8 #include <sys/resource.h>
16 #include "compiler/compiler.h"
25 #include "arch/arch.h"
29 #ifdef FIO_HAVE_SYSLET
45 TD_DDIR_READ = 1 << 0,
46 TD_DDIR_WRITE = 1 << 1,
47 TD_DDIR_RAND = 1 << 2,
48 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
49 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
50 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
51 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
55 * Use for maintaining statistics
58 unsigned long max_val;
59 unsigned long min_val;
60 unsigned long samples;
67 * A single data sample
76 * Dynamically growing data sample log
79 unsigned long nr_samples;
80 unsigned long max_samples;
81 struct io_sample *log;
85 * When logging io actions, this matches a single sent io_u
89 struct rb_node rb_node;
90 struct list_head list;
94 struct fio_file *file;
96 unsigned long long offset;
101 unsigned int file_action;
105 #ifdef FIO_HAVE_SYSLET
107 struct syslet_uatom atom; /* the atom to submit */
108 struct syslet_uatom *head; /* head of the sequence */
109 long ret; /* syscall return value */
114 IO_U_F_FREE = 1 << 0,
115 IO_U_F_FLIGHT = 1 << 1,
125 #ifdef FIO_HAVE_LIBAIO
128 #ifdef FIO_HAVE_POSIXAIO
132 struct sg_io_hdr hdr;
134 #ifdef FIO_HAVE_SYSLET
135 struct syslet_req req;
137 #ifdef FIO_HAVE_GUASI
141 struct timeval start_time;
142 struct timeval issue_time;
145 * Allocated/set buffer and length
148 unsigned long buflen;
149 unsigned long long offset;
150 unsigned long long endpos;
153 * IO engine state, may be different from above when we get
154 * partial transfers / residual data counts
157 unsigned long xfer_buflen;
165 * io engine private data
174 struct fio_file *file;
176 struct list_head list;
179 * Callback for io completion
181 int (*end_io)(struct thread_data *, struct io_u *);
185 * io_ops->queue() return values
188 FIO_Q_COMPLETED = 0, /* completed sync */
189 FIO_Q_QUEUED = 1, /* queued, will complete async */
190 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
193 #define FIO_HDR_MAGIC 0xf00baaef
196 VERIFY_NONE = 0, /* no verification */
197 VERIFY_MD5, /* md5 sum data blocks */
198 VERIFY_CRC64, /* crc64 sum data blocks */
199 VERIFY_CRC32, /* crc32 sum data blocks */
200 VERIFY_CRC16, /* crc16 sum data blocks */
201 VERIFY_CRC7, /* crc7 sum data blocks */
202 VERIFY_NULL, /* pretend to verify */
206 * A header structure associated with each checksummed data block
208 struct verify_header {
209 unsigned int fio_magic;
211 unsigned int verify_type;
213 char md5_digest[MD5_HASH_WORDS * 4];
214 unsigned long long crc64;
216 unsigned short crc16;
221 struct group_run_stats {
222 unsigned long long max_run[2], min_run[2];
223 unsigned long long max_bw[2], min_bw[2];
224 unsigned long long io_kb[2];
225 unsigned long long agg[2];
229 * What type of allocation to use for io buffers
232 MEM_MALLOC = 0, /* ordinary malloc */
233 MEM_SHM, /* use shared memory segments */
234 MEM_SHMHUGE, /* use shared memory segments with huge pages */
235 MEM_MMAP, /* use anonynomous mmap */
236 MEM_MMAPHUGE, /* memory mapped huge file */
240 * The type of object we are working on
243 FIO_TYPE_FILE = 1, /* plain file */
244 FIO_TYPE_BD, /* block device */
245 FIO_TYPE_CHAR, /* character device */
246 FIO_TYPE_PIPE, /* pipe */
249 enum fio_ioengine_flags {
250 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
251 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
252 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
253 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
254 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
255 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
258 enum fio_file_flags {
259 FIO_FILE_OPEN = 1 << 0, /* file is open */
260 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
261 FIO_FILE_EXISTS = 1 << 2, /* file there */
262 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
263 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
264 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
265 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
269 * Each thread_data structure has a number of files associated with it,
270 * this structure holds state information for a single file.
273 enum fio_filetype filetype;
276 * A file may not be a file descriptor, let the io engine decide
279 unsigned long file_data;
284 * filename and possible memory mapping
288 unsigned int major, minor;
291 * size of the file, offset into file, and io size from that offset
293 unsigned long long real_file_size;
294 unsigned long long file_offset;
295 unsigned long long io_size;
297 unsigned long long last_pos;
298 unsigned long long last_completed_pos;
301 * block map for random io
303 unsigned long *file_map;
304 unsigned int num_maps;
305 unsigned int last_free_lookup;
308 enum fio_file_flags flags;
312 * How many depth levels to log
314 #define FIO_IO_U_MAP_NR 8
315 #define FIO_IO_U_LAT_U_NR 10
316 #define FIO_IO_U_LAT_M_NR 12
327 struct io_log *slat_log;
328 struct io_log *clat_log;
329 struct io_log *bw_log;
332 * bandwidth and latency stats
334 struct io_stat clat_stat[2]; /* completion latency */
335 struct io_stat slat_stat[2]; /* submission latency */
336 struct io_stat bw_stat[2]; /* bandwidth stats */
338 unsigned long long stat_io_bytes[2];
339 struct timeval stat_sample_time[2];
342 * fio system usage accounting
344 struct rusage ru_start;
345 struct rusage ru_end;
346 unsigned long usr_time;
347 unsigned long sys_time;
351 * IO depth and latency stats
353 unsigned int io_u_map[FIO_IO_U_MAP_NR];
354 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
355 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
356 unsigned long total_io_u[2];
357 unsigned long short_io_u[2];
359 unsigned long long io_bytes[2];
360 unsigned long runtime[2];
361 unsigned long total_run_time;
364 struct thread_options {
372 enum td_ddir td_ddir;
373 unsigned int ddir_nr;
374 unsigned int iodepth;
375 unsigned int iodepth_low;
376 unsigned int iodepth_batch;
378 unsigned long long size;
379 unsigned long long file_size_low;
380 unsigned long long file_size_high;
381 unsigned long long start_offset;
384 unsigned int min_bs[2];
385 unsigned int max_bs[2];
387 unsigned int nr_files;
388 unsigned int open_files;
390 unsigned int odirect;
391 unsigned int invalidate_cache;
392 unsigned int create_serialize;
393 unsigned int create_fsync;
394 unsigned int end_fsync;
395 unsigned int sync_io;
397 unsigned int verifysort;
398 unsigned int header_interval;
399 unsigned int use_thread;
401 unsigned int do_disk_util;
402 unsigned int override_sync;
403 unsigned int rand_repeatable;
404 unsigned int write_lat_log;
405 unsigned int write_bw_log;
406 unsigned int norandommap;
407 unsigned int bs_unaligned;
408 unsigned int fsync_on_close;
410 unsigned int hugepage_size;
411 unsigned int rw_min_bs;
412 unsigned int thinktime;
413 unsigned int thinktime_spin;
414 unsigned int thinktime_blocks;
415 unsigned int fsync_blocks;
416 unsigned int start_delay;
417 unsigned long long timeout;
418 unsigned int overwrite;
419 unsigned int bw_avg_time;
421 unsigned long long zone_size;
422 unsigned long long zone_skip;
423 enum fio_memtype mem_type;
425 unsigned int stonewall;
426 unsigned int new_group;
427 unsigned int numjobs;
428 os_cpu_mask_t cpumask;
429 unsigned int cpumask_set;
431 unsigned int rwmixcycle;
432 unsigned int rwmix[2];
434 unsigned int file_service_type;
435 unsigned int group_reporting;
436 unsigned int fadvise_hint;
437 unsigned int zero_buffers;
438 unsigned int time_based;
440 char *read_iolog_file;
441 char *write_iolog_file;
444 * Pre-run and post-run shell
450 unsigned int ratemin;
451 unsigned int ratecycle;
452 unsigned int rate_iops;
453 unsigned int rate_iops_min;
458 * CPU "io" cycle burner
460 unsigned int cpuload;
461 unsigned int cpucycle;
464 #define FIO_VERROR_SIZE 128
467 * This describes a single thread/process executing a fio job.
470 struct thread_options o;
471 char verror[FIO_VERROR_SIZE];
475 struct thread_stat ts;
476 struct fio_file *files;
477 unsigned int files_index;
478 unsigned int nr_open_files;
479 unsigned int nr_done_files;
480 unsigned int nr_normal_files;
482 unsigned int next_file;
483 os_random_state_t next_file_state;
489 size_t orig_buffer_size;
490 volatile int terminate;
491 volatile int runstate;
493 unsigned int last_was_sync;
503 os_random_state_t bsrange_state;
504 os_random_state_t verify_state;
509 * IO engine hooks, contains everything needed to submit an io_u
510 * to any of the available IO engines.
512 struct ioengine_ops *io_ops;
515 * Current IO depth and list of free and busy io_u's.
517 unsigned int cur_depth;
518 struct list_head io_u_freelist;
519 struct list_head io_u_busylist;
520 struct list_head io_u_requeues;
521 unsigned int io_u_queued;
526 unsigned long rate_usec_cycle;
527 long rate_pending_usleep;
528 unsigned long rate_bytes;
529 unsigned long rate_blocks;
530 struct timeval lastrate;
532 unsigned long long total_io_size;
534 unsigned long io_issues[2];
535 unsigned long long io_blocks[2];
536 unsigned long long io_bytes[2];
537 unsigned long long io_skip_bytes;
538 unsigned long long this_io_bytes[2];
539 unsigned long long zone_bytes;
540 struct fio_sem *mutex;
543 * State for random io, a bitmap of blocks done vs not done
545 os_random_state_t random_state;
547 struct timeval start; /* start of this loop */
548 struct timeval epoch; /* time job was started */
549 struct timeval rw_end[2];
550 struct timeval last_issue;
551 unsigned int rw_end_set[2];
554 * read/write mixed workload state
556 os_random_state_t rwmix_state;
557 unsigned long long rwmix_bytes;
558 struct timeval rwmix_switch;
559 enum fio_ddir rwmix_ddir;
560 unsigned int ddir_nr;
563 * IO history logs for verification. We use a tree for sorting,
564 * if we are overwriting. Otherwise just use a fifo.
566 struct rb_root io_hist_tree;
567 struct list_head io_hist_list;
572 struct list_head io_log_list;
577 struct timeval timeout_end;
578 struct itimerval timer;
581 * for fileservice, how often to switch to a new file
583 unsigned int file_service_nr;
584 unsigned int file_service_left;
585 struct fio_file *file_service_file;
588 * For generating file sizes
590 os_random_state_t file_size_state;
594 * roundrobin available files, or choose one at random.
597 FIO_FSERVICE_RANDOM = 1,
602 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
603 * the timer on each queue operation.
605 #define IO_U_TIMEOUT_INC 5
606 #define IO_U_TIMEOUT 30
608 #define __td_verror(td, err, msg, func) \
614 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
618 #define td_verror(td, err, func) \
619 __td_verror((td), (err), strerror((err)), (func))
620 #define td_vmsg(td, err, msg, func) \
621 __td_verror((td), (err), (msg), (func))
623 extern int exitall_on_terminate;
624 extern int thread_number;
625 extern int nr_process, nr_thread;
628 extern int terse_output;
631 extern int temp_stall_ts;
632 extern unsigned long long mlock_size;
633 extern unsigned long page_mask, page_size;
635 extern struct thread_data *threads;
637 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
638 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
639 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
640 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
642 #define BLOCKS_PER_MAP (8 * sizeof(long))
643 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
644 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
645 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
647 #define MAX_JOBS (1024)
649 static inline int should_fsync(struct thread_data *td)
651 if (td->last_was_sync)
655 if (td_write(td) || td_rw(td) || td->o.override_sync)
662 * Disk utils as read in /sys/block/<dev>/stat
664 struct disk_util_stat {
667 unsigned long long sectors[2];
670 unsigned time_in_queue;
674 * Per-device disk util management
677 struct list_head list;
684 struct disk_util_stat dus;
685 struct disk_util_stat last_dus;
691 #define DISK_UTIL_MSEC (250)
702 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
703 extern void log_io_u(struct thread_data *, struct io_u *);
704 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
705 extern int __must_check init_iolog(struct thread_data *td);
706 extern void log_io_piece(struct thread_data *, struct io_u *);
707 extern void prune_io_piece_log(struct thread_data *);
708 extern void write_iolog_close(struct thread_data *);
713 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
714 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
715 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
716 extern void show_run_stats(void);
717 extern void init_disk_util(struct thread_data *);
718 extern void update_rusage_stat(struct thread_data *);
719 extern void update_io_ticks(void);
720 extern void disk_util_timer_arm(void);
721 extern void setup_log(struct io_log **);
722 extern void finish_log(struct thread_data *, struct io_log *, const char *);
723 extern void __finish_log(struct io_log *, const char *);
724 extern struct io_log *agg_io_log[2];
725 extern int write_bw_log;
726 extern void add_agg_sample(unsigned long, enum fio_ddir);
731 extern unsigned long utime_since(struct timeval *, struct timeval *);
732 extern unsigned long utime_since_now(struct timeval *);
733 extern unsigned long mtime_since(struct timeval *, struct timeval *);
734 extern unsigned long mtime_since_now(struct timeval *);
735 extern unsigned long time_since_now(struct timeval *);
736 extern unsigned long mtime_since_genesis(void);
737 extern void __usec_sleep(unsigned int);
738 extern void usec_sleep(struct thread_data *, unsigned long);
739 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
740 extern void fill_start_time(struct timeval *);
741 extern void fio_gettime(struct timeval *, void *);
742 extern void set_genesis_time(void);
745 * Init/option functions
747 extern int __must_check parse_options(int, char **);
748 extern int fio_option_parse(struct thread_data *, const char *);
749 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
750 extern void fio_fill_default_options(struct thread_data *);
751 extern int fio_show_option_help(const char *);
752 extern void fio_options_dup_and_init(struct option *);
753 extern void options_mem_dupe(struct thread_data *);
754 extern void options_mem_free(struct thread_data *);
755 #define FIO_GETOPT_JOB 0x89988998
756 #define FIO_NR_OPTIONS 128
759 * File setup/shutdown
761 extern void close_files(struct thread_data *);
762 extern int __must_check setup_files(struct thread_data *);
763 extern int __must_check open_files(struct thread_data *);
764 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
765 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
766 extern void generic_close_file(struct thread_data *, struct fio_file *);
767 extern int add_file(struct thread_data *, const char *);
768 extern void get_file(struct fio_file *);
769 extern void put_file(struct thread_data *, struct fio_file *);
770 extern int add_dir_files(struct thread_data *, const char *);
771 extern int init_random_map(struct thread_data *);
772 extern void dup_files(struct thread_data *, struct thread_data *);
773 extern int get_fileno(struct thread_data *, const char *);
774 extern void free_release_files(struct thread_data *);
779 extern void print_thread_status(void);
780 extern void print_status_init(int);
785 #ifdef FIO_HAVE_DISK_UTIL
786 extern void show_disk_util(void);
787 extern void disk_util_timer_arm(void);
788 extern void init_disk_util(struct thread_data *);
789 extern void update_io_ticks(void);
791 #define show_disk_util()
792 #define disk_util_timer_arm()
793 #define init_disk_util(td)
794 #define update_io_ticks()
798 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
799 * will never back again. It may cycle between running/verififying/fsyncing.
800 * Once the thread reaches TD_EXITED, it is just waiting for the core to
817 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
818 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
819 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
824 extern int __must_check fio_pin_memory(void);
825 extern void fio_unpin_memory(void);
826 extern int __must_check allocate_io_mem(struct thread_data *);
827 extern void free_io_mem(struct thread_data *);
832 #define queue_full(td) list_empty(&(td)->io_u_freelist)
833 extern struct io_u *__get_io_u(struct thread_data *);
834 extern struct io_u *get_io_u(struct thread_data *);
835 extern void put_io_u(struct thread_data *, struct io_u *);
836 extern void requeue_io_u(struct thread_data *, struct io_u **);
837 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
838 extern long __must_check io_u_queued_complete(struct thread_data *, int);
839 extern void io_u_queued(struct thread_data *, struct io_u *);
840 extern void io_u_log_error(struct thread_data *, struct io_u *);
841 extern void io_u_init_timeout(void);
842 extern void io_u_set_timeout(struct thread_data *);
843 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
846 * io engine entry points
848 extern int __must_check td_io_init(struct thread_data *);
849 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
850 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
851 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
852 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
853 extern int __must_check td_io_commit(struct thread_data *);
854 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
855 extern void td_io_close_file(struct thread_data *, struct fio_file *);
860 #ifdef FIO_HAVE_BLKTRACE
861 extern int is_blktrace(const char *);
862 extern int load_blktrace(struct thread_data *, const char *);
866 * If logging output to a file, stderr should go to both stderr and f_err
868 #define log_err(args...) do { \
869 fprintf(f_err, ##args); \
870 if (f_err != stderr) \
871 fprintf(stderr, ##args); \
874 #define log_info(args...) fprintf(f_out, ##args)
876 FILE *get_f_out(void);
877 FILE *get_f_err(void);
879 struct ioengine_ops {
880 struct list_head list;
884 int (*setup)(struct thread_data *);
885 int (*init)(struct thread_data *);
886 int (*prep)(struct thread_data *, struct io_u *);
887 int (*queue)(struct thread_data *, struct io_u *);
888 int (*commit)(struct thread_data *);
889 int (*getevents)(struct thread_data *, int, int, struct timespec *);
890 struct io_u *(*event)(struct thread_data *, int);
891 int (*cancel)(struct thread_data *, struct io_u *);
892 void (*cleanup)(struct thread_data *);
893 int (*open_file)(struct thread_data *, struct fio_file *);
894 void (*close_file)(struct thread_data *, struct fio_file *);
899 #define FIO_IOOPS_VERSION 7
901 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
902 extern void register_ioengine(struct ioengine_ops *);
903 extern void unregister_ioengine(struct ioengine_ops *);
904 extern void close_ioengine(struct thread_data *);
907 * Mark unused variables passed to ops functions as unused, to silence gcc
909 #define fio_unused __attribute((__unused__))
910 #define fio_init __attribute__((constructor))
911 #define fio_exit __attribute__((destructor))
913 #define for_each_td(td, i) \
914 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
915 #define for_each_file(td, f, i) \
916 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
918 #define fio_assert(td, cond) do { \
921 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
922 (td)->runstate = TD_EXITED; \
923 (td)->error = EFAULT; \
928 static inline void clear_error(struct thread_data *td)
931 td->verror[0] = '\0';