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 #define FIO_RANDSEED            (0xb1899bedUL)
  22
  23 #define td_var_offset(var)      ((size_t) &((struct thread_data *)0)->var)
  24
  25 static int str_mem_cb(void *, const char *);
  26 static int str_lockmem_cb(void *, unsigned long *);
  27 #ifdef FIO_HAVE_IOPRIO
  28 static int str_prio_cb(void *, unsigned int *);
  29 static int str_prioclass_cb(void *, unsigned int *);
  30 #endif
  31 static int str_exitall_cb(void);
  32 static int str_cpumask_cb(void *, unsigned int *);
  33 static int str_fst_cb(void *, const char *);
  34
  35 #define __stringify_1(x)        #x
  36 #define __stringify(x)          __stringify_1(x)
  37
  38 /*
  39  * Map of job/command line options
  40  */
  41 static struct fio_option options[] = {
  42         {
  43                 .name   = "description",
  44                 .type   = FIO_OPT_STR_STORE,
  45                 .off1   = td_var_offset(description),
  46                 .help   = "Text job description",
  47         },
  48         {
  49                 .name   = "name",
  50                 .type   = FIO_OPT_STR_STORE,
  51                 .off1   = td_var_offset(name),
  52                 .help   = "Name of this job",
  53         },
  54         {
  55                 .name   = "directory",
  56                 .type   = FIO_OPT_STR_STORE,
  57                 .off1   = td_var_offset(directory),
  58                 .help   = "Directory to store files in",
  59         },
  60         {
  61                 .name   = "filename",
  62                 .type   = FIO_OPT_STR_STORE,
  63                 .off1   = td_var_offset(filename),
  64                 .help   = "Force the use of a specific file",
  65         },
  66         {
  67                 .name   = "rw",
  68                 .type   = FIO_OPT_STR,
  69                 .off1   = td_var_offset(td_ddir),
  70                 .help   = "IO direction",
  71                 .def    = "read",
  72                 .posval = {
  73                           { .ival = "read", .oval = TD_DDIR_READ },
  74                           { .ival = "write", .oval = TD_DDIR_WRITE },
  75                           { .ival = "randread", .oval = TD_DDIR_RANDREAD },
  76                           { .ival = "randwrite", .oval = TD_DDIR_RANDWRITE },
  77                           { .ival = "rw", .oval = TD_DDIR_RW },
  78                           { .ival = "randrw", .oval = TD_DDIR_RANDRW },
  79                           },
  80         },
  81         {
  82                 .name   = "ioengine",
  83                 .type   = FIO_OPT_STR_STORE,
  84                 .off1   = td_var_offset(ioengine),
  85                 .help   = "IO engine to use",
  86                 .def    = "sync",
  87                 .posval = {
  88                           { .ival = "sync", },
  89 #ifdef FIO_HAVE_LIBAIO
  90                           { .ival = "libaio", },
  91 #endif
  92 #ifdef FIO_HAVE_POSIXAIO
  93                           { .ival = "posixaio", },
  94 #endif
  95                           { .ival = "mmap", },
  96 #ifdef FIO_HAVE_SPLICE
  97                           { .ival = "splice", },
  98 #endif
  99 #ifdef FIO_HAVE_SGIO
 100                           { .ival = "sg", },
 101 #endif
 102                           { .ival = "null", }, { .ival = "net", },
 103 #ifdef FIO_HAVE_SYSLET
 104                           { .ival = "syslet-rw", },
 105 #endif
 106                           { .ival = "cpuio", },
 107                           { .ival = "external", },
 108                           },
 109         },
 110         {
 111                 .name   = "iodepth",
 112                 .type   = FIO_OPT_INT,
 113                 .off1   = td_var_offset(iodepth),
 114                 .help   = "Amount of IO buffers to keep in flight",
 115                 .def    = "1",
 116         },
 117         {
 118                 .name   = "iodepth_batch",
 119                 .type   = FIO_OPT_INT,
 120                 .off1   = td_var_offset(iodepth_batch),
 121                 .help   = "Number of IO to submit in one go",
 122         },
 123         {
 124                 .name   = "iodepth_low",
 125                 .type   = FIO_OPT_INT,
 126                 .off1   = td_var_offset(iodepth_low),
 127                 .help   = "Low water mark for queuing depth",
 128         },
 129         {
 130                 .name   = "size",
 131                 .type   = FIO_OPT_STR_VAL,
 132                 .off1   = td_var_offset(total_file_size),
 133                 .help   = "Size of device or file",
 134         },
 135         {
 136                 .name   = "bs",
 137                 .type   = FIO_OPT_STR_VAL_INT,
 138                 .off1   = td_var_offset(bs[DDIR_READ]),
 139                 .off2   = td_var_offset(bs[DDIR_WRITE]),
 140                 .help   = "Block size unit",
 141                 .def    = "4k",
 142         },
 143         {
 144                 .name   = "bsrange",
 145                 .type   = FIO_OPT_RANGE,
 146                 .off1   = td_var_offset(min_bs[DDIR_READ]),
 147                 .off2   = td_var_offset(max_bs[DDIR_READ]),
 148                 .off3   = td_var_offset(min_bs[DDIR_WRITE]),
 149                 .off4   = td_var_offset(max_bs[DDIR_WRITE]),
 150                 .help   = "Set block size range (in more detail than bs)",
 151         },
 152         {
 153                 .name   = "bs_unaligned",
 154                 .type   = FIO_OPT_STR_SET,
 155                 .off1   = td_var_offset(bs_unaligned),
 156                 .help   = "Don't sector align IO buffer sizes",
 157         },
 158         {
 159                 .name   = "offset",
 160                 .type   = FIO_OPT_STR_VAL,
 161                 .off1   = td_var_offset(start_offset),
 162                 .help   = "Start IO from this offset",
 163                 .def    = "0",
 164         },
 165         {
 166                 .name   = "randrepeat",
 167                 .type   = FIO_OPT_BOOL,
 168                 .off1   = td_var_offset(rand_repeatable),
 169                 .help   = "Use repeatable random IO pattern",
 170                 .def    = "1",
 171         },
 172         {
 173                 .name   = "norandommap",
 174                 .type   = FIO_OPT_STR_SET,
 175                 .off1   = td_var_offset(norandommap),
 176                 .help   = "Accept potential duplicate random blocks",
 177         },
 178         {
 179                 .name   = "nrfiles",
 180                 .type   = FIO_OPT_INT,
 181                 .off1   = td_var_offset(nr_files),
 182                 .help   = "Split job workload between this number of files",
 183                 .def    = "1",
 184         },
 185         {
 186                 .name   = "openfiles",
 187                 .type   = FIO_OPT_INT,
 188                 .off1   = td_var_offset(open_files),
 189                 .help   = "Number of files to keep open at the same time",
 190         },
 191         {
 192                 .name   = "file_service_type",
 193                 .type   = FIO_OPT_STR,
 194                 .cb     = str_fst_cb,
 195                 .off1   = td_var_offset(file_service_type),
 196                 .help   = "How to select which file to service next",
 197                 .def    = "roundrobin",
 198                 .posval = {
 199                           { .ival = "random", .oval = FIO_FSERVICE_RANDOM },
 200                           { .ival = "roundrobin", .oval = FIO_FSERVICE_RR },
 201                           },
 202         },
 203         {
 204                 .name   = "fsync",
 205                 .type   = FIO_OPT_INT,
 206                 .off1   = td_var_offset(fsync_blocks),
 207                 .help   = "Issue fsync for writes every given number of blocks",
 208                 .def    = "0",
 209         },
 210         {
 211                 .name   = "direct",
 212                 .type   = FIO_OPT_BOOL,
 213                 .off1   = td_var_offset(odirect),
 214                 .help   = "Use O_DIRECT IO (negates buffered)",
 215                 .def    = "0",
 216         },
 217         {
 218                 .name   = "buffered",
 219                 .type   = FIO_OPT_BOOL,
 220                 .off1   = td_var_offset(odirect),
 221                 .neg    = 1,
 222                 .help   = "Use buffered IO (negates direct)",
 223                 .def    = "1",
 224         },
 225         {
 226                 .name   = "overwrite",
 227                 .type   = FIO_OPT_BOOL,
 228                 .off1   = td_var_offset(overwrite),
 229                 .help   = "When writing, set whether to overwrite current data",
 230                 .def    = "0",
 231         },
 232         {
 233                 .name   = "loops",
 234                 .type   = FIO_OPT_INT,
 235                 .off1   = td_var_offset(loops),
 236                 .help   = "Number of times to run the job",
 237                 .def    = "1",
 238         },
 239         {
 240                 .name   = "numjobs",
 241                 .type   = FIO_OPT_INT,
 242                 .off1   = td_var_offset(numjobs),
 243                 .help   = "Duplicate this job this many times",
 244                 .def    = "1",
 245         },
 246         {
 247                 .name   = "startdelay",
 248                 .type   = FIO_OPT_INT,
 249                 .off1   = td_var_offset(start_delay),
 250                 .help   = "Only start job when this period has passed",
 251                 .def    = "0",
 252         },
 253         {
 254                 .name   = "runtime",
 255                 .alias  = "timeout",
 256                 .type   = FIO_OPT_STR_VAL_TIME,
 257                 .off1   = td_var_offset(timeout),
 258                 .help   = "Stop workload when this amount of time has passed",
 259                 .def    = "0",
 260         },
 261         {
 262                 .name   = "mem",
 263                 .type   = FIO_OPT_STR,
 264                 .cb     = str_mem_cb,
 265                 .off1   = td_var_offset(mem_type),
 266                 .help   = "Backing type for IO buffers",
 267                 .def    = "malloc",
 268                 .posval = {
 269                           { .ival = "malloc", .oval = MEM_MALLOC },
 270                           { .ival = "shm", .oval = MEM_SHM },
 271 #ifdef FIO_HAVE_HUGETLB
 272                           { .ival = "shmhuge", .oval = MEM_SHMHUGE },
 273 #endif
 274                           { .ival = "mmap", .oval = MEM_MMAP },
 275 #ifdef FIO_HAVE_HUGETLB
 276                           { .ival = "mmaphuge", .oval = MEM_MMAPHUGE },
 277 #endif
 278                           },
 279         },
 280         {
 281                 .name   = "verify",
 282                 .type   = FIO_OPT_STR,
 283                 .off1   = td_var_offset(verify),
 284                 .help   = "Verify sum function",
 285                 .def    = "0",
 286                 .posval = {
 287                           { .ival = "0", .oval = VERIFY_NONE },
 288                           { .ival = "crc32", .oval = VERIFY_CRC32 },
 289                           { .ival = "md5", .oval = VERIFY_MD5 },
 290                           },
 291         },
 292         {
 293                 .name   = "write_iolog",
 294                 .type   = FIO_OPT_STR_STORE,
 295                 .off1   = td_var_offset(write_iolog_file),
 296                 .help   = "Store IO pattern to file",
 297         },
 298         {
 299                 .name   = "read_iolog",
 300                 .type   = FIO_OPT_STR_STORE,
 301                 .off1   = td_var_offset(read_iolog_file),
 302                 .help   = "Playback IO pattern from file",
 303         },
 304         {
 305                 .name   = "exec_prerun",
 306                 .type   = FIO_OPT_STR_STORE,
 307                 .off1   = td_var_offset(exec_prerun),
 308                 .help   = "Execute this file prior to running job",
 309         },
 310         {
 311                 .name   = "exec_postrun",
 312                 .type   = FIO_OPT_STR_STORE,
 313                 .off1   = td_var_offset(exec_postrun),
 314                 .help   = "Execute this file after running job",
 315         },
 316 #ifdef FIO_HAVE_IOSCHED_SWITCH
 317         {
 318                 .name   = "ioscheduler",
 319                 .type   = FIO_OPT_STR_STORE,
 320                 .off1   = td_var_offset(ioscheduler),
 321                 .help   = "Use this IO scheduler on the backing device",
 322         },
 323 #endif
 324         {
 325                 .name   = "zonesize",
 326                 .type   = FIO_OPT_STR_VAL,
 327                 .off1   = td_var_offset(zone_size),
 328                 .help   = "Give size of an IO zone",
 329                 .def    = "0",
 330         },
 331         {
 332                 .name   = "zoneskip",
 333                 .type   = FIO_OPT_STR_VAL,
 334                 .off1   = td_var_offset(zone_skip),
 335                 .help   = "Space between IO zones",
 336                 .def    = "0",
 337         },
 338         {
 339                 .name   = "lockmem",
 340                 .type   = FIO_OPT_STR_VAL,
 341                 .cb     = str_lockmem_cb,
 342                 .help   = "Lock down this amount of memory",
 343                 .def    = "0",
 344         },
 345         {
 346                 .name   = "rwmixcycle",
 347                 .type   = FIO_OPT_INT,
 348                 .off1   = td_var_offset(rwmixcycle),
 349                 .help   = "Cycle period for mixed read/write workloads (msec)",
 350                 .def    = "500",
 351         },
 352         {
 353                 .name   = "rwmixread",
 354                 .type   = FIO_OPT_INT,
 355                 .off1   = td_var_offset(rwmixread),
 356                 .maxval = 100,
 357                 .help   = "Percentage of mixed workload that is reads",
 358                 .def    = "50",
 359         },
 360         {
 361                 .name   = "rwmixwrite",
 362                 .type   = FIO_OPT_INT,
 363                 .off1   = td_var_offset(rwmixwrite),
 364                 .maxval = 100,
 365                 .help   = "Percentage of mixed workload that is writes",
 366                 .def    = "50",
 367         },
 368         {
 369                 .name   = "nice",
 370                 .type   = FIO_OPT_INT,
 371                 .off1   = td_var_offset(nice),
 372                 .help   = "Set job CPU nice value",
 373                 .minval = -19,
 374                 .maxval = 20,
 375                 .def    = "0",
 376         },
 377 #ifdef FIO_HAVE_IOPRIO
 378         {
 379                 .name   = "prio",
 380                 .type   = FIO_OPT_INT,
 381                 .cb     = str_prio_cb,
 382                 .help   = "Set job IO priority value",
 383                 .minval = 0,
 384                 .maxval = 7,
 385         },
 386         {
 387                 .name   = "prioclass",
 388                 .type   = FIO_OPT_INT,
 389                 .cb     = str_prioclass_cb,
 390                 .help   = "Set job IO priority class",
 391                 .minval = 0,
 392                 .maxval = 3,
 393         },
 394 #endif
 395         {
 396                 .name   = "thinktime",
 397                 .type   = FIO_OPT_INT,
 398                 .off1   = td_var_offset(thinktime),
 399                 .help   = "Idle time between IO buffers (usec)",
 400                 .def    = "0",
 401         },
 402         {
 403                 .name   = "thinktime_spin",
 404                 .type   = FIO_OPT_INT,
 405                 .off1   = td_var_offset(thinktime_spin),
 406                 .help   = "Start think time by spinning this amount (usec)",
 407                 .def    = "0",
 408         },
 409         {
 410                 .name   = "thinktime_blocks",
 411                 .type   = FIO_OPT_INT,
 412                 .off1   = td_var_offset(thinktime_blocks),
 413                 .help   = "IO buffer period between 'thinktime'",
 414                 .def    = "1",
 415         },
 416         {
 417                 .name   = "rate",
 418                 .type   = FIO_OPT_INT,
 419                 .off1   = td_var_offset(rate),
 420                 .help   = "Set bandwidth rate",
 421         },
 422         {
 423                 .name   = "ratemin",
 424                 .type   = FIO_OPT_INT,
 425                 .off1   = td_var_offset(ratemin),
 426                 .help   = "The bottom limit accepted",
 427         },
 428         {
 429                 .name   = "ratecycle",
 430                 .type   = FIO_OPT_INT,
 431                 .off1   = td_var_offset(ratecycle),
 432                 .help   = "Window average for rate limits (msec)",
 433                 .def    = "1000",
 434         },
 435         {
 436                 .name   = "invalidate",
 437                 .type   = FIO_OPT_BOOL,
 438                 .off1   = td_var_offset(invalidate_cache),
 439                 .help   = "Invalidate buffer/page cache prior to running job",
 440                 .def    = "1",
 441         },
 442         {
 443                 .name   = "sync",
 444                 .type   = FIO_OPT_BOOL,
 445                 .off1   = td_var_offset(sync_io),
 446                 .help   = "Use O_SYNC for buffered writes",
 447                 .def    = "0",
 448         },
 449         {
 450                 .name   = "bwavgtime",
 451                 .type   = FIO_OPT_INT,
 452                 .off1   = td_var_offset(bw_avg_time),
 453                 .help   = "Time window over which to calculate bandwidth (msec)",
 454                 .def    = "500",
 455         },
 456         {
 457                 .name   = "create_serialize",
 458                 .type   = FIO_OPT_BOOL,
 459                 .off1   = td_var_offset(create_serialize),
 460                 .help   = "Serialize creating of job files",
 461                 .def    = "1",
 462         },
 463         {
 464                 .name   = "create_fsync",
 465                 .type   = FIO_OPT_BOOL,
 466                 .off1   = td_var_offset(create_fsync),
 467                 .help   = "Fsync file after creation",
 468                 .def    = "1",
 469         },
 470         {
 471                 .name   = "cpuload",
 472                 .type   = FIO_OPT_INT,
 473                 .off1   = td_var_offset(cpuload),
 474                 .help   = "Use this percentage of CPU",
 475         },
 476         {
 477                 .name   = "cpuchunks",
 478                 .type   = FIO_OPT_INT,
 479                 .off1   = td_var_offset(cpucycle),
 480                 .help   = "Length of the CPU burn cycles (usecs)",
 481                 .def    = "50000",
 482         },
 483 #ifdef FIO_HAVE_CPU_AFFINITY
 484         {
 485                 .name   = "cpumask",
 486                 .type   = FIO_OPT_INT,
 487                 .cb     = str_cpumask_cb,
 488                 .help   = "CPU affinity mask",
 489         },
 490 #endif
 491         {
 492                 .name   = "end_fsync",
 493                 .type   = FIO_OPT_BOOL,
 494                 .off1   = td_var_offset(end_fsync),
 495                 .help   = "Include fsync at the end of job",
 496                 .def    = "0",
 497         },
 498         {
 499                 .name   = "unlink",
 500                 .type   = FIO_OPT_BOOL,
 501                 .off1   = td_var_offset(unlink),
 502                 .help   = "Unlink created files after job has completed",
 503                 .def    = "0",
 504         },
 505         {
 506                 .name   = "exitall",
 507                 .type   = FIO_OPT_STR_SET,
 508                 .cb     = str_exitall_cb,
 509                 .help   = "Terminate all jobs when one exits",
 510         },
 511         {
 512                 .name   = "stonewall",
 513                 .type   = FIO_OPT_STR_SET,
 514                 .off1   = td_var_offset(stonewall),
 515                 .help   = "Insert a hard barrier between this job and previous",
 516         },
 517         {
 518                 .name   = "thread",
 519                 .type   = FIO_OPT_STR_SET,
 520                 .off1   = td_var_offset(use_thread),
 521                 .help   = "Use threads instead of forks",
 522         },
 523         {
 524                 .name   = "write_bw_log",
 525                 .type   = FIO_OPT_STR_SET,
 526                 .off1   = td_var_offset(write_bw_log),
 527                 .help   = "Write log of bandwidth during run",
 528         },
 529         {
 530                 .name   = "write_lat_log",
 531                 .type   = FIO_OPT_STR_SET,
 532                 .off1   = td_var_offset(write_lat_log),
 533                 .help   = "Write log of latency during run",
 534         },
 535         {
 536                 .name   = "hugepage-size",
 537                 .type   = FIO_OPT_STR_VAL,
 538                 .off1   = td_var_offset(hugepage_size),
 539                 .help   = "When using hugepages, specify size of each page",
 540                 .def    = __stringify(FIO_HUGE_PAGE),
 541         },
 542         {
 543                 .name   = "group_reporting",
 544                 .type   = FIO_OPT_STR_SET,
 545                 .off1   = td_var_offset(group_reporting),
 546                 .help   = "Do reporting on a per-group basis",
 547         },
 548         {
 549                 .name = NULL,
 550         },
 551 };
 552
 553 #define FIO_JOB_OPTS    (sizeof(options) / sizeof(struct fio_option))
 554 #define FIO_CMD_OPTS    (16)
 555 #define FIO_GETOPT_JOB  (0x89988998)
 556
 557 /*
 558  * Command line options. These will contain the above, plus a few
 559  * extra that only pertain to fio itself and not jobs.
 560  */
 561 static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
 562         {
 563                 .name           = "output",
 564                 .has_arg        = required_argument,
 565                 .val            = 'o',
 566         },
 567         {
 568                 .name           = "timeout",
 569                 .has_arg        = required_argument,
 570                 .val            = 't',
 571         },
 572         {
 573                 .name           = "latency-log",
 574                 .has_arg        = required_argument,
 575                 .val            = 'l',
 576         },
 577         {
 578                 .name           = "bandwidth-log",
 579                 .has_arg        = required_argument,
 580                 .val            = 'b',
 581         },
 582         {
 583                 .name           = "minimal",
 584                 .has_arg        = optional_argument,
 585                 .val            = 'm',
 586         },
 587         {
 588                 .name           = "version",
 589                 .has_arg        = no_argument,
 590                 .val            = 'v',
 591         },
 592         {
 593                 .name           = "help",
 594                 .has_arg        = no_argument,
 595                 .val            = 'h',
 596         },
 597         {
 598                 .name           = "cmdhelp",
 599                 .has_arg        = optional_argument,
 600                 .val            = 'c',
 601         },
 602         {
 603                 .name           = NULL,
 604         },
 605 };
 606
 607 static int def_timeout = 0;
 608
 609 static char fio_version_string[] = "fio 1.13";
 610
 611 static char **ini_file;
 612 static int max_jobs = MAX_JOBS;
 613
 614 struct thread_data def_thread;
 615 struct thread_data *threads = NULL;
 616
 617 int exitall_on_terminate = 0;
 618 int terse_output = 0;
 619 unsigned long long mlock_size = 0;
 620 FILE *f_out = NULL;
 621 FILE *f_err = NULL;
 622
 623 static int write_lat_log = 0;
 624 int write_bw_log = 0;
 625
 626 FILE *get_f_out()
 627 {
 628         return f_out;
 629 }
 630
 631 FILE *get_f_err()
 632 {
 633         return f_err;
 634 }
 635
 636 /*
 637  * Return a free job structure.
 638  */
 639 static struct thread_data *get_new_job(int global, struct thread_data *parent)
 640 {
 641         struct thread_data *td;
 642
 643         if (global)
 644                 return &def_thread;
 645         if (thread_number >= max_jobs)
 646                 return NULL;
 647
 648         td = &threads[thread_number++];
 649         *td = *parent;
 650
 651         td->thread_number = thread_number;
 652         return td;
 653 }
 654
 655 static void put_job(struct thread_data *td)
 656 {
 657         if (td == &def_thread)
 658                 return;
 659
 660         if (td->error)
 661                 fprintf(f_out, "fio: %s\n", td->verror);
 662
 663         memset(&threads[td->thread_number - 1], 0, sizeof(*td));
 664         thread_number--;
 665 }
 666
 667 /*
 668  * Lazy way of fixing up options that depend on each other. We could also
 669  * define option callback handlers, but this is easier.
 670  */
 671 static void fixup_options(struct thread_data *td)
 672 {
 673         if (!td->rwmixread && td->rwmixwrite)
 674                 td->rwmixread = 100 - td->rwmixwrite;
 675
 676         if (td->write_iolog_file && td->read_iolog_file) {
 677                 log_err("fio: read iolog overrides write_iolog\n");
 678                 free(td->write_iolog_file);
 679                 td->write_iolog_file = NULL;
 680         }
 681
 682         if (td->io_ops->flags & FIO_SYNCIO)
 683                 td->iodepth = 1;
 684         else {
 685                 if (!td->iodepth)
 686                         td->iodepth = td->open_files;
 687         }
 688
 689         /*
 690          * only really works for sequential io for now, and with 1 file
 691          */
 692         if (td->zone_size && td_random(td) && td->open_files == 1)
 693                 td->zone_size = 0;
 694
 695         /*
 696          * Reads can do overwrites, we always need to pre-create the file
 697          */
 698         if (td_read(td) || td_rw(td))
 699                 td->overwrite = 1;
 700
 701         if (!td->min_bs[DDIR_READ])
 702                 td->min_bs[DDIR_READ]= td->bs[DDIR_READ];
 703         if (!td->max_bs[DDIR_READ])
 704                 td->max_bs[DDIR_READ] = td->bs[DDIR_READ];
 705         if (!td->min_bs[DDIR_WRITE])
 706                 td->min_bs[DDIR_WRITE]= td->bs[DDIR_WRITE];
 707         if (!td->max_bs[DDIR_WRITE])
 708                 td->max_bs[DDIR_WRITE] = td->bs[DDIR_WRITE];
 709
 710         td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]);
 711
 712         if (td_read(td) && !td_rw(td))
 713                 td->verify = 0;
 714
 715         if (td->norandommap && td->verify != VERIFY_NONE) {
 716                 log_err("fio: norandommap given, verify disabled\n");
 717                 td->verify = VERIFY_NONE;
 718         }
 719         if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO))
 720                 log_err("fio: bs_unaligned may not work with raw io\n");
 721
 722         /*
 723          * O_DIRECT and char doesn't mix, clear that flag if necessary.
 724          */
 725         if (td->filetype == FIO_TYPE_CHAR && td->odirect)
 726                 td->odirect = 0;
 727
 728         /*
 729          * thinktime_spin must be less than thinktime
 730          */
 731         if (td->thinktime_spin > td->thinktime)
 732                 td->thinktime_spin = td->thinktime;
 733
 734         /*
 735          * The low water mark cannot be bigger than the iodepth
 736          */
 737         if (td->iodepth_low > td->iodepth || !td->iodepth_low) {
 738                 /*
 739                  * syslet work around - if the workload is sequential,
 740                  * we want to let the queue drain all the way down to
 741                  * avoid seeking between async threads
 742                  */
 743                 if (!strcmp(td->io_ops->name, "syslet-rw") && !td_random(td))
 744                         td->iodepth_low = 1;
 745                 else
 746                         td->iodepth_low = td->iodepth;
 747         }
 748
 749         /*
 750          * If batch number isn't set, default to the same as iodepth
 751          */
 752         if (td->iodepth_batch > td->iodepth || !td->iodepth_batch)
 753                 td->iodepth_batch = td->iodepth;
 754
 755         if (td->open_files > td->nr_files || !td->open_files)
 756                 td->open_files = td->nr_files;
 757 }
 758
 759 /*
 760  * This function leaks the buffer
 761  */
 762 static char *to_kmg(unsigned int val)
 763 {
 764         char *buf = malloc(32);
 765         char post[] = { 0, 'K', 'M', 'G', 'P', 'E', 0 };
 766         char *p = post;
 767
 768         do {
 769                 if (val & 1023)
 770                         break;
 771
 772                 val >>= 10;
 773                 p++;
 774         } while (*p);
 775
 776         snprintf(buf, 31, "%u%c", val, *p);
 777         return buf;
 778 }
 779
 780 /* External engines are specified by "external:name.o") */
 781 static const char *get_engine_name(const char *str)
 782 {
 783         char *p = strstr(str, ":");
 784
 785         if (!p)
 786                 return str;
 787
 788         p++;
 789         strip_blank_front(&p);
 790         strip_blank_end(p);
 791         return p;
 792 }
 793
 794 /*
 795  * Adds a job to the list of things todo. Sanitizes the various options
 796  * to make sure we don't have conflicts, and initializes various
 797  * members of td.
 798  */
 799 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 800 {
 801         const char *ddir_str[] = { NULL, "read", "write", "rw", NULL,
 802                                    "randread", "randwrite", "randrw" };
 803         struct stat sb;
 804         int numjobs, i;
 805         struct fio_file *f;
 806         const char *engine;
 807
 808         /*
 809          * the def_thread is just for options, it's not a real job
 810          */
 811         if (td == &def_thread)
 812                 return 0;
 813
 814         engine = get_engine_name(td->ioengine);
 815         td->io_ops = load_ioengine(td, engine);
 816         if (!td->io_ops) {
 817                 log_err("fio: failed to load engine %s\n", engine);
 818                 return 1;
 819         }
 820
 821         if (td->use_thread)
 822                 nr_thread++;
 823         else
 824                 nr_process++;
 825
 826         if (td->odirect)
 827                 td->io_ops->flags |= FIO_RAWIO;
 828
 829         if (!td->filename)
 830                 td->filename = strdup(jobname);
 831
 832         td->filetype = FIO_TYPE_FILE;
 833         if (!lstat(td->filename, &sb)) {
 834                 if (S_ISBLK(sb.st_mode))
 835                         td->filetype = FIO_TYPE_BD;
 836                 else if (S_ISCHR(sb.st_mode))
 837                         td->filetype = FIO_TYPE_CHAR;
 838         }
 839
 840         fixup_options(td);
 841
 842         if (td->filename)
 843                 td->nr_uniq_files = 1;
 844         else
 845                 td->nr_uniq_files = td->open_files;
 846
 847         if (td->filetype == FIO_TYPE_FILE || td->filename) {
 848                 char tmp[PATH_MAX];
 849                 int len = 0;
 850
 851                 if (td->directory && td->directory[0] != '\0') {
 852                         if (lstat(td->directory, &sb) < 0) {
 853                                 log_err("fio: %s is not a directory\n", td->directory);
 854                                 td_verror(td, errno, "lstat");
 855                                 return 1;
 856                         }
 857                         if (!S_ISDIR(sb.st_mode)) {
 858                                 log_err("fio: %s is not a directory\n", td->directory);
 859                                 return 1;
 860                         }
 861                         len = sprintf(tmp, "%s/", td->directory);
 862                 }
 863
 864                 td->files = malloc(sizeof(struct fio_file) * td->open_files);
 865
 866                 for_each_file(td, f, i) {
 867                         memset(f, 0, sizeof(*f));
 868                         f->fd = -1;
 869
 870                         if (td->filename)
 871                                 sprintf(tmp + len, "%s", td->filename);
 872                         else
 873                                 sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i);
 874                         f->file_name = strdup(tmp);
 875                 }
 876         } else {
 877                 td->open_files = td->nr_files = 1;
 878                 td->files = malloc(sizeof(struct fio_file));
 879                 f = &td->files[0];
 880
 881                 memset(f, 0, sizeof(*f));
 882                 f->fd = -1;
 883                 f->file_name = strdup(jobname);
 884         }
 885
 886         for_each_file(td, f, i) {
 887                 f->file_size = td->total_file_size / td->nr_files;
 888                 f->file_offset = td->start_offset;
 889         }
 890
 891         td->mutex = fio_sem_init(0);
 892
 893         td->ts.clat_stat[0].min_val = td->ts.clat_stat[1].min_val = ULONG_MAX;
 894         td->ts.slat_stat[0].min_val = td->ts.slat_stat[1].min_val = ULONG_MAX;
 895         td->ts.bw_stat[0].min_val = td->ts.bw_stat[1].min_val = ULONG_MAX;
 896
 897         if (td->stonewall && td->thread_number > 1)
 898                 groupid++;
 899
 900         td->groupid = groupid;
 901
 902         if (setup_rate(td))
 903                 goto err;
 904
 905         if (td->write_lat_log) {
 906                 setup_log(&td->ts.slat_log);
 907                 setup_log(&td->ts.clat_log);
 908         }
 909         if (td->write_bw_log)
 910                 setup_log(&td->ts.bw_log);
 911
 912         if (!td->name)
 913                 td->name = strdup(jobname);
 914
 915         if (!terse_output) {
 916                 if (!job_add_num) {
 917                         if (!strcmp(td->io_ops->name, "cpuio"))
 918                                 fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
 919                         else {
 920                                 char *c1, *c2, *c3, *c4;
 921
 922                                 c1 = to_kmg(td->min_bs[DDIR_READ]);
 923                                 c2 = to_kmg(td->max_bs[DDIR_READ]);
 924                                 c3 = to_kmg(td->min_bs[DDIR_WRITE]);
 925                                 c4 = to_kmg(td->max_bs[DDIR_WRITE]);
 926
 927                                 fprintf(f_out, "%s: (g=%d): rw=%s, bs=%s-%s/%s-%s, ioengine=%s, iodepth=%u\n", td->name, td->groupid, ddir_str[td->td_ddir], c1, c2, c3, c4, td->io_ops->name, td->iodepth);
 928
 929                                 free(c1);
 930                                 free(c2);
 931                                 free(c3);
 932                                 free(c4);
 933                         }
 934                 } else if (job_add_num == 1)
 935                         fprintf(f_out, "...\n");
 936         }
 937
 938         /*
 939          * recurse add identical jobs, clear numjobs and stonewall options
 940          * as they don't apply to sub-jobs
 941          */
 942         numjobs = td->numjobs;
 943         while (--numjobs) {
 944                 struct thread_data *td_new = get_new_job(0, td);
 945
 946                 if (!td_new)
 947                         goto err;
 948
 949                 td_new->numjobs = 1;
 950                 td_new->stonewall = 0;
 951                 job_add_num = numjobs - 1;
 952
 953                 if (add_job(td_new, jobname, job_add_num))
 954                         goto err;
 955         }
 956         return 0;
 957 err:
 958         put_job(td);
 959         return -1;
 960 }
 961
 962 /*
 963  * Initialize the various random states we need (random io, block size ranges,
 964  * read/write mix, etc).
 965  */
 966 int init_random_state(struct thread_data *td)
 967 {
 968         unsigned long seeds[5];
 969         int fd, num_maps, blocks, i;
 970         struct fio_file *f;
 971
 972         if (td->io_ops->flags & FIO_DISKLESSIO)
 973                 return 0;
 974
 975         fd = open("/dev/urandom", O_RDONLY);
 976         if (fd == -1) {
 977                 td_verror(td, errno, "open");
 978                 return 1;
 979         }
 980
 981         if (read(fd, seeds, sizeof(seeds)) < (int) sizeof(seeds)) {
 982                 td_verror(td, EIO, "read");
 983                 close(fd);
 984                 return 1;
 985         }
 986
 987         close(fd);
 988
 989         os_random_seed(seeds[0], &td->bsrange_state);
 990         os_random_seed(seeds[1], &td->verify_state);
 991         os_random_seed(seeds[2], &td->rwmix_state);
 992
 993         if (td->file_service_type == FIO_FSERVICE_RANDOM)
 994                 os_random_seed(seeds[3], &td->next_file_state);
 995
 996         if (!td_random(td))
 997                 return 0;
 998
 999         if (td->rand_repeatable)
1000                 seeds[4] = FIO_RANDSEED * td->thread_number;
1001
1002         if (!td->norandommap) {
1003                 for_each_file(td, f, i) {
1004                         blocks = (f->real_file_size + td->rw_min_bs - 1) / td->rw_min_bs;
1005                         num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP;
1006                         f->file_map = malloc(num_maps * sizeof(long));
1007                         if (!f->file_map) {
1008                                 log_err("fio: failed allocating random map. If running a large number of jobs, try the 'norandommap' option\n");
1009                                 return 1;
1010                         }
1011                         f->num_maps = num_maps;
1012                         memset(f->file_map, 0, num_maps * sizeof(long));
1013                 }
1014         }
1015
1016         os_random_seed(seeds[4], &td->random_state);
1017         return 0;
1018 }
1019
1020 static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu)
1021 {
1022 #ifdef FIO_HAVE_CPU_AFFINITY
1023         unsigned int i;
1024
1025         CPU_ZERO(&cpumask);
1026
1027         for (i = 0; i < sizeof(int) * 8; i++) {
1028                 if ((1 << i) & cpu)
1029                         CPU_SET(i, &cpumask);
1030         }
1031 #endif
1032 }
1033
1034 static int is_empty_or_comment(char *line)
1035 {
1036         unsigned int i;
1037
1038         for (i = 0; i < strlen(line); i++) {
1039                 if (line[i] == ';')
1040                         return 1;
1041                 if (line[i] == '#')
1042                         return 1;
1043                 if (!isspace(line[i]) && !iscntrl(line[i]))
1044                         return 0;
1045         }
1046
1047         return 1;
1048 }
1049
1050 /*
1051  * Check if mmap/mmaphuge has a :/foo/bar/file at the end. If so, return that.
1052  */
1053 static char *get_opt_postfix(const char *str)
1054 {
1055         char *p = strstr(str, ":");
1056
1057         if (!p)
1058                 return NULL;
1059
1060         p++;
1061         strip_blank_front(&p);
1062         strip_blank_end(p);
1063         return strdup(p);
1064 }
1065
1066 static int str_mem_cb(void *data, const char *mem)
1067 {
1068         struct thread_data *td = data;
1069
1070         if (td->mem_type == MEM_MMAPHUGE || td->mem_type == MEM_MMAP) {
1071                 td->mmapfile = get_opt_postfix(mem);
1072                 if (td->mem_type == MEM_MMAPHUGE && !td->mmapfile) {
1073                         log_err("fio: mmaphuge:/path/to/file\n");
1074                         return 1;
1075                 }
1076         }
1077
1078         return 0;
1079 }
1080
1081 static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
1082 {
1083         mlock_size = *val;
1084         return 0;
1085 }
1086
1087 #ifdef FIO_HAVE_IOPRIO
1088 static int str_prioclass_cb(void *data, unsigned int *val)
1089 {
1090         struct thread_data *td = data;
1091
1092         td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
1093         return 0;
1094 }
1095
1096 static int str_prio_cb(void *data, unsigned int *val)
1097 {
1098         struct thread_data *td = data;
1099
1100         td->ioprio |= *val;
1101         return 0;
1102 }
1103 #endif
1104
1105 static int str_exitall_cb(void)
1106 {
1107         exitall_on_terminate = 1;
1108         return 0;
1109 }
1110
1111 static int str_cpumask_cb(void *data, unsigned int *val)
1112 {
1113         struct thread_data *td = data;
1114
1115         fill_cpu_mask(td->cpumask, *val);
1116         return 0;
1117 }
1118
1119 static int str_fst_cb(void *data, const char *str)
1120 {
1121         struct thread_data *td = data;
1122         char *nr = get_opt_postfix(str);
1123
1124         td->file_service_nr = 1;
1125         if (nr)
1126                 td->file_service_nr = atoi(nr);
1127
1128         return 0;
1129 }
1130
1131 /*
1132  * This is our [ini] type file parser.
1133  */
1134 static int parse_jobs_ini(char *file, int stonewall_flag)
1135 {
1136         unsigned int global;
1137         struct thread_data *td;
1138         char *string, *name;
1139         fpos_t off;
1140         FILE *f;
1141         char *p;
1142         int ret = 0, stonewall;
1143
1144         f = fopen(file, "r");
1145         if (!f) {
1146                 perror("fopen job file");
1147                 return 1;
1148         }
1149
1150         string = malloc(4096);
1151         name = malloc(256);
1152         memset(name, 0, 256);
1153
1154         stonewall = stonewall_flag;
1155         do {
1156                 p = fgets(string, 4095, f);
1157                 if (!p)
1158                         break;
1159                 if (is_empty_or_comment(p))
1160                         continue;
1161                 if (sscanf(p, "[%255s]", name) != 1)
1162                         continue;
1163
1164                 global = !strncmp(name, "global", 6);
1165
1166                 name[strlen(name) - 1] = '\0';
1167
1168                 td = get_new_job(global, &def_thread);
1169                 if (!td) {
1170                         ret = 1;
1171                         break;
1172                 }
1173
1174                 /*
1175                  * Seperate multiple job files by a stonewall
1176                  */
1177                 if (!global && stonewall) {
1178                         td->stonewall = stonewall;
1179                         stonewall = 0;
1180                 }
1181
1182                 fgetpos(f, &off);
1183                 while ((p = fgets(string, 4096, f)) != NULL) {
1184                         if (is_empty_or_comment(p))
1185                                 continue;
1186
1187                         strip_blank_front(&p);
1188
1189                         if (p[0] == '[')
1190                                 break;
1191
1192                         strip_blank_end(p);
1193
1194                         fgetpos(f, &off);
1195
1196                         /*
1197                          * Don't break here, continue parsing options so we
1198                          * dump all the bad ones. Makes trial/error fixups
1199                          * easier on the user.
1200                          */
1201                         ret |= parse_option(p, options, td);
1202                 }
1203
1204                 if (!ret) {
1205                         fsetpos(f, &off);
1206                         ret = add_job(td, name, 0);
1207                 } else {
1208                         log_err("fio: job %s dropped\n", name);
1209                         put_job(td);
1210                 }
1211         } while (!ret);
1212
1213         free(string);
1214         free(name);
1215         fclose(f);
1216         return ret;
1217 }
1218
1219 static int fill_def_thread(void)
1220 {
1221         memset(&def_thread, 0, sizeof(def_thread));
1222
1223         if (fio_getaffinity(getpid(), &def_thread.cpumask) == -1) {
1224                 perror("sched_getaffinity");
1225                 return 1;
1226         }
1227
1228         /*
1229          * fill default options
1230          */
1231         fill_default_options(&def_thread, options);
1232
1233         def_thread.timeout = def_timeout;
1234         def_thread.write_bw_log = write_bw_log;
1235         def_thread.write_lat_log = write_lat_log;
1236
1237 #ifdef FIO_HAVE_DISK_UTIL
1238         def_thread.do_disk_util = 1;
1239 #endif
1240
1241         return 0;
1242 }
1243
1244 static void usage(void)
1245 {
1246         printf("%s\n", fio_version_string);
1247         printf("\t--output\tWrite output to file\n");
1248         printf("\t--timeout\tRuntime in seconds\n");
1249         printf("\t--latency-log\tGenerate per-job latency logs\n");
1250         printf("\t--bandwidth-log\tGenerate per-job bandwidth logs\n");
1251         printf("\t--minimal\tMinimal (terse) output\n");
1252         printf("\t--version\tPrint version info and exit\n");
1253         printf("\t--help\t\tPrint this page\n");
1254         printf("\t--cmdhelp=cmd\tPrint command help, \"all\" for all of them\n");
1255 }
1256
1257 static int parse_cmd_line(int argc, char *argv[])
1258 {
1259         struct thread_data *td = NULL;
1260         int c, ini_idx = 0, lidx, ret, dont_add_job = 0;
1261
1262         while ((c = getopt_long_only(argc, argv, "", long_options, &lidx)) != -1) {
1263                 switch (c) {
1264                 case 't':
1265                         def_timeout = atoi(optarg);
1266                         break;
1267                 case 'l':
1268                         write_lat_log = 1;
1269                         break;
1270                 case 'w':
1271                         write_bw_log = 1;
1272                         break;
1273                 case 'o':
1274                         f_out = fopen(optarg, "w+");
1275                         if (!f_out) {
1276                                 perror("fopen output");
1277                                 exit(1);
1278                         }
1279                         f_err = f_out;
1280                         break;
1281                 case 'm':
1282                         terse_output = 1;
1283                         break;
1284                 case 'h':
1285                         usage();
1286                         exit(0);
1287                 case 'c':
1288                         ret = show_cmd_help(options, optarg);
1289                         exit(ret);
1290                 case 'v':
1291                         printf("%s\n", fio_version_string);
1292                         exit(0);
1293                 case FIO_GETOPT_JOB: {
1294                         const char *opt = long_options[lidx].name;
1295                         char *val = optarg;
1296
1297                         if (!strncmp(opt, "name", 4) && td) {
1298                                 ret = add_job(td, td->name ?: "fio", 0);
1299                                 if (ret) {
1300                                         put_job(td);
1301                                         return 0;
1302                                 }
1303                                 td = NULL;
1304                         }
1305                         if (!td) {
1306                                 int global = !strncmp(val, "global", 6);
1307
1308                                 td = get_new_job(global, &def_thread);
1309                                 if (!td)
1310                                         return 0;
1311                         }
1312
1313                         ret = parse_cmd_option(opt, val, options, td);
1314                         if (ret)
1315                                 dont_add_job = 1;
1316                         break;
1317                 }
1318                 default:
1319                         break;
1320                 }
1321         }
1322
1323         if (td) {
1324                 if (dont_add_job)
1325                         put_job(td);
1326                 else {
1327                         ret = add_job(td, td->name ?: "fio", 0);
1328                         if (ret)
1329                                 put_job(td);
1330                 }
1331         }
1332
1333         while (optind < argc) {
1334                 ini_idx++;
1335                 ini_file = realloc(ini_file, ini_idx * sizeof(char *));
1336                 ini_file[ini_idx - 1] = strdup(argv[optind]);
1337                 optind++;
1338         }
1339
1340         return ini_idx;
1341 }
1342
1343 static void free_shm(void)
1344 {
1345         struct shmid_ds sbuf;
1346
1347         if (threads) {
1348                 shmdt((void *) threads);
1349                 threads = NULL;
1350                 shmctl(shm_id, IPC_RMID, &sbuf);
1351         }
1352 }
1353
1354 /*
1355  * The thread area is shared between the main process and the job
1356  * threads/processes. So setup a shared memory segment that will hold
1357  * all the job info.
1358  */
1359 static int setup_thread_area(void)
1360 {
1361         /*
1362          * 1024 is too much on some machines, scale max_jobs if
1363          * we get a failure that looks like too large a shm segment
1364          */
1365         do {
1366                 size_t size = max_jobs * sizeof(struct thread_data);
1367
1368                 shm_id = shmget(0, size, IPC_CREAT | 0600);
1369                 if (shm_id != -1)
1370                         break;
1371                 if (errno != EINVAL) {
1372                         perror("shmget");
1373                         break;
1374                 }
1375
1376                 max_jobs >>= 1;
1377         } while (max_jobs);
1378
1379         if (shm_id == -1)
1380                 return 1;
1381
1382         threads = shmat(shm_id, NULL, 0);
1383         if (threads == (void *) -1) {
1384                 perror("shmat");
1385                 return 1;
1386         }
1387
1388         atexit(free_shm);
1389         return 0;
1390 }
1391
1392 /*
1393  * Copy the fio options into the long options map, so we mirror
1394  * job and cmd line options.
1395  */
1396 static void dupe_job_options(void)
1397 {
1398         struct fio_option *o;
1399         unsigned int i;
1400
1401         i = 0;
1402         while (long_options[i].name)
1403                 i++;
1404
1405         o = &options[0];
1406         while (o->name) {
1407                 long_options[i].name = o->name;
1408                 long_options[i].val = FIO_GETOPT_JOB;
1409                 if (o->type == FIO_OPT_STR_SET)
1410                         long_options[i].has_arg = no_argument;
1411                 else
1412                         long_options[i].has_arg = required_argument;
1413
1414                 i++;
1415                 o++;
1416                 assert(i < FIO_JOB_OPTS + FIO_CMD_OPTS);
1417         }
1418 }
1419
1420 int parse_options(int argc, char *argv[])
1421 {
1422         int job_files, i;
1423
1424         f_out = stdout;
1425         f_err = stderr;
1426
1427         options_init(options);
1428
1429         dupe_job_options();
1430
1431         if (setup_thread_area())
1432                 return 1;
1433         if (fill_def_thread())
1434                 return 1;
1435
1436         job_files = parse_cmd_line(argc, argv);
1437
1438         for (i = 0; i < job_files; i++) {
1439                 if (fill_def_thread())
1440                         return 1;
1441                 if (parse_jobs_ini(ini_file[i], i))
1442                         return 1;
1443                 free(ini_file[i]);
1444         }
1445
1446         free(ini_file);
1447
1448         if (!thread_number) {
1449                 log_err("No jobs defined(s)\n");
1450                 return 1;
1451         }
1452
1453         return 0;
1454 }