[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 <unistd.h>
#include <string.h>

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

#include "syslet.h"

enum fio_ddir {
	DDIR_READ = 0,
	DDIR_WRITE,
	DDIR_SYNC,
};

/*
 * Use for maintaining statistics
 */
struct io_stat {
	unsigned long max_val;
	unsigned long min_val;
	unsigned long samples;

	double mean;
	double S;
};

/*
 * A single data sample
 */
struct io_sample {
	unsigned long time;
	unsigned long val;
	enum fio_ddir ddir;
};

/*
 * Dynamically growing data sample log
 */
struct io_log {
	unsigned long nr_samples;
	unsigned long max_samples;
	struct io_sample *log;
};

/*
 * When logging io actions, this matches a single sent io_u
 */
struct io_piece {
	struct list_head list;
	struct fio_file *file;
	unsigned long long offset;
	unsigned long len;
	enum fio_ddir ddir;
};

#ifdef FIO_HAVE_SYSLET
struct syslet_req {
	struct syslet_uatom atom;
	long ret;
};
#endif

/*
 * 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
#ifdef FIO_HAVE_SYSLET
		struct syslet_req req;
#endif
	};
	struct timeval start_time;
	struct timeval issue_time;

	/*
	 * Allocated/set buffer and length
	 */
	void *buf;
	unsigned long buflen;
	unsigned long long offset;

	/*
	 * IO engine state, may be different from above when we get
	 * partial transfers / residual data counts
	 */
	void *xfer_buf;
	unsigned long xfer_buflen;

	unsigned int resid;
	unsigned int error;

	enum fio_ddir ddir;

	/*
	 * io engine private data
	 */
	union {
		unsigned int index;
		unsigned int seen;
	};

	struct fio_file *file;

	struct list_head list;
};

#define FIO_HDR_MAGIC	0xf00baaef

enum {
	VERIFY_NONE = 0,		/* no verification */
	VERIFY_MD5,			/* md5 sum data blocks */
	VERIFY_CRC32,			/* crc32 sum data blocks */
};

/*
 * A header structure associated with each checksummed data block
 */
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];
};

/*
 * What type of allocation to use for io buffers
 */
enum fio_memtype {
	MEM_MALLOC = 0,	/* ordinary malloc */
	MEM_SHM,	/* use shared memory segments */
	MEM_SHMHUGE,	/* use shared memory segments with huge pages */
	MEM_MMAP,	/* use anonynomous mmap */
	MEM_MMAPHUGE,	/* memory mapped huge file */
};

/*
 * The type of object we are working on
 */
enum fio_filetype {
	FIO_TYPE_FILE = 1,		/* plain file */
	FIO_TYPE_BD,			/* block device */
	FIO_TYPE_CHAR,			/* character device */
};

enum fio_ioengine_flags {
	FIO_SYNCIO	= 1 << 0,	/* io engine has synchronous ->queue */
	FIO_CPUIO	= 1 << 1,	/* cpu burner, doesn't do real io */
	FIO_MMAPIO	= 1 << 2,	/* uses memory mapped io */
	FIO_RAWIO	= 1 << 3,	/* some sort of direct/raw io */
	FIO_NETIO	= 1 << 4,	/* networked io */
	FIO_NULLIO	= 1 << 5,	/* no real data transfer (cpu/null) */
};

/*
 * Each thread_data structure has a number of files associated with it,
 * this structure holds state information for a single file.
 */
struct fio_file {
	/*
	 * A file may not be a file descriptor, let the io engine decide
	 */
	union {
		unsigned long file_data;
		long fd;
	};
	char *file_name;
	void *mmap;
	unsigned long long file_size;
	unsigned long long real_file_size;
	unsigned long long file_offset;
	unsigned long long last_pos;
	unsigned long long last_completed_pos;

	/*
	 * block map for random io
	 */
	unsigned long *file_map;
	unsigned int num_maps;
	unsigned int last_free_lookup;

	unsigned int unlink;
};

/*
 * How many depth levels to log
 */
#define FIO_IO_U_MAP_NR	8

/*
 * This describes a single thread/process executing a fio job.
 */
struct thread_data {
	char *description;
	char *name;
	char *directory;
	char *filename;
	char verror[80];
	pthread_t thread;
	int thread_number;
	int groupid;
	enum fio_filetype filetype;
	struct fio_file *files;
	unsigned int nr_files;
	unsigned int nr_uniq_files;
	unsigned int next_file;
	int error;
	pid_t pid;
	char *orig_buffer;
	size_t orig_buffer_size;
	volatile int terminate;
	volatile int runstate;
	enum fio_ddir ddir;
	unsigned int iomix;
	unsigned int ioprio;
	unsigned int last_was_sync;

	unsigned int sequential;
	unsigned int odirect;
	unsigned int invalidate_cache;
	unsigned int create_serialize;
	unsigned int create_fsync;
	unsigned int end_fsync;
	unsigned int sync_io;
	unsigned int verify;
	unsigned int use_thread;
	unsigned int unlink;
	unsigned int do_disk_util;
	unsigned int override_sync;
	unsigned int rand_repeatable;
	unsigned int write_lat_log;
	unsigned int write_bw_log;
	unsigned int norandommap;
	unsigned int bs_unaligned;

	unsigned int bs[2];
	unsigned int min_bs[2];
	unsigned int max_bs[2];
	unsigned int hugepage_size;
	unsigned int rw_min_bs;
	unsigned int thinktime;
	unsigned int thinktime_blocks;
	unsigned int fsync_blocks;
	unsigned int start_delay;
	unsigned long timeout;
	unsigned int overwrite;
	unsigned int bw_avg_time;
	unsigned int loops;
	unsigned long long zone_size;
	unsigned long long zone_skip;
	enum fio_memtype mem_type;
	char *mmapfile;
	int mmapfd;
	unsigned int stonewall;
	unsigned int numjobs;
	unsigned int iodepth;
	os_cpu_mask_t cpumask;
	unsigned int iolog;
	unsigned int read_iolog;
	unsigned int rwmixcycle;
	unsigned int rwmixread;
	unsigned int rwmixwrite;
	unsigned int nice;

	char *read_iolog_file;
	char *write_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;

	/*
	 * IO engine hooks, contains everything needed to submit an io_u
	 * to any of the available IO engines.
	 */
	struct ioengine_ops *io_ops;

	/*
	 * Current IO depth and list of free and busy io_u's.
	 */
	unsigned int cur_depth;
	unsigned int io_u_map[FIO_IO_U_MAP_NR];
	unsigned long total_io_u;
	struct list_head io_u_freelist;
	struct list_head io_u_busylist;

	/*
	 * Rate state
	 */
	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_file_size;
	unsigned long long start_offset;
	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];
	volatile int mutex;

	/*
	 * State for random io, a bitmap of blocks done vs not done
	 */
	os_random_state_t random_state;

	/*
	 * CPU "io" cycle burner
	 */
	unsigned int cpuload;
	unsigned int cpucycle;

	/*
	 * 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 timeval end_time;/* time job ended */

	/*
	 * fio system usage accounting
	 */
	struct rusage ru_start;
	struct rusage ru_end;
	unsigned long usr_time;
	unsigned long sys_time;
	unsigned long ctx;

	/*
	 * read/write mixed workload state
	 */
	os_random_state_t rwmix_state;
	struct timeval rwmix_switch;
	enum fio_ddir rwmix_ddir;

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

	/*
	 * IO historic logs
	 */
	struct list_head io_hist_list;
	struct list_head io_log_list;
};

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


#define td_verror(td, err)	__td_verror((td), (err), strerror((err)))
#define td_vmsg(td, err, msg)	__td_verror((td), (err), (msg))

extern int exitall_on_terminate;
extern int thread_number;
extern int shm_id;
extern int groupid;
extern int terse_output;
extern FILE *f_out;
extern FILE *f_err;
extern int temp_stall_ts;
extern unsigned long long mlock_size;

extern struct thread_data *threads;

#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, f, b)	((b) - ((f)->file_offset / (td)->rw_min_bs))
#define RAND_MAP_IDX(td, f, b)	(TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
#define RAND_MAP_BIT(td, f, b)	(TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))

#define MAX_JOBS	(1024)

static inline int should_fsync(struct thread_data *td)
{
	if (td->last_was_sync)
		return 0;
	if (td->odirect)
		return 0;
	if (td_write(td) || td_rw(td) || td->override_sync)
		return 1;

	return 0;
}

/*
 * Disk utils as read in /sys/block/<dev>/stat
 */
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;
};

/*
 * Per-device disk util management
 */
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;
};

/*
 * Used for passing io_u completion data
 */
struct io_completion_data {
	int nr;				/* input */

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

#define DISK_UTIL_MSEC	(250)

#ifndef min
#define min(a, b)	((a) < (b) ? (a) : (b))
#endif
#ifndef max
#define max(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 *, enum fio_ddir, unsigned long);
extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
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 void __finish_log(struct io_log *, const char *);
extern int setup_rate(struct thread_data *);
extern struct io_log *agg_io_log[2];
extern int write_bw_log;
extern void add_agg_sample(unsigned long, enum fio_ddir);

/*
 * Time functions
 */
extern unsigned long utime_since(struct timeval *, struct timeval *);
extern unsigned long utime_since_now(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 unsigned long mtime_since_genesis(void);
extern void __usec_sleep(unsigned int);
extern void usec_sleep(struct thread_data *, unsigned long);
extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
extern void fill_start_time(struct timeval *);
extern void fio_gettime(struct timeval *, void *);

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

/*
 * File setup/shutdown
 */
extern void close_files(struct thread_data *);
extern int setup_files(struct thread_data *);
extern int open_files(struct thread_data *);
extern int file_invalidate_cache(struct thread_data *, struct fio_file *);

/*
 * ETA/status stuff
 */
extern void print_thread_status(void);
extern void print_status_init(int);

/*
 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
 * will never back again. It may cycle between running/verififying/fsyncing.
 * Once the thread reaches TD_EXITED, it is just waiting for the core to
 * reap it.
 */
enum {
	TD_NOT_CREATED = 0,
	TD_CREATED,
	TD_INITIALIZED,
	TD_RUNNING,
	TD_VERIFYING,
	TD_FSYNCING,
	TD_EXITED,
	TD_REAPED,
};

/*
 * Verify helpers
 */
extern void populate_verify_io_u(struct thread_data *, struct io_u *);
extern int get_next_verify(struct thread_data *td, struct io_u *);
extern int do_io_u_verify(struct thread_data *, struct io_u **);

/*
 * Memory helpers
 */
extern int fio_pin_memory(void);
extern void fio_unpin_memory(void);
extern int allocate_io_mem(struct thread_data *);
extern void free_io_mem(struct thread_data *);

/*
 * io unit handling
 */
#define queue_full(td)	list_empty(&(td)->io_u_freelist)
extern struct io_u *__get_io_u(struct thread_data *);
extern struct io_u *get_io_u(struct thread_data *, struct fio_file *);
extern void put_io_u(struct thread_data *, struct io_u *);
extern void ios_completed(struct thread_data *, struct io_completion_data *);
extern void io_completed(struct thread_data *, struct io_u *, struct io_completion_data *);

/*
 * io engine entry points
 */
extern int td_io_init(struct thread_data *);
extern int td_io_prep(struct thread_data *, struct io_u *);
extern int td_io_queue(struct thread_data *, struct io_u *);
extern int td_io_sync(struct thread_data *, struct fio_file *);
extern int td_io_getevents(struct thread_data *, int, int, struct timespec *);

/*
 * 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 *sem, int val)
{
	*sem = val;
}

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

	(*sem)--;
}

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

/*
 * If logging output to a file, stderr should go to both stderr and f_err
 */
#define log_err(args...)	do {		\
	fprintf(f_err, ##args);			\
	if (f_err != stderr)			\
		fprintf(stderr, ##args);	\
	} while (0)

struct ioengine_ops {
	struct list_head list;
	char name[16];
	int version;
	int flags;
	int (*setup)(struct thread_data *);
	int (*init)(struct thread_data *);
	int (*prep)(struct thread_data *, struct io_u *);
	int (*queue)(struct thread_data *, struct io_u *);
	int (*getevents)(struct thread_data *, int, int, struct timespec *);
	struct io_u *(*event)(struct thread_data *, int);
	int (*cancel)(struct thread_data *, struct io_u *);
	void (*cleanup)(struct thread_data *);
	void *data;
	void *dlhandle;
};

#define FIO_IOOPS_VERSION	3

extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
extern int register_ioengine(struct ioengine_ops *);
extern void unregister_ioengine(struct ioengine_ops *);
extern void close_ioengine(struct thread_data *);

/*
 * Mark unused variables passed to ops functions as unused, to silence gcc
 */
#define fio_unused	__attribute((__unused__))
#define fio_init	__attribute__((constructor))
#define fio_exit	__attribute__((destructor))

#define for_each_td(td, i)	\
	for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
#define for_each_file(td, f, i)	\
	for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)

#define fio_assert(td, cond)	do {	\
	if (!(cond)) {			\
		int *__foo = NULL;	\
		fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond);	\
		(td)->runstate = TD_EXITED;	\
		(td)->error = EFAULT;		\
		*__foo = 0;			\
	}	\
} while (0)

#endif
Commit	Line	Data
ebac4655 JA	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>
3c39a379 JA	9	#include <errno.h>
	10	#include <stdlib.h>
	11	#include <stdio.h>
6d6f031f	12	#include <unistd.h>
34cfcdaf	13	#include <string.h>
ebac4655 JA	14
	15	#include "list.h"
	16	#include "md5.h"
	17	#include "crc32.h"
	18	#include "arch.h"
	19	#include "os.h"
	20
a4f4fdd7 JA	21	#include "syslet.h"
a4f4fdd7 JA	22
1e97cce9 JA	23	enum fio_ddir {
	24	DDIR_READ = 0,
	25	DDIR_WRITE,
	26	DDIR_SYNC,
	27	};
	28
1c9e06ee JA	29	/*
	30	* Use for maintaining statistics
	31	*/
ebac4655	32	struct io_stat {
ebac4655 JA	33	unsigned long max_val;
	34	unsigned long min_val;
	35	unsigned long samples;
68704084 JA	36
	37	double mean;
	38	double S;
ebac4655 JA	39	};
ebac4655 JA	40
1c9e06ee JA	41	/*
	42	* A single data sample
	43	*/
ebac4655 JA	44	struct io_sample {
	45	unsigned long time;
	46	unsigned long val;
1e97cce9	47	enum fio_ddir ddir;
ebac4655 JA	48	};
ebac4655 JA	49
1c9e06ee JA	50	/*
	51	* Dynamically growing data sample log
	52	*/
ebac4655 JA	53	struct io_log {
	54	unsigned long nr_samples;
	55	unsigned long max_samples;
	56	struct io_sample *log;
	57	};
	58
1c9e06ee JA	59	/*
	60	* When logging io actions, this matches a single sent io_u
	61	*/
ebac4655 JA	62	struct io_piece {
ebac4655 JA	63	struct list_head list;
53cdc686	64	struct fio_file *file;
ebac4655	65	unsigned long long offset;
a4f4fdd7	66	unsigned long len;
1e97cce9	67	enum fio_ddir ddir;
ebac4655 JA	68	};
ebac4655 JA	69
a4f4fdd7 JA	70	#ifdef FIO_HAVE_SYSLET
	71	struct syslet_req {
	72	struct syslet_uatom atom;
a4f4fdd7 JA	73	long ret;
	74	};
	75	#endif
	76
ebac4655 JA	77	/*
	78	* The io unit
	79	*/
	80	struct io_u {
	81	union {
	82	#ifdef FIO_HAVE_LIBAIO
	83	struct iocb iocb;
	84	#endif
	85	#ifdef FIO_HAVE_POSIXAIO
	86	struct aiocb aiocb;
	87	#endif
	88	#ifdef FIO_HAVE_SGIO
	89	struct sg_io_hdr hdr;
a4f4fdd7 JA	90	#endif
a4f4fdd7 JA	91	#ifdef FIO_HAVE_SYSLET
a2e1b08a	92	struct syslet_req req;
ebac4655 JA	93	#endif
	94	};
	95	struct timeval start_time;
	96	struct timeval issue_time;
	97
1c9e06ee JA	98	/*
	99	* Allocated/set buffer and length
	100	*/
6b9cea23	101	void *buf;
a4f4fdd7	102	unsigned long buflen;
ebac4655 JA	103	unsigned long long offset;
ebac4655 JA	104
1c9e06ee JA	105	/*
	106	* IO engine state, may be different from above when we get
	107	* partial transfers / residual data counts
	108	*/
cec6b55d	109	void *xfer_buf;
a4f4fdd7	110	unsigned long xfer_buflen;
cec6b55d	111
ebac4655 JA	112	unsigned int resid;
	113	unsigned int error;
	114
1e97cce9	115	enum fio_ddir ddir;
ebac4655	116
dfd7bc2c JA	117	/*
	118	* io engine private data
	119	*/
	120	union {
	121	unsigned int index;
	122	unsigned int seen;
	123	};
	124
53cdc686 JA	125	struct fio_file *file;
53cdc686 JA	126
ebac4655 JA	127	struct list_head list;
	128	};
	129
	130	#define FIO_HDR_MAGIC 0xf00baaef
	131
	132	enum {
1c9e06ee JA	133	VERIFY_NONE = 0, /* no verification */
	134	VERIFY_MD5, /* md5 sum data blocks */
	135	VERIFY_CRC32, /* crc32 sum data blocks */
ebac4655 JA	136	};
ebac4655 JA	137
1c9e06ee JA	138	/*
	139	* A header structure associated with each checksummed data block
	140	*/
ebac4655 JA	141	struct verify_header {
	142	unsigned int fio_magic;
	143	unsigned int len;
	144	unsigned int verify_type;
	145	union {
	146	char md5_digest[MD5_HASH_WORDS * 4];
	147	unsigned long crc32;
	148	};
	149	};
	150
	151	struct group_run_stats {
9104f874 JA	152	unsigned long long max_run[2], min_run[2];
9104f874 JA	153	unsigned long long max_bw[2], min_bw[2];
e9b2a3fa	154	unsigned long long io_kb[2];
9104f874	155	unsigned long long agg[2];
ebac4655 JA	156	};
ebac4655 JA	157
e9c047a0 JA	158	/*
	159	* What type of allocation to use for io buffers
	160	*/
	161	enum fio_memtype {
	162	MEM_MALLOC = 0, /* ordinary malloc */
	163	MEM_SHM, /* use shared memory segments */
74b025b0	164	MEM_SHMHUGE, /* use shared memory segments with huge pages */
e9c047a0	165	MEM_MMAP, /* use anonynomous mmap */
d0bdaf49	166	MEM_MMAPHUGE, /* memory mapped huge file */
e9c047a0 JA	167	};
	168
	169	/*
	170	* The type of object we are working on
	171	*/
	172	enum fio_filetype {
1c9e06ee JA	173	FIO_TYPE_FILE = 1, /* plain file */
	174	FIO_TYPE_BD, /* block device */
	175	FIO_TYPE_CHAR, /* character device */
e9c047a0 JA	176	};
e9c047a0 JA	177
2866c82d	178	enum fio_ioengine_flags {
1c9e06ee JA	179	FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
	180	FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
	181	FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
	182	FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
	183	FIO_NETIO = 1 << 4, /* networked io */
	184	FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
e9c047a0 JA	185	};
e9c047a0 JA	186
1c9e06ee JA	187	/*
	188	* Each thread_data structure has a number of files associated with it,
	189	* this structure holds state information for a single file.
	190	*/
53cdc686 JA	191	struct fio_file {
	192	/*
	193	* A file may not be a file descriptor, let the io engine decide
	194	*/
	195	union {
	196	unsigned long file_data;
a4f4fdd7	197	long fd;
53cdc686 JA	198	};
	199	char *file_name;
	200	void *mmap;
	201	unsigned long long file_size;
	202	unsigned long long real_file_size;
	203	unsigned long long file_offset;
	204	unsigned long long last_pos;
02bcaa8c	205	unsigned long long last_completed_pos;
53cdc686	206
1c9e06ee JA	207	/*
	208	* block map for random io
	209	*/
53cdc686 JA	210	unsigned long *file_map;
53cdc686 JA	211	unsigned int num_maps;
1c9e06ee	212	unsigned int last_free_lookup;
9b031dc8 JA	213
9b031dc8 JA	214	unsigned int unlink;
53cdc686 JA	215	};
53cdc686 JA	216
71619dc2 JA	217	/*
	218	* How many depth levels to log
	219	*/
	220	#define FIO_IO_U_MAP_NR 8
	221
e9c047a0 JA	222	/*
	223	* This describes a single thread/process executing a fio job.
	224	*/
ebac4655	225	struct thread_data {
61697c37	226	char *description;
b4692828	227	char *name;
ef899b63	228	char *directory;
13f8e2d2	229	char *filename;
ebac4655 JA	230	char verror[80];
	231	pthread_t thread;
	232	int thread_number;
	233	int groupid;
e9c047a0	234	enum fio_filetype filetype;
53cdc686 JA	235	struct fio_file *files;
53cdc686 JA	236	unsigned int nr_files;
13f8e2d2	237	unsigned int nr_uniq_files;
53cdc686	238	unsigned int next_file;
ebac4655	239	int error;
ebac4655 JA	240	pid_t pid;
	241	char *orig_buffer;
	242	size_t orig_buffer_size;
	243	volatile int terminate;
	244	volatile int runstate;
e9c047a0	245	enum fio_ddir ddir;
3d60d1ed	246	unsigned int iomix;
ebac4655	247	unsigned int ioprio;
87dc1ab1	248	unsigned int last_was_sync;
e9c047a0	249
9158d2f7 JA	250	unsigned int sequential;
	251	unsigned int odirect;
	252	unsigned int invalidate_cache;
	253	unsigned int create_serialize;
	254	unsigned int create_fsync;
	255	unsigned int end_fsync;
	256	unsigned int sync_io;
	257	unsigned int verify;
	258	unsigned int use_thread;
	259	unsigned int unlink;
	260	unsigned int do_disk_util;
	261	unsigned int override_sync;
	262	unsigned int rand_repeatable;
	263	unsigned int write_lat_log;
	264	unsigned int write_bw_log;
bb8895e0	265	unsigned int norandommap;
690adba3	266	unsigned int bs_unaligned;
e9c047a0	267
a00735e6 JA	268	unsigned int bs[2];
	269	unsigned int min_bs[2];
	270	unsigned int max_bs[2];
56bb17f2	271	unsigned int hugepage_size;
a00735e6	272	unsigned int rw_min_bs;
ebac4655	273	unsigned int thinktime;
9c1f7434	274	unsigned int thinktime_blocks;
ebac4655 JA	275	unsigned int fsync_blocks;
ebac4655 JA	276	unsigned int start_delay;
906c8d75	277	unsigned long timeout;
ebac4655	278	unsigned int overwrite;
ebac4655	279	unsigned int bw_avg_time;
ebac4655	280	unsigned int loops;
20dc95c4 JA	281	unsigned long long zone_size;
20dc95c4 JA	282	unsigned long long zone_skip;
e9c047a0	283	enum fio_memtype mem_type;
313cb206 JA	284	char *mmapfile;
313cb206 JA	285	int mmapfd;
ebac4655 JA	286	unsigned int stonewall;
ebac4655 JA	287	unsigned int numjobs;
ebac4655 JA	288	unsigned int iodepth;
ebac4655 JA	289	os_cpu_mask_t cpumask;
aea47d44	290	unsigned int iolog;
843a7413	291	unsigned int read_iolog;
a6ccc7be JA	292	unsigned int rwmixcycle;
a6ccc7be JA	293	unsigned int rwmixread;
e1f36503	294	unsigned int rwmixwrite;
b6f4d880	295	unsigned int nice;
aea47d44	296
076efc7c JA	297	char *read_iolog_file;
076efc7c JA	298	char *write_iolog_file;
843a7413 JA	299	void *iolog_buf;
843a7413 JA	300	FILE *iolog_f;
ebac4655	301
da86774e	302	char *sysfs_root;
da86774e JA	303	char *ioscheduler;
da86774e JA	304
6dfd46b9 JA	305	os_random_state_t bsrange_state;
6dfd46b9 JA	306	os_random_state_t verify_state;
ebac4655 JA	307
	308	int shm_id;
	309
e9c047a0 JA	310	/*
	311	* IO engine hooks, contains everything needed to submit an io_u
	312	* to any of the available IO engines.
	313	*/
2866c82d	314	struct ioengine_ops *io_ops;
ebac4655	315
e9c047a0 JA	316	/*
	317	* Current IO depth and list of free and busy io_u's.
	318	*/
ebac4655	319	unsigned int cur_depth;
71619dc2 JA	320	unsigned int io_u_map[FIO_IO_U_MAP_NR];
71619dc2 JA	321	unsigned long total_io_u;
ebac4655 JA	322	struct list_head io_u_freelist;
	323	struct list_head io_u_busylist;
	324
e9c047a0 JA	325	/*
	326	* Rate state
	327	*/
ebac4655 JA	328	unsigned int rate;
	329	unsigned int ratemin;
	330	unsigned int ratecycle;
	331	unsigned long rate_usec_cycle;
	332	long rate_pending_usleep;
	333	unsigned long rate_bytes;
	334	struct timeval lastrate;
	335
	336	unsigned long runtime[2]; /* msec */
	337	unsigned long long io_size;
53cdc686 JA	338	unsigned long long total_file_size;
53cdc686 JA	339	unsigned long long start_offset;
ebac4655 JA	340	unsigned long long total_io_size;
ebac4655 JA	341
9104f874 JA	342	unsigned long long io_blocks[2];
	343	unsigned long long io_bytes[2];
	344	unsigned long long zone_bytes;
	345	unsigned long long this_io_bytes[2];
bbfd6b00	346	volatile int mutex;
ebac4655	347
e9c047a0 JA	348	/*
	349	* State for random io, a bitmap of blocks done vs not done
	350	*/
6dfd46b9	351	os_random_state_t random_state;
ebac4655	352
b990b5c0 JA	353	/*
	354	* CPU "io" cycle burner
	355	*/
	356	unsigned int cpuload;
	357	unsigned int cpucycle;
	358
ebac4655 JA	359	/*
	360	* bandwidth and latency stats
	361	*/
	362	struct io_stat clat_stat[2]; /* completion latency */
	363	struct io_stat slat_stat[2]; /* submission latency */
	364	struct io_stat bw_stat[2]; /* bandwidth stats */
	365
9104f874	366	unsigned long long stat_io_bytes[2];
ebac4655 JA	367	struct timeval stat_sample_time[2];
	368
	369	struct io_log *slat_log;
	370	struct io_log *clat_log;
	371	struct io_log *bw_log;
	372
	373	struct timeval start; /* start of this loop */
	374	struct timeval epoch; /* time job was started */
69008999	375	struct timeval end_time;/* time job ended */
ebac4655	376
e9c047a0 JA	377	/*
	378	* fio system usage accounting
	379	*/
ebac4655 JA	380	struct rusage ru_start;
	381	struct rusage ru_end;
	382	unsigned long usr_time;
	383	unsigned long sys_time;
	384	unsigned long ctx;
	385
e9c047a0 JA	386	/*
	387	* read/write mixed workload state
	388	*/
6dfd46b9	389	os_random_state_t rwmix_state;
a6ccc7be	390	struct timeval rwmix_switch;
e9c047a0	391	enum fio_ddir rwmix_ddir;
a6ccc7be	392
4e0ba8af JA	393	/*
	394	* Pre-run and post-run shell
	395	*/
	396	char *exec_prerun;
	397	char *exec_postrun;
	398
e9c047a0 JA	399	/*
	400	* IO historic logs
	401	*/
ebac4655	402	struct list_head io_hist_list;
aea47d44	403	struct list_head io_log_list;
ebac4655 JA	404	};
ebac4655 JA	405
b990b5c0	406	#define __td_verror(td, err, msg) \
ebac4655 JA	407	do { \
	408	int e = (err); \
	409	(td)->error = e; \
b990b5c0	410	snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, (msg)); \
ebac4655 JA	411	} while (0)
ebac4655 JA	412
b990b5c0 JA	413
	414	#define td_verror(td, err) __td_verror((td), (err), strerror((err)))
	415	#define td_vmsg(td, err, msg) __td_verror((td), (err), (msg))
	416
ebac4655 JA	417	extern int exitall_on_terminate;
	418	extern int thread_number;
	419	extern int shm_id;
	420	extern int groupid;
c6ae0a5b	421	extern int terse_output;
eb8bbf48 JA	422	extern FILE *f_out;
eb8bbf48 JA	423	extern FILE *f_err;
53cdc686	424	extern int temp_stall_ts;
1e97cce9	425	extern unsigned long long mlock_size;
ebac4655 JA	426
	427	extern struct thread_data *threads;
	428
ebac4655 JA	429	#define td_read(td) ((td)->ddir == DDIR_READ)
ebac4655 JA	430	#define td_write(td) ((td)->ddir == DDIR_WRITE)
3d60d1ed	431	#define td_rw(td) ((td)->iomix != 0)
ebac4655 JA	432
ebac4655 JA	433	#define BLOCKS_PER_MAP (8 * sizeof(long))
a00735e6	434	#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
53cdc686 JA	435	#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
53cdc686 JA	436	#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
ebac4655 JA	437
	438	#define MAX_JOBS (1024)
	439
87dc1ab1 JA	440	static inline int should_fsync(struct thread_data *td)
	441	{
	442	if (td->last_was_sync)
	443	return 0;
	444	if (td->odirect)
	445	return 0;
	446	if (td_write(td) \|\| td_rw(td) \|\| td->override_sync)
	447	return 1;
	448
	449	return 0;
	450	}
	451
1c9e06ee JA	452	/*
	453	* Disk utils as read in /sys/block/<dev>/stat
	454	*/
ebac4655 JA	455	struct disk_util_stat {
	456	unsigned ios[2];
	457	unsigned merges[2];
	458	unsigned long long sectors[2];
	459	unsigned ticks[2];
	460	unsigned io_ticks;
	461	unsigned time_in_queue;
	462	};
	463
1c9e06ee JA	464	/*
	465	* Per-device disk util management
	466	*/
ebac4655 JA	467	struct disk_util {
	468	struct list_head list;
	469
	470	char *name;
	471	char path[256];
	472	dev_t dev;
	473
	474	struct disk_util_stat dus;
	475	struct disk_util_stat last_dus;
	476
	477	unsigned long msec;
	478	struct timeval time;
	479	};
	480
1c9e06ee JA	481	/*
	482	* Used for passing io_u completion data
	483	*/
ebac4655 JA	484	struct io_completion_data {
	485	int nr; /* input */
	486
	487	int error; /* output */
	488	unsigned long bytes_done[2]; /* output */
02bcaa8c	489	struct timeval time; /* output */
ebac4655 JA	490	};
	491
	492	#define DISK_UTIL_MSEC (250)
	493
6a0106a0 JA	494	#ifndef min
	495	#define min(a, b) ((a) < (b) ? (a) : (b))
	496	#endif
a00735e6 JA	497	#ifndef max
	498	#define max(a, b) ((a) > (b) ? (a) : (b))
	499	#endif
6a0106a0	500
6796209a JA	501	/*
	502	* Log exports
	503	*/
	504	extern int read_iolog_get(struct thread_data , struct io_u );
	505	extern void write_iolog_put(struct thread_data , struct io_u );
	506	extern int init_iolog(struct thread_data *td);
	507	extern void log_io_piece(struct thread_data , struct io_u );
	508	extern void prune_io_piece_log(struct thread_data *);
	509	extern void write_iolog_close(struct thread_data *);
	510
	511	/*
	512	* Logging
	513	*/
1e97cce9 JA	514	extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
	515	extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
	516	extern void add_bw_sample(struct thread_data , enum fio_ddir, struct timeval );
6796209a JA	517	extern void show_run_stats(void);
	518	extern void init_disk_util(struct thread_data *);
	519	extern void update_rusage_stat(struct thread_data *);
	520	extern void update_io_ticks(void);
	521	extern void disk_util_timer_arm(void);
8914a9d8 JA	522	extern void setup_log(struct io_log **);
8914a9d8 JA	523	extern void finish_log(struct thread_data , struct io_log , const char *);
bb3884d8	524	extern void __finish_log(struct io_log , const char );
8914a9d8	525	extern int setup_rate(struct thread_data *);
bb3884d8 JA	526	extern struct io_log *agg_io_log[2];
	527	extern int write_bw_log;
	528	extern void add_agg_sample(unsigned long, enum fio_ddir);
6796209a JA	529
	530	/*
	531	* Time functions
	532	*/
	533	extern unsigned long utime_since(struct timeval , struct timeval );
69008999	534	extern unsigned long utime_since_now(struct timeval *);
6796209a JA	535	extern unsigned long mtime_since(struct timeval , struct timeval );
	536	extern unsigned long mtime_since_now(struct timeval *);
	537	extern unsigned long time_since_now(struct timeval *);
263e529f	538	extern unsigned long mtime_since_genesis(void);
b990b5c0	539	extern void __usec_sleep(unsigned int);
6796209a	540	extern void usec_sleep(struct thread_data *, unsigned long);
a00735e6	541	extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
6043c579	542	extern void fill_start_time(struct timeval *);
02bcaa8c	543	extern void fio_gettime(struct timeval , void );
6796209a	544
8914a9d8 JA	545	/*
	546	* Init functions
	547	*/
	548	extern int parse_options(int, char **);
	549	extern int init_random_state(struct thread_data *);
	550
53cdc686 JA	551	/*
	552	* File setup/shutdown
	553	*/
	554	extern void close_files(struct thread_data *);
	555	extern int setup_files(struct thread_data *);
21972cde	556	extern int open_files(struct thread_data *);
e5b401d4	557	extern int file_invalidate_cache(struct thread_data , struct fio_file );
53cdc686	558
263e529f JA	559	/*
	560	* ETA/status stuff
	561	*/
	562	extern void print_thread_status(void);
	563	extern void print_status_init(int);
	564
	565	/*
	566	* Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
	567	* will never back again. It may cycle between running/verififying/fsyncing.
	568	* Once the thread reaches TD_EXITED, it is just waiting for the core to
	569	* reap it.
	570	*/
	571	enum {
	572	TD_NOT_CREATED = 0,
	573	TD_CREATED,
	574	TD_INITIALIZED,
	575	TD_RUNNING,
	576	TD_VERIFYING,
	577	TD_FSYNCING,
	578	TD_EXITED,
	579	TD_REAPED,
	580	};
	581
e29d1b70 JA	582	/*
	583	* Verify helpers
	584	*/
	585	extern void populate_verify_io_u(struct thread_data , struct io_u );
	586	extern int get_next_verify(struct thread_data td, struct io_u );
a9619d44	587	extern int do_io_u_verify(struct thread_data , struct io_u *);
e29d1b70	588
2f9ade3c JA	589	/*
	590	* Memory helpers
	591	*/
	592	extern int fio_pin_memory(void);
	593	extern void fio_unpin_memory(void);
	594	extern int allocate_io_mem(struct thread_data *);
	595	extern void free_io_mem(struct thread_data *);
	596
10ba535a JA	597	/*
	598	* io unit handling
	599	*/
	600	#define queue_full(td) list_empty(&(td)->io_u_freelist)
	601	extern struct io_u __get_io_u(struct thread_data );
	602	extern struct io_u get_io_u(struct thread_data , struct fio_file *);
	603	extern void put_io_u(struct thread_data , struct io_u );
	604	extern void ios_completed(struct thread_data , struct io_completion_data );
	605	extern void io_completed(struct thread_data , struct io_u , struct io_completion_data *);
	606
	607	/*
	608	* io engine entry points
	609	*/
8c16d840	610	extern int td_io_init(struct thread_data *);
10ba535a JA	611	extern int td_io_prep(struct thread_data , struct io_u );
	612	extern int td_io_queue(struct thread_data , struct io_u );
	613	extern int td_io_sync(struct thread_data , struct fio_file );
	614	extern int td_io_getevents(struct thread_data , int, int, struct timespec );
	615
bbfd6b00 JA	616	/*
	617	* This is a pretty crappy semaphore implementation, but with the use that fio
	618	* has (just signalling start/go conditions), it doesn't have to be better.
	619	* Naturally this would not work for any type of contended semaphore or
	620	* for real locking.
	621	*/
1056eaad	622	static inline void fio_sem_init(volatile int *sem, int val)
bbfd6b00 JA	623	{
	624	*sem = val;
	625	}
	626
1056eaad	627	static inline void fio_sem_down(volatile int *sem)
bbfd6b00 JA	628	{
	629	while (*sem == 0)
	630	usleep(10000);
	631
	632	(*sem)--;
	633	}
	634
1056eaad	635	static inline void fio_sem_up(volatile int *sem)
bbfd6b00 JA	636	{
	637	(*sem)++;
	638	}
	639
3b70d7e5 JA	640	/*
	641	* If logging output to a file, stderr should go to both stderr and f_err
	642	*/
	643	#define log_err(args...) do { \
	644	fprintf(f_err, ##args); \
	645	if (f_err != stderr) \
	646	fprintf(stderr, ##args); \
	647	} while (0)
	648
2866c82d	649	struct ioengine_ops {
5f350952	650	struct list_head list;
2866c82d JA	651	char name[16];
	652	int version;
	653	int flags;
ea2877a4	654	int (setup)(struct thread_data );
2866c82d JA	655	int (init)(struct thread_data );
	656	int (prep)(struct thread_data , struct io_u *);
	657	int (queue)(struct thread_data , struct io_u *);
	658	int (getevents)(struct thread_data , int, int, struct timespec *);
	659	struct io_u (event)(struct thread_data *, int);
	660	int (cancel)(struct thread_data , struct io_u *);
	661	void (cleanup)(struct thread_data );
2866c82d JA	662	void *data;
	663	void *dlhandle;
	664	};
	665
87dc1ab1	666	#define FIO_IOOPS_VERSION 3
2866c82d	667
b4692828	668	extern struct ioengine_ops load_ioengine(struct thread_data , const char *);
5f350952 JA	669	extern int register_ioengine(struct ioengine_ops *);
5f350952 JA	670	extern void unregister_ioengine(struct ioengine_ops *);
2866c82d JA	671	extern void close_ioengine(struct thread_data *);
	672
	673	/*
	674	* Mark unused variables passed to ops functions as unused, to silence gcc
	675	*/
	676	#define fio_unused __attribute((__unused__))
5f350952 JA	677	#define fio_init __attribute__((constructor))
5f350952 JA	678	#define fio_exit __attribute__((destructor))
2866c82d	679
34572e28 JA	680	#define for_each_td(td, i) \
34572e28 JA	681	for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686	682	#define for_each_file(td, f, i) \
34572e28	683	for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
53cdc686	684
0032bf9f JA	685	#define fio_assert(td, cond) do { \
0032bf9f JA	686	if (!(cond)) { \
340fd243	687	int *__foo = NULL; \
0032bf9f JA	688	fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
0032bf9f JA	689	(td)->runstate = TD_EXITED; \
437c9b71	690	(td)->error = EFAULT; \
340fd243	691	*__foo = 0; \
0032bf9f JA	692	} \
	693	} while (0)
	694
ebac4655	695	#endif