xref: /freebsd/usr.bin/top/machine.c (revision e52d92164754cbfff84767d4c6eb3cc93e8c21ae)
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  *
15  * $FreeBSD$
16  */
17 
18 #include <sys/errno.h>
19 #include <sys/file.h>
20 #include <sys/param.h>
21 #include <sys/proc.h>
22 #include <sys/resource.h>
23 #include <sys/rtprio.h>
24 #include <sys/signal.h>
25 #include <sys/sysctl.h>
26 #include <sys/time.h>
27 #include <sys/user.h>
28 #include <sys/vmmeter.h>
29 
30 #include <assert.h>
31 #include <err.h>
32 #include <kvm.h>
33 #include <math.h>
34 #include <nlist.h>
35 #include <paths.h>
36 #include <pwd.h>
37 #include <stdio.h>
38 #include <stdbool.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <strings.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 #define	SMPUNAMELEN	13
55 #define	UPUNAMELEN	15
56 
57 extern struct timeval timeout;
58 static int smpmode;
59 enum displaymodes displaymode;
60 static int namelength = 8;
61 /* TOP_JID_LEN based on max of 999999 */
62 #define TOP_JID_LEN 7
63 #define TOP_SWAP_LEN 6
64 static int jidlength;
65 static int swaplength;
66 static int cmdlengthdelta;
67 
68 /* get_process_info passes back a handle.  This is what it looks like: */
69 
70 struct handle {
71 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
72 	int remaining;			/* number of pointers remaining */
73 };
74 
75 
76 /* define what weighted cpu is.  */
77 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
78 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
79 
80 /* what we consider to be process size: */
81 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
82 
83 #define RU(pp)	(&(pp)->ki_rusage)
84 #define RUTOT(pp) \
85 	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
86 
87 #define	PCTCPU(pp) (pcpu[pp - pbase])
88 
89 /* definitions for indices in the nlist array */
90 
91 /*
92  *  These definitions control the format of the per-process area
93  */
94 
95 static const char io_header[] =
96     "  PID%*s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
97 
98 static const char io_Proc_format[] =
99     "%5d%*s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s";
100 
101 static const char smp_header_thr_and_pid[] =
102     "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
103 static const char smp_header_tid_only[] =
104     "  THR%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
105 
106 static const char smp_Proc_format[] =
107     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s %2d%7s %6.2f%% %.*s";
108 
109 static char up_header_thr_and_pid[] =
110     "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
111 static char up_header_tid_only[] =
112     "  THR%*s %-*.*s "   "PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
113 
114 static char up_Proc_format[] =
115     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s%.0d%7s %6.2f%% %.*s";
116 
117 
118 /* process state names for the "STATE" column of the display */
119 /* the extra nulls in the string "run" are for adding a slash and
120    the processor number when needed */
121 
122 static const char *state_abbrev[] = {
123 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
124 };
125 
126 
127 static kvm_t *kd;
128 
129 /* values that we stash away in _init and use in later routines */
130 
131 static double logcpu;
132 
133 /* these are retrieved from the kernel in _init */
134 
135 static load_avg  ccpu;
136 
137 /* these are used in the get_ functions */
138 
139 static int lastpid;
140 
141 /* these are for calculating cpu state percentages */
142 
143 static long cp_time[CPUSTATES];
144 static long cp_old[CPUSTATES];
145 static long cp_diff[CPUSTATES];
146 
147 /* these are for detailing the process states */
148 
149 static const char *procstatenames[] = {
150 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
151 	" zombie, ", " waiting, ", " lock, ",
152 	NULL
153 };
154 static int process_states[nitems(procstatenames)];
155 
156 /* these are for detailing the cpu states */
157 
158 static int cpu_states[CPUSTATES];
159 static const char *cpustatenames[] = {
160 	"user", "nice", "system", "interrupt", "idle", NULL
161 };
162 
163 /* these are for detailing the memory statistics */
164 
165 static const char *memorynames[] = {
166 	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
167 	"K Free", NULL
168 };
169 static int memory_stats[nitems(memorynames)];
170 
171 static const char *arcnames[] = {
172 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
173 	NULL
174 };
175 static int arc_stats[nitems(arcnames)];
176 
177 static const char *carcnames[] = {
178 	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
179 	NULL
180 };
181 static int carc_stats[nitems(carcnames)];
182 
183 static const char *swapnames[] = {
184 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
185 	NULL
186 };
187 static int swap_stats[nitems(swapnames)];
188 
189 
190 /* these are for keeping track of the proc array */
191 
192 static int nproc;
193 static int onproc = -1;
194 static int pref_len;
195 static struct kinfo_proc *pbase;
196 static struct kinfo_proc **pref;
197 static struct kinfo_proc *previous_procs;
198 static struct kinfo_proc **previous_pref;
199 static int previous_proc_count = 0;
200 static int previous_proc_count_max = 0;
201 static int previous_thread;
202 
203 /* data used for recalculating pctcpu */
204 static double *pcpu;
205 static struct timespec proc_uptime;
206 static struct timeval proc_wall_time;
207 static struct timeval previous_wall_time;
208 static uint64_t previous_interval = 0;
209 
210 /* total number of io operations */
211 static long total_inblock;
212 static long total_oublock;
213 static long total_majflt;
214 
215 /* these are for getting the memory statistics */
216 
217 static int arc_enabled;
218 static int carc_enabled;
219 static int pageshift;		/* log base 2 of the pagesize */
220 
221 /* define pagetok in terms of pageshift */
222 
223 #define pagetok(size) ((size) << pageshift)
224 
225 /* swap usage */
226 #define ki_swap(kip) \
227     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
228 
229 /*
230  * Sorting orders.  The first element is the default.
231  */
232 static const char *ordernames[] = {
233 	"cpu", "size", "res", "time", "pri", "threads",
234 	"total", "read", "write", "fault", "vcsw", "ivcsw",
235 	"jid", "swap", "pid", NULL
236 };
237 
238 /* Per-cpu time states */
239 static int maxcpu;
240 static int maxid;
241 static int ncpus;
242 static u_long cpumask;
243 static long *times;
244 static long *pcpu_cp_time;
245 static long *pcpu_cp_old;
246 static long *pcpu_cp_diff;
247 static int *pcpu_cpu_states;
248 
249 static int compare_swap(const void *a, const void *b);
250 static int compare_jid(const void *a, const void *b);
251 static int compare_pid(const void *a, const void *b);
252 static int compare_tid(const void *a, const void *b);
253 static const char *format_nice(const struct kinfo_proc *pp);
254 static void getsysctl(const char *name, void *ptr, size_t len);
255 static int swapmode(int *retavail, int *retfree);
256 static void update_layout(void);
257 static int find_uid(uid_t needle, int *haystack);
258 
259 static int
260 find_uid(uid_t needle, int *haystack)
261 {
262 	size_t i = 0;
263 
264 	for (; i < TOP_MAX_UIDS; ++i)
265 		if ((uid_t)haystack[i] == needle)
266 			return 1;
267 	return 0;
268 }
269 
270 void
271 toggle_pcpustats(void)
272 {
273 
274 	if (ncpus == 1)
275 		return;
276 	update_layout();
277 }
278 
279 /* Adjust display based on ncpus and the ARC state. */
280 static void
281 update_layout(void)
282 {
283 
284 	y_mem = 3;
285 	y_arc = 4;
286 	y_carc = 5;
287 	y_swap = 4 + arc_enabled + carc_enabled;
288 	y_idlecursor = 5 + arc_enabled + carc_enabled;
289 	y_message = 5 + arc_enabled + carc_enabled;
290 	y_header = 6 + arc_enabled + carc_enabled;
291 	y_procs = 7 + arc_enabled + carc_enabled;
292 	Header_lines = 7 + arc_enabled + carc_enabled;
293 
294 	if (pcpu_stats) {
295 		y_mem += ncpus - 1;
296 		y_arc += ncpus - 1;
297 		y_carc += ncpus - 1;
298 		y_swap += ncpus - 1;
299 		y_idlecursor += ncpus - 1;
300 		y_message += ncpus - 1;
301 		y_header += ncpus - 1;
302 		y_procs += ncpus - 1;
303 		Header_lines += ncpus - 1;
304 	}
305 }
306 
307 int
308 machine_init(struct statics *statics)
309 {
310 	int i, j, empty, pagesize;
311 	uint64_t arc_size;
312 	bool carc_en;
313 	size_t size;
314 
315 	size = sizeof(smpmode);
316 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
317 	    NULL, 0) != 0 &&
318 	    sysctlbyname("kern.smp.active", &smpmode, &size,
319 	    NULL, 0) != 0) ||
320 	    size != sizeof(smpmode))
321 		smpmode = 0;
322 
323 	size = sizeof(arc_size);
324 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
325 	    NULL, 0) == 0 && arc_size != 0)
326 		arc_enabled = 1;
327 	size = sizeof(carc_en);
328 	if (arc_enabled &&
329 	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
330 	    NULL, 0) == 0 && carc_en == 1)
331 		carc_enabled = 1;
332 
333 	namelength = MAXLOGNAME;
334 	if (smpmode && namelength > SMPUNAMELEN)
335 		namelength = SMPUNAMELEN;
336 	else if (namelength > UPUNAMELEN)
337 		namelength = UPUNAMELEN;
338 
339 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
340 	if (kd == NULL)
341 		return (-1);
342 
343 	GETSYSCTL("kern.ccpu", ccpu);
344 
345 	/* this is used in calculating WCPU -- calculate it ahead of time */
346 	logcpu = log(loaddouble(ccpu));
347 
348 	pbase = NULL;
349 	pref = NULL;
350 	pcpu = NULL;
351 	nproc = 0;
352 	onproc = -1;
353 
354 	/* get the page size and calculate pageshift from it */
355 	pagesize = getpagesize();
356 	pageshift = 0;
357 	while (pagesize > 1) {
358 		pageshift++;
359 		pagesize >>= 1;
360 	}
361 
362 	/* we only need the amount of log(2)1024 for our conversion */
363 	pageshift -= LOG1024;
364 
365 	/* fill in the statics information */
366 	statics->procstate_names = procstatenames;
367 	statics->cpustate_names = cpustatenames;
368 	statics->memory_names = memorynames;
369 	if (arc_enabled)
370 		statics->arc_names = arcnames;
371 	else
372 		statics->arc_names = NULL;
373 	if (carc_enabled)
374 		statics->carc_names = carcnames;
375 	else
376 		statics->carc_names = NULL;
377 	statics->swap_names = swapnames;
378 	statics->order_names = ordernames;
379 
380 	/* Allocate state for per-CPU stats. */
381 	cpumask = 0;
382 	ncpus = 0;
383 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
384 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
385 	if (times == NULL)
386 		err(1, "calloc %zu bytes", size);
387 	size = sizeof(long) * maxcpu * CPUSTATES;
388 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
389 		err(1, "sysctlbyname kern.cp_times");
390 	pcpu_cp_time = calloc(1, size);
391 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
392 	for (i = 0; i <= maxid; i++) {
393 		empty = 1;
394 		for (j = 0; empty && j < CPUSTATES; j++) {
395 			if (times[i * CPUSTATES + j] != 0)
396 				empty = 0;
397 		}
398 		if (!empty) {
399 			cpumask |= (1ul << i);
400 			ncpus++;
401 		}
402 	}
403 	assert(ncpus > 0);
404 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
405 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
406 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
407 	statics->ncpus = ncpus;
408 
409 	update_layout();
410 
411 	/* all done! */
412 	return (0);
413 }
414 
415 const char *
416 format_header(const char *uname_field)
417 {
418 	static char Header[128];
419 	const char *prehead;
420 
421 	if (ps.jail)
422 		jidlength = TOP_JID_LEN + 1;	/* +1 for extra left space. */
423 	else
424 		jidlength = 0;
425 
426 	if (ps.swap)
427 		swaplength = TOP_SWAP_LEN + 1;  /* +1 for extra left space */
428 	else
429 		swaplength = 0;
430 
431 	switch (displaymode) {
432 	case DISP_CPU:
433 		/*
434 		 * The logic of picking the right header format seems reverse
435 		 * here because we only want to display a THR column when
436 		 * "thread mode" is off (and threads are not listed as
437 		 * separate lines).
438 		 */
439 		prehead = smpmode ?
440 		    (ps.thread ? smp_header_tid_only : smp_header_thr_and_pid) :
441 		    (ps.thread ? up_header_tid_only : up_header_thr_and_pid);
442 		snprintf(Header, sizeof(Header), prehead,
443 		    jidlength, ps.jail ? " JID" : "",
444 		    namelength, namelength, uname_field,
445 		    swaplength, ps.swap ? " SWAP" : "",
446 		    ps.wcpu ? "WCPU" : "CPU");
447 		break;
448 	case DISP_IO:
449 		prehead = io_header;
450 		snprintf(Header, sizeof(Header), prehead,
451 		    jidlength, ps.jail ? " JID" : "",
452 		    namelength, namelength, uname_field);
453 		break;
454 	case DISP_MAX:
455 		assert("displaymode must not be set to DISP_MAX");
456 	}
457 	cmdlengthdelta = strlen(Header) - 7;
458 	return (Header);
459 }
460 
461 static int swappgsin = -1;
462 static int swappgsout = -1;
463 
464 
465 void
466 get_system_info(struct system_info *si)
467 {
468 	struct loadavg sysload;
469 	int mib[2];
470 	struct timeval boottime;
471 	uint64_t arc_stat, arc_stat2;
472 	int i, j;
473 	size_t size;
474 
475 	/* get the CPU stats */
476 	size = (maxid + 1) * CPUSTATES * sizeof(long);
477 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
478 		err(1, "sysctlbyname kern.cp_times");
479 	GETSYSCTL("kern.cp_time", cp_time);
480 	GETSYSCTL("vm.loadavg", sysload);
481 	GETSYSCTL("kern.lastpid", lastpid);
482 
483 	/* convert load averages to doubles */
484 	for (i = 0; i < 3; i++)
485 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
486 
487 	/* convert cp_time counts to percentages */
488 	for (i = j = 0; i <= maxid; i++) {
489 		if ((cpumask & (1ul << i)) == 0)
490 			continue;
491 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
492 		    &pcpu_cp_time[j * CPUSTATES],
493 		    &pcpu_cp_old[j * CPUSTATES],
494 		    &pcpu_cp_diff[j * CPUSTATES]);
495 		j++;
496 	}
497 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
498 
499 	/* sum memory & swap statistics */
500 	{
501 		static unsigned int swap_delay = 0;
502 		static int swapavail = 0;
503 		static int swapfree = 0;
504 		static long bufspace = 0;
505 		static uint64_t nspgsin, nspgsout;
506 
507 		GETSYSCTL("vfs.bufspace", bufspace);
508 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
509 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
510 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
511 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
512 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
513 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
514 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
515 		/* convert memory stats to Kbytes */
516 		memory_stats[0] = pagetok(memory_stats[0]);
517 		memory_stats[1] = pagetok(memory_stats[1]);
518 		memory_stats[2] = pagetok(memory_stats[2]);
519 		memory_stats[3] = pagetok(memory_stats[3]);
520 		memory_stats[4] = bufspace / 1024;
521 		memory_stats[5] = pagetok(memory_stats[5]);
522 		memory_stats[6] = -1;
523 
524 		/* first interval */
525 		if (swappgsin < 0) {
526 			swap_stats[4] = 0;
527 			swap_stats[5] = 0;
528 		}
529 
530 		/* compute differences between old and new swap statistic */
531 		else {
532 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
533 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
534 		}
535 
536 		swappgsin = nspgsin;
537 		swappgsout = nspgsout;
538 
539 		/* call CPU heavy swapmode() only for changes */
540 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
541 			swap_stats[3] = swapmode(&swapavail, &swapfree);
542 			swap_stats[0] = swapavail;
543 			swap_stats[1] = swapavail - swapfree;
544 			swap_stats[2] = swapfree;
545 		}
546 		swap_delay = 1;
547 		swap_stats[6] = -1;
548 	}
549 
550 	if (arc_enabled) {
551 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
552 		arc_stats[0] = arc_stat >> 10;
553 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
554 		arc_stats[1] = arc_stat >> 10;
555 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
556 		arc_stats[2] = arc_stat >> 10;
557 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
558 		arc_stats[3] = arc_stat >> 10;
559 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
560 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
561 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
562 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
563 		arc_stats[5] = arc_stat >> 10;
564 		si->arc = arc_stats;
565 	}
566 	if (carc_enabled) {
567 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
568 		carc_stats[0] = arc_stat >> 10;
569 		carc_stats[2] = arc_stat >> 10; /* For ratio */
570 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
571 		carc_stats[1] = arc_stat >> 10;
572 		si->carc = carc_stats;
573 	}
574 
575 	/* set arrays and strings */
576 	if (pcpu_stats) {
577 		si->cpustates = pcpu_cpu_states;
578 		si->ncpus = ncpus;
579 	} else {
580 		si->cpustates = cpu_states;
581 		si->ncpus = 1;
582 	}
583 	si->memory = memory_stats;
584 	si->swap = swap_stats;
585 
586 
587 	if (lastpid > 0) {
588 		si->last_pid = lastpid;
589 	} else {
590 		si->last_pid = -1;
591 	}
592 
593 	/*
594 	 * Print how long system has been up.
595 	 * (Found by looking getting "boottime" from the kernel)
596 	 */
597 	mib[0] = CTL_KERN;
598 	mib[1] = KERN_BOOTTIME;
599 	size = sizeof(boottime);
600 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
601 	    boottime.tv_sec != 0) {
602 		si->boottime = boottime;
603 	} else {
604 		si->boottime.tv_sec = -1;
605 	}
606 }
607 
608 #define NOPROC	((void *)-1)
609 
610 /*
611  * We need to compare data from the old process entry with the new
612  * process entry.
613  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
614  * structure to cache the mapping.  We also use a negative cache pointer
615  * of NOPROC to avoid duplicate lookups.
616  * XXX: this could be done when the actual processes are fetched, we do
617  * it here out of laziness.
618  */
619 static const struct kinfo_proc *
620 get_old_proc(struct kinfo_proc *pp)
621 {
622 	struct kinfo_proc **oldpp, *oldp;
623 
624 	/*
625 	 * If this is the first fetch of the kinfo_procs then we don't have
626 	 * any previous entries.
627 	 */
628 	if (previous_proc_count == 0)
629 		return (NULL);
630 	/* negative cache? */
631 	if (pp->ki_udata == NOPROC)
632 		return (NULL);
633 	/* cached? */
634 	if (pp->ki_udata != NULL)
635 		return (pp->ki_udata);
636 	/*
637 	 * Not cached,
638 	 * 1) look up based on pid.
639 	 * 2) compare process start.
640 	 * If we fail here, then setup a negative cache entry, otherwise
641 	 * cache it.
642 	 */
643 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
644 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
645 	if (oldpp == NULL) {
646 		pp->ki_udata = NOPROC;
647 		return (NULL);
648 	}
649 	oldp = *oldpp;
650 	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
651 		pp->ki_udata = NOPROC;
652 		return (NULL);
653 	}
654 	pp->ki_udata = oldp;
655 	return (oldp);
656 }
657 
658 /*
659  * Return the total amount of IO done in blocks in/out and faults.
660  * store the values individually in the pointers passed in.
661  */
662 static long
663 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
664     long *vcsw, long *ivcsw)
665 {
666 	const struct kinfo_proc *oldp;
667 	static struct kinfo_proc dummy;
668 	long ret;
669 
670 	oldp = get_old_proc(pp);
671 	if (oldp == NULL) {
672 		bzero(&dummy, sizeof(dummy));
673 		oldp = &dummy;
674 	}
675 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
676 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
677 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
678 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
679 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
680 	ret =
681 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
682 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
683 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
684 	return (ret);
685 }
686 
687 /*
688  * If there was a previous update, use the delta in ki_runtime over
689  * the previous interval to calculate pctcpu.  Otherwise, fall back
690  * to using the kernel's ki_pctcpu.
691  */
692 static double
693 proc_calc_pctcpu(struct kinfo_proc *pp)
694 {
695 	const struct kinfo_proc *oldp;
696 
697 	if (previous_interval != 0) {
698 		oldp = get_old_proc(pp);
699 		if (oldp != NULL)
700 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
701 			    / previous_interval);
702 
703 		/*
704 		 * If this process/thread was created during the previous
705 		 * interval, charge it's total runtime to the previous
706 		 * interval.
707 		 */
708 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
709 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
710 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
711 			return ((double)pp->ki_runtime / previous_interval);
712 	}
713 	return (pctdouble(pp->ki_pctcpu));
714 }
715 
716 /*
717  * Return true if this process has used any CPU time since the
718  * previous update.
719  */
720 static int
721 proc_used_cpu(struct kinfo_proc *pp)
722 {
723 	const struct kinfo_proc *oldp;
724 
725 	oldp = get_old_proc(pp);
726 	if (oldp == NULL)
727 		return (PCTCPU(pp) != 0);
728 	return (pp->ki_runtime != oldp->ki_runtime ||
729 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
730 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
731 }
732 
733 /*
734  * Return the total number of block in/out and faults by a process.
735  */
736 static long
737 get_io_total(const struct kinfo_proc *pp)
738 {
739 	long dummy;
740 
741 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
742 }
743 
744 static struct handle handle;
745 
746 void *
747 get_process_info(struct system_info *si, struct process_select *sel,
748     int (*compare)(const void *, const void *))
749 {
750 	int i;
751 	int total_procs;
752 	long p_io;
753 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
754 	long nsec;
755 	int active_procs;
756 	struct kinfo_proc **prefp;
757 	struct kinfo_proc *pp;
758 	struct timespec previous_proc_uptime;
759 
760 	/* these are copied out of sel for speed */
761 	int show_idle;
762 	int show_jid;
763 	int show_self;
764 	int show_system;
765 	int show_uid;
766 	int show_pid;
767 	int show_kidle;
768 
769 	/*
770 	 * If thread state was toggled, don't cache the previous processes.
771 	 */
772 	if (previous_thread != sel->thread)
773 		nproc = 0;
774 	previous_thread = sel->thread;
775 
776 	/*
777 	 * Save the previous process info.
778 	 */
779 	if (previous_proc_count_max < nproc) {
780 		free(previous_procs);
781 		previous_procs = calloc(nproc, sizeof(*previous_procs));
782 		free(previous_pref);
783 		previous_pref = calloc(nproc, sizeof(*previous_pref));
784 		if (previous_procs == NULL || previous_pref == NULL) {
785 			fprintf(stderr, "top: Out of memory.\n");
786 			quit(TOP_EX_SYS_ERROR);
787 		}
788 		previous_proc_count_max = nproc;
789 	}
790 	if (nproc) {
791 		for (i = 0; i < nproc; i++)
792 			previous_pref[i] = &previous_procs[i];
793 		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
794 		qsort(previous_pref, nproc, sizeof(*previous_pref),
795 		    ps.thread ? compare_tid : compare_pid);
796 	}
797 	previous_proc_count = nproc;
798 	previous_proc_uptime = proc_uptime;
799 	previous_wall_time = proc_wall_time;
800 	previous_interval = 0;
801 
802 	pbase = kvm_getprocs(kd, sel->thread ? KERN_PROC_ALL : KERN_PROC_PROC,
803 	    0, &nproc);
804 	(void)gettimeofday(&proc_wall_time, NULL);
805 	if (clock_gettime(CLOCK_UPTIME, &proc_uptime) != 0)
806 		memset(&proc_uptime, 0, sizeof(proc_uptime));
807 	else if (previous_proc_uptime.tv_sec != 0 &&
808 	    previous_proc_uptime.tv_nsec != 0) {
809 		previous_interval = (proc_uptime.tv_sec -
810 		    previous_proc_uptime.tv_sec) * 1000000;
811 		nsec = proc_uptime.tv_nsec - previous_proc_uptime.tv_nsec;
812 		if (nsec < 0) {
813 			previous_interval -= 1000000;
814 			nsec += 1000000000;
815 		}
816 		previous_interval += nsec / 1000;
817 	}
818 	if (nproc > onproc) {
819 		pref = realloc(pref, sizeof(*pref) * nproc);
820 		pcpu = realloc(pcpu, sizeof(*pcpu) * nproc);
821 		onproc = nproc;
822 	}
823 	if (pref == NULL || pbase == NULL || pcpu == NULL) {
824 		(void) fprintf(stderr, "top: Out of memory.\n");
825 		quit(TOP_EX_SYS_ERROR);
826 	}
827 	/* get a pointer to the states summary array */
828 	si->procstates = process_states;
829 
830 	/* set up flags which define what we are going to select */
831 	show_idle = sel->idle;
832 	show_jid = sel->jid != -1;
833 	show_self = sel->self == -1;
834 	show_system = sel->system;
835 	show_uid = sel->uid[0] != -1;
836 	show_pid = sel->pid != -1;
837 	show_kidle = sel->kidle;
838 
839 	/* count up process states and get pointers to interesting procs */
840 	total_procs = 0;
841 	active_procs = 0;
842 	total_inblock = 0;
843 	total_oublock = 0;
844 	total_majflt = 0;
845 	memset(process_states, 0, sizeof(process_states));
846 	prefp = pref;
847 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
848 
849 		if (pp->ki_stat == 0)
850 			/* not in use */
851 			continue;
852 
853 		if (!show_self && pp->ki_pid == sel->self)
854 			/* skip self */
855 			continue;
856 
857 		if (!show_system && (pp->ki_flag & P_SYSTEM))
858 			/* skip system process */
859 			continue;
860 
861 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
862 		    &p_vcsw, &p_ivcsw);
863 		total_inblock += p_inblock;
864 		total_oublock += p_oublock;
865 		total_majflt += p_majflt;
866 		total_procs++;
867 		process_states[(unsigned char)pp->ki_stat]++;
868 
869 		if (pp->ki_stat == SZOMB)
870 			/* skip zombies */
871 			continue;
872 
873 		if (!show_kidle && pp->ki_tdflags & TDF_IDLETD)
874 			/* skip kernel idle process */
875 			continue;
876 
877 		PCTCPU(pp) = proc_calc_pctcpu(pp);
878 		if (sel->thread && PCTCPU(pp) > 1.0)
879 			PCTCPU(pp) = 1.0;
880 		if (displaymode == DISP_CPU && !show_idle &&
881 		    (!proc_used_cpu(pp) ||
882 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
883 			/* skip idle or non-running processes */
884 			continue;
885 
886 		if (displaymode == DISP_IO && !show_idle && p_io == 0)
887 			/* skip processes that aren't doing I/O */
888 			continue;
889 
890 		if (show_jid && pp->ki_jid != sel->jid)
891 			/* skip proc. that don't belong to the selected JID */
892 			continue;
893 
894 		if (show_uid && !find_uid(pp->ki_ruid, sel->uid))
895 			/* skip proc. that don't belong to the selected UID */
896 			continue;
897 
898 		if (show_pid && pp->ki_pid != sel->pid)
899 			continue;
900 
901 		*prefp++ = pp;
902 		active_procs++;
903 	}
904 
905 	/* if requested, sort the "interesting" processes */
906 	if (compare != NULL)
907 		qsort(pref, active_procs, sizeof(*pref), compare);
908 
909 	/* remember active and total counts */
910 	si->p_total = total_procs;
911 	si->p_pactive = pref_len = active_procs;
912 
913 	/* pass back a handle */
914 	handle.next_proc = pref;
915 	handle.remaining = active_procs;
916 	return ((void*)&handle);
917 }
918 
919 static char fmt[512];	/* static area where result is built */
920 
921 char *
922 format_next_process(void* xhandle, char *(*get_userid)(int), int flags)
923 {
924 	struct kinfo_proc *pp;
925 	const struct kinfo_proc *oldp;
926 	long cputime;
927 	double pct;
928 	struct handle *hp;
929 	char status[22];
930 	int cpu;
931 	size_t state;
932 	struct rusage ru, *rup;
933 	long p_tot, s_tot;
934 	const char *proc_fmt;
935 	char thr_buf[6];
936 	char jid_buf[TOP_JID_LEN + 1], swap_buf[TOP_SWAP_LEN + 1];
937 	char *cmdbuf = NULL;
938 	char **args;
939 	const int cmdlen = 128;
940 
941 	/* find and remember the next proc structure */
942 	hp = (struct handle *)xhandle;
943 	pp = *(hp->next_proc++);
944 	hp->remaining--;
945 
946 	/* get the process's command name */
947 	if ((pp->ki_flag & P_INMEM) == 0) {
948 		/*
949 		 * Print swapped processes as <pname>
950 		 */
951 		size_t len;
952 
953 		len = strlen(pp->ki_comm);
954 		if (len > sizeof(pp->ki_comm) - 3)
955 			len = sizeof(pp->ki_comm) - 3;
956 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
957 		pp->ki_comm[0] = '<';
958 		pp->ki_comm[len + 1] = '>';
959 		pp->ki_comm[len + 2] = '\0';
960 	}
961 
962 	/*
963 	 * Convert the process's runtime from microseconds to seconds.  This
964 	 * time includes the interrupt time although that is not wanted here.
965 	 * ps(1) is similarly sloppy.
966 	 */
967 	cputime = (pp->ki_runtime + 500000) / 1000000;
968 
969 	/* calculate the base for cpu percentages */
970 	pct = PCTCPU(pp);
971 
972 	/* generate "STATE" field */
973 	switch (state = pp->ki_stat) {
974 	case SRUN:
975 		if (smpmode && pp->ki_oncpu != NOCPU)
976 			sprintf(status, "CPU%d", pp->ki_oncpu);
977 		else
978 			strcpy(status, "RUN");
979 		break;
980 	case SLOCK:
981 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
982 			sprintf(status, "*%.6s", pp->ki_lockname);
983 			break;
984 		}
985 		/* fall through */
986 	case SSLEEP:
987 		sprintf(status, "%.6s", pp->ki_wmesg);
988 		break;
989 	default:
990 
991 		if (state < nitems(state_abbrev)) {
992 			sprintf(status, "%.6s", state_abbrev[state]);
993 		} else {
994 			sprintf(status, "?%5zu", state);
995 		}
996 		break;
997 	}
998 
999 	cmdbuf = calloc(cmdlen + 1, 1);
1000 	if (cmdbuf == NULL) {
1001 		warn("calloc(%d)", cmdlen + 1);
1002 		return NULL;
1003 	}
1004 
1005 	if (!(flags & FMT_SHOWARGS)) {
1006 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1007 		    pp->ki_tdname[0]) {
1008 			snprintf(cmdbuf, cmdlen, "%s{%s%s}", pp->ki_comm,
1009 			    pp->ki_tdname, pp->ki_moretdname);
1010 		} else {
1011 			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
1012 		}
1013 	} else {
1014 		if (pp->ki_flag & P_SYSTEM ||
1015 		    pp->ki_args == NULL ||
1016 		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
1017 		    !(*args)) {
1018 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1019 		    	    pp->ki_tdname[0]) {
1020 				snprintf(cmdbuf, cmdlen,
1021 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
1022 				    pp->ki_moretdname);
1023 			} else {
1024 				snprintf(cmdbuf, cmdlen,
1025 				    "[%s]", pp->ki_comm);
1026 			}
1027 		} else {
1028 			const char *src;
1029 			char *dst, *argbuf;
1030 			const char *cmd;
1031 			size_t argbuflen;
1032 			size_t len;
1033 
1034 			argbuflen = cmdlen * 4;
1035 			argbuf = calloc(argbuflen + 1, 1);
1036 			if (argbuf == NULL) {
1037 				warn("calloc(%zu)", argbuflen + 1);
1038 				free(cmdbuf);
1039 				return NULL;
1040 			}
1041 
1042 			dst = argbuf;
1043 
1044 			/* Extract cmd name from argv */
1045 			cmd = strrchr(*args, '/');
1046 			if (cmd == NULL)
1047 				cmd = *args;
1048 			else
1049 				cmd++;
1050 
1051 			for (; (src = *args++) != NULL; ) {
1052 				if (*src == '\0')
1053 					continue;
1054 				len = (argbuflen - (dst - argbuf) - 1) / 4;
1055 				strvisx(dst, src,
1056 				    MIN(strlen(src), len),
1057 				    VIS_NL | VIS_CSTYLE);
1058 				while (*dst != '\0')
1059 					dst++;
1060 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1061 					*dst++ = ' '; /* add delimiting space */
1062 			}
1063 			if (dst != argbuf && dst[-1] == ' ')
1064 				dst--;
1065 			*dst = '\0';
1066 
1067 			if (strcmp(cmd, pp->ki_comm) != 0) {
1068 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1069 				    pp->ki_tdname[0])
1070 					snprintf(cmdbuf, cmdlen,
1071 					    "%s (%s){%s%s}", argbuf,
1072 					    pp->ki_comm, pp->ki_tdname,
1073 					    pp->ki_moretdname);
1074 				else
1075 					snprintf(cmdbuf, cmdlen,
1076 					    "%s (%s)", argbuf, pp->ki_comm);
1077 			} else {
1078 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1079 				    pp->ki_tdname[0])
1080 					snprintf(cmdbuf, cmdlen,
1081 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1082 					    pp->ki_moretdname);
1083 				else
1084 					strlcpy(cmdbuf, argbuf, cmdlen);
1085 			}
1086 			free(argbuf);
1087 		}
1088 	}
1089 
1090 	if (ps.jail == 0)
1091 		jid_buf[0] = '\0';
1092 	else
1093 		snprintf(jid_buf, sizeof(jid_buf), "%*d",
1094 		    jidlength - 1, pp->ki_jid);
1095 
1096 	if (ps.swap == 0)
1097 		swap_buf[0] = '\0';
1098 	else
1099 		snprintf(swap_buf, sizeof(swap_buf), "%*s",
1100 		    swaplength - 1,
1101 		    format_k2(pagetok(ki_swap(pp)))); /* XXX */
1102 
1103 	if (displaymode == DISP_IO) {
1104 		oldp = get_old_proc(pp);
1105 		if (oldp != NULL) {
1106 			ru.ru_inblock = RU(pp)->ru_inblock -
1107 			    RU(oldp)->ru_inblock;
1108 			ru.ru_oublock = RU(pp)->ru_oublock -
1109 			    RU(oldp)->ru_oublock;
1110 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1111 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1112 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1113 			rup = &ru;
1114 		} else {
1115 			rup = RU(pp);
1116 		}
1117 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1118 		s_tot = total_inblock + total_oublock + total_majflt;
1119 
1120 		snprintf(fmt, sizeof(fmt), io_Proc_format,
1121 		    pp->ki_pid,
1122 		    jidlength, jid_buf,
1123 		    namelength, namelength, (*get_userid)(pp->ki_ruid),
1124 		    rup->ru_nvcsw,
1125 		    rup->ru_nivcsw,
1126 		    rup->ru_inblock,
1127 		    rup->ru_oublock,
1128 		    rup->ru_majflt,
1129 		    p_tot,
1130 		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
1131 		    screen_width > cmdlengthdelta ?
1132 		    screen_width - cmdlengthdelta : 0,
1133 		    printable(cmdbuf));
1134 
1135 		free(cmdbuf);
1136 
1137 		return (fmt);
1138 	}
1139 
1140 	/* format this entry */
1141 	if (smpmode) {
1142 		if (state == SRUN && pp->ki_oncpu != NOCPU)
1143 			cpu = pp->ki_oncpu;
1144 		else
1145 			cpu = pp->ki_lastcpu;
1146 	} else
1147 		cpu = 0;
1148 	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1149 	if (ps.thread != 0)
1150 		thr_buf[0] = '\0';
1151 	else
1152 		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1153 		    (int)(sizeof(thr_buf) - 2), pp->ki_numthreads);
1154 
1155 	snprintf(fmt, sizeof(fmt), proc_fmt,
1156 	    (ps.thread) ? pp->ki_tid : pp->ki_pid,
1157 	    jidlength, jid_buf,
1158 	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1159 	    thr_buf,
1160 	    pp->ki_pri.pri_level - PZERO,
1161 	    format_nice(pp),
1162 	    format_k2(PROCSIZE(pp)),
1163 	    format_k2(pagetok(pp->ki_rssize)),
1164 	    swaplength, swaplength, swap_buf,
1165 	    status,
1166 	    cpu,
1167 	    format_time(cputime),
1168 	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1169 	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1170 	    printable(cmdbuf));
1171 
1172 	free(cmdbuf);
1173 
1174 	/* return the result */
1175 	return (fmt);
1176 }
1177 
1178 static void
1179 getsysctl(const char *name, void *ptr, size_t len)
1180 {
1181 	size_t nlen = len;
1182 
1183 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1184 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1185 		    strerror(errno));
1186 		quit(TOP_EX_SYS_ERROR);
1187 	}
1188 	if (nlen != len) {
1189 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1190 		    name, (unsigned long)len, (unsigned long)nlen);
1191 		quit(TOP_EX_SYS_ERROR);
1192 	}
1193 }
1194 
1195 static const char *
1196 format_nice(const struct kinfo_proc *pp)
1197 {
1198 	const char *fifo, *kproc;
1199 	int rtpri;
1200 	static char nicebuf[4 + 1];
1201 
1202 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1203 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1204 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1205 	case PRI_ITHD:
1206 		return ("-");
1207 	case PRI_REALTIME:
1208 		/*
1209 		 * XXX: the kernel doesn't tell us the original rtprio and
1210 		 * doesn't really know what it was, so to recover it we
1211 		 * must be more chummy with the implementation than the
1212 		 * implementation is with itself.  pri_user gives a
1213 		 * constant "base" priority, but is only initialized
1214 		 * properly for user threads.  pri_native gives what the
1215 		 * kernel calls the "base" priority, but it isn't constant
1216 		 * since it is changed by priority propagation.  pri_native
1217 		 * also isn't properly initialized for all threads, but it
1218 		 * is properly initialized for kernel realtime and idletime
1219 		 * threads.  Thus we use pri_user for the base priority of
1220 		 * user threads (it is always correct) and pri_native for
1221 		 * the base priority of kernel realtime and idletime threads
1222 		 * (there is nothing better, and it is usually correct).
1223 		 *
1224 		 * The field width and thus the buffer are too small for
1225 		 * values like "kr31F", but such values shouldn't occur,
1226 		 * and if they do then the tailing "F" is not displayed.
1227 		 */
1228 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1229 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1230 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1231 		    kproc, rtpri, fifo);
1232 		break;
1233 	case PRI_TIMESHARE:
1234 		if (pp->ki_flag & P_KPROC)
1235 			return ("-");
1236 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1237 		break;
1238 	case PRI_IDLE:
1239 		/* XXX: as above. */
1240 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1241 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1242 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1243 		    kproc, rtpri, fifo);
1244 		break;
1245 	default:
1246 		return ("?");
1247 	}
1248 	return (nicebuf);
1249 }
1250 
1251 /* comparison routines for qsort */
1252 
1253 static int
1254 compare_pid(const void *p1, const void *p2)
1255 {
1256 	const struct kinfo_proc * const *pp1 = p1;
1257 	const struct kinfo_proc * const *pp2 = p2;
1258 
1259 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1260 
1261 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1262 }
1263 
1264 static int
1265 compare_tid(const void *p1, const void *p2)
1266 {
1267 	const struct kinfo_proc * const *pp1 = p1;
1268 	const struct kinfo_proc * const *pp2 = p2;
1269 
1270 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1271 
1272 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1273 }
1274 
1275 /*
1276  *  proc_compare - comparison function for "qsort"
1277  *	Compares the resource consumption of two processes using five
1278  *	distinct keys.  The keys (in descending order of importance) are:
1279  *	percent cpu, cpu ticks, state, resident set size, total virtual
1280  *	memory usage.  The process states are ordered as follows (from least
1281  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1282  *	array declaration below maps a process state index into a number
1283  *	that reflects this ordering.
1284  */
1285 
1286 static int sorted_state[] = {
1287 	0,	/* not used		*/
1288 	3,	/* sleep		*/
1289 	1,	/* ABANDONED (WAIT)	*/
1290 	6,	/* run			*/
1291 	5,	/* start		*/
1292 	2,	/* zombie		*/
1293 	4	/* stop			*/
1294 };
1295 
1296 
1297 #define ORDERKEY_PCTCPU(a, b) do { \
1298 	double diff; \
1299 	if (ps.wcpu) \
1300 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1301 		    weighted_cpu(PCTCPU((a)), (a)); \
1302 	else \
1303 		diff = PCTCPU((b)) - PCTCPU((a)); \
1304 	if (diff != 0) \
1305 		return (diff > 0 ? 1 : -1); \
1306 } while (0)
1307 
1308 #define ORDERKEY_CPTICKS(a, b) do { \
1309 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1310 	if (diff != 0) \
1311 		return (diff > 0 ? 1 : -1); \
1312 } while (0)
1313 
1314 #define ORDERKEY_STATE(a, b) do { \
1315 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1316 	if (diff != 0) \
1317 		return (diff > 0 ? 1 : -1); \
1318 } while (0)
1319 
1320 #define ORDERKEY_PRIO(a, b) do { \
1321 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1322 	if (diff != 0) \
1323 		return (diff > 0 ? 1 : -1); \
1324 } while (0)
1325 
1326 #define	ORDERKEY_THREADS(a, b) do { \
1327 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1328 	if (diff != 0) \
1329 		return (diff > 0 ? 1 : -1); \
1330 } while (0)
1331 
1332 #define ORDERKEY_RSSIZE(a, b) do { \
1333 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1334 	if (diff != 0) \
1335 		return (diff > 0 ? 1 : -1); \
1336 } while (0)
1337 
1338 #define ORDERKEY_MEM(a, b) do { \
1339 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1340 	if (diff != 0) \
1341 		return (diff > 0 ? 1 : -1); \
1342 } while (0)
1343 
1344 #define ORDERKEY_JID(a, b) do { \
1345 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1346 	if (diff != 0) \
1347 		return (diff > 0 ? 1 : -1); \
1348 } while (0)
1349 
1350 #define ORDERKEY_SWAP(a, b) do { \
1351 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1352 	if (diff != 0) \
1353 		return (diff > 0 ? 1 : -1); \
1354 } while (0)
1355 
1356 /* compare_cpu - the comparison function for sorting by cpu percentage */
1357 
1358 static int
1359 compare_cpu(const void *arg1, const void *arg2)
1360 {
1361 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1362 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1363 
1364 	ORDERKEY_PCTCPU(p1, p2);
1365 	ORDERKEY_CPTICKS(p1, p2);
1366 	ORDERKEY_STATE(p1, p2);
1367 	ORDERKEY_PRIO(p1, p2);
1368 	ORDERKEY_RSSIZE(p1, p2);
1369 	ORDERKEY_MEM(p1, p2);
1370 
1371 	return (0);
1372 }
1373 
1374 /* compare_size - the comparison function for sorting by total memory usage */
1375 
1376 static int
1377 compare_size(const void *arg1, const void *arg2)
1378 {
1379 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1380 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1381 
1382 	ORDERKEY_MEM(p1, p2);
1383 	ORDERKEY_RSSIZE(p1, p2);
1384 	ORDERKEY_PCTCPU(p1, p2);
1385 	ORDERKEY_CPTICKS(p1, p2);
1386 	ORDERKEY_STATE(p1, p2);
1387 	ORDERKEY_PRIO(p1, p2);
1388 
1389 	return (0);
1390 }
1391 
1392 /* compare_res - the comparison function for sorting by resident set size */
1393 
1394 static int
1395 compare_res(const void *arg1, const void *arg2)
1396 {
1397 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1398 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1399 
1400 	ORDERKEY_RSSIZE(p1, p2);
1401 	ORDERKEY_MEM(p1, p2);
1402 	ORDERKEY_PCTCPU(p1, p2);
1403 	ORDERKEY_CPTICKS(p1, p2);
1404 	ORDERKEY_STATE(p1, p2);
1405 	ORDERKEY_PRIO(p1, p2);
1406 
1407 	return (0);
1408 }
1409 
1410 /* compare_time - the comparison function for sorting by total cpu time */
1411 
1412 static int
1413 compare_time(const void *arg1, const void *arg2)
1414 {
1415 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1416 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1417 
1418 	ORDERKEY_CPTICKS(p1, p2);
1419 	ORDERKEY_PCTCPU(p1, p2);
1420 	ORDERKEY_STATE(p1, p2);
1421 	ORDERKEY_PRIO(p1, p2);
1422 	ORDERKEY_RSSIZE(p1, p2);
1423 	ORDERKEY_MEM(p1, p2);
1424 
1425 	return (0);
1426 }
1427 
1428 /* compare_prio - the comparison function for sorting by priority */
1429 
1430 static int
1431 compare_prio(const void *arg1, const void *arg2)
1432 {
1433 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1434 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1435 
1436 	ORDERKEY_PRIO(p1, p2);
1437 	ORDERKEY_CPTICKS(p1, p2);
1438 	ORDERKEY_PCTCPU(p1, p2);
1439 	ORDERKEY_STATE(p1, p2);
1440 	ORDERKEY_RSSIZE(p1, p2);
1441 	ORDERKEY_MEM(p1, p2);
1442 
1443 	return (0);
1444 }
1445 
1446 /* compare_threads - the comparison function for sorting by threads */
1447 static int
1448 compare_threads(const void *arg1, const void *arg2)
1449 {
1450 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1451 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1452 
1453 	ORDERKEY_THREADS(p1, p2);
1454 	ORDERKEY_PCTCPU(p1, p2);
1455 	ORDERKEY_CPTICKS(p1, p2);
1456 	ORDERKEY_STATE(p1, p2);
1457 	ORDERKEY_PRIO(p1, p2);
1458 	ORDERKEY_RSSIZE(p1, p2);
1459 	ORDERKEY_MEM(p1, p2);
1460 
1461 	return (0);
1462 }
1463 
1464 /* compare_jid - the comparison function for sorting by jid */
1465 static int
1466 compare_jid(const void *arg1, const void *arg2)
1467 {
1468 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1469 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1470 
1471 	ORDERKEY_JID(p1, p2);
1472 	ORDERKEY_PCTCPU(p1, p2);
1473 	ORDERKEY_CPTICKS(p1, p2);
1474 	ORDERKEY_STATE(p1, p2);
1475 	ORDERKEY_PRIO(p1, p2);
1476 	ORDERKEY_RSSIZE(p1, p2);
1477 	ORDERKEY_MEM(p1, p2);
1478 
1479 	return (0);
1480 }
1481 
1482 /* compare_swap - the comparison function for sorting by swap */
1483 static int
1484 compare_swap(const void *arg1, const void *arg2)
1485 {
1486 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1487 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1488 
1489 	ORDERKEY_SWAP(p1, p2);
1490 	ORDERKEY_PCTCPU(p1, p2);
1491 	ORDERKEY_CPTICKS(p1, p2);
1492 	ORDERKEY_STATE(p1, p2);
1493 	ORDERKEY_PRIO(p1, p2);
1494 	ORDERKEY_RSSIZE(p1, p2);
1495 	ORDERKEY_MEM(p1, p2);
1496 
1497 	return (0);
1498 }
1499 
1500 /* assorted comparison functions for sorting by i/o */
1501 
1502 static int
1503 compare_iototal(const void *arg1, const void *arg2)
1504 {
1505 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1506 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1507 
1508 	return (get_io_total(p2) - get_io_total(p1));
1509 }
1510 
1511 static int
1512 compare_ioread(const void *arg1, const void *arg2)
1513 {
1514 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1515 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1516 	long dummy, inp1, inp2;
1517 
1518 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1519 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1520 
1521 	return (inp2 - inp1);
1522 }
1523 
1524 static int
1525 compare_iowrite(const void *arg1, const void *arg2)
1526 {
1527 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1528 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1529 	long dummy, oup1, oup2;
1530 
1531 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1532 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1533 
1534 	return (oup2 - oup1);
1535 }
1536 
1537 static int
1538 compare_iofault(const void *arg1, const void *arg2)
1539 {
1540 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1541 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1542 	long dummy, flp1, flp2;
1543 
1544 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1545 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1546 
1547 	return (flp2 - flp1);
1548 }
1549 
1550 static int
1551 compare_vcsw(const void *arg1, const void *arg2)
1552 {
1553 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1554 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1555 	long dummy, flp1, flp2;
1556 
1557 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1558 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1559 
1560 	return (flp2 - flp1);
1561 }
1562 
1563 static int
1564 compare_ivcsw(const void *arg1, const void *arg2)
1565 {
1566 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1567 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1568 	long dummy, flp1, flp2;
1569 
1570 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1571 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1572 
1573 	return (flp2 - flp1);
1574 }
1575 
1576 int (*compares[])(const void *arg1, const void *arg2) = {
1577 	compare_cpu,
1578 	compare_size,
1579 	compare_res,
1580 	compare_time,
1581 	compare_prio,
1582 	compare_threads,
1583 	compare_iototal,
1584 	compare_ioread,
1585 	compare_iowrite,
1586 	compare_iofault,
1587 	compare_vcsw,
1588 	compare_ivcsw,
1589 	compare_jid,
1590 	compare_swap,
1591 	NULL
1592 };
1593 
1594 
1595 /*
1596  * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1597  *		the process does not exist.
1598  */
1599 
1600 int
1601 proc_owner(int pid)
1602 {
1603 	int cnt;
1604 	struct kinfo_proc **prefp;
1605 	struct kinfo_proc *pp;
1606 
1607 	prefp = pref;
1608 	cnt = pref_len;
1609 	while (--cnt >= 0) {
1610 		pp = *prefp++;
1611 		if (pp->ki_pid == (pid_t)pid)
1612 			return ((int)pp->ki_ruid);
1613 	}
1614 	return (-1);
1615 }
1616 
1617 static int
1618 swapmode(int *retavail, int *retfree)
1619 {
1620 	int n;
1621 	struct kvm_swap swapary[1];
1622 	static int pagesize = 0;
1623 	static u_long swap_maxpages = 0;
1624 
1625 	*retavail = 0;
1626 	*retfree = 0;
1627 
1628 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1629 
1630 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1631 	if (n < 0 || swapary[0].ksw_total == 0)
1632 		return (0);
1633 
1634 	if (pagesize == 0)
1635 		pagesize = getpagesize();
1636 	if (swap_maxpages == 0)
1637 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1638 
1639 	/* ksw_total contains the total size of swap all devices which may
1640 	   exceed the maximum swap size allocatable in the system */
1641 	if ( swapary[0].ksw_total > swap_maxpages )
1642 		swapary[0].ksw_total = swap_maxpages;
1643 
1644 	*retavail = CONVERT(swapary[0].ksw_total);
1645 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1646 
1647 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1648 	return (n);
1649 }
1650