1/*
2 * top - a top users display for Unix
3 *
4 * DESCRIPTION:
5 * Originally written for BSD4.4 system by Christos Zoulas.
6 * Ported to FreeBSD 2.x by Steven Wallace && Wolfram Schneider
7 * Order support hacked in from top-3.5beta6/machine/m_aix41.c
8 *   by Monte Mitzelfelt (for latest top see http://www.groupsys.com/topinfo/)
9 *
10 * AUTHOR:  Christos Zoulas <christos@ee.cornell.edu>
11 *          Steven Wallace  <swallace@FreeBSD.org>
12 *          Wolfram Schneider <wosch@FreeBSD.org>
13 *          Thomas Moestl <tmoestl@gmx.net>
14 *          Eitan Adler <eadler@FreeBSD.org>
15 */
16
17#include <sys/param.h>
18#include <sys/cpuset.h>
19#include <sys/errno.h>
20#include <sys/fcntl.h>
21#include <sys/priority.h>
22#include <sys/proc.h>
23#include <sys/resource.h>
24#include <sys/sbuf.h>
25#include <sys/sysctl.h>
26#include <sys/time.h>
27#include <sys/user.h>
28
29#include <assert.h>
30#include <err.h>
31#include <libgen.h>
32#include <kvm.h>
33#include <math.h>
34#include <paths.h>
35#include <stdio.h>
36#include <stdbool.h>
37#include <stdint.h>
38#include <stdlib.h>
39#include <string.h>
40#include <time.h>
41#include <unistd.h>
42#include <vis.h>
43
44#include "top.h"
45#include "display.h"
46#include "machine.h"
47#include "loadavg.h"
48#include "screen.h"
49#include "utils.h"
50#include "layout.h"
51
52#define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
53
54extern struct timeval timeout;
55static int smpmode;
56enum displaymodes displaymode;
57static const int namelength = 10;
58/* TOP_JID_LEN based on max of 999999 */
59#define TOP_JID_LEN 6
60#define TOP_SWAP_LEN 5
61
62/* get_process_info passes back a handle.  This is what it looks like: */
63
64struct handle {
65	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
66	int remaining;			/* number of pointers remaining */
67};
68
69
70/* define what weighted cpu is.  */
71#define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
72			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
73
74/* what we consider to be process size: */
75#define PROCSIZE(pp) ((pp)->ki_size / 1024)
76
77#define RU(pp)	(&(pp)->ki_rusage)
78
79#define	PCTCPU(pp) (pcpu[pp - pbase])
80
81/* process state names for the "STATE" column of the display */
82/* the extra nulls in the string "run" are for adding a slash and
83   the processor number when needed */
84
85static const char *state_abbrev[] = {
86	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
87};
88
89
90static kvm_t *kd;
91
92/* values that we stash away in _init and use in later routines */
93
94static double logcpu;
95
96/* these are retrieved from the kernel in _init */
97
98static load_avg  ccpu;
99
100/* these are used in the get_ functions */
101
102static int lastpid;
103
104/* these are for calculating cpu state percentages */
105
106static long cp_time[CPUSTATES];
107static long cp_old[CPUSTATES];
108static long cp_diff[CPUSTATES];
109
110/* these are for detailing the process states */
111
112static const char *procstatenames[] = {
113	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
114	" zombie, ", " waiting, ", " lock, ",
115	NULL
116};
117static int process_states[nitems(procstatenames)];
118
119/* these are for detailing the cpu states */
120
121static int cpu_states[CPUSTATES];
122static const char *cpustatenames[] = {
123	"user", "nice", "system", "interrupt", "idle", NULL
124};
125
126/* these are for detailing the memory statistics */
127
128static const char *memorynames[] = {
129	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
130	"K Free", NULL
131};
132static int memory_stats[nitems(memorynames)];
133
134static const char *arcnames[] = {
135	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
136	NULL
137};
138static int arc_stats[nitems(arcnames)];
139
140static const char *carcnames[] = {
141	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
142	NULL
143};
144static int carc_stats[nitems(carcnames)];
145
146static const char *swapnames[] = {
147	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
148	NULL
149};
150static int swap_stats[nitems(swapnames)];
151
152static int has_swap;
153
154/* these are for keeping track of the proc array */
155
156static int nproc;
157static int onproc = -1;
158static int pref_len;
159static struct kinfo_proc *pbase;
160static struct kinfo_proc **pref;
161static struct kinfo_proc *previous_procs;
162static struct kinfo_proc **previous_pref;
163static int previous_proc_count = 0;
164static int previous_proc_count_max = 0;
165static int previous_thread;
166
167/* data used for recalculating pctcpu */
168static double *pcpu;
169static struct timespec proc_uptime;
170static struct timeval proc_wall_time;
171static struct timeval previous_wall_time;
172static uint64_t previous_interval = 0;
173
174/* total number of io operations */
175static long total_inblock;
176static long total_oublock;
177static long total_majflt;
178
179/* these are for getting the memory statistics */
180
181static int arc_enabled;
182static int carc_enabled;
183static int pageshift;		/* log base 2 of the pagesize */
184
185/* define pagetok in terms of pageshift */
186
187#define pagetok(size) ((size) << pageshift)
188
189/* swap usage */
190#define ki_swap(kip) \
191    ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
192
193/*
194 * Sorting orders.  The first element is the default.
195 */
196static const char *ordernames[] = {
197	"cpu", "size", "res", "time", "pri", "threads",
198	"total", "read", "write", "fault", "vcsw", "ivcsw",
199	"jid", "swap", "pid", NULL
200};
201
202/* Per-cpu time states */
203static int maxcpu;
204static int maxid;
205static int ncpus;
206static cpuset_t cpumask;
207static long *times;
208static long *pcpu_cp_time;
209static long *pcpu_cp_old;
210static long *pcpu_cp_diff;
211static int *pcpu_cpu_states;
212
213/* Battery units and states */
214static int battery_units;
215static int battery_life;
216
217static int compare_swap(const void *a, const void *b);
218static int compare_jid(const void *a, const void *b);
219static int compare_pid(const void *a, const void *b);
220static int compare_tid(const void *a, const void *b);
221static const char *format_nice(const struct kinfo_proc *pp);
222static void getsysctl(const char *name, void *ptr, size_t len);
223static int swapmode(int *retavail, int *retfree);
224static void update_layout(void);
225static int find_uid(uid_t needle, int *haystack);
226static int cmd_matches(struct kinfo_proc *, const char *);
227
228static int
229find_uid(uid_t needle, int *haystack)
230{
231	size_t i = 0;
232
233	for (; i < TOP_MAX_UIDS; ++i)
234		if ((uid_t)haystack[i] == needle)
235			return 1;
236	return (0);
237}
238
239void
240toggle_pcpustats(void)
241{
242
243	if (ncpus == 1)
244		return;
245	update_layout();
246}
247
248/* Adjust display based on ncpus and the ARC state. */
249static void
250update_layout(void)
251{
252
253	y_mem = 3;
254	y_arc = 4;
255	y_carc = 5;
256	y_swap = 3 + arc_enabled + carc_enabled + has_swap;
257	y_idlecursor = 4 + arc_enabled + carc_enabled + has_swap;
258	y_message = 4 + arc_enabled + carc_enabled + has_swap;
259	y_header = 5 + arc_enabled + carc_enabled + has_swap;
260	y_procs = 6 + arc_enabled + carc_enabled + has_swap;
261	Header_lines = 6 + arc_enabled + carc_enabled + has_swap;
262
263	if (pcpu_stats) {
264		y_mem += ncpus - 1;
265		y_arc += ncpus - 1;
266		y_carc += ncpus - 1;
267		y_swap += ncpus - 1;
268		y_idlecursor += ncpus - 1;
269		y_message += ncpus - 1;
270		y_header += ncpus - 1;
271		y_procs += ncpus - 1;
272		Header_lines += ncpus - 1;
273	}
274}
275
276int
277machine_init(struct statics *statics)
278{
279	int i, j, empty, pagesize;
280	uint64_t arc_size;
281	int carc_en, nswapdev;
282	size_t size;
283
284	size = sizeof(smpmode);
285	if (sysctlbyname("kern.smp.active", &smpmode, &size, NULL, 0) != 0 ||
286	    size != sizeof(smpmode))
287		smpmode = 0;
288
289	size = sizeof(arc_size);
290	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
291	    NULL, 0) == 0 && arc_size != 0)
292		arc_enabled = 1;
293	size = sizeof(carc_en);
294	if (arc_enabled &&
295	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
296	    NULL, 0) == 0 && carc_en == 1) {
297		uint64_t uncomp_sz;
298
299		/*
300		 * Don't report compression stats if no data is in the ARC.
301		 * Otherwise, we end up printing a blank line.
302		 */
303		size = sizeof(uncomp_sz);
304		if (sysctlbyname("kstat.zfs.misc.arcstats.uncompressed_size",
305		    &uncomp_sz, &size, NULL, 0) == 0 && uncomp_sz != 0)
306			carc_enabled = 1;
307	}
308
309	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
310	if (kd == NULL)
311		return (-1);
312
313	size = sizeof(nswapdev);
314	if (sysctlbyname("vm.nswapdev", &nswapdev, &size, NULL, 0) == 0 &&
315	    nswapdev != 0)
316		has_swap = 1;
317
318	GETSYSCTL("kern.ccpu", ccpu);
319
320	/* this is used in calculating WCPU -- calculate it ahead of time */
321	logcpu = log(loaddouble(ccpu));
322
323	pbase = NULL;
324	pref = NULL;
325	pcpu = NULL;
326	nproc = 0;
327	onproc = -1;
328
329	/* get the page size and calculate pageshift from it */
330	pagesize = getpagesize();
331	pageshift = 0;
332	while (pagesize > 1) {
333		pageshift++;
334		pagesize >>= 1;
335	}
336
337	/* we only need the amount of log(2)1024 for our conversion */
338	pageshift -= LOG1024;
339
340	/* fill in the statics information */
341	statics->procstate_names = procstatenames;
342	statics->cpustate_names = cpustatenames;
343	statics->memory_names = memorynames;
344	if (arc_enabled)
345		statics->arc_names = arcnames;
346	else
347		statics->arc_names = NULL;
348	if (carc_enabled)
349		statics->carc_names = carcnames;
350	else
351		statics->carc_names = NULL;
352	if (has_swap)
353		statics->swap_names = swapnames;
354	else
355		statics->swap_names = NULL;
356	statics->order_names = ordernames;
357
358	/* Allocate state for per-CPU stats. */
359	GETSYSCTL("kern.smp.maxcpus", maxcpu);
360	times = calloc(maxcpu * CPUSTATES, sizeof(long));
361	if (times == NULL)
362		err(1, "calloc for kern.smp.maxcpus");
363	size = sizeof(long) * maxcpu * CPUSTATES;
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 = MIN(size / CPUSTATES / sizeof(long) - 1, CPU_SETSIZE - 1);
368	CPU_ZERO(&cpumask);
369	for (i = 0; i <= maxid; i++) {
370		empty = 1;
371		for (j = 0; empty && j < CPUSTATES; j++) {
372			if (times[i * CPUSTATES + j] != 0)
373				empty = 0;
374		}
375		if (!empty)
376			CPU_SET(i, &cpumask);
377	}
378	ncpus = CPU_COUNT(&cpumask);
379	assert(ncpus > 0);
380	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
381	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
382	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
383	statics->ncpus = ncpus;
384
385	/* Allocate state of battery units reported via ACPI. */
386	battery_units = 0;
387	size = sizeof(int);
388	sysctlbyname("hw.acpi.battery.units", &battery_units, &size, NULL, 0);
389	statics->nbatteries = battery_units;
390
391	update_layout();
392
393	/* all done! */
394	return (0);
395}
396
397char *
398format_header(const char *uname_field)
399{
400	static struct sbuf* header = NULL;
401
402	/* clean up from last time. */
403	if (header != NULL) {
404		sbuf_clear(header);
405	} else {
406		header = sbuf_new_auto();
407	}
408
409	switch (displaymode) {
410	case DISP_CPU: {
411		sbuf_printf(header, "  %s", ps.thread_id ? " THR" : "PID");
412		sbuf_printf(header, "%*s", ps.jail ? TOP_JID_LEN : 0,
413									ps.jail ? " JID" : "");
414		sbuf_printf(header, " %-*.*s  ", namelength, namelength, uname_field);
415		if (!ps.thread) {
416			sbuf_cat(header, "THR ");
417		}
418		sbuf_cat(header, "PRI NICE   SIZE    RES ");
419		if (ps.swap) {
420			sbuf_printf(header, "%*s ", TOP_SWAP_LEN - 1, "SWAP");
421		}
422		sbuf_cat(header, "STATE    ");
423		if (smpmode) {
424			sbuf_cat(header, "C   ");
425		}
426		sbuf_cat(header, "TIME ");
427		sbuf_printf(header, " %6s ", ps.wcpu ? "WCPU" : "CPU");
428		sbuf_cat(header, "COMMAND");
429		sbuf_finish(header);
430		break;
431	}
432	case DISP_IO: {
433		sbuf_printf(header, "  %s%*s %-*.*s",
434			ps.thread_id ? " THR" : "PID",
435		    ps.jail ? TOP_JID_LEN : 0, ps.jail ? " JID" : "",
436		    namelength, namelength, uname_field);
437		sbuf_cat(header, "   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND");
438		sbuf_finish(header);
439		break;
440	}
441	case DISP_MAX:
442		assert("displaymode must not be set to DISP_MAX");
443	}
444
445	return sbuf_data(header);
446}
447
448static int swappgsin = -1;
449static int swappgsout = -1;
450
451
452void
453get_system_info(struct system_info *si)
454{
455	struct loadavg sysload;
456	int mib[2];
457	struct timeval boottime;
458	uint64_t arc_stat, arc_stat2;
459	int i, j;
460	size_t size;
461
462	/* get the CPU stats */
463	size = (maxid + 1) * CPUSTATES * sizeof(long);
464	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
465		err(1, "sysctlbyname kern.cp_times");
466	GETSYSCTL("kern.cp_time", cp_time);
467	GETSYSCTL("vm.loadavg", sysload);
468	GETSYSCTL("kern.lastpid", lastpid);
469
470	/* convert load averages to doubles */
471	for (i = 0; i < 3; i++)
472		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
473
474	/* convert cp_time counts to percentages */
475	for (i = j = 0; i <= maxid; i++) {
476		if (!CPU_ISSET(i, &cpumask))
477			continue;
478		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
479		    &pcpu_cp_time[j * CPUSTATES],
480		    &pcpu_cp_old[j * CPUSTATES],
481		    &pcpu_cp_diff[j * CPUSTATES]);
482		j++;
483	}
484	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
485
486	/* sum memory & swap statistics */
487	{
488		static unsigned int swap_delay = 0;
489		static int swapavail = 0;
490		static int swapfree = 0;
491		static long bufspace = 0;
492		static uint64_t nspgsin, nspgsout;
493
494		GETSYSCTL("vfs.bufspace", bufspace);
495		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
496		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
497		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
498		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
499		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
500		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
501		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
502		/* convert memory stats to Kbytes */
503		memory_stats[0] = pagetok(memory_stats[0]);
504		memory_stats[1] = pagetok(memory_stats[1]);
505		memory_stats[2] = pagetok(memory_stats[2]);
506		memory_stats[3] = pagetok(memory_stats[3]);
507		memory_stats[4] = bufspace / 1024;
508		memory_stats[5] = pagetok(memory_stats[5]);
509		memory_stats[6] = -1;
510
511		/* first interval */
512		if (swappgsin < 0) {
513			swap_stats[4] = 0;
514			swap_stats[5] = 0;
515		}
516
517		/* compute differences between old and new swap statistic */
518		else {
519			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
520			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
521		}
522
523		swappgsin = nspgsin;
524		swappgsout = nspgsout;
525
526		/* call CPU heavy swapmode() only for changes */
527		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
528			swap_stats[3] = swapmode(&swapavail, &swapfree);
529			swap_stats[0] = swapavail;
530			swap_stats[1] = swapavail - swapfree;
531			swap_stats[2] = swapfree;
532		}
533		swap_delay = 1;
534		swap_stats[6] = -1;
535	}
536
537	if (arc_enabled) {
538		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
539		arc_stats[0] = arc_stat >> 10;
540		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
541		arc_stats[1] = arc_stat >> 10;
542		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
543		arc_stats[2] = arc_stat >> 10;
544		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
545		arc_stats[3] = arc_stat >> 10;
546		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
547		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
548		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
549		GETSYSCTL("kstat.zfs.misc.arcstats.bonus_size", arc_stat);
550		arc_stats[5] = arc_stat >> 10;
551		GETSYSCTL("kstat.zfs.misc.arcstats.dnode_size", arc_stat);
552		arc_stats[5] += arc_stat >> 10;
553		GETSYSCTL("kstat.zfs.misc.arcstats.dbuf_size", arc_stat);
554		arc_stats[5] += arc_stat >> 10;
555		si->arc = arc_stats;
556	}
557	if (carc_enabled) {
558		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
559		carc_stats[0] = arc_stat >> 10;
560		carc_stats[2] = arc_stat >> 10; /* For ratio */
561		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
562		carc_stats[1] = arc_stat >> 10;
563		si->carc = carc_stats;
564	}
565
566	/* set arrays and strings */
567	if (pcpu_stats) {
568		si->cpustates = pcpu_cpu_states;
569		si->ncpus = ncpus;
570	} else {
571		si->cpustates = cpu_states;
572		si->ncpus = 1;
573	}
574	si->memory = memory_stats;
575	si->swap = swap_stats;
576
577
578	if (lastpid > 0) {
579		si->last_pid = lastpid;
580	} else {
581		si->last_pid = -1;
582	}
583
584	/*
585	 * Print how long system has been up.
586	 * (Found by looking getting "boottime" from the kernel)
587	 */
588	mib[0] = CTL_KERN;
589	mib[1] = KERN_BOOTTIME;
590	size = sizeof(boottime);
591	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
592	    boottime.tv_sec != 0) {
593		si->boottime = boottime;
594	} else {
595		si->boottime.tv_sec = -1;
596	}
597
598	battery_life = 0;
599	if (battery_units > 0) {
600		GETSYSCTL("hw.acpi.battery.life", battery_life);
601	}
602	si->battery = battery_life;
603}
604
605#define NOPROC	((void *)-1)
606
607/*
608 * We need to compare data from the old process entry with the new
609 * process entry.
610 * To facilitate doing this quickly we stash a pointer in the kinfo_proc
611 * structure to cache the mapping.  We also use a negative cache pointer
612 * of NOPROC to avoid duplicate lookups.
613 * XXX: this could be done when the actual processes are fetched, we do
614 * it here out of laziness.
615 */
616static const struct kinfo_proc *
617get_old_proc(struct kinfo_proc *pp)
618{
619	const struct kinfo_proc * const *oldpp, *oldp;
620
621	/*
622	 * If this is the first fetch of the kinfo_procs then we don't have
623	 * any previous entries.
624	 */
625	if (previous_proc_count == 0)
626		return (NULL);
627	/* negative cache? */
628	if (pp->ki_udata == NOPROC)
629		return (NULL);
630	/* cached? */
631	if (pp->ki_udata != NULL)
632		return (pp->ki_udata);
633	/*
634	 * Not cached,
635	 * 1) look up based on pid.
636	 * 2) compare process start.
637	 * If we fail here, then setup a negative cache entry, otherwise
638	 * cache it.
639	 */
640	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
641	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
642	if (oldpp == NULL) {
643		pp->ki_udata = NOPROC;
644		return (NULL);
645	}
646	oldp = *oldpp;
647	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
648		pp->ki_udata = NOPROC;
649		return (NULL);
650	}
651	pp->ki_udata = __DECONST(void *, oldp);
652	return (oldp);
653}
654
655/*
656 * Return the total amount of IO done in blocks in/out and faults.
657 * store the values individually in the pointers passed in.
658 */
659static long
660get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
661    long *vcsw, long *ivcsw)
662{
663	const struct kinfo_proc *oldp;
664	static struct kinfo_proc dummy;
665	long ret;
666
667	oldp = get_old_proc(__DECONST(struct kinfo_proc *, pp));
668	if (oldp == NULL) {
669		memset(&dummy, 0, sizeof(dummy));
670		oldp = &dummy;
671	}
672	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
673	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
674	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
675	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
676	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
677	ret =
678	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
679	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
680	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
681	return (ret);
682}
683
684/*
685 * If there was a previous update, use the delta in ki_runtime over
686 * the previous interval to calculate pctcpu.  Otherwise, fall back
687 * to using the kernel's ki_pctcpu.
688 */
689static double
690proc_calc_pctcpu(struct kinfo_proc *pp)
691{
692	const struct kinfo_proc *oldp;
693
694	if (previous_interval != 0) {
695		oldp = get_old_proc(pp);
696		if (oldp != NULL)
697			return ((double)(pp->ki_runtime - oldp->ki_runtime)
698			    / previous_interval);
699
700		/*
701		 * If this process/thread was created during the previous
702		 * interval, charge it's total runtime to the previous
703		 * interval.
704		 */
705		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
706		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
707		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
708			return ((double)pp->ki_runtime / previous_interval);
709	}
710	return (pctdouble(pp->ki_pctcpu));
711}
712
713/*
714 * Return true if this process has used any CPU time since the
715 * previous update.
716 */
717static int
718proc_used_cpu(struct kinfo_proc *pp)
719{
720	const struct kinfo_proc *oldp;
721
722	oldp = get_old_proc(pp);
723	if (oldp == NULL)
724		return (PCTCPU(pp) != 0);
725	return (pp->ki_runtime != oldp->ki_runtime ||
726	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
727	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
728}
729
730/*
731 * Return the total number of block in/out and faults by a process.
732 */
733static long
734get_io_total(const struct kinfo_proc *pp)
735{
736	long dummy;
737
738	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
739}
740
741static struct handle handle;
742
743void *
744get_process_info(struct system_info *si, struct process_select *sel,
745    int (*compare)(const void *, const void *))
746{
747	int i;
748	int total_procs;
749	long p_io;
750	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
751	long nsec;
752	int active_procs;
753	struct kinfo_proc **prefp;
754	struct kinfo_proc *pp;
755	struct timespec previous_proc_uptime;
756
757	/*
758	 * If thread state was toggled, don't cache the previous processes.
759	 */
760	if (previous_thread != sel->thread)
761		nproc = 0;
762	previous_thread = sel->thread;
763
764	/*
765	 * Save the previous process info.
766	 */
767	if (previous_proc_count_max < nproc) {
768		free(previous_procs);
769		previous_procs = calloc(nproc, sizeof(*previous_procs));
770		free(previous_pref);
771		previous_pref = calloc(nproc, sizeof(*previous_pref));
772		if (previous_procs == NULL || previous_pref == NULL) {
773			fprintf(stderr, "top: Out of memory.\n");
774			quit(TOP_EX_SYS_ERROR);
775		}
776		previous_proc_count_max = nproc;
777	}
778	if (nproc) {
779		for (i = 0; i < nproc; i++)
780			previous_pref[i] = &previous_procs[i];
781		memcpy(previous_procs, pbase, nproc * sizeof(*previous_procs));
782		qsort(previous_pref, nproc, sizeof(*previous_pref),
783		    ps.thread ? compare_tid : compare_pid);
784	}
785	previous_proc_count = nproc;
786	previous_proc_uptime = proc_uptime;
787	previous_wall_time = proc_wall_time;
788	previous_interval = 0;
789
790	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
791	    0, &nproc);
792	gettimeofday(&proc_wall_time, NULL);
793	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
794		memset(&proc_uptime, 0, sizeof(proc_uptime));
795	else if (previous_proc_uptime.tv_sec != 0 &&
796	    previous_proc_uptime.tv_nsec != 0) {
797		previous_interval = (proc_uptime.tv_sec -
798		    previous_proc_uptime.tv_sec) * 1000000;
799		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
800		if (nsec < 0) {
801			previous_interval -= 1000000;
802			nsec += 1000000000;
803		}
804		previous_interval += nsec / 1000;
805	}
806	if (nproc > onproc) {
807		pref = realloc(pref, sizeof(*pref) * nproc);
808		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
809		onproc = nproc;
810	}
811	if (pref == NULL || pbase == NULL || pcpu == NULL) {
812		fprintf(stderr, "top: Out of memory.\n");
813		quit(TOP_EX_SYS_ERROR);
814	}
815	/* get a pointer to the states summary array */
816	si->procstates = process_states;
817
818	/* count up process states and get pointers to interesting procs */
819	total_procs = 0;
820	active_procs = 0;
821	total_inblock = 0;
822	total_oublock = 0;
823	total_majflt = 0;
824	memset(process_states, 0, sizeof(process_states));
825	prefp = pref;
826	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
827
828		if (pp->ki_stat == 0)
829			/* not in use */
830			continue;
831
832		if (!sel->self && pp->ki_pid == mypid && sel->pid == -1)
833			/* skip self */
834			continue;
835
836		if (!sel->system && (pp->ki_flag & P_SYSTEM) && sel->pid == -1)
837			/* skip system process */
838			continue;
839
840		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
841		    &p_vcsw, &p_ivcsw);
842		total_inblock += p_inblock;
843		total_oublock += p_oublock;
844		total_majflt += p_majflt;
845		total_procs++;
846		process_states[(unsigned char)pp->ki_stat]++;
847
848		if (pp->ki_stat == SZOMB)
849			/* skip zombies */
850			continue;
851
852		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD && sel->pid == -1)
853			/* skip kernel idle process */
854			continue;
855
856		PCTCPU(pp) = proc_calc_pctcpu(pp);
857		if (sel->thread && PCTCPU(pp) > 1.0)
858			PCTCPU(pp) = 1.0;
859		if (displaymode == DISP_CPU && !sel->idle &&
860		    (!proc_used_cpu(pp) ||
861		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
862			/* skip idle or non-running processes */
863			continue;
864
865		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
866			/* skip processes that aren't doing I/O */
867			continue;
868
869		if (sel->jid != -1 && pp->ki_jid != sel->jid)
870			/* skip proc. that don't belong to the selected JID */
871			continue;
872
873		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
874			/* skip proc. that don't belong to the selected UID */
875			continue;
876
877		if (sel->pid != -1 && pp->ki_pid != sel->pid)
878			continue;
879
880		if (!cmd_matches(pp, sel->command))
881			/* skip proc. that doesn't match grep string */
882			continue;
883
884		*prefp++ = pp;
885		active_procs++;
886	}
887
888	/* if requested, sort the "interesting" processes */
889	if (compare != NULL)
890		qsort(pref, active_procs, sizeof(*pref), compare);
891
892	/* remember active and total counts */
893	si->p_total = total_procs;
894	si->p_pactive = pref_len = active_procs;
895
896	/* pass back a handle */
897	handle.next_proc = pref;
898	handle.remaining = active_procs;
899	return (&handle);
900}
901
902static int
903cmd_matches(struct kinfo_proc *proc, const char *term)
904{
905	char **args = NULL;
906
907	if (!term) {
908		/* No command filter set */
909		return 1;
910	} else {
911		/* Filter set, does process name contain term? */
912		if (strstr(proc->ki_comm, term))
913			return 1;
914		/* Search arguments only if arguments are displayed */
915		if (show_args) {
916			args = kvm_getargv(kd, proc, 1024);
917			if (args == NULL) {
918				/* Failed to get arguments so can't search them */
919				return 0;
920			}
921			while (*args != NULL) {
922				if (strstr(*args, term))
923					return 1;
924				args++;
925			}
926		}
927	}
928	return 0;
929}
930
931char *
932format_next_process(struct handle * xhandle, char *(*get_userid)(int), int flags)
933{
934	struct kinfo_proc *pp;
935	const struct kinfo_proc *oldp;
936	long cputime;
937	char status[22];
938	size_t state;
939	struct rusage ru, *rup;
940	long p_tot, s_tot;
941	char *cmdbuf = NULL;
942	char **args;
943	static struct sbuf* procbuf = NULL;
944
945	/* clean up from last time. */
946	if (procbuf != NULL) {
947		sbuf_clear(procbuf);
948	} else {
949		procbuf = sbuf_new_auto();
950	}
951
952
953	/* find and remember the next proc structure */
954	pp = *(xhandle->next_proc++);
955	xhandle->remaining--;
956
957	/* get the process's command name */
958	if ((pp->ki_flag & P_INMEM) == 0) {
959		/*
960		 * Print swapped processes as <pname>
961		 */
962		size_t len;
963
964		len = strlen(pp->ki_comm);
965		if (len > sizeof(pp->ki_comm) - 3)
966			len = sizeof(pp->ki_comm) - 3;
967		memmove(pp->ki_comm + 1, pp->ki_comm, len);
968		pp->ki_comm[0] = '<';
969		pp->ki_comm[len + 1] = '>';
970		pp->ki_comm[len + 2] = '\0';
971	}
972
973	/*
974	 * Convert the process's runtime from microseconds to seconds.  This
975	 * time includes the interrupt time although that is not wanted here.
976	 * ps(1) is similarly sloppy.
977	 */
978	cputime = (pp->ki_runtime + 500000) / 1000000;
979
980	/* generate "STATE" field */
981	switch (state = pp->ki_stat) {
982	case SRUN:
983		if (smpmode && pp->ki_oncpu != NOCPU)
984			sprintf(status, "CPU%d", pp->ki_oncpu);
985		else
986			strcpy(status, "RUN");
987		break;
988	case SLOCK:
989		if (pp->ki_kiflag & KI_LOCKBLOCK) {
990			sprintf(status, "*%.6s", pp->ki_lockname);
991			break;
992		}
993		/* fall through */
994	case SSLEEP:
995		sprintf(status, "%.6s", pp->ki_wmesg);
996		break;
997	default:
998
999		if (state < nitems(state_abbrev)) {
1000			sprintf(status, "%.6s", state_abbrev[state]);
1001		} else {
1002			sprintf(status, "?%5zu", state);
1003		}
1004		break;
1005	}
1006
1007	cmdbuf = calloc(screen_width + 1, 1);
1008	if (cmdbuf == NULL) {
1009		warn("calloc(%d)", screen_width + 1);
1010		return NULL;
1011	}
1012
1013	if (!(flags & FMT_SHOWARGS)) {
1014		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1015		    pp->ki_tdname[0]) {
1016			snprintf(cmdbuf, screen_width, "%s{%s%s}", pp->ki_comm,
1017			    pp->ki_tdname, pp->ki_moretdname);
1018		} else {
1019			snprintf(cmdbuf, screen_width, "%s", pp->ki_comm);
1020		}
1021	} else {
1022		if (pp->ki_flag & P_SYSTEM ||
1023		    (args = kvm_getargv(kd, pp, screen_width)) == NULL ||
1024		    !(*args)) {
1025			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1026		    	    pp->ki_tdname[0]) {
1027				snprintf(cmdbuf, screen_width,
1028				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
1029				    pp->ki_moretdname);
1030			} else {
1031				snprintf(cmdbuf, screen_width,
1032				    "[%s]", pp->ki_comm);
1033			}
1034		} else {
1035			const char *src;
1036			char *dst, *argbuf;
1037			const char *cmd;
1038			size_t argbuflen;
1039			size_t len;
1040
1041			argbuflen = screen_width * 4;
1042			argbuf = calloc(argbuflen + 1, 1);
1043			if (argbuf == NULL) {
1044				warn("calloc(%zu)", argbuflen + 1);
1045				free(cmdbuf);
1046				return NULL;
1047			}
1048
1049			dst = argbuf;
1050
1051			/* Extract cmd name from argv */
1052			cmd = basename(*args);
1053
1054			for (; (src = *args++) != NULL; ) {
1055				if (*src == '\0')
1056					continue;
1057				len = (argbuflen - (dst - argbuf) - 1) / 4;
1058				strvisx(dst, src,
1059				    MIN(strlen(src), len),
1060				    VIS_NL | VIS_TAB | VIS_CSTYLE | VIS_OCTAL);
1061				while (*dst != '\0')
1062					dst++;
1063				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1064					*dst++ = ' '; /* add delimiting space */
1065			}
1066			if (dst != argbuf && dst[-1] == ' ')
1067				dst--;
1068			*dst = '\0';
1069
1070			if (strcmp(cmd, pp->ki_comm) != 0) {
1071				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1072				    pp->ki_tdname[0])
1073					snprintf(cmdbuf, screen_width,
1074					    "%s (%s){%s%s}", argbuf,
1075					    pp->ki_comm, pp->ki_tdname,
1076					    pp->ki_moretdname);
1077				else
1078					snprintf(cmdbuf, screen_width,
1079					    "%s (%s)", argbuf, pp->ki_comm);
1080			} else {
1081				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1082				    pp->ki_tdname[0])
1083					snprintf(cmdbuf, screen_width,
1084					    "%s{%s%s}", argbuf, pp->ki_tdname,
1085					    pp->ki_moretdname);
1086				else
1087					strlcpy(cmdbuf, argbuf, screen_width);
1088			}
1089			free(argbuf);
1090		}
1091	}
1092
1093	if (displaymode == DISP_IO) {
1094		oldp = get_old_proc(pp);
1095		if (oldp != NULL) {
1096			ru.ru_inblock = RU(pp)->ru_inblock -
1097			    RU(oldp)->ru_inblock;
1098			ru.ru_oublock = RU(pp)->ru_oublock -
1099			    RU(oldp)->ru_oublock;
1100			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1101			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1102			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1103			rup = &ru;
1104		} else {
1105			rup = RU(pp);
1106		}
1107		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1108		s_tot = total_inblock + total_oublock + total_majflt;
1109
1110		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1111
1112		if (ps.jail) {
1113			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1114		}
1115		sbuf_printf(procbuf, "%-*.*s", namelength, namelength, (*get_userid)(pp->ki_ruid));
1116		sbuf_printf(procbuf, "%6ld ", rup->ru_nvcsw);
1117		sbuf_printf(procbuf, "%6ld ", rup->ru_nivcsw);
1118		sbuf_printf(procbuf, "%6ld ", rup->ru_inblock);
1119		sbuf_printf(procbuf, "%6ld ", rup->ru_oublock);
1120		sbuf_printf(procbuf, "%6ld ", rup->ru_majflt);
1121		sbuf_printf(procbuf, "%6ld ", p_tot);
1122		sbuf_printf(procbuf, "%6.2f%% ", s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot));
1123
1124	} else {
1125		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1126		if (ps.jail) {
1127			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1128		}
1129		sbuf_printf(procbuf, "%-*.*s ", namelength, namelength, (*get_userid)(pp->ki_ruid));
1130
1131		if (!ps.thread) {
1132			sbuf_printf(procbuf, "%4d ", pp->ki_numthreads);
1133		} else {
1134			sbuf_printf(procbuf, " ");
1135		}
1136
1137		sbuf_printf(procbuf, "%3d ", pp->ki_pri.pri_level - PZERO);
1138		sbuf_printf(procbuf, "%4s", format_nice(pp));
1139		sbuf_printf(procbuf, "%7s ", format_k(PROCSIZE(pp)));
1140		sbuf_printf(procbuf, "%6s ", format_k(pagetok(pp->ki_rssize)));
1141		if (ps.swap) {
1142			sbuf_printf(procbuf, "%*s ",
1143				TOP_SWAP_LEN - 1,
1144				format_k(pagetok(ki_swap(pp))));
1145		}
1146		sbuf_printf(procbuf, "%-6.6s ", status);
1147		if (smpmode) {
1148			int cpu;
1149			if (state == SRUN && pp->ki_oncpu != NOCPU) {
1150				cpu = pp->ki_oncpu;
1151			} else {
1152				cpu = pp->ki_lastcpu;
1153			}
1154			sbuf_printf(procbuf, "%3d ", cpu);
1155		}
1156		sbuf_printf(procbuf, "%6s ", format_time(cputime));
1157		sbuf_printf(procbuf, "%6.2f%% ", ps.wcpu ? 100.0 * weighted_cpu(PCTCPU(pp), pp) : 100.0 * PCTCPU(pp));
1158	}
1159	sbuf_printf(procbuf, "%s", cmdbuf);
1160	free(cmdbuf);
1161	return (sbuf_data(procbuf));
1162}
1163
1164static void
1165getsysctl(const char *name, void *ptr, size_t len)
1166{
1167	size_t nlen = len;
1168
1169	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1170		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1171		    strerror(errno));
1172		quit(TOP_EX_SYS_ERROR);
1173	}
1174	if (nlen != len) {
1175		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1176		    name, (unsigned long)len, (unsigned long)nlen);
1177		quit(TOP_EX_SYS_ERROR);
1178	}
1179}
1180
1181static const char *
1182format_nice(const struct kinfo_proc *pp)
1183{
1184	const char *fifo, *kproc;
1185	int rtpri;
1186	static char nicebuf[4 + 1];
1187
1188	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1189	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1190	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1191	case PRI_ITHD:
1192		return ("-");
1193	case PRI_REALTIME:
1194		/*
1195		 * XXX: the kernel doesn't tell us the original rtprio and
1196		 * doesn't really know what it was, so to recover it we
1197		 * must be more chummy with the implementation than the
1198		 * implementation is with itself.  pri_user gives a
1199		 * constant "base" priority, but is only initialized
1200		 * properly for user threads.  pri_native gives what the
1201		 * kernel calls the "base" priority, but it isn't constant
1202		 * since it is changed by priority propagation.  pri_native
1203		 * also isn't properly initialized for all threads, but it
1204		 * is properly initialized for kernel realtime and idletime
1205		 * threads.  Thus we use pri_user for the base priority of
1206		 * user threads (it is always correct) and pri_native for
1207		 * the base priority of kernel realtime and idletime threads
1208		 * (there is nothing better, and it is usually correct).
1209		 *
1210		 * The field width and thus the buffer are too small for
1211		 * values like "kr31F", but such values shouldn't occur,
1212		 * and if they do then the tailing "F" is not displayed.
1213		 */
1214		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1215		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1216		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1217		    kproc, rtpri, fifo);
1218		break;
1219	case PRI_TIMESHARE:
1220		if (pp->ki_flag & P_KPROC)
1221			return ("-");
1222		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1223		break;
1224	case PRI_IDLE:
1225		/* XXX: as above. */
1226		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1227		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1228		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1229		    kproc, rtpri, fifo);
1230		break;
1231	default:
1232		return ("?");
1233	}
1234	return (nicebuf);
1235}
1236
1237/* comparison routines for qsort */
1238
1239static int
1240compare_pid(const void *p1, const void *p2)
1241{
1242	const struct kinfo_proc * const *pp1 = p1;
1243	const struct kinfo_proc * const *pp2 = p2;
1244
1245	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1246
1247	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1248}
1249
1250static int
1251compare_tid(const void *p1, const void *p2)
1252{
1253	const struct kinfo_proc * const *pp1 = p1;
1254	const struct kinfo_proc * const *pp2 = p2;
1255
1256	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1257
1258	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1259}
1260
1261/*
1262 *  proc_compare - comparison function for "qsort"
1263 *	Compares the resource consumption of two processes using five
1264 *	distinct keys.  The keys (in descending order of importance) are:
1265 *	percent cpu, cpu ticks, state, resident set size, total virtual
1266 *	memory usage.  The process states are ordered as follows (from least
1267 *	to most important):  run, zombie, idle, interrupt wait, stop, sleep.
1268 *	The array declaration below maps a process state index into a
1269 *	number that reflects this ordering.
1270 */
1271
1272static const int sorted_state[] = {
1273	[SIDL] =	3,	/* being created	*/
1274	[SRUN] =	1,	/* running/runnable	*/
1275	[SSLEEP] =	6,	/* sleeping		*/
1276	[SSTOP] =	5,	/* stopped/suspended	*/
1277	[SZOMB] =	2,	/* zombie		*/
1278	[SWAIT] =	4,	/* intr			*/
1279	[SLOCK] =	7,	/* blocked on lock	*/
1280};
1281
1282
1283#define ORDERKEY_PCTCPU(a, b) do { \
1284	double diff; \
1285	if (ps.wcpu) \
1286		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1287		    weighted_cpu(PCTCPU((a)), (a)); \
1288	else \
1289		diff = PCTCPU((b)) - PCTCPU((a)); \
1290	if (diff != 0) \
1291		return (diff > 0 ? 1 : -1); \
1292} while (0)
1293
1294#define ORDERKEY_CPTICKS(a, b) do { \
1295	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1296	if (diff != 0) \
1297		return (diff > 0 ? 1 : -1); \
1298} while (0)
1299
1300#define ORDERKEY_STATE(a, b) do { \
1301	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1302	if (diff != 0) \
1303		return (diff > 0 ? 1 : -1); \
1304} while (0)
1305
1306#define ORDERKEY_PRIO(a, b) do { \
1307	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1308	if (diff != 0) \
1309		return (diff > 0 ? 1 : -1); \
1310} while (0)
1311
1312#define	ORDERKEY_THREADS(a, b) do { \
1313	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1314	if (diff != 0) \
1315		return (diff > 0 ? 1 : -1); \
1316} while (0)
1317
1318#define ORDERKEY_RSSIZE(a, b) do { \
1319	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1320	if (diff != 0) \
1321		return (diff > 0 ? 1 : -1); \
1322} while (0)
1323
1324#define ORDERKEY_MEM(a, b) do { \
1325	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1326	if (diff != 0) \
1327		return (diff > 0 ? 1 : -1); \
1328} while (0)
1329
1330#define ORDERKEY_JID(a, b) do { \
1331	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1332	if (diff != 0) \
1333		return (diff > 0 ? 1 : -1); \
1334} while (0)
1335
1336#define ORDERKEY_SWAP(a, b) do { \
1337	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1338	if (diff != 0) \
1339		return (diff > 0 ? 1 : -1); \
1340} while (0)
1341
1342/* compare_cpu - the comparison function for sorting by cpu percentage */
1343
1344static int
1345compare_cpu(const void *arg1, const void *arg2)
1346{
1347	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1348	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1349
1350	ORDERKEY_PCTCPU(p1, p2);
1351	ORDERKEY_CPTICKS(p1, p2);
1352	ORDERKEY_STATE(p1, p2);
1353	ORDERKEY_PRIO(p1, p2);
1354	ORDERKEY_RSSIZE(p1, p2);
1355	ORDERKEY_MEM(p1, p2);
1356
1357	return (0);
1358}
1359
1360/* compare_size - the comparison function for sorting by total memory usage */
1361
1362static int
1363compare_size(const void *arg1, const void *arg2)
1364{
1365	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1366	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1367
1368	ORDERKEY_MEM(p1, p2);
1369	ORDERKEY_RSSIZE(p1, p2);
1370	ORDERKEY_PCTCPU(p1, p2);
1371	ORDERKEY_CPTICKS(p1, p2);
1372	ORDERKEY_STATE(p1, p2);
1373	ORDERKEY_PRIO(p1, p2);
1374
1375	return (0);
1376}
1377
1378/* compare_res - the comparison function for sorting by resident set size */
1379
1380static int
1381compare_res(const void *arg1, const void *arg2)
1382{
1383	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1384	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1385
1386	ORDERKEY_RSSIZE(p1, p2);
1387	ORDERKEY_MEM(p1, p2);
1388	ORDERKEY_PCTCPU(p1, p2);
1389	ORDERKEY_CPTICKS(p1, p2);
1390	ORDERKEY_STATE(p1, p2);
1391	ORDERKEY_PRIO(p1, p2);
1392
1393	return (0);
1394}
1395
1396/* compare_time - the comparison function for sorting by total cpu time */
1397
1398static int
1399compare_time(const void *arg1, const void *arg2)
1400{
1401	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1402	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1403
1404	ORDERKEY_CPTICKS(p1, p2);
1405	ORDERKEY_PCTCPU(p1, p2);
1406	ORDERKEY_STATE(p1, p2);
1407	ORDERKEY_PRIO(p1, p2);
1408	ORDERKEY_RSSIZE(p1, p2);
1409	ORDERKEY_MEM(p1, p2);
1410
1411	return (0);
1412}
1413
1414/* compare_prio - the comparison function for sorting by priority */
1415
1416static int
1417compare_prio(const void *arg1, const void *arg2)
1418{
1419	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1420	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1421
1422	ORDERKEY_PRIO(p1, p2);
1423	ORDERKEY_CPTICKS(p1, p2);
1424	ORDERKEY_PCTCPU(p1, p2);
1425	ORDERKEY_STATE(p1, p2);
1426	ORDERKEY_RSSIZE(p1, p2);
1427	ORDERKEY_MEM(p1, p2);
1428
1429	return (0);
1430}
1431
1432/* compare_threads - the comparison function for sorting by threads */
1433static int
1434compare_threads(const void *arg1, const void *arg2)
1435{
1436	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1437	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1438
1439	ORDERKEY_THREADS(p1, p2);
1440	ORDERKEY_PCTCPU(p1, p2);
1441	ORDERKEY_CPTICKS(p1, p2);
1442	ORDERKEY_STATE(p1, p2);
1443	ORDERKEY_PRIO(p1, p2);
1444	ORDERKEY_RSSIZE(p1, p2);
1445	ORDERKEY_MEM(p1, p2);
1446
1447	return (0);
1448}
1449
1450/* compare_jid - the comparison function for sorting by jid */
1451static int
1452compare_jid(const void *arg1, const void *arg2)
1453{
1454	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1455	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1456
1457	ORDERKEY_JID(p1, p2);
1458	ORDERKEY_PCTCPU(p1, p2);
1459	ORDERKEY_CPTICKS(p1, p2);
1460	ORDERKEY_STATE(p1, p2);
1461	ORDERKEY_PRIO(p1, p2);
1462	ORDERKEY_RSSIZE(p1, p2);
1463	ORDERKEY_MEM(p1, p2);
1464
1465	return (0);
1466}
1467
1468/* compare_swap - the comparison function for sorting by swap */
1469static int
1470compare_swap(const void *arg1, const void *arg2)
1471{
1472	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1473	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1474
1475	ORDERKEY_SWAP(p1, p2);
1476	ORDERKEY_PCTCPU(p1, p2);
1477	ORDERKEY_CPTICKS(p1, p2);
1478	ORDERKEY_STATE(p1, p2);
1479	ORDERKEY_PRIO(p1, p2);
1480	ORDERKEY_RSSIZE(p1, p2);
1481	ORDERKEY_MEM(p1, p2);
1482
1483	return (0);
1484}
1485
1486/* assorted comparison functions for sorting by i/o */
1487
1488static int
1489compare_iototal(const void *arg1, const void *arg2)
1490{
1491	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1492	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1493
1494	return (get_io_total(p2) - get_io_total(p1));
1495}
1496
1497static int
1498compare_ioread(const void *arg1, const void *arg2)
1499{
1500	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1501	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1502	long dummy, inp1, inp2;
1503
1504	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1505	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1506
1507	return (inp2 - inp1);
1508}
1509
1510static int
1511compare_iowrite(const void *arg1, const void *arg2)
1512{
1513	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1514	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1515	long dummy, oup1, oup2;
1516
1517	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1518	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1519
1520	return (oup2 - oup1);
1521}
1522
1523static int
1524compare_iofault(const void *arg1, const void *arg2)
1525{
1526	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1527	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1528	long dummy, flp1, flp2;
1529
1530	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1531	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1532
1533	return (flp2 - flp1);
1534}
1535
1536static int
1537compare_vcsw(const void *arg1, const void *arg2)
1538{
1539	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1540	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1541	long dummy, flp1, flp2;
1542
1543	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1544	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1545
1546	return (flp2 - flp1);
1547}
1548
1549static int
1550compare_ivcsw(const void *arg1, const void *arg2)
1551{
1552	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1553	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1554	long dummy, flp1, flp2;
1555
1556	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1557	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1558
1559	return (flp2 - flp1);
1560}
1561
1562int (*compares[])(const void *arg1, const void *arg2) = {
1563	compare_cpu,
1564	compare_size,
1565	compare_res,
1566	compare_time,
1567	compare_prio,
1568	compare_threads,
1569	compare_iototal,
1570	compare_ioread,
1571	compare_iowrite,
1572	compare_iofault,
1573	compare_vcsw,
1574	compare_ivcsw,
1575	compare_jid,
1576	compare_swap,
1577	compare_pid,
1578	NULL
1579};
1580
1581
1582static int
1583swapmode(int *retavail, int *retfree)
1584{
1585	int n;
1586	struct kvm_swap swapary[1];
1587	static int pagesize = 0;
1588	static unsigned long swap_maxpages = 0;
1589
1590	*retavail = 0;
1591	*retfree = 0;
1592
1593#define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1594
1595	n = kvm_getswapinfo(kd, swapary, 1, 0);
1596	if (n < 0 || swapary[0].ksw_total == 0)
1597		return (0);
1598
1599	if (pagesize == 0)
1600		pagesize = getpagesize();
1601	if (swap_maxpages == 0)
1602		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1603
1604	/* ksw_total contains the total size of swap all devices which may
1605	   exceed the maximum swap size allocatable in the system */
1606	if ( swapary[0].ksw_total > swap_maxpages )
1607		swapary[0].ksw_total = swap_maxpages;
1608
1609	*retavail = CONVERT(swapary[0].ksw_total);
1610	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1611
1612#undef CONVERT
1613
1614	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1615	return (n);
1616}
1617