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_NULL, /* pretend to verify */
204 * A header structure associated with each checksummed data block. It is
205 * followed by a checksum specific header that contains the verification
208 struct verify_header {
209 unsigned int fio_magic;
211 unsigned int verify_type;
215 uint32_t md5_digest[16];
236 struct group_run_stats {
237 unsigned long long max_run[2], min_run[2];
238 unsigned long long max_bw[2], min_bw[2];
239 unsigned long long io_kb[2];
240 unsigned long long agg[2];
244 * What type of allocation to use for io buffers
247 MEM_MALLOC = 0, /* ordinary malloc */
248 MEM_SHM, /* use shared memory segments */
249 MEM_SHMHUGE, /* use shared memory segments with huge pages */
250 MEM_MMAP, /* use anonynomous mmap */
251 MEM_MMAPHUGE, /* memory mapped huge file */
255 * The type of object we are working on
258 FIO_TYPE_FILE = 1, /* plain file */
259 FIO_TYPE_BD, /* block device */
260 FIO_TYPE_CHAR, /* character device */
261 FIO_TYPE_PIPE, /* pipe */
264 enum fio_ioengine_flags {
265 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
266 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
267 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
268 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
269 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
270 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
273 enum fio_file_flags {
274 FIO_FILE_OPEN = 1 << 0, /* file is open */
275 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
276 FIO_FILE_EXISTS = 1 << 2, /* file there */
277 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
278 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
279 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
280 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
284 * Each thread_data structure has a number of files associated with it,
285 * this structure holds state information for a single file.
288 enum fio_filetype filetype;
291 * A file may not be a file descriptor, let the io engine decide
294 unsigned long file_data;
299 * filename and possible memory mapping
303 unsigned int major, minor;
306 * size of the file, offset into file, and io size from that offset
308 unsigned long long real_file_size;
309 unsigned long long file_offset;
310 unsigned long long io_size;
312 unsigned long long last_pos;
313 unsigned long long last_completed_pos;
316 * block map for random io
318 unsigned long *file_map;
319 unsigned int num_maps;
320 unsigned int last_free_lookup;
323 enum fio_file_flags flags;
327 * How many depth levels to log
329 #define FIO_IO_U_MAP_NR 8
330 #define FIO_IO_U_LAT_U_NR 10
331 #define FIO_IO_U_LAT_M_NR 12
342 struct io_log *slat_log;
343 struct io_log *clat_log;
344 struct io_log *bw_log;
347 * bandwidth and latency stats
349 struct io_stat clat_stat[2]; /* completion latency */
350 struct io_stat slat_stat[2]; /* submission latency */
351 struct io_stat bw_stat[2]; /* bandwidth stats */
353 unsigned long long stat_io_bytes[2];
354 struct timeval stat_sample_time[2];
357 * fio system usage accounting
359 struct rusage ru_start;
360 struct rusage ru_end;
361 unsigned long usr_time;
362 unsigned long sys_time;
366 * IO depth and latency stats
368 unsigned int io_u_map[FIO_IO_U_MAP_NR];
369 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
370 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
371 unsigned long total_io_u[2];
372 unsigned long short_io_u[2];
374 unsigned long long io_bytes[2];
375 unsigned long runtime[2];
376 unsigned long total_run_time;
379 struct thread_options {
387 enum td_ddir td_ddir;
388 unsigned int ddir_nr;
389 unsigned int iodepth;
390 unsigned int iodepth_low;
391 unsigned int iodepth_batch;
393 unsigned long long size;
394 unsigned long long file_size_low;
395 unsigned long long file_size_high;
396 unsigned long long start_offset;
399 unsigned int min_bs[2];
400 unsigned int max_bs[2];
402 unsigned int nr_files;
403 unsigned int open_files;
405 unsigned int odirect;
406 unsigned int invalidate_cache;
407 unsigned int create_serialize;
408 unsigned int create_fsync;
409 unsigned int end_fsync;
410 unsigned int sync_io;
412 unsigned int verifysort;
413 unsigned int verify_interval;
414 unsigned int verify_offset;
415 unsigned int verify_pattern;
416 unsigned int verify_pattern_bytes;
417 unsigned int use_thread;
419 unsigned int do_disk_util;
420 unsigned int override_sync;
421 unsigned int rand_repeatable;
422 unsigned int write_lat_log;
423 unsigned int write_bw_log;
424 unsigned int norandommap;
425 unsigned int bs_unaligned;
426 unsigned int fsync_on_close;
428 unsigned int hugepage_size;
429 unsigned int rw_min_bs;
430 unsigned int thinktime;
431 unsigned int thinktime_spin;
432 unsigned int thinktime_blocks;
433 unsigned int fsync_blocks;
434 unsigned int start_delay;
435 unsigned long long timeout;
436 unsigned int overwrite;
437 unsigned int bw_avg_time;
439 unsigned long long zone_size;
440 unsigned long long zone_skip;
441 enum fio_memtype mem_type;
443 unsigned int stonewall;
444 unsigned int new_group;
445 unsigned int numjobs;
446 os_cpu_mask_t cpumask;
447 unsigned int cpumask_set;
449 unsigned int rwmixcycle;
450 unsigned int rwmix[2];
452 unsigned int file_service_type;
453 unsigned int group_reporting;
454 unsigned int fadvise_hint;
455 unsigned int zero_buffers;
456 unsigned int time_based;
458 char *read_iolog_file;
459 char *write_iolog_file;
462 * Pre-run and post-run shell
468 unsigned int ratemin;
469 unsigned int ratecycle;
470 unsigned int rate_iops;
471 unsigned int rate_iops_min;
476 * CPU "io" cycle burner
478 unsigned int cpuload;
479 unsigned int cpucycle;
482 #define FIO_VERROR_SIZE 128
485 * This describes a single thread/process executing a fio job.
488 struct thread_options o;
489 char verror[FIO_VERROR_SIZE];
493 struct thread_stat ts;
494 struct fio_file *files;
495 unsigned int files_index;
496 unsigned int nr_open_files;
497 unsigned int nr_done_files;
498 unsigned int nr_normal_files;
500 unsigned int next_file;
501 os_random_state_t next_file_state;
507 size_t orig_buffer_size;
508 volatile int terminate;
509 volatile int runstate;
511 unsigned int last_was_sync;
521 os_random_state_t bsrange_state;
522 os_random_state_t verify_state;
527 * IO engine hooks, contains everything needed to submit an io_u
528 * to any of the available IO engines.
530 struct ioengine_ops *io_ops;
533 * Current IO depth and list of free and busy io_u's.
535 unsigned int cur_depth;
536 struct list_head io_u_freelist;
537 struct list_head io_u_busylist;
538 struct list_head io_u_requeues;
539 unsigned int io_u_queued;
544 unsigned long rate_usec_cycle;
545 long rate_pending_usleep;
546 unsigned long rate_bytes;
547 unsigned long rate_blocks;
548 struct timeval lastrate;
550 unsigned long long total_io_size;
552 unsigned long io_issues[2];
553 unsigned long long io_blocks[2];
554 unsigned long long io_bytes[2];
555 unsigned long long io_skip_bytes;
556 unsigned long long this_io_bytes[2];
557 unsigned long long zone_bytes;
558 struct fio_sem *mutex;
561 * State for random io, a bitmap of blocks done vs not done
563 os_random_state_t random_state;
565 struct timeval start; /* start of this loop */
566 struct timeval epoch; /* time job was started */
567 struct timeval rw_end[2];
568 struct timeval last_issue;
569 unsigned int rw_end_set[2];
572 * read/write mixed workload state
574 os_random_state_t rwmix_state;
575 unsigned long long rwmix_bytes;
576 struct timeval rwmix_switch;
577 enum fio_ddir rwmix_ddir;
578 unsigned int ddir_nr;
581 * IO history logs for verification. We use a tree for sorting,
582 * if we are overwriting. Otherwise just use a fifo.
584 struct rb_root io_hist_tree;
585 struct list_head io_hist_list;
590 struct list_head io_log_list;
595 struct timeval timeout_end;
596 struct itimerval timer;
599 * for fileservice, how often to switch to a new file
601 unsigned int file_service_nr;
602 unsigned int file_service_left;
603 struct fio_file *file_service_file;
606 * For generating file sizes
608 os_random_state_t file_size_state;
612 * roundrobin available files, or choose one at random.
615 FIO_FSERVICE_RANDOM = 1,
620 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
621 * the timer on each queue operation.
623 #define IO_U_TIMEOUT_INC 5
624 #define IO_U_TIMEOUT 30
626 #define __td_verror(td, err, msg, func) \
632 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
636 #define td_verror(td, err, func) \
637 __td_verror((td), (err), strerror((err)), (func))
638 #define td_vmsg(td, err, msg, func) \
639 __td_verror((td), (err), (msg), (func))
641 extern int exitall_on_terminate;
642 extern int thread_number;
643 extern int nr_process, nr_thread;
646 extern int terse_output;
649 extern int temp_stall_ts;
650 extern unsigned long long mlock_size;
651 extern unsigned long page_mask, page_size;
653 extern struct thread_data *threads;
655 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
656 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
657 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
658 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
660 #define BLOCKS_PER_MAP (8 * sizeof(long))
661 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
662 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
663 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
665 #define MAX_JOBS (1024)
667 static inline int should_fsync(struct thread_data *td)
669 if (td->last_was_sync)
673 if (td_write(td) || td_rw(td) || td->o.override_sync)
680 * Disk utils as read in /sys/block/<dev>/stat
682 struct disk_util_stat {
685 unsigned long long sectors[2];
688 unsigned time_in_queue;
692 * Per-device disk util management
695 struct list_head list;
702 struct disk_util_stat dus;
703 struct disk_util_stat last_dus;
709 #define DISK_UTIL_MSEC (250)
720 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
721 extern void log_io_u(struct thread_data *, struct io_u *);
722 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
723 extern int __must_check init_iolog(struct thread_data *td);
724 extern void log_io_piece(struct thread_data *, struct io_u *);
725 extern void prune_io_piece_log(struct thread_data *);
726 extern void write_iolog_close(struct thread_data *);
731 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
732 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
733 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
734 extern void show_run_stats(void);
735 extern void init_disk_util(struct thread_data *);
736 extern void update_rusage_stat(struct thread_data *);
737 extern void update_io_ticks(void);
738 extern void disk_util_timer_arm(void);
739 extern void setup_log(struct io_log **);
740 extern void finish_log(struct thread_data *, struct io_log *, const char *);
741 extern void __finish_log(struct io_log *, const char *);
742 extern struct io_log *agg_io_log[2];
743 extern int write_bw_log;
744 extern void add_agg_sample(unsigned long, enum fio_ddir);
749 extern unsigned long utime_since(struct timeval *, struct timeval *);
750 extern unsigned long utime_since_now(struct timeval *);
751 extern unsigned long mtime_since(struct timeval *, struct timeval *);
752 extern unsigned long mtime_since_now(struct timeval *);
753 extern unsigned long time_since_now(struct timeval *);
754 extern unsigned long mtime_since_genesis(void);
755 extern void __usec_sleep(unsigned int);
756 extern void usec_sleep(struct thread_data *, unsigned long);
757 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
758 extern void fill_start_time(struct timeval *);
759 extern void fio_gettime(struct timeval *, void *);
760 extern void set_genesis_time(void);
763 * Init/option functions
765 extern int __must_check parse_options(int, char **);
766 extern int fio_option_parse(struct thread_data *, const char *);
767 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
768 extern void fio_fill_default_options(struct thread_data *);
769 extern int fio_show_option_help(const char *);
770 extern void fio_options_dup_and_init(struct option *);
771 extern void options_mem_dupe(struct thread_data *);
772 extern void options_mem_free(struct thread_data *);
773 #define FIO_GETOPT_JOB 0x89988998
774 #define FIO_NR_OPTIONS 128
777 * File setup/shutdown
779 extern void close_files(struct thread_data *);
780 extern int __must_check setup_files(struct thread_data *);
781 extern int __must_check open_files(struct thread_data *);
782 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
783 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
784 extern void generic_close_file(struct thread_data *, struct fio_file *);
785 extern int add_file(struct thread_data *, const char *);
786 extern void get_file(struct fio_file *);
787 extern void put_file(struct thread_data *, struct fio_file *);
788 extern int add_dir_files(struct thread_data *, const char *);
789 extern int init_random_map(struct thread_data *);
790 extern void dup_files(struct thread_data *, struct thread_data *);
791 extern int get_fileno(struct thread_data *, const char *);
792 extern void free_release_files(struct thread_data *);
797 extern void print_thread_status(void);
798 extern void print_status_init(int);
803 #ifdef FIO_HAVE_DISK_UTIL
804 extern void show_disk_util(void);
805 extern void disk_util_timer_arm(void);
806 extern void init_disk_util(struct thread_data *);
807 extern void update_io_ticks(void);
809 #define show_disk_util()
810 #define disk_util_timer_arm()
811 #define init_disk_util(td)
812 #define update_io_ticks()
816 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
817 * will never back again. It may cycle between running/verififying/fsyncing.
818 * Once the thread reaches TD_EXITED, it is just waiting for the core to
835 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
836 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
837 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
842 extern int __must_check fio_pin_memory(void);
843 extern void fio_unpin_memory(void);
844 extern int __must_check allocate_io_mem(struct thread_data *);
845 extern void free_io_mem(struct thread_data *);
850 #define queue_full(td) list_empty(&(td)->io_u_freelist)
851 extern struct io_u *__get_io_u(struct thread_data *);
852 extern struct io_u *get_io_u(struct thread_data *);
853 extern void put_io_u(struct thread_data *, struct io_u *);
854 extern void requeue_io_u(struct thread_data *, struct io_u **);
855 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
856 extern long __must_check io_u_queued_complete(struct thread_data *, int);
857 extern void io_u_queued(struct thread_data *, struct io_u *);
858 extern void io_u_log_error(struct thread_data *, struct io_u *);
859 extern void io_u_init_timeout(void);
860 extern void io_u_set_timeout(struct thread_data *);
861 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
864 * io engine entry points
866 extern int __must_check td_io_init(struct thread_data *);
867 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
868 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
869 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
870 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
871 extern int __must_check td_io_commit(struct thread_data *);
872 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
873 extern void td_io_close_file(struct thread_data *, struct fio_file *);
878 #ifdef FIO_HAVE_BLKTRACE
879 extern int is_blktrace(const char *);
880 extern int load_blktrace(struct thread_data *, const char *);
884 * If logging output to a file, stderr should go to both stderr and f_err
886 #define log_err(args...) do { \
887 fprintf(f_err, ##args); \
888 if (f_err != stderr) \
889 fprintf(stderr, ##args); \
892 #define log_info(args...) fprintf(f_out, ##args)
894 FILE *get_f_out(void);
895 FILE *get_f_err(void);
897 struct ioengine_ops {
898 struct list_head list;
902 int (*setup)(struct thread_data *);
903 int (*init)(struct thread_data *);
904 int (*prep)(struct thread_data *, struct io_u *);
905 int (*queue)(struct thread_data *, struct io_u *);
906 int (*commit)(struct thread_data *);
907 int (*getevents)(struct thread_data *, int, int, struct timespec *);
908 struct io_u *(*event)(struct thread_data *, int);
909 int (*cancel)(struct thread_data *, struct io_u *);
910 void (*cleanup)(struct thread_data *);
911 int (*open_file)(struct thread_data *, struct fio_file *);
912 void (*close_file)(struct thread_data *, struct fio_file *);
917 #define FIO_IOOPS_VERSION 7
919 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
920 extern void register_ioengine(struct ioengine_ops *);
921 extern void unregister_ioengine(struct ioengine_ops *);
922 extern void close_ioengine(struct thread_data *);
925 * Mark unused variables passed to ops functions as unused, to silence gcc
927 #define fio_unused __attribute((__unused__))
928 #define fio_init __attribute__((constructor))
929 #define fio_exit __attribute__((destructor))
931 #define for_each_td(td, i) \
932 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
933 #define for_each_file(td, f, i) \
934 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
936 #define fio_assert(td, cond) do { \
939 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
940 (td)->runstate = TD_EXITED; \
941 (td)->error = EFAULT; \
946 static inline void clear_error(struct thread_data *td)
949 td->verror[0] = '\0';