xref: /freebsd/usr.bin/top/machine.c (revision c5fda9bac0325eb8c5b447717862d279006f318f)
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/sysctl.h>
26 #include <sys/time.h>
27 #include <sys/user.h>
28 
29 #include <assert.h>
30 #include <err.h>
31 #include <kvm.h>
32 #include <math.h>
33 #include <paths.h>
34 #include <stdio.h>
35 #include <stdbool.h>
36 #include <stdint.h>
37 #include <stdlib.h>
38 #include <string.h>
39 #include <time.h>
40 #include <unistd.h>
41 #include <vis.h>
42 
43 #include "top.h"
44 #include "display.h"
45 #include "machine.h"
46 #include "loadavg.h"
47 #include "screen.h"
48 #include "utils.h"
49 #include "layout.h"
50 
51 #define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
52 #define	SMPUNAMELEN	13
53 #define	UPUNAMELEN	15
54 
55 extern struct timeval timeout;
56 static int smpmode;
57 enum displaymodes displaymode;
58 static int namelength = 8;
59 /* TOP_JID_LEN based on max of 999999 */
60 #define TOP_JID_LEN 7
61 #define TOP_SWAP_LEN 6
62 static int jidlength;
63 static int swaplength;
64 static int cmdlengthdelta;
65 
66 /* get_process_info passes back a handle.  This is what it looks like: */
67 
68 struct handle {
69 	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
70 	int remaining;			/* number of pointers remaining */
71 };
72 
73 
74 /* define what weighted cpu is.  */
75 #define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
76 			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
77 
78 /* what we consider to be process size: */
79 #define PROCSIZE(pp) ((pp)->ki_size / 1024)
80 
81 #define RU(pp)	(&(pp)->ki_rusage)
82 #define RUTOT(pp) \
83 	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
84 
85 #define	PCTCPU(pp) (pcpu[pp - pbase])
86 
87 /*
88  *  These definitions control the format of the per-process area
89  */
90 
91 static const char io_header[] =
92     "  PID%*s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
93 
94 static const char io_Proc_format[] =
95     "%5d%*s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s";
96 
97 /* XXX: build up header instead of statically defining them.
98  * This will also allow for a "format string" to be supplied
99  * as an argument to top(1) instead of having predefined options */
100 static const char smp_header_thr_and_pid[] =
101     "  %s%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
102 static const char smp_header_id_only[] =
103     "  %s%*s %-*.*s  PRI NICE   SIZE    RES%*s STATE   C   TIME %7s COMMAND";
104 static const char smp_Proc_format[] =
105     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s %2d%7s %6.2f%% %.*s";
106 
107 static char up_header_thr_and_pid[] =
108     "  PID%*s %-*.*s  THR PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
109 static char up_header_id_only[] =
110     "  %s%*s %-*.*s   PRI NICE   SIZE    RES%*s STATE    TIME %7s COMMAND";
111 static char up_Proc_format[] =
112     "%5d%*s %-*.*s %s%3d %4s%7s %6s%*.*s %-6.6s%.0d%7s %6.2f%% %.*s";
113 
114 
115 /* process state names for the "STATE" column of the display */
116 /* the extra nulls in the string "run" are for adding a slash and
117    the processor number when needed */
118 
119 static const char *state_abbrev[] = {
120 	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
121 };
122 
123 
124 static kvm_t *kd;
125 
126 /* values that we stash away in _init and use in later routines */
127 
128 static double logcpu;
129 
130 /* these are retrieved from the kernel in _init */
131 
132 static load_avg  ccpu;
133 
134 /* these are used in the get_ functions */
135 
136 static int lastpid;
137 
138 /* these are for calculating cpu state percentages */
139 
140 static long cp_time[CPUSTATES];
141 static long cp_old[CPUSTATES];
142 static long cp_diff[CPUSTATES];
143 
144 /* these are for detailing the process states */
145 
146 static const char *procstatenames[] = {
147 	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
148 	" zombie, ", " waiting, ", " lock, ",
149 	NULL
150 };
151 static int process_states[nitems(procstatenames)];
152 
153 /* these are for detailing the cpu states */
154 
155 static int cpu_states[CPUSTATES];
156 static const char *cpustatenames[] = {
157 	"user", "nice", "system", "interrupt", "idle", NULL
158 };
159 
160 /* these are for detailing the memory statistics */
161 
162 static const char *memorynames[] = {
163 	"K Active, ", "K Inact, ", "K Laundry, ", "K Wired, ", "K Buf, ",
164 	"K Free", NULL
165 };
166 static int memory_stats[nitems(memorynames)];
167 
168 static const char *arcnames[] = {
169 	"K Total, ", "K MFU, ", "K MRU, ", "K Anon, ", "K Header, ", "K Other",
170 	NULL
171 };
172 static int arc_stats[nitems(arcnames)];
173 
174 static const char *carcnames[] = {
175 	"K Compressed, ", "K Uncompressed, ", ":1 Ratio, ",
176 	NULL
177 };
178 static int carc_stats[nitems(carcnames)];
179 
180 static const char *swapnames[] = {
181 	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
182 	NULL
183 };
184 static int swap_stats[nitems(swapnames)];
185 
186 
187 /* these are for keeping track of the proc array */
188 
189 static int nproc;
190 static int onproc = -1;
191 static int pref_len;
192 static struct kinfo_proc *pbase;
193 static struct kinfo_proc **pref;
194 static struct kinfo_proc *previous_procs;
195 static struct kinfo_proc **previous_pref;
196 static int previous_proc_count = 0;
197 static int previous_proc_count_max = 0;
198 static int previous_thread;
199 
200 /* data used for recalculating pctcpu */
201 static double *pcpu;
202 static struct timespec proc_uptime;
203 static struct timeval proc_wall_time;
204 static struct timeval previous_wall_time;
205 static uint64_t previous_interval = 0;
206 
207 /* total number of io operations */
208 static long total_inblock;
209 static long total_oublock;
210 static long total_majflt;
211 
212 /* these are for getting the memory statistics */
213 
214 static int arc_enabled;
215 static int carc_enabled;
216 static int pageshift;		/* log base 2 of the pagesize */
217 
218 /* define pagetok in terms of pageshift */
219 
220 #define pagetok(size) ((size) << pageshift)
221 
222 /* swap usage */
223 #define ki_swap(kip) \
224     ((kip)->ki_swrss > (kip)->ki_rssize ? (kip)->ki_swrss - (kip)->ki_rssize : 0)
225 
226 /*
227  * Sorting orders.  The first element is the default.
228  */
229 static const char *ordernames[] = {
230 	"cpu", "size", "res", "time", "pri", "threads",
231 	"total", "read", "write", "fault", "vcsw", "ivcsw",
232 	"jid", "swap", "pid", NULL
233 };
234 
235 /* Per-cpu time states */
236 static int maxcpu;
237 static int maxid;
238 static int ncpus;
239 static unsigned long cpumask;
240 static long *times;
241 static long *pcpu_cp_time;
242 static long *pcpu_cp_old;
243 static long *pcpu_cp_diff;
244 static int *pcpu_cpu_states;
245 
246 static int compare_swap(const void *a, const void *b);
247 static int compare_jid(const void *a, const void *b);
248 static int compare_pid(const void *a, const void *b);
249 static int compare_tid(const void *a, const void *b);
250 static const char *format_nice(const struct kinfo_proc *pp);
251 static void getsysctl(const char *name, void *ptr, size_t len);
252 static int swapmode(int *retavail, int *retfree);
253 static void update_layout(void);
254 static int find_uid(uid_t needle, int *haystack);
255 
256 static int
257 find_uid(uid_t needle, int *haystack)
258 {
259 	size_t i = 0;
260 
261 	for (; i < TOP_MAX_UIDS; ++i)
262 		if ((uid_t)haystack[i] == needle)
263 			return 1;
264 	return (0);
265 }
266 
267 void
268 toggle_pcpustats(void)
269 {
270 
271 	if (ncpus == 1)
272 		return;
273 	update_layout();
274 }
275 
276 /* Adjust display based on ncpus and the ARC state. */
277 static void
278 update_layout(void)
279 {
280 
281 	y_mem = 3;
282 	y_arc = 4;
283 	y_carc = 5;
284 	y_swap = 4 + arc_enabled + carc_enabled;
285 	y_idlecursor = 5 + arc_enabled + carc_enabled;
286 	y_message = 5 + arc_enabled + carc_enabled;
287 	y_header = 6 + arc_enabled + carc_enabled;
288 	y_procs = 7 + arc_enabled + carc_enabled;
289 	Header_lines = 7 + arc_enabled + carc_enabled;
290 
291 	if (pcpu_stats) {
292 		y_mem += ncpus - 1;
293 		y_arc += ncpus - 1;
294 		y_carc += ncpus - 1;
295 		y_swap += ncpus - 1;
296 		y_idlecursor += ncpus - 1;
297 		y_message += ncpus - 1;
298 		y_header += ncpus - 1;
299 		y_procs += ncpus - 1;
300 		Header_lines += ncpus - 1;
301 	}
302 }
303 
304 int
305 machine_init(struct statics *statics)
306 {
307 	int i, j, empty, pagesize;
308 	uint64_t arc_size;
309 	bool carc_en;
310 	size_t size;
311 
312 	size = sizeof(smpmode);
313 	if ((sysctlbyname("machdep.smp_active", &smpmode, &size,
314 	    NULL, 0) != 0 &&
315 	    sysctlbyname("kern.smp.active", &smpmode, &size,
316 	    NULL, 0) != 0) ||
317 	    size != sizeof(smpmode))
318 		smpmode = 0;
319 
320 	size = sizeof(arc_size);
321 	if (sysctlbyname("kstat.zfs.misc.arcstats.size", &arc_size, &size,
322 	    NULL, 0) == 0 && arc_size != 0)
323 		arc_enabled = 1;
324 	size = sizeof(carc_en);
325 	if (arc_enabled &&
326 	    sysctlbyname("vfs.zfs.compressed_arc_enabled", &carc_en, &size,
327 	    NULL, 0) == 0 && carc_en == 1)
328 		carc_enabled = 1;
329 
330 	namelength = MAXLOGNAME;
331 	if (smpmode && namelength > SMPUNAMELEN)
332 		namelength = SMPUNAMELEN;
333 	else if (namelength > UPUNAMELEN)
334 		namelength = UPUNAMELEN;
335 
336 	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
337 	if (kd == NULL)
338 		return (-1);
339 
340 	GETSYSCTL("kern.ccpu", ccpu);
341 
342 	/* this is used in calculating WCPU -- calculate it ahead of time */
343 	logcpu = log(loaddouble(ccpu));
344 
345 	pbase = NULL;
346 	pref = NULL;
347 	pcpu = NULL;
348 	nproc = 0;
349 	onproc = -1;
350 
351 	/* get the page size and calculate pageshift from it */
352 	pagesize = getpagesize();
353 	pageshift = 0;
354 	while (pagesize > 1) {
355 		pageshift++;
356 		pagesize >>= 1;
357 	}
358 
359 	/* we only need the amount of log(2)1024 for our conversion */
360 	pageshift -= LOG1024;
361 
362 	/* fill in the statics information */
363 	statics->procstate_names = procstatenames;
364 	statics->cpustate_names = cpustatenames;
365 	statics->memory_names = memorynames;
366 	if (arc_enabled)
367 		statics->arc_names = arcnames;
368 	else
369 		statics->arc_names = NULL;
370 	if (carc_enabled)
371 		statics->carc_names = carcnames;
372 	else
373 		statics->carc_names = NULL;
374 	statics->swap_names = swapnames;
375 	statics->order_names = ordernames;
376 
377 	/* Allocate state for per-CPU stats. */
378 	cpumask = 0;
379 	ncpus = 0;
380 	GETSYSCTL("kern.smp.maxcpus", maxcpu);
381 	times = calloc(maxcpu * CPUSTATES, sizeof(long));
382 	if (times == NULL)
383 		err(1, "calloc for kern.smp.maxcpus");
384 	size = sizeof(long) * maxcpu * CPUSTATES;
385 	if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1)
386 		err(1, "sysctlbyname kern.cp_times");
387 	pcpu_cp_time = calloc(1, size);
388 	maxid = (size / CPUSTATES / sizeof(long)) - 1;
389 	for (i = 0; i <= maxid; i++) {
390 		empty = 1;
391 		for (j = 0; empty && j < CPUSTATES; j++) {
392 			if (times[i * CPUSTATES + j] != 0)
393 				empty = 0;
394 		}
395 		if (!empty) {
396 			cpumask |= (1ul << i);
397 			ncpus++;
398 		}
399 	}
400 	assert(ncpus > 0);
401 	pcpu_cp_old = calloc(ncpus * CPUSTATES, sizeof(long));
402 	pcpu_cp_diff = calloc(ncpus * CPUSTATES, sizeof(long));
403 	pcpu_cpu_states = calloc(ncpus * CPUSTATES, sizeof(int));
404 	statics->ncpus = ncpus;
405 
406 	update_layout();
407 
408 	/* all done! */
409 	return (0);
410 }
411 
412 const char *
413 format_header(const char *uname_field)
414 {
415 	static char Header[128];
416 	const char *prehead;
417 
418 	if (ps.jail)
419 		jidlength = TOP_JID_LEN + 1;	/* +1 for extra left space. */
420 	else
421 		jidlength = 0;
422 
423 	if (ps.swap)
424 		swaplength = TOP_SWAP_LEN + 1;  /* +1 for extra left space */
425 	else
426 		swaplength = 0;
427 
428 	switch (displaymode) {
429 	case DISP_CPU:
430 		/*
431 		 * The logic of picking the right header is confusing, and
432 		 * depends on too much. We should instead have a struct of
433 		 * "header name", and "header format" which we build up.
434 		 * This would also fix the duplicate of effort into up vs smp
435 		 * mode.
436 		 */
437 		if (smpmode) {
438 			prehead = ps.thread ?
439 				smp_header_id_only : smp_header_thr_and_pid;
440 			snprintf(Header, sizeof(Header), prehead,
441 					ps.thread_id ? " THR" : "PID",
442 					jidlength, ps.jail ? " JID" : "",
443 					namelength, namelength, uname_field,
444 					swaplength, ps.swap ? " SWAP" : "",
445 					ps.wcpu ? "WCPU" : "CPU");
446 		} else {
447 			prehead = ps.thread ?
448 				up_header_id_only : up_header_thr_and_pid;
449 			snprintf(Header, sizeof(Header), prehead,
450 					ps.thread_id ? " THR" : "PID",
451 					jidlength, ps.jail ? " JID" : "",
452 					namelength, namelength, uname_field,
453 					swaplength, ps.swap ? " SWAP" : "",
454 					ps.wcpu ? "WCPU" : "CPU");
455 		}
456 		break;
457 	case DISP_IO:
458 		prehead = io_header;
459 		snprintf(Header, sizeof(Header), prehead,
460 		    jidlength, ps.jail ? " JID" : "",
461 		    namelength, namelength, uname_field);
462 		break;
463 	case DISP_MAX:
464 		assert("displaymode must not be set to DISP_MAX");
465 	}
466 	cmdlengthdelta = strlen(Header) - 7;
467 	return (Header);
468 }
469 
470 static int swappgsin = -1;
471 static int swappgsout = -1;
472 
473 
474 void
475 get_system_info(struct system_info *si)
476 {
477 	struct loadavg sysload;
478 	int mib[2];
479 	struct timeval boottime;
480 	uint64_t arc_stat, arc_stat2;
481 	int i, j;
482 	size_t size;
483 
484 	/* get the CPU stats */
485 	size = (maxid + 1) * CPUSTATES * sizeof(long);
486 	if (sysctlbyname("kern.cp_times", pcpu_cp_time, &size, NULL, 0) == -1)
487 		err(1, "sysctlbyname kern.cp_times");
488 	GETSYSCTL("kern.cp_time", cp_time);
489 	GETSYSCTL("vm.loadavg", sysload);
490 	GETSYSCTL("kern.lastpid", lastpid);
491 
492 	/* convert load averages to doubles */
493 	for (i = 0; i < 3; i++)
494 		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
495 
496 	/* convert cp_time counts to percentages */
497 	for (i = j = 0; i <= maxid; i++) {
498 		if ((cpumask & (1ul << i)) == 0)
499 			continue;
500 		percentages(CPUSTATES, &pcpu_cpu_states[j * CPUSTATES],
501 		    &pcpu_cp_time[j * CPUSTATES],
502 		    &pcpu_cp_old[j * CPUSTATES],
503 		    &pcpu_cp_diff[j * CPUSTATES]);
504 		j++;
505 	}
506 	percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
507 
508 	/* sum memory & swap statistics */
509 	{
510 		static unsigned int swap_delay = 0;
511 		static int swapavail = 0;
512 		static int swapfree = 0;
513 		static long bufspace = 0;
514 		static uint64_t nspgsin, nspgsout;
515 
516 		GETSYSCTL("vfs.bufspace", bufspace);
517 		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
518 		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
519 		GETSYSCTL("vm.stats.vm.v_laundry_count", memory_stats[2]);
520 		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[3]);
521 		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
522 		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
523 		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
524 		/* convert memory stats to Kbytes */
525 		memory_stats[0] = pagetok(memory_stats[0]);
526 		memory_stats[1] = pagetok(memory_stats[1]);
527 		memory_stats[2] = pagetok(memory_stats[2]);
528 		memory_stats[3] = pagetok(memory_stats[3]);
529 		memory_stats[4] = bufspace / 1024;
530 		memory_stats[5] = pagetok(memory_stats[5]);
531 		memory_stats[6] = -1;
532 
533 		/* first interval */
534 		if (swappgsin < 0) {
535 			swap_stats[4] = 0;
536 			swap_stats[5] = 0;
537 		}
538 
539 		/* compute differences between old and new swap statistic */
540 		else {
541 			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
542 			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
543 		}
544 
545 		swappgsin = nspgsin;
546 		swappgsout = nspgsout;
547 
548 		/* call CPU heavy swapmode() only for changes */
549 		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
550 			swap_stats[3] = swapmode(&swapavail, &swapfree);
551 			swap_stats[0] = swapavail;
552 			swap_stats[1] = swapavail - swapfree;
553 			swap_stats[2] = swapfree;
554 		}
555 		swap_delay = 1;
556 		swap_stats[6] = -1;
557 	}
558 
559 	if (arc_enabled) {
560 		GETSYSCTL("kstat.zfs.misc.arcstats.size", arc_stat);
561 		arc_stats[0] = arc_stat >> 10;
562 		GETSYSCTL("vfs.zfs.mfu_size", arc_stat);
563 		arc_stats[1] = arc_stat >> 10;
564 		GETSYSCTL("vfs.zfs.mru_size", arc_stat);
565 		arc_stats[2] = arc_stat >> 10;
566 		GETSYSCTL("vfs.zfs.anon_size", arc_stat);
567 		arc_stats[3] = arc_stat >> 10;
568 		GETSYSCTL("kstat.zfs.misc.arcstats.hdr_size", arc_stat);
569 		GETSYSCTL("kstat.zfs.misc.arcstats.l2_hdr_size", arc_stat2);
570 		arc_stats[4] = (arc_stat + arc_stat2) >> 10;
571 		GETSYSCTL("kstat.zfs.misc.arcstats.other_size", arc_stat);
572 		arc_stats[5] = arc_stat >> 10;
573 		si->arc = arc_stats;
574 	}
575 	if (carc_enabled) {
576 		GETSYSCTL("kstat.zfs.misc.arcstats.compressed_size", arc_stat);
577 		carc_stats[0] = arc_stat >> 10;
578 		carc_stats[2] = arc_stat >> 10; /* For ratio */
579 		GETSYSCTL("kstat.zfs.misc.arcstats.uncompressed_size", arc_stat);
580 		carc_stats[1] = arc_stat >> 10;
581 		si->carc = carc_stats;
582 	}
583 
584 	/* set arrays and strings */
585 	if (pcpu_stats) {
586 		si->cpustates = pcpu_cpu_states;
587 		si->ncpus = ncpus;
588 	} else {
589 		si->cpustates = cpu_states;
590 		si->ncpus = 1;
591 	}
592 	si->memory = memory_stats;
593 	si->swap = swap_stats;
594 
595 
596 	if (lastpid > 0) {
597 		si->last_pid = lastpid;
598 	} else {
599 		si->last_pid = -1;
600 	}
601 
602 	/*
603 	 * Print how long system has been up.
604 	 * (Found by looking getting "boottime" from the kernel)
605 	 */
606 	mib[0] = CTL_KERN;
607 	mib[1] = KERN_BOOTTIME;
608 	size = sizeof(boottime);
609 	if (sysctl(mib, nitems(mib), &boottime, &size, NULL, 0) != -1 &&
610 	    boottime.tv_sec != 0) {
611 		si->boottime = boottime;
612 	} else {
613 		si->boottime.tv_sec = -1;
614 	}
615 }
616 
617 #define NOPROC	((void *)-1)
618 
619 /*
620  * We need to compare data from the old process entry with the new
621  * process entry.
622  * To facilitate doing this quickly we stash a pointer in the kinfo_proc
623  * structure to cache the mapping.  We also use a negative cache pointer
624  * of NOPROC to avoid duplicate lookups.
625  * XXX: this could be done when the actual processes are fetched, we do
626  * it here out of laziness.
627  */
628 static const struct kinfo_proc *
629 get_old_proc(struct kinfo_proc *pp)
630 {
631 	const struct kinfo_proc * const *oldpp, *oldp;
632 
633 	/*
634 	 * If this is the first fetch of the kinfo_procs then we don't have
635 	 * any previous entries.
636 	 */
637 	if (previous_proc_count == 0)
638 		return (NULL);
639 	/* negative cache? */
640 	if (pp->ki_udata == NOPROC)
641 		return (NULL);
642 	/* cached? */
643 	if (pp->ki_udata != NULL)
644 		return (pp->ki_udata);
645 	/*
646 	 * Not cached,
647 	 * 1) look up based on pid.
648 	 * 2) compare process start.
649 	 * If we fail here, then setup a negative cache entry, otherwise
650 	 * cache it.
651 	 */
652 	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
653 	    sizeof(*previous_pref), ps.thread ? compare_tid : compare_pid);
654 	if (oldpp == NULL) {
655 		pp->ki_udata = NOPROC;
656 		return (NULL);
657 	}
658 	oldp = *oldpp;
659 	if (memcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
660 		pp->ki_udata = NOPROC;
661 		return (NULL);
662 	}
663 	pp->ki_udata = oldp;
664 	return (oldp);
665 }
666 
667 /*
668  * Return the total amount of IO done in blocks in/out and faults.
669  * store the values individually in the pointers passed in.
670  */
671 static long
672 get_io_stats(const struct kinfo_proc *pp, long *inp, long *oup, long *flp,
673     long *vcsw, long *ivcsw)
674 {
675 	const struct kinfo_proc *oldp;
676 	static struct kinfo_proc dummy;
677 	long ret;
678 
679 	oldp = get_old_proc(pp);
680 	if (oldp == NULL) {
681 		memset(&dummy, 0, sizeof(dummy));
682 		oldp = &dummy;
683 	}
684 	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
685 	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
686 	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
687 	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
688 	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
689 	ret =
690 	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
691 	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
692 	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
693 	return (ret);
694 }
695 
696 /*
697  * If there was a previous update, use the delta in ki_runtime over
698  * the previous interval to calculate pctcpu.  Otherwise, fall back
699  * to using the kernel's ki_pctcpu.
700  */
701 static double
702 proc_calc_pctcpu(struct kinfo_proc *pp)
703 {
704 	const struct kinfo_proc *oldp;
705 
706 	if (previous_interval != 0) {
707 		oldp = get_old_proc(pp);
708 		if (oldp != NULL)
709 			return ((double)(pp->ki_runtime - oldp->ki_runtime)
710 			    / previous_interval);
711 
712 		/*
713 		 * If this process/thread was created during the previous
714 		 * interval, charge it's total runtime to the previous
715 		 * interval.
716 		 */
717 		else if (pp->ki_start.tv_sec > previous_wall_time.tv_sec ||
718 		    (pp->ki_start.tv_sec == previous_wall_time.tv_sec &&
719 		    pp->ki_start.tv_usec >= previous_wall_time.tv_usec))
720 			return ((double)pp->ki_runtime / previous_interval);
721 	}
722 	return (pctdouble(pp->ki_pctcpu));
723 }
724 
725 /*
726  * Return true if this process has used any CPU time since the
727  * previous update.
728  */
729 static int
730 proc_used_cpu(struct kinfo_proc *pp)
731 {
732 	const struct kinfo_proc *oldp;
733 
734 	oldp = get_old_proc(pp);
735 	if (oldp == NULL)
736 		return (PCTCPU(pp) != 0);
737 	return (pp->ki_runtime != oldp->ki_runtime ||
738 	    RU(pp)->ru_nvcsw != RU(oldp)->ru_nvcsw ||
739 	    RU(pp)->ru_nivcsw != RU(oldp)->ru_nivcsw);
740 }
741 
742 /*
743  * Return the total number of block in/out and faults by a process.
744  */
745 static long
746 get_io_total(const struct kinfo_proc *pp)
747 {
748 	long dummy;
749 
750 	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
751 }
752 
753 static struct handle handle;
754 
755 void *
756 get_process_info(struct system_info *si, struct process_select *sel,
757     int (*compare)(const void *, const void *))
758 {
759 	int i;
760 	int total_procs;
761 	long p_io;
762 	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
763 	long nsec;
764 	int active_procs;
765 	struct kinfo_proc **prefp;
766 	struct kinfo_proc *pp;
767 	struct timespec previous_proc_uptime;
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 		memcpy(previous_procs, pbase, 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 	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 		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 	/* count up process states and get pointers to interesting procs */
831 	total_procs = 0;
832 	active_procs = 0;
833 	total_inblock = 0;
834 	total_oublock = 0;
835 	total_majflt = 0;
836 	memset(process_states, 0, sizeof(process_states));
837 	prefp = pref;
838 	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
839 
840 		if (pp->ki_stat == 0)
841 			/* not in use */
842 			continue;
843 
844 		if (!sel->self && pp->ki_pid == mypid)
845 			/* skip self */
846 			continue;
847 
848 		if (!sel->system && (pp->ki_flag & P_SYSTEM))
849 			/* skip system process */
850 			continue;
851 
852 		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
853 		    &p_vcsw, &p_ivcsw);
854 		total_inblock += p_inblock;
855 		total_oublock += p_oublock;
856 		total_majflt += p_majflt;
857 		total_procs++;
858 		process_states[(unsigned char)pp->ki_stat]++;
859 
860 		if (pp->ki_stat == SZOMB)
861 			/* skip zombies */
862 			continue;
863 
864 		if (!sel->kidle && pp->ki_tdflags & TDF_IDLETD)
865 			/* skip kernel idle process */
866 			continue;
867 
868 		PCTCPU(pp) = proc_calc_pctcpu(pp);
869 		if (sel->thread && PCTCPU(pp) > 1.0)
870 			PCTCPU(pp) = 1.0;
871 		if (displaymode == DISP_CPU && !sel->idle &&
872 		    (!proc_used_cpu(pp) ||
873 		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
874 			/* skip idle or non-running processes */
875 			continue;
876 
877 		if (displaymode == DISP_IO && !sel->idle && p_io == 0)
878 			/* skip processes that aren't doing I/O */
879 			continue;
880 
881 		if (sel->jid != -1 && pp->ki_jid != sel->jid)
882 			/* skip proc. that don't belong to the selected JID */
883 			continue;
884 
885 		if (sel->uid[0] != -1 && !find_uid(pp->ki_ruid, sel->uid))
886 			/* skip proc. that don't belong to the selected UID */
887 			continue;
888 
889 		if (sel->pid != -1 && pp->ki_pid != sel->pid)
890 			continue;
891 
892 		*prefp++ = pp;
893 		active_procs++;
894 	}
895 
896 	/* if requested, sort the "interesting" processes */
897 	if (compare != NULL)
898 		qsort(pref, active_procs, sizeof(*pref), compare);
899 
900 	/* remember active and total counts */
901 	si->p_total = total_procs;
902 	si->p_pactive = pref_len = active_procs;
903 
904 	/* pass back a handle */
905 	handle.next_proc = pref;
906 	handle.remaining = active_procs;
907 	return (&handle);
908 }
909 
910 static char fmt[512];	/* static area where result is built */
911 
912 char *
913 format_next_process(void* xhandle, char *(*get_userid)(int), int flags)
914 {
915 	struct kinfo_proc *pp;
916 	const struct kinfo_proc *oldp;
917 	long cputime;
918 	double pct;
919 	struct handle *hp;
920 	char status[22];
921 	int cpu;
922 	size_t state;
923 	struct rusage ru, *rup;
924 	long p_tot, s_tot;
925 	const char *proc_fmt;
926 	char thr_buf[6];
927 	char jid_buf[TOP_JID_LEN + 1], swap_buf[TOP_SWAP_LEN + 1];
928 	char *cmdbuf = NULL;
929 	char **args;
930 	const int cmdlen = 128;
931 
932 	/* find and remember the next proc structure */
933 	hp = (struct handle *)xhandle;
934 	pp = *(hp->next_proc++);
935 	hp->remaining--;
936 
937 	/* get the process's command name */
938 	if ((pp->ki_flag & P_INMEM) == 0) {
939 		/*
940 		 * Print swapped processes as <pname>
941 		 */
942 		size_t len;
943 
944 		len = strlen(pp->ki_comm);
945 		if (len > sizeof(pp->ki_comm) - 3)
946 			len = sizeof(pp->ki_comm) - 3;
947 		memmove(pp->ki_comm + 1, pp->ki_comm, len);
948 		pp->ki_comm[0] = '<';
949 		pp->ki_comm[len + 1] = '>';
950 		pp->ki_comm[len + 2] = '\0';
951 	}
952 
953 	/*
954 	 * Convert the process's runtime from microseconds to seconds.  This
955 	 * time includes the interrupt time although that is not wanted here.
956 	 * ps(1) is similarly sloppy.
957 	 */
958 	cputime = (pp->ki_runtime + 500000) / 1000000;
959 
960 	/* calculate the base for cpu percentages */
961 	pct = PCTCPU(pp);
962 
963 	/* generate "STATE" field */
964 	switch (state = pp->ki_stat) {
965 	case SRUN:
966 		if (smpmode && pp->ki_oncpu != NOCPU)
967 			sprintf(status, "CPU%d", pp->ki_oncpu);
968 		else
969 			strcpy(status, "RUN");
970 		break;
971 	case SLOCK:
972 		if (pp->ki_kiflag & KI_LOCKBLOCK) {
973 			sprintf(status, "*%.6s", pp->ki_lockname);
974 			break;
975 		}
976 		/* fall through */
977 	case SSLEEP:
978 		sprintf(status, "%.6s", pp->ki_wmesg);
979 		break;
980 	default:
981 
982 		if (state < nitems(state_abbrev)) {
983 			sprintf(status, "%.6s", state_abbrev[state]);
984 		} else {
985 			sprintf(status, "?%5zu", state);
986 		}
987 		break;
988 	}
989 
990 	cmdbuf = calloc(cmdlen + 1, 1);
991 	if (cmdbuf == NULL) {
992 		warn("calloc(%d)", cmdlen + 1);
993 		return NULL;
994 	}
995 
996 	if (!(flags & FMT_SHOWARGS)) {
997 		if (ps.thread && pp->ki_flag & P_HADTHREADS &&
998 		    pp->ki_tdname[0]) {
999 			snprintf(cmdbuf, cmdlen, "%s{%s%s}", pp->ki_comm,
1000 			    pp->ki_tdname, pp->ki_moretdname);
1001 		} else {
1002 			snprintf(cmdbuf, cmdlen, "%s", pp->ki_comm);
1003 		}
1004 	} else {
1005 		if (pp->ki_flag & P_SYSTEM ||
1006 		    pp->ki_args == NULL ||
1007 		    (args = kvm_getargv(kd, pp, cmdlen)) == NULL ||
1008 		    !(*args)) {
1009 			if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1010 		    	    pp->ki_tdname[0]) {
1011 				snprintf(cmdbuf, cmdlen,
1012 				    "[%s{%s%s}]", pp->ki_comm, pp->ki_tdname,
1013 				    pp->ki_moretdname);
1014 			} else {
1015 				snprintf(cmdbuf, cmdlen,
1016 				    "[%s]", pp->ki_comm);
1017 			}
1018 		} else {
1019 			const char *src;
1020 			char *dst, *argbuf;
1021 			const char *cmd;
1022 			size_t argbuflen;
1023 			size_t len;
1024 
1025 			argbuflen = cmdlen * 4;
1026 			argbuf = calloc(argbuflen + 1, 1);
1027 			if (argbuf == NULL) {
1028 				warn("calloc(%zu)", argbuflen + 1);
1029 				free(cmdbuf);
1030 				return NULL;
1031 			}
1032 
1033 			dst = argbuf;
1034 
1035 			/* Extract cmd name from argv */
1036 			cmd = strrchr(*args, '/');
1037 			if (cmd == NULL)
1038 				cmd = *args;
1039 			else
1040 				cmd++;
1041 
1042 			for (; (src = *args++) != NULL; ) {
1043 				if (*src == '\0')
1044 					continue;
1045 				len = (argbuflen - (dst - argbuf) - 1) / 4;
1046 				strvisx(dst, src,
1047 				    MIN(strlen(src), len),
1048 				    VIS_NL | VIS_CSTYLE);
1049 				while (*dst != '\0')
1050 					dst++;
1051 				if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
1052 					*dst++ = ' '; /* add delimiting space */
1053 			}
1054 			if (dst != argbuf && dst[-1] == ' ')
1055 				dst--;
1056 			*dst = '\0';
1057 
1058 			if (strcmp(cmd, pp->ki_comm) != 0) {
1059 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1060 				    pp->ki_tdname[0])
1061 					snprintf(cmdbuf, cmdlen,
1062 					    "%s (%s){%s%s}", argbuf,
1063 					    pp->ki_comm, pp->ki_tdname,
1064 					    pp->ki_moretdname);
1065 				else
1066 					snprintf(cmdbuf, cmdlen,
1067 					    "%s (%s)", argbuf, pp->ki_comm);
1068 			} else {
1069 				if (ps.thread && pp->ki_flag & P_HADTHREADS &&
1070 				    pp->ki_tdname[0])
1071 					snprintf(cmdbuf, cmdlen,
1072 					    "%s{%s%s}", argbuf, pp->ki_tdname,
1073 					    pp->ki_moretdname);
1074 				else
1075 					strlcpy(cmdbuf, argbuf, cmdlen);
1076 			}
1077 			free(argbuf);
1078 		}
1079 	}
1080 
1081 	if (ps.jail == 0)
1082 		jid_buf[0] = '\0';
1083 	else
1084 		snprintf(jid_buf, sizeof(jid_buf), "%*d",
1085 		    jidlength - 1, pp->ki_jid);
1086 
1087 	if (ps.swap == 0)
1088 		swap_buf[0] = '\0';
1089 	else
1090 		snprintf(swap_buf, sizeof(swap_buf), "%*s",
1091 		    swaplength - 1,
1092 		    format_k2(pagetok(ki_swap(pp)))); /* XXX */
1093 
1094 	if (displaymode == DISP_IO) {
1095 		oldp = get_old_proc(pp);
1096 		if (oldp != NULL) {
1097 			ru.ru_inblock = RU(pp)->ru_inblock -
1098 			    RU(oldp)->ru_inblock;
1099 			ru.ru_oublock = RU(pp)->ru_oublock -
1100 			    RU(oldp)->ru_oublock;
1101 			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
1102 			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
1103 			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
1104 			rup = &ru;
1105 		} else {
1106 			rup = RU(pp);
1107 		}
1108 		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
1109 		s_tot = total_inblock + total_oublock + total_majflt;
1110 
1111 		snprintf(fmt, sizeof(fmt), io_Proc_format,
1112 		    pp->ki_pid,
1113 		    jidlength, jid_buf,
1114 		    namelength, namelength, (*get_userid)(pp->ki_ruid),
1115 		    rup->ru_nvcsw,
1116 		    rup->ru_nivcsw,
1117 		    rup->ru_inblock,
1118 		    rup->ru_oublock,
1119 		    rup->ru_majflt,
1120 		    p_tot,
1121 		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
1122 		    screen_width > cmdlengthdelta ?
1123 		    screen_width - cmdlengthdelta : 0,
1124 		    printable(cmdbuf));
1125 
1126 		free(cmdbuf);
1127 
1128 		return (fmt);
1129 	}
1130 
1131 	/* format this entry */
1132 	if (smpmode) {
1133 		if (state == SRUN && pp->ki_oncpu != NOCPU)
1134 			cpu = pp->ki_oncpu;
1135 		else
1136 			cpu = pp->ki_lastcpu;
1137 	} else
1138 		cpu = 0;
1139 	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
1140 	if (ps.thread != 0)
1141 		thr_buf[0] = '\0';
1142 	else
1143 		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
1144 		    (int)(sizeof(thr_buf) - 2), pp->ki_numthreads);
1145 
1146 	snprintf(fmt, sizeof(fmt), proc_fmt,
1147 	    (ps.thread_id) ? pp->ki_tid : pp->ki_pid,
1148 	    jidlength, jid_buf,
1149 	    namelength, namelength, (*get_userid)(pp->ki_ruid),
1150 	    thr_buf,
1151 	    pp->ki_pri.pri_level - PZERO,
1152 	    format_nice(pp),
1153 	    format_k2(PROCSIZE(pp)),
1154 	    format_k2(pagetok(pp->ki_rssize)),
1155 	    swaplength, swaplength, swap_buf,
1156 	    status,
1157 	    cpu,
1158 	    format_time(cputime),
1159 	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
1160 	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
1161 	    printable(cmdbuf));
1162 
1163 	free(cmdbuf);
1164 
1165 	/* return the result */
1166 	return (fmt);
1167 }
1168 
1169 static void
1170 getsysctl(const char *name, void *ptr, size_t len)
1171 {
1172 	size_t nlen = len;
1173 
1174 	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
1175 		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
1176 		    strerror(errno));
1177 		quit(TOP_EX_SYS_ERROR);
1178 	}
1179 	if (nlen != len) {
1180 		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
1181 		    name, (unsigned long)len, (unsigned long)nlen);
1182 		quit(TOP_EX_SYS_ERROR);
1183 	}
1184 }
1185 
1186 static const char *
1187 format_nice(const struct kinfo_proc *pp)
1188 {
1189 	const char *fifo, *kproc;
1190 	int rtpri;
1191 	static char nicebuf[4 + 1];
1192 
1193 	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
1194 	kproc = (pp->ki_flag & P_KPROC) ? "k" : "";
1195 	switch (PRI_BASE(pp->ki_pri.pri_class)) {
1196 	case PRI_ITHD:
1197 		return ("-");
1198 	case PRI_REALTIME:
1199 		/*
1200 		 * XXX: the kernel doesn't tell us the original rtprio and
1201 		 * doesn't really know what it was, so to recover it we
1202 		 * must be more chummy with the implementation than the
1203 		 * implementation is with itself.  pri_user gives a
1204 		 * constant "base" priority, but is only initialized
1205 		 * properly for user threads.  pri_native gives what the
1206 		 * kernel calls the "base" priority, but it isn't constant
1207 		 * since it is changed by priority propagation.  pri_native
1208 		 * also isn't properly initialized for all threads, but it
1209 		 * is properly initialized for kernel realtime and idletime
1210 		 * threads.  Thus we use pri_user for the base priority of
1211 		 * user threads (it is always correct) and pri_native for
1212 		 * the base priority of kernel realtime and idletime threads
1213 		 * (there is nothing better, and it is usually correct).
1214 		 *
1215 		 * The field width and thus the buffer are too small for
1216 		 * values like "kr31F", but such values shouldn't occur,
1217 		 * and if they do then the tailing "F" is not displayed.
1218 		 */
1219 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1220 		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
1221 		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
1222 		    kproc, rtpri, fifo);
1223 		break;
1224 	case PRI_TIMESHARE:
1225 		if (pp->ki_flag & P_KPROC)
1226 			return ("-");
1227 		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
1228 		break;
1229 	case PRI_IDLE:
1230 		/* XXX: as above. */
1231 		rtpri = ((pp->ki_flag & P_KPROC) ? pp->ki_pri.pri_native :
1232 		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
1233 		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
1234 		    kproc, rtpri, fifo);
1235 		break;
1236 	default:
1237 		return ("?");
1238 	}
1239 	return (nicebuf);
1240 }
1241 
1242 /* comparison routines for qsort */
1243 
1244 static int
1245 compare_pid(const void *p1, const void *p2)
1246 {
1247 	const struct kinfo_proc * const *pp1 = p1;
1248 	const struct kinfo_proc * const *pp2 = p2;
1249 
1250 	assert((*pp2)->ki_pid >= 0 && (*pp1)->ki_pid >= 0);
1251 
1252 	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
1253 }
1254 
1255 static int
1256 compare_tid(const void *p1, const void *p2)
1257 {
1258 	const struct kinfo_proc * const *pp1 = p1;
1259 	const struct kinfo_proc * const *pp2 = p2;
1260 
1261 	assert((*pp2)->ki_tid >= 0 && (*pp1)->ki_tid >= 0);
1262 
1263 	return ((*pp1)->ki_tid - (*pp2)->ki_tid);
1264 }
1265 
1266 /*
1267  *  proc_compare - comparison function for "qsort"
1268  *	Compares the resource consumption of two processes using five
1269  *	distinct keys.  The keys (in descending order of importance) are:
1270  *	percent cpu, cpu ticks, state, resident set size, total virtual
1271  *	memory usage.  The process states are ordered as follows (from least
1272  *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
1273  *	array declaration below maps a process state index into a number
1274  *	that reflects this ordering.
1275  */
1276 
1277 static int sorted_state[] = {
1278 	0,	/* not used		*/
1279 	3,	/* sleep		*/
1280 	1,	/* ABANDONED (WAIT)	*/
1281 	6,	/* run			*/
1282 	5,	/* start		*/
1283 	2,	/* zombie		*/
1284 	4	/* stop			*/
1285 };
1286 
1287 
1288 #define ORDERKEY_PCTCPU(a, b) do { \
1289 	double diff; \
1290 	if (ps.wcpu) \
1291 		diff = weighted_cpu(PCTCPU((b)), (b)) - \
1292 		    weighted_cpu(PCTCPU((a)), (a)); \
1293 	else \
1294 		diff = PCTCPU((b)) - PCTCPU((a)); \
1295 	if (diff != 0) \
1296 		return (diff > 0 ? 1 : -1); \
1297 } while (0)
1298 
1299 #define ORDERKEY_CPTICKS(a, b) do { \
1300 	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
1301 	if (diff != 0) \
1302 		return (diff > 0 ? 1 : -1); \
1303 } while (0)
1304 
1305 #define ORDERKEY_STATE(a, b) do { \
1306 	int diff = sorted_state[(unsigned char)(b)->ki_stat] - sorted_state[(unsigned char)(a)->ki_stat]; \
1307 	if (diff != 0) \
1308 		return (diff > 0 ? 1 : -1); \
1309 } while (0)
1310 
1311 #define ORDERKEY_PRIO(a, b) do { \
1312 	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1313 	if (diff != 0) \
1314 		return (diff > 0 ? 1 : -1); \
1315 } while (0)
1316 
1317 #define	ORDERKEY_THREADS(a, b) do { \
1318 	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1319 	if (diff != 0) \
1320 		return (diff > 0 ? 1 : -1); \
1321 } while (0)
1322 
1323 #define ORDERKEY_RSSIZE(a, b) do { \
1324 	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1325 	if (diff != 0) \
1326 		return (diff > 0 ? 1 : -1); \
1327 } while (0)
1328 
1329 #define ORDERKEY_MEM(a, b) do { \
1330 	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1331 	if (diff != 0) \
1332 		return (diff > 0 ? 1 : -1); \
1333 } while (0)
1334 
1335 #define ORDERKEY_JID(a, b) do { \
1336 	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1337 	if (diff != 0) \
1338 		return (diff > 0 ? 1 : -1); \
1339 } while (0)
1340 
1341 #define ORDERKEY_SWAP(a, b) do { \
1342 	int diff = (int)ki_swap(b) - (int)ki_swap(a); \
1343 	if (diff != 0) \
1344 		return (diff > 0 ? 1 : -1); \
1345 } while (0)
1346 
1347 /* compare_cpu - the comparison function for sorting by cpu percentage */
1348 
1349 static int
1350 compare_cpu(const void *arg1, const void *arg2)
1351 {
1352 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1353 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1354 
1355 	ORDERKEY_PCTCPU(p1, p2);
1356 	ORDERKEY_CPTICKS(p1, p2);
1357 	ORDERKEY_STATE(p1, p2);
1358 	ORDERKEY_PRIO(p1, p2);
1359 	ORDERKEY_RSSIZE(p1, p2);
1360 	ORDERKEY_MEM(p1, p2);
1361 
1362 	return (0);
1363 }
1364 
1365 /* compare_size - the comparison function for sorting by total memory usage */
1366 
1367 static int
1368 compare_size(const void *arg1, const void *arg2)
1369 {
1370 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1371 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1372 
1373 	ORDERKEY_MEM(p1, p2);
1374 	ORDERKEY_RSSIZE(p1, p2);
1375 	ORDERKEY_PCTCPU(p1, p2);
1376 	ORDERKEY_CPTICKS(p1, p2);
1377 	ORDERKEY_STATE(p1, p2);
1378 	ORDERKEY_PRIO(p1, p2);
1379 
1380 	return (0);
1381 }
1382 
1383 /* compare_res - the comparison function for sorting by resident set size */
1384 
1385 static int
1386 compare_res(const void *arg1, const void *arg2)
1387 {
1388 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1389 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1390 
1391 	ORDERKEY_RSSIZE(p1, p2);
1392 	ORDERKEY_MEM(p1, p2);
1393 	ORDERKEY_PCTCPU(p1, p2);
1394 	ORDERKEY_CPTICKS(p1, p2);
1395 	ORDERKEY_STATE(p1, p2);
1396 	ORDERKEY_PRIO(p1, p2);
1397 
1398 	return (0);
1399 }
1400 
1401 /* compare_time - the comparison function for sorting by total cpu time */
1402 
1403 static int
1404 compare_time(const void *arg1, const void *arg2)
1405 {
1406 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const  *)arg1;
1407 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *) arg2;
1408 
1409 	ORDERKEY_CPTICKS(p1, p2);
1410 	ORDERKEY_PCTCPU(p1, p2);
1411 	ORDERKEY_STATE(p1, p2);
1412 	ORDERKEY_PRIO(p1, p2);
1413 	ORDERKEY_RSSIZE(p1, p2);
1414 	ORDERKEY_MEM(p1, p2);
1415 
1416 	return (0);
1417 }
1418 
1419 /* compare_prio - the comparison function for sorting by priority */
1420 
1421 static int
1422 compare_prio(const void *arg1, const void *arg2)
1423 {
1424 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1425 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1426 
1427 	ORDERKEY_PRIO(p1, p2);
1428 	ORDERKEY_CPTICKS(p1, p2);
1429 	ORDERKEY_PCTCPU(p1, p2);
1430 	ORDERKEY_STATE(p1, p2);
1431 	ORDERKEY_RSSIZE(p1, p2);
1432 	ORDERKEY_MEM(p1, p2);
1433 
1434 	return (0);
1435 }
1436 
1437 /* compare_threads - the comparison function for sorting by threads */
1438 static int
1439 compare_threads(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 
1444 	ORDERKEY_THREADS(p1, p2);
1445 	ORDERKEY_PCTCPU(p1, p2);
1446 	ORDERKEY_CPTICKS(p1, p2);
1447 	ORDERKEY_STATE(p1, p2);
1448 	ORDERKEY_PRIO(p1, p2);
1449 	ORDERKEY_RSSIZE(p1, p2);
1450 	ORDERKEY_MEM(p1, p2);
1451 
1452 	return (0);
1453 }
1454 
1455 /* compare_jid - the comparison function for sorting by jid */
1456 static int
1457 compare_jid(const void *arg1, const void *arg2)
1458 {
1459 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1460 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1461 
1462 	ORDERKEY_JID(p1, p2);
1463 	ORDERKEY_PCTCPU(p1, p2);
1464 	ORDERKEY_CPTICKS(p1, p2);
1465 	ORDERKEY_STATE(p1, p2);
1466 	ORDERKEY_PRIO(p1, p2);
1467 	ORDERKEY_RSSIZE(p1, p2);
1468 	ORDERKEY_MEM(p1, p2);
1469 
1470 	return (0);
1471 }
1472 
1473 /* compare_swap - the comparison function for sorting by swap */
1474 static int
1475 compare_swap(const void *arg1, const void *arg2)
1476 {
1477 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1478 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1479 
1480 	ORDERKEY_SWAP(p1, p2);
1481 	ORDERKEY_PCTCPU(p1, p2);
1482 	ORDERKEY_CPTICKS(p1, p2);
1483 	ORDERKEY_STATE(p1, p2);
1484 	ORDERKEY_PRIO(p1, p2);
1485 	ORDERKEY_RSSIZE(p1, p2);
1486 	ORDERKEY_MEM(p1, p2);
1487 
1488 	return (0);
1489 }
1490 
1491 /* assorted comparison functions for sorting by i/o */
1492 
1493 static int
1494 compare_iototal(const void *arg1, const void *arg2)
1495 {
1496 	const struct kinfo_proc * const p1 = *(const struct kinfo_proc * const *)arg1;
1497 	const struct kinfo_proc * const p2 = *(const struct kinfo_proc * const *)arg2;
1498 
1499 	return (get_io_total(p2) - get_io_total(p1));
1500 }
1501 
1502 static int
1503 compare_ioread(const void *arg1, const void *arg2)
1504 {
1505 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1506 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1507 	long dummy, inp1, inp2;
1508 
1509 	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1510 	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1511 
1512 	return (inp2 - inp1);
1513 }
1514 
1515 static int
1516 compare_iowrite(const void *arg1, const void *arg2)
1517 {
1518 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1519 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1520 	long dummy, oup1, oup2;
1521 
1522 	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1523 	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1524 
1525 	return (oup2 - oup1);
1526 }
1527 
1528 static int
1529 compare_iofault(const void *arg1, const void *arg2)
1530 {
1531 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1532 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1533 	long dummy, flp1, flp2;
1534 
1535 	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1536 	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1537 
1538 	return (flp2 - flp1);
1539 }
1540 
1541 static int
1542 compare_vcsw(const void *arg1, const void *arg2)
1543 {
1544 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1545 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1546 	long dummy, flp1, flp2;
1547 
1548 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1549 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1550 
1551 	return (flp2 - flp1);
1552 }
1553 
1554 static int
1555 compare_ivcsw(const void *arg1, const void *arg2)
1556 {
1557 	const struct kinfo_proc *p1 = *(const struct kinfo_proc * const *)arg1;
1558 	const struct kinfo_proc *p2 = *(const struct kinfo_proc * const *)arg2;
1559 	long dummy, flp1, flp2;
1560 
1561 	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1562 	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1563 
1564 	return (flp2 - flp1);
1565 }
1566 
1567 int (*compares[])(const void *arg1, const void *arg2) = {
1568 	compare_cpu,
1569 	compare_size,
1570 	compare_res,
1571 	compare_time,
1572 	compare_prio,
1573 	compare_threads,
1574 	compare_iototal,
1575 	compare_ioread,
1576 	compare_iowrite,
1577 	compare_iofault,
1578 	compare_vcsw,
1579 	compare_ivcsw,
1580 	compare_jid,
1581 	compare_swap,
1582 	NULL
1583 };
1584 
1585 
1586 /*
1587  * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1588  *		the process does not exist.
1589  */
1590 
1591 int
1592 proc_owner(int pid)
1593 {
1594 	int cnt;
1595 	struct kinfo_proc **prefp;
1596 	struct kinfo_proc *pp;
1597 
1598 	prefp = pref;
1599 	cnt = pref_len;
1600 	while (--cnt >= 0) {
1601 		pp = *prefp++;
1602 		if (pp->ki_pid == (pid_t)pid)
1603 			return ((int)pp->ki_ruid);
1604 	}
1605 	return (-1);
1606 }
1607 
1608 static int
1609 swapmode(int *retavail, int *retfree)
1610 {
1611 	int n;
1612 	struct kvm_swap swapary[1];
1613 	static int pagesize = 0;
1614 	static unsigned long swap_maxpages = 0;
1615 
1616 	*retavail = 0;
1617 	*retfree = 0;
1618 
1619 #define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1620 
1621 	n = kvm_getswapinfo(kd, swapary, 1, 0);
1622 	if (n < 0 || swapary[0].ksw_total == 0)
1623 		return (0);
1624 
1625 	if (pagesize == 0)
1626 		pagesize = getpagesize();
1627 	if (swap_maxpages == 0)
1628 		GETSYSCTL("vm.swap_maxpages", swap_maxpages);
1629 
1630 	/* ksw_total contains the total size of swap all devices which may
1631 	   exceed the maximum swap size allocatable in the system */
1632 	if ( swapary[0].ksw_total > swap_maxpages )
1633 		swapary[0].ksw_total = swap_maxpages;
1634 
1635 	*retavail = CONVERT(swapary[0].ksw_total);
1636 	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1637 
1638 	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1639 	return (n);
1640 }
1641