[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"

#ifdef FIO_HAVE_SYSLET
#include "syslet.h"
#endif

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
b6959b55	21	#ifdef FIO_HAVE_SYSLET
a4f4fdd7	22	#include "syslet.h"
b6959b55	23	#endif
a4f4fdd7	24
1e97cce9 JA	25	enum fio_ddir {
	26	DDIR_READ = 0,
	27	DDIR_WRITE,
	28	DDIR_SYNC,
	29	};
	30
1c9e06ee JA	31	/*
	32	* Use for maintaining statistics
	33	*/
ebac4655	34	struct io_stat {
ebac4655 JA	35	unsigned long max_val;
	36	unsigned long min_val;
	37	unsigned long samples;
68704084 JA	38
	39	double mean;
	40	double S;
ebac4655 JA	41	};
ebac4655 JA	42
1c9e06ee JA	43	/*
	44	* A single data sample
	45	*/
ebac4655 JA	46	struct io_sample {
	47	unsigned long time;
	48	unsigned long val;
1e97cce9	49	enum fio_ddir ddir;
ebac4655 JA	50	};
ebac4655 JA	51
1c9e06ee JA	52	/*
	53	* Dynamically growing data sample log
	54	*/
ebac4655 JA	55	struct io_log {
	56	unsigned long nr_samples;
	57	unsigned long max_samples;
	58	struct io_sample *log;
	59	};
	60
1c9e06ee JA	61	/*
	62	* When logging io actions, this matches a single sent io_u
	63	*/
ebac4655 JA	64	struct io_piece {
ebac4655 JA	65	struct list_head list;
53cdc686	66	struct fio_file *file;
ebac4655	67	unsigned long long offset;
a4f4fdd7	68	unsigned long len;
1e97cce9	69	enum fio_ddir ddir;
ebac4655 JA	70	};
ebac4655 JA	71
a4f4fdd7 JA	72	#ifdef FIO_HAVE_SYSLET
	73	struct syslet_req {
	74	struct syslet_uatom atom;
a4f4fdd7 JA	75	long ret;
	76	};
	77	#endif
	78
ebac4655 JA	79	/*
	80	* The io unit
	81	*/
	82	struct io_u {
	83	union {
	84	#ifdef FIO_HAVE_LIBAIO
	85	struct iocb iocb;
	86	#endif
	87	#ifdef FIO_HAVE_POSIXAIO
	88	struct aiocb aiocb;
	89	#endif
	90	#ifdef FIO_HAVE_SGIO
	91	struct sg_io_hdr hdr;
a4f4fdd7 JA	92	#endif
a4f4fdd7 JA	93	#ifdef FIO_HAVE_SYSLET
a2e1b08a	94	struct syslet_req req;
ebac4655 JA	95	#endif
	96	};
	97	struct timeval start_time;
	98	struct timeval issue_time;
	99
1c9e06ee JA	100	/*
	101	* Allocated/set buffer and length
	102	*/
6b9cea23	103	void *buf;
a4f4fdd7	104	unsigned long buflen;
ebac4655 JA	105	unsigned long long offset;
ebac4655 JA	106
1c9e06ee JA	107	/*
	108	* IO engine state, may be different from above when we get
	109	* partial transfers / residual data counts
	110	*/
cec6b55d	111	void *xfer_buf;
a4f4fdd7	112	unsigned long xfer_buflen;
cec6b55d	113
ebac4655 JA	114	unsigned int resid;
	115	unsigned int error;
	116
1e97cce9	117	enum fio_ddir ddir;
ebac4655	118
dfd7bc2c JA	119	/*
	120	* io engine private data
	121	*/
	122	union {
	123	unsigned int index;
	124	unsigned int seen;
	125	};
	126
53cdc686 JA	127	struct fio_file *file;
53cdc686 JA	128
ebac4655 JA	129	struct list_head list;
	130	};
	131
	132	#define FIO_HDR_MAGIC 0xf00baaef
	133
	134	enum {
1c9e06ee JA	135	VERIFY_NONE = 0, /* no verification */
	136	VERIFY_MD5, /* md5 sum data blocks */
	137	VERIFY_CRC32, /* crc32 sum data blocks */
ebac4655 JA	138	};
ebac4655 JA	139
1c9e06ee JA	140	/*
	141	* A header structure associated with each checksummed data block
	142	*/
ebac4655 JA	143	struct verify_header {
	144	unsigned int fio_magic;
	145	unsigned int len;
	146	unsigned int verify_type;
	147	union {
	148	char md5_digest[MD5_HASH_WORDS * 4];
	149	unsigned long crc32;
	150	};
	151	};
	152
	153	struct group_run_stats {
9104f874 JA	154	unsigned long long max_run[2], min_run[2];
9104f874 JA	155	unsigned long long max_bw[2], min_bw[2];
e9b2a3fa	156	unsigned long long io_kb[2];
9104f874	157	unsigned long long agg[2];
ebac4655 JA	158	};
ebac4655 JA	159
e9c047a0 JA	160	/*
	161	* What type of allocation to use for io buffers
	162	*/
	163	enum fio_memtype {
	164	MEM_MALLOC = 0, /* ordinary malloc */
	165	MEM_SHM, /* use shared memory segments */
74b025b0	166	MEM_SHMHUGE, /* use shared memory segments with huge pages */
e9c047a0	167	MEM_MMAP, /* use anonynomous mmap */
d0bdaf49	168	MEM_MMAPHUGE, /* memory mapped huge file */
e9c047a0 JA	169	};
	170
	171	/*
	172	* The type of object we are working on
	173	*/
	174	enum fio_filetype {
1c9e06ee JA	175	FIO_TYPE_FILE = 1, /* plain file */
	176	FIO_TYPE_BD, /* block device */
	177	FIO_TYPE_CHAR, /* character device */
e9c047a0 JA	178	};
e9c047a0 JA	179
2866c82d	180	enum fio_ioengine_flags {
1c9e06ee JA	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) */
e9c047a0 JA	187	};
e9c047a0 JA	188
1c9e06ee JA	189	/*
	190	* Each thread_data structure has a number of files associated with it,
	191	* this structure holds state information for a single file.
	192	*/
53cdc686 JA	193	struct fio_file {
	194	/*
	195	* A file may not be a file descriptor, let the io engine decide
	196	*/
	197	union {
	198	unsigned long file_data;
a4f4fdd7	199	long fd;
53cdc686 JA	200	};
	201	char *file_name;
	202	void *mmap;
	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;
02bcaa8c	207	unsigned long long last_completed_pos;
53cdc686	208
1c9e06ee JA	209	/*
	210	* block map for random io
	211	*/
53cdc686 JA	212	unsigned long *file_map;
53cdc686 JA	213	unsigned int num_maps;
1c9e06ee	214	unsigned int last_free_lookup;
9b031dc8 JA	215
9b031dc8 JA	216	unsigned int unlink;
53cdc686 JA	217	};
53cdc686 JA	218
71619dc2 JA	219	/*
	220	* How many depth levels to log
	221	*/
	222	#define FIO_IO_U_MAP_NR 8
	223
e9c047a0 JA	224	/*
	225	* This describes a single thread/process executing a fio job.
	226	*/
ebac4655	227	struct thread_data {
61697c37	228	char *description;
b4692828	229	char *name;
ef899b63	230	char *directory;
13f8e2d2	231	char *filename;
ebac4655 JA	232	char verror[80];
	233	pthread_t thread;
	234	int thread_number;
	235	int groupid;
e9c047a0	236	enum fio_filetype filetype;
53cdc686 JA	237	struct fio_file *files;
53cdc686 JA	238	unsigned int nr_files;
13f8e2d2	239	unsigned int nr_uniq_files;
53cdc686	240	unsigned int next_file;
ebac4655	241	int error;
ebac4655 JA	242	pid_t pid;
	243	char *orig_buffer;
	244	size_t orig_buffer_size;
	245	volatile int terminate;
	246	volatile int runstate;
e9c047a0	247	enum fio_ddir ddir;
3d60d1ed	248	unsigned int iomix;
ebac4655	249	unsigned int ioprio;
87dc1ab1	250	unsigned int last_was_sync;
e9c047a0	251
9158d2f7 JA	252	unsigned int sequential;
	253	unsigned int odirect;
	254	unsigned int invalidate_cache;
	255	unsigned int create_serialize;
	256	unsigned int create_fsync;
	257	unsigned int end_fsync;
	258	unsigned int sync_io;
	259	unsigned int verify;
	260	unsigned int use_thread;
	261	unsigned int unlink;
	262	unsigned int do_disk_util;
	263	unsigned int override_sync;
	264	unsigned int rand_repeatable;
	265	unsigned int write_lat_log;
	266	unsigned int write_bw_log;
bb8895e0	267	unsigned int norandommap;
690adba3	268	unsigned int bs_unaligned;
e9c047a0	269
a00735e6 JA	270	unsigned int bs[2];
	271	unsigned int min_bs[2];
	272	unsigned int max_bs[2];
56bb17f2	273	unsigned int hugepage_size;
a00735e6	274	unsigned int rw_min_bs;
ebac4655	275	unsigned int thinktime;
9c1f7434	276	unsigned int thinktime_blocks;
ebac4655 JA	277	unsigned int fsync_blocks;
ebac4655 JA	278	unsigned int start_delay;
906c8d75	279	unsigned long timeout;
ebac4655	280	unsigned int overwrite;
ebac4655	281	unsigned int bw_avg_time;
ebac4655	282	unsigned int loops;
20dc95c4 JA	283	unsigned long long zone_size;
20dc95c4 JA	284	unsigned long long zone_skip;
e9c047a0	285	enum fio_memtype mem_type;
313cb206 JA	286	char *mmapfile;
313cb206 JA	287	int mmapfd;
ebac4655 JA	288	unsigned int stonewall;
ebac4655 JA	289	unsigned int numjobs;
ebac4655 JA	290	unsigned int iodepth;
ebac4655 JA	291	os_cpu_mask_t cpumask;
aea47d44	292	unsigned int iolog;
843a7413	293	unsigned int read_iolog;
a6ccc7be JA	294	unsigned int rwmixcycle;
a6ccc7be JA	295	unsigned int rwmixread;
e1f36503	296	unsigned int rwmixwrite;
b6f4d880	297	unsigned int nice;
aea47d44	298
076efc7c JA	299	char *read_iolog_file;
076efc7c JA	300	char *write_iolog_file;
843a7413 JA	301	void *iolog_buf;
843a7413 JA	302	FILE *iolog_f;
ebac4655	303
da86774e	304	char *sysfs_root;
da86774e JA	305	char *ioscheduler;
da86774e JA	306
6dfd46b9 JA	307	os_random_state_t bsrange_state;
6dfd46b9 JA	308	os_random_state_t verify_state;
ebac4655 JA	309
	310	int shm_id;
	311
e9c047a0 JA	312	/*
	313	* IO engine hooks, contains everything needed to submit an io_u
	314	* to any of the available IO engines.
	315	*/
2866c82d	316	struct ioengine_ops *io_ops;
ebac4655	317
e9c047a0 JA	318	/*
	319	* Current IO depth and list of free and busy io_u's.
	320	*/
ebac4655	321	unsigned int cur_depth;
71619dc2 JA	322	unsigned int io_u_map[FIO_IO_U_MAP_NR];
71619dc2 JA	323	unsigned long total_io_u;
ebac4655 JA	324	struct list_head io_u_freelist;
	325	struct list_head io_u_busylist;
	326
e9c047a0 JA	327	/*
	328	* Rate state
	329	*/
ebac4655 JA	330	unsigned int rate;
	331	unsigned int ratemin;
	332	unsigned int ratecycle;
	333	unsigned long rate_usec_cycle;
	334	long rate_pending_usleep;
	335	unsigned long rate_bytes;
	336	struct timeval lastrate;
	337
	338	unsigned long runtime[2]; /* msec */
	339	unsigned long long io_size;
53cdc686 JA	340	unsigned long long total_file_size;
53cdc686 JA	341	unsigned long long start_offset;
ebac4655 JA	342	unsigned long long total_io_size;
ebac4655 JA	343
9104f874 JA	344	unsigned long long io_blocks[2];
	345	unsigned long long io_bytes[2];
	346	unsigned long long zone_bytes;
	347	unsigned long long this_io_bytes[2];
bbfd6b00	348	volatile int mutex;
ebac4655	349
e9c047a0 JA	350	/*
	351	* State for random io, a bitmap of blocks done vs not done
	352	*/
6dfd46b9	353	os_random_state_t random_state;
ebac4655	354
b990b5c0 JA	355	/*
	356	* CPU "io" cycle burner
	357	*/
	358	unsigned int cpuload;
	359	unsigned int cpucycle;
	360
ebac4655 JA	361	/*
	362	* bandwidth and latency stats
	363	*/
	364	struct io_stat clat_stat[2]; /* completion latency */
	365	struct io_stat slat_stat[2]; /* submission latency */
	366	struct io_stat bw_stat[2]; /* bandwidth stats */
	367
9104f874	368	unsigned long long stat_io_bytes[2];
ebac4655 JA	369	struct timeval stat_sample_time[2];
	370
	371	struct io_log *slat_log;
	372	struct io_log *clat_log;
	373	struct io_log *bw_log;
	374
	375	struct timeval start; /* start of this loop */
	376	struct timeval epoch; /* time job was started */
69008999	377	struct timeval end_time;/* time job ended */
ebac4655	378
e9c047a0 JA	379	/*
	380	* fio system usage accounting
	381	*/
ebac4655 JA	382	struct rusage ru_start;
	383	struct rusage ru_end;
	384	unsigned long usr_time;
	385	unsigned long sys_time;
	386	unsigned long ctx;
	387
e9c047a0 JA	388	/*
	389	* read/write mixed workload state
	390	*/
6dfd46b9	391	os_random_state_t rwmix_state;
a6ccc7be	392	struct timeval rwmix_switch;
e9c047a0	393	enum fio_ddir rwmix_ddir;
a6ccc7be	394
4e0ba8af JA	395	/*
	396	* Pre-run and post-run shell
	397	*/
	398	char *exec_prerun;
	399	char *exec_postrun;
	400
e9c047a0 JA	401	/*
	402	* IO historic logs
	403	*/
ebac4655	404	struct list_head io_hist_list;
aea47d44	405	struct list_head io_log_list;
ebac4655 JA	406	};
ebac4655 JA	407
b990b5c0	408	#define __td_verror(td, err, msg) \
ebac4655 JA	409	do { \
	410	int e = (err); \
	411	(td)->error = e; \
b990b5c0	412	snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, error=%s", __FILE__, __LINE__, (msg)); \
ebac4655 JA	413	} while (0)
ebac4655 JA	414
b990b5c0 JA	415
	416	#define td_verror(td, err) __td_verror((td), (err), strerror((err)))
	417	#define td_vmsg(td, err, msg) __td_verror((td), (err), (msg))
	418
ebac4655 JA	419	extern int exitall_on_terminate;
	420	extern int thread_number;
	421	extern int shm_id;
	422	extern int groupid;
c6ae0a5b	423	extern int terse_output;
eb8bbf48 JA	424	extern FILE *f_out;
eb8bbf48 JA	425	extern FILE *f_err;
53cdc686	426	extern int temp_stall_ts;
1e97cce9	427	extern unsigned long long mlock_size;
ebac4655 JA	428
	429	extern struct thread_data *threads;
	430
ebac4655 JA	431	#define td_read(td) ((td)->ddir == DDIR_READ)
ebac4655 JA	432	#define td_write(td) ((td)->ddir == DDIR_WRITE)
3d60d1ed	433	#define td_rw(td) ((td)->iomix != 0)
ebac4655 JA	434
ebac4655 JA	435	#define BLOCKS_PER_MAP (8 * sizeof(long))
a00735e6	436	#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
53cdc686 JA	437	#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
53cdc686 JA	438	#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
ebac4655 JA	439
	440	#define MAX_JOBS (1024)
	441
87dc1ab1 JA	442	static inline int should_fsync(struct thread_data *td)
	443	{
	444	if (td->last_was_sync)
	445	return 0;
	446	if (td->odirect)
	447	return 0;
	448	if (td_write(td) \|\| td_rw(td) \|\| td->override_sync)
	449	return 1;
	450
	451	return 0;
	452	}
	453
1c9e06ee JA	454	/*
	455	* Disk utils as read in /sys/block/<dev>/stat
	456	*/
ebac4655 JA	457	struct disk_util_stat {
	458	unsigned ios[2];
	459	unsigned merges[2];
	460	unsigned long long sectors[2];
	461	unsigned ticks[2];
	462	unsigned io_ticks;
	463	unsigned time_in_queue;
	464	};
	465
1c9e06ee JA	466	/*
	467	* Per-device disk util management
	468	*/
ebac4655 JA	469	struct disk_util {
	470	struct list_head list;
	471
	472	char *name;
	473	char path[256];
	474	dev_t dev;
	475
	476	struct disk_util_stat dus;
	477	struct disk_util_stat last_dus;
	478
	479	unsigned long msec;
	480	struct timeval time;
	481	};
	482
1c9e06ee JA	483	/*
	484	* Used for passing io_u completion data
	485	*/
ebac4655 JA	486	struct io_completion_data {
	487	int nr; /* input */
	488
	489	int error; /* output */
	490	unsigned long bytes_done[2]; /* output */
02bcaa8c	491	struct timeval time; /* output */
ebac4655 JA	492	};
	493
	494	#define DISK_UTIL_MSEC (250)
	495
6a0106a0 JA	496	#ifndef min
	497	#define min(a, b) ((a) < (b) ? (a) : (b))
	498	#endif
a00735e6 JA	499	#ifndef max
	500	#define max(a, b) ((a) > (b) ? (a) : (b))
	501	#endif
6a0106a0	502
6796209a JA	503	/*
	504	* Log exports
	505	*/
	506	extern int read_iolog_get(struct thread_data , struct io_u );
	507	extern void write_iolog_put(struct thread_data , struct io_u );
	508	extern int init_iolog(struct thread_data *td);
	509	extern void log_io_piece(struct thread_data , struct io_u );
	510	extern void prune_io_piece_log(struct thread_data *);
	511	extern void write_iolog_close(struct thread_data *);
	512
	513	/*
	514	* Logging
	515	*/
1e97cce9 JA	516	extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
	517	extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
	518	extern void add_bw_sample(struct thread_data , enum fio_ddir, struct timeval );
6796209a JA	519	extern void show_run_stats(void);
	520	extern void init_disk_util(struct thread_data *);
	521	extern void update_rusage_stat(struct thread_data *);
	522	extern void update_io_ticks(void);
	523	extern void disk_util_timer_arm(void);
8914a9d8 JA	524	extern void setup_log(struct io_log **);
8914a9d8 JA	525	extern void finish_log(struct thread_data , struct io_log , const char *);
bb3884d8	526	extern void __finish_log(struct io_log , const char );
8914a9d8	527	extern int setup_rate(struct thread_data *);
bb3884d8 JA	528	extern struct io_log *agg_io_log[2];
	529	extern int write_bw_log;
	530	extern void add_agg_sample(unsigned long, enum fio_ddir);
6796209a JA	531
	532	/*
	533	* Time functions
	534	*/
	535	extern unsigned long utime_since(struct timeval , struct timeval );
69008999	536	extern unsigned long utime_since_now(struct timeval *);
6796209a JA	537	extern unsigned long mtime_since(struct timeval , struct timeval );
	538	extern unsigned long mtime_since_now(struct timeval *);
	539	extern unsigned long time_since_now(struct timeval *);
263e529f	540	extern unsigned long mtime_since_genesis(void);
b990b5c0	541	extern void __usec_sleep(unsigned int);
6796209a	542	extern void usec_sleep(struct thread_data *, unsigned long);
a00735e6	543	extern void rate_throttle(struct thread_data *, unsigned long, unsigned int, int);
6043c579	544	extern void fill_start_time(struct timeval *);
02bcaa8c	545	extern void fio_gettime(struct timeval , void );
6796209a	546
8914a9d8 JA	547	/*
	548	* Init functions
	549	*/
	550	extern int parse_options(int, char **);
	551	extern int init_random_state(struct thread_data *);
	552
53cdc686 JA	553	/*
	554	* File setup/shutdown
	555	*/
	556	extern void close_files(struct thread_data *);
	557	extern int setup_files(struct thread_data *);
21972cde	558	extern int open_files(struct thread_data *);
e5b401d4	559	extern int file_invalidate_cache(struct thread_data , struct fio_file );
53cdc686	560
263e529f JA	561	/*
	562	* ETA/status stuff
	563	*/
	564	extern void print_thread_status(void);
	565	extern void print_status_init(int);
	566
	567	/*
	568	* Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
	569	* will never back again. It may cycle between running/verififying/fsyncing.
	570	* Once the thread reaches TD_EXITED, it is just waiting for the core to
	571	* reap it.
	572	*/
	573	enum {
	574	TD_NOT_CREATED = 0,
	575	TD_CREATED,
	576	TD_INITIALIZED,
	577	TD_RUNNING,
	578	TD_VERIFYING,
	579	TD_FSYNCING,
	580	TD_EXITED,
	581	TD_REAPED,
	582	};
	583
e29d1b70 JA	584	/*
	585	* Verify helpers
	586	*/
	587	extern void populate_verify_io_u(struct thread_data , struct io_u );
	588	extern int get_next_verify(struct thread_data td, struct io_u );
a9619d44	589	extern int do_io_u_verify(struct thread_data , struct io_u *);
e29d1b70	590
2f9ade3c JA	591	/*
	592	* Memory helpers
	593	*/
	594	extern int fio_pin_memory(void);
	595	extern void fio_unpin_memory(void);
	596	extern int allocate_io_mem(struct thread_data *);
	597	extern void free_io_mem(struct thread_data *);
	598
10ba535a JA	599	/*
	600	* io unit handling
	601	*/
	602	#define queue_full(td) list_empty(&(td)->io_u_freelist)
	603	extern struct io_u __get_io_u(struct thread_data );
	604	extern struct io_u get_io_u(struct thread_data , struct fio_file *);
	605	extern void put_io_u(struct thread_data , struct io_u );
	606	extern void ios_completed(struct thread_data , struct io_completion_data );
	607	extern void io_completed(struct thread_data , struct io_u , struct io_completion_data *);
	608
	609	/*
	610	* io engine entry points
	611	*/
8c16d840	612	extern int td_io_init(struct thread_data *);
10ba535a JA	613	extern int td_io_prep(struct thread_data , struct io_u );
	614	extern int td_io_queue(struct thread_data , struct io_u );
	615	extern int td_io_sync(struct thread_data , struct fio_file );
	616	extern int td_io_getevents(struct thread_data , int, int, struct timespec );
	617
bbfd6b00 JA	618	/*
	619	* This is a pretty crappy semaphore implementation, but with the use that fio
	620	* has (just signalling start/go conditions), it doesn't have to be better.
	621	* Naturally this would not work for any type of contended semaphore or
	622	* for real locking.
	623	*/
1056eaad	624	static inline void fio_sem_init(volatile int *sem, int val)
bbfd6b00 JA	625	{
	626	*sem = val;
	627	}
	628
1056eaad	629	static inline void fio_sem_down(volatile int *sem)
bbfd6b00 JA	630	{
	631	while (*sem == 0)
	632	usleep(10000);
	633
	634	(*sem)--;
	635	}
	636
1056eaad	637	static inline void fio_sem_up(volatile int *sem)
bbfd6b00 JA	638	{
	639	(*sem)++;
	640	}
	641
3b70d7e5 JA	642	/*
	643	* If logging output to a file, stderr should go to both stderr and f_err
	644	*/
	645	#define log_err(args...) do { \
	646	fprintf(f_err, ##args); \
	647	if (f_err != stderr) \
	648	fprintf(stderr, ##args); \
	649	} while (0)
	650
2866c82d	651	struct ioengine_ops {
5f350952	652	struct list_head list;
2866c82d JA	653	char name[16];
	654	int version;
	655	int flags;
ea2877a4	656	int (setup)(struct thread_data );
2866c82d JA	657	int (init)(struct thread_data );
	658	int (prep)(struct thread_data , struct io_u *);
	659	int (queue)(struct thread_data , struct io_u *);
	660	int (getevents)(struct thread_data , int, int, struct timespec *);
	661	struct io_u (event)(struct thread_data *, int);
	662	int (cancel)(struct thread_data , struct io_u *);
	663	void (cleanup)(struct thread_data );
2866c82d JA	664	void *data;
	665	void *dlhandle;
	666	};
	667
87dc1ab1	668	#define FIO_IOOPS_VERSION 3
2866c82d	669
b4692828	670	extern struct ioengine_ops load_ioengine(struct thread_data , const char *);
5f350952 JA	671	extern int register_ioengine(struct ioengine_ops *);
5f350952 JA	672	extern void unregister_ioengine(struct ioengine_ops *);
2866c82d JA	673	extern void close_ioengine(struct thread_data *);
	674
	675	/*
	676	* Mark unused variables passed to ops functions as unused, to silence gcc
	677	*/
	678	#define fio_unused __attribute((__unused__))
5f350952 JA	679	#define fio_init __attribute__((constructor))
5f350952 JA	680	#define fio_exit __attribute__((destructor))
2866c82d	681
34572e28 JA	682	#define for_each_td(td, i) \
34572e28 JA	683	for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686	684	#define for_each_file(td, f, i) \
34572e28	685	for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
53cdc686	686
0032bf9f JA	687	#define fio_assert(td, cond) do { \
0032bf9f JA	688	if (!(cond)) { \
340fd243	689	int *__foo = NULL; \
0032bf9f JA	690	fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
0032bf9f JA	691	(td)->runstate = TD_EXITED; \
437c9b71	692	(td)->error = EFAULT; \
340fd243	693	*__foo = 0; \
0032bf9f JA	694	} \
	695	} while (0)
	696
ebac4655	697	#endif