machine.c revision 170774
1/*
2 * top - a top users display for Unix
3 *
4 * SYNOPSIS:  For FreeBSD-2.x and later
5 *
6 * DESCRIPTION:
7 * Originally written for BSD4.4 system by Christos Zoulas.
8 * Ported to FreeBSD 2.x by Steven Wallace && Wolfram Schneider
9 * Order support hacked in from top-3.5beta6/machine/m_aix41.c
10 *   by Monte Mitzelfelt (for latest top see http://www.groupsys.com/topinfo/)
11 *
12 * This is the machine-dependent module for FreeBSD 2.2
13 * Works for:
14 *	FreeBSD 2.2.x, 3.x, 4.x, and probably FreeBSD 2.1.x
15 *
16 * LIBS: -lkvm
17 *
18 * AUTHOR:  Christos Zoulas <christos@ee.cornell.edu>
19 *          Steven Wallace  <swallace@freebsd.org>
20 *          Wolfram Schneider <wosch@FreeBSD.org>
21 *          Thomas Moestl <tmoestl@gmx.net>
22 *
23 * $FreeBSD: head/usr.bin/top/machine.c 170774 2007-06-15 12:03:07Z bde $
24 */
25
26#include <sys/param.h>
27#include <sys/errno.h>
28#include <sys/file.h>
29#include <sys/proc.h>
30#include <sys/resource.h>
31#include <sys/rtprio.h>
32#include <sys/signal.h>
33#include <sys/sysctl.h>
34#include <sys/time.h>
35#include <sys/user.h>
36#include <sys/vmmeter.h>
37
38#include <kvm.h>
39#include <math.h>
40#include <nlist.h>
41#include <paths.h>
42#include <pwd.h>
43#include <stdio.h>
44#include <stdlib.h>
45#include <string.h>
46#include <strings.h>
47#include <unistd.h>
48#include <vis.h>
49
50#include "top.h"
51#include "machine.h"
52#include "screen.h"
53#include "utils.h"
54
55#define GETSYSCTL(name, var) getsysctl(name, &(var), sizeof(var))
56#define	SMPUNAMELEN	13
57#define	UPUNAMELEN	15
58
59extern struct process_select ps;
60extern char* printable(char *);
61static int smpmode;
62enum displaymodes displaymode;
63static int namelength = 8;
64static int cmdlengthdelta;
65
66/* Prototypes for top internals */
67void quit(int);
68
69/* get_process_info passes back a handle.  This is what it looks like: */
70
71struct handle {
72	struct kinfo_proc **next_proc;	/* points to next valid proc pointer */
73	int remaining;			/* number of pointers remaining */
74};
75
76/* declarations for load_avg */
77#include "loadavg.h"
78
79/* define what weighted cpu is.  */
80#define weighted_cpu(pct, pp) ((pp)->ki_swtime == 0 ? 0.0 : \
81			 ((pct) / (1.0 - exp((pp)->ki_swtime * logcpu))))
82
83/* what we consider to be process size: */
84#define PROCSIZE(pp) ((pp)->ki_size / 1024)
85
86#define RU(pp)	(&(pp)->ki_rusage)
87#define RUTOT(pp) \
88	(RU(pp)->ru_inblock + RU(pp)->ru_oublock + RU(pp)->ru_majflt)
89
90
91/* definitions for indices in the nlist array */
92
93/*
94 *  These definitions control the format of the per-process area
95 */
96
97static char io_header[] =
98    "  PID%s %-*.*s   VCSW  IVCSW   READ  WRITE  FAULT  TOTAL PERCENT COMMAND";
99
100#define io_Proc_format \
101    "%5d%s %-*.*s %6ld %6ld %6ld %6ld %6ld %6ld %6.2f%% %.*s"
102
103static char smp_header_thr[] =
104    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE  C   TIME %6s COMMAND";
105static char smp_header[] =
106    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE  C   TIME %6s COMMAND";
107
108#define smp_Proc_format \
109    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s %1x%7s %5.2f%% %.*s"
110
111static char up_header_thr[] =
112    "  PID%s %-*.*s  THR PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
113static char up_header[] =
114    "  PID%s %-*.*s "   "PRI NICE   SIZE    RES STATE    TIME %6s COMMAND";
115
116#define up_Proc_format \
117    "%5d%s %-*.*s %s%3d %4s%7s %6s %-6.6s%.0d%7s %5.2f%% %.*s"
118
119
120/* process state names for the "STATE" column of the display */
121/* the extra nulls in the string "run" are for adding a slash and
122   the processor number when needed */
123
124char *state_abbrev[] = {
125	"", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB", "WAIT", "LOCK"
126};
127
128
129static kvm_t *kd;
130
131/* values that we stash away in _init and use in later routines */
132
133static double logcpu;
134
135/* these are retrieved from the kernel in _init */
136
137static load_avg  ccpu;
138
139/* these are used in the get_ functions */
140
141static int lastpid;
142
143/* these are for calculating cpu state percentages */
144
145static long cp_time[CPUSTATES];
146static long cp_old[CPUSTATES];
147static long cp_diff[CPUSTATES];
148
149/* these are for detailing the process states */
150
151int process_states[8];
152char *procstatenames[] = {
153	"", " starting, ", " running, ", " sleeping, ", " stopped, ",
154	" zombie, ", " waiting, ", " lock, ",
155	NULL
156};
157
158/* these are for detailing the cpu states */
159
160int cpu_states[CPUSTATES];
161char *cpustatenames[] = {
162	"user", "nice", "system", "interrupt", "idle", NULL
163};
164
165/* these are for detailing the memory statistics */
166
167int memory_stats[7];
168char *memorynames[] = {
169	"K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ",
170	"K Free", NULL
171};
172
173int swap_stats[7];
174char *swapnames[] = {
175	"K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
176	NULL
177};
178
179
180/* these are for keeping track of the proc array */
181
182static int nproc;
183static int onproc = -1;
184static int pref_len;
185static struct kinfo_proc *pbase;
186static struct kinfo_proc **pref;
187static struct kinfo_proc *previous_procs;
188static struct kinfo_proc **previous_pref;
189static int previous_proc_count = 0;
190static int previous_proc_count_max = 0;
191
192/* total number of io operations */
193static long total_inblock;
194static long total_oublock;
195static long total_majflt;
196
197/* these are for getting the memory statistics */
198
199static int pageshift;		/* log base 2 of the pagesize */
200
201/* define pagetok in terms of pageshift */
202
203#define pagetok(size) ((size) << pageshift)
204
205/* useful externals */
206long percentages();
207
208#ifdef ORDER
209/*
210 * Sorting orders.  The first element is the default.
211 */
212char *ordernames[] = {
213	"cpu", "size", "res", "time", "pri", "threads",
214	"total", "read", "write", "fault", "vcsw", "ivcsw",
215	"jid", NULL
216};
217#endif
218
219static int compare_jid(const void *a, const void *b);
220static int compare_pid(const void *a, const void *b);
221static const char *format_nice(const struct kinfo_proc *pp);
222static void getsysctl(const char *name, void *ptr, size_t len);
223static int swapmode(int *retavail, int *retfree);
224
225int
226machine_init(struct statics *statics)
227{
228	int pagesize;
229	size_t modelen;
230	struct passwd *pw;
231
232	modelen = sizeof(smpmode);
233	if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen,
234	    NULL, 0) != 0 &&
235	    sysctlbyname("kern.smp.active", &smpmode, &modelen,
236	    NULL, 0) != 0) ||
237	    modelen != sizeof(smpmode))
238		smpmode = 0;
239
240	while ((pw = getpwent()) != NULL) {
241		if (strlen(pw->pw_name) > namelength)
242			namelength = strlen(pw->pw_name);
243	}
244	if (smpmode && namelength > SMPUNAMELEN)
245		namelength = SMPUNAMELEN;
246	else if (namelength > UPUNAMELEN)
247		namelength = UPUNAMELEN;
248
249	kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, "kvm_open");
250	if (kd == NULL)
251		return (-1);
252
253	GETSYSCTL("kern.ccpu", ccpu);
254
255	/* this is used in calculating WCPU -- calculate it ahead of time */
256	logcpu = log(loaddouble(ccpu));
257
258	pbase = NULL;
259	pref = NULL;
260	nproc = 0;
261	onproc = -1;
262
263	/* get the page size and calculate pageshift from it */
264	pagesize = getpagesize();
265	pageshift = 0;
266	while (pagesize > 1) {
267		pageshift++;
268		pagesize >>= 1;
269	}
270
271	/* we only need the amount of log(2)1024 for our conversion */
272	pageshift -= LOG1024;
273
274	/* fill in the statics information */
275	statics->procstate_names = procstatenames;
276	statics->cpustate_names = cpustatenames;
277	statics->memory_names = memorynames;
278	statics->swap_names = swapnames;
279#ifdef ORDER
280	statics->order_names = ordernames;
281#endif
282
283	/* all done! */
284	return (0);
285}
286
287char *
288format_header(char *uname_field)
289{
290	static char Header[128];
291	const char *prehead;
292
293	switch (displaymode) {
294	case DISP_CPU:
295		/*
296		 * The logic of picking the right header format seems reverse
297		 * here because we only want to display a THR column when
298		 * "thread mode" is off (and threads are not listed as
299		 * separate lines).
300		 */
301		prehead = smpmode ?
302		    (ps.thread ? smp_header : smp_header_thr) :
303		    (ps.thread ? up_header : up_header_thr);
304		snprintf(Header, sizeof(Header), prehead,
305		    ps.jail ? " JID" : "",
306		    namelength, namelength, uname_field,
307		    ps.wcpu ? "WCPU" : "CPU");
308		break;
309	case DISP_IO:
310		prehead = io_header;
311		snprintf(Header, sizeof(Header), prehead,
312		    ps.jail ? " JID" : "",
313		    namelength, namelength, uname_field);
314		break;
315	}
316	cmdlengthdelta = strlen(Header) - 7;
317	return (Header);
318}
319
320static int swappgsin = -1;
321static int swappgsout = -1;
322extern struct timeval timeout;
323
324void
325get_system_info(struct system_info *si)
326{
327	long total;
328	struct loadavg sysload;
329	int mib[2];
330	struct timeval boottime;
331	size_t bt_size;
332	int i;
333
334	/* get the cp_time array */
335	GETSYSCTL("kern.cp_time", cp_time);
336	GETSYSCTL("vm.loadavg", sysload);
337	GETSYSCTL("kern.lastpid", lastpid);
338
339	/* convert load averages to doubles */
340	for (i = 0; i < 3; i++)
341		si->load_avg[i] = (double)sysload.ldavg[i] / sysload.fscale;
342
343	/* convert cp_time counts to percentages */
344	total = percentages(CPUSTATES, cpu_states, cp_time, cp_old, cp_diff);
345
346	/* sum memory & swap statistics */
347	{
348		static unsigned int swap_delay = 0;
349		static int swapavail = 0;
350		static int swapfree = 0;
351		static int bufspace = 0;
352		static int nspgsin, nspgsout;
353
354		GETSYSCTL("vfs.bufspace", bufspace);
355		GETSYSCTL("vm.stats.vm.v_active_count", memory_stats[0]);
356		GETSYSCTL("vm.stats.vm.v_inactive_count", memory_stats[1]);
357		GETSYSCTL("vm.stats.vm.v_wire_count", memory_stats[2]);
358		GETSYSCTL("vm.stats.vm.v_cache_count", memory_stats[3]);
359		GETSYSCTL("vm.stats.vm.v_free_count", memory_stats[5]);
360		GETSYSCTL("vm.stats.vm.v_swappgsin", nspgsin);
361		GETSYSCTL("vm.stats.vm.v_swappgsout", nspgsout);
362		/* convert memory stats to Kbytes */
363		memory_stats[0] = pagetok(memory_stats[0]);
364		memory_stats[1] = pagetok(memory_stats[1]);
365		memory_stats[2] = pagetok(memory_stats[2]);
366		memory_stats[3] = pagetok(memory_stats[3]);
367		memory_stats[4] = bufspace / 1024;
368		memory_stats[5] = pagetok(memory_stats[5]);
369		memory_stats[6] = -1;
370
371		/* first interval */
372		if (swappgsin < 0) {
373			swap_stats[4] = 0;
374			swap_stats[5] = 0;
375		}
376
377		/* compute differences between old and new swap statistic */
378		else {
379			swap_stats[4] = pagetok(((nspgsin - swappgsin)));
380			swap_stats[5] = pagetok(((nspgsout - swappgsout)));
381		}
382
383		swappgsin = nspgsin;
384		swappgsout = nspgsout;
385
386		/* call CPU heavy swapmode() only for changes */
387		if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
388			swap_stats[3] = swapmode(&swapavail, &swapfree);
389			swap_stats[0] = swapavail;
390			swap_stats[1] = swapavail - swapfree;
391			swap_stats[2] = swapfree;
392		}
393		swap_delay = 1;
394		swap_stats[6] = -1;
395	}
396
397	/* set arrays and strings */
398	si->cpustates = cpu_states;
399	si->memory = memory_stats;
400	si->swap = swap_stats;
401
402
403	if (lastpid > 0) {
404		si->last_pid = lastpid;
405	} else {
406		si->last_pid = -1;
407	}
408
409	/*
410	 * Print how long system has been up.
411	 * (Found by looking getting "boottime" from the kernel)
412	 */
413	mib[0] = CTL_KERN;
414	mib[1] = KERN_BOOTTIME;
415	bt_size = sizeof(boottime);
416	if (sysctl(mib, 2, &boottime, &bt_size, NULL, 0) != -1 &&
417	    boottime.tv_sec != 0) {
418		si->boottime = boottime;
419	} else {
420		si->boottime.tv_sec = -1;
421	}
422}
423
424#define NOPROC	((void *)-1)
425
426/*
427 * We need to compare data from the old process entry with the new
428 * process entry.
429 * To facilitate doing this quickly we stash a pointer in the kinfo_proc
430 * structure to cache the mapping.  We also use a negative cache pointer
431 * of NOPROC to avoid duplicate lookups.
432 * XXX: this could be done when the actual processes are fetched, we do
433 * it here out of laziness.
434 */
435const struct kinfo_proc *
436get_old_proc(struct kinfo_proc *pp)
437{
438	struct kinfo_proc **oldpp, *oldp;
439
440	/*
441	 * If this is the first fetch of the kinfo_procs then we don't have
442	 * any previous entries.
443	 */
444	if (previous_proc_count == 0)
445		return (NULL);
446	/* negative cache? */
447	if (pp->ki_udata == NOPROC)
448		return (NULL);
449	/* cached? */
450	if (pp->ki_udata != NULL)
451		return (pp->ki_udata);
452	/*
453	 * Not cached,
454	 * 1) look up based on pid.
455	 * 2) compare process start.
456	 * If we fail here, then setup a negative cache entry, otherwise
457	 * cache it.
458	 */
459	oldpp = bsearch(&pp, previous_pref, previous_proc_count,
460	    sizeof(*previous_pref), compare_pid);
461	if (oldpp == NULL) {
462		pp->ki_udata = NOPROC;
463		return (NULL);
464	}
465	oldp = *oldpp;
466	if (bcmp(&oldp->ki_start, &pp->ki_start, sizeof(pp->ki_start)) != 0) {
467		pp->ki_udata = NOPROC;
468		return (NULL);
469	}
470	pp->ki_udata = oldp;
471	return (oldp);
472}
473
474/*
475 * Return the total amount of IO done in blocks in/out and faults.
476 * store the values individually in the pointers passed in.
477 */
478long
479get_io_stats(struct kinfo_proc *pp, long *inp, long *oup, long *flp,
480    long *vcsw, long *ivcsw)
481{
482	const struct kinfo_proc *oldp;
483	static struct kinfo_proc dummy;
484	long ret;
485
486	oldp = get_old_proc(pp);
487	if (oldp == NULL) {
488		bzero(&dummy, sizeof(dummy));
489		oldp = &dummy;
490	}
491	*inp = RU(pp)->ru_inblock - RU(oldp)->ru_inblock;
492	*oup = RU(pp)->ru_oublock - RU(oldp)->ru_oublock;
493	*flp = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
494	*vcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
495	*ivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
496	ret =
497	    (RU(pp)->ru_inblock - RU(oldp)->ru_inblock) +
498	    (RU(pp)->ru_oublock - RU(oldp)->ru_oublock) +
499	    (RU(pp)->ru_majflt - RU(oldp)->ru_majflt);
500	return (ret);
501}
502
503/*
504 * Return the total number of block in/out and faults by a process.
505 */
506long
507get_io_total(struct kinfo_proc *pp)
508{
509	long dummy;
510
511	return (get_io_stats(pp, &dummy, &dummy, &dummy, &dummy, &dummy));
512}
513
514static struct handle handle;
515
516caddr_t
517get_process_info(struct system_info *si, struct process_select *sel,
518    int (*compare)(const void *, const void *))
519{
520	int i;
521	int total_procs;
522	long p_io;
523	long p_inblock, p_oublock, p_majflt, p_vcsw, p_ivcsw;
524	int active_procs;
525	struct kinfo_proc **prefp;
526	struct kinfo_proc *pp;
527	struct kinfo_proc *prev_pp = NULL;
528
529	/* these are copied out of sel for speed */
530	int show_idle;
531	int show_self;
532	int show_system;
533	int show_uid;
534	int show_command;
535
536	/*
537	 * Save the previous process info.
538	 */
539	if (previous_proc_count_max < nproc) {
540		free(previous_procs);
541		previous_procs = malloc(nproc * sizeof(*previous_procs));
542		free(previous_pref);
543		previous_pref = malloc(nproc * sizeof(*previous_pref));
544		if (previous_procs == NULL || previous_pref == NULL) {
545			(void) fprintf(stderr, "top: Out of memory.\n");
546			quit(23);
547		}
548		previous_proc_count_max = nproc;
549	}
550	if (nproc) {
551		for (i = 0; i < nproc; i++)
552			previous_pref[i] = &previous_procs[i];
553		bcopy(pbase, previous_procs, nproc * sizeof(*previous_procs));
554		qsort(previous_pref, nproc, sizeof(*previous_pref),
555		    compare_pid);
556	}
557	previous_proc_count = nproc;
558
559	pbase = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nproc);
560	if (nproc > onproc)
561		pref = realloc(pref, sizeof(*pref) * (onproc = nproc));
562	if (pref == NULL || pbase == NULL) {
563		(void) fprintf(stderr, "top: Out of memory.\n");
564		quit(23);
565	}
566	/* get a pointer to the states summary array */
567	si->procstates = process_states;
568
569	/* set up flags which define what we are going to select */
570	show_idle = sel->idle;
571	show_self = sel->self == -1;
572	show_system = sel->system;
573	show_uid = sel->uid != -1;
574	show_command = sel->command != NULL;
575
576	/* count up process states and get pointers to interesting procs */
577	total_procs = 0;
578	active_procs = 0;
579	total_inblock = 0;
580	total_oublock = 0;
581	total_majflt = 0;
582	memset((char *)process_states, 0, sizeof(process_states));
583	prefp = pref;
584	for (pp = pbase, i = 0; i < nproc; pp++, i++) {
585
586		if (pp->ki_stat == 0)
587			/* not in use */
588			continue;
589
590		if (!show_self && pp->ki_pid == sel->self)
591			/* skip self */
592			continue;
593
594		if (!show_system && (pp->ki_flag & P_SYSTEM))
595			/* skip system process */
596			continue;
597
598		p_io = get_io_stats(pp, &p_inblock, &p_oublock, &p_majflt,
599		    &p_vcsw, &p_ivcsw);
600		total_inblock += p_inblock;
601		total_oublock += p_oublock;
602		total_majflt += p_majflt;
603		total_procs++;
604		process_states[pp->ki_stat]++;
605
606		if (pp->ki_stat == SZOMB)
607			/* skip zombies */
608			continue;
609
610		if (displaymode == DISP_CPU && !show_idle &&
611		    (pp->ki_pctcpu == 0 ||
612		     pp->ki_stat == SSTOP || pp->ki_stat == SIDL))
613			/* skip idle or non-running processes */
614			continue;
615
616		if (displaymode == DISP_IO && !show_idle && p_io == 0)
617			/* skip processes that aren't doing I/O */
618			continue;
619
620		if (show_uid && pp->ki_ruid != (uid_t)sel->uid)
621			/* skip proc. that don't belong to the selected UID */
622			continue;
623
624		/*
625		 * When not showing threads, take the first thread
626		 * for output and add the fields that we can from
627		 * the rest of the process's threads rather than
628		 * using the system's mostly-broken KERN_PROC_PROC.
629		 */
630		if (sel->thread || prev_pp == NULL ||
631		    prev_pp->ki_pid != pp->ki_pid) {
632			*prefp++ = pp;
633			active_procs++;
634			prev_pp = pp;
635		} else {
636			prev_pp->ki_pctcpu += pp->ki_pctcpu;
637		}
638	}
639
640	/* if requested, sort the "interesting" processes */
641	if (compare != NULL)
642		qsort(pref, active_procs, sizeof(*pref), compare);
643
644	/* remember active and total counts */
645	si->p_total = total_procs;
646	si->p_active = pref_len = active_procs;
647
648	/* pass back a handle */
649	handle.next_proc = pref;
650	handle.remaining = active_procs;
651	return ((caddr_t)&handle);
652}
653
654static char fmt[128];	/* static area where result is built */
655
656char *
657format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
658{
659	struct kinfo_proc *pp;
660	const struct kinfo_proc *oldp;
661	long cputime;
662	double pct;
663	struct handle *hp;
664	char status[16];
665	int state;
666	struct rusage ru, *rup;
667	long p_tot, s_tot;
668	char *proc_fmt, thr_buf[6], jid_buf[6];
669	char *cmdbuf = NULL;
670	char **args;
671
672	/* find and remember the next proc structure */
673	hp = (struct handle *)handle;
674	pp = *(hp->next_proc++);
675	hp->remaining--;
676
677	/* get the process's command name */
678	if ((pp->ki_sflag & PS_INMEM) == 0) {
679		/*
680		 * Print swapped processes as <pname>
681		 */
682		size_t len;
683
684		len = strlen(pp->ki_comm);
685		if (len > sizeof(pp->ki_comm) - 3)
686			len = sizeof(pp->ki_comm) - 3;
687		memmove(pp->ki_comm + 1, pp->ki_comm, len);
688		pp->ki_comm[0] = '<';
689		pp->ki_comm[len + 1] = '>';
690		pp->ki_comm[len + 2] = '\0';
691	}
692
693	/*
694	 * Convert the process's runtime from microseconds to seconds.  This
695	 * time includes the interrupt time although that is not wanted here.
696	 * ps(1) is similarly sloppy.
697	 */
698	cputime = (pp->ki_runtime + 500000) / 1000000;
699
700	/* calculate the base for cpu percentages */
701	pct = pctdouble(pp->ki_pctcpu);
702
703	/* generate "STATE" field */
704	switch (state = pp->ki_stat) {
705	case SRUN:
706		if (smpmode && pp->ki_oncpu != 0xff)
707			sprintf(status, "CPU%d", pp->ki_oncpu);
708		else
709			strcpy(status, "RUN");
710		break;
711	case SLOCK:
712		if (pp->ki_kiflag & KI_LOCKBLOCK) {
713			sprintf(status, "*%.6s", pp->ki_lockname);
714			break;
715		}
716		/* fall through */
717	case SSLEEP:
718		if (pp->ki_wmesg != NULL) {
719			sprintf(status, "%.6s", pp->ki_wmesg);
720			break;
721		}
722		/* FALLTHROUGH */
723	default:
724
725		if (state >= 0 &&
726		    state < sizeof(state_abbrev) / sizeof(*state_abbrev))
727			sprintf(status, "%.6s", state_abbrev[state]);
728		else
729			sprintf(status, "?%5d", state);
730		break;
731	}
732
733	cmdbuf = (char *)malloc(cmdlengthdelta + 1);
734	if (cmdbuf == NULL) {
735		warn("malloc(%d)", cmdlengthdelta + 1);
736		return NULL;
737	}
738
739	if (!(flags & FMT_SHOWARGS)) {
740		snprintf(cmdbuf, cmdlengthdelta, "%s", pp->ki_comm);
741	}
742	else if (pp->ki_args == NULL ||
743	    (args = kvm_getargv(kd, pp, cmdlengthdelta)) == NULL || !(*args))
744		snprintf(cmdbuf, cmdlengthdelta, "[%s]", pp->ki_comm);
745	else {
746		char *src, *dst, *argbuf;
747		char *cmd;
748		size_t argbuflen;
749		size_t len;
750
751		argbuflen = cmdlengthdelta * 4;
752		argbuf = (char *)malloc(argbuflen + 1);
753		if (argbuf == NULL) {
754			warn("malloc(%d)", argbuflen + 1);
755			free(cmdbuf);
756			return NULL;
757		}
758
759		dst = argbuf;
760
761		/* Extract cmd name from argv */
762		cmd = strrchr(*args, '/');
763		if (cmd == NULL)
764			cmd = *args;
765		else
766			cmd++;
767
768		for (; (src = *args++) != NULL; ) {
769			if (*src == '\0')
770				continue;
771			len = (argbuflen - (dst - argbuf) - 1) / 4;
772			strvisx(dst, src, strlen(src) < len ? strlen(src) : len,
773			    VIS_NL | VIS_CSTYLE);
774			while (*dst != '\0')
775				dst++;
776			if ((argbuflen - (dst - argbuf) - 1) / 4 > 0)
777				*dst++ = ' '; /* add delimiting space */
778		}
779		if (dst != argbuf && dst[-1] == ' ')
780			dst--;
781		*dst = '\0';
782
783		if (strcmp(cmd, pp->ki_comm) != 0 )
784			snprintf(cmdbuf, cmdlengthdelta, "%s (%s)",argbuf, \
785				 pp->ki_comm);
786		else
787			strlcpy(cmdbuf, argbuf, cmdlengthdelta);
788
789		free(argbuf);
790	}
791
792	if (ps.jail == 0)
793		jid_buf[0] = '\0';
794	else
795		snprintf(jid_buf, sizeof(jid_buf), " %*d",
796		    sizeof(jid_buf) - 3, pp->ki_jid);
797
798	if (displaymode == DISP_IO) {
799		oldp = get_old_proc(pp);
800		if (oldp != NULL) {
801			ru.ru_inblock = RU(pp)->ru_inblock -
802			    RU(oldp)->ru_inblock;
803			ru.ru_oublock = RU(pp)->ru_oublock -
804			    RU(oldp)->ru_oublock;
805			ru.ru_majflt = RU(pp)->ru_majflt - RU(oldp)->ru_majflt;
806			ru.ru_nvcsw = RU(pp)->ru_nvcsw - RU(oldp)->ru_nvcsw;
807			ru.ru_nivcsw = RU(pp)->ru_nivcsw - RU(oldp)->ru_nivcsw;
808			rup = &ru;
809		} else {
810			rup = RU(pp);
811		}
812		p_tot = rup->ru_inblock + rup->ru_oublock + rup->ru_majflt;
813		s_tot = total_inblock + total_oublock + total_majflt;
814
815		sprintf(fmt, io_Proc_format,
816		    pp->ki_pid,
817		    jid_buf,
818		    namelength, namelength, (*get_userid)(pp->ki_ruid),
819		    rup->ru_nvcsw,
820		    rup->ru_nivcsw,
821		    rup->ru_inblock,
822		    rup->ru_oublock,
823		    rup->ru_majflt,
824		    p_tot,
825		    s_tot == 0 ? 0.0 : (p_tot * 100.0 / s_tot),
826		    screen_width > cmdlengthdelta ?
827		    screen_width - cmdlengthdelta : 0,
828		    printable(cmdbuf));
829
830		free(cmdbuf);
831
832		return (fmt);
833	}
834
835	/* format this entry */
836	proc_fmt = smpmode ? smp_Proc_format : up_Proc_format;
837	if (ps.thread != 0)
838		thr_buf[0] = '\0';
839	else
840		snprintf(thr_buf, sizeof(thr_buf), "%*d ",
841		    sizeof(thr_buf) - 2, pp->ki_numthreads);
842
843	sprintf(fmt, proc_fmt,
844	    pp->ki_pid,
845	    jid_buf,
846	    namelength, namelength, (*get_userid)(pp->ki_ruid),
847	    thr_buf,
848	    pp->ki_pri.pri_level - PZERO,
849	    format_nice(pp),
850	    format_k2(PROCSIZE(pp)),
851	    format_k2(pagetok(pp->ki_rssize)),
852	    status,
853	    smpmode ? pp->ki_lastcpu : 0,
854	    format_time(cputime),
855	    ps.wcpu ? 100.0 * weighted_cpu(pct, pp) : 100.0 * pct,
856	    screen_width > cmdlengthdelta ? screen_width - cmdlengthdelta : 0,
857	    printable(cmdbuf));
858
859	free(cmdbuf);
860
861	/* return the result */
862	return (fmt);
863}
864
865static void
866getsysctl(const char *name, void *ptr, size_t len)
867{
868	size_t nlen = len;
869
870	if (sysctlbyname(name, ptr, &nlen, NULL, 0) == -1) {
871		fprintf(stderr, "top: sysctl(%s...) failed: %s\n", name,
872		    strerror(errno));
873		quit(23);
874	}
875	if (nlen != len) {
876		fprintf(stderr, "top: sysctl(%s...) expected %lu, got %lu\n",
877		    name, (unsigned long)len, (unsigned long)nlen);
878		quit(23);
879	}
880}
881
882static const char *
883format_nice(const struct kinfo_proc *pp)
884{
885	const char *fifo, *kthread;
886	int rtpri;
887	static char nicebuf[4 + 1];
888
889	fifo = PRI_NEED_RR(pp->ki_pri.pri_class) ? "" : "F";
890	kthread = (pp->ki_flag & P_KTHREAD) ? "k" : "";
891	switch (PRI_BASE(pp->ki_pri.pri_class)) {
892	case PRI_ITHD:
893		return ("-");
894	case PRI_REALTIME:
895		/*
896		 * XXX: the kernel doesn't tell us the original rtprio and
897		 * doesn't really know what it was, so to recover it we
898		 * must be more chummy with the implementation than the
899		 * implementation is with itself.  pri_user gives a
900		 * constant "base" priority, but is only initialized
901		 * properly for user threads.  pri_native gives what the
902		 * kernel calls the "base" priority, but it isn't constant
903		 * since it is changed by priority propagation.  pri_native
904		 * also isn't properly initialized for all threads, but it
905		 * is properly initialized for kernel realtime and idletime
906		 * threads.  Thus we use pri_user for the base priority of
907		 * user threads (it is always correct) and pri_native for
908		 * the base priority of kernel realtime and idletime threads
909		 * (there is nothing better, and it is usually correct).
910		 *
911		 * The field width and thus the buffer are too small for
912		 * values like "kr31F", but such values shouldn't occur,
913		 * and if they do then the tailing "F" is not displayed.
914		 */
915		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
916		    pp->ki_pri.pri_user) - PRI_MIN_REALTIME;
917		snprintf(nicebuf, sizeof(nicebuf), "%sr%d%s",
918		    kthread, rtpri, fifo);
919		break;
920	case PRI_TIMESHARE:
921		if (pp->ki_flag & P_KTHREAD)
922			return ("-");
923		snprintf(nicebuf, sizeof(nicebuf), "%d", pp->ki_nice - NZERO);
924		break;
925	case PRI_IDLE:
926		/* XXX: as above. */
927		rtpri = ((pp->ki_flag & P_KTHREAD) ? pp->ki_pri.pri_native :
928		    pp->ki_pri.pri_user) - PRI_MIN_IDLE;
929		snprintf(nicebuf, sizeof(nicebuf), "%si%d%s",
930		    kthread, rtpri, fifo);
931		break;
932	default:
933		return ("?");
934	}
935	return (nicebuf);
936}
937
938/* comparison routines for qsort */
939
940static int
941compare_pid(const void *p1, const void *p2)
942{
943	const struct kinfo_proc * const *pp1 = p1;
944	const struct kinfo_proc * const *pp2 = p2;
945
946	if ((*pp2)->ki_pid < 0 || (*pp1)->ki_pid < 0)
947		abort();
948
949	return ((*pp1)->ki_pid - (*pp2)->ki_pid);
950}
951
952/*
953 *  proc_compare - comparison function for "qsort"
954 *	Compares the resource consumption of two processes using five
955 *	distinct keys.  The keys (in descending order of importance) are:
956 *	percent cpu, cpu ticks, state, resident set size, total virtual
957 *	memory usage.  The process states are ordered as follows (from least
958 *	to most important):  WAIT, zombie, sleep, stop, start, run.  The
959 *	array declaration below maps a process state index into a number
960 *	that reflects this ordering.
961 */
962
963static int sorted_state[] = {
964	0,	/* not used		*/
965	3,	/* sleep		*/
966	1,	/* ABANDONED (WAIT)	*/
967	6,	/* run			*/
968	5,	/* start		*/
969	2,	/* zombie		*/
970	4	/* stop			*/
971};
972
973
974#define ORDERKEY_PCTCPU(a, b) do { \
975	long diff; \
976	if (ps.wcpu) \
977		diff = floor(1.0E6 * weighted_cpu(pctdouble((b)->ki_pctcpu), \
978		    (b))) - \
979		    floor(1.0E6 * weighted_cpu(pctdouble((a)->ki_pctcpu), \
980		    (a))); \
981	else \
982		diff = (long)(b)->ki_pctcpu - (long)(a)->ki_pctcpu; \
983	if (diff != 0) \
984		return (diff > 0 ? 1 : -1); \
985} while (0)
986
987#define ORDERKEY_CPTICKS(a, b) do { \
988	int64_t diff = (int64_t)(b)->ki_runtime - (int64_t)(a)->ki_runtime; \
989	if (diff != 0) \
990		return (diff > 0 ? 1 : -1); \
991} while (0)
992
993#define ORDERKEY_STATE(a, b) do { \
994	int diff = sorted_state[(b)->ki_stat] - sorted_state[(a)->ki_stat]; \
995	if (diff != 0) \
996		return (diff > 0 ? 1 : -1); \
997} while (0)
998
999#define ORDERKEY_PRIO(a, b) do { \
1000	int diff = (int)(b)->ki_pri.pri_level - (int)(a)->ki_pri.pri_level; \
1001	if (diff != 0) \
1002		return (diff > 0 ? 1 : -1); \
1003} while (0)
1004
1005#define	ORDERKEY_THREADS(a, b) do { \
1006	int diff = (int)(b)->ki_numthreads - (int)(a)->ki_numthreads; \
1007	if (diff != 0) \
1008		return (diff > 0 ? 1 : -1); \
1009} while (0)
1010
1011#define ORDERKEY_RSSIZE(a, b) do { \
1012	long diff = (long)(b)->ki_rssize - (long)(a)->ki_rssize; \
1013	if (diff != 0) \
1014		return (diff > 0 ? 1 : -1); \
1015} while (0)
1016
1017#define ORDERKEY_MEM(a, b) do { \
1018	long diff = (long)PROCSIZE((b)) - (long)PROCSIZE((a)); \
1019	if (diff != 0) \
1020		return (diff > 0 ? 1 : -1); \
1021} while (0)
1022
1023#define ORDERKEY_JID(a, b) do { \
1024	int diff = (int)(b)->ki_jid - (int)(a)->ki_jid; \
1025	if (diff != 0) \
1026		return (diff > 0 ? 1 : -1); \
1027} while (0)
1028
1029/* compare_cpu - the comparison function for sorting by cpu percentage */
1030
1031int
1032#ifdef ORDER
1033compare_cpu(void *arg1, void *arg2)
1034#else
1035proc_compare(void *arg1, void *arg2)
1036#endif
1037{
1038	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1039	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1040
1041	ORDERKEY_PCTCPU(p1, p2);
1042	ORDERKEY_CPTICKS(p1, p2);
1043	ORDERKEY_STATE(p1, p2);
1044	ORDERKEY_PRIO(p1, p2);
1045	ORDERKEY_RSSIZE(p1, p2);
1046	ORDERKEY_MEM(p1, p2);
1047
1048	return (0);
1049}
1050
1051#ifdef ORDER
1052/* "cpu" compare routines */
1053int compare_size(), compare_res(), compare_time(), compare_prio(),
1054    compare_threads();
1055
1056/*
1057 * "io" compare routines.  Context switches aren't i/o, but are displayed
1058 * on the "io" display.
1059 */
1060int compare_iototal(), compare_ioread(), compare_iowrite(), compare_iofault(),
1061    compare_vcsw(), compare_ivcsw();
1062
1063int (*compares[])() = {
1064	compare_cpu,
1065	compare_size,
1066	compare_res,
1067	compare_time,
1068	compare_prio,
1069	compare_threads,
1070	compare_iototal,
1071	compare_ioread,
1072	compare_iowrite,
1073	compare_iofault,
1074	compare_vcsw,
1075	compare_ivcsw,
1076	compare_jid,
1077	NULL
1078};
1079
1080/* compare_size - the comparison function for sorting by total memory usage */
1081
1082int
1083compare_size(void *arg1, void *arg2)
1084{
1085	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1086	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1087
1088	ORDERKEY_MEM(p1, p2);
1089	ORDERKEY_RSSIZE(p1, p2);
1090	ORDERKEY_PCTCPU(p1, p2);
1091	ORDERKEY_CPTICKS(p1, p2);
1092	ORDERKEY_STATE(p1, p2);
1093	ORDERKEY_PRIO(p1, p2);
1094
1095	return (0);
1096}
1097
1098/* compare_res - the comparison function for sorting by resident set size */
1099
1100int
1101compare_res(void *arg1, void *arg2)
1102{
1103	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1104	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1105
1106	ORDERKEY_RSSIZE(p1, p2);
1107	ORDERKEY_MEM(p1, p2);
1108	ORDERKEY_PCTCPU(p1, p2);
1109	ORDERKEY_CPTICKS(p1, p2);
1110	ORDERKEY_STATE(p1, p2);
1111	ORDERKEY_PRIO(p1, p2);
1112
1113	return (0);
1114}
1115
1116/* compare_time - the comparison function for sorting by total cpu time */
1117
1118int
1119compare_time(void *arg1, void *arg2)
1120{
1121	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1122	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1123
1124	ORDERKEY_CPTICKS(p1, p2);
1125	ORDERKEY_PCTCPU(p1, p2);
1126	ORDERKEY_STATE(p1, p2);
1127	ORDERKEY_PRIO(p1, p2);
1128	ORDERKEY_RSSIZE(p1, p2);
1129	ORDERKEY_MEM(p1, p2);
1130
1131	return (0);
1132}
1133
1134/* compare_prio - the comparison function for sorting by priority */
1135
1136int
1137compare_prio(void *arg1, void *arg2)
1138{
1139	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1140	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1141
1142	ORDERKEY_PRIO(p1, p2);
1143	ORDERKEY_CPTICKS(p1, p2);
1144	ORDERKEY_PCTCPU(p1, p2);
1145	ORDERKEY_STATE(p1, p2);
1146	ORDERKEY_RSSIZE(p1, p2);
1147	ORDERKEY_MEM(p1, p2);
1148
1149	return (0);
1150}
1151
1152/* compare_threads - the comparison function for sorting by threads */
1153int
1154compare_threads(void *arg1, void *arg2)
1155{
1156	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1157	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1158
1159	ORDERKEY_THREADS(p1, p2);
1160	ORDERKEY_PCTCPU(p1, p2);
1161	ORDERKEY_CPTICKS(p1, p2);
1162	ORDERKEY_STATE(p1, p2);
1163	ORDERKEY_PRIO(p1, p2);
1164	ORDERKEY_RSSIZE(p1, p2);
1165	ORDERKEY_MEM(p1, p2);
1166
1167	return (0);
1168}
1169
1170/* compare_jid - the comparison function for sorting by jid */
1171static int
1172compare_jid(const void *arg1, const void *arg2)
1173{
1174	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1175	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1176
1177	ORDERKEY_JID(p1, p2);
1178	ORDERKEY_PCTCPU(p1, p2);
1179	ORDERKEY_CPTICKS(p1, p2);
1180	ORDERKEY_STATE(p1, p2);
1181	ORDERKEY_PRIO(p1, p2);
1182	ORDERKEY_RSSIZE(p1, p2);
1183	ORDERKEY_MEM(p1, p2);
1184
1185	return (0);
1186}
1187#endif /* ORDER */
1188
1189/* assorted comparison functions for sorting by i/o */
1190
1191int
1192#ifdef ORDER
1193compare_iototal(void *arg1, void *arg2)
1194#else
1195io_compare(void *arg1, void *arg2)
1196#endif
1197{
1198	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1199	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1200
1201	return (get_io_total(p2) - get_io_total(p1));
1202}
1203
1204#ifdef ORDER
1205int
1206compare_ioread(void *arg1, void *arg2)
1207{
1208	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1209	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1210	long dummy, inp1, inp2;
1211
1212	(void) get_io_stats(p1, &inp1, &dummy, &dummy, &dummy, &dummy);
1213	(void) get_io_stats(p2, &inp2, &dummy, &dummy, &dummy, &dummy);
1214
1215	return (inp2 - inp1);
1216}
1217
1218int
1219compare_iowrite(void *arg1, void *arg2)
1220{
1221	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1222	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1223	long dummy, oup1, oup2;
1224
1225	(void) get_io_stats(p1, &dummy, &oup1, &dummy, &dummy, &dummy);
1226	(void) get_io_stats(p2, &dummy, &oup2, &dummy, &dummy, &dummy);
1227
1228	return (oup2 - oup1);
1229}
1230
1231int
1232compare_iofault(void *arg1, void *arg2)
1233{
1234	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1235	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1236	long dummy, flp1, flp2;
1237
1238	(void) get_io_stats(p1, &dummy, &dummy, &flp1, &dummy, &dummy);
1239	(void) get_io_stats(p2, &dummy, &dummy, &flp2, &dummy, &dummy);
1240
1241	return (flp2 - flp1);
1242}
1243
1244int
1245compare_vcsw(void *arg1, void *arg2)
1246{
1247	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1248	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1249	long dummy, flp1, flp2;
1250
1251	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &flp1, &dummy);
1252	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &flp2, &dummy);
1253
1254	return (flp2 - flp1);
1255}
1256
1257int
1258compare_ivcsw(void *arg1, void *arg2)
1259{
1260	struct kinfo_proc *p1 = *(struct kinfo_proc **)arg1;
1261	struct kinfo_proc *p2 = *(struct kinfo_proc **)arg2;
1262	long dummy, flp1, flp2;
1263
1264	(void) get_io_stats(p1, &dummy, &dummy, &dummy, &dummy, &flp1);
1265	(void) get_io_stats(p2, &dummy, &dummy, &dummy, &dummy, &flp2);
1266
1267	return (flp2 - flp1);
1268}
1269#endif /* ORDER */
1270
1271/*
1272 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
1273 *		the process does not exist.
1274 *		It is EXTREMLY IMPORTANT that this function work correctly.
1275 *		If top runs setuid root (as in SVR4), then this function
1276 *		is the only thing that stands in the way of a serious
1277 *		security problem.  It validates requests for the "kill"
1278 *		and "renice" commands.
1279 */
1280
1281int
1282proc_owner(int pid)
1283{
1284	int cnt;
1285	struct kinfo_proc **prefp;
1286	struct kinfo_proc *pp;
1287
1288	prefp = pref;
1289	cnt = pref_len;
1290	while (--cnt >= 0) {
1291		pp = *prefp++;
1292		if (pp->ki_pid == (pid_t)pid)
1293			return ((int)pp->ki_ruid);
1294	}
1295	return (-1);
1296}
1297
1298static int
1299swapmode(int *retavail, int *retfree)
1300{
1301	int n;
1302	int pagesize = getpagesize();
1303	struct kvm_swap swapary[1];
1304
1305	*retavail = 0;
1306	*retfree = 0;
1307
1308#define CONVERT(v)	((quad_t)(v) * pagesize / 1024)
1309
1310	n = kvm_getswapinfo(kd, swapary, 1, 0);
1311	if (n < 0 || swapary[0].ksw_total == 0)
1312		return (0);
1313
1314	*retavail = CONVERT(swapary[0].ksw_total);
1315	*retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
1316
1317	n = (int)(swapary[0].ksw_used * 100.0 / swapary[0].ksw_total);
1318	return (n);
1319}
1320