machine.c revision 178071
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: head/usr.bin/top/machine.c 178071 2008-04-10 16:17:54Z ru $
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 <kvm.h>
39#include <math.h>
40#include <nlist.h>
41#include <paths.h>
42#include <pwd.h>
43#include <stdio.h>
44#include <stdlib.h>
45#include <string.h>
46#include <strings.h>
47#include <unistd.h>
48#include <vis.h>
49
50#include "top.h"
51#include "machine.h"
52#include "screen.h"
53#include "utils.h"
54#include "layout.h"
55
56#define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
57#define	SMPUNAMELEN	13
58#define	UPUNAMELEN	15
59
60extern struct process_select ps;
61extern char* printable(char *);
62static int smpmode;
63enum displaymodes displaymode;
64#ifdef TOP_USERNAME_LEN
65static int namelength = TOP_USERNAME_LEN;
66#else
67static int namelength = 8;
68#endif
69static int cmdlengthdelta;
70
71/* Prototypes for top internals */
72void quit(int);
73
74/* get_process_info passes back a handle.  This is what it looks like: */
75
76struct handle {
77	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
78	int remaining;			/* number of pointers remaining */
79};
80
81/* declarations for load_avg */
82#include "loadavg.h"
83
84/* define what weighted cpu is.  */
85#define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
86			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
87
88/* what we consider to be process size: */
89#define PROCSIZE(pp) ((pp)->ki_size / 1024)
90
91#define RU(pp)	(&(pp)->ki_rusage)
92#define RUTOT(pp) \
93	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
94
95
96/* definitions for indices in the nlist array */
97
98/*
99 *  These definitions control the format of the per-process area
100 */
101
102static char io_header[] =
103    "  PID%s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
104
105#define io_Proc_format \
106    "%5d%s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s"
107
108static char smp_header_thr[] =
109    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE  C   TIME %6s COMMAND";
110static char smp_header[] =
111    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE  C   TIME %6s COMMAND";
112
113#define smp_Proc_format \
114    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s %1x%7s %5.2f%% %.*s"
115
116static char up_header_thr[] =
117    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
118static char up_header[] =
119    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
120
121#define up_Proc_format \
122    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s%.0d%7s %5.2f%% %.*s"
123
124
125/* process state names for the "STATE" column of the display */
126/* the extra nulls in the string "run" are for adding a slash and
127   the processor number when needed */
128
129char *state_abbrev[] = {
130	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
131};
132
133
134static kvm_t *kd;
135
136/* values that we stash away in _init and use in later routines */
137
138static double logcpu;
139
140/* these are retrieved from the kernel in _init */
141
142static load_avg  ccpu;
143
144/* these are used in the get_ functions */
145
146static int lastpid;
147
148/* these are for calculating cpu state percentages */
149
150static long cp_time[CPUSTATES];
151static long cp_old[CPUSTATES];
152static long cp_diff[CPUSTATES];
153
154/* these are for detailing the process states */
155
156int process_states[8];
157char *procstatenames[] = {
158	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
159	" zombie, ", " waiting, ", " lock, ",
160	NULL
161};
162
163/* these are for detailing the cpu states */
164
165int cpu_states[CPUSTATES];
166char *cpustatenames[] = {
167	"user", "nice", "system", "interrupt", "idle", NULL
168};
169
170/* these are for detailing the memory statistics */
171
172int memory_stats[7];
173char *memorynames[] = {
174	"K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ",
175	"K Free", NULL
176};
177
178int swap_stats[7];
179char *swapnames[] = {
180	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
181	NULL
182};
183
184
185/* these are for keeping track of the proc array */
186
187static int nproc;
188static int onproc = -1;
189static int pref_len;
190static struct kinfo_proc *pbase;
191static struct kinfo_proc **pref;
192static struct kinfo_proc *previous_procs;
193static struct kinfo_proc **previous_pref;
194static int previous_proc_count = 0;
195static int previous_proc_count_max = 0;
196
197/* total number of io operations */
198static long total_inblock;
199static long total_oublock;
200static long total_majflt;
201
202/* these are for getting the memory statistics */
203
204static int pageshift;		/* log base 2 of the pagesize */
205
206/* define pagetok in terms of pageshift */
207
208#define pagetok(size) ((size) << pageshift)
209
210/* useful externals */
211long percentages();
212
213#ifdef ORDER
214/*
215 * Sorting orders.  The first element is the default.
216 */
217char *ordernames[] = {
218	"cpu", "size", "res", "time", "pri", "threads",
219	"total", "read", "write", "fault", "vcsw", "ivcsw",
220	"jid", NULL
221};
222#endif
223
224/* Per-cpu time states */
225static int maxcpu;
226static int maxid;
227static int ncpus;
228static u_long cpumask;
229static long *times;
230static long *pcpu_cp_time;
231static long *pcpu_cp_old;
232static long *pcpu_cp_diff;
233static int *pcpu_cpu_states;
234
235static int compare_jid(const void *a, const void *b);
236static int compare_pid(const void *a, const void *b);
237static const char *format_nice(const struct kinfo_proc *pp);
238static void getsysctl(const char *name, void *ptr, size_t len);
239static int swapmode(int *retavail, int *retfree);
240
241int
242machine_init(struct statics *statics, char do_unames)
243{
244	int pagesize;
245	size_t modelen;
246	struct passwd *pw;
247
248	modelen = sizeof(smpmode);
249	if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen,
250	    NULL, 0) != 0 &&
251	    sysctlbyname("kern.smp.active", &smpmode, &modelen,
252	    NULL, 0) != 0) ||
253	    modelen != sizeof(smpmode))
254		smpmode = 0;
255
256	if (do_unames) {
257	    while ((pw = getpwent()) != NULL) {
258		if (strlen(pw->pw_name) > namelength)
259			namelength = strlen(pw->pw_name);
260	    }
261	}
262	if (smpmode && namelength > SMPUNAMELEN)
263		namelength = SMPUNAMELEN;
264	else if (namelength > UPUNAMELEN)
265		namelength = UPUNAMELEN;
266
267	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
268	if (kd == NULL)
269		return (-1);
270
271	GETSYSCTL("kern.ccpu", ccpu);
272
273	/* this is used in calculating WCPU -- calculate it ahead of time */
274	logcpu = log(loaddouble(ccpu));
275
276	pbase = NULL;
277	pref = NULL;
278	nproc = 0;
279	onproc = -1;
280
281	/* get the page size and calculate pageshift from it */
282	pagesize = getpagesize();
283	pageshift = 0;
284	while (pagesize > 1) {
285		pageshift++;
286		pagesize >>= 1;
287	}
288
289	/* we only need the amount of log(2)1024 for our conversion */
290	pageshift -= LOG1024;
291
292	/* fill in the statics information */
293	statics->procstate_names = procstatenames;
294	statics->cpustate_names = cpustatenames;
295	statics->memory_names = memorynames;
296	statics->swap_names = swapnames;
297#ifdef ORDER
298	statics->order_names = ordernames;
299#endif
300
301	/* Adjust display based on ncpus */
302	if (pcpu_stats) {
303		int i, j, empty;
304		size_t size;
305
306		cpumask = 0;
307		ncpus = 0;
308		GETSYSCTL("kern.smp.maxcpus", maxcpu);
309		size = sizeof(long) * maxcpu * CPUSTATES;
310		times = malloc(size);
311		if (times == NULL)
312			err(1, "malloc %zd bytes", size);
313		if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
314			err(1, "sysctlbyname kern.cp_times");
315		maxid = (size / CPUSTATES / sizeof(long)) - 1;
316		for (i = 0; i <= maxid; i++) {
317			empty = 1;
318			for (j = 0; empty && j < CPUSTATES; j++) {
319				if (times[i * CPUSTATES + j] != 0)
320					empty = 0;
321			}
322			if (!empty) {
323				cpumask |= (1ul << i);
324				ncpus++;
325			}
326		}
327
328		if (ncpus > 1) {
329			y_mem += ncpus - 1;	/* 3 */
330			y_swap += ncpus - 1;	/* 4 */
331			y_idlecursor += ncpus - 1; /* 5 */
332			y_message += ncpus - 1;	/* 5 */
333			y_header += ncpus - 1;	/* 6 */
334			y_procs += ncpus - 1;	/* 7 */
335			Header_lines += ncpus - 1; /* 7 */
336		}
337		size = sizeof(long) * ncpus * CPUSTATES;
338		pcpu_cp_time = calloc(1, size);
339		pcpu_cp_old = calloc(1, size);
340		pcpu_cp_diff = calloc(1, size);
341		pcpu_cpu_states = calloc(1, size);
342		statics->ncpus = ncpus;
343	} else {
344		statics->ncpus = 1;
345	}
346
347	/* all done! */
348	return (0);
349}
350
351char *
352format_header(char *uname_field)
353{
354	static char Header[128];
355	const char *prehead;
356
357	switch (displaymode) {
358	case DISP_CPU:
359		/*
360		 * The logic of picking the right header format seems reverse
361		 * here because we only want to display a THR column when
362		 * "thread mode" is off (and threads are not listed as
363		 * separate lines).
364		 */
365		prehead = smpmode ?
366		    (ps.thread ? smp_header : smp_header_thr) :
367		    (ps.thread ? up_header : up_header_thr);
368		snprintf(Header, sizeof(Header), prehead,
369		    ps.jail ? " JID" : "",
370		    namelength, namelength, uname_field,
371		    ps.wcpu ? "WCPU" : "CPU");
372		break;
373	case DISP_IO:
374		prehead = io_header;
375		snprintf(Header, sizeof(Header), prehead,
376		    ps.jail ? " JID" : "",
377		    namelength, namelength, uname_field);
378		break;
379	}
380	cmdlengthdelta = strlen(Header) - 7;
381	return (Header);
382}
383
384static int swappgsin = -1;
385static int swappgsout = -1;
386extern struct timeval timeout;
387
388
389void
390get_system_info(struct system_info *si)
391{
392	long total;
393	struct loadavg sysload;
394	int mib[2];
395	struct timeval boottime;
396	size_t bt_size;
397	int i, j;
398	size_t size;
399
400	/* get the cp_time array */
401	if (pcpu_stats) {
402		size = (maxid + 1) * CPUSTATES * sizeof(long);
403		if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
404			err(1, "sysctlbyname kern.cp_times");
405	} else {
406		GETSYSCTL("kern.cp_time", cp_time);
407	}
408	GETSYSCTL("vm.loadavg", sysload);
409	GETSYSCTL("kern.lastpid", lastpid);
410
411	/* convert load averages to doubles */
412	for (i = 0; i < 3; i++)
413		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
414
415	if (pcpu_stats) {
416		for (i = j = 0; i <= maxid; i++) {
417			if ((cpumask & (1ul << i)) == 0)
418				continue;
419			/* convert cp_time counts to percentages */
420			percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
421			    &pcpu_cp_time[j * CPUSTATES],
422			    &pcpu_cp_old[j * CPUSTATES],
423			    &pcpu_cp_diff[j * CPUSTATES]);
424			j++;
425		}
426	} else {
427		/* convert cp_time counts to percentages */
428		percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
429	}
430
431	/* sum memory & swap statistics */
432	{
433		static unsigned int swap_delay = 0;
434		static int swapavail = 0;
435		static int swapfree = 0;
436		static int bufspace = 0;
437		static int nspgsin, nspgsout;
438
439		GETSYSCTL("vfs.bufspace", bufspace);
440		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
441		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
442		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[2]);
443		GETSYSCTL("vm.stats.vm.v_cache_count", memory_stats[3]);
444		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
445		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
446		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
447		/* convert memory stats to Kbytes */
448		memory_stats[0] = pagetok(memory_stats[0]);
449		memory_stats[1] = pagetok(memory_stats[1]);
450		memory_stats[2] = pagetok(memory_stats[2]);
451		memory_stats[3] = pagetok(memory_stats[3]);
452		memory_stats[4] = bufspace / 1024;
453		memory_stats[5] = pagetok(memory_stats[5]);
454		memory_stats[6] = -1;
455
456		/* first interval */
457		if (swappgsin < 0) {
458			swap_stats[4] = 0;
459			swap_stats[5] = 0;
460		}
461
462		/* compute differences between old and new swap statistic */
463		else {
464			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
465			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
466		}
467
468		swappgsin = nspgsin;
469		swappgsout = nspgsout;
470
471		/* call CPU heavy swapmode() only for changes */
472		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
473			swap_stats[3] = swapmode(&swapavail, &swapfree);
474			swap_stats[0] = swapavail;
475			swap_stats[1] = swapavail - swapfree;
476			swap_stats[2] = swapfree;
477		}
478		swap_delay = 1;
479		swap_stats[6] = -1;
480	}
481
482	/* set arrays and strings */
483	if (pcpu_stats) {
484		si->cpustates = pcpu_cpu_states;
485		si->cpumask = cpumask;
486		si->ncpus = ncpus;
487	} else {
488		si->cpustates = cpu_states;
489		si->cpumask = 1;
490		si->ncpus = 1;
491	}
492	si->memory = memory_stats;
493	si->swap = swap_stats;
494
495
496	if (lastpid > 0) {
497		si->last_pid = lastpid;
498	} else {
499		si->last_pid = -1;
500	}
501
502	/*
503	 * Print how long system has been up.
504	 * (Found by looking getting "boottime" from the kernel)
505	 */
506	mib[0] = CTL_KERN;
507	mib[1] = KERN_BOOTTIME;
508	bt_size = sizeof(boottime);
509	if (sysctl(mib, 2, &boottime, &bt_size, NULL, 0) != -1 &&
510	    boottime.tv_sec != 0) {
511		si->boottime = boottime;
512	} else {
513		si->boottime.tv_sec = -1;
514	}
515}
516
517#define NOPROC	((void *)-1)
518
519/*
520 * We need to compare data from the old process entry with the new
521 * process entry.
522 * To facilitate doing this quickly we stash a pointer in the kinfo_proc
523 * structure to cache the mapping.  We also use a negative cache pointer
524 * of NOPROC to avoid duplicate lookups.
525 * XXX: this could be done when the actual processes are fetched, we do
526 * it here out of laziness.
527 */
528const struct kinfo_proc *
529get_old_proc(struct kinfo_proc *pp)
530{
531	struct kinfo_proc **oldpp, *oldp;
532
533	/*
534	 * If this is the first fetch of the kinfo_procs then we don't have
535	 * any previous entries.
536	 */
537	if (previous_proc_count == 0)
538		return (NULL);
539	/* negative cache? */
540	if (pp->ki_udata == NOPROC)
541		return (NULL);
542	/* cached? */
543	if (pp->ki_udata != NULL)
544		return (pp->ki_udata);
545	/*
546	 * Not cached,
547	 * 1) look up based on pid.
548	 * 2) compare process start.
549	 * If we fail here, then setup a negative cache entry, otherwise
550	 * cache it.
551	 */
552	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
553	    sizeof(*previous_pref), compare_pid);
554	if (oldpp == NULL) {
555		pp->ki_udata = NOPROC;
556		return (NULL);
557	}
558	oldp = *oldpp;
559	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
560		pp->ki_udata = NOPROC;
561		return (NULL);
562	}
563	pp->ki_udata = oldp;
564	return (oldp);
565}
566
567/*
568 * Return the total amount of IO done in blocks in/out and faults.
569 * store the values individually in the pointers passed in.
570 */
571long
572get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
573    long *vcsw, long *ivcsw)
574{
575	const struct kinfo_proc *oldp;
576	static struct kinfo_proc dummy;
577	long ret;
578
579	oldp = get_old_proc(pp);
580	if (oldp == NULL) {
581		bzero(&dummy, sizeof(dummy));
582		oldp = &dummy;
583	}
584	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
585	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
586	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
587	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
588	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
589	ret =
590	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
591	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
592	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
593	return (ret);
594}
595
596/*
597 * Return the total number of block in/out and faults by a process.
598 */
599long
600get_io_total(struct kinfo_proc *pp)
601{
602	long dummy;
603
604	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
605}
606
607static struct handle handle;
608
609caddr_t
610get_process_info(struct system_info *si, struct process_select *sel,
611    int (*compare)(const void *, const void *))
612{
613	int i;
614	int total_procs;
615	long p_io;
616	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
617	int active_procs;
618	struct kinfo_proc **prefp;
619	struct kinfo_proc *pp;
620	struct kinfo_proc *prev_pp = NULL;
621
622	/* these are copied out of sel for speed */
623	int show_idle;
624	int show_self;
625	int show_system;
626	int show_uid;
627	int show_command;
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
669	/* count up process states and get pointers to interesting procs */
670	total_procs = 0;
671	active_procs = 0;
672	total_inblock = 0;
673	total_oublock = 0;
674	total_majflt = 0;
675	memset((char *)process_states, 0, sizeof(process_states));
676	prefp = pref;
677	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
678
679		if (pp->ki_stat == 0)
680			/* not in use */
681			continue;
682
683		if (!show_self && pp->ki_pid == sel->self)
684			/* skip self */
685			continue;
686
687		if (!show_system && (pp->ki_flag & P_SYSTEM))
688			/* skip system process */
689			continue;
690
691		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
692		    &p_vcsw, &p_ivcsw);
693		total_inblock += p_inblock;
694		total_oublock += p_oublock;
695		total_majflt += p_majflt;
696		total_procs++;
697		process_states[pp->ki_stat]++;
698
699		if (pp->ki_stat == SZOMB)
700			/* skip zombies */
701			continue;
702
703		if (displaymode == DISP_CPU && !show_idle &&
704		    (pp->ki_pctcpu == 0 ||
705		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
706			/* skip idle or non-running processes */
707			continue;
708
709		if (displaymode == DISP_IO && !show_idle && p_io == 0)
710			/* skip processes that aren't doing I/O */
711			continue;
712
713		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
714			/* skip proc. that don't belong to the selected UID */
715			continue;
716
717		/*
718		 * When not showing threads, take the first thread
719		 * for output and add the fields that we can from
720		 * the rest of the process's threads rather than
721		 * using the system's mostly-broken KERN_PROC_PROC.
722		 */
723		if (sel->thread || prev_pp == NULL ||
724		    prev_pp->ki_pid != pp->ki_pid) {
725			*prefp++ = pp;
726			active_procs++;
727			prev_pp = pp;
728		} else {
729			prev_pp->ki_pctcpu += pp->ki_pctcpu;
730		}
731	}
732
733	/* if requested, sort the "interesting" processes */
734	if (compare != NULL)
735		qsort(pref, active_procs, sizeof(*pref), compare);
736
737	/* remember active and total counts */
738	si->p_total = total_procs;
739	si->p_active = pref_len = active_procs;
740
741	/* pass back a handle */
742	handle.next_proc = pref;
743	handle.remaining = active_procs;
744	return ((caddr_t)&handle);
745}
746
747static char fmt[128];	/* static area where result is built */
748
749char *
750format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
751{
752	struct kinfo_proc *pp;
753	const struct kinfo_proc *oldp;
754	long cputime;
755	double pct;
756	struct handle *hp;
757	char status[16];
758	int state;
759	struct rusage ru, *rup;
760	long p_tot, s_tot;
761	char *proc_fmt, thr_buf[6], jid_buf[6];
762	char *cmdbuf = NULL;
763	char **args;
764
765	/* find and remember the next proc structure */
766	hp = (struct handle *)handle;
767	pp = *(hp->next_proc++);
768	hp->remaining--;
769
770	/* get the process's command name */
771	if ((pp->ki_flag & P_INMEM) == 0) {
772		/*
773		 * Print swapped processes as <pname>
774		 */
775		size_t len;
776
777		len = strlen(pp->ki_comm);
778		if (len > sizeof(pp->ki_comm) - 3)
779			len = sizeof(pp->ki_comm) - 3;
780		memmove(pp->ki_comm + 1, pp->ki_comm, len);
781		pp->ki_comm[0] = '<';
782		pp->ki_comm[len + 1] = '>';
783		pp->ki_comm[len + 2] = '\0';
784	}
785
786	/*
787	 * Convert the process's runtime from microseconds to seconds.  This
788	 * time includes the interrupt time although that is not wanted here.
789	 * ps(1) is similarly sloppy.
790	 */
791	cputime = (pp->ki_runtime + 500000) / 1000000;
792
793	/* calculate the base for cpu percentages */
794	pct = pctdouble(pp->ki_pctcpu);
795
796	/* generate "STATE" field */
797	switch (state = pp->ki_stat) {
798	case SRUN:
799		if (smpmode && pp->ki_oncpu != 0xff)
800			sprintf(status, "CPU%d", pp->ki_oncpu);
801		else
802			strcpy(status, "RUN");
803		break;
804	case SLOCK:
805		if (pp->ki_kiflag & KI_LOCKBLOCK) {
806			sprintf(status, "*%.6s", pp->ki_lockname);
807			break;
808		}
809		/* fall through */
810	case SSLEEP:
811		if (pp->ki_wmesg != NULL) {
812			sprintf(status, "%.6s", pp->ki_wmesg);
813			break;
814		}
815		/* FALLTHROUGH */
816	default:
817
818		if (state >= 0 &&
819		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
820			sprintf(status, "%.6s", state_abbrev[state]);
821		else
822			sprintf(status, "?%5d", state);
823		break;
824	}
825
826	cmdbuf = (char *)malloc(cmdlengthdelta + 1);
827	if (cmdbuf == NULL) {
828		warn("malloc(%d)", cmdlengthdelta + 1);
829		return NULL;
830	}
831
832	if (!(flags & FMT_SHOWARGS)) {
833		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
834		    pp->ki_ocomm[0]) {
835			snprintf(cmdbuf, cmdlengthdelta, "{%s}", pp->ki_ocomm);
836		} else {
837			snprintf(cmdbuf, cmdlengthdelta, "%s", pp->ki_comm);
838		}
839	} else {
840		if (pp->ki_flag & P_SYSTEM ||
841		    pp->ki_args == NULL ||
842		    (args = kvm_getargv(kd, pp, cmdlengthdelta)) == NULL ||
843		    !(*args)) {
844			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
845		    	pp->ki_ocomm[0]) {
846				snprintf(cmdbuf, cmdlengthdelta,
847				    "{%s}", pp->ki_ocomm);
848			} else {
849				snprintf(cmdbuf, cmdlengthdelta,
850				    "[%s]", pp->ki_comm);
851			}
852		} else {
853			char *src, *dst, *argbuf;
854			char *cmd;
855			size_t argbuflen;
856			size_t len;
857
858			argbuflen = cmdlengthdelta * 4;
859			argbuf = (char *)malloc(argbuflen + 1);
860			if (argbuf == NULL) {
861				warn("malloc(%d)", argbuflen + 1);
862				free(cmdbuf);
863				return NULL;
864			}
865
866			dst = argbuf;
867
868			/* Extract cmd name from argv */
869			cmd = strrchr(*args, '/');
870			if (cmd == NULL)
871				cmd = *args;
872			else
873				cmd++;
874
875			for (; (src = *args++) != NULL; ) {
876				if (*src == '\0')
877					continue;
878				len = (argbuflen - (dst - argbuf) - 1) / 4;
879				strvisx(dst, src,
880				    strlen(src) < len ? strlen(src) : len,
881				    VIS_NL | VIS_CSTYLE);
882				while (*dst != '\0')
883					dst++;
884				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
885					*dst++ = ' '; /* add delimiting space */
886			}
887			if (dst != argbuf && dst[-1] == ' ')
888				dst--;
889			*dst = '\0';
890
891			if (strcmp(cmd, pp->ki_comm) != 0 )
892				snprintf(cmdbuf, cmdlengthdelta,
893				    "%s (%s)",argbuf,  pp->ki_comm);
894			else
895				strlcpy(cmdbuf, argbuf, cmdlengthdelta);
896
897			free(argbuf);
898		}
899	}
900
901	if (ps.jail == 0)
902		jid_buf[0] = '\0';
903	else
904		snprintf(jid_buf, sizeof(jid_buf), " %*d",
905		    sizeof(jid_buf) - 3, pp->ki_jid);
906
907	if (displaymode == DISP_IO) {
908		oldp = get_old_proc(pp);
909		if (oldp != NULL) {
910			ru.ru_inblock = RU(pp)->ru_inblock -
911			    RU(oldp)->ru_inblock;
912			ru.ru_oublock = RU(pp)->ru_oublock -
913			    RU(oldp)->ru_oublock;
914			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
915			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
916			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
917			rup = &ru;
918		} else {
919			rup = RU(pp);
920		}
921		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
922		s_tot = total_inblock + total_oublock + total_majflt;
923
924		sprintf(fmt, io_Proc_format,
925		    pp->ki_pid,
926		    jid_buf,
927		    namelength, namelength, (*get_userid)(pp->ki_ruid),
928		    rup->ru_nvcsw,
929		    rup->ru_nivcsw,
930		    rup->ru_inblock,
931		    rup->ru_oublock,
932		    rup->ru_majflt,
933		    p_tot,
934		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
935		    screen_width > cmdlengthdelta ?
936		    screen_width - cmdlengthdelta : 0,
937		    printable(cmdbuf));
938
939		free(cmdbuf);
940
941		return (fmt);
942	}
943
944	/* format this entry */
945	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
946	if (ps.thread != 0)
947		thr_buf[0] = '\0';
948	else
949		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
950		    sizeof(thr_buf) - 2, pp->ki_numthreads);
951
952	sprintf(fmt, proc_fmt,
953	    pp->ki_pid,
954	    jid_buf,
955	    namelength, namelength, (*get_userid)(pp->ki_ruid),
956	    thr_buf,
957	    pp->ki_pri.pri_level - PZERO,
958	    format_nice(pp),
959	    format_k2(PROCSIZE(pp)),
960	    format_k2(pagetok(pp->ki_rssize)),
961	    status,
962	    smpmode ? pp->ki_lastcpu : 0,
963	    format_time(cputime),
964	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
965	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
966	    printable(cmdbuf));
967
968	free(cmdbuf);
969
970	/* return the result */
971	return (fmt);
972}
973
974static void
975getsysctl(const char *name, void *ptr, size_t len)
976{
977	size_t nlen = len;
978
979	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
980		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
981		    strerror(errno));
982		quit(23);
983	}
984	if (nlen != len) {
985		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
986		    name, (unsigned long)len, (unsigned long)nlen);
987		quit(23);
988	}
989}
990
991static const char *
992format_nice(const struct kinfo_proc *pp)
993{
994	const char *fifo, *kthread;
995	int rtpri;
996	static char nicebuf[4 + 1];
997
998	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
999	kthread = (pp->ki_flag & P_KTHREAD) ? "k" : "";
1000	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1001	case PRI_ITHD:
1002		return ("-");
1003	case PRI_REALTIME:
1004		/*
1005		 * XXX: the kernel doesn't tell us the original rtprio and
1006		 * doesn't really know what it was, so to recover it we
1007		 * must be more chummy with the implementation than the
1008		 * implementation is with itself.  pri_user gives a
1009		 * constant "base" priority, but is only initialized
1010		 * properly for user threads.  pri_native gives what the
1011		 * kernel calls the "base" priority, but it isn't constant
1012		 * since it is changed by priority propagation.  pri_native
1013		 * also isn't properly initialized for all threads, but it
1014		 * is properly initialized for kernel realtime and idletime
1015		 * threads.  Thus we use pri_user for the base priority of
1016		 * user threads (it is always correct) and pri_native for
1017		 * the base priority of kernel realtime and idletime threads
1018		 * (there is nothing better, and it is usually correct).
1019		 *
1020		 * The field width and thus the buffer are too small for
1021		 * values like "kr31F", but such values shouldn't occur,
1022		 * and if they do then the tailing "F" is not displayed.
1023		 */
1024		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1025		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1026		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1027		    kthread, rtpri, fifo);
1028		break;
1029	case PRI_TIMESHARE:
1030		if (pp->ki_flag & P_KTHREAD)
1031			return ("-");
1032		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1033		break;
1034	case PRI_IDLE:
1035		/* XXX: as above. */
1036		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1037		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1038		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1039		    kthread, rtpri, fifo);
1040		break;
1041	default:
1042		return ("?");
1043	}
1044	return (nicebuf);
1045}
1046
1047/* comparison routines for qsort */
1048
1049static int
1050compare_pid(const void *p1, const void *p2)
1051{
1052	const struct kinfo_proc * const *pp1 = p1;
1053	const struct kinfo_proc * const *pp2 = p2;
1054
1055	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1056		abort();
1057
1058	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1059}
1060
1061/*
1062 *  proc_compare - comparison function for "qsort"
1063 *	Compares the resource consumption of two processes using five
1064 *	distinct keys.  The keys (in descending order of importance) are:
1065 *	percent cpu, cpu ticks, state, resident set size, total virtual
1066 *	memory usage.  The process states are ordered as follows (from least
1067 *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1068 *	array declaration below maps a process state index into a number
1069 *	that reflects this ordering.
1070 */
1071
1072static int sorted_state[] = {
1073	0,	/* not used		*/
1074	3,	/* sleep		*/
1075	1,	/* ABANDONED (WAIT)	*/
1076	6,	/* run			*/
1077	5,	/* start		*/
1078	2,	/* zombie		*/
1079	4	/* stop			*/
1080};
1081
1082
1083#define ORDERKEY_PCTCPU(a, b) do { \
1084	long diff; \
1085	if (ps.wcpu) \
1086		diff = floor(1.0E6 * weighted_cpu(pctdouble((b)->ki_pctcpu), \
1087		    (b))) - \
1088		    floor(1.0E6 * weighted_cpu(pctdouble((a)->ki_pctcpu), \
1089		    (a))); \
1090	else \
1091		diff = (long)(b)->ki_pctcpu - (long)(a)->ki_pctcpu; \
1092	if (diff != 0) \
1093		return (diff > 0 ? 1 : -1); \
1094} while (0)
1095
1096#define ORDERKEY_CPTICKS(a, b) do { \
1097	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1098	if (diff != 0) \
1099		return (diff > 0 ? 1 : -1); \
1100} while (0)
1101
1102#define ORDERKEY_STATE(a, b) do { \
1103	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1104	if (diff != 0) \
1105		return (diff > 0 ? 1 : -1); \
1106} while (0)
1107
1108#define ORDERKEY_PRIO(a, b) do { \
1109	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1110	if (diff != 0) \
1111		return (diff > 0 ? 1 : -1); \
1112} while (0)
1113
1114#define	ORDERKEY_THREADS(a, b) do { \
1115	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1116	if (diff != 0) \
1117		return (diff > 0 ? 1 : -1); \
1118} while (0)
1119
1120#define ORDERKEY_RSSIZE(a, b) do { \
1121	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1122	if (diff != 0) \
1123		return (diff > 0 ? 1 : -1); \
1124} while (0)
1125
1126#define ORDERKEY_MEM(a, b) do { \
1127	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1128	if (diff != 0) \
1129		return (diff > 0 ? 1 : -1); \
1130} while (0)
1131
1132#define ORDERKEY_JID(a, b) do { \
1133	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1134	if (diff != 0) \
1135		return (diff > 0 ? 1 : -1); \
1136} while (0)
1137
1138/* compare_cpu - the comparison function for sorting by cpu percentage */
1139
1140int
1141#ifdef ORDER
1142compare_cpu(void *arg1, void *arg2)
1143#else
1144proc_compare(void *arg1, void *arg2)
1145#endif
1146{
1147	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1148	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1149
1150	ORDERKEY_PCTCPU(p1, p2);
1151	ORDERKEY_CPTICKS(p1, p2);
1152	ORDERKEY_STATE(p1, p2);
1153	ORDERKEY_PRIO(p1, p2);
1154	ORDERKEY_RSSIZE(p1, p2);
1155	ORDERKEY_MEM(p1, p2);
1156
1157	return (0);
1158}
1159
1160#ifdef ORDER
1161/* "cpu" compare routines */
1162int compare_size(), compare_res(), compare_time(), compare_prio(),
1163    compare_threads();
1164
1165/*
1166 * "io" compare routines.  Context switches aren't i/o, but are displayed
1167 * on the "io" display.
1168 */
1169int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1170    compare_vcsw(), compare_ivcsw();
1171
1172int (*compares[])() = {
1173	compare_cpu,
1174	compare_size,
1175	compare_res,
1176	compare_time,
1177	compare_prio,
1178	compare_threads,
1179	compare_iototal,
1180	compare_ioread,
1181	compare_iowrite,
1182	compare_iofault,
1183	compare_vcsw,
1184	compare_ivcsw,
1185	compare_jid,
1186	NULL
1187};
1188
1189/* compare_size - the comparison function for sorting by total memory usage */
1190
1191int
1192compare_size(void *arg1, void *arg2)
1193{
1194	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1195	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1196
1197	ORDERKEY_MEM(p1, p2);
1198	ORDERKEY_RSSIZE(p1, p2);
1199	ORDERKEY_PCTCPU(p1, p2);
1200	ORDERKEY_CPTICKS(p1, p2);
1201	ORDERKEY_STATE(p1, p2);
1202	ORDERKEY_PRIO(p1, p2);
1203
1204	return (0);
1205}
1206
1207/* compare_res - the comparison function for sorting by resident set size */
1208
1209int
1210compare_res(void *arg1, void *arg2)
1211{
1212	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1213	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1214
1215	ORDERKEY_RSSIZE(p1, p2);
1216	ORDERKEY_MEM(p1, p2);
1217	ORDERKEY_PCTCPU(p1, p2);
1218	ORDERKEY_CPTICKS(p1, p2);
1219	ORDERKEY_STATE(p1, p2);
1220	ORDERKEY_PRIO(p1, p2);
1221
1222	return (0);
1223}
1224
1225/* compare_time - the comparison function for sorting by total cpu time */
1226
1227int
1228compare_time(void *arg1, void *arg2)
1229{
1230	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1231	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1232
1233	ORDERKEY_CPTICKS(p1, p2);
1234	ORDERKEY_PCTCPU(p1, p2);
1235	ORDERKEY_STATE(p1, p2);
1236	ORDERKEY_PRIO(p1, p2);
1237	ORDERKEY_RSSIZE(p1, p2);
1238	ORDERKEY_MEM(p1, p2);
1239
1240	return (0);
1241}
1242
1243/* compare_prio - the comparison function for sorting by priority */
1244
1245int
1246compare_prio(void *arg1, void *arg2)
1247{
1248	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1249	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1250
1251	ORDERKEY_PRIO(p1, p2);
1252	ORDERKEY_CPTICKS(p1, p2);
1253	ORDERKEY_PCTCPU(p1, p2);
1254	ORDERKEY_STATE(p1, p2);
1255	ORDERKEY_RSSIZE(p1, p2);
1256	ORDERKEY_MEM(p1, p2);
1257
1258	return (0);
1259}
1260
1261/* compare_threads - the comparison function for sorting by threads */
1262int
1263compare_threads(void *arg1, void *arg2)
1264{
1265	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1266	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1267
1268	ORDERKEY_THREADS(p1, p2);
1269	ORDERKEY_PCTCPU(p1, p2);
1270	ORDERKEY_CPTICKS(p1, p2);
1271	ORDERKEY_STATE(p1, p2);
1272	ORDERKEY_PRIO(p1, p2);
1273	ORDERKEY_RSSIZE(p1, p2);
1274	ORDERKEY_MEM(p1, p2);
1275
1276	return (0);
1277}
1278
1279/* compare_jid - the comparison function for sorting by jid */
1280static int
1281compare_jid(const void *arg1, const void *arg2)
1282{
1283	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1284	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1285
1286	ORDERKEY_JID(p1, p2);
1287	ORDERKEY_PCTCPU(p1, p2);
1288	ORDERKEY_CPTICKS(p1, p2);
1289	ORDERKEY_STATE(p1, p2);
1290	ORDERKEY_PRIO(p1, p2);
1291	ORDERKEY_RSSIZE(p1, p2);
1292	ORDERKEY_MEM(p1, p2);
1293
1294	return (0);
1295}
1296#endif /* ORDER */
1297
1298/* assorted comparison functions for sorting by i/o */
1299
1300int
1301#ifdef ORDER
1302compare_iototal(void *arg1, void *arg2)
1303#else
1304io_compare(void *arg1, void *arg2)
1305#endif
1306{
1307	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1308	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1309
1310	return (get_io_total(p2) - get_io_total(p1));
1311}
1312
1313#ifdef ORDER
1314int
1315compare_ioread(void *arg1, void *arg2)
1316{
1317	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1318	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1319	long dummy, inp1, inp2;
1320
1321	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1322	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1323
1324	return (inp2 - inp1);
1325}
1326
1327int
1328compare_iowrite(void *arg1, void *arg2)
1329{
1330	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1331	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1332	long dummy, oup1, oup2;
1333
1334	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1335	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1336
1337	return (oup2 - oup1);
1338}
1339
1340int
1341compare_iofault(void *arg1, void *arg2)
1342{
1343	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1344	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1345	long dummy, flp1, flp2;
1346
1347	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1348	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1349
1350	return (flp2 - flp1);
1351}
1352
1353int
1354compare_vcsw(void *arg1, void *arg2)
1355{
1356	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1357	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1358	long dummy, flp1, flp2;
1359
1360	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1361	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1362
1363	return (flp2 - flp1);
1364}
1365
1366int
1367compare_ivcsw(void *arg1, void *arg2)
1368{
1369	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1370	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1371	long dummy, flp1, flp2;
1372
1373	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1374	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1375
1376	return (flp2 - flp1);
1377}
1378#endif /* ORDER */
1379
1380/*
1381 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1382 *		the process does not exist.
1383 *		It is EXTREMLY IMPORTANT that this function work correctly.
1384 *		If top runs setuid root (as in SVR4), then this function
1385 *		is the only thing that stands in the way of a serious
1386 *		security problem.  It validates requests for the "kill"
1387 *		and "renice" commands.
1388 */
1389
1390int
1391proc_owner(int pid)
1392{
1393	int cnt;
1394	struct kinfo_proc **prefp;
1395	struct kinfo_proc *pp;
1396
1397	prefp = pref;
1398	cnt = pref_len;
1399	while (--cnt >= 0) {
1400		pp = *prefp++;
1401		if (pp->ki_pid == (pid_t)pid)
1402			return ((int)pp->ki_ruid);
1403	}
1404	return (-1);
1405}
1406
1407static int
1408swapmode(int *retavail, int *retfree)
1409{
1410	int n;
1411	int pagesize = getpagesize();
1412	struct kvm_swap swapary[1];
1413
1414	*retavail = 0;
1415	*retfree = 0;
1416
1417#define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1418
1419	n = kvm_getswapinfo(kd, swapary, 1, 0);
1420	if (n < 0 || swapary[0].ksw_total == 0)
1421		return (0);
1422
1423	*retavail = CONVERT(swapary[0].ksw_total);
1424	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1425
1426	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1427	return (n);
1428}
1429