machine.c revision 336206
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/11/usr.bin/top/machine.c 336206 2018-07-11 19:39:29Z kib $
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
70/* TOP_JID_LEN based on max of 999999 */
71#define TOP_JID_LEN 7
72#define TOP_SWAP_LEN 6
73static int jidlength;
74static int swaplength;
75static int cmdlengthdelta;
76
77/* Prototypes for top internals */
78void quit(int);
79
80/* get_process_info passes back a handle.  This is what it looks like: */
81
82struct handle {
83	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
84	int remaining;			/* number of pointers remaining */
85};
86
87/* declarations for load_avg */
88#include "loadavg.h"
89
90/* define what weighted cpu is.  */
91#define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
92			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
93
94/* what we consider to be process size: */
95#define PROCSIZE(pp) ((pp)->ki_size / 1024)
96
97#define RU(pp)	(&(pp)->ki_rusage)
98#define RUTOT(pp) \
99	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
100
101#define	PCTCPU(pp) (pcpu[pp - pbase])
102
103/* definitions for indices in the nlist array */
104
105/*
106 *  These definitions control the format of the per-process area
107 */
108
109static char io_header[] =
110    "  PID%*s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
111
112#define io_Proc_format \
113    "%5d%*s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s"
114
115static char smp_header_thr[] =
116    "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
117static char smp_header[] =
118    "  PID%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
119
120#define smp_Proc_format \
121    "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s %2d%7s %6.2f%% %.*s"
122
123static char up_header_thr[] =
124    "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
125static char up_header[] =
126    "  PID%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
127
128#define up_Proc_format \
129    "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s%.0d%7s %6.2f%% %.*s"
130
131
132/* process state names for the "STATE" column of the display */
133/* the extra nulls in the string "run" are for adding a slash and
134   the processor number when needed */
135
136char *state_abbrev[] = {
137	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
138};
139
140
141static kvm_t *kd;
142
143/* values that we stash away in _init and use in later routines */
144
145static double logcpu;
146
147/* these are retrieved from the kernel in _init */
148
149static load_avg  ccpu;
150
151/* these are used in the get_ functions */
152
153static int lastpid;
154
155/* these are for calculating cpu state percentages */
156
157static long cp_time[CPUSTATES];
158static long cp_old[CPUSTATES];
159static long cp_diff[CPUSTATES];
160
161/* these are for detailing the process states */
162
163int process_states[8];
164char *procstatenames[] = {
165	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
166	" zombie, ", " waiting, ", " lock, ",
167	NULL
168};
169
170/* these are for detailing the cpu states */
171
172int cpu_states[CPUSTATES];
173char *cpustatenames[] = {
174	"user", "nice", "system", "interrupt", "idle", NULL
175};
176
177/* these are for detailing the memory statistics */
178
179int memory_stats[7];
180char *memorynames[] = {
181	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
182	"K Free", NULL
183};
184
185int arc_stats[7];
186char *arcnames[] = {
187	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
188	NULL
189};
190
191int carc_stats[4];
192char *carcnames[] = {
193	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
194	NULL
195};
196
197int swap_stats[7];
198char *swapnames[] = {
199	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
200	NULL
201};
202
203
204/* these are for keeping track of the proc array */
205
206static int nproc;
207static int onproc = -1;
208static int pref_len;
209static struct kinfo_proc *pbase;
210static struct kinfo_proc **pref;
211static struct kinfo_proc *previous_procs;
212static struct kinfo_proc **previous_pref;
213static int previous_proc_count = 0;
214static int previous_proc_count_max = 0;
215static int previous_thread;
216
217/* data used for recalculating pctcpu */
218static double *pcpu;
219static struct timespec proc_uptime;
220static struct timeval proc_wall_time;
221static struct timeval previous_wall_time;
222static uint64_t previous_interval = 0;
223
224/* total number of io operations */
225static long total_inblock;
226static long total_oublock;
227static long total_majflt;
228
229/* these are for getting the memory statistics */
230
231static int arc_enabled;
232static int carc_enabled;
233static int pageshift;		/* log base 2 of the pagesize */
234
235/* define pagetok in terms of pageshift */
236
237#define pagetok(size) ((size) << pageshift)
238
239/* swap usage */
240#define ki_swap(kip) \
241    ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
242
243/* useful externals */
244long percentages();
245
246#ifdef ORDER
247/*
248 * Sorting orders.  The first element is the default.
249 */
250char *ordernames[] = {
251	"cpu", "size", "res", "time", "pri", "threads",
252	"total", "read", "write", "fault", "vcsw", "ivcsw",
253	"jid", "swap", "pid", NULL
254};
255#endif
256
257/* Per-cpu time states */
258static int maxcpu;
259static int maxid;
260static int ncpus;
261static u_long cpumask;
262static long *times;
263static long *pcpu_cp_time;
264static long *pcpu_cp_old;
265static long *pcpu_cp_diff;
266static int *pcpu_cpu_states;
267
268static int compare_swap(const void *a, const void *b);
269static int compare_jid(const void *a, const void *b);
270static int compare_pid(const void *a, const void *b);
271static int compare_tid(const void *a, const void *b);
272static const char *format_nice(const struct kinfo_proc *pp);
273static void getsysctl(const char *name, void *ptr, size_t len);
274static int swapmode(int *retavail, int *retfree);
275static void update_layout(void);
276static int find_uid(uid_t needle, int *haystack);
277
278static int
279find_uid(uid_t needle, int *haystack)
280{
281	size_t i = 0;
282
283	for (; i < TOP_MAX_UIDS; ++i)
284		if ((uid_t)haystack[i] == needle)
285			return 1;
286	return 0;
287}
288
289void
290toggle_pcpustats(void)
291{
292
293	if (ncpus == 1)
294		return;
295	update_layout();
296}
297
298/* Adjust display based on ncpus and the ARC state. */
299static void
300update_layout(void)
301{
302
303	y_mem = 3;
304	y_arc = 4;
305	y_carc = 5;
306	y_swap = 4 + arc_enabled + carc_enabled;
307	y_idlecursor = 5 + arc_enabled + carc_enabled;
308	y_message = 5 + arc_enabled + carc_enabled;
309	y_header = 6 + arc_enabled + carc_enabled;
310	y_procs = 7 + arc_enabled + carc_enabled;
311	Header_lines = 7 + arc_enabled + carc_enabled;
312
313	if (pcpu_stats) {
314		y_mem += ncpus - 1;
315		y_arc += ncpus - 1;
316		y_carc += ncpus - 1;
317		y_swap += ncpus - 1;
318		y_idlecursor += ncpus - 1;
319		y_message += ncpus - 1;
320		y_header += ncpus - 1;
321		y_procs += ncpus - 1;
322		Header_lines += ncpus - 1;
323	}
324}
325
326int
327machine_init(struct statics *statics, char do_unames)
328{
329	int i, j, empty, pagesize;
330	uint64_t arc_size;
331	boolean_t carc_en;
332	size_t size;
333	struct passwd *pw;
334
335	size = sizeof(smpmode);
336	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
337	    NULL, 0) != 0 &&
338	    sysctlbyname("kern.smp.active", &smpmode, &size,
339	    NULL, 0) != 0) ||
340	    size != sizeof(smpmode))
341		smpmode = 0;
342
343	size = sizeof(arc_size);
344	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
345	    NULL, 0) == 0 && arc_size != 0)
346		arc_enabled = 1;
347	size = sizeof(carc_en);
348	if (arc_enabled &&
349	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
350	    NULL, 0) == 0 && carc_en == 1)
351		carc_enabled = 1;
352
353	if (do_unames) {
354	    while ((pw = getpwent()) != NULL) {
355		if (strlen(pw->pw_name) > namelength)
356			namelength = strlen(pw->pw_name);
357	    }
358	}
359	if (smpmode && namelength > SMPUNAMELEN)
360		namelength = SMPUNAMELEN;
361	else if (namelength > UPUNAMELEN)
362		namelength = UPUNAMELEN;
363
364	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
365	if (kd == NULL)
366		return (-1);
367
368	GETSYSCTL("kern.ccpu", ccpu);
369
370	/* this is used in calculating WCPU -- calculate it ahead of time */
371	logcpu = log(loaddouble(ccpu));
372
373	pbase = NULL;
374	pref = NULL;
375	pcpu = NULL;
376	nproc = 0;
377	onproc = -1;
378
379	/* get the page size and calculate pageshift from it */
380	pagesize = getpagesize();
381	pageshift = 0;
382	while (pagesize > 1) {
383		pageshift++;
384		pagesize >>= 1;
385	}
386
387	/* we only need the amount of log(2)1024 for our conversion */
388	pageshift -= LOG1024;
389
390	/* fill in the statics information */
391	statics->procstate_names = procstatenames;
392	statics->cpustate_names = cpustatenames;
393	statics->memory_names = memorynames;
394	if (arc_enabled)
395		statics->arc_names = arcnames;
396	else
397		statics->arc_names = NULL;
398	if (carc_enabled)
399		statics->carc_names = carcnames;
400	else
401		statics->carc_names = NULL;
402	statics->swap_names = swapnames;
403#ifdef ORDER
404	statics->order_names = ordernames;
405#endif
406
407	/* Allocate state for per-CPU stats. */
408	cpumask = 0;
409	ncpus = 0;
410	GETSYSCTL("kern.smp.maxcpus", maxcpu);
411	size = sizeof(long) * maxcpu * CPUSTATES;
412	times = malloc(size);
413	if (times == NULL)
414		err(1, "malloc %zu bytes", size);
415	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
416		err(1, "sysctlbyname kern.cp_times");
417	pcpu_cp_time = calloc(1, size);
418	maxid = (size / CPUSTATES / sizeof(long)) - 1;
419	for (i = 0; i <= maxid; i++) {
420		empty = 1;
421		for (j = 0; empty && j < CPUSTATES; j++) {
422			if (times[i * CPUSTATES + j] != 0)
423				empty = 0;
424		}
425		if (!empty) {
426			cpumask |= (1ul << i);
427			ncpus++;
428		}
429	}
430	size = sizeof(long) * ncpus * CPUSTATES;
431	pcpu_cp_old = calloc(1, size);
432	pcpu_cp_diff = calloc(1, size);
433	pcpu_cpu_states = calloc(1, size);
434	statics->ncpus = ncpus;
435
436	update_layout();
437
438	/* all done! */
439	return (0);
440}
441
442char *
443format_header(char *uname_field)
444{
445	static char Header[128];
446	const char *prehead;
447
448	if (ps.jail)
449		jidlength = TOP_JID_LEN + 1;	/* +1 for extra left space. */
450	else
451		jidlength = 0;
452
453	if (ps.swap)
454		swaplength = TOP_SWAP_LEN + 1;  /* +1 for extra left space */
455	else
456		swaplength = 0;
457
458	switch (displaymode) {
459	case DISP_CPU:
460		/*
461		 * The logic of picking the right header format seems reverse
462		 * here because we only want to display a THR column when
463		 * "thread mode" is off (and threads are not listed as
464		 * separate lines).
465		 */
466		prehead = smpmode ?
467		    (ps.thread ? smp_header : smp_header_thr) :
468		    (ps.thread ? up_header : up_header_thr);
469		snprintf(Header, sizeof(Header), prehead,
470		    jidlength, ps.jail ? " JID" : "",
471		    namelength, namelength, uname_field,
472		    swaplength, ps.swap ? " SWAP" : "",
473		    ps.wcpu ? "WCPU" : "CPU");
474		break;
475	case DISP_IO:
476		prehead = io_header;
477		snprintf(Header, sizeof(Header), prehead,
478		    jidlength, ps.jail ? " JID" : "",
479		    namelength, namelength, uname_field);
480		break;
481	}
482	cmdlengthdelta = strlen(Header) - 7;
483	return (Header);
484}
485
486static int swappgsin = -1;
487static int swappgsout = -1;
488extern struct timeval timeout;
489
490
491void
492get_system_info(struct system_info *si)
493{
494	long total;
495	struct loadavg sysload;
496	int mib[2];
497	struct timeval boottime;
498	uint64_t arc_stat, arc_stat2;
499	int i, j;
500	size_t size;
501
502	/* get the CPU stats */
503	size = (maxid + 1) * CPUSTATES * sizeof(long);
504	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
505		err(1, "sysctlbyname kern.cp_times");
506	GETSYSCTL("kern.cp_time", cp_time);
507	GETSYSCTL("vm.loadavg", sysload);
508	GETSYSCTL("kern.lastpid", lastpid);
509
510	/* convert load averages to doubles */
511	for (i = 0; i < 3; i++)
512		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
513
514	/* convert cp_time counts to percentages */
515	for (i = j = 0; i <= maxid; i++) {
516		if ((cpumask & (1ul << i)) == 0)
517			continue;
518		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
519		    &pcpu_cp_time[j * CPUSTATES],
520		    &pcpu_cp_old[j * CPUSTATES],
521		    &pcpu_cp_diff[j * CPUSTATES]);
522		j++;
523	}
524	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
525
526	/* sum memory & swap statistics */
527	{
528		static unsigned int swap_delay = 0;
529		static int swapavail = 0;
530		static int swapfree = 0;
531		static long bufspace = 0;
532		static int nspgsin, nspgsout;
533
534		GETSYSCTL("vfs.bufspace", bufspace);
535		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
536		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
537		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
538		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
539		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
540		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
541		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
542		/* convert memory stats to Kbytes */
543		memory_stats[0] = pagetok(memory_stats[0]);
544		memory_stats[1] = pagetok(memory_stats[1]);
545		memory_stats[2] = pagetok(memory_stats[2]);
546		memory_stats[3] = pagetok(memory_stats[3]);
547		memory_stats[4] = bufspace / 1024;
548		memory_stats[5] = pagetok(memory_stats[5]);
549		memory_stats[6] = -1;
550
551		/* first interval */
552		if (swappgsin < 0) {
553			swap_stats[4] = 0;
554			swap_stats[5] = 0;
555		}
556
557		/* compute differences between old and new swap statistic */
558		else {
559			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
560			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
561		}
562
563		swappgsin = nspgsin;
564		swappgsout = nspgsout;
565
566		/* call CPU heavy swapmode() only for changes */
567		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
568			swap_stats[3] = swapmode(&swapavail, &swapfree);
569			swap_stats[0] = swapavail;
570			swap_stats[1] = swapavail - swapfree;
571			swap_stats[2] = swapfree;
572		}
573		swap_delay = 1;
574		swap_stats[6] = -1;
575	}
576
577	if (arc_enabled) {
578		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
579		arc_stats[0] = arc_stat >> 10;
580		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
581		arc_stats[1] = arc_stat >> 10;
582		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
583		arc_stats[2] = arc_stat >> 10;
584		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
585		arc_stats[3] = arc_stat >> 10;
586		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
587		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
588		arc_stats[4] = arc_stat + arc_stat2 >> 10;
589		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
590		arc_stats[5] = arc_stat >> 10;
591		si->arc = arc_stats;
592	}
593	if (carc_enabled) {
594		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
595		carc_stats[0] = arc_stat >> 10;
596		carc_stats[2] = arc_stat >> 10; /* For ratio */
597		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
598		carc_stats[1] = arc_stat >> 10;
599		si->carc = carc_stats;
600	}
601
602	/* set arrays and strings */
603	if (pcpu_stats) {
604		si->cpustates = pcpu_cpu_states;
605		si->ncpus = ncpus;
606	} else {
607		si->cpustates = cpu_states;
608		si->ncpus = 1;
609	}
610	si->memory = memory_stats;
611	si->swap = swap_stats;
612
613
614	if (lastpid > 0) {
615		si->last_pid = lastpid;
616	} else {
617		si->last_pid = -1;
618	}
619
620	/*
621	 * Print how long system has been up.
622	 * (Found by looking getting "boottime" from the kernel)
623	 */
624	mib[0] = CTL_KERN;
625	mib[1] = KERN_BOOTTIME;
626	size = sizeof(boottime);
627	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
628	    boottime.tv_sec != 0) {
629		si->boottime = boottime;
630	} else {
631		si->boottime.tv_sec = -1;
632	}
633}
634
635#define NOPROC	((void *)-1)
636
637/*
638 * We need to compare data from the old process entry with the new
639 * process entry.
640 * To facilitate doing this quickly we stash a pointer in the kinfo_proc
641 * structure to cache the mapping.  We also use a negative cache pointer
642 * of NOPROC to avoid duplicate lookups.
643 * XXX: this could be done when the actual processes are fetched, we do
644 * it here out of laziness.
645 */
646const struct kinfo_proc *
647get_old_proc(struct kinfo_proc *pp)
648{
649	struct kinfo_proc **oldpp, *oldp;
650
651	/*
652	 * If this is the first fetch of the kinfo_procs then we don't have
653	 * any previous entries.
654	 */
655	if (previous_proc_count == 0)
656		return (NULL);
657	/* negative cache? */
658	if (pp->ki_udata == NOPROC)
659		return (NULL);
660	/* cached? */
661	if (pp->ki_udata != NULL)
662		return (pp->ki_udata);
663	/*
664	 * Not cached,
665	 * 1) look up based on pid.
666	 * 2) compare process start.
667	 * If we fail here, then setup a negative cache entry, otherwise
668	 * cache it.
669	 */
670	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
671	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
672	if (oldpp == NULL) {
673		pp->ki_udata = NOPROC;
674		return (NULL);
675	}
676	oldp = *oldpp;
677	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
678		pp->ki_udata = NOPROC;
679		return (NULL);
680	}
681	pp->ki_udata = oldp;
682	return (oldp);
683}
684
685/*
686 * Return the total amount of IO done in blocks in/out and faults.
687 * store the values individually in the pointers passed in.
688 */
689long
690get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
691    long *vcsw, long *ivcsw)
692{
693	const struct kinfo_proc *oldp;
694	static struct kinfo_proc dummy;
695	long ret;
696
697	oldp = get_old_proc(pp);
698	if (oldp == NULL) {
699		bzero(&dummy, sizeof(dummy));
700		oldp = &dummy;
701	}
702	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
703	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
704	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
705	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
706	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
707	ret =
708	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
709	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
710	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
711	return (ret);
712}
713
714/*
715 * If there was a previous update, use the delta in ki_runtime over
716 * the previous interval to calculate pctcpu.  Otherwise, fall back
717 * to using the kernel's ki_pctcpu.
718 */
719static double
720proc_calc_pctcpu(struct kinfo_proc *pp)
721{
722	const struct kinfo_proc *oldp;
723
724	if (previous_interval != 0) {
725		oldp = get_old_proc(pp);
726		if (oldp != NULL)
727			return ((double)(pp->ki_runtime - oldp->ki_runtime)
728			    / previous_interval);
729
730		/*
731		 * If this process/thread was created during the previous
732		 * interval, charge it's total runtime to the previous
733		 * interval.
734		 */
735		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
736		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
737		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
738			return ((double)pp->ki_runtime / previous_interval);
739	}
740	return (pctdouble(pp->ki_pctcpu));
741}
742
743/*
744 * Return true if this process has used any CPU time since the
745 * previous update.
746 */
747static int
748proc_used_cpu(struct kinfo_proc *pp)
749{
750	const struct kinfo_proc *oldp;
751
752	oldp = get_old_proc(pp);
753	if (oldp == NULL)
754		return (PCTCPU(pp) != 0);
755	return (pp->ki_runtime != oldp->ki_runtime ||
756	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
757	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
758}
759
760/*
761 * Return the total number of block in/out and faults by a process.
762 */
763long
764get_io_total(struct kinfo_proc *pp)
765{
766	long dummy;
767
768	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
769}
770
771static struct handle handle;
772
773caddr_t
774get_process_info(struct system_info *si, struct process_select *sel,
775    int (*compare)(const void *, const void *))
776{
777	int i;
778	int total_procs;
779	long p_io;
780	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
781	long nsec;
782	int active_procs;
783	struct kinfo_proc **prefp;
784	struct kinfo_proc *pp;
785	struct timespec previous_proc_uptime;
786
787	/* these are copied out of sel for speed */
788	int show_idle;
789	int show_jid;
790	int show_self;
791	int show_system;
792	int show_uid;
793	int show_command;
794	int show_kidle;
795
796	/*
797	 * If thread state was toggled, don't cache the previous processes.
798	 */
799	if (previous_thread != sel->thread)
800		nproc = 0;
801	previous_thread = sel->thread;
802
803	/*
804	 * Save the previous process info.
805	 */
806	if (previous_proc_count_max < nproc) {
807		free(previous_procs);
808		previous_procs = malloc(nproc * sizeof(*previous_procs));
809		free(previous_pref);
810		previous_pref = malloc(nproc * sizeof(*previous_pref));
811		if (previous_procs == NULL || previous_pref == NULL) {
812			(void) fprintf(stderr, "top: Out of memory.\n");
813			quit(23);
814		}
815		previous_proc_count_max = nproc;
816	}
817	if (nproc) {
818		for (i = 0; i < nproc; i++)
819			previous_pref[i] = &previous_procs[i];
820		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
821		qsort(previous_pref, nproc, sizeof(*previous_pref),
822		    ps.thread ? compare_tid : compare_pid);
823	}
824	previous_proc_count = nproc;
825	previous_proc_uptime = proc_uptime;
826	previous_wall_time = proc_wall_time;
827	previous_interval = 0;
828
829	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
830	    0, &nproc);
831	(void)gettimeofday(&proc_wall_time, NULL);
832	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
833		memset(&proc_uptime, 0, sizeof(proc_uptime));
834	else if (previous_proc_uptime.tv_sec != 0 &&
835	    previous_proc_uptime.tv_nsec != 0) {
836		previous_interval = (proc_uptime.tv_sec -
837		    previous_proc_uptime.tv_sec) * 1000000;
838		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
839		if (nsec < 0) {
840			previous_interval -= 1000000;
841			nsec += 1000000000;
842		}
843		previous_interval += nsec / 1000;
844	}
845	if (nproc > onproc) {
846		pref = realloc(pref, sizeof(*pref) * nproc);
847		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
848		onproc = nproc;
849	}
850	if (pref == NULL || pbase == NULL || pcpu == NULL) {
851		(void) fprintf(stderr, "top: Out of memory.\n");
852		quit(23);
853	}
854	/* get a pointer to the states summary array */
855	si->procstates = process_states;
856
857	/* set up flags which define what we are going to select */
858	show_idle = sel->idle;
859	show_jid = sel->jid != -1;
860	show_self = sel->self == -1;
861	show_system = sel->system;
862	show_uid = sel->uid[0] != -1;
863	show_command = sel->command != NULL;
864	show_kidle = sel->kidle;
865
866	/* count up process states and get pointers to interesting procs */
867	total_procs = 0;
868	active_procs = 0;
869	total_inblock = 0;
870	total_oublock = 0;
871	total_majflt = 0;
872	memset((char *)process_states, 0, sizeof(process_states));
873	prefp = pref;
874	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
875
876		if (pp->ki_stat == 0)
877			/* not in use */
878			continue;
879
880		if (!show_self && pp->ki_pid == sel->self)
881			/* skip self */
882			continue;
883
884		if (!show_system && (pp->ki_flag & P_SYSTEM))
885			/* skip system process */
886			continue;
887
888		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
889		    &p_vcsw, &p_ivcsw);
890		total_inblock += p_inblock;
891		total_oublock += p_oublock;
892		total_majflt += p_majflt;
893		total_procs++;
894		process_states[pp->ki_stat]++;
895
896		if (pp->ki_stat == SZOMB)
897			/* skip zombies */
898			continue;
899
900		if (!show_kidle && pp->ki_tdflags & TDF_IDLETD)
901			/* skip kernel idle process */
902			continue;
903
904		PCTCPU(pp) = proc_calc_pctcpu(pp);
905		if (sel->thread && PCTCPU(pp) > 1.0)
906			PCTCPU(pp) = 1.0;
907		if (displaymode == DISP_CPU && !show_idle &&
908		    (!proc_used_cpu(pp) ||
909		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
910			/* skip idle or non-running processes */
911			continue;
912
913		if (displaymode == DISP_IO && !show_idle && p_io == 0)
914			/* skip processes that aren't doing I/O */
915			continue;
916
917		if (show_jid && pp->ki_jid != sel->jid)
918			/* skip proc. that don't belong to the selected JID */
919			continue;
920
921		if (show_uid && !find_uid(pp->ki_ruid, sel->uid))
922			/* skip proc. that don't belong to the selected UID */
923			continue;
924
925		*prefp++ = pp;
926		active_procs++;
927	}
928
929	/* if requested, sort the "interesting" processes */
930	if (compare != NULL)
931		qsort(pref, active_procs, sizeof(*pref), compare);
932
933	/* remember active and total counts */
934	si->p_total = total_procs;
935	si->p_active = pref_len = active_procs;
936
937	/* pass back a handle */
938	handle.next_proc = pref;
939	handle.remaining = active_procs;
940	return ((caddr_t)&handle);
941}
942
943static char fmt[512];	/* static area where result is built */
944
945char *
946format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
947{
948	struct kinfo_proc *pp;
949	const struct kinfo_proc *oldp;
950	long cputime;
951	double pct;
952	struct handle *hp;
953	char status[16];
954	int cpu, state;
955	struct rusage ru, *rup;
956	long p_tot, s_tot;
957	char *proc_fmt, thr_buf[6];
958	char jid_buf[TOP_JID_LEN + 1], swap_buf[TOP_SWAP_LEN + 1];
959	char *cmdbuf = NULL;
960	char **args;
961	const int cmdlen = 128;
962
963	/* find and remember the next proc structure */
964	hp = (struct handle *)handle;
965	pp = *(hp->next_proc++);
966	hp->remaining--;
967
968	/* get the process's command name */
969	if ((pp->ki_flag & P_INMEM) == 0) {
970		/*
971		 * Print swapped processes as <pname>
972		 */
973		size_t len;
974
975		len = strlen(pp->ki_comm);
976		if (len > sizeof(pp->ki_comm) - 3)
977			len = sizeof(pp->ki_comm) - 3;
978		memmove(pp->ki_comm + 1, pp->ki_comm, len);
979		pp->ki_comm[0] = '<';
980		pp->ki_comm[len + 1] = '>';
981		pp->ki_comm[len + 2] = '\0';
982	}
983
984	/*
985	 * Convert the process's runtime from microseconds to seconds.  This
986	 * time includes the interrupt time although that is not wanted here.
987	 * ps(1) is similarly sloppy.
988	 */
989	cputime = (pp->ki_runtime + 500000) / 1000000;
990
991	/* calculate the base for cpu percentages */
992	pct = PCTCPU(pp);
993
994	/* generate "STATE" field */
995	switch (state = pp->ki_stat) {
996	case SRUN:
997		if (smpmode && pp->ki_oncpu != NOCPU)
998			sprintf(status, "CPU%d", pp->ki_oncpu);
999		else
1000			strcpy(status, "RUN");
1001		break;
1002	case SLOCK:
1003		if (pp->ki_kiflag & KI_LOCKBLOCK) {
1004			sprintf(status, "*%.6s", pp->ki_lockname);
1005			break;
1006		}
1007		/* fall through */
1008	case SSLEEP:
1009		if (pp->ki_wmesg != NULL) {
1010			sprintf(status, "%.6s", pp->ki_wmesg);
1011			break;
1012		}
1013		/* FALLTHROUGH */
1014	default:
1015
1016		if (state >= 0 &&
1017		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
1018			sprintf(status, "%.6s", state_abbrev[state]);
1019		else
1020			sprintf(status, "?%5d", state);
1021		break;
1022	}
1023
1024	cmdbuf = (char *)malloc(cmdlen + 1);
1025	if (cmdbuf == NULL) {
1026		warn("malloc(%d)", cmdlen + 1);
1027		return NULL;
1028	}
1029
1030	if (!(flags & FMT_SHOWARGS)) {
1031		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1032		    pp->ki_tdname[0]) {
1033			snprintf(cmdbuf, cmdlen, "%s{%s%s}", pp->ki_comm,
1034			    pp->ki_tdname, pp->ki_moretdname);
1035		} else {
1036			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
1037		}
1038	} else {
1039		if (pp->ki_flag & P_SYSTEM ||
1040		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
1041		    !(*args)) {
1042			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1043		    	    pp->ki_tdname[0]) {
1044				snprintf(cmdbuf, cmdlen,
1045				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
1046				    pp->ki_moretdname);
1047			} else {
1048				snprintf(cmdbuf, cmdlen,
1049				    "[%s]", pp->ki_comm);
1050			}
1051		} else {
1052			char *src, *dst, *argbuf;
1053			char *cmd;
1054			size_t argbuflen;
1055			size_t len;
1056
1057			argbuflen = cmdlen * 4;
1058			argbuf = (char *)malloc(argbuflen + 1);
1059			if (argbuf == NULL) {
1060				warn("malloc(%zu)", argbuflen + 1);
1061				free(cmdbuf);
1062				return NULL;
1063			}
1064
1065			dst = argbuf;
1066
1067			/* Extract cmd name from argv */
1068			cmd = strrchr(*args, '/');
1069			if (cmd == NULL)
1070				cmd = *args;
1071			else
1072				cmd++;
1073
1074			for (; (src = *args++) != NULL; ) {
1075				if (*src == '\0')
1076					continue;
1077				len = (argbuflen - (dst - argbuf) - 1) / 4;
1078				strvisx(dst, src,
1079				    MIN(strlen(src), len),
1080				    VIS_NL | VIS_CSTYLE);
1081				while (*dst != '\0')
1082					dst++;
1083				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1084					*dst++ = ' '; /* add delimiting space */
1085			}
1086			if (dst != argbuf && dst[-1] == ' ')
1087				dst--;
1088			*dst = '\0';
1089
1090			if (strcmp(cmd, pp->ki_comm) != 0) {
1091				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1092				    pp->ki_tdname[0])
1093					snprintf(cmdbuf, cmdlen,
1094					    "%s (%s){%s%s}", argbuf,
1095					    pp->ki_comm, pp->ki_tdname,
1096					    pp->ki_moretdname);
1097				else
1098					snprintf(cmdbuf, cmdlen,
1099					    "%s (%s)", argbuf, pp->ki_comm);
1100			} else {
1101				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1102				    pp->ki_tdname[0])
1103					snprintf(cmdbuf, cmdlen,
1104					    "%s{%s%s}", argbuf, pp->ki_tdname,
1105					    pp->ki_moretdname);
1106				else
1107					strlcpy(cmdbuf, argbuf, cmdlen);
1108			}
1109			free(argbuf);
1110		}
1111	}
1112
1113	if (ps.jail == 0)
1114		jid_buf[0] = '\0';
1115	else
1116		snprintf(jid_buf, sizeof(jid_buf), "%*d",
1117		    jidlength - 1, pp->ki_jid);
1118
1119	if (ps.swap == 0)
1120		swap_buf[0] = '\0';
1121	else
1122		snprintf(swap_buf, sizeof(swap_buf), "%*s",
1123		    swaplength - 1,
1124		    format_k2(pagetok(ki_swap(pp)))); /* XXX */
1125
1126	if (displaymode == DISP_IO) {
1127		oldp = get_old_proc(pp);
1128		if (oldp != NULL) {
1129			ru.ru_inblock = RU(pp)->ru_inblock -
1130			    RU(oldp)->ru_inblock;
1131			ru.ru_oublock = RU(pp)->ru_oublock -
1132			    RU(oldp)->ru_oublock;
1133			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1134			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1135			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1136			rup = &ru;
1137		} else {
1138			rup = RU(pp);
1139		}
1140		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1141		s_tot = total_inblock + total_oublock + total_majflt;
1142
1143		snprintf(fmt, sizeof(fmt), io_Proc_format,
1144		    pp->ki_pid,
1145		    jidlength, jid_buf,
1146		    namelength, namelength, (*get_userid)(pp->ki_ruid),
1147		    rup->ru_nvcsw,
1148		    rup->ru_nivcsw,
1149		    rup->ru_inblock,
1150		    rup->ru_oublock,
1151		    rup->ru_majflt,
1152		    p_tot,
1153		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
1154		    screen_width > cmdlengthdelta ?
1155		    screen_width - cmdlengthdelta : 0,
1156		    printable(cmdbuf));
1157
1158		free(cmdbuf);
1159
1160		return (fmt);
1161	}
1162
1163	/* format this entry */
1164	if (smpmode) {
1165		if (state == SRUN && pp->ki_oncpu != NOCPU)
1166			cpu = pp->ki_oncpu;
1167		else
1168			cpu = pp->ki_lastcpu;
1169	} else
1170		cpu = 0;
1171	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1172	if (ps.thread != 0)
1173		thr_buf[0] = '\0';
1174	else
1175		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1176		    (int)(sizeof(thr_buf) - 2), pp->ki_numthreads);
1177
1178	snprintf(fmt, sizeof(fmt), proc_fmt,
1179	    pp->ki_pid,
1180	    jidlength, jid_buf,
1181	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1182	    thr_buf,
1183	    pp->ki_pri.pri_level - PZERO,
1184	    format_nice(pp),
1185	    format_k2(PROCSIZE(pp)),
1186	    format_k2(pagetok(pp->ki_rssize)),
1187	    swaplength, swaplength, swap_buf,
1188	    status,
1189	    cpu,
1190	    format_time(cputime),
1191	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1192	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1193	    printable(cmdbuf));
1194
1195	free(cmdbuf);
1196
1197	/* return the result */
1198	return (fmt);
1199}
1200
1201static void
1202getsysctl(const char *name, void *ptr, size_t len)
1203{
1204	size_t nlen = len;
1205
1206	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1207		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1208		    strerror(errno));
1209		quit(23);
1210	}
1211	if (nlen != len) {
1212		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1213		    name, (unsigned long)len, (unsigned long)nlen);
1214		quit(23);
1215	}
1216}
1217
1218static const char *
1219format_nice(const struct kinfo_proc *pp)
1220{
1221	const char *fifo, *kproc;
1222	int rtpri;
1223	static char nicebuf[4 + 1];
1224
1225	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1226	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1227	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1228	case PRI_ITHD:
1229		return ("-");
1230	case PRI_REALTIME:
1231		/*
1232		 * XXX: the kernel doesn't tell us the original rtprio and
1233		 * doesn't really know what it was, so to recover it we
1234		 * must be more chummy with the implementation than the
1235		 * implementation is with itself.  pri_user gives a
1236		 * constant "base" priority, but is only initialized
1237		 * properly for user threads.  pri_native gives what the
1238		 * kernel calls the "base" priority, but it isn't constant
1239		 * since it is changed by priority propagation.  pri_native
1240		 * also isn't properly initialized for all threads, but it
1241		 * is properly initialized for kernel realtime and idletime
1242		 * threads.  Thus we use pri_user for the base priority of
1243		 * user threads (it is always correct) and pri_native for
1244		 * the base priority of kernel realtime and idletime threads
1245		 * (there is nothing better, and it is usually correct).
1246		 *
1247		 * The field width and thus the buffer are too small for
1248		 * values like "kr31F", but such values shouldn't occur,
1249		 * and if they do then the tailing "F" is not displayed.
1250		 */
1251		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1252		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1253		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1254		    kproc, rtpri, fifo);
1255		break;
1256	case PRI_TIMESHARE:
1257		if (pp->ki_flag & P_KPROC)
1258			return ("-");
1259		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1260		break;
1261	case PRI_IDLE:
1262		/* XXX: as above. */
1263		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1264		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1265		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1266		    kproc, rtpri, fifo);
1267		break;
1268	default:
1269		return ("?");
1270	}
1271	return (nicebuf);
1272}
1273
1274/* comparison routines for qsort */
1275
1276static int
1277compare_pid(const void *p1, const void *p2)
1278{
1279	const struct kinfo_proc * const *pp1 = p1;
1280	const struct kinfo_proc * const *pp2 = p2;
1281
1282	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
1283		abort();
1284
1285	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1286}
1287
1288static int
1289compare_tid(const void *p1, const void *p2)
1290{
1291	const struct kinfo_proc * const *pp1 = p1;
1292	const struct kinfo_proc * const *pp2 = p2;
1293
1294	if ((*pp2)->ki_tid < 0 || (*pp1)->ki_tid < 0)
1295		abort();
1296
1297	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1298}
1299
1300/*
1301 *  proc_compare - comparison function for "qsort"
1302 *	Compares the resource consumption of two processes using five
1303 *	distinct keys.  The keys (in descending order of importance) are:
1304 *	percent cpu, cpu ticks, state, resident set size, total virtual
1305 *	memory usage.  The process states are ordered as follows (from least
1306 *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1307 *	array declaration below maps a process state index into a number
1308 *	that reflects this ordering.
1309 */
1310
1311static int sorted_state[] = {
1312	0,	/* not used		*/
1313	3,	/* sleep		*/
1314	1,	/* ABANDONED (WAIT)	*/
1315	6,	/* run			*/
1316	5,	/* start		*/
1317	2,	/* zombie		*/
1318	4	/* stop			*/
1319};
1320
1321
1322#define ORDERKEY_PCTCPU(a, b) do { \
1323	double diff; \
1324	if (ps.wcpu) \
1325		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1326		    weighted_cpu(PCTCPU((a)), (a)); \
1327	else \
1328		diff = PCTCPU((b)) - PCTCPU((a)); \
1329	if (diff != 0) \
1330		return (diff > 0 ? 1 : -1); \
1331} while (0)
1332
1333#define ORDERKEY_CPTICKS(a, b) do { \
1334	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1335	if (diff != 0) \
1336		return (diff > 0 ? 1 : -1); \
1337} while (0)
1338
1339#define ORDERKEY_STATE(a, b) do { \
1340	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
1341	if (diff != 0) \
1342		return (diff > 0 ? 1 : -1); \
1343} while (0)
1344
1345#define ORDERKEY_PRIO(a, b) do { \
1346	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1347	if (diff != 0) \
1348		return (diff > 0 ? 1 : -1); \
1349} while (0)
1350
1351#define	ORDERKEY_THREADS(a, b) do { \
1352	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1353	if (diff != 0) \
1354		return (diff > 0 ? 1 : -1); \
1355} while (0)
1356
1357#define ORDERKEY_RSSIZE(a, b) do { \
1358	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1359	if (diff != 0) \
1360		return (diff > 0 ? 1 : -1); \
1361} while (0)
1362
1363#define ORDERKEY_MEM(a, b) do { \
1364	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1365	if (diff != 0) \
1366		return (diff > 0 ? 1 : -1); \
1367} while (0)
1368
1369#define ORDERKEY_JID(a, b) do { \
1370	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1371	if (diff != 0) \
1372		return (diff > 0 ? 1 : -1); \
1373} while (0)
1374
1375#define ORDERKEY_SWAP(a, b) do { \
1376	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1377	if (diff != 0) \
1378		return (diff > 0 ? 1 : -1); \
1379} while (0)
1380
1381/* compare_cpu - the comparison function for sorting by cpu percentage */
1382
1383int
1384#ifdef ORDER
1385compare_cpu(void *arg1, void *arg2)
1386#else
1387proc_compare(void *arg1, void *arg2)
1388#endif
1389{
1390	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1391	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1392
1393	ORDERKEY_PCTCPU(p1, p2);
1394	ORDERKEY_CPTICKS(p1, p2);
1395	ORDERKEY_STATE(p1, p2);
1396	ORDERKEY_PRIO(p1, p2);
1397	ORDERKEY_RSSIZE(p1, p2);
1398	ORDERKEY_MEM(p1, p2);
1399
1400	return (0);
1401}
1402
1403#ifdef ORDER
1404/* "cpu" compare routines */
1405int compare_size(), compare_res(), compare_time(), compare_prio(),
1406    compare_threads();
1407
1408/*
1409 * "io" compare routines.  Context switches aren't i/o, but are displayed
1410 * on the "io" display.
1411 */
1412int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1413    compare_vcsw(), compare_ivcsw();
1414
1415int (*compares[])() = {
1416	compare_cpu,
1417	compare_size,
1418	compare_res,
1419	compare_time,
1420	compare_prio,
1421	compare_threads,
1422	compare_iototal,
1423	compare_ioread,
1424	compare_iowrite,
1425	compare_iofault,
1426	compare_vcsw,
1427	compare_ivcsw,
1428	compare_jid,
1429	compare_swap,
1430	NULL
1431};
1432
1433/* compare_size - the comparison function for sorting by total memory usage */
1434
1435int
1436compare_size(void *arg1, void *arg2)
1437{
1438	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1439	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1440
1441	ORDERKEY_MEM(p1, p2);
1442	ORDERKEY_RSSIZE(p1, p2);
1443	ORDERKEY_PCTCPU(p1, p2);
1444	ORDERKEY_CPTICKS(p1, p2);
1445	ORDERKEY_STATE(p1, p2);
1446	ORDERKEY_PRIO(p1, p2);
1447
1448	return (0);
1449}
1450
1451/* compare_res - the comparison function for sorting by resident set size */
1452
1453int
1454compare_res(void *arg1, void *arg2)
1455{
1456	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1457	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1458
1459	ORDERKEY_RSSIZE(p1, p2);
1460	ORDERKEY_MEM(p1, p2);
1461	ORDERKEY_PCTCPU(p1, p2);
1462	ORDERKEY_CPTICKS(p1, p2);
1463	ORDERKEY_STATE(p1, p2);
1464	ORDERKEY_PRIO(p1, p2);
1465
1466	return (0);
1467}
1468
1469/* compare_time - the comparison function for sorting by total cpu time */
1470
1471int
1472compare_time(void *arg1, void *arg2)
1473{
1474	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1475	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1476
1477	ORDERKEY_CPTICKS(p1, p2);
1478	ORDERKEY_PCTCPU(p1, p2);
1479	ORDERKEY_STATE(p1, p2);
1480	ORDERKEY_PRIO(p1, p2);
1481	ORDERKEY_RSSIZE(p1, p2);
1482	ORDERKEY_MEM(p1, p2);
1483
1484	return (0);
1485}
1486
1487/* compare_prio - the comparison function for sorting by priority */
1488
1489int
1490compare_prio(void *arg1, void *arg2)
1491{
1492	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1493	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1494
1495	ORDERKEY_PRIO(p1, p2);
1496	ORDERKEY_CPTICKS(p1, p2);
1497	ORDERKEY_PCTCPU(p1, p2);
1498	ORDERKEY_STATE(p1, p2);
1499	ORDERKEY_RSSIZE(p1, p2);
1500	ORDERKEY_MEM(p1, p2);
1501
1502	return (0);
1503}
1504
1505/* compare_threads - the comparison function for sorting by threads */
1506int
1507compare_threads(void *arg1, void *arg2)
1508{
1509	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1510	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1511
1512	ORDERKEY_THREADS(p1, p2);
1513	ORDERKEY_PCTCPU(p1, p2);
1514	ORDERKEY_CPTICKS(p1, p2);
1515	ORDERKEY_STATE(p1, p2);
1516	ORDERKEY_PRIO(p1, p2);
1517	ORDERKEY_RSSIZE(p1, p2);
1518	ORDERKEY_MEM(p1, p2);
1519
1520	return (0);
1521}
1522
1523/* compare_jid - the comparison function for sorting by jid */
1524static int
1525compare_jid(const void *arg1, const void *arg2)
1526{
1527	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1528	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1529
1530	ORDERKEY_JID(p1, p2);
1531	ORDERKEY_PCTCPU(p1, p2);
1532	ORDERKEY_CPTICKS(p1, p2);
1533	ORDERKEY_STATE(p1, p2);
1534	ORDERKEY_PRIO(p1, p2);
1535	ORDERKEY_RSSIZE(p1, p2);
1536	ORDERKEY_MEM(p1, p2);
1537
1538	return (0);
1539}
1540
1541/* compare_swap - the comparison function for sorting by swap */
1542static int
1543compare_swap(const void *arg1, const void *arg2)
1544{
1545	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1546	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1547
1548	ORDERKEY_SWAP(p1, p2);
1549	ORDERKEY_PCTCPU(p1, p2);
1550	ORDERKEY_CPTICKS(p1, p2);
1551	ORDERKEY_STATE(p1, p2);
1552	ORDERKEY_PRIO(p1, p2);
1553	ORDERKEY_RSSIZE(p1, p2);
1554	ORDERKEY_MEM(p1, p2);
1555
1556	return (0);
1557}
1558#endif /* ORDER */
1559
1560/* assorted comparison functions for sorting by i/o */
1561
1562int
1563#ifdef ORDER
1564compare_iototal(void *arg1, void *arg2)
1565#else
1566io_compare(void *arg1, void *arg2)
1567#endif
1568{
1569	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1570	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1571
1572	return (get_io_total(p2) - get_io_total(p1));
1573}
1574
1575#ifdef ORDER
1576int
1577compare_ioread(void *arg1, void *arg2)
1578{
1579	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1580	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1581	long dummy, inp1, inp2;
1582
1583	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1584	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1585
1586	return (inp2 - inp1);
1587}
1588
1589int
1590compare_iowrite(void *arg1, void *arg2)
1591{
1592	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1593	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1594	long dummy, oup1, oup2;
1595
1596	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1597	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1598
1599	return (oup2 - oup1);
1600}
1601
1602int
1603compare_iofault(void *arg1, void *arg2)
1604{
1605	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1606	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1607	long dummy, flp1, flp2;
1608
1609	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1610	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1611
1612	return (flp2 - flp1);
1613}
1614
1615int
1616compare_vcsw(void *arg1, void *arg2)
1617{
1618	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1619	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1620	long dummy, flp1, flp2;
1621
1622	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1623	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1624
1625	return (flp2 - flp1);
1626}
1627
1628int
1629compare_ivcsw(void *arg1, void *arg2)
1630{
1631	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1632	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1633	long dummy, flp1, flp2;
1634
1635	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1636	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1637
1638	return (flp2 - flp1);
1639}
1640#endif /* ORDER */
1641
1642/*
1643 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1644 *		the process does not exist.
1645 *		It is EXTREMELY IMPORTANT that this function work correctly.
1646 *		If top runs setuid root (as in SVR4), then this function
1647 *		is the only thing that stands in the way of a serious
1648 *		security problem.  It validates requests for the "kill"
1649 *		and "renice" commands.
1650 */
1651
1652int
1653proc_owner(int pid)
1654{
1655	int cnt;
1656	struct kinfo_proc **prefp;
1657	struct kinfo_proc *pp;
1658
1659	prefp = pref;
1660	cnt = pref_len;
1661	while (--cnt >= 0) {
1662		pp = *prefp++;
1663		if (pp->ki_pid == (pid_t)pid)
1664			return ((int)pp->ki_ruid);
1665	}
1666	return (-1);
1667}
1668
1669static int
1670swapmode(int *retavail, int *retfree)
1671{
1672	int n;
1673	struct kvm_swap swapary[1];
1674	static int pagesize = 0;
1675	static u_long swap_maxpages = 0;
1676
1677	*retavail = 0;
1678	*retfree = 0;
1679
1680#define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1681
1682	n = kvm_getswapinfo(kd, swapary, 1, 0);
1683	if (n < 0 || swapary[0].ksw_total == 0)
1684		return (0);
1685
1686	if (pagesize == 0)
1687		pagesize = getpagesize();
1688	if (swap_maxpages == 0)
1689		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1690
1691	/* ksw_total contains the total size of swap all devices which may
1692	   exceed the maximum swap size allocatable in the system */
1693	if ( swapary[0].ksw_total > swap_maxpages )
1694		swapary[0].ksw_total = swap_maxpages;
1695
1696	*retavail = CONVERT(swapary[0].ksw_total);
1697	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1698
1699	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1700	return (n);
1701}
1702