8 #include <sys/resource.h>
21 #ifdef FIO_HAVE_SYSLET
32 * Use for maintaining statistics
35 unsigned long max_val;
36 unsigned long min_val;
37 unsigned long samples;
44 * A single data sample
53 * Dynamically growing data sample log
56 unsigned long nr_samples;
57 unsigned long max_samples;
58 struct io_sample *log;
62 * When logging io actions, this matches a single sent io_u
65 struct list_head list;
66 struct fio_file *file;
67 unsigned long long offset;
72 #ifdef FIO_HAVE_SYSLET
74 struct syslet_uatom atom;
84 #ifdef FIO_HAVE_LIBAIO
87 #ifdef FIO_HAVE_POSIXAIO
93 #ifdef FIO_HAVE_SYSLET
94 struct syslet_req req;
97 struct timeval start_time;
98 struct timeval issue_time;
101 * Allocated/set buffer and length
104 unsigned long buflen;
105 unsigned long long offset;
108 * IO engine state, may be different from above when we get
109 * partial transfers / residual data counts
112 unsigned long xfer_buflen;
120 * io engine private data
127 struct fio_file *file;
129 struct list_head list;
132 #define FIO_HDR_MAGIC 0xf00baaef
135 VERIFY_NONE = 0, /* no verification */
136 VERIFY_MD5, /* md5 sum data blocks */
137 VERIFY_CRC32, /* crc32 sum data blocks */
141 * A header structure associated with each checksummed data block
143 struct verify_header {
144 unsigned int fio_magic;
146 unsigned int verify_type;
148 char md5_digest[MD5_HASH_WORDS * 4];
153 struct group_run_stats {
154 unsigned long long max_run[2], min_run[2];
155 unsigned long long max_bw[2], min_bw[2];
156 unsigned long long io_kb[2];
157 unsigned long long agg[2];
161 * What type of allocation to use for io buffers
164 MEM_MALLOC = 0, /* ordinary malloc */
165 MEM_SHM, /* use shared memory segments */
166 MEM_SHMHUGE, /* use shared memory segments with huge pages */
167 MEM_MMAP, /* use anonynomous mmap */
168 MEM_MMAPHUGE, /* memory mapped huge file */
172 * The type of object we are working on
175 FIO_TYPE_FILE = 1, /* plain file */
176 FIO_TYPE_BD, /* block device */
177 FIO_TYPE_CHAR, /* character device */
180 enum fio_ioengine_flags {
181 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
182 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
183 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
184 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
185 FIO_NETIO = 1 << 4, /* networked io */
186 FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
190 * Each thread_data structure has a number of files associated with it,
191 * this structure holds state information for a single file.
195 * A file may not be a file descriptor, let the io engine decide
198 unsigned long file_data;
203 unsigned long long file_size;
204 unsigned long long real_file_size;
205 unsigned long long file_offset;
206 unsigned long long last_pos;
207 unsigned long long last_completed_pos;
210 * block map for random io
212 unsigned long *file_map;
213 unsigned int num_maps;
214 unsigned int last_free_lookup;
220 * How many depth levels to log
222 #define FIO_IO_U_MAP_NR 8
223 #define FIO_IO_U_LAT_NR 12
226 * This describes a single thread/process executing a fio job.
237 enum fio_filetype filetype;
238 struct fio_file *files;
239 unsigned int nr_files;
240 unsigned int nr_uniq_files;
241 unsigned int next_file;
245 size_t orig_buffer_size;
246 volatile int terminate;
247 volatile int runstate;
251 unsigned int last_was_sync;
253 unsigned int sequential;
254 unsigned int odirect;
255 unsigned int invalidate_cache;
256 unsigned int create_serialize;
257 unsigned int create_fsync;
258 unsigned int end_fsync;
259 unsigned int sync_io;
261 unsigned int use_thread;
263 unsigned int do_disk_util;
264 unsigned int override_sync;
265 unsigned int rand_repeatable;
266 unsigned int write_lat_log;
267 unsigned int write_bw_log;
268 unsigned int norandommap;
269 unsigned int bs_unaligned;
272 unsigned int min_bs[2];
273 unsigned int max_bs[2];
274 unsigned int hugepage_size;
275 unsigned int rw_min_bs;
276 unsigned int thinktime;
277 unsigned int thinktime_blocks;
278 unsigned int fsync_blocks;
279 unsigned int start_delay;
280 unsigned long timeout;
281 unsigned int overwrite;
282 unsigned int bw_avg_time;
284 unsigned long long zone_size;
285 unsigned long long zone_skip;
286 enum fio_memtype mem_type;
289 unsigned int stonewall;
290 unsigned int numjobs;
291 unsigned int iodepth;
292 os_cpu_mask_t cpumask;
294 unsigned int read_iolog;
295 unsigned int rwmixcycle;
296 unsigned int rwmixread;
297 unsigned int rwmixwrite;
300 char *read_iolog_file;
301 char *write_iolog_file;
308 os_random_state_t bsrange_state;
309 os_random_state_t verify_state;
314 * IO engine hooks, contains everything needed to submit an io_u
315 * to any of the available IO engines.
317 struct ioengine_ops *io_ops;
320 * Current IO depth and list of free and busy io_u's.
322 unsigned int cur_depth;
323 unsigned int io_u_map[FIO_IO_U_MAP_NR];
324 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
325 unsigned long total_io_u;
326 struct list_head io_u_freelist;
327 struct list_head io_u_busylist;
333 unsigned int ratemin;
334 unsigned int ratecycle;
335 unsigned long rate_usec_cycle;
336 long rate_pending_usleep;
337 unsigned long rate_bytes;
338 struct timeval lastrate;
340 unsigned long runtime[2]; /* msec */
341 unsigned long long io_size;
342 unsigned long long total_file_size;
343 unsigned long long start_offset;
344 unsigned long long total_io_size;
346 unsigned long long io_blocks[2];
347 unsigned long long io_bytes[2];
348 unsigned long long zone_bytes;
349 unsigned long long this_io_bytes[2];
353 * State for random io, a bitmap of blocks done vs not done
355 os_random_state_t random_state;
358 * CPU "io" cycle burner
360 unsigned int cpuload;
361 unsigned int cpucycle;
364 * bandwidth and latency stats
366 struct io_stat clat_stat[2]; /* completion latency */
367 struct io_stat slat_stat[2]; /* submission latency */
368 struct io_stat bw_stat[2]; /* bandwidth stats */
370 unsigned long long stat_io_bytes[2];
371 struct timeval stat_sample_time[2];
373 struct io_log *slat_log;
374 struct io_log *clat_log;
375 struct io_log *bw_log;
377 struct timeval start; /* start of this loop */
378 struct timeval epoch; /* time job was started */
379 struct timeval end_time;/* time job ended */
382 * fio system usage accounting
384 struct rusage ru_start;
385 struct rusage ru_end;
386 unsigned long usr_time;
387 unsigned long sys_time;
391 * read/write mixed workload state
393 os_random_state_t rwmix_state;
394 struct timeval rwmix_switch;
395 enum fio_ddir rwmix_ddir;
398 * Pre-run and post-run shell
406 struct list_head io_hist_list;
407 struct list_head io_log_list;
410 #define __td_verror(td, err, msg) \
414 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, (msg)); \
418 #define td_verror(td, err) __td_verror((td), (err), strerror((err)))
419 #define td_vmsg(td, err, msg) __td_verror((td), (err), (msg))
421 extern int exitall_on_terminate;
422 extern int thread_number;
425 extern int terse_output;
428 extern int temp_stall_ts;
429 extern unsigned long long mlock_size;
431 extern struct thread_data *threads;
433 #define td_read(td) ((td)->ddir == DDIR_READ)
434 #define td_write(td) ((td)->ddir == DDIR_WRITE)
435 #define td_rw(td) ((td)->iomix != 0)
437 #define BLOCKS_PER_MAP (8 * sizeof(long))
438 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
439 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
440 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
442 #define MAX_JOBS (1024)
444 static inline int should_fsync(struct thread_data *td)
446 if (td->last_was_sync)
450 if (td_write(td) || td_rw(td) || td->override_sync)
457 * Disk utils as read in /sys/block/<dev>/stat
459 struct disk_util_stat {
462 unsigned long long sectors[2];
465 unsigned time_in_queue;
469 * Per-device disk util management
472 struct list_head list;
478 struct disk_util_stat dus;
479 struct disk_util_stat last_dus;
486 * Used for passing io_u completion data
488 struct io_completion_data {
491 int error; /* output */
492 unsigned long bytes_done[2]; /* output */
493 struct timeval time; /* output */
496 #define DISK_UTIL_MSEC (250)
499 #define min(a, b) ((a) < (b) ? (a) : (b))
502 #define max(a, b) ((a) > (b) ? (a) : (b))
508 extern int read_iolog_get(struct thread_data *, struct io_u *);
509 extern void write_iolog_put(struct thread_data *, struct io_u *);
510 extern int init_iolog(struct thread_data *td);
511 extern void log_io_piece(struct thread_data *, struct io_u *);
512 extern void prune_io_piece_log(struct thread_data *);
513 extern void write_iolog_close(struct thread_data *);
518 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
519 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
520 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
521 extern void show_run_stats(void);
522 extern void init_disk_util(struct thread_data *);
523 extern void update_rusage_stat(struct thread_data *);
524 extern void update_io_ticks(void);
525 extern void disk_util_timer_arm(void);
526 extern void setup_log(struct io_log **);
527 extern void finish_log(struct thread_data *, struct io_log *, const char *);
528 extern void __finish_log(struct io_log *, const char *);
529 extern int setup_rate(struct thread_data *);
530 extern struct io_log *agg_io_log[2];
531 extern int write_bw_log;
532 extern void add_agg_sample(unsigned long, enum fio_ddir);
537 extern unsigned long utime_since(struct timeval *, struct timeval *);
538 extern unsigned long utime_since_now(struct timeval *);
539 extern unsigned long mtime_since(struct timeval *, struct timeval *);
540 extern unsigned long mtime_since_now(struct timeval *);
541 extern unsigned long time_since_now(struct timeval *);
542 extern unsigned long mtime_since_genesis(void);
543 extern void __usec_sleep(unsigned int);
544 extern void usec_sleep(struct thread_data *, unsigned long);
545 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
546 extern void fill_start_time(struct timeval *);
547 extern void fio_gettime(struct timeval *, void *);
552 extern int parse_options(int, char **);
553 extern int init_random_state(struct thread_data *);
556 * File setup/shutdown
558 extern void close_files(struct thread_data *);
559 extern int setup_files(struct thread_data *);
560 extern int open_files(struct thread_data *);
561 extern int file_invalidate_cache(struct thread_data *, struct fio_file *);
566 extern void print_thread_status(void);
567 extern void print_status_init(int);
570 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
571 * will never back again. It may cycle between running/verififying/fsyncing.
572 * Once the thread reaches TD_EXITED, it is just waiting for the core to
589 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
590 extern int get_next_verify(struct thread_data *td, struct io_u *);
591 extern int do_io_u_verify(struct thread_data *, struct io_u **);
596 extern int fio_pin_memory(void);
597 extern void fio_unpin_memory(void);
598 extern int allocate_io_mem(struct thread_data *);
599 extern void free_io_mem(struct thread_data *);
604 #define queue_full(td) list_empty(&(td)->io_u_freelist)
605 extern struct io_u *__get_io_u(struct thread_data *);
606 extern struct io_u *get_io_u(struct thread_data *, struct fio_file *);
607 extern void put_io_u(struct thread_data *, struct io_u *);
608 extern void ios_completed(struct thread_data *, struct io_completion_data *);
609 extern void io_completed(struct thread_data *, struct io_u *, struct io_completion_data *);
612 * io engine entry points
614 extern int td_io_init(struct thread_data *);
615 extern int td_io_prep(struct thread_data *, struct io_u *);
616 extern int td_io_queue(struct thread_data *, struct io_u *);
617 extern int td_io_sync(struct thread_data *, struct fio_file *);
618 extern int td_io_getevents(struct thread_data *, int, int, struct timespec *);
621 * This is a pretty crappy semaphore implementation, but with the use that fio
622 * has (just signalling start/go conditions), it doesn't have to be better.
623 * Naturally this would not work for any type of contended semaphore or
626 static inline void fio_sem_init(volatile int *sem, int val)
631 static inline void fio_sem_down(volatile int *sem)
639 static inline void fio_sem_up(volatile int *sem)
645 * If logging output to a file, stderr should go to both stderr and f_err
647 #define log_err(args...) do { \
648 fprintf(f_err, ##args); \
649 if (f_err != stderr) \
650 fprintf(stderr, ##args); \
653 struct ioengine_ops {
654 struct list_head list;
658 int (*setup)(struct thread_data *);
659 int (*init)(struct thread_data *);
660 int (*prep)(struct thread_data *, struct io_u *);
661 int (*queue)(struct thread_data *, struct io_u *);
662 int (*getevents)(struct thread_data *, int, int, struct timespec *);
663 struct io_u *(*event)(struct thread_data *, int);
664 int (*cancel)(struct thread_data *, struct io_u *);
665 void (*cleanup)(struct thread_data *);
670 #define FIO_IOOPS_VERSION 3
672 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
673 extern int register_ioengine(struct ioengine_ops *);
674 extern void unregister_ioengine(struct ioengine_ops *);
675 extern void close_ioengine(struct thread_data *);
678 * Mark unused variables passed to ops functions as unused, to silence gcc
680 #define fio_unused __attribute((__unused__))
681 #define fio_init __attribute__((constructor))
682 #define fio_exit __attribute__((destructor))
684 #define for_each_td(td, i) \
685 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
686 #define for_each_file(td, f, i) \
687 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
689 #define fio_assert(td, cond) do { \
692 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
693 (td)->runstate = TD_EXITED; \
694 (td)->error = EFAULT; \