Home | History | Annotate | Download | only in fio
      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 #include <unistd.h>
     13 #include <string.h>
     14 #include <inttypes.h>
     15 #include <assert.h>
     16 
     17 #include "compiler/compiler.h"
     18 #include "thread_options.h"
     19 #include "flist.h"
     20 #include "fifo.h"
     21 #include "arch/arch.h"
     22 #include "os/os.h"
     23 #include "mutex.h"
     24 #include "log.h"
     25 #include "debug.h"
     26 #include "file.h"
     27 #include "io_ddir.h"
     28 #include "ioengine.h"
     29 #include "iolog.h"
     30 #include "helpers.h"
     31 #include "options.h"
     32 #include "profile.h"
     33 #include "fio_time.h"
     34 #include "gettime.h"
     35 #include "lib/getopt.h"
     36 #include "lib/rand.h"
     37 #include "lib/rbtree.h"
     38 #include "client.h"
     39 #include "server.h"
     40 #include "stat.h"
     41 #include "flow.h"
     42 #include "io_u_queue.h"
     43 
     44 #ifdef CONFIG_SOLARISAIO
     45 #include <sys/asynch.h>
     46 #endif
     47 
     48 #ifdef CONFIG_LIBNUMA
     49 #include <linux/mempolicy.h>
     50 #include <numa.h>
     51 
     52 /*
     53  * "local" is pseudo-policy
     54  */
     55 #define MPOL_LOCAL MPOL_MAX
     56 #endif
     57 
     58 /*
     59  * offset generator types
     60  */
     61 enum {
     62 	RW_SEQ_SEQ	= 0,
     63 	RW_SEQ_IDENT,
     64 };
     65 
     66 enum {
     67 	TD_F_VER_BACKLOG	= 1,
     68 	TD_F_TRIM_BACKLOG	= 2,
     69 	TD_F_READ_IOLOG		= 4,
     70 	TD_F_REFILL_BUFFERS	= 8,
     71 	TD_F_SCRAMBLE_BUFFERS	= 16,
     72 	TD_F_VER_NONE		= 32,
     73 	TD_F_PROFILE_OPS	= 64,
     74 	TD_F_COMPRESS		= 128,
     75 	TD_F_NOIO		= 256,
     76 	TD_F_COMPRESS_LOG	= 512,
     77 	TD_F_VSTATE_SAVED	= 1024,
     78 };
     79 
     80 enum {
     81 	FIO_RAND_BS_OFF		= 0,
     82 	FIO_RAND_VER_OFF,
     83 	FIO_RAND_MIX_OFF,
     84 	FIO_RAND_FILE_OFF,
     85 	FIO_RAND_BLOCK_OFF,
     86 	FIO_RAND_FILE_SIZE_OFF,
     87 	FIO_RAND_TRIM_OFF,
     88 	FIO_RAND_BUF_OFF,
     89 	FIO_RAND_SEQ_RAND_READ_OFF,
     90 	FIO_RAND_SEQ_RAND_WRITE_OFF,
     91 	FIO_RAND_SEQ_RAND_TRIM_OFF,
     92 	FIO_RAND_START_DELAY,
     93 	FIO_DEDUPE_OFF,
     94 	FIO_RAND_NR_OFFS,
     95 };
     96 
     97 /*
     98  * This describes a single thread/process executing a fio job.
     99  */
    100 struct thread_data {
    101 	struct thread_options o;
    102 	unsigned long flags;
    103 	void *eo;
    104 	char verror[FIO_VERROR_SIZE];
    105 	pthread_t thread;
    106 	unsigned int thread_number;
    107 	unsigned int subjob_number;
    108 	unsigned int groupid;
    109 	struct thread_stat ts;
    110 
    111 	int client_type;
    112 
    113 	struct io_log *slat_log;
    114 	struct io_log *clat_log;
    115 	struct io_log *lat_log;
    116 	struct io_log *bw_log;
    117 	struct io_log *iops_log;
    118 
    119 	struct tp_data *tp_data;
    120 
    121 	uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
    122 	struct timeval bw_sample_time;
    123 
    124 	uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
    125 	struct timeval iops_sample_time;
    126 
    127 	/*
    128 	 * Tracks the last iodepth number of completed writes, if data
    129 	 * verification is enabled
    130 	 */
    131 	uint64_t *last_write_comp;
    132 	unsigned int last_write_idx;
    133 
    134 	volatile int update_rusage;
    135 	struct fio_mutex *rusage_sem;
    136 	struct rusage ru_start;
    137 	struct rusage ru_end;
    138 
    139 	struct fio_file **files;
    140 	unsigned char *file_locks;
    141 	unsigned int files_size;
    142 	unsigned int files_index;
    143 	unsigned int nr_open_files;
    144 	unsigned int nr_done_files;
    145 	unsigned int nr_normal_files;
    146 	union {
    147 		unsigned int next_file;
    148 		struct frand_state next_file_state;
    149 	};
    150 	int error;
    151 	int sig;
    152 	int done;
    153 	int stop_io;
    154 	pid_t pid;
    155 	char *orig_buffer;
    156 	size_t orig_buffer_size;
    157 	volatile int terminate;
    158 	volatile int runstate;
    159 	unsigned int last_was_sync;
    160 	enum fio_ddir last_ddir;
    161 
    162 	int mmapfd;
    163 
    164 	void *iolog_buf;
    165 	FILE *iolog_f;
    166 
    167 	char *sysfs_root;
    168 
    169 	unsigned long rand_seeds[FIO_RAND_NR_OFFS];
    170 
    171 	struct frand_state bsrange_state;
    172 	struct frand_state verify_state;
    173 	struct frand_state trim_state;
    174 	struct frand_state delay_state;
    175 
    176 	struct frand_state buf_state;
    177 	struct frand_state buf_state_prev;
    178 	struct frand_state dedupe_state;
    179 
    180 	unsigned int verify_batch;
    181 	unsigned int trim_batch;
    182 
    183 	struct thread_io_list *vstate;
    184 
    185 	int shm_id;
    186 
    187 	/*
    188 	 * IO engine hooks, contains everything needed to submit an io_u
    189 	 * to any of the available IO engines.
    190 	 */
    191 	struct ioengine_ops *io_ops;
    192 
    193 	/*
    194 	 * Queue depth of io_u's that fio MIGHT do
    195 	 */
    196 	unsigned int cur_depth;
    197 
    198 	/*
    199 	 * io_u's about to be committed
    200 	 */
    201 	unsigned int io_u_queued;
    202 
    203 	/*
    204 	 * io_u's submitted but not completed yet
    205 	 */
    206 	unsigned int io_u_in_flight;
    207 
    208 	/*
    209 	 * List of free and busy io_u's
    210 	 */
    211 	struct io_u_ring io_u_requeues;
    212 	struct io_u_queue io_u_freelist;
    213 	struct io_u_queue io_u_all;
    214 	pthread_mutex_t io_u_lock;
    215 	pthread_cond_t free_cond;
    216 
    217 	/*
    218 	 * async verify offload
    219 	 */
    220 	struct flist_head verify_list;
    221 	pthread_t *verify_threads;
    222 	unsigned int nr_verify_threads;
    223 	pthread_cond_t verify_cond;
    224 	int verify_thread_exit;
    225 
    226 	/*
    227 	 * Rate state
    228 	 */
    229 	uint64_t rate_bps[DDIR_RWDIR_CNT];
    230 	long rate_pending_usleep[DDIR_RWDIR_CNT];
    231 	unsigned long rate_bytes[DDIR_RWDIR_CNT];
    232 	unsigned long rate_blocks[DDIR_RWDIR_CNT];
    233 	struct timeval lastrate[DDIR_RWDIR_CNT];
    234 
    235 	uint64_t total_io_size;
    236 	uint64_t fill_device_size;
    237 
    238 	/*
    239 	 * Issue side
    240 	 */
    241 	uint64_t io_issues[DDIR_RWDIR_CNT];
    242 	uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
    243 	uint64_t loops;
    244 
    245 	/*
    246 	 * Completions
    247 	 */
    248 	uint64_t io_blocks[DDIR_RWDIR_CNT];
    249 	uint64_t this_io_blocks[DDIR_RWDIR_CNT];
    250 	uint64_t io_bytes[DDIR_RWDIR_CNT];
    251 	uint64_t io_skip_bytes;
    252 	uint64_t this_io_bytes[DDIR_RWDIR_CNT];
    253 	uint64_t zone_bytes;
    254 	struct fio_mutex *mutex;
    255 
    256 	/*
    257 	 * State for random io, a bitmap of blocks done vs not done
    258 	 */
    259 	struct frand_state random_state;
    260 
    261 	struct timeval start;	/* start of this loop */
    262 	struct timeval epoch;	/* time job was started */
    263 	struct timeval last_issue;
    264 	long time_offset;
    265 	struct timeval tv_cache;
    266 	struct timeval terminate_time;
    267 	unsigned int tv_cache_nr;
    268 	unsigned int tv_cache_mask;
    269 	unsigned int ramp_time_over;
    270 
    271 	/*
    272 	 * Time since last latency_window was started
    273 	 */
    274 	struct timeval latency_ts;
    275 	unsigned int latency_qd;
    276 	unsigned int latency_qd_high;
    277 	unsigned int latency_qd_low;
    278 	unsigned int latency_failed;
    279 	uint64_t latency_ios;
    280 	int latency_end_run;
    281 
    282 	/*
    283 	 * read/write mixed workload state
    284 	 */
    285 	struct frand_state rwmix_state;
    286 	unsigned long rwmix_issues;
    287 	enum fio_ddir rwmix_ddir;
    288 	unsigned int ddir_seq_nr;
    289 
    290 	/*
    291 	 * rand/seq mixed workload state
    292 	 */
    293 	struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
    294 
    295 	/*
    296 	 * IO history logs for verification. We use a tree for sorting,
    297 	 * if we are overwriting. Otherwise just use a fifo.
    298 	 */
    299 	struct rb_root io_hist_tree;
    300 	struct flist_head io_hist_list;
    301 	unsigned long io_hist_len;
    302 
    303 	/*
    304 	 * For IO replaying
    305 	 */
    306 	struct flist_head io_log_list;
    307 
    308 	/*
    309 	 * For tracking/handling discards
    310 	 */
    311 	struct flist_head trim_list;
    312 	unsigned long trim_entries;
    313 
    314 	struct flist_head next_rand_list;
    315 
    316 	/*
    317 	 * for fileservice, how often to switch to a new file
    318 	 */
    319 	unsigned int file_service_nr;
    320 	unsigned int file_service_left;
    321 	struct fio_file *file_service_file;
    322 
    323 	unsigned int sync_file_range_nr;
    324 
    325 	/*
    326 	 * For generating file sizes
    327 	 */
    328 	struct frand_state file_size_state;
    329 
    330 	/*
    331 	 * Error counts
    332 	 */
    333 	unsigned int total_err_count;
    334 	int first_error;
    335 
    336 	struct fio_flow *flow;
    337 
    338 	/*
    339 	 * Can be overloaded by profiles
    340 	 */
    341 	struct prof_io_ops prof_io_ops;
    342 	void *prof_data;
    343 
    344 	void *pinned_mem;
    345 };
    346 
    347 /*
    348  * when should interactive ETA output be generated
    349  */
    350 enum {
    351 	FIO_ETA_AUTO,
    352 	FIO_ETA_ALWAYS,
    353 	FIO_ETA_NEVER,
    354 };
    355 
    356 #define __td_verror(td, err, msg, func)					\
    357 	do {								\
    358 		unsigned int ____e = (err);				\
    359 		if ((td)->error)					\
    360 			break;						\
    361 		(td)->error = ____e;					\
    362 		if (!(td)->first_error)					\
    363 			snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg));		\
    364 	} while (0)
    365 
    366 
    367 #define td_clear_error(td)		\
    368 	(td)->error = 0;
    369 #define td_verror(td, err, func)	\
    370 	__td_verror((td), (err), strerror((err)), (func))
    371 #define td_vmsg(td, err, msg, func)	\
    372 	__td_verror((td), (err), (msg), (func))
    373 
    374 #define __fio_stringify_1(x)	#x
    375 #define __fio_stringify(x)	__fio_stringify_1(x)
    376 
    377 extern int exitall_on_terminate;
    378 extern unsigned int thread_number;
    379 extern unsigned int stat_number;
    380 extern int shm_id;
    381 extern int groupid;
    382 extern int output_format;
    383 extern int append_terse_output;
    384 extern int temp_stall_ts;
    385 extern uintptr_t page_mask, page_size;
    386 extern int read_only;
    387 extern int eta_print;
    388 extern int eta_new_line;
    389 extern unsigned long done_secs;
    390 extern char *job_section;
    391 extern int fio_gtod_offload;
    392 extern int fio_gtod_cpu;
    393 extern enum fio_cs fio_clock_source;
    394 extern int fio_clock_source_set;
    395 extern int warnings_fatal;
    396 extern int terse_version;
    397 extern int is_backend;
    398 extern int nr_clients;
    399 extern int log_syslog;
    400 extern int status_interval;
    401 extern const char fio_version_string[];
    402 extern int helper_do_stat;
    403 extern pthread_cond_t helper_cond;
    404 extern char *trigger_file;
    405 extern char *trigger_cmd;
    406 extern char *trigger_remote_cmd;
    407 extern long long trigger_timeout;
    408 
    409 extern struct thread_data *threads;
    410 
    411 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
    412 {
    413 	assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
    414 }
    415 
    416 #define REAL_MAX_JOBS		2048
    417 
    418 static inline int should_fsync(struct thread_data *td)
    419 {
    420 	if (td->last_was_sync)
    421 		return 0;
    422 	if (td_write(td) || td_rw(td) || td->o.override_sync)
    423 		return 1;
    424 
    425 	return 0;
    426 }
    427 
    428 /*
    429  * Init/option functions
    430  */
    431 extern int __must_check fio_init_options(void);
    432 extern int __must_check parse_options(int, char **);
    433 extern int parse_jobs_ini(char *, int, int, int);
    434 extern int parse_cmd_line(int, char **, int);
    435 extern int fio_backend(void);
    436 extern void reset_fio_state(void);
    437 extern void clear_io_state(struct thread_data *);
    438 extern int fio_options_parse(struct thread_data *, char **, int, int);
    439 extern void fio_keywords_init(void);
    440 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
    441 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
    442 extern void fio_fill_default_options(struct thread_data *);
    443 extern int fio_show_option_help(const char *);
    444 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
    445 extern void fio_options_dup_and_init(struct option *);
    446 extern void fio_options_mem_dupe(struct thread_data *);
    447 extern void options_mem_dupe(void *data, struct fio_option *options);
    448 extern void td_fill_rand_seeds(struct thread_data *);
    449 extern void add_job_opts(const char **, int);
    450 extern char *num2str(uint64_t, int, int, int, int);
    451 extern int ioengine_load(struct thread_data *);
    452 extern int parse_dryrun(void);
    453 extern int fio_running_or_pending_io_threads(void);
    454 extern int fio_set_fd_nonblocking(int, const char *);
    455 
    456 extern uintptr_t page_mask;
    457 extern uintptr_t page_size;
    458 extern int initialize_fio(char *envp[]);
    459 
    460 #define FIO_GETOPT_JOB		0x89000000
    461 #define FIO_GETOPT_IOENGINE	0x98000000
    462 #define FIO_NR_OPTIONS		(FIO_MAX_OPTS + 128)
    463 
    464 /*
    465  * ETA/status stuff
    466  */
    467 extern void print_thread_status(void);
    468 extern void print_status_init(int);
    469 extern char *fio_uint_to_kmg(unsigned int val);
    470 
    471 /*
    472  * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
    473  * will never back again. It may cycle between running/verififying/fsyncing.
    474  * Once the thread reaches TD_EXITED, it is just waiting for the core to
    475  * reap it.
    476  */
    477 enum {
    478 	TD_NOT_CREATED = 0,
    479 	TD_CREATED,
    480 	TD_INITIALIZED,
    481 	TD_RAMP,
    482 	TD_SETTING_UP,
    483 	TD_RUNNING,
    484 	TD_PRE_READING,
    485 	TD_VERIFYING,
    486 	TD_FSYNCING,
    487 	TD_FINISHING,
    488 	TD_EXITED,
    489 	TD_REAPED,
    490 };
    491 
    492 extern void td_set_runstate(struct thread_data *, int);
    493 extern int td_bump_runstate(struct thread_data *, int);
    494 extern void td_restore_runstate(struct thread_data *, int);
    495 
    496 /*
    497  * Allow 60 seconds for a job to quit on its own, otherwise reap with
    498  * a vengeance.
    499  */
    500 #define FIO_REAP_TIMEOUT	60
    501 
    502 #define TERMINATE_ALL		(-1)
    503 extern void fio_terminate_threads(int);
    504 extern void fio_mark_td_terminate(struct thread_data *);
    505 
    506 /*
    507  * Memory helpers
    508  */
    509 extern int __must_check fio_pin_memory(struct thread_data *);
    510 extern void fio_unpin_memory(struct thread_data *);
    511 extern int __must_check allocate_io_mem(struct thread_data *);
    512 extern void free_io_mem(struct thread_data *);
    513 extern void free_threads_shm(void);
    514 
    515 /*
    516  * Reset stats after ramp time completes
    517  */
    518 extern void reset_all_stats(struct thread_data *);
    519 
    520 /*
    521  * blktrace support
    522  */
    523 #ifdef FIO_HAVE_BLKTRACE
    524 extern int is_blktrace(const char *, int *);
    525 extern int load_blktrace(struct thread_data *, const char *, int);
    526 #endif
    527 
    528 /*
    529  * Latency target helpers
    530  */
    531 extern void lat_target_check(struct thread_data *);
    532 extern void lat_target_init(struct thread_data *);
    533 extern void lat_target_reset(struct thread_data *);
    534 
    535 #define for_each_td(td, i)	\
    536 	for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
    537 #define for_each_file(td, f, i)	\
    538 	if ((td)->files_index)						\
    539 		for ((i) = 0, (f) = (td)->files[0];			\
    540 	    	 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
    541 		 (i)++)
    542 
    543 #define fio_assert(td, cond)	do {	\
    544 	if (!(cond)) {			\
    545 		int *__foo = NULL;	\
    546 		fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond);	\
    547 		td_set_runstate((td), TD_EXITED);	\
    548 		(td)->error = EFAULT;		\
    549 		*__foo = 0;			\
    550 	}	\
    551 } while (0)
    552 
    553 static inline int fio_fill_issue_time(struct thread_data *td)
    554 {
    555 	if (td->o.read_iolog_file ||
    556 	    !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
    557 		return 1;
    558 
    559 	return 0;
    560 }
    561 
    562 static inline int __should_check_rate(struct thread_data *td,
    563 				      enum fio_ddir ddir)
    564 {
    565 	struct thread_options *o = &td->o;
    566 
    567 	/*
    568 	 * If some rate setting was given, we need to check it
    569 	 */
    570 	if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
    571 	    o->rate_iops_min[ddir])
    572 		return 1;
    573 
    574 	return 0;
    575 }
    576 
    577 static inline int should_check_rate(struct thread_data *td,
    578 				    uint64_t *bytes_done)
    579 {
    580 	int ret = 0;
    581 
    582 	if (bytes_done[DDIR_READ])
    583 		ret |= __should_check_rate(td, DDIR_READ);
    584 	if (bytes_done[DDIR_WRITE])
    585 		ret |= __should_check_rate(td, DDIR_WRITE);
    586 	if (bytes_done[DDIR_TRIM])
    587 		ret |= __should_check_rate(td, DDIR_TRIM);
    588 
    589 	return ret;
    590 }
    591 
    592 static inline unsigned int td_max_bs(struct thread_data *td)
    593 {
    594 	unsigned int max_bs;
    595 
    596 	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
    597 	return max(td->o.max_bs[DDIR_TRIM], max_bs);
    598 }
    599 
    600 static inline unsigned int td_min_bs(struct thread_data *td)
    601 {
    602 	unsigned int min_bs;
    603 
    604 	min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
    605 	return min(td->o.min_bs[DDIR_TRIM], min_bs);
    606 }
    607 
    608 static inline int is_power_of_2(uint64_t val)
    609 {
    610 	return (val != 0 && ((val & (val - 1)) == 0));
    611 }
    612 
    613 /*
    614  * We currently only need to do locking if we have verifier threads
    615  * accessing our internal structures too
    616  */
    617 static inline void td_io_u_lock(struct thread_data *td)
    618 {
    619 	if (td->o.verify_async)
    620 		pthread_mutex_lock(&td->io_u_lock);
    621 }
    622 
    623 static inline void td_io_u_unlock(struct thread_data *td)
    624 {
    625 	if (td->o.verify_async)
    626 		pthread_mutex_unlock(&td->io_u_lock);
    627 }
    628 
    629 static inline void td_io_u_free_notify(struct thread_data *td)
    630 {
    631 	if (td->o.verify_async)
    632 		pthread_cond_signal(&td->free_cond);
    633 }
    634 
    635 extern const char *fio_get_arch_string(int);
    636 extern const char *fio_get_os_string(int);
    637 
    638 #ifdef FIO_INTERNAL
    639 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
    640 #endif
    641 
    642 enum {
    643 	FIO_OUTPUT_TERSE	= 0,
    644 	FIO_OUTPUT_JSON,
    645 	FIO_OUTPUT_NORMAL,
    646 };
    647 
    648 enum {
    649 	FIO_RAND_DIST_RANDOM	= 0,
    650 	FIO_RAND_DIST_ZIPF,
    651 	FIO_RAND_DIST_PARETO,
    652 };
    653 
    654 #define FIO_DEF_ZIPF		1.1
    655 #define FIO_DEF_PARETO		0.2
    656 
    657 enum {
    658 	FIO_RAND_GEN_TAUSWORTHE = 0,
    659 	FIO_RAND_GEN_LFSR,
    660 };
    661 
    662 enum {
    663 	FIO_CPUS_SHARED		= 0,
    664 	FIO_CPUS_SPLIT,
    665 };
    666 
    667 extern void exec_trigger(const char *);
    668 extern void check_trigger_file(void);
    669 
    670 #endif
    671