top: document a debugging script
[oweals/busybox.git] / procps / top.c
1 /* vi: set sw=4 ts=4: */
2 /*
3  * A tiny 'top' utility.
4  *
5  * This is written specifically for the linux /proc/<PID>/stat(m)
6  * files format.
7  *
8  * This reads the PIDs of all processes and their status and shows
9  * the status of processes (first ones that fit to screen) at given
10  * intervals.
11  *
12  * NOTES:
13  * - At startup this changes to /proc, all the reads are then
14  *   relative to that.
15  *
16  * (C) Eero Tamminen <oak at welho dot com>
17  *
18  * Rewritten by Vladimir Oleynik (C) 2002 <dzo@simtreas.ru>
19  *
20  * Sept 2008: Vineet Gupta <vineet.gupta@arc.com>
21  * Added Support for reporting SMP Information
22  * - CPU where Process was last seen running
23  *   (to see effect of sched_setaffinity() etc)
24  * - CPU Time Split (idle/IO/wait etc) PER CPU
25  *
26  * Copyright (c) 1992 Branko Lankester
27  * Copyright (c) 1992 Roger Binns
28  * Copyright (C) 1994-1996 Charles L. Blake.
29  * Copyright (C) 1992-1998 Michael K. Johnson
30  *
31  * Licensed under GPLv2, see file LICENSE in this source tree.
32  */
33 /* Howto snapshot /proc for debugging top problems:
34  * for f in /proc/[0-9]*""/stat; do
35  *         n=${f#/proc/}
36  *         n=${n%/stat}_stat
37  *         cp $f $n
38  * done
39  * cp /proc/stat /proc/meminfo .
40  * top -bn1 >top.out
41  */
42
43 #include "libbb.h"
44
45
46 typedef struct top_status_t {
47         unsigned long vsz;
48 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
49         unsigned long ticks;
50         unsigned pcpu; /* delta of ticks */
51 #endif
52         unsigned pid, ppid;
53         unsigned uid;
54         char state[4];
55         char comm[COMM_LEN];
56 #if ENABLE_FEATURE_TOP_SMP_PROCESS
57         int last_seen_on_cpu;
58 #endif
59 } top_status_t;
60
61 typedef struct jiffy_counts_t {
62         /* Linux 2.4.x has only first four */
63         unsigned long long usr, nic, sys, idle;
64         unsigned long long iowait, irq, softirq, steal;
65         unsigned long long total;
66         unsigned long long busy;
67 } jiffy_counts_t;
68
69 /* This structure stores some critical information from one frame to
70    the next. Used for finding deltas. */
71 typedef struct save_hist {
72         unsigned long ticks;
73         pid_t pid;
74 } save_hist;
75
76 typedef int (*cmp_funcp)(top_status_t *P, top_status_t *Q);
77
78
79 enum { SORT_DEPTH = 3 };
80
81
82 struct globals {
83         top_status_t *top;
84         int ntop;
85 #if ENABLE_FEATURE_TOPMEM
86         smallint sort_field;
87         smallint inverted;
88 #endif
89 #if ENABLE_FEATURE_TOP_SMP_CPU
90         smallint smp_cpu_info; /* one/many cpu info lines? */
91 #endif
92 #if ENABLE_FEATURE_USE_TERMIOS
93         struct termios initial_settings;
94 #endif
95 #if !ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
96         cmp_funcp sort_function[1];
97 #else
98         cmp_funcp sort_function[SORT_DEPTH];
99         struct save_hist *prev_hist;
100         int prev_hist_count;
101         jiffy_counts_t cur_jif, prev_jif;
102         /* int hist_iterations; */
103         unsigned total_pcpu;
104         /* unsigned long total_vsz; */
105 #endif
106 #if ENABLE_FEATURE_TOP_SMP_CPU
107         /* Per CPU samples: current and last */
108         jiffy_counts_t *cpu_jif, *cpu_prev_jif;
109         int num_cpus;
110 #endif
111         char line_buf[80];
112 }; //FIX_ALIASING; - large code growth
113 enum { LINE_BUF_SIZE = COMMON_BUFSIZE - offsetof(struct globals, line_buf) };
114 #define G (*(struct globals*)&bb_common_bufsiz1)
115 struct BUG_bad_size {
116         char BUG_G_too_big[sizeof(G) <= COMMON_BUFSIZE ? 1 : -1];
117         char BUG_line_buf_too_small[LINE_BUF_SIZE > 80 ? 1 : -1];
118 };
119 #define INIT_G() do { } while (0)
120 #define top              (G.top               )
121 #define ntop             (G.ntop              )
122 #define sort_field       (G.sort_field        )
123 #define inverted         (G.inverted          )
124 #define smp_cpu_info     (G.smp_cpu_info      )
125 #define initial_settings (G.initial_settings  )
126 #define sort_function    (G.sort_function     )
127 #define prev_hist        (G.prev_hist         )
128 #define prev_hist_count  (G.prev_hist_count   )
129 #define cur_jif          (G.cur_jif           )
130 #define prev_jif         (G.prev_jif          )
131 #define cpu_jif          (G.cpu_jif           )
132 #define cpu_prev_jif     (G.cpu_prev_jif      )
133 #define num_cpus         (G.num_cpus          )
134 #define total_pcpu       (G.total_pcpu        )
135 #define line_buf         (G.line_buf          )
136
137 enum {
138         OPT_d = (1 << 0),
139         OPT_n = (1 << 1),
140         OPT_b = (1 << 2),
141         OPT_m = (1 << 3),
142         OPT_EOF = (1 << 4), /* pseudo: "we saw EOF in stdin" */
143 };
144 #define OPT_BATCH_MODE (option_mask32 & OPT_b)
145
146
147 #if ENABLE_FEATURE_USE_TERMIOS
148 static int pid_sort(top_status_t *P, top_status_t *Q)
149 {
150         /* Buggy wrt pids with high bit set */
151         /* (linux pids are in [1..2^15-1]) */
152         return (Q->pid - P->pid);
153 }
154 #endif
155
156 static int mem_sort(top_status_t *P, top_status_t *Q)
157 {
158         /* We want to avoid unsigned->signed and truncation errors */
159         if (Q->vsz < P->vsz) return -1;
160         return Q->vsz != P->vsz; /* 0 if ==, 1 if > */
161 }
162
163
164 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
165
166 static int pcpu_sort(top_status_t *P, top_status_t *Q)
167 {
168         /* Buggy wrt ticks with high bit set */
169         /* Affects only processes for which ticks overflow */
170         return (int)Q->pcpu - (int)P->pcpu;
171 }
172
173 static int time_sort(top_status_t *P, top_status_t *Q)
174 {
175         /* We want to avoid unsigned->signed and truncation errors */
176         if (Q->ticks < P->ticks) return -1;
177         return Q->ticks != P->ticks; /* 0 if ==, 1 if > */
178 }
179
180 static int mult_lvl_cmp(void* a, void* b)
181 {
182         int i, cmp_val;
183
184         for (i = 0; i < SORT_DEPTH; i++) {
185                 cmp_val = (*sort_function[i])(a, b);
186                 if (cmp_val != 0)
187                         return cmp_val;
188         }
189         return 0;
190 }
191
192 static NOINLINE int read_cpu_jiffy(FILE *fp, jiffy_counts_t *p_jif)
193 {
194 #if !ENABLE_FEATURE_TOP_SMP_CPU
195         static const char fmt[] = "cpu %llu %llu %llu %llu %llu %llu %llu %llu";
196 #else
197         static const char fmt[] = "cp%*s %llu %llu %llu %llu %llu %llu %llu %llu";
198 #endif
199         int ret;
200
201         if (!fgets(line_buf, LINE_BUF_SIZE, fp) || line_buf[0] != 'c' /* not "cpu" */)
202                 return 0;
203         ret = sscanf(line_buf, fmt,
204                         &p_jif->usr, &p_jif->nic, &p_jif->sys, &p_jif->idle,
205                         &p_jif->iowait, &p_jif->irq, &p_jif->softirq,
206                         &p_jif->steal);
207         if (ret >= 4) {
208                 p_jif->total = p_jif->usr + p_jif->nic + p_jif->sys + p_jif->idle
209                         + p_jif->iowait + p_jif->irq + p_jif->softirq + p_jif->steal;
210                 /* procps 2.x does not count iowait as busy time */
211                 p_jif->busy = p_jif->total - p_jif->idle - p_jif->iowait;
212         }
213
214         return ret;
215 }
216
217 static void get_jiffy_counts(void)
218 {
219         FILE* fp = xfopen_for_read("stat");
220
221         /* We need to parse cumulative counts even if SMP CPU display is on,
222          * they are used to calculate per process CPU% */
223         prev_jif = cur_jif;
224         if (read_cpu_jiffy(fp, &cur_jif) < 4)
225                 bb_error_msg_and_die("can't read /proc/stat");
226
227 #if !ENABLE_FEATURE_TOP_SMP_CPU
228         fclose(fp);
229         return;
230 #else
231         if (!smp_cpu_info) {
232                 fclose(fp);
233                 return;
234         }
235
236         if (!num_cpus) {
237                 /* First time here. How many CPUs?
238                  * There will be at least 1 /proc/stat line with cpu%d
239                  */
240                 while (1) {
241                         cpu_jif = xrealloc_vector(cpu_jif, 1, num_cpus);
242                         if (read_cpu_jiffy(fp, &cpu_jif[num_cpus]) <= 4)
243                                 break;
244                         num_cpus++;
245                 }
246                 if (num_cpus == 0) /* /proc/stat with only "cpu ..." line?! */
247                         smp_cpu_info = 0;
248
249                 cpu_prev_jif = xzalloc(sizeof(cpu_prev_jif[0]) * num_cpus);
250
251                 /* Otherwise the first per cpu display shows all 100% idles */
252                 usleep(50000);
253         } else { /* Non first time invocation */
254                 jiffy_counts_t *tmp;
255                 int i;
256
257                 /* First switch the sample pointers: no need to copy */
258                 tmp = cpu_prev_jif;
259                 cpu_prev_jif = cpu_jif;
260                 cpu_jif = tmp;
261
262                 /* Get the new samples */
263                 for (i = 0; i < num_cpus; i++)
264                         read_cpu_jiffy(fp, &cpu_jif[i]);
265         }
266 #endif
267         fclose(fp);
268 }
269
270 static void do_stats(void)
271 {
272         top_status_t *cur;
273         pid_t pid;
274         int i, last_i, n;
275         struct save_hist *new_hist;
276
277         get_jiffy_counts();
278         total_pcpu = 0;
279         /* total_vsz = 0; */
280         new_hist = xmalloc(sizeof(new_hist[0]) * ntop);
281         /*
282          * Make a pass through the data to get stats.
283          */
284         /* hist_iterations = 0; */
285         i = 0;
286         for (n = 0; n < ntop; n++) {
287                 cur = top + n;
288
289                 /*
290                  * Calculate time in cur process.  Time is sum of user time
291                  * and system time
292                  */
293                 pid = cur->pid;
294                 new_hist[n].ticks = cur->ticks;
295                 new_hist[n].pid = pid;
296
297                 /* find matching entry from previous pass */
298                 cur->pcpu = 0;
299                 /* do not start at index 0, continue at last used one
300                  * (brought hist_iterations from ~14000 down to 172) */
301                 last_i = i;
302                 if (prev_hist_count) do {
303                         if (prev_hist[i].pid == pid) {
304                                 cur->pcpu = cur->ticks - prev_hist[i].ticks;
305                                 total_pcpu += cur->pcpu;
306                                 break;
307                         }
308                         i = (i+1) % prev_hist_count;
309                         /* hist_iterations++; */
310                 } while (i != last_i);
311                 /* total_vsz += cur->vsz; */
312         }
313
314         /*
315          * Save cur frame's information.
316          */
317         free(prev_hist);
318         prev_hist = new_hist;
319         prev_hist_count = ntop;
320 }
321
322 #endif /* FEATURE_TOP_CPU_USAGE_PERCENTAGE */
323
324 #if ENABLE_FEATURE_TOP_CPU_GLOBAL_PERCENTS && ENABLE_FEATURE_TOP_DECIMALS
325 /* formats 7 char string (8 with terminating NUL) */
326 static char *fmt_100percent_8(char pbuf[8], unsigned value, unsigned total)
327 {
328         unsigned t;
329         if (value >= total) { /* 100% ? */
330                 strcpy(pbuf, "  100% ");
331                 return pbuf;
332         }
333         /* else generate " [N/space]N.N% " string */
334         value = 1000 * value / total;
335         t = value / 100;
336         value = value % 100;
337         pbuf[0] = ' ';
338         pbuf[1] = t ? t + '0' : ' ';
339         pbuf[2] = '0' + (value / 10);
340         pbuf[3] = '.';
341         pbuf[4] = '0' + (value % 10);
342         pbuf[5] = '%';
343         pbuf[6] = ' ';
344         pbuf[7] = '\0';
345         return pbuf;
346 }
347 #endif
348
349 #if ENABLE_FEATURE_TOP_CPU_GLOBAL_PERCENTS
350 static void display_cpus(int scr_width, char *scrbuf, int *lines_rem_p)
351 {
352         /*
353          * xxx% = (cur_jif.xxx - prev_jif.xxx) / (cur_jif.total - prev_jif.total) * 100%
354          */
355         unsigned total_diff;
356         jiffy_counts_t *p_jif, *p_prev_jif;
357         int i;
358 # if ENABLE_FEATURE_TOP_SMP_CPU
359         int n_cpu_lines;
360 # endif
361
362         /* using (unsigned) casts to make operations cheaper */
363 # define  CALC_TOTAL_DIFF do { \
364         total_diff = (unsigned)(p_jif->total - p_prev_jif->total); \
365         if (total_diff == 0) total_diff = 1; \
366 } while (0)
367
368 # if ENABLE_FEATURE_TOP_DECIMALS
369 #  define CALC_STAT(xxx) char xxx[8]
370 #  define SHOW_STAT(xxx) fmt_100percent_8(xxx, (unsigned)(p_jif->xxx - p_prev_jif->xxx), total_diff)
371 #  define FMT "%s"
372 # else
373 #  define CALC_STAT(xxx) unsigned xxx = 100 * (unsigned)(p_jif->xxx - p_prev_jif->xxx) / total_diff
374 #  define SHOW_STAT(xxx) xxx
375 #  define FMT "%4u%% "
376 # endif
377
378 # if !ENABLE_FEATURE_TOP_SMP_CPU
379         {
380                 i = 1;
381                 p_jif = &cur_jif;
382                 p_prev_jif = &prev_jif;
383 # else
384         /* Loop thru CPU(s) */
385         n_cpu_lines = smp_cpu_info ? num_cpus : 1;
386         if (n_cpu_lines > *lines_rem_p)
387                 n_cpu_lines = *lines_rem_p;
388
389         for (i = 0; i < n_cpu_lines; i++) {
390                 p_jif = &cpu_jif[i];
391                 p_prev_jif = &cpu_prev_jif[i];
392 # endif
393                 CALC_TOTAL_DIFF;
394
395                 { /* Need a block: CALC_STAT are declarations */
396                         CALC_STAT(usr);
397                         CALC_STAT(sys);
398                         CALC_STAT(nic);
399                         CALC_STAT(idle);
400                         CALC_STAT(iowait);
401                         CALC_STAT(irq);
402                         CALC_STAT(softirq);
403                         /*CALC_STAT(steal);*/
404
405                         snprintf(scrbuf, scr_width,
406                                 /* Barely fits in 79 chars when in "decimals" mode. */
407 # if ENABLE_FEATURE_TOP_SMP_CPU
408                                 "CPU%s:"FMT"usr"FMT"sys"FMT"nic"FMT"idle"FMT"io"FMT"irq"FMT"sirq",
409                                 (smp_cpu_info ? utoa(i) : ""),
410 # else
411                                 "CPU:"FMT"usr"FMT"sys"FMT"nic"FMT"idle"FMT"io"FMT"irq"FMT"sirq",
412 # endif
413                                 SHOW_STAT(usr), SHOW_STAT(sys), SHOW_STAT(nic), SHOW_STAT(idle),
414                                 SHOW_STAT(iowait), SHOW_STAT(irq), SHOW_STAT(softirq)
415                                 /*, SHOW_STAT(steal) - what is this 'steal' thing? */
416                                 /* I doubt anyone wants to know it */
417                         );
418                         puts(scrbuf);
419                 }
420         }
421 # undef SHOW_STAT
422 # undef CALC_STAT
423 # undef FMT
424         *lines_rem_p -= i;
425 }
426 #else  /* !ENABLE_FEATURE_TOP_CPU_GLOBAL_PERCENTS */
427 # define display_cpus(scr_width, scrbuf, lines_rem) ((void)0)
428 #endif
429
430 static unsigned long display_header(int scr_width, int *lines_rem_p)
431 {
432         FILE *fp;
433         char buf[80];
434         char scrbuf[80];
435         unsigned long total, used, mfree, shared, buffers, cached;
436
437         /* read memory info */
438         fp = xfopen_for_read("meminfo");
439
440         /*
441          * Old kernels (such as 2.4.x) had a nice summary of memory info that
442          * we could parse, however this is gone entirely in 2.6. Try parsing
443          * the old way first, and if that fails, parse each field manually.
444          *
445          * First, we read in the first line. Old kernels will have bogus
446          * strings we don't care about, whereas new kernels will start right
447          * out with MemTotal:
448          *                              -- PFM.
449          */
450         if (fscanf(fp, "MemTotal: %lu %s\n", &total, buf) != 2) {
451                 fgets(buf, sizeof(buf), fp);    /* skip first line */
452
453                 fscanf(fp, "Mem: %lu %lu %lu %lu %lu %lu",
454                         &total, &used, &mfree, &shared, &buffers, &cached);
455                 /* convert to kilobytes */
456                 used /= 1024;
457                 mfree /= 1024;
458                 shared /= 1024;
459                 buffers /= 1024;
460                 cached /= 1024;
461                 total /= 1024;
462         } else {
463                 /*
464                  * Revert to manual parsing, which incidentally already has the
465                  * sizes in kilobytes. This should be safe for both 2.4 and
466                  * 2.6.
467                  */
468                 fscanf(fp, "MemFree: %lu %s\n", &mfree, buf);
469
470                 /*
471                  * MemShared: is no longer present in 2.6. Report this as 0,
472                  * to maintain consistent behavior with normal procps.
473                  */
474                 if (fscanf(fp, "MemShared: %lu %s\n", &shared, buf) != 2)
475                         shared = 0;
476
477                 fscanf(fp, "Buffers: %lu %s\n", &buffers, buf);
478                 fscanf(fp, "Cached: %lu %s\n", &cached, buf);
479
480                 used = total - mfree;
481         }
482         fclose(fp);
483
484         /* output memory info */
485         if (scr_width > (int)sizeof(scrbuf))
486                 scr_width = sizeof(scrbuf);
487         snprintf(scrbuf, scr_width,
488                 "Mem: %luK used, %luK free, %luK shrd, %luK buff, %luK cached",
489                 used, mfree, shared, buffers, cached);
490         /* go to top & clear to the end of screen */
491         printf(OPT_BATCH_MODE ? "%s\n" : "\033[H\033[J%s\n", scrbuf);
492         (*lines_rem_p)--;
493
494         /* Display CPU time split as percentage of total time
495          * This displays either a cumulative line or one line per CPU
496          */
497         display_cpus(scr_width, scrbuf, lines_rem_p);
498
499         /* read load average as a string */
500         buf[0] = '\0';
501         open_read_close("loadavg", buf, sizeof(buf) - 1);
502         buf[sizeof(buf) - 1] = '\n';
503         *strchr(buf, '\n') = '\0';
504         snprintf(scrbuf, scr_width, "Load average: %s", buf);
505         puts(scrbuf);
506         (*lines_rem_p)--;
507
508         return total;
509 }
510
511 static NOINLINE void display_process_list(int lines_rem, int scr_width)
512 {
513         enum {
514                 BITS_PER_INT = sizeof(int) * 8
515         };
516
517         top_status_t *s;
518         char vsz_str_buf[8];
519         unsigned long total_memory = display_header(scr_width, &lines_rem); /* or use total_vsz? */
520         /* xxx_shift and xxx_scale variables allow us to replace
521          * expensive divides with multiply and shift */
522         unsigned pmem_shift, pmem_scale, pmem_half;
523 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
524         unsigned tmp_unsigned;
525         unsigned pcpu_shift, pcpu_scale, pcpu_half;
526         unsigned busy_jifs;
527 #endif
528
529         /* what info of the processes is shown */
530         printf(OPT_BATCH_MODE ? "%.*s" : "\033[7m%.*s\033[0m", scr_width,
531                 "  PID  PPID USER     STAT   VSZ %MEM"
532                 IF_FEATURE_TOP_SMP_PROCESS(" CPU")
533                 IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE(" %CPU")
534                 " COMMAND");
535         lines_rem--;
536
537 #if ENABLE_FEATURE_TOP_DECIMALS
538 # define UPSCALE 1000
539 # define CALC_STAT(name, val) div_t name = div((val), 10)
540 # define SHOW_STAT(name) name.quot, '0'+name.rem
541 # define FMT "%3u.%c"
542 #else
543 # define UPSCALE 100
544 # define CALC_STAT(name, val) unsigned name = (val)
545 # define SHOW_STAT(name) name
546 # define FMT "%4u%%"
547 #endif
548         /*
549          * MEM% = s->vsz/MemTotal
550          */
551         pmem_shift = BITS_PER_INT-11;
552         pmem_scale = UPSCALE*(1U<<(BITS_PER_INT-11)) / total_memory;
553         /* s->vsz is in kb. we want (s->vsz * pmem_scale) to never overflow */
554         while (pmem_scale >= 512) {
555                 pmem_scale /= 4;
556                 pmem_shift -= 2;
557         }
558         pmem_half = (1U << pmem_shift) / (ENABLE_FEATURE_TOP_DECIMALS ? 20 : 2);
559 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
560         busy_jifs = cur_jif.busy - prev_jif.busy;
561         /* This happens if there were lots of short-lived processes
562          * between two top updates (e.g. compilation) */
563         if (total_pcpu < busy_jifs) total_pcpu = busy_jifs;
564
565         /*
566          * CPU% = s->pcpu/sum(s->pcpu) * busy_cpu_ticks/total_cpu_ticks
567          * (pcpu is delta of sys+user time between samples)
568          */
569         /* (cur_jif.xxx - prev_jif.xxx) and s->pcpu are
570          * in 0..~64000 range (HZ*update_interval).
571          * we assume that unsigned is at least 32-bit.
572          */
573         pcpu_shift = 6;
574         pcpu_scale = UPSCALE*64 * (uint16_t)busy_jifs;
575         if (pcpu_scale == 0)
576                 pcpu_scale = 1;
577         while (pcpu_scale < (1U << (BITS_PER_INT-2))) {
578                 pcpu_scale *= 4;
579                 pcpu_shift += 2;
580         }
581         tmp_unsigned = (uint16_t)(cur_jif.total - prev_jif.total) * total_pcpu;
582         if (tmp_unsigned != 0)
583                 pcpu_scale /= tmp_unsigned;
584         /* we want (s->pcpu * pcpu_scale) to never overflow */
585         while (pcpu_scale >= 1024) {
586                 pcpu_scale /= 4;
587                 pcpu_shift -= 2;
588         }
589         pcpu_half = (1U << pcpu_shift) / (ENABLE_FEATURE_TOP_DECIMALS ? 20 : 2);
590         /* printf(" pmem_scale=%u pcpu_scale=%u ", pmem_scale, pcpu_scale); */
591 #endif
592
593         /* Ok, all preliminary data is ready, go through the list */
594         scr_width += 2; /* account for leading '\n' and trailing NUL */
595         if (lines_rem > ntop)
596                 lines_rem = ntop;
597         s = top;
598         while (--lines_rem >= 0) {
599                 unsigned col;
600                 CALC_STAT(pmem, (s->vsz*pmem_scale + pmem_half) >> pmem_shift);
601 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
602                 CALC_STAT(pcpu, (s->pcpu*pcpu_scale + pcpu_half) >> pcpu_shift);
603 #endif
604
605                 if (s->vsz >= 100000)
606                         sprintf(vsz_str_buf, "%6ldm", s->vsz/1024);
607                 else
608                         sprintf(vsz_str_buf, "%7ld", s->vsz);
609                 /* PID PPID USER STAT VSZ %MEM [%CPU] COMMAND */
610                 col = snprintf(line_buf, scr_width,
611                                 "\n" "%5u%6u %-8.8s %s%s" FMT
612                                 IF_FEATURE_TOP_SMP_PROCESS(" %3d")
613                                 IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE(FMT)
614                                 " ",
615                                 s->pid, s->ppid, get_cached_username(s->uid),
616                                 s->state, vsz_str_buf,
617                                 SHOW_STAT(pmem)
618                                 IF_FEATURE_TOP_SMP_PROCESS(, s->last_seen_on_cpu)
619                                 IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE(, SHOW_STAT(pcpu))
620                 );
621                 if ((int)(col + 1) < scr_width)
622                         read_cmdline(line_buf + col, scr_width - col, s->pid, s->comm);
623                 fputs(line_buf, stdout);
624                 /* printf(" %d/%d %lld/%lld", s->pcpu, total_pcpu,
625                         cur_jif.busy - prev_jif.busy, cur_jif.total - prev_jif.total); */
626                 s++;
627         }
628         /* printf(" %d", hist_iterations); */
629         bb_putchar(OPT_BATCH_MODE ? '\n' : '\r');
630         fflush_all();
631 }
632 #undef UPSCALE
633 #undef SHOW_STAT
634 #undef CALC_STAT
635 #undef FMT
636
637 static void clearmems(void)
638 {
639         clear_username_cache();
640         free(top);
641         top = NULL;
642         ntop = 0;
643 }
644
645 #if ENABLE_FEATURE_USE_TERMIOS
646
647 static void reset_term(void)
648 {
649         tcsetattr_stdin_TCSANOW(&initial_settings);
650         if (ENABLE_FEATURE_CLEAN_UP) {
651                 clearmems();
652 # if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
653                 free(prev_hist);
654 # endif
655         }
656 }
657
658 static void sig_catcher(int sig UNUSED_PARAM)
659 {
660         reset_term();
661         _exit(EXIT_FAILURE);
662 }
663
664 #endif /* FEATURE_USE_TERMIOS */
665
666 /*
667  * TOPMEM support
668  */
669
670 typedef unsigned long mem_t;
671
672 typedef struct topmem_status_t {
673         unsigned pid;
674         char comm[COMM_LEN];
675         /* vsz doesn't count /dev/xxx mappings except /dev/zero */
676         mem_t vsz     ;
677         mem_t vszrw   ;
678         mem_t rss     ;
679         mem_t rss_sh  ;
680         mem_t dirty   ;
681         mem_t dirty_sh;
682         mem_t stack   ;
683 } topmem_status_t;
684
685 enum { NUM_SORT_FIELD = 7 };
686
687 #define topmem ((topmem_status_t*)top)
688
689 #if ENABLE_FEATURE_TOPMEM
690
691 static int topmem_sort(char *a, char *b)
692 {
693         int n;
694         mem_t l, r;
695
696         n = offsetof(topmem_status_t, vsz) + (sort_field * sizeof(mem_t));
697         l = *(mem_t*)(a + n);
698         r = *(mem_t*)(b + n);
699         if (l == r) {
700                 l = ((topmem_status_t*)a)->dirty;
701                 r = ((topmem_status_t*)b)->dirty;
702         }
703         /* We want to avoid unsigned->signed and truncation errors */
704         /* l>r: -1, l=r: 0, l<r: 1 */
705         n = (l > r) ? -1 : (l != r);
706         return inverted ? -n : n;
707 }
708
709 /* display header info (meminfo / loadavg) */
710 static void display_topmem_header(int scr_width, int *lines_rem_p)
711 {
712         enum {
713                 TOTAL = 0, MFREE, BUF, CACHE,
714                 SWAPTOTAL, SWAPFREE, DIRTY,
715                 MWRITE, ANON, MAP, SLAB,
716                 NUM_FIELDS
717         };
718         static const char match[NUM_FIELDS][12] = {
719                 "\x09" "MemTotal:",  // TOTAL
720                 "\x08" "MemFree:",   // MFREE
721                 "\x08" "Buffers:",   // BUF
722                 "\x07" "Cached:",    // CACHE
723                 "\x0a" "SwapTotal:", // SWAPTOTAL
724                 "\x09" "SwapFree:",  // SWAPFREE
725                 "\x06" "Dirty:",     // DIRTY
726                 "\x0a" "Writeback:", // MWRITE
727                 "\x0a" "AnonPages:", // ANON
728                 "\x07" "Mapped:",    // MAP
729                 "\x05" "Slab:",      // SLAB
730         };
731         char meminfo_buf[4 * 1024];
732         const char *Z[NUM_FIELDS];
733         unsigned i;
734         int sz;
735
736         for (i = 0; i < NUM_FIELDS; i++)
737                 Z[i] = "?";
738
739         /* read memory info */
740         sz = open_read_close("meminfo", meminfo_buf, sizeof(meminfo_buf) - 1);
741         if (sz >= 0) {
742                 char *p = meminfo_buf;
743                 meminfo_buf[sz] = '\0';
744                 /* Note that fields always appear in the match[] order */
745                 for (i = 0; i < NUM_FIELDS; i++) {
746                         char *found = strstr(p, match[i] + 1);
747                         if (found) {
748                                 /* Cut "NNNN" out of "    NNNN kb" */
749                                 char *s = skip_whitespace(found + match[i][0]);
750                                 p = skip_non_whitespace(s);
751                                 *p++ = '\0';
752                                 Z[i] = s;
753                         }
754                 }
755         }
756
757         snprintf(line_buf, LINE_BUF_SIZE,
758                 "Mem total:%s anon:%s map:%s free:%s",
759                 Z[TOTAL], Z[ANON], Z[MAP], Z[MFREE]);
760         printf(OPT_BATCH_MODE ? "%.*s\n" : "\033[H\033[J%.*s\n", scr_width, line_buf);
761
762         snprintf(line_buf, LINE_BUF_SIZE,
763                 " slab:%s buf:%s cache:%s dirty:%s write:%s",
764                 Z[SLAB], Z[BUF], Z[CACHE], Z[DIRTY], Z[MWRITE]);
765         printf("%.*s\n", scr_width, line_buf);
766
767         snprintf(line_buf, LINE_BUF_SIZE,
768                 "Swap total:%s free:%s", // TODO: % used?
769                 Z[SWAPTOTAL], Z[SWAPFREE]);
770         printf("%.*s\n", scr_width, line_buf);
771
772         (*lines_rem_p) -= 3;
773 }
774
775 static void ulltoa6_and_space(unsigned long long ul, char buf[6])
776 {
777         /* see http://en.wikipedia.org/wiki/Tera */
778         smart_ulltoa5(ul, buf, " mgtpezy");
779         buf[5] = ' ';
780 }
781
782 static NOINLINE void display_topmem_process_list(int lines_rem, int scr_width)
783 {
784 #define HDR_STR "  PID   VSZ VSZRW   RSS (SHR) DIRTY (SHR) STACK"
785 #define MIN_WIDTH sizeof(HDR_STR)
786         const topmem_status_t *s = topmem;
787
788         display_topmem_header(scr_width, &lines_rem);
789         strcpy(line_buf, HDR_STR " COMMAND");
790         line_buf[5 + sort_field * 6] = '*';
791         printf(OPT_BATCH_MODE ? "%.*s" : "\e[7m%.*s\e[0m", scr_width, line_buf);
792         lines_rem--;
793
794         if (lines_rem > ntop)
795                 lines_rem = ntop;
796         while (--lines_rem >= 0) {
797                 /* PID VSZ VSZRW RSS (SHR) DIRTY (SHR) COMMAND */
798                 ulltoa6_and_space(s->pid     , &line_buf[0*6]);
799                 ulltoa6_and_space(s->vsz     , &line_buf[1*6]);
800                 ulltoa6_and_space(s->vszrw   , &line_buf[2*6]);
801                 ulltoa6_and_space(s->rss     , &line_buf[3*6]);
802                 ulltoa6_and_space(s->rss_sh  , &line_buf[4*6]);
803                 ulltoa6_and_space(s->dirty   , &line_buf[5*6]);
804                 ulltoa6_and_space(s->dirty_sh, &line_buf[6*6]);
805                 ulltoa6_and_space(s->stack   , &line_buf[7*6]);
806                 line_buf[8*6] = '\0';
807                 if (scr_width > (int)MIN_WIDTH) {
808                         read_cmdline(&line_buf[8*6], scr_width - MIN_WIDTH, s->pid, s->comm);
809                 }
810                 printf("\n""%.*s", scr_width, line_buf);
811                 s++;
812         }
813         bb_putchar(OPT_BATCH_MODE ? '\n' : '\r');
814         fflush_all();
815 #undef HDR_STR
816 #undef MIN_WIDTH
817 }
818
819 #else
820 void display_topmem_process_list(int lines_rem, int scr_width);
821 int topmem_sort(char *a, char *b);
822 #endif /* TOPMEM */
823
824 /*
825  * end TOPMEM support
826  */
827
828 enum {
829         TOP_MASK = 0
830                 | PSSCAN_PID
831                 | PSSCAN_PPID
832                 | PSSCAN_VSZ
833                 | PSSCAN_STIME
834                 | PSSCAN_UTIME
835                 | PSSCAN_STATE
836                 | PSSCAN_COMM
837                 | PSSCAN_CPU
838                 | PSSCAN_UIDGID,
839         TOPMEM_MASK = 0
840                 | PSSCAN_PID
841                 | PSSCAN_SMAPS
842                 | PSSCAN_COMM,
843 };
844
845 //usage:#if ENABLE_FEATURE_SHOW_THREADS || ENABLE_FEATURE_TOP_SMP_CPU
846 //usage:# define IF_SHOW_THREADS_OR_TOP_SMP(...) __VA_ARGS__
847 //usage:#else
848 //usage:# define IF_SHOW_THREADS_OR_TOP_SMP(...)
849 //usage:#endif
850 //usage:#define top_trivial_usage
851 //usage:       "[-b] [-nCOUNT] [-dSECONDS]" IF_FEATURE_TOPMEM(" [-m]")
852 //usage:#define top_full_usage "\n\n"
853 //usage:       "Provide a view of process activity in real time."
854 //usage:   "\n""Read the status of all processes from /proc each SECONDS"
855 //usage:   "\n""and display a screenful of them."
856 //usage:   "\n""Keys:"
857 //usage:   "\n""        N/M"
858 //usage:                IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE("/P")
859 //usage:                IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE("/T")
860 //usage:           ": " IF_FEATURE_TOPMEM("show CPU usage, ") "sort by pid/mem"
861 //usage:                IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE("/cpu")
862 //usage:                IF_FEATURE_TOP_CPU_USAGE_PERCENTAGE("/time")
863 //usage:        IF_FEATURE_TOPMEM(
864 //usage:   "\n""        S: show memory, R: reverse memory sort"
865 //usage:        )
866 //usage:        IF_SHOW_THREADS_OR_TOP_SMP(
867 //usage:   "\n""        "
868 //usage:                IF_FEATURE_SHOW_THREADS("H: toggle threads")
869 //usage:                IF_FEATURE_SHOW_THREADS(IF_FEATURE_TOP_SMP_CPU(", "))
870 //usage:                IF_FEATURE_TOP_SMP_CPU("1: toggle SMP")
871 //usage:        )
872 //usage:   "\n""        Q,^C: exit"
873
874 int top_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
875 int top_main(int argc UNUSED_PARAM, char **argv)
876 {
877         int iterations;
878         unsigned lines, col;
879         int lines_rem;
880         unsigned interval;
881         char *str_interval, *str_iterations;
882         unsigned scan_mask = TOP_MASK;
883 #if ENABLE_FEATURE_USE_TERMIOS
884         struct termios new_settings;
885         struct pollfd pfd[1];
886         unsigned char c;
887
888         pfd[0].fd = 0;
889         pfd[0].events = POLLIN;
890 #endif
891
892         INIT_G();
893
894         interval = 5; /* default update interval is 5 seconds */
895         iterations = 0; /* infinite */
896 #if ENABLE_FEATURE_TOP_SMP_CPU
897         /*num_cpus = 0;*/
898         /*smp_cpu_info = 0;*/  /* to start with show aggregate */
899         cpu_jif = &cur_jif;
900         cpu_prev_jif = &prev_jif;
901 #endif
902
903         /* all args are options; -n NUM */
904         opt_complementary = "-"; /* options can be specified w/o dash */
905         col = getopt32(argv, "d:n:b"IF_FEATURE_TOPMEM("m"), &str_interval, &str_iterations);
906 #if ENABLE_FEATURE_TOPMEM
907         if (col & OPT_m) /* -m (busybox specific) */
908                 scan_mask = TOPMEM_MASK;
909 #endif
910         if (col & OPT_d) {
911                 /* work around for "-d 1" -> "-d -1" done by getopt32
912                  * (opt_complementary == "-" does this) */
913                 if (str_interval[0] == '-')
914                         str_interval++;
915                 /* Need to limit it to not overflow poll timeout */
916                 interval = xatou16(str_interval);
917         }
918         if (col & OPT_n) {
919                 if (str_iterations[0] == '-')
920                         str_iterations++;
921                 iterations = xatou(str_iterations);
922         }
923
924         /* change to /proc */
925         xchdir("/proc");
926 #if ENABLE_FEATURE_USE_TERMIOS
927         tcgetattr(0, (void *) &initial_settings);
928         memcpy(&new_settings, &initial_settings, sizeof(new_settings));
929         /* unbuffered input, turn off echo */
930         new_settings.c_lflag &= ~(ISIG | ICANON | ECHO | ECHONL);
931
932         bb_signals(BB_FATAL_SIGS, sig_catcher);
933         tcsetattr_stdin_TCSANOW(&new_settings);
934 #endif
935
936 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
937         sort_function[0] = pcpu_sort;
938         sort_function[1] = mem_sort;
939         sort_function[2] = time_sort;
940 #else
941         sort_function[0] = mem_sort;
942 #endif
943
944         while (1) {
945                 procps_status_t *p = NULL;
946
947                 lines = 24; /* default */
948                 col = 79;
949 #if ENABLE_FEATURE_USE_TERMIOS
950                 /* We output to stdout, we need size of stdout (not stdin)! */
951                 get_terminal_width_height(STDOUT_FILENO, &col, &lines);
952                 if (lines < 5 || col < 10) {
953                         sleep(interval);
954                         continue;
955                 }
956 #endif
957                 if (col > LINE_BUF_SIZE-2) /* +2 bytes for '\n', NUL, */
958                         col = LINE_BUF_SIZE-2;
959
960                 /* read process IDs & status for all the processes */
961                 while ((p = procps_scan(p, scan_mask)) != NULL) {
962                         int n;
963 #if ENABLE_FEATURE_TOPMEM
964                         if (scan_mask != TOPMEM_MASK)
965 #endif
966                         {
967                                 n = ntop;
968                                 top = xrealloc_vector(top, 6, ntop++);
969                                 top[n].pid = p->pid;
970                                 top[n].ppid = p->ppid;
971                                 top[n].vsz = p->vsz;
972 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
973                                 top[n].ticks = p->stime + p->utime;
974 #endif
975                                 top[n].uid = p->uid;
976                                 strcpy(top[n].state, p->state);
977                                 strcpy(top[n].comm, p->comm);
978 #if ENABLE_FEATURE_TOP_SMP_PROCESS
979                                 top[n].last_seen_on_cpu = p->last_seen_on_cpu;
980 #endif
981                         }
982 #if ENABLE_FEATURE_TOPMEM
983                         else { /* TOPMEM */
984                                 if (!(p->smaps.mapped_ro | p->smaps.mapped_rw))
985                                         continue; /* kernel threads are ignored */
986                                 n = ntop;
987                                 /* No bug here - top and topmem are the same */
988                                 top = xrealloc_vector(topmem, 6, ntop++);
989                                 strcpy(topmem[n].comm, p->comm);
990                                 topmem[n].pid      = p->pid;
991                                 topmem[n].vsz      = p->smaps.mapped_rw + p->smaps.mapped_ro;
992                                 topmem[n].vszrw    = p->smaps.mapped_rw;
993                                 topmem[n].rss_sh   = p->smaps.shared_clean + p->smaps.shared_dirty;
994                                 topmem[n].rss      = p->smaps.private_clean + p->smaps.private_dirty + topmem[n].rss_sh;
995                                 topmem[n].dirty    = p->smaps.private_dirty + p->smaps.shared_dirty;
996                                 topmem[n].dirty_sh = p->smaps.shared_dirty;
997                                 topmem[n].stack    = p->smaps.stack;
998                         }
999 #endif
1000                 } /* end of "while we read /proc" */
1001                 if (ntop == 0) {
1002                         bb_error_msg("no process info in /proc");
1003                         break;
1004                 }
1005
1006                 if (scan_mask != TOPMEM_MASK) {
1007 #if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
1008                         if (!prev_hist_count) {
1009                                 do_stats();
1010                                 usleep(100000);
1011                                 clearmems();
1012                                 continue;
1013                         }
1014                         do_stats();
1015                         /* TODO: we don't need to sort all 10000 processes, we need to find top 24! */
1016                         qsort(top, ntop, sizeof(top_status_t), (void*)mult_lvl_cmp);
1017 #else
1018                         qsort(top, ntop, sizeof(top_status_t), (void*)(sort_function[0]));
1019 #endif
1020                 }
1021 #if ENABLE_FEATURE_TOPMEM
1022                 else { /* TOPMEM */
1023                         qsort(topmem, ntop, sizeof(topmem_status_t), (void*)topmem_sort);
1024                 }
1025 #endif
1026                 lines_rem = lines;
1027                 if (OPT_BATCH_MODE) {
1028                         lines_rem = INT_MAX;
1029                 }
1030                 if (scan_mask != TOPMEM_MASK)
1031                         display_process_list(lines_rem, col);
1032 #if ENABLE_FEATURE_TOPMEM
1033                 else
1034                         display_topmem_process_list(lines_rem, col);
1035 #endif
1036                 clearmems();
1037                 if (iterations >= 0 && !--iterations)
1038                         break;
1039 #if !ENABLE_FEATURE_USE_TERMIOS
1040                 sleep(interval);
1041 #else
1042                 if (option_mask32 & (OPT_b|OPT_EOF))
1043                          /* batch mode, or EOF on stdin ("top </dev/null") */
1044                         sleep(interval);
1045                 else if (safe_poll(pfd, 1, interval * 1000) > 0) {
1046                         if (safe_read(STDIN_FILENO, &c, 1) != 1) { /* error/EOF? */
1047                                 option_mask32 |= OPT_EOF;
1048                                 continue;
1049                         }
1050                         if (c == initial_settings.c_cc[VINTR])
1051                                 break;
1052                         c |= 0x20; /* lowercase */
1053                         if (c == 'q')
1054                                 break;
1055                         if (c == 'n') {
1056                                 IF_FEATURE_TOPMEM(scan_mask = TOP_MASK;)
1057                                 sort_function[0] = pid_sort;
1058                         }
1059                         if (c == 'm') {
1060                                 IF_FEATURE_TOPMEM(scan_mask = TOP_MASK;)
1061                                 sort_function[0] = mem_sort;
1062 # if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
1063                                 sort_function[1] = pcpu_sort;
1064                                 sort_function[2] = time_sort;
1065 # endif
1066                         }
1067 # if ENABLE_FEATURE_SHOW_THREADS
1068                         if (c == 'h'
1069                          IF_FEATURE_TOPMEM(&& scan_mask != TOPMEM_MASK)
1070                         ) {
1071                                 scan_mask ^= PSSCAN_TASKS;
1072                         }
1073 # endif
1074 # if ENABLE_FEATURE_TOP_CPU_USAGE_PERCENTAGE
1075                         if (c == 'p') {
1076                                 IF_FEATURE_TOPMEM(scan_mask = TOP_MASK;)
1077                                 sort_function[0] = pcpu_sort;
1078                                 sort_function[1] = mem_sort;
1079                                 sort_function[2] = time_sort;
1080                         }
1081                         if (c == 't') {
1082                                 IF_FEATURE_TOPMEM(scan_mask = TOP_MASK;)
1083                                 sort_function[0] = time_sort;
1084                                 sort_function[1] = mem_sort;
1085                                 sort_function[2] = pcpu_sort;
1086                         }
1087 #  if ENABLE_FEATURE_TOPMEM
1088                         if (c == 's') {
1089                                 scan_mask = TOPMEM_MASK;
1090                                 free(prev_hist);
1091                                 prev_hist = NULL;
1092                                 prev_hist_count = 0;
1093                                 sort_field = (sort_field + 1) % NUM_SORT_FIELD;
1094                         }
1095                         if (c == 'r')
1096                                 inverted ^= 1;
1097 #  endif
1098 #  if ENABLE_FEATURE_TOP_SMP_CPU
1099                         /* procps-2.0.18 uses 'C', 3.2.7 uses '1' */
1100                         if (c == 'c' || c == '1') {
1101                                 /* User wants to toggle per cpu <> aggregate */
1102                                 if (smp_cpu_info) {
1103                                         free(cpu_prev_jif);
1104                                         free(cpu_jif);
1105                                         cpu_jif = &cur_jif;
1106                                         cpu_prev_jif = &prev_jif;
1107                                 } else {
1108                                         /* Prepare for xrealloc() */
1109                                         cpu_jif = cpu_prev_jif = NULL;
1110                                 }
1111                                 num_cpus = 0;
1112                                 smp_cpu_info = !smp_cpu_info;
1113                                 get_jiffy_counts();
1114                         }
1115 #  endif
1116 # endif
1117                 }
1118 #endif /* FEATURE_USE_TERMIOS */
1119         } /* end of "while (1)" */
1120
1121         bb_putchar('\n');
1122 #if ENABLE_FEATURE_USE_TERMIOS
1123         reset_term();
1124 #endif
1125         return EXIT_SUCCESS;
1126 }