8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
21 #include "arch/arch.h"
25 #ifdef FIO_HAVE_SYSLET
41 TD_DDIR_READ = 1 << 0,
42 TD_DDIR_WRITE = 1 << 1,
43 TD_DDIR_RAND = 1 << 2,
44 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
45 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
46 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
47 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
51 * Use for maintaining statistics
54 unsigned long max_val;
55 unsigned long min_val;
56 unsigned long samples;
63 * A single data sample
72 * Dynamically growing data sample log
75 unsigned long nr_samples;
76 unsigned long max_samples;
77 struct io_sample *log;
81 * When logging io actions, this matches a single sent io_u
85 struct rb_node rb_node;
86 struct list_head list;
90 struct fio_file *file;
92 unsigned long long offset;
97 unsigned int file_action;
101 #ifdef FIO_HAVE_SYSLET
103 struct syslet_uatom atom; /* the atom to submit */
104 struct syslet_uatom *head; /* head of the sequence */
105 long ret; /* syscall return value */
110 IO_U_F_FREE = 1 << 0,
111 IO_U_F_FLIGHT = 1 << 1,
121 #ifdef FIO_HAVE_LIBAIO
124 #ifdef FIO_HAVE_POSIXAIO
128 struct sg_io_hdr hdr;
130 #ifdef FIO_HAVE_SYSLET
131 struct syslet_req req;
133 #ifdef FIO_HAVE_GUASI
137 struct timeval start_time;
138 struct timeval issue_time;
141 * Allocated/set buffer and length
144 unsigned long buflen;
145 unsigned long long offset;
146 unsigned long long endpos;
149 * IO engine state, may be different from above when we get
150 * partial transfers / residual data counts
153 unsigned long xfer_buflen;
161 * io engine private data
170 struct fio_file *file;
172 struct list_head list;
175 * Callback for io completion
177 int (*end_io)(struct thread_data *, struct io_u *);
180 * Called when io_u is about to be freed
182 void (*unmap)(struct thread_data *, struct io_u *);
186 * io_ops->queue() return values
189 FIO_Q_COMPLETED = 0, /* completed sync */
190 FIO_Q_QUEUED = 1, /* queued, will complete async */
191 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
194 #define FIO_HDR_MAGIC 0xf00baaef
197 VERIFY_NONE = 0, /* no verification */
198 VERIFY_MD5, /* md5 sum data blocks */
199 VERIFY_CRC64, /* crc64 sum data blocks */
200 VERIFY_CRC32, /* crc32 sum data blocks */
201 VERIFY_CRC16, /* crc16 sum data blocks */
202 VERIFY_CRC7, /* crc7 sum data blocks */
203 VERIFY_SHA256, /* sha256 sum data blocks */
204 VERIFY_SHA512, /* sha512 sum data blocks */
205 VERIFY_META, /* block_num, timestamp etc. */
206 VERIFY_PATTERN, /* verify a specific pattern */
207 VERIFY_NULL, /* pretend to verify */
211 * A header structure associated with each checksummed data block. It is
212 * followed by a checksum specific header that contains the verification
215 struct verify_header {
216 unsigned int fio_magic;
218 unsigned int verify_type;
222 uint32_t md5_digest[16];
244 unsigned char thread;
245 unsigned short numberio;
246 unsigned long time_sec;
247 unsigned long time_usec;
250 struct group_run_stats {
251 unsigned long long max_run[2], min_run[2];
252 unsigned long long max_bw[2], min_bw[2];
253 unsigned long long io_kb[2];
254 unsigned long long agg[2];
258 * What type of allocation to use for io buffers
261 MEM_MALLOC = 0, /* ordinary malloc */
262 MEM_SHM, /* use shared memory segments */
263 MEM_SHMHUGE, /* use shared memory segments with huge pages */
264 MEM_MMAP, /* use anonynomous mmap */
265 MEM_MMAPHUGE, /* memory mapped huge file */
269 * The type of object we are working on
272 FIO_TYPE_FILE = 1, /* plain file */
273 FIO_TYPE_BD, /* block device */
274 FIO_TYPE_CHAR, /* character device */
275 FIO_TYPE_PIPE, /* pipe */
278 enum fio_ioengine_flags {
279 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
280 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
281 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
282 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
283 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
284 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
287 enum fio_file_flags {
288 FIO_FILE_OPEN = 1 << 0, /* file is open */
289 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
290 FIO_FILE_EXISTS = 1 << 2, /* file there */
291 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
292 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
293 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
294 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
298 * Each thread_data structure has a number of files associated with it,
299 * this structure holds state information for a single file.
302 enum fio_filetype filetype;
305 * A file may not be a file descriptor, let the io engine decide
308 unsigned long file_data;
313 * filename and possible memory mapping
317 unsigned int major, minor;
320 * size of the file, offset into file, and io size from that offset
322 unsigned long long real_file_size;
323 unsigned long long file_offset;
324 unsigned long long io_size;
326 unsigned long long last_pos;
327 unsigned long long last_completed_pos;
330 * block map for random io
332 unsigned long *file_map;
333 unsigned int num_maps;
334 unsigned int last_free_lookup;
337 enum fio_file_flags flags;
341 * How many depth levels to log
343 #define FIO_IO_U_MAP_NR 8
344 #define FIO_IO_U_LAT_U_NR 10
345 #define FIO_IO_U_LAT_M_NR 12
356 struct io_log *slat_log;
357 struct io_log *clat_log;
358 struct io_log *bw_log;
361 * bandwidth and latency stats
363 struct io_stat clat_stat[2]; /* completion latency */
364 struct io_stat slat_stat[2]; /* submission latency */
365 struct io_stat bw_stat[2]; /* bandwidth stats */
367 unsigned long long stat_io_bytes[2];
368 struct timeval stat_sample_time[2];
371 * fio system usage accounting
373 struct rusage ru_start;
374 struct rusage ru_end;
375 unsigned long usr_time;
376 unsigned long sys_time;
380 * IO depth and latency stats
382 unsigned int io_u_map[FIO_IO_U_MAP_NR];
383 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
384 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
385 unsigned long total_io_u[2];
386 unsigned long short_io_u[2];
388 unsigned long long io_bytes[2];
389 unsigned long runtime[2];
390 unsigned long total_run_time;
393 struct thread_options {
401 enum td_ddir td_ddir;
402 unsigned int ddir_nr;
403 unsigned int iodepth;
404 unsigned int iodepth_low;
405 unsigned int iodepth_batch;
407 unsigned long long size;
408 unsigned long long file_size_low;
409 unsigned long long file_size_high;
410 unsigned long long start_offset;
413 unsigned int min_bs[2];
414 unsigned int max_bs[2];
416 unsigned int nr_files;
417 unsigned int open_files;
419 unsigned int odirect;
420 unsigned int invalidate_cache;
421 unsigned int create_serialize;
422 unsigned int create_fsync;
423 unsigned int end_fsync;
424 unsigned int sync_io;
426 unsigned int do_verify;
427 unsigned int verifysort;
428 unsigned int verify_interval;
429 unsigned int verify_offset;
430 unsigned int verify_pattern;
431 unsigned int verify_pattern_bytes;
432 unsigned int verify_fatal;
433 unsigned int use_thread;
435 unsigned int do_disk_util;
436 unsigned int override_sync;
437 unsigned int rand_repeatable;
438 unsigned int write_lat_log;
439 unsigned int write_bw_log;
440 unsigned int norandommap;
441 unsigned int bs_unaligned;
442 unsigned int fsync_on_close;
444 unsigned int hugepage_size;
445 unsigned int rw_min_bs;
446 unsigned int thinktime;
447 unsigned int thinktime_spin;
448 unsigned int thinktime_blocks;
449 unsigned int fsync_blocks;
450 unsigned int start_delay;
451 unsigned long long timeout;
452 unsigned int overwrite;
453 unsigned int bw_avg_time;
455 unsigned long long zone_size;
456 unsigned long long zone_skip;
457 enum fio_memtype mem_type;
459 unsigned int stonewall;
460 unsigned int new_group;
461 unsigned int numjobs;
462 os_cpu_mask_t cpumask;
463 unsigned int cpumask_set;
465 unsigned int rwmixcycle;
466 unsigned int rwmix[2];
468 unsigned int file_service_type;
469 unsigned int group_reporting;
470 unsigned int fadvise_hint;
471 unsigned int zero_buffers;
472 unsigned int time_based;
474 char *read_iolog_file;
475 char *write_iolog_file;
478 * Pre-run and post-run shell
484 unsigned int ratemin;
485 unsigned int ratecycle;
486 unsigned int rate_iops;
487 unsigned int rate_iops_min;
492 * CPU "io" cycle burner
494 unsigned int cpuload;
495 unsigned int cpucycle;
498 #define FIO_VERROR_SIZE 128
501 * This describes a single thread/process executing a fio job.
504 struct thread_options o;
505 char verror[FIO_VERROR_SIZE];
509 struct thread_stat ts;
510 struct fio_file *files;
511 unsigned int files_index;
512 unsigned int nr_open_files;
513 unsigned int nr_done_files;
514 unsigned int nr_normal_files;
516 unsigned int next_file;
517 os_random_state_t next_file_state;
523 size_t orig_buffer_size;
524 volatile int terminate;
525 volatile int runstate;
527 unsigned int last_was_sync;
537 os_random_state_t bsrange_state;
538 os_random_state_t verify_state;
543 * IO engine hooks, contains everything needed to submit an io_u
544 * to any of the available IO engines.
546 struct ioengine_ops *io_ops;
549 * Current IO depth and list of free and busy io_u's.
551 unsigned int cur_depth;
552 struct list_head io_u_freelist;
553 struct list_head io_u_busylist;
554 struct list_head io_u_requeues;
555 unsigned int io_u_queued;
560 unsigned long rate_usec_cycle;
561 long rate_pending_usleep;
562 unsigned long rate_bytes;
563 unsigned long rate_blocks;
564 struct timeval lastrate;
566 unsigned long long total_io_size;
568 unsigned long io_issues[2];
569 unsigned long long io_blocks[2];
570 unsigned long long io_bytes[2];
571 unsigned long long io_skip_bytes;
572 unsigned long long this_io_bytes[2];
573 unsigned long long zone_bytes;
574 struct fio_sem *mutex;
577 * State for random io, a bitmap of blocks done vs not done
579 os_random_state_t random_state;
581 struct timeval start; /* start of this loop */
582 struct timeval epoch; /* time job was started */
583 struct timeval rw_end[2];
584 struct timeval last_issue;
585 unsigned int rw_end_set[2];
588 * read/write mixed workload state
590 os_random_state_t rwmix_state;
591 unsigned long long rwmix_bytes;
592 struct timeval rwmix_switch;
593 enum fio_ddir rwmix_ddir;
594 unsigned int ddir_nr;
597 * IO history logs for verification. We use a tree for sorting,
598 * if we are overwriting. Otherwise just use a fifo.
600 struct rb_root io_hist_tree;
601 struct list_head io_hist_list;
606 struct list_head io_log_list;
611 struct timeval timeout_end;
612 struct itimerval timer;
615 * for fileservice, how often to switch to a new file
617 unsigned int file_service_nr;
618 unsigned int file_service_left;
619 struct fio_file *file_service_file;
622 * For generating file sizes
624 os_random_state_t file_size_state;
628 * roundrobin available files, or choose one at random.
631 FIO_FSERVICE_RANDOM = 1,
636 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
637 * the timer on each queue operation.
639 #define IO_U_TIMEOUT_INC 5
640 #define IO_U_TIMEOUT 30
642 #define __td_verror(td, err, msg, func) \
648 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
652 #define td_verror(td, err, func) \
653 __td_verror((td), (err), strerror((err)), (func))
654 #define td_vmsg(td, err, msg, func) \
655 __td_verror((td), (err), (msg), (func))
657 extern int exitall_on_terminate;
658 extern int thread_number;
659 extern int nr_process, nr_thread;
662 extern int terse_output;
665 extern int temp_stall_ts;
666 extern unsigned long long mlock_size;
667 extern unsigned long page_mask, page_size;
669 extern struct thread_data *threads;
671 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
672 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
673 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
674 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
676 #define BLOCKS_PER_MAP (8 * sizeof(long))
677 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
678 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
679 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
681 #define MAX_JOBS (1024)
683 static inline int should_fsync(struct thread_data *td)
685 if (td->last_was_sync)
689 if (td_write(td) || td_rw(td) || td->o.override_sync)
696 * Disk utils as read in /sys/block/<dev>/stat
698 struct disk_util_stat {
701 unsigned long long sectors[2];
704 unsigned time_in_queue;
708 * Per-device disk util management
711 struct list_head list;
718 struct disk_util_stat dus;
719 struct disk_util_stat last_dus;
725 #define DISK_UTIL_MSEC (250)
736 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
737 extern void log_io_u(struct thread_data *, struct io_u *);
738 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
739 extern int __must_check init_iolog(struct thread_data *td);
740 extern void log_io_piece(struct thread_data *, struct io_u *);
741 extern void prune_io_piece_log(struct thread_data *);
742 extern void write_iolog_close(struct thread_data *);
747 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
748 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
749 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
750 extern void show_run_stats(void);
751 extern void init_disk_util(struct thread_data *);
752 extern void update_rusage_stat(struct thread_data *);
753 extern void update_io_ticks(void);
754 extern void disk_util_timer_arm(void);
755 extern void setup_log(struct io_log **);
756 extern void finish_log(struct thread_data *, struct io_log *, const char *);
757 extern void __finish_log(struct io_log *, const char *);
758 extern struct io_log *agg_io_log[2];
759 extern int write_bw_log;
760 extern void add_agg_sample(unsigned long, enum fio_ddir);
765 extern unsigned long utime_since(struct timeval *, struct timeval *);
766 extern unsigned long utime_since_now(struct timeval *);
767 extern unsigned long mtime_since(struct timeval *, struct timeval *);
768 extern unsigned long mtime_since_now(struct timeval *);
769 extern unsigned long time_since_now(struct timeval *);
770 extern unsigned long mtime_since_genesis(void);
771 extern void __usec_sleep(unsigned int);
772 extern void usec_sleep(struct thread_data *, unsigned long);
773 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
774 extern void fill_start_time(struct timeval *);
775 extern void fio_gettime(struct timeval *, void *);
776 extern void set_genesis_time(void);
779 * Init/option functions
781 extern int __must_check parse_options(int, char **);
782 extern int fio_option_parse(struct thread_data *, const char *);
783 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
784 extern void fio_fill_default_options(struct thread_data *);
785 extern int fio_show_option_help(const char *);
786 extern void fio_options_dup_and_init(struct option *);
787 extern void options_mem_dupe(struct thread_data *);
788 extern void options_mem_free(struct thread_data *);
789 #define FIO_GETOPT_JOB 0x89988998
790 #define FIO_NR_OPTIONS 128
793 * File setup/shutdown
795 extern void close_files(struct thread_data *);
796 extern int __must_check setup_files(struct thread_data *);
797 extern int __must_check open_files(struct thread_data *);
798 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
799 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
800 extern void generic_close_file(struct thread_data *, struct fio_file *);
801 extern int add_file(struct thread_data *, const char *);
802 extern void get_file(struct fio_file *);
803 extern void put_file(struct thread_data *, struct fio_file *);
804 extern int add_dir_files(struct thread_data *, const char *);
805 extern int init_random_map(struct thread_data *);
806 extern void dup_files(struct thread_data *, struct thread_data *);
807 extern int get_fileno(struct thread_data *, const char *);
808 extern void free_release_files(struct thread_data *);
813 extern void print_thread_status(void);
814 extern void print_status_init(int);
819 #ifdef FIO_HAVE_DISK_UTIL
820 extern void show_disk_util(void);
821 extern void disk_util_timer_arm(void);
822 extern void init_disk_util(struct thread_data *);
823 extern void update_io_ticks(void);
825 #define show_disk_util()
826 #define disk_util_timer_arm()
827 #define init_disk_util(td)
828 #define update_io_ticks()
832 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
833 * will never back again. It may cycle between running/verififying/fsyncing.
834 * Once the thread reaches TD_EXITED, it is just waiting for the core to
851 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
852 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
853 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
858 extern int __must_check fio_pin_memory(void);
859 extern void fio_unpin_memory(void);
860 extern int __must_check allocate_io_mem(struct thread_data *);
861 extern void free_io_mem(struct thread_data *);
866 #define queue_full(td) list_empty(&(td)->io_u_freelist)
867 extern struct io_u *__get_io_u(struct thread_data *);
868 extern struct io_u *get_io_u(struct thread_data *);
869 extern void put_io_u(struct thread_data *, struct io_u *);
870 extern void requeue_io_u(struct thread_data *, struct io_u **);
871 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
872 extern long __must_check io_u_queued_complete(struct thread_data *, int);
873 extern void io_u_queued(struct thread_data *, struct io_u *);
874 extern void io_u_log_error(struct thread_data *, struct io_u *);
875 extern void io_u_init_timeout(void);
876 extern void io_u_set_timeout(struct thread_data *);
877 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
880 * io engine entry points
882 extern int __must_check td_io_init(struct thread_data *);
883 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
884 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
885 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
886 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
887 extern int __must_check td_io_commit(struct thread_data *);
888 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
889 extern void td_io_close_file(struct thread_data *, struct fio_file *);
894 #ifdef FIO_HAVE_BLKTRACE
895 extern int is_blktrace(const char *);
896 extern int load_blktrace(struct thread_data *, const char *);
900 * If logging output to a file, stderr should go to both stderr and f_err
902 #define log_err(args...) do { \
903 fprintf(f_err, ##args); \
904 if (f_err != stderr) \
905 fprintf(stderr, ##args); \
908 #define log_info(args...) fprintf(f_out, ##args)
910 FILE *get_f_out(void);
911 FILE *get_f_err(void);
913 struct ioengine_ops {
914 struct list_head list;
918 int (*setup)(struct thread_data *);
919 int (*init)(struct thread_data *);
920 int (*prep)(struct thread_data *, struct io_u *);
921 int (*queue)(struct thread_data *, struct io_u *);
922 int (*commit)(struct thread_data *);
923 int (*getevents)(struct thread_data *, int, int, struct timespec *);
924 struct io_u *(*event)(struct thread_data *, int);
925 int (*cancel)(struct thread_data *, struct io_u *);
926 void (*cleanup)(struct thread_data *);
927 int (*open_file)(struct thread_data *, struct fio_file *);
928 void (*close_file)(struct thread_data *, struct fio_file *);
933 #define FIO_IOOPS_VERSION 7
935 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
936 extern void register_ioengine(struct ioengine_ops *);
937 extern void unregister_ioengine(struct ioengine_ops *);
938 extern void close_ioengine(struct thread_data *);
941 * Mark unused variables passed to ops functions as unused, to silence gcc
943 #define fio_unused __attribute((__unused__))
944 #define fio_init __attribute__((constructor))
945 #define fio_exit __attribute__((destructor))
947 #define for_each_td(td, i) \
948 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
949 #define for_each_file(td, f, i) \
950 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
952 #define fio_assert(td, cond) do { \
955 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
956 (td)->runstate = TD_EXITED; \
957 (td)->error = EFAULT; \
962 static inline void clear_error(struct thread_data *td)
965 td->verror[0] = '\0';