[PATCH] Allow io engine to do the file setup
[fio.git] / stat.c
1 #include <stdio.h>
2 #include <string.h>
3 #include <sys/time.h>
4 #include <sys/types.h>
5 #include <sys/stat.h>
6 #include <dirent.h>
7 #include <libgen.h>
8 #include <math.h>
9
10 #include "fio.h"
11
12 static struct itimerval itimer;
13 static struct list_head disk_list = LIST_HEAD_INIT(disk_list);
14
15 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
16 {
17         unsigned in_flight;
18         char line[256];
19         FILE *f;
20         char *p;
21
22         f = fopen(du->path, "r");
23         if (!f)
24                 return 1;
25
26         p = fgets(line, sizeof(line), f);
27         if (!p) {
28                 fclose(f);
29                 return 1;
30         }
31
32         if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
33                 fclose(f);
34                 return 1;
35         }
36
37         fclose(f);
38         return 0;
39 }
40
41 static void update_io_tick_disk(struct disk_util *du)
42 {
43         struct disk_util_stat __dus, *dus, *ldus;
44         struct timeval t;
45
46         if (get_io_ticks(du, &__dus))
47                 return;
48
49         dus = &du->dus;
50         ldus = &du->last_dus;
51
52         dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
53         dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
54         dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
55         dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
56         dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
57         dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
58         dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
59         dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
60         dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
61         dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
62
63         gettimeofday(&t, NULL);
64         du->msec += mtime_since(&du->time, &t);
65         memcpy(&du->time, &t, sizeof(t));
66         memcpy(ldus, &__dus, sizeof(__dus));
67 }
68
69 void update_io_ticks(void)
70 {
71         struct list_head *entry;
72         struct disk_util *du;
73
74         list_for_each(entry, &disk_list) {
75                 du = list_entry(entry, struct disk_util, list);
76                 update_io_tick_disk(du);
77         }
78 }
79
80 static int disk_util_exists(dev_t dev)
81 {
82         struct list_head *entry;
83         struct disk_util *du;
84
85         list_for_each(entry, &disk_list) {
86                 du = list_entry(entry, struct disk_util, list);
87
88                 if (du->dev == dev)
89                         return 1;
90         }
91
92         return 0;
93 }
94
95 static void disk_util_add(dev_t dev, char *path)
96 {
97         struct disk_util *du = malloc(sizeof(*du));
98
99         memset(du, 0, sizeof(*du));
100         INIT_LIST_HEAD(&du->list);
101         sprintf(du->path, "%s/stat", path);
102         du->name = strdup(basename(path));
103         du->dev = dev;
104
105         gettimeofday(&du->time, NULL);
106         get_io_ticks(du, &du->last_dus);
107
108         list_add_tail(&du->list, &disk_list);
109 }
110
111 static int check_dev_match(dev_t dev, char *path)
112 {
113         unsigned int major, minor;
114         char line[256], *p;
115         FILE *f;
116
117         f = fopen(path, "r");
118         if (!f) {
119                 perror("open path");
120                 return 1;
121         }
122
123         p = fgets(line, sizeof(line), f);
124         if (!p) {
125                 fclose(f);
126                 return 1;
127         }
128
129         if (sscanf(p, "%u:%u", &major, &minor) != 2) {
130                 fclose(f);
131                 return 1;
132         }
133
134         if (((major << 8) | minor) == dev) {
135                 fclose(f);
136                 return 0;
137         }
138
139         fclose(f);
140         return 1;
141 }
142
143 static int find_block_dir(dev_t dev, char *path)
144 {
145         struct dirent *dir;
146         struct stat st;
147         int found = 0;
148         DIR *D;
149
150         D = opendir(path);
151         if (!D)
152                 return 0;
153
154         while ((dir = readdir(D)) != NULL) {
155                 char full_path[256];
156
157                 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
158                         continue;
159
160                 sprintf(full_path, "%s/%s", path, dir->d_name);
161
162                 if (!strcmp(dir->d_name, "dev")) {
163                         if (!check_dev_match(dev, full_path)) {
164                                 found = 1;
165                                 break;
166                         }
167                 }
168
169                 if (lstat(full_path, &st) == -1) {
170                         perror("stat");
171                         break;
172                 }
173
174                 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
175                         continue;
176
177                 found = find_block_dir(dev, full_path);
178                 if (found) {
179                         strcpy(path, full_path);
180                         break;
181                 }
182         }
183
184         closedir(D);
185         return found;
186 }
187
188 void init_disk_util(struct thread_data *td)
189 {
190         struct stat st;
191         char foo[256], tmp[256];
192         dev_t dev;
193         char *p;
194
195         if (!td->do_disk_util)
196                 return;
197
198         if (!stat(td->file_name, &st)) {
199                 if (S_ISBLK(st.st_mode))
200                         dev = st.st_rdev;
201                 else
202                         dev = st.st_dev;
203         } else {
204                 /*
205                  * must be a file, open "." in that path
206                  */
207                 strcpy(foo, td->file_name);
208                 p = dirname(foo);
209                 if (stat(p, &st)) {
210                         perror("disk util stat");
211                         return;
212                 }
213
214                 dev = st.st_dev;
215         }
216
217         if (disk_util_exists(dev))
218                 return;
219                 
220         sprintf(foo, "/sys/block");
221         if (!find_block_dir(dev, foo))
222                 return;
223
224         /*
225          * If there's a ../queue/ directory there, we are inside a partition.
226          * Check if that is the case and jump back. For loop/md/dm etc we
227          * are already in the right spot.
228          */
229         sprintf(tmp, "%s/../queue", foo);
230         if (!stat(tmp, &st)) {
231                 p = dirname(foo);
232                 sprintf(tmp, "%s/queue", p);
233                 if (stat(tmp, &st)) {
234                         log_err("unknown sysfs layout\n");
235                         return;
236                 }
237                 sprintf(foo, "%s", p);
238         }
239
240         td->sysfs_root = strdup(foo);
241         disk_util_add(dev, foo);
242 }
243
244 void disk_util_timer_arm(void)
245 {
246         itimer.it_value.tv_sec = 0;
247         itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
248         setitimer(ITIMER_REAL, &itimer, NULL);
249 }
250
251 void update_rusage_stat(struct thread_data *td)
252 {
253         if (!(td->runtime[0] + td->runtime[1]))
254                 return;
255
256         getrusage(RUSAGE_SELF, &td->ru_end);
257
258         td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
259         td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
260         td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
261
262         
263         memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
264 }
265
266 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
267                     double *mean, double *dev)
268 {
269         double n;
270
271         if (is->samples == 0)
272                 return 0;
273
274         *min = is->min_val;
275         *max = is->max_val;
276
277         n = (double) is->samples;
278         *mean = (double) is->val / n;
279         *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1));
280
281         return 1;
282 }
283
284 static void show_group_stats(struct group_run_stats *rs, int id)
285 {
286         fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
287
288         if (rs->max_run[DDIR_READ])
289                 fprintf(f_out, "   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
290         if (rs->max_run[DDIR_WRITE])
291                 fprintf(f_out, "  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
292 }
293
294 static void show_disk_util(void)
295 {
296         struct disk_util_stat *dus;
297         struct list_head *entry;
298         struct disk_util *du;
299         double util;
300
301         fprintf(f_out, "\nDisk stats (read/write):\n");
302
303         list_for_each(entry, &disk_list) {
304                 du = list_entry(entry, struct disk_util, list);
305                 dus = &du->dus;
306
307                 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
308                 if (util > 100.0)
309                         util = 100.0;
310
311                 fprintf(f_out, "  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
312         }
313 }
314
315 static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
316                              int ddir)
317 {
318         char *ddir_str[] = { "read ", "write" };
319         unsigned long min, max;
320         unsigned long long bw;
321         double mean, dev;
322
323         if (!td->runtime[ddir])
324                 return;
325
326         bw = td->io_bytes[ddir] / td->runtime[ddir];
327         fprintf(f_out, "  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
328
329         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
330                 fprintf(f_out, "    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
331
332         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
333                 fprintf(f_out, "    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
334
335         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
336                 double p_of_agg;
337
338                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
339                 fprintf(f_out, "    bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
340         }
341 }
342
343 static void show_thread_status(struct thread_data *td,
344                                struct group_run_stats *rs)
345 {
346         double usr_cpu, sys_cpu;
347
348         if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
349                 return;
350
351         fprintf(f_out, "%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error);
352
353         show_ddir_status(td, rs, td->ddir);
354         if (td->io_bytes[td->ddir ^ 1])
355                 show_ddir_status(td, rs, td->ddir ^ 1);
356
357         if (td->runtime[0] + td->runtime[1]) {
358                 double runt = td->runtime[0] + td->runtime[1];
359
360                 usr_cpu = (double) td->usr_time * 100 / runt;
361                 sys_cpu = (double) td->sys_time * 100 / runt;
362         } else {
363                 usr_cpu = 0;
364                 sys_cpu = 0;
365         }
366
367         fprintf(f_out, "  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
368 }
369
370 static void show_ddir_status_terse(struct thread_data *td,
371                                    struct group_run_stats *rs, int ddir)
372 {
373         unsigned long min, max;
374         unsigned long long bw;
375         double mean, dev;
376
377         bw = 0;
378         if (td->runtime[ddir])
379                 bw = td->io_bytes[ddir] / td->runtime[ddir];
380
381         fprintf(f_out, ",%llu,%llu,%lu", td->io_bytes[ddir] >> 10, bw, td->runtime[ddir]);
382
383         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
384                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
385         else
386                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
387
388         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
389                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
390         else
391                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
392
393         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
394                 double p_of_agg;
395
396                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
397                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", min, max, p_of_agg, mean, dev);
398         } else
399                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", 0UL, 0UL, 0.0, 0.0, 0.0);
400                 
401 }
402
403
404 static void show_thread_status_terse(struct thread_data *td,
405                                      struct group_run_stats *rs)
406 {
407         double usr_cpu, sys_cpu;
408
409         fprintf(f_out, "%s,%d,%d",td->name, td->groupid, td->error);
410
411         show_ddir_status_terse(td, rs, 0);
412         show_ddir_status_terse(td, rs, 1);
413
414         if (td->runtime[0] + td->runtime[1]) {
415                 double runt = td->runtime[0] + td->runtime[1];
416
417                 usr_cpu = (double) td->usr_time * 100 / runt;
418                 sys_cpu = (double) td->sys_time * 100 / runt;
419         } else {
420                 usr_cpu = 0;
421                 sys_cpu = 0;
422         }
423
424         fprintf(f_out, ",%f%%,%f%%,%lu\n", usr_cpu, sys_cpu, td->ctx);
425 }
426
427 void show_run_stats(void)
428 {
429         struct group_run_stats *runstats, *rs;
430         struct thread_data *td;
431         int i;
432
433         runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
434
435         for (i = 0; i < groupid + 1; i++) {
436                 rs = &runstats[i];
437
438                 memset(rs, 0, sizeof(*rs));
439                 rs->min_bw[0] = rs->min_run[0] = ~0UL;
440                 rs->min_bw[1] = rs->min_run[1] = ~0UL;
441         }
442
443         for (i = 0; i < thread_number; i++) {
444                 unsigned long long rbw, wbw;
445
446                 td = &threads[i];
447
448                 if (td->error) {
449                         fprintf(f_out, "%s: %s\n", td->name, td->verror);
450                         continue;
451                 }
452
453                 rs = &runstats[td->groupid];
454
455                 if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
456                         rs->min_run[0] = td->runtime[0];
457                 if (td->runtime[0] > rs->max_run[0])
458                         rs->max_run[0] = td->runtime[0];
459                 if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
460                         rs->min_run[1] = td->runtime[1];
461                 if (td->runtime[1] > rs->max_run[1])
462                         rs->max_run[1] = td->runtime[1];
463
464                 rbw = wbw = 0;
465                 if (td->runtime[0])
466                         rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
467                 if (td->runtime[1])
468                         wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
469
470                 if (rbw < rs->min_bw[0])
471                         rs->min_bw[0] = rbw;
472                 if (wbw < rs->min_bw[1])
473                         rs->min_bw[1] = wbw;
474                 if (rbw > rs->max_bw[0])
475                         rs->max_bw[0] = rbw;
476                 if (wbw > rs->max_bw[1])
477                         rs->max_bw[1] = wbw;
478
479                 rs->io_kb[0] += td->io_bytes[0] >> 10;
480                 rs->io_kb[1] += td->io_bytes[1] >> 10;
481         }
482
483         for (i = 0; i < groupid + 1; i++) {
484                 rs = &runstats[i];
485
486                 if (rs->max_run[0])
487                         rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
488                 if (rs->max_run[1])
489                         rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
490         }
491
492         /*
493          * don't overwrite last signal output
494          */
495         if (!terse_output)
496                 printf("\n");
497
498         for (i = 0; i < thread_number; i++) {
499                 td = &threads[i];
500                 rs = &runstats[td->groupid];
501
502                 if (terse_output)
503                         show_thread_status_terse(td, rs);
504                 else
505                         show_thread_status(td, rs);
506         }
507
508         if (!terse_output) {
509                 for (i = 0; i < groupid + 1; i++)
510                         show_group_stats(&runstats[i], i);
511
512                 show_disk_util();
513         }
514 }
515
516 static inline void add_stat_sample(struct io_stat *is, unsigned long val)
517 {
518         if (val > is->max_val)
519                 is->max_val = val;
520         if (val < is->min_val)
521                 is->min_val = val;
522
523         is->val += val;
524         is->val_sq += val * val;
525         is->samples++;
526 }
527
528 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
529                            unsigned long val, int ddir)
530 {
531         if (iolog->nr_samples == iolog->max_samples) {
532                 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
533
534                 iolog->log = realloc(iolog->log, new_size);
535                 iolog->max_samples <<= 1;
536         }
537
538         iolog->log[iolog->nr_samples].val = val;
539         iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch);
540         iolog->log[iolog->nr_samples].ddir = ddir;
541         iolog->nr_samples++;
542 }
543
544 void add_clat_sample(struct thread_data *td, int ddir, unsigned long msec)
545 {
546         add_stat_sample(&td->clat_stat[ddir], msec);
547
548         if (td->clat_log)
549                 add_log_sample(td, td->clat_log, msec, ddir);
550 }
551
552 void add_slat_sample(struct thread_data *td, int ddir, unsigned long msec)
553 {
554         add_stat_sample(&td->slat_stat[ddir], msec);
555
556         if (td->slat_log)
557                 add_log_sample(td, td->slat_log, msec, ddir);
558 }
559
560 void add_bw_sample(struct thread_data *td, int ddir)
561 {
562         unsigned long spent = mtime_since_now(&td->stat_sample_time[ddir]);
563         unsigned long rate;
564
565         if (spent < td->bw_avg_time)
566                 return;
567
568         rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
569         add_stat_sample(&td->bw_stat[ddir], rate);
570
571         if (td->bw_log)
572                 add_log_sample(td, td->bw_log, rate, ddir);
573
574         gettimeofday(&td->stat_sample_time[ddir], NULL);
575         td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
576 }
577
578