8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
40 TD_DDIR_READ = 1 << 0,
41 TD_DDIR_WRITE = 1 << 1,
42 TD_DDIR_RAND = 1 << 2,
43 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
44 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
45 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
46 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
56 * Use for maintaining statistics
59 unsigned long max_val;
60 unsigned long min_val;
61 unsigned long samples;
68 * A single data sample
77 * Dynamically growing data sample log
80 unsigned long nr_samples;
81 unsigned long max_samples;
82 struct io_sample *log;
86 * When logging io actions, this matches a single sent io_u
90 struct rb_node rb_node;
91 struct list_head list;
95 struct fio_file *file;
97 unsigned long long offset;
102 unsigned int file_action;
107 IO_U_F_FREE = 1 << 0,
108 IO_U_F_FLIGHT = 1 << 1,
118 #ifdef FIO_HAVE_LIBAIO
121 #ifdef FIO_HAVE_POSIXAIO
125 struct sg_io_hdr hdr;
127 #ifdef FIO_HAVE_GUASI
131 struct timeval start_time;
132 struct timeval issue_time;
135 * Allocated/set buffer and length
138 unsigned long buflen;
139 unsigned long long offset;
140 unsigned long long endpos;
143 * IO engine state, may be different from above when we get
144 * partial transfers / residual data counts
147 unsigned long xfer_buflen;
155 * io engine private data
164 struct fio_file *file;
166 struct list_head list;
169 * Callback for io completion
171 int (*end_io)(struct thread_data *, struct io_u *);
175 * io_ops->queue() return values
178 FIO_Q_COMPLETED = 0, /* completed sync */
179 FIO_Q_QUEUED = 1, /* queued, will complete async */
180 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
183 #define FIO_HDR_MAGIC 0xf00baaef
186 VERIFY_NONE = 0, /* no verification */
187 VERIFY_MD5, /* md5 sum data blocks */
188 VERIFY_CRC64, /* crc64 sum data blocks */
189 VERIFY_CRC32, /* crc32 sum data blocks */
190 VERIFY_CRC16, /* crc16 sum data blocks */
191 VERIFY_CRC7, /* crc7 sum data blocks */
192 VERIFY_SHA256, /* sha256 sum data blocks */
193 VERIFY_SHA512, /* sha512 sum data blocks */
194 VERIFY_META, /* block_num, timestamp etc. */
195 VERIFY_NULL, /* pretend to verify */
199 * A header structure associated with each checksummed data block. It is
200 * followed by a checksum specific header that contains the verification
203 struct verify_header {
204 unsigned int fio_magic;
206 unsigned int verify_type;
210 uint32_t md5_digest[16];
232 unsigned char thread;
233 unsigned short numberio;
234 unsigned long time_sec;
235 unsigned long time_usec;
238 struct group_run_stats {
239 unsigned long long max_run[2], min_run[2];
240 unsigned long long max_bw[2], min_bw[2];
241 unsigned long long io_kb[2];
242 unsigned long long agg[2];
246 * What type of allocation to use for io buffers
249 MEM_MALLOC = 0, /* ordinary malloc */
250 MEM_SHM, /* use shared memory segments */
251 MEM_SHMHUGE, /* use shared memory segments with huge pages */
252 MEM_MMAP, /* use anonynomous mmap */
253 MEM_MMAPHUGE, /* memory mapped huge file */
257 * The type of object we are working on
260 FIO_TYPE_FILE = 1, /* plain file */
261 FIO_TYPE_BD, /* block device */
262 FIO_TYPE_CHAR, /* character device */
263 FIO_TYPE_PIPE, /* pipe */
266 enum fio_ioengine_flags {
267 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
268 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
269 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
270 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
271 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
272 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
273 FIO_NOIO = 1 << 6, /* thread does only pseudo IO */
274 FIO_SIGQUIT = 1 << 7, /* needs SIGQUIT to exit */
277 enum fio_file_flags {
278 FIO_FILE_OPEN = 1 << 0, /* file is open */
279 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
280 FIO_FILE_EXTEND = 1 << 2, /* needs extend */
281 FIO_FILE_DONE = 1 << 3, /* io completed to this file */
282 FIO_SIZE_KNOWN = 1 << 4, /* size has been set */
283 FIO_FILE_HASHED = 1 << 5, /* file is on hash */
287 * Each thread_data structure has a number of files associated with it,
288 * this structure holds state information for a single file.
291 struct list_head hash_list;
292 enum fio_filetype filetype;
295 * A file may not be a file descriptor, let the io engine decide
298 unsigned long file_data;
303 * filename and possible memory mapping
307 unsigned int major, minor;
310 * size of the file, offset into file, and io size from that offset
312 unsigned long long real_file_size;
313 unsigned long long file_offset;
314 unsigned long long io_size;
316 unsigned long long last_pos;
319 * if io is protected by a semaphore, this is set
321 struct fio_mutex *lock;
323 unsigned int lock_batch;
324 enum fio_ddir lock_ddir;
327 * block map for random io
329 unsigned long *file_map;
330 unsigned int num_maps;
331 unsigned int last_free_lookup;
334 enum fio_file_flags flags;
338 * How many depth levels to log
340 #define FIO_IO_U_MAP_NR 8
341 #define FIO_IO_U_LAT_U_NR 10
342 #define FIO_IO_U_LAT_M_NR 12
353 struct io_log *slat_log;
354 struct io_log *clat_log;
355 struct io_log *bw_log;
358 * bandwidth and latency stats
360 struct io_stat clat_stat[2]; /* completion latency */
361 struct io_stat slat_stat[2]; /* submission latency */
362 struct io_stat bw_stat[2]; /* bandwidth stats */
364 unsigned long long stat_io_bytes[2];
365 struct timeval stat_sample_time[2];
368 * fio system usage accounting
370 struct rusage ru_start;
371 struct rusage ru_end;
372 unsigned long usr_time;
373 unsigned long sys_time;
375 unsigned long minf, majf;
378 * IO depth and latency stats
380 unsigned int io_u_map[FIO_IO_U_MAP_NR];
381 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
382 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
383 unsigned long total_io_u[2];
384 unsigned long short_io_u[2];
386 unsigned long long io_bytes[2];
387 unsigned long runtime[2];
388 unsigned long total_run_time;
396 struct thread_options {
404 enum td_ddir td_ddir;
405 unsigned int ddir_nr;
406 unsigned int iodepth;
407 unsigned int iodepth_low;
408 unsigned int iodepth_batch;
410 unsigned long long size;
411 unsigned int fill_device;
412 unsigned long long file_size_low;
413 unsigned long long file_size_high;
414 unsigned long long start_offset;
417 unsigned int min_bs[2];
418 unsigned int max_bs[2];
419 struct bssplit *bssplit;
420 unsigned int bssplit_nr;
422 unsigned int nr_files;
423 unsigned int open_files;
424 enum file_lock_mode file_lock_mode;
425 unsigned int lockfile_batch;
427 unsigned int odirect;
428 unsigned int invalidate_cache;
429 unsigned int create_serialize;
430 unsigned int create_fsync;
431 unsigned int end_fsync;
432 unsigned int sync_io;
434 unsigned int do_verify;
435 unsigned int verifysort;
436 unsigned int verify_interval;
437 unsigned int verify_offset;
438 unsigned int verify_pattern;
439 unsigned int verify_pattern_bytes;
440 unsigned int verify_fatal;
441 unsigned int use_thread;
443 unsigned int do_disk_util;
444 unsigned int override_sync;
445 unsigned int rand_repeatable;
446 unsigned int write_lat_log;
447 unsigned int write_bw_log;
448 unsigned int norandommap;
449 unsigned int softrandommap;
450 unsigned int bs_unaligned;
451 unsigned int fsync_on_close;
453 unsigned int hugepage_size;
454 unsigned int rw_min_bs;
455 unsigned int thinktime;
456 unsigned int thinktime_spin;
457 unsigned int thinktime_blocks;
458 unsigned int fsync_blocks;
459 unsigned int start_delay;
460 unsigned long long timeout;
461 unsigned int overwrite;
462 unsigned int bw_avg_time;
464 unsigned long long zone_size;
465 unsigned long long zone_skip;
466 enum fio_memtype mem_type;
468 unsigned int stonewall;
469 unsigned int new_group;
470 unsigned int numjobs;
471 os_cpu_mask_t cpumask;
472 unsigned int cpumask_set;
474 unsigned int rwmixcycle;
475 unsigned int rwmix[2];
477 unsigned int file_service_type;
478 unsigned int group_reporting;
479 unsigned int fadvise_hint;
480 unsigned int zero_buffers;
481 unsigned int time_based;
483 char *read_iolog_file;
484 char *write_iolog_file;
487 * Pre-run and post-run shell
493 unsigned int ratemin;
494 unsigned int ratecycle;
495 unsigned int rate_iops;
496 unsigned int rate_iops_min;
501 * CPU "io" cycle burner
503 unsigned int cpuload;
504 unsigned int cpucycle;
507 #define FIO_VERROR_SIZE 128
510 * This describes a single thread/process executing a fio job.
513 struct thread_options o;
514 char verror[FIO_VERROR_SIZE];
518 struct thread_stat ts;
519 struct fio_file **files;
520 unsigned int files_index;
521 unsigned int nr_open_files;
522 unsigned int nr_done_files;
523 unsigned int nr_normal_files;
525 unsigned int next_file;
526 os_random_state_t next_file_state;
532 size_t orig_buffer_size;
533 volatile int terminate;
534 volatile int runstate;
536 unsigned int ioprio_set;
537 unsigned int last_was_sync;
547 os_random_state_t bsrange_state;
548 os_random_state_t verify_state;
553 * IO engine hooks, contains everything needed to submit an io_u
554 * to any of the available IO engines.
556 struct ioengine_ops *io_ops;
559 * Current IO depth and list of free and busy io_u's.
561 unsigned int cur_depth;
562 struct list_head io_u_freelist;
563 struct list_head io_u_busylist;
564 struct list_head io_u_requeues;
565 unsigned int io_u_queued;
570 unsigned long rate_usec_cycle;
571 long rate_pending_usleep;
572 unsigned long rate_bytes;
573 unsigned long rate_blocks;
574 struct timeval lastrate;
576 unsigned long long total_io_size;
578 unsigned long io_issues[2];
579 unsigned long long io_blocks[2];
580 unsigned long long io_bytes[2];
581 unsigned long long io_skip_bytes;
582 unsigned long long this_io_bytes[2];
583 unsigned long long zone_bytes;
584 struct fio_mutex *mutex;
587 * State for random io, a bitmap of blocks done vs not done
589 os_random_state_t random_state;
591 struct timeval start; /* start of this loop */
592 struct timeval epoch; /* time job was started */
593 struct timeval rw_end[2];
594 struct timeval last_issue;
595 unsigned int rw_end_set[2];
598 * read/write mixed workload state
600 os_random_state_t rwmix_state;
601 unsigned long rwmix_issues;
602 enum fio_ddir rwmix_ddir;
603 unsigned int ddir_nr;
606 * IO history logs for verification. We use a tree for sorting,
607 * if we are overwriting. Otherwise just use a fifo.
609 struct rb_root io_hist_tree;
610 struct list_head io_hist_list;
615 struct list_head io_log_list;
620 struct timeval timeout_end;
621 struct itimerval timer;
624 * for fileservice, how often to switch to a new file
626 unsigned int file_service_nr;
627 unsigned int file_service_left;
628 struct fio_file *file_service_file;
631 * For generating file sizes
633 os_random_state_t file_size_state;
637 * roundrobin available files, or choose one at random.
640 FIO_FSERVICE_RANDOM = 1,
645 * when should interactive ETA output be generated
654 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
655 * the timer on each queue operation.
657 #define IO_U_TIMEOUT_INC 5
658 #define IO_U_TIMEOUT 30
660 #define __td_verror(td, err, msg, func) \
666 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
670 #define td_verror(td, err, func) \
671 __td_verror((td), (err), strerror((err)), (func))
672 #define td_vmsg(td, err, msg, func) \
673 __td_verror((td), (err), (msg), (func))
675 extern int exitall_on_terminate;
676 extern int thread_number;
677 extern int nr_process, nr_thread;
680 extern int terse_output;
681 extern int temp_stall_ts;
682 extern unsigned long long mlock_size;
683 extern unsigned long page_mask, page_size;
684 extern int read_only;
685 extern int eta_print;
686 extern unsigned long done_secs;
687 extern char *job_section;
689 extern struct thread_data *threads;
691 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
692 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
693 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
694 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
695 #define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
697 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
699 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
702 #define BLOCKS_PER_MAP (8 * sizeof(long))
703 #define TO_MAP_BLOCK(td, f, b) (b)
704 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
705 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
707 #define MAX_JOBS (1024)
709 static inline int should_fsync(struct thread_data *td)
711 if (td->last_was_sync)
715 if (td_write(td) || td_rw(td) || td->o.override_sync)
722 * Disk utils as read in /sys/block/<dev>/stat
724 struct disk_util_stat {
727 unsigned long long sectors[2];
730 unsigned time_in_queue;
734 * Per-device disk util management
737 struct list_head list;
744 struct disk_util_stat dus;
745 struct disk_util_stat last_dus;
751 #define DISK_UTIL_MSEC (250)
763 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
764 extern void log_io_u(struct thread_data *, struct io_u *);
765 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
766 extern int __must_check init_iolog(struct thread_data *td);
767 extern void log_io_piece(struct thread_data *, struct io_u *);
768 extern void queue_io_piece(struct thread_data *, struct io_piece *);
769 extern void prune_io_piece_log(struct thread_data *);
770 extern void write_iolog_close(struct thread_data *);
775 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
776 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
777 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
778 extern void show_run_stats(void);
779 extern void init_disk_util(struct thread_data *);
780 extern void update_rusage_stat(struct thread_data *);
781 extern void update_io_ticks(void);
782 extern void disk_util_timer_arm(void);
783 extern void setup_log(struct io_log **);
784 extern void finish_log(struct thread_data *, struct io_log *, const char *);
785 extern void __finish_log(struct io_log *, const char *);
786 extern struct io_log *agg_io_log[2];
787 extern int write_bw_log;
788 extern void add_agg_sample(unsigned long, enum fio_ddir);
793 extern unsigned long long utime_since(struct timeval *, struct timeval *);
794 extern unsigned long long utime_since_now(struct timeval *);
795 extern unsigned long mtime_since(struct timeval *, struct timeval *);
796 extern unsigned long mtime_since_now(struct timeval *);
797 extern unsigned long time_since_now(struct timeval *);
798 extern unsigned long mtime_since_genesis(void);
799 extern void __usec_sleep(unsigned int);
800 extern void usec_sleep(struct thread_data *, unsigned long);
801 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
802 extern void fill_start_time(struct timeval *);
803 extern void fio_gettime(struct timeval *, void *);
804 extern void set_genesis_time(void);
807 * Init/option functions
809 extern int __must_check parse_options(int, char **);
810 extern int fio_option_parse(struct thread_data *, const char *);
811 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
812 extern void fio_fill_default_options(struct thread_data *);
813 extern int fio_show_option_help(const char *);
814 extern void fio_options_dup_and_init(struct option *);
815 extern void options_mem_dupe(struct thread_data *);
816 extern void options_mem_free(struct thread_data *);
817 #define FIO_GETOPT_JOB 0x89988998
818 #define FIO_NR_OPTIONS 128
821 * File setup/shutdown
823 extern void close_files(struct thread_data *);
824 extern void close_and_free_files(struct thread_data *);
825 extern int __must_check setup_files(struct thread_data *);
826 extern int __must_check open_files(struct thread_data *);
827 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
828 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
829 extern int __must_check generic_close_file(struct thread_data *, struct fio_file *);
830 extern int add_file(struct thread_data *, const char *);
831 extern void get_file(struct fio_file *);
832 extern int __must_check put_file(struct thread_data *, struct fio_file *);
833 extern void lock_file(struct thread_data *, struct fio_file *, enum fio_ddir);
834 extern void unlock_file(struct thread_data *, struct fio_file *);
835 extern void unlock_file_all(struct thread_data *, struct fio_file *);
836 extern int add_dir_files(struct thread_data *, const char *);
837 extern int init_random_map(struct thread_data *);
838 extern void dup_files(struct thread_data *, struct thread_data *);
839 extern int get_fileno(struct thread_data *, const char *);
840 extern void free_release_files(struct thread_data *);
845 extern void print_thread_status(void);
846 extern void print_status_init(int);
851 #ifdef FIO_HAVE_DISK_UTIL
852 extern void show_disk_util(void);
853 extern void disk_util_timer_arm(void);
854 extern void init_disk_util(struct thread_data *);
855 extern void update_io_ticks(void);
857 #define show_disk_util()
858 #define disk_util_timer_arm()
859 #define init_disk_util(td)
860 #define update_io_ticks()
864 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
865 * will never back again. It may cycle between running/verififying/fsyncing.
866 * Once the thread reaches TD_EXITED, it is just waiting for the core to
883 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
884 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
885 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
890 extern int __must_check fio_pin_memory(void);
891 extern void fio_unpin_memory(void);
892 extern int __must_check allocate_io_mem(struct thread_data *);
893 extern void free_io_mem(struct thread_data *);
898 #define queue_full(td) list_empty(&(td)->io_u_freelist)
899 extern struct io_u *__get_io_u(struct thread_data *);
900 extern struct io_u *get_io_u(struct thread_data *);
901 extern void put_io_u(struct thread_data *, struct io_u *);
902 extern void requeue_io_u(struct thread_data *, struct io_u **);
903 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
904 extern long __must_check io_u_queued_complete(struct thread_data *, int);
905 extern void io_u_queued(struct thread_data *, struct io_u *);
906 extern void io_u_log_error(struct thread_data *, struct io_u *);
907 extern void io_u_init_timeout(void);
908 extern void io_u_set_timeout(struct thread_data *);
909 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
912 * io engine entry points
914 extern int __must_check td_io_init(struct thread_data *);
915 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
916 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
917 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
918 extern int __must_check td_io_getevents(struct thread_data *, unsigned int, unsigned int, struct timespec *);
919 extern int __must_check td_io_commit(struct thread_data *);
920 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
921 extern int td_io_close_file(struct thread_data *, struct fio_file *);
926 #ifdef FIO_HAVE_BLKTRACE
927 extern int is_blktrace(const char *);
928 extern int load_blktrace(struct thread_data *, const char *);
931 struct ioengine_ops {
932 struct list_head list;
936 int (*setup)(struct thread_data *);
937 int (*init)(struct thread_data *);
938 int (*prep)(struct thread_data *, struct io_u *);
939 int (*queue)(struct thread_data *, struct io_u *);
940 int (*commit)(struct thread_data *);
941 int (*getevents)(struct thread_data *, unsigned int, unsigned int, struct timespec *);
942 struct io_u *(*event)(struct thread_data *, int);
943 int (*cancel)(struct thread_data *, struct io_u *);
944 void (*cleanup)(struct thread_data *);
945 int (*open_file)(struct thread_data *, struct fio_file *);
946 int (*close_file)(struct thread_data *, struct fio_file *);
951 #define FIO_IOOPS_VERSION 9
953 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
954 extern void register_ioengine(struct ioengine_ops *);
955 extern void unregister_ioengine(struct ioengine_ops *);
956 extern void close_ioengine(struct thread_data *);
959 * Mark unused variables passed to ops functions as unused, to silence gcc
961 #define fio_unused __attribute((__unused__))
962 #define fio_init __attribute__((constructor))
963 #define fio_exit __attribute__((destructor))
965 #define for_each_td(td, i) \
966 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
967 #define for_each_file(td, f, i) \
968 if ((td)->files_index) \
969 for ((i) = 0, (f) = (td)->files[0]; \
970 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
973 #define fio_assert(td, cond) do { \
976 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
977 (td)->runstate = TD_EXITED; \
978 (td)->error = EFAULT; \
983 static inline void clear_error(struct thread_data *td)
986 td->verror[0] = '\0';
990 static inline void dprint_io_u(struct io_u *io_u, const char *p)
992 struct fio_file *f = io_u->file;
994 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
995 io_u->offset, io_u->buflen, io_u->ddir);
996 if (fio_debug & (1 << FD_IO)) {
998 log_info("/%s", f->file_name);
1004 #define dprint_io_u(io_u, p)