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 *);
181 * io_ops->queue() return values
184 FIO_Q_COMPLETED = 0, /* completed sync */
185 FIO_Q_QUEUED = 1, /* queued, will complete async */
186 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
189 #define FIO_HDR_MAGIC 0xf00baaef
192 VERIFY_NONE = 0, /* no verification */
193 VERIFY_MD5, /* md5 sum data blocks */
194 VERIFY_CRC64, /* crc64 sum data blocks */
195 VERIFY_CRC32, /* crc32 sum data blocks */
196 VERIFY_CRC16, /* crc16 sum data blocks */
197 VERIFY_CRC7, /* crc7 sum data blocks */
198 VERIFY_SHA256, /* sha256 sum data blocks */
199 VERIFY_SHA512, /* sha512 sum data blocks */
200 VERIFY_META, /* block_num, timestamp etc. */
201 VERIFY_NULL, /* pretend to verify */
205 * A header structure associated with each checksummed data block. It is
206 * followed by a checksum specific header that contains the verification
209 struct verify_header {
210 unsigned int fio_magic;
212 unsigned int verify_type;
216 uint32_t md5_digest[16];
238 unsigned char thread;
239 unsigned short numberio;
240 unsigned long time_sec;
241 unsigned long time_usec;
244 struct group_run_stats {
245 unsigned long long max_run[2], min_run[2];
246 unsigned long long max_bw[2], min_bw[2];
247 unsigned long long io_kb[2];
248 unsigned long long agg[2];
252 * What type of allocation to use for io buffers
255 MEM_MALLOC = 0, /* ordinary malloc */
256 MEM_SHM, /* use shared memory segments */
257 MEM_SHMHUGE, /* use shared memory segments with huge pages */
258 MEM_MMAP, /* use anonynomous mmap */
259 MEM_MMAPHUGE, /* memory mapped huge file */
263 * The type of object we are working on
266 FIO_TYPE_FILE = 1, /* plain file */
267 FIO_TYPE_BD, /* block device */
268 FIO_TYPE_CHAR, /* character device */
269 FIO_TYPE_PIPE, /* pipe */
272 enum fio_ioengine_flags {
273 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
274 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
275 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
276 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
277 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
278 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
281 enum fio_file_flags {
282 FIO_FILE_OPEN = 1 << 0, /* file is open */
283 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
284 FIO_FILE_EXISTS = 1 << 2, /* file there */
285 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
286 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
287 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
288 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
292 * Each thread_data structure has a number of files associated with it,
293 * this structure holds state information for a single file.
296 enum fio_filetype filetype;
299 * A file may not be a file descriptor, let the io engine decide
302 unsigned long file_data;
307 * filename and possible memory mapping
311 unsigned int major, minor;
314 * size of the file, offset into file, and io size from that offset
316 unsigned long long real_file_size;
317 unsigned long long file_offset;
318 unsigned long long io_size;
320 unsigned long long last_pos;
321 unsigned long long last_completed_pos;
324 * block map for random io
326 unsigned long *file_map;
327 unsigned int num_maps;
328 unsigned int last_free_lookup;
331 enum fio_file_flags flags;
335 * How many depth levels to log
337 #define FIO_IO_U_MAP_NR 8
338 #define FIO_IO_U_LAT_U_NR 10
339 #define FIO_IO_U_LAT_M_NR 12
350 struct io_log *slat_log;
351 struct io_log *clat_log;
352 struct io_log *bw_log;
355 * bandwidth and latency stats
357 struct io_stat clat_stat[2]; /* completion latency */
358 struct io_stat slat_stat[2]; /* submission latency */
359 struct io_stat bw_stat[2]; /* bandwidth stats */
361 unsigned long long stat_io_bytes[2];
362 struct timeval stat_sample_time[2];
365 * fio system usage accounting
367 struct rusage ru_start;
368 struct rusage ru_end;
369 unsigned long usr_time;
370 unsigned long sys_time;
374 * IO depth and latency stats
376 unsigned int io_u_map[FIO_IO_U_MAP_NR];
377 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
378 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
379 unsigned long total_io_u[2];
380 unsigned long short_io_u[2];
382 unsigned long long io_bytes[2];
383 unsigned long runtime[2];
384 unsigned long total_run_time;
387 struct thread_options {
395 enum td_ddir td_ddir;
396 unsigned int ddir_nr;
397 unsigned int iodepth;
398 unsigned int iodepth_low;
399 unsigned int iodepth_batch;
401 unsigned long long size;
402 unsigned long long file_size_low;
403 unsigned long long file_size_high;
404 unsigned long long start_offset;
407 unsigned int min_bs[2];
408 unsigned int max_bs[2];
410 unsigned int nr_files;
411 unsigned int open_files;
413 unsigned int odirect;
414 unsigned int invalidate_cache;
415 unsigned int create_serialize;
416 unsigned int create_fsync;
417 unsigned int end_fsync;
418 unsigned int sync_io;
420 unsigned int do_verify;
421 unsigned int verifysort;
422 unsigned int verify_interval;
423 unsigned int verify_offset;
424 unsigned int verify_pattern;
425 unsigned int verify_pattern_bytes;
426 unsigned int verify_fatal;
427 unsigned int use_thread;
429 unsigned int do_disk_util;
430 unsigned int override_sync;
431 unsigned int rand_repeatable;
432 unsigned int write_lat_log;
433 unsigned int write_bw_log;
434 unsigned int norandommap;
435 unsigned int bs_unaligned;
436 unsigned int fsync_on_close;
438 unsigned int hugepage_size;
439 unsigned int rw_min_bs;
440 unsigned int thinktime;
441 unsigned int thinktime_spin;
442 unsigned int thinktime_blocks;
443 unsigned int fsync_blocks;
444 unsigned int start_delay;
445 unsigned long long timeout;
446 unsigned int overwrite;
447 unsigned int bw_avg_time;
449 unsigned long long zone_size;
450 unsigned long long zone_skip;
451 enum fio_memtype mem_type;
453 unsigned int stonewall;
454 unsigned int new_group;
455 unsigned int numjobs;
456 os_cpu_mask_t cpumask;
457 unsigned int cpumask_set;
459 unsigned int rwmixcycle;
460 unsigned int rwmix[2];
462 unsigned int file_service_type;
463 unsigned int group_reporting;
464 unsigned int fadvise_hint;
465 unsigned int zero_buffers;
466 unsigned int time_based;
468 char *read_iolog_file;
469 char *write_iolog_file;
472 * Pre-run and post-run shell
478 unsigned int ratemin;
479 unsigned int ratecycle;
480 unsigned int rate_iops;
481 unsigned int rate_iops_min;
486 * CPU "io" cycle burner
488 unsigned int cpuload;
489 unsigned int cpucycle;
492 #define FIO_VERROR_SIZE 128
495 * This describes a single thread/process executing a fio job.
498 struct thread_options o;
499 char verror[FIO_VERROR_SIZE];
503 struct thread_stat ts;
504 struct fio_file *files;
505 unsigned int files_index;
506 unsigned int nr_open_files;
507 unsigned int nr_done_files;
508 unsigned int nr_normal_files;
510 unsigned int next_file;
511 os_random_state_t next_file_state;
517 size_t orig_buffer_size;
518 volatile int terminate;
519 volatile int runstate;
521 unsigned int last_was_sync;
531 os_random_state_t bsrange_state;
532 os_random_state_t verify_state;
537 * IO engine hooks, contains everything needed to submit an io_u
538 * to any of the available IO engines.
540 struct ioengine_ops *io_ops;
543 * Current IO depth and list of free and busy io_u's.
545 unsigned int cur_depth;
546 struct list_head io_u_freelist;
547 struct list_head io_u_busylist;
548 struct list_head io_u_requeues;
549 unsigned int io_u_queued;
554 unsigned long rate_usec_cycle;
555 long rate_pending_usleep;
556 unsigned long rate_bytes;
557 unsigned long rate_blocks;
558 struct timeval lastrate;
560 unsigned long long total_io_size;
562 unsigned long io_issues[2];
563 unsigned long long io_blocks[2];
564 unsigned long long io_bytes[2];
565 unsigned long long io_skip_bytes;
566 unsigned long long this_io_bytes[2];
567 unsigned long long zone_bytes;
568 struct fio_sem *mutex;
571 * State for random io, a bitmap of blocks done vs not done
573 os_random_state_t random_state;
575 struct timeval start; /* start of this loop */
576 struct timeval epoch; /* time job was started */
577 struct timeval rw_end[2];
578 struct timeval last_issue;
579 unsigned int rw_end_set[2];
582 * read/write mixed workload state
584 os_random_state_t rwmix_state;
585 unsigned long long rwmix_bytes;
586 struct timeval rwmix_switch;
587 enum fio_ddir rwmix_ddir;
588 unsigned int ddir_nr;
591 * IO history logs for verification. We use a tree for sorting,
592 * if we are overwriting. Otherwise just use a fifo.
594 struct rb_root io_hist_tree;
595 struct list_head io_hist_list;
600 struct list_head io_log_list;
605 struct timeval timeout_end;
606 struct itimerval timer;
609 * for fileservice, how often to switch to a new file
611 unsigned int file_service_nr;
612 unsigned int file_service_left;
613 struct fio_file *file_service_file;
616 * For generating file sizes
618 os_random_state_t file_size_state;
622 * roundrobin available files, or choose one at random.
625 FIO_FSERVICE_RANDOM = 1,
630 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
631 * the timer on each queue operation.
633 #define IO_U_TIMEOUT_INC 5
634 #define IO_U_TIMEOUT 30
636 #define __td_verror(td, err, msg, func) \
642 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
646 #define td_verror(td, err, func) \
647 __td_verror((td), (err), strerror((err)), (func))
648 #define td_vmsg(td, err, msg, func) \
649 __td_verror((td), (err), (msg), (func))
651 extern int exitall_on_terminate;
652 extern int thread_number;
653 extern int nr_process, nr_thread;
656 extern int terse_output;
659 extern int temp_stall_ts;
660 extern unsigned long long mlock_size;
661 extern unsigned long page_mask, page_size;
663 extern struct thread_data *threads;
665 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
666 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
667 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
668 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
670 #define BLOCKS_PER_MAP (8 * sizeof(long))
671 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
672 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
673 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
675 #define MAX_JOBS (1024)
677 static inline int should_fsync(struct thread_data *td)
679 if (td->last_was_sync)
683 if (td_write(td) || td_rw(td) || td->o.override_sync)
690 * Disk utils as read in /sys/block/<dev>/stat
692 struct disk_util_stat {
695 unsigned long long sectors[2];
698 unsigned time_in_queue;
702 * Per-device disk util management
705 struct list_head list;
712 struct disk_util_stat dus;
713 struct disk_util_stat last_dus;
719 #define DISK_UTIL_MSEC (250)
730 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
731 extern void log_io_u(struct thread_data *, struct io_u *);
732 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
733 extern int __must_check init_iolog(struct thread_data *td);
734 extern void log_io_piece(struct thread_data *, struct io_u *);
735 extern void prune_io_piece_log(struct thread_data *);
736 extern void write_iolog_close(struct thread_data *);
741 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
742 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
743 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
744 extern void show_run_stats(void);
745 extern void init_disk_util(struct thread_data *);
746 extern void update_rusage_stat(struct thread_data *);
747 extern void update_io_ticks(void);
748 extern void disk_util_timer_arm(void);
749 extern void setup_log(struct io_log **);
750 extern void finish_log(struct thread_data *, struct io_log *, const char *);
751 extern void __finish_log(struct io_log *, const char *);
752 extern struct io_log *agg_io_log[2];
753 extern int write_bw_log;
754 extern void add_agg_sample(unsigned long, enum fio_ddir);
759 extern unsigned long utime_since(struct timeval *, struct timeval *);
760 extern unsigned long utime_since_now(struct timeval *);
761 extern unsigned long mtime_since(struct timeval *, struct timeval *);
762 extern unsigned long mtime_since_now(struct timeval *);
763 extern unsigned long time_since_now(struct timeval *);
764 extern unsigned long mtime_since_genesis(void);
765 extern void __usec_sleep(unsigned int);
766 extern void usec_sleep(struct thread_data *, unsigned long);
767 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
768 extern void fill_start_time(struct timeval *);
769 extern void fio_gettime(struct timeval *, void *);
770 extern void set_genesis_time(void);
773 * Init/option functions
775 extern int __must_check parse_options(int, char **);
776 extern int fio_option_parse(struct thread_data *, const char *);
777 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
778 extern void fio_fill_default_options(struct thread_data *);
779 extern int fio_show_option_help(const char *);
780 extern void fio_options_dup_and_init(struct option *);
781 extern void options_mem_dupe(struct thread_data *);
782 extern void options_mem_free(struct thread_data *);
783 #define FIO_GETOPT_JOB 0x89988998
784 #define FIO_NR_OPTIONS 128
787 * File setup/shutdown
789 extern void close_files(struct thread_data *);
790 extern int __must_check setup_files(struct thread_data *);
791 extern int __must_check open_files(struct thread_data *);
792 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
793 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
794 extern void generic_close_file(struct thread_data *, struct fio_file *);
795 extern int add_file(struct thread_data *, const char *);
796 extern void get_file(struct fio_file *);
797 extern void put_file(struct thread_data *, struct fio_file *);
798 extern int add_dir_files(struct thread_data *, const char *);
799 extern int init_random_map(struct thread_data *);
800 extern void dup_files(struct thread_data *, struct thread_data *);
801 extern int get_fileno(struct thread_data *, const char *);
802 extern void free_release_files(struct thread_data *);
807 extern void print_thread_status(void);
808 extern void print_status_init(int);
813 #ifdef FIO_HAVE_DISK_UTIL
814 extern void show_disk_util(void);
815 extern void disk_util_timer_arm(void);
816 extern void init_disk_util(struct thread_data *);
817 extern void update_io_ticks(void);
819 #define show_disk_util()
820 #define disk_util_timer_arm()
821 #define init_disk_util(td)
822 #define update_io_ticks()
826 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
827 * will never back again. It may cycle between running/verififying/fsyncing.
828 * Once the thread reaches TD_EXITED, it is just waiting for the core to
845 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
846 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
847 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
852 extern int __must_check fio_pin_memory(void);
853 extern void fio_unpin_memory(void);
854 extern int __must_check allocate_io_mem(struct thread_data *);
855 extern void free_io_mem(struct thread_data *);
860 #define queue_full(td) list_empty(&(td)->io_u_freelist)
861 extern struct io_u *__get_io_u(struct thread_data *);
862 extern struct io_u *get_io_u(struct thread_data *);
863 extern void put_io_u(struct thread_data *, struct io_u *);
864 extern void requeue_io_u(struct thread_data *, struct io_u **);
865 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
866 extern long __must_check io_u_queued_complete(struct thread_data *, int);
867 extern void io_u_queued(struct thread_data *, struct io_u *);
868 extern void io_u_log_error(struct thread_data *, struct io_u *);
869 extern void io_u_init_timeout(void);
870 extern void io_u_set_timeout(struct thread_data *);
871 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
874 * io engine entry points
876 extern int __must_check td_io_init(struct thread_data *);
877 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
878 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
879 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
880 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
881 extern int __must_check td_io_commit(struct thread_data *);
882 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
883 extern void td_io_close_file(struct thread_data *, struct fio_file *);
888 #ifdef FIO_HAVE_BLKTRACE
889 extern int is_blktrace(const char *);
890 extern int load_blktrace(struct thread_data *, const char *);
894 * If logging output to a file, stderr should go to both stderr and f_err
896 #define log_err(args...) do { \
897 fprintf(f_err, ##args); \
898 if (f_err != stderr) \
899 fprintf(stderr, ##args); \
902 #define log_info(args...) fprintf(f_out, ##args)
904 FILE *get_f_out(void);
905 FILE *get_f_err(void);
907 struct ioengine_ops {
908 struct list_head list;
912 int (*setup)(struct thread_data *);
913 int (*init)(struct thread_data *);
914 int (*prep)(struct thread_data *, struct io_u *);
915 int (*queue)(struct thread_data *, struct io_u *);
916 int (*commit)(struct thread_data *);
917 int (*getevents)(struct thread_data *, int, int, struct timespec *);
918 struct io_u *(*event)(struct thread_data *, int);
919 int (*cancel)(struct thread_data *, struct io_u *);
920 void (*cleanup)(struct thread_data *);
921 int (*open_file)(struct thread_data *, struct fio_file *);
922 void (*close_file)(struct thread_data *, struct fio_file *);
927 #define FIO_IOOPS_VERSION 7
929 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
930 extern void register_ioengine(struct ioengine_ops *);
931 extern void unregister_ioengine(struct ioengine_ops *);
932 extern void close_ioengine(struct thread_data *);
935 * Mark unused variables passed to ops functions as unused, to silence gcc
937 #define fio_unused __attribute((__unused__))
938 #define fio_init __attribute__((constructor))
939 #define fio_exit __attribute__((destructor))
941 #define for_each_td(td, i) \
942 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
943 #define for_each_file(td, f, i) \
944 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
946 #define fio_assert(td, cond) do { \
949 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
950 (td)->runstate = TD_EXITED; \
951 (td)->error = EFAULT; \
956 static inline void clear_error(struct thread_data *td)
959 td->verror[0] = '\0';