machine.c revision 266280
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: stable/10/usr.bin/top/machine.c 266280 2014-05-17 02:39:20Z bdrewery $
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
61extern struct process_select ps;
62extern char* printable(char *);
63static int smpmode;
64enum displaymodes displaymode;
65#ifdef TOP_USERNAME_LEN
66static int namelength = TOP_USERNAME_LEN;
67#else
68static int namelength = 8;
69#endif
70static int cmdlengthdelta;
71
72/* Prototypes for top internals */
73void quit(int);
74
75/* get_process_info passes back a handle.  This is what it looks like: */
76
77struct 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
103static 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
109static char smp_header_thr[] =
110    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE   C   TIME %7s COMMAND";
111static char smp_header[] =
112    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE   C   TIME %7s COMMAND";
113
114#define smp_Proc_format \
115    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s %2d%7s %6.2f%% %.*s"
116
117static char up_header_thr[] =
118    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE    TIME %7s COMMAND";
119static char up_header[] =
120    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE    TIME %7s COMMAND";
121
122#define up_Proc_format \
123    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s%.0d%7s %6.2f%% %.*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
130char *state_abbrev[] = {
131	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
132};
133
134
135static kvm_t *kd;
136
137/* values that we stash away in _init and use in later routines */
138
139static double logcpu;
140
141/* these are retrieved from the kernel in _init */
142
143static load_avg  ccpu;
144
145/* these are used in the get_ functions */
146
147static int lastpid;
148
149/* these are for calculating cpu state percentages */
150
151static long cp_time[CPUSTATES];
152static long cp_old[CPUSTATES];
153static long cp_diff[CPUSTATES];
154
155/* these are for detailing the process states */
156
157int process_states[8];
158char *procstatenames[] = {
159	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
160	" zombie, ", " waiting, ", " lock, ",
161	NULL
162};
163
164/* these are for detailing the cpu states */
165
166int cpu_states[CPUSTATES];
167char *cpustatenames[] = {
168	"user", "nice", "system", "interrupt", "idle", NULL
169};
170
171/* these are for detailing the memory statistics */
172
173int memory_stats[7];
174char *memorynames[] = {
175	"K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ",
176	"K Free", NULL
177};
178
179int arc_stats[7];
180char *arcnames[] = {
181	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
182	NULL
183};
184
185int swap_stats[7];
186char *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
194static int nproc;
195static int onproc = -1;
196static int pref_len;
197static struct kinfo_proc *pbase;
198static struct kinfo_proc **pref;
199static struct kinfo_proc *previous_procs;
200static struct kinfo_proc **previous_pref;
201static int previous_proc_count = 0;
202static int previous_proc_count_max = 0;
203static int arc_enabled;
204
205/* total number of io operations */
206static long total_inblock;
207static long total_oublock;
208static long total_majflt;
209
210/* these are for getting the memory statistics */
211
212static 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 */
219long percentages();
220
221#ifdef ORDER
222/*
223 * Sorting orders.  The first element is the default.
224 */
225char *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 */
233static int maxcpu;
234static int maxid;
235static int ncpus;
236static u_long cpumask;
237static long *times;
238static long *pcpu_cp_time;
239static long *pcpu_cp_old;
240static long *pcpu_cp_diff;
241static int *pcpu_cpu_states;
242
243static int compare_jid(const void *a, const void *b);
244static int compare_pid(const void *a, const void *b);
245static int compare_tid(const void *a, const void *b);
246static const char *format_nice(const struct kinfo_proc *pp);
247static void getsysctl(const char *name, void *ptr, size_t len);
248static int swapmode(int *retavail, int *retfree);
249static void update_layout(void);
250
251void
252toggle_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. */
261static void
262update_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
286int
287machine_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
391char *
392format_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
424static int swappgsin = -1;
425static int swappgsout = -1;
426extern struct timeval timeout;
427
428
429void
430get_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 */
576const struct kinfo_proc *
577get_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 */
619long
620get_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 */
647long
648get_io_total(struct kinfo_proc *pp)
649{
650	long dummy;
651
652	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
653}
654
655static struct handle handle;
656
657caddr_t
658get_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_jid;
672	int show_self;
673	int show_system;
674	int show_uid;
675	int show_command;
676	int show_kidle;
677
678	/*
679	 * Save the previous process info.
680	 */
681	if (previous_proc_count_max < nproc) {
682		free(previous_procs);
683		previous_procs = malloc(nproc * sizeof(*previous_procs));
684		free(previous_pref);
685		previous_pref = malloc(nproc * sizeof(*previous_pref));
686		if (previous_procs == NULL || previous_pref == NULL) {
687			(void) fprintf(stderr, "top: Out of memory.\n");
688			quit(23);
689		}
690		previous_proc_count_max = nproc;
691	}
692	if (nproc) {
693		for (i = 0; i < nproc; i++)
694			previous_pref[i] = &previous_procs[i];
695		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
696		qsort(previous_pref, nproc, sizeof(*previous_pref),
697		    ps.thread ? compare_tid : compare_pid);
698	}
699	previous_proc_count = nproc;
700
701	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
702	    0, &nproc);
703	if (nproc > onproc)
704		pref = realloc(pref, sizeof(*pref) * (onproc = nproc));
705	if (pref == NULL || pbase == NULL) {
706		(void) fprintf(stderr, "top: Out of memory.\n");
707		quit(23);
708	}
709	/* get a pointer to the states summary array */
710	si->procstates = process_states;
711
712	/* set up flags which define what we are going to select */
713	show_idle = sel->idle;
714	show_jid = sel->jid != -1;
715	show_self = sel->self == -1;
716	show_system = sel->system;
717	show_uid = sel->uid != -1;
718	show_command = sel->command != NULL;
719	show_kidle = sel->kidle;
720
721	/* count up process states and get pointers to interesting procs */
722	total_procs = 0;
723	active_procs = 0;
724	total_inblock = 0;
725	total_oublock = 0;
726	total_majflt = 0;
727	memset((char *)process_states, 0, sizeof(process_states));
728	prefp = pref;
729	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
730
731		if (pp->ki_stat == 0)
732			/* not in use */
733			continue;
734
735		if (!show_self && pp->ki_pid == sel->self)
736			/* skip self */
737			continue;
738
739		if (!show_system && (pp->ki_flag & P_SYSTEM))
740			/* skip system process */
741			continue;
742
743		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
744		    &p_vcsw, &p_ivcsw);
745		total_inblock += p_inblock;
746		total_oublock += p_oublock;
747		total_majflt += p_majflt;
748		total_procs++;
749		process_states[pp->ki_stat]++;
750
751		if (pp->ki_stat == SZOMB)
752			/* skip zombies */
753			continue;
754
755		if (!show_kidle && pp->ki_tdflags & TDF_IDLETD)
756			/* skip kernel idle process */
757			continue;
758
759		if (displaymode == DISP_CPU && !show_idle &&
760		    (pp->ki_pctcpu == 0 ||
761		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
762			/* skip idle or non-running processes */
763			continue;
764
765		if (displaymode == DISP_IO && !show_idle && p_io == 0)
766			/* skip processes that aren't doing I/O */
767			continue;
768
769		if (show_jid && pp->ki_jid != sel->jid)
770			/* skip proc. that don't belong to the selected JID */
771			continue;
772
773		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
774			/* skip proc. that don't belong to the selected UID */
775			continue;
776
777		*prefp++ = pp;
778		active_procs++;
779	}
780
781	/* if requested, sort the "interesting" processes */
782	if (compare != NULL)
783		qsort(pref, active_procs, sizeof(*pref), compare);
784
785	/* remember active and total counts */
786	si->p_total = total_procs;
787	si->p_active = pref_len = active_procs;
788
789	/* pass back a handle */
790	handle.next_proc = pref;
791	handle.remaining = active_procs;
792	return ((caddr_t)&handle);
793}
794
795static char fmt[512];	/* static area where result is built */
796
797char *
798format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
799{
800	struct kinfo_proc *pp;
801	const struct kinfo_proc *oldp;
802	long cputime;
803	double pct;
804	struct handle *hp;
805	char status[16];
806	int cpu, state;
807	struct rusage ru, *rup;
808	long p_tot, s_tot;
809	char *proc_fmt, thr_buf[6], jid_buf[6];
810	char *cmdbuf = NULL;
811	char **args;
812	const int cmdlen = 128;
813
814	/* find and remember the next proc structure */
815	hp = (struct handle *)handle;
816	pp = *(hp->next_proc++);
817	hp->remaining--;
818
819	/* get the process's command name */
820	if ((pp->ki_flag & P_INMEM) == 0) {
821		/*
822		 * Print swapped processes as <pname>
823		 */
824		size_t len;
825
826		len = strlen(pp->ki_comm);
827		if (len > sizeof(pp->ki_comm) - 3)
828			len = sizeof(pp->ki_comm) - 3;
829		memmove(pp->ki_comm + 1, pp->ki_comm, len);
830		pp->ki_comm[0] = '<';
831		pp->ki_comm[len + 1] = '>';
832		pp->ki_comm[len + 2] = '\0';
833	}
834
835	/*
836	 * Convert the process's runtime from microseconds to seconds.  This
837	 * time includes the interrupt time although that is not wanted here.
838	 * ps(1) is similarly sloppy.
839	 */
840	cputime = (pp->ki_runtime + 500000) / 1000000;
841
842	/* calculate the base for cpu percentages */
843	pct = pctdouble(pp->ki_pctcpu);
844
845	/* generate "STATE" field */
846	switch (state = pp->ki_stat) {
847	case SRUN:
848		if (smpmode && pp->ki_oncpu != 0xff)
849			sprintf(status, "CPU%d", pp->ki_oncpu);
850		else
851			strcpy(status, "RUN");
852		break;
853	case SLOCK:
854		if (pp->ki_kiflag & KI_LOCKBLOCK) {
855			sprintf(status, "*%.6s", pp->ki_lockname);
856			break;
857		}
858		/* fall through */
859	case SSLEEP:
860		if (pp->ki_wmesg != NULL) {
861			sprintf(status, "%.6s", pp->ki_wmesg);
862			break;
863		}
864		/* FALLTHROUGH */
865	default:
866
867		if (state >= 0 &&
868		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
869			sprintf(status, "%.6s", state_abbrev[state]);
870		else
871			sprintf(status, "?%5d", state);
872		break;
873	}
874
875	cmdbuf = (char *)malloc(cmdlen + 1);
876	if (cmdbuf == NULL) {
877		warn("malloc(%d)", cmdlen + 1);
878		return NULL;
879	}
880
881	if (!(flags & FMT_SHOWARGS)) {
882		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
883		    pp->ki_tdname[0]) {
884			snprintf(cmdbuf, cmdlen, "%s{%s}", pp->ki_comm,
885			    pp->ki_tdname);
886		} else {
887			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
888		}
889	} else {
890		if (pp->ki_flag & P_SYSTEM ||
891		    pp->ki_args == NULL ||
892		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
893		    !(*args)) {
894			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
895		    	    pp->ki_tdname[0]) {
896				snprintf(cmdbuf, cmdlen,
897				    "[%s{%s}]", pp->ki_comm, pp->ki_tdname);
898			} else {
899				snprintf(cmdbuf, cmdlen,
900				    "[%s]", pp->ki_comm);
901			}
902		} else {
903			char *src, *dst, *argbuf;
904			char *cmd;
905			size_t argbuflen;
906			size_t len;
907
908			argbuflen = cmdlen * 4;
909			argbuf = (char *)malloc(argbuflen + 1);
910			if (argbuf == NULL) {
911				warn("malloc(%d)", argbuflen + 1);
912				free(cmdbuf);
913				return NULL;
914			}
915
916			dst = argbuf;
917
918			/* Extract cmd name from argv */
919			cmd = strrchr(*args, '/');
920			if (cmd == NULL)
921				cmd = *args;
922			else
923				cmd++;
924
925			for (; (src = *args++) != NULL; ) {
926				if (*src == '\0')
927					continue;
928				len = (argbuflen - (dst - argbuf) - 1) / 4;
929				strvisx(dst, src,
930				    strlen(src) < len ? strlen(src) : len,
931				    VIS_NL | VIS_CSTYLE);
932				while (*dst != '\0')
933					dst++;
934				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
935					*dst++ = ' '; /* add delimiting space */
936			}
937			if (dst != argbuf && dst[-1] == ' ')
938				dst--;
939			*dst = '\0';
940
941			if (strcmp(cmd, pp->ki_comm) != 0) {
942				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
943				    pp->ki_tdname[0])
944					snprintf(cmdbuf, cmdlen,
945					    "%s (%s){%s}", argbuf, pp->ki_comm,
946					    pp->ki_tdname);
947				else
948					snprintf(cmdbuf, cmdlen,
949					    "%s (%s)", argbuf, pp->ki_comm);
950			} else {
951				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
952				    pp->ki_tdname[0])
953					snprintf(cmdbuf, cmdlen,
954					    "%s{%s}", argbuf, pp->ki_tdname);
955				else
956					strlcpy(cmdbuf, argbuf, cmdlen);
957			}
958			free(argbuf);
959		}
960	}
961
962	if (ps.jail == 0)
963		jid_buf[0] = '\0';
964	else
965		snprintf(jid_buf, sizeof(jid_buf), " %*d",
966		    sizeof(jid_buf) - 3, pp->ki_jid);
967
968	if (displaymode == DISP_IO) {
969		oldp = get_old_proc(pp);
970		if (oldp != NULL) {
971			ru.ru_inblock = RU(pp)->ru_inblock -
972			    RU(oldp)->ru_inblock;
973			ru.ru_oublock = RU(pp)->ru_oublock -
974			    RU(oldp)->ru_oublock;
975			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
976			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
977			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
978			rup = &ru;
979		} else {
980			rup = RU(pp);
981		}
982		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
983		s_tot = total_inblock + total_oublock + total_majflt;
984
985		snprintf(fmt, sizeof(fmt), io_Proc_format,
986		    pp->ki_pid,
987		    jid_buf,
988		    namelength, namelength, (*get_userid)(pp->ki_ruid),
989		    rup->ru_nvcsw,
990		    rup->ru_nivcsw,
991		    rup->ru_inblock,
992		    rup->ru_oublock,
993		    rup->ru_majflt,
994		    p_tot,
995		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
996		    screen_width > cmdlengthdelta ?
997		    screen_width - cmdlengthdelta : 0,
998		    printable(cmdbuf));
999
1000		free(cmdbuf);
1001
1002		return (fmt);
1003	}
1004
1005	/* format this entry */
1006	if (smpmode) {
1007		if (state == SRUN && pp->ki_oncpu != 0xff)
1008			cpu = pp->ki_oncpu;
1009		else
1010			cpu = pp->ki_lastcpu;
1011	} else
1012		cpu = 0;
1013	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1014	if (ps.thread != 0)
1015		thr_buf[0] = '\0';
1016	else
1017		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1018		    sizeof(thr_buf) - 2, pp->ki_numthreads);
1019
1020	snprintf(fmt, sizeof(fmt), proc_fmt,
1021	    pp->ki_pid,
1022	    jid_buf,
1023	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1024	    thr_buf,
1025	    pp->ki_pri.pri_level - PZERO,
1026	    format_nice(pp),
1027	    format_k2(PROCSIZE(pp)),
1028	    format_k2(pagetok(pp->ki_rssize)),
1029	    status,
1030	    cpu,
1031	    format_time(cputime),
1032	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1033	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1034	    printable(cmdbuf));
1035
1036	free(cmdbuf);
1037
1038	/* return the result */
1039	return (fmt);
1040}
1041
1042static void
1043getsysctl(const char *name, void *ptr, size_t len)
1044{
1045	size_t nlen = len;
1046
1047	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1048		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1049		    strerror(errno));
1050		quit(23);
1051	}
1052	if (nlen != len) {
1053		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1054		    name, (unsigned long)len, (unsigned long)nlen);
1055		quit(23);
1056	}
1057}
1058
1059static const char *
1060format_nice(const struct kinfo_proc *pp)
1061{
1062	const char *fifo, *kthread;
1063	int rtpri;
1064	static char nicebuf[4 + 1];
1065
1066	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1067	kthread = (pp->ki_flag & P_KTHREAD) ? "k" : "";
1068	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1069	case PRI_ITHD:
1070		return ("-");
1071	case PRI_REALTIME:
1072		/*
1073		 * XXX: the kernel doesn't tell us the original rtprio and
1074		 * doesn't really know what it was, so to recover it we
1075		 * must be more chummy with the implementation than the
1076		 * implementation is with itself.  pri_user gives a
1077		 * constant "base" priority, but is only initialized
1078		 * properly for user threads.  pri_native gives what the
1079		 * kernel calls the "base" priority, but it isn't constant
1080		 * since it is changed by priority propagation.  pri_native
1081		 * also isn't properly initialized for all threads, but it
1082		 * is properly initialized for kernel realtime and idletime
1083		 * threads.  Thus we use pri_user for the base priority of
1084		 * user threads (it is always correct) and pri_native for
1085		 * the base priority of kernel realtime and idletime threads
1086		 * (there is nothing better, and it is usually correct).
1087		 *
1088		 * The field width and thus the buffer are too small for
1089		 * values like "kr31F", but such values shouldn't occur,
1090		 * and if they do then the tailing "F" is not displayed.
1091		 */
1092		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1093		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1094		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1095		    kthread, rtpri, fifo);
1096		break;
1097	case PRI_TIMESHARE:
1098		if (pp->ki_flag & P_KTHREAD)
1099			return ("-");
1100		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1101		break;
1102	case PRI_IDLE:
1103		/* XXX: as above. */
1104		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
1105		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1106		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1107		    kthread, rtpri, fifo);
1108		break;
1109	default:
1110		return ("?");
1111	}
1112	return (nicebuf);
1113}
1114
1115/* comparison routines for qsort */
1116
1117static int
1118compare_pid(const void *p1, const void *p2)
1119{
1120	const struct kinfo_proc * const *pp1 = p1;
1121	const struct kinfo_proc * const *pp2 = p2;
1122
1123	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1124		abort();
1125
1126	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1127}
1128
1129static int
1130compare_tid(const void *p1, const void *p2)
1131{
1132	const struct kinfo_proc * const *pp1 = p1;
1133	const struct kinfo_proc * const *pp2 = p2;
1134
1135	if ((*pp2)->ki_tid < 0 || (*pp1)->ki_tid < 0)
1136		abort();
1137
1138	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1139}
1140
1141/*
1142 *  proc_compare - comparison function for "qsort"
1143 *	Compares the resource consumption of two processes using five
1144 *	distinct keys.  The keys (in descending order of importance) are:
1145 *	percent cpu, cpu ticks, state, resident set size, total virtual
1146 *	memory usage.  The process states are ordered as follows (from least
1147 *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1148 *	array declaration below maps a process state index into a number
1149 *	that reflects this ordering.
1150 */
1151
1152static int sorted_state[] = {
1153	0,	/* not used		*/
1154	3,	/* sleep		*/
1155	1,	/* ABANDONED (WAIT)	*/
1156	6,	/* run			*/
1157	5,	/* start		*/
1158	2,	/* zombie		*/
1159	4	/* stop			*/
1160};
1161
1162
1163#define ORDERKEY_PCTCPU(a, b) do { \
1164	long diff; \
1165	if (ps.wcpu) \
1166		diff = floor(1.0E6 * weighted_cpu(pctdouble((b)->ki_pctcpu), \
1167		    (b))) - \
1168		    floor(1.0E6 * weighted_cpu(pctdouble((a)->ki_pctcpu), \
1169		    (a))); \
1170	else \
1171		diff = (long)(b)->ki_pctcpu - (long)(a)->ki_pctcpu; \
1172	if (diff != 0) \
1173		return (diff > 0 ? 1 : -1); \
1174} while (0)
1175
1176#define ORDERKEY_CPTICKS(a, b) do { \
1177	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1178	if (diff != 0) \
1179		return (diff > 0 ? 1 : -1); \
1180} while (0)
1181
1182#define ORDERKEY_STATE(a, b) do { \
1183	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1184	if (diff != 0) \
1185		return (diff > 0 ? 1 : -1); \
1186} while (0)
1187
1188#define ORDERKEY_PRIO(a, b) do { \
1189	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1190	if (diff != 0) \
1191		return (diff > 0 ? 1 : -1); \
1192} while (0)
1193
1194#define	ORDERKEY_THREADS(a, b) do { \
1195	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1196	if (diff != 0) \
1197		return (diff > 0 ? 1 : -1); \
1198} while (0)
1199
1200#define ORDERKEY_RSSIZE(a, b) do { \
1201	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1202	if (diff != 0) \
1203		return (diff > 0 ? 1 : -1); \
1204} while (0)
1205
1206#define ORDERKEY_MEM(a, b) do { \
1207	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1208	if (diff != 0) \
1209		return (diff > 0 ? 1 : -1); \
1210} while (0)
1211
1212#define ORDERKEY_JID(a, b) do { \
1213	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1214	if (diff != 0) \
1215		return (diff > 0 ? 1 : -1); \
1216} while (0)
1217
1218/* compare_cpu - the comparison function for sorting by cpu percentage */
1219
1220int
1221#ifdef ORDER
1222compare_cpu(void *arg1, void *arg2)
1223#else
1224proc_compare(void *arg1, void *arg2)
1225#endif
1226{
1227	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1228	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1229
1230	ORDERKEY_PCTCPU(p1, p2);
1231	ORDERKEY_CPTICKS(p1, p2);
1232	ORDERKEY_STATE(p1, p2);
1233	ORDERKEY_PRIO(p1, p2);
1234	ORDERKEY_RSSIZE(p1, p2);
1235	ORDERKEY_MEM(p1, p2);
1236
1237	return (0);
1238}
1239
1240#ifdef ORDER
1241/* "cpu" compare routines */
1242int compare_size(), compare_res(), compare_time(), compare_prio(),
1243    compare_threads();
1244
1245/*
1246 * "io" compare routines.  Context switches aren't i/o, but are displayed
1247 * on the "io" display.
1248 */
1249int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1250    compare_vcsw(), compare_ivcsw();
1251
1252int (*compares[])() = {
1253	compare_cpu,
1254	compare_size,
1255	compare_res,
1256	compare_time,
1257	compare_prio,
1258	compare_threads,
1259	compare_iototal,
1260	compare_ioread,
1261	compare_iowrite,
1262	compare_iofault,
1263	compare_vcsw,
1264	compare_ivcsw,
1265	compare_jid,
1266	NULL
1267};
1268
1269/* compare_size - the comparison function for sorting by total memory usage */
1270
1271int
1272compare_size(void *arg1, void *arg2)
1273{
1274	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1275	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1276
1277	ORDERKEY_MEM(p1, p2);
1278	ORDERKEY_RSSIZE(p1, p2);
1279	ORDERKEY_PCTCPU(p1, p2);
1280	ORDERKEY_CPTICKS(p1, p2);
1281	ORDERKEY_STATE(p1, p2);
1282	ORDERKEY_PRIO(p1, p2);
1283
1284	return (0);
1285}
1286
1287/* compare_res - the comparison function for sorting by resident set size */
1288
1289int
1290compare_res(void *arg1, void *arg2)
1291{
1292	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1293	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1294
1295	ORDERKEY_RSSIZE(p1, p2);
1296	ORDERKEY_MEM(p1, p2);
1297	ORDERKEY_PCTCPU(p1, p2);
1298	ORDERKEY_CPTICKS(p1, p2);
1299	ORDERKEY_STATE(p1, p2);
1300	ORDERKEY_PRIO(p1, p2);
1301
1302	return (0);
1303}
1304
1305/* compare_time - the comparison function for sorting by total cpu time */
1306
1307int
1308compare_time(void *arg1, void *arg2)
1309{
1310	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1311	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1312
1313	ORDERKEY_CPTICKS(p1, p2);
1314	ORDERKEY_PCTCPU(p1, p2);
1315	ORDERKEY_STATE(p1, p2);
1316	ORDERKEY_PRIO(p1, p2);
1317	ORDERKEY_RSSIZE(p1, p2);
1318	ORDERKEY_MEM(p1, p2);
1319
1320	return (0);
1321}
1322
1323/* compare_prio - the comparison function for sorting by priority */
1324
1325int
1326compare_prio(void *arg1, void *arg2)
1327{
1328	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1329	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1330
1331	ORDERKEY_PRIO(p1, p2);
1332	ORDERKEY_CPTICKS(p1, p2);
1333	ORDERKEY_PCTCPU(p1, p2);
1334	ORDERKEY_STATE(p1, p2);
1335	ORDERKEY_RSSIZE(p1, p2);
1336	ORDERKEY_MEM(p1, p2);
1337
1338	return (0);
1339}
1340
1341/* compare_threads - the comparison function for sorting by threads */
1342int
1343compare_threads(void *arg1, void *arg2)
1344{
1345	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1346	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1347
1348	ORDERKEY_THREADS(p1, p2);
1349	ORDERKEY_PCTCPU(p1, p2);
1350	ORDERKEY_CPTICKS(p1, p2);
1351	ORDERKEY_STATE(p1, p2);
1352	ORDERKEY_PRIO(p1, p2);
1353	ORDERKEY_RSSIZE(p1, p2);
1354	ORDERKEY_MEM(p1, p2);
1355
1356	return (0);
1357}
1358
1359/* compare_jid - the comparison function for sorting by jid */
1360static int
1361compare_jid(const void *arg1, const void *arg2)
1362{
1363	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1364	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1365
1366	ORDERKEY_JID(p1, p2);
1367	ORDERKEY_PCTCPU(p1, p2);
1368	ORDERKEY_CPTICKS(p1, p2);
1369	ORDERKEY_STATE(p1, p2);
1370	ORDERKEY_PRIO(p1, p2);
1371	ORDERKEY_RSSIZE(p1, p2);
1372	ORDERKEY_MEM(p1, p2);
1373
1374	return (0);
1375}
1376#endif /* ORDER */
1377
1378/* assorted comparison functions for sorting by i/o */
1379
1380int
1381#ifdef ORDER
1382compare_iototal(void *arg1, void *arg2)
1383#else
1384io_compare(void *arg1, void *arg2)
1385#endif
1386{
1387	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1388	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1389
1390	return (get_io_total(p2) - get_io_total(p1));
1391}
1392
1393#ifdef ORDER
1394int
1395compare_ioread(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, inp1, inp2;
1400
1401	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1402	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1403
1404	return (inp2 - inp1);
1405}
1406
1407int
1408compare_iowrite(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, oup1, oup2;
1413
1414	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1415	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1416
1417	return (oup2 - oup1);
1418}
1419
1420int
1421compare_iofault(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, &flp1, &dummy, &dummy);
1428	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1429
1430	return (flp2 - flp1);
1431}
1432
1433int
1434compare_vcsw(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, &flp1, &dummy);
1441	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1442
1443	return (flp2 - flp1);
1444}
1445
1446int
1447compare_ivcsw(void *arg1, void *arg2)
1448{
1449	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1450	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1451	long dummy, flp1, flp2;
1452
1453	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1454	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1455
1456	return (flp2 - flp1);
1457}
1458#endif /* ORDER */
1459
1460/*
1461 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1462 *		the process does not exist.
1463 *		It is EXTREMELY IMPORTANT that this function work correctly.
1464 *		If top runs setuid root (as in SVR4), then this function
1465 *		is the only thing that stands in the way of a serious
1466 *		security problem.  It validates requests for the "kill"
1467 *		and "renice" commands.
1468 */
1469
1470int
1471proc_owner(int pid)
1472{
1473	int cnt;
1474	struct kinfo_proc **prefp;
1475	struct kinfo_proc *pp;
1476
1477	prefp = pref;
1478	cnt = pref_len;
1479	while (--cnt >= 0) {
1480		pp = *prefp++;
1481		if (pp->ki_pid == (pid_t)pid)
1482			return ((int)pp->ki_ruid);
1483	}
1484	return (-1);
1485}
1486
1487static int
1488swapmode(int *retavail, int *retfree)
1489{
1490	int n;
1491	int pagesize = getpagesize();
1492	struct kvm_swap swapary[1];
1493
1494	*retavail = 0;
1495	*retfree = 0;
1496
1497#define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1498
1499	n = kvm_getswapinfo(kd, swapary, 1, 0);
1500	if (n < 0 || swapary[0].ksw_total == 0)
1501		return (0);
1502
1503	*retavail = CONVERT(swapary[0].ksw_total);
1504	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1505
1506	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1507	return (n);
1508}
1509