[fio.git] / fio.h

#ifndef FIO_H
#define FIO_H

#include <sched.h>
#include <limits.h>
#include <pthread.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <errno.h>
#include <stdlib.h>
#include <stdio.h>

#include "list.h"
#include "md5.h"
#include "crc32.h"
#include "arch.h"
#include "os.h"

struct io_stat {
	unsigned long val;
	unsigned long val_sq;
	unsigned long max_val;
	unsigned long min_val;
	unsigned long samples;
};

struct io_sample {
	unsigned long time;
	unsigned long val;
	unsigned int ddir;
};

struct io_log {
	unsigned long nr_samples;
	unsigned long max_samples;
	struct io_sample *log;
};

struct io_piece {
	struct list_head list;
	unsigned long long offset;
	unsigned int len;
	int ddir;
};

/*
 * The io unit
 */
struct io_u {
	union {
#ifdef FIO_HAVE_LIBAIO
		struct iocb iocb;
#endif
#ifdef FIO_HAVE_POSIXAIO
		struct aiocb aiocb;
#endif
#ifdef FIO_HAVE_SGIO
		struct sg_io_hdr hdr;
#endif
	};
	struct timeval start_time;
	struct timeval issue_time;

	char *buf;
	unsigned int buflen;
	unsigned long long offset;
	unsigned int index;

	unsigned int resid;
	unsigned int error;

	unsigned char seen;
	unsigned char ddir;

	struct list_head list;
};

#define FIO_HDR_MAGIC	0xf00baaef

enum {
	VERIFY_NONE = 0,
	VERIFY_MD5,
	VERIFY_CRC32,
};

struct verify_header {
	unsigned int fio_magic;
	unsigned int len;
	unsigned int verify_type;
	union {
		char md5_digest[MD5_HASH_WORDS * 4];
		unsigned long crc32;
	};
};

struct group_run_stats {
	unsigned long long max_run[2], min_run[2];
	unsigned long long max_bw[2], min_bw[2];
	unsigned long long io_kb[2];
	unsigned long long agg[2];
};

struct thread_data {
	char name[64];
	char file_name[256];
	char *directory;
	char verror[80];
	pthread_t thread;
	int thread_number;
	int groupid;
	int filetype;
	int error;
	int fd;
	void *mmap;
	pid_t pid;
	char *orig_buffer;
	size_t orig_buffer_size;
	volatile int terminate;
	volatile int runstate;
	volatile int old_runstate;
	unsigned int ddir;
	unsigned int iomix;
	unsigned int ioprio;
	unsigned int sequential;
	unsigned int bs;
	unsigned int min_bs;
	unsigned int max_bs;
	unsigned int odirect;
	unsigned int thinktime;
	unsigned int fsync_blocks;
	unsigned int start_delay;
	unsigned int timeout;
	unsigned int io_engine;
	unsigned int create_file;
	unsigned int overwrite;
	unsigned int invalidate_cache;
	unsigned int bw_avg_time;
	unsigned int create_serialize;
	unsigned int create_fsync;
	unsigned int end_fsync;
	unsigned int loops;
	unsigned long long file_size;
	unsigned long long real_file_size;
	unsigned long long file_offset;
	unsigned long long zone_size;
	unsigned long long zone_skip;
	unsigned int sync_io;
	unsigned int mem_type;
	unsigned int verify;
	unsigned int stonewall;
	unsigned int numjobs;
	unsigned int use_thread;
	unsigned int iodepth;
	os_cpu_mask_t cpumask;
	unsigned int jobnum;
	unsigned int iolog;
	unsigned int read_iolog;
	unsigned int write_iolog;
	unsigned int rwmixcycle;
	unsigned int rwmixread;
	unsigned int nice;

	char *iolog_file;
	void *iolog_buf;
	FILE *iolog_f;

	char *sysfs_root;

	char *ioscheduler;

	os_random_state_t bsrange_state;
	os_random_state_t verify_state;

	int shm_id;

	void *io_data;
	char io_engine_name[16];
	int (*io_prep)(struct thread_data *, struct io_u *);
	int (*io_queue)(struct thread_data *, struct io_u *);
	int (*io_getevents)(struct thread_data *, int, int, struct timespec *);
	struct io_u *(*io_event)(struct thread_data *, int);
	int (*io_cancel)(struct thread_data *, struct io_u *);
	void (*io_cleanup)(struct thread_data *);
	int (*io_sync)(struct thread_data *);

	unsigned int cur_depth;
	struct list_head io_u_freelist;
	struct list_head io_u_busylist;

	unsigned int rate;
	unsigned int ratemin;
	unsigned int ratecycle;
	unsigned long rate_usec_cycle;
	long rate_pending_usleep;
	unsigned long rate_bytes;
	struct timeval lastrate;

	unsigned long runtime[2];		/* msec */
	unsigned long long io_size;
	unsigned long long total_io_size;

	unsigned long long io_blocks[2];
	unsigned long long io_bytes[2];
	unsigned long long zone_bytes;
	unsigned long long this_io_bytes[2];
	unsigned long long last_pos;
	volatile int mutex;

	os_random_state_t random_state;
	unsigned long *file_map;
	unsigned int num_maps;

	/*
	 * bandwidth and latency stats
	 */
	struct io_stat clat_stat[2];		/* completion latency */
	struct io_stat slat_stat[2];		/* submission latency */
	struct io_stat bw_stat[2];		/* bandwidth stats */

	unsigned long long stat_io_bytes[2];
	struct timeval stat_sample_time[2];

	struct io_log *slat_log;
	struct io_log *clat_log;
	struct io_log *bw_log;

	struct timeval start;	/* start of this loop */
	struct timeval epoch;	/* time job was started */

	struct rusage ru_start;
	struct rusage ru_end;
	unsigned long usr_time;
	unsigned long sys_time;
	unsigned long ctx;

	unsigned int do_disk_util;
	unsigned int override_sync;

	os_random_state_t rwmix_state;
	struct timeval rwmix_switch;
	int rwmix_ddir;

	/*
	 * Pre-run and post-run shell
	 */
	char *exec_prerun;
	char *exec_postrun;

	struct list_head io_hist_list;
	struct list_head io_log_list;
};

#define td_verror(td, err)						\
	do {								\
		int e = (err);						\
		(td)->error = e;					\
		snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, strerror(e));	\
	} while (0)

extern struct io_u *__get_io_u(struct thread_data *);
extern void put_io_u(struct thread_data *, struct io_u *);

extern int rate_quit;
extern int write_lat_log;
extern int write_bw_log;
extern int exitall_on_terminate;
extern int thread_number;
extern int shm_id;
extern int groupid;

extern struct thread_data *threads;

enum {
	DDIR_READ = 0,
	DDIR_WRITE,
};

/*
 * What type of allocation to use for io buffers
 */
enum {
	MEM_MALLOC,	/* ordinary malloc */
	MEM_SHM,	/* use shared memory segments */
	MEM_MMAP,	/* use anonynomous mmap */
};

/*
 * The type of object we are working on
 */
enum {
	FIO_TYPE_FILE = 1,
	FIO_TYPE_BD,
	FIO_TYPE_CHAR,
};

enum {
	FIO_SYNCIO	= 1 << 0,
	FIO_MMAPIO	= 1 << 1 | FIO_SYNCIO,
	FIO_LIBAIO	= 1 << 2,
	FIO_POSIXAIO	= 1 << 3,
	FIO_SGIO	= 1 << 4,
	FIO_SPLICEIO	= 1 << 5 | FIO_SYNCIO,
};

#define td_read(td)		((td)->ddir == DDIR_READ)
#define td_write(td)		((td)->ddir == DDIR_WRITE)
#define td_rw(td)		((td)->iomix != 0)

#define BLOCKS_PER_MAP		(8 * sizeof(long))
#define TO_MAP_BLOCK(td, b)	((b) - ((td)->file_offset / (td)->min_bs))
#define RAND_MAP_IDX(td, b)	(TO_MAP_BLOCK(td, b) / BLOCKS_PER_MAP)
#define RAND_MAP_BIT(td, b)	(TO_MAP_BLOCK(td, b) & (BLOCKS_PER_MAP - 1))

#define MAX_JOBS	(1024)

struct disk_util_stat {
	unsigned ios[2];
	unsigned merges[2];
	unsigned long long sectors[2];
	unsigned ticks[2];
	unsigned io_ticks;
	unsigned time_in_queue;
};

struct disk_util {
	struct list_head list;

	char *name;
	char path[256];
	dev_t dev;

	struct disk_util_stat dus;
	struct disk_util_stat last_dus;

	unsigned long msec;
	struct timeval time;
};

struct io_completion_data {
	int nr;				/* input */

	int error;			/* output */
	unsigned long bytes_done[2];	/* output */
};

#define DISK_UTIL_MSEC	(250)

#ifndef min
#define min(a, b)	((a) < (b) ? (a) : (b))
#endif

/*
 * Log exports
 */
extern int read_iolog_get(struct thread_data *, struct io_u *);
extern void write_iolog_put(struct thread_data *, struct io_u *);
extern int init_iolog(struct thread_data *td);
extern void log_io_piece(struct thread_data *, struct io_u *);
extern void prune_io_piece_log(struct thread_data *);
extern void write_iolog_close(struct thread_data *);

/*
 * Logging
 */
extern void add_clat_sample(struct thread_data *, int, unsigned long);
extern void add_slat_sample(struct thread_data *, int, unsigned long);
extern void add_bw_sample(struct thread_data *, int);
extern void show_run_stats(void);
extern void init_disk_util(struct thread_data *);
extern void update_rusage_stat(struct thread_data *);
extern void update_io_ticks(void);
extern void disk_util_timer_arm(void);
extern void setup_log(struct io_log **);
extern void finish_log(struct thread_data *, struct io_log *, const char *);
extern int setup_rate(struct thread_data *);

/*
 * Time functions
 */
extern unsigned long utime_since(struct timeval *, struct timeval *);
extern unsigned long mtime_since(struct timeval *, struct timeval *);
extern unsigned long mtime_since_now(struct timeval *);
extern unsigned long time_since_now(struct timeval *);
extern void usec_sleep(struct thread_data *, unsigned long);
extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);

/*
 * Init functions
 */
extern int parse_options(int, char **);
extern int init_random_state(struct thread_data *);

/*
 * This is a pretty crappy semaphore implementation, but with the use that fio
 * has (just signalling start/go conditions), it doesn't have to be better.
 * Naturally this would not work for any type of contended semaphore or
 * for real locking.
 */
static inline void fio_sem_init(volatile int volatile *sem, int val)
{
	*sem = val;
}

static inline void fio_sem_down(volatile int volatile *sem)
{
	while (*sem == 0)
		usleep(10000);

	(*sem)--;
}

static inline void fio_sem_up(volatile int volatile *sem)
{
	(*sem)++;
}

#endif
Commit	Line	Data
	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
	13	#include "list.h"
	14	#include "md5.h"
	15	#include "crc32.h"
	16	#include "arch.h"
	17	#include "os.h"
	18
	19	struct io_stat {
	20	unsigned long val;
	21	unsigned long val_sq;
	22	unsigned long max_val;
	23	unsigned long min_val;
	24	unsigned long samples;
	25	};
	26
	27	struct io_sample {
	28	unsigned long time;
	29	unsigned long val;
	30	unsigned int ddir;
	31	};
	32
	33	struct io_log {
	34	unsigned long nr_samples;
	35	unsigned long max_samples;
	36	struct io_sample *log;
	37	};
	38
	39	struct io_piece {
	40	struct list_head list;
	41	unsigned long long offset;
	42	unsigned int len;
	43	int ddir;
	44	};
	45
	46	/*
	47	* The io unit
	48	*/
	49	struct io_u {
	50	union {
	51	#ifdef FIO_HAVE_LIBAIO
	52	struct iocb iocb;
	53	#endif
	54	#ifdef FIO_HAVE_POSIXAIO
	55	struct aiocb aiocb;
	56	#endif
	57	#ifdef FIO_HAVE_SGIO
	58	struct sg_io_hdr hdr;
	59	#endif
	60	};
	61	struct timeval start_time;
	62	struct timeval issue_time;
	63
	64	char *buf;
	65	unsigned int buflen;
	66	unsigned long long offset;
	67	unsigned int index;
	68
	69	unsigned int resid;
	70	unsigned int error;
	71
	72	unsigned char seen;
	73	unsigned char ddir;
	74
	75	struct list_head list;
	76	};
	77
	78	#define FIO_HDR_MAGIC 0xf00baaef
	79
	80	enum {
	81	VERIFY_NONE = 0,
	82	VERIFY_MD5,
	83	VERIFY_CRC32,
	84	};
	85
	86	struct verify_header {
	87	unsigned int fio_magic;
	88	unsigned int len;
	89	unsigned int verify_type;
	90	union {
	91	char md5_digest[MD5_HASH_WORDS * 4];
	92	unsigned long crc32;
	93	};
	94	};
	95
	96	struct group_run_stats {
	97	unsigned long long max_run[2], min_run[2];
	98	unsigned long long max_bw[2], min_bw[2];
	99	unsigned long long io_kb[2];
	100	unsigned long long agg[2];
	101	};
	102
	103	struct thread_data {
	104	char name[64];
	105	char file_name[256];
	106	char *directory;
	107	char verror[80];
	108	pthread_t thread;
	109	int thread_number;
	110	int groupid;
	111	int filetype;
	112	int error;
	113	int fd;
	114	void *mmap;
	115	pid_t pid;
	116	char *orig_buffer;
	117	size_t orig_buffer_size;
	118	volatile int terminate;
	119	volatile int runstate;
	120	volatile int old_runstate;
	121	unsigned int ddir;
	122	unsigned int iomix;
	123	unsigned int ioprio;
	124	unsigned int sequential;
	125	unsigned int bs;
	126	unsigned int min_bs;
	127	unsigned int max_bs;
	128	unsigned int odirect;
	129	unsigned int thinktime;
	130	unsigned int fsync_blocks;
	131	unsigned int start_delay;
	132	unsigned int timeout;
	133	unsigned int io_engine;
	134	unsigned int create_file;
	135	unsigned int overwrite;
	136	unsigned int invalidate_cache;
	137	unsigned int bw_avg_time;
	138	unsigned int create_serialize;
	139	unsigned int create_fsync;
	140	unsigned int end_fsync;
	141	unsigned int loops;
	142	unsigned long long file_size;
	143	unsigned long long real_file_size;
	144	unsigned long long file_offset;
	145	unsigned long long zone_size;
	146	unsigned long long zone_skip;
	147	unsigned int sync_io;
	148	unsigned int mem_type;
	149	unsigned int verify;
	150	unsigned int stonewall;
	151	unsigned int numjobs;
	152	unsigned int use_thread;
	153	unsigned int iodepth;
	154	os_cpu_mask_t cpumask;
	155	unsigned int jobnum;
	156	unsigned int iolog;
	157	unsigned int read_iolog;
	158	unsigned int write_iolog;
	159	unsigned int rwmixcycle;
	160	unsigned int rwmixread;
	161	unsigned int nice;
	162
	163	char *iolog_file;
	164	void *iolog_buf;
	165	FILE *iolog_f;
	166
	167	char *sysfs_root;
	168
	169	char *ioscheduler;
	170
	171	os_random_state_t bsrange_state;
	172	os_random_state_t verify_state;
	173
	174	int shm_id;
	175
	176	void *io_data;
	177	char io_engine_name[16];
	178	int (io_prep)(struct thread_data , struct io_u *);
	179	int (io_queue)(struct thread_data , struct io_u *);
	180	int (io_getevents)(struct thread_data , int, int, struct timespec *);
	181	struct io_u (io_event)(struct thread_data *, int);
	182	int (io_cancel)(struct thread_data , struct io_u *);
	183	void (io_cleanup)(struct thread_data );
	184	int (io_sync)(struct thread_data );
	185
	186	unsigned int cur_depth;
	187	struct list_head io_u_freelist;
	188	struct list_head io_u_busylist;
	189
	190	unsigned int rate;
	191	unsigned int ratemin;
	192	unsigned int ratecycle;
	193	unsigned long rate_usec_cycle;
	194	long rate_pending_usleep;
	195	unsigned long rate_bytes;
	196	struct timeval lastrate;
	197
	198	unsigned long runtime[2]; /* msec */
	199	unsigned long long io_size;
	200	unsigned long long total_io_size;
	201
	202	unsigned long long io_blocks[2];
	203	unsigned long long io_bytes[2];
	204	unsigned long long zone_bytes;
	205	unsigned long long this_io_bytes[2];
	206	unsigned long long last_pos;
	207	volatile int mutex;
	208
	209	os_random_state_t random_state;
	210	unsigned long *file_map;
	211	unsigned int num_maps;
	212
	213	/*
	214	* bandwidth and latency stats
	215	*/
	216	struct io_stat clat_stat[2]; /* completion latency */
	217	struct io_stat slat_stat[2]; /* submission latency */
	218	struct io_stat bw_stat[2]; /* bandwidth stats */
	219
	220	unsigned long long stat_io_bytes[2];
	221	struct timeval stat_sample_time[2];
	222
	223	struct io_log *slat_log;
	224	struct io_log *clat_log;
	225	struct io_log *bw_log;
	226
	227	struct timeval start; /* start of this loop */
	228	struct timeval epoch; /* time job was started */
	229
	230	struct rusage ru_start;
	231	struct rusage ru_end;
	232	unsigned long usr_time;
	233	unsigned long sys_time;
	234	unsigned long ctx;
	235
	236	unsigned int do_disk_util;
	237	unsigned int override_sync;
	238
	239	os_random_state_t rwmix_state;
	240	struct timeval rwmix_switch;
	241	int rwmix_ddir;
	242
	243	/*
	244	* Pre-run and post-run shell
	245	*/
	246	char *exec_prerun;
	247	char *exec_postrun;
	248
	249	struct list_head io_hist_list;
	250	struct list_head io_log_list;
	251	};
	252
	253	#define td_verror(td, err) \
	254	do { \
	255	int e = (err); \
	256	(td)->error = e; \
	257	snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, strerror(e)); \
	258	} while (0)
	259
	260	extern struct io_u __get_io_u(struct thread_data );
	261	extern void put_io_u(struct thread_data , struct io_u );
	262
	263	extern int rate_quit;
	264	extern int write_lat_log;
	265	extern int write_bw_log;
	266	extern int exitall_on_terminate;
	267	extern int thread_number;
	268	extern int shm_id;
	269	extern int groupid;
	270
	271	extern struct thread_data *threads;
	272
	273	enum {
	274	DDIR_READ = 0,
	275	DDIR_WRITE,
	276	};
	277
	278	/*
	279	* What type of allocation to use for io buffers
	280	*/
	281	enum {
	282	MEM_MALLOC, /* ordinary malloc */
	283	MEM_SHM, /* use shared memory segments */
	284	MEM_MMAP, /* use anonynomous mmap */
	285	};
	286
	287	/*
	288	* The type of object we are working on
	289	*/
	290	enum {
	291	FIO_TYPE_FILE = 1,
	292	FIO_TYPE_BD,
	293	FIO_TYPE_CHAR,
	294	};
	295
	296	enum {
	297	FIO_SYNCIO = 1 << 0,
	298	FIO_MMAPIO = 1 << 1 \| FIO_SYNCIO,
	299	FIO_LIBAIO = 1 << 2,
	300	FIO_POSIXAIO = 1 << 3,
	301	FIO_SGIO = 1 << 4,
	302	FIO_SPLICEIO = 1 << 5 \| FIO_SYNCIO,
	303	};
	304
	305	#define td_read(td) ((td)->ddir == DDIR_READ)
	306	#define td_write(td) ((td)->ddir == DDIR_WRITE)
	307	#define td_rw(td) ((td)->iomix != 0)
	308
	309	#define BLOCKS_PER_MAP (8 * sizeof(long))
	310	#define TO_MAP_BLOCK(td, b) ((b) - ((td)->file_offset / (td)->min_bs))
	311	#define RAND_MAP_IDX(td, b) (TO_MAP_BLOCK(td, b) / BLOCKS_PER_MAP)
	312	#define RAND_MAP_BIT(td, b) (TO_MAP_BLOCK(td, b) & (BLOCKS_PER_MAP - 1))
	313
	314	#define MAX_JOBS (1024)
	315
	316	struct disk_util_stat {
	317	unsigned ios[2];
	318	unsigned merges[2];
	319	unsigned long long sectors[2];
	320	unsigned ticks[2];
	321	unsigned io_ticks;
	322	unsigned time_in_queue;
	323	};
	324
	325	struct disk_util {
	326	struct list_head list;
	327
	328	char *name;
	329	char path[256];
	330	dev_t dev;
	331
	332	struct disk_util_stat dus;
	333	struct disk_util_stat last_dus;
	334
	335	unsigned long msec;
	336	struct timeval time;
	337	};
	338
	339	struct io_completion_data {
	340	int nr; /* input */
	341
	342	int error; /* output */
	343	unsigned long bytes_done[2]; /* output */
	344	};
	345
	346	#define DISK_UTIL_MSEC (250)
	347
	348	#ifndef min
	349	#define min(a, b) ((a) < (b) ? (a) : (b))
	350	#endif
	351
	352	/*
	353	* Log exports
	354	*/
	355	extern int read_iolog_get(struct thread_data , struct io_u );
	356	extern void write_iolog_put(struct thread_data , struct io_u );
	357	extern int init_iolog(struct thread_data *td);
	358	extern void log_io_piece(struct thread_data , struct io_u );
	359	extern void prune_io_piece_log(struct thread_data *);
	360	extern void write_iolog_close(struct thread_data *);
	361
	362	/*
	363	* Logging
	364	*/
	365	extern void add_clat_sample(struct thread_data *, int, unsigned long);
	366	extern void add_slat_sample(struct thread_data *, int, unsigned long);
	367	extern void add_bw_sample(struct thread_data *, int);
	368	extern void show_run_stats(void);
	369	extern void init_disk_util(struct thread_data *);
	370	extern void update_rusage_stat(struct thread_data *);
	371	extern void update_io_ticks(void);
	372	extern void disk_util_timer_arm(void);
	373	extern void setup_log(struct io_log **);
	374	extern void finish_log(struct thread_data , struct io_log , const char *);
	375	extern int setup_rate(struct thread_data *);
	376
	377	/*
	378	* Time functions
	379	*/
	380	extern unsigned long utime_since(struct timeval , struct timeval );
	381	extern unsigned long mtime_since(struct timeval , struct timeval );
	382	extern unsigned long mtime_since_now(struct timeval *);
	383	extern unsigned long time_since_now(struct timeval *);
	384	extern void usec_sleep(struct thread_data *, unsigned long);
	385	extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
	386
	387	/*
	388	* Init functions
	389	*/
	390	extern int parse_options(int, char **);
	391	extern int init_random_state(struct thread_data *);
	392
	393	/*
	394	* This is a pretty crappy semaphore implementation, but with the use that fio
	395	* has (just signalling start/go conditions), it doesn't have to be better.
	396	* Naturally this would not work for any type of contended semaphore or
	397	* for real locking.
	398	*/
	399	static inline void fio_sem_init(volatile int volatile *sem, int val)
	400	{
	401	*sem = val;
	402	}
	403
	404	static inline void fio_sem_down(volatile int volatile *sem)
	405	{
	406	while (*sem == 0)
	407	usleep(10000);
	408
	409	(*sem)--;
	410	}
	411
	412	static inline void fio_sem_up(volatile int volatile *sem)
	413	{
	414	(*sem)++;
	415	}
	416
	417	#endif