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,
36 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
37 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_READ,
38 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
42 * Use for maintaining statistics
45 unsigned long max_val;
46 unsigned long min_val;
47 unsigned long samples;
54 * A single data sample
63 * Dynamically growing data sample log
66 unsigned long nr_samples;
67 unsigned long max_samples;
68 struct io_sample *log;
72 * When logging io actions, this matches a single sent io_u
75 struct list_head list;
76 struct fio_file *file;
77 unsigned long long offset;
82 #ifdef FIO_HAVE_SYSLET
84 struct syslet_uatom atom;
91 IO_U_F_FLIGHT = 1 << 1,
99 #ifdef FIO_HAVE_LIBAIO
102 #ifdef FIO_HAVE_POSIXAIO
106 struct sg_io_hdr hdr;
108 #ifdef FIO_HAVE_SYSLET
109 struct syslet_req req;
112 struct timeval start_time;
113 struct timeval issue_time;
116 * Allocated/set buffer and length
119 unsigned long buflen;
120 unsigned long long offset;
123 * IO engine state, may be different from above when we get
124 * partial transfers / residual data counts
127 unsigned long xfer_buflen;
135 * io engine private data
144 struct fio_file *file;
146 struct list_head list;
150 * io_ops->queue() return values
153 FIO_Q_COMPLETED = 0, /* completed sync */
154 FIO_Q_QUEUED = 1, /* queued, will complete async */
155 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
158 #define FIO_HDR_MAGIC 0xf00baaef
161 VERIFY_NONE = 0, /* no verification */
162 VERIFY_MD5, /* md5 sum data blocks */
163 VERIFY_CRC32, /* crc32 sum data blocks */
167 * A header structure associated with each checksummed data block
169 struct verify_header {
170 unsigned int fio_magic;
172 unsigned int verify_type;
174 char md5_digest[MD5_HASH_WORDS * 4];
179 struct group_run_stats {
180 unsigned long long max_run[2], min_run[2];
181 unsigned long long max_bw[2], min_bw[2];
182 unsigned long long io_kb[2];
183 unsigned long long agg[2];
187 * What type of allocation to use for io buffers
190 MEM_MALLOC = 0, /* ordinary malloc */
191 MEM_SHM, /* use shared memory segments */
192 MEM_SHMHUGE, /* use shared memory segments with huge pages */
193 MEM_MMAP, /* use anonynomous mmap */
194 MEM_MMAPHUGE, /* memory mapped huge file */
198 * The type of object we are working on
201 FIO_TYPE_FILE = 1, /* plain file */
202 FIO_TYPE_BD, /* block device */
203 FIO_TYPE_CHAR, /* character device */
206 enum fio_ioengine_flags {
207 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
208 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
209 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
210 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
211 FIO_NETIO = 1 << 4, /* networked io */
212 FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
216 * Each thread_data structure has a number of files associated with it,
217 * this structure holds state information for a single file.
221 * A file may not be a file descriptor, let the io engine decide
224 unsigned long file_data;
229 unsigned long long file_size;
230 unsigned long long real_file_size;
231 unsigned long long file_offset;
232 unsigned long long last_pos;
233 unsigned long long last_completed_pos;
236 * block map for random io
238 unsigned long *file_map;
239 unsigned int num_maps;
240 unsigned int last_free_lookup;
246 struct io_log *slat_log;
247 struct io_log *clat_log;
248 struct io_log *bw_log;
251 * bandwidth and latency stats
253 struct io_stat clat_stat[2]; /* completion latency */
254 struct io_stat slat_stat[2]; /* submission latency */
255 struct io_stat bw_stat[2]; /* bandwidth stats */
257 unsigned long long stat_io_bytes[2];
258 struct timeval stat_sample_time[2];
261 * fio system usage accounting
263 struct rusage ru_start;
264 struct rusage ru_end;
265 unsigned long usr_time;
266 unsigned long sys_time;
271 * How many depth levels to log
273 #define FIO_IO_U_MAP_NR 8
274 #define FIO_IO_U_LAT_NR 12
277 * This describes a single thread/process executing a fio job.
289 struct thread_stat ts;
290 enum fio_filetype filetype;
291 struct fio_file *files;
292 unsigned int nr_files;
293 unsigned int nr_uniq_files;
295 unsigned int next_file;
296 os_random_state_t next_file_state;
301 size_t orig_buffer_size;
302 volatile int terminate;
303 volatile int runstate;
304 enum td_ddir td_ddir;
306 unsigned int last_was_sync;
308 unsigned int odirect;
309 unsigned int invalidate_cache;
310 unsigned int create_serialize;
311 unsigned int create_fsync;
312 unsigned int end_fsync;
313 unsigned int sync_io;
315 unsigned int use_thread;
317 unsigned int do_disk_util;
318 unsigned int override_sync;
319 unsigned int rand_repeatable;
320 unsigned int write_lat_log;
321 unsigned int write_bw_log;
322 unsigned int norandommap;
323 unsigned int bs_unaligned;
326 unsigned int min_bs[2];
327 unsigned int max_bs[2];
328 unsigned int hugepage_size;
329 unsigned int rw_min_bs;
330 unsigned int thinktime;
331 unsigned int thinktime_spin;
332 unsigned int thinktime_blocks;
333 unsigned int fsync_blocks;
334 unsigned int start_delay;
335 unsigned long timeout;
336 unsigned int overwrite;
337 unsigned int bw_avg_time;
339 unsigned long long zone_size;
340 unsigned long long zone_skip;
341 enum fio_memtype mem_type;
344 unsigned int stonewall;
345 unsigned int numjobs;
346 unsigned int iodepth;
347 unsigned int iodepth_low;
348 os_cpu_mask_t cpumask;
350 unsigned int read_iolog;
351 unsigned int rwmixcycle;
352 unsigned int rwmixread;
353 unsigned int rwmixwrite;
355 unsigned int file_service_type;
357 char *read_iolog_file;
358 char *write_iolog_file;
365 os_random_state_t bsrange_state;
366 os_random_state_t verify_state;
371 * IO engine hooks, contains everything needed to submit an io_u
372 * to any of the available IO engines.
374 struct ioengine_ops *io_ops;
377 * Current IO depth and list of free and busy io_u's.
379 unsigned int cur_depth;
380 unsigned int io_u_map[FIO_IO_U_MAP_NR];
381 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
382 unsigned long total_io_u;
383 struct list_head io_u_freelist;
384 struct list_head io_u_busylist;
385 struct list_head io_u_requeues;
391 unsigned int ratemin;
392 unsigned int ratecycle;
393 unsigned long rate_usec_cycle;
394 long rate_pending_usleep;
395 unsigned long rate_bytes;
396 struct timeval lastrate;
398 unsigned long runtime[2]; /* msec */
399 unsigned long long io_size;
400 unsigned long long total_file_size;
401 unsigned long long start_offset;
402 unsigned long long total_io_size;
404 unsigned long io_issues[2];
405 unsigned long long io_blocks[2];
406 unsigned long long io_bytes[2];
407 unsigned long long this_io_bytes[2];
408 unsigned long long zone_bytes;
412 * State for random io, a bitmap of blocks done vs not done
414 os_random_state_t random_state;
417 * CPU "io" cycle burner
419 unsigned int cpuload;
420 unsigned int cpucycle;
422 struct timeval start; /* start of this loop */
423 struct timeval epoch; /* time job was started */
424 struct timeval end_time;/* time job ended */
427 * read/write mixed workload state
429 os_random_state_t rwmix_state;
430 struct timeval rwmix_switch;
431 enum fio_ddir rwmix_ddir;
434 * Pre-run and post-run shell
442 struct list_head io_hist_list;
443 struct list_head io_log_list;
448 struct timeval timeout_end;
449 struct itimerval timer;
453 * roundrobin available files, or choose one at random.
456 FIO_FSERVICE_RANDOM = 1,
461 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
462 * the timer on each queue operation.
464 #define IO_U_TIMEOUT_INC 5
465 #define IO_U_TIMEOUT 30
467 #define __td_verror(td, err, msg, func) \
473 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
477 #define td_verror(td, err, func) \
478 __td_verror((td), (err), strerror((err)), (func))
479 #define td_vmsg(td, err, msg, func) \
480 __td_verror((td), (err), (msg), (func))
482 extern int exitall_on_terminate;
483 extern int thread_number;
486 extern int terse_output;
489 extern int temp_stall_ts;
490 extern unsigned long long mlock_size;
492 extern struct thread_data *threads;
494 #define td_read(td) ((td)->td_ddir & TD_DDIR_READ)
495 #define td_write(td) ((td)->td_ddir & TD_DDIR_WRITE)
496 #define td_rw(td) (((td)->td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
497 #define td_random(td) ((td)->td_ddir & TD_DDIR_RAND)
499 #define BLOCKS_PER_MAP (8 * sizeof(long))
500 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
501 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
502 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
504 #define MAX_JOBS (1024)
506 static inline int should_fsync(struct thread_data *td)
508 if (td->last_was_sync)
512 if (td_write(td) || td_rw(td) || td->override_sync)
519 * Disk utils as read in /sys/block/<dev>/stat
521 struct disk_util_stat {
524 unsigned long long sectors[2];
527 unsigned time_in_queue;
531 * Per-device disk util management
534 struct list_head list;
540 struct disk_util_stat dus;
541 struct disk_util_stat last_dus;
548 * Callback for io completion
550 typedef int (endio_handler)(struct io_u *);
552 #define DISK_UTIL_MSEC (250)
555 #define min(a, b) ((a) < (b) ? (a) : (b))
558 #define max(a, b) ((a) > (b) ? (a) : (b))
564 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
565 extern void write_iolog_put(struct thread_data *, struct io_u *);
566 extern int __must_check init_iolog(struct thread_data *td);
567 extern void log_io_piece(struct thread_data *, struct io_u *);
568 extern void prune_io_piece_log(struct thread_data *);
569 extern void write_iolog_close(struct thread_data *);
574 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
575 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
576 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
577 extern void show_run_stats(void);
578 extern void init_disk_util(struct thread_data *);
579 extern void update_rusage_stat(struct thread_data *);
580 extern void update_io_ticks(void);
581 extern void disk_util_timer_arm(void);
582 extern void setup_log(struct io_log **);
583 extern void finish_log(struct thread_data *, struct io_log *, const char *);
584 extern void __finish_log(struct io_log *, const char *);
585 extern int setup_rate(struct thread_data *);
586 extern struct io_log *agg_io_log[2];
587 extern int write_bw_log;
588 extern void add_agg_sample(unsigned long, enum fio_ddir);
593 extern unsigned long utime_since(struct timeval *, struct timeval *);
594 extern unsigned long utime_since_now(struct timeval *);
595 extern unsigned long mtime_since(struct timeval *, struct timeval *);
596 extern unsigned long mtime_since_now(struct timeval *);
597 extern unsigned long time_since_now(struct timeval *);
598 extern unsigned long mtime_since_genesis(void);
599 extern void __usec_sleep(unsigned int);
600 extern void usec_sleep(struct thread_data *, unsigned long);
601 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
602 extern void fill_start_time(struct timeval *);
603 extern void fio_gettime(struct timeval *, void *);
604 extern void set_genesis_time(void);
609 extern int __must_check parse_options(int, char **);
610 extern int __must_check init_random_state(struct thread_data *);
613 * File setup/shutdown
615 extern void close_files(struct thread_data *);
616 extern int __must_check setup_files(struct thread_data *);
617 extern int __must_check open_files(struct thread_data *);
618 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
623 extern void print_thread_status(void);
624 extern void print_status_init(int);
627 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
628 * will never back again. It may cycle between running/verififying/fsyncing.
629 * Once the thread reaches TD_EXITED, it is just waiting for the core to
646 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
647 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
648 extern int __must_check verify_io_u(struct io_u *);
653 extern int __must_check fio_pin_memory(void);
654 extern void fio_unpin_memory(void);
655 extern int __must_check allocate_io_mem(struct thread_data *);
656 extern void free_io_mem(struct thread_data *);
661 #define queue_full(td) list_empty(&(td)->io_u_freelist)
662 extern struct io_u *__get_io_u(struct thread_data *);
663 extern struct io_u *get_io_u(struct thread_data *);
664 extern void put_io_u(struct thread_data *, struct io_u *);
665 extern void requeue_io_u(struct thread_data *, struct io_u **);
666 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *, endio_handler *);
667 extern long __must_check io_u_queued_complete(struct thread_data *, int, endio_handler *);
668 extern void io_u_queued(struct thread_data *, struct io_u *);
669 extern void io_u_init_timeout(void);
670 extern void io_u_set_timeout(struct thread_data *);
673 * io engine entry points
675 extern int __must_check td_io_init(struct thread_data *);
676 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
677 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
678 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
679 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
680 extern int __must_check td_io_commit(struct thread_data *);
683 * This is a pretty crappy semaphore implementation, but with the use that fio
684 * has (just signalling start/go conditions), it doesn't have to be better.
685 * Naturally this would not work for any type of contended semaphore or
688 static inline void fio_sem_init(volatile int *sem, int val)
693 static inline void fio_sem_down(volatile int *sem)
701 static inline void fio_sem_up(volatile int *sem)
707 * If logging output to a file, stderr should go to both stderr and f_err
709 #define log_err(args...) do { \
710 fprintf(f_err, ##args); \
711 if (f_err != stderr) \
712 fprintf(stderr, ##args); \
715 struct ioengine_ops {
716 struct list_head list;
720 int (*setup)(struct thread_data *);
721 int (*init)(struct thread_data *);
722 int (*prep)(struct thread_data *, struct io_u *);
723 int (*queue)(struct thread_data *, struct io_u *);
724 int (*commit)(struct thread_data *);
725 int (*getevents)(struct thread_data *, int, int, struct timespec *);
726 struct io_u *(*event)(struct thread_data *, int);
727 int (*cancel)(struct thread_data *, struct io_u *);
728 void (*cleanup)(struct thread_data *);
734 #define FIO_IOOPS_VERSION 5
736 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
737 extern void register_ioengine(struct ioengine_ops *);
738 extern void unregister_ioengine(struct ioengine_ops *);
739 extern void close_ioengine(struct thread_data *);
742 * Mark unused variables passed to ops functions as unused, to silence gcc
744 #define fio_unused __attribute((__unused__))
745 #define fio_init __attribute__((constructor))
746 #define fio_exit __attribute__((destructor))
748 #define for_each_td(td, i) \
749 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
750 #define for_each_file(td, f, i) \
751 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
753 #define fio_assert(td, cond) do { \
756 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
757 (td)->runstate = TD_EXITED; \
758 (td)->error = EFAULT; \