| 1 | #ifndef FIO_H |
| 2 | #define FIO_H |
| 3 | |
| 4 | #include <sched.h> |
| 5 | #include <limits.h> |
| 6 | #include <pthread.h> |
| 7 | #include <sys/time.h> |
| 8 | #include <sys/resource.h> |
| 9 | #include <errno.h> |
| 10 | #include <stdlib.h> |
| 11 | #include <stdio.h> |
| 12 | #include <unistd.h> |
| 13 | #include <string.h> |
| 14 | #include <inttypes.h> |
| 15 | #include <assert.h> |
| 16 | |
| 17 | struct thread_data; |
| 18 | |
| 19 | #include "compiler/compiler.h" |
| 20 | #include "thread_options.h" |
| 21 | #include "flist.h" |
| 22 | #include "fifo.h" |
| 23 | #include "lib/rbtree.h" |
| 24 | #include "arch/arch.h" |
| 25 | #include "os/os.h" |
| 26 | #include "mutex.h" |
| 27 | #include "log.h" |
| 28 | #include "debug.h" |
| 29 | #include "file.h" |
| 30 | #include "io_ddir.h" |
| 31 | #include "ioengine.h" |
| 32 | #include "iolog.h" |
| 33 | #include "helpers.h" |
| 34 | #include "options.h" |
| 35 | #include "profile.h" |
| 36 | #include "time.h" |
| 37 | #include "gettime.h" |
| 38 | #include "lib/getopt.h" |
| 39 | #include "lib/rand.h" |
| 40 | #include "client.h" |
| 41 | #include "server.h" |
| 42 | #include "stat.h" |
| 43 | #include "flow.h" |
| 44 | |
| 45 | #ifdef FIO_HAVE_GUASI |
| 46 | #include <guasi.h> |
| 47 | #endif |
| 48 | |
| 49 | #ifdef FIO_HAVE_SOLARISAIO |
| 50 | #include <sys/asynch.h> |
| 51 | #endif |
| 52 | |
| 53 | /* |
| 54 | * offset generator types |
| 55 | */ |
| 56 | enum { |
| 57 | RW_SEQ_SEQ = 0, |
| 58 | RW_SEQ_IDENT, |
| 59 | }; |
| 60 | |
| 61 | /* |
| 62 | * This describes a single thread/process executing a fio job. |
| 63 | */ |
| 64 | struct thread_data { |
| 65 | struct thread_options o; |
| 66 | void *eo; |
| 67 | char verror[FIO_VERROR_SIZE]; |
| 68 | pthread_t thread; |
| 69 | unsigned int thread_number; |
| 70 | unsigned int groupid; |
| 71 | struct thread_stat ts; |
| 72 | |
| 73 | int client_type; |
| 74 | |
| 75 | struct io_log *slat_log; |
| 76 | struct io_log *clat_log; |
| 77 | struct io_log *lat_log; |
| 78 | struct io_log *bw_log; |
| 79 | struct io_log *iops_log; |
| 80 | |
| 81 | uint64_t stat_io_bytes[2]; |
| 82 | struct timeval bw_sample_time; |
| 83 | |
| 84 | uint64_t stat_io_blocks[2]; |
| 85 | struct timeval iops_sample_time; |
| 86 | |
| 87 | struct rusage ru_start; |
| 88 | struct rusage ru_end; |
| 89 | |
| 90 | struct fio_file **files; |
| 91 | unsigned int files_size; |
| 92 | unsigned int files_index; |
| 93 | unsigned int nr_open_files; |
| 94 | unsigned int nr_done_files; |
| 95 | unsigned int nr_normal_files; |
| 96 | union { |
| 97 | unsigned int next_file; |
| 98 | os_random_state_t next_file_state; |
| 99 | struct frand_state __next_file_state; |
| 100 | }; |
| 101 | int error; |
| 102 | int done; |
| 103 | pid_t pid; |
| 104 | char *orig_buffer; |
| 105 | size_t orig_buffer_size; |
| 106 | volatile int terminate; |
| 107 | volatile int runstate; |
| 108 | unsigned int ioprio; |
| 109 | unsigned int ioprio_set; |
| 110 | unsigned int last_was_sync; |
| 111 | enum fio_ddir last_ddir; |
| 112 | |
| 113 | char *mmapfile; |
| 114 | int mmapfd; |
| 115 | |
| 116 | void *iolog_buf; |
| 117 | FILE *iolog_f; |
| 118 | |
| 119 | char *sysfs_root; |
| 120 | |
| 121 | unsigned long rand_seeds[8]; |
| 122 | |
| 123 | union { |
| 124 | os_random_state_t bsrange_state; |
| 125 | struct frand_state __bsrange_state; |
| 126 | }; |
| 127 | union { |
| 128 | os_random_state_t verify_state; |
| 129 | struct frand_state __verify_state; |
| 130 | }; |
| 131 | union { |
| 132 | os_random_state_t trim_state; |
| 133 | struct frand_state __trim_state; |
| 134 | }; |
| 135 | |
| 136 | struct frand_state buf_state; |
| 137 | |
| 138 | unsigned int verify_batch; |
| 139 | unsigned int trim_batch; |
| 140 | |
| 141 | int shm_id; |
| 142 | |
| 143 | /* |
| 144 | * IO engine hooks, contains everything needed to submit an io_u |
| 145 | * to any of the available IO engines. |
| 146 | */ |
| 147 | struct ioengine_ops *io_ops; |
| 148 | |
| 149 | /* |
| 150 | * Queue depth of io_u's that fio MIGHT do |
| 151 | */ |
| 152 | unsigned int cur_depth; |
| 153 | |
| 154 | /* |
| 155 | * io_u's about to be committed |
| 156 | */ |
| 157 | unsigned int io_u_queued; |
| 158 | |
| 159 | /* |
| 160 | * io_u's submitted but not completed yet |
| 161 | */ |
| 162 | unsigned int io_u_in_flight; |
| 163 | |
| 164 | /* |
| 165 | * List of free and busy io_u's |
| 166 | */ |
| 167 | struct flist_head io_u_freelist; |
| 168 | struct flist_head io_u_busylist; |
| 169 | struct flist_head io_u_requeues; |
| 170 | pthread_mutex_t io_u_lock; |
| 171 | pthread_cond_t free_cond; |
| 172 | |
| 173 | /* |
| 174 | * async verify offload |
| 175 | */ |
| 176 | struct flist_head verify_list; |
| 177 | pthread_t *verify_threads; |
| 178 | unsigned int nr_verify_threads; |
| 179 | pthread_cond_t verify_cond; |
| 180 | int verify_thread_exit; |
| 181 | |
| 182 | /* |
| 183 | * Rate state |
| 184 | */ |
| 185 | unsigned long long rate_bps[2]; |
| 186 | long rate_pending_usleep[2]; |
| 187 | unsigned long rate_bytes[2]; |
| 188 | unsigned long rate_blocks[2]; |
| 189 | struct timeval lastrate[2]; |
| 190 | |
| 191 | unsigned long long total_io_size; |
| 192 | unsigned long long fill_device_size; |
| 193 | |
| 194 | unsigned long io_issues[2]; |
| 195 | unsigned long long io_blocks[2]; |
| 196 | unsigned long long this_io_blocks[2]; |
| 197 | unsigned long long io_bytes[2]; |
| 198 | unsigned long long io_skip_bytes; |
| 199 | unsigned long long this_io_bytes[2]; |
| 200 | unsigned long long zone_bytes; |
| 201 | struct fio_mutex *mutex; |
| 202 | |
| 203 | /* |
| 204 | * State for random io, a bitmap of blocks done vs not done |
| 205 | */ |
| 206 | union { |
| 207 | os_random_state_t random_state; |
| 208 | struct frand_state __random_state; |
| 209 | }; |
| 210 | |
| 211 | struct timeval start; /* start of this loop */ |
| 212 | struct timeval epoch; /* time job was started */ |
| 213 | struct timeval last_issue; |
| 214 | struct timeval tv_cache; |
| 215 | unsigned int tv_cache_nr; |
| 216 | unsigned int tv_cache_mask; |
| 217 | unsigned int ramp_time_over; |
| 218 | |
| 219 | /* |
| 220 | * read/write mixed workload state |
| 221 | */ |
| 222 | union { |
| 223 | os_random_state_t rwmix_state; |
| 224 | struct frand_state __rwmix_state; |
| 225 | }; |
| 226 | unsigned long rwmix_issues; |
| 227 | enum fio_ddir rwmix_ddir; |
| 228 | unsigned int ddir_seq_nr; |
| 229 | |
| 230 | /* |
| 231 | * IO history logs for verification. We use a tree for sorting, |
| 232 | * if we are overwriting. Otherwise just use a fifo. |
| 233 | */ |
| 234 | struct rb_root io_hist_tree; |
| 235 | struct flist_head io_hist_list; |
| 236 | unsigned long io_hist_len; |
| 237 | |
| 238 | /* |
| 239 | * For IO replaying |
| 240 | */ |
| 241 | struct flist_head io_log_list; |
| 242 | |
| 243 | /* |
| 244 | * For tracking/handling discards |
| 245 | */ |
| 246 | struct flist_head trim_list; |
| 247 | unsigned long trim_entries; |
| 248 | |
| 249 | /* |
| 250 | * for fileservice, how often to switch to a new file |
| 251 | */ |
| 252 | unsigned int file_service_nr; |
| 253 | unsigned int file_service_left; |
| 254 | struct fio_file *file_service_file; |
| 255 | |
| 256 | unsigned int sync_file_range_nr; |
| 257 | |
| 258 | /* |
| 259 | * For generating file sizes |
| 260 | */ |
| 261 | union { |
| 262 | os_random_state_t file_size_state; |
| 263 | struct frand_state __file_size_state; |
| 264 | }; |
| 265 | |
| 266 | /* |
| 267 | * Error counts |
| 268 | */ |
| 269 | unsigned int total_err_count; |
| 270 | int first_error; |
| 271 | |
| 272 | struct fio_flow *flow; |
| 273 | |
| 274 | /* |
| 275 | * Can be overloaded by profiles |
| 276 | */ |
| 277 | struct prof_io_ops prof_io_ops; |
| 278 | void *prof_data; |
| 279 | }; |
| 280 | |
| 281 | /* |
| 282 | * when should interactive ETA output be generated |
| 283 | */ |
| 284 | enum { |
| 285 | FIO_ETA_AUTO, |
| 286 | FIO_ETA_ALWAYS, |
| 287 | FIO_ETA_NEVER, |
| 288 | }; |
| 289 | |
| 290 | #define __td_verror(td, err, msg, func) \ |
| 291 | do { \ |
| 292 | if ((td)->error) \ |
| 293 | break; \ |
| 294 | int e = (err); \ |
| 295 | (td)->error = e; \ |
| 296 | if (!(td)->first_error) \ |
| 297 | snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \ |
| 298 | } while (0) |
| 299 | |
| 300 | |
| 301 | #define td_clear_error(td) \ |
| 302 | (td)->error = 0; |
| 303 | #define td_verror(td, err, func) \ |
| 304 | __td_verror((td), (err), strerror((err)), (func)) |
| 305 | #define td_vmsg(td, err, msg, func) \ |
| 306 | __td_verror((td), (err), (msg), (func)) |
| 307 | |
| 308 | #define __fio_stringify_1(x) #x |
| 309 | #define __fio_stringify(x) __fio_stringify_1(x) |
| 310 | |
| 311 | extern int exitall_on_terminate; |
| 312 | extern unsigned int thread_number; |
| 313 | extern int shm_id; |
| 314 | extern int groupid; |
| 315 | extern int terse_output; |
| 316 | extern int temp_stall_ts; |
| 317 | extern unsigned long long mlock_size; |
| 318 | extern unsigned long page_mask, page_size; |
| 319 | extern int read_only; |
| 320 | extern int eta_print; |
| 321 | extern unsigned long done_secs; |
| 322 | extern char *job_section; |
| 323 | extern int fio_gtod_offload; |
| 324 | extern int fio_gtod_cpu; |
| 325 | extern enum fio_cs fio_clock_source; |
| 326 | extern int warnings_fatal; |
| 327 | extern int terse_version; |
| 328 | extern int is_backend; |
| 329 | extern int nr_clients; |
| 330 | extern int log_syslog; |
| 331 | extern const char fio_version_string[]; |
| 332 | extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN]; |
| 333 | |
| 334 | extern struct thread_data *threads; |
| 335 | |
| 336 | static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u) |
| 337 | { |
| 338 | assert(!(io_u->ddir == DDIR_WRITE && !td_write(td))); |
| 339 | } |
| 340 | |
| 341 | #define BLOCKS_PER_MAP (8 * sizeof(unsigned long)) |
| 342 | #define TO_MAP_BLOCK(f, b) (b) |
| 343 | #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP) |
| 344 | #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1)) |
| 345 | |
| 346 | #define REAL_MAX_JOBS 2048 |
| 347 | |
| 348 | #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ) |
| 349 | |
| 350 | static inline enum error_type td_error_type(enum fio_ddir ddir, int err) |
| 351 | { |
| 352 | if (err == EILSEQ) |
| 353 | return ERROR_TYPE_VERIFY; |
| 354 | if (ddir == DDIR_READ) |
| 355 | return ERROR_TYPE_READ; |
| 356 | return ERROR_TYPE_WRITE; |
| 357 | } |
| 358 | |
| 359 | static inline void update_error_count(struct thread_data *td, int err) |
| 360 | { |
| 361 | td->total_err_count++; |
| 362 | if (td->total_err_count == 1) |
| 363 | td->first_error = err; |
| 364 | } |
| 365 | |
| 366 | static inline int should_fsync(struct thread_data *td) |
| 367 | { |
| 368 | if (td->last_was_sync) |
| 369 | return 0; |
| 370 | if (td->o.odirect) |
| 371 | return 0; |
| 372 | if (td_write(td) || td_rw(td) || td->o.override_sync) |
| 373 | return 1; |
| 374 | |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | /* |
| 379 | * Init/option functions |
| 380 | */ |
| 381 | extern int __must_check fio_init_options(void); |
| 382 | extern int __must_check parse_options(int, char **); |
| 383 | extern int parse_jobs_ini(char *, int, int, int); |
| 384 | extern int parse_cmd_line(int, char **, int); |
| 385 | extern int fio_backend(void); |
| 386 | extern void reset_fio_state(void); |
| 387 | extern void clear_io_state(struct thread_data *); |
| 388 | extern int fio_options_parse(struct thread_data *, char **, int); |
| 389 | extern void fio_keywords_init(void); |
| 390 | extern int fio_cmd_option_parse(struct thread_data *, const char *, char *); |
| 391 | extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *); |
| 392 | extern void fio_fill_default_options(struct thread_data *); |
| 393 | extern int fio_show_option_help(const char *); |
| 394 | extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td); |
| 395 | extern void fio_options_dup_and_init(struct option *); |
| 396 | extern void fio_options_mem_dupe(struct thread_data *); |
| 397 | extern void options_mem_dupe(void *data, struct fio_option *options); |
| 398 | extern void td_fill_rand_seeds(struct thread_data *); |
| 399 | extern void add_job_opts(const char **, int); |
| 400 | extern char *num2str(unsigned long, int, int, int); |
| 401 | extern int ioengine_load(struct thread_data *); |
| 402 | |
| 403 | extern unsigned long page_mask; |
| 404 | extern unsigned long page_size; |
| 405 | extern int initialize_fio(char *envp[]); |
| 406 | |
| 407 | #define FIO_GETOPT_JOB 0x89000000 |
| 408 | #define FIO_GETOPT_IOENGINE 0x98000000 |
| 409 | #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128) |
| 410 | |
| 411 | /* |
| 412 | * ETA/status stuff |
| 413 | */ |
| 414 | extern void print_thread_status(void); |
| 415 | extern void print_status_init(int); |
| 416 | extern char *fio_uint_to_kmg(unsigned int val); |
| 417 | |
| 418 | /* |
| 419 | * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it |
| 420 | * will never back again. It may cycle between running/verififying/fsyncing. |
| 421 | * Once the thread reaches TD_EXITED, it is just waiting for the core to |
| 422 | * reap it. |
| 423 | */ |
| 424 | enum { |
| 425 | TD_NOT_CREATED = 0, |
| 426 | TD_CREATED, |
| 427 | TD_INITIALIZED, |
| 428 | TD_RAMP, |
| 429 | TD_RUNNING, |
| 430 | TD_PRE_READING, |
| 431 | TD_VERIFYING, |
| 432 | TD_FSYNCING, |
| 433 | TD_EXITED, |
| 434 | TD_REAPED, |
| 435 | }; |
| 436 | |
| 437 | extern void td_set_runstate(struct thread_data *, int); |
| 438 | #define TERMINATE_ALL (-1) |
| 439 | extern void fio_terminate_threads(int); |
| 440 | |
| 441 | /* |
| 442 | * Memory helpers |
| 443 | */ |
| 444 | extern int __must_check fio_pin_memory(void); |
| 445 | extern void fio_unpin_memory(void); |
| 446 | extern int __must_check allocate_io_mem(struct thread_data *); |
| 447 | extern void free_io_mem(struct thread_data *); |
| 448 | |
| 449 | /* |
| 450 | * Reset stats after ramp time completes |
| 451 | */ |
| 452 | extern void reset_all_stats(struct thread_data *); |
| 453 | |
| 454 | /* |
| 455 | * blktrace support |
| 456 | */ |
| 457 | #ifdef FIO_HAVE_BLKTRACE |
| 458 | extern int is_blktrace(const char *); |
| 459 | extern int load_blktrace(struct thread_data *, const char *); |
| 460 | #endif |
| 461 | |
| 462 | /* |
| 463 | * Mark unused variables passed to ops functions as unused, to silence gcc |
| 464 | */ |
| 465 | #define fio_unused __attribute((__unused__)) |
| 466 | #define fio_init __attribute__((constructor)) |
| 467 | #define fio_exit __attribute__((destructor)) |
| 468 | |
| 469 | #define for_each_td(td, i) \ |
| 470 | for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++) |
| 471 | #define for_each_file(td, f, i) \ |
| 472 | if ((td)->files_index) \ |
| 473 | for ((i) = 0, (f) = (td)->files[0]; \ |
| 474 | (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \ |
| 475 | (i)++) |
| 476 | |
| 477 | #define fio_assert(td, cond) do { \ |
| 478 | if (!(cond)) { \ |
| 479 | int *__foo = NULL; \ |
| 480 | fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \ |
| 481 | td_set_runstate((td), TD_EXITED); \ |
| 482 | (td)->error = EFAULT; \ |
| 483 | *__foo = 0; \ |
| 484 | } \ |
| 485 | } while (0) |
| 486 | |
| 487 | static inline int fio_fill_issue_time(struct thread_data *td) |
| 488 | { |
| 489 | if (td->o.read_iolog_file || |
| 490 | !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw) |
| 491 | return 1; |
| 492 | |
| 493 | return 0; |
| 494 | } |
| 495 | |
| 496 | static inline int __should_check_rate(struct thread_data *td, |
| 497 | enum fio_ddir ddir) |
| 498 | { |
| 499 | struct thread_options *o = &td->o; |
| 500 | |
| 501 | /* |
| 502 | * If some rate setting was given, we need to check it |
| 503 | */ |
| 504 | if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] || |
| 505 | o->rate_iops_min[ddir]) |
| 506 | return 1; |
| 507 | |
| 508 | return 0; |
| 509 | } |
| 510 | |
| 511 | static inline int should_check_rate(struct thread_data *td, |
| 512 | unsigned long *bytes_done) |
| 513 | { |
| 514 | int ret = 0; |
| 515 | |
| 516 | if (bytes_done[0]) |
| 517 | ret |= __should_check_rate(td, 0); |
| 518 | if (bytes_done[1]) |
| 519 | ret |= __should_check_rate(td, 1); |
| 520 | |
| 521 | return ret; |
| 522 | } |
| 523 | |
| 524 | static inline int is_power_of_2(unsigned int val) |
| 525 | { |
| 526 | return (val != 0 && ((val & (val - 1)) == 0)); |
| 527 | } |
| 528 | |
| 529 | /* |
| 530 | * We currently only need to do locking if we have verifier threads |
| 531 | * accessing our internal structures too |
| 532 | */ |
| 533 | static inline void td_io_u_lock(struct thread_data *td) |
| 534 | { |
| 535 | if (td->o.verify_async) |
| 536 | pthread_mutex_lock(&td->io_u_lock); |
| 537 | } |
| 538 | |
| 539 | static inline void td_io_u_unlock(struct thread_data *td) |
| 540 | { |
| 541 | if (td->o.verify_async) |
| 542 | pthread_mutex_unlock(&td->io_u_lock); |
| 543 | } |
| 544 | |
| 545 | static inline void td_io_u_free_notify(struct thread_data *td) |
| 546 | { |
| 547 | if (td->o.verify_async) |
| 548 | pthread_cond_signal(&td->free_cond); |
| 549 | } |
| 550 | |
| 551 | extern const char *fio_get_arch_string(int); |
| 552 | extern const char *fio_get_os_string(int); |
| 553 | |
| 554 | #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0]))) |
| 555 | |
| 556 | #endif |