xref: /freebsd/usr.bin/top/machine.c (revision 74521224)
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 
154 /* these are for keeping track of the proc array */
155 
156 static int nproc;
157 static int onproc = -1;
158 static int pref_len;
159 static struct kinfo_proc *pbase;
160 static struct kinfo_proc **pref;
161 static struct kinfo_proc *previous_procs;
162 static struct kinfo_proc **previous_pref;
163 static int previous_proc_count = 0;
164 static int previous_proc_count_max = 0;
165 static int previous_thread;
166 
167 /* data used for recalculating pctcpu */
168 static double *pcpu;
169 static struct timespec proc_uptime;
170 static struct timeval proc_wall_time;
171 static struct timeval previous_wall_time;
172 static uint64_t previous_interval = 0;
173 
174 /* total number of io operations */
175 static long total_inblock;
176 static long total_oublock;
177 static long total_majflt;
178 
179 /* these are for getting the memory statistics */
180 
181 static int arc_enabled;
182 static int carc_enabled;
183 static 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  */
196 static 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 */
203 static int maxcpu;
204 static int maxid;
205 static int ncpus;
206 static unsigned long cpumask;
207 static long *times;
208 static long *pcpu_cp_time;
209 static long *pcpu_cp_old;
210 static long *pcpu_cp_diff;
211 static int *pcpu_cpu_states;
212 
213 static int compare_swap(const void *a, const void *b);
214 static int compare_jid(const void *a, const void *b);
215 static int compare_pid(const void *a, const void *b);
216 static int compare_tid(const void *a, const void *b);
217 static const char *format_nice(const struct kinfo_proc *pp);
218 static void getsysctl(const char *name, void *ptr, size_t len);
219 static int swapmode(int *retavail, int *retfree);
220 static void update_layout(void);
221 static int find_uid(uid_t needle, int *haystack);
222 
223 static int
224 find_uid(uid_t needle, int *haystack)
225 {
226 	size_t i = 0;
227 
228 	for (; i < TOP_MAX_UIDS; ++i)
229 		if ((uid_t)haystack[i] == needle)
230 			return 1;
231 	return (0);
232 }
233 
234 void
235 toggle_pcpustats(void)
236 {
237 
238 	if (ncpus == 1)
239 		return;
240 	update_layout();
241 }
242 
243 /* Adjust display based on ncpus and the ARC state. */
244 static void
245 update_layout(void)
246 {
247 
248 	y_mem = 3;
249 	y_arc = 4;
250 	y_carc = 5;
251 	y_swap = 4 + arc_enabled + carc_enabled;
252 	y_idlecursor = 5 + arc_enabled + carc_enabled;
253 	y_message = 5 + arc_enabled + carc_enabled;
254 	y_header = 6 + arc_enabled + carc_enabled;
255 	y_procs = 7 + arc_enabled + carc_enabled;
256 	Header_lines = 7 + arc_enabled + carc_enabled;
257 
258 	if (pcpu_stats) {
259 		y_mem += ncpus - 1;
260 		y_arc += ncpus - 1;
261 		y_carc += ncpus - 1;
262 		y_swap += ncpus - 1;
263 		y_idlecursor += ncpus - 1;
264 		y_message += ncpus - 1;
265 		y_header += ncpus - 1;
266 		y_procs += ncpus - 1;
267 		Header_lines += ncpus - 1;
268 	}
269 }
270 
271 int
272 machine_init(struct statics *statics)
273 {
274 	int i, j, empty, pagesize;
275 	uint64_t arc_size;
276 	int carc_en;
277 	size_t size;
278 
279 	size = sizeof(smpmode);
280 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
281 	    NULL, 0) != 0 &&
282 	    sysctlbyname("kern.smp.active", &smpmode, &size,
283 	    NULL, 0) != 0) ||
284 	    size != sizeof(smpmode))
285 		smpmode = 0;
286 
287 	size = sizeof(arc_size);
288 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
289 	    NULL, 0) == 0 && arc_size != 0)
290 		arc_enabled = 1;
291 	size = sizeof(carc_en);
292 	if (arc_enabled &&
293 	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
294 	    NULL, 0) == 0 && carc_en == 1)
295 		carc_enabled = 1;
296 
297 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
298 	if (kd == NULL)
299 		return (-1);
300 
301 	GETSYSCTL("kern.ccpu", ccpu);
302 
303 	/* this is used in calculating WCPU -- calculate it ahead of time */
304 	logcpu = log(loaddouble(ccpu));
305 
306 	pbase = NULL;
307 	pref = NULL;
308 	pcpu = NULL;
309 	nproc = 0;
310 	onproc = -1;
311 
312 	/* get the page size and calculate pageshift from it */
313 	pagesize = getpagesize();
314 	pageshift = 0;
315 	while (pagesize > 1) {
316 		pageshift++;
317 		pagesize >>= 1;
318 	}
319 
320 	/* we only need the amount of log(2)1024 for our conversion */
321 	pageshift -= LOG1024;
322 
323 	/* fill in the statics information */
324 	statics->procstate_names = procstatenames;
325 	statics->cpustate_names = cpustatenames;
326 	statics->memory_names = memorynames;
327 	if (arc_enabled)
328 		statics->arc_names = arcnames;
329 	else
330 		statics->arc_names = NULL;
331 	if (carc_enabled)
332 		statics->carc_names = carcnames;
333 	else
334 		statics->carc_names = NULL;
335 	statics->swap_names = swapnames;
336 	statics->order_names = ordernames;
337 
338 	/* Allocate state for per-CPU stats. */
339 	cpumask = 0;
340 	ncpus = 0;
341 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
342 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
343 	if (times == NULL)
344 		err(1, "calloc for kern.smp.maxcpus");
345 	size = sizeof(long) * maxcpu * CPUSTATES;
346 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
347 		err(1, "sysctlbyname kern.cp_times");
348 	pcpu_cp_time = calloc(1, size);
349 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
350 	for (i = 0; i <= maxid; i++) {
351 		empty = 1;
352 		for (j = 0; empty && j < CPUSTATES; j++) {
353 			if (times[i * CPUSTATES + j] != 0)
354 				empty = 0;
355 		}
356 		if (!empty) {
357 			cpumask |= (1ul << i);
358 			ncpus++;
359 		}
360 	}
361 	assert(ncpus > 0);
362 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
363 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
364 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
365 	statics->ncpus = ncpus;
366 
367 	update_layout();
368 
369 	/* all done! */
370 	return (0);
371 }
372 
373 char *
374 format_header(const char *uname_field)
375 {
376 	static struct sbuf* header = NULL;
377 
378 	/* clean up from last time. */
379 	if (header != NULL) {
380 		sbuf_clear(header);
381 	} else {
382 		header = sbuf_new_auto();
383 	}
384 
385 	switch (displaymode) {
386 	case DISP_CPU: {
387 		sbuf_printf(header, "  %s", ps.thread_id ? " THR" : "PID");
388 		sbuf_printf(header, "%*s", ps.jail ? TOP_JID_LEN : 0,
389 									ps.jail ? " JID" : "");
390 		sbuf_printf(header, " %-*.*s  ", namelength, namelength, uname_field);
391 		sbuf_cat(header, "THR PRI NICE  SIZE   RES ");
392 		if (ps.swap) {
393 			sbuf_printf(header, "%*s ", TOP_SWAP_LEN - 1, "SWAP");
394 		}
395 		sbuf_cat(header, "STATE    ");
396 		if (smpmode) {
397 			sbuf_cat(header, "C   ");
398 		}
399 		sbuf_cat(header, "TIME ");
400 		sbuf_printf(header, " %6s ", ps.wcpu ? "WCPU" : "CPU");
401 		sbuf_cat(header, "COMMAND");
402 		sbuf_finish(header);
403 		break;
404 	}
405 	case DISP_IO: {
406 		sbuf_printf(header, "  %s%*s %-*.*s",
407 			ps.thread_id ? " THR" : "PID",
408 		    ps.jail ? TOP_JID_LEN : 0, ps.jail ? " JID" : "",
409 		    namelength, namelength, uname_field);
410 		sbuf_cat(header, "   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND");
411 		sbuf_finish(header);
412 		break;
413 	}
414 	case DISP_MAX:
415 		assert("displaymode must not be set to DISP_MAX");
416 	}
417 
418 	return sbuf_data(header);
419 }
420 
421 static int swappgsin = -1;
422 static int swappgsout = -1;
423 
424 
425 void
426 get_system_info(struct system_info *si)
427 {
428 	struct loadavg sysload;
429 	int mib[2];
430 	struct timeval boottime;
431 	uint64_t arc_stat, arc_stat2;
432 	int i, j;
433 	size_t size;
434 
435 	/* get the CPU stats */
436 	size = (maxid + 1) * CPUSTATES * sizeof(long);
437 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
438 		err(1, "sysctlbyname kern.cp_times");
439 	GETSYSCTL("kern.cp_time", cp_time);
440 	GETSYSCTL("vm.loadavg", sysload);
441 	GETSYSCTL("kern.lastpid", lastpid);
442 
443 	/* convert load averages to doubles */
444 	for (i = 0; i < 3; i++)
445 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
446 
447 	/* convert cp_time counts to percentages */
448 	for (i = j = 0; i <= maxid; i++) {
449 		if ((cpumask & (1ul << i)) == 0)
450 			continue;
451 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
452 		    &pcpu_cp_time[j * CPUSTATES],
453 		    &pcpu_cp_old[j * CPUSTATES],
454 		    &pcpu_cp_diff[j * CPUSTATES]);
455 		j++;
456 	}
457 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
458 
459 	/* sum memory & swap statistics */
460 	{
461 		static unsigned int swap_delay = 0;
462 		static int swapavail = 0;
463 		static int swapfree = 0;
464 		static long bufspace = 0;
465 		static uint64_t nspgsin, nspgsout;
466 
467 		GETSYSCTL("vfs.bufspace", bufspace);
468 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
469 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
470 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
471 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
472 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
473 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
474 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
475 		/* convert memory stats to Kbytes */
476 		memory_stats[0] = pagetok(memory_stats[0]);
477 		memory_stats[1] = pagetok(memory_stats[1]);
478 		memory_stats[2] = pagetok(memory_stats[2]);
479 		memory_stats[3] = pagetok(memory_stats[3]);
480 		memory_stats[4] = bufspace / 1024;
481 		memory_stats[5] = pagetok(memory_stats[5]);
482 		memory_stats[6] = -1;
483 
484 		/* first interval */
485 		if (swappgsin < 0) {
486 			swap_stats[4] = 0;
487 			swap_stats[5] = 0;
488 		}
489 
490 		/* compute differences between old and new swap statistic */
491 		else {
492 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
493 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
494 		}
495 
496 		swappgsin = nspgsin;
497 		swappgsout = nspgsout;
498 
499 		/* call CPU heavy swapmode() only for changes */
500 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
501 			swap_stats[3] = swapmode(&swapavail, &swapfree);
502 			swap_stats[0] = swapavail;
503 			swap_stats[1] = swapavail - swapfree;
504 			swap_stats[2] = swapfree;
505 		}
506 		swap_delay = 1;
507 		swap_stats[6] = -1;
508 	}
509 
510 	if (arc_enabled) {
511 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
512 		arc_stats[0] = arc_stat >> 10;
513 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
514 		arc_stats[1] = arc_stat >> 10;
515 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
516 		arc_stats[2] = arc_stat >> 10;
517 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
518 		arc_stats[3] = arc_stat >> 10;
519 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
520 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
521 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
522 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
523 		arc_stats[5] = arc_stat >> 10;
524 		si->arc = arc_stats;
525 	}
526 	if (carc_enabled) {
527 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
528 		carc_stats[0] = arc_stat >> 10;
529 		carc_stats[2] = arc_stat >> 10; /* For ratio */
530 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
531 		carc_stats[1] = arc_stat >> 10;
532 		si->carc = carc_stats;
533 	}
534 
535 	/* set arrays and strings */
536 	if (pcpu_stats) {
537 		si->cpustates = pcpu_cpu_states;
538 		si->ncpus = ncpus;
539 	} else {
540 		si->cpustates = cpu_states;
541 		si->ncpus = 1;
542 	}
543 	si->memory = memory_stats;
544 	si->swap = swap_stats;
545 
546 
547 	if (lastpid > 0) {
548 		si->last_pid = lastpid;
549 	} else {
550 		si->last_pid = -1;
551 	}
552 
553 	/*
554 	 * Print how long system has been up.
555 	 * (Found by looking getting "boottime" from the kernel)
556 	 */
557 	mib[0] = CTL_KERN;
558 	mib[1] = KERN_BOOTTIME;
559 	size = sizeof(boottime);
560 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
561 	    boottime.tv_sec != 0) {
562 		si->boottime = boottime;
563 	} else {
564 		si->boottime.tv_sec = -1;
565 	}
566 }
567 
568 #define NOPROC	((void *)-1)
569 
570 /*
571  * We need to compare data from the old process entry with the new
572  * process entry.
573  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
574  * structure to cache the mapping.  We also use a negative cache pointer
575  * of NOPROC to avoid duplicate lookups.
576  * XXX: this could be done when the actual processes are fetched, we do
577  * it here out of laziness.
578  */
579 static const struct kinfo_proc *
580 get_old_proc(struct kinfo_proc *pp)
581 {
582 	const struct kinfo_proc * const *oldpp, *oldp;
583 
584 	/*
585 	 * If this is the first fetch of the kinfo_procs then we don't have
586 	 * any previous entries.
587 	 */
588 	if (previous_proc_count == 0)
589 		return (NULL);
590 	/* negative cache? */
591 	if (pp->ki_udata == NOPROC)
592 		return (NULL);
593 	/* cached? */
594 	if (pp->ki_udata != NULL)
595 		return (pp->ki_udata);
596 	/*
597 	 * Not cached,
598 	 * 1) look up based on pid.
599 	 * 2) compare process start.
600 	 * If we fail here, then setup a negative cache entry, otherwise
601 	 * cache it.
602 	 */
603 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
604 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
605 	if (oldpp == NULL) {
606 		pp->ki_udata = NOPROC;
607 		return (NULL);
608 	}
609 	oldp = *oldpp;
610 	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
611 		pp->ki_udata = NOPROC;
612 		return (NULL);
613 	}
614 	pp->ki_udata = oldp;
615 	return (oldp);
616 }
617 
618 /*
619  * Return the total amount of IO done in blocks in/out and faults.
620  * store the values individually in the pointers passed in.
621  */
622 static long
623 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
624     long *vcsw, long *ivcsw)
625 {
626 	const struct kinfo_proc *oldp;
627 	static struct kinfo_proc dummy;
628 	long ret;
629 
630 	oldp = get_old_proc(pp);
631 	if (oldp == NULL) {
632 		memset(&dummy, 0, sizeof(dummy));
633 		oldp = &dummy;
634 	}
635 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
636 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
637 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
638 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
639 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
640 	ret =
641 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
642 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
643 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
644 	return (ret);
645 }
646 
647 /*
648  * If there was a previous update, use the delta in ki_runtime over
649  * the previous interval to calculate pctcpu.  Otherwise, fall back
650  * to using the kernel's ki_pctcpu.
651  */
652 static double
653 proc_calc_pctcpu(struct kinfo_proc *pp)
654 {
655 	const struct kinfo_proc *oldp;
656 
657 	if (previous_interval != 0) {
658 		oldp = get_old_proc(pp);
659 		if (oldp != NULL)
660 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
661 			    / previous_interval);
662 
663 		/*
664 		 * If this process/thread was created during the previous
665 		 * interval, charge it's total runtime to the previous
666 		 * interval.
667 		 */
668 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
669 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
670 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
671 			return ((double)pp->ki_runtime / previous_interval);
672 	}
673 	return (pctdouble(pp->ki_pctcpu));
674 }
675 
676 /*
677  * Return true if this process has used any CPU time since the
678  * previous update.
679  */
680 static int
681 proc_used_cpu(struct kinfo_proc *pp)
682 {
683 	const struct kinfo_proc *oldp;
684 
685 	oldp = get_old_proc(pp);
686 	if (oldp == NULL)
687 		return (PCTCPU(pp) != 0);
688 	return (pp->ki_runtime != oldp->ki_runtime ||
689 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
690 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
691 }
692 
693 /*
694  * Return the total number of block in/out and faults by a process.
695  */
696 static long
697 get_io_total(const struct kinfo_proc *pp)
698 {
699 	long dummy;
700 
701 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
702 }
703 
704 static struct handle handle;
705 
706 void *
707 get_process_info(struct system_info *si, struct process_select *sel,
708     int (*compare)(const void *, const void *))
709 {
710 	int i;
711 	int total_procs;
712 	long p_io;
713 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
714 	long nsec;
715 	int active_procs;
716 	struct kinfo_proc **prefp;
717 	struct kinfo_proc *pp;
718 	struct timespec previous_proc_uptime;
719 
720 	/*
721 	 * If thread state was toggled, don't cache the previous processes.
722 	 */
723 	if (previous_thread != sel->thread)
724 		nproc = 0;
725 	previous_thread = sel->thread;
726 
727 	/*
728 	 * Save the previous process info.
729 	 */
730 	if (previous_proc_count_max < nproc) {
731 		free(previous_procs);
732 		previous_procs = calloc(nproc, sizeof(*previous_procs));
733 		free(previous_pref);
734 		previous_pref = calloc(nproc, sizeof(*previous_pref));
735 		if (previous_procs == NULL || previous_pref == NULL) {
736 			fprintf(stderr, "top: Out of memory.\n");
737 			quit(TOP_EX_SYS_ERROR);
738 		}
739 		previous_proc_count_max = nproc;
740 	}
741 	if (nproc) {
742 		for (i = 0; i < nproc; i++)
743 			previous_pref[i] = &previous_procs[i];
744 		memcpy(previous_procs, pbase, nproc * sizeof(*previous_procs));
745 		qsort(previous_pref, nproc, sizeof(*previous_pref),
746 		    ps.thread ? compare_tid : compare_pid);
747 	}
748 	previous_proc_count = nproc;
749 	previous_proc_uptime = proc_uptime;
750 	previous_wall_time = proc_wall_time;
751 	previous_interval = 0;
752 
753 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
754 	    0, &nproc);
755 	gettimeofday(&proc_wall_time, NULL);
756 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
757 		memset(&proc_uptime, 0, sizeof(proc_uptime));
758 	else if (previous_proc_uptime.tv_sec != 0 &&
759 	    previous_proc_uptime.tv_nsec != 0) {
760 		previous_interval = (proc_uptime.tv_sec -
761 		    previous_proc_uptime.tv_sec) * 1000000;
762 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
763 		if (nsec < 0) {
764 			previous_interval -= 1000000;
765 			nsec += 1000000000;
766 		}
767 		previous_interval += nsec / 1000;
768 	}
769 	if (nproc > onproc) {
770 		pref = realloc(pref, sizeof(*pref) * nproc);
771 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
772 		onproc = nproc;
773 	}
774 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
775 		fprintf(stderr, "top: Out of memory.\n");
776 		quit(TOP_EX_SYS_ERROR);
777 	}
778 	/* get a pointer to the states summary array */
779 	si->procstates = process_states;
780 
781 	/* count up process states and get pointers to interesting procs */
782 	total_procs = 0;
783 	active_procs = 0;
784 	total_inblock = 0;
785 	total_oublock = 0;
786 	total_majflt = 0;
787 	memset(process_states, 0, sizeof(process_states));
788 	prefp = pref;
789 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
790 
791 		if (pp->ki_stat == 0)
792 			/* not in use */
793 			continue;
794 
795 		if (!sel->self && pp->ki_pid == mypid && sel->pid == -1)
796 			/* skip self */
797 			continue;
798 
799 		if (!sel->system && (pp->ki_flag & P_SYSTEM) && sel->pid == -1)
800 			/* skip system process */
801 			continue;
802 
803 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
804 		    &p_vcsw, &p_ivcsw);
805 		total_inblock += p_inblock;
806 		total_oublock += p_oublock;
807 		total_majflt += p_majflt;
808 		total_procs++;
809 		process_states[(unsigned char)pp->ki_stat]++;
810 
811 		if (pp->ki_stat == SZOMB)
812 			/* skip zombies */
813 			continue;
814 
815 		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD && sel->pid == -1)
816 			/* skip kernel idle process */
817 			continue;
818 
819 		PCTCPU(pp) = proc_calc_pctcpu(pp);
820 		if (sel->thread && PCTCPU(pp) > 1.0)
821 			PCTCPU(pp) = 1.0;
822 		if (displaymode == DISP_CPU && !sel->idle &&
823 		    (!proc_used_cpu(pp) ||
824 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
825 			/* skip idle or non-running processes */
826 			continue;
827 
828 		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
829 			/* skip processes that aren't doing I/O */
830 			continue;
831 
832 		if (sel->jid != -1 && pp->ki_jid != sel->jid)
833 			/* skip proc. that don't belong to the selected JID */
834 			continue;
835 
836 		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
837 			/* skip proc. that don't belong to the selected UID */
838 			continue;
839 
840 		if (sel->pid != -1 && pp->ki_pid != sel->pid)
841 			continue;
842 
843 		*prefp++ = pp;
844 		active_procs++;
845 	}
846 
847 	/* if requested, sort the "interesting" processes */
848 	if (compare != NULL)
849 		qsort(pref, active_procs, sizeof(*pref), compare);
850 
851 	/* remember active and total counts */
852 	si->p_total = total_procs;
853 	si->p_pactive = pref_len = active_procs;
854 
855 	/* pass back a handle */
856 	handle.next_proc = pref;
857 	handle.remaining = active_procs;
858 	return (&handle);
859 }
860 
861 char *
862 format_next_process(struct handle * xhandle, char *(*get_userid)(int), int flags)
863 {
864 	struct kinfo_proc *pp;
865 	const struct kinfo_proc *oldp;
866 	long cputime;
867 	char status[22];
868 	size_t state;
869 	struct rusage ru, *rup;
870 	long p_tot, s_tot;
871 	char *cmdbuf = NULL;
872 	char **args;
873 	static struct sbuf* procbuf = NULL;
874 
875 	/* clean up from last time. */
876 	if (procbuf != NULL) {
877 		sbuf_clear(procbuf);
878 	} else {
879 		procbuf = sbuf_new_auto();
880 	}
881 
882 
883 	/* find and remember the next proc structure */
884 	pp = *(xhandle->next_proc++);
885 	xhandle->remaining--;
886 
887 	/* get the process's command name */
888 	if ((pp->ki_flag & P_INMEM) == 0) {
889 		/*
890 		 * Print swapped processes as <pname>
891 		 */
892 		size_t len;
893 
894 		len = strlen(pp->ki_comm);
895 		if (len > sizeof(pp->ki_comm) - 3)
896 			len = sizeof(pp->ki_comm) - 3;
897 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
898 		pp->ki_comm[0] = '<';
899 		pp->ki_comm[len + 1] = '>';
900 		pp->ki_comm[len + 2] = '\0';
901 	}
902 
903 	/*
904 	 * Convert the process's runtime from microseconds to seconds.  This
905 	 * time includes the interrupt time although that is not wanted here.
906 	 * ps(1) is similarly sloppy.
907 	 */
908 	cputime = (pp->ki_runtime + 500000) / 1000000;
909 
910 	/* generate "STATE" field */
911 	switch (state = pp->ki_stat) {
912 	case SRUN:
913 		if (smpmode && pp->ki_oncpu != NOCPU)
914 			sprintf(status, "CPU%d", pp->ki_oncpu);
915 		else
916 			strcpy(status, "RUN");
917 		break;
918 	case SLOCK:
919 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
920 			sprintf(status, "*%.6s", pp->ki_lockname);
921 			break;
922 		}
923 		/* fall through */
924 	case SSLEEP:
925 		sprintf(status, "%.6s", pp->ki_wmesg);
926 		break;
927 	default:
928 
929 		if (state < nitems(state_abbrev)) {
930 			sprintf(status, "%.6s", state_abbrev[state]);
931 		} else {
932 			sprintf(status, "?%5zu", state);
933 		}
934 		break;
935 	}
936 
937 	cmdbuf = calloc(screen_width + 1, 1);
938 	if (cmdbuf == NULL) {
939 		warn("calloc(%d)", screen_width + 1);
940 		return NULL;
941 	}
942 
943 	if (!(flags & FMT_SHOWARGS)) {
944 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
945 		    pp->ki_tdname[0]) {
946 			snprintf(cmdbuf, screen_width, "%s{%s%s}", pp->ki_comm,
947 			    pp->ki_tdname, pp->ki_moretdname);
948 		} else {
949 			snprintf(cmdbuf, screen_width, "%s", pp->ki_comm);
950 		}
951 	} else {
952 		if (pp->ki_flag & P_SYSTEM ||
953 		    (args = kvm_getargv(kd, pp, screen_width)) == NULL ||
954 		    !(*args)) {
955 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
956 		    	    pp->ki_tdname[0]) {
957 				snprintf(cmdbuf, screen_width,
958 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
959 				    pp->ki_moretdname);
960 			} else {
961 				snprintf(cmdbuf, screen_width,
962 				    "[%s]", pp->ki_comm);
963 			}
964 		} else {
965 			const char *src;
966 			char *dst, *argbuf;
967 			const char *cmd;
968 			size_t argbuflen;
969 			size_t len;
970 
971 			argbuflen = screen_width * 4;
972 			argbuf = calloc(argbuflen + 1, 1);
973 			if (argbuf == NULL) {
974 				warn("calloc(%zu)", argbuflen + 1);
975 				free(cmdbuf);
976 				return NULL;
977 			}
978 
979 			dst = argbuf;
980 
981 			/* Extract cmd name from argv */
982 			cmd = basename(*args);
983 
984 			for (; (src = *args++) != NULL; ) {
985 				if (*src == '\0')
986 					continue;
987 				len = (argbuflen - (dst - argbuf) - 1) / 4;
988 				strvisx(dst, src,
989 				    MIN(strlen(src), len),
990 				    VIS_NL | VIS_CSTYLE);
991 				while (*dst != '\0')
992 					dst++;
993 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
994 					*dst++ = ' '; /* add delimiting space */
995 			}
996 			if (dst != argbuf && dst[-1] == ' ')
997 				dst--;
998 			*dst = '\0';
999 
1000 			if (strcmp(cmd, pp->ki_comm) != 0) {
1001 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1002 				    pp->ki_tdname[0])
1003 					snprintf(cmdbuf, screen_width,
1004 					    "%s (%s){%s%s}", argbuf,
1005 					    pp->ki_comm, pp->ki_tdname,
1006 					    pp->ki_moretdname);
1007 				else
1008 					snprintf(cmdbuf, screen_width,
1009 					    "%s (%s)", argbuf, pp->ki_comm);
1010 			} else {
1011 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1012 				    pp->ki_tdname[0])
1013 					snprintf(cmdbuf, screen_width,
1014 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1015 					    pp->ki_moretdname);
1016 				else
1017 					strlcpy(cmdbuf, argbuf, screen_width);
1018 			}
1019 			free(argbuf);
1020 		}
1021 	}
1022 
1023 	if (displaymode == DISP_IO) {
1024 		oldp = get_old_proc(pp);
1025 		if (oldp != NULL) {
1026 			ru.ru_inblock = RU(pp)->ru_inblock -
1027 			    RU(oldp)->ru_inblock;
1028 			ru.ru_oublock = RU(pp)->ru_oublock -
1029 			    RU(oldp)->ru_oublock;
1030 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1031 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1032 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1033 			rup = &ru;
1034 		} else {
1035 			rup = RU(pp);
1036 		}
1037 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1038 		s_tot = total_inblock + total_oublock + total_majflt;
1039 
1040 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1041 
1042 		if (ps.jail) {
1043 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1044 		}
1045 		sbuf_printf(procbuf, "%-*.*s", namelength, namelength, (*get_userid)(pp->ki_ruid));
1046 		sbuf_printf(procbuf, "%6ld ", rup->ru_nvcsw);
1047 		sbuf_printf(procbuf, "%6ld ", rup->ru_nivcsw);
1048 		sbuf_printf(procbuf, "%6ld ", rup->ru_inblock);
1049 		sbuf_printf(procbuf, "%6ld ", rup->ru_oublock);
1050 		sbuf_printf(procbuf, "%6ld ", rup->ru_majflt);
1051 		sbuf_printf(procbuf, "%6ld ", p_tot);
1052 		sbuf_printf(procbuf, "%6.2f%% ", s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot));
1053 
1054 	} else {
1055 		sbuf_printf(procbuf, "%5d ", (ps.thread_id) ? pp->ki_tid : pp->ki_pid);
1056 		if (ps.jail) {
1057 			sbuf_printf(procbuf, "%*d ", TOP_JID_LEN - 1, pp->ki_jid);
1058 		}
1059 		sbuf_printf(procbuf, "%-*.*s ", namelength, namelength, (*get_userid)(pp->ki_ruid));
1060 
1061 		if (!ps.thread) {
1062 			sbuf_printf(procbuf, "%4d ", pp->ki_numthreads);
1063 		}
1064 
1065 		sbuf_printf(procbuf, "%3d ", pp->ki_pri.pri_level - PZERO);
1066 		sbuf_printf(procbuf, "%4s", format_nice(pp));
1067 		sbuf_printf(procbuf, "%6s ", format_k(PROCSIZE(pp)));
1068 		sbuf_printf(procbuf, "%5s ", format_k(pagetok(pp->ki_rssize)));
1069 		if (ps.swap) {
1070 			sbuf_printf(procbuf, "%*s ",
1071 				TOP_SWAP_LEN - 1,
1072 				format_k(pagetok(ki_swap(pp))));
1073 		}
1074 		sbuf_printf(procbuf, "%-6.6s ", status);
1075 		if (smpmode) {
1076 			int cpu;
1077 			if (state == SRUN && pp->ki_oncpu != NOCPU) {
1078 				cpu = pp->ki_oncpu;
1079 			} else {
1080 				cpu = pp->ki_lastcpu;
1081 			}
1082 			sbuf_printf(procbuf, "%3d ", cpu);
1083 		}
1084 		sbuf_printf(procbuf, "%6s ", format_time(cputime));
1085 		sbuf_printf(procbuf, "%6.2f%% ", ps.wcpu ? 100.0 * weighted_cpu(PCTCPU(pp), pp) : 100.0 * PCTCPU(pp));
1086 	}
1087 	sbuf_printf(procbuf, "%s", printable(cmdbuf));
1088 	free(cmdbuf);
1089 	return (sbuf_data(procbuf));
1090 }
1091 
1092 static void
1093 getsysctl(const char *name, void *ptr, size_t len)
1094 {
1095 	size_t nlen = len;
1096 
1097 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1098 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1099 		    strerror(errno));
1100 		quit(TOP_EX_SYS_ERROR);
1101 	}
1102 	if (nlen != len) {
1103 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1104 		    name, (unsigned long)len, (unsigned long)nlen);
1105 		quit(TOP_EX_SYS_ERROR);
1106 	}
1107 }
1108 
1109 static const char *
1110 format_nice(const struct kinfo_proc *pp)
1111 {
1112 	const char *fifo, *kproc;
1113 	int rtpri;
1114 	static char nicebuf[4 + 1];
1115 
1116 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1117 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1118 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1119 	case PRI_ITHD:
1120 		return ("-");
1121 	case PRI_REALTIME:
1122 		/*
1123 		 * XXX: the kernel doesn't tell us the original rtprio and
1124 		 * doesn't really know what it was, so to recover it we
1125 		 * must be more chummy with the implementation than the
1126 		 * implementation is with itself.  pri_user gives a
1127 		 * constant "base" priority, but is only initialized
1128 		 * properly for user threads.  pri_native gives what the
1129 		 * kernel calls the "base" priority, but it isn't constant
1130 		 * since it is changed by priority propagation.  pri_native
1131 		 * also isn't properly initialized for all threads, but it
1132 		 * is properly initialized for kernel realtime and idletime
1133 		 * threads.  Thus we use pri_user for the base priority of
1134 		 * user threads (it is always correct) and pri_native for
1135 		 * the base priority of kernel realtime and idletime threads
1136 		 * (there is nothing better, and it is usually correct).
1137 		 *
1138 		 * The field width and thus the buffer are too small for
1139 		 * values like "kr31F", but such values shouldn't occur,
1140 		 * and if they do then the tailing "F" is not displayed.
1141 		 */
1142 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1143 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1144 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1145 		    kproc, rtpri, fifo);
1146 		break;
1147 	case PRI_TIMESHARE:
1148 		if (pp->ki_flag & P_KPROC)
1149 			return ("-");
1150 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1151 		break;
1152 	case PRI_IDLE:
1153 		/* XXX: as above. */
1154 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1155 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1156 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1157 		    kproc, rtpri, fifo);
1158 		break;
1159 	default:
1160 		return ("?");
1161 	}
1162 	return (nicebuf);
1163 }
1164 
1165 /* comparison routines for qsort */
1166 
1167 static int
1168 compare_pid(const void *p1, const void *p2)
1169 {
1170 	const struct kinfo_proc * const *pp1 = p1;
1171 	const struct kinfo_proc * const *pp2 = p2;
1172 
1173 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1174 
1175 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1176 }
1177 
1178 static int
1179 compare_tid(const void *p1, const void *p2)
1180 {
1181 	const struct kinfo_proc * const *pp1 = p1;
1182 	const struct kinfo_proc * const *pp2 = p2;
1183 
1184 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1185 
1186 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1187 }
1188 
1189 /*
1190  *  proc_compare - comparison function for "qsort"
1191  *	Compares the resource consumption of two processes using five
1192  *	distinct keys.  The keys (in descending order of importance) are:
1193  *	percent cpu, cpu ticks, state, resident set size, total virtual
1194  *	memory usage.  The process states are ordered as follows (from least
1195  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1196  *	array declaration below maps a process state index into a number
1197  *	that reflects this ordering.
1198  */
1199 
1200 static int sorted_state[] = {
1201 	0,	/* not used		*/
1202 	3,	/* sleep		*/
1203 	1,	/* ABANDONED (WAIT)	*/
1204 	6,	/* run			*/
1205 	5,	/* start		*/
1206 	2,	/* zombie		*/
1207 	4	/* stop			*/
1208 };
1209 
1210 
1211 #define ORDERKEY_PCTCPU(a, b) do { \
1212 	double diff; \
1213 	if (ps.wcpu) \
1214 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1215 		    weighted_cpu(PCTCPU((a)), (a)); \
1216 	else \
1217 		diff = PCTCPU((b)) - PCTCPU((a)); \
1218 	if (diff != 0) \
1219 		return (diff > 0 ? 1 : -1); \
1220 } while (0)
1221 
1222 #define ORDERKEY_CPTICKS(a, b) do { \
1223 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1224 	if (diff != 0) \
1225 		return (diff > 0 ? 1 : -1); \
1226 } while (0)
1227 
1228 #define ORDERKEY_STATE(a, b) do { \
1229 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1230 	if (diff != 0) \
1231 		return (diff > 0 ? 1 : -1); \
1232 } while (0)
1233 
1234 #define ORDERKEY_PRIO(a, b) do { \
1235 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1236 	if (diff != 0) \
1237 		return (diff > 0 ? 1 : -1); \
1238 } while (0)
1239 
1240 #define	ORDERKEY_THREADS(a, b) do { \
1241 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1242 	if (diff != 0) \
1243 		return (diff > 0 ? 1 : -1); \
1244 } while (0)
1245 
1246 #define ORDERKEY_RSSIZE(a, b) do { \
1247 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1248 	if (diff != 0) \
1249 		return (diff > 0 ? 1 : -1); \
1250 } while (0)
1251 
1252 #define ORDERKEY_MEM(a, b) do { \
1253 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1254 	if (diff != 0) \
1255 		return (diff > 0 ? 1 : -1); \
1256 } while (0)
1257 
1258 #define ORDERKEY_JID(a, b) do { \
1259 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1260 	if (diff != 0) \
1261 		return (diff > 0 ? 1 : -1); \
1262 } while (0)
1263 
1264 #define ORDERKEY_SWAP(a, b) do { \
1265 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1266 	if (diff != 0) \
1267 		return (diff > 0 ? 1 : -1); \
1268 } while (0)
1269 
1270 /* compare_cpu - the comparison function for sorting by cpu percentage */
1271 
1272 static int
1273 compare_cpu(const void *arg1, const void *arg2)
1274 {
1275 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1276 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1277 
1278 	ORDERKEY_PCTCPU(p1, p2);
1279 	ORDERKEY_CPTICKS(p1, p2);
1280 	ORDERKEY_STATE(p1, p2);
1281 	ORDERKEY_PRIO(p1, p2);
1282 	ORDERKEY_RSSIZE(p1, p2);
1283 	ORDERKEY_MEM(p1, p2);
1284 
1285 	return (0);
1286 }
1287 
1288 /* compare_size - the comparison function for sorting by total memory usage */
1289 
1290 static int
1291 compare_size(const void *arg1, const void *arg2)
1292 {
1293 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1294 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1295 
1296 	ORDERKEY_MEM(p1, p2);
1297 	ORDERKEY_RSSIZE(p1, p2);
1298 	ORDERKEY_PCTCPU(p1, p2);
1299 	ORDERKEY_CPTICKS(p1, p2);
1300 	ORDERKEY_STATE(p1, p2);
1301 	ORDERKEY_PRIO(p1, p2);
1302 
1303 	return (0);
1304 }
1305 
1306 /* compare_res - the comparison function for sorting by resident set size */
1307 
1308 static int
1309 compare_res(const void *arg1, const void *arg2)
1310 {
1311 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1312 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1313 
1314 	ORDERKEY_RSSIZE(p1, p2);
1315 	ORDERKEY_MEM(p1, p2);
1316 	ORDERKEY_PCTCPU(p1, p2);
1317 	ORDERKEY_CPTICKS(p1, p2);
1318 	ORDERKEY_STATE(p1, p2);
1319 	ORDERKEY_PRIO(p1, p2);
1320 
1321 	return (0);
1322 }
1323 
1324 /* compare_time - the comparison function for sorting by total cpu time */
1325 
1326 static int
1327 compare_time(const void *arg1, const void *arg2)
1328 {
1329 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1330 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1331 
1332 	ORDERKEY_CPTICKS(p1, p2);
1333 	ORDERKEY_PCTCPU(p1, p2);
1334 	ORDERKEY_STATE(p1, p2);
1335 	ORDERKEY_PRIO(p1, p2);
1336 	ORDERKEY_RSSIZE(p1, p2);
1337 	ORDERKEY_MEM(p1, p2);
1338 
1339 	return (0);
1340 }
1341 
1342 /* compare_prio - the comparison function for sorting by priority */
1343 
1344 static int
1345 compare_prio(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_PRIO(p1, p2);
1351 	ORDERKEY_CPTICKS(p1, p2);
1352 	ORDERKEY_PCTCPU(p1, p2);
1353 	ORDERKEY_STATE(p1, p2);
1354 	ORDERKEY_RSSIZE(p1, p2);
1355 	ORDERKEY_MEM(p1, p2);
1356 
1357 	return (0);
1358 }
1359 
1360 /* compare_threads - the comparison function for sorting by threads */
1361 static int
1362 compare_threads(const void *arg1, const void *arg2)
1363 {
1364 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1365 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1366 
1367 	ORDERKEY_THREADS(p1, p2);
1368 	ORDERKEY_PCTCPU(p1, p2);
1369 	ORDERKEY_CPTICKS(p1, p2);
1370 	ORDERKEY_STATE(p1, p2);
1371 	ORDERKEY_PRIO(p1, p2);
1372 	ORDERKEY_RSSIZE(p1, p2);
1373 	ORDERKEY_MEM(p1, p2);
1374 
1375 	return (0);
1376 }
1377 
1378 /* compare_jid - the comparison function for sorting by jid */
1379 static int
1380 compare_jid(const void *arg1, const void *arg2)
1381 {
1382 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1383 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1384 
1385 	ORDERKEY_JID(p1, p2);
1386 	ORDERKEY_PCTCPU(p1, p2);
1387 	ORDERKEY_CPTICKS(p1, p2);
1388 	ORDERKEY_STATE(p1, p2);
1389 	ORDERKEY_PRIO(p1, p2);
1390 	ORDERKEY_RSSIZE(p1, p2);
1391 	ORDERKEY_MEM(p1, p2);
1392 
1393 	return (0);
1394 }
1395 
1396 /* compare_swap - the comparison function for sorting by swap */
1397 static int
1398 compare_swap(const void *arg1, const void *arg2)
1399 {
1400 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1401 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1402 
1403 	ORDERKEY_SWAP(p1, p2);
1404 	ORDERKEY_PCTCPU(p1, p2);
1405 	ORDERKEY_CPTICKS(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 /* assorted comparison functions for sorting by i/o */
1415 
1416 static int
1417 compare_iototal(const void *arg1, const void *arg2)
1418 {
1419 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1420 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1421 
1422 	return (get_io_total(p2) - get_io_total(p1));
1423 }
1424 
1425 static int
1426 compare_ioread(const void *arg1, const void *arg2)
1427 {
1428 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1429 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1430 	long dummy, inp1, inp2;
1431 
1432 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1433 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1434 
1435 	return (inp2 - inp1);
1436 }
1437 
1438 static int
1439 compare_iowrite(const void *arg1, const void *arg2)
1440 {
1441 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1442 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1443 	long dummy, oup1, oup2;
1444 
1445 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1446 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1447 
1448 	return (oup2 - oup1);
1449 }
1450 
1451 static int
1452 compare_iofault(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 	long dummy, flp1, flp2;
1457 
1458 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1459 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1460 
1461 	return (flp2 - flp1);
1462 }
1463 
1464 static int
1465 compare_vcsw(const void *arg1, const void *arg2)
1466 {
1467 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1468 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1469 	long dummy, flp1, flp2;
1470 
1471 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1472 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1473 
1474 	return (flp2 - flp1);
1475 }
1476 
1477 static int
1478 compare_ivcsw(const void *arg1, const void *arg2)
1479 {
1480 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1481 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1482 	long dummy, flp1, flp2;
1483 
1484 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1485 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1486 
1487 	return (flp2 - flp1);
1488 }
1489 
1490 int (*compares[])(const void *arg1, const void *arg2) = {
1491 	compare_cpu,
1492 	compare_size,
1493 	compare_res,
1494 	compare_time,
1495 	compare_prio,
1496 	compare_threads,
1497 	compare_iototal,
1498 	compare_ioread,
1499 	compare_iowrite,
1500 	compare_iofault,
1501 	compare_vcsw,
1502 	compare_ivcsw,
1503 	compare_jid,
1504 	compare_swap,
1505 	NULL
1506 };
1507 
1508 
1509 static int
1510 swapmode(int *retavail, int *retfree)
1511 {
1512 	int n;
1513 	struct kvm_swap swapary[1];
1514 	static int pagesize = 0;
1515 	static unsigned long swap_maxpages = 0;
1516 
1517 	*retavail = 0;
1518 	*retfree = 0;
1519 
1520 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1521 
1522 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1523 	if (n < 0 || swapary[0].ksw_total == 0)
1524 		return (0);
1525 
1526 	if (pagesize == 0)
1527 		pagesize = getpagesize();
1528 	if (swap_maxpages == 0)
1529 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1530 
1531 	/* ksw_total contains the total size of swap all devices which may
1532 	   exceed the maximum swap size allocatable in the system */
1533 	if ( swapary[0].ksw_total > swap_maxpages )
1534 		swapary[0].ksw_total = swap_maxpages;
1535 
1536 	*retavail = CONVERT(swapary[0].ksw_total);
1537 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1538 
1539 #undef CONVERT
1540 
1541 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1542 	return (n);
1543 }
1544