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         },
 146         {
 147                 .name   = "read_bs",
 148                 .type   = FIO_OPT_STR_VAL_INT,
 149                 .off1   = td_var_offset(bs[DDIR_READ]),
 150         },
 151         {
 152                 .name   = "write_bs",
 153                 .type   = FIO_OPT_STR_VAL_INT,
 154                 .off1   = td_var_offset(bs[DDIR_WRITE]),
 155         },
 156         {
 157                 .name   = "offset",
 158                 .type   = FIO_OPT_STR_VAL,
 159                 .off1   = td_var_offset(start_offset),
 160         },
 161         {
 162                 .name   = "zonesize",
 163                 .type   = FIO_OPT_STR_VAL,
 164                 .off1   = td_var_offset(zone_size),
 165         },
 166         {
 167                 .name   = "zoneskip",
 168                 .type   = FIO_OPT_STR_VAL,
 169                 .off1   = td_var_offset(zone_skip),
 170         },
 171         {
 172                 .name   = "lockmem",
 173                 .type   = FIO_OPT_STR_VAL,
 174                 .cb     = str_lockmem_cb,
 175         },
 176         {
 177                 .name   = "bsrange",
 178                 .type   = FIO_OPT_RANGE,
 179                 .off1   = td_var_offset(min_bs[DDIR_READ]),
 180                 .off2   = td_var_offset(max_bs[DDIR_READ]),
 181         },
 182         {
 183                 .name   = "read_bsrange",
 184                 .type   = FIO_OPT_RANGE,
 185                 .off1   = td_var_offset(min_bs[DDIR_READ]),
 186                 .off2   = td_var_offset(max_bs[DDIR_READ]),
 187         },
 188         {
 189                 .name   = "write_bsrange",
 190                 .type   = FIO_OPT_RANGE,
 191                 .off1   = td_var_offset(min_bs[DDIR_WRITE]),
 192                 .off2   = td_var_offset(max_bs[DDIR_WRITE]),
 193         },
 194         {
 195                 .name   = "nrfiles",
 196                 .type   = FIO_OPT_INT,
 197                 .off1   = td_var_offset(nr_files),
 198         },
 199         {
 200                 .name   = "iodepth",
 201                 .type   = FIO_OPT_INT,
 202                 .off1   = td_var_offset(iodepth),
 203         },
 204         {
 205                 .name   = "fsync",
 206                 .type   = FIO_OPT_INT,
 207                 .off1   = td_var_offset(fsync_blocks),
 208         },
 209         {
 210                 .name   = "rwmixcycle",
 211                 .type   = FIO_OPT_INT,
 212                 .off1   = td_var_offset(rwmixcycle),
 213         },
 214         {
 215                 .name   = "rwmixread",
 216                 .type   = FIO_OPT_INT,
 217                 .off1   = td_var_offset(rwmixread),
 218                 .max_val= 100,
 219         },
 220         {
 221                 .name   = "rwmixwrite",
 222                 .type   = FIO_OPT_INT,
 223                 .off1   = td_var_offset(rwmixwrite),
 224                 .max_val= 100,
 225         },
 226         {
 227                 .name   = "nice",
 228                 .type   = FIO_OPT_INT,
 229                 .off1   = td_var_offset(nice),
 230         },
 231 #ifdef FIO_HAVE_IOPRIO
 232         {
 233                 .name   = "prio",
 234                 .type   = FIO_OPT_INT,
 235                 .cb     = str_prio_cb,
 236         },
 237         {
 238                 .name   = "prioclass",
 239                 .type   = FIO_OPT_INT,
 240                 .cb     = str_prioclass_cb,
 241         },
 242 #endif
 243         {
 244                 .name   = "thinktime",
 245                 .type   = FIO_OPT_INT,
 246                 .off1   = td_var_offset(thinktime)
 247         },
 248         {
 249                 .name   = "rate",
 250                 .type   = FIO_OPT_INT,
 251                 .off1   = td_var_offset(rate)
 252         },
 253         {
 254                 .name   = "ratemin",
 255                 .type   = FIO_OPT_INT,
 256                 .off1   = td_var_offset(ratemin)
 257         },
 258         {
 259                 .name   = "ratecycle",
 260                 .type   = FIO_OPT_INT,
 261                 .off1   = td_var_offset(ratecycle)
 262         },
 263         {
 264                 .name   = "startdelay",
 265                 .type   = FIO_OPT_INT,
 266                 .off1   = td_var_offset(start_delay)
 267         },
 268         {
 269                 .name   = "timeout",
 270                 .type   = FIO_OPT_STR_VAL_TIME,
 271                 .off1   = td_var_offset(timeout)
 272         },
 273         {
 274                 .name   = "invalidate",
 275                 .type   = FIO_OPT_INT,
 276                 .off1   = td_var_offset(invalidate_cache)
 277         },
 278         {
 279                 .name   = "sync",
 280                 .type   = FIO_OPT_INT,
 281                 .off1   = td_var_offset(sync_io)
 282         },
 283         {
 284                 .name   = "bwavgtime",
 285                 .type   = FIO_OPT_INT,
 286                 .off1   = td_var_offset(bw_avg_time)
 287         },
 288         {
 289                 .name   = "create_serialize",
 290                 .type   = FIO_OPT_INT,
 291                 .off1   = td_var_offset(create_serialize)
 292         },
 293         {
 294                 .name   = "create_fsync",
 295                 .type   = FIO_OPT_INT,
 296                 .off1   = td_var_offset(create_fsync)
 297         },
 298         {
 299                 .name   = "loops",
 300                 .type   = FIO_OPT_INT,
 301                 .off1   = td_var_offset(loops)
 302         },
 303         {
 304                 .name   = "numjobs",
 305                 .type   = FIO_OPT_INT,
 306                 .off1   = td_var_offset(numjobs)
 307         },
 308         {
 309                 .name   = "cpuload",
 310                 .type   = FIO_OPT_INT,
 311                 .off1   = td_var_offset(cpuload)
 312         },
 313         {
 314                 .name   = "cpuchunks",
 315                 .type   = FIO_OPT_INT,
 316                 .off1   = td_var_offset(cpucycle)
 317         },
 318         {
 319                 .name   = "direct",
 320                 .type   = FIO_OPT_INT,
 321                 .off1   = td_var_offset(odirect)
 322         },
 323         {
 324                 .name   = "overwrite",
 325                 .type   = FIO_OPT_INT,
 326                 .off1   = td_var_offset(overwrite)
 327         },
 328 #ifdef FIO_HAVE_CPU_AFFINITY
 329         {
 330                 .name   = "cpumask",
 331                 .type   = FIO_OPT_INT,
 332                 .cb     = str_cpumask_cb,
 333         },
 334 #endif
 335         {
 336                 .name   = "end_fsync",
 337                 .type   = FIO_OPT_INT,
 338                 .off1   = td_var_offset(end_fsync)
 339         },
 340         {
 341                 .name   = "unlink",
 342                 .type   = FIO_OPT_STR_SET,
 343                 .off1   = td_var_offset(unlink),
 344         },
 345         {
 346                 .name   = "exitall",
 347                 .type   = FIO_OPT_STR_SET,
 348                 .cb     = str_exitall_cb,
 349         },
 350         {
 351                 .name   = "stonewall",
 352                 .type   = FIO_OPT_STR_SET,
 353                 .off1   = td_var_offset(stonewall),
 354         },
 355         {
 356                 .name   = "thread",
 357                 .type   = FIO_OPT_STR_SET,
 358                 .off1   = td_var_offset(thread),
 359         },
 360         {
 361                 .name   = "write_bw_log",
 362                 .type   = FIO_OPT_STR_SET,
 363                 .off1   = td_var_offset(write_bw_log),
 364         },
 365         {
 366                 .name   = "write_lat_log",
 367                 .type   = FIO_OPT_STR_SET,
 368                 .off1   = td_var_offset(write_lat_log),
 369         },
 370         {
 371                 .name   = "norandommap",
 372                 .type   = FIO_OPT_STR_SET,
 373                 .off1   = td_var_offset(norandommap),
 374         },
 375         {
 376                 .name   = "bs_unaligned",
 377                 .type   = FIO_OPT_STR_SET,
 378                 .off1   = td_var_offset(bs_unaligned),
 379         },
 380         {
 381                 .name = NULL,
 382         },
 383 };
 384
 385 #define FIO_JOB_OPTS    (sizeof(options) / sizeof(struct fio_option))
 386 #define FIO_CMD_OPTS    (16)
 387 #define FIO_GETOPT_JOB  (0x89988998)
 388
 389 /*
 390  * Command line options. These will contain the above, plus a few
 391  * extra that only pertain to fio itself and not jobs.
 392  */
 393 static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
 394         {
 395                 .name           = "output",
 396                 .has_arg        = required_argument,
 397                 .val            = 'o',
 398         },
 399         {
 400                 .name           = "timeout",
 401                 .has_arg        = required_argument,
 402                 .val            = 't',
 403         },
 404         {
 405                 .name           = "latency-log",
 406                 .has_arg        = required_argument,
 407                 .val            = 'l',
 408         },
 409         {
 410                 .name           = "bandwidth-log",
 411                 .has_arg        = required_argument,
 412                 .val            = 'b',
 413         },
 414         {
 415                 .name           = "minimal",
 416                 .has_arg        = optional_argument,
 417                 .val            = 'm',
 418         },
 419         {
 420                 .name           = "version",
 421                 .has_arg        = no_argument,
 422                 .val            = 'v',
 423         },
 424         {
 425                 .name           = NULL,
 426         },
 427 };
 428
 429 static int def_timeout = DEF_TIMEOUT;
 430
 431 static char fio_version_string[] = "fio 1.7";
 432
 433 static char **ini_file;
 434 static int max_jobs = MAX_JOBS;
 435
 436 struct thread_data def_thread;
 437 struct thread_data *threads = NULL;
 438
 439 int rate_quit = 0;
 440 int exitall_on_terminate = 0;
 441 int terse_output = 0;
 442 unsigned long long mlock_size = 0;
 443 FILE *f_out = NULL;
 444 FILE *f_err = NULL;
 445
 446 static int write_lat_log = DEF_WRITE_LAT_LOG;
 447 static int write_bw_log = DEF_WRITE_BW_LOG;
 448
 449 /*
 450  * Return a free job structure.
 451  */
 452 static struct thread_data *get_new_job(int global, struct thread_data *parent)
 453 {
 454         struct thread_data *td;
 455
 456         if (global)
 457                 return &def_thread;
 458         if (thread_number >= max_jobs)
 459                 return NULL;
 460
 461         td = &threads[thread_number++];
 462         *td = *parent;
 463
 464         td->thread_number = thread_number;
 465         return td;
 466 }
 467
 468 static void put_job(struct thread_data *td)
 469 {
 470         if (td == &def_thread)
 471                 return;
 472
 473         memset(&threads[td->thread_number - 1], 0, sizeof(*td));
 474         thread_number--;
 475 }
 476
 477 /*
 478  * Lazy way of fixing up options that depend on each other. We could also
 479  * define option callback handlers, but this is easier.
 480  */
 481 static void fixup_options(struct thread_data *td)
 482 {
 483         if (!td->rwmixread && td->rwmixwrite)
 484                 td->rwmixread = 100 - td->rwmixwrite;
 485
 486         if (td->write_iolog_file && td->read_iolog_file) {
 487                 log_err("fio: read iolog overrides write_iolog\n");
 488                 free(td->write_iolog_file);
 489                 td->write_iolog_file = NULL;
 490         }
 491
 492         if (td->io_ops->flags & FIO_SYNCIO)
 493                 td->iodepth = 1;
 494         else {
 495                 if (!td->iodepth)
 496                         td->iodepth = td->nr_files;
 497         }
 498
 499         /*
 500          * only really works for sequential io for now, and with 1 file
 501          */
 502         if (td->zone_size && !td->sequential && td->nr_files == 1)
 503                 td->zone_size = 0;
 504
 505         /*
 506          * Reads can do overwrites, we always need to pre-create the file
 507          */
 508         if (td_read(td) || td_rw(td))
 509                 td->overwrite = 1;
 510
 511         if (td->bs[DDIR_READ] != DEF_BS)
 512                 td->bs[DDIR_WRITE] = td->bs[DDIR_READ];
 513         if (!td->min_bs[DDIR_READ])
 514                 td->min_bs[DDIR_READ]= td->bs[DDIR_READ];
 515         if (!td->max_bs[DDIR_READ])
 516                 td->max_bs[DDIR_READ] = td->bs[DDIR_READ];
 517         if (!td->min_bs[DDIR_WRITE])
 518                 td->min_bs[DDIR_WRITE]= td->bs[DDIR_WRITE];
 519         if (!td->max_bs[DDIR_WRITE])
 520                 td->max_bs[DDIR_WRITE] = td->bs[DDIR_WRITE];
 521
 522         td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]);
 523
 524         if (td_read(td) && !td_rw(td))
 525                 td->verify = 0;
 526
 527         if (td->norandommap && td->verify != VERIFY_NONE) {
 528                 log_err("fio: norandommap given, verify disabled\n");
 529                 td->verify = VERIFY_NONE;
 530         }
 531         if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO))
 532                 log_err("fio: bs_unaligned may not work with raw io\n");
 533 }
 534
 535 /*
 536  * Adds a job to the list of things todo. Sanitizes the various options
 537  * to make sure we don't have conflicts, and initializes various
 538  * members of td.
 539  */
 540 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 541 {
 542         char *ddir_str[] = { "read", "write", "randread", "randwrite",
 543                              "rw", NULL, "randrw" };
 544         struct stat sb;
 545         int numjobs, ddir, i;
 546         struct fio_file *f;
 547
 548         /*
 549          * the def_thread is just for options, it's not a real job
 550          */
 551         if (td == &def_thread)
 552                 return 0;
 553
 554         /*
 555          * Set default io engine, if none set
 556          */
 557         if (!td->io_ops) {
 558                 td->io_ops = load_ioengine(td, DEF_IO_ENGINE_NAME);
 559                 if (!td->io_ops) {
 560                         log_err("default engine %s not there?\n", DEF_IO_ENGINE_NAME);
 561                         return 1;
 562                 }
 563         }
 564
 565         if (td->odirect)
 566                 td->io_ops->flags |= FIO_RAWIO;
 567
 568         fixup_options(td);
 569
 570         td->filetype = FIO_TYPE_FILE;
 571         if (!stat(jobname, &sb)) {
 572                 if (S_ISBLK(sb.st_mode))
 573                         td->filetype = FIO_TYPE_BD;
 574                 else if (S_ISCHR(sb.st_mode))
 575                         td->filetype = FIO_TYPE_CHAR;
 576         }
 577
 578         if (td->filename)
 579                 td->nr_uniq_files = 1;
 580         else
 581                 td->nr_uniq_files = td->nr_files;
 582
 583         if (td->filetype == FIO_TYPE_FILE || td->filename) {
 584                 char tmp[PATH_MAX];
 585                 int len = 0;
 586                 int i;
 587
 588                 if (td->directory && td->directory[0] != '\0')
 589                         sprintf(tmp, "%s/", td->directory);
 590
 591                 td->files = malloc(sizeof(struct fio_file) * td->nr_files);
 592
 593                 for_each_file(td, f, i) {
 594                         memset(f, 0, sizeof(*f));
 595                         f->fd = -1;
 596
 597                         if (td->filename)
 598                                 sprintf(tmp + len, "%s", td->filename);
 599                         else
 600                                 sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i);
 601                         f->file_name = strdup(tmp);
 602                 }
 603         } else {
 604                 td->nr_files = 1;
 605                 td->files = malloc(sizeof(struct fio_file));
 606                 f = &td->files[0];
 607
 608                 memset(f, 0, sizeof(*f));
 609                 f->fd = -1;
 610                 f->file_name = strdup(jobname);
 611         }
 612
 613         for_each_file(td, f, i) {
 614                 f->file_size = td->total_file_size / td->nr_files;
 615                 f->file_offset = td->start_offset;
 616         }
 617
 618         fio_sem_init(&td->mutex, 0);
 619
 620         td->clat_stat[0].min_val = td->clat_stat[1].min_val = ULONG_MAX;
 621         td->slat_stat[0].min_val = td->slat_stat[1].min_val = ULONG_MAX;
 622         td->bw_stat[0].min_val = td->bw_stat[1].min_val = ULONG_MAX;
 623
 624         if (td->stonewall && td->thread_number > 1)
 625                 groupid++;
 626
 627         td->groupid = groupid;
 628
 629         if (setup_rate(td))
 630                 goto err;
 631
 632         if (td->write_lat_log) {
 633                 setup_log(&td->slat_log);
 634                 setup_log(&td->clat_log);
 635         }
 636         if (td->write_bw_log)
 637                 setup_log(&td->bw_log);
 638
 639         if (!td->name)
 640                 td->name = strdup(jobname);
 641
 642         ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
 643
 644         if (!terse_output) {
 645                 if (!job_add_num) {
 646                         if (td->io_ops->flags & FIO_CPUIO)
 647                                 fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
 648                         else
 649                                 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);
 650                 } else if (job_add_num == 1)
 651                         fprintf(f_out, "...\n");
 652         }
 653
 654         /*
 655          * recurse add identical jobs, clear numjobs and stonewall options
 656          * as they don't apply to sub-jobs
 657          */
 658         numjobs = td->numjobs;
 659         while (--numjobs) {
 660                 struct thread_data *td_new = get_new_job(0, td);
 661
 662                 if (!td_new)
 663                         goto err;
 664
 665                 td_new->numjobs = 1;
 666                 td_new->stonewall = 0;
 667                 job_add_num = numjobs - 1;
 668
 669                 if (add_job(td_new, jobname, job_add_num))
 670                         goto err;
 671         }
 672         return 0;
 673 err:
 674         put_job(td);
 675         return -1;
 676 }
 677
 678 /*
 679  * Initialize the various random states we need (random io, block size ranges,
 680  * read/write mix, etc).
 681  */
 682 int init_random_state(struct thread_data *td)
 683 {
 684         unsigned long seeds[4];
 685         int fd, num_maps, blocks, i;
 686         struct fio_file *f;
 687
 688         if (td->io_ops->flags & FIO_CPUIO)
 689                 return 0;
 690
 691         fd = open("/dev/urandom", O_RDONLY);
 692         if (fd == -1) {
 693                 td_verror(td, errno);
 694                 return 1;
 695         }
 696
 697         if (read(fd, seeds, sizeof(seeds)) < (int) sizeof(seeds)) {
 698                 td_verror(td, EIO);
 699                 close(fd);
 700                 return 1;
 701         }
 702
 703         close(fd);
 704
 705         os_random_seed(seeds[0], &td->bsrange_state);
 706         os_random_seed(seeds[1], &td->verify_state);
 707         os_random_seed(seeds[2], &td->rwmix_state);
 708
 709         if (td->sequential)
 710                 return 0;
 711
 712         if (td->rand_repeatable)
 713                 seeds[3] = DEF_RANDSEED;
 714
 715         if (!td->norandommap) {
 716                 for_each_file(td, f, i) {
 717                         blocks = (f->file_size + td->rw_min_bs - 1) / td->rw_min_bs;
 718                         num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP;
 719                         f->file_map = malloc(num_maps * sizeof(long));
 720                         f->num_maps = num_maps;
 721                         memset(f->file_map, 0, num_maps * sizeof(long));
 722                 }
 723         }
 724
 725         os_random_seed(seeds[3], &td->random_state);
 726         return 0;
 727 }
 728
 729 static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu)
 730 {
 731 #ifdef FIO_HAVE_CPU_AFFINITY
 732         unsigned int i;
 733
 734         CPU_ZERO(&cpumask);
 735
 736         for (i = 0; i < sizeof(int) * 8; i++) {
 737                 if ((1 << i) & cpu)
 738                         CPU_SET(i, &cpumask);
 739         }
 740 #endif
 741 }
 742
 743 static int is_empty_or_comment(char *line)
 744 {
 745         unsigned int i;
 746
 747         for (i = 0; i < strlen(line); i++) {
 748                 if (line[i] == ';')
 749                         return 1;
 750                 if (!isspace(line[i]) && !iscntrl(line[i]))
 751                         return 0;
 752         }
 753
 754         return 1;
 755 }
 756
 757 static int str_rw_cb(void *data, const char *mem)
 758 {
 759         struct thread_data *td = data;
 760
 761         if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) {
 762                 td->ddir = DDIR_READ;
 763                 td->sequential = 1;
 764                 return 0;
 765         } else if (!strncmp(mem, "randread", 8)) {
 766                 td->ddir = DDIR_READ;
 767                 td->sequential = 0;
 768                 return 0;
 769         } else if (!strncmp(mem, "write", 5) || !strncmp(mem, "1", 1)) {
 770                 td->ddir = DDIR_WRITE;
 771                 td->sequential = 1;
 772                 return 0;
 773         } else if (!strncmp(mem, "randwrite", 9)) {
 774                 td->ddir = DDIR_WRITE;
 775                 td->sequential = 0;
 776                 return 0;
 777         } else if (!strncmp(mem, "rw", 2)) {
 778                 td->ddir = 0;
 779                 td->iomix = 1;
 780                 td->sequential = 1;
 781                 return 0;
 782         } else if (!strncmp(mem, "randrw", 6)) {
 783                 td->ddir = 0;
 784                 td->iomix = 1;
 785                 td->sequential = 0;
 786                 return 0;
 787         }
 788
 789         log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n");
 790         return 1;
 791 }
 792
 793 static int str_verify_cb(void *data, const char *mem)
 794 {
 795         struct thread_data *td = data;
 796
 797         if (!strncmp(mem, "0", 1)) {
 798                 td->verify = VERIFY_NONE;
 799                 return 0;
 800         } else if (!strncmp(mem, "md5", 3) || !strncmp(mem, "1", 1)) {
 801                 td->verify = VERIFY_MD5;
 802                 return 0;
 803         } else if (!strncmp(mem, "crc32", 5)) {
 804                 td->verify = VERIFY_CRC32;
 805                 return 0;
 806         }
 807
 808         log_err("fio: verify types: md5, crc32\n");
 809         return 1;
 810 }
 811
 812 static int str_mem_cb(void *data, const char *mem)
 813 {
 814         struct thread_data *td = data;
 815
 816         if (!strncmp(mem, "malloc", 6)) {
 817                 td->mem_type = MEM_MALLOC;
 818                 return 0;
 819         } else if (!strncmp(mem, "shm", 3)) {
 820                 td->mem_type = MEM_SHM;
 821                 return 0;
 822         } else if (!strncmp(mem, "mmap", 4)) {
 823                 td->mem_type = MEM_MMAP;
 824                 return 0;
 825         }
 826
 827         log_err("fio: mem type: malloc, shm, mmap\n");
 828         return 1;
 829 }
 830
 831 static int str_ioengine_cb(void *data, const char *str)
 832 {
 833         struct thread_data *td = data;
 834
 835         td->io_ops = load_ioengine(td, str);
 836         if (td->io_ops)
 837                 return 0;
 838
 839         log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice, cpu\n");
 840         return 1;
 841 }
 842
 843 static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
 844 {
 845         mlock_size = *val;
 846         return 0;
 847 }
 848
 849 #ifdef FIO_HAVE_IOPRIO
 850 static int str_prioclass_cb(void *data, unsigned int *val)
 851 {
 852         struct thread_data *td = data;
 853
 854         td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
 855         return 0;
 856 }
 857
 858 static int str_prio_cb(void *data, unsigned int *val)
 859 {
 860         struct thread_data *td = data;
 861
 862         td->ioprio |= *val;
 863         return 0;
 864 }
 865 #endif
 866
 867 static int str_exitall_cb(void)
 868 {
 869         exitall_on_terminate = 1;
 870         return 0;
 871 }
 872
 873 static int str_cpumask_cb(void *data, unsigned int *val)
 874 {
 875         struct thread_data *td = data;
 876
 877         fill_cpu_mask(td->cpumask, *val);
 878         return 0;
 879 }
 880
 881 /*
 882  * This is our [ini] type file parser.
 883  */
 884 int parse_jobs_ini(char *file, int stonewall_flag)
 885 {
 886         unsigned int global;
 887         struct thread_data *td;
 888         char *string, *name;
 889         fpos_t off;
 890         FILE *f;
 891         char *p;
 892         int ret = 0, stonewall;
 893
 894         f = fopen(file, "r");
 895         if (!f) {
 896                 perror("fopen job file");
 897                 return 1;
 898         }
 899
 900         string = malloc(4096);
 901         name = malloc(256);
 902         memset(name, 0, 256);
 903
 904         stonewall = stonewall_flag;
 905         do {
 906                 p = fgets(string, 4095, f);
 907                 if (!p)
 908                         break;
 909                 if (is_empty_or_comment(p))
 910                         continue;
 911                 if (sscanf(p, "[%255s]", name) != 1)
 912                         continue;
 913
 914                 global = !strncmp(name, "global", 6);
 915
 916                 name[strlen(name) - 1] = '\0';
 917
 918                 td = get_new_job(global, &def_thread);
 919                 if (!td) {
 920                         ret = 1;
 921                         break;
 922                 }
 923
 924                 /*
 925                  * Seperate multiple job files by a stonewall
 926                  */
 927                 if (!global && stonewall) {
 928                         td->stonewall = stonewall;
 929                         stonewall = 0;
 930                 }
 931
 932                 fgetpos(f, &off);
 933                 while ((p = fgets(string, 4096, f)) != NULL) {
 934                         if (is_empty_or_comment(p))
 935                                 continue;
 936
 937                         strip_blank_front(&p);
 938
 939                         if (p[0] == '[')
 940                                 break;
 941
 942                         strip_blank_end(p);
 943
 944                         fgetpos(f, &off);
 945
 946                         /*
 947                          * Don't break here, continue parsing options so we
 948                          * dump all the bad ones. Makes trial/error fixups
 949                          * easier on the user.
 950                          */
 951                         ret |= parse_option(p, options, td);
 952                 }
 953
 954                 if (!ret) {
 955                         fsetpos(f, &off);
 956                         ret = add_job(td, name, 0);
 957                 } else {
 958                         log_err("fio: job %s dropped\n", name);
 959                         put_job(td);
 960                 }
 961         } while (!ret);
 962
 963         free(string);
 964         free(name);
 965         fclose(f);
 966         return ret;
 967 }
 968
 969 static int fill_def_thread(void)
 970 {
 971         memset(&def_thread, 0, sizeof(def_thread));
 972
 973         if (fio_getaffinity(getpid(), &def_thread.cpumask) == -1) {
 974                 perror("sched_getaffinity");
 975                 return 1;
 976         }
 977
 978         /*
 979          * fill globals
 980          */
 981         def_thread.ddir = DDIR_READ;
 982         def_thread.iomix = 0;
 983         def_thread.bs[DDIR_READ] = DEF_BS;
 984         def_thread.bs[DDIR_WRITE] = DEF_BS;
 985         def_thread.min_bs[DDIR_READ] = def_thread.min_bs[DDIR_WRITE] = 0;
 986         def_thread.max_bs[DDIR_READ] = def_thread.max_bs[DDIR_WRITE] = 0;
 987         def_thread.odirect = DEF_ODIRECT;
 988         def_thread.ratecycle = DEF_RATE_CYCLE;
 989         def_thread.sequential = DEF_SEQUENTIAL;
 990         def_thread.timeout = def_timeout;
 991         def_thread.overwrite = DEF_OVERWRITE;
 992         def_thread.invalidate_cache = DEF_INVALIDATE;
 993         def_thread.sync_io = DEF_SYNCIO;
 994         def_thread.mem_type = MEM_MALLOC;
 995         def_thread.bw_avg_time = DEF_BWAVGTIME;
 996         def_thread.create_serialize = DEF_CREATE_SER;
 997         def_thread.create_fsync = DEF_CREATE_FSYNC;
 998         def_thread.loops = DEF_LOOPS;
 999         def_thread.verify = DEF_VERIFY;
1000         def_thread.stonewall = DEF_STONEWALL;
1001         def_thread.numjobs = DEF_NUMJOBS;
1002         def_thread.use_thread = DEF_USE_THREAD;
1003         def_thread.rwmixcycle = DEF_RWMIX_CYCLE;
1004         def_thread.rwmixread = DEF_RWMIX_READ;
1005         def_thread.nice = DEF_NICE;
1006         def_thread.rand_repeatable = DEF_RAND_REPEAT;
1007         def_thread.nr_files = DEF_NR_FILES;
1008         def_thread.unlink = DEF_UNLINK;
1009         def_thread.write_bw_log = write_bw_log;
1010         def_thread.write_lat_log = write_lat_log;
1011         def_thread.norandommap = DEF_NO_RAND_MAP;
1012 #ifdef FIO_HAVE_DISK_UTIL
1013         def_thread.do_disk_util = 1;
1014 #endif
1015
1016         return 0;
1017 }
1018
1019 static void usage(void)
1020 {
1021         printf("%s\n", fio_version_string);
1022         printf("\t--output\tWrite output to file\n");
1023         printf("\t--timeout\tRuntime in seconds\n");
1024         printf("\t--latency-log\tGenerate per-job latency logs\n");
1025         printf("\t--bandwidth-log\tGenerate per-job bandwidth logs\n");
1026         printf("\t--minimal\tMinimal (terse) output\n");
1027         printf("\t--version\tPrint version info and exit\n");
1028 }
1029
1030 static int parse_cmd_line(int argc, char *argv[])
1031 {
1032         struct thread_data *td = NULL;
1033         int c, ini_idx = 0, lidx, ret;
1034
1035         while ((c = getopt_long(argc, argv, "", long_options, &lidx)) != -1) {
1036                 switch (c) {
1037                 case 't':
1038                         def_timeout = atoi(optarg);
1039                         break;
1040                 case 'l':
1041                         write_lat_log = 1;
1042                         break;
1043                 case 'w':
1044                         write_bw_log = 1;
1045                         break;
1046                 case 'o':
1047                         f_out = fopen(optarg, "w+");
1048                         if (!f_out) {
1049                                 perror("fopen output");
1050                                 exit(1);
1051                         }
1052                         f_err = f_out;
1053                         break;
1054                 case 'm':
1055                         terse_output = 1;
1056                         break;
1057                 case 'h':
1058                         usage();
1059                         exit(0);
1060                 case 'v':
1061                         printf("%s\n", fio_version_string);
1062                         exit(0);
1063                 case FIO_GETOPT_JOB: {
1064                         const char *opt = long_options[lidx].name;
1065                         char *val = optarg;
1066
1067                         if (!strncmp(opt, "name", 4) && td) {
1068                                 ret = add_job(td, td->name ?: "fio", 0);
1069                                 if (ret) {
1070                                         put_job(td);
1071                                         return 0;
1072                                 }
1073                                 td = NULL;
1074                         }
1075                         if (!td) {
1076                                 int global = !strncmp(val, "global", 6);
1077
1078                                 td = get_new_job(global, &def_thread);
1079                                 if (!td)
1080                                         return 0;
1081                         }
1082
1083                         ret = parse_cmd_option(opt, val, options, td);
1084                         if (ret) {
1085                                 log_err("fio: job dropped\n");
1086                                 put_job(td);
1087                                 td = NULL;
1088                         }
1089                         break;
1090                 }
1091                 default:
1092                         printf("optarg <<%s>>\n", argv[optind]);
1093                         break;
1094                 }
1095         }
1096
1097         if (td) {
1098                 ret = add_job(td, td->name ?: "fio", 0);
1099                 if (ret)
1100                         put_job(td);
1101         }
1102
1103         while (optind < argc) {
1104                 ini_idx++;
1105                 ini_file = realloc(ini_file, ini_idx * sizeof(char *));
1106                 ini_file[ini_idx - 1] = strdup(argv[optind]);
1107                 optind++;
1108         }
1109
1110         return ini_idx;
1111 }
1112
1113 static void free_shm(void)
1114 {
1115         struct shmid_ds sbuf;
1116
1117         if (threads) {
1118                 shmdt((void *) threads);
1119                 threads = NULL;
1120                 shmctl(shm_id, IPC_RMID, &sbuf);
1121         }
1122 }
1123
1124 /*
1125  * The thread area is shared between the main process and the job
1126  * threads/processes. So setup a shared memory segment that will hold
1127  * all the job info.
1128  */
1129 static int setup_thread_area(void)
1130 {
1131         /*
1132          * 1024 is too much on some machines, scale max_jobs if
1133          * we get a failure that looks like too large a shm segment
1134          */
1135         do {
1136                 size_t size = max_jobs * sizeof(struct thread_data);
1137
1138                 shm_id = shmget(0, size, IPC_CREAT | 0600);
1139                 if (shm_id != -1)
1140                         break;
1141                 if (errno != EINVAL) {
1142                         perror("shmget");
1143                         break;
1144                 }
1145
1146                 max_jobs >>= 1;
1147         } while (max_jobs);
1148
1149         if (shm_id == -1)
1150                 return 1;
1151
1152         threads = shmat(shm_id, NULL, 0);
1153         if (threads == (void *) -1) {
1154                 perror("shmat");
1155                 return 1;
1156         }
1157
1158         atexit(free_shm);
1159         return 0;
1160 }
1161
1162 /*
1163  * Copy the fio options into the long options map, so we mirror
1164  * job and cmd line options.
1165  */
1166 static void dupe_job_options(void)
1167 {
1168         struct fio_option *o;
1169         unsigned int i;
1170
1171         i = 0;
1172         while (long_options[i].name)
1173                 i++;
1174
1175         o = &options[0];
1176         while (o->name) {
1177                 long_options[i].name = o->name;
1178                 long_options[i].val = FIO_GETOPT_JOB;
1179                 if (o->type == FIO_OPT_STR_SET)
1180                         long_options[i].has_arg = no_argument;
1181                 else
1182                         long_options[i].has_arg = required_argument;
1183
1184                 i++;
1185                 o++;
1186                 assert(i < FIO_JOB_OPTS + FIO_CMD_OPTS);
1187         }
1188 }
1189
1190 int parse_options(int argc, char *argv[])
1191 {
1192         int job_files, i;
1193
1194         f_out = stdout;
1195         f_err = stderr;
1196
1197         dupe_job_options();
1198
1199         if (setup_thread_area())
1200                 return 1;
1201         if (fill_def_thread())
1202                 return 1;
1203
1204         job_files = parse_cmd_line(argc, argv);
1205
1206         for (i = 0; i < job_files; i++) {
1207                 if (fill_def_thread())
1208                         return 1;
1209                 if (parse_jobs_ini(ini_file[i], i))
1210                         return 1;
1211                 free(ini_file[i]);
1212         }
1213
1214         free(ini_file);
1215
1216         if (!thread_number) {
1217                 log_err("No jobs defined(s)\n");
1218                 return 1;
1219         }
1220
1221         return 0;
1222 }