8 #include <sys/resource.h>
28 * Use for maintaining statistics
31 unsigned long max_val;
32 unsigned long min_val;
33 unsigned long samples;
40 * A single data sample
49 * Dynamically growing data sample log
52 unsigned long nr_samples;
53 unsigned long max_samples;
54 struct io_sample *log;
58 * When logging io actions, this matches a single sent io_u
61 struct list_head list;
62 struct fio_file *file;
63 unsigned long long offset;
73 #ifdef FIO_HAVE_LIBAIO
76 #ifdef FIO_HAVE_POSIXAIO
83 struct timeval start_time;
84 struct timeval issue_time;
87 * Allocated/set buffer and length
91 unsigned long long offset;
94 * IO engine state, may be different from above when we get
95 * partial transfers / residual data counts
98 unsigned int xfer_buflen;
106 * io engine private data
113 struct fio_file *file;
115 struct list_head list;
118 #define FIO_HDR_MAGIC 0xf00baaef
121 VERIFY_NONE = 0, /* no verification */
122 VERIFY_MD5, /* md5 sum data blocks */
123 VERIFY_CRC32, /* crc32 sum data blocks */
127 * A header structure associated with each checksummed data block
129 struct verify_header {
130 unsigned int fio_magic;
132 unsigned int verify_type;
134 char md5_digest[MD5_HASH_WORDS * 4];
139 struct group_run_stats {
140 unsigned long long max_run[2], min_run[2];
141 unsigned long long max_bw[2], min_bw[2];
142 unsigned long long io_kb[2];
143 unsigned long long agg[2];
147 * What type of allocation to use for io buffers
150 MEM_MALLOC = 0, /* ordinary malloc */
151 MEM_SHM, /* use shared memory segments */
152 MEM_SHMHUGE, /* use shared memory segments with huge pages */
153 MEM_MMAP, /* use anonynomous mmap */
154 MEM_MMAPHUGE, /* memory mapped huge file */
158 * The type of object we are working on
161 FIO_TYPE_FILE = 1, /* plain file */
162 FIO_TYPE_BD, /* block device */
163 FIO_TYPE_CHAR, /* character device */
166 enum fio_ioengine_flags {
167 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
168 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
169 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
170 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
171 FIO_NETIO = 1 << 4, /* networked io */
172 FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
176 * Each thread_data structure has a number of files associated with it,
177 * this structure holds state information for a single file.
181 * A file may not be a file descriptor, let the io engine decide
184 unsigned long file_data;
189 unsigned long long file_size;
190 unsigned long long real_file_size;
191 unsigned long long file_offset;
192 unsigned long long last_pos;
193 unsigned long long last_completed_pos;
196 * block map for random io
198 unsigned long *file_map;
199 unsigned int num_maps;
200 unsigned int last_free_lookup;
206 * How many depth levels to log
208 #define FIO_IO_U_MAP_NR 8
211 * This describes a single thread/process executing a fio job.
222 enum fio_filetype filetype;
223 struct fio_file *files;
224 unsigned int nr_files;
225 unsigned int nr_uniq_files;
226 unsigned int next_file;
230 size_t orig_buffer_size;
231 volatile int terminate;
232 volatile int runstate;
236 unsigned int last_was_sync;
238 unsigned int sequential;
239 unsigned int odirect;
240 unsigned int invalidate_cache;
241 unsigned int create_serialize;
242 unsigned int create_fsync;
243 unsigned int end_fsync;
244 unsigned int sync_io;
246 unsigned int use_thread;
248 unsigned int do_disk_util;
249 unsigned int override_sync;
250 unsigned int rand_repeatable;
251 unsigned int write_lat_log;
252 unsigned int write_bw_log;
253 unsigned int norandommap;
254 unsigned int bs_unaligned;
257 unsigned int min_bs[2];
258 unsigned int max_bs[2];
259 unsigned int hugepage_size;
260 unsigned int rw_min_bs;
261 unsigned int thinktime;
262 unsigned int thinktime_blocks;
263 unsigned int fsync_blocks;
264 unsigned int start_delay;
265 unsigned long timeout;
266 unsigned int overwrite;
267 unsigned int bw_avg_time;
269 unsigned long long zone_size;
270 unsigned long long zone_skip;
271 enum fio_memtype mem_type;
274 unsigned int stonewall;
275 unsigned int numjobs;
276 unsigned int iodepth;
277 os_cpu_mask_t cpumask;
279 unsigned int read_iolog;
280 unsigned int rwmixcycle;
281 unsigned int rwmixread;
282 unsigned int rwmixwrite;
285 char *read_iolog_file;
286 char *write_iolog_file;
293 os_random_state_t bsrange_state;
294 os_random_state_t verify_state;
299 * IO engine hooks, contains everything needed to submit an io_u
300 * to any of the available IO engines.
302 struct ioengine_ops *io_ops;
305 * Current IO depth and list of free and busy io_u's.
307 unsigned int cur_depth;
308 unsigned int io_u_map[FIO_IO_U_MAP_NR];
309 unsigned long total_io_u;
310 struct list_head io_u_freelist;
311 struct list_head io_u_busylist;
317 unsigned int ratemin;
318 unsigned int ratecycle;
319 unsigned long rate_usec_cycle;
320 long rate_pending_usleep;
321 unsigned long rate_bytes;
322 struct timeval lastrate;
324 unsigned long runtime[2]; /* msec */
325 unsigned long long io_size;
326 unsigned long long total_file_size;
327 unsigned long long start_offset;
328 unsigned long long total_io_size;
330 unsigned long long io_blocks[2];
331 unsigned long long io_bytes[2];
332 unsigned long long zone_bytes;
333 unsigned long long this_io_bytes[2];
337 * State for random io, a bitmap of blocks done vs not done
339 os_random_state_t random_state;
342 * CPU "io" cycle burner
344 unsigned int cpuload;
345 unsigned int cpucycle;
348 * bandwidth and latency stats
350 struct io_stat clat_stat[2]; /* completion latency */
351 struct io_stat slat_stat[2]; /* submission latency */
352 struct io_stat bw_stat[2]; /* bandwidth stats */
354 unsigned long long stat_io_bytes[2];
355 struct timeval stat_sample_time[2];
357 struct io_log *slat_log;
358 struct io_log *clat_log;
359 struct io_log *bw_log;
361 struct timeval start; /* start of this loop */
362 struct timeval epoch; /* time job was started */
363 struct timeval end_time;/* time job ended */
366 * fio system usage accounting
368 struct rusage ru_start;
369 struct rusage ru_end;
370 unsigned long usr_time;
371 unsigned long sys_time;
375 * read/write mixed workload state
377 os_random_state_t rwmix_state;
378 struct timeval rwmix_switch;
379 enum fio_ddir rwmix_ddir;
382 * Pre-run and post-run shell
390 struct list_head io_hist_list;
391 struct list_head io_log_list;
394 #define __td_verror(td, err, msg) \
398 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, (msg)); \
402 #define td_verror(td, err) __td_verror((td), (err), strerror((err)))
403 #define td_vmsg(td, err, msg) __td_verror((td), (err), (msg))
405 extern int exitall_on_terminate;
406 extern int thread_number;
409 extern int terse_output;
412 extern int temp_stall_ts;
413 extern unsigned long long mlock_size;
415 extern struct thread_data *threads;
417 #define td_read(td) ((td)->ddir == DDIR_READ)
418 #define td_write(td) ((td)->ddir == DDIR_WRITE)
419 #define td_rw(td) ((td)->iomix != 0)
421 #define BLOCKS_PER_MAP (8 * sizeof(long))
422 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
423 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
424 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
426 #define MAX_JOBS (1024)
428 static inline int should_fsync(struct thread_data *td)
430 if (td->last_was_sync)
434 if (td_write(td) || td_rw(td) || td->override_sync)
441 * Disk utils as read in /sys/block/<dev>/stat
443 struct disk_util_stat {
446 unsigned long long sectors[2];
449 unsigned time_in_queue;
453 * Per-device disk util management
456 struct list_head list;
462 struct disk_util_stat dus;
463 struct disk_util_stat last_dus;
470 * Used for passing io_u completion data
472 struct io_completion_data {
475 int error; /* output */
476 unsigned long bytes_done[2]; /* output */
477 struct timeval time; /* output */
480 #define DISK_UTIL_MSEC (250)
483 #define min(a, b) ((a) < (b) ? (a) : (b))
486 #define max(a, b) ((a) > (b) ? (a) : (b))
492 extern int read_iolog_get(struct thread_data *, struct io_u *);
493 extern void write_iolog_put(struct thread_data *, struct io_u *);
494 extern int init_iolog(struct thread_data *td);
495 extern void log_io_piece(struct thread_data *, struct io_u *);
496 extern void prune_io_piece_log(struct thread_data *);
497 extern void write_iolog_close(struct thread_data *);
502 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
503 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
504 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
505 extern void show_run_stats(void);
506 extern void init_disk_util(struct thread_data *);
507 extern void update_rusage_stat(struct thread_data *);
508 extern void update_io_ticks(void);
509 extern void disk_util_timer_arm(void);
510 extern void setup_log(struct io_log **);
511 extern void finish_log(struct thread_data *, struct io_log *, const char *);
512 extern void __finish_log(struct io_log *, const char *);
513 extern int setup_rate(struct thread_data *);
514 extern struct io_log *agg_io_log[2];
515 extern int write_bw_log;
516 extern void add_agg_sample(unsigned long, enum fio_ddir);
521 extern unsigned long utime_since(struct timeval *, struct timeval *);
522 extern unsigned long utime_since_now(struct timeval *);
523 extern unsigned long mtime_since(struct timeval *, struct timeval *);
524 extern unsigned long mtime_since_now(struct timeval *);
525 extern unsigned long time_since_now(struct timeval *);
526 extern unsigned long mtime_since_genesis(void);
527 extern void __usec_sleep(unsigned int);
528 extern void usec_sleep(struct thread_data *, unsigned long);
529 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
530 extern void fill_start_time(struct timeval *);
531 extern void fio_gettime(struct timeval *, void *);
536 extern int parse_options(int, char **);
537 extern int init_random_state(struct thread_data *);
540 * File setup/shutdown
542 extern void close_files(struct thread_data *);
543 extern int setup_files(struct thread_data *);
544 extern int open_files(struct thread_data *);
545 extern int file_invalidate_cache(struct thread_data *, struct fio_file *);
550 extern void print_thread_status(void);
551 extern void print_status_init(int);
554 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
555 * will never back again. It may cycle between running/verififying/fsyncing.
556 * Once the thread reaches TD_EXITED, it is just waiting for the core to
573 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
574 extern int get_next_verify(struct thread_data *td, struct io_u *);
575 extern int do_io_u_verify(struct thread_data *, struct io_u **);
580 extern int fio_pin_memory(void);
581 extern void fio_unpin_memory(void);
582 extern int allocate_io_mem(struct thread_data *);
583 extern void free_io_mem(struct thread_data *);
588 #define queue_full(td) list_empty(&(td)->io_u_freelist)
589 extern struct io_u *__get_io_u(struct thread_data *);
590 extern struct io_u *get_io_u(struct thread_data *, struct fio_file *);
591 extern void put_io_u(struct thread_data *, struct io_u *);
592 extern void ios_completed(struct thread_data *, struct io_completion_data *);
593 extern void io_completed(struct thread_data *, struct io_u *, struct io_completion_data *);
596 * io engine entry points
598 extern int td_io_init(struct thread_data *);
599 extern int td_io_prep(struct thread_data *, struct io_u *);
600 extern int td_io_queue(struct thread_data *, struct io_u *);
601 extern int td_io_sync(struct thread_data *, struct fio_file *);
602 extern int td_io_getevents(struct thread_data *, int, int, struct timespec *);
605 * This is a pretty crappy semaphore implementation, but with the use that fio
606 * has (just signalling start/go conditions), it doesn't have to be better.
607 * Naturally this would not work for any type of contended semaphore or
610 static inline void fio_sem_init(volatile int *sem, int val)
615 static inline void fio_sem_down(volatile int *sem)
623 static inline void fio_sem_up(volatile int *sem)
629 * If logging output to a file, stderr should go to both stderr and f_err
631 #define log_err(args...) do { \
632 fprintf(f_err, ##args); \
633 if (f_err != stderr) \
634 fprintf(stderr, ##args); \
637 struct ioengine_ops {
638 struct list_head list;
642 int (*setup)(struct thread_data *);
643 int (*init)(struct thread_data *);
644 int (*prep)(struct thread_data *, struct io_u *);
645 int (*queue)(struct thread_data *, struct io_u *);
646 int (*getevents)(struct thread_data *, int, int, struct timespec *);
647 struct io_u *(*event)(struct thread_data *, int);
648 int (*cancel)(struct thread_data *, struct io_u *);
649 void (*cleanup)(struct thread_data *);
654 #define FIO_IOOPS_VERSION 3
656 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
657 extern int register_ioengine(struct ioengine_ops *);
658 extern void unregister_ioengine(struct ioengine_ops *);
659 extern void close_ioengine(struct thread_data *);
662 * Mark unused variables passed to ops functions as unused, to silence gcc
664 #define fio_unused __attribute((__unused__))
665 #define fio_init __attribute__((constructor))
666 #define fio_exit __attribute__((destructor))
668 #define for_each_td(td, i) \
669 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
670 #define for_each_file(td, f, i) \
671 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
673 #define fio_assert(td, cond) do { \
675 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
676 (td)->runstate = TD_EXITED; \
677 (td)->error = EFAULT; \