2 * top - a top users display for Unix
4 * SYNOPSIS: For FreeBSD-2.x and later
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/)
12 * This is the machine-dependent module for FreeBSD 2.2
14 * FreeBSD 2.2.x, 3.x, 4.x, and probably FreeBSD 2.1.x
18 * AUTHOR: Christos Zoulas <christos@ee.cornell.edu>
19 * Steven Wallace <swallace@freebsd.org>
20 * Wolfram Schneider <wosch@FreeBSD.org>
21 * Hiten Pandya <hmp@backplane.com>
23 * $FreeBSD: src/usr.bin/top/machine.c,v 1.29.2.2 2001/07/31 20:27:05 tmm Exp $
24 * $DragonFly: src/usr.bin/top/machine.c,v 1.19 2006/10/03 12:20:11 y0netan1 Exp $
29 #include <sys/types.h>
30 #include <sys/signal.h>
31 #include <sys/param.h>
39 #include <sys/errno.h>
40 #include <sys/sysctl.h>
44 #include <sys/vmmeter.h>
45 #include <sys/resource.h>
46 #include <sys/rtprio.h>
52 #include <osreldate.h> /* for changes in kernel structures */
54 #include <sys/kinfo.h>
59 static int check_nlist(struct nlist *);
60 static int getkval(unsigned long, int *, int, char *);
61 extern char* printable(char *);
62 int swapmode(int *retavail, int *retfree);
64 static int namelength;
68 * needs to be a global symbol, so wrapper can be
69 * modified accordingly.
71 static int show_threads = 0;
73 /* get_process_info passes back a handle. This is what it looks like: */
77 struct kinfo_proc **next_proc; /* points to next valid proc pointer */
78 int remaining; /* number of pointers remaining */
81 /* declarations for load_avg */
84 #define PP(pp, field) ((pp)->kp_proc . field)
85 #define EP(pp, field) ((pp)->kp_eproc . field)
86 #define TP(pp, field) ((pp)->kp_thread . field)
87 #define VP(pp, field) ((pp)->kp_eproc.e_vm . field)
89 /* define what weighted cpu is. */
90 #define weighted_cpu(pct, pp) (PP((pp), p_swtime) == 0 ? 0.0 : \
91 ((pct) / (1.0 - exp(PP((pp), p_swtime) * logcpu))))
93 /* what we consider to be process size: */
94 #define PROCSIZE(pp) (VP((pp), vm_map.size) / 1024)
97 * These definitions control the format of the per-process area
100 static char smp_header[] =
101 " PID %-*.*s PRI NICE SIZE RES STATE C TIME WCPU CPU COMMAND";
103 #define smp_Proc_format \
104 "%5d %-*.*s %3d %3d%7s %6s %-6.6s %1x%7s %5.2f%% %5.2f%% %.*s"
106 static char up_header[] =
107 " PID %-*.*s PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND";
109 #define up_Proc_format \
110 "%5d %-*.*s %3d %3d%7s %6s %-6.6s%.0d%7s %5.2f%% %5.2f%% %.*s"
114 /* process state names for the "STATE" column of the display */
115 /* the extra nulls in the string "run" are for adding a slash and
116 the processor number when needed */
118 char *state_abbrev[] =
120 "", "START", "RUN\0\0\0", "SLEEP", "STOP", "ZOMB",
126 /* values that we stash away in _init and use in later routines */
128 static double logcpu;
134 /* these are for calculating cpu state percentages */
136 static struct kinfo_cputime *cp_time, *cp_old;
138 /* these are for detailing the process states */
140 int process_states[6];
141 char *procstatenames[] = {
142 "", " starting, ", " running, ", " sleeping, ", " stopped, ",
147 /* these are for detailing the cpu states */
150 char *cpustatenames[CPU_STATES + 1] = {
151 "user", "nice", "system", "interrupt", "idle", NULL
154 /* these are for detailing the memory statistics */
157 char *memorynames[] = {
158 "K Active, ", "K Inact, ", "K Wired, ", "K Cache, ", "K Buf, ", "K Free",
163 char *swapnames[] = {
165 "K Total, ", "K Used, ", "K Free, ", "% Inuse, ", "K In, ", "K Out",
170 /* these are for keeping track of the proc array */
173 static int onproc = -1;
175 static struct kinfo_proc *pbase;
176 static struct kinfo_proc **pref;
178 /* these are for getting the memory statistics */
180 static int pageshift; /* log base 2 of the pagesize */
182 /* define pagetok in terms of pageshift */
184 #define pagetok(size) ((size) << pageshift)
187 /* sorting orders. first is default */
188 char *ordernames[] = {
189 "cpu", "size", "res", "time", "pri", "thr", NULL
194 cputime_percentages(int out[CPU_STATES], struct kinfo_cputime *new,
195 struct kinfo_cputime *old)
197 struct kinfo_cputime diffs;
198 uint64_t total_change, half_total;
203 diffs.cp_user = new->cp_user - old->cp_user;
204 diffs.cp_nice = new->cp_nice - old->cp_nice;
205 diffs.cp_sys = new->cp_sys - old->cp_sys;
206 diffs.cp_intr = new->cp_intr - old->cp_intr;
207 diffs.cp_idle = new->cp_idle - old->cp_idle;
208 total_change = diffs.cp_user + diffs.cp_nice + diffs.cp_sys +
209 diffs.cp_intr + diffs.cp_idle;
210 old->cp_user = new->cp_user;
211 old->cp_nice = new->cp_nice;
212 old->cp_sys = new->cp_sys;
213 old->cp_intr = new->cp_intr;
214 old->cp_idle = new->cp_idle;
216 /* avoid divide by zero potential */
217 if (total_change == 0)
220 /* calculate percentages based on overall change, rounding up */
221 half_total = total_change >> 1;
223 out[0] = ((diffs.cp_user * 1000LL + half_total) / total_change);
224 out[1] = ((diffs.cp_nice * 1000LL + half_total) / total_change);
225 out[2] = ((diffs.cp_sys * 1000LL + half_total) / total_change);
226 out[3] = ((diffs.cp_intr * 1000LL + half_total) / total_change);
227 out[4] = ((diffs.cp_idle * 1000LL + half_total) / total_change);
231 machine_init(struct statics *statics)
234 register int pagesize;
239 if (kinfo_get_cpus(&n_cpus))
240 err(1, "kinfo_get_cpus failed");
242 modelen = sizeof(smpmode);
243 if ((sysctlbyname("machdep.smp_active", &smpmode, &modelen, NULL, 0) < 0 &&
244 sysctlbyname("smp.smp_active", &smpmode, &modelen, NULL, 0) < 0) ||
245 modelen != sizeof(smpmode))
248 while ((pw = getpwent()) != NULL) {
249 if (strlen(pw->pw_name) > namelength)
250 namelength = strlen(pw->pw_name);
254 if (smpmode && namelength > 13)
256 else if (namelength > 15)
259 if ((kd = kvm_open(NULL, NULL, NULL, O_RDONLY, "kvm_open")) == NULL)
262 if (kinfo_get_sched_ccpu(&ccpu)) {
263 fprintf(stderr, "top: kinfo_get_sched_ccpu failed\n");
267 /* this is used in calculating WCPU -- calculate it ahead of time */
268 logcpu = log(loaddouble(ccpu));
274 /* get the page size with "getpagesize" and calculate pageshift from it */
275 pagesize = getpagesize();
283 /* we only need the amount of log(2)1024 for our conversion */
284 pageshift -= LOG1024;
286 /* fill in the statics information */
287 statics->procstate_names = procstatenames;
288 statics->cpustate_names = cpustatenames;
289 statics->memory_names = memorynames;
290 statics->swap_names = swapnames;
292 statics->order_names = ordernames;
299 char *format_header(register char *uname_field)
302 static char Header[128];
304 snprintf(Header, sizeof(Header), smpmode ? smp_header : up_header,
305 namelength, namelength, uname_field);
307 if (screen_width <= 79)
312 cmdlength = cmdlength - strlen(Header) + 6;
317 static int swappgsin = -1;
318 static int swappgsout = -1;
319 extern struct timeval timeout;
322 get_system_info(struct system_info *si)
326 struct timeval boottime;
331 if (cpu_states == NULL) {
332 cpu_states = malloc(sizeof(*cpu_states) * CPU_STATES * n_cpus);
333 if (cpu_states == NULL)
335 bzero(cpu_states, sizeof(*cpu_states) * CPU_STATES * n_cpus);
337 if (cp_time == NULL) {
338 cp_time = malloc(2 * n_cpus * sizeof(cp_time[0]));
341 cp_old = cp_time + n_cpus;
343 len = n_cpus * sizeof(cp_old[0]);
345 if (sysctlbyname("kern.cputime", cp_old, &len, NULL, 0))
346 err(1, "kern.cputime");
349 len = n_cpus * sizeof(cp_time[0]);
351 if (sysctlbyname("kern.cputime", cp_time, &len, NULL, 0))
352 err(1, "kern.cputime");
354 getloadavg(si->load_avg, 3);
358 /* convert cp_time counts to percentages */
359 for (cpu = 0; cpu < n_cpus; ++cpu) {
360 cputime_percentages(cpu_states + cpu * CPU_STATES,
361 &cp_time[cpu], &cp_old[cpu]);
364 /* sum memory & swap statistics */
368 int vms_size = sizeof(vms);
369 int vmm_size = sizeof(vmm);
370 static unsigned int swap_delay = 0;
371 static int swapavail = 0;
372 static int swapfree = 0;
373 static int bufspace = 0;
375 if (sysctlbyname("vm.vmstats", &vms, &vms_size, NULL, 0))
376 err(1, "sysctlbyname: vm.vmstats");
378 if (sysctlbyname("vm.vmmeter", &vmm, &vmm_size, NULL, 0))
379 err(1, "sysctlbyname: vm.vmmeter");
381 if (kinfo_get_vfs_bufspace(&bufspace))
382 err(1, "kinfo_get_vfs_bufspace");
384 /* convert memory stats to Kbytes */
385 memory_stats[0] = pagetok(vms.v_active_count);
386 memory_stats[1] = pagetok(vms.v_inactive_count);
387 memory_stats[2] = pagetok(vms.v_wire_count);
388 memory_stats[3] = pagetok(vms.v_cache_count);
389 memory_stats[4] = bufspace / 1024;
390 memory_stats[5] = pagetok(vms.v_free_count);
391 memory_stats[6] = -1;
399 /* compute differences between old and new swap statistic */
401 swap_stats[4] = pagetok(((vmm.v_swappgsin - swappgsin)));
402 swap_stats[5] = pagetok(((vmm.v_swappgsout - swappgsout)));
405 swappgsin = vmm.v_swappgsin;
406 swappgsout = vmm.v_swappgsout;
408 /* call CPU heavy swapmode() only for changes */
409 if (swap_stats[4] > 0 || swap_stats[5] > 0 || swap_delay == 0) {
410 swap_stats[3] = swapmode(&swapavail, &swapfree);
411 swap_stats[0] = swapavail;
412 swap_stats[1] = swapavail - swapfree;
413 swap_stats[2] = swapfree;
419 /* set arrays and strings */
420 si->cpustates = cpu_states;
421 si->memory = memory_stats;
422 si->swap = swap_stats;
426 si->last_pid = lastpid;
432 * Print how long system has been up.
433 * (Found by looking getting "boottime" from the kernel)
436 mib[1] = KERN_BOOTTIME;
437 bt_size = sizeof(boottime);
438 if (sysctl(mib, 2, &boottime, &bt_size, NULL, 0) != -1 &&
439 boottime.tv_sec != 0) {
440 si->boottime = boottime;
442 si->boottime.tv_sec = -1;
446 static struct handle handle;
448 caddr_t get_process_info(struct system_info *si, struct process_select *sel,
452 register int total_procs;
453 register int active_procs;
454 register struct kinfo_proc **prefp;
455 register struct kinfo_proc *pp;
457 /* these are copied out of sel for speed */
461 int show_only_threads;
466 pbase = kvm_getprocs(kd, KERN_PROC_ALL, 0, &nproc);
468 pref = (struct kinfo_proc **) realloc(pref, sizeof(struct kinfo_proc *)
470 if (pref == NULL || pbase == NULL) {
471 (void) fprintf(stderr, "top: Out of memory.\n");
474 /* get a pointer to the states summary array */
475 si->procstates = process_states;
477 /* set up flags which define what we are going to select */
478 show_idle = sel->idle;
479 show_self = sel->self;
480 show_system = sel->system;
481 show_threads = sel->threads;
482 show_only_threads = sel->only_threads;
483 show_uid = sel->uid != -1;
484 show_command = sel->command != NULL;
486 /* count up process states and get pointers to interesting procs */
489 memset((char *)process_states, 0, sizeof(process_states));
491 for (pp = pbase, i = 0; i < nproc; pp++, i++)
494 * Place pointers to each valid proc structure in pref[].
495 * Process slots that are actually in use have a non-zero
496 * status field. Processes with P_SYSTEM set are system
497 * processes---these get ignored unless show_sysprocs is set.
499 if ((show_threads && (TP(pp, td_proc) == NULL)) ||
500 (!show_only_threads && (PP(pp, p_stat) != 0 &&
501 (show_self != PP(pp, p_pid)) &&
502 (show_system || ((PP(pp, p_flag) & P_SYSTEM) == 0)))))
505 process_states[(unsigned char) PP(pp, p_stat)]++;
506 if ((show_threads && (TP(pp, td_proc) == NULL)) ||
507 (!show_only_threads && (PP(pp, p_stat) != SZOMB) &&
508 (show_idle || (PP(pp, p_pctcpu) != 0) ||
509 (PP(pp, p_stat) == SRUN)) &&
510 (!show_uid || EP(pp, e_ucred.cr_ruid) == (uid_t)sel->uid)))
518 /* if requested, sort the "interesting" processes */
521 qsort((char *)pref, active_procs, sizeof(struct kinfo_proc *), compare);
524 /* remember active and total counts */
525 si->p_total = total_procs;
526 si->p_active = pref_len = active_procs;
528 /* pass back a handle */
529 handle.next_proc = pref;
530 handle.remaining = active_procs;
531 return((caddr_t)&handle);
534 char fmt[128]; /* static area where result is built */
536 char *format_next_process(caddr_t handle, char *(*get_userid)())
538 struct kinfo_proc *pp;
547 /* find and remember the next proc structure */
548 hp = (struct handle *)handle;
549 pp = *(hp->next_proc++);
552 /* set the wrapper for the process/thread name */
553 if ((PP(pp, p_flag) & P_SWAPPEDOUT))
554 wrapper = "[]"; /* swapped process [pname] */
555 else if (((PP(pp, p_flag) & P_SYSTEM) != 0) && (TP(pp, td_proc) != NULL))
556 wrapper = "()"; /* system process (pname) */
557 else if (show_threads && (TP(pp, td_proc) == NULL))
558 wrapper = "<>"; /* pure kernel threads <thread> */
562 /* get the process's command name */
563 if (wrapper != NULL) {
564 char *comm = TP(pp, td_comm);
565 #define COMSIZ sizeof(TP(pp, td_comm))
567 (void) strncpy(buf, comm, COMSIZ);
568 comm[0] = wrapper[0];
569 (void) strncpy(&comm[1], buf, COMSIZ - 2);
570 comm[COMSIZ - 2] = '\0';
571 (void) strncat(comm, &wrapper[1], COMSIZ - 1);
572 comm[COMSIZ - 1] = '\0';
576 * Convert the process's runtime from microseconds to seconds. This
577 * time includes the interrupt time although that is not wanted here.
578 * ps(1) is similarly sloppy.
580 cputime = (EP(pp, e_uticks) + EP(pp, e_sticks)) / 1000000;
582 /* calculate the base for cpu percentages */
583 pct = pctdouble(PP(pp, p_pctcpu));
585 /* generate "STATE" field */
586 switch (state = PP(pp, p_stat)) {
588 if (smpmode && TP(pp, td_flags) & TDF_RUNNING)
589 sprintf(status, "CPU%d", EP(pp, e_cpuid));
591 strcpy(status, "RUN");
594 if (TP(pp, td_wmesg) != NULL) {
595 sprintf(status, "%.6s", EP(pp, e_wmesg));
602 state < sizeof(state_abbrev) / sizeof(*state_abbrev))
603 sprintf(status, "%.6s", state_abbrev[(unsigned char) state]);
605 sprintf(status, "?%5d", state);
610 * idle time 0 - 31 -> nice value +21 - +52
611 * normal time -> nice value -20 - +20
612 * real time 0 - 31 -> nice value -52 - -21
613 * thread 0 - 31 -> nice value -53 -
615 switch(PP(pp, p_rtprio.type)) {
616 case RTP_PRIO_REALTIME:
617 nice = PRIO_MIN - 1 - RTP_PRIO_MAX + PP(pp, p_rtprio.prio);
620 nice = PRIO_MAX + 1 + PP(pp, p_rtprio.prio);
622 case RTP_PRIO_THREAD:
623 nice = PRIO_MIN - 1 - RTP_PRIO_MAX - PP(pp, p_rtprio.prio);
626 nice = PP(pp, p_nice);
631 /* format this entry */
632 snprintf(fmt, sizeof(fmt),
633 smpmode ? smp_Proc_format : up_Proc_format,
635 namelength, namelength,
636 (*get_userid)(EP(pp, e_ucred.cr_ruid)),
637 (show_threads && (TP(pp, td_proc) == NULL)) ? TP(pp, td_pri) :
638 PP(pp, p_usdata.bsd4.priority),
640 format_k2(PROCSIZE(pp)),
641 format_k2(pagetok(VP(pp, vm_rssize))),
643 smpmode ? EP(pp, e_cpuid) : 0,
644 format_time(cputime),
645 100.0 * weighted_cpu(pct, pp),
648 printable(TP(pp, td_comm)));
650 /* return the result */
656 * check_nlist(nlst) - checks the nlist to see if any symbols were not
657 * found. For every symbol that was not found, a one-line
658 * message is printed to stderr. The routine returns the
659 * number of symbols NOT found.
662 static int check_nlist(register struct nlist *nlst)
666 /* check to see if we got ALL the symbols we requested */
667 /* this will write one line to stderr for every symbol not found */
670 while (nlst->n_name != NULL)
672 if (nlst->n_type == 0)
674 /* this one wasn't found */
675 (void) fprintf(stderr, "kernel: no symbol named `%s'\n",
685 /* comparison routines for qsort */
688 * proc_compare - comparison function for "qsort"
689 * Compares the resource consumption of two processes using five
690 * distinct keys. The keys (in descending order of importance) are:
691 * percent cpu, cpu ticks, state, resident set size, total virtual
692 * memory usage. The process states are ordered as follows (from least
693 * to most important): WAIT, zombie, sleep, stop, start, run. The
694 * array declaration below maps a process state index into a number
695 * that reflects this ordering.
698 static unsigned char sorted_state[] =
702 1, /* ABANDONED (WAIT) */
710 #define ORDERKEY_PCTCPU \
711 if (lresult = (long) PP(p2, p_pctcpu) - (long) PP(p1, p_pctcpu), \
712 (result = lresult > 0 ? 1 : lresult < 0 ? -1 : 0) == 0)
714 #define CPTICKS(p) (EP(p, e_uticks) + EP(p, e_sticks))
716 #define ORDERKEY_CPTICKS \
717 if ((result = CPTICKS(p2) > CPTICKS(p1) ? 1 : \
718 CPTICKS(p2) < CPTICKS(p1) ? -1 : 0) == 0)
720 #define ORDERKEY_STATE \
721 if ((result = sorted_state[(unsigned char) PP(p2, p_stat)] - \
722 sorted_state[(unsigned char) PP(p1, p_stat)]) == 0)
724 #define ORDERKEY_PRIO \
725 if ((result = PP(p2, p_usdata.bsd4.priority) - PP(p1, p_usdata.bsd4.priority)) == 0)
727 #define ORDERKEY_KTHREADS \
728 if ((result = (TP(p1, td_proc) == NULL) - (TP(p2, td_proc) == NULL)) == 0)
730 #define ORDERKEY_KTHREADS_PRIO \
731 if ((result = TP(p2, td_pri) - TP(p1, td_pri)) == 0)
733 #define ORDERKEY_RSSIZE \
734 if ((result = VP(p2, vm_rssize) - VP(p1, vm_rssize)) == 0)
736 #define ORDERKEY_MEM \
737 if ( (result = PROCSIZE(p2) - PROCSIZE(p1)) == 0 )
739 /* compare_cpu - the comparison function for sorting by cpu percentage */
743 compare_cpu(struct proc **pp1, struct proc **pp2)
745 proc_compare(struct proc **pp1, struct proc **pp2)
748 register struct kinfo_proc *p1;
749 register struct kinfo_proc *p2;
751 register pctcpu lresult;
753 /* remove one level of indirection */
754 p1 = *(struct kinfo_proc **) pp1;
755 p2 = *(struct kinfo_proc **) pp2;
769 /* compare routines */
770 int compare_size(), compare_res(), compare_time(), compare_prio(), compare_thr();
772 int (*proc_compares[])() = {
782 /* compare_size - the comparison function for sorting by total memory usage */
785 compare_size(struct proc **pp1, struct proc **pp2)
787 register struct kinfo_proc *p1;
788 register struct kinfo_proc *p2;
790 register pctcpu lresult;
792 /* remove one level of indirection */
793 p1 = *(struct kinfo_proc **) pp1;
794 p2 = *(struct kinfo_proc **) pp2;
807 /* compare_res - the comparison function for sorting by resident set size */
810 compare_res(struct proc **pp1, struct proc **pp2)
812 register struct kinfo_proc *p1;
813 register struct kinfo_proc *p2;
815 register pctcpu lresult;
817 /* remove one level of indirection */
818 p1 = *(struct kinfo_proc **) pp1;
819 p2 = *(struct kinfo_proc **) pp2;
832 /* compare_time - the comparison function for sorting by total cpu time */
835 compare_time(struct proc **pp1, struct proc **pp2)
837 register struct kinfo_proc *p1;
838 register struct kinfo_proc *p2;
840 register pctcpu lresult;
842 /* remove one level of indirection */
843 p1 = *(struct kinfo_proc **) pp1;
844 p2 = *(struct kinfo_proc **) pp2;
849 ORDERKEY_KTHREADS_PRIO
859 /* compare_prio - the comparison function for sorting by cpu percentage */
862 compare_prio(struct proc **pp1, struct proc **pp2)
864 register struct kinfo_proc *p1;
865 register struct kinfo_proc *p2;
867 register pctcpu lresult;
869 /* remove one level of indirection */
870 p1 = *(struct kinfo_proc **) pp1;
871 p2 = *(struct kinfo_proc **) pp2;
874 ORDERKEY_KTHREADS_PRIO
887 compare_thr(struct proc **pp1, struct proc **pp2)
889 register struct kinfo_proc *p1;
890 register struct kinfo_proc *p2;
892 register pctcpu lresult;
894 /* remove one level of indirection */
895 p1 = *(struct kinfo_proc **) pp1;
896 p2 = *(struct kinfo_proc **) pp2;
899 ORDERKEY_KTHREADS_PRIO
914 * proc_owner(pid) - returns the uid that owns process "pid", or -1 if
915 * the process does not exist.
916 * It is EXTREMLY IMPORTANT that this function work correctly.
917 * If top runs setuid root (as in SVR4), then this function
918 * is the only thing that stands in the way of a serious
919 * security problem. It validates requests for the "kill"
920 * and "renice" commands.
923 int proc_owner(int pid)
926 register struct kinfo_proc **prefp;
927 register struct kinfo_proc *pp;
934 if (PP(pp, p_pid) == (pid_t)pid)
936 return((int)EP(pp, e_ucred.cr_ruid));
944 * swapmode is based on a program called swapinfo written
945 * by Kevin Lahey <kml@rokkaku.atl.ga.us>.
948 swapmode(int *retavail, int *retfree)
951 int pagesize = getpagesize();
952 struct kvm_swap swapary[1];
957 #define CONVERT(v) ((quad_t)(v) * pagesize / 1024)
959 n = kvm_getswapinfo(kd, swapary, 1, 0);
960 if (n < 0 || swapary[0].ksw_total == 0)
963 *retavail = CONVERT(swapary[0].ksw_total);
964 *retfree = CONVERT(swapary[0].ksw_total - swapary[0].ksw_used);
966 n = (int)((double)swapary[0].ksw_used * 100.0 /
967 (double)swapary[0].ksw_total);