > namelength)
310 			namelength = strlen(pw->pw_name);
311 	    }
312 	}
313 	if (smpmode && namelength > SMPUNAMELEN)
314 		namelength = SMPUNAMELEN;
315 	else if (namelength > UPUNAMELEN)
316 		namelength = UPUNAMELEN;
317 
318 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
319 	if (kd == NULL)
320 		return (-1);
321 
322 	GETSYSCTL("kern.ccpu", ccpu);
323 
324 	/* this is used in calculating WCPU -- calculate it ahead of time */
325 	logcpu = log(loaddouble(ccpu));
326 
327 	pbase = NULL;
328 	pref = NULL;
329 	nproc = 0;
330 	onproc = -1;
331 
332 	/* get the page size and calculate pageshift from it */
333 	pagesize = getpagesize();
334 	pageshift = 0;
335 	while (pagesize > 1) {
336 		pageshift++;
337 		pagesize >>= 1;
338 	}
339 
340 	/* we only need the amount of log(2)1024 for our conversion */
341 	pageshift -= LOG1024;
342 
343 	/* fill in the statics information */
344 	statics->procstate_names = procstatenames;
345 	statics->cpustate_names = cpustatenames;
346 	statics->memory_names = memorynames;
347 	if (arc_enabled)
348 		statics->arc_names = arcnames;
349 	else
350 		statics->arc_names = NULL;
351 	statics->swap_names = swapnames;
352 #ifdef ORDER
353 	statics->order_names = ordernames;
354 #endif
355 
356 	/* Allocate state for per-CPU stats. */
357 	cpumask = 0;
358 	ncpus = 0;
359 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
360 	size = sizeof(long) * maxcpu * CPUSTATES;
361 	times = malloc(size);
362 	if (times == NULL)
363 		err(1, "malloc %zd bytes", size);
364 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
365 		err(1, "sysctlbyname kern.cp_times");
366 	pcpu_cp_time = calloc(1, size);
367 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
368 	for (i = 0; i <= maxid; i++) {
369 		empty = 1;
370 		for (j = 0; empty && j < CPUSTATES; j++) {
371 			if (times[i * CPUSTATES + j] != 0)
372 				empty = 0;
373 		}
374 		if (!empty) {
375 			cpumask |= (1ul << i);
376 			ncpus++;
377 		}
378 	}
379 	size = sizeof(long) * ncpus * CPUSTATES;
380 	pcpu_cp_old = calloc(1, size);
381 	pcpu_cp_diff = calloc(1, size);
382 	pcpu_cpu_states = calloc(1, size);
383 	statics->ncpus = ncpus;
384 
385 	update_layout();
386 
387 	/* all done! */
388 	return (0);
389 }
390 
391 char *
392 format_header(char *uname_field)
393 {
394 	static char Header[128];
395 	const char *prehead;
396 
397 	switch (displaymode) {
398 	case DISP_CPU:
399 		/*
400 		 * The logic of picking the right header format seems reverse
401 		 * here because we only want to display a THR column when
402 		 * "thread mode" is off (and threads are not listed as
403 		 * separate lines).
404 		 */
405 		prehead = smpmode ?
406 		    (ps.thread ? smp_header : smp_header_thr) :
407 		    (ps.thread ? up_header : up_header_thr);
408 		snprintf(Header, sizeof(Header), prehead,
409 		    ps.jail ? " JID" : "",
410 		    namelength, namelength, uname_field,
411 		    ps.wcpu ? "WCPU" : "CPU");
412 		break;
413 	case DISP_IO:
414 		prehead = io_header;
415 		snprintf(Header, sizeof(Header), prehead,
416 		    ps.jail ? " JID" : "",
417 		    namelength, namelength, uname_field);
418 		break;
419 	}
420 	cmdlengthdelta = strlen(Header) - 7;
421 	return (Header);
422 }
423 
424 static int swappgsin = -1;
425 static int swappgsout = -1;
426 extern struct timeval timeout;
427 
428 
429 void
430 get_system_info(struct system_info *si)
431 {
432 	long total;
433 	struct loadavg sysload;
434 	int mib[2];
435 	struct timeval boottime;
436 	uint64_t arc_stat, arc_stat2;
437 	int i, j;
438 	size_t size;
439 
440 	/* get the CPU stats */
441 	size = (maxid + 1) * CPUSTATES * sizeof(long);
442 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
443 		err(1, "sysctlbyname kern.cp_times");
444 	GETSYSCTL("kern.cp_time", cp_time);
445 	GETSYSCTL("vm.loadavg", sysload);
446 	GETSYSCTL("kern.lastpid", lastpid);
447 
448 	/* convert load averages to doubles */
449 	for (i = 0; i < 3; i++)
450 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
451 
452 	/* convert cp_time counts to percentages */
453 	for (i = j = 0; i <= maxid; i++) {
454 		if ((cpumask & (1ul << i)) == 0)
455 			continue;
456 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
457 		    &pcpu_cp_time[j * CPUSTATES],
458 		    &pcpu_cp_old[j * CPUSTATES],
459 		    &pcpu_cp_diff[j * CPUSTATES]);
460 		j++;
461 	}
462 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
463 
464 	/* sum memory & swap statistics */
465 	{
466 		static unsigned int swap_delay = 0;
467 		static int swapavail = 0;
468 		static int swapfree = 0;
469 		static long bufspace = 0;
470 		static int nspgsin, nspgsout;
471 
472 		GETSYSCTL("vfs.bufspace", bufspace);
473 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
474 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
475 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[2]);
476 		GETSYSCTL("vm.stats.vm.v_cache_count", memory_stats[3]);
477 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
478 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
479 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
480 		/* convert memory stats to Kbytes */
481 		memory_stats[0] = pagetok(memory_stats[0]);
482 		memory_stats[1] = pagetok(memory_stats[1]);
483 		memory_stats[2] = pagetok(memory_stats[2]);
484 		memory_stats[3] = pagetok(memory_stats[3]);
485 		memory_stats[4] = bufspace / 1024;
486 		memory_stats[5] = pagetok(memory_stats[5]);
487 		memory_stats[6] = -1;
488 
489 		/* first interval */
490 		if (swappgsin < 0) {
491 			swap_stats[4] = 0;
492 			swap_stats[5] = 0;
493 		}
494 
495 		/* compute differences between old and new swap statistic */
496 		else {
497 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
498 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
499 		}
500 
501 		swappgsin = nspgsin;
502 		swappgsout = nspgsout;
503 
504 		/* call CPU heavy swapmode() only for changes */
505 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
506 			swap_stats[3] = swapmode(&swapavail, &swapfree);
507 			swap_stats[0] = swapavail;
508 			swap_stats[1] = swapavail - swapfree;
509 			swap_stats[2] = swapfree;
510 		}
511 		swap_delay = 1;
512 		swap_stats[6] = -1;
513 	}
514 
515 	if (arc_enabled) {
516 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
517 		arc_stats[0] = arc_stat >> 10;
518 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
519 		arc_stats[1] = arc_stat >> 10;
520 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
521 		arc_stats[2] = arc_stat >> 10;
522 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
523 		arc_stats[3] = arc_stat >> 10;
524 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
525 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
526 		arc_stats[4] = arc_stat + arc_stat2 >> 10;
527 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
528 		arc_stats[5] = arc_stat >> 10;
529 		si->arc = arc_stats;
530 	}
531 
532 	/* set arrays and strings */
533 	if (pcpu_stats) {
534 		si->cpustates = pcpu_cpu_states;
535 		si->ncpus = ncpus;
536 	} else {
537 		si->cpustates = cpu_states;
538 		si->ncpus = 1;
539 	}
540 	si->memory = memory_stats;
541 	si->swap = swap_stats;
542 
543 
544 	if (lastpid > 0) {
545 		si->last_pid = lastpid;
546 	} else {
547 		si->last_pid = -1;
548 	}
549 
550 	/*
551 	 * Print how long system has been up.
552 	 * (Found by looking getting "boottime" from the kernel)
553 	 */
554 	mib[0] = CTL_KERN;
555 	mib[1] = KERN_BOOTTIME;
556 	size = sizeof(boottime);
557 	if (sysctl(mib, 2, &boottime, &size, NULL, 0) != -1 &&
558 	    boottime.tv_sec != 0) {
559 		si->boottime = boottime;
560 	} else {
561 		si->boottime.tv_sec = -1;
562 	}
563 }
564 
565 #define NOPROC	((void *)-1)
566 
567 /*
568  * We need to compare data from the old process entry with the new
569  * process entry.
570  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
571  * structure to cache the mapping.  We also use a negative cache pointer
572  * of NOPROC to avoid duplicate lookups.
573  * XXX: this could be done when the actual processes are fetched, we do
574  * it here out of laziness.
575  */
576 const struct kinfo_proc *
577 get_old_proc(struct kinfo_proc *pp)
578 {
579 	struct kinfo_proc **oldpp, *oldp;
580 
581 	/*
582 	 * If this is the first fetch of the kinfo_procs then we don't have
583 	 * any previous entries.
584 	 */
585 	if (previous_proc_count == 0)
586 		return (NULL);
587 	/* negative cache? */
588 	if (pp->ki_udata == NOPROC)
589 		return (NULL);
590 	/* cached? */
591 	if (pp->ki_udata != NULL)
592 		return (pp->ki_udata);
593 	/*
594 	 * Not cached,
595 	 * 1) look up based on pid.
596 	 * 2) compare process start.
597 	 * If we fail here, then setup a negative cache entry, otherwise
598 	 * cache it.
599 	 */
600 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
601 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
602 	if (oldpp == NULL) {
603 		pp->ki_udata = NOPROC;
604 		return (NULL);
605 	}
606 	oldp = *oldpp;
607 	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
608 		pp->ki_udata = NOPROC;
609 		return (NULL);
610 	}
611 	pp->ki_udata = oldp;
612 	return (oldp);
613 }
614 
615 /*
616  * Return the total amount of IO done in blocks in/out and faults.
617  * store the values individually in the pointers passed in.
618  */
619 long
620 get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
621     long *vcsw, long *ivcsw)
622 {
623 	const struct kinfo_proc *oldp;
624 	static struct kinfo_proc dummy;
625 	long ret;
626 
627 	oldp = get_old_proc(pp);
628 	if (oldp == NULL) {
629 		bzero(&dummy, sizeof(dummy));
630 		oldp = &dummy;
631 	}
632 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
633 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
634 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
635 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
636 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
637 	ret =
638 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
639 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
640 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
641 	return (ret);
642 }
643 
644 /*
645  * Return the total number of block in/out and faults by a process.
646  */
647 long
648 get_io_total(struct kinfo_proc *pp)
649 {
650 	long dummy;
651 
652 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
653 }
654 
655 static struct handle handle;
656 
657 caddr_t
658 get_process_info(struct system_info *si, struct process_select *sel,
659     int (*compare)(const void *, const void *))
660 {
661 	int i;
662 	int total_procs;
663 	long p_io;
664 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
665 	int active_procs;
666 	struct kinfo_proc **prefp;
667 	struct kinfo_proc *pp;
668 
669 	/* these are copied out of sel for speed */
670 	int show_idle;
671 	int show_self;
672 	int show_system;
673 	int show_uid;
674 	int show_command;
675 	int show_kidle;
676 
677 	/*
678 	 * Save the previous process info.
679 	 */
680 	if (previous_proc_count_max < nproc) {
681 		free(previous_procs);
682 		previous_procs = malloc(nproc * sizeof(*previous_procs));
683 		free(previous_pref);
684 		previous_pref = malloc(nproc * sizeof(*previous_pref));
685 		if (previous_procs == NULL || previous_pref == NULL) {
686 			(void) fprintf(stderr, "top: Out of memory.\n");
687 			quit(23);
688 		}
689 		previous_proc_count_max = nproc;
690 	}
691 	if (nproc) {
692 		for (i = 0; i < nproc; i++)
693 			previous_pref[i] = &previous_procs[i];
694 		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
695 		qsort(previous_pref, nproc, sizeof(*previous_pref),
696 		    ps.thread ? compare_tid : compare_pid);
697 	}
698 	previous_proc_count = nproc;
699 
700 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
701 	    0, &nproc);
702 	if (nproc > onproc)
703 		pref = realloc(pref, sizeof(*pref) * (onproc = nproc));
704 	if (pref == NULL || pbase == NULL) {
705 		(void) fprintf(stderr, "top: Out of memory.\n");
706 		quit(23);
707 	}
708 	/* get a pointer to the states summary array */
709 	si->procstates = process_states;
710 
711 	/* set up flags which define what we are going to select */
712 	show_idle = sel->idle;
713 	show_self = sel->self == -1;
714 	show_system = sel->system;
715 	show_uid = sel->uid != -1;
716 	show_command = sel->command != NULL;
717 	show_kidle = sel->kidle;
718 
719 	/* count up process states and get pointers to interesting procs */
720 	total_procs = 0;
721 	active_procs = 0;
722 	total_inblock = 0;
723 	total_oublock = 0;
724 	total_majflt = 0;
725 	memset((char *)process_states, 0, sizeof(process_states));
726 	prefp = pref;
727 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
728 
729 		if (pp->ki_stat == 0)
730 			/* not in use */
731 			continue;
732 
733 		if (!show_self && pp->ki_pid == sel->self)
734 			/* skip self */
735 			continue;
736 
737 		if (!show_system && (pp->ki_flag & P_SYSTEM))
738 			/* skip system process */
739 			continue;
740 
741 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
742 		    &p_vcsw, &p_ivcsw);
743 		total_inblock += p_inblock;
744 		total_oublock += p_oublock;
745 		total_majflt += p_majflt;
746 		total_procs++;
747 		process_states[pp->ki_stat]++;
748 
749 		if (pp->ki_stat == SZOMB)
750 			/* skip zombies */
751 			continue;
752 
753 		if (!show_kidle && pp->ki_tdflags & TDF_IDLETD)
754 			/* skip kernel idle process */
755 			continue;
756 
757 		if (displaymode == DISP_CPU && !show_idle &&
758 		    (pp->ki_pctcpu == 0 ||
759 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
760 			/* skip idle or non-running processes */
761 			continue;
762 
763 		if (displaymode == DISP_IO && !show_idle && p_io == 0)
764 			/* skip processes that aren't doing I/O */
765 			continue;
766 
767 		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
768 			/* skip proc. that don't belong to the selected UID */
769 			continue;
770 
771 		*prefp++ = pp;
772 		active_procs++;
773 	}
774 
775 	/* if requested, sort the "interesting" processes */
776 	if (compare != NULL)
777 		qsort(pref, active_procs, sizeof(*pref), compare);
778 
779 	/* remember active and total counts */
780 	si->p_total = total_procs;
781 	si->p_active = pref_len = active_procs;
782 
783 	/* pass back a handle */
784 	handle.next_proc = pref;
785 	handle.remaining = active_procs;
786 	return ((caddr_t)&handle);
787 }
788 
789 static char fmt[512];	/* static area where result is built */
790 
791 char *
792 format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
793 {
794 	struct kinfo_proc *pp;
795 	const struct kinfo_proc *oldp;
796 	long cputime;
797 	double pct;
798 	struct handle *hp;
799 	char status[16];
800 	int state;
801 	struct rusage ru, *rup;
802 	long p_tot, s_tot;
803 	char *proc_fmt, thr_buf[6], jid_buf[6];
804 	char *cmdbuf = NULL;
805 	char **args;
806 	const int cmdlen = 128;
807 
808 	/* find and remember the next proc structure */
809 	hp = (struct handle *)handle;
810 	pp = *(hp->next_proc++);
811 	hp->remaining--;
812 
813 	/* get the process's command name */
814 	if ((pp->ki_flag & P_INMEM) == 0) {
815 		/*
816 		 * Print swapped processes as <pname>
817 		 */
818 		size_t len;
819 
820 		len = strlen(pp->ki_comm);
821 		if (len > sizeof(pp->ki_comm) - 3)
822 			len = sizeof(pp->ki_comm) - 3;
823 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
824 		pp->ki_comm[0] = '<';
825 		pp->ki_comm[len + 1] = '>';
826 		pp->ki_comm[len + 2] = '\0';
827 	}
828 
829 	/*
830 	 * Convert the process's runtime from microseconds to seconds.  This
831 	 * time includes the interrupt time although that is not wanted here.
832 	 * ps(1) is similarly sloppy.
833 	 */
834 	cputime = (pp->ki_runtime + 500000) / 1000000;
835 
836 	/* calculate the base for cpu percentages */
837 	pct = pctdouble(pp->ki_pctcpu);
838 
839 	/* generate "STATE" field */
840 	switch (state = pp->ki_stat) {
841 	case SRUN:
842 		if (smpmode && pp->ki_oncpu != 0xff)
843 			sprintf(status, "CPU%d", pp->ki_oncpu);
844 		else
845 			strcpy(status, "RUN");
846 		break;
847 	case SLOCK:
848 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
849 			sprintf(status, "*%.6s", pp->ki_lockname);
850 			break;
851 		}
852 		/* fall through */
853 	case SSLEEP:
854 		if (pp->ki_wmesg != NULL) {
855 			sprintf(status, "%.6s", pp->ki_wmesg);
856 			break;
857 		}
858 		/* FALLTHROUGH */
859 	default:
860 
861 		if (state >= 0 &&
862 		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
863 			sprintf(status, "%.6s", state_abbrev[state]);
864 		else
865 			sprintf(status, "?%5d", state);
866 		break;
867 	}
868 
869 	cmdbuf = (char *)malloc(cmdlen + 1);
870 	if (cmdbuf == NULL) {
871 		warn("malloc(%d)", cmdlen + 1);
872 		return NULL;
873 	}
874 
875 	if (!(flags & FMT_SHOWARGS)) {
876 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
877 		    pp->ki_tdname[0]) {
878 			snprintf(cmdbuf, cmdlen, "%s{%s}", pp->ki_comm,
879 			    pp->ki_tdname);
880 		} else {
881 			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
882 		}
883 	} else {
884 		if (pp->ki_flag & P_SYSTEM ||
885 		    pp->ki_args == NULL ||
886 		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
887 		    !(*args)) {
888 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
889 		    	    pp->ki_tdname[0]) {
890 				snprintf(cmdbuf, cmdlen,
891 				    "[%s{%s}]", pp->ki_comm, pp->ki_tdname);
892 			} else {
893 				snprintf(cmdbuf, cmdlen,
894 				    "[%s]", pp->ki_comm);
895 			}
896 		} else {
897 			char *src, *dst, *argbuf;
898 			char *cmd;
899 			size_t argbuflen;
900 			size_t len;
901 
902 			argbuflen = cmdlen * 4;
903 			argbuf = (char *)malloc(argbuflen + 1);
904 			if (argbuf == NULL) {
905 				warn("malloc(%d)", argbuflen + 1);
906 				free(cmdbuf);
907 				return NULL;
908 			}
909 
910 			dst = argbuf;
911 
912 			/* Extract cmd name from argv */
913 			cmd = strrchr(*args, '/');
914 			if (cmd == NULL)
915 				cmd = *args;
916 			else
917 				cmd++;
918 
919 			for (; (src = *args++) != NULL; ) {
920 				if (*src == '\0')
921 					continue;
922 				len = (argbuflen - (dst - argbuf) - 1) / 4;
923 				strvisx(dst, src,
924 				    strlen(src) < len ? strlen(src) : len,
925 				    VIS_NL | VIS_CSTYLE);
926 				while (*dst != '\0')
927 					dst++;
928 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
929 					*dst++ = ' '; /* add delimiting space */
930 			}
931 			if (dst != argbuf && dst[-1] == ' ')
932 				dst--;
933 			*dst = '\0';
934 
935 			if (strcmp(cmd, pp->ki_comm) != 0) {
936 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
937 				    pp->ki_tdname[0])
938 					snprintf(cmdbuf, cmdlen,
939 					    "%s (%s){%s}", argbuf, pp->ki_comm,
940 					    pp->ki_tdname);
941 				else
942 					snprintf(cmdbuf, cmdlen,
943 					    "%s (%s)", argbuf, pp->ki_comm);
944 			} else {
945 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
946 				    pp->ki_tdname[0])
947 					snprintf(cmdbuf, cmdlen,
948 					    "%s{%s}", argbuf, pp->ki_tdname);
949 				else
950 					strlcpy(cmdbuf, argbuf, cmdlen);
951 			}
952 			free(argbuf);
953 		}
954 	}
955 
956 	if (ps.jail == 0)
957 		jid_buf[0] = '\0';
958 	else
959 		snprintf(jid_buf, sizeof(jid_buf), " %*d",
960 		    sizeof(jid_buf) - 3, pp->ki_jid);
961 
962 	if (displaymode == DISP_IO) {
963 		oldp = get_old_proc(pp);
964 		if (oldp != NULL) {
965 			ru.ru_inblock = RU(pp)->ru_inblock -
966 			    RU(oldp)->ru_inblock;
967 			ru.ru_oublock = RU(pp)->ru_oublock -
968 			    RU(oldp)->ru_oublock;
969 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
970 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
971 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
972 			rup = &ru;
973 		} else {
974 			rup = RU(pp);
975 		}
976 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
977 		s_tot = total_inblock + total_oublock + total_majflt;
978 
979 		snprintf(fmt, sizeof(fmt), io_Proc_format,
980 		    pp->ki_pid,
981 		    jid_buf,
982 		    namelength, namelength, (*get_userid)(pp->ki_ruid),
983 		    rup->ru_nvcsw,
984 		    rup->ru_nivcsw,
985 		    rup->ru_inblock,
986 		    rup->ru_oublock,
987 		    rup->ru_majflt,
988 		    p_tot,
989 		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
990 		    screen_width > cmdlengthdelta ?
991 		    screen_width - cmdlengthdelta : 0,
992 		    printable(cmdbuf));
993 
994 		free(cmdbuf);
995 
996 		return (fmt);
997 	}
998 
999 	/* format this entry */
1000 	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1001 	if (ps.thread != 0)
1002 		thr_buf[0] = '\0';
1003 	else
1004 		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1005 		    sizeof(thr_buf) - 2, pp->ki_numthreads);
1006 
1007 	snprintf(fmt, sizeof(fmt), proc_fmt,
1008 	    pp->ki_pid,
1009 	    jid_buf,
1010 	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1011 	    thr_buf,
1012 	    pp->ki_pri.pri_level - PZERO,
1013 	    format_nice(pp),
1014 	    format_k2(PROCSIZE(pp)),
1015 	    format_k2(pagetok(pp->ki_rssize)),
1016 	    status,
1017 	    smpmode ? pp->ki_lastcpu : 0,
1018 	    format_time(cputime),
1019 	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1020 	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1021 	    printable(cmdbuf));
1022 
1023 	free(cmdbuf);
1024 
1025 	/* return the result */
1026 	return (fmt);
1027 }
1028 
1029 static void
1030 getsysctl(const char *name, void *ptr, size_t len)
1031 {
1032 	size_t nlen = len;
1033 
1034 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1035 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1036 		    strerror(errno));
1037 		quit(23);
1038 	}
1039 	if (nlen != len) {
1040 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1041 		    name, (unsigned long)len, (unsigned long)nlen);
1042 		quit(23);
1043 	}
1044 }
1045 
1046 static const char *
1047 format_nice(const struct kinfo_proc *pp)
1048 {
1049 	const char *fifo, *kthread;
1050 	int rtpri;
1051 	static char nicebuf[4 + 1];
1052 
1053 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1054 	kthread = (pp->ki_flag & P_KTHREAD) ? "k" : "";
1055 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1056 	case PRI_ITHD:
1057 		return ("-");
1058 	case PRI_REALTIME:
1059 		/*
1060 		 * XXX: the kernel doesn't tell us the original rtprio and
1061 		 * doesn't really know what it was, so to recover it we
1062 		 * must be more chummy with the implementation than the
1063 		 * implementation is with itself.  pri_user gives a
1064 		 * constant "base" priority, but is only initialized
1065 		 * properly for user threads.  pri_native gives what the
1066 		 * kernel calls the "base" priority, but it isn't constant
1067 		 * since it is changed by priority propagation.  pri_native
1068 		 * also isn't properly initialized for all threads, but it
1069 		 * is properly initialized for kernel realtime and idletime
1070 		 * threads.  Thus we use pri_user for the base priority of
1071 		 * user threads (it is always correct) and pri_native for
1072 		 * the base priority of kernel realtime and idletime threads
1073 		 * (there is nothing better, and it is usually correct).
1074 		 *
1075 		 * The field width and thus the buffer are too small for
1076 		 * values like "kr31F", but such values shouldn't occur,
1077 		 * and if they do then the tailing "F" is not displayed.
1078 		 */
1079 		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1080 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1081 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1082 		    kthread, rtpri, fifo);
1083 		break;
1084 	case PRI_TIMESHARE:
1085 		if (pp->ki_flag & P_KTHREAD)
1086 			return ("-");
1087 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1088 		break;
1089 	case PRI_IDLE:
1090 		/* XXX: as above. */
1091 		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1092 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1093 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1094 		    kthread, rtpri, fifo);
1095 		break;
1096 	default:
1097 		return ("?");
1098 	}
1099 	return (nicebuf);
1100 }
1101 
1102 /* comparison routines for qsort */
1103 
1104 static int
1105 compare_pid(const void *p1, const void *p2)
1106 {
1107 	const struct kinfo_proc * const *pp1 = p1;
1108 	const struct kinfo_proc * const *pp2 = p2;
1109 
1110 	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1111 		abort();
1112 
1113 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1114 }
1115 
1116 static int
1117 compare_tid(const void *p1, const void *p2)
1118 {
1119 	const struct kinfo_proc * const *pp1 = p1;
1120 	const struct kinfo_proc * const *pp2 = p2;
1121 
1122 	if ((*pp2)->ki_tid < 0 || (*pp1)->ki_tid < 0)
1123 		abort();
1124 
1125 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1126 }
1127 
1128 /*
1129  *  proc_compare - comparison function for "qsort"
1130  *	Compares the resource consumption of two processes using five
1131  *	distinct keys.  The keys (in descending order of importance) are:
1132  *	percent cpu, cpu ticks, state, resident set size, total virtual
1133  *	memory usage.  The process states are ordered as follows (from least
1134  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1135  *	array declaration below maps a process state index into a number
1136  *	that reflects this ordering.
1137  */
1138 
1139 static int sorted_state[] = {
1140 	0,	/* not used		*/
1141 	3,	/* sleep		*/
1142 	1,	/* ABANDONED (WAIT)	*/
1143 	6,	/* run			*/
1144 	5,	/* start		*/
1145 	2,	/* zombie		*/
1146 	4	/* stop			*/
1147 };
1148 
1149 
1150 #define ORDERKEY_PCTCPU(a, b) do { \
1151 	long diff; \
1152 	if (ps.wcpu) \
1153 		diff = floor(1.0E6 * weighted_cpu(pctdouble((b)->ki_pctcpu), \
1154 		    (b))) - \
1155 		    floor(1.0E6 * weighted_cpu(pctdouble((a)->ki_pctcpu), \
1156 		    (a))); \
1157 	else \
1158 		diff = (long)(b)->ki_pctcpu - (long)(a)->ki_pctcpu; \
1159 	if (diff != 0) \
1160 		return (diff > 0 ? 1 : -1); \
1161 } while (0)
1162 
1163 #define ORDERKEY_CPTICKS(a, b) do { \
1164 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1165 	if (diff != 0) \
1166 		return (diff > 0 ? 1 : -1); \
1167 } while (0)
1168 
1169 #define ORDERKEY_STATE(a, b) do { \
1170 	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1171 	if (diff != 0) \
1172 		return (diff > 0 ? 1 : -1); \
1173 } while (0)
1174 
1175 #define ORDERKEY_PRIO(a, b) do { \
1176 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1177 	if (diff != 0) \
1178 		return (diff > 0 ? 1 : -1); \
1179 } while (0)
1180 
1181 #define	ORDERKEY_THREADS(a, b) do { \
1182 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1183 	if (diff != 0) \
1184 		return (diff > 0 ? 1 : -1); \
1185 } while (0)
1186 
1187 #define ORDERKEY_RSSIZE(a, b) do { \
1188 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1189 	if (diff != 0) \
1190 		return (diff > 0 ? 1 : -1); \
1191 } while (0)
1192 
1193 #define ORDERKEY_MEM(a, b) do { \
1194 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1195 	if (diff != 0) \
1196 		return (diff > 0 ? 1 : -1); \
1197 } while (0)
1198 
1199 #define ORDERKEY_JID(a, b) do { \
1200 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1201 	if (diff != 0) \
1202 		return (diff > 0 ? 1 : -1); \
1203 } while (0)
1204 
1205 /* compare_cpu - the comparison function for sorting by cpu percentage */
1206 
1207 int
1208 #ifdef ORDER
1209 compare_cpu(void *arg1, void *arg2)
1210 #else
1211 proc_compare(void *arg1, void *arg2)
1212 #endif
1213 {
1214 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1215 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1216 
1217 	ORDERKEY_PCTCPU(p1, p2);
1218 	ORDERKEY_CPTICKS(p1, p2);
1219 	ORDERKEY_STATE(p1, p2);
1220 	ORDERKEY_PRIO(p1, p2);
1221 	ORDERKEY_RSSIZE(p1, p2);
1222 	ORDERKEY_MEM(p1, p2);
1223 
1224 	return (0);
1225 }
1226 
1227 #ifdef ORDER
1228 /* "cpu" compare routines */
1229 int compare_size(), compare_res(), compare_time(), compare_prio(),
1230     compare_threads();
1231 
1232 /*
1233  * "io" compare routines.  Context switches aren't i/o, but are displayed
1234  * on the "io" display.
1235  */
1236 int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1237     compare_vcsw(), compare_ivcsw();
1238 
1239 int (*compares[])() = {
1240 	compare_cpu,
1241 	compare_size,
1242 	compare_res,
1243 	compare_time,
1244 	compare_prio,
1245 	compare_threads,
1246 	compare_iototal,
1247 	compare_ioread,
1248 	compare_iowrite,
1249 	compare_iofault,
1250 	compare_vcsw,
1251 	compare_ivcsw,
1252 	compare_jid,
1253 	NULL
1254 };
1255 
1256 /* compare_size - the comparison function for sorting by total memory usage */
1257 
1258 int
1259 compare_size(void *arg1, void *arg2)
1260 {
1261 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1262 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1263 
1264 	ORDERKEY_MEM(p1, p2);
1265 	ORDERKEY_RSSIZE(p1, p2);
1266 	ORDERKEY_PCTCPU(p1, p2);
1267 	ORDERKEY_CPTICKS(p1, p2);
1268 	ORDERKEY_STATE(p1, p2);
1269 	ORDERKEY_PRIO(p1, p2);
1270 
1271 	return (0);
1272 }
1273 
1274 /* compare_res - the comparison function for sorting by resident set size */
1275 
1276 int
1277 compare_res(void *arg1, void *arg2)
1278 {
1279 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1280 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1281 
1282 	ORDERKEY_RSSIZE(p1, p2);
1283 	ORDERKEY_MEM(p1, p2);
1284 	ORDERKEY_PCTCPU(p1, p2);
1285 	ORDERKEY_CPTICKS(p1, p2);
1286 	ORDERKEY_STATE(p1, p2);
1287 	ORDERKEY_PRIO(p1, p2);
1288 
1289 	return (0);
1290 }
1291 
1292 /* compare_time - the comparison function for sorting by total cpu time */
1293 
1294 int
1295 compare_time(void *arg1, void *arg2)
1296 {
1297 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1298 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1299 
1300 	ORDERKEY_CPTICKS(p1, p2);
1301 	ORDERKEY_PCTCPU(p1, p2);
1302 	ORDERKEY_STATE(p1, p2);
1303 	ORDERKEY_PRIO(p1, p2);
1304 	ORDERKEY_RSSIZE(p1, p2);
1305 	ORDERKEY_MEM(p1, p2);
1306 
1307 	return (0);
1308 }
1309 
1310 /* compare_prio - the comparison function for sorting by priority */
1311 
1312 int
1313 compare_prio(void *arg1, void *arg2)
1314 {
1315 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1316 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1317 
1318 	ORDERKEY_PRIO(p1, p2);
1319 	ORDERKEY_CPTICKS(p1, p2);
1320 	ORDERKEY_PCTCPU(p1, p2);
1321 	ORDERKEY_STATE(p1, p2);
1322 	ORDERKEY_RSSIZE(p1, p2);
1323 	ORDERKEY_MEM(p1, p2);
1324 
1325 	return (0);
1326 }
1327 
1328 /* compare_threads - the comparison function for sorting by threads */
1329 int
1330 compare_threads(void *arg1, void *arg2)
1331 {
1332 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1333 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1334 
1335 	ORDERKEY_THREADS(p1, p2);
1336 	ORDERKEY_PCTCPU(p1, p2);
1337 	ORDERKEY_CPTICKS(p1, p2);
1338 	ORDERKEY_STATE(p1, p2);
1339 	ORDERKEY_PRIO(p1, p2);
1340 	ORDERKEY_RSSIZE(p1, p2);
1341 	ORDERKEY_MEM(p1, p2);
1342 
1343 	return (0);
1344 }
1345 
1346 /* compare_jid - the comparison function for sorting by jid */
1347 static int
1348 compare_jid(const void *arg1, const void *arg2)
1349 {
1350 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1351 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1352 
1353 	ORDERKEY_JID(p1, p2);
1354 	ORDERKEY_PCTCPU(p1, p2);
1355 	ORDERKEY_CPTICKS(p1, p2);
1356 	ORDERKEY_STATE(p1, p2);
1357 	ORDERKEY_PRIO(p1, p2);
1358 	ORDERKEY_RSSIZE(p1, p2);
1359 	ORDERKEY_MEM(p1, p2);
1360 
1361 	return (0);
1362 }
1363 #endif /* ORDER */
1364 
1365 /* assorted comparison functions for sorting by i/o */
1366 
1367 int
1368 #ifdef ORDER
1369 compare_iototal(void *arg1, void *arg2)
1370 #else
1371 io_compare(void *arg1, void *arg2)
1372 #endif
1373 {
1374 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1375 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1376 
1377 	return (get_io_total(p2) - get_io_total(p1));
1378 }
1379 
1380 #ifdef ORDER
1381 int
1382 compare_ioread(void *arg1, void *arg2)
1383 {
1384 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1385 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1386 	long dummy, inp1, inp2;
1387 
1388 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1389 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1390 
1391 	return (inp2 - inp1);
1392 }
1393 
1394 int
1395 compare_iowrite(void *arg1, void *arg2)
1396 {
1397 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1398 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1399 	long dummy, oup1, oup2;
1400 
1401 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1402 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1403 
1404 	return (oup2 - oup1);
1405 }
1406 
1407 int
1408 compare_iofault(void *arg1, void *arg2)
1409 {
1410 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1411 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1412 	long dummy, flp1, flp2;
1413 
1414 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1415 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1416 
1417 	return (flp2 - flp1);
1418 }
1419 
1420 int
1421 compare_vcsw(void *arg1, void *arg2)
1422 {
1423 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1424 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1425 	long dummy, flp1, flp2;
1426 
1427 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1428 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1429 
1430 	return (flp2 - flp1);
1431 }
1432 
1433 int
1434 compare_ivcsw(void *arg1, void *arg2)
1435 {
1436 	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1437 	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1438 	long dummy, flp1, flp2;
1439 
1440 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1441 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1442 
1443 	return (flp2 - flp1);
1444 }
1445 #endif /* ORDER */
1446 
1447 /*
1448  * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1449  *		the process does not exist.
1450  *		It is EXTREMELY IMPORTANT that this function work correctly.
1451  *		If top runs setuid root (as in SVR4), then this function
1452  *		is the only thing that stands in the way of a serious
1453  *		security problem.  It validates requests for the "kill"
1454  *		and "renice" commands.
1455  */
1456 
1457 int
1458 proc_owner(int pid)
1459 {
1460 	int cnt;
1461 	struct kinfo_proc **prefp;
1462 	struct kinfo_proc *pp;
1463 
1464 	prefp = pref;
1465 	cnt = pref_len;
1466 	while (--cnt >= 0) {
1467 		pp = *prefp++;
1468 		if (pp->ki_pid == (pid_t)pid)
1469 			return ((int)pp->ki_ruid);
1470 	}
1471 	return (-1);
1472 }
1473 
1474 static int
1475 swapmode(int *retavail, int *retfree)
1476 {
1477 	int n;
1478 	int pagesize = getpagesize();
1479 	struct kvm_swap swapary[1];
1480 
1481 	*retavail = 0;
1482 	*retfree = 0;
1483 
1484 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1485 
1486 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1487 	if (n < 0 || swapary[0].ksw_total == 0)
1488 		return (0);
1489 
1490 	*retavail = CONVERT(swapary[0].ksw_total);
1491 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1492 
1493 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1494 	return (n);
1495 }
1496