xref: /freebsd/usr.bin/top/machine.c (revision 25cdacf79b06356c929e59d5074d26c9dac41bdf)
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  * $FreeBSD$
17  */
18 
19 #include <sys/errno.h>
20 #include <sys/fcntl.h>
21 #include <sys/param.h>
22 #include <sys/priority.h>
23 #include <sys/proc.h>
24 #include <sys/resource.h>
25 #include <sys/sbuf.h>
26 #include <sys/sysctl.h>
27 #include <sys/time.h>
28 #include <sys/user.h>
29 
30 #include <assert.h>
31 #include <err.h>
32 #include <libgen.h>
33 #include <kvm.h>
34 #include <math.h>
35 #include <paths.h>
36 #include <stdio.h>
37 #include <stdbool.h>
38 #include <stdint.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <time.h>
42 #include <unistd.h>
43 #include <vis.h>
44 
45 #include "top.h"
46 #include "display.h"
47 #include "machine.h"
48 #include "loadavg.h"
49 #include "screen.h"
50 #include "utils.h"
51 #include "layout.h"
52 
53 #define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
54 
55 extern struct timeval timeout;
56 static int smpmode;
57 enum displaymodes displaymode;
58 static const int namelength = 10;
59 /* TOP_JID_LEN based on max of 999999 */
60 #define TOP_JID_LEN 6
61 #define TOP_SWAP_LEN 5
62 
63 /* get_process_info passes back a handle.  This is what it looks like: */
64 
65 struct handle {
66 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
67 	int remaining;			/* number of pointers remaining */
68 };
69 
70 
71 /* define what weighted cpu is.  */
72 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
73 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
74 
75 /* what we consider to be process size: */
76 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
77 
78 #define RU(pp)	(&(pp)->ki_rusage)
79 
80 #define	PCTCPU(pp) (pcpu[pp - pbase])
81 
82 /* process state names for the "STATE" column of the display */
83 /* the extra nulls in the string "run" are for adding a slash and
84    the processor number when needed */
85 
86 static const char *state_abbrev[] = {
87 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
88 };
89 
90 
91 static kvm_t *kd;
92 
93 /* values that we stash away in _init and use in later routines */
94 
95 static double logcpu;
96 
97 /* these are retrieved from the kernel in _init */
98 
99 static load_avg  ccpu;
100 
101 /* these are used in the get_ functions */
102 
103 static int lastpid;
104 
105 /* these are for calculating cpu state percentages */
106 
107 static long cp_time[CPUSTATES];
108 static long cp_old[CPUSTATES];
109 static long cp_diff[CPUSTATES];
110 
111 /* these are for detailing the process states */
112 
113 static const char *procstatenames[] = {
114 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
115 	" zombie, ", " waiting, ", " lock, ",
116 	NULL
117 };
118 static int process_states[nitems(procstatenames)];
119 
120 /* these are for detailing the cpu states */
121 
122 static int cpu_states[CPUSTATES];
123 static const char *cpustatenames[] = {
124 	"user", "nice", "system", "interrupt", "idle", NULL
125 };
126 
127 /* these are for detailing the memory statistics */
128 
129 static const char *memorynames[] = {
130 	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
131 	"K Free", NULL
132 };
133 static int memory_stats[nitems(memorynames)];
134 
135 static const char *arcnames[] = {
136 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
137 	NULL
138 };
139 static int arc_stats[nitems(arcnames)];
140 
141 static const char *carcnames[] = {
142 	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
143 	NULL
144 };
145 static int carc_stats[nitems(carcnames)];
146 
147 static const char *swapnames[] = {
148 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
149 	NULL
150 };
151 static int swap_stats[nitems(swapnames)];
152 
153 static int has_swap;
154 
155 /* these are for keeping track of the proc array */
156 
157 static int nproc;
158 static int onproc = -1;
159 static int pref_len;
160 static struct kinfo_proc *pbase;
161 static struct kinfo_proc **pref;
162 static struct kinfo_proc *previous_procs;
163 static struct kinfo_proc **previous_pref;
164 static int previous_proc_count = 0;
165 static int previous_proc_count_max = 0;
166 static int previous_thread;
167 
168 /* data used for recalculating pctcpu */
169 static double *pcpu;
170 static struct timespec proc_uptime;
171 static struct timeval proc_wall_time;
172 static struct timeval previous_wall_time;
173 static uint64_t previous_interval = 0;
174 
175 /* total number of io operations */
176 static long total_inblock;
177 static long total_oublock;
178 static long total_majflt;
179 
180 /* these are for getting the memory statistics */
181 
182 static int arc_enabled;
183 static int carc_enabled;
184 static int pageshift;		/* log base 2 of the pagesize */
185 
186 /* define pagetok in terms of pageshift */
187 
188 #define pagetok(size) ((size) << pageshift)
189 
190 /* swap usage */
191 #define ki_swap(kip) \
192     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
193 
194 /*
195  * Sorting orders.  The first element is the default.
196  */
197 static const char *ordernames[] = {
198 	"cpu", "size", "res", "time", "pri", "threads",
199 	"total", "read", "write", "fault", "vcsw", "ivcsw",
200 	"jid", "swap", "pid", NULL
201 };
202 
203 /* Per-cpu time states */
204 static int maxcpu;
205 static int maxid;
206 static int ncpus;
207 static unsigned long cpumask;
208 static long *times;
209 static long *pcpu_cp_time;
210 static long *pcpu_cp_old;
211 static long *pcpu_cp_diff;
212 static int *pcpu_cpu_states;
213 
214 /* Battery units and states */
215 static int battery_units;
216 static int battery_life;
217 
218 static int compare_swap(const void *a, const void *b);
219 static int compare_jid(const void *a, const void *b);
220 static int compare_pid(const void *a, const void *b);
221 static int compare_tid(const void *a, const void *b);
222 static const char *format_nice(const struct kinfo_proc *pp);
223 static void getsysctl(const char *name, void *ptr, size_t len);
224 static int swapmode(int *retavail, int *retfree);
225 static void update_layout(void);
226 static int find_uid(uid_t needle, int *haystack);
227 
228 static int
229 find_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 
239 void
240 toggle_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. */
249 static void
250 update_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 
276 int
277 machine_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 		carc_enabled = 1;
298 
299 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
300 	if (kd == NULL)
301 		return (-1);
302 
303 	size = sizeof(nswapdev);
304 	if (sysctlbyname("vm.nswapdev", &nswapdev, &size, NULL,
305 		0) == 0 && nswapdev != 0)
306 			has_swap = 1;
307 
308 	GETSYSCTL("kern.ccpu", ccpu);
309 
310 	/* this is used in calculating WCPU -- calculate it ahead of time */
311 	logcpu = log(loaddouble(ccpu));
312 
313 	pbase = NULL;
314 	pref = NULL;
315 	pcpu = NULL;
316 	nproc = 0;
317 	onproc = -1;
318 
319 	/* get the page size and calculate pageshift from it */
320 	pagesize = getpagesize();
321 	pageshift = 0;
322 	while (pagesize > 1) {
323 		pageshift++;
324 		pagesize >>= 1;
325 	}
326 
327 	/* we only need the amount of log(2)1024 for our conversion */
328 	pageshift -= LOG1024;
329 
330 	/* fill in the statics information */
331 	statics->procstate_names = procstatenames;
332 	statics->cpustate_names = cpustatenames;
333 	statics->memory_names = memorynames;
334 	if (arc_enabled)
335 		statics->arc_names = arcnames;
336 	else
337 		statics->arc_names = NULL;
338 	if (carc_enabled)
339 		statics->carc_names = carcnames;
340 	else
341 		statics->carc_names = NULL;
342 	if (has_swap)
343 		statics->swap_names = swapnames;
344 	else
345 		statics->swap_names = NULL;
346 	statics->order_names = ordernames;
347 
348 	/* Allocate state for per-CPU stats. */
349 	cpumask = 0;
350 	ncpus = 0;
351 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
352 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
353 	if (times == NULL)
354 		err(1, "calloc for kern.smp.maxcpus");
355 	size = sizeof(long) * maxcpu * CPUSTATES;
356 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
357 		err(1, "sysctlbyname kern.cp_times");
358 	pcpu_cp_time = calloc(1, size);
359 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
360 	for (i = 0; i <= maxid; i++) {
361 		empty = 1;
362 		for (j = 0; empty && j < CPUSTATES; j++) {
363 			if (times[i * CPUSTATES + j] != 0)
364 				empty = 0;
365 		}
366 		if (!empty) {
367 			cpumask |= (1ul << i);
368 			ncpus++;
369 		}
370 	}
371 	assert(ncpus > 0);
372 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
373 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
374 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
375 	statics->ncpus = ncpus;
376 
377 	/* Allocate state of battery units reported via ACPI. */
378 	battery_units = 0;
379 	size = sizeof(int);
380 	sysctlbyname("hw.acpi.battery.units", &battery_units, &size, NULL, 0);
381 	statics->nbatteries = battery_units;
382 
383 	update_layout();
384 
385 	/* all done! */
386 	return (0);
387 }
388 
389 char *
390 format_header(const char *uname_field)
391 {
392 	static struct sbuf* header = NULL;
393 
394 	/* clean up from last time. */
395 	if (header != NULL) {
396 		sbuf_clear(header);
397 	} else {
398 		header = sbuf_new_auto();
399 	}
400 
401 	switch (displaymode) {
402 	case DISP_CPU: {
403 		sbuf_printf(header, "  %s", ps.thread_id ? " THR" : "PID");
404 		sbuf_printf(header, "%*s", ps.jail ? TOP_JID_LEN : 0,
405 									ps.jail ? " JID" : "");
406 		sbuf_printf(header, " %-*.*s  ", namelength, namelength, uname_field);
407 		if (!ps.thread) {
408 			sbuf_cat(header, "THR ");
409 		}
410 		sbuf_cat(header, "PRI NICE   SIZE    RES ");
411 		if (ps.swap) {
412 			sbuf_printf(header, "%*s ", TOP_SWAP_LEN - 1, "SWAP");
413 		}
414 		sbuf_cat(header, "STATE    ");
415 		if (smpmode) {
416 			sbuf_cat(header, "C   ");
417 		}
418 		sbuf_cat(header, "TIME ");
419 		sbuf_printf(header, " %6s ", ps.wcpu ? "WCPU" : "CPU");
420 		sbuf_cat(header, "COMMAND");
421 		sbuf_finish(header);
422 		break;
423 	}
424 	case DISP_IO: {
425 		sbuf_printf(header, "  %s%*s %-*.*s",
426 			ps.thread_id ? " THR" : "PID",
427 		    ps.jail ? TOP_JID_LEN : 0, ps.jail ? " JID" : "",
428 		    namelength, namelength, uname_field);
429 		sbuf_cat(header, "   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND");
430 		sbuf_finish(header);
431 		break;
432 	}
433 	case DISP_MAX:
434 		assert("displaymode must not be set to DISP_MAX");
435 	}
436 
437 	return sbuf_data(header);
438 }
439 
440 static int swappgsin = -1;
441 static int swappgsout = -1;
442 
443 
444 void
445 get_system_info(struct system_info *si)
446 {
447 	struct loadavg sysload;
448 	int mib[2];
449 	struct timeval boottime;
450 	uint64_t arc_stat, arc_stat2;
451 	int i, j;
452 	size_t size;
453 
454 	/* get the CPU stats */
455 	size = (maxid + 1) * CPUSTATES * sizeof(long);
456 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
457 		err(1, "sysctlbyname kern.cp_times");
458 	GETSYSCTL("kern.cp_time", cp_time);
459 	GETSYSCTL("vm.loadavg", sysload);
460 	GETSYSCTL("kern.lastpid", lastpid);
461 
462 	/* convert load averages to doubles */
463 	for (i = 0; i < 3; i++)
464 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
465 
466 	/* convert cp_time counts to percentages */
467 	for (i = j = 0; i <= maxid; i++) {
468 		if ((cpumask & (1ul << i)) == 0)
469 			continue;
470 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
471 		    &pcpu_cp_time[j * CPUSTATES],
472 		    &pcpu_cp_old[j * CPUSTATES],
473 		    &pcpu_cp_diff[j * CPUSTATES]);
474 		j++;
475 	}
476 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
477 
478 	/* sum memory & swap statistics */
479 	{
480 		static unsigned int swap_delay = 0;
481 		static int swapavail = 0;
482 		static int swapfree = 0;
483 		static long bufspace = 0;
484 		static uint64_t nspgsin, nspgsout;
485 
486 		GETSYSCTL("vfs.bufspace", bufspace);
487 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
488 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
489 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
490 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
491 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
492 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
493 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
494 		/* convert memory stats to Kbytes */
495 		memory_stats[0] = pagetok(memory_stats[0]);
496 		memory_stats[1] = pagetok(memory_stats[1]);
497 		memory_stats[2] = pagetok(memory_stats[2]);
498 		memory_stats[3] = pagetok(memory_stats[3]);
499 		memory_stats[4] = bufspace / 1024;
500 		memory_stats[5] = pagetok(memory_stats[5]);
501 		memory_stats[6] = -1;
502 
503 		/* first interval */
504 		if (swappgsin < 0) {
505 			swap_stats[4] = 0;
506 			swap_stats[5] = 0;
507 		}
508 
509 		/* compute differences between old and new swap statistic */
510 		else {
511 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
512 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
513 		}
514 
515 		swappgsin = nspgsin;
516 		swappgsout = nspgsout;
517 
518 		/* call CPU heavy swapmode() only for changes */
519 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
520 			swap_stats[3] = swapmode(&swapavail, &swapfree);
521 			swap_stats[0] = swapavail;
522 			swap_stats[1] = swapavail - swapfree;
523 			swap_stats[2] = swapfree;
524 		}
525 		swap_delay = 1;
526 		swap_stats[6] = -1;
527 	}
528 
529 	if (arc_enabled) {
530 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
531 		arc_stats[0] = arc_stat >> 10;
532 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
533 		arc_stats[1] = arc_stat >> 10;
534 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
535 		arc_stats[2] = arc_stat >> 10;
536 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
537 		arc_stats[3] = arc_stat >> 10;
538 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
539 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
540 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
541 		GETSYSCTL("kstat.zfs.misc.arcstats.bonus_size", arc_stat);
542 		arc_stats[5] = arc_stat >> 10;
543 		GETSYSCTL("kstat.zfs.misc.arcstats.dnode_size", arc_stat);
544 		arc_stats[5] += arc_stat >> 10;
545 		GETSYSCTL("kstat.zfs.misc.arcstats.dbuf_size", arc_stat);
546 		arc_stats[5] += arc_stat >> 10;
547 		si->arc = arc_stats;
548 	}
549 	if (carc_enabled) {
550 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
551 		carc_stats[0] = arc_stat >> 10;
552 		carc_stats[2] = arc_stat >> 10; /* For ratio */
553 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
554 		carc_stats[1] = arc_stat >> 10;
555 		si->carc = carc_stats;
556 	}
557 
558 	/* set arrays and strings */
559 	if (pcpu_stats) {
560 		si->cpustates = pcpu_cpu_states;
561 		si->ncpus = ncpus;
562 	} else {
563 		si->cpustates = cpu_states;
564 		si->ncpus = 1;
565 	}
566 	si->memory = memory_stats;
567 	si->swap = swap_stats;
568 
569 
570 	if (lastpid > 0) {
571 		si->last_pid = lastpid;
572 	} else {
573 		si->last_pid = -1;
574 	}
575 
576 	/*
577 	 * Print how long system has been up.
578 	 * (Found by looking getting "boottime" from the kernel)
579 	 */
580 	mib[0] = CTL_KERN;
581 	mib[1] = KERN_BOOTTIME;
582 	size = sizeof(boottime);
583 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
584 	    boottime.tv_sec != 0) {
585 		si->boottime = boottime;
586 	} else {
587 		si->boottime.tv_sec = -1;
588 	}
589 
590 	battery_life = 0;
591 	if (battery_units > 0) {
592 		GETSYSCTL("hw.acpi.battery.life", battery_life);
593 	}
594 	si->battery = battery_life;
595 }
596 
597 #define NOPROC	((void *)-1)
598 
599 /*
600  * We need to compare data from the old process entry with the new
601  * process entry.
602  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
603  * structure to cache the mapping.  We also use a negative cache pointer
604  * of NOPROC to avoid duplicate lookups.
605  * XXX: this could be done when the actual processes are fetched, we do
606  * it here out of laziness.
607  */
608 static const struct kinfo_proc *
609 get_old_proc(struct kinfo_proc *pp)
610 {
611 	const struct kinfo_proc * const *oldpp, *oldp;
612 
613 	/*
614 	 * If this is the first fetch of the kinfo_procs then we don't have
615 	 * any previous entries.
616 	 */
617 	if (previous_proc_count == 0)
618 		return (NULL);
619 	/* negative cache? */
620 	if (pp->ki_udata == NOPROC)
621 		return (NULL);
622 	/* cached? */
623 	if (pp->ki_udata != NULL)
624 		return (pp->ki_udata);
625 	/*
626 	 * Not cached,
627 	 * 1) look up based on pid.
628 	 * 2) compare process start.
629 	 * If we fail here, then setup a negative cache entry, otherwise
630 	 * cache it.
631 	 */
632 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
633 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
634 	if (oldpp == NULL) {
635 		pp->ki_udata = NOPROC;
636 		return (NULL);
637 	}
638 	oldp = *oldpp;
639 	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
640 		pp->ki_udata = NOPROC;
641 		return (NULL);
642 	}
643 	pp->ki_udata = __DECONST(void *, oldp);
644 	return (oldp);
645 }
646 
647 /*
648  * Return the total amount of IO done in blocks in/out and faults.
649  * store the values individually in the pointers passed in.
650  */
651 static long
652 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
653     long *vcsw, long *ivcsw)
654 {
655 	const struct kinfo_proc *oldp;
656 	static struct kinfo_proc dummy;
657 	long ret;
658 
659 	oldp = get_old_proc(__DECONST(struct kinfo_proc *, pp));
660 	if (oldp == NULL) {
661 		memset(&dummy, 0, sizeof(dummy));
662 		oldp = &dummy;
663 	}
664 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
665 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
666 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
667 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
668 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
669 	ret =
670 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
671 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
672 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
673 	return (ret);
674 }
675 
676 /*
677  * If there was a previous update, use the delta in ki_runtime over
678  * the previous interval to calculate pctcpu.  Otherwise, fall back
679  * to using the kernel's ki_pctcpu.
680  */
681 static double
682 proc_calc_pctcpu(struct kinfo_proc *pp)
683 {
684 	const struct kinfo_proc *oldp;
685 
686 	if (previous_interval != 0) {
687 		oldp = get_old_proc(pp);
688 		if (oldp != NULL)
689 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
690 			    / previous_interval);
691 
692 		/*
693 		 * If this process/thread was created during the previous
694 		 * interval, charge it's total runtime to the previous
695 		 * interval.
696 		 */
697 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
698 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
699 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
700 			return ((double)pp->ki_runtime / previous_interval);
701 	}
702 	return (pctdouble(pp->ki_pctcpu));
703 }
704 
705 /*
706  * Return true if this process has used any CPU time since the
707  * previous update.
708  */
709 static int
710 proc_used_cpu(struct kinfo_proc *pp)
711 {
712 	const struct kinfo_proc *oldp;
713 
714 	oldp = get_old_proc(pp);
715 	if (oldp == NULL)
716 		return (PCTCPU(pp) != 0);
717 	return (pp->ki_runtime != oldp->ki_runtime ||
718 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
719 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
720 }
721 
722 /*
723  * Return the total number of block in/out and faults by a process.
724  */
725 static long
726 get_io_total(const struct kinfo_proc *pp)
727 {
728 	long dummy;
729 
730 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
731 }
732 
733 static struct handle handle;
734 
735 void *
736 get_process_info(struct system_info *si, struct process_select *sel,
737     int (*compare)(const void *, const void *))
738 {
739 	int i;
740 	int total_procs;
741 	long p_io;
742 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
743 	long nsec;
744 	int active_procs;
745 	struct kinfo_proc **prefp;
746 	struct kinfo_proc *pp;
747 	struct timespec previous_proc_uptime;
748 
749 	/*
750 	 * If thread state was toggled, don't cache the previous processes.
751 	 */
752 	if (previous_thread != sel->thread)
753 		nproc = 0;
754 	previous_thread = sel->thread;
755 
756 	/*
757 	 * Save the previous process info.
758 	 */
759 	if (previous_proc_count_max < nproc) {
760 		free(previous_procs);
761 		previous_procs = calloc(nproc, sizeof(*previous_procs));
762 		free(previous_pref);
763 		previous_pref = calloc(nproc, sizeof(*previous_pref));
764 		if (previous_procs == NULL || previous_pref == NULL) {
765 			fprintf(stderr, "top: Out of memory.\n");
766 			quit(TOP_EX_SYS_ERROR);
767 		}
768 		previous_proc_count_max = nproc;
769 	}
770 	if (nproc) {
771 		for (i = 0; i < nproc; i++)
772 			previous_pref[i] = &previous_procs[i];
773 		memcpy(previous_procs, pbase, nproc * sizeof(*previous_procs));
774 		qsort(previous_pref, nproc, sizeof(*previous_pref),
775 		    ps.thread ? compare_tid : compare_pid);
776 	}
777 	previous_proc_count = nproc;
778 	previous_proc_uptime = proc_uptime;
779 	previous_wall_time = proc_wall_time;
780 	previous_interval = 0;
781 
782 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
783 	    0, &nproc);
784 	gettimeofday(&proc_wall_time, NULL);
785 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
786 		memset(&proc_uptime, 0, sizeof(proc_uptime));
787 	else if (previous_proc_uptime.tv_sec != 0 &&
788 	    previous_proc_uptime.tv_nsec != 0) {
789 		previous_interval = (proc_uptime.tv_sec -
790 		    previous_proc_uptime.tv_sec) * 1000000;
791 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
792 		if (nsec < 0) {
793 			previous_interval -= 1000000;
794 			nsec += 1000000000;
795 		}
796 		previous_interval += nsec / 1000;
797 	}
798 	if (nproc > onproc) {
799 		pref = realloc(pref, sizeof(*pref) * nproc);
800 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
801 		onproc = nproc;
802 	}
803 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
804 		fprintf(stderr, "top: Out of memory.\n");
805 		quit(TOP_EX_SYS_ERROR);
806 	}
807 	/* get a pointer to the states summary array */
808 	si->procstates = process_states;
809 
810 	/* count up process states and get pointers to interesting procs */
811 	total_procs = 0;
812 	active_procs = 0;
813 	total_inblock = 0;
814 	total_oublock = 0;
815 	total_majflt = 0;
816 	memset(process_states, 0, sizeof(process_states));
817 	prefp = pref;
818 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
819 
820 		if (pp->ki_stat == 0)
821 			/* not in use */
822 			continue;
823 
824 		if (!sel->self && pp->ki_pid == mypid && sel->pid == -1)
825 			/* skip self */
826 			continue;
827 
828 		if (!sel->system && (pp->ki_flag & P_SYSTEM) && sel->pid == -1)
829 			/* skip system process */
830 			continue;
831 
832 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
833 		    &p_vcsw, &p_ivcsw);
834 		total_inblock += p_inblock;
835 		total_oublock += p_oublock;
836 		total_majflt += p_majflt;
837 		total_procs++;
838 		process_states[(unsigned char)pp->ki_stat]++;
839 
840 		if (pp->ki_stat == SZOMB)
841 			/* skip zombies */
842 			continue;
843 
844 		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD && sel->pid == -1)
845 			/* skip kernel idle process */
846 			continue;
847 
848 		PCTCPU(pp) = proc_calc_pctcpu(pp);
849 		if (sel->thread && PCTCPU(pp) > 1.0)
850 			PCTCPU(pp) = 1.0;
851 		if (displaymode == DISP_CPU && !sel->idle &&
852 		    (!proc_used_cpu(pp) ||
853 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
854 			/* skip idle or non-running processes */
855 			continue;
856 
857 		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
858 			/* skip processes that aren't doing I/O */
859 			continue;
860 
861 		if (sel->jid != -1 && pp->ki_jid != sel->jid)
862 			/* skip proc. that don't belong to the selected JID */
863 			continue;
864 
865 		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
866 			/* skip proc. that don't belong to the selected UID */
867 			continue;
868 
869 		if (sel->pid != -1 && pp->ki_pid != sel->pid)
870 			continue;
871 
872 		*prefp++ = pp;
873 		active_procs++;
874 	}
875 
876 	/* if requested, sort the "interesting" processes */
877 	if (compare != NULL)
878 		qsort(pref, active_procs, sizeof(*pref), compare);
879 
880 	/* remember active and total counts */
881 	si->p_total = total_procs;
882 	si->p_pactive = pref_len = active_procs;
883 
884 	/* pass back a handle */
885 	handle.next_proc = pref;
886 	handle.remaining = active_procs;
887 	return (&handle);
888 }
889 
890 char *
891 format_next_process(struct handle * xhandle, char *(*get_userid)(int), int flags)
892 {
893 	struct kinfo_proc *pp;
894 	const struct kinfo_proc *oldp;
895 	long cputime;
896 	char status[22];
897 	size_t state;
898 	struct rusage ru, *rup;
899 	long p_tot, s_tot;
900 	char *cmdbuf = NULL;
901 	char **args;
902 	static struct sbuf* procbuf = NULL;
903 
904 	/* clean up from last time. */
905 	if (procbuf != NULL) {
906 		sbuf_clear(procbuf);
907 	} else {
908 		procbuf = sbuf_new_auto();
909 	}
910 
911 
912 	/* find and remember the next proc structure */
913 	pp = *(xhandle->next_proc++);
914 	xhandle->remaining--;
915 
916 	/* get the process's command name */
917 	if ((pp->ki_flag & P_INMEM) == 0) {
918 		/*
919 		 * Print swapped processes as <pname>
920 		 */
921 		size_t len;
922 
923 		len = strlen(pp->ki_comm);
924 		if (len > sizeof(pp->ki_comm) - 3)
925 			len = sizeof(pp->ki_comm) - 3;
926 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
927 		pp->ki_comm[0] = '<';
928 		pp->ki_comm[len + 1] = '>';
929 		pp->ki_comm[len + 2] = '\0';
930 	}
931 
932 	/*
933 	 * Convert the process's runtime from microseconds to seconds.  This
934 	 * time includes the interrupt time although that is not wanted here.
935 	 * ps(1) is similarly sloppy.
936 	 */
937 	cputime = (pp->ki_runtime + 500000) / 1000000;
938 
939 	/* generate "STATE" field */
940 	switch (state = pp->ki_stat) {
941 	case SRUN:
942 		if (smpmode && pp->ki_oncpu != NOCPU)
943 			sprintf(status, "CPU%d", pp->ki_oncpu);
944 		else
945 			strcpy(status, "RUN");
946 		break;
947 	case SLOCK:
948 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
949 			sprintf(status, "*%.6s", pp->ki_lockname);
950 			break;
951 		}
952 		/* fall through */
953 	case SSLEEP:
954 		sprintf(status, "%.6s", pp->ki_wmesg);
955 		break;
956 	default:
957 
958 		if (state < nitems(state_abbrev)) {
959 			sprintf(status, "%.6s", state_abbrev[state]);
960 		} else {
961 			sprintf(status, "?%5zu", state);
962 		}
963 		break;
964 	}
965 
966 	cmdbuf = calloc(screen_width + 1, 1);
967 	if (cmdbuf == NULL) {
968 		warn("calloc(%d)", screen_width + 1);
969 		return NULL;
970 	}
971 
972 	if (!(flags & FMT_SHOWARGS)) {
973 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
974 		    pp->ki_tdname[0]) {
975 			snprintf(cmdbuf, screen_width, "%s{%s%s}", pp->ki_comm,
976 			    pp->ki_tdname, pp->ki_moretdname);
977 		} else {
978 			snprintf(cmdbuf, screen_width, "%s", pp->ki_comm);
979 		}
980 	} else {
981 		if (pp->ki_flag & P_SYSTEM ||
982 		    (args = kvm_getargv(kd, pp, screen_width)) == NULL ||
983 		    !(*args)) {
984 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
985 		    	    pp->ki_tdname[0]) {
986 				snprintf(cmdbuf, screen_width,
987 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
988 				    pp->ki_moretdname);
989 			} else {
990 				snprintf(cmdbuf, screen_width,
991 				    "[%s]", pp->ki_comm);
992 			}
993 		} else {
994 			const char *src;
995 			char *dst, *argbuf;
996 			const char *cmd;
997 			size_t argbuflen;
998 			size_t len;
999 
1000 			argbuflen = screen_width * 4;
1001 			argbuf = calloc(argbuflen + 1, 1);
1002 			if (argbuf == NULL) {
1003 				warn("calloc(%zu)", argbuflen + 1);
1004 				free(cmdbuf);
1005 				return NULL;
1006 			}
1007 
1008 			dst = argbuf;
1009 
1010 			/* Extract cmd name from argv */
1011 			cmd = basename(*args);
1012 
1013 			for (; (src = *args++) != NULL; ) {
1014 				if (*src == '\0')
1015 					continue;
1016 				len = (argbuflen - (dst - argbuf) - 1) / 4;
1017 				strvisx(dst, src,
1018 				    MIN(strlen(src), len),
1019 				    VIS_NL | VIS_TAB | VIS_CSTYLE | VIS_OCTAL);
1020 				while (*dst != '\0')
1021 					dst++;
1022 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1023 					*dst++ = ' '; /* add delimiting space */
1024 			}
1025 			if (dst != argbuf && dst[-1] == ' ')
1026 				dst--;
1027 			*dst = '\0';
1028 
1029 			if (strcmp(cmd, pp->ki_comm) != 0) {
1030 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1031 				    pp->ki_tdname[0])
1032 					snprintf(cmdbuf, screen_width,
1033 					    "%s (%s){%s%s}", argbuf,
1034 					    pp->ki_comm, pp->ki_tdname,
1035 					    pp->ki_moretdname);
1036 				else
1037 					snprintf(cmdbuf, screen_width,
1038 					    "%s (%s)", argbuf, pp->ki_comm);
1039 			} else {
1040 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1041 				    pp->ki_tdname[0])
1042 					snprintf(cmdbuf, screen_width,
1043 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1044 					    pp->ki_moretdname);
1045 				else
1046 					strlcpy(cmdbuf, argbuf, screen_width);
1047 			}
1048 			free(argbuf);
1049 		}
1050 	}
1051 
1052 	if (displaymode == DISP_IO) {
1053 		oldp = get_old_proc(pp);
1054 		if (oldp != NULL) {
1055 			ru.ru_inblock = RU(pp)->ru_inblock -
1056 			    RU(oldp)->ru_inblock;
1057 			ru.ru_oublock = RU(pp)->ru_oublock -
1058 			    RU(oldp)->ru_oublock;
1059 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1060 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1061 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1062 			rup = &ru;
1063 		} else {
1064 			rup = RU(pp);
1065 		}
1066 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1067 		s_tot = total_inblock + total_oublock + total_majflt;
1068 
1069 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1070 
1071 		if (ps.jail) {
1072 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1073 		}
1074 		sbuf_printf(procbuf, "%-*.*s", namelength, namelength, (*get_userid)(pp->ki_ruid));
1075 		sbuf_printf(procbuf, "%6ld ", rup->ru_nvcsw);
1076 		sbuf_printf(procbuf, "%6ld ", rup->ru_nivcsw);
1077 		sbuf_printf(procbuf, "%6ld ", rup->ru_inblock);
1078 		sbuf_printf(procbuf, "%6ld ", rup->ru_oublock);
1079 		sbuf_printf(procbuf, "%6ld ", rup->ru_majflt);
1080 		sbuf_printf(procbuf, "%6ld ", p_tot);
1081 		sbuf_printf(procbuf, "%6.2f%% ", s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot));
1082 
1083 	} else {
1084 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1085 		if (ps.jail) {
1086 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1087 		}
1088 		sbuf_printf(procbuf, "%-*.*s ", namelength, namelength, (*get_userid)(pp->ki_ruid));
1089 
1090 		if (!ps.thread) {
1091 			sbuf_printf(procbuf, "%4d ", pp->ki_numthreads);
1092 		} else {
1093 			sbuf_printf(procbuf, " ");
1094 		}
1095 
1096 		sbuf_printf(procbuf, "%3d ", pp->ki_pri.pri_level - PZERO);
1097 		sbuf_printf(procbuf, "%4s", format_nice(pp));
1098 		sbuf_printf(procbuf, "%7s ", format_k(PROCSIZE(pp)));
1099 		sbuf_printf(procbuf, "%6s ", format_k(pagetok(pp->ki_rssize)));
1100 		if (ps.swap) {
1101 			sbuf_printf(procbuf, "%*s ",
1102 				TOP_SWAP_LEN - 1,
1103 				format_k(pagetok(ki_swap(pp))));
1104 		}
1105 		sbuf_printf(procbuf, "%-6.6s ", status);
1106 		if (smpmode) {
1107 			int cpu;
1108 			if (state == SRUN && pp->ki_oncpu != NOCPU) {
1109 				cpu = pp->ki_oncpu;
1110 			} else {
1111 				cpu = pp->ki_lastcpu;
1112 			}
1113 			sbuf_printf(procbuf, "%3d ", cpu);
1114 		}
1115 		sbuf_printf(procbuf, "%6s ", format_time(cputime));
1116 		sbuf_printf(procbuf, "%6.2f%% ", ps.wcpu ? 100.0 * weighted_cpu(PCTCPU(pp), pp) : 100.0 * PCTCPU(pp));
1117 	}
1118 	sbuf_printf(procbuf, "%s", cmdbuf);
1119 	free(cmdbuf);
1120 	return (sbuf_data(procbuf));
1121 }
1122 
1123 static void
1124 getsysctl(const char *name, void *ptr, size_t len)
1125 {
1126 	size_t nlen = len;
1127 
1128 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1129 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1130 		    strerror(errno));
1131 		quit(TOP_EX_SYS_ERROR);
1132 	}
1133 	if (nlen != len) {
1134 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1135 		    name, (unsigned long)len, (unsigned long)nlen);
1136 		quit(TOP_EX_SYS_ERROR);
1137 	}
1138 }
1139 
1140 static const char *
1141 format_nice(const struct kinfo_proc *pp)
1142 {
1143 	const char *fifo, *kproc;
1144 	int rtpri;
1145 	static char nicebuf[4 + 1];
1146 
1147 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1148 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1149 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1150 	case PRI_ITHD:
1151 		return ("-");
1152 	case PRI_REALTIME:
1153 		/*
1154 		 * XXX: the kernel doesn't tell us the original rtprio and
1155 		 * doesn't really know what it was, so to recover it we
1156 		 * must be more chummy with the implementation than the
1157 		 * implementation is with itself.  pri_user gives a
1158 		 * constant "base" priority, but is only initialized
1159 		 * properly for user threads.  pri_native gives what the
1160 		 * kernel calls the "base" priority, but it isn't constant
1161 		 * since it is changed by priority propagation.  pri_native
1162 		 * also isn't properly initialized for all threads, but it
1163 		 * is properly initialized for kernel realtime and idletime
1164 		 * threads.  Thus we use pri_user for the base priority of
1165 		 * user threads (it is always correct) and pri_native for
1166 		 * the base priority of kernel realtime and idletime threads
1167 		 * (there is nothing better, and it is usually correct).
1168 		 *
1169 		 * The field width and thus the buffer are too small for
1170 		 * values like "kr31F", but such values shouldn't occur,
1171 		 * and if they do then the tailing "F" is not displayed.
1172 		 */
1173 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1174 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1175 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1176 		    kproc, rtpri, fifo);
1177 		break;
1178 	case PRI_TIMESHARE:
1179 		if (pp->ki_flag & P_KPROC)
1180 			return ("-");
1181 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1182 		break;
1183 	case PRI_IDLE:
1184 		/* XXX: as above. */
1185 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1186 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1187 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1188 		    kproc, rtpri, fifo);
1189 		break;
1190 	default:
1191 		return ("?");
1192 	}
1193 	return (nicebuf);
1194 }
1195 
1196 /* comparison routines for qsort */
1197 
1198 static int
1199 compare_pid(const void *p1, const void *p2)
1200 {
1201 	const struct kinfo_proc * const *pp1 = p1;
1202 	const struct kinfo_proc * const *pp2 = p2;
1203 
1204 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1205 
1206 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1207 }
1208 
1209 static int
1210 compare_tid(const void *p1, const void *p2)
1211 {
1212 	const struct kinfo_proc * const *pp1 = p1;
1213 	const struct kinfo_proc * const *pp2 = p2;
1214 
1215 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1216 
1217 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1218 }
1219 
1220 /*
1221  *  proc_compare - comparison function for "qsort"
1222  *	Compares the resource consumption of two processes using five
1223  *	distinct keys.  The keys (in descending order of importance) are:
1224  *	percent cpu, cpu ticks, state, resident set size, total virtual
1225  *	memory usage.  The process states are ordered as follows (from least
1226  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1227  *	array declaration below maps a process state index into a number
1228  *	that reflects this ordering.
1229  */
1230 
1231 static int sorted_state[] = {
1232 	0,	/* not used		*/
1233 	3,	/* sleep		*/
1234 	1,	/* ABANDONED (WAIT)	*/
1235 	6,	/* run			*/
1236 	5,	/* start		*/
1237 	2,	/* zombie		*/
1238 	4	/* stop			*/
1239 };
1240 
1241 
1242 #define ORDERKEY_PCTCPU(a, b) do { \
1243 	double diff; \
1244 	if (ps.wcpu) \
1245 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1246 		    weighted_cpu(PCTCPU((a)), (a)); \
1247 	else \
1248 		diff = PCTCPU((b)) - PCTCPU((a)); \
1249 	if (diff != 0) \
1250 		return (diff > 0 ? 1 : -1); \
1251 } while (0)
1252 
1253 #define ORDERKEY_CPTICKS(a, b) do { \
1254 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1255 	if (diff != 0) \
1256 		return (diff > 0 ? 1 : -1); \
1257 } while (0)
1258 
1259 #define ORDERKEY_STATE(a, b) do { \
1260 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1261 	if (diff != 0) \
1262 		return (diff > 0 ? 1 : -1); \
1263 } while (0)
1264 
1265 #define ORDERKEY_PRIO(a, b) do { \
1266 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1267 	if (diff != 0) \
1268 		return (diff > 0 ? 1 : -1); \
1269 } while (0)
1270 
1271 #define	ORDERKEY_THREADS(a, b) do { \
1272 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1273 	if (diff != 0) \
1274 		return (diff > 0 ? 1 : -1); \
1275 } while (0)
1276 
1277 #define ORDERKEY_RSSIZE(a, b) do { \
1278 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1279 	if (diff != 0) \
1280 		return (diff > 0 ? 1 : -1); \
1281 } while (0)
1282 
1283 #define ORDERKEY_MEM(a, b) do { \
1284 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1285 	if (diff != 0) \
1286 		return (diff > 0 ? 1 : -1); \
1287 } while (0)
1288 
1289 #define ORDERKEY_JID(a, b) do { \
1290 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1291 	if (diff != 0) \
1292 		return (diff > 0 ? 1 : -1); \
1293 } while (0)
1294 
1295 #define ORDERKEY_SWAP(a, b) do { \
1296 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1297 	if (diff != 0) \
1298 		return (diff > 0 ? 1 : -1); \
1299 } while (0)
1300 
1301 /* compare_cpu - the comparison function for sorting by cpu percentage */
1302 
1303 static int
1304 compare_cpu(const void *arg1, const void *arg2)
1305 {
1306 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1307 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1308 
1309 	ORDERKEY_PCTCPU(p1, p2);
1310 	ORDERKEY_CPTICKS(p1, p2);
1311 	ORDERKEY_STATE(p1, p2);
1312 	ORDERKEY_PRIO(p1, p2);
1313 	ORDERKEY_RSSIZE(p1, p2);
1314 	ORDERKEY_MEM(p1, p2);
1315 
1316 	return (0);
1317 }
1318 
1319 /* compare_size - the comparison function for sorting by total memory usage */
1320 
1321 static int
1322 compare_size(const void *arg1, const void *arg2)
1323 {
1324 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1325 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1326 
1327 	ORDERKEY_MEM(p1, p2);
1328 	ORDERKEY_RSSIZE(p1, p2);
1329 	ORDERKEY_PCTCPU(p1, p2);
1330 	ORDERKEY_CPTICKS(p1, p2);
1331 	ORDERKEY_STATE(p1, p2);
1332 	ORDERKEY_PRIO(p1, p2);
1333 
1334 	return (0);
1335 }
1336 
1337 /* compare_res - the comparison function for sorting by resident set size */
1338 
1339 static int
1340 compare_res(const void *arg1, const void *arg2)
1341 {
1342 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1343 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1344 
1345 	ORDERKEY_RSSIZE(p1, p2);
1346 	ORDERKEY_MEM(p1, p2);
1347 	ORDERKEY_PCTCPU(p1, p2);
1348 	ORDERKEY_CPTICKS(p1, p2);
1349 	ORDERKEY_STATE(p1, p2);
1350 	ORDERKEY_PRIO(p1, p2);
1351 
1352 	return (0);
1353 }
1354 
1355 /* compare_time - the comparison function for sorting by total cpu time */
1356 
1357 static int
1358 compare_time(const void *arg1, const void *arg2)
1359 {
1360 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1361 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1362 
1363 	ORDERKEY_CPTICKS(p1, p2);
1364 	ORDERKEY_PCTCPU(p1, p2);
1365 	ORDERKEY_STATE(p1, p2);
1366 	ORDERKEY_PRIO(p1, p2);
1367 	ORDERKEY_RSSIZE(p1, p2);
1368 	ORDERKEY_MEM(p1, p2);
1369 
1370 	return (0);
1371 }
1372 
1373 /* compare_prio - the comparison function for sorting by priority */
1374 
1375 static int
1376 compare_prio(const void *arg1, const void *arg2)
1377 {
1378 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1379 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1380 
1381 	ORDERKEY_PRIO(p1, p2);
1382 	ORDERKEY_CPTICKS(p1, p2);
1383 	ORDERKEY_PCTCPU(p1, p2);
1384 	ORDERKEY_STATE(p1, p2);
1385 	ORDERKEY_RSSIZE(p1, p2);
1386 	ORDERKEY_MEM(p1, p2);
1387 
1388 	return (0);
1389 }
1390 
1391 /* compare_threads - the comparison function for sorting by threads */
1392 static int
1393 compare_threads(const void *arg1, const void *arg2)
1394 {
1395 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1396 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1397 
1398 	ORDERKEY_THREADS(p1, p2);
1399 	ORDERKEY_PCTCPU(p1, p2);
1400 	ORDERKEY_CPTICKS(p1, p2);
1401 	ORDERKEY_STATE(p1, p2);
1402 	ORDERKEY_PRIO(p1, p2);
1403 	ORDERKEY_RSSIZE(p1, p2);
1404 	ORDERKEY_MEM(p1, p2);
1405 
1406 	return (0);
1407 }
1408 
1409 /* compare_jid - the comparison function for sorting by jid */
1410 static int
1411 compare_jid(const void *arg1, const void *arg2)
1412 {
1413 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1414 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1415 
1416 	ORDERKEY_JID(p1, p2);
1417 	ORDERKEY_PCTCPU(p1, p2);
1418 	ORDERKEY_CPTICKS(p1, p2);
1419 	ORDERKEY_STATE(p1, p2);
1420 	ORDERKEY_PRIO(p1, p2);
1421 	ORDERKEY_RSSIZE(p1, p2);
1422 	ORDERKEY_MEM(p1, p2);
1423 
1424 	return (0);
1425 }
1426 
1427 /* compare_swap - the comparison function for sorting by swap */
1428 static int
1429 compare_swap(const void *arg1, const void *arg2)
1430 {
1431 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1432 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1433 
1434 	ORDERKEY_SWAP(p1, p2);
1435 	ORDERKEY_PCTCPU(p1, p2);
1436 	ORDERKEY_CPTICKS(p1, p2);
1437 	ORDERKEY_STATE(p1, p2);
1438 	ORDERKEY_PRIO(p1, p2);
1439 	ORDERKEY_RSSIZE(p1, p2);
1440 	ORDERKEY_MEM(p1, p2);
1441 
1442 	return (0);
1443 }
1444 
1445 /* assorted comparison functions for sorting by i/o */
1446 
1447 static int
1448 compare_iototal(const void *arg1, const void *arg2)
1449 {
1450 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1451 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1452 
1453 	return (get_io_total(p2) - get_io_total(p1));
1454 }
1455 
1456 static int
1457 compare_ioread(const void *arg1, const void *arg2)
1458 {
1459 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1460 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1461 	long dummy, inp1, inp2;
1462 
1463 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1464 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1465 
1466 	return (inp2 - inp1);
1467 }
1468 
1469 static int
1470 compare_iowrite(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 	long dummy, oup1, oup2;
1475 
1476 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1477 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1478 
1479 	return (oup2 - oup1);
1480 }
1481 
1482 static int
1483 compare_iofault(const void *arg1, const void *arg2)
1484 {
1485 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1486 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1487 	long dummy, flp1, flp2;
1488 
1489 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1490 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1491 
1492 	return (flp2 - flp1);
1493 }
1494 
1495 static int
1496 compare_vcsw(const void *arg1, const void *arg2)
1497 {
1498 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1499 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1500 	long dummy, flp1, flp2;
1501 
1502 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1503 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1504 
1505 	return (flp2 - flp1);
1506 }
1507 
1508 static int
1509 compare_ivcsw(const void *arg1, const void *arg2)
1510 {
1511 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1512 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1513 	long dummy, flp1, flp2;
1514 
1515 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1516 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1517 
1518 	return (flp2 - flp1);
1519 }
1520 
1521 int (*compares[])(const void *arg1, const void *arg2) = {
1522 	compare_cpu,
1523 	compare_size,
1524 	compare_res,
1525 	compare_time,
1526 	compare_prio,
1527 	compare_threads,
1528 	compare_iototal,
1529 	compare_ioread,
1530 	compare_iowrite,
1531 	compare_iofault,
1532 	compare_vcsw,
1533 	compare_ivcsw,
1534 	compare_jid,
1535 	compare_swap,
1536 	NULL
1537 };
1538 
1539 
1540 static int
1541 swapmode(int *retavail, int *retfree)
1542 {
1543 	int n;
1544 	struct kvm_swap swapary[1];
1545 	static int pagesize = 0;
1546 	static unsigned long swap_maxpages = 0;
1547 
1548 	*retavail = 0;
1549 	*retfree = 0;
1550 
1551 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1552 
1553 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1554 	if (n < 0 || swapary[0].ksw_total == 0)
1555 		return (0);
1556 
1557 	if (pagesize == 0)
1558 		pagesize = getpagesize();
1559 	if (swap_maxpages == 0)
1560 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1561 
1562 	/* ksw_total contains the total size of swap all devices which may
1563 	   exceed the maximum swap size allocatable in the system */
1564 	if ( swapary[0].ksw_total > swap_maxpages )
1565 		swapary[0].ksw_total = swap_maxpages;
1566 
1567 	*retavail = CONVERT(swapary[0].ksw_total);
1568 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1569 
1570 #undef CONVERT
1571 
1572 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1573 	return (n);
1574 }
1575