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         fio_gettime(&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         fio_gettime(&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 fio_file *f;
 191         struct stat st;
 192         char foo[PATH_MAX], tmp[PATH_MAX];
 193         dev_t dev;
 194         char *p;
 195
 196         if (!td->do_disk_util)
 197                 return;
 198
 199         /*
 200          * Just use the same file, they are on the same device.
 201          */
 202         f = &td->files[0];
 203         if (!stat(f->file_name, &st)) {
 204                 if (S_ISBLK(st.st_mode))
 205                         dev = st.st_rdev;
 206                 else
 207                         dev = st.st_dev;
 208         } else {
 209                 /*
 210                  * must be a file, open "." in that path
 211                  */
 212                 strncpy(foo, f->file_name, PATH_MAX - 1);
 213                 p = dirname(foo);
 214                 if (stat(p, &st)) {
 215                         perror("disk util stat");
 216                         return;
 217                 }
 218
 219                 dev = st.st_dev;
 220         }
 221
 222         if (disk_util_exists(dev))
 223                 return;
 224
 225         sprintf(foo, "/sys/block");
 226         if (!find_block_dir(dev, foo))
 227                 return;
 228
 229         /*
 230          * If there's a ../queue/ directory there, we are inside a partition.
 231          * Check if that is the case and jump back. For loop/md/dm etc we
 232          * are already in the right spot.
 233          */
 234         sprintf(tmp, "%s/../queue", foo);
 235         if (!stat(tmp, &st)) {
 236                 p = dirname(foo);
 237                 sprintf(tmp, "%s/queue", p);
 238                 if (stat(tmp, &st)) {
 239                         log_err("unknown sysfs layout\n");
 240                         return;
 241                 }
 242                 strncpy(tmp, p, PATH_MAX - 1);
 243                 sprintf(foo, "%s", tmp);
 244         }
 245
 246         if (td->ioscheduler)
 247                 td->sysfs_root = strdup(foo);
 248
 249         disk_util_add(dev, foo);
 250 }
 251
 252 void disk_util_timer_arm(void)
 253 {
 254         itimer.it_value.tv_sec = 0;
 255         itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
 256         setitimer(ITIMER_REAL, &itimer, NULL);
 257 }
 258
 259 void update_rusage_stat(struct thread_data *td)
 260 {
 261         getrusage(RUSAGE_SELF, &td->ru_end);
 262
 263         td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
 264         td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
 265         td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
 266
 267
 268         memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
 269 }
 270
 271 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
 272                     double *mean, double *dev)
 273 {
 274         double n;
 275
 276         if (is->samples == 0)
 277                 return 0;
 278
 279         *min = is->min_val;
 280         *max = is->max_val;
 281
 282         n = (double) is->samples;
 283         *mean = (double) is->val / n;
 284         *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1));
 285
 286         return 1;
 287 }
 288
 289 static void show_group_stats(struct group_run_stats *rs, int id)
 290 {
 291         fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
 292
 293         if (rs->max_run[DDIR_READ])
 294                 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]);
 295         if (rs->max_run[DDIR_WRITE])
 296                 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]);
 297 }
 298
 299 static void show_disk_util(void)
 300 {
 301         struct disk_util_stat *dus;
 302         struct list_head *entry, *next;
 303         struct disk_util *du;
 304         double util;
 305
 306         fprintf(f_out, "\nDisk stats (read/write):\n");
 307
 308         list_for_each(entry, &disk_list) {
 309                 du = list_entry(entry, struct disk_util, list);
 310                 dus = &du->dus;
 311
 312                 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
 313                 if (util > 100.0)
 314                         util = 100.0;
 315
 316                 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);
 317         }
 318
 319         /*
 320          * now free the list
 321          */
 322         list_for_each_safe(entry, next, &disk_list) {
 323                 list_del(entry);
 324                 du = list_entry(entry, struct disk_util, list);
 325                 free(du->name);
 326                 free(du);
 327         }
 328 }
 329
 330 static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
 331                              int ddir)
 332 {
 333         const char *ddir_str[] = { "read ", "write" };
 334         unsigned long min, max;
 335         unsigned long long bw;
 336         double mean, dev;
 337
 338         if (!td->runtime[ddir])
 339                 return;
 340
 341         bw = td->io_bytes[ddir] / td->runtime[ddir];
 342         fprintf(f_out, "  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
 343
 344         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
 345                 fprintf(f_out, "    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
 346
 347         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
 348                 fprintf(f_out, "    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
 349
 350         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
 351                 double p_of_agg;
 352
 353                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
 354                 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);
 355         }
 356 }
 357
 358 static void show_thread_status(struct thread_data *td,
 359                                struct group_run_stats *rs)
 360 {
 361         double usr_cpu, sys_cpu;
 362         unsigned long runtime;
 363
 364         if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
 365                 return;
 366
 367         fprintf(f_out, "%s: (groupid=%d): err=%2d:\n",td->name, td->groupid, td->error);
 368
 369         show_ddir_status(td, rs, td->ddir);
 370         if (td->io_bytes[td->ddir ^ 1])
 371                 show_ddir_status(td, rs, td->ddir ^ 1);
 372
 373         runtime = mtime_since(&td->epoch, &td->end_time);
 374         if (runtime) {
 375                 double runt = (double) runtime;
 376
 377                 usr_cpu = (double) td->usr_time * 100 / runt;
 378                 sys_cpu = (double) td->sys_time * 100 / runt;
 379         } else {
 380                 usr_cpu = 0;
 381                 sys_cpu = 0;
 382         }
 383
 384         fprintf(f_out, "  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
 385 }
 386
 387 static void show_ddir_status_terse(struct thread_data *td,
 388                                    struct group_run_stats *rs, int ddir)
 389 {
 390         unsigned long min, max;
 391         unsigned long long bw;
 392         double mean, dev;
 393
 394         bw = 0;
 395         if (td->runtime[ddir])
 396                 bw = td->io_bytes[ddir] / td->runtime[ddir];
 397
 398         fprintf(f_out, ",%llu,%llu,%lu", td->io_bytes[ddir] >> 10, bw, td->runtime[ddir]);
 399
 400         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
 401                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
 402         else
 403                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
 404
 405         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
 406                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
 407         else
 408                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
 409
 410         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
 411                 double p_of_agg;
 412
 413                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
 414                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", min, max, p_of_agg, mean, dev);
 415         } else
 416                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", 0UL, 0UL, 0.0, 0.0, 0.0);
 417
 418 }
 419
 420
 421 static void show_thread_status_terse(struct thread_data *td,
 422                                      struct group_run_stats *rs)
 423 {
 424         double usr_cpu, sys_cpu;
 425
 426         fprintf(f_out, "%s,%d,%d",td->name, td->groupid, td->error);
 427
 428         show_ddir_status_terse(td, rs, 0);
 429         show_ddir_status_terse(td, rs, 1);
 430
 431         if (td->runtime[0] + td->runtime[1]) {
 432                 double runt = (double) (td->runtime[0] + td->runtime[1]);
 433
 434                 usr_cpu = (double) td->usr_time * 100 / runt;
 435                 sys_cpu = (double) td->sys_time * 100 / runt;
 436         } else {
 437                 usr_cpu = 0;
 438                 sys_cpu = 0;
 439         }
 440
 441         fprintf(f_out, ",%f%%,%f%%,%lu\n", usr_cpu, sys_cpu, td->ctx);
 442 }
 443
 444 void show_run_stats(void)
 445 {
 446         struct group_run_stats *runstats, *rs;
 447         struct thread_data *td;
 448         int i;
 449
 450         runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
 451
 452         for (i = 0; i < groupid + 1; i++) {
 453                 rs = &runstats[i];
 454
 455                 memset(rs, 0, sizeof(*rs));
 456                 rs->min_bw[0] = rs->min_run[0] = ~0UL;
 457                 rs->min_bw[1] = rs->min_run[1] = ~0UL;
 458         }
 459
 460         for_each_td(td, i) {
 461                 unsigned long long rbw, wbw;
 462
 463                 if (td->error) {
 464                         fprintf(f_out, "%s: %s\n", td->name, td->verror);
 465                         continue;
 466                 }
 467
 468                 rs = &runstats[td->groupid];
 469
 470                 if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
 471                         rs->min_run[0] = td->runtime[0];
 472                 if (td->runtime[0] > rs->max_run[0])
 473                         rs->max_run[0] = td->runtime[0];
 474                 if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
 475                         rs->min_run[1] = td->runtime[1];
 476                 if (td->runtime[1] > rs->max_run[1])
 477                         rs->max_run[1] = td->runtime[1];
 478
 479                 rbw = wbw = 0;
 480                 if (td->runtime[0])
 481                         rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
 482                 if (td->runtime[1])
 483                         wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
 484
 485                 if (rbw < rs->min_bw[0])
 486                         rs->min_bw[0] = rbw;
 487                 if (wbw < rs->min_bw[1])
 488                         rs->min_bw[1] = wbw;
 489                 if (rbw > rs->max_bw[0])
 490                         rs->max_bw[0] = rbw;
 491                 if (wbw > rs->max_bw[1])
 492                         rs->max_bw[1] = wbw;
 493
 494                 rs->io_kb[0] += td->io_bytes[0] >> 10;
 495                 rs->io_kb[1] += td->io_bytes[1] >> 10;
 496         }
 497
 498         for (i = 0; i < groupid + 1; i++) {
 499                 rs = &runstats[i];
 500
 501                 if (rs->max_run[0])
 502                         rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
 503                 if (rs->max_run[1])
 504                         rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
 505         }
 506
 507         /*
 508          * don't overwrite last signal output
 509          */
 510         if (!terse_output)
 511                 printf("\n");
 512
 513         for_each_td(td, i) {
 514                 rs = &runstats[td->groupid];
 515
 516                 if (terse_output)
 517                         show_thread_status_terse(td, rs);
 518                 else
 519                         show_thread_status(td, rs);
 520         }
 521
 522         if (!terse_output) {
 523                 for (i = 0; i < groupid + 1; i++)
 524                         show_group_stats(&runstats[i], i);
 525
 526                 show_disk_util();
 527         }
 528
 529         free(runstats);
 530 }
 531
 532 static inline void add_stat_sample(struct io_stat *is, unsigned long val)
 533 {
 534         if (val > is->max_val)
 535                 is->max_val = val;
 536         if (val < is->min_val)
 537                 is->min_val = val;
 538
 539         is->val += val;
 540         is->val_sq += val * val;
 541         is->samples++;
 542 }
 543
 544 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
 545                            unsigned long val, enum fio_ddir ddir)
 546 {
 547         if (iolog->nr_samples == iolog->max_samples) {
 548                 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
 549
 550                 iolog->log = realloc(iolog->log, new_size);
 551                 iolog->max_samples <<= 1;
 552         }
 553
 554         iolog->log[iolog->nr_samples].val = val;
 555         iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch);
 556         iolog->log[iolog->nr_samples].ddir = ddir;
 557         iolog->nr_samples++;
 558 }
 559
 560 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
 561                      unsigned long msec)
 562 {
 563         add_stat_sample(&td->clat_stat[ddir], msec);
 564
 565         if (td->clat_log)
 566                 add_log_sample(td, td->clat_log, msec, ddir);
 567 }
 568
 569 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
 570                      unsigned long msec)
 571 {
 572         add_stat_sample(&td->slat_stat[ddir], msec);
 573
 574         if (td->slat_log)
 575                 add_log_sample(td, td->slat_log, msec, ddir);
 576 }
 577
 578 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
 579                    struct timeval *t)
 580 {
 581         unsigned long spent = mtime_since(&td->stat_sample_time[ddir], t);
 582         unsigned long rate;
 583
 584         if (spent < td->bw_avg_time)
 585                 return;
 586
 587         rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
 588         add_stat_sample(&td->bw_stat[ddir], rate);
 589
 590         if (td->bw_log)
 591                 add_log_sample(td, td->bw_log, rate, ddir);
 592
 593         fio_gettime(&td->stat_sample_time[ddir], NULL);
 594         td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
 595 }
 596
 597