init.c

   1 /*
   2  * This file contains job initialization and setup functions.
   3  */
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <unistd.h>
   7 #include <fcntl.h>
   8 #include <ctype.h>
   9 #include <string.h>
  10 #include <errno.h>
  11 #include <getopt.h>
  12 #include <assert.h>
  13 #include <sys/ipc.h>
  14 #include <sys/shm.h>
  15 #include <sys/types.h>
  16 #include <sys/stat.h>
  17
  18 #include "fio.h"
  19 #include "parse.h"
  20
  21 /*
  22  * The default options
  23  */
  24 #define DEF_BS                  (4096)
  25 #define DEF_TIMEOUT             (0)
  26 #define DEF_RATE_CYCLE          (1000)
  27 #define DEF_ODIRECT             (1)
  28 #define DEF_IO_ENGINE           (FIO_SYNCIO)
  29 #define DEF_IO_ENGINE_NAME      "sync"
  30 #define DEF_SEQUENTIAL          (1)
  31 #define DEF_RAND_REPEAT         (1)
  32 #define DEF_OVERWRITE           (1)
  33 #define DEF_INVALIDATE          (1)
  34 #define DEF_SYNCIO              (0)
  35 #define DEF_RANDSEED            (0xb1899bedUL)
  36 #define DEF_BWAVGTIME           (500)
  37 #define DEF_CREATE_SER          (1)
  38 #define DEF_CREATE_FSYNC        (1)
  39 #define DEF_LOOPS               (1)
  40 #define DEF_VERIFY              (0)
  41 #define DEF_STONEWALL           (0)
  42 #define DEF_NUMJOBS             (1)
  43 #define DEF_USE_THREAD          (0)
  44 #define DEF_FILE_SIZE           (1024 * 1024 * 1024UL)
  45 #define DEF_ZONE_SIZE           (0)
  46 #define DEF_ZONE_SKIP           (0)
  47 #define DEF_RWMIX_CYCLE         (500)
  48 #define DEF_RWMIX_READ          (50)
  49 #define DEF_NICE                (0)
  50 #define DEF_NR_FILES            (1)
  51 #define DEF_UNLINK              (0)
  52 #define DEF_WRITE_BW_LOG        (0)
  53 #define DEF_WRITE_LAT_LOG       (0)
  54 #define DEF_NO_RAND_MAP         (0)
  55
  56 #define td_var_offset(var)      ((size_t) &((struct thread_data *)0)->var)
  57
  58 static int str_rw_cb(void *, const char *);
  59 static int str_ioengine_cb(void *, const char *);
  60 static int str_mem_cb(void *, const char *);
  61 static int str_verify_cb(void *, const char *);
  62 static int str_lockmem_cb(void *, unsigned long *);
  63 #ifdef FIO_HAVE_IOPRIO
  64 static int str_prio_cb(void *, unsigned int *);
  65 static int str_prioclass_cb(void *, unsigned int *);
  66 #endif
  67 static int str_exitall_cb(void);
  68 static int str_cpumask_cb(void *, unsigned int *);
  69
  70 /*
  71  * Map of job/command line options
  72  */
  73 static struct fio_option options[] = {
  74         {
  75                 .name   = "name",
  76                 .type   = FIO_OPT_STR_STORE,
  77                 .off1   = td_var_offset(name),
  78         },
  79         {
  80                 .name   = "directory",
  81                 .type   = FIO_OPT_STR_STORE,
  82                 .off1   = td_var_offset(directory),
  83         },
  84         {
  85                 .name   = "filename",
  86                 .type   = FIO_OPT_STR_STORE,
  87                 .off1   = td_var_offset(filename),
  88         },
  89         {
  90                 .name   = "rw",
  91                 .type   = FIO_OPT_STR,
  92                 .cb     = str_rw_cb,
  93         },
  94         {
  95                 .name   = "ioengine",
  96                 .type   = FIO_OPT_STR,
  97                 .cb     = str_ioengine_cb,
  98         },
  99         {
 100                 .name   = "mem",
 101                 .type   = FIO_OPT_STR,
 102                 .cb     = str_mem_cb,
 103         },
 104         {
 105                 .name   = "verify",
 106                 .type   = FIO_OPT_STR,
 107                 .cb     = str_verify_cb,
 108         },
 109         {
 110                 .name   = "write_iolog",
 111                 .type   = FIO_OPT_STR_STORE,
 112                 .off1   = td_var_offset(write_iolog_file),
 113         },
 114         {
 115                 .name   = "read_iolog",
 116                 .type   = FIO_OPT_STR_STORE,
 117                 .off1   = td_var_offset(read_iolog_file),
 118         },
 119         {
 120                 .name   = "exec_prerun",
 121                 .type   = FIO_OPT_STR_STORE,
 122                 .off1   = td_var_offset(exec_prerun),
 123         },
 124         {
 125                 .name   = "exec_postrun",
 126                 .type   = FIO_OPT_STR_STORE,
 127                 .off1   = td_var_offset(exec_postrun),
 128         },
 129 #ifdef FIO_HAVE_IOSCHED_SWITCH
 130         {
 131                 .name   = "ioscheduler",
 132                 .type   = FIO_OPT_STR_STORE,
 133                 .off1   = td_var_offset(ioscheduler),
 134         },
 135 #endif
 136         {
 137                 .name   = "size",
 138                 .type   = FIO_OPT_STR_VAL,
 139                 .off1   = td_var_offset(total_file_size),
 140         },
 141         {
 142                 .name   = "bs",
 143                 .type   = FIO_OPT_STR_VAL_INT,
 144                 .off1   = td_var_offset(bs[DDIR_READ]),
 145                 .off2   = td_var_offset(bs[DDIR_WRITE]),
 146         },
 147         {
 148                 .name   = "offset",
 149                 .type   = FIO_OPT_STR_VAL,
 150                 .off1   = td_var_offset(start_offset),
 151         },
 152         {
 153                 .name   = "zonesize",
 154                 .type   = FIO_OPT_STR_VAL,
 155                 .off1   = td_var_offset(zone_size),
 156         },
 157         {
 158                 .name   = "zoneskip",
 159                 .type   = FIO_OPT_STR_VAL,
 160                 .off1   = td_var_offset(zone_skip),
 161         },
 162         {
 163                 .name   = "lockmem",
 164                 .type   = FIO_OPT_STR_VAL,
 165                 .cb     = str_lockmem_cb,
 166         },
 167         {
 168                 .name   = "bsrange",
 169                 .type   = FIO_OPT_RANGE,
 170                 .off1   = td_var_offset(min_bs[DDIR_READ]),
 171                 .off2   = td_var_offset(max_bs[DDIR_READ]),
 172                 .off3   = td_var_offset(min_bs[DDIR_WRITE]),
 173                 .off4   = td_var_offset(max_bs[DDIR_WRITE]),
 174         },
 175         {
 176                 .name   = "nrfiles",
 177                 .type   = FIO_OPT_INT,
 178                 .off1   = td_var_offset(nr_files),
 179         },
 180         {
 181                 .name   = "iodepth",
 182                 .type   = FIO_OPT_INT,
 183                 .off1   = td_var_offset(iodepth),
 184         },
 185         {
 186                 .name   = "fsync",
 187                 .type   = FIO_OPT_INT,
 188                 .off1   = td_var_offset(fsync_blocks),
 189         },
 190         {
 191                 .name   = "rwmixcycle",
 192                 .type   = FIO_OPT_INT,
 193                 .off1   = td_var_offset(rwmixcycle),
 194         },
 195         {
 196                 .name   = "rwmixread",
 197                 .type   = FIO_OPT_INT,
 198                 .off1   = td_var_offset(rwmixread),
 199                 .max_val= 100,
 200         },
 201         {
 202                 .name   = "rwmixwrite",
 203                 .type   = FIO_OPT_INT,
 204                 .off1   = td_var_offset(rwmixwrite),
 205                 .max_val= 100,
 206         },
 207         {
 208                 .name   = "nice",
 209                 .type   = FIO_OPT_INT,
 210                 .off1   = td_var_offset(nice),
 211         },
 212 #ifdef FIO_HAVE_IOPRIO
 213         {
 214                 .name   = "prio",
 215                 .type   = FIO_OPT_INT,
 216                 .cb     = str_prio_cb,
 217         },
 218         {
 219                 .name   = "prioclass",
 220                 .type   = FIO_OPT_INT,
 221                 .cb     = str_prioclass_cb,
 222         },
 223 #endif
 224         {
 225                 .name   = "thinktime",
 226                 .type   = FIO_OPT_INT,
 227                 .off1   = td_var_offset(thinktime)
 228         },
 229         {
 230                 .name   = "rate",
 231                 .type   = FIO_OPT_INT,
 232                 .off1   = td_var_offset(rate)
 233         },
 234         {
 235                 .name   = "ratemin",
 236                 .type   = FIO_OPT_INT,
 237                 .off1   = td_var_offset(ratemin)
 238         },
 239         {
 240                 .name   = "ratecycle",
 241                 .type   = FIO_OPT_INT,
 242                 .off1   = td_var_offset(ratecycle)
 243         },
 244         {
 245                 .name   = "startdelay",
 246                 .type   = FIO_OPT_INT,
 247                 .off1   = td_var_offset(start_delay)
 248         },
 249         {
 250                 .name   = "timeout",
 251                 .type   = FIO_OPT_STR_VAL_TIME,
 252                 .off1   = td_var_offset(timeout)
 253         },
 254         {
 255                 .name   = "invalidate",
 256                 .type   = FIO_OPT_INT,
 257                 .off1   = td_var_offset(invalidate_cache)
 258         },
 259         {
 260                 .name   = "sync",
 261                 .type   = FIO_OPT_INT,
 262                 .off1   = td_var_offset(sync_io)
 263         },
 264         {
 265                 .name   = "bwavgtime",
 266                 .type   = FIO_OPT_INT,
 267                 .off1   = td_var_offset(bw_avg_time)
 268         },
 269         {
 270                 .name   = "create_serialize",
 271                 .type   = FIO_OPT_INT,
 272                 .off1   = td_var_offset(create_serialize)
 273         },
 274         {
 275                 .name   = "create_fsync",
 276                 .type   = FIO_OPT_INT,
 277                 .off1   = td_var_offset(create_fsync)
 278         },
 279         {
 280                 .name   = "loops",
 281                 .type   = FIO_OPT_INT,
 282                 .off1   = td_var_offset(loops)
 283         },
 284         {
 285                 .name   = "numjobs",
 286                 .type   = FIO_OPT_INT,
 287                 .off1   = td_var_offset(numjobs)
 288         },
 289         {
 290                 .name   = "cpuload",
 291                 .type   = FIO_OPT_INT,
 292                 .off1   = td_var_offset(cpuload)
 293         },
 294         {
 295                 .name   = "cpuchunks",
 296                 .type   = FIO_OPT_INT,
 297                 .off1   = td_var_offset(cpucycle)
 298         },
 299         {
 300                 .name   = "direct",
 301                 .type   = FIO_OPT_INT,
 302                 .off1   = td_var_offset(odirect)
 303         },
 304         {
 305                 .name   = "overwrite",
 306                 .type   = FIO_OPT_INT,
 307                 .off1   = td_var_offset(overwrite)
 308         },
 309 #ifdef FIO_HAVE_CPU_AFFINITY
 310         {
 311                 .name   = "cpumask",
 312                 .type   = FIO_OPT_INT,
 313                 .cb     = str_cpumask_cb,
 314         },
 315 #endif
 316         {
 317                 .name   = "end_fsync",
 318                 .type   = FIO_OPT_INT,
 319                 .off1   = td_var_offset(end_fsync)
 320         },
 321         {
 322                 .name   = "unlink",
 323                 .type   = FIO_OPT_STR_SET,
 324                 .off1   = td_var_offset(unlink),
 325         },
 326         {
 327                 .name   = "exitall",
 328                 .type   = FIO_OPT_STR_SET,
 329                 .cb     = str_exitall_cb,
 330         },
 331         {
 332                 .name   = "stonewall",
 333                 .type   = FIO_OPT_STR_SET,
 334                 .off1   = td_var_offset(stonewall),
 335         },
 336         {
 337                 .name   = "thread",
 338                 .type   = FIO_OPT_STR_SET,
 339                 .off1   = td_var_offset(thread),
 340         },
 341         {
 342                 .name   = "write_bw_log",
 343                 .type   = FIO_OPT_STR_SET,
 344                 .off1   = td_var_offset(write_bw_log),
 345         },
 346         {
 347                 .name   = "write_lat_log",
 348                 .type   = FIO_OPT_STR_SET,
 349                 .off1   = td_var_offset(write_lat_log),
 350         },
 351         {
 352                 .name   = "norandommap",
 353                 .type   = FIO_OPT_STR_SET,
 354                 .off1   = td_var_offset(norandommap),
 355         },
 356         {
 357                 .name   = "bs_unaligned",
 358                 .type   = FIO_OPT_STR_SET,
 359                 .off1   = td_var_offset(bs_unaligned),
 360         },
 361         {
 362                 .name = NULL,
 363         },
 364 };
 365
 366 #define FIO_JOB_OPTS    (sizeof(options) / sizeof(struct fio_option))
 367 #define FIO_CMD_OPTS    (16)
 368 #define FIO_GETOPT_JOB  (0x89988998)
 369
 370 /*
 371  * Command line options. These will contain the above, plus a few
 372  * extra that only pertain to fio itself and not jobs.
 373  */
 374 static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
 375         {
 376                 .name           = "output",
 377                 .has_arg        = required_argument,
 378                 .val            = 'o',
 379         },
 380         {
 381                 .name           = "timeout",
 382                 .has_arg        = required_argument,
 383                 .val            = 't',
 384         },
 385         {
 386                 .name           = "latency-log",
 387                 .has_arg        = required_argument,
 388                 .val            = 'l',
 389         },
 390         {
 391                 .name           = "bandwidth-log",
 392                 .has_arg        = required_argument,
 393                 .val            = 'b',
 394         },
 395         {
 396                 .name           = "minimal",
 397                 .has_arg        = optional_argument,
 398                 .val            = 'm',
 399         },
 400         {
 401                 .name           = "version",
 402                 .has_arg        = no_argument,
 403                 .val            = 'v',
 404         },
 405         {
 406                 .name           = NULL,
 407         },
 408 };
 409
 410 static int def_timeout = DEF_TIMEOUT;
 411
 412 static char fio_version_string[] = "fio 1.7";
 413
 414 static char **ini_file;
 415 static int max_jobs = MAX_JOBS;
 416
 417 struct thread_data def_thread;
 418 struct thread_data *threads = NULL;
 419
 420 int rate_quit = 0;
 421 int exitall_on_terminate = 0;
 422 int terse_output = 0;
 423 unsigned long long mlock_size = 0;
 424 FILE *f_out = NULL;
 425 FILE *f_err = NULL;
 426
 427 static int write_lat_log = DEF_WRITE_LAT_LOG;
 428 static int write_bw_log = DEF_WRITE_BW_LOG;
 429
 430 /*
 431  * Return a free job structure.
 432  */
 433 static struct thread_data *get_new_job(int global, struct thread_data *parent)
 434 {
 435         struct thread_data *td;
 436
 437         if (global)
 438                 return &def_thread;
 439         if (thread_number >= max_jobs)
 440                 return NULL;
 441
 442         td = &threads[thread_number++];
 443         *td = *parent;
 444
 445         td->thread_number = thread_number;
 446         return td;
 447 }
 448
 449 static void put_job(struct thread_data *td)
 450 {
 451         if (td == &def_thread)
 452                 return;
 453
 454         memset(&threads[td->thread_number - 1], 0, sizeof(*td));
 455         thread_number--;
 456 }
 457
 458 /*
 459  * Lazy way of fixing up options that depend on each other. We could also
 460  * define option callback handlers, but this is easier.
 461  */
 462 static void fixup_options(struct thread_data *td)
 463 {
 464         if (!td->rwmixread && td->rwmixwrite)
 465                 td->rwmixread = 100 - td->rwmixwrite;
 466
 467         if (td->write_iolog_file && td->read_iolog_file) {
 468                 log_err("fio: read iolog overrides write_iolog\n");
 469                 free(td->write_iolog_file);
 470                 td->write_iolog_file = NULL;
 471         }
 472
 473         if (td->io_ops->flags & FIO_SYNCIO)
 474                 td->iodepth = 1;
 475         else {
 476                 if (!td->iodepth)
 477                         td->iodepth = td->nr_files;
 478         }
 479
 480         /*
 481          * only really works for sequential io for now, and with 1 file
 482          */
 483         if (td->zone_size && !td->sequential && td->nr_files == 1)
 484                 td->zone_size = 0;
 485
 486         /*
 487          * Reads can do overwrites, we always need to pre-create the file
 488          */
 489         if (td_read(td) || td_rw(td))
 490                 td->overwrite = 1;
 491
 492         if (!td->min_bs[DDIR_READ])
 493                 td->min_bs[DDIR_READ]= td->bs[DDIR_READ];
 494         if (!td->max_bs[DDIR_READ])
 495                 td->max_bs[DDIR_READ] = td->bs[DDIR_READ];
 496         if (!td->min_bs[DDIR_WRITE])
 497                 td->min_bs[DDIR_WRITE]= td->bs[DDIR_WRITE];
 498         if (!td->max_bs[DDIR_WRITE])
 499                 td->max_bs[DDIR_WRITE] = td->bs[DDIR_WRITE];
 500
 501         td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]);
 502
 503         if (td_read(td) && !td_rw(td))
 504                 td->verify = 0;
 505
 506         if (td->norandommap && td->verify != VERIFY_NONE) {
 507                 log_err("fio: norandommap given, verify disabled\n");
 508                 td->verify = VERIFY_NONE;
 509         }
 510         if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO))
 511                 log_err("fio: bs_unaligned may not work with raw io\n");
 512 }
 513
 514 /*
 515  * Adds a job to the list of things todo. Sanitizes the various options
 516  * to make sure we don't have conflicts, and initializes various
 517  * members of td.
 518  */
 519 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 520 {
 521         char *ddir_str[] = { "read", "write", "randread", "randwrite",
 522                              "rw", NULL, "randrw" };
 523         struct stat sb;
 524         int numjobs, ddir, i;
 525         struct fio_file *f;
 526
 527         /*
 528          * the def_thread is just for options, it's not a real job
 529          */
 530         if (td == &def_thread)
 531                 return 0;
 532
 533         /*
 534          * Set default io engine, if none set
 535          */
 536         if (!td->io_ops) {
 537                 td->io_ops = load_ioengine(td, DEF_IO_ENGINE_NAME);
 538                 if (!td->io_ops) {
 539                         log_err("default engine %s not there?\n", DEF_IO_ENGINE_NAME);
 540                         return 1;
 541                 }
 542         }
 543
 544         if (td->odirect)
 545                 td->io_ops->flags |= FIO_RAWIO;
 546
 547         fixup_options(td);
 548
 549         td->filetype = FIO_TYPE_FILE;
 550         if (!stat(jobname, &sb)) {
 551                 if (S_ISBLK(sb.st_mode))
 552                         td->filetype = FIO_TYPE_BD;
 553                 else if (S_ISCHR(sb.st_mode))
 554                         td->filetype = FIO_TYPE_CHAR;
 555         }
 556
 557         if (td->filename)
 558                 td->nr_uniq_files = 1;
 559         else
 560                 td->nr_uniq_files = td->nr_files;
 561
 562         if (td->filetype == FIO_TYPE_FILE || td->filename) {
 563                 char tmp[PATH_MAX];
 564                 int len = 0;
 565                 int i;
 566
 567                 if (td->directory && td->directory[0] != '\0')
 568                         sprintf(tmp, "%s/", td->directory);
 569
 570                 td->files = malloc(sizeof(struct fio_file) * td->nr_files);
 571
 572                 for_each_file(td, f, i) {
 573                         memset(f, 0, sizeof(*f));
 574                         f->fd = -1;
 575
 576                         if (td->filename)
 577                                 sprintf(tmp + len, "%s", td->filename);
 578                         else
 579                                 sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i);
 580                         f->file_name = strdup(tmp);
 581                 }
 582         } else {
 583                 td->nr_files = 1;
 584                 td->files = malloc(sizeof(struct fio_file));
 585                 f = &td->files[0];
 586
 587                 memset(f, 0, sizeof(*f));
 588                 f->fd = -1;
 589                 f->file_name = strdup(jobname);
 590         }
 591
 592         for_each_file(td, f, i) {
 593                 f->file_size = td->total_file_size / td->nr_files;
 594                 f->file_offset = td->start_offset;
 595         }
 596
 597         fio_sem_init(&td->mutex, 0);
 598
 599         td->clat_stat[0].min_val = td->clat_stat[1].min_val = ULONG_MAX;
 600         td->slat_stat[0].min_val = td->slat_stat[1].min_val = ULONG_MAX;
 601         td->bw_stat[0].min_val = td->bw_stat[1].min_val = ULONG_MAX;
 602
 603         if (td->stonewall && td->thread_number > 1)
 604                 groupid++;
 605
 606         td->groupid = groupid;
 607
 608         if (setup_rate(td))
 609                 goto err;
 610
 611         if (td->write_lat_log) {
 612                 setup_log(&td->slat_log);
 613                 setup_log(&td->clat_log);
 614         }
 615         if (td->write_bw_log)
 616                 setup_log(&td->bw_log);
 617
 618         if (!td->name)
 619                 td->name = strdup(jobname);
 620
 621         ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
 622
 623         if (!terse_output) {
 624                 if (!job_add_num) {
 625                         if (td->io_ops->flags & FIO_CPUIO)
 626                                 fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
 627                         else
 628                                 fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d/%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs[DDIR_READ], td->max_bs[DDIR_READ], td->min_bs[DDIR_WRITE], td->max_bs[DDIR_WRITE], td->rate, td->io_ops->name, td->iodepth);
 629                 } else if (job_add_num == 1)
 630                         fprintf(f_out, "...\n");
 631         }
 632
 633         /*
 634          * recurse add identical jobs, clear numjobs and stonewall options
 635          * as they don't apply to sub-jobs
 636          */
 637         numjobs = td->numjobs;
 638         while (--numjobs) {
 639                 struct thread_data *td_new = get_new_job(0, td);
 640
 641                 if (!td_new)
 642                         goto err;
 643
 644                 td_new->numjobs = 1;
 645                 td_new->stonewall = 0;
 646                 job_add_num = numjobs - 1;
 647
 648                 if (add_job(td_new, jobname, job_add_num))
 649                         goto err;
 650         }
 651         return 0;
 652 err:
 653         put_job(td);
 654         return -1;
 655 }
 656
 657 /*
 658  * Initialize the various random states we need (random io, block size ranges,
 659  * read/write mix, etc).
 660  */
 661 int init_random_state(struct thread_data *td)
 662 {
 663         unsigned long seeds[4];
 664         int fd, num_maps, blocks, i;
 665         struct fio_file *f;
 666
 667         if (td->io_ops->flags & FIO_CPUIO)
 668                 return 0;
 669
 670         fd = open("/dev/urandom", O_RDONLY);
 671         if (fd == -1) {
 672                 td_verror(td, errno);
 673                 return 1;
 674         }
 675
 676         if (read(fd, seeds, sizeof(seeds)) < (int) sizeof(seeds)) {
 677                 td_verror(td, EIO);
 678                 close(fd);
 679                 return 1;
 680         }
 681
 682         close(fd);
 683
 684         os_random_seed(seeds[0], &td->bsrange_state);
 685         os_random_seed(seeds[1], &td->verify_state);
 686         os_random_seed(seeds[2], &td->rwmix_state);
 687
 688         if (td->sequential)
 689                 return 0;
 690
 691         if (td->rand_repeatable)
 692                 seeds[3] = DEF_RANDSEED;
 693
 694         if (!td->norandommap) {
 695                 for_each_file(td, f, i) {
 696                         blocks = (f->file_size + td->rw_min_bs - 1) / td->rw_min_bs;
 697                         num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP;
 698                         f->file_map = malloc(num_maps * sizeof(long));
 699                         f->num_maps = num_maps;
 700                         memset(f->file_map, 0, num_maps * sizeof(long));
 701                 }
 702         }
 703
 704         os_random_seed(seeds[3], &td->random_state);
 705         return 0;
 706 }
 707
 708 static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu)
 709 {
 710 #ifdef FIO_HAVE_CPU_AFFINITY
 711         unsigned int i;
 712
 713         CPU_ZERO(&cpumask);
 714
 715         for (i = 0; i < sizeof(int) * 8; i++) {
 716                 if ((1 << i) & cpu)
 717                         CPU_SET(i, &cpumask);
 718         }
 719 #endif
 720 }
 721
 722 static int is_empty_or_comment(char *line)
 723 {
 724         unsigned int i;
 725
 726         for (i = 0; i < strlen(line); i++) {
 727                 if (line[i] == ';')
 728                         return 1;
 729                 if (!isspace(line[i]) && !iscntrl(line[i]))
 730                         return 0;
 731         }
 732
 733         return 1;
 734 }
 735
 736 static int str_rw_cb(void *data, const char *mem)
 737 {
 738         struct thread_data *td = data;
 739
 740         if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) {
 741                 td->ddir = DDIR_READ;
 742                 td->sequential = 1;
 743                 return 0;
 744         } else if (!strncmp(mem, "randread", 8)) {
 745                 td->ddir = DDIR_READ;
 746                 td->sequential = 0;
 747                 return 0;
 748         } else if (!strncmp(mem, "write", 5) || !strncmp(mem, "1", 1)) {
 749                 td->ddir = DDIR_WRITE;
 750                 td->sequential = 1;
 751                 return 0;
 752         } else if (!strncmp(mem, "randwrite", 9)) {
 753                 td->ddir = DDIR_WRITE;
 754                 td->sequential = 0;
 755                 return 0;
 756         } else if (!strncmp(mem, "rw", 2)) {
 757                 td->ddir = 0;
 758                 td->iomix = 1;
 759                 td->sequential = 1;
 760                 return 0;
 761         } else if (!strncmp(mem, "randrw", 6)) {
 762                 td->ddir = 0;
 763                 td->iomix = 1;
 764                 td->sequential = 0;
 765                 return 0;
 766         }
 767
 768         log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n");
 769         return 1;
 770 }
 771
 772 static int str_verify_cb(void *data, const char *mem)
 773 {
 774         struct thread_data *td = data;
 775
 776         if (!strncmp(mem, "0", 1)) {
 777                 td->verify = VERIFY_NONE;
 778                 return 0;
 779         } else if (!strncmp(mem, "md5", 3) || !strncmp(mem, "1", 1)) {
 780                 td->verify = VERIFY_MD5;
 781                 return 0;
 782         } else if (!strncmp(mem, "crc32", 5)) {
 783                 td->verify = VERIFY_CRC32;
 784                 return 0;
 785         }
 786
 787         log_err("fio: verify types: md5, crc32\n");
 788         return 1;
 789 }
 790
 791 static int str_mem_cb(void *data, const char *mem)
 792 {
 793         struct thread_data *td = data;
 794
 795         if (!strncmp(mem, "malloc", 6)) {
 796                 td->mem_type = MEM_MALLOC;
 797                 return 0;
 798         } else if (!strncmp(mem, "shm", 3)) {
 799                 td->mem_type = MEM_SHM;
 800                 return 0;
 801         } else if (!strncmp(mem, "mmap", 4)) {
 802                 td->mem_type = MEM_MMAP;
 803                 return 0;
 804         }
 805
 806         log_err("fio: mem type: malloc, shm, mmap\n");
 807         return 1;
 808 }
 809
 810 static int str_ioengine_cb(void *data, const char *str)
 811 {
 812         struct thread_data *td = data;
 813
 814         td->io_ops = load_ioengine(td, str);
 815         if (td->io_ops)
 816                 return 0;
 817
 818         log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice, cpu\n");
 819         return 1;
 820 }
 821
 822 static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
 823 {
 824         mlock_size = *val;
 825         return 0;
 826 }
 827
 828 #ifdef FIO_HAVE_IOPRIO
 829 static int str_prioclass_cb(void *data, unsigned int *val)
 830 {
 831         struct thread_data *td = data;
 832
 833         td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
 834         return 0;
 835 }
 836
 837 static int str_prio_cb(void *data, unsigned int *val)
 838 {
 839         struct thread_data *td = data;
 840
 841         td->ioprio |= *val;
 842         return 0;
 843 }
 844 #endif
 845
 846 static int str_exitall_cb(void)
 847 {
 848         exitall_on_terminate = 1;
 849         return 0;
 850 }
 851
 852 static int str_cpumask_cb(void *data, unsigned int *val)
 853 {
 854         struct thread_data *td = data;
 855
 856         fill_cpu_mask(td->cpumask, *val);
 857         return 0;
 858 }
 859
 860 /*
 861  * This is our [ini] type file parser.
 862  */
 863 int parse_jobs_ini(char *file, int stonewall_flag)
 864 {
 865         unsigned int global;
 866         struct thread_data *td;
 867         char *string, *name;
 868         fpos_t off;
 869         FILE *f;
 870         char *p;
 871         int ret = 0, stonewall;
 872
 873         f = fopen(file, "r");
 874         if (!f) {
 875                 perror("fopen job file");
 876                 return 1;
 877         }
 878
 879         string = malloc(4096);
 880         name = malloc(256);
 881         memset(name, 0, 256);
 882
 883         stonewall = stonewall_flag;
 884         do {
 885                 p = fgets(string, 4095, f);
 886                 if (!p)
 887                         break;
 888                 if (is_empty_or_comment(p))
 889                         continue;
 890                 if (sscanf(p, "[%255s]", name) != 1)
 891                         continue;
 892
 893                 global = !strncmp(name, "global", 6);
 894
 895                 name[strlen(name) - 1] = '\0';
 896
 897                 td = get_new_job(global, &def_thread);
 898                 if (!td) {
 899                         ret = 1;
 900                         break;
 901                 }
 902
 903                 /*
 904                  * Seperate multiple job files by a stonewall
 905                  */
 906                 if (!global && stonewall) {
 907                         td->stonewall = stonewall;
 908                         stonewall = 0;
 909                 }
 910
 911                 fgetpos(f, &off);
 912                 while ((p = fgets(string, 4096, f)) != NULL) {
 913                         if (is_empty_or_comment(p))
 914                                 continue;
 915
 916                         strip_blank_front(&p);
 917
 918                         if (p[0] == '[')
 919                                 break;
 920
 921                         strip_blank_end(p);
 922
 923                         fgetpos(f, &off);
 924
 925                         /*
 926                          * Don't break here, continue parsing options so we
 927                          * dump all the bad ones. Makes trial/error fixups
 928                          * easier on the user.
 929                          */
 930                         ret |= parse_option(p, options, td);
 931                 }
 932
 933                 if (!ret) {
 934                         fsetpos(f, &off);
 935                         ret = add_job(td, name, 0);
 936                 } else {
 937                         log_err("fio: job %s dropped\n", name);
 938                         put_job(td);
 939                 }
 940         } while (!ret);
 941
 942         free(string);
 943         free(name);
 944         fclose(f);
 945         return ret;
 946 }
 947
 948 static int fill_def_thread(void)
 949 {
 950         memset(&def_thread, 0, sizeof(def_thread));
 951
 952         if (fio_getaffinity(getpid(), &def_thread.cpumask) == -1) {
 953                 perror("sched_getaffinity");
 954                 return 1;
 955         }
 956
 957         /*
 958          * fill globals
 959          */
 960         def_thread.ddir = DDIR_READ;
 961         def_thread.iomix = 0;
 962         def_thread.bs[DDIR_READ] = DEF_BS;
 963         def_thread.bs[DDIR_WRITE] = DEF_BS;
 964         def_thread.min_bs[DDIR_READ] = def_thread.min_bs[DDIR_WRITE] = 0;
 965         def_thread.max_bs[DDIR_READ] = def_thread.max_bs[DDIR_WRITE] = 0;
 966         def_thread.odirect = DEF_ODIRECT;
 967         def_thread.ratecycle = DEF_RATE_CYCLE;
 968         def_thread.sequential = DEF_SEQUENTIAL;
 969         def_thread.timeout = def_timeout;
 970         def_thread.overwrite = DEF_OVERWRITE;
 971         def_thread.invalidate_cache = DEF_INVALIDATE;
 972         def_thread.sync_io = DEF_SYNCIO;
 973         def_thread.mem_type = MEM_MALLOC;
 974         def_thread.bw_avg_time = DEF_BWAVGTIME;
 975         def_thread.create_serialize = DEF_CREATE_SER;
 976         def_thread.create_fsync = DEF_CREATE_FSYNC;
 977         def_thread.loops = DEF_LOOPS;
 978         def_thread.verify = DEF_VERIFY;
 979         def_thread.stonewall = DEF_STONEWALL;
 980         def_thread.numjobs = DEF_NUMJOBS;
 981         def_thread.use_thread = DEF_USE_THREAD;
 982         def_thread.rwmixcycle = DEF_RWMIX_CYCLE;
 983         def_thread.rwmixread = DEF_RWMIX_READ;
 984         def_thread.nice = DEF_NICE;
 985         def_thread.rand_repeatable = DEF_RAND_REPEAT;
 986         def_thread.nr_files = DEF_NR_FILES;
 987         def_thread.unlink = DEF_UNLINK;
 988         def_thread.write_bw_log = write_bw_log;
 989         def_thread.write_lat_log = write_lat_log;
 990         def_thread.norandommap = DEF_NO_RAND_MAP;
 991 #ifdef FIO_HAVE_DISK_UTIL
 992         def_thread.do_disk_util = 1;
 993 #endif
 994
 995         return 0;
 996 }
 997
 998 static void usage(void)
 999 {
1000         printf("%s\n", fio_version_string);
1001         printf("\t--output\tWrite output to file\n");
1002         printf("\t--timeout\tRuntime in seconds\n");
1003         printf("\t--latency-log\tGenerate per-job latency logs\n");
1004         printf("\t--bandwidth-log\tGenerate per-job bandwidth logs\n");
1005         printf("\t--minimal\tMinimal (terse) output\n");
1006         printf("\t--version\tPrint version info and exit\n");
1007 }
1008
1009 static int parse_cmd_line(int argc, char *argv[])
1010 {
1011         struct thread_data *td = NULL;
1012         int c, ini_idx = 0, lidx, ret;
1013
1014         while ((c = getopt_long(argc, argv, "", long_options, &lidx)) != -1) {
1015                 switch (c) {
1016                 case 't':
1017                         def_timeout = atoi(optarg);
1018                         break;
1019                 case 'l':
1020                         write_lat_log = 1;
1021                         break;
1022                 case 'w':
1023                         write_bw_log = 1;
1024                         break;
1025                 case 'o':
1026                         f_out = fopen(optarg, "w+");
1027                         if (!f_out) {
1028                                 perror("fopen output");
1029                                 exit(1);
1030                         }
1031                         f_err = f_out;
1032                         break;
1033                 case 'm':
1034                         terse_output = 1;
1035                         break;
1036                 case 'h':
1037                         usage();
1038                         exit(0);
1039                 case 'v':
1040                         printf("%s\n", fio_version_string);
1041                         exit(0);
1042                 case FIO_GETOPT_JOB: {
1043                         const char *opt = long_options[lidx].name;
1044                         char *val = optarg;
1045
1046                         if (!strncmp(opt, "name", 4) && td) {
1047                                 ret = add_job(td, td->name ?: "fio", 0);
1048                                 if (ret) {
1049                                         put_job(td);
1050                                         return 0;
1051                                 }
1052                                 td = NULL;
1053                         }
1054                         if (!td) {
1055                                 int global = !strncmp(val, "global", 6);
1056
1057                                 td = get_new_job(global, &def_thread);
1058                                 if (!td)
1059                                         return 0;
1060                         }
1061
1062                         ret = parse_cmd_option(opt, val, options, td);
1063                         if (ret) {
1064                                 log_err("fio: job dropped\n");
1065                                 put_job(td);
1066                                 td = NULL;
1067                         }
1068                         break;
1069                 }
1070                 default:
1071                         printf("optarg <<%s>>\n", argv[optind]);
1072                         break;
1073                 }
1074         }
1075
1076         if (td) {
1077                 ret = add_job(td, td->name ?: "fio", 0);
1078                 if (ret)
1079                         put_job(td);
1080         }
1081
1082         while (optind < argc) {
1083                 ini_idx++;
1084                 ini_file = realloc(ini_file, ini_idx * sizeof(char *));
1085                 ini_file[ini_idx - 1] = strdup(argv[optind]);
1086                 optind++;
1087         }
1088
1089         return ini_idx;
1090 }
1091
1092 static void free_shm(void)
1093 {
1094         struct shmid_ds sbuf;
1095
1096         if (threads) {
1097                 shmdt((void *) threads);
1098                 threads = NULL;
1099                 shmctl(shm_id, IPC_RMID, &sbuf);
1100         }
1101 }
1102
1103 /*
1104  * The thread area is shared between the main process and the job
1105  * threads/processes. So setup a shared memory segment that will hold
1106  * all the job info.
1107  */
1108 static int setup_thread_area(void)
1109 {
1110         /*
1111          * 1024 is too much on some machines, scale max_jobs if
1112          * we get a failure that looks like too large a shm segment
1113          */
1114         do {
1115                 size_t size = max_jobs * sizeof(struct thread_data);
1116
1117                 shm_id = shmget(0, size, IPC_CREAT | 0600);
1118                 if (shm_id != -1)
1119                         break;
1120                 if (errno != EINVAL) {
1121                         perror("shmget");
1122                         break;
1123                 }
1124
1125                 max_jobs >>= 1;
1126         } while (max_jobs);
1127
1128         if (shm_id == -1)
1129                 return 1;
1130
1131         threads = shmat(shm_id, NULL, 0);
1132         if (threads == (void *) -1) {
1133                 perror("shmat");
1134                 return 1;
1135         }
1136
1137         atexit(free_shm);
1138         return 0;
1139 }
1140
1141 /*
1142  * Copy the fio options into the long options map, so we mirror
1143  * job and cmd line options.
1144  */
1145 static void dupe_job_options(void)
1146 {
1147         struct fio_option *o;
1148         unsigned int i;
1149
1150         i = 0;
1151         while (long_options[i].name)
1152                 i++;
1153
1154         o = &options[0];
1155         while (o->name) {
1156                 long_options[i].name = o->name;
1157                 long_options[i].val = FIO_GETOPT_JOB;
1158                 if (o->type == FIO_OPT_STR_SET)
1159                         long_options[i].has_arg = no_argument;
1160                 else
1161                         long_options[i].has_arg = required_argument;
1162
1163                 i++;
1164                 o++;
1165                 assert(i < FIO_JOB_OPTS + FIO_CMD_OPTS);
1166         }
1167 }
1168
1169 int parse_options(int argc, char *argv[])
1170 {
1171         int job_files, i;
1172
1173         f_out = stdout;
1174         f_err = stderr;
1175
1176         dupe_job_options();
1177
1178         if (setup_thread_area())
1179                 return 1;
1180         if (fill_def_thread())
1181                 return 1;
1182
1183         job_files = parse_cmd_line(argc, argv);
1184
1185         for (i = 0; i < job_files; i++) {
1186                 if (fill_def_thread())
1187                         return 1;
1188                 if (parse_jobs_ini(ini_file[i], i))
1189                         return 1;
1190                 free(ini_file[i]);
1191         }
1192
1193         free(ini_file);
1194
1195         if (!thread_number) {
1196                 log_err("No jobs defined(s)\n");
1197                 return 1;
1198         }
1199
1200         return 0;
1201 }