8 #include <sys/resource.h>
24 #ifdef FIO_HAVE_SYSLET
39 TD_DDIR_READ = 1 << 0,
40 TD_DDIR_WRITE = 1 << 1,
41 TD_DDIR_RAND = 1 << 2,
42 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
43 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
44 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
45 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
49 * Use for maintaining statistics
52 unsigned long max_val;
53 unsigned long min_val;
54 unsigned long samples;
61 * A single data sample
70 * Dynamically growing data sample log
73 unsigned long nr_samples;
74 unsigned long max_samples;
75 struct io_sample *log;
79 * When logging io actions, this matches a single sent io_u
83 struct rb_node rb_node;
84 struct list_head list;
86 struct fio_file *file;
87 unsigned long long offset;
92 #ifdef FIO_HAVE_SYSLET
94 struct syslet_uatom atom; /* the atom to submit */
95 struct syslet_uatom *head; /* head of the sequence */
96 long ret; /* syscall return value */
101 IO_U_F_FREE = 1 << 0,
102 IO_U_F_FLIGHT = 1 << 1,
112 #ifdef FIO_HAVE_LIBAIO
115 #ifdef FIO_HAVE_POSIXAIO
119 struct sg_io_hdr hdr;
121 #ifdef FIO_HAVE_SYSLET
122 struct syslet_req req;
124 #ifdef FIO_HAVE_GUASI
128 struct timeval start_time;
129 struct timeval issue_time;
132 * Allocated/set buffer and length
135 unsigned long buflen;
136 unsigned long long offset;
139 * IO engine state, may be different from above when we get
140 * partial transfers / residual data counts
143 unsigned long xfer_buflen;
151 * io engine private data
160 struct fio_file *file;
162 struct list_head list;
165 * Callback for io completion
167 int (*end_io)(struct thread_data *, struct io_u *);
171 * io_ops->queue() return values
174 FIO_Q_COMPLETED = 0, /* completed sync */
175 FIO_Q_QUEUED = 1, /* queued, will complete async */
176 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
179 #define FIO_HDR_MAGIC 0xf00baaef
182 VERIFY_NONE = 0, /* no verification */
183 VERIFY_MD5, /* md5 sum data blocks */
184 VERIFY_CRC32, /* crc32 sum data blocks */
185 VERIFY_NULL, /* pretend to verify */
189 * A header structure associated with each checksummed data block
191 struct verify_header {
192 unsigned int fio_magic;
194 unsigned int verify_type;
196 char md5_digest[MD5_HASH_WORDS * 4];
201 struct group_run_stats {
202 unsigned long long max_run[2], min_run[2];
203 unsigned long long max_bw[2], min_bw[2];
204 unsigned long long io_kb[2];
205 unsigned long long agg[2];
209 * What type of allocation to use for io buffers
212 MEM_MALLOC = 0, /* ordinary malloc */
213 MEM_SHM, /* use shared memory segments */
214 MEM_SHMHUGE, /* use shared memory segments with huge pages */
215 MEM_MMAP, /* use anonynomous mmap */
216 MEM_MMAPHUGE, /* memory mapped huge file */
220 * The type of object we are working on
223 FIO_TYPE_FILE = 1, /* plain file */
224 FIO_TYPE_BD, /* block device */
225 FIO_TYPE_CHAR, /* character device */
228 enum fio_ioengine_flags {
229 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
230 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
231 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
232 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
233 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
236 enum fio_file_flags {
237 FIO_FILE_OPEN = 1 << 0, /* file is open */
238 FIO_FILE_UNLINK = 1 << 1, /* unlink on close */
239 FIO_FILE_CLOSING = 1 << 2, /* file being closed */
240 FIO_FILE_EXISTS = 1 << 3, /* no need to create */
244 * Each thread_data structure has a number of files associated with it,
245 * this structure holds state information for a single file.
248 enum fio_filetype filetype;
251 * A file may not be a file descriptor, let the io engine decide
254 unsigned long file_data;
259 unsigned long long file_size;
260 unsigned long long real_file_size;
261 unsigned long long file_offset;
262 unsigned long long last_pos;
263 unsigned long long last_completed_pos;
266 * block map for random io
268 unsigned long *file_map;
269 unsigned int num_maps;
270 unsigned int last_free_lookup;
273 enum fio_file_flags flags;
277 * How many depth levels to log
279 #define FIO_IO_U_MAP_NR 8
280 #define FIO_IO_U_LAT_NR 12
291 struct io_log *slat_log;
292 struct io_log *clat_log;
293 struct io_log *bw_log;
296 * bandwidth and latency stats
298 struct io_stat clat_stat[2]; /* completion latency */
299 struct io_stat slat_stat[2]; /* submission latency */
300 struct io_stat bw_stat[2]; /* bandwidth stats */
302 unsigned long long stat_io_bytes[2];
303 struct timeval stat_sample_time[2];
306 * fio system usage accounting
308 struct rusage ru_start;
309 struct rusage ru_end;
310 unsigned long usr_time;
311 unsigned long sys_time;
315 * IO depth and latency stats
317 unsigned int io_u_map[FIO_IO_U_MAP_NR];
318 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
319 unsigned long total_io_u[2];
321 unsigned long long io_bytes[2];
322 unsigned long runtime[2];
323 unsigned long total_run_time;
326 struct thread_options {
334 enum td_ddir td_ddir;
335 unsigned int ddir_nr;
336 unsigned int iodepth;
337 unsigned int iodepth_low;
338 unsigned int iodepth_batch;
340 unsigned long long size;
341 unsigned long long file_size_low;
342 unsigned long long file_size_high;
343 unsigned long long start_offset;
346 unsigned int min_bs[2];
347 unsigned int max_bs[2];
349 unsigned int nr_files;
350 unsigned int open_files;
352 unsigned int odirect;
353 unsigned int invalidate_cache;
354 unsigned int create_serialize;
355 unsigned int create_fsync;
356 unsigned int end_fsync;
357 unsigned int sync_io;
359 unsigned int use_thread;
361 unsigned int do_disk_util;
362 unsigned int override_sync;
363 unsigned int rand_repeatable;
364 unsigned int write_lat_log;
365 unsigned int write_bw_log;
366 unsigned int norandommap;
367 unsigned int bs_unaligned;
368 unsigned int fsync_on_close;
370 unsigned int hugepage_size;
371 unsigned int rw_min_bs;
372 unsigned int thinktime;
373 unsigned int thinktime_spin;
374 unsigned int thinktime_blocks;
375 unsigned int fsync_blocks;
376 unsigned int start_delay;
377 unsigned long timeout;
378 unsigned int overwrite;
379 unsigned int bw_avg_time;
381 unsigned long long zone_size;
382 unsigned long long zone_skip;
383 enum fio_memtype mem_type;
385 unsigned int stonewall;
386 unsigned int new_group;
387 unsigned int numjobs;
388 os_cpu_mask_t cpumask;
390 unsigned int read_iolog;
391 unsigned int rwmixcycle;
392 unsigned int rwmix[2];
394 unsigned int file_service_type;
395 unsigned int group_reporting;
396 unsigned int fadvise_hint;
398 char *read_iolog_file;
399 char *write_iolog_file;
402 * Pre-run and post-run shell
408 unsigned int ratemin;
409 unsigned int ratecycle;
410 unsigned int rate_iops;
411 unsigned int rate_iops_min;
416 * CPU "io" cycle burner
418 unsigned int cpuload;
419 unsigned int cpucycle;
422 #define FIO_VERROR_SIZE 128
425 * This describes a single thread/process executing a fio job.
428 struct thread_options o;
429 char verror[FIO_VERROR_SIZE];
433 struct thread_stat ts;
434 struct fio_file *files;
435 unsigned int files_index;
436 unsigned int nr_open_files;
437 unsigned int nr_normal_files;
439 unsigned int next_file;
440 os_random_state_t next_file_state;
445 size_t orig_buffer_size;
446 volatile int terminate;
447 volatile int runstate;
449 unsigned int last_was_sync;
459 os_random_state_t bsrange_state;
460 os_random_state_t verify_state;
465 * IO engine hooks, contains everything needed to submit an io_u
466 * to any of the available IO engines.
468 struct ioengine_ops *io_ops;
471 * Current IO depth and list of free and busy io_u's.
473 unsigned int cur_depth;
474 struct list_head io_u_freelist;
475 struct list_head io_u_busylist;
476 struct list_head io_u_requeues;
477 unsigned int io_u_queued;
482 unsigned long rate_usec_cycle;
483 long rate_pending_usleep;
484 unsigned long rate_bytes;
485 unsigned long rate_blocks;
486 struct timeval lastrate;
488 unsigned long long io_size;
489 unsigned long long total_io_size;
491 unsigned long io_issues[2];
492 unsigned long long io_blocks[2];
493 unsigned long long io_bytes[2];
494 unsigned long long this_io_bytes[2];
495 unsigned long long zone_bytes;
496 struct fio_sem *mutex;
499 * State for random io, a bitmap of blocks done vs not done
501 os_random_state_t random_state;
503 struct timeval start; /* start of this loop */
504 struct timeval epoch; /* time job was started */
505 struct timeval rw_end[2];
506 unsigned int rw_end_set[2];
509 * read/write mixed workload state
511 os_random_state_t rwmix_state;
512 unsigned long long rwmix_bytes;
513 struct timeval rwmix_switch;
514 enum fio_ddir rwmix_ddir;
515 unsigned int ddir_nr;
518 * IO history logs for verification. We use a tree for sorting,
519 * if we are overwriting. Otherwise just use a fifo.
521 struct rb_root io_hist_tree;
522 struct list_head io_hist_list;
527 struct list_head io_log_list;
532 struct timeval timeout_end;
533 struct itimerval timer;
536 * for fileservice, how often to switch to a new file
538 unsigned int file_service_nr;
539 unsigned int file_service_left;
540 struct fio_file *file_service_file;
543 * For generating file sizes
545 os_random_state_t file_size_state;
549 * roundrobin available files, or choose one at random.
552 FIO_FSERVICE_RANDOM = 1,
557 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
558 * the timer on each queue operation.
560 #define IO_U_TIMEOUT_INC 5
561 #define IO_U_TIMEOUT 30
563 #define __td_verror(td, err, msg, func) \
569 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
573 #define td_verror(td, err, func) \
574 __td_verror((td), (err), strerror((err)), (func))
575 #define td_vmsg(td, err, msg, func) \
576 __td_verror((td), (err), (msg), (func))
578 extern int exitall_on_terminate;
579 extern int thread_number;
580 extern int nr_process, nr_thread;
583 extern int terse_output;
586 extern int temp_stall_ts;
587 extern unsigned long long mlock_size;
588 extern unsigned long page_mask, page_size;
590 extern struct thread_data *threads;
592 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
593 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
594 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
595 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
597 #define BLOCKS_PER_MAP (8 * sizeof(long))
598 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
599 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
600 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
602 #define MAX_JOBS (1024)
604 static inline int should_fsync(struct thread_data *td)
606 if (td->last_was_sync)
610 if (td_write(td) || td_rw(td) || td->o.override_sync)
617 * Disk utils as read in /sys/block/<dev>/stat
619 struct disk_util_stat {
622 unsigned long long sectors[2];
625 unsigned time_in_queue;
629 * Per-device disk util management
632 struct list_head list;
638 struct disk_util_stat dus;
639 struct disk_util_stat last_dus;
645 #define DISK_UTIL_MSEC (250)
648 #define min(a, b) ((a) < (b) ? (a) : (b))
651 #define max(a, b) ((a) > (b) ? (a) : (b))
657 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
658 extern void write_iolog_put(struct thread_data *, struct io_u *);
659 extern int __must_check init_iolog(struct thread_data *td);
660 extern void log_io_piece(struct thread_data *, struct io_u *);
661 extern void prune_io_piece_log(struct thread_data *);
662 extern void write_iolog_close(struct thread_data *);
667 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
668 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
669 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
670 extern void show_run_stats(void);
671 extern void init_disk_util(struct thread_data *);
672 extern void update_rusage_stat(struct thread_data *);
673 extern void update_io_ticks(void);
674 extern void disk_util_timer_arm(void);
675 extern void setup_log(struct io_log **);
676 extern void finish_log(struct thread_data *, struct io_log *, const char *);
677 extern void __finish_log(struct io_log *, const char *);
678 extern struct io_log *agg_io_log[2];
679 extern int write_bw_log;
680 extern void add_agg_sample(unsigned long, enum fio_ddir);
685 extern unsigned long utime_since(struct timeval *, struct timeval *);
686 extern unsigned long utime_since_now(struct timeval *);
687 extern unsigned long mtime_since(struct timeval *, struct timeval *);
688 extern unsigned long mtime_since_now(struct timeval *);
689 extern unsigned long time_since_now(struct timeval *);
690 extern unsigned long mtime_since_genesis(void);
691 extern void __usec_sleep(unsigned int);
692 extern void usec_sleep(struct thread_data *, unsigned long);
693 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
694 extern void fill_start_time(struct timeval *);
695 extern void fio_gettime(struct timeval *, void *);
696 extern void set_genesis_time(void);
699 * Init/option functions
701 extern int __must_check parse_options(int, char **);
702 extern int fio_option_parse(struct thread_data *, const char *);
703 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
704 extern void fio_fill_default_options(struct thread_data *);
705 extern int fio_show_option_help(const char *);
706 extern void fio_options_dup_and_init(struct option *);
707 extern void options_mem_dupe(struct thread_data *);
708 extern void options_mem_free(struct thread_data *);
709 #define FIO_GETOPT_JOB 0x89988998
710 #define FIO_NR_OPTIONS 128
713 * File setup/shutdown
715 extern void close_files(struct thread_data *);
716 extern int __must_check setup_files(struct thread_data *);
717 extern int __must_check open_files(struct thread_data *);
718 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
719 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
720 extern void generic_close_file(struct thread_data *, struct fio_file *);
721 extern void add_file(struct thread_data *, const char *);
722 extern void get_file(struct fio_file *);
723 extern void put_file(struct thread_data *, struct fio_file *);
724 extern int add_dir_files(struct thread_data *, const char *);
725 extern int init_random_map(struct thread_data *);
726 extern void dup_files(struct thread_data *, struct thread_data *);
731 extern void print_thread_status(void);
732 extern void print_status_init(int);
735 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
736 * will never back again. It may cycle between running/verififying/fsyncing.
737 * Once the thread reaches TD_EXITED, it is just waiting for the core to
754 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
755 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
756 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
761 extern int __must_check fio_pin_memory(void);
762 extern void fio_unpin_memory(void);
763 extern int __must_check allocate_io_mem(struct thread_data *);
764 extern void free_io_mem(struct thread_data *);
769 #define queue_full(td) list_empty(&(td)->io_u_freelist)
770 extern struct io_u *__get_io_u(struct thread_data *);
771 extern struct io_u *get_io_u(struct thread_data *);
772 extern void put_io_u(struct thread_data *, struct io_u *);
773 extern void requeue_io_u(struct thread_data *, struct io_u **);
774 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
775 extern long __must_check io_u_queued_complete(struct thread_data *, int);
776 extern void io_u_queued(struct thread_data *, struct io_u *);
777 extern void io_u_log_error(struct thread_data *, struct io_u *);
778 extern void io_u_init_timeout(void);
779 extern void io_u_set_timeout(struct thread_data *);
780 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
783 * io engine entry points
785 extern int __must_check td_io_init(struct thread_data *);
786 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
787 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
788 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
789 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
790 extern int __must_check td_io_commit(struct thread_data *);
791 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
792 extern void td_io_close_file(struct thread_data *, struct fio_file *);
795 * If logging output to a file, stderr should go to both stderr and f_err
797 #define log_err(args...) do { \
798 fprintf(f_err, ##args); \
799 if (f_err != stderr) \
800 fprintf(stderr, ##args); \
803 #define log_info(args...) fprintf(f_out, ##args)
805 FILE *get_f_out(void);
806 FILE *get_f_err(void);
808 struct ioengine_ops {
809 struct list_head list;
813 int (*setup)(struct thread_data *);
814 int (*init)(struct thread_data *);
815 int (*prep)(struct thread_data *, struct io_u *);
816 int (*queue)(struct thread_data *, struct io_u *);
817 int (*commit)(struct thread_data *);
818 int (*getevents)(struct thread_data *, int, int, struct timespec *);
819 struct io_u *(*event)(struct thread_data *, int);
820 int (*cancel)(struct thread_data *, struct io_u *);
821 void (*cleanup)(struct thread_data *);
822 int (*open_file)(struct thread_data *, struct fio_file *);
823 void (*close_file)(struct thread_data *, struct fio_file *);
828 #define FIO_IOOPS_VERSION 6
830 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
831 extern void register_ioengine(struct ioengine_ops *);
832 extern void unregister_ioengine(struct ioengine_ops *);
833 extern void close_ioengine(struct thread_data *);
836 * Mark unused variables passed to ops functions as unused, to silence gcc
838 #define fio_unused __attribute((__unused__))
839 #define fio_init __attribute__((constructor))
840 #define fio_exit __attribute__((destructor))
842 #define for_each_td(td, i) \
843 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
844 #define for_each_file(td, f, i) \
845 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
847 #define fio_assert(td, cond) do { \
850 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
851 (td)->runstate = TD_EXITED; \
852 (td)->error = EFAULT; \