8 #include <sys/resource.h>
21 #ifdef FIO_HAVE_SYSLET
32 TD_DDIR_READ = 1 << 0,
33 TD_DDIR_WRITE = 1 << 1,
34 TD_DDIR_RAND = 1 << 2,
35 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
39 * Use for maintaining statistics
42 unsigned long max_val;
43 unsigned long min_val;
44 unsigned long samples;
51 * A single data sample
60 * Dynamically growing data sample log
63 unsigned long nr_samples;
64 unsigned long max_samples;
65 struct io_sample *log;
69 * When logging io actions, this matches a single sent io_u
72 struct list_head list;
73 struct fio_file *file;
74 unsigned long long offset;
79 #ifdef FIO_HAVE_SYSLET
81 struct syslet_uatom atom;
88 IO_U_F_FLIGHT = 1 << 1,
96 #ifdef FIO_HAVE_LIBAIO
99 #ifdef FIO_HAVE_POSIXAIO
103 struct sg_io_hdr hdr;
105 #ifdef FIO_HAVE_SYSLET
106 struct syslet_req req;
109 struct timeval start_time;
110 struct timeval issue_time;
113 * Allocated/set buffer and length
116 unsigned long buflen;
117 unsigned long long offset;
120 * IO engine state, may be different from above when we get
121 * partial transfers / residual data counts
124 unsigned long xfer_buflen;
132 * io engine private data
141 struct fio_file *file;
143 struct list_head list;
147 * io_ops->queue() return values
150 FIO_Q_COMPLETED = 0, /* completed sync */
151 FIO_Q_QUEUED = 1, /* queued, will complete async */
152 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
155 #define FIO_HDR_MAGIC 0xf00baaef
158 VERIFY_NONE = 0, /* no verification */
159 VERIFY_MD5, /* md5 sum data blocks */
160 VERIFY_CRC32, /* crc32 sum data blocks */
164 * A header structure associated with each checksummed data block
166 struct verify_header {
167 unsigned int fio_magic;
169 unsigned int verify_type;
171 char md5_digest[MD5_HASH_WORDS * 4];
176 struct group_run_stats {
177 unsigned long long max_run[2], min_run[2];
178 unsigned long long max_bw[2], min_bw[2];
179 unsigned long long io_kb[2];
180 unsigned long long agg[2];
184 * What type of allocation to use for io buffers
187 MEM_MALLOC = 0, /* ordinary malloc */
188 MEM_SHM, /* use shared memory segments */
189 MEM_SHMHUGE, /* use shared memory segments with huge pages */
190 MEM_MMAP, /* use anonynomous mmap */
191 MEM_MMAPHUGE, /* memory mapped huge file */
195 * The type of object we are working on
198 FIO_TYPE_FILE = 1, /* plain file */
199 FIO_TYPE_BD, /* block device */
200 FIO_TYPE_CHAR, /* character device */
203 enum fio_ioengine_flags {
204 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
205 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
206 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
207 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
208 FIO_NETIO = 1 << 4, /* networked io */
209 FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
213 * Each thread_data structure has a number of files associated with it,
214 * this structure holds state information for a single file.
218 * A file may not be a file descriptor, let the io engine decide
221 unsigned long file_data;
226 unsigned long long file_size;
227 unsigned long long real_file_size;
228 unsigned long long file_offset;
229 unsigned long long last_pos;
230 unsigned long long last_completed_pos;
233 * block map for random io
235 unsigned long *file_map;
236 unsigned int num_maps;
237 unsigned int last_free_lookup;
243 struct io_log *slat_log;
244 struct io_log *clat_log;
245 struct io_log *bw_log;
248 * bandwidth and latency stats
250 struct io_stat clat_stat[2]; /* completion latency */
251 struct io_stat slat_stat[2]; /* submission latency */
252 struct io_stat bw_stat[2]; /* bandwidth stats */
254 unsigned long long stat_io_bytes[2];
255 struct timeval stat_sample_time[2];
258 * fio system usage accounting
260 struct rusage ru_start;
261 struct rusage ru_end;
262 unsigned long usr_time;
263 unsigned long sys_time;
268 * How many depth levels to log
270 #define FIO_IO_U_MAP_NR 8
271 #define FIO_IO_U_LAT_NR 12
274 * This describes a single thread/process executing a fio job.
285 struct thread_stat ts;
286 enum fio_filetype filetype;
287 struct fio_file *files;
288 unsigned int nr_files;
289 unsigned int nr_uniq_files;
291 unsigned int next_file;
292 os_random_state_t next_file_state;
297 size_t orig_buffer_size;
298 volatile int terminate;
299 volatile int runstate;
300 enum td_ddir td_ddir;
302 unsigned int last_was_sync;
304 unsigned int odirect;
305 unsigned int invalidate_cache;
306 unsigned int create_serialize;
307 unsigned int create_fsync;
308 unsigned int end_fsync;
309 unsigned int sync_io;
311 unsigned int use_thread;
313 unsigned int do_disk_util;
314 unsigned int override_sync;
315 unsigned int rand_repeatable;
316 unsigned int write_lat_log;
317 unsigned int write_bw_log;
318 unsigned int norandommap;
319 unsigned int bs_unaligned;
322 unsigned int min_bs[2];
323 unsigned int max_bs[2];
324 unsigned int hugepage_size;
325 unsigned int rw_min_bs;
326 unsigned int thinktime;
327 unsigned int thinktime_spin;
328 unsigned int thinktime_blocks;
329 unsigned int fsync_blocks;
330 unsigned int start_delay;
331 unsigned long timeout;
332 unsigned int overwrite;
333 unsigned int bw_avg_time;
335 unsigned long long zone_size;
336 unsigned long long zone_skip;
337 enum fio_memtype mem_type;
340 unsigned int stonewall;
341 unsigned int numjobs;
342 unsigned int iodepth;
343 unsigned int iodepth_low;
344 os_cpu_mask_t cpumask;
346 unsigned int read_iolog;
347 unsigned int rwmixcycle;
348 unsigned int rwmixread;
349 unsigned int rwmixwrite;
351 unsigned int file_service_type;
353 char *read_iolog_file;
354 char *write_iolog_file;
361 os_random_state_t bsrange_state;
362 os_random_state_t verify_state;
367 * IO engine hooks, contains everything needed to submit an io_u
368 * to any of the available IO engines.
370 struct ioengine_ops *io_ops;
373 * Current IO depth and list of free and busy io_u's.
375 unsigned int cur_depth;
376 unsigned int io_u_map[FIO_IO_U_MAP_NR];
377 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
378 unsigned long total_io_u;
379 struct list_head io_u_freelist;
380 struct list_head io_u_busylist;
381 struct list_head io_u_requeues;
387 unsigned int ratemin;
388 unsigned int ratecycle;
389 unsigned long rate_usec_cycle;
390 long rate_pending_usleep;
391 unsigned long rate_bytes;
392 struct timeval lastrate;
394 unsigned long runtime[2]; /* msec */
395 unsigned long long io_size;
396 unsigned long long total_file_size;
397 unsigned long long start_offset;
398 unsigned long long total_io_size;
400 unsigned long io_issues[2];
401 unsigned long long io_blocks[2];
402 unsigned long long io_bytes[2];
403 unsigned long long this_io_bytes[2];
404 unsigned long long zone_bytes;
408 * State for random io, a bitmap of blocks done vs not done
410 os_random_state_t random_state;
413 * CPU "io" cycle burner
415 unsigned int cpuload;
416 unsigned int cpucycle;
418 struct timeval start; /* start of this loop */
419 struct timeval epoch; /* time job was started */
420 struct timeval end_time;/* time job ended */
423 * read/write mixed workload state
425 os_random_state_t rwmix_state;
426 struct timeval rwmix_switch;
427 enum fio_ddir rwmix_ddir;
430 * Pre-run and post-run shell
438 struct list_head io_hist_list;
439 struct list_head io_log_list;
444 struct timeval timeout_end;
445 struct itimerval timer;
449 * roundrobin available files, or choose one at random.
452 FIO_FSERVICE_RANDOM = 1,
457 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
458 * the timer on each queue operation.
460 #define IO_U_TIMEOUT_INC 5
461 #define IO_U_TIMEOUT 30
463 #define __td_verror(td, err, msg, func) \
469 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
473 #define td_verror(td, err, func) \
474 __td_verror((td), (err), strerror((err)), (func))
475 #define td_vmsg(td, err, msg, func) \
476 __td_verror((td), (err), (msg), (func))
478 extern int exitall_on_terminate;
479 extern int thread_number;
482 extern int terse_output;
485 extern int temp_stall_ts;
486 extern unsigned long long mlock_size;
488 extern struct thread_data *threads;
490 #define td_read(td) ((td)->td_ddir & TD_DDIR_READ)
491 #define td_write(td) ((td)->td_ddir & TD_DDIR_WRITE)
492 #define td_rw(td) (((td)->td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
493 #define td_random(td) ((td)->td_ddir & TD_DDIR_RAND)
495 #define BLOCKS_PER_MAP (8 * sizeof(long))
496 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
497 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
498 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
500 #define MAX_JOBS (1024)
502 static inline int should_fsync(struct thread_data *td)
504 if (td->last_was_sync)
508 if (td_write(td) || td_rw(td) || td->override_sync)
515 * Disk utils as read in /sys/block/<dev>/stat
517 struct disk_util_stat {
520 unsigned long long sectors[2];
523 unsigned time_in_queue;
527 * Per-device disk util management
530 struct list_head list;
536 struct disk_util_stat dus;
537 struct disk_util_stat last_dus;
544 * Callback for io completion
546 typedef int (endio_handler)(struct io_u *);
548 #define DISK_UTIL_MSEC (250)
551 #define min(a, b) ((a) < (b) ? (a) : (b))
554 #define max(a, b) ((a) > (b) ? (a) : (b))
560 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
561 extern void write_iolog_put(struct thread_data *, struct io_u *);
562 extern int __must_check init_iolog(struct thread_data *td);
563 extern void log_io_piece(struct thread_data *, struct io_u *);
564 extern void prune_io_piece_log(struct thread_data *);
565 extern void write_iolog_close(struct thread_data *);
570 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
571 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
572 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
573 extern void show_run_stats(void);
574 extern void init_disk_util(struct thread_data *);
575 extern void update_rusage_stat(struct thread_data *);
576 extern void update_io_ticks(void);
577 extern void disk_util_timer_arm(void);
578 extern void setup_log(struct io_log **);
579 extern void finish_log(struct thread_data *, struct io_log *, const char *);
580 extern void __finish_log(struct io_log *, const char *);
581 extern int setup_rate(struct thread_data *);
582 extern struct io_log *agg_io_log[2];
583 extern int write_bw_log;
584 extern void add_agg_sample(unsigned long, enum fio_ddir);
589 extern unsigned long utime_since(struct timeval *, struct timeval *);
590 extern unsigned long utime_since_now(struct timeval *);
591 extern unsigned long mtime_since(struct timeval *, struct timeval *);
592 extern unsigned long mtime_since_now(struct timeval *);
593 extern unsigned long time_since_now(struct timeval *);
594 extern unsigned long mtime_since_genesis(void);
595 extern void __usec_sleep(unsigned int);
596 extern void usec_sleep(struct thread_data *, unsigned long);
597 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
598 extern void fill_start_time(struct timeval *);
599 extern void fio_gettime(struct timeval *, void *);
600 extern void set_genesis_time(void);
605 extern int __must_check parse_options(int, char **);
606 extern int __must_check init_random_state(struct thread_data *);
609 * File setup/shutdown
611 extern void close_files(struct thread_data *);
612 extern int __must_check setup_files(struct thread_data *);
613 extern int __must_check open_files(struct thread_data *);
614 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
619 extern void print_thread_status(void);
620 extern void print_status_init(int);
623 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
624 * will never back again. It may cycle between running/verififying/fsyncing.
625 * Once the thread reaches TD_EXITED, it is just waiting for the core to
642 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
643 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
644 extern int __must_check verify_io_u(struct io_u *);
649 extern int __must_check fio_pin_memory(void);
650 extern void fio_unpin_memory(void);
651 extern int __must_check allocate_io_mem(struct thread_data *);
652 extern void free_io_mem(struct thread_data *);
657 #define queue_full(td) list_empty(&(td)->io_u_freelist)
658 extern struct io_u *__get_io_u(struct thread_data *);
659 extern struct io_u *get_io_u(struct thread_data *);
660 extern void put_io_u(struct thread_data *, struct io_u *);
661 extern void requeue_io_u(struct thread_data *, struct io_u **);
662 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *, endio_handler *);
663 extern long __must_check io_u_queued_complete(struct thread_data *, int, endio_handler *);
664 extern void io_u_queued(struct thread_data *, struct io_u *);
665 extern void io_u_init_timeout(void);
666 extern void io_u_set_timeout(struct thread_data *);
669 * io engine entry points
671 extern int __must_check td_io_init(struct thread_data *);
672 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
673 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
674 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
675 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
676 extern int __must_check td_io_commit(struct thread_data *);
679 * This is a pretty crappy semaphore implementation, but with the use that fio
680 * has (just signalling start/go conditions), it doesn't have to be better.
681 * Naturally this would not work for any type of contended semaphore or
684 static inline void fio_sem_init(volatile int *sem, int val)
689 static inline void fio_sem_down(volatile int *sem)
697 static inline void fio_sem_up(volatile int *sem)
703 * If logging output to a file, stderr should go to both stderr and f_err
705 #define log_err(args...) do { \
706 fprintf(f_err, ##args); \
707 if (f_err != stderr) \
708 fprintf(stderr, ##args); \
711 struct ioengine_ops {
712 struct list_head list;
716 int (*setup)(struct thread_data *);
717 int (*init)(struct thread_data *);
718 int (*prep)(struct thread_data *, struct io_u *);
719 int (*queue)(struct thread_data *, struct io_u *);
720 int (*commit)(struct thread_data *);
721 int (*getevents)(struct thread_data *, int, int, struct timespec *);
722 struct io_u *(*event)(struct thread_data *, int);
723 int (*cancel)(struct thread_data *, struct io_u *);
724 void (*cleanup)(struct thread_data *);
730 #define FIO_IOOPS_VERSION 5
732 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
733 extern void register_ioengine(struct ioengine_ops *);
734 extern void unregister_ioengine(struct ioengine_ops *);
735 extern void close_ioengine(struct thread_data *);
738 * Mark unused variables passed to ops functions as unused, to silence gcc
740 #define fio_unused __attribute((__unused__))
741 #define fio_init __attribute__((constructor))
742 #define fio_exit __attribute__((destructor))
744 #define for_each_td(td, i) \
745 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
746 #define for_each_file(td, f, i) \
747 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
749 #define fio_assert(td, cond) do { \
752 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
753 (td)->runstate = TD_EXITED; \
754 (td)->error = EFAULT; \